You are on page 1of 272

ISSN 1018-5593

E U R O P E A N
C O M M I S S I O N

RESEARCH
DEVELOPMENT

Practical information and Programmes

Fatigue of
materials and
components for
wind turbine
rotor blades

Edited by
C. W. Kensche

A German
Sj Aerospace
\ f Research
V DLR Establishment

EUR 16684 EN
EUROPEAN COMMISSION

Fatigue of
materials and components
for wind turbine
rotor blades
Editor

C. W. Kensche

Authors

S. I. Andersen, Risø, P. W. Bach, ECN, W. J. A. Bonnee, NLR,


C. W. Kensche, DLR, H. Lilholt, Risø, Α. Lystrup, Risø, W. Sys, RUG

Directorate-General XII
Science, Research and Development

1996 EUR 16684 EN


Published by the
EUROPEAN COMMISSION
Directorate-General XIII
Telecommunications, Information Market and Exploitation of Research
L-2920 Luxembourg

LEGAL NOTICE
Neither the European Commission nor any person acting on behalf of
the Commission is responsible for the use which might be made of the
following information

Cataloguing data can be found at the end of this publication

Luxembourg: Office for Official Publications of the European Communities, 1996

ISBN 92-827-4361-6

© ECSC-EC-EAEC, Brussels · Luxembourg, 1996


Printed in Belgium
Contributing Institutions

DLR Deutsche Forschungsanstalt für Luft- und Raumfahrt e.V., Institute of


Structures and Design, Pfaffenwaldring 38-40, 70569 Stuttgart, Germany

ECN Energieonderzoek Centrum Nederland, Chemistry and Materials Department


P.O.Box 1, 1755 ZG Petten, The Netherlands

FFA') Flygtekniska Försöksanstalten, Sweden


P.O.Box 110 21, 161 11 Bromma, Sweden

NLR Nationaal Lucht- en Ruimtevaartlaboratorium


Postbus 90502, 1006 BM Amsterdam, The Netherlands '

RUG Rijksuniversiteit Gent, Laboratorium Soete vor Weerstand van Materialien en


Lastechniek, Sint-Pietersnieuwstraat, 41, 9000 Gent, Belgium

Risø Forskningscenter Risø, Materials Department


P.O.Box 49, 4000 Roskilde, Denmark

') Data from FFA are included in Section 4.

- in -
Contents

Foreword IX

Preface XI

1. Introduction 1

1.1 General 1
1.2 Fatigue and Cracks 3

1.3 The WISPER Load Sequence 4

2. Materials 10

2.1 Glass-Polyester Composites (GI-UP) 10


2.1.1 ECN-lnvestigations of GI-UP Materials 10
2.1.2 NLR-lnvestigations of GI-UP Materials ' 39
2.1.3 Risø-lnvestigations of GI-UP Materials 71
2.1.4 RUG-lnvestigations of GI-UP Materials 120
2.2 Glass-Epoxy Composites (Gl-Ep) 131
2.2.1 D LR-lnvestigations 131
2.3 Steel 158
2.3.1 ECN, Welded Steel D etails 158
3. Components 170

3.1 RUG, Root Parts of Glass-Polyester Rotor Blades 170


3.2 D LR, Spar Beams of Glass-Epoxy Rotor Blades 189

4. Statistical Evaluation and Comparison of Fatigue D ata 223

ν-
Abbreviations

CEC Commission of European Communities


DLR German Aerospace Research Establishment (Germany)
EC European Community
ECN Netherlands Energy Research Foundation (The Netherlands)
FFA Aeronautical Research Institute of Sweden (Sweden)
GFRP Glass Fibre Reinforces Plastics
Gl-Ep Glass-Epoxy
GI-UP Glass-Polyester
GRP Glass Fibre Reinforced Polyester
HAWT Horizontal Axis Wind Turbine
IEA International Energy Agency
ILSS Interlaminar Shear Strength
JOULE Joint Opportunities for Unconventional or Long-term Energy supply
LBF Laboratorium für Betriebsfestigkeit, Darmstadt
MLE Maximum Likelihood Estimator
NLR National Aerospace Laboratory (The Netherlands)
R Stress Ratio
Risø Risø National Laboratory (Denmark)
RT Room Temperature
RUG Rijksuniversiteit Gent (Belgium)
UD Uni-Directional
UP Unsaturated Polyester
UTS Ultimate Tensile Strength
WEC Wind Energy Converter
WISPER Wind SPECtrum Reference
WISPERX Short time derivation of WISPER

- VII
Foreword
This foreword can only be written on behalf of my good friend Ulrich Hütter, who unfortunately
passed away much too early. He was the pioneer of the utilization of wind energy and the
application of composite materials in his rotor blades. An impressive witness of this
development is a blade of his 34 m diameter wind turbine, built in 1957, which was placed in
front of his former institute. It undergoes there, however, much less load cycles than in the
turbine.

At that time very little was known about fatigue of glass fibre composites. It was just hoped
that the properties of composite materials would be at least as good as those of metals. The
propagation of micro cracks in the matrix was expected to stop at the fibres.

Since then composite materials have successfully been applied in many other areas of
engineering, not only in gliders and aircrafts where they allowed tremendous improvements
in structures and surfaces. Until now it is a fundamental problem to obtain reliable fatigue data
of glass fibre composites which are not too conservative. Wind turbines and airplanes become
just too heavy if conventional theories are applied to determine safe'design allowables from
different fatigue data sources.

The present book includes fatigue data from many research groups. This yields less
conservative design allowables for glass fibre composite structures. It is thus an important
milestone for the wind energy community. Moreover, also aircraft engineers and many users
of glass fibre composites will profit from the book.

Stuttgart, June 6, 1995 Richard Eppler

ix -
Preface

Glass fibre reinforced plastics (GFRP) have been used as a structural material in technical
fields since the fifties. In 1957 a 34 m diameter wind turbine, the W 34, was established under
the leadership of Ulrich Hütter. The rotor blades were for many years the world's largest
primary structures made of glass fibre composites. In the same year the Phoenix, the first
sailplane also with a GFRP structure, was developed by Richard Eppler together with Nägele
and Lindner.

Both designs caused a revolution in their fields of application. In the following glass fibre
composites were used for other new gliders causing already in the sixties the establishment
of relatively high design allowables necessary for achieving as light weight designs as
possible. For rotor blades, the need for more knowlegde about the limits of this material came
up later, i.e. in the eighties after the renaissance in wind energy.

Within a research program settled by the EU at this time, the long term fatigue behaviour of
the glass fibre composites was investigated by several research institutes. The aim was to get
basic information for possible design limits of new rotor blades. Up' to that time they were
relatively conservative compared to those in sailplanes, since for the application on a rotor no
extreme weight saving was needed.

This book collects and evaluates the results of the project achieved on the fatigue data of
GFRP in materials and components of rotor blades. The content of the Sections 1, 2 and 3
was taken from the individual final reports of each co-author and I would like to sincerely thank
them for their contributions, and also for doing their best to conform to my planning and time
table for publishing. Section 4 presents the data of Section 2 in a statistically evaluated
manner to make the data comparable and to give a basis for safe design allowables. I would
like to acknowledge here the Swedish FFA for providing additional data for use in the
comparison. Moreover, I would like to especially thank Wilfried Sys (RUG, Belgium) for his
help with the statistical evaluation in Section 4. Without these signifant contributions, this book
would not have appeared in its present form.

Although the data handling procedure caused an unforeseen delay in publishing the work, the
authors believe that the book nevertheless will contribute to an important increase of
knowledge of GFRP fatigue and thus will be useful for all people using glass fibre composites
in wind turbine rotor blades and also in other areas of technical application.

On behalf of my co-authors I would like to take this opportunity to thank Dr. Wolfgang Palz,
Head of Renewable Energies division in DG XII, for first supporting the research projects on
the fatigue of GFRP rotor blades at several European research institutes, and then for his
initiative to collect this information and publish it in book form so that it is of greater benefit
for the European industry. Finally, I want to acknowledge Mrs. Susan Giegerich for her
considerable help in proof reading and improving the final English text.

Christoph W. Kensche

XI
1. Introduction

1.1 General (Ch. W. Kensche)


Wind turbines nowadays are designed with rotor blades usually made of GFRP (glass fibre
reinforced plastics). These structures are safe life constructions, i.e. their full integrity is
guaranteed over the total foreseeable service life of the wind turbine which, in general, is 20
years. In this time period the rotor blades are expected to endure more than 10B load cycles.
Since no relevant fatigue data for the material used were available, the blades had to be
designed with relatively low design levels, e.g. a longitudinal strain of 0.3 per cent for the
tension, and -0.2 per cent for the compression area of a component. These conservative
allowables lead to an unnecessarily high structural mass of the rotor and, thus, also of the
drive train, nacelle, and tower.

The establishment of these values was useful particularly in the starting phase of series
production of wing blades and also with respect to small machines. However, the trend is to
increase the size of wind generators. In this case lightweight rotor blades are more favour­
able, since, for diameters larger than about 25 m, it has been shown that the blade mass
increases with a higher exponential factor than below that figure. Vice versa, the less the
mass of the blades, the more it is worthwhile to keep large machines in mind. Lighter blades
have the advantages of reduced costs and since they require less material natural resources
can be conserved.

However, for low mass products with a reliable fatigue life prediction, higher design allow­
ables are necessary than is state-of-the-art. Thus, data on long time fatigue properties of the
most commonly used rotor blade materials must be proven. For light aircrafts and sailplanes
made of GFRP certification allowables exist which are a factor 2-3 higher than the strain
limits mentioned above for rotor blades [1]. Due to some similarities in the designs of wings
and rotor blades, higher design levels are conceivable for the latter, too. In a material testing
program from 1986 to 1990 within a concerted action of several EC countries, investigations
therefore were carried out on typical wing blade materials and structures to improve know­
ledge about high cycle fatigue of specific compounds in small and large scale components.

This publication comprises the results of these tests. Since the participating institutes all have
their individually developed history and experience, the test methods used and described in
the individual sections differ more or less one from another. Each author is responsible for
the content of his contribution. Therefore, no summarizing explanations are made, in general,
for differences in the results of e.g. static strength and stiffness. However, for an objective
comparison, a statistical evaluation of the fatigue data is included in the last section. Thus,
the book contains useful information for design engineers, manufacturers, and those who
establish rules and guidelines for the design of rotor blades.

In this first international material and fatigue program for wind energy applications naturally
quite different interests had to be covered under the one subject, high cycle fatigue. To sat­
isfy these aims, constant amplitude fatigue tests and also variable amplitude tests were
carried out not only on small coupon specimens but also on large scale rotor blade compo­
nents. The participating countries and institutions were
• Belgium, Rijksuniversiteit Gent (RUG)
• Denmark, Risø National Laboratory, Roskilde
• Germany, German Aerospace Research Establishment, Stuttgart (DLR)
• The Netherlands, Netherlands Energy Research Foundation (ECN)
• The Netherlands, National Aerospace Laboratory (NLR)
Some national programs were incorporated in the concerted action of the EC. Open
exchange of experience and data was agreed. This demanded close cooperation between
the participants in terms of one or two meetings per year. Thus, knowledge of a material's
behaviour could increase more rapidly and extensively than in purely national projects.
The need for such a forum is underlined by the fact that the FFA (The Aeronautical Research
Institute of Sweden) ­ as a non­EC member at that time ­ joined the concerted action during
the project phase as an associated partner. Some results of an FFA program could be
included in this book (see chapter 4, Statistical Evaluation and Comparison of Fatigue
Data). Meanwhile the Swedish institute is fully integrated in the new Material Program of
JOULE.
In addition to the common aim of all participants in the concerted action, to get fatigue data
for GFRP up to 10s load cycles by means of coupon specimens, some welded steel com­
ponents were also tested as a material necessary for attaching the rotor blade to the hub.
There is a fundamental difference between the growth of cracks in metals and in composites.
A short review of the nature of both kinds of fatigue propagation is therefore given in the next
section of the introduction. The statements here, referring to glass­polyester (GI­UP), in
principle are valid for glass­epoxy (Gl­Ep), too.
Each laboratory focussed its interest on special items which are described in the following
sections. In general, the constant amplitude tests were carried out with different stress ratios
R which is defined by the quotient of the lower stress σ, by the upper stress ou:

The following stress ratios were used by all institutes:

R = 0.1 (tension­tension)
R = ­1 (tension­compression)
R = 10 (compression­compression)

The fatigue investigations were generally carried out at room temperature (RT). A dditionally,
material properties in hot and cold surrounding conditions were tested by RUG.
For the variable amplitude tests the IEA WISPER standard, a wind energy specific load
sequence developed especially for comparing materials, was used. A more detailed
description is given in Section 1.3 of the introduction. It also decribes the short version of the
sequence, WISPERX.
Glass­polyester was chosen by RUG, Risø, ECN and NLR, while glass­epoxy was of special
interest for DLR.
In most cases rotor blades will not be built with a pure 0° stacking sequence of the glass
layers. Due to their tapered geometry they show misalignments up to 10°. To be able to
transmit shear stresses, the shell of the blades is often designed with some layers of +45°
orientation. Thus, in the programs not only 0° fibre specimens were used but also ±10° and
±45°. For special purposes a few tests were also carried out on materials with ±30° and
±60° fibre directions.
To gain a better understanding of the overall behaviour not only of the material itself but also
of the GFRP structure and load attachment parts of a blade, real components of rotor blades
including the whole hub connection were also tested by RUG and DLR. Constant as well as
variable amplitude tests were carried out with the WISPER and WISPERX sequences. In
both the specimen tests and the spar beam and blade component tests, stiffness decrease
was measured to obtain possible early input about impending fatigue.

­ 2 ­
1.2 Fatigue and Cracks (H.Lilholt)
For fatigue loadings, it is the component lifetime which is relevant. Lifetime is often measured
as the number of cyclic loadings which the material (the component) can sustain before
"failure" or "nonfunction", which must be defined according to a certain criterion.
Under fatigue loading the material is normally degraded by plastic deformation and/or
cracking, both of which are irreversible processes. In principle therefore, a limit can be set
for a certain amount of damage in the material; such a limit can be used as a design limit for
a component made of the material. The practical design limit is often a certain residual stiff­
ness or a certain residual strength, and these are governed by the microstructural architec­
ture and the damage suffered by the material.
In steel and other metallic materials of nearly isotropic nature and moderate-to-large ductility,
plastic deformation will be the first irreversible process in the material, and it will lead to
cracks, of which one will normally become dominant and grow to a critical size for fast, final
fracture of the material and thus of the component.
In glass-polyester composite materials there is normally little or no plastic deformation, but
many small cracks will form in the polyester matrix and/or at the interface between glass
fibres and polyester matrix. When a sufficient number of cracks has'formed, some will grow
together into a larger group, which will lead to final failure due to the reduced load bearing
cross-sectional area of the material (component).
For both steel (and metallic alloys) and glass-polyester, the fatigue data (experimental
results) are presented in so-called S-N curves, which show the (maximum) load S, plotted
against the logarithm of the number of cyclic loadings N before failure. For both materials
there is normally a so-called fatigue limit at a large value of N, typically 106 to 107; this fatigue
limit means that loads (stresses) below the limit will not lead to failure, i.e. the lifetime is
infinitely long.
In steel, plastic deformation takes place by dislocation movements on the slip planes, and
this leads to so-called extrusions and intrusions (on a microscale); especially the intrusions
are the site for potential initial, small cracks, which then can grow under subsequent cyclic
loadings. Normally (only) one of these cracks will grow more and faster than the others, and
eventually reach a critical size, which starts the final, fast fracture.
Fracture mechanics relates the size, c, of the crack to the resulting failure stress, σ, via the
fracture toughness parameter K,c:

σ= **

Fracture mechanics analysis assumes that the crack has the same shape during growth;
normally the shape is taken to be ellipsoidal.
This analysis and idea describe (at least qualitatively) the decreasing section of the fatigue
S-N curve (Figure 1 on page 6). The fatigue limit is then the load (stress) below which there
is no (or very little) plastic deformation, so that no intrusions and initial microcracks can
develop. The fatigue limit is probably equal to or lower than the conventional yield stress.
In glass-polyester composite materials many cracks will nucleate and grow in the polyester
matrix; each crack will follow the principles of fracture mechanics because polyester is itself
a brittle material. The presence of fibres causes the matrix cracks to (temporarily) stop at the
fibres, because the interface can open up by debonding, as illustrated in Figure 2 on page
6. This process of debonding is critically dependent on the (shear) strength of the interface;
the ability to stop or "deviate" the (initial) crack is higher the weaker the interface strength.
This temporary stopping of the growing crack allows other (small) cracks to grow, and the

3 -
whole process therefore leads to many, moderately large cracks (rather than a single very
large crack of a final critical size).
This situation of many (noncritical) cracks is established partly because of the above process,
and partly because the fibre spacing directly limits the maximum available crack size. Final
failure is therefore reached when so many cracks have grouped that the remaining cross-
sectional area of the material cannot sustain the applied load.
These aspects describe the decreasing part of the S-N curve for (glass-polyester) composite
materials (Figure 3 on page 6). At the fatigue limit the load (stress) is so low that the initial,
small cracks cannot grow, or, at very low loads, do not form at all. The situation of non-
growing or nonexisting cracks corresponds to low or no plastic deformation in steel.
A further comparison between steel (metallic materials) and glass-polyester (composite
materials) in terms of cracks and their development under (fatigue) loading can be made with
reference to Figure 4 on page 7. In steel cracks grow with similar shape and thus are self-
accelerating; in glass-polyester cracks grow until stop/deviation at the fibre/matrix interface
and thus are self-decelerating (self-stopping).
Many composite materials, like glass-polyester, can have, and often do have, different vol­
ume fractions of fibres, and thus form a family of (similar) materials. It has been shown that
a plot of the strain, instead of the stress, in the S-N diagrams leads to a common curve for
the family of composite materials. Such an ε-Ν diagram is shown in Figure 5 on page 8 with
regions III, II and I referring to growing cracks, nongrowing cracks and no cracks, respec­
tively.
Example of S-N curves for some glass-polyester materials are shown in Figure 6 on page
8; this illustrates the stresses in fatigue and the large differences depending on fibre volume
fraction and fibre orientation.
The various aspects of fatigue in composite materials described above are relatively well
supported for glass-polyester, and from these design limits and design allowables can be
derived.

1.3 The WISPER Load Sequence (P. W. Bach)


Current design procedures are primarily based on constant amplitude data. However, this
simple kind of loading is usually insufficient to represent the interaction effects of high and
low cycles that occur in a realistic type of loading.
Especially in the aerospace field variable amplitude fatigue testing is widely used and stan­
dard loading sequences have been developed in order to predict material fatigue behaviour
under realistic conditions. The main applications of standardized load sequences are for
ranking materials, structural design alternatives, fabrication techniques, etc..
This was also realized by a group of European wind turbine manufacturers and research
institutes, which led to an international program (IEA working group) to establish a relevant
loading standard for HAWT blades [2].
The resulting standard load sequence for wind turbine blades called WISPER (Wind SPEc-
trum Reference) is based on flap load measurements of 9 different HAWT's made of different
blade materials (steel, glass fibre reinforced plastic, wood epoxy), with a wide range of
diameters (11.7 to 80.0 m) and at different locations. It consists of 265,423 loading reversal
points, i.e. 132,711 cycles. The WISPER sequence is a row of integers ranging from 1 to 64,
with zero load at level 25 and maximum load at level 64. In practice the WISPER levels are
multiplied by a certain factor in order to obtain the desired maximum load level.
The stress ratio R of the sequence can be expressed in two ways:

R = -0.6: the ratio based on the most extreme peak and the most extreme trough
R = +0.4: the average of the R-values of the individual cycles

An example of a part of the WISPER sequence is shown in Figure 7 on page 9. A variant


of the WISPER spectrum has been derived (WISPERX), in which the ranges below level 17
have been omitted. The number of cycles in this spectrum is reduced by almost a factor 10,
12,831 versus 132,711. If the damage developement caused by these spectra does not differ
too much, testing time can be significantly reduced by using WISPERX.
The WISPER spectrum represents a mean 2-month flapwise loading of an imaginary wind
turbine blade. As a result WISPER does not simulate sufficiently well the typical loading in
a blade of a particular turbine in a particular wind regime. This means that it can never be
used instead of a design spectrum.
The WISPER load sequence is intended for comparative purposes only, i.e. to evaluate
materials and structural details, dimensions, design alternatives, 'and lifetime prediction
methods.
A lifetime prediction can be based on the rainflowed WISPER or WISPERX 64x64 from-to
matrix. A damage 64x64 matrix can be obtained by dividing the number of cycles in each
matrix element by the number of cycles to failure calculated with the relevant S-N equation
or constant life diagram. The relative damage of one WISPER sequence is determined by
summation of the damage data of the elements.

1.4 Bibliography

111 Kensche, Ch.W. Fatigue of Composite Materials in Sailplanes and Rotor Blades XIX.OSTIV-
Congress, Rieti/ltaly, 2-10 August 1985, Publ.XVIII
[2] Ten Have, A.A.V. WISPER: A Standardized Fatigue Load Sequence for HAWT-Blades Proc.
EWEC, ed W. Palz, Pubi. Stephens & Ass, Bedford, England, p. 448, (1988)
1.5 Figures

Steel

log N
Figure 1. Fatigue curve for steel (schematic).

Figure 2. Sketch of crack, which temporarily stops at a fibre because of debonding at the
interface.

Glass/polyester
σ
A

Vf increases

^logN
Figure 3. Fatigue curves for glass-polyester, with different volume fractions of fibres (sche­
matic).
Figure 4. Cracks in steel (a) and in glass-polyester (b), at three successive time steps.
ε
A

Jog Ν
Figure 5. Master fatigue curve for glass­polyester (with different volume fractions), regions
I, II, and III correspond to no cracks, nongrowing cracks, and growing cracks,
respectively (see text for details).

MPa Δσ
' ' * *-'

400
\ Glass/polyester
\ ±5°
300

\0790°\
200
mat \^
100
. log Ν
0 1 1 1 ,. ¡ .,,, .,_.„...«.

0 8
Figure 6. Fatigue curves for glass­polyester with different fibre configurations, maximum
stress is used for the y­axis.
WISPER load sequence

20

10

Q
O 640 128Θ 1320 25G0 3200
nr of points

NISPER load sequence


60

50

40
iLi

iD 30

20

Θ
0 6400 12800 13200 25600 32000
nr of points
Figure 7. Example of a part of the WISPER load sequence.
2. Materials

2.1 Glass-Polyester Composites (GI-UP)

2.1.1 ECN Investigations of GI-UP Materials (P.W. Bach)

2.1.1.1 General
The most fatigue-critical part of a wind turbine is the rotor blade structure and its connection
to the hub. In a design life extending over more than 20 years it has to endure high cycle
( > 108) fatigue loading of variable amplitude due to gusts, wind shear, start and stop proce­
dures, etc. Very few data are available in this high cycle fatigue range for the materials and
substructures used in wind turbines. One of the materials that is extensively used in wind-
turbine blades is glass fibre reinforced polyester (GRP).
The design strength and stiffness of rotor blades have to be maintained during the lifetime
of a wind turbine. Stiffness is especially important because stiffness degradation of the
material has an effect on the aerodynamics and the eigenfrequency of the rotor blade. In
fatigue investigation special attention is paid to monitoring the stiffness degradation of the
specimens.

2.1.1.2 Materials and Methods

Unidirectional Glass Fibre Reinforced Polyester Coupons


For the fatigue investigation, coupon specimens were cut on-axis from a unidirectional rein­
forced glass fibre/polyester plate (5 mm thick) containing 6 layers of 700 g/m2 nonwoven
unidirectional and 100 g/m2 chopped glass fibre. The plate was produced under industrial
conditions by hand lay-up and with a glass-polyester ratio of - 35 vol. %.
Load introduction to unidirectionally reinforced composites is very critical due to the multi-
axial stress distribution and stress concentration in the grips. For smooth load introduction,
tabs of glass fibre/epoxy were bonded to the specimens and they were machined to a dog-
bone shape.
The specimens had the following dimensions: length 170 mm, test section length 80 mm,
width 25 mm, width at waist 21 mm, thickness 5 mm (see Figure 1 on page 23).
The radius of the waist was chosen in such a way that there is enough fatigue strength for
shear in the 0° layers. The interlaminar shear strength value for glass fibre reinforced epoxy
is about 30 M Pa at 106 cycles [4].
The ultimate tensile strength of the specimen was measured with a speed of 0.5 mm/min,
which gave a loading rate of about 300 N/sec. The mean strength of six specimens is 600
MPa. The tensile strength of glass fibre reinforced plastics, however, is dependent on the
loading rate [6]. The low loading rate used is supposed to give a conservative value for
tensile strength. With a loading rate of 300 kN/sec the measured tensile strength is 685 MPa.
The Ε-modulus of the material was calculated after measuring the strain with an extensom-
eter. The mean Ε-modulus of 12 specimens is 26.3 GPa, which is close to the calculated
value of 26.5 GPa.

Cross-plied Glass Fibre Reinforced Polyester Coupons


Glass fibre reinforced polyester plates about seven mm thick were produced under industrial
conditions by hand lay-up and with a glass-polyester ratio of 35 vol %. The plates contain

10
eight layers of 700 g/m2 nonwoven unidirectional glass fibre and 100 g/m2 chopped fibre in
a symmetrical (+45%45°) lay-up (the same material as for the unidirectionally reinforced
coupons). Specimens with a width of 45 mm and a length of 170 mm were cut from the
plates. For smooth load introduction, tabs of glass fibre/epoxy were bonded to the specimens
(see F igure 2 on page 24).
The Ε-modulus of the material was calculated after measuring of the strain with an exten-
someter. The mean Ε-modulus of six specimens is 11.2 GPa. The tensile strength was
measured with a loading rate of 1 kN/sec. The mean tensile strength of three specimens is
115 MPa.

Constant Amplitude Tests


The constant amplitude tests were performed on load controlled servo-hydraulic testing
machines (Instron 1342 and Instron 1343) with a tension-tension or a tension-compression
load (stress ratio R = amin/omax = 0.1 or -1). Due to the expected high cycle fatigue life, the
tests were performed with a frequency of 5-20 Hz. The temperature of the specimen was kept
below 25°C by a cool air flow.
After measuring initial stiffness, stiffness degradation was determined on-line during the test
at regular intervals (100-10.000 cycles). Stiffness was calculated as the delta load divided
by the delta elongation of the specimen as measured by the displacement transducer of the
actuator. The stiffness degradation of the specimens will influence load performance in the
closed-loop servo-hydraulic system. In order to have a constant load until failure, the load
was corrected on-line during the fatigue life of a specimen.

Loading of composites with stress ratio R = -1 is, relative to the magnitude of the stress, the
most severe constant amplitude test.
In addition to tension and compression, the specimens were also subjected to buckling. The
design of the specimen is such that it is stable to Euler buckling. But on a local scale
microbuckling of the fibres can occur due to misalignment of the 0° reinforcement or small
voids in the polyester. Although recommended by Schütz, antibuckling guides were not
applied because they will also restrain the development of microbuckling damage and cause
extra heating of the specimens [10].

Variable Amplitude Tests


The variable amplitude tests were performed on load controlled servo-hydraulic testing
machines (Instron 1342 and Instron 1343) equipped with the Harrier computer interface and
a Hewlett-Packard 9816 microcomputer. A special sofware programme was developed in
order to control the fatigue machines in the WISPER sequence tests. After measuring initial
stiffness, stiffness degradation was determined on-line at regular intervals.
Variable amplitude testing is possible in two modes and with two wave shapes.
1. With constant frequency. Only the amplitudes are variable.
2. With a constant load application rate. F or each individual load cycle the frequency must
be calculated in order to obtain a constant loading rate. In this mode both the frequency
and amplitude are variable.
3. Tests can be performed with a sawtooth or a sine as the wave shape.
The majority of the WISPER fatigue tests were performed with a constant load application
rate and a sinusoidal wave shape.

11
2.1.1.3 Results
Constant Amplitude Tests
Unidirectionally glass fibre reinforced polyester coupons
In Figure 3 on page 25 the typical stiffness degradation S/S0 of a specimen loaded with a
stress ratio of 0.1 is shown together with the test parameters. The fatigue damage observed
in the specimens proceeds gradually and is consistent with the measured stiffness degra­
dation. In an early stage of fatigue life (10-30%) damage starts with cracks in the polyester
perpendicular to the load. This causes a very small reduction in stiffness, because the con­
tribution of the polyester to the initial stiffness is only about 10%. In a later stage (30-70%)
interfacial shear damage occurs and the first delaminations appear. Due to the multiaxial
stress state and stress concentration near the load introduction, the delaminations usually
originate from the grips. The delaminations expand in the last stage across the whole spec­
imen until final separation occurs.

In Figure 4 on page 25 a typical stiffness degradation plot of a test with a stress ratio R =
-1 is depicted. The stiffness degradation in the first 10-20% of fatigue life is quite strong and
is probably due to microbuckling effects. After this initial stage interfacial shear damage
occurs and small delaminations start to grow and expand in the last stage across the speci­
men.
In order to obtain fatigue endurance curves (ε-Ν or σ-Ν), tests up to the high cycle range
(10 7 -10 8 cycles) were conducted. In Table 1 on page 19 the results of the tests are listed
for the stress ratios R = 0.1 and R = - 1 . In Figure 5 on page 26 the number of cycles until
specimen fracture versus the maximum strain level is shown for R = 0.1 and R = - 1 . Some
specimens did not fail during the planned fatigue test. Such a so-called "runout" is indicated
with an arrow. The results of the constant amplitude fatigue investigation can be compared
with the empirical relation for the fatigue strength Sn. of glass fibre reinforced plastics found
by Mandell [9].
Sn = UTS (1 - D log n) (3.1)

• UTS = ultimate tensile strength


• D = ' f ( R , material)-0.1
• η = number of cycles.
The reduction in fatigue strength Sn is about 10% per decade of cycles. The fatigue lifetimes
measured in the present investigation are higher than those calculated from Equation 3.1.
This could be expected, because relation 3.1 is based mostly on fatigue data with n<10 6
and is somewhat conservative for a high number of cycles, as it neglects a possible fatigue
limit. Appel and Olthoff [1] have modified the degradation factor D after analyzing data from
literature:
D=0.015A + 0.08 (3.2)
with A = 1 - R and R defined here as:
R = signSmin | S m i n | /signSmax \ Smax | (3.3)

with |S m J the absolute minimum stress or strain and |S max | the absolute maximum stress
or strain.
These modified ε-Ν curves are also shown in Figure 5 on page 26. The Appel/Olthoff mod­
ification predicts fatigue lifetimes which are safe for R = 0.1 but are optimistic for the higher
load levels of the R = -1 tests.

12
The investigated nonwoven glass fibre reinforcement is, according to Mandell [8], more
fatigue resistant than woven material in the higher stress range. The fatigue of laminates of
woven material is affected by local friction and microdelamination near the crossover point.
Only at low stress levels near a possible fatigue limit will the fatigue life of woven material
approximate the fatigue performance of laminates with nonwoven reinforcement.

Failure criteria
The choice of a failure criterion, other than fracture is a rather arbitrary matter. Fatigue
damage develops rather gradually and can be represented by stiffness degradation. Stiffness
degradation is an important parameter for the design of wind turbine rotor blades because
a certain stiffness has to be maintained in order to retain the aerodynamic profile of the blade
and to keep the deflection and the eigenfrequency within the design envelope. In a survey
of the long-term durability of fibre glass reinforced plastic structures, Lieblein [7] found that
static strength can be retained for a considerable period of time under real service conditions.
Structures which are continuously exposed to ultraviolet radiation and weathering require
appropriate surface protection. The stiffness modulus, however, decreases about 10% over
some 20 years. Such a 10% reduction in stiffness can be used as a failure criterion for rotor
blades with a design life of approximately 20 years. In Figure 6 on page 27 the number of
cycles until 10% stiffness reduction is shown. The fatigue data can be analysed by Weibull
statistics. This is shown in Appendix A.

At the lowest applied stress level (100 MPa, R = -1) no fatigue damage is observed at
n - 5.10 7 . The slope of the ε-Ν or σ-Ν curves decreases with increasing n. Both facts indicate
that a fatigue limit will exist only in the very high cycle range (~ 109 cycles) and at a rather
low load (<0.15 UTS).

Cross-plied glass fibre reinforced polyester coupons


Constant amplitude tests with R = 0.1 and R = -1 are performed on load controlled servo-
hydraulic testing machines. In Figure 7 on page 28 and Figure 8 on page 28 typical stiffness
degradation curves for the specimens at R = 0.1 and R = -1 are shown. The stiffness
degradation of the ± 45° reinforced specimens is very strong. In the diagonally reinforced
material stiffness reduction can be as high as 40-60% before fracture. Especially in the tests
with R = -1 at the higher stress levels, stiffness can be progressively reduced during fatigue
loading. Damage develops much more rapidly in the 45° angle plied specimens than in the
unidirectionally reinforced material. The observed damage mechanism increases interfacial
shear failure (glass-polyester) in the final stage, followed by delamination and fracture. In
Table 2 on page 20 the results of the tests are listed for stress ratios R = 0.1 and R = - 1 .

The ε-Ν fatigue diagram determined for the 45° angle plied specimens is shown in Figure 8
on page 28 for R = 0.1 and R = -1 and for the failure criteria: fracture and 10% stiffness
reduction. Due to the initial steep stiffness degradation at the stress ratio R = 0.1 there is a
large difference in fatigue life for the two criteria.
The observed damage mechanisms for the unidirectionally and cross-plied glass fibre rein­
forced polyester specimens are consistent with the conceptual framework for the interpreta­
tion of fatigue damage by Talreja [13]. In this model the fatigue limit for unidirectionally
reinforced composites (on-axis) is correlated with the fatigue limit of the matrix materials. For
symmetric angle plied laminates the fatigue limit is determined by the (glass-matrix) interfa­
cial behaviour, which depends on fibre orientation.
For orientations < 30°, fatigue limit is governed by matrix properties. At orientations > 60°,
transverse fibre debonding is the dominant mechanism with a lower fatigue limit (= 0 . 1 %
strain).

- 13
At intermediate orientations a transition occurs from mixed mode matrix and interfacial
damage to transverse fibre debonding.
It is expected that the fatigue limit for unidirectionally reinforced polyester is approximately
0.3% strain for the stress ratio R = 0.1 and about 0.2% strain for R = - 1 . For the 45° angle
plied GRP these fatigue limits are expected to be situated at a 0.2% strain level for R = 0.1
and about 0.1% for R = -1 [3].

Two Stress Levels Fatigue Performance


An experimental programme was performed to investigate the effect of preloading (high-low
step tests) on the constant amplitude fatigue performance of the glass fibre reinforced
polyester coupons. This programme is a preparation for the fatigue tests which will be carried
out with a reference load sequence.
The cumulative fatigue damage of unnotched composite laminates is usuaily characterized
by continuous degradation of residual strength up to the level of the actual stress applied.
When a specimen is subjected to low-high or high-low sequences, fatigue failure will occur
at different levels of residual strength. Another aspect contributing to the load sequence effect
is the memory of the materials with respect to the previously experienced load sequences,
for instance due to the visco-elastic behaviour of the composite.

The effect of two stress levels on fatigue life was investigated with the same waisted coupon
specimens and the same test procedure as used in the constant amplitude tests. High-low
step tests were carried out with the stress ratios R = 0.1 and R = - 1 . The specimens were
preloaded (100-10,000 cycles) with a preload factor of 1.3 to 2 which, roughly, simulates the
contribution of high wind speeds and the start and stop procedures to the fatigue load
spectrum of a wind turbine.
The stiffness degradation of a coupon specimen during two stress levels (high-low) of con­
stant amplitude fatigue is shown in Figure 10 on page 30 for R = 0.1 and in Figure 11 on
page 31 for R = - 1 . During preloading the stiffness remains almost constant with a low
number of preloading cycles, but there can be stiffness reduction up to 10% for a larger
number of cycles, depending on the fraction of the fatigue life n/N¡ ( n¡ is the number of cycles
applied at a stress level o¡ and N¡ is the number of cycles to failure at this stress level).
Both for R = 0.1 and R = -1 stiffness degradation at the second stress level is similar to the
degradation observed in the standard constant amplitude tests.
In Figure 12 on page 32 the results of the two stress level tests are summarized, with the
number of cycles to fracture at the second stress level as failure criterion. The results are
also listed in Table 4 on page 21 for R = 0.1 and R = - 1 . For comparison the results of the
standard constant amplitude fatigue tests are depicted with horizontal lines. In Table 5 on
page 22 the mean damage sum according to Miner is given.

The results of the high-low step tests do not show, contrary to expectation, the detrimental
effect on constant amplitude fatigue life (horizontal line in Figure 12 on page 32 ) at stress
ratio R = 0.1. The results are even located at the high cycle side of the constant amplitude
fatigue life distribution. The cause for this effect is not clear but might be due to a proof
testing influence, the dispersion of load stress concentrations on microcracks, and of the
setting of the specimens during preloading. In fact, it is often observed that the stiffness of
composite specimens increases in the first few cycles.
At stress ratio R = -1 fatigue life is hardly influenced in the lower stress range by high pre­
loading. However, in the higher stress range fatigue life is distinctly reduced, probably due
to an additional damage mechanism (microbuckling) [13], [2].

14
The extension of fatigue life due to precycling has also been observed by others [5],[14]. It
should be noted that the above discussion can by no means be conclusive, as the evidence
in literature is incidental and often contradictory.
For metallic materials many cumulative damage theories have been developed which, except
for Miner's linear damage theory ( Ση/Ν, = 1), are based on a crystalline multigrain micro
structure and thus are not applicable to the fatigue behaviour of glass fibre reinforced plas­
tics.
The fatigue behaviour of fibre reinforced plastics is different from that of metallic materials
where most of the time is spent on developing the first crack. In the case of fibre reinforced
plastic laminates a first crack can already be formed in the first load cycle. The crack, how­
ever, is confined to the ply in which it is initiated for most of the fatigue life. A reduction in
stiffness and in strength will only be measured after the formation of many, usually dispersed,
small cracks.

Because of the statistical nature of the fatigue behaviour of composite materials, research
on cumulative damage models has been directed in recent years to a statistical approach
based on the Miner theory [16].
The presently observed increase in fatigue life due to precycling, however, cannot be
explained by the (modified) Miner's rule or other related damage accumulation theories. In
order to obtain a better understanding of the long-term mechanical behaviour of glass fibre
reinforced polyester, this material has to be investigated in more detail with respect to
stress-strain history effects.

Variable Amplitude Tests


Load application
During the preparation of the variable amplitude tests there was some uncertainty about the
effect of load application rate and wave shape on the lifetime of the specimens. It is known
that for glass fibre reinforced polyester the load application rate has an effect on (fatigue)
strength due to the visco-elastic behaviour of the material.

The constant amplitude tests were performed with a constant frequency which was adapted
to the maximum load level in order to obtain rather comparable heat generation in tests at
different load levels.
If the WISPER test is performed with a constant frequency, the loading rate will vary by the
large difference in load intervals of the individual cycles. Performing the tests with a constant
loading rate will result in a variable frequency.
Also, the shape of the cycle can affect fatigue performance; a sawtooth shape will give a real
constant loading rate; a sinusoidal shape is more realistic for the stress cycles in a turbine
blade.
In order to check the influence of variables in load application on the fatigue lifetime of the
specimen, a series of tests was performed with the WISPER load sequence. In Figure 13
on page 33 through Figure 15 on page 35 the results of the tests with the conditions constant
frequency/sine, constant loading rate/sine and constant loading rate/sawtooth are shown.
Within the usual scatter of fatigue tests there is no difference in fatigue life. The remaining
WISPER fatigue tests were thus performed with a constant loading rate and a sinusoidal
wave shape.

15
WISPER tests on unidirectionally glass fibre reinforced polyester coupons
In order to determine the fatigue lifetime of wind turbine materials with a realistic load
sequence, tests were performed with the WISPER load spectrum. In Figure 16 on page 36
the stiffness reduction of a glass fibre reinforced polyester coupon is shown. This type of
stiffness reduction is comparable to the stiffness reduction in constant amplitude tests with
a stress ratio R = - 1 . The relatively small number of through zero cycles in the WISPER
sequence is obviously responsible for the stiffness reduction in the first 10% of fatigue life.
This is probably caused by damage initiation due to microbuckling during compression of the
specimen.

A number of fatigue tests with several maximum load levels was performed. The data and
the results of the fatigue test with the WISPER sequence are listed in Table 3 on page 20 .
In Figure 16 on page 36 the results of the WISPER fatigue tests are shown in an ε-Ν diagram
with the maximum strain of the WISPER sequence as the ordinate.
There are no norms or standard procedures for evaluating load sequence fatigue tests of
glass fibre reinforced plastics. For steel a Palmgren-Miner summation based on the stress
range works quite well, but for glass fibre reinforced plastics large deviations between pre­
dicted and measured fatigue life can be expected for the higher load levels with this method
[11]. One of the problems is that the once occurring largest range is responsible for about
60% of the calculated damage. A practical way of coping with the WISPER results, which is
used by several manufacturers [12], is a lifetime prediction with a linear damage accumu­
lation calculation (Palmgren-Miner summation) based on a modified Mandell type of ε-Ν
curve.

For this ε-Ν curve the Appel/Olthoff equation (3.2) can be used. For calculating the damage
of a WISPER sequence a rainflowed WISPER 64x64 from-to matrix is used. This matrix is
connected with the stress (level 64 is maximum, level 25 is zero) in a fatigue test, and the
R-ratio of each matrix element is calculated. A damage 64x64 matrix is determined by cal­
culating the number of cycles to failure for each matrix element with the Appel/Olthoff
equation and applying the Palmgren-Miner rule.

In Figure 16 on page 36 these predicted fatigue lifetimes are indicated with line A/O. These
predicted lifetimes are rather close to the experimental lifetimes where they are conservative
for the higher stress levels and become more optimistic for lower maximum stresses.
From Figure 5 on page 26 it is obvious that the Appel/Olthoff equation is conservative for
R = 0.1 and optimistic for R = - 1 . In WISPER the majority of the cycle has an R-ratio > 0 so
a safe prediction can be expected.
If the Mandell equation is adapted to the experimental constant amplitude lifetimes in
Figure 5 on page 26, the degradation factor D becomes
D = 0.028A + 0.062. (3.4)
The lifetime prediction of WISPER fatigue tests with this ε-Ν curve goes through the cloud
of experimental points (line B/D) in Figure 16 on page 36 .This means that the lifetimes in
running WISPER tests can be predicted within the usual scatter of fatigue tests by this
method. In this prediction linear damage is assumed. The results show that spectrum loading
is not more detrimental than constant amplitude loading, so loading sequence effects prob­
ably did not play an important role in fatiguing glass fibre reinforced polyester.

- 16
2.1.1.4 Conclusions
From constant amplitude fatigue tests on glass fibre reinforced polyester it is concluded that
a fatigue limit exists only at a low load (<0.15 UTS) and a large number of load cycles (>
109).
The results of the high-low step tests do not show, contrary to expectation, a detrimental
effect on constant amplitude fatigue life at R = 0.1. However, at R = -1 constant amplitude
fatigue life is reduced in the higher stress range, probably due to an additional damage
mechanism (microbuckling).
Predicting lifetimes using variable amplitude fatigue tests and the WISPER spectrum on glass
fibre reinforced polyester specimens is possible within the usual scatter by a linear damage
accumulation calculation based on a modified Mandell type of ε-Ν curve.

2.1.1.5 Bibliography
[11 Appel, Ν. en Olthoff, J. Voorontwerpstudie NEWECS-45 Polymarin rapport (1988)
[2] Bach, P.W. High Cycle Fatigue Investigation into Wind Turbine Materials Proc. EWEC, ed. W.
Palz, Stevens & Ass., Bedford, England, p. 337, (1988)
13] Bach, P.W. High Cycle Fatigue of Glass Fibre Reinforced Polyester Proc. IEA Workshop on
Fatigue in Wind Turbines, ed. K.F. McAnulty, ESTU-N-113, (1989)
[4] Beardmore, P. Fatigue Response of Struc tural Composites Proc. 2nd Int. Conf. on F atigue
Thresholds, ed. Beevers, C.J., Brimingham, p. 1091, (1984)
[5] Boiler, K.H. Effec t of Pre-cyclic Stresses on Fatigue Life of RP Laminates Modern Plastics, p.
1962, (1965)
[6] Jones, C.J. Environmental Fatigue of Reinforced Plastic Composites 14, nr. 3, p. 288, (1983)
[7] Lieblein, S. Survey of Long Term Durability of Fiberglass Reinforc ed Plastic Structures Report
DOE/NASA/90599-1, NASA CR-165320 TRS 106, (1981)
[8] Mandell, J.F. Composite Reliability STP 580, ASTM, p. 515, (1975)
[9] Mandell, J.F. Development in Reinforced Plastics-2 ed. O. Pritchard, Appi. Science Pubi., p. 67,
(1982)
[10] Metondang, T.H., Schütz, D. The Influence of Anti-buckling Guides on the Compression Fatigue
Behaviour of Carbon Fibre Reinforced Lamiates LFB-report, (1981)
[ 111 Poppen, M. Fatigue Testing of Glass Fibre Reinforced Epoxy using the WISPER Sequence FFA
TN 1989-45, (1989)
|12| Roorda, B. Analyse variabele amplitude proeven met WISPER belastingsspectrum aan pen/gat
proefstukken Polymarin rapport R&D-900212, (1990)
[13] Talreja, R. Fatigue of Composite Materials Technomic Pubi., (1987)
[14] Tanimoto, T. et al Fatigue life and reliability of FRP under muitistep loading Composite Materials,
ed. Kawata, K. and Akasaka, C, Proc. Japan, US Conf., Tokyo, (1981)
[15] Ten Have, A.A.V. WISPER: A Standardized Fatigue Load Sequence for HAWT-blades Proc.
EWEC, ed. W. Palz, Pubi. Stephens & Ass, Bedford, England, p. 448, (1988)
[16] Yang, J.N. and Jones, D.L. Load Sequence Effec ts on the Fatigue of Unnotc hed Composite
Materials Fatigue of Fibrous Composite Materials, ASTM-STP 723, p. 213, (1981)

17
2.1.1.6 Appendix A: Weibull Analysis of Fatigue Data
Both the fatigue data for fracture and for the 10% stiffness reduction can be analysed sta­
tistically by means of a Weibull distribution. The twoparameter Weibull cumulative distribution
function for a fraction F of failed specimens in fatigue is:

[ -%J]
yn
(ai)
N = number of cycles
N0 = characteristic number of cycles
m = Weibull modulus
The Weibull modulus can be calculated from the data of a number of specimens and is a
measure for the scatter of the data. In Figure 18 on page 38 the data of the fatigue tests
with R = 0.1 are analysed in a Weibull plot. From the Weibull plot the number of cycles can
be determined with 95% probability of survival (= 5% fraction failed) for fatigue at the partic­
ular stress level.
A summary of the fatigue and tensile data of the unidirectional coupon specimens can be
depicted in a R-N fatigue curve. For the σ-Ν curve (see Figure 19 on page 38 ) several cri­
teria derived from the Weibull statistical analysis are used:
• fracture
characteristic number of cycles
number of cycles at 5% fraction failed
• stiffness reduction
characteristic number of cycles at 10% stiffness reduction
5% fraction with less than 10% stiffness reduction

18
2.1.1.7 Tables

Code I Fiai Sigia Eps Freq C yc 101 sr.cyc C ooent


kH HPa io"3 Hi

1UPP ­1.0 10.50 100.00 3.8 20 28900000 run out


Π0ΡΡ ­1.0 10.50 100.00 3.3 20 47000000 run out
108PP ­1.0 10.50 100.00 3.8 20 54700000 run out
152PP ­1.0 13.12 125.00 4.7 15 10600000 1600000
135PP ­1.0 13.12 125.00 4.7 20 11700000 1000000
181PP ­1.0 13.12 125.00 4.7 5 15800000 5000000
148PP ­1.0 13.12 125.00 4.7 20 18300000 1200000
140PP ­1.0 13.12 125.00 4.7 15 23600000 2800000
124PP ­1.0 13.12 125.00 4.7 15 45000000 5000000
131PP ­1.0 15.75 150.00 5.7 17 1050000 120000
120PP ­1.0 15.75 150.00 5.7 15 1080000
101PP ­1.0 15.75 150.00 5.7 17 1310000 2000000
128PP ­1.0 15.75 150.00 5.7 15 1510000
126PP ­1.0 15.75 150.00 5.7 15 1630000
129PP ­1.0 15.75 150.00 5.7 17 1750000 60000
127PP ­1.0 15.75 150.00 5.7 15 2540000
100PP ­1.0 15.75 150.00 5.7 17 5600000
95PP ­1.0 15.75 150.00 5.7 17 6000000
158PP ­1.0 18.90 180.00 6.3 10 52013 10000
139PP ­1.0 18.90 180.00 6.8 10 273000 14000
137PP ­1.0 18.90 180.00 6.3 10 420000 25000
141PP ­1.0 18.90 180.00 6.8 10 660000 30000
168PP ­1.0 22.10 210.00 8.0 5 50000 4000
169PP ­1.0 22.10 210.00 3.0 5 76000 20000
171PP ­1.0 22.10 210.00 3.0 5 77000 10000
151PP ­1.0 22.10 210.00 3.0 2 120000 30000
156PP ­1.0 22.10 210.00 3.0 2 283201 50000
153PP ­1.0 22.10 210.00 8.0 2 690413 60000
160PP ­1.0 22.10 210.00 3.0 2 1165000 300000
90PP 0.1 22.10 210.00 3.0 17 24270000 20000000
38PP o.i 22.10 210.00 8.0 20 31950000 16000000
91PP 0.1 22.10 210.00 8.0 20 33950000 18000000
87PP 0.1 22.10 210.00 8.0 17 47310000 31000000
92PP 0.1 22.10 210.00 8.0 17 50040000 20000000
'¡m 0.1 22.10 210.00 8.0 20 52320000 21000000
79PP 0.1 26.25 250.00 9.5 17 9390000 5000000
83PP 0.1 26.25 250.00 9.5 17 10200000
93PP 0.1 26.25 250.00 9.5 17 10340000 7700000
84PP 0.1 26.25 250.00 9.5 20 11220000
82PP 0.1 26.25 250.00 9.5 17 11500000 7000000
31PP 0.1 26.25 250.00 9.5 20 14200000
80PP 0.1 26.25 250.00 9.5 17 16700000 7500000
119PP 0.1 31.50 300.00 11.4 15 440000 350000
U6PP 0.1 31.50 300.00 11.4 15 460000
117PP 0.1 31.50 300.00 11.4 15 310000 550000
122PP 0.1 31.50 300.00 11.4 15 370000 600000
123PP 0.1 31.50 300.00 11.4 15 1380000 700000
118PP 0.1 31.50 300.00 11.4 15 2020000 800000
103PP 0.1 31.50 300.00 11.4 15 3090000 1400000
102PP 0.1 31.50 300.00 11.4 15 3790000 1600000
145PP 0.1 44.20 420.00 16.0 1 2001
143PP 0.1 44.20 420.00 16.0 5 2002
175PP 0.1 44.20 420.00 16.0 5 5500
144PP 0.1 44.20 420.00 16.0 2 38009
134PP 0.1 44.20 422.00 16.0 1 101000

Table 1. Results of fatigue tests with the unidirectionally reinforced specimens.

- 19
Code R Fnax Signa Eps Freq Cyc 10i sr.cyc Coment
kN MPa IO"3 Hz
19PSL -1.0 7.50 23.80 2.3 10 6500000 1400000 run out
6PSL -1.0 7.50 23.80 2.3 10 20000000 run out
llPSL -1.0 10.00 31.50 3.0 10 108000 50000
7PSL -1.0 10.00 31.50 3.0 10 176000 35000
8PSL -1.0 10.00 31.50 3.0 10 260000 60000
10PSL -1.0 10.00 31.50 3.0 10 300000 80000
12PSL -1.0 11.75 37.30 3.5 8 26100 9000
13PSL -1.0 11.75 37.30 3.5 8 26500 11000
15PSL -1.0 18.00 57.10 5.3 8 1100 800
20PSL 0.1 5.00 15.90 1.6 20 50000000 run out
18PSL 0.1 12.00 38.10 3.6 15 10000000 1800000 run out
9PSL 0.1 12.00 38.10 3.6 15 33000000 3000000 run out
2PSL 0.1 15.00 47.60 4.5 10 500013 35000
1PSL 0.1 15.00 47.60 4.5 10 869002 45000
14PSL 0.1 15.00 47.60 4.5 10 992000 60000
3PSL 0.1 15.00 47.60 4.5 10 2430000
17PSL 0.1 18.00 57.10 5.3 8 44000 3500
16PSL 0.1 18.00 57.10 5.3 8 65300 4500

Table 2. Results of fatigue tests with the cross-plied reinforced specimens.

Code R Fmax Signa Eps Freq Cyc 10? sr.cyc Coment


kN MPa 10" 3 Hz
A1SL 0.1 18.00 35 1160000
A13SL 0.1 30.00 167.00 35 46160000 run out
A20SL 0.1 38.00 210.00 25 140083
A21SL 0.1 38.00 211.00 25 590025
A14SL 0.1 38.00 211.00 35 19360000 run out
A19SL 0.1 45.00 250.00 30 60048
A18SL 0.1 45.00 250.00 30 140084
A16SL 0.1 45.00 250.00 35 310000
A15SL 0.1 45.00 250.00 35 480000
A17SL 0.1 45.00 250.00 30 1020000

Table 3. Results of WISPER tests with the unidirectionally reinforced specimens.

20 -
Code R Fnax Signa Eps Freq Cyc 101 sr.cyc Coment
kN MPa io" 3 Hz
HLL195PP -1.0 13.12 125.00 4.7 20 4600000 400000 run out
HLL198PP -1.0 13.12 125.00 4.7 20 8700000 1000000 run out
HLL193PP -1.0 13.12 125.00 4.7 20 10600000 1000000 run out
HLL197PP -1.0 13.12 125.00 4.7 20 15400000 3500000 run out
HLL196PP -1.0 13.12 125.00 4.7 20 20800000 2000000
HLL194PP -1.0 13.12 125.00 4.7 15 34800000 2000000 run out
HLL189PP -1.0 15.75 150.00 5.7 17 23600 2000
HLL191PP -1.0 15.75 150.00 5.7 17 45500 3000
BLL190PP -1.0 15.75 150.00 5.7 17 52600 2000
HLL203PP -1.0 15.75 150.00 5.7 17 65200 3000 run out
HLL192PP -1.0 15.75 150.00 5.7 17 146000 10000
HLL206PP -1.0 15.75 150.00 5.7 15 200000 30000 run out
HLL204PP -1.0 15.75 150.00 5.7 15 554000 20000 run out
HLH206PP -1.0 22.10 210.00 8.0 1 1000
HLH204PP -1.0 22.10 210.00 8.0 1 1000
HLH203PP -1.0 22.10 211.00 8.0 5 1000
HLB194PP -1.0 22.10 210.00 8.0 5 1000
HLH193PP -1.0 22.10 210.00 8.0 5 1000
HLH192PP -1.0 22.10 210.00 3.0 5 1000
HLH198PP -1.0 22.10 210.00 8.0 5 10000 4000
BLH197PP -1.0 22.10 210.00 8.0 5 10000 7000
HLH196PP -1.0 22.10 210.00 8.0 5 10000 7000
HLH195PP -1.0 22.10 210.00 8.0 5 10000 2500
HLH191PP -1.0 22.10 210.00 8.0 5 10000 3000
HLH190PP -1.0 22.10 210.00 8.0 5 10000 3000
HLH189PP -1.0 22.10 210.00 8.0 5 10000 2000
HLL199PP 0.1 22.10 210.00 8.0 20 34200000 33000000
HLL201PP 0.1 22.10 210.00 8.0 20 67100000 45000000
HLL164PP 0.1 22.10 210.00 3.0 20 81790000
HLL184PP 0.1 26.25 250.00 9.5 17 13087000 10000000
HLL185PP 0.1 26.25 250.00 9.5 20 13430000 2000000
HLL188PP 0.1 26.25 250.00 9.5 20 18300000 2000000
HLL186PP 0.1 26.25 250.00 9.5 20 23340000 10000000
HLL187PP 0.1 26.25 250.00 9.5 20 33300000 6000000
HLH164PP 0.1 31.50 300.00 11.4 15 99848 4000
HLH185PP 0.1 44.20 420.00 16.0 1 997
HLH201PP 0.1 44.20 420.00 16.0 1 1000
HLH186PP 0.1 44.20 420.00 16.0 1 1000
HLH199PP 0.1 44.20 420.00 16.0 1 1000
HLH188PP 0.1 44.20 420.00 16.0 1 1000
HLH187PP 0.1 44.20 420.00 1 1000
HLH184PP 0.1 52.50 500.00 19.0 0 1
LHL146PP 0.1 22.10 20 100000
LHL159PP 0.1 22.10 210.00 20 100000
LHL182PP 0.1 22.10 20 100000
LHL183PP 0.1 22.10 20 100000
LHH146PP 0.1 31.50 15 130015
LHH183PP 0.1 31.50 15 2540000
LHH159PP 0.1 31.50 289.00 15 3370000
LHH182PP 0.1 31.50 15 4190000

Table 4. Results of the two stress level fatigue tests.

21 -
'
R S1/S2 [MPa] e2[f] d total

low-high 0.1 210(lE5)/300 1.11 1.5

high-low 0.1 iJ20(lE3)/250 0.95 1.6


high-low 0.1 ¿Ì20(1E3)/210 0.80 1.3
high-low 0.1 300(1E5)/210 0.80 1.6
high-low -1 210(1E4)/150 0.57 0.2
high-low -1 210(1E4)/125 0.47 1.5

Table 5. The mean damage sum (dlolal) according to Miner of the low-high and the high-low
fatigue tests.

- 22
2.1.1.8 Figures

1.5 „..,,.1.5

RiOO_

25

ΑΛΝ- |
T
*L I MATERIAAL EN/OF TEK. NR. CODE NUMMER
VORM EN PLAATSTOLERANTIES 25 - ί 2 25.0 - ± 0.2 25.00 - ± 0.02 RUWHEID VLGS.
VLGS. NEN 3311 25 - 1. 1 25.Õ ­ ± 0.1 2S.0Õ ­ i. 0.01 NEN 6301 · 3632

T.B.
DATUM
8-12.07
25 ­ ± 0.5

TITEL:
25.0 ­ ± 0.05 25.00 ­ ± 0,005
­ 25 ­ : SYMM. t.o.v. HA RTLIJN Φ<3 3634 - 3638 ■ 3639

HFD.TEK.K.
OPDR. GEVER WIJZ. DA T. PA R

ONDERWERP:

ENERGIEONDERZOEK CENTRUM NEDERLA ND


AUTEURSRECHT VOORBEHOUDEN BIJ DE WET OPDRACHT NR. FORMAAT A «

Figure 1. Unidirectional glass fibre reinforced polyester coupon specimen.

23 ­
iL* )Λ5

45 10
-> -f-

MATERIAAL EN/OF TEK. NR. CODE NUMMER


VORM EN PLAATSTOLERANTIES 25 ­ t 2 25.0 ­ i 0.2 25.00 ­ i 0.02 RUWHEID VLGS.
VLGS. NEN 3311 25­11 25.Õ­J.0.1 2S.0Õ­±0.01 NEN 6301 ■ 3632
i:i
K DE WIT
DATUM 25 ­ 1 0.5 25.0 ­ ± 0.05 25.00 ­ i 0.006

27­10­ΘΘ ­ 25 ­ : SYMM. t.o.v. HA RTLIJN


®ξ} 3634 ­ 3638 · 3639

TITEL:
HFD. GROEP

OPDR. GEVER WIJZ. DA T. PA R

ONDERWERP:

ENERGIEONDERZOEK CENTRUM NEDERLAND


AUTEURSRECHT VOORBEHOUDEN BIJ DE WET OPDRACHT NR. FORMAAT A «

Figure 2. Cross­plied glass fibre reinforced polyester coupon specimen.

24
S p e c i men 87PP

­ nf : 4.73F.+7

'\tJ\jjV\pf- Fmax: 22.10 kN


UI
N
UI
(ii · β
w Fmax: 208.7 MPa

UI
Ld
Ζ
^
L·. ­ c
l­H
e


Ì ls/1,
R : .10

Freq: 17.0 Hz

Temp: 18 ±3 °C
H
U)
J ·« RHum: 65 ±15 V.
u
Oi. ■

.2

I I I I ■ ■ I I
mm
1E+7 2E+7 3E+7 4E+7 5E+7
Cyc1 es

Figure 3. Stiffness degradation S/S0 of unidirectionally reinforced specimen during fatigue


with R = 0.1.

Spec imen 148PP

nf : 1.83E+7

O Fmax: 13.13 kN
\
ω Β Fmax: 124.9 MPa
ω ·
ui R : ­1.00
υ ^~" V ^S'So
ζ
Η­ G Freq: 20.0 Hz
Ij. ·"

UI Temp: 1B±3°C

LU RHum: G5 ±15 Y.
a:

W&M
fι ΙΕ+7 2E+7
Cycles

Figure 4. Stiffness degradation S/S„ of unidirectionally reinforced specimen during fatigue


with R = ­ 1 .

25 ­
αϊ

É
οι
E
'o
Ol
(Λ ■D
OI
3.0 FIBER/RESIN
•α u
ni
E­700 U0> 100 H/ kυ. α
01
­ POLYESTER
υ ■D
c OU)
01
k.
ro
2.5 >. αϊ
c ro

NON­KOVEN

k. 6 * ( 0 DECI o
4­,
o
II
lm tr

2.0
. ^ ■a -a
c c
VOLUME I FIBRE
3
ro
v.
J=
■*­« o
­ II

1.5
"♦♦♦< < ♦ o Ο
01
k.
er
L·.
' 'v. "■ n
ΗΛΧ. ' .V. ­)
(1 «^
0)
m
u.
STRAIN ·*­ e
­ 3
O
O
m TI
01
1.0 7
'v. W^ ω ■c

­ JR_­ -J_ E o
•4 ro
σι ξ
ε
ro
.5 χ+*
-a
(1)
­ï i o
ttl
rn η
α.
LL <
ι ι ιιιιιι ι ι ι um 1 1 I I Mil ι ι ι nm ι ι ιιιιιι ι ι ιιιιιι ι ι ιιιιιι ι ι ι mu
0.0 ui
2 3 4 5 6 7 JL
LOG Ν ai
k.
r
σ
LL
(Q
C
FIBER/RESIN LAY­UP VOLUME X FIBRE Ε­MOD CGPa) UTS (MPA )
O)
E ­ 7 0 0 UD* 100 Η / NON­WOVEN 35 26.3 600
1 POLYESTER 6 · CO DEG5
ss
5' -t.
­ * (Q

°S
16
·++-¥■—♦—t FRACTURE
Ig:
il 14


+ R­0. 1
Φ Q}
• R­ ­1
12
2 o
11
o
oigmx**
C 10X STIFFNESS-
—\
O 10 REDUCTION
Dl
3 MAX. <3S8H*
α STRAIN O R-0. 1
θ

C10­­33
<5-Q
OOGXD
1
mm* O R- -1

a: 6
3 O
η
V)
V) iXDO O
-%
(B
α
c *éé
o
3 2 ­

3-
:[fcfl
(B Q I I I I ' I [ill I ■■ ιιm im·' ι ι ι ι '"■' ι ι ι mul ι—ι ι ι IIII
C 3 4 5 6 7 Β 9
3 LOG Ν
α
η
α
c
3
Specimen 14PSL
I .2

nf : 9.92Ε+5
1 .0

V Fmax: 15.00 kN
O
UI
S
If)
S
fmax: 49.0 HPa

UI R : .10
LU •
Ζ
Freq: 10.0 Hz
υ. · Ε
Η
U1
■ Λ Temp: 24 ±3 °C
J ·« RHum: 52 ±15 '/.
U
oc ­
.2

­ m&M
0
() 1Ε+5 2Ε+5 3Ε+5 4Ε+5 5Ε+5 GE+5 7Ε+5 8Ε+5 9Ε+5 ΙΕ+Β
ι
Cyc 1 es

Figure 7. Stiffness degradation S/S0 of a cross­plied specimen during fatigue with R = 0.1.

Specimen 12PSL

nf : 2.61E+4
1 .0
O Fmax: 11.75 kN
in
\
in
^ ^
ïmax: 3B.3 MPa
in · Β
u> R : ­1.00
id
Ζ
Freq: B.0 Hz
1­ ■
Temp: 2 0 ± 3 ° C
ω »S^So
J ·« RHum: 47 ±15 X
u
0¿

.2
\m&m
0
iï lE+4 2E+4 3E+4 4E+4 5E +4
Cyc1 es

Figure 8. Stiffness degradation S/S0 of a cross­plied specimen during fatigue with R = ­1.

­ 28
­π
cã'
c
­ι
ro FIBER/RESIN LAY-UP VOLUME X FIBRE E-MOO (GPa) UTS (MPa)

Ε-700 UT> 100 Μ/ NON-WOVEN 35 11.2 115


POLYESTER 4·<f/-45DEG> s

(Λ φ 16
α
'S - ♦ R­0. 1
ο
| ( oδ
3ω M
S3 FRACTURE
i» ­κ
ο
_ • R­ ­ 1
12
01
ο
c _,
(Β 10 O R­0.1
01 MAX.
3 - ­10X STIFFNESS. RED­
α STRAIN
—ι Β O R­ ­ i
ο "
~-9
[10--3)
3!
3
(D
0)
01 iQO^ ­
-*

[ -<38S>- ♦ <> ♦
α
c
ο : ^D
δ'
^3D00 · « « —
3

3"
(11 1 1 1 1 I 111 1 1 1 1 Mil ι ι ι ι mi ι ι ι ι nu 1 1 I i n n 1 1 1 1 IIII
¡D^D
0
3 4 5 6 7
Ο
<Λ LOG N
01

α

5'
Sp e c i m e n HLH199PP

nf : run out

Fmax : 4 4 . 2 0 kN
in
\
in Cmax : 4 2 1 . 0 MPa
.B
in
R : . 10
U)
UJ
­
ζ
Freq: 1.0 Hz

Temp : 28 ± 3 ° C

UJ RHum: 35 ± 1 5 X
o: ­

­ mm
ø lE+3
Cyc i es

Specimen HLL199PP

nf : 3.42E+7

Fmax: 22.10 kN

fmax: 210.5 MPa


.B

^Sz­So R : .10

.G Freq: 20.0 Hz

­ Temp: 20 ±3 °C
·« RHum: 35 ±15 V.
­
. 2

­
I . , 1
mm
R t 1 1 1 1 1 1 1

1E+7 2E+7 3E+7 4E+7 5E+7


Cyc1 e s

Figure 10. Stiffness degradation of a unidirectionally reinforced specimen during high­low


step fatigue for R = 0.1.

­ 30 ­
Specimen HLH204PP

nf : run out

Fmax: 22.10 kN

fmax: 210.8 MPa

R : -1.00

Freq: 1.0 Hz

Temp: 17 ±3 °C

RHum: 80 ±15 V.

mm
lE+3
Cyc1 es

Specimen HLL204PP
1 .2

nf : 5.54E+5
1 .0
O Fmax: 15.75 kN
in
\
Ul fmax: 149.3 MPa

in R : -1.00
Ld ■

ζ 'S'! o
Freq: 15.0 Hz
u. 'G
t-t
h- • Temp: 23 ±3 °C
ΙΠ
J ■* RHum: G0 ± 15 '/.
u
OL
-
.2

- mmm
0
(i 1E+5 2E+5 3E+5 4E+5 5E+5 6E+5 7E+5 8E+5
Cyc1 es

Figure 11. Stiffness degradation of a unidirectionally reinforced specimen during high-low


step fatigue for R = -1.

- 31
TI
(Q
C
FIBER/RESIN LAY­UP VOLUME X FIBRE Ε­MOD (GPa) UTS CMPA )

E­700 UD+ 100 M/ NON­WOVEN 35 26.3 600


POLYESTER 6*(0 DEG)
IIII
οι οι --Γιο. 16
420
■ (Β c
­ PRE­LOADING
°3 01 Φ
(Β Q.
S o oi­ HIGH­LOW CMPaD
ro 3<Û 14
Isa ­
R­0. 1
▼300(1E5)­210
&s ° 12 315
2- i r <B D420(1E2)-210
öS s ­
O420ClE3)-210\250
§2­1
= Q) Ol 10 V420C1E4)-210
"< Ol £♦ „ rviY^ MAX.
-ι -χ (B MAX.
(B Ol Ol ­ STRESS
STRAIN
r4 Α Ι Λ ^ 210
9,3? B ^ ™^l^eui* R- - 1

il» C10­­3) ­ CHPal


Θ2Ι0(1Ε3)-125/150
«Η 6
­ O CD HAU νΠ Ο210(1Ε4)-125/150
(Β 3 Ol
3 H¡
da YVTVTÏ ηΥ
105 LOW­HIGH R­0.1
3? Ζ k210ClE5)­300
I» 3 J
«.
fil
ro
* Λ
Ol ;
~ CBc
α» 3

äs« _
ι ιιιιιι ι ι ι ι I I 1 III ■ 1 1 1 IIII 1 1 1 1 MII 1 | ιιιιιι
Φ Q "*
01 «<
6
LOG Ν
s ff α
S
3- 3α ro
οι
_ 01 -t
^ 2 O
2. ­k T
Ν 01 fil
o zt. o
3<0
ff £ c3Φ
—ro
ID
C 3.00
S ­
Ci

3 § 2.50
o. ω
0) "D
Ol
m ­
3D
3
C
οι ro 2.00

I
o_ ui
α' f*

I
= · Ol

I
fil O


Φ Q.
ro
4—*
1.50 I
­
ΙΙΙΙΙΙ
ΙΙΙΙΙΙ
Ol ^'
οι 2. CO
■°
Φ =·
o
οι 1.00
­
ro
3'
o
­t
0.50
o
ro
α
ui
■□
(Β 0.00 1 1 1 1 ! Ml ! 1 1 1 1 III 1 1 1 1 1 III 1 1 1 I I I M II! 1 1 1 1 1 1 11 ι ι ι ι 1 III

o 1 2 3 d 5 6 7
3'
ro 10° 10 10 10 I0 10 10 ΙΟ 106
3
Ol

number of cycles
η
o
3
UI
»*■
01
3

ro

c
Φ
3
O
■<
(O
c 3.00 ­

­
Ol
3 Ü 2.50
α en
■α
Dl
m ­
J1
3
C φ
UI ω
2.00
η
η οι
­
η υ
3
S C
c
01 3
< α 1.50 CO
φ ro ­
01 —ι
3­ η
01 *♦
■ο Ο ­
(Β 3
1.00 ­

­
Ο
0.50 ­

Ω.
01
­ο ­
φ I 1 I I 1 III 1 1 1 1 1 III 1 1 1 1 1 III I
ο I t i l i 'ι ι ι I I M III
3' S 1 2 3 4 5 6 7
φ
3
1Q 10 10 10 10 10 10 10 10 !
U1
S
number. of cycles
ο
ο
3
Ά
01
3
.­»■

Õ
01
α
5'
U3

Ι
IO
c 3.00 ­

­
S ë
Φ U)
2.50 ­
οι Τ
3 m
a n ­
fil s·
.. φ
Ol Ol
fil —
2.00 ­
o o Χ'
o => *—' ­
3­C
_ 1.50 ­
e< l3 α
ΑΛΑ
φ 2 ._
4­*
m S co ­

01 3
■g ni 1.00
Φ = ­
" ><
ro
3 ­
O 0.50 ­
­*
O
ro
α.
UI
Τ3
ro 0.00 I 1 1 ! Ill 1 1 1 1 'III 1 I I I 1 Ml 1 1 1 1 1 III 1 1 1 1 1 1 1 1 in
o
1 2 3 4 5 6 7
3
Φ
3
10° 10 10 10 10 10 10 10 10 8
UI

number of cycles
o
o
3
Ul

3

O
ai
α
5'
U3
TI
CD'
c Sp e c i men 9PPN
ro
1 .2

sa -
eoa
Τ) 3 1 .0
O
"τη' S<" en
- ^v___^ S/So
■" οι
^» \
Sa LO
^ "1
ro ai LO
' ωa
LO
õ' U -
3 Ζ
ω LL.
δ5 .G
ω ι­ -
3
C
οί
3
ã
ro'
o u
I-f -
õ'
3
.2

Φ
3
O
1 1 1 ι 1
O
φ
a.
.4 .6
o
N/Nf
3
Φ
3
a
c
Tl
lu'
c
constan t amplitude test 3.0
predicte d lifetimes with
mens. L ine A-O rep rese
Fatigue diagram ol the

2.5

2.0
Α
­
^ ΑΑ
?ff -
™ ~<n
a. 3- Tj
1.5 ν.
Φ Φ m MAX.
I
«„JJ ­ . ^
ω fi· Ό sr STRAIN 0
°-v s
-*J
ι Bl Φ ¡2. CX] *
­ ■ ~ UI
\ 0
°­k= 3*oSs
Ζ3 1.0
fil O 3­
aa;„
5­5?
Φ Q. E;
.5
iδ αsi
fil φ ­■
3
o _ o
3 5; =r. ­
coS§
u g DI I 1 HIM ι ι umi ι ι ι ιιιιι ι ι ι ιιιιι ι ι ιιιιιι ι ι ι ιιιιι ι 1 Ι ι
0.0
aS<
() 2 3 4 5 6 7 8
i" 3 LOG Ν
Φ 3 —.
3.— o \Μ(£Μ
ine Β
ned f

rced

O i . Ol
­ι ÖTj
rt. r* Φ
3­ 3­ O
FIBER/RESIN LAY­UP VOLUME X FIBRE E­HOO CGPol UTS (MPa)

E­700 UD­1D0 H / NON­KOVEN 35 26.3 600


POLYESTER 6 ­ ( 0 DEC)

.99 / j j fEIBULL PLOT R­0. 1

.9

.S
i
_ CH
A R. LIFE
J 101 STIFFNESS RED.

·- 22. 1 kN 17­20HX

f- 2 6 . 3 kN 17­20HZ
FRAC
FAILED
.1 L ~ 3 1 . S kN 15 H i
Ζ 5r FRA crim

.01

Ζ / H ­ 1.95 M ­ / 4. B2

/H­ 2. B5

H©lfcD
IE 3 IE 4 IE S IE 6 IE 7 IE Β IE 9
CYCLES

Figure 18. Weibull analysis of fatigue data of unidirectional reinforced specimens.

VOLUME X FIBRE Ε­MOD CCPo)

E-700 UD» 100 M/ NON­WOVEN


POLYESTER 6 · < 0 DEC)

700

k
• R-0. 1
■R—1
sooC
MAX.
STRESS _
400
^v
mPtj \ ■ CHAR. LIFE
G S X FRACTION

300
Δ~Α
10X STIFFNESS. RED
£ A ■ '>.
200 Δ GA A CHAR. LIFE
: A 3' "*■·■
Δ Δ 51 FRACTION
Α 0 * · —m

100

I I " " " ι ι mill 1 1 Mtttl 1 1 ι ιιιιι 1 1 IMI!. 1 1 DHU t ι itiin Ι Ι MIHI
[M]
0 3 4
LOG Ν

Figure 19. Fatigue diagram for fracture and 10% stiffness reduction after Weibull statistical
analysis of the data.

38
2.1.2 NLR Investigations of GI­UP Materials (W.J.
A . Bonnee)

2.1.2.1 General
Glass­polyester (GI­UP) is a material that is often used in the production of wind turbine
blades. However, despite the frequent use of this material, there is still a need for fatigue
data. In particular, variable amplitude test data are not available. Recently a (variable
amplitude) fatigue spectrum called WISPER was developed for use in the fatigue evaluation
of horizontal axis wind turbine blade materials.
In this investigation constant and variable amplitude fatigue tests were carried out on glass
fibre reinforced polyester coupon specimens with representative glass fibre lay­ups that
combine good longitudinal, transverse and shear properties. The lay­ups that have been
chosen are generally a mix of 0° and ± 45° fibres.
Constant amplitude tests were conducted in order to build up a data base in cooperation with
international partners. These results were compared with spectrum fatigue data obtained with
the WISPER sequence.
It is recommended to continue the testing of sequence effects of the WISPER spectrum and
to evaluate effects of omission and truncation.

2.1.2.2 Material
As Table 1 on page 47 and Table 2 on page 47 show, a total of six batches of material were
used in these tests. The laminate of batches I through III consisted of eight layers of Univer
61 unidirectional (UD) Ε­glass roving and BA SF A420 polyester resin. The glass roving had
a weight of 610 g/m2. The unidirectional glass fibres were held together by a limited amount
of synthetic fibres in the transverse direction.
In the batch IV material four layers of Univer 61 and Cetarol 3606 MV polyester resin were
used.
Batches I through IV were manufactured using the hand lay­up technique. Nominal laminate
thicknesses were 4.8 and 2.4 mm for eight and four layers, respectively.
The material for batches V and VI was manufactured by Risø. Two lay­ups were tested,
namely ±10° (batch V) and 07±45° material (batch VI). The 07± 45°­material had the fol­
lowing stacking sequence: (0°2, ± 45°, 0°)sym. The nominal thickness per layer was 0.2 mm.
The basic material was Ε­glass roving in a UP 333 polyester matrix. The roving was wound
around a flat plate using an automatic winding machine. A fter curing, two flat plates were
obtained. Table 2 on page 47 shows some of the material characteristics.

2.1.2.3 Specimens
From the test plates (batches I through IV) prismatic specimens were cut with the longitudinal
direction parallel to the glass fibre orientation. For load introduction to the specimens, glass
polyester or aluminium tabs were bonded to the specimens.
Figure 1 on page 58 gives an overview of the types of specimens. The dimensions depended
on the thickness of the laminate and the stress ratio, R = omin/omax, selected for the test. The
length between the tabs, Lm, was such that no antibuckling device was necessary.
Aluminium tabs were introduced in order to avoid crushing the specimens as a result of the
clamping forces. A s Figure 1 on page 58 (b) shows, the tabs consisted of a U­profile with
internal dimensions equal to the width of the specimen (25 mm) and an aluminium plate on
top.

- 39
This configuration provides lateral support for the specimen when clamped between the grips
of the testing machine.
The specimens of batch V and VI were 250 mm long and 12 mm wide. The thickness varied
between 1.8 and 2.2 mm, with a nominal value of 2.0 mm.
In order to be able to test the specimens in compression a steel "antibuckling guide" was
developed. Figure 2 on page 59 shows this device. The specimen is supported at the edges,
leaving a window of 8 mm in the width direction and a free length of about 80 mm between
the tabs. The tabs are of steel and are an integral part of the antibuckling guide.
Figure 3 on page 60 gives the dimensions of the various steel components. The "filler plate"
has a thickness of 1.8 mm. For specimens with a thickness in excess of 1.8 mm shims are
added to the filler plate to attain the same thickness as the specimen. Note that the four
"antibuckling plates" support but do not grip the sides of the specimen, i.e. there is no load
transfer between the specimen and the antibuckling plates.

2.1.2.4 Testing Machines and Testing Parameters


Two fatigue testing machines were used. One was a mechanical pulsator with mechanical
grips (batches I through III). The other was a servo-hydraulic machine with hydraulic grips
(batch IV).
The testing frequency was 25 Hz for tests with batches I through III, whereas batch IV
material was tested with frequencies of 5, 10 and 25 Hz. All material was tested at R-values
of 0.1 and - 1 .
For batch IV, wet specimens were also tested at an R-value of - 1 . The specimens were
preconditioned in water at 50 °C for 24 hours. Thereafter fatigue testing took place with the
specimens wrapped in a wet cloth.
All static tests (see Tabie 2 on page 47) and fatigue tests for batches V and VI were carried
out on a servo-hydraulic testing machine fitted with hydraulic grips. Temperature measure­
ments were done at the surface of the specimens during testing. The fatigue cycle frequency
was adjusted so as not to exceed a temperature of 30°C.
The following tests were done:
• Constant amplitude tests at R values of -1 and 10 (compression-compression). The cycle
frequency during these tests was 10 Hz. Batch V material (±10° lay-up) was tested only
at R = 10.
• High-low and low-high tests consisting of blocks of 104 large or small constant amplitude
load cycles, followed by a number of smaller or larger constant amplitude load cycles
until failure. These tests were done for R = -1 as well as for R = 10. The testing fre­
quency was 3 Hz;
• Variable amplitude testing on batch VI material using the WISPER and the WISPERX
spectrum described in Section 1.3. Table 3 on page 48 shows the range-pair-range
counting result for the spectrum.
The cyclic loading wave form for the tests was sinusoidal. A constant loading rate was used
for the variable amplitude testing. This means that the testing frequency depends on the load
range applied. In most cases the frequency was 7.5 Hz for a load range of 11 WISPER lev­
els. For larger load ranges the frequency was less and for smaller load ranges the frequency
was higher, with a maximum of 20 Hz for most test specimens.
During WISPER and WISPERX testing the maximum and minimum load levels and corre­
sponding displacements of the grips were measured at certain intervals. From these values
the relative stiffness of the specimens was calculated. Table 4 on page 49 shows an

- 40
example of the computer output. Fluctuations of the Ε-values were reduced by using the
moving average of 19 values, as shown in column 8 of Table 4 on page 49.
The spectrum fatigue tests were preceded by an "adjusting cycle" in order to adjust the load
of the testing machine. This was necessary because it was decided to express the fatigue
loads in strain levels. This adjusting cycle ranged from zero load to emax and back to zero,
or from WISPER level 25 to 64 to 25.

2.1.2.5 Results
Static tests
For batch I through IV material, tensile and compression strength, modulus of elasticity (E),
and interlaminar shear strength (ILSS) were obtained from tests, see Table 1 on page 47.
The fibre volume fraction was estimated from the E-modulus.
Three tensile and three compressive tests were done on the 07+45° material (batch VI). The
results are given in Table 2 on page 47. Figure 4 and Figure 5 on page 61 show examples
of failed tensile and compression specimens. Failure of the tensile specimen was typically
"brush-like" (Figure 4 on page 61). The compression specimen failed at one end between
the tabs and one side of the antibuckling guide.

Fatigue tests
Figure 6 on page 62 shows the fatigue test results for material from batches I through III.
Both R = 0.1 and a limited number of R = -1 results are given. As can be seen, there is
considerable scatter. However, this is not unusual for unidirectional material.
The results for R = 0.1 were processed using a Weibull analysis. The solid line in Figure 6
on page 62 indicates 95 % survival rate with a probability of 95 %. Table 5 on page 50 shows
the test results in tabular form, and some characteristic values of the Weibull analysis. In this
table α and β are the shape parameter and the scale parameter, respectively. The column
"95 %" indicates the life for 95% survival with 95% probability. In the column "mean" the
Weibull mean life is given.
In Figure 7 on page 62 the results for batch IV material are given for the test conditions R
- 0.1 and - 1 . Scatter is rather small. This is because the effect of clamping is smaller for the
thin specimens than for thick specimens. Tables 6 and 7 on page 51 show the test results
and Weibull characteristics for R = 0.1 and R = - 1 , respectively.
Figure 8 on page 63 and Table 8 on page 52 show the results for the R - -1 WET tests.
Also, Figure 8 shows that the wet specimens had shorter lives then the dry specimens.
In Figure 9 and Figure 10 on page 64 two examples are shown for the appearance of
specimens after testing. Both specimens are of the batch IV material and are fitted with alu­
minium tabs. Figure 9 shows a specimen that failed in tension (R = 0.1), whereas the
specimen shown in Figure 10 failed in compression (R = -1). In general, delamination of the
specimens started from underneath the tabs and grew toward the area between the tabs.
Final fracture was mostly underneath the tabs (Figure 9). However, when compressive
strains are involved the specimen may also fail in compression between the tabs
(Figure 10).

Constant amplitude tests


Table 9 on page 53 and Figure 11 on page 64 through Figure 13 on page 65 show the
results of the constant amplitude tests for the ±10° and 07±45° lay-ups (batches V and VI).
The stress ratio R was 10 for the ± 10° tests, whereas the 07± 45° material was tested at

41
R = 10 and - 1 . The vertical axes in Figure 11 through Figure 13 represent the lowest strain
£mln in the load cycle.
A Weibull analysis was performed on the test results. The Weibull characteristics are given
in the last four columns of Table 9 on page 53. The probability curves for 95 % survival with
a confidence level of 95 % are shown in Figure 11 through Figure 13.
In Figure 14 on page 66 a failed ±10° specimen (batch V) is shown tested at R = 10. There
are three areas to be distinguished, namely: a central delaminated area (white) where final
fracture took place, an undamaged part (black) and the tab area. From this it follows that
delamination started in the middle of the specimen. Figure 15 on page 66 shows a
07± 45°-specimen also tested at R = 10. Failure was also in the central part of the specimen.
No damage was observed initiating from underneath the tabs. Both specimens are typical
examples.

High-low versus low-high tests (batch VI material)


In Table 10 on page 54 the results for the high-low and low-high tests are shown. To
understand these results, the first line can be taken as an example. This was a test at R =
10 and the specimen was given 104 load cycles at a load level of ε™ = -1.2 %. The damage
d resulting from these 10" cycles can, according to Miner, be calculated from the constant
amplitude life at emin = -1.2 %. These data are given in Table 11 on page 54, which is an
excerpt of Table 9 on page 53. The value for

1
98,364
After this the specimen was given 31,170 load cycles at a level of £„,,„ = -1.5 % before final
failure occurred. This resulted in a damage contribution of

d2 = %™ = 5.034
^ 6,192
The total damage of the two loading blocks then amounts to 0.102 + 5.034 = 5.136. The total
damage was calculated for all 20 specimens as shown in Table 10 on page 54. From the five
results for each condition the highest and lowest value was omitted to reduce scatter, and
the remaining three were averaged. This resulted in the values given in Table 12 on page
54.

WISPER and WISPERX tests (batch VI material)


A total of 20 specimens were tested using the WISPER spectrum and 10 specimens using
the WISPERX spectrum. The results of these tests are given in Table 13 on page 55. The
last column of this table gives the life expressed in number of passes through a complete
WISPER or WISPERX spectrum. The results expressed in number of cycles are divided by
132,711 for WISPER and by 12,831 for WISPERX. In Figure 16 on page 67 the results are
given in an ε-Ν diagram.
The results of both spectra were compared by performing a Weibull analysis on the results
expressed in number of passes through the spectrum. The results of this analysis are given
in Table 15 on page 57. Furtheremore, the 95 % probability of survival curves and the mean
curves for both spectra are given in Figure 17 on page 68.
As was described in paragraph 2.1.2.4, stiffness measurements were made during fatigue
testing. Some typical results are shown in Figure 18 on page 69, where the relative stiffness
for five specimens during WISPERX testing is plotted against the fraction of life. In most
cases the relative stiffness increased to above 100 % in the first load cycles. Due to scatter
and/or the spectrum on the stiffness curves.

- 42
Comparison of spectrum and constant amplitude loading
An important exercise to be done is to compare the WISPER and WISPERX test results with
the constant amplitude data. This comparison will show whether spectrum loading is more
detrimental than constant amplitude loading. Comparison was done as described below.
Hagg and van Delft [2] developed the following equation for the constant amplitude fatigue
behaviour of glass fibre reinforced plastics:

logA/=10­10.3log(10~^f (3.1)

where:
omax is the maximum stress in the fatigue cycle
UTS is the ultimate tensile strength.
This relation can be checked with the R = ­1 constant amplitude test data given in
Figure 13 on page 65. If the following values are substituted in Equation 3.1:

R = ­1
UTS = 752 MPa
o™ = 29257 ε ^ (MPa)

it follows that
log Ν = 10 ­ 1 0 . 3 log (518.7 ε ^ ) (3.2)

In Figure 19 on page 70 the results for the R = ­ 1 tests are given and compared with the
relation described by Equation 3.2. This relation correlates very well with the 95% probability
of survival curve for the test results.
For this reason Equation 3.1 was used in a Palmgren­Miner calculation. This calculation is
shown in Table 14 on page 56 for the WISPER spectrum with zma¡¡ = 2 %. For every range
size the mean R­value, £mean and ε„,8Χ are calculated. The R­value and ε„,ΒΧ were substituted
in Equation 3.1, which resulted in the life Ν expressed in number of ranges. For every range
size the damage factor n/N could then be calculated. The sum of all damage factors:

( 3 3 )
ZJNΝ ·
is the Palmgren­Miner sum. The expected life according to Palmgren­Miner's rule LPM is:

LpM = ^ (3­4)

expressed in the number of passes through the spectrum.


The first line in Table 14 on page 56 shows the damage (n/N)adj done by the adjusting cycle
preceding the testing. This adjusting cycle was necessary in order to obtain the correct emax
level for the test. If this additional one­time damage factor is incorporated, the life will be:

1-(
-w)
LPM = ^ - ^ (3.5)

Because of the low value of (n/N)ad¡ the influence of this adjusting cycle in the Palmgren­Miner
calculation is negligible.

43
Table 16 on page 57 shows the expected life for the WISPER and WISPERX spectrum for
various values of emax. These results are also given in Figure 17 on page 68. As can be seen,
in the calculation there is little difference between WISPER and WISPERX life expressed in
number of passes through the spectrum. Furthermore, the calculated life is higher by a factor
of 20 to 60 than the actual results from the tests, indicating that there is probably a sequence
effect involved for the present material, resulting in lower lives.

2.1.2.6 Discussion
Fatigue testing of unidirectional composites is always a difficult task, because the clamping
arrangement of the composite specimen is in general a weaker joint than the material itself.
In practice this means that the onset of failure and final failure are influenced by the grips.
In testing the batch I through IV material, splitting and delamination started from underneath
the tabs and then the damage grew into the area between the tabs. Final failure generally
occurred under the tabs in tension and between the tabs in compression. As a result of the
influence of gripping, the fatigue results will be conservative.
Secondary effects such as gripping result in scatter of the test results. This could be
observed when comparing the results with glass and aluminium tabs. From the tests it could
also be concluded that the use of hydraulic grips for composites is preferable.
Comparison of the fatigue results for R = 0.1 and - 1 shows the ε™χ - log Ν curve for R = -1
to be at Ema, levels aproximately half of those for R = 0.1. (See Figure 6 on page 62). If these
fatigue results were presented on the basis of strain range (Δε - ε„,3Χ - £min) the curves for both
R-values would almost have overlapped, indicating only a Δε dependency and no R
dependency for these two R-values. As expected, the results for the wet tests at R = - 1
showed slightly shorter live.
Although there was some thickness variation in the specimens used from batches V and VI,
their general quality was very good. The differences in thickness could easily be adjusted
by using shims in the antibuckling guide.
The antibuckling guide gave good support to the edges of the specimens without obscuring
them. Therefore, the initiation of damage could be seen very clearly. Continuous temperature
measurements on the specimen surfaces were easily made. This meant that testing fre­
quency could be controlled to avoid overheating the specimens.
Elastic modulus measurements were made using the load and displacement of the testing
machine grips. This gave significant scatter in the results, which was reduced by using a
moving average of 19 values. In most cases there was an increase of the Ε-modulus after
some cycles. This is most probably the result of alignment of U|D fibres with the loading
direction. After approximately 30% of the life, the Ε-modulus had decreased to the initial
value, and from then on a gradual decrease occurred until failure. There were no clear dif­
ferences found in the behaviour of specimens tested at different load levels.
The constant amplitude fatigue data of the 07±45° laminate are different for R = 10 and R
= - 1 . The εΠ!1χ levels of the R = -1 curve are not at half height of the R = 10 curve, as
expected, but higher (see Figure 12 and Figure 13 on page 65). This means that the ranges
in R = 10 testing are more severe than the same ranges in R = -1 testing. In the batch I
through IV tests it was seen that for R = 0.1 and R = -1 testing there was no R dependency.
The above implies that compression-compression testing is relatively severe for the present
material.
With respect to the high-low and low-high tests, it is expected that the effect of large load
cycles preceding smaller load cycles for composites is unfavourable. This is because dam­
age that occurred as a result of the large load cycles can grow under the influence of the
smaller cycles, whereas this would not have happened if no damage were present.

- 44
If no sequence effects were present, the damage sum for the high-low and low-high tests
would equal 1. A result smaller than 1 would indicate that a sequence effect is present with
a negative effect on life. A result larger than 1 indicates a positive effect on life. The test
results (see Table 12 on page 54) show for three of the four conditions a damage sum less
than 1. Only for high-low testing at an R-value of 10 was the damage sum above 1. The
above results should be treated with care since the number of cycles in the first load block
of the high-low and low-high tests is rather small, namely 104. This means that a very small
amount of damage has been done (0.102 for R = 10 and 0.054 for R = -1). The resulting
number of cycles until failure in the second load block falls in the scatter band of the constant
amplitude results at that load level.

A possible explanation for the behaviour of the high-low tests at R = 10 and R = -1 is as


follows: At R = 10 the high compressive loads may reduce the stress concentrations in the
material without causing fibre fracture, whereas at R = -1 fibre fracture and matrix cracks in
the ±45° layers may occur. Reducing stress concentrations in R = 10 tests then results in
longer lives when testing at the lower load level. On the other hand because of the initiation
of cracks at the high load level in the R = -1 tests, damage may then develop at lower load
levels, causing a total damage factor less than 1.
From the WISPER and WISPERX tests it is reasonable· to state that omission of the smaller
ranges does not have a large influence on life (see Figure 17 on page 68). The same con­
clusion can be drawn from the Palmgren-Miner calculation. The above implies that the
WISPER spectrum can be reduced to 10 % of its size. This would greatly decrease testing
time. However, this conclusion is based only on a limited number of tests. Further testing on
the effect of other omission and truncation levels is recommended.
The comparison of the spectrum fatigue tests with constant amplitude data was done using
Palmgren-Miner's rule and the relation of Hagg and van Delft [2] The latter relation was
checked only for R = -1 test results. The results show a calculated life which is higher by a
factor of 20 to 60 than the test results. This is a large diference and, what is more important,
the calculation results in an unconservative value for the expected life. A possible explanation
for the relatively low lives in the tests could be the presence of the adjusting cycle preceding
the test. However, in the Palmgren-Miner calculation the contribution of this cycle was
already shown to be negligible, and in the real test larger ranges than the adjusting cycle
appear early in the WISPER spectrum. This means that only in tests with a life much shorter
than 1 complete pass through the spectrum would a considerable effect of the adjusting cycle
be expected to be present. In fact, the discrepancy between test results and calculation
(Figure 17 on page 68) is larger at the higher strain levels, but it is not clear whether this is
entirely due to the adjusting cycle.

2.1.2.7 Conclusions and Recommendations


From the present investigation the following conclusions can be drawn:
• Testing of unidirectional composite material resulted in conservative fatigue results.
• The use of thin specimens and hydraulic grips improved the test results.
• In tests at R-values of 0.1 and -1 no R dependency was observed. The fatigue results
were influenced only by the load range.
• Fatigue life for wet specimens was slightly lower than for dry specimens.
• The antibuckling guide, as used in the CEC programme, was very practical. Small dif­
ferences in specimen thickness could easily be compensated for.
• Measurement of the Ε-modulus using displacement of the fatigue testing machine grips
was probably not discriminating enough to show differences for various load levels.
• Based on load ranges, the fatigue behaviour for R = 10 (compression-compression) was
unfavourable compared to the R = -1 tests.

45 -
• The high-low and low-high test results showed some indications of possible sequence
effects. However, the limited number of tests done so far makes it difficult to draw firm
conclusions.
• The omission spectrum WISPERX and the basic WISPER spectrum resulted in compa­
rable lives. The Palmgren-Miner calculation for WISPER and WISPERX gave the same
result.
• The Palmgren-Miner calculation based on mean R-values for each load range and the
relation of Hagg and van Delft [2] resulted in unconservative values.
From the above it is recommended to:
• Continue testing for sequence effects in glass fibre polyester material.
• Investigate effects of other omission levels and clipping for a complete evaluation of the
WISPER spectrum.

2.1.2.8 Bibliography
[11 Ten Have A.A WISPER: A standardized fatigue load sequence for H AWT-blades NLR MP 88029
U (1988)
[2] Hagg. F. and van Delft D.R.V. Design process from site data until built turbine (1989)

46
2.1.2.9 Tables

Batch
I II III IV Remarks
Glass-fibre Univer 61 UD E-glass Univer 61 UD E-glass
Hatrix BASF A 420 polyester Cetarol 3606 MV polyester
Number of layers 8 4
Production technique Hand lay-up Hand lay-up
Tensile strength at (MPa) 670
Compressive strength σ (MPa) 380 475
Modulus of e l a s t i c i t y E (GPa) 29.5 36.7 29.5 30.0
Average thickness t flV (mm) 4.7 4.1 4.8 2.4 Nominal thickness 0.6 mm/layer
Fibre volume fraction'(%) -36 -46 -36 -36 Estimate r e s u l t i n g from E-mod
Inter laminar
shear strength ILSS (MPa) 50.6 52.7

Table 1. Static properties of the material of batches I through IV.

LAY-UP v f (%) at (MPa) E c (MPa ec (%) ac (MPa) REMARKS

±10° 546 40900 Risø-data,

0°,±45 47,9 Risø-data,


corrected to
2mm n o m i n a l
thickness

752 29257 2,57 1892 NLR - d a t a


Average of
three results

Table 2. Characteristics of Risø material.

47 -
RANGE-SIZE NUMBER CUMUL. AV. MEAN MIN MAX
(LEVELS) NUMBER MEAN (RMS) (LEVEL) (LEVEL)
(LEVEL)

63 1 1 32.5 0.0 1 64
54 1 2 28.0 0.0 1 55
48 4 6 32.0 1.0 7 57
46 6 12 32.0 0.0 9 55
45 4 16 32.5 0.0 10 55
43 3 19 36.5 4.2 12 64
42 2 21 31.0 0.0 10 52
41 4 25 33.0 1.5 11 55
40 2 27 35.0 0.0 15 55
39 4 31 33.5 2.0 12 55
38 2 33 38.0 0.0 19 57
36 6 39 34.0 6.5 7 58
35 9 48 33.6 4.6 11 56
34 10 58 34.8 5.6 10 57
33 2 60 40.5 0.0 24 57
32 8 68 36.5 4.3 13 55
31 19 87 39.6 2.3 19 57
30 7 94 39.4 1.8 23 57
29 60 154. 40.3 1.0 23 57
28 36 190 39.4 4.0 10 55
27 26 216 41.2 .7 26 55
26 144 360 41.2 3.7 11 57
25 30 390 40.2 1.4 26 56
24 586 976 39.3 .8 26 57
23 560 1536 39.2 3.2 21 57
22 1264 2800 44.3 1.7 26 56
21 2666 5466 45.8 1.2 28 57
20 2450 7916 42.2 1.9 22 55
19 3218 11134 43.3 1.0 22 55
18 8614 19748 39.9 3.3 22 55
17 5914 25662 39.3 2.4 23 54
16 17292 42954 44.8 1.4 24 54
15 26364 69318 40.6 3.4 25 53
14 62122 131440 41.8 3.3 25 53
13 46492 177932 39.3 .7 27 46
12 16484 194416 35.5 .6 25 46
11 70650 265066 41.5 .1 29 47
10 180 265246 40.6 1.6 31 48
9 50 265296 40.5 .2 36 46
8 10 265306 39.2 1.9 33 46
7 22 265328 39.5 0.0 36 43
5 92 265420 38.5 0.0 36 41
4 2 265422 35.0 0.0 33 37

Table 3. Range-pair-range counting result.

48
8.10000E+06 2110 1037 100 -337 4.63 93.4 90 .8 17.5
8.12000E+Û6 1733 531 -59 -602 4.45 Si'. 3 30 . S 17.4
8.14000E+Û6 1555 433 -151 -646 4.51 90 50 .3 ± í . 0
8.16000E+06 1326 621 13 -563 4.53 90.4 30 .8 17.6
3.180ÜOE+O6 2671 1366 350 -222 4.56 31 30 .3 17.6
3.2Û3Û0E+Û6 20 IS 304 47 -496 4.47 33.2 30.3 17.3
8.22Û0ÛE-Î-06 2020 811 50 -484 4.53 50.4 90.3 17.8
8.2400ÚE+06 2576 1177 308 -300 4.6 31.8 30.8 17.8
S.260GÜE+06 2110 1087 3S -353 4.54 30.6 30.8 17.3
3.2800ÚE+06 1323 523 17 -607 4.47 35.2 30 .8 17.9
8.30000E+OS 2103 1Û87 99 -342 4.63 52.4 90.6 18
8.32000E+06 2203 333 134 -433 4.55 30.8 30.7 18.1
8.340ÛÛE+06 2018 527 52 -609 4.51 90 30.8 13.2
8.36OO0E+O6 2107 10 33 89 -348 4 ,5S 93 30 .3 13. 3
3.33000E+06 2433 301 259 -443 4.51 90 30.8 1o c
3.4ÛÛÛ0E+06 1Û91 2115 -353 31 4.55. 30.8 31 13.7
8.4200ÛE+Û6 1642 529 -122 -615 4.52 30.2 31 18.8
3.44Q00E+06 21Û 7 1085 86 -360 4.58 91.4 30.8 18.8
8.4600ÛE+Û6 2112 1034 37 -358 4.62 92.2 50.7 IS.9
8.43000E+06 2016 800 53 -433 4.45 33.8 50 .3 18.3
8.5UOOOE+OÕ 2015 3Ü5 44 -483 4.56 91 90.3 13.3
8.52000E+06 2016 302 44 -475 4.64 52.6 30. S 18.3
8.54GOÖE+Ü6 2576 1084 311 -345 4.55 90.8 30 .7 13. 6
S.56000E+Û6 1547 435 -157 -646 4.55 30.3 30 .6 18.5

Column 1 Number of load levels


21■ Load levels

• Displacement values
6 Modulus of elasticity E
7 Relative modulus of elasticity w i t h
respect to first measurement (%)·
8 Moving average for 19 values of E (%)
9 Temperature of the specimen surface (°C)

Table 4. Computer­output for WISPER and WISPERX tests.

49 ­
WEIBULL
£
max(%) N α β 95 % mean

0.47 27.700.000

0.55 20.172.000 9.8 24.232.920 17.622.410 23.032.020


25.783.280

0.57 55.000.000

0.61 18.852.600 9.9 22.579.840 16.506.330 21.475.870


24.000.000

0.65 3.031.520

0.67 11.271.000 4.7 16.512.630 8.507.047 15.106.500


18.788.500

0.69 4.069.000

0.71 17.474.500 2.7 13.254.870 4.686.758 11.794.450


17.345.400
8.108.300
6.425.100
9.270.300

0.75 2.282.700

0.79 11.145.200 2.8 9.008.411 3.018.730 8.026.895


4.800.000

0.85 10.250.000

0.91 1.302.800

1.01 1.396.200

1.17 196.200 ,

Table 5. Batch I, II and III fatigue test results and Weibull characteristics (R = 0.1).

50
WEIBULL
£
max<%> f(Hz) L (mm) N α ß 95 % mean
m
1.6 10 40 487.000 2.1 363.476 80.883 321.967
1.6 10 40 153.000

1.4 10 40 546.000
1.4 25 40 535.000

1.2 10 40 3.187.000 3.1 2.613.764 943.900 2.336.050


1.2 25 40 1.454.000

1.0 25 40 6.491.000

0.8 25 40 19.680.000

0.7 25 40 28.579.000

0.6 25 40 16.976.000-»

0.4 25 40 17.676.000-

Table 6. Fatigue test results for batch IV material (R = 0.1).

WEIBULL
£
max<*> f(Hz) L (mm) Ν a ß 95 % mean
m
1.2 5 20 29.000

1.0 5 20 145.300 2.0 95.372 22.384 84.515


1.0 10 20 95.000
1.0 5 20 75.200
1.0 5 20 23.000
0.8 5 20 836.700 1.2 335.659 27.943 317.776
0.8 10 20 146.000
0.8 10 20 138.500
0.8 10 40 138.000
0.6 10 40 6.123.000 2.8 4.261.204 1 541.370 3.794.823
0.6 10 20 4.465.000
0.6 10 20 3.544.620
0.6 10 20 3.148.000
0.6 10 20 1.645.000
0.4 10 40 8.000.000-

Table 7. Fatigue test results for batch IV material (R = ­1 dry).

51 -
WEIBULL

e
max(%> f(Hz) L^mm) N α ß 95 % mean

1.0 25 20 45.000 2.6 34.834 11.629 30.927


1.0 25 20 45.000
1.0 25 20 37.000
1.0 25 20 30.000
1.0 5 20 21.000
1.0 5 20 7.970

0.8 10 20 251.000 1.8 147.079 27.877 130.987


0.8 10 20 120.000
0.8 10 20 116.940
0.8 25 20 35.000

0.6 25 20 3.569.900 1.1 1.264.249 86.254 1.234.229


0.6 10 20 1.055.940
0.6 10 20 973.880
0.6 25 20 354.000
0.6 10 20 202.370

Table 8. Fatigue test results for batch IV material (R = ­1 wet).

52
WEIBULL
SHAPE SCALE
lay-up R «n.ln<*> N a ß 95 % mean

±10% 10 0,5 2.282.880


0,5 9.495.980 2.1 6.541.680 1.573.212 5.794.006
0.5 5.533.340

0.6 515.380 2.0 383.175 83.693 339.466


0.6 159.460

0.63 320.060
0.7 76.290 32.6 80.606 73.259 79.252
0.7 82.120

0°,±45° 10 0.5 3.800.000-


0.5 10.000.000-

0.7 10.000.000-

0.8 4.900.000 5.5 6.802.202 3.848.137 6.277.721


0.8 7.600.000

1.0 720.000
1.0 260.000 3.1 640.645 244.988 573.041
1.0 730.000

1.2 79.000
1.2 170.000 2.0 110.974 24.572 98.364
1.2 44.000

1.5 12.000
1.5 5.900 1.2 6.511 486 6.192
1.5 760

0°,±45° -1 0.4 10.000.000-

0.5 1.400.000 2.2 3.125.165 774.846 2.767.904


0.5 4.100.000

0.7 72.000
0.7 340.000 1.7 207.930 36.633 185.361
0.7 140.000

0.8 52.000

1.0 13.000
1.0 2.600 1.4 6.849 779 6.260
1.0 3.000

1.2 300
1.2 1.300 1.4 690 81 629
1.2 270

1.5 200

Table 9. Results of constant amplitude testing.

53
d
f total
R min/%) N
l d
l
£
min2(*> N2 d2 d-,+d2
10 -1.2 0.102 -1.5 31.170 5.034 5.136
10 -1.2 0.102 -1.5 5.680 0.917 1.019
10 -1.2 0.102 -1.5 2.900 0.468 0.570
10 -1.2 0.102 -1.5 3.610 0.583 0.685
10 -1.2 10 4 0.102 -1.5 540 0.087 0.189

10 -1.2 10* 0.102 -1.0 854.150 1.491 1.593


10 -1.2 0.102 -1.0 0 0 0.102
10 -1.2 io4 0.102 -1.0 597.020 1.042 1.144
10 -1.2 0.102 -1.0 1.124.280 1.962 2.064
10 -1.2 10 4 0.102 -1.0 736.700 1.286 1.388
-1.0 -0.7 10 0.054 -1.0 5.850 0.935 0.989
-1.0 -0.7 4 0.054 -1.0
10
Λ
5.430 0.867 0.921
-1.0 -0.7 0.054 -1.0 3.280 0.524 0.578
-1.0 -0.7 io 4 0.054 -1.0 130 0.021 0.075
-1.0 -0.7 IO 4 0.054 -1.0 3.670 0.586 0.640
-1.0 -0.7 io4 0.054 -0.5 30 0 0.054
-1.0 -0.7 10 0.054 -0.5 1.635.210 0.591 0.645
-1.0 -0.7 10
2 0.054 -0.5 925.240 0.334 0.388
-1.0 -0.7 4 0.054 -0.5 1.509.550 0.545 0.599
10
-1.0 -0.7 IOA4 0.054 -0.5 830.050 0.300 0.354

Table 10. Results of thé high-low and low-high tests (07±45° lay-up).

WEIBULL MEAN
£
R min(%) LIFE

10 -1.5 6.192
-1.2 98.364
-1.0 573.041

-1 -1.0 6.260
-0.7 185.361
-0.5 2.767.904

Table 11. Excerpt from Table 9 on page 53

d
total ( m e a n o f
R 3 values)

Low-high 10 0.758
high-low 10 1.375
low-high -1 0.713
high-low -1 0.447

Table 12. Mean total damage of duplex load level block testing.

54
£
maxW N(cycles) Number of passes
in spectrum through spectrum

WISPER 2.0 10 4 0.08

1.8 3.8.104 0.29


1.8 3.2.103 0.02

1.6 1.6.105 1.21


1.6 1.7.105 1.28
1.6 4.4.IO3 0.03
1.6 2.5.105 1.88

1.4 1.1.IO6 8.29


1.4 1.4.IO5 1.05
1.4 8.4.104 0.63
1.4 2.8.106 21.10

1.36 3.5 IO6 26.37

1.2 IO5 0.75


1.2 2.4.IO6 18.08
1.2 10 7 - 75.35-
1.2 1.4.IO5 1.05
1.2 6.6.105 4.97
1.2 4.0. IO6"* 30.14-

1.0 1.2.IO6 9.04


1.0 1.1. IO7"* 82.89

WISPX 1.6 2.2.IO3 0.17


1.6 1.2.IO4 0.94

1.4 3.2.IO4 2.49


1.4 3.4.IO4 2.65

1.2 9.1.IO3 0.71


1.2 9.7.IO4 7.56
1.2 IO6 77.94
1.2 1.1.IO6 85.73
1.2 2.8.IO5 21.82

1.0 3.4.IO6 264.98

Table 13. Results of the WISPER and WISPERX tests.

- 55
spectrum = WISPER «max = 2.0000 %
range--size number n «mean R N n/N
(levels) (ranges) (ranges)

39. 2 0,.010000 0,.00 0.87E+03 0 .231E-02

63. 1 0,.003846 -0,.62 0.62E+01 0,. 162E+00


54. 1 0..001538 -0 .80 0.30E+02 0 .330E-01
48. 4 0..003590 -0,,55 0.10E+03 0,.393E-01
46. 6 0..003590 -0,.53 0.16E+03 0,.380E-01
45. 4 0,.003846 -0 .50 0.20E+03 0..202E-01
43. 3 0..005897 -0,.30 0.32E+03 0..948E-02
42. 2 0..003077 -0,.56 0.40E+03 0,. 496E-02
41. 4 0..004103 - 0 ,.44 0.52E+03 0.■774E-02
40. 2 0..005128 -0,.33 0.67E+03 0,. 300E-02
39. 4 0,.004359 -0,.39 0.87E+03 0 .462E-02
38. 2 0..006667 - 0 ,.19 0.11E+04 0.. 177E-02
36. 6 0..004615 -0,.33 0.20E+04 0,. 304E-02
35. 9 0.,004410 - 0 ..34 0.26E+04 0..341E-02
34. 10 0..005026 - 0 ..27 0.36E+04 0,.281E-02
33. 2 0.,007949 - 0 ..03 0.48E+04 0..414E-03
32. 8 0.,005897 - 0 ,.16 0.66E+04 0.. 121E-02
31. 19 0..007487 - 0 .,03 0.92E+04 0.. 206E-02
30. 7 0..007385 - 0 ..02 0.13E+05 0. 543E-03
29. 60 0,,007846 0.,03 0.18E+05 0. 328E-02
28. 36 0..007385 0.,01 0.26E+05 0., 137E-02
27. 26 0..008308 0..09 0.38E+05 0.. 681E-03
26. 144 0..008308 0.,11 0.56E+05 0.,256E-02
25. 30 0..007795 0.,10 0.84E+05 0..356E-03
24. 586 0..007333 0.,09 0.13É+06 0.,456E-02
23. 560 0.,007282 0.,11 0.20E+06 0. 281E-02
22. 1264 0.,009897 0..27 0.31E+06 0. 402E-02
21. 2666 0.,010667 0..33 0.51E+06 0. 524E-02
20. 2450 0..008821 0.,26 0.84E+06 0. 292E-02
19. 3218 0..009385 0..32 0.14E+07 0. 226E-02
18. 8614 0..007641 0..25 0.25E+07 0.. 346E-02
17. 5914 0.,007333 0,.25 0.45E+07 0., 132E-02
16. 17292 0.,010154 0,.42 0.84E+07 0..207E-02
15. 26364 0..008000 0..35 0.16E+08 0..162E-02
14. 62122 0.,008615 0..41 0.33E+08 0,. 188E-02
13. 46492 0..007333 0,.38 0.71E+08 0.. 655E-03
12. 16484 0,,005385 0,.27 0.16E+09 0,.102E-03
11. 70650 0..008462 0,.50 0.40E+09 0 . 178E-03
10. 180 0..008000 0,.51 0.11E+10 0 . 170E-06
9. 50 0..007949 0,.55 0.31E+10 0..159E-07
8. 10 0,.007282 0,.56 0.11E+11 0,.948E-09
7. 22 0..007436 0..61 0.42E+11 0..527E-09
5. 92 0..006923 0,.69 0.13E+13 0,.689E-10
4. 2 0..005128 0..67 0.13E+14 0.,150E-12

Ση = 265422 ranges D = Σ(η/Ν) = 0.3783


expected life duration = 2.6 WISPER sequences
or 350769. cycles

Table 14. Palmgren­Miner calculation.

56 ­
WEIBULL
£
max< )%
a
*
β 95 % * mean

WISPER 1.8 0.90 0.15 0.005 0.156


1.6 1.02 1.11 0.063 1.097
1.4 0.78 6.76 0.164 7.780
1.2 0.83 5.58 0.166 6.167

WISPX 1.6 1.40 0.61 0.066 0.556


1.4 38.53 2.61 2.406 2.571
1.2 0.77 34.01 '0.822 39.767

Table 15. Comparison of WISPER and WISPERX results.

Palmgren-Miner sum EXPECTED LIFE

£
max(%) D-Z§ Number of spectra Number of cycles

WISPER 2.0 0.3783 2.6 3.5.105


1.8 0.1278 7.8 1.0.106
1.6 0.0380 26.3 3.5.106
1.4 0.0096 104.1 1.4.107
1.2 0.0020 509.5 6.8.107
1.0 0.0003 3332.1 4.4.108

WISPX 1.6 0.0373 26.8 3.4.105


1.4 0.0094 106.0 1.4.106
1.2 0.0019 518.4 6.7.106
1.0 0.0003 3390.4 4.4.107

Table 16. Palmgren-Miner results.

57
2.1.2.10 Figures

0 . 8 ­ 1.8 ­25­

^^

a) Specimen with glass-tabs, batches I and Π

­25­

■I
ι
I |_
Η— L
m

b) Specimen with aluminium -tabs, batches HI and _Z3Z~

DIMENSI ONS IN mm |

BATCH

ï +π+m DZ

R 0.1 ­1.0 0.1 ­1.0

L 250 190 180 160 180

L
m 110 50 40 20 40

Figure 1. Dimensions of the test specimens of batches I through IV.

­ 58
Figure 2. The antibuckling guide.

- 59
Tl
lu'

O
3'
ro
3 F I L L E R PLA TE
in TAB (4x1 1 1 0 x 17 χ 1.8 (2x)
õ
3
Ui
12
a
Ol
3
£■
σ
c
o

m Α Ν Τ Ι ­ B U C K L I N G PLA TE
c
α
ni
110x20x4(4x1

o 60
o
3 M6 BOLT A ND NUT (6x)
■a
o
Θ
3
<I>
3

20
SPECIMEN­
CROSS SUPPORT
5 0 x 2 5 χ 4 (4x)

7.5

ΙΑνΑΛΛΛΛ^ν^

DIMENSIONS IN mm
Figure 4. Typical tension failure.

Figure 5. Typical compression failure.

61
2.0 ­
­
Φ R = 0.1
Χ R = ­1.0
­ 9 5 % PROBA BILITY
OF SURVIVA L
­
• •

1.0­ ­

>· · •
— Χ ­


■ ' ' ' 'I I ' ι ,l ι 1 1 I1I I 1 , 1 , ,,,,1 , ,

— log Ν (CYCLES!

Figure 6. Fatigue test results for batch I, II and III material.

• R C.I
X R = ­1.0
­ 9 5 % PROBA BILITY L · ·
OF SURVIVA L

\ «·
X >( · ·

bt χ χ χ ·

] ~ ^ ^ ^ XXX X ·

~­—|x xx χ χ ^ —

Χ— ·—­

7 8

log Ν (CYCLES)

Figure 7. Fatigue test results for batch IV material.

62 -
20

Χ R = ­ 1.0 WET
95% R = ­ 1 . 0 WET

— 95% R = ­ 1 . 0 DRY

χ χ ""**­■( χ

I , I il I I ι ιιιιι! ι nJ ι I _l ■ , ι ι , ι

6 7 8

«­ log Ν (CYCLESI

Figure 8. Fatigue test results for batch IV material.

Figure 9. Batch IV specimen tested at R = 0.1.

­ 63 ­
Figure 10. Batch IV specimen tested at R = - 1 .

1.0
min
(%) 0.8

-0.6

(14
X TEST DATA ± 1 0 , R=10
95% PROBABILITY OF SURVIVAL
-0.2

, ι ι "Ι ι ι ι ι ) ι 11! Ι ι I I I I III ι ι! Ι Ι ι ι ι I I ι! ι ι ι ι ι 111!


1 6 7
Ina Ν (CYCLES)

Figure 1 1 . Results of the constant amplitude tests for ±10° lay-up (R = 10).

- 64
­2.Or

­1.8­ ­
min
(%) ­1.6­
**^κ X X
­1.4 ■

­1.2 ■ _ ^*^^­­^^x X X

2 RESULTS
­1.0­ ­
­0.8­ _ ^"**­· X X
x—
­0.6­ ­ TESTDATA X F
A ILURE 1 0 ^ R . „ ,
> X­* X­*­
X— NO FA ILURE J
­0.4­
9 5 % PROBA BILITY OF SURVIVA L

­0.2­ ­
I ι ' ' I , , , , ml , , ι ι

► log N (CYCLES)

Figure 12. Results of the constant amplitude tests for 07+45° lay-up (R = 10).

-2.0

-1.8
mm

-1.6

-1.4

-1.2

-1.0

-0.8
κ X X
-0.6
TEST DATA χ FAILURE
± 4 5 I, R = - 1 . 0
X— NO FAILURE
-0.4
9 5 % PROBABILITY OF SURVIVAL

-0,2

0 ι ι ι ι ill 1—ι—ι ι l i n i _l I J 1—ι ι ι ι ml 1—ι—ι ι ι n il ι ' I


' I 5 6 7

► log N (CYCLES)

Figure 13. Results of the constant amplitude tests for 0 7-45 lay-up (R = -1).

65
Figure 14. Example of failed ±10° specimen (R = 10).

Figure 15. Example of failed 07±45° specimen (R = 10).

66
2.0

e
max <%>
χ · χ

XX · ·

• Χ · XX · · — ·—

IO

• WISPER
x WISPX

η \ Ι Ι Ι Ι Ι Ι ι il Ι Ι Ι .1 ι Ι Ι Ι Ι J | Ι Ι 1 1 1 M i l l i 1 1 1 l i l i l í 1 1 1 Ι Ι Ι ΙΙ
2 3 4 5 6 7

-*- log Ν (CYCLES)

Figure 16. Results of the W I S P E R a n d W I S P E R X tests.

67
Tl
IQ'

3
TJ 2.0
Dl
-t
ui
o
PALMGREN - MINER RESULT
Cmax <%' FOR WISPER AND WISPX

co
-*--_
TJ
m
30
ω
3
Q.


TJ
m 1.0
Jj
χ
3
«1
c • WISPER
X WISPX

MEAN
95% PROBABILITY
OF SURVIVAL

OL- _l I ι ι 11 ι II II i n i I I I I I I I ll ι ι ιιιιιι _1 I l_
-4 -2 1 2 3

log (NUMBER OF PASSES THROUGH SPECTRUM)


Tl
ta

ro
pa no

CO

+ +
Φ
£* xoo
ui loo
«1
α

RELATIVE Δ
STIFFNESS +
o
ο α
(%)
i

i 90

ro
3
vQ

80
CO
TJ
m

S 70 WISPX
max
LD
C
ro Ο 53T3
* * ■

(D x 75TB
UI
Δ 75T3 5 SPECIMENS
60 D 73T3
+ 75T4

50­
_l I I I ' I _l I I I l l i l _i ι ι ι ι ι ι ι I
0.01 0.1 10 100
FRACTION OF LIFE (%)
IQ
C

O
o 2.0-
3
TJ
Bl
-i
ui
(%)
TJ
II

ro
V)
1.0-
c
ui

TEST RESULTS R = - 1 . 0
HAGG AND V A N D E L F T
o (1989) EQUATION 3.2
3

Ol
ta
(Q
_j ι I _l I I I I I II I I I I I I I I I I I I I I ll I I ' I I I I ll

<
0) log Ν (CYCLES)
3
D
ro
*
IO
2.1.3 Risø Investigations of GI­UP Materials
(S.I. Andersen, P. Brøndsted, H. Lilholt, Aa. Lystrup)

2.1.3.1 General
For design calculations and lifetime prediction for wing blades of horizontal axis wind tur­
bines, there is a need for data on the long time fatigue properties of the most commonly used
materials. One group of these is glass fibre reinforced polyester. The philosophy of the pro­
gramme is to study selected materials and properties which are considered to be repre­
sentative of current materials and loadings for wind turbine wing blades. Well defined mate­
rials are used, with volume fractions of 50% fibres and fibre orientations of 0°, 10°, 45°, 60°
and combinations thereof. Fatigue testing is carried out under constant load conditions, with
up to 10e cycles. Results for these materials will be discussed in relation to fatigue curves,
change in stiffness during fatigue, models for damage development and property changes,
and the possible design considerations which can be derived from these results.

2.1.3.2 Materials
Glass fibre reinforced polyester is selected with a volume fraction of fibres of about 50% and
fibre orientations including 0° and angle plied orientations, as well as combinations. The fol­
lowing well­defined microstructure configurations are selected, so that a systematic study can
be made of the effect of microstructure, especially fibre orientation, on fatigue performance:
0°, +10°, +45°, +60°, and 07+45°. The data are listed in Table 2 on page 82.
The damage and cracks developed during fatigue loading are important microstructural
changes, and in order to study these easily, the material is made transparent to light by using
a polyester with a refractive index similar to that of the glass fibres. The specifications of the
polyester and the glass fibres are listed in Table 1 on page 82.
Furthermore the angle ply materials must not contain a fibre crossover point between the
+Θ plies and the ­Θ plies; this is achieved by a winding technique where a complete ply is
laid down first and then followed by a second ply of the next orientation. The angle ply
materials must also be symmetrical laminates in order to avoid couplings between different
deformation modes (e.g. tension, shear and bending).
The 0°, ±10°, ±45° and ±60° materials were fabricated on a filament winding machine where
the fibres are laid down on a flat plate­mandrel by wet winding (Figure 1 on page 88). This
method fulfils the above requirements, and two laminates of 2 mm thickness are produced
in each winding operation. The (cold) curing of the polyester matrix is done in an autoclave
under low pressure, where the laminate is consolidated between two steel plates to control
the thickness and fibre content of the cured laminate. The 07+45° materials are fabricated
along the same principles as the simple angle­ply laminates. The procedure is somewhat
more complex, because three fibre directions have to be laid down, and because the plate­
mandrel must be remounted in the winding machine for each fibre orientation.

The laminates are characterised with respect to microstructure and (static) mechanical
properties (stiffness, strength and ductility). The characteristics are listed in Table 2 on page
82; the mechanical properties refer to the directions indicated in Figure 2 on page 89.

2.1.3.3 Specimens
From the laminate plates, specimens of dog­bone shape are cut with a narrow, central gauge
section, and tabs at the ends for gripping in the fixtures of the fatigue machine. For angle
ply laminates there is a potential risk that early damage is initiated at the free edges of the
specimens; this may not occur (as early) in a large specimen or in a component with no

- 71 -
edges. Fatigue testing is continued until excessive damage, but not necessarily complete
separation of the specimen, has occurred.
The specimen geometry is illustrated in Figure 3 on page 89.

2.1.3.4 Fatigue testing


Frequency and wave form
Fatigue testing was performed at constant amplitude loading, with a few additional tests with
the variable amplitude spectrum WISPER.
All tests were made with constant frequency of 5 Hz and with constant loading rate (strain
rate), so that the wave form of the fatigue loading was "triangular", as illustrated in the sketch
in Figure 4 on page 90. The applied load range (omax - omin = Δσ) or the corresponding strain
range (Δε) is related to frequency and (load) strain rate by the equation:

έ = Δε·2/ (1)
This implies that a test series at constant f and various Δε gives a different strain (load) rate
in each test, with the highest έ at the largest Δε; this is illustrated in Figure 4 on page 90.
An alternative fatigue loading strategy would be to apply the same strain (load) rate at all
strain ranges Δε, and therefore perform the individual tests in a test series with a different
frequency f according to Equation (1).
The second strategy is perhaps the most fundamental, because material properties (strength)
are normally dependent on strain rate, with higher strengths at higher strain rates. A detailed
experimental and analytical discussion has been given by Sims and Gladman (1978,1980)
[1],[2]. A sketch of the situation is given in the S-N diagram in Figure 5 on page 91.

In the present project the tests are performed at constant frequency for the following reasons:
• most other data (from literature and reports) are based on constant frequency perform­
ance;
• the difference between the two strategies is relatively small, in particular in comparison
with the general experimental error of fatigue testing;
• the external loads imposed on the material (wing blades) are more likely to be such that
a (nearly) constant frequency is seen by the material.

Loading types
The loading on wing blades brings the material into tension-tension fatigue and compresi-
son-compression fatigue, with some cases of tension-compression fatigue. The loading type
is described by the parameter R:

Tests are made with the following loading types:


• R = 0.1 (tension-tension)
• R = 10 (compression-compression)
• R = -1 (tension-compression)
• WISPER (spectrum)
The load-time patterns are shown schematically in Figure 6 on page 91.

72
Temperature
During fatigue loading the polyester material will experience irreversible, visco­elastic defor­
mation, which leads to dissipation of energy in the material (hysteresis effect). Therefore the
polyester, and thus the composite, will heat up during fatigue. This heat is conducted out of
the material; hysteresis heat production and heat conduction will, together with external
parameters, control the resulting permanent temperature difference between the centre of the
material (specimen) and the surface of the material.

The external parameters are the thickness t of the material (specimen), the (maximum) load
(stress), and the frequency of fatigue loading; the temperature difference is approximately
proportional to these external parameters:
Δ7~= constant- f ■ σ · f
The value of ΔΤ is fixed for constant parameters; this means that in practice a reasonably low
value of f is required. Measurements of the temperature in the middle of the material and on
the surface of the material have shown that ΔΤ is 2­5°C for f in the range 5­10 Hz.
It should be noted that ΔΤ is permanent, but the absolute values of temperature (middle and
surface of material) can be kept low, e.g. by simple air cooling the surface of the specimen,
such that the surface temperature is, e.g., 20°C and thus the middle temperature will be
24­25°C.
In the present tests the frequency is 5 Hz without cooling. This implies ΔΤ = 2­5°C, and the
average temperature of the material (specimen) is thus about 22­23°C, which is taken as the
(reference) temperature for (all) tests.

Controls during fatigue testing


External wind loading on wing blades imposes certain loads (load patterns) on the material.
Fatigue testing is therefore performed under load control on the testing machine; this means
that maximum load and minimum load are the same during the whole test period for a
specimen. Because the material (normally) suffers a reduction in stiffness, this means that
the displacements (strains) are (slowly) increasing during the test period.

The alternative procedure of using displacement control is considered less realistic for
materials intended for wing blades.

Test machine and facilities


Fatigue testing is performed on a servo­hydraulic test machine with the following specifica­
tions:
Type: Instron 1332
Max. load: 100 kN
The grips for the specimens can accommodate the flat specimen shown in Figure 3 on page
89, and their tapered sections provide a constraint from the edges of the specimen, so that
the clamping pressure does not crush the gripping ends of the specimen. The tapered section
also provides a positive self­tightening of the grips onto the specimen. The geometry and
appearance of the grips are shown in Figure 7 on page 92 The experimental arrangement
in the test machine is shown in Figure 8 on page 93.
The antibuckling device is designed to support the specimen during compressive loading (R
= 10; R = ­1), so that the specimen does not buckle (elastically) out of the flat plane. This
device allows the use of the same type and size of specimens as in testing with (only) tensile
loads.

73
The alternative method would be to use (very) "fat" specimens so that the specimen geom­
etry itself discourages buckling. This will in practice mean that the overall size of the speci­
men will be (much) smaller than the specimens for tension loading. This method therefore
has the disadvantage that a (much) smaller volume of material would be tested in com­
pression loadings than in tension loading. This could have an effect on the statistics of crack
nucleation and growth, not to mention the practical inconvenience of using different types of
specimens.
The antibuckling device has the disadvantage of (possible) friction between the specimen
and the supporting (steel) plates; in view of the above considerations this is acceptable.
The grips are modified for compression loading by inserts which fix the end edges of the
specimen against slipping "out of the grips" during the compression part of the cycle.
The grips and the antibuckling device are shown in Figure 9 on page 94 to Figure 11 on
page 96.
The load during fatigue testing is recorded by a conventional load cell.
The strain during fatigue testing is recorded by strain gauge extensometers, with one
mounted at each side of the specimen. This allows compensation for any (unintentional)
bending of the specimen.

Test conditions
The load control settings are established by first measuring the (start) modulus E0 of the
composite, and then calculating the loads (max. and min.) from the equation
P=E0-z0.A

via the chosen strains (max. and min.) and the chosen R-value:

The strains are the initial values and they will change (increase) during testing. The strains
are recorded from the strain gauge extensometers.
Two material parameters are recorded during testing. The (material) modulus E (with start
value E0) is measured as the initial slope of the stress-strain curve. The secant modulus (with
start value E0) is measured from ΔΡ and Δε as
AP
E -
C e - -Α -Δε
where A is the cross-sectional area of the specimen.
During testing E and Es are measured/recorded at every tenth of a decade of cyclic loading,
where a static tensile test to low strain is performed to measure E.
The number of cyclic loadings Ν is recorded during testing. The test is stopped (automat­
ically) when the displacements (strains) during cycling have reached preset limits; this nor­
mally corresponds to "excessive damage" in the material, but not necessarily to complete
separation of the specimen into two (or more) parts.
All experimental data are collected in a computer for further analysis and presentation.
The damage in the material during testing is (for some specimens) recorded by a video
camera. The damage is cracks and delaminations, and appears as "whitening" in the trans­
parent specimens. The video recording allows a replay of the sequence of damage, from
initiation to final failure.

74 -
2.1.3.5 Results presentation
The fatigue results are presented in S-N diagrams with a replacement of S (stress) by ε
(strain); the maximum, initial strain is used in the diagrams. The values of E and Es are
plotted versus log N. The values Nmax refer to "failure".
The type of diagrams are:

t vs log Ν
E vs log Ν
Es vs log Ν
E/E0 vs log Ν
Es/Eo vs log Ν
E/E0 vs log N/ log N,
Es/Eo vs log N/ log N,

2.1.3.6 Results overview


The glass-polyester composite materials which were tested are listed in Table 3 on page
82.

2.1.3.7 Results for ε-Ν diagrams


The data for all materials and all loading conditions are collected in Table 4 on page 83 to
Table 10 on page 86 and presented in Figure 12 on page 97 to Figure 18 on page 103.
Most tests were made at R = 0.1 and several comparisons and observations will be made
on the basis of the results. The loadings at R = 10, R = -1 and WISPER-spectrum will be
compared to the data for R = 0.1.
In all ε-Ν diagrams the regression line for the data for the 0° laminate at R = 0.1 will be
reproduced to serve as a reference for comparison.

R = 0.1, comparison between laboratories


The results for 0° laminates in Figure 12 on page 97 show the expected trend in the curve
with a possible fatigue limit, at large number of cycles, of ε = 0.8%. The same 0° material
(fabricated at Risø) was also tested at Gent University, Belgium, and the results in
Figure 19 on page 104 show very good agreement between the two sets of results. This
encourages the pooling of data, which makes more detailed analyses possible.

R = 0.1, comparison of all materials


These materials are all made at Risø, as described in 2.1.3.2 on page 71 and are all of high
quality, with especially low porosity. They serve as "perfect" reference materials, to be used
in, e.g., evaluating industrially made glass-polyester laminates. They are collected in
Figure 20 on page 105.
The 0° laminate is the reference material in this comparison with +Θ laminates. The ±10°
laminate gives data a little, but clearly below, the 0° data. The results for +45° laminates and
especially ±60° laminates are well below the 0° reference; this illustrates the significance of
fibre orientation, and in particular the detrimental effect of large off-axis angles.

75 -
The 07±45° laminate is one (of many) combination of 0° fibres and (some) off-axis oriented
fibres. The data points are very close to the 0° reference, possibly a little below at large
number of cycles. This is qualitatively not surprising, because the 0° layers are carrying the
major part of the (total) load, and the data are plotted as strain. The total stress carried by
the 07±45° laminate is, of course, smaller than the stress carried by the 0° laminate, because
the modulus is lower; E (07+45°) = 29.7 GPa and E (0°) = 47.2 GPa, as listed in Table 2
on page 82.

R = 0.1, comparison between "Risø" perfect material and "Nibe" material


In Figure 21 on page 106 a comparison of the "Risø" perfect laminates is made with existing
data on the "Nibe" material with approximately ±5° fibre orientation and lower quality matrix
(more porosity). A trend of similarity is seen, with possible fatigue limits.
It is, however, clear that the +5° material has lower strain values than the ±10° material. This
(unexpected) situation is probably caused by the lower quality of the ±5° material: about 9%
of the total amount of fibres are at right angle to most of the fibres (i.e. at about ±85° orien­
tation); the fibres have crossover points (originating from the woven tape used in fabrication);
and most fibres (at ±5°) have a length of 20 cm, so that fibre ends are present in the material.
An important parameter causing lower fatigue properties is the larger porosity content;
porosity in the material can be considered as pre-existing "cracks" present at the start of the
test.

R = 10 and R = 0.1 compared


The data for ±10° laminates and ±45° laminates are shown in Figure 22 on page 107 (with
0° reference), for the two R-values. It is clear that ±10° laminates have higher fatigue strains
in tension-tension than in compression-compression, while the opposite is found for ±45°
laminates. The result for +10° laminates is in qualitative agreement with existing data from
other studies (although it is not clear how the compression fatigue tests have been per­
formed, and in particular, what risk is involved for premature failure because of the possibility
of buckling). The ±45° laminate results are unexpected, although no other data seem to be
available for comparison.

In practice the ±10° laminates and their performance are more relevant, and therefore sup­
port the need for lower strain values for compression than for tension.

R = -1, R =10 and R = 0.1 compared


These three R-values are compared for the ±10° laminates in Figure 23 on page 108 (with
0° reference). The single data point for R = -1 is comparable to the data line for R = 10, or
perhaps a little below the line.
It is not possible to illustrate the R = -1 situation further.

WISPER spectrum and R = 0.1 compared


The comparison of spectrum loading with other data is made for the 07±45° laminate in
Figure 24 on page 109. The data for the WISPER spectrum are placed higher than the data
line for constant amplitude loading. This may be a real effect, but it should be noted that the
strain used for the strain axis is the absolute maximum amplitude reached in the period of
the spectrum (which contains 132,711 individual cycles); this is probably a too "optimistic"
way of plotting, since most of the cycles have lower amplitudes.
It is not possible to evaluate the use of a (WISPER) spectrum further.

- 76
Fatigue curve analysis
The ε­Ν curves for "Risø" perfect materials and "Nibe" material all show a possible fatigue
limit and a (nearly) constant slope for the decreasing part of the curve. A preliminary analysis
is made by estimating these parameters, as shown in Figure 25 on page 110 and Table 11
on page 86. It is clear that the identification of a fatigue limit is difficult because of the rather
few data available for a very large number of cycles. The slopes are reasonably well defined
when many data support the line.
The parameters are plotted for ±Θ laminates in Figure 26 on page 111 ; there is no clear
pattern for all data, although the fatigue limit decreases with increasing Θ for R = 0.1 data
on Risø laminates.
This is generally expected, on the basis of the trend in strength of ±Θ laminates.
The (two) data points for R = 10 do not give a clear picture in relation to the data for R = 0.1.
The "Nibe" material (of lower quality) shows a lower fatigue limit; this is expected qualitatively.
The numerically large slope is perhaps also to be expected because of the lower quality
(higher porosity content).
The combined 07+45° laminate is compared to the pure 0° and ±45° laminates in Figure 27
on page 112. No firm analysis can be made, except that a "simple mixture" of the two "limits"
may describe the behaviour of the 07±45° laminate.
It should be noted that a (more) accurate description of the fatigue limit and the slope of the
ε­Ν curve will improve the possibilities for identification of relevant design parameters relating
to, in particular, very long lifetimes.

2.1.3.8 Results for E and Es


Normalization
The data for stiffness E and Es can conveniently be normalized by the initial modulus E0, and
further plotted versus the lifetime normalized as log N/ log Nmax.
The recorded changes (reductions) in stiffness are plotted in a normalized form in Figure 28
on page 113 for Ε­modulus and in Figure 29 on page 113 for secant­modulus; these data
are for the ±10° material. The superposition of curves for two strain levels indicates the
similarity in damage development and the related stiffness reduction. This is useful because
master curves may be established, and thus reduce the need for extensive, individual testing.

Material modulus E
For the two glass­polyester materials of ±10° and +60° fibre orientation the values of E/E0
are shown in Figure 30 and Figure 31 on page 114, each at three different levels of fatigue
loading, measured as the maximum (initial) strain of the load cycle. The normalized plot
allows a direct comparison of the three strain levels, even if the corresponding lifetimes are
very different.
The strain levels for the two materials are rather different, but the fatigue curves of
Figure 25 on page 110 can be used to illustrate a correspondence in terms of lifetimes, as
shown in Table 12 on page 87:
The shapes of the E/E0 curves are very similar for the two materials at the same strain level,
as defined in Table 12 on page 87.
A possible interpretation of the shapes of the E/E0 curves at the three strain levels corre­
sponding to short, medium and long (infinite) lifetimes, is the following:

­ 77
• At high strains, rather few cracks can develop and grow before a crack configuration is
formed which results in (final) failure. The lifetime is short and the stiffness reduction is
moderate.

• At medium strains, many cracks can form and grow before a critical configuration is
formed. The lifetime is relatively long and the stiffness reduction is significant, with E
decreasing to a level of 70-75% of E0.
• At low strains, which are close to or perhaps below a fatigue limit, relatively few cracks
will form, and their growth is restricted. The lifetime is very long (infinite) and the stiffness
reduction is small and stabilizes at a certain level towards the end of life. At present it
is not clear why the ±60° laminate shows an increase in stiffness before stabilization.

Material modulus E and secant modulus Es


To study the general shape of the stress-strain curve of glass-polyester after (the same)
fatigue loading, a detailed comparison between E and Es has been made for the ±10° lami­
nate at medium strain level. Two strain levels of 0.8% and 1.0% are used in Figure 32 on
page 115, for the normalized plots of E and Es, respectively.
At this medium strain level the reduction in stiffness is generally large, and a comparison
between E and Es shows a tendency, which is enhanced schematically in Figure 33 on page
116. Three regions can be identified and their characteristics are listed in Table 13 on page
87.
The damage mechanisms which can be responsible for the observed changes in stiffness
parameters E and Es are based on the following type of cracks (Figure 34 on page 116) and
their implied effect on the deformation behaviour of the composite material:
• Matrix cracks parallel to the loading direction and fibre direction (+10°) will not lead to
(measurable) changes in E, but may cause some increase in the significance of the
matrix visco-elastic behaviour on the (overall) deformation of the composite, i.e. Es <
E.
• Debonding (nearly) parallel to the loading direction and at the interface between fibres
and matrix can have two effects (Figure 34 on page 116):
debonding will most likely start at the fibre ends, causing the effective fibre length
for load transfer to decrease; this will result in a (small) reduction in E.
the significance of the matrix visco-elastic behaviour on composite deformation will
increase, i.e. Es < E.
• Matrix cracks at right angles to the loading direction and the fibre direction (±10°) will
lead to a reduction in E, but not to any change (increase) in the matrix visco-elastic
contribution, i.e. Es= E ( < E0). There may be a contribution to composite deformation
from the (matrix) crack opening, and this will in principle mean that Es < E.
On the basis of these possible types of cracks in glass-polyester, an interpretation of the
changes observed in E and Es can be given; it is presented in Table 14 on page 87.
A possible identification of likely mechanisms of crack formation during the (normalized)
fatigue life can therefore be the following, based on the glass-polyester ±10° laminate:
1. At short lifetimes, log N/ log Nmaii 0.4, very few, if any, cracks have formed, and no
(measurable) changes in stiffness can be recorded.
2. At medium lifetimes, log N/ log Nmax between 0.4 and 0.7, cracks at right angles to the
loading direction develop, and they lead to reduction in E, but to no significant change
in the matrix visco-elastic contribution, so that Es = E.

78 -
3. At long lifetimes, log N/ log Nmax between 0.7 and 1.0, additional cracking occurs parallel
to the loading direction (matrix cracks and/or debonding), so that the matrix visco-elastic
contribution becomes more significant, and thus Es < E.
The (schematic) stress-strain curves corresponding to this interpretation of cracks and
deformation mechanisms are shown in Figure 35 on page 117. This may be a simplified
picture, because interactions can occur between the different types of cracks, with respect
to their formation, number and growth.

2.1.3.9 Damage development


Damage in glass-polyester composite materials has been followed by video recording of
selected specimens during fatigue testing.
The materials were ±10° laminates, ±60° laminates and 07±45° laminates. The loading
conditions were R = 0.1. The individual specimens are marked in Figure 37 on page 118 to
show the conditions of strain and number of cycles at the end of the observation period.
A general observation for all glass-polyester materials is that a fairly large amount of damage
occurs before the end of the observation period, although the three materials behave differ­
ently. Another observation is that some cracks start at the edge of the specimen, while others
start internally in the specimens.
The types of damage observed are edge cracks and delaminations, sectional delaminations
and general (often internal) damage.
The characteristics of the three types of glass-polyester laminates are:
• ±10° laminates:
- early life: edge cracks along fibre directions and delaminations
- middle life: sectional delaminations
- final life: general damage in middle of specimen
• ±60° laminates:
- early life: no damage
middle life: no damage
final life: local cracks along fibre directions, starting inside specimen
• 07±45° laminates:
- early life: some longitudinal splitting along 0° fibre direction
middle life: very little or no cracks and delaminations along ±45° fibre direction;
longitudinal splitting
final life: longitudinal splitting

A general comparison is that the "high strain" laminates (±10° and 07±45°) show early start
of damage, with widespread damage at the end of life, while the "low strain" laminate
(±60°) shows no early damage and only little and localised damage at the end of life. The
damage is shown in Figure 36 on page 117 and Figure 38 on page 119.
Further observations are required before more detailed statements can be made about
damage development.

2.1.3.10 Design limits


The fatigue data for all materials in the ε-Ν diagram (Figure 20 on page 105) form the basis
for the following tasks:
• establishment of design limits,

79
• improvement in material performance,
• comparison of industrial materials with the reference (perfect) materials
The stiffness data in the Ε-N and Es-N diagrams (Figure 28 to Figure 32) form the basis for
the following tasks:
• establishment of design limits defined by stiffness,
• estimate of strength and deformation of materials after fatigue,
• estimate of internal damage to the microstructure.
The master diagram for ε-Ν data (Figure 20 on page 105) shows a general similarity in the
curves for all materials. This allows new and/or improved materials to be "tested" against this
basis. Rather few fatigue tests are probably required to establish the "position" of the ε-Ν
curve for a potential (new) material; the new curve is expected to follow the trend of the
existing family, i.e. no crossover points, at least for the same R-value.

The master diagram of ε-Ν data can also be used to evaluate an industrially produced
material against the existing family of curves; the "Nibe" material is such a material and the
comparison was made in Figure 21 on page 106.
Generally the results in Figure 28 to Figure 32 indicate a significant reduction in stiffness
before (final) failure. This emphasises the need for defining acceptable stiffness values in
design, and, following this, the need for fatigue curves for the corresponding, acceptable
stiffness reduction.
The stiffness changes can be used to construct ε-Ν curves referring to a given, acceptable
stiffness reduction during fatigue, rather than the (conventional) reference to "failure" as cri­
terion for the ε-Ν curve.

The design philosophy for glass-polyester materials can be based on cracks (damage) in the
material, on a certain acceptable reduction in stiffness, or on a certain acceptable reduction
in strength of the material.
The damage concept implies a certain level of damage (cracks, delaminations), which then
defines sufficient (residual) stiffness and strength. The damage will also be important in
relation to the risk of environmental attack on the material.
The stiffness concept defines a sufficient stiffness of the material (without specific reference
to the level of damage).
The strength concept defines a sufficient (residual) strength (without specific reference to the
level of damage); the strength should be sufficient for the material to survive predictable
overloading (e.g. gale conditions).
The most fundamental concept is the damage idea, because it governs the mechanical
properties. A possible requirement of no damage would be very safe, but also very con­
servative. This would place the design limit below the (possible) fatigue limit.
At present the recommended design limits in Denmark are strains less than 0.3% for tension
loading and less than 0.2% for compressive loading. These limits are probably conservative
in the sense that they refer to strain levels at which no cracks are likely to form. The design
limits are not related to specific materials (in terms of volume fraction and fibre orientation),
and are derived (initially) from the data in Figure 21 on page 106 for the ±5° material.
Materials with most fibres oriented close to 0° are common at present.
The design limits for no cracks should also ensure only a minimum effect of (water) corrosion
of the material.

80
2.1.3.11 Conclusions
The project has yielded the following results and information:
Composite materials of glass fibre reinforced polyester were fabricated with well-defined
microstructure and high quality, especially low porosity content.
Fatigue data for these glass/polyester materials were obtained and presented in ε-Ν
diagrams as a family of curves (master diagrams).
The ε-Ν curves are similar (i.e. no cross-over points) for R = 0.1.
The data for R = 10 (compression-fatigue) are close to data for R = 0.1 (tension-fatigue),
with some differences.
Industrial glass-polyester material ("Nibe") fits into the family of ε-Ν curves, but at a lower
level than expected (probably due to lower material quality).
Stiffness changes were measured during fatigue, and can form the basis for establishing
of design limits related to material-stiffness.
Stiffness changes can be used to estimate the stress-strain curve and strength of the
materials after fatigue.
Stiffness changes were used to evaluate the possible development of cracks and
delaminations in the materials during fatigue.
Microstructural changes (damage) in the form of cracks and delaminations were
recorded by video camera to establish the time sequence of damage development.
Design philosophies and design limits were discussed.

2.1.3.12 Bibliography
[1 ] G.D. Sims, D.G. Gladman Effect of test conditions on the fatigue strength of a glass-fabric lami­
nate. Part A - Frequency Plastics and Rubber: Materials and Applications, May 1978, 41 -48
[2] G.D. Sims, D.G. Gladman Effect of test conditions on the fatigue strength of a glass-fabric lami­
nate: Pan Β - Specimen conditions Plastics and Rubber: Materials and Applications, August 1980,
122-128.

81 -
2.1.3.13 Tables

Glass-fibres: type: E
300 filamentsAow
code: RPA 3820/21
supplier: Scandinavian Glassfibres
Polyester: type: UP 333
additional styrene: 10%
accelerator: 0.7% N49P
curing agent: 1.2% Methyl ethyl keton peroxide
forsinker: NLC 10: 0.7 g pr. 100 g
supplier: Hoechst/Polyplex

Table 1. Glass fibres and polyester.

Composite material: 0° ±10° ±45° ±60° 0°/±45°


No. of plies 8 8 8 8 10, with
stacking
sequence,
[0°2, ±45°,
0°ls
Vol.fraction of qlass% 56.6 51.1 - 58.1 55.0
Porosity content % 0.51 0.35 - 0.38 0.91
Thickness of laminate mm 1.9 -2.0 -2.0 -2.0 -2.3
Mechanical properties:
Ει GPa 47.2 40.9 16.5 29.7
E2GPa 14.7
V12 0.31
Ol MPa 862 546 55 655
02MPa 43
ει % 2.3 1.5 0.5 2.7
82% 0.3

Table 2. Laminates and their characteristics.

Materials Fatigue load ratio R


0.1 10 -1 WISPER
ε-Ν E&Es ε-Ν E&Es ε-Ν E&Es ε-Ν E&Es
"Risø" 0° χ
±10° χ χ χ χ
±45° χ χ
±60° χ
0°/±45° χ χ
"Nibe"-±5° χ

Table 3. Fatigue testing conditions and measurements for all laminates.

- 82
PROJECT : Fat igue Glass/Polyeste r R = 0.1
DATE 13 Dee 1990 V f = 52.8%
SPECIMEN 0 0 (unidirectional) porosity = 0.83%
File : FAT -OG.tab
Spec. no. load level load type Fracture max
[strain](%) R yes/no N

0°:
111-34-20-11 2.104 STATIC yes 1
111-34-20-12 2.56 STATIC yes 1
111-34-20-5 .9 .1 no 8300000
111-34-20-6 1 .1 yes 5209100
111-34-20-7 1.5 .1 yes 36800
111-34-20-8 1.2 .1 yes 295000
111-34-19-1 1.1 .1 yes 1471000
111-34-19-2 1.15 .1 yes 231000
111-34-29-2 1 .1 yes 102801
111-34-29-9 1.2 .1 yes 84267
111-34-29-5 1.07 .1 yes 969138
111-34-29-12--1 1.1 .1 yes 584531
111-34-29-13--1 1.5 .1 yes 489
111-34-30-10--1 1.7 .1 yes 2360.9
111-34-29-11--1 2.1 .1 yes 57.5

Table 4. Fatigue data for 0° laminate at R = 0.1.

PROJECT Fatigue Glass/Polyeste r V f = 49.1%


DATE 13 Dec L990 porosity = 0.20%
SPECIMEN 10 · R== 0.1
File FAT-10G .tab
Spec. no. load level load type Fracture max
[strain](%) R yes/no N
10° :
111-34-1 10° 1.58 static yes 1
111-34-2 10° 1.44 static yes 1
111-34-4 10° 1.41 static yes 1
111-34-34-8 10' 1 .1 yes 33523
111-34-34-6 10° .8 .1 no 411690
111-34-34-12 10° .6 .1 no. 11496625
111-34-34-11 10° .8 .1 yes 184401
111-34-34-10 10° 1 .1 yes 37101
111-34-34-7 10° .7 .1 yes 594308
111-34-34-5 10° .9 .1 yes 40072
111-34-36-8 10° 1.4 .1 yes 131
111-34-36-7 10° 1.3 .1 yes 1448
111-34-36-6 10° 1.2 .1 yes 1465
111-34-36-5 10° 1.2 .1 yes 2628
111-34-36-4 10° 1.1 .1 yes 3376
111-34-36-2 10° 1 .1 yes 24454
111-34-36-1 10° 1.1 .1 yes 3001
111-34-36-10 10° .6 .1 yes 332801

Table 5. Fatigue data for ±10° laminate at R = 0.1.

83
PROJECT : Fatigue 3lass/Poly ester
SPECIMEN: ±45 ° R =0.1
File : FAT-4501 .tab

Spec. no. load level load type Fracture max


[strain]( %) R yes/no N

±45
111-34-92-22 ±45 .28 .1 no 12578204
111-34-95-22 ±45 .63 .1 yes 800
111-34-96-11 ±45 .37 .1 yes 25834
111-34-92-11 ±45 .37 .1 yes 39707
111-34-93-12 ±45 .49 .1 yes 10194
111-34-91-12 ±45 .48 .1 yes 6352
111-34-92-12 ±45 .46 .1 yes 13615
111-34-93-22 ±45 .55 .1 yes 6301
111-34-96-22 ±45 .49 .1 yes 3201
111-34-94-22 ±45 .54 .1 yes 1976
111-34-94-11 ±45 .62 .1 yes 1676
111-34-96-12 ±45 .62 .1 yes 1090
111-34-96-22 ±45 .64 .1 yes 836

Table 6. Fatigue data for +45° laminate at R = 0.1.

PROJECT Fatigue Glass/Polyester V ( = 58.1%


DATE 13 Dec 1990
SPECIMEN 60 ° R=0.1 porosity = U 38%
File FAT-60G .tab
Spec. no. load level load type Fracture max
[strain](%) R yes/no N
60°
111-34-43-1 60° .536 static yes 1
111-34-43-2 60° .4742 static yes 1
111-34-44-4 60° .4536 static yes 1
111-34-43-3 60° .172 .1 yes 376116
111-34-44-1 60° .174 .1 yes 170567
111-34-46-02 60° .153 .1 yes 290438
111-34-43-04 60° .172 .1 yes 130014
111-34-46-01 60° .312 .1 yes 35
111-34-47-02 60° .279 .1 yes 401
111-34-47-01 60° .216 .1 yes 10104
111-34-42-03 60° .187 .1 yes 38146
111-34-45-03 60° .216 .1 yes 15081
111-34-44-03 60° .1 .1 no 10352408
111-34-42-4 60° .18 .1 no. 61426

Table 7. Fatigue data for ±60° laminate at R = 0.1.

84
PROJECT : Fatigue Glass/Polyeste r R = 0.1
DATE l-dec-1990 V f = 55.0%
SPECIMEN 45°/0°
File FAT45-0 tab porosity = 0.91%

Spec. no. load level load typ e Fracture max


[strain](%) R yes/no N

±45/0°
111-34-56-15 ±45/0° .916 .1 1052534
111-34-56-16 ±45/0° .669 .1 no 12097602
111-34-58-02 ±45/0° .86 .1 1388367
111-34-61-02 ±45/0° .828 .1 5796801
111-34-64-01 ±45/0° .919 .1 2937514
111-34-58-01 ±45/0° .8 . .1 obs! 7964865
111-34-55-st +45/0° 2.59 static 1
111-34-54-st ±45/0° 2.708 static 1
111-34-65-01 ±45/0° 1.3 .1 yes 94007
111-34-65-02 ±45/0° 1.28 .1 yes 54741
111-34-67-01 ±45/0° 1.27 .1 yes 154801
111-34-67-02 ±45/0° 1.42 .1 yes 35801
111-34-67-02 ±45/0° 1.5 .1 yes 19381

Table 8. Fatigue data for 07145° laminate at R = 0.1.

PROJECT Fatigue Glass/Polyester V f = 52. 7%


DATE 13 Dec 1990
SPECIMEN ± 10° R=10 porosity — 0.55%
File FAT-10CG.tab
Spec, no load level load type Fracture max
[strain](%) R yes/no Ν

±10
111-34-4C -7 ±10 .5 10 no 4849403
111-34-4C -5 ±10 .7 10 yes 13746
111-34-4] -6 ±10 .8 10 yes 16886
lll-34-3£ -10 ±10 .9 10 yes 8947

Table 9. Fatigue data for ±10° laminate at R = 10.

85
PROJECT : Fatigue Glass/Polyester
DATE 13 Dec 1990
SPECIMEN ±45 ° R=10
File FAT-4510.tab

Spec. no. load level load type Fracture max


[strain](% ) S yes/no N

±45
111-34-95-12 ±45 .6 10 yes 50779
111-34-94-2 ±45 1 10 yes 3883
111-34-95-1 ±45 1 10 - 3616
111-34-90-21 ±45 .3 10 no 15035801
111-34-93-2 ±45 .5 10 yes 632323
111-34-90-11 ±45 .5 10 yes 318790
111-34-90-1 ±45 .8 10 yes 27316
111-34-89-1 ±45 .6 10 yes 96171
111-34-91-2 ±45 .8 10 yes 18752

Table 10. Fatigue data for ±45° laminate at R = 10.

Material R 8fatlim. slope (ε-log N)


% %/decade
0° 0.1 0.86 -0.25
±10° 0.1 0.66 -0.15
±45° 0.1 0.27 -0.13
±60° 0.1 0.08 -0.060

07±45° 0.1 0.71 -0.27

±10° 10 0.51 -0.11


±45° 10 0.29 -0.21

-±5° (Nibe) 0.1 0.44 -0.19

Table 11. Data for (possible) fatigue limit and slope of ε-Ν curves.

86
Material
strain level ±10° ±60° log Nmax (tip. 25)
"hiqh" 1.18% 0.32% -3.2
"medium" 0.93% 0.20% -4.6
"low" 0.58% 0.10% >7

Table 12. Correspondence of lifetimes for ±10° and ±60° laminates.

Region Stiffness Lifetime


log N/loq Nmax
a Es=E=Eo 0-0.4
b Es=E<Eo 0.4 - 0.7
c Es<E<Eo 0.7-1.0

Table 13. Stiffness and lifetimes during fatigue.

Region Crack types (fiq. 34)


a none
b 3
c 1 and 2

Table 14. Relation between stiffness curves and cracks (damage).

87
2.1.3.14 Figures

Figure 1. Filament winding machine for fabricating glass-polyester laminate plates.

88
Figure 2. Directions in composite laminates

2^ -¿I

zr ~<c
Figure 3. Fatigue specimen dimensions.

- 89
σ

σmax

σ.min
time

f,

min

* time
Figure 4. Fatigue loading sequences; (a) definitions; (b) fatigue with two different (strain)
amplitudes, at constant frequency.

90
f = constant

ε = constant

_ log Ν
Figure 5. Fatigue curves (schematic) for constant frequency and constant strain rate,
respectively.

R = 0.1

R = -l

-> time

R=10
Figure 6. Fatigue loading sequences for R = 0.1, R = 10 and R = -1, respectively.

- 91 -
Specimen holder, upper
f~^^~~"
i ζ

, ,_
m w
© ©
© >

'/.■//'s////>>~

IH.
'Wïêm /,/,. 'y//.

ili '■■/­

%#${·,

Specimen holder, lower

Figure 7. (a) Sketch of grips; (b) grips with specimen.

­ 92
Figure 8. Fatigue testing machine with fixture for gripping specimens and set-up for video
recording of damage development.

93
ta
c
πι
co

ω
F*
O
ZT

ai
3
f*
σ
c
o
¡Γ
5'
ta
a
οι
<
õ'
ni
Figure 10. Antibuckling support plates and specimen.

95 -
Figure 11. Antibuckling device mounted on specimen.

96
0 o - laminate
CQ
C

tu
—k
NI

-Olπ :]
tu' 2.5
c
ro
Q.
Ol

sr
-*
o
■Ί :] ^ ^^ π
o ι 1
o υ 2
1 ί\—'
3
3
C
0)
φ
Π ^n
L.
XI
II W 1.5 ι—1
» ^ 1 I
o
σ
»i

c π
TD
LJ ^ 1——
L­^>

0.5

C) 1
r
> :; 5)
¿ E» 7 8
Log Ν
10 o - laminate
to
c


EU
t5'
C
2.5
φ
α
ω

ι+ 2
ο
ο
&?
ω"
3 Ç
5'
0) 'σ
ι­«·
φ L· c:
0)
*+ W 1.5
Π
30
II "ο π
ο
c ππ
'Ε [Ο
1 II Μ
ο ,
π r-—-,.
5
LI LI l^ *

Π
π 3 O
0.5

0
r ■
c
0 1 ; L > Εi 7 8
±45° laminate
ta 3­
c

ta
c
2.5­
Φ
a.
ω
S

It '—' 2­
ω
3 C
5'
m
Φ 'σ
S
II
co 1.5­
o

c

α DD
π
η□ Rr
DL 0
DD
Q L>

ο­
o 4 8
Log Ν
± 6 0 ° laminaite
ta 3­
c

­π
ai
ta' 2.5­
c
Φ

a
ET ­

S Τ1 0

01
3
5'
c
m Ö
φ
51
33 00 1 .D
II
o ö
c
E
0 H
1
Ζ

Γ\ c: E]
U.D F Γ "

□ Π
C=° πα OD r­jp
[3 O
η —O
υ

r ­; A ei E 8
0/±45° laminate, R=0.1
ta
c

α
τι C]
Ol

~. 2.5
ta'
c
Φ
a
Ol

sr

1+ fe*
J>
en

3
3' o
ai L·
**■
φ (Λ 1.5 ι—Ι
0)
TJ
^-^π
II
o c

't^ö
Q O
0.5

0
c r
) E 8
Γ
0 1 ί Α
Log N
±10° laminates
to
c

+
s? R=10
2.5 ite
α R= 0.1
'
I
ι+
­i.
o ^
o
õ7
3
S'
Dl
.-* α
Φ
Ä
κ
Μ 1.5
TJ
II M >x
—t
o

Ol
3
α Έ VU/
/Π.Ν

TJ
II o 1 ■+
yfviRk ^■s^^

o
+
+
* D>
0.5 1
1
Λ^~-

0
r*
0 1 1 L C
6 7 8
TI
±45° laminate, R= 10
to' 3­
c
5

D?
tö'
c
2.5­
Φ
α
I
Ol
^

ω
3 C
5'
ω 'σ
Φ
m ­ι—
TJ 1.5­
ιι
Ο
C

ο 1­ o
ππ
π π
η ^
α—uzi—

π ^ >

η
c 1 r ι :
; A c
> £ ' 8
Log Ν
0°-laminate, R=0.1
ια
c n
RISØ dita
3
χ Ghent data
ω
3S 1 n
"',?
Q ια 2.5
Zn

δ" a
3 Ol

2 α n
h £
Ξ SJ
o 2. c
gl 'ö Π
D. Ol ι_
ι—I
Φ (/) 1.5 1 1 I I
3 II
c?
3 . "ö
<o c π
Φ o

ai Έ Γ ­1 A>v / I 1
o 1
L J •VN I I

s»m' U^
O X
α
3

I 0.5
0)
^>
X
3
3
αϊ
σ
o

SI
o 0
-ι 0 1
r
' > ι E
> 6 77 8
φ'
tn

Log N
Unidirectional and Angle­ply laminates
ta

ro
o +
­π
Ol

ta'
c
2.5 0 R=0.1
φ
a.
ST 10R=0.1

Ol
45 R=0.1
3 X
3'
Ol 60 R=0.1
πι
UI 'θ 0/45 R=0.1
ai Χ
.­* "cõ 1 . 5 Kl
TJ
ιι æ 45R = 10
ρ
Li
ai
3
a.
TJ
II >IOX<

KIKI >K >K_


­O
D>
:>«­
0.5 Kl Ki

X X
□□ LSZ^>
><x x x XX χ Χ
—c> X—O
0
Log Ν
0°, ± 5° and ± 10° laminates
Tl
tõ'
φ
ro

τι
Ol
ια' 2.5
c
φ
0 R=0.1
a
Ol

sr 10R=0.1
f/l Ρ
o LLi 2 "Nibe"
3
Φ
õT C
3
5'
Ol
φ

tn
οι
3
w 1.5 >x
Q.
.*·
3- "σ
Φ
c
NWZ
σ
9
'E affi ρ >I0X<—
3 o 1 ■ O
SI
Φ ρ □ rn¡ : O
ρ ano >κ
0.5
*—o
a
TJ
ιι
o

0 ¿. 7
Log Ν
Angle­ply laminates, R=0.1 and 10
Tl
ια'

ro
ro
+
o τι 10 R ==10
to'
2.5
5K
o. 10 R ==0.1
0)

s Ρ
45 R ==0.1
1+
o LLL Bl
01
C 45 R ==10
3
3'
01 'o
01
X
3
CL
oi 1.5
M
UI "o
01
3 •WS.
5'
01
φ
>ιοχ<
o
o
3 ­KIISI {>
5>' RDP + Kl Œ3 ■M £ >
o
0.5 ­°β­Β Ε
ρ-
El El
TJ PP
II [S_£^
o
oi
3
α 0
TJ 0 8
II
Log N
±10° laminates
τι
to'

ro
ω


2.5 +
Φ R = 10
α ÍK
R=0.1

ι &* ©
R=­1
οι
3
5'
01 'ö
ο
ο ω '.5
κ
3
-ο >κ
αϊ
-ι ~δ
¡Ä'
Ο
3
•ÍWS.

TJ
ιι
>ιοκ<
ι
­1
ÍK *L
TJ
ιι ^
0>
—i.
ο
ω
>Κ ΙΚ­ ■ο
0.5
3
α ­ο­
TJ
II
ο

0
8
Log Ν
00¿±450 laminates
τι
ια'
c
φ
ro II
+
j>

ai II
= ■ .

ta 2.5 WISPER
c
Φ
a.

οι
S R=0.1

2
tn
S
ai C
3
5'
01 'a +
o
o Μ 1.5
+
3
TJ ö
ai
¡ñ' c - ' -
O
3
Έ
O 1
co
TJ
m
'^m mm
TJ
■—L>
0.5
01
3
a
TJ
II
o
0 1 2 "2
» 4 C
6 8
Log N
Unidirectional and Angle-ply laminates
TI
ια'

ro
en
­τ­
0 R = 0.1
S 2.5
to' >K
c
φ
α. 10 R==0.1
Ol

sr Ρ
-*» 45
o
=
οι &* Χ
αϊ
3 60
Ξ'
ω
φ Ό + ■
ΙΛ
?<£ 0/45 R=0.1
φ
UI ω 1.5 E
i' "ο m 45 R:= 10
ω
φ
νι c ,­t­
­t­
ια : >Ι0Χ<
c
Φ

KIKI $κ ^e o
αϊ
3
α El Œ3 5*; :te-
2L
ο
0.5 -Oe-SΠ Ϊ & ­ E El
■σ
Φ PP
ο χ χ
XX XX χ Χ
x
\ χ
0
Φ
1/1

Log Ν
^/decade slope R-ratio :
0.1 10
0 "Risø" D O
"Nibe" χ
0.1 o
I)

0.2

0.3

θ
0 40 80 degree
' fatigue-limit

0.8

0.6

0.4
o
π
0.2
D

0 _L_
θ
o 40 80 degree

Figure 26. Fatigue limit and ε -Ν slope as a function of fibre orientation for +Θ laminates, R
= 0.1 and R = 10.

111
%/decade slope

-0.1

-0.2

-0.3 h
fraction of
► ±45° - plies
0 0.4 0.8 1.0
[0°] [0°2,±45o,0°]s [±45°]
fO *- fatigue limit

0.8
Δ

0.6

0.4-

0.2
fraction of
0 + ±45° - plies
0 0.4 0.8 1.0
[0o] [0°2,±45°,0°]i [±45°]
Figure 27. Fatigue limit and ε-Ν slope as a function of fraction of +45° layers in 07+45° lami­
nates, R = 0.1.

112
Nominal load lev els [%]:
1.25 D .8
X 1

X
* χ * « Of χ 1X1 χ
^ fl.

.75

.25

0 .2 .4 .6 .8 1
Lop N/Log Nmax

Figure 28. Ε­modulus as a function of N, normalized curve, for +10° laminate, at R = 0.1, for
max. strain of 0.8% and 1.0%, respectively.

Nominal load lev els [%]:


1.25 □ .8
X 1

** 0 X*
Xc
bo
X
.75
*x%x

ï *1

.25

0 .2 .4 .6 .8 1
Logr N/Log Nmax

Figure 29. Secant modulus Es as a function of N, normalized curves, for +10° laminate, at R
= 0.1, for max. strain of 0.8% and 1.0%, respectively.

­ 113
1.25

a fa-·—
Χ
Ε Α * ^

.75
"^x STA
\ ' 'i I

.25

O .2 .4 .6 .3 1
log {Ν)/log (Nmax)
Figure 30. E-modulus as a function of Ν, normalized curves, for ±10° laminates, at R = 0.1,
for strains of 1.18%, 0.93% and 0.58%, respectively.

1.25

' ^\r
-Vt^ f #n
.75
jiJ'
X
.25

.4 .6
lo g (N)/lo g (Nmax)

Figure 31. E-modulus as a function of N, normalized curves, for +60° laminates, at R = 0.1,
for strains of 0.32%, 0.20% and 0.1%, respectively.

114
E-mod. variation 10 —degree
Nominal load level 0.8%
1.25 ο E
X Es

Ιψ

75

.5

25

1 2 3 4 5 6 7
Log Ν

E-mo d. variation 10 — degree


Nominal load levels 1.0%
1.25 D E
Χ Ea

ρ "□# α Sc α

χ χ ΙΪ3-
75 1,
χ1
χ
> °Χ
S?

.25

o 1 3 4 5
Log Ν

Figure 32. Comparison of E-modulus and secant modulus E5, both as a function of N, nor­
malized curves, strains 0.8% (above) and 1.0%.

115
E/E &- "ETE
*-" - " o
s' 0
1

a b ""^""^v^

C \ :^E
Es

ι
1
„ log N/log Nmax
·.
0 0.5 1.0
Figure 33. Schematic curves of E and Es as a function of Ν, normalized curves, with three
time periods a, b and c, as described in the text.

L 'effective J
Figure 34. Sketch of microstructure for polyester matrix and glass fibres, with parallel cracks
(1), debonding (2) and transverse cracks (3).

116
Figure 35. Schematic stress-strain curves during fatigue, corresponding to conditions a, b
and c in Fig. 37.

Figure 36. Damage development in ±10° laminates, after fatigue at R = 0.1; strains are (a)
1.18%, (b) 0.93%, and (c) 0.58%.

117
Studies of damage development
•ti
in'
c
■ι
a
II
ω
α
■■
αϊ
2.5
3
αϊ
ια *κ
ID
α 10R=0.1
π
<
2. Χ
o
2
■o
3
ID
3
£i ­
60 R=0.1

Î 0/45 R=0.1
UI
'ö ■

3 χ
o
Ο
-\ w 1.5
α
ID Ν ^ ■ ,65­1
α ~ö ite
o
ite ite
■ ■■■
3 c 34­10
ΙΑ
■Ό
64­1
viwttry
ID
O O 1
3 56­15~«% ^ ­61­2
ID
3
UI if ite Τ
3
οι 34­5 34­¿1* 58­2 ■
te
i 0.5 χ
2^
42­2 Χ χ 44­1
(D
χ χ 44­3
Ζ
χ χ Χχ'χΧ
α
δ'
ια
/­ ";; c 7
αϊ
3
0 1 4 6 8
Log Ν
Figure 38. Damage development in ±60° laminate after fatigue at R = 0.1 ; strain is 0.3%.

119
2.1.4 RUG Investigations of GI-UP Materials (W. Sys)

2.1.4.1 General
This chapter describes the results of fatigue tests on glass-polyester coupons. This material
is widely used in the construction of rotor blades. In general, a rotor blade is a component
typically subjected to a very large number of load cycles in its lifetime. The particular aim of
the present work is to generate high cycle fatigue data up to more than N7 cycles.

2.1.4.2 Description of Material


The material selected was a well-defined glass fiber-reinforced-polyester made in the Risø
National Laboratory in Roskilde, Denmark. The volume fraction of fibers is 50% and the fiber
orientations are +10°. The material was fabricated on a filament winding machine where the
fibers are laid down on a flat plate-mandrel by wet winding. Two laminates of 2 mm thickness
are produced in each winding operation. The curing is done in an autoclave under low
pressure, where the laminate is consolidated between two steel plates to control thickness
and fiber content. Four laminates were delivered: plate 33, plate 37 top, plate 37 bottom and
plate 38 top.

2.1.4.3 Specimens
The tension-tension fatigue tests were conducted with straight-sided specimens of constant
cross section. The specimens were cut from the 2 mm thick laminates. Before cutting, end
tab material was glued to the plates. The material for the end tabs is a glass fiber-reinforced
epoxy about 2 mm thick, with fiber directions of 0° and 90°. The bond surfaces were prepared
by sanding and cleaning with a solvent. The tabs were beveled to a taper of about 10°. A
diamond impregnated saw with water cooling was used to cut the specimens in order to
obtain smooth cut edges. The test specimen geometry is: overall length 230-240 mm, gauge
length 150 mm, width 12 mm, and the specimen axis is at ± 10° of the fiber directions.

The tension-compression fatigue tests were performed - with side support from an anti­
buckling guide - on straight-sided specimens with nonbonded steel end tabs. The anti-buck-
ling guide and specimen are shown in Figure 1 on page 128.

2.1.4.4 Fatigue Machines


All fatigue tests are performed in servo-hydraulic universal ESH machines. These machines
are suited for tests in load control, displacement control or in third parameter (e.g. strain)
control. Each machine is equipped with peak detectors which can hold the peak value (in +
and in -) of load, displacement or external parameter. This is particularly interesting for
monitoring damage; for example, in load-controlled tests one can detect the gradual increase
of displacement if this has been selected by the peak device. Such an increase in displace­
ment is commonly the outcome of damage.

The machines are equipped with wave generators for different wave forms such as: sine,
square or triangle. For tension-tension tests, wedge-type grips are used, whereas for ten­
sion-compression tests, an active grip pressure is needed and grips with screw action are
mounted. Each machine has heating and cooling facilities to provide test specimen environ­
ments in the temperature range of -30 to 100°C.

- 120 -
2.1.4.5 Test Program
A concise test program must be written when conducting more than 100 high­cycle fatigue
tests. For the present case, the complexity of the plan is readily reduced as many potential
variables are kept constant. The "constants" include:
• a single material: glass fiber­reinforced­polyester with 50% volume fraction and a lay­up
of ±10°;
• constant­amplitude cyclic loading and sine wave forms;
• load­controlled fatigue: in general load­controlled tests are prefered for high­cycle fatigue
testing. As a consequence of this choice, a concise failure criterion, i.e. specimen break,
can be adopted.
The variables are:
• the temperature: fatigue testing is performed at three different temperatures: ­20°C,
+20°C and +50°C. These temperatures are surface temperatures, measured with ther­
mocouples glued onto the surface at mid­length;
• the stress range or R­ratio: the first series of tests are carried out under R = 0.1, the
second under R = ­ 1 . A s reported in Section 2.1.4.3,­ the tension­tension (R = 0.1)
specimens differ from the tension­compression (R = ­1) ones. The former have bonded
composite tabs and the latter nonbonded steel tabs. Consequently, the specimen design,
not being considered a real variable, may have an effect on fatigue behaviour. The
observed effect of the stress range may be partially due to the specimen design.

Frequency is not considered a variable because the temperatures are under control. How­
ever, it can be objected that the temperature distribution in the volume of the specimen is
not under control. For this reason, an acceptable frequency of 8 Hz was chosen for the +20°C
tests to avoid autogenous heating. For the ­20°C and the +50°C tests, a frequency of 15 Hz
was selected. It is, however, questionable wether the 15 Hz choice for the ­20°C tests was
a good one.

Tests at three temperatures and under two stress ranges are normally good for six different
fatigue S­N curves. It was decided that each curve should be defined at four stress levels
with a replication percentage of at least 75%. Hence, a minimum of 16 specimens is needed
per S­N curve. The material received from Risø yielded approximately 120 specimens. If four
spare specimens are considered for each series, no specimens are left for static tests and
preliminary tests. Hence, it was agreed by the participating laboratories to produce only five
curves; the S­N curve for R = ­1 and +20°C was not produced in order to assure sufficient
data for each curve. The remaining 20 specimens were used to:

• determine the Ε­modulus and the static strength;


• test on an exploratory basis in order to obtain an estimate of the position of the curve
in the S­N coordinate system. In this way, a good selection of stress levels could be
made: i.e., sufficient data in the 107 to 108 range and some data below 10" cycles­to­
failure.

The test plan is summarized in Table 1 on page 125.


A critical point in the test procedure is the determination of the cross­sectional area of the
specimens. Indeed, the specimens are not machined to exact dimensions as is the case with
steel specimens. The variation in the thickness direction is of the order of 0.1 to 0.2 mm,
which represents 5 to 10% of the area. In the absence of standardisation or recommended
test methods, it was decided to measure the thickness in three to five positions, and to take
the minimum measured area. The arbitrary character of the method should be recognized.

­ 121 ­
2.1.4.6 Results
The core results of the fatigue tests are summarized in Table 2 on page 125 to Table 6 on
page 127.

Statistical Characterization
From a practical design standpoint, fatigue data must be presented in terms of probability
of survival. One is more interested in survivability above 95% than in good estimates of the
mean time to failure.
In most practical situations of fatigue testing, a limited number of data are available. In our
programme of fatigue testing, four levels of stress are commonly used and four specimens
are tested at each level of stress. A method is now needed which provides a means of
forming distribution functions with some statistical value. For fatigue testing of composite
materials, a two­parameter Weibull distribution is widely used for estimating the number of
cycles to failure. It is commonly assumed that the shape parameter of the two­parameter
Weibull distribution is independent of stress level (Hahn and Kim [1]). Then a data pooling
technique can be utilized to determine the common shape parameter. The problem of
obtaining a value for the shape parameter from a limited number of test results is an esti­
mation problem for which various solutions are available. For fatigue testing, the maximum
likelihood method is widely used.
The approach described by Whitney, Daniel and Pipes [2] is used to define the two estima­
tors of the two­parameter Weibull distribution. A brief description of this approach is given
here.
The approach is based on the following assumptions:
• the shape parameter "a" of the Weibull distribution is independent of stress level;
• the scale parameter Np of the Weibull distribution can be written as a function of stress
(or strain) by a power law
κ b
Afe = (­§■) 2­ 1

where:
S = maximum stress;
Κ and b denote constants depending upon the material and loading conditions.
The probability of survival R(N) takes the form:

R(/V) = e x p [ ­ ( ­ ^ ­ ) a ] 2.2

where:
R(N) = P r {result>N};
Ν = number of cycles to failure.
The data set of the fatigue results can be described by N¡¡, where:
• i = 1, ... m : number of the stress level;
• j = 1, ... η : number of the replicate;
• N¡¡ : number of cycles to failure at stress level i for the replicate j .
The first step in the evaluation of the fatigue data is to fit at each stress level " i " a two­pa­
rameter Weibull distribution in order to obtain the best estimates of the scale parameters
Npi. This is achieved by using the method of maximum likelihood estimator (MLE).

­ 122
In the second step, the fatigue data is pooled by dividing, at each stress level, the test results
by the scale parameter Npl of the current stress level "i". The new data set X¡¡ is called the
"normalized data set". This data set, containing all the results of the fatigue tests, is now fitted
to a two­parameter Weibull distribution in order to obtain the best estimate of the pooled
common shape parameter "a". This is achieved again by using the method of MLE. The scale
parameter Xp of the Weibull distribution of the pooled data set should be unity. This is seldom
the case because the fit is seldom perfect. The scale parameters from the first step Npi will
now be multiplied by Xp, yielding:
Nai = XpNpi 2.3

where Nai is the "adjusted" scale parameter.


It should be noted that the adjusted "normalized data set" XPX¡, will yield the same shape
parameter "a"; the scale parameter will now be unity. The set of the values S¡, Na¡ is now fit
by the method of least squares to the power law (Equation 2.1) in order to determine the
material constants K and b.
If NP from Equation 2.1 is substituted into Equation 2.2, the following result is obtained:

S=K{[-inR(N)]Vab}N~Vb 2.4

For any value of the probability of survival, the S­N curve or ε­Ν curve can now be drawn.
It should be noted that the S­N curves obtained do not generate conservative estimates.
These can be calculated by utilizing lower boundary values for "a" and Np.
The calculated shape parameters "a" and the constants "b" and "K" are listed in Table 7 on
page 127. The units for Equation 2.1 are: cycles to failure for Np and MPa for S.
The results are illustrated in Figure 2 on page 128 to Figure 6 on page 130. Each figure
shows the individual test results (dots) and at the same time the 95% survival curve and the
5% survival curve. As can be seen, the data points lie reasonably well between the 95% and
the 5% survival curves. Several data points could not be used in the calculations for two
reasons:

• runouts at a certain stress level without broken specimens at this level cannot be used;
• specimens with different Ε­moduli tested at the same stress level cannot be evaluated
properly, because the strains are different.
Nevertheless, these points are represented in the figures (e.g. two runouts at 108 for R = 0.1,
Τ =+50°C and S = 100MPa.
The estimates of the endurance limit corresponding to Ν = 10e cycles at a probability of
survival of 95% are listed in Table 8 on page 127.

2.1.4.7 Conclusions and Discussion


This paper has described the activities performed in a joint research program. The present
results will be used in a general analysis which should form part of a final report.
The results of fatigue tests performed on ±10° laminates are described in strains versus
number of cycles. In fact, the tests are performed in load control; as such, strain is not a
direct measure. This is important because the error on a calculated strain can be significant;
as in actual practice, specimen thickness variations are unavoidable, and stress­strain curves
are not perfectly linear in the region of interest. It would be useful to develop standard pro­
cedures for determining cross sections and E­moduli.

­ 123
The results of the fatigue tests are presented in terms of probability of survival by using a
two-parameter Weibull distribution and a power law to describe the number of cycles N as
a function of the stress S.
The fatigue strength at -20°C is definitely higher than at +20°C or +50°C.

2.1.4.8 Bibliography
[1] Hahn, H.T. and Kim, R.Y. Fatigue Behavior of Composite Laminate Journal of Composite Mate­
rials, Vol. 10, No. 2, 156-180, April 1976.
[2] Whitney, J.M., Daniel, I.M., Pipes, R.B. Experimental Mechanics of Fiber Reinforced Composite
Materials Society for Experimental Mechanics Monograph No. 3, ISBN 0-912053-01-1, revised
edition, 1984.

124
2.1.4.9 Tables

Stress ratio Temperature (°C) N umber of specimens

0.1 -20 16 - 20
0.1 +20 16 - 20
0.1 +50 16 - 20
-1 -20 16 - 20
-] +50 16 - 20

Table 1. T est plan for fatigue tests.

MATERIAL: NUMBER OF CYCLES TO FAILURE


GRP, ±10°, Vf = 50%
Code: plate 37/top 250 200 175 150
E = Al GPa, UTS = 551 MPa MPa MPa MPa MPa
TEST CONDITIONS:
R = 0.1, f = 15 Hz, Τ = -20°C 1.32E5 6 .57E5 A.76E6 3..68E7
R H U ) = 50±10, load control,
sine wave 9 ■ 13EA 8 .18E5 1.A7E7 1..23E7
SPECIMEN:
1 χ b χ t = 2A0 χ 12 χ 2 1..00E5 7 .93E5 5.50E6 5..70E7
gage length = 150 mm
tab length and taper = A5 mm and 10° 1.16E5 1.A3E6 3.A6E6 5.57E7

Table 2. T est results for R = 0.1 and Τ = -20°C.

- 125 -
MATERIAL: NUMBER OF CYCLES TO FAILURE
GRP, ±10°, Vf = 50%
Code: plate 37/bot and 37/top 300 250 200 150
E = Al GPa, UTS = 551 MPa MPa MPa MPa MPa
TEST CONDITIONS:
R = 0.1, f= 8 Hz, Τ = +20°C 1.66EA 5.50EA 1.31E5 3.02E6
RH(%) = 50±10, load control,
sine wave 1.68EA 5.31EA 3.37E5 8.26E6
SPECIMEN:
1 χ b χ t = 2A0 χ 12 χ 2 1.63EA 6.AAEA 2.A2E5 9.95E6
gage length = 150 mm
tab length and taper = A5 mm and 10° 1.93EA 5.28EA 3.02E5 5.98E6

Table 3. Test results for R = 0.1 and Τ = +20°C.

MATERIAL: NUMBER OF CYCLES TO FAILURE


GRP, ±10°, Vf = 50%
Code: plate 33 350 250 200 150
E = 33 GPa, UTS = A98 MPa MPa MPa MPa MPa
TEST CONDITIONS:
R = 0.1, f = 15 Hz, Τ = +50°C A.30E3 2.99EA A.73EA 1.03E6
RH(%) = 50±10, load control, "
sine wave 1.51E3 1.90EA 9.92EA 3.08E6
SPECIMEN:
1 χ b χ t = 2A0 χ 12 χ 2 0.90E3 2.66EA 6.92EA 2.58E6
gage length = 150 mm
tab length and taper = A5 mm and 10° 2.2AE3 1.03EA 1.07E5 2.38E6

Table 4. Test results for R = 0.1 and T = +50°C.

MATERIAL: NUMBER OF CYCLES TO FAILURE


GRP, +10°, Vf = 50%
Code: plate 38/top 200 150 125 100
E = A0,3 GPa, UTS = A87 MPa MPa MPa MPa MPa
TEST CONDITIONS:
R = -1, f= 15 Hz, T = -20°C 3.2AEA 2.52E5 A.10E6 7.05E6
RH(%) = 50+10, load control,
sine wave 3.06EA 3.65E5 A.62E6 3.27E7
SPECIMEN:
1 χ b χ t = 2A0 χ 12 χ 2 3.08EA 2.68E5 2.72E6 8.99E6
gage l e n g t h = 150 mm
a n t i - b u c k l i n g guide 3.67EA 2.33E5 A.80E6 2.08E7

Table 5. Test results for R = -1 and Τ = -20°C.

- 126
MATERIAL: NUMBER OF CYCLES TO FAILURE
GRP, ±10°, Vf = 50%
Code: plate 37 and 38/top 200 150 100 80
E = Al GPa, UTS = 551 MPa MPa MPa MPa MPa
TEST CONDITIONS:
R = -1, f= 15 Hz, Τ = +50°C 2.31E3 2.A5E3 3.25E5 A.25E6
RH(%) = 50+10, load control
sine wave 3.80E3 1.A5EA 9.87EA A.57E6
SPECIMEN:
1 χ b χ t = 2A0 χ 12 χ 2 3.20E3 6.27E3 2.1AE5 2.89E7
gage l e n g t h = 150 mm
a n t i - b u c k l i n g guide 2.A0E3 9.70E3 3.55E5 2.15E7

Table 6. Test results for R = -1 and Τ = +50°C.

R T°C a b k

0.1 -20 2.55 11.96 65A.6


0.1 +20 5.05 8.88 870.0
0.1 +50 2.72 8.17 872.5
-1 -20 3.37 9.7A 57A.9
-1 +50 2.37 9.A0 A28.3

Table 7. The shape parameter and constants b and K.

R T°C Max. s t r a i n (%)

0.1 -20 0.32


0.1 +20 0.25
0.1 +50 0.25
-1 -20 0.20
-1 +50 0.13

Table 8. Maximum strain for 95% survivability.

127
2.1.4.10 Figures

F I L L E R (110 x l 7 χ 1,8 1

Α Ν Τ Ι - BUCKLING PLATE
(110 x 2 0 x i >

CROSS SUPPORT
(SOx 2 S x i I

Dimensions in mm.

Figurei. Antibuckling guide.

GRR C VF = O . 5 / = 1O° :

c ι .o
α

Χ O. 5

Figure 2. Results for R = 0.1 and Τ = -20 and equation 2.4 for R(N) = 0.05 and 0.95.

128
GRR C Vr = O . 5 / ί 10° J

c 1.0
α

5
α °-
Σ:

Figure 3. Results for R = 0.1 and Τ = +20 and equation 2.4 for R(N) = 0.05 and 0.95.

GRR C V

OL

Χ
a

Figure 4. Results for R = 0.1 and Τ = +50 and equation 2.4 for R(N) = 0.05 and 0.95.

- 129
GRR C V, ­ O . 5 / ­ 10° 3
^ Ä t .

ΙΙΙΙΙΙ
RIJKSUNIVERSITEIT
GENT

c 1.0
α
" ­ ^ ^ s ^

9 5 ^

2 3 s e
io i o i o io­* 1 0 1 0 io"7 1 0 e

Nt c y c l e s 3

Figure 5. Results for R = ­1 and Τ = ­20 and equation 2.4 for R(N) = 0.05 and 0.95.

mïïT
GRR C V^ = 0 . 5 / Ï 10° 3

R 1 Τ ­ * S OC
°
β Ε ­ 4 0 . Ξ GPo
­ Ε ­ 3 3 GPa
c 1 . α

χ
a
Σ:

e 3 4 5 e
IO 1 0 IO I O IO 1 0 io"7 1 0 e

NC c y c l e s D

Figure 6. Results for R = ­1 and Τ = +50 and equation 2.4 for R(N) = 0.05 and 0.95.

130
2.2 Glass-Epoxy Composites (Gl-Ep) (Ch.W. Kensche)

2.2.1 DLR Investigations

2.2.1.1 General
Horizontal axis wind turbines are being designed for more than 20 years of operation. That
means that for the rotor blades, which commonly are made of GRP, load cycle numbers of
more than 10° will occur in that period. However, at the beginning of the project, no fatigue
data for that material exceeding 106 - 107 load cycles were available. A reliable lifetime
evaluation was not possible, and the rotor blades had to be designed conservatively, leading
to unnecessarily high design weights.
Thus, the aim of this investigation was to test the behaviour of different glass-epoxy (Gl-Ep)
systems up to 10B load cycles under constant amplitude and also under the wind energy-
specific load spectra WISPER of the IEA (International Energy Agency) as well as under the
short version WISPERX.
The tests were carried out on flat coupon specimens with laminates typical for wind turbine
rotor blades.

2.2.1.2 Material
The Gl-Ep material used in these investigations was fabricated and delivered by industry.
The lay-up of the specimens consisted of unidirectional (UD) fabric, the ravings being ori­
ented in the loading direction, and ±45° cloth for shear forces. It was similar to that used in
the AEROMAN rotor blade. For comparison purposes, a pure UD lay-up was additionally
investigated.
Two matrix systems were chosen, the cold curing and hand lay-up resin system L20/SL by
Bakelite and the low temperature/low pressure prepreg systems M10 and M9 by Brochier
and Ciba. Table 1 and Table 2 on page 138 show the specification of the material for the
specimens and the woven glass reinforcement.
In principle an attempt was made to get specimens of laboratory quality to avoid excessive
scatter in lifetime induced by materials of different or bad quality. Nevertheless, small voids
occurred in the specimens with both types of resin systems used. The medium fibre content
was 36.9 vol% for the L20/SL material with pure UD laminate (Type 1A), 38.6 vol% for the
material with the mixed lay-up (Type 2A) and 53.0 vol% for the prepreg coupons (Type 2C),
see also Table 1.

2.2.1.3 Specimens
The plates for the specimens were laminated by MAN-Technology. The curing and postcuring
procedure for the material with the L20/SL resin system was:
precuring at RT (room temperature) between steel plates at 50°C
postcuring 8 h at 60°C
postcuring 15 h at 80°C
The M9/M10 prepreg system was cured and postcured following Brochier/Ciba specifications.
After fabrication the plates were sent to DLR to be cut as received parallel to the UD fibre
orientation into pieces 250 mm long and stored in the uncontrolled climate of the laboratory
until usage.
For the prepreg material, which was the first to be tested, a dog bone geometry was chosen.
The coupons with the L20/SL system, however, had a rectangular shape. Both coupon types

131
had a minimum width of 20 mm. Figure 1 and Figure 2 on page 146 show the geometries
of the different types. The thickness differed generally between 2.6 and 3.0 mm for Type 2C,
and 2.0 and 2.4 mm for Types 1A and 2A, see also Table 1.
The steel tabs used for the first batch of prepreg material (dog bone I, see Table 3) were
replaced by Gl­Ep tabs with +45° lay­up for all following tests. Since almost all tests were
done partly or totally In axial compression loading, the use of antibuckling guides was inevi­
table. Different types of these stability supports had to be used, DLR Type I (Figure 3), DLR
Type II (Figure 4), and Type LBF (Figure 5), a development of the Laboratorium für
Betriebsfestigkeit in Darmstadt. Table 3 on page 138 gives a survey of which geometry, type
of tab and antibuckling guide was chosen for the different specimen types. In the case of the
LBF support, the Gl­Ep tabs were glued onto the steel and not onto the specimen itself.

2.2.1.4 Testing Procedure


Static and fatigue tests were carried out to supply all necessary data for an ε­Ν curve. The
static tests were partly performed in a spindle test machine (Zwick), partly in hydraulic pul­
sators which were used also for the fatigue testing. In this case the strain was measured by
strain gauges only, in the Zwick machine clip gauges were additionally used. Table 4 gives
an overview of the failure strains from static tests. A ll these tests were carried out in a
Celanese testing device except for Type 2C, which was tested in the Type I antibuckling
guide.
Fatigue testing was load controlled. Constant amplitude loading was carried out sinusoidally,
variable amplitude loading in a triangular wave form in servo­hydraulic cylinders, all pulsators
being equipped with mechanical grips. Table 3 gives a survey of which type of material was
investigated with the relevant fatigue loading.
For testing the effects of variable amplitude loading, the WISPERX load sequence was
applied which is a short time derivation of the WISPER standard needing only the tenth part
of testing time from WISPER. Both are shortly described in Section 1.3. The reason to use
WISPERX was twofold. Primarily, the good fatigue behaviour of the material used would have
prolonged the testing time extensively. Even with WISPERX an actuator time of one year
was necessary to complete the program. Secondly, the decision was supported by results
of the NLR investigation, which did not show remarkable differences in lifetime between
WISPER or WISPERX In terms of passes through the sequence [3].
For the range­pair­range counting results of the sequence, which were also used by DLR for
the lifetime evaluation described below, see Tab. 3 in Section 2.1.2.
The maximum load level of class 64 in either the WISPER or WISPERX spectrum is sup­
posed to be a positive load, i.e. it represents the tension area or lower spar cap, respectively,
of a blade. However, the upper cap will then be loaded in compression, which is supposed
to be more critical. Consequently, reverse WISPERX was also applied at the coupons, where
level 64 is the maximum negative and 25 the zero load level.
The fatigue life of GRP is often accompanied by a decrease of stiffness, the extent of which
depends on the lay­up and the resin system of the compound used. In many cases a severe
reduction can be observed at the beginning of fatigue loading and shortly before failure,
whereas the long phase in between is relatively constant. A s the beginning of the larger
decrease of stiffness shortly before failure can be used to estimate the end of lifetime, the
measurement of stiffness reduction during fatigue loading is an important task.
Therefore each specimen was loaded prior to its fatigue test up to the maximum tension
strain, which was measured by an extensometer. Load recording enabled a determination
of the Ε­modulus. The mean value of the Ε­modulus for the various types of laminates is
shown in Table 1. Stiffness degradation during the load­controlled fatigue tests was moni­

132
tored by measuring either the extensometer signals (at intervals) or the change of the dis­
placement of the grips (on-line) of a representative number of specimens during the tests.

2.2.1.5 Results
Constant Amplitude Tests on Prepreg Material (Type 2C)
The first tests were carried out with Type 2C prepreg material under tension-compression
loading (R = -1). The statically tested specimens achieved mean strains of 2.67 per cent in
tension and 1.96 per cent in compression, see Table 4. The mean Ε-modulus (of the
dynamically tested coupons) was 31.7 GPa, see Table 1. This is a very high value. It can
be assigned to the high fibre content of the compound (53 vol%), see Table 1, and the high
ratio of the fibre content in warp direction to that in fill direction in the load bearing glass fabric
UD 940 (19.1/1), see Table 2. The fatigue tests on the specimens were performed at maxi­
mum strain levels ranging from ±0.4 to ±1.4 per cent. The testing frequencies depended on
the strain level, with the purpose of not exceeding the surrounding temperatures at the sur­
face by more than about 10°C. They were between 10 and 25 Hz for 0.4 per cent, and
2.5 Hz for 1.4 per cent strain. An antibuckling guide was used, as shown in Figure 3 on
page 147. Stiffness reduction could be measured continuously with a Schenck extensometer,
the temperature on the surface of the middle section with a PT 100 resistance sensor. A
typical on-line analogue plot shows the surrounding and surface temperatures as well as the
load and the change in strain of a specimen loaded with a strain of ±0.8 per cent, see
Figure 7. Strain and surface temperature begin to increase shortly before failure, at about
96 % of lifetime. The increase of strain during this time was not larger than about 4 per cent.
This behaviour is similar for all types of material used and all R-ratios.

Unfortunately the steel tabs used for these specimens caused wear in the gripping area.
Consequently, several coupons failed at this location. The scatter at the low-cycle side of the
data points in Figure 8 can be attributed to that reason. Figure 6 shows the wear appear­
ance of those gripping areas after testing.

Thus, the 2C specimens for R = 10 (compression-compression) got Gl-Ep tabs, see


Figure 1. The material for these specimens was laminated, using the same charge of pre­
preg as for the R = -1 coupons, however about six months later. It seems that this time of
storage had changed the quality of the prepreg system, for the material contained visibly
more flaws compared to the first lot which was tested with R = - 1 . However, the mean
Ε-modulus was still higher (34.0 GPa), see Table 1.

The measurement of stiffness change in these specimens was still performed with the clip
gauge, but only at certain stages of life. The reason was twofold. A practical one was that
the very expensive sensor can be damaged in the case of failure; on the other hand it had
been shown in previous tests that stiffness reduction did not exceed 4 % during the main
part of lifetime for the material used.

Figure 8 on page 150 shows the data points in an ε-Ν plot. Surprisingly, in the high strain
region the points are below those of the R = -1-specimens, whereas they have superior val­
ues In the low strain area. It was supposed before that the curve would have overall higher
values. Two possible reasons are seen. Firstly, the material contained obviously more flaws
than that used for the R = -1 tests, as mentioned above. Secondly, many of the coupons
failed due to intralaminar shear induced by the dog bone shape.

A survey of load cycle numbers, applied strain and frequency as well as stiffness degradation
of the 2C specimens tested is given in Table 5 and Table 6 on page 140.

133 -
Constant Amplitude Tests on Cold Curing Material of Type 2A
Due to experience with scatter in the fatigue results of dog bone shaped specimens, for the
following specimens (Type 2A) a rectangular shape was used, see Figure 2. They were built
with the cold curing resin system L20/SL, and also tested with R ratios R = -1 and R = 10,
to have a good basis for comparison with the prepreg material. Static tension tests showed
that the mean failure strain is 2.21 per cent, which is close to that of the compression results
of 2.14 per cent. The scatter is relatively low, see Table 4. The mean Ε-modulus was
24.6 GPa for R = - 1 , and 22.7 GPa for the R = 10 material, see Table 1. This is relatively
low compared to the prepreg material of Type 2C, which is probably due to the lower fibre
content of 38.6 vol%.

The fatigue tests at R = -1 were performed between 0.4 and 1.2 per cent maximum strain.
Two types of antibuckling guides were used, DLR II and LBF, see Figure 4 and Figure 5
on page 148. The first type had the advantage that it could be dismantled for the purpose
of measuring stiffness with a clip gauge. Those specimens had Gl-Ep tabs. However, at
strains of more than 0.8 per cent, some of them failed very early due to instabilities in the
area between the tabs and the antibuckling guide. Thus, it was decided to use the LBF type
antibuckling guide for strains beyond 0.6 per cent. In this case the tabs were glued onto the
steel guide. In spite of smooth load introduction in the grippings, several coupons failed in
that area, causing a scatter of fatigue data points. However, some samples tested at 0.4 per
cent achieved 10e load cycles and unfortunately failed only because of a fault in the electro-
hydraulic net. These runouts indicate that a possible linear fatigue curve of 50 % survivability
could go flat here, i.e. that endurance limit may exist for this Gl-Ep.

When tested in compression-compression, the material showed strains about twice as high
as those in tension-compression relative to the load cycle number, Figure 9. This is in con­
trast to experience with the prepreg material, where bad quality, the dog bone shape of the
specimens, and possibly a too high fibre content seem to have caused lower values and
another slope for the ε-η data of the R = 10 tested material.
The frequencies were between 3 and 16 Hz for the different strains applied.
Stiffness reduction could only be measured from time to time on those specimens with DLR
Type II antibuckling guides, since those with the LBF type could not be dismantled without
opening the grippings of the loading machine. Experience has shown that in many cases
sudden failure occurs in the gripping area of those coupon specimens which were removed
from the machine and later reinstalled. The reduction was little more than 0.4 to 0.5 %, see
also Table 7 and Table 8.
In view of the tests described later on large scale components, a small but important inves­
tigation still has to be mentioned which was carried out by courtesy of another DLR institute
[4]. Two specimens of Type 2A were tested at R = -1 and a strain level of 0.8 per cent. The
width was 40 mm and buckling was enabled in a window of 36 · 36 mm2. By buckling at a
frequency of 5 Hz the temperature also increased steadily, and finally fatigue occurred at
about 3 · 10" load cycles, far earlier than expected from the relevant existing ε-Ν data. Thus,
it seems that these results must be attributed not only to buckling-induced stresses, but also
to temperature. The data points are shown in Figure 15 on page 157.

Constant Amplitude Tests on Cold Curing Material of Type 1A


The UD material specimens (Type 1A) were also tested at R-ratios of R = - 1 , but additionally
in tension-tension (R = 0.1), see Figure 10 on page 152. The Ε-moduli were 29.6 GPa and
31.9 GPa, respectively, see also Table 1.
The mean static strain in tension was about 2.1, in compression 2.4 per cent, see Table 4.
Also the fatigue data points for R = -1 are in good correspondence to those of Type 2A

134
including two 10B-runouts at the 0.4 per cent level. This similarity of the curves was expected,
as the UD material was the same. The ±45° cloth in Type 2A does not seem to have a great
influence on fatigue properties compared to those of Type 1A.
The fatigue strains of the tension-tension (R = 0.1) tested 1A material are also higher than
the R = -1 data points, but lie beneath the R = 10 data of the 2A coupons.
The relation between testing frequencies and applied strains was similar to that for Type 2A.
For strains higher than 0.8 per cent, in general the LBF antibuckling guide had to be used
to avoid early failure in the gripping area due to local buckling, as was seen when testing the
2A material. Thus, a stiffness reduction measurement by means of clip gauges could only
be performed on the low strain specimens when using the DLR Type II antibuckling guide.
The reduction showed similarly low values as for Type 2A specimens, see Table 9 and
Table 10.

Load Sequence Tests


Load sequence tests were carried out in the normal and the reverse version of the WISPERX
spectrum on seven specimens each. These tests were performed in one hydraulic pulsor only
and needed a testing time of more than one year.
The coupons were of Type 2A. Table 11 and Table 12 give a survey of the lifetime and the
stiffness reduction which was measured at certain stages of service life by means of an
extensometer. However, the rated load signals for the computer, the load and the displace­
ment of the hydraulic jack were also observed on-line by a Gould and Brush recorder. For
all specimens the change of displacement over lifetime was evaluated. The change in stiff­
ness, calculated from the clip gauge signal, was compared with that calculated from the
change of the displacements amplitudes. Figure 11 on page 153 demonstrates that the high
stiffness reduction of 10 % and more, computed on the basis of the displacement change,
Is misleading. It includes, e.g., also the stiffness change of the whole gripping area, and thus
cannot be used for a correct fatigue estimate. However, the extensometer measurements
show stiffness changes which are relatively low (not more than 4 %) for all maximum strain
levels which were applied.

For the normal WISPERX spectrum (high load peak in tension) four maximum strain levels
were tested, the lowest being 1.0 per cent, see also Figure 12. At this level one specimen
was fatigue loaded up to 1200 passes through the sequence without showing any signs of
fatigue. On that specimen a residual tension test was carried out with a failure strain beyond
2 per cent, i.e. like the static strength of the virgin coupons, also plotted in that figure. Also
at 1.2 and 1.6 per cent strain one specimen each was investigated with the sequence, while
at the 1.4 per cent level four specimens were fatigue-loaded. A least square regression line,
calculated from the points between the static strength data and the 1.2 per cent strain point,
indicates that the lifetime of the 1.0 per cent coupon could be perhaps one decade higher
than tested.
The specimens with reverse WISPERX were tested at the -1.2, -1.4, and -1.6 per cent levels,
see Figure 12. The data points of lifetime are much closer to those of the tensionally tested
coupons than would be expected having in mind the large difference between the R = 0.1
and R = 10 values of the ε-Ν curves. The reason was found in the failure mechanism. The
evaluations of the analogue plots showed that some of the specimens investigated with
reverse WISPERX failed at the maximum compression load in the area between the Gl-Ep
gripping and the DLR Type II antibuckling guide. That means that instability effects had
produced early failure of the reverse loaded pieces. This instability does not seem to have
similar detrimental effects on the normally loaded coupons.

- 135
2.2.1.6 Fatigue Life Evaluation
For a life prediction of metals and metal joints the linear Palmgrene­Miner rule is state of the
art. It assumes linear damage accumulation due to the different amplitudes of a load
sequence. This results in the cumulative damage coefficient

¡=1 '

where k is the sum of the load steps, n, the number of spectrum load cycles at strain ε, and
N¡ the number of load cycles up to failure at ε,. D depends mainly on
the load spectrum,
the stress level,
the structural design,
the composite lay­up.
Since experience has shown that the value of D can vary over a wide range, from 10~1 to
10+1, this theory is not very satisfactory. However, due to its simplicity it is widely used also
for composite materials, and was applied in our case, too.
A constant amplitude life diagram was constructed from the ε­Ν data of the 2A and 1A
materials. This seems to be justified as long as no data for 2A material at R = ­1 are avail­
able, because the R = ­1 values are similar, and thus a similarity in tension fatigue can be
assumed for Type 1A and 2A. Figure 13 shows the diagram based on the logarithmic mean
values of the single constant amplitude data.
In order to enable accurate lifetime prediction for all R­ratios which exist in the load
sequence, the relevant fatigue data should be available. A s this is not the case they must
be found by means of interpolation. A polygon spline is performed, see Figure 13. But a
cubic spline could also be possible, as shown in Figure 14. For our calculation the R­values
of the load sequence were taken from the range­pair­range counting results of Table 3 in
Section 2.1.2. The standard deviation (RMS = Root Mean Square) in column 5 of that table
shows that when relating all equal range size levels to an average mean level, the maximum
standard deviation is not higher than 6.5. Most RMS values are lower than half of 6.5, i.e. the
average mean level for the different R­values of equal range sizes is very close to the
demanded accuracy.

The calculation was performed for both normal and reverse WISPERX as well as WISPER
sequences at the 1.2 and the 1.4 per cent strain levels. Both forms of interpolation were used.
Table 14 on page 145 shows for the cubic spline that at a strain level of 1.2 per cent the
calculated as well as the tested lifetime for the normal WISPERX achieved 710 passes
through the sequence, i.e. the damage accumulation factor would be exactly 1.0. For the 1.4
per cent level the Miner sum would be 1.28. Calculation shows, however, that for the whole
WISPER sequence the lifetime would be only half that value.
For reverse WISPERX the damage factor is about 0.1 at 1.2 per cent, and 0.2 at the 1.4 per
cent strain level. This is on the unsafe side. The reason is that, as explained above, in the
load sequence experiments the coupons failed too early compared to the R = 10 constant
amplitude tests of the 2A material, due to instability effects.
In Table 13, it can be seen that linear interpolation shifts the Miner sum to the unsafe side,
but the tendencies are the same as described for the cubic spline.
Considering the results of the tension­tested WISPERX specimens it may be assumed that
the cubic and the polygon spline give the upper and lower limits of possible data for different
R values. However, the calculated damage factors for the linear spline are only about 0.6

136
of those for the cubic spline. These differences are remarkably low and would allow a rela­
tively good lifetime prediction. This demonstrates that linear interpolation may give satisfac­
tory results for the construction of a constant amplitude life diagram when R values ­ 1 , 0.1
and 10 are known.
In a concluding consideration on lifetime evaluation it does seem to be acceptable to apply
the linear Palmgren­Mlner rule also on composite materials, if reliable data are available. In
the case of large differences from the ideal damage sum, the failure mechanism should be
studied carefully, which may lead to a satisfactory explanation. Nevertheless, it should be
taken into account that these results were achieved under ¡deal conditions and represent an
optimum case.

2.2.1.7 Conclusions
High cycle fatigue behaviour of several Gl­Ep materials was investigated with coupon spec­
imens at stress ratios of ­ 1 , 0.1 and 10. It is shown that at R = ­1 the most conservative
results were achieved in comparison to R = 0.1 or R = 10. However, at the lowest strain
applied in the investigations (ε = 0.4 per cent) the fatigue limit seems to be still higher than
10° load cycles. From these results a Haigh diagram was constructed.
Load sequence effects were tested on coupon specimens of a selected material using WIS­
PERX, the short version of the wind energy specific standard WISPER. The standard was
used in a normal mode, representing a tension loaded cap of a spar beam, and also in a
reverse mode, simulating the compression cap. The linear Palmgren­Miner rule was applied
to evaluate the lifetime of the material for a maximum sequence strain of 1.2 and 1.4 per cent
in tension and compression.
For tension the cumulative damage factor is shown to be near 1.0 depending on whether a
cubic or a linear interpolation is chosen in the Haigh diagram between the known R­values.
The Miner sum for the compression mode, however, is near 10"\ It was shown that this
disagreement is due to different failure modes of the specimens due to the use of different
antibuckling guides.

2.2.1.8 Bibliography
(1 ] A .A . Ten Have WI SPER: A Standardized Fatigue Load Sequence for HAWT-Blades European
Community Wind Energy Conference (ECWEC '88) Proceedings 1988, Herning/Denmark
[2] Ch. Kensche Research Programme for Fatigue-Critical Structures of Small and Medium-sized
Wind Turbines European Community Wind Energy Conference (ECWEC '88) Proceedings 1988,
Herning/Denmark
[3] W.J.A . Bonnee Constant Amplitude and Spectrum Fatigue Loading of Glass Fibre Reinforced
Polyester Coupon Specimens NLR CR 90316 L, August 1990
[4] L. Kirschke Fatigue Testing of Coupon Specimens under Admittance of Buckling DLR
Braunschweig, Personal communication, 1 Nov 1990

137
2.2.1.9 Tables

Resin Mean Fibre Mean Thick­


Type Lay-Up System Content E-modulus ness
vol% GPa mm
1A 5x92146/0° L20/SL 36.9 29.6 (R=-1) 2.0 - 2.4
31.9 (R=.1)
2A (1x92125/±45°; L20/SL 38.6 24.6 (R=-1) 2.0 - 2.4
2x92146/0°) sym 22.7 (R=10)
2C (1x1023/±45°; M9/M10 53.0 31.7 (R=-1) 2.6 - 3.0
1.5xUD940/0°) sym (Prepreg) 34.0 (R=10)

Table 1. Lay-up, fibre content, modulus and thickness of specimens

Style Producer Weight Weave Thickness


g/m2 Pattern mm
92146 Interglas 425 UD (10/1) Plain 0.47
92125 Interglas 280 2/2 Twill 0.40
UD940 Brochier 920 UD (ca.19.1/1) 0.83
(with M9-Prepreg) (Ciba) (1483) Plain
1023 Brochier 390 2/2 Twill 0.45
(with M10-Prepreg) (Ciba) (650)

Table 2. Data of glass-reinforcement.

Material R Geometry of Type of Type of Anti­


Type Specimens Tabs buckling Guide
1A 0.1 straight GRP
-1 straight GRP DLR ll/LBF
2A -1 straight GRP DLR ll/LBF
10 straight GRP DLR ll/LBF
WISPERX straight GRP DLR II
Reverse WISPERX straight GRP DLR II
2C -1 dogbone I Steel DLR I
10 dogbone II GRP DLR II

Table 3. Geometry of specimens and type of antibuckling guide.

138 -
Material Static Failure Mean Standard
Type Loading Strain, % Strain, % Deviation, %
1A Tension 2.060 2.092 0.200
1A Tension 2.015
1A Tension 2.005
1A Tension 2.257
1A Tension 2.412
1A Compression 1.933 2.418 0.231
1A Compression 2.574
1A Compression 2.396
1A Compression 2.523
1A Compression 2.498
1A Compression 2.703
1A Compression 2.296
2A Tension 2.326 2.212 0.153
2A Tension 2.285
2A Tension 2.284
2A Tension 1.948
2A Compression 2.081 2.140 0.109
2A Compression 2.029
2A Compression 2.336
2A Compression 2.051
2A Compression 2.272
2A Compression 2.074
2A Compression 2.135
2C Tension 2.780 2.670 0.110
2C Tension 2.560
2C Compression 2.200 1.960 0.240
2C Compression 1.720

Table 4. Static tension and compression strain values of material types 1A, 2A
and 2C.

- 139
Anti­ Stiffn. Reduction
Loadcycles Strain Specimen Freq. buckling After Temp.
[%] Code [Hz] Guide [%] Cycles [°C]

28787900 0,4 01-05 10,0 DLR I 3 2.5E + 07 26-38


187270 0,6 7-1 10,0 DLR I 21-40
245440 0,6 02-24 10,0 DLR I 3 245440 25-44
705290 0,6 02-14 10,0 DLR I 2 705290 26-48
1437500 0,6 02-20 10,0 DLR I 25-31
3238290 0,6 02-18 10,0 DLR I 3 1599770 24-52
3339010 0,6 02-17 10,0 DLR I 0 3339010 25-38
120660 0,8 01-11 5,0 DLR I 2 120660 26-45
165460 0,8 01-07 5,0 DLR I 2 165460 31-48
277950 0,8 01-06 5,0 DLR I 1 277950 26-42
8625 1,0 01-12 2,5 DLR I 24-31
33610 1,0 02-19 2,5 DLR I 7 33610 25-39
36409 1,0 02-15 2,5 DLR I 3 36409 24-36
1851 1,2 02-13 2,0 DLR I 3 1851 21-32
3930 1,2 02-22 2,0 DLR i 3 3900 24-38
1043 1,4 02-21 1,5 DLR I 3 1043 25-36
1314 1,4 02-23 2,0 DLR I 13 1314 27-36
Specimens: Dog bone, Steel tabs; E-Reccirding with Clip Gauges Continuously and Statically

Table 5. Results with material type 2C, (1x1023;±4571.5xUD940;0°)8ym / M10/M9, R = -1.

Anti­ Stiffn. Reduction


Loadcycles Strain Specimen Freq. buckling After Temperatur
[%] Code [Hz] Guide [%] Cycles [UC1
9813370 0,6 2C/03 10 DLR II
1844000 0,8 11 8 DLR II 20-25
2149460 0,8 10-10 10 DLR II 2 2149460 22-28
5907740 0,8 10-12 8 DLR II 12 5907740 19-24
7913420 0,8 2C/20 8 DLR II 5 7912670 20-29
8213130 0,8 10-07 10 DLR II
9220 1,0 10-05 10 DLR II 22-28
17400 1,0 10-04 10 DLR II
26010 1,0 2C/13 6 DLR II 19-24
36980 1,0 2C/14 6 DLR II 5 36980 20-25
185980 1,0 10-06 10 DLR II
780 1,2 10-08 5 DLR II 22-25
871 1,2 2C/18 5 DLR II 24-26
3700 1,2 10-09 3 DLR II
5430 1,2 2C/21 3 DLR II
5600 1,2 2C/16 5 DLR II 3 5600 24-31
7290 1,2 2C/22 3 DLR II 1 7290 27-30
Specimens: Dof ibone, Gl-E D tabs; E-Re(:ording by Displáceme nt of Act jator

Table 6. Results with material type 2C, (1x1023;±4571.5xUD940;0°)sym / M10/M9; R = 10.

140
Anti­
Loadcycles Strain Specimen Freq. buckling Comment
[%] Code [Hz] Guide

100061600 0,4 2A-21 16 DLR II Run Out


76421820 0,4 2 A-13 16 DLR II
56637600 0,4 2A-20 12,5 DLR II
6156400 0,6 2 A-14 12,5 DLR II
4383690 0,6 2A-50 4 DLR II
3337630 0,6 2A-5 6 DLR II
563630 0,8 2A-21 4 LBF
551000 0,8 2A-22 4 LBF
357760 0,8 2A-28 5 LBF
290000 0,8 2A-27 5 LBF
90150 0,8 2A-25 5 LBF
43980 1 2A-26 2,5 LBF
28600 1 2A-24 2,5 · LBF
36100 1 2A-20 2,5 LBF
250 1,2 2A-23 3 LBF

Specimens: Rectangular, Gl-Ep tabs; Ε-Recording by Static Tests with Clip Gauges

Table 7. Results with material type 2A, (1x92125;±45°/2x92146;0°)8ym / L20/SL; R = - 1 .

Anti­ Stiffn. Reduction


Loadcycles Strain Specimen Freq. buckling After Comment
[%] Code [Hz] guide [%] Cycles
109039710 0,6 2A-52 12,5 DLR I I 1,4 78322000 Run Out
39254000 1 2A-30 5 DLR I I
30314920 1 2A-40 4 DLR I I
16249670 1 2A-31 5 DLR I I
1869840 1 2A-51 3 DLR I I
1297270 1,4 2A-38 3 LBF
412370 1,4 2A-39 3 LBF
Specimens: Rectar igular. Gl -Ep tabs; E-R ecording t >y Static Tests with Clip Gauges

Table 8. Results with material type 2A, (1x92125;±4572x92146;0°)sym / L20/SL; R = 10.

141
Anti­ Stiffn. Reduction
Loadcycles Strain Specimen Freq. buckling After Temp. Comment
[%] Code [Hz] guide [%] Cycles [°C]

100000000 0,4 03-20 16 DLR II Run Out


100000000 0,4 03-19 16 DLR II 0 3,8E + 07 Run Out
27800000 0,4 03-16 16 DLR II
57220600 0,4 03-18 16 DLR II 4 3,6E + 07
5340000 0,6 3.4 5 DLR II 19-22
6560900 0,6 02-8 6 DLR II 30-33
260530 0,8 1A-3.1 5 DLR II 27-29
349050 0,8 1A-1 5 DLR II
606340 0,8 03-5 5
873440 0,8 02-2 5 DLR II 25-28
1016760 0,8 1A-3.3 4 DLR II 0 572000 26-33
16000 1 X1 3/2 DLR II 20-50
23160 1 v5 3 LBF
26180 1 v3 3 LBF
29530 1 vi 3 LBF
33810 1 v6 3 LBF
40330 1 X2 3 DLR II 25-38
6190 1,2 v9 3 LBF
6610 1,2 v11 3 LBF
7110 1,2 v7 3 LBF
7560 1,2 v10 3 LBF
8580 1,2 v8 3 LBF

Specimens: Rei:tangular, C¡l-Ep tabs; IE-Recordi i g by Static Tests \vith Clip Gauges

Table 9. Results with material type 1A, (5x92146;0°) / L20/SL; R = -1.

142
Stiff n. Reduction
Loadcycles Strain Specimen Freq. Tab After Comment
[%] Code [Hz] [%] Cycles

10190370 0,6 03-3 10 Gl-Ep 0 1E + 07 Run Out


32976300 0,7 03-24 12,5 Gl-Ep 0 2,4E + 07
38798600 0,7 03-23 12,5 Gl-Ep , 0 2.4E + 07
17549300 0,7 03-26 12,5
42675000 0,7 03-25 12,5
22889850 0,7 1A-27 12,5
2397210 0,8 02-7 8
2057270 0,8 03-10 8
3224920 0,8 03-8 8 Steel 0 676390
4967200 0,8 3,2 5 Gl-Ep 1 2180000
5359050 0,8 03-7 8 Gl-Ep 2 2800000
5863800 0,8 02-11 8 Gl-Ep
6183280 0,8 03-1 5 Gl-Ep 0 1730000
556830 1 03-13 5 Gl-Ep
751610 1 02-6 5 Gl-Ep
826140 1 03-2 5 Gl-Ep
1162770 1 03-12 5 Gl-Ep
1606000 1 03-6 5 Gl-Ep
60220 1,2 03-9 2,5 Gl-Ep 0 1000
165428 1,2 03-20 5 Gl-Ep
175870 1,2 03-11 2,5 Gl-Ep
302190 1,2 02-10 2,5 Gl-Ep
410880 1,2 03-21 5 Gl-Ep
41470 1,4 03-14 2,5 Gl-Ep
54530 1,4 03-15 2,5 Gl-Ep
12770 1,4 1A-29 5
35720 1,4 1A-30 5
37300 1,4 1A-28
60320 1,5 1A-C34 2,5 Gl-Ep
26970 1,5 03-17 2,5 Gl-Ep
23330 1,5 03-18 2,5 Gl-Ep
1240 1,6 03-17 2,5 Gl-Ep

Shape of Specirr en: Rectan guiar; E-Rec<>rding by Sal ic Tests wit!ι Clip G auges

Table 10. Results with material type 1A, (5x92146;0°) / L20/SL; R = 0.1.

143 -
Code Tension Passes Load Cycles Max. Strain, % Failure
2A-16 Τ 76 975156 1.6 normal
2A-13 Τ 120 1539720 1.4 normal
2A-10 τ 122 1565382 1.4 normal
2A-18 τ 314 4028934 1.4 normal
2A-12 τ 340 4362540 1.4 normal
2A-3 τ 710 9110010 1.2 normal
2A-1 τ 1240 15910440 1 Runout
2A-1 Res. Strength 2.1 Tension

Table 11. Results of normal WISPERX tests.

Code Compression Passes Load Cycles Max. Strain, % Failure


2A-53 C 60 769860 1.6 max. Compr.
load
2A-50 C 86 1-090635 1.6 max. Compr.
load
2A-49 C 165 2117115 1.6 normal
2A-51 c 183 2348073 1.4 max. Compr.
load
2A-52 c 289 3708159 1.4 normal
2A-46 c 655 8404305 1.4 near max.
Compr.-load
2A-47 c 1715 22005165 1.2 normal

Table 12. Results of reverse WISPERX tests.

144 -
Max. Strain WISPER WISPERX
[%] Normal Reverse Normal Reverse
1.2 (calculation) 950 8690 1200 15300
1.2 (test) —- 710 1715
Damage Accumulation Factor D: — 0.59 0.11
1.4 (calculation) 168 1880 235 2518
1.4 (test) —- 199 326
Damage Accumulation Factor D: — 0.85 0.13

Table 13. Palmgren-Miner calculation with polygon spline.

Max. Strain WISPER WISPX


[%] Normal Reverse Normal Reverse
1.2 (calculation) 370 7650 710 18030
1.2 (test) —- 710 1715
Damage Accumulation Factor D: — 1 0.1
1.4 (calculation) 70 765 155 1604
1.4 (test) —- 199 326
Damage Accumulation Factor D: — 1.28 0.2

Table 14. Palmgren-Miner calculation with cubic spline.

145
2.2.1.10 Figures

Figure 1. Specimen with dog bone shape.

110 70

.o
CM

250

CM
•O)

τ:
'C\i
Γ di
without tabs, when GRP tabs (±45°) for (N
LBF- antibuckling guide antibuckling guide,
was used type DLR II

Figure 2. Specimen with straight shape.

146
Figure 3. A ntibuckling guide, DLR Type I.

­Φ­ <î>
­f­
4 -è·

J+L
ππ^ζιυζ τττ­
Æ
7|7. ■
T=Èj5= ­tn tn

CUUFIKC PLATE PTFE-INTERLINIHO

Hl·
7- g <?>

li "3 -&· -φ
­ββ­

•118­
t"
Figure 4. A ntibuckling guide, DLR Type II.

147
"hi« W Ψ

$
Ή-1
/CENTRE BORE HOLE
k- M -*i
PECIMEN 250·20·2.5
ΠΊ KV'v'-v I GFRP-TAB 55.20.2.2

Τ' * V'»» ' PTFE-INTERLINING

Figure 5. Antibuckling guide, Type LBF. .

Figure 6. Grippings affected by steel tabs.

- 148
Figure 7. Plot of constant amplitude test at R = -1 for specimen type 2C loaded with ±0.8%
strain.

149
3.0
co
c Δ Static Tension Test
Zi
V Static Compression Test
(D
00

2.5 ¿ï- RCS = Residual Compression Strain


α
ω

3
ω
2.0
φ
χ RCS
ω
I

a w
Τ3
ο-
(D
α
ΓΟ
Ο sco 1.5
L_
CO
ΟΙ Χ \ΚΏ
1.0 %}-EE& -B—
[XXX UD CD

X R = ­1
X χ:κ χ l/
D R = 10 χ
(1 χ 1023/45° / 1 , 5 χ UD940/0°)8ym / Μ10/Μ9 (Type 2C)

.0
1E0 1E1 1E2 1E3 1E4 1E5 1E6 1E7 1E8
Load Cycles η
(û 3.0
c
­ι
ω
Δ Static Tension Test
io V Static Compression Test
o.
m
2.5
S

3
Bl

BL .o :F
o

­5
"O
m
ro C
> 'CO 1.5
ι— X X
CO
D
1.0 * X
­BEl· * l · ^
D LTUD

D R = ­1
X R = 10
LU
Κ
DIED
(1 χ 92125/45° / 2 χ 92146/0°)sym / L20/SL (Type 2A)

1E0 1E1 1E2 1E3 1E4 1E5 1E6 1E7 1E8


Load Cycles η
(O
3.0
c
Δ Static Tension Test
V Static Compression Test
2.5
Q.

O
3
ft
(B 2.0
­ι
i!
o
­σ
ra O
g
CO 1.5
1— o «*>
co
i m
o OOO
sjvvuautø/
1.0 /TVJRMS ^ ^ ^

/Tvr\ 7 P ^ T N o«» <8>


< /
0 R = 0.1
X R = ­1 * * 3
5 χ 92146/0° / L20/SL (Type 1Α)

.0
1E0 1E1 1E2 1E3 1E4 1E5 1E6 1E7 1E8
Load Cycles η
105

0 100 200 300 400 500 600 700 800


Passes Through Load Sequence WISPERX

Figure 11. Stiffness changes, displacement- and extensometer-measured.

153
­π
iE'
3.0
c

33
ro 2.5
ui
c
¡Λ
O

o
ω Χ Residual
α Δ Static Tension Test
Ui
ro 2.0 " Tension
J2
C c V Static Compression Test
ro 'CO Strain
3 i—
O +­·
ro CO
3
ro
ω OJ ΟΧ> Ο
ui
c E
C 1.5
ro
3
)<Χ Ο <*( ο
ro
3
χ O
cl
en 1.0 ^
TJ
m
33
X

O Compression
X Tension
(1 χ 92125/±45° / 2 χ 92146/0°)8ym / L20/SL (Type 2A)
.0
1E0 1E1 3 1E2 3 1E3 1E4
Passes through Load Sequence WISPERX
-π ,i()
vQ
C

ro
—L

ω
5 χ 92146/0° / L20/SL (Type 1A)
O
0
3 2.5 (1 χ 92125/±45° / 2 χ 92146/0°)sym / L20/SL (Type 2A)

B
3
T
**
ai
3
■o

~
c
α
ro 0s- 2.0
ro C
α CO
l_
ω
CQ +-■
3 CO
3 1.5
S
5
T3 CO
o c
l_
><
(O
o ω
■*->

3
UI < 1.0
■o

-2.5 -2.0 •1.5 -1.0 .5 .0 .5 1.0 1.5 2.0 2.5


Mean Strain, %

ta a.u
c
-*
ra

J>
5 χ 92146/0° / L20/SL (Type 1A)
O
o
3 2.5 (1 χ 92125/±45° / 2 χ 92146/0°)sym / L20/SL (Type 2A)
UI

sr
3
*+
αϊ
3
-o
­♦
c
Q.
ra 0s· 2.0
ro C
α CO
αϊ
IQ
1_
■*-·

CO
3 D) 1.5
—i. S C
ZT ■*-·
o CO
' c 1c_
σ
o α>
UI +->
■o
5' < 1.0
ro

.0
-2.5 -2.0 -1.5 ■1.0 -.5 .0 .5 1.0 1.5 2.0
Mean Strain, %
ία
3.0
c

33 =ί O
o
2.5
'.'ii CD
uiro 3 O Gl.­Ep. Specimens of Types 1A , 2A, 2C
ωo
3 -* 2.0 CT
ÜL°
Sa
S'
^!ff
m ft ■I 1.6
äs- (ID
CO
<",? O O O
-α (θ
ro c
o ro
3 α
o o>
1.0 R = ­1 ­θ
3 ff
ui
S. O
S3
0 0 C ODODOvID
çjo
S° V Spar Beam P03 O OÜD O
ε­σ /

( Q =>
Δ Spar Beam P07 V Ο 0(

ou!
Q S
c 2.
S ra
2 3 O Specimens Type 2A with Buckling and Temperature Increase
S" 1 1 1 ι ι ί -
ö­l. 1E0 1E1 1E2 1E3 1E4 1E5
oi 5 1E6 1E7 1E8
ro ­,
o. ra Load Cycles η
Se
5ÕT
2.3 Steel

2.3.1 ECN, Welded Steel Details (P.W. Bach)

2.3.1.1 General
Another type of material used in wind turbine blades, hub and tower construction is steel.
For a mechanical construction the fatigue of a wind turbine blade is a quite unusual matter
due to the large number of cycles and the typical load spectrum. For welded steel con­
structions there are standards to calculate the fatigue life, but it is not clear to what extent
these standards are conservative.

2.3.1.2 Materials and Specimens


Steel is also used by some manufacturers for the blades of wind turbines. For this material
the weldings are usually the critical part for fatigue of the construction. For the investigation
two welded structural details were chosen which are also described as welded details no.
111 and no. 301 in the Dutch standard for fatigue of welded steel structures [1 ].
One welded structural detail was a transverse joint tapered in thickness and made from 4
mm and 8 mm thick plates of Fe 37-2 (see Figure 1 on page 162 for dimensions).
The other detail was a cruciform joint made of a 50 mm thick block of Fe 37-3 and 8 mm thick
plate of Fe 37-2 (see Figure 1 on page 162 for dimensions). The specimens were welded
according to a weld method specification and were checked for defects using an X-ray or
ultrasonic method.
The weld method specification, the nondestructive test results and the material data are listed
in Appendix A (in Dutch). The specimens were cleaned with steel grit no. 50 and coated with
a primer, as is practice in manufacturing steel tilades.

2.3.1.3 Results
Constant Amplitude Tests
Fatigue problems in steel often occur at the welded joint. Fatigue resistance is strongly
dependent on the quality of the weld. Cracks are often initiated near the weld toe where a
combination of a complex metallurgical state, geometrical notch factors and weld defects is
present. The magnitude and nature of stresses that will cause crack propagation are affected
by the presence of residual stresses, inherent defects in welds and adjacent parent metal,
surface defects and other stress raisers interfering with the flow of stress. As preparation for
the test with spectrum loading, the two types of welded structural details were fatigued with
an R-ratio 0.1. The results of these tests on transverse joints tapered in thickness and on
cruciform joints are depicted in Figure 3 on page 164 and Figure 4 on page 165 and listed
in Table 1 and Table 2 on page 160 .

According to the "Wöhlerlinienkatalog" [2] the fatigue limit for the cruciform joint is about 160
MPa. This is in agreement with the fatigue results bserved in this investigation.
Specimen failure was in all cases initiated from the weld toe at a small geometrical defect.
In Figure 5 on page 166 a series of photographs shows the fracture surface of the cruciform
joint, which is illustrative of the damage mechanism.
The obvious characteristics of a fatigue fracture surface are visible:
• crack initiation from a geometrical or metallurgical defect

- 158
• crack growth in concentric rings
• rather smooth fracture surface due to small microplastic deformation until final macro-
plastic rupture.

WISPER Tests
In a mechanical construction the fatigue of a wind turbine blade is a quite unusual matter due
to the large number of cycles involved (~109) and the typical load spectrum. In order to
determine to what extent the fatigue lifetimes which can be calculated from a standard for the
WISPER spectrum are conservative, WISPER tests are performed on two welded structural
details, a transverse joint tapered in thickness and a cruciform joint.
The results of the WISPER tests for several maximum load levels are listed in Table 3 and
Table 4 on page 161. These results are also shown in the log-log fatigue diagrams
Figure 6 on page 167 and Figure 7 on page 168, where the maximum stress in the Weibull
sequence is taken as ordinate.
For several maximum stress levels of the WISPER spectrum the fatigue lifetimes were also
calculated with the Dutch standard NEN 2063 without the load factor 1.2.
The range size levels of the WISPER spectrum, the number of cycles of WISPER, the stress
range and the K-value of the welded structural detail (K45 for the transverse joint and K35
for the cruciform joint) are put in a spread sheet. With this spread sheet the D3 and D5
damage is calculated and after summation of the damage according to NEN 2063, the
number of cycles until failure is determined.
These calculated lifetimes are also indicated in Figure 6 on page 167 and Figure 7 on page
168.
It is obvious that the calculated lifetimes are conservative, although the margin becomes
small in the higher stress ranges.
For the lower stress ranges which will be used in practice, NEN 2063 is at least one order
of magnitude more conservative as to lifetime.

2.3.1.4 Conclusions
Using variable amplitude fatigue tests with the WISPER spectrum on two welded structural
details it is concluded that the standard for fatigue NEN 2063 is conservative, although the
margin becomes small in the higher stress ranges.

2.3.1.5 Bibliography
[1] NEN 2063 Op vermoeiing belaste constructie NEN 2063, Nederlands Normalisatie Instituut,
Booglassen, (1988)
[2] Olivier, R. und Ritter, W. Wöhlerlinienkatalog für Schweissverbindungen an Baustählen, Teil 3:
Doppel-T-Stoss DVS Berichte 56/III, (1981)

159
2.3.1.6 Tables

Code R Fmax Signa Eps Freq Cyc 10? sr.cyc Coment


kN MPa 10" 3 Hz
A1SL 0.1 18.00 35 1160000
A13SL 0.1 30.00 167.00 35 46160000 run out
A20SL 0.1 38.00 210.00 25 140083
A21SL 0.1 38.00 211.00 25 590025
A14SL 0.1 38.00 211.00 35 19360000 run out
A19SL 0.1 45.00 250.00 30 60048
A18SL 0.1 45.00 250.00 30 140084
A16SL 0.1 45.00 250.00 35 310000
A15SL 0.1 45.00 250.00 35 480000
A17SL 0.1 45.00 250.00 30 1020000

Table 1. Results of fatigue tests with the transverse joints tapered in thickness.

Code E Pnax Signa Eps Freq Cyc 10? sr.cyc Coment


kN MPa Hz
B3FL 0.1 50.00 139.00 30 39000000 run out
B6FL 0.1 68.00 189.00 25 500000
B7FL 0.1 68.00 189.00 25 920000
B4FL 0.1 68.00 189.00 30 1330000
B9FL 0.1 68.00 189.00 25 9000000 run out
B5FL 0.1 68.00 189.00 30 10000000 run out
B8FL 0.1 68.00 189.00 25 10000000 run out
B2FL 0.1 70.00 194.00 30 1730000
B10FL 0.1 100.00 278.00 20 88031
B12FL 0.1 100.00 278.00 20 120150
B11FL 0.1 100.00 278.00 20 170033

Table 2. Results of fatigue tests with the cruciform joints.

160
Specimen F a RS N N„ Remarks
max
RT-A-K45 max MPa kN/s WISP
kN

26 27 150 200 7^9 1.0E8 Run out


23 36 200 25Ο 93 1.2E7
24 36 200 250 101 1.3E7
25 36 200 250 201 2.7E7
36 45 250 45Ο 21 2.8E6
42 '45 25O 45Ο 27 3.6E6
'β '15 25O 450 35 4.5E6
30 54 300 500 13 1.8E6
38 51 300 600 6 8.4E5
'to 54 300 500 11 1.5E6
31 63 350 600 4 5.5E5
35 63 350 300 4 5.2E5
39 63 350 600 6 6.1E5

Table 3. Results of WISPER tests with the transverse joints tapered in thickness.

Specimen F σ RS Ν N Remarks
max f
max
CJ-B-K40 kN kN/s WISP
MPa

17 72 200 500 749 1.0E8 Run Out


13 90 250 600 59 7.8E6
15 90 250 600 32 4.3E6
16 90 250 600 15 6.0E6
19 108 300 600 27 3.5E6
18 108 300 600 20 2.7E6
20 108 300 600 18 2.4E6
21 126 350 600 5 6.4E5
22 126 350 600 7 9.3E5
23 126 350 600 7 8.8E5

Table 4. Results of WISPER tests with the cruciform joints.

- 161
2.3.1.7 Figures

¿s

St> Ψ

HlllliniHI.il

STUK
NR. MATERIAAL EN/OF TEK. NR. CODE NUMMER
VORM EN PLAATSTOLERANTIES 25 - t 2 25.Õ - ± 0.2 25.00 - ± 0.02
RUWHEID VLGS.
VLGS. NEN 3311 25-3:1 25.Õ-10.1 25.00-i0.01 NEN 6301 ■ 3632
1.2
T.B.
DATUM
4J2.87
25 - ± 0,5 25.0 - ± 0,05 25.00 - ± 0,005
- 25 - : SYMM. t.o.w. HARTLIJN €3 3634 - 3638 - 3639

GECONTR. TITEL:
HFD. GROEP PROEFSTUK 2
HFD. TEK. K.
OPDR.GEVER WIJZ. OAT. P AR

ONDERWERP:

LM
ENERGIEONDERZOEK CENTRUM NEDERLAND
AUTEURSRECHT VOORBEHOUDEN BIJ DE WET OPDRACHT NR. FORMAAT A<

Figure 1. Welded structural detail: transverse joint tapered in thickness.

162
¿5

Lir

s>
ψ \ll 0.5 A

tudens het stellen In lasmal 7


STUK
NR. MATERIAAL EN/OF TEK. NR. CODE NUMMER
VORM EN PLA A TSTOLERA NTIES 25 ­ t 2 2S.0 ­ ± 0,2 25,00 ­ ± 0,02 PROJ. METH. RUWHEID VLGS.
VLGS. NEN 3311 25­±1 25.Õ­Í0.1 2S.0Õ­±0.01 NEN 6301 ­ 3632
DATUM 25 ­ ± 0,5 25,0 ­ i 0.05 25.00 ­ ± 0,005 3634 - 3638 - 3639
V­2
­ 25 ­ : SYMM. t.o.v. HA RTLIJN
GETEKEND T.B. 3.12.87
GECONTