Sause 2016

Download as pdf or txt
Download as pdf or txt
You are on page 1of 647

Springer Series in Materials Science 242

Markus G.R. Sause

In Situ
Monitoring of
Fiber-Reinforced
Composites
Theory, Basic Concepts, Methods, and
Applications
Springer Series in Materials Science

Volume 242

Series editors
Robert Hull, Charlottesville, USA
Chennupati Jagadish, Canberra, Australia
Yoshiyuki Kawazoe, Sendai, Japan
Richard M. Osgood, New York, USA
J€
urgen Parisi, Oldenburg, Germany
Tae-Yeon Seong, Seoul, Korea, Republic of
Shin-ichi Uchida, Tokyo, Japan
Zhiming M. Wang, Chengdu, China

More information about this series at http://www.springer.com/series/856


The Springer Series in Materials Science covers the complete spectrum of
materials physics, including fundamental principles, physical properties, materials
theory and design. Recognizing the increasing importance of materials science in
future device technologies, the book titles in this series reflect the state-of-the-art
in understanding and controlling the structure and properties of all important
classes of materials.
Markus G.R. Sause

In Situ Monitoring of
Fiber-Reinforced Composites

Theory, Basic Concepts, Methods,


and Applications
Markus G.R. Sause
Institute of Physics and Institute of Materials
Resource Management
University of Augsburg
Augsburg, Germany

ISSN 0933-033X ISSN 2196-2812 (electronic)


Springer Series in Materials Science
ISBN 978-3-319-30953-8 ISBN 978-3-319-30954-5 (eBook)
DOI 10.1007/978-3-319-30954-5

Library of Congress Control Number: 2016940044

© Springer International Publishing Switzerland 2016


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.

Printed on acid-free paper

This Springer imprint is published by Springer Nature


The registered company is Springer International Publishing AG Switzerland
To Kerstin
and my parents Brigitte and Dieter
Preface

The design of materials with essentially new properties by modification of existing


materials is an ongoing journey in human history. Today, more than one hundred
thousand engineering materials are known and are used to fabricate the goods used
in our daily life and to enable some of the most challenging technological applica-
tions. Still there is a strong desire to increase the performance of some materials or
to substitute existing materials to overcome technological limitations or to reduce
the ecological footprint. In materials science, one powerful trend always has been to
compose materials by incorporation of spurious dopants in an otherwise pure
material. This rather simple concept gave rise to the large amount of metallic
alloys, ceramic compounds, and polymer blends we know today. One specific
type of microscopic dopants to incorporate in a material is reinforcement fibers.
These are often found to substantially increase the mechanical performance of the
so-obtained composite material. Hence, this concept of fiber reinforcement has led
to unique materials, exhibiting some of the most extraordinary properties available
for today’s engineering applications. While those fiber-reinforced composites have
now been in use for a long time in aerospace applications, other industry branches
are still at the edge to use these materials for their products. Some of the main
challenges of these materials are the production process itself, but also the theoret-
ical description of the material. Still the failure behavior of this class of materials is
subject to numerous investigations, and theoretical predictions of their ultimate
limits lack in accuracy. One reason for this lack of theoretical description is the
complexity of material structure found in fiber-reinforced materials. Apart from the
existence of two (or more) material ingredients with different properties, a technical
composite exhibits a hierarchical structure covering at least four orders of magni-
tude in length scale. The material itself typically exhibits nonlinear mechanical
behavior and nonlinear temperature response as well as pronounced anisotropy of
all material properties.
This naturally causes a strong need for adequate test methods that are able to
provide the relevant materials properties on every length scale required for the
description of the material. In the past decades, many new test procedures have

vii
viii Preface

been established to specifically address the need of characterization of composite


materials. With the advent of modern theories to describe the complex failure
mechanisms occurring in composite materials, even more properties are required
to be measured nowadays. While this has already led to several new test concepts,
one particular field of research, which has seen a strong interest within the last
years, is the in situ characterization of composite materials. As a major benefit
compared to other test concepts, the in situ approach does not just provide a single
value after the measurement, but instead is able to record the full evolution of
damage in the material until the occurrence of final failure. Consequently, all this
information can be used to understand the evolution of failure in composite
materials.
The aim of this book is to provide an overview of some of the emerging in situ
test concepts and method combinations to demonstrate their applicability to assist
in research of composite materials. It is thought to provide a first insight into the
operational principles of each method and to present the different ways of inter-
pretation of the obtained results. It further provides an overview of some practical
applications ready to be used in quality control and discusses the applicability of the
individual methods as well as their strengths and weaknesses.
The book is organized in seven chapters. Chapter 1 provides an introduction to
the relevance of in situ testing for fiber-reinforced composites. Chapter 2 briefly
reviews the present state of failure theories available to predict failure of fiber-
reinforced polymers, since a majority of the experimental methods need a thorough
understanding of the failure mechanisms occurring in these materials in order to
interpret their sensitivity for detection. Chapters 3–6 introduce different experi-
mental methods used for in situ characterization. In all cases, a focus is given on the
operation principles and recent developments. Reference to standard literature is
made as much as possible to facilitate the reading. Chapter 3 deals with the
applications of digital image correlation (DIC), and Chap. 4 emphasizes the devel-
opments in acoustic emission (AE) analysis. Chapter 5 has its focus on electromag-
netic emission (EME) analysis, while Chap. 6 discusses possibilities of in situ
computed tomography (CT). In Chap. 7, a comparison of the capabilities of the
methods in application to in situ testing of composite materials is made, and
suitable method combinations are discussed.
As supplementary data, Appendix A provides a comprehensive list of common
experimental parameters influencing comparability of acoustic emission measure-
ments. Appendix B lists all of the materials properties used for the modeling work
throughout the chapters. Appendix C provides an overview of acoustic emission
parameter definitions, and common definitions and abbreviations are found in
Appendix D.

Augsburg, Germany Markus G.R. Sause


Acknowledgments

First of all, I would like to express my gratitude to Prof. Dr. Siegfried Horn, Prof.
Dr. Ferdinand Haider, and Prof. Dr. Marvin Hamstad for guiding me in the process
of Habilitation and for providing every level of understanding and support I needed
during this process.
My gratitude is also to Prof. Kanji Ono, Prof. Stepan Lomov, and Prof. Ian
Sinclair to review this work as part of my Habilitation thesis review. My gratitude is
also to Prof. Dr. Marvin Hamstad and Prof. Dipl.-Ing. Dr. techn. Roland Hinterh€olzl
for proofreading and reviewing parts of this book and thus providing me the
opportunity to improve the content.
I would also like to thank all of the students, who have contributed to this book
by providing experimental or modeling data or by intensive discussion of their
interpretation. The work would not have been possible without the valuable con-
tribution of numerous Ph.D. students. I would like to thank Stefan Richler for
micromechanical test data, Sebastian Gade for performing EME measurements and
modeling work, Sinan Kalafat for acoustic emission measurements and in situ
computed tomography measurements, Thomas Guglh€or for computed tomography
imaging, Andreas Monden and Michael Greisel for digital image correlation data,
and Ursula Weiss for accompanying modeling work.
Further, I am indebted to all other students who have contributed within their
master’s or bachelor’s thesis. I would like to thank Marina Pl€ockl, Marcus
Bornschlegl, and Wolfgang Skopalik for acoustic emission measurements, Eva
Laukmanis for micromechanical test data, Florian Staab for electromagnetic emis-
sion data, Nora Schorer for a majority of the digital image correlation data, Uli
Buchner and Tobias Krones for computing tomography investigations, and Philipp
Potstada and Benjamin Alaca for electromagnetic emission measurements.
Further thanks also go to the technical staff for their ongoing support in testing
issues and for taking care of the students’ needs. I would like to thank Stefan
Schmitt for the good laboratory organization, Sabine Bessel for sensor calibration,
and Katja Ziebe-Eiffler for digital image correlation and mechanical testing data as
well as for specimen fabrication.

ix
x Acknowledgments

Next, I would like to thank all academic colleagues for their collaboration in the
various fields covered by this book. Substantial parts of my present work were
inspired by the work of Dr.-Ing. habil. J€urgen Bohse, Dr. Andreas Brunner, Prof.
Dr.-Ing. habil. Dipl.-Geophys. Christian Große, Prof. Emmanuel Ramasso, Prof.
Vincent Placet, Dr.-Ing. Tobias M€uller, Prof. Dr.-Ing. Alexander Horoschenkoff,
and Prof. Antonios Kontsos. The contributions of various modeling aspects by my
colleague Dr.-Ing. Andreea-Manuela Zelenyak are highly appreciated. Many
thanks are also to those professionals from industry for their constant feedback
regarding applicability of the method in industrial environments. In particular, I
would like to thank Birte H€ock, Joachim Scharringhausen, and Dr. Gunter Geiss
from MT Aerospace AG, Dr. Matthias Goldammer and Dr. Detlef Rieger from
Siemens AG, and Dr. Rui de Oliveira from BASF AG. Special recognition is also
noted for all persons allowing me to coauthor their scientific publications, namely
Dr. Wolfgang M€ uller, Dr. Judith Moosburger-Will, Dr. G€unter Obermeier, Torben
Priess, Michael Greisel, and Florian Henne. My deep gratitude is to Dr. Lidewei
Vergeynst, Dr. Franziska Baensch, and Prof. Dr. Peter Niemz for the opportunity to
get a deep insight into in situ methods application to wood materials.
I would also like to thank all of those I have forgotten to mention and all of those
who have used their valuable time to discuss any scientific aspects of this work.
Contents

1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2 Failure of Fiber-Reinforced Composites . . . . . . . . . . . . . . . . . . . . . . 5
2.1 Classification of Failure Mechanisms . . . . . . . . . . . . . . . . . . . . . . 6
2.1.1 Microscale . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.1.2 Mesoscale . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.1.3 Macroscale . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.2 Failure Theories for Fiber-Reinforced Composites . . . . . . . . . . . . 18
2.2.1 Quasi-Static Failure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.2.2 Quasi-Static Failure Including Growth of Damage,
Damage Mechanics, and Degradation . . . . . . . . . . . . . . . . 21
2.2.3 Long-Term Behavior . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.2.4 High Velocity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.3 Challenges in Mechanical Testing of Fiber-Reinforced
Materials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
2.3.1 Detection of First Failure Onsets . . . . . . . . . . . . . . . . . . . 34
2.3.2 Tracking Failure Evolution . . . . . . . . . . . . . . . . . . . . . . . 35
2.3.3 Ductile Matrix Materials . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.4 What Can In Situ Methods Contribute to Mechanical Testing? . . . 36
2.4.1 In Situ Microscopy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
2.4.2 Digital Image Correlation . . . . . . . . . . . . . . . . . . . . . . . . 37
2.4.3 X-Ray Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
2.4.4 Thermography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
2.4.5 Shearography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
2.4.6 Ultrasonic Measurements . . . . . . . . . . . . . . . . . . . . . . . . . 42
2.4.7 Acoustic Emission . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
2.4.8 Electromagnetic Emission . . . . . . . . . . . . . . . . . . . . . . . . 44
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45

xi
xii Contents

3 Digital Image Correlation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57


3.1 Principle of Operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
3.2 System Accuracy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
3.2.1 Error Sources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
3.2.2 Resolution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
3.3 Strain Concentration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
3.3.1 Measurement of Strain Concentration Due
to Internal Defects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
3.3.2 FEM Modeling of Strain Concentration Due
to Internal Defects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
3.3.3 Detectability of Defects Using DIC . . . . . . . . . . . . . . . . . 97
3.4 Application to Composites . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
3.4.1 DIC as Optical Extensometer . . . . . . . . . . . . . . . . . . . . . . 111
3.4.2 Detection of Failure Onsets . . . . . . . . . . . . . . . . . . . . . . . 118
3.4.3 Detectability of Failure Mechanisms . . . . . . . . . . . . . . . . . 124
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
4 Acoustic Emission . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
4.1 Principle of Operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
4.2 Source Mechanics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
4.2.1 AE Rise-Times and Plate Waves . . . . . . . . . . . . . . . . . . . 135
4.2.2 AE Source Model Implementation for Fiber
Reinforced Materials . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
4.2.3 Case Studies for Acoustic Emission Sources
in Fiber Reinforced Materials . . . . . . . . . . . . . . . . . . . . . . 147
4.2.4 Detectability of Failure Mechanisms . . . . . . . . . . . . . . . . . 187
4.3 Wave Propagation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192
4.3.1 Attenuation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202
4.3.2 Influence of Geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . 206
4.3.3 External and Internal Obstacles . . . . . . . . . . . . . . . . . . . . 214
4.4 Detection of Acoustic Emission Signals . . . . . . . . . . . . . . . . . . . . 226
4.4.1 Comparison of Sensor Types . . . . . . . . . . . . . . . . . . . . . . 230
4.4.2 Sensor Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 233
4.4.3 Waveguides . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 239
4.4.4 Other Factors Affecting Sensor Sensitivity . . . . . . . . . . . . 245
4.5 Signal Classification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 250
4.5.1 Recommended Practices Before Starting Signal
Classification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 252
4.5.2 Pattern Recognition Method to Detect
Natural Clusters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 256
4.5.3 Uncertainty of Classification . . . . . . . . . . . . . . . . . . . . . . 260
4.5.4 Factors of Influence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 264
4.6 Source Localization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 282
4.6.1 Determination of Signal Onset . . . . . . . . . . . . . . . . . . . . . 285
4.6.2 Classical Localization Methods . . . . . . . . . . . . . . . . . . . . 287
Contents xiii

4.6.3 Source Localization Methods Based


on Neural Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 291
4.7 Application to Composites . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 305
4.7.1 AE Source Identification Using FEM Results . . . . . . . . . . 305
4.7.2 Detection of Failure Onset . . . . . . . . . . . . . . . . . . . . . . . . 309
4.7.3 Comparison to Failure Criteria Predictions . . . . . . . . . . . . 317
4.7.4 Tracking Failure Evolution . . . . . . . . . . . . . . . . . . . . . . . 327
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 349
5 Electromagnetic Emission . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 361
5.1 Principle of Operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 361
5.2 Source Mechanism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 363
5.2.1 Modeling of Electromagnetic Emission Sources . . . . . . . . 371
5.2.2 Test Sources for Electromagnetic Emission . . . . . . . . . . . . 389
5.3 Signal Propagation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 396
5.4 Detection of Electromagnetic Emission Signals . . . . . . . . . . . . . . 400
5.4.1 EME Detector Concepts . . . . . . . . . . . . . . . . . . . . . . . . . 401
5.4.2 Electromagnetic Shielding . . . . . . . . . . . . . . . . . . . . . . . . 413
5.5 Application to Composites . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 427
5.5.1 Measurement of EME Due to Crack Formation . . . . . . . . . 427
5.5.2 Measurement of Fracture Surface Orientation . . . . . . . . . . 445
5.5.3 Detectability of Failure Mechanisms . . . . . . . . . . . . . . . . . 450
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 453
6 Computed Tomography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 457
6.1 Principle of Operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 458
6.2 Detail Visibility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 461
6.2.1 Object Resolution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 462
6.2.2 Artifacts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 464
6.2.3 Detectability of Defects in Fiber Reinforced Materials . . . . 475
6.3 Volumetric Inspection of Materials . . . . . . . . . . . . . . . . . . . . . . . 481
6.3.1 Concepts for Ex Situ Loading . . . . . . . . . . . . . . . . . . . . . 481
6.3.2 Concepts for In Situ Loading . . . . . . . . . . . . . . . . . . . . . . 482
6.4 Digital Volume Correlation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 493
6.5 Application to Composites . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 498
6.5.1 Ex Situ Testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 502
6.5.2 In Situ Testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 515
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 527
7 Combination of Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 533
7.1 Comparison of In Situ Methods . . . . . . . . . . . . . . . . . . . . . . . . . . 534
7.1.1 In Situ Capabilities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 536
7.1.2 Detectability of Failure Mechanisms . . . . . . . . . . . . . . . . . 538
7.1.3 Detection Sensitivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . 540
7.1.4 Extension of Methods to Large Scale
and Field Testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 543
xiv Contents

7.2 Established Method Combinations . . . . . . . . . . . . . . . . . . . . . . . . 548


7.2.1 Imaging Methods and Acoustic Emission . . . . . . . . . . . . . 548
7.2.2 Digital Image Correlation and Acoustic Emission . . . . . . . 555
7.2.3 Thermography and Acoustic Emission . . . . . . . . . . . . . . . 558
7.2.4 Computed Tomography and Acoustic Emission . . . . . . . . 563
7.2.5 Computed Tomography and Digital Image Correlation . . . 590
7.2.6 Electromagnetic Emission and Acoustic Emission . . . . . . . 590
7.2.7 Acousto-Ultrasonics . . . . . . . . . . . . . . . . . . . . . . . . . . . . 596
7.3 Future Developments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 601
7.3.1 Digital Image Correlation . . . . . . . . . . . . . . . . . . . . . . . . 602
7.3.2 Acoustic Emission . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 602
7.3.3 Electromagnetic Emission . . . . . . . . . . . . . . . . . . . . . . . . 604
7.3.4 In Situ Computed Tomography . . . . . . . . . . . . . . . . . . . . 605
7.3.5 Method Combinations . . . . . . . . . . . . . . . . . . . . . . . . . . . 606
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 606

Appendix A: Acoustic Emission—Parameters of Influence . . . . . . . . . . 611

Appendix B: Material Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 617

Appendix C: Acoustic Emission Signal Parameters . . . . . . . . . . . . . . . . 621

Appendix D: Definitions and Abbreviations . . . . . . . . . . . . . . . . . . . . . . 625

Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 629
Financial Support

Financial support for the research described in this book was provided by a number
of sources. I gratefully acknowledge the financial support by the Federal Ministry
of Education and Research (BMBF) for funding the joint research projects MAI zfp
and MAI plast within the leading edge cluster MAI Carbon, the joint research
project FORCiMA funded by the Bavarian Research Foundation, the German
Research Foundation (DFG) for funding the research project HO 955/8-1 and
HO 955/8-2, the Free state of Bavaria for funding of the project ComBo within the
program “BayernFIT,” and the Free state of Bavaria for funding the project
CFK-Integral within the program “Neue Materialien f€ ur Bayern.”

xv
About the Author

Markus Sause studied Physics at the University of Augsburg and earned his
doctoral degree in 2010 in Experimental Physics at the same institution. He
received the “Erich-Krautz-Preis” in 2010 for his outstanding contribution to the
interpretation of acoustic emission of fiber-reinforced materials. In 2015, he was
awarded his Habilitation in Experimental Physics. He is lecturer at the University of
Augsburg and head of the Materials Testing Research Group at the Institute of
Physics and at the Institute of Materials Resource Management. Since 2014, he has
been a member of the EWGAE executive committee and active in several other
committees dedicated to the testing and analysis of fiber-reinforced materials. His
research interests span the mechanics of fiber-reinforced composites, their destruc-
tive and nondestructive testing, as well as numerical methods to interpret the
materials behavior. A special focus is given to bridge the gap between destructive
testing approaches and nondestructive inspection to perform in situ analysis of
materials failure.

xvii
Chapter 1
Introduction

Failure of engineering materials is a phenomenon, which has been present since the
early days of mankind and has been investigated on a modern scientific basis
starting at the advent of the industrial age. Only in recent decades it has become
convenient to build structures based on previous numerical calculations to predict
failure on the material level rather than based on empirical knowledge. The
pioneering work of those researchers with a focus on engineering materials like
metals, concrete, ceramics, and polymers has enabled us to allow construction of
technically safe structures out of these materials. In recent time it has become a
significant trend in material science to fabricate composite materials in order to
combine benefits of different material classes. Such composite materials are applied
for a variety of reasons. Some material combinations aim to improve electrical
conductivity by inclusion of particles, others try to improve the mechanical behav-
ior by adding reinforcing elements. The increasing number of interfaces inherent to
composite materials generally induces a more complex failure behavior than
encountered in pure materials.
Fiber reinforced polymers (FRPs) are a class of materials that show an extraor-
dinary strength-to-weight ratio and stiffness-to-weight ratio. However, the current
limited predictability of material failure typically has resulted in high safety
margins for the allowable design limits in construction of FRP structures. There-
fore, in safety relevant structures, the full potential of this material class is seldom
used. Many attempts have been made to transfer concepts of fracture mechanics and
failure prediction from pure materials to such heterogeneous and very often aniso-
tropic materials. Since structural failure in FRP is a complex evolution of various
microscopic failure mechanisms, understanding of their evolution is key to under-
stand global failure and to reach reliable predictions. Also, the structural integrity is
affected differently by the occurrence of different failure mechanisms. Thus, it is a
requirement to distinguish between different types of failure mechanisms that occur
within the structure under load.
One comprehensive attempt to assess the status of today’s failure theories for
FRPs is known as world-wide failure exercise. The organizers of this round-robin

© Springer International Publishing Switzerland 2016 1


M.G.R. Sause, In Situ Monitoring of Fiber-Reinforced Composites,
Springer Series in Materials Science 242, DOI 10.1007/978-3-319-30954-5_1
2 1 Introduction

test invited leading researchers world-wide to predict failure of selected laminate


configurations for a set of load cases. This comparative assessment of theoretical
failure prediction has brought together researchers working on the same field but
using fairly different approaches. Due to the complexity of failure found in fiber
reinforced materials, it was one of the essential results of the “World-Wide Failure
Exercise I” [1–4] and “World-Wide Failure Exercise II” [5–7], that for some load
cases theoretical predictions of the different participants were differing by a large
amount relative to each other and relative to the experimental data. Since in the
world-wide failure exercises the benchmark was the experimentally obtained
values, it was also argued [8, 9] whether the experimental data is reliable enough
to be used for comparison.
While the required parameters for failure prediction in metals, ceramics, or
polymers are in many cases straight forward to measure, the testing of fiber
reinforced composites remains a challenge by itself. Moreover, due to the anisot-
ropy the number of required material parameters is substantially larger in fiber
reinforced materials than for isotropic engineering materials.
There is an ongoing discussion regarding the testing methods involved to obtain
material properties of FRPs. Present standards to obtain material properties for fiber
reinforced composites focus on the signatures of load–displacement curves or
stress–strain curves to deduce angle dependent moduli or strength values. However,
for some cases such global stress–strain curves have only limited relevance to the
determination of the true (or required) material properties. Since microscopic
failure precedes the ultimate failure of the test specimen on the macroscopic
scale, it is doubtful if maximum values of stress–strain curves are the correct way
to derive relevant material properties. The heterogeneity and anisotropy on the
mesoscopic scale can often induce extremely different stress and strain states than
expected from the macroscopic loading condition. In particular, the individual
layers of a laminate are subject to constraining effects of the neighboring layers
and thus exhibit stress states distinctly different to the global average. After
initiation of first failure on the microscopic scale, a complex interplay of failure
mechanisms initiates that involves mechanisms on length-scales spanning from the
atomistic scale to the dimensions of the test specimen. The loading conditions, the
fiber architecture, fiber volume fraction, and the fracture behavior of the matrix
material all contribute to the ability of the material to survive upon further loading
until finally ultimate failure occurs on the macroscopic scale. Due to the statistically
driven initiation and evolution of microscopic failure, FRPs have the reputation to
typically exhibit a large scatter of material properties. The significance of this
statistical distribution of initiation sites for failure is known as a volume effect
and has been intensively investigated [10].
The present work does not aim to improve the mechanical testing part or the
failure theories themselves, but to provide the necessary experimental methods to
allow monitoring of the progression of failure in FRPs. In this sense, failure refers
to the occurrence of microscopic or macroscopic networks of cracks in the material.
If failure initiation can be detected in an appropriate way, imaging methods can be
used to capture the accumulated damage state present in the material at that stage.
1 Introduction 3

Such imaging can then be incorporated in a modeling environment and be used to


allow a detailed forward prediction of the failure behavior or it can be used to
deduce more relevant material properties as an input for failure theories.
The work presented herein is thus motivated by two possibilities of composite
analysis enabled by the presence of validated experimental methods for:
1. Detection of failure initiation
This deals with the applicability of experimental methods to detect the first onset
of a particular failure mechanism in a fiber reinforced composite. If such failure
mechanisms can be detected with certain reliability, the resulting local or global
stress and strain levels can be related to this failure onset to derive more
representative material properties.
2. Tracking failure evolution
After failure initiation, the proposed experimental methods should allow the
continuous monitoring of failure evolution in FRPs. This allows for continuous
comparison of the experimental damage states relative to theoretical predictions.
Moreover, this allows deducing subsequent stages of damage progression in
order to improve the understanding of failure in general.
These two possibilities are exemplified in application to detect crack initiation
and crack propagation in Fig. 1.1. To investigate initiation and growth of cracks in
FRPs, a variety of experimental methods are applied. Imaging methods are fre-
quently used to monitor the damage progress after loading and unloading of test
specimens. This allows visualization of the damage progress in discrete steps, but
the unloading of the specimens causes existent cracks to close and makes them

load
load Detection of
failure initiation

Tracking
failure evolution

displacement

Fig. 1.1 Detection of failure initiation and tracking of failure evolution illustrated for the
occurrence of crack formation on the microscopic scale and subsequent propagation on the
macroscopic scale
4 1 Introduction

harder to spot. To overcome this problem, one possibility is the application of in situ
microscopy. This yields continuous observation, but the imaging process is either
restricted to surface observations using optical microscopy or electron microscopy
and is limited to the specimen size or resolution as in computed tomography (CT).
As an alternative for surface and near surface, digital image correlation (DIC) can
add valuable information to spot damaged areas by sudden changes in the strain
field. In the same way as the ear complements the eye, the detection of acoustic
emission (AE) can act as complementary method to imaging methods in order to
detect failure initiation and to track growth of damage in the full volume of the
specimen. Also the generation of electromagnetic emission (EME) during crack
initiation and crack growth is a nondestructive method useful for online monitoring
of failure in FRPs covering the full volume of the specimen.

References

1. Soden, P.D., Hinton, M.J., Kaddour, A.S.: Biaxial test results for strength and deformation of a
range of E-glass and carbon fibre reinforced composite laminates: failure exercise benchmark
data. Compos. Sci. Technol. 62, 1489–1514 (2002)
2. Hinton, M.J., Kaddour, A.S., Soden, P.D.: Evaluation of failure prediction in composite
laminates: background to “part B” of the exercise. Compos. Sci. Technol. 62, 1481–1488
(2002)
3. Hinton, M.J., Soden, P.D.: Predicting failure in composite laminates: the background to the
exercise. Compos. Sci. Technol. 58, 1001–1010 (1998)
4. Soden, P.D., Hinton, M.J., Kaddour, A.S.: Lamina properties, lay-up configurations and
loading conditions for a range of fibre-reinforced composite laminates. Compos. Sci. Technol.
58, 1011–1022 (1998)
5. Kaddour, A.S., Hinton, M.J.: Maturity of 3D failure criteria for fibre-reinforced composites:
comparison between theories and experiments: part B of WWFE-II. J. Compos. Mater. 47,
925–966 (2013)
6. Hinton, M., Kaddour, A.: The background to part B of the Second World-Wide Failure
Exercise: evaluation of theories for predicting failure in polymer composite laminates under
three-dimensional states of stress. J. Compos. Mater. 47, 643–652 (2013)
7. Hinton, M., Kaddour, A.: Triaxial test results for fibre reinforced composites: the Second
World-Wide Failure Exercise benchmark data. J. Compos. Mater. 47, 925–966 (2013)
8. Christensen, R.M.: A physically based cumulative damage formalism. In: Major Accomplish-
ments in Composite Materials and Sandwich Structures: An Anthology of ONR Sponsored
Research, pp. 51–65 (2009)
9. Cuntze, R.: Comparison between experimental and theoretical results using Cuntze’s “failure
mode concept” model for composites under triaxial loadings-part B of the Second World-Wide
Failure Exercise. J. Compos. Mater. 47, 893–924 (2013)
10. Wisnom, M.R.: Size effects in the testing of fibre-composite materials. Compos. Sci. Technol.
59, 1937–1957 (1999)
Chapter 2
Failure of Fiber-Reinforced Composites

Generally, we start to refer to failure of a material, when the material no longer


fulfills its intended use. Therefore, the physical phenomenon associated with the
failure is subject to the anticipated use as well. A scratch in the hood of an
automobile is an annoying thing but may not necessarily stop the hood from
operating in its intentional use. Still the scratch may cause initiation of corrosion,
since the protective coating may have been partially removed. In contrast to the
automobile hood, a scratch in a painting directly affects the product quality. The
painting may not look as beautiful anymore and therefore may fail in its intended
use. In the context of this book, the term “failure” always refers to the structural
failure of the material due to mechanical load. The primary source for this failure
are cracks forming within a material causing it to reduce its effective cross section
or introducing instability causing buckling, kinking, or likewise failure modes.
The mechanisms of crack formation and propagation within materials is a
physical phenomenon which is still not fully understood due to the complex
interplay between the local microstructure of the material and its impact on the
macroscopic values of stress fields. In fracture mechanics, the calculation of
material specific quantities like critical strain energy release rates is based on
macroscopically observable quantities like forces and displacements. Microscopi-
cally, the strain energy required for crack propagation is converted to surface
energy, heat, elastic energy release, and plastic deformation. Still one of the
unsolved problems in fracture mechanics is the description of tensile extensions
ahead of the crack tip and the dynamic displacement field surrounding the crack tip
[1]. This is owed to the fact that a direct observation of crack nucleation and
propagation is usually not possible, for example, by optical means, due to the
crack propagation velocities in solid materials. Also the influence of dynamical
instabilities is not fully understood [2].
Despite of such missing scientific details, high-quality phenomenological
approaches exist to describe failure of typical engineering materials due to crack
formation. The fracture mechanics concepts of Griffith and Irwin [3, 4] form the
basis for the description of crack initiation and propagation in metals, polymers and

© Springer International Publishing Switzerland 2016 5


M.G.R. Sause, In Situ Monitoring of Fiber-Reinforced Composites,
Springer Series in Materials Science 242, DOI 10.1007/978-3-319-30954-5_2
6 2 Failure of Fiber-Reinforced Composites

ceramics. However, in fiber-reinforced materials, the hierarchical and heteroge-


neous microstructure superimposes the isotropic and homogenous properties of the
matrix materials. Additional effects arise due to crack branching, crack deflection,
and mixed mode conditions. Therefore, some simple assumptions inherent to the
classical descriptions are no longer fulfilled. Hence, a broad variety of theories is
currently applied for the description of failure in fiber-reinforced polymers.
This chapter starts first with an overview on failure mechanisms in composites
and introduces their hierarchical classification. The second part of this chapter
briefly reviews failure theories currently applied in the field of fiber-reinforced
composites. The third part of this chapter then has its focus on the mechanical
methods used to provide material properties necessary as input data for these failure
theories. In particular, test methods still controversially discussed in the community
are critically reviewed. The last section then finally provides a first overview on
“secondary” methods as being used to improve the “primary” mechanical methods
to obtain valid material data.

2.1 Classification of Failure Mechanisms

During mechanical loading of fiber-reinforced plastics various microscopic damage


mechanisms occur. The accumulated damage finally results in failure of the com-
posite. As seen in Fig. 2.1a, on the microscale cracks can evolve within the matrix
material, breakage of the fibers can occur, or cracks develop at the interface
between both constituents. From the fracture mechanics point of view, the situation
grows more complex, since each of these fundamental mechanisms can be induced
by various types of macroscopic loading conditions. On the mesoscopic scale seen
in Fig. 2.1b, as a function of external loading, these mechanisms can interact, grow,
and can form even more complex mechanisms. Finally their growth develops into a
critical state, which then causes ultimate failure of the composite structure due to
the macroscopic loading condition as seen on the macroscopic scale of Fig. 2.1c.
In the following, an introduction is given for the different failure mechanisms

a b c

Fig. 2.1 Schematic illustration of failure evolution in a cross-ply used to define microscale (a),
mesoscale (b), and macroscale (c)
2.1 Classification of Failure Mechanisms 7

according to a hierarchical classification with respect to the length scale of occur-


rence. Parts of the following discussion are based on the considerations found in
standard literature [5–7] as well as the author’s previous work [8].

2.1.1 Microscale

Fundamentally, failure initiates at the atomistic level. As consequence of excessive


mechanical forces, bonds are separated and cracks can initiate at weak spots or
grow further. There are valuable attempts found in literature to conduct ab initio
predictions of failure based on Monte–Carlo simulations [9, 10]. However, most of
these approaches are still limited in the amount of complexity being captured by the
algorithms. In particular, the consideration of an explicit microstructure is usually
beyond the capabilities of such approaches, that is, solid media may be assumed to
be homogeneous.
In the context of fiber-reinforced composites, the microscale is determined by
the typical sizes of the constituents. In this case, the diameter of the fibers and their
characteristic length are used to define this scale. Therefore, the microscale is in the
range of micrometers, which is considerably larger than the atomistic scales. Since
this allows distinguishing different failure mechanisms beyond failure of the
individual constituents, this is the first relevant scale to explicitly discuss for
fiber-reinforced materials.
Stress components causing the corresponding failure in unidirectional compos-
ites are noted using subset directions indicating parallel (k ) and transverse (⊥ )
orientation relative to the fiber axis. On this scale, the distribution of the reinforce-
ment fibers within the matrix material is of particular relevance. The fraction of
fiber filaments embedded in the matrix material is usually referred to as fiber
volume fraction. Other than on the macroscale, the fiber volume fraction in a real
composite material may change substantially as a function of position. Especially
for fiber architectures such as woven fabrics or for materials with comparatively
low fiber volume fractions, the extreme cases of regions of pure matrix and dense
fiber volume fractions may occur (cf. Fig. 2.2). In order to quantify these effects,
automated image processing of cross sections [11] as well as volumetric analysis
of fiber distributions by computed tomography have emerged as suitable
methods [12].
One of the key hypotheses in fracture mechanics is the fact that anomalies are
always present in real materials. In the context of fiber-reinforced materials, there
are several characteristic defects located at the microscale, which can be understood
as such anomalies. In Figs. 2.3, 2.4, and 2.5 such typical defects found in real fiber-
reinforced composites are shown.
As a residue of the thermal processing, polymer matrix materials often show
porosity at various scales. These pores can result from gases of the chemical curing
process or from missing consolidation pressure during the solidification. The type
of porosity smaller in extent than fiber diameters is often referred to as
8 2 Failure of Fiber-Reinforced Composites

Fig. 2.2 Examples of fiber distribution on the microscale in glass/epoxy-woven fabrics (a) and
unidirectional carbon/epoxy (b)

Fig. 2.3 Schematic illustration of microporosity and mesoporosity (left) and according scanning
electron microscopy images of carbon/epoxy material (right)

microporosity. At this scale, many small pores reside within the matrix material,
sometimes resulting in foam-like microstructure as seen in Fig. 2.3.
At the next length scale (size of fiber diameters), the pores often stick to the fiber
filaments or fiber bundles and therefore cause regions, where the fibers are not
supported by the matrix material. This can cause preliminary buckling if the
2.1 Classification of Failure Mechanisms 9

thickness thickness

Fig. 2.4 Schematic illustration of in-plane (left) and out-of-plane (right) undulations

2 mm

Fig. 2.5 Schematic illustration of particle inclusion, resin rich areas and illustrative microscopy
images of a carbon/epoxy material

composite is subject to compressive load or shear load. In deviation to the termi-


nology of length scale used herein, such types of pores are already referred to as
mesoporosity. This is to reflect the length scale of microporosity found below.
Another type of defect which is always present to some extent within fiber-
reinforced materials is undulation (waviness) of the fibers. As seen in Fig. 2.4, one
can distinguish between in-plane and out-of-plane undulations relative to the
thickness direction of the laminate. Due to the waviness of the fiber filaments or
fiber bundles, the local stress exposure can dramatically change. For a tensile load,
10 2 Failure of Fiber-Reinforced Composites

the neighboring fiber filaments may thus be subject to higher stress levels, while in
compressive load the waviness may induce early fiber buckling.
Other types of anomalies frequently encountered in fiber-reinforced composites
are shown in Fig. 2.5. Inclusions of dust particles, resin-rich areas, binder yarns or
fiber knots all share the common impact of disturbing the local fiber distribution.
This is either realized in form of fiber undulations or in affecting the local fiber
volume fraction. Both effects considerably change the local stress exposure of fiber
and matrix material and therefore can cause preliminary failure. However, some of
these defects can also cause benign failure behavior. One of the main improvements
in terms of damage tolerance is the ability of the fibers to cause crack deflection of
interfiber cracks. Here, every crack deflection or crack branching is advantageous,
since the additional energy necessary to elongate the crack propagation path
contributes directly to the macroscopic fracture toughness.
But even fiber-reinforced materials free of defects have intrinsic weak spots.
Due to production process at elevated temperatures, most fiber-reinforced polymers
are subject to a residual thermal stress between fibers and matrix. Here, the matrix
material between two adjacent fibers is exposed to a geometrically induced stress
concentration and therefore acts as weakest link when loading the material trans-
verse to the fiber axis [5, 13–15].
Perhaps, the most critical type of failure induced within fiber-reinforced poly-
mers is fiber failure. As the reinforcement part of the composite, this inherently
degrades the structures ability to bear increased load. For uniaxial tensile loading
parallel to the fiber axis, fiber breakage is observed if the local strength of the
fiber filament is exceeded. However, such single fiber breakage does not neces-
sarily result in ultimate failure of the composite. On the microscopic scale fibers
show a broad distribution in their ultimate strength due to variations of the fiber
diameter and internal flaws [5, 16]. This can result in early failure of numerous
fibers, while the remaining fibers still act as load-bearing elements. Due to this
strength distribution of the fibers, weak fibers are the first to fail. As a conse-
quence of the fracture of a single fiber, the stress concentration at the position of
rupture initiates debonding between fiber and matrix as well as fracture of the
surrounding matrix material. This type of failure is schematically shown in
Fig. 2.6 and is denoted fiber breakage in the following. For comparison, a
respective scanning electron microscopy image of fiber breakage caused by
tensile loading is shown as well.
For compressive loading, the failure behavior of fibers is characteristically
different. Due to the compressive load, the reinforcement fibers typically buckle,
once the matrix material cannot support their orientation any longer. After that
buckling point, band kinking occurs, which finally results in crushing of the fibers at
distinct positions [17] as schematically shown in Fig. 2.7. For comparison, a
corresponding scanning electron microscopy image of compressive fiber failure is
shown in Fig. 2.7. As indicated in the schematic illustration, this type of failure
naturally results in significant fiber fragmentation, when observed at the surface
level.
2.1 Classification of Failure Mechanisms 11

+σ|| +σ||

Fig. 2.6 Schematic illustration of fiber breakage due to tensile loading and corresponding electron
microscopy image of a carbon/epoxy material in surface view (based on [8])

–σ|| –σ||

Fig. 2.7 Schematic illustration of fiber crushing due to compressive loading and corresponding
electron microscopy image of a carbon/epoxy material in surface view (based on [8])

The type of failure observed most often in fiber-reinforced structures is interfiber


fracture. As schematically shown in Fig. 2.8, this failure of the matrix material can
originate from various loading conditions. In all cases, cracks initiate or propagate
within the matrix or at the interface between fiber and matrix, if the respective
loading condition of normal stresses or shear stresses exceeds the local strength of
the matrix or interface, respectively.
In principle, the process of interfiber fracture can be further decomposed into
different microscopic failure mechanisms. While fracture of the matrix material
itself is always involved to some extent, the failure of the fiber–matrix interface is
also likely to occur. At the microscopic scale, one can distinguish between the
process of fiber–matrix debonding and the fiber–matrix pull-out.
As debonding the partial or full loss of contact between fiber and matrix due to
external load is understood, which can occur in various load configurations
(cf. Fig. 2.9).
As second failure process at the interface between fiber and matrix, the fiber–
matrix pull-out is often discussed in literature [18–23]. As seen in Fig. 2.10, this
involves a fiber–matrix debonding process as initial damage state, but then is
followed by a sliding of the fiber filament relative to the matrix. The filament is
not necessarily destroyed, but often spans the evolving crack in the matrix material.
However, this failure mechanism is only likely to occur for tensile loading in
12 2 Failure of Fiber-Reinforced Composites

+σ⊥
+σ⊥

–σ⊥

–σ⊥

τ||⊥

τ⊥⊥ τ⊥⊥

Fig. 2.8 Schematic illustration of interfiber fracture (matrix cracking) and corresponding electron
microscopy cross-sectional images of a carbon/epoxy material for typical load cases (based on [8])

+σ⊥ +σ||

Fig. 2.9 Schematic illustration of fiber–matrix debonding due to transverse stress and due to fiber
parallel stress
2.1 Classification of Failure Mechanisms 13

pull-out

+σ||
+σ||

Fig. 2.10 Schematic illustration of fiber–matrix pull-out due to fiber parallel stress with respec-
tive scanning electron microscopy image of a carbon/epoxy material

parallel to the fiber axis and for fiber–matrix combinations with relatively low
bonding strength. If the adhesive bonding between fiber and matrix is too large, no
debonding can initiate due to shear forces at the interface and fiber failure is
observed instead.

2.1.2 Mesoscale

Beyond the length scale defined by the composite constituents, the mesoscale is
found. In fiber-reinforced materials, the relevance of this scale typically arises due
to the textile architecture chosen to build laminates. In technical applications,
laminates with only one fiber orientation are usually not relevant since they exhibit
very poor mechanical properties transverse to the fiber orientation. Therefore,
laminates are frequently fabricated as stacks of individual plies consisting of
unidirectional layers in various orientations, woven fabrics, other textiles, or mix-
tures thereof. In the following, the term mesoscale is used to define the length scale
of interaction between these different plies forming the laminate; thus, it is located
in the range of several hundred micrometers.
For loading in the out-of-plane direction of the laminate, intralaminar delami-
nation (splitting) and interlaminar delamination (inter-ply) can occur. The relevant
material property determining inter-ply failure is the fiber–matrix adhesion. As seen
in the microscopy images in Fig. 2.12, the magnitude of the bonding strength
between fiber and matrix dominates the laminates interlaminar fracture resistance.
As a function of increased bonding strength, the fracture surfaces show a pro-
nounced hackle structure [7], which is caused by an increased amount of crack
deflection and crack branching and therefore requires more energy for the same
amount of macroscopically relevant crack propagation length (cf. Fig. 2.12).
For the interlaminar failure, the weak link is the individual ply itself. As for the
inter-ply delamination, the relevant material property is the fiber–matrix adhesion.
If the individual ply is exposed to high out-of-plane stress, it is also likely that
failure will occur within a layer. As seen in Fig. 2.13, there is no predescribed
fracture plane, so interlaminar delamination typically is associated with substantial
14 2 Failure of Fiber-Reinforced Composites

inter fiber fracture

inter-ply
delamination
+σ +σ

Fig. 2.11 Schematic illustration of inter-ply delamination with corresponding electron micros-
copy images of a carbon/epoxy material in cross-sectional view (based on [8])

fiber direction fiber direction

Fig. 2.12 Influence of fiber–matrix adhesive strength on delamination behavior in scanning


electron microscopy images of a carbon/epoxy material in top view

Interlaminar splitting

Fig. 2.13 Schematic illustration of interlaminar delamination (splitting) in unidirectional fiber-


reinforced layers and corresponding fracture surface of a carbon/epoxy material in scanning
electron microscopy image

splitting of the fiber layers and rough fracture surfaces as seen from the scanning
electron microscopy image.
If such laminates are subject to uniaxial stresses, the individual plies are exposed
to fairly different stress levels. As long as there is no damage at the interface
between the plies, the assumption of strain coupling is fulfilled [5]. Since the elastic
properties and strength values of the individual plies vary significantly as function
of fiber orientation, characteristically different stress states are reached.
As seen in Fig. 2.14, tensile loading of a laminate often causes early interfiber
cracks in the off-axis oriented plies. Dependent on the respective external load
2.1 Classification of Failure Mechanisms 15

+σ +σ

Fig. 2.14 Schematic illustration of interfiber cracks in off-axis layers

+σ|| +σ||

Fig. 2.15 Schematic illustration of fiber bridging together with corresponding electron micros-
copy images of a carbon/epoxy material in top view (based on [8])

configuration and the stacking sequence, these interfiber cracks can occur either in
parallel or transverse to the load axis. Adjacent plies of different fiber orientation
typically act as crack stopper, since the presence of fibers oriented with the load axis
avoids failure of the matrix material due to local strain coupling. However, such
interfiber cracks can induce further damage as shown in Fig. 2.11. The existence of
such cracks typically causes strong stress concentration at the crack tip. This is a
typical site for initiation of inter-ply delamination propagating along the interface
between two plies.
Another process of mesoscopic failure, linked to the strength of the fiber–matrix
interface is the phenomenon of fiber bridging. If the fiber–matrix bonding is
sufficiently low, fiber pull-out can occur if the laminate is subject to tensile loading.
If the fiber filaments are strong enough, they can survive the pull-out to a certain
extend and span a mesoscopic crack propagating transverse to the fiber axis as
schematically shown in Fig. 2.15. In this case the crack progress is somewhat
constrained, since the intact fibers still bear the applied load. This failure mecha-
nism is also visible in the scanning electron microscopy image in Fig. 2.15.
Even at the mesoscale porosity can exist as a residue of the production process.
As seen from Fig. 2.16, these pores are located between adjacent plies and can have
a substantial impact on the mechanical properties of the laminate. Since the pores
do not carry any load, the laminate is subject to high local stress concentration and
may fail due to the onset of delamination, due to buckling, or due to the increased
stress level in the reduced cross section.
16 2 Failure of Fiber-Reinforced Composites

1mm

Fig. 2.16 Schematic illustration of macroporosity including corresponding computed tomogra-


phy image of a carbon/epoxy material in cross-sectional view

2.1.3 Macroscale

Due to these large number of failure mechanisms on the microscale and mesoscale,
the definition of macroscopic composite failure usually is subject to the type of
application and the type of loading. The scale associated with this term ranges from
the thickness of the laminate to the full dimension of the structure. Therefore, the
length scale is within the range of few millimeters to some meters.
Although interfiber fracture does not usually cause ultimate failure of the
composite on the macroscopic scale and may even be acceptable in some applica-
tions, it can still alter the mechanical properties, for example, the stiffness of the
structure. In addition, such failure can change the surface properties of the com-
posite or can raise secondary damage from aggressive media penetrating the
structure or an increased rate of moisture absorption. Also, microscopic fiber
breakage does not necessarily result in ultimate failure. In addition, the damage
progression induced by the stacking sequence often causes the global structure to
compensate the failure of one full layer.
An example of a macroscopic fracture surface is shown in Fig. 2.17 as originat-
ing from T-Pull testing. As seen from this example, all of these macroscopic
fracture surfaces are composed of different microscopic failure mechanisms,
whereas the final type of fracture is often dominated by the processes on the
mesoscale.
Another example to demonstrate the link between microscale, mesoscale, and
macroscale is the damage pattern observed in a woven fabric structure. As seen
from the computed tomography image in Fig. 2.18, the macroscopically visible
crack ranging through the laminate thickness develops many branches on the
mesoscale causing vast inter-ply delamination and microbuckling of individual
layers as seen by the cross-sectional images. On the length scale below the
individual plies (the microscale), all sorts of interfiber cracks and fiber breakage
can be observed.
One particularly relevant macroscopic failure mode for fiber-reinforced com-
posites is the impact damage. As seen schematically in Fig. 2.19, the damage state
2.1 Classification of Failure Mechanisms 17

Fig. 2.17 Typical example of a fracture surfaces. Photography of full fracture surface and detail
of fracture surface as electron microscopy images of a carbon/epoxy material in top view

Fig. 2.18 3D image of damaged woven fabric and virtual 2D cross sections of the volume
showing details of the damage state

Impact
1mm

1mm

glass fiber bundle

Fig. 2.19 Schematic drawing of impact damage and respective computed tomography scan of
impact zone (carbon/epoxy with embedded glass fiber bundles)
18 2 Failure of Fiber-Reinforced Composites

reached within a multiaxial laminate due to a low velocity impact is rather severe.
On the surface level, the impactor sometimes causes only small damage in the form
of local matrix damage or breakage of some surface plies. Therefore, such low
velocity impacts are often barely visible from the outside. However, on the inside of
the laminate, the large bending stresses during impact cause considerable
interlaminar shear stresses resulting in initiation and growth of interfiber failure
and delamination. This weakens the stiffness of the laminate structure considerably
and therefore comprises a major damage type that is well within the macroscopic
range.

2.2 Failure Theories for Fiber-Reinforced Composites

Within the last decades many research groups developed criteria to predict failure
in fiber-reinforced composites. As discussed in Sect. 2.1, there are various micro-
scopic failure mechanisms causing macroscopic failure. Moreover, their occurrence
and evolution is distinctly different for the type of macroscopic loading condition.
Hence, numerous theories are found in literature aiming to describe the complex
initiation and growth of damage in composites. In order to provide an overview of
those different theories, a first categorization can be made with respect to their
range of application. One can distinguish between theories focusing on description
of:
• Quasi-static failure.
• Quasi-static failure including growth of damage, damage mechanics, and
degradation.
• Long-term behavior (creep, stress rupture, and fatigue).
• High-velocity (high strain rates).
In the following, a brief review on the most important developments in these
areas is given. Focus rests on recent developments and the usage of experimentally
obtained values as input parameters for the theories.

2.2.1 Quasi-Static Failure

For the range of quasi-static loading conditions, many failure criteria have been
developed to predict the first onset of failure (first ply failure) as well as the
respective ultimate failure (last ply failure). In 1986, Reifsnider and Nahas
presented and compared already 30 different failure criteria for that purpose
[24]. Since that time many refinements have been proposed and essentially new
criteria have been developed.
In principle, stress and strain failure criteria are equivalent descriptions since the
two quantities are directly linked by the material’s constitutive law. But as pointed
2.2 Failure Theories for Fiber-Reinforced Composites 19

out by Puck, strain criteria are more difficult in their mathematical description,
since the superposition by transverse strains causes an inherent complexity of the
theoretical formulation [25]. Moreover, to obtain the respective input data a uni-
axial strain state is required to measure valid fracture strains. This is not feasible in
many experimental configurations.
In general, a distinction can be made between global criteria and individual
criteria differentiating between failure mechanisms. The global criteria apply
classical continuum mechanics theory to describe fiber-reinforced composites as
homogenous and anisotropic solid. In contrast, the differentiating criteria consider
the microstructure of the material and take into account the failure of fibers or
matrix material.
In 1971, Tsai and Wu published a generalized failure theory for anisotropic
solids known as the Tsai–Wu criteria [26]. Three years later, Wu demonstrated that
all global criteria published until that point could be directly derived from the Tsai–
Wu criteria [27]. Later Hashin [28] discussed limitations of those theories in terms
of the physical principles and pointed out the lack of consistency found for the
Tsai–Wu criteria under some stress states. Essentially this initiated the development
of further criteria which explicitly distinguish between fiber failure and interfiber
failure.
The type of failure criteria based on this failure hypothesis are called differen-
tiating criteria. They provide separate failure conditions for fiber failure and for
interfiber failure. In most failure criteria formulations simple maximum stress
criteria are used to describe fiber failure. But for interfiber failure it quickly turned
out that more sophisticated failure criteria are required to describe the fracture
behavior under different loading conditions. First attempts have been made by
Hashin [28], who adapted the fracture criterion of Mohr [29] formulated for failure
of brittle materials. The basic rule in Mohr’s concept states that fracture is caused
by the stress in the fracture plane. While this concept is mathematically straight-
forward to formulate in isotropic materials, it is much more difficult to formulate
for fiber-reinforced materials. Here, the detection of the correct fracture plane can
only be achieved by iterative algorithms in the general case. Therefore, formula-
tions are usually given for the simplified case as introduced by the presence of
transversely isotropic symmetry in unidirectional fiber-reinforced materials. This
yields the formulation of the failure criterion as quadratic–additive interaction
criteria based on four of the five stress invariants given by the transversal isotropy.
An alternative formulation for the interfiber failure criterion is given by Cuntze
[30], who uses five stress invariants. Based on this formulation, as function of
external loading, three independent interfiber failure modes for composites can
occur as shown in Fig. 2.20. However, the transition range between the different
failure modes is challenging, so probabilistic methods are applied in the formula-
tion by Cuntze.
As alternative approach, Mohr’s fracture criterion uses a solid physical basis, so
Puck started to continue Hashin’s work to determine the correct fracture plane
throughout a series of publications [25, 31–36]. He established a parabolic inter-
action criterion for interfiber failure based on Mohr’s fracture criterion and three
20 2 Failure of Fiber-Reinforced Composites

+σ⊥ –σ⊥ –σ⊥

τ||⊥ τ||⊥ τ||⊥

mode A mode B mode C

Fig. 2.20 Fracture modes in plane stress conditions according to Puck

different interfiber failure modes for plane stress conditions as shown in


Fig. 2.20 [25].
Fracture mode A occurs due to a tensile stress component σ ⊥ superimposed by a
shear stress component τk⊥ . The resulting crack growth in a unidirectional layer is
of a mixed crack opening mode (mode I) and shear mode (mode II) caused by the
superimposed stress conditions. Fracture mode B is caused by a compressive stress
component σ ⊥ superimposed by a shear stress component τk⊥ . Here, no crack
opening mode occurs, so crack growth is mostly due to the shear mode (mode II).
Finally, fracture mode C is caused by a compressive stress component σ ⊥
superimposed by a shear stress component τk⊥ . In contrast to mode B, σ ⊥ is the
dominating stress component. As consequence crack growth occurs angled to the
load axis of the compressive stress component as seen in Fig. 2.20. The fracture
angle is a result of the additional shear stress component acting in the inclined
fracture plane.
However, in addition to the four basic material strengths, the definition of the
slopes linking the tensile and compressive stress states requires the measurement of
the so-called inclination parameters.
One noteworthy effort to compare the predictive capabilities of different failure
criteria was carried out by Hinton, Soden, and Kaddour throughout a series of
exercises. These became known as “World-Wide Failure Exercise I” [37–40] and
“World-Wide Failure Exercise II” [41–43] due to the multitude of participants from
countries all over the world. The procedure chosen by the organizers comprises two
stages. During part A of the exercise, the participants conduct blind predictions
given a set of material properties and loading conditions. At the beginning of part B,
the experimental data is revealed to the participants and they are allowed to adjust
their predictions accordingly. Afterwards, comparison is made between the predic-
tion results of the different theories applied by the participants.
Within the “World-Wide failure exercise I” the focus was on the response of
continuous fiber laminates subject to in-plane biaxial loading situations, in the
absence of stress concentrations. As a short summary of the World-Wide failure
exercise I, the failure criteria using differentiating criteria performed reasonably
well as compared to the global failure criteria. In particular, the failure criteria
2.2 Failure Theories for Fiber-Reinforced Composites 21

given by Puck [35, 44], Zinoviev [45, 46], and Cuntze [47] were found to yield
highest predictive capabilities [48].
One of the major gaps of the theories present at that time was the lack of
predictive capabilities under 3D (i.e., triaxial) states of stress. Such stress states
are typically found in thick composite components after impact or as a result of
stress concentrators. Therefore, in the “World-Wide Failure exercise II” 3D states
of stress including triaxial failure and associated theories were also considered.
As the outcome of this second exercise was more diverse than for the first
exercise, an overall recommendation for a particular theory seems premature.
According to [41], the overall impression of the 3D failure theories provided by
Pinho [49, 50], Carrere [51, 52], Puck [53, 54], and Cuntze [30] were ranked best.
However, many discussions have accompanied the “World-Wide Failure Exercise
II” regarding the reliability of the experimental data provided. In particular, for
those test cases dealing with 3D stress states many involved researchers commented
on the experimental issues associated with the measurements [30, 41, 49,
54–57]. Beyond those failure theories compared in the “World-Wide Failure
exercise” other groups also presented derivatives of existing theories or proposed
essentially new theories [58, 59]. Such new approaches are particularly relevant for
the case of textile composites. In these materials, the undulations of the fiber yarns
and the pronounced mesostructure cause further challenges to prediction of damage
progression and ultimate failure.

2.2.2 Quasi-Static Failure Including Growth of Damage,


Damage Mechanics, and Degradation

As mentioned in Sect. 2.1, there are various microscopic failure mechanisms, which
finally cause ultimate failure of a macroscopic structure. The consideration of their
contribution to degradation of macroscopic properties and to the initiation of
subsequent damage is a challenge, which goes beyond the prediction of global
stress or strain limits. Several failure theories presented in Sect. 2.1.1 already take
into account the degradation of laminate properties due to occurrence of damage
inside the plies.
However, the description of the macroscopic failure in terms of accumulation of
microscopic damage should theoretically be possible. More specifically, the micro-
scopic scale is the correct length scale to account for the microstructure introduced
in fiber-reinforced composites [5, 60–64]. However, the derivation of accurate
microscopic mechanical properties is experimentally challenging. In particular,
the measurement of strength values for small microscopic volumes is very difficult
and predictions based on microscopic measurements extrapolating to the macro-
scopic scale are also influenced by the statistical distribution of flaws [65]. More-
over, using a micromechanical approach in practice typically comes with high
computational intensity. Therefore, analytical methods using homogenized
22 2 Failure of Fiber-Reinforced Composites

properties, defect-free unidirectional plies, orthotropic symmetry, and macroscopic


stress–strain relations are typically preferred. In this context, the concept known as
classical laminate theory is usually applied.
This concept uses a layer-wise failure of plies and applies respective degradation
models to account for damage accumulated in a layer. For this approach, the
question arises, if macroscopically measured strength values of unidirectional
layers are applicable to the plies inside a laminate. As reported by various authors
[66–70], crack propagation inside such internal layers depends on the thickness of
the ply as well as on the stiffness of the neighboring layers. For sufficiently thin
layers, strain coupling to the adjacent layers can inhibit the opening of existing
cracks. According to Parvizi [68] this effect is most pronounced for thin layers, but
vanishes for ply thicknesses above 0.4 mm.
To describe the propagation of existing cracks inside a solid usually concepts of
fracture mechanics are applied. These approaches could in principle also be used to
describe the damage progression in fiber-reinforced composites. The formulations
of the stress-intensity concept, the critical strain energy release or the J-Integral are
well-established concepts for pure materials [71–75]. Since flaws are also present in
fiber-reinforced materials, a direct transfer of the fracture mechanics concepts
seems straightforward. However, the computational intensity and the lack of
consistent micromechanical parameters have prevented the application of
micromechanical fracture mechanics for a long time. In recent years, some
approaches used fracture mechanics concept as hybrid approaches [50, 58] or in
their pure form [76] to describe damage progression within laminates. Similar to the
strength based failure criteria, there is still room for improvement of the experi-
mental methods to determine microscopic fracture toughness values for various
load conditions [15, 77, 78].
As a consequence of the results of the “World-Wide Failure exercise II” more
focus is then given to the prediction of damage. Within the “World-Wide Failure
exercise III,” the prediction of matrix crack initiation, delamination initiation due to
transverse cracks, deformation, and final fracture is investigated. While this exer-
cise is still going on, first results of the blind predictions indicate a clear diversity of
predictions regarding the effects of ply thickness, the stacking sequence and
interaction of existing cracks with neighboring layers, in particular to induce
delamination [79]. However, a larger tendency is seen to use finite element model-
ing [80–86] and micromechanical approaches [80, 81, 85, 87–89] to obtain failure
predictions. A specific challenge for the existing theories is the forward prediction
of crack density and crack saturation of interfiber cracks. In fact, many theories
require the crack density of some cases as input parameter, rather than allowing a
physically based forward prediction. Therefore, experimental methods are required
which allow suitable tracking of this crack evolution as function of loading for the
reference cases, but as well to benchmark failure theory predictions for arbitrary
laminate configurations and load conditions.
Similarly, many efforts have been commenced in the last years in order to
improve the prediction of damage initiation and damage progression for textile
composites [90–95]. For these materials, even the prediction of first failure onset is
2.2 Failure Theories for Fiber-Reinforced Composites 23

comparatively complex because of the explicit microstructure and mesostructure of


textile composites. The textile architecture of the fiber reinforcement (e.g., woven
fabrics, noncrimp fabrics, and braids) cause a complex multiaxial residual stress
state after curing which needs to be accounted for. Since this involves calculations
on various length scales, classical homogenization approaches can effectively be
combined with finite element based routines to account for such multiscale prob-
lems [90, 93–96].
In the following, a brief review is made for three particular failure modes
typically addressed in current failure theories for laminates composed of unidirec-
tional plies. The failure criteria given by Puck, Sch€urmann, and Deuschle [5, 53, 54,
97] is taken as an example1 to discuss current trends in the development of failure
theories and to discuss the relation to recent experimental findings on damage
evolution in fiber-reinforced composites.
For the consideration of fiber failure due to tensile or compressive loading σ 1,
usually simple maximum stress criteria are applied. In recent years, 3D failure
criteria evolved, which partially consider the influence of the stress components
transverse to the load axis. The impact of the off-axis components σ 2 and σ 3 on fiber
failure has already been noted by Edge [98] and Hart-Smith [99], but was neglected
in the original work by Puck [25]. One example of an improved theoretical
formulation was proposed by Puck, Sch€urmann, and Deuschle [5, 53, 54,
97]. The simple maximum stress criterion using the strength of the unidirectional
layer under tensile load Rþ 
k and under compressive load Rk is given as:

jσ 1 j
¼1 for σ1  0 ð2:1aÞ

k

jσ 1 j
¼1 for σ1 < 0 ð2:1bÞ
R
k

The stress exposure in a single fiber filament is given by the relation applying the
empirical magnification factor m (e.g., 1.3 for GFRP and 1.1 for CFRP) to respect
the influence of the off-axis components σ 2 and σ 3.

σ f k ¼ Ef k  ε1 þ νf ⊥k  m  ðσ 2 þ σ 3 Þ ð2:2Þ

Due to strain coupling, the laminate strain equals the filament strain, therefore:
σ 1 ν⊥k ν⊥k
ε1 ¼  σ2  σ3 ð2:3Þ
Ek Ek Ek

Combining (2.2) and (2.3), this yields the refined form of the fiber failure criterion:

1
There is still much discussion regarding the general applicability of such failure criteria, so the
reader should always be aware of the potential drawbacks of a particular theory.
24 2 Failure of Fiber-Reinforced Composites

     
σ  
 1 Ek σ 2 þ σ 3 
    ν⊥k  νf ⊥k m  ¼1 ð2:4Þ
 Rk  Ef k  Rk 

However, in the meaning of strength-based failure criteria, fiber failure is usually


understood as failure of fiber bundles or failure of full layers. In the context of first
failure analysis by secondary methods, it is worth to have a closer look on the
failure evolution causing macroscopic failure due to loading of a unidirectional
material parallel to the fiber axis. It is well known that fiber filament strength is
subject to a broad strength distribution. Therefore, even in a defect-free laminate it
is likely that failure of some fiber filaments precedes failure of the full layer. As
demonstrated in recent work by Scott et al. [100, 101] and Bunsell et al. [102, 103],
there is indeed an evolution of fiber filament failure preceding the ultimate failure of
a layer. According to their findings, failure of a single filament is likely to occur
after a certain local stress exceeds the strength of a fiber filament, that is, the failure
position is of statistical nature. The surrounding filaments are able to compensate
the breakage of the individual, so one fiber filament failure is not of large relevance
to the durability of the structure. The situation changes if other fiber filaments fail in
close vicinity. This increases the stress exposure of the surrounding filaments.
As quantified by in situ synchrotron tomography, a critical health state of fiber-
reinforced materials is reached if the number N of localized fiber failures reaches a
critical value. Such a set of N localized fiber failures was entitled N-plets by Scott
et al. [101]. A scheme of the evolution of single fiber filament failures starting at a
1-plet and growing to a 3-plet and 6-plet configuration is shown in a cross-sectional
view in Fig. 2.21. However, in reality the cumulative N-plet count refers to a 3D local
volume rather than just a local 2D cross section of the material. Remarkably enough
N was found to typically stay < 20 before ultimate failure of the structure occurs.
As discussed in Sect. 2.2.1, the failure of fiber-reinforced composites due to
interfiber fracture is much more challenging to capture in a failure theory. Based on
the results of the World-Wide Failure exercise I and II Puck’s and Cuntze’s failure,

1-plet 3-plet 6-plet

Fig. 2.21 Example of local accumulation of fiber failures starting at 1-plet (a) and increasing to
3-plet (b), and 6-plet (c) configuration
2.2 Failure Theories for Fiber-Reinforced Composites 25

τnt
3 3 3
τnψ σn
ψ

θ
τn1

θ
2 2 2
1 1 1

Fig. 2.22 Definition of fracture plane at angle θ and definition of stress components acting at
fracture plane

criteria were ranked among the best available [41, 48, 104]. In the following, the 3D
failure criterion as given in [5, 53, 54, 97] is used as an example.
As shown in Fig. 2.22 and discussed in Sect. 2.2.1, every cross-sectional plane
perpendicular to the fiber axis is a potential fracture plane. Therefore, only three
stress components (and their superposition) are causing interfiber failure. These are
the components σ n, τnt and τn1 with reference to the normal (n), tangential (t), and
fiber parallel (1) axis of the fracture plane as shown in Fig. 2.22.
For such an arbitrary fracture plane orthogonal to σ n, the superposition of the
shear stress components τnt and τn1 can be replaced by an effective shear stress
component τnψ using:
τn1
¼ tan ψ ð2:5Þ
τnt
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
τnψ ¼ τ2n1 þ τ2nt ð2:6Þ

Therefore, interfiber failure is governed by the stress σ n and τnψ acting on the
fracture plane. The failure criterion can thus be formulated using the respective
unidirectional material strength values Rþ A
⊥ , R⊥⊥ , and R⊥k and the inclination
parameters pþ 
⊥ψ and p⊥ψ (cf. Fig. 2.23) defined as:

dτnψ 
¼ pþ ð2:7aÞ
dσ n σ n ¼0 ⊥ψ


dτnψ 
¼ p ð2:7bÞ
dσ n σ n ¼0 ⊥ψ

In the tensile load regime (σ n > 0), one can define:


!2 !
þ
τnψ pþ
⊥ψ σ n pþ
⊥ψ R⊥ σ 2n
A
þ2 A
þ 12 A
 2 ¼ 1 ð2:8Þ
R⊥ψ R⊥ψ R⊥ψ Rþ⊥

For compressive load (σ n < 0)


26 2 Failure of Fiber-Reinforced Composites

inclination p– τnψ Rσ

elliptical
inclination p+

parabolic

σn– 0 σn+

Fig. 2.23 Definition of the inclination parameters

!2
τnψ p
⊥ψ σ n
A
þ2 A
¼1 ð2:9Þ
R⊥ψ R⊥ψ

And for pure shear (σ n ¼ 0)


!2  2 !2
τnψ τnt τn1
A
¼ A
þ A
ð2:10Þ
R⊥ψ R⊥⊥ R⊥k

The combination of (2.8), (2.9), and (2.10) then finally yields Puck’s formulation
of the interfiber failure criterion for an arbitrary fracture plane.
For σ n  0:
v"ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
! #2  ffi
u 2  2
u p þ
τnt τn1 pþ
t 1

⊥ψ
 σ þ þ þ
⊥ψ
 σn ¼ 1 ð2:11aÞ
Rþ A n A A
⊥ R ⊥ψ R ⊥⊥ R ⊥k R ⊥ψ

For σ n < 0:
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
!2
u
u τ 2  τ 2 p p
t nt
þ
n1
þ
⊥ψ
 σ þ
⊥ψ
 σn ¼ 1 ð2:11bÞ
A A n A
R⊥⊥ R⊥k R⊥ψ R⊥ψ

Up to this point only failure initiation is captured by the failure criterion. In order to
predict damage progression the theory has to take into account the growth of such
cracks, their impact on triggering of new damage and their impact on macroscopic
stress and strain states. In the sense of Puck’s failure theory, fiber failure comprises
the ultimate limit of the laminate. Therefore, description of damage progression
spans the range between the first occurrence of interfiber failure and final occur-
rence of fiber failure.
There are two ways to represent interfiber cracks in failure theories. Either those
are represented as explicit description in a micromechanical or multiscale approach
2.2 Failure Theories for Fiber-Reinforced Composites 27


force flux

90° (reduced stiffness)


+σ +σ

discrete cracks smeared representation

Fig. 2.24 Representation of interfiber cracks in discrete and smeared form

or the cracks are taken into account as smeared representation in a continuum


damage mechanics approach. A comparison of both concepts is shown in Fig. 2.24.
The approach taken by Deuschle and Puck [53, 54, 97] is the smeared represen-
tation allowing a layer-wise degradation of properties. At the occurrence of
interfiber failure spanning the full thickness of internal layers, this usually does
not disintegrate the whole laminate. Instead a sequential increase of crack density
occurs. Accordingly, the laminate properties (i.e., stiffness) degrade. For the exam-
ple shown in Fig. 2.24, the occurrence of interfiber cracks increases the amount of
load to be carried by the neighboring layers. Therefore, such a damaged laminate
will deform more under the same load as the same laminate configuration without
any damage. Hence, the degradation of the laminate properties due to such
interfiber cracks is represented by a reduction of stiffness using the degradation
parameter ~η in the interval [1; 0]

reduced stiffness ¼ ~η  original stiffness ð2:12Þ

For degradation the parameter ~η goes down from 1 (no damage) to 0 if saturated
crack density is reached. Particular rules for more complex degradation rules for the
3D case are discussed by Deuschle [97]. Such smeared crack representations
facilitate fast computation of the degradation process, but obviously introduce
assumptions regarding their impact on the local interaction with neighboring layers.
Recently, Trappe et al. applied X-ray refraction techniques to measure
the occurrence of interfiber cracks in glass-fiber-reinforced composites
[105–108]. According to their findings, the inhomogeneous microstructure of
fiber-reinforced composites triggers a statistical distribution of the occurrence of
interfiber cracks. Therefore, occurrence of such cracks cannot be predicted by
layer-wise failure analysis. As demonstrated by Scott et al., additional aspects
come into play if the existence of voids in the laminate is considered
[109]. Using high-resolution synchrotron tomography, the interfiber cracks were
shown to interact to a substantial degree with such defects, since they tend to
intersect with voids and grow directed along such weak spots. Those aspects are
28 2 Failure of Fiber-Reinforced Composites

hardly reflected by layer-wise degradation analysis, but should be considered if the


void content is of significant contribution.
It is noted that there is some ambiguity between the usage of the term interfiber
fracture as above and the term delamination. However, delamination is different to
other interfiber failure modes due to its orientation relative to the thickness direc-
tion of the laminate. As explained in Sect. 2.1, the laminate can fail by separation of
two layers due to interfiber crack growth in parallel to the in-plane direction. Such
delamination is mostly caused by through-thickness tension components or shear
stress components. Usually, the initiation and growth of delamination is a problem
well formulated by fracture mechanics approaches using mode I, mode II, mode III,
or superimposed fracture modes. In experiments, it is frequently observed that
delamination often initiates at the interface between adjacent plies at the position
of the crack tip of an interfiber crack. This is due to the stress concentration at the
crack tip, which has been accounted for in various theories (e.g., [69]).

2.2.3 Long-Term Behavior

Much progress has been made to develop failure criteria applicable for quasi-static
loading conditions. However, for investigation of the long-term behavior of com-
posite structures fewer investigations have been conducted so far. In the following,
a brief review is made comprising the particular challenges of failure prediction in
the fields of creep, stress rupture, and fatigue, and the associated experimental
approaches to assist in the understanding of failure for these load cases. Following
the classification used by [110], strain rates below 106 s1 are considered to be
creep loading and strain rates at or below 103 s1 are considered to represent
quasi-static deformations. As an extension beyond the quasi-static range, strain
rates above 102 s1 are classified as high strain rates, strain rates above 104 s1 are
called very high strain rates, and strain rates above 106 s1 are referred to as ultra-
high strain rates.

2.2.3.1 Creep and Stress Rupture

For fiber-reinforced polymer materials, creep loading is naturally associated with


severe drawbacks based on the general experience of pure polymers. However,
creep and relaxation effects are not always considered to be of negative impact for
the composite. As discussed by Sch€urmann [5], the stress relaxation within the
matrix material can cause significant reduction of residual thermal stress and a
redistribution of stress into the fiber reinforcements. Still, the structure is subject to
permanent deformation, which might not be acceptable in some applications. For
pure polymers, the time–temperature superposition principle is usually applied to
allow extrapolation of long-term stability based on short-term measurements.
Thermosets are the dominating type of polymer used as matrix material, but there
2.2 Failure Theories for Fiber-Reinforced Composites 29

is a growing interest in thermoplastic matrix materials. Although thermoplastic


matrix materials are not essentially new in applications and despite of the higher
creep rates faced in thermoplastic polymers, less investigations have been
conducted. A comparison between creep rates of a fiber-reinforced epoxy resin
and fiber-reinforced Polyetheretherketone has been performed by [111]. As further
demonstrated for glass-fiber-reinforced Polyamide [112], carbon-fiber-reinforced
polyimide [113], and glass-fiber-reinforced polycarbonate [114], the application of
the Arrhenius type time–temperature superposition principle is applicable in these
cases. In particular, if influences of physical and thermal ageing are taken into
account explicitly in analytical equations, stable predictions of creep strength as
function of crystallinity and fiber volume fraction seem feasible [112].
For the prediction of stress rupture, there is no fundamental difference to the
considerations presented in Sects. 2.2.1 and 2.2.2. In principle, the same strength-
based formulations or fracture mechanics approaches can be adopted. However,
one has to explicitly consider the viscoelastic nature of the polymer matrix material
in the case of creep loading and the evolution of damage in the static load case.
Although this is straightforward in theory, the according mathematical formulation
is not trivial in most cases. Only few attempts have been made to explicitly account
for viscoelastic material behavior in present failure theories. Guedes reviewed
different formulations using continuum damage mechanics and cumulative damage
formulations [115, 116]. For the case of stress-rupture behavior, comparison was
made between maximum work stress criterion (Reiner–Weissenberg criterion [117]
and modifications thereof), maximum strain criterion, and linear cumulative dam-
age formulation of Miner [118] as well as Christensen’s cumulative damage
formalism [119]. For the purpose of lifetime prediction under constant stress, it
was demonstrated that the different formulations discussed produce relatively
similar predictions [115]. However, experimental evidence indicates the limited
predictability of stress rupture in real composite structures such as pressure vessels
[120]. Failure of such safety relevant structures are very critical and the predictive
capability of existent theories not yet sufficient to decrease safety margins. It is well
known that glass fibers, carbon fibers, and Kevlar fibers all suffer from long-term
static loads. Even for the best performant carbon fibers, the maximum allowable
pressure needs to be reduced to less than 60 % of the static burst pressure when
designed for more than 100,000 h operation and an intended probability of stress-
rupture failure less than 0.001 [121]. Therefore, accelerated testing using the time–
temperature superposition discussed by [122–124] needs further validation to
assess its relevance for stress-rupture predictions.
From the viewpoint of secondary methods in conjunction with creep and stress-
rupture experiments, the biggest improvement might be further shortening of the
test duration. At present, the overall duration of a test campaign to obtain creep
properties for only one laminate configuration can still take more than 10,000 h to
extrapolate the stability for two decades [115]. Therefore, in situ methods being
able to detect the onset of secondary or tertiary stages of creep as well as the ability
to differentiate between different mechanisms during stress-rupture testing are
much appreciated.
30 2 Failure of Fiber-Reinforced Composites

2.2.3.2 Fatigue

Another load scenario fiber-reinforced materials are facing in application is fatigue


loading. From the testing point of view, the term fatigue is usually associated with
the test procedure first proposed by W€ohler [125]. Here, a specimen is cyclically
loaded with constant amplitude levels at a constant frequency. In fatigue testing, the
load configurations are usually determined by the R-ratio, which is the minimum
load level divided by the maximum load level of the specimen. One can distinguish
between:
• Tensile–tensile fatigue (upper and lower amplitude  0, R-Ratio  0).
• Tensile–compression fatigue (upper amplitude > 0, lower amplitude < 0,
R-Ratio  0).
• Compression–compression fatigue (upper and lower amplitude  0,
R-Ratio  0).
Other specific load cases are known as unidirectional loading (upper
amplitude > 0, lower amplitude ¼ 0, R-Ratio ¼ 0) and fully reversed loading (mag-
nitude of upper amplitude equal to magnitude of lower amplitude, R-Ratio ¼ 1).
In all cases, the testing is continued until a significant stiffness reduction or
ultimate failure occurs. The test campaign is carried out for different amplitude
levels (S) between the static amplitude limit and sufficiently small amplitude
values. The number of cycles to failure (N) is plotted versus the amplitude level.
This S–N curve is used to establish the long-term limit of materials, that is, to define
the amplitude value below no failure of the material will occur for an infinite
number of cycles.
In isotropic materials, the fatigue life is often estimated from the original version
of the linear cumulative damage law (LCD), referred to as Miner’s rule [118] or
improvements thereof [126, 127]. But even for isotropic materials, it has been
questioned, if simple LCD approaches will yield realistic lifetime predictions in
fatigue loading [126].
Some fiber-reinforced composites, such as CFRP, are associated with superior
lifetime in fatigue. However, this is not essentially true as far as first damage
initiation and damage progression is concerned [128]. One of the key problems
with composites is the sequence of amplitudes faced in the loading–unloading
cycles [115, 129–131]. Since the real “load spectrum” differs from application to
application, it is hard to standardize. But for fiber-reinforced composites, the load
spectrum has significant impact on the number of load cycles to failure. In a fiber-
reinforced material, damage can initiate at comparatively low load levels and
continuously grow until final failure is reached. The type of damage growth may
change as function of the applied load and may also interact with other damage
existent in the material (e.g., small matrix cracks initiating large growth of delam-
ination). Therefore, preliminary initiation of damage at higher load levels can
extend further in subsequent cycles with low load levels. This can cause growth
of damage at load levels, which usually would not result in damage initiation or
growth by themselves as been evident from a series of bucket truck failures [132].
2.2 Failure Theories for Fiber-Reinforced Composites 31

Therefore, different failure theories have been proposed to explicitly account for
the load spectrum effect. A good review on the major developments is given by
Degrieck [128] and a more recent comparison of selected failure theory predictions
is presented by Guedes [115]. A basic distinction in terms of failure criteria can be
made according to Sendeckyj between macroscopic strength criteria, criteria based
on residual strength, those based on residual stiffness, and failure criteria taking into
account an explicit formulation of an actual failure mechanism [133].
One class of models comprises the so-called fatigue life models, which are
essentially based on the S–N curve behavior as discussed above. These models
usually neglect the interaction of different failure mechanisms such as interfiber
failure or fiber breakage. Examples for these failure models are given in
[122, 134–141]. As a result, these types of models yield the fatigue life of the
composite studied based on the extrapolation of the measured S–N curves. The
second class of models take into account the degradation of composite properties
based on phenomenological descriptions. Some of these degradation models are
formulated for the residual stiffness [129, 142–148], others are using the residual
strength based on deterministic descriptions or based on statistical strength distri-
butions [149–156]. The last type of models uses an explicit formulation of the
damage progression correlating one or more damage variables to the damage
extent. This quantitatively accounts for the damage progression in a physically
sound way. Distinction can be made between models able to predict the damage
growth [157–162] and models relating the damage growth to property degradation
[163–172].
A common challenge of the models provided until today is the extrapolation of
data acquired in model specimens to real structures. Usually, input data for failure
theories is collected using a specific stacking sequence and uniaxial loading in one
test configuration. The corresponding theories may be able to describe these
situations quite accurately, but for the general situation of multiaxial loading
conditions in stacking sequences changing as function of position in a real structure,
their predictive capabilities are still questionable.
For the experimental approaches, comprehensive reviews in application to
composites have been provided in [133, 173]. One of the key challenges yet to
overcome is the ability of test methods to accompany long-term testing with
nondestructive assessment of failure evolution. At least for the models using
explicit formulations of damage growth such input parameters are essential to
further advance the reliability of the model and to improve their predictive capa-
bilities for generalized load situations. Another factor is similar to the main
challenge in creep experiments, which is the shortening of the test duration. Fatigue
testing with low amplitudes can take several weeks. In order to extrapolate the
lifetime such testing is still required and since composites lack from the applica-
bility of the Miner’s rule, there is currently no suitable approach to circumvent
realistic load spectra. Alternative termination criteria other than residual stiffness
values or ultimate failure are highly appreciated. In particular, for testing of
structures accompanying in situ measurements are suitable tools to improve the
cost effectiveness of the test campaigns.
32 2 Failure of Fiber-Reinforced Composites

2.2.4 High Velocity

The increasing application of fiber-reinforced materials in crash-relevant structures


requires corresponding experimental methods to characterize these materials under
high strain rates. The failure modes found for high strain rates cannot be expected to
be the same as for the strain rates used in quasi-static conditions. During fast
damage accumulation, no relaxation effects and, hence, stress redistribution can
occur. Moreover, elastic properties and failure behavior of the matrix material can
change significantly due to transition from ductile to brittle failure modes.
The failure modes considered in high strain rate experiments are typically of
ultimate nature, that is, the experiments result in immediate failure of the laminate
rather than partial or sequential damage accumulation. Therefore, description of
material failure does not consider evolution of damage to the extent as the criteria
used for creep, fatigue, or quasi-static load situations. As a function of strain rates
one can distinguish between experimental configurations considered being in the
creep domain < 106 s1, the quasi-static deformation range < 103 s1, and those
reached by servohydraulic test rigs < 100 s1. Above this limit specialized equip-
ment, such as Kolsky bars (also known as Split–Hopkinson bars) are used to reach
strain rates > 106 s1 [174]. For even higher strain rates, impactors driven by
gravity or projectiles shot by gas cannons are used to test materials subject to
high-velocity impact. A good review on the recent developments of the experimen-
tal techniques in high-velocity testing is given by Ramesh [110] and a critical
review of the Kolsky bars is given by Gama [175].
From the view point of failure theories, recently Kawai compared the predictive
capabilities of classical failure theories of Tsai–Wu, Tsai–Hill, and Hashin–Rotem
with respect to the off-axis strength values of unidirectional laminates under high
strain rate tension and compression [176]. A detailed discussion of the failure
modes and fracture angles was given and some key modifications to the failure
criteria were proposed.
For fiber-reinforced composites various experimental efforts have been made to
characterize material properties under tensile, compressive, and shear loading
[176–183]. Until recent times, the focus of the experimental work was to develop
suitable test rigs, assuring the anticipated load situation at the constant strain rates.
Secondary methods evolved with the usage of high-speed cameras to directly
observe the fracture of fiber-reinforced materials [184]. Improvements have been
made by [181–183, 185] introducing digital image correlation systems to measure
the full strain field during the test. This validated fundamental assumptions regarding
the principle of operation of Kolsky bars. Beyond the point of image acquisition and
processing, secondary methods are not used extensively. Beyond a certain strain rate,
acoustic methods are generally problematic, since the load is introduced by an
acoustic wave itself. Secondary methods like acoustic emission, guided wave testing,
vibrometry, or ultrasonic tomography are hence not of big relevance for the exper-
iments used in the field of materials testing at strain rates up to 106 s1. However,
applications exist, where impacts themselves are monitored by acoustic emission
sensors to localize their position as well as to deduce the impact energy [186–189].
2.3 Challenges in Mechanical Testing of Fiber-Reinforced Materials 33

2.3 Challenges in Mechanical Testing of Fiber-Reinforced


Materials

In general, mechanical testing of fiber-reinforced materials is used to obtain mate-


rial properties that are used for quality control or for designing composite struc-
tures. While quality control often solely requires comparative measurements
carried out under reproducible conditions, the design and engineering requires the
measurement of true material properties. Therefore, material testing has to assure
that the obtained data is in accordance with the boundary conditions and assump-
tions of the respective theory. For the linear elastic range of the material, the major
requirement is the generation of homogeneous stress–strain fields within the test
specimen. Moreover, suitable test specimen geometries with defined superimposed
loading conditions are still challenging. For the measurement of the material
properties after first onset of damage inside the material several challenges arise.
The first challenge is the detection of first damage itself. Subsequent to this load
level, the second challenge has its focus on the usage of appropriate specimen
geometries to assure the intended type of failure. Many times, the specimens fail in
an unexpected fashion due to failure modes different from the intended mode.
Almost all of these challenges stem from the hierarchical structure of fiber-
reinforced materials. Compared to “pure” engineering materials as metals, poly-
mers, or ceramics, fiber-reinforced materials intrinsically have at least two addi-
tional characteristic length scales relevant for failure progression. The first is the
length scale of their fiber reinforcements and the second is the length scale of the
plies composed into a laminate. For textile composites, an additional length scale
defined by the characteristic dimensions of the elementary cell of the textile
reinforcement also comes into play.
Therefore, during testing one is always subject to two fundamental questions:
• Which is the appropriate scale to obtain material parameters?
• What are valid “material” parameters?
In contrast to many classical engineering materials, which are mostly homoge-
nous in their properties on several length scales, the fiber reinforcements in com-
posites induce heterogeneity, inhomogeneity, or both. The relevance of the fiber
distribution is linked to the observation scale. While on the macroscopic scale, one
can hardly resolve the individual fiber filaments, their distribution and orientation is
quite obvious on the length scale of micrometers. Therefore, the measurable
properties can be completely different for measurements on the microscopic
scale. As consequence, one has to provide different material properties for failure
theories driven by microscopic volume elements or failure theories based on a
macroscopic homogenized continuum representation.
Also, the fiber reinforcements induce a certain degree of anisotropy in the
material properties. Compared to isotropic materials this inevitably causes a larger
amount of required testing methods. Therefore, specific test geometries are
required, which allow for testing under load conditions typically not applied for
classical engineering materials.
34 2 Failure of Fiber-Reinforced Composites

While many of these aspects are sufficiently taken into account for some test
procedures, for other cases there still is an active discussion in the community
regarding the reliability and comparability of different test procedures. As
discussed by Bleier [190], Cuntze [30], Pinho [49, 84], Deuschle [97], and Basan
[191], there is a strong need to provide true material properties as input for modern
failure criteria of fiber-reinforced composites. Juhasz points out that failure predic-
tion generally should be possible without prior knowledge of the exact load
configuration and according experimental test results of a specific structure [59].
Therefore, in order to verify different failure criteria, there is a strong need for
testing methods assuring the validity of superimposed loading conditions and for
methods allowing tracking the evolution of failure until the point of rupture.
In the following, three typical challenges for presently used test standards are
presented. A particular focus is given to the applicability of secondary methods to
improve the understanding of the test results and their relevance to the work
presented herein.

2.3.1 Detection of First Failure Onsets

As discussed in Sect. 2.2.1 various failure theories aim to predict the first occur-
rence of a particular failure mode within fiber-reinforced materials. But for many
laminate configurations, there is no possibility to quantify the onset of interfiber
failure in the individual plies from load–displacement curves. Moreover, the detec-
tion of the onset of fiber failure is also not feasible using stress–strain curve
information unless this is considered to be identical to ultimate failure. Apart
from the failure theories, there are several other standardized methods, which rely
on the information obtained from the first onset of damage.
One test setup, which requires the determination of the initiation of interlaminar
delamination, is the short-beam-shear test according to the standards ISO 14130,
ASTM D 2344, DIN EN 2377, and DIN EN 2563. But sometimes the occurrence of
delamination onset is not identical to the maximum load or the first significant load
drop as demonstrated in Chap. 3. However, the latter is a requirement to obtain the
apparent interlaminar shear strength. Despite of the usage of this method mostly in
quality control, this already indicates the ambiguous level of confidence associated
with the proposed data reduction schemes.
Another experimental setup, which monitors crack propagation, is the configu-
ration to deduce mode I interlaminar fracture toughness according to ISO 15024,
ASTM D 5528, prEN 6033, or AITM 1.0005. Besides the valuable data reduction
routines to deduce the fracture toughness included with the ASTM D 5528 standard
several methods are noted to quantify the level of the initial fracture toughness. This
is owed to the fact that none of the onset criteria included within the standard is
applicable for all types of materials. Therefore, secondary methods can be valuable
tools to assist in the interpretation of such methods and to obtain true material
properties.
2.3 Challenges in Mechanical Testing of Fiber-Reinforced Materials 35

Similarly the standards ASTM D7905, prEN 6034, and AITM 1.0006 used to
obtain mode II fracture toughness by the End-Notched-Flexure (ENF) configuration
attempt to derive the onset of interlaminar crack growth using signatures of force-
displacement curves. Here, the type of crack insert, the measurement of the crack
length, and the load fixture itself can have significant impact on the fracture
toughness value [192]. But the critical fracture toughness value is also affected
by the load value picked for the first onset of delamination. Here, again the load
maximum is not necessarily the correct value to choose for the onset as demon-
strated in Chap. 4. Another method aiming to deduce valid mode II fracture
toughness values is the Transverse Crack Tension (TCT) specimen. Although the
TCT tests are currently not covered by any standard, the key challenge associated
with this configuration stems again from the validity of the detected onset of
delamination.

2.3.2 Tracking Failure Evolution

From a general point of view, there are two good reasons to detect the failure
evolution in a test specimen made from fiber-reinforced material. The first reason is
to compare the failure evolution predicted by a failure theory to an actual measure-
ment. The second reason is the interpretation of data from stress–strain curves after
onset of initial failure.
The comparison of predicted and measured failure evolution is most interesting
to validate attempts of degradation analysis or to detect subsequent onsets of
particular failure mechanisms in angle-ply laminates. This type of comparison
makes even more sense for a more complex structure than a test specimen. Here,
simple laminate degradation analysis faces more challenges. However, this requires
validated experimental methods capable of recording the evolution of failure.
A comparison of the individual capabilities and specific combinations of such
in situ methods are made in Chap. 7.
In order to deduce meaningful data from stress–strain curves, one has to assure
that the underlying assumptions are fulfilled. One prominent example is tensile
testing of unidirectional specimens with load axis parallel to the fiber axis. In such
configurations, first failure in form of interfiber fracture occurs already at 70 % of
the ultimate load [25]. In addition, fiber alignment is far from perfect and therefore
fiber failure initiates at the edge of the specimen due to off-axis loading or stress
concentration effects. Hence, the measured ultimate strength used as first-ply
failure strength in this load configuration is a value obtained in a material state,
which is typically far away from the undamaged state. This is in contrast to the
usual assumptions made in failure theories. But the impact of these preliminary
failures on the achievable ultimate strength is certainly dependent on the individual
material type and production process. In such cases, methods capable to track the
evolution of damage are suitable to assist the interpretation of stress–strain curves.
36 2 Failure of Fiber-Reinforced Composites

2.3.3 Ductile Matrix Materials

Traditionally, fiber-reinforced polymers are fabricated from brittle matrix systems,


such as epoxy resins. Consequently, many standards used for testing of fiber-
reinforced polymers implicitly assume brittle failure behavior of the matrix system.
Therefore, some of the assumptions made for the data reduction routines cannot be
directly transferred to matrix systems with ductile failure behavior, for example,
due to nonlinearities in the stress–strain behavior of these materials.
In recent years, an increasing interest in thermoplastic matrix materials has been
evident. However, the testing of fiber-reinforced thermoplastic polymers is not a
straightforward procedure using established standards. Nevertheless, some of the
existing standards are still applicable if the data reduction routine is accompanied
by additional information using in situ methods.
For all test methods, the validity of specimen geometries is subject to discussion.
In many thermoplastic materials, the interlaminar properties are substantially
higher than for typical epoxy resins. Therefore, many test specimens prepared to
fail via interlaminar or transverse stresses may fail in unintended failure modes.
Suitable adjustment of the specimen geometries or changes to the concept of load
introduction can be used to avoid such failure. Assistance by secondary methods
monitoring the occurrence of failure outside the intended position is of great
practical usage for such reinvention of testing procedures.
For fracture mechanics applications, a particular challenge is the suitable detec-
tion of first delamination onsets, since the increased ductility of the matrix material
tends to reduce the stress concentration at the crack tip. At the same time, the
validity of boundary conditions given by linear elastic fracture mechanics is subject
to discussion. Secondary methods can be used to provide better insight into the
progression of damage and therefore allow better interpretation of the respective
crack propagation energies.

2.4 What Can In Situ Methods Contribute to Mechanical


Testing?

Based on the difficulties to obtain reliable material properties from stress–strain or


load–displacement curves several attempts have been made to improve the level of
reliability. In general, the idea is to aid the interpretation of the material behavior by
a secondary method independent in its measurement principle from the macro-
scopic stress–strain response. The relevance and suitability of the secondary exper-
imental methods stems from the intended purpose and is briefly reviewed for
several methods in the following.
2.4 What Can In Situ Methods Contribute to Mechanical Testing? 37

2.4.1 In Situ Microscopy

The application of in situ microscopy during testing has become a standard in some
test configurations for fiber-reinforced materials. Namely, testing under mode I
condition facilitates usage of a travelling microscope to measure the accurate crack
propagation length. But also other test methods can benefit from accompanied
in situ microscopy as seen in Fig. 2.25, for a simple unidirectional test specimen
with fiber orientation parallel to the load axis. The simultaneous recording of
images can easily aid to interpret load drops preceding the maximum load. In this
test configuration, such behavior is often found for preliminary fiber failure induced
at the edges.
However, the major drawback of such imaging techniques is the lack of
sensitivity associated with their field of view. Naturally, this kind of imaging is
limited to detection of failure at the surface level unless the internal failure is so
severe that their occurrence causes corresponding damage at the surface. Also, a
proper trade-off between a high resolution and a large field of view has to be
made beforehand.

2.4.2 Digital Image Correlation

Digital image correlation also counts as an in situ imaging technique, but is


discussed independently, since it comprises various new aspects to be used for
detection of failure in fiber-reinforced materials. In the context of materials testing,
the method is used to deduce the spatially resolved strain field using a series of

2500

2000
stress [MPa]

1500

1000

500

0
0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8
strain [%]

+σ||

Fig. 2.25 Stress–strain curve from tensile testing of unidirectional fiber-reinforced polymer.
Visual observation of the test region at distinct load steps. Arrows indicate position of
corresponding failure at each load level
38 2 Failure of Fiber-Reinforced Composites

10 mm 1st principal strain [%] z


1.0
0.9
0.8
y
load axis
0.7
z reduced region
y 0.6
0.5
x 0.4
0.3

strain concentration 0.2

0.1
clamping 0.0

tapered region

reinforcement
region

Fig. 2.26 DIC image of inhomogeneous strain field at wedge between reinforcement region and
reduced region of tensile specimen

sequential images with speckle-type patterns (see also Chap. 3 for a detailed
description). In terms of the challenges associated with testing of fiber-reinforced
materials, there is one major improvement by application of DIC. This is the
possibility to perform full-field measurements of strain values to deduce material
properties [95, 193, 194]. As seen in the example in Fig. 2.26, the DIC method is able
to compute the strain values not only between two particular points, but within all
parts of the specimen covered by a speckle pattern. This effectively allows selecting
only a particular area to obtain strain values for data reduction. Therefore, regions
with inhomogeneous strain fields (e.g., at the reinforcements) can be excluded and
valid strain values (e.g., within the reduced section) can be obtained. At the same
time, such full-field measurements allow to optimize specimen geometries to assure
the anticipated strain state is reached in every part of the test volume.
In recent years, DIC using high-resolution camera systems became available.
This provides even further possibilities for the analysis of composite materials. In
addition to the possibility to perform full-field measurements, one can observe
disturbances and anomalies of the strain field during such experiments. Intuitively
one would associate such signatures with other anomalies such as the occurrence of
failure. As seen in Fig. 2.27, the presence of these signatures in strain field
measurements encourages the investigation of the limits of the DIC method in
this aspect as presented in Chap. 3.

2.4.3 X-Ray Methods

Complementary to the conventional in situ imaging methods, several research


groups demonstrated the use of X-ray imaging to obtain information on the damage
progress in fiber-reinforced materials. The main advantage of X-Rays compared to
2.4 What Can In Situ Methods Contribute to Mechanical Testing? 39

10 mm
2nd principal strain 1st principal strain

disturbance of
strain anomaly strain field

load axis

Fig. 2.27 Anomalies and disturbances of the strain field as readily visible by DIC systems

light is their ability to penetrate the specimen and therefore to obtain volumetric
information. The possibility to use computed tomography for visualization of
internal damage states has become a standard method already.
One application of X-rays considers the possibility to detect the onset and
growth of interfiber failure during mechanical loading [106, 195]. Using X-ray
refraction,2 these groups were able to detect the onset and growth of internal defects
such as fiber–matrix debonding due to changes in the refraction index of the
specimen.
Other groups have been using synchrotron radiation in combination with in situ
loading stages to carry out volumetric imaging of miniature specimens under
mechanical load [101, 109, 196, 197]. Such microscopic imaging of the specimen
intuitively allows to track the initiation of damage in the interior and to deduce the
interaction between different failure mechanisms at increasing load levels. As seen
in Fig. 2.28 in an exemplary image using a commercial X-ray computed tomogra-
phy device, the level of detail reached in small specimens is sufficiently high to
visualize details of the fracture mode and even to spot single fiber filaments. This
gave rise to numerous investigations to track the occurrence and accumulation of
failure in composite materials under mechanical load. Other groups have adopted

2
X-Ray refraction is similar to refraction of the visible light and is not identical to the established
techniques of X-ray diffraction (see [195]).
40 2 Failure of Fiber-Reinforced Composites

0.8 mm

z
x
y

Fig. 2.28 3D image of small carbon/epoxy composite specimen with details of the fracture mode

this analysis routine and investigated the failure of fiber-reinforced materials under
various load conditions [94, 101, 103, 109, 197, 198].
In particular, it is technically feasible to combine every mechanical test config-
uration with a computed tomography device in situ. The term “in situ” in this
context typically refers to load-hold cycles with intermediate scanning. That way
damage is first introduced and existing damage states are kept under load. For
imaging stress relaxation effects have to be considered in general and creep effects
of the matrix polymer have to be taken into account in particular. Moreover, there is
currently no feasible technical solution to scale the method to allow specimen
dimensions larger than the few millimeters of cross section currently used. But
the possibilities of this approach to assist in interpretation of failure of fiber-
reinforced materials are so encouraging that a particular focus on this topic is
given in Chap. 6.

2.4.4 Thermography

Another type of in situ measurement that is applied in combination with mechanical


testing of fiber-reinforced materials is thermography imaging. As the name indi-
cates, this also allows for continuous imaging of the specimen under load, but the
origin of the visual information is fairly different to the previously presented
techniques. In contrast to the imaging methods discussed before, the thermography
cameras record the spatial generation of heat via detection of the generated infrared
radiation. The generation of heat can be attributed to two source mechanisms. The
first one is the heat release during fracture of a material. For strong failure events,
this dissipative energy contribution is readily detectable by commercial equipment
as seen in Fig. 2.29. Here, tensile testing of an open-hole specimen was monitored
by thermography imaging. The heat release during macroscopic fracture saturates
the measurement range of the equipment. Before ultimate failure, several heat
signatures could be spotted in the area surrounding the hole, which allows moni-
toring evolution of damage to a certain extent. The second type of heat source is
friction inside the material or between the support and the test specimen. While the
generation of heat due to the latter is a general problem especially during fatigue
testing of composites [199–201], the friction inside the material is key to assist in
2.4 What Can In Situ Methods Contribute to Mechanical Testing? 41

high

low
t1 t2 t3 t4

Fig. 2.29 Thermography images of four subsequent times (t1 < t2 < t3 < t4) during fracture of an
open-hole tensile test

interpretation of material failure. During periodic cyclic loading experiments,


existing crack surfaces can rub against each other and therefore generate a notice-
able amount of heat. Using thermography imaging, this allows spotting damaged
areas and to quantify their evolution to a certain extent (Fig. 2.29).
However, there are two problems associated with both types of heat sources. As
mentioned above, the transient heat pulse due to material fracture is only detectable
for larger failure events. In addition, the heat transfer from the damaged spot to the
surface causes strong dispersion of the spatial information as function of subsurface
position and time. This effect is also relevant for the generation of heat by cyclic
friction of crack surfaces, but is superimposed additionally by the accumulation of
heat in the specimen itself. Therefore no attempts have been made to directly
correlate the occurrence of individual failure types in fiber-reinforced materials to
signatures in thermography images. However, it is worth noting that the resolution
of commercial infrared cameras is well suited to measure temperature rises of few
milli Kelvin with sufficient spatial resolution.

2.4.5 Shearography

Shearography is another optical method, which is distinct to the above mentioned


methods in the way it is used for in situ monitoring. The specimen under investi-
gation is illuminated using an expanded laser beam and images are taken using a
CCD camera via an optical shearing element [202]. The shearing element allows a
coherent superposition of two laterally displaced images of the surface of the
specimen in the image plane. The lateral displacement introduced by the shearing
element is called the shear of the images. The superposition of the two images is
called the shearogram, which is an interferogram of an object wave with the sheared
object wave as a reference wave. To allow for defect detection, two shearograms
are recorded for different loading conditions of the sample. The type of loading
should induce some deformation of the specimen, which is visible at the surface. In
42 2 Failure of Fiber-Reinforced Composites

practice, thermal, acoustical, or mechanical loading are applied in a static or


dynamic way [203–208]. Due to internal referencing shearography is inherently
insensitive to the environmental vibrations and therefore applicable to testing in
field environments.
The absolute difference of the shearograms recorded at different loading states
results in an interference fringe pattern, which is directly correlated to the local
slope of the specimen deformation. Hence, anomalies in the strain field can be
detected and interpreted in a similar fashion as discussed in Sect. 2.4.2 for digital
image correlation. This has been actively pursued for defect detection in composite
materials in the past years but has been mostly directed towards field inspection and
to detect mostly macroscopic defect sizes [206–209].

2.4.6 Ultrasonic Measurements

For the field of ultrasonic measurements, various approaches are reported in


literature to aid the interpretation of occurrence of damage in fiber-reinforced
materials. Well known are the inspection methods using active pulse-echo or
through-transmission configurations to visualize damage inside test specimens.
However, the imaging mode of these configurations (i.e., B-scans or C-scans) are
not straightforward to be used in situ with mechanically loaded specimens. But due
to the close relation between the elastic properties of a material and the ultrasound
velocity and attenuation, guided wave ultrasound pulses can be used to track the
damage state in a fiber-reinforced material. The sensitivity of guided acoustic
waves to such internal defects is very high as long as the wavelength is sufficiently
small to interact with the defect. Such in situ monitoring using two or more sensors
is always limited to the occurrence of damage within the full propagation path
between the sensors. New attempts apply ultrasonic tomography and use sensor
arrays to localize damaged areas by beam forming techniques [210–213]. This
allows a substantial improvement of the spatial resolution to detect the occurrence
of failure in a material, but the temporal resolution is still limited to the acquisition
rate of the active pulses.
Another method suitable to detect the occurrence of failure in situ is the
application of (nonlinear) vibrometry. In this configuration an active pulser is
used to generate cyclic acoustic waves in the specimen under load. One can
image the defect positions by observation of the acoustic wave field using a
scanning laser vibrometer. The occurrence of failure mechanisms of a characteristic
size will cause a noticeable change in the resonances visible as acoustic wave field.
For a particular position this can be quantified in terms of amplitude changes and
frequency or phase shifts [214–216].
2.4 What Can In Situ Methods Contribute to Mechanical Testing? 43

2.4.7 Acoustic Emission

Many approaches are reported in literature to monitor acoustic emission during


testing of fiber-reinforced materials [93, 217–234]. In general, all mechanisms
described in Sect. 2.1 generate acoustic emission during their formation. The
detection of the respective acoustic emission waves by several sensors allows the
inverse calculation of the source position based on the arrival time of the signals at
the different sensors. At the same time the detection and accumulation of signals
allows an in situ monitoring of the damage progress in fiber-reinforced materials. In
this context, it is worth noting that the sensitivity of the acoustic emission method is
extraordinary high, which easily allows detection of single fiber filament failure as
demonstrated in Chap. 4. However, the level of detail seen by acoustic emission is
sometimes hard to relate to the level of description by failure theories or the level of
observation of other experimental methods. For instance, it is arguable whether the
onset of fiber failure as within a failure theory is identical to the first detection of
fiber failure by acoustic emission. Also, the propagation of the acoustic emission
signals in a test specimen has significant impact on the interpretability of the signal.
Therefore, only a thorough understanding of the acoustic emission source, the
signal propagation, and the detection process can lead to a reliable interpretation
of the measurement data. Chapter 4 presents recent developments in this field and is
used to provide an overview of test configurations in conjunction with acoustic
emission acquisition.
Beyond the application of acoustic emission to obtain properties of fiber-
reinforced materials, a large amount of work has also been conducted to use the
method in situ for inspection of structures [132, 235–238] or to conduct failure
prediction. Most noteworthy are the approaches to predict burst pressures in fiber-
reinforced materials [231, 239, 240]. In this context, the usage of acoustic emission
analysis to support interpretation of material failure has already been covered by
several internationally recognized standards such as ISO, EN, ASTM, DIN, or other
national or proprietary standards.

2.4.8 Electromagnetic Emission

Similarly to acoustic emission, the origin of electromagnetic emission is the active


occurrence of damage in the material. As demonstrated in Chap. 5, typical failure
types in fiber-reinforced composites also cause detectable electromagnetic
emission.
However, the details of the source mechanism of electromagnetic emission due
to material failure are still under controversial discussion in literature. Also, today
no commercial equipment is available to readily detect electromagnetic emission
signals. Therefore, electromagnetic emission itself is currently understood as a
method under development and is not yet covered by any technical standards.
44 2 Failure of Fiber-Reinforced Composites

However, the method is very promising for failure analysis of fiber-reinforced


materials and might emerge as a future in situ testing method.
Similar to acoustic emission, the electromagnetic emission benefits from the
high discrimination capability, that is, it only senses active material failure. There
are also indications of even higher discriminative capabilities than for acoustic
emission, since it might be speculated if friction does provide electromagnetic
emission signals. Since friction sources form a typical, but unwanted acoustic
emission source, this discrimination would help to improve the reliability of signal
interpretation. Also, the electromagnetic signals show negligible influence of the
propagation medium and sensor systems are available that show relatively flat
frequency response. Source orientations can be detected using sensor arrays and
can be used to quantify the resulting orientations of interfiber crack planes. As a
technical drawback, the present systems signal-to-noise ratio is one to two orders of
magnitude lower than for commercially available acoustic emission measurement
systems. Moreover, the signal amplitude shows a strong decay as function of source
sensor distance. Despite of these drawbacks, Chap. 5 demonstrates first attempts
dedicated to use electromagnetic emission analysis to monitor failure in fiber-
reinforced materials.

References

1. Yamada, I., Masuda, K., Mizutani, H.: Electromagnetic and acoustic emission associated
with rock fracture. Phys. Earth Planet. Inter. 57, 157–168 (1989)
2. Frid, V., Rabinovitch, A., Bahat, D.: Fracture induced electromagnetic radiation. J. Phys. D
Appl. Phys. 36, 1620–1628 (2003)
3. Griffith, A.A.: The phenomena of rupture and flow in solids. Philos. Trans. R. Soc. Lond. Ser.
A 221, 163–198 (1921)
4. Irwin, G.: Analysis of stresses and strains near the end of a crack traversing a plate. J. Appl.
Mech. 24, 361–364 (1957)
5. Sch€urmann, H.: Konstruieren mit Faser-Kunststoff-Verbunden. Springer, Berlin (2005)
6. Matthews, F.L., Rawlings, R.D.: Composite Materials: Engineering and Science. Woodhead,
Cambridge (1999)
7. Greenhalgh, E.: Failure Analysis and Fractography of Polymer Composites. Woodhead,
Oxford (2009)
8. Sause, M.: Identification of Failure Mechanisms in Hybrid Materials Utilizing Pattern
Recognition Techniques Applied to Acoustic Emission Signals. mbv-Verlag, Berlin (2010)
9. Kermode, J.R., Albaret, T., Sherman, D., Bernstein, N., Gumbsch, P., Payne, M.C., Csányi,
G., De Vita, A.: Low-speed fracture instabilities in a brittle crystal. Nature 455, 1224–1227
(2008)
10. He, F., Tan, C.M., Zhang, S., Cheng, S.: Monte Carlo simulation of fatigue crack initiation at
elevated temperature. In: 13th Conference on Fracture, pp. 1–10. Beijing, China (2013)
11. Guglhoer, T., Manger, F., Sause, M.G.R.: Quantification of local fiber distribution for
optimization of tape laying techniques. In: ECCM16—16th European Conference on Com-
posite Materials, pp. 1–9. Sevilla, Spain (2014)
12. Salaberger, D., Arikan, M., Paier, T., Kastner, J.: Characterization of damage mechanisms in
glass fibre reinforced polymers using x-ray computed tomography. In: 11th European Con-
ference on Non-Destructive Testing (ECNDT 2014), pp. 1–9. Prague, Czech Republic (2014)
References 45

13. Hobbiebrunken, T., Hojo, M., Fiedler, B., Tanaka, M., Ochiai, S., Schulte, K.:
Thermomechanical analysis of micromechanical formation of residual stresses and initial
matrix failure in CFRP. JSME Int. J. Ser. A 47, 349–356 (2004)
14. Hobbiebrunken, T., Fiedler, B., Hojo, M., Ochiai, S., Schulte, K.: Microscopic yielding of
CF/epoxy composites and the effect on the formation of thermal residual stresses. Compos.
Sci. Technol. 65, 1626–1635 (2005)
15. Greisel, M., Jäger, J., Moosburger-Will, J., Sause, M.G.R., Mueller, W.M., Horn, S.: Influ-
ence of residual thermal stress in carbon fiber-reinforced thermoplastic composites on
interfacial fracture toughness evaluated by cyclic single-fiber push-out tests. Compos. Part
A Appl. Sci. Manuf. 66, 117–127 (2014)
16. Tanaka, T., Nakayama, H., Sakaida, A., Horikawa, N.: Estimation of tensile strength distri-
bution for carbon fiber with diameter variation along fiber. Mater. Sci. Res. Int. 5, 90–97
(1999)
17. Jumahat, A., Soutis, C., Jones, F., Hodzic, A.: Fracture mechanisms and failure analysis of
carbon fibre/toughened epoxy composites subjected to compressive loading. Compos. Struct.
92, 295–305 (2010)
18. Alhashmi, H., Kumar, S.: Micro-mechanical modeling of fiber pull-out stresses in an axi-
symmetric composite system. In: ECCM16—16th European Conference on Composite
Materials, pp. 1–9. Sevilla, Spain (2014)
19. Yang, L., Thomason, J.L.: Interface strength in glass fibre-polypropylene measured using the
fibre pull-out and microdebond methods. In: ICCM 17—17th International Conference on
Composite Materials, pp. 1–9, Edinburgh (2009)
20. Lin, Y., Scheuring, T., Friedrich, K.: Matrix morphology and fibre pull-out strength of T700/
PPS and T700/PET thermoplastic composite. J. Mater. 30, 4761–4769 (1995)
21. Pisanova, E., Zhandarov, S., Mäder, E.: How can adhesion be determined from
micromechanical tests? Compos. Part A Appl. Sci. Manuf. 32, 425–434 (2001)
22. Novak, J., Pearce, C.J., Grassl, P., Yang, L., Thomason, J.: Analysis of the microbond test
using nonlinear fracture mechanics. In: ICCM 17—17th International Conference on Com-
posite Materials, pp. 1—13. Edinburgh (2009)
23. Fuentes, C.A., Brughmans, G., Verpoest, I., van Vuure, A.W., Tran, L.Q.N., Dupont-Gillain,
C.: Effect of roughness on the interface in natural fibre composites: physical adhesion and
mechanical interlocking. In: ECCM16—16th European Conference on Composite Materials.
Sevilla, Spain (2014)
24. Reifsnider, K., Sendeckyj, G., Wang, S., Johnson, W., Stinchcomb, W., Pagano, N., Nahas,
M.: Survey of failure and post-failure theories of laminated fiber-renforced composites.
J. Compos. Technol. Res. 8, 138 (1986)
25. Puck, A.: Festigkeitsanalyse von Faser-Matrix-Laminaten Modelle f€ ur die Praxis. Carl
Hanser Verlag, Munich, Germany (1996)
26. Tsai, S.W., Wu, E.M.: A general theory of strength for anisotropic materials. J. Compos.
Mater. 5, 58–80 (1971)
27. Wu, E.: Phenomenological anisotropic failure criteria. Mech. Compos. Mater. 2, 353–431
(1974)
28. Hashin, Z.: Failure criteria for unidirectional fiber composites. J. Appl. Mech. 47, 329 (1980)
29. Mohr, O.: Welche Umstände bedingen die Elastizitätsgrenze und den Bruch eines materials?
Z. Ver. Dtsch. Ing. 44, 1524–1530 (1900)
30. Cuntze, R.: Comparison between experimental and theoretical results using Cuntze’s “failure
mode concept” model for composites under triaxial loadings—part B of the second world-
wide failure exercise. J. Compos. Mater. 47, 893–924 (2013)
31. Puck, A.: Festigkeitsberechnung an Glasfaser/Kunststoff-Laminaten bei zusammengesetzter
Beanspruchung. Kunststoffe 59, 780–787 (1969)
32. Puck, A.: GFK-Drehrohrfedern sollen h€ochstbeanspruchte Stahlfedern ersetzen. Kunststoffe
80, 1380–1384 (1990)
33. Puck, A.: Ein Bruchkriterium gibt die Richtung an. Kunststoffe 82, 607–610 (1992)
46 2 Failure of Fiber-Reinforced Composites

34. Puck, A., Kopp, J., Knops, M.: Guidelines for the determination of the parameters in Puck’s
action plane strength criterion. Compos. Sci. Technol. 62, 371–378 (2002)
35. Puck, A., Sch€ urmann, H.: Failure analysis of FRP laminates by means of physically based
phenomenological models. Compos. Sci. Technol. 62, 1633–1662 (2002)
36. Puck, A., Mannigel, M.: Physically based non-linear stress–strain relations for the inter-fibre
fracture analysis of FRP laminates. Compos. Sci. Technol. 67, 1955–1964 (2007)
37. Soden, P.D., Hinton, M.J., Kaddour, A.S.: Biaxial test results for strength and deformation of
a range of E-glass and carbon fibre reinforced composite laminates: failure exercise bench-
mark data. Compos. Sci. Technol. 62, 1489–1514 (2002)
38. Hinton, M.J., Kaddour, A.S., Soden, P.D.: Evaluation of failure prediction in composite
laminates: background to “part B” of the exercise. Compos. Sci. Technol. 62, 1481–1488
(2002)
39. Hinton, M.J., Soden, P.D.: Predicting failure in composite laminates: the background to the
exercise. Compos. Sci. Technol. 58, 1001–1010 (1998)
40. Soden, P.D., Hinton, M.J., Kaddour, A.S.: Lamina properties, lay-up configurations and
loading conditions for a range of fibre-reinforced composite laminates. Compos. Sci.
Technol. 58, 1011–1022 (1998)
41. Kaddour, A.S., Hinton, M.J.: Maturity of 3D failure criteria for fibre-reinforced composites:
comparison between theories and experiments: part B of WWFE-II. J. Compos. Mater. 47,
925–966 (2013)
42. Hinton, M., Kaddour, A.: The background to part B of the second world-wide failure
exercise: evaluation of theories for predicting failure in polymer composite laminates under
three-dimensional states of stress. J. Compos. Mater. 47, 643–652 (2013)
43. Hinton, M., Kaddour, A.: Triaxial test results for fibre reinforced composites: the second
world-wide failure exercise benchmark data. J. Compos. Mater. 47, 925–966 (2013)
44. Puck, A.: Failure analysis of FRP laminates by means of physically based phenomenological
models. Compos. Sci. Technol. 58, 1045–1067 (1998)
45. Zinoviev, P.A., Lebedeva, O.V., Tairova, L.P.: A coupled analysis of experimental and
theoretical results on the deformation and failure of composite laminates under a state of
plane stress. Compos. Sci. Technol. 62, 1711–1723 (2002)
46. Zinoviev, P.: The strength of multilayered composites under a plane-stress state. Compos.
Sci. Technol. 58, 1209–1223 (1998)
47. Cuntze, R.G., Freund, A.: The predictive capability of failure mode concept-based strength
criteria for multidirectional laminates. Compos. Sci. Technol. 64, 343–377 (2004)
48. Soden, P.D., Kaddour, A.S., Hinton, M.J.: Recommendations for designers and researchers
resulting from the world-wide failure exercise. Compos. Sci. Technol. 64, 589–604 (2004)
49. Pinho, S., Vyas, G., Robinson, P.: Material and structural response of polymer-matrix fibre-
reinforced composites: part B. J. Compos. Mater. 47, 679–696 (2013)
50. Pinho, S., Darvizeh, R., Robinson, P., Schuecker, C., Camanho, P.: Material and structural
response of polymer-matrix fibre-reinforced composites. J. Compos. Mater. 46, 2313–2341
(2012)
51. Carrere, N., Laurin, F., Maire, J.-F.: Micromechanical-based hybrid mesoscopic 3D approach
for non-linear progressive failure analysis of composite structures. J. Compos. Mater. 46,
2389–2415 (2012)
52. Carrere, N., Laurin, F., Maire, J.-F.: Micromechanical-based hybrid mesoscopic three-
dimensional approach for non-linear progressive failure analysis of composite structures—
part B: comparison with experimental data. J. Compos. Mater. 47, 743–762 (2012)
53. Matthias Deuschle, H., Kroplin, B.-H.: Finite element implementation of Puck’s failure
theory for fibre-reinforced composites under three-dimensional stress. J. Compos. Mater.
46, 2485–2513 (2012)
54. Deuschle, H.M., Puck, A.: Application of the Puck failure theory for fibre-reinforced com-
posites under three-dimensional stress: comparison with experimental results. J. Compos.
Mater. 47, 827–846 (2013)
References 47

55. Kaddour, A., Hinton, M., Smith, P., Li, S.: The background to the third world-wide failure
exercise. J. Compos. Mater. 47, 2417–2426 (2013)
56. Huang, Y., Jin, C., Ha, S.K.: Strength prediction of triaxially loaded composites using a
progressive damage model based on micromechanics of failure. J. Compos. Mater. 47,
777–792 (2013)
57. Zhang, D., Xu, L., Ye, J.: Prediction of failure envelopes and stress–strain curves of fiber
composite laminates under triaxial loads: comparison with experimental results. J. Compos.
Mater. 47, 763–776 (2013)
58. Tessmer, J.: Theoretische und algorithmische Beitraege zur Berechnung von
Faserverbundschalen. Ph.D. Thesis, University of Hannover (2000)
59. Juhasz, T.: Ein neues physikalisch basiertes Versagenskriterium f€ ur schwach 3D-verstärkte
Faserverbundlaminate. Ph.D. Thesis, University Carolo-Wilhelmina Braunschweig (2003)
60. Axelrad, D.: Micromechanics of Solids. Elsevier Scientific, Amsterdam (1978)
61. Huang, Z.M., Ramakrishna, S.: Micromechanical modeling approaches for the stiffness and
strength of knitted fabric composites: a review and comparative study. Compos. Part A Appl.
Sci. Manuf. 31, 479–501 (2000)
62. Kuksenko, V.S., Tamuzs, V.P.: Fracture Micromechanics of Polymer Materials. Springer,
Dordrecht (1981)
63. Stellbrink, K.: Micromechanics of Composites: Composite Properties of Fibre and Matrix
Constituents. Carl Hanser Verlag, Munich (1996)
64. Voyiadjis, G.Z., Deliktas, B.: A coupled anisotropic damage model for the inelastic response
of composite materials. Comput. Meth. Appl. Mech. Eng. 183, 159–199 (2000)
65. Wisnom, M.R.: Size effects in the testing of fibre-composite materials. Compos. Sci.
Technol. 59, 1937–1957 (1999)
66. Flaggs, D.L., Kural, M.H.: Experimental determination of the in-situ transverse lamina
strength in graphite/epoxy laminates. J. Compos. Mater. 16, 103–116 (1982)
67. Ogin, S.L., Smith, P.A.: Fast fracture and fatigue growth of transverse ply cracks in com-
posite laminates. Scr. Metall. 19, 779–784 (1985)
68. Parvizi, A., Garrett, K.W., Bailey, J.E.: Constrained cracking in glass fibre-reinforced epoxy
cross-ply laminates. J. Mater. Sci. 13, 195–201 (1978)
69. Nairn, J., Hu, S.: The initiation and growth of delaminations induced by matrix microcracks
in laminated composites. Int. J. Fract. 57, 1–24 (1992)
70. Nairn, J.A., Hu, S.: The formation and effect of outer-ply microcracks in cross-ply laminates:
a variational approach. Eng. Fract. Mech. 41, 203–221 (1992)
71. Anderson, T.L.: Fracture Mechanics: Fundamentals and Applications, 3rd edn. Taylor &
Francis Group, Boca Raton (2005)
72. Groos, D.: Bruchmechanik. Springer, Berlin (2007)
73. Rice, J.R.: A path independent integral and the approximate analysis of strain concentration
by notches and cracks. J. Appl. Mech. 35, 379 (1968)
74. Rice, J.R.: Elastic fracture mechanics concepts for interfacial cracks. J. Appl. Mech. 55,
98 (1988)
75. Dowling, N.E.: Mechanical Behavior of Materials. Prentice Hall, Upper Saddle River (1999)
76. Doudican, B.M., Zand, B., Amaya, P., Butalia, T.S., Wolfe, W.E., Schoeppner, G.A.: Strain
energy based failure criterion: comparison of numerical predictions and experimental obser-
vations for symmetric composite laminates subjected to triaxial loading. J. Compos. Mater.
47, 847–866 (2013)
77. Mueller, W.M., Moosburger-Will, J., Sause, M.G.R., Horn, S.: Microscopic analysis of
single-fiber push-out tests on ceramic matrix composites performed with Berkovich and
flat-end indenter and evaluation of interfacial fracture toughness. J. Eur. Ceram. Soc. 33,
441–451 (2013)
78. Jäger, J., Sause, M.G.R., Burkert, F., Moosburger-Will, J., Greisel, M., Horn, S.: Influence of
plastic deformation on single-fiber push-out tests of carbon fiber reinforced epoxy resin.
Compos. Part A Appl. Sci. Manuf. 71, 157–167 (2015)
48 2 Failure of Fiber-Reinforced Composites

79. Kaddour, A., Hinton, M., Smith, P., Li, S.: A comparison between the predictive capability of
matrix cracking, damage and failure criteria for fibre reinforced composite laminates: part A
of the third world-wide failure exercise. J. Compos. Mater. 47, 2749–2779 (2013)
80. Chamis, C.C., Abdi, F., Garg, M., Minnetyan, L., Baid, H., Huang, D., Housner, J., Talagani,
F.: Micromechanics-based progressive failure analysis prediction for WWFE-III composite
coupon test cases. J. Compos. Mater. 47, 2695–2712 (2013)
81. Laurin, F., Carrere, N., Huchette, C., Maire, J.-F.: A multiscale hybrid approach for damage
and final failure predictions of composite structures. J. Compos. Mater. 47, 2713–2747 (2013)
82. Daghia, F., Ladeveze, P.: Identification and validation of an enhanced mesomodel for
laminated composites within the WWFE-III. J. Compos. Mater. 47, 2675–2693 (2013)
83. Flatscher, T., Schuecker, C., Pettermann, H.: A constitutive ply model for stiffness degrada-
tion and plastic strain accumulation: its application to the Third World Wide Failure Exercise
(part A). J. Compos. Mater. 47, 2575–2593 (2013)
84. Pinho, S., Vyas, G., Robinson, P.: Response and damage propagation of polymer-matrix
fibre-reinforced composites: predictions for WWFE-III part A. J. Compos. Mater. 47,
2595–2612 (2013)
85. Sapozhnikov, S.B., Cheremnykh, S.I.: The strength of fibre reinforced polymer under a
complex loading. J. Compos. Mater. 47, 2525–2552 (2013)
86. Forghani, A., Zobeiry, N., Poursartip, A., Vaziri, R.: A structural modelling framework for
prediction of damage development and failure of composite laminates. J. Compos. Mater. 47,
2553–2573 (2013)
87. Singh, C.V., Talreja, R.: A synergistic damage mechanics approach to mechanical response
of composite laminates with ply cracks. J. Compos. Mater. 47, 2475–2501 (2012)
88. McCartney, L.: Derivations of energy-based modelling for ply cracking in general symmetric
laminates. J. Compos. Mater. 47, 2641–2673 (2013)
89. McCartney, L.: Energy methods for modelling damage in laminates. J. Compos. Mater. 47,
2613–2640 (2012)
90. Gorbatikh, L., Ivanov, D., Lomov, S., Verpoest, I.: On modelling of damage evolution in
textile composites on meso-level via property degradation approach. Compos. Part A Appl.
Sci. Manuf. 38, 2433–2442 (2007)
91. Lomov, S., Ivanov, D., Verpoest, I., Zako, M., Kurashiki, T., Nakai, H., Hirosawa, S.: Meso-
FE modelling of textile composites: road map, data flow and algorithms. Compos. Sci.
Technol. 67, 1870–1891 (2007)
92. Whitcomb, J.D., Chapman, C.D., Tang, X.: Derivation of boundary conditions for
micromechanics analyses of plain and satin weave composites. J. Compos. Mater. 34,
724–747 (2000)
93. Mikhaluk, D.S., Truong, T.C., Borovkov, A.I., Lomov, S.V., Verpoest, I.: Experimental
observations and finite element modelling of damage initiation and evolution in carbon/
epoxy non-crimp fabric composites. Eng. Fract. Mech. 75, 2751–2766 (2008)
94. Ivanov, D.S., Baudry, F., Van Den Broucke, B., Lomov, S.V., Xie, H., Verpoest, I.: Failure
analysis of triaxial braided composite. Compos. Sci. Technol. 69, 1372–1380 (2009)
95. Lomov, S.V., Ivanov, D.S., Verpoest, I., Zako, M., Kurashiki, T., Nakai, H., Molimard, J.,
Vautrin, A.: Full-field strain measurements for validation of meso-FE analysis of textile
composites. Compos. Part A Appl. Sci. Manuf. 39, 1218–1231 (2008)
96. Ivanov, S.G., Gorbatikh, L., Lomov, S.V.: Interlaminar fracture behaviour of textile com-
posites with thermoplastic matrices. In: ECCM16—16th European Conference on Composite
Materials, pp. 1–8. Sevilla, Spain (2014)
97. Deuschle, H.M.: 3D failure analysis of UD fibre reinforced composites: Puck’s theory within
FEA. Ph.D. Thesis, University of Stuttgart (2010)
98. Edge, E.C.: Does transverse and shear loading affect the compression strength of unidirec-
tional CFC? A reply to Dr Hart-Smith. Composites 25, 159–161 (1994)
99. Hart-Smith, L.: Fibrous composite failure criteria-fact and fantasy. In: 7th International
Conference of Composite Structures. Paisley, (1993)
References 49

100. Scott, A.E., Mavrogordato, M., Wright, P., Sinclair, I., Spearing, S.M.: In-situ fibre fracture
measurement in carbon-epoxy laminates using high resolution computed tomography.
Compos. Sci. Technol. 71, 1471–1477 (2011)
101. Scott, A.E., Sinclair, I., Spearing, S.M., Mavrogordato, M., Bunsell, A.R., Thionnet, A.:
Comparison of the accumulation of fibre breaks occurring in a unidirectional carbon/epoxy
composite identified in a multi-scale micro-mechanical model with that of experimental
observations using high resolution computed tomography. In: Matériaux 2010, pp. 1–9.
Nantes, France (2010)
102. Fuwa, M., Harris, B., Bunsell, A.R.: Acoustic emission during cyclic loading of carbon-fibre-
reinforced plastics. J. Phys. D Appl. Phys. 8, 1460–1471 (2001)
103. Scott, A.E., Sinclair, I., Spearing, S.M., Thionnet, A., Bunsell, A.R.: Damage accumulation in
a carbon/epoxy composite: comparison between a multiscale model and computed tomogra-
phy experimental results. Compos. Part A Appl. Sci. Manuf. 43, 1514–1522 (2012)
104. Kaddour, A.S., Hinton, M.J., Soden, P.D.: A comparison of the predictive capabilities of
current failure theories for composite laminates: additional contributions. Compos. Sci.
Technol. 64, 449–476 (2004)
105. Ortwein, H.-P., Bohse, J., Trappe, V.: Untersuchung der Mikrorissbildung in
Faserkunststoffverbunden mittels R€ontgen- refraktions- und Schallemissionsmessung. In:
18. Kolloquium Schallemission, pp. 1–7. Wetzlar, Germany (2011)
106. Trappe, V., G€ unzel, S., Hickmann, S.: Non-destructive evaluation of micro cracking in short
fibre reinforced thermoplastics with X-ray-refraction. In: ICCM 17—17th International
Conference on Composite Material. Edinburgh (2009)
107. Trappe, V., Hickmann, S., Sturm, H.: Bestimmung des Zwischenfaserbruchversagens in
textilfaserverstärkten Glasfaserkunststoff mittels Refraktionstopographie. Mater. Test. 50,
615–622 (2008)
108. Metzkes, K., Trappe, V.: Damage Evolution in short fibre reinforced polyamide caused by
biaxial loading. In: ECCM16—16th European Conference on Composite Materials, pp. 1–8.
Sevilla, Spain (2014)
109. Scott, A.E., Sinclair, I., Spearing, S.M., Mavrogordato, M.N., Hepples, W.: Influence of voids
on damage mechanisms in carbon/epoxy composites determined via high resolution com-
puted tomography. Compos. Sci. Technol. 90, 147–153 (2014)
110. Ramesh, K.T.: High strain rate and impact experiments. In: Springer Handbook of Experi-
mental Solid Mechanics, pp. 1–31. Springer, Berlin (2008)
111. Horoschenkoff, A.: Beitrag zur Charakterisierung des nichtlinear thermoviskoelastischen
Kriechverhaltens von Faserverbundwerkstoffen. Ph.D. Thesis, Technical University of
Munich (1995)
112. Sakai, T., Somiya, S.: Analysis of creep behavior in thermoplastics based on visco-elastic
theory. Mech. Time-Dependent Mater. 15, 293–308 (2011)
113. Somiya, S.: Creep behavior of a carbon-fiber-reinforced thermoplastic polyimide resin.
J. Thermoplast. Compos. Mater. 7, 91–99 (1994)
114. Sakai, T., Somiya, S.: Analysis of creep behavior in thermoplastic based on visco-elastic
theory. In: Proceedings of the SEM Annual Conference. Albuquerque, New Mexico (2009)
115. Guedes, R.M.: Creep and fatigue lifetime prediction of polymer matrix composites based on
simple cumulative damage laws. Compos. Part A Appl. Sci. Manuf. 39, 1716–1725 (2008)
116. Guedes, R.M.: Relationship between lifetime under creep and constant stress rate for
polymer-matrix composites. Compos. Sci. Technol. 69, 1200–1205 (2009)
117. Reiner, M., Weissenberg, K.: A thermodynamic theory of the strength of the materials. Rheol.
Leafl. 10, 12–20 (1939)
118. Miner, M.: Cumulative damage in fatigue. J. Appl. Mech. 12, 159–164 (1945)
119. Christensen, R.: A physically based cumulative damage formalism. Int. J. Fatigue 30,
595–602 (2008)
50 2 Failure of Fiber-Reinforced Composites

120. Saulsberry, R.L., Greene, N.J., Hernandez, L.: Stress rupture nondestructive evaluation of
composite overwrapped pressure vessels. In: Lulla, K. (ed.) 2011 Biennial Research and
Technology Development Report, pp. 287–289. (2011)
121. Chang, J.B.: Implementation guidelines for ANSI/AIAA S-081: space systems composite
overwrapped pressure vessels. In: Aerospace Report No. TR-2003(8504)-1, AD A413531,
p. 83. (2003)
122. Miyano, Y., Nakada, M., Sekine, N.: Accelerated testing for long-term durability of FRP
laminates for marine use. J. Compos. Mater. 39, 5–20 (2005)
123. Miyano, Y., Nakada, M., Nishigaki, K.: Prediction of long-term fatigue life of quasi-isotropic
CFRP laminates for aircraft use. Int. J. Fatigue 28, 1217–1225 (2006)
124. Nakada, M., Okuya, T., Miyano, Y.: Statistical prediction of tensile creep failure time for
unidirectional CFRP. Adv. Compos. Mater. 23, 451–460 (2014)
125. W€ ohler, A.: Über die Festigkeits-Versuche mit Eisen und Stahl. Z. Bauwesen 1–3, 74–106
(1870)
126. Haibach, E.: Betriebsfestigkeit. Springer, Berlin (2006)
127. Palmgren, A.: Die Lebensdauer von Kugellagern. Z. Ver. Dtsch. Ing. 68, 339–341 (1924)
128. Degrieck, J., Van Paepegem, W.: Fatigue damage modeling of fibre-reinforced composite
materials: review. Appl. Mech. Rev. 54, 279 (2001)
129. Hwang, W., Han, K.S.: Cumulative damage models and multi-stress fatigue life prediction.
J. Compos. Mater. 20, 125–153 (1986)
130. Barnard, P.M., Butler, R.J., Curtis, P.T.: Composite Structures 3. Springer, Dordrecht (1985)
131. Farrow, I.R.: Damage accumulation and degradation of composite laminates under aircraft
service loading: assessment and prediction. Ph.D. Thesis, Cranfield Institute of Technology
(1989)
132. Bingham, A.H., Ek, C.W.: Acoustic emission testing of aerial devices and associated
equipment used in the utility industries, Ausgabe 1139. ASTM International (1992)
133. Sendeckyj, G.P.: Life prediction for resin-matrix composite materials. In: Reifsnider,
K.L. (ed.) Fatigue of Composite Materials. Elsevier B.V., Amsterdam (1990)
134. Hashin, Z., Rotem, A.: A fatigue failure criterion for fiber reinforced materials. J. Compos.
Mater. 7, 448–464 (1973)
135. Ellyin, F., El-Kadi, H.: A fatigue failure criterion for fiber reinforced composite laminae.
Compos. Struct. 15, 61–74 (1990)
136. Reifsnider, K., Gao, Z.: A micromechanics model for composites under fatigue loading. Int.
J. Fatigue 13, 149–156 (1991)
137. Wu, C.M.L.: Thermal and mechanical fatigue analysis of CFRP laminates. Compos. Struct.
25, 339–344 (1993)
138. Jen, M.: Strength and life in thermoplastic composite laminates under static and fatigue loads.
Part II: formulation. Int. J. Fatigue 20, 617–629 (1998)
139. Jen, M.: Strength and life in thermoplastic composite laminates under static and fatigue loads.
Part I: experimental. Int. J. Fatigue 20, 605–615 (1998)
140. Philippidis, T.P., Vassilopoulos, A.P.: Fatigue strength prediction under multiaxial stress.
J. Compos. Mater. 33, 1578–1599 (1999)
141. Castillo, E., Fernández-Canteli, A., Hadi, A.S.: On fitting a fatigue model to data. Int.
J. Fatigue 21, 97–106 (1999)
142. Hwang, W., Han, K.S.: Fatigue of composites–fatigue modulus concept and life prediction.
J. Compos. Mater. 20, 154–165 (1986)
143. Sidoroff, F., Subagio, B.: Fatigue damage modelling of composite materials from bending
tests. In: Sixth International Conference on Composite Materials (ICCM-VI) & Second
European Conference on Composite Materials (ECCM-II), vol. 4, pp. 4.32–4.39. London
(1987)
144. Van Paepegem, W., Degrieck, J.: Experimental set-up for and numerical modelling of
bending fatigue experiments on plain woven glass/epoxy composites. Compos. Struct. 51,
1–8 (2001)
References 51

145. Yang, J.N., Lee, L.J., Sheu, D.Y.: Modulus reduction and fatigue damage of matrix domi-
nated composite laminates. Compos. Struct. 21, 91–100 (1992)
146. Lee, L.J., Fu, K.E., Yang, J.N.: Prediction of fatigue damage and life for composite laminates
under service loading spectra. Compos. Sci. Technol. 56, 635–648 (1996)
147. Brøndsted, P., Lilholt, H., Andersen, S.I.: Fatigue damage prediction by measurements of the
stiffness degradation in polymer matrix composites. In: 8th International Spring Meeting of
ICFC. Paris, France (1997)
148. Whitworth, H.A.: A stiffness degradation model for composite laminates under fatigue
loading. Compos. Struct. 40, 95–101 (1997)
149. Chou, P., Croman, R.: Degradation and sudden-death models of fatigue of graphite/epoxy
composites. In: Composite Materials: Testing and Design (5th Conf) ASTM STP, (1979)
150. Halpin, C., Jerina, K., Johnson, T.: Characterization of composites for the purpose of
reliability evaluation. In: Analysis of the Test Methods for High Modulus Fibers and
Composites: A Symposium Presented at a Meeting of Committee D-30 on High Modulus
Fibers and Their Composites. (1973)
151. Yang, J., Jones, D.: Load sequence effects on the fatigue of unnotched composite materials.
In: Fatigue of Fibrous Composite Materials, pp. 213–232. ASTM STP, (1981)
152. Daniel, I.M., Charewicz, A.: Fatigue damage mechanisms and residual properties of graphite/
epoxy laminates. Eng. Fract. Mech. 25, 793–808 (1986)
153. Rotem, A.: Fatigue and residual strength of composite laminates. Eng. Fract. Mech. 25,
819–827 (1986)
154. Schaff, J.R., Davidson, B.D.: Life prediction methodology for composite structures. Part II—
spectrum fatigue. J. Compos. Mater. 31, 158–181 (1997)
155. Caprino, G.: Predicting fatigue life of composite laminates subjected to tension–tension
fatigue. J. Compos. Mater. 34, 1334–1355 (2000)
156. Yao, W.X., Himmel, N.: A new cumulative fatigue damage model for fibre-reinforced
plastics. Compos. Sci. Technol. 60, 59–64 (2000)
157. Biner, S.B., Yuhas, V.C.: Growth of short fatigue cracks at notches in woven fiber glass
reinforced polymeric composites. J. Eng. Mater. Technol. 111, 363 (1989)
158. Bergmann, H.W., Prinz, R.: Fatigue life estimation of graphite/epoxy laminates under
consideration of delamination growth. Int. J. Numer. Meth. Eng. 27, 323–341 (1989)
159. Dahlen, C., Springer, G.S.: Delamination growth in composites under cyclic loads.
J. Compos. Mater. 28, 732–781 (1994)
160. Bartley-Cho, J., Gyu Lim, S., Hahn, H.T., Shyprykevich, P.: Damage accumulation in quasi-
isotropic graphite/epoxy laminates under constant-amplitude fatigue and block loading.
Compos. Sci. Technol. 58, 1535–1547 (1998)
161. Bucinell, R.B.: Development of a stochastic free edge delamination model for laminated
composite materials subjected to constant amplitude fatigue loading. J. Compos. Mater. 32,
1138–1156 (1998)
162. Sch€ on, J.: A model of fatigue delamination in composites. Compos. Sci. Technol. 60,
553–558 (2000)
163. Highsmith, A., Reifsnider, K.: Stiffness-reduction mechanisms in composite laminates. In:
Damage in Composite Materials, pp. 103–117. ASTM International, Philadelphia, (1982)
164. Reifsnider, K.L.: The critical element model: a modeling philosophy. Eng. Fract. Mech. 25,
739–749 (1986)
165. Subramanian, S., Reifsnider, K.L., Stinchcomb, W.W.: A cumulative damage model to
predict the fatigue life of composite laminates including the effect of a fibre-matrix inter-
phase. Int. J. Fatigue 17, 343–351 (1995)
166. Talreja, R.: Stiffness properties of composite laminates with matrix cracking and interior
delamination. Eng. Fract. Mech. 25, 751–762 (1986)
167. Ogin, S.L., Smith, P.A., Beaumont, P.W.R.: Matrix cracking and stiffness reduction during
the fatigue of a (0/90)s GFRP laminate. Compos. Sci. Technol. 22, 23–31 (1985)
52 2 Failure of Fiber-Reinforced Composites

168. Spearing, S.M., Beaumont, P.W.R., Ashby, M.F.: Fatigue damage mechanics of composite
materials. II: a damage growth model. Compos. Sci. Technol. 44, 169–177 (1992)
169. Spearing, S.M., Beaumont, P.W.R.: Fatigue damage mechanics of composite materials. I:
experimental measurement of damage and post-fatigue properties. Compos. Sci. Technol. 44,
159–168 (1992)
170. Ladevèze, P.: A damage computational method for composite structures. Comput. Struct. 44,
79–87 (1992)
171. Shokrieh, M.M., Lessard, L.B.: Progressive fatigue damage modeling of composite materials.
Part I: modeling. J. Compos. Mater. 34, 1056–1080 (2000)
172. Shokrieh, M.M., Lessard, L.B.: Progressive fatigue damage modeling of composite materials.
Part II: material characterization and model verification. J. Compos. Mater. 34, 1081–1116
(2000)
173. Reifsnider, K.: Fatigue of Composite Materials. Elsevier, Amsterdam (1990)
174. Field, J.E., Walley, S.M., Proud, W.G., Goldrein, H.T., Siviour, C.R.: Review of experimen-
tal techniques for high rate deformation and shock studies. Int. J. Impact Eng. 30, 725–775
(2004)
175. Gama, B.A., Lopatnikov, S.L., Gillespie, J.W.: Hopkinson bar experimental technique: a
critical review. Appl. Mech. Rev. 57, 223 (2004)
176. Kawai, M., Saito, S.: Off-axis strength differential effects in unidirectional carbon/epoxy
laminates at different strain rates and predictions of associated failure envelopes. Compos.
Part A Appl. Sci. Manuf. 40, 1632–1649 (2009)
177. Vogler, T.J., Kyriakides, S.: Inelastic behavior of an AS4/PEEK composite under combined
transverse compression and shear. Part I: experiments. Int. J. Plast. 15, 783–806 (1999)
178. Hsiao, H.M., Daniel, I.M., Cordes, R.D.: Strain rate effects on the transverse compressive and
shear behavior of unidirectional composites. J. Compos. Mater. 33, 1620–1642 (1999)
179. Hosur, M.V., Alexander, J., Vaidya, U.K., Jeelani, S.: High strain rate compression response
of carbon/epoxy laminate composites. Compos. Struct. 52, 405–417 (2001)
180. Tsai, J.-L., Sun, C.T.: Strain rate effect on in-plane shear strength of unidirectional polymeric
composites. Compos. Sci. Technol. 65, 1941–1947 (2005)
181. Koerber, H., Xavier, J., Camanho, P.P.: High strain rate characterisation of unidirectional
carbon-epoxy IM7-8552 in transverse compression and in-plane shear using digital image
correlation. Mech. Mater. 42, 1004–1019 (2010)
182. K€orber, H., Camanho, P.P.: Characterisation of unidirectional carbon-epoxy IM7-8552 in
longitudinal compression under high strain rates. In: DYMAT 2009—9th International
Conferences on the Mechanical and Physical Behaviour of Materials Under Dynamic Load-
ing, pp. 185–191. EDP Sciences, Les Ulis, France (2009)
183. Koerber, H., Vogler, M., Kuhn, P., Camanho, P.P.: Experimental characterisation and
modelling of non-linear stress–strain behaviour and strain rate effects for unidirectional
carbon-epoxy. In: ECCM16—16th European Conference on Composite Materials, pp. 1–8.
Sevilla, Spain (2014)
184. Lee, D., Tippur, H., Bogert, P.: Quasi-static and dynamic fracture of graphite/epoxy com-
posites: an optical study of loading-rate effects. Compos. Part B Eng. 41, 462–474 (2010)
185. Gilat, A., Schmidt, T.E., Walker, A.L.: Full field strain measurement in compression and
tensile split hopkinson bar experiments. Exp. Mech. 49, 291–302 (2008)
186. Ciampa, F., Meo, M., Barbieri, E.: Impact localization in composite structures of arbitrary
cross section. Struct. Health Monit. 11(6), 643–655 (2012)
187. Ciampa, F., Meo, M.: Impact detection in anisotropic materials using a time reversal
approach. Struct. Health Monit. 11, 43–49 (2011)
188. Gorman, M.R., Ziola, S.M.: Hypervelocity impact (HVI). WLE Small-Scale Fiberglass Panel
Flat Target C-2, vol. 4. (2007)
189. Gorman, M.R., Ziola, S.M.: Hypervelocity impact (HVI). WLE High Fidelity Specimen Fg
(RCC)-2, vol. 6. (2007)
References 53

190. Bleier, A.: Pr€ufverfahren zur Bestimmung exakter Werkstoffkennwerte einer


unidirektionalen Schicht unter besonderer Ber€ucksichtigung physikalischer Nichtlinearitäten.
Ph.D. Thesis, Technical University Darmstadt (2011)
191. Basan, R.: Untersuchung der intralaminaren Schubeigenschaften von
Faserverbundwerkstoffen mit Epoxidharzmatrix unter Ber€ ucksichtigung nichtlinearer
Effekte. Ph.D. Thesis, Technical University Berlin (2011)
192. Blackman, B.R.K., Brunner, A.J., Williams, J.G.: Mode II fracture testing of composites: a
new look at an old problem. Eng. Fract. Mech. 73, 2443–2455 (2006)
193. Lomov, S.V., Boisse, P., Deluycker, E., Morestin, F., Vanclooster, K., Vandepitte, D.,
Verpoest, I., Willems, A.: Full-field strain measurements in textile deformability studies.
Compos. Part A Appl. Sci. Manuf. 39, 1232–1244 (2008)
194. Pierron, F.: Application of full-field measurement techniques to composite materials and
structures. Compos. Part A Appl. Sci. Manuf. 39, 1193 (2008)
195. M€ uller, B.R., Hentschel, M.P., Harbich, K.-W., Lange, A., Schors, J.: X-ray refraction
topography for non-destructive evaluation of advanced materials. In: 15th World Conference
on NDT, Roma, Italy (2000)
196. Baensch, F., Zauner, M., Sanabria, S.J., Sause, M.G.R., Pinzer, B.R., Brunner, A.J.,
Stampanoni, M., Niemz, P.: Damage evolution in wood: synchrotron radiation micro-
computed tomography (SRμCT) as a complementary tool for interpreting acoustic emission
(AE) behavior. Holzforschung 69 (2015)
197. Scott, A.E., Hepples, W., Kalantzis, N., Wright, P., Mavrogordato, M.N., Sinclair, I.,
Spearing, S.M.: High resolution damage detection of loaded carbon/epoxy laminates using
synchrotron radiation computed tomography. In: ICCM-18 18th International Conference on
Composite Materials, pp. 1–6. ICC Jeju, Korea (2011)
198. Patel, A.J., Sottos, N.R., Wetzel, E.D., White, S.R.: Autonomic healing of low-velocity
impact damage in fiber-reinforced composites. Compos. Part A Appl. Sci. Manuf. 41,
360–368 (2010)
199. Bagheri, Z.S., Tavakkoli Avval, P., Bougherara, H., Aziz, M.S.R., Schemitsch, E.H., Zdero,
R.: Biomechanical analysis of a new carbon fiber/flax/epoxy bone fracture plate shows less
stress shielding compared to a standard clinical metal plate. J. Biomech. Eng. 136, 091002
(2014)
200. Abello, L.S., Marco, Y., Le Saux, V., Robert, G., Charrier, P.: Fast prediction of the fatigue
behavior of short fiber reinforced thermoplastics from heat build-up measurements. Procedia
Eng. 66, 737–745 (2013)
201. Gornet, L., Wesphal, O., Burtin, C., Bailleul, J.-L., Rozycki, P., Stainier, L.: Rapid determi-
nation of the high cycle fatigue limit curve of carbon fiber epoxy matrix composite laminates
by thermography methodology: tests and finite element simulations. Procedia Eng. 66,
697–704 (2013)
202. Francis, D., Tatam, R.P., Groves, R.M.: Shearography technology and applications: a review.
Meas. Sci. Technol. 21, 102001 (2010)
203. Hung, Y.Y., Liang, C.Y.: Image-shearing camera for direct measurement of surface strains.
Appl. Opt. 18, 1046–51 (1979)
204. Hung, Y.Y.: Shearography for non-destructive evaluation of composite structures. Opt.
Lasers Eng. 24, 161–182 (1996)
205. Hung, Y.Y.: Shearography: a new optical method for strain measurement and nondestructive
testing. Opt. Eng. 21, 213391 (1982)
206. Wang, W.-C., Su, C.-W., Liu, P.-W.: Full-field non-destructive analysis of composite plates.
Compos. Part A Appl. Sci. Manuf. 39, 1302–1310 (2008)
207. Gryzagoridis, J., Findeis, D.: Benchmarking shearographic NDT for composites. Insight
non-destructive test. Cond. Monit. 50, 249–252 (2008)
208. De Angelis, G., Meo, M., Almond, D.P., Pickering, S.G., Angioni, S.L.: A new technique to
detect defect size and depth in composite structures using digital shearography and
unconstrained optimization. NDT E Int. 45, 91–96 (2012)
54 2 Failure of Fiber-Reinforced Composites

209. Pezzoni, R., Krupka, R.: Laser-shearography for non-destructive testing of large-area com-
posite helicopter structures. In: 15th World Conference on NDT, Roma, Italy (2000)
210. Leonard, K.R., Malyarenko, E.V., Hinders, M.K.: Ultrasonic Lamb wave tomography.
Inverse Probl. 18, 1795–1808 (2002)
211. Hay, T.R., Royer, R.L., Gao, H., Zhao, X., Rose, J.L.: A comparison of embedded sensor
Lamb wave ultrasonic tomography approaches for material loss detection. Smart Mater.
Struct. 15, 946–951 (2006)
212. Jansen, D.P., Hutchins, D.A., Mottram, J.T.: Lamb wave tomography of advanced composite
laminates containing damage. Ultrasonics 32, 83–90 (1994)
213. Prasad, S.M., Balasubramaniam, K., Krishnamurthy, C.: V: Structural health monitoring of
composite structures using Lamb wave tomography. Smart Mater. Struct. 13, N73–N79
(2004)
214. Solodov, I., Wackerl, J., Pfleiderer, K., Busse, G.: Nonlinear self-modulation and
subharmonic acoustic spectroscopy for damage detection and location. Appl. Phys. Lett.
84, 5386 (2004)
215. Solodov, I., Busse, G.: Nonlinear air-coupled emission: the signature to reveal and image
microdamage in solid materials. Appl. Phys. Lett. 91, 251910 (2007)
216. Krohn, N., Stoessel, R., Busse, G.: Acoustic non-linearity for defect selective imaging.
Ultrasonics 40, 633–637 (2002)
217. Pappas, Y.Z., Kostopoulos, V.: Toughness characterization and acoustic emission monitoring
of a 2-D carbon/carbon composite. Eng. Fract. Mech. 68, 1557–1573 (2001)
218. Maire, E., Carmona, V., Courbon, J., Ludwig, W.: Fast X-ray tomography and acoustic
emission study of damage in metals during continuous tensile tests. Acta Mater. 55,
6806–6815 (2007)
219. Bohse, J., Kroh, G.: Micromechanics and acoustic emission analysis of the failure process of
thermoplastic composites. J. Mater. Sci. 27, 298–306 (1992)
220. Haselbach, W., Lauke, B.: Acoustic emission of debonding between fibre and matrix to
evaluate local adhesion. Compos. Sci. Technol. 63, 2155–2162 (2003)
221. Bohse, J.: Damage analysis of polymer matrix composites by acoustic emission testing. In:
EWGAE 2004—26th European Conference on Acoustic Emission Testing, pp. 339–348.
Berlin, Germany (2004)
222. Cuadra, J., Vanniamparambil, P.A., Hazeli, K., Bartoli, I., Kontsos, A.: Damage quantifica-
tion in polymer composites using a hybrid NDT approach. Compos. Sci. Technol. 83, 11–21
(2013)
223. Unnthorsson, R., Runarsson, T., Jonsson, M.: Monitoring the evolution of individual AE
sources in cyclically loaded FRP composites. J. Acoust. Emiss. 25, 253–259 (2007)
224. Surgeon, M., Vanswijgenhoven, E., Wevers, M., Van Der Biest, O.: Acoustic emission during
tensile testing of SiC-fibre-reinforced BMAS glass-ceramic composites. Compos. Part A
Appl. Sci. Manuf. 28, 473–480 (1997)
225. Ramirez-Jimenez, C.R., Papadakis, N., Reynolds, N., Gan, T.H., Purnell, P., Pharaoh, M.:
Identification of failure modes in glass/polypropylene composites by means of the primary
frequency content of the acoustic emission event. Compos. Sci. Technol. 64, 1819–1827
(2004)
226. Narisawa, I., Oba, H.: An evaluation of acoustic emission from fibre-reinforced composites.
J. Mater. Sci. 19, 1777–1786 (1984)
227. Huguet, S., Godin, N., Gaertner, R., Salmon, L., Villard, D.: Use of acoustic emission to
identify damage modes in glass fibre reinforced polyester. Compos. Sci. Technol. 62,
1433–1444 (2002)
228. Anastassopoulos, A.A., Philippidis, T.P.: Clustering methodology for the evaluation of
acoustic emission from composites. J. Acoust. Emiss. 13, 11–21 (1995)
229. Giordano, M., Calabro, A., Esposito, C., D’Amore, A., Nicolais, L.: An acoustic-emission
characterization of the failure modes in polymer-composite materials. Compos. Sci. Technol.
58, 1923–1928 (1998)
References 55

230. Scholey, J.J., Wilcox, P.D., Wisnom, M.R., Friswell, M.I.: Quantitative experimental mea-
surements of matrix cracking and delamination using acoustic emission. Compos. Part A
Appl. Sci. Manuf. 41, 612–623 (2010)
231. Waller, J.M., Nichols, C.T., Wentzel, D.J., Saulsberry, R.L., Thompson, D.O., Chimenti, D.
E.: Use of modal acoustic emission to monitor damage progression in carbon fiber∕epoxy
composites. In: AIP Conference Proceedings, pp. 919–926. San Diego, (2011)
232. Pl€ockl, M., Sause, M.G.R., Scharringhausen, J., Horn, S.: Failure analysis of nol-ring
specimens by acoustic emission. In: 30th European Conference on Acoustic Emission,
pp. 1–12. Granada, Spain (2012)
233. Sause, M.G.R., M€uller, T., Horoschenkoff, A., Horn, S.: Quantification of failure mecha-
nisms in mode-I loading of fiber reinforced plastics utilizing acoustic emission analysis.
Compos. Sci. Technol. 72, 167–174 (2012)
234. Ono, K., Gallego, A.: Research and applications of AE on advanced composites. J Acoust.
Emiss. 30, 180–229 (2012)
235. Anastassopoulos, A., Tsimogiannis, A., Kouroussis, D.: Unsupervised classification of acous-
tic emission sources from aerial man lift devices. In: 15th World Conference on NDT, Roma,
Italy (2000)
236. Anastassopoulos, A.A., Kouroussis, D.A., Nikolaidis, V.N., Proust, A., Dutton, A.G., Blanch,
M.J., Jones, L.E., Vionis, P., Lekou, D.J., van Delft, D.R. V, Joosse, P.A., Philippidis, T.P.,
Kossivas, T., Fernando, G.: Structural integrity evaluation of wind turbine blades using
pattern recognition analysis on acoustic emission data. In: Proceedings of the 25th
European Conference on Acoustic Emission Testing. Prague, Czech Republic (2002)
237. Grosse, C.U., Ohtsu, M.: Acoustic Emission Testing. Springer, Berlin (2008)
238. Weihnacht, B., Schulze, E., Frankenstein, B.: Acoustic emission analysis in the dynamic
fatigue testing of fiber composite components. In: 31st Conference of the European Working
Group on Acoustic Emission, pp. 1–8. Dresden, Germany (2014)
239. Abraham, A.R.A., Johnson, K.L., Nichols, C.T., Saulsberry, R.L., Waller, J.M.: Use of
statistical analysis of acoustic emission data on carbon-epoxy COPV materials-of-construc-
tion for enhanced felicity ratio onset determination—JSC-CN-26080. (2011)
240. Downs, K.S., Hamstad, M.A.: Acoustic emission from depressurization to detect/evaluate
significance of impact damage to graphite/epoxy pressure vessels. J. Compos. Mater. 32,
258–307 (1998)
Chapter 3
Digital Image Correlation

In order to detect the occurrence of failure in a material, visual observation is a very


intuitive approach. Although global failure is many times easier to observe visually,
more dedicated analysis is required to detect small changes indicating the occur-
rence of first failure. One specific type of visual observation follows the idea to
track and measure changes between subsequently acquired images of an object. The
concept used for that purpose is based on digital image acquisition of a moving and
deforming object and is known as digital image correlation (DIC). The background
of this technique stems from the basic principles of optics, perspective imaging, and
digital image analysis.
The aim of this chapter is not to cover all aspects of the topic of DIC, but to
introduce the key concepts and point out the relevant principles for measurement of
strain fields of solid materials. For the interested reader, a comprehensive review on
the topic is given by Sutton, Orteu, and Schreier [1] and a concise review on the
application of 2D-DIC is also found in [2].
One of the first attempts to use image correlation as a measurement technique to
register features in images of different views was made by Hobrough in the 1950s.
A few years later in 1961 he also developed a system to perform first quantitative
positional measurements extracted from an image correlation process. With the
availability of digitized images as provided by CMOS or CCD cameras in the 1960s
and 1970s until the present days the branch of vision-based algorithms with a focus
on DIC techniques has become a growing field of research. The development of the
algorithms of the DIC method is driven by the accompanying applications in
material science. Since the first experimental application in materials science by
Peters and Ranson in 1982 [3], many experimental studies in the field of fiber
reinforced composites were accompanied by DIC as a measurement technique
[4–12]. In recent decades, suitable approaches have been developed to measure
deformations with high magnification (e.g., inside a scanning electron microscope)
and high-speed camera systems have been used to allow digital image correlation
during high strain rate testing [13, 14].

