University of Twente

Department of
Mechanical Engineering

Chair of Structural Dynamics and Acoustics

Chair of Production Technology

Detection of delaminations
based on dynamic behaviour
Application of fibre bragg gratings for
dynamic damage detection

Wouter Grouve
December 2006

CTW.06/TM-5543
Detection of delaminations based
on dynamic behaviour
Application of fibre bragg gratings for dynamic
damage detection

W.J.B. Grouve
December 7, 2006

Chair of Structural Dynamics and Acoustics,

Chair of Production Technology,
Department of Mechanical Engineering
University of Twente, Enschede
 SUMMARY / SAMENVATTING

English version

Nowadays, fibre reinforced plastics enjoy an increase in popularity. Their high spe-
cific strength and stiffness together with the high degree of design freedom have
resulted in whole new variety of applications. However, due to their nature com-
posites can suffer from complex damage types unknown to conventional materials.
Common encountered examples include matrix cracking and delaminations. The
detection of these damages can be a real challenge as it generally cannot be iden-
tified by visual inspection. The currently available non-destructive evaluation me-
thods, such as ultrasonic evaluation, are time consuming and costly. In search for
alternatives, part of the research concentrated on damage detection by monitoring
changes in dynamic behaviour of structures. The first part of this thesis gives an
overview of the existing literature on this subject. Various techniques and methods
are discussed and compared.

The second part will underline the possibilities of damage detection based on dy-
namic behaviour. A theoretical analysis will show that delamination detection
based on shifts in flexural resonance frequencies can give very accurate results.
Also it is shown that measured frequency shifts of multiple modes can be used to
locate the delaminations. The theory is experimentally validated with delaminated
composite beam shaped test specimens.

Finally, an experimental programme is conducted to investigate the use of fibre

bragg gratings for dynamic health monitoring. Fibre bragg grating have some
advantages over conventional methods for capturing the response of a vibrating
composite structure. The most notable is the fact that fibre bragg gratings can be
embedded in composite structures, which protects them from environmental influ-
ences. More advantages can be found in the possibility of multiplexing, i.e. having
multiple sensors on one fibre, and the strong resistance against fatigue. It is con-
cluded that, despite the necessity of additional research, fibre bragg gratings will
aid in the development of an applicable online health monitoring system.

i
ii Delamination detection based on dynamic behaviour

Nederlandse versie

De populariteit van vezelversterkte kunststoffen groeit momenteel enorm. De hoge

specifieke sterkte en stijfheid in combinatie met de grote mate van ontwerpvrijheid
heeft geresulteerd in een heel nieuw scala van toepassingen. Echter, de complexe
aard van composieten brengt ingewikkelde schadevormen met zich mee. Veel
voorkomende voorbeelden kunnen worden gevonden in delaminaties of matrix
cracking. Door middel van visuele inspectie kunnen deze schadevormen doorgaans
niet worden herkend. In de huidige detectie methoden, zoals ultrasoon geluid,
moet het gehele oppervlak gescand worden. Dit is een tijdrovende en dus kost-
bare bezigheid. De zoektocht naar alternatieve methoden heeft zich onder andere
gericht op detectie van schade aan de hand van veranderingen in dynamisch gedrag
van constructies. In het eerste deel van dit rapport worden de verschillende moge-
lijkheden op dit gebied vergeleken.
In het tweede deel worden de mogelijkheden van schade detectie op basis van ver-
anderingen in dynamisch gedrag benadrukt. Een theoretisch model wordt gebruikt
om aan te tonen dat delaminatie met grote nauwkeurigheid gedetecteerd kunnen
worden op basis van gemeten eigenfrequenties. Daarnaast kan, door het meten van
de verschuivingen van eigenfrequenties van verschillende trilmodes de delaminatie
gelokaliseerd worden. Het theoretische model is experimenteel gevalideerd.
Als laatste wordt, aan de hand van een experimenteel programma, de mogelijkheid
van het gebruik van fibre bragg gratings voor dynamische schade detectie onder-
zocht. Ten opzichte van meer conventionele sensoren hebben fibre bragg gratings
enkele voordelen. Het feit dat the fibres ingebed kunnen worden in een composiet
materiaal is de meest opvallende. De sensor wordt hierdoor bescherm tegen invloe-
den van buitenaf. Ook de mogelijkheid van multiplexing, dat wil zeggen het plaat-
sen van meerdere sensors op een vezel, en de grote weerstand tegen vermoeiing
kunnen worden gezien als voordelen. Ondanks het feit dat er nog veel onderzoek
nodig is, kan geconcludeerd worden dat fibre bragg gratings een positieve invloed
zullen hebben op de ontwikkeling van een toepasbaar online schade detectie sys-
teem.
 TABLE OF CONTENTS

Summary / Samenvatting i

Preface v

1 Introduction 1
1.1 General introduction . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Thesis lay-out . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 Background information . . . . . . . . . . . . . . . . . . . . . . 2
1.3.1 Composites . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3.2 Vibration based structural health monitoring . . . . . . . . 6
1.3.3 Fibre bragg gratings . . . . . . . . . . . . . . . . . . . . 7

2 Literature survey 9
2.1 General information . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.1.1 Factors to consider . . . . . . . . . . . . . . . . . . . . . 10
2.1.2 Classification . . . . . . . . . . . . . . . . . . . . . . . . 11
2.1.3 Monitoring modal parameters . . . . . . . . . . . . . . . 11
2.2 Frequency based detection methods . . . . . . . . . . . . . . . . 12
2.2.1 Literature on damage detection methods based on shifting
resonance frequencies . . . . . . . . . . . . . . . . . . . 12
2.2.2 Factors to consider when using resonance frequencies for
health monitoring . . . . . . . . . . . . . . . . . . . . . . 18
2.3 Damping based detection methods . . . . . . . . . . . . . . . . . 18
2.3.1 Literature on damage detection methods based on changes
in damping . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.4 Other methods worth mentioning . . . . . . . . . . . . . . . . . . 22
2.5 Summary and concluding remarks . . . . . . . . . . . . . . . . . 23

3 Theoretical modelling 27
3.1 Flexural vibration of delaminated beams . . . . . . . . . . . . . . 27
3.1.1 Basic assumptions and modelling . . . . . . . . . . . . . 27
3.1.2 Static solution . . . . . . . . . . . . . . . . . . . . . . . 29
3.1.3 Dynamic solution: Bernoulli-Euler beam equations . . . . 32
3.1.4 Dynamic Solution: Governing equations . . . . . . . . . . 34

iii
iv Delamination detection based on dynamic behaviour

3.1.5 Dynamic solution: Boundary and continuity conditions . . 37

3.1.6 Dynamic solution: Analytical results . . . . . . . . . . . 38
3.2 Application of the model for composite beams . . . . . . . . . . . 42
3.2.1 Determination of flexural rigidity . . . . . . . . . . . . . 42
3.2.2 Static solution . . . . . . . . . . . . . . . . . . . . . . . 44
3.2.3 Dynamic solution . . . . . . . . . . . . . . . . . . . . . . 46
3.3 Finite element analysis . . . . . . . . . . . . . . . . . . . . . . . 48
3.3.1 Beam elements with node coupling . . . . . . . . . . . . 48
3.3.2 Contact elements in combination with brick elements . . . 49
3.3.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.4 Summary and concluding remarks . . . . . . . . . . . . . . . . . 50
3.5 Nomenclature . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52

4 Experimental validation 53
4.1 Programme outline . . . . . . . . . . . . . . . . . . . . . . . . . 53
4.2 Experimental set-up . . . . . . . . . . . . . . . . . . . . . . . . . 54
4.3 Stage A: Validation of the experimental set-up and sensor evaluation 55
4.3.1 Specimen preparation . . . . . . . . . . . . . . . . . . . . 56
4.3.2 Evaluation of the experimental results . . . . . . . . . . . 57
4.3.3 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . 64
4.4 Stage B: Application of fibre bragg gratings for health monitoring 65
4.4.1 Goals and proposed testing scheme . . . . . . . . . . . . 65
4.4.2 Experimental set-up for static testing . . . . . . . . . . . 66
4.4.3 Specimen preparation . . . . . . . . . . . . . . . . . . . . 67
4.4.4 Evaluation of the experimental results . . . . . . . . . . . 68
4.4.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . 71
4.4.6 Conclusions and recommendations . . . . . . . . . . . . . 73
4.5 Stage C: Validation of theoretical analysis . . . . . . . . . . . . . 74
4.5.1 Testing scheme . . . . . . . . . . . . . . . . . . . . . . . 74
4.5.2 Evaluation and discussion . . . . . . . . . . . . . . . . . 75
4.5.3 Conclusions and recommendations . . . . . . . . . . . . . 78

5 Discussion and application 81

5.1 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
5.2 Application . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83

6 Conclusions and recommendations 85

6.1 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
6.2 Recommendations . . . . . . . . . . . . . . . . . . . . . . . . . . 87

A Data acquisition and used hardware equipment 93

B Specimen fabrication scheme 97

C Specimen dimensions and delamination parameters 99

 PREFACE

”After exhaustive research it can confidentially be concluded that the cow indeed
got killed.”
-W OUTER, Oct. 2006

The work reported in this thesis forms the climax of my study Mechanical En-
gineering at the University of Twente and provides a free pass to a sedulous and
civil life. Curious for the time to come, I realize that the last six tedious years of
eduction and personal development will inevitable be missed in the future.
The accomplishment of this thesis has been aided by a number of people. First,
I want to thank Laurent Warnet for his infectious enthusiasm and great assistance
during this research. I also want to thank the company FOS&S for providing the
tested fibre bragg gratings and for their hospitality and helpfulness during our stay
in Belgium. An honourable mention is deserved by Bert Wolbert, for his assistance
during the validation programme, and Andre de Boer, for his supervision of this
project. Last, but certainly not least, I want to thank my girlfriend for her support
and continuous encouragement.

Exam committee:
· Prof. dr. ir. De Boer
· Dr. ir. Warnet
· Dr. ir. Van der Hoogt
· Dr. ir. Vlekken

v
 CHAPTER

ONE
INTRODUCTION

I n this chapter the reader is provided with the background information necessary
to understand the theme of this thesis. The first two sections give a general problem
description and the lay-out of this thesis. In the third section general information
about composites and fibre bragg gratings is presented.

1.1 General introduction

An increase in quality together with the development of new production methods

have resulted in a growth of demand in composite materials. This goes especially
for aerospace applications. Airbus and Boeing, for example, both have plans to
manufacture an airplane in which a substantial part consists of fibre reinforced
plastics. This is because composites combine high strength and stiffness with low
weight. The application of composites however is also a cause of concern in the
field of maintenance. The complexity of fibre reinforced plastics results in un-
conventional types of damage, like transverse cracking or delaminations. These
damages can severely influence the mechanical properties of the composite mate-
rials, which in the past even resulted in catastrophic failure with loss of life.
Present methods which are able to detect these damage types are laborious and
expensive. The lack of a cheap and effective damage detection system can be seen
as a hindrance in the application of composite materials. Research in alternative
non-destructive detection methods has increased enormously over the past decades.
A part of this research concentrated on the development of an online detection
system based on changes in dynamic behaviour. This idea is based on the fact
that dynamic properties, like resonance frequencies or damping values, are directly
related to the structural properties. A change in these mechanical properties, due
to damage, results therefore in a change in dynamic behaviour. An advantage is
that some dynamic characteristics can easily and cheaply be obtained by response

1
2 Delamination detection based on dynamic behaviour

measurements at a single point on the structure, which implies that only a part of
the structure needs to be accessible. Furthermore it is proved in this thesis that
dynamic damage detection can give very good and reliable results.

1.2 Thesis lay-out

The first part of this thesis concentrates on the investigation of the possibilities and
advantages of health monitoring based on dynamic behaviour. A literature survey
is conducted in which the various possibilities on this subject are discussed and
compared. A broad overview of different techniques is given and conclusions are
drawn on present and future applicability. An analytical model is presented in
which the flexural resonance frequencies of delaminated beams can be determined.
It is shown that damage detection based on shifts of resonance frequencies is able
to detect smaller delaminations than a similar static analysis. It is also proved that
laminate lay-up plays an important role. As an addition two finite element models,
describing a delaminated beam, are also presented. The theoretical findings are
experimentally validated in a testing programme.
Furthermore the possibilities of fibre bragg gratings for damage detection based on
dynamic behaviour are investigated experimentally. Fibre bragg gratings consist
of a glass fibre with sensors for strain measurements and can be used to capture
the response of vibrating structures. Fibre bragg gratings possess some advantages
over conventional sensing systems. The first and most notable lies in the fact that
the glass fibre can be embedded in the composite structure, which allows strain
measurements inside the host material. Also embedding the fibres can be preferred
over gluing strain gauges or accelerometers for practical reasons. The fact that
strain can be measured at multiple location with only one fibre can be an advantage
as well. Various tests are conducted to investigate the applicability of these fibre
bragg gratings for health monitoring purposes.

1.3 Background information

Today, composites are widely used in all kind of fields. Generally, modern com-
posite materials combine high strength and stiffness with low density. This makes
them extremely useful in applications were weight plays an important role. First,
the reader is provided with basic background information about composites. In the
second part background information about vibration based health monitoring and
fibre bragg gratings is provided. Knowledge of this section is desired to fully un-
derstand theme of this thesis. Readers with experience in these fields can skip this
section and continue with the next chapter on page 9.
Chapter 1. Introduction 3

1.3.1 Composites

By the broadest definition, a composite material is one in which two or more ma-
terials that are different are combined to form a single component. It has to be
stipulated that the constituents retain their identity; that is, they can still be physi-
cally identified and do not merge or dissolve in one another. Composites combine
the material properties of its components in order to obtain properties, unavailable
in natural occurring materials. The constituent materials can be divided in two
categories: matrix and reinforcement. At least one fraction of both is required.
Due to the enormous variety in available matrix and reinforcement materials the
potential is tremendous.
One commonly used composite in the modern world is reinforced concrete. This
building material combines concrete and steel. Concrete is a hard material with a
very high compressive strength. In contrast, steel has a very high tensile strength.
Combining these materials results in a composite with superior mechanical proper-
ties. However when speaking about composite materials or composites today, few
refer to reinforced concrete, instead one often refers to the highly engineered com-
binations of polymers and fibre materials such as graphite, carbon or glass. This
type of composite is ordinarily known as fibre reinforced plastic or FRP.

Fibre reinforced plastics

Fibre reinforced plastics are widely used in aerospace, automotive and marine in-
dustries. They comprise a polymer matrix reinforced with fibres. The fibres in
these composites are, in most cases, used for their high strength and stiffness, while
the matrix transfers loads, binds the fibres together and protects them from harsh
environmental influences. Given the nature of FRP’s there are virtually infinite
ways of combining matrix and fibres. Every composite material can be tailored
specifically for a certain application. Fibre reinforced materials have some great
advantages over ordinary natural occurring materials. These advantages may be
summarized as:
• High specific strength and stiffness: Most fibre reinforced plastics com-
bine a high stiffness and high strength with a low weight. Compared to steel
and aluminium, fibre reinforced plastics have a much higher specific stiffness
and strength.
• High durability: Fibre reinforced plastics possess a high resistance against
corrosion. Besides FRP’s also have a fair resistance against fatigue. These
properties make fibre reinforced plastics extremely useful for applications in
harsh environments.
• Design flexibility: The anisotropic character of composites opens new pos-
sibilities for designers. Material properties can be tailored, in some extend,
4 Delamination detection based on dynamic behaviour

to meet certain design demands, which allows the designer to save material
amounts. Furthermore, fibre reinforced plastics can be formed into many
complex shapes during the manufacturing process.
• Cost-effective: Using fibre reinforced plastics can effectively reduce ma-
nufacturing costs. Parts that formerly consisted of several smaller steel com-
ponents can, with composites, be manufactured into one larger part. Also,
compared to steel tooling bodies, the initial investment in manufacturing
equipment is relatively low.
The benefits of fibre reinforced plastics over steel are notable. The research in
composites has flourished over the past decades and composites are more and more
employed over a variety of applications.

Carbon fibre reinforced plastic

Carbon fibre reinforced plastic (CFRP or CRP), is a strong, light and very expen-
sive composite material or fibre reinforced plastic. As with most composites, the
material is commonly referred to by the name of its reinforcing fibres.
Carbon fibres possess high strength
and stiffness. The plastic is most of-
ten an epoxy, but other plastics, such
as polyester, vinylester but also some
thermoplatics, are also used. All these
plastics have a low density. As stated
C-PEI layer
before the plastic binds and protects
the fibres and transfers loads between
them. The high specific strength and
stiffness causes carbon fibre reinforced
plastics to have many applications in
aerospace and automotive fields. The
production process of fibre reinforced
Stacked laminate
plastic is based on the bonding of fibres
and matrix and obtaining the purposed
product shape. Vital step in this pro-
cess is to impregnate the matrix mate-
rial in order to enclose the fibres and to
obtain the product shape. The process
in which most CFRP is made varies,
Figure 1.1: Carbon-polyetherimide layer depending on the piece being created
and stacked laminate. and how many of this particular pro-
duct are going to be produced.

The used processes vary from traditional autoclave pressing to liquid composite
Chapter 1. Introduction 5

moulding. An overview of these methods can be found in [37]. Test specimens

used in this thesis were all fabricated by hot pressing prepreg carbon reinforced
polyetherimide, the production scheme is given in appendix B. In this process la-
yers consisting of pre-impregnated carbon fibres in one direction are stacked in the
desired laminate lay-up, this is illustrated in figure 1.1. The pile of prepreg mate-
rial is then placed in a heated press in which the polyetherimide matrix is melted.
Pressure is applied to ensure good consolidation of the laminate. After the cooling
process, in which the pressure is preserved, a laminated plate is manufactured.

FRP damage types: transverse cracking and delaminations

Due to their complex nature and the required production process, FRPs suffer from
various damage types unknown to conventional materials. One of the first damage
mechanisms to occur is known as transverse cracking or matrix cracking. This type
of crack grows parallel to the fibre and in the thickness direction of the laminate.
The formation of transverse cracks does rarely mean the total fracture of a laminate,
as it does not affect the load carrying capacity of the fibres. However, transverse
cracking influences the mechanical and thermal properties of the laminate. Most
importantly, this type of cracking forms a trigger for further damage mechanisms.
Figure 1.3 shows a simple illustration of a transverse crack. In figure 1.2 it can
be seen that the crack runs between the fibre interfaces. Transverse cracks can be
caused by in-service loading, e.g. due to impact. Also the difference in thermal
expansion coefficient between fibre and matrix can, in the production phase, be a
cause of matrix cracking.
Delamination is a damage type that generally is preceded by transverse cracking.
The term delamination is directly related to the process of laminating. Just in
the same way as producing a laminate is about bonding different layers to each
other, delaminating is about debonding the layers. It is therefore a crack, which
again runs in a plane parallel to the fibres, but at the interface between two layers.
Chronologically, it is recognized that a delamination mostly initiates from the tip
of a transverse crack. Figure 1.3 also shows a delamination at the interface be-
tween the 0o and the 90o layer. Figure 1.4 shows a micrograph of a cross-section

Figure 1.2: Micrograph of cross-sectioned 90o layer in the region of a transverse crack.
6 Delamination detection based on dynamic behaviour

Transverse crack
Delamination

Figure 1.3: Schematic representation of transverse crack and delamination in a laminate.

Figure 1.4: Cross-section of a laminate with transverse cracks and delaminations.

of a laminate with both transverse cracks and delaminations. Delaminations can

seriously affect the thermo-mechanical properties of the laminate.

1.3.2 Vibration based structural health monitoring

Health monitoring consist of the continuous assessment of the condition of a struc-

tural of mechanical system. Ideally, a health monitoring system is able to detect
damage at the moment it originates. A working system should then be able to
replace current expensive maintenance routines. The idea to monitor dynamic pa-
rameters is based on the fact that modal parameters are function of the physical
properties, e.g. Youngs modulus or geometry, of the structure. Changes in these
physical properties will therefore result in detectable changes in modal parameters.
The process of modal based damage detection reduces eventually to some form of
pattern recognition problem [10]. In practice the structure under consideration is
excited, either forced or by environmental forces. The response is captured and
dynamic properties, like mode shapes, resonance frequencies or damping values,
are determined. The obtained modal data is compared to either a healthy reference
model or to some form of mathematical model. Deviations are used as a damage
indicator.
Chapter 1. Introduction 7

1.3.3 Fibre bragg gratings

A fibre bragg grating is a segment of an optical fibre that reflects particular wave-
lengths of light. This is achieved by altering parts of the fibre core so that their
index of refraction is slightly higher than normal. As a result, the structure will
transmit most wavelengths of light, but will reflect certain specific wavelengths.
This is illustrated in figure 1.5. The grating is created by burning a periodic vari-
ation in the index of refraction into the core of an optic glassfibre. The reflected
wavelength is related to the refractive index of the material and the grating period.
The basis of a fibre bragg grating based sensor system lies in the variation of this
grating period as a result of stretching or compressing of the optical fibre. Moni-
toring the shifts of wavelengths can be used to determine the strain of the grating.
By embedding fibre bragg gratings in a host structure the strain inside the host
material can be determined. This embeddability is an advantage over conventional
strain gauges. Other advantages can be found in the possibility of multiplexing,
the strong resistance against fatigue and the fact that no electric signal is required.
Multiplexing uses several bragg gratings with different reflective wavelengths on
one fibre. This way the strain at different locations can be obtained with only one
fibre. By coupling the strain measurements to material properties like thermal ex-
pansion coefficient or Youngs modulus other parameters, like temperature or stress,
can be acquired. However in this thesis only the pure strain measurements are used.

Wide spectrum Transmitted spectrum

Grating period

Reflected spectrum

Figure 1.5: Working principle of fibre bragg gratings.

 CHAPTER

TWO
LITERATURE SURVEY: HEALTH
MONITORING BASED ON DYNAMIC
BEHAVIOUR

A combination of failure of equipment, technological advancements and eco-

nomic considerations have resulted in an increase in research and publications on
the subject of health monitoring over the past few decades. Especially the prospect
of reducing costs on maintenance has been a major motive in health monitoring
research. A fully operating and reliable monitoring system could replace expen-
sive routine checks. A large part of this research concentrates on the application
of health monitoring based on dynamic behaviour. This section gives an overview
of the available literature on this subject.

2.1 General information

The general concept of dynamic health monitoring is based on the fact that modal
parameters, like natural frequencies, modal damping or mode shapes, are a function
of physical properties like mass, stiffness or geometry. A change in structural
integrity, induced by damage, will therefore result in a change in modal parameters.
Of course the opposite is also true. A measured change in modal parameters might
be preceded by a change in structural integrity and geometry. This fact could be
used in practice by subjecting the structure under consideration to a modal analysis
and compare the results with some sort of healthy reference model. Every deviation
in modal parameters (resonance frequency, modal damping or mode shapes) from
the model signifies some form of damage. This means that in fact, as stated in
[10], the process of modal-based damage detection reduces to some form of pattern
recognition problem.

9
10 Delamination detection based on dynamic behaviour

Cawley and Adams [4] noted in 1979 that the measurement of dynamic characteris-
tics of a structure as a form of structural health monitoring had a promising future.
The dynamic characteristics (e.g. resonance frequencies or damping values) of a
structure can be measured at a single point on the structure and are independent
of the chosen position, with exception of nodal points of modes of interest. This
means that dynamic testing does not require access to the whole structure, which is
an advantage over conventional ’scanning’ types of health monitoring, like the use
of ultrasound. Salawu [28] added that modal parameters can easily and cheaply be
obtained. A transducer monitors structural responses to artificially or environmen-
tally induced excitations. In general low input excitation levels are required since
the input energy is dynamically amplified.

2.1.1 Factors to consider

Besides the obvious advantages, Farrar and Doebling [10] listed some factors to
consider when applying modal data for health monitoring. The first, and most
important, is that standard modal data represents some form of data compression,
which effectively results in loss of data. The main reason of this loss can be found
in the fact that for a linear system the modal parameters are independent of ex-
citation signal characteristics (amplitude and frequency) and excitation location,
while the time histories are not. Furthermore the sampling parameters may limit
the ability to resolve the higher frequencies, which effectively also results in loss
of data.
Identifying different modes contributing to the response can in practice also prove
to be difficult because of coupling between modes that are closely spaced in fre-
quency. Especially at higher frequencies this problem is often encountered, as
modal density tends to be greater. Also, the introduction of systematic errors from
windowing data or those that arise from changing environmental conditions during
the test, will tend to make the identified modal parameters less representative. A
practical issue to take seriously into account concerns the fact that damage is gen-
erally a local phenomenon. In order to detect these local integrity changes one is
particulary interested in the local response of the structure. The fact that local re-
sponses are generally captured by higher frequency modes causes difficulties [28].
In order to produce measurable responses from a high frequency excitation more
energy is required in comparison with lower frequency excitation. Together with
the loss of information these factors form the major difficulties in current modal
based health monitoring. One logical solution would be to use the time-history
data for damage detection. However, despite the difficulties that arise with de-
tection based on modal data, it is far more difficult to detect changes in material
properties directly based on measured time histories.
Chapter 2. Literature survey 11

2.1.2 Classification

Health monitoring or damage identification methods can be classified in different

ways. Two of them are commonly known and widely used. The first classification
distinguishes two approaches, namely non-model and model based techniques. In
a non-model based technique the results are compared with the results of a mea-
surement prior to setting the structure in service. Deviances in modal parameters
indicate the presence of damage. In a model based technique the modal analysis is
compared with some form of mathematical model. This can either be an analytical
or finite element model. The advantage of using model based techniques is that
these could well be extended to give information about the location and severity of
the detected damage. General model based techniques imply that the damage on
the structure can be classified as linear. This means that an initially linear-elastic
structure remains linear-elastic after damage. The response of the structure can
still be modelled using linear equations of motion. However, when a structure be-
haves non-linear after damage, the effect is classified as non-linear damage. An
example is the opening and closing of a formed delamination, see figure 2.3. Lin-
ear equations of motion do not hold any longer and modeling the behaviour of the
structure becomes problematic. However the observation of nonlinear behaviour
in a supposedly linear-elastic structure can be a strong indicator for damage.
Another classification system for health monitoring is given by Rytter [27]. Rytter
defines four levels of damage identification.
• Level 1: Determination that damage is present in the structure.
• Level 2: Level 1 plus determination of the geometric location of the
damage.
• Level 3: Level 2 plus quantification of the severity of the damage.
• Level 4: Level 3 plus prediction of the remaining service life of the
structure.
In order to obtain level 3 damage detection a mathematical model is required, this
makes level 3 and 4 a model based technique. Non-model based techniques can
only provide health monitoring up to level 2 [27].

2.1.3 Monitoring modal parameters

There exists a wide variety of literature about modal based health monitoring. The
earliest literature studies the relation between changes in resonance frequencies or
modal damping and structural damage. Adams and Cawley [1, 2, 3, 4, 5] were
among the first to actively research this subject. A part of the available literature
about these methods will be summarized in the following two sections. Subse-
quently research concentrated also on other parameters. Zou et al. [40] and Farrar
12 Delamination detection based on dynamic behaviour

and Doebling [10] mention several options like mode shape changes, curvature
changes or thermography. The most promising techniques will be mentioned in
section 2.4.

2.2 Frequency based detection methods

Damage can influence mass, Young’s modulus or the second moment of area,
which in turn can result in a change in resonance frequencies. By monitoring these
frequencies it should be possible to acquire information about structural health.
Salawu [28] noted that shifts in resonance frequencies are probably the most use-
ful for detecting damage in structures. An additional advantage, besides the general
ones for dynamic measurements mentioned before, can be found in the fact that res-
onance frequencies show less statistical variation from random error sources than
other modal parameters. Also the knowledge on resonance frequencies is relative
great.

2.2.1 Literature on damage detection methods based on shifting

resonance frequencies

Adams et al. [1] found that damage could be roughly located by examining fre-
quency shifts of multiple vibration modes. This is based on the fact that the stress
distribution through a vibrating structure is non-uniform and different for each
mode. This means that every mode will be influenced differently by local changes
in dynamic stiffness and thus damage. If the damage is located at a point with
minimum stress for some mode, the effect on the natural frequency for this mode
will be minimal. Naturally, the opposite is also true. The effect of damage on nat-
ural frequencies is maximal in case the damage is located at points with maximum
stress. Adams also added that the size of the frequency changes may be related to
the severity of the damage.
Cawley and Adams [5] used this to develop a method to locate and roughly quantify
damage in fibre reinforced composite structures by using shifts of natural frequen-
cies only. For determining the location and severity of the damage, which is in fact
a level 3 problem, a mathematical model is required. Cawley and Adams based
damage location on the fact that the ratio, δωi /δωj , of the frequency changes due
to damage in two modes i and j is only a function of the position of damage within
the structure. The ratio δωi /δωj for different damage locations is theoretically de-
termined. Positions where the calculated ratio equals the experimentally measured
value are then possible damage sites. The theoretical frequency changes as a result
of the damage were determined by performing a sensitivity analysis. The analy-
sis of one mode pair gives a locus of possible damage sites. In order to achieve
a unique solution, several mode pairs have to be used. In the case of symmetry a
Chapter 2. Literature survey 13

minimum number of solutions equal to the degree of geometric symmetry can be

found. The most likely damage site is calculated with an error function.
Cawley and Adams [3] used this method to detect different types of damage in-
flicted on composite structures, which were representative of those encountered
in aerospace. Damage types included holes, saw cuts, crushing, impact and local
heating. The lower resonance frequencies of various structures were determined by
Cawley and Adams. It was shown that it was possible to detect, locate and roughly
quantify all these types of damage. The testing showed that it was possible to de-
tect damage equivalent to the removal of approximately 0.1% of the area of a shell
structure.
Hassiotis and Jeong [14] noted that this method has the disadvantage that a sensi-
tivity analysis of all the individual elements has to be carried out. This is a time
consuming activity and calls for a large data storage capacity for complex struc-
tures. Also, the possibility of damage in more than one location is excluded. In-
stead Hassiotis and Jeong [14] tried to reconstruct the element stiffness matrices
from changes in natural frequencies. A first-order perturbation of the eigenvalue
problem is used to yield the relation between the variation in global stiffness ma-
trix and eigenvalues. Then, the variation of the stiffness matrix is expressed as the
summation of changes proportional to the element stiffness matrices. The result is
a set of simultaneous equations that relate changes in eigenvalues to those of ele-
ments stiffnesses. Using only a small number of resonance frequencies, Hassiotis
and Jeong were able to successfully detect simulated damage of a cantilever beam
and an iron frame. Damage, in the order of 10% reduction in stiffness, was identi-
fied extremely accurately when exact data was used. However, when the data was
corrupted with random noise it proved very difficult to successfully detect damage.
This was because the changes in eigenvalues due to damage was sometimes less
then the change due to noise. If the data was corrupted with random white noise it
was only possible to detect damage equal to a reduction of 40% of the stiffness.

Transverse cracks or notch detection

The articles discussed above all describe methods for damage detection based on a
general decrease in stiffness due to damage. These methods could well be applied
to all types of damage. However, a wide variety of literature is dedicated to the
investigation of the relation between resonance frequencies and specific types of
damage, for example notches or delaminations. These damage types are often, for
research purposes, inflicted on simplified structures, like cantilevered beams. The
following three articles comprise a damage type that can be quantified as a notch
or transverse crack.
Gudmundson [11] presented a method which predicts changes in resonance fre-
quencies resulting from cracks, notches or other general geometric changes. Chang-
14 Delamination detection based on dynamic behaviour

es in resonance frequencies are shown to be dependant on the strain energy of a

static solution. Patil and Maiti [24] explained that the strain energy of a beam
containing a crack reduces because the beam can deform easier to the same ex-
tent as the uncracked beam. The reduction is found to equal the extend of energy
stored in a fictitious bending spring, which represents the crack. See figure 2.1.
Gudmundson couples this reduction of strain energy with changes in resonance
frequencies. The proposed technique was checked on three different cases. In the
first case analytical data was compared with experimental data gathered by Wendt-
land [38]. The two other cases compared the analytical results with a finite element
analysis. In all cases the analytical results compared well with the experimental re-
sults.
Transverse crack or notch

Bending spring

Figure 2.1: Bending spring representing a transverse crack or notch.

Salawu [28] noted that through width transverse cracks could well be represented
by bending springs. The local spring stiffness can be computed from the strain en-
ergy function and depends on crack depth and beam thickness. The beam is divided
in segments separated by bending springs. For every segment the Bernoulli-Euler
beam equations are solved for appropriate boundary and continuity conditions. Os-
tachowicz and Krawczuk [22] determined a flexibility function which describes the
equivalent spring stiffness for both single and double sided cracks. Chondros et al.
[7] developed a similar function, with slight differences.
Tomasel et al. [33] carried out some experiments in order to compare both [22, 7]
methods for determining equivalent spring stiffness. The measurements deter-
mined resonance frequencies using an optical experimental set-up. Both methods
showed good agreement with measured values for crack depths up to 80% of the
total height of the beam. It was, however, evident that the best approach was given
by Chondros et al. [7].

Delamination detection

Delaminations are another damage type. In fact this damage type can be seen
as the most commonly observed form of damage. Delaminations may originate
during fabrication, due to incomplete wetting or entrapped air pockets, or may
Chapter 2. Literature survey 15

be service induced, such as by impact or fatigue loading. The difficulties with

delaminations can be found in the fact that they arise inside the laminate. This
type of damage is therefore difficult to identify by visual inspection. However the
resulting degradation in strength and stiffness can be disastrous. There is a wide
variety of literature available on the subject of detecting delaminations through
dynamic analysis of structures.

(a)

(b)

Figure 2.2: Delaminated layers without (a) and with (b) coupling effects.

Ramkumar et al. [26] were among the first to present a simplified analysis to
estimate the free flexural vibration frequencies of a laminated beam with an inter-
laminar delamination. The through width delamination can exist at an arbitrary
axial location. The flawed beam is analyzed as four separate beams that are joined
together with the appropriate boundary and continuity conditions. However the
theory did not match experimental results very well. The analytical prediction was
found to underestimate the flexural stiffness.
Wang et al. [35] proposed a similar delaminated beam theory. An analytical model
was proposed based on the classical beam theory. However Wang did, in contrast
with Ramkumar, include the coupling effects between longitudinal and flexural
motion, see figure 2.2. The calculated results did now show good agreements with
the experimental results. However, Mujumdar and Suryanarayan [21] found that
the theory proposed by Wang et al. could lead to mode shapes that are not phy-
sically admissible, as in figure 2.3. They extended the theory with the assumption
that the delaminated layers of the beam are constrained to have identical transverse
displacement. Results of an experimental study on vibration of beams with mid-
plane as well as off-midplane delaminations verified the validity of the analytical
model over a wide range of values of the delamination parameters. In both studies
the basic assumption is that in vibration the delamination always remains either
open [35] or closed [21].
Lee attempted [20] to include the interaction between opening and closing vibra-
tion modes of the delaminated region. The opening and closing of the delamination
can be classified as non-linear motion which causes considerable difficulties. To
overcome these problems, Lee assumed that the period of motion can be split in
two states of mode shapes which are considered separately; one in which a dela-
mination opens and one in which delamination closes. He stated that, assuming
16 Delamination detection based on dynamic behaviour

Figure 2.3: Example of a mode shape that is physically not admissible.

that both layers have the same velocity and deflection at the transition between the
two modes, the total period T can be approximated by:

1
T = (Topen + Tclosed )
2

As result, from the relation between period and resonance frequency, the resonance
frequencies can be estimated as follows:

2 · ωopen · ωclosed
ω=
ωopen + ωclosed

In which ωclosed and ωopen represent the resonance frequencies for a particular
mode in the case of constrained or free delaminated layers. Lee noted that the
velocity matching assumption is somewhat inappropriate for longer, i.e. 60% of
the specimens length, delaminations. This is because the first vibration mode for
delamination opening show almost pure local vibration of the thin upper layer,
while delamination opening-suppressed mode is global. The result is a substantial
difference between the two cases. For the case of large delaminations a more rigor-
ous approach, which can predict interaction between open and closed delamination
states accurately, is necessary.

(a)

(b)

(c)

Figure 2.4: Vibration mode change during the period of motion for a clamped beam.
Chapter 2. Literature survey 17

Diaz Valdes and Soutis [34] measured the effect of enlarging a small delamination
in a cantilevered beam on natural frequencies. The measurements were taken in the
frequency range from 8 to 13 kHz, because it was expected that the effects of small
delamination are more significant at higher modes. A sharp scalpel was used to en-
large an artificial introduced delamination. Every time the size of the delamination
was increased a modal analysis was conducted in order to monitor changes in nat-
ural frequency and damping. In this fashion the delamination was gradually grown
from 0.23% to 3.84% of the total surface area. This type of artificial delamina-
tion is thought to better represent damage patterns observed in fatigue loading than
through width delaminations with sizes covering 5% to 60% of the total surface
area. The results show clearly that all resonance frequencies shift to lower values
with increasing damage size. The absolute frequency changes are quite large and
detectable, i.e. in the order of ten to a few hundred Hz. There is also a small attenu-
ation of the peak level and sharpness as damage grows, which indicates an increase
in damping. More on this subject can be found in section 2.3.
Tenek et al. [32] studied the effect of delaminations on the natural frequencies of
plates. A finite element method was adopted for this purpose. The method showed
good comparison with experimental results for the case of cantilever laminated
plates. The results for delaminated plates indicated that even for large flaw sizes,
the first few resonance frequencies are not significantly affected. However, for
higher frequencies an overall reduction proportional to the delamination size was
observed. There is a possibility that at certain excitation frequencies, delamina-
tions exhibit independent dynamic plate behaviour and vibrate out of phase with
the rest of the structure. Tenek et al. also studied the possibilities of utilizing ther-
mographic and thermoelastic non-destructive evaluation techniques. They found
that at particular frequency regions, local flaw vibration causes dissipation of me-
chanical energy into heat, which in turn can be detected. This is further discussed
in section 2.4.
Chrysochoidis and Saravanos [8] found an apparent linear relation between the
number of resonance peaks in a frequency response plot and the size of a delam-
ination. Experiments show a clear increase in resonance peaks between 0 and
2kHz. This phenomenon can be partially attributed to the appearance of additional
opening and sliding mode shapes in the delaminated region, which vibrate either
individually or coupled with the bending modes of the beam. Also the shifting of
resonance frequencies to lower values as the delamination increases contributes to
the increased modal density.
The methods described above all concern a linear damage effect. The structures
all behave linearly after the inflicted damage. In reality however, it’s questionable
whether this assumption will hold. One can easily imagine that an originated crack
will open and close under varying load. Damage detection can in this case reduce
to identifying non-linear behaviour.
18 Delamination detection based on dynamic behaviour

2.2.2 Factors to consider when using resonance frequencies for

health monitoring

In some of the techniques mentioned above the damage is modelled as either a

bending spring or delamination. Changes in resonance frequencies are calculated
for these damage types. A fundamental limitation of these methods is that it should
be possible to represent the actual damage with these models. This will not always
be the case. Salawu [28] lists other factors to consider when using shifts in re-
sonance frequencies for damage detection. For successful application of vibration
data, measurements should be taken at points where all the modes of interest are
well represented. The simplest way of achieving this is to perform a theoretical vi-
bration analysis prior to testing in order to identify the desired measurement points.
A full understanding of the effect of damage on resonance frequencies is vital.
Also an idea of the sensitivity of changes in support or environmental, such as tem-
perature or humidity, conditions is required. Changes in resonance frequencies of
about 5% are measured as a result of changes in ambient conditions only. Quan-
titative relationships between resonance frequencies and environmental factors are
rare. This is because many variables are involved, which makes it difficult to assign
numerical factors. The effect of pre-loading and residual stress on frequency re-
sponse was tested by Kawiecki [16]. From the various tests it can be concluded that
changes in loading conditions or residual stresses can cause resonance frequencies
to shift. As a result of these external influences, the changes in frequencies should
be considerable in order to detect damage with confidence.
Other factors to consider are the consistency and reliability of measuring proce-
dures. Besides, Doebling et al. [9] mention that the somewhat low sensitivity
of the lower resonance frequency shifts to damage requires very precise measure-
ments or large levels of damage. Currently, using frequency shifts to detect damage
appears to be more practical in applications where such shifts can be measured very
precisely in a controlled environment, such as for quality control in manufacturing.

2.3 Damping based detection methods

There is little literature available on the effect of damage on damping. Zou et al.
[40] mention that damping is more sensitive to damage than stiffness or mass. This
makes damping a potential interesting parameter to monitor for damage detection.
Chandra et al. [6] give two types of damping due to damage. The first consists
of frictional damping, which is a result of slip in the unbound regions between
fibre and matrix interface or delaminations. The other type is damping due to
dissipation of energy as a result of visco-elastic deformation in the area of matrix
cracks, broken fibres etc. In the following section a part of the available literature
is reviewed.
Chapter 2. Literature survey 19

2.3.1 Literature on damage detection methods based on changes

in damping

Guild and Adams [12] described a method in which the measurements of specific
damping capacity (SDC) are used to detect very small cracks in unidirectional
glass-fibre reinforced polyester. The specific damping capacity is defined as the
ratio of the energy absorbed by the beam during one cycle to the total strain energy
stored by the beam during that cycle. The energy stored can be calculated from
the power input to the exciting coils. The total strain energy absorbed by the beam
can be calculated from the deflection measured at a point, together with the the-
oretical mode shape. A change of SDC is based on the principle that cracks will
cause a significant loss of energy. Guild and Adams extended a method proposed
by Adams and Bacon [2] in which measurements of SDC of composite materials
could provide relative insight into their lay up. Previous result in the article by
Adams and Bacon indicated that the method could be used for the detection of
cracks. Guild and Adams developed an apparatus that is able to excite beams in a
free-free flexural vibration, the apparatus is described in [13]. Guild and Adams
[12] introduced cracks in the central region of glass fibre reinforced plastic beams
using four-point flexural loading. The growth of a crack was monitored using
acoustic emission. The beams were excited and the specific damping capacity was
determined. With the proposed method Guild and Adams were able to successfully
detect small cracks in glass fibre reinforced plastic beams. The measurements indi-
cated a notable change in SDC after cracking was observed by acoustic emission.
These cracks were small and not visible by naked eye.

Kyriazoglou et al. [18] measured the specific damping capacity of different types
of composite beams in flexural vibration. An experimental set-up similar to the
one used by Guild and Adams [13] was used. Tests have been conducted before
and after damage was introduced. Homogeneous damage was induced using quasi-
static loading in tension or as a result of cyclic loading. The results showed that
the specific damping capacity is more sensitive to damage than resonance frequen-
cies. Transverse cracks in continuous GFRP as a result of quasi-static loading were
clearly measurable. Also for damaged cross-ply GFRP specimens a clear change
in SDC was noticed. It should be mentioned however that the change is SDC is
dependant on laminate lay-up. The cracks in the transverse layer have far greater
effect on the SDC when they are in the outer plies of the beam. The explanation
for this behaviour is believed to arise from the constraint imposed on the crack
opening by the outer plies. However, a change in specific damping capacity was
also measured when cracking occurred in the inner plies. The changes in SDC for
woven GFRP laminates were not so consistent. When, in quasi-static loading, the
applied strain exceeded 0.6% a significant increase in SDC was observed. How-
ever, no consistent change was measured after application of higher levels of strain.
As for the cyclic loading, the specific damping capacity clearly increases with an
increasing number of cycles. It is concluded by Kyriazoglou et al. that damping
20 Delamination detection based on dynamic behaviour

properties are sensitive to the presence of cracks. By monitoring the specific damp-
ing capacity damage can successfully be detected.

Lee et al. [19] measured damping coefficients with use of an impulse technique
developed by Sun et al. [31]. Specimens are cantilevered and excited by a hammer
blow. A non-contact motion transducer monitored the displacement of the free
end. Signals acquired by the motion transducer and hammer were fed into a FFT
analyzer which displayed a frequency response function. Damping was derived
from this function. Lee et al. utilized the technique to measure the influence of
different types of damage on damping. The following conclusions were drawn
from the measurements:

• For notches, the experimental results show that the changes in damping for
different modes depend on the location of notches. That is, the change in
damping may be significant in one mode and may not be in another mode.
A similar phenomenon was encountered earlier in section 2.2 where it is
explained that the ratio of frequency change for two modes could also be
used to predict damage location. The same technique could be employed
with damping changes for two modes.

• Rather disappointing results were obtained when Lee et al. tried to measure
damping changes as a result of matrix cracking. Cracks were introduced
in specimens by means of a three point bending test. Even for the speci-
men which was bend with a force equal to 90% of the maximum bending
force, before failure, no significant increase in damping was observed. Lee
et al. believed that the changes in damping due to matrix cracking may be
overwhelmed by the changes due material imperfections. In addition it was
noticed that the damping measurement was strongly dependent on clamping
conditions.

• For artificial induced delaminations the increase in damping is proportional

with the size of the delaminated area. Generally speaking it seemed that the
damping measurements of the second and the third modes show significant
changes while the first mode has the lowest sensitivity.

• In order to analyze the effect of material imperfections, specimens with in-

clusions were studied. The measurements show little change in damping.
This is probably due to the fact that the inserts are well bonded with the
laminate, so that there is no real friction between insert and laminate.

Overall Lee et al. concluded that the impulse technique is quite capable of detect-
ing both severity and location of the notches. Also for delamination the technique
is quite promising. For matrix cracks and material imperfections however, the
changes of damping are insignificant. Lee et al. suggested that in order to detect
microdamage the wave length of the impulse should be in the same order or less
than that of microdamage.
Chapter 2. Literature survey 21

Kawiecki [16] used piezotranducers to record a frequency transfer function of spe-

cimens. The modal damping magnitudes were determined using the halfpower
point approach. Damage was simulated with viscoelastic patches. All measure-
ments show that changes in damping are clearly visible and detectable. Kawiecki
concludes that the results are an important complement to monitoring methods
based on detection of changes in stiffness.
Although sufficient literature is available on crack or delamination prediction and
propagation, the theoretical work on damping behaviour in delaminated compo-
site structures is almost non existent [6]. Saravanos and Hopkins [30] developed a
laminate theory which predicts resonance frequencies and modal damping. Ana-
lytical results of modal damping in graphite epoxy composite beams with a central
delamination are compared with experimental measurements. The analytical re-
sults show reasonable comparison for modal damping for small and medium, i.e.
smaller than 30% of the specimens length, delaminations. However, the effects of
delaminations on the dynamic characteristics (resonance frequencies or damping)
are very dependent on laminate configuration.
Diaz Valdes and Soutis [34] measured the effect of enlarging a small delamination
on the modal damping of a cantilevered composite beam. The measurements were
conducted in the frequency range of 8 to 13kHz, because it was expected that the
effects of small delamination are more significant at higher modes. It was shown
that damping increases with increasing delamination length.
Chrysochoidis and Saravanos [8] also carried out some experiments in order to in-
vestigate the effect of delaminations on modal damping. It is found that damping
seems to increase with the size of delamination. However the relation seems less
apparent than the relation between resonance frequencies and delamination size.
Also the effect becomes more notable at higher modes. Chrysochoidis and Sara-
vanos concluded that, for the time being, modal damping may not be effectively
used for delamination detection.
The fact that damping is more sensitive to damage [40] makes it an interesting
parameter for health monitoring. However there is not much literature available
on the subject, which calls for an increase in research. The effects of damage on
material damping have to be investigated more thoroughly, in order to successfully
use changes in damping for damage detection. Also there are some factors to
consider. Most of the ones mentioned in section 2.2.2 apply as well for the case
of damping. For the external effects for example, Kyriazoglou mentioned that
air damping has been found to significantly affect damping measurements [18].
Kawiecki [16] also investigated the effect of pre-loading and residual stress on
damping and found that these factors also influence measured modal damping.
22 Delamination detection based on dynamic behaviour

2.4 Other methods worth mentioning

Considerable effort has been spent in obtaining a relation between resonance fre-
quencies or damping values and the presence and location of damage. However,
a part of the research investigated alternative techniques. The techniques worth
mentioning are discussed in this section.

Yuen [39] investigated the influence of damage on the mode shapes of a cantilever
beam. A finite element programme was used and damage was modelled by a re-
duction in elasticity. Yuen expected that the mode shapes representing each of the
degrees of freedom would be affected differently by the presence of damage. Also
it was expected that the changes in the mode shapes would reflect the location and
extend of the damage. The damage zone should behave like a spring allowing a
greater degree of motion for points beyond the damage zone. The mode shapes of
a cantilever beam with varying damage location were calculated. Results indicate
that there is a systematic change in the first mode shape with respect to the dam-
age location. However for higher modes this does not apply. The mode shapes of
higher modes are too complex and further investigation is needed in order to use
these modes for damage location.

Pandey et al. [23] extended this research and introduced a new parameter called
curvature mode shape. Curvature mode shapes are related to the flexural rigidity of
a beam. Curvature is defined by ν ′′ = M/(EI), in which ν ′′ is the curvature, M is
the bending moment and EI is the flexural rigidity. In case of damage the flexural
rigidity EI will reduce in that region, as a result the curvature will increase there.
As curvature is a function of EI one could easily relate the change in curvature
to the severity of damage. Finite element experiments were carried out on both a
cantilevered and a simply supported beam. Differences between curvature mode
shapes of damaged beams and intact ones were analyzed. The largest difference
could be found at damage regions. The differences in curvature are relatively small
outside the damage region, which makes it easy to identify damage. This is an
advantage over the method proposed by Yuen [39], in which it proved difficult to
identify the exact damage location. Also the method of Pandey is not restricted
to the first vibration mode. The fact that curvature is proportional to the bending
strain, makes it an interesting parameter. Curvature mode shapes can be easily
obtained by measuring strains instead of displacement or acceleration.

Sampaio et al. [29] extended the theory given by Pandey [23] and developed the
frequency response function curvature method. This method uses the curvature on
all frequencies in the measurement range and not just the modal frequencies. In
order to obtain the curvature for each frequency, responses at different locations of
the damaged structure are needed. The curvature is given by

α(ω)i+1,j − 2α(ω)i,j + α(ω)i−1,j

α′′ (ω)i,j = ,
h2
Chapter 2. Literature survey 23

where αi,j is the receptance measured at location i for a force input at location j
and h is the step size. Positions where the difference between intact and damaged
FRF curvature maximizes, can be identified as probable damage locations. Two
tests show that the FRF curvature method performed well in detecting and locating
damage. Its main advantage is its simplicity the fact that no modal analysis needs
to be performed for the identification of mode shapes or resonance frequencies.
In an article by Kim [17], damage identification is based on reconstructed fre-
quency response functions (FRF). The possibility of using the residual FRF, i.e.
the difference between intact and damaged FRFs, as a damage indicator is investi-
gated. Also changes of the area under the FRF-curve are used as damage indicator.
Kim concluded that use of changes in area under the FRF-curves cannot be used as
a decisive damage detection parameter. However in combination with changes in
resonance frequencies and damping it can prove reliable information on the detec-
tion of damage.
Hu et al. [15] used mode shapes, obtained from a modal analysis, from both exper-
imental and finite element analysis of laminated plates. The mode shapes were
then used to calculate strain energy using a technique known as the differential
quadrature method. The local strain energies of both intact and damaged plates
were used to identify surface crack location. Experimental results showed that
surface crack location in various composite laminates can successfully be detected.
A different technique was proposed by Pye and Adams [25]. Infra-red thermo-
graphy was used to give an indication of variation in the surface temperature of the
structure under test. As the structure is subjected to cyclic stresses, as a result of
resonance, temperature variations are caused by local increase in energy dissipation
which then again is a result of damage. Testing showed that this can be an effective
technique for location damage. Temperature differences in the order of 0.5o C were
measured. However for CFRP materials this technique might cause difficulties
because of the high thermal conductivity of carbon.

2.5 Summary and concluding remarks

Several techniques and methods for damage detection based on dynamic behaviour
have been discussed. Most of them consisted of monitoring shifts in resonance
frequencies or changes in modal damping. An overview of the advantages and
disadvantages is given in this section.
In section 2.2 health monitoring based on frequency shifts are discussed. The arti-
cles discussed can be summarized:
• Resonance frequencies show less statistical variation compared to other modal
parameter, this is large advantage over other methods [28].
24 Delamination detection based on dynamic behaviour

• Resonance frequencies can be used as a damage indicator. As a result of

changes in stiffness or second moment of area, resonance frequencies tend
to shift [1, 3, 5, 14]. Since stress distribution is non-uniform during vibration
every mode will be affected differently by damage location. As a result shifts
of multiple modes can be used for locating damage [3, 5].
• The influence of several damage types on resonance frequencies shifts have
been modelled in the past. The models concerned transverse cracks and
notches [7, 11, 22, 33] or delaminations [8, 20, 21, 26, 32, 34, 35]. These
methods could be used as a basis for level 3 damage detection [27]. From
these models it is shown that laminate configuration has a large influence
on resonance frequency shifts. A fundamental limitation, however, of these
methods is that it should be possible to represent the actual damage with
these models. This will not always be the case.
• Exogenic effect can influence measurements, resonance frequency shifts ex-
ceeding 5% have been recorded as a result of environmental changes only
[28, 16]. Also the low sensitivity of resonance frequency changes to certain
damage can cause difficulties [9]. Successful damage detection, based on
resonance frequency shifts only, calls for either very precise measurements
or severe damage.
Section 2.3 discussed health monitoring based on changes in modal damping:
• Experiments showed that damping is more sensitive to damage than reso-
nance frequencies [40]. Measurements showed that changes in damping are
highly related to laminate lay-up [30]. Despite the sensitivity, the relation
between damage and changes in damping seems not very apparent [8].
• Damping seems to increase with damage [12, 18]. However for matrix crack-
ing or material imperfections, changes in modal damping are rather small
[19].
• Also for damping it applies that exogenic effect has an influence on measure-
ments. It has been reported that air damping can be found to significantly
affect damping measurements [18]. Also pre-loading and residual stress can
influence the measured modal damping values [16].
There are other methods available for dynamic measurement based damage detec-
tion, these are discussed in section 2.4. Changes in mode shapes [39] or curvature
mode shapes [23, 29] were shown to be promising parameters for damage detec-
tion. Especially, results obtained with help of the curvature mode shapes were
encouraging. A disadvantage of these methods is that several measurement points
are required to qualify the different mode or curvature mode shapes.
 Authors Year Ref. Frequency Damping Model Predefined Remarks
based based based damage
R EVIEW ARTICLES
Farrar, Doebling, Prime and Shevitz 1997 [10] yes yes na na
Salawu 1997 [28] yes no na na
Zou, Tong and Steven 2000 [40] yes yes na na
T HEORETICAL MODELS FOR DAMAGE DETECTION
Cawley and Adams 1978 [5] yes no yes no
Wang, Liu and Gibby 1982 [35] yes no yes yes models delaminations
Gudmundson 1982 [11] yes no yes yes
Yuen 1985 [39] no no yes no uses mode shape changes
Mujumdar and Suryanarayan 1988 [21] yes no yes yes models delaminations
Pandey, Biswas and Samman 1991 [23] no no yes no uses curvature changes of beams
Hassiotis and Jeong 1993 [14] yes no yes no
Tenek, Henneke and Gunzburger 1993 [32] yes no yes yes
Saravanos and Hopkins 1996 [30] no yes no no
Chondros, Dimarogonas and Yao 1998 [7] yes no yes yes models notches
Sampaio, Maia and Silva 1999 [29] no no yes no uses curvature changes of beams
Lee 2000 [20] yes no yes yes models delaminations
E XPERIMENTS ON DAMAGE DETECTION
Cawley and Adams 1978 [3] yes no yes no
Cawley and Adams 1981 [1] no yes no no
Lee, Sun and Liu 1987 [19] no yes no no
Tomasel, Larrondo and Laura 1999 [33] yes no yes yes notch detection
Diaz Valdes and Soutis 1999 [34] yes yes no yes delamination detection
25

Kawiecki 2001 [16] yes yes no no

Kyriazoglou, Page and Guild 2004 [18] no yes no no

Table 2.1: A selection of literature on damage detection based on dynamic behaviour.

 CHAPTER

THREE
DETERMINING THE RESONANCE
FREQUENCIES OF A DELAMINATED
BEAM

I n the following chapter the model presented by Mujumdar and Suryanarayan

[21] is outlined. This model determines the resonance frequencies of a beam con-
taining one ’through width’ delamination. To provide some insight into the me-
chanics of the model a solution to the static problem is added. In the second section
the model is modified to make it applicable for composite materials. Finally two
finite element models are presented.

3.1 Flexural vibration of delaminated beams

Delamination is a common detected type of damage in composite components and

can have a great influence on mechanical properties. As stated in previous sections
delaminations may originate during fabrication or from in-service loading. The
notion that delaminations not only affect strength but also cause a reduction in
bending stiffness the basis of the model described by Mujumdar and Suryanarayan
[21]. The model is an extension of the theory proposed by Ramkumar et al. [26]
and provides an analytical solution to the eigenvalue problem and determines the
resonance frequencies of delaminated beams in flexural vibration.

3.1.1 Basic assumptions and modelling

The model presented by Mujumdar and Suryanarayan [21] describes a beam with
an arbitrarily located through-width delamination. Figure 3.1 shows a delaminated
beam subdivided in three separate sections, namely two integral sections and one

27
28 Delamination detection based on dynamic behaviour

delaminated section. The delaminated region is then again subdivided in com-

ponent segments above and below the delamination plane. The four segments are
each treated as Bernoulli-Euler beams. The solution for the whole beam is obtained
in terms of the solutions of all the separate sections by satisfying the appropriate
boundary and continuity conditions. For the sake of simplicity, Mujumdar and
Suryanarayan assumed the beam to be homogeneous and isotropic. Also only a
single delamination is considered, though the analysis can be easily extended to
multiple delaminations. It is further assumed that there exists no gap between the
delaminated layers.

Integral section Delaminated section Integral section

V2 ,M2 ,N2 V2 ,M2 ,N2

V1 ,M1 x2 z 2 V4 ,M4
x1 z x4 z4
1
x3 z 3

V3 ,M3 ,N3 V3 ,M3 ,N3

a b c

Figure 3.1: Geometry, co-ordinate system and modelling of a beam with a delamination.

The basic theory is based on variations in the distribution of shear stress. In a

beam that is in nonuniform bending both normal and shear stress is developed. In
a beam without delamination the shear stress is distributed as shown in figure 3.2a.
However, in case a delamination is present, the layers are free to slide and the shear
stress on the delaminated surface therefore equals zero per definition. The shear
stress distribution for this case is shown in figure 3.2b. This discontinuity in shear
stress results in a decrease of bending stiffness which then again will cause the
flexural resonance frequencies to shift to lower values.
Chapter 3. Theoretical modelling 29

N
τ a.
N
τ
τ
b.
a. b.

Figure 3.2: Distribution of shear stress in Figure 3.3: Different continuity condi-
(a) an intact beam section and (b) a delam- tions at junction between delaminated and
inated beam section. integral section.

The difference between this model and the one proposed by Ramkumar et al. [26]
lies in the axial continuity conditions at the junctions between integral and de-
laminated section. Ramkumar, as opposed to Mujumdar and Suryanarayan [21],
neglected the coupling effects between axial and flexural deformation. This is best
illustrated in figure 3.3. Mujumdar and Suryanarayan demand that the free ends
of the delaminated segments stay in the same plane. In order to realize this the
normal force N is introduced, as in figure 3.3b. Figure 3.3a shows the situation
as proposed by Ramkumar et al. The absence of normal force N results in a dis-
continuity at the end plane. The delaminated section acts entirely equivalent to a
beam with a step reduction in flexural rigidity corresponding to a reduced second
moment of area equal to (I2 + I3 ). It was already shown by Ramkumar et al. [26]
that the flexural rigidity of a delaminated beam is considerable underestimated this
way. The compatibility condition used by Mujumdar and Suryanarayan introduces
an axial load system of equal and opposite forces. These normal forces compress
one segment and stretch the other such that their ends lie in the same plane. This
load system results in an internal bending moment, which contributes to the total
bending stiffness of the beam.

3.1.2 Static solution

In order to gain insight in the theory proposed by Mujumdar and Suryanarayan

a solution to the static problem is presented first. The results are compared with
results obtained by the model proposed by Ramkumar et al. [26]. As stated above,
Mujumdar and Suryanarayan included the coupling effects between longitudinal
and flexural motion. For this static analysis, the delaminated section is clamped
at one end, as is shown in figure 3.4, and a force F0 and bending moment M0 are
applied to the free end. Figure 3.5 show the forces that act on each separate beam
segment. The normal forces N2 and N3 stretch one beam segment and compress
the other such that their free ends lie in the same plane.
30 Delamination detection based on dynamic behaviour

EI2
x2
h3 z2
M0
h x3 Delamination surface
h2 z3

EI3 F0

Figure 3.4: Geometry, co-ordinate system and modelling of a delaminated beam section.

Equilibrium of force in both z-

and x-direction and bending moment
D2 ,N2 ,M2 h2
give the following equations:
del. surf.
N2 + N3 = 0 θ
h3
D2 + D3 = F0 D3 ,N3 ,M3
F0 , M0
1
M2 + M3 − N 2 h = M0 (3.1)
2
Figure 3.5: Free body diagram of
The bending moment is taken with re- end plane.
spect to the junction between delami-
nation surface and endplane.

In case the deflection is assumed to be small the slope and displacement in x-

direction at the free end for the separate beams follow from the bending moment
equation and are given by

Mi L Di L2
u′i (L) = ϕi (L) = +
EIi 2EIi
Mi L2 Di L3
ui (L) = + (3.2)
2EIi 3EIi

with i = 2, 3. The displacement in axial direction is calculated by:

Ni L
wi (L) = (3.3)
EAi

The Young’s modulus is represented by E, the second moment of area by Ii and the
cross sectional area by Ai . Continuity of x-displacement and slope of both beam
segments at the free end is demanded. Also the free end of both beams should
remain in the same plane. This results, with figure 3.5, in the following continuity
Chapter 3. Theoretical modelling 31

conditions:

ϕ2 (L) = ϕ3 (L)
u2 (L) = u3 (L)
1
w3 = w2 + δ = w2 + hϕ2 (L) (3.4)
2

Substituting equations 3.1, 3.2 and 3.3 in the continuity conditions 3.4 the follow-
ing system of equations is found:

L L L2 L2 hL L L2
M2 ( + ) + D2 ( + ) − N2 = M0 + F0
EI2 EI3 2EI2 EI3 2EI3 EI3 2EI3
L2 L 2 L 3 L 3 hL 2 L 2 L3
M2 ( + ) + D2 ( + ) − N2 = M0 + F0
2EI2 2EI3 3EI2 3EI3 4EI3 2EI3 3EI3
hL d1 L 2 L L
M2 + D2 + N2 ( + ) = 0
2EI2 4EI2 EA2 EA3

From this set of equations the three unknown parameters M2 , D2 and N2 can be
derived and the transversal displacement and slope at z = L can be determined
by substituting these in equations 3.2. For a fictive beam the results are obtained
and listed in table 3.1. The table also shows the results for a beam without a
delamination and in case the delamination is treated as a step in second moment
of area, as proposed by Ramkumar [26]. The delamination is located at a fraction
h2 /h of the beams thickness.
It can be seen from table 3.1 that, in both models, the effect of the delamina-
tion increases as it is positioned closer to the neutral line. The solution of both
displacement and slope approach the intact case when alpha approaches zero or
one. The differences between the results obtained with Ramkumar et al. and Mu-
jumdar and Suryanarayan are significant. In the model used by Mujumdar and
Suryanarayan the loss of flexural rigidity is compensated by the additional bending
moment caused by the stretching and compressing of the layers.

No delamination Ramkumar Mujumdar

et al. Suryanarayan
h2 /h F0 M0 u(L) ϕ(L) u(L) ϕ(L) u(L) ϕ(L)
0.2 1 0 .5 .75 1.0 1.4 .615 .75
0.2 0 1 .75 1.5 1.4 2.9 .75 1.5
0.4 1 0 .5 .75 1.8 2.7 .821 .75
0.4 0 1 .75 1.5 2.7 5.4 .75 1.5
0.5 1 0 .5 .75 2.0 3.0 .875 .75
0.5 0 1 .75 1.5 3.0 6.0 .75 1.5

Table 3.1: Comparison of static solutions

32 Delamination detection based on dynamic behaviour

In case the beam is in pure bending

the deflection and slope equal the in- 1 Del.
tact case. This makes sense, since no

displacement
Intact

Normalized
shear stress is developed in the beam
and therefore no sliding of the layers 0.5
occurs. On the other hand when a
shear force is applied the layers will
slide and the effect of a delamination 0
0 0.5 1
is noticed. The normalized shape of a Dimensionless distance z/L
delaminated beam subjected to a shear
force at the fee end is given in figure Figure 3.6: Shape of a beam subjected to
3.6. It can be seen that only the deflec- a shear force.
tion at the free end is influenced, the
slope equals the intact case.

It seems, from figure 3.6, that the delamination has a huge effect on the total dis-
placement of the beam. However normally only a fraction of the beam will be
delaminated and the influence on the total displacement will be minimal. This is
illustrated in figure 3.7 where the normalized shape of a delaminated beam under
a shear force is plotted. The dimensionless length of the delamination is varied.
Even for delaminations with a length up to 40% of the beam span the increase in
displacement is negligible.

1 b/L=.0
b/L=.9
displacement

b/L=.7
Normalized

b/L=.4
0.5
b/L=.2

0
0 0.2 0.4 0.6 0.8 1
Dimensionless distance z/L

Figure 3.7: Shape of beams with varying midplane delamination lengths.

3.1.3 Dynamic solution: Bernoulli-Euler beam equations

The model proposed by Mujumdar and Suryanarayan uses Bernoulli-Euler beam

equations for every beam section. In this paragraph the equations of motion for
such a beam section are derived. Figure 3.81 shows a beam under uniform loading,
Chapter 3. Theoretical modelling 33

figure 3.91 displays the free-body diagram corresponding to a beam element of

length dz. The transverse displacement at point z and time t is denoted by x(z, t)
and the force per unit length is p(z, t). The parameters involved are mass per unit
length m(z) and the flexural rigidity EI(z), where I(z) is the second moment of
area about the y-axis and E(z) is Young’s modulus of elasticity. It is assumed that
the rotation of the beam is negligible compared to the vertical displacement. For
this to be true the ratio between length and height should be relatively large and the
beam should not become too ’wrinkled’, with other words only the lower modes of
vibration should be considered.

x p(z,t)
p(z,t) ∂V (z,t)
V (z,t)+
M (z,t) ∂z

dz
x(z,t) m(z),EI(z)
∂M (z,t)
M (z,t)+
z x(z,t) V (z,t) ∂z
z dz

Figure 3.8: Beam in flexure under uni- Figure 3.9: Free body diagram of a
form loading. beam element.

First, from figure 3.9, the force equation of motion in x-direction is derived:

∂V (z, t) ∂ 2 x(z, t)
[V (z, t) + dz] − V (z, t) + p(z, t)dz = m(z)dz (3.5)
∂z ∂t2

Since the rotation induced inertia torque is ignored the following equation holds
for the moment about the center of the left edge:

∂M (z, t)
[M (z, t) + dz] − M (z, t) + [V (z, t)+
∂z (3.6)
∂V (z, t) 1
]dz + p(z, t)dz dz = 0
∂z 2

Canceling appropriate terms and discarding terms involving second or higher pow-
ers in dz, equation 3.6 reduces to:

∂M (z, t)
− = V (z, t) (3.7)
∂z
1
Figures are taken from: L. Meirovitch. Elements of Vibration analysis, 1986.
34 Delamination detection based on dynamic behaviour

EI(z)∂ 2 x(z, t)
M (z, t) = (3.8)
∂z 2

Substituting equation 3.7 into 3.5 and using the bending moment equation 3.8 the
following differential equation for the flexural vibration of a beam is found:

EI(z)∂ 4 x(z, t) ∂ 2 x(z, t)

− + p(z, t) = m(z) (3.9)
∂z 4 ∂t2

Separation of variables x(z, t) = u(z)f (t) and letting p(z, t) = 0, give:

EI(z) 1 d4 u(z)
− = −ω 2
m(z) u(z) dz 4
1 d2 f (t)
= −ω 2 (3.10)
f (t) dt2

Rearranging the first of equation 3.10 the eigenvalue problem formulation reduces
to the following differential equation (3.11), with solution (3.12):

d4 u(z) ω 2 m(z) d4 u(z)

− u(z) = − β 4 u(z) = 0 (3.11)
dz 4 EI(z) dz 4
u(z) = A sin(βz) + B cos(βz) + C sinh(βz) + D cosh(βz) (3.12)

The solution depends on the constants of integration A, B, C and D, these can be

found by applying the four boundary conditions. For finding a solution to the eigen-
value problem, the value of the constants is not needed. The boundary conditions
can be written as a set of simultaneous linear homogeneous algebraic equations in
the unknown constants. A non-trivial solution is only found when the determinant
of the coefficient matrix vanishes. All the components of the matrix are functions
of the unknown fundamental frequency β, which can thus be solved.

3.1.4 Dynamic Solution: Governing equations

In section 3.1.1 it was explained that Mujumdar and Suryanarayan [21] modelled a
delaminated beam as four separate Bernoulli-Euler beams. Since the integral beam
regions, with lengths a and c in figure 3.1, have the same loading as the beam
segment explained above, the equation and solution derived there will be used.
However, the equation will be written in a dimensionless form,

d4 ui (z i )
− β 4 ui (z i ) = 0, i = 1, 4 (3.13)
dz 4i
Chapter 3. Theoretical modelling 35

M2 ,V2 ,N M2 ,V2 ,N

M3 ,V3 ,N p M3 ,V3 ,N

Figure 3.10: Free body diagrams of the layers above and below a delamination.

in which:
ω 2 mL4
β4 = (3.14)
EI

The dimensionless axial coordinate and transverse displacement are defined as

z i = zi /L and ui = ui /d. A full list of the used notation can be found in section
3.5. Since it is assumed that the delaminated layers remain in contact a pressure,
p(z, t), exists between the two layers, see figure 3.10 and 3.11. The equations of
motion for both segments can be written as:

EI2 ∂ 4 x2 (z2 , t) ∂ 2 x2 (z2 , t) m2 ∂ 2 x2 (z2 , t)

− N − +p=0
∂z24 ∂z22 ∂t2
EI3 ∂ 4 x3 (z3 , t) ∂ 2 x3 (z3 , t) m3 ∂ 2 x3 (z3 , t)
4 +N − −p=0 (3.15)
∂z3 ∂z32 ∂t2

p
∂V3
V3 + ∂z
∂M
M3 + ∂z3
M2 ,N
dz
N + ∂N
∂z

V2
x(z,t)

Figure 3.11: Free body diagram of a delaminated beam element.

As stated before the delaminated layers remain in contact with each other, this
means that the displacement in the x-direction of segment 2 equals the displace-
ment of segment 3, or u2 (z, t) = u3 (z, t). Furthermore it follows from the origins
shown in figure 3.1 that z2 equals z3 . Replacing u3 and z3 by u2 and z2 respectively
36 Delamination detection based on dynamic behaviour

in the last of equation 3.15 and adding this to the first of equation 3.15, the gov-
erning differential equation describing the dynamic behaviour of de delaminated
section is obtained:

E(I3 + I2 )∂ 4 x2 (z2 , t) (m2 + m3 )∂ 2 x2 (z2 , t)

− =0
∂z24 ∂t2

After separation of variables and writing the solution in a dimensionless form one
obtains:

d4 u2 (z 2 )
− β24 u2 (z 2 ) = 0 (3.16)
dz 42

in which:
ω 2 (m2 + m3 )L4i
β24 = (3.17)
E(I2 + I3 )

Introducing the dimensionless variables h2 = h2 /h and h3 = h3 /h, with h2 and

h3 the thickness of the upper and lower delaminated beam segments respectively
and h the total thickness, equation 3.17 can be rewritten using equation 3.14:

ω 2 (m2 + m3 )L4 β4
β24 = = 3 3 (3.18)
E(I2 + I3 ) h2 + h3

Recapitulating, a general differential equation is found for two integral sections

and one combined delaminated section,

d4 ui (z i )
− βi4 ui (z i ) = 0, i = 1, 2, 4 (3.19)
dz 4i

in which:

ω 2 mL4
β14 = β44 =
EI
ω 2 (m2 + m3 )L4 β4
β24 = = 3 3 (3.20)
E(I2 + I3 ) h2 + h3

The general solution, which describes the displacement for each section (i =
1, 2, 4) is found to be:

ui (z i ) = Ai sin βi z i + Bi cos βi z i + Ci sinh βi z i + Di cosh βi z i (3.21)

Chapter 3. Theoretical modelling 37

3.1.5 Dynamic solution: Boundary and continuity conditions

The solution presented in the last section contains 12 unknown coefficients, these
are determined by the boundary and continuity conditions. The beam, presented
in figure 3.1, has two boundaries (z1 = 0) and (z4 = 0); for every boundary two
conditions are obtained. In table 3.2 several ways of supporting a beam are listed
along with the resulting boundary conditions.

Support Boundary conditions

Clamped ui (0) = 0 u′i (0) = 0
Simply supported ui (0) = 0 u′′i (0) = 0
Free u′′i (0) = 0 u′′′
i (0) = 0

Table 3.2: Boundary conditions

A total of four boundary conditions is found; another eight conditions are needed
to solve the twelve unknown constants in equation 3.21. These follow from the
conditions of continuity at the two junctions (z 1 = a) and (z 4 = −c). Continuity
of displacement and slope:

u1 (a) = u2 (−b/2) u4 (−c) = u2 (b/2)

u′1 (a) = u′2 (−b/2) u′4 (−c) = u′2 (b/2) (3.22)

The remaining four conditions follow from the continuity in shear force and ben-
ding moment at the two junctions. Continuity of shear force between the integral
and delaminated section gives Vi = V2 +V3 , with i = 1, 4. In terms of displacement
this results in

d 3 ui d3 u2
EI1 = (EI 2 + EI 3 )
dzi3 dz23

writing this in a dimensionless form one obtains:

d3 ui 3
3
3 d u2
= (h 2 + h 3 ) , i = 1, 4 (3.23)
dz 3i dz 32

The equilibrium of bending moments about the midplane of section 1 or 4 can be

found to be Mi = M2 + M3 + N (h/2), with again i = 1, 4. This also can be
written in terms of displacements:

d2 ui d2 u2
EI1 2 = (EI2 + EI3 ) 2 + N (h/2)
dzi dz2
38 Delamination detection based on dynamic behaviour

In dimensionless form:

d2 ui 2
2
2 d u2
= (h 2 + h3 ) + N L2 /2EI1 , i = 1, 4 (3.24)
dz 2i dz 22

The force N on the left of equation 3.24 depends on the stretching and compressing
of the segments 2 and 3. This was also shown in the static solution. Force N is
given in a dimensionless form by:

EI du1 (a) du4 (−c)

N= (6h2 h3 /b)[− + ] (3.25)
L2 dz 1 dz 4

A derivation of this term can be found in [21]. The resulting equation for continuity
of bending moment is obtained, with i = 1, 4:

d2 ui 2
2
2 d u2 du1 (a) du4 (−c)
= (h 2 + h 3 ) − (3h2 h3 /b)[− + ] (3.26)
dz 2i dz 22 dz 1 dz 4

A total of 12 boundary and continuity conditions are formulated. These condi-

tions can be written as a matrix vector equation in twelve unknown constants. The
system possesses a non-trivial solution when the determinant of the matrix equals
zero. All components in the matrix are functions of the unknown fundamental
frequency β, which can thus be solved.

3.1.6 Dynamic solution: Analytical results

Analytical results for a cantilevered beam, obtained with the model described above,
are shown in figures 3.12 to 3.15. The figures show the influence of the delami-
nation parameters like delamination length b and both axial ac and thickness-wise

ac = .5 a = .5
c
β − Fundamental Frequency

1.9 1.9
β − Fundamental Frequency

1.8 1.85

1.7 1.8
h /h=.1 b/L=.1
2
b/L=.3
1.6 h2/h=.3 1.75
b/L=.5
h2/h=.5 b/L=.7
1.7
1.5 0 0.1 0.2 0.3 0.4 0.5
0 0.2 0.4 0.6 0.8 1
Delamination length b/L Del. location h2/h

Figure 3.12: Influence of delamination length b and thickness-wise location h2 on first

vibration frequency.
Chapter 3. Theoretical modelling 39

h /h=.33 h2/h=.5
2
1.9
β − Fundamental Frequency

1.9

1.85
1.85
b/L=.1 b/L=.1
1.8
b/L=.2 b/L=.2
1.8
b/L=.3 1.75 b/L=.3
b/L=.4 b/L=.4
1.75 b/L=.5 1.7 b/L=.5
b/L=.7 b/L=.7
1.7 1.65
0 0.2 0.4 0.6 0.8 1 0.2 0.4 0.6 0.8
Spanwise location ac Spanwise location a
c

Figure 3.13: Effect of axial position ac of a delamination on the first vibration frequency.

h2/h=.33 h2/h=.5
β − Fundamental Frequency

4.6 4.6

b/L=.1 4.4 b/L = .1

4.4 b/L=.2 b/L = .2
b/L=.3 4.2 b/L = .3
4.2 b/L=.4 b/L = .4
b/L=.5 4 b/L = .5
b/L=.7 b/L = .7
4 3.8
0.2 0.4 0.6 0.8 0.2 0.4 0.6 0.8
Spanwise location ac Spanwise location ac

Figure 3.14: Effect of axial position ac of a delamination on the second vibration fre-
quency.

h2/h=.33 h2/h=.5
β − Fundamental Frequency

7.5 7.5
b/L = .1 b/L=.1
b/L = .2 7 b/L=.2
7
b/L = .3 b/L=.3
b/L = .4 6.5 b/L=.4
6.5 b/L = .5 b/L=.5
b/L = .7 6 b/L=.7
6
0.2 0.4 0.6 0.8 0.2 0.4 0.6 0.8
Spanwise location a Spanwise location ac
c

Figure 3.15: Effect of axial position ac of a delamination on the third vibration frequency.

h2 location on the fundamental frequencies β of the delaminated beam. The fun-

damental frequency can be substituted in equation 3.14 to determine the resonance
frequencies of the delaminated beam.
Figure 3.12 shows that both the delamination length and thickness-wise location
have a big influence on the first fundamental frequency. In these cases the delam-
ination is located at half the total length of the beam. For an intact beam the first
40 Delamination detection based on dynamic behaviour

fundamental frequency equals 1.875. It can be seen that the solution converges to
this value in case h2 approaches zero and the delamination shifts to the surface of
the beam. Furthermore it is observed that the effect is maximal for a delamination
near the midplane (h2 = .5). This was also observed in the static analysis. How-
ever it has to be mentioned that only for relatively large delaminations (b > .3) the
effect on the first mode is noticeable. This can also be seen in figure 3.13 where
the influence of the axial location on the first fundamental frequency is shown. It
is shown that the delamination has the greatest influence when it is positioned near
the clamped end. However, a delamination with a length of b = .3 only causes a re-
duction of the first resonance frequency ω of 2.6 %, which is probably not enough
to detect it with confidence. Figure 3.14 shows that the second fundamental fre-
quency is influenced more by the delamination. The second fundamental frequency
for an intact beam equals 4.694. The greatest effect again is observed when the de-
lamination is positioned close to the clamped end. The peak near ac = .55 attracts
the attention and indicates that the effect of a delamination is negligible in case it
is located at .55 times the beams length. In figure 3.15, which shows the effect on
the third fundamental frequency, similar peaks are observed; in this case however
these are located at approximately ac = .32 and ac = .73. The third fundamental
frequency for an intact beam equals 7.855.
As mentioned in the static solution the effect of the delamination is a combination
of two effects: one is the reduction in flexural stiffness over the beams thickness,
the other is the stretching and compressing of the delaminated beam segments. The
first effect results in a reduction of flexural rigidity, while the second effect tends
to compensate this. It was pointed out in section 3.1.2 that the weakening effect of
a delamination is only noticed in case a shear force is applied. This means that the
effect of a delamination can be neglected if the delamination is located in regions
where the shear force approaches zero. The shear force can be expressed as a
function of the third derivative of the displacement:

∂ 3 u(z)
V = EI
∂z 3
The displacement u(z) of an intact cantilever beam at location z is given by:
u(z) = Ar [(sinβL − sinhβL)(sinβz − sinhβz) +
cosβL + coshβL)(cosβz − coshβz)] (3.27)

In figure 3.16 the displacement u(z) and third derivative ∂ 3 u(z)/∂z 3 for the second
mode of vibration of an intact beam is plotted. Figure 3.17 plots the result for the
third vibration mode. Both the displacement and third derivative are normalized to
the maximum value. As can be seen from both figures, ∂u(z)3 /∂z 3 equals zero
where the influence of the delamination vanishes. Also the delamination has the
biggest influence where ∂u(z)3 /∂z 3 maximizes, which is in fact near the clamped
end.
Chapter 3. Theoretical modelling 41

1 1 3 3
u(z) u(z) ∂ u(z)/∂z
3 3
0.5 ∂ u(z)/∂z 0.5

0 0

−0.5 −0.5

−1 −1
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
Distance z/L Distance z/L

Figure 3.16: Mode shape and third Figure 3.17: Mode shape and third
derivative for second vibration mode. derivative for third vibration mode.

This influence on spanwise location also shows up in the vibration mode shapes.
Figure 3.18 shows the mode shapes for the second vibration mode of an intact and
delaminated beam. It can be seen that the effect of the delamination is minimal
in case it is located at ac = .55. For ac = .3 a clear difference between the
delaminated and intact case is observed. Figure 3.19 shows similar results for the
third vibration mode. The mode shapes of an intact or delaminated beam do not
differ much in case the delamination is located near regions of minimum shear
stress. If its located near regions of high shear stress a clear difference is observed.
Overall it can be concluded from figure 3.12 to 3.15 that the influence of a de-
lamination increases with mode number. The third vibration mode is clearly more
affected than the first. It is also observed that the influence increases in case the
delamination shifts to the midplane (h2 = .5) and vanishes at the surface (h2 = 0).
Also the spanwise location has a great influence. An arbitrary located delamina-
tion affects each mode differently depending on the location. This is caused by
the nonuniform distribution of shear stress. This effect was used by Cawley and
Adams [3] to locate delaminations.

ac = .3, b/L = .3 ac=.55, b/L=.3

1 1
displacement

displacement
Normalized

Normalized

0 0
Delaminated Delaminated
Intact Intact
−1 −1
0 0.5 1 0 0.5 1
Distance z/L Distance z/L

Figure 3.18: Mode shapes for second mode of vibration for delaminated and intact beam.
42 Delamination detection based on dynamic behaviour

a =.3, b/L=.3 a =.5, b=.3

c c
1 1

displacement
displacement

Normalized
Normalized

0 0

−1 Delaminated Intact −1 Delaminated Intact

0 0.5 1 0 0.5 1
Distance z/L Distance z/L

Figure 3.19: Mode shapes for third mode of vibration for delaminated and intact beam.

3.2 Application of the model for composite beams

The model by Mujumdar and Suryanarayan [21] can easily be applied to composite
beams. Unlike isotropic materials, the Young’s modulus of composites varies along
the thickness. This is caused by changes in fibre orientation over the thickness. Mu-
jumdar and Suryanarayan simplified their model for application with homogeneous
materials. This simplification however, can easily be undone. To do so the ratio
(EI2 + EI3 )/EI1 in the continuity equations for shear force, bending moment and
normal force cannot be rewritten in terms of dimensionless beam thickness, as in
equations 3.23, 3.24 and 3.25. This is because this ratio is now also a function of
laminate lay-up, and not only of the thickness anymore. Also the moment caused
by the stretching and compressing of the layers depends now on lay-up. In the fol-
lowing section it is shown how this laminate lay-up influences both the static and
dynamic behaviour of a delaminated composite beam.

3.2.1 Determination of flexural rigidity

The flexural rigidity of the segments can be determined by calculating the internal
bending moment and combining this with the bending moment equation. Before
calculating the flexural rigidity, the location of the neutral surface has to be deter-
mined. This is done by subjecting the beam to a normal force F which causes the
beam to compress, as shown in figure 3.20. In case the force is applied at the neu-
tral surface the separate layers will all deform equally with ǫl . To determine this
location the sum of moments is taken about the bottom-left point of the laminate.
The force generated by each layer can be derived from the strain of the laminate:

σl F
ǫl = =
El EAl
Chapter 3. Theoretical modelling 43

ǫl
Ek+1 ρ
dθ
M x0
Ek
zk−1 x1
xk
Ek−1 1
6t 2t
F E3
2
z2 3 M
−2t
E2 4
δi z1 x
E1 z
dz

Figure 3.20: Determination of neu- Figure 3.21: Composite beam under

tral surface. bending moment.

In which σl represent the stress in the laminate and El the Youngs modulus of the
laminate. EAl can be determined as follows:
N
X
EAl = b Ek (xk−1 − xk ) (3.28)
k=1

With Ek representing the Youngs modulus of a layer and xk the distance of the
layer edge to the delaminated surface. The summation of the moment exerted by
each layer should equal the moment caused by F multiplied by the distance δi from
the delaminated surface.
N N N
X X X F
F δi = dk σk bt = dk ǫl Ek bt = dk Ek bt
EAl
k=1 k=1 k=1

In which dk = zk−1 + (1/2)(zk − zk−1 ) represents the distance from delaminated

surface to the midplane of the layer. This results in:
N
X dk Ek bt
δi = (3.29)
EAl
k=1

To calculate the flexural rigidity figure 3.21 is considered. A laminated beam is

subjected to a bending moment M which results in a radius of curvature of ρ. The
normal stress at a distance x from the neutral line in a beam in flexure is calculated
from the normal strain
x
σz (x) = Ek
ρ
in which ρ is the radius of curvature of the laminate and Ek the local Young’s
modulus. The resulting bending moment can be determined by:
Z 2t
M= σz (x)bxdx
−2t
44 Delamination detection based on dynamic behaviour

With xk and xk−1 the distances from the neutral surface the following is obtained:
N Z zk N
X −bx2 −b X
M= Ek dx = Ek (x3k−1 − x3k ) (3.30)
zk−1 ρ 3ρ
k=1 k=1
The bending moment equation can be used to express the bending moment as a
function of the elasticity modulus of the laminate Ezl :
El Iy u′′ = M → El Iy = M ρ (3.31)
Combining the last two equations gives the flexural rigidity of the laminate as a
whole:
N
−b X
EIl = Ek (x3k−1 − x3k ) (3.32)
3
k=1

This relation can be used to determine

the bending stiffness of each beam seg- 1 Isotropic
ment. The distances xk and xk−1 (EI2+EI3)/EI1 material
then relate to the neutral line of eve- [903/0/904]s
ry separate beam segment. As an 0.5
example the ratio (EI2 + EI3 )/EI1
of a [903 /0/904 ]s laminate, with ma-
terial properties as in table 3.3, is plot- 0
0 2 4 6 8
ted against delamination location. For o
Layer N
comparison the ratio for an isotropic
beam is also plotted. From figure 3.22
Figure 3.22: Influence of delamination lo-
it is seen that laminate lay-up plays an cation on flexural rigidity.
important role.
The ratio (EI2 + EI3 )/EI1 shows a great jump in case the delamination is located
after the fourth layer. This is caused by the fact that the stiff fourth layer now be-
longs to the upper laminate. The effective distance from the neutral line is reduced
which results in a reduction of the total flexural rigidity.

carbon-PEI E0 E90 G12 ν12 ν23 ρ

(GP a) (GP a) (GP a) (kg/m3 )
120 8.1 3.5 0.32 0.45 1580

Table 3.3: Mechanical properties of carbon-polyetherimide layer.

3.2.2 Static solution

The flexural rigidity of the delaminated segments derived above can be substituted
in the static solution presented in section 3.1.2. Besides this, also the location
Chapter 3. Theoretical modelling 45

where force Ni is applied has changed. Force Ni causes either a shortening or an

elongation of the delaminated layers. In order to prevent nonuniform strain this
force should be applied at the neutral surface. This distance was also calculated in
the paragraph above.
Figure 3.23 shows the free body diagram of the endplane. Equilibrium of force in
both x- and z-direction and bending moment about the junction between delami-
nation and end plane give now:

N2 + N3 = 0 M2 ,V2 ,N2

V2 + V3 = 0 Neutral surf. δ2
M2 + M3 + N2 (δ2 + δ3 ) = 0
Del surf. δ3
A similar change is observed in the M3 ,V3 ,N3
continuity conditions:
Neutral surf.
ϕ2 (L) = ϕ3 (L)
u2 (L) = u3 (L)
w3 (L) = w2 (L) + (δ2 + δ3 )ϕ2 (L) Figure 3.23: Free body diagram of
end plane.

As was observed in section 3.1.2, also in case of composites only the displacement
as result of a shear force is affected. Figure 3.24 shows the displacement at the
free end in case a shear force is applied as function of delamination location. The
graph shows a considerable increase of deflection after the delamination is located
under the fourth layer. The stiff fourth layer becomes part of the upper segment, as
a result its distance to the neutral surface is reduced which decreases the flexural
rigidity of the delaminated segment.
In figure 3.25 the effect of delamination location on the displacement is shown for

[90 /0/90 ] [90/0/90/0/90/0/90 ]

3 4 s 2 s
3
2.5
displacement

displacement
Normalized

Normalized

2.5
2
2
1.5 1.5
1 1
0 1 2 3 4 5 6 7 8 0 1 2 3 4 5 6 7 8
Delamination after layer No Delamination after layer N o

Figure 3.24: Effect of delamination loca- Figure 3.25: Effect of delamination loca-
tion on displacement. tion on displacement.
46 Delamination detection based on dynamic behaviour

a [90/0/90/0/90/0/902 ]s laminate. Every time a stiff zero degree layer is shifted

to the upper segment a jump in displacement is observed. From the figures 3.22,
3.24 and 3.25 it can be concluded that in case of laminates the decrease in flexural
rigidity is heavily influenced by the delamination location and laminate lay-up.

3.2.3 Dynamic solution

As stated above the flexural rigidity of the delaminated section depends on laminate
lay-up. The general solution for the delaminated segment was found to be:

u2 (z i ) = A2 sin β2 z 2 + B2 cos β2 z 2 + C2 sinh β2 z 2 + D2 cosh β2 z 2 (3.33)

In this equation β2 cannot be rewritten in the dimensionless transversal delamina-

tion location h2 anymore. For β2 the following relation is found:

ω 2 (m2 + m3 )L4 β 4 EI1

β24 = =
EI2 + EI3 EI2 + EI3

Furthermore the continuity conditions for shear force, bending moment and nor-
mal force, equations 3.23, 3.24 and 3.25, have to be rewritten. These cannot be
simplified to the dimensionless transversal delamination location anymore. The
continuity equations for shear force and bending moment become:

∂ 3 ui EI2 + EI3 ∂ 3 u2
µ ¶
= , i = 1, 4 (3.34)
∂z 3i EI1 ∂z 32
∂ 2 ui EI2 + EI3 ∂ 2 u2 P L2 (δ2 + δ3 )
µ ¶
= + , i = 1, 4 (3.35)
∂z 2i EI1 ∂z 22 dEI1

The force P can be determined from the extend of stretching of the delaminated
layers, as shown in static analysis. From [21] it follows that P equals:

1 −1
· ¸ · ¸
d(δ2 + δ3 ) 1 ∂u1 (a) ∂u4 (−c)
P = + − + (3.36)
bL EA2 EA3 ∂z 1 ∂z 4

In which EAi is determined above and given in equation 3.28. The twelve condi-
tions are again written as a matrix vector equation. The fundamental frequency β
is found by the requirement for the determinant of the matrix to equal zero.
For a [903 /0/904 ]s laminate the influence of delamination location on the first and
second fundamental frequency is respectively shown in figure 3.26 and 3.27. The
delamination length equals 20% of the total beam length. The delamination has
almost no effect if it’s located anywhere between the first or fourth layer. However
in case it’s located after layer number four a considerable influence in noticed.
These results correspond well with the static solution presented above. There also
Chapter 3. Theoretical modelling 47

[90 /0/90 ] (b/L=.2) [90 /0/90 ] (b/L=.2)

3 4 s 3 4 s
4.7

1.875
β − Fundamental Frequency

β − Fundamental Frequency
4.6

1.87
4.5

1.865 Layer 1 4.4 Layer 1

Layer 3
Layer 3
Layer 4 Layer 4
1.86 4.3
Layer 6 Layer 6
Layer 8 Layer 8
1.855 4.2
0 0.5 1 0 0.5 1
Spanwise location ac Spanwise location ac

Figure 3.26: Effect of delamination loca- Figure 3.27: Effect of delamination loca-
tion on first resonance frequency. tion on second resonance frequency.

[903/0/904]s [903/0/904]s
1.9 4.8
β − Fundamental Frequency
β − Fundamental Frequency

1.85 4.6

1.8 4.4

d=3,b=.2 d=3,b=.2
1.75 d=3,b=.5 4.2 d=3,b=.5
d=3,b=.7 d=3,b=.7
1.7 d=4,b=.2 4 d=4,b=.2
d=4,b=.5 d=4,b=.5
d=4,b=.7 d=4,b=.7
1.65 3.8
0 0.5 1 0 0.5 1
Spanwise location ac Spanwise location ac

Figure 3.28: Effect of delamination loca- Figure 3.29: Effect of delamination loca-
tion on first resonance frequency. tion on second resonance frequency.

a sudden decrease in flexural rigidity was observed in case the delamination was
located after the fourth layer. In figure 3.28 and 3.29 the effects of delamination
length for different transversal and longitudinal locations on the first and second
fundamental frequency is shown. Again the effect of delamination location on
flexural rigidity is underlined. The parameter d indicates after which layer the
delamination is located.
48 Delamination detection based on dynamic behaviour

3.3 Finite element analysis

Two finite element models, representing a delaminated beam, are described in this
section. Finite element models provide an easy way of investigating the effect of
delaminations in more complex structures. In this section two models are proposed.
In the first model the delaminated beam is modelled with two node beam elements.
As in the analytical model this model consists of four separate beam segments.
The second model uses contact elements in combination with four node 2D brick
elements. The contact elements in this last model could be used in the future for
modelling more complex delaminated structures.

3.3.1 Beam elements with node coupling

This model is based on 2D two node beam elements. As stated above four segments
can be distinguished in this model, this is illustrated in figure 3.30. The transversal
displacement of axially corresponding nodes in the delaminated layers are coupled:

u a n = u bn (3.37)

This effectively means that the delaminated layers stay in contact during the vi-
bration of the beam. However because the axial displacement and rotation are not
coupled the layers are free to slide with respect to each other. At the junction be-
tween integral and delaminated segments all degrees of freedom are coupled. For
the transversal displacement and rotation of the nodes at the junction the following
holds:

φi = φj = φk
ui = uj = uk

For the axial displacement of nodes i, j, k in figure 3.30 one obtains:

wi = wj − φj δ2
wk = wj + φj δ3

an−1 an an+1
i
bn−1 bn bn+1
j
k

x
z

Figure 3.30: Representation of a delaminated beam with 2-node beam elements.

Chapter 3. Theoretical modelling 49

In which δi represents the distance of the neutral plane of segment i to the neutral
plane of the first or last beam segment. The beam is modelled in ANSYS using
BEAM3 elements to represent the separate segments. The elements are created on
the neutral axis of the beam segments.

3.3.2 Contact elements in combination with brick elements

A two dimensional finite element model was created in ANSYS. The beam was
modelled using PLANE42 elements. This brick element is mainly used for 2D
modelling of solid structures. The element is defined by four nodes, all having
two degrees of freedom: translations in nodal x and y. A plane strain situation
was assumed. The delaminated surface was created using contact elements. For
this purpose two elements were needed, namely a contact element CONTA171 and
a target element TARGE169. It was assumed that the delaminated layers stay in
contact and only slide, without friction, in relation to each other. Figure 3.31 shows
the finite element representation of a delaminated section of the beam.
PLANE42 element

Contact elements

Figure 3.31: Representation of a section of a delaminated beam with contact elements and
2D-brick plane elements.

A common problem encountered using four node plane elements in bending ana-
lysis, is that the flexural rigidity can be overestimated. This effect is minimized by
taking more elements over the beam thickness. A study showed that the extend of
this overestimation, compared with results obtained by using eight node 2D brick
elements, is found to be less than .3% for resonance frequencies near 1200Hz.

3.3.3 Results

A modal analysis of both models was carried out to extract the resonance frequen-
cies of the first five modes of vibrations for an isotropic cantilevered beam. The
material properties used are listed in table 3.3. The beams length and width were
respectively 400mm and 20mm, the thickness was 2.0mm. In table 3.4 the reso-
nance frequencies obtained by the analytical analysis are compared with the results
obtained by these finite element analysis. It can be seen that the results from the
finite element models correspond well with the analytical case. Both the coupling
of nodes and the use of contact elements are very suitable for representing sliding
50 Delamination detection based on dynamic behaviour

delaminations. In more complex structures, however, it might be hard to repre-

sent a delamination by coupling nodes. Contact elements provide, as can be seen
in table 3.4, an excellent alternative. Also contact elements can be used to intro-
duce friction between the layers or allow the delamination to open and close during
vibration.

D ELAMINATION PARAMETERS
ac = .4, b = .6, h2 = .25 ac = .5, b = .2, h2 = .5
Mode Analytical BEAM3 PLANE42 Analytical BEAM3 PLANE42
1 16.52 16.52 16.51 17.56 17.56 17.56
2 94.25 94.25 94.06 110.46 110.46 110.40
3 241.53 241.53 241.07 279.41 279.39 278.92
4 484.34 484.33 483.89 596.70 596.63 595.68
5 801.05 801.12 800.93 821.93 821.79 821.17

Table 3.4: Comparison between analytical and FEM results.

3.4 Summary and concluding remarks

The influence of a delamination on the resonance frequencies of a cantilever beam

was studied in this chapter. For this purpose an analytical model, introduced by
Mujumdar and Suryanarayan [21], was investigated and modified for use with com-
posite structures. Furthermore two finite element models were presented which
both could be used to investigate the dynamic behaviour of delaminated structures.
The results of the two finite element models compared extremely well with the
analytical analysis.

Regarding the effects of a single ’through width’ delamination on an isotropic,

homogeneous beam the follwing general conclusions can be drawn:

• A static analysis, subjecting a delaminated beam to a shear force, showed

that the influence of a delamination on bending shape or displacement of
free end can be neglected. Only for large delaminations, over 40% of the
beams length, a detectable change in displacement is observed.

• The axial, in contrast with transversal, location of the delamination has no

influence on static bending behaviour. Transversally, however, the effect of
a delamination maximizes in case it is located on the neutral plane of the
beams thickness.

• The influence of a delamination on resonance frequencies is significant and

increases with mode number. In all cases resonance frequencies shift to
lower values as result of a delamination.
Chapter 3. Theoretical modelling 51

• As in the static case, resonance frequencies are influenced most in case the
delamination is located at half the thickness.
• Each mode is affected differently by the axial location of the delamina-
tion. This fact results from the fact that the shear force is distributed non-
uniformly accross the beams length an can be used to locate the delamina-
tion.
An investigation of the influence of a delamination on the resonance frequencies
of composite beams showed that, laminate lay-up plays an important role.
From this theoretical analysis it can be concluded that shifts in resonance frequen-
cies can very well be used as a damage indicator. Monitoring resonance frequen-
cies provides a more accurate way for detecting delaminations than a static analy-
sis.
52 Delamination detection based on dynamic behaviour

3.5 Nomenclature

Ai ,Bi ,Ci ,Di constants in mode shape expression

a length of first integral segment
b delamination length
β fundamental frequency
c length of third integral segment
a,b,c dimensionless delamination parameters, a/L, b/L, c/L
ac dimensionless spanwise delamination location
h thickness of beam
hi thickness of segment i
h, hi dimensionless thickness h/L, dimensionless transversal delami-
nation location hi /h
EIi flexural rigidity of segment i
i subscript representing each segment
L beam length
Mi bending moment in each segment
m mass per unit length
Ni normal force in each segment
p contact pressure between each segment
t time
ui transversal displacement
ui dimensionless transversal displacement ui /d
Vi shear force in each segment
wi axial displacement
wi dimensionless axial displacement wi /d
zi dimensionless axial co-ordinate zi /L
ω radian frequency of vibration
 CHAPTER

FOUR
EXPERIMENTAL VALIDATION

I n the following section the validation programme will be described. Both static
and dynamic tests have been conducted. Two goals will be pursued in this pro-
gramme. The first consists of validating the analytical and finite element models
provided in chapter 3. The second goal is to investigate the potential of fibre bragg
gratings for dynamic health monitoring based on dynamic behaviour.

4.1 Programme outline

The validation programme consists of three stages, of which a detailed outline is

given below. The programme is aimed at validating the statements found in the
theoretical analysis. A modal analysis is performed on cantilevered beam shaped
specimens with varying delamination parameters. The experimental set-up is dis-
cussed in section 4.2. Delaminated specimens are easily manufactured by inserting
a thin sheet of kapton foil between two pre-impregnated carbon-polyetherimide
plies prior to the consolidation of the laminate. The foil prevents the two plies
to melt together and thus a delamination is formed. The size and position of the
delamination is highly controllable which is an advantage over other artificially
induced damage types.
Stage A: Validation of the experimental set-up and sensor evaluation
In order to validate the conclusions drawn in the theoretical analysis a reliable test-
ing approach should be adopted. The main goal of this stage therefore is to check
the testing approach in terms of reproducibility and to study the influence of extern
parameters like clampforce and tension in the spring, supporting the shaker. The
set-up is shown in figure 4.1. Various literature state that it is difficult to achieve
exact cantilever vibration in practice. It is checked to what extend this might cause
problems during this experimental programme. The secondary goal is to evaluate
different types of sensors to capture the response of the beam specimens. In this

53
54 Delamination detection based on dynamic behaviour

stage an accelerometer, strain gauge and a microphone are evaluated. Furthermore

results obtained from the models derived in section 3.1 are validated. To achieve
these goals, nine beam-shaped specimens are fabricated. Three of these specimens
are labeled as intact and contain no delamination. The other six are divided in
two groups, of which three specimens have a delamination at 50% of the beam
thickness and the remaining three have a delamination at 25% of the thickness.
Stage B: Application of fibre bragg gratings for health monitoring
Despite the encountered difficulties, stage A proved that strain measurements could
be used to determine the dynamic properties of a structure. This stage will inves-
tigate the application of fibre bragg gratings for strain measurements in composite
structures. These measurements are used to derive the dynamic properties of the
specimens. The main problem concerning strain measurements lied in the fact that
for higher vibration modes the strains became too small to measure. The laminate
lay-up for the different specimens will be tailored in order to maximize the strain
for higher modes. Static tests will be conducted to investigate the resolution and
accuracy of the fibre bragg gratings. The data acquisition system is tested in order
to evaluate the sampling rate.
Stage C: Validation of theoretical results
In this stage the analytical and finite element models will be validated. The reso-
nance frequencies of beam specimens containing varying delamination parameters
are obtained. Stage A demonstrated that the experimental set-up, as described in
section 4.2, did not satisfy the requirements. Test specimens are therefore excited
by hammer-impact. The delaminations have a length of 30% of the beam span at
most. Also the effect of a growing delamination is closely monitored. For this pur-
pose a specimen containing a small delamination is used. Between several modal
analyses the delamination is extended by using a sharp razorblade.

4.2 Experimental set-up

In figure 4.1 the experimental set-up is outlined. Beam shaped specimens are can-
tilevered and excited using a shaker. In order to reduce the influence of environ-
mental vibrations the shaker is hung from a spring connected to the fixed world.
The force excited by the shaker on the specimen is recorded with a force transducer.
The response of the beam is measured using accelerometers, strain gauges and fi-
bre bragg gratings. Also the sound radiated from the vibrating beam is recorded by
microphone. The signals are amplified before these are fed in to a data analyser,
see appendix A. With help of a computer the gathered time responses are analysed
and transformed by means of a fast fourier transformation (FFT). In appendix A
the used hardware is also listed.
Chapter 4. Experimental validation 55

Spring
Shaker Radiated sound
Force sensor

Clamped end

Strain gauges
Accelerometer

Figure 4.1: Experimental set-up (shaker) for modal analysis.

4.3 Stage A: Validation of the experimental set-up

and sensor evaluation

The main goal of stage A was to validate the test approach and the experimental
set-up in terms of reproducibility and robustness. Also the use of accelerometers,
strain gauges and a microphone were evaluated for capturing the responses of the
beam. At last the effects of delaminations on the frequency response of beams were
obtained and compared to the theoretical analysis. The following testing scheme
was adopted to pursue these goals.
1. Validation of the testing approach
• Effect of clamping conditions The effect of clamping conditions on
the frequency response plot was studied. The conditions were system-
atically varied.
• Influence of spring force The effect of the tension in the spring was
investigated. The spring force was varied by means of changing the de-
flection of the spring in rest. Three cases were examined, the deflection
was varied between 0, 7 and 15mm.
• Reproducibility The reproducibility was studied by performing sepa-
rate tests with constant conditions at distinct times. The specimen was
removed after each test. The resulting variations gave an indication of
the reproducibility of the testing method.
All these tests were performed using an intact specimen. The shaker excited
the beams with a chirp signal from 0 to 2kHz. The response of the beam was
captured with an accelerometer attached to the free end of the specimen.
2. Evaluation of used sensors The response of the beam specimens was ob-
tained by means of different sensors. In this stage measurements were ob-
tained using an accelerometer, strain gauges and a microphone. The data
obtained by the different sensors was compared and conclusions were drawn
56 Delamination detection based on dynamic behaviour

on their applicability for dynamic health monitoring. The information gath-

ered by the strain gauges was used to gain insight in the application of fibre
bragg gratings in later research. The tests were performed using the same
specimen as used above.
3. Evaluation of the effect of delaminations Two series of three specimens
contained a through width delamination at 50% and 25% of the thickness.
The effect of these delaminations on the resonance frequencies was inves-
tigated. The theoretical study indicates a shift of resonance frequencies to
lower values. The obtained resonance frequencies were compared to theo-
retical values.

Clamped end 1
4 2
a. b. c. 3

Cut-out

Figure 4.2: Specimen cut-outs. Figure 4.3: Clamping construction.

4.3.1 Specimen preparation

A unidirectional plate was fabricated from 12 layers of prepreg C-PEI. The fabri-
cation process is documented in appendix B. The resulting plate had a thickness
of approximately 1.6mm and a fibre mass fraction of 40% [36]. The material pro-
perties are listed in table 4.1. By using 16 µm thick kapton foil two delaminations
at different thickness-wise locations were created. The delaminations, indicated in
figure 4.2 by a and c, were located on respectively 25% and 50% of the thickness
and measured 60% of the beams total length. The region indicated by b contained
no delamination. The large delaminations were thought to have a great influence
on the beams dynamic behaviour and the effect will be easy to note. Three beams,

carbon-PEI E1 (GP a) E2 (GP a) G12 (GP a) ν12 ν23 ρ (kg/m3 )

120 8.1 3.5 0.32 0.45 1580

Table 4.1: Mechanical properties of carbon-polyetherimide layer.

Chapter 4. Experimental validation 57

with a length of 336mm and a width of 20mm, were cut from every section. The
axial direction of the beam shaped cut-outs corresponded with the fibre direction
of the laminated plate. Table 4.2 and appendix C list the specimen dimensions and
delamination parameters. An explanation of the delamination parameters can also
be found in appendix C.

Specimen Specimen size Del. parameters

ID L (mm) w (mm) h (mm) ac b h2
U1.00.000 320 20.1 1.67 na. na. na.
U1.06.025 320 20.1 1.66 .5 .6 .25
U1.06.050 320 19.9 1.67 .5 .6 .5

Table 4.2: Specimen dimensions and configuration stage A.

4.3.2 Evaluation of the experimental results

Validation of the experimental set-up

The experimental set-up was validated on two distinct points: the reproducibility
of the experiments and the influence of extern parameters like clamping conditions
or spring tension.
The influence of the clamp, see figure 4.3, was studied by performing several test
under varying clamping conditions. In figure 4.4 four obtained frequency response
functions are plotted. The conditions are listed in the table 4.3. The results rep-
resent worst case scenarios. Tests with other variations in clamping conditions
showed less or no deviation at all in frequency response plots compared to the
fixed case.

Condition Description
Fixed Fully bolted
Condition 1 Six layers of scotch tape added between the two clamp parts
Condition 2 Left side (bolt 1 and 4)β of the clamp bolted
Condition 3 Back side (bolt 1 and 2)β of the clamp bolted
β See figure 4.3

Table 4.3: Clamping conditions.

The results indicate that only in the case of condition 4 the frequency response
plot shows a large deviation compared to the fixed clamp. A significant shift in
resonance frequencies is observed. However, since the clamp was only bolted at
the back side, this case hardly represents cantilever vibration. For the remaining
conditions the resonance frequencies do not vary significantly. Based on the results
obtained it may be concluded that a small variation in clamping conditions has a
58 Delamination detection based on dynamic behaviour

Frequency Response Plot

100
Fixed
Condition 1
80 Condition 2
Condition 3

60
Magnitude [dB]

40

20

−20
0 500 1000 1500
Frequency [Hz]

Figure 4.4: Frequency response plots of intact specimen U1.00.000 with varying clamping
conditions.

Frequency Response Plot

100
Deflection 7mm
Deflection 15mm
80 No deflection

60
Magnitude [dB]

40

20

−20
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Frequency [Hz]

Figure 4.5: Frequency response plots of intact specimen U1.00.000 with varying spring
tension.

negligible influence on the obtained frequency response functions. The clamp can
therefore safely be used in this experimental programme and does not need any
refinement.

The influence of the tension in the spring on the obtained result was also investi-
gated. Similar conditions concerning excitation and response capturing were used
as for the tests above. The tension was varied by changing the spring deflection
Chapter 4. Experimental validation 59

Frequency Response Plot

100

80

60
Magnitude [dB]

40

20

−20
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Frequency [Hz]

Figure 4.6: Frequency response plots of specimen U1.00.000 at similar conditions, but at
distinct times.

in rest between 0, 7 and 15mm. Figure 4.5 shows the frequency response plots for
three different cases. As can be seen from the figure this has no influence on the
obtained frequency response plots.

The reproducibility of the experimental set-up was tested by performing three tests
under constant conditions at distinct times. An intact specimen was used. The
beam was after each test removed from the clamp. Figure 4.6 shows three measured
frequency response plots for an intact specimen, at different times. As can be
seen the different curves compare very well. The highest standard deviation over
the three tests was found to be .35% for the last peak. The averaged standard
deviation over all peaks equalled 0.2%. Similar results were obtained for the other
beam specimens. It is therefore concluded that the experimental set-up and method
satisfies the demands in terms of reproducibility. In literature the small statistical
variation of resonance frequencies was already mentioned [28]. The small standard
deviation allows the detection of very small shifts of resonance frequency and thus
the detection of small delaminations.

Recapitulating, the following conclusions are drawn:

• The influence of external parameters can be neglected. However, the clamped

end should be used with care. Small deviations in clamping conditions have
no influence on the obtained results, while erroneous usage can influence the
results heavily.

• The experiments are easy to reproduce and the obtained results show negli-
gible differences.
60 Delamination detection based on dynamic behaviour

• The experimental set-up and method suffices the conditions in terms of re-
producibility and influence of external parameters and can therefore be used
in this programme.

Evaluation of different sensors

The second part of stage A evaluated different sensors for capturing the beams re-
sponse. Besides the accelerometers used in the first part of this stage, the response
of the beam was also captured by strain gauges and a microphone. The strain
gauges were fixed near the clamped end of the beam in order to capture all flexural
vibration modes. Radiated sound from the vibrating beam was captured by a mi-
crophone. The obtained time signals were fed into a data analyser and transformed
by means of FFT.

Figures 4.7 and 4.8 show respectively the frequency transfer plots obtained by an
accelerometer and strain gauges. The upper graph in both figures shows the cohe-
rence of the measurement. The coherence is a guide to the quality of the transfer
function measurement. The coherence measures in what extend the power of the
response of the beam is caused by the driving power of the excitator. For a good
measurement the coherence equals 1 over the whole frequency spectrum. As can
be seen from figure 4.7 the quality of the measurement obtained by the accelerom-
eter is very good. For the measurements obtained by the strain gauges, figure 4.8,
however the coherence is poor. Only at resonance frequencies the coherence tends
to equal 1. The beam vibrates heavily in its resonance frequency, allowing the

Frequency Response Plot

1
Coherence

0.5

0
0 200 400 600 800 1000 1200 1400 1600 1800 2000

100
Magnitude [dB]

50

−50
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Frequency [Hz]

Figure 4.7: Frequency response plots of specimen U1.00.000 obtained by an accelerome-

ter attached to the free end of the beam specimen.
Chapter 4. Experimental validation 61

Frequency Response Plot

1
Coherence

0.5

0
0 200 400 600 800 1000 1200 1400 1600 1800 2000

40
Magnitude [dB]

20
0
−20
−40
−60
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Frequcncy [Hz]

Figure 4.8: Frequency response plots of specimen U1.00.000 obtained by strain gauges
glued near the clamped end of the beam specimen.

strain sensor to distinguish the induced strain from the noise level. Outside these
resonance frequencies the strain becomes very small, resulting in a poor signal to
noise ratio. For the higher modes with frequencies over 2kHz, the strain gauges
were unable to provide any signal at all. It is also noted that the resonance frequen-
cies obtained by the accelerometer are slightly lower than the ones obtained by the
strain gauges. This can be explained by the fact that the accelerometer adds mass
to the system.

Also the use of a microphone for capturing the response of vibrating structures
was investigated. Figure 4.9 shows the frequency response plot, up to 10kHz, of an
intact specimen. In the upper graph the signal was captured by the accelerometer.
The lower graph gives the frequency response plot as obtained by the microphone.
The frequency response function obtained by the microphone shows more peaks
and also a more erratic profile. The additional peaks can be caused by the fact that
a microphone is also able to capture torsional vibration modes, while the accelero-
meter is located in the middle of the beams width to prevent this. Furthermore the
surrounding noise can be the reason for the erratic profile shown in the graph. To
prevent the influence of surrounding noise, measurements with a microflown can
be considered. Nevertheless the general profile of the two measurements compare
reasonably.

Recapitulating, the following conclusions are drawn:

• Accelerometers provide good results for obtaining the response of the can-
tilevered specimens. The measurements show good coherence and resonance
62 Delamination detection based on dynamic behaviour

FRF − accelerometer
50

Magnitude [dB] 0

−50
0 2000 4000 6000 8000 10000
FRF − microphone
50
Magnitude [dB]

−50
0 2000 4000 6000 8000 10000
Frequency [Hz]

Figure 4.9: Frequency response plots of intact specimen U1.00.000 obtained by an ac-
celerometer and by capturing the radiated pressure waves with a microphone.

peaks are easy to identify.

• The strain gauges showed bad coherence outside the resonance frequencies.
This can be explained by the poor signal to noise ratio. The occurring strain
becomes too small to distinguish it from the noise level. This problem plays
a bigger role for higher vibration modes, as the strain will become smaller.
Tailoring the specimens in terms of laminate lay-up might proof useful in an
attempt to increase the measured strains.

• Obtaining the frequency response function by means of a microphone looks

quite good. At higher frequencies the obtained coherence was still good.
The influence of surrounding noise however can limit the applicability of a
microphone for capturing responses.

The effect of a single ’through width’ delamination

Three types of specimens were cut-out from the fabricated unidirectional plate.
Two types contained a delamination; one type at 50% of the thickness and one
type at 25% of the thickness. The response of these specimens was measured.

Figure 4.10 shows the frequency response plots of the three specimens. It can
clearly be seen that the resonance frequencies of the delaminated beams are much
lower than for the intact beam. This qualitatively corresponds well with literature
[21, 8, 20, 26, 32, 34, 35]. It can be observed that the reduction in resonance fre-
quencies is the greatest for the specimen with a delamination at 50% of the thick-
ness. This was also determined using the model by Mujumdar and Suryanarayan
[21], described in chapter 3. The closer the delamination plane lies to the neutral
Chapter 4. Experimental validation 63

Frequency Response Plot

100
No delamination
.25 x thickness
80 .50 x thickness

60
Magnitude [dB]

40

20

−20
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Frequency [Hz]

Figure 4.10: Frequency response plots of three specimens with varying delamination pa-
rameters.

line the larger the reduction in resonance frequencies. This is explained by the fact
that the flexural rigidity of the delaminated region (EI2 + EI3 ) decreases as the
plane moves to this neutral line.

In table 4.4 the obtained resonance frequencies are compared with the theoretical
values calculated with the beam element model. This model is used as reference
since it provides an easy way of adding the mass for the accelerometer. The diffe-
rences between theory and validation for the intact specimen stand out. These
differences question the reliability of the used experimental set-up. In order to
investigate this theory, the shaker was removed from the set-up and the specimens
were excited by a hammer impact, as can be seen in figure 4.11. The accelerometer
again was used to record the response of the excited beam. The hammer-impact
was applied at the same location the shaker was fixed to make sure all modes of
interest were excited. The obtained results, listed in table 4.5, show for the intact

U1.00.000 U1.06.025 U1.06.050

Mode Theory Val. Theory Val. Theory Val.
1st 18.7 18.1 18.6 17.5 17.5 16.3
2nd 121.3 150.6 116.6 175.5 108.3 124.4
3rd 339.8 343.6 262.8 273.8 216.3 243.8
4th 671.1 634.4 530.1 488.1 440.4 413.1
5th 1118.1 1073.1 909.0 862.5 744.1 653.8

Table 4.4: Comparison between theoretical and validated (shaker) resonance frequencies
[Hz].
64 Delamination detection based on dynamic behaviour

Impact hammer

Force
Accelerometer

Figure 4.11: Experimental set-up (hammer-impact) for modal analysis.

beam far better comparison with the theoretical values. From this is concluded that
the hammer-impact is a reliable way of obtaining the resonance frequencies. For
the delaminated specimens the results show reasonable comparison with theory.
The differences observed can be explained by the fact that the delaminations had a
length of 60% of the beams length. The analytical model required the delaminated
segments to stay in contact and slide among each other. With delaminations this
size it is practically impossible to satisfy this requirement. Also if there exists
contact between the layers there probably occurs some friction, which can also
influence the results.

U1.00.000 U1.06.025 U1.06.050

Mode Theory % Val. Theory % Val. Theory % Val.
1st 18.7 .5 18.8 18.6 6.3 17.5 17.5 3.6 16.9
2nd 121.3 0 121.3 116.6 10.3 130 108.3 9.3 119.4
3rd 339.8 .8 342.6 262.8 1.1 260.0 216.3 2.6 210.8
4th 671.7 .2 673.1 532.8 .5 530.1 440.4 4.4 421.9
5th 1118.1 .7 1110.0 909.0 1.3 910.2 744.1 5.5 705.6

Table 4.5: Comparison between theoretical and validated (hammer-impact) resonance fre-
quencies [Hz].

4.3.3 Conclusions

In this first testing stage the experimental set-up was validated in terms of repro-
ducibility and robustness. Furthermore different sensors for capturing the response
of vibrating test specimens have been evaluated. At last the obtained resonance fre-
quencies of specimens, with varying delamination parameters, have been compared
with the theoretical analysis in section 3.1. From this the following conclusions can
be drawn:

• In terms of reproducibility and robustness the experimental set-up does sat-

isfy the requirements for scientific research. However, the shaker has a
significant influence on the found resonance frequencies, especially for the
lower modes. For experiments in which the results are to be compared with a
Chapter 4. Experimental validation 65

theoretical analysis this set-up can therefore not be used. The resonance fre-
quencies can in this case be obtained by an impact analysis. The specimen is
excited by a hammer-impact and the response is captured by an accelerome-
ter. Results obtained this way show better comparison with theory.
• Capturing the response by an accelerometer gives the best results; the mea-
surements showed good coherence and the resonance frequencies can easily
identified. It is harder to measure the beams response with strain gauges.
This is caused by a poor signal to noise ratio. Only for low vibration modes
the strain is large enough to distinguish it from the noise level. Capturing the
response of a vibrating beam by means of recording the radiated noise seems
a reasonable alternative. It might be difficult, however, to identify torsional
or flexural vibration modes. Also the surrounding noise has a large influence
on the obtained frequency response function.
• The results obtained by the hammer-impact for the intact beam specimens
correspond well with theory. For the delaminated specimens differences
were found. However these discrepancies were small and are probably caused
by the size of the delaminations. Theory requires that the delaminated seg-
ments stay in contact and slide amongst each other. However, since the de-
laminations measured 60% of the beams length, it can be argued whether
this requirement can be satisfied. The experimental programme needs to be
extended to validate the theoretical models.

4.4 Stage B: Application of fibre bragg gratings for

health monitoring

The goal of this stage was to investigate the possibility of using the fibre bragg
gratings for dynamic health monitoring. Fibre bragg gratings can be embedded
in composite structures and are able to measure strain at different positions, this
makes them a potential interesting monitoring technique. Also the fact that fibre
bragg gratings show good resistance against corrosion and fatigue is an advantage
over conventional strain gauges. Various tests were conducted and finally conclu-
sions were drawn on future applicability of fibre bragg gratings for dynamic health
monitoring.

4.4.1 Goals and proposed testing scheme

The accuracy of the strain measurements with fibre bragg gratings was evaluated
with a static test. Beam specimens containing the sensing fibres were cantilevered
and a force was applied at the free end. The strain near the clamped region was
measured by an embedded fibre bragg grating and compared with theory. The goal
66 Delamination detection based on dynamic behaviour

of these tests was to determine the general accuracy of the gratings. Both dis-
placement and applied force were measured at the free end. The validity of the
obtained data was checked by the calculated flexural rigidity. The laminate lay-up
of the beam specimens was tailored to ensure that the fibres stayed on their de-
sired location during production. In order to investigate the applicability of fibre
bragg gratings for dynamic health monitoring dynamic experiments were also con-
ducted. These tests also determined the suitability of the data acquisition system
for dynamic signal processing. The following testing programme was adopted:

1. Static testing

• Force-displacement Beam shaped specimens were cantilevered. A

force cell pushed the free end of the specimen down. The applied force
and displacement at the free end were captured. The results were ana-
lysed and the flexural rigidity of the beam specimens was obtained.

• Strain measurements The strain near the clamped end was measured
with the fibre bragg gratings. The results were compared with the theo-
retical case. Conclusions were drawn on the accuracy of the strain
measurements.

2. Dynamic testing For the dynamic experiments the set-up provided in stage
A was used. The resonance frequencies of the beam were obtained with
help of the accelerometer. The shaker was then used to excite the beam in
a number of these resonance frequencies. The strain near the clamped end
was obtained by the fibre bragg gratings. The quality of the response was
visually investigated. The data was captured by a data acquisition system
with a sampling rate of 50kHz. The system only allowed a record time of one
second. The time signal for various driving frequencies were transformed by
means of a fast fourier transform. These tests gave insight in the maximum
frequency up to which the data acquisition system is still able to capture the
beams response.

4.4.2 Experimental set-up for static testing

In figure 4.12 the experimental set-up is outlined. A force cell is used to push
the cantilevered specimen down. The displacement of the free end is recorded by
means of a laser distance measurement device. Both the force and displacement
at the free end are recorded. The data acquisition scheme is shown in appendix
A. The maximum delamination length in the test specimens in this experimental
stage equals 30% of the beams length. As was shown in section 3.1 the influence
of such small delaminations on stiffness can be neglected. No real difference in
bending stiffness between a delaminated and intact beam is thus expected. The
Chapter 4. Experimental validation 67

Laser distance meter

Force cell

Figure 4.12: Experimental set-up for static tests.

flexural rigidity of the specimens can be determined as follows:

F L3
EI =
u 3
In which F/u represents the beam stiffness and is obtained from the experimen-
tal set-up and L represents the beams length. The theoretical flexural rigidity of
laminated beams is derived in section 3.2.1.

4.4.3 Specimen preparation

A total of six beam specimens, of which five contained a sensing fibre, were sawn
from a laminated CFRP plate. In order to maximize the measured strain, the fibre
bragg gratings were positioned far from the neutral bending surface. To make sure
the fibre gratings stayed on their desired distance from the neutral surface during
the manufacturing process, the fibres were placed between two 90o layers. This
is illustrated in figure 4.13. The carbon fibres in the 90o layers will prevent the
layers to move in the melted matrix material during production. Space was created
in the 0o to minimize deformation of the sensing fibres. However, as can be seen
from the figure the fibre diameter (approx. 195µm) is larger than the thickness
of the individual layers (approx. 150µm). During production the sensing fibre is
therefore compressed.

The gratings were placed in the sev- Sensing fibre

enth layer from the symmetry plane in
90o
a [90/0/90/02 /90/0/90]s plate. De-
laminations were created on the sym- 0o
metry plane of the laminate. The lay- 90o
up ascertained that the segments above
and below the delamination are sym-
metric, which prevents failure due to Figure 4.13: Cross-section of the top side
thermal residual stress induced curva- of the laminate, showing the placement of
ture. The theoretical flexural rigidity the fibre bragg grating between two 90o
layers.
of the beams equals 1.154 N m2 .
68 Delamination detection based on dynamic behaviour

One intact and five delaminated specimens were fabricated. Table 4.6 lists the
various beam specimens. The delaminations were specifically located to validate
the statement that each mode is affected differently by delamination location. From
the analytical analysis in section 3.1 it followed that the effect of the delamination
on the second resonance frequency is minimal for specimens with a delamination
at .52 of the length. For the other specimens, with a delamination at .32 times the
length, the effect is minimal on the third resonance frequency. The last specimen,
U2.grow, will be used in stage C to monitor the effects of a growing delamination.
The middle of delamination is located at a dimensionless distance ac from the
clamped end and has a dimensionless length of b. The parameters h2 gives the
dimensionless thickness-wise location. Again this is summarized in appendix C.

Specimen Specimen size Del. parameters

ID L (mm) w (mm) h (mm) ac b h2
U2.000.00 455 20.1 2.31 na. na. na.
U2.052.02 455 20.1 2.32 .52 .2 .5
U2.052.03 455 19.9 2.32 .52 .3 .5
U2.032.02 455 20.0 2.32 .32 .2 .5
U2.032.03 455 20.0 2.32 .32 .3 .5
U2.grow 435 19.9 2.31 .32 .05 .5

Table 4.6: Specimen dimensions and configuration stage B.

4.4.4 Evaluation of the experimental results

Validation of the static test set-up

To validate the experimental static set-up, described in section 4.4.2, several tests
were conducted on a [0]16 carbon-polyetherimide test beam specimen of which
the material properties are known. After several distinct tests a standard devia-
tion of approximately .7% was obtained. Also the determined Youngs modulus
corresponded well with the theoretical value. From this is concluded that the ex-
perimental set-up shows reproducible and reliable results.
Evaluation of strain measurements
Static measurements on three of the five specimens were conducted to investigate
the general accuracy and resolution of the gratings for strain measurements. A
50kHz data acquisition system was used to investigate the resolution of the grat-
ings. It was found that the signal was considerably corrupted with noise. The
measured strain in the specimens showed, in the worst case, a standard deviation
of 3.7µǫ. The specimen with the best signal to noise ratio showed a standard devi-
ation of 1.3µǫ. Normally a standard deviation of 1.0pǫ is measured in similar fibre
bragg gratings. These, however, are not embedded in a laminate. Possible causes
of this high level of noise are discussed in section 4.4.5.
Chapter 4. Experimental validation 69

Table 4.7 lists the obtained results from the static tests. The obtained flexural rigid-
ity, measured with force and laser displacement devices, correspond well with the-
oretical value of 1.15N m2 . The strain was measured using a 1Hz data acquisition
system. It can be seen, in table 4.7 that the measured strain deviates a lot from
the theoretical case. A maximum difference of 27% is observed. The difference
between validated and theoretical strain can have various causes, which are again
discussed in section 4.4.5.

Specimen STDEV Measured Measured Analytical Difference

ID noise (µǫ) rigidity (N m2 ) strain (µǫ) strain (µǫ) (%)
U2.000.00 3.7 1.133 168.0 201.7 17
U2.052.02 2.3 1.161 163.1 224.1 27
U2.052.03 1.6 1.152 194.5 257.5 24

Table 4.7: Results from static experiments.

Evaluation of dynamic results

The dynamic analysis was performed in order to investigate the suitability of the
50kHz data acquisition system for fast digital signal processing and the capacity of
the gratings to pick-up the different resonance frequencies. Since the data acquisi-
tion system was still in its development phase no software was available to perform
a modal analysis, also the recording time was limited to one second. Therefore the
quality of the obtained time signal was investigated visually. First the resonance
frequencies of the specimens were determined with an accelerometer. Then the
beam specimens were excited in their resonance frequencies and the time-strain
signal was acquired. The obtained data was transformed by means of a fast fourier
transform.
Table 4.8 shows the obtained results. It can be seen that the fibre bragg gratings is
able to resolve the right frequencies. The upper graphs in figures 4.14 to 4.16 give
the obtained fourier transform of the intact specimen vibrating in its resonance

Specimen Driving Obtained FBG

ID frequency (Hz) frequency (Hz)
U2.000.00 11 11
- 368 369
- 910 910
U2.052.02 96 96
- 355 355
- 771 771
U2.032.02 87 87
- 591 591

Table 4.8: Results from dynamic experiments.

70 Delamination detection based on dynamic behaviour

frequencies. In the lower graph the wavelength of the reflected light is plotted
against the elapsed time. As stated in the introduction the shift in wavelength is
coupled to the strain of the grating.From figure 4.16 can be seen that for higher
vibration modes the strain becomes very small.

FFT FFT
200 200
Amplitude

Amplitude
100 100

0 0
0 200 400 600 800 1000 0 500 1000 1500
Frequency [Hz] Frequency [Hz]
Wavelength [nm]

Wavelength [nm]
1560.42
1560.4
1560.4
1560.39
1560.38
1560.38
0 0.02 0.04 0.06 0.08 0 0.05 0.1 0.15
Time [s] Time [s]

Figure 4.14: FFT and time signal for a Figure 4.15: FFT and time signal for a
driving frequency of 196Hz. driving frequency of 534Hz.

The driving frequency of 910Hz can

FFT
still be found in the fourier plot, but 30
Amplitude

it is a very weak signal. Also this 20

FFT plot shows peaks at lower fre- 10

quencies. The peaks are located at 50 0

0 500 1000
Frequency [Hz]
to 550Hz with intervals of 100Hz. The
Wavelength [nm]

same peaks are also observed in fig- 1560.4

ure 4.15. However because of the fact 1560.39

that the strain, caused by the shaker vi- 1560.38

0 0.02 0.04 0.06 0.08 0.1
bration, is a lot higher for this vibra- Time [s]

tion mode the lower frequency peaks

are small compared to the driving fre- Figure 4.16: FFT and time signal
quency peak. for a driving frequency of 910Hz.

A detailed examination showed that in all signals these lower frequencies were rep-
resented. The origin of frequencies can be found in the AC/DC power conversion.
After some an investigation the problem was solved.

The limiting factor in this testing phase was found to be the poor signal to noise
ratio. The strain can not be distinguished from the background noise level. This
plays a bigger role for higher vibration modes as the occurring strain becomes
smaller. In stage A, where the response of the beam specimen was captured with
strain gauges this problem was also observed. Earlier research in the suitability
of the data acquisition system had shown that the system was able to capture the
 Chapter 4. Experimental validation 71

signal reliably up to a driving frequency of 5kHz. However, further research in this

field is desired.

4.4.5 Discussion

The possible use of fibre bragg gratings for dynamic health monitoring has been
investigated in this experimental programme. The largest problem encountered lies
in the poor signal to noise ratio. A standard deviation of 3.7µǫ was measured. This
can be caused by several factors. Generally a standard deviation of 1.0pǫ is ob-
served, however as stated these fibres are not embedded in a laminate. The produc-
tion process of the laminate in which the fibres were embedded can have a severe
influence on the signal strength. It is known that high temperatures can cause the
grating marks to fade out, which results in an attenuation of signal strength. Since
matrix material polyetherimide is consolidated during production at a temperature
of 300o C this might have had a bad influence on signal strength. An experiment,
conducted during another research project, in which the signal strength was moni-
tored during the consolidation of the matrix material confirms this theory.

Coating

Cladding
195µm 6µm 128µm
Core

Figure 4.17: Dimensions of fibre bragg Figure 4.18: Micrograph of embedded

grating. fibre bragg grating (25x).

Also the applied pressure of 9 bar during the production process can have had a
negative influence on signal strength. As stated before the sensing fibres are em-
bedded between two 90o layers. Figure 4.17 indicates that the outer diameter of the
fibre equals 195µm. With a carbon-PEI layer thickness of approximately 150µm
this effectively means that the grating is compressed between two layers. Micro-
graphs of an cross-section of an embedded fibre, shown in figure 4.18, confirm this.
The cross-section shows that the fibre has an elliptic form, while originally the fibre
should show a circular cross-section. The drawn white ellipses show the position
of the coating and cladding. It can be seen that they are not perfectly concentric.
Figure 4.19 shows this in detail. Similar results were obtained for other specimens,
i.e. figure 4.20. It is not known if this has a negative influence on accuracy or signal
72 Delamination detection based on dynamic behaviour

strength. The dark areas in the micrographs are caused by water seeping through
the voids in the laminate. Figure 4.20 shows that the ceramic coating of the fibre
bragg grating is broken. Because the broken pieces are embedded by the matrix
material it can be concluded that this occurred during the production phase of the
laminate. It is unknown if such damage has a negative effect on signal strength
and accuracy. Extended research in the influence of the production process on the
quality of the fibres is therefore required.

Figure 4.19: Micrograph of embedded Figure 4.20: Micrograph of embedded

fibre bragg grating (200x). fibre bragg grating (200x).

The difference between theoretical and experimental measured strain is the second
problem encountered. The difference mounts up to 27%. The theoretical strain
near the clamped end followed from:
3δ · t
ǫ= (4.1)
L2
In which δ is the displacement at the free end of the beam, L is the length of
the beam and t the distance of the grating to the neutral plane. This equation
assumes that the strain develops linearly over the thickness. For an ideal clamping
condition, indicated in figure 4.21a, this might be the case. However, practically
such a clamping condition is impossible. Figure 4.21b gives the used clamping
condition. A finite element analysis in ANSYS evaluated the influence of the used
clamping conditions. The results for both clamps are given in figure 4.21c and
4.21d. A clear difference in strain distribution is observed for the two cases. The
analysis further showed that the used clamp can result in a difference of the strain
near the gratings of at most 10%.
Another uncertainty in the strain measurements is caused by the fact that the exact
axial location of the gratings is not known. During the placement of the sensing
fibres an error of approximately 10mm can be introduced. This is caused by the
fact that it is impossible to identify the gratings by naked eye. Additionally the
grating has a length of 8mm over which the strain is measured.
However, because the mentioned causes cannot explain the large difference be-
tween measured and theoretical strain, of at most 27%, additional tests were per-
formed. In order to investigate if embedding a fibre bragg grating has a negative
Chapter 4. Experimental validation 73

a. Theoretical clamp b. Used clamp

c. Strain field d. Strain field

Figure 4.21: Finite element analysis of clamping condition.

influence on accuracy a tensile-strength test was conducted. For this purpose two
specimens were fabricated. One unidirectional 16-ply carbon-PEI beam with a
fibre bragg grating embedded on the symmetry plane. The other specimen was
similar to the ones used in this stage and had a [90/0/90/0/0/90/0/90]s lay-up.
In this specimen the fibre bragg grating was embedded in the seventh layer from
the symmetry plane. Besides measuring the strain with the embedded fibre, this
was also done with a fibre bragg grating glued to the surface of the test specimen.
Additionally, the strain was measured by an external strain meter. The tests for both
specimens showed that no differences at all were found in the different strain mea-
surements. These tests show, as opposed to the previous tests, that the strain can
very well be measured with an embedded fibre bragg grating. However, the strains
that occurred during the bending test were significantly lower than the strain that
arose during the tensile stress. Further research in order to explain the differences
between theoretical and measured strain is required.

4.4.6 Conclusions and recommendations

Both static and dynamic experiments were conducted to investigate the possibilities
of using fibre bragg gratings for dynamic health monitoring. The obtained results
indicate a good perspective for future applicability of the fibre bragg gratings. The
sampling rate of the data acquisition system is probably high enough to resolve
frequencies up to 5kHz, however more research in this field is desired. One of the
problems encountered was the lack of a software package to extract an averaged
autospectrum of multiple measurements. However, the system is still in its devel-
opment phase and sufficient improvement in this field can be expected. The largest
problem at this moment is the low signal resolution and accuracy. Due to the low
strain for higher vibration modes and high noise level it was impossible to resolve
these higher resonance frequencies. The poor signal quality is known to be caused
by the production process of the test specimens in which the fibres were embedded.
It is recommended to concentrate future research in the following fields:
• To verify the accuracy of the fibre bragg gratings a four point bending test
on a test specimen should be performed. The strain is then constant between
74 Delamination detection based on dynamic behaviour

the two loading points. Also the problems described in the previous section
with the clamping conditions are avoided this way.
• The effect of different processing conditions on general resolution and accu-
racy should be investigated thoroughly. It is known that high temperatures
can cause the gratings to fade out. It has to be examined in what extend the
applied temperature and pressure influences the accuracy and resolution.
• As an extension to the previous point it might be valuable to monitor the
wavelength signal during the production process. The attenuation of sig-
nal strength can give insight in the influences of different production steps.
While monitoring the changes in wavelengths can give interesting informa-
tion about the development of residual stresses inside the laminate when
cooling down.
• In order to test the potential a fibre bragg grating can be glued to the surface
of a beam specimen. Signal attenuation due to the production process are
avoided this way. The high accuracy allows an thorough investigation of the
data acquisition system.
• Furthermore software should be developed for the data acquisition system to
enable easy extraction of the resonance frequencies of test specimens.

4.5 Stage C: Validation of theoretical analysis

The goal of this stage was to validate the theoretical model provided in section 3.1.
The specimens used in stage B were subjected to a modal analysis. It was observed
in stage A that the experimental set-up with the shaker did not produce accurate
results. The resonance frequencies of the specimens were therefore determined by
measuring the response of the beam excited by a hammer-impact, as given in figure
4.11.

4.5.1 Testing scheme

The test specimens fabricated in stage B were all excited by a hammer-impact.

The response was captured with an accelerometer. The found resonance frequen-
cies were then compared to the ones calculated with the finite beam element model
described in section 3.3.1. The finite element model was chosen as a reference be-
cause this provided an easy way of adding mass for the accelerometer. The flexural
rigidity of the beam segments was determined as given in section 3.2.1. Further-
more the effect of a growing delamination was studied. A specimen containing
a small delamination was manufactured and its resonance frequencies were deter-
mined. Then a razorblade was used to enlarge the delamination in a controlled
Chapter 4. Experimental validation 75

way prior to subsequent measurements. This way the effect of a delamination can
closely be monitored.

4.5.2 Evaluation and discussion

Pre-delaminated specimens
Several hammer impact measurements were conducted and averaged. The data was
used to obtain an autospectrum of the response. Figures 4.22 and 4.23 shows the
response of respectively the intact and U2.052.02 specimen to the hammer-impact.
The resonance frequencies can easily be identified by the peaks in the plots. A
quick glance shows that the peaks in figure 4.22 appear to be steeper. The wider
peaks in figure 4.23 are caused by an increase in damping, probably as a result of
friction between the two delaminated segments. This subject, however, is not co-
vered in this thesis. In table 4.9 the obtained resonance frequencies are compared
to the ones calculated with the beam element model, provided in section 3.3.1. It
can be seen that for the intact beam specimen the theory corresponds very well with
the validated resonance frequencies. An averaged difference of 2.0% exists. How-
ever if the first resonance frequency, due to the high influence of the sampling rate,
is excluded, an averaged difference of 0.8% is found. A similar result, an averaged
difference of 0.4%, was observed in stage A, table 4.4. It can again be concluded
that the results obtained by the hammer-impact are reliable. From table 4.9 it fol-
lows that the first and second resonance frequency for the intact specimen is lower
than for the delaminated beams. The smaller theoretical resonance frequencies are
caused by a differences in beam geometry. The intact specimen, U2.000.00 had
a thickness of 2.31mm compared to 2.32mm for the delaminated specimens. This
causes a reduction in flexural rigidity of 1.3%. This difference can be caused by the
fact that the intact specimen, U2.000.00 was sawn from the edge of the laminated
plate.
Further it can be seen that for the delaminated specimens the theoretical results

Autospectrum Autospectrum
−50 −50
Magnitude [dB]
Magnitude [dB]

−100
−100
−150
−150
−200

−200 −250
0 500 1000 1500 2000 0 500 1000 1500 2000
Frequency [Hz] Frequency [Hz]

Figure 4.22: Autospectrum of intact spec- Figure 4.23: Autospectrum of delami-

imen U2.000.00 excited by a hammer- nated specimen U2.052.02 excited by a
impact. hammer-impact.
76 Delamination detection based on dynamic behaviour

show more deviation to the obtained values than for the intact specimen. An av-
eraged deviation of 3.1% is found. The maximum difference of 10.2% is found
for the fourth resonance frequency of specimen U2.052.03. The differences can be
caused by several factors:
• The delaminated segments are modelled to stay in contact during vibration.
This however is a requirement that is practically very difficult to satisfy. This
also depends on the vibration mode considered.
• When contact occurs the theory demands that the delaminated segments are
able to slide freely in relation to each other. However practically there will
exist friction between the two segments. This friction can hinder the seg-
ments to slide, which then will result in an increase of flexural rigidity. The
inserted kapton foil can have an additional negative effect on this friction.
The foil thickness causes the contact pressure to increase with negative in-
fluence on friction forces. Also the friction coefficient of the foil can have a
negative influence.
• In conjunction with previous point; the actual delamination, i.e. the region
where the segments are able to slide, is probably smaller than the size of the
inserted kapton foil. The kapton foil pushes the layers outwards, which re-
sults in a large contact pressure near the junction. This prevents the layers to
slide freely near the junction and this effectively means that the delamination
is probably shorter than intended.
It can be seen, despite the differences, from table 4.9 that the theoretical and vali-
dated results show a similar tendency.
In figure 4.24 it is again underlined to what extend delamination location influ-
ences frequency shifts of different modes. This figure plots the autospectra of two
delaminated specimens. The delaminations were created on different locations.
The first specimen, U2.052.03, has a delamination on .52 of the beam length. The
lower graph in figure 4.24 indicates that for the second vibration mode the third
derivative, and thus the shear force, is zero in this location. This means that the de-
lamination has only a small influence on the second vibration mode. This also goes
for the fourth vibration mode, as can be seen from figure 4.24. The delamination in
the other specimen, U2.032.03, will have a small influence on the third resonance
frequency. The delamination is located on a point of minimum shear force for
this vibration mode. The upper graph of figure 4.24 shows the obtained autospec-
tra of the two specimens. The graph shows that the second and fourth resonance
frequency has shifted more for the U2.032.03 specimen than for the U2.052.03
specimen. The opposite is true for the third resonance frequency. This corresponds
qualitatively very well with the theoretical analysis.
Chapter 4. Experimental validation 77

Autospectrum
−50 U2.052.03
Magnitude [dB]

U3.032.03
−100

−150

−200

−250
0 50 100 150 200 250 300 350 400 450 500
Frequency [Hz]

Third derivative of mode shapes

1 2nd Mode 3rd Mode 4th Mode
0.5

−0.5

−1
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Dimensionless distance

Figure 4.24: Influence of delamination location on resonance frequency shifts for different
vibration modes.

Growing delamination
The effects of a delamination on the resonance frequencies is also investigated by
enlarging a small delamination in between several modal analyses. This prevents
unwanted frequencies shifts as a result of variations in specimen dimensions or
material properties. The delamination was enlarged with a sharp razorblade. The
effects of a delamination on the resonance frequencies can be monitored closely
this way.
Figure 4.25 visualizes this effect. Several measured frequency response plots are
plotted in this graph. As can be seen from the figure the effects of a growing de-
lamination are most noticed for the higher vibration modes; the sixth and seventh
resonance peak clearly shift to the lower values. For the lower vibration modes
the effect is less significant. The sudden shift for the third and fifth resonance fre-
quency for a delamination length of 142mm stands out. In table 4.9 the results
are compared to the theoretical analysis. For a small, up to 42mm, delamination
length the obtained theoretical results compare very well to the validated resonance
frequencies. However as the delamination is enlarged the results show less compa-
rison. Possible explanations were already listed above. One additional reason can
be found in the fact that the delaminated surface was very rough. This is caused
by the fact that the delamination was forced to grow with the razorblade. This can
increase the occurring friction. Previous experiments were only conducted on a
smooth predefined delamination surface.
78 Delamination detection based on dynamic behaviour

Autospectrum

b= 142mm

b= 89mm
Magnitude

b= 65mm

b= 42mm

b= 31mm

0 200 400 600 800 1000 1200 1400 1600 1800 2000
Frequency [Hz]

Figure 4.25: Effects of a growing delamination.

4.5.3 Conclusions and recommendations

From the results obtained in this experimental stage it can be concluded that the
developed model gives a good indication of the effects of a delamination in vi-
brating beams. The theoretical and validated resonance frequencies show a similar
trend. An average difference of 3.1% between theoretical and measured resonance
frequencies of delaminated beams was found. This value was found for the pre-
defined delaminations. Less comparison, 3.7%, is found when the delamination
was forced to grow. This is probably caused by the newly created rough delami-
nation surface. Because it is unknown in practice how much the measured reso-
nance frequencies deviate from the theoretical model it is hard to use the model
for damage localization. For instance, a difference of 3% on the measured second
resonance frequency gives significant other results than the same difference on the
third vibration mode. Also a more accurate model is desired to be able to predict
the shifts in resonance frequencies better. The new model needs improvement in
the following fields:
• Implement the effect of an opening delamination. As stated above the re-
quirement that the delamination stays closed during vibration is hard to sat-
isfy practically. The finite element model using contact elements can be
extended for this purpose. This will, however, make the analysis far more
time consuming since the vibration becomes non-linear.
 Vibration U3.000.00 U3.052.02 S3.052.03 U3.032.02 S3.032.03
Mode Theory Validation Theory Validation Theory Validation Theory Validation Theory Validation
1st 12.0 11 12.1 12 12.0 12 11.7 12 11.4 12
2nd 77.2 76 77.9 80 77.5 83 73.4 74 65.9 65
3rd 216.4 221 191.7 199 158.6 170 220.2 220 208.8 208
4th 428.3 432 424.8 440 374.6 417 360.2 371 313.1 339
5th 713.3 716 568.6 580 532.7 546 627.3 609 555.4 536
6th 1071.4 1068 957.0 993 793.8 813 963.3 979 806.5 775
7th 1502.1 na.ζ 1248.3 1233 1175.6 1204 1247.9 1239 1183.4 1119
Vibration Del size: 31mm Del size: 42mm Del size: 65mm Del size: 89mm Del size: 142mm
Mode Theory Validation Theory Validation Theory Validation Theory Validation Theory Validation
1st 13.03 13 13.0 13 12.9 13 12.7 13 12.5 13
2nd 82.5 82 82.5 82 80.7 81 78.6 81 77.9 84
3rd 235.3 236 234.8 237 226.5 237 187.3 223 164.4 208
4th 458.2 462 446.2 449 389.9 401 369.7 373 355.5 383
5th 758.7 764 747.1 748 733.7 751 591.1 746 551.1 557
6th 1160.1 1160 1142 1159 939.4 1021 882.6 878 832.9 821
7th 1545.9 1570 1445.7 1516 1392.9 1368 1209.8 1261 1169.4 1159
ζ Unable to identify from autospectrum

Table 4.9: Comparison of theoretical and validated resonance frequencies of specimens with varying delamination parameters and with a growing
delamination
79
80 Delamination detection based on dynamic behaviour

• Implement the friction between the two delaminated segments. This also can
best be achieved using the contact elements.
In order to validate the analytical and finite element models proposed in chapter
3 other beam specimens should be produced. The uncertainties regarding contact
pressure and surface roughness should be avoided. To minimize these issues, spec-
imens of for instance aluminium should be considered. Delaminations can then be
created by gluing two aluminium beams. The glue will probably influence the mea-
sured damping coefficient but its influence on resonance frequencies will probably
be small.
 CHAPTER
FIVE
DISCUSSION AND APPLICATION

W ith the presented theoretical models the resonance frequencies of a delami-

nated beam can be determined. In this section it is discussed in what extend this
model can be used to detect and locate originating delaminations.

5.1 Discussion

A general problem concerning dynamic ’health monitoring’ consists of the sensi-

tivity of dynamic parameters to environmental changes. In the literature survey it
was already stated that these properties are highly influenced by changes in tem-
perature or humidity. Kawiecki [16] reported that the resonance frequencies of a
steel bridge showed variations up to 5% as a result of temperature changes only.
This dependency relinquishes the use of non-model based techniques outside tem-
perature and humidity controlled environments. A model or extensive database is
required to distinguish damage or environmentally induced changes. In this light
the application of such systems in for instance airplanes or wind turbine blades
still has a long way to go. However in the field of quality control such problems
play no role as a stable environment can easily be created. Dynamic ’health mon-
itoring’ could therefore be a strong short-term alternative to conventional quality
check procedures.
An analytical and two finite element models were presented to determine the res-
onance frequencies of delaminated beam shaped specimens. The model proved
that with the simple measurement of resonance frequencies, delaminations can be
detected and located. An advantage of resonance frequencies over other dynamic
parameters can be found in the fact that resonance frequencies show insignificant
statistical variation. It was experimentally shown that the standard deviation be-
tween three measurements at distinct times was found to be less than 0.2%. This
allows to detect changes of 1%, which makes resonance frequencies a very accu-
rate parameter to model. For the proposed model however an averaged difference

81
82 Delamination detection based on dynamic behaviour

between theory and validated resonance frequencies of 3.1% was found. It has to
be added that this figure applies to the pre-delaminated specimens with a relatively
smooth delaminated surface. For the specimen with the enlarged delamination
an averaged difference of 3.7% was obtained. For the detection of damage these
figures play no role. For the localization however this difference might prove prob-
lematic. As indicated in the theoretical model the location is determined by the
combination of frequency shifts for various vibration modes. Only for the higher
modes a frequency drop of more than 5% is expected. By taking more vibration
modes into account the determined results will naturally become more reliable. In
order to be able to use the lower vibration modes a more accurate model is required.
This thesis also investigated the possible application of fibre bragg gratings for
health monitoring. The advantages over conventional methods to capture vibra-
tions are obvious. The most notable can be found in the fact that fibre bragg grat-
ings can easily be embedded in composite structures. This allows strain measure-
ments inside the host material. Also embedding sensing fibres can be preferred
over gluing strain gauges or accelerometers for practical reasons. Additionally, the
matrix material will protect the sensors which allows application in harsh environ-
ments. Other advantages can be found in the ability of multiplexing, i.e. having
multiple sensors on one fibre, and the high resistance against fatigue. In an ex-
perimental programme the fibre bragg gratings were evaluated. The main problem
encountered was found to be the low signal to noise ratio. Together with the low
strains for higher vibration modes, this made it impossible to identify the higher,
over 1kHz, resonance frequencies. The data acquisition system itself has a sam-
pling rate of 50kHz, which is certainly fast enough to resolve such frequencies.
The poor signal to noise ratio is probably caused by the process conditions of the
host material in which the fibres are embedded. The high temperature is known
to have a negative influence on signal strength. Extensive research in this field is
required.
Despite the low signal to noise ratio, fibre bragg gratings will certainly have a pos-
itive influence on the development of dynamic based health monitoring systems.
The ability of embedding sensing fibres inside a composite together with the possi-
bility of multiplexing gives a potentially strong measurement system. A network of
gratings could be used to survey local strain development over a large area. Besides
determining resonance frequencies, a network of sensors will then give the possi-
bility to define mode shapes or determine cross-spectra between different gratings.
New algorithms can be developed to make use of all this information simultane-
ously. Also the development of small lamb waves through the host material can
be monitored for damage detection. The development of fibre bragg gratings thus
provides new opportunities for dynamic health monitoring.
Chapter 5. Discussion and application 83

5.2 Application

The model proposed in section 3 determines the resonance frequencies of delami-

nated beams. Naturally, the opposite is desired for dynamic health monitoring, i.e.
determining the delamination parameters from multiple measured resonance fre-
quencies. In order to achieve this, the presented analytical solution can be rewritten
to determine these three parameters (ac , b, h2 , see appendix C). A total of three
frequency shifts are required to resolve these unknowns. However, because this is a
non-linear function, every small error introduced in the measured frequency shifts
can result in large deviations. In an attempt to get around this problem, the model
is used to construct a database with delamination parameters and their accompany-
ing relative resonance frequency shifts. The fact that multiple shifts are taken into
account means that errors in the measured resonance frequencies will have less
influence on the found delamination parameters. The number of vibration modes
included, determines the accuracy of the prediction. Also a larger database, with
smaller increments in delamination parameters, allows a more accurate localiza-
tion. The general idea is to compare relative resonance frequency shifts for an
arbitrary specimen to the values in the database. The differences for each mode
between the measured shifts and the database values are squared. By summing the
squared differences for all modes some form of error function is obtained. The
delamination parameters for which this sum is minimal can be seen as actual pa-
rameters.
This concept was tested with a database containing a total of 924 delamination
cases. The first six resonance frequencies were considered. The frequency shifts
for several arbitrary delamination parameters were then calculated and compared to
the database. In case the exact input resonance frequencies were used, the database
proved to be large enough to locate the damage in approximately 98% of the cases.
However, if an error, e.g. caused by temperature changes, of 5% was introduced
on one of the input frequencies it proved to be very difficult to resolve the delam-
ination parameters. This corresponds very well with the statements made earlier.
The model is, for now, not accurate enough to use it to locate a delamination. Also
any error in the frequency measurements can have a serious influence on the ob-
tained results. The short-term applicability of dynamic measurements for health
monitoring should therefore be limited to controlled environments.
 CHAPTER

SIX
CONCLUSIONS AND
RECOMMENDATIONS

T his thesis investigated the possibilities of on line health monitoring based on dy-
namic behaviour. A literature survey described the research done in this field. Var-
ious techniques based on resonance frequencies, damping values or other modal
parameters were discussed. It was chosen to investigate how the resonance fre-
quencies of beam specimens are influenced by a delamination. An analytical model
was investigated and extended to determine the resonance frequencies of a delam-
inated composite beam. In addition two finite element models were constructed.
One of these is based on contact elements, which makes it applicable for future
more complex analysis. The model was validated with carbon-PEI test specimens
with varying delamination parameters. The model could be used to locate delam-
ination based on shifts in the resonance frequencies. A neural network could for
instance be used for this purpose. Further this thesis investigated the use of fibre
bragg gratings for dynamic health monitoring; experiments concerned accuracy,
resolution and data acquisition.

6.1 Conclusions

With respect to the various parts of this thesis, the following conclusions can be
drawn:

General dynamic health monitoring

• Health monitoring based on variations in dynamic behavior is a potentially

interesting technique. Damage can influence the material or geometrical
properties of a structure. As a result the dynamic behaviour changes. De-
tected changes in modal parameters, like resonance frequencies, damping

85
86 Delamination detection based on dynamic behaviour

values or mode shapes, are then probably caused by damage. Monitor-

ing these parameters might prove interesting as an alternative for laborious
’scanning’ non destructive damage detection methods.

• Damage detection algorithms can be divided in two groups, namely model

and non-model methods. As stated in the literature survey model based tech-
niques allow localization and quantification of occurring damage. The dis-
advantage of model based health monitoring can be found in the fact that it
should be able to represent the actual damage with these models. This will
not always be the case. Non-model based algorithms are harder to formulate.

• Shifts in resonance frequencies can very well be used as damage indicator.

Damage can be detected at the moment it occurs. Also the potential damage
location can be extracted by monitoring shifts of resonance frequencies of
multiple modes. Resonance frequencies can easily and cheaply be obtained
and do not require access to the whole structure under consideration. Also
resonance frequencies show relatively little statistical variation.

• Damping is another potential interesting dynamic parameter to monitor. Damp-

ing seems to be more sensitive to damage, however no apparent relation ex-
ists.

Delamination detection

• The decrease in flexural rigidity as a result of a delamination is compensated

by an internal bending moment. This moment is caused by the stretching
and compressing of the delaminated segments. In case a delaminated beam
is subjected to a pure bending moment this internal moment will fully com-
pensate the loss in flexural rigidity. Only in case a shear force is present and
the layers are free to slide, the effect of the delamination can be noticed.

• The decrease in flexural rigidity due to a delamination can be detected with

more confidence and accuracy by monitoring shifts of resonance frequency
than by a static analysis. Especially frequencies belonging to higher vibra-
tion modes are affected considerably.

• The shifts in frequencies are higher in case the delamination is located closer
to the neutral plane. Also the length of the delamination has a positive effect
on frequency shift.

• By measuring the shifts of resonance frequencies of different modes the axial

location of the delamination can be determined. This is caused by the non-
uniform shear stress distribution in a vibrating beam. If, for a particular
vibration mode, the delamination is located in a region with low shear stress,
the effect on resonance frequency for that mode will be minimal. This is
caused by the fact that sliding of the delaminated segments will only occur
in case a shear force is present.
Chapter 6. Conclusions and recommendations 87

Fibre bragg gratings

• Fibre bragg gratings can have a positive influence on the development of an

applicable online health monitoring system. The ability to measure strain
on multiple locations inside the host material offers a huge potential. A
network of gratings provides a whole new range of opportunities for damage
detection.

• The differences in accuracy and resolution between embedded and free fi-
bre bragg gratings are a cause for concern. The poor signal to noise ratio
prevented accurate measurements for higher modes. Research should con-
centrate on the influence of production parameters on signal strength and
noise ratio.

• The tested 50kHz data acquisition system provides the ability to resolve res-
onance frequencies up to 5kHz confidentially. However the low signal to
noise ratio prevented extensive research in this field.

6.2 Recommendations

Online health monitoring has a great potential and will in the future replace expen-
sive routine check-ups. Fibre bragg gratings can play an important role. To aid in
its development, the research on this subject should concentrate on the following
fields:

• The current finite element model with the contact elements should be ex-
tended in order to allow the delamination to open and close during vibration.
Also a friction coefficient between the layers should be added. With this
new model delaminations can be detected with more accuracy. However,
the opening and closing of the delamination results in non-linear vibration.
This, however, will seriously complicate the analysis as the solution has to
be obtained in the time domain.

• New algorithms should be developed which use the full potential of fibre
bragg gratings. A network of gratings will be used to map the strain field
in a host structure during vibration. The additional information can be used
to extract mode shapes and cross-spectra between different gratings. A new
algorithm should be able to use all this information for health monitoring.

• In order to use the gratings for health monitoring the problems regarding loss
of accuracy and poor resolution should be solved. The influence of produc-
tion parameters, like consolidation temperature and pressure, on embedded
fibre bragg gratings has to be investigated. Embedding a sensing fibre should
have no influence on its resolution or accuracy.
88 Delamination detection based on dynamic behaviour

• Data acquisition system software has to be developed. New software should

be able to extract autospectra or even frequency response functions from
measured data. The system should be able to average a series of measure-
ments.
 BIBLIOGRAPHY

[1] R. D. Adams, D. Walton, J.E. Flitcroft, and D. Short. Vibration testing as a

non-destructive test tool for composite materials. Composite reliability, page
159, 1975.
[2] R.D. Adams and D.G.C. Bacon. Dynamic properties of unidirectional fibre
reinforced composites in flexure and torsion. Journal of composite materials,
7:53–67, 1973.
[3] P. Cawley and R. D. Adams. A vibration technique for non-destructive testing
of fibre composite structures. Journal of composite materials, 13:161–175,
1978.
[4] P. Cawley and R.D. Adams. Improved frequency resolution from transient
tests with short record lengths. Journal of sound and vibration, 64:123–132,
1979.
[5] P. Cawley and R.D. Adams. The location of defects in structures from mea-
surements of natural frequencies. Journal of strain analysis, 14(4):49–57,
1979.
[6] R. Chandra, S.P. Singh, and K. Gupta. Damping studies in fiber-reinforced
composites -a review. Composite structures, 46:41–51, 1999.
[7] T.G. Chondros, A.D. Dimarogonas, and J. Yao. A continuous cracked beam
vibration theory. Journal of sound and vibration, 215:17–34, 1998.
[8] N.A. Chrysochoidis and D.A. Saravanos. Assessing the effects of delamina-
tion on the damped dynamic response of composite beams with piezoelectric
actuators and sensors. Smart materials and structures, 13:733–742, 2004.
[9] S.W. Doebling, C.R. Farrar, M.B. Prime, and D.W. Shevitz. Damage iden-
tification and health monitoring of structural and mechanical systems from
changes in their vibration characteristics: A literature review. Technical Re-
port LA-13070-MS, Los Alamos National Laboratory, May 1996.
[10] C.R. Farrar and S.W. Doebling. An overview of modal-based damage identi-
fication methods. Proc. of DAMAS Conference, 1997.

89
90 Delamination detection based on dynamic behaviour

[11] P. Gudmundson. Eigenfrequency changes of structures due to cracks, notches

or other geometrical changes. Journal of mechanics, physics and solids,
30(5):339–353, 1982.
[12] F.J. Guild and R.D. Adams. The detction of cracks in damaged composite.
Journal of physics D: Applied physics, 14(8):1651–1573, 1981.
[13] F.J. Guild and R.D. Adams. A new technique for the measurement of the
specific damping capacity of beams in flexure. Journal of physics E: Scientific
instruments, 14:355–363, 1981.
[14] S. Hassiotis and G.D. Jeong. Assessment of structureal damage from natural
frequency measurements. Computers and structures, 49(4):679–691, 1993.
[15] H.H. Hu, B-T. Wang, C-H. Lee, and J-S. Su. Damage detection of surface
cracks in composite laminates using modal analysis and strain energy method.
Composite structures, 74:399–405, 2006.
[16] G. Kawiecki. Modal damping measurement for damage detection. Smart
materials and structures, 10:466–471, 2001.
[17] H.Y. Kim. Vibration-based damage identification using reconstructed frfs in
composite structures. Journal of sound and vibration, 259(5):1131–1146,
2003.
[18] C. Kyriazoglou, B.H. Le Page, and F.J. Guild. Vibration damping for crack
detection in composite laminates. Composites part A: applied sience and
manufacturing, 35:945–953, 2004.
[19] B.T. Lee, C.T. Sun, and D. Liu. An assessment of damping measurement in
the evaluation of integrity of composite beams. Journal of reinforced plastics
and composites, 6:114–125, April 1987.
[20] J. Lee. Free vibration analysis of delaminated composite beams. Computers
and structures, 74:121–129, 2000.
[21] P.M. Mujumdar and S. Suryanarayan. Flexural vibrations of beams with de-
laminations. Journal of sound and vibration, 128(3):441–461, 1988.
[22] W.M. Ostachowicz and M. Krawczuk. Analysis of the effect of cracks on
the natrual frequencies of a cantilever beam. Journal of sound and vibration,
150(2):191–201, 1991.
[23] A.K. Pandey, M. Biswas, and M.M. Samman. Damage detection from
changes in curvature mode shapes. Journal of sound and vibration,
145(2):321–332, 1991.
[24] D.P. Patil and S.K. Maiti. Experimental verification of a method of detection
of multiple cracks in beams based on frequency measurements. Journal of
sound and vibration, 281:439–451, 2005.
Chapter 6. BIBLIOGRAPHY 91

[25] C.J. Pye and R.D. Adams. Detection of damage in fibre reinforced plastics
using thermal fields generated during resonant vibration. NDT International,
pages 111–118, 1981.
[26] R.L. Ramkumar, S.V. Kulkari, and R.B. Pipes. Free vibration frequencies of a
delaminated beam. In Annual Technical Conference, pages 1–5. Reinforced/
Composites Institute, Society of plastics industry, 1979.
[27] A. Rytter. Vibration based inspection of civil engineering structures. PhD
thesis, Aalborg University, Denmark, 1993.
[28] O.S. Salawu. Detection of structural damage through changes in frequency:
a review. Engineering structures, 19:718–723, 1997.
[29] R.P.C. Sampaio, N.M.M. Maia, and J.M.M. Silva. Damage detection using
the frequency-response-function curvature method. Journal of sound and vi-
bration, 226(5):1029–1042, 1999.
[30] D.A. Saravanos and D.A. Hopkins. Effects of elaminations on the damped
dynamic characteristics of composite laminates: Analysis and experiments.
Journal of sound and vibration, 195(5):977–993, 1996.
[31] C.T. Sun, B.T. Lee, and S.K. Chaturvedi. Composite material damping using
impulse technique. In L. Rogers, editor, Vibration Damping 1984; Workshop
proceedings, pages 1–24, November 1984.
[32] L.H. Tenek, E.G. Henneke, and M.D. Gunzburger. Vibration of delaminated
composite plates and some applications to non-destructive testing. Composite
structures, 23:253–262, 1993.
[33] F.G. Tomasel, H.A. Larrondo, and P.A.A. Laura. Detecion of cracks in can-
tilever beams: experimental set-up using optical tecniques and theoretical
modelling. Journal of sound and vibration, 228(5):1198–1204, 1999.
[34] S.H. Diaz Valdes and C. Soutis. Delamination detection in composite lam-
inates from variations of their modal characteristics. Journal of sound and
vibration, 228:1–9, 1999.
[35] J.T.S. Wang, Y.Y. Liu, and J.A. Gibby. Vibration of split beams. Journal of
sound and vibration, 84:491–502, 1982.
[36] L. Warnet. On the effect of residual stresses on the transverse cracking
in cross-ply carbon-polyetherimide laminates. PhD thesis, University of
Twente, 2000.
[37] L. Warnet and R. Akkerman. Composite Course 2006. University of Twente,
2006.
[38] D. Wendtland. Aenderung der Biegeeigenfrequenzen einer idealisierten
Schaufel durch Risse. PhD thesis, University of Karlsruhe, 1972.
92 Delamination detection based on dynamic behaviour

[39] M.M.F. Yuen. A numerical study of the eigenparameters of a damaged can-

tilever. Journal of sound and vibration, 103(3):301–310, 1985.
[40] Y. Zou, L. Tong, and G.P. Steven. Vibration based model dependant damage
(delamination) identification and health monitoring for composite structures
-a review. Journal of sound and vibration, 230:357–378, 2000.
 APPENDIX
A
DATA ACQUISITION AND USED
HARDWARE EQUIPMENT

strain 1.

6.

sound 2.

F
3.
a
5.

4.

shaker

Figure A.1: Data acquisition scheme for the dynamic set-up.

93
94 Delamination detection based on dynamic behaviour

1.
4.
F

2.

Figure A.2: Data acquisition scheme for the static set-up.

3.

FBG 1.

2.

Figure A.3: Data acquisition scheme for the fibre bragg gratings.
Appendix A. Data acquisition and used hardware equipment 95

No Hardware Description
DYNAMIC SET- UP FIG . 4.1
1 Shaker B&K 4802
2 Force transducer B&K 8203
3 Accelerometer B&K 4517C-001
4 Strain gauges TML 350Ω, length 6mm
5 Fibre Bragg grating FOS&S
6 Microphone B&K 4192L
7 Spring
DYNAMIC DATA ACQUISITION FIG . A.1
1 Bridge amplifier Fylde DC 500kHz
2 Charge condition amplifier B&K Nexus
3 Charge condition amplifier B&K Nexus
4 Power amplifier B&K 2706
5 Digital sensor processor Siglab 20-42
6 PC
S TATIC SET- UP FIG . 4.12
1 Force cell HBM 50N
2 Laser distance meter Micro-epsilon non-contact
S TATIC DATA ACQUISITION FIG . C.1
1 Bridge amplifier HBM AC amplifier
2 Data acquisition NI USB-6008 Data Acq. Syst.
3 PC
FBG DATA ACQUISITION FIG . A.3
1 Slow data acquisition SpectralEye 600
2 Fast data acquisition 50kHz Slope DAQ
3 PC

Table A.1: Used measurement hardware.

 APPENDIX
B
SPECIMEN FABRICATION SCHEME

1. The needed amount of plies are cut out of a roll of pre-impregnated carbon-
polyetherimide. Before the material can be used, they are put in a vacuum
oven at 80o C for about 24 hours. The vacuum and high temperature makes
sure that any moisture evaporates.
2. The plies are stacked, in the preferred lay-up, inside a mould. A sheet of
polyamide is placed between mould and stacked pile to prevent the polyether-
imide to stick to the mould. In case fibre bragg gratings are embedded, space
is created by cutting a small strip out of the ply. The sensing fibre is fixed
by sticking it to a ply with a polyamide adhesive tape. Delaminations are
created by placing polyamide kapton foil at desired places. The foil prevents
two plies to melt together and thus a delamination is formed. Meanwhile a
press is heated to a temperature of 316o C1 .
3. The mould is closed and placed inside the press. A pressure of 2bar is
applied for five minutes. The high temperature allows the polyetherimide
matrix to melt. The relatively low pressure prevents that sensing fibre is
crushed.
4. After the five minutes a pressure of 9bar is applied. During an additional five
minutes the temperature of 316o C is maintained. Hereafter the temperature
is gradually, in one hour, decreased to 20o C. The melted matrix material
solidifies and binds the carbon fibres and fibre bragg gratings together.
5. The result is a laminated carbon-polyetherimide plate. Test specimens are
sawn out of the plate and used in the experimental validation programme.

1
In case sensing fibres were embedded a consolidation temperature of 300o C was used. High
temperatures can have a negative influence on signal strength.

97
 APPENDIX
C
SPECIMEN DIMENSIONS AND
DELAMINATION PARAMETERS

b h2 h

ac

Figure C.1: Specimen and delamination parameters.

Specimen Specimen size Del. parameters FBG

ID L (mm) w (mm) h (mm) ac b h2
U1.00.000 320 20.1 1.67 na. na. na. no
U1.06.025 320 20.1 1.66 .5 .6 .25 no
U1.06.050 320 19.9 1.67 .5 .6 .5 no
U2.000.00 455 20.1 2.31 na. na. na. yes
U2.052.02 455 20.1 2.32 .52 .2 .5 yes
U2.052.03 455 19.9 2.32 .52 .3 .5 yes
U2.032.02 455 20.0 2.32 .32 .2 .5 yes
U2.032.03 455 20.0 2.32 .32 .3 .5 yes
U2.grow 435 19.9 2.31 .32 .05 .5 no

Table C.1: Specimen dimensions and configuration.

99