© Springer International Publishing Switzerland 2016 57


M.G.R. Sause, In Situ Monitoring of Fiber-Reinforced Composites,
Springer Series in Materials Science 242, DOI 10.1007/978-3-319-30954-5_3
58 3 Digital Image Correlation

This chapter starts with a description of the mathematical and physical principles
used by 2D-DIC and 3D-DIC. Focus is given on the parameters that limit detect-
ability for state-of-the-art equipment and comparison is made between theoretical
predictions and measurements under laboratory conditions. The second part of the
chapter has its focus on the theoretical description of strain concentration due to
internal flaws, especially for those types found in fiber reinforced composites.
Measurements with a variety of defect types are compared to calculations using
finite element modeling. Detectability of failure mechanisms in fiber reinforced
composites is discussed as function of failure type, failure size, and depth position
below surface. The last section elaborates some typical applications of DIC mea-
surements in conjunction with testing of fiber reinforced composites.

3.1 Principle of Operation

The idea of DIC techniques is to yield a quantitative measure of motion, rotation,


and deformation occurring between two subsequently acquired images of an object
as seen in Fig. 3.1.
In general, it is not possible to find the correspondence of one pixel in a
subsequent image, since the gray value of the single pixel can be found at many
other pixel positions in the subsequent image too. Therefore, there is no unique
correspondence of one single pixel position in subsequently acquired images.
Instead, a small neighborhood around the pixel of interest is considered to form a
more characteristic pattern to detect in the subsequent images (see Fig. 3.1). This
small neighborhood is usually referred to as subset, window, or facet. However, if
repeating structures such as grids of points, lines, or otherwise regular patterns are
used, there is still no unique correspondence between the images. This matching
problem is known as the correspondence problem. One effective way to circumvent
this lack of unique correspondence is to use a non-periodic, stochastic pattern as the
speckle pattern shown in Fig. 3.1. In order to allow for comparable spatial

rotation deformation

displacement small neighborhood (subset)

Fig. 3.1 Tracking deformation states between images and speckle pattern including small neigh-
borhoods (subset) used for image matching process
3.1 Principle of Operation 59

resolution within the image a narrow size distribution of the image features (e.g.,
black dots on white background) is required.
For the matching process between two images, various algorithms are applied in
practice [1, 15]. In the following, the Lucas-Kanade tracker algorithm [16] is used
as an example to demonstrate the basic principles of the digital image correlation
process. Using the small neighborhoods around the pixel of interest, a subset of the
full image is defined. The intensity distribution of the reference subset is denoted as
e ðrÞ and the intensity distribution of the subset after displacement by G
F e ðrÞ. Now,
the goal is to find the 2D-displacement vector uopt between these two subsets, which
can be done by minimization of the sum of squares deviation:
X  2
uopt ¼ min G e ðrÞ
e ð r þ uÞ  F ð3:1Þ

Such optimization routines are carried out using a cost function χ written as first-
order Taylor series
 2
X  ∂ e ∂ e 
χ 2 ðux þ Δx , uy þ Δy Þ ¼ e þ uÞ 
Gðr
G
Δx 
G e 
Δy  FðrÞ ð3:2Þ
 ∂x ∂y 

This starts with the present estimates ux and uy for the initial 2D-displacement
vector and the incremental motion updates Δx and Δy in the current iteration.
Taking partials of (3.2) with respect to Δx and Δy and setting them to zero results
in a linear equation system being solved incrementally
0 !2 11
X ∂Ge X e e 0 1
X ∂G 
  BB
∂G ∂G C
∂x ∂y C e
eG
Δx B ∂x C B ∂x 
F C
¼B !2 C B @ X ∂G
C
A ð3:3Þ
Δy B X X ∂Ge C e
eG
@ e ∂e
∂G G A F
∂x ∂y ∂y
∂y

This optimization procedure iteratively refines the estimate for uopt in p iterations
using upþ1 ¼ up þ Δ until convergence is achieved. This algorithm can suitably
detect arbitrarily large 2D-displacements as long as the initial estimate is within the
convergence radius of the method. However, in many applications the image
matching process needs to account for high distortions, such as additional rotations
of the image. This is taken into account using shape functions ξ(r, p) that transform
pixel coordinates in the reference subset into the coordinates of the deformed
subset. This requires modification of the cost function of (3.2) in the following way
X  2
χ 2 ð pÞ ¼ ~ ξðr, pÞ  F
ðG ~ ðrÞ ð3:4Þ
60 3 Digital Image Correlation

Optimization is carried out for the parameter vector p of the shape function. Taking
derivatives of (3.4) with respect to all parameters of p for the affine shape functions,
the parameter update can be written as

Δp ¼ H1 q ð3:5Þ

Using the symmetric Hessian matrix


0X 2 X X X X X 1
e
G e xG
G ey e 2x
G e 2y
G e xG
G e yx e xG
G e yy
B
x X X x X x X X
B e2
G e xG
G e yx e xG
G e yy e 2x
G e 2y C
G C
B y X X X y X y C
B e x2 2 e xy
2 e xG
e yx 2 e y xy C
e xG
B G x X
G x X
G
X
G C
H¼B
B Ge 2 y2 e xG
G e y xy G e y y2 C
e xG C
B x X X C
B e 2 2
G y xy C
e 2
@ Gy x
X A
e 2 y2
G y

ð3:6Þ

and
0 X  1
ex F
G e
eG
B X  C
B ey F e
eG C
B G C
B X  C
B C
B e xx F
G e
eG C
B  C
q¼ B X C ð3:7Þ
B e xy F
G e
eG C
B  C
B X C
B e yx F e
eG C
B G C
@ X  A
e yy F
G e
eG

denoting partial differentiations by subscripts, e.g., Ge x ¼ ∂e


G
.
∂x
 
The coordinates (xi, yj) and (xi , yj ) are related by the deformation that occurs
between the two images. If the motion is perpendicular to the optical axis of the
camera, then the relation between (xi, yj) and (xi , yj ) can be approximated by a
shape function resembling a 2D affine transformation:

∂u ∂u
x* ¼ x þ u þ Δx þ Δy ð3:8aÞ
∂x ∂y
∂v ∂v
y* ¼ y þ v þ Δx þ Δy ð3:8bÞ
∂x ∂y

Here u and v are translations of the center of the sub-image in the x and y direction,
respectively. The distances from the center of the sub-image to the point (x, y) are
3.1 Principle of Operation 61

denoted by Δx and Δy, and ∂u , ∂u , ∂v , ∂v    are the displacement gradients. The


∂x ∂y ∂x ∂y
latter are straightforward to be interpreted as components of the local strain tensor.
If evaluated relative to the global x- and y-direction the displacement gradients can
readily be converted to the axial and transverse strain components with respect to
the axial load axis of a test specimen. Another usual conversion of the local strain
tensor is to calculate a principal component analysis to obtain the first, second, and
third1 principal strain values. The principal strain values have been found to be
particularly useful to detect strain anomalies (see Sect. 3.3), whereas for homoge-
neous strain fields the first principal strain value simply coincides with the axial
strain value (if this is the dominant deformation direction) and the second principal
strain coincides with the transverse strain value, respectively.
Additional considerations are required if offsets and scaling of the lighting
conditions occur during acquisition of the images. In practical applications these
effects will always be present to some extent and may falsify the matching process
due to systematic shifts or scales of the recorded gray values. As has been shown
in [1] such effects can be compensated using a modified gray value distribution
e0 ¼ e
G aGe þe b introducing an offset e b and a scale parameter e a . Consequently, the
cost function formulated in (3.4) is modified to yield
X 2
χ2 ¼ e
aGei þ e ei
bF ð3:9Þ

Therefore the iterative minimization procedure has to account for two additional
a and e
parameters e b. Optimal estimators of these two parameters can be derived from
two equations

∂χ 2 X
¼2 e
aGei þ ebF ei G ei ð3:10aÞ
∂ea
X
∂χ 2 ei  e
ðF bÞGei
¼0!e a opt ¼ X
∂ea Ge2
i

∂χ 2 X 
¼2 e
aG ei þ ebF ei ð3:10bÞ
∂eb
X
∂χ 2 Fei  e ei
aG
¼0!e b opt ¼ X
∂e
b 1
D E D E D E D E
Using ei ¼ F
F e and G
ei  F ei ¼ G e one can write the optimal
ei  G
parameter estimates as

1
The third principal strain component is only applicable for 3D-DIC measurements.
62 3 Digital Image Correlation

X D ED E
ei G
F ei
a opt ¼ X
e ð3:11aÞ
hGi i2
X D ED E
D E D E ei G
F ei
e e e
b opt ¼ F  F G e ð3:11bÞ
X D E2
ei
G

which allows for writing the cost functional solely in terms of the gray value
e
e and G
distributions F
00X D ED E X D ED E1 12
X BB e
F e
G D E e
F e
G i C D E
e C
i i i
χ2 ¼ @@ X ei  Ge e
2
G X D E2 A  F i þ F A ð3:12Þ
hG i i Ge i

Until this point it was assumed that the object shows only negligible out-of-plane
motion between subsequent images. Using a single camera system noticeable out-
of-plane motion of the object with respect to the camera will introduce virtual
displacement gradients. These will corrupt the in-plane measurements and may
make it difficult or impossible to separate the true deformations and “pseudo image
deformations” introduced by the out-of-plane motion. As shown in [17] a relative
motion away from the image plane decreases image magnification and introduces a
negative normal strain. In [17] it was demonstrated that measurements by a regular
2D-DIC system result in a relationship of the normal strain ε to the out-of-plane
motion w in a linear fashion resulting in a proportionality of ε ¼ 1524  106  w.
In the typical 2D-DIC configuration only one camera is used to acquire images
of the object. Such a single camera performs a perspective transformation of a 3D
object point to a 2D image point. Therefore, information of the third dimension is
irreversibly reduced during this operation. As shown in Fig. 3.2 this causes two
points O1 and O2 being projected to the same image point P1 ¼ P2 . In contrast, the
usage of two camera systems with different viewing angles as shown in Fig. 3.2
allows for recovery of the 3D position of the object point.
Using such a stereo vision system the measurement error due to out-of-plane
motion can almost be compensated. Experiments for a 3D-DIC system demon-
strated the measured strain error to be linear with the out-of-plane motion by a
factor of only 1:31  106 [17]. This is four orders of magnitude lower than for the
recordings using a 2D-DIC system.
However, for such stereo matching processes, further modifications to (3.10a)
and (3.10b) are required, since the nonlinearity of perspective projection does
not allow usage of affine transformations even if the object is a plane (for details
see [1]).
3.2 System Accuracy 63

02 02
01
01

P2
P1 = P2 P1 = P2
P1

COOS COOS
COOS‘
2D-DIC 3D-DIC

Fig. 3.2 Camera configuration as usually used in 2D-DIC (a) and stereo vision process to measure
3D object positions used in 3D-DIC (b)

time

discretization

min χ2
object lens CCD camera

Fig. 3.3 Acquisition chain of DIC comprising steps of optical imaging using lens system, digital
image acquisition, and image correlation

3.2 System Accuracy

For the purpose of materials testing, the extraction of quantitative displacement and
strain values out of the acquired images is of largest practical interest. Therefore
concise knowledge of the measurement accuracy is required. In this context one can
distinguish between several factors, which influence the finally obtained measure-
ment values. As seen in Fig. 3.3, one part deals with the measurement errors
introduced by the optics of the measurement system. Lens distortions such as
spherical aberration, coma distortion, astigmatism, curvature of field, and linear
distortion are introduced by the system optics used in the image acquisition process.
However, these effects can be sufficiently compensated using accurate calibration
procedures to map the object coordinates and the image coordinates [1]. Descrip-
tions of these effects are extensively covered in standard literature on optics
[18, 19]. In the following, the discussion is thus focused on the measurement errors
arising due to the digital image acquisition, due to the image correlation process and
the computations in the stereo-vision process.
64 3 Digital Image Correlation

3.2.1 Error Sources

The DIC method is known to measure displacements with sub-pixel accuracy and
lateral surface strains in the range <103. There are several sources influencing the
system resolution, which can be due to intrinsic noise of the acquired images,
statistical and systematical errors introduced by the system calibration, sub-pixel
effects resulting from the camera resolution [1, 20–22] or due to intrinsic uncer-
tainties of the image correlation algorithm.
As first category, the correlation errors describe uncertainties of the correlation
of corresponding subset positions in subsequent images. These correlation errors
can be further divided in systematic and random errors. A second category arises for
the usage of 3D-DIC methods. Here, the 3D-coordinates reconstruction method
introduces further error sources for the finally computed values. Although the
displacement and strain values are the quantities of interest for the purpose of
materials testing, the primary measurement value in DIC is the pixel (px). The link
between these pixel shifts and the computed displacement and strain values will
thus be established in Sect. 3.2.2.

3.2.1.1 Systematic Errors

Systematical errors are caused by sub-pixel effects, which occur because of the
discretization of the speckle pattern by the CCD camera pixels. In order to obtain
sub-pixel accuracy of the image correlation, the cost function χ 2 has to be evaluated
at non-integer locations of the pixel array. Therefore, gray values have to be
interpolated between the sample points (pixels). As interpolation is a method of
approximating a value between two samples, the use of gray value interpolation in
the matching algorithm can be expected to introduce errors. Another factor likely to
introduce errors in the sub-pixel image correlation is the highly nonlinear distor-
tions of the subsets.
As has been investigated in detail in [1, 21, 23] there are three main factors
considered as sources for systematic errors:
• Image noise level
• Choice of subset size
• Choice of speckle pattern size
The typical procedure to assess the influence of these parameters is the evalu-
ation of synthetic or experimental speckle patterns. These speckle patterns are
moved in one direction by a known value resulting in a sub-pixel displacement.
Therefore, the reference displacement is known and the mismatch of the values
calculated by the DIC method can be compared relative to that reference.
A qualitative result of such an investigation is shown in Fig. 3.4. As reported
previously [1, 21, 23, 24], there is a typical sinusoidal fluctuation of the matching
error, which is common to most of the DIC algorithms. For most of the DIC
3.2 System Accuracy 65

0.01
systematic error [px] algorithm A
algorithm B

0.00
0.0 0.5 1.0 pixel shift

–0.01

Fig. 3.4 Schema representation of the systematic matching error as function of the DIC matching
algorithm. Absolute values of pixel mismatch and detailed behavior depend on the respective
algorithm (see [21, 23] for details)

0.15
without noise
systematic error [px]

with noise

0.00
0.0 0.5 1.0 pixel shift

–0.15

Fig. 3.5 Schema representation of the systematic matching error as function of noise present in
the images. Absolute values of pixel mismatch and detailed behavior depend on the respective
algorithm (see [21, 23] for details)

algorithms investigated the matching error approaches zero at the integer positions,
i.e., at 0 and 1 px displacement. Some of the algorithms also approach zero
matching error at 0.5 px but others tend to fluctuate without noticeable structure.
Therefore, Fig. 3.4 is only understood as qualitative representation, whereas the
quantitative values of the matching error depend on the respective DIC algorithm
and on the subset shape functions used therein.
For random noise added to the image, the overall performance of the matching
process starts to decrease. As has been shown by [21, 23], there are some DIC
algorithms, which are almost unaffected by the presence of random noise in the
images, while other algorithms show almost linear dependence on random noise
superimposed to a displaced speckle pattern (cf. Fig. 3.5). For the typical noise
values as expected in real measurements a broad range of matching errors was
found. Whilst the best algorithm was found to be constantly below 0.005 px
mismatch, some of the algorithms result in maximum values of up to 0.25 px
mismatch.
Similar effects have been found for the subset sizes. Most of the algorithms yield
almost no characteristics in the mismatch error as function of different subset sizes
66 3 Digital Image Correlation

(e.g., spanning a typical range from 6 to 32 px). However, some algorithms yield
pronounced dependence on the subset size, potentially causing a magnification of
the mismatch error by a factor of four for a decreasing subset size.
In general, the same effect could be expected for the speckle pattern size. Since
the subset is the discretized version of the real speckle pattern, both size choices can
have a similar impact. Therefore it is more convenient to investigate the effect of
the speckle pattern normalized by the subset size to assess the effects of this
discretization. The findings of [21, 23] can be compared to those derived from the
expression of the systematic error developed by [22]. In their expression,
the systematic error is calculated taking into account the gray value gradients
of the initial and translated (or deformed) images as well as the noise level.
Hence, there is a common tendency of most of the algorithms to increase their
sensitivity to the presence of noise, when coarse speckle patterns are used.

3.2.1.2 Random Errors

Random matching errors occur because of the limited number of pixels and
corresponding gray values in each subset, and the sub-pixel matching process.
The corresponding statistical error decreases with the square root of the number
of subset pixels used for the matching process. Additional random error sources are
statistical noise of the gray values, different illumination settings for the two
cameras, image contrast, and size of the speckle pattern on the specimen surface.
Such random matching errors can often be compensated by smoothing operations or
by averaging of the acquired images.
The random matching error is evaluated using the same procedures as the
systematic matching error. The main difference is the evaluation of the error
quantity. While the systematic matching error is understood as global displacement
error of the image correlation process, the random matching error is understood as
standard deviation of all of the individual subset results.
As demonstrated by [21, 23], the mean random matching error generally
decreases with the subset size and increases with the noise level as schematically
seen in Fig. 3.6. Roux and Hild [25] have proposed an analytical expression of the
random matching error showing that it scales linearly with image noise and pixel
size and with the inverse of the mean square gradient and subset size.
However, based on the different implementations of DIC algorithms, some of
the algorithms show a mean random matching error independent of the speckle size.
As for the systematic matching error, the random matching error is a function of the
interpolation process which generates individual gray value gradients. Therefore,
each algorithm tends to show unique sensitivity on these factors, causing intrinsic
complexity to the prediction of the random matching error.
3.2 System Accuracy 67

0.10 algorithm A – small Speckle pattern


mean random error [px] algorithm A – larger Speckle pattern
algorithm B – small Speckle pattern
algorithm B – larger Speckle pattern

0.05

0.00
0.0 0.5 1.0
noise level [a.u.]

Fig. 3.6 Schema representation of the mean random matching error as function of the speckle
pattern size and the noise level. Absolute values of pixel mismatch and detailed behavior depend
on the respective algorithm (see [21, 23] for details)

3.2.1.3 Stereo Vision

As indicated in Fig. 3.2, the 3D-DIC approach uses the stereo vision concept to
derive the 3D position of an object. The errors introduced by this matching
procedure add to the errors discussed for the image matching process. Here, the
uncertainties of the calibration parameters translate into errors, when reconstructing
the 3D coordinates from correlated subsets of the two cameras.
Calibration errors appear in a systematical manner as a function of the subset
positions in the camera images, causing local distortions of the reconstructed 3D
space. In order to get an understanding of the impact on the evaluated data, the
distortion effects can be described by the following approximation:
If P ¼ ðx, y, zÞ is the real position of a reconstructed point, the measured
coordinate Pexp will differ from P by a deviation vector ΔP. In the vicinity of a
point P0 ¼ ðx0 , y0 , z0 Þ the deviation vector can be approximated using the distor-
tion matrix K:
   
ΔP P0 þ r ¼ ΔP P0 þ K  r ð3:13Þ

The evaluation of the components of the distortion matrix K and the known
displacement r thus allows the calculation of the deviation vector ΔP associated
with the error in reconstruction of the 3D position.

3.2.2 Resolution

For the final interpretation of measured displacement and strain values, the resolu-
tion of the imaging system has to be considered. Based on the principles of optics,
there is a straightforward derivation of the pixel resolution. As seen for the simple
68 3 Digital Image Correlation

Fig. 3.7 Principle of S1 S2


pinhole camera establishing
CCD pixels
relation between size of
object and single pixel size
on camera system
optical axis

object lens projection

1st principal strain [%]

15 mm strain
anomaly

20 px× 15 px 15 px× 10 px 10 px× 08 px 08px× 06 px 05px× 04 px


size × distance

Fig. 3.8 Example of strain field calculated for identical images with 0.028 mm/px resolution
(upper row) and different subset window sizes in magnified section of speckle pattern (lower row)

example of a pinhole camera in Fig. 3.7, the measured object is projected to the
detector using a lens system. Accordingly, the system resolution is given by the size
of the object, divided by the number of pixels in the used array. This assumes, the
projected image is within the depth of field of the camera system (i.e., is projected
without blurring) and the whole configuration is calibrated to compensate for the
effect of lens distortions.
This theoretical system resolution is thus given as pixels per unit length and is
very dependent on the selected configuration. Choosing lens systems with different
magnification factors and adjusting the distance to the object can be used to
optimize this value. Therefore, the measured displacement value is directly
obtained from the change in pixels times the resolution. Similarly, the measured
strain values are usually derived from the change in pixels relative to the local
reference length. Therefore, the latter sometimes is more tolerable to the occurrence
of distortions, since the displacement and the reference length are affected by such
distortions in the same way.
Although the DIC systems provide sub-pixel accuracy as discussed in
Sect. 3.2.1, this does not necessarily allow interpretation of strain values at this
length scale. Beyond the single pixel resolution, the limiting factor is the choice of
subset sizes. As seen in Fig. 3.8 with 0.028 mm/px, a certain trade-off between the
3.3 Strain Concentration 69

increasing spatial resolution and the increasing matching errors has to be achieved.
Here the same image data was processed using different subset window sizes.
Beyond the effects of interaction between the subset size and the speckle pattern
size, a clear trend is observed. For larger subset sizes no anomalies are visible in the
strain field. With decreasing subset size, such strain anomalies appear and tend to
grow sharper and more detailed, which is caused by the increased spatial resolution
of the strain field as provided by the increased amount of subset windows. However,
for very low subset window sizes (such as 5 px) the increasing matching errors start
to dominate the appearance of the strain field. In addition to the increased error of
the absolute values this also prevents the detection of such strain anomalies and
poses further challenges to overcome when detecting the occurrence of failure in
fiber reinforced materials.
As a rough estimate of current systems capabilities, for a typical commercial
3D-DIC system the respective accuracy as function of different lens systems was
evaluated by [26]. Taking into account the error sources introduced in Sect. 3.2.1 it
was observed that displacement errors are present on the order of less than 0.02 px,
translating into strain errors being limited to 0.05  103, when using a lens with
50 mm focal length, or 0.2  103 strain error for a 17 mm lens for the selected field
of view. If the displacement between images is small (less than 50 px), the
respective errors were found to scale linearly. For a wide angle lens system with
a focal length of 4.8 mm, distortions introduced by the camera could not be
sufficiently compensated by the calibration routine.

3.3 Strain Concentration

Based on the history of solid mechanics, the presence of geometric discontinuities


is usually associated with the concept of stress concentration. As seen in Fig. 3.9,
the presence of a hole causes a change of stress flux in the residual cross section
when subject to a tensile load. This is a well-known concept in structural mechanics
and an established routine for design of structural components [27, 28]. The same
principle holds valid for the presence of internal defects such as cracks. These
cracks cause a stress concentration at the crack tip and may propagate after
exceeding a critical stress value. For typical engineering materials like metals,
polymers, or ceramics the critical size of these cracks and the critical stress values
can directly be obtained using the concepts of fracture mechanics [29–31]. Due
to the complexity of the microstructure and the shape of the cracks, similar
approaches for fiber reinforced composites are still subject to recent research efforts
[27, 32–34].
The maximum stress near a crack occurs in the area of lowest radius of
curvature. For the simple case of an elliptical crack of length 2a and width 2b
(cf. Fig. 3.9) for an applied external stress σ, the stress at the ends of the major axes
of the crack is given by:
70 3 Digital Image Correlation

a b
load load

crack

2b

2a 2a

Fig. 3.9 Schema of stress flux around an elliptical hole (a) and stress flux around a crack (b)

  rffiffiffiffi
a a
σ max ¼ σ 1 þ 2 ¼ σ 1 þ 2 ð3:14Þ
b rc

This uses rc as radius of curvature of the crack tip. The so-called stress concentra-
tion factor is the ratio of the highest stress (σ max) to a reference stress (σ) of the
gross cross section. As the radius of curvature approaches zero, the maximum stress
approaches infinity. It is worth noting that the stress concentration factor is only a
function of the shape of a crack but not of the absolute crack length.
Based on the stress–strain relation implied by Hooke’s law, there is an expected
relation between the stress concentration and the strain concentration. Although
linear elastic material behavior does not hold valid for the region close the crack tip,
there is still a distinct proportionality between the values of stress concentration and
strain concentration as discussed in detail for the case of fiber reinforced materials
by [27, 32–34].
Various experimental methods have been applied to derive stress or strain
concentration, such as photoelastic stress analysis, shearography, strain gages, or
digital image correlation. On the theoretical side, many analytical formulations
have been provided for design of structures, most of them going back to [28]. How-
ever, for the complexity faced in composite materials in terms of anisotropy,
boundary conditions, and heterogeneity, numerical methods based on finite element
modeling have been recently proposed [33, 34].
Combining these simple considerations on strain concentration and the avail-
ability of high resolution cameras for full-field strain measurements, it is feasible to
evaluate the capabilities of DIC systems to detect strain concentrations due to
typical failure types in fiber reinforced composites. This is motivated by recent
literature indicating strain concentration due to the microstructure of the material
[35] or the formation of internal cracks [36].
3.3 Strain Concentration 71

However, for the existence of strain concentrations at a certain load level, there
are several options how these indications may relate to the formation of (micro-)
cracks on the interior of the material. All kinds of defects, such as voids, particle
inclusions, or fiber disorientations, are well known to act as crack initiation sites.
The presence of these defects will cause a local disturbance of the strain field due to
the presence of geometric discontinuities (voids) or due to the locally different
stiffness properties (fiber orientation, inclusions).
In all cases, after crack initiation there is a geometrical discontinuity located in
the composite, which is given by the presence of the crack itself. Accordingly, the
strain field is affected by the presence of this discontinuity. The effect of the latter
will substantially depend on the size of the crack and the type of stiffness reduction
(e.g., fiber failure vs. inter-fiber failure).
In order to estimate the visibility of such defects in full-field strain measure-
ments, Sects. 3.3.1 and 3.3.2 present results of recent studies used to assess the
applicability of presently available high resolution DIC systems [37, 38]. For these
first studies, focus is given to the typical specimen dimensions used in materials
testing, spanning a range of several centimeters in size.

3.3.1 Measurement of Strain Concentration Due


to Internal Defects

For the experimental part of the study tensile specimens are prepared, which are
either free of defects (reference) or comprise certain model defects in various depth
positions below the surface. The respective specimen dimensions are shown in
Fig. 3.17 including the position of the speckle pattern. As material system a Sigrafil
CE1250-230-39 prepreg system was used, which was cured according to the
material supplier’s specifications. All experiments are carried out by calibrated
3D-DIC measurements using an ARAMIS 12 M system applying the acquisition
parameters reported in Tables 3.1 and 3.2. A photographic representation of a
typical experimental configuration is shown in Fig. 3.10. The tensile specimens
are subject to a tensile load using a universal testing machine with a cross-head
velocity of 1 mm/min.
To assess the limits of detectability for specific failure mechanisms in compos-
ites it is necessary to prepare defects of a known size at a certain depth of the
laminate. Since this is impractical to be done by using mechanical experiments (i.e.,
the size and position of a defect is arbitrary in such a case), artificial defects are
prepared to represent the existence of real defects. To this end, defects are embed-
ded in the specimens during the fabrication process.
To fabricate an artificial fiber breakage one ply of the stacking sequence is cut
for the full thickness for a certain length. As seen in the computed tomography
image in Fig. 3.11 after curing of the laminate, the fiber cuts are clearly visible and
accurate dimensional measurements are possible to obtain the real depth position
72 3 Digital Image Correlation

Table 3.1 Acquisition settings of DIC system used for tensile tests to detect strain concentration
effects
Measurement setup Value
Camera Toshiba CMOS camera CSC12M25BMP19-01B with 4096  3072 pixels
Lenses Titanar, focal length 5.6/100 mm
Lighting KSP 0495-0001A LED 20 W/24 V white 30
Filter Polarization-filter Schneider-Kreuznach
Field of view (55.9  41.7) mm2
Scale factor 0.0136 mm/px
Subset size 21  21 pixel
Acquisition rate 2 Hz
Software ARAMIS
Configuration 3D

Table 3.2 Acquisition settings of DIC system for tensile tests on unidirectional fiber reinforced
composites
Measurement setup Value
Camera Toshiba CMOS camera CSC12M25BMP19-01B with 4096  3072 pixels
Lenses Titanar, focal length 2.8/50 mm
Lighting KSP 0495-0001A LED 20 W/24 V white 30
Filter Polarization-filter Schneider-Kreuznach
Field of view (113.4  86.7) mm2
Scale factor 0.028 mm/px
Subset size 19  19 pixel
Acquisition rate 2 Hz
Software ARAMIS
Configuration 3D

and size of the artificial defect. In the example the nominal size of the fiber cuts was
chosen to be 10 mm in length and one ply thickness of 0.22 mm. However, the cut is
not always perfectly oriented inside the laminate, i.e., it develops an angle relative
to the z-axis as seen from the right image in Fig. 3.11.
Also, the crack front is not exactly linear, but tends to deviate slightly from an
ideal line as seen from the xy-cross section in the left image (this shows a cross
section inside the laminate, i.e., the top-surface seen in the left image is not the
outer surface of the laminate).
To fabricate an artificial delamination, inserts of thin Ethylene
tetrafluoroethylene (ETFE) sheets are used. In the present case two ETFE sheets
of 10 μm thickness are molded together at the edges using a hot wire processing
technique. This yields ETFE balloons of well-defined cross-sectional dimensions
filled with air. These balloons are embedded in the laminate during fabrication and
are easy to detect in computed tomography images of the cured laminate
(cf. Fig. 3.12). In the left image the yz-cross section reveals the width of the
delamination (nominally 10 mm) and the wavy orientation of the embedded balloon
3.3 Strain Concentration 73

load axis

averaging
area

camera 1
specimen

x
camera 2
y
LED light

Fig. 3.10 Typical experimental setup and specimen dimensions including speckle pattern posi-
tion and area marked by dashed line as used to obtain mean strain values

z fiber breakage
x
y

z
x y

Fig. 3.11 3D-slices of computed tomography images of artificially prepared fiber breakage in a
unidirectional laminate
74 3 Digital Image Correlation

z
x delamination
y

z
x y

Fig. 3.12 3D-slices of computed tomography images of artificially prepared inter-ply delamina-
tion in a unidirectional laminate

along the z-direction. This is a consequence of the material contraction during


laminate curing and thus cannot be avoided using this fabrication process. How-
ever, the orientation in parallel to the fiber axis is nearly a straight line as seen from
the xz-cross section in the lower right figure. Here the nominal length of the
delamination was also chosen to be 10 mm, yielding a cross section of the prepared
inter-ply delamination of 100 mm2.
Comparatively challenging is the fabrication of artificial defects resembling
inter-fiber cracks. Using 50 μm thin Polyimide (PI) stripes it is possible to embed
an out-of-plane discontinuity as seen in Fig. 3.13. As seen from the xy-cross section
in the left image, the nominal length of this inter-fiber crack was chosen to be
10 mm, which can be implemented rather accurately. However, the angle of this
discontinuity in the yz-plane is hard to control during laminate fabrication. For the
example shown in Fig. 3.13, the resulting angle was evaluated to be approximately
61 as seen from the yz-cross section in the right image.
It was found to be particularly challenging to fabricate different angles in a well-
defined orientation. In a real composite, the occurrence of inter-fiber cracks at other
angles than 90 is readily expected from failure theory and experimental observa-
tions. Accordingly, the orientation will also cause distinct changes with respect to
detectability of such inter-fiber cracks as demonstrated in Sect. 3.3.2.
Based on these three representative examples of artificially prepared defects, a
systematic study was conducted. In all cases investigated, the artificially prepared
defects are meant to represent an already existing crack surface. Therefore, the
values discussed in the following are understood as strain concentration effects due
to the existence of these defects, not due to the formation of these defects. The
implication of this assumption will be discussed in Sect. 3.3.3.
For each failure type artificial defects of different size and depth position were
prepared. The systematic aspects of this study will be discussed in Sect. 3.3.2. As a
3.3 Strain Concentration 75

inter-fiber crack
z
x
y

θ = 61°
1.7 mm

z
x
inter-fiber crack
y

Fig. 3.13 3D-slices of computed tomography images of artificially prepared inter-fiber crack at
61 angle in a unidirectional laminate

representative example of the resultant experimental measurements, the case of


fiber breakage embedded in a unidirectional laminate will be presented in the
following.
As seen from the false-color representation of Fig. 3.14, the presence of an
artificial fiber breakage within the laminate causes a readily visible strain concen-
tration at its position. All examples shown in Fig. 3.14 are shown for a mean
(global) strain value of 0.42 % as indicated by the false-color histogram of the x-
strain (in parallel to load axis). Each measurement represents a different length of
fiber breakage starting at nominally 10 mm (Fig. 3.14a) and decreasing to 7.5 mm
(Fig. 3.14b), 5.0 mm (Fig. 3.14c), and 2.5 mm (Fig. 3.14d). All fiber breakages were
located at the second ply (i.e., 0.22 mm below the surface). For each measurement
the strain values above and below the crack position exceed the surrounding
(global) strain value by 0.10–0.27 %. As seen in the figures, there is a distinct
relationship between the size of the crack and the height of the strain value
exceeding the background value. As expected from (3.14), smaller defect dimen-
sions cause smaller strain concentration.
76 3 Digital Image Correlation

x-strain x-strain
a b
[%] [%]
0.96 0.96
0.69 % 0.61 %
0.79 0.79

0.62 0.62
10 mm 7.5 mm

0.45 0.45

0.28 x 0.28

y
0.11 0.11

x-strain x-strain
c [%] d [%]
0.96 0.96
0.67 % 0.52 %
0.79 0.79

5 mm 0.62 2.5 mm 0.62

0.45 0.45

0.28 0.28

0.11 0.11

Fig. 3.14 Typical measurement result for artificial fiber breakage in a unidirectional laminate
located at 0.22 mm below surface (2nd ply) at 0.42 % mean strain and changing spatial size
ranging from 10 mm (a) to 7.5 mm (b), 5.0 mm (c), and 2.5 mm (d)

Keeping the spatial dimensions constant at 10 mm size, the depth position of the
fiber breakage is now systematically varied. As seen from the measured strain fields
of Fig. 3.15 at a constant mean strain value of 0.42 %, this causes a significant drop
in the measured strain concentration when moving from near surface positions to
deeper within the laminate. For convenience, the depth position of the artificial fiber
breakage is noted in terms of the cut ply assuming an individual ply thickness of
0.22 mm. Thus the 2nd ply is located 0.22 mm below the surface, the fiber breakage
itself spanning the full extent of the second ply, i.e., ranging to 0.44 mm. As
indicated in Fig. 3.11, these values constitute only nominal values, but are good
enough to discuss the overall trend. Evaluating the images in Fig. 3.15 according to
(3.15), the strain concentration value at the defect position was found to exceed the
mean value by 0.27 % for the 2nd ply, by 0.23 % for the 3rd ply, and by 0.08 % for
the 4th ply position depending on the exact position used for the strain quantifica-
tion. The reason for this decay will be discussed using the respective modeling
results in Sect. 3.3.2. With respect to the fluctuation of strain values in the
surrounding area, the question arises if the measured strain concentration is signif-
icant relative to the background noise. This requires a discussion of the background
3.3 Strain Concentration 77

x-strain x-strain
a b
[%] [%]
0.69 % 0.96 0.96
0.55 %

0.79 0.79

0.62 0.62
10 mm

0.45 0.45

0.28 0.28

2nd ply 3rd ply


0.11 0.11

x-strain
c
[%]
0.50 % 0.96

0.79

0.62

0.45
x

0.28
y
th
4 ply
0.11

Fig. 3.15 Typical measurement result for artificial fiber breakage in a unidirectional laminate at
0.42 % mean strain and 10 mm size located at different depth positions below surface ranging from
2nd ply (a) to 3rd ply (b) and 4th ply (c)

noise level and a refinement of the strategy to quantify strain concentration. To this
end, the discussion of the background noise floor is found in Sect. 3.3.3. In order to
quantify a representative strain concentration value for each defect, the experimen-
tally used data reduction strategy is introduced next.
The experimental measurements were carried out using static tensile tests with
an acquisition rate of 2 Hz yielding a large data base for a single type of defect. For
a representative subset of the images as function of the applied stress values, two
measurements are made each. First, the average tensile stress within the cross-
hatched area of Fig. 3.17 is evaluated by calculating the arithmetic average of all
subset values excluding the region of the embedded defect. This is denoted as strain
average hεiavg in Fig. 3.16a and is plotted as function of the applied stress value.
Second, at the positions of the embedded artificial defect the strain concentration
value hεiconc is evaluated at distinct points. As seen from the linear regression line
in Fig. 3.16a, the strain concentration values show almost linear behavior as
function of the applied stress value, but exceed the strain average at all times. For
the given example of a fiber breakage of Fig. 3.14a, this is indicated by the region of
78 3 Digital Image Correlation

a
1.8
strain concentration
1.6 strain average
1.4
1.2
strain [%]

1.0

0.8
0.6
0.4

0.2
0.0
0 200 400 600 800 1000 1200 1400
stress [MPa]
b
1.0

0.8
Δ-strain / stress [% GPa–1]

0.6

0.4 0.457 +/- 0.037

0.2

0.0
0 200 400 600 800 1000 1200 1400
stress [MPa]

Fig. 3.16 Evaluation of strain average and strain concentration of artificial fiber breakage of
10 mm size located at 2nd ply below surface as function of stress value (a) and respective
quantification of the strain exceedance normalized by the applied stress as function of stress (b)

increased compliance in the vicinity of the cut fibers. The origin of this local change
of compliance is straight forward to understand by the partial absence of fibers
contributing to the stiffness of the laminate.
While the example of Fig. 3.16a is rather obvious in terms of absolute values, the
same finding was valid for each defect investigated. All artificial defects investi-
gated exhibit similar behavior, being either systematically greater or lower than the
strain average, depending on the type of artificial defect and the strain direction
used for evaluation (see Sect. 3.3.2).
3.3 Strain Concentration 79

As measure of strain concentration relative to the surrounding area Δε the


following equation is evaluated for each stress value:
 
 
Δε ¼ hεiconc  hεiavg  ð3:15Þ

using the averaging procedure

1X m
hεiconc ¼ εj ð3:15aÞ
m j¼1

1X n
hεiavg ¼ εk ð3:15bÞ
n k¼1

where m is the total number of subsets evaluated in the global strain field and n is
the number of evaluation points of strain concentration in the region of the defect.
Taking the absolute values in (3.15) allows to interpret Δε as positive scalar
deviation relative to hεiavg. This definition was chosen since the detectability of a
defect does not distinguish between positive or negative exceedance of the strain
average.
For the values of the example in Fig. 3.16a, the value of Δε is observed to
increase with the applied stress value, thus indications of such embedded defects
are easier to observe at higher stress values. By dividing the value of Δε by the
applied stress σ the normalized value Δε/σ shown in Fig. 3.16b is obtained.
Remarkably, this value is almost independent of the applied stress, which is
attributed to a linear elastic relationship of both hεiconc and hεiavg to the applied
stress. Taking the arithmetic average of these values and their according standard
deviation this allows to obtain a characteristic value for this type of artificial defect
at its depth position and evaluates for this example as (0.457  0.037) [% GPa1
]. The specific units of Δε/σ are chosen to be percentage per Gigapascal to provide
numeric values close to one, which are easier to discuss in the following. However,
the use of GPa1 should not be misinterpreted as compliance value of the laminate,
since the used strain value does not reflect the global or local compliance value, but
the difference between both.
To reach a statistically more significant representation for one single defect, the
same evaluation is performed at n different points of strain concentration, i.e.,
Fig. 3.16 shows only one of these evaluations. Thus, the n locations for each defect
evaluation are chosen at spots of highest strain concentration, since the extreme
values will determine visibility of the strain signature. In contrast, averaging of
many points in the area of the embedded defects (i.e., n > 50) was not found to yield
appropriate values, since the extreme values of strain concentration will be sub-
stantially reduced in the averaging procedure. Thus n ¼ 5 positions were typically
used for the evaluation.
80 3 Digital Image Correlation

Based on this type of measurements, the following section is used to validate a


modeling strategy to predict the strain concentration effects in a laminate due to the
existence of typical defects before starting the discussion of system detectability
limits in Sect. 3.3.3.

3.3.2 FEM Modeling of Strain Concentration Due


to Internal Defects

In order to begin a systematic study of the strain concentration caused by internal


defects in fiber reinforced materials, a finite element modeling approach is devel-
oped and validated using DIC measurements on reference specimens, open-hole
tensile specimens, and specimens with embedded artificial defects.
For the finite element method modeling the “structural mechanics module” of
the commercial software package Comsol Multiphysics is used. A full 3D repre-
sentation of the experimentally used geometries is implemented and discretized by
a mesh using linear order shape functions and a maximum edge length of 0.5 mm
with substantial mesh refinement down to 10 μm when approaching the crack tips.
As material model, the data of the experimentally used Sigrafil CE 1250-230-39
material given in Table B.1 in Appendix B is used in the following. The individual
plies of the test specimen are chosen as homogenous unidirectional material with an
orientation in accordance with the experimental configuration. The reinforcement
sections are modeled as stacking sequence of [þ45/45/0]sym using the Sigrafil CE
1250-230-39 material data. For modeling of the strain concentration, a stationary
tensile load is applied as force at the top and bottom surfaces of the reinforcements
on one end of the specimen as seen in Fig. 3.17a. The opposite side of the specimen
uses fixed displacement conditions at the surface of the reinforcements
corresponding to the experimental configuration. All geometric dimensions of the
tensile specimens are matched to the experimental values, with their nominal
dimensions shown in Fig. 3.17.
To validate the finite element model in its global stress–strain prediction, a first
comparison of calculation results is made relative to experimental data for a case
study without incorporated defects. Figure 3.18 shows a comparison between
experimental results of a unidirectional tensile specimen with load axis in parallel
to the fiber axis and the respective modeling result. For the experimental results, the
strain values are obtained from the 3D-DIC measurements using the acquisition
parameters noted in Table 3.1. The experimental strain values are obtained as mean
value from area inside the dashed line as marked in Fig. 3.10 on the surface of the
specimen. Comparison is made to the simulated mean strain value as calculated on
the cross-hatched surface in the area indicated in Fig. 3.17. As seen from Fig. 3.18,
there is very good agreement of the experimental and simulated axial and transverse
stress–strain curves. Based on this validation, the linear-elastic material model is
3.3 Strain Concentration 81

force F

8 plies

area for averaging


z 1.9 mm
y fixed constraint
x

1.9 mm

Fig. 3.17 Model setup for validation of material properties and cross-hatched area for averaging
of strain values (a) and model setup for open-hole tensile test including dimensions (b)

assumed to be valid for the investigated stress limit of 800 MPa, an axial strain limit
of 0.65 %, and a transverse strain limit of 0.21 %.

3.3.2.1 Full-Field Comparison

To further validate the accuracy of the FEM model, a typical well-known macro-
scopic discontinuity is introduced to cause a strain concentration at a preferred
position. This is done in the form of an open-hole tensile test using a drilled hole of
nominally 3 mm diameter as seen in Fig. 3.17. Figure 3.19 shows the axial and
transverse strain fields of the open-hole tensile test at 346 MPa. Comparison is
made in Fig. 3.19 for the strain concentration in x- and y-direction in the vicinity of
the hole as obtained from 3D-DIC measurements (b,d) and as computed by finite
element modeling (a,c). As seen from the figures, the absolute values of strains of
measurement and computation agree well to each other. Some differences arise due
to the noise being present in the experimental measurement causing fluctuations of
the strain values and the spatial resolution of the measurement causing less resolved
peak structures when compared to the computation result. Using this full-field
validation it may thus be further concluded that the chosen modeling parameters,
such as discretization settings, are sufficient to compute similar strain concentration
values as experimentally measured.
82 3 Digital Image Correlation

800 experimental
modeling

600
stress [MPa]

400

200

0
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
axial strain [%]
b

800 experimental
modeling

600
stress [MPa]

400

200

0
0.00 –0.05 –0.10 –0.15 –0.20
transverse strain [%]

Fig. 3.18 Comparison of axial (a) and transverse (b) strain curves as obtained by 3D-DIC and by
finite element modeling

Due to the good agreement of the computed strain values for the open-hole
discontinuity, it is assumed that further finite element based predictions of strain
concentrations due to internal defects are valid within a certain level of accuracy.
The validated material model is thus used to calculate strain concentration effects
due to embedded artificial defects in fiber reinforced materials. To this end, one
representative example of each defect is shown in Figs. 3.20, 3.21, and 3.22 each.
As first example, Fig. 3.20 presents a comparison of an experimentally measured
strain concentration of a fiber breakage of 7.5 mm size located at the 2nd ply below
the surface at 0.42 % mean strain and the corresponding modeling result. Due to the
identical strain scales of both figures (measured in parallel to the load axis), a direct
discussion of the observed differences is possible. To model the cut fibers the
3.3 Strain Concentration 83

approach makes use of a “thin elastic layer” oriented in the yz-plane. The stiffness
of the “thin elastic layer” is set to zero, resembling fully separated surfaces, i.e., a
fully developed crack at this position. Due to the presence of the resin after
specimen fabrication, this assumption may only be fulfilled after exceeding a

Fig. 3.19 Comparison of the strain-concentration at 346 MPa of open-hole tensile test in x-strain
field and y-strain field as computed by finite element modeling (a,c) and as obtained from 3D-DIC
measurements (b,d)
84 3 Digital Image Correlation

Fig. 3.19 (continued)

certain minimum stress. This is required to crack the resin at this position and to
open a gap representative for broken fibers. At the applied 569 MPa stress this is
likely the case. The modeling result shows a primary strain concentration above and
below the position of the modeled crack. Highest values occur at the center of the
crack, i.e., at 3.75 mm distance to each edge of the crack. In contrast, the example of
Fig. 3.20a yields highest strain concentration values at the left edge of the crack.
This can likely be due to imperfections during the preparation, since the cut fibers
3.3 Strain Concentration 85

a x-strain b x-strain
[%] [%]
0.81 0.81

0.72 0.72

0.63 0.63

0.54 0.54

0.45 0.45

0.36 0.36
x
0.27 0.27
5 mm 0.18
y 5 mm 0.18

Fig. 3.20 Comparison of the strain-concentration at 569 MPa for an artificial fiber breakage of
7.5 mm size in a unidirectional laminate located at the 2nd ply below surface as measured (a) and
as computed by finite element modeling (b) (based on [37])

x-strain x-strain
a b
[%] [%]
0.82 0.82

0.76 0.76

0.70 0.70

0.63 0.63

0.57 0.57

0.50 0.50
x
0.44 0.44
5 mm y 5 mm
0.37 0.37

Fig. 3.21 Comparison of the strain-concentration at 695 MPa of an artificial delamination of


10 mm  10 mm size in a unidirectional laminate located at the interface between 1st and 2nd ply
below surface as measured (a) and as computed by finite element modeling (b) (based on [37])

tend to deviate from an ideal orientation along the x-axis as revealed by the 3D
images of Fig. 3.11. The orientation within the xz-plane may develop an angle,
which is not identical at each position along the y-axis. These effects may cause
distinctly different strain concentration values, since their absolute orientation
exhibit some intrinsic scatter. Moreover, secondary damage might be present at
the measurement position or might develop during specimen loading. For the case
shown in Fig. 3.20a, there is a certain likelihood of delamination formation during
laminate curing as consequence of the cut fibers. Furthermore, such delamination
may also initiate at low load levels, as e.g. used as measurement principle in
transverse-crack tension specimens (see Sect. 4.7 or [39]). Neglecting these effects,
the overall match of strain concentration predicted by the finite element model was
found to be in good agreement to the experimental measurement.
86 3 Digital Image Correlation

a y-strain b y-strain
[%] [%]
0.24 0.24

0.15 0.15

0.055 0.055

–0.039 –0.039

–0.13 –0.13

–0.23 –0.23

–0.32
x –0.32

5 mm –0.42
y 5 mm –0.42

Fig. 3.22 Comparison of the strain-concentration at 535 MPa of an artificial inter-fiber crack of
10 mm size in a unidirectional laminate as measured (a) and as computed by finite element
modeling (b) (based on [37])

As second example, Fig. 3.21 presents a comparison of an experimentally


measured strain concentration of a delamination of 10 mm  10 mm size located
at the interface between 1st and 2nd ply below surface stressed by 695 MPa and the
corresponding modeling result. To model such inter-ply delamination, the defect
was embedded as “thin elastic layer” of same dimensions in x- and y-direction as
the experimental defect. To mimic the presence of the embedded ETFE balloon, the
“thin elastic layer” was given the stiffness properties of ETFE along the z-axis for
compressive stresses normal to the surface and zero stiffness in x- and y-direction.
Use of a perfectly flat rectangular representation was found to underestimate the
measured strain concentration at least by a factor of two. However, the true shape of
the artificially embedded defect exhibits certain waviness along the z-axis as seen
from the computed tomography image of Fig. 3.12. Using a geometry with similar
wavelength and z-amplitude as for the experimental defect, the computation result
of Fig. 3.20b was obtained. Given the identical strain value range there is good
agreement between experiment and model. Primary differences between both
figures are attributed to the non-periodic waviness of the experimentally prepared
delamination and the difference in edge representation of the embedded ETFE
balloon. The latter exhibits a pronounced topology in the experiment, but was
merely accounted for as straight line in the model. This may likely cause the higher
strain concentration values seen in the model result. However, for the overall
quantification of strain concentration, this will be negligible as demonstrated in
the following.
As third exemplary representation a comparison is made between an experimen-
tally prepared artificial inter-fiber crack and the corresponding modeling result.
Figure 3.22 shows the comparison for a stress value of 535 MPa and 10 mm size of
the inter-fiber crack. As described above, this type of defect is prepared using an
3.3 Strain Concentration 87

embedded PI stripe reaching from the front to the back of the laminate, i.e.,
spanning the full z-extent of the laminate. Accordingly, the model uses an embed-
ded cube of 50 μm length along the y-axis, 10 mm length along the x-axis, and
1.8 mm along the z-axis (full laminate thickness). The domain is given the elastic
properties of PI as noted in Table B.1 in Appendix B. Using this representation, the
image of Fig. 3.22b is computed, yielding good agreement to the experimental
result of Fig. 3.22a.
Corresponding to the experiment, the higher Δε values are observed for the
y-axis providing a better visible signature. Thus in the regions of the embedded
inter-fiber crack larger Poisson contraction seems to occur, translating into lower
absolute y-strain values. The main differences between experimental and modeling
results are the waviness and position of the signature. The waviness may readily be
understood from the computed tomography images of Fig. 3.13 and is due to the
same preparation artifacts as previously described for fiber breakage. In the exper-
iment, the strain concentration seems to occur preferentially at the positions close to
the edge. This is also seen in the model result, yet the zone of highest strain
concentration was found to be smaller and the transition more continuous. Given
the geometric imperfections of the prepared inter-fiber crack it is thus likely that
this overrides the continuous transition of strain values as expected from the model
predictions.

3.3.2.2 Signatures of Artificial Defects

To investigate a broad range of possible configurations in unidirectional laminates,


variations of defect type, depth position, and size as shown in Fig. 3.23 are studied.
The positions of discontinuities represented by the embedded defects are marked as
colored lines in Fig. 3.23. All modeling results are compared to measurement
values of experimentally prepared artificial internal defects following the data
reduction strategy explained previously. For extraction of quantitative values
from the modeling results a different strategy was applied, since the data reduction
does not benefit from a statistical approach, given the model uses a numerically
convergent approach. Thus the extreme values of strain concentration may directly
be extracted from the model results at sufficient accuracy without further averaging.
To demonstrate the procedure, a typical result from a modeled fiber breakage
located at the 4th ply below surface is shown in Fig. 3.24. For each model, five
different dimensions have been tested. As indicated by the strain concentration in
close vicinity to the crack position, there is a clear visibility of the discontinuity. To
quantitatively interpret the modeling results, the following procedure was applied.
First, a cross-section line is drawn at the surface level of the specimen as indicated
in Fig. 3.24. The resulting evaluation of the x-strain and y-strain as function of
position along this line is shown in Fig. 3.25a, b, respectively.
Due to the presence of the reinforcements at the specimen ends and the applied
boundary conditions (cf. Fig. 3.17), there is a characteristic background deforma-
tion for the x-strain and the y-strain. The presence of the fiber breaks causes some
88 3 Digital Image Correlation

a b
viewing direction viewing direction

force force
d
z z
fiber axis fiber axis
x x

c
viewing direction viewing direction viewing direction

force

z
fiber axis 67.5° 45°
x
viewing direction viewing direction

22.5° 0°

Fig. 3.23 Depth configuration and orientation of defects to study strain concentration in unidi-
rectional materials for fiber breakage (a), inter-ply delamination (b) and inter-fiber failure in xz-
and yz-cross section (c)

readily visible indications, which are seen as spikes in Fig. 3.25a and as dips in
Fig. 3.25b. The strain concentration was evaluated as maximum height of the peak
relative to the strain average following (3.15). The cross-section line was chosen
according to the position of maximum strain concentration. The strain average was
obtained from the mean values of the cross-hatched area of the reference specimen
free of defects at the same stress level.
As first type of defects, the resultant strain concentration for fiber breakage is
evaluated. Figure 3.26 shows the Δε/σ evaluated from the strain in x-direction for
the modeled cases of fiber breakage as solid rectangles. In addition, experimental
results are added to the same graph by open circles including their measurement
error as discussed above. The colors indicate the depth position below surface, with
the 1st ply being the surface ply. As seen from Fig. 3.26, the presence of fiber
breakage at the surface ply causes strain concentration of approximately equal
height. As long as the spatial resolution of the DIC system is sufficient to detect
the fiber breakage, this should still be valid even for smaller sizes than the 1.25 mm
investigated. For the model results there is even a small trend towards higher Δε for
smaller defect sizes. For instance, in Fig. 3.25a there is a zone of strain concentra-
tion at each edge of the modeled fiber break. When decreasing the size of the defect
these two individual regions of strain concentration merge into one larger signature.
3.3 Strain Concentration 89

a
x-strain

cross-section line

z
y x

b
y-strain

cross-section line

Fig. 3.24 x-component of strain field (a) and y-component of strain field (b) with presence of fiber
breakage of varying defect size buried at 4th ply below surface of laminate

This causes a noticeable increase for the detectable Δε in this case (Fig. 3.30a
shows a respective example for the case of inter-ply delamination).
The situation changes for fiber breakages covered by one or more intact plies,
i.e., the fiber breakage is located within the volume of the material and does not
reach the surface. For all depth positions studied herein, there is a general trend
from large to small lateral dimensions. In all cases, the strain concentration
decreases with decreasing defect dimensions. According to (3.14) the behavior as
function of defect size should obey a square root law, which is found to be
applicable in the present case and is shown as solid line fit to the modeled data in
90 3 Digital Image Correlation

a
0.210

0.205
x-strain [%]

0.200

0.195

0.190
-60 -40 -20 0 20 40 60
x-position [mm]
b
-0.055

-0.060
y-strain [%]

-0.065

-0.070

-0.075
-60 -40 -20 0 20 40 60
x-position [mm]
Fig. 3.25 Evaluation of x-strain (a) and y-strain (b) along cross-section line of Fig. 3.24

Fig. 3.26. The modeled data is supported well by the experimental data given the
measurement errors of the experimental data. Due to the implication discussed in
the model validation above, it is likely that not all experimentally prepared defects
conform with the nominal values used in Fig. 3.26. Especially for the smaller
3.3 Strain Concentration 91

modeled data
experimental data
3.45
3.30 1st ply
3.15 (surface level)
3.00
Δstrain / stress [% GPa–1]

0.45

2nd ply

0.30
3rd ply
4th ply

0.15

0.00
0 2 4 6 8 10
size [mm]

Fig. 3.26 Δε/σ due to fiber breakage of varying defect size buried at several depths below surface
(based on [37])

defects, the preparation may easily cause a false orientation of the intended fracture
plane. Especially for the 2.5 mm fiber cut located in the 2nd ply such effects may be
the reason for the substantial deviation to the predicted trend. Furthermore, the
same size defect located in the 3rd ply agrees well with the model prediction and is
found at a similar level of strain concentration as the 2nd ply defect. The latter
contradicts the expected behavior as function of defect depth as will be discussed in
the following. Thus it may be speculated that the measured strain concentration
value of 2.5 mm at the 2nd ply might underestimate the real value for such
defect size.
For a constant defect size there is a distinct behavior expected as function of
defect depth. For defects buried deeper inside the material a respective decrease of
detectable strain concentration is expected at the surface level. As observed from
the trend in Fig. 3.26 for each defect size, this observation is found to be applicable.
To better quantify this decrease with defect depth, Fig. 3.27 shows the x-stress
component and the x-strain component in the xz-cross section of the model of a fiber
break with 10 mm extent in y-direction. The cross section is chosen at the center of
the crack, i.e., with 5 mm distance to each edge along the y-direction. As seen from
these cross sections, stress field and strain field exhibit very similar signatures,
which becomes even more apparent when computing a measure2 of the local

2
Due to the anisotropy of the material and the Poisson’s contraction effect, a specific discussion of
the local stiffness response by means of the components ∂σ=∂ε essentially yields the same result.
92 3 Digital Image Correlation

a Stress [MPa]
900
0.9 mm z
800
x
700

600

500

400
b Strain [10-3]

z 8

7
x
6

3
c Stiffness [105Pa]

z 1.4

1.36
x
1.32

1.28

1.24

1.20

Fig. 3.27 Zone of strain field (a), stress field (b), and local stiffness (c) around modeled fiber
breakage of 10 mm size in y-direction plotted in xz-cross section at stress value of 694 MPa

stiffness response by evaluating σ x/Δεx as seen in Fig. 3.27c. Except for the
expected locations close to the edges of the crack tip the stiffness response is
almost constant indicating the direct proportionality of stress and strain concentra-
tion. In particular, for the direction of highest strain concentration (indicated by
dashed lines in Fig. 3.27b) the stress–strain response is almost linear. Thus, the use
of a linear elastic fracture mechanics description seems applicable to discuss the
behavior of strain concentration. For an isotropic material the stress field around
such a mode I crack would decrease in x-direction σ x as function of z-distance z to
the crack tip following

1
σ x ¼ K I  pffiffiffiffiffiffiffi ð3:16Þ
2πz
3.3 Strain Concentration 93

a
4 Model data
Square root fit of 5 mm size

Δ strain /stress [% GPa–1] A = 0.038 +/-0.002


3 B = 0.133 +/- 0.013

0
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
depth below surface [mm]
b
viewing direction

0.0 mm
ply 1
ply 2
ply 3
ply 4
0.9 mm
ply 5
ply 6
z
ply 7
x ply 8

fiber axis

Fig. 3.28 Change in Δε/σ due to fiber breakage as function of defect depth with colors
corresponding to depth position (a) and respective scheme to quantify depth position of defect (b)

For the given case of an anisotropic material a similar relationship, may be


assumed which will give rise to a distinctly different stress-concentration factor
KI. However, this is expected to be a constant material property, which does not
inhibit the following discussion. As function of depth position z, the strain concen-
tration can thus be estimated to be proportional to the stress concentration. Conse-
quently, a decay of strain concentration Δε as function of z1=2 is derived as first
order approximation from analytical theories.
Figure 3.28 presents the same data as Fig. 3.26, now plotted as function of defect
depth below surface. As measure of defect depth, the edge of the crack closest to the
surface is selected as indicated in the scheme of Fig. 3.27b (first ply is at z ¼ 0 mm).
94 3 Digital Image Correlation

As seen from the square root fit to the modeled data in Fig. 3.28a, the decrease of
Δε/σ with defect depth is well described by the relationship Δε=σ ¼ A e þ Bz
e 1=2
using the fit parameters Ae and B.
e For better visibility, Fig. 3.28a shows only the fit of
the 5 mm size fiber cut, since the remaining fits essentially provide the same trend,
but almost superimpose in Fig. 3.28a. Thus the consideration regarding the general
trend described by (3.14) seems readily applicable for the case of fiber cuts studied
herein.
As second type of defects, the occurrence of inter-ply delamination was studied.
Figure 3.29 shows the strain concentration Δε/σ evaluated from the strain in
x-direction for the modeled cases of inter-ply delamination as solid rectangles.
Experimental results are added to the same graph by open circles including their
measurement error. The colors indicate the depth position of the interface studied.
Although, inter-ply delamination in unidirectional materials may occur at arbi-
trary depth positions, the discretization steps taken for the analysis are chosen in
accordance with the interface positions of the plies used to build the laminate. This
is due to the experimental preparation of the artificial defects, since the ETFE
balloons may only be prepared between the individual plies. Thus the colors in
Fig. 3.29 indicate the interface location, i.e., ply 1-2 indicates the position between
1st and 2nd ply. All defects are prepared inside the volume of the material, so there
is no defect, which extends to the surface. As consequence, all Δε values obey the
same trend as function of defect size following the square root law as given in
(3.14). However, the fit values used to obtain the lines of Fig. 3.29 return distinctly
different values than for fiber breakage. This is due to the geometric difference

0.20 modeled data


0.18 experimental data

0.16 ply 1-2


Δstrain / stress [% GPa–1]

0.14

0.12

0.10

0.08
ply 2-3
0.06
ply 3-4
0.04

0.02

0.00
0 2 4 6 8 10
size [mm]

Fig. 3.29 Δε/σ due to inter-ply delamination of varying defect size buried at several depths below
surface as strain concentration per GPa external stress (based on [37])
3.3 Strain Concentration 95

between the modeled defects. Although the inter-ply delamination does produce a
noticeable strain concentration, the modeled waviness and its size along the x-axis
cause a distinctly different signature. As seen from the evaluation of the x-strain in
Fig. 3.30a for five different sizes of inter-ply delamination, the signature basically

a
0.210

0.205
x-strain [%]

individual signatures merge


0.200

0.195

0.190
–60 –40 –20 0 20 40 60
x-position [mm]
b
1.00
Model data
Square root fit of 5 mm size

A = 0.001 +/-0.002
0.75 B = 0.105 +/- 0.017
Δstrain /stress [% GPa–1]

0.50

0.25

0.00
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
depth below surface [mm]

Fig. 3.30 Evaluation of x-strain along cross-section line according to Fig. 3.24 for depth position
1-2 (a) and evaluation of Δε/σ as function of depth position (b)
96 3 Digital Image Correlation

consists of two sharp strain peaks located at the edge of the delamination and a
plateau of increased strain values in between. Since the waviness is implemented
along the y-axis, there is no strain value modulation seen in Fig. 3.30a. For smaller
defect sizes the two peaks start to approach each other, merging into one signature
below 1.25 mm delamination size and thus possibly deviate from the square root
dependency of (3.14).
Plotting the data of Fig. 3.29 as function of defect depth below surface returns
the graphical representation shown in Fig. 3.30b. As defect depth position, the mean
depth with respect to the sinusoidal waviness amplitude was chosen which directly
corresponds to the position of the interface between the individual plies. For better
visibility only the fit for the 5 mm size delamination is shown in Fig. 3.30b as solid
line. Similar to the case study of fiber cuts above the trend of Δε/σ as function of
defect depth is well captured by the relationship Δε=σ ¼ A e þ Bz
e 1=2 . The resulting
fit parameters A e and B e are characteristically different to the fiber cuts, which is
dominated by the absolute difference of Δε/σ. Nevertheless, the overall trend
confirms the applicability of the abovementioned considerations regarding the
decay of Δε/σ with z1=2 .
As further defect type, the occurrence of inter-fiber failure has been investigated.
Since inter-fiber failure in the case of a unidirectional material is assumed to span
the full thickness of the laminate, the depth position is not varied for this case, i.e.,
the experimentally prepared and modeled defects always reach to the surface for
angles <90 . For the specific case of an inclination angle of 90 the defect is
oriented within the xy-plane with its fracture plane as seen in Fig. 3.23. This would
correspond to a delamination at the midplane of the laminate. The evaluation of
Δε/σ as function of defect size and inclination angle is shown in Fig. 3.31 for the
cases of inclination angles of 0 , 22.5 , 45 , and 67.5 following the angle definition
given in the scheme of Fig. 3.23. For the angle 90 experimental data and modeled
data is added to the figure obtained from a respective delamination case. Evidently,
the observed behavior in Fig. 3.31 is distinctly different to the findings for fiber cuts
and delamination which requires some further discussion.
Based on the previous discussions, the value of Δε/σ is influenced by the size of
the defect and the depth position. For the strain concentration as function of defect
size, Fig. 3.31 reveals opposing trends above and below the inclination of 45 . The
behavior observed at 67.5 and 90 clearly corresponds to the trend observed for
delamination obeying a square root law as function of size. At an inclination angle
of 45 almost no change as function of defect size is found. Above 45 the behavior
cannot be described by a square root dependency and Δε/σ is instead found to
increase with decreasing defect size.
This is due to the definition of Δε in (3.15) using the absolute value of strain
concentration. For the cases above 45 , the highest Δε are obtained for the y-strain
values hεiconc higher than hεiavg in good agreement with the observations for
delamination. For inclination angles smaller than 45 , the highest Δε signatures
occur for hεiconc smaller than hεiavg as indicated in Fig. 3.32b. Thus, a turnover of
3.3 Strain Concentration 97

modeled data
0.60 experimental data

0.55
0.50
Δstrain / stress [% GPa-1]


0.45
0.40
0.35 22.5°

0.30 45°
0.25
0.20 67.5°
0.15
0.10
0.05 90°

0.00
0 2 4 6 8 10
size [mm]

Fig. 3.31 Δε/σ due to inter-fiber cracks as function of varying defect size and inclination angles
(based on [37])

the behavior may be expected at an angle of 45 resulting in the almost constant


values as function of defect size.
To investigate this effect more precisely, the data of Fig. 3.31 is plotted as
function of inclination angle inFig. 3.32. The strain signatures are best described by
a cosine fit Δε=σ ¼ A eþB e cos φ  C e of the 5 mm defect size using fit parameters
e B,
A, e and Ce which is shown as solid line. This fit is found to describe the observed
dependency very well. Thus, the decay of strain concentration as function of angle
may readily be interpreted as sinusoidal change between the two extreme cases at
0 orientation and 90 orientation. The change in dimension of the embedded crack
is then translated into different amplitudes B.

3.3.3 Detectability of Defects Using DIC

Based on the results of Sects. 3.3.1 and 3.3.2, the purpose of this section is the
discussion of the proposed measurement routine for the application to evaluate
composite failure. Beyond the quantification of the strain concentration presented
in the previous section, this requires further considerations to be made. First,
the influence of measurement parameters and experimental issues intrinsic to the
measurement principle are discussed. Second, the predictive capabilities of the
presented modeling approach and the applicability to real defects are reviewed.
98 3 Digital Image Correlation

a
Model data
0.40
Sine fit
0.35
A = 0.21 +/- 0.01
Δstrain / stress [% GPa–1]

0.30 B = -0.18 +/- 0.01


C = 45°
0.25

0.20

0.15

0.10

0.05

0.00
0 10 20 30 40 50 60 70 80 90
angle [°]
b
y-strain

22.5°
z

y x

strain minimum

67.5°

strain maximum

Fig. 3.32 Evaluation of Δε/σ as function of inclination angle (a) and regions for evaluation above
and below 45 (b)

Finally, the system limits of presently used 3D DIC systems are established and
compared to the experimental findings of Sect. 3.3.2 to assess the detectability limit
of the DIC method to spot buried defects.
3.3 Strain Concentration 99

3.3.3.1 Experimental Parameters

For the experimental measurement of full-field strain distributions, several aspects


need to be considered when discussing signatures attributed to strain concentra-
tions. For the exemplary measurement in Fig. 3.33a at a mean strain of 0.44 % there
is an obvious gradient along the y-direction in the x-strain values. The origin of this
gradient can be manifold, starting with specimen imperfections, and bad bonding of
reinforcement tabs or wrong specimen alignment. In all cases, the consequence can
be a systematic variation of the resultant strain level as function of position. This is
not only bad for measurement of material properties, but also poses further chal-
lenges for the interpretation of strain concentration. In this context, the obtained
strain reference hεiavg may need to be interpreted differently. To demonstrate this
effect, Fig. 3.33b shows a model result of the same 7.5 mm fiber breakage cases in
the 2nd and 3rd ply below surface with a superimposed gradient in the tensile force.
This change in loading conditions causes a visible strain gradient similar to the
experimental data. Nevertheless, the strain concentration due to the embedded
defects is still clearly visible for both cases. However, for application of (3.15)
the question of the correct interpretation of hεiavg arises. Since the load gradient is
balanced, the same strain average hεiavg is calculated as for the case of Fig. 3.26.
Thus, for specimens with unbalanced strain gradients, the correct average strain
should be taken from the local neighborhood of the defect, since the strain concen-
tration relative to this reference will define the detectability.
Consequently, the occurrence of such strain gradients does not affect the general
possibilities to spot defects as strain concentration in full-field strain measurements.
Moreover, almost the same quantitative information may be extracted, if the strain
reference is taken from the local neighborhood for cases of inhomogeneous (global)

a x-strain b x-strain
[%] [%]
0.57 0.57

0.54 0.54

0.51 0.51

0.48 0.48

0.46 0.46

0.43 0.43
x
0.40 0.40
y 7.5mm
0.37 0.37

Fig. 3.33 Experimental data of embedded fiber breaks exhibiting gradient in x-strain (a) and
corresponding model with embedded defects (b)
100 3 Digital Image Correlation

a b

particle
position

spurious
artefacts

x
invalid strain
y 2.2 mm computation 3.6 mm

Fig. 3.34 Dust particle artifact (a) and artifacts due to incorrect subset tracking (b)

strain fields. The latter is especially relevant for real structures, because these will
hardly exhibit equal strain values at all locations on the measured surface.
Another important effect to consider for measurement of strain concentrations
arises from the image correlation algorithm and the speckle pattern used for the
measurement. As seen in the images in Fig. 3.34, there are characteristic signatures,
which might readily be interpreted as local strain concentration.
There are various effects, which might cause a false computation of a local strain
value leading to characteristic artifacts:
1. Dust particles on lens
2. Incorrect subset tracking
3. Corrupt speckle patterns
4. Wrong choice of subset size relative to speckle pattern
For the first effect, the principle of DIC explained in Sect. 3.2 may readily be
applied to understand the impact. If a dust particle sits on the camera objective or at
any other position included within the optical path of the system it may readily
cause a distortion of the optical beam path. This directly translates into a wrong
computation of the real strain value at the position of the dust particle within the
field of view. Consequently, the position of this artifact will not change during
the measurement, i.e., the position of the strain signature will be constant during the
measurement and will not move with the specimen during loading. Thus, the
position of the artifact within the image will be constant for the case of small
deformations. For larger deformations, its position may still start to displace, due to
the wrong reference image taken with the dust particle in place. As additional
source to generate similar artifacts in the strain field, the occurrence of dead pixels
on the camera chip may be named. In such cases, one or more pixels on the camera
chip are not operating with equivalent sensitivity compared to the remaining pixels
or do not operate at all. For some systems it is possible to eliminate such dead pixels
by calibration routines, but for systems without such elimination the recorded
3.3 Strain Concentration 101

images have systematic deficiencies at the locations of dead pixels. The conse-
quence will be an erroneous computation of the strain values for those subsets
involving the dead pixels. Thus, the position of the artifact within the image will
stay at a constant position and will not move with the loaded specimen.
Another class of artifacts, which may readily be interpreted as strain concentra-
tion, are those due to incorrect subset tracking. As seen in the example of Fig. 3.34b,
this may cause two characteristic effects. If the subsets are not recovered by both
cameras in a 3D-DIC configuration, some software packages indicate this by
returning a distinctly different color in the image (white colored subsets in the
case shown in Fig. 3.34b). This marks subsets with obviously invalid strain values.
For high resolution measurements, spurious artifacts tend to appear as well. These
are found as characteristic pair of high and low strain values in the direct vicinity of
one subset. Their occurrence is sometimes not permanent and may vary in each step
of image correlation. This is most likely due to imperfections of the speckle pattern
causing incorrect subset tracking from image to image and, consequently, unreal-
istically high strain values.
Even more computation artifacts may occur, when a corrupt speckle pattern is
used. The definition of “corrupt” covers several aspects, all of them resulting in an
erroneous measurement of strain values. First of all, the speckle pattern may show
unintended variation of the particle sizes. For sprayed patterns this may typically
occur due to formation of larger drops or irregular particle distribution
(cf. Fig. 3.35a). Moreover, for speckle patterns which do not permanently stick to
the specimen the pattern might get harmed during handling of the specimen prior to
testing or during mounting of the specimen in a test machine (cf. Fig. 3.35b). For all
these cases, the speckle pattern no longer fulfills the quality required for the image
correlation process and the derived displacement vectors will suffer in accuracy to a
substantial extent. Consequently, the strain values calculated for such areas will not
provide sufficient precision to evaluate the small strain concentrations expected for
most of the defects studied in Sect. 3.3.2.
Another source for artifacts is the choice of subset size relative to the applied
speckle pattern particle distribution size. For the purpose of detection of failure

a b

2.6 mm 2.6 mm

Fig. 3.35 Corrupt speckle pattern showing irregular particle distribution (a) and partially
destroyed speckle pattern (b)
102 3 Digital Image Correlation

types as presented in Sect. 3.3.2 a sufficiently high spatial resolution is required to


spot smaller defect sizes. This relationship will be discussed in Sect. 3.3.3.3 in more
detail. However, a smaller subset size will give rise to several effects. Very small
subset sizes tend to generate unreliable strain values as seen for the subset size cases
of 10, 20, and 30 px seen in Fig. 3.36. Even for a perfect speckle pattern, this is

a b

15 mm

embedded foil

50% 30% 10% 50% 30% 10%


subset 30 px × 30 px subset 20 px × 20 px

c y-strain

50% 30% 10%


subset 10 px ×10 px

Fig. 3.36 Influence of subset size and subset intersection on strain noise and visibility of strain
signatures for five embedded inter-fiber cracks of various dimensions
3.3 Strain Concentration 103

owed to the fact that the image correlation process is based on a correlation
algorithm, which generally benefits from a larger number of data points to correlate
(cf. Sect. 3.2). Thus, subset values lower than 20 px need to be used with care when
interpreting the computed values. Furthermore, the stochastic pattern used for the
image correlation procedure needs to match the chosen subset size in order to avoid
erroneous computations. Clearly, if the speckle pattern exhibits additional inhomo-
geneity as discussed above, the local deviations will also give rise to artifacts,
which will appear more often for smaller subset sizes, even if the subset size is well
adopted to the global average.

3.3.3.2 Applicability of the Modeling Approach to Real Defects

As pointed out in Sect. 3.3.2, the artificial defects are prepared by means of
embedded foils or cut fibers. This naturally opens the discussion, whether or not
these defects are representative for real failure mechanisms as observed in
composites.
For the artificial fiber failure, the filaments are cut during the fabrication of the
laminate and thus a resin rich area is expected in between the position of the
filaments (cf. Fig. 3.11). This is an obvious difference to a real fiber failure,
where the space between the filament ends would be filled by a cavity. The
model uses an almost realistic assumption of broken filaments, since the thin elastic
layer is fully decoupled. As indicated in Sect. 3.3.2 the strain concentration
measured for the fiber cuts agrees reasonably well to the model predictions. Thus
the estimated values of Δε/σ are expected to be also applicable for real fiber failure
cases. However, the general drawback for this prepared defect arises from the
lateral sizes investigated. For real laminates, the expected defect sizes for fiber
failure prior to ultimate failure are mostly in the order of few tens of micrometers.
This is due to the fact that clustering of only few fibers in immediate vicinity will
induce global failure of the respective ply [40, 41]. Since these defect sizes are
difficult to prepare by the experimental means of manual cutting, it is difficult to
evaluate smaller defect sizes experimentally. However, even for smaller sizes of
fiber failure than investigated experimentally the model predicts almost comparable
values of Δε/σ. Given a reasonable dimension of fiber failure of a rectangular area
of 20 μm  20 μm (i.e., approximately 5–9 fiber filaments, dependent on filament
diameter and fiber volume fraction) a corresponding model was evaluated to
provide a strain concentration of 0.06 % at 695 MPa. This is close to the system
accuracy as will be discussed below and, therefore, may be at the edge of
detectability.
For the case of delamination, the investigated defect sizes are quite realistic. The
main discrepancy to a real delamination originates from the use of an ETFE foil as
embedded balloon. Whilst this approach seems to provide a reasonable approxi-
mation at first, it does cause a substantial bias for the local strain concentration.
Figure 3.37 compares a computation result of the x-strain evaluated for a 10 mm
size delamination including the embedded ETFE foil as used for the validation in
104 3 Digital Image Correlation

a b
x-strain
10 mm load axis 10 mm load axis

x x

y y

Fig. 3.37 Modeling result for inter-ply delamination at interface between 1st and 2nd ply with
embedded ETFE (a) and without (b)

Sect. 3.3.2 and without such a foil as expected for a real delamination. Given the
same color range of the x-strain in Fig. 3.37, the difference is quite obvious. Clearly,
the ETFE foil acts like an intermediate layer of reduced stiffness, causing high
strain concentration in the region of the embedded balloon. In comparison, the layer
modeled in Fig. 3.37b uses the same waviness as in Fig. 3.37a, but the out-of-plane
compressive stiffness of the layer is chosen to be compatible with the remaining
composite as expected for a real delamination. As consequence, the calculated
strain concentration drops from Δε ¼ 0:108 % to Δε ¼ 0:007 %. Clearly, this will
cause a substantial reduction of the possibilities to spot the occurrence of delami-
nation failure in real laminates regardless of their size.
Similar to the previous case, the dimensions of the artificial inter-fiber failure
defects studied in Sect. 3.3.2 are in a typical range as expected for real composites.
Again the main discrepancy to the real failure mode arises from the inclusion of a
thin PI foil to mimic the presence of an internally separated material. Although the
tensile and shear load transfer by this thin foil is negligible, it still provides an
artificial layer of low stiffness under compressive load. This local change in
compliance is responsible for the generation of the high transverse strain values
evaluated as strain concentration effect in Sect. 3.3.2. However, if there is no such
layer in a real specimen, the appearance of the inter-fiber crack will likely change.
Thus, in a tensile load scenario the model of an inter-fiber crack with identical
stiffness of the thin elastic layer as for the surrounding composite did not even
provide a strain signature exceeding the numerical noise level. However, since the
perfect orientation of such inter-fiber failure is quite unlikely, the result of a
computation with a minimal rotation of 2 around the z-axis is shown in
3.3 Strain Concentration 105

a b
load axis load axis
y-strain

10 mm 10 mm

x x

y y

Fig. 3.38 Modeling result for inter-fiber failure with 0 inclination angle and embedded PI (a) and
without PI, but additional rotation of 2 around z-axis (b)

Fig. 3.38b instead. As seen due to the same y-strain color range in Fig. 3.38, this
immediately yields a strain concentration signature of equivalent intensity as for the
embedded PI foil. Given the typical deviations of the inter-fiber crack orientations
in a real laminate it may thus be assumed that this type of failure is readily visible
by means of 3D DIC observations. For the macroscopically visible crack shown in
Fig. 3.52 in Sect. 3.4, this finding may also be directly concluded from visual
observations during the measurement.
However, more interesting applications arise for those inter-fiber cracks not
reaching the surface since this inhibits a direct visual monitoring of their occur-
rence. As seen in the measurements of the x-strain field of a multi-axial laminate in
Fig. 3.39 characteristic strain concentrations aligned with the ply directions (45
and 90 ) indicate the occurrence of inter-fiber failure in these layers. Thus it may be
concluded from modeling and experiment that the observation of inter-fiber failure
should generally be possible given the system accuracy is sufficient as will be
discussed in the next section.

3.3.3.3 Limitations Due to System Accuracy

So far, some first limitations of the experimental measurement technologies were


presented and the visibility of defects was mostly discussed in terms of modeling
results. However, the modeling results come with zero background noise and
106 3 Digital Image Correlation

εavg = 1.05 % x-strain [%]


1.5
y
1.2
8 mm x
0.9

90° cracks 0.6

45° cracks 0.3

0.0

Fig. 3.39 Experimental tensile tests showing clear signatures of inter-fiber cracks of 45 and 90
orientation to load-axis (x-direction)

neglects aperture effects which arise due to the experimentally used subset sizes.
Thus, the final step to judge on the detectability of specific failure mechanisms
consists in the assessment of limitations of the measurement system. To this end,
there are two essential influences to account for. First, the detectability will be
discussed in terms of the detectable strain concentration Δε relative to the back-
ground noise. Second, the aperture effect given by the subset size will be evaluated
to establish the limit of spatial resolution of the system.
With reference to system accuracy, the measurement equipment needs to detect
very small local increases of strain concentration Δε relative to surrounding areas.
The absolute limit of detectability may be defined in terms of the values hεiavg  Δ
hεiavg and hεiconc  Δhεiconc taking into account the measurement uncertainty of the
global average strain value Δhεiavg and the local strain value Δhεiconc.
Detectability of a signature is then defined as

hεiconc  hεiavg þ Δhεiconc þ Δhεiavg > 0 ð3:17Þ

The average value of hεiconc thus needs to be larger than the value of hεiavg plus the
uncertainties of both strain measurements Δhεiavg and Δhεiconc. This definition
assumes the validity of (3.15), i.e., hεiconc > hεiavg needs to be generally applicable.
The overall strain concentration Δε will determine the absolute detectability of a
failure mechanism. If the strain concentration values get small, the detectability
will be mostly limited by the uncertainty Δhεiavg (cf. Fig. 3.40).
However, given the investigation in Sect. 3.3.2 it is evident that the strain
concentration may be approximated using a linear elastic material behavior. Con-
sequently, the absolute value of Δε will scale linearly with the applied stress.
Hence, defects which fall below the detection limit at low load levels may readily
3.3 Strain Concentration 107

a
1.4 global strain (100mm lens)
local strain (100mm lens)
1.2 global strain (50mm lens)
local strain (50mm lens)
1.0
strain [%]

0.8

0.6

0.4

0.2

0.0
0.0 0.2 0.4 0.6 0.8 1.0 1.2
subset size [mm]

b
x-strain
[%]
50mm lens
0.91

0.80

5mm 0.69

0.58
100mm lens 0.47

0.36

0.25
5 mm
0.14

Fig. 3.40 Comparison of evaluated strain values as function of subset size for two lenses (a) and
corresponding full-field measurement showing the strain concentration close to the measured fiber
cut (b) (in part based on [37])

be visible at higher load levels, since the values of Δhεiavg and Δhεiconc were not
found to increase with the applied stress.
For the experimental investigation of Sect. 3.3.2, the settings according to
Table 3.1 allow an estimate of the absolute strain limit based on Fig. 3.40. Given
the subset size of 21 px and the resolution of 0.0136 mm/px the minimum value of
Δε to be detected may be estimated based on the uncertainty of Δhεiavg ¼ 0.029 %
and Δhεiconc ¼ 0.023 % and therefore evaluates as Δε ¼ 0:052 %.
108 3 Digital Image Correlation

The spatial resolution of the measured strain field is determined by the camera
resolution and the choice of subset parameters. The camera resolution first of all
depends on the pixel density of the camera sensor and the measurement distance to
the object. Due to the fact that modern cameras already provide 5–50 Megapixels
on their camera chips, an increase of pixel density by one order of magnitude is not
readily expected. Except for some specific fields of applications like astronomy,
there is no strong market requirement driving a development towards larger pixel
densities.
For a particular system a direct benefit towards higher spatial resolution may
thus be obtained by an increase of the magnification factor. Given the general laws
of optics as introduced in Sect. 3.1 an increased magnification at constant detector
size comes with the drawback of decreased field of view. So for experimental work
using 3D DIC, the following discussion will be limited to object sizes of reasonable
extent, i.e., in the range of typical laboratory size specimens.
To demonstrate the effect of the magnification factor on the system accuracy,
Fig. 3.40a provides a comparison of measurements of a fiber cut monitored by the
same DIC system using a 50 mm lens and a 100 mm lens. The respective magni-
fication factors correspond to 0.0291 and 0.0136 mm/px, respectively. For this
study, the size of the chosen subsets dsubset was systematically varied keeping the
intersection of subsets at approximately 50 %. To allow direct comparison of both
lenses, subset size values are reported as metric values instead of pixels. Whilst the
global strain average hεiavg is almost unaffected by the changed lens for all subset
sizes, a distinct change in the values of local strain hεiconc is observed for subset
sizes of lengths <0.6 mm. This is caused by the increased spatial resolution
provided by the 100 mm lens.
However, the findings shown in Fig. 3.40 indicate that the spatial resolution of
the strain field is also the result of the size of the chosen subsets and thus will add to
the ability to detect smaller defects. As discussed previously, the intuitive assump-
tion is that smaller subset sizes will provide better spatial resolution and, therefore,
detectability. However, a decrease of subset size comes with the price of increased
artifacts and less stable strain values. As consequence, an optimum trade-off
between the accuracy of the strain value and the spatial resolution is required.
This relationship is already visible in Fig. 3.40a and is indicated by the decreas-
ing uncertainty Δhεiavg as function of subset size. Thus it is not advisable to go
below a certain subset size in pixels. For the 50 mm lens a subset length of 0.6 mm
and for the 100 mm lens a subset size of 0.25 mm may thus be used as a proper
choice.
Since some systems average the local strain value by usage of neighboring
subsets, an additional aperture effect needs to be considered for the strain evalua-
tion. As seen in the schema in Fig. 3.41a, the strain value at a certain subset position
is evaluated from the average of a local neighborhood (a 3  3 neighborhood with
50 % intersection in this example). Therefore, assuming a typical Gaussian weight
function, the spatial resolution of the system may be estimated from the half-width
3.3 Strain Concentration 109

a
50%

d
su
bs
et

dsubset

b
0.7
global strain
local strain

0.6
strain [%]

0.5

0.4

0.3
0 20 40 60 80 100
intersection [%]

Fig. 3.41 Schema to evaluate spatial relation of subsets (a) and evaluated strain values as function
of subset intersection for example of Fig. 3.40 (b) (in part based on [37])

at half-height. The evaluation of the strain value at the position of the embedded
fiber cut shown in Fig. 3.41b clearly reveals the importance of this contribution to
the visibility of the defect. As function of subset intersection the global strain value
hεiavg is not affected much, but the local strain value hεiconc undergoes an almost
linear increase from 0 to 95 % subset intersection. This is due to the highly localized
strain concentration observed for this case and thus provides an improvement in
detectability by a factor of two. Therefore an increase in subset intersection
110 3 Digital Image Correlation

constitutes the better approach to improve spatial resolution as compared to a


decrease of subset size. For the limit of intersection approaching 100 % the size
of the single subset starts to dominate the measurement again. Consequently, the
subset size will constitute the absolute limit in terms of spatial resolution, i.e., the
minimum size dconc of a strain concentration, which can be detected may be
approximated for a quadratic subset size dsubset as:
pffiffiffi
dconc > 2dsubset ð3:18Þ

However, the computation of full-field strain values with small subset sizes and
high intersection values approaches the limits of presently available computation
capacities. As more realistic estimate, one may thus instead consider an intersection
of approximately 50 % leading to a modification of (3.18) by accounting for a 3  3
neighborhood according to the scheme in Fig. 3.41a:
pffiffiffi
dconc > 2 2dsubset ð3:19Þ

For the type of high-resolution system ARAMIS 12 M used in the study in


Sect. 3.3.2 the subset size of 21 pixels corresponds to a physical length of
0.29 mm, which turns into an expected lower limit for defect visibility of 0.81 mm.
Combining the limitations with respect to detectability given by (3.17) and
(3.19), some final conclusions can be obtained.
Due to the spatial resolution of presently available systems, it will be difficult to
detect strain concentrations as indications in the range of one millimeter or below.
This does not necessarily imply that smaller defects may not be visible, since the
size of the strain concentrations does typically exceed the physical size of the defect
by at least a factor of two (cf. Figs. 3.20, 3.21, and 3.22). Therefore many failure
types occurring in fiber reinforced composites may be sufficiently resolved.
A more critical parameter for the visibility of a buried defect the visible strain
concentration value at the surface of the material was evaluated. The present
approach using leading edge image correlation algorithms is at the limit of detect-
ability when considering the small strain concentrations expected for some of the
smaller size failure mechanisms. Especially for the case of delamination, a low
expectation for detectability is given by the low values of strain concentration. For
such cases, the use of shearography as described, i.e., in ASTM E 2581 was found to
be much more sensitive [42–45]. However, large size defects will cause noticeable
strain concentration values and can therefore be used as valuable tool to monitor the
occurrence of failure in composites as will be presented in Sect. 3.4. Another
natural extension of the presented approach considers the use of digital volume
correlation to evaluate strain fields not at the surface of a composite material, but on
the inside as will be demonstrated in Sect. 6.4.
3.4 Application to Composites 111

3.4 Application to Composites

As discussed in Sect. 3.3 the possibility to detect the occurrence of particular


defects in composite materials is a question of the spatial resolution of the measured
strain field as well as the depth position of occurrence. In this section, some
traditional fields of application for DIC during testing of composites and some
new developments are presented to apply DIC for evaluation of material properties
of composite materials.

3.4.1 DIC as Optical Extensometer

One of the traditional fields of application for DIC is the replacement of strain gages
or mechanical extensometers in mechanical testing. Several drawbacks associated
with the use of strain gages are well known and are found in standard literature (i.e.,
[46]):
• There is always a mismatch of thermal expansion coefficients at least in one
direction. Since the outermost layer usually has anisotropic thermal properties,
changes of temperature during testing may have substantial impact on the
measurement accuracy.
• The high strain levels sometimes reached in fiber reinforced materials may cause
preliminary failure of the strain gage due to overstraining.
• Another factor is the possible interaction between periodic structures of the
composite (e.g., woven fabric) and the grid length, causing error-prone
measurements.
• The long-term stability during creep or fatigue experiments are well-known
challenges for strain gages as well.
• For extensometers, knife edges may act as stress concentrators causing prelim-
inary failure at the mounting positions.
On the positive side, strain gages and extensometers benefit from a long tradition
in mechanical testing, which is accompanied by well-developed technical equip-
ment and skilled personnel.
Some of the possibilities of optical measurement techniques were already
presented in Chap. 2.4 demonstrating the selection of appropriate regions to derive
mechanical properties. In this context it is worth noting that the accuracy of 3D-DIC
methods has nowadays reached the same level as strain gages [1, 47–50]. Moreover,
the evaluation of the full strain field allows distinguishing between global strains
(corresponding to strain gage values) and local strains (as caused by the presence of
strain concentrators). Measurement of the full set of material properties as required
by the advanced failure criteria presented in Sect. 2.3 requires a comprehensive
amount of material testing. In the following, some examples are presented how DIC
techniques can be used to obtain material properties of composites.
112 3 Digital Image Correlation

areal evaluation

line evaluation
point evaluation

Fig. 3.42 Drawing of typical tensile test specimen and scheme of possible strain component
evaluations using DIC

Since the measurement by DIC results in full-field information of the strain


components of the material under test, there is a variety of possibilities for offline
analysis. As schematically shown in Fig. 3.42 the computed strain components can
be evaluated, e.g., at specific points or averaged along lines or areas of interest.
Using areal or line evaluations strain components may also be visualized based on
their position to yield results as seen in Fig. 3.25 or Fig. 3.19.

3.4.1.1 Tensile Testing of Unidirectional Fiber Reinforced Polymers

As a basic example for characterization of composites, the tensile test specimen and
the measurement result of a unidirectional fiber reinforced polymer tested with fiber
orientation in parallel to the load axis is shown in Fig. 3.43. For that purpose
standard specimen types as required by DIN EN ISO 527-5, DIN EN 2561, or
similar standards are used. The speckle pattern is applied to the surface of the test
specimen as shown in Fig. 3.43. To produce the speckle pattern, a white base coat
with subsequent application of a black spray pattern has proven to be useful for
materials characterization of composites. The opposite approach using a white
pattern directly applied on the naturally black color of carbon fiber reinforced
composites does not provide equivalent results due to the opaqueness of the outer
surface. Thus a well-defined background such as given by the white base coating is
the best choice for a good signal-to-noise ratio of the measurement. In application
to composites, the quality of the surface finish can be expected to have an impact on
the measurement signal. Due to the rough residues of peel-plies as compared to a
smooth finish from tooling surfaces different surface roughness values are found in
practice. Comparative measurements of surfaces with peel-ply residues and
surfaces subject to subsequent machining did not reveal substantial differences in
the measured absolute strain values. However, slight differences were found in the
width of the strain distribution (i.e., the scatter around the absolute value) within the
same area used for averaging.
3.4 Application to Composites 113

fiber orientation

evaluation area

a 2500

2000
stress [MPa]

1500 E11 = 149.2 GPa

1000

500

0
0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8
axial strain [%]

b –0.20
ν12 = 0.34
transverse strain [%]

–0.15

–0.10

–0.05

0.00
0.0 0.1 0.2 0.3 0.4 0.5 0.6
axial strain [%]

Fig. 3.43 Drawing of specimen used for tensile testing and exemplary stress–strain curve to
evaluate modulus (a) and exemplary diagram of axial and transverse strain (b) to evaluate
Poisson’s ratio υ12

As exemplary evaluation, strain values in parallel to the fiber axis and perpen-
dicular to the fiber axis are averaged in the marked area and are used to evaluate the
modulus and the Poisson’s ratio υ12 as shown in Fig. 3.43.
For tensile testing perpendicular to the fiber axis 3D-DIC can be used to measure
the transverse modulus and the Poisson’s ratio υ21 or the Poisson’s ratio υ23. For the
114 3 Digital Image Correlation

evaluation area fiber orientation

fiber orientation

evaluation area

Fig. 3.44 Position of speckle pattern used to measure values for transverse modulus and Poisson’s
ratio υ21 (top) and the Poisson’s ratio υ23 (bottom) including evaluation area

measurement, standard specimens in accordance with DIN-EN-ISO 527-5, DIN EN


2597, or similar standards are used. As seen for the specimen configuration in
Fig. 3.44, a speckle pattern can be applied to the flat side of the laminate to measure
the Poisson’s ratio υ21. Applying a speckle pattern to the thin side of the laminate
allows to measure the Poisson’s ratio υ23 (cf. Fig. 3.44 bottom). This measurement
is particularly relevant to establish the 3D elasticity tensor of a unidirectional fiber
reinforced polymer. For measurement of strain fields on the thin side of the laminate
a sufficiently high resolution of the subset regions across the 2 mm thickness
direction is required to avoid influences of the free edges. Therefore the averaging
area chosen slightly offset to the side surfaces, so a border of approximately the size
of the plate thickness is neglected for the averaging procedure.
An exemplary configuration for the 3D-DIC setup used for such measurements is
given in Table 3.2. To derive the global strain field, the given resolution is
comparatively high. As shown in [1] lower resolutions will still yield comparable
results to conventional extensometers.

3.4.1.2 Compressive Testing of Unidirectional Fiber Reinforced


Polymers

Similar in concept to tensile testing, evaluation of the strain field in compression


testing as function of the fiber orientation and side surface allows measuring the
Poisson’s ratios υ12, υ13, υ21, and υ23. It is worth noting that the Poisson ratio’s υ13
and υ23 are usually not measured in experimental setups according to ASTM D
6641, DIN EN 2850, DIN EN ISO 14126, or similar standards. As seen in Fig. 3.45,
the specimen types correspond to the respective standards and the speckle pattern is
3.4 Application to Composites 115

fiber orientation

evaluation area

a
160
E22 = 8.1 GPa
140

120
stress [MPa]

100

80

60

40

20

0
0.0 0.5 1.0 1.5 2.0 2.5
b negative axial strain [%]
1.5
transverse strain [%]

1.0 ν23 = 0.53

0.5

0.0
0.0 0.5 1.0 1.5 2.0 2.5
negative axial strain [%]

Fig. 3.45 Drawing of specimen used for compressive testing and exemplary stress–strain curve to
evaluate transverse modulus (a) and exemplary diagram of axial and transverse strain to measure
Poisson’s ratio υ23 (b)

applied to the respective surfaces of the specimen as described before. An exem-


plary stress–strain curve to obtain the transverse compressive modulus and a plot of
axial and transverse strain to obtain the Poisson’s ratio υ23 is shown at the bottom of
116 3 Digital Image Correlation

Table 3.3 Acquisition settings of DIC system for compression tests of unidirectional fiber
reinforced composites
Measurement setup Value
Camera Toshiba CMOS camera CSC12M25BMP19-01B with 4096  3072 pixels
Lenses Titanar, focal length 2.8/50 mm
Lighting KSP 0495-0001A LED 20 W/24 V white 30
Filter Polarization-filter Schneider-Kreuznach
Field of view (103.4  77.4) mm2
Scale factor 0.025 mm/px
Subset size 19  19 pixel
Acquisition rate 2 Hz
Software ARAMIS
Configuration 3D

Fig. 3.45. In this test configuration it is highly recommended to use 3D-DIC


measurements instead of 2D-DIC measurements. The reason for this recommenda-
tion is the strong tendency of fiber reinforced polymers to show unintended
buckling modes during compressive loading, which might cause substantial out-
of-plane motion with respect to the focal plane of the camera system. This could
cause potentially high errors in 2D-DIC measurements. At the same time, some of
the established standards involve some monitoring of the specimen buckling.
ASTM D 6641 requires measurement of the deformation values using two strain
gages mounted on each of the two specimen sides. If more than 10 % deviation
between the deformations of the two strain gages is observed, the test results are
rejected. Similar measurements can be performed using 3D-DIC and can be used to
assure, the specimen has failed under acceptable compressive strain states. Typical
DIC configurations confirm with those given in Table 3.3. In comparison to the
tensile test the main difference is the smaller area used for extraction of the strain
field. Therefore, a fine speckle pattern and higher resolution in px/mm is required
for these measurements.

3.4.1.3 V-Notched Rail Shear Testing of Unidirectional Fiber


Reinforced Polymers

Another reasonable application for full strain field measurements is to assure


specific strain conditions during testing. Especially for testing under shear loading
there is a long history in development of test rigs suitable for producing constant
shear stress as recently reviewed by [51, 52]. One of the key requirements for
unidirectional fiber reinforced polymers is the homogeneity of the produced shear
stress field. While this can be achieved with various rail shear test fixtures, the usage
of bolts to fix the specimen often induced preliminary failure. The test concept
favored by ASTM D 7078 uses a combination of a rail-shear type movement and
Iosipescu-type notch geometry to induce a zone of theoretically pure shear stress in
3.4 Application to Composites 117

Speckle pattern averaging strain gages

Fig. 3.46 Speckle pattern position in V-Notched rail shear configuration (left) and strain gage
positions with averaging area for DIC measurements marked in red (right)

Table 3.4 Acquisition settings of DIC system for V-notched rail shear tests of unidirectional fiber
reinforced composites
Measurement setup Value
Camera Toshiba CMOS camera CSC12M25BMP19-01B with 4096x3072 pixels
Lenses Titanar, focal length 2.8/50 mm
Lighting KSP 0495-0001A LED 20 W/24 V white 30
Filter Polarization-filter Schneider-Kreuznach
Field of view (81.9  61.4) mm2
Scale factor 0.02 mm/px
Subset size 19  17 pixel square
Acquisition rate 2 Hz
Software ARAMIS
Configuration 3D

the region between the notches. However, the standard recommends measurement
of the shear strain value in the center of the specimen using two strain gages
mounted in 45 orientation relative to the vertical direction. The resulting mea-
surement of the strain values is likely error-prone if the shear stress zone is not as
sharp as theoretically expected. Figure 3.47 shows a comparison between two
specimens of the same unidirectional fiber reinforced polymer with fiber orientation
in parallel to the movement of the fixture (vertical direction in Fig. 3.46) measured
using the DIC setup given in Table 3.4.
Whereas one specimen shows a rather undefined shear strain field (specimen A),
the other specimen exhibits a sharp concentration of the shear strain field as
expected from theory (specimen B). Failure in specimen A occurs preliminary
due to crack initiation offset to the notch position, which is likely due to defects
present at that position. If the shear strain is evaluated at the marked position, the
resulting stress–strain curves shown in Fig. 3.47 are also falsified by this shear
strain field. Therefore, direct DIC observation of the strain field can be used as
additional factor to judge on the validity of the measurement.
118 3 Digital Image Correlation

a
45 specimen A
specimen B
40

35
shear stress [MPa]

30

25

20

15

10

0
0 2 4 6 8 10 12 14 16 18 20 22 24
shear strain [%]

specimen A specimen B

Fig. 3.47 Measured stress–strain curve for V-notched rail shear specimens including shear strain
field for two different specimens at a shear strain of 4 %

3.4.2 Detection of Failure Onsets

In the context of fracture mechanics, the detection of first failure onsets is usually of
huge relevance to obtain fracture toughness values for composite materials. Espe-
cially for brittle matrix materials such as epoxy, the first onset of crack growth is
often used to deduce the critical fracture toughness value as relevant material
property. For more ductile matrix materials such as some thermoplastic polymers,
some of the established experimental setups in fracture mechanics may fail in their
concept or in their interpretation. In the following some examples of test standards
are presented to demonstrate these challenges. Therefore, using the detection of the
3.4 Application to Composites 119

full strain field during loading, the central aim in the next applications is not the
quantitative measurement of the strain field or the replacement of strain gages, but
the correct interpretation of the strain concentration shortly before and after crack
growth. This is somewhat different to approaches to quantitatively measure the
strain field around crack tips to directly calculate the J-integral values during crack
growth [1].

3.4.2.1 Short-Beam Shear Tests

One example of configurations being used frequently in composites testing is the


short-beam shear test. Although this test configuration is not meant to measure a
true material property relevant to fracture mechanics, it still is a well-suited method
being used in quality assurance and screening of materials. In the experiment the
load is applied at the center of the specimen utilizing a load nose midways between
the supports (cf. Fig. 3.48).
The apparent interlaminar shear strength value τILSS is related to the onset of
delamination under shear forces parallel to the layers of the laminate and is
calculated by

3Fonset
τILSS ¼ ð3:20Þ
4bh

using the load at the moment of delamination onset Fonset, the width of the specimen
b, and the thickness of the specimen h.
During the test, samples are loaded with a constant displacement rate of the
loading nose of 1 mm/min and at the same time the speckle pattern on the side
surface of the specimens was monitored by 3D-DIC techniques as seen in Fig. 3.49
using the parameters given in Table 3.5.
As shown for the image in Fig. 3.49, sudden increase of the shear angle in
positive and negative direction indicates interlaminar failure at the center plane.
The first signature in the measured load–displacement curve coincides with the
delamination onset. With delamination onset (and growth), the load drops to a
lower value. As the bending stiffness of the specimen is reduced, the slope of the
force–displacement curve decreases after delamination onset. Usually further load
drops can be observed when loading the specimen further. Although this observa-
tion does seem trivial, the usual procedure recommended in standards ISO 14130,

Fig. 3.48 Schema of short load nose


beam shear test
configuration including Speckle pattern
speckle pattern position

supports
120 3 Digital Image Correlation

2.5

2.0

1.5
force [kN]

1.0

0.5

0.0
0.0 0.1 0.2 0.3 0.4 0.5
displacement [mm]

Fig. 3.49 Force–displacement curve including DIC image at onset of interlaminar crack growth
with false-color indicating shear angle. First onset is marked in red in force–displacement curve

Table 3.5 Acquisition settings of DIC system for short beam shear tests
Measurement setup Value
Camera Toshiba CMOS camera CSC12M25BMP19-01B with 4096  3072 pixels
Lenses Titanar, focal length 2.8/50 mm
Lighting KSP 0495-0001A LED 20 W/24 V white 30
Filter Polarization-filter Schneider-Kreuznach
Subset size 11  8 pixel square
Acquisition rate 1 Hz
Software ARAMIS
Configuration 2D

ASTM D 2344, DIN EN 2377, and DIN EN 2563 is the usage of the force maximum
of the curve as value of Fonset. This systematically overestimates the apparent
interlaminar shear strength for the present example by 4.9 %. Also, ductile matrix
materials often show substantial deviations to the sequence of damage expected by
3.4 Application to Composites 121

load nose Speckle pattern

notch tip
supports

Fig. 3.50 Schema of ENF test configuration including speckle pattern position

the test standards and therefore cannot be interpreted based solely on load–dis-
placement curves. For such cases, the DIC measurements are a suitable alternative
to securely assign signatures in the load–displacement profile to the occurrence of
interlaminar failure [38].

3.4.2.2 End-Notched Flexure Tests

Another typical test configuration used in fracture mechanics of fiber reinforced


composites is the end-notched flexure (ENF) test as seen in Fig. 3.50. The aim of the
test is to initiate an interlaminar crack under a mode II load situation. Although
there is an ongoing discussion if this test configuration does produce a pure mode II
load situation [53, 54], it is used here as another representative example for
measurement of material properties utilizing DIC techniques.
Conceptually similar to the interpretation of the short-beam shear test, the ENF
test uses the force value Fonset at the onset of interlaminar crack growth to derive the
critical mode II fracture toughness according to

3mc Fonset 2 a0 2
GIIc ¼ ð3:21Þ
2b

This uses the initial crack length a0, the width of the specimen b, and a compliance
calibration factor mc for the specimen evaluated prior to testing. Therefore, the only
critical parameter to obtain during the final test procedure is the value of Fonset.
To demonstrate the usage of DIC in this context, ENF tests were carried out
using a layup consisting of 16 layers of Sigrafil CE1250-230-39 unidirectional
epoxy-based prepreg with a pre-crack prepared at the interface at the midplane.
Samples of 170 mm length, 25 mm width, and 3.6 mm thickness were cut from the
cured plate. Simultaneously, images were recorded for DIC following the param-
eters reported in Table 3.6.
As seen for the example in Fig. 3.51, the force–displacement curve is straight-
forward to interpret in accordance with existing standards. Indicated by the two
subsequent DIC images at the time of maximum force, the first load drop is clearly
associated with the onset of interlaminar crack growth. However, for some cases it
Table 3.6 Acquisition settings of DIC system for ENF tests
Measurement setup Value
Camera Toshiba CMOS camera CSC12M25BMP19-01B with 4096  3072 pixels
Lenses Titanar, focal length 2.8/50 mm
Lighting KSP 0495-0001A LED 20 W/24 V white 30
Filter Polarization-filter Schneider-Kreuznach
Subset size 11  11 pixel
Acquisition rate 1 Hz
Software ARAMIS
Configuration 2D

0.6

0.4
force [kN]

0.2

0.0
0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8
displacement [mm]

Fig. 3.51 Force–displacement curve including two DIC images before and after onset of
interlaminar crack growth with false-color indicating shear angle. Onset of crack growth is marked
in red in force–displacement curve
3.4 Application to Composites 123

0.6

0.4
force [kN]

0.2

0.0
0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8
displacement [mm]

crack progression

Fig. 3.52 Force–displacement curve including two DIC images before and after onset of
interlaminar crack growth with false-color indicating shear angle. Onset of crack growth is marked
in red in force–displacement curve

is harder to determine the correct onset of crack growth as shown in the second
example of a unidirectional laminate in Fig. 3.52. Here the change of slope in the
load–displacement curve indicates a change in the material, but is hard to evaluate
in quantitative terms. The DIC images clearly reveal the onset of crack growth at
the position marked in the load–displacement curve. For this example the onset of
interlaminar crack growth was found to precede the force maximum. Therefore the
evaluation according to common test standards would overestimate the onset load
by 6.3 %, while the combination with DIC allows for evaluation of a more relevant
and conservative load level.
124 3 Digital Image Correlation

3.4.3 Detectability of Failure Mechanisms

In Sect. 3.3 it was demonstrated that certain types of defects in fiber reinforced
materials can be found by DIC. However, these investigations were limited to
artificial defects as strain concentrators. In the following, some results are presented
for tensile loading of carbon fiber reinforced materials monitored by 3D-DIC with
parameters as given in Table 3.2. Tensile tests on unidirectional fiber reinforced
polymers were carried out in accordance with DIN EN 2561. As shown in the
sequence of images in Fig. 3.53, it is easy to spot macroscopic crack growth in the
DIC images due to the local decrease of the 2nd principal strain. In this tensile test
parallel to the fiber axis, such crack growth is not intended prior to ultimate failure,
since such splicing significantly affects the stress distribution in the cross section of
the test specimen. Hence ultimate failure below the real material strength is quite
likely. In this context, DIC imaging can act as indicator to spot such failure modes
long before (i.e., at εz,avg ¼ 0.30 %) the ultimate rupture strain of 1.30 % for the
present example. In comparison to DIC, the detection of such crack growth by
conventional optical methods suffers from the weak perceptibility of inter-fiber
cracks in carbon fiber reinforced materials. However, the detection of this type of
crack is trivial by DIC, since the inter-fiber cracks typically span the whole

2nd principal strain

y
fiber axis

15 mm

εz, avg = 0.15 % εz, avg = 0.30 % εz, avg = 0.47 % εz, avg = 0.68 % εz, avg = 1.05 %

Fig. 3.53 Strain field of 2nd principal strain at various load levels indicated by the strain averaged
over the full area seen εz,avg. Growth of inter-fiber crack parallel to fiber axis is clearly seen by
presence of areas with strain values < 0
3.4 Application to Composites 125

50 MPa 150 MPa 200 MPa

2 mm 2 mm 2 mm

Fig. 3.54 Optical microscopy results for [0/902]sym stacking sequence of GFRP. Crack growth in
the 90 layers at three different load levels is seen well due to the change in reflectivity of the crack
surface

Table 3.7 Acquisition settings of DIC system for tensile tests on [0/90]sym fiber reinforced
composites
Measurement setup Value
Camera Toshiba CMOS camera CSC12M25BMP19-01B with 4096  3072 pixels
Lenses Titanar, focal length 2.8/50 mm
Lighting KSP 0495-0001A LED 20 W/24 V white 30
Filter Polarization-filter Schneider-Kreuznach
Field of view (107.4  82.1) mm2
Scale factor 0.026 mm/px
Subset size 30  20 pixel square
Acquisition rate 1 Hz
Software ARAMIS
Configuration 3D

thickness of the unidirectional laminate in this load situation. Therefore the crack is
also at the surface level, which is a situation easy to detect by DIC as discussed in
Sect. 3.3.
As discussed in Sect. 3.3, the detectability of failure types below the first ply or
even deeper below the surface of a laminate is much more challenging than the
previous example. For glass fiber reinforced materials the optically transparent
fibers and matrix systems allow a direct observation of inter-fiber cracks in the
plies 90 to the load axis. This is shown in exemplary microscopy investigations in
Fig. 3.54. With increasing load, an increasing number of inter-fiber cracks can be
observed due to the change in optical reflectivity of the crack surface. For carbon
fiber reinforced materials, such direct observation is not possible. Therefore DIC
can be used as suitable alternative.
To this end tensile tests of an [0/90]sym stacking of an epoxy prepreg system
were monitored using the DIC parameters given in Table 3.7. As shown in the
sequence of images in Fig. 3.55, the formation of cracks in the 90 plies is clearly
observed by the strain concentrations transverse to the load axis (vertical direction
in Fig. 3.55). In this case, the visibility originates from the increase in z-strain
values at the surface level, i.e., in the 0 layer. Due to the presence of the inter-fiber
crack below this surface layer an increased stress exposure occurs in the 0 layers,
126 3 Digital Image Correlation

z-strain z-strain z-strain

15 mm

εavg = 0.35 % εavg = 0.95 % εavg = 1.05 %

z-strain z-strain

εavg = 1.15 % εavg = 1.35 %

Fig. 3.55 Strain field of strain in z-axis at various load levels indicated by the strain averaged over
the full area seen εavg. Growth of inter-fiber cracks in the 90 layers is clearly seen by the presence
of areas of increased strain values

which translates into a local increase of strain. However, the appearance of these
cracks is smoother than for optical imaging. This is due to the resolution of the
subsets used in DIC and the smearing of the strain field throughout the surface ply.
Therefore detectability of such cracks is limited to the outermost plies as discussed
in Sect. 3.3. Moreover, the onset of inter-ply delamination at the crack tip further
reduces the strain concentration and therefore may reduce the visibility of the
existing crack.

References

1. Sutton, M.A., Orteu, J.J., Schreier, H.: Image Correlation for Shape, Motion and Deformation
Measurements: Basic Concepts, Theory and Applications. Springer, Berlin (2009)
2. Pan, B., Qian, K., Xie, H., Asundi, A.: Two-dimensional digital image correlation for in-plane
displacement and strain measurement: a review. Meas. Sci. Technol. 20, 062001 (2009)
References 127

3. Peters, W.H., Ranson, W.F.: Digital imaging techniques in experimental stress analysis. Opt.
Eng. 21, 213427 (1982)
4. Fouinneteau, M.R.C., Pickett, A.K.: Shear mechanism modelling of heavy tow braided
composites using a meso-mechanical damage model. Compos. Part A Appl. Sci. Manuf. 38,
2294–2306 (2007)
5. Fouinneteau, M.R.C., Pickett, A.K.: Failure characterisation of heavy tow braided composites
using digital image correlation (DIC). In: Applied Mechanics and Materials, pp. 399–406,
(2006). doi:10.4028/www.scientific.net/AMM.5-6.399
6. Geers, M.G.D., de Borst, R., Peijs, T.: Mixed numerical-experimental identification of
non-local characteristics of random-fibre-reinforced composites. Compos. Sci. Technol. 59,
1569–1578 (1999)
7. Machida, K., Sato, M., Ogihara, S.: Application of stress analysis by digital image correlation
and intelligent hybrid method to unidirectional CFRP laminate. In: Key Engineering Materials,
pp. 1121–1124, (2007). doi:10.4028/www.scientific.net/KEM.345-346.1121
8. Nicoletto, G., Anzelotti, G., Riva, E., Roncella, R.: Mesomechanic strain fields in woven
composites: experiments vs. FEM modeling. In: Proceedings of 3rd Workshop on Optical
Measurement Techniques for Structures and Systems (OPTIMESS’2007). Leuven, Belgium
(2007)
9. McGowan, D.M., Ambur, D.R., Hanna, T.G., McNeill, S.R.: Evaluating the compressive
response of notched composite panels using full-field displacements. J. Aircr. 38, 122–129
(2001)
10. Périé, J.-N., Calloch, S., Cluzel, C., Hild, F.: Analysis of a multiaxial test on a C/C composite
by using digital image correlation and a damage model. Exp. Mech. 42, 318–328 (2002)
11. Rae, P.J., Palmer, S.J.P., Goldrein, H.T., Lewis, A.L., Field, J.E.: White-light digital image
cross-correlation (DICC) analysis of the deformation of composite materials with random
microstructure. Opt. Lasers Eng. 41, 635–648 (2004)
12. Ruggy, K.L., Cox, B.N.: Deformation mechanisms of dry textile preforms under mixed
compressive and shear loading. J. Reinf. Plast. Compos. 23, 1425–1442 (2004)
13. Koerber, H., Xavier, J., Camanho, P.P.: High strain rate characterisation of unidirectional
carbon-epoxy IM7-8552 in transverse compression and in-plane shear using digital image
correlation. Mech. Mater. 42, 1004–1019 (2010)
14. Koerber, H., Vogler, M., Kuhn, P., Camanho, P.P.: Experimental characterisation and model-
ling of non-linear stress-strain behaviour and strain rate effects for unidirectional carbon-
epoxy. In: ECCM16—16th European Conference on Composite Materials, pp. 1–8. Sevilla,
Spain (2014)
15. Xavier, J., Sousa, A.M.R., Morais, J.J.L., Filipe, V.M.J., Vaz, M.: Measuring displacement
fields by cross-correlation and a differential technique: experimental validation. Opt. Eng. 51,
043602 (2012)
16. Lucas, B.D., Kanade, T.: An iterative image registration technique with an application to
stereo vision. In: Proceedings of the 1981 DARPA Image Understanding Workshop,
pp. 121–130, (1981)
17. Sutton, M.A., Yan, J.H., Tiwari, V., Schreier, H.W., Orteu, J.J.: The effect of out-of-plane
motion on 2D and 3D digital image correlation measurements. Opt. Lasers Eng. 46, 746–757
(2008)
18. Fowles, G.R.: Introduction to Modern Optics. Dover, New York (2012)
19. Pedrotti, F.L., Pedrotti, L.S.: Introduction to Optics, 2nd edn. Prentice Hall, Upper Saddle
River (1993)
20. Schreier, H.W.: Systematic errors in digital image correlation caused by intensity interpola-
tion. Opt. Eng. 39, 2915 (2000)
21. Dupré, J.C., Bornert, M., Robert, L., Wattrisse, B.: Digital image correlation: displacement
accuracy estimation. In: ICEM 14—14th International Conference on Experimental Mechan-
ics. Poitiers, France (2010)
128 3 Digital Image Correlation

22. Wang, Y.Q., Sutton, M.A., Bruck, H.A., Schreier, H.W.: Quantitative error assessment in
pattern matching: effects of intensity pattern noise, interpolation, strain and image contrast on
motion measurements. Strain 45, 160–178 (2009)
23. Amiot, F., Bornert, M., Doumalin, P., Dupré, J.-C.C., Fazzini, M., Orteu, J.-J.J., Poil^ane, C.,
Robert, L., Rotinat, R., Toussaint, E., Wattrisse, B., Wienin, J.S.: Assessment of digital image
correlation measurement accuracy in the ultimate error regime: main results of a collaborative
benchmark. Strain 49, 483–496 (2013)
24. Tong, W.: An evaluation of digital image correlation criteria for strain mapping applications.
Strain 41, 167–175 (2005)
25. Roux, S., Hild, F.: Stress intensity factor measurements from digital image correlation: post-
processing and integrated approaches. Int. J. Fract. 140, 141–157 (2006)
26. Becker, T., Splitthof, K., Siebert, T., Kletting, P.: Technical Note—T-Q-400-Accuracy-
3DCORR-003-EN, (2006)
27. Yeh, H.-Y., Chern, C.: The Yeh-Stratton criterion for stress concentrations in fiber-reinforced
composite materials. J. Compos. Mater. 32, 141–157 (1998)
28. Pilkey, W.D., Pilkey, D.F.: Peterson’s Stress Concentration Factors. Wiley, Hoboken (2008)
29. Griffith, A.A.: The phenomena of rupture and flow in solids. Philos. Trans. R. Soc. Lond. Ser.
A 221, 163–198 (1921)
30. Irwin, G.: Analysis of stresses and strains near the end of a crack traversing a plate. J. Appl.
Mech. 24, 361–364 (1957)
31. Tada, H., Paris, P.C., Irwin, G.R.: The Stress Analysis of Cracks Handbook, 3rd edn. ASME,
New York, (2000)
32. O’Higgins, R.M., McCarthy, M.A., McCarthy, C.T.: Comparison of open hole tension char-
acteristics of high strength glass and carbon fibre-reinforced composite materials. Compos.
Sci. Technol. 68, 2770–2778 (2008)
33. Kushch, V.I., Sevostianov, I., Mishnaevsky, L.: Stress concentration and effective stiffness of
aligned fiber reinforced composite with anisotropic constituents. Int. J. Solids Struct. 45,
5103–5117 (2008)
34. Chang, K.-Y., Llu, S., Chang, F.-K.: Damage tolerance of laminated composites containing an
open hole and subjected to tensile loadings. J. Compos. Mater. 25, 274–301 (1991)
35. Nicoletto, G., Anzelotti, G., Riva, E.: Mesoscopic strain fields in woven composites: experi-
ments vs. finite element modeling. Opt. Lasers Eng. 47, 352–359 (2009)
36. Godara, A., Raabe, D., Bergmann, I., Putz, R., M€uller, U.: Influence of additives on the global
mechanical behavior and the microscopic strain localization in wood reinforced polypropylene
composites during tensile deformation investigated using digital image correlation. Compos.
Sci. Technol. 69, 139–146 (2009)
37. Schorer, N., Sause, M.G.R.: Identification of failure mechanisms in CFRP laminates using 3D
digital image correlation. In: 20th International Conference on Composite Materials, pp. 1–8.
Copenhagen, Denmark (2015)
38. Monden, A., Sause, M.G.R., Hartwig, A., Hammerl, C., Karl, H., Horn, S.: Evaluation of
surface modified CFRP-metal hybrid laminates. In: Euro Hybrid Materials and Structures
2014, pp. 1–8. Stade, Germany (2014)
39. Priess, T., Sause, M.G., Fischer, D., Middendorf, P.: Detection of delamination onset in laser-
cut carbon fiber transverse crack tension specimens using acoustic emission. J. Compos.
Mater. 49, 2639–2647 (2015)
40. Scott, A.E., Sinclair, I., Spearing, S.M., Mavrogordato, M., Bunsell, A.R., Thionnet, A.:
Comparison of the accumulation of fibre breaks occurring in a unidirectional carbon/epoxy
composite identified in a multi-scale micro-mechanical model with that of experimental
observations using high resolution computed tomography. In: Matériaux 2010, pp. 1–9.
Nantes, France (2010)
41. Scott, A.E., Mavrogordato, M., Wright, P., Sinclair, I., Spearing, S.M.: In-situ fibre fracture
measurement in carbon-epoxy laminates using high resolution computed tomography.
Compos. Sci. Technol. 71, 1471–1477 (2011)
References 129

42. De Angelis, G., Meo, M., Almond, D.P., Pickering, S.G., Angioni, S.L.: A new technique to
detect defect size and depth in composite structures using digital shearography and
unconstrained optimization. NDT E Int. 45, 91–96 (2012)
43. Pezzoni, R., Krupka, R.: Laser-shearography for non-destructive testing of large-area com-
posite helicopter structures. In: 15th World Conference on NDT. Roma, Italy (2000)
44. Lomov, S.V., Ivanov, D.S., Verpoest, I., Zako, M., Kurashiki, T., Nakai, H., Molimard, J.,
Vautrin, A.: Full-field strain measurements for validation of meso-FE analysis of textile
composites. Compos. Part A Appl. Sci. Manuf. 39, 1218–1231 (2008)
45. Hung, Y.Y.: Shearography for non-destructive evaluation of composite structures. Opt. Lasers
Eng. 24, 161–182 (1996)
46. Hoffmann, K.: An Introduction to Measurements Using Strain Gages. Hottinger Baldwin
Messtechnik GmbH, Darmstadt, Germany (1987)
47. Robert, L., Nazaret, F., Cutard, T., Orteu, J.-J.: Use of 3-D digital image correlation to
characterize the mechanical behavior of a fiber reinforced refractory castable. Exp. Mech.
47, 761–773 (2007)
48. Grédiac, M., Pierron, F., Avril, S., Toussaint, E.: The virtual fields method for extracting
constitutive parameters from full-field measurements: a review. Strain 42, 233–253 (2008)
49. Yoneyama, S., Morimoto, Y., Takashi, M.: Automatic evaluation of mixed-mode stress
intensity factors utilizing digital image correlation. Strain 42, 21–29 (2006)
50. Hild, F., Roux, S.: Digital image correlation: from displacement measurement to identification
of elastic properties—a review. Strain 42, 69–80 (2006)
51. Basan, R.: Untersuchung der intralaminaren Schubeigenschaften von
Faserverbundwerkstoffen mit Epoxidharzmatrix unter Ber€ ucksichtigung nichtlinearer Effekte.
Ph.D. Thesis, Technical University Berlin (2011)
52. Trappe, V., Basan, R., Grasse, F.: Stiffness and fracture of shear loaded laminates with
unidirectional and biaxial fibre orientation investigated with a picture frame test. In: 16th
European Conference on Composite Materials, pp. 22–26. Seville (2014)
53. Blackman, B.R.K., Brunner, A.J., Williams, J.G.: Mode II fracture testing of composites: a
new look at an old problem. Eng. Fract. Mech. 73, 2443–2455 (2006)
54. Brunner, A.J., Blackman, B.R.K., Davies, P.: A status report on delamination resistance testing
of polymer-matrix composites. Eng. Fract. Mech. 75, 2779–2794 (2008)
Chapter 4
Acoustic Emission

Acoustic emission (AE) analysis is about the detection and interpretation of


ultrasonic waves caused by rapid internal displacements. In the context used herein,
the formation and propagation of cracks in fiber reinforced materials are understood
as the most relevant acoustic emission source. This acoustic emission source is
generally described as spatial crack surface oscillation [1, 2]. During propagation of
the emitted acoustic wave, the characteristics of the signal (e.g., frequency content)
suffer from attenuation, dispersion, and propagation in guiding media. In addition,
the characteristics of the signals detected at the surface of the solid are further
altered by the detection process using piezoelectric sensors.
This chapter starts with a short introduction to the principle of operation followed
by a sequential review on the acoustic emission source, aspects of wave propagation
in guided media, and the signal detection process. Subsequently, signal classifica-
tion techniques and source localization algorithms are discussed with a particular
focus on recent developments. The chapter closes with some applications of acous-
tic emission as in situ technique to monitor failure of fiber reinforced composites.

4.1 Principle of Operation

AE signals originate from sudden microscopic displacements inside solid media.


One major source for AE in fiber reinforced materials is the initiation and growth of
damage. Microscopically this relates to the generation and propagation of cracks
inside the matrix material, along the interface between matrix and fiber or the
rupture of the fiber filaments. Another relevant AE source found in fiber reinforced
polymers is friction of existent crack walls, which is frequently encountered in
unloading and reloading of damaged composites. In any case, in situ monitoring by
AE sensors on the structure under loading can identify active areas in the composite
and can identify their time of occurrence. In the past, the detection of AE signals
and the according interpretation has led to a variety of approaches and analysis

© Springer International Publishing Switzerland 2016 131


M.G.R. Sause, In Situ Monitoring of Fiber-Reinforced Composites,
Springer Series in Materials Science 242, DOI 10.1007/978-3-319-30954-5_4
132 4 Acoustic Emission

AE signal

signal counting localization classification


prediction (PoD) (UoL) (UoC)

amount position type

of damage

material
analysis

Fig. 4.1 Analysis methods to use with AE signals and AE signal prediction as forward approach

methods. One possibility to categorize the potential of AE analysis techniques is


shown in Fig. 4.1. Starting with the detected AE signals, a first distinction can be
made between counting the AE activity, localizing the AE source position, and
identifying classes of AE signals. If carried out successfully, this yields the amount
of damage, the position of the damage zone, and the type of damage. Thus, the
combination of these three types of information can yield valuable information for
the interpretation of failure of fiber reinforced composites. The application of AE to
this intent is not essentially new and has seen substantial development in the past
decades. A good review on the present status in this field can be found in, e.g., [3]
for the work until 2000 or in [4] for the ongoing research since 2000.
For the analysis options illustrated in Fig. 4.1, counting the number of signals is
considered to be a standard technique in AE analysis since this is already inherent to
the detection of the AE signals. However, the counting of the signals is associated
with an uncertainty, given by the probability of detection (PoD).
The localization of source positions is already well established but comes with
additional challenges, when acoustically anisotropic and inhomogeneous media
like fiber reinforced polymers are considered [5]. Established localization tech-
niques can yield useful source positions in laboratory size fiber reinforced polymer
specimens. However, their application in realistic structures including hybrid joints
or multi-material combinations is still challenging; therefore, every calculated
source location is subject to an uncertainty of location (UoL).
The identification of a particular type of damage in fiber reinforced polymers has
seen several substantial advancements in the last two decades. For discussion of this
analysis method, it is suitable to distinguish between signal classification and
source identification. The first task is the grouping of signals based on their
similarity relative to each other. This is often achieved by analysis of signal
4.2 Source Mechanics 133

features, i.e., parameters calculated from the detected signals. The methods to
group AE signals range from very simple approaches comprising discrete feature
values to automated or semiautomated pattern recognition strategies. However,
every classification result is subject to uncertainty of the assignment of the signals
to a particular group. This is named uncertainty of classification (UoC).
After grouping of the signals, the second task is the assignment of one group of
signals to a particular source mechanism. This is typically done by phenomenolog-
ical observations [6], by comparative measurement of test specimens with known
types of AE sources [7, 8] or subsequent microscopy [9]. Recent advances also
allow the forward prediction of the emitted AE signal of a specific source type by
analytical methods [10] or finite element modeling (FEM) [11–13].
As shown in Fig. 4.1, the possibility to predict the AE signal of a known type of
failure is key to establish traceable acoustic emission analysis techniques and to add
a sound physically based theory for generation of acoustic emission signals.

4.2 Source Mechanics

The formation and propagation of cracks in solid media are a field of research that
has been active for decades. Still, the theoretical description of the physics at the
crack tip and the crack dynamics are active fields of research [14, 15]. The gener-
ation of acoustic emission signals due to the crack motion is a closely related topic.
Despite of the broad range of technical applications, only little work has been
performed recently to advance the understanding of the physical processes involved
in the generation of acoustic emission.
In order to interpret the detected acoustic emission signals in terms of their
relevance to material failure, it is required to have concise knowledge of the
underlying source mechanics. In the past, various valuable attempts have been
made to provide a theoretical description of acoustic emission sources.
In general, the excitation process of an elastic wave as a result of crack surface
movement is given by the elastic wave equation as long as the deformation remains
elastic, i.e., for small displacements [16]. In the case of crack growth, the stress
level at the newly formed surface area drops from the initial value σ frac to zero. As a
result the crack surface is deflected within a characteristic time te and starts to
oscillate around its new energetic steady state [17]. The crack kinematic behavior
can then be described by a crack motion vector b (Burgers vector) and a vector
normal to the crack surface d. As demonstrated by [18], the characteristics of crack
deflection are fully described by the moment tensor concept, which couples the
crack kinematic behavior with the elastic properties:
ð
f
M ¼ C  b  d dV ð4:1Þ
V

The resulting excitation of a small elastic wave released in the surrounding medium
is then derived from the theory of elastic waves [19].
134 4 Acoustic Emission

It is important to note that the duration for the initial deflection te is strictly
linked to the elastic properties of the cracking material and the local loading
conditions. This implies that the excited elastic wave is a characteristic fingerprint
for the failure process and thus a valuable source of information that is accessible
macroscopically by AE measurements. In general the duration for the initial
deflection decreases with the brittleness of the material [20] and thus determines
the oscillation frequency. Since the initial deflection time te typically is in the order
of magnitude between 108 and 104 s, the frequencies of the excited elastic waves
at the crack tip are located in the ultrasonic range between 10 kHz and 100 MHz.
While the correlation of te to the elastic properties may seem obvious, the
influence of the local loading conditions on the excited elastic wave is more
complex. Because crack formation can result from different loading conditions,
the oscillation of the crack surface is strongly influenced in intensity, direction of
radiation, and damping behavior by the type of failure.
To this end, M. Ohtsu and K. Ono have examined the correlation between the
response function on the surface of linear elastic solids and dislocations arising
from acoustic emission sources located within [21, 22]. According to the general-
ized theory of acoustic emission, the elastic wave amplitude observed at position r
due to a microscopic displacement d(t) located at rsource in an isotropic medium can
be written as

uðr; tÞ ¼ Γðr; rsource ; tÞ C  ΔV  dðrsource ; tÞ ð4:2Þ

Here denotes the convolution integral between the Green function Γ(r, rsource, t)
and elasticity tensor C times crack volume ΔV. The crack kinematics are taken into
account by the source-time function using a magnitude of displacement d and a rise-
time te. Various step-function descriptions exist, which are used to define the three-
dimensional spatial displacement of the crack surface during crack formation
[2, 20, 22–24]. In particular the rise-time of the initial crack surface displacement
is an essential parameter to model the crack surface motion [25].
However, there are no reports in literature of successful measurements of rise-
times of real acoustic emission sources, e.g., due to cracking. Instead, the rise-time
is typically estimated based on the elastic properties of the bulk material. Despite of
that drawback, this type of source modeling has been successfully applied to many
cases and the basic concept has been used within the generalized theory of acoustic
emission by Ono and Ohtsu [21, 22], the work of Scruby and Wadley [26–28], and
numerous other analytical descriptions [2, 10, 20, 29, 30].
In recent years it has become convenient to use numerical methods to model
acoustic emission sources. In this field, Prosser, Hamstad, and Gary applied FEM to
simulate acoustic emission sources based on body forces acting as point source in a
solid [31, 32]. Hora and Cervena investigated the difference between nodal sources,
line sources, and cylindrical sources to build geometrically more representative
acoustic emission sources [33]. At the same time, a finite element approach was
proposed utilizing an acoustic emission source model taking into account the
geometry of a crack and the inhomogeneous elastic properties in the vicinity of
the acoustic emission source [24].
4.2 Source Mechanics 135

In the following several investigation results as achieved by means of FEM are


used to demonstrate some of the key principles in acoustic emission source
mechanics. The main advantage of FEM in this context is that precise control of
the source characteristics (such as rise-time and source orientation), source loca-
tion, and propagation distance and direction is feasible. Another closely related
advantage is the possibility to compute the surface displacement directly, whereas
in the experiment, this evaluation suffers from the characteristics of the AE sensor.

4.2.1 AE Rise-Times and Plate Waves

As the first step towards a better understanding of the interaction between source
types and the detected acoustic emission signals, the relationship between the
source and the guided wave formation shall be investigated. To this end, there is
a limited set of factors to consider, whose influence shall be discussed first:
• Source magnitude
• Source rise-time
• Source-sensor distance (x source position)
• Source-sensor angle (x and y source position)
• Source depth position (z source position)
To address these different factors, a simple case of an aluminum plate with
buried dipole sources is considered. This kind of investigation follows the
established procedures of previous work conducted by Hamstad et al. [25, 31, 34].
The model configuration uses a 2D-axisymmetric representation of a plate with
4.7 mm thickness and material properties of AlMg3 as given in Table B.1 in
Appendix B. As source, a dipole of 150 μm axis length using a cosine-bell source
function with varying rise-time te and strength of 1 N was modeled. The out-of-plane
displacement is calculated for a distance of 240 mm using mesh settings with
maximum edge length of 0.5 mm and a time step of 0.1 μs. All these settings were
found to yield convergent results based on the coherence level of >98 % in the
frequency range up to 1 MHz, when compared to a solution with higher mesh
resolution and lower time step. Also the computations using these settings were
found to be in very good agreement to the respective calculations in [25, 31, 34],
therefore extending these previous findings.
The influence of the source magnitude is covered by the generalized theory of
acoustic emission and is addressed in terms of FEM work in [24, 32, 35]. Since this
is not of large relevance for the discussion as long as linear elastic computations are
used, the reader is referred to the respective original work, but also to Sect. 4.2.2 for
more recent work in this field. For the present configuration, the source magnitude
was kept constant. The source rise-time dominates the bandwidth of the source and
therefore is explicitly accounted for in the present investigation following the
previous work of [25]. Effects of changes of the source-sensor distance and the
source-sensor angle are also excluded from the present investigation by focusing on
136 4 Acoustic Emission

a 2D problem, evaluating the signals at constant distance. However, the respective


effects due to different wave propagation paths are extensively discussed in
Sect. 4.3 for various configurations as faced in composite materials. As pointed
out by Downs et al., the source depth position causes a systematic shift of the
intensity contribution of the A0 wave mode as soon as the source position moves
away from the center plane of the plate [31, 32]. Therefore, even for the same type
of source, there is a systematic change of the detected mode ratio.
This influence of the depth position is emphasized in Fig. 4.2 in two Choi-
Williams distributions (CWDs) as calculated for the respective relative depth

a
0.0E+00
depth = 2.350 mm 2.5E+01
S0 -mode
800 5.0E+01
frequency [kHz]

7.5E+01
1.0E+02
600 1.3E+02
1.5E+02
400 1.8E+02
2.0E+02
2.2E+02
200 A0 -mode

0 50 100 150 200 250


displacement [fm]

0.05

0.00

–0.05
0 50 100 150 200 250
time [μs]
b
0.0E+00
depth = 1.324 mm 2.4E+04
S0 -mode
800 4.9E+04
frequency [kHz]

7.3E+04
9.8E+04
600 1.2E+05
1.5E+05
400 1.7E+05
2.0E+05
2.2E+05
200 A0 -mode
displacement [fm]

50 100 150 200


0.5
0.0
–0.5

0 50 100 150 200 250


time [μs]

Fig. 4.2 Choi-Williams distribution of two signals at 240 mm distance due to an in-plane dipole
source positioned at 2.350 mm (a) and 1.324 mm (b) below the surface
4.2 Source Mechanics 137

positions. The superimposed dispersion curves clearly allow identifying the indi-
vidual modal contributions in the present configuration, which are the fundamental
symmetric mode (S0) and antisymmetric mode (A0), often referred to as extensional
and flexural mode.
Another important factor of influence for the modal composition of an acoustic
emission signal was found to be the rise-time. As visualized in Fig. 4.87 in Sect. 4.4,
the bandwidth of the source determines if a particular wave mode can be excited or
not. If the bandwidth of the source cuts off at too low frequency, there is no
possibility to excite guided wave modes propagating preferentially above that
cutoff frequency. The bandwidth of an acoustic emission source is dominantly
influenced by its rise-time, a factor which was also investigated previously using
FEM [25]. For the present case of buried in-plane dipoles, the influence of the rise-
time was studied and is shown in exemplary CWDs for an in-plane source at the
same depth position in Fig. 4.3a, b. As reported previously, the change in rise-time
substantially changes the excited wave modes [25]. For the shortest rise-time, the
bandwidth is sufficiently high, so higher-order modes are excited. But with decreas-
ing bandwidth, the S0 mode cannot be excited beyond certain intensity and the
signal is now dominated by the A0 mode.
From the viewpoint of source identification procedures, one interesting question
is the relative influence of these two factors. Therefore, signals within a range of
depth position and a range of rise-times were calculated and signals evaluated at a
distance of 240 mm. For each of the signals, the CWD was calculated, and the
maximum intensity of the coefficients for each of the wave modes was determined
to yield a measure of intensity. Subsequently, the matrix of rise-times and relative
depth positions was interpolated to yield the 3D representation of the coefficient
intensity as shown in Fig. 4.4 for the S0 mode and in Fig. 4.5 for the A0 mode.
As seen from the intensity map of the S0 mode in Fig. 4.4, there is some influence
of the relative depth position, which is expected from the according shift from the
center plane to the surface of the plate. However, the influence of the rise-time is
apparently much more relevant, since it is almost impossible to generate a reason-
able intensity of the S0 mode above a rise-time of 2.5 μs.
For the A0 mode, the behavior is strictly different. Here the relative depth
position shows a pronounced effect as expected by the shift of source position
from the center plane to the surface of the plate. Within the selected range of rise-
times, there is no substantial influence of the rise-time for a constant depth position.
This relates to the dominant propagation of the A0 mode at low frequencies. This
frequency range is hard to cut off since it is always found in the bandwidth of
sources operating with steplike source functions.
Although these results in combination with previous publications [25, 31, 32, 34]
are only reported for an isotropic material, the key principles are still applicable to
anisotropic materials as fiber reinforced laminates. However, there are numerous
additional effects, which have to be considered during wave propagation, which are
discussed in more detail in Sect. 4.3.
138 4 Acoustic Emission

a
0.0E+00
rise-time 0.5 μs 4.5E+03
frequency [kHz] 800 9.1E+03
1.4E+04
1.8E+04
600 2.3E+04
2.7E+04
400 3.2E+04
3.6E+04
4.0E+04
200
d is p la c e m e n t [fm ]

50 100 150 200


0.5
0.0
–0.5

0 50 100 150 200 250


time [μs]
b
0.0E+00
rise-time 15.0 μs 2.2E+04
800 4.4E+04
frequency [kHz]

6.6E+04
8.8E+04
600 1.1E+05
1.3E+05
400 1.5E+05
1.8E+05
1.9E+05
200

50 100 150 200


displacement [fm]

0.5
0.0
–0.5

0 50 100 150 200 250


time [μs]

Fig. 4.3 Choi-Williams distribution of two signals at 240 mm distance due to an in-plane dipole
source positioned at 1.324 mm below the surface using a rise-time of 0.5 μs (a) and 15.0 μs (b)

4.2.2 AE Source Model Implementation for Fiber Reinforced


Materials

Despite the high value of dipole-source representation to deduce some basic


conclusions on source operation, such description still lacks in the level of detail,
when compared to real acoustic emission sources. Therefore different approaches
4.2 Source Mechanics 139

Fig. 4.4 Intensity of CWD coefficient of the S0 mode as function source depth and rise-time

Fig. 4.5 Intensity of CWD coefficient of the A0 mode as function source depth and rise-time
140 4 Acoustic Emission

a b c
+F

+F +F

-F -F

z
y static geometry static geometry change of geometry
x -F during crack growth

Fig. 4.6 Different types of acoustic emission source model descriptions used in literature
employing point sources (a) or extended sources (b) in conjunction with analytic source functions.
New source model description presented herein using dynamic changes of the source geometry
based on fracture mechanics (c) (reprinted from [13])

have been published in the past in order to improve the accuracy of acoustic
emission source models. Based on previous investigations, one can categorize the
different modeling strategies to describe acoustic emission sources of crack prop-
agation as shown in Fig. 4.6. The first type of source models considers point-like
sources explicitly defining the source dynamics utilizing analytical source functions
(cf. Fig. 4.6a). As the second type, we can name those attempts that have been made
to incorporate more accurate source geometries, while the modeled crack dynamics
are still based on analytical source functions (cf. Fig. 4.6b). The third type uses
accurate artificial source geometries and does not need an analytical source function
to generate acoustic emission. Instead, this type of source model is capable to
generate the crack dynamics based on experimentally accessible parameters and
fracture mechanics laws.
Currently most source models proposed in the literature are of type one or type
two, thus requiring an explicit definition of a source function. Accordingly, no
details of the dynamics arising from the crack formation process and the subsequent
crack surface motion are taken into account. To overcome these limitations, a
source model description of type three would be preferred.
From a mathematical modeling and simulation point of view, there are two main
challenges in providing a numerically based acoustic emission source model of
the third type. The first challenge consists of the different scales involved in the
problem (crack length of the order of microns versus signal wavelength of the order
of millimeters to centimeters) and the proper scale bridging. Owing to the vastly
different observations scales, a full multi-scale approach is thus necessary. The
second challenge stems from the calculation of temporal and spatial evolution of
the surfaces of the crack. This is a level of detail that is typically not accounted for
in modeling approaches used to describe crack formation by means of cohesive
zone elements, extended finite element methods, or similar implementations.
4.2 Source Mechanics 141

Throughout the publications [13, 36, 37], a new source model description was
developed and validated. Its application to fiber reinforced materials is presented in
the following. This source modeling approach consists of three sequential modeling
steps as schematically presented in Fig. 4.7. The first step is derived from classical
structural mechanics. Suitable displacement boundary conditions are defined for the
geometry considered to restrict some of the displacement components on one end.
The other end is loaded by a force high enough to initiate fracture at the crack plane
considered. If the respective force value is unknown, the implementation of a
fracture criterion (e.g., fracture toughness, max. stress, etc.) to deduce the onset
load for crack initiation is a straightforward procedure using a stationary solver
sequence with incremental loading. If the external force is known from experi-
ments, the measured force value can directly be used for the stationary solver. For
the example shown in Fig. 4.7, the presence of the notch causes stress concentration
at the tip of the notch, which will cause crack initiation at this position.
In the second step, the initial conditions for the displacement u and stress state σ
are chosen to be identical to the static values ustatic and σ static as calculated in the
previous step. Boundary conditions for restricted displacement components and
external loads are kept identical to the previous step. In contrast to the previous
step, a transient calculation of the displacement field is now performed. In addition,
boundary conditions at the crack plane are selected to allow for crack opening
according to a fracture mechanics law. The duration of this transient calculation tfrac
is chosen to be sufficient until crack propagation has come to a rest. As seen from
Fig. 4.7, the presence of the static displacement field causes crack propagation with
an accompanying excitation of an acoustic emission wave. This spatial movement
is seen best in the velocity field, since the static displacement dominates the
displacement scale and therefore inhibits to observe the very small displacements
caused by the acoustic emission wave.
For the third1 step, the initial conditions ðt ¼ 0Þ for the displacement, velocity,
and stress states of the last time value of the previous step (tfrac) are used. Boundary
conditions for restricted displacement components and external loads are kept
identical to the first and second step. The boundary conditions applied at the
crack plane are chosen to allow for independent movement of the new crack
surfaces without allowing penetration of each other. This transient calculation is
continued for a sufficient duration tend to allow for signal propagation in the test
geometry as shown in Fig. 4.7.
The present implementation of crack growth requires the definition of a fracture
plane, similar to conventional cohesive zone modeling. In the example given in
Fig. 4.7, the fracture plane is the horizontal xy-plane located at the z-position of the
notch tip. The extension of the fracture plane in x-direction is chosen sufficiently
large that the crack will not grow beyond the end of that plane.

1
The third step is only required due to the technical implementation chosen. For the selected
modeling environment, different solver settings were required for the second and the third
computation step making it necessary to split this part of the computation in two steps.
142

t= 0 t= 0
z-displacement field
= =

t= 0 t= 0 z-stress field
= =

/ t= 0 z-velocity field
=
/
z
y
x

t = 0; = 0; Č = 1 t£ ; Č = Č( , ) < t£ ;Č = 0

first step second step third step


(loading) (crack propagation) (signal propagation)

Fig. 4.7 Schematic of source model description using three subsequent modeling steps (reprinted from [13])
4 Acoustic Emission
4.2 Source Mechanics 143

e of this fracture plane is written in terms of the boundary


The stiffness vector k
coordinate system (t1, t2, n), the normal modulus En of the fracture plane, and the
in-plane shear modulus Gt12 and the in-plane Poisson’s ratio νt12 as
0 1
kt1
e ¼ @ kt2 A
k ð4:3Þ
kn
Gt12
kt1 ¼ kt2 ¼ ð4:4Þ
h

En ð1  νt12 Þ
kn ¼ ð4:5Þ
hð1 þ νt12 Þð1  2νt12 Þ

The parameter h is an effective thickness associated with the internal surface. The
thickness value h is chosen sufficiently small, so that the value of ke has negligible
influence on the overall compliance of the model.
To model crack growth, the stiffness vector is multiplied by a degradation
function Č(r) evaluated as a function of the position on the fracture surface r.

4.2.2.1 Source Models for Microscale

In [13, 36] the von Mises equivalent stress σ v was chosen to model failure in
isotropic materials. For a general stress state, σ v is written in terms of the normal
stresses σ i and shear stresses τij as
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi

σ v ¼ σ 2x þ σ 2y þ σ 2z  σ x σ y  σ x σ z  σ y σ z þ 3 τ2xy þ τ2xz þ τ2yz ð4:6Þ

e thus occurs if σ v exceeds the materials tensile


Degradation of the stiffness vector k
strength R.
For technical reasons, it is required to track the maximum value σ max of σ v
throughout the computation. Therefore, the current implementation evaluates
whether the fracture condition is fulfilled in the present time step i or if it was
fulfilled in any previous time step. Therefore the degradation function is written in
terms of the maximum value of either σ max or σ v:

1 if maxðσ max ðrÞ, σ v ðrÞÞ < R


ðrÞ ¼ ð4:7Þ
0 if maxðσ max ðrÞ, σ v ðrÞÞ  R

While this procedure was found to yield valid results in application to failure of
homogeneous and isotropic media, such as pure matrix polymer or carbon fiber
filaments [13], further modifications are required to apply this approach to AE
sources in fiber reinforced materials.
144 4 Acoustic Emission

First, the microstructural complexity of the material has to be accounted for.


Since some of the failure mechanisms discussed in Sect. 2.1 start on the scale of
several micrometers and specimens are in the length scale of several centimeters or
meters, a suitable multi-scale approach needs to be developed to bridge those
scales. Second, the time scales of fracture for fiber materials (less than nanosec-
onds) and the typical time scales for signal propagation (many microseconds) are
different by several orders of magnitude. Therefore, suitable adoption of the time
resolution is required during the progress of the simulation.
To achieve the transition between the inhomogeneous material properties on the
microscopic scale and the homogeneous material properties on the macroscopic
scale, the usage of a perfectly matched layer region has been proposed in
[12, 24]. Based on the wavelengths faced in the simulation and the fast divergence
of the waves (i.e., fast increase of wavelength) when moving away from the source,
the resolution of the microstructure can be limited to some micrometers. For a fiber
reinforced material, this comprises the typical representative volume element
(RVE) approach as seen in Fig. 4.8. In previous investigations values between
30 and 75 μm were found to be sufficient to explicitly model the microstructure and
the crack process in this RVE. With reference to (4.8) and the scheme in Fig. 4.9,
the radius of this RVE with pure material properties C0 (i.e., fiber and matrix) is
designated rpure. Adjacent to this inner zone, a perfectly matched layer (PML)
sphere with radius rpml is modeled. Here the microstructure is still geometrically
resolved, but the material properties C 0 are gradually changed using intermediate
properties as function of radial position r (with r ¼ 0 at the center of the RVE). The
material properties in the PML domains C0 for r pure  r  r pml are defined as

embedded RVE

z
y
x single fiber filament

Fig. 4.8 Geometry of RVE embedded in macroscopic geometry for wave propagation. The RVE
includes a PML domain allowing smooth transition of properties between the microstructural part
and the homogenized layers. Calculations within the embedded RVE and the macroscopic
geometry are carried out simultaneously
4.2 Source Mechanics 145

material
52 MPa stress
density component
[g/m3] x-direction
[GPa]

rPML
rpure

z z
100 μm 2400 MPa 3700 MPa
y y

Fig. 4.9 Visualization of the transition of material properties between rpure and rPML using
material density (left) and according tensile stress states for macroscopically applied tensile load
in x-direction (right)

  
0 r
C ¼ C0 þ  1  ðC1  C0 Þ ð4:8Þ
r pure

The required size of this RVE zone is governed by the fiber architecture, the fiber
volume fraction, the size of the modeled crack growth, and the wavelengths faced for
the respective failure mechanism. For the inner dimensions of the RVE stated above,
dimensions of rpml in the range between 50 and 100 μm were found to yield
convergent solutions for the cases presented in the following. On the macroscopic
scale (outside the RVE and PML region), the individual layers of the composite are
modeled as homogenized anisotropic continuum C1. A visualization of the transition
of material properties between rpure and rPML is shown in Fig. 4.9 using the material
density in the yz-cross-section of the RVE as an example. The operation of the RVE
in terms of the stress distribution is visualized in Fig. 4.9 as well. For a macroscop-
ically applied tensile stress of 2400 MPa in x-direction, the stresses within the RVE
are based on the individual stiffness of fiber and matrix and thus evaluate as
3700 MPa within the modeled fiber filaments and 52 MPa within the matrix.
Before fracture occurs, materials deform with substantial contributions of plastic
deformation. While this contribution is almost negligible for most of the reinforce-
ment fibers (i.e., carbon, glass, Kevlar), the polymeric matrix materials typically
undergo plastic deformation to a certain extent. Although the AE release (i.e.,
generation of elastic waves) is dominated by the elastic energy release, the influ-
ence of plastic deformation should still be considered in the material models within
the RVE. However, the extent of the RVE needs to be enlarged to fully include the
size of the plastic deformation zone inside rpure, since it is not practicable to rewrite
(4.8) in a trivial way to consider the transition of plastic material models between
the microscale and the macroscale.
For the change in time resolution, the wavelengths or frequencies faced in the
source modeling procedure are the relevant quantity to discuss. Since breakage of a
single filament of a carbon fiber is in the order of one nanosecond, this requires a
146 4 Acoustic Emission

sufficiently small time step (e.g., in the order of 1011 s). While this time step is still
straightforward to use during crack growth, this high time resolution is impracti-
cable for calculation of signal propagation with durations of several 100 μs. With
reference to Fig. 4.7, this requires a modification of the approach in the third step of
the modeling dealing with signal propagation. Based on publications on AE source
modeling [12, 24], one way is an iterative increase in the time step until the final
resolution is reached. For modeling of fiber breakage with a duration of crack
growth of tcrack ¼ 1 ns, this could be done in a sequence of resolutions doubling the
time step every 100 time increments starting at 0.01 ns and ending at a 10 ns
time step.

4.2.2.2 Source Models for Meso- and Macroscale

While the RVE approach is well suited to describe the propagation of cracks within
the size scale of several fiber diameters, the approach becomes numerically imprac-
tical beyond a certain length scale. This is due to the fact that the RVE itself
comprises most of the computational intensity, and thus it is not reasonable to
extend much beyond the dimensions shown in Fig. 4.8. Also, the failure on the
mesoscopic or macroscopic length scale in fiber reinforced materials is expected to
be governed more by the properties of the homogenized anisotropic solid rather
than the individual fiber and matrix materials contributing to the fracture surface.
To this end another approach was used in the following to describe the occurrence
of failure on this length scale.
As introduced in Sect. 2.2.1, Puck’s failure criteria for inter-fiber failure and
fiber failure are frequently applied to describe the occurrence of failure in fiber
reinforced materials. Considering now a homogeneous anisotropic solid in the
following, (2.11a) and (2.11b) are implemented within the “structural mechanics
module” of the software package COMSOL Multiphysics and are used to describe
layer-wise failure in unidirectional fiber reinforced materials on the mesoscopic and
macroscopic scale. For the stiffness vector k e of the fracture plane, the values of
normal modulus En, the shear modulus Gt12 , and the Poisson’s ratio νt12 are
approximated based on the orientation of the fracture plane with respect to the
orientation of the modeled layer. Although this yields only a first-order approxi-
mation of the values, a proper choice of h was found to be much more relevant to
yield negligible compliance of the closed crack surface.
For the mesoscale and macroscale, the degradation function is written in terms of
the PUCK criteria as follows:

1 if maxðPUCKmax ðrÞ, PUCKðrÞÞ < 1


ðrÞ ¼ ð4:9Þ
0 if maxðPUCKmax ðrÞ, PUCKðrÞÞ  1

Similar to (4.7) this makes use of the maximum value PUCKmax of Puck’s failure
criteria at a given position to compare the present value with the results of all
previous solutions.
4.2 Source Mechanics 147

4.2.3 Case Studies for Acoustic Emission Sources


in Fiber Reinforced Materials

One advantage of the newly proposed type three approach compared to other
formulations for acoustic emission source models is the experimental availability
of the input parameters. In the proposed model, crack growth and acoustic emission
are solely defined by the macroscopic loading condition and the failure criterion
used. In particular, no explicit source function comprising internal forces or rise-
times is necessary to initiate an acoustic emission signal. In this context, the validity
of the new source modeling approach has been verified against experimental results
from micromechanical experiments and was found to be in good agreement
[13, 37]. This validated modeling approach is used in the following to demonstrate
the excitation process of acoustic emission signals in several load cases as encoun-
tered in fiber reinforced materials.
To systematically address the different AE source mechanisms found in fiber
reinforced composites, a convenient approach is to begin with a unidirectional fiber
reinforced material and then extend the approach to more complex configurations.
For all modeling work presented in this context in the following, the material
properties of T700/PPS as reported in Table B.1 in Appendix B were used. The
elastoplastic material properties for the PPS matrix were derived from tensile tests
of the pure PPS material.
In order to avoid any influence of the thickness, or the geometry of the propa-
gation medium, a simple plate as shown in Fig. 4.10 was used as macroscopic
geometry for all investigations. The 1 mm thickness of the plate only allows
propagation of the S0 and A0 Lamb wave modes. As stacking sequence, unidirec-
tional plates with fiber axis in parallel to the x-axis and a [0/90/90/0] cross-ply
stacking were used. For evaluation of the AE signals, the out-of-plane displacement
as detected in 0 , 45 , and 90 orientation relative to the x-axis is used. The
observation positions were chosen at 25 mm distance to the source position located
at the center of the plate (see Fig. 4.16). All calculations were performed for 100 μs

signal detection 45°


1 mm

signal detection 0°

90° 0° source position

z
y
x
signal detection 90°

Fig. 4.10 Model configuration used to investigate AE signals of typical failure mechanisms in
fiber reinforced materials including detection positions of the out-of-plane displacement signals
148 4 Acoustic Emission

90° 0°
tensileper pendicular
tensile parallel σ┴
σ║

compression parallel compression perpendicular


–σ║ –σ┴

shear parallel
τ║┴ bending
σn
z
y
x

Fig. 4.11 Scheme of macroscopic load cases studied to induce AE sources in fiber reinforced
materials

duration of signal propagation. For convenience, the displacement offset due to the
applied macroscopic load is subtracted from the simulated out-of-plane displace-
ments in the following to plot AE signals starting at zero amplitude.
In the following, various failure mechanisms as typical in fiber reinforced
materials are modeled. In order to induce these different failure types, various
load conditions are required. The chosen macroscopic load situations are summa-
rized in Fig. 4.11. Based on the considerations of failure theories as discussed in
Sects. 2.1 and 2.2, there are specific failure mechanisms expected for each of the
load cases shown in Fig. 4.11. An overview of these failure mechanisms is found in
Table 4.1 including a sketch of their micromechanical representation.
Since some of the load cases induce buckling of the plate, suitable boundary
conditions were chosen for all of the cases to minimize this effect and to yield the
intended stress components on the macroscopic scale and on the microscopic scale
within the RVE. The macroscopic load levels of each case were gradually increased
until the failure condition within the RVE or at the designated fracture plane was
fulfilled. In addition, the depth position of the microscopic failure mechanisms was
varied to study its influence in analogy to Sect. 4.2.1. Details of the respective
source configurations are provided in the following subsections.
4.2 Source Mechanics 149

Table 4.1 Summary of load cases and expected failure mechanisms including micromechanical
representation
Load case Failure mechanism Micromechanical representation
Tensile parallel σ k Fiber breakage

+σ||

Fiber bridging

+σ||

Tensile perpen- Inter-fiber failure


dicular σ ⊥

+σ⊥

Compression Fiber breakage


parallel σ k

−σ||

Compression Inter-fiber failure


perpendicular
σ ⊥

−σ⊥

Shear parallel τk⊥ Inter-fiber failure

τ||⊥

(continued)
150 4 Acoustic Emission

Table 4.1 (continued)


Load case Failure mechanism Micromechanical representation
Inter-ply delamination

τ||^

Bending σ n Inter-fiber failure in 90 plies

Inter-ply delamination at inter-


face between 90 and 0 ply

4.2.3.1 Inter-fiber Failure

As one of the fundamental failure mechanisms in fiber reinforced composites, inter-


fiber failure is understood as formation of cracks in the matrix region between the
fibers and along the interface between fiber and matrix (cf. Sect. 2.1). For a
unidirectional fiber reinforced material, a broad distribution of crack sizes is
expected [38]. As shown in Fig. 2.1, such inter-fiber failure may occur on the
length scale of few fiber diameters, but may also span the full thickness of a ply and
may propagate several centimeters in one step. Therefore several AE source models
need to be investigated in terms of the dimensions of the inter-fiber cracks. In the
following, the AE source models were implemented as described in the previous
Sect. 4.2.2 employing a thin elastic layer in combination with Puck’s inter-fiber
failure criterion. Before starting the modeling work, an analysis using Puck’s failure
criterion is made to which of the above load situations are able to induce inter-fiber
failure.
4.2 Source Mechanics 151

 
As the first load case, the tensile parallel σ k situation is studied. As discussed
by Puck [39], inter-fiber failure in such a load situation is usually not covered by
most of the failure theories. Still it is evident from experimental observations that
such load situations may sometimes cause inter-fiber failure in the form of splitting
of the fiber layers. This is likely due to the generation of shear stresses at imper-
fections of the material or other breakdowns of the perfect tensile parallel load case.
However, for the AE source modeling applying Puck’s failure criterion, it was
found impossible to generate inter-fiber failure at any orientation of the fracture
plane before the occurrence of fiber failure. Thus it is not feasible to calculate AE
sources due
 to inter-fiber failure following Puck’s failure criterion in the tensile
parallel σ k load case and similar in the compressive parallel σ k load case.
Consequently, only results of the remaining four load cases will be discussed in the
following.

Tensile Perpendicular σ ⊥

As the first example, tensile loading perpendicular to the fiber axis ðσ ⊥ Þ is presented
and used as an example to demonstrate some key aspects of acoustic emission
source operation. For a unidirectional composite material, this load case typically
results in ultimate failure due to macroscopic inter-fiber failure. However, the
ultimate (macroscopic) separation of the material is not the focus of this study.
Herein inter-fiber failure is assumed to stop after a certain propagation distance to
generate an AE signal for comparison with other failure mechanisms. Hence, the
crack growth was confined to a distinct size. The geometry of the crack and the
according load configuration is shown in Fig. 4.12. To initiate the crack at the center
position of the plate, an artificial flaw of 10 μm length spanning the full thickness
direction was included to act as a precrack.
In the stationary model, the static load F⊥ was increased to 1000 N before Puck’s
inter-fiber failure criterion indicated crack growth at the position of the edge of the

tensile 1 mm
perpendicular
precrack
σ┴
(10 μm)
fracture surface
z
y (x,y,z) = (0,0,0)
x
symmetry plane
symmetry plane

Fig. 4.12 Model configuration including position of symmetry plane and details of crack region
152 4 Acoustic Emission

t = 50 ns t = 250 ns

z
y
x

t = 500 ns

Fig. 4.13 Extension of fracture surface during crack growth at t ¼ 50 ns, t ¼ 250 ns, and t ¼ 500 ns
after initiation of crack propagation

precrack yielding σ ⊥ ¼ 20 MPa. Due to the presence of the static load, a static
displacement field exists within the specimen. As pointed out in the previous
section, these static load and displacement field are transferred to a second com-
putation step. This initiates the transient degradation of the stiffness vector of the
thin elastic layer based on the evaluation of Puck’s failure criterion. A sequence of
images during the duration of crack propagation is shown in Fig. 4.13. Here the
value of the degradation function is plotted for t ¼ 50 ns, t ¼ 250 ns, and t ¼ 500 ns
after initiation of crack propagation. The false-color range of the degradation
function Č(r) indicates the extension of the crack by red areas, while blue colors
resemble areas, where the material is still in contact.
Due to the orientation of the local stress components, the crack walls are subject
to a dominant mode I type load situation. This causes a crack opening during
propagation of the crack, and the resulting movement of the fracture surfaces
constitutes the actual source mechanism. Since the static displacement values
dominate the displacement scale, it is hard to spot the minimal displacement values
superimposed to this background. Thus it is more convenient to use diagrams of the
acoustic velocity magnitude to visualize the generation of the acoustic emission
wave around the crack. As seen from Fig. 4.14, the near field of this crack type
consists of a radiation perpendicular to the crack surfaces, which extends along the
x-axis with the progression of the crack.
4.2 Source Mechanics 153

t = 100 ns t = 250 ns

z
y
x

acoustic
t = 500 ns velocity
magnitude
[m/s]

Fig. 4.14 Plot of acoustic velocity magnitude during crack growth at t ¼ 100 ns, t ¼ 250 ns, and
t ¼ 500 ns after initiation of crack propagation

In order to obtain the source function of the acoustic emission source, the
y-displacement at the position (x, y, z) ¼ (0, 0, 0.5) mm is evaluated. This position
is at the center of the growing crack and thus the position with the maximum crack
opening. As seen in Fig. 4.15, during the duration of crack propagation, the
y-displacement increases continuously. However, at the moment of crack arrest,
the y-displacement does not stop or cease, but increases further until a maximum
value is reached. Subsequently, the crack surface starts to vibrate and finally settles
at a new equilibrium position.
Based on the theory of Green and Zerna [40], the expected position of the crack
surface in a static crack opening situation of a crack length a may be estimated for
an isotropic material with modulus E and Poisson’s ratio ν for a given stress σ as

4  ð1  ν Þ
utheory ¼ σa ð4:10Þ
πE

Following the intent of (4.10) for a mode I type load, the calculated value using the
material stiffness E22, ν21, σ ⊥ ¼ 20 MPa, and a ¼ 2 mm is marked as dashed line in
Fig. 4.15b. As can be seen from the comparison, in the initial part, these values are
systematically lower than the present model predictions. This originates from the
underlying assumptions in the theory. In [40] static crack opening is assumed, i.e.,
the length of the crack is already present as flaw in the material and merely opens
due to an external force. If the same assumptions are made in the present model,
154 4 Acoustic Emission

a
4.0
duration of crack growth
3.5 crack surface motion

3.0
displacement [μm]

2.5

2.0

1.5

1.0

0.5 crack arrest

0.0
0 1 2 3 4 5 6 7 8 9 10
time [μs]

b
4.0
duration of crack growth
3.5 crack surface motion

3.0
displacement [μm]

ytheory = 2.9 μm
2.5

2.0
signature of reflected A0-mode
1.5

1.0

0.5

0.0
0 20 40 60 80 100
time [μs]

Fig. 4.15 Evaluation of source displacement in y-direction during crack growth (a) and for full
duration of computation (b)

the achieved deformation state is in very good agreement to the values predicted
by [40]. However, the dynamic crack propagation seems to generate initial acoustic
emission amplitudes being larger than predicted by the analytical theory.
At the moment of crack arrest at t ¼ 1.1 μs, the model starts with the third
computation step to calculate the spreading of the acoustic emission signal in the
plate. In the far field, the displacement and velocity components generated by
the crack source are converted into plate waves and start to propagate toward the
detection positions. As seen from the acoustic velocity fields in Fig. 4.16 at six
in-plane
t = 1 μs t = 2 μs t = 3μs velocity
[m/s]
4.2 Source Mechanics

radiation radiation
perpendicular parallel
z
y
x

t = 4 μs t = 5 μs t = 7.5 μs

edge
reflection

Fig. 4.16 Propagation of acoustic waves in the far field around the crack at distinct time steps after initiation of crack propagation
155
156 4 Acoustic Emission

distinct time steps, there are two major contributions to the wave field. The first part
is the radiation of the wave in orientation perpendicular to the fracture surface. This
constitutes the dominant contribution to the wave field. As the second part, the
propagation of the crack causes a wave in parallel to the direction of crack growth.
This less-intense contribution spreads out faster due to the acoustic anisotropy of
the unidirectional fiber reinforced plate.
Due to the different radiation of the source in each direction, the resulting signals
obtained at 0 , 45 , and 90 start to appear significantly different. A comparison of
the signals detected at 25 mm distance to the source position is shown in Fig. 4.17.
Based on the angular dependency of the guided wave modes, the initial arrival of
the S0 mode is observed first in 0 orientation and last for 90 orientation. The mode
intensities are characteristically different for each orientation, which is a conse-
quence of the radiation pattern discussed above. The appearance of the modes
themselves is governed by the shape of the dispersion curves as shown in Fig. 4.59
in Sect. 4.3.

Compression Perpendicular σ ⊥

As the next type of load configuration, a compressive load perpendicular to the fiber
axis σ ⊥ was studied. As discussed in Sect. 2.1, such load situations are expected to
cause inter-fiber failure with a fracture surface inclined by θ ¼ 53 as shown in
Fig. 4.18. Here crack growth was initiated by a precrack at the center of the crack
with a width of 10 μm. In the stationary step of the modeling procedure, the static
load F⊥ was increased to 9000 N before Puck’s inter-fiber failure criterion
indicated crack growth at the position of the edge of the precrack yielding
σ ⊥ ¼ 90 MPa.
Due to the presence of the compression load, a static displacement field exists
within the specimen causing crack propagation at the designated fracture plane.
A sequence of images during the duration of crack propagation is shown in
Fig. 4.19. Here the degradation function is shown for t ¼ 50 ns, t ¼ 250 ns, and
t ¼ 500 ns after initiation of crack propagation. As seen from Fig. 4.19, the crack
starts to grow beginning at the precrack and propagates along the x-axis direction.
The false-color range of the degradation function Č(r) indicates the extension of the
crack by red areas, while blue colors resemble areas, where the material is still in
contact.
Similar to the previous load case, the near field of the crack is shown as velocity
magnitude at three distinct time steps in Fig. 4.20. The observed crack propagation
causes a motion perpendicular to the formation of the crack surface which is
inclined by 53 . In comparison to the load case σ ⊥ , the intensity of the surface
motion is not balanced along the y-axis at each edge of the crack surface. Instead,
there is an asymmetric intensity observed radiating in the positive y-axis direction
at the top surface of the plate and in the negative y-axis direction on the bottom of
the plate (cf. Fig. 4.20).
4.2 Source Mechanics 157

a 90°
0.20 45°
0.18 0°

0.16
0.14
offset displacement [μm]

0.12
0.10
0.08
0.06
0.04
0.02
0.00
–0.02

0 20 40 60 80 100
time [μs]
b 90°
2.5x10–5 45°

2.0x10–5
FFT-magnitude [1/m]

1.5x10–5

1.0x10–5

5.0x10–6

0.0
0 100 200 300 400 500 600 700 800 900 1000
frequency [kHz]

Fig. 4.17 Acoustic emission signals of inter-fiber failure due to σ ⊥ detected in unidirectional plate
at 0 , 45 , and 90 relative to fiber orientation at 25 mm distance to source position in time domain
(a) and frequency domain (b)

The acoustic emission signals detected in 25 mm distance to the source position


are shown in Fig. 4.21 for three different angles relative to the source position.
Again, the combination of the radiation pattern of the source and the anisotropy of
158 4 Acoustic Emission

compression
perpendicular fracture surface
precrack
–σ┴ (10 μm)
θ
z
y (x,y,z) = (0,0,0)
x
symmetry plane 1mm
symmetry plane

Fig. 4.18 Model configuration including position of symmetry plane and details of crack region

t = 50 ns t = 250 ns

z
y
x

t = 500 ns

Fig. 4.19 Extension of fracture surface during crack growth at t ¼ 50 ns, t ¼ 250 ns, and t ¼ 500 ns
after initiation of crack propagation

the propagation medium causes signals, which look distinctly different although
they emanate from the same source. This finding is not unexpected and was also
experimentally verified (cf. Sect. 4.5.4.3). In direct comparison to the σ ⊥ load case,
the signals detected at each of the angles show reasonable similarity.

Shear Parallel τk⊥

As further load case relevant for inter-fiber failure, shear loading parallel τk⊥ to the
fiber axis is investigated. This shear stress situation indicated by the load arrows in
Fig. 4.22 is expected to yield a fracture surface orientation as for the σ ⊥ case.
4.2 Source Mechanics 159

t = 100 ns t = 250 ns

z
y
x

acoustic
t = 500 ns velocity
magnitude
[m/s]

1 mm

Fig. 4.20 Plot of acoustic velocity magnitude during crack growth at t ¼ 100 ns, t ¼ 250 ns, and
t ¼ 500 ns after initiation of crack propagation

As shown in the schematically drawing in Fig. 4.22, a precrack of 10 μm length was


chosen to facilitate crack initiation. To allow for better comparison with the other
inter-fiber failure cases, the extension of the fracture surface was chosen to be 1 mm
along the x-axis. A macroscopic load of Fk⊥ ¼ 8000 N results in a shear stress value
of τk⊥ ¼ 80 MPa at the position of the fracture surface.
A visualization of crack propagation according to Puck’s failure criterion is
shown in three distinct time steps in Fig. 4.23. Similar to the previous load cases,
the crack front advances steadily along the direction of the x-axis due to the
presence of the macroscopic stress field. While the speed of crack propagation
along the x-axis is almost unaltered compared to the previous cases, slight differ-
ences arise in the shape of the crack front. This is attributed to the different load
components faced in this shear load case.
The acoustic velocity magnitude is shown as false-color plot in Fig. 4.24 at three
distinct time steps after crack initiation. While the first impression of the crack front
resembles some close similarity to the near field emanating from the crack growth under
σ ⊥ , there are some substantial differences. In the τk⊥ load case, the extent of the motion
along the y-axis does reach as far as for the σ ⊥ case, and some noticeable additional
motions with higher frequencies are observed. The latter are seen best as shades ahead
of the intense part of the near-field zone spreading out in x- and y-direction.
160 4 Acoustic Emission

a 90°
1.1 45°
1.0 0°
0.9
0.8
offset displacement [μm]

0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.0
–0.1
–0.2
0 20 40 60 80 100
time [μs]

b 90°
45°
2.8x10–4 0°

2.4x10–4
FFT-magnitude [1/m]

2.0x10–4

1.6x10–4

1.2x10–4

8.0x10–5

4.0x10–5

0.0
0 100 200 300 400 500 600 700 800 900 1000
frequency [kHz]

Fig. 4.21 Acoustic emission signals of inter-fiber failure due to σ ⊥ detected in unidirectional
plate at 0 , 45 , and 90 relative to fiber orientation at 25 mm distance to source position in time
domain (a) and frequency domain (b)

The change in source operation as compared to the σ ⊥ load case also translates
into distinct differences in the detected acoustic emission signals. In Fig. 4.25 a
comparison of the signals detected at three different angles relative to the source
position is shown. In contrast to the signals of Figs. 4.17 and 4.21, these acoustic
4.2 Source Mechanics 161

shear parallel 1mm


τ║┴ precrack
(10 μm)
fracture surface
z
y (x,y,z) = (0,0,0)
x
antisymmetry plane antisymmetry plane

Fig. 4.22 Model configuration including position of antisymmetry plane and details of crack
region

t = 50 ns t = 250 ns

z
y
x

t = 500 ns

Fig. 4.23 Extension of fracture surface during crack growth at t ¼ 50 ns, t ¼ 250 ns, and t ¼ 500 ns
after initiation of crack propagation

emission signals share the common aspect of lack of lower frequency components.
This is due to the confined possibilities of crack surface movement. Since the
macroscopic τk⊥ load does not cause a mode I type crack opening movement, but
rather a mode II type sliding crack propagation, the dominant vibration of the crack
surface is a movement in the xz-plane.

Bending σ n

As the last type of load scenario for the occurrence of inter-fiber failure, a bending
load σ n is investigated. Here, a different laminate type is selected to induce
162 4 Acoustic Emission

t = 100 ns t = 250 ns

z
y
x

acoustic
t = 500 ns velocity
magnitude
[m/s]

Fig. 4.24 Plot of acoustic velocity magnitude during crack growth at t ¼ 100 ns, t ¼ 250 ns, and
t ¼ 500 ns after initiation of crack propagation

inter-fiber failure in a [0/90]sym stacking sequence. The thickness was chosen to be


1 mm in accordance with the previously studied unidirectional plate. A schemati-
cally view of the load configuration with details of the crack model is shown in
Fig. 4.26. The inter-fiber failure fracture plane is defined directly below the position
of applied load and to occur within the full thickness of the 90 layer. For better
comparability with the previous cases, the size of the crack along the y-axis
direction is artificially constrained to yield a length of 500 μm.
To initiate crack growth in the designated fracture plane, a precrack of 10 μm
length in the z-axis direction was modeled at the top edge of the 90 layer. The
required external load was evaluated to be Fn ¼ 500 N so that Puck’s failure
criterion indicated crack initiation at the position of the precrack. The crack
propagation is visualized for three distinct time steps in Fig. 4.27 using the false-
color representation of the degradation function. In this load case, the crack initiates
at the precrack position and starts to propagate with an almost plane crack front
along the z-axis direction until it reaches the position of the bottom 0 layer.
The near field of the growing crack is shown in Fig. 4.28 for three distinct time
steps as plot of the acoustic velocity magnitude. Since the crack propagation
direction is along the z-axis direction, the view of the acoustic velocity magnitude
within the xz-plane needs some further explanation. Because the crack starts at the
top of the 90 layers, this is also the start position of the crack surface motion. With
increasing time the motion of the crack walls turn on further displacements along
4.2 Source Mechanics 163

a 90°
0.060 45°
0.055 0°
0.050
offset displacement [μm]

0.045
0.040
0.035
0.030
0.025
0.020
0.015
0.010
0.005
0.000
–0.005
–0.010
–0.015
0 20 40 60 80 100
time [μs]
b 90°
45°
1.0x10–5 0°
FFT-magnitude [1/m]

8.0x10–6

6.0x10–6

4.0x10–6

2.0x10–6

0.0
0 100 200 300 400 500 600 700 800 900 1000
frequency [kHz]
Fig. 4.25 Acoustic emission signals of inter-fiber failure due to τk⊥ detected in unidirectional
plate at 0 , 45 , and 90 relative to fiber orientation at 25 mm distance to source position in time
domain (a) and frequency domain (b)

the crack propagation direction causing a stripe-like appearance of the near-field


zone. After the crack has reached the bottom of the 90 layers, the motion solely
spreads out from this position. The effect of the different fiber orientations in the
0 and 90 layers also adds to the formation of the near field. Since wave propaga-
tion along the x-axis is faster in the top and bottom 0 layer, the motion in this part
164 4 Acoustic Emission

z
y precrack
x (10 μm)
1mm
fracture
bending surface
σn 0°
90° symmetry
0° (x,y,z) = (0,0,0) plane
symmetry plane

Fig. 4.26 Model configuration including position of symmetry plane and details of crack region

t = 10 ns t = 50 ns
precrack

z
y
x

t = 100 ns


90°

Fig. 4.27 Extension of fracture surface during crack growth at t ¼ 10 ns, t ¼ 50 ns, and t ¼ 100 ns
after initiation of crack propagation

of the plate propagates faster than in the 90 layer before all motions finally turn
into a common guided wave mode.
As expected from the different stacking sequence, but also due to the different
source type, the acoustic emission signals appear to be different to the previous
cases. As shown for three distinct angles, the acoustic emission signals detected in
25 mm distance to the source position show a pronounced high-frequency vibration
superimposed by an out-of-plane movement with comparatively low frequency.
The high-frequency part originates from similar confined crack motions as
discussed for the τk⊥ case. In the bending case, the upper half of the fracture
4.2 Source Mechanics 165

t = 100 ns t = 250 ns

z
y
x

acoustic
t = 500 ns velocity
magnitude
[m/s]
1 mm

Fig. 4.28 Plot of acoustic velocity magnitude during crack growth at t ¼ 100 ns, t ¼ 250 ns, and
t ¼ 500 ns after initiation of crack propagation

plane is subject to a dominant compressive normal load, while the lower part is
subject to a tensile normal load. In combination with the maximum shear load at the
neutral axis of the plate, this leads to superimposed constraints with respect to
the movement of the fracture surface. These constraints inhibit a free vibration of
the full fracture surface (as, i.e., observed in [13]) and cause the vibration seen in
Fig. 4.29. The low-frequency contribution however has a distinctly different origin.
Due to the formation of the crack, the plate loses its stiffness, which results in a
macroscopic settlement at the center of the plate. This settlement causes a plate
motion, which is directed along the 0 and 90 axis orientation and hence visible in
the contributions of the respective signals.

4.2.3.2 Delamination

As the second category of failure mechanisms studied, the focus is now on the
occurrence of inter-ply delamination. Many failure theories do not distinguish
between inter-ply delamination and inter-fiber failure, since crack growth is occur-
ring between the fibers in both cases. However, due to the different orientation of
the fracture surface relative to the thickness direction of the plate, this failure type is
acoustically different to the cases studied in Sect. 4.2.3.1.
166 4 Acoustic Emission

a 90°
0.026 45°
0.024 0°
0.022
0.020
0.018
offset displacement [μm]

0.016
0.014
0.012
0.010
0.008
0.006
0.004
0.002
0.000
–0.002
–0.004
0 20 40 60 80 100
time [μs]
b 90°
6.0x10–6 45°

FFT-magnitude [1/m]

4.0x10–6

2.0x10–6

0.0
0 100 200 300 400 500 600 700 800 900 1000
frequency [kHz]

Fig. 4.29 Acoustic emission signals of inter-fiber failure due to σ n detected in [0/90]sym plate at
0 , 45 , and 90 relative to fiber orientation at 25 mm distance to source position in time domain
(a) and frequency domain (b)

Using Puck’s failure criterion to evaluate the occurrence of delamination (i.e.,


inter-fiber failure with θ ¼ 90 ), the same considerations as in Sect. 4.2.3.1 apply.
For the load cases σ k and σ k , there is no inter-fiber failure observed before the
occurrence of fiber failure. Similarly, the fracture plane orientation for the load case
4.2 Source Mechanics 167

σ ⊥ in a unidirectional plate is 0 , and for the load case σ ⊥ , the fracture plane
orientation is calculated as θ ¼ 53 . Consequently, none of those load cases is
discussed in the following. Only the τk⊥ load case was found to yield a delamination
type failure in a unidirectional material. In addition, the important scenario of inter-
ply delamination in a multiaxial laminate is investigated. To this end, an analysis
using a [0/90]sym laminate subject to bending load σ n is presented.

Shear Parallel τk⊥

As shown in Fig. 4.30, the fracture surface is oriented parallel to the top and bottom
surface ðθ ¼ 90 Þ of the laminate plate. The width of inter-ply delamination was
chosen as 1 mm, and crack growth was confined to a maximum length of 1 mm
allowing better comparison to the inter-fiber failure cases studied previously. To
allow crack initiation at the designated position, a precrack of 10 μm width was
modeled at the edge to the yz-plane.
At an applied force of Fk⊥ ¼ 6300 N, Puck’s failure criterion indicates crack
growth at the precrack position. A visual representation of crack growth found for
this load configuration is shown in Fig. 4.31. At three distinct time steps, the value
of the degradation function is plotted as false-color values. Unlike the inter-fiber
failure cases, the crack propagation in the present case seems to be substantially
slower.
For this load case, the direction of crack propagation is along the x-axis. The type
of crack motion seen in the acoustic velocity magnitude images in Fig. 4.32 is
mostly due to the perpendicular motion of the crack surface. As indicated from the
multiple wave peaks in Fig. 4.32, the crack motion is subject to a sequence of crack
arrest and crack initiations. This turns into a high-frequency contribution as seen in
the acoustic emission signals in Fig. 4.33 and a characteristic resonance like
frequency. However, due to Poisson’s contraction, the crack wall motion along
the z-axis is energetically limited and therefore the absolute amplitudes of the
signals are one to two orders of magnitude lower than for the inter-fiber failure
cases of similar crack surface dimensions.

shearparallel 1mm
τ║┴
precrack
(10 μm)
z fracture
y (x,y,z) = (0,0,0) surface
x
antisymmetry plane
antisymmetry plane

Fig. 4.30 Model configuration including position of antisymmetry plane and details of crack
region
168 4 Acoustic Emission

t = 50 ns t = 2000 ns

z
y
x

t = 5000 ns

Fig. 4.31 Extension of fracture surface during crack growth at t ¼ 50 ns, t ¼ 2000 ns, and
t ¼ 5000 ns after initiation of crack propagation

t = 100 ns t = 500 ns

z
y
x

t = 1000 ns acoustic
velocity
magnitude
[m/s]

1 mm

Fig. 4.32 Plot of acoustic velocity magnitude during crack growth at t ¼ 100 ns, t ¼ 500 ns, and
t ¼ 1000 ns after initiation of crack propagation
4.2 Source Mechanics 169

a 90°
2.8 45°
2.6 0°
2.4
2.2
2.0
offset displacement [nm]

1.8
1.6
1.4
1.2
1.0
0.8
0.6
0.4
0.2
0.0
–0.2
–0.4
–0.6
–0.8
0 20 40 60 80 100
time [μs]

b 90°
1.2x10–6 45°

1.0x10–6
FFT-magnitude [1/m]

8.0x10–7

6.0x10–7

4.0x10–7

2.0x10–7

0.0
0 100 200 300 400 500 600 700 800 900 1000
frequency [kHz]

Fig. 4.33 Acoustic emission signals of inter-ply delamination due to τk⊥ detected in unidirectional
plate at 0 , 45 , and 90 relative to fiber orientation at 25 mm distance to source position in time (a)
and frequency domain (b)

Bending σ n

Another configuration encountered in fiber reinforced materials, which causes


initiation of inter-ply delamination, is the presence of an inter-fiber crack in
off-axis plies relative to the load axis. As representative example for this scenario,
170 4 Acoustic Emission

precrack

90°
1mm
bending

σn

symmetry
plane
symmetry plane
fracture
z surfaces (x,y,z) = (0,0,0)
y
x

Fig. 4.34 Model configuration including position of symmetry plane and details of crack region

the load case bending σ n in conjunction with a [0/90]sym laminate is investigated.


The schematically arrangement of this source type is shown in Fig. 4.34.
As precrack a fully grown inter-fiber crack spanning the thickness of the two 90
plies is chosen. The extent of this precrack is identical to the previously studied case
of inter-fiber failure due to bending load σ n. In such a case, inter-ply delamination is
expected to occur at the interface between 0 and 90 plies at the top and at the
bottom, respectively. The size of each of the delamination areas was chosen to be
1.0 mm  1.0 mm.
At an applied bending load Fn ¼ 750 N, Puck’s failure criterion indicated
initiation of crack growth at both of the interfaces. As seen from the evaluation of
the degradation function as false-color plot at three distinct time steps in Fig. 4.35,
the crack front advances along the x-axis starting at the position of the precrack.
Due to the distinctly different load conditions at the top and at the bottom interface,
the crack front at the top interface spreads out first and advances faster than at the
bottom. However, both cracks are active at the same time, thus yielding a combined
contribution to the near-field zone and thus effectively yielding one acoustic
emission signal.
The combined activity of both delamination fronts is also seen in the acoustic
velocity magnitude shown in Fig. 4.36 at three distinct time steps. Here the zone at
the top interface appears larger, which is due to the delamination at the top interface
preceding the delamination at the bottom interface. The near field is shaped by the
direction of crack propagation along the x-axis but is also facilitated by the acoustic
anisotropy of the 0 layers causing faster wave propagation along the x-axis than in
the perpendicular directions. Therefore the near-field zone appears rather elliptical
at the top and at the bottom position.
Due to the close physical distance and the close time-wise distance, the activity
of the two delamination fronts causes only one acoustic emission signal. This is
demonstrated by the calculated signals in 25 mm distance to the source positions
for three different angles as shown in Fig. 4.37. The acoustic emission signals do
not reveal any indications of two merged signals but rather appear as typical
transient signal with one distinct onset and according activity after that. The
4.2 Source Mechanics 171

t = 100 ns t = 500 ns

z
y
x

t = 1000 ns

Fig. 4.35 Extension of fracture surface during crack growth at t ¼ 100 ns, t ¼ 500 ns, and
t ¼ 1000 ns after initiation of crack propagation

t = 500 ns t = 1000 ns

z
y
x

acoustic
t = 2500 ns velocity
magnitude
[m/s]

1 mm

Fig. 4.36 Plot of acoustic velocity magnitude during crack growth at t ¼ 500 ns, t ¼ 1000 ns, and
t ¼ 2500 ns after initiation of crack propagation
172 4 Acoustic Emission

a 90°
0.18 45°
0.16 0°

0.14
offset displacement [μm]

0.12
0.10
0.08
0.06
0.04
0.02
0.00
–0.02
–0.04
0 20 40 60 80 100
time [μs]
b 90°
8.0x10–5 45°

6.0x10–5
FFT-magnitude [1/m]

4.0x10–5

2.0x10–5

0.0
0 100 200 300 400 500 600 700 800 900 1000
frequency [kHz]

Fig. 4.37 Acoustic emission signals of inter-ply delamination due to σ n detected in [0/90]sym plate
at 0 , 45 , and 90 relative to fiber orientation at 25 mm distance to source position in time domain
(a) and frequency domain (b)
4.2 Source Mechanics 173

dominant contributions seen in the signals can be attributed to similar origins as


before. The high-frequency portion is mostly caused by the elementary crack
arrest and crack advance of the fracture surface, whereas the low-frequency
portion is again due to a settlement of the damage zone induced by the change
in stiffness reduction.

4.2.3.3 Fiber Failure

Distinctly different to the two previous categories of failure mechanisms, fiber


failure is a type of crack propagation associated with several challenges in source
modeling. As evident from experimental findings, fiber failure is a mechanism
which is usually not found on the macroscopic scale (e.g., crack zones with mm2)
prior to ultimate failure. Instead composite failure was observed to be caused by
accumulation of clusters of less than 20 single fiber filament failures [41]. With
some exceptions, rupture of a larger number of loaded fibers will initiate ultimate
failure, and the according acoustic emission signal will be useless in the current
context, since it will overlap with many other mechanisms and therefore cannot be
related to a particular microscopic mechanism. Consequently, the following model-
ing work aims for the failure of single filaments and small fiber bundles in relation
to their acoustic emission signals. Consequently, the RVE approach was used to
model a microscopic part of the unidirectional fiber reinforced plate embedded in
the macroscopic plate. Distinction is made between two load cases typically
causing failure of fiber filaments in unidirectional laminates. One is the tensile
parallel load case σ k and one is the compression parallel load case σ k .

Tension Parallel σ k

For initiation of fiber breakage, tensile loading parallel σ k to the fiber axis is the only
relevant situation for a unidirectional laminate. This model configuration is shown
in Fig. 4.38 illustrating the macroscopic load configuration and then presenting
details of the RVE dimensions and position within the plate. For the crack position,
a further magnification of the RVE provides some more details with respect to the
chosen model configuration. In the present RVE configuration, all fibers have
identical dimensions and mechanical properties.2 To initiate failure in one single
filament, two modifications to the perfect RVE arrangement were required. First,
within the designated yz-plane, the matrix area surrounding the fiber filament was
defined as fully cracked, resembling presence of a void or an inter-fiber crack acting
as stress concentrator at the surface of the fiber. Second, to avoid failure of other

2
This assumption is not absolutely justified in reality, since it is well known that fibers show
Weibull-type strength distributions, and may have significant variation in their cross-sections and
their modulus.
174 4 Acoustic Emission

100 μm
tensile parallel
σ║
symmetry
plane

(x,y,z) = (0,0,0)
symmetry plane

z
y
x
7 μm
fracture surface

pore diameter
350 nm

precrack

Fig. 4.38 Model configuration including position of symmetry plane, details of RVE position,
dimensions and details of crack region

surrounding fibers (subject to the same stress concentration effect), the designated
fiber filament was additionally weakened by the presence of a small pore with
350 nm diameter. This is one tenth of the fiber radius and therefore causes enough
stress concentration at one particular fiber filament to enable crack propagation only
within its cross-section. That way it is possible to systematically vary the filament
failure strength between 300 and 4061 MPa to investigate its relation to the
detectable acoustic emission signal amplitude. The value of 4061 MPa was taken
from the material supplier’s datasheet and is interpreted as mean fiber strength.
In order to reach a stress concentration of 4061 MPa at the designated fracture
plane, a macroscopic load of Fk ¼ 263 kN was necessary. As failure criterion the
von Mises equivalent stress as defined in (4.6) was used. Due to the presence of the
modeled pore, failure initiates at the bottom of the fiber as seen in the false-color
plots of the degradation function in Fig. 4.39. The crack propagation takes about
1.3 ns until the fracture surface is finally established. This duration seems quite
reasonable when compared to the value of 1.2 ns calculated for a free fiber filament
in [13] which has been validated against experimental data. As distinct difference to
the case of a free fiber filament, the crack front seems to initiate at the position of the
pore, but later starts to advance along the radial direction of the fiber starting at all
directions reducing the cross-section of the fiber upon crack growth.
The near field of the single fiber filament failure is visualized as acoustic velocity
magnitude in Fig. 4.40 for three distinct time steps after fiber failure. The shape of
the near-field zone resembles close similarity to a dipole radiation pattern.
Fig. 4.39 Extension of fracture surface during crack growth at t ¼ 0.1 ns, t ¼ 0.5 ns, and t ¼ 1.0 ns
after initiation of crack propagation

t = 2.5 ns t = 5.0 ns

100 μm

z
y
x

acoustic
t = 10.0 ns
velocity
magnitude
[m/s]

Fig. 4.40 Plot of acoustic velocity magnitude during crack growth at t ¼ 2.5 ns, t ¼ 5.0 ns, and
t ¼ 10.0 ns after initiation of crack propagation
176 4 Acoustic Emission

This pattern is slightly distorted over time, because of the high acoustic velocity
along the fiber axis causing a faster spreading along the fiber axis than in the
surrounding matrix material. The dominant radiation direction is found to be in
the in-plane direction which is in good accordance with results from previous
source models [24, 35].
Despite of the high macroscopic failure load and the intense acoustic velocity
magnitudes seen in Fig. 4.40, the amplitudes of the detected acoustic emission
signals are substantially lower than for inter-fiber failure or delamination
(cf. Fig. 4.41). For the three acoustic emission signals detected at 25 mm to the
source position at three different angles, the values are found to be three to four
orders of magnitude lower than for the previous cases. This is caused by the
substantially lower fracture surface area in the present case. Compared to the
previous fracture surface area of the order of 1  106 m2 in the present case, the
fiber cross-section is of the order of only 1  1010 m2 having a difference of about
four orders of magnitude. The main cause for the difference in intensity for the
0 orientation and the 45 and 90 orientation is the unidirectional arrangement of
the fiber and the according radiation pattern having a dominant radiation in parallel
to the fiber axis. Independent of the angle, the detected acoustic emission signals are
dominated by high-frequency contributions, which are caused by the short rise-time
of the source.

Compression Parallel σ k

As further load case for single filament failure, the compression load parallel σ k to
the fiber axis is investigated. As seen from the micromechanical representation in
Fig. 4.42, this type of failure usually causes fragmentation of the fiber filaments due
to the occurrence of band kinking prior to failure. Therefore the resulting AE signal
is caused by failure of more than one material and a mixed mode fracture of the
individual microscopic components. Being a representative case of this class of
failure, the kinking of one single fiber filament was investigated.
As seen from Fig. 4.42, the RVE was embedded in the laminate at the center
position of the plate. To facilitate the visibility of the relevant modifications, the
fibers of the RVE surrounding the failing filament are not shown in Fig. 4.42. The
single fiber filament subject to failure was modified to include two pores at opposite
edges along the z-axis. Due to this pore configuration, a skew microscopic load
situation occurs, which causes kinking of the designated fiber filament.
Using a macroscopic stress level of 800 MPa at the designated fracture planes, a
microscopic load of σ k ¼ 1800 MPa was observed. As failure criterion in this
case, the von Mises equivalent stress as defined in (4.6) was used for this case.
Similar to the previous case, the presence of the modeled pores initiates crack
growth at these positions as seen from the false-color plots of the degradation
function in Fig. 4.43. The crack propagation takes about 1.3 ns until the full extent
of the fracture surface is reached, which is the identical duration as observed for the
4.2 Source Mechanics 177

a 90°
0.40 45°
0.35 0°

0.30
0.25
offset displacement [fm]

0.20
0.15
0.10
0.05
0.00
–0.05
–0.10
–0.15
0 20 40 60 80 100
time [μs]
b 90°
1.0x10–10 45°

8.0x10–11
FFT-magnitude [1/m]

6.0x10–11

4.0x10–11

2.0x10–11

0.0
0 100 200 300 400 500 600 700 800 900 1000
frequency [kHz]

Fig. 4.41 Acoustic emission signals of fiber failure due to σ k detected in unidirectional plate at 0 ,
45 , and 90 relative to fiber orientation at 25 mm distance to source position in time domain (a)
and frequency domain (b)

tensile fiber failure case. This seems well applicable since the crack speed is
dominated by the carbon fiber properties in both cases.
The signal propagation after filament failure is visualized as acoustic velocity
magnitude in Fig. 4.44 for three distinct time steps after failure initiation. Different
100 μm
compression parallel
–σ║
symmetry
plane

(x,y,z) = (0,0,0)
symmetry plane

z
y
x
fracture surface
7 μm

pore diameter
350 nm

Fig. 4.42 Model configuration including position of symmetry plane, details of RVE position,
dimensions and details of crack region

Fig. 4.43 Extension of fracture surface during crack growth at t ¼ 0.1 ns, t ¼ 0.5 ns, and t ¼ 1.0 ns
after initiation of crack propagation
4.2 Source Mechanics 179

t = 1.0 ns t = 5.0 ns

z
y
x

acoustic
t = 10.0 ns velocity
magnitude
[m/s]

100 μm

Fig. 4.44 Plot of acoustic velocity magnitude during crack growth at t ¼ 1.0 ns, t ¼ 5.0 ns, and
t ¼ 10.0 ns after initiation of crack propagation

to the tensile load case, the presence of two moving fracture surfaces causes
formation of two dipole-like radiation patterns instead of one. Their superposition
causes the wave propagation in the near field seen in Fig. 4.44. The dominant
radiation direction is found to be in the in-plane direction with stronger intensity
than for the tensile load case.
The resulting acoustic emission signals of compressive fiber failure are shown in
Fig. 4.45. They obey the same trend as function of angle like the signals of tensile
fiber failure, but tend to be larger in amplitude by almost one order of magnitude.
This can be partially attributed to the formation of two times the fracture surface
when compared to the tensile load case. Another factor is the tilted movement of the
fracture zone (including the fractured filament and the surrounding matrix) which
turns into larger microscopic displacements than for the tensile fiber failure case.

4.2.3.4 Fiber Bridging

Beyond the individual failure of the matrix material, the fibers, or the interface,
there are several failure mechanisms involving a specific arrangement of
micromechanical processes. As discussed in the previous section, compressive
180 4 Acoustic Emission

a 90°
45°

4
offset displacement [fm]

-2

0 20 40 60 80 100
time [μs]

b 1.4x10-9
90°
45°

1.2x10-9
FFT-magnitude [1/m]

1.0x10-9

8.0x10-10

6.0x10-10

4.0x10-10

2.0x10-10

0.0
0 100 200 300 400 500 600 700 800 900 1000
frequency [kHz]
Fig. 4.45 Acoustic emission signals of fiber failure due to σ k detected in unidirectional plate at
0 , 45 , and 90 relative to fiber orientation at 25 mm distance to source position in time domain
(a) and frequency domain (b)

fiber failure is such a process. Another failure mechanism in this context is fiber
bridging. Here the crack propagates transverse to the fiber axis, but the individual
filaments remain intact during crack propagation. Instead of fiber failure, a combi-
nation of fiber-matrix debonding, matrix cracking, and interfacial failure occurs
4.2 Source Mechanics 181

50 μm
tensile parallel
σ║

symmetry plane

(x,y,z) = (0,0,0) symmetry plane

z
y
x
fracture surface

7 μm
precrack

Fig. 4.46 Model configuration including position of symmetry plane, details of RVE position,
dimensions and details of crack region

(cf. Sect. 2.1). To model this phenomenon on the microscopic scale, the RVE
approach was used, and the von Mises criterion was applied to evaluate failure of
the matrix material within the RVE. As macroscopic load configuration, the tensile
load parallel case σ k was studied. To initiate the crack used for fiber bridging, a
precrack starting at the surface position of the laminate was chosen as indicated in
Fig. 4.46.
The selected position and dimensions of the precrack assure that due to a
macroscopic stress level of 2000 MPa, a microscopic load of σ k ¼ 52 MPa is
reached at the edge of the precrack. This fulfills the fracture condition for matrix
cracking and thus drives the crack propagation plotted in Fig. 4.47. As seen for
three distinct time steps after failure initiation, the growth of the fracture plane is
not straightforward along the z- or y-axis, but is influenced by the stress concen-
tration effect of the nearby fibers. Accordingly, the full fracture surface is finally
established by merging of three individual matrix areas between the surrounding
fibers after 5 ns. Since the model is calculated with symmetry conditions at the xz-
plane, this configuration would approximate an elliptical crack front, which
includes the precrack area.
After crack growth, the signal spreads out and propagates along the free surface
and into the volume of the laminate. As seen in the acoustic velocity magnitude
from Fig. 4.48, the intensity of the wave is comparatively low, and the
corresponding acoustic emission signals shown in Fig. 4.49 are among the lowest
intensities found for the cases studied herein. This is readily understood from the
small size of the fracture zone and the reduced mobility of the crack wall move-
ment. The latter is bridged by the fibers and these in turn limit the deflection of the
Fig. 4.47 Extension of fracture surface during crack growth at t ¼ 0.5 ns, t ¼ 1.0 ns, and t ¼ 5.0 ns
after initiation of crack propagation

t = 15 ns t = 20 ns

z
y
x

t = 25 ns acoustic
velocity
magnitude
[m/s]

50 μm

Fig. 4.48 Plot of acoustic velocity magnitude during crack growth at t ¼ 15 ns, t ¼ 20 ns, and
t ¼ 25 ns after initiation of crack propagation
4.2 Source Mechanics 183

a 90°
45°
0.10 0°

0.05
offset displacement [fm]

0.00

–0.05

–0.10

–0.15
0 20 40 60 80 100
time [μs]

b 90°
4x10–10 45°

3x10–10
FFT-magnitude [1/m]

2x10–10

1x10–10

0
0 100 200 300 400 500 600 700 800 900 1000
frequency [kHz]

Fig. 4.49 Acoustic emission signals of fiber bridging failure due to σ k detected in unidirectional
plate at 0 , 45 , and 90 relative to fiber orientation at 25 mm distance to source position in time
domain (a) and frequency domain (b)

crack walls along the fiber axis. Hence, the acoustic emission magnitudes are
expected to be rather low.
Especially for the acoustic emission signal detected along the 0 axis, some
high-frequency contributions are visible for the S0 mode at the beginning of the
184 4 Acoustic Emission

signal shown in Fig. 4.49. These are similar to the frequencies seen for the fiber
breakage cases, which are in turn linked to the size of the fracture zones. But the
dominant contribution in the present case are lower frequencies originating from
the A0 mode contributions starting after approximately 10 μs. These are predomi-
nantly due to the chosen source position at the surface of the laminate. Thus a
decrease of their intensity is expected if fiber bridging would occur inside the
laminate (cf. Sect. 4.2.3.7).

4.2.3.5 Influence of Crack Length

Beyond the distinct relationship of individual source types and their acoustic
emission signals, it is worthwhile to consider some further factors of influence.
First a comparison is made between the motions of the crack surface caused by
three different crack propagation lengths for inter-fiber failure in the σ ⊥ load case.
Therefore the y-displacement of the crack surface position is evaluated following
the procedure described in Sect. 4.2.3.1. To this end, the length of the thin elastic
layer is adjusted to allow crack propagation only to the designated maximum
length. The respective evaluation of the y-displacement of the crack surface is
shown in Fig. 4.50 as comparison of the three cases. During the duration of crack
propagation, the y-displacement increases almost linearly. However, at the moment
of crack arrest, the y-displacement does not stop or cease, but increases further until
a maximum value is reached as discussed previously in Sect. 4.2.3.1. Subsequently,
the crack surface starts to vibrate and finally settles at a new equilibrium position.
The predictions of the y-displacement derived from the theory of Green [40] are

22 length 5 mm
20 length 1 mm
length 0.1 mm
18
16
displacement [μm]

14.5 μm
14
12
10
8
6
4
2.9 μm
2
0 0.3 μm
0 5 10 15 20 25 30 35 40
time [μs]

Fig. 4.50 Evaluation of source displacement in y-direction for σ ⊥ load case with three different
maximum propagation lengths and dashed lines to mark predictions from analytical theories
4.2 Source Mechanics 185

marked as dashed lines in Fig. 4.50. As can be seen from the comparison, in the
initial part, these values are systematically lower than the present model predic-
tions. This originates from the underlying assumptions in the theory. In [40] static
crack opening is assumed, i.e., the length of the crack is already present as a flaw in
the material and merely opens due to an external force. If the same assumptions are
made in the present model, the achieved deformation state is in good agreement to
the values predicted by [40]. However, the dynamic crack propagation seems to
generate initial acoustic emission amplitudes larger than predicted by analytical
theories, which has also been reported for the case of isotropic materials
before [36].

4.2.3.6 Influence of Fracture Surface

Another factor of influence contributing to the intensity of the acoustic emission


signal is the size of the fracture surface. To demonstrate this effect, the tensile
parallel load case σ k for fiber failure is chosen. As major difference to the failure
model shown in Fig. 4.38, the position of the pore acting as precrack and the
number of failing filaments were varied. Studies are carried out for failure of one,
three, and five filaments. In this context, the number of filaments failing simulta-
neously determines the effective fracture surface as shown in blue in Fig. 4.51.

fracture surface
0.25 1 filament
3 filaments
5 filaments
offset displacement [fm]

0.20
crack growth
σ||
0.15
100 μm

0.10

0.05
crack growth
σ||

0.00

0 20 40 60 80 100

7 μm z
time [μs]
y
x
crack growth
σ||

Fig. 4.51 Schematically representation of fracture surface within the RVE marked in blue (left
side) and evaluation of acoustic emission signals due to fiber failure with one, three, and five
filaments in unidirectional laminate at 25 mm distance to source position at 0 angle relative to the
fiber orientation (right side)
186 4 Acoustic Emission

The position of the pore was chosen at the bottom of the lowest fiber filament, and
crack propagation was found to be mostly oriented along the z-axis direction as
indicated by the arrow in Fig. 4.51. Based on the generalized theory of acoustic
emission [21, 22], a larger fracture surface translates into larger vibrating volume ΔV
if the same stress level is applied to the fracture surface. Since the external load level
was kept identical for all model configurations at Fk ¼ 263 kN, the dominant change
is solely due to the change in fracture surface. As seen from the detected acoustic
emission signals in Fig. 4.51, this change in fracture surface directly translates into a
change in signal amplitude. Here the higher the numbers of failing filaments, the
higher the signal amplitudes were calculated. The observed frequencies of the signals
are still comparable, since the rise-time is still in the same order of magnitude.

4.2.3.7 Influence of Depth Position

So far all crack positions were chosen symmetric with respect to the neutral axis of
the plate in order to avoid any influences due to asymmetric source positions as
described in Sect. 4.2.1. However, for the unidirectional fiber reinforced composite
used in this study, there is no preferential position for fiber breakage along the
z-axis. Therefore, various depth positions below the surface were studied to com-
pare the resulting out-of-plane displacement signals as seen in Fig. 4.52. Except for
the position of the fiber breakage sources, all model parameters are chosen identical
to the description given in Sect. 4.2.3.3 for single fiber filament failure.

0.35 depth = 0.750 mm


depth = 0.625 mm
0.30 depth = 0.500 mm
0.25
0.500 mm

offset displacement [fm]

0.20
σ║
0.15

0.10
0.625 mm

0.05

0.00
σ║ –0.05

–0.10
0.750 mm

0 20 40 60 80 100
time [μs]
z
σ║
y
x

Fig. 4.52 Evaluation of acoustic emission signals due to fiber failure at three different depth
positions in unidirectional laminate at 25 mm distance to source position at 0 angle relative to the
fiber orientation
4.2 Source Mechanics 187

Starting at the midplane of the plate (z ¼ 0.50 mm), the RVE region was
gradually moved to a position shortly below the surface with the fiber breakage
being at z ¼ 0.75 mm (cf. Fig. 4.52). For better visibility only one half of the RVE
region is shown. For the tensile parallel load case σ k , failure of one fiber filament
results in the out-of-plane displacement signals shown in Fig. 4.52. As expected by
the principles of modal AE, the gradual shift from the midplane to the surface
causes a noticeable increase of the A0 mode. However, there is still noticeable
contribution of the high-frequency components propagating as S0 mode. In fact,
there is no reduction of bandwidth for the three signals. But there is a distinct shift
of the relative intensity of these contributions, causing higher contributions at lower
frequencies with increased distance to the midplane. Here the A0 mode also causes
increased absolute signal amplitudes, which is beneficial from the detection point
of view.

4.2.4 Detectability of Failure Mechanisms

Based on the multitude of calculated signals for different failure mechanisms


presented in the previous section, it is possible to assess the detectability of different
failure mechanisms by AE measurements. As will be demonstrated in Sect. 4.4, the
sensitivity of AE sensors is extraordinary high, but still there are physical limita-
tions. Various authors claim the expected out-of-plane sensitivity of acoustic
emission sensors to be in the order of 1013–1014 m [42, 43] and thus will act as
ultimate detection limit in the following. Using the investigations of the previous
section, all signals are evaluated with respect to their maximum absolute amplitude
within the first 100 μs. This procedure is identically carried out for signals of all
source mechanisms and for all detection angles to account for these effects as
scatter of the obtained values.
The aim of this subsection is twofold. First, the general relationship between the
detectable acoustic emission signal amplitude and the physical size of the failure
mechanism is established. Second, the aspect of detectability in terms of the
distance and orientation relative to the source position is discussed.
Figure 4.53 presents the results of the calculated out-of-plane displacement of
the acoustic emission signal plotted as function of the “crack area.” For the
quantification of the acoustic emission signal, the absolute value of the out-of-
plane displacement is used, since detectability by acoustic emission sensors does
not distinguish between positive or negative amplitude values. The “crack area” is
quantified as area of the fracture surface ΔA finally reached by the respective failure
type (i.e., for single fiber filament failure, this is equivalent to the cross-section of
one filament). In general, the vibrating volume ΔV is the acoustically relevant
quantity. The vibrating volume is related to ΔA by considering the respective
displacement normal to the surface A. However, the quantity ΔV cannot be readily
derived from simple geometrical considerations and therefore does not constitute a
188 4 Acoustic Emission

0.01 25 mm distance 10 mm x 1mm inter-fiber failure


1E-3 delamination
1E-4 fiber breakage
10μm x 10μm 100μm x 100μm
1E-5
AE surface displacement [m]

1E-6 100plet scatter


1E-7
1E-8 1plet

1E-9 scatter
1E-10
1E-11
1E-12
1E-13
1E-14
strength absolute
1E-15 distribution detection limit
1E-16 fiber bridging
1E-17
1E-11 1E-10 1E-9 1E-8 1E-7 1E-6 1E-5 1E-4 1E-3
crack area [mm2]

Fig. 4.53 Out-of-plane component of acoustic emission signal detected at 25 mm distance to the
source for different failure types as function of the crack area in 1 mm thick T700/PPS plate

useful quantity to assess detectability. Instead, the size of the fracture zone ΔA can
be understood in close analogy to the flaw size as typically used in NDT standards
and therefore allows better interpretability of Fig. 4.53. To facilitate reading further,
some exemplary crack dimensions representative of the scales are noted in Fig. 4.53
as black vertical lines.
As expected from the physical dimensions of the fracture zones modeled, the
detectable crack areas span five orders of magnitude. Similarly, the detected signal
amplitudes cover eight orders of magnitude. Hence, both axes are shown in a
logarithmic scale to allow for better visibility of the individual signals. There is a
general trend common to all failure mechanisms showing larger signal amplitudes
with larger crack areas. Based on the fundamental considerations made in (4.1) and
(4.2), this finding is not surprising.
For the failure mechanism of fiber breakage, the variation of fiber filament
strength causes a characteristic distribution of expected AE signal amplitudes for
one particular crack area since the latter was kept constant. This is indicated
exemplarily for the case of one fiber filament. Additional scatter arises from the
intensity radiated to the different angles relative to the source. Both effects define a
mean value of fiber filament failure for the present case and a respective scatter
expected due to the underlying Weibull strength distribution and the extreme cases
of observation angles relative to the source position. For an increasing number of
fiber filaments failing simultaneously, the detected AE signal amplitude increases
4.2 Source Mechanics 189

as does the crack area. Given the applied macroscopically tensile stress is kept
constant, a power law fit indicated by the solid red line was found to describe the
relationship between these two quantities best. This allows a slight extrapolation of
this relationship beyond the case of 100 fibers failing simultaneously (which was
the largest number studied by numerical methods). Accordingly, the extrapolation
of the scatter band establishes an approximate area in Fig. 4.53 where signal
amplitudes of all fiber failures at a detector distance of 25 mm are expected
regardless of the orientation of the detector position relative to the source. It is
worth noting that at the distance of 25 mm, some of the weak fiber filament failures
are approaching the absolute detection limit of 1014 m and thus may not be
detected in an experimental setup.
Signals of fiber bridging are found close to the intensities of single fiber failure
and due to the crack areas studied herein also have closely related crack areas. In
average these signal amplitudes are even lower, which is mostly due to the lower
macroscopic tensile load required to generate this failure mechanism and the
constrained movement due to the fibers bridging the crack walls. Hence these
signals also partially fall below the detection limit at 25 mm distance.
The different cases of inter-fiber failure studied in the previous section typically
show larger crack areas and, consequently, also larger signal amplitudes. For a
particular source mechanism, there is an almost linear relationship in the log-log
scale as indicated by the solid blue lines as a result from two power law fits for the
cases of inter-fiber failure due to σ ⊥ and the inter-fiber failure due to τk⊥ , both with
varying crack lengths. However, it is difficult to spot a general relationship covering
all of the different load cases. Also, data points in Fig. 4.53 originate from all angles
of orientation relative to the source position as well as results from the unidirec-
tional plate and the cross-ply plate. Therefore the blue rectangle is sufficient to
cover all cases of inter-fiber failure studied herein and can thus be used as areal
representation of inter-fiber failure. Within this area all AE signals of inter-fiber
failure detected at 25 mm to the source should locate, regardless of their specific
type and angular position relative to the detector. The lower and upper boundary of
the rectangle in terms of the crack area are chosen arbitrarily at dimensions
approaching the range of the microscopic constituents of the composite on one
end and at dimensions approaching the cross-section of the plate investigated on the
other end. Hence these boundaries do not constitute absolute limits of the relation-
ship, but are limits dictated by the chosen size of the model.
For the case of inter-ply delamination, results of the unidirectional plate and the
cross-ply plate are shown in Fig. 4.53. Marked in green color, these seem to
generally exhibit lower AE signal amplitudes for the same crack area when
compared to inter-fiber failure. The result of a power law fit for the case of inter-
ply delamination due to τk⊥ with varying crack area is shown as solid green line.
Based on the same considerations as for inter-fiber failure, the green rectangle is
established as area of largest likelihood for the location of AE signals due to inter-
ply delamination.
190 4 Acoustic Emission

To summarize the findings shown in Fig. 4.53, the marked areas related to the
occurrence of inter-fiber failure, delamination, and fiber breakage include all
extreme cases of signals as arising due to:
(a) Variation of detector position by an angle from 0 to 90
(b) Signals of a unidirectional and a cross-ply laminate
(c) Variation of source strength at identical crack area
(d) Variation of source depth position within the laminate
As seen from Fig. 4.53, reasonable overlap between the calculated AE signal
amplitudes for the different failure mechanisms occurs. So based on this investiga-
tion, there is a general trend for the intensity range of the individual mechanisms, but
there is no decisive boundary to separate failure mechanisms based on their intensity
for the types of failure mechanisms studied herein. Such classification can instead be
achieved by means of suitable frequency analysis as discussed in Sect. 4.5.
To allow more general conclusions on detectability of AE signals in fiber
reinforced materials, the results of Fig. 4.53 may be further generalized with respect
to the attenuation of signals with distance to the source position. Considering the
signal attenuation, this allows establishing a 3D detectability diagram as derivation
of the 2D cross-section at 25 mm distance shown in Fig. 4.53. In the following a
value of 0.1 dB/mm was chosen as extreme case found for the T700/PPS plates
investigated.
In Figs. 4.54, 4.55, and 4.56, an extrapolation of the envelopes of each failure
type presented in Fig. 4.53 is extrapolated to a distance of 500 mm between source
and detector. For comparison the gray plane shows the upper level of the absolute

Fig. 4.54 Extrapolation of envelope of out-of-plane component of acoustic emission signal due to
inter-fiber failure to 500 mm detection distance
4.2 Source Mechanics 191

Fig. 4.55 Extrapolation of envelope of out-of-plane component of acoustic emission signal due to
interfacial failure to 500 mm detection distance

Fig. 4.56 Extrapolation of envelope of out-of-plane component of acoustic emission signal due to
fiber breakage to 500 mm detection distance

threshold limit at 1013 m. As seen in Fig. 4.54 for the case of inter-fiber failure, a
part of the signals falls already below this limit at 25 mm distance. Accordingly,
due to attenuation a larger fraction of signals falls below this limit at higher
distances. The 3D envelope shown in Fig. 4.54 may be used to estimate if a certain
192 4 Acoustic Emission

crack size will be detectable by means of AE measurements. As mentioned previ-


ously, the envelope was found to be applicable for all angles of detection relative to
the source and unidirectional and cross-ply stackings. However, results are based on
the material properties of T700/PPS, and the attenuation coefficient was assumed to
be linear and isotropic and constitutes an extreme value. Hence, the envelope may
be considered as conservative approach for detection probability of the AE method.
Similar diagrams are shown in Figs. 4.55 and 4.56 for the cases of interfacial
failure and fiber breakage. The case of interfacial failure may be readily understood
in analogy to the matrix cracking case, yet the overall number of signals that may
fall below the detection limit was assessed to be higher. Accordingly, less PoD may
be expected at higher distances as well. For the extrapolation of the envelope of
fiber breakage shown in Fig. 4.56, one remarkable difference arises. It was already
shown in Fig. 4.53 that a part of the signals corresponding to failure of weak
filaments falls below the detection limit already at 25 mm distance. Thus
Fig. 4.56 predicts the loss of all signals of single fiber filament failure after a certain
distance to the source. Since the occurrence of fiber breakage is a crucial indicator
to the structural health of a fiber reinforced composite, this is of vital importance.
For the T700/PPS material investigated here, this distance was evaluated at
400 mm distance. Keeping in mind the conservative assumptions made regarding
the attenuation coefficients, this distance might be estimated slightly too short but is
certainly within a realistic range.

4.3 Wave Propagation

After the emission of an acoustic wave by the source, the signal propagates into the
solid and is subject to geometric spreading. Upon propagation the wave is affected
by dispersion and attenuation effects due to dissipative mechanisms in the material
and formation of distinct wave modes in finite specimen geometries.
According to Landau and Lifschitz, the deformation of a solid causes heat to be
generated at the deformation zone which fluctuates in time and space. But if the
spatial heat propagation is slower than the oscillatory movements within the solid,
the deformations can assumed to be adiabatic [16]. For this case, the theory of
elastic waves describes the propagation of deformations within the solid for small
initial deformations. This yields the basic theoretical formulation for the under-
standing of acoustic emission given by the theory of elastic waves, which in general
requires solving the momentum balance equation for the case of negligible
Lagrangian inertia:
2
∂ u ∂σ
ρ ¼ ð4:11Þ
∂t2 ∂r
4.3 Wave Propagation 193

Here ρ is the material density, u is the displacement, and σ is the stress tensor
derived from the generalized formulation of Hooke’s law:

σ¼Cε ð4:12Þ

This equation, also known as constitutive relation, couples the stress tensor σ with
the deformation tensor ε using the elasticity tensor C for the case of small elastic
deformations in a homogeneous medium. In general C is a fourth-order tensor with
81 elastic coefficients. Since the deformation tensor ε is symmetric, the elasticity
tensor C exhibits the same symmetry constraints regarding index permutation,
which reduces the number of independent coefficients for a solid material to
21 [16]. The occurrence of symmetries within the solid reduces the number of
independent coefficients further. For the case of orthotropic materials like a com-
posite laminate, the elasticity tensor C can be reduced to a 6  6 matrix with nine
independent elasticity coefficients and one angle defining the axis orientation in the
xy-plane. Using the Voigt notation, the stress tensor σ can be reduced to a vector
with six independent components, and the deformation tensor ε is also written as a
vector with six independent components. Hence, (4.12) may be rewritten in matrix
form as
0 1 0 1 0 1
σ1 C11 C12 C13 0 0 0 ε1
B σ 2 C B C12 C22 C23 0 0 0 C B ε2 C
B C B C B C
B σ 3 C B C13 0 C B C
B C¼B C23 C33 0 0 C  B ε3 C ð4:13Þ
B σ4 C B 0 C
0 C B C
B C B 0 0 C44 0 B ε4 C
@ σ5 A @ 0 0 0 0 C55 0 A @ ε5 A
σ6 0 0 0 0 0 C66 ε6

If the solid is isotropic, (4.13) is reduced to


0 1 0 1 0 1
σ1 κ þ 2ς κ κ 0 0 0 ε1
B σ2 C B κ κ þ 2ς κ 0 0 0C B ε2 C
B C B C B C
B σ3 C B κ κ κ þ 2ς 0C B C
B C¼B 0 0 C  B ε3 C ð4:14Þ
B σ4 C B 0 ς 0C B C
B C B 0 0 0 C B ε4 C
@ σ5 A @ 0 0 0 0 ς 0 A @ ε5 A
σ6 0 0 0 0 0 ς ε6

The remaining two independent elements κ and ς are named Lame’s constants.
These completely describe all elastic parameters for isotropic materials like
Young’s modulus EY, compressive modulus EC , shear modulus ES, and Poisson’s
ratio ν.

3κ þ 2ς
EY ¼ ς ð4:15Þ
κþς
194 4 Acoustic Emission

2
EC ¼ κ þ ς ð4:16Þ
3
κð1  2ςÞ
ES ¼ ð4:17Þ

κ
ν¼ ð4:18Þ
2ðκ þ ςÞ

For such isotropic materials, (4.11) can be simplified using the Lame’s constants to
2
∂ u
ρ ¼ ðλ þ ςÞ∇ð∇  uÞ þ μΔu ð4:19Þ
∂t2

The solution of (4.19) can then be formulated with scalar and vector potentials as

u ¼ ∇Χ þ ∇  ψ: ð4:20Þ

This yields the two independent wave equations in solid isotropic media:
2
1 ∂ ς
ΔΧ ¼ ð4:21Þ
c2L ∂t2
2
1 ∂ ψ
Δψ ¼ ð4:22Þ
c2T ∂t2

Equations (4.21) and (4.22) use the definition of the longitudinal and transversal
propagation velocities cL and cT for an elastic wave given as
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
EY ð1  νÞ
cL ¼ ð4:23Þ
ρð1 þ νÞð1  2νÞ
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
EY
cT ¼ ð4:24Þ
2ρð1 þ νÞ

With respect to the boundary conditions, (4.21) and (4.22) can be used to obtain
solutions for wave propagation in arbitrary geometries. In infinite isotropic and
homogeneous media equations, (4.21) and (4.22) have solutions of the form of a
monochromatic plane wave with an amplitude u, wave vector k, and angular
frequency ω:

uðr; tÞ ¼ u  eiðkrωtÞ : ð4:25Þ

Here the propagation velocities cL and cT are identical to the respective phase
velocities cP of the plane waves. The propagation of an elastic wave in this case
can thus be described by independent longitudinal wave and transversal wave
propagation.
4.3 Wave Propagation 195

In general, for non-monochromatic waves and anisotropic or finite media, the


propagation velocity of the wave is the group velocity cG defined as

∂ωðkÞ
cG ¼ ð4:26Þ
∂k

Similarly, the phase velocity cP is defined by the relationship between k and ω,


which is called dispersion relation:

ωðkÞ
cP ¼ ð4:27Þ
k

For geometries of finite extent, acoustic wave propagation has to take into account
the respective boundary conditions. This ultimately results in wave propagation
which is “guided” by the geometry of the propagation medium. The most prominent
types of guided waves are the surface waves, which occur for a solid with free
surface. These were first predicted by Lord Rayleigh in 1885 for the special case of
an infinite solid with free surface. Later Stoneley and Love generalized Rayleigh’s
description for the case of a solid with an adjacent liquid half-space and planar
interfaces located between two half-spaces [44, 45]. The guided waves in such
structures with liquid-solid interfaces are hence often called Stoneley or Love
waves. For infinitely extended solids with two parallel free surfaces, H. Lamb
published the respective solutions for the wave equation in 1917 [46]. According
to the geometry of the propagation medium, the respective guided elastic waves are
commonly called “plate waves.” These guided waves are the dominant mode of
wave propagation in fiber reinforced structures, since these are often designed in the
geometry of shell-like structures. In other types of geometries like rods or cylinders,
similar types of guided waves exist. Due to the planar structure of the experimental
specimens used for the present study, the propagation behavior of guided waves is
very important and will be discussed in more detail in the following. For a closer
review about Rayleigh waves, the reader is referred to [47], while other types of
guided waves are comprehensively discussed in [48–50].
In order to understand guided waves, it is necessary to understand the behavior
of elastic waves at interfaces. The relevant physical quantity for this case is the
acoustic impedance Z, which is defined using the density ρ and the respective sound
velocity ci as
Z ¼ ρci : ð4:28Þ

If a planar acoustic wave arrives with orthogonal incidence to an interface, the wave
is partially reflected and transmitted dependent on the acoustic impedance of the
two materials. When the wave propagates from medium one with impedance Z1 to
medium two with impedance Z2 as shown in Fig. 4.57, the coefficients for reflection
Re and transmission Tr are given by

Z2  Z1
Re ¼ ð4:29Þ
Z2 þ Z1
196 4 Acoustic Emission

Fig. 4.57 Schematic inc


ide Z1 Z2
representation of law of nt
wa
refraction for acoustic ve c1 c2
waves (based on [35]) θ1
transm θ2
w ave itted w
ave
R ec ted
refl R

2Z2
Tr ¼ ð4:30Þ
Z2 þ Z1

For incidence angles other than θ1 ¼ 0 , the law of refraction is applied. In contrast
to electromagnetic waves, elastic waves usually undergo modal conversion during
transmittance; thus, longitudinal waves can be completely or partially converted to
transversal (shear) waves dependent on the incidence angle. In the most general
case, the law of refraction can be written for arbitrary combinations of longitudinal
and transversal sound velocities as

sin ðθ1 Þ c1
¼ : ð4:31Þ
sin ðθ2 Þ c2

For certain incident angles, the transmitted parts of the elastic waves are of very
special nature. The most important among them are the surface waves which occur
for elastic wave incidence from gaseous or liquid media to solids under certain
angles. The total reflection condition for the longitudinal wave is termed first
critical angle and causes excitation of a creep wave propagating along the surface
of the solid having a propagation velocity identical to the longitudinal wave and a
fast energetic decay. More important for practical applications is the Rayleigh wave
which occurs under the second critical angle given by the total reflection condition
of the transversal wave. L. Bergman proposed an approximation formula for the
sound velocity of Rayleigh waves [51]:

0:874 þ 1:12ν
cR cT ð4:32Þ
1þν

The velocity of the Rayleigh wave is generally slower than the transversal wave
velocity of the solid. The amplitude of the Rayleigh wave decays exponentially
with depth position below surface level. This is the reason why these types of waves
are called “surface” waves as they propagate and interact only close to the solid
surface.
If the dimension of the solid is reduced and approaches the shape of a plate with
thickness zd, the surface wave is further influenced and another type of guided wave
propagation results. Such a plate with infinite extensions in the x and y direction and
a finite thickness in z-direction defines boundary conditions for the solution of the
4.3 Wave Propagation 197

acoustic wave equation which result in a classic eigenvalue problem. The respec-
tive solution of the acoustic wave equation is given by:

ux ¼ f x ðzÞeiðkxωtÞ
uy ¼ 0 ð4:33Þ
iðkxωtÞ
uz ¼ f z ðzÞe

Equation (4.33) describes the propagation of an elastic wave in x-direction as


function of wave number k, frequency ω, and as function of the z-position fi(z).
The (x, y) origin is located in the medial plane of the plate as shown in Fig. 4.58. The
specific type of guided plate waves described by (4.33) is referred to as Lamb waves
which exhibit no displacement in y-direction for an infinite plate. Complementary
to the Lamb waves, the so-called shear-horizontal waves exist, which have no
displacement in the x- or z-directions and thus propagate only in y-direction.
If the boundary condition σ ¼ 0 is applied at the surface level 0:5 h of a plate
with thickness h, this results in two characteristic periodic equations for symmetric
and antisymmetric motion with respect to the plane z ¼ 0 as schematically
represented in Fig. 4.58:

e
tan 0:5 Bh e Bk
e 2
4A
 ¼  ð4:34Þ
e
tan 0:5 Ah k2  Be2
2

  2
e
tan 0:5 Bh k2  Be2
 ¼ ð4:35Þ
e
tan 0:5 Ah 4Ae Bk
e 2

particle movement particle movement


surface surface

neutral axis neutral axis


h

surface surface

(x,y,z)=(0,0,0) (x,y,z)=(0,0,0)
propagation direction propagation direction

z Antisymmetric (A) mode Symmetric (S) mode

Fig. 4.58 Schematic representation of symmetric and antisymmetric Lamb wave modes propa-
gating in x-direction with strongly emphasized surface motion (based on [35])
198 4 Acoustic Emission

Using the abbreviations A~ and B:


e

e 2 ¼ ω  k2
2
A ð4:36Þ
c2L

e 2 ¼ ω  k2
2
B ð4:37Þ
c2T

The equations (4.34) and (4.35) implicitly relate the wave number k and the
frequency ω to the plate thickness h. Because the velocity of propagation is
dependent on the frequency, this results in dispersion. For Lamb waves the relation
between phase velocity cP and group velocity cG is thus given by
0 1
B C
B C
B 1 C
cG ¼ cP ðωhÞB1  cP C: ð4:38Þ
B 1 C
@ dcP A
ωh
dðωhÞ

The relationship between frequency, plate thickness, and phase or group velocity is
typically visualized within a set of dispersion curves for symmetric (Si) and
antisymmetric (Ai) Lamb wave modes as well as the shear-horizontal modes
(SHi). An example of such dispersion curves for a 1.0 mm thick unidirectional
T800/913 composite plate using the material properties of Table B.1 in Appendix B
is shown in Fig. 4.59. Since the wave propagation is dependent on the orientation,
Fig. 4.59a displays the dispersion curves for a propagation direction parallel to the
fiber orientation. For a propagation direction perpendicular to the fiber orientation,
the dispersion curves appear differently as seen in Fig. 4.59b. This is due to the
acoustical anisotropy introduced by the orientation of the fibers.
The dispersion curves show the characteristic dependency of the phase velocity
and group velocity on the product ω  h for the symmetric and antisymmetric wave
modes. The two fundamental modes exhibiting the lowest frequency for a given
wave number start at index i ¼ 0, whereas higher-order modes are iterated consec-
utively. In principle, there are infinite numbers of Lamb wave modes, which can
exist within a plate. However, the types of modes faced most often in thin-walled
fiber reinforced structures are the fundamental symmetric mode (S0) and antisym-
metric mode (A0), often referred to as extensional and flexural mode.
In guided wave testing, Lamb waves are typically excited by a plane elastic wave
with appropriate angle of incidence ϕ as shown in Fig. 4.60. The excited wave
modes in turn depend on the used frequency and plate thickness. If the wavelength λ
of the used frequency is chosen inappropriately, the incident elastic wave is
reflected at the top and bottom boundaries and propagates in the same way as in a
bulk medium (Fig. 4.60, top). If the wavelength is sufficiently high or the thickness
4.3 Wave Propagation 199

a A0
A1
10000 SH0
S0
8000 S1
group velocity [m/s]

S2

6000

4000

2000

0
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5
.
frequency thickness [MHz.mm]
b
A0
A1
3000
SH0
S0
group velocity [m/s]

S1
S2
2000

1000

0
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5
. .
frequency thickness [MHz mm]

Fig. 4.59 Set of dispersion curves calculated for a 1.0 mm thick unidirectional fiber reinforced
material for propagation direction parallel to the fiber orientation (a) and perpendicular to the fiber
orientation (b)

zd is small enough, at the same angle of incidence, interference occurs and causes
excitation of Lamb waves (Fig. 4.60, bottom).
As discussed in Sect. 4.2, the excitation process of Lamb waves due to acoustic
emission sources is different and is affected by various additional factors.
One particular difference to conventional guided wave testing is the bandwidth
of AE sources. In guided wave testing, usually a center frequency is chosen, which
200 4 Acoustic Emission

neutral axis
h

φ λ1

neutral axis
z h

x λ2>λ1 Lamb-wave propagation

Fig. 4.60 Excitation by planar acoustic wave with angle of incidence ϕ and small wavelength λ1
(top) and larger wavelength λ2 causing Lamb wave excitation due to interference (bottom) (based
on [35])

is enveloped by a Gaussian bell to yield a finite duration wave package. Although


the frequency content of such an enveloped center frequency is theoretically
infinite, still the dominating component is the center frequency. Therefore, one
can select a particular propagation velocity of a wave mode at a particular fre-
quency. Compared to that, AE sources typically show broad bandwidth, which is
not controlled by the operator, but controlled by the AE source. Consequently,
multimode excitation and variation of intensity as function of frequency may occur.
But even for equal intensity within a certain bandwidth, unequal excitation of the
according wave modes is found (cf. [31, 34]). As seen in Fig. 4.61 for
the calculations of a dipole source in aluminum plate as discussed in Sect. 4.2.1,
the source bandwidth falls in the range up to several MHz. Despite of the source
spectrum in Fig. 4.61a being smooth with frequency in the range up to 1 MHz, the
branches of the S0 mode and A0 mode are not excited equally, but predominantly in
local regions (see Fig. 4.61b). This translates in characteristic frequencies of
the final AE signal as seen in the respective frequency spectrum of the AE signal
(see Fig. 4.61a).
For other geometries as rods or cylinders, other types of guided waves exist,
which can also be excited by acoustic emission sources. Common to all guided
waves is the fact that these need a certain distance to fully develop. Typically this is
dominated by the relevant geometrical boundary constraint such as the plate
thickness or the rod diameter. Thus guided waves may only be interpreted after
4.3 Wave Propagation 201

a
1E-8 spectrum of signal at 240 mm distance
spectrum of source function

1E-9
FFT-magnitude [1/m]

1E-10

1E-11

1E-12

1E-13
0 100 200 300 400 500 600 700 800 900 1000
frequency [kHz]

b
0.0E+00
4.5E+03
800 preferential 9.1E+03
frequency [kHz]

excitation 1.4E+04
1.8E+04
600 2.3E+04
2.7E+04
400 3.2E+04
almost no 3.6E+04
excitation 4.0E+04
200

50 100 150 200


displacement [fm]

0.5

0.0

–0.5

0 50 100 150 200 250


time [μs]

Fig. 4.61 Source spectrum of a buried dipole in an aluminum plate (a) and according Choi-
Williams distribution at 234 mm distance to the source showing preferential excitation of Lamb
wave modes (b)

some distance to the source. More information on this topic can be found, e.g., in
[48, 49, 52].
Lamb wave propagation in undamaged carbon fiber reinforced polymers (CFRP)
has been investigated extensively before [53–56]. Scattering of Lamb waves at
202 4 Acoustic Emission

internal damage, like cracks or delamination, is the key principle for structural
health monitoring of CFRP by guided wave testing [57]. The impact of such
discontinuities on acoustic emission analysis has been investigated less and is
therefore reviewed in more detail in Sect. 4.3.3. Since the guided wave propagation
is the carrier of information of AE source activity in the material, distortion of this
information due to interaction of Lamb waves with internal discontinuities is closely
related to the question of reliability of the information. Changes of modal intensity or
occurrence of alternative propagation paths due to scattering can readily affect source
localization accuracy and/or complicate source identification procedures.
In principle, numerical methods for analytical solutions to describe wave prop-
agation in multilayered structures were already given by Thomson and Haskell
[58, 59]. However, many numerical challenges like the problem of ill-conditioned
matrices as used for the transfer-matrix method had to be overcome. A good review
on those developments is found in [54]. Also, analytical predictions for wave
propagation in multilayered structures are given by Mal and Kundu for the case
of infinite half-spaces [60]. Compared to analytical solutions, a finite element
approach to model wave propagation in fiber reinforced materials is advantageous
in two ways. First, arbitrary geometries can be implemented without substantial
modifications of the chosen approach. In comparison, for analytical approaches
boundary conditions along with complex geometries are often difficult to imple-
ment. Second, three-dimensional FEM takes into account the boundary reflections
of edges and additional interfaces, which are rarely captured adequately by analyt-
ical approaches. As has been demonstrated by measurements using a laser
vibrometer, the accuracy of the FEM approach used in the following is sufficient
to describe acoustic emission signal propagation in a variety of cases [53, 61, 62].
In general, the propagation of the acoustic waves is subject to the boundary
conditions formed by the propagation medium geometry. In addition, the propaga-
tion of acoustic waves is subject to attenuation which significantly alters the
appearance of the waves as function of propagation distance and is therefore
discussed in the subsequent section.

4.3.1 Attenuation

The propagation of acoustic waves in an infinite isotropic medium exhibits no loss


of energy. However, for real materials, the elastic waves experience several attenu-
ation effects. As demonstrated by Auld, attenuation of acoustic waves propagating in
solids close to room temperature can be formulated by a viscous damping term
[63]. As shown in [55, 63–65], the acoustic wave attenuation is sufficiently described
by introducing a complex stiffness tensor C* linking the stress tensor σ to the
deformation tensor ε:

σ ¼ C* ε ð4:39Þ
4.3 Wave Propagation 203

This formulation can be rewritten using complex elastic coefficients if the respec-
tive acoustic wave field has eiωt time dependence:

C* ¼ ðC þ iωC 00 Þ ð4:40Þ

Acoustic attenuation is then defined as decay of amplitude of an acoustic wave


between two observation positions 1 and 2:
 2
Attenuation ¼ 10 log eαðu1 u2 Þ dB ð4:41Þ

This introduces the attenuation coefficient α used as measure for the damping
properties of a material. The attenuation is typically expressed in the dB scale to
cover the large range of attenuation effects. As discussed by Pollock [66] and
Prosser [67] from the viewpoint of acoustic emission, there are five contributions
for attenuation of acoustic emission signals, which are discussed in more detail in
the following.

4.3.1.1 Geometric Spreading

In the near field, close to the acoustic emission source, geometric spreading is the
main reason for attenuation. This is caused by the spherical radiation of energy into
the solid volume, which results in an energetic decay per dihedral angle with
inverse propagation distance r. When Lamb waves are excited, the attenuation in
the near-field range will even be more pronounced, since the elastic wave energy is
separated into the distinct symmetrical and antisymmetrical modes. In addition, the
Lamb waves suffer from velocity dispersion as will be described below. According
to [67] for the case of Lamb waves, the distance of dominating attenuation by
geometric spreading was estimated as r < 4:34=α.

4.3.1.2 Thermoelastic and Akhiezer Dissipation

In the far field of the acoustic emission source, the main contribution to attenuation
at room temperature originates from the thermoelastic mechanism and the Akhiezer
mechanism. The first one results from irreversible heat conduction originating from
compression during propagation of the longitudinal wave. The second mechanism
arises from the disturbance of the equilibrium distribution of thermal acoustic
waves (phonons). The propagation of a coherent acoustic wave disturbs the equi-
librium positions of the phonons and thus leads to a dissipative energy contribution.
For homogeneous isotropic solids, the thermoelastic and the Akhiezer mechanism
result in attenuation for elastic waves dependent on the square of the oscillation
frequency ω2 .
204 4 Acoustic Emission

4.3.1.3 Dispersion

For guided waves in anisotropic solids, even more effects contribute to the total
acoustic attenuation. Specifically for the case of Lamb waves, the total attenuation
is affected by spatial dispersion and frequency dispersion [68]. Attenuation from
spatial dispersion is caused by the dependency between phase velocity and wave
vector inherent in (4.34) and (4.35). As discussed by Ward et al. for dispersive
media, the phase velocity is directly linked to the attenuation by Kramers-Kronig
relations deduced from the principle of causality [69]. The effect of attenuation due
to frequency dispersion arises in propagation of non-monochromatic wave pack-
ages. Since each frequency component travels at a distinct velocity, an initially
short pulse begins to spread in time during propagation. This causes attenuation
with propagation length. For the case of Lamb waves, this is of great importance
since the different modes exhibit different α-coefficients. For CFRP plates the
attenuation of the antisymmetrical modes is still higher than the attenuation of the
symmetrical modes. This effect superimposes the thermoelastic effect, since the S0
Lamb wave mode typically shows higher frequencies than the A0 Lamb wave mode
and consequently is attenuated more [67].

4.3.1.4 Scattering

In addition, for wavelengths approaching the dimensions of included inhomogene-


ities, strong scattering of the elastic waves can occur. Typical examples are the
grain structure within metals, fibers in composites, or voids. This can also result in
severe attenuation and has to be taken into account when discussing attenuation
effects.

4.3.1.5 Dissipation into Adjacent Media

In experimental environments acoustic emission signals propagate within a mate-


rial under testing, which typically is in contact with adjacent media like, e.g.,
fixtures or supply pipes. Because a partial transmission of the acoustic emission
signal into adjacent media results in a respective loss of the reflected wave, such
losses can be fairly large and thus can even dominate the acoustic attenuation.

4.3.1.6 Modeling Attenuation

Moreover, attenuation in fiber reinforced structures also depends on the orientation


of the fibers in the laminates. To demonstrate this effect of acoustic anisotropy, Ono
and Gallego conducted measurements of the signal attenuation in AS4/3502 plates
(see Table B.1 of Appendix B) using piezoelectric pulsers (see [70] for details).
4.3 Wave Propagation 205

a
1.0
experimental data
modeling results
0.8
normalized amplitude

0.6

0.4

0.2

0.0
20 40 60 80 100 120 140
distance [mm]
b

1.0
experimental data
modeling results
0.8
normalized Amplitude

0.6

0.4

0.2

0.0
20 40 60 80 100 120 140
distance [mm]

Fig. 4.62 Comparison between normalized experimental attenuation measurements and simula-
tion result using complex elastic coefficients for the S0 mode (a) and the A0 mode (b)

Figure 4.62 shows a comparison of these measurements of a unidirectional plate to


respective results of modeling work using the complex elastic coefficients given in
Appendix B. As seen from Fig. 4.62 as a function of the distance to the source, the
detected amplitudes decrease significantly. In the model the total attenuation is due
to geometric spreading (result of using real part of stiffness tensor) but also due to a
contribution of thermoelastic dissipation (taken into account by the complex elastic
coefficients). Moreover, there is a difference in the attenuation of the two funda-
mental modes as seen by comparison of Fig. 4.62a, b.
206 4 Acoustic Emission

a b
unidirectional unidirectional
cross-ply cross-ply
90° 80° quasi-isotropic
90° 80° quasi-isotropic
70° 70°
60° 60°
50° 50°

40° 40°

30° 30°

20° 20°

10° 10°

0° 0°
attenuation attenuation

Fig. 4.63 Comparison between normalized attenuation levels of unidirectional, cross-ply, and
quasi-isotropic stacking for the S0 mode (a) and the A0 mode (b)

The specific challenges arising in arbitrary stackings of laminates are shown in


Fig. 4.63a, b. As seen in the comparison of the three different stackings in one polar
diagram, each layup causes different attenuation levels as a function of distance to
the source. Also, the ratio of attenuation between S0 mode and A0 mode is
influenced by the layup investigated. In particular for the S0 mode, the effect of
the change in in-plane stiffness (dominated by the fiber orientation) is clearly seen.
In general, these attenuation effects will cause difficulties to interpret the signals
after a certain distance of propagation, since the information included in the signals
is altered, and in polymeric materials, the higher-frequency components are atten-
uated faster due to the thermoelastic dissipation. This will render signals useless for
frequency interpretation after a certain distance and has to be included in predictive
tools to account for this effect. For the modeling part, the usage of complex elastic
coefficients according to the approach by [55, 64] is found to be a suitable approach
to account for this effect in FEM. However, care has to be used for those frequency
ranges, where strong changes to the complex elastic coefficients occur. As reported
by [64], the values of the complex stiffness tensor are usually a function of the
frequency of propagation and therefore have to be implemented as such in a
modeling environment if not constant within the investigated bandwidth.

4.3.2 Influence of Geometry

Similar as the propagation within an attenuating medium, a change of the boundary


conditions of the propagation medium can have a substantial impact on the acoustic
4.3 Wave Propagation 207

emission signal. For the case of fiber reinforced materials, this mostly refers to
changes in the specimen geometry, changes of the stacking sequence, and changes
to the material configuration.
In Fig. 4.64a, b, modeled signals for a fiber breakage source are shown as
calculated for a unidirectional T800/913 material using the material properties

a
initial part signal reflections

0.012
width = 10mm
0.010

0.008
displacement [nm]

0.006
width = 20mm
0.004

0.002

0.000

–0.002 width = 30mm

–0.004

–0.006
0 10 20 30 40 50
time [μs]
b
initial part signal reflections

0.012

0.010 thickness = 1mm

0.008
displacement [nm]

0.006
thickness = 2mm
0.004

0.002

0.000 thickness = 3mm

–0.002

0 10 20 30 40 50
time [μs]

Fig. 4.64 Influence of specimen width at 1 mm thickness (a) and specimen thickness at 10 mm
width (b) on detected signals of a fiber breakage source model in 30 mm distance
208 4 Acoustic Emission

from Table B.1 in Appendix B. The acoustic emission signals are evaluated as out-
of-plane displacement in 30 mm distance to the source position. The width of the
specimen and the thickness of the specimen are varied as noted in the Fig. 4.64a, b,
respectively. While changes of the specimen width mainly affect the presence and
intensity of the edge reflections, changes of the thickness apparently have larger
impact on the signal frequency content. Due to the nature of plate waves, their
frequency content is implicitly related to the thickness of the plate. This causes a
substantial change of the AE signal frequency, being the highest for the thinnest
specimen thickness.
Similar changes to the frequency content are observed, when the acoustic wave
passes a plate section with a gradual increase of the thickness. Such gradual changes
of the specimen thickness have similar impact on the signal frequency content as a
propagation medium with different thickness. However, additional attenuation
effects occur due to the change in propagation volume. Figure 4.65 presents a

signal detection point

z signal excitation
y
x
12

10

6
displacement [fm]

–2

–4 reference signal
–6 detected on wedge

0 20 40 60 80 100
time [μs]

Fig. 4.65 Model setup showing wedge profile and comparison of signals after propagating
100 mm distance with and without presence of a wedge to increase the plate thickness
4.3 Wave Propagation 209

initial part signal reflections

0.012

quasi-isotropic
0.010
displacement [nm]

0.008

0.006 cross-ply

0.004

0.002

0.000 unidirectional

–0.002

–0.004
0 10 20 30 40 50
time [μs]
Fig. 4.66 Influence of stacking sequence on detected signals of a fiber breakage source model in
30 mm distance

comparison of signals detected in a 1 mm thick T800/913 unidirectional fiber


reinforced polymer plate with properties of Table B.1 in Appendix B. One curve
shows the reference signal detected as out-of-plane displacement in 100 mm
distance to a dipole source buried in the plate. The other curve is the signal detected
in the same distance after passing through a wedge section increasing the thickness
to 2 mm using a wedge length of 10 mm.
As further parameter, the stacking sequence is varied, and its influence on the
signal emitted by the fiber breakage source model is shown in Fig. 4.66. Three
different configurations of identical plate thickness of 3 mm were chosen, compris-
ing a unidirectional, a cross-ply [0/90/90/0]sym, and a quasi-isotropic [90/-45/45/
0]sym stacking sequence. As seen in Sect. 4.3.1, these different stacking sequences
translate into different acoustic properties, which have a significant impact on the
visual appearance and frequency content of the AE signals, as well as on the arrival
time of the signals.
Further complications arise, when the material is not only anisotropic but also
inhomogeneous as for the case of metallic liners attached to the composite or hybrid
plates of fiber reinforced materials interspersed with metallic sheets. First, the
influence of the thickness ratio between layers of aluminum and a fiber reinforced
material is investigated as demonstrated in [53]. Using the results of aluminum and
T800/913 unidirectional CFRP with values as given in Table B.1 of Appendix B, a
combination of both materials is investigated as a two-layer system. The thickness
ratio between aluminum and CFRP is systematically varied while keeping the total
thickness of 1 mm. Simulated signals shown in Fig. 4.67 are for an AE dipole
210 4 Acoustic Emission

a 1.00 mm CFRP

1400 1.000E-03
A0 - 1mm (CFRP)
1200
0.005545
frequency [kHz]

1000
0.01091
800
600 0.05636

400 S0 - 1mm (CFRP)


0.1000
200
displacement [nm]

1.5
1.0
0.5
0.0
–0.5
–1.0
–1.5
0 25 50 75 100
time [μs]
b 0.50 mm Aluminum / 0.50 mm CFRP

1400 S0 - 1mm hybrid plate 1.000E-04

1200 5.545E-04
frequency [kHz]

1000 0.001091

800 A0 - 1mm hybrid plate


0.005636
600
guided wave 0.01182
400 in CFRP layer
0.05000
200
0
displacement [nm]

0.6
0.3
0.0
–0.3
–0.6

0 25 50 75 100
time [μs]

Fig. 4.67 Simulated signals of AE dipole source in two-layer system with varying thickness ratio
between aluminum and T800/913 unidirectional CFRP obtained at 100 mm distance for 90
propagation angle. (a–c) show Choi-Williams distribution (logarithmic scale) of selected config-
urations with superimposed dispersion curves; (d) shows comparison of all configurations (based
on [53])
4.3 Wave Propagation 211

c 1.00 mm Aluminum
1400 1.000E-04

1200
S0 - 1mm (Alu)
frequency [kHz]

6.000E-04
1000
800 0.002000

600 A0 - 1mm (Alu)


0.007000
400
0.01000
200
0
displacement [nm]

0.2

0.0

–0.2
0 25 50 75 100
time [μs]
d
4

3
displacement [nm]

2 Al 0.0 mm CFRP 1.0 mm

1 Al 0.5 mm CFRP 0.5 mm

0 Al 1.0 mm CFRP 0.0 mm

0 20 40 60 80 100 120 140


time [μs]
Fig. 4.67 (continued)

source buried in the CFRP part with axis orientation along the y-axis that is detected
at 100 mm distance under 90 propagation angle on the top surface (composite side)
of the plate. To identify particular Lamb wave modes, the CWDs shown in Fig. 4.67
are superimposed by results of dispersion curve calculations following
reference [71].
212 4 Acoustic Emission

The separation in time of the S0 mode and A0 mode is observed strongest in


homogeneous CFRP and weakest in homogenous aluminum. The modes of the
signal shown in Fig. 4.67b are identified as symmetric and antisymmetric motion
relative to the midplane of the plate. However, there is no match between the
detected Lamb wave modes and the dispersion curve calculations for the individual
materials. Therefore, the Lamb wave propagation seems to occur as a motion of the
complete plate for this case.
For varying ratio between aluminum and CFRP, a systematic transition of signal
shape and propagation velocities is found. Similar to the S0 mode, the signal shape
of the A0 mode is changed continuously between the two extreme values of only
CFRP and only aluminum, while the amplitude is subject to a nonlinear transition.
There is significant influence on the propagation velocity of the S0 mode and the A0
mode when the thickness ratio between aluminum and CFRP is changed. As
demonstrated in [53], the agreement between the simulated velocities and those
obtained from the rule of mixture of the velocities of both materials indicates that
the velocity of Lamb wave propagation of the S0 mode is governed by the average
elastic properties of the materials involved. This is of particular interest for AE
source localization approaches that are based on the initial threshold crossing. A
similar transition of propagation velocities is found for the A0 mode. However, a
direct application of the rule of mixture does not capture the changes in propagation
velocity. This is due to the correlation of the A0 mode velocity with the overall
bending stiffness of the plate, which cannot be obtained from a direct rule of
mixture calculation of the individual sound velocities of the materials involved.
Also, the ratio between anisotropic and isotropic material seems to be the key
factor for the changes in signal propagation. Thus for thickness ratios as typically
faced for thin liner materials (<0.125 of the total thickness), the influence of the
liner on signal propagation can assumed to be minimalistic.
For an increased total thickness of the plates, more than the two fundamental
Lamb wave modes can be identified in the CWD. For a case with 2.00 mm
thickness, two propagation modes are found at the beginning of the signal. These
originate from a S0 mode Lamb wave of the hybrid plate and an additional guided
wave propagating in the CFRP layer [53].
This effect is seen best in the signals detected at 0 propagation angle, as shown
in Fig. 4.68. The comparison of the signal detected on the aluminum side
(Fig. 4.68a) and detected on the composite side (Fig. 4.68b) demonstrates the
distinct difference of the signals. While both signals exhibit the same intensity of
A0 mode and S1 mode, the additional propagation of the S0 mode is seen only for the
detection on the composite side.
This demonstrates that Lamb wave propagation in such multilayered plates can
take place as motion of the complete plate, but also as a guided wave in only one of
the layers.
4.3 Wave Propagation 213

a 1.0 mm Aluminum / 1.0 mm CFRP


1000
5.000E-05

800
2.973E-04
frequency [kHz]

S1 - 2mm hybrid plate


600
9.451E-04

400 S0 - 2mm hybrid plate 0.003418


A0 - 2mm hybrid plate
200 0.009945

0
displacement [nm]

0.03
0.02
0.01
0.00
–0.01
–0.02 Aluminum side
–0.03
0 25 50 75 100
time [μs]

b 1.00 mm Aluminum / 1.00 mm CFRP


1000
5.000E-05

800 3.247E-04
frequency [kHz]

S1 - 2mm hybrid plate 0.001094


600
0.004116

400 S0 - 2mm hybrid plate


0.01802

200 0.05000
A0 - 2mm hybrid plate
0
displacement [nm]

0.03
0.02
0.01
0.00
–0.01
–0.02 Composite side
–0.03
0 25 50 75 100
time [μs]

Fig. 4.68 Choi-Williams distribution (logarithmic scale) of simulated signals of AE dipole source
in a two-layer system with total thickness of 2.00 mm obtained at 100 mm distance for 0 prop-
agation angle on aluminum side (a) and composite side (b) (based on [53])
214 4 Acoustic Emission

4.3.3 External and Internal Obstacles

So far the acoustic wave propagation was discussed for undisturbed media, which is
certainly not the case for a loaded composite structure with a continuously increas-
ing amount of damage. Therefore, the influence of pre-damaged structures on wave
propagation has to be considered for valid interpretation of acoustic emission
signals from composites. Due to the controlled situation, the previously validated
finite element method is further applied to account for the presence of several types
of external and internal obstacles. Within this section only aspects of wave prop-
agation are discussed. The influence on source localization methods will be
presented in Sect. 4.6 and the impact on source identification in Sect. 4.5.
The model setup used in the following is shown in Fig. 4.69. A rectangular plate
of 200 mm  200 mm size and 1 mm thickness is used as propagation medium. For
evaluation of the signals, a unidirectional CFRP plate was selected as propagation
medium, since this exhibits a maximal elastic anisotropy.
The elastic properties of a unidirectional T800/913 CFRP plate are used as given
in Table B.1 of Appendix B. The fiber direction of the unidirectional laminate is
oriented along the 0 axis as noted in Fig. 4.69. Two symmetry planes were chosen
to reduce the volume modeled to one quarter of the overall volume. The symmetry
planes are the yz- and xz-plane with respect to the origin of the coordinate system.
As acoustic emission source, a point couple was defined following [61, 62]. The
position of the source was at (x, y, z) ¼ (0.10, 0.00, 0.53) mm, slightly asymmetric
with respect to the midplane of the plate, to excite a reasonable amount of S0 and A0
Lamb wave modes at the same time. A linear ramp function with excitation time
te ¼ 1 μs and maximum force Fmax ¼ 3 N is used:

CFRP (unidirectional 0°)


mm 20
200 0m points for signal detection
m 0°: (x,y,z) = (100,0,1) mm
90° 0° 45°: (x,y,z) = (71,71,1) mm
90°: (x,y,z) = (0,100,1) mm
yz
-sy ne
mm pla
try
etr me
yp sym
lan xz-
e
model source

1.0 mm
z damaged area (e.g. crack)
y origin (x,y,z) = (0,0,0) mm
x

Fig. 4.69 Three-dimensional model setup used for simulation of Lamb wave propagation includ-
ing details for signal detection points and position of obstacle (e.g., crack position)
4.3 Wave Propagation 215

Fig. 4.70 Three-


dimensional model
geometry used for
simulation of bolts and
holes

width: 25 μm width: 25 μm width: 25 μm


z
y x

mm
20 mm
10 m
5m
50 mm distance to AE source

Fig. 4.71 Three-dimensional model geometry used for simulation of inter-fiber cracks (cut
parallel to fiber axis direction) (reprinted from [62])

Fmax  ðt=te Þ t  te
FðtÞ ¼ ð4:42Þ
Fmax t > te

The obstacle was placed with midpoint at a distance of 50 mm in x-direction to the


source. In the following, the influence of the presence of rivets, bolts, holes, inter-
fiber cracks, broken fibers, and delamination is discussed. Each of these obstacles
refers to one prototype of discontinuity. Rivets and bolts are through-thickness
metallic enclosures that allow direct transmission of elastic waves, but are expected
to cause wave scattering due to their acoustic impedance mismatch. All signals are
evaluated at 100 mm distance to the AE source in 0 , 45 , and 90 propagation
direction as z-displacement of the plate. For further details of this study, the reader
is referred to [61, 62].
Details of the geometries chosen for rivets and holes are given in Fig. 4.70. All
metallic parts marked in red are modeled as high-strength alloy steel parts, with
elastic properties as listed in Table B.1 in Appendix B.
Inter-fiber cracks (also called transverse cracks) weaken the link between neigh-
boring fibers and can thus affect the transmission of acoustic waves. In the chosen
configuration, the growth direction of such cracks is along the fiber axis direction
(0 ). From the perspective of wave propagation, the worst case is a growth of such
cracks through the complete thickness. This configuration is studied here as cut of
25 μm width ranging from the top to the bottom of the plate and various lengths as
shown in Fig. 4.71.
As further prototype, rupture of fibers can accumulate locally and can become a
macroscopic crack. This is modeled as a 100 μm thick cut perpendicular to the fiber
axis direction as shown in Fig. 4.72. Since such crack growth typically initiates
ultimate failure of the load-carrying structure, three depths of the cut were
216 4 Acoustic Emission

20 20 20
m m m
m m m

z
y x

0.500 mm 0.250 mm 0.125 mm


50 mm distance to AE source

Fig. 4.72 Three-dimensional model geometry used for simulation of broken fibers (cut perpen-
dicular to fiber axis direction) (reprinted from [62])

z
y 20
x 20 m
20 m
m
m m
m
height: height:
50 μm 50 μm mm height:
m m m 20 50 μm
5m 10

50 mm distance to AE source

Fig. 4.73 Three-dimensional model geometry used for simulation of inter-ply delamination (cut
parallel to fiber axis direction) (reprinted from [62])

investigated to study the influence on wave propagation. The depth of 0.5 mm


already refers to a significant amount of damage, reflecting a rupture of more than
800,000 fiber filaments. Also, for bending loads, the initiation of these cracks is
typically found on the surface of the structure. Therefore, the current geometry of
broken fibers starting at the surface exhibits an asymmetric location of the crack
with respect to the midplane.
Finally, inter-ply delamination is an in-plane discontinuity that weakens the
local bending stiffness of the plate. This is often encountered in fiber reinforced
structures as a consequence of impact damage, excessive loading, or even as residue
of the manufacturing process. Since this can occur in various extents, three different
sizes were investigated to reflect reasonable variability of the size of the
delaminated area. For all configurations a thickness of the delamination of 50 μm
is chosen centered within the plate (Fig. 4.73).

4.3.3.1 Reference Case

In the following the influence of the modeled obstacles is demonstrated. Figure 4.74
shows the CWD of the calculation result for signal propagation along the 0 axis of
the unidirectional CFRP plate. The CWD in Fig. 4.74a uses a different color,
4.3 Wave Propagation 217

a 2000
1800 CFRP Reference S0-mode (100mm) 0.000

1600 S0-mode (300mm) 0.004000


Frequency [kHz]

1400
0.008000
1200
1000 0.01200
800
0.01600
600
400 0.02000

200 edge reflection


0
Displacement [fm]

10
5
0
–5
–10
0.00 0.01 0.02 0.03 0.04 0.05
Time [ms]

b 1000
CFRP Reference 0.000

800 4.000
A0-mode (100mm)
Frequency [kHz]

S0-mode (100mm) 8.000


600
12.00

400
16.00

200 20.00

0
Displacement [nm]

0.10
0.05
0.00
–0.05
–0.10
0.000 0.025 0.050 0.075 0.100
Time [ms]

Fig. 4.74 Simulation results of signal propagation at 100 mm distance along 0 direction in
unidirectional CFRP. Truncated time scale shows S0 mode (a) and full time scale shows A0 mode
(b) (reprinted from [62])

frequency, and time range as Fig. 4.74b, which is necessary to visualize the S0 mode
at the initial part of the signal. To identify particular Lamb wave modes, the
dispersion curves of the fundamental modes were calculated.
As indicated by the superimposed dispersion curves for the S0 mode at 100 and
300 mm source distance, the S0 mode is reflected at the edge of the plate in
218 4 Acoustic Emission

Fig. 4.75 Comparison of simulated wave fields z-displacement for reference case at t ¼ 20 μs (a)
and at t ¼ 40 μs (b) after signal excitation (reprinted from [62])

x-direction and is detected twice. The shape of the A0 mode shown in Fig. 4.74b
agrees well with the calculated result of the dispersion curve for 100 mm distance.
The small dips in the curve seen at 52 and 62 μs are caused by multiple reflections
of the S0 mode, which has been reflected at all edges of the plate at this time. The
calculation results for the CFRP plate without defects in Fig. 4.74 will act as a
reference case to evaluate the influence of any obstacles modeled within the
propagation path.
The wave fields obtained for the reference case without obstacles are shown in
Fig. 4.75a, b for two distinct times t ¼ 20 μs and t ¼ 40 μs after signal excitation.
Effects of edge reflections and modal conversions are easy to spot in these stills of
the wave field.

4.3.3.2 Rivets and Bolts

Bolts and rivets are commonly used as fasteners in CFRP structures and are
frequently encountered during testing of real structures. The calculation result for
signal propagation in the presence of such obstacles is shown in Fig. 4.76 with time,
frequency, and coefficient range settings slightly different to Fig. 4.74. As observed
from Fig. 4.76a, shape and intensity of the S0 mode are influenced by the presence
of the rivet within the propagation path. The intensity is less than for the reference
structure. On the one hand, this is caused by the acoustic transmission coefficient
(0.75) of the CFRP-metal and metal-CFRP interface in 0 direction calculated
under the assumption of planar wave incidence. On the other hand, interfaces are
likely to cause modal conversion. For the current case, a significant part of the
incident S0 mode is converted to an A0 mode radiating away from the rivet (cf. also
Fig. 4.77). This causes an additional emission of an A0 mode that is ahead of the A0
mode originally emitted by the point source. Thus, determination of the correct
arrival time of the A0 mode with respect to the point source is almost impossible.
The presence of the rivet causes a strong interaction with the incident Lamb
waves. As seen in Fig. 4.77a, the interaction of the rivet with the S0 mode causes
excitation of a strong secondary A0 mode with almost circular radiation pattern
around the center of the rivet. The only explanation for this is a modal conversion of
4.3 Wave Propagation 219

a
2000
1800 Rivet 0.000

1600 Second emission of S0-mode 0.04000


Frequency [kHz]

1400 Initial S0-mode Emitted A0-mode 0.08000


1200
1000 0.1200
800
0.1600
600
400 0.2000

200
0
Displacement [fm]

20
10
0
–10
–20
0.00 0.01 0.02 0.03 0.04 0.05
Time [ms]

b 1000
Rivet 0.000

800 Initial A0-mode 2.000


Frequency [kHz]

4.000
600 Emitted A0-mode
6.000
400
8.000

200 10.00

0
Displacement [nm]

0.10
0.05
0.00
–0.05
–0.10
0.000 0.025 0.050 0.075 0.100
Time [ms]

Fig. 4.76 Simulation results of signal propagation at 100 mm distance along 0 direction in
unidirectional CFRP with included rivet. Truncated time scale shows S0 mode (a) and full time
scale shows A0 mode (b)

the primary S0 mode to a secondary A0 mode, since the primary A0 mode did not
arrive at the rivet for t < 20 μs. In Fig. 4.77b, the wave field for t ¼ 40 μs shows the
interaction of the A0 mode with the rivet. Similar as for the incident S0 mode, the A0
mode is partially reflected at the rivet.
220 4 Acoustic Emission

Fig. 4.77 Comparison of simulated wave fields z-displacement for rivet obstacle at t ¼ 20 μs (a)
and at t ¼ 40 μs (b) after signal excitation (reprinted from [61])

4.3.3.3 Holes

Holes in CFRP are another type of acoustic obstacle often encountered in technical
CFRP structures. These can originate from drilling processes or represent seriously
damaged areas, e.g., after high-velocity impacts. As shown in the CWDs in
Fig. 4.78 with settings identical to Fig. 4.74, the influence of this kind of obstacle
is more severe than the influence of the rivet. As seen in Fig. 4.78a, the presence of
the through-thickness hole causes less-intense transmission of the initial S0 mode as
compared to the reference case. Also, the shape of the hole results in a second
emission of a S0 mode with a virtual source position located at the center of the hole.
This process is observed in more detail in the wave fields of Fig. 4.79a, b. As soon
as the wavelength of the A0 mode becomes sufficiently larger than the diameter of
the hole, there is no interaction with the hole. Thus, only the short wavelengths of
the A0 mode (i.e., the initial part) are affected by the presence of the hole and are
scattered at the obstacle.

4.3.3.4 Inter-fiber Cracks

Inter-fiber cracks are one of the basic damage types found in fiber reinforced
composites. These are cracks with propagation direction perpendicular to the
fiber axis direction. Since the strength of the direction perpendicular to the fiber
axis is the weak link between the fibers, this type of damage is observed frequently
in composite structures. In the current configuration, the inter-fiber crack is
modeled as crack-through process, i.e., the crack reaches from the top to the bottom
of the plate. The length of the inter-fiber crack is varied between 5 and 20 mm to
cover a broad range of macroscopic crack sizes.
As seen in Fig. 4.80, the S0 mode detected in these damaged structures has only
negligible differences when compared to the reference case (black line). But there is
a detectable influence of the presence of the inter-fiber cracks on the initial part of
the A0 mode. For the various lengths of 5, 10, and 20 mm of the inter-fiber crack, no
4.3 Wave Propagation 221

a
2000
1800 Hole 0.000

1600 Second emission 0.004000


Frequency [kHz]

1400
0.008000
1200 edge reflection
1000 0.01200
800
0.01600
600
400 0.02000

200
0
Displacement [fm]

10
5
0
–5
–10
0.00 0.01 0.02 0.03 0.04 0.05
Time [ms]

b
1000
Hole 0.000

800 4.000
Frequency [kHz]

8.000
600
12.00
400
16.00

200 20.00

0
Displacement [nm]

0.10
0.05
0.00
–0.05
–0.10
0.000 0.025 0.050 0.075 0.100
Time [ms]

Fig. 4.78 Simulation results of signal propagation at 100 mm distance along 0 direction in
unidirectional CFRP with hole. Truncated time scale shows S0 mode (a) and full time scale shows
A0 mode (b) excitation (reprinted from [61])

significant differences are observed relative to each other besides a small signature
at 63 μs.
For the inter-fiber cracks, only weak interaction with the S0 mode and A0 mode is
found as seen in Fig. 4.81a, b.
222 4 Acoustic Emission

Fig. 4.79 Comparison of simulated wave fields z-displacement for hole obstacle at t ¼ 20 μs (a)
and at t ¼ 40 μs (b) after signal excitation (reprinted from [61])

0.35
Reference
5mm
0.30 10mm
20mm
0.25
Displacement [nm]

0.20

0.15

0.10

0.05

0.00

–0.05

0 20 40 60 80 100
Time [μs]

Fig. 4.80 Simulation results of signal after propagation of 100 mm along 0 direction in
unidirectional CFRP with inter-fiber cracks of various lengths (reprinted from [62])

Fig. 4.81 Comparison of simulated wave fields z-displacement for inter-fiber crack at t ¼ 20 μs
(a) and at t ¼ 40 μs (b) after signal excitation (reprinted from [62])
4.3 Wave Propagation 223

4.3.3.5 Broken Fibers

Among the damage types found in fiber reinforced composites, a rupture of the
load-bearing constituents is the most severe damage. In the model, the presence of
such discontinuous fiber filaments is taken into account by an air gap with extension
perpendicular to the fiber axis direction (cf. Fig. 4.72). To consider the various
degrees of damage for this type of failure, the depth of the air gap in the plate
thickness direction is varied between 0.125 and 0.500 mm.
As seen in Fig. 4.82, the influence of the discontinuity on the first arrival of the S0
mode (t < 20 μs) is negligible for all cases investigated. Dependent on the through-
thickness dimension of the modeled discontinuity, a secondary peak arises around
35 μs. This is caused by modal conversion of the incident primary S0 mode at the
discontinuity into a secondary A0 mode and into a secondary S0 mode (see
Fig. 4.83a, b). The later parts of the signal (t > 40 μs) are also affected by this

0.50
Reference
0.45 0.125mm
0.250mm
0.40 0.500mm
0.35
Displacement [nm]

0.30
0.25
0.20
0.15
0.10
0.05
0.00
–0.05

0 20 40 60 80 100
Time [μs]

Fig. 4.82 Simulation results of signal after propagation of 100 mm along 0 direction in
unidirectional CFRP with broken fibers modeled with various dimensions (reprinted from [62])

Fig. 4.83 Comparison of simulated wave fields z-displacement for fiber breakage at t ¼ 20 μs (a)
and at t ¼ 40 μs (b) after signal excitation (reprinted from [62])
224 4 Acoustic Emission

interaction, which results in a superposition of the incident A0 mode with the


multiple reflections of the secondary modes. It is worth noting that the model is
energy conservative. Still, the calculated amplitude of the signals is distinctly
different. This is solely caused by the presence of the broken fibers, which are
responsible for a preferential distortion of the displacement field resulting in a
transfer of energy from the in-plane displacement to the out-of-plane displacement.

4.3.3.6 Delamination

Inter-ply delamination is one of the most common types of damage found in fiber
reinforced composites, since it is often caused by local impact or as residue of a
manufacturing error. During mechanical testing of fiber reinforced structures,
delamination can evolve step by step and can affect the elastic properties along
the signal propagation path. Thus, the influence of delamination on the detected AE
signals is of considerable interest. To resemble the variety of dimensions of inter-
ply delamination, the size of the delaminated area in the in-plane direction was
varied between 5 and 20 mm (cf. Fig. 4.73).
Figure 4.84 shows the calculated signals for all three inter-ply delamination sizes
investigated and the reference signal for comparison. For the S0 mode, there are
only negligible differences observed to the reference case. Since propagation of the
S0 mode is dominated by the in-plane stiffness, this is explained by considering a
multilayer specimen composed of 950 μm CFRP and 50 μm air. The in-plane
stiffness of such a plate is almost 95 % of the stiffness of a 1000 μm CFRP plate.

0.40
Reference
0.35 5mm
10mm
0.30 20mm

0.25
Displacement [nm]

0.20

0.15

0.10

0.05

0.00

–0.05

0 20 40 60 80 100
Time [μs]

Fig. 4.84 Simulation results of signal after propagation of 100 mm along 0 direction in
unidirectional CFRP with inter-ply delamination of various dimensions (reprinted from [62])
4.3 Wave Propagation 225

Fig. 4.85 Comparison of simulated wave fields z-displacement for delamination at t ¼ 20 μs (a)
and at t ¼ 40 μs (b) after signal excitation (reprinted from [62])

A larger influence is observed for the propagation behavior of the A0 mode. The
shape and intensity of the A0 modes differ from the reference signal due to the
change of local bending stiffness as introduced by the inter-ply delamination. With
increasing size of the inter-ply delamination, the deviation compared to the refer-
ence case increases as well.
As seen in the wave field in Fig. 4.85a, b, for the delamination case, only weak
interaction with the S0 mode and the A0 mode is found.

4.3.3.7 Influence on Signal Arrival Times

As pointed out in the previous sections, obstacles like rivets, holes, or inclusions as
well as the occurrence of damage can have severe impact on the transmittance of
acoustic waves. Since damage may evolve during loading of the material, their
sudden presence may alter the transmission path between an AE source and the AE
sensor. Therefore, the acoustic emission signals calculated for the damage config-
urations are also evaluated with respect to the arrival time of signal.
In Fig. 4.86, the extracted initial arrival times are shown for all damage config-
urations (inter-fiber failure, inter-ply delamination, and broken fibers) and the
reference case in 0 , 45 , and 90 propagation direction. The largest influence of
the internal damage types is found for the 90 propagation direction, which dem-
onstrates that the complete wave field is affected by the presence of the damage and
not solely the part of the wave propagating through the obstacle (0 direction).
Based on the maximum deviation of 4.2 μs to the reference case of the calculated
arrival time of the S0 mode in 90 propagation direction, a corresponding error of
localization in the range of several centimeters can be expected. The estimation is
based on the calculated group velocity of the S0 mode in 90 propagation direction.
A significantly larger influence was found for the arrival time of the A0 mode (not
shown in Fig. 4.86). Due to modal conversion occurring at the internal damage, the
arrival of the first detectable A0 mode is 48.6 μs ahead compared to the reference
case in one configuration. However, this is not the arrival of the primary A0 mode
as discussed before. For the majority of the cases studied, the arrival times of
the A0 mode are within 10 μs difference to the reference case. Consequently,
226 4 Acoustic Emission

40

35 90°

30
S0 arrival time [μs]

25 45°

20

15 Reference case
Inter-Fiber cracks
0° Broken Fibers
10
Inter-ply Delamination

5
0 5 10 15 20 25 30
configuration

Fig. 4.86 Calculated arrival time of S0 mode for the various configurations of internal obstacles
and internal damage in comparison to the reference case for 0 , 45 , and 90 propagation direction
(reprinted from [62])

for localization methods using arrival times of both fundamental Lamb wave
modes, the observed difference in arrival time of the A0 mode might have a large
impact as well.

4.4 Detection of Acoustic Emission Signals

The detection of an acoustic emission signal occurs at the surface of a solid. The
most common type of measurement uses the sensitivity of piezoelectric materials to
detect the wave amplitudes down to the order of 1013–1014 m [42, 43]. In many
cases this piezoelectric element is included within a protective housing and
connected to the surface by a viscous coupling medium [72]. There is a large
variety of acoustic emission sensors available today, which are almost completely
based on piezoelectric detector materials, such as lead zirconate titanate Pb
[ZrxTi1  x]O3 (PZT) or lead metaniobate PbNb2O6 (PM) [73]. As technical alter-
native to piezoelectric materials, various other detection methods were proposed.
While mostly capacitance sensors were used in the past decades for broadband
applications [74, 75], recently other methods using fiber Bragg gratings (resonant)
[76–79] or laser interferometers (broadband) have demonstrated their feasibility as
approaches to detect AE [80, 81].
4.4 Detection of Acoustic Emission Signals 227

The main purpose of the acoustic emission sensor is the conversion of surface
motion of the solid to an electric signal. For the piezoelectric sensors, this conversion
is based on the piezoelectric effect which is directly related to the occurrence of
electric dipole moments within piezoelectric materials. For the case of PZT, these
dipole moments result from the asymmetric charge density located within the oxygen
octahedron. The magnitude of the dipole moment along the axis of the octahedral is
directly proportional to the applied mechanical stress along that direction.
The fundamental equation for the electrical displacement D of a material with
electrical permittivity ξ within an electric field with strength E is given by

D ¼ ξ  E: ð4:43Þ

In combination with Hooke’s law, this yields the coupled equations of the piezo-
electric effect in stress-charge form:

D¼SεþξE ð4:44Þ
σ ¼CεS E t
ð4:45Þ

The coupling matrix is given by the direct S and inverse St piezoelectric constants,
which can be written in the stress-charge form for a C4v-crystal class as for the case
of PZT using the Voigt notation as 0 1
ε1
0 1 0 1B ε2 C
D1 0 0 0 0 S15 0 B C B
@ D2 A ¼ @ 0 ε3 C
0 0 S24 0 0 AB B ε4 C
C
D3 S31 S32 S33 0 0 0 @ C B
ε5 A
ε6
0 10 1
ξ11 0 0 E1
þ @ 0 ξ22 0 A@ E2 A ð4:46Þ
0 0 ξ33 E3
0 1 0 10 1
σ1 C11 C12 C13 0 0 0 ε1
B σ 2 C B C12 C22 C23 0 0 0 C B ε2 C
B C B CB C
B σ 3 C B C13 C23 0 C B C
B C¼B C33 0 0 CB ε3 C
B σ4 C B 0 C
0 CB C
B C B 0 0 C44 0 B ε4 C
@ σ5 A @ 0 0 0 0 C55 0 A @ ε5 A
σ6 0 0 0 0 0 C66 ε6
0 1
0 0 S31
B 0 0 S32 C 0 1
B C E1
B 0 0 S33 C
BB 0 S24
C@ E 2 A ð4:47Þ
B 0 CC E3
@ S15 0 0 A
0 0 0
228 4 Acoustic Emission

Thus the lattice deformation of a piezoelectric solid results in an electric field,


which is called the direct piezoelectric effect and is typically used in sensor
applications. Complementary, the application of an electrical field on a piezoelec-
tric material causes a mechanical displacement known as inverse piezoelectric
effect which is instead often used in actuator elements.
Although such piezoelectric sensors are very sensitive to detection of surface
displacements, they show intrinsically resonant behavior. This resonant frequency
response is caused by the piezoelectric material properties and the finite dimensions
of the sensing element. For the frequently used geometrical shape of a disk, the
possible resonances are for movements along the thickness direction, movements
along the radial direction, as well as due to torsional movement. These effects
superimpose the frequency characteristic of the detected elastic wave. Conse-
quently, for interpretation of the frequency distribution of the acoustic emission
signals, the effect of the sensor has to be taken into account, which is discussed in
more detail in Sect. 4.4.1.
A related concept dealing with such characteristic system responses is the
convolution theory. The output function hA(t) of a system is given by the convo-
lution integral of input function hE(t) and the transfer function Ξ(t):
ð þ1
hA ðtÞ ¼ hE ðtÞ  Ξðt0  tÞdt0 : ð4:48Þ
1

The transfer function itself is defined as the response of the system to a δ-function in
the time domain as
ð þ1
ΞðtÞ ¼ hE ðtÞ  δðt0  tÞdt0 : ð4:49Þ
1

For most considerations, (4.48) is transferred to the frequency domain by Fourier or


Laplace transformation using the complex frequency ω*:
     *
e
h A ω* ¼ e e ω
h E ω*  Ξ ð4:50Þ

Thus in the frequency domain, the system output is simply given by multiplication
of the input function by the transfer function. Moreover, transfer functions of
concatenated systems can be written as product of each singular transfer function.
With respect to the case of acoustic emission detection, this can be expressed
according to [2, 52, 82] as
           
u AE ω* ¼ des ω*  Ξ
e e med ω*  Ξ
e geo ω*  Ξ
e sen ω*  Ξ
e amp ω* ð4:51Þ
 
This includes the acoustic emission source-time function des ω* in the frequency
 *  
domain, the influence of the propagation medium Ξ e med ω , its geometry Ξe geo ω* ,
 
the sensors response Ξ e sen ω* , and the characteristic of the amplifier Ξamp(ω*) of
the recording equipment. Figure 4.87 shows exemplary frequency spectra of these
transfer functions illustrating their influence on the final acoustic emission signal.
6E-18 1E-15
5.5E-18
5E-18
source material and sensor
geometry
4.5E-18

4E-18 1E-16 1000

3.5E-18

3E-18

FFT-Magnitude
FFT-Magnitude
1E-17

FFT-Magnitude
2.5E-18

2E-18 100
0 200 400 600 800 1000 1200 1400 1600 1800 2000 0 200 400 600 800 1000 1200 1400 1600 1800 2000 0 200 400 600 800 1000 1200 1400 1600 1800 2000
Frequency [kHz] 1E-12 Frequency [kHz] Frequency [kHz]

1E-13
0.1
amplifier recorded acoustic emission signal
1E-14

FFT-Magnitude
FFT-Magnitude
0.01

1E-15

1E-3
0 200 400 600 800 1000 1200 1400 1600 1800 2000 0 200 400 600 800 1000 1200 1400 1600 1800 2000
Frequency [kHz] Frequency [kHz]

Fig. 4.87 Contributions of the different transfer functions of source, material and geometry, sensor and acoustic emission equipment obtained from
FEM-simulations to the final acoustic emission signals, visualized following Bohse [82] (figure reprinted from [35])
230 4 Acoustic Emission

As already pointed out by Giordano et al. [2] and Wilcox et al. [10], it is possible
to evaluate each transfer function of (4.51) independently. Consequently, an appli-
cation of the deconvolution analysis according to [27] to (4.51) can deduce the
 
source function des ω* from the recorded signal e u AE (ω*) if all other transfer
functions are known.
 
Usually the influence of the propagation medium Ξ e med ω* and its geometry
 
e geo ω* cannot be separated sufficiently. Moreover their exact appearance is also
Ξ
correlated to the propagation path of the acoustic wave. Therefore analytical
solutions for these two contributions are usually approximated by idealistic
assumptions on the medium influence and assumptions with respect to the real
geometry, typically applying infinite extensions along one or more axis [21, 22,
29, 30].
In contrast, the estimation of the transfer function of the sensor and the amplifier
is an experimentally well-established procedure. While the characteristic of the
amplifier can easily be obtained by a network analyzer, the measurement of the
transfer function of acoustic emission sensors, also known as primary absolute
calibration, is more difficult. Several experimental methods are thus established for
primary absolute calibration of acoustic emission sensors. The most common are
the step-function calibration according to ASTM E 1106 and the reciprocity
calibration according to NDIS 2109. Originally, Breckenridge et al. applied a
capacitive sensor for primary absolute calibration of piezoelectric sensors [83]. In
recent years laser vibrometers are frequently applied to the same purpose [84, 85].
As pointed out by Goujon et al., there is no expectation for an absolute transfer
function of the sensor, since the frequency sensitivity characteristic of the sensor
also depends on the elastic properties of the medium it is applied on [85]. This
effect is caused by the mismatch of acoustic impedance between sensor and
medium and is discussed in Sect. 4.4.2.3.

4.4.1 Comparison of Sensor Types

Despite of the common usage of piezoelectric materials as sensing element, there


are many different acoustic emission sensor designs available. The selection of the
appropriate sensor type follows obvious aspects like operation temperature, but also
more complicated parameters like bandwidth and flatness of frequency response.
For many applications it is only relevant to detect acoustic emission signals in
general, so the frequency characteristics might not be that relevant. Other
approaches use the frequency information of the detected signal to perform source
identification or modal analysis, which requires sensors with a certain bandwidth.
Also, a flat sensor response allows better detection of signal arrival at a certain
frequency. This leads to the type of sensor classification into resonant, multi-
resonant, and wideband sensor types. Some examples of commercially available
4.4 Detection of Acoustic Emission Signals 231

Resonant Multi-Resonant Wideband

Micro30 WD 1045S 1025T KRNBB-PC


(Mistras) (Mistras) (Fujicera) (Digital Wave) (KRN)

Diameter 10 mm Diameter 17.8 mm Diameter 20 mm Diameter 9.3 mm


Height 12 mm Height 16.5 mm Height 20 mm Height 13.3 mm
Resonance 225 kHz Main resonance 450 kHz
Range 150-400 kHz Range 125–1000 kHz Range 200–1300 kHz Range 70 – 1000 kHz

Fig. 4.88 Selection of different acoustic emission sensor types including specifications according
to manufacturer’s data sheets

sensor types are shown in Fig. 4.88 including their most relevant specifications
according to the manufacturer’s data sheets.3
To compare the response of the different sensors, an aluminum plate of
1500 mm  1000 mm  3 mm is chosen with the sensor positioned at the center
of the plate. Pencil lead breaks with 0.5 mm diameter and 2H hardness were
performed at a distance of 250 mm to the sensor midpoint. Pencil lead breaks
were repeated 20 times, and AE signals were detected using a preamplifier with
10 kΩ input impedance and 15 pF input capacitance operating at fixed gain. The
mean frequency spectrum was calculated from the 20 detected signals for compar-
ison of sensor response. As seen for the representative data in the time domain, the
choice of sensor type has significant impact on the amplitude and shape of the
signal. The different sensitivity is owed to the fact that vastly different geometries
of sensing elements are used in the type of sensors studied. Clearly seen in the
frequency spectra in Fig. 4.89b are the resonances of the Micro30 sensor at around
250 kHz. Here the sensing element is close to its (damped) Eigen resonance and
therefore allows for preferential conversion of the incident wave into an electric
signal.
However, such resonances limit the effective bandwidth of the sensor and
usually cause strong drops of sensitivity outside the resonance range. Improvements
of this are given by multi-resonant sensors such as the WD sensor. Using a
combination of two sensing elements (cf. [24, 35]), this allows for signal detection
with a higher bandwidth, since resonances at 300 and 600 kHz are intrinsic to this
geometry. For those sensor types referred to as wideband sensors, a good match to the
theoretical frequency content of the pencil lead break source is found. While the
1045S and the 1025T still show some minor resonances, the KRNBB-PC seems to

3
According to ASTM E976 calibration standard [V/μbar]
232 4 Acoustic Emission

a
20

18

16

14
amplitude [V]

12

10

0
0 25 50 75 100
time [μs]
b
10000 WD (PAC)
Micro30 (PAC)
KRNBB-PC (KRN)
1045S (Fujicera)
1000
1025T (DigitalWave)

100
FFT-Magnitude

10

0.1

0 250 500 750 1000 1250 1500


frequency [kHz]

Fig. 4.89 Comparison of detected PLB signals in 250 mm distance in the time domain (a) and the
respective frequency spectrum averaged over 20 signals (b)

match best with the theoretically expected smooth falloff with frequency. The latter
sensor type uses a conical sensing element which was found to yield very flat response
in the frequency range of interest for acoustic emission analysis [42, 86–89]. However,
as seen in Fig. 4.89b within the range of the resonances of other sensor types,
4.4 Detection of Acoustic Emission Signals 233

the KRNBB-PC has less sensitivity. Therefore, a suitable trade-off has to be made in
practice between sensitivity and flatness with frequency when selecting sensor types
for a particular application.

4.4.2 Sensor Modeling

To better understand some of the principles of the sensor systems used, several
attempts demonstrate the feasibility to model the response of piezoelectric sensors
by FEM [33, 87, 90, 91]. For realistic modeling of sensor signals, aspects of signal
propagation and sensor modeling have to be combined as demonstrated for a
commercial sensor type [35] and a parameter study of conical PZT elements
[87]. The modeling strategy presented in this section enables a dedicated analysis
of the key factors in sensor design, which are responsible for the different transfer
functions observed for the various sensor types [92]. Also, a comprehensive
modeling strategy of acoustic emission sensors enhances the possibilities to inves-
tigate possible new sensor designs [93–95] and allows new insights into other
sensor concepts to detect acoustic emission [95–97].
While acoustic wave propagation is sufficiently described by solving the con-
stitutive equations of structural mechanics, the interaction between acoustic waves
and a piezoelectric sensor requires new approaches using coupled partial differen-
tial equations [35, 98]. In particular, the generation of electrical charges due to
deformation of the piezoelectric material requires simultaneous solving of the
electrical and structural mechanics constitutive equations (4.44) and (4.45).
Another important aspect in simulation of transfer functions of piezoelectric sen-
sors is the interaction between the sensing material and the attached circuitry. In
contrast to conventional open-loop simulations of sensor response, additional input
impedances of preamplifiers and cables can change the sensor response signifi-
cantly and have to be taken into account for realistic prediction of the sensor
behavior. To include these effects, a FEM approach is demonstrated in this section
following reference [87] with technical details reported therein.
As seen in Fig. 4.90, signal propagation due to a pencil lead break in 250 mm
distance is studied in a 3.0 mm thick AlMg3 aluminum plate. All material proper-
ties used in the modeling are given in Tables B.1 and B.2 in Appendix B. As a first
step, the interaction of the sensor as function of the attached circuitry is discussed.
The respective geometric details of the sensor type investigated are shown in
Fig. 4.91 with the layout of the attached circuit given in Fig. 4.92.
A comparison of the experimentally detected signal and the modeling result is
shown in Fig. 4.93a for the conical sensor type and for the WD sensor type in
Fig. 4.93b. Despite of some differences at the beginning of the A0 mode, the main
characteristics of the two different sensor types are well captured by the modeling
approach. In particular it is possible to calculate the absolute values of the sensor
output since the strength of the pencil lead break source is known (cf. [53, 99]).
234 4 Acoustic Emission

backing mass
aluminum plate

3.0 mm

pencil lead break

mm
50 750
0 mm
m
m 250

ne
z pla
try
me
y sym
x xz-

Fig. 4.90 Three-dimensional model setup of experimental configuration (based on [87])

Fig. 4.91 Three-dimensional geometry of conical sensor type used in [87] (a) and WD sensor type
used in [35] (b) (images based on [35] and [87])

sensor sensor cable preamplifier

10Ω
10kΩ 10 kHz
90pF 15pF

Fig. 4.92 Drawing of electrical circuit used for simulation of connection cable and preamplifier
(based on [87])
4.4 Detection of Acoustic Emission Signals 235

0.15
a experimental signal

sig. voltage [V]


0.10
7.0 simulated signal
0.05
6.0 0.00 conical sensor
5.0 -0.05
-0.10
4.0 -0.15
40 45 50 55 60 65 70 75
signal voltage [V]

3.0 time [μs]

2.0
1.0
0.0
–1.0
–2.0
–3.0
–4.0
40 50 60 70 80 90 100 110 120 130 140
time [μs]
0.4
b experimental signal
sig. voltage [V]

0.3
6.0 0.2
simulated signal
0.1
5.0 0.0 WD sensor
-0.1
-0.2
4.0 -0.3
-0.4
3.0 40 45 50 55 60 65 70 75
signal voltage [V]

time [μs]
2.0

1.0

0.0

–1.0

–2.0

–3.0
40 50 60 70 80 90 100 110 120 130 140
time [μs]

Fig. 4.93 Comparison of experimentally detected pencil lead break signals in 250 mm distance
and simulated signals for conical sensor type (a) (based on [87]) and WD sensor type (b). Insets
show S0 mode arrival between 40 and 75 μs after signal excitation

4.4.2.1 Influence of Attached Circuit

As the first parameter study, the capacitance of the cable model was systematically
changed between 10 and 1000 pF, while holding all other circuit parts of Fig. 4.92
constant. The resulting sensor signals of the simulation are shown in Fig. 4.94a.
236 4 Acoustic Emission

a
15pF
50 150pF
1500pF
15000pF

25
signal voltage [mV]

–25

–50
40 60 80 100 120 140
time [μs]

b
0.3 1k
10 k
100 k
0.2
signal voltage [V]

0.1

0.0

–0.1

–0.2

–0.3

40 60 80 100 120 140


time [μs]

Fig. 4.94 Variation of cable capacitance for 1.50 mm conical element diameter (a) and pream-
plifier impedance (b)

In a second study, the resistance value of the preamplifier was changed from 1 to
100 kΩ (see Fig. 4.94b), while the cable capacitance was kept constant. Since the
ranges of both cover a broad range of the intrinsic sensor impedance and capaci-
tance (cf. [87]), a strong interaction between the piezoelectric element and the
circuitry is found.
4.4 Detection of Acoustic Emission Signals 237

The change in signal amplitude due to changes in the resistance and capacitance
was of similar magnitude as the geometrical changes of the conical elements. For
example, an increase in the capacitance of the piezoelectric element has the same
effect as a decrease in the circuit capacitance. This is a consequence of the coupled
electrical system of piezoelectric material and the attached circuit. Therefore, a
suitable match between the sensor type and the preamplifier has to be assured to
avoid a negative impact on the sensitivity of the system and the transmitted bandwidth.

4.4.2.2 Influence of Sensor Aperture

In addition to the attached circuitry, other effects also contribute to the final
appearance of the signal. As discussed previously [12, 24, 87, 100, 101], one
major influence on the signal shape is the aperture effect due to the wavelengths
of the propagating wave. This influence is pointed out in Fig. 4.95. A comparison of
a simulated signal without the presence of the sensor reveals the ideal shape of the
Lamb wave in the aluminum plate used.
An overlay with the normalized simulated sensor voltage signal for sensor tip
diameters of 0.25 and 4.00 mm demonstrates the characteristic differences in the
signal shape due to the increase of aperture. The out-of-plane signal is in phase with
the sensor signals at the beginning of the A0 mode, but increasingly deviates at

0.010 0.04
norm. signal voltage
displacement [nm]

0.005 0.02
displacement
1.0 0.25 mm tip diameter 1.4
0.000 0.00 4.00 mm tip diameter
1.2
0.8 –0.005 –0.02
1.0
–0.010 –0.04
0.6 40 45 50 55 60 65 70 75 0.8
time [μs]
displacement [nm]

normalized signal voltage

0.4 0.6
0.4
0.2
0.2
0.0 0.0
–0.2
–0.2
–0.4
–0.4 –0.6
–0.8
–0.6
–1.0
40 60 80 100 120 140
time [μs]

Fig. 4.95 Comparison between out-of-plane displacement at position of sensor midpoint without
the presence of sensor model and simulated signal voltage for conical elements with 0.25 and
4.00 mm tip diameter (based on [87])
238 4 Acoustic Emission

times larger than about 80 μs. This is caused by the intrinsic phase shift of the
piezoelectric conversion and the attached circuit. Despite the phase shift, a better
match to the simulated out-of-plane displacement is found for the 0.25 mm tip
diameter. Since this diameter approaches a point contact, the effects of aperture
become more negligible. In comparison, the 4.00 mm tip diameter suffers from
aperture more and thus shows a signal with more difference compared to the
original Lamb wave.

4.4.2.3 Influence of Impedance Mismatch

For the detection of an acoustic emission signal, the incident wave needs to be
transmitted from the propagation medium to the outermost part of the sensor. This
is the casing, the wear plate, or the piezoelectric element. This transmission is
subject to the considerations of the transmission and reflections coefficients as
given in (4.29) and (4.30). Therefore, the mismatch of acoustic impedance between
the sensor and the propagation medium is expected to show a significant role. In
addition, the detecting sensor can also be understood as spring-mass system.
Consequently, a relation to the stiffness of the propagation medium is also
expected. While it is hard to provide a general description of this relationship4
between the detected signal amplitude, the acoustic impedance, and the stiffness of
the structure, it is still worthwhile to assess the magnitude of this effect for an
exemplary case to yield the map of expected signal amplitudes shown in Fig. 4.96.
To this end the setup shown in Fig. 4.90 using the conical sensor model is used
and the properties of the plate are varied to investigate the difference in detected
signal amplitudes. The material properties of the isotropic materials listed in
Table B.1 in Appendix B span a broad range of acoustic impedances and Young’s
modulus values. In addition to the “real” materials, “artificial” materials were used
having values of acoustic impedance and Young’s modulus with equidistant spac-
ing. This allows interpolating between the different materials and allows computing
the full range of possible material properties.
Following [87] the incident acoustic waves are converted into electrical signals
using simulation of the piezoelectric conversion process and the attached circuitry.
Due to the different sound velocities in the materials, the durations for the individ-
ual computations are modified appropriately to allow for efficient computations.
The amplitude evaluated in Fig. 4.96 is the peak value of the A0 mode propa-
gating in the structure divided by the incident out-of-plane component of the elastic
wave. Computations for “real” materials are given by the black dots, and the
interpolated field was created from an array of 9  12 “artificial” materials spanning
the range between 5.4 and 54.4 MRayl and 2.5–1000.0 GPa.
It can be concluded from the results in Fig. 4.96 that neither the acoustic impedance
nor the stiffness of the plate shows strictly linear dependence on the detected signal

4
Note that [48, 73] discuss the effect of the acoustic impedance but not for incident plate waves.
4.4 Detection of Acoustic Emission Signals 239

Fig. 4.96 Map of detected signal amplitude for configuration used in [87]. Plate materials are
varied using “real” materials (black dots), and interpolated sensor response field is computed from
“artificial” materials

amplitudes. Moreover, the computed behavior seems to behave nonproportional in


some regions. However, there is a strong tendency seen for a change of the Young’s
modulus of the material. Here, the stiffness effect seems to overrule the effect of
acoustic impedance mismatches. For those materials with the highest Young’s mod-
ulus, the detected signal amplitudes are more than 300 times larger than for the
materials with low Young’s modulus. In comparison, the largest effect for a variation
of the acoustic impedances was observed to be in the range of a factor of 1.3.
These findings strongly indicate that the detectable signal amplitudes for the
same type of sensor are unique for every material. Therefore, calibration curves of a
sensor are unlikely to be representative for inspection of other materials than the
material used for the calibration. Since these calibration routines are usually carried
out on metallic materials (i.e., steel or aluminum), the direct transfer to fiber
reinforced composites might be questionable.

4.4.3 Waveguides

Another particular challenge arises if test temperatures exceed the operation tem-
peratures of piezoelectric sensors. In the past, suitable alternatives to PZT were
investigated to extend the accessible temperature range. One solution to avoid using
such PZT sensors at operation temperatures outside their specifications was found
240 4 Acoustic Emission

by the researchers like Lynnworth et al. and involves the usage of thin rods, clad
rods, or hollow tubes as ultrasonic waveguides for signal detection [102]. They
demonstrated the use of such waveguides to duplicate acoustic emission signals
propagating as guided waves in a plate. Also various short waveguides with
different material properties, length, diameter, and detection angles were tested in
order to investigate the effects upon pulsed events [102, 103]. However, the
selection of appropriate waveguides is still an experimental procedure and relies
on the experience to a large extent. Although the principle description of guided
wave modes in rods is found in standard literature [104], numerical methods are
nowadays also found to be a convenient tool to assess the properties of such
waveguides prior to experimental design [105, 106].
Following the established and validated numerical procedure described in
[105, 106], the response of a model sensor is evaluated in various geometrical
configurations and material pairings. In the following, effects of the waveguide
termination, the diameter, material, and temperature gradients are briefly discussed.
As seen in Fig. 4.97, the computations are performed for one reference case with
the sensor directly attached to a 3.0 mm thick aluminum plate. As an acoustic
emission source, a pencil lead break test in 100 mm distance to the sensor location
is used. To reduce the computational intensity of the model, a symmetry condition is
chosen along the xz-plane. The details of the model sensor are indicated in Fig. 4.97
and confirm with the previous choice in [87] (see also Sect. 4.4.2). As the second
model, the propagation path between source and sensor is modified by incorporation
of an acoustic waveguide as seen in Fig. 4.97. Based on an experimentally used
design, the diameter is selected as 1.59 mm and the length as 306.4 mm.

a
3.0 mm
source location
(x,y,z)=(0,0,3) mm
signal detection location
(x,y,z)=(100,0,3) mm

b
signal detection location
c (x,y,z)=(100,0,309.4) mm
19.0 mm
22.5 mm

backing
mass 3.0 mm

conical source location


(x,y,z)=(0,0,3) mm
9.3 mm

element
PZT -5A

Fig. 4.97 Geometry and dimensions of numerical case studies for plate (a), waveguide (b) and
sensor model (c) following [105, 106]
4.4 Detection of Acoustic Emission Signals 241

a
30
Reference Signal
Waveguide Signal
20

10
Amplitude [mV]

–10

–20

waveguide signal is 60.2 μs offset to superimpose


–30
0 20 40 60 80 100
Time [μs]

b
–44
Reference Signal
–55 Waveguide Signal

–66
FFT magnitude [dB]

–77

–88

–99

–110

–121

–132

0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
Frequency [MHz]

Fig. 4.98 Comparison of detected acoustic emission signals after propagation in aluminum plate
and after additional propagation in acoustic waveguide (a) and respective frequency spectra (b).
Signal detected at the top of waveguide in (a) is shifted forward by 60.2 μs to compensate for the
effect of signal propagation within the waveguide

The result of both computations is shown in Fig. 4.98. In order to allow better
comparison of both acoustic emission signals, the result including the computation
with the waveguide is shifted forward by 60.2 μs to compensate for the effect of
different propagation distances.
242 4 Acoustic Emission

Overall, both signals still show some resemblance, but the frequency content and
the amplitude of the individual guided wave modes are not fully retained. This is
also seen in the comparison of the frequency spectra of both signals in Fig. 4.98b. In
the present case, the signals propagate as guided plate waves in the aluminum plate.
At the position of the waveguide, the out-of-plane components are transmitted into
the waveguide. At this position, the signal transmittance is governed by the
impedance match of the waveguide to the plate and the aperture of the waveguide.
After transmittance into the waveguide, the wave propagation starts to transform
into the guided wave modes of the waveguide geometry (e.g., rod modes). The final
signals’ time-frequency characteristic at the top of the waveguide thus suffers from
two types of guided wave propagation and modal conversions, respectively. This
causes characteristic deviations to the acoustic emission signals detected by a
sensor mounted directly on the structure of interest, which is further denoted as
reference signal.

4.4.3.1 Shape of Waveguide

In order to ensure stable mechanical attachment of the waveguide and the acoustic
emission sensor, several terminations are typically used in practice. Therefore it is
helpful to understand their impact on the signal transmission when being used in an
application. To this end, [105] has investigated several terminations, keeping the
diameter and material constant. As seen in the frequency spectra of the acoustic
emission signals detected on the top of the waveguide in Fig. 4.99, the various
configurations do cause substantial differences in the transmitted signals.

Reference signal Reference Signal


Cylindrical Waveguide
–60 Conical truncation
Hollow Waveguide
Sharp tip truncation
FFT magnitude [dB]

–80

–100

–120

0.0 0.2 0.4 0.6 0.8 1.0


Frequency [MHz]

Fig. 4.99 Comparison of frequency spectra of signals detected with different waveguide
geometries
4.4 Detection of Acoustic Emission Signals 243

Relative to a simple circular waveguide truncations like cones or sharp tips may
either cause resonances (especially the conical truncation at the sensor position) or
may reduce the detectable signal amplitude by up to 17 dB. In comparison, the use
of hollow waveguides or such filled with liquids was found to be of less impact
concerning the waveguide transmission characteristics in the investigated diameter
and frequency range (for details on dimensions, see [105]).

4.4.3.2 Diameter of Waveguide

Based on the intended application, it might be required to select a particular


minimum waveguide diameter to offer a certain mechanical strength or stiffness.
On the contrary it might be necessary to select a certain maximum diameter to limit
temperature reduction at the contact position with the object or to confirm with
certain lead-through dimensions. For cylindrical geometries, the increase of the
diameter will directly increase the possible number of guided wave modes (see
Sect. 7.2.4.2 for an example). Mixed mode propagation and frequency specific
attenuation will thus affect the transmitted acoustic emission signals. Considering
the same model geometry as before, Fig. 4.100a presents some signals detected on
top of cylindrical waveguides with different diameters. Apart from the shift in
intense frequency regions, the signal amplitude is also affected. As seen in the
evaluation of the signal amplitude at a constant frequency of 450 kHz in
Fig. 4.100b, there is a preferred waveguide diameter at around 2.2 mm, which
offers highest signal detectability at the top of the waveguide. However, this is not a
simple resonance of the waveguide, but a specific transmission characteristic of the
waveguide at 450 kHz. Based on the material of the waveguide and additional
effects of the truncation at the end of the waveguide, the regions of preferred
frequency transmission may shift and therefore another diameter would be the
optimal choice. Thus, for practical applications, it is recommended to perform a
sensitivity analysis for typical acoustic emission signals found in the structure
under test.

4.4.3.3 Material of Waveguide

As indicated in [106], the impedance mismatch between the structure under test and
the waveguide will also cause noticeable differences in the detectable acoustic
emission signals. This is owed to the law of refraction as seen in Fig. 4.57 and
causes the same consequences as already pointed out in Sect. 4.4.2.3. Thus it is
recommended to use waveguide materials with acoustic impedance close to the
material under inspection to avoid strong impedance gaps. If this is not feasible,
the acoustic impedance of the waveguide should be selected larger than that of the
material under investigation to provide good acoustic transmission into the wave-
guide. However, this can cause additional modal conversion, therefore reducing the
effective transmitted amplitude.
244 4 Acoustic Emission

a
0.03 Reference signal
Waveguide 0.398 mm diameter
Waveguide 3.980 mm diameter
0.02

0.01
Amplitude [V]

0.00

–0.01

–0.02

–0.03
0 10 20 30 40 50 60
Time [μs]
b
0.8
CWT coefficient intensity at 450 kHz frequency

0.7

0.6

0.5
Amplitude [mV]

0.4

0.3

0.2

0.1

0.0
0 1 2 3 4 5 6 7
Diameter [mm]

Fig. 4.100 Comparison of different acoustic emission signals detected at the top of waveguide
with different diameter (a) and evaluation of maximum signal amplitude at constant frequency of
450 kHz using continuous wavelet transformation (b)
4.4 Detection of Acoustic Emission Signals 245

4.4.3.4 Influence of Temperature Gradient

Many waveguide applications deal with objects operating or being loaded at


elevated temperatures. Since typical acoustic emission sensors may not exceed
their operational temperature range, the waveguide is selected long enough to
reduce the temperature from the elevated temperature to safe operation conditions
for the acoustic emission sensor. In some applications this temperature gradient
within the waveguide can be in the order of several 100 C or even up to 2000 C.
Accordingly, the waveguide material changes its material properties as function of
the local temperature. Depending on the sensitivity of Young’s modulus, Poisson’s
ratio, and density of the material, the wave propagation is noticeably changed. An
exemplary calculation for an UNS K12211 steel alloy with a temperature gradient
of 500 C was previously presented in [106]. The findings did not reveal a strong
impact of the waveguide operation due to the existence of the temperature gradient.
Except for a delay in signal arrival of 1.65 μs, both signals were almost identical.
However, other waveguide materials may exhibit stronger changes of their material
properties as function of temperature and therefore may affect signal propagation
more than this.

4.4.4 Other Factors Affecting Sensor Sensitivity

Beyond the specific choice of the sensor type, several other factors will affect the
detectable frequency range and the sensitivity of the acoustic emission sensor.
Apart from aging of the sensor itself and environmental test conditions, these are
the choice of the coupling medium and the way of sensor fixation.

4.4.4.1 Coupling Medium

In principle, the purpose of the coupling medium is to avoid air gaps in between the
object under test and the acoustic emission sensor. Typically, the test object surface
is not fully flat or comes with a certain roughness, so the planar sensor surface
would only be coupled at distinct locations. Having a grease or a liquid as coupling
medium can thus fill these gaps and ensure a good acoustic transmission. Typically,
coupling media are viscoelastic materials causing higher acoustic attenuation at
higher frequencies. The choice of coupling media may thus limit the bandwidth of
the measurement chain as seen in Fig. 4.101. The spectra shown are averaged out of
20 pencil lead break signals, which were detected by a WD type sensor under
identical conditions on an aluminum plate except for the choice of couplant.
Obviously the hot glue reduces the sensitivity at higher frequencies when compared
to the less viscous Baysilone silicone grease couplant.
246 4 Acoustic Emission

Hot glue
160000 Baysilone

140000

120000
FFT-Magnitude [1/mV]

100000

80000

60000

40000

20000

0
0 200 400 600 800 1000
Frequency [kHz]

Fig. 4.101 Comparison of frequency spectra recorded under identical conditions except for the
choice of couplant

In combination with temperature ranges other than room temperature, several


other factors may limit the applicability of coupling media. At temperatures <0 C
included water or the coupling medium itself may freeze, causing microcracking
with noticeable acoustic emission. Although this acoustic emission is often with
relatively low amplitudes, it may cause a substantial amount of additional signals,
which may interfere with acoustic emission signals emitted by the test object. For
such applications a coupling medium needs to be selected which does not show
intrinsic acoustic emission activity in the intended temperature range.
In the range above room temperature, the primary challenge is to avoid degra-
dation of the coupling medium. This can be assessed prior to the monitoring task by
a face-to-face arrangement. Here, one sensor is used as a pulser and another sensor
is used as a detector, both coupled together in a face-to-face arrangement using the
coupling medium to be tested. While the pulser is emitting a defined broadband
pulse at distinct intervals, the whole configuration is subject to continuous temper-
ature variations. The evaluation of the detected signal amplitudes is shown in
Fig. 4.102 for three different types of coupling media and one dry contact coupling.
As seen from the difference in signal amplitude at room temperature, the different
coupling choices come with different initial sensitivities being worst for the dry
coupling. During the temperature increase up to 200 C, the detected signal
amplitude decreases, which is mostly due to the thermally induced expansion of
the overall face-to-face arrangement and the respective changes in the contact
force. Still, the overall loss in signal amplitude is still different for each coupling
medium. Upon cooling, the coupling media show hysteresis effects of the detected
4.4 Detection of Acoustic Emission Signals 247

signal amplitudes. Back at room temperature, only the Apiezon has returned to its
previous sensitivity. The other two choices do not get back to their original
sensitivity within a range of 2 dB, thus indicating possible degradation effects
of the coupling medium. Such procedures allow to evaluate the performance of the
coupling medium prior to any testing.

a
98
96 heating
cooling
94
92
90
88
Amplitude [dB]

86
84
82
80 Baysilone
78
76
74
72
70
20 40 60 80 100 120 140 160 180 200 220
Temperature [°C]
b
98
96 heating
cooling
94
92
90
88
Amplitude [dB]

86
84
82
80 Apiezon
78
76
74
72
70
20 40 60 80 100 120 140 160 180 200 220
Temperature [°C]

Fig. 4.102 Detectable signal amplitude for various coupling media (a, b, c) as function of
temperature in a face-to-face configuration. Reference case without coupling medium in (d)
248 4 Acoustic Emission

c
98
96 heating
cooling
94
92
90
88
Amplitude [dB]

86
84
82
80 Triboflon
78
76
74
72
70
20 40 60 80 100 120 140 160 180 200 220
Temperature [°C]
d
98
96
94
92
90
88
Amplitude [dB]

86
84
82
80 dry coupling
78
76
74
72
70
20 40 60 80 100 120 140 160 180 200 220
Temperature [°C]
Fig. 4.102 (continued)

4.4.4.2 Sensor Fixation

Apart from the coupling medium, another critical item for acoustic emission
sensors is the way of sensor fixation during loading of the test object. Basically,
the coupling medium may be categorized in those that bond the acoustic emission
sensor to the structure and those which do not allow to carry its weight.
4.4 Detection of Acoustic Emission Signals 249

For the bond couplings, all types of permanent adhesives, hot glue, as well as
two-sided duct tapes are typically used. Despite of their ease of handling, the main
drawback is to assure good bonding quality during loading of the test structure and
possibly also during temperature changes and other superimposed environmental
factors. Especially for fiber reinforced materials, the adhesive needs to be compat-
ible with the matrix material and the sensor casing material (typically metallic or
ceramic). Upon large deformations or superimposed temperatures, this may often
lead to failure of the adhesive, thus causing additional acoustic emission signals or
drop-down of the attached sensor. Also, the choice of adhesive may affect the
frequency-dependent sensitivity as seen in Fig. 4.102.
As alternative, the acoustic emission sensor can be attached to the test object
using a viscous or liquid coupling medium, but a mechanical force is required to
carry the weight of the sensor and to keep it in its designated position. To this end,
several fixation systems are established. Custom-made systems operating from
mechanically stable platforms or fixtures can be used as well as clamp systems,
rubber bands, (external) tacky tapes, suction cups, or magnetic holders. In
Fig. 4.103 some of these concepts are shown in exemplary applications. Note that
the magnetic holders are somewhat restricted in terms of composite applications,
since these require a ferromagnetic material to be attached to. For many of these
systems, their main advantage is the ease of handling and the instant contact of the
sensor with the structure (as compared to adhesives which might need time to cure).
Also, unmounting and repositioning of the sensor is fairly easy, since no permanent
contact is made with the test structure.

Fig. 4.103 Sensor fixation using clamp system (a), (external) tacky tape (b), and suction cup
holders (c)
250 4 Acoustic Emission

4.5 Signal Classification

In recent years, many new approaches were proposed that add the possibility of AE
source identification to the established approaches of AE signal accumulation and
AE source localization. Namely, these are moment tensor inversion, time reversal
approaches, guided wave analysis, and pattern recognition techniques.
In volumetric media, one suitable approach to deduce the source type is to use
the moment tensor inversion. A comprehensive review on this topic was recently
given in [52], so only a brief overview will be provided in the following. Based on
the type of source description by the moment tensor as given in Sect. 4.2, the basic
measurement principle stems from the acquisition of a set of signals at different
angles to the source position. Using these different signals and knowing the Green’s
function of the medium, the moment tensor components are readily computed by a
deconvolution analysis after Wadley [27]. In this field, Ohtsu and Ono demonstrated
that the moment tensors from different crack modes in volumetric media are readily
distinguishable (see [18, 21, 52]). Green extended this approach used in isotropic
media to the needs of acoustic emission sources in composite laminates
[29, 30]. Despite these valuable efforts, there are several drawbacks associated with
the moment tensor inversion in application to fiber reinforced composites. One
challenge stems from the reduced field of view on the source in the typically thin
composite materials. Since signals propagate as guided waves, much of the informa-
tion on source orientation is lost at the first millimeters of propagation. Therefore,
even for thick laminates, the different observation angles of the source required for
moment tensor inversion are difficult to obtain in real testing of composite materials
or structures. Another challenge is the precise knowledge of the Green’s functions of
the medium. Although it was demonstrated that these can be computed practically for
every piece of material, it is still burdensome to achieve in practice.
However, if moment tensor inversion is performed on a composite material, it is
a well-suitable technique to classify signal origins in different source radiation
types. This yields valuable insight on the material failure at this level.
In recent years, the approach of time reversal acoustic emission has emerged as
an alternative source localization strategy [107–109]. Here the detected signals are
inversed and basically back projected to determine their signal origin. Beyond the
spatial source position, this also yields further information on the radiation pattern
of the source. This was successfully demonstrated for measurement of the crack
orientation in thin plates [109]. Consequently, the time reversal approach might be
used for composites to extract further information on the source origin. But still the
approach faces the same challenge as the moment tensor inversion, which is the
knowledge of the Green’s function of the system required to allow a valid inter-
pretation of the source activity.
Another classification technique more adapted to composite materials is the
analysis of guided waves. This approach has a prime focus on the relationship
between certain source types and the type of guided wave modes found in platelike
structures [6, 67, 110–117]. The aim is to find characteristic ratios of certain guided
wave modes, which are characteristic of a particular source type. Although such
4.5 Signal Classification 251

relations are partially expected, it is still hard to allow for unique assignments of
such guided wave ratios to source mechanisms (see Sect. 4.2). The first publications
of this approach investigating the relation between source type and guided wave
formation attracted much attention and allowed for better understanding of the
signal propagation in platelike materials. Also the pattern recognition techniques
presented in the following might be understood as more general classification
routine following the idea of guided wave acoustic emission analysis.
In the context of acoustic emission analysis, the purpose of pattern recognition
techniques is the identification of similarities in the recorded acoustic emission
signals. In contrast to conventional feature-based interpretation of the signals,
pattern recognition techniques use a multitude of features and consequently can
identify distinct signal types even if they cannot be described by fixed feature
limits. Basically, pattern recognition methods can be separated into two
approaches, namely, supervised and unsupervised pattern recognition.
The task of unsupervised pattern recognition is to separate a set of given objects
into distinct groups according to their similarity to each other without any previous
knowledge. Various approaches with a focus on detection of characteristic similar-
ities of the recorded signals have been published [7, 24, 77, 118–130].
As distinctly different approach, supervised pattern recognition techniques con-
sist of two subsequent stages. In the supervised stage, a set of objects with known
assignment to the respective classes is prepared. This assignment is usually denoted
labeling. Here an algorithm is trained to recognize these types of objects based on a
given set of features. In the subsequent stage, the algorithm is applied on objects
with unknown assignment and classifies them based upon their similarity to the
object classes provided in the supervised stage. To this end, discrimination between
noise and non-noise acoustic emission signals is often achieved by unsupervised
pattern recognition techniques [118, 119]. Acoustic emission signals originating
from friction and electromagnetic inductions can easily be identified in this respect,
due to their inherent characteristic difference to transient acoustic emission signals.
Utilizing suitable experimental considerations and finite element simulations, the
respective signal types can also be associated with specific failure mechanisms
[7, 24, 120, 122–125, 131–134].
In terms of pattern recognition techniques, the problem faced is the identification
of natural clusters of acoustic emission signals. One of the major difficulties in this
concern is an adequate evaluation of the classification results of different clustering
approaches. There are many ways to choose features, normalization procedures,
and algorithms that will influence the result of the classification process.
The first step in every signal classification procedure is a suitable data reduction
of the acquired AE signals. This comprises elimination of obvious noise signals and
a suitable strategy to focus on the AE signals relevant for material failure (see
Sect. 4.5.1). Subsequently, the detected AE signals are reduced to a number of
features calculated from the signals as defined in Appendix C. A number of those
extracted features are then used as dataset and are investigated by an unsupervised
pattern recognition method to yield groups of similar AE signals (AE signal
clusters). These basic steps used in the following are schematically shown in
Fig. 4.104. As a last step, a suitable validation of the resulting partition is
252 4 Acoustic Emission

feature 1 feature 2
0.10
Intensity
Intensität

0.05 feature 1
0.00
0 500 1000
Frequenz [kHz]
Frequency [kHz]
Feature
0.10
Intensity
Intensität

0.05

0.00
0 500 1000
extraction
Frequenz [kHz]
Frequency [kHz]
0.10
Intensity
Intensität

0.05
feature 2
0.00
0 500 1000
Frequenz [kHz]
Frequency [kHz]

Fig. 4.104 Three exemplary frequency spectra resulting in three data points after feature extrac-
tion process. Evaluation of many signals potentially results in cluster formation for frequency
intensity at position feature 1 and feature 2. Ellipsoids indicate extension of clusters and can be
evaluated by outer diameters and relative distances

mandatory. Since a partition is by definition an arbitrary grouping of signals, one


has to assure that the grouping is meaningful in a physical sense. For that purpose,
cluster validity measures can be used to compare partitions obtained throughout
different pattern recognition approaches.

4.5.1 Recommended Practices Before Starting Signal


Classification

As seen in Fig. 4.105, generally one can distinguish between continuous acoustic
emission and transient acoustic emission (burst) signals. Both types are encountered
in fiber reinforced materials, and the transition between the observations of either
type is also subject to the loading rate of the experiment, the inspected volume, and
the general acoustic emission activity of the inspected material. Ultimately if
damage progression in the material is fast and many AE sources are operative,
the individual burst signals superimpose and merge into a continuous AE release.
In the following all descriptions refer to the burst-type signals, since they can be
attributed to a particular source type, whereas the interpretation of continuous
acoustic emission has to use fairly different concepts [52].
Before starting a source discrimination routine in terms of failure discrimination,
some practices are suggested in order to allow for better interpretation of the data.
The first part consists of a suitable removal of noise signals potentially present in
the recorded dataset. Some recommendations are to look at the distribution of
feature values as seen, e.g., in Fig. 4.105:
• Signals with durations at the limits of the detection system indicate signals with
long durations, potentially from continuous acoustic emission sources, such as
friction.
4.5 Signal Classification 253

25000
0.003

0.002

0.001
20000

amplitude [V]
0.000
duration [μs]

-0.001

15000 -0.002

-0.003

-0.004
0 100 200 300 400 500 600 700 800 900 1000 1100 1200
10000 time [μs]

continuous signals

5000 10

amplitude [V]
0 0
40 50 60 70 80 90 100
amplitude [dB AE] -5

-10

0.10 0 100 200 300 400 500 600 700 800 900 1000 1100 1200
0.002
time [μs]

0.001
0.05 saturated signals
amplitude [V]

amplitude [V]

0.00
0.000

-0.05
-0.001

-0.10
-0.002

-0.15
0 100 200 300 400 0 100 200 300 400 500 600 700 800 900 1000 1100 1200
time [μs] time [μs]

poorly digitized signals transient signals

Fig. 4.105 Typical diagram of amplitude versus duration of acoustic emission signals with noise
signals

• Signals with very short durations are sometimes just electromagnetically


induced noise (spikes).
• Signals with very low amplitudes may fall into the digitization limit and may not
yield accurate feature values.
Another recommendation is to use source localization routines to increase the
level of relevance of a particular signal origin. If the acoustic emission of a source is
recorded by more than one sensor within a certain time interval, this provides
additional evidence that the signal is within the region of interest of interpretation
(e.g., the gage section of a specimen). This excludes all signals outside the region of
interest and consequently is sometimes referred to as Δt-filter. This can inhibit
triggering if there is no signal arrival within a certain Δt at two or more sensors (see
also Fig. 4.122). Also, the same concept can be used in a postprocessing step,
eliminating all signals for further analysis, which are not detected at more than one
sensor within a certain time interval Δt. If the source can also be localized5 in this
region of interest, this indicates even higher relevance for the interpretation of the
respective acoustic emission signals.

5
Note that the arrival of a signal at two or more sensors within a certain time interval does not
necessarily imply that the source position can be determined by a source localization algorithm.
254 4 Acoustic Emission

If there is no possibility to remove “noise” signals, such as spurious signals or


continuous signals from transient pulses, one can also specifically train neural
networks to detect the “noise” signals based on prototypes of “useful” signals and
“noise” signals. For such an application, one can choose a multitude of features for
description of the signal types and train a neural network to distinguish the two
groups of signals. As seen in Fig. 4.106, reapplication of such a neural network then
can be used to automatically classify and select the signal types of interest.

a
25000 "noise" signals
useful signals

20000
duration [μs]

15000

10000

5000

0
40 50 60 70 80 90 100
amplitude [dBAE]
b
108
"noise" signals
useful signals
107

106
absolute energy [aJ]

105

104

103

102

101
40 50 60 70 80 90 100
amplitude [dBAE]

Fig. 4.106 Neural net-based removal of signals attributed to noise sources as plots of signal
amplitude vs. duration (a) and as signal amplitude vs. absolute energy (b)
4.5 Signal Classification 255

However, if there are changes to the type of experiment such as type of loading,
type of material, stacking sequences, or other parameters, a direct application of
such trained neural networks to new datasets is often not possible.
Another challenge arises due to the feature extraction process itself (see
Appendix C for mathematical definitions of typical AE features). First of all, feature
values are always linked to the acquisition settings of the equipment and the sensor
type. Therefore it is hard to compare feature values of different setups in a
quantitative fashion. Moreover, the feature values will strictly depend on the
selected portion of the signal. As seen in Fig. 4.107 in the time and frequency

a
300μs
0.04
200μs

0.02 100μs
amplitude [V]

0.00

–0.02

–0.04

200 300 400 500 600 700 800 900 1000 1100 1200
time [μs]
b
4 window after threshold crossing
fpeak = 104 kHz 100μs
200μs
fpeak = 327 kHz 300μs
3
FFT-magnitude [1/V]

fpeak = 361 kHz

0
0 100 200 300 400 500 600 700 800 900 1000
frequency [kHz]

Fig. 4.107 Feature values for feature extraction process using three different window lengths in
time domain (a) and frequency domain (b)
256 4 Acoustic Emission

domain, the change of the time window to extract feature values has significant
impact. In particular for small specimens with many occurrences of edge reflec-
tions, the relevant portion of the signal is at the beginning [35, 56]. Therefore,
feature extraction is recommended to start at the time of first arrival and then to
extend until a significant part of the signal has arrived at the sensor position. In
particular for guided waves, this should imply arrival of all wave modes propagat-
ing in the structure to a certain amount. For larger specimens with vast differences
in the arrival time of the different modes, a suitable compromise has to be achieved
between the length of the feature extraction window and the inclusion of a large
amount of reflections of the specimen edges.
Although the method proposed in [125] can potentially deal with arbitrary
numbers of features, it is recommended to preselect a number of features for such
an investigation. One way to reduce some redundancy of the features is to consider
feature correlation dendrograms as shown in [35, 118]. Another approach is to
consider the individual feature distributions as recommended in [135, 136]. If the
feature distribution does not indicate more than one maximum, it is not likely a
feature of large relevance to segment data. However, all of these feature reduction
methods should be used carefully as higher-order correlation between different
features might be lost in such attempts [135, 136].

4.5.2 Pattern Recognition Method to Detect Natural Clusters

As indicated in the previous section, there have been many attempts to perform
pattern recognition in order to allow a classification of AE source mechanisms. In
general, such attempts may be distinguished in two phases. The first phase is the
statistical data analysis, which is used to form groups of signals. The second phase
is the attempt to correlate each group with a particular AE source mechanism. In
this section a data analysis algorithm is presented, which aims for the detection of
“natural clusters.” These are understood as clusters of entities, which provide
highest discrimination to the members of other groups. In the context of acoustic
emission, it is important to mention that these natural clusters are only potentially
related to different AE source types. It may also happen that the natural clusters of
an AE dataset are linked to different sensor types, due to propagation effects or
similar effects which also alter the detected AE signals. Thus it is worthwhile to
consider the occurrence of natural clusters, but this is not necessarily enough to
allow full distinguishability of different AE source types. Moreover, the direct
relation between particular cluster structures and respective AE source mechanisms
needs to be validated in every experimental configuration. However, it has been
demonstrated by various groups that the cluster strategy proposed in [125] works
for source discrimination of AE signals of composite materials and wood fracture
and in plant science [24, 124, 129, 137–143].
In order to find the natural clusters of a dataset without previous assumptions on
the number of clusters and previous feature selection, an automated screening of
4.5 Signal Classification 257

feature combinations is a suitable approach. To evaluate the quality of the partition


of a dataset, several cluster validity indices are established in literature
[144–147]. Since each cluster validity index shows a different performance
depending on the occurrence of outliers and the shape of the clusters, a combined
evaluation of multiple cluster validation indices is a possibility to increase the
reliability of the cluster identification method. Such combined evaluation was
already proposed by G€unter and Bunke [148] in the form of a voting scheme to
measure the overall performance of the different indices.
The goal of the presented unsupervised pattern recognition methodology is to
introduce an automated technique to detect promising feature combinations for
clustering of acoustic emission signals as presented in [125] based on the method
proposed in [35]. In contrast to feature reduction procedures (e.g., based on
correlation dendrograms [118]), an exhaustive search of global optimal combina-
tions of the signal features used is performed. As already pointed out by Polikar,
such exhaustive search methods are conceptually simple but computationally
devastating since the number of feature combinations to investigate grows combi-
natorial with the number of available features [136].
Consequently, before application of the presented exhaustive search method, an
educated preselection of promising features for description of the respective acous-
tic emission signals is suggested. In principle the proposed method has no limitation
regarding the number of features, but the exhaustive screening of a large number of
feature combinations will lead to increased computation times. A visualization of
the complete method as described in the following is shown in Fig. 4.108 as a
flowchart diagram.
Determine all feature combinations:
The first step consists of selection of all possible feature combinations with a given
minimum number of features Kmin to be used for clustering. The total number Ntot
of combinations of K previously selected features is thus
XK  
K
N tot ¼ ð4:52Þ
X¼K min X

Preprocessing:
As a second step, all preprocessing options for each feature combination are
applied. In the current investigation, the preprocessing consists solely of feature
normalization by dividing by the standard deviation. This is based on preceding
investigations on classification of acoustic emission signals [24, 124, 125, 137]. For
a generalized approach, the settings should be modified to fit the specific needs of
the classification problem investigated.
Clustering:
Subsequently the respective partitions for 2, 3, 4, . . . Ω clusters are calculated
utilizing a cluster algorithm with Ω as the maximum number of clusters. The
k-means algorithm is one of the most widely used algorithms to this end. However,
depending on the data structure, the k-means algorithm may fail to detect
258 4 Acoustic Emission

Fig. 4.108 Flowchart of


Determine all feature
generalized pattern combinations N
recognition method used in
the following (based on
[125]) for j = 1, …, N

Preprocessing (once)
Normalization

Clustering (for i = 2, …, P)
Apply cluster algorithm

Calculate cluster
validity indices

Voting scheme

for k = 1, …, Q

Sort (for j = 1, …, N)
Evaluate cluster
validity indices

Voting scheme

Number of Best feature


natural clusters combination

meaningful partitions [126, 149, 150]. Hence, other cluster algorithms, such as the
Gustafson-Kessel algorithm, LVQ, SOM, or other computationally intense
approaches may yield improved results. To increase the chances of finding the
global minimum using the k-means algorithm, MacQueen’s implementation with
ten random cluster center initializations and a maximum number of 100 iterations to
convergence was applied [151].
For the present implementation, as a measure of similarity, the Euclidean
distance was used. For each partition the Tou index TOU, the Davies-Bouldin
index DB, Rousseeuw’s silhouette value SIL, and Hubert’s Gamma statistic
GAMMA were calculated. Since these cluster validity indices are comprehensively
described in the respective authors’ original publications [144–147], they are not
4.5 Signal Classification 259

explicitly repeated. It is worth pointing out that the TOU index, the SIL value, and
the GAMMA statistics indicate good partitions by value maximization, while the
DB index indicates good partitions by value minimization.
Voting scheme:
Based on these four cluster validity measures, the numerical performance of each
partition is evaluated in the following scheme adapted from G€unter and Bunke [148]:
1. The number of clusters with the best index performance is given Ω points.
2. The number of clusters with second-best performance is given ðΩ  1Þ points.
3. The number of clusters with third-best performance is given ðΩ  2Þ points.
4. ...
Subsequently the points of all four cluster validity indices are accumulated as a
function of the number of clusters and are evaluated for their global maximum in
points. The respective number of clusters is defined as numerically optimal and thus
selected for the current feature combination. As discussed in [148], this combined
evaluation of multiple indices can improve automated determination of the optimal
number of clusters. Since cluster validity indices are often affected differently by
outliers and the dataset structure, strengths of different indices are effectively
combined by the voting strategy, while weaknesses are reduced.
In this context it is worth pointing out that the cluster validity indices chosen
have low numerical complexity. There are numerous other cluster separation
measures available, but these typically come with increased computational costs.
Thus they are less desirable for automated screening of large numbers of feature
combinations.
Detection of globally best feature combination:
As a final step, the feature combination with best global performance is determined.
By definition, the partition with best global performance is detectable by the
extreme values of the associated cluster validity indices. Therefore the 25 best
partitions for each cluster validity index are evaluated as follows:
1. The partition (feature combination) with best index performance is given
25 points.
2. The partition with second-best performance is given 24 points.
3. The partition with third-best performance is given 23 points.
4. . . .
Subsequently, the individual results of the cluster validity indices are cumulated.
Thus a partition which is favored by all cluster validity indices receives exactly
100 points. Similarly, good partitions would get high numbers of points, while less
favored or ambiguous partitions get fewer points. Typically, the partition with best
global performance has less than 100 points. Thus the individual cluster validity
indices would suggest different partitions. This indicates why the voting strategy is
better than simple extreme value search, since it takes into account an averaged vote
of the individual cluster validity indices.
260 4 Acoustic Emission

Since the first publication of the method, it has been adopted by various groups
to classify acoustic emission signals [129, 139, 140, 142, 143, 152]. The overall task
of the method is to detect the most significant clusters of the entirety of acoustic
emission signals with a minimum of initial assumptions on the cluster structure.
Therefore, no assumptions are made on the exact number of signal classes or the
number of AE features or the type of AE features.
Compared to signal classification methods based on single features, pattern
recognition methods are computationally intense. But single AE features like
peak-frequency or signal amplitudes can have significant dependency on the type
of sensor, the details of the specimen geometry (cf. Sect. 4.5.3), or the propagation
distance between source and sensor [153]. Consequently, source classification by
static AE feature ranges cannot be generalized beyond certain limits. In contrast,
pattern recognition techniques are adaptive to the problem investigated and do not
rely on static AE feature ranges.

4.5.3 Uncertainty of Classification

Any classification of acoustic emission ultimately results in a partition. This


grouping of signals might show an overlap of some of the feature values to a certain
degree. This leads to ambiguity for some of the signals located at the boundaries of
the groups, since they could potentially fall into two or more groups. In order to
quantify such uncertainty of the classification process, recently a stochastically
method was proposed to estimate the overlap of such cluster structures as resulting
from pattern recognition [154]. In the following, the pattern recognition process is
used as an example to quantify the UoC, but the method is straightforward to apply
on any type of classification achieved by other methods.
It is demonstrated how the values of cluster validity measures DB, TOU, SIL,
and GAMMA can be used to estimate the UoC. In order to establish an analytical
correlation between the cluster validity measures and the UoC, artificially gener-
ated datasets were investigated [154]. Following the approach of Milligan [155] and
the refinement by Qiu and Joe [156], datasets according to the implementation
within the software package “R” by Qiu et al. were generated. As a measure of the
degree of separation, J as introduced by [156] is used. The measure J is based on the
separation of two clusters generated from two univariate normal distributions (0, 1)
~ [156]. For the values of A
and (0, A) e ¼ 8 and A e ¼ 4, the measure of the degree of
separation ranges from “well-separated” ðJ ¼ 0:342Þ to “close” ðJ ¼ 0:010Þ as
shown in the scatterplots in Fig. 4.109.
The proposed pattern recognition method can identify the number of clusters
accurately down to values of J ¼ 0:010 [125]. Since the internal structure of the
artificial datasets is initially known, this allows quantifying the classification error
using the Rand index [157]. This index relates the pairwise assignment of objects in
two partitions to all pairs of the partitions. In the present case, one partition is the
4.5 Signal Classification 261

a
15
J = 0.342 Class 1
Class 2
10 Class 3

5
Feature 2

–5

–10

–15
-25 -20 -15 -10 -5 0 5 10 15 20 25
Feature 1

b J = 0.010
Class 1
15
Class 2
Class 3
10

5
Feature 2

–5

–10

–15

–20 –15 –10 –5 0 5 10 15 20


Feature 1

Fig. 4.109 Visualization of cluster structure for degree of separation of J ¼ 0.342 (a) and
J ¼ 0.010 (b) (based on [154])

initially known assignment of the objects to a particular cluster. The other partition
is the resulting assignment of objects of the pattern recognition method. A true
positive (TP) decision is a classification of both objects to the same cluster; a true
negative (TN) decision assigns two dissimilar objects to dissimilar clusters. These
quantities are related to the false-positive (FP) decisions, which assigns two
262 4 Acoustic Emission

dissimilar objects to the same cluster and false negative (FN) decisions that assign
two similar objects to different clusters:

TP þ TN
RAND ¼ ð4:53Þ
TP þ TN þ FP þ FN

The Rand index is a direct measure of the percentage of decisions that are correct
and can be employed as direct measure of the UoC. Next the correlation between
the calculated cluster validity measures DB, TOU, SIL, and GAMMA and the Rand
index is considered. To this end, a number of artificial datasets with varying degree
of separation J between 0.45 and 0.45 are investigated. The number of objects in
each cluster was randomly chosen within the range [50, 200] which reflects rea-
sonable variation of the cluster sizes.
In the following, the case of three clusters and five features is discussed as an
example to demonstrate the relationship between the cluster validity measures and
the Rand index. In Fig. 4.110 values of the Rand index are plotted as a function of
DB, TOU, SIL, and GAMMA. Since all cluster validity measures are by definition
linked to the quality of the partition, a correlation to the Rand index is expected.
Here, DB indicates a partition of high quality by a numerical minimum, while TOU,
SIL, and GAMMA maximize their values for high cluster separation. The scatter in
Fig. 4.110 indicates that there is no unique analytical correlation between a partic-
ular cluster validity measure and the Rand index. This is owed to the statistical
distribution of clusters elements, which can cause nearly identical values of DB,
TOU, SIL, or GAMMA for two partitions, while the respective number of correct
classifications is still different for both partitions. In order to bring this behavior
down to an analytical relationship, a five-parameter logistic function is used to fit
the cluster validity measure c to the Rand index:

Amax  Amin
RAND ¼ Amin þ   h s ð4:54Þ
1 þ cc0

The boundary conditions of the Rand index with lower limit of Amin ¼ 0 and upper
limit of Amax ¼ 1 reduce the number of independent fit parameters by two. The
resulting fit including the 95 % prediction band is shown in Fig. 4.110. Values of the
fit parameters c0, h, and s are shown in the respective inset.
As indicated by the values of the adjusted least square errors (adj. R2), the
various cluster validity measures show different performance in their correlation
to the Rand index value. As noted in [154], best fit quality was observed for SIL and
GAMMA. A strong correlation was found for DB, while the size of the 95 %
prediction band for TOU was found to be insufficient for reliable estimation of
the UoC.
As demonstrated in [154], the application of (4.54) to the cluster validity
measures of an experimental dataset using the fit parameters from Fig. 4.110
directly yields the UoC given by the estimated Rand index of the experimental
4.5 Signal Classification 263

partition. The UoC is understood as fraction of the signals, which are potentially
classified in the wrong cluster. However, influences of the number of features
selected for the process and the number of clusters have to be considered, when
using values for the fit parameters c0, h, and s (cf. [154]).

a
c0= 1.73 +/- 0.21
1.0
h = -9.59 +/- 1.86
s = 1.01 +/- 0.84
2
0.8 adj. R = 0.85823
Rand Index

0.6

0.4

0.2 data
fit
95% prediction band
0.0

0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0 2.2 2.4 2.6 2.8 3.0
Davies-Bouldin Index
b
1.0

0.8
Rand Index

0.6
c0= 1.10 +/- 0.37
h = 8.37 +/- 2.33
0.4 s = 0.57 +/- 0.24
2
adj. R = 0.66218
data
0.2
fit
95% prediction band

0.0

0.0 0.5 1.0 1.5 2.0 2.5 3.0


Tou Index

Fig. 4.110 Rand index as a function of cluster validity statistics. Plots include results of fit model
and their corresponding 95 % prediction band. Diagrams are shown for Davies-Bouldin index (a),
Tou index (b), Hubert’s Gamma coefficient (c), and Rousseeuw’s silhouette value (d) (based on
[154])
264 4 Acoustic Emission

c
1.0

0.8
Rand Index

0.6
c0= 0.28 +/- 0.01
h = 5.22 +/- 0.23
0.4 s = 5.59 +/- 4.79
adj. R2 = 0.93266

0.2 data
fit
95% prediction band
0.0

0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
Hubert's Gamma coefficient
d
1.0

0.8
Rand Index

0.6
c0= 0.12 +/- 0.07
h = 4.27 +/- 0.42
0.4 s = 6.40 +/- 5.81
adj. R2 = 0.93071

0.2 data
fit
95% prediction band
0.0

0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
Rousseeuw’s Silhouette value
Fig. 4.110 (continued)

4.5.4 Factors of Influence

In the following some of the factors influencing the quality of signal classification
procedures are presented. All factors are discussed in terms of the respective origin,
and their effect on the pattern recognition approach proposed in references
[35, 125] is used as an example to demonstrate their impact. To judge on the
different factors, the quantification method from [154] is applied to yield a measure
of the classification accuracy in form of the UoC, whenever appropriate.
4.5 Signal Classification 265

4.5.4.1 Source-Sensor Distance

Since attenuation and dispersion affect the acoustic emission signal during propa-
gation, these naturally have a significant impact on the possibilities to distinguish
different signal sources. The impact of this effect can be experimentally studied in a
test specimen using test sources being able to excite acoustic emission signals at
various frequency ranges. To demonstrate this effect, reference [154] investigated
this effect using large double-cantilever beam specimens. Three types of acoustic
emission test sources were applied at source-sensor distances between 80 and
280 mm as seen in Fig. 4.111. In analogy to the signals detected during testing of
double-cantilever beams, signal features Partial Power 2 and Weighted Peak Fre-
quency were calculated from the obtained signals as defined in Appendix C. As
demonstrated in [154], each of the three acoustic emission test sources excites a
unique combination of S0 mode, A0 mode, and SH0 mode. This translates into a
unique feature range as seen in Fig. 4.112.
Since the depth position of the pencil lead breaks is constant and the source type
is identical, the variation in feature values shown in Fig. 4.112 is solely caused by
the change in propagation length to the sensor. Similarly, the feature range of
signals of pencil lead breaks at the edge of the plate starts at different feature values
and spans a range of feature values in the diagram of Partial Power 2 and Weighted
Peak Frequency that coincides well with the cluster attributed to interfacial failure.
Here, both feature ranges (Partial Power 2 and Weighted Peak Frequency) are well
reflected by the signals detected after different propagation lengths. For the test
signals from the WD pulser, the change of propagation length to the sensor also
causes a significant shift in the feature values Partial Power 2 and Weighted Peak
Frequency. However, the trajectories of the values extracted for the different test
sources do not intersect with each other, despite of the change in propagation

WD-sensor (detection)
40
pencil lead breaks (top surface) m
+ pulser positions m

mm 6.4 mm
570

pencil lead breaks (side


surface) positions
z
y x
Fig. 4.111 Scheme of experimental setup used for generation of AE test sources including
dimensions of DCB specimen used as propagation medium and subsequent mechanical test
(based on [154])
266 4 Acoustic Emission

a
100
PLB top surface
90 PLB side surface
piezoelectric pulser
80

70
Partial Power 2 [%]

60

50

40

30

20

10

0
0 100 200 300 400 500 600 700
Weighted Peak-Frequency [kHz]
b
100
Matrix cracking
90 Interfacial failure
Fiber breakage
80

70
Partial Power 2 [%]

60

50

40

30

20

10

0
0 100 200 300 400 500 600 700
Weighted Peak-Frequency [kHz]

Fig. 4.112 Change of feature values for AE test source measurements with different distance
between source and sensor (a) and experimental dataset partition as obtained from pattern
recognition method from test on DCB specimen shown in Fig. 4.111 (b)

distance between 80 and 280 mm. In combination with investigations on typical


size geometries for mode I testing (cf. Sect. 4.7.4.1), this indicates that distinction
of sources operating at different frequency ranges should be possible for propaga-
tion distances up to 280 mm.
4.5 Signal Classification 267

a b
sensor width distribution

mean source-sensor
distance

AE-source

Fig. 4.113 Scheme for selection of source position and distribution to calculate cluster validity
indices as function of source-sensor distance (a) and as function of width distribution (b)

In order to evaluate the UoC error as function of source-sensor distance, the


double-cantilever beam shown in Fig. 4.111 was tested as reported in detail
in [154]. The x-position of the acoustic emission signals was localized using a Δt
localization procedure. In such an experiment, the majority of the acoustic emission
signals originate from positions close to the crack tip. Only few signals are localized
significantly ahead of or behind the position of the crack tip. This allows calculating
the distance to the detecting sensor. In addition, the depth position (z-position) of
the acoustic emission sources is expected to be dominantly around the medial plane
of the specimen, and the y-position is expected to be randomly distributed.
Two approaches were evaluated to investigate the influence of the source-sensor
distance on the UoC. As shown in the scheme in Fig. 4.113a, as the first approach
the signals were analyzed at constant width distribution as function of source-sensor
distance. As the second approach the source-sensor distance is kept constant and the
width distribution is varied. The subsets were chosen in stepwise increments of
500 signals until the full dataset size is reached.
For each subset of the whole dataset, Rousseeuw’s silhouette value and Hubert’s
Gamma coefficient were calculated based on the features Partial Power 1, Partial
Power 2, Partial Power 4, Peak Frequency, and Weighted Peak Frequency as
defined in Table C.2 in Appendix C. The resulting UoC for both approaches is
shown in Fig. 4.114a, b, respectively.
For the shifted approach, no clear trend of the cluster validity measures or their
respective UoC is observed as a function of the source-sensor distance. The
cumulative approach reveals an increased UoC with increasing width distribution
of the source positions. This leads to the conclusion that the source-sensor distance
itself is not a critical parameter for valid source identification within the investi-
gated range below 375 mm. In contrast, the width distribution of sources within the
specimen seems to influence the accuracy of the source identification procedure
significantly. For these experiments, an increased width distribution was found to
cause an increased UoC.
268 4 Acoustic Emission

a 3.0
UoC < 2.86 %

Shifted
2.5
uncertainty of classification [%]

2.0

1.5

1.0

0.5

0.0
150 200 250 300 350 400
mean source-sensor distance [mm]
b
3.0
Cumulative

2.5
uncertainty of classification [%]

UoC < 2.35 %

2.0

1.5

1.0

0.5

0.0
0 50 100 150 200 250 300
width distribution of sources [mm]

Fig. 4.114 Calculation of uncertainty of classification as function of distance for shifted (a) and
cumulative (b) approach

4.5.4.2 Plate Thickness

Similar as for the source-sensor distance, a change in the propagation medium


geometry is likely to affect the UoC. In order to demonstrate the effect of thickness
changes, laminates used for bearing strength testing were investigated. The test
fixture for the 10.8 mm thick specimens is seen in Fig. 4.115. For the variation of
the specimen dimensions, an approximate ratio of 0.5:1:1.5 was selected, which are
4.5 Signal Classification 269

5 mm

54 mm
load axis

y 108 mm WD sensor position


z
x

Fig. 4.115 Test fixture for smallest specimen size used for bearing strength tests including AE
sensor positions

applied to the lateral dimensions, the thickness of the specimen, as well as to the
diameter of the bolts. Thus the smallest specimen is of dimensions of
108 mm  54 mm  5 mm (height  width  thickness), the medium size is of
dimensions of 220 mm  110 mm  10 mm (length  width  thickness), and the
largest size is of dimensions of 330 mm  165 mm  15 mm
(length  width  thickness). This scaling of specimen dimensions inevitably
involves a change of the average source-sensor distance. Since for this test, the
position of damage spots is well localized behind the pins connecting the laminate
to the test fixtures, their width distribution remains rather narrow (<10 mm). Based
on the findings in Sect. 4.5.4.1, the lateral scaling is thus unlikely to contribute
much to the UoC. To allow for comparable propagation paths, only signals of one of
the sensors indicated in Fig. 4.115 are used in the following. Due to the bearing
failure mode observed in all of the specimens, it is also likely to generate equal
ratios of failure mechanisms for each of the three specimen dimensions. For signal
acquisition, a type WD sensor with 20 dBAE preamplification and a band-pass from
20 kHz to 1 MHz is used. Further details on the experimental setup and the data
acquisition parameters are found in [158, 159].
One representative partition of each specimen type as achieved by the pattern
recognition method following [125] is shown in Fig. 4.116. The respective assign-
ment to failure mechanisms is based on accompanying FEM modeling work
following the strategies presented in Sects. 4.2 and 4.7. Clearly, all three thickness
types show clusters of acoustic emission signals, which can be partitioned using
pattern recognition methods. Remarkably, the UoC is nearly identical for all three
specimen types. This gives rise to the conclusion that the thickness of the specimen
does not have a substantial impact on the uncertainty involved in the classification
procedure. However, larger discrepancies are expected, when such thickness
changes arise during signal propagation (see Sect. 4.3.2).

4.5.4.3 Stacking Sequence

As already seen from the studies on wave propagation in Sect. 4.3.2, the stacking
sequence has significant impact on the shape of the signal, even for identical
sources. Therefore it is necessary to investigate the impact of this parameter on
270 4 Acoustic Emission

the distinguishability of different AE source types. To this end three different


stacking sequences of Sigrafil CE1250-230-39 prepreg material were prepared.
Plates of 570  570 mm and 1.3 mm thickness were fabricated following the
production cycle recommend by the material supplier. As stacking sequence,

a
100
thickness 5 mm
Matrix cracking
80 Interfacial failure
Fiber breakage

UoC < 8.3 %


Partial Power 2 [%]

60

40

20

0
0 200 400 600 800 1000
weighted Peak-Frequency [kHz]

b
100
thickness 10 mm
Matrix cracking
80 Interfacial failure
Fiber breakage

UoC < 10.1 %


Partial Power 2 [%]

60

40

20

0
0 200 400 600 800 1000
weighted Peak-Frequency [kHz]

Fig. 4.116 Classification results of thickness 5 mm (a), 10 mm (b) and 15 mm (c) with similar
ratios to all axis (nearly quasi-isotropic) including calculated uncertainty of classification
4.5 Signal Classification 271

c
100
thickness 15 mm
Matrix cracking
80 Interfacial failure
Fiber breakage

UoC < 9.0 %


Partial Power 2 [%]

60

40

20

0
0 200 400 600 800 1000
weighted Peak-Frequency [kHz]
Fig. 4.116 (continued)

1.3 mm

1.3 mm 1.3 mm

in-plane source out-of-plane source


z
y
x

Fig. 4.117 Scheme of test configurations used including test source position and sensor positions

a unidirectional plate [03]sym, a cross-ply laminate [0/902]sym, and an quasi-


isotropic plate [0/60/-60]sym were used. To study wave propagation in these plates,
four different configurations were selected with an arrangement as shown in
Fig. 4.117. In every configuration two different test sources are used to generate
ten test signals. As the first test source, pencil lead breaks in the out-of-plane (OoP)
direction were selected, which cause signals with dominant A0 mode having mostly
low-frequency content. As the second test source type, the pencil lead breaks were
carried out at the edge of the plate. This in-plane (IP) direction at the middle plane
of the plate causes dominant S0 modes having substantially higher-frequency
content. To avoid influences of distance changes between source and sensor, the
test source position was kept constant. Every test source signal was monitored by
five type WD sensors arranged in a quarter circle around the source position at a
distance of 100 mm. Prior to testing it was assured that the sensors show similar
272 4 Acoustic Emission

a
100
90°
90 67,5°
45°
80 22,5°

70 average value Fiber
Partial Power 2 [%]

90°
60
50

40
30
20
10
0
100 150 200 250 300 350 400 450 500 550
Weighted Peak-Frequency [kHz]

b
100
90°
90 67,5°
45°
80 22,5°

70 average value Fiber
Partial Power 2 [%]

90°
60
50

40
30
20
10
0
100 150 200 250 300 350 400 450 500 550
Weighted Peak-Frequency [kHz]

Fig. 4.118 Feature values as extracted for configuration 1 (unidirectional) for the IP-source (a)
and the OoP-source (b)

frequency response and sensitivity. For the unidirectional plate, the two fundamen-
tal orientations were investigated, since different test source signals are expected
based on the difference in stiffness along these two directions.
Feature extraction using a time window of 100 μs after signal arrival yields the
diagrams shown in Figs. 4.118, 4.119, 4.120, and 4.121. The angles noted through-
out Figs. 4.118, 4.119, 4.120, and 4.121 refer to the orientation of the sensor relative
to the source position, whereas the orientation of the different plies is indicated by
4.5 Signal Classification 273

a
100
90°
90 67,5°
45°
80 22,5°

70 average value Fiber
90°
Partial Power 2 [%]

60

50

40

30

20

10

0
100 150 200 250 300 350 400 450 500 550
Weighted Peak-Frequency [kHz]

b
100
90°
90 67,5°
45°
80 22,5°

70 average value Fiber
90°
Partial Power 2 [%]

60

50

40

30

20

10

0
100 150 200 250 300 350 400 450 500 550
Weighted Peak-Frequency [kHz]

Fig. 4.119 Feature values as extracted for configuration 2 (unidirectional, rotated) for the
IP-source (a) and the OoP-source (b)

the fiber arrows. For the configuration 1 (unidirectional plate, source parallel to
fiber axis), the result is given in Fig. 4.118a, b. A first comparison of these two
figures clearly indicates the difference as expected from the different test sources.
For the case of the IP-source shown in Fig. 4.118a, the detected frequency content
expressed by the feature values Partial Power 2 and Weighted Peak Frequency is
274 4 Acoustic Emission

a
100
90°
90 67,5°
45°
80 22,5°

70 average value Fiber
90°
Partial Power 2 [%]

60

50

40

30

20

10

0
100 150 200 250 300 350 400 450 500 550
Weighted Peak-Frequency [kHz]

b
100
90°
90 67,5°
45°
80 22,5°

70 average value Fiber
90°
Partial Power 2 [%]

60

50

40

30

20

10

0
100 150 200 250 300 350 400 450 500 550
Weighted Peak-Frequency [kHz]

Fig. 4.120 Feature values as extracted for configuration 3 (cross-ply) for the IP-source (a) and the
OoP-source (b)

clearly different for the different sensor positions. Moreover, the feature range
covered by the individual sensors spans the typical feature ranges also used to
sort out failure types based on previous work [24, 125, 138, 160]. For the
OoP-source results seen in Fig. 4.118b, this effect is less pronounced, but still
visible due to the formation of individual clouds for each sensor. It is worth noting
4.5 Signal Classification 275

a
100
90°
90 67,5°
45°
80 22,5°
0° Fiber
70 average value
90°
Partial Power 2 [%]

60

50

40

30

20

10

0
100 150 200 250 300 350 400 450 500 550
Weighted Peak-Frequency [kHz]

b
100
90°
90 67,5°
45°
80 22,5°
0° Fiber
70 average value
90°
Partial Power 2 [%]

60

50

40

30

20

10

0
100 150 200 250 300 350 400 450 500 550
Weighted Peak-Frequency [kHz]

Fig. 4.121 Feature values as extracted for configuration 4 (quasi-isotropic) for the IP-source (a)
and the OoP-source (b)

that for each sensor, there is distinct difference in the cluster position for the
IP-source and the OoP-source. Therefore, source discrimination would be still
possible if only signals of one sensor are investigated at a time. The problem of
overlapping cluster ranges solely arises, if signals of all sensors are mixed and are
interpreted at the same time. In order to overcome this problem, it is possible to
calculate average feature values for each source event.
276 4 Acoustic Emission

If signals are known to originate from the same source and are detected by
N sensors, it is straightforward to calculate the average feature value hqi:
XN
h qi ¼ i¼1
qi ð4:55Þ

These average feature values are superimposed as black dots in Fig. 4.118a, b,
respectively. These cluster positions of q are found well separated and therefore
will allow meaningful source type distinction.
Identical investigations on the remaining three cases yield Figs. 4.119, 4.120,
and 4.121 comparing the results of the IP-source and the OoP-source. For config-
uration 2 (unidirectional plate, source perpendicular to fiber axis), similar behavior
as for configuration 1 is observed. However, the average frequency content of the
IP-source is found to be significantly lower, which is caused by the source mech-
anism rather than the orientation aspect. Therefore, the final separation of the
averaged features seems to be less pronounced.
For the configurations 3 (cross-ply) and 4 (quasi-isotropic), the decreasing
degree of acoustic anisotropy causes less distinctly separated cluster for each of
the detecting sensors. Nevertheless, the basic findings established for configurations
1 and 2 are still found to be valid.
Based on these findings, it seems to be possible to distinguish different failure
types in the stacking sequences investigated above. More generally, the distinction
based on frequency content should always be possible if sources are monitored at the
same propagation angle (i.e., the different sources are subject to the same propagation
path). If this condition cannot be fulfilled (i.e., a technical structure with multiple
possible source positions), it is recommended to use the average feature values hNi of
a source event instead of the feature values of the individual sensors. As a second
requirement for the latter, a significant portion of orientations relative to the source
should be covered (i.e., spanning the range between 0 and 90 ).

4.5.4.4 Internal Damage

As discussed in Sect. 4.3.3 and demonstrated in Sect. 7.2.7, the interaction of


propagating waves with external or internal obstacles may have substantial influ-
ence on the detectable AE signals. Therefore it is straightforward to expect a
respective change in the feature values extracted from such signals. For acoustic
emission sources detected during mechanical loading of composites, this influence
is quite relevant, since there are lots of microscopic and mesoscopic failure process
within the propagation path between potential sources and the sensor positions until
final failure.
As demonstrated in [61, 62], it is useful to evaluate the changes of feature values
due to the presence of such internal damage. One example is the signal amplitude
shown in Fig. 4.122 for each configuration described in Sect. 4.3.3, evaluated in 0 ,
45 , and 90 propagation directions. Compared to the reference case (red), the
4.5 Signal Classification 277

1.0
0.9 Reference case
0.8 Inter-Fiber cracks
Broken Fibers
0.7
Inter-ply Delamination
0.6
0.5
amplitude [V]

0.4
0.3

0.07
0.06
0.05
0.04
0.03
0.02
0.01
0.00
0 5 10 15 20 25 30
configuration

Fig. 4.122 Calculated signal amplitude of signals for the various configurations of internal
damage in comparison to the reference case for 0 , 45 , and 90 propagation direction (reprinted
from [62])

signals passing through the damaged region experience significant changes of their
signal amplitude. This is due to the modal conversion at the damage position and
the scattering introduced by some of the damage types. Except for one of the
models with broken fibers, the overall deviation of amplitude values was found to
be within 0.25 V (i.e., 5 dBAE) to the reference case. Compared to the 3 dBAE
recommendation of ASTM E 2191 for sensor replacement, this amplitude range is
well within typically encountered uncertainties of measurement setups used.
The calculated Weighted Peak Frequency for the signals of different damage
configurations and the reference case are shown in Fig. 4.123 for 0 , 45 , and 90
propagation directions. Dependent on the propagation angle, different values for the
Weighted Peak Frequency are observed. This is caused by the anisotropic propa-
gation of Lamb wave modes in the unidirectional plate used in this modeling work
[61, 62]. Similar to the signal amplitude, deviations compared to the reference case
are up to 100 %. Again, features evaluated in all propagation directions are affected
by the presence of the modeled damage types.
As discussed in Sect. 4.5.4.3, the difference in feature values to the different
angles is expected from the anisotropy of the unidirectional plate studied in the
modeling work. For valid source identification, one can use the average frequency
features detected by a set of sensors as described in Sect. 4.5.4.3. The distribution
range of feature values as caused by the modeled damage types within the propa-
gation path is within the typical experimentally encountered distribution range for
an individual feature. Therefore, the proposed pattern recognition methods are
likely to deal with these changes [61, 62]. However, the presence of such internal
obstacles is expected to increase the mean distribution of feature values of the
278 4 Acoustic Emission

Reference case 90°


500 Inter-Fiber cracks
Broken Fibers
Weigted Peak Frequency [kHz] Inter-ply Delamination
400

300
45°

200


100

0
0 5 10 15 20 25 30
configuration

Fig. 4.123 Calculated Weighted Peak Frequency of signals for the various configurations of
internal damage in comparison to the reference case for 0 , 45 , and 90 propagation direction
(reprinted from [62])

cluster belonging to a particular failure type. Hence, with increasing amount of


damage, cluster structures may overlap to a significant amount and may cause an
increase in the UoC.

4.5.4.5 Sensor Type

As discussed in Sect. 4.4, the detection process of AE signals by an AE sensor may


have significant impact on the frequency content of the signal. The commercially
available types of AE sensors can be categorized as resonant, multi-resonant, or
wideband types. Naturally, resonant sensor types reduce the detectable frequency
spectrum, while broadband types are sensitive in a wide frequency range. Some
sensor designs show almost flat frequency response, minimizing the influence of the
sensor on the detection process [87–89].
To estimate the influence of the chosen AE sensor type on the outcome of
frequency-based signal classification, three different sensor types are chosen and
their results are compared. As typical representatives, the Digital-Wave 1025T
(wideband), the Physical Acoustics WD (multi-resonant), and the Physical Acous-
tics Micro30 (resonant) were chosen. The sensors were mounted during mode I
testing of Sigrafil CE1250-230-39 prepreg at identical locations. The according
experimental setup and data reduction routine to yield the graphs in Fig. 4.124 are
described in more detail in [154].
4.5 Signal Classification 279

Figure 4.124 shows the results of the proposed pattern recognition procedure as
scatterplots of Partial Power 2 over Weighted Peak Frequency with definition of
these features found in Appendix C. AE signals are assigned to matrix cracking,
interfacial failure, and fiber breakage following the FEM-based strategy as outlined
in Sects. 4.2–4.4 and as described in detail in [138].

a
100
Matrix cracking
90 Interfacial failure
Fiber breakage
80
Micro30-sensor
70
Partial Power 2 [%]

60 UoC < 19.6%

50

40

30

20

10

0
0 100 200 300 400 500 600 700
weighted Peak-Frequency [kHz]
b
100
Matrix cracking
90 Interfacial failure
Fiber breakage
80
WD-sensor
70
Partial Power 2 [%]

60 UoC < 1.2 %

50

40

30

20

10

0
0 100 200 300 400 500 600 700
weighted Peak-Frequency [kHz]

Fig. 4.124 Positions of signal clusters achieved by pattern recognition approach for three types of
AE sensors used with same type of experiment. Figures show resonant sensor type (a), multi-
resonant sensor type (b), and broadband sensor type (c) including calculated uncertainty of
classification
280 4 Acoustic Emission

c
100
Matrix cracking
90 Interfacial failure
Fiber breakage
80
1025T-sensor
70
Partial Power 2 [%]

(20 kHz high-pass)


60

50 UoC < 8.3%

40
30
20

10

0
0 100 200 300 400 500 600 700
weighted Peak-Frequency [kHz]
Fig. 4.124 (continued)

The AE signals detected by the Micro30 sensor shown in Fig. 4.124a are
centered in their Weighted Peak Frequency around the sensor resonant frequency
of 200 kHz. Still, the pattern recognition algorithm is able to distinguish between
three types of AE signals, but the clusters associated with matrix cracking and
interfacial failure show substantial overlap. Thus, the AE signals lose their discrim-
inative frequency information by the detection process with a narrow bandwidth
sensor type, which is expressed in a high UoC of 19.6 %.
In Fig. 4.124b the results of the same experiment using the WD sensor are
shown. Compared to the results in Fig. 4.124a, the values of the Weighted Peak
Frequency are spread out over a range of 500 kHz. Although the WD sensor
exhibits multiple resonant frequencies, these do not directly correspond to the
positions or number of the clusters. Instead, the concentration of AE signals around
distinct Weighted Peak Frequencies is influenced mostly by the specimen geometry
and the stacking sequence (cf. also [35]). The overlap of the clusters is found to be
less than for the resonant-type sensor yielding an UoC of only 1.2 %.
Initially, the AE signals detected by the broadband-type sensor 1025T did not
show any cluster structure. All feature values were located within a small band-
width of (150  100) kHz Weighted Peak Frequency and (5  5) % Partial Power
2. The result shown in Fig. 4.124c is the same dataset after postprocessing by a
20 kHz Butterworth high-pass filter of eighth order. Using this step, high sensitivity
of this sensor type at frequencies below 20 kHz is suppressed. The data structure
revealed is similar to that achieved by the other two sensors and can be classified
accordingly with an UoC of 8.3 %.
4.5 Signal Classification 281

It was found that the bandwidth of the sensor type used has significant influence
on the separability degree of the cluster structure. Thus the according error of
assignment to a failure mechanism will depend on the choice of sensor type as
well as on the suitability of the postprocessing steps.

4.5.4.6 Signal-to-Noise Ratio

AE signals of composite materials usually cover a broad range of signal amplitudes.


Even acquisition equipment with substantially high digitalization depth cannot
assure an equal signal-to-noise (S/N ) ratio for every AE signal. This raises the
question if feature values extracted from AE signals with poor S/N ratio are
different to those extracted from AE signals with good S/N ratio.
Therefore typical AE signals of each failure mechanism were chosen to inves-
tigate the influence of S/N ratio on the cluster structure. White noise Unoise was
superimposed to the AE signal UAE with increasing intensity as shown in
Fig. 4.125a:
0
e  Unoise ðtÞ þ UAE ðtÞ
U AE ðtÞ ¼ α ð4:56Þ

The noise level α was chosen relative to the maximum signal amplitude Umax of the
AE signal:

e ¼ U noise =Umax
α ð4:57Þ

The resulting AE signals based on one matrix cracking signal are shown in
Fig. 4.125a for five different noise levels α e. The noise levels were selected to
span the full range of S/N ratios encountered in AE testing. The resulting extracted
feature values are plotted in Fig. 4.125b as scatterplot of Partial Power 2 over
Weighted Peak Frequency. The dependency of the feature values on α is indicated
by an arrow starting at the original AE signal’s feature value.
One observation is the tendency of high α e -values to cause shifts of the feature
values to higher frequencies. For the Weighted Peak Frequency, this is observed
directly from Fig. 4.125, while the influence of the S/N ratio on Partial Power
2 originates from the negative correlation to the other Partial Powers. Thus for
low-frequency dominated AE signals (e.g., matrix cracking), Partial Power 2 values
increase with α e. Contrary, AE signals dominated by intermediate or high frequen-
cies (e.g., interfacial failure, fiber breakage) decrease their Partial Power 2 values
with αe, because Partial Powers 3 and 4 increase instead.
Another observation is the path of the trajectories for increasing αe-values. Even
for the extreme values of S/N ratio considered here, the feature values shown in
Fig. 4.125b reside inside the typical extent of respective signal clusters.
282 4 Acoustic Emission

a
120

100
offset signal voltage [V]

= 5.000
80

60
= 0.500
40
= 0.050
20
= 0.005

0 = 0.000

200 250 300 350 400 450 500 550 600


time [μs]
b
100
Matrix cracking
90 Interfacial failure
Fiber breakage
80
70
Partial Power 2 [%]

60
50
40
30
20
10
0
0 100 200 300 400 500 600 700 800 900
weighted Peak-Frequency [kHz]

e -values on
Fig. 4.125 Increase of S/N ratio for AE signal of matrix cracking (a). Influence of α
extracted frequency feature values (b)

4.6 Source Localization

Another important concept for interpretation of acoustic emission signals is the


localization of the acoustic emission source position. Once the acoustic emission
source is localized, its origin can be correlated with microscopic observations and
usually is called an acoustic emission event. For most of the algorithms used for
source localization, the calculation of the source position is based on the differences
of the arrival times between several sensor positions. As shown in Fig. 4.126,
4.6 Source Localization 283

Fig. 4.126 Scheme for


localization of acoustic difference in propagation
emission source positions length
based on differences in AE-source
signal arrival at different t0
t2
sensor positions (reprinted
from [35]) rSource r2

r1 t1
(0,0,0)

the elastic wave originating from an acoustic emission source radiates into the solid
volume and “arrives” at the respective sensor positions with distinct differences in
arrival time Δt ¼ t2  t1 .
A comprehensive review on state-of-the-art source localization procedures was
recently given by [52]. The purpose of this section is thus not to repeat the
substantial developments that have been carried out previously. Instead, the focus
of this section rests on the specific challenges that arise for the application of source
localization procedures in fiber reinforced materials and some recent developments
to address these needs.
A first challenge arises due to the type of wave propagation faced in most of the
composite applications. For source localization in volumetric media, the relevant
sound velocities are the longitudinal cL and transversal cT bulk wave velocities.
Although these velocities suffer from dispersion, their values can be assumed to be
reasonably constant within typical AE applications. But as extensively discussed in
Sect. 4.3, the acoustic waves found in thin platelike structures are guided waves.
For these Lamb-type guided waves, the propagation velocities are strongly depen-
dent on frequency (cf. Sect. 4.3) and therefore need to be considered explicitly.
Moreover, for these guided waves, every wave mode has an individual dependency
of propagation velocity as a function of frequency. Finally, the formation of such
guided waves reduces a 3D localization problem to a 2D localization problem, since
the depth position of the AE source cannot be localized based on differences in
arrival time at the AE sensors.
Especially in the case of composite plates, an acoustic anisotropy is introduced
by the orientation of the fiber reinforcements. Their dominating elastic modulus
results in highly anisotropic sound velocities and therefore is another significant
challenge for source localization. For the nonuniform stacking sequences typically
used in composites, these acoustic anisotropies can result in complex propagation
behavior of the acoustic emission signals and have to be taken into account for
localization and interpretation. An example of the acoustic anisotropy faced in such
fiber reinforced composites is given in Fig. 4.127.
Here the angular dependency of the group velocity of the fundamental Lamb
wave modes and the fundamental shear-horizontal mode is plotted for a frequency
of 250 kHz. The calculated group velocities of the S0 mode show strongest
284 4 Acoustic Emission

a b
S0-mode S0-mode
90° 80° A0-mode 90° 80° A0-mode
70° SH0-mode 70° SH0-mode
60° 60°
50° 50°

40° 40°

30° 30°

20° 20°

10° 10°

0° 0°
0 2000 4000 6000 8000 10000 0 2000 4000 6000 8000 10000
group velocity [m/s] group velocity [m/s]
[0/0/0/0/0/0/0/0] - 1mm [0/22.5/-22-5/90/90/-22.5/22.5/0] - 1mm
c d
S0-mode S0-mode
90° 80° A0-mode 90° 80° A0-mode
70° SH0-mode 70° SH0-mode
60° 60°
50° 50°

40° 40°

30° 30°

20° 20°

10° 10°

0° 0°
0 2000 4000 6000 8000 10000 0 2000 4000 6000 8000 10000
group velocity [m/s] group velocity [m/s]
[0/90/0/90/90/0/90/0] - 1mm [0/45/-45/90/90/-45/45/0] - 1mm

Fig. 4.127 Calculated sound velocities at 250 kHz for S0, SH0, and A0 mode in unidirectional
stacking (a), oriented cross-ply stacking (b), cross-ply stacking (c) and quasi-isotropic stacking
(d) of 1 mm thick CFRP plates

anisotropy for the unidirectional stacking sequence as seen in Fig. 4.127a. The
degree of anisotropy in the group velocities tends to vanish as more layers are
orientated in different directions as seen in Fig. 4.127b, c. For the quasi-isotropic
stacking sequence, the group velocity is almost equal to every angle as seen in
4.6 Source Localization 285

Fig. 4.127d. The A0 mode is less affected by the stacking sequence than the S0 mode
and shows almost isotropic nature for all of the stacking sequences. Only for the
unidirectional stacking sequence a slight elliptical group velocity profile is visible
in Fig. 4.127a.
While acoustic anisotropy and guided wave propagation are challenges closely
linked to fiber reinforced materials, even further challenges arise in realistic
structures made from these materials. These challenges comprise complex propa-
gation paths including reflections at boundaries and interfaces as well as changing
acoustic properties due to external stresses or due to growth of damage.
In the following, first the important step of onset picking is reviewed, since this
comprises one of the fundamental tasks in AE source localization. Subsequently,
the traditional way of Δt-based source localization is briefly introduced to act as
reference for new developments. In the last section, a recently developed approach
of AE source localization using neural networks is presented in some applications.

4.6.1 Determination of Signal Onset

The main experimental challenge in AE source localization is the exact determi-


nation of the arrival time of a transient AE signal. As depicted in Fig. 4.128a, the
first occurrence of a threshold crossing can be used as onset of the signal. The
so-obtained arrival time is obviously linked to the chosen threshold level and
consequently can falsify the calculated source coordinates. As alternative
approaches other methods for determination of the arrival time of elastic waves
were adopted from seismology in the past. The most important techniques to
mention are the autoregressive methods. The modified Hinkley criteria use the
position of the energetic minimum of transient signals at the onset [161]. An
application of the Hinkley criterion using two different scale parameters a is
shown in Fig. 4.128b. In contrast to the threshold-based techniques, this method
yields a more reliable determination of the signal onset. Another alternative is the
Akaike information criterion (AIC), which is obtained from the signals envelope
after applying a Hilbert transformation [162]. The result for the AIC criterion in
application to an acoustic emission signal is shown in Fig. 4.128c.
In comparison to the previous techniques, the accuracy for determination of the
signal onset is substantially increased. However, both autoregressive methods
benefit from previous knowledge of the approximate region of signal onset to
provide stable results. In this case, these methods yield the most reliable results
to determine the signal onset [161]. Further techniques focus on direct determina-
tion of time of arrival differences by inverse calculation of attenuation effects, by
beam forming [163] or by extracting arrival times at constant frequencies using
time-frequency analysis [164, 165]. Beyond these standard methods, various
approaches have been proposed, which are found in a comprehensive review on
the different established techniques in [52].
The difference in arrival time between two sensor positions can also be used to
filter signals based on the Δt-value. As an example of its usage, the frequently used
286 4 Acoustic Emission

test configuration of a tensile test is shown in Fig. 4.130. Here the goal is to interpret
AE signals originating from the reduced section of the specimen. In such test
configurations, failure of the grips and additional damage at the reinforcement
section may cause a substantial amount of acoustic emission signals. To avoid
occupying system resources in detecting these unwanted signals, a Δt-filtering
method can be used. As indicated in Fig. 4.129, all sources originating from a
certain part of the specimen fulfill the condition jΔtj  20 μs, which can be used to

a
0.6
AE signal

0.4
tB=29.8 μs

0.2
amplitude [V]

threshold B

0.0 threshold A

-0.2

-0.4
tA=26.7μs

-0.6
0 10 20 30 40 50
time [μs]
b
0.6
AE signal
Hinkley criterion a=5
0.4 Hinkley criterion a=10
t10=26.9 μs
0.2
amplitude [V]

0.0

-0.2

-0.4
t5=27.4μs

-0.6
0 10 20 30 40 50
time [μs]

Fig. 4.128 Determination of arrival time showing influence of threshold level (a) results of
modified Hinkley criterion (b) and results of AIC criterion (c)
4.6 Source Localization 287

c
0.6
AE signal
AIC criterion
0.4
tAIC =26.0 μs
0.2
amplitude [V]

0.0

–0.2

–0.4

–0.6
0 10 20 30 40 50
time [μs]
Fig. 4.128 (continued)

Fig. 4.129 Application of a b


Δt-filter to tensile test setup
detecting signals
symmetrically between the
sensors (a) and 1 1
t1 t1
|Δt| < 20μs

unsymmetrically between
-5μs < Δt < 20μs

the sensors (b) x x

t2 t2

2 2

acquire only signals out of the cross-hatched area. Using the same technique, one
can define areal sections for configurations with more than two sensors or define
unsymmetrical portions between sensors, if different Δt limits are defined in the
positive and negative direction (see Fig. 4.129b).

4.6.2 Classical Localization Methods

Based on the measured arrival time differences as discussed in Sect. 4.6.1, the
simplest approach to perform source localization is the zonal localization. For
undisturbed signal spreading, the sensor showing first arrival of the wave is closest
288 4 Acoustic Emission

to the source position. Therefore the vicinity of the sensor is selected as according
source “zone.” For larger objects, this may already give useful indications of areas
with high source activity and might be used to identify such spots. However, the
source position can rarely be determined with an accuracy higher than a few
centimeters. Typically the accuracy is expected to be roughly around half the sensor
spacing. Based on a set of arrival time values of different sensors, the source
position r0 and time of acoustic emission t0 can be calculated as:

j ri  r0 j ¼ c  ðti  t0 Þ: ð4:58Þ

This equation relates the positions of the different sensors ri and their signal arrival
times ti to the acoustic emission source location r0 using the sound velocity c. The
localization of acoustic emission signals in isotropic media is closely related to the
problems faced in GPS (Global Positioning System) applications. Consequently,
the corresponding algorithm of Bancroft is typically applied to obtain the acoustic
emission source position in a direct algebraic approach [166]. As an alternative,
algorithms adapted from seismology science are often applied, which use an inverse
approach. Here the specimen is subdivided into elements with finite resolution and
hypothetical runtimes are calculated for each element. The element showing suffi-
cient agreement with the measured runtimes is chosen as source position [52].
In general, (4.58) applies to arbitrary dimensions. Since the unknowns are the
source position r0 and the time of occurrence t0 for localization in N-dimensions,
N þ 1 sensors are required to solve (4.58). Thus, for a three-dimensional localiza-
tion, at least four sensors are necessary to obtain solutions. Typically more than
N þ 1 sensors are used to localize the source positions which results in overdeter-
mined equation systems. This additional information can be used to estimate the
source localization error or to improve the accuracy of localization by multiple
regression analysis.
In practice the localization accuracy depends on the sampling rate ΔT 1 used to
record the acoustic emission signals, the aperture of the sensors, the validity of the
assumptions made on signal propagation, and the uncertainty in determining the
signal arrival. The basic inaccuracy of the measured signal runtime can be calcu-
lated from the inverse sampling rate. This is directly correlated to a metric local-
ization error via multiplication of the sampling interval ΔT with the sound velocity.
The physical size of the sensor results in an additional inaccuracy of the determi-
nation of the arrival time caused by aperture effects. If the sensor radius is small in
comparison to the signal propagation length (e.g., if the sensor can be assumed as
point sensor), this error contribution can be neglected. For complex geometries and
anisotropic media, the largest errors arise from deviations in the assumed signal
propagation path. If the used algorithm does not take into account the reflections at
specimen boundaries, the assumed signal propagation velocity might be over- or
underestimated. In the case of planar specimens, two-dimensional localization of
acoustic emission sources generating Lamb waves can be treated in this respect by
4.6 Source Localization 289

using the propagation velocity of the S0 mode for localization. As depicted in


Fig. 4.127, for the frequency range of acoustic emission signals and reasonable
plate thickness, the S0 mode shows higher group velocity than the other Lamb wave
modes and thus arrives first at the sensor position.
When the propagation medium exhibits elastic anisotropy, (4.58) is no longer
valid and should be replaced by a respective approach for anisotropic sound
velocities. In general this boosts the complexity of the equations. For simple
two-dimensional localization problems like in CFRP plates, the varying ply layups
will always result in elastic anisotropy to a certain degree. A common approach is to
scale the geometry to the ratio of anisotropy in the two principal sound velocities.
Afterwards the common algorithms are used to infer the source location based upon
equation (4.58). A direct implementation for cases where a bidirectional velocity
characteristic is derived by rewriting (4.58) to yield:
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi vu
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
u 2
tðx0  xj Þ þ ðy0  yj Þ
2 2 2
ðx0  xi Þ ðy0  yi Þ
ti  t0 ¼ þ  ð4:59Þ
c2x c2y c2x c2y

Here xi, yi and xj,yj denote the x- and y-coordinates of the sensor positions on the
surface of the test object. The values xo and yo are the x- and y-coordinate of the
source position, and cx and cy are the sound velocity in the respective directions. As
a solver for this equation system, usually the Nelder-Mead-based algorithm is used.
One of the major uncertainties entering the calculation of the source position is
given by the extraction of the arrival time value from the signal as mentioned in
Sect. 4.6.1. For example, constant threshold values to detect the signal arrival time
depend substantially on the slope of the initial part of the signal as well as on its
amplitude and, therefore, may cause scatter of the so-obtained Δt-values. If the
signal amplitude is sufficiently high above the threshold level, the scatter in
Δt-values is dominated by the slopes of the acoustic emission signal. For signal
amplitudes close to the noise level, it is hardly possible to determine the correct
arrival time, which may cause even stronger scatter of the extracted values. Other
factors to take into account for the calculation of the uncertainty of localization are
the aperture of the sensor and the sampling rate of the system. Further factors
governing the accuracy of the method are the system sensitivity and the chosen
couplant. These are known to be relevant factors that may have a systematic
influence on the quality of the localization results, but are neglected in the follow-
ing, since these are difficult to quantify and are subject to the individual experiment.
The uncertainty of localization is also dependent on the sensor positions and the
position of the acoustic emission source. Source localization based on Δt-values
leads to a spatial variation of the localization sensitivity. This means that an
identical shift in position Δr at different locations on the test object will lead to
different shifts in the respective Δt-values.
The Δt-values of two sensors are directly obtained using formula (4.58). To
determine the source localization error, the gradient of the Δt-values may be
290 4 Acoustic Emission

Δ =
a 4 3 Δ =
Δ =
= =
Δ =
Δ =
Δ =

y=9 In general:
1 2 = =Δ

= x=8

b 4 3 4 3

Δ =

1 2 1 2

Fig. 4.130 Scheme of Δt-mapping source localization technique to establish Δt-map (a) and
during application (b)

calculated. This vector field is defined as gradient of formula (4.58) in all spatial
directions:
 
jri  r0 j
∇ð t i  t 0 Þ ¼ ∇ ð4:60Þ
c

This gradient of Δt-values also determines the maximum source localization


accuracy that can be achieved for a particular source position since it directly
affects the accuracy of the localization procedure. Because the errors discussed so
far are mutually uncorrelated, the resulting temporal and spatial errors may be
derived using the Gauss formalism to yield a final estimate for the uncertainty of
localization.
One strategy to overcome the problem of acoustic anisotropy in source locali-
zation procedures was proposed by Pullin, Baxter, and Eaton and is referred to as
Δt-mapping [167, 168] schematically shown in Fig. 4.130.
As schematically shown in Fig. 4.130a, for a structure with arbitrary geometry
and acoustic properties, first a subdivision into discrete zones is made. Given a
sensor network at fixed positions, for every discrete zone, a theoretical Δt-value is
calculated for every sensor combination. This effectively yields a Δt-map of the
structure. For the Δt-values of the experimentally detected signals, the closest
combination of Δt-values can then be estimated by automatic search algorithms
[167, 169, 170]. This provides an effective localization procedure for fiber
reinforced materials. However, the drawback of this approach is the necessity to
rely on theoretical Δt-values. Material imperfections, erroneous sound velocities,
attenuation, and reflections may cause distinct differences between theoretical
values and those Δt-values measured in a real structure.
4.6 Source Localization 291

4.6.3 Source Localization Methods Based on Neural


Networks

First recommendations have been made to use neural networks for source locali-
zation in geometrically complex metallic structures [171]. For the neural network
approaches proposed so far, the basic strategy involves two subsequent stages
(training and application) as shown schematically in Fig. 4.131 and is explained
in more detail in the next passages.
Similar to the Δt-mapping concept, artificial neural networks can be used to
establish a symbolic relationship between input data (Δt-values) and output values
(source position coordinates).
By conception, the application of a neural network can be subdivided in two
stages:
1. Training of a functional relationship based on Δt-values of known source
positions
2. Application of this trained relationship to Δt-values with unknown source position
For the first part (the training stage), Δt-values need to be extracted from signals
detected for known source positions. This can be achieved by experimental test
sources (see Sect. 4.6.3.1) or by respective modeling of the signals (see [172]).
While for the modeling approaches, a precise knowledge of the acoustical proper-
ties is necessary, the experimental approach using test sources does not even require
an explicit knowledge of the sound velocities in the material.
Based on either of these input data, a neural network can be used to approximate
a functional relationship to the source coordinates. To achieve such training of

a 4 3 Input nodes Output nodes


Δt12
Δt13 x
Δt14
y
Δt23

1 2

Δ =
=

b 4 3 4 3

Δ =

1 2 1 2

Fig. 4.131 Scheme of source localization technique applying neural networks including training
stage (a) and application stage (b)
292 4 Acoustic Emission

functional relationships, typically back-propagating algorithms such as the back-


propagation multilayer perceptron (MLP) or the SuperSAB are applied [136]. In
both cases, the training error is minimized by a gradient method using a feedback of
the resultant training error. In the following a review of the resilient back-
propagating MLP is given, since this was used for the work presented in
[158, 159, 173, 174].
The architecture of a MLP is a neural network as schematically shown in
Fig. 4.131a. The training of such neural network structures can be understood as
function approximation problem, where the “function” maps values given at the
input nodes to values at the output nodes. Since this correlation is generally a
nonlinear problem, multiple artificial neurons are typically used, arranged in a
number of hidden layers between input and output nodes (see Fig. 4.131a), power-
ful enough to allow approximation of arbitrary functions.
The response of a single neuron node hA to various inputs hEi with weights wi is
formulated using the activation function factivation as
X
hA ¼ f activation i
wi hEi : ð4:61Þ

The
X activation function is used to modify the linear decision boundary given by
w h and operates as decision criteria if the neuron is active or inactive. This
i i Ei
can either be realized by a threshold function (hard-limit function) or by continuous
(linear or sigmoidal) functions. The functionality of the neural network now is the
adaptive adjustment of the node weights wi to generate an optimized mapping between
the input layer and the output layer. This is done by adjusting the weights of the nodes
in order to find a global minimum of the quadratic error surface function L: e

XOtrain X O  2
eðwÞ ¼ 1
L qref , j  qj ð4:62Þ
2 i¼1 j¼1

Here w represents the vector of all node weights, qref,j the output value of the
reference objects, and qj the present output value. The dependency of the error
surface function on wtot is implicitly given in (4.62), since the node weights strictly
depend on the values of qj. The back-propagating multilayer perceptron updates
these weights w iteratively based upon a gradient search of the global minimum
with a learning rate ι:
e ðw Þ
∂L
wði þ 1Þ ¼ wðiÞ  ι : ð4:63Þ
∂w

In order to achieve convergence of this iterative process, a minimum in L eðwÞ is


necessary. As pointed out by Polikar, such minima of the error surface function can
either be local or global [136]. Thus, convergence of (4.62) is dependent on the initial
settings of the weights and does not necessarily converge to the global minimum.
4.6 Source Localization 293

Other typical problems faced in back-propagating MLPs are the appropriate


choices of the learning rate, the network structure, and the threshold level of the
error function. The latter one is the termination criterion which is used as the lower
boundary assuming that the network is sufficiently trained when this threshold level
is reached. While the learning rate and the network structure can often be deter-
mined empirically, a low threshold level of the error function can result in
overtraining of neural networks. Then the neural network tries to capture very
detailed the characteristics of the input objects.
After successful training of such neural networks, their ultimate strength is the
computationally fast reapplication to new input data. This is the second stage and is
known as application stage. Here the neural network is able to approximate source
positions based on new Δt-value combinations.
In the following a review is made of the implementation introduced in [158, 159,
173, 174] to yield a source localization technique particularly useful for fiber
reinforced materials. To this end, the next subsections present some aspects of
experimental test sources to generate input data. Subsequently, it is demonstrated
how source localization can be performed in highly anisotropic media such as
unidirectional fiber reinforced composites and how the influence of obstacles in
the propagation path is compensated. The section closes with some examples of
source localization in composite structures subject to mechanical loading.

4.6.3.1 Training Stage Using Test Sources

For the training stage, various experimental test sources can be used. These test
sources are applied at different positions on the test object to generate a dataset of
representative Δt-values. Therefore test sources need to be placed at sufficient
spatial density. Moreover, the test sources need to fulfill two further requirements:
1. At every test position, the signal of the test source needs to be detected by a
sufficient number of sensors.
2. The test source signals’ modal composition and frequency content need to be
representative for the experimentally detected signals in the application stage.
The most frequently used test source in AE is the pencil lead break source
(Hsu-Nielsen source) according to standard ASTM E976. The signals of pencil lead
breaks are typically of high intensity and are therefore useful to allow signal
detection also at sensors with far distance to the source position. However, pencil
lead breaks are usually performed in the out-of-plane direction and their bandwidth
is dominantly below 200 kHz. This causes a preferential stimulation of the funda-
mental antisymmetric mode in platelike structures as seen in a CWD in Fig. 4.132a.
Since various sources in fiber reinforced materials tend to stimulate signals with
dominant contributions of other than A0 modes, it is necessary to also take into
account their propagation behavior. As seen in the CWD in Fig. 4.132b, piezoelec-
tric pulsers can be used to excite fairly different signals in the same test structure.
Although being placed in the out-of-plane direction, they can stimulate also sym-
metric modes, as seen by the presence of the S0 mode in Fig. 4.132b.
294 4 Acoustic Emission

a 800
0.0
700 4.3E+06
8.6E+06
600 1.3E+07
frequency [kHz]

500 1.7E+07
2.2E+07
400 2.6E+07
3.0E+07
300
3.5E+07
200 3.8E+07

100
0
10
amplitude [V]

5
0
–5
–10
0 50 100 150 200
time [μs]
b
800 0.0
700 1.8E+02
3.5E+02
600
frequency [kHz]

5.3E+02
500 7.0E+02
8.8E+02
400 1.1E+03
1.2E+03
300
1.4E+03
200 1.6E+03

100
0
0.2
amplitude [V]

0.1
0.0
–0.1
–0.2
0 50 100 150 200
time [μs]

Fig. 4.132 AE signals of different artificial test sources evaluated at 60 mm source-sensor


distance. Graphs show Choi-Williams distribution for signal of top pencil lead break (a) and
piezoelectric pulser (b) detected by KRNBB-PC sensor

An alternative approach to the use of physically different test sources is to have a


type of test source, which can generate a broad range of frequencies. To this end,
piezoelectric pulsers operating within a reasonably broad frequency range can be
applied. Using a piezoelectric pulse with 20 ns rise-time, 1 μs width, and 10 V
amplitude applied to the conical type sensor described in Sect. 4.4.2, a KRNBB-PC
4.6 Source Localization 295

sensor is able to detect both fundamental Lamb wave modes with reasonable
intensities as shown in the CWD in Fig. 4.132b. After detection, it is possible to
use high- or low-pass filters to select particular frequency ranges of the signals to
generate different input signals as basis for the neural network training. As seen in
Fig. 4.133a, a seventh-order Butterworth high-pass filter of 400 kHz generates a

a
800 0.0
700 22
44
600
frequency [kHz]

66
500 88
1.1E+02
400 1.3E+02
1.5E+02
300
1.8E+02
200 1.9E+02

100
0
0.1
amplitude [V]

0.0

–0.1
0 50 100 150 200
time [μs]

b
0.0
700 1.8E+02
3.5E+02
600
frequency [kHz]

5.3E+02
500 7.1E+02
8.9E+02
400 1.1E+03
1.2E+03
300
1.4E+03
200 1.6E+03

100

0 20 40 60 80 100 120 140 160 180


0.2
amplitude [V]

0.1
0.0
–0.1
–0.2
0 50 100 150 200
time [μs]

Fig. 4.133 AE signals of piezoelectric pulser filtered by seventh-order Butterworth high-pass of


400 kHz (a) and low-pass of 200 kHz (b)
296 4 Acoustic Emission

signal dominated by the S0 mode showing a signal arrival at 52.6 μs, whereas a
seventh-order Butterworth low-pass filter of 200 kHz yields a signal dominated by
the A0 mode (cf. Fig. 4.133b), which has a signal arrival at 77.3 μs due to the lower
propagation velocity of this mode. Using such filtering, this effectively allows to
generate signals with frequency contents being representative for a broad range of
source types as faced in fiber reinforced materials and therefore allows to train the
neural network with a variety of input signals without the need for repetitive testing
at a single test source position.
An obvious alternative to test sources with broad frequency range is to filter the
detected signals of the application stage to yield a characteristic similar to those of
pencil lead break signals. Practically this means using low-pass filtering to yield only
the low-frequency portion of the signal. However, this may yield poor signal-to-noise
ratios for the signals that are dominated by frequencies outside the filtering range.
Further considerations are required with respect to the density of training points.
For a simple example as a plate structure, a homogeneous distribution of training
points might be expected to yield appropriate results. Judgment on the suitability of
the density and position of training points can be made based on the appearance of
Δt-maps. Here the Δt-values of a particular sensor pair is plotted as function of the
spatial coordinates. This is exemplarily shown for a set of laminates ranging from a
unidirectional stacking sequence to a cross-ply stacking and a quasi-isotropic
stacking in Fig. 4.134. Due to the different fiber orientations, a corresponding
change in the acoustic anisotropy may be expected as shown in the diagrams in

Fig. 4.134 Example of Δt-map for one sensor pair mounted on laminate plates with 400 mm edge
length as discussed in Sect. 4.6.3.2. Comparison is made between unidirectional stacking, cross-
ply stacking, and quasi-isotropic stacking
4.6 Source Localization 297

Fig. 4.127. This directly translates into the expected variance of Δt-values, which
are governed by the wave velocities of the material. As seen from the examples in
Fig. 4.134, the individual Δt-maps are reasonably smooth, and their steepness and
curvature is just owed to the degree of acoustical anisotropy in the laminate. Hence
the full character of the laminates may already be suitably approximated by subsets
of the training data points shown. This allows the reduction of the overall number of
training data points in these cases as will be shown in Sect. 4.6.3.2.
However, for a geometrically more complex structure, the ideal distribution and
position of training points is not so straightforward to conclude. In such cases, the
Δt-map may be used to detect divergences or areas with less smooth structures,
which might benefit from refinement. An example for such an object made from
fiber reinforced composites is shown in Fig. 4.135. This structural CFRP part
contains several sinks, holes, and steep curvatures as well as local changes in the
stacking sequence and the thickness.
As discussed in Sect. 4.3, all of these aspects influence the signal propagation
and thus will add to substantial deviations in the signal arrivals at the sensor
position. After applying a test source at several points on a test grid, the respective
Δt-map of the structure is obtained as shown in Fig. 4.135. Here only the Δt-value
of sensors 3 and 12 was used for visualization of the Δt-map. Due to the size of the
component, the overall arrival time difference spans a range of 250 μs. The Δt-map
does not resemble a topological map of the part, but instead it reveals the degree of
acoustic anisotropy as well as regions with strong discontinuities as seen close to

Fig. 4.135 Example of complex structural part made from CFRP with respective Δt-map
298 4 Acoustic Emission

the position of the holes. This allows to identify regions, which benefit from a
higher number of training points to be used for the neural network approach. Also,
the scatter of Δt-values at x-positions below 200 mm indicates that the detected
signals approach the level of the background noise, thus making the correct
evaluation of the signal arrival time more difficult.

4.6.3.2 Compensation of Acoustic Anisotropy

To demonstrate the capabilities of neural network-based source localization, an


investigation on carbon fiber reinforced laminates is presented. Plates of
570  570 mm edge length and 1.3 mm thickness were fabricated using three stacking
sequences. All plates are made from unidirectional epoxy prepreg Sigrafil CE1250-
230-39 and were cured according to the material supplier recommendations at 120 C.
The first plate is a quasi-isotropic laminate with [0/60/-60]sym stacking sequence, the
second is a cross-ply laminate with [0/902]sym stacking, and the third is a unidirectional
laminate with [03]sym stacking. Therefore, the three plates show an increasing amount
of acoustical anisotropy due to the orientation of the fibers in the individual layers.
Using a rectangular training point grid of 11  11 points, a total number of
121 source positions is distributed across the plate in equal mutual distances of
5 cm. At every position a pencil lead break is performed to yield the training data
(i.e., to generate signals at known source positions). Four WD sensors mounted at
the corner of the plates are used to detect the AE signals using an acquisition rate of
10 MSP/s, 40 dBAE preamplification, and a band-pass filter ranging from 100 kHz
to 1 MHz. The arrival time of the AE signals is determined using the AIC.
In Fig. 4.136 the localization error of the classical methods based on (4.58) is
shown. The source localization error Δr is obtained from the calculated source
position r0 and the known test source position rtest as

Δr ¼ jr0  rtest j ð4:64Þ

The false-color range in Fig. 4.136 is used to represent the source localization error.
As clearly seen in the figures, the classical method suffers from the increasing
amount of acoustical anisotropy. While the quasi-isotropic laminate still allows a
successful localization of all test signals, this is not the case for the cross-ply
laminate and the unidirectional laminate. Here a decreasing number of only
37 out of 121 source positions can be calculated within the dimensions of the
cross-ply laminate. For the unidirectional laminate, this is reduced even further to
6 out of 121 source positions. Also the average source localization error increases
substantially. Especially in the areas outside the rectangle spanned by the four
sensors, the source localization accuracy is very poor.
For the neural network-based approach, 62 test positions are chosen as training
dataset. An artificial neural network using two hidden layers and ten neurons is applied.
As activation function of the neurons, a sigmoidal hyperbolic tangent function is
defined. For every laminate the trained neural network allows a localization of all
4.6 Source Localization 299

a
delta-t (classical) result neural network result
600 600
0.000
15.00
500 500
30.00
45.00
y-coordinate [mm]

y-coordinate [mm]
400 400 60.00
75.00
90.00
300 300 105.0
120.0
135.0
200 200 150.0

[mm]
100 100

0 0
0 100 200 300 400 500 600 0 100 200 300 400 500 600
x-coordinate [mm] x-coordinate [mm]
b
delta-t (classical) result neural network result
600 600
0.000
3.000
500 500
6.000
9.000
y-coordinate [mm]

y-coordinate [mm]

400 400 12.00


15.00
18.00
300 300 21.00
24.00
27.00
200 200 30.00

[mm]
100 100

0 0
0 100 200 300 400 500 600 0 100 200 300 400 500 600
x-coordinate [mm] x-coordinate [mm]
c
delta-t (classical) result neural network result
600 600
0.000
15.00
500 500
30.00
45.00
y-coordinate [mm]

y-coordinate [mm]

400 400 60.00


75.00
90.00
300 300 105.0
120.0
135.0
200 200 150.0

[mm]
100 100

0 0
0 100 200 300 400 500 600 0 100 200 300 400 500 600
x-coordinate [mm] x-coordinate [mm]

Fig. 4.136 Comparison of AE source localization accuracy for classical localization (left) and
neural network-based approach (right) for quasi-isotropic laminate (a), cross-ply laminate (b), and
unidirectional laminate (c)
300 4 Acoustic Emission

121 source positions within the dimensions of the plate. Moreover, the source locali-
zation error is four to five times less than for the classical localization method, and the
source localization accuracy is almost identical in all parts of the plate.
This demonstrates the high flexibility of neural network-based source localiza-
tion with respect to acoustically anisotropic media. However, from a practical point
of view, the number of 62 test source positions used for the area of 60 cm  60 cm is
unnecessarily high, since the Δt-fields of these plates do not exhibit strong gradi-
ents. Hence, similar source localization accuracy should be expected for systemat-
ically less number of training data positions. To demonstrate the impact of a
reduced number of training points, Fig. 4.137 shows the false-color representation
of the source localization error of the quasi-isotropic plate.

a b
26 training points 22 training points
600 600
0.000
10.00
500 500
20.00
30.00
y-coordinate [mm]

y-coordinate [mm]

400 400 40.00


50.00
60.00
300 300 70.00
80.00
90.00
200 200 100.0

[mm]
100 100

0 0
0 100 200 300 400 500 600 0 100 200 300 400 500 600
x-coordinate [mm] x-coordinate [mm]

c 17 training points d 13 training points


600 600
0.000
10.00
500 500
20.00
30.00
y-coordinate [mm]

y-coordinate [mm]

400 400 40.00


50.00
60.00
300 300 70.00
80.00
90.00
200 200 100.0

[mm]
100 100

0 0
0 100 200 300 400 500 600 0 100 200 300 400 500 600
x-coordinate [mm] x-coordinate [mm]

Fig. 4.137 AE source localization accuracy for systematic reduction of training data positions to
26 (a), 22 (b), 17 (c), and 13 (d)
4.6 Source Localization 301

As seen in the source localization error indicated in the figures, a systematic


reduction of the number of training data points from 62 (cf. Fig. 4.129a) to 26, 22,
17, and 13 in Fig. 4.137 still yields a reasonable source localization accuracy. The
average source localization error was found to increase from 3.5 to 25.4 mm in the
case with only 13 training data positions.
So far, the results were obtained only for test source signals using pencil lead
breaks. But as previously discussed in Sect. 4.6.3.1, the typical AE signals of fiber
reinforced materials may show distinct differences in their modal composition
and frequency range, which is not sufficiently represented by pencil lead break
sources.
A suitable alternative is the usage of two different test sources as the piezoelec-
tric pulser and the pencil lead break at identical test source positions. Subsequently
the neural network is trained using the same approach as before but with one
additional input node. This input node is set to 0 for the pencil lead break input
data and to 1 for the input data of the piezoelectric pulser. As a consequence, the
neural network virtually forms two branches to deal with the source localization
task of the one or the other set of input data. Figure 4.138 presents the resulting
source localization error using this approach. In both cases, a combined set of two
times 45 training data points was used. Figure 4.138a shows the application of this
trained neural network to the signals of the piezoelectric pulser only and Fig. 4.138b
to the signals of the pencil lead breaks only. Clearly in both cases, the neural
network provides similar source localization accuracy being mostly dominated by
the number of training data points rather than the presence of two different sets of
input data.

a b
piezoelectric pulser pencil lead break
600 600
0.000
6.000
500 500
12.00
18.00
y-coordinate [mm]
y-coordinate [mm]

400 400 24.00


30.00
36.00
300 300 42.00
48.00
54.00
200 200 60.00

[mm]
100 100

0 0
0 100 200 300 400 500 600 0 100 200 300 400 500 600
x-coordinate [mm] x-coordinate [mm]

Fig. 4.138 AE source localization accuracy for quasi-isotropic laminate with two times 45 train-
ing data points based on simultaneous training data of piezoelectric pulser and pencil lead break
source applied to data of piezoelectric pulser (a) and applied to data of pencil lead break (b)
302 4 Acoustic Emission

4.6.3.3 Discontinuous Δt-Fields

Other than the plates studied in Sect. 4.6.3.2, realistic structures usually show more
geometrical complexity. This may involve curvatures, internal interfaces, holes,
supports, or other factors, which affect the way of acoustical propagation. All these
disturbances in acoustical propagation can also be interpreted as disturbances in the
arrival time differences Δt. Therefore, the Δt-field may start to exhibit steep
gradients or may start to show discontinuities. Overall, the quality of the neural
network-based source localization approach is directly linked to the approximation
quality of this Δt-field.
To introduce some further complexity to the source localization problem, the
specimen geometry typically used for bearing strength test of bolted laminates is a
suitable candidate. T700/PPS specimens of 220 mm  110 mm  10 mm
(length  width  thickness) with 18 mm diameter bolts were used to this end.
During the test, four type WD AE sensors were mounted on the specimens using
medium viscosity couplants and clamp systems. AE signals were acquired using
10 MSP/s acquisition rate at 40 dBAE preamplification and a band-pass ranging
from 20 kHz to 1 MHz. The complex situation of AE signal propagation encoun-
tered in this test environment is shown in Fig. 4.139 in an exemplary calculation
result using finite element simulations based on the implementation described in
Sect. 4.2.
In this test configuration, the presence of a bolt in the laminate causes a
pronounced discontinuity in the Δt-field. Following the previously outlined
method, a neural network with four neurons and three hidden layers with sigmoidal
activation functions is used. Therefore, all effects of the disturbed wave propaga-
tion are accounted for in the training stage of the neural network.

a b
source position

z signal inside fixture


y
low-reflecting x
t = 10 μs t = 20 μs
boundary

Fig. 4.139 FEM result of acoustic emission wave field of modeled fiber breakage failure at
t ¼ 10 μs (a) and at t ¼ 20 μs after source excitation
4.6 Source Localization 303

a b
delta-t (classical) result neural network result
deviation to test source
position [mm]
200 200
0.000
3.000
6.000
9.000
12.00
150 150 15.00
18.00
y-coordinate [mm]

y-coordinate [mm]
21.00
24.00
27.00
30.00
100 100

50 50

0 0
0 50 100 0 50 100
x-coordinate [mm] x-coordinate [mm]

Fig. 4.140 Comparison of AE source localization accuracy for classical localization (a) and
neural network-based approach (b)

A comparison between the source localization error of the classical method and
the source localization error of the neural network-based approach (based on
208 training positions) is shown in Fig. 4.140. As average of the whole specimen,
the classical method has a source localization accuracy of 14.5 mm. Out of the
210 test source positions, only 196 were localized within the geometry of the
specimen. Compared to that, the neural network approach has a mean source
localization accuracy of only 3.3 mm, and 209 out of 210 source positions were
localized within the specimen geometry.

4.6.3.4 Scale Invariance and Portability

Another interesting aspect of the neural network-based source localization is the


ability to scale existing training data to similar objects of larger size. To demon-
strate these capabilities, the specimens of the bearing strength test are used. Here,
three types of test specimens with different dimensions are investigated. The
smallest specimen investigated has dimensions of 108 mm  54 mm  5 mm
(length  width  thickness), the medium size dimensions of 220 mm 
110 mm  10 mm (length  width  thickness), and the largest size dimensions of
330 mm  165 mm  15 mm (length  width  thickness). Starting with the
304 4 Acoustic Emission

small medium large AE source density


high

medium

low

54 mm

110 mm

165 mm

Fig. 4.141 Exemplary localization results by neural network-based technique using training data
of the medium-sized specimen

smallest size, the diameter of the bolt is 9 mm with an increase of a factor of 2 and
3 for the medium and large size, respectively. Test signals of pencil lead breaks are
detected as described in Sect. 4.6.3.3.
To investigate the scalability and portability of trained neural networks, only one
database will be used for the results shown in the following. The training data was
generated using the medium size plate. Since the three different specimen types vary
in all dimensions by a factor of 2 and 3 relative to the smallest plate, the Δt-values of
the training dataset are multiplied by a factor of 0.5 and 1.5, respectively.
The comparison in Fig. 4.141 shows the AE source positions as localized during
a bearing strength test. For the small and large cases in Fig. 4.141, the localization
results are obtained for a neural network based on the training data obtained on the
medium specimen size. The obtained source localization error for the individual
training datasets of each specimen type was found to be almost identical to the
source localization error of the scaled training datasets.
Similar to the scaling procedure of the training dataset, it is also feasible to
transfer the trained neural networks from specimen to specimen with identical
geometries. This is particularly relevant for testing of specimen series, where no
substantial changes of the material, the geometry, the sensor positions, and the
types occur. In such cases a direct reapplication of a trained neural network is an
efficient way to perform fast source localization.
4.7 Application to Composites 305

4.7 Application to Composites

Based on the possibility to detect the occurrence of all sorts of microscopic failure
mechanisms in composite materials, there is a broad range of applications for the
acoustic emission method. In this section, a focus to monitoring of materials testing
under laboratory conditions is given. However, it is attractive to extend the presented
approaches to structural health-monitoring applications. This comprises a field of
research on its own and is therefore not covered in this book. For more details on this
part, the reader is referred to other literature dealing with this topic [52, 175, 176].

4.7.1 AE Source Identification Using FEM Results

In this section, a methodology to compare experimental data to FEM results is


reviewed. The modeling procedure to compute acoustic emission signals used for
this purpose is extensively described in Sects. 4.2–4.4 and is not repeated in
detail here.
In general, such FEM investigations can be performed for different reasons. In
any case it is required to have a validated modeling environment as presented in the
previous sections. To allow source identification based on modeling results, one can
use different approaches. As schematically shown in Fig. 4.142, the intuitive way is

a
Experimental signal Modeled signal
Amplitude [V]
Amplitude [V]

0.2 0.2
0.1 0.1
0.0 0.0
-0.1 -0.1
-0.2 -0.2

0.00 0.02 0.04 0.06 0.08 0.10 0.00 0.02 0.04 0.06 0.08 0.10
Time [ms] Time [ms]

Feature extraction Feature extraction

b
feature 1 feature 1

feature 2 feature 2
Experimental partition Modeled partition

Fig. 4.142 Scheme of data processing to compare experimental AE data and FEM-based AE data
based on signals (a) and partitions (b)
306 4 Acoustic Emission

to assume a certain source type at a position within a solid and directly compute the
respective signal in a forward approach. Being demonstrated in [13, 35, 87], this
approach can yield quantitative prediction of AE signals if all details of the
experiment are known. In particular, the accurate source position of an experimen-
tal signal is usually hard to obtain, since source localization approaches show some
uncertainty. Moreover, for platelike structures, the depth position of the source can
usually not be retrieved by the source localization approach (see Sect. 4.6).
Because for composite materials, many acoustic emission signals are detected
during the experiment, one can also address the problem of source identification by
a statistical approach.
As first presented in [177], the procedure is carried out in three subsequent steps.
The volume of interest is defined and modeling of different source types is carried
out at representative positions. With reference to a simple tensile test of a unidi-
rectional specimen discussed in [178], the consideration would be as follows:
• The volume of interest is given by the tapered area of the specimen.
• Failure mechanisms are assumed to occur everywhere within the volume of
interest.
For cross-ply specimens on the other hand, preferential depth positions of inter-
ply delamination arise at the interfaces between the plies. Therefore modeling of
this failure type should be considered focused at those depth positions. If notches
are present, these likely act as stress concentrator and may initiate all sorts of failure
mechanisms in their vicinity. Then the volume of interest has to be adapted
accordingly. After the modeling work is completed, a set of waves due to a variety
of source positions and different failure types is available.
The second step is to apply the same feature extraction procedure to the modeled
signals as for the respective experimental signals. This yields a partition of acoustic
emission signal features with known origin and is indicated as modeled partition in
Fig. 4.142b.
Accordingly, the third step then consists in comparison of the feature value
positions of the modeled signals and the experimental signals. This procedure has
been demonstrated in various publications [24, 35, 125, 138, 177, 179] and is
demonstrated in the following for one example of a tensile test as described in
[178]. All acoustic emission signals were acquired using the settings reported in
Table 4.2 with sensor positions at the end of the tapered area of the specimen.
Figure 4.143a shows the partition of acoustic emission signals obtained by
unsupervised pattern recognition for one representative specimen. To visualize
the position of the signal clusters, a projection to the features Weighted Peak
Frequency and Partial Power 2 was used. For the experimental data, the clusters
are well defined, but their edges are close together. As discussed in Sect. 4.5, this
may cause an uncertainty in the assignment of the signals to the respective cluster.
Accordingly, simulations for the failure mechanisms as described in Sect. 4.2.2.2
were performed for the type of tensile specimen considered. For the simulated AE
signals, the extracted feature values are plotted in Fig. 4.143b.
4.7 Application to Composites 307

Table 4.2 Acquisition settings and postprocessing parameters used for tensile tests
Acquisition Postprocessing
settings Value parameters Value
Preamplification 20 dBAE Feature extraction 100 μs after
window threshold
Threshold 35 dBAE Δt-filter (7  2) μs
Triggering Individual channel
Acquisition rate 10 MS/s
Band-pass range 20 kHz to 1 MHz
Couplant Medium viscosity silicone
grease
Mounting system Clamp
Sensor type WD
Number of 2
sensors

The signals simulated for fiber breakage are well separated from the rest of the
signals. Compared to the experimental data, the simulated signals show slightly
higher frequency contributions, but overall coincide with the experimental feature
range.
The feature values extracted from modeled signals for inter-fiber failure and
inter-ply delamination are observed within similar ranges as for the experimental
data. Signals originating from inter-fiber failure with dominant crack opening
(mode I) components are found with lowest “Partial Power 2.” The simulations
of inter-fiber failure with significant shear (mode II) contributions are found with
higher “Partial Power 2” values up to 40 %. The variation of the orientation of the
fracture plane causes variability in the absolute values as seen in Fig. 4.143b, but
does not substantially change the extracted feature values.
For inter-ply delamination, a small overlap to the feature value ranges of inter-
fiber failure is found. This is in very good agreement to the experimental observa-
tions. However, the separation to signals associated with fiber breakage is much
more pronounced in the simulation data than in the experimental data. One likely
explanation is the exclusion of the signal-to-noise effects as described in
Sect. 4.5.4.6. Since the signals of 1-plet fiber breakage is of small amplitude, the
presence of noise will likely add to a larger variability of feature values in the
experimental data than currently observed in the modeled data.
As example to demonstrate the relation between source position and feature
ranges, the feature trajectory for a change in source-sensor distance is marked for
one case of inter-ply delamination. Here the same source is modeled at different
source-sensor distances, and the feature values are extracted consistently. This
evidently causes a substantial shift in the values of Partial Power 2 and Weighted
Peak Frequency. However, the length and shape of the trajectories are typically
responsible for the cluster dimensions and do not result in crossing between clusters
for the example of Fig. 4.143.
308 4 Acoustic Emission

a
Experiment
Matrix cracking
60 Interfacial failure
Fiber breakage
Partial Power 2 [%]

40

20

0
0 200 400 600 800 1000 1200
weighted Peak Frequency [kHz]
b
Simulation
Inter-fiber failure, all angles
60 Inter-ply delamination
Fiber breakage
Partial Power 2 [%]

40
source-sensor distance

20

0
0 200 400 600 800 1000 1200
weighted Peak Frequency [kHz]

Fig. 4.143 Comparison between feature values extracted from experimental (a) and simulated (b)
AE signals

Based on these findings, it is thus possible to conclude that the clusters detected
by the pattern recognition approach allow meaningful distinction between the
occurrences of three types of signal classes. In the modeling part, these were strictly
termed as fiber breakage, inter-fiber failure, and inter-ply delamination. In previous
work it was demonstrated that effects such as crack formation in pure matrix
material also fall within the feature ranges observed for the class of inter-fiber
failure. Moreover, other interfacial failure types such as fiber pullout and
4.7 Application to Composites 309

intralaminar delamination were observed to fall within the feature range modeled
for inter-ply delamination. To account for this difference, the experimentally
derived signal classes are termed as fiber breakage, matrix cracking, and interfacial
failure in Fig. 4.143 and in the other parts of this chapter.
Beyond the nature of the AE source, other factors were previously discussed to
affect the feature range of the clusters and their overlap. In addition to the source-
sensor distance and the signal-to-noise ratio mentioned above, the ply layup,
complex 3D-geometries of the individual plies (e.g., fabrics), and the AE sensor
type will influence the quality of the partition. In many cases, the accumulation of
these effects will cause significant overlap of the clusters. For such cases, any
attempts to distinguish AE signals based on the proposed feature values using
unsupervised pattern recognition strategies are unlikely to yield meaningful
partitions of clusters. However, using the modeling strategies presented in
Sects. 4.2–4.4, it is possible to evaluate the severity of these effects. In particular,
the presented approaches were found to be sufficient to predict cluster overlaps and
merging of clusters, which constitutes the ultimate limit to perform source discrim-
ination using signal features.

4.7.2 Detection of Failure Onset

As already discussed for the application of DIC in Chap. 3, one of the main tasks in
the context of fracture mechanics is the detection of first failure onsets. The
following section demonstrates some applications, where acoustic emission mon-
itoring is used to provide better input quantities for mechanical testing in this
context. It seems natural to associate the onset of acoustic emission signals with
the onset of damage in a material. However, this direct association is only valid if
no noise sources are present in the experimental setup. But even for such ideal
laboratory conditions, detection of first acoustic emission is not identical to detec-
tion of first damage onset. Depending on the type of testing, it is also required to
detect the initiation of the failure mode relevant for the given mechanical test
procedure.

4.7.2.1 Apparent Interlaminar Shear Strength Tests

The measurement of the apparent interlaminar shear strength using 3-point bending
is one typical application requiring the detection of failure initiation. As previously
described in Sect. 3.4.2.1 at the moment of onset of interlaminar crack growth, the
maximum shear force occurs at the neutral axis of the beam. According to the test
standards, this onset is reached at the first load drop or at the maximum force
recorded during the test. The value for the apparent interlaminar shear strength is
then deduced from the force at onset Fonset, the thickness h of the specimen, and the
width t according to (3.20).
310 4 Acoustic Emission

However, there are several cases, where these assumptions are not fulfilled. One
possibility is to have failure at the load introduction points rather than at the neutral
axis. Accordingly, the load drop may not be interpreted as interlaminar failure.
Another possibility is plastic deformation prior to interlaminar failure. This also
causes signatures in the load-displacement curve, which may not be interpreted as
failure onset. Moreover, the real onset of interlaminar failure may likely precede
any signatures visible in load-displacement curves.
As demonstrated by [180], a combination of the short-beam shear test with
acoustic emission monitoring is a suitable approach to increase the reliability of
the test. Figure 4.144 shows test results of a specimen made from Sigrafil CE1250-
230-39 prepreg subject to a short-beam shear test according to DIN EN 2563. For
acoustic emission monitoring, a type WD sensor was mounted on the load nose
fixture, and all signals were acquired using the acquisition settings given in
Table 4.3. As seen in Fig. 4.144, there is a clear signature in a load-displacement
curve indicated by Fvisible. The maximum of the curve appears at later stages and is
obviously not the onset of damage. As revealed by the superimposed curve of

500
WD sensor
1600
Accumulated number of AE signals

450
1400
Fvisible 400
1200 350
Force
Force [N]

1000 FAE 300

800 AEonset 250

200
600
150
400 neutral axis
100
200
50 y
0
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
20 mm x
displacement [mm]

Fig. 4.144 Exemplary load-displacement curve superimposed by detected acoustic emission


signals (left) and according scheme of experimental setup (right)

Table 4.3 Acquisition Acquisition settings Value


settings used for short-beam
Preamplification 40 dBAE
shear tests
Threshold 35 dBAE
Triggering Individual channel
Acquisition rate 10 MS/s
Band-pass range 20 kHz to 1 MHz
Couplant Medium viscosity silicone grease
Mounting system Clamp
Sensor type WD
Number of sensors 1
4.7 Application to Composites 311

accumulated acoustic emission signals, a first onset of damage is detected at a much


lower load level FAE. Compared to the possibility of DIC monitoring as described
in Sect. 3.4.2.1, the possibility to leave the acoustic emission sensor permanently on
the test fixture seems advantageous.

4.7.2.2 End-Notched Flexure Tests

A test configuration frequently used in fracture mechanics of fiber reinforced


polymers is the end-notched flexure test to measure the mode II fracture toughness.
As described previously in Sect. 3.4.2.2, the aim of this test is also to determine the
load at the moment of onset of interlaminar crack growth. It has already been
demonstrated in Sect. 3.4.2.2 that it is not absolutely necessary to accompany tests
of fiber reinforced polymers with brittle epoxy matrix systems. These materials tend
to fail very brittle in mode II loading, and the first load drop was found to coincide
well with the first occurrence on interlaminar crack growth. However, this test
procedure becomes particularly more challenging, when ductile matrix materials,
such as polyphenylensulfide (PPS), are tested.
To demonstrate the deviations to the failure behavior expected by the established
test standards, ENF tests using carbon fiber reinforced PPS laminates were
performed [181]. Specimen geometries in accordance with ASTM D7905 were
prepared and tested including acoustic emission monitoring as seen in Fig. 4.145.
All acoustic emission signals were acquired using the acquisition setting given in
Table 4.4 using the sensor position indicated in Fig. 4.145.
The detected signals are analyzed by the pattern recognition method described in
[125], which results in three natural clusters that can be assigned to three types of
failure mechanisms as indicated in Fig. 4.146.

z marker for WD sensor


Compliance calibration force
3.5mm
3mm

x
specimen

crack tip

130 mm

initial crack length 52 mm

Fig. 4.145 Test configuration used for mode II ENF testing in combination with acoustic
emission detection
312 4 Acoustic Emission

Table 4.4 Acquisition settings and postprocessing parameters used for mode II ENF tests
Acquisition Postprocessing
settings Value parameters Value
Preamplification 40 dBAE Feature extraction 100 μs after
Threshold 35 dBAE window threshold
Triggering Individual channel
Acquisition rate 10 MS/s
Band-pass range 20 kHz to 1 MHz
Couplant Medium viscosity silicone
grease
Mounting system Clamp
Sensor type WD
Number of 1
sensors

Interpreting the load-time curves in accordance with standard ASTM D7905


using the force maximum as value for Fonset yields the values of GIIc as noted in
Table 4.5. These differ to existing values in literature of 802–933 J/m2 [182, 183] by
a factor of 4. However, clear nonlinearity is observed prior to the load maximum. At
the same time, a significant amount of acoustic emission signals are detected as
seen in Fig. 4.146b. This indicates the onset of subcritical crack growth and
therefore suggests the usage of the acoustic emission onsets to deduce the onset
of crack growth.
If no distinction is made between different failure types, the evaluation of the
overall onset of all acoustic emission signals yields a very conservative value of
647 J/m2 having a coefficient of variation of 36.1 %. However, the significance of
the acoustic emission analysis can be increased substantially by distinction of
different failure types.
For the onset of interlaminar crack growth, the relevant class of acoustic
emission signals is interfacial failure. Evaluation of the specific onset of this class
of signals yields a value of 980 J/m2 with a significantly reduced coefficient of
variation of 18.8 %. This value was found to be more reasonable given the reports in
literature [182, 183] and the typically expected range of mode II fracture toughness
values for fiber reinforced polymers.

4.7.2.3 Transverse Crack Tension Tests

A further example of a test procedure based on valid detection of damage onsets is


the transverse crack tension test. Standard tensile test specimen geometries with a
tapered area and tab reinforcement are usually used. As a major modification, one
of the center plies is fabricated discontinuous, i.e., exhibits a cut transverse to the
load direction. Due the applied tensile load, this translates into a shear component
acting between the inner plies at the discontinuity. Therefore, this refers to a mode
II load situation. Using the relation between the force at onset Fonset, the thickness
4.7 Application to Composites 313

a
50
Matrix cracking
Interfacial failure
40 Fiber breakage
Partial Power 2 [%]

30

20

10

0
0 100 200 300 400 500 600 700 800 900 1000
weighted Peak Frequency [kHz]
b
700
Force-displacement curve
250 Matrix cracking
accumulated number of AE signals

Interfacial failure 600


Fiber breakage
200 500
AEonset

force [N]
400
150

AEonset, interface 300


100
200

50
100

0 0
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5
displacement [mm]

Fig. 4.146 Partition of signals as obtained by pattern recognition method applied to signals
detected in ENF tests (a) and accumulated number of acoustic emission signals with superimposed
load-time curve (b)

h of the specimen, the thickness ht of one ply, the width b, and the modulus of
the specimen E, it is possible to calculate the mode II fracture toughness GIIc
according to
F2onset  ht
GIIc ¼ ð4:65Þ
4b2 E  hðh  ht Þ

As an example for this test procedure in combination with acoustic emission


analysis, test results of a unidirectional carbon fiber fabric made from Toray
T700 50C, 24k are presented in [143]. The transverse crack in the center ply was
314 4 Acoustic Emission

Table 4.5 Evaluation results of mode II ENF test using standard procedure and using acoustic
emission analysis
Energy release rate ASTM D7905 AEonset, total AEonset, interface
Mean [J/m2] 3722 647 980
Standard deviation [J/m2] 1227 234 185
Coefficient of variation [%] 32.9 36.1 18.8

Table 4.6 Acquisition settings and postprocessing parameters used for transverse crack tension
tests
Acquisition Postprocessing
settings Value parameters Value
Preamplification 20 dBAE Feature extraction 100 μs after
window threshold
Threshold 35 dBAE Δt-filter (15  2) μs
Triggering Individual channel
Acquisition rate 10 MS/s
Band-pass range 20 kHz to 1 MHz
Couplant Medium viscosity silicone
grease
Mounting system Clamp
Sensor type WD
Number of 2
sensors

Table 4.7 Acquisition Measurement


settings of DIC system setup Value
for transverse crack
Camera Toshiba CMOS camera CSC12M25BMP19-
tension tests
01B with 4096  3072 pixels
Lenses Titanar, focal length 2.8/50 mm
Lighting KSP 0495-0001A LED 20W/24V white 30
Filter Polarization-filter Schneider-Kreuznach
Field of view (113.4  86.7) mm2
Scale factor 0.028 mm/pixel
Subset size 20  10 pixel
Acquisition rate 2 Hz
Software ARAMIS
Configuration 3D

introduced by a standard cutter. Mechanical testing is carried out with a 250 kN


load cell and displacement rate of 1 mm/min. All acoustic emission signals were
acquired using the parameters given in Table 4.6 using two AE sensors attached to
the specimen in 125 mm distance to each other. For digital image correlation, the
edge of the specimen was monitored. Acquisition is done by 2D-measurements
applying the parameters found in Table 4.7.
4.7 Application to Composites 315

An application of the pattern recognition method described in [125] yields three


signal classes for all specimens investigated. Based on the method proposed in
[154], the uncertainty of the classification was evaluated to be less than 0.20 % for
all specimens investigated. Assignment of the signal classes to respective failure
mechanisms is carried out as described in Sect. 4.7.1.
This allows comparing different data reduction methods in one test. As the first
technique, the classical approach to deduce the onset level of the propagation load,
which is normally used for the quantification of mode II interlaminar fracture
toughness, was evaluated [184, 185]. This classical approach is essentially based
on signatures of the recorded force-displacement curves. As the second technique,
DIC was used to measure the load corresponding to first initiation of delamination.
To obtain a quantitative measure of the strain field, the displacement in the through-
thickness direction (z-direction) was evaluated at several x-positions in the vicinity
of the discontinuous center ply. The distance between two points was measured at
positions as indicated in the top image in Fig. 4.147. However, as seen in the
exemplary evaluation of the measured distance of two positions as a function of
loading in Fig. 4.147, there were no clear signatures found in the curves. Instead,
the change in slope of the signals was mostly gradual, which makes picking of a
trend change challenging. Also, there was no mutual agreement in the shape of the
individual curves recorded for different specimens. Therefore, the optical detection
of the delamination onset was found to yield unreliable results for this kind of
application, although the process of damage progression was easily visible in the
DIC images.

z
x distance 2
1st principal strain

load axis
70
distance 1
60

50
force [kN]

40

30 distance 1
distance 2
20

10

0
0.0 0.1 0.2 0.3 0.4 0.5
z-displacement [mm]

Fig. 4.147 Evaluation of distance between points in z-direction as function of loading at two
exemplary positions in the vicinity of the discontinuous center ply
316 4 Acoustic Emission

10
Matrix cracking
60 Interfacial failure
Fiber breakage
8
50

Fonset = 37.7kN

x-position [cm]
40 6
force [kN]

position of transverse
30 crack
4
20

2
10

0 0
0 50 100 150 200 250
time [s]

Fig. 4.148 Exemplary evaluation results of acoustic emission source positions as function of time
with superimposed load-time curve for a transverse crack tension specimen

As third method, the onset of localized acoustic emission signals was evaluated.
At the time of acoustic emission onset (see Fig. 4.148), no evident signature could
be detected in the DIC data or in the force-time curve. Based on previous results
from tensile specimens [177, 186], the detected acoustic emission signals could be
assigned to the occurrence of matrix cracking, interfacial failure, and fiber break-
age. One representative classification result is shown in Fig. 4.148. With increased
load, acoustic emission signals initiate at weak spots inside the laminate. Although
the discontinuous center ply is intentionally designed as a weak spot, the first
acoustic emission signals are localized at various positions along the x-axis of the
specimen. Since these signals were classified as matrix cracks, it is likely that they
originate from failure at the specimen edges, failure at resin rich areas, or failure at
randomly distributed minor fabrication defects. At a distinct load level, signals
classified as interfacial failure initiate specifically at the position of the discontin-
uous center ply (x-position 5 cm).
Subsequently, acoustic emission signals are localized in the vicinity of the
discontinuous center ply, which densify around x ¼ 50 mm and spread outward
with increased load level. This is indicative of interlaminar crack growth, starting at
the interrupted center ply. At increased load levels, acoustic emission signals are
localized in a broad range between the two sensors. This reveals the occurrence of
additional damage occurring at several positions inside the specimen. Among these
signals some are classified as fiber breakage signals. Since the tensile strength of
carbon fibers typically shows Weibull-type distributions, failure of individual fiber
filaments is well expected before the ultimate load level of the laminate is reached
[187]. Therefore occurrence of sporadic single fiber breakage long before ultimate
failure is a likely source for these acoustic emission signals.
4.7 Application to Composites 317

Remarkably, the relevant acoustic emission signals occur long before significant
signatures are visible in the load-displacement curves. Since initiation of interfacial
failure at the position of the discontinuous center ply is of key interest for further
analysis, the load level of initiation was also quantified by acoustic emission
measurements. Therefore, the first occurrence (onset) of the acoustic emission
signals, classified as interfacial failure within the range of the discontinuous center
ply at (50  5) mm x-coordinate, was used to obtain the respective load level of
failure initiation Fonset.
The occurrence of early acoustic emission indicates that the initiation of delam-
ination is not a uniform process over the width of the transverse crack tension
specimens, but may also initiate within the specimen volume. Consequently,
detection of the delamination onset based solely on the DIC information obtained
at the specimen surface may be error prone.
In summary, the traditional approach based on signatures in force-displacement
curves allows to deduce the fracture toughness for crack propagation. However,
more conservative values for the fracture toughness values for crack initiation can
be obtained by the acoustic emission method proposed above. As has been shown in
[143], the sensitivity of the acoustic emission method is sufficient to assess the
quality of different dry fiber-cutting processes such as mechanical cutting or laser
cutting. In this application no differences were found by the traditional approach,
and therefore this data reduction would significantly underestimate the influence of
the cutting process [143].

4.7.3 Comparison to Failure Criteria Predictions

Beyond the measurement of first onset of damage in specimens explicitly designed


for fracture mechanics testing, it would also be interesting to have methods capable
of visualizing the onset of specific failure modes, e.g., as predicted by failure
criteria. In the following, some examples are presented, which compare measure-
ment results using acoustic emission analysis to the predicted onsets of inter-fiber
failure and fiber failure using Puck’s failure criterion. Simple tensile test specimens
were selected for this purpose, since this load scenario is expected to yield a high
predictive capability by the theory. However, the application of the experimental
method to more complicated load scenarios is straightforward and will be demon-
strated in Sect. 4.7.4.3 for a structural component.

4.7.3.1 Unidirectional Laminates

As the first application, the case of tensile test of unidirectional laminates with load
axis in parallel to the fiber axis is discussed. According to the failure theory by
Puck, no inter-fiber failure is expected prior to ultimate failure due to fiber failure.
318 4 Acoustic Emission

Table 4.8 Acquisition settings and postprocessing parameters used for tensile tests
Acquisition Postprocessing
settings Value parameters Value
Preamplification 20 dBAE Feature extraction 100 μs after
window threshold
Threshold 35 dBAE Δt-filter (7  2) μs
Triggering Individual channel
Acquisition rate 10 MS/s
Band-pass range 20 kHz to 1 MHz
Couplant Medium viscosity silicone
grease
Mounting system Clamp
Sensor type WD
Number of 2
sensors

However, a comment is made in [39] that experimental findings indicate the


occurrence of inter-fiber failure at about 70 % of the ultimate tensile strength of
the specimen.
To evaluate the acoustic emission release of specimens under this load scenario,
tensile tests according to DIN EN 2561 using specimens made from T700/PPS were
performed [178]. Acoustic emission signals were detected using the acquisition
settings given in Table 4.8. All signals were localized within the tapered area of the
specimen and were subject to the pattern recognition method described in Sect. 4.5.
Relevant for confirmation of failure theories is the experimental detection of
initiation of specific failure modes. As seen in Fig. 4.149, contributions of all failure
mechanisms are detected during loading. The usual sequence of their occurrence
starts with signals associated with matrix cracking, followed by interfacial failure
and fiber breakage. However, for the present case, distinction between the occur-
rence of interfacial failure and matrix cracking is hardly justified, since both of
these types are associated with inter-fiber failure in the sense of Puck’s failure
criterion. Therefore, distinction is made between the average onset of signals from
matrix cracking and interfacial failure and the onset of acoustic emission signals of
fiber breakage.
The ratio of the load level at acoustic emission onset and the ultimate failure load
is shown in Fig. 4.149a. The average onset of acoustic emission signals of matrix
cracking is found at load levels between 5 and 30 % of the ultimate load. Based on
observations of simultaneously recorded videos, some of these first onsets of
acoustic emission signals correspond to the onset of splitting at the edges of the
specimen. Subsequently, damage progression can be expected, which is likely
responsible for the early onset of acoustic emission. However, as seen in
Fig. 4.149b, many of the unidirectional specimens show a certain change in
acoustic emission activity at (76.0  4.1) % load to failure. This is likely due to
the formation of inter-fiber cracks in parallel to the load axis as predicted by [39].
4.7 Application to Composites 319

a
100
Fiber breakage
Matrix cracking

80
load to failure [%]

60
54%

40

20

0
specimen

b
2000 600
stress-strain curve
1750 Matrix cracking

accumulated number of AE signals


Interfacial failure 500
Fiber breakage
1500
400
1250
stress [MPa]

1000 300

750
200
change in activity
500
100
250

0 0
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.1 1.2 1.3
strain [%]

Fig. 4.149 Evaluation results of acoustic emission onsets for different specimens for unidirec-
tional tensile specimens (a) and exemplary result of change in activity of matrix cracking signals at
75 % of ultimate load level (b)

The average onset of fiber breakage signals is quantified to be at around 54 % of


the ultimate failure load. Based on the theory of Zhou et al. [188] and the values for
the T700 filaments from [189], it is quite probable to have fiber failure at 54 % of
the ultimate strength. In fact, the typical Weibull strength distribution of filaments
would even expect fiber failure to occur at lower load levels as recently demon-
strated by Lomov et al. [190]. But as demonstrated in Sect. 4.2.3, the energy release
320 4 Acoustic Emission

of weak filaments is likely to fall below the detection limit of the sensors used. This
may therefore explain the general discrepancy between expected onset of fiber
breakage signals and AE measurements. However, the constant onset level of fiber
failure signals indicates that first measurable fiber breakage is taking place in the
loaded plies at a systematic load level prior to ultimate failure of the specimens.

4.7.3.2 Cross-Ply Laminates

As the second application, investigations of tensile tests performed on cross-ply


laminates are presented in this subsection. To this end six different cross-ply
stackings were fabricated using Sigrafil CE1250-230-39 prepreg material with a
nominal ply thickness of 220 μm. In the tapered area, the stacking sequences were
chosen as [02/90/02]sym, [02/902/0]sym, [0/903/0]sym, [0/90]sym, [(0/90)2]sym, and
[(0/90)3]sym. For production of the reinforcements, the ply-drop method was used.
The respective interspersed plies are shown exemplarily in Fig. 4.150. The data
reduction follows reference [186], and data interpretation uses the set of quantities
described in [35, 138]. All acoustic emission signals were detected using the
acquisition settings given in Table 4.9, and only signals localized in the tapered
area were used for further analysis.
For each specimen of each stacking sequence, the pattern recognition method as
described in Sect. 4.5 was applied. Representative results for four exemplary
laminate configurations are shown in Fig. 4.151 as a diagram of the signal features
Partial Power 4 over Weighted Peak Frequency.
Based on the different ratio of 0 and 90 plies of the four stacking sequences, a
different ratio of microscopic failure mechanisms is expected. As seen from
Fig. 4.151, qualitatively the total number of signals correlated with matrix cracking
increases for an increasing number of 90 plies. Similarly, the absolute number of
signals correlated with fiber breakage increases for an increasing number of

Fig. 4.150 Cross-sectional [0/0/90/0/0]sym


images of reinforcement
4.2 mm

sections with interspersed


plies for three out of six
stacking sequences [0/0/90/90/0]sym

investigated (reprinted from 45° 0°
[186]) 0° 90°
0° 90°
90° 0°
0° 0°
90° [0/90/90/90/0]sym
135°

2.0 mm
2.2 mm


4.7 Application to Composites 321

Table 4.9 Acquisition settings and postprocessing parameters used for tensile tests
Acquisition Postprocessing
settings Value parameters Value
Preamplification 20 dBAE Feature extraction 100 μs after
Threshold 35 dBAE window threshold
Triggering Individual channel
Acquisition rate 10 MS/s
Band-pass range 20 kHz to 1 MHz
Couplant Medium viscosity silicone
grease
Mounting system Clamp
Sensor type WD
Number of 2
sensors

0 plies. The absolute number of signals correlated with interfacial failure was
found to be similar for all stacking sequences with identical number of 0 /90
interfaces. This type of signals is attributed to the occurrence of inter-ply delami-
nation. The different laminate configurations produce a noticeable change of the
cluster positions in the diagrams for each failure mechanism. This is caused by
changes of the frequency spectra of the excited Lamb waves and originates from the
different elastic properties of the propagation medium and the characteristic depth
changes of the acoustic emission sources.
In the following, the pattern recognition results of all specimens are evaluated
using the following quantities as introduced in [35, 138, 186]:
The relative number of signals Nrel,i and the relative amplitude of signals Urel,i,
for the ith of the Itot failure mechanisms with Ni of the total signals Ntot, are given as

Ni
N rel, i ¼ ð4:66Þ
N tot
XN i XI XN 1
U rel, i ¼ U ij  ð4:67Þ
tot i

j¼1 i¼1 j¼1


U ij

Thus, Nrel,i and Urel,i express the contribution of a particular failure type relative to
XItot XNi
all the recorded signals Ntot with accumulated signal amplitudes i¼1
U .
j¼1 ij
In addition, the average signal amplitude hUii of the ith failure mechanism is
defined as
XN i
hU i i ¼ j¼1
U ij  ðN i Þ1 ð4:68Þ

In comparison to these normalized measures, the absolute values usually lack in


comparability due to differences in sensor sensitivities, differences in coupling
quality, and numerous other factors affecting the strength and PoD of acoustic
emission signals. Therefore the quantities derived in (4.66)–(4.68) are meant to
322 4 Acoustic Emission

provide an improved possibility to compare and interpret the failure of fiber


reinforced composites.
An evaluation of the cross-ply specimens with stacking sequences [02/90/02]sym,
[02/902/0]sym, and [0/903/0]sym in terms of (4.66)–(4.68) is shown in Fig. 4.152. The
design of these three laminate configurations gradually substitutes the 0 plies by
90 plies forming 90 layers with a thickness of one, two, and finally three plies.

a
100 Laminate [0/0/90/0/0]
sym
Matrix Cracking
Interfacial failure
80 Fiber breakage
Partial Power 4 [%]

60

40

20

0
0 100 200 300 400 500 600 700
weighted Peak Frequency [kHz]

b
100 Laminate [0/0/90/90/0]
sym
Matrix Cracking
Interfacial failure
80 Fiber breakage
Partial Power 4 [%]

60

40

20

0
0 100 200 300 400 500 600 700 800
weighted Peak Frequency [kHz]

Fig. 4.151 Representative results for the identified natural clusters for different laminate config-
urations [02/90/02]sym (a), [02/902/0]sym (b), [0/903/0]sym (c), and [0/90]sym (d) (partially based on
[186])
4.7 Application to Composites 323

c
100 Laminate [0/90/90/90/0]sym
Matrix Cracking
Interfacial failure
80 Fiber breakage
Partial Power 4 [%]

60

40

20

0
0 100 200 300 400 500 600 700
weighted Peak Frequency [kHz]
d
100 Laminate [0/90]sym
Matrix Cracking
Interfacial failure
80 Fiber breakage
Partial Power 4 [%]

60

40

20

0
0 100 200 300 400 500 600 700
weighted Peak Frequency [kHz]
Fig. 4.151 (continued)

As discussed before, Nrel,i and Urel,i show quite similar trends [35]. In view of the
physical correlation of Urel,i with the source strength, Urel,i is the relevant quantity
to discuss. In both diagrams (Fig. 4.152a, b), the contribution of matrix cracking
increases as the number of 90 plies increases. Generally, a higher critical crack
density is expected for [0/90n/0]sym laminates for lower n before the initiation of the
delamination at the interface between the 0 and 90 plies [191, 192]. This seems to
324 4 Acoustic Emission

be in contrast to the quantification of matrix cracking as seen from Fig. 4.152a, b.


However, the major amount of signals attributed to matrix cracking for n ¼ 2 and
n ¼ 3 occurs after the onset of delamination and is thus not included in Nairn’s
theory [191, 192]. The changes observed are thus attributed to damage occurring
after the onset of inter-ply delamination.

a
100 Matrix Cracking
Interfacial failure
90 Fiber breakage
R e la tiv e N u m b e r o f S ig n a ls [ % ]

80

70

60

50

40

30

20

10

0
[0/0/90/0/0]sym [0/0/90/90/0]sym [0/90/90/90/0]sym

b
100 Matrix Cracking
Interfacial failure
90 Fiber breakage
R e la tiv e A m p litu d e o f S ig n a ls [ % ]

80

70

60

50

40

30

20

10

0
[0/0/90/0/0]sym [0/0/90/90/0]sym [0/90/90/90/0]sym

Fig. 4.152 Quantification of relative number of signals (a), relative amplitude of signals (b), and
average amplitude per signal (c) for laminate types [02/90/02]sym, [02/902/0]sym, and [0/903/0]sym
(based on [186])
4.7 Application to Composites 325

c
120
Matrix Cracking
A v e r a g e A m p litu d e p e r S ig n a l [m V ] Matrix cracking (mean)
Interfacial failure
100 Fiber breakage

80

60

40

20

0
[0/0/90/0/0]sym [0/0/90/90/0]sym [0/90/90/90/0]sym
Fig. 4.152 (continued)

The number of fiber breakages increases with increasing number of 0 plies.


Based on probabilistic considerations, the frequency of the occurrence of fiber
breakage (i.e., single filament failure) before the ultimate failure of the specimen
is expected to be proportional to the cross-section of 0 plies.
Contributions of interfacial failure scatter between 10% and 70 %, but exhibit no
significant trend as a function of stacking sequence. The large scatter of signals
attributed to interfacial failure is consistent with optical observations of the mac-
roscopic failure mode. Here the specimen exhibits vast differences in the visible
delamination lengths as will be further discussed in the following.
As predicted by the generalized theory of acoustic emission [21, 22] and also
demonstrated with finite element simulations [35], the amplitude UAE of an acoustic
emission signal detected at a distance r is proportional to the vibrating crack volume
ΔV:

  ΔV
U AE ¼ βU c2L  ð4:69Þ
r

The proportionality constant βU is a linear function of the squared longitudinal


sound velocity c2L of the cracking medium and the orientation of the crack surface
movement (not taken into account in (4.69)). For the inter-fiber failure case, the
vibrating crack volume ΔV is expected to be proportional to the cross-sectional area
ΔA of the adjoining 90 plies. Consequently, based on the difference in stacking
sequence, the signal amplitudes should increase by a factor of 2 for the [02/902/
0]sym stacking sequence and by a factor of 3 for the [0/903/0]sym stacking sequence,
compared to the [02/90/02]sym stacking sequence. The mean Ui for matrix cracking
(indicated by the dashed line in Fig. 4.152c) amounts to 11.5 mV (81 dBAE),
326 4 Acoustic Emission

17.9 mV (85 dBAE), and 43.9 mV (92.8 dBAE) for the three stacking sequences,
respectively. Although strong scattering is observed in Ui of the [0/903/0]sym
stacking sequence, overall the values are found to be in good agreement with the
expected relation derived from (4.69). The average amplitude of the fiber breakage
also exhibits a characteristic dependency on the stacking sequence. The mean Ui
decreases as the number of 0 plies decreases. Because the majority of the signals
belonging to this class originate from single filament failure, the vibrating crack
volume ΔV is not proportional to the thickness of the 0 plies. Instead, the
amplitude UAE is influenced by the strain energy release, which is higher in the
stacking sequences with higher number of 0 plies.
While the diagrams of Fig. 4.152 are mostly useful to visualize the accumulated
damage causing final failure of the specimens, the evolution of microscopic failure
mechanisms is another relevant factor for comparison to predictions from analytical
failure models. Based on the values from Table B.1 and Table B.4 in Appendix B,
the stress and strain values for first-ply failure and last-ply failure (ultimate failure)
were calculated for the six stacking sequences investigated according to Puck’s
nonlinear laminate failure model [193–196]. The stress and strain values for the
onset of the three failure mechanisms were obtained from the acoustic emission
measurements, and their mean value was calculated from all samples of each
stacking sequence. A graphical comparison of the obtained values including their
margin of error is shown in Fig. 4.153.
The calculated stress and strain values for last-ply failure are found to be in good
agreement with the measured values for maximum stress. Only for the [(0/90)3]sym
stacking sequence a laminate strength lower than the predicted value was measured.

Calculated
First ply failure
1400
Last ply failure
Measured
1200 Onset Matrix cracking
Onset Interfacial failure
Onset Fiber breakage
1000 Maximum stress
Stress [MPa]

800

600

400

200

0
[02/90/02]sym [02/902/0]sym [0/903/0]sym [0/90]sym [(0/90)2]sym [(0/90)3]sym

Fig. 4.153 Calculated first-ply failure and last-ply failure based on Puck’s nonlinear failure model
and respective values derived from acoustic emission measurements
4.7 Application to Composites 327

The systematic decrease of the laminate strength for the [(0/90)n]sym laminates is
likely caused by the volume effect [197], since the stacking sequence design is
expected to yield identical failure strength based on Puck’s nonlinear failure model.
Generally, last-ply failure is caused by fiber breakage in the 0 plies of the stacking
sequences. In this study, the quantified onset of fiber breakage is found to be
systematically lower than the stress values calculated for ultimate failure. This is
explained by the fact that acoustic emission measurements are sensitive enough to
detect the occurrence of single filament failure. Thus, detection of fiber breakage by
pattern recognition can effectively act as early indicator for imminent structural
failure down to a range of 70 % of ultimate failure (cf. also Sect. 4.7.3.1).
A good agreement is also found between the calculated stress values for first-ply
failure and the initiation of matrix cracking and interfacial failure. Overall, the
mean onset of matrix cracking is systematically lower than the values calculated for
first-ply failure, while the onset of interfacial failure sometimes exceeds the calcu-
lated value. This might be caused by acoustic emission signals originating from the
edges of the specimen due to inhomogeneities and improper specimen preparation.
Although this kind of initial type of failure is not covered by the analytical
description [193–196], it is easily detectable by acoustic emission measurements.
Similar to the case of fiber breakage, detection of matrix cracking and interfacial
failure can be used to quantify the onset of failure in composites by acoustic
emission analysis as a nondestructive measurement technique.

4.7.4 Tracking Failure Evolution

After the detection of first failure onsets, one of the key tasks for monitoring
methods is to keep track of the further damage evolution. In the following sections,
some typical applications are presented, which have used acoustic emission anal-
ysis for this purpose.

4.7.4.1 Double-Cantilever Beam Testing

In general, double-cantilever beam tests according to standards like ISO 15024 or


ASTM D 5528 are used to evaluate the mode I interlaminar fracture toughness
value of fiber reinforced laminates. A schematic drawing of such a typical test setup
is shown in Fig. 4.154. As has been demonstrated previously, acoustic emission
analysis can significantly improve the interpretation of composite material failure
[35, 131, 138, 181, 198]. In this section, two routines using acoustic emission are
demonstrated, which improve the testing procedure in the double-cantilever beam
setup.
Typically traveling microscopes or high-resolution camera systems are used to
follow the position of the crack tip. The progress of delamination is then obtained
via measurement of the crack progress between subsequent images. Evaluating the
328 4 Acoustic Emission

z
AE sensors

5.7 mm
force
x
specimen
1 2

crack tip

250 mm
initial crack length 67mm

Fig. 4.154 Experimental setup used for double-cantilever beam tests in combination with acous-
tic emission acquisition

220 raw data


smoothed
200
180
160
source position [mm]

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

Fig. 4.155 Acoustic emission source positions during double-cantilever beam test as function of
time (reprinted from [198])

delamination length as a function of loading, it is thus possible to calculate the


delamination resistance curve (R-curve). However, the number of images used to
quantify the delamination lengths is usually limited, since this involves manual
processing of each image.
In the following, acoustic emission source localization is applied to track the
position of the crack tip. This approach follows previous investigations by
[181, 199] and can be used to generate continuous R-curves. In Fig. 4.155 the result
from a measurement of a HTA/RTM6 specimen as reported in [198] is shown.
4.7 Application to Composites 329

Table 4.10 Acquisition settings and postprocessing parameters used for double-cantilever beam
tests
Acquisition Postprocessing
settings Value parameters Value
Preamplification 40 dBAE Feature extraction 200 μs after
Threshold 35 dBAE window threshold
Triggering Individual channel
Acquisition rate 10 MS/s
Band-pass range 20 kHz to 1 MHz
Couplant Medium viscosity silicone
grease
Mounting system Clamp
Sensor type WD
Number of 2
sensors

All acoustic emission signals were detected according to the parameters reported in
Table 4.10. Subsequent to testing, the acoustic emission signals were localized in
one dimension between the two sensors and are shown as function of time in
Fig. 4.155. In order to obtain an effective crack length as function of time, the
raw data was smoothed by a percentile filter with 1500 points window size.
Compared to other averaging techniques (adjacent averaging, Savitzky-Golay
method), this filter is less affected by outliers and better reflects the achieved
crack length. In particular, for those cases with sudden crack progress, steps in
the delamination length occur. The filter setting chosen is able to preserve these
sharp jumps and therefore resembles a reasonable method to obtain effective crack
lengths of the damage zone surrounding the crack tip.
Utilizing an interpolation scheme on the smoothed acoustic emission source
position, it is possible to obtain a continuous representation of the delamination
length a. In combination with the recorded force F, the cross-head displacement e δ,
and specimen width b, this allows for continuous calculation of the GI value by the
modified beam theory method recommended in ASTM D5528. The corresponding
fit quality to calculate the offset value Δ used in (4.70) is shown in Fig. 4.156:

3Feδ
GI ¼ ð4:70Þ
2bða þ jΔjÞ

By knowing the onset of acoustic emission, it is also possible to define a measure


for the onset of crack growth. During the experiments it was observed that this
definition for the crack onset precedes the visible crack onset and can thus be
understood as the most conservative estimation for GIc. Subsequently the continu-
ous R-curve (strain energy release as a function of delamination length) as seen in
Fig. 4.157 can be further used to evaluate the interlaminar delamination process.
330 4 Acoustic Emission

0.9

0.8
compliance C1/3 [N/mm] 0.7

0.6

0.5

0.4

0.3

0.2

0.1 1/3
C = -(0.030 +/- 0.001) + (0.00632+/-0.00001) x a
0.0
0 20 40 60 80 100 120 140
delamination length a [mm]

Fig. 4.156 Compliance data and linear fit to obtain shift of effective delamination length Δ
(reprinted from [198])

450

400
strain energy release rate [J/m2]

350

300 <GI > = 305.6 J/m2

250 G Ic = 268.7 J/m2

200

150

100

50

0
40 50 60 70 80 90 100 110 120 130 140 150 160 170 180
delamination length a [mm]

Fig. 4.157 R-curve of sample including GIc onset value and GIc mean value (reprinted from [198])

The minimum value of the strain energy release rate at the onset of acoustic
emission is then used as fracture toughness value GIc and can be estimated as
268.7 J/m2 for the example in Fig. 4.157. Accordingly it is feasible to define a mean
GIc value occurring during crack propagation as arithmetic average of all values of
the R-curve. This mean GIc value amounts to 305.6 J/m2.
4.7 Application to Composites 331

Table 4.11 Results of mode I energy release rate evaluation


NL VIS 5%/MAX AEonset AEstable
Energy release rate [J/m2] [J/m2] [J/m2] [J/m2] [J/m2]
Mean [J/m2] 222 700 968 289 1028
Standard deviation [J/m2] 22 93 88 80 –
Coefficient of variation [%] 9.9 13.3 9.1 27.7 –

Being typical for a brittle matrix system as RTM6, the fracture toughness value
for crack initiation is lowest, followed by a region with stable crack propagation of
higher fracture toughness.
However, this behavior is not necessarily identical for ductile matrix materials as
will be demonstrated for T700/PPS specimens in the following. Such ductile
thermoplastic materials may exhibit a transition between subcritical failure during
crack initiation mixed with stable and unstable crack growth. Using the test setup of
Fig. 4.154 and acoustic emission acquisition parameters as given in Table 4.10,
such behavior can be measured by acoustic emission analysis. To this end, the
recorded acoustic emission signals were localized in one dimension, and the pattern
recognition method described in Sect. 4.5 was applied. According to the ASTM D
5528 standard, the energy release rate is calculated for the onset of the first
nonlinear (NL) force-displacement behavior, for the onset of the first visible
crack growth (VIS), and for 5 % increased compliance or the force maximum
(5 %/MAX). The corresponding evaluation of these values is given in Table 4.11
for comparison.
Based on the comparison in Fig. 4.158b, the onset of acoustic emission signals
coincides well with the first onset of nonlinear force-displacement behavior. This
indicates that prior to the force maximum, significant damage occurs within the
specimen. With increased loading it is observed that the contributions of signal
classes of matrix cracking and interfacial failure dominate, while fiber breakage
signals show only minor contributions.
Looking more precisely into the contributions of the individual failure mecha-
nisms, Fig. 4.159a reveals some interesting details. Here the relative contribution of
the individual mechanisms was quantified according to (4.67). There are two
distinct ranges of substantially different contributions of the failure mechanisms
found in all of the tested specimens. At the initiation of crack growth, the contri-
butions of the individual signal classes tend to scatter by up to 30 %. After reaching
the force maximum, their contributions stabilize and remain constant within a range
of scatter of 5 %. This transition between the two ranges is hardly visible in the
respective force-displacement curve.
After carrying out a series of measurements, this allows to combine measure-
ment results of different specimens at different times during the experiment. A
compilation of such results is shown in Fig. 4.159b. Here fracture toughness values
were obtained for each specimen at these times, where the VIS criterion is fulfilled.
332 4 Acoustic Emission

a
100 Matrix cracking
Interfacial failure
Fiber breakage
80
Partial Power 2 [%]

60

40

20

0
0 100 200 300 400 500 600 700 800 900 1000
weighted Peak-Frequency [kHz]
b
force-displacement curve
Matrix cracking 22000
80 Interfacial failure 20000
Fiber breakage

accumulated number of AE signals


70 18000
AEonset
60 16000
14000
50
force [N]

12000
40 10000

30 8000
6000
20
4000
10
2000
0 0
0 10 20 30 40 50 60 70 80 90 100
displacement [mm]

Fig. 4.158 Result of pattern recognition result for one exemplary specimen (a) and accumulated
number of acoustic emission signals superimposed to force-displacement curve (b)

The fracture toughness values are used as x-axis in this diagram. As y-axis, the
corresponding value of the relative amplitude at the time of reaching the VIS
criteria is plotted. In addition, the mean values of the fracture toughness as
computed from the R-curves in the region of stable crack propagation were
4.7 Application to Composites 333

a
100
initiation stable crack growth
90 80
80
relative amplitude [%]

70
60
60

force [N]
50
40
40 force-displacement curve
Matrix cracking
30
Interfacial failure
20 Fiber breakage 20

10

0 0
0 10 20 30 40 50 60 70 80 90 100
displacement [mm]

b
100 Matrix cracking
inititation stable crack growth
Linear Fit
90 Interfacial failure
80 Linear Fit
Fiber breakage
relative amplitude [%]

70
PPS: 210 J/m2
60
GIc-value stable: 1028 J/m2
50

40

30

20

10

0
0 200 400 600 800 1000 1200 1400
GIc-value [J/m2]

Fig. 4.159 Relative amplitude of failure mechanisms for one representative specimen (a) and
combined evaluation of all specimens of a series of measurements (b)

determined for each specimen. Accordingly, the relative amplitude value of each of
the specimens as found in the region of stable crack propagation is plotted on the
y-axis.
The plot of data points in Fig. 4.159b allows several conclusions for the material
system investigated. As reported previously in [35, 138], the back extrapolation of
334 4 Acoustic Emission

the contributions of matrix cracking and interfacial failure intersects at around 50 %


contribution and at an abscissa of the fracture toughness value of the pure matrix
material. For PPS this is expected around 210 J/m2 [182]. This value is also in good
agreement with the most conservative fracture toughness value obtained by the NL
criterion and the value obtained by the onset of acoustic emission AEonset as
reported in Table 4.11. This allows concluding that the onset of nonlinear behavior
is merely due to subcritical crack growth, which precedes the macroscopic crack
growth [200].
As the second conclusion, it is possible to deduce a lower boundary for the
fracture toughness value for stable crack growth. Based on the significant trend
change in Fig. 4.159b, this can be evaluated as (1028  50) J/m2. This value is also
in good agreement with the values obtained from the 5 %/MAX criterion (see
Table 4.11). Therefore, the acoustic emission analysis allows for reinterpretation
of the values of ASTM D 5528 in terms of fracture toughness values for subcritical
crack initiation and fracture toughness values for stable crack growth.

4.7.4.2 Tensile Testing

While the double-cantilever beam test is specifically designed to allow for moni-
toring of crack propagation, this is not the case for most of the tensile specimens. In
many cases, without secondary methods, the only noticeable observation of damage
occurs at the time of rupture of the specimen. Sometimes preliminary failure is
observable due to failure of filaments at the edges or due to audible cracking of the
specimen. However, none of these phenomena can be interpreted in terms of
quantitative values. As has been demonstrated in Sect. 4.7.3, acoustic emission
detected during loading of tensile specimens can be used to detect failure mode
specific onsets. Beyond that, the continuous recording of acoustic emission signals
also allows to record the evolution of damage. This possibility is demonstrated in
the following using the measurements previously described in Sect. 4.7.3.2.
In Fig. 4.160 the accumulated number of signals corresponding to each of the
three failure mechanisms is plotted together with the specimen’s stress-strain curve.
Although the specimens in Fig. 4.160a, b originate from the same sample plate and
have the same [02/90/02]sym stacking sequence, they show a significant difference in
their failure behavior. While the specimen in Fig. 4.160a exhibits the theoretically
expected order of failure mechanism onsets (matrix cracking, interfacial failure,
fiber breakage), this is different for the specimen results in Fig. 4.160b (matrix
cracking, fiber breakage, interfacial failure). However, a significant increase in the
activity rate of fiber breakage signals is observed above 1 % strain in both speci-
mens in Fig. 4.160. This is expected to be the onset of fiber failure that ultimately
causes last-ply failure.
Also, the weight of contributions of fiber breakage and interfacial failure is
contrary in both specimens. This is found to be consistent with the following
macroscopic observations. Specimen 1 shows large areas of delamination
(bottom Fig. 4.161), while specimen 2 fails predominantly by fiber breakage (top
4.7 Application to Composites 335

a
1600 220
Laminate [0/0/90/0/0]sym 200
1400 Matrix cracking Specimen 1
180

accumulated number of signals


Interfacial failure
1200 Fiber breakage
Stress-strain curve 160

1000 140
stress [MPa]

120
800
100
600 80
60
400
40
200
20
0 0
0.0 0.2 0.4 0.6 0.8 1.0 1.2
strain [%]
b
1600 220
Laminate [0/0/90/0/0]sym 200
1400 Matrix cracking Specimen 2

accumulated number of AE signals


Interfacial failure 180
1200 Fiber breakage
160
Stress-strain curve
1000 140
stress [MPa]

120
800
100
600 80
60
400
40
200
20
0 0
0.0 0.2 0.4 0.6 0.8 1.0 1.2
strain [%]

Fig. 4.160 Comparison between stress-strain curves and accumulated number of acoustic emis-
sion signals for two specimens with [02/90/02]sym stacking sequence (specimen 1 in a, specimen
2 in b)

Fig. 4.161). These differences also account for the substantial differences in the
respective contributions of failure mechanisms as seen in Fig. 4.160.
Apart from monotonically increasing the applied load, repetitive loading and
unloading cycles are a popular concept used for acoustic emission testing. This
testing scheme is shown in Fig. 4.162. Here the specimen is subject to a monotonic
336 4 Acoustic Emission

fiber breakage

Specimen 2
16.2 mm

Specimen 1

0∞ inter-ply delamination

Fig. 4.161 Macroscopic observed failure mode for two different specimens with identical [02/90/
02]sym stacking sequence cut from the same plate

Fig. 4.162 Scheme for L max


definition of Felicity-ratio L2 number of AE signals

B2
L1 1
B1 load profile

time
AE onset

increasing load until a certain load level L1 is reached. Then the specimen is fully
(or partially) unloaded. Subsequently the load is again increased monotonically
until a certain load level L2 is reached, which exceeds the previous load level. This
process is repeated N-times up to a certain maximum load level Lmax or up to
specimen failure. During the whole process, acoustic emission is recorded to
evaluate the progression of damage. Focus is given on the correct evaluation of
the onset of a significant amount of acoustic emission signals.
For the scheme shown below, the onset of signals in the second cycle is at the
load level L2, which allows calculating the Felicity6-ratio FR of cycle 2 as follows:

FR ¼ B2 = L1 ð4:71Þ

Accordingly, the load-ratio LR of cycle 2 can be calculated using the failure load
Lmax:

LR ¼ L2 =Lmax ð4:72Þ

6
The term “Felicity” was established by Timothy Fowler, honoring the contributions of his
daughter Felicity to his scientific work [218].
4.7 Application to Composites 337

a b
load load

number of AE signals

load profile

time time
second cascade of AE signals
AE onset by linear regression
first AE signals

Fig. 4.163 Approaches to establish significant onset of acoustic emission signals at repetitive
loading based on first AE signal or cascades (a) or based on linear regression (b)

As Kaiser has demonstrated for the case of metallic materials, the acoustic emission
usually starts after exceeding the previous load level, which implies FR  1
[201]. This behavior of materials has been termed Kaiser effect to honor his
pioneering work in the field of acoustic emission. But in most of the cases for
fiber reinforced materials, the acoustic emission signals initiate at lower load levels
than previously reached. This fundamental observation goes back to the work of
Timothy Fowler providing the definition of (4.71) in [202, 203]. Since the first
publication in 1977, the evaluation of the Felicity-ratio has become an important
concept to understand the progression of failure in fiber reinforced materials using
acoustic emission.
One particular challenge arises from the correct interpretation of the “signifi-
cant” onset of acoustic emission. As the term significant is also used in standards
like ASTM E 1067, it is important to point out that there is no unique definition of
significant acoustic emission. Generally, the acoustic emission onset may not
necessarily be identical to the occurrence of the first signal, since this may likely
be a noise signal or a premature (unique) failure on the microscale, which is not
representative for the full specimen or structure under inspection. Especially for
cases with large gaps between the occurrence of the first signal and the occurrence
of the next cascade of acoustic emission signals (cf. Fig. 4.163a), the assignment of
the signal onset based on the first signal is sometimes doubtful.
For cases, where there are a lot of acoustic emission signals starting already at
the moment of load application, other aspects might need to be considered. Here
trend monitoring of the activity curves may yield an improved approach to detect
the onset of “significant” acoustic emission. Especially for cases, where the first
signals show only negligible energy compared to the rest of the detected signals,
this may yield more meaningful signal onset values. Such can be done by backward
extrapolation of linear regression if the activity is rather constant or by other
approaches to systematically establish a measure for the trend change in the knee
of an acoustic emission activity curve (cf. Fig. 4.163b).
For interpretation of the failure progression in fiber reinforced materials, an
exemplary evaluation of a unidirectional T700/PPS tensile specimen is shown in
338 4 Acoustic Emission

a
Felicity-ratio (based on first signal)
1.2 Felicity-ratio (based on average of first 10 signals)

1.0

0.8
Felicity-ratio

0.6

0.4

0.2

0.0
0.0 0.2 0.4 0.6 0.8 1.0
load-ratio
b
1.2

1.0

0.8
Felicity-ratio

0.6
Fiber breakage
Fnterfacial failure
0.4 Matrix cracking

0.2

0.0
0.0 0.2 0.4 0.6 0.8 1.0
load-ratio

Fig. 4.164 Felicity-ratio evaluation of tensile specimens (UD-T700/PPS) based on first signal and
average of first ten signals (a) and evaluated according to individual failure mechanisms (b)

Fig. 4.164. Except for the load scheme, the acquisition setup confirms with the
configuration reported in Sect. 4.7.1. The specimen has been tested until failure, so
Lmax is known and the load-ratio of the respective cycles can be calculated
according to (4.72). Figure 4.164a holds an evaluation of the Felicity-ratio based
on the occurrence of the first signal (black dots) and based on the average load level
4.7 Application to Composites 339

of the first ten signals (red dots). Only signals localized in the test section of the
specimen have been taken into account for the evaluation. At the cycle of first
occurrence of acoustic emission, the calculated Felicity-ratio based on the first
signal is already at 0.9. However, the respective Felicity-ratio based on the average
of the first ten signals is at 1.1 indicating an undamaged specimen. For both
evaluation types, there is a continuous decrease of Felicity-ratios with increased
loading. This is indicative of a growing damage state and can be directly related to
the increased load-ratio. For the Felicity-ratio of the average of the first ten signals,
this is almost a direct linear correlation as indicated by the linear regression.
However, the example demonstrates that the onset of the first signal is not neces-
sarily the most stable means of evaluating the Felicity-ratio. Especially for the
higher load-ratios, these single signal-based Felicity-ratios drop to very low levels
(indicating early onsets), which is due to the presence of many damaged areas in the
material. Also, from the comparison of both types of evaluation, it is obvious that
the definition of the onset will have significant impact on the absolute values of the
Felicity-ratio for the same type of experiment.
In Fig. 4.164b the evaluation of the same specimen is shown using the results of
a pattern recognition process as described in Sect. 4.7.1. Here the Felicity-ratios are
based on the signal onsets of the individual classes related to fiber breakage,
interfacial failure, and matrix cracking. For convenience only the evaluation results
using the average load level of the first ten signals per class are plotted. As expected
from failure theory, the first onset is usually given by the occurrence of matrix
cracking, yielding lowest Felicity-ratios. This is followed by the class of (single)
fiber breakage and then by the class of interfacial failure. However, there are some
cycles, where the order of occurrence is different. For the first load cycle, the first
signal onset is given by fiber breakage, which has been visually observed during the
test procedure to originate from failure of filaments at the edge of the specimen.
Similar observation was made for the load-ratio 0.85, which explains the early onset
of fiber breakage signals in these cycles.

4.7.4.3 Structural Components

So far, the application of acoustic emission analysis was mostly demonstrated for
experimental setups used for material testing. However, the method is by far not
limited to such applications. Therefore, a representative example of a more com-
plex structure made from fiber reinforced polymers such as an angle-ply torsion
shaft as shown in Fig. 4.165 is presented. The aim is to demonstrate the possibilities
of the methods introduced in the previous sections in a structural component and to
provide an example on how to visualize the evolution of failure in such a structure.
The torsion shafts were fabricated by a wet filament winding process based on a
Sigrafil C30 carbon fiber and SIKA CR132/CH132-7 resin as matrix polymer. All
specimens are wound as a 30 layup with fibers in circumferential direction as
reinforcement. At the load introduction points and at two further locations, which
340 4 Acoustic Emission

force
AE sensor position torque

50.0 mm 39.8 mm

y
370.9 mm
z
bearing support

Fig. 4.165 Scheme of torsion shaft experiment including AE sensor positions and load introduc-
tion concept

Table 4.12 Values of torque Cycle Torque [Nm] Bending moment [Nm] Ratio
and resultant bending moment
1 500 323 1.55
as applied to the torsion shafts
2 750 482 1.56
3 1000 647 1.55
4 1250 798 1.57

are used as support position for a roller bearing, additional 90 layers were added as
reinforcement. For load introduction a superimposed bending/torsion load profile as
reported in Table 4.12 was chosen.
To apply a static transversal force, a special load frame was used as introduced in
[204]. The transverse load is introduced perpendicular to the cylinder axis and is
supported by two roller bearings mounted on the torsion shaft. In addition to the
transverse load, a torque is superimposed to the full shaft by means of a universal
testing machine. In four subsequent cycles, bending force and maximum torque were
increased following the levels reported in Table 4.12. Testing was carried out in three
steps for each load cycle. First the static bending force was manually increased to the
force level corresponding to the bending moment of Table 4.12. As the second step,
the torque was automatically increased until the maximum value was reached and
subsequently unloaded to zero torque. As final step, the bending force was manually
released to reach a zero-stress state for the start of the next load cycle.
During all steps the specimen was monitored by eight acoustic emission sensors
mounted at the positions marked in Fig. 4.165 applying the acquisition parameters
given in Table 4.13. Given the geometrical arrangement, no mounting of sensors by
clamp systems or suction cup holders was feasible. Hence, hot glue was used as
couplant and mounting system. This causes somewhat higher attenuation than for
silicone grease and also causes a slight shift in frequency sensitivity. Therefore
reference measurements on laboratory-scale specimens using different couplant
systems are recommended to check the performance of the sensing system with
respect to sensitivity and signal distinguishability.
Figure 4.166 shows the classification result of the acoustic emission measure-
ments. The pattern recognition method is able to classify the signals as matrix
cracking, interfacial failure, and fiber breakage based on the method proposed in
4.7 Application to Composites 341

Table 4.13 Acquisition settings and postprocessing parameters used for torsion shaft test
Acquisition settings Value Postprocessing parameters Value
Preamplification 40 dBAE Feature extraction window 100 μs after threshold
Threshold 35 dBAE
Triggering Individual channel
Acquisition rate 10 MS/s
Band-pass range 20 kHz to 1 MHz
Couplant Hot glue
Mounting system Hot glue
Sensor type WD
Number of sensors 8

Table 4.14 Acquisition settings and postprocessing parameters used for large-scale pressure
vessel testTable 4.14 Citation is missing. Kindly check and provide.was added
Acquisition
settings Value Postprocessing parameters Value
Preamplification 60 dBAE Feature extraction window 250 μs after threshold
Threshold 40 dBAE
Triggering Individual channel
Acquisition rate 10 MS/s
Band-pass range 100 kHz to 1 MHz
Couplant Hot glue
Mounting system Hot glue
Sensor type WD
Number of sensors 16

100
Matrix cracking
Interfacial failure
Fiber breakage
80
Partial Power 2 [%]

60

40

20

0
0 100 200 300 400 500 600 700 800 900 1000
weighted Peak-Frequency [kHz]

Fig. 4.166 Diagram of Partial Power 2 over Weighted Peak Frequency for one representative
specimen (signals shown are result of all load cycles)
342 4 Acoustic Emission

Sect. 4.4. The according UoC was found to be below 1.9 % for all specimens
investigated. Due to the partitions found for other specimens and the relatively
short source-sensor distances, the assignment is expected to be valid.
In Fig. 4.167, the classification results are shown as function of time of the
experiment superimposed to the torque as function of time. First acoustic emission
signals are observed at the application of the static bending force as well as after
some additional application of torque. Hence, first occurrence of damage on the
microscale should occur, but as indicated by the few signals during the first three

a
700 101
torque-time curve
600 Matrix cracking 100

10-1
500

AE energy [pJ]
10-2
torque [Nm]

400
10-3
300
10-4
200
10-5

100 10-6

0 10-7
0 500 1000 1500
time [s]
b
700 101
torque-time curve
600 Interfacial failure 100

10-1
500
AE energy [pJ]

10-2
torque [Nm]

400
10-3
300
10-4
200
10-5

100 10-6

0 10-7
0 500 1000 1500
time [s]

Fig. 4.167 Evolution of failure mechanisms classified as matrix cracking (a), interfacial fail-
ure (b), and fiber breakage (c) as function of load cases and accumulated AE energy of all failure
mechanisms as function of load cases (d)
4.7 Application to Composites 343

c
700 101
torque-time curve
600 Fiber breakage 100

10-1
500

AE energy [pJ]
10-2
torque [Nm]

400
10-3
300
10-4
200
10-5

100 10-6

0 10-7
0 500 1000 1500
time [s]
d
torque-time curve
700 Matrix cracking 80
Interfacial failure
Fiber breakage 70
600

accumulated AE energy [pJ]


application of bending moment
60
500
50
torque [Nm]

400
40
300
30
200
20

100 10

0 0
0 500 1000 1500
time [s]
Fig. 4.167 (continued)

load cycles, there is not much damage expected to be present at these load levels.
The source positions of these load cycles are plotted as density diagrams
superimposed to the torque shaft structure in Fig. 4.168a for matrix cracking
sources and in Fig. 4.168b for interfacial failure sources. As seen from these
diagrams, no acoustic emission sources are localized in the areas of torque intro-
duction but preferentially within the area of applied transverse load.
344 4 Acoustic Emission

a Matrix cracking AE source density


high
y
medium
z low

b Interfacial failure

Fig. 4.168 Density diagrams of failure mechanisms classified as matrix cracking (a) and interfa-
cial failure (b)

At load cycle four, the number and energy of acoustic emission signals increase
substantially during the application of the static bending moment. Subsequently the
application of the torque causes further increase of the acoustic emission activity as
seen from the accumulated acoustic emission energy in Fig. 4.167d. Accordingly,
first acoustic emission sources are localized in the area of load introduction of the
static bending moment. A further increase of acoustic emission activity is then
observed at the beginning of load cycle five causing numerous localized source
positions dominantly in the central part of the torsion shaft, but also at the locations
of the roller bearings acting as structural supports for the static bending force. Here,
Fig. 4.168a, b can be used to visualize the preferential positions of matrix cracking
and interfacial failure, which can then be applied to optimize the chosen layup for
damage tolerance or to modify the torsion shaft geometry to withstand a higher load
level. Finally, the accumulated acoustic emission energies can also be used for
comparative purpose to estimate the mechanical energy release due to crack growth
(cf. [138]). However, the latter requires comparable acquisition conditions for the
experiments including the geometry of the specimen, the sensor types and position,
digitization settings, and further parameters (also see Appendix A for a compre-
hensive list).
Another type of application, which has been investigated using acoustic emis-
sion to a substantial extent, are composite pressure vessels as being used for storage
of pressurized gases such as compressed natural gas (CNG) and liquefied petroleum
gas (LPG) for energy storage, as compressed air cylinder, or as low-pressure tank
system for lightweight applications. For filament winding as fabrication technique,
one can distinguish between five different categories of pressure vessels as shown
schematically in Fig. 4.169:
• Type I: Fully metallic pressure vessel, which does not contain fiber reinforced
composite materials as structural reinforcement
• Type II: Mostly metallic vessel with some fiber reinforced material applied
along the hoop direction
• Type III: Metallic liner material with full overwrap of fiber reinforced material
to carry the structural load (inside and outside pressure, impacts, etc.)
4.7 Application to Composites 345

Fig. 4.169 The five basic a b c


type categories of
composite pressure vessels
as seen in (a-e)

type I type II type III

d e

type IV type V

• Type IV: Polymeric liner material with full overwrap of fiber reinforced material
to carry the structural load (inside and outside pressure, impacts, etc.)
• Type V: No liner material, but solely fiber reinforced material carrying all
structural loads and having sufficient tightness and matrix crack resistance to
avoid leakage of contained gases or liquids
Due to the safety criticality of pressure vessels, there is a long history of methods
evaluating the performance of these structures. The application of acoustic emis-
sion to pursue quality control, to monitor aging, or to predict burst pressures has
been followed for decades [205–212]. For some cases, the monitoring of acoustic
emission has been covered by standards like ASTM E 2863, ASTM E 1067, ASTM
E 1419, ISO 16148, and ISO 19016, whereas in other fields of application, there is
still a lack of established standards.
From the viewpoint of acoustic emission monitoring of composite pressure
vessels, there are several factors governing the potential success of the measure-
ment. First it is helpful to note that for cylindrical pressure vessels, the usual ply
layout chosen to fabricate the vessels is different compared to the typical laminates
since the hoop load and the axial load are not identical. Hence, the number of
filaments to the axial and hoop directions differs, which causes a pronounced
acoustic anisotropy. Second, the filament winding process typically uses a layup
sequence, which is not balanced, i.e., is not symmetric around the midplane. Both
effects cause a complex type of signal propagation, which is hardly captured by
analytical considerations and sometimes counterintuitive to understand. However,
numerical methods may be used to visualize and interpret signals of different source
mechanisms even in such complex structures [53, 213–215]. Also the possibility of
propagation paths through the contained liquid may lead to additional complexity in
the interpretation of the detected acoustic emission signals [215, 216].
In the following an example of a large-scale burst pressure test is shown to
demonstrate the possibilities of acoustic emission analysis in such an application.
The type IV pressure vessel inspected is shown in Fig. 4.170, including the
positions of the 16 mounted acoustic emission sensors using the settings of
Table 4.14. The pressure vessel was fabricated using a T700/PPS unidirectional
346 4 Acoustic Emission

y
1.3 m AE sensor
x
position

y
z 4.0 m
y z
x

Fig. 4.170 Scheme of burst pressure configuration including 16 AE sensor

tape in situ consolidated using a high-power laser in a filament winding process and
is reported in detail in [212, 217]. The axial dimension of the vessel is 4.0 m, with
an outer diameter of 1.3 m as seen in Fig. 4.170. The full vessel is composed out of
two composite parts joint by a two-sided clevis fastener with two rows of bolts. For
the chosen sensor positions, a maximum distance of 0.5 m between the source and
sensor is expected for every point on the vessel.
Based on the typical amplitude distribution of sources generated by such a
vessel, the PoD of the measurement chain can be calculated. This procedure is
based on the measured attenuation curves along the hoop direction and along the
axial direction of the vessel. The results of these calculations are reported in
Fig. 4.171 based on the 60 dBAE gain of the preamplifier and the acquisition
threshold of 40 dBAE. To estimate the lower bound of the PoD, the attenuation
curves of the higher-frequency portions of the signals are considered, since these
are subject to higher attenuation as pointed out in Sect. 4.3. As seen from
Fig. 4.171a, b, respectively, this allows estimation of the PoD in the vessel to be
larger than 60 %. Accordingly, the maximum distance between source and sensor to
allow a PoD > 90 % is below 150 mm. However, the application of such a dense
sensor network is usually not economical in a field testing application; the acquired
acoustic emission signals need to be interpreted from a statistical point of view.
Since fiber reinforced composites generate numerous acoustic emission signals
prior to failure and the signals with highest intensity have a much higher PoD
than the signals at the detection limit of the acquisition system, approaches with an
overall PoD down to 60 % are sometimes justified.
There are various established protocols to test pressure vessels in combination
with acoustic emission. The loading scheme being used depends on the aim of test,
such as quality control, repeated testing after being in service, proof testing, or burst
pressure evaluation.
As recommended by ASTM E 1067, the load scheme for the pressure vessel
shown in Fig. 4.170 was based on subsequent cycles with incremental peak load and
intermediate unload to 70 % of the peak load. The peak load was increased until
4.7 Application to Composites 347

a
100

80
axial direction

60
PoD %

40

20

60dB preamplification
0
0 50 100 150 200 250
source-sensor distance [mm]

100

hoop direction
80
PoD %

60

40 60dB preamplification

20

0
0 500 1000 1500 2000
source-sensor distance [mm]

Fig. 4.171 Calculated probability of detection along axial direction (a) and along hoop
direction (b)

final burst failure of the vessel. Using the sensor network mounted on the pressure
vessel, it is possible to localize acoustic emission source positions and visualize
them as function of the applied pressure. As seen from the density diagram of
localized acoustic emission sources in Fig. 4.172 at the moment of vessel rupture,
there are numerous sources active at the wedge sections of the composite parts of
the pressure vessel. The region of highest acoustic emission density is the actual
position of failure of the whole vessel, which was already indicative in the first
pressure cycle (24.7 % of burst pressure) as shown in Fig. 4.172.
348 4 Acoustic Emission

aer first cycle

3m

2m

1m

0° 180° 360°

at burst pressure

3m

2m

1m

0° 180° 360°

Fig. 4.172 Acoustic emission source localization results as density diagram at burst pressure

An analysis of the position of first occurrence of acoustic emission signals within


each load cycle is shown in Fig. 4.173. Here the vertical coordinate including the
standard deviation of the first ten signals is plotted as function of the load-ratio. For
the first two cycles, the first acoustic emission signals originate from positions in the
cylindrical part of the upper half of the pressure vessel. For later cycles, their
position changes and stays within the area of locally increased thickness around
1500 mm. For the last load cycles, the position moves to the range covered by the
clevis fastener. At this stage, the damage has grown within the laminate and causes
increased stress concentration at the bolts. Within the last cycle, the position of
occurrence of the first acoustic emission signals precisely indicates the failure
location.
References 349

3400
3200
3000
2800 range of AE onset
2600
axial source position [mm]

2400
2200
2000
1800
1600
1400
1200
1000 clevis fastener
800
600
400
200
0
–200
0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.1
load ratio

Fig. 4.173 Axial acoustic emission source position as function of the load-ratio indicating shift
toward failure location for the eight load cycles with incremental peak load (left to right)

References

1. Lacidogna, G., Carpinteri, A., Manuello, A., Durin, G., Schiavi, A., Niccolini, G., Agosto, A.:
Acoustic and electromagnetic emissions as precursor phenomena in failure processes. Strain
47, 144–152 (2011)
2. Giordano, M., Condelli, L., Nicolais, L.: Acoustic emission wave propagation in a viscoelas-
tic plate. Compos. Sci. Technol. 59, 1735–1743 (1999)
3. Hamstad, M.A.: Thirty years of advances and some remaining challenges in the application of
acoustic emission to composite materials. In: Kishi, T., Ohtsu, M., Yuyama, S. (eds.)
Acoustic Emission Beyond the Millennium, pp. 77–91. Elsevier Science, Amsterdam (2000)
4. Ono, K., Gallego, A.: Research and applications of AE on advanced composites. J. Acoust.
Emiss. 30, 180–229 (2012)
5. Reinhardt, H.W., Grosse, C.U., Kurz, J.H.: Localization and mode determination of fracture
events by acoustic emission. In: Carpinteri, A., Lacidogna, G. (eds.) Acoustic Emission and
Critical Phenomena. Taylor & Francis Group, London (2008)
6. Prosser, W.H., Jackson, K.E., Kellas, S., Smith, B.T., McKeon, J., Friedman, A.: Advanced
waveform-based acoustic emission detection of matrix cracking in composites. Mater. Eval.
53, 1052–1058 (1995)
7. Huguet, S., Godin, N., Gaertner, R., Salmon, L., Villard, D.: Use of acoustic emission to
identify damage modes in glass fibre reinforced polyester. Compos. Sci. Technol. 62,
1433–1444 (2002)
8. Scholey, J.J., Wilcox, P.D., Wisnom, M.R., Friswell, M.I.: Quantitative experimental mea-
surements of matrix cracking and delamination using acoustic emission. Compos. Part A
Appl. Sci. Manuf. 41, 612–623 (2010)
9. Giordano, M., Calabro, A., Esposito, C., D’Amore, A., Nicolais, L.: An acoustic-emission
characterization of the failure modes in polymer-composite materials. Compos. Sci. Technol.
58, 1923–1928 (1998)
350 4 Acoustic Emission

10. Wilcox, P.D., Lee, C.K., Scholey, J.J., Friswell, M.I., Wisnom, M.R., Drinkwater, B.W.:
Progress towards a forward model of the complete acoustic emission process. Adv. Mater.
Res. 13–14, 69–75 (2006)
11. Prosser, W.H., Hamstad, M.A., Gary, J., Gallagher, A.O.: Finite element and plate theory
modeling of acoustic emission waveforms. J. Nondestruct. Eval. 18, 83–90 (1999)
12. Sause, M.G.R., Horn, S.: Simulation of Lamb wave excitation for different elastic properties
and acoustic emission source geometries. J. Acoust. Emiss. 28, 142–154 (2010)
13. Sause, M.G.R., Richler, S.: Finite element modelling of cracks as acoustic emission sources.
J. Nondestruct. Eval. 34, 1–13 (2015)
14. Livne, A., Bouchbinder, E., Svetlizky, I., Fineberg, J.: The near-tip fields of fast cracks.
Science 327, 1359–1363 (2010)
15. Livne, A., Bouchbinder, E., Fineberg, J.: Breakdown of linear elastic fracture mechanics near
the tip of a rapid crack. Phys. Rev. Lett. 101, 1–4 (2008)
16. Landau, L.D., Lifschitz, E.M.: Elastizitätstheorie. Akademie Verlag GmbH, Berlin (1987)
17. Freund, L.B.: The initial wave front emitted by a suddenly extending crack in an elastic solid.
J. Appl. Mech. 39, 601–602 (1972)
18. Aki, K., Richards, P.G.: Quantitative Seismology, Theory and Methods. University Science,
Sausalito (1980)
19. Ohtsu, M.: Source mechanism and waveform analysis of acoustic emission in concrete.
J. Acoust. Emiss. 2, 103–112 (1982)
20. Lysak, M.V.: Development of the theory of acoustic emission by propagating cracks in terms
of fracture mechanics. Eng. Fract. Mech. 55, 443–452 (1996)
21. Ohtsu, M., Ono, K.: A generalized theory of acoustic emission and Green’s function in a half
space. J. Acoust. Emiss. 3, 27–40 (1984)
22. Ohtsu, M., Ono, K.: The generalized theory and source representation of acoustic emission.
J. Acoust. Emiss. 5, 124–133 (1986)
23. Hamstad, M.A., O’Gallagher, A., Gary, J.: Modeling of buried monopole and dipole sources
of acoustic emission with a finite element technique. J. Acoust. Emiss. 17, 97–110 (1999)
24. Sause, M.G.R., Horn, S.: Simulation of acoustic emission in planar carbon fiber reinforced
plastic specimens. J. Nondestruct. Eval. 29, 123–142 (2010)
25. Hamstad, M.A.: Frequencies and amplitudes of AE signals in a plate as a function of source
rise time. In: 29th European Conference on Acoustic Emission Testing, Vienna, Austria,
2010, pp. 1–8.
26. Scruby, C.B., Buttle, D.J.: Quantitative fatigue crack measurement by acoustic emission. In:
Marsh, K.J., Smith, R., Ritchie, R.O. (eds.) Crack Measurement: Techniques and Applica-
tions, pp. 207–287. Engineering Materials Advisory Services Ltd., West Midlands (1992)
27. Wadley, H.N.G., Scruby, C.B.: Acoustic Emission Source Characterization. Advances in
Acoustic Emission. Dunhart, Knoxville (1981)
28. Scruby, C.B.: Quantitative acoustic emission techniques. Nondestruct. Test. 8, 141–208
(1985)
29. Green, E.R.: Acoustic emission sources in a cross-ply laminated plate. Compos. Eng. 5,
1453–1469 (1995)
30. Green, E.R.: Acoustic emission in composite laminates. J. Nondestruct. Eval. 17, 117–127
(1998)
31. Hamstad, M.A., O’Gallagher, A., Gary, J.: A wavelet transform applied to acoustic emission
signals: part 1: source identification. J. Acoust. Emiss. 20, 39–61 (2002)
32. Downs, K.S., Hamstad, M.A., O’Gallagher, A.: Wavelet transform signal processing to
distinguish different acoustic emission sources. J. Acoust. Emiss. 21, 52–69 (2003)
33. Hora, P., Cervena, O.: Acoustic emission source modeling. Appl. Comput. Mech. 4, 25–36
(2010)
34. Hamstad, M.A., O’Gallagher, A., Gary, J.: A wavelet transform applied to acoustic emission
signals: part 2: source location. J. Acoust. Emiss. 20, 62–82 (2002)
References 351

35. Sause, M.: Identification of Failure Mechanisms in Hybrid Materials Utilizing Pattern
Recognition Techniques Applied to Acoustic Emission Signals. mbv-Verlag, Berlin (2010)
36. Sause, M.G.R.: Modelling of crack growth based acoustic emission release in aluminum
alloys. In: 31st Conference of the European Working Group on Acoustic Emission, pp. 1–8.
Dresden, Germany (2014)
37. Kalafat, S., Zelenyak, A.-M., Sause, M.G.R.: In-situ monitoring of composite failure by
computing tomography and acoustic emission. In: 20th International Conference on Com-
posite Materials, pp. 1–8. Copenhagen, Denmark (2015)
38. Juhasz, T.: Ein neues physikalisch basiertes Versagenskriterium f€ ur schwach 3D-verstärkte
Faserverbundlaminate. PhD-thesis, University Carolo-Wilhelmina Braunschweig (2003)
39. Puck, A.: Festigkeitsanalyse von Faser-Matrix-Laminaten Modelle f€ ur die Praxis. Carl
Hanser Verlag, Munich (1996)
40. Green, A.E., Zerna, W.: Theoretical Elasticity. Oxford University Press, New York (2002)
41. Scott, A.E., Sinclair, I., Spearing, S.M., Mavrogordato, M., Bunsell, A.R., Thionnet, A.:
Comparison of the accumulation of fibre breaks occurring in a unidirectional carbon/epoxy
composite identified in a multi-scale micro-mechanical model with that of experimental
observations using high resolution computed tomography. In: Matériaux 2010, pp. 1–9.
Nantes, France (2010)
42. Boltz, E.S., Fortunko, C.M., Hamstad, M.A., Renken, M.C.: Absolute sensitivity of air, light
and direct-coupled wideband acoustic emission transducers. In: Thompson, D.O., Chimenti,
D.E. (eds.) Review of Progress in Quantitative Nondestructive Evaluation, pp. 967–974.
Springer, Boston (1995)
43. Tatro, C.A.: Design criteria for acoustic emission experimentation. In: Acoustic Emission
ASTM STP 505, pp. 84–99 (1972)
44. Stoneley, R.: Elastic waves at the surface of separation of two solids. Proc. R. Soc. Lond.
A. 106, 416–428 (1924)
45. Love, A.E.H.: Some Problems of Geodynamics. University Press, Cambridge (1911)
46. Lamb, H.: On waves in an elastic plate. Proc. R. Soc. Lond. A. 93, 114–128 (1917)
47. Rayleigh, L.: On waves propagated along the plane surface of an elastic solid. Proc. Lond.
Math. Soc. s1–17, 4–11 (1885)
48. Krautkrämer, J., Krautkrämer, H.: Ultrasonic Testing of Materials. Springer, Berlin (1983)
49. Cremer, L., Heckl, M.: K€orperschall Physikalische Grundlagen und technische
Anwendungen. Springer Verlag, Berlin (1996)
50. Redwood, M.: Mechanical Waveguides; The Propagation of Acoustic and Ultrasonic Waves
in Fluids and Solids with Boundaries. Pergamon Press, New York (1960)
51. Bergman, E.H., Shahbender, R.: Effect of statically applied stresses on the velocity of
propagation of ultrasonic waves. J. Appl. Phys. 29, 1736–1738 (1958)
52. Grosse, C.U., Ohtsu, M.: Acoustic Emission Testing. Springer, Berlin (2008)
53. Sause, M.G.R., Hamstad, M.A., Horn, S.: Finite element modeling of Lamb wave propaga-
tion in anisotropic hybrid materials. Compos. Part B Eng. 53, 249–257 (2013)
54. Lowe, M.J.S.: Matrix techniques for modeling ultrasonic waves in multilayered media. IEEE
Trans. Ultrason. Ferroelectr. Freq. Control 42, 525–542 (1995)
55. Castaings, M., Bacon, C., Hosten, B., Predoi, M.V.: Finite element predictions for the
dynamic response of thermo-viscoelastic material structures. J. Acoust. Soc. Am. 115,
1125 (2004)
56. Heidary, Z., Ozevin, D.: On the influences of boundary reflections and piezoelectric sensors
to the characteristics of elastic waves for pattern recognition methods. J. Nondestruct. Eval.
34, 271 (2014)
57. Raghavan, A., Cesnik, C.E.S.: Review of guided-wave structural health monitoring. Shock
Vib. Dig. 39, 91–114 (2007)
58. Thomson, W.T.: Transmission of elastic waves through a stratified solid medium. J. Appl.
Phys. 21, 89 (1950)
352 4 Acoustic Emission

59. Haskell, N.A.: The dispersion of surface waves on multilayered media. Bull. Seismol. Soc.
Am. 43, 17–34 (1953)
60. Kundu, T., Mal, A.: Elastic waves in a multilayered solid due to a point source. Wave Motion
7, 459–471 (1985)
61. Sause, M.G.R., Horn, S.: Influence of internal discontinuities on ultrasonic signal propagation
in carbon fiber reinforced plastics. In: 30th European Conference on Acoustic Emissionm
pp. 1–11. Granada, Spain (2012)
62. Sause, M.G.R.: Acoustic emission signal propagation in damaged composite structures.
J. Acoust. Emiss. 31, 1–18 (2013)
63. Auld, B.A.: Acoustic Fields ans Waves in Solids. Krieger, Malabar (1990)
64. Hosten, B.: Heterogeneous structure of modes and Kramers-Kronig relationship in aniso-
tropic viscoelastic materials. J. Acoust. Soc. Am. 104, 1382–1388 (1998)
65. Calomfirescu, M., Herrmann, A.: Attenuation of Lamb waves in composites: models and
possible applications. In: Proceedings of the 6th International Workshop on Structural Health
Monitoring, pp. 1–8. Stanford, CA, USA (2007)
66. Pollock, A.A.: Classical wave theory in practical AE testing. In: Proceedings of the 8th
International AE Symposium, pp. 708–721. Tokyo, Japan (1986)
67. Prosser, W.H.: Advanced AE techniques in composite materials research. J. Acoust. Emiss.
14, 1–11 (1996)
68. Neau, G., Deschamps, M., Lowe, M.J.S.: Group velocity of Lamb waves in anisotropic
plates: comparison between theory and experiments. In: AIP Conference Proceedings,
pp. 81–88 (2001)
69. Ward, I.M.: Mechanical Properties of Solid Polymers. Wiley, New York (1971)
70. Gallego, A., Ono, K.: An improved acousto-ultrasonic scheme with Lamb wave mode
separation and damping factor in CFRP plates. J. Acoust. Emiss. 30, 109–123 (2012)
71. Choi, H.-I., Williams, W.: Improved time-frequency representation of multicomponent
signals using exponential kernels. IEEE Trans. Acoust. Speech Signal Process. 37,
862–872 (1989)
72. Beattie, A.G.: Acoustic emission, principles and Instrumentation. J. Acoust. Emiss. 2, 95–128
(1983)
73. Gautschi, G.: Piezoelectric Sensorics. Springer, Berlin (2002)
74. Breckenridge, F.R.: Acoustic emission: some applications of Lamb’s problem. J. Acoust.
Soc. Am. 57, 626 (1975)
75. Scruby, C.B., Wadley, H.N.G.: A calibrated capacitance transducer for the detection of
acoustic emission. J. Phys. D Appl. Phys. 11, 1487–1494 (1978)
76. Read, I., Foote, P., Murray, S.: Optical fibre acoustic emission sensor for damage detection in
carbon fibre composite structures. Meas. Sci. Technol. 13, N5–N9 (2002)
77. de Oliveira, R., Fraz~ao, O., Santos, J.L., Marques, A.T.: Optic fibre sensor for real-time
damage detection in smart composite. Comput. Struct. 82, 1315–1321 (2004)
78. Wild, G., Hinckley, S.: Acousto-ultrasonic optical fiber sensors: overview and state-of-the-
art. IEEE Sens. J. 8, 1184–1193 (2008)
79. Wild, G., Hinckley, S.: Fiber Bragg grating sensors for acoustic emission and transmission
detection applied to robotic NDE in structural health monitoring. In: 2007 I.E. Sensors
Applications Symposium, pp. 1–6. IEEE (2007)
80. Watanabe, M., Enoki, M., Kishi, T.: Fracture behavior of ceramic coatings during thermal
cycling evaluated by acoustic emission method using laser interferometers. Mater. Sci. Eng.
A 359, 368–374 (2003)
81. Enoki, M., Watanabe, M., Chivavibul, P., Kishi, T.: Non-contact measurement of acoustic
emission in materials by laser interferometry. Sci. Technol. Adv. Mater. 1, 157–165 (2000)
82. Bohse, J.: Damage analysis of polymer matrix composites by acoustic emission testing. In:
EWGAE 2004—26th European Conference on Acoustic Emission Testing, pp. 339–348.
Berlin, Germany (2004)
References 353

83. Breckenridge, F.R., Greenspan, M.: Surface-wave displacement: absolute measurements


using a capacitance transducer. J. Acoust. Soc. Am. 69, 1177–1185 (1981)
84. Grosse, C.U.: Quantitative non-destructive testing of construction materials using acoustic
emission technique and ultrasound. PhD-thesis, University of Stuttgart (1996)
85. Goujon, L., Baboux, J.C.: Behaviour of acoustic emission sensors using broadband calibra-
tion techniques. Meas. Sci. Technol. 14, 903–908 (2003)
86. Hamstad, M.A.: Re-examination of NIST acoustic emission absolute sensor calibration: part
II: finite element modeling of acoustic emission signal from glass capillary fracture.
J. Acoust. Emiss. 29, 175–183 (2011)
87. Sause, M.G.R., Hamstad, M.A., Horn, S.: Finite element modeling of conical acoustic
emission sensors and corresponding experiments. Sens. Actuators A Phys. 184, 64–71 (2012)
88. Hamstad, M.A.: Improved signal-to-noise wideband acoustic/ultrasonic contact displacement
sensors for wood and polymers. Wood Fiber Sci. 29, 239–248 (1997)
89. McLaskey, G.C., Glaser, S.D.: Acoustic emission sensor calibration for absolute source
measurements. J. Nondestruct. Eval. 31, 157–168 (2012)
90. Huang, Y.-H., Ma, C.-C.: Forced vibration analysis of piezoelectric quartz plates in reso-
nance. Sens. Actuators A Phys. 149, 320–330 (2009)
91. Ohtsu, M., Ono, K.: Resonance analysis of piezoelectric transducer elements. J. Acoust.
Emiss. 2, 247–260 (1983)
92. Ono, K., Cho, H., Matsuo, T.: Transfer functions of acoustic emission sensors. J. Acoust.
Emiss. 26, 72–90 (2008)
93. Feng, G.-H., Tsai, M.-Y.: Acoustic emission sensor with structure-enhanced sensing mech-
anism based on micro-embossed piezoelectric polymer. Sens. Actuators A Phys. 162,
100–106 (2010)
94. Or, S.W., Chan, H.L.W., Choy, C.L.: P(VDF-TrFE) copolymer acoustic emission sensors.
Sens. Actuators A Phys. 80, 237–241 (2000)
95. Barbezat, M., Brunner, A.J., Fl€ueler, P., Huber, C., Kornmann, X.: Acoustic emission sensor
properties of active fibre composite elements compared with commercial acoustic emission
sensors. Sens. Actuators A Phys. 114, 13–20 (2004)
96. Marin-Franch, P., Martin, T., Tunnicliffe, D.L., Das-Gupta, D.K.: PTCa/PEKK piezo-
composites for acoustic emission detection. Sens. Actuators A Phys. 99, 236–243 (2002)
97. Zheng, S.X., McBride, R., Barton, J.S., Jones, J.D.C., Hale, K.F., Jones, B.E.: Intrinsic optical
fibre sensor for monitoring acoustic emission. Sens. Actuators A Phys. 31, 110–114 (1992)
98. Nieuwenhuis, J.H., Neumann, J.J., Greve, D.W., Oppenheim, I.J.: Generation and detection
of guided waves using PZT wafer transducers. IEEE Trans. Ultrason. Ferroelectr. Freq.
Control 52, 2103–2111 (2005)
99. Sause, M.G.R.: Investigation of pencil-lead breaks as acoustic emission sources. J. Acoust.
Emiss. 29, 184–196 (2011)
100. Greenspan, M.: The NBS conical transducer: analysis. J. Acoust. Soc. Am. 81, 173 (1987)
101. McLaskey, G.C., Glaser, S.D.: Hertzian impact: experimental study of the force pulse and
resulting stress waves. J. Acoust. Soc. Am. 128, 1087–1096 (2010)
102. Lynnworth, L.C., Umina, J.A.: Extensional bundle waveguide techniques for measuring flow
of hot fluids. IEEE Trans. Ultrason. Ferroelectr. Freq. Control 52, 538–544 (2005)
103. Hamstad, M.A.: Small diameter waveguide for wideband acoustic emission. J. Acoust.
Emiss. 24, 234–247 (2006)
104. Rose, J.L.: Ultrasonic Waves in Solid Media. University Press, Cambridge (2004)
105. Zelenyak, A.-M., Hamstad, M.A., Sause, M.G.R.: Finite element modeling of acoustic
emission signal propagation with various shaped waveguides. In: 31st Conference of the
European Working Group on Acoustic Emission, pp. 1–8. Dresden, Germany (2014)
106. Zelenyak, A.-M., Hamstad, M., Sause, M.: Modeling of acoustic emission signal propagation
in waveguides. Sensors 15, 11805–11822 (2015)
354 4 Acoustic Emission

107. Ernst, R., Dual, J.: Acoustic emission source detection using the time reversal principle on
dispersive waves in beams. In: Proceedings of the 2013 International Congress on Ultrasonics
(ICU 2013), pp. 87–92. Singapore (2013)
108. Ciampa, F., Meo, M.: Acoustic emission source localization and velocity determination of the
fundamental mode A0 using wavelet analysis and a Newton-based optimization technique.
Smart Mater. Struct. 19, 045027 (2010)
109. Ozevin, D., Heidary, Z.: Acoustic emission source orientation based on time scale. J. Acoust.
Emiss. 29, 123–132 (2011)
110. Gorman, M.R., Prosser, W.H.: AE source orientation by plate wave analysis. J. Acoust.
Emiss. 9, 283–288 (1991)
111. Gorman, M.R.: Plate wave acoustic emission. J. Acoust. Soc. Am. 90, 358 (1991)
112. Gorman, M.R., Ziola, S.M.: Plate waves produced by transverse matrix cracking. Ultrasonics
29, 245–251 (1991)
113. Prosser, W.H.: The Propagation Characteristics of the Plate Modes of Acoustic Emission
Waves in Thin Aluminum Plates and Thin Graphite/Epoxy Composite Plates and Tubes
(1991)
114. Morscher, G.N.: Modal acoustic emission of damage accumulation in a woven SiC/SiC
composite. Vacuum 59, 687–697 (1999)
115. Surgeon, M., Wevers, M.: Modal analysis of acoustic emission signals from CFRP laminates.
NDT E Int. 32, 311–322 (1999)
116. Prosser, W.H., Jackson, K.E., Kellas, S., Smith, B.T., McKeon, J., Friedman, A.: Evaluation
of damage in metal matrix composites by means of acoustic emission monitoring. NDT E Int.
30, 108 (1997)
117. Gorman, M.R.: Modal AE analysis of fracture and failure in composite materials, and the
quality and life of high pressure composite pressure cylinders. J. Acoust. Emiss. 29, 1–28
(2011)
118. Anastassopoulos, A.A., Philippidis, T.P.: Clustering methodology for the evaluation of
acoustic emission from composites. J. Acoust. Emiss. 13, 11–21 (1995)
119. Philippidis, T., Nikolaidis, V., Anastassopoulos, A.: Damage characterisation of C/C lami-
nates using neural network techniques on AE signals. NDT E Int. 31, 329–340 (1998)
120. Richardson, J.M., Elsley, R.K., Graham, L.J.: Nonadaptive, semi-adaptive and adaptive
approaches to signal processing problems in nondestructive evaluation. Pattern Recognit.
Lett. 2, 387–394 (1984)
121. Vi-Tong, E., Gaillard, P.: An algorithm for non-supervised sequential classification of
signals. Pattern Recognit. Lett. 5, 307–313 (1987)
122. Ramirez-Jimenez, C.R., Papadakis, N., Reynolds, N., Gan, T.H., Purnell, P., Pharaoh, M.:
Identification of failure modes in glass/polypropylene composites by means of the primary
frequency content of the acoustic emission event. Compos. Sci. Technol. 64, 1819–1827
(2004)
123. Marec, A., Thomas, J.-H., Guerjouma, R.: Damage characterization of polymer-based com-
posite materials: multivariable analysis and wavelet transform for clustering acoustic emis-
sion data. Mech. Syst. Signal Process. 22, 1441–1464 (2008)
124. Sause, M.G.R., Haider, F., Horn, S.: Quantification of metallic coating failure on carbon fiber
reinforced plastics using acoustic emission. Surf. Coat. Technol. 204, 300–308 (2009)
125. Sause, M.G.R., Gribov, A., Unwin, A.R., Horn, S.: Pattern recognition approach to identify
natural clusters of acoustic emission signals. Pattern Recognit. Lett. 33, 17–23 (2012)
126. Doan, D.D., Ramasso, E., Placet, V., Boubakar, L., Zerhouni, N.: Application of an
unsupervised pattern recognition approach for AE data originating from fatigue tests on
CFRP. In: 31st Conference of the European Working Group on Acoustic Emission, pp. 1–8.
Dresden, Germany (2014)
127. Anastassopoulos, A.A., Nikolaidis, V.N., Philippidis, T.P.: A comparative study of pattern
recognition algorithms for classification of ultrasonic signals. Neural Comput. Appl. 8, 53–66
(1999)
References 355

128. Yu, P., Anastassopoulos, V., Venetsanopoulos, A.N.: Pattern recognition based on morpho-
logical shape analysis and neural networks. Math. Comput. Simul. 40, 577–595 (1996)
129. Baensch, F., Sause, M.G.R., Brunner, A.J., Niemz, P.: Damage evolution in wood—pattern
recognition based on acoustic emission (AE) frequency spectra. Holzforschung 69, 1–9
(2015)
130. Kostopoulos, V., Loutas, T., Kontsos, A., Sotiriadis, G., Pappas, Y.: On the identification of
the failure mechanisms in oxide/oxide composites using acoustic emission. NDT E Int. 36,
571–580 (2003)
131. Bohse, J., Chen, J.: Acoustic emission examination of mode I, mode II and mixed-mode I/II
interlaminar fracture of unidirectional fiber-reinforced polymers. J. Acoust. Emiss. 19, 1–10
(2001)
132. Haselbach, W., Lauke, B.: Acoustic emission of debonding between fibre and matrix to
evaluate local adhesion. Compos. Sci. Technol. 63, 2155–2162 (2003)
133. Li, L., Lomov, S.V., Yan, X., Carvelli, V.: Cluster analysis of acoustic emission signals for
2D and 3D woven glass/epoxy composites. Compos. Struct. 116, 286–299 (2014)
134. Li, L., Lomov, S.V., Yan, X.: Correlation of acoustic emission with optically observed
damage in a glass/epoxy woven laminate under tensile loading. Compos. Struct. 123,
45–53 (2015)
135. Bishop, C.M.: Neural Networks for Pattern Recognition. Clarendon, Oxford (1995)
136. Polikar, R.: Pattern recognition. In: Akay, M. (ed.) Wiley Encyclopedia of Biomedical
Engineering, pp. 1–22. Wiley, Hoboken (2006)
137. Sause, M.G.R., Schultheiß, D., Horn, S.: Acoustic emission investigation of coating fracture
and delamination in hybrid carbon fiber reinforced plastic structures. J. Acoust. Emiss. 26,
1–13 (2008)
138. Sause, M.G.R., M€uller, T., Horoschenkoff, A., Horn, S.: Quantification of failure mecha-
nisms in mode-I loading of fiber reinforced plastics utilizing acoustic emission analysis.
Compos. Sci. Technol. 72, 167–174 (2012)
139. Ritschel, F., Sause, M.G.R., Brunner, A.J., Niemz, P.: Acoustic emission (AE) signal clas-
sification from tensile tests on plywood and layered wood. In: 31st Conference of the
European Working Group on Acoustic Emission, pp. 1–7. Dresden, Germany (2014)
140. Vergeynst, L.L., Sause, M.G.R., Steppe, K.: Acoustic emission signal detection in drought-
stressed trees: beyond counting hits. In: 31st Conference of the European Working Group on
Acoustic Emission, pp. 1–8. Dresden, Germany (2014)
141. Njuhovic, E., Bräu, M., Wolff-Fabris, F., Starzynski, K., Altstädt, V.: Identification of
interface failure mechanisms of metallized glass fibre reinforced composites using acoustic
emission analysis. Compos. Part B Eng. 66, 443–452 (2014)
142. Kempf, M., Skrabala, O., Altstädt, V.: Acoustic emission analysis for characterisation of
damage mechanisms in fibre reinforced thermosetting polyurethane and epoxy. Compos. Part
B Eng. 56, 477–483 (2014)
143. Priess, T., Sause, M.G., Fischer, D., Middendorf, P.: Detection of delamination onset in laser-
cut carbon fiber transverse crack tension specimens using acoustic emission. J. Compos.
Mater. 49, 2639–2647 (2015)
144. Tou, J.T.: DYNOC—a dynamic optimal cluster-seeking technique. Int. J. Comput. Inf. Sci. 8,
541–547 (1979)
145. Davies, D.L., Bouldin, D.W.: A cluster separation measure. IEEE Trans. Pattern Anal. Mach.
Intell. 1, 224–227 (1979)
146. Rousseeuw, P.J.: Silhouettes: a graphical aid to the interpretation and validation of cluster
analysis. J. Comput. Appl. Math. 20, 53–65 (1987)
147. Hubert, L.J., Arabie, P.: Comparing partitions. J. Classif. 2, 193–218 (1985)
148. G€unter, S., Bunke, H.: Validation indices for graph clustering. Pattern Recognit. Lett. 24,
1107–1113 (2003)
149. Placet, V., Ramasso, E., Boubakar, L., Zerhouni, N.: Online segmentation of acoustic
emission data streams for detection of damages in composites structures in unconstrained
356 4 Acoustic Emission

environments. In: 11th International Conference on Structural Safety & Reliability, pp. 1–8.
New York, USA (2013)
150. Serir, L., Ramasso, E., Zerhouni, N.: Evidential evolving Gustafson–Kessel algorithm for
online data streams partitioning using belief function theory. Int. J. Approx. Reason. 53,
747–768 (2012)
151. MacQueen, J.B.: Some methods for classification and analysis of multivariate observations.
In: Proceedings of 5th Berkeley Symposium on Mathematical Statistics and Probability,
pp. 281–297 (1967)
152. Baensch, F., Zauner, M., Sanabria, S.J., Sause, M.G.R., Pinzer, B.R., Brunner, A.J.,
Stampanoni, M., Niemz, P.: Damage evolution in wood: synchrotron radiation micro-
computed tomography (SRμCT) as a complementary tool for interpreting acoustic emission
(AE) behavior. Holzforschung 69 (2015)
153. Eaton, M.J., Holford, K.M., Featherston, C.A., Pullin, R.: Damage in carbon fibre compos-
ites: the discrimination of acoustic emission signals using frequency. J. Acoust. Emiss. 25,
140–148 (2007)
154. Sause, M.G.R., Horn, S.: Quantification of the uncertainty of pattern recognition approaches
applied to acoustic emission signals. J. Nondestruct. Eval. 32, 242–255 (2013)
155. Milligan, G.W.: An algorithm for generating artificial test clusters. Psychometrika 50,
123–127 (1985)
156. Qiu, W., Joe, H.: Generation of random clusters with specified degree of separation.
J. Classif. 23, 315–334 (2006)
157. Rand, W.M.: Objective criteria for the evaluation of clustering methods. J. Am. Stat. Assoc.
66, 846–850 (1971)
158. Kalafat, S., Sause, M.G.R.: Localization of acoustic emission sources in fiber composites
using artificial neural networks. In: 31st Conference of the European Working Group on
Acoustic Emission, pp. 1–8. Dresden, Germany (2014)
159. Sause, M.G.R., Kalafat, S., Zelenyak, A., Hoeck, B., Horn, S.: Acoustic emission source
localization in bearing tests of fiber reinforced polymers by neural networks. In: 16th
International Conference on Experimental Mechanics, pp. 1–3. Cambridge (2014)
160. Pl€ockl, M., Sause, M.G.R., Scharringhausen, J., Horn, S.: Failure analysis of NOL-ring
specimens by acoustic emission. In: 30th European Conference on Acoustic Emission,
pp. 1–12. Granada, Spain (2012)
161. Kurz, J.H.: Verifikation von Bruchprozessen bei gleichzeitiger Automatisierung der
Schallenmissionsanalyse an Stahl- und Stahlfaserbeton. University of Stuttgart, Stuttgart
(2006)
162. Akaike, H.: Markovian representation of stochastic process and its application to the analysis
of autoregressive moving average processes. Ann. Inst. Stat. Math. 26, 363–387 (1974)
163. K€uhnicke, H., Schulze, E., Voigt, D.: Verbesserte Lokalisation mittels Signalformanalyse. In:
DGZfP-BB, pp. 44–51 (2007)
164. Hamstad, M.A.: Comparison of wavelet transform and Choi-Williams distribution to deter-
mine group velocities for different acoustic emission sensors. J. Acoust. Emiss. 26, 40–59
(2008)
165. Hamstad, M.A., O’Gallagher, A.: Effects of noise on Lamb-mode acoustic-emission arrival
times determined by wavelet transform. J. Acoust. Emiss. 23, 1–24 (2005)
166. Bancroft, S.: An algebraic solution of the GPS equations. IEEE Trans. Aerosp. Electron. Syst.
21, 56–59 (1985)
167. Pullin, R., Baxter, M., Eaton, M.M.J., Holford, K.M., Evans, S.L.: Novel acoustic emission
source location. J. Acoust. Emiss. 25, 215–223 (2007)
168. Eaton, M.J., Pullin, R., Holford, K.M., Featherston, C.A.: AE wave propagation and novel
source location in composite plates. In: 28th European Conference on Acoustic Emission
Testing, Berlin, Germany (2008)
169. Baxter, M.G., Pullin, R., Holford, K.M., Evans, S.L.: Delta T source location for acoustic
emission. Mech. Syst. Signal Process. 21, 1512–1520 (2007)
References 357

170. Eaton, M.J., Pullin, R., Holford, K.M.: Acoustic emission source location in composite
materials using Delta T Mapping. Compos. Part A Appl. Sci. Manuf. 43, 856–863 (2012)
171. Chlada, M., Prevorovsky, Z., Blahacek, M.: Neural network AE source location apart from
structure size and material. J. Acoust. Emiss. 28, 99–108 (2010)
172. Blahacek, M., Chlada, M., Prevorovsky, Z.: Acoustic emission source location based on
signal features. Adv. Mater. Res. 13–14, 77–82 (2006)
173. Kalafat, S., Sause, M.G.R.: Lokalisierung von Schallemissionsquellen in
Faserverbundwerkstoffen mit k€unstlichen neuronalen Netzwerken. In: 19. Kolloquium
Schallemission, pp. 1–8. Augsburg, Germany (2013)
174. Kalafat, S., Sause, M.G.: Acoustic emission source localization by artificial neural networks.
Struct. Heal. Monit. 1–15 (2015).
175. Balageas, D., Fritzen, C.-P., Gemes, A.: Structural Health Monitoring. ISTE, London (2006)
176. Ciang, C.C., Lee, J.-R., Bang, H.-J.: Structural health monitoring for a wind turbine system: a
review of damage detection methods. Meas. Sci. Technol. 19, 122001 (2008)
177. Sause, M.G.R., Horn, S.R.: Influence of specimen geometry on acoustic emission signals in
fiber reinforced composites: FEM-simulations and experiments. In: 29th European Confer-
ence on Acoustic Emission Testing, pp. 1–8. Vienna, Austria (2010)
178. Sause, M.G.R., Scharringhausen, J., Horn, S.R.: Identification of failure mechanisms in
thermoplastic composites by acoustic emission measurements. In: 19th International Confer-
ence on Composite Materials, Montreal, Canada (2013)
179. Burks, B., Kumosa, M.: A modal acoustic emission signal classification scheme derived from
finite element simulation. Int. J. Damage Mech. 23, 43–62 (2013)
180. Pahr, D.H., Rammerstorfer, F.G., Rosenkranz, P., Humer, K., Weber, H.W.: A study of short-
beam-shear and double-lap-shear specimens of glass fabric/epoxy composites. Compos. Part
B Eng. 33, 125–132 (2002)
181. Sause, M.G.R., Pl€ockl, M., Horn, S.R., Forberich, B., Scharringhausen, J.: Bestimmung der
GIc und GIIc Kennwerte von thermoplastischen Faserverbundwerkstoffen mittels
Schallemissionsanalyse. In: Verbundwerkstoffe und Werkstoffverbunde 2013, Bayreuth,
Germany (2013)
182. Spruiell, J.E.: A Review of the Measurement and Development of Crystallinity and Its
Relation to Properties in Neat Poly(Phenylene Sulfide) and Its Fiber Reinforced Composites.
Oak Ridge, Tennesse, USA (2005)
183. Hahn, H.T., Lagace, P.A., O’Brien, T.K.: Composite Materials: Fatigue and Fracture. ASTM
International, West Conshohocken (1991)
184. Wisnom, M.R.: On the increase in fracture energy with thickness in delamination of unidi-
rectional glass fibre-epoxy with cut central plies. J. Reinf. Plast. Compos. 11, 897–909 (1992)
185. Cui, W., Wisnom, M.R., Jones, M.: An experimental and analytical study of delamination of
unidirectional specimens with cut central plies. J. Reinf. Plast. Compos. 13, 722–739 (1994)
186. Sause, M.G.R., Monden, A.: Comparison of predicted onset of failure mechanisms by
nonlinear failure theory and by acoustic emission measurements. In: 16th European Confer-
ence on Composite Materials. Sevilla, Spain (2014)
187. Scott, A.E., Mavrogordato, M., Wright, P., Sinclair, I., Spearing, S.M.: In-situ fibre fracture
measurement in carbon-epoxy laminates using high resolution computed tomography.
Compos. Sci. Technol. 71, 1471–1477 (2011)
188. Zhou, Y., Jiang, D., Xia, Y.: Tensile mechanical behavior of T300 and M40J fiber bundles at
different strain rate. J. Mater. Sci. 36, 919–922 (2001)
189. Durand, L.P.: Composite Materials Research Progress. Nova Science, New York (2008)
190. Lomov, S., Karahan, M., Bogdanovich, A., Verpoest, I.: Monitoring of acoustic emission
damage during tensile loading of 3D woven carbon/epoxy composites. Text. Res. J. 84,
1373–1384 (2014)
191. Nairn, J.A., Hu, S.: The formation and effect of outer-ply microcracks in cross-ply laminates:
a variational approach. Eng. Fract. Mech. 41, 203–221 (1992)
358 4 Acoustic Emission

192. Nairn, J., Hu, S.: The initiation and growth of delaminations induced by matrix microcracks
in laminated composites. Int. J. Fract. 57, 1–24 (1992)
193. Matthias Deuschle, H., Kroplin, B.-H.: Finite element implementation of Puck’s failure
theory for fibre-reinforced composites under three-dimensional stress. J. Compos. Mater.
46, 2485–2513 (2012)
194. Deuschle, H.M.: 3D Failure Analysis of UD Fibre Reinforced Composites: Puck’s Theory
Within FEA. PhD-thesis, University of Stuttgart (2010)
195. Puck, A., Mannigel, M.: Physically based stress-strain relations for the inter-fibre-fracture
analysis of FRP laminates. Compos. Sci. Technol. 67, 1955–1964 (2007)
196. Puck, A., Mannigel, M.: Physically based non-linear stress-strain relations for the inter-fibre
fracture analysis of FRP laminates. Compos. Sci. Technol. 67, 1955–1964 (2007)
197. Wisnom, M.R.: Size effects in the testing of fibre-composite materials. Compos. Sci.
Technol. 59, 1937–1957 (1999)
198. Moosburger-Will, J., Sause, M.G.R., Horny, R., Horn, S., Scholler, J., Llopard Prieto, L.:
Joining of carbon fiber reinforced polymer laminates by a novel partial cross-linking process.
J. Appl. Polym. Sci. 132, 42159 (2015)
199. Bohse, J., Chen, J., Brunner, A.: Acoustic emission analysis and micro-mechanical interpre-
tation of mode I fracture toughness tests on composite materials. Fract. Polym. Compos.
Adhes. 27, 15–26 (2000)
200. Carlsson, L.A., Gillespie, J.W., Trethewey, B.R.: Mode II interlaminar fracture of graphite/
epoxy and graphite/PEEK. J. Reinf. Plast. Compos. 5, 170–187 (1986)
201. Kaiser, J.: Untersuchungen €uber das Auftreten von Geräuschen beim Zugversuch. Disserta-
tion, Technische Hochschule M€unchen (1950)
202. Fowler, T.J.: Acoustic emission testing of fiber reinforced plastics. In: Preprint 3092.
American Society of Civil Engineers, New York (1977)
203. Fowler, T.J.: Acoustic Emission Testing of Fiber Reinforced Plastics. Proc. Pap. J. Tech.
Counc. ASCE. 105(TC2), 281–289 (1979)
204. Rodriguez, G.: Development and implementation of a testing concept for the bearing of a
carbon fiber shaft. Thesis, Technical University Munich (2013)
205. Tonatto, M.L.P., Faria, H., Marques, A.T., Amico, S.C.: Effect of environmental condition-
ing on burst pressure of carbon/epoxy filament wound composite. In: 16th European Confer-
ence on Composite Materials, pp. 22–26. Sevilla, Spain (2014)
206. Hill, E.K., Dion, S.T., Karl, J.O., Spivey, N.S., Ii, J.L.W.: Neural network burst pressure
prediction in composite overwrapped pressure vessels. Neural Netw. 25, 187–193 (2007)
207. Walker, J.L., Workman, G.L., Russell, S.S., Hill, E.V.K.: Neural network/acoustic emission
burst pressure prediction for impact damaged composite pressure vessels. Mater. Eval. 55
(1997)
208. Gorman, M.R.: Burst prediction by acoustic emission in filament-wound pressure vessels.
J. Acoust. Emiss. 9, 131–139 (1990)
209. Hamstad, M.A., Patterson, R.G.: Considerations for acoustic emission monitoring of spher-
ical kevlar/epoxy composite pressure vessels. In: ASME Energy Technology Conference on
Composites in Pressure Vessels and Piping, Houston, TX, USA (1977)
210. Bunsell, A.R., Thionnet, A.: Life prediction for carbon fibre filament wound composite
structures. Philos. Mag. 90, 4129–4146 (2010)
211. Dong, L., Mistry, J.: Acoustic emission monitoring of composite cylinders. Compos. Struct.
40, 43–53 (1997)
212. H€ock, B., Regnet, M., Bickelmaier, S., Henne, F., Sause, M.G.R., Schmidt, T., Geiss, G.:
Innovative and efficient manufacturing technologies for highly advanced composite pressure
vessels. In: Proceedings of 13th European Conference on Spacecraft Structures, Materials þ
Environmental Testing, Braunschweig, Germany (2014)
213. Hamstad, M.A., Sause, M.G.R.: Acoustic emission signals versus propagation direction for
hybrid composite layup with large stiffness differences versus direction. In: 31st Conference
of the European Working Group on Acoustic Emission, pp. 1–8. Dresden, Germany (2014)
References 359

214. Burks, B., Hamstad, M.A: On the anisotropic attenuation behavior of the flexure mode of
carbon fiber composites. In: 19th International Conference on Composite Materials, pp. 1–9.
Montreal, Canada (2013)
215. Burks, B., Hamstad, M.A.: The impact of solid–fluid interaction on transient stress wave
propagation due to Acoustic Emissions in multi-layer plate structures. Compos. Struct. 117,
411–422 (2014)
216. Hamstad, M.A.: A waveform-based study of AE wave propagation by use of eight wide-band
sensors on a composite pressure vessel. In: 30th European Conference on Acoustic Emission,
pp. 12–15. Granada, Spain (2012)
217. Henne, F., Ehard, S., Kollmannsberger, A., Hoeck, B., Sause, M., Drechsler, K.: Thermo-
plastic in-situ fiber placement for future solid rocket motor casing manufacturing. In: SAMPE
Europe SETEC 14—Efficient Composite Solutions to Foster Economic Growth, Tampere,
Finland (2014)
218. Fowler, T.J.: The origin of CARP and the term “Felicity Effect”. In: 31st Conference of the
European Working Group on Acoustic Emission, pp. 1–8. Dresden, Germany (2014)
Chapter 5
Electromagnetic Emission

Originating from the field of geophysics science, electromagnetic emission (EME)


analysis is a nondestructive measurement technique to monitor crack formation and
propagation. Similar to AE analysis, the EME method is capable of providing real-
time information on microscopic failure mechanisms on a qualitative basis. In both
cases the reliability of quantitative information still hinges on the lack of a detailed
understanding of the correlation between the source mechanism and the measured
signal. Therefore, EME analysis is currently considered a method under develop-
ment and is mostly limited to applications in materials research and has not yet been
used for structural health monitoring applications.
In this chapter the principle of operation of electromagnetic emission is
presented first. Subsequent to that, aspects of the source mechanism, the detection
systems, and some applications of the method used as in situ technique are
introduced. Due to the novelty of the method, some applications are demonstrated
to monitor failure of reference materials such as polymers and carbon fibers to
elucidate basic relationships between failure mechanisms and the EME phenome-
non. At the end of the chapter, the established approaches are then extended to
applications monitoring failure of fiber reinforced composites.

5.1 Principle of Operation

EME signal generation is due to a temporary electrification of newly generated


crack surfaces caused by charge separation [1]. Based on physical considerations,
spatial movement of these charges is detectable as electromagnetic fluctuation in
time, which is referred to as electromagnetic emission. However, there is no
generally established theory, which adequately describes the mechanical and elec-
trical dynamics of this charge movement. A comprehensive review on this lack of
consistency of current models and their discrepancy with recent experiments is

© Springer International Publishing Switzerland 2016 361


M.G.R. Sause, In Situ Monitoring of Fiber-Reinforced Composites,
Springer Series in Materials Science 242, DOI 10.1007/978-3-319-30954-5_5
362 5 Electromagnetic Emission

given by Frid et al. [2]. Some authors attributed the dynamic to the mechanical
vibration of the crack surface [1, 3]. To overcome these discrepancies, new sug-
gestions were made to describe EME in the form of charge surface vibrational
waves [2, 4, 5]. However, there are still inconsistencies of the proposed models and
experimental results, e.g., regarding the occurrence of EME under shear crack
propagation [2, 5–8].
In comparison to AE analysis, the major advantage of EME analysis is the
negligible influence of dispersion and attenuation on the EME signals. But EME
signals are affected by the conductive properties of the cracking material [2, 4, 9–12].
This determines the time constant of charge equalization. Most investigations
of the EME phenomena were carried out in minerals (granite, marble, concrete)
[1–5, 10, 11]. In recent years various authors report on EME in reinforced polymer
materials, like carbon nanotube reinforced polyether ether ketone [13], glass fiber
reinforced epoxy resins [9], and carbon fiber reinforced epoxy resins [14, 15]. Other
authors also report on EME in highly conductive materials like metals [16–18], which
suggest that, in principle, EME is observable in all types of materials.
Another advantage of EME analysis is the possibility to fabricate sensor systems
with ideally flat response within the frequency range of the signals. Among these
are capacitive sensor systems and coil sensors operating far away from their
resonance frequencies. But currently no commercial sensor systems are available
for EME detection, although transient recorder cards and preamplifiers as being
used for AE acquisition are well suited for digitization of the detected EME signals.
Due to their different source mechanism and their weak interaction with the
nearby medium, the detected EME signals are a valuable alternative to investigate
the formation and propagation of cracks in solid materials. Some of the possibilities
of EME analysis are schematically visualized in Fig. 5.1. The occurrence of EME

EME signal

signal counng classificaon


predicon (PoD) (UoC)

amount type

of damage

material
analysis

Fig. 5.1 Analysis methods used with EME signals and EME signal prediction as forward
approach
5.2 Source Mechanism 363

signals can be used to determine the time of failure and their accumulation. In
addition their activity can be interpreted in terms of the accumulated damage and
the activity of failure in the material. As with every detection system, the successful
acquisition of an EME signal still is subject to a probability of detection (PoD)
governed by the sensitivity of the detection system and several other factors that
will be pointed out in Sect. 5.4.
For materials like fiber reinforced composites, the occurrence of different types
of failure mechanisms is also associated with different types of EME signals being
released. Given a known relationship between the source mechanism and the
signatures of the EME signals, the classification of the EME signals allows to
differentiate between different source types. This provides unique access to the
dynamics of a crack source as will be demonstrated in Sect. 5.2. Similar to AE
analysis, the classification procedure is subject to an uncertainty of classification
(UoC), which is dominated by the classification strategy and the uniqueness of
relation between EME signatures and their underlying source types.
Other than for AE, a source localization procedure based on arrival time
differences is hardly practical, since the propagation speed of the EME waves is
close to the speed of light. For the usual distances between sensor systems, the
arrival time differences are in the order of fractions of nanoseconds, which is still
below the typical temporal resolution of ultrasonic transient recorder cards. How-
ever, in combination with simultaneous AE acquisition, the time of occurrence of
EME signals can precisely determine the time of emission t0 and therefore supple-
ment the AE localization procedure [3].
Based on a suitable EME source model, it is convenient to perform a forward
prediction of the EME signal being detected at certain positions in the vicinity of
the source [12]. This signal prediction aspect is vital to establish physically based
EME source models and to allow for an improved understanding of the EME source
process and the influence of the signal detection chain.

5.2 Source Mechanism

In order to understand the EME signals of material failure, it is of vital importance to


understand their source mechanism. This involves some particular challenges regard-
ing the physical origin of the electrical charges and their relation to the formation and
propagation of cracks in materials. As a common denominator of most theories, the
type of EME source discussed in the following is restricted to the active formation
and propagation of a crack in a material. On the atomistic scale, this process of crack
formation is directly related to breakage of bonds at the position of the crack tip. This
somehow links the phenomenon of fracture and the generation of charges as release
of chemical bonds, but is not sufficient to understand the nature of electromagnetic
emission sources. Instead there are numerous published theories trying to explain the
details of the relationship between the occurrence of fracture and the finally detected
EME signal. Since there is no finally accepted theory explaining the EME
364 5 Electromagnetic Emission

phenomenon, the aim of this section is to provide an overview on the existing model
descriptions and to present some recent advances toward forward prediction of EME
signals by means of finite element modeling.
For discussion of effects involving EME, the basis for all further discussion are
Maxwell’s equations. In combination with Lorentz force law, these yield the
foundation of classical electrodynamics. As a convenient way to write Maxwell’s
equation, the differential form is used in the following:

ρ~
∇E¼ ð5:1Þ
ξ0
∇B¼ 0 ð5:2Þ

∂B
∇ E ¼  ð5:3Þ
∂t
 
∂E
∇  B ¼ μ 0 J þ ξ0 ð5:4Þ
∂t

Here the conventional abbreviations of the electric field E and magnetic field B are
applied, including the permittivity of free space ξ0, permeability of free space μ0,
the electric charge density (charge per unit volume) ρ~, and the electric current
density (current per unit area) J.
To derive the nature of electromagnetic emission sources, it is easy to infer from
the first Maxwell’s equation (Gauss’ law) that electric charges showing a temporal
or spatial oscillation in time produce a corresponding oscillating electric field.
Introducing homogeneous, isotropic, nondispersive, linear materials, the electric
field is directly linked to the displacement-field D by the electric permittivity ξ0 and
the polarization field P:

D ¼ ξ0 E þ P ð5:5Þ

Similarly, the magnetizing field H is defined by the permeability μ0 and the


magnetization field M:

1
H¼ BM ð5:6Þ
μ0

Therefore it is straightforward to conclude that oscillations of electric charge


density on the microscale are potentially a source to generate a macroscopically
detectable electric field oscillation.
However, there are two key questions which need to be addressed to establish a
meaningful description of the full EME phenomenon:
(a) What is the source of the variation of electrical charge density?
(b) What is the link between the electrical charge density variation and the
mechanical (fracture) dynamics?
5.2 Source Mechanism 365

load locally trapped electrons


lattice ion
+ + - + + --
+ + +- + - + + + +
plasticity zone - - -
+ +- + + - + + - + +
+ + + - +-- + + + +
+ + + + - + +- + - +
- -
+ + - + + -+ + + +
+ - + + +- + - + + + conduction band
- -
+ + + +- + + + +
- electron
+ -+ + +
crack
vibration due
to sudden stop Lüders band position

Fig. 5.2 Dislocation model as possible source description of EME

Despite the variety of proposed EME source models, the individual approaches
can be roughly categorized as dislocation models, discharge models, capacitor
models, and the more recently proposed electrical surface wave model.
Being one of the first proposals to describe the EME phenomenon, Misra
et al. suggested in 1978 the dislocation model [19, 20]. Their source description
is based on the elastic-plastic transition region ahead of the crack tip. As visualized
in Fig. 5.2, after a certain crack increment, the zone of plastic deformation arrests in
a mechanically stable configuration. This is associated with a sudden arrest of the
conduction electrons being trapped at the position of the dislocation band. This
sudden stop of the L€uders band dislocations is assumed to induce a type of
electromagnetic radiation similar to the Bremsstrahlung process used to generate
electromagnetic waves in the X-ray frequency range. During propagation of the
dislocations, Misra et al. further assumed a discrete redistribution of the conduction
electrons yielding a Hertzian dipole formed by the conduction electrons moving
relative to the positive ions of the crystal lattice [19]. Although parts of this EME
source mechanism are plausible, there is some experimental evidence which is in
conflict to the proposed description. One contrasting argument was provided by
Molotskii et al., who calculated the expected maximum EME frequency based on
the propagation velocity of the dislocation bands [21]. For some materials this
calculated frequency was found to be one order of magnitude less than the mea-
sured EME frequencies. Other experimental evidence is the occurrence of EME in a
broad range of solid materials ranging from rather ductile materials like metals and
polymers to comparatively brittle types like ceramics, glasses, and minerals. Espe-
cially for brittle materials, it is well known that the movement of plasticity bands is
negligible, yet the EME has even been found to be stronger in such materials than in
ductile materials [8, 22–30]. Also for all materials other than metals, there is no
such trap for the conduction electrons, rendering this attempt impossible to describe
EME for solids in general [12, 18, 26].
Another early attempt to describe the occurrence of EME was given by Finkel
et al. [31]. They demonstrated that a rapid crack in an alkali halide crystal can
produce a stochastic arrangement of positive and negative electric charges on both
sides of the crack walls causing a strong electrostatic field of the potential order of
366 5 Electromagnetic Emission

Fig. 5.3 Discharge model load


proposed as source of EME,
generating electrical
discharges due to electric discharges
stochastically arrangement
of local dipoles - - + + + +
- + +
- -
electric
field
- + +
- + -
+ + - - -
local dipole

Fig. 5.4 Capacitor model load


using opposite charges on
the crack surfaces to oscillating movement
generate EME
of crack wall
+ + + + + +
+ + +
+ +

- - -
- - -
- - - - -

107 V/cm. As a consequence of such high field strengths and the close distance of
the crack faces, Finkel et al. speculated whether this may cause electrical discharges
and these may form the origin of EME (cf. Fig. 5.3). However, the EME spectrum
of such electric discharges are well known to be of the white noise type [6] rather
than the spectra of EME pulses exhibiting some preferred frequencies as observed
in most of the materials [8, 12, 14, 23, 24, 26, 27, 32, 33]. Therefore this type of
source mechanism is not able to sufficiently describe the experimentally observed
EME signals.
A more recently proposed type of EME source is the capacitor model, which
goes back to the considerations of Finkel et al. and early experiments of
Miroshnichenko et al. [6, 31]. The actual EME source type was proposed by
O’Keefe and Thiel assuming the presence of electrical charges on the crack walls
moving apart due to a crack opening process [7]. This is schematically shown in
Fig. 5.4. As has been demonstrated by Miroshnichenko et al., the movement of such
charged capacitance plates is indeed linked to the generation of an electric field [6].
Similarly, following the consideration of Finkel et al. [31], one could assume that
5.2 Source Mechanism 367

the crack process itself is able to generate a reasonable amount of charges on the
crack walls. As a first model, the capacitor model is able to capture the temporal
decay of an EME signal due to the separation of crack walls being drawn apart. This
increases the capacitance causing an associated voltage drop being measured as loss
in electric field strength. However, there are several arguments which are not
adequately addressed by just the capacitor model. One of the key arguments against
this type of EME model is the symmetry-break argument. In order to operate in the
proposed way, there needs to be a charge imbalance during formation of the crack.
Consequently, a break of electrical symmetry needs to occur at some point. Several
ideas have been proposed in order to justify the occurrence of the symmetry break.
Some of these arguments are a possible pre-polarization of the material, applied
physical gradients due to piezoelectricity or pseudo-piezoelectricity, temperature
gradients, deformation gradients, or an impurity concentration gradient [7]. Other
authors claimed that the electrification of the crack walls could originate from
inhomogeneous elastic strains in the vicinity of the crack [34] or that crack side
electrification stems from piezoelectrification and contact (or separating) electrifi-
cation [35]. Even for the assumption that the two crack walls would be fully
charged in a stochastically pattern as discussed by Finkel et al. [31] and thus
retaining an overall charge neutrality, the induced EME of the local dipoles
would be very weak due to the cancelation of the macroscopic field. Also for
such a scenario, frequent changes of the electric field polarity would occur. Besides
the symmetry-break argument, Frid et al. pointed out the lack of applicability of the
capacitor model to the description of EME signals for shear cracks [2], since a
relative movement of the charges in parallel to the crack extension would not
sufficiently generate EME signals of similar strength as for the movement in
parallel to the applied tensile load. This was found to be in contrast to the
experimental findings in [27, 28]. Also, the absence of EME signals under simul-
taneous occurrence of AE signals and vice versa was raised as an argument against
the capacitor model [1, 22, 27].
Based on the lack of consistency of the previous EME source descriptions, Frid
et al. proposed an alternative EME source model as seen in Fig. 5.5 [2]. Given the
typical shape of an experimental EME pulse, they concluded the overall signal to
originate from three contributions:
1. Generation of an electrical charge source with strength proportional to area of
crack growth
2. Charge oscillations on both sides of the crack due to electrical surface waves
3. Decay of electrical surface waves as function of time
For the first contribution, there have been several indications being reported in
literature, which indicate the proportionality between crack length and the gener-
ated EME signal strength [12, 23, 36]. Also, the crack arrest has been clearly
correlated to the occurrence of amplitude decrease [23, 36]. Therefore it is con-
cluded that progression of the crack tip (and likewise generation of new crack
surface) is causing a generation of new charges at the crack tip since this also
involves a progressive fracture of bonds.
368 5 Electromagnetic Emission

Fig. 5.5 Electrical surface load


wave model as source
description of EME oscillating movement
of crack wall
+ + + + +
- - - +
- -
+ + + -
+ + +

- - exponential decay
- - +
- - + of charge amplitude
+ + + -
+ - -
- - -

For the second contribution, Frid et al. propose the existence of an electrical
surface wave [2]. During the rupture of atomic bonds at the crack tip, the atoms on
both sides of the bonds are moved to nonequilibrium positions relative to their
steady state. If each atom vibrates individually, the situation would resemble the
Einstein model of lattice vibrations, and the frequency of these oscillations would
approach 1015 Hz. However, in a solid the movement of one atom is coupled to
the surrounding atoms resulting in a Debye bulk model rather than an Einstein
model causing slightly lower effective frequencies, but still in similar orders of
magnitude [37]. During rupture of one atomic bond, the positive charges on these
surfaces move away together in one direction from the equilibrium position (one
crack wall), while the negative charges move in unison in the other direction from
the equilibrium position (the same crack wall). Therefore this retains overall charge
neutrality, while there is an initiation of an electrical charge wave propagating
along the crack wall. Their speed and mode of propagation is assumed to be similar
to other types of surface waves like Rayleigh waves [38] or surface vibrational
optical waves [39]; thus, their amplitude decays exponentially into the bulk
material.
The model also readily explains the decay of amplitude as typically observed in
EME signals. Given the presence of charges at the newly formed fracture surface,
there is relaxation of charges over time dominated by the dispersive propagation of
the electrical charge wave and the flow of charges from the position of the crack tip
into the bulk material, which is governed by the electrical conductivity of the
material.
In the publication by Frid et al., plenty of experimental evidence has been given
that the electrical surface wave model is a valid description of the EME source [2].
However, the origin of the oscillating crack wall movement is subject to further
discussion. In recent experiments it was demonstrated that the presence of the
oscillating part in the detected EME signal is subject to the experimental setup,
namely, the bandwidth of the recording setup [12]. Also, recent advances in AE
source modeling point out the existence of further contributions to the mechanical
5.2 Source Mechanism 369

movements occurring during crack propagation which are likely to act as source for
the oscillating part observed in EME signals.
Therefore getting back to the question “What is the link between the electrical
charge density variation and the mechanical (fracture) dynamics?” one has to
further consider the dynamics involved in the generation and propagation of cracks.
For the case of a mode I crack, Freund et al. have comprehensively reviewed the
crack propagation using a linearly elastic isotropic continuum [40–44]. The rela-
tionship between the instantaneous dynamic energy release rate per unit area of
crack plane Gd (J/m2) and the instantaneous dynamic stress intensity KId (Pa m1/2)
of the moving crack in (5.7) is given by [44] as

ð1  νÞ2 ð1  νÞ2
Gd ¼ ðK Id Þ2 ¼ gc ðK Ic Þ2 ð5:7Þ
E E

where KIc (Pa m1/2) is the equivalent static stress intensity factor of a stationary
crack having the same instantaneous length as the dynamic crack. The function gc is
the complex universal function of crack speed which may be approximated by the
simple relation in (5.8) as defined by [40, 41, 45]:
ccr cR  ccr
gc 1 ¼ ð5:8Þ
cR cR

where ccr is the crack tip speed and cR is the Rayleigh wave velocity. This classical
linear elastic fracture mechanics result points out that the ultimate limit of crack
propagation speed is the Rayleigh wave velocity. However, in reality most mate-
rials do not tend to approach the Rayleigh wave velocity. Some reasons for this
behavior were reported by Tromans [46]. Based on ab initio calculations of the
crystal lattice structure and the interatomic potentials in polycrystalline matter, he
was able to demonstrate the decrease in critical crack tip speed ccr due to the effects
of tip bifurcation, branching, and the coalescence of cracks.
Naturally, at the moment of initiation, the propagation speed starts at zero and
may approach the speed limit for crack propagation after a certain acceleration
phase. Similarly at the end of crack propagation, another change of crack propaga-
tion speed may be expected. This is either due to a lack of strain energy concen-
tration at the crack tip, causing a crack arrest or due to the changes in the boundary
conditions as induced by the presence of a free edge. Thus, the temporal stages of
crack growth may be subdivided in three phases as schematically shown in
Fig. 5.6a. First the crack starts to accelerate (phase 1) and may reach the critical
propagation velocity (phase 2) if the boundary conditions allow such. Subsequently
the crack decelerates (phase 3) to adapt to the changing boundary conditions (crack
arrest or approaching free surface). Clearly, the relative share of the three phases
may look characteristically different, possibly hiding the contributions of phase
1 and phase 3 and, hence, leaving an almost straight line. In the context of acoustic
emission source functions, such different possibilities of crack activity have been
also described by C. Scruby in [47] for the case of a linear elastic material.
370 5 Electromagnetic Emission

a b propagation below
a phase 2 a critical velocity

phase 3

propagation with
phase 1
critical velocity

acceleration deceleration t acceleration deceleration t

Fig. 5.6 Schematically representations of temporal stages of crack growth for case reaching
critical propagation velocity (a) and case staying below critical propagation velocity (b)

a b c
precrack y relaxation at velocity ccr ΔV(t) crack arrest
a
x
t<0 0 < t < a/ccr t > a/ccr

crack extension at propagation direction


velocity ccr

Fig. 5.7 Visualization of crack growth in a simplified 2D model for a linear elastic material
(based on [48])

In the same context, Wadley and Scruby proposed a simplified 2D model to


explain the dynamics of crack growth in a linear elastic material for a mode I load
situation [48]. As shown in Fig. 5.7, the parabolic precrack is subject to a tensile
load in the y-direction causing a crack extension a along the x-axis within a certain
duration a/ccr and within a certain velocity ccr. Accordingly, the crack front passes a
certain volume ΔV driving the excitation of an acoustic emission signal as
discussed in Sect. 4.2. However, this simplified model already predicts the exis-
tence of a backward motion propagating along the crack walls exhibiting Rayleigh
speed velocity cR. This additional “vibration” of the crack surface is another
important contribution to the overall crack dynamics, which likely causes another
spatial movement of the charges being generated by the progression of the crack tip.
While the simplified 2D model in Fig. 5.7 and the corresponding 1D represen-
tation of the crack tip position in Fig. 5.6 are valuable to understand the individual
contributions to the generation of the dynamics of the crack movement, it is still
challenging to transfer these observations to a real experimental situation. As
evident from fracture surface analysis, the occurrence of fast crack growth in brittle
materials typically does not come with a plane crack front. Except for some
specifically designed single crystal structures [49–52], the usual engineering
5.2 Source Mechanism 371

Fig. 5.8 Microscopy image


of a crack surface of Wallner lines
a RTM6 resin. The red
arrows indicate 2 mm
the direction of crack
propagation starting at
a distinct point on the
notch of the specimen x
crack origin crack branching notch
z

materials exhibit a distinct microstructure, grain boundaries, include defects or


impurities, and simply do not have a perfect geometry. As a consequence, the
failure of these materials is usually initiated at flaws, and the direction of crack
growth is governed by the nonsymmetric and inhomogeneous stress field caused by
the imperfect material and specimen geometry. As seen from the exemplary
microscopy image of a crack surface of a RTM6 resin in Fig. 5.8, the crack initiates
at a distinct point at the edge of the specimen. Subsequently, the crack grows in two
different stages: a rough area with almost radial marks extruding from the position
of the crack origin and a smooth area with rib marks at the upper half of the crack
surface. The radial marks are parallel to the direction of crack propagation, and the
sliver-like patterns indicate regions where crack branching takes place and there-
fore indicates an accelerating, unstable crack. The upper half of the fracture surface
is mirror smooth with straight, horizontal rib marks (Wallner lines). This indicates
smaller crack propagation velocities and a propagation direction perpendicular to
these lines, which is induced by the presence of tensile and compressive stress states
in the lower and upper part of the bending specimen cross-section shown in Fig. 5.8.
Hence, the overall assumptions of acceleration and deceleration can be related to
morphological investigations of the fracture surface, but the position of the crack is
hardly be captured by a 1D description along the x-axis. Instead, the 3D orientation
of real crack surfaces is likely to cause further differences to the simplified model
geometries discussed so far.

5.2.1 Modeling of Electromagnetic Emission Sources

In order to overcome some of the limitations of established analytical descriptions


for the investigation of arbitrary 3D geometries, the development of a source model
based on finite element modeling is a suitable approach. Recently, Gade
et al. proposed a phenomenological EME source model using a finite element
modeling approach [12]. The general idea of this modeling strategy is to generate
charges at the position of the crack tip, to allow for charge relaxation by incorpo-
rating the electrical conductivity of the material, and to couple the electromagnetic
calculation with structural mechanics to perform modeling of crack growth follow-
ing the implementation of Sect. 4.2.
372 5 Electromagnetic Emission

a b c
+F

+d(t)
+Q(t) +Q(t) +Q(r(t))

-Q(t)
-Q(t) -Q(r,(t))
-d(t)
z
y static geometry static geometry change of geometry
x -F during crack growth

Fig. 5.9 Different types of electromagnetic emission source model descriptions employing
charging of fracture surface as function of time (a), charging of fracture surface in combination
with mechanical vibration (b), and a full source model description coupling the dynamic changes
of the source geometry based on fracture mechanics with the generation of charges in space and
time (c)

In analogy to the approach to categorize acoustic emission source models in


Sect. 4.2, the different modeling strategies to describe electromagnetic emission
sources of crack propagation may be distinguished as shown in Fig. 5.9.
The first type of electromagnetic emission source models considers a static crack
geometry and explicitly defines the position and intensity of charges present on the
crack surface (cf. Fig. 5.9a). This type of model may be understood as a direct
implementation of analytical source functions, since it requires defining all dynam-
ics of the charges as function of time Q(t).
The second type of electromagnetic emission source model is the type of model
of Fig. 5.9b, which now includes a stationary crack geometry with superimposed
crack wall vibration prescribed by an analytical function d(t). The generation of
charges Q(t) is based on the progression of the crack tip and needs to be derived
from a known crack growth function. The latter could be inversely derived from
experimental EME measurements or could be computed by an AE source model
such as presented in Sect. 4.2.
Finally, the third type of electromagnetic emission source model is a model
capable to calculate the crack dynamics based on experimental loading conditions
and fracture mechanics laws (cf. Fig. 5.9c). At the same time charges are generated
on the fracture surface according to the position of the crack tip, i.e., Q(r, t).

5.2.1.1 Implementation and Validation

In order to implement a source model of the first type (Fig. 5.9a), fracture of a
single-edge-notched beam (SENB) specimen was modeled using the “AC/DC
module” and the “structural mechanics module” of the software COMSOL
5.2 Source Mechanism 373

Multiphysics. To reduce the computational intensity of the model, several simpli-


fication steps may be introduced. A transient computation of the electric field
components using the reduced version of the Maxwell’s equations (5.1) to (5.6)
above was chosen, since the magnetic field components are not detected by the
sensor systems and are thus not relevant to the interpretation of the experimental
findings.
Further considerations to simplify the modeling approach can be made regarding
the nature of charge relaxation.
As experimentally observed and described by [36], the EME signals are not
affected by charge relaxation as long as the underlying dynamics are much faster
than the relaxation process. The latter involves Ohm’s law using the material’s
electric conductivity σ~

J ¼ σ~ E ð5:9Þ

the equation of continuity

∂~
ρ
þ∇J¼0 ð5:10Þ
∂t

and Gauss’ law

∇  D ¼ ρ~ ð5:11Þ

The combination of (5.9), (5.10), and (5.11) results in the differential equation for
electric charge density in a homogeneous medium:

∂~
ρ σ~
þ ρ~ ¼ 0 ð5:12Þ
∂t ξr

This equation has a solution of the form

ρ~ðtÞ ¼ ρ~0 et=~τ ð5:13Þ

introducing the charge relaxation time ~τ ¼ ξr =~σ . For highly conductive mate-
rials, ~τ may approach 1019 s, whereas poor conductors reach ~τ -values in the
order of 102 s and approach infinity for ideal insulators. For modeling of charge
relaxation, it is thus important to consider the observation scale of the compu-
tation relative to the expected ~τ -values.
In [12] an implementation assuming ~τ ffi t is presented. In this case, the
computation requires to solve (5.10), which may be written in terms of the material
properties ξr and σ~ as


∇  ðξ ∇U þ PÞ  ∇  ðσ~ ∇U  JÞ ¼ 0 ð5:14Þ
∂t r
374 5 Electromagnetic Emission

As long as the induced electric fields are free of curls, this is sufficient for a near-
field approximation, i.e., the expected wavelengths need to be much larger than the
geometrical dimensions considered. As will be discussed in Sect. 5.3, the wave-
lengths of the electric fields in the frequency range of interest are indeed much
larger than the typical dimensions of the experimental setup. Hence, the computa-
tion is carried out for the reactive near field around the source and this assumption is
applicable. Moreover, the skin effect and wave propagation effects need to be
negligible as well. This approach was found to be suitable to describe the dynamics
intrinsic to the EME source as presented in [12].
However, for materials such as polymers, carbon fibers, or carbon fiber
reinforced polymers, it may readily be estimated from the conductivity values
and the materials permittivity given in Table B.3 of Appendix B that ~τ t
(cf. Sect. 5.2.1.4). For the model implementation, this observation allows to reduce
the computational intensity further by switching to an electrostatic formulation of
the problem, yielding a reduced version of (5.14):

∇  ðξr ∇U  PÞ ¼ ρ~ ð5:15Þ

In this context, the charge distribution can be considered as static input to (5.15)
resulting in a corresponding electric potential U. However, this does not necessarily
imply that the electric charge density is constant. Instead, the use of Gauss’ law in
this reduced variant does allow definition of an electric charge density as function
of time ρ~ðtÞ as used for the EME source model outlined in Fig. 5.9 and reduces the
computational intensity of the coupled approaches.
Figure 5.10 shows the model geometry, consisting of the SENB specimen made
from RTM6 epoxy resin, the platelike EME sensors, and the surrounding air
volume. To avoid interaction with nearby materials and to eliminate influences of
the size of surrounding conductive media, a spherical infinite element domain of
5 mm thickness was used to stretch the electric field to infinity. Such an infinite
element domain applies a rational coordinate scaling to a layer of domains

5mm
EME sensor

fracture surface
2mm
10mm

190mm

grounded
plate specimen
z
y
x

Fig. 5.10 Overall dimensions of the model configuration (left) including details of the modeled
specimen and EME detector system (right)
5.2 Source Mechanism 375

surrounding the region of interest. When the dependent variables vary slowly with
radial distance from the center of the region of interest, the finite elements coordi-
nate can be stretched in the radial direction such that boundary conditions on the
outside of the infinite element layer effectively operate at a very large distance from
the region of interest. The coordinate scaling along the radial direction was
implemented as nearly singular 1/r stretching.
For the specimen and the EME sensors, the corresponding material properties of
RTM6 epoxy resin and copper as reported in Tables B.1 and B.3 in Appendix B
were used. Capacitance plate sensors of 10 mm  10 mm width and height and
12 mm distance to the fracture plane were modeled and implemented as will be
described in detail in Sect. 5.4. As has been previously shown in [12], the explicit
modeling of the EME sensors and incorporation of the system transfer function
allow to compute EME signals of very close similarity to the experimentally
measured signals. For the present configuration, the attached circuitry was modeled
as combination of a 10 MΩ resistor and a 70 pF capacitor to account for the first
stage of the preamplifier circuit used in the experiments (cf. Sect. 5.4). This is
sufficient to emulate the high-pass behavior of the EME sensor system and enables
a direct comparison of the modeled signals with experimental signals.
For an EME source model of the first type, as source function for the EME, a
time-dependent surface charge Q(t) is used. Within the model this was implemented
as surface charge density ρ~ðtÞ located at the hypothetical fracture surface position
(cf. Fig. 5.10). Based on the EME source operation hypothesis formulated above, it
is assumed that one contribution of Q(t) is due to charge separation following the
path of the crack tip causing an increasing accumulation of charges at the fracture
surface during crack growth as seen in Fig. 5.11a. The second contribution is
attributed to the movement of charges owing to mechanical movement (such as
vibration of the crack surface) and is modeled as rise and decay of a sinusoidal
oscillation with a frequency of 80 kHz (cf. Fig. 5.11b). The superposition of both
contributions defines the assumed analytical change of charges Q(t) existing on the
fracture surface as function of time.
Electrically insulated boundaries are simulated by the zero-charge condition
n  D ¼ 0, while at grounded boundaries the electric potential is set equal to zero.
As initial conditions for the model, a zero potential is chosen everywhere.
To achieve convergent numerical solutions, quadratic Lagrange elements were
used for the spatial discretization, and a maximum mesh element scale of 0.5 mm
was selected for the RTM6 specimen. For the two small copper plates of the EME
sensors, a resolution with maximum mesh element size of 1 mm proved sufficient.
The air domain was meshed with a maximum element size of 11 mm and with a
maximum element growth rate of 1.4 per element. This ensures an adequate
resolution in the area between specimen and EME sensor and reduces the compu-
tationally required degrees of freedom. The transient calculation is done via a
generalized-α algorithm with a time step size of 0.5 μs.
The computation results are shown in Fig. 5.12a as voltage signal detected at
the EME sensor compared to the surface charge density on the fracture surface. The
major differences between the two curves are due to the high-pass behavior of the
376 5 Electromagnetic Emission

a
0.20
surface charge density [mC/m2]]

0.16

0.12

0.08

0.04

0.00
0 50 100 150 200 250
time [μs]

b
0.004
surface charge density [mC/m2]

0.002

0.000

–0.002

–0.004

0 50 100 150 200 250


time [μs]

Fig. 5.11 Implementation of analytical source function in EME source model comprising (a)
contribution of charge accumulation due to crack growth and (b) contribution due to crack surface
oscillation

modeled EME sensors causing a noticeable decay of signal voltage compared to the
buildup of charge density. However, the step-function type of charge generation is
well preserved in the EME sensor signal and constitutes the dominating part of the
detected signal. A comparison of both signals after application of a 50 kHz
5.2 Source Mechanism 377

a
12

0.20
10

surface charge density [mC/m2]


0.16
8
voltage [V]

0.12
6 sensor signal
source function
0.08
4

2 0.04

0 0.00
0 50 100 150 200 250
time [μs]

b filtered by 50 kHz Butterworth high-pass 10th order


0.0035
0.05
0.0030

surface charge density [mC/m2]


0.0025
0.00
0.0020
voltage [V]

–0.05 sensor signal 0.0015


source function
0.0010

–0.10 0.0005

0.0000
–0.15
–0.0005

–0.0010
–0.20
0 50 100 150 200 250
time [μs]

Fig. 5.12 Resulting signal voltage at EME sensor compared to generation of surface charges as
function of time (a) and same comparison for filtered sensor signal compared to oscillating part of
source function (b)

Butterworth high-pass filter of tenth order reveals close similarity between the
fluctuation of charges assumed due to crack surface oscillation and the detected
signal at the sensor position (cf. Fig. 5.12b). The overall shape of the vibrating
part of the surface charges is well captured in the detected EME signal, and the
minor differences as well as the observed phase shift between both signals are
solely caused by the high-pass filter. Thus, as previously demonstrated in [12],
378 5 Electromagnetic Emission

such generation of surface charges and variation of their strength in time will cause
an EME signal of close similarity to those seen in the experiment.
As extension of the previous EME source model description, the origin of the
oscillating part of the source function is now modified. Thus, the EME source
model of second type (cf. Fig. 5.9b) is implemented using a generation of charges as
function of time as caused by the progression of the crack tip. The increase of
surface charges due to crack growth is chosen identical to the previous model as
seen in Fig. 5.11a. As second contribution, the crack walls are now forced to vibrate
in an oscillatory motion using a prescribed displacement condition dx(t) applied to
the hypothetical fracture surface as given in Fig. 5.11b with a maximum amplitude
of 0.5 μm. All other computation settings were kept identical to the previous model.
As seen from the resulting sensor voltage in Fig. 5.13a, the overall rise and decay as
expected due to the growth of charges and the high-pass character of the sensing
system is found in close analogy to the findings of the previous model. However,
the superimposed oscillations seen in the sensor signal do not match the expected
shape of the source function of Fig. 5.11b. Instead, close resemblance is observed
for the source displacement component evaluated along the y-direction. This is
shown in the filtered sensor signal of Fig. 5.13b. In addition, the strength of the
EME signal rises until 150 μs, which is owed to the increased number of charges
applied on the fracture plane. This oscillating contribution is already easy to see in
the sensor signal of Fig. 5.13a giving rise to the question of correct choice of the
displacement magnitude.
Consequently, the EME source model of Fig. 5.9c combines the previously
presented acoustic emission source model and the generation of charges as function
of time. To this end, the same geometry as used in the two previous models is
chosen. The specimen is loaded up to 23 N static force in three-point bending until
the fracture condition is fulfilled (cf. implementation in Sect. 4.2). In a subsequent
modeling step, charges are applied to the fracture surface according to the evalu-
ation of the crack tip position, i.e., following the degradation function. All dynamic
displacements of the fracture plane are directly computed and are not prescribed by
an analytical function. The duration of crack growth in this case was found to be
approximately 150 μs. The evaluated surface charge density follows the previous
choice and was selected to provide the same amount of charges at the end of
fracture to comply with the previous cases.
Therefore, the sensor signal of Fig. 5.14a yields close resemblance to the two
previous cases. Systematic differences are observed for the oscillating part of the
signal as seen in the filtered version of Fig. 5.14b. For the EME signal, a first
oscillation occurs at the first instance of crack growth followed by a damped
oscillation until the crack reaches the upper part of the specimen. At the moment
of rupture, a second oscillation is observed in the EME signal. This part of the
EME signal is described best by the mean source displacement along the x- and y-
axis. The first vibration of the formed fracture surface is in good correspondence
with the respective signature in the EME signal. However, the oscillation was
found to be damped much faster. The most likely reason for this discrepancy is
the contributions of other displacement components, i.e., along the z-direction.
5.2 Source Mechanism 379

a
12
0.20
10

surface charge density [mC/m2]


0.16
8
voltage [V]

0.12
6 sensor signal
charge growth
0.08
4

2 0.04

0 0.00
0 50 100 150 200 250
time [μs]
b
filtered by 50 kHz Butterworth high-pass 10th order
30
0.5
25

0.0 20

displacement [nm]
15
voltage [V]

–0.5
sensor signal 10
source displacement
–1.0 5

0
–1.5
–5

–2.0 –10
0 50 100 150 200 250
time [μs]

Fig. 5.13 Resulting signal voltage at EME sensors compared to generation of surface charges as
function of time (a) and same for filtered sensor signal compared to source oscillation (b)

Although similar signatures were observed in these displacement signals, it was


not suitable to combine them into one single “surface vibration” being directly
comparable to the EME signal. At the moment of rupture of the specimen,
another high displacement oscillation occurs along the y-direction. Due to the
averaging process, the propagation time of the acoustic wave is slightly delayed
when compared to the EME signal.
380 5 Electromagnetic Emission

a
12
0.20
10

surface charge density [mC/m2]


0.16
8
sensor signal
voltage [V]

charge growth 0.12


6

0.08
4

2 0.04

0 0.00
0 50 100 150 200 250
time [μs]
b
filtered by 50 kHz Butterworth high-pass 10th order
0.02 1000

800

0.00 600

400 displacement [μm]


voltage [V]

200
–0.02
0

–200
sensor signal
–0.04 source displacement
–400

0 50 100 150 200 250


time [μs]

Fig. 5.14 Resulting signal voltage at EME sensors compared to generation of surface charges as
function of time (a) and same for filtered sensor signal compared to source oscillation (b)

Overall, the proposed EME source models allow to compare and discuss the
origin of the individual contributions of the EME signal. In combination with
validated sensor modeling (see Sect. 5.3), this provides a solid basis to investigate
some more details of EME source operation.
5.2 Source Mechanism 381

5.2.1.2 Source Radiation Pattern

As first aspect to study, the effect of source orientation relative to the sensor shall be
investigated. Based on the distinct orientation of the fracture surface within the
specimen, it may be expected that the emitted electromagnetic waves exhibit some
sort of radiation pattern. If the EME source has a preferential orientation relative to
the crack surface, a noticeable change in the intensity should be detectable as
function of detection angle. In order to investigate this effect, the single-edge-
notched beam (SENB) configuration as shown in Fig. 5.15 with a specimen size of
25 mm  5 mm  5 mm (length  height  width) is used. The notch depth was
prepared with 2 mm. All tests were carried out using a three-point bending fixture
made from PMMA and supports made from PVC in combination with copper
capacitance plates of 6 mm  8 mm (height  width) as EME sensors. Acquisition
settings were chosen in analogy to acoustic emission measurements and are sum-
marized in Table 5.1.
Keeping the sensor orientation constant, the bending fixture was rotated by an
angle ϕ around the z-direction. This allows a systematic variation of the angle
between the crack surface normal and the sensor plate normal between 0 (parallel)
and 90 (perpendicular) in steps of 10 . The distance d of the sensor to the rotation
axis (see Fig. 5.15) was kept constant at 14 mm. From the detected EME signals, the
absolute energies were quantified using the definition of Appendix C. In addition, a
tenth order Butterworth band-pass filter ranging from 20 to 100 kHz was applied to

Force
fracture surface z

d
z
6mm

20 mm notch

AE-Sensor

16.5 mm
15 mm

EME sensor

support ø = 3.1 mm

Fig. 5.15 Experimental setup for flexural testing of SENB specimens with EME acquisition (left)
and definition of rotation angle and source-sensor distance (right)
382 5 Electromagnetic Emission

Table 5.1 EME acquisition Acquisition settings Value


settings used for SENB tests
Preamplification 40 dB
Threshold 35 dB
Triggering Independent
Acquisition rate 10 MS/s
Bandpass range 1 kHz to 3 MHz
Mounting system Floating arm
Sensor type Capacitance plate
Number of sensors 1 pair

a 0
1500

1200
abs. energy [aJ]

900

600

300

0 –90 90

b 100 0

80
abs. energy [aJ]

60

40

20

0 –90 90

Fig. 5.16 EME signal energies as function of detection angle for full signal (a) and filtered
signal (b)

the signals to independently investigate the oscillating part of the signals. Subse-
quently, the absolute energies of the filtered signals were quantified as well.
As shown in Fig. 5.16 for the filtered and unfiltered signal, a substantial
difference of the signal energy was measured as function of the detection angle.
While the strongest signals are detected for a parallel orientation of crack surface
and sensor plate, the measured signals decrease significantly in energy with
increasing angle. This indicates a directional field distribution. The angular direc-
tivity differs for the different parts of the signals. Figure 5.16a, b shows the energies
for the detected signals quantified for the base part and for the oscillating part. The
total signal energy is dominated by the energy of the base part. As discussed in the
5.2 Source Mechanism 383

previous sections, the base signal is attributed to the separation of charges during
crack growth. This part exhibits some kind of dipole characteristic, since two crack
surfaces with opposite charges form a dipole moment with a direction parallel to the
crack surface normal. The detected energies show a stronger angular dependence
than the energies of the oscillating part. Only for angles up to 40 energies
significantly exceeding the level of the noise were detected with a maximum at
0 . At ϕ ¼ 10 the energy of the signals has dropped to 53 % when compared to the
energy at ϕ ¼ 0 . The oscillating part of the signals is generated by the vibration
of the crack surface as has been demonstrated by the source models discussed
previously. When this vibration is assumed to be perpendicular to the crack surface,
i.e., has a strong directional orientation, a similar angular dependence of the signal
intensity is expected. The measured behavior seems to be more complex than one
would expect for a simple dipole characteristic. For a point dipole, the potential
scales with cos(ϕ), so one could expect the energy scaling with cos 2 ðϕÞ
(as indicated by the dashed line in Fig. 5.16b). The detected signal energies only
partially confirm this simple dipole characteristic. However, since the sensor is in
the reactive near-field zone of the source and may not be approximated by a point-
like position, it is likely to always detect some field components of other angles.
This could account for the stronger detection of energies for small angles than one
would expect for a simple dipole characteristic. Additionally the spatial character-
istics of an electrical field depend on the type of source and on the surrounding
matter. For example, the presence of conductors near the EME source affects the
formation of the electric field as shown in Fig. 5.28 in Sect. 5.3.

5.2.1.3 Influence of Distance Between Source and Sensor

Similar to AE signals, it may be expected that due to geometric spreading, signal


intensity will decay with increasing distance to the sensing system. To evaluate the
detectable signal energy as function of distance, the EME signal energy was
measured in the SENB configuration shown in Fig. 5.15 as function of source-
sensor distance along the detection angles 0 , 45 , and 90 . All acquisition settings
were chosen following Table 5.1. For the different angles, the distance of the EME
sensor to the source was varied ranging from the closest distance possible up to the
distance where the signal-to-noise ratio inhibits detection of the signal. The
recorded signals and the filtered signals (band-pass setting as in Sect. 5.2.1.2)
were then analyzed in terms of the detected energy. In Fig. 5.17a, the results for
the calculated energies are displayed for the full signals and for the oscillating part
of the signals in Fig. 5.17b, respectively.
As expected, the measured energy decreases with increasing sensor distance.
This was observed for all three angles. As discussed in the previous subsection, the
different parts of the signals show a different angular dependence. The dependence
on the distance appears to be almost the same for the base signal and the oscillating
part. For both parts the signals are only detectable within a distance of a few
millimeters to the source. The characteristics of the plotted data show no intuitive
384 5 Electromagnetic Emission

a
1800

1600 45°
90°
1400
absolute energy [aJ]

1200

1000

800

600

400

200

0
0 2 4 6 8 10 12 14 16 18 20 22 24 26
distance [mm]
b
120

45°
100 90°
absolute energy [aJ]

80

60

40

20

0
0 2 4 6 8 10 12 14 16 18 20 22 24 26
distance [mm]

Fig. 5.17 EME signal energies as function of source-sensor distance for full signal (a) and filtered
signal (b) (based on [12])

dependence of distance d. This may be due to multiple influences that all depend on
the position of the sensor. The main effect is assumed to arise from the spatial
characteristic of the electric field. The potential generated by a dipole decreases
with a 1=d2 dependence. This would result in a decrease with 1/d4 for the measured
energies. However, such dependence as function of distance was not observed.
Since the real charge distribution is unclear, multipole moments of different order
may also appear during the fracture process exhibiting different kinds of distance
dependencies. Furthermore, with an increase of d, the distance between the
5.2 Source Mechanism 385

capacitor plates also increases and thus the capacitance decreases. For constant field
strength, the smaller the capacitance gets, the higher the voltage between the plates
would be.
Another effect on the signal intensity, which might even be of larger relevance
than the increase of capacitance with distance, is the influence of other conducting
parts of the experimental setup. Although the fixture was built from nonconducting
materials, some other elements inevitably consist of conducting materials. The most
important one is an acoustic emission sensor which is positioned 25 mm above the
specimen and was used for simultaneous detection of acoustic emission signals.
Since all conductors near the source influence the voltage on the sensor plate, this
influence becomes more important at larger sensor distances.

5.2.1.4 Influence of Electrical Properties

As next fundamental investigation, the influence of the electrical properties of


the cracking medium is investigated. Based on the modeling approach considered
in [12], the properties of the modeled materials are now systematically varied. As
first parameter, the surface charge density is investigated. Figure 5.18a shows
computation results of the reference case of an RTM-6 specimen using the source
functions plotted in Fig. 5.11.
Compared to the reference case with a final surface charge density of 0.02 C/m2,
the two other results were computed for a surface charge density reduced by a factor
of two and four, respectively. As seen from Fig. 5.18a, the resulting EME signal
amplitude decreases directly proportional to the amount of charges used as source.
Consequently, the level of charge imbalance will contribute significantly to the
detection of EME signals.
As second numerical investigation, the material of the bending geometry of
Fig. 5.10 was changed to properties of a typical carbon fiber reinforced epoxy taken
from [53] and to those of steel as noted in Table B.3 of Appendix B. As seen from
the comparison of EME signals with equal surface charge density, there is only little
influence of the change in electrical conductivity and electric permittivity. For the
computation results of RTM6 and CFRP, almost no difference is observed (black
and red lines superimpose in Fig. 5.18b). Only the substantial change in conduc-
tivity of steel causes a noticeable change in the EME signal as found in the inset of
Fig. 5.18b. This allows to conclude for the cases of materials studied herein that the
influence of charge relaxation is expected to be negligible within the typical
durations of a few 100 μs.

5.2.1.5 Sources in Fiber Reinforced Composites

For application of EME measurements to fiber reinforced composites, several other


factors regarding the EME source need to be considered. For these materials, the
formation of a crack is due to fracture of one or more of the constituent materials or
386 5 Electromagnetic Emission

a RTM-6 (Reference)
0.5 x surface charges
10 0.25 x surface charges

8
voltage [V]

0
0 100 200 300 400
time [μs]

b
RTM-6
10 CFRP
Steel

8
9.70
voltage [V]

9.68
6 9.66
9.64
voltage [V]

9.62
4 9.60
9.58
9.56
9.54
2 9.52
9.50
160 165 170 175 180
time [μs]
0
0 100 200 300 400
time [μs]

Fig. 5.18 Variation of charge density applied to the fracture surface (a) and variation of materials (b)

due to crack formation at their interfaces. In this case, the difference in electrical
properties of the fibers and the matrix will contribute to the generation of charges
along the fracture plane, and the anisotropic electrical conductivity of the compos-
ite will govern charge relaxation effects. Based on previous studies [12] and other
findings in literature [9, 13–15], it may readily be expected that failure mechanisms
5.2 Source Mechanism 387

in composites cause measurable electromagnetic emission (see also Sect. 5.5). This
underlines that charge relaxation effects in fiber reinforced composites will not
completely suppress the EME signatures arising from crack growth and the oscil-
lation of the fracture surface as described in Sect. 5.2.1.1.
Following the validated modeling approach, three basic types of failure modes
are investigated in order to point out some basic possibilities of EME measurements
in composite materials. To this end, inter-fiber failure, inter-ply delamination, and
fiber breakage are studied. As model configuration, a typical tensile test specimen
with 195 mm  20 mm  1.9 mm (length  width  thickness) according to DIN-
EN-ISO 527-4 is investigated. The electrical material properties used for the
electrostatic simulations within the “AC/DC module” of COMSOL Multiphysics
are those of Sigrafil CE1250-230-39 as given in Table B.3 of Appendix B. No
explicit modeling of the stacking sequence was carried out in the following, since
the electric conductivity tensor is not required in electrostatic modeling. Instead,
only the orientation of the fracture surface was varied to account for directivity
effects. As for the configuration of Fig. 5.10, the tensile specimen was located
within a sphere of 300 mm diameter and adjacent infinite element domain to avoid
interaction with any nearby boundaries. In all cases investigated, the charge density
on the fracture surface was chosen to be 0.02 C/m2. Since this value is arbitrarily
chosen, it is important to note that the appearance of the field lines and the
comparability of the iso-voltage surfaces throughout Figs. 5.19, 5.20, and 5.21
are not affected by this choice. By keeping this value constant throughout the

a b c

1.9 mm

z
y
x electric field lines
iso-voltage surface

195 mm

20 mm

y
x

Fig. 5.19 Electric field lines surrounding charged fracture surface and iso-voltage surfaces as 3D
view and projection to xy-plane for inter-fiber failure at 0 angle (a), 45 angle (b), and 90 angle (c)
388 5 Electromagnetic Emission

y iso-voltage surface

195 mm
1.9 mm

20 mm

z
y
x electric field lines

Fig. 5.20 Electric field lines surrounding charged fracture surface and iso-voltage surfaces as 3D
view and projection to xy-plane for inter-ply delamination

y
z x
y 195 mm
x

20 mm
electric field lines

iso-voltage surface

1.9 mm

Fig. 5.21 Electric field lines surrounding charged fracture surface and iso-voltage surfaces as 3D
view and projection to xy-plane for fiber breakage including details close to fracture surface

simulations, it is instead assumed that a similar amount of charges per crack area is
generated regardless of the fracture type. This might only be partially justified for
the fiber breakage case as compared to the other two cases. All mesh settings
confirm with the choices reported in Sect. 5.2.1.1.
5.2 Source Mechanism 389

As a first example, a study of the electric field lines due to inter-fiber failure is
presented. The size of the fracture surface was implemented as 2.5 mm  0.5 mm
(width  height). The result of the computation is shown in Fig. 5.19a–c for three
different orientations of fracture surfaces within the xy-plane. In addition, the
iso-voltage surfaces surrounding the modeled fracture surface are shown in red
(positive) and green (negative) color. The iso-voltage surfaces may readily be
understood as radiation pattern as discussed in Sect. 5.2.1.2. As clearly seen from
the orientation of the iso-voltage surfaces as well as the formation of electric field
lines, the orientation of the inter-fiber failure will directly translate into the radia-
tion pattern of the respective EME source. Given the findings of Sect. 5.2.1.2, this is
not surprising, but points out the possibility to measure the orientation of fracture
surfaces by EME as discussed in Sect. 5.5.2.
Choosing another orientation of the fracture surface as typical for inter-ply
delamination results in the electric field lines and iso-voltage surface shown in
Fig. 5.20. Here the size of the delamination fracture surface was implemented as
5 mm  5 mm (length  width). The orientation of the delamination was chosen to
be within the xy-plane at the center of the laminate with respect to the z-axis. Due to
the change in orientation of the charged fracture surface, the radiation pattern is
now found to be dominant in the out-of-plane direction of the laminate. Despite of
the small increase in the fracture surface area, the iso-voltage planes are of almost
similar size as for the inter-fiber failure cases.
For the configuration of a charged fracture surface as typical for fiber breakage,
some obvious differences arise as seen in Fig. 5.21. The size of the charged fracture
surface was chosen to be 240 μm  240 μm (width  height) being oriented within
the yz-plane. This corresponds to a typical fracture surface of more than 1000
filaments. Overall, this is much smaller than the fracture surfaces investigated
before. Given the assumption of constant charges per surface area, this translates
into a much smaller signature, having iso-voltage planes of the same level being
much closer to the source position. Nevertheless, the overall radiation pattern seen
from the magnification of details in Fig. 5.21 also resembles a dipole characteristic.
Due to the difference in rise-times of the sources and the resulting oscillations as
presented in Sect. 4.2, it may readily be expected that the signature of the
corresponding EME signals will be characteristically different to those of inter-
fiber failure of the same orientation.

5.2.2 Test Sources for Electromagnetic Emission

In order to validate measurement setups and to qualify sensor systems, it is quite


useful to have an artificial type of EME source. However, there is no national or
international standard covering this topic, and there are not even fully established
routines found in literature. Consequently, various designs have been proposed,
which exhibit different strengths and weaknesses. In this section some simple test
source concepts are presented, which are easy to fabricate and easy to use for
390 5 Electromagnetic Emission

generation of artificial electromagnetic emission signals. The natural choice to emit


an electromagnetic wave is a metallic conductor with a driving voltage applied.
This combination is further referred to as emitter1 for simplicity. To act as a test
source for electromagnetic emission, three criteria may be raised to distinguish
different artificial source types:
1. The emitter should operate sufficiently far outside its electric resonance fre-
quency to ensure a reasonably flat frequency characteristic.
2. The dimensions of the emitter should be chosen such that it can be located at the
same position as the real electromagnetic emission source to be detected.
3. The emitter should exhibit a source radiation characteristic similar to a real EME
source.
In general, all sorts of metallic objects are used as emitter, whereas different
electric systems and excitation functions are used as signal drivers. In many cases,
simple plates or open-ended wires were found to be most useful as emitter
component.
In order to fulfill the first criterion, the emitter and its circuit need to be selected
appropriately. Theoretically it is possible to estimate the resonance frequency of
any electrical system based on the geometry and the electrical properties of the
antenna material and concise knowledge of the emitter circuit. While this is
straightforward to derive for some emitter configurations, in reality the near-field
aspect [54–56] may cause additional resonances, which are difficult to predict by
the analytical calculations. Thus it is advisable to directly measure the resonance
frequency of a test source emitter based on the electric impedance of the system.
Such measurement is readily be achieved by use of a network analyzer or by means
of monitoring the detected signal at a sensor system, given the latter is reasonably
flat with frequency.
In order to comply with the second criterion, the dimensions of the emitter
system should be reasonably small. Since this is a general matter of the experimen-
tal setup, a typical emitter size should not be much larger than a few millimeters to
be able to resemble a point-like EME source originating from material failure. If the
EME source to be investigated cannot be approximated as point source, this
criterion becomes almost obsolete. In cases of an EME source size ranging from
some millimeters to few centimeters, the emitter is certainly more appropriate when
it is of similar dimensions.
Compared to the other two criteria, the third one is the most challenging one.
Based on the discussion of source radiation characteristics of EME sources in
Sect. 5.2.1.2, it is obvious that this effect needs to be incorporated to form a
representative artificial EME source. As schematically shown in Fig. 5.22, it is
possible to fabricate such EME test sources by using two aligned metallic parts.

1
The term emitter should not be misunderstood as system operating with far-field characteristics,
since the EME configuration is usually in the reactive near-field (cf. Sect. 5.3).
5.2 Source Mechanism 391

a b
z z
y sensor y sensor
x x
center
conductor

shielded cable
shielded cable
U(t) U(t)

foilshield –U(t)
and braided
shield

Fig. 5.22 Emitter configuration used for generation of artificial EME using single shielded cable
(a) and similar configuration using two shielded cables yielding artificial EME source with
characteristic source radiation pattern (b)

For this purpose, cables as typically used for Bayonet Neill-Concelman (BNC)
cables may be used. At the end of the cable, parts of the shielding are removed to
reveal the center conductor to a length of few millimeters. It is important to retain
the shielding up to the position, where the center conductor is revealed to open
space to avoid symmetry breaks between the two center conductors. Using only one
such cable, an easy EME test source is fabricated, which can be driven by a signal
voltage U(t). Keeping a constant distance to the capacitance sensor position (a plate
in Fig. 5.22a, b), this allows to test the acquisition settings and allows to evaluate
the system transfer function (cf. Sect. 5.4). To fabricate an EME source with a
characteristic source radiation, two shielded cables may be placed next to each
other. Applying a signal voltage U(t) of opposite polarity and identical phase to the
two cables as seen in Fig. 5.22b, this yields an EME source with characteristic
radiation pattern.
Since the relative movement between source and detector will also cause
noticeable EME signals, it is recommended to generally use nonconductive holders
or supporting arms to position the test source at the designated spot and to keep it at
this constant position during the measurements.
For all geometrical configurations, the question arises, which type of source
function U(t) to use for generating artificial EME. This certainly depends on the
intended purpose. For calibration of sensor systems as carried out in Sect. 5.4,
Gaussian enveloped sinusoidal signals with varying center frequency may be used
to sweep a certain frequency bandwidth. For generation of artificial EME signals,
i.e., signals representative for a real EME source, different source functions are
necessary. As demonstrated in Sect. 5.2.1, the most likely type of source function
seen for an EME source are step-function types with superimposed oscillations.
Such source signals are straightforward to implement in arbitrary waveform
392 5 Electromagnetic Emission

generators and may then be applied as driving voltage to the emitter configurations
shown in Fig. 5.22. As one of the intents of the test source, this allows checking the
detector systems prior to the measurement of the test specimen. And it also allows
optimization of the sensor positions, given the chosen test source exhibits the same
type of source radiation characteristic as the real EME source.
Using the configuration of Fig. 5.22a and a negligible distance between emitter
and detector plate, the EME signal as seen in Fig. 5.23a is obtained. The input
function to the arbitrary waveform generator consists of a step-function signal with
a superimposed 80 kHz oscillation of small amplitude. The latter is only visible
after high-pass filtering by a Butterworth filter of tenth order as seen in Fig. 5.23b.
The experimental EME sensor consists of a small rectangular copper plate of
5 mm  8 mm  1 mm (height  width  thickness). Signals are detected by a
commercial acoustic emission system using a preamplification factor of 37 dB.
Comparing the measured EME signal before (Fig. 5.23a) and after filtering
(Fig. 5.23b) readily reveals both contributions of the input signal. However, the
overall shape of the detected EME signal and phase shifts occur due to the nonlinear
transfer function and the band-pass characteristic of the measurement system
(cf. Sect. 5.4).
Using a test source configuration as schematically show in Fig. 5.24a and the
same procedure as for the single BNC cable as described above, it is possible to
evaluate the detected signal intensities as function of detection angle. Experimen-
tally, two capacitance plate sensors were mounted in orthogonal arrangement, i.e.,
one sensor plate normal in parallel to the y-axis and one sensor plate normal in
parallel to the z-axis. Following the sensor modeling approach of Sect. 5.4, this
configuration was implemented within the “AC/DC module” of COMSOL
Multiphysics and modeled using the geometry of the experimental configuration
accordingly. The distance between the center of the artificial source and the sensor
plate was 22 mm, and as test signal a sinusoidal voltage of 1 kHz frequency was
chosen. Measurements and simulations were performed for angle increments of 10
ranging from 0 to 90 rotation of the test source. A comparison of the obtained
experimental and numerical peak-to-peak voltages is given in Fig. 5.24b. Within
the margin of error, the artificial test source shows the expected cosine shape
(dashed line). Also, the detected signal voltages at sensor 1 and sensor 2 exhibit
the expected phase shift of 90 due to their orthonormal orientation relative to the
test source.
One of the drawbacks of arbitrary waveform generators is their low operational
voltage (typically 20 V or less). In typical experimental configurations involving a
distance between test source and detector of a few centimeters, this might already
be at the limit of detectability. Hence, these comparatively low voltages do not
sufficiently resemble strong EME sources as sometimes seen in the experiment.
However, it is not advisable to change the distance between detector and source to
apparently increase the test source strength, since this will have an impact on the
relative position of EME sensors and their position relative to other metallic objects
close by. Thus such modifications might falsify the measurements of the test
sources. To overcome this problem, either more powerful arbitrary waveform
5.2 Source Mechanism 393

10

8
voltage [V]

6
input signal
measured signal
4

0
0 50 100 150 200 250
time [μs]

b
0.06 input signal
measured signal
0.04

0.02
voltage [V]

0.00

–0.02

–0.04
filtered by 100 kHz Butterworth high-pass 10th order
–0.06

0 50 100 150 200 250


time [μs]

Fig. 5.23 Comparison of test source signals detected by capacitance sensor in negligible distance
to test source using BNC wire configuration (a) and same signals after application of 10 kHz high-
pass filter (b)

generators may be applied to generate stronger EME signals or an alternative test


source may be applied. Experiments using a configuration as shown in Fig. 5.25a
involving a piezoelectric lighter as voltage source have demonstrated the possibility
to yield highly reproducible and strong EME signals.
394 5 Electromagnetic Emission

a b sensor 1
sensor 1 (model)
80 sensor 2
sensor 2 (model)
sensor 1 70 cos(angle)

60

voltagePP [mV]
50

40

30

20
sensor 2
z 10
y 0
0 10 20 30 40 50 60 70 80 90
x angle [°]

Fig. 5.24 Artificial test source arrangement (a) and evaluation of peak-to-peak voltages for
experiment and model in comparison to cosine function (b)

a b
sensor
z z
y
x x

inner spring
piezoelectric
disc impactor
piezoelectric
lighter anvil

Fig. 5.25 EME test source configuration using piezoelectric lighter in close distance to the EME
sensor (a) and part of the spring mechanism on the interior of ignition device (b)

In a piezoelectric lighter, a spring system is used to impact a piezoelectric


crystal, which produces a strong electric field in the kilovolt range. This spring
mechanism is schematically shown in Fig. 5.25b, but may slightly vary in different
commercial lighters. In its original application, this spring-driven impact on the
piezoelectric material generates a spark that ignites the inflammable gaseous
medium. After removal of the gas of a commercial lighter, the ignition device is
still operating and may be used as EME test source. At a constant distance of 30 mm
and constant orientation relative to the sensor, this allows generation of reproduc-
ible and strong EME signals as seen in Fig. 5.26a detected by wire loop sensors (see
Sect. 5.4) with a total preamplification factor of 25 dB. In contrast to the approach
5.2 Source Mechanism 395

a
0.8

0.6

0.4

0.2
voltage [V]

0.0

–0.2
signal 1
signal 2
–0.4
signal 3
–0.6 signal 4

–0.8

–1.0
0 200 400 600 800 1000 1200 1400
time [μs]
b
0.8
signal 1
0.6 signal 2
signal 3
0.4 signal 4

0.2
voltage [V]

0.0

–0.2

–0.4

–0.6

–0.8

–1.0
250 255 260 265 270 275
time [μs]

Fig. 5.26 Typical EME signals of piezoelectric lighter source detected by wire loop sensors in
30 mm distance (a) and same signals zoomed to reveal first steplike signature in the test signal (b)

described above, this EME source does not show a distinct source radiation
characteristic. This is due to the metallic parts surrounding the piezoelectric
material, which tend to partially shield and distribute the generated electromagnetic
field. Experiments using various orientations to the detector did not reveal a
characteristic radiation pattern.
396 5 Electromagnetic Emission

5.3 Signal Propagation

After the charges are present at the crack tip, according to Maxwell’s equations, the
spatial movements of such charges will implicitly cause a fluctuation of the electric
and magnetic field strength and therefore form the source for an electromagnetic
wave [57]. The basic physical principles are well established, and for a compre-
hensive reading on this topic, the reader is referred to some recent literature in this
field [54–56]. To understand the propagation of electromagnetic waves in the
context of electromagnetic emission, it is first of all necessary to distinguish the
relevant aspects unique to this particular physical phenomenon from well-known
terms and phenomena in classical physics. A first distinction shall be made between
the term electromagnetic radiation and the term electromagnetic emission.
Electromagnetic radiation is associated with electromagnetic fields free to prop-
agate without the continuing influence of the moving charges that produced them,
because they have achieved sufficient distance to those charges. The electromag-
netic wave travels through space carrying radiant energy and thus are decoupled
from their source. This description is only valid in the far field of the source and,
therefore, can make use of several simplifications regarding the source and the
structure of the electromagnetic wave. For these electromagnetic waves in free
space, Maxwell’s equations directly result in two main classes of solutions, namely,
plane waves and spherical waves propagating at the speed of light in vacuum
conditions. These are convenient to describe using the angular frequency ω and
the wave number k. The propagation velocity of an electromagnetic wave of
particular frequency is given by the phase velocity cP:
ω
cP ¼ ð5:16Þ
k

Frequency and wave number are directly related to each other by the dispersion
relationship:

ω ¼ f ðk Þ ð5:17Þ

In the case of electromagnetic wave propagation in solid materials, the frequency is


not directly proportional to the wave number, causing dispersion upon propagation.
However, the impact of this dispersion for the interpretation of electromagnetic
emission is not nearly as severe as discussed for the guided acoustic emission waves
in Sect. 4.3. For a wave package of a narrow range of frequencies, the group
velocity is defined by the gradient of the dispersion relation:

∂ω
cG ¼ : ð5:18Þ
∂k

In an identical meaning as for acoustic waves, the group velocity is the velocity at
which the energy propagates through the medium.
5.3 Signal Propagation 397

source

reactive radiative
(Fresnel)
λ/2π
transition zone
near field region far field region

Fig. 5.27 Distinction between near-field and far-field regions based on wavelength and distance
to source

For the electromagnetic emission phenomenon discussed herein, the frequency


range from 10 Hz to 10 MHz results in typical wavelengths ranging from 30,000 km
to 30 m. This covers the full range from the ELF (extremely low frequencies) to the
HF (high frequencies) band. Still, the typical spatial dimensions of the source are
much smaller than the emitted wavelength.
In analogy to electromagnetically short antennas, this gives rise to the distinction
of four characteristic distances to the source position as shown in Fig. 5.27. Here the
near-field and far-field regions are defined by ratios of the distance r of the source to
the wavelength λ of the emitted wave. The near field in this case is defined as
distance r  λ. Within the near field, further distinction can be made between the
reactive and radiative region. The outer boundary of the reactive near-field region is
commonly considered to be at a distance r ¼ λ=2π. The radiative near field (also
known as the “Fresnel region”) covers the rest of the near-field region. The latter is
the boundary to the far field of the source, which is defined for any distance r 2λ.
The transition zone is understood as any distance λ  r  2λ and forms an
intermediate region in which both near-field and far-field effects are important. In
this region, near-field behavior decreases gradually, leaving far-field effects as
dominant behavior.
For electromagnetic emission, the used frequencies and typical source sizes
imply that the position of the sensing systems are located in the reactive near
field of the source, which e.g., for frequencies below 10 MHz is any distance
r < 4:77 m. Due to the limited sensitivity of the detection systems, the typical
source-sensor distance can even be expected in the centimeter range, and thus the
reactive near field would still be applicable for frequencies up to 100 MHz. Hence, a
particular focus is given in the following to explain the effects of wave propagation
in this reactive near-field region.
In the region very close to the source, the prediction of the electric and magnetic
field strength is comparatively complex, because geometric details of the source
and its vicinity as well as the electrical boundary conditions can have a strong
impact on the field lines. For a simple and ideal configuration like a dipole antenna,
the near field may still appear predictable, but is already hard to achieve by use of
analytical tools since multipole expansion is no longer valid and the full solution of
398 5 Electromagnetic Emission

c
a antenna b conductive obstacle

z z z
y y y
x x x
electric field lines

Fig. 5.28 Near field surrounding dipole antenna visualized using the electric field lines (a) and
same antenna with added conductive obstacle directly in parallel to dipole axis (b) and perpen-
dicular to dipole axis (c)

Maxwell’s equations is generally required. Using numerical methods as demon-


strated by the respective finite element solution shown in Fig. 5.28a, the electric
field lines are straightforward to obtain.2
Moreover, the addition of a metallic obstacle in the vicinity of the antenna causes
a strong distortion of the near-field appearance as visualized in Fig. 5.28b, which is
readily obtained using Maxwell’s equation in conjunction with numerical methods,
but is hard to predict solely by using analytical methods due to the boundary
conditions faced in a geometrically complex situation.
The term reactive near field indicates the nature of interaction within this
particular distance to the source. Here the detection process of electromagnetic
waves includes a substantial interaction with the source. This is in contrast to the
detection process in the far field, where the electromagnetic wave is fully decoupled
from the source, and hence, no interaction takes place. It is well known from other
fields of science that measurements in the reactive near field may give ambiguous
results. Hence it is of vital importance to concisely understand the interaction
between the source and the detection system.
For the case of an antenna as source representation, this interaction can readily
be understood by electrostatic and magnetostatic effects. For an alternating current
flowing in the antenna, a purely magnetic component is created in the near field,
which starts to collapse as the antenna current begins to reverse. This causes a
transfer of the magnetic field energy back to electrons in the antenna due to self-
induction. Consequently, this alternating process returns energy from the magnetic
field back to the antenna in a regenerative way. A similar consideration can be made
for the electric field as electric charges are generated. If charges are forced to locate
in one section of the antenna by the excitation voltage, a local electric field forms

2
It is worth noting that in the near-field region, the vectors of electric and magnetic field are not
always perpendicular, but may exhibit a phase shift relative to each other.
5.3 Signal Propagation 399

around that section of antenna due to the self-capacitance of the antenna. When the
excitation signal reverses, charges can dislocate from their forced regions, and the
previously generated electric field assists in pushing electrons back in the new
direction of their flow in a similar fashion as the discharge of a unipolar capacitor.
Due to this process, energy returns from the electric field back to the antenna
current.
Because of this energy storage and return, it is straightforward to conclude that
any additional inductive or electrostatic interaction in the range of the reactive near
field will cause a transfer of field energy from the antenna to another conductor in
close distance. Some prominent technical examples making use of this effect are
voltage transformers or transponder systems.
The amplitude of the electromagnetic field components close to the source can
be quite high, but decay with 1/r behavior if there is no disturbance of the
electromagnetic field. Hence, the field energy is strongest close to the region near
the source and may fall below detection limits very fast if the source strength is
weak. However, for an antenna as source, the usual understanding is that there is no
loss of field energy unless there is another object, such as a sensor in the area close
to the antenna. Thus, the near fields only transfer energy to very close conductors
and cause a measurable drop of energy at the antenna, when interacting with the
conductors. In the context of electromagnetic emission, these additional conductors
could be the detecting sensors, but also all other nearby objects such as shielding or
load frame components. To demonstrate this effect, the full configuration is
enclosed in a typical rectangular shielding chamber of 100 mm  100 mm  100 mm
(length  width  height) as seen in Fig. 5.29a (two surfaces are chosen transparent
to visualize interior). For experimental measurements, the shielding chamber typ-
ically comprises the largest piece of conductive material in the reactive near field of
the EME source. In addition, the source configuration of Sect. 5.2.1 is modified to
include two EME capacitance sensors with orthogonal orientation relative to each

a z
b
y 12
x
EME sensor
10

8
voltage [V]

4
specimen one sensor
grounded 2 two sensors - shielded
two sensors - unshielded
plate
0
0 50 100 150 200 250
time [μs]

Fig. 5.29 Model geometry including shielding chamber (a) and comparison of signal voltages at
sensor 1 with and without the presence of sensor 2 in shielded and unshielded configuration (b)
400 5 Electromagnetic Emission

other. Simulations of the detected EME signal at sensor position 1 are performed
with and without the presence of the second EME sensor system. All other model
settings confirm to the approach of Sect. 5.2.1.
As consequence of the addition of a second EME sensing system in the reactive
near-field range of the EME source, the measurable voltage drops slightly from a
maximum of 10.5–10.1 V. This effect demonstrates the impact of the reactive near
field and the direct interaction with the EME sensor system. However, given the
presence of the large conductive shielding chamber, the addition of the small
amount of conductive material of the EME sensor is almost negligible for practical
measurements. Nevertheless, other metallic pieces in close vicinity to the EME
source will have similar impact and may reduce the applicability of the presented
possibilities to detect EME signals as such or to measure the orientation of the
fracture surface. In direct comparison, a model using a spherical infinite element
domain surrounding the specimen as in Fig. 5.10 would mimic an unshielded case.
If an additional EME sensor is added to this unshielded case, a significant voltage
drop by a factor of 0.5 occurs. In this case, the amount of conductive material is
doubled by the presence of the second sensor system, and the electric field strength
redistributes accordingly.
For a real EME source, the behavior in the reactive near field is generally
expected to be more complex than for an antenna. Another important distinction
between an EME source and an antenna is given by the resulting near field of an
EME source. This is of transient nature since it is linked to the generation and decay
of charges of the fracture surface. So the description of alternating currents or
charges is only correct as far as the spatial movement is concerned. The driving
energy for the EME signal is the crack process, and the resulting field energy is
therefore increased as long as charges are generated at the crack tip. Hence, the flow
of energy back and forth from the field to the source as described for the antenna
above is of limited relevance in this context. Instead, the electromagnetic emission
process is better described by a single cycle unipolar generation of field energy. In
principle, the generated electric fields may couple back to the charges via self-
capacitance. But a coupling of field energy back to the charges is not expected
during the duration of source activity, since the respective frequencies and wave-
lengths do not allow back-coupling within the durations of a few 100 μs.

5.4 Detection of Electromagnetic Emission Signals

Unlike the other testing methods reviewed, the type of electromagnetic emission
measurements addressed herein are not yet covered by technical standards and have
not been turned into a commercial measurement technique so far. Consequently, no
common standard for EME detection is existent today. The standardization of a
detector concept is also challenging due to the variety of reasons already raised in
the previous sections. First, the detector establishes a coupled interaction with the
near field of the EME source being influenced in sensitivity and response by the
5.4 Detection of Electromagnetic Emission Signals 401

presence of nearby conductors. Hence, no unique transfer function can be expected


for one sensing system, but rather a unique transfer function for each particular
configuration. Second, no general guidelines exist for fabricating an EME sensing
system. To this end, various proposals for EME detection systems have been made
in literature [1–8, 12], all of them resulting in unique sensing systems.
The physical principle for EME detection can be fully established from Max-
well’s equations. Thus, the mode of operation can be readily obtained for every
individual configuration by use of analytical or numerical methods. Therefore, the
intent of this section is to provide an overview of sensing systems that have been
proposed in literature so far and to elucidate their strength and weakness in terms of
EME detection and EME signal interpretation. To facilitate an objective compar-
ison, most of the effects are discussed based on modeling results using the validated
approach of EME sensor modeling established in [12] applying a test source
configuration as presented in Sect. 5.2.
The second goal of this section is reviewing the important aspect of electromag-
netic shielding. Since this comprises a standard approach in the design of modern
electronic equipment, only the major aspects as relevant to EME measurements are
discussed. Along this line, the reduction of shielding effectiveness for different
challenges faced in practical EME testing is outlined, and results of experimental
shielding measurements are presented. Focus is given to the bandwidth of the
frequency spectrum relevant to the detection of EME and to provide some guide-
lines for design of shielding chambers to carry out mechanical testing in combina-
tion with EME detection.

5.4.1 EME Detector Concepts

As discussed in Sect. 5.3, EME detection constitutes a measurement in the near-


field range of the source. The relevant quantities to detect are the emitted electric or
magnetic field components. Whether the one or the other will be detected is subject
to the individual sensing system, since every sensor may be built to detect either
electrical field components or magnetic field components or a mixture of both. For
detection of the electric field component, most detector systems apply the genera-
tion of surface charges on a conductive sensing material to convert the incident
electric field to a measurable electric potential. For this field component detector
geometries like plates or other conductive parts acting as capacitance plates are
typically used. For magnetic fields the detector principle often relies on inductance
using such geometrical configurations like coils. Other detection systems like Hall
sensors, magnetoresistive devices, or superconducting quantum interference
devices (SQUID) magnetometers could be applied to the same end. However,
based on the considerations of the previous sections and the detector systems
used in the presented experiments, the following discussion of sensing systems
will focus only on the detection of the electrical field components.
Gade et al. previously presented a modeling approach, which allows computing
EME signals of almost identical shape to experimental EME signals [12].
402 5 Electromagnetic Emission

To perform a validation of the modeling approach, all required details of the


experimental configuration were implemented as 3D geometry. Using a system
transfer function accounting for the influence of the preamplifier and acquisition
system, the modeled signals at the EME sensor in Fig. 5.30a may be converted to an
EME signal for comparison with a respective experimental signal as plotted in
Fig. 5.30b.

a
5.5 modeled signal at detector
5.0
4.5
4.0
3.5
voltage [V]

3.0
2.5
2.0
1.5
1.0
0.5
0.0
0 100 200 300 400 500 600
time [μs]

0.04 experimental signal


modeled signal

0.02
voltage [V]

0.00

–0.02

–0.04

0 100 200 300 400 500 600


time [μs]

Fig. 5.30 Validation of EME sensor modeling using modeled signal voltage at detector (a)
subject to system transfer function and comparison to experimental signal (b) (based on [12])
5.4 Detection of Electromagnetic Emission Signals 403

An input surface charge density with a maximum of 3.3 nC/m2 applied as


steplike source function with superimposed oscillations resulted in a calculated
electric potential at the surface of a capacitance plate sensor as seen in Fig. 5.30a.
After convolution by the system transfer function, the modeled signal matches the
amplitude and shape of the experimental signals as compared in Fig. 5.30b. This
demonstrates that the proposed EME sensor modeling approach is suitable to
describe the experimentally obtained signals.
To compare different sensing systems and to point out some basic principles, all
detector geometries are evaluated within the same validated modeling environment
in the following. As modeled source type, the SENB test specimen geometry as
presented in the source modeling of Sect. 5.2 is used. This comprises a bar of solid
material (10 mm  2 mm  2 mm) with a constant charge density of 0.020 C/m2 on
the fracture surface at the center of the specimen. A sphere of 100 mm diameter
surrounds the test object and the sensing systems to allow visualization of the
electric field lines in a spherical volume of free space. The modeled sphere is
terminated by infinite element domains of 5 mm thickness, causing a virtual scaling
of the spatial coordinates to infinity as seen in Fig. 5.10 in Sect. 5.2. Thus the
modeled setup for the source and detector system has no nearby boundaries to
interact with. The floating potential at the conductive surfaces of the detector
system is input to a circuit simulation consisting of a parallel series of a 70 pF
capacitor and a 10 MΩ resistor to implicitly account for the measurement systems
band-pass characteristic. The resulting EME signal is then evaluated as voltage
across the 10 MΩ resistor. All computations are carried out using the “AC/DC
module” of COMSOL Multiphysics.
In the following, the results of different EME detector concepts are presented.
First, the geometrical configuration of the different concepts is outlined, and the
interaction of the detector geometry with the electric field lines is shown. Second,
the resultant transfer functions are discussed in terms of their absolute values and
their frequency characteristics. To facilitate the reading, these two aspects are
presented for three basic categories of sensor types. These are capacitance plate
sensors, wire sensors, and coil sensors. Finally, some basic factors of influence in
sensor configuration and selection are analyzed, and their relative sensitivity to
detect EME is discussed.

5.4.1.1 Capacitance Plate Sensors

As first basic category of EME detector systems, the results from capacitance plate
sensors are demonstrated. The geometrical configuration of the capacitance plate
sensor is shown in Fig. 5.31a including the modeled lateral dimensions of the plate
of a thickness of 1 mm. The configuration in Fig. 5.31a uses two solid plates
modeled using electrical properties of copper of Table B.3 in Appendix B. While
the surfaces of one of the plates is set as electrical ground (i.e., U ¼ 0 V), the
generated charges on the surface of the other plate are used as input to the electrical
circuit simulation. Modeling of a charged surface at the center of the test specimen
404 5 Electromagnetic Emission

a b c
test specimen
charged surface

z
solid plates 10mm z y
y y electric field lines
x x x

Fig. 5.31 Geometrical configuration of capacitance plate sensor (a), electric field lines in 3D
view (b), and electric field lines projected to xy-plane (c)

a b c
test specimen
charged surface
10mm

solid plate
z z y
y y
x x x
electric field lines

Fig. 5.32 Geometrical configuration of floating capacitance plate sensor (a), electric field lines in
3D view (b), and electric field lines projected to xy-plane (c)

causes the formation of electric field lines as shown in Fig. 5.31b, c. As seen from
the 3D view in Fig. 5.31b, the field lines emerge at the charged surface and
terminate at the conductive capacitance plate sensors, either on the grounded
plate or on the detector plate. The resultant pattern of electric field lines is even
better visualized in the projection of the field lines to the xy-plane shown in
Fig. 5.31c. The field lines “cutoff” at some distance to the detector plates is due
to the presence of the infinite element domain at this position. As mentioned above
this boundary domain allows a scaling of the spatial coordinates toward infinity, so
the field lines are truncated at the entry to the infinite element domain.
As small modification to the configuration of Fig. 5.31a is the removal of the
grounded capacitance plate as seen in Fig. 5.32a. For such configurations, the
ground wire is left unconnected, so the signal wire is operating on a floating ground.
This small modification causes some distinct differences in the shape of the field
lines. In contrast to the field lines in Fig. 5.32b, c, the floating capacitance plate
sensor forms a dipole-like pattern together with the charged surface. All field lines
emerge at the charged surface and terminate in the capacitance plate thus transmit-
ting all field energy into the sensing system. However, such ideal situations as given
by the infinite element boundary domain are hardly realized in practice. Instead,
5.4 Detection of Electromagnetic Emission Signals 405

a b c
charged test
surface specimen

z z y
y y
x x
electric field lines x

Fig. 5.33 Geometrical configuration of two orthogonal capacitance plate sensors (a), electric field
lines in 3D view (b), and electric field lines projected to xy-plane (c)

a b c
charged test
surface specimen

z z y
y y
x x electric field lines x

Fig. 5.34 Geometrical configuration of three orthogonal capacitance plate sensors (a), electric
field lines in 3D view (b), and electric field lines projected to xy-plane (c)

field lines will terminate into other nearby conductive media (cf. Fig. 5.28 in
Sect. 5.3) and therefore reduce the effectiveness of this sensor configuration.
In order to detect EME signals with more than one distinct angle relative to the
source, the geometrical arrangements of Figs. 5.33a and 5.34a are also interesting.
Using an assembly of either two capacitance plate pairs with perpendicular orien-
tation relative to each other or three capacitance plate pairs with different angles
relative to the source allows measuring the orientation of the fracture surface as will
be demonstrated in Sect. 5.5.2. However, as pointed out in Sect. 5.3, the increased
amount of conductive materials in the surrounding area of the EME sources reduces
the detection sensitivity of the individual sensor. This may also be readily derived
from the reduced density of electric field lines terminating in the detector plates in
Figs. 5.33b, c and 5.34b, c. Moreover, due to the orientation of the grounded
capacitance plates, the electric field lines are also oriented along the principal
axis of the geometrical configuration (i.e., 45 orientation in the xy-plane for
configuration in Fig. 5.33).
406 5 Electromagnetic Emission

5.4.1.2 Wire Sensors

The second basic group of EME detector systems is identified as all configurations
using single-ended wires. In contrast to the capacitance plate sensors, the basic
geometry of the detector system is thus build from a conductive wire instead of a
conductive platelike material. As simplest of all configurations, the open wire
sensor is shown in Fig. 5.35a. Here a small piece of a conductive wire is placed
close to the EME source or is formed into a wire loop. In practice, various
orientations of the wire loop may be used following the considerations of the source
radiation patterns established in Sect. 5.2.1.2. For the arrangement shown in
Fig. 5.35a, a 100 μm diameter wire made of copper was modeled as open loop
enclosing the charged surface. As seen from the resultant electric field lines in
Fig. 5.35b, c, this yields a similar dipole-like pattern as observed for the floating
capacitance sensor of Fig. 5.32.
Another geometrical arrangement using wire sensors is shown in Fig. 5.36a.
Here the wire is formed into a spiral loop of 10 mm diameter with open end. This
flat structure may then be placed next to an EME source in similar configurations
as presented for the capacitance plates in the previous section. Based on the
dipole-like appearance of the electric field lines, it may be concluded that

a b c
test specimen

charged surface

z z y
y open wire loop y
x x x
electric field lines

Fig. 5.35 Geometrical configuration of open wire sensor (a), electric field lines in 3D view (b),
and electric field lines projected to xy-plane (c)

a b c
test specimen
charged surface
10mm

open wire loop


z z y
y y
x x
electric field lines x

Fig. 5.36 Geometrical configuration of open wire plate sensor (a), electric field lines in 3D
view (b), and electric field lines projected to xy-plane (c)
5.4 Detection of Electromagnetic Emission Signals 407

a b c
test specimen
charged surface

10mm
closed wire
z z y
y y
x x electric field lines x

Fig. 5.37 Geometrical configuration of closed wire plate sensor (a), electric field lines in 3D
view (b), and electric field lines projected to xy-plane (c)

this configuration does effectively operate almost identical to a floating


capacitance plate. Since the overall dimensions of the modeled sensors are approx-
imately identical, the good match of the respective transfer functions shown later
in Fig. 5.41 is not surprising. Hence, this configuration is termed “open wire
plate” sensor.
A further possibility of wire-based sensor configurations is the use of closed wire
loops shown in Fig. 5.37a. Similar to the previous configuration, this resembles the
overall shape of a floating capacitance plate and is thus referred to as “closed wire
plate” sensor in the following. Here the signal wire is attached at an arbitrary
position along the circumference and the ground wire is left unconnected. As
seen from the Fig. 5.37b, c, the electric field lines interact with the sensing system
in a similar fashion as for the floating capacitance plate. As for the configuration
shown in Fig. 5.36a, this is due to the large wavelength of the electric field in the
frequency range up to 10 MHz, which does not allow to resolve the geometrical
difference between the configurations in Figs. 5.32a, 5.36a, and 5.37a.

5.4.1.3 Coil Sensors

The third basic group of EME detector systems comprises all sorts of coil config-
urations. Traditionally, coils are associated with inductance as primary principle of
interaction with electromagnetic fields. However, since the electric field component
is the quantity of interest, different approaches are required.
Together with the attached circuit, all coils form an electrical resonator. But as
long as the resonance frequency is sufficiently far away from the frequency range of
interest, such systems may still be used to detect EME signals with comparatively
flat transfer characteristics. And likewise, one can design resonator circuits with
resonance frequencies within the frequency range of interest, which are sensitive,
but narrow-band EME detector systems.
For the two modeled coil configurations shown in Figs. 5.38a and 5.39a, a wire
of 200 μm diameter was wound into coils of ten turns in circular cross-section
yielding an inductance of 250 nH. Together with the attached circuit, this yields a
408 5 Electromagnetic Emission

a b c
test specimen
charged surface

10mm
closed wire loop
z z y
y y
x x
electric field lines x

Fig. 5.38 Geometrical configuration of coil sensor with main axis in x-direction (a), electric field
lines in 3D view (b), and electric field lines projected to xy-plane (c)

a b c
test specimen
charged surface
10mm

closed wire loop


z z y
y y
x x
electric field lines x

Fig. 5.39 Geometrical configuration of coil sensor with main axis in y-direction (a), electric field
lines in 3D view (b), and electric field lines projected to xy-plane (c)

resonance frequency of 38 MHz and a bandwidth of 227 Hz with reasonable


sensitivity, which is sufficiently far away from the investigated frequency range
up to 5 MHz. To resemble a real experimental configuration, one cross-section of the
coil is grounded and the other cross-section is input the modeled electrical circuit.
In such conditions, the coil operates as simple metallic conductor, which causes
the formation of field lines as seen in Figs. 5.38b, c and 5.39b, c for the two different
coil configurations, respectively. Hence, the configuration of electric field lines is
again best described by a dipole-like shape with emerging lines at the charged
surface and terminating lines at the surface of the coil.

5.4.1.4 Comparison and Common Aspects

To compare the sensitivity of the different EME sensor systems investigated in the
previous section, it is useful to calculate their transfer functions. To this end, the
modeled test specimen is replaced by a small copper antenna, which is driven by a
sinusoidal voltage with 5 V peak amplitude. A frequency sweep is carried out for
each configuration ranging from 10 Hz to 5 MHz with nonlinear frequency steps
5.4 Detection of Electromagnetic Emission Signals 409

adapted to the logarithmic frequency scale. The circuit is kept identical to the
previous section. The transfer function is then derived from the voltage drop at
the 10 MΩ resistor divided by the source voltage. To validate this procedure,
experimental measurement results using the described configuration for the floating
capacitance plate and the wire sensor are shown in Fig. 5.40a, b as reference.

a
40
measurement
35 simulation

30

25
sensitivity [dB]

20

15

10

0
102 103 104 105 106 107
frequency [Hz]
b
40
measurement
35 simulation

30

25
sensitivity [dB]

20

15

10

0
102 103 104 105 106 107
frequency [Hz]

Fig. 5.40 Frequency characteristic of EME sensor using floating capacitance plate (a) and open
wire sensor (b)
410 5 Electromagnetic Emission

The corresponding modeling results were amplified by 26 dB to conform with the


experimentally used preamplifier gain. Quantitatively, the 1.8 dB difference in
sensitivity of the floating capacitance plate and the open wire sensor are well
described by the model configuration. While the high-pass characteristic of both
sensor systems is captured by the modeled high-pass given by the resistor and
capacitor in the attached circuit, the falloff at higher frequencies is not reflected
well in the present model configuration. This low-pass characteristic is owed to the
3 MHz cutoff frequency of the acquisition system used to record the experimental
data and therefore is not accounted for in the EME sensor model. However, since
the frequencies found for the experimental EME signals so far do not exceed
1 MHz, the present implementation of EME sensor modeling is assumed to be
sufficiently validated within the frequency range of interest.
Based on this experimental validation, it may be assumed that the evaluated
transfer functions of the remaining EME sensor systems are sufficiently accurate to
allow quantitative comparison of the different systems. A comparison of the
modeling results for the different sensor systems described in the previous sections
is shown in Fig. 5.41.
The floating capacitance plate was found to show the highest sensitivity among
the cases studied. Also, except for the high-pass characteristic given by the modeled
circuit, there is no deviation from a flat frequency response for this detector system.
All other floating sensor systems yield almost similar sensitivity, their differences
being attributed to the size of the detector surface and their relative orientation to
the source. The latter is seen best for the coil sensor investigated. For this sensor
type, there is no indication of resonances in the plotted frequency range as expected
from the calculated resonance frequency of 38 MHz.

40

20
sensitivity [dB]

0
floating capacitance plate
open wire sensor
capacitance plate
coil
–20 open wire plate
closed wire plate

102 103 104 105 106 107


frequency [Hz]

Fig. 5.41 Comparison of transfer functions of different EME sensor systems


5.4 Detection of Electromagnetic Emission Signals 411

One obvious result is the reduced sensitivity of the capacitance plate sensor with
grounded plate as compared to all other EME sensor types in Fig. 5.41. This is due
to the chosen configuration involving an explicitly grounded metallic plate close to
the EME source. As discussed in Sect. 5.4.1.1, this causes a partial termination of
electric field lines at the grounded plate and therefore reduces the sensitivity of the
sensing plate by 23 dB compared to the floating capacitance plate (an EME detector
of otherwise identical geometry).
However, these results need to be interpreted carefully. In reality all sensor
systems will need a respective electrical ground to allow a voltage measurement.
Surrounding conductive materials (i.e., fixtures, shielding chamber) which are
electrically grounded will thus reduce the sensitivity of the sensor system. In
addition, an increase in distance to the EME source will cause a significant reduction
of the measured signal strength as demonstrated in Sect. 5.2.1.3. But based on the
modeling results of Sect. 5.2, it may be concluded that for a fixed source-sensor
distance, the potential at the position of the sensor is a linear function of the electric
charges present at the crack surface. Consequently, the EME sensor transfer function
could also be expressed in terms of measured voltage per surface charges, yielding
the same curves as shown in Fig. 5.41 but resulting in different dB values.
Many other EME sensor configurations may be derived from the basic categories
presented in the previous sections, all of which will yield unique properties if the
presented aspect ratios are changed. To judge on their sensitivity and bandwidth, a
direct measurement of the sensor (and system) transfer function using the artificial
EME sources as described in Sect. 5.2.2 is recommended for every new design.
Apart from this, all EME sensor configurations share some common aspects, which
shall be discussed next. Among these aspects is the choice of sensor material, the
positioning of the sensor system, the preamplifier choice, and the acquisition system.

Material

The basic choice for fabrication of the EME detectors is a conductive material.
Trials with different electrically conductive materials did not reveal substantial
differences in the efficiency of the sensor system as long as the material is a
reasonable electrical conductor. This naturally includes all types of metals and
their alloys. Due to the good availability and high electrical conductivity, copper
alloys are the first choice for an EME detector material. Moreover, copper is easy to
machine and comes with good surface compatibility to allow soldering of electrical
wires for connecting the EME detector to cables.

Positioning

Similar to the considerations made for the artificial EME sources in Sect. 5.2.2,
there are some common guidelines to the positioning of EME sensor systems. It is
advantageous to fabricate all types of clamps or fixtures from nonconductive
412 5 Electromagnetic Emission

materials. Otherwise the presence of these sensor holders will adversely affect the
sensor performance. Furthermore, the EME sensor needs to be kept at a constant
position during the experiment, since any relative movement to the EME source
will induce an additional signal voltage. Hence, EME sensor fixtures may either rest
at a fixed (global) position or may be attached to the test specimen to retain a fixed
position relative to the specimen. The decision between either of these configura-
tions is subject to the respective experiment. Some practical examples for each
configuration are found in Sect. 5.5 to assist in an appropriate selection.

Preamplifier

As indicated in the previous sections, the system response of an EME sensor is


directly influenced by the attached circuit. For all practical considerations, this
means a cable ranging from the EME detector to the preamplifier and a second
cable ranging from the preamplifier to the acquisition system. The importance of
these settings may readily be understood from the change in the detected EME
signals shown in Fig. 5.42b for a variation of input capacitance ranging from 70 pF
to 140 pF and 210 pF. The value of the connected capacitance is dominated by the
input capacitance of the selected preamplifier and the length of the cable between
sensor system and preamplifier.
In analogy to the developments in acoustic emission measurement technology, it
was further found to be suitable to use a cascade of preamplifiers in order to
optimize the signal-to-noise ratio of the measurement chain. Thus, one J-FET
preamplifier is preferentially located directly attached or physically close to the
EME detector system and provides a first amplification of the detected signal
voltage. In a subsequent stage, the amplified signal is fed into another preamplifier
with broad operational bandwidth to retain the large bandwidth provided by the
EME sensor systems. A typical circuit for the first preamplifier stage is shown in
Fig. 5.42a and is based on the related work used for acoustic emission sensors by
Hamstad et al. and Shiwa et al. [58–60].

Acquisition

As final piece of the measurement chain, the amplified signal needs to be recorded
by an acquisition system. To this end, digital storage oscilloscopes or transient
recorders may be used. Similarly, typical commercial acoustic emission systems
capable of transient signal acquisition can be applied for this purpose. Compared to
the typical acoustic emission signals, one difference of the acquired signals is the
characteristic shape of EME signals, which may require some modifications to
classically used trigger settings. Also, EME signals of larger (macroscopically)
fracture events were found to have durations ranging into the millisecond ranges. If
these are meant to be stored as single event, the acquisition depth needs to be
adequate to store such long signals at the required digitization rate.
5.4 Detection of Electromagnetic Emission Signals 413

a
sensor J-FET

preamplifier
RV

18 V

10 MΩ 10 Ω

b
12.5 70pF
140pF
210pF
10.0
voltage [V]

7.5

5.0

2.5

0.0
0 100 200 300 400
time [μs]

Fig. 5.42 Typical preamplifier circuit for first stage (a) and variation of attached circuit
parameters (b)

5.4.2 Electromagnetic Shielding

Although the detection of EME signals may seem experimentally straightforward


based on the descriptions of the previous section, there is one considerable exper-
imental drawback involved in their measurement. Since electromagnetic fields obey
the superposition principle, any other electromagnetic field present during the
detection process may interfere with the useful EME signal of the source of interest.
Other external electromagnetic fields within the frequency range of the measure-
ment may also significantly contribute to the detected signal level and shape.
In most locations, there is a tremendous amount of sources for electromagnetic
waves in the relevant frequency spectrum. The sum of the contribution of all of the
sources is usually considered as electromagnetic noise, consisting of several char-
acteristic peaks and a comparatively continuous background noise floor. Hence,
electromagnetic shielding is the practice of reducing this electromagnetic noise by
blocking external fields with barriers made of conductive or magnetic materials.
414 5 Electromagnetic Emission

Thus, EME measurements are carried out within a shielding chamber suitably
reducing the surrounding electromagnetic noise level and thus allowing detection
of the electromagnetic field of the EME source. Generally, shielding effectiveness
needs to be distinguished for electric fields and magnetic fields, because different
physical mechanisms are relevant to reach the shielding of the one or the other.
As general measure for the respective shielding effectiveness (SE), the following
definition is conventionally used in electronics engineering for electric fields and
magnetic fields [61–63]:

jEunshielded j
SE ¼ 20 log ð5:19aÞ
jEshielded j

jHunshielded j
SE ¼ 20 log ð5:19bÞ
jHshielded j

The amount of reduction depends very much on the material properties, the
chamber thickness, the size of the shielded volume, and the frequency of the
electromagnetic fields. Moreover, the size, shape, and orientation of apertures
(e.g., penetrations, holes) in a shielding chamber may reduce the shielding effec-
tiveness and therefore need to be considered explicitly.
Especially in the reactive near-field distance, the electric and magnetic field
requires different methods of shielding, since the electric waves are typically being
reflected, whereas magnetic fields are absorbed instead.
Hence it is convenient to distinguish between three categories of electromag-
netic shielding:
1. Electric field shielding
2. Magnetic field shielding
3. Shielding enclosures
This distinction is quite useful, since electromagnetic fields first of all consist of
coupled electric and magnetic fields with individual requirements to shield the one
or the other field component.
The electric field results in forces acting on the charge carriers in the shielding
material. If this material is an ideal conductor, the electric field component induces
a current that causes internal charge displacement generating an opposed electric
field canceling the externally applied electric field. In contrast, oscillating magnetic
fields generate eddy currents inside a conductive material that drive an opposing
magnetic field canceling the externally applied magnetic field. For static magnetic
fields, there is no shielding effect unless the shielding material moves relative to the
static magnetic field. In both cases, the result is that electromagnetic fields do not
penetrate the shielding material. As a distinctly different category, shielding enclo-
sures are also used to protect a setup from unwanted external electromagnetic fields.
With increasing frequencies, it may happen that a simple shield starts to emit an
electromagnetic wave, which sometimes is even more intense than the original
source. This effect starts to dominate, when the dimensions of the shield approach
5.4 Detection of Electromagnetic Emission Signals 415

Fig. 5.43 Plane wave


incident on a conductive η0 ηS
material
Ein

Etr

x=0 x

Ere

fractions of the wavelength and thus act as antenna more than a shield. For such
cases and to shield electromagnetic fields in a broad frequency range, it is useful to
encapsulate a setup completely inside a perfectly conductive material.
Several factors limit the shielding effectiveness in practice. The finite electrical
resistance of the shielding material causes the excited internal field not to
completely cancel the incident external field. Also, for the case of magnetic
shielding, many conductive materials exhibit a ferromagnetic response at low
frequencies including hysteretic losses during generation of the eddy currents,
and thus the external field is not completely attenuated. As technical implication,
any geometric imperfections of the shield itself, such as apertures, holes, or
electrical connectors, reduce the overall shielding effectiveness.
In order to understand the shielding of electromagnetic fields, one can first of all
consider the incidence of a plane wave propagating from one material into another.
As with other laws of refraction, the incident electromagnetic wave is partially
reflected and partially transmitted when incident to an interface between two
materials of different electrical properties. For the case of orthogonal incidence of
a plane wave from one material with impedance η0 to another material with
impedance ηS, the incident electric field component Ein is partially transmitted Etr
and partially reflected Ere (Fig. 5.43).
The magnetic field orientation in a plane wave is perpendicular to the electric
field, and the magnetic field amplitude in free space with intrinsic impedance η0 is
related to the electric field amplitude by
sffiffiffiffiffi
ξ0 1
jHin j ¼ jEin j ¼ jEin j ð5:20Þ
μ0 η0

Similarly, the magnetic field amplitude in the shielding material is directly linked to
the electric field amplitude by

1
jHtr j ¼ jEtr j ð5:21Þ
ηS
416 5 Electromagnetic Emission

In addition, at the interface between both materials, continuity of the electric field
vector and the magnetic field vector is required. Thus the amplitude of the reflected
electric field component may be written in terms of a characteristic ratio of both
intrinsic impedances and the incident amplitude:
ηS  η0
jEre j ¼ jEin j ð5:22Þ
ηS þ η0

And the transmitted electric field amplitude in a similar way as

2ηS
jEtr j ¼ jEin j ð5:23Þ
ηS þ η0

Equations (5.22) and (5.23) imply that as ηS approaches η0, the transmission
increases and the reflection decreases. For matching impedances ηS ¼ η0 , the
incident field is fully transmitted.
As additional effect, a real material has finite conductivity (i.e., σ~ > 0), and the
transmitted electric field amplitude will decrease with distance to the surface due to
the skin effect. The electric field amplitude as function of depth below surface may
thus be written as function of the skin depth δ of the material:

jEtr ðxÞj ¼ jEtr ðx ¼ 0Þjeδ


x
ð5:24Þ

The skin depth δ is usually well approximated assuming


sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2
δffi ð5:25Þ
ω~ σ μr μ0

using the conductivity of the material σ~ , the angular frequency ω, the relative
magnetic permeability of the material μr, and the permeability of free space μ0.
When the electromagnetic wave is incident to a material with finite thickness h,
the transmission through the shielding material is readily obtained by combination
of the above effects. The incident wave is partially reflected upon penetration of the
material, then attenuated within the material, and finally partially transmitted into
free space as schematically shown in Fig. 5.44. The law of reflection for transmis-
sion at the interface from inside the shielding into free space at x ¼ h is given as:

2η0
jEtr j ¼ jEtr ðx ¼ hÞj ð5:26Þ
η0 þ ηS

By combination of the equations above, the finally transmitted electric field


amplitude may be written for all materials being substantially thicker than one
skin depth δ as:
 
2ηS 2η0
e δ
h
jEtr j ¼ jEin j ð5:27Þ
ηS þ η0 η0 þ ηS
5.4 Detection of Electromagnetic Emission Signals 417

Fig. 5.44 Plane wave


η0 ηS η0
incident on a shielding
material of finite thickness Ein

Etr

x=0 x

Ere
h

It can be derived from (5.27) that good conductors (~ σ ωξ) will yield the best
shielding effectiveness, since this implies ηS  η0 .
Based on the definition of the shielding effectiveness in (5.19) and the
components of (5.27) the overall contribution to the shielding effectiveness may
be split into two parts accounting for the reflection loss R(dB) and the absorption
loss A(dB):
!
4ηS η0  h
SE ¼ 20 log þ 20 log eδ ¼ RðdBÞ þ AðdBÞ ð5:28Þ
ðη0 þ ηS Þ2

The magnitude of R(dB) is thus independent of the thickness and comprises a


constant contribution to the shielding effectiveness, while the contribution of A
(dB) is directly influenced by the ratio between thickness and the skin depth of the
material.
Since the reduction of the electrical field component is of primary interest for
EME measurements, some shielding effectiveness curves are shown in Fig. 5.45 for
a variation of material thickness. Using aluminum with electrical properties as
listed in Table B.3 in Appendix B, the calculated shielding effectiveness following
(5.28) as function of thickness is plotted in Fig. 5.45. For the frequency range of
interest for EME measurements, the thinnest configuration of 100 μm thickness
(i.e., aluminum foil) already provides a shielding effectiveness above 100 dB. For
practical applications a shielding effectiveness >40 dB was found to be sufficient to
reduce the surrounding electromagnetic noise beyond the detectable limit. How-
ever, handling of such thin shields in the framework of mechanical testing is quite
difficult, so it is more realistic to use shielding chambers within the thickness range
of some millimeters. For a thickness of 1 mm, the contribution of A(dB) starts to
dominate and is found even more pronounced for a thickness of 10 mm both of
them yielding even higher shielding effectiveness.
As materials being used for electromagnetic shielding, metals are obviously the
first choice. In order to construct shielding chambers of a decent size and with a
thickness providing sufficient mechanical stability, this limits the type of metallic
materials to typical engineering alloys based on aluminum, steel, or copper. Other
metals, such as silver, gold, or platinum would also be interesting for shielding
applications but are economically not reasonable to apply. Using (5.28) the
418 5 Electromagnetic Emission

500

400
shielding effectiveness [dB]

300

200

100
Aluminum - thickness 0.1 mm
Aluminum - thickness 1.0 mm
Aluminum - thickness 10.0 mm
0
103 104 105 106
frequency [Hz]

Fig. 5.45 Shielding effectiveness of aluminum shield as function of thickness and frequency

500

400
shielding effectiveness [dB

300

200

100
Aluminum - thickness 1.0 mm
Copper - thickness 1.0 mm
Steel - thickness 1.0 mm
0
103 104 105 106
frequency [Hz]

Fig. 5.46 Shielding effectiveness of shield with 1.0 mm thickness as function of material and
frequency

shielding effectiveness for a chamber made from 1.0 mm thick, aluminum, steel,
and copper applying the electrical properties of Table B.3 in Appendix B are plotted
in Fig. 5.46. Based on their different properties, it can be observed that steel would
yield the best shielding solution in the relevant frequency range. However, based on
handling considerations, a shielding chamber made of steel is less desirable than
5.4 Detection of Electromagnetic Emission Signals 419

one made of an aluminum alloy. Since the gain in shielding effectiveness at


frequencies below 10 kHz is also fairly low, this makes aluminum the best choice
for a solid shielding chamber for an EME application.
As alternative to metallic materials, attempts have been made to use carbon fiber
reinforced materials to shield electromagnetic fields [64]. Another commonly used
material modification is to coat the inside of the enclosure with a metallic ink. The
ink itself consists of a carrier material loaded by suitable metallic particles and thus
provides some shielding capability.
Although the sensing systems discussed in Sect. 5.4.1 are not intended to detect
the magnetic field component, it may occur that the detector system or the pream-
plifier electronics also need protective shielding against external magnetic fields.
For frequencies above 100 kHz, the dominating effect of shielding is the generation
of eddy currents. However, for parts of the relevant frequency range (below about
100 kHz), static or slowly varying magnetic fields are not well shielded by the
generation of eddy currents. This is also readily explained by the skin effect.
Similar to the electric field, also the magnetic field penetrates the shielding material
inducing a voltage. For an incident magnetic field causing a charge flux

QðtÞ ¼ Q0 sin ðωtÞ ð5:29Þ

this results in a voltage

dQ
U ðt Þ ¼ ¼ Q0 ω cos ðωtÞ ð5:30Þ
dt

Therefore, the induced voltage increases with increasing ω, causing increasing eddy
currents to cancel the incident magnetic field. In a similar fashion as the electric
field described above, the amplitude of the magnetic field also decreases exponen-
tially with distance to the surface. This skin effect is minimal for low frequencies,
allowing most of the magnetic field to be transmitted through the shielding material.
But for sufficiently high frequencies, the shielding maximizes and will reach a
sufficient level.
To improve the shielding effectiveness at lower frequencies, a modification of
the shielding material toward high magnetic permeability metal alloys or ferro-
magnetic metal coatings is suitable. Since the shielding material is acting as path for
the magnetic field lines, the best design in this context is a fully closed container
surrounding the setup (i.e., a shielding enclosure).
As general drawback with shielding enclosures of finite size, the geometrical
arrangement itself may be subject to resonances, thus reducing the shielding
effectiveness at these frequencies. For a simple rectangular box of length a, height
h, and width b, the resonance frequencies are given for an arbitrary material by
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 2  2  2ffi
c iπ jπ kπ
f ijk ¼ pffiffiffiffiffiffiffiffi þ þ ð5:31Þ
2π μr ξr a h b
420 5 Electromagnetic Emission

This equation can be simplified for the case of an empty box (vacuum conditions) to
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 2  2  2
i j k
f ijk ¼ 150 þ þ ð5:32Þ
a h b

The fundamental resonance case i ¼ j ¼ k ¼ 0 is forbidden by definition of (5.32).


Thus for a realistic size of a shielding enclosure of dimensions 200 mm 
300 mm  100 mm (length  width  height), the first resonances are expected
for 901, 1250, and 1580 MHz. Since these are well outside the frequency range
of EME signals, the primary resonances and their higher orders are not relevant to
consider for construction of shielding applications.
So far all considerations have been made for shields or enclosures of sufficient
geometrical size to prevent anything within the shielded volume from electrically
coupling to anything outside. Such an ideal shielded enclosure with sufficient
conductivity and no electrical discontinuities would perfectly shield whatever is
inside the enclosure from whatever is outside. In reality all shielding enclosures
exhibit deficiencies which reduce the shielding effectiveness and thus require
separate discussion in the following subsections. In most practical applications,
there are apertures, cable penetrations, or seams which need to be considered in
their effect on the shielding effectiveness.

5.4.2.1 Apertures

For practical use of EME in combination with mechanical testing, the aim is to
shield the specimen and the EME detector system. Since the specimen itself is
attached to load fixtures, there are only two obvious possibilities:
1. Shielding of the specimen, detectors, fixtures, and testing machine within a
shielding chamber
2. Fabrication of shielding chamber for specimen and detectors with apertures to
connect to fixtures
While the first one would not only require a sufficiently large shielding chamber
to enclose the full test equipment, it even comes with the disadvantage of enclosing
other potential noise sources, such as the electromotor or computer control systems
with the specimen and the detector.3 Thus it may not necessarily constitute the best
solution for this kind of application. Instead, the fabrication of shielding chambers
with apertures was found to be the best economical and technical solution. As basic
rule of thumb, any apertures must be significantly smaller than the incident wave-
length of the electromagnetic waves to be shielded. Else parts of the incident waves

3
Noise measurements with the same EME sensor system did not reveal a substantial difference of
the noise floor when comparing a commercial walk-in shielding chamber and a self-made
shielding chamber directly attached around the specimen (as used in [12]).
5.4 Detection of Electromagnetic Emission Signals 421

a b aperture
voltage voltage
1 1
0.95 0.95
0.9 0.9
0.85 0.85
0.8 0.8
0.75 0.75
0.7 0.7
0.65 0.65
0.6 0.6
0.55 0.55
0.5 0.5
0.45 0.45
0.4 0.4
0.35 0.35
0.3 0.3
0.25 0.25
0.2 0.2
0.15 0.15
0.1 0.1
0.05 0.05
0 0

100 Hz 10 MHz
shielding

Fig. 5.47 Leakage of electric field component due to 20 mm aperture at 100 Hz (a) and 10 MHz
(b). Contour lines show electric potential and gray scale represents magnitude of electric field
strength (white being zero)

might be transmitted into the shielded volume, and the total shielding effectiveness
might be reduced, because of the discontinuity of electric conductivity of the
shielding enclosure. This “leakage” is shown as a result from a numerical study
using an electromagnetic wave incident to a metallic enclosure as calculated within
the AC/DC module of COMSOL Multiphysics. The enclosure includes an aperture
of 20 mm at the top edge. At 100 Hz, the aperture is still smaller than the
wavelength and the shielding effectiveness of the chamber is not affected
(cf. Fig. 5.47a). As seen from the same calculation at 10 MHz frequency in
Fig. 5.47b, the electric field component is able to penetrate into the shielded
volume, causing a reduction of the overall shielding effectiveness.
The wavelengths of electromagnetic waves in the frequency range up to 10 MHz
are larger than 30 m. Hence, all practically relevant aperture sizes for laboratory
size specimens (order of few millimeters) may be assumed to be much smaller than
the incident wavelengths.
Some measurements of the shielding effectiveness using an aluminum chamber
of 320 mm  320 mm  120 mm (length  width  height) and 5 mm thickness are
shown in Fig. 5.48a. First, the EME sensor is mounted unshielded in free space to
collect the locally present noise floor, which is the black FFT spectrum in
Fig. 5.48a. Mounting the sensor inside the box, but leaving one side surface open,
causes a substantial reduction of the collected noise background as seen by the blue
line. If the cover plate is mounted on the box, the shield chamber is fully closed and
the incident background noise almost eliminated within the frequency range up to
5 MHz. The remaining falloff in the detected intensity as function of frequency is
dominated by the characteristics of the EME sensor and the acquisition system. In
the fully enclosed shield chamber, a circular aperture of 10 mm diameter is drilled
and then the measurement is repeated. There is no significant increase of the noise
floor. Even after drilling of five apertures of 10 mm diameter close by, no change in
the detected noise floor is observed.
422 5 Electromagnetic Emission

a
4
unshielded
shield chamber closed
shield chamber open

3
FFT magnitude [1/V]

0
103 104 105 106
frequency [Hz]
b
2.0
shield chamber closed
10 mm aperture
5 x 10 mm aperture
1.5
FFT magnitude [1/V]

1.0

0.5

0.0
103 104 105 106
frequency [Hz]

Fig. 5.48 Measurements of the electric noise floor using shield chamber (a) and repetitive
measurements after application of apertures (b)

Such apertures are obviously required as lead through for mechanical fixtures or
cables, but are also useful to look into the shielding chamber to monitor specimen
behavior. If some leakage occurs due to the presence of apertures, it is possible to
increase the depth of an aperture by adding a metallic skirt, which acts as a
waveguide below cutoff frequency.
5.4 Detection of Electromagnetic Emission Signals 423

waveguide
a b
voltage voltage
1 1
0.95 0.95
0.9 0.9
0.85 0.85
0.8 0.8
0.75 0.75
0.7 0.7
0.65 0.65
0.6 0.6
0.55 0.55
0.5 0.5
0.45 0.45
0.4 0.4
0.35 0.35
0.3 0.3
0.25 0.25
0.2 0.2
0.15 0.15
0.1 0.1
0.05 0.05
0 0
100 Hz 10 MHz
shielding

Fig. 5.49 Reduction of leakage of electric field component using waveguide shown at 100 Hz (a)
and 10 MHz (b). Contour lines show electric potential and gray scale represents magnitude of
electric field strength (white being zero)

5.4.2.2 Waveguides Below Cutoff

Using the geometry and computation setup of the previous section, the aperture is
modified by adding a metallic skirt perpendicular to the aperture diameter as seen in
Fig. 5.49. This modification is motivated by the idea that energy will not propagate
within an electromagnetic waveguide at frequencies below its cutoff frequency.
Based on the cross-section of the obtained waveguide, the attenuation as function of
waveguide length d may be estimated as:
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 2
d f
attenuation ½dB ffi 27 1 ð5:33Þ
a fc

for a rectangular waveguide of cross-sectional lengths a and b with b < a.


The cutoff frequency of the lowest propagation mode (the TE10 mode) for a
hollow waveguide is given by using the propagation velocity c ¼ 3  108 m/s of
the wave as
c c
fc ¼ ¼ ð5:34Þ
λc 2a

Similarly, for a circular waveguide of diameter a, one may estimate the


attenuation as:
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 2
d f
attenuation ½dB ffi 32 1 ð5:35Þ
a fc
424 5 Electromagnetic Emission

Equation (5.35) uses the cutoff frequency of the lowest frequency propagation mode
(the TE11 mode in this case):

0:586 c
fc ¼ ð5:36Þ
a
An example of the reduced field leakage into the shielded volume is seen for
10 MHz frequency in Fig. 5.49b when compared to Fig. 5.47b without a waveguide
attached to the aperture. At frequencies below the cutoff, such as 100 Hz, no
penetration into the shielding chamber occurs.
Another practical implication for EME testing is the inverse use of a waveguide.
If a mechanical fixture is applied to load the specimen inside the shield chamber,
the typical material choices are metallic fixtures due their high stiffness, strength,
and wear resistance. However, metals come with high electric conductivity and
will constitute a preferential path of leakage into the shielding chamber. This effect
is shown in Fig. 5.50a using the same modeling concept as described above.
Clearly, the metallic rod acts as waveguide allowing the electric field to penetrate
the shielding chamber. In the EME experiments, it was found that a highly
conductive material will not work as fixture system, since it compromises the
effect of the shielding chamber almost completely. The usage of low conductive
materials, such as a polymer rod seen in Fig. 5.50b, does not affect the shielding
effectiveness of the chamber. Hence, fixtures for EME measurements need either
to be fabricated out of materials with poor electric conductivity or metallic fixtures
need to be electrically isolated so that no conductive path from the outside to the
inside of the shielding chamber exists. In this context, PMMA was found to be a
primary material for fabrication of fixtures at lower loads, while GFRP was found
to be a reasonable alternative for construction of fixtures subject to higher mechan-
ical loads.

a metallic rod b polymer rod


voltage voltage
1 1
0.95 0.95
0.9 0.9
0.85 0.85
0.8 0.8
0.75 0.75
0.7 0.7
0.65 0.65
0.6 0.6
0.55 0.55
0.5 0.5
0.45 0.45
0.4 0.4
0.35 0.35
0.3 0.3
0.25 0.25
0.2 0.2
0.15 0.15
0.1 0.1
0.05 0.05
0 0

10 kHz shielding 10 kHz

Fig. 5.50 Penetration of electric field components in shield chamber due to metallic rod pene-
trating the shield (a) and reduced penetration using polymer rod (b). Contour lines show electric
potential and gray scale represents magnitude of electric field strength (white being zero)
5.4 Detection of Electromagnetic Emission Signals 425

5.4.2.3 Seams

Similar to the leakage effects due to apertures, other effects come into play when
fabricating shielding chambers in practice. Obviously, no metallic box exists,
which is fully closed, since this is hardly useful in practical applications. Instead,
shield chambers either have some larger apertures or are closed by lids or covers
during operation. In the context of EME measurements, these are used to mount the
specimen and the sensor system as well as to wire the cables inside the shield
chamber. Potentially a leakage of the electric field may occur at the positions of
bad electrical contact, which typically are the interfaces between the shielding
chamber and the covers and lids. As seen in Fig. 5.51, there is a variety of
connector mechanisms available to ensure a good electrical connection between
two metallic parts.
Generally, seams between the cover and the shield chamber may form a more
significant source of leakage than apertures, since they typically are of much larger
dimension. Seams in the order of half-wavelengths may act as dipole-like radiation
source causing a substantial increase of the field strength within the shielded
volume. Generally, it is not advisable to have simple touching seams as seen in
Fig. 5.51 for constructing shield chambers. The disadvantage of these configura-
tions is that they may have bad electrical contact over long distances, while they
physically still appear to have contact. Similar challenges arise with rivets or
screws, since they provide a good local contact, but may cause waviness of the
casing and therefore inhibit good electrical contact in the area between their spots.
Seams of shielding chambers are thus typically constructed using overlapping
seams, finger stocks, or conductive gaskets to assure a good electrical contact
along the full length of the seam.
The leakage of such seams is shown in Fig. 5.52a, b applying the same modeling
setup as in the previous sections for two geometries. Instead of a perfect shield
chamber, the upper edge is modified by incorporating the geometric dimensions of
a seam as seen in the magnified images. At frequencies of 50 MHz and above, the
two seams allow a significant portion of the incident field strength to penetrate the
shield chamber. However, this may easily be compensated by use of a longer
overlapping distance applying the concept of using a waveguide below cutoff
frequency.

touching seam screwed seam finger stock

touching seam overlapping seam conductive gasket

Fig. 5.51 Scheme of different types of connector mechanisms used to combine electrically
conductive materials for shielding chambers
426 5 Electromagnetic Emission

a seam
voltage
voltage

50 MHz shielding

b seam
voltage
voltage

50 MHz shielding

Fig. 5.52 Leakage of electric field component due to two different seams at 50 MHz. Contour
lines show electric potential and gray scale represents magnitude of electric field strength (white
being zero)

5.4.2.4 Cable Penetrations

As final component required to build an EME shield chamber for mechanical


testing, the EME detectors need to be connected with acquisition equipment on
the outside of the chamber. Thus it is necessary to consider cable penetrations. In
the first instance, these consist of an aperture to allow physical penetration of the
shield chamber by the cables. However, the conductive materials in the cable will
cause a similar reduction of the shielding effectiveness of the chamber as discussed
for the metallic rod used as fixture system above. Here, a single unshielded,
unfiltered wire penetrating a shielded enclosure can completely eliminate any
shielding benefit that the enclosure will provide.
For practical EME applications, it has been found a useful practice to keep the
shield of the incident wires at the same electric potential as the shielding chamber at
all relevant frequencies. For this reason, the wire shielding needs to make a
low-inductance connection to the enclosure as seen in Fig. 5.53a. This is typically
achieved by a conductive shell enclosing the signal and ground wire penetrating the
5.5 Application to Composites 427

a b
shield shield

conductive shell non-conductive shell

cable shield signal pigtail connection


and
ground

Fig. 5.53 Shielding of cable connections when penetrating shielding chambers using conductive
(a) and nonconductive (b) shell

shield. Such conductive shell connectors are readily available for BNC cables as
female-to-female connectors allowing direct connection of a BNC cable on both
sides of the shield. Using nonconductive shells in combination with a pigtail
connection as in Fig. 5.53b is not recommended, since this will result in a connec-
tion with significant inductance.

5.5 Application to Composites

Based on the previously discussed possibilities of the EME method, the focus of
this section is the presentation of some experimental studies carried out under
laboratory conditions, which have been conducted to assess the capabilities of
EME measurements of material failure. The section starts with presentation of
some exemplary studies and measurement setups used to detect EME signals due
to crack formation. Subsequently some first applications with respect to measure-
ment of fracture surface orientation are provided. The section closes with a discus-
sion of the method sensitivity to detect material failure.

5.5.1 Measurement of EME Due to Crack Formation

Within this section examples of various experimental configurations are given,


which have been used to study EME due to crack formation in materials. Since
the direct application of a barely understood measurement method to the complex-
ity of failure modes found in composites is quite challenging, some results in
application to study polymer failure and failure of carbon fibers are presented
first. This introduces some basic test concepts and allows understanding the gener-
ation of EME signals in general.
428 5 Electromagnetic Emission

5.5.1.1 Polymer Failure

In order to induce crack growth in pure polymeric material, several test concepts are
established in fracture mechanics. In general, distinction is made between the
suitability of an experimental setup for particular fracture modes, such as a pure
mode I loading. Due to the broad range of fracture behavior of polymers ranging
from brittle to ductile, a common test standard for a particular fracture mode is
challenging.
An experimental setup used for mode I testing has to fulfill several requirements.
From the mechanical point of view, the load introduction has to ensure that a crack
opening mode is induced in a notched geometry without significant contributions of
plastic deformation at other position than at the notch tip (e.g., at the positions of
load introduction). In order to allow a direct comparability of the detected EME
signals, identical test geometries are beneficial. To this end simple compact-tension
(CT) specimens, double-edge-notched tension (DENT) specimens, and tapered
double-cantilever beam (TDCB) specimens as shown in Fig. 5.54 were evaluated
for their applicability with a broad range of technically relevant polymers such as
epoxy resins (RTM6), polypropylene (PP), polyether ether ketone (PEEK), and
polytetrafluoroethylene (PTFE).
For the RTM6 specimen, fully unstable crack growth was observed for the CT
and DENT geometry once a crack has initiated. Therefore, only one EME signal
was detected during these measurements. For the more ductile thermoplastic
polymer types, no crack initiation was found in the CT and DENT geometry.
Instead, only large plastic deformation occurred at the positions of the pinholes
used for load introduction. Thus no EME signals were recorded. In contrast, the
TDCB geometry shown in Fig. 5.54 provided many EME signals for the RTM6

Fig. 5.54 Comparison of different specimen geometries as investigated for mode I testing of
polymers (all dimensions in mm)
5.5 Application to Composites 429

specimens investigated. This is due to the self-inhibiting crack growth caused by


the tapered design of the cantilever beams. For the more ductile materials, large
plastic deformation was observed at the pinholes used for load introduction.
Consequently the tapered design of the cantilever beams was retained, but the
contour was optimized to avoid excessive plastic deformations at the pinholes.
An increase of specimen thickness (along y-axis) was not feasible, since curing of
RTM6 specimens causes exothermic reactions which lead to specimen degradation
for thicknesses values above 5 mm. As a result of a numerical geometry optimiza-
tion procedure, the most promising geometry that was found to allow crack growth
for brittle and ductile specimens is shown in Fig. 5.55. For this configuration, the
EME sensor was realized by two horizontally oriented wire loops as shown in red.
Those wires form two effective capacitance plates with plate normal directed along
the z-axis.
As seen in Fig. 5.55, besides the EME sensor, two acoustic emission sensors of
type KRNBB-PC are also attached to the TDCB specimens, which are used to
detect the respective acoustic emission signal. All acquisition settings are listed in
Table 5.2. All measurements are carried out in an aluminum shielding chamber of
320 mm  120 mm  320 mm (length  width  height) and 5 mm wall thickness.
All TDCB specimens were loaded by the fixture shown in Fig. 5.56a. The cross-
head velocity was chosen as 5 mm/min with force measurement performed by a
5 kN load cell. The fixture was machined out of stainless steel components with a
pull-rod out of polyoxymethylene. This is to avoid the presence of a conductive
path to the volume outside the shielding chamber. The pull-rod was designed to
yield negligible compliance for the load capacity of 5 kN. Representative load-
displacement curves for each of the four polymers are plotted in Fig. 5.56b. While
RTM6 and PEEK specimens exhibit almost brittle failure, the two types PP and

Fig. 5.55 Optimized TDCB specimen geometry used for mode I testing of polymers including AE
and EME sensor positions (all dimensions in mm)
430 5 Electromagnetic Emission

Table 5.2 EME and AE acquisition settings used for TDCB tests
Acquisition settings Value EME Value AE
Preamplification 23 dB 20 dB
Threshold 35 dB 35 dB
Triggering Individual channel Individual channel
Acquisition rate 5 MS/s 10 MS/s
Bandpass range 1 kHz to 3 MHz 1 kHz to 3 MHz
Mounting system Fixed on specimen PMMA nut and clamp
Sensor type Closed wire loop KRNBB-PC
Number of sensors 1 pair 2

a b
Force
2000
POM rod PEEK
RTM6
z PP
x 1500
PTFE
force [N]

y
1000

500

0
EME sensor 0 1 2 3 4 5
displacement [mm]
stainless steel fixture

Fig. 5.56 Schematic of measurement setup (a) and respective force-displacement curves for the
investigated polymer types (b)

PTFE undergo substantial plastic deformation prior to failure, which is expressed in


the pronounced nonlinearity of the curves seen in Fig. 5.56b.
A representative collection of EME signals is given in Fig. 5.57 for each of the
four different polymers. Comparison is made relative to the acoustic emission
signals detected by AE sensor 2 shown as black curve. The EME signals in
Fig. 5.57 are plotted as detected (red lines) and after filtering by a seventh order
Butterworth high-pass filter of 20 kHz to reveal the oscillating part of the signal. As
seen from the EME signals of Fig. 5.57, characteristic signals are obtained for each
polymer type. First of all this is expressed by the different shapes of the signals,
comprising different rise-times and decay times as may be perceived in the unfil-
tered signals. Based on the theory presented in Sect. 5.2, this corresponds to
different durations of crack propagation. The latter may arise due to the differences
5.5 Application to Composites 431

in the overall propagation length or due to differences in the crack velocity. Since
the crack length may readily be derived from microscopic measurements of the
fracture surfaces after failure, this allows evaluation of the maximum crack velocity
for each of the cases shown in Fig. 5.57. These evaluate as 1475 m/s for RTM6,

a
2.00 RTM6

1.00
voltage [V]

0.00

–1.00

–2.00

10
0.15 filtered by 20 kHz Butterworth high-pass 7th order
0.10 5
unfiltered

voltage [mV]
voltage [V]

0.05
0.00 0
–0.05
–0.10 –5
–0.15
–10
0 50 100 150 200 250 300 350 400
time [μs]

b
4.00
PP
2.00
voltage [V]

0.00
–2.00
–4.00
–6.00

10
0.10
5
voltage [mV]

0.05
voltage [V]

0.00 0
–0.05
unfiltered –5
–0.10 filtered by 20 kHz Butterworth high-pass 7th order
–10
0 50 100 150 200 250 300 350 400
time [μs]

Fig. 5.57 Set of representative AE signals (black) and EME signals (red and blue) as acquired
during tests of TDCB specimen geometries in RTM6 (a), PP (b), PTFE (c), and PEEK (d)
432 5 Electromagnetic Emission

c
0.45 PTFE
0.30
voltage [V]

0.15
0.00
–0.15
–0.30

1.0
0.09 filtered by 20 kHz Butterworth high-pass 7th order
0.06
unfiltered 0.5

voltage [mV]
voltage [V]

0.03
0.00 0.0
–0.03
–0.5
–0.06
–0.09
–1.0
0 50 100 150 200 250 300 350 400
time [μs]

d
6.00
PEEK
4.00
2.00
voltage [V]

0.00
–2.00
–4.00
–6.00

15
0.20 filtered by 20 kHz Butterworth high-pass 7th order
10
voltage [mV]

0.10 unfiltered
voltage [V]

5
0.00 0

–0.10 –5
–10
–0.20
–15
0 50 100 150 200 250 300 350 400
time [μs]
Fig. 5.57 (continued)

1596 m/s for PP, 294 m/s for PTFE, and 2130 m/s for PEEK and therefore are
well below the expected Rayleigh wave velocities of the four materials
ðcR ¼ 2586 m=s for RTM6, cR ¼ 2625 m=s for PP, cR ¼ 1321 m=s for
PTFE, and cR ¼ 2493 m=s for PEEK).
5.5 Application to Composites 433

In addition, characteristically different amplitude levels of the EME signals are


observed. These can be attributed to systematic differences in the number of
generated charges per fracture surface as well as to the differences in absolute
crack propagation lengths.
After application of the 20 kHz high-pass, the EME signals tend to appear
similar to the respective AE signals. Since the “oscillating” part of the EME signal
should be directly proportional to the movement of the crack walls, which act as AE
source, this similarity is not unexpected. However, only the first few cycles of the
AE signal are expected to show close resemblance, since the later parts of the AE
signal are additionally superimposed by edge reflections incident at the sensor
position. Still, for the present investigations, the similarities of the AE and EME
signals were found to be significant enough to attribute them to the same source
mechanism.

5.5.1.2 Fiber Filament Failure

For the investigation of fiber filament failure, an experimental setup as shown in


Fig. 5.58 was developed. To this end, the fiber filament is first bonded to a PEEK
rod mounted in a testing machine and then moved into a drop of two component
epoxy adhesive. The adhesive is cured in situ, and after sufficient curing time, the
fiber filament is loaded to failure by a high-precision universal testing machine
using 20 μm/min test speed. The drop of adhesive rests on a replaceable PMMA
plate, which is attached to an aluminum frame by two PA6 screws. Within the

a b
Force

PEEK rod
z adhesive
y adhesive
x

PMMA plate

PMMA
screw EME sensor fiber

AE sensor

Fig. 5.58 Schematic of measurement setup (a) and magnified section close to fiber filament (b)
434 5 Electromagnetic Emission

Table 5.3 EME and AE acquisition settings used for single fiber filament tensile tests
Acquisition settings Value EME Value AE
Preamplification 40 dB 40 dB
Threshold 30 dB 32 dB
Triggering Individual channel Individual channel
Acquisition rate 40 MS/s 40 MS/s
Bandpass range 1 kHz to 3 MHz 1 kHz to 3 MHz
Mounting system Floating arm Screw thread
Sensor type Open wire loop KRN-BB PC
Number of sensors 1 1

aluminum frame, a KRNBB-PC type AE sensor is mounted to detect the generated


AE signal. The acoustic propagation path thus comprises the adhesive drop and the
PMMA plate along the thickness direction, since the tip of the AE sensor is
positioned directly below the fiber position. As EME sensor system, an open wire
configuration is used as schematically seen in Fig. 5.58b.
Acquisition settings for AE and EME are reported in Table 5.3. The advantage of
this sensor system is the flat transfer function and the ease of handling to adjust the
sensor position relative to the fiber position. Moreover, due to the small dimensions
of the sensor system, a very short source-sensor distance can be realized. Since this
was found to be a major parameter to improve the detectable signal energy, it is a
key to detect potentially weak EME signals of carbon fiber fracture. All measure-
ments are carried out in an aluminum shielding chamber of 240 mm  90 mm 
240 mm (length  width  height) and 2.5 mm wall thickness.
Some exemplary EME signals for testing of Sigrafil C030 type carbon fiber
filaments with a mean diameter of (6.5  1.2) μm are shown in Fig. 5.59. For single
fiber filament tensile tests, failure typically occurs after exceeding the ultimate
strength of the filament, so only one EME signal can be detected. It was found that
the EME signals of carbon fiber breakage are fundamentally different to the EME
signals found for polymers. Two representative EME signals obtained from a
tensile test of one fiber are plotted in Fig. 5.59a, b. As first noticeable difference,
the polarity of the EME signals of Fig. 5.59 is found to be different for both cases,
while the AE signals have the same polarity. Since no unique polarity was observed
in the present experimental setup, this indicates that the assumed charge disbalance
does not occur in a preferential orientation.
The unfiltered EME signal has strong similarity to the system response of a
sharp step function; thus, these signals are dominated by the 1 kHz high-pass filter
used in the acquisition card. On the high- frequency side, the signal bandwidth is
limited by the 3 MHz low-pass filter. Therefore, it is feasible to conclude that the
rise-time of the signal is less or equal to the cutoff frequency and thus is in the
order of 333 ns. Compared to the calculated crack duration of 1.2 ns for the free
fiber filament in [65], this result is still plausible, but without faster digitization
rates than 40 MS/s and an increased cutoff frequency no direct measurement can
be made.
5.5 Application to Composites 435

a
0.4

0.2
voltage [V]

0.0

–0.2

–0.4

0.10 10
0.08
0.06
5

voltage [mV]
0.04
voltage [V]

0.02
0.00 0
–0.02
–0.04
unfiltered –5
–0.06
–0.08 filtered by 20 kHz Butterworth high-pass 7th order
–0.10 –10
0 200 400 600
time [μs]

b
0.4

0.2
voltage [V]

0.0

–0.2

–0.4

0.10 filtered by 20 kHz Butterworth high-pass 7th order 10


0.08
0.06 unfiltered
5
voltage [mV]

0.04
voltage [V]

0.02
0.00 0
–0.02
–0.04
–5
–0.06
–0.08
–0.10 –10
0 200 400 600
time [μs]

Fig. 5.59 Set of representative AE signals (black) and EME signals (red and blue) as acquired
during tensile testing of single carbon fiber filaments
436 5 Electromagnetic Emission

After application of a 20 kHz high-pass filter, a close similarity to the detected


AE signals is observed. This is again attributed to the same origin of the oscillating
part of the EME signal and the source of the acoustic emission.

5.5.1.3 Composites

Beyond the detection of EME signals in pure materials, the use of these signals for
analysis of composite failure is also a promising approach. As presented in several
publications before [9, 13–15], fiber reinforced composites are known to generate
electromagnetic emission signals. The present section demonstrates the detection of
such EME signals in configurations as typically used to measure the mode I and
mode II fracture toughness values in double-cantilever-beam (DCB) configuration
and end-notched flexure (ENF) configuration.
As pointed out in Sect. 5.3, the presence of conductive materials in the proximity
of the EME source may lead to adverse effects with respect to the alignment of the
electric field lines. Due to these restrictions, all specimen fixtures are completely
machined from nonconductive materials, and the full setup is completely included
in a shielding chamber. For the present experimental setups, the only metallic
components are the EME sensor itself, the AE sensor casing, the attached cables,
and the shielding chamber.
The first configuration is used to measure the interlaminar crack growth in mode
I loading as seen in Fig. 5.60. The bulk of the fixture is machined from PMMA
parts, which are connected by epoxy adhesive and PA6 screws. The load is
introduced by a pultruded GFRP rod, and the precracked specimen is connected
by PVC bolts to the load blocks fabricated from PMMA. The load blocks in turn are
adhesively bonded to the test specimen. In addition, a wire is glued to the surface of
the specimen to act as EME sensor, and a KRNBB-PC AE sensor is mounted on the
specimen using a clamp system. All acquisition settings for AE and EME are listed
in Table 5.4. The cross-head displacement was chosen as 2 mm/min in accordance
with the speed of typical test standards. Since the measurement of the mode I
fracture toughness value was not envisaged in this test scenario, no optical tracking
of the crack tip or localization of the AE sources by means of a second sensor
(cf. Sect. 4.7.4.1) was carried out. All specimens used are fabricated as unidirec-
tional Sigrafil CE1250-230-39 prepreg laminate cured according to the material
supplier’s specification. All specimens were cut to dimensions of 155 mm 
25 mm  3 mm (length  width  thickness) with an embedded ETFE foil of
60 mm length acting as crack initiation side.
During the test, numerous AE and EME signals can be detected, thus requiring a
different approach than for the single events discussed and presented in the
previous two sections. Among the multitude of EME signals, four characteristic
groups can be identified, which are shown in Fig. 5.61 as unfiltered (red) and
filtered signals (blue). Since these signal types occurred in all tests on fiber
reinforced composites, they may already indicate some sort of distinctly different
EME source groups.
5.5 Application to Composites 437

Force

GFRP rod

z
x

y
PMMA fixture

A E sensor

EME sensor

precrack

PVC bolts
PA6 screws

Fig. 5.60 Schematic of measurement setup showing details of specimen fixture for DCB test and
AE and EME sensor arrangement

Table 5.4 EME and AE acquisition settings used for mode I (DCB) testing
Acquisition settings Value EME Value AE
Preamplification 40 dB 20 dBAE
Threshold 28 dB 40 dBAE
Triggering Individual channel Individual channel
Acquisition rate 5 MS/s 10 MS/s
Bandpass range 1 kHz to 3 MHz 1 kHz to 3 MHz
Mounting system Epoxy PMMA nut and clamp
Sensor type Open wire KRN-BB PC
Number of sensors 1 1

The first signal type shown in Fig. 5.61a exhibits a very quick rise of the
unfiltered EME signal being superimposed by the 1 kHz response of the acquisition
system filter. The corresponding filtered signals reveal some lower-frequency
oscillations. However, the fast rise in the beginning of the signal indicates a fast
crack procedure with similar or faster rise-times than those observed for fiber
failure in the single filament tensile tests.
Signals of similar characteristic, but lower rise-time, were found as seen in
Fig. 5.61b. Their rise-time was in the range between 10 μs and 200 μs indicating
either growth of rather large cracks or growth of smaller cracks with low propaga-
tion velocity.
438 5 Electromagnetic Emission

As distinctly different type of EME signals, the oscillating type seen in


Fig. 5.61c was recorded. For the unfiltered case, these EME signals do not show
a signal, which is dominated by the 1 kHz high-pass filter. Instead their bandwidth
seems to be systematically above 1 kHz yielding almost similar shape after appli-
cation of the 20 kHz high-pass. These signals could arise from a variety of
mechanisms. As first possibility, a sequence of fracture events could cause a
a
75
50
voltage [mV]

25
0
–25
–50 unfiltered
–75

20
voltage [mV]

–20

–40
filtered by 20 kHz Butterworth high-pass 7th order
0 100 200 300 400 500 600 700
time [μs]

b
15
10
voltage [mV]

5
0
-5
–10
–15 unfiltered
–20
5.0

2.5
voltage [mV]

0.0

–2.5
filtered by 20 kHz Butterworth high-pass 7th order
–5.0
0 100 200 300 400 500 600 700
time [μs]

Fig. 5.61 Set of typical EME signals recorded during DCB test in unfiltered (red) and filtered
state (blue)
5.5 Application to Composites 439

c
15
10
voltage [mV]

5
0
–5
–10 unfiltered
–15
15
10
voltage [mV]

5
0
–5
–10
filtered by 20 kHz Butterworth high-pass 7th order
–15
0 100 200 300 400 500 600 700
time [μs]

d
50

25
voltage [mV]

–25
unfiltered
–50

20
voltage [mV]

–20
filtered by 20 kHz Butterworth high-pass 7th order
0 100 200 300 400 500 600 700
time [μs]
Fig. 5.61 (continued)

periodic appearance of an EME signal like this. Second, the EME signal could be
the result of a wave motion of a previously charged surface. Such could happen
subsequently to a larger fracture event, when a substantial fracture surface is still
charged and acoustic waves are passing these areas. Finally, these EME signals
could also result from cross talk between AE and EME sensors. The latter cases can
440 5 Electromagnetic Emission

be excluded in scenarios where no AE sensors are used or no AE signal is recorded


synchronous in time.
The last type of EME signal found to be characteristic in failure of fiber
reinforced composites is shown in Fig. 5.61d. This signal is an obvious example
of a cascade of fracture events recorded in one signal. At around 50 μs time, a first
small EME signature is seen. After approximately 400 μs a second, larger EME
signature is seen, which likely constitutes a new fracture event. Although this effect
is obvious from Fig. 5.61d, there are many EME signals with merged signatures,
which cannot be clearly separated. Since the trigger mode for the present setup
operates on the unfiltered signals, this is not surprising, given the rising slope of the
unfiltered signal of Fig. 5.61d. Since this does not fall below the threshold level, no
termination criterion is reached to stop the acquisition of the first event. If only
filtered signals are used, this might limit the number of merged EME signals, but
would also limit the approaches to estimate the rise-time of the underlying source.
Beyond the many AE and EME signals recorded after crack initiation in DCB
tests, some specific signal pairs are shown exemplary in Fig. 5.62. These are the AE
and EME signals corresponding to the crack initiation at the ETFE insert in
Fig. 5.62a and the initiation of interlaminar crack growth starting from a precrack
in Fig. 5.62b. For comparison to the AE signal, only the filtered EME signals are
shown. As for the previous investigations on polymer fracture and fiber failure, the
filtered EME signals show close similarity to the AE signals especially in the first
100 μs. Also, the EME signals show a systematic onset which precedes the AE
signals by 50–100 μs. This is due to the difference in propagation velocities of the
acoustic waves and the electromagnetic waves. Despite of the close amplitude
values of the AE signals, the EME signals of interlaminar crack growth starting
at the precrack were generally found to be of higher amplitude. This could indicate
a systematic difference in the nature of the EME source induced by the presence of
the ETFE insert. Alternatively, this could also be due to a systematic difference in
the interlaminar crack growth, which is known to behave differently when starting
from precracks or inserts.
A quantification of the signal amplitudes and their absolute energy is shown in
Fig. 5.63 for the detected AE and the detected EME signals during crack propaga-
tion. Since both sensor systems use different amplitude threshold levels, a distinct
difference on the low end of the distribution is found. However, for the strongest
signals, a difference of 26 dB is observed for the corresponding unfiltered EME and
AE signals. Hence, it is likely that the majority of the EME signals corresponding to
the weaker AE signals are not detected with the present system, since they likely
fall below the detection threshold. Given the presently used measurement chain of a
J-FET preamplifier mounted close to the EME sensor and the subsequent UBBV-
NF-35 preamplifier, it is not possible to increase the preamplification factor without
saturating stronger signals. Thus, refinement of the signal acquisition electronics is
required to account for the broad dynamic range of signal amplitudes seen in fiber
reinforced composites.
As second configuration used to demonstrate the use of EME measurements
for composite materials, the mode II test shown in Fig. 5.64 was selected. Like the
5.5 Application to Composites 441

a
0.3
0.2
0.1
voltage [V]

0.0
–0.1
–0.2
–0.3

20
voltage [mV]

–20

–40
filtered by 20 kHz Butterworth high-pass 7th order
0 100 200 300 400 500 600 700 800 900 1000
time [μs]

b
0.3
0.2
0.1
voltage [V]

0.0
–0.1
–0.2
–0.3

80
60
voltage [mV]

40
20
0
–20
–40
–60
–80 filtered by 20 kHz Butterworth high-pass 7th order
0 100 200 300 400 500 600 700 800 900 1000
time [μs]

Fig. 5.62 Two exemplary pairs of AE signals (black) and filtered EME signals (blue) detected at
crack initiation starting at ETFE insert (a) and starting from precrack (b)

previous experimental setups, all components are fabricated from nonconductive


materials. Namely, the bulk fixtures are made from PMMA with supports out of
PVC and a pultruded GFRP rod for load transfer to the outside of the shielding
chamber. Same as above, the specimens used in the tests are fabricated as unidi-
rectional Sigrafil CE1250-230-39 prepreg laminate cured according to the material
442 5 Electromagnetic Emission

26 dB

1000000
AE threshold
100000
EME threshold
absolute energy [aJ]

10000

1000

100

10

0.1
AE signals
0.01 EME signals

1E-3
20 30 40 50 60 70 80 90
amplitude [dB]
Fig. 5.63 Comparison of AE signals energy and EME signals energy as function of amplitudes

Fig. 5.64 Schematic of


measurement setup showing
details of specimen fixture Force
for ENF test and AE and
EME sensor arrangement GFRP rod

z
x

y
PMMA fixture

EME sensors

AE sensor

PVC supports
precrack
5.5 Application to Composites 443

Table 5.5 EME and AE acquisition settings used for mode II (ENF) testing
Acquisition settings Value EME Value AE
Preamplification 40 dB 20 dBAE
Threshold 27 dB 40 dBAE
Triggering Individual channel Individual channel
Acquisition rate 2 MS/s 10 MS/s
Bandpass range 1 kHz to 3 MHz 1 kHz to 3 MHz
Mounting system Duct tape PMMA nut and clamp
Sensor type Open wire KRN-BB PC
Number of sensors 2 1

supplier’s specification. Specimens were cut to dimensions 180 mm  25 mm 


3.5 mm (length  width  thickness) with an embedded ETFE foil of 45 mm length
acting as crack initiation side.
The occurrence of acoustic emission during the test is monitored by a KRNBB-
PC sensor attached to the specimen by means of a clamp system at the position seen
in Fig. 5.64. Electromagnetic emission is monitored by two EME sensors realized
by open-ended wires arranged in an array seen in Fig. 5.64 and more detailed in
Fig. 5.66b in the xz-cross-section. The acquisition settings were chosen as reported
in Table 5.5.
All tests are carried out in displacement-controlled mode using a universal
testing machine with a 5 kN load cell. Specimens are first loaded to generate a
precrack in the material starting from the ETFE foil using a cross-head velocity of
2 mm/min. Subsequently, specimens are shifted within the three-point bending
setup to move the position of the crack tip to the center of the EME sensor array.
Then the specimen is loaded again by a cross-head velocity of 2 mm/min until crack
propagation is observed.
Two exemplary pairs of AE and EME signals are shown in Fig. 5.65 as recorded
during the crack initiation starting at the ETFE foil and as starting from the
precrack. For this comparison only the signal of one of the two EME sensors is
shown and compared to the AE sensor signal after filter application of a seventh
order Butterworth high-pass of 20 kHz. As for the previous findings, close similar-
ity of the EME signal appearance to those of the AE signals is found. Similar to the
EME signals of the mode I investigation, there is a systematic difference in
the measured EME amplitude of both cases. Here the EME signals starting from
the precrack appear to be more intense than the first crack initiation starting at the
embedded ETFE foil. Therefore, the specific load condition (mode I or mode II)
does not seem to hinder the detectability or occurrence of noticeable EME signals.
Given the 90 rotation in the present case when compared to the DCB arrangement
of Fig. 5.60, it even seems likely to be able to detect EME signals in various
orientations. For the ENF experiments, AE and EME signals were even found to be
of similar amplitudes, so almost no difference in signal amplitude was measured.
This is likely due to the close distance of the EME sensors relative to the position of
444 5 Electromagnetic Emission

a
6

4
voltage [V]

–2

–4
2

1
voltage [V]

–1

–2 filtered by 20 kHz Butterworth high-pass 7th order


0 100 200 300 400 500 600 700 800 900 1000
time [μs]

b
6

4
voltage [V]

–2

–4
2

1
voltage [V]

–1

–2 filtered by 20 kHz Butterworth high-pass 7th order


0 100 200 300 400 500 600 700 800 900 1000
time [μs]

Fig. 5.65 Two exemplary pairs of AE signals (black) and filtered EME signals (blue) detected at
crack initiation starting at ETFE insert (a) and starting from precrack (b)

the EME source. This was confirmed by the lack of detection of EME signatures,
when the crack tip has propagated sufficiently far away. Thus no EME signals were
detected by the present configuration after the crack tip has propagated beyond
20 mm distance to the sensor system.
5.5 Application to Composites 445

a b
0.06
sensor 1
0.05 10 mm specimen
sensor 2
0.04 sensor 2

0.03
voltage [V]

0.02

0.01 sensor 1 z

0.00
ETFE foil
–0.01 crack growth x

–0.02

0 100 200 300 400 500 600 700 800 900 1000
time [μs]

Fig. 5.66 Pairs of EME signals detected at sensor 1 and sensor 2 at crack initiation (a) and
respective schematic of sensor arrangement (b)

Given the arrangement of the two EME sensors, signal pairs were detected in
most of the cases. One example of such an EME signal pair is seen in Fig. 5.66a.
The specific sensor array used in this experiment is seen in Fig. 5.66b. For this
experiment, two of the wires were chosen at ground potential, while one wire on top
and bottom was selected as sensing system. Due to the intersected arrangement, this
allows to analyze the nature of an EME source in various means. Due to the position
of the ETFE foil at the center of the laminate, the interlaminar crack is initiated at
this ply level and thus propagates at the center between the two EME sensors. As
seen from the EME signals detected at the top and bottom of the plate, the two
signals are detected with opposite polarity. This indicates a directed electric field
orientation as expected from a crack surface charged with opposing polarity on each
side. Since the crack propagates toward the sensor 1, this likely explains the
stronger signal amplitude when compared to the sensor 2. Moreover, this finding
underlines the possibility to use the characteristic radiation pattern of the EME
source to perform orientation measurements of the fracture surface.

5.5.2 Measurement of Fracture Surface Orientation

To demonstrate the use of EME measurements for measurement of the fracture


surface orientation, other test configurations than DCB and ENF test are required,
since the fracture surface is at an approximately constant angle in these setups.
Thus, cross-ply specimens with [0/90/90/0/90]sym were fabricated from the
unidirectional Sigrafil CE1250-230-39 prepreg laminate cured according to the
material supplier’s specification. All specimens were cut to dimensions of
95 mm  15 mm  2.2 mm (length  width  thickness) according to the recom-
mendations of DIN-EN-ISO 14125. As test configuration a three-point bending
setup as seen in Fig. 5.67 was chosen. This ensures predominant failure at the
position of the load nose, i.e., at the middle of the specimen.
446 5 Electromagnetic Emission

Fig. 5.67 Schematic of


measurement setup showing
details of specimen fixture Force
for three-point bending test
and AE and EME sensor GFRP rod
arrangement
z
x

y
PMMA fixture
AE sensor

PVC support

EME sensor 1

EME sensor 2

a b sensor 2
PVC load nose
U2
sensor 2 fracture surface

3 mm θ
sensor 1
sensor 1
specimen z U1

8 mm
x

Fig. 5.68 Arrangement of EME sensors used in three-point bending test (a) and concept to derive
orientation of fracture surface by measurement with two sensors (b)

In analogy to the previous setups, the test was carried out inside a shielding
chamber using only nonconductive materials for the specimen fixtures
(cf. Fig. 5.67). For all tests a universal test machine with a slow cross-head velocity
of 0.02 mm/s was applied to allow good separation of single failure events. The
occurrence of acoustic emission during the test is monitored by a KRNBB-PC
sensor attached to the specimen by means of a clamp system at the position seen in
Fig. 5.67. The concept for the arrangement of EME sensors for the measurement is
shown in the xz-cross-section of Fig. 5.68a with acquisition parameters as reported
in Table 5.6. Similar to the principle of the ENF measurements above, four
5.5 Application to Composites 447

Table 5.6 EME and AE acquisition settings used for flexural testing
Acquisition settings Value EME Value AE
Preamplification 54 dB 20 dBAE
Threshold 30 dB 40 dBAE
Triggering On acoustic emission channel
Acquisition rate 10 MS/s 10 MS/s
Bandpass range 1 kHz to 3 MHz 1 kHz to 3 MHz
Mounting system Floating arm PMMA nut and clamp
Sensor type Open wire KRN-BB PC
Number of sensors 2 1

open-ended wires are used as EME sensors. Two wires are grounded as marked in
Fig. 5.68a. The wire acting as sensor 1 and the corresponding ground are glued to
the bottom of the specimen. The wire of sensor 2 is embedded in an open hole of the
PVC load nose, and the corresponding ground is floating sufficiently low below
the bottom of the specimen to avoid collision during the test. Due to this alignment,
this allows to evaluate two components of the electric field at approximately
orthogonal angle.
The concept used to determine the inclination angle θ from the measurement is
shown in Fig. 5.68b. Given the principle of operation of the EME source in
Sect. 5.2, the two EME sensors should be able to detect orthogonal components
of the electric field. As seen from the measurement using the test source in
Sect. 5.2.2, the orientation of the source is directly obtained from a sinusoidal
relationship of the two signal amplitudes. The inclination angle θ of Fig. 5.68b is
obtained using the EME sensor voltages U1 and U2 of the oscillating signal part as:
 
1 U1
θ ¼ tan ð5:37Þ
U2

The use of the oscillating part is recommended, since the exact amplitude of the
low-frequency components are typically difficult to obtain as quantitative values
due to saturation of system components as well as due to the 1 kHz high-pass
characteristic.
In the following, failure of one exemplary measurement is evaluated using this
concept and is discussed relative to microscopy observations after failure based on
the considerations made in [66]. Typically, the cross-ply specimens investigated
herein fail in compressive mode at the position of the load nose with subsequent
inter-ply delamination and failure of multiple plies. Due to the failure within the
0 plies, the orientation of inter-ply delamination and the orientation of failure
within the 90 plies, fracture surfaces of macroscopically visible extent are gener-
ated with distinctly different orientation.
During the first significant load drop of one specimen, a sequence of two strong
EME signatures was detected. The first event is shown in Fig. 5.69a as unfiltered
signal and in Fig. 5.69b as signal filtered by a 20 kHz Butterworth high-pass of
448 5 Electromagnetic Emission

a
2.0
sensor 1
1.5
voltage [V]

1.0

0.5

0.0

–0.5
2.0

1.5 sensor 2
voltage [V]

1.0

0.5

0.0

–0.5
0 100 200 300 400 500 600 700 800 900 1000
time [μs]

b
1.0
sensor 1
0.5
voltage [V]

0.0

–0.5

–1.0
1.0

0.5 sensor 2
voltage [V]

0.0

–0.5

–1.0
0 100 200 300 400 500 600 700 800 900 1000
time [μs]

Fig. 5.69 First EME event detected during substantial load drop recorded by two EME sensors in
unfiltered view (a) and as filtered signals (b)

seventh order. Since EME signals of both sensors are shown in identical scales, it is
obvious that there is almost no difference in signal amplitude seen before and after
filtering, when comparing signals of both sensors. An application of (5.37) to the
measured EME signal amplitudes thus results in an inclination angle of θ ¼ 44:5 .
This corresponds well to the orientation of the failure in the 0 ply seen in Fig. 5.71,
when taking into account the superimposed deflection and rotation during loading.
5.5 Application to Composites 449

a
2.5
2.0 sensor 1
1.5
voltage [V]

1.0
0.5
0.0
–0.5
–1.0
–1.5
2.5
2.0
1.5 sensor 2
voltage [V]

1.0
0.5
0.0
–0.5
–1.0
–1.5
0 100 200 300 400 500 600 700 800 900 1000
time [μs]

b
0.2 sensor 1
0.1
voltage [V]

0.0

–0.1

–0.2

0.2
sensor 2
0.1
voltage [V]

0.0

–0.1

–0.2

0 100 200 300 400 500 600 700 800 900 1000
time [μs]

Fig. 5.70 Second EME event detected during substantial load drop recorded by two EME sensors
in unfiltered view (a) and as filtered signals (b)

Only few milliseconds later, a second strong event was observed as seen in
Fig. 5.70a as unfiltered and in Fig. 5.70b as filtered signal in analogy to Fig. 5.69.
While the unfiltered signals reveal almost no difference in signal amplitude, the
oscillating part of Fig. 5.70b exhibits a distinct difference in signal intensity.
Applying equation (5.37) to these amplitudes, a fracture angle of θ ¼ 11:0 is
450 5 Electromagnetic Emission

a b
first event PVC load nose

second event θ

z
specimen
z delamination
x x

Fig. 5.71 Microscopy image of fracture mode as corresponding to the EME events shown in
Figs. 5.69 and 5.70 and corresponding scheme of situation under load

obtained. This narrow angle can directly be related to the orientation of the inter-ply
delamination of Fig. 5.71 when taking into account the curvature of the specimen
under load as seen from the scheme on the right side of Fig. 5.71.

5.5.3 Detectability of Failure Mechanisms

As has been demonstrated in the previous sections and throughout various literature
sources [9, 13–15], EME is generally detectable for the failure mechanisms encoun-
tered in fiber reinforced composites. Since measurement of EME signals of single
filament failure was found to be possible, it may readily be assumed that the method
is generally suitable to aid in interpretation of composite failure.
However, given the novelty of the method and the lack of established technical
standards, it is difficult to assess the detectability in a similar fashion as for digital
image correlation in Sect. 3.3.3 or for acoustic emission in Sect. 4.2.4. Thus, more
intense research and standardization efforts will be required to establish the prob-
ability of EME detection in a similar level of detail.
Instead, this section will provide a guideline as first estimate for a couple of
experimental factors to consider when conducting EME measurements. Similar to
the other methods investigated, the first focus is given to the absolute limit of the
measurement system.

5.5.3.1 Absolute System Limits

Based on the experimental investigations presented in this chapter and the efforts to
establish a thorough implementation of the measurement chain in Sect. 5.4, it is
possible to allow for some first conclusions regarding the absolute system limits.
Other than for AE sensors, there is no technically relevant lower detection limit of
5.5 Application to Composites 451

the EME sensor system itself. Since this EME measurement is in principle only a
measurement of an AC voltage, the sensitivity of this measurement can practically
be realized for all technically relevant orders of magnitude of voltages. Instead, the
dominating factor of influence with respect to the detection limits of the measure-
ment system is currently owed to shielding aspects, since the electromagnetic noise
detected by the sensor system constitutes the real practical limit of applicability of
the measurement method. Because this will be discussed in the next step explicitly,
this is neglected for the first discussion. Apart from the EME sensor itself, the only
remaining part in the measurement chain are the system electronics. Since EME
signals were observed to span similar orders of magnitude in induced signal voltage
as for AE sensor signals, this constitutes the primary challenge of the signal
acquisition stage. For an AE sensor, the generated voltage may be assumed to be
proportional to the surface motion present at the sensor’s contact area to the solid
[67–69]. For an EME sensor, the generated signal voltage will depend primarily on
the distance and orientation relative to the source. Thus the generated sensor signal
voltage may not be directly related to the charges on the fracture surface, but
possibly to the incident electric field strength.
Consequently, the requirement to the acquisition chain is to assure a sufficient
digitization, without saturating the preamplifier or the digitization equipment.
Given the assumption of negligible saturation effects, this causes the lower detec-
tion limit to be dominated by the electronic noise floor of the system electronics,
which is below 1 mV for a typical commercial system operating in the frequency
range from 1 kHz to 1 MHz.

5.5.3.2 Signal-to-Noise Ratio

As indicated in the previous section, the relevant detection limit for EME measure-
ments was found to be given by the suppression of electromagnetic noise. For a
specific sensor sensitivity and generated electric field strength of the EME source,
the ability to detect the corresponding EME signal will substantially depend on the
background noise level. To this end, the noise floor of the system electronics was
already introduced in the previous section. However, there are numerous other
sources for electromagnetic noise included within the measurement chain. As a first
step to gain a high signal-to-noise ratio, it is thus required to avoid detection of
electromagnetic noise.
The EME sensor itself is susceptible to all incident electromagnetic radiation,
not only from the EME source. Therefore, the use of a shielding chamber as
presented in Sect. 5.4 is a recommended step to resolve the weak signatures of
EME sources as originating from material failure. The level of shielding effective-
ness will determine the overall quality of the measurement, but may certainly not be
reduced beyond a certain level. As additional sources to detect electromagnetic
noise, all cable connections and preamplifier circuits may be considered. In order to
reduce the incident electromagnetic noise to a minimum, one step consists of
mounting a low-noise J-FET preamplifier directly to the EME sensor. In the spirit
452 5 Electromagnetic Emission

of [58–60], this was implemented for the presented experiments as seen in Fig. 5.42
and was found to increase the signal-to-noise ratio by 18 %. While this approach
helps to improve the signal quality, it is further required to keep a high shielding
effectiveness for all further cables and remaining electrical components (filters,
preamplifiers). In practice it was found to be suitable to include the cables to the
main preamplifier stage within the shielding chamber. This effectively avoids
detection of external electromagnetic noise sources and keeps the signal-to-noise
ratio at a high level.
As a second step to increase the overall signal-to-noise ratio, it is generally
possible to increase the sensitivity of the sensor system. However, for the present
EME sensor system discussed in Sect. 5.4, there was no significant difference
observed as function of sensor geometry. Thus, the simple detection by using a
metallic conductor seems to be limited in this aspect. Therefore, other sensor
concepts, such as Hall effect sensors, magnetometers, or superconducting quantum
interference devices (SQUIDs) could be explored as alternatives to detect electro-
magnetic waves.
Another obvious optimization routine consists in considering the source-sensor
distance and the radiation pattern of the expected source types. Based on the decay
of electric field strength as function of distance to the charged fracture surface, it is
advisable to position the EME sensor as close as possible to the fracture surface to
obtain high EME signal amplitudes. Similarly, the expected EME source radiation
patterns may allow selecting preferred EME sensor positions to optimize the
incident electric field strength.

5.5.3.3 Acquisition Mode

Another factor of practical relevance to the detectability of EME signals in exper-


iments deals with the chosen acquisition mode. Like AE signals, EME signals are of
transient nature. Thus the acquisition mode of system electronics may either
operate in continuous storage mode causing a large data amount for reasonable
acquisition rates or may operate in triggered mode to store only parts of the stream
after exceeding a trigger event.
The continuous storage mode comes with the distinct advantage of extensive
capabilities for offline analysis, but requires respective hardware and software
systems. For the triggered mode, other challenges arise, which shall be discussed
briefly.
For typical commercial transient recorders, the triggered acquisition mode is
based on exceedance of a certain voltage threshold. Depending on system quality,
further parameters are used to determine if the trigger event is a valid transient
signal to be stored or not. For acquisition systems optimized for AE signal detec-
tion, the shape of EME signals comprises a first challenge to overcome by modified
hardware and software settings. Since this was hard to be implemented in practice,
another approach was followed throughout some of the experiments presented in
this chapter. Typically the trigger condition was chosen to simultaneously record
References 453

the EME signals, once an AE signal is detected. This provided reliable detection
conditions, but also indicated an important factor only partially solved so far.
As previously discussed in literature [1, 3, 5, 10, 14], there is a general discrep-
ancy of the number of AE signals versus the number of EME signals. In general,
less EME signals than AE signals are acquired during an experiment. While it might
be intuitive to attribute this discrepancy to the lack of sensitivity of EME sensors, it
is likely that this is not the only reason. Apart from the aspects of the source
radiation pattern, it was demonstrated that EME measurements are sensitive enough
to detect even the weakest failure modes (i.e., fiber breakage signals) also detect-
able by AE measurements. Instead, one essential difference could be given by the
origin of the source mechanism for EME signals and AE signals. Thus it can be
speculated, if every AE source is also an EME source. Since the generation of an
EME signal requires the generation of surface charges due to breakage of bonds,
this might not be the case for some AE sources. For stick-slip friction or internal
friction, this might apply. Since these are most relevant AE sources, the absence of
EME for such sources could be used to securely identify this AE source type.
Another possible scenario could be the formation of cracks not exhibiting a charge
imbalance as required for the proposed EME source model description in Sect. 5.2.
In this context it cannot be excluded that the EME phenomenon could potentially be
suppressed in some exotic material configurations. If crack formation happens
without charge imbalance in these materials or the process is of stochastically
nature in its strength, the detection of EME signals might not be a reliable process
to detect crack formation in such materials. However, the latter seems not applica-
ble for the experiments and materials presented herein, since these showed out-
standing repeatability in their EME occurrence.

References

1. Yamada, I., Masuda, K., Mizutani, H.: Electromagnetic and acoustic emission associated with
rock fracture. Phys. Earth Planet. Inter. 57, 157–168 (1989)
2. Frid, V., Rabinovitch, A., Bahat, D.: Fracture induced electromagnetic radiation. J. Phys.
D. Appl. Phys. 36, 1620–1628 (2003)
3. Sedlak, P., Sikula, J., Lokajicek, T., Mori, Y.: Acoustic and electromagnetic emission as a tool
for crack localization. Meas. Sci. Technol. 19, 045701 (2008)
4. Rabinovitch, A., Frid, V., Bahat, D.: Surface oscillations—a possible source of fracture
induced electromagnetic radiation. Tectonophysics 431, 15–21 (2007)
5. Lacidogna, G., Carpinteri, A., Manuello, A., Durin, G., Schiavi, A., Niccolini, G., Agosto, A.:
Acoustic and electromagnetic emissions as precursor phenomena in failure processes. Strain
47, 144–152 (2011)
6. Miroshnichenko, M., Kuksenko, V.: Study of electromagnetic pulses in initiation of cracks in
solid dielectrics. Sov. Phys. Solid State 22, 1531–1533 (1980)
7. O’Keefe, S.G., Thiel, D.V.: A mechanism for the production of electromagnetic radiation
during fracture of brittle materials. Phys. Earth Planet. Inter. 89, 127–135 (1995)
8. Rabinovitch, A., Frid, V., Bahat, D., Goldbaum, J.: Fracture area calculation from electro-
magnetic radiation and its use in chalk failure analysis. Int. J. Rock Mech. Min. Sci. 37,
1149–1154 (2000)
454 5 Electromagnetic Emission

9. Koktavy, P.: Experimental study of electromagnetic emission signals generated by crack


generation in composite materials. Meas. Sci. Technol. 20, 015704 (2008)
10. Koktavy, P., Pavelka, J., Sikula, J.: Characterization of acoustic and electromagnetic emission
sources. Meas. Sci. Technol. 15, 973–977 (2004)
11. Sikula, J., Majzner, J., Sedlak, P., Mori, Y.: Electromagnetic and acoustic emission fine
spectra. Adv. Mater. Res. 13–14, 169–174 (2006)
12. Gade, S.O., Weiss, U., Peter, M.A., Sause, M.G.R.: Relation of electromagnetic emission and
crack dynamics in epoxy resin materials. J. Nondestruct. Eval. 33, 711–723 (2014)
13. Sedlak, P., Enoki, M., Ogasawara, T., Sikula, J.: Electromagnetic and acoustic emission in
PEEK/carbon nanotube composites. In: 29th European Conference on Acoustic Emission
Testing, Vienna (2010)
14. Sklarczyk, C., Winkler, S., Thielicke, B.: Die elektrische Emission beim Versagen von
Faserverbundwerkstoffen und ihren Komponenten. Materwiss. Werkstofftech. 27, 559–566
(1996)
15. Aman, S., Aman, A., Morgner, W.: Monitoring of carbon fibre breakage in composites based
on microwave emission. Compos. Sci. Technol. 84, 58–64 (2013)
16. Misra, A., Prasad, R.C., Chauhan, V.S., Srilakshmi, B.: A theoretical model for the electro-
magnetic radiation emission during plastic deformation and crack propagation in metallic
materials. Int. J. Fract. 145, 99–121 (2007)
17. Winkler, S.R.: Field emissions caused by fracture and yielding. Int. J. Fract. 136, 221–235
(2005)
18. Jagasivamani, V., Iyer, K.J.L.: Electromagnetic emission during the fracture of heat-treated
spring steel. Mater. Lett. 6, 418–422 (1988)
19. Misra, A.: A physical model for the stress-induced electromagnetic effect in metals. Appl.
Phys. 16, 195–199 (1978)
20. Misra, A., Ghosh, S.: Electromagnetic radiation characteristics during fatigue crack propaga-
tion and failure. Appl. Phys. 23, 387–390 (1980)
21. Molotskii, M.I.: Dislocation mechanism for the misra effect. Sov. Tech. Phys. Lett. 6, 22–23
(1980)
22. Rabinovitch, A., Bahat, D., Frid, V.: Comparison of electromagnetic radiation and acoustic
emission in granite fracturing. Int. J. Fract. 71, R33–R41 (1995)
23. Rabinovitch, A., Frid, V., Bahat, D.: Parametrization of electromagnetic radiation pulses
obtained by triaxial fracture of granite samples. Philos. Mag. Lett. 77, 289–293 (1998)
24. Rabinovitch, A.: A note on the amplitude-frequency relation of electromagnetic radiation
pulses induced by material failure. Philos. Mag. Lett. 79, 195–200 (1999)
25. Rabinovitch, A., Frid, V., Bahat, D.: Gutenberg-Richter-type relation for laboratory fracture-
induced electromagnetic radiation. Phys. Rev. E. 65, 011401 (2001)
26. Frid, V.: Electromagnetic radiation associated with induced triaxial fracture in granite. Philos.
Mag. Lett. 79, 79–86 (1999)
27. Frid, V.: Electromagnetic radiation method water-infusion control in rockburst-prone strata.
J. Appl. Geophys. 43, 5–13 (2000)
28. Bahat, D., Rabinovitch, A., Frid, V.: Fracture characterization of chalk in uniaxial and triaxial
tests by rock mechanics, fractographic and electromagnetic radiation methods. J. Struct. Geol.
23, 1531–1547 (2001)
29. Bahat, D., Frid, V., Rabinovitch, A., Palchik, V.: Exploration via electromagnetic radiation
and fractographic methods of fracture properties induced by compression in glass-ceramic. Int.
J. Fract. 116, 179–194 (2002)
30. Goldbaum, J., Frid, V., Rabinovitch, A., Bahat, D.: Electromagnetic radiation induced by
percussion drilling. Int. J. Fract. 111, 15–20 (2001)
31. Finkel, V.M., Golovin, Y.I., Sereda, V.E.: Electric effects at fracture of LiF crystals in
connection with the crack control problem. Sov. Phys. Solid State 17, 492–495 (1975)
32. Rabinovitch, A., Bahat, D., Frid, V.: Similarity and dissimilarity of electromagnetic radiation
from carbonate rocks under compression, drilling and blasting. Int. J. Rock Mech. Min. Sci. 39,
125–129 (2002)
References 455

33. Sklarczyk, C., Altpeter, I.: The electric emission from mortar and concrete subjected to
mechanical impact. Scr. Mater. 44, 2537–2541 (2001)
34. Petrenko, V.F.: On the nature of electrical polarization of materials caused by cracks. Appli-
cation to ice electromagnetic emission. Philos. Mag. Part B 67, 301–315 (1993)
35. Ogawa, T., Oike, K., Miura, T.: Electromagnetic radiations from rocks. J. Geophys. Res. 90,
6245 (1985)
36. Rabinovitch, A., Frid, V., Bahat, D., Goldbaum, J.: Decay mechanism of fracture induced
electromagnetic pulses. J. Appl. Phys. 93, 5085 (2003)
37. Hill, T.L.: An Introduction to Statistical Thermodynamics. Dover, New York (1986)
38. Rayleigh, L.: On waves propagated along the plane surface of an elastic solid. Proc. London
Math. Soc. s1–s17, 4–11 (1885)
39. Srivastava, G.P.: The Physics of Phonons. Taylor & Francis, New York (1990)
40. Freund, L.B.: Crack propagation in an elastic solid subjected to general loading—I. Constant
rate of extension. J. Mech. Phys. Solids 20, 129–140 (1972)
41. Freund, L.B.: Crack propagation in an elastic solid subjected to general loading—II.
Non-uniform rate of extension. J. Mech. Phys. Solids 20, 141–152 (1972)
42. Freund, L.B.: Crack propagation in an elastic solid subjected to general loading—III. Stress
wave loading. J. Mech. Phys. Solids. 21, 47–61 (1973)
43. Freund, L.B.: Crack propagation in an elastic solid subjected to general loading—IV.
Obliquely incident stress pulse. J. Mech. Phys. Solids 22, 137–146 (1974)
44. Freund, L.B.: Dynamic Fracture Mechanics. Cambridge University Press, Cambridge (1990)
45. Rose, L.R.F.: Recent theoretical and experimental results on fast brittle fracture. Int. J. Fract.
12, 799–813 (1976)
46. Tromans, D.: Crack propagation in brittle materials: relevance to minerals comminution. Int.
J. Res. Rev. Appl. Sci. 13, 406–427 (2012)
47. Scruby, C.B.: Quantitative acoustic emission techniques. Nondestruct. Test. 8, 141–208 (1985)
48. Wadley, H.N.G., Scruby, C.B.: Elastic wave radiation from cleavage crack extension. Int.
J. Fract. 23, 111–128 (1983)
49. Guozden, T.M., Jagla, E.A.: Supersonic crack propagation in a class of lattice models of mode
III brittle fracture. Phys. Rev. Lett. 95, 224302 (2005)
50. Guo, G., Yang, W., Huang, Y.: Supersonic crack growth in a solid of upturn stress–strain
relation under anti-plane shear. J. Mech. Phys. Solids 51, 1971–1985 (2003)
51. Sanders, W.T.: On the possibility of a supersonic crack in a crystal lattice. Eng. Fract. Mech. 4,
145–153 (1972)
52. Rosakis, A.J.: Cracks faster than the shear wave speed. Science 284, 1337–1340 (1999)
53. Todoroki, A., Tanaka, M., Shimamura, Y.: Measurement of orthotropic electric conductance
of CFRP laminates and analysis of the effect on delamination monitoring with an electric
resistance change method. Compos. Sci. Technol. 62, 619–628 (2002)
54. Girard, C., Joachim, C., Gauthier, S.: The physics of the near-field. Reports Prog. Phys. 63,
893–938 (2000)
55. Mikki, S.M., Antar, Y.M.M.: A theory of antenna electromagnetic near field—Part I. IEEE
Trans. Antennas Propag. 59, 4691–4705 (2011)
56. Schmitt, R.: Electromagnetics Explained: A Handbook for Wireless/RF, EMC, and High-
Speed Electronics. Newnes, New York (2002)
57. Maxwell, J.C.: A dynamical theory of the electromagnetic field. Philos. Trans. R. Soc. London
155, 459–512 (1865)
58. Hamstad, M.A.: Improved signal-to-noise wideband acoustic/ultrasonic contact displacement
sensors for wood and polymers. Wood Fiber Sci. 29, 239–248 (1997)
59. Hamstad, M.A., Fortunko, C.M.: Development of practical wideband high-fidelity acoustic
emission sensors. In: Chase, S.B. (ed.) Nondestructive Evaluation of Aging Infrastructure,
pp. 281–288. International Society for Optics and Photonics, San Diego (1995)
60. Shiwa, M., Inaba, H., Carpenter, S.H., Kishi, T.: Development of high-sensitivity and
low-noise integrated acoustic emission sensor. Mater. Eval. 50, 868–874 (1992)
456 5 Electromagnetic Emission

61. Smuga, J.R.: Conductively filled poly(methyl methacrylate) composites: manufacture


and testing processes for EMI shielding effectiveness. PhD-thesis, Edinburgh Napier
University (2012)
62. Ott, H.: Electromagnetic Compatibility Engineering. Wiley, New York (2009)
63. Paul, C.R.: Introduction to Electromagnetic Compatibility. Wiley Series in Microwave and
Optical Engineering. Wiley, New York (2006)
64. Préault, V., Corcolle, R., Daniel, L., Pichon, L.: Influence of skin effect on the effective
shielding effectiveness of composite materials. J. Appl. Phys. 115, 154904 (2014)
65. Sause, M.G.R., Richler, S.: Finite element modelling of cracks as acoustic emission sources.
J. Nondestruct. Eval. 34, 1–13 (2015)
66. Alaca, B.B., Gade, S.O., Sause, M.G.R.: Untersuchung von elektromagnetischen und
akustischen Emissionen während Bruchvorgängen in faserverstärkten Kunststoffen. In: 20.
Kolloquium Schallemission. pp. 1–8, Garmisch-Partenkirchen, Germany (2015).
67. McLaskey, G.C., Glaser, S.D.: Acoustic emission sensor calibration for absolute source
measurements. J. Nondestruct. Eval. 31, 157–168 (2012)
68. Sause, M.G.R., Hamstad, M.A., Horn, S.: Finite element modeling of conical acoustic emis-
sion sensors and corresponding experiments. Sensors Actuators A Phys. 184, 64–71 (2012)
69. Ono, K., Cho, H., Matsuo, T.: Transfer functions of acoustic emission sensors. J. Acoust.
Emiss. 26, 72–90 (2008)
Chapter 6
Computed Tomography

Many of the early developments of X-ray-based computed tomography (CT) go


back to the pioneering work of Cosslett and Nixon [1–4]. The method of computed
tomography is understood as any computer-aided tomographic process (usually
involving X-rays) to produce 3D representations of the scanned object both exter-
nally and internally. This inspection method has been substantially improved since
its technological beginnings in the 1970s and allows inspecting fiber reinforced
composite materials nowadays with remarkable precision. Some of the key uses for
industrial CT scanning involve flaw detection, failure analysis, metrology, assem-
bly analysis, and reverse engineering applications [5]. X-ray-based computed
tomography is an established technology, which is readily available from commer-
cial companies. Consequently, the focus of this chapter is not to cover all funda-
mentals of the measurement method but its application and limitation in the context
of defect sensing in fiber reinforced materials. For a more detailed review on the
method of computed tomography, the reader is instead referred to some recent
books on this topic [6–8].
To start with the chapter, first an overview on the principle of operation of CT is
provided, and some basic configurations of CT are presented. The subsequent
section has its focus on the present limitations in terms of CT resolution, including
artifacts and typical failure modes as encountered in fiber reinforced materials. One
section is used to review the important concepts involving combinations of
mechanical loading of materials with CT imaging. As further development of
DIC methods, the important aspect of digital volume correlation (DVC) and its
application to fiber reinforced materials are discussed. Finally some applications of
CT to assess the failure progression in composite materials along with possibilities
to use CT information in modeling approaches are presented.

© Springer International Publishing Switzerland 2016 457


M.G.R. Sause, In Situ Monitoring of Fiber-Reinforced Composites,
Springer Series in Materials Science 242, DOI 10.1007/978-3-319-30954-5_6
458 6 Computed Tomography

6.1 Principle of Operation

A conventional X-ray inspection is based on a simple projection of the test object


onto a detector using a point-like X-ray source. This simple geometrical arrange-
ment as seen in Fig. 6.1 causes a projection of the object volume to the detector and
has been termed “shadow microscopy.” The magnification factor of the test object
is primarily determined by the distance between source and object as well as the
distance between the object and the detector.
In order to yield a difference in detected X-ray intensity, two main effects can be
distinguished:
1. The detected intensity may be decreased due to a material with higher X-ray
absorption.
2. The detected intensity may be decreased due to a test object with higher
thickness.
Hence, one projection direction may not allow to distinguish between both
effects, since a reduction in detected X-ray intensity could be due to either effect
or a combination of both.
In contrast to the conventional 2D projection, the process of CT uses projections
of the test object acquired at different angles and combines their information to
yield a volumetric structure of the test object. This is only possible by usage of a
reconstruction algorithm operating on discretized versions of the projection images.
The typical CT configuration basically requires three components, an X-ray
source, a moveable table carrying the test object, and a detector system. Although
there are many types of CT configurations available, only the most common type of
cone-beam CT will be explained in the following, since it was used for the
experimental results shown in this chapter.
For a CT scan, the test object is placed on a rotary table. Similar to the principle
of “shadow microscopy,” in CT the test object can be moved closer to the source to
yield higher magnification or closer to the detector to yield lower magnification.
One very important factor for cone-beam CT is that the part must fit in the field of
view of the detector to allow proper reconstruction (Fig. 6.2). After the initial
alignment, the test object is rotated 360 and 2D X-ray projections are taken at

Fig. 6.1 Arrangement of


source, test object, and
detector in “shadow
microscopy”

light source
object projection
6.1 Principle of Operation 459

Fig. 6.2 Typical cone-beam


arrangement of cone-beam object z
CT θ

x
y
source table detector

distinct angles. For typical CT acquisitions, the number of X-ray images ranges
from several hundreds to several thousands in a single rotation of the test object.
The individual images are then reconstructed to create a 3D volume representation
of the test object. If the test object does not allow a full 360 rotation, special
reconstruction algorithms are used to allow scans based on projected images
acquired at fractions of the full 360 rotation.
In the following, some key aspects of CT are discussed in more detail to allow a
better understanding on how the X-ray source, the detector, and the reconstruction
algorithm influence the imaging process. In the context of this book, the main
application is to visualize the interior volume of fiber reinforced materials.
In order to improve the resolution of X-ray microscopes, one of the key chal-
lenges was to achieve a smaller source size, while keeping or increasing the
brightness of the source to allow for reasonable exposure times. This lead to the
geometrical arrangement of X-ray tubes nowadays referred to as micro-focus or
nano-focus tubes as seen in Fig. 6.3 which constitute the most frequently used X-ray
source.1
As with any conventional X-ray tube, the first step is a generation of a free
electron gas using thermionic emission by a heating voltage UH applied to a coil
located inside a vacuum tube. The free electrons are focused to leave a filament grid
and form an electron beam using a voltage UG. By using an acceleration voltage
Uacc between the emission coil and an anode, the electron beam is further acceler-
ated and directed towards a target material. The target material (typically tungsten
or molybdenum) is hit by the electron beam, and X-rays are released in form of
deceleration radiation and characteristic radiation. To achieve a high resolution, a
small focal spot size and thus a small interaction volume between electron beam
and target material are necessary. In order to minimize the focal spot size, the
electron beam is typically focused using magnetic lens systems minimizing its
cross-section at the position of target impact. In addition, high-resolution tube
systems use target materials, which are comparatively thin and are arranged per-
pendicular to the incident beam (transmission tubes). This is in contrast to the usual

1
In principle, radioisotopes could also be used to generate 2D projections of objects as frequently
done in field applications. However, the handling aspects and the intensity decay of these sources
practically inhibit their use for computed tomography applications.
460 6 Computed Tomography

a b
UH UH
filament grid
UG U UG U
ACC ACC
anode

deflection unit

magnetic lens
X-rays
target
X-rays
target

Fig. 6.3 Configurations of X-ray transmission tubes (a) and tubes using directional design (b)

configuration with oblique incidence of the electron beam onto a thick target
material, a configuration which is known as directional tubes. In any case the
X-rays are released within the target material and radiate outward from their
position. This causes a typical cone-shaped X-ray beam spreading starting at the
source spot.
A superior technological alternative as an X-ray source for CT is synchrotron
radiation. When electrons traveling at ultrarelativistic speeds are forced by a
magnetic field to travel in a curved path, synchrotron radiation is produced. In
theory, the ultrarelativistic speed will change the observed radiation frequency due
to the Doppler effect. The relativistic length contraction changes the radiated
frequencies further, thus effectively yielding electromagnetic radiation in the
X-ray spectrum. The radiation pattern can be distorted from an isotropic dipole
pattern into an extremely forward-pointing cone of radiation depending on the
velocity of the electrons. Synchrotron radiation is the brightest artificial source of
X-rays and may be achieved artificially in synchrotrons or storage rings. The
electromagnetic radiation produced in this way has a characteristic polarization,
and the high intensity photon beam allows fast exposure times for CT scanning. In
addition, the high brilliance of the synchrotron beam generates a small divergence
and enables a small source spot size.
Another important component to achieve high-resolution volumetric represen-
tation of objects is the detector system. For computed tomography this is a device
that captures X-rays and usually converts them to light in the visible spectrum.
Among the most relevant detector concepts used, there are scintillator systems and
flat-panel detectors. In practice this includes line scanners and array scanners to
produce 1D or 2D projections of the test object.
For the scintillator systems, the incident X-rays are first converted by a scintil-
lator crystal into visible light. A photo cathode is then used to convert the visible
light into electrons, which are accelerated and focused onto a fluorescent screen.
The projected image on the fluorescent screen is then detected by a CCD camera
system and constitutes the final 2D digital image. As technological alternative, the
6.2 Detail Visibility 461

digital flat-panel detectors use a thin scintillation foil directly applied on a matrix
array of photodiodes operating in the range of emitted visible light. The detected
photon intensity is then turned into a 2D digital image. Some of the recent
developments consider the overall pixel resolution of the detector arrays, which
has a direct impact on the detail visibility. In addition, digital detectors come in a
variety of pixel pitches which is the space between pixels. Typically a detector with
a smaller pixel pitch relates to longer image acquisition time and higher costs and
requires a higher photon flux to produce an image. A larger pixel pitch requires less
energy to light up the pixels and is cheaper to manufacture but reduces resolution.
As fundamental principle of computed tomography, the 2D projections by itself
are meaningless without further processing by a reconstruction algorithm. The task
of the latter is the generation of 2D slices transverse to the axis of rotation and the
subsequent stacking of these individual slices to produce a 3D volume. The usual
algorithm, which is applied to this end, is the filtered back projection, which is
based on the Radon transformation developed in 1917 by Radon [9]. Since the
ongoing developments in this particular field expand beyond the level of detail
intended for this chapter, the interested reader is referred to some recent literature in
this area for more precise information [6–8].

6.2 Detail Visibility

Within this section the important aspect of detail visibility is reviewed. The term
detail visibility accounts for our perception of the scanned objects. Since this is not
equivalent to the system resolution, it needs some further specification. In most of
the imaging technologies, detail visibility is used for the (human) ability to identify
a certain object as what it is. Thus it can be understood as the necessary resolution
of the image in terms of pixels or voxels sufficient to allow spotting a detail, which
is different to its surrounding. However, in a general situation, an exact definition of
detail visibility is quite challenging, since our perception is substantially influenced
on whether the relevant structure is isolated against a background or whether the
structure is periodic with a certain grid length. In the context of this chapter, the
term will be used in conjunction with the aspect of visualization of defects in fiber
reinforced materials. Some examples of CT images are presented at the end of this
section to provide a subjective feeling of detail visibility as achieved by the state-of-
the art computed tomography devices.
Before, the two main contributions influencing detail visibility are discussed.
First, the overall object resolution as well as induced uncertainties of the experi-
mental configuration of the 2D projections is derived. Second, the aspects of image
artifacts arising from material combinations, reconstruction algorithms, and other
effects are presented.
462 6 Computed Tomography

6.2.1 Object Resolution

Based on the concept of shadow microscopy, the geometrical magnification M is


readily derived by the length of the focus-object-distance (FOD) and the length of
the focus-detector-distance (FDD) as

FDD
M¼ ð6:1Þ
FOD

This simple geometrical relationship is shown in Fig. 6.4a schematically and as


measurement result of a composite material in Fig. 6.5.
In principle, (6.1) would allow arbitrary large magnification factors, given the
size of the detector is scaled accordingly. However, the actual object resolution is
not just determined by the geometrical magnification factor. To assess the final
resolution of the measurement system, there are three contributions from the
experimental equipment to take into account:
e
1. Pixel size P
e
2. Focal spot size F
3. Mechanical stability

a b
FOD
FDD

point-source object detector real source object detector

Fig. 6.4 Definition of geometrical magnification factor (a) and relation between penumbra effect
and focal spot size (b)

M = 2.4 M = 5.5
1 mm 1 mm

Fig. 6.5 Comparison of CT scan slice of same object with different magnification factors
6.2 Detail Visibility 463

The acquired digital images are spatially discretized by the pixel array of
e This constitutes the fundamental length scale of spatial resolution
individual size P.
at the position of the detector. The absolute resolution of the volume produced from
such 2D images is generally referred to as voxel size V˜, which is defined by the
e and the geometric magnification factor M:
pixel size P

e
e¼ P
V ð6:2Þ
M

The overall resolution of the system is further limited by the effect of unsharpness
resulting in blurred images. The geometric unsharpness dg is mostly due to the focal
e The latter causes a penumbra effect as seen in Fig. 6.4b and may be
spot size F.
written in terms of the geometric magnification as

e ð M  1Þ
dg ¼ F ð6:3Þ

Thus, the higher M, the larger the impact on dg as seen by the schematically
comparison in Fig. 6.6.
Finally, the mechanical stability of the configuration may result in unintended
movement of the object and thus may introduce additional movement unsharpness
dmov. The latter is linked to the rotation of the test object and may be induced by
tilting of the object during scanning. However, given the test object is tightly fixed
and has sufficient stiffness, this effect is usually negligible. As another factor of
mechanical movement during image acquisition, temperature gradients may cause
a slow change in the dimensions of the test object or the geometric arrangement as
such. Under usual operating conditions, the CT systems are designed in a way that
the thermal expansion of the system is minimal and therefore the effect is usually
also negligible. For the dimensional change of the test object, the thermal conduc-
tivity and the thermal expansion coefficients are relevant quantities. These strictly
depend on the scanned material and thus may induce according thermal expansion
of the test object. However, within the usual change of temperature, this effect is
negligible for most materials. The sum of all of these movement effects will lead to

a b
FOD FOD

dg dg

F F
P P

high magnification low magnification

Fig. 6.6 Relation between magnification factor and geometrical unsharpness for high (a) and low
(b) magnification factor
464 6 Computed Tomography

an additional unsharpness proportional to M in the acquired image and, hence, will


directly influence the resolution of the volume.
The total unsharpness dtot of the system may then be derived by combination of
the geometric unsharpness, the movement unsharpness, and the pixel size, yielding
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
!2 ffi
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi u
ud 2 e
1 e 2 ¼ t mov þ F ð M  1 Þ e2
dtot ¼ d 2mov þ d2g þ P þV ð6:4Þ
M M M

Thus for a detector pixel size P e  d tot and dmov ! 0, the focal spot size is the
limiting factor for image resolution. In contrast, for pixel size P e d tot and
dmov ! 0, the pixel size will be the limiting factor.
In high-resolution imaging, the ultimate resolution is usually limited by the focal
e For the case of microstructure as found in fiber reinforced materials, an
spot size F.
empirical value of the maximum detail detectability may be approximately 1/3 of
e
the focal spot size F.
As an example of this relationship, a series of cross-sectional images of the same
object are shown as function of the focal spot size F e in Fig. 6.7 for a constant
magnification factor M ¼ 5:5. Clearly, the smallest spot size comes with best detail
visibility. When changing F e to larger sizes, this induces noise in the images,
therefore reducing the possibility to resolve details of the composite microstructure.

6.2.2 Artifacts

In the context of computed tomography, the term artifact is applied to any system-
atic discrepancy between the attenuation values in the reconstructed volume and the
true attenuation coefficients of the test object [10, 11]. CT images are easily
affected by various artifacts since the final volumetric image is based on a multitude
of independent measurements. The reconstruction algorithm assumes that all these
measurements are consistent, so any inconsistency during the measurement will
directly transfer into an error in the reconstructed volume. As general terminology
used for artifacts in computed tomography, Barrett and Keat suggest the following
four artifact types [10]:
(a) Streaking
(b) Shading
(c) Rings
(d) Image distortion
Further it is possible to group the artifacts according to their origins into four
categories:
6.2 Detail Visibility 465

Fig. 6.7 Comparison of CT


scan slice of same object
e at
with different spot sizes F
constant magnification
factor M ¼ 5:5

1. Physics-based artifacts, which result from the physical processes involved in the
acquisition of the projection images
2. Hardware-based artifacts, which result from imperfections in the acquisition
system
3. Reconstruction artifacts, which are caused by the deficiencies in the image
reconstruction process
4. Motion artifacts, which are caused by movement of the test object during
acquisition
In the following, typical artifacts of these four categories are discussed following
the considerations made by [10] for medical applications, but are generalized in
their meaning to the field of material analysis, and distinct examples in application
to fiber reinforced materials are presented.
466 6 Computed Tomography

6.2.2.1 Physics-Based Artifacts

Beam Hardening

As a polychromatic X-ray beam passes through matter, low-energy photons are


preferentially absorbed, and the attenuation is no longer a linear function of
absorber thickness. Hence, the beam becomes “harder” because the lower-energy
photons are absorbed more rapidly than the higher-energy photons, and therefore
the mean energy of the beam increases [10, 12]. This may induce several types of
artifacts frequently occurring during scans of fiber reinforced materials. Among the
types of beam hardening artifacts, there are the so-called cupping artifacts, the
appearance of dark bands or streaks between objects with high density, and
the glooming of high-density inclusions. Several algorithms are available in mod-
ern reconstruction programs to minimize the effect of beam hardening to some
extent [13–15].

Cupping Artifact

The occurrence of cupping artifacts is readily explained following [12] by consid-


ering the attenuation profile observed for a uniform cylinder as seen in Fig. 6.8a.
Ideally, the attenuation profile would resemble a certain shape allowing a recon-
struction of the cylinder in its expected shape. However, the beams passing through
the center of the cylinder suffer more from beam hardening than those passing the
edges. Since higher-energy X-rays are attenuated less during propagation, the
projection areas with more beam hardening suffer less from attenuation than
theoretically expected. The reconstruction of these falsified projections thus causes
a characteristic cupped shape at the edges of the cylinder.
This type of artifact is frequently observed at the edge of the material, i.e., the
interface between the surrounding air and the solid. This causes the white shade at
the edge of the cross-section of the fiber reinforced material seen in Fig. 6.8b and
should not be misinterpreted as material with higher density.

a b
attenuation cupping

uniform cylinder

ideal projection
projection with
beam hardening

detector position 65 μm

Fig. 6.8 Attenuation profiles obtained with and without beam hardening for an X-ray beam
passing through a uniform cylinder (a) and example of cupping artifact in CT scan (b)
6.2 Detail Visibility 467

a b c
dark bands
X-ray beam
attenuation attenuation
CFRP

steel
low density

high density

detector position detector position

3 mm

Fig. 6.9 Origin of dark bands due to beam hardening effects (a, b) and example of dark bands due
to high-density inclusion (steel layer) in a carbon fiber reinforced polymer (c)

Dark Bands

Another type of artifact, which may occur due to beam hardening, are dark bands
caused by spatially close high-density inclusions or oriented high-density materials.
In such a situation, the incident beam is affected more by beam hardening for the
path between source and detector where the high-density object is aligned with the
beam direction (Fig. 6.9a) and suffers less for the path perpendicular to this
arrangement (Fig. 6.9b). As consequence, the regions behind projection directions
with strong beam hardening appear with higher density than in reality. This effect is
visible as dark bands extruding from these projection directions. As an example for
this effect, a reconstructed slice of a CT scan of a fiber-metal laminate is shown in
Fig. 6.9c. In this case, the thin steel layer has much higher density than the
surrounding fiber reinforced material and therefore acts as high-density inclusion.
Along the long axis of the metal sheet, strong dark bands extrude from the position
of the specimen edge.

High-Density Foreign Material Artifact

Similar to the previous two effects, the presence of a material with comparatively
high density to the surrounding material can generally lead to visible streaking
artifacts. These typically occur because the density of the material may exceed the
measurement range of the detector system and thus results in an incorrect attenu-
ation measurement. In the context of fiber reinforced materials, any metallic
inclusions may cause this type of artifact. Since metallic materials attenuate the
X-ray beam much more than the polymeric matrix system or the typical reinforce-
ment fibers, the attenuation values of objects behind the metallic inclusion are
measured too high. Due to the application of the reconstruction algorithm to this
468 6 Computed Tomography

a b c

attenuation

streak inclusion
detector bands
limit

700 μm 200 μm

detector position true width of steel foil

Fig. 6.10 Wrong measurement of attenuation profile due to detector limit (a) and example of gloom
due to high-density inclusion (steel layer) in a carbon fiber reinforced polymer (b) and example of
streak bands due to high-density particle inclusion in a carbon fiber reinforced polymer (c)

incorrect measurement of the attenuation profile, bright and dark streaks in trans-
verse directions to the rotation axis may appear in the CT image.
The effect of this incorrect measurement of the density is visible as bright glow
around the high-density object. For the case of the fiber-metal laminate shown in
Fig. 6.10b, this causes a characteristic glow at the interface between the thin
metallic sheet and the fiber reinforced material reducing visibility of details at
this interface. As another example, streak bands due to a particle of high density
included in a fiber reinforced polymer are shown in Fig. 6.10c.

Partial Volume Averaging

The algorithms used in CT data reconstruction assume that the object is completely
covered by the detector at all view angles and that the attenuation is caused by the
object only. In cases where this situation is not fulfilled, the CT reconstruction starts
to show truncated sections. Due to the divergence of the X-ray beam, this is usually
visible at the end of the reconstructed volume if the scanned region is smaller than
the object (see Fig. 6.11a). This occurs because the finite size of the detector does
not allow measuring the full attenuation profile of the cone beam, since the red areas
are not equally scanned at all angles. This effect is exemplarily shown for a
reconstruction of a piece of fiber reinforced material in Fig. 6.11b.
Another artifact related to the partial volume effect is mostly relevant to the
specific scanning configuration of helix tomography. Here line scanners instead of
array scanners are used to produce each reconstructed slice. The feed of the test
object is done mechanically, so the thickness of each slice is determined by the
feeding speed. Considering the scheme in Fig. 6.12, the challenge is to provide an
equal lateral resolution along all three axes for reconstructing the volume image. If
the slice thickness is chosen large compared to the lateral size of the pixel, its aspect
ratio gets distorted and the contained information is smeared out (cf. Fig. 6.12a).
Here the length of the voxel along the feed axis is much larger than in the directions
6.2 Detail Visibility 469

a b
object

reconstructed
volume

detector detector
2 mm
position 1 position 2

Fig. 6.11 Scan configuration (a) and example of partial volume effect upon reconstruction (b)

a b material A
thick slices thin slices

feed axis
feed axis

slice 1
slice 1 slice 2
slice 3

material B
evaluation evaluation

slice 1 slice 1

pixel gray values


slice 2

slice 3

pixel gray values

Fig. 6.12 Origin of spillover effect and distortion of voxel aspect ratio causing blurred informa-
tion in the feed direction due to thick slices (a) compared to thin slices (b)

perpendicular. This averaging effect is called spillover and manifests itself in a


blurred image information along the direction of object feed. Selecting thin slices
instead allows keeping a decent aspect ratio of the voxel and retains sharp gradients
of material density (cf. Fig. 6.12b).

Quantum Mottle (Noise)

Another trivial type of artifact is due to quantum mottle. The origin of this effect is
the incident background radiation seen at the detector, which superimposes to the
scan beam. This results in a grain noise structure seen on the image and thus
470 6 Computed Tomography

Fig. 6.13 Example of grain


noise in CT image
grain
50 μm noise

Fig. 6.14 Scheme of measurement configuration without (a) and with (b) additional collimator
and apertures to reduce the intensity of secondary X-rays incident to the detector

resembles an artificial scatter of density. This noise is composed of the general


cosmic background noise and the secondary reflections of the X-ray beam, which
are incident to the detector at other angles than through the test object. As seen in
Fig. 6.13, this is typically pronounced for scans with a low signal-to-noise ratio. The
latter may occur for setups with X-rays barely able to penetrate the test object due to
its thickness or absorption, but also when attempting to distinguish materials with
close-by density such as carbon fibers and polymer matrix systems (see Fig. 6.13).
Some X-ray CT devices use additional shielding measures to reduce the amount
of secondary X-ray photons to the detector. In this context “secondary” refers to any
backscatter produced by interaction of the primary X-ray beam with other pieces of
the equipment, such as the shielding cabinet or the included mechanics. In the
configuration shown in Fig. 6.14b, the presence of apertures collimates the primary
beam and shields the cabinet from the incident X-rays so the amount of secondary
X-rays gets substantially reduced.
6.2 Detail Visibility 471

Photon Starvation

Another potential source of streaking artifacts is photon starvation. This may occur
for objects with highly attenuating areas in some of the projection directions.
Similar to the dark bands, these are caused by the relative alignment of the test
object and the source-detector axis. Other than for beam hardening the present
effect is due to a lack of photons arriving at the detector in general, rather than a
falsified attenuation profile. Therefore some of the orientations suffer from noisy
projection images. The consequences are streaks in the reconstruction plane, i.e.,
perpendicular to the rotation axis. For scanning of materials, this effect may be
avoided by previously selecting the orientation with highest attenuation and suit-
ably adjusting the X-ray intensity to avoid photon starvation at these angles.
However, for geometries with strongly distorted x/y aspect ratio, this may come
with the cost of saturating the X-ray detector at the other projection angles, so
suitable trade-off might be required.

Undersampling

As a fundamental principle to computed tomography, the number of projections


should be sufficient to allow a reasonable reconstruction of the specimen volume. If
the number of projections gets too low (undersampling), the first artifact to observe
is ray aliasing at sharp edges and at narrow details. This is seen for the object in
Fig. 6.15 being visible as discrete stripes at some distance to the scanned object.
If the overall number of projections gets even lower, severe imperfections of the
scanned object can be expected. This is seen in Fig. 6.16 for a decreasing number of
projection images as basis for the same reconstruction algorithm. For the full set of
736 images, the rectangular cross-section is well reconstructed, but also the occur-
rence of dark bands is seen as extension to the longer axis edges of the material.
With reduced number of images, the rectangular cross-section is hard to be
reconstructed accurately, and an increasing amount of ray aliases occur.

Fig. 6.15 Example of ray


aliasing due to
undersampling 350 μm

ray aliasing
472 6 Computed Tomography

23 images 46 images 92 images

184 images 368 images 736 images

Fig. 6.16 Example of image reconstructed with different numbers of projections

6.2.2.2 Hardware-Based Artifacts

Ring Artifact

The ring artifact is probably the most common hardware-based artifact. It appears
as image of one or more “rings” within the reconstruction plane. If one of the pixels
of the detector is corrupted or badly calibrated, the detector will give a consistently
false reading at each angular position. During the reconstruction process, this turns
into a circular artifact with the circle extension being perpendicular to the rotation
axis as seen in Fig. 6.17. The appearance of such ring artifacts naturally lowers
the image quality, since it tends to overlap with parts of the image reducing the
visibility of details in these areas. Modern CT systems allow recalibrating the
detector system to reduce the occurrence of ring artifacts by simply removing
the information of dead pixels in the detector unit. Similarly, some reconstruction
algorithms explicitly focus on suppression of ring artifacts by removing the dead
pixel content during the reconstruction process.

Tube Arcing

During operation of X-ray tubes, the occurrence of short circuits within the tube
may cause image artifacts due to tube arcing. The electrical shortcut typically
emanates from a spark between the cathode and tube cover, which causes a
spontaneous loss of X-ray intensity. The occurrence of this effect is linked to the
tube design and the particular reliability of the individual tubes.
6.2 Detail Visibility 473

a b

200 μm 100 μm

Fig. 6.17 Examples of ring artifacts due to dead pixel (a) and due to bad calibration range (b)

6.2.2.3 Reconstruction Artifacts

The next class of artifacts arises from the (physical) imperfection of the incident
X-ray beams and the according assumptions made in the reconstruction process.
Since this is different for the configurations of cone-beam CT and helical CT, the
resulting types of artifacts are distinguished based on these two arrangements in the
following.

Cone-Beam Effect

For the cone-beam CT configurations, further distinction is made between line-scan


detectors resulting in fan-shaped beam geometries and the array detectors resulting
in a cone-beam shape of the X-ray beam (cf. Fig. 6.18). Due to the physical size of
the beam geometries, the collimators can be closed more for the line scanners but
need to stay open more for the array detectors. As a consequence of the cone-beam
geometry, the photons collected at the detector position do not only originate from
the ideal transmission path along an ideally flat plane configuration achieved in the
fan-shape geometry but are also incident at other angles. This effect becomes more
pronounced the farther away the scan object is from the ideal rotation axis. Thus to
minimize the effect, it is most suitable to position the test object at the center of the
rotation axis. However, modern CT systems are readily equipped with cone-beam
reconstruction algorithms, so this effect does not play an important role in such
systems anymore.

Windmill Artifacts

The windmill artifacts are linked to the use of helical CT device for the scan. The
two most well-known artifacts for these systems are the zebra artifact and the stair-
step artifact. As zebra artifact, a periodic modulation of light stripes parallel to the
474 6 Computed Tomography

a b
fan-shaped cone-beam

line-detector

array-detector

Fig. 6.18 Difference between line-scan arrangement (a) and array detectors (b)

axis of rotation may appear in the images, which is due to the helical interpolation
process causing a modulation of the noise level. This effect starts to be more
pronounced at some distance to the axis of rotation. As a stair-step artifact, the
misalignment of the individual slices during stacking along the axis of rotation is
known. This causes steplike contours responsible for the name of this artifact.

6.2.2.4 Motion Artifacts

Motion artifacts may result from sudden movement or relaxation of a test specimen
being scanned. This results in unequal projections of the angles before and after
movement and therefore needs to be avoided for proper reconstruction of the
volume. For most of the CT scans involving fiber reinforced materials, motion
artifacts may not seem to be of huge relevance. If the specimen is tightly fixed to the
rotary table, there is no strong likelihood of sudden movement of the specimen.
From medical imaging, the consequence of such sudden movement in CT imaging
is well known, and motion artifacts usually appear as shading or streaks in the
volumetric image or may even render the whole scan useless if the movement was
too high.
However, for the in situ load stages discussed herein, there is a certain likelihood
of motion artifacts during scanning. For the test object being scanned still subject to
mechanical load, the occurrence of partial failure during the acquisition process
may cause small movements of the scan object and thus cause the aforementioned
artifacts. This is exemplarily seen in the example of a fiber reinforced material
recorded within a tensile load stage as seen in Fig. 6.19. While the bottom part of
the CT scan is free of motion artifacts, the upper half of the fractured material was
subject to sudden secondary failure causing an additional motion during the acqui-
sition procedure turning into the blurred representation seen best at the edges.
6.2 Detail Visibility 475

Fig. 6.19 Motion artifact in


upper half of fractured
material as observed in
in situ load stage

motion artefacts

300 μm

6.2.3 Detectability of Defects in Fiber Reinforced Materials

High-resolution computed tomography can be used to identify microstructures in


composite materials, such as fiber distributions [16–18], void volume fractions
[19–24], and damage mechanisms such as cracks and fiber breaks [25–34] occur-
ring within the volume of the material. A high level of confidence can be placed in
the results as the features are viewed directly at the relevant length scale. In the
following some representative examples of computed tomography images of fiber
reinforced materials are presented to demonstrate the current capabilities of high-
resolution computed tomography devices in application to fiber reinforced mate-
rials. Such high-resolution X-ray CT has gained considerable interest as a powerful
tool for material studies directed to understand interaction of composite failure
mechanisms [31, 35, 36]. Advances in the last decade now provide established 3D
imaging routines in the micrometer and sub-micrometer range. This method is
gaining importance in engineering and material science to determine the internal
characteristics of materials and their behavior [37]. In the case of fiber reinforced
materials, CT has been used to identify material characteristics like fiber orientation
and volume fraction at the sub-ply level, and failure mechanisms, such as ply cracks
[31, 32, 34, 38] and fiber failure [39, 40]. Among the multitude of possibilities, a
particular focus is given to the detectability of damage mechanisms in the follow-
ing, since this is the main topic within this book.
Figure 6.20a–d shows CT results of a fiber reinforced laminate with [(0/90)2]sym
stacking sequence made from a Sigrafil CE1250-230-39 prepreg. The
17 mm  150 mm  2.2 mm specimen was first subject to tensile load according
to DIN-EN-ISO 527-4 to induce some failure and was unloaded prior to ultimate
failure, unmounted and scanned without further preparation. The according acqui-
sition parameters of the scan are summarized in Table 6.1.
As seen in Fig. 6.20, the individual layers can clearly be distinguished based on
their different fiber orientations. Also, a high level of detail can be observed for the
inter-fiber cracks and inter-ply delamination present in the laminate. The details in
Fig. 6.20b, c allow to spot inter-fiber failure in the 90 plies, inter-ply delamination,
476 6 Computed Tomography

a b

z
x
z y
x
y

c d z
y

z
x
y
540 μm

Fig. 6.20 3D representation of [(0/90)2]sym laminate (a) and magnifications allowing to spot inter-
fiber failure and inter-ply delamination in 3D volume according (b, c) and respective virtual cross-
section (d)

Table 6.1 Acquisition parameters for scan of tensile specimens


Measurement parameter Value
Detector type Amorphous Silicon CsI detector (3072  2400)px2
Filter None
X-ray tube voltage 90 kV
X-ray tube current 150 μA
Focal spot size 1.5 μm
Voxel size 7.8 μm
Acquisition angles 2000
Image exposure time 1.0 s
3D reconstruction software Phoenix datos|x2 reconstruction

and inclusions like voids. Clearly one key advantage beyond classical microscopy
investigations is the possibility to produce virtual cross-sections at arbitrary posi-
tions of the scanned volume. In contrast, optical and electron microscopy may offer
even higher spatial resolution, but can only be carried out at surfaces, requiring
either a preparation step after loading (losing any relation to the applied load or
damage evolution) or via in situ analysis under a nonrepresentative stress state (i.e.,
monitored at free surface as opposed to the state within the bulk material). Polishing
and cutting techniques commonly required for cross-sectional imaging can also
induce artifacts which may be mistaken for features of interest.
6.2 Detail Visibility 477

a b
x x
420 μm 420 μm
y y

single filament failure


c
x
420 μm
y

fiber bundle failure

Fig. 6.21 Visibility of single carbon fiber filament failure (a, b) and fiber bundle failure (c) in 2D
cross-sections of [(0/90)2]sym laminate

Using a cross-section in the xy-plane, this allows inspecting the 0 layers for the
occurrence of fiber breakage. As seen from Fig. 6.21a, b, the occurrence of single
carbon filament failure (typical diameter 7 μm) is clearly visible from the cross-
sectional images. Moreover, the initiation of matrix cracks at these positions
causing splitting of the laminate in parallel to the fiber axis is also readily observed.
In addition, Fig. 6.21c shows an example of a fiber bundle failure.
Another advantage of CT is to segment the scanned volume to retrieve only a
particular part of the specimen. As seen in the 3D visualization in Fig. 6.22, this
allows to artificially remove one part of a fractured body to reveal the internal
fracture surface in a unidirectional T800/913 prepreg laminate. The high-resolution
scan settings reported in Table 6.2 allow spotting a high level of detail at the
fracture surface, such as the morphology and broken fiber filaments.
Using similar high-resolution settings reported in Table 6.3, CT scans also allow
removal of the matrix material based on its density and to spot the remaining single
carbon fiber filaments (cf. Fig. 6.23). This can be used to evaluate their orientation,
to quantify their distribution, or to spot fiber filament breaks as seen in the lower left
corner of the present HTA fiber bundle in a polyurethane resin. However, these
478 6 Computed Tomography

Fig. 6.22 Visibility of


details on fracture surface of 120 μm
fiber reinforced polymer in
high-resolution scan

z
x

y
fracture surface

Table 6.2 Acquisition parameters for scan of fracture surface


Measurement parameter Value
Detector type Amorphous Silicon CsI detector (3072  2400)px2
Filter None
X-ray tube voltage 50 kV
X-ray tube current 170 μA
Focal spot size 1.1 μm
Voxel size 3.4 μm
Acquisition angles 1000
Image exposure time 1.5 s
3D reconstruction software Phoenix datos|x2 reconstruction

Table 6.3 Acquisition parameters for high-resolution scan


Measurement parameter Value
Detector type Amorphous Silicon CsI detector (3072  2400)px2
Filter None
X-ray tube voltage 70 kV
X-ray tube current 220 μA
Focal spot size 0.6 μm
Voxel size 0.7 μm
Acquisition angles 1800
Image exposure time 2.0 s
3D reconstruction software Phoenix datos|x2 reconstruction

high-resolution images are only possible for specimen sizes small enough to permit
short focus-object distances.
In particular for fiber reinforced materials, the damage zone is sometimes
geometrically complex, and severe fragmentation during fracture is happening
frequently. Thus conventional cross-sectioning techniques using resin embedding
6.2 Detail Visibility 479

Fig. 6.23 Visibility of


single carbon fiber filaments 120 μm
fiber filament
and fiber failure in 3D
volume

z
x
fiber breakage
y

Table 6.4 Acquisition parameters for scan of impact damage


Measurement parameter Value
Detector type Amorphous Silicon CsI detector (3072  2400)px2
Filter None
X-ray tube voltage 50 kV
X-ray tube current 200 μA
Focal spot size 1.5 μm
Voxel size 27.5 μm
Acquisition angles 1000
Image exposure time 1.5 s
3D reconstruction software Phoenix datos|x2 reconstruction

and mechanical grinding and polishing are somewhat limited as many of the small
fragments are not conserved by this approach. Using a CT scan for such damage
zones, these small details can readily be visualized. As seen for the example of
impact damage in a carbon fiber/epoxy laminate with glass fiber rovings in
Fig. 6.24, all small fiber filaments residues and epoxy resin fragments can be
spotted, and the geometrical complexity is much better represented in a 3D image
than in a simple 2D cross-section.
The analysis of porosity inside a laminate has achieved high attention within the
last decade [19–24]. Due to the direct visibility of gaseous inclusions within the
laminate, the direct quantification of their volumetric fraction is tempting, but
standardized procedures are still subject to recent research efforts. As an example
of the visibility of voids on the micro- and mesoscale, two examples are shown in
Fig. 6.25a, b for a CT scan with the parameter settings of Table 6.1. Extensive
studies have been carried out to investigate the visibility and accuracy of void
dimensions as function of CT scan parameters [21, 23, 24]. Since the quantification
of such inclusions is not within the scope of this book, this aspect will not be
480 6 Computed Tomography

a z
y
x

Impact
3.1 mm

z
y
x
b

Impact
3.1 mm

Fig. 6.24 Scan of damage zone as due to impact damage in 3D view for low- (a) and high-energy
impact (b)

a z
y

voids

1 mm

b z
x

particles
0.4 mm

Fig. 6.25 Detectability of voids and foreign material inclusion in [0/90/0/90]sym laminate

covered in the following. Also visible in Fig. 6.25b are examples of foreign material
inclusions identified as bright particles. Such inclusions may result from impurities
during the production process, but may also be part of the resin systems, i.e., as
flame retardants.
6.3 Volumetric Inspection of Materials 481

6.3 Volumetric Inspection of Materials

As demonstrated within the previous sections of this chapter, computed tomogra-


phy provides a microscopy method suitable to visualize the interior structure of a
fiber reinforced composite. Still the approaches presented so far are merely a
substitute to other microscopy techniques. In the following, distinction is thus
made between concepts used for ex situ loading, removal of the specimen, and
subsequent scanning in contrast to approaches using in situ loading of materials and
subsequent scanning.

6.3.1 Concepts for Ex Situ Loading

The basic concept for ex situ loading is a straightforward combination of a test rig
as conventionally used for mechanical loading of a material and a CT scanner. As
schematically shown in Fig. 6.26, the specimen is first loaded within the test rig
until a certain damage state is reached. The relevance of the load level for inspec-
tion may either be based on signatures in stress-strain curves or based on secondary
methods such as acoustic emission. The next step consists of dismounting
the specimen and taking to the CT device. Subsequently a CT scan is made and
the specimen is potentially remounted in the mechanical load rig to increase the
mechanical load level. To this end a sequence of damage states may be recorded

a
force
mount in CT scanner
θ

indication

b
force force
embed mount in CT scanner
θ

Fig. 6.26 Scheme of ex situ loading in combination with CT scan (a) and by using potting
material to embed damage zone (b)
482 6 Computed Tomography

as discrete volumetric images to allow assessment of the present damage status of


the specimen [29, 41–45].
However, the release of the applied mechanical load during dismounting may
cause a closure of small cracks and, therefore, render them invisible during the CT
inspection. Hence it might be necessary to preserve the stress-strain state within the
material during scanning. To this end an approach using epoxy resin as potting
material to embed the specimen under load can be applied [44]. Typically fast-
curing epoxy resins without high-density fillers are used to avoid generation of
artifacts within the epoxy resin. Optical transparence is not of vital importance
since the X-rays will penetrate the material, but it helps to later select the correct
portion of the specimen for a precise cutting process. For the purpose of segmen-
tation and to increase visibility of the specimen, a noticeable difference in material
density between the potting material and the embedded fiber reinforced polymer is
recommended. This may be subject to trial and error, since typically the difference
in density values is very low, when comparing values of epoxy resins used as
potting material and values of the epoxy resins used as matrix materials. Ultimately,
this procedure will assure that a certain portion of the original stress-strain state is
preserved within the specimen, and small cracks will be better visible. As disad-
vantage, the specimen cannot be remounted to the test rig for further loading, and
there is some justified doubt, whether this procedure fully inhibits specimen
relaxation. Therefore any quantitative information like separation of crack walls
or other geometric measurements need to be taken with necessary care. Also, from a
practical point of view, some initial experience is required to perform the potting
process within a mechanical test rig since the relevant portion of the specimen
usually is in the middle of free air or barely accessible due to the load fixtures. In
practice this means that suitable potting compounds need to be fabricated for each
specimen type. Moreover, with the specimen being under load during this proce-
dure, according safety measures are required to avoid unintended failure during a
hold cycle.

6.3.2 Concepts for In Situ Loading

In order to perform in situ observations of material failure, further modifications to


the established test setups are required. As major difference to the ex situ approach,
the configurations discussed in this section use an arrangement as shown in
Fig. 6.27. Here, the load rig is mounted in situ within the CT scanning device
[25, 26, 31, 34, 35, 40, 46–61]. For the implementation of these concepts,
three experimental factors may be distinguished and shall be discussed in the
following. These are the type of X-ray source, the type of X-ray detector, and the
type of load rig.
As general remark to all of the techniques mentioned within this section, it is
important to understand that none of these CT techniques really operates in a real-
time sense. Although comparatively fast scans are possible nowadays and the
6.3 Volumetric Inspection of Materials 483

Fig. 6.27 Experimental force


configuration for in situ
loading during CT scan
θ

source in situ applied detector


load

reconstruction time may be of similar order, the specimen still is required to be kept
under a certain load for durations of minutes to hours during scan operation. This
will allow stress relaxation within the material and therefore will alter the “true”
geometry of the specimen. Hence all practical implementations make use of a short
duration to allow for stress relaxation of the specimen before scan operation in
order to avoid motion artifacts. Therefore, the name in situ is only considered to be
correct in the sense that the specimen does not need to be dismounted from an
external load rig to be inspected by CT. Nevertheless, this approach enables several
new possibilities for analysis of failure behavior of fiber reinforced materials, so it
is worthwhile to consider some of the technical details of this approach more
precisely.

6.3.2.1 Type of X-ray Source

The majority of the published approaches were carried out using synchrotron-based
CT scanners in conjunction with specifically developed load rigs [25–28, 31, 32,
34–36, 38, 40, 42, 45, 46, 49, 57, 61–64]. The main reason for this trend is the high
brilliance and high intensity of the synchrotron beams in combination with a small
spot size assuring high-resolution imaging as discussed in Sect. 6.1. Since com-
mercial devices hardly could compete with any of these three aspects, only few
attempts have been made to use commercial CT scanners for in situ inspection with
a level of detail as discussed in Sect. 6.2. Nowadays there are some first CT
scanners which provide voxel resolutions in the sub-micrometer range and are
thus suitable to resolve the microstructure of a fiber reinforced polymer. One of
the key arguments in favor for commercial X-ray CT scanners is without doubt the
price and the availability. However, the intensity of commercial X-ray tubes is
hardly comparable with the photon flux reached at typical synchrotron beam lines,
so the contrast resolution is still not within the same range, and scan durations are
typically much longer. Since longer scan times increase the chance of motion
484 6 Computed Tomography

artifacts and may also induce motion unsharpness, these should be kept at a
minimum. Therefore, a suitable trade-off between the necessary detail visibility2
and the scan duration is required.

6.3.2.2 Type of X-ray Detector

Similar to the X-ray source, the type of X-ray detector is of relevance when
considering in situ inspection of materials. Because both devices are usually
coupled to form one scanning device, it is not straightforward to discuss them
independently, but since a variety of configurations is encountered in practice, the
relevant parameters of the detector system shall also be briefly discussed. Regard-
less of the X-ray source, one can distinguish between line detector and array
detector systems as previously discussed in Sect. 6.2. The key advantage of line
scanners in the present context is their fast scan times as compared to array
detectors. If the in situ inspection is restricted to a particular cross-section within
the material (cf. Fig. 6.28a), this enables very fast scan times even with typical
commercial X-ray sources. Here typical scan durations were found to be in the
order of few minutes and thus reducing the likelihood of motion artifacts. Obvi-
ously, the drawback of this detector configuration is the lack of 3D information of
the material interior. So unless damage is expected to occur at a distinct position
(i.e., at a notch) or is expected to be randomly distributed, it is very likely to miss
damage formation in some parts of the specimen under load. Hence, to retrieve
information out of a certain volume of the material, either subsequent line scanning
or an inspection using an array detector system is the only choice (see Fig. 6.28b).
For all other technical aspects of the detector system, the same considerations as for
conventional X-ray CT scans apply.

6.3.2.3 Type of Load Rig

As key component of the in situ loading concept, the load rig is discussed next.
From the mechanical engineering perspective, one challenge is to apply a certain
stress on the test specimen, while being able to rotate the whole configuration
ideally by 360 without having any additional obstacles within the projection area.
Since any applied load also needs to be compensated within the rig, the typical
scheme of a load test rig is of composed of the components shown in Fig. 6.29.
These comprise the specimen fixtures, a stationary (fixed) part and a moving part
driven by an actuator imposing a displacement to the specimen. In typical labora-
tory size machines, the vertical glides are simple columns, and the stationary and

2
As the first approximation, an increase of the number of projections and an increase of the
exposure time can be assumed to be beneficial for detail visibility if low photon intensity is given.
Both parameters add to the overall scan duration.
6.3 Volumetric Inspection of Materials 485

a
force
z
x
y
x
y
slice
reconstruction
defect indication
line-detector

b
force defect indication
z

x
y volume
z
reconstruction
x
y

array-detector

Fig. 6.28 Defect indication as retrieved from in situ measurement using line detector (a) and
using array detector (b)

a b
moving part

force flux force flux

specimen
region of specimen
interest
moving part
columns
cylindrical
sleeve

fixed boundary fixed boundary

Fig. 6.29 Scheme of conventional load test rig using columns (a) and load test rig using
cylindrical sleeve for load deflection (b)

movable parts of the machine are simple beams. If the whole configuration now is
rotated around its center axis, it will be subject to substantial artifacts caused by the
presence of the columns when passing the projection area. Hence, a configuration
using a radially symmetric geometry as shown in Fig. 6.29b is favored from this
perspective. Moreover, having the bottom fixture as moving part also enables a
simple mechanical design for the upper part of the test rig. Thus the upper part is
486 6 Computed Tomography

chosen as stationary fixture, and the lower part is fully instrumented by an actuator
and a load cell. The force flux for both cases is indicated by blue dashed arrows.
In order to ensure high detail visibility, further considerations with respect to the
material used for the load rig are required. Since metallic parts exhibit compara-
tively high X-ray attenuation, a presence of these parts within the projection area
should be avoided as much as possible. To scan the region of interest shown in
Fig. 6.29b, materials with low X-ray attenuation should be applied at all positions
within the projection path. Nevertheless, these parts are meant to carry the full
mechanical load as applied to the specimen, so they need to be of a sufficient cross-
section to avoid buckling or failure of the load rig. Here polymer materials are
among the first choices when designing this part of the load rig, which is typically
done in the shape of a tube. Transparent polymers are technically not required since
the X-rays will penetrate the polymer material, but they allow for visual inspection
of the specimen alignment and also were found to improve the handling during
mounting of specimen. To select a transparent polymer, further aspects may also be
considered. First, the material should be able to withstand X-ray intensity for an
extended period without optical or mechanical degradation. Second, the micro-
structure of the material should be homogeneous and free of textures, which could
cause artifacts reducing the detail visibility of the scanned specimen.
In order to achieve highest possible resolution, the limiting factor in terms of the
described configuration is the distance between the center of the test specimen and
the outer dimension of the test rig. Hence, this value should be minimized to
achieve high-resolution images. Due to the combination of the aforementioned
aspects, the load deflection concept using a tube (cylindrical sleeve) instead of
columns is generally favored, since this geometry is fairly stable to withstand
buckling and comes with a larger effective cross-section than two or more columns
retaining the same outer dimensions of the test rig.
In order to measure the applied force, a load cell is included in the test
configuration. Since this requires some electronic readout, the preferred spot to
mount the load cell is at the lower part of the test rig. That way, the cables can easily
be attached to the rotary table and do not need to hang in free space avoiding
possible interference with the X-ray source and the projection directions.
Finally, the actuator may consist of either a manually operated mechanical
system or an electrically driven motor. In the following some exemplary configu-
rations as used for fiber reinforced materials are presented to explain some
established concepts for in situ CT scanning. In order to classify the configurations,
distinction is made between the different load cases.

Tensile or Compressive Load Test

As pointed out in the previous section, the basis for design of load rigs is a
cylindrical tube made from a material with low X-ray attenuation able to carry
the applied load. Therefore only cross-sectional drawings of the load rigs will be
shown in the following. One characteristic difference between the concepts in use is
6.3 Volumetric Inspection of Materials 487

a b
load screw

bearing

anti-torsion unit
tube specimen
pin pin
tube
specimen bearing

load cell
load cell F
F
motor

M U

Fig. 6.30 Comparison between manually driven load rig for tensile tests (a) and computer-
controlled load rig driven by electronic stepper motor (b)

the type of load control. While manual load drivers are mechanically more straight-
forward to implement and more economical, the advantage of a fully load con-
trolled systems certainly is the possibility to automatize the full procedure. In
Fig. 6.30 a comparison of two systems as commercially available is shown.
Figure 6.30a shows a simple load screw in combination with an anti-torsion unit
to stress a specimen pinned to the load rig. This concept is well suitable for isotropic
materials causing no failure in the pin section when designing the dimensions of the
tapered region and the thickness of the specimen accordingly. The anti-torsion unit
is vital in this configuration to avoid any torque being applied to the specimen
during the loading procedure. Otherwise superimposed tensile-torsion stress states
could be the result, or the mounted specimen could even fail due to excessive torque
during load application.
For the configuration of Fig. 6.30b, the specimen is also attached using pins. The
basic difference to Fig. 6.30a is the use of an electric stepper motor in combination
with a mechanical driver system to stress the specimen in close relation to the
operation principle of typical macroscopic load test rigs. Here, the moving part is
typically supported by additional bearings and also protected to avoid the occur-
rence of excessive torque applied to the test specimen. In both configurations, the
load applied to the specimen is measured by a load cell positioned at the bottom of
the load path.
In contrast to the concepts of Fig. 6.30, for fiber reinforced composites, a differ-
ent fixture concept was found to be necessary. Similar to macroscopic test speci-
mens, a connection to the load test rig by pins is not reasonable due to shear-out
488 6 Computed Tomography

failure for many laminate configurations. Following the established test standards
for tensile or compression testing, the typical approach is to produce tapered test
specimens by bonding reinforcements at the end of the specimen. The specimen is
then gripped using mechanical or hydraulic jaws to apply load.
In very small load test rigs, the fabrication of jaws comes with some additional
challenges, so a more practical solution was found to directly bond the specimen
within the load rig using standard two-component epoxy adhesives. Since the
duration of one experiment with CT scanning (e.g., hours) largely exceed the
duration for bonding and curing (e.g., minutes), this seems to be a reasonable
approach. For macroscopic specimens this is quite different because the total test
duration is usually within few minutes, and therefore a quick replacement of
specimens is appreciated. Using a suitable embedding length of the material within
the test rig, this allows failure in the free section and avoids failure of the adhesive
bond. A major disadvantage of this concept is the necessity to have replaceable
parts of the load rig tensile or compression bars, since the removal of specimen and
adhesive from the bars degrades the geometry and therefore requires replacement
after few un-mounting and re-mounting cycles. The schematic arrangement of the
load rig as used for tensile tests and for compression test of fiber reinforced
composites is shown in Fig. 6.31a, b, respectively. Similar to the macroscopic
test concepts, one key difference is to keep the free length of the specimen (i.e.,
distance between edges of lower and upper bar) smaller for compression tests to
avoid buckling.

a b
force flux force flux

tensile bar compression bar


glue glue
free length
free length
specimen

specimen movement

movement

F F

M +U M –U

Fig. 6.31 Load rig used for tensile tests (a) and compression tests (b) of fiber reinforced materials
6.3 Volumetric Inspection of Materials 489

Mode I Test

As distinct difference to tensile or compression load, the generation of other


relevant load states in fiber reinforced composites requires some further modifica-
tion to the test rigs of Fig. 6.31. As for the macroscopic investigations of laminates,
the damage progression under mode I load is of key interest to the design and
material optimization of composites. To observe such damage progression with
volumetric imaging as offered by CT scanning, the modifications shown in
Fig. 6.32 are introduced. In this load rig, the specimen is attached to the lower
load bar and moved in the vertical direction. This relative movement may either be
due to an actuator at the bottom as seen in Fig. 6.32 or by a load screw at the top as
seen in Fig. 6.30a. The material is prepared including a precrack as in the case of
macroscopic specimens. At the top load bar a sharp wedge is mounted sliding into
the precrack. Due to the steepness of the wedge, the laminate splits and a damage
zone develops around the precrack tip.
If the curvature of the wedge is sufficiently large and the material is brittle
enough, the crack tip position is sufficiently far away from the wedge tip. Thus the
crack progression occurs under a mode I load situation. If the wedge curvature is
too sharp, the laminate is instead “cut“ and the real load type causing crack
progression will likely be a superimposed load state. The force in this configuration
is measured at the bottom of the load path and needs to be carefully interpreted
since friction effects may add a substantial amount to the force value and thus may
not be directly used in the same way as in macroscopic load configurations (i.e.,
double-cantilever beam configurations). Instead, closer relationship of the mechan-
ical properties is expected to macroscopic wedge-peel tests as carried out for
adhesives or thermoplastic tapes [65–69].

Fig. 6.32 Test rig force flux


configuration for wedge-
loaded mode I investigation
of fiber reinforced materials
precrack

specimen

glue

movement

M –U
490 6 Computed Tomography

Flexural Test and Mode II Test

Another load configuration being distinctly different to the previous cases, the
flexural load case is shown in Fig. 6.33. Such loads can either be used to generate
flexural stresses or to induce mode II-driven cracks as in the macroscopically
applied end-notched flexure specimens. Distinction is made between a load rig
applying manual load introduction (cf. Fig. 6.33a) and a computer-controlled
system using a motor (cf. Fig. 6.33b). In both cases one challenge is to have
supports being present at different positions to allow a deflection of the mounted
specimen. Since the support at the center of the tube will always be present within
the projected images, the same considerations as for the tube apply. If possible, the
material of the support should be affected less by X-rays to assure mechanical
stability during exposure, and it should cause a minimum of artifacts. For the case in
Fig. 6.33a, the center support is tightly mounted to the tube and therefore remains at
a constant position. To introduce a flexural load, two load screws transfer their
displacement to the specimen using a slide bearing on which the supports are
located. Here one disadvantage is the lack of a proper load cell measurement and
the challenge to synchronize the upper and lower load screw in order to avoid skew
load situations.
An alternative option is shown in Fig. 6.33b. Here the upper and lower supports
remain at a constant position, while the center support is the moving part. By an
elongated drill hole within the tube, the support is guided to allow only movement
in the designated direction (i.e., the horizontal axis in Fig. 6.33b). Using an actuator

a b
unloaded loaded

specimen

sliding
support
sliding
guided
fixed contact
support
load screw supports
fixed
support

movement
F F
force
flux

M M

Fig. 6.33 Test rig configuration for flexural testing of fiber reinforced materials using load screws
(a) and computer-controlled load rig (b)
6.3 Volumetric Inspection of Materials 491

system at the bottom of the test rig, a pull bar is moved in the vertical direction to
stress the specimen. A wedge at the tip of the bar allows pressing the support firmly
to the specimen and causes deflection. During this operation, the load bar is sliding
against the tube, and thus the acting loads are also transferred to the tube in order to
avoid significant deflection of the bar itself. The load cell is mounted at the bottom
of the configuration to measure the required load. However, this load is composed
of contributions due to friction and due to specimen deflection and may not be
directly used. Hence using a material with known stiffness for calibration of the
system, the measured load may be corrected to obtain true mechanical properties of
the measured material.

Thermal Loading

As a fundamentally different type of load source relevant to fiber reinforced


materials, a change in temperature may also generate stresses in the material.
These stresses are solely caused by the thermal expansion of the material, which
may induce some strong gradients in the case of anisotropic materials as in fiber
reinforced composite laminates. Moreover, joints between metallic parts and fiber
reinforced polymers are subject to a strong gradient in the thermal expansion
properties which may induce severe stress at the interface between both materials.
Here the basic concept is to generate such stresses in situ to allow accompanied CT
scanning. As seen in the schematically drawing in Fig. 6.34, the concept of the test
stage is relatively similar to the designs explained in the previous paragraphs. The
specimen is mounted at the bottom part and is enclosed in a sealed cylindrical tube
made from X-ray stable material. To avoid the occurrence of artifacts, the cylin-
drical design is kept. To monitor the temperature within the chamber, a temperature
element is positioned close to the specimen. Either electric heat elements or Peltiér
elements are applied to cause a change in temperature reaching either above or
below room temperature.

Fig. 6.34 Test rig to


generate thermal stress in DT
material specimen
specimen

thermal expansion
tube
temperature
sensor

heatelement
U
492 6 Computed Tomography

Imaging Requirements

Whatever type of load is being applied to the specimen, one requirement in CT


scanning is to have the specimen immobile during the scan duration. Otherwise
motion artifacts will appear which may render the measurement useless or cause
less detail visibility.
In the case of fiber reinforced polymers, the stress relaxation effect of the matrix
system needs to be considered in this context. Consider the load scheme shown in
Fig. 6.35 for two subsequent steps of in situ CT scans. The aim is to load a specimen
up until a distinct load level F1 and perform a CT scan at this load level to finish step
one. Without material creep, the specimen would simply keep its internal stress
state at load F and scanning could be performed immediately. In reality all materials
will show some stress relaxation after reaching the programmed load. Hence after
the quasi-static load phase to reach F1 (blue arrows in Fig. 6.35), a wait period needs
0
to be present to allow the material to reach a relaxation load F1 . At this relaxation
load, the CT scan can take place to avoid motion artifacts. Subsequently, the load is
further increased to F2 and the next CT scan is performed in step 2 after allowing
0
stress relaxation of the material to the load level F2 . This sequence is carried on
until the specimen finally fails.
This procedure immediately allows concluding that the mechanical experiments
need to be performed in displacement-controlled mode. In force-controlled mode,
the material would undergo a more or less constant creep and therefore

Fig. 6.35 Typical a


displacement profile (a) and step 1 step 2
load profile (b) during in situ
CT scanning including CT scan 1
displacement

effects of relaxation and


sequence of actions CT scan 2

loading 1 loading 2

me
b
load

step 1 step 2
F2

F2‘
F2

F1‘

waiting me
6.4 Digital Volume Correlation 493

continuously deform causing no stable conditions to perform CT scanning. It


should be noted that the ratio of durations for loading and scanning is not to scale
from Fig. 6.35, since the CT scan operation may easily fall in the order of hours,
whereas the load duration typically still is within a few minutes. Also the level of
stress relaxation is exaggerated for better visibility. However, careful consider-
ations are required when interpreting the imaged damage state. Firstly, the quanti-
fied dimensions and corresponding stress-strain states might be of reduced amount,
since the specimen undergoes relaxation prior to scanning. Secondly, upon subse-
quent loading, the material starts from a different stationary state, so the material
behavior concluded from such interrupted load profiles may be altered compared to
the conclusions drawn from continuous load profiles until failure.

6.4 Digital Volume Correlation

The approach to use the correlation of a sequence of digital images is already well
established and is being applied more and more in material testing applications.
Volumetric imaging-based X-ray computed tomography [25, 26, 31, 34, 35, 40,
46–61], magnetic resonance imaging [70], or, for partially transparent or opaque
materials, laser confocal scanning microscopy [71–75] are also used frequently.
Based on the volumetric image information, a consequent extension of the digital
image correlation method presented in Chap. 3 was suggested by Bay
et al. [76]. They were among the first to consider the correlation of two subsequent
volume images [76]. This method of DVC allows to measure quantitatively the 3D
internal displacement and deformation of a material. The main distinction between
approaches of 2D-DIC, 3D-DIC, and DVC is shown in Fig. 6.36. The primary
difference between DIC and DVC is the hierarchy of information derived from the
measurement. In simple 2D-DIC, a plane surface measurement is performed, not
being able to detect out-of-plane components w and thus resulting in simple 2D

a b c
z0

z y z y z y
x x x

2D image correlation 3D image correlation Digital volume correlation

u= u= u=

Fig. 6.36 Comparison of displacement vectors u as obtained from 2D-DIC (a), 3D-DIC (b), and
DVC (c) including schematically representation of information hierarchy
494 6 Computed Tomography

displacement information u of the outer surface. The extension to 3D-DIC allows


measuring the out-of-plane component of the displacement and results in a 3D
displacement measurement u of the surface. For the DVC approach, the displace-
ment is not just measured at the surface of the material but originates from the full
volume and therefore constitutes a true 3D measurement of the displacement vector
u in every point of the volume.
Among the applications of the DVC technique, there are obvious candidates like
the ability to validate the result of 3D-FEM computations. Successful applications
include materials such as wood [41, 42, 49] and composite materials [58].
In full analogy to the procedure used in DIC, the basic steps for DVC are the
following:
1. First recording a volumetric image of the material
2. Deforming the volume due to applied loads
3. Second recording of a volumetric image of the material
4. Correlation of both volumetric images using subvolumes and computation of
displacement vector for each subvolume
The acquisition of a volumetric image and the in situ deformation of the material
were presented throughout Sects. 6.1, 6.2, and 6.3, and the only remaining step now
is the correlation of two volumetric images. The basic concept is shown in Fig. 6.37.
Since a computation of the displacement vector of a single voxel is not suitable due
to the ambiguity faced in the image correlation process (cf. Sect. 3.1), a local
neighborhood around one voxel is used. As seen in Fig. 6.37, in DVC this is a local
subvolume subset rather than a 2D subset as introduced for DIC. Apart from that,
the subvolume is subject to a general displacement, rotation, and deformation as
introduced in DIC before.
In order to perform the correlation process in DIC, speckle-type patterns are
applied to the inspected surface to provide locally unique subsets. Since such
application of speckle patterns is not possible in the interior of the material, other
image signatures are required to perform the correlation process. To this end, the
material microstructure itself can be used to provide a unique correspondence of

a b
subvolume

point u

z z
y y
Reference volume Deformed volume
x x

Fig. 6.37 Tracking deformation states between reference volumetric image (a) including local
volumetric neighborhood (subvolume) and deformed volume (b) as being used for DVC
6.4 Digital Volume Correlation 495

a b
50 μm 10 mm

Fig. 6.38 Examples of CT cross-sections of fiber reinforced material (a) and typical speckle
pattern used in DIC (b)

subvolumes. For some materials a stochastically distribution of particles or voids is


found on the microscale. Accordingly, subvolumes of such materials yield unique
information to perform a correlation process. For fiber reinforced materials, the
arrangement of the fiber filament cross-section is of such a length scale to allow
successful DVC [58]. A comparison of two cross-sectional images of a fiber
reinforced material in Fig. 6.38 to a speckle-type pattern reveals the close
similarities.
Although the principle of DVC constitutes a straightforward extension of the
DIC approach, one of the key differences is the computational intensity faced. To
this end, further distinction is required between two typical implementations of
DVC. The first algorithms developed were based on a correlation procedure using
only discrete voxel sizes. This comes with lower computational intensity and is
readily be implemented by search algorithms in the spatial [76] or frequency
domain [74]. However, this allows a calculation of the displacement vector only
by increments of more than half the voxel size, which is typically not feasible for
many materials being subject to small deformations only. Thus the strain resolution
of this simple implementation would be 16.7 % strain for a 3-voxel cube, 10.0 %
strain for a 5-voxel cube, 7.1 % strain for a 7-voxel cube, and so on. For the typical
strain values in the linear elastic range of many materials with fractions of 1 %, a
sub-voxel accuracy of the correlation procedure is required. Therefore, the present
focus of the method development is on the improvement of the algorithms used for
such sub-voxel correlation [76–81]. Since an extensive discussion of the image
correlation algorithms is beyond the scope of this book, the reader is referred to [82]
on the basics of image correlation and to Pan et al. [77] for a recent review and
modification of existing correlation algorithms.
To demonstrate the use of this method, Figs. 6.39 and 6.40 show some exem-
plary results of volume cross-sections as collected from an in situ CT scan of a
carbon fiber reinforced epoxy material. Following the procedure outlined in
496 6 Computed Tomography

Fig. 6.39 Arbitrary xy-cross-section out of volumetric image of in situ loaded fiber reinforced
polymer. Computed displacement fields in y-direction (a) and x-direction (b) are superimposed to
the CT cross-sectional image

Sect. 6.5.2.1, miniaturized specimens were prepared and loaded by a compressive


force in situ in parallel to the y-axis. CT scans were performed in the initial state
and at a load level prior to ultimate failure. For the DVC process, the software
package DaVis was applied, and the resultant displacement field is visualized as
6.4 Digital Volume Correlation 497

Fig. 6.40 Arbitrary xy-cross-section out of volumetric image of in situ loaded fiber reinforced
polymer. Computed strain fields in y-direction (a) and x-direction (b) are superimposed to the CT
cross-sectional image

x-displacement and y-displacement in Fig. 6.39 superimposed to the selected cross-


section. For technical reasons, the displacement scale is given in pixels with one
pixel corresponding to 1.36 μm. As seen from the black regions in the xy-slice, the
material includes several voids of diameters up to 100 μm. In the computed
displacement field, the material predominantly shows the compression due to the
498 6 Computed Tomography

applied compressive force in y-direction and the effect of Poisson’s contraction in


the x-direction. For the corresponding local strains shown in Fig. 6.40, the situation
is different. Here the larger voids show a significant strain concentration indicating
regions of higher compliance. As will be seen from the analysis of the failure modes
in Sect. 6.5.2, these regions form a weak link in the material contributing signifi-
cantly to the final failure mode. Similar to the analysis of DIC images presented in
Chap. 3, this analysis can thus be used to identify strain concentrations in the
volume of a material to perform assessment of possible failure locations prior to
failure or to analyze the evolution of failure initiating at such spots.

6.5 Application to Composites

The use of computed tomography to inspect the microstructure of fiber reinforced


materials has been outlined in the previous sections. Based on the scope of the book,
the last section of this chapter has its focus on capabilities of the method to detect
and quantify damage initiation and damage progression. Thus only exemplary
applications related to in situ testing and ex situ testing approaches are presented
in the following. Other fields of application of computed tomography as relevant to
composites such as porosity quantification or measurement of fiber orientation are
instead covered by recent literature [16–24].
Before proceeding to the first subsection on ex situ testing approaches, a brief
introduction to the use of segmentation tools to visualize the internal microstructure
is given. This step is necessary to provide the basic analysis routines used to
evaluate the damage progression as presented in Sects. 6.5.1 and 6.5.2.
As discussed in Sect. 6.1 and 6.2, the gray scale images obtained in computed
tomography basically represent the material density. In the absence of image
artifacts and for reasonable resolution, this may thus be used to segment different
materials based on their difference in density. A trivial example for this procedure
is given in Fig. 6.41. The selected xz-slices of a computed tomography scan result
from an inspection of a Sigrafil CE1250-230-39 prepreg fabricated with consolida-
tion pressures below the manufacturer recommendations. Due to the missing
pressure, this results in the formation of pores upon curing and is readily visible
in Fig. 6.41. Both laminates were scanned at a time, so the histogram of gray values
on the left side of Fig. 6.41 is representative for both laminates. Within the
histogram two clear peaks are found, which correspond to the density values of
air and the composite material. Based on the large distances of the peak maximums,
it may seem trivial to segment this image. The threshold of Fig. 6.41a is thus
selected at the half distance between both peak maxima. The corresponding seg-
mentation is indicated as yellow line in the xz-slice in Fig. 6.41a on the right side.
Obviously, some of the needle pores in the lower laminate are not segmented
correct. This is due to the presence of image artifacts as discussed in Sect. 6.2,
causing slightly different gray values for the case of air enclosed by the composite
material. By moving the segmentation threshold towards the material peak, more
6.5 Application to Composites 499

a
counts threshold 0.5 mm
background material

threshold

gray value

b
counts threshold 0.5 mm

gray value

Fig. 6.41 Selection of threshold for image segmentation at half distance between peak maxima
(a) and close to material peak (b)

realistic segmentation is reached as seen in Fig. 6.41b. Since this procedure is


driven by visual decision criteria, its suitability for extraction of quantitative
information is generally questionable. Consequently, various strategies to obtain
valid decision boundaries have been proposed in literature. Recently, this has seen a
strong focus on the application to quantify the degree of porosity as well as to
segment matrix and fibers [16–24]. Despite of these strong research efforts, there is
still no standardized procedure to perform image segmentation in computed tomog-
raphy, and consequently, extraction of quantitative volumetric information is still
challenging.
After image segmentation, there is a broad range of possibilities to analyze and
interpret the data, ranging from simple volumetric ratios to topological analysis. As
an example of both, the two laminates of Fig. 6.41 alongside with a reference
laminate are subject to a topological analysis with their results shown in Fig. 6.42.
On the top of Fig. 6.42a–c, the 3D visualization of the inspected laminates is shown.
All three laminates were scanned at the same time and are segmented using the
threshold of Fig. 6.41b. At the bottom of each figure, a topological analysis is
shown for the detected porosity. The false-color range indicates the volumetric size
500 6 Computed Tomography

z
y x

volume
large

small

z
y x

volume
large

small

z
y x

volume
large

small

Fig. 6.42 Visualization of porosity for well-consolidated laminate (a), laminate with some needle
pores (b), and laminate with extraneous porosity (c)

of the single pores (or the pore network if existent). As additional guide to the eye,
two black lines are added to the figures, representing the edges of the laminates.
A reduced consolidation pressure results in the formation of a substantial amount of
needle-shaped pores (cf. Fig. 6.42b) in combination with extraneous porosity
6.5 Application to Composites 501

(cf. Fig. 6.42c). Despite of the difficulty to extract quantitative information from
these measurements, the visualization of these voids prior to testing certainly helps
to assure that the specimen is free of porosity as for the laminate of Fig. 6.42a.
However, it is important to mention that the scan resolution will obviously influ-
ence the porosity assessment as well. As demonstrated by Kastner et al. [21], the
voxel resolution as well as the choice of the threshold will have tremendous impact
on the quantified porosity. Thus, the choice of scan parameters as well as the size of
the test specimen will need to be considered for such analysis routines.
In the context of damage assessment, it is worth noting that it is generally hard to
distinguish between the indications of porosity and those given by the formation of
cracks. This is owed to the fact that cracks in the composite material are basically
air inclusions of a certain shape and orientation. Since the visualization of these
indications is based on their density, it is thus impossible to separate both indica-
tions without further knowledge. As practical example for this challenge, a 3D
visualization of a tapered section of a laminate is shown in Fig. 6.43. This scan
visualizes the tapered section of a Sigrafil CE1250-230-39 prepreg test specimen of
[(0/90)2]sym stacking sequence fabricated with interspersed plies as described in
Sect. 4.7.3.2.
The specimen was subject to a tensile load until failure within the reduced
section (not shown in Fig. 6.43). For Fig. 6.43b, the gray value range of the
composite material is chosen transparent, so that only the volumes with density
close to air remain. For improved visibility, the air volume surrounding the spec-
imen was also chosen transparent, so only enclosed air volumes remain visible in

a b

z
z y
y x
x
3.5 mm
3.5 mm

Fig. 6.43 3D CT scan of tapered laminate section (a) and visualization of cracks and porosity by
setting composite material and surrounding air transparent (b)
502 6 Computed Tomography

Fig. 6.43b. This kind of visualization allows evaluating the orientation and position
of pores and cracks. However, there is no way to distinguish between the occur-
rences of either kind. The laminate has certainly seen some damage in the scanned
region before final failure of the specimen, so parts of the indications of Fig. 6.43b
are due to the presence of cracks. But as evident from computed tomography scans
prior to testing, there is always a certain amount of pores in the reinforcement
section, which is difficult to avoid by the chosen processing conditions. Hence, a
considerable amount of the indications of Fig. 6.43b are due to needle pores.
Because of their shape and orientation in parallel to the fibers of the individual
plies, it is not possible to distinguish them from the occurrence of inter-fiber cracks.
Also, it is likely, that the presence of a pore acts as initiation site for an inter-fiber
crack. Consequently, mixtures of both indications are also likely to occur.
Thus for damage assessment by computed tomography, the following procedure
is recommended:
1. Inspection of the relevant part of the test specimen prior to mechanical testing to
assure there are no significant pores or other inclusions
2. Comparative inspection after testing of same part of the test specimen
This procedure allows evaluating the difference before and after testing and thus
may be used to resolve the ambiguity of indications. In cases of almost negligible
indications prior to testing (such as in Fig. 6.42a), the indications after testing may
be directly used to perform damage assessment.

6.5.1 Ex Situ Testing

Based on the consideration presented in Sect. 6.3, it is suitable to distinguish


between inspection approaches performed ex situ and in situ. For the ex situ
approaches, the specimen is loaded in an external universal test machine until a
certain damage state is reached. Subsequently, the specimen is dismounted and
inspected by a CT scanner. Depending on the type of analysis, the specimen is then
mounted again in the universal test machine and loaded until the next damage state
is reached. In contrast, in situ approaches combine the capabilities of a universal
testing machine with a CT scanning device and are described in their applications in
Sect. 6.5.2.

6.5.1.1 Visualization of Damage Progress

In this subsection exemplary applications for tracking of damage progress using ex


situ CT approaches are shown. Since there are numerous applications found in
literature that could be added to this section [25, 26, 31, 34, 35, 40, 46–61], focus
will be given on two representative examples. As first example, the growth of a
mode II interlaminar crack is investigated. This kind of example is representative
6.5 Application to Composites 503

2.0 mm
z
x 2 mm

z
x
y

fracture surface

y
x 2 mm

Fig. 6.44 3D visualization of mode II fracture surface in unidirectional laminate and according
cross-sections in xz-plane and xy-plane

for a macroscopic indication and is used to demonstrate some basic concepts in


tracking of damage progression. As second example, the growth of inter-fiber
cracks in cross-ply laminates is presented. For this example, the major challenge
stems from the visualization of relevant indications and the compromise between
high image resolution and large inspection volume.
As first example, a typical specimen as used in end-notched flexure test to
determine the mode II fracture toughness is analyzed (cf. Sect. 4.7.2.2). In this
configuration the crack initiates at the precrack position and starts to grow
interlaminar due to the applied bending load. The typical appearance of the fracture
surface after testing is shown as 3D volume in Fig. 6.44. The xz-cross-section and
the xy-cross-section of Fig. 6.44 are used in addition to visualize the complexity of
the fracture surface achieved in this test condition. For this unidirectional prepreg
laminate fabricated out of the Sigrafil CE1250-230-39 material, the crack growth in
mode II condition causes splitting of the laminate in parallel to the fiber axis. Thus,
the real crack front is hardly described as an ideal rectangle, but has certain
waviness along the z-axis. In addition, the crack extends to different distances
along the y-axis depending on the z-position.
In order to evaluate the damage progression during the mode II test, the
specimen was first scanned directly after precracking and a second time after the
first load drop according to the evaluation procedure of ASTM D7905. A compar-
ison of both scans is shown in Fig. 6.45. For this visualization, the composite
material and the surrounding air volume were chosen transparent, so that only
indications of the enclosed volume remain visible. To improve the visibility of
the crack front, a small wedge was inserted at the precrack position to open the
interlaminar crack during the CT inspection. For the evaluation the 3D result is
projected to the xy-plane rendering all indications semitransparent.
504 6 Computed Tomography

Fig. 6.45 Comparison of mode II fracture surface in unidirectional laminate after precrack
procedure (a) and after mode II ENF test (b) using visualization of cracks and porosity by setting
material as transparent

The position of the crack front after precracking is marked as dashed line in
Fig. 6.45a. There is not a plane crack front, but a distinct profile with different crack
extension along the y-axis as function of the x-position. Also, there is a certain level
of pores readily visible inside the laminate. In this application they were found to be
quite useful, since their shapes can be used as characteristic fingerprint to determine
the offset between the first scan and the second scan as indicated by the black
arrows of Fig. 6.45. During the mode II test procedure, the crack front advances by
17.4 mm in average as seen from Fig. 6.45b. The crack front retains a distinct
profile. A 3D evaluation of the newly generated crack surface of Fig. 6.45 allows to
quantify the total crack surface, which evaluates as 1074 mm2 for the case shown.
Unlike other attempts to measure crack lengths using X-ray projection methods, the
use of the 3D volume considers the full topology of the crack surface and therefore
especially considers the waviness along the z-axis. For metallic materials such
approaches have already demonstrated their capabilities for quantitative estimation
of mixed-mode load conditions [83, 84]. The quantification based on visual obser-
vation of the crack front (averaged at both edges) according to the ASTM D7905
results in 348 mm2. Thus the real fracture surface is underestimated by a factor of
3. However, the data reduction routine in ASTM D7905 does not directly apply the
value of the fracture surface and, hence, does not benefit from such improved
evaluation methods. Instead, the volumetric measurement of the fracture surface
may constitute an alternative approach of fracture toughness determination in the
spirit of Griffith’s energy relation.
As further example, the growth of an inter-fiber network in cross-ply laminates is
presented. The damage sequence recorded for a [0/903/0]sym laminate made from
6.5 Application to Composites 505

800

700

600 566 MPa


Stress [MPa]

500 485 MPa


second change
404 MPa in stiffness
400
323 MPa
300 deviation from
linear elastic
behaviour
200

100

0
0.0 0.2 0.4 0.6 0.8 1.0 1.2
Strain [%]

Fig. 6.46 Stress-strain curve of exemplary [0/903/0]sym specimen with levels of CT inspection
marked in red

a b c
y y y
crack in 90° ply
x x x

first inter-fiber cracks

delamination

crack in 0° ply

404 MPa 3 mm 485 MPa 3 mm 566 MPa 3 mm

Fig. 6.47 Sequence of images recorded at distinct stress levels shown as projection of the 3D
volume to the xy-plane with material set transparent

Sigrafil CE1250-230-39 prepreg subject to the load profile shown in Fig. 6.46 is
given in Fig. 6.47. The specimen was loaded along the y-axis to the first level of
323 MPa, unmounted from the test machine and then scanned using the computed
tomography parameters given in Table 6.1. Subsequent to the scan, the specimen
506 6 Computed Tomography

was remounted and loaded until the second stress level of 404 MPa is reached.
These steps are repeated until reaching the stress level of 566 MPa. As seen from
the stress-strain curve of a specimen from the same plate loaded until failure, there
are several changes in specimen stiffness, which can readily be associated with
microstructural changes within the materials, such as crack formation.
In Fig. 6.47, the scans are shown as projection of the scanned volume to the xy-
plane following the visualization described above with the composite material set
as transparent. Since almost no porosity was observed in the scan at the lowest
stress level of 323 MPa, all indications visible in Fig. 6.47 may readily be under-
stood as formation of damage in the material.
In Fig. 6.47a, recorded at the region of first nonlinearity of the stress-strain
curve, the formation of a microscopic crack network occurs. This is indicated by the
light black lines in the xy-projection, corresponding to the formation of first inter-
fiber cracks in the 90 and 0 plies. At increased stress levels, the black indications
densify at the previous positions and appear at new locations (cf. Fig. 6.47b).
Depending on the orientation of the lines, these may readily be correlated to the
occurrence of further inter-fiber cracks, which can be related to their specific depth
position by the inspection of the 3D volume. Some areas show larger black clouds
extending from the position of previously grown inter-fiber cracks. These areas are
located at the depth positions between the individual plies and may thus easily be
understood as formation of delamination. At highest stress levels, a significant
amount of microscopic damage has already been accumulated and has led to the
formation of the crack network seen in Fig. 6.47c. In addition to numerous inter-
fiber cracks along the 0 and 90 directions, there is a substantial area showing
inter-ply delamination. Beyond this stress level, the damage is expected to accu-
mulate further and finally cause the rupture of full plies. Since this is typically not
compensated by the remaining plies, this will ultimately cause the rupture of the full
laminate.
Thus for stress-strain curves with significant signatures as seen in Fig. 6.46, the
direct transfer of those procedures to other load scenarios, specimen types, and
material combinations is straightforward. However, depending on the selected CT
scan parameters, the presence of image artifacts, and the volume size, it may result
in less visible details. Also, for several applications no significant deviations of the
stress-strain behavior from linearity occur prior to ultimate failure. To this end,
Sect. 7.2.4.1 presents a suitable combination between acoustic emission analysis
and CT to deal with such conditions.

6.5.1.2 Extraction of Geometries from Volume Data

Apart from visualization of damage progression itself, the extraction of existing


damage from volumetric images can be used as input for modeling approaches.
The application of computed tomography to perform dimensional measurements
is already standard in quality control and has seen a long history of developments.
Using the image segmentation methods presented above, it is feasible to extract 3D
6.5 Application to Composites 507

representations of the segmented entities. This process is known as reverse engi-


neering and is referring to the creation of a digital dataset based on the physical
representation of the object. Compared to alternative methods, like tactile or optical
measurement systems, computed tomography comes with the large benefit to also
show internal structures and thus allows virtual disassembly of parts. The outer
contour of these parts can be directly turned into point clouds or into tessellated
surfaces (e.g., as triangulated STL files). These files can then be compared relative
to a CAD drawing or may be used to inspect the dimensional accuracy of a part [5].
In the context of damage tracking followed herein, these possibilities are con-
sequently extended by the extraction of crack geometries as digital objects for
damage progression modeling or for validation of measurement methods. To this
end, two examples are presented in the following to demonstrate the possibilities of
this approach for assessment of material failure.
In a first example, the extraction of a fracture surface of a mode II crack is
presented. As schematically shown in Fig. 6.48, the full procedure basically con-
sists of two subsequent steps. In the first step, the crack surface is extracted from a
CT scan and digitized to yield a digital representation of the 3D geometry. Within
the second step, this digital object is imported to a FEM program and embedded to a
CAD object for further processing. Although these steps are conceptually easy to
follow, the practical handling of the surface extraction, digitization, and embedding
comes with several challenges and may result in inaccuracies if carried out inap-
propriately. Thus, the sequence is described in more detail in the following using

extracted
crack
front

z
x crack tip
y
fracture surface

Fig. 6.48 Scheme of procedure used to transfer CT scan of fracture surface into FEM-model
508 6 Computed Tomography

the software packages VGStudio MAX, MeshLab, and COMSOL Multiphysics as


representative tools to perform surface extraction, mesh processing and geometry
embedding, and FEM computation.
To extract the fracture surface, a suitable CT scan of the region surrounding the
damage zone is performed. In this context, the volume may already be reduced to
yield only the region of interest close to the fracture surface for further processing
as seen in Fig. 6.48.
The CT scan should be of sufficient resolution and contrast, to allow proper
image segmentation using the threshold techniques discussed in the beginning of
Sect. 6.5. For proper extraction of the crack tip geometry, the selection of the
segmentation threshold is of particular relevance and should be considered in detail
when performing the segmentation step (cf. Fig. 6.49). In this context it is worth
noting that most of the postprocessing steps cause a reduction of the volume of the
extracted geometry. Thus shrinkage of the extracted object is technically possible,
while artificial growth of the geometry is generally not deemed a feasible approach.
As a consequence, the threshold selection should be carried out in a way to slightly
overestimate the crack front position as seen in the example of Fig. 6.49c. Within
VGStudio MAX it is possible to perform a surface extraction step using
thresholding techniques. The obtained segmentation boundary can then be exported
as STL file format to yield a tessellated surface for further processing. In this
context, a “normal” level of detail was found to be appropriate for the surface
extraction algorithm in VGStudio MAX. Nevertheless, a high level of accuracy in

a b c
y

x
threshold

counts threshold counts threshold counts threshold

gray value gray value gray value

Fig. 6.49 Influence of selected threshold on geometry of crack tip


6.5 Application to Composites 509

this step may result in large files for further processing and may need adjustment
subject to the extracted volume, scan resolution, and computational capabilities for
processing.
As next step, the obtained STL file typically needs further processing before
used in a FEM environment. This is mostly due to several issues arising from the
tessellation algorithm applied to the extracted surface. First, there is a large
likelihood that the obtained fracture surface is not fully closed, which is caused
by imperfections of the CT scan as well as the selected segmentation threshold.
Such holes need to be closed, to yield an appropriate representation of the fracture
surface. Second, the tessellation algorithm is likely to generate topologically
non-manifold faces and vertices in regions of geometric singularities (i.e., at the
crack tip). Since these cause difficulties for generation of computational meshes,
these need to be removed as well. And finally, the presence of pores or crack
bifurcations may cause isolated entities, which may be considered in FEM
approaches, but may also lead to difficulties in mesh generations. Thus it was
found to be useful to remove such isolated surfaces before further processing.
Since all these operations are hardly implemented by manual mesh processing
strategies, they require an appropriate environment to perform such STL file
modifications. To this end, the software platform MeshLab was used for processing
and simplification of the extracted surface.
After import of the STL file, the first step consists of a removal of isolated
entities by the command “remove isolated pieces” applied to the imported mesh.
Based on the selected option, this is used to remove all isolated objects depending
on their diameter or number of faces as seen in the transition between Fig. 6.50a, b.

a b
1.5 mm

730000 elements
z pores
y x

c d

holes

350000 elements 75000 elements

Fig. 6.50 Use of “remove isolated pieces” command to remove isolated entities (a) and “qua-
dratic based edge collapse strategy” to simplify the mesh to different number of elements (b–d)
510 6 Computed Tomography

For further processing within a FEM program, the topology of the obtained mesh
may need some processing. Some of the mesh refinement strategies present in the
next steps will require a mesh with two-manifold edges. This means that each edge
should be connected to exactly two neighboring faces. Non-closed objects are
typically one-manifold (or even zero-manifold for plane edges) and objects
containing artifact faces are three- or more manifold. Before further processing,
the vertices, edges, or faces causing this lack of topological integrity of the surface
need to be removed. Examples for such non-two-manifold edges are enclosed faces
or overlapping faces. These are difficult to spot by the bare eye, but are easy to
select and delete by using the “select non-manifold edge” filter. Careful removal
without compromising the integrity of the geometry is used to yield a suitable mesh
representation for further processing.
Subsequently, the mesh is subject to a simplification scheme. Among the
different choices, the “quadratic based edge collapse strategy” option following
the scheme of Hoppe was found to be particularly useful [85]. Two stages of mesh
simplification are seen in Fig. 6.50c, d as compared to the exported mesh from
VGStudio MAX in Fig. 6.50b. For such mesh refinements, it is key to preserve the
topology of the extracted surface and to avoid removal of geometric details, which
are required for the computation result. The latter aspect is hard to determine at this
stage, so the simplification routines should be evaluated in their impact on the
computation results by comparing results of different simplification levels. Clearly,
this step is used to increase the numerical efficiency of the modeling procedure and
should thus be considered carefully to reach a satisfying level of detail, while
keeping the numerical intensity at a minimum. Newly generated or existing holes
are filled using the “fill holes” or “close holes” command as shown in Fig. 6.51.
Since a fracture surface is by definition free of holes, the latter is a requirement for
further computation tasks. As final step, the simplified and repaired surface is
exported to another STL file.
Within the FEM program COMSOL Multiphysics, STL files may be imported
during geometry creation. For the exemplary fracture surface of a mode II crack,
only the crack tip was scanned by CT and is imported into the FEM program. The
remaining parts of the specimen may either be imported from other CAD formats or

a b
1.5 mm

z holes
y x

Fig. 6.51 Use of “fill holes” command to artificially close the generated holes
6.5 Application to Composites 511

a b
Force Force
z
x

y
2.8 mm

1.4 mm

cohesive zone elements

Fig. 6.52 FEM-model configuration used to calculate strain fields due to static load using
extracted fracture surface (a) and reference case with plane fracture surface (b)

are being generated directly within the FEM program. From the imported crack
surface, only the lower half was kept, and the remaining crack surface was joined
with a surrounding volume using a sequence of intersection and union commands to
yield the volume shown in Fig. 6.48. The so-obtained volume was then embedded
within the full 3D geometry of a mode II specimen as shown in Fig. 6.52a. The full
fracture surface of the specimen is reached by extrusion of the crack contour at the
left edge of the embedded volume and finally yields the blue fracture surface seen in
Fig. 6.52a. For comparison a second model with identical outer geometry is
generated. As a major difference, the second model uses a simple rectangular
crack representation at the half-thickness position of the laminate as seen in
Fig. 6.52b.
For further computations, a line load acting in z-direction is applied at the center
of the specimen, while at the position of the lower supports, displacement constraints
are chosen, inhibiting movement in x-direction and z-direction. Material properties
are those of Sigrafil CE1250-230-39 as listed in Table B.1 in Appendix B.
The full fracture surface is modeled by cohesive zone elements following the
implementation described in Sect. 4.2.
Within the “Structural Mechanics module” of COMSOL Multiphysics, the
configurations of Fig. 6.52 are subject to a stationary analysis using an applied
force equal to the residual load after crack growth as obtained in the experiment.
The computation results are visualized as false-color plots of the shear angle in
Fig. 6.53. This representation of the calculated deformation is chosen in analogy to
the evaluation of mode II crack growth using DIC techniques presented in Sect. 3.4.
The typical 2D image seen by a DIC system is shown as an additional inset for both
cases.
It is evident from Fig. 6.53 that the different implementations of crack geometry
result in noticeable differences in the computed shear angle values around the crack
position and along the delaminated area. Both cases allow identifying the position
of the crack tip by the steplike change in shear angle from 0 to -0.5 or less.
However, the remaining signatures of the strain field around the position of the
crack tip are affected by the real shape of the crack tip, causing less distinct areas of
positive shear angle above and below the crack tip for the case shown in Fig. 6.53b.
512 6 Computed Tomography

a shear angle [°]


0.5
0.4
z 0.3
x
0.2
crack tip
y z 0.1
0
x
-0.1
-0.2
-0.3
-0.4
-0.5

shear angle [°]


b 0.5
0.4
z 0.3
x
0.2
crack tip
0.1
y z
0
x -0.1
-0.2
-0.3
-0.4
-0.5

Fig. 6.53 Comparison of calculated shear angle in ENF test for laminate with perfect center crack
(a) and shear angle due to real fracture surface as obtained from CT scan (b)

Moreover, the present finding is in contrast to the typical observations of crack


growth by DIC methods as seen in Figs. 3.48, 3.50, and 3.51 in Sect. 3.4. Thus, it
may be speculated if the shear strain signatures seen by DIC methods are really due
to the deformation of the speckle pattern or are more a result of the local chipping of
the speckle pattern due to interlaminar crack growth.
As second example for numerical computation utilizing an extracted fracture
surface, the configuration shown in Fig. 6.54 is presented. Applying the same
software environments as for the previous example, the fracture surface of a
RTM6 specimen as used for flexural testing in Sect. 5.2 is extracted from a CT
scan and is joined with a 3D volume of the full specimen. This test specimen is then
embedded within a model of the full test setup including a 3D representation of the
lower supports and the geometry of the pushrod introducing the mechanical load.
The dimensions of the model as well as the selected boundary constraints are given
in Fig. 6.55. To induce a mechanical load, the upper surface of the pushrod was
subject to a force of 78.8 N acting in z-direction, while movement of the edges at the
bottom of the support was constrained along the x-direction and the z-direction. As
further boundary constraints, the extracted fracture surface was modeled by cohe-
sive zone elements following the approach presented in Sect. 4.2 with a von Mises
failure criterion and the material properties of RTM6 as given in Tables B.1 and B.4
of Appendix B.
6.5 Application to Composites 513

push rod

surface
extraction
z
x

extracted fracture surface

supports

Fig. 6.54 Scheme of procedure used to transfer CT scan of fracture surface into FEM-model

Within the “Structural Mechanics module” of COMSOL Multiphysics, the


configuration of Fig. 6.55 is subject to a transient analysis. The acting force was
identical to the failure load of the specimen. All dimensions including the fracture
plane are chosen identical to the experimentally used specimen. Based on the
material properties and the von Mises failure criteria, this causes crack initiation
and crack propagation at this load level.
As seen from the false-color visualization of the degradation function Č(r) in
Fig. 6.56, the applied external load results in crack initiation at the notch position
with crack growth from the bottom to the top of the specimen. The position of the
crack front is determined by the evaluation of Č(r) and does not obey a spatially and
temporally homogeneous growth. Instead, the topology of the fracture surface
causes an inhomogeneous growth of the crack front along the y-axis. Due to the
initial acceleration, the crack velocity increases first (cf. t ¼ 12.5 μs to t ¼ 25.0 μs)
and decelerates when approaching the top of the specimen (cf. t ¼ 50.0 μs to
t ¼ 100.0 μs). The dynamics of this crack growth confirm well with results of a
fractographic analysis of the resultant fracture plane (cf. Fig. 5.8). Here a crack
initiation zone with sliver-like pattern is observed followed by pronounced Wallner
lines when reaching the deceleration zone. In order to perform a physically correct
description of the dynamics of the crack growth, it is thus required to consider the
topology of the fracture plane. A simple approach defining a straight rectangular
plane was found to yield inaccurate results even for such relatively simple failure
modes [86]. Especially the complex fracture planes of composite materials are
514 6 Computed Tomography

Fig. 6.55 FEM-model Force


configuration used to
calculate crack growth in
flexural test configuration
(based on [86]) z
x

20 mm

16.5 mm
15 mm
4.6 mm

support ø = 3.1 mm

Fig. 6.56 Crack growth in t = 12.5 μs t = 25.0 μs


RTM6 specimen as function
of time visualized as false-
color diagram of the
degradation function Č(r)
(based on [86])

z
x

t = 50.0 μs t = 100.0 μs
1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
6.5 Application to Composites 515

expected to benefit from approaches to directly obtain the geometric dimensions


from computed tomography measurements (see also Sect. 7.2.4.2). However, other
approaches using meshless methods [87–93] or refined FEM approaches [84, 94] do
not need to define an explicit fracture plane and therefore should be able to capture
the dynamics intrinsically. For such advanced modeling methods, the extraction of
fracture planes may be then directly be used to compare model predictions to
experimental results.

6.5.2 In Situ Testing

As primary difference to the previous section, the approaches considered as in situ


testing using computed tomography typically combine a mechanical test rig with a
CT device as presented in Sect. 6.4. Many applications to visualize and detect
damage progression in composite materials by in situ approaches have already been
published [25, 26, 31, 34, 35, 40, 46–61]. Thus, the aim of this section is not to
repeat their findings but to demonstrate the capabilities of such approaches using
some representative examples. To this end, results of a study are presented, which
investigate a subset of the fundamental microscopic failure modes of fiber
reinforced composites.
In order to induce a particular failure mechanism, the load frame concept
introduced in Sect. 6.3 is applied. This comprises an aluminum load frame
component in combination with a PMMA tube used as load deflection. The load
stage itself is a fully computer-controlled commercial device of type “Microtest”
applying a 500 N load cell for precise force measurements. For later combination
with AE measurements (see Sect. 7.2.4.2), the load frame was modified to form a
waveguide to transfer the AE signal from the source position to the sensor position
[60]. A 3D CT scan of a mounted specimen showing the details of the transfer
components is found in Fig. 6.57 next to a photography image of the system. As
explained in Sect. 6.3, the system applies a load deflection concept introducing
tensile or compressive load by movement of the lower bar. The upper bar is
mounted to the surrounding PMMA tube and acts as fixed counterpart. To improve
the acoustic propagation path between the specimen and the sensor, a waveguide
concept is applied, which is explained in more detail in Sect. 7.2.4.2. The AE
sensor itself is mounted in a screw thread machined into the upper part of the
stage.
The specimens investigated are fabricated as prepreg laminate using the carbon/
epoxy system Sigrafil CE1250-230-39 following the curing cycle recommended by
the manufacturer. Miniaturized samples of nominally 1.7 mm  2.0 mm  22.0 mm
(depth  width  height) are cut from the laminates utilizing a water-cooled
low-speed saw with precision diamond blade. Samples are attached to the alumi-
num load bars by an UHU Plus endfest 300 epoxy adhesive. Mechanical testing is
carried out in displacement-controlled mode with 0.2 mm/min displacement rate.
516 6 Computed Tomography

Fig. 6.57 3D scan result of


load rig showing cross-
section of the mounted AE sensor
specimen, the tensile bars,
and the mounted acoustic acoustic
emission sensor (left) and decoupling
according photography waveguide
image (right)
load bar

specimen

150 mm
load bar
PMMA tube

movement

y x

The generated load-displacement curves are recorded by the software program


“Deben Microtest.”
The different types of load bars are shown in Fig. 6.58. Depending on the
intended type of load, several load bar types are used. Tension and compression
tests were carried out by using the load bars of Fig. 6.58a. Here the specimen is
directly bonded to the load bars by an epoxy adhesive. The large embedding length
seen in Fig. 6.57 offers a comparatively large contact surface to the adhesive and is
sufficient to transfer tensile or compressive loads into the specimen until failure in
the free section. For compressive loading, the adhesive bonding has the further
advantage to support the specimen at the edges, thus avoiding end crushing by a
shear load introduction. As further load concept, a mode I configuration was
realized by a wedge cracking method. This is shown in Fig. 6.58b holding the
sharp wedge on the bottom load bar and a respective slit for the specimen machined
into the upper load bar.
In mechanical testing, a comparatively stiff load frame should allow a direct
measurement of the specimen strain by evaluation of the cross-head displacement,
but the strain rate intrinsically changes during increased loading [95]. For the given
case, the stiffness of the full load frame was evaluated as 635 N/mm for tensile and
compressive load and was not found to be sufficiently high to evaluate the specimen
strain directly from the movement of the stepper motor. A typical measurement
curve for the stress-strain relationship during tensile testing transverse to the fiber
6.5 Application to Composites 517

b c
a
140

120

100

stress [MPa]
80
150 mm

60

40

20

0
0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0
strain [%]

change in system
compliance

Fig. 6.58 Several load bars as used for tension and compression (a) and mode I (b) loading.
Typical stress-strain curve from tensile test exhibiting change of system compliance (c)

direction is shown in Fig. 6.58c. As seen from the figure, the curve exhibits a
characteristic change of compliance which unlikely originates from the material.
Instead such changes in system compliance were found in various load configura-
tions, being partially at characteristic load levels and partially being unique for each
measurement. Consequently, a direct compensation of the load frame compliance
by a linear correction factor is not possible. For this case a nonlinear compensation
would be required to account for changes in system compliance, i.e., due to screw
threads, bolt connections, and other connectors. Since this was found to be partially
at unique load levels, no generally applicable compliance correction could be made.
Therefore, further discussion will be limited to the observed failure mode and the
stress level during failure.
In the following, results are shown for the occurrence of inter-fiber failure in a
unidirectional composite due to tensile and compressive load as well as fiber
failure due to tensile and compressive load. Further results are presented on
tracking of interlaminar crack growth resulting from mode I loading. All scans
were performed using a Nanotom 180 CT scanner applying the scan parameters of
Table 6.5 and the “phoenix datos|x2 reconstruction” software for image acquisi-
tion and VGStudio MAX for postprocessing. Given the high voxel resolution,
details of the fracture plane and inclusions (e.g., pores) in the bulk composite are
readily visible.
518 6 Computed Tomography

Table 6.5 Acquisition parameters for CT scan


Measurement parameter Value
Detector type Amorphous Silicon CsI detector (3072  2400)px2
Filter None
X-ray tube voltage 50 kV
X-ray tube current 170 μA
Focal spot size 0.6 μm
Voxel size 1.7–3.8 μm
Acquisition angles 1000
Image exposure time 1s
3D reconstruction software Phoenix datos|x2 reconstruction

force fixture

fracture surface

z adhesive bonding
y x

Fig. 6.59 3D scan of the failure mode due to tensile load transverse to fiber axis including details
of the fracture surface (reprinted from [60])

6.5.2.1 Inter-fiber Failure

For tensile loading transverse to the fiber direction, the expected failure mode is a
comparatively smooth fracture surface normal to the applied load. An exemplary
CT scan after final failure of the specimen as shown in Fig. 6.59 confirms this
assumption. Although the details of the fracture plane reveal a certain roughness of
the fracture surface, the overall orientation is found to be normal to the load axis.
Beyond the possibility to extract the fracture surface as discussed in
Sect. 6.5.1.2, the 3D capabilities of the scan allow further inspections to be made.
Since failure of unidirectional composite transverse to the fiber axis is dominated by
the presence of laminate imperfections (e.g., inclusions, dry spots, and resin rich
6.5 Application to Composites 519

a b

0.4 mm

z
y

Fig. 6.60 Exemplary yz-cross-section before (a) and after failure (b) due to tensile load transverse
to fiber axis

areas), the 3D analysis may readily be used to spot such regions and to investigate
their relation to formation of the fracture plane. In the spirit of [28], an example of
such inspection is shown in Fig. 6.60.
Here the scan of Fig. 6.60a reveals a substantial amount of porosity in the
laminate, which has been quantified to be in the order of 5 %. From the position
of the pores, it is obvious that they link and agglomerate at the positions between
the individual plies, which can readily be observed in Fig. 6.60a. A direct compar-
ison to the contour of the fracture plane shown in Fig. 6.60b can thus be made. As
seen from the red line in Fig. 6.60a, the later position of the fracture surface links
several pores on the right side of the specimen. However, based on the cross-section
shown in Fig. 6.60, a deviation of the crack from the y-axis would be expected
including some of the bigger pores seen in Fig. 6.60a. Instead, the crack stays
almost in parallel to the y-axis. There are several reasons for this behavior. First, for
the given load scenario, the presence of a crack tip will immediately result in
transverse failure of the material due to the superimposed normal stress. Energet-
ically it is not favorable to deviate from this angle (cf. Sect. 2.2). Unless there are
strong stress concentrators, the crack will keep its propagation direction as seen in
Fig. 6.59. Moreover, the specific cross-section shown in Fig. 6.60 is only a part of
the full volume. Since some of the larger pores seen in Fig. 6.60a do not extend
along the full dimension of the specimen along the x-axis, they do not necessarily
constitute weak paths to dominate crack propagation.
To investigate the relation between porosity and formation of the fracture plane
further, two examples of compression transverse to the fiber axis are discussed next.
A direct comparison of the specimen shape before and after failure is shown in
Fig. 6.61.
Based on Puck’s failure theory, an angle of approximately 53 would be
expected for an otherwise perfect and homogenous laminate [96, 97]. As seen
from the 3D scan result in Fig. 6.61, the shape and orientation of the fracture
surface in this load case does coincide well with the prediction. However, parts of
the specimen were certainly subject to bending loads due to the partial or full
buckling of the specimen after crack initiation. Accordingly, smaller fragments of
the material have separated from the remaining specimen and are thus not included
520 6 Computed Tomography

peel ply texture


force

fracture surface

z
y x

Fig. 6.61 Comparison of specimen loaded by compressive force transverse to fiber axis before
and after failure

in the scan. Using an yz-cross-section, the resultant fracture angle can readily be
measured as shown in the example of Fig. 6.62. As average of ten measurements,
this evaluates as (53.6  1.5) and therefore agrees well to the prediction of Puck’s
inter-fiber failure criterion.
To study the influence of porosity in such load cases, a relevant yz-cross-section
of a laminate before and after fracture is shown in Fig. 6.63. The dashed line in
Fig. 6.63a resembles the contour of the final fracture surface seen in Fig. 6.63b. As a
result of the acting forces, a distinct fracture surface develops. This is partially due
to the collapse of pores and due to multiple fractures along the paths dictated by the
position of the pores. The red line of Fig. 6.63a clearly indicates that the connection
of individual pores finally results in the formation of the fracture plane and thus acts
as preferential path for crack growth.

6.5.2.2 Fiber Failure

In order to investigate fiber failure due to tensile and compressive load, different
specimen sizes were prepared to conform with the maximum force capacity of
500 N of the used load cell. Specimens are fabricated from the Sigrafil CE1250-
230-39 carbon/epoxy prepreg laminate using miniaturized samples of
6.5 Application to Composites 521

Fig. 6.62 Evaluation of


fracture angle due to
compressive force
transverse to fiber axis

Fig. 6.63 Exemplary yz-cross-section before (a) and after failure (b) due to compressive load
transverse to fiber axis

0.14 mm  1.35 mm  22.0 mm (depth  width  height). All other settings were
chosen as for the previous cases.
As seen from the comparison of the specimen before and after failure in
Fig. 6.64, the failure mode is distinctly different to the inter-fiber failure cases
522 6 Computed Tomography

force

adhesive
bonding

0.14 mm

fracture surface

z
y x

Fig. 6.64 Comparison of specimen loaded with tensile force parallel to fiber axis before and after
failure

since this load configuration requires the rupture of the reinforcement fibers.
Inspired by macroscopic test configurations for tensile tests parallel to the fiber
axis, a tapered region as shown in the xz-cross-section of Fig. 6.65a was prepared
using the epoxy adhesive. This ensures failure of the material in the reduced part of
the specimen as seen in Fig. 6.64. In addition to the rupture of the fiber filaments,
splitting along the z-axis occurs during failure. The latter is seen well in the
comparison of the specimen cross-section in Fig. 6.65b, c. Here several inter-fiber
cracks occur in parallel to the fiber axis.
The influence of pores on the occurrence of fiber filament failure has been
studied in [28]. Therein it was concluded that the presence of voids increases the
likelihood of a single filament fiber break in the immediate vicinity. Other exper-
imental studies with a focus on the accumulation of single fiber filament breaks and
their relevance to global failure of the laminate have provided substantial insight on
this failure mode [25, 26, 56]. Moreover, substantial refinement of the theoretical
framework to describe the agglomeration of fiber failure has been proposed by
Swolfs et al. [98–100].
To induce failure of the specimens with compressive load applied parallel to the
fiber axis, the same experimental configuration as for the tensile test is used. The
taper provided by the epoxy was found to be quite useful to stabilize the specimen
and to avoid buckling failure modes. As visualized the comparison of 3D scans
before and after failure of the composite in Fig. 6.66, the predominant failure mode
6.5 Application to Composites 523

Fig. 6.65 Comparison of tensile specimen cross-section in xz-plane (a) and xy-plane (b) before
loading and in xy-plane after loading (c)

force

0.41 mm

fracture surface

z
y x

Fig. 6.66 Comparison of specimen loaded with compressive force parallel to fiber axis before and
after failure
524 6 Computed Tomography

observed for the carbon/epoxy material used in this load configuration, is a


brooming failure mode at the center of the specimen. This is likely induced by
band kinking of the fibers prior to the final failure. As seen from the cross-sectional
images in Fig. 6.67, the upper half of the specimen impacts the bottom half upon
failure as typically observed in macroscopic testing. This impact causes much of the
substantial damage visible in Figs. 6.66 and 6.67. The used scan resolution allows to
resolve many details on the fracture mode, such as fiber fragments, the extent of
delamination, splitting of the beams, and other geometrical changes. Note also the
rotated alignment of the upper part of the beam compared to the bottom part. This is
seen best by comparing the cross-sectional images in the xy-plane of Fig. 6.67 taken
at the dashed red lines as indicated in the xz-plane cross-section.

Fig. 6.67 Cross-sectional


0.3 mm

x
view of specimen after

y
failure in directions
perpendicular to applied
load (xy-plane) and as cross-
section aligned with applied
load (xz-plane)

x
0.3 mm

y
0.3 mm

x
y

1.2 mm
z
x
6.5 Application to Composites 525

6.5.2.3 Interlaminar Failure

For initiation of an interlaminar mode I crack, a different load configuration than for
tensile or compressive testing is required. As discussed in Sect. 6.3, the specimen is
hold in an upright position using a fixture at the bottom part. At the top position, a
sharp knife is used to induce a mode I crack due to wedge loading of the two beams.
To facilitate crack initiation at the center part of the prepared specimen, a precrack
is introduced prior to testing. For this test procedure, specimens are fabricated from
the Sigrafil CE1250-230-39 carbon/epoxy prepreg laminate using miniaturized
samples of 1.7 mm  2.2 mm  32.0 mm (length  width  height). The displace-
ment rate of the indentation knife was chosen as 1 mm/min.
The resultant crack propagation after exceeding the 121 N for initiation of the
precrack is shown in Fig. 6.68 for two subsequent states of knife indentation. As
clearly revealed by these images, the knife indentation causes a splitting of the
beam resulting in a typical mode I load condition at the position of the crack tip.
The in situ scan of this mode I crack allows examining details of the crack front,
such as bifurcations or branching of the crack tip, surface roughness or interaction
with particles, and voids or other defects. As an example for this kind of analysis,
Fig. 6.69 presents cross-sections in the xy-plane (i.e., plane perpendicular to the
direction of crack growth) taken at different positions along the z-axis. Figure 6.69a
is taken at a position close to the present position of the crack tip. As noticeable
from the absence of the crack on the left side of the cross-sectional image, the crack
does not grow with a planar crack front along the y-axis (cf. also Fig. 6.44). When

Fig. 6.68 3D visualization force


of interlaminar crack force
growth at two subsequent knife
stages of knife indentation

crack tip

2.2 mm

z
y x
526 6 Computed Tomography

Fig. 6.69 Cross-sections in the xy-plane at z-positions of approximately 2 mm distance to the


precrack

a
x
2.2 mm
z

b
x
z

c
x
branching of crack tip z

Fig. 6.70 Sequence of images in xz-cross-section recorded in subsequent load stages to visualize
interlaminar crack propagation and branching of crack tip

moving towards the position of the indentation knife, Fig. 6.69b–d reveals the
roughness of the fracture surface along the y-axis as well as some crack bifurcations
and splitting of the laminate upon crack growth. The latter is particularly well seen
for the fiber filament residues visible as small gray particles between the two
fracture surfaces.
The progression of the crack tip may also be visualized for a sequence of images
taken at different indentation depths of the knife. As visible in cross-sectional
images in the xz-plane in Fig. 6.70 at identical y-position, the crack tip starts to
propagate starting at the precrack position (Fig. 6.70a). Due to the loading of the
indentation knife, the mode I condition at the crack tip causes interlaminar crack
propagation at a distance of more than 3 mm ahead of the knife position. Thus, the
crack is solely driven by a mode I load and is not simply cut by the indentation
knife. As seen from the cross-sectional images in Fig. 6.70b, c, the crack splits the
laminate and propagates with occasional branching of the crack tip. This sequential
imaging mode is therefore well suited to detect influences of enclosed particles or to
study deviations in the crack front due to the presence of defects, the textile
architecture, or other likely crack deflectors.
References 527

References

1. Cosslett, V.E., Nixon, W.C.: X-ray shadow microscope. Nature 168, 24–25 (1951)
2. Cosslett, V.E., Nixon, W.C.: X-ray shadow microscopy. Nature 170, 436–438 (1952)
3. Cosslett, V.E., Nixon, W.C.: X-Ray Microscopy. Cambridge University Press, Cambridge
(1960)
4. Nixon, W.C.: Improved resolution with the x-ray projection microscope. Nature 175,
1078–1079 (1955)
5. Flisch, A., Wirth, J., Zanini, R., Breitenstein, M.: Industrial computed tomography in reverse
engineering applications. In: Computerized Tomography for Industrial Applications and
Image Processing in Radiology, Berlin, Germany (1999)
6. Buzug, T.M.: Computed Tomography: From Photon Statistics to Modern Cone-Beam
CT. Springer, Berlin (2008)
7. Kalender, W.A.: Computed Tomography: Fundamentals, System Technology, Image Qual-
ity, Applications. Wiley, New York (2011)
8. Shull, P.J.: Nondestructive Evaluation: Theory, Techniques, and Applications. Marcel Dek-
ker, Inc., New York (2002)
9. Radon, J.: Über die Bestimmung von Funktionen durch ihre Integralwerte längs gewisser
Mannigfaltigkeiten. Berichte Sächsische Akad. der Wissenschaften. 69, 262–277 (1917)
10. Barrett, J.F., Keat, N.: Artifacts in CT: recognition and avoidance. Radiographics 24,
1679–1691 (2004)
11. Herman, G.T.: Fundamentals of Computerized Tomography: Image Reconstruction from
Projections. Springer, Berlin (2009)
12. Brooks, R.A., Di Chiro, G.: Beam hardening in X-ray reconstructive tomography. Phys. Med.
Biol. 21, 390–398 (1976)
13. Jin, P., Bouman, C.A., Sauer, K.D.: A method for simultaneous image reconstruction and
beam hardening correction. In: 2013 I.E. Nuclear Science Symposium and Medical Imaging
Conference (2013 NSS/MIC), pp. 1–5. IEEE (2013)
14. Van de Casteele, E., Van Dyck, D., Sijbers, J., Raman, E.: A model-based correction method
for beam hardening artefacts in X-ray microtomography. J. Xray. Sci. Technol. 12, 43–57
(2004)
15. Van Gompel, G., Van Slambrouck, K., Defrise, M., Batenburg, K.J., de Mey, J., Sijbers, J.,
Nuyts, J.: Iterative correction of beam hardening artifacts in CT. Med. Phys. 38, S36 (2011)
16. Kastner, J., Plank, B., Salaberger, D.: High resolution X-ray computed tomography of fibre-
and particle filled polymers. In: 18th World Conference on Nondestructive Testing, 16-20
April 2012, Durban, South Africa, pp. 16–20 (2012)
17. Salaberger, D., Arikan, M., Paier, T., Kastner, J.: Characterization of damage mechanisms in
glass fibre reinforced polymers using X-ray computed tomography. In: 11th European
Conference on Non-destructive Testing (ECNDT 2014), Prague, Czech Republic, pp. 1–9
(2014)
18. Krause, M., Hausherr, J.M., Burgeth, B., Herrmann, C., Krenkel, W.: Determination of the
fibre orientation in composites using the structure tensor and local X-ray transform. J. Mater.
Sci. 45, 888–896 (2009)
19. Sugimori, M., Lam, F.: Macro-void distribution analysis in strand-based wood composites
using an X-ray computer tomography technique. J. Wood Sci. 45, 254–257 (1999)
20. Little, J.E., Yuan, X., Jones, M.I.: Characterisation of voids in fibre reinforced composite
materials. NDT E Int. 46, 122–127 (2012)
21. Kastner, J., Plank, B., Salaberger, D., Sekelja, J.: Defect and porosity determination of fibre
reinforced polymers by X-ray computed tomography. In: 2nd International Symposium on
NDT in Aerospace, Hamburg, Germany, pp. 1–12 (2010)
22. Stoessel, R., Kiefel, D., Oster, R., Diewel, B., Llopard Prieto, L.: μ-computed tomography for
3d porosity evaluation in Carbon Fibre Reinforced Plastics (CFRP). In: International Sym-
posium on Digital Radiology and Computed Tomography, Berlin, Germany (2011)
528 6 Computed Tomography

23. Plank, B., Mayr, G., Reh, A., Kiefel, D., Stoessel, R., Kastner, J.: Evaluation and visualisation
of shape factors in dependence of the void content within CFRP by means of X-ray computed
tomography. In: 11th European Conference on Non-destructive Testing (ECNDT 2014),
Prague, Czech Republic (2014)
24. Kiefel, D., Stoessel, R., Plank, B., Heinzl, C., Kastner, J.: CFRP porosity characterisation
using μ-computed tomography with optimized test parameters supported by XCT-simulation.
In: Proceedings of Conference on Industrial Computed Tomography (iCT2014), Wels,
Austria, pp. 35–43 (2014)
25. Scott, A.E., Mavrogordato, M., Wright, P., Sinclair, I., Spearing, S.M.: In-situ fibre fracture
measurement in carbon-epoxy laminates using high resolution computed tomography.
Compos. Sci. Technol. 71, 1471–1477 (2011)
26. Scott, A.E., Sinclair, I., Spearing, S.M., Mavrogordato, M., Bunsell, A.R., Thionnet, A.:
Comparison of the accumulation of fibre breaks occurring in a unidirectional carbon/epoxy
composite identified in a multi-scale micro-mechanical model with that of experimental
observations using high resolution computed tomography. In: Matériaux 2010, Nantes,
France, pp. 1–9 (2010)
27. Scott, A.E., Hepples, W., Kalantzis, N., Wright, P., Mavrogordato, M.N., Sinclair, I.,
Spearing, S.M.: High resolution damage detection of loaded carbon/epoxy laminates using
synchrotron radiation computed tomography. In: ICCM-18 18th International Conference on
Composite Materials, pp. 1–6. ICC Jeju, Korea (2011)
28. Scott, A.E., Sinclair, I., Spearing, S.M., Mavrogordato, M.N., Hepples, W.: Influence of voids
on damage mechanisms in carbon/epoxy composites determined via high resolution com-
puted tomography. Compos. Sci. Technol. 90, 147–153 (2014)
29. Allen, T., Scott, A., Hepples, W., Spearing, S.M., Reed, P.A., Sinclair, I., Testing, N.,
Resistance, D.: Investigating damage resistance of hybrid composite—metallic structures
using multi-scale computed In: ECCM16—16th European Conference on Composite Mate-
rials, Sevilla, Spain, pp. 22–26 (2014)
30. Schilling, P.J., Karedla, B.R., Tatiparthi, A.K., Verges, M.A., Herrington, P.D.: X-ray
computed microtomography of internal damage in fiber reinforced polymer matrix compos-
ites. Compos. Sci. Technol. 65, 2071–2078 (2005)
31. Moffat, A.J., Wright, P., Buffière, J.Y., Sinclair, I., Spearing, S.M.: Micromechanisms of
damage in 0 splits in a [90/0]s composite material using synchrotron radiation computed
tomography. Scr. Mater. 59, 1043–1046 (2008)
32. Moffat, A.J., Wright, P., Helfen, L., Baumbach, T., Johnson, G., Spearing, S.M., Sinclair, I.:
In-situ synchrotron computed laminography of damage in carbon fibre-epoxy [90/0]s lami-
nates. Scr. Mater. 62, 97–100 (2010)
33. Moffat, A.J., Wright, P., Renault, A., Sinclair, I., Spearing, S.M.: Analysis of transverse ply
cracks using computed tomography. In: 17th International Conference on Composite Mate-
rials, Edinburgh, UK (2009)
34. Wright, P., Moffat, A., Sinclair, I., Spearing, S.M.: High resolution tomographic imaging and
modelling of notch tip damage in a laminated composite. Compos. Sci. Technol. 70,
1444–1452 (2010)
35. Wright, P., Moffat, A., Renault, A., Sinclair, I., Spearing, S.M.: High resolution computed
tomography for modelling laminate damage. In: 17th International Conference on Composite
Materials, Edinburgh, UK (2009)
36. Garcea, S.C., Sinclair, I., Spearing, S.M.: Characterisation of fatigue micromechanisms in
toughened carbon fiber-polymer composites using synchrotron radiation computed tomogra-
phy. In: 16th European Conference on Composite Materials, Sevilla, Spain (2014)
37. Busignies, V., Leclerc, B., Porion, P., Evesque, P., Couarraze, G., Tchoreloff, P.: Quantitative
measurements of localized density variations in cylindrical tablets using X-ray
microtomography. Eur. J. Pharm. Biopharm. 64, 38–50 (2006)
References 529

38. Wright, P., Fu, X., Sinclair, I., Spearing, S.M.: Ultra high resolution computed tomography of
damage in notched carbon fiber—epoxy composites. J. Compos. Mater. 42, 1993–2002
(2008)
39. Sinclair, R., Preuss, M., Maire, E., Buffiere, J.Y., Bowen, P., Withers, P.J.: The effect of fibre
fractures in the bridging zone of fatigue cracked Ti-6Al-4V/SiC fibre composites. Acta
Mater. 52, 1423–1438 (2004)
40. Aroush, D.R.-B., Maire, E., Gauthier, C., Youssef, S., Cloetens, P., Wagner, H.D.: A study of
fracture of unidirectional composites using in-situ high-resolution synchrotron X-ray
microtomography. Compos. Sci. Technol. 66, 1348–1353 (2006)
41. Forsberg, F., Sj€odahl, M., Mooser, R., Hack, E., Wyss, P.: Full three-dimensional strain
measurements on wood exposed to three-point bending: analysis by use of digital volume
correlation applied to synchrotron radiation micro-computed tomography image data. Strain
46, 47–60 (2010)
42. Forsberg, F., Mooser, R., Arnold, M., Hack, E., Wyss, P.: 3D micro-scale deformations of
wood in bending: synchrotron radiation muCT data analyzed with digital volume correlation.
J. Struct. Biol. 164, 255–62 (2008)
43. Badel, P., Vidal-Sallé, E., Maire, E., Boisse, P.: Simulation and tomography analysis of
textile composite reinforcement deformation at the mesoscopic scale. Compos. Sci. Technol.
68, 2433–2440 (2008)
44. Scott, A.E., Clinch, M., Hepples, W., Kalantzis, N., Sinclair, I., Spearing, S.M.: Advanced
micro-mechanical analysis of highly loaded hybrid composite structures. In: ICCM 17—17th
International Conference on Composite Materials, Edinburgh, UK (2009)
45. Zauner, M., Keunecke, D., Mokso, R., Stampanoni, M., Niemz, P.: Synchrotron-based
tomographic microscopy (SbTM) of wood: development of a testing device and observation
of plastic deformation of uniaxially compressed Norway spruce samples. Holzforschung. 66,
(2012)
46. Stampanoni, M., Groso, A., Isenegger, A., Mikuljan, G., Chen, Q., Bertrand, A., Henein, S.,
Betemps, R., Frommherz, U., B€ohler, P., Meister, D., Lange, M., Abela, R., Boehler, P.:
Trends in synchrotron-based tomographic imaging: the SLS experience. In: Developments in
X-ray Tomography V, pp. 63180M–63180M–14 (2006)
47. Groso, A., Abela, R., Stampanoni, M.: Implementation of a fast method for high resolution
phase contrast tomography. Opt. Express. 14, 8103 (2006)
48. Maire, E., Carmona, V., Courbon, J., Ludwig, W.: Fast X-ray tomography and acoustic
emission study of damage in metals during continuous tensile tests. Acta Mater. 55,
6806–6815 (2007)
49. Baensch, F., Zauner, M., Sanabria, S.J., Sause, M.G.R., Pinzer, B.R., Brunner, A.J.,
Stampanoni, M., Niemz, P.: Damage evolution in wood: synchrotron radiation micro-
computed tomography (SRμCT) as a complementary tool for interpreting acoustic emission
(AE) behavior. Holzforschung 69(8), 1015–1025 (2015)
50. Cazaux, J., Erre, D., Mouze, D., Patat, J.M., Rondot, S., Sasov, A., Trebbia, P., Zolfaghari,
A.: Recent developments in X-ray projection microscopy and X-ray microtomography
applied to materials science. J. Phys. IV 03, 2099–2104 (1993)
51. Chotard, T.J., Boncoeur-Martel, M.P., Smith, A., Dupuy, J.P., Gault, C.: Application of X-ray
computed tomography to characterise the early hydration of calcium aluminate cement. Cem.
Concr. Compos. 25, 145–152 (2003)
52. Steppe, K., Cnudde, V., Girard, C., Lemeur, R., Cnudde, J.-P., Jacobs, P.: Use of X-ray
computed microtomography for non-invasive determination of wood anatomical character-
istics. J. Struct. Biol. 148, 11–21 (2004)
53. Trtik, P., Dual, J., Keunecke, D., Mannes, D., Niemz, P., Stähli, P., Kaestner, A., Groso, A.,
Stampanoni, M.: 3D imaging of microstructure of spruce wood. J. Struct. Biol. 159, 46–55
(2007)
530 6 Computed Tomography

54. Van den Bulcke, J., Masschaele, B., Dierick, M., Van Acker, J., Stevens, M., Van Hoorebeke,
L.: Three-dimensional imaging and analysis of infested coated wood with X-ray submicron
CT. Int. Biodeterior. Biodegrad. 61, 278–286 (2008)
55. Helliwell, J.R., Sturrock, C.J., Grayling, K.M., Tracy, S.R., Flavel, R.J., Young, I.M.,
Whalley, W.R., Mooney, S.J.: Applications of X-ray computed tomography for examining
biophysical interactions and structural development in soil systems: a review. Eur. J. Soil Sci.
64, 279–297 (2013)
56. Scott, A.E., Sinclair, I., Spearing, S.M., Thionnet, A., Bunsell, A.R.: Damage accumulation in
a carbon/epoxy composite: comparison between a multiscale model and computed tomogra-
phy experimental results. Compos. Part A Appl. Sci. Manuf. 43, 1514–1522 (2012)
57. Hu, X., Wang, L., Xu, F., Xiao, T., Zhang, Z.: In-situ observations of fractures in short carbon
fiber/epoxy composites. Carbon 67, 368–376 (2014)
58. Brault, R., Germaneau, A., Dupré, J.C., Doumalin, P., Mistou, S., Fazzini, M.: In-situ analysis
of laminated composite materials by X-ray micro-computed tomography and digital volume
correlation. Exp. Mech. 53, 1143–1151 (2013)
59. Rolland, H., Saintier, N., Robert, G.: Damage mechanisms into short glass fibre reinforced
thermoplastic during in-situ microtomographic tensile tests. In: ECCM16—16th European
Conference on Composite Materials, Sevilla, Spain (2014).
60. Kalafat, S., Zelenyak, A.-M., Sause, M.G.R.: In-situ monitoring of composite failure by
computing tomography and acoustic emission. In: 20th International Conference on Com-
posite Materials, Copenhagen, Denmark, pp. 1–8 (2015)
61. Borstnar, G., Mavrogordato, M.N., Sinclair, I., Spearing, S.M.: Micro-mechanistic analysis
of in-situ crack growth in toughened carbon/epoxy laminates to develop micro-mechanical
fracture models. In: ECCM16—16th European Conference on Composite Materials, Sevilla,
Spain (2014)
62. Silva, F.A., Williams, J.J., Chawla, N.: 3D Microstructure Visualization of SiC Particle
Reinforced Al Matrix Composites by X-ray Synchrotron Tomography. In: ICCM 17—17th
International Conference on Composite Materials, Edinburgh, UK (2009)
63. Borstnar, G., Mavrogordato, M.N., Helfen, L., Sinclair, I., Spearing, S.M.: Interlaminar
fracture micro-mechanisms in toughened carbon fibre reinforced plastics investigated via
synchrotron radiation computed tomography and laminography. Compos. Part A Appl. Sci.
Manuf. 71, 176–183 (2015)
64. Rodrı́guez Hortalá, M., Requena, G., Seiser, B., Degischer, P., Di Michiel, M., Buslaps, T.:
3D-characterization of continuous fibre reinforced composites. In: ICCM 17—17th Interna-
tional Conference on Composite Materials, Edinburgh, UK
65. Henne, F., Ehard, S., Kollmannsberger, A., Hoeck, B., Sause, M., Drechsler, K.: Thermo-
plastic in-situ fiber placment for future solid rocket motor casing manufacturing. In: SAMPE
Europe SETEC 14—Efficient Composite Solutions to Foster Economic Growth, Tampere,
Finland (2014)
66. Sargent, J.P.: Durability studies for aerospace applications using peel and wedge tests. Int.
J. Adhes. Adhes. 25, 247–256 (2005)
67. Martiny, P., Lani, F., Kinloch, A.J., Pardoen, T.: Numerical analysis of the energy contribu-
tions in peel tests: a steady-state multilevel finite element approach. Int. J. Adhes. Adhes. 28,
222–236 (2008)
68. Ferracin, T., Landis, C.M., Delannay, F., Pardoen, T.: On the determination of the cohesive
zone properties of an adhesive layer from the analysis of the wedge-peel test. Int. J. Solids
Struct. 40, 2889–2904 (2003)
69. Khan, M.A., Mitschang, P., Schledjewski, R.: Identification of some optimal parameters to
achieve higher laminate quality through tape placement process. Adv. Polym. Technol. 29,
98–111 (2010)
70. Reiser, M.F., Semmler, W., Hricak, H.: Magnetic Resonance Tomography. Springer, Berlin
(2007)
References 531

71. Kalinina, O., Kumacheva, E.: A “core-shell” approach to producing 3D polymer


nanocomposites. Macromolecules 32, 4122–4129 (1999)
72. Zhang, Y., Ouyang, H., Lim, C.T., Ramakrishna, S., Huang, Z.-M.: Electrospinning of gelatin
fibers and gelatin/PCL composite fibrous scaffolds. J. Biomed. Mater. Res. B. Appl.
Biomater. 72, 156–165 (2005)
73. Chestnut, M.H.: Confocal microscopy of colloids. Curr. Opin. Colloid Interface Sci. 2,
158–161 (1997)
74. Franck, C., Hong, S., Maskarinec, S.A., Tirrell, D.A., Ravichandran, G.: Three-dimensional
full-field measurements of large deformations in soft materials using confocal microscopy
and digital volume correlation. Exp. Mech. 47, 427–438 (2007)
75. Germaneau, A., Doumalin, P., Dupré, J.-C.: Comparison between X-ray micro-computed
tomography and optical scanning tomography for full 3D strain measurement by digital
volume correlation. NDT E Int. 41, 407–415 (2008)
76. Bay, B.K., Smith, T.S., Fyhrie, D.P., Saad, M.: Digital volume correlation: three-dimensional
strain mapping using X-ray tomography. Exp. Mech. 39, 217–226 (1999)
77. Pan, B., Wu, D., Wang, Z.: Internal displacement and strain measurement using digital
volume correlation: a least-squares framework. Meas. Sci. Technol. 23, 045002 (2012)
78. Leclerc, H., Périé, J.N., Roux, S., Hild, F.: Voxel-scale digital volume correlation. Exp.
Mech. 51, 479–490 (2011)
79. Smith, T.S., Bay, B.K., Rashid, M.M.: Digital volume correlation including rotational
degrees of freedom during minimization. Exp. Mech. 42, 272–278 (2002)
80. Verhulp, E., van Rietbergen, B., Huiskes, R.: A three-dimensional digital image correlation
technique for strain measurements in microstructures. J. Biomech. 37, 1313–1320 (2004)
81. Lee, S.Y., Kim, G.W., Han, B.H., Cho, M.H.: Strain measurement from 3D micro-CT images
of a breast-mimicking phantom. Comput. Biol. Med. 41, 123–130 (2011)
82. Sutton, M.A., Orteu, J.J., Schreier, H.: Image Correlation for Shape, Motion and Deformation
Measurements: Basic Concepts,Theory and Applications. Springer, Berlin (2009)
83. Toda, H.: A 3D measurement procedure for internal local crack driving forces via synchro-
tron X-ray microtomography. Acta Mater. 52, 1305–1317 (2004)
84. Ferrié, E., Buffière, J.-Y., Ludwig, W., Gravouil, A., Edwards, L.: Fatigue crack propagation:
in-situ visualization using X-ray microtomography and 3D simulation using the extended
finite element method. Acta Mater. 54, 1111–1122 (2006)
85. Hoppe, H.: New quadric metric for simplifying meshes with appearance attributes. In:
VISUALIZATION’99 Proceedings of the 10th IEEE Visualization 1999 Conference (VIS
’99), Washingto, DC, USA (1999)
86. Sause, M.G.R., Zelenyak, A.-M.: Modellierung von Schallemissionsquellen auf Basis von
volumetrischen Bildinformationen. In: 20. Kolloquium Schallemission, Garmisch-
Partenkirchen, Germany, pp. 1–9 (2015)
87. Guo, Y., Nairn, J.: Calculation of J-integral and stress intensity factors using the material
point method. Comput. Model. Eng. Sci. 6, 295–308 (2004)
88. Guo, Y., Nairn, J.: Three-dimensional dynamic fracture analysis using the material point
method. Comput. Model. Eng. Sci. 16, 11–25 (2006)
89. Nairn, J.: Material point method calculations with explicit cracks. Comput. Model. Eng. Sci.
4, 649–663 (2003)
90. Guilkey, J.E., Weiss, J.A.: Implicit time integration for the material point method: quantita-
tive and algorithmic comparisons with the finite element method. Int. J. Numer. Methods
Eng. 57, 1323–1338 (2003)
91. Wie˛ckowski, Z.: The material point method in large strain engineering problems. Comput.
Methods Appl. Mech. Eng. 193, 4417–4438 (2004)
92. Wang, B., Karuppiah, V., Lu, H., Komanduri, R., Roy, S.: Two-dimensional mixed mode
crack simulation using the material point method. Mech. Adv. Mater. Struct. 12, 471–484
(2005)
532 6 Computed Tomography

93. Tan, H., Nairn, J.A.: Hierarchical, adaptive, material point method for dynamic energy
release rate calculations. Comput. Methods Appl. Mech. Eng. 191, 2123–2137 (2002)
94. Moës, N., Belytschko, T.: Extended finite element method for cohesive crack growth. Eng.
Fract. Mech. 69, 813–833 (2002)
95. Hamstad, M.A., Gillis, P.P.: Effective strain rates in low-speed uniaxial tension tests. Mater.
Res. Stand. 6, 569–573 (1966)
96. Sch€urmann, H.: Konstruieren mit Faser-Kunststoff-Verbunden. Springer, Berlin (2005)
97. Puck, A., Sch€ urmann, H.: Failure analysis of FRP laminates by means of physically based
phenomenological models. Compos. Sci. Technol. 62, 1633–1662 (2002)
98. Swolfs, Y., Verpoest, I., Gorbatikh, L.: Issues in strength models for unidirectional fibre-
reinforced composites related to Weibull distributions, fibre packings and boundary effects.
Compos. Sci. Technol. 114, 42–49 (2015)
99. Swolfs, Y., Gorbatikh, L., Verpoest, I.: Stress concentrations in hybrid unidirectional fibre-
reinforced composites with random fibre packings. Compos. Sci. Technol. 85, 10–16 (2013)
100. Swolfs, Y., Gorbatikh, L., Romanov, V., Orlova, S., Lomov, S.V., Verpoest, I.: Stress
concentrations in an impregnated fibre bundle with random fibre packing. Compos. Sci.
Technol. 74, 113–120 (2013)
Chapter 7
Combination of Methods

For the previous chapters on the methods of DIC, AE, EME, and CT the intent was
to discuss the scientific and technological basis and to provide factors of influence
as relevant in application to composite materials. However, attention was given to
the capabilities of the individual methods only. The intent of the present chapter is
to demonstrate the synergetic effects that arise by suitable combinations of the
methods reviewed intensively in Chaps. 3–6 as well as some additional methods not
extensively discussed in the previous chapters. Still the main focus is on the
application to in situ detection of failure mechanisms occurring in fiber-reinforced
composites under mechanical load.
In this context, some method combinations are well-established and are
presented with reference to relevant literature. For other combinations, recent
experimental developments have been made, which require a more detailed
presentation.
This chapter starts with a comparison of the established methods. Focus is given
to the possibilities to distinguish between different failure types as well as to the
sensitivity of the individual methods with respect to the size of the damage zone.
The important aspect of portability from laboratory environments to field applica-
tions is discussed and practical technological scaling aspects as well as physical
limitations of some of the methods are elucidated to provide an overview of
possible field applications. Subsequently several method combinations and their
applications are presented. Since a comprehensive review on this topic would be
beyond the scope of this book only representative examples are used to demonstrate
the strength of some combinations. Finally this book closes by providing an outlook
on currently ongoing developments of the experimental methods that are likely
having an impact on further applicability of the methods within the next decades.

© Springer International Publishing Switzerland 2016 533


M.G.R. Sause, In Situ Monitoring of Fiber-Reinforced Composites,
Springer Series in Materials Science 242, DOI 10.1007/978-3-319-30954-5_7
534 7 Combination of Methods

7.1 Comparison of In Situ Methods

In order to select a particular method for in situ testing of a composite material a


comparison of the individual capabilities is useful, but quite challenging. Due to the
different principles of operation involved in the detection process, the sheer amount
of different failure types occurring in a composite material under load and the
partially diffuse limits of application inhibit an entirely equivalent judgement of the
individual methods. Hence an attempt of comparison is made, which is based on
three different aspects used to provide a first overview to the reader.
As first aspect, the suitability of a method to detect a particular failure mechanism
as occurring in fiber-reinforced materials is discussed. To this end some generali-
zation of failure mechanisms as extensively discussed in Sect. 2.1 is made.
As second aspect the detection sensitivity of the individual methods for these
defects is assessed. This includes some considerations of the absolute detection
limits of the methods but also empirical limits that have been obtained from
experimental work.
As third aspect, the applicability of the individual methods to field testing is
reviewed. In this context “field testing” considers two basic implications. The first
implication is the change in size from a material specimen in the range of few
centimeters to a range of some meters as realistic for some composite material
applications. Second, the implications of outdoor situations such as temperature,
noise level, humidity, or portability are covered as opposed to classical indoor
situations.
One important aspect covered within this book is the general suitability of the
method to detect failure mechanisms in composite materials in situ. In this context
the broader meaning of “in situ” is being able to resolve the failure process as a
function of loading either online or inline assuring a certain discretization of the
load stages until final failure. Moreover, for a downselection of methods, only the
aspect of volumetric imaging is considered. Since for the evolution of failure in
composite materials, internal damage is as relevant as surface damage, only
methods are discussed in the following that have subsurface-sensing capabilities.
This basically excludes all sorts of in situ microscopy, since these attempts (despite
their high value) are limited to the detection of surface damage.
In order to facilitate the reading of the subsequent sections, the prerequisites of
the individual in situ method configurations are first of all defined in order to avoid
misunderstandings.
Digital image correlation (DIC): Speckle patterns are applied to relevant surfaces
of the investigated part. Images of the speckle pattern are recorded as a function of
loading with reasonable high speed (e.g., in the order of Hz for quasi-static load
conditions) and sufficient spatial resolution to be able to spot the relevant damage
type as discussed in Chap. 3.
Acoustic emission analysis (AE): Acoustic emission sensors with reasonable broad
bandwidth of sensitivity are mounted on the investigated part. The spacing and
position is chosen adequately so that a sufficient probability of detection for
7.1 Comparison of In Situ Methods 535

interpretation is reached and source localization is possible with an acceptable


localization error. Acquisition settings are chosen appropriately to also allow for
source discrimination as described in Chap. 4.
Electromagnetic emission analysis (EME): Electromagnetic emission sensors are
placed close to the investigated part with sufficient number to be able to detect the
release of EME. Shielding of the test configuration is appropriate to reach ade-
quately low background noise. Acquisition settings are chosen to allow for detec-
tion of EME signals following the recommendations of Chap. 5.
Passive Thermography (PTT): The camera system is positioned to monitor the
relevant surface of the investigated part. Temperature images of the surface are
recorded as a function of loading with reasonable high speed (e.g., in the order of
Hz for quasi-static load conditions) and sufficient spatial resolution to spot the heat
signatures of the relevant damage type.
Shearography (SH): The shearography probehead is positioned to monitor the
relevant surface of the investigated part. Shearography images of the surface are
recorded as a function of load at reasonable intervals to track damage evolution.
System resolution is assumed to be sufficient to spot the strain signatures of the
relevant damage type.
In situ computed tomography (CT): The investigated part is inspected in situ using a
load stage concept as described in Chap. 6. Acquisition settings of the CT-scan are
chosen adequately to reach sufficient spatial resolution to spot the intended failure.
X-Ray diffraction (XRD): The orientation of the laminate is chosen appropriately to
allow detection of the relevant failure types. The part is mounted inside a load stage
following considerations of Chap. 6 and imaging parameters are chosen adequately
to achieve sufficient spatial resolution to spot the relevant failure type.
Acousto-Ultrasonics (AU): Pulsable acoustic emission sensors with reasonable
broad bandwidth are mounted on the investigated part. The spacing and position
is chosen adequately so that a sufficient probability of detection for interpretation is
reached and source localization is possible with an acceptable localization error and
acquisition settings are appropriate to also allow for source discrimination as
described in Chap. 4. Further, the density and position of sensors is chosen to
allow for detection of the mutual pulses by a sufficient number of other sensors.
As a general rule-of-thumb for the discussions in the following subsections, all
methods are assumed to operate under technically optimal conditions. All devia-
tions to this ideal state (i.e., change of position affecting resolution) will be
specifically discussed in terms of the detectability, the loss of sensitivity, or the
limitations caused by field measurements.
536 7 Combination of Methods

7.1.1 In Situ Capabilities

A first distinction among the methods can be made with respect to their in situ
capabilities. The basic difference is the moment in time when the individual
methods provide some information on the failure progress inside the inspected
part. Distinction can be made between online, time resolved and inline. In this
context, online capabilities are understood in a way, that the method is able to
provide first results during the duration of the experiment. In contrast, time-
resolved means the method is able to track failure evolution in a timewise discrete
way, but more or less extensive post-processing of the recorded data is required
before information is accessible. As further distinction, the experiment might be
carried out in situ, but the method may need some interruption of the load process to
record data. The latter approach enables applicability of many other nondestructive
testing methods. Within the scope of this book restriction is made to in situ
computed tomography as typical example of this class of nondestructive testing
methods. A summary of the capabilities of the individual methods is given in
Table 7.1.
For DIC, the images of the speckle pattern are recorded during the measurement,
which does not interrupt the loading process. But the displacement field information
is only readily interpretable after a post-processing step, thus the method cannot be
fully considered as online method. This is not true for the application of
DIC-techniques used as optical extensometers. These can readily be applied for
strain sensing of the specimen, but make use of lower resolution and only regional
calculation of the strain components, which renders them useless in the context of
defect detection since this requires full-field measurements. For inline purposes the
DIC method is readily applicable, given a reference image at zero load is available.
AE is detected during the measurement with high temporal discretization.
It can be considered as an online method, since display of signal activity or
source localization are standards in commercial software packages providing an
almost instant view on the damage progression. Some approaches still need
post-processing steps, which reduces the speed of information to a time-resolved

Table 7.1 Categorization of Method Online Time resolved Inline


in situ capabilities of methods
DIC (x) x (x)
discussed in the following
AE x (x) –
EME x (x) –
PTT x (x) (x)
SH – x x
CT – – x
XRD x (x) x
AU x (x) x
Note: x: applicable, (x): partially applicable, – : not
applicable
7.1 Comparison of In Situ Methods 537

method in these cases. However, the method requires an active damage progression
or active friction in damaged areas to record data, so it cannot be applied to
parts subject to constant load without any damage progression or even to
dismounted parts.
Similar to AE, the detection of EME requires an active damage progression
during the measurement. Thus the method is also considered as an online method,
since commercial AE equipment is capable of detecting and processing EME
signals in a similar fashion as AE signals. For the time being, the information is
limited to activity curves and data acquisition, since no further software packages
are currently available to make use of the remaining information of the EME
signals. The further categorization follows the considerations of AE signals above.
For PTT the images are recorded during the measurement and can be displayed
fully online during the measurement. But the application of filters or background
subtractions usually involves offline steps, leaving the method as time-resolved
image series. As further advantage, an existing passive thermography setup can
easily be modified to turn into an active thermography setup by addition of an
external heat source. That way it is possible to detect the presence of damage also
inline, e.g., without applied load or during load-hold steps.
A measurement using SH is typically not performed online, since the data
processing step requires some computational intensity to turn the recorded data
into an image allowing spotting defect positions. As additional benefit, SH is also
applicable to inspect parts inline. To this end, SH operates as classical NDT method
using an external heat source to generate a small deformation or by applying a small
amount of mechanical load directly to the part.
For CT inspection in the aforementioned sense, the only applicable category is
inline inspection, since the post-processing time and scan duration is so large, the
specimen is required to be kept at a load-hold step. Of course, as classical NDT
method, CT is capable of performing scans also for parts without any applied loads.
In contrast to CT-scans, XRD inspection is carried out during the continuous
measurement of a part and allows viewing the projection images online. Similar to
the other imaging techniques it is still hard to evaluate the image content online, so
generally post-processing will yield more detailed information to examine. Because
this information may still be linked to the previously applied load this method is
fully time resolved. Finally, XRD is also capable of inline measurements, since the
X-ray diffraction constitutes an absolute quantity, which can also be evaluated
during load-holds.
AU generally comes with the same capabilities as AE since it fully includes this
method. As further aspects, the active pulsing and detection of signals contribute to
the information value and is available online, but also after post-processing. As
major distinction to AE, the possibility to actively scan the part under investigation
enables the possibility to perform a damage assessment also during load-hold steps
or even for a part completely free of external load.
538 7 Combination of Methods

7.1.2 Detectability of Failure Mechanisms

A key aspect of an inspection technology is their ability to sense a particular defect


in a material. As pointed out in Sect. 2.1, the range of failure mechanisms observed
in fiber-reinforced materials is quite complex, and it is an intuitive conclusion that
not all methods will be able to detect all sorts of failure mechanisms. In the present
section, focus is given to the aspect whether the method is by concept capable to
sense the different failure mechanisms. The detection sensitivity to each mecha-
nism will be discussed in the subsequent section.
Before starting a discussion of the individual method’s capabilities, the failure
mechanisms reported in Table 7.2 are briefly defined with reference to Sect. 2.1 for
a more detailed explanation of each type. All types of inter-fiber failure involving
fracture of the matrix material or at the interface between fiber and matrix will be
termed matrix cracking. This is regardless of the size of the crack, which can cover
a broad range from the order of few micrometers to several centimeters. Although
the failure type delamination is technically also a matrix crack, it is still distin-
guished as a different failure type in Table 7.2, due to its characteristically different
orientation. Further distinction is made between the occurrence of single filament
failure and fiber bundle failure. While the first considers the occurrence of fiber
breakage within few filaments in close spatial distance (i.e., >100 μm), the second
type considers fiber failure in the form of substantial dimensions like complete
rovings or even layerwise failure. Fiber pullout is explicitly understood as
debonding between the fiber filaments and the matrix material and is thus expected
to reside on the microscale. The phenomenon of fiber bridging ranges in size from
the microscale to the mesoscale, therefore spanning several orders of magnitude.
Main distinction to matrix cracking is due to the orientation of the crack front being
perpendicular to the fiber direction.

Table 7.2 Applicability of individual in situ methods to detect failure mechanisms and anomalies
in fiber-reinforced composites
Method Failure mechanisms Anomalies
Single
fiber Fiber
Matrix filament bundle Fiber Fiber
cracking Delamination failure failure pullout bridging Voids Undulations
DIC x x no x no ? ? ?
AE x x x x x x no no
EME x x x x ? x no no
PTT x x x x no ? (x) (x)
SH x x no x no x ? x
CT x x x x x x x x
XRD x (x) no no no (x) x (x)
AU x x x x x x ? ?
Note: x: applicable, (x): partially applicable, no: not applicable, ?: not validated but theoretically
possible
7.1 Comparison of In Situ Methods 539

Similarly, specific types of anomalies are of interest to be detected by in situ


methods, since they often act as stress concentrator, therefore may induce damage
at preferential spots. Among these anomalies, Table 7.2 distinguishes between
voids as gaseous inclusions of various size and undulations of the fiber filaments.
The latter is understood as waviness of many fiber filaments simultaneously rather
than the specific orientation of one particular fiber.
For DIC, the Chap. 3 has demonstrated the broad range of applicability to sense
the failure mechanisms marked in Table 7.2. General deficits exist in sensing of
very small defect types causing weak strain signatures. Among these are single
filament failure and fiber pullout. Hence it is not possible to detect these failure
mechanisms when recordings of macroscopic specimens are used. The aspect of
fiber bridging has not been explicitly investigated in Chap. 3, so it remains unclear
if the method is applicable to this end. However, if the crack is of a certain size,
there is no general reason, why the resulting strain signature should not be detect-
able. For voids and undulations, the applicability has also not been investigated in
Chap. 3. In this case it is questionable, whether the presence of these anomalies will
cause significant distortion of the strain field so they can be detected by DIC.
As presented throughout Chap. 4, AE is generally able to sense all sorts of active
failure types in fiber-reinforced materials. Due to operational principle being based
on active damage progression, this restricts detection of voids and undulations.
Similar considerations apply to EME detection, since this also requires an active
failure type. However, the occurrence of fiber pullout has not been explicitly
investigated by EME so it remains unclear, whether the method is suitable to this
end. Given the detectability of the other failure mechanisms, there is no general
physical limitation why this should not be detectable. But the current experimental
setups still suffer from several drawbacks with respect to the sensitivity when
compared to AE (cf. Sect. 5.5).
For PTT imaging, the image resolution and the strength of the heat signature are
the relevant quantities to discuss regarding the detectability. As demonstrated in
Sect. 7.2, PTT is capable to detect the heat signatures of single filament failure as
well as several sorts of matrix cracking, including delamination and possibly fiber
bridging. However, the occurrence of failure types with minimal heat release such
as fiber pullout is quite unlikely and therefore assumed to be generally not appli-
cable. Given the modification to active thermography as described in the previous
subsection, the method is also in principle capable of detecting anomalies like voids
or undulations.
For the suitability of SH, similar considerations as for DIC apply. Although the
strain sensing capabilities of SH are superior to DIC, there is currently no likely
solution to detect single filament failure or fiber pull out by means of SH. For the
anomaly types, SH has been reported to be applicable for detection of undulations
[1], but there are no indications in literature, that SH testing may be used for the
detection of voids.
Using appropriate settings as high-resolution configuration, there is no technical
limitation for CT to spot the failure types and anomalies listed in Table 7.2. The
general limitation for CT is the detail visibility defined by the chosen measurement
540 7 Combination of Methods

configuration and the specimen size as discussed in detail in Chap. 6 and will be
emphasized in more detail in the following subsection.
For the measurement setup using XRD, a particular focus is given to the
detection of matrix cracks in the form of inter-fiber cracks. Hence this method is
specifically developed for that purpose making it insensitive for other failure
mechanisms. There is limited applicability to delamination and fiber bridging, in
case the orientation of the incident beam is chosen accordingly. In addition, the
XRD method provides sensitivity by means of absorption measurements to quantify
voids and some limited applications to quantify undulations by means of diffraction
pattern analysis.
As a further method considered herein, AU again includes all sensing capabil-
ities of AE, since it fully includes this method. As extension to the aspects
mentioned before AU is in principle applicable to detect undulations or voids,
since both anomalies will cause a local change of the acoustic properties. However
this has not become a standard application, yet.

7.1.3 Detection Sensitivity

Besides the general applicability of an in situ method to detect a particular type of


defect, the question of detection sensitivity arises. Certainly some of the presented
methods might be able to spot a particular type of damage given some boundary
conditions are fulfilled, but might easily fail to detect the same type of damage with
smaller dimensions. Thus for practical implications it is worthwhile to compare the
methods with respect to their sensitivity to detect a given type of failure mechanism
or anomaly. In order to allow a fair comparison, three conditions are assumed in the
following:
1. The method generally operates under in situ conditions as discussed in
Sect. 7.1.1.
2. The method is subject to ideal operating conditions as outlined in the beginning
of Sect. 7.1.
3. The detectable size of the anomaly or failure mechanism is used as a benchmark
to compare methods.
Likewise, if the method requires any modifications to the above conditions (e.g.,
to move from online to time-resolved in order to perform the operation) the
sensitivity of the method is generally ranked lower than without any modifications.
For DIC, the possibility to detect failure mechanisms has been discussed inten-
sively in Chap. 3. As concluded from the experimental investigations and the
modelling, there is a certain limitation in detection sensitivity as soon as the damage
is located deep enough below the surface. Similarly, small defect sizes inhibit
detection of particular failure types, since they may easily fall below the detection
limit. For some the anomalies listed in Table 7.3 the method might be applicable,
but has not been tested so far.
7.1 Comparison of In Situ Methods 541

Table 7.3 Detection sensitivity of individual in situ methods for failure mechanisms and anom-
alies in fiber-reinforced composites
Method Failure mechanisms Anomalies
Single Fiber
Matrix filament bundle Fiber Fiber
cracking Delamination failure failure pullout bridging Voids Undulations
DIC x x no x no (x) (x) (x)
AE xxx xxx xx xxx xx xxx no no
EME xx xx x xx (x) xx no no
PTT x x x x no (x) x x
SH xx xx no xx no xx (x) x
CT xx xx xx xx x xx xxx xx
XRD xx x no no no x xx (x)
AU xxx xxx xx xxx xx xxx (x) (x)
Note: x: working, xx: sensitive, xxx: very sensitive, (x): not validated but theoretically possible,
no: not applicable

The method of AE testing has been investigated for many decades and the
current technical limitations have been extensively discussed in Chap. 4. In general,
AE has extraordinary benefits in terms of the detection sensitivity, because the
sensing systems are able to detect active failure mechanisms covering several
orders of magnitude in size (cf. Sect. 4.2). Some limitations occur for failure
mechanisms being very small in size (e.g., single filament failure) because of
attenuation effects, the respective AE signal may fall below the detection threshold
after short distances of wave propagation. For passive defects, such as voids or
undulations, AE is not applicable.
Being similar to AE testing, EME is generally able to detect the same types of
failure mechanisms as AE. But the distance between sensing system and source
position has a strong impact on the detection system, which currently turns the
detectability into a unique assessment for each setup. Also, the present detection
systems are not standardized or readily available from commercial companies
turning their use into a very individual configuration for every experiment. Thus,
the method is ranked in a similar fashion as AE testing but with generally lower
detection sensitivity. However, as presented in Chap. 5, for a particular configura-
tion the EME systems may show similar detection sensitivity as an AE system.
In PTT imaging, the detection sensitivity is dominated by the resolution of the
images and the thermal resolution of the camera and the heat signature intensity as
compared to the thermal background. For the camera resolution, the present
technology of thermography cameras does not show as high of a resolution as
conventional systems. Accordingly, detail visibility is generally less than for
imaging methods operating in the visual spectrum. The thermal resolution of
current camera systems was found to be suitable to detect transient heat signatures
as occurring during loading of fiber-reinforced materials. However, the main
challenge arises from the discrimination of these heat signatures relative to the
thermal background. First, the material generally heats up during mechanical
542 7 Combination of Methods

loading (much more pronounced for cyclic loading). With increasingly strong heat
signatures at already damaged zones (i.e., due to internal friction) it becomes
impossible to detect superimposed transient heat signatures indicating further
failure. Second, the heat signature itself needs to be strong enough to be discrim-
inated against the noisy thermal background that exists for fiber-reinforced mate-
rials. For the cases studied in Sect. 7.2.3 it was found, that only very strong acoustic
emission sources could be related to such heat signatures. Thus the method may
generally be assumed to be less sensitive to a majority of the more microscopic
failure mechanisms occurring in the material. For the branch of active thermogra-
phy however, the method has some sensitivity to detect the occurrence of voids or
the local change in heat flux due to undulations, but again is not the most sensitive
method among the list in Table 7.3.
The detection of failure mechanisms using SH has seen substantial applications
in conventional nondestructive testing. Since the method is readily transferable to
in situ applications, some of the general conclusions on detection sensitivity of SH
testing are still valid. The method is well suited to detect these failure mechanisms
which are indicated by strong out-of-plane deformation signatures. In contrast to
DIC, the method generally benefits from a higher detection sensitivity in these
cases. But for failure mechanism being small in size or located below the surface,
the same limitations arise. Here the method in its present technological state is not
assumed to be suitable to detect the occurrence of single filament failure or likewise
dimensions of the other failure mechanisms. For the anomalies, SH testing has seen
some applications to visualize undulations with certain sensitivity [1] but it is
questionable, whether the systems are ready to detect the presence of voids.
Volumetric imaging using CT generally benefits from a high level of detail
visibility. Thus, given suitable scan duration and sufficiently small volumes of the
material, the method is able to detect all failure mechanisms and anomalies with
very high detection sensitivity (cf. Sect. 6.2). In the context of in situ testing,
general drawbacks originate from the lack of full online capabilities and the
limitations in scanned volume. Whenever the limits of the ideal operating condi-
tions are approached, the method loses detection sensitivity to small failure mech-
anisms and anomalies very quickly. Thus the overall ranking in Table 7.3 is lower
than it would be for a CT system operating in standard scan operations, i.e., without
the specific implications of the in situ mode with superimposed mechanical or
thermal loads.
For XRD only a subset of the failure mechanisms listed in Table 7.3 are
detectable. The method has been specifically adapted to detect the occurrence of
matrix cracking or delamination (and also fiber bridging if large enough) and shows
reasonable sensitivity for these cases. Due to the nature of diffraction techniques,
very precise alignment of the apparatus and the specimen is required with strong
reduction in the detection capabilities when the limits of the ideal operating
conditions are approached. For detection of voids, XRD generally has a high
sensitivity because of the change in X-ray attenuation. Similarly the change in
fiber orientation due to undulation causes noticeable changes in the diffraction
spectrum and could be used to detect the presence of undulations to a certain extent.
7.1 Comparison of In Situ Methods 543

AU testing extends the detection sensitivity of the AE method to detect the


presence of anomalies. Since the presence of voids or undulations generally may
influence the guided wave propagation, the method may show some sensitivity to
detect the presence of either of these defects. Although the interaction of guided
waves with internal damage as discussed in Sect. 4.3 may result in measurable
quantities it is certainly not most sensitive to the occurrence of these two types of
anomalies. Hence, it currently remains an open question, whether the AU method is
able to sense these anomalies with reasonable accuracy.

7.1.4 Extension of Methods to Large Scale and Field Testing

One further challenge for all test methods is the harsh change in operating condi-
tions when moving from a laboratory environment to an application in the field. In
the context of the in situ testing approach followed within this book, the term “field
testing” needs some further specification prior to the discussion.
In the following field, testing is understood to apply to all kinds of testing
applications that leave the well-defined operating conditions as outlined in the
beginning of Sect. 7.1. This obviously includes all sorts of outside inspection of
fiber-reinforced material parts and thus includes all aspects of harsh climates or
remote inspection sites. But for many methods, strong limitations arise long before
reaching these rather severe inspection conditions. Many methods suffer from
strong drawbacks, when the size of the part to be inspected is already larger than
a few centimeters, but the part being tested still resides within a laboratory with
standardized climate. Other limitations arise due to typical noise sources as being
present for large-scale testing or outside conditions. Further, some specific limita-
tions, which may or may not exist within a laboratory environment or in outside
conditions will be discussed for each method. Consequently, the aim of the present
section is to discuss the key limitations of each of the methods in their ability to
inspect larger parts made from fiber-reinforced materials, potentially in an outside
inspection scenario as summarized in Table 7.4.
The method of DIC is generally applicable for field-testing applications, since
the experimental setup basically consists of a camera system recording a speckle-
type pattern. Challenges in terms of failure detection arise from the loss of sensi-
tivity, when investigating larger objects. As pointed out in Sect. 3.2, the resolution
of DIC depends on the field of view which in turn defines the image resolution.
Thus monitoring of a larger object with only one camera system causes a strong
loss of detection sensitivity when compared to monitoring of a smaller object.
As a consequence, one could use multiple camera systems, or if attempts are made
to solely measure the macroscopic displacement and deformation of a structural
part, a full monitoring approach by one camera system may still be feasible. But
many of the other analysis types proposed in Chap. 3 will lose their significance
beyond a certain spatial resolution. Therefore, a scale-up to monitor failure types as
outlined in Sect. 3.3 is not assumed to be successful beyond an imaged object size
544 7 Combination of Methods

Table 7.4 Applicability of individual in situ methods in the field


External noise
Method Applicability Sensitivity sources Other relevant limitations
DIC x Scales with Lighting condi- Visibility of specific defects
field of view tions, structure not feasible beyond a certain
movement resolution, large amount of
data for long-term monitoring
AE x Density of Extraneous noise, Acoustic signal spreading,
network vibrations and attenuation, signal
EMI signals interpretation
EME ? Density of EMI signals, Housing in shielding chamber
network vibrations
PTT x Scales with Transient or con- Visibility of specific defects
field of view stant heat sources, not feasible beyond a certain
turbulences resolution, large amount of
data for long-term monitoring
SH x Constant for Vibrations, heat Laser handling
one system sources
CT (x) Scales with Vibrations, heat Closed system, X-Ray han-
field of view sources dling, large amount of data for
long-term monitoring
XRD ? Scales with Vibrations Geometric alignment, X-Ray
field of view handling
AU x Density of Extraneous noise, Acoustic signal spreading,
network vibrations and attenuation, change of material
EMI properties and dimension, sig-
nal interpretation
Note: x: applicable, (x): partially applicable, ?: not validated but theoretically possible

of few cm2. If critical spots in structural parts are known, one attempt is to directly
observe these areas instead of monitoring the whole part.
Among the external noise sources, a sudden or periodic change in lighting
conditions may be named first. This happens frequently in outside conditions, but
may also happen in laboratory environments if no constant lighting conditions (e.g.,
closing sun-flaps) can be assured during the test. The consequence may be useless
sequences of images or falsified quantitative displacement and strain information.
As second important noise source, an absolute movement or vibration of the
structure may occur. Since the cameras field of view is usually fixed, a large
displacement or rotation of a structure might move the relevant part to be imaged
outside the cameras field of view. In particular for large-scale testing, some parts
may undergo strong deformation before the first occurrence of detectable damage.
Thus the camera position may need to be adjusted previously, so the relevant part of
the structure is within the field of view after reaching this critical deformation stage.
Finally, one general challenge in long-term testing is the amount of data collected.
At constant image size, the major impact on the amount of data collected is the
acquisition rate. For a long-term monitoring operation, the acquisition rates may
7.1 Comparison of In Situ Methods 545

easily fall in the ranges below 0.1 Hz causing less chance to track the details of
damage evolution within the monitored structure.
AE testing has a long tradition as a method being applied in the field to monitor
fiber-reinforced parts [2]. This is due to several beneficial aspects as has been
reviewed so far. The loss of sensitivity of one AE sensor may be compensated by
adding more sensors to the same structure until a certain density of the AE sensor
network is reached. In modern equipment it is possible to record signals from
several channels simultaneously, thus the sensitivity could in principle be kept at
a similar level as in laboratory testing, given the operator is willing to mount the
required number of sensors to the structure.
In large-scale testing and in outside monitoring applications a variety of external
noise sources may occur which will reduce the effective sensitivity, since the
acquisition settings may need to be modified. In this context, the occurrence of
extraneous acoustic noise is the most relevant source. In some test configurations,
the acoustic noise floor may easily render any attempts by AE monitoring useless.
Potential reasons are manifold, but among the typical sources are all sorts of friction
(e.g., between inspected part and supports or fixtures), valves or other parts of
hydraulic equipment as well as rain drops, snow, or hail. If technically possible, the
potential noise sources may be removed from the structural part (e.g., mounting
valves offset to the structure) or lubricants may be used to minimize stick–slip
friction. However some of the noise sources may not be eliminated leaving the
operator with the only chance to decrease the sensitivity of the equipment, so to
acquire only those AE signals being stronger than the background noise. As a
further type of external noise source, the equipment may detect electromagnetically
induced (EMI) signals. These may originate from any electrical devices operating
nearby and are typically detected because of insufficient shielding of the AE
measurement chain. An improvement of the electrical shielding of the measurement
equipment or spatial decoupling of the potential noise source may be used to avoid
detection of EMI signals. Beyond the external noise sources, some further limita-
tions of the AE method arise when moving from small-scale testing to large-scale
testing. First, the aspects of signal propagation in larger structures need to be
carefully considered. Thus either the same source-sensor distance as in a small-
scale test is required, or wave propagation and attenuation will have an adverse
effect on the information included in the detected AE signals. Also, the wave
propagation in structural parts may generally be more complicated as shown in
the simple cases of Sect. 4.3. Therefore a careful revision of the approaches to
perform source localization and source classification are required when being
applied to a test object of larger size.
The method of EME testing has so far not seen any application to testing of
larger parts or outside applications. Instead, the method is still under development
and is hence discussed with an optimistic view on potential application in such
scenarios. Similar to AE testing, the sensitivity of the method is a function of the
number of sensors applied during the test. Therefore the same limitations in terms
of the network density arise as for AE systems. However, since the overall sensi-
tivity of EME sensors is much less than for commercial AE sensors and the
546 7 Combination of Methods

maximum source-sensor distance is still very small (very few centimeters) a


comparatively dense sensor network would be required for monitoring of a larger
part. Further, a multitude of external noise sources exist. As demonstrated in
Sect. 5.4, one of the key requirements to perform EME measurements is to operate
within a shielding chamber. This is to reduce the electromagnetic background to a
level, where EME signatures are not fully hidden and are thus detectable. The
scaling of such shielding chambers as presented in Chap. 5 to test scenarios of
larger parts is generally challenging, but technically possible. Commercially avail-
able walk-in shielding chambers have been tested with present EME equipment and
were found to show sufficient shielding capabilities from external electromagnetic
background. Another potential challenge in such testing scenarios is seen by
structural vibrations of the test rig or the structure. Since the EME signals are
generated by movement of charges, there is no distinction possible between charge
movements due to active defects (e.g., vibration of the crack wall) or due to the
vibration of the material itself. It may even be expected that the macroscopic
movement or vibration of the structure will form the dominant EME signature
and therefore may cover the EME signatures resulting from damage formation.
Testing with PTT is done in a variety of field applications, so there is general
applicability of the method despite of some general limitations of the equipment
concerning outside applications. As being an imaging technique, for the sensitivity
of the method the same considerations as outlined for DIC apply. Thus, the
sensitivity to detect details will suffer from an increasing field of view. In addition
to these limitations, the temperature information may also be subject to changes
when moving away from the test object. Any turbulence in the air may cause visible
artifacts and may cause difficulties to spot tiny heat signatures as relevant to
damage detection. However, the overall loss of sensitivity due to scaling is not as
severe as for DIC, so this imaging technique may allow monitoring of larger parts,
while keeping reasonable detection sensitivity.
Among the external noise sources the occurrence of additional heat sources is
the most severe. If the part to be tested is exposed to sunshine or if the mechanical
actuator dissipates its heat to the tested part then additional heat backgrounds are
superimposed and may mask transient heat signatures. Also artificial lights may
cause reflections visible in the thermal images and may cause zones of the part not
able to be inspected. The latter is even more relevant if the inspected parts are
subject to larger displacements or rotations, since this might cause additional
movement of reflection spots on the structure. Especially for outside applications,
movement of the air between the test object and the camera system as well as the
local cooling due to the air flow may cause additional artifacts in the camera
images. Some of these effects may be avoided by shielding the test object against
any external heat source and moving artificial light sources to avoid reflections
being visible within the camera images. For monitoring applications the image
acquisition rate needs to be considered further, since this is the key quantity to
define the overall amount of data being recorded.
As a further method to be considered for field applications, SH testing is among
the methods being frequently applied in such conditions. The systems are typically
7.1 Comparison of In Situ Methods 547

used as handheld devices able to scan several cm2 at once or as specific large-scale
systems to allow inspection of larger structures at once. The general benefit of these
systems is that their detection sensitivity is constant for one system, since this is
more or less fixed in its distance to the test object. Among the various external noise
sources, the occurrence of any superimposed displacement is one of the limitations
of the systems. Thus additional vibrations or transient heat sources causing defor-
mation due to thermal expansion may need specific considerations of the operator to
distinguish image artifacts from defect indications. An additional aspect to be
considered for the operation of shearography devices is the protective measures
required to allow laser operation. Since the interference of laser-based speckle
patterns forms the basis of this method, potentially hazardous reflections of the
projected area need to be shielded to avoid any health risks.
Considering CT for a field application comes with some conceptual barriers that
need to be overcome. For currently existing systems, the size of the scanned object
is limited by the size of the CT-scanner. While nowadays some large-scale
scanners are developed which allow scanning of objects in the meters range, the
general principle of shadow microscopy is not overcome by those attempts. Thus
one always needs to consider the trade-off between the scanned volume and the
level of detail to be detected. Also, the application of in situ stages is an emerging
field and so far no test rigs have been reported in literature, which are located
beyond the range of laboratory test rigs (i.e., 250 kN machines). However, there is
no existing limitation in extending the in situ test concept to even larger scales.
Among the limitations of the method as far as external noise sources are
concerned, a majority of the typical image artifacts have been shown in
Sect. 6.2. Specifically for a large-scale test, the lack of penetration of the test
object as well as geometry-related image artifacts are likely to occur. Further,
motion of the test object may originate from structural vibrations or thermal
expansion due to localized heat sources. Most of these effects (including stress-
relaxation) will cause visible artifacts and therefore reduce the visibility of details
in the volumetric images. Since CT using X-rays is the current technological
standard, suitable protective measures are required. X-ray CT-scanners usually
operate in sealed rooms or chambers to avoid any hazardous X-ray dose for the
operator. Any testing equipment and the test object itself needs to be contained
inside such a space. This obviously limits any attempts to perform such operation
in remote site inspection scenarios. Another relevant limitation is the amount of
data being recorded for one volumetric scan. This amount of data is much more
challenging than for the 2D imaging methods discussed so far. Since one volume
scan typically consists of several hundred to thousand projection images, the data
amount increases accordingly for just one volumetric image. On the positive side
the amount of volumetric images taken during one experiment is still compara-
tively low due to the long scan durations.
Although the principle of XRD has been known for decades, there have been no
substantial attempts to bring the method into a large-scale or field-testing applica-
tion. In contrast to conventional X-ray inspection, which is used in remote inspec-
tion, the major drawback is the precise alignment of XRD device and specimen.
Whilst this is also a requirement for X-ray-based CT application, the applications
548 7 Combination of Methods

using XRD in combination with inspection of fiber-reinforced materials may seem


to be too limited to enforce further development of the technique beyond the
laboratory scale. This is certainly to a large portion also caused by the required
protective measures to avoid exposure to X-rays and thus causes similar limitations
as discussed for CT-scanning.
The method of AU has already found its way into field applications as it
complements the AE technology by an interesting further aspect. As such, the
inspection of fiber-reinforced materials by guided waves alone is also carried out
frequently and comes with a couple of further challenges in addition to those
already mentioned in the AE description above. For the aspect of guided wave
inspection alone, the density of the sensor network will also define the overall
inspection sensitivity of the method. If the distance between sensors is chosen too
large, the guided waves excited by one actuator may not be detected by another
sensor. However, the average distance from actuator to sensor is allowed to be
much larger than for AE operation alone, since the excited guided waves are
typically of much larger amplitude and hence may propagate much longer distances
before falling below the detection threshold.
Because the mode of propagation in fiber-reinforced materials in AE and AU is
typically a guided wave, the same limitations in terms of wave propagation apply as
discussed previously. Hence, the changes in signatures of the guided waves may
need careful revision when brought from geometrically simple structures (such as
material test specimens) to more complex structural parts. For operation in outside
conditions, additional factors, such as changes in material properties due to
temperature changes as well as dimensional changes due to thermal expansion or
mechanical loads may need to be considered.

7.2 Established Method Combinations

While the previous section had a strong focus on the strength and weakness of each
individual method, one clear strategy is to combine methods in order to reduce
some of the weaknesses of the single method. To this end, some particular method
combinations are discussed in the following subsections to provide an overview of
established combinations and to demonstrate their application to fiber-reinforced
composites. In general, the fusion of information obtained by individual methods
has become an interesting strategy to yield improved insight into the processes
occurring in materials during failure [3–7].

7.2.1 Imaging Methods and Acoustic Emission

One conceptual drawback of acoustic emission analysis is the lack of imaging


capabilities with respect to defects not being active. Thus acoustic emission signals
only appear as a transient type of information at the moment, when the source is
7.2 Established Method Combinations 549

active. Using sensor arrays and localization algorithms as presented in Sect. 4.6 the
detected acoustic emission signals can be used to calculate the spatial location of
the source. Depending on the number of sensors, this may result in 1D, 2D, or 3D
source coordinates, which visualize the damage progression. However, the locali-
zation result itself is often difficult to interpret without an additional visual obser-
vation of the specimen.
Imaging methods benefit from their ability to capture damage progression in a
visual sense, but often lack sensitivity to detect the occurrence of damage and may
not cover several orders of length scale in observation. Therefore skilled decision
needs to be made, when selecting the type of resolution and the field of view. But
the combination of both methods is able to compensate some of the drawbacks
associated with the individual methods.
Consider the example of a tensile test of a unidirectional T700/PPS material seen
in Fig. 7.1. Here, the localized source position in x-direction between the two
acoustic emission sensors is shown as false-color density scale in addition to the
simultaneously recorded images of the tensile test. Due to the sensor arrangement,
no 2D source localization is possible in this test configuration. As consequence, the
source positions may not be attributed to particular (x, y) coordinates using solely
acoustic emission measurement. The corresponding images reveal the occurrence
of fiber filament and fiber bundle failure at distinct load levels. Simultaneously
recorded acoustic emission signals localized at the respective x-coordinate of the
observed failure can therefore be attributed to these particular locations.
Also, the ability to have the full image of the specimen in its present damaged
state as reference provides better understanding of the formation and densification
of acoustic emission source positions at specific spots. Hence, weak spots in the
material may be spotted by acoustic emission analysis very early and may be

Fig. 7.1 Localized AE source positions during tensile test of unidirectional T700/PPS specimen
synchronized with camera images at two distinct load levels
550 7 Combination of Methods

a strain markers b
AE source
density
x high

medium
y
low

specimen
Fiber orientation
sensor positions 12.5 mm

Fig. 7.2 Localized AE source positions during V-notched rail shear test of unidirectional T700/
PPS specimen superimposed to camera images

interpreted as formation of the according damage visible in the recorded images at


later stages of the experiment.
Another example of combination between acoustic emission measurements and
imaging is shown in Fig. 7.2. Here V-notched rail shear testing of an unidirectional
T700/PPS laminate was monitored by four acoustic emission sensors to allow
source localization in 2D as seen in Fig. 7.2a. The experiment was carried out
inside a temperature chamber at 160 C and images of the specimen were taken at
discrete time intervals to allow visual monitoring of the damage progression.
As seen by the superposition of localized (x, y) source positions to the deformed
specimen, this allows visualizing the damage progression within the material.
As seen from the image in Fig. 7.2b, the majority of the acoustic emission sources
start to develop on one side of the specimen and shift in position in accordance with
the specimen deformation, the latter being caused by severe inter-ply failure and
several further failure mechanisms occurring at the notch positions.
Such superposition of the acoustic emission source localization on recorded
images of the test structure was found to be of particular use for larger test objects.
Here structurally weak spots are easy to identify by early onsets of acoustic
emission signals and are much easier to correlate to the relevant part of the
geometry when put into the same coordinate system and merged together to yield
one sequence of images to interpret as a function of load.
Another natural application of the combination between imaging and acoustic
emission is the field of micromechanics. Specifically developed micromechanical
experiments allow a direct microscopic observation of the acoustic emission
source. As recently presented in [8], this allows to generate particular source
types and allows to benchmark modelling approaches in their ability to predict
acoustic emission signals in a quantitative way. As seen in Fig. 7.3, the core of the
7.2 Established Method Combinations 551

force F

39 mm

tensile bar
aluminum pin

z piezoelectric elements
y
x WD sensor
fixed constraint

Fig. 7.3 3D-model of the experimental setup including cross-sectional view of acoustic emission
sensor used (reprinted from [8])

micromechanics test stage consists of an aluminum block with dimensions of


120 mm  120 mm  39 mm (length  width  height). At the center of the top
side of the aluminum block, a pin with 2 mm diameter and 1 mm height was
machined out of the block. At the opposite side of the block, an acoustic emission
sensor is positioned at the center. In order to ensure reproducible mounting condi-
tions, a spring system to firmly press the sensor to the aluminum block is used (not
shown in Fig. 7.3).
For the presented cases, the first part of the preparation is to apply a small droplet
of epoxy resin on top of the aluminum pin. To facilitate the positioning of the epoxy
resin droplet a small depression was machined into the end of the aluminum pin.
To prepare the micromechanical stage for generation of a matrix crack a tensile bar
made from polyether ether ketone (PEEK) of 2 mm diameter and 80 mm length is
used. The tensile bar is first moved into the liquid epoxy resin and then retracted to
yield a tapered contour of 380–800 μm (see Fig. 7.4). Subsequently, curing of the
liquid resin is carried out using heating foils attached to the aluminum block and an
additional heating sleeve wrapped around a small cylinder covering the aluminum
pin and parts of the tensile bar. A curing cycle comprising a heating rate of 2 C/min
up to the curing temperature of 180 C is chosen. The curing temperature is kept
constant for 150 min with subsequent cooling to room temperature at a rate between
0.5 and 2 C/min. Because of its low thermal conductivity, the tensile bar made
from PEEK minimizes the dissipative heat flux and therefore assures a constant
temperature of the resin during curing. The tensile bar and the aluminum block are
mounted in a universal test machine so that thermal expansion of the components
and chemical shrinkage of the resin can be compensated by a closed loop force
control. The test machine control adjusts the tensile bar position to guarantee zero
552 7 Combination of Methods

Fig. 7.4 Microscopy images of the failure mechanisms investigated: matrix cracking (left) and
fiber breakage (right) (reprinted from [8])

Table 7.5 Acquisition Acquisition settings Value


settings used for
Preamplification 20 dBAE/40 dBAE
micromechanics experiments
Threshold 45 dBAE
Triggering Individual channel
Acquisition rate 40 MS/s
Band-pass range 20 kHz to 1 MHz
Couplant Apiezon-L grease
Mounting system Spring loaded
Sensor type WD
Number of sensors 1

force acting during curing, which is necessary to avoid excessive forces acting on
the test specimen, potentially causing preliminary failure due to thermal expansion
and cure shrinkage of the resin.
To prepare fiber breakage, a two-component epoxy is used to bond a HTA
carbon fiber to the end of a flat-topped tensile bar made from PEEK. The fiber is
then moved into the resin droplet utilizing a micrometer stage and an optical
microscope. The free fiber length was between 350 and 450 μm. The embed length
was chosen larger than 100 μm to reach fiber breakage before fiber pullout occurs.
After embedding, the resin droplet is cured in situ.
After preparation of the test geometry, the universal test machine applies a
tensile force using a displacement controlled mode with velocities dependent on
the selected failure mechanism. A velocity of 20 μm/min for fiber breakage and
50 μm/min for matrix cracking was found to be suitable. The occurrence of failure
was monitored by an optical microscope with a magnification factor of 100.
The images obtained after failure are shown in Fig. 7.4 for the two failure types.
The respective acoustic emission acquisition settings are reported in Table 7.5.
Having such direct microscopic observation of the failure mechanism allows
performing numerical simulation following the approach reported in [8], which has
been outlined in Sect. 4.2. The modelled signals are the result from a full 3D
7.2 Established Method Combinations 553

computation including piezoelectric conversion within the sensor model along with
application of a P-Spice circuit simulation and subsequent band-pass filtering. After
amplification of the modelled signals by 20 or 40 dBAE in accordance with the
experimental settings, this allows for direct comparison of the acoustic emission
signal amplitudes in the voltage scale.
For the case of fiber breakage, the comparison is found in Fig. 7.5a, b. As seen
from the voltage scale and the time-frequency signature given in the Choi-Williams
distribution there is very good agreement between the modelled and the

a
1200 0.0
2.6E+05
1000 5.2E+05
frequency [kHz]

7.8E+05
800 1.0E+06
1.3E+06
600 1.6E+06
1.8E+06
400 2.1E+06
2.3E+06
200

0
5 10 15 20 25 30 35
amplitude [V]

4
2
0
–2
–4
5 10 15 20 25 30 35
time [μs]
b
1200 0.0
2.6E+05
1000 5.2E+05
frequency [kHz]

7.8E+05
800 1.0E+06
1.3E+06
600 1.6E+06
1.8E+06
400 2.1E+06
2.3E+06
200

0
5 10 15 20 25 30 35
amplitude [V]

4
2
0
–2
–4
5 10 15 20 25 30 35
time [μs]

Fig. 7.5 Comparison between simulated (a), (c), and experimental (b), (d) results of fiber
breakage and matrix cracking, respectively (reprinted from [8])
554 7 Combination of Methods

c
1200 0.0
2.6E+05
frequency [kHz] 1000 5.2E+05
7.8E+05
800 1.0E+06
1.3E+06
600 1.6E+06
1.8E+06
400 2.1E+06
2.3E+06
200

0
5 10 15 20 25 30 35
amplitude [V]

4
2
0
–2
–4
5 10 15 20 25 30 35
time [μs]
d
1200 0.0
2.6E+05
1000 5.2E+05
frequency [kHz]

7.8E+05
800 1.0E+06
1.3E+06
600 1.6E+06
1.8E+06
400 2.1E+06
2.3E+06
200

0
5 10 15 20 25 30 35
amplitude [V]

4
2
0
–2
–4
5 10 15 20 25 30 35
time [μs]
Fig. 7.5 (continued)

experimental signal. In particular, the signal amplitudes show almost identical peak
values and the echoes of the initial pulse are adequately captured.
Also for the matrix cracking case shown in Fig. 7.5c, d there is very good
agreement in the voltage scales and the time-frequency signature seen in the
Choi-Williams distribution. Slight differences arise in the modelled signal after
t ¼ 10 μs. This is due to the repetitive approach-retract cycles of the newly formed
fracture surface. In the modelling part, those fracture planes are partially restricted
in their relative motion because of the selected symmetry plane. Therefore their
7.2 Established Method Combinations 555

dominant movement direction is in the z-direction. In the experimental part, the


fracture surface might experience additional sliding and torsional motions as well
as additional interlocking of rough surface parts, which may account for the
smoother appearance of the spikes in the modelled signals.
The combination of imaging methods and acoustic emission analysis allows to
visually examine particular source types and to directly correlate the respective AE
signals. Since imaging methods in this sense are restricted to surface observations,
their results may be error-prone when compared to a volumetric analysis method
such as acoustic emission. But for many applications, the correspondence is unique
and allows safe concluding on the occurrence of particular failure modes in fiber-
reinforced materials.

7.2.2 Digital Image Correlation and Acoustic Emission

As an extension of the combination of information outlined in the previous section,


the results of digital image correlation may also be merged with results from
acoustic emission measurements. Similar to simple imaging of the test object,
this concept has recently emerged and has already been applied for various config-
urations and materials [3, 4, 9–13]. When compared to simple imaging of the
specimen, DIC keeps the beneficial aspect of being an imaging technique, but
certainly facilitates the detection of damage locations due to visible strain concen-
trations as described in Sects. 3.3 and 3.4. Beyond the initial comparison made in
Sect. 4.7.2.3, the following example is used to elucidate the link between both
methods.
To this end, tensile tests of the Sigrafil CE1250-230-39 cross-ply specimens of
Sect. 4.7.3.2 were simultaneously monitored by acoustic emission and DIC using
the acquisition parameters of Tables 4.2 and 3.7, respectively. Here the focus is on
the formation and growth of inter-fiber cracks in the 90 plies of the [0/90]sym,
[(0/90)2]sym and [(0/90)3]sym stacking sequences. Based on the considerations in
Chap. 3 such transverse cracks should be detectable to a certain extent below
the specimen surface. Hence a comparison between the localized AE sources
as function of macroscopic stress level and the DIC results is made in Figs. 7.6,
7.7, and 7.8.
The left side of Fig. 7.6 presents the localized AE sources identified as matrix
cracking as function of the specimen stress. The right side shows the evaluation
results of identified positions of transverse cracks from DIC measurements shown
as horizontal black lines. These are reported integrated over time, i.e., all identified
transverse crack positions until final failure of the specimens are marked. The DIC
image shows the field of the first principal strain in the last image prior to ultimate
failure of the specimen. Clearly some of the transverse cracks are still visible by the
strain concentrations found in the DIC image. However, some of the signatures are
hard to spot, since the occurrence of further transverse cracks in close neighborhood
smears out the signatures or starts to dominate the strain scales.
556 7 Combination of Methods

1st principal strain


[%]
60 60 1.300
AE source position
1.200
50 50 1.125

1.050
40 40
y-position [mm]

0.975
0.900
30 30
0.825

20 20 0.750

0.675
10 10 0.600

0 0
200 300 400 500 600 7.5 mm
stress [MPa] crack position [mm]

Fig. 7.6 Localized AE source positions of matrix cracking in [0/90]sym specimen as function
of the specimen stress level (left) and DIC results of localized strain concentrations (black
lines indicate y-position) with snapshot of strain field of first principal strain at last image prior
to failure (right)

1st principal strain

60 60
AE source position

50 50

40 40
y-position [mm]

30 30

20 20

10 10

0 0
200 300 400 500 600 700 7.5 mm
stress [MPa] crack position [mm]

Fig. 7.7 Localized AE source positions of matrix cracking in [(0/90)2]sym specimen as function
of the specimen stress level (left) and DIC results of localized strain concentrations (black
lines indicate y-position) with snapshot of strain field of first principal strain at last image
prior to failure (right)
7.2 Established Method Combinations 557

1st principal strain

60 60
AE source position

50 50

40 40
y-position [mm]

30 30

20 20

10 10

0 0
200 300 400 500 600 700 7.5 mm
stress [MPa] crack position [mm]

Fig. 7.8 Localized AE source positions of matrix cracking in [(0/90)3]sym specimen as function
of the specimen stress level (left) and DIC results of localized strain concentrations (black
lines indicate y-position) with snapshot of strain field of first principal strain at last image prior
to failure (right)

For the three different stacking sequences shown in Figs. 7.6, 7.7, and 7.8, good
agreement was found in the observed positions of transverse crack growth in AE
and DIC. Considering that the source localization accuracy of the AE signals
suffers from the growing number of cracks acting as an internal obstacle (cf. also
Sect. 4.7), the occurrence of the signatures in DIC indicating the occurrence of
transverse crack growth coincides well with the occurrence of the respective AE
source locations. This allows tracking of the characteristic concentration of trans-
verse cracks as seen in the examples of the [(0/90)2]sym and the [(0/90)3]sym
stacking sequence.
Based on the conclusions in Sects. 3.3 and 4.2, acoustic emission measurements
show superior sensitivity to detect failure mechanisms. However, for the present
case only one-dimensional localization was possible using acoustic emission,
whereas the DIC results provide full-field 2D information of the size of the defects.
For a reasonable field of view, the DIC measurements provide significantly better
spatial representations of failure locations and allow easy tracking of their growth.
In summary, the combination of information of both methods substantially
increases the reliability of the drawn conclusions. While AE measurements still
serve as benchmark in terms of detection sensitivity, the DIC signatures add full-
field information and are easier to interpret in terms of their significance, once
conforming AE source positions are found at the same spot.
558 7 Combination of Methods

7.2.3 Thermography and Acoustic Emission

In order to complement acoustic emission measurements by a visual method, still


other options exist. One approach which has been reported in literature recently is
the use of thermography in combination with acoustic emission monitoring
[3, 7]. To this end, two approaches exist, that can be employed in practice:
1. Load-hold experiments in combination with active thermography to perform
classical NDT inspection.
2. Passive monitoring of the experiment using thermography camera.
The first option is highly valuable in terms of the derived information, such as
visual representation of damaged zones and the possibility to track and locate
damage as an independent method. However, using the state-of-the-art equipment,
the second option comes with much more benefits as to merge information of AE
measurements and thermography since it constitutes a similar in situ approach with
high temporal resolution.
To demonstrate the high synergy between the two methods and to provide an
estimate to which level thermography and acoustic emission may supplement each
other, two representative examples are provided in the following. As first example,
results of an open-hole tensile test using a Sigrafil CE1250-230-39 cross-ply
specimens as described in Sect. 4.7.3.2 with [(0/90)3]sym stacking sequence are
presented. As second example, an edge-notched tensile test result applied to a
[(0/90)2]sym laminate made from the same material is shown. Acquisition settings
for acoustic emission measurements confirm with Table 4.2, while the recording
parameters of the thermography equipment are reported in Tables 7.6 and 7.8.
During quasi-static loading of an open-hole specimen, several acoustic emission
signals were detected and have been classified into the occurrence of matrix
cracking, interfacial failure, and fiber breakage as outlined in Sect. 4.5. After
reaching a static load of 21 kN, a cyclic load with 0.4 Hz and 2 kN was applied
to allow damage progression due to periodic loading and unloading. Figure 7.9
shows the localization results as a function of the load time with superimposed
load-time curve and a schematic representation of the specimen dimension. As
expected from the stress concentration effect of the hole inside the laminate, most
of the AE source positions fall into the expected area of the damage zone close to
the position of the hole.

Table 7.6 Thermography acquisition parameters


Measurement setup Value
Camera FLIR SC7000 with InSb-Detektor of 640  512 pixels
Lenses 50 mm, spectral range 3–5 μm
Thermal resolution 20 mK
Acquisition rate 100 Hz
Averaging 4
Integration time 1250 μs
7.2 Established Method Combinations 559

a b
25 force-time curve
Matrix cracking
Interfacial failure 60 AE sensor
20 Fiber breakage 1

40
15
3 mm hole
force [kN]

10 20

x-position [mm]
5
x
0 2
y
0
0 25 50 75 100 125 150
time [s] 17 mm

Fig. 7.9 Localized AE source positions during open-hole tensile test as function of the load time
with superimposed load-time curve (a) and schematic representation of the specimen (b)

a b high

low
x heat signature x

y y

6 mm 6 mm

Fig. 7.10 Thermography images recorded during loading and unloading stage at t ¼ 132 s (a) and
at t ¼ 142 s (b). Temperature scale is adapted for each image to provide highest visibility

Within the simultaneously recorded thermography images, first heat signatures


appear during the first cycles of the loading and unloading stage. As seen in
Fig. 7.10a, a first strong transient heat signature occurs at the onset of the first
cyclic loading period (around 132 s in Fig. 7.9) at a designated position. This is
followed by generation of more heat in the vicinity of the hole exhibiting a stripe-
like appearance before ultimate failure occurs. The stripe-like heat signatures are
likely caused by failure initiated at the hole in between the fibers oriented in parallel
to the load axis. Due to the high image acquisition rate and the known position of
the heat signatures this allows comparing the occurrence of these heat signatures
with the acoustic emission source positions. Therefore the first heat signature can be
560 7 Combination of Methods

directly related to the occurrence of matrix cracking and interfacial failure at around
132 s. The subsequent indication of the thermography images (cf. Fig. 7.10b) are
more difficult to be correlated to a particular signal. This is because of the close
sequence of the acoustic emission signals resulting from multiple failure types
including fiber breakage. Since each failure type will cause a heat signature, the
challenge arises from the heat flux in the material, which inhibits spotting individ-
ual heat spots. Moreover, the cyclic loading will generate heat from friction at the
damaged areas. Thus the heat signatures of the individual failure types all merge
into one signature and form a good visual representation of the overall affected
damage zone. The extent of this zone along the x-direction is in good accordance
with the calculated acoustic emission source positions.
The second example considers a tensile test of a double-edge-notched specimen.
As seen from the load profile in Fig. 7.11, this type of specimen was subject to a
sequence of static load increases followed by a short period of loading and
unloading cycles of 0.4 Hz with an amplitude of 2 kN. In the cyclic loading
stages, the experiment was monitored by a thermography camera with acquisition
settings of Table 7.7. During the experiment, many acoustic emission signals were

a b
force-time curve
16
Matrix cracking AE sensor
60
Interfacial failure
14
Fiber breakage 1
12
40
10 5 mm cut
force [kN]

8
20
6
x-position [mm]

4 x
0 2
2
y
0
0 500 1000 1500 2000
time [s] 17 mm

Fig. 7.11 Localized AE source positions during double-edge-notched tensile test as function of
the load time with superimposed load-time curve (a) and schematically representation of the
specimen (b)

Table 7.7 Thermography Measurement setup Value


acquisition parameters
Camera FLIR SC7000 with InSb-Detektor
of 640  512 pixels
Lenses 50 mm, spectral range 3–5 μm
Thermal resolution 20 mK
Acquisition rate 100 Hz
Averaging 2
Integration time 1250 μs
7.2 Established Method Combinations 561

recorded and were post-processed analogous to the previous example. The resulting
localized source positions are plotted as function of load time in Fig. 7.11 with
color-coded information of the source type. Since the presence of the notch
causes stress concentration at the center of the specimen it is expected to see
most of the AE source locations being present at this position. However, there is
a reasonable distribution of source positions extending well above and below the
position of the notches.
In Fig. 7.12 a selection of thermography images exhibiting heat signatures at
distinct positions is shown for the loading and unloading periods with center load
levels of 6, 8, 9, and 10 kN. For all cases, the moment of occurrence and the extent
of the heat signatures allow direct comparison to the detected acoustic emission
signals. As seen from the comparison of damage zone sizes evaluated based on the
heat signatures and as based on acoustic emission measurements there is good
agreement between both methods (cf. Table 7.8). However, the 7 kN load step lacks
a clear heat signature, while still some acoustic emission signals are detected and
localized. Also, for the broad heat signatures, it is difficult to correlate the occur-
rence of a particular signal to the heat signature. For small heat signatures such as
the 8 kN load case, most of the corresponding acoustic emission signals are all
classified as matrix cracking, so the occurrence of inter-fiber failure at this position
is likely the case for these heat signatures. For the heat signatures of Fig. 7.12 at

6 kN 8 kN 9 kN 10 kN high

x
heat signature
y
low
10 mm

Fig. 7.12 Set of thermography images exhibiting heat signatures recorded during loading and
unloading stage at 6, 8, 9, and 10 kN. Temperature scale is adapted for each image to provide
highest visibility

Table 7.8 Comparison of Damage zone size Damage zone


damage zone size evaluated Load (kN) thermography (mm) size AE (mm)
based on heat signatures and
6 7 15
based on localized acoustic
emission sources 7 – 3
8 1 1
9 5 5
10 7 5.5
11 10 14.2
12 15 17.5
562 7 Combination of Methods

12 kN 12 kN 12 kN high

heat
heat
signature
signature

low
x
heat
signature 10 mm
y

Fig. 7.13 Set of thermography images exhibiting heat signatures recorded during loading and
unloading stage at (12  2) kN. Temperature scale is adapted for each image to provide highest
visibility

6, 9, and 10 kN, the corresponding acoustic emission signals originate from


different source types, so the signature is likely caused by an extended damage
zone covering more than one particular failure mechanism.
A further set of spatially localized heat signatures is shown in Fig. 7.13 all
recorded during the 12 kN loading and unloading stage. AE signals at these source
positions in space and time all correspond to the occurrence of matrix cracking.
Thus it can be concluded, that these signatures also originate from the occurrence of
inter-fiber failure in the laminate.
At the first cycle of (14  2) kN load, the specimen finally fails. This rupture
process is shown in the thermography images of Fig. 7.14. As seen by the
generation of heat at the left edge a first crack starts to appear, quickly propagating
in parallel to the applied load. The heat zone starts to broaden, possibly also
indicating damage growth in perpendicular to the applied load, which could be
due to inter-ply delamination. Subsequent to this first signature, a second heat
signature appears at the opposite edge and starts to propagate in parallel and
perpendicular to the applied load. Finally both zones merge at the notch position
and cause severe fragmentation with heat signatures of particles being visible.
After rupture, the amount of heat generated by the macroscopic failure spreads out
in the remaining specimen.
The exemplary results shown above demonstrate the high synergistic potential of
thermography imaging in combination with acoustic emission measurements. Still
the detection sensitivity of acoustic emission measurements is better than the
detection of heat signatures. However, the appearance of transient heat signatures
allows spotting damage in a similar fashion as acoustic emission measurements, yet
the signatures are difficult to be attributed to a particular failure mechanism. But the
usage of thermography visualizes damage progression with reasonable spatial and
temporal resolution [14], while the classification and localization of acoustic
emission sources may be used to add quantitative information in terms of the
underlying failure mechanism.
7.2 Established Method Combinations 563

t = t0 t = t0 +280ms t = t0 +480ms t = t0 +540ms

x
first heat
signature second heat
y signature

t = t0 +560ms t = t0 +1020ms 20 mm high

fragments
low

Fig. 7.14 Sequence of thermography images recorded during rupture of the specimen. Temper-
ature scale is adapted for each image to provide highest visibility

7.2.4 Computed Tomography and Acoustic Emission

Computed tomography constitutes another approach to combine an imaging tech-


nique with the results of acoustic emission measurements. In contrast to the
previous imaging techniques, computed tomography is used as an example for a
volumetric imaging technique and thus forms a method combination able to
interpret and understand damage progression within the full volume of the material.
As discussed in Sect. 6.3 it is useful to distinguish between ex situ approaches and
in situ approaches in this context.

7.2.4.1 Ex Situ Computed Tomography

In addition to the previous benefits of acoustic emission measurements, two addi-


tional possibilities arise for investigating fiber-reinforced materials. First, the whole
concept of ex situ inspection is based on the idea to image a certain damage state at
a given damage progression. But without any secondary method being applied
during loading of the test specimen, this is rather difficult to accomplish. As a
representative example, consider the test results of three different Sigrafil prepreg
laminate types subject to a tensile test shown in Fig. 7.15. The experimental setup
confirms with the description given in Sect. 4.7.3.2. Based on the stress–strain curve
564 7 Combination of Methods

a
108
Laminate [02/90/02]sym visible signature
1200
Matrix cracking
Interfacial failure 107
1000 Fiber breakage

accumulated AE energy [aJ]


Stress-strain curve

800 106
Stress [MPa]

600
stage 1
stage 3 105

400 stage 2

104
200

0 103
0.0 0.2 0.4 0.6 0.8 1.0 1.2
Strain [%]

b 109
Laminate [02/902/0]sym
1000
Matrix cracking
Interfacial failure 108
Fiber breakage

accumulated AE energy [aJ]


800 Stress-strain curve
107
Stress [MPa]

600
106

400
stage 1 stage 2 105
stage 3

200
104

0 103
0.0 0.2 0.4 0.6 0.8 1.0 1.2
Strain [%]

Fig. 7.15 Example of three tensile tests with different laminate stacking including accumulated
acoustic emission energy in logarithmic scale
7.2 Established Method Combinations 565

c
1010
800 Laminate [0/903/0]sym barely visible non-linearity
Matrix cracking 109
700 Interfacial failure
Fiber breakage

accumulated AE energy [aJ]


600 Stress-strain curve 108
Stress [MPa]

500 107

400
106
300
stage 1 stage 2 stage 3 105
200

100 104

0 103
0.0 0.2 0.4 0.6 0.8 1.0 1.2
Strain [%]
Fig. 7.15 (continued)

of the materials it is often not possible to determine the damage onset or to evaluate
the level of damage progression. Thus without any secondary method, such as
digital image correlation, thermography or acoustic emission there is no intuitive
way to define distinct load levels relevant to be inspected by CT-scanning. Hence,
the specimen could only be scanned at arbitrary load levels defined relative to the
average ultimate tensile strength. This could potentially result in redundant scans
without damage being present or may select load levels of unstable damage states
and cause false conclusions. As alternative approach, the onset or levelling of
failure mechanisms based on acoustic emission measurements as shown in
Fig. 7.15 may be used to define characteristic stages of damage progression and,
consequently, characteristic load levels to perform CT inspection. For these exam-
ples, distinction can be made for three significant stages of damage progression. As
the first damage occurs almost immediately after load onset, this stage is considered
as load level of only microscopic damage being present within the material. In this
context note the logarithmic acoustic emission energy scale in Fig. 7.15, so the
total amount of acoustic emission energy during stage 1 is almost negligible
compared to the remaining signals. In some cases (e.g., Fig. 7.15a) the energy
level settles which indicates a state of crack arrest after first occurrence of micro-
scopic damage. Such load levels are well suited for a first inspection to track the
potential damage initiation sites like matrix cracks in off-axis plies or first delam-
ination (cf. examples in Fig. 7.16).
Within stage 2, energetically relevant acoustic emission release is recorded.
More than the spurious signals detected within the first stage, this is indicative of
566 7 Combination of Methods

Fig. 7.16 Exemplary set of a


CT-scan images recorded 3 mm
at the end of stage 1 in
[02/90/02]sym laminate
(a) and [02/902/0]sym
laminate (b)

x
z
y
crack in 90° ply

b 90° orientation
0° orientation

3 mm

z x
y delamination

a damage onset which affects the structural integrity of the material. Whilst this
acoustic emission still originates from cracks on the microscopic scale, their
accumulation will reach an extent which will leave the structure in a
pre-damaged state after unloading from this stage. For the cross-ply stackings
investigated here, this damage consists mostly of inter-fiber cracking in the 0 and
90 plies and interfacial delamination in between the plies. Also the first occurrence
of fiber breakage is noted and is due to the fact of weak filaments failing long prior
to ultimate failure. As seen from the corresponding CT-scans in Fig. 7.17 there are
numerous indications of inter-fiber cracks and delamination, which can readily be
observed after unloading a specimen from stage 2.
Subsequently, stage 3 ranges from this pre-damaged state to ultimate failure of
the specimen. Depending on the type of stacking, this region might also show
several distinct signatures, which might be useful to identify further levels for
inspection. For some cases, within stage 3, also significant signatures in the stress–
strain curve can be identified. This may range from substantial load drops
(cf. Fig. 7.15a) to weak nonlinearity of the stress–strain response originating from
7.2 Established Method Combinations 567

Fig. 7.17 Exemplary set of a


CT-scan images recorded at
3 mm
the end of stage 2 in [02/90/
02]sym laminate (a) and [02/
902/0]sym laminate (b)

x
z
y

b 90° orientation
0° orientation
3 mm

z x

the reduction of specimen stiffness (cf. Fig. 7.15c). Within stage 3, the detected
acoustic emission energy tends to increase substantially, sometimes reaching the
exponential growth behavior. This is because of the immense amount of micro-
scopic damage that accumulates, interacts, and finally causes failure of individual
layers or the full cross-section. In stage 3, all sorts of failure types can be found and
it is almost impossible to directly relate the occurrence of one acoustic emission
signal with the CT-images due to the multitude of options for correlation, given the
amount of signals within a time interval and the accuracy of source localization
procedures.
However, for lower load levels, an obvious strength of the combination between
CT and AE is the possibility to validate the source mechanism of the detected AE
signals. To this end, sufficiently high resolution may be achieved by state-of-the-art
CT-scanners as discussed in Sect. 6.2. In this context the detail visibility and
accuracy of spatial localization in CT is superior to AE source localization and
thus comprises a major challenge for a direct comparison. Since the typical plate-
like structure also inhibits source localization along the thickness direction of the
laminate, the AE source position is always a projection of all failure mechanisms
occurring within the laminate to the 3D-surface. Despite of these difficulties, it is
possible to perform a direct correlation for some cases. The first approach, which
568 7 Combination of Methods

was already successfully used for coating materials in [15] is to rely on the
characteristic source positions of a partition of signals, rather than the individual
signals. To this end, a Sigrafil prepreg laminate with [0/90]sym stacking was
investigated. The experimental setup confirms with the description given in Sect.
4.7.3.2. The specimen was loaded to stage 3 (strain level 1.2 %) as discussed
previously, was removed from the test machine and was scanned in this damaged
state. Subsequently, the specimen was loaded until ultimate failure. For this spec-
imen quite different source positions and signal densities were found for each
classified failure mechanism, which makes it particularly interesting to correlate
with CT-scan observations.
In Fig. 7.18 a comparison of the signals attributed to matrix cracking and a scan
of the center part of the specimen is shown. The AE signals’ source positions along
the x-axis are plotted as the function of strain and are used to demonstrate the
evolution of the occurrence of matrix cracking. The information on source location
is thus only one-dimensional, i.e., the source coordinate is only known for the
x-axis. The CT-scan volume was chosen to provide high spatial resolution
inhibiting the scan of a larger volume than presented in Fig. 7.18. For visibility
of the cracks, the material was rendered transparent, so only air inclusions remain
to reveal the pattern of inter-fiber cracks, inter-ply delamination and pores.

Photography CT-Scan Photography

1.3
1.2
1.1
1.0
Strain [%]

0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
10 15 20 25 30 35 40 45
x-coordinate [mm]

Fig. 7.18 Comparison between AE source localization pattern for matrix cracking and CT-scan
after reaching 1.2 % strain (only signals of highest AE energy are related by arrows)
7.2 Established Method Combinations 569

Remarkably, the signals of matrix cracking falling within the range of x ¼ 20 mm to


x ¼ 35 mm show very high acoustic emission energies. Therefore they should
correspond to the occurrence of large inter-fiber cracks, which was found to be
valid for the CT-scan shown in Fig. 7.18.
However, some inter-fiber cracks readily visible in Fig. 7.18 are not accompa-
nied by an according AE signal source position. The reasons can be manifold. First,
it could be that the occurrence of the crack has fallen into the dead-time of the
recording equipment (see Appendix C). Second, it could be, that the signal was
detected, but no valid source position could be calculated due to overlapping
signals. Third, the signal could be attributed to interfacial failure instead. The latter
is in fact quite likely, since some of the inter-fiber cracks seen in the CT-scan also
cause substantial amount of inter-ply delamination. Apart from that, the localiza-
tion pattern was found to be in reasonable agreement and demonstrates how AE and
CT can be combined to increase the reliability of the AE method.
For the group of signals identified as interfacial failure, the corresponding
localization as function of strain is shown in Fig. 7.19. For comparison, the same
CT-scan as in Fig. 7.18 is shown. Compared to the matrix cracking signals, the
overall number of signals of interfacial failure is much higher. Hence it is difficult
to correlate individual signals to particular signatures in the CT-scan. Nevertheless,

Photography CT-Scan Photography

1.3
1.2
1.1
1.0
Strain [%]

0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
10 15 20 25 30 35 40 45
x-coordinate [mm]

Fig. 7.19 Comparison between AE source localization pattern for interfacial failure and CT-scan
after reaching 1.2 % strain
570 7 Combination of Methods

two positions of repetitive identical source locations are marked in Fig. 7.19 and
indicate two positions of large inter-fiber cracks, where no associated matrix
cracking signal has been found. These cracks also are accompanied by a zone of
inter-ply delamination (seen as gray band at half the specimen width). The growth
of these inter-fiber cracks may have initiated inter-ply delamination simultaneously
and thus may have caused a signal corresponding rather to this type of mechanism.
The cracks in parallel to the load axis (x-direction) and the extent of the inter-ply
delamination are within the same range as the position of signals belonging to the
group of interfacial failure. Therefore, this cluster may be understood and credited
to these phenomena in this specimen.
An obviously different localization pattern is observed for the fiber breakage AE
signals. Their source position is shown as the function of strain in Fig. 7.20. A large
amount of signals localizes at two distinct positions at x ¼ 15 mm and x ¼ 38 mm.
As seen from the photography image after ultimate failure of the specimen, the left
agglomeration of fiber breakage signals corresponds to the position of specimen
failure. Since this is due to the rupture of the plies with fibers in parallel to the load
axis, this finding is not surprising. But the distinctly different source location
positions may readily be understood as further evidence for the difference in source

Photography

1.3
1.2
1.1
1.0
Strain [%]

0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
10 15 20 25 30 35 40 45
x-coordinate [mm]

Fig. 7.20 Comparison between AE source localization pattern for fiber breakage and photography
after final failure
7.2 Established Method Combinations 571

mechanism, which is well supported by the CT-scans and the failure mode of the
specimen. Likewise, other specimens without preferential densification of fiber
breakage source positions were observed to fail either at multiple positions or by
vast fragmentation of the specimen.
Further possibilities to combine ex situ CT observations with AE measurements
are attempts to perform a direct one-to-one correlation of an acoustic emission
signal with signatures in a CT-scan. To this end, investigations using a double-
edge-notched tensile specimen were performed. As seen from the 3D CT-scan of
the specimen after failure, the dominant position of failure is at the position of the
reduced cross-section (cf. Fig. 7.21). However, based on the crack-pattern seen in
Fig. 7.22, there is also a substantial amount of damage located at distances far away
from the position of the notches. Since only few acoustic emission signals were

Fig. 7.21 3D image of CT-scan of double-edge-notched tensile specimen after ultimate failure

y
45

40
y-coordinate [mm]

35

30

notch position
25

20
0 500 1000 1500 2000
Time [s]

Fig. 7.22 Comparison between AE source localization pattern of matrix cracking and CT-scan
after failure of specimen
572 7 Combination of Methods

localized outside the position of the notches, this allows a direct correlation of the
isolated signals with respective signatures in the CT-scan.
In Fig. 7.22 a comparison of the signals attributed to matrix cracking and a
projection of all inter-fiber cracks, inter-ply delamination and pores is shown. The
AE signals’ source positions along the y-axis are plotted as the function of time of
the experiment. The CT-scan volume was chosen to provide high spatial resolution,
thus allowing spotting the occurrence of single fiber filament failure. Similarly as
for the previous example, it is possible to identify an inter-fiber crack signature for
all AE signals in the corresponding CT-scan. For all source positions above
y ¼ 32 mm, the correspondence was found to be unambiguous. For the remaining
source positions, the situation is more difficult, since the area close to the notches
suffers from the final rupture of the specimen, and it is not straightforward to
correlate the multiple AE source positions with the details of the CT-scan.
For the case of interfacial failure shown in comparison to the CT-scan in
Fig. 7.23, again good and unambiguous assignment of the source positions can be
made with signatures corresponding to the occurrence of inter-ply delamination.
The latter was mostly observed in areas close to cracks in parallel to the y-axis.
Finally, for the AE signals identified as fiber breakage, only three distinct signals
were found outside the notch positions, which fall within the volume of the
CT-scan. As seen in Fig. 7.24, these do not occur at positions of the inter-fiber
cracks discussed before. Accordingly, virtual xy-cross-sections were made within
the CT-volume at different z-positions in the region of localized fiber breakages. To
account for the uncertainty of the AE source coordinate, several cross-sections were

y
45

40
y-coordinate [mm]

35

30

notch position
25

20
0 500 1000 1500 2000
Time [s]

Fig. 7.23 Comparison between AE source localization pattern of interfacial failure and CT-scan
after failure of specimen
7.2 Established Method Combinations 573

x
45
y

40
y-coordinate [mm]

35

30

notch position
25
0.2 mm

20
0 500 1000 1500 2000
Time [s]

Fig. 7.24 Comparison between AE source localization pattern of fiber breakage and CT-scan after
failure of specimen with details of particular cross-sections revealing the occurrence of single
filament failure

evaluated along the y-axis, considering an uncertainty of localization of 2 mm.


Based on the acoustic emission energy of the signals, it is expected that the signals
correspond to the occurrence of fiber breakage likely with a cross-section of less
than three filaments.
Only three distinct positions with signatures of fiber breakage were located in the
designated regions of the CT-scan as indicated in the magnified xy-cross-section
planes shown in Fig. 7.24. Two of them are part of an inter-fiber crack running along
the y-axis. The latter might either be the consequence of the fiber breakage or the
cause for the filament failure. Given the resolution of the CT-scan it is clear, that all
three cross-sections are of a typical dimension as expected for one single filament.

7.2.4.2 In Situ Computed Tomography

In contrast to the ex situ approach, the combination of in situ CT-scanning with AE


measurements requires some modifications to the approaches described in Sect. 6.3.
In addition to the challenges faced by the load introduction and the geometrical
stability, the generated acoustic emission signals also need to be detected by
attached sensor systems. For reasonable specimen sizes and miniaturized AE
sensors these may be attached to the specimen in a conventional way as reported
in literature before [16–18].
But in order to achieve the highest resolution, i.e., to resolve single
fiber filaments, the specimen cross-section will typically approach dimensions of
574 7 Combination of Methods

acoustic waveguide
sensor
attachement
6.95 mm

load bars specimen

z PMMA tube
x adhesive
(180° removed )
y

motor case

Fig. 7.25 3D drawing of in situ load rig optimized for AE acquisition showing full device (left)
and interior design with some parts removed (center) as well as details of specimen attachment
(right) (reprinted from [19])

few mm2, which does not allow a direct mounting of the substantially larger AE
sensors. Hence, in such cases another approach is necessary to allow AE monitoring
of specimen failure. To this end the load rigs presented in Sect. 6.3 can be modified
to include a waveguide to transfer the AE signal to an attached AE sensor.
Figure 7.25 shows a 3D drawing of such a load rig as used for tensile/compression
testing. The top of the load rig has been adapted to allow acoustic emission
monitoring by using a KRNBB-PC-type sensor. The primary modification involves
a slender waveguide-like design of the specimen fixture and an acoustic decoupling
to the surrounding parts before reaching the sensor position. Using these modifica-
tions allow a reasonable interpretation of the first arrival of the recorded AE signal
without significant interference with signal reflections due to the surrounding
components.
Considering the waveguide design used in the following experiments, it is
helpful to calculate the dispersion curves of the design. The L-modes, F-modes,
and T-modes of an Aluminum rod of 6.95 mm diameter made from AlMg3 as found
in Table B.1 of Appendix B are shown in Figs. 7.26, 7.27, and 7.28. The indexing of
guided wave families follows the software used for the computation, which is
described in [20]. For the frequency range up to 1 MHz there are numerous guided
wave modes found for this rod diameter. Thus, it may happen that the detected AE
signal may suffer from the transmission due to the generation of multiple guided
wave modes.
7.2 Established Method Combinations 575

a
10000
L(0,0)
L(0,1)
L(0,2)
8000
phase velocity [m/s]

6000

4000

2000

0
0 200 400 600 800 1000
frequency [kHz]
b
6000
5500 L(0,0)
L(0,1)
5000
L(0,2)
4500
group velocity [m/s]

4000
3500
3000
2500
2000
1500
1000
500
0
0 200 400 600 800 1000
frequency [kHz]

Fig. 7.26 Calculated phase velocities (a) and group velocities (b) of L-modes for an Aluminum
rod of 6.95 mm diameter

To evaluate the severity of this effect, the transfer function of the waveguide was
assessed using numerical methods following [21, 22]. Several failure mechanisms
were modelled at the position of the mounted specimen and their source function
was compared to the detected AE signal. As a simple source function, a cosine-bell
step-function with 1 μs rise-time and 3 N force was used. The resulting wave
576 7 Combination of Methods

a
10000
F(1,0)
F(2,0)
F(3,0)
8000 F(1,1)
F(2,1)
phase velocity [m/s]

F(3,1)
F(1,2)
6000
F(2,2)
F(3,2)
F(1,3)
4000 F(2,3)
F(1,4)

2000

0
0 200 400 600 800 1000
frequency [kHz]
b
6000
F(1,0) F(1,2)
5500 F(2,0) F(2,2)
5000 F(3,0) F(1,3)
F(1,1) F(2,3)
4500 F(2,1) F(1,4)
F(3,1)
group velocity [m/s]

4000
3500
3000
2500
2000
1500
1000
500
0
0 200 400 600 800 1000
frequency [kHz]

Fig. 7.27 Calculated phase velocities (a) and group velocities (b) of F-modes for an Aluminum
rod of 6.95 mm diameter

propagation within the rod is shown in Fig. 7.29 at three distinct time steps. As seen
from the false-color images, the dominant wave propagation mode for this case is a
longitudinal pressure wave moving along the rod, corresponding to the L(0,0) mode.
For the source position in this case being at the axis of the rod (i.e., at the center)
it is not likely to excite any other mode than the longitudinal mode, since no torque
7.2 Established Method Combinations 577

a
10000
T(0,0)
T(0,1)

8000
phase velocity [m/s]

6000

4000

2000

0
0 200 400 600 800 1000
frequency [kHz]
b
6000
5500 T(0,0)
T(0,1)
5000
4500
group velocity [m/s]

4000
3500
3000
2500
2000
1500
1000
500
0
0 200 400 600 800 1000
frequency [kHz]

Fig. 7.28 Calculated phase velocities (a) and group velocities (b) of T-modes for an Aluminum
rod of 6.95 mm diameter

or bending moment is induced by the source. To check this influence, a second


study was conducted using a source position offset to the center by half a radius.
Despite of a noticeable excitation of the F(1,0) mode, the overall appearance of the
wave propagation and the calculated transfer function was found to be quite similar.
578 7 Combination of Methods

Fig. 7.29 Sequence of acoustic velocity


5 μs 10 μs 15 μs
images showing wave magnitude
propagation in the
waveguide

z
x
y

Thus, the transmission for the present case seems to be comparatively flat for the L
(0,0) mode and the F(1,0) mode propagation within the rod.
An evaluation of the source function and the detected out-of-plane displacement at
the detector position is given in Fig. 7.30a. Although the two signals seem to be
substantially different at first sight, this observation is no longer valid after transfor-
mation to the frequency domain as evident from the FFT of both signals in Fig. 7.30b.
In fact, both signals are found to be very similar, resulting in a transfer function,
which is almost unity as seen in Fig. 7.30b. The reason for the visual difference in
Fig. 7.30a is the drift of the rod superimposed to the evaluation at the detector
position. This also causes the noticeable deviation to unity of the transfer function
below 100 kHz. However, for the present applications, the design of the waveguide is
assumed to be sufficient to detect the acoustic emission signal of material failure.
Using this experimental setup, four different fundamental AE source configura-
tions were studied. These comprise the occurrence of inter-fiber failure in unidi-
rectional composites due to tensile and compressive load perpendicular to the fiber
axis and fiber failure due to tensile and compressive load parallel to the fiber axis.

Inter-Fiber Failure

To induce inter-fiber failure, the mechanical testing procedure follows the


techniques described in Sect. 6.5.2. For the following tests, the specimens are
fabricated as prepreg laminate from the carbon/epoxy system Sigrafil
7.2 Established Method Combinations 579

a
0.14
detector position
source position
0.12

0.10
z-displacement [μm]

0.08

0.06

0.04

0.02

0.00
0 10 20 30 40 50
time [μs]
b
2x100

100
z-displacement [1/m]

detector position
source position
transfer function
10–4

10–5

10–6

10–7

10–8
0 500 1000 1500 2000
frequency [kHz]

Fig. 7.30 Comparison of source function displacement and out-of-plane displacement at detector
position (a) and respective signals in frequency domain including calculated transfer function of
waveguide

CE1250-230-39 following the curing cycle recommended by the manufacturer.


Miniaturized samples of nominally 1.2 mm  1.9 mm  22.0 mm (depth  width 
height) are cut from the laminates using a water-cooled low-speed saw with
precision diamond blade. Samples are attached to the aluminum load bars by an
580 7 Combination of Methods

Table 7.9 Acquisition Acquisition settings Value


settings used for in situ
Preamplification 20 dBAE
experiments
Threshold 38 dBAE
Triggering Individual channel
Acquisition rate 20 MS/s
Band-pass range 1 kHz to 3 MHz
Couplant None
Mounting system Screw thread
Sensor type KRNBB-PC
Number of sensors 1

UHU Plus endfest 300 epoxy adhesive. Mechanical testing is carried out in dis-
placement controlled mode applying 0.2 mm/min displacement rate. During
mechanical loading and CT-scan operation, AE measurements are performed by a
KRNBB-PC-type AE sensor with flat frequency response up to 1 MHz and a PCI-2
data acquisition card with acquisition settings reported in Table 7.9. For evaluation
of the failure mode, a CT-scan was made after recording an AE signal.
For comparison to the experimental signal, a respective AE signal was modelled
for the presented case. All computation results shown in the following were
obtained by the “Structural mechanics module” in the software program Comsol
Multiphysics following established routines for modelling of acoustic emission
sources [8, 23], signal propagation [24] and detection [25]. Explicit modelling of
the KRNBB-PC sensor is not necessary, since the voltage signal for this type is
directly proportional to the out-of-plane displacement. Hence, the modelled AE
signals are evaluated as displacement normal to the top surface of the load bar, i.e.,
the z-direction in Fig. 7.29. All mesh settings were sufficient to provide >99 %
coherence as established in [24], with a maximal mesh size of 0.5 mm and
respective refinement in regions of narrow details. The modelling procedure con-
sists of two subsequent analysis steps following the way of implementation as
described in Sect. 4.2 and for the present example specifically described in [19].
As first step, the model is subject to a static load applied along the z-axis
direction with a value identical to the fracture load in the experiment (i.e.,
200 N). In the second step, a transient analysis is performed allowing crack growth
and AE release. As time step 10 ns were chosen to provide sufficient resolution
during crack growth and signal propagation. The cohesive zone element approach
requires a definition of an internal fracture plane. Assuming a planar fracture plane
for this purpose might not take into account roughness or curvatures as existent in
reality. Therefore, the present implementation uses the procedure described in
Sect. 6.5.1.2, to extract the resultant fracture surface. The example of Fig. 7.31
shows the fracture surface formed due to tensile load perpendicular to the fiber axis,
which was extracted from the CT-scan. The extracted and simplified fracture surface
was then embedded within a FEM-model to calculate the respective acoustic
emission signal. This procedure is schematically shown in Fig. 7.31, including the
specific dimensions of the fracture surface. The extraction of the fracture surface was
performed using “VGStudio MAX” with subsequent simplification in “MeshLab”.
7.2 Established Method Combinations 581

waveguide

z
y x

0.50 mm

extracted fracture surface specimen

Fig. 7.31 Scheme of procedure used to transfer CT-scan of inter-fiber fracture surface into
FEM-model (reprinted from [19])

Due to the presence of the static load in z-direction, a static displacement field
exists within the specimen. As pointed out in Sect. 4.2, this static load and
displacement field is used as an initial boundary condition for a second computation
step. This initiates the transient degradation of the stiffness vector of the thin elastic
layer based on the evaluation of Puck’s inter-fiber failure criterion. A sequence of
images during the crack propagation is shown in Fig. 7.32.
The false-color range of the degradation function Č(r) indicates the extension of
the crack by red areas, while blue colors resemble areas, where the material is still
in contact. The images visualize the state of crack growth between t ¼ 0.2 μs and
t ¼ 1.2 μs. The initiation point is located at the lower left corner of the embedded
fracture plane. The initiation is solely owed to the stress concentration at this edge
based on the 3D topology of the fracture plane and the applied load level. After
initiation, the crack grows along the fracture plane due to the stress concentration at
the crack front. The extension of the fracture plane is driven by the roughness of the
fracture surface and the velocity of crack growth approaches the speed of a
Rayleigh wave, which is the general limit of crack propagation velocity.
Due to the orientation of the local stress components, the newly formed fracture
surface is subject to a dominant mode-I-type load situation. This causes a crack
opening during propagation of the crack and the resulting movement of the fracture
surfaces constitutes the actual source mechanism for AE release. The initiation of
the dynamic motion of the fracture planes are conveniently visualized using dia-
grams of the acoustic velocity magnitude to visualize the birth of the AE signal
around the crack. As seen from Fig. 7.33, the near-field of this crack type consists of
a dominant radiation perpendicular to the crack surfaces, which extends along the
582 7 Combination of Methods

t = 0.2 μs t = 0.4 μs t = 0.6 μs 1

z 0.9

y x 0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0

t = 0.8 μs t = 1.0 μs t = 1.2 μs

Fig. 7.32 Calculated crack growth in in situ tensile test with fiber orientation perpendicular to
load axis at distinct time intervals with false-color range to indicate the value of the degradation
function Č(r) (reprinted from [19])

acoustic
t = 0.2 μs t = 1.0 μs t = 2.0 μs velocity
magnitude
[m/s]
25
z
y x
20

15

10

0
t = 3.0 μs t = 4.0 μs t = 5.0 μs

Fig. 7.33 Plot of acoustic velocity magnitude during and after crack growth (based on [19])
7.2 Established Method Combinations 583

acoustic
t = 6 μs t = 11 μs t = 16 μs t = 21 μs t = 26 μs t = 31 μs velocity
magnitude
[m/s]
2
z
y x 1.8

1.6

1.4
F(1,0) mode
1.2

0.8

secondary 0.6
F(1,0) mode 0.4

0.2

Fig. 7.34 Plot of acoustic velocity magnitude during signal spreading in waveguide geometry
(based on [19])

x- and y-axis during the progression of the crack. After t ¼ 3.0 μs the first reflections
at the edges of the small specimen superimpose to the emitted waves of the source.
Thus it is more convenient to adapt the false-color scale and observe the signal
spreading in the attached waveguide.
Within the waveguide (i.e., tensile bar) the incident waves are converted into
guided waves, which propagate towards the AE sensor position. As seen from the
acoustic velocity fields in Fig. 7.34 at six distinct time steps, there are several
guided wave modes visible within the waveguide. The two wave packages with
highest intensity were identified as primary and secondary F(1,0) modes by means
of time-frequency analysis using Choi-Williams diagrams and calculated disper-
sion curves of the 6.95 mm aluminum rod using the software provided by [20]
following the routines established in [26, 27].
The displacement normal to the top surface of the waveguide is evaluated as AE
signal and is plotted in Fig. 7.35 for the calculated duration of t ¼ 60 μs. For
comparison, the respective experimental AE signal obtained for the investigated
fracture surface is shown in Fig. 7.35 as well. The behavior of the experimentally
used KRNBB-PC sensor can readily be estimated as flat frequency response with
584 7 Combination of Methods

Experimental signal
4 Modelled signal

voltage [V] 2

–2

–4

–6
0 10 20 30 40 50 60
time [μs]

Fig. 7.35 Comparison of experimental and simulated acoustic emission signal at detector position
(reprinted from [19])

constant conversion factor of 6 kV/m also accounting for the preamplifier gain.
Thus the modelled sensor voltage may directly be compared to the experimental
sensor voltage. In addition, the modelled signal was filtered by a sixth-order
Butterworth high-pass of 1 kHz to confirm with the bandwidth limitations of the
experimental setup. As seen from the good quantitative agreement between both
signals, the proposed model setup is able to describe the generated AE signal very
well. Some differences arise with respect to the number and position of dips after
25 μs. Based on visual observations of the signal propagation path, this is likely
owed to interactions with parts not included within the model, which cause addi-
tional reflections arriving at the sensor position after the first wave mode has passed
beyond the sensor position (cf. Fig. 7.34). Since the experimental signal will also
move into the sensor and will interact with the surrounding metallic sensor fixture
this will contribute to the shape of the signal to a certain extent. However, except for
the first rise in the signal, the shape is dominated by the experimentally used band-
pass filter, which causes the sharp falling slope between 20 and 40 μs and the rising
slope from 40 μs on.
On an extended time scale, the existence of the guided wave modes and their
reflections become more apparent when superimposing the calculated dispersion
curves to a Choi-Williams distribution of the experimental signal as shown in
Fig. 7.36. All relevant modes exhibit a first arrival within a time window of about
10 μs (between 52 and 62 μs in Fig. 7.36) for the given distance between the source
and sensor. For the detected frequency components only the L(0,0) as first signal
arrival and later the dominant F(1,0) are present within the signal. The discrepancy
in the dominant frequency components of the F(1,0) dispersion curve and the Choi-
Williams distribution is dominated by the high-pass filter and can still be attributed
to this wave mode, since no other guided wave mode exists at the observed
7.2 Established Method Combinations 585

800 0.0
700 5.6E+05
F(1,0) 1.1E+06
600
frequency [kHz]

1.7E+06
500 F(1,1) 2.2E+06
second 2.8E+06
400 arival 3.3E+06
3.9E+06
300 L(0,0) 4.4E+06
200 4.9E+06

100
0

4
amplitude [V]

2
0
–2
–4
–6
0 25 50 75 100 125 150 175 200
time [μs]

Fig. 7.36 Choi-Williams distribution (top) for acoustic emission signal (bottom) recorded during
failure in tensile load transverse to fiber axis

frequency. The second wave package arrival seen in Fig. 7.36 also matches the
signature of a F(1,0) wave mode. This is emitted by the source as second event,
which was evaluated from the model to be due to the first vibration cycle of the two
fracture planes. Such a vibration cycle is given by the retracting fracture planes for
about 8 μs, which are reflected at the load bars and start to approach each other at
about 15 μs. This impact causes the second signature and in superposition to the first
F(1,0) signal the highest amplitude in the signal.
A distinctly different failure mode is induced by compressive load applied
transverse to the fiber axis. As seen from the Choi-Williams distribution of
Fig. 7.37, the acoustic emission signals are dominated by the occurrence of the F
(1,0) guided wave mode. At the beginning of the signal, a small signature due to the
first arrival of the L(0,0) mode is visible.
Compared to the tensile load case, the signal appears to be different in several
aspects. First, the overall signal intensity is less by almost a factor of five.
The primary reason for this difference in amplitude is the orientation of the
fracture surfaces and their vibration. Unlike the dominant mode-I failure of the
tensile case, the compressive load is likely to induce a sequence of individual
failures under mixed-mode conditions. Such a sequence may correspond to the
formation of the fracture planes A, B, C, and D seen in Fig. 7.38 either individually
or several at a time. This is confirmed by the sequence of F(1,0) mode arrivals in the
Choi-Williams distribution of Fig. 7.37, which are unlikely reflections, since these
are not present in other signals detected in this configuration (cf. e.g., Fig. 7.36).
Instead, they might be the individual release of one fracture plane after another
586 7 Combination of Methods

800 0.0
700 3.6E+04
7.1E+04
600 F(1,0)
frequency [kHz]

1.1E+05
500 F(1,1) 1.4E+05
multiple 1.8E+05
400 arivals 2.1E+05
2.5E+05
300 L(0,0) 2.9E+05
200 3.2E+05

100
0
amplitude [V]

–1

0 25 50 75 100 125 150 175 200


time [μs]

Fig. 7.37 Choi-Williams distribution (top) for acoustic emission signal (bottom) recorded during
failure in compressive load transverse to fiber axis

Fig. 7.38 Fracture mode force


for failure due to
compressive load transverse
to fiber axis

fracture A

fracture C
fracture B

fracture D

z
y x

causing the final appearance of the specimen as seen in Fig. 7.38. From the
viewpoint of failure theories, the full sequence is still considered as single failure
mode, but given the temporal resolution of the acoustic emission measurement,
such sequential damage can readily be resolved.
7.2 Established Method Combinations 587

Fiber Failure

In order to investigate fiber failure due to tensile and compressive load, different
specimen sizes were prepared to comply with the maximum force capacity of 500 N
of the in situ test frame and load cell. Specimens are fabricated from the Sigrafil
CE1250-230-39 carbon/epoxy prepreg laminate using miniaturized samples of
0.14 mm  1.35 mm  22.0 mm (depth  width  height). All other settings were
chosen as for the previous cases.
As seen from the Choi-Williams distribution in Fig. 7.39, the acoustic emission
signal is significantly different compared to the recorded signals of inter-fiber
failure. This is caused by the failure mode of the signals as shown in Fig. 7.40.
The dominant movement of the fracture surface is expected to be along the z-
direction giving rise to a strong excitation of the L(0,0) mode as identified by the
respective dispersion curve in Fig. 7.39. In addition, the failure mode is not
symmetric along the y-axis likely causing a superimposed movement turning into
a simultaneous excitation of a flexural mode. Since the dominant frequency content
of the source mechanism seems to be located around 250 kHz, the resultant guided
wave mode is the F(1,1) instead of the previously encountered F(1,0) wave mode.
A possible secondary arrival of a L(0,0) or F(1,0) is found around 110 μs and may
be caused by a sequence of damage as described previously. However, given the
nature of tensile failure it is not very likely to obtain a sequence of damage with an
intermediate settling of 30 μs.
For failure of the test specimens subject to compressive loading parallel to the
fiber axis, the acoustic emission signal shown in Fig. 7.41 was found to be

800 0.0
700 8.3E+04
F(1,0) 1.7E+05
600
frequency [kHz]

2.5E+05
500 F(1,1) 3.3E+05
4.2E+05
400 5.0E+05

300 5.8E+05
L(0,0) 6.7E+05
200 7.4E+05

100
0
amplitude [V]

–2

0 25 50 75 100 125 150 175 200


time [μs]

Fig. 7.39 Choi-Williams distribution (top) for acoustic emission signal (bottom) recorded during
failure in tensile load parallel to fiber axis
588 7 Combination of Methods

Fig. 7.40 Fracture mode


for failure due to tensile
load parallel to fiber axis

z dominant
movement
y
x

superimposed
movement

800 0.0
8.9E+03
1.8E+04
600 F(1,0)
frequency [kHz]

2.7E+04
F(1,1) 3.6E+04
4.5E+04
400 5.4E+04
6.3E+04
L(0,0) 7.2E+04
200 7.9E+04

0
2
amplitude [V]

–2
0 25 50 75 100 125 150 175 200
time [μs]

Fig. 7.41 Choi-Williams distribution (top) for acoustic emission signal (bottom) recorded during
failure in compressive load parallel to fiber axis

representative. Unlike the tensile load, the detected acoustic emission signals are
dominated by an almost pure F(1,0) wave mode. To analyze the resultant fracture
plane, the specimen fragments were moved apart subsequent to fracture to reveal
the scan result shown in Fig. 7.42. As seen in the resultant fracture surface, the
expected displacement components during fracture are in parallel to the z-axis, but
7.2 Established Method Combinations 589

Fig. 7.42 Band-kinking


fracture mode for failure
due to compressive load
parallel to fiber axis
sliding
displacement
z
y fracture surface
x fragments

displacement

also along a sliding displacement in the yz-plane. The latter component is expected
to cause a preferential stimulation of a flexural wave mode incident to the wave-
guide. It can also be observed from Fig. 7.41, that there are no noticeable secondary
arrivals of other wave modes. Based on the conclusions for the other failure modes
above, this indicates that the observed damage of Fig. 7.42 occurs not as pro-
nounced sequence of individual failure events but merges all into one acoustic
emission signal.
The present findings demonstrate the valuable possibilities to investigate com-
posite failure mechanisms offered by an in situ combination of computed tomog-
raphy and acoustic emission detection. Among other aspects, the experiments
demonstrate the sensitivity of the acoustic emission method to detect and analyze
sequences of damage macroscopically understood as a single failure event (such as
in the case of compressive inter-fiber failure). Moreover, it could be demonstrated,
that all failure mechanisms investigated cause distinctly different signal types. Most
of all, these could be distinguished by the presence of specific guided wave modes.
The use of a flat-with-frequency acoustic emission sensor in combination with a
waveguide of very flat transfer function provides a good basis to investigate the
frequency content of the detected signals. Here it was found, that the tensile fiber
failure stimulates guided wave modes at higher frequencies. For the fiber failure
due to compressive load, this was not the case, but it is likely, that the dominant
contribution to generate the acoustic emission signal is not the band-kinking of the
fibers, but the longitudinal splitting of the laminate. For the inter-fiber failure
modes, dominant stimulation at lower frequencies was observed.
However, the detected frequencies in these experiments might not be directly
related to experiments in macroscopic specimens. This is due to the used wave-
guide, which allows only distinct wave modes to propagate at distinct frequencies
590 7 Combination of Methods

(cf. also Sect. 4.3). Also, the overall size of the fracture zone will add to the
expected rise-time, thus dominating the excited bandwidth (cf. Sect. 4.2.1). Since
failure of fiber filaments will consist of much smaller fracture surfaces than
investigated here a substantial shift of bandwidth towards higher frequencies is
likely. Instead, the present in situ studies might act as reference for validation of
computational methods as demonstrated for the case of inter-fiber failure due to
tensile load. Such validated models of failure mechanisms may then be embedded
in macroscopic models as seen in Sect. 4.2 to compute quantitative acoustic
emission signals for larger structures.

7.2.5 Computed Tomography and Digital Image Correlation

As further suitable combination of the methods presented herein, the application of


digital image correlation techniques to computed tomography images may be
named. This has been established in the form of “digital volume correlation” and
has already turned into an active field of research and is ready to be used for
composite materials. The principles of this approach have already been introduced
in Sect. 6.4 and are thus not repeated here.

7.2.6 Electromagnetic Emission and Acoustic Emission

Acoustic emission analysis and electromagnetic emission analysis are established


measurement techniques for monitoring of crack formation and propagation as
discussed throughout Chaps. 4 and 5. While acoustic emission is already used for
monitoring of the durability of engineering structures and is properly established
in technical terms, electromagnetic emission analysis still is subject to basic
research efforts and has not yet turned into a technologically relevant measurement
technique. However, both methods share an important common link, which is the
source mechanism driving the excitation of an acoustic wave as well as the
electromagnetic wave.
Both techniques are capable of providing qualitative and quantitative statements,
but the reliability of quantitative information relies on the detailed understanding of
the correlation between the source mechanism and the measured signal. In this
context, the temporal evolution of the displacement field around the crack tip is the
source of acoustic emission and also constitutes a relevant contribution to the
generation of the electromagnetic emission signal. The combination of both mea-
surements is thus expected to provide valuable information to understand crack
formation and propagation on the microscale. However, during propagation of the
acoustic wave, the characteristics of the signal (e.g., frequency content) suffer from
attenuation, dispersion, and propagation in guiding media. In addition, the charac-
teristics of the signals detected at the surface of the solid are further altered by the
7.2 Established Method Combinations 591

detection process using piezoelectric sensors. Consequently, to derive a fundamen-


tal correlation between the (visco-) elastic properties of the cracking medium and
the crack surface motion, an experimental investigation by a method which is not
influenced by these mechanisms is desirable.
In comparison to acoustic emission analysis, the major advantage of electro-
magnetic analysis is the negligible influence of the propagation medium on the
detected signals, as well as the possibility to ensure flat response of the detecting
sensor within the frequency range of the signals. As demonstrated in Chap. 5,
electromagnetic emission signals provide a direct observation to investigate the
crack wall movement during and after crack formation. As major difficulties of this
method, the intensity of electromagnetic emission signals is strictly dependent on
the orientation between crack surface and the detecting sensors and the distance to
the sensor. In comparison to these effects, the electromagnetic emission signals
show only negligible influence on the conductive properties of the material inves-
tigated. Given the observation of electromagnetic emission in a broad range of
materials [28–37] it can be assumed, that measurement of electromagnetic emission
of fiber-reinforced polymers should generally be possible for a broad range of fiber/
matrix combinations.
Consequently, acoustic emission and electromagnetic emission measurements
may be combined to provide a comprehensive measurement approach for each
mechanism, which generates both signals.
The simultaneous detection of acoustic emission and electromagnetic emission
can then been used to aid in a variety of aspects:
1. Measure the source excitation time t0 for source localization procedures
2. Measurement of crack duration
3. Measurement of crack tip velocity
4. Measurement of source dynamics for acoustic emission source modelling
5. Measurement of crack orientation
Since Chap. 5 already presented some examples to demonstrate the combined
use of electromagnetic emission and acoustic emission, only some specific aspects
mentioned above are discussed in the following.
The use of the source excitation time t0 has been previously proposed in [37] and
for a detailed discussion, the interested reader is referred to this original publica-
tion. As has been pointed out in Sect. 4.6, the application of some source localiza-
tion procedures requires to solve for the unknown source coordinates and the
unknown source excitation time t0. Typically this is derived by solving an overde-
termined equation system. An alternative strategy has been proposed by [37], since
the EME measurement allows directly obtaining t0. Because electromagnetic waves
propagate at the speed of light, their signal onset directly corresponds to the
absolute value of t0. As seen in the exemplary graphs of Fig. 7.43, the EME signals
recorded during failure of single fiber filaments systematically precede the AE
signals by some microseconds. For the given experimental setup of Fig. 5.58 in
Sect. 5.5, this is a well-expected value, since the distance between the source and
sensor is about (4  1) mm (depending on the exact location of filament failure)
592 7 Combination of Methods

resulting in approximately (3  1) μs time of propagation to reach the sensor


position. Similar behavior was found for all pairs of EME and AE signals investi-
gated in the experiments throughout Chap. 5. Therefore, the measured onset of the
EME signal may directly be used for source localization procedures, i.e., to obtain

a
1.0

0.5 Δt = 3.0 μs
voltage [V]

0.0

–0.5

–1.0

0.10
0.08
0.06
voltage [V]

0.04
0.02
0.00
–0.02
–0.04
–0.06
250 260 270 280 290 300 310 320 330 340 350
time [μs]

b
1.0

0.5
voltage [V]

Δt = 2.9 μs
0.0

–0.5

–1.0

0.10
0.08
0.06
voltage [V]

0.04
0.02
0.00
–0.02
–0.04
–0.06
260 280 300 320 340
time [μs]

Fig. 7.43 Representative set of EME signals (red) and AE signals (black) for failure of single
filament failure as presented in Sect. 5.5
7.2 Established Method Combinations 593

c
1.0

0.5
voltage [V]

Δt = 3.1 μs
0.0

–0.5

–1.0

0.10
0.08
0.06
voltage [V]

0.04
0.02
0.00
–0.02
–0.04
–0.06
250 260 270 280 290 300 310 320 330 340 350
time [μs]

d
1.0

0.5
voltage [V]

Δt = 1.6 μs
0.0

–0.5

–1.0

0.10
0.08
0.06
voltage [V]

0.04
0.02
0.00
–0.02
–0.04
–0.06
250 260 270 280 290 300 310 320 330 340 350
time [μs]
Fig. 7.43 (continued)

Δt-values not between sensors, but as absolute measure relative to the time of
excitation t0 as seen in Fig. 7.43. However, this procedure is certainly restricted to
the simultaneous occurrence of EME and AE signals, which is not always the case
as discussed in Chap. 5.
The extraction of quantitative data to describe the crack dynamics is exemplified
in Fig. 7.44. As demonstrated in Sect. 5.2, the EME source provides a direct
594 7 Combination of Methods

a
~te
60
50 unfiltered
voltage [mV]

40 crack wall displacement


30
20
10
0
–10
–20
20
crack wall "oscillation"
10
voltage [mV]

–10
filtered by 20 kHz Butterworth high-pass 7th order
–20
0 100 200 300 400 500
time [μs]

b max. slope
12 amax = 12 mm
crack length [mm]

10
8
6
4
2
0

600 cmax = 594 m/s


crack velocity [m/s]

500
400
300
200
100
0
50 55 60 65 70 75 80 85 90 95 100 105 110
time [μs]

Fig. 7.44 Exemplary signal of an EME signal in composite failure to determine crack duration
and source oscillation (a) and respective results to obtain crack tip position and crack velocity (b)

window to observe the movement of the charged crack walls as the function of time.
In the “unfiltered” signals the dominating contribution is due to the generation of
charges as a consequence of the newly formed fracture surface. For reasonably flat
frequency sensors the first rise of the signal can thus be related to the duration of
crack growth te. The subsequent decay is partially owed to the charge relaxation,
7.2 Established Method Combinations 595

but was found to be dominated by the 1 kHz high-pass used in the presented
experimental setup of Fig. 5.64 in Sect. 5.5. If the respective fracture surface is
known (e.g., by macroscopic or microscopic measurements) the voltage scale may
directly be converted into a length scale of crack growth, since the initial and final
boundary condition is known. For the given case in Fig. 7.44b, this was evaluated as
12 mm and thus allows measuring the crack tip position as the function of time.
Likewise, the derivative of this signal may be used to measure the crack velocity as
the function of time. In the present example this evaluates 594 m/s as the maximum
velocity, which is reasonably less than the Rayleigh wave speed expected for
fracture of a carbon fiber-reinforced polymer failing parallel to the fiber axis.
Using a high-pass filter setting as shown in Fig. 7.44a, the “unfiltered” signal
may also be reduced to reveal only the oscillating part of the signal. As presented
throughout Sect. 5.5, this oscillation is in good agreement to the detected AE signals
of flat-with-frequency sensors. Based on the considerations on EME source oper-
ation in Sect. 5.2, it may be readily understood as the oscillation of the newly
formed crack walls. Thus, the information on its frequency, amplitude rise, and
decay are valuable measures to compare to respective modelling attempts. As has
been demonstrated in Sect. 5.4, the EME signals of a fracture event can be
quantitatively predicted by respective modelling means. Likewise, the results of
AE models presented in Sect. 5.2 may directly be validated in the near-field of the
source by using such EME measurements.
Finally, Sect. 5.5.2 has presented the possibility to perform orientation measure-
ments of the fracture surface by means of EME. Based on the distinct EME
radiation pattern, this allows to inversely determine the internal orientation of the
fracture plane. Since such information may not be obtained by acoustic emission
analysis using guided waves,1 the use of an array of EME sensors complements the
possibilities of AE in this respect. Therefore, the recording of multiple EME and
AE signals of one fracture event in a composite material can be used to obtain a
multitude of information:
1. 2D position of fracture (AE and EME source localization)
2. Type of fracture (AE source classification procedures)
3. Orientation of fracture plane (EME radiation pattern analysis)
The availability of this information provides a substantial increase in the knowl-
edge on failure evolution in composite materials provided in situ. However, one of
the general difficulties is given by the absence of EME signals for many AE signals.
As discussed in Sect. 5.5, this may likely occur due to two reasons.
First, the overall sensitivity of the EME equipment may not be at the same level
as for comparable AE equipment. There is an ongoing development to increase the
sensitivity of the EME sensors, but so far it is not absolutely clear, if all EME

1
For volumetric propagation media, moment tensor inversion procedure constitutes a possibility
to analyze the orientation of the fracture plane.
596 7 Combination of Methods

signals can be detected by the used sensor systems, or what fraction of signals is
hidden in the background noise.
As second reason, it may be speculated if every AE source is also an EME
source. For all types of friction-based AE sources, there is no breaking of bonds on
the microscopic scale and, consequently, no EME should be expected. Also, the
amount of charges present on the surface may change as the function of material. If
this drops below a certain value, no measurable EME will be generated. Similarly,
if no charge imbalance occurs, no EME signal is to be expected. For the latter
effect, it is required to clarify if the charging of the fracture surface discussed in
Sect. 5.2 is a statistical phenomenon or if the amount of generated charges obeys a
strict relationship to the material properties and the spatial dimension of the
fracture zone.

7.2.7 Acousto-Ultrasonics

As discussed in Sects. 7.1 and 2.4.6, the method of acousto-ultrasonics is a


combination of guided wave testing with acoustic emission analysis. In this
approach the capabilities of acoustic emission measurements are fully incorporated.
The major improvement consists in the ability to additionally sense defects not
currently active in the material by means of guided wave analysis. As has been
demonstrated in a variety of publications in the past [38–42] this allows continuous
assessment of the damage progression and allows visualization of damaged areas in
a broad range of materials and applications. Since this has been covered by many
scientific publications it is not intended to repeat these findings in detail within this
book. Instead the reader may refer to specific publications dealing with the method,
i.e., a special issue in the Journal of Acoustic Emission [43].
In the context of this book, the acousto-ultrasonics method is used in the
following to provide additional insight on the limiting factors of the acoustic
emission method in terms of source classification and source localization. As
pointed out in Sects. 4.5 and 4.6 the uncertainty of localization and uncertainty of
classification are directly affected by changes of the propagation path between
source and sensor. Thus the question arises, whether the formation of damage and
the dimensional change of the material will affect the extracted feature values, such
as Δt-values used for source localization or frequency parameters used for source
identification. To assess the severity of the effect, tensile tests of six different
stacking sequences are presented as representative data basis for some first
conclusions.
The basic experimental configuration is identical to the setup presented in Sect.
4.7.3.2, but uses an additionally mounted piezoelectric pulser directly below the
bottom AE sensor as seen in Fig. 7.45. By an arbitrary waveform generator, a 10 V
pulse with duration of 10 μs and 20 ns rise-time is generated to excite an acoustic
emission signal having a characteristic bandwidth between 1 and 800 kHz
(cf. Fig. 7.46a, b). The use of a signal bandwidth with reasonably high frequencies
7.2 Established Method Combinations 597

Fig. 7.45 Experimental


setup used for acousto- AE sensor
ultrasonics and tensile
2
specimen geometry

specimen

x
1
y
pulser
P

17 mm

is based on the findings of Sect. 4.3, which indicate higher sensitivity to interact
with damage for shorter wavelengths.
Pulses are repeated every 2 s and are detected in addition to the regular AE
signals as obtained from specimen failure during tensile loading. All recorded AE
signals are post-processed in the same way to obtain the typical AE features. Due to
the periodicity of the signals of the piezoelectric pulser, these are easy to distinguish
from the rest of the AE signals and are hence marked as pulser signals in the
following. For further interpretation of these signals, only signals of detector 2 were
evaluated, since these signals have passed through 45–50 mm of potentially dam-
aged specimen length.
Figure 7.47 shows an example of the calculated Δt-values of the pulse travelling
from one sensor to the other. As seen by the gradual decrease of the Δt-value, the
increasing amount of damage and the gradual change of specimen dimensions
cause a measurable shift of almost 1 μs during the experiment. Using the group
velocity of the S0 mode in this configuration, this translates into a resultant shift in
source position of 5.8 mm. This finding constitutes the extreme case of evaluation
results found for the investigated laminates. Compared to the other influences on
the uncertainty of localization as described in Sect. 4.6, this value is within the same
order as other effects that affect the source localization accuracy. Since the shift in
Δt-values is expected to be proportional to the propagation length this effect is
likely even more severe for larger structures. Hence, substantial contributions to
source localization errors may occur. For the other laminate configurations the Δt-
value changes were typically less, but also faced nonmonotonic trends. For the
quasi-isotropic laminate, first an average decrease of around 0.40 μs was found and
with the onset of substantial damage, the trend changed, resulting in an average
increase of the Δt-values by 0.97 μs. The findings of the other laminates including
their corresponding deviation in source positions are given in Table 7.10.
For the other extracted feature values, the plot of Partial Power 2 (150–300 kHz)
versus weighted Peak-Frequency is used as a representative diagram. Superimposed
to the partition of signals due to specimen failure, a small cluster corresponding
to the detected pulser signals is readily visible for each case shown in Fig. 7.48.
598 7 Combination of Methods

a
initial
0.45 intermediate
0.40 end

0.35
0.30
shifted voltage [V]

0.25
0.20
0.15
0.10
0.05
0.00
–0.05
–0.10
0 50 100 150 200
time [μs]

b
initial
16 intermediate
end
14

12
FFT-magnitude [1/V]

10

0
0 200 400 600 800 1000
frequency [kHz]

Fig. 7.46 Comparison of initial, intermediate, and end signal detected at sensor 2 (a) and
corresponding frequency spectra (b) for a [0/þ45/45/90]sym laminate

The absolute number of AE signals is subject to the respective laminate type. For
better visibility, the extent of the cluster corresponding to the pulser signals is
marked as black ellipsoid. During the progression of the experiment (and hence
damage) the pulser signals frequency content does not change substantially. This is
already seen quite well in the three representative signals shown in Fig. 7.46. Since
7.2 Established Method Combinations 599

9.0

8.5
Δt = 1μs

8.0
Δt-value [μs]

7.5

7.0

6.5

6.0

5.5

5.0
0 200 400 600 800 1000 1200
time [s]

Fig. 7.47 Calculated Δt-values for [0/90]sym laminate as the function of experiment duration

Table 7.10 Evaluation results of average Δt-values of all laminate configurations


Laminate type Δt-value (μs) Deviation of source location (mm)
[04]sym 0.27 1.9
[0/þ45/45/90]sym First drops 0.40 then increases 0.97 First drops 2.2 then increases 5.4
[02/90/02]sym 0.20 1.3
[02/902/0]sym 0.20 1.0
[0/903/0]sym 0.35 1.9
[0/90]sym 1.00 5.8

highest frequencies are expected to be more sensitive to the interaction with


damage within the laminate, the findings below may be considered as extreme
case. For all laminate configurations, the position of the pulser signals cluster falls
at the range of weighted Peak-Frequency between 500 and 600 kHz at lowest
Partial Power 2 levels. Despite the damage growth in the specimen, the resulting
spread of the signals does not move outside the corresponding fiber breakage cluster
or switches to another cluster. Compared to the other factors discussed in Sect. 4.5
the minor shift in frequency positions observed in the partitions of Fig. 7.48 would
not substantially interfere with attempts to perform source classification.
600 7 Combination of Methods

a
100 Matrix cracking
[04 ]sym
90 Interfacial failure
Fiber breakage
80 Test source

70
Partial Power 2 [%]

60

50

40

30

20

10

0
0 100 200 300 400 500 600 700 800
weighted Peak-Frequency [kHz]
b
100 Matrix cracking
[0/+45/-45/90]sym
Interfacial failure
Fiber breakage
80 Test source
Partial Power 2 [%]

60

40

20

0
0 100 200 300 400 500 600 700 800
weighted Peak-Frequency [kHz]

Fig. 7.48 Representative partitions for each laminate configuration with superimposed cluster of
pulser signals marked with black ellipsoid
7.3 Future Developments 601

c
100 Matrix cracking
[02/90/02]sym
90 Interfacial failure
Fiber breakage
80 Test source

70
Partial Power 2 [%]

60

50

40

30

20

10

0
0 100 200 300 400 500 600 700 800
weighted Peak-Frequency [kHz]
d
100 Matrix cracking
[02/902 /0]sym
90 Interfacial failure
Fiber breakage
80 Test source

70
Partial Power 2 [%]

60

50

40

30

20

10

0
0 100 200 300 400 500 600 700 800
weighted Peak-Frequency [kHz]
Fig. 7.48 (continued)

7.3 Future Developments

The intent of this book is to present the current state-of-the-art of in situ test
methods as available for composites. Still some of the challenges and drawbacks
associated with each method are at the edge of being overcome by technological
improvements. For every method discussed herein, there are ongoing developments
and a constant flow of new ideas may push the capabilities of the various techniques
602 7 Combination of Methods

beyond currently existing limits. Therefore, the aim of this final section is to
provide a short outlook on emerging developments of each method and to discuss
the future prospects of method combinations as presented in Sect. 7.2.

7.3.1 Digital Image Correlation

Digital image correlation has already made its way to a mature measurement
technology. One urgent issue to overcome is the missing standardization in the
optical measurement technologies. This is currently discussed intensively in vari-
ous standardization organizations and deals with the general applicability of DIC as
strain measurement tool as well as the explicit allowance to perform strain mea-
surements by an optical technique instead of a strain gauge. A further item achiev-
ing much attention these days is the standardization of the equipment as such and
the need to calibrate systems having a standardized measurement reference, i.e.,
objects with optimal speckle patterns undergoing known displacements and defor-
mations. Being an imaging technique DIC benefits from the technological evolution
in camera systems. From an economical perspective, the use of high-quality and
high-resolution camera systems becomes more and more feasible. Also, a further
boost in resolution of camera systems extending the range beyond 20 megapixels
will substantially impact the detail visibility of DIC systems.
One further challenge stems from the fusion of information of different DIC
systems. As a technical alternative to monitor one test object by one system, it is
also possible to use several systems to perform such monitoring. In this case, the
primary task is to combine the recorded information in one coordinate system.
At the heart of the DIC method, the image correlation algorithm itself is still an
ongoing development. Pushing the limits in correlation with sub-pixel accuracy is
the key to get to even better quantitative strain measures and to allow spotting of
strain concentrations in fiber-reinforced materials. For this material another chal-
lenge still originates from the distinction between natural anomalies and defect
indications. This is a general problem of fiber-reinforced materials, which due to
their inhomogeneity and local fluctuation in reinforcement fibers start to produce
strain signatures which are not necessarily linked to the formation of defects. This
part of DIC will certainly benefit from the development of damage tracking
algorithms in machine vision, since discrimination between indications and mate-
rial structure is a current field of research common to many imaging-based inspec-
tion methods applied to fiber-reinforced materials.

7.3.2 Acoustic Emission

As an established measurement technology, acoustic emission is already applied to


many composite applications and has already seen some standardization attempts.
7.3 Future Developments 603

Still deficiencies exist in the primary calibration of the sensing equipment and more
specific standards for composite applications. Also, new analysis routines are
hardly incorporated in existing standards. Since the quality of interpretation
might benefit by incorporation of source classification procedures or improved
localization routines it would be appropriate to perform some standardization in
their context for future applications. For the field of AE source identification, the
methods proposed herein are generally of high value and were found to be well
suited for the presented purposes. In order to extend the proposed classification
strategies to larger, thicker, or geometrically more complex structures made from
fiber-reinforced materials, still more research efforts are required. The same applies
for attempts to perform valid source classification in the presence of external noise.
In particular the ability to update source classification strategies as damage starts to
develop or as initiation of further AE sources happens [44, 45] is of vital importance
to get to a generally accepted source classification strategy. Similarly, the process
of source localization may need even closer attention than seen so far. Based on the
change of propagation medium and the growth of damage, simple localization
algorithms come with the price of increased uncertainty at higher load or damage
levels. Accordingly an implementation using an automatic update of the source
localization algorithm during operation could be used to overcome such loss of
accuracy.
A particular type of AE source, which has received only little attention in the
past, is the AE from stick–slip friction. For a fiber-reinforced materials, the rough
fracture surfaces that form within the material is naturally causing such AE sources
to a certain extent. It may be speculated that friction AE is of a different charac-
teristic than the AE of fracture processes, but there is currently no proof by
experiments or modelling work. Overall, the reliable implementation of fracture
mechanics-based modelling strategies to yield quantitative AE signals for friction
sources and material fracture could be understood as key to increase the overall
confidence put into the AE method.
Using such validated AE methods, they could certainly serve to indicate the
initiation and growth of damage as discussed herein, but also to perform damage
prediction and lifetime prediction of structures in service. Currently existing stan-
dards are partially making use of this concept already, but their applicability is
always limited to the generic type of structures the standard covers (e.g., lift devices
[46, 47] or pressure vessels [48, 49]).
Another limitation to overcome in the future is the design of AE sensors. The
currently used commercial devices are comparatively bulky and are not well suited
for a permanent monitoring of lightweight materials. To this end various groups
have already recommended embedded sensor systems for AE monitoring [50, 51],
most of them being still not as sensitive as existing solutions and are not appropriate
for frequency analysis due to system resonances. However, the general trend to
fabricate integrated sensing systems is directly compatible with fabrication of fiber-
reinforced materials and may be of stronger relevance in future structural designs.
Alongside with these attempts, the ongoing development in MEMS is generally of
high relevance to the field of AE. Following the idea to have commercially
604 7 Combination of Methods

available sensor types of low cost and low weight, the integration of AE as
permanent monitoring technology is much more economical and technically com-
petitive. In the same context, the wireless operation of AE sensors is receiving
much attention. The removal of wires is a weight-saving factor by itself, but also
limits the need for access to the structure and especially in the field of remote
inspection sites might be superior to currently existing solutions.

7.3.3 Electromagnetic Emission

Electromagnetic emission is a phenomenon, which has already demonstrated its


value in basic research applications. It nicely complements acoustic emission
technology and adds more information to interpret material failure. To become a
standalone measurement technology, there are several aspects that need to be
addressed by future research. First, the technology is still emerging, so substantial
improvement with respect to the sensing technologies is necessary. Here the key
parameters to optimize are the signal-to-noise ratio within the experimentally used
bandwidth and likewise the sensitivity of the sensing system relative to background
noise. If any of these aspects can be improved by several orders of magnitude, the
necessity of shielding chambers might become obsolete. A closely related topic is
the uniqueness of the transfer function of each sensing system, which makes
comparison of the measurement results very challenging. To this end, first guide-
lines have been established by means of this book, but these are not generally
applicable and might fail to be applicable for other experimental configurations
than those presented herein. Likewise, the development of miniaturized (MEMS
type) sensing systems or other sensor concepts as presented in Chap. 5 could boost
the technology if the current drawbacks can be overcome by such sensor systems.
From the basic research point of view, a scientifically sound description of the
source mechanism is required. Beyond the initial assessments presented within this
book, there is need for validation of the presented theories in other material systems
and material combinations. Based on the electrical properties of different materials,
the present description of the EME source could turn out to be valid or could lack in
the level of detail to describe the EME source operation.
Having such a validated EME method, it may act as an ideal tool to complement
AE or may act as an independent tool for inspection of fiber-reinforced materials.
The insight gained is likely to be used for material development and improvement
of failure theories, because it provides essentially new information on the formation
and propagation of cracks. Particular attention should be paid to the difference in
the absolute number of AE and EME signals. Despite the generally less sensitivity
of the EME system it might be speculated if every AE source also operates as an
EME source. The phenomenon of stick–slip friction is well-known to cause notice-
able AE but since there is no real crack formation it can be hypothesized if such
mechanisms will also cause EME. If this is not the case, such simultaneous
recording might well be suited for advanced source discrimination. This would be
7.3 Future Developments 605

a valuable approach to improve the understanding of a well-known phenomenon in


acoustic emission. In cyclic loading of composites, the Felicity-ratio becomes
smaller than one when approaching the ultimate limit of the material. Some
researchers have speculated if the AE signals detected before the previous load
limit are mostly due to friction sources [52, 53]. If EME is able to safely discrim-
inate between AE due to crack growth and AE due to friction this would help to
clarify a long-living ambiguity in interpretation of AE results in such load
scenarios.

7.3.4 In Situ Computed Tomography

For the applications involving in situ computed tomography, many technological


developments have already moved the frontiers of the method in the last 10 years.
Naturally, the method benefits from all advances in the design of the computed
tomography devices. There is still a trend to achieve smaller focal spot sizes since
this has still not reached its conceptual limits. Also, the detector systems still profit
from ongoing developments. In the context of in situ CT the most important factor
would be the considerable reduction of the scan duration. This could be achieved
by more sensitive detector systems or by substantial reduction of the signal-to-
noise ratio of the detector. Alongside with the attempts to achieve higher resolu-
tion, there are various efforts to scale the whole CT concept to be able to inspect
much larger parts. Similar benefits in terms of detail visibility are expected from
the reconstruction algorithms, which are still subject to numerical improvements.
Furthermore, new concepts to reduce scan durations or to reduce the required
X-ray intensities have been proposed by means of dual-source-CT and dual-energy
CT [54–57]. Also other imaging modes like bright-field, dark-field, and phase-
contrast imaging have demonstrated their potential for improving the visibility of
details [58, 59].
In the specific field of in situ CT, the amount of data being processed for
damage tracking or digital volume correlation is one of the key challenges for
algorithms. Hence, many efforts are currently put into development of smarter
algorithms to address existing fields of application like damage mark-up or
damage tracking as well as digital volume correlation. The extension of the latter
to a technological level currently seen for DIC will enable a much wider field of
application of digital volume correlation to perform research on the mechanical
behavior of materials.
Finally, X-rays are just one type of electromagnetic waves, which are well suited
for volumetric imaging of fiber-reinforced materials. Other volumetric imaging
sources, like microwave tomography or Terahertz-tomography are currently emerg-
ing and might be used for future inspection of fiber-reinforced materials.
606 7 Combination of Methods

7.3.5 Method Combinations

The combination of various measurement systems has become a field of research on


its own. Instead of relying on the information provided by one measurement
system, there is a strong trend to fuse information of different equipment. One
challenge directly related to these approaches is the lack of compatibility between
different measurement systems. Frequently the equipment is provided by different
manufacturers and there is no technical solution to directly merge the acquired
information of different systems. In principle there are common data exchange
formats, which are capable to store and distribute the acquired information. How-
ever, there is a general lack of implementation of these formats in the commercial
equipment due to the missing standardization. But data fusion of measurement
systems has been identified as key requirement for future technologies. With the
advances in information technology, it is likely that such data exchange might
become a standard at some point. Until then, specific software needs to be devel-
oped, which allows an easy fusion of discrete data pools originating from many
different measurement systems. This software needs to be ready for 3D visualiza-
tion and needs to show a high level of integration in the existing software packages
of commercial devices as the currently used portable document file (pdf) standard
according to ISO 32000-1.

References

1. Lee, J.-R., Molimard, J., Vautrin, A., Surrel, Y.: Application of grating shearography and
speckle shearography to mechanical analysis of composite material. Compos. Part A Appl. Sci.
Manuf. 35, 965–976 (2004)
2. Ono, K.: Structural integrity evaluation using acoustic emission techniques. J Acoust. Emiss.
25, 1–20 (2007)
3. Cuadra, J., Vanniamparambil, P.A., Hazeli, K., Bartoli, I., Kontsos, A.: Damage quantification
in polymer composites using a hybrid NDT approach. Compos. Sci. Technol. 83, 11–21 (2013)
4. Iliopoulos, S., Tsangouri, E., Aggelis, D.G., Pyl, L., Vantomme, J., Van Marcke, P., Areias, L.:
Digital image correlation, acoustic emission and ultrasonic pulse velocity for the detection of
cracks in the concrete buffer of the Belgian nuclear supercontainer. In: 31st Conference of the
European Working Group on Acoustic Emission. pp. 1–7. Dresden (2014)
5. Chotard, T.J., Smith, A., Boncoeur, M.P., Fargeot, D., Gault, C.: Characterisation of early
stage calcium aluminate cement hydration by combination of non-destructive techniques:
acoustic emission and X-ray tomography. J. Eur. Ceram. Soc. 23, 2211–2223 (2003)
6. Scott, A.E., Clinch, M., Hepples, W., Kalantzis, N., Sinclair, I., Spearing, S.M.: Advanced
micro-mechanical analysis of highly loaded hybrid composite structures. In: ICCM 17—17th
International Conference on Composite Materials, Edinburgh (2009)
7. Munoz, V., Vales, B., Perrin, M., Pastor, M.L., Welemane, H., Cantarel, A., Karama, M.:
Coupling infrared thermography and acoustic emission for damage study in CFRP composites.
In: The 12th International Conference on Quantitative InfraRed Thermography—QIRT 2014,
Bordeaux (2014)
8. Sause, M.G.R., Richler, S.: Finite element modelling of cracks as acoustic emission sources.
J. Nondestruct. Eval. 34, 1–13 (2015)
References 607

9. Vanniamparambil, P.A., Bolhassani, M., Carmi, R., Khan, F., Bartoli, I., Moon, F.L., Hamid,
A., Kontsos, A.: A data fusion approach for progressive damage quantification in reinforced
concrete masonry walls. Smart Mater. Struct. 23, 015007 (2014)
10. Fischer, G., Bohse, J.: Observation and analysis of fracture processes in concrete with acoustic
emission (AE) and digital image correlation (DIC). In: 31st Conference of the European
Working Group on Acoustic Emission. pp. 1–8. Dresden (2014)
11. Mao, W.G., Wu, D.J., Yao, W.B., Zhou, M., Lu, C.: Multiscale monitoring of interface failure
of brittle coating/ductile substrate systems: a non-destructive evaluation method combined
digital image correlation with acoustic emission. J. Appl. Phys. 110, 084903 (2011)
12. Rouchier, S., Foray, G., Godin, N., Woloszyn, M., Roux, J.-J.: Damage monitoring in fibre
reinforced mortar by combined digital image correlation and acoustic emission. Constr. Build.
Mater. 38, 371–380 (2013)
13. Aggelis, D.G., Verbruggen, S., Tsangouri, E., Tysmans, T., Van Hemelrijck, D.: Characteri-
zation of mechanical performance of concrete beams with external reinforcement by acoustic
emission and digital image correlation. Constr. Build. Mater. 47, 1037–1045 (2013)
14. Yekani Fard, M., Sadat, S.M., Raji, B.B., Chattopadhyay, A.: Damage characterization of
surface and sub-surface defects in stitch-bonded biaxial carbon/epoxy composites. Compos.
Part B Eng. 56, 821–829 (2014)
15. Sause, M.G.R., Haider, F., Horn, S.: Quantification of metallic coating failure on carbon fiber
reinforced plastics using acoustic emission. Surf. Coatings Technol. 204, 300–308 (2009)
16. Baensch, F., Zauner, M., Sanabria, S.J., Sause, M.G.R., Pinzer, B.R., Brunner, A.J.,
Stampanoni, M., Niemz, P.: Damage evolution in wood: synchrotron radiation micro-
computed tomography (SRμCT) as a complementary tool for interpreting acoustic emission
(AE) behavior. Holzforschung 69, 1015–1025 (2015)
17. Maire, E., Carmona, V., Courbon, J., Ludwig, W.: Fast X-ray tomography and acoustic
emission study of damage in metals during continuous tensile tests. Acta Mater. 55,
6806–6815 (2007)
18. Katsaga, T., Young, R.P.: Acoustic emission and x-ray tomography imaging of shear fracture
formation in large reinforced concrete beam. J. Acoust. Emiss. 25, 294–307 (2007)
19. Kalafat, S., Zelenyak, A.-M., Sause, M.G.R.: In-situ monitoring of composite failure by
computing tomography and acoustic emission. In: 20th International Conference on Compos-
ite Materials. pp. 1–8. Copenhagen (2015)
20. Seco, F., Jiménez, A.R.: Modelling the generation and propagation of ultrasonic signals in
cylindrical waveguides. In: Santos, A. (ed.) Ultrasonic Waves, pp. 1–28. Intech Open Access,
Rijeka (2012)
21. Zelenyak, A.-M., Hamstad, M., Sause, M.: Modeling of acoustic emission signal propagation
in waveguides. Sensors 15, 11805–11822 (2015)
22. Zelenyak, A.-M., Hamstad, M.A., Sause, M.G.R.: Finite element modeling of acoustic emis-
sion signal propagation with various shaped waveguides. In: 31st Conference of the European
Working Group on Acoustic Emission. pp. 1–8. Dresden (2014)
23. Sause, M.G.R., Horn, S.: Simulation of acoustic emission in planar carbon fiber reinforced
plastic specimens. J. Nondestruct. Eval. 29, 123–142 (2010)
24. Sause, M.G.R., Hamstad, M.A., Horn, S.: Finite element modeling of lamb wave propagation
in anisotropic hybrid materials. Compos. Part B Eng. 53, 249–257 (2013)
25. Sause, M.G.R., Hamstad, M.A., Horn, S.: Finite element modeling of conical acoustic emis-
sion sensors and corresponding experiments. Sensors Actuators A Phys. 184, 64–71 (2012)
26. Hamstad, M.A.: Comparison of wavelet transform and Choi-Williams distribution to deter-
mine group velocities for different acoustic emission sensors. J. Acoust. Emiss. 26, 40–59
(2008)
27. Choi, H.-I., Williams, W.: Improved time-frequency representation of multicomponent signals
using exponential kernels. IEEE Trans. Acoust. Speech Signal Process. 37, 862–872 (1989)
28. Rabinovitch, A.: A note on the amplitude-frequency relation of electromagnetic radiation
pulses induced by material failure. Philos. Mag. Lett. 79, 195–200 (1999)
608 7 Combination of Methods

29. Rabinovitch, A., Frid, V., Bahat, D., Goldbaum, J.: Fracture area calculation from electro-
magnetic radiation and its use in chalk failure analysis. Int. J. Rock Mech. Min. Sci. 37,
1149–1154 (2000)
30. Rabinovitch, A., Bahat, D., Frid, V.: Similarity and dissimilarity of electromagnetic radiation
from carbonate rocks under compression, drilling and blasting. Int. J. Rock Mech. Min. Sci. 39,
125–129 (2002)
31. Frid, V.: Electromagnetic radiation method water-infusion control in rockburst-prone strata.
J. Appl. Geophys. 43, 5–13 (2000)
32. Rabinovitch, A., Frid, V., Bahat, D.: Parameterization of electromagnetic radiation pulses
obtained by triaxial fracture of granite samples. Philos. Mag. Lett. 77, 289–293 (1998)
33. Frid, V.: Electromagnetic radiation associated with induced triaxial fracture in granite. Philos.
Mag. Lett. 79, 79–86 (1999)
34. Gade, S.O., Weiss, U., Peter, M.A., Sause, M.G.R.: Relation of electromagnetic emission and
crack dynamics in epoxy resin materials. J. Nondestruct. Eval. 33, 711–723 (2014)
35. Sklarczyk, C., Altpeter, I.: The electric emission from mortar and concrete subjected to
mechanical impact. Scr. Mater. 44, 2537–2541 (2001)
36. Sklarczyk, C., Winkler, S., Thielicke, B.: Die elektrische Emission beim Versagen von
Faserverbundwerkstoffen und ihren Komponenten. Materwiss. Werkstofftech. 27, 559–566
(1996)
37. Sedlak, P., Sikula, J., Lokajicek, T., Mori, Y.: Acoustic and electromagnetic emission as a tool
for crack localization. Meas. Sci. Technol. 19, 045701 (2008)
38. Gallego, A., Ono, K.: An improved acousto-ultrasonic scheme with lamb wave mode separa-
tion and damping factor in CFRP plates. J Acoust. Emiss. 30, 109–123 (2012)
39. Vary, A.: The acousto-ultrasonic approach. In: Duke, J.C. (ed.) Acousto-Ultrasonics, pp. 1–21.
Springer, Boston (1988)
40. Guo, N., Cawley, P.: Lamb wave propagation in composite laminates and its relationship with
acousto-ultrasonics. NDT E Int. 26, 75–84 (1993)
41. Kwon, O.-Y., Lee, S.-H.: Acousto-ultrasonic evaluation of adhesively bonded CFRP-
aluminum joints. NDT E Int. 32, 153–160 (1999)
42. Philippidis, T.P., Aggelis, D.G.: An acousto-ultrasonic approach for the determination of
water-to-cement ratio in concrete. Cem. Concr. Res. 33, 525–538 (2003)
43. Ono, K.: Special issue: acousto-ultrasonics. J. Acoust. Emiss. 12, 1–102 (1994)
44. Serir, L., Ramasso, E., Zerhouni, N.: Evidential evolving Gustafson–Kessel algorithm for
online data streams partitioning using belief function theory. Int. J. Approx. Reason. 53,
747–768 (2012)
45. Doan, D.D., Ramasso, E., Placet, V., Boubakar, L., Zerhouni, N.: Application of an
unsupervised pattern recognition approach for AE data originating from fatigue tests on
CFRP. In: 31st Conference of the European Working Group on Acoustic Emission. pp. 1–8.
Dresden (2014)
46. Anastassopoulos, A., Tsimogiannis, A., Kouroussis, D.: Unsupervised classification of acous-
tic emission sources from aerial man lift devices. In: 15th World Conference on NDT, Roma
(2000)
47. Bingham, A.H., Ek, C.W.: Acoustic emission testing of aerial devices and associated equip-
ment used in the utility industries, Ausgabe 1139. ASTM International (1992)
48. Downs, K.S., Hamstad, M.A.: Acoustic emission from depressurization to detect/evaluate
significance of impact damage to graphite/epoxy pressure vessels. J. Compos. Mater. 32,
258–307 (1998)
49. Hill, E.K., Dion, S.T., Karl, J.O., Spivey, N.S., Ii, J.L.W.: Neural network burst pressure
prediction in composite overwrapped pressure vessels. Neural Netw. 25, 187–193 (2007)
50. Marin-Franch, P., Martin, T., Tunnicliffe, D.L., Das-Gupta, D.K.: PTCa/PEKK piezo-
composites for acoustic emission detection. Sensors Actuators A Phys. 99, 236–243 (2002)
51. Barbezat, M., Brunner, A.J., Fl€ueler, P., Huber, C., Kornmann, X.: Acoustic emission sensor
properties of active fibre composite elements compared with commercial acoustic emission
sensors. Sensors Actuators A Phys. 114, 13–20 (2004)
References 609

52. Hamstad, M.A.: A discussion of the basic understanding of the felicity effect in fiber compos-
ites. J. Acoust. Emiss. 5, 95–102 (1986)
53. Awerbuch, J., Gorman, M.R., Madhukar, M.: Monitoring acoustic emission during quasi-static
loading-unloading cycles of filament-wound graphic-epoxy laminate coupons. Mater. Eval.
43, 754–764 (1985)
54. Robert, C., Dinten, J.M., Rizo, P.: Dual-energy computed tomography for ceramics and
composite materials. In: Review of Progress in Quantitative Nondestructive Evaluation.
pp. 481–488. Springer, Boston (1996)
55. Kalender, W.A.: Computed Tomography: Fundamentals, System Technology, Image Quality,
Applications. Wiley, New York (2011)
56. Petersilka, M., Bruder, H., Krauss, B., Stierstorfer, K., Flohr, T.G.: Technical principles of dual
source CT. Eur. J. Radiol. 68, 362–368 (2008)
57. Flohr, T.G., McCollough, C.H., Bruder, H., Petersilka, M., Gruber, K., S€ uss, C., Grasruck, M.,
Stierstorfer, K., Krauss, B., Raupach, R., Primak, A.N., K€ uttner, A., Achenbach, S., Becker, C.,
Kopp, A., Ohnesorge, B.M.: First performance evaluation of a dual-source CT (DSCT) system.
Eur. Radiol. 16, 256–268 (2006)
58. Momose, A., Takeda, T., Itai, Y., Hirano, K.: Phase-contrast X-ray computed tomography for
observing biological soft tissues. Nat. Med. 2, 473–475 (1996)
59. Takeda, T., Momose, A., Hirano, K., Haraoka, S., Watanabe, T., Itai, Y.: Human carcinoma:
early experience with phase-contrast X-ray CT with synchrotron radiation—comparative
specimen study with optical microscopy. Radiology 214, 298–301 (2000)
Appendix A: Acoustic Emission—Parameters
of Influence

For detection of acoustic emission signals, a variety of parameters need to be


considered during the experiment. The aim of the following list is to provide an
overview on the potential influence of some of these settings. Moreover, it is also
meant to act as a guide for selecting suitable parameters for detection of acoustic
emission signals in composites.

A.1 Acquisition System Electronics

• Threshold setting should be 3–4 dB above noise floor. If threshold is set to high a
potential loss of signals below the threshold level could occur.
• Choice of trigger settings (e.g., PDT/HDT/HLT). Erroneous setup may cause
triggering on echoes or system may not trigger on relevant signals.
• High-pass or low-pass filtered signals may be used for triggering another set of
sensors, whereas triggered channels record without filtering but higher
reliability.
• Independent triggering of channels may possibly not yield a set of signals to
allow source localization because of missing signals.
• Synchronous triggering of channels may record some signals, which are hidden
in the noise floor.
• Dynamic range of waveform recording relative to the range of amplitudes
present in the experiment. Possible saturation of signals if chosen to low or
loss of signals if they fall below threshold.
• Suitable selection of gain. Possible saturation of signals if chosen to low or loss
of signals if they fall below threshold.
• Bit resolution of acquisition card. If not sufficient, wrong frequency content for
digitized signals may result and nonlinear digitization effects may occur.
• Poor digitization rate may cause wrong recording of signal frequency content
and may increase the uncertainty for localization.

© Springer International Publishing Switzerland 2016 611


M.G.R. Sause, In Situ Monitoring of Fiber-Reinforced Composites,
Springer Series in Materials Science 242, DOI 10.1007/978-3-319-30954-5
612 Appendix A: Acoustic Emission—Parameters of Influence

• High acquisition rates of signals may lead to downtimes of system due to data
storage.
• Signal termination due to missing signal depth (amount of memory of recording
system for individual signal) may cause lack of information for signals with long
durations.
• Memory handling concepts of either onboard storage or hard disk storage. Both
concepts may lose signals during storage if interfering with other high-priority
tasks running on the system.
• Usage of waveform streaming during experiment may possibly limit the avail-
able digitization rate.
• Bad signal-to-noise ratios due to the quality of equipment or induced electronic
noise may cause wrong computation of signal features.
• High number of channels processed by one computer may saturate the electronic
processing capabilities.
• Band pass settings (either analog in preamplifier/card or digital band pass in
acquisition settings) may cut-off frequency ranges of interest.
• Dead-time after triggered signal acquisition may lose further signals arriving
within this time window.
• Application of energy or amplitude filters may skip low amplitudes or energy
signals.
• Arrival-time filters allowing only recording of signals within certain arrival time
at different sensors may lose signals not detected by more than one sensor.
• Use of load filter in fatigue testing to record only signals above a certain load
level.

A.2 Sensing Technology

• Too low sensor density may provide detectability of AE sources only within
certain range around the sensors
• Signal-to-noise ratio can be influenced by quality of the sensor and preamplifier
and the cables used
• Sensor types (e.g., resonant, multi-resonant, near flat with frequency) can have
significant impact on detected frequency spectra and signal amplitudes
• Sensor frequency characteristics (e.g., flat, resonant) can have significant impact
on detected frequency spectra and signal amplitudes
• Sensor response type (e.g., displacement sensitive, velocity sensitive, combina-
tion) can have significant impact on detected frequency spectra and signal
amplitudes
• Couplants may produce intrinsic AE during temperature exposure, may cause
additional attenuation, and may reduce transmission of shear (in-plane)
components
Appendix A: Acoustic Emission—Parameters of Influence 613

• Use of waveguides may cause frequency filtering, additional attenuation effects,


resonances and may show different sensitivity to the in-plane and out-of-plane
components than for sensor system mounted directly on specimen
• The position of the sensor relative to radiation pattern of a oriented sources may
cause substantial differences in the detected energy or frequency of the signal
• The propagation direction between the source and the sensors in anisotropic
materials may cause substantial changes to the detected signals
• The aperture of the sensor may reduce the ability to see certain (higher)
frequencies and the presence of sensors with larger apertures will also disturb
the wave propagation
• The mismatch in acoustic impedance between the sensors’ wear plate and the
material under test may cause different sensitivity
• The sensor system needs to match the input capacitance and impedance of the
preamplifier system, since they form a coupled system
• Usage of integrated preamplifier may increase signal-to-noise ratio, but comes
with less flexibility for selection of preamplifier gain
• Use of integrated preamplifiers may also simplify mounting procedure of mea-
surement chain, e.g., in industrial environments operating on cranes
• The detector position (surface side) of an AE sensor on hybrid materials may
show different sensitivity due to guided waves in individual layers

A.3 Material and Geometry

• Material may have anisotropic propagation velocities to consider for sensor


placement
• As a function of frequency (anisotropic) propagation velocities may change as
function fo frequency due to dispersion effects
• The attenuation of the inspected material needs to be considered to select
appropriate sensor positions
• Attenuation due to geometric spreading or other effects (e.g., thermoelastic
effect) may cause fast decay above a certain frequency level and may not be
detectable by narrow band acquisition systems after some distance
• The effective rise-time of an AE source is linked to the generated frequency
bandwidth and may not allow detection of certain source types with acquisition
systems of narrow bandwidth
• The physical size of the source will partially determine the amount of energy
release (e.g., bundle of filaments vs. single filament) and may cause substantial
differences in detectability of the AE signals
• For materials with high AE activity, one continuous source or multiple sources
may exist at the same spot
• The orientation of the source inside material may cause characteristic radiation
patterns and may limit detectability for certain sensor positions
614 Appendix A: Acoustic Emission—Parameters of Influence

• The depth-position of the AE source (level below surface) will govern the
relative excitation of different Lamb-wave modes
• For materials with pronounced ductility (high fracture toughness) the source
rise-times may become too long to produce AE in the ultrasonic or audible range
• Fracture of inner parts of the material (e.g., inclusions, fibers, etc.) may produce
AE due to different fracture toughness on the microscale with sufficiently short
rise-times
• For materials with very low modulus large rise-times are expected which may
become too long to produce AE in the ultrasonic or audible range
• The sample thickness will govern the number of Lamb-wave modes in the
typical AE bandwidth
• The specific specimen layup will have substantial impact on the degree of
anisotropic signal spreading and anisotropic attenuation
• The volume fraction of the constituents of the composite will define the elastic
properties and therefore also the acoustic wave velocities and acoustic
attenuation
• The diameter of reinforcement fibers may have an impact on the AE signal
amplitude, since larger diameter fibers cause a larger fracture surface and at
identical fracture toughness will produce a higher amplitude signal
• A change in temperature may induce change to the material properties or may
induce a brittle to ductile transition, which will impact the AE source rise-times
and the size of fracture surfaces, which in turn will influence the detectability,
and possible also the overall AE activity
• Environmental changes such as moisture uptake may change the material prop-
erties or the overall AE activity
• Prior or initial damage may cause increased attenuation or changes in the signal
path and potentially higher initial AE activity
• Presence of flaws near a source may affect radiation pattern in the near-field,
while larger defects may cause changes in the propagation path, e.g., reflections
• Void content may cause higher attenuation levels, or initiation of damage due to
stress concentration at pores
• Residual stresses are superimposed to external loads, which may cause an early
initial AE onset
• Autofrettage cycles may cause continuous AE signals, which cannot be detected
as transient signals
• Thermal expansion may also cause residual stresses which may cause AE signals
• Signals with multiple paths arriving at sensor location (e.g., water path, reflec-
tions at edges) may have a wrong signal arrival and may not allow identification
of guided wave modes
• Signals superimposed by echoes or reflections may not be useful for frequency
analysis or pattern recognition techniques based on frequency analysis
• Close position of the sensor relative to source may detect waves, which have not
fully developed into guided wave modes
• Distinction is necessary between the generation of volume waves, guided waves,
and surface waves
Appendix A: Acoustic Emission—Parameters of Influence 615

• The possibility of leakage of waves into fluids, attached solids, or air may need
to be considered during sensor placement and signal interpretation
• Repeatability of AE of a specimen geometry also depends on the accuracy of
geometry and the occurrence of typical manufacturing defects
• Strength distributions of the individual materials may cause scatter and charac-
teristic ranges of signal amplitudes
• Fiber orientation relative to load “axis” may cause preliminary failure of specific
(possibly disoriented) fiber bundles, while small misalignment of fibers may also
cause early matrix cracking
• Some fibers are tightly aligned and carry load before other (loose) fibers are
starting to carry the load, which may cause shifts of expected signal onsets
corresponding to fiber breakage to lower load levels
• Stress concentration effects caused by geometry intended (e.g., fracture tough-
ness testing or open-hole-tensile) or unintended (e.g., tapered areas, broken
filaments) may cause AE onsets at unexpected (low) load levels
• Generation of relative magnitude of extensional versus flexural modes may
affect source localization accuracy when based on a particular mode, since
wave modes are subject to different attenuation levels

A.4 Experimental Configuration

• Fast loading rates may cause superimposed signals or continuous AE, which will
not allow distinguishability of signals and, hence, possibilities to localize and
classify AE signals
• Attenuation effects due to propagation into adjacent media may cause a sub-
stantial reduction of detection sensitivity at some spots
• Bad signal-to-noise ratio might be caused by general background noise, presence
of continuous AE, presence of friction sources, bad cable shielding or other
effects
• Is the testing scheme suitable to generate AE or induce damage in general?
• Sufficiency of load or test level to record AE data relative to the average strength
of the item (Is the analysis based only on precursor signals, or even absence of
signals?)
• Fixtures and grips may cause additional AE (i.e., damage is induced in these
sections)
• Fatigue testing may cause extraneous noise due to friction at fixtures
• Hydraulic test equipment may cause extra noise signals due to servo valve
operation
• Pneumatic pressurized grips or actuators may cause even more noise signals
• Uniformity of loading or pressurization may cause less repeatable loading
conditions and therefore may induce non-repeatable AE onsets
616 Appendix A: Acoustic Emission—Parameters of Influence

• Uniaxial and biaxial loading conditions may cause different evolution of failure
mechanisms from specimen to specimen and therefore different generation of
AE sources, resulting in different energy release of AE sources
• For material with high AE activity, the method may also be seen as statistical
approach (lots of AE data per experiment, hence loss of some signals in dead-
time and related settings is not as relevant)
• For material with low AE activity each AE signal is important to capture since it
contributes a substantial amount to the total signature
• For the detection of AE onsets a high gain to capture low amplitude signals may
result in different interpretation than low gain measurements
• For monitoring of structural integrity, the gain needs to be chosen adequate to
capture signals representative of the damage state (these may be low or high
amplitude signals)
• For detection of failure evolution (e.g., materials testing) the gain needs to be set
to avoid saturation of high amplitude signals, while keeping it high enough to
detect low amplitude signals
Appendix B: Material Properties

The following tables list the material properties used in the analytical and numerical
calculations. For better readability, distinction is made between elastic and piezo-
electric properties as well as values used for failure criteria (Tables B.1, B.2, B.3
and B.4).

Table B.1 Elastic properties of materials used in modeling


Density [kg/m3] Poisson-Ratio Stiffness [GPa]
T800/913 (unidirectional, 60 % 1550 – C11 ¼ 154.0
fiber volume fraction) C12 ¼ C13 ¼ 3.7
C22 ¼ C33 ¼ 9.5
C23 ¼ 5.2
C44 ¼ 2.5
C66 ¼ C55 ¼ 4.2
Sigrafil CE 1250-230-39 (unidi- 1500 – C11 ¼ 131.0
rectional, 60 % fiber volume C12 ¼ C13 ¼ 3.7
fraction) C22 ¼ C33 ¼ 8.1
C23 ¼ 3.1
C44 ¼ 2.1
C55 ¼ C66 ¼ 6.1
T700/PPS (unidirectional, 60 % 1600 – C11 ¼ 152.7
fiber volume fraction) C12 ¼ C13 ¼ 4.7
C22 ¼ C33 ¼ 11.7
C23 ¼ 4.4
C44 ¼ 3.6
C55 ¼ C66 ¼ 4.5
(continued)

© Springer International Publishing Switzerland 2016 617


M.G.R. Sause, In Situ Monitoring of Fiber-Reinforced Composites,
Springer Series in Materials Science 242, DOI 10.1007/978-3-319-30954-5
618 Appendix B: Material Properties

Table B.1 (continued)


Density [kg/m3] Poisson-Ratio Stiffness [GPa]
AS4/3502 (unidirectional, 60 % 1550 – C11 ¼ 147.1 þ i3.0
fiber volume fraction) C12 ¼ C13 ¼ 4.1 þ i0.9
C22 ¼ C33 ¼ 10.6 þ i0.3
C23 ¼ 3.1 þ i0.2
C44 ¼ 3.8 þ i0.1
C55 ¼ 6.0 þ i0.2
C66 ¼ 6.0 þ i0.3
High strength alloy steel 7850 0.33 200.0
Stainless steel 7970 0.29 219.0
Aluminum AlMg3 2660 0.33 70.0
Aluminum 6063-T83 2700 0.33 69.0
Titanium Ti-6Al-4V 4430 0.40 110.2
Brass 8530 0.33 113.4
Al-filled epoxy 2700 0.40 30.0
Ag-filled epoxy 1700 0.45 2.7
HDPE 959 0.38 1.1
PTFE 2200 0.33 0.4
ETFE 1750 0.33 1.5
PPS 1350 0.36 3.8
PEI 1290 0.40 2.9
PI 1430 0.40 3.3
PVC 1760 0.40 2.9
PMMA 1190 0.32 6.2
RTM6 1140 0.38 2.8
HTA40 1770 0.20 240.0
T700S 1800 0.20 230.0
Al2O3 3965 0.22 400.0
Appendix B: Material Properties 619

Table B.2 Piezoelectric PZT-5A


properties of PZT-5A as used
Density [kg/m3] 7750
in modeling
Stiffness [GPa] C11 ¼ C22 ¼ 120.3
C12 ¼ 75.2
C13 ¼ C23 ¼ 75.1
C33 ¼ 110.9
C44 ¼ C55 ¼ 21.1
C66 ¼ 22.6
Piezoelectric constants [C/m2] S31 ¼ S32 ¼ 5.4
S33 ¼ 15.8
S24 ¼ S15 ¼ 12.3
Electrical permittivity ξ11 ¼ 919:1
ξ22 ¼ 919:1
ξ33 ¼ 826:6
Subscript indices follow the conventions for piezoelec-
tric materials

Table B.3 Electric properties of materials used in modeling


Electrical Relative
conductivity [Sm1] permittivity
RTM 6 (@10 kHz) 6.6  109 4.26
Copper 5.998  107 1
Aluminum 3.030  107 1
High strength alloy steel 4.032  106 1
Sigratex CE 1250-230-39 (unidirectional-parallel) 5.50  106 4.26
Sigratex CE 1250-230-39 (unidirectional-perpendicular) 0.204  106 4.26
Carbon fiber 1.0  103 1
PEEK (@10 kHz) 7.5  1011 3.21
PTFE (@10 kHz) <1018 2.10
PP (@10 kHz) 3.5  1011 1.59
PE (@10 kHz) 8.7  1011 2.32
PA6 (@10 kHz) 4.0  109 4.97
For non-literature values, measurement frequencies are denoted in brackets
620 Appendix B: Material Properties

Table B.4 Failure strength values of material used for failure criteria
Sigratex CE
1250-230-39 T700/PPS
(unidirectional) (unidirectional) RTM6 PPS T700S HTA40
Parallel tensile strength 1660 2400 – – – –

k [MPa]
Transverse tensile 54 20 – – – –
strength Rþ⊥ [MPa]
Parallel compressive 703 800 – – – –
strength Rk [MPa]
Transverse compres- 88 70 – – – –
sive strength R
⊥ [MPa]-
In-plane shear strength 80 63 – – – –
R⊥k [MPa]
Tensile strength – – 76 20 4061 3006
R [MPa]
Subscript notation refers to conventions used in Puck’s failure criterion
Appendix C: Acoustic Emission Signal
Parameters

This appendix is used to define the common acoustic emission parameters as used
in this work and in technical standards. The definition of a characteristic parameter
or “feature” like an energy derived from a transient signal is related to the concept
known as feature extraction. In acoustic emission these signal features have long
been used to interpret the type of failure which caused the signal. Basically, a
feature is used to reduce the amount of information carried by the signal to a single
value. As shown in Fig. C.1 for a typical voltage signal in the time domain U(t) and
frequency domain U ~ ð f Þ large numbers of features are available to describe the
signal. Although many features are already part of national and international
standards the exact feature definitions of different providers for acoustic emission
equipment slightly differ. Further, for the interpretation of acoustic emission signals
with pattern recognition techniques it is useful to define new features like the
“weighted Peak Frequency.” Since the definition and calculation of signal features
is an important step preceding the analysis routines of Chap. 4, all features used in
this book are briefly summarized. While the basic signal features in Table C.1 are
almost self-explanatory and covered by the above mentioned standards, this is not
necessarily the case for the derived features presented in Table C.2. These show
discrepancies between the different providers of acoustic emission software and are
hence defined here in the way used within this book.
Most of the features used to describe an acoustic emission signal are composed
of extreme values, derived from the signal in time- or frequency domain as shown
in Fig. C.1. The “RMS”- and “ASL”-values are used in a similar manner like in
electrical engineering, where these are used to describe the effective voltage within
a characteristic time interval. The three different count-based frequencies “Initia-
tion-”, “Reverberation-”and “Average Frequency” are used to provide an estimate
of the characteristic frequency before and after the peak-maximum, and of the
complete signal. This should not be understood as exact frequency analysis since
the number of threshold crossings is often very low and thus yields very inaccurate
results. A similar measure can be obtained from the “Rise-”and “Decay Angle.”

© Springer International Publishing Switzerland 2016 621


M.G.R. Sause, In Situ Monitoring of Fiber-Reinforced Composites,
Springer Series in Materials Science 242, DOI 10.1007/978-3-319-30954-5
622 Appendix C: Acoustic Emission Signal Parameters

a t0 tpeak

0.010 Umax

0.005
Amplitude [mV]

threshold level

0.000

–0.005

–0.010
tAE

0.00 0.02 0.04 0.06 0.08 0.10


Time [ms]

b fpeak
1.0

0.8 fcentroid
FFT-Magnitude [mV]

0.6

0.4

0.2

0.0
0 200 400 600 800 1000
Frequency [kHz]

Fig. C.1 Examples of acoustic emission signal features in the time and frequency domain

These are used to describe the angle of the slope formed by the point of first
threshold crossing and the signal maximum (rise) and the angle of the slope from
signal maximum to the last threshold crossing (decay). The “Absolute Energy” is an
absolute measure of the electrical energy measured for an acoustic emission signal
and is given in Atto-Joule [1018 J] due to their small energy. The “Frequency
Appendix C: Acoustic Emission Signal Parameters 623

Table C.1 Basic signal parameters used for feature definition


Feature Definition Unit
t0 Time of first threshold crossing (arrival time) [s]
NAE Number of threshold crossings [#]
tAE Time between first and last threshold crossing of signal [μs]
Umax Maximum signal voltage [mV]
tpeak Time of maximum signal voltage [μs]
fpeak Frequency of maximum signal contribution [Hz]
Npeak Number of threshold crossings between t0 and tpeak [#]

Table C.2 Definition of acoustic emission signal features used within this book
Feature Definition Unit
Amplitude dBAE ¼ 20logðU max=1μVÞ  dBpreamplifier [dB]
Counts NAE [#]
Duration tAE [μs]
Rise time t0  tpeak [μs]
Counts to peak Npeak [#]
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Z t0 þT
Root mean square 1 [mV]
(RMS) RMS ¼ U 2 ðtÞdt with a characteristic time TRMS for
T RMS t0
averaging ranging from 10 to 1000 ms
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Z t0 þT
Average signal 1   [dB]
level (ASL) ASL ¼ 20logðU ðtÞ=1μVÞ  dBpreamplifier dt with a
T ASL t0
characteristic time TASL for ranging from 10 to 1000 ms
Average hf i ¼ N AE =tAE [Hz]
frequency
Reverberation N AE  N peak [Hz]
f rev ¼
frequency tAE  tpeak
Initiation N peak [Hz]
f init ¼
frequency tpeak
 
Rise angle U max [rad]
ϕrise ¼ tan
tpeak
 
Decay angle U max [rad]
ϕdecay ¼ tan
tAE  tpeak
Z tAE
Absolute energy ðU ðtÞÞ2 [aJ]
W AE ¼ dt with 10kΩ input impedance of the
0 10kΩ
recording equipment
Peak frequency fpeak [Hz]
Z
Frequency ~ ðf Þdf [Hz]
f U
centroid
f centroid ¼ Z
U~ ðf Þdf
D E pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Weighted peak f peak ¼ f peak  f centroid [Hz]
frequency
Z f2 Z f end
Partial power [%]
~ 2 ðf Þdf =
U ~ 2 ðf Þdf
U
f1 f start
Frequency range of interest [f1;f2]
Frequency range of investigation [fstart;fend]
624 Appendix C: Acoustic Emission Signal Parameters

Centroid” is used to characterize the overall frequency content of an acoustic


emission signal in a similar manner like the center of mass is used to describe the
properties of geometric object with uniform density. Thus it can be understood as
additional information on the characteristic average frequency of the signal and is
in general not equal to the “Peak Frequency.” This in turn is a highly efficient
feature for discrimination of acoustic emission signals, since it is very discrete. To
improve the representation of a characteristic frequency associated with an acoustic
emission signal the “weighted Peak Frequency” was introduced to combine the
discriminative efficiency of the “Peak Frequency” with the information on the
average frequency content of the signal expressed by the “Frequency Centroid.”
A very similar approach is used by the different “Partial Power” levels. They
measure the signals frequency contribution within a given interval and are thus
used to capture the frequency composition of the signal in more than one charac-
teristic value. For the “Partial Powers” used within this work, the frequency ranges
of Table C.3 were selected as discretization of the full frequency range between
f start ¼ 0 kHz and f end ¼ 1200 kHz.

Table C.3 Definition of Partial Power frequency ranges


Feature Definition Unit
Partial power 1 f start ¼ f 1 ¼ 0 kHz; f 2 ¼ 150 kHz [%]
Partial power 2 f 1 ¼ 150 kHz; f 2 ¼ 300 kHz [%]
Partial power 3 f 1 ¼ 300 kHz; f 2 ¼ 450 kHz [%]
Partial power 4 f 1 ¼ 450 kHz; f 2 ¼ 600 kHz [%]
Partial power 5 f 1 ¼ 600 kHz; f 2 ¼ 900 kHz [%]
Partial power 6 f 1 ¼ 900 kHz; f 2 ¼ f end ¼ 12000 kHz [%]
Appendix D: Definitions and Abbreviations

This Appendix is used to introduce all abbreviations and definitions used within this
book. For better readability, symbols are grouped into different tables with refer-
ence to particular fields of science (Tables D.1, D.2, D.3, D.4, D.5, D.6, and D.7).
Table D.1 Symbols used for optics and geometry
Optics and geometry
Small increment Δ
Area A
Volume V
Angle (various uses) θ, ψ
Distances (various uses) a, b, h, s, d
Subset indices and counters (various uses) i, j, k, l, m, n
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Cartesian distance r ¼ x2 þ y2 þ z 2
Cartesian vector, coordinates r, (x, y, z)
Vector normal to surface n
Number of entities N
Intensity distribution reference subset ~ ðrÞ
F
Intensity distribution displaced subset ~ ðrÞ
G
Cost function χ
Shape function ξ
Scale and shift parameter ~ ~
a, b
Point locations P, Ō
Geometrical magnification M
Pixel size ~
P
Focal spot size ~
F
Voxel size ~
V
Distortion matrix K
Parallel to fiber axis k
Perpendicular to fiber axis ⊥

© Springer International Publishing Switzerland 2016 625


M.G.R. Sause, In Situ Monitoring of Fiber-Reinforced Composites,
Springer Series in Materials Science 242, DOI 10.1007/978-3-319-30954-5
626 Appendix D: Definitions and Abbreviations

Table D.2 Symbols used for description of waves


Waves
Scalar potential Χ
Vector potential ψ
Sound velocity, longitudinal, transversal, Rayleigh, phase, group ci, cL, cT, cR, cP, cG
Wave vector, number k, k
Wave length λ
Time t
Frequency, angular, complex angular f, ω, ω*
Acoustic impedance Z
Coefficient reflection, transmission Re, Tr
See (4.36) A~
See (4.37) ~
B
Attenuation coefficient α
Sampling interval, rate ΔT, ΔT 1

Table D.3 Symbols used for Continuum mechanics


continuum mechanics
Load L
Force F
Displacement vector, magnitude u, u
Displacement, cartesian coordinates u, v, w
Stress-tensor and vector σ
Normal stress σ
Shear stress τ
Deformation-tensor and vector ε
Normal strain ε
Shear strain γ
Elasticity-tensor, complex C, C*
Modulus E
Shear modulus G
Poisson number υ
Lame’s constants κ, ς
Density ρ
Cross-head displacement ~δ
Function input, output hE(t), hA(t)
Transfer function Ξ(t)
Appendix D: Definitions and Abbreviations 627

Table D.4 Symbols used for fracture mechanics and failure theory
Fracture mechanics and failure theory
Fracture toughness, dynamic fracture toughness Gc, Gd
Stress-intensity factor, critical, dynamic KI, KIc, KId
Crack tip speed ccr
Fracture strength R
Crack motion vector, normal b, d
Moment tensor M~
Green function Γ
^
Degradation function C ðrÞ
Stiffness vector ~
k
Unidirectional laminate strength, parallel, perpendicular R 
k , R⊥
Magnification factor m
Inclination parameters p
⊥ψ
Degradation parameter ~η
Compliance calibration factor mc
Source emission time te
Crack duration tfrac

Table D.5 Symbols used for Electrodynamics and piezoelectricity


electrodynamics and
Electric field E
piezoelectrics
Magnetic field B
Electric displacement-field D
Polarization field P
Electric charge density ~
ρ
Electric current density J
Magnetizing field H
Magnetization field M
Electric charge Q
Electric potential U
Permittivity of free space, material, tensor ξ0, ξr, ξ
Permeability of free space, material μ0, μr
Electric conductivity σ~
Electric impedance η
Skin depth δ
Charge relaxation time ~τ
Piezoelectric constants, inverse S, St
628 Appendix D: Definitions and Abbreviations

Table D.6 Pattern Pattern recognition and neural networks


recognition and neural
Number of clusters Ω
networks
Number of features, minimum K, Kmin
Number of objects in dataset O
Number of feature combinations Ntot
Separation measure between clusters J
Feature vector, element q, qi
Node weight vector, element w, wi
Error surface function ~
L
Learning rate ι
Rand index RAND
Tou index TOU
Davies-bouldin index DB
Hubert’s gamma statistics GAMMA
Silhouette value SIL

Table D.7 Common definitions for mathematical relations


Other

∂f ∂f
Nabla-operator ∇f ¼ e
∂x x
þ ∂y ey þ ∂f e
∂z z

∂f x ∂f y
∇f ¼ ∂x
þ ∂y
þ ∂f
∂z
z

  
∂f z ∂f y ∂f x ∂f y
∇f ¼ ∂y
 ∂z
ex þ ∂z
 ∂f
∂x
z
ey þ ∂x
 ∂f
∂y
x
ez
Laplace-operator Δ¼∇∇
Mean value of O objects XO
xi
h xi ¼
i¼1
O
Variance of O objects XO x 
i hxi
VarðxÞ ¼
i¼1
O
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Standard deviation of O objects u O 
u1 X 2
stdevðxÞ ¼ t xi  h xi
O i¼1
Index

A strain, 41, 42, 61, 69, 539


Acoustic emission, 44, 132, 134–146, 157, 190, Artificial defect, 71, 74, 77–80, 82,
230, 231, 245–247, 250–281, 87–97, 103, 124
285–309, 323, 337 Auld, B.A., 202
detection, 131, 192, 226–249
sensor aperture, 237–238
sensor (sensor type) B
coupling medium (couplant), 245–247 Barrett, J.F., 464
flat, 44, 230, 278 Basan, R., 33
multi-resonant, 230, 231, 278–280 Baxter, M., 290
resonant, 230, 278, 279 Bay, B.K., 493
signal Bleier, A., 33
attenuation, 190 Bunke, H., 257, 259
classification, 250–281 Bunsell, A.R., 24
counting, 132
feature extraction, 252, 255, 272, 306
noise, 251–254, 337 C
onset, 285–287 Carrere, N., 21
source (mechanism, type) Cervena, O., 134
classical localization, 287–290 Choi, H.-I., 553, 554
Δt–mapping, 290 Classification, 6–18
identification, 305–309 uncertainty of (UoC), 133, 260–263,
model, 138–146 267–270, 278–280, 315, 342
neural net based localization, 291–304 Cluster (clustering)
radiation pattern, 157, 250 natural, 251, 256–260, 311, 322
rise time, 134–137 validity (measure, index), 252, 257–260,
strength, 135, 143, 190, 323 262, 267
uncertainty of localization (UoL), Cohesive zone (model, modeling, elements),
132 140, 141
source-sensor distance, 135, 265–267 Combination of methods (method
test source (pencil lead break, piezoelectric combinations), 533–606
pulser), 293–298 Cosslett, V.E., 457
waveguide, 239–245 Creep, 18, 28–29, 31, 32, 40, 111, 196
Acousto-ultrasonics, 535, 596–599 Cuntze, R., 20, 21, 33
Anomaly, 7, 10, 38, 69

© Springer International Publishing Switzerland 2016 629


M.G.R. Sause, In Situ Monitoring of Fiber-Reinforced Composites,
Springer Series in Materials Science 242, DOI 10.1007/978-3-319-30954-5
630 Index

D source
Degradation function, 143, 146, 152, 156, 162, models, 363, 365–367, 371, 372,
167, 170, 174 374–376, 378, 380, 453
Degrieck, J., 30 radiation pattern, 381–383, 391,
Delamination 452, 453
inter-ply (interlaminar), 13–16, 34, 74, 86, test source, 389–395
89, 94, 95, 104, 149, 165, 167, 169, Electro magnetic shielding, 420–424,
170, 172, 189, 216, 224, 225, 426–427
306–309, 321, 324, 329, 387, effectiveness, 401, 414, 415, 417–421, 424,
447, 450, 475, 476, 506, 562, 426, 451, 452
568–570, 572 reduction due to
intra-ply (intralaminar), 13, 309 apertures, 420–422
Detectability of failure mechanisms (detection cable penetrations, 426–427
sensitivity) seams, 425–426
in acoustic emission, 187–192 waveguides below cut-off, 423–424
in digital image correlation, 106, 124–126 Electro magnetic wave
in electro magnetic emission, 538–540 dispersion, 362, 396
in X-ray computed tomography near-field, 397, 414
(detailvisibility), 493 propagation, 396
Deuschle, H.M., 23, 27
Digital image correlation, 102
artefact, 87, 100, 101 F
resolution, 4, 67–69 Failure, 19–21, 23, 26, 519, 520, 522, 581
stereo vision, 62, 63, 67 criterion
subset (neighborhood) anisotropic, 19
intersection, influence of, 102 differentiating, 19, 21
size, influence of, 102 global, 19, 21, 522
system accuracy, 63–69 Puck’s, 20, 23, 26, 519, 520, 581
3D, 58, 62–64, 67, 69, 71, 80–83, 101, 111, mechanism, 1–3, 6–19, 21, 30, 35, 39,
113, 114, 116, 119, 124 43, 149, 361, 363, 386, 450,
Digital volume correlation, 110, 457, 493–498, 475, 515, 533, 534, 538–542,
590, 605 550, 552, 557, 562, 565, 567,
Downs, K.S., 136 568, 575, 589, 590
type, 1, 10, 11, 19, 33, 533–535, 538–540,
543, 552, 560, 567
E Failure evolution (damage progress, damage
Eato, M.M.J., 290 progression), 3, 6, 24, 31, 35,
Edge, E.C., 23 327–348
Elasticity tensor (stiffness tensor, elastic Fatigue, 18, 28–32, 40, 111
coefficients), 114, 134, 193, 205 R-ratio, 29
complex, 202, 205, 206 Felicity effect (Felicity ratio), 336–339, 605
Electro magnetic emission (EME), 362, 363, Fiber
365–367, 371, 372, 374–376, breakage, 10, 11, 16, 30, 71, 73, 75–78, 82,
378, 380–383, 391, 401, 85, 87–89, 91–94, 99, 146, 149, 173,
403–410, 452, 453 184, 186–188, 190–192, 207, 209,
sensor (detector) 223, 279, 281, 302, 307–309, 316,
capacitive (capacitance plate), 362 318–320, 325–327, 331, 334, 335,
coil, 362, 401, 403–405, 339, 342, 387–389, 434, 453, 477,
407–408, 410 538, 552, 553, 558, 560, 566,
wire, 403, 404, 406–407, 409, 410 570–573, 599
signal bridging, 15, 149, 179–184, 189,
classification, 363 538–542
counting, 132
Index 631

filament, 7–11, 15, 24, 33, 39, 43, 103, 131, propagation, 196, 202, 212, 242,
143, 145, 173, 174, 176, 186–189, 250, 285, 543
192, 216, 223, 316, 433–436, 477, testing, 32, 198, 199, 202, 596
479, 495, 522, 526, 538, 539, 549, G€unter, S., 257, 259
572, 573, 590, 591
N-plets, 24
Fiber-matrix H
debonding, 11–13, 39, 180, 538 Hamstad, M.A., 134, 135, 412
pull-out, 11, 13, 552 Hart-Smith, L., 23
Fiber reinforced (material, composite), 2, 3 Hashin, Z., 19
fabric, 7, 8, 13, 17, 23 Haskell, N.A., 202
unidirectional, 7, 8, 14, 19, 38 High-velocity (high strain rates), 18, 31–32
Field of view Hinton, M.J., 20
in digital image correlation, 69, 72, 100, Hobrough, G.L., 57
108, 116, 117, 125, 250, 314 Hora, P., 134
In X-ray computed tomography, 493
Field testing, 346, 534, 543–547
Finkel, V.M., 365–367 I
First failure (failure onset), 2, 23, 34–35, Image segmentation, 499, 506, 508
118, 309 threshold, 499
Fracture, 158, 389 Impact damage, 16, 17, 216, 479, 480
mechanics, 1, 5–7, 22, 27, 28, 36, 69, 92, Impedance
118, 119, 121, 140, 141, 309, 311, acoustic, 195, 215, 230, 238, 239, 243
317, 369, 372, 428, 603 electric, 390
surface, 13, 14, 16, 17, 368, 370–372, 375, Indications, 44, 71, 79, 88, 110, 170, 288
378, 381, 386–389, 400, 403, 405, In situ monitoring (in situ capabilities), 42, 43,
427, 431, 433, 439, 445–452, 477, 131, 536–537
478, 503, 504, 507–513, 518–520, Internal defects, 39, 43, 69, 71–97
526, 554, 555, 580, 581, 583, 585, Irvin, G., 5
587, 588, 590, 594–596, 603
orientation, 158, 389
Freund, L.B., 369 J
Frid, V., 361, 367, 368 Joe, H., 260
Full-field
information, 112, 557
measurement, 37, 38, 70, 71, 99, K
107, 536 Kaddour, A.S., 20
Kaiser, J., 337
Kaiser effect, 337
G Kastner, J., 501
Gade, S.O., 371, 401 Keat, N., 464
Gama, B.A., 32 Kundu, T., 202
Gary, J., 134
Geometrical unsharpness, 463
Geometric spreading, 192, 203, 205, L
383, 583 Lamb, H., 195
Geometry extraction, 506–515 Lamb waves (plate waves), 135–137, 147,
Goujon, L., 230 197–201, 203, 204, 211, 212, 214,
Griffith, A.A., 5 218, 226, 237, 238, 277, 283, 288,
Guedes, R.M., 30 295
Guided waves Laminate
attenuation, 192, 204–206 cross-ply, 190, 271, 298, 299, 320–327
dispersion, 192, 204 failure, 326
632 Index

Laminate (cont.) Pattern recognition (method)


multi-axial, 18, 105, 167 supervised, 251
quasi-isotropic, 298, 299, 301 unsupervised, 251, 257, 306, 309
unidirectional, 32, 73–77, 85–87, 123, 125, Perfectly matched layer (PML), 144, 145
173, 185, 186, 214, 298, 299, Peters, W.H., 57
317–320 Piezoelectric effect, 227, 228
Landau, L.D., 192 Pinho, S., 21, 33
Lifschitz, E.M., 192 Poisson’s ratio, 113–115, 143, 146,
Lomov, S.V., 319 153, 193, 245
Lynnworth, L.C., 240 Polikar, R., 257, 292
Pollock, A.A., 203
Porosity (voids), 7, 27, 71, 204, 479, 498–501,
M 504, 506, 519, 520
Magnification factor, 24, 68, 108, 458, Pressure vessel, 29, 341, 344–347, 603
462–465, 552 Probability of detection (PoD), 132, 192, 321,
Mal, A., 202 346, 347, 363, 535
Matrix cracking, 180, 181, 192, 279–282, 309, Prosser, W.H., 203
316, 318–320, 323–325, 327, 331, Puck, A., 20, 21, 23, 27, 317
334, 339, 340, 342–344, 538, 539, Pullin, R., 290
541, 542, 552–558, 560–562,
568–572
Maxwell’s equations, 364, 396, 401 Q
Mesostructure, 21, 23 Qiu, W., 260
Microscopy
in situ, 4, 36–37
shadow, 458 R
Milligan, G.W., 260 Radon, J., 461
Miner, M., 29 Ramesh, K.T., 32
Miroshnichenko, M., 366 Ranson, W.F., 57
Misra, A., 365 Rayleigh waves, 195, 196, 368,
Mohr, O., 19 369, 432
Molotskii, M.I., 365 Reifsnider, K., 19
Moment tensor, 133, 250, 595 Representative volume element (RVE),
144–146, 148, 173, 174, 176, 178,
181, 185, 187
N Resin rich area, 9, 10, 103, 316
Nahas, M., 19
Nixon, W.C., 457
S
Schreier, H., 57
O Sch€urmann, H., 23, 28
Obstacle Scott, A.E., 24
external, 214–226, 276 Scruby, C.B., 134, 369, 370
internal, 214–226, 276, 277, 557 Seams, 0, 425, 426
Ohtsu, M., 134, 250 Sendeckyj, G., 30
Ono, K., 134, 250 Shear-horizontal waves, 197
Optical extensometer, 111–117, 536 Shearography, 42, 70, 110, 535, 546
Orteu, J.J., 57 Shiwa, M., 412
Signal
attenuation, 362
P onset, 591, 592
Pan, B., 495 Signal-to-noise ratio
Particle inclusion (foreign materials inclusion), in acoustic emission, 281
9, 71, 468 in digital image correlation (detectability of
Parvizi, A., 22 failure mechanisms), 281–285
Index 633

in electromagnetic emission, 44, 383, U


451–452, 604 Ultrasonics (active), 596–599
in X-ray computed tomography Uncertainty of classification (UoC), 363, 596
(quantummottle), 605 Undulation, 9, 10, 21, 538–543
Skin depth (penetration depth), 416, 417
Soden, P.D., 20
Sound velocity V
angular dependency, 383 Vibrometry (laser vibrometry), 32
group, 396, 575–577, 597 nonlinear, 43
modal, 195, 196
phase, 396, 575–577
Speckle pattern, 37, 58, 64–69, 71, 73, W
100–102, 112, 114–117, 119, 121, Wadley, H.N.G., 134, 250, 370
494, 495, 512, 534, 536, 546, 602 Wallner lines, 371, 513
Strain concentration (stress concentration), Wave scattering, 215
69–110, 498, 555–557, 602 Williams, W., 553, 554
Stress rupture, 18, 28–29 Wires form, 429
Sutton, M.A., 57 W€ohler, A., 29
Swolfs, Y., 522 Wu, E.M., 19

T X
Testing X-ray, 459, 460, 462–464, 473, 476, 478, 479,
bearing strength, 268, 269, 302–304 481–483, 492, 495, 502, 517, 518,
compressive, 113–116 535, 571, 573, 605
double cantilever beam, 327–334 attenuation, 486, 542
end-notched flexure, 121–123, 311–312 computed tomography
mechanical, 2, 32–44, 111, 224, 309, 314 artifacts, 464
short-beam shear, 34, 119–121, 310 cone-beam, 459, 473
single filament, 437 ex situ, 481, 502, 571
tapered double cantilever beam, 428 fan-shaped, 473
tensile, 35, 38, 40, 112–114, 334–339 in situ, 482, 483, 492, 495, 535,
transverse crack tension, 312–317 573, 605
V-notched rail shear, 116–117 voxel resolution, 483, 517
Thermoelastic dissipation (Akhieser detector, 471, 482, 484
dissipation), 203 refraction, 415
Thermography, 40–42, 535, 537, 539, 541, source, 458–460, 482–484, 486
542, 558–563, 565 focal spot size, 459, 460, 462–464, 476,
Thomson, W.T., 202 478, 479, 483, 518, 605
Transfer function, 228, 230, 233, 375, 391,
392, 401–403, 407–411, 434, 575,
577–579, 589, 604 Z
Trappe, V., 27 Zinoviev, P.A., 21
Tsai, S.W., 19

You might also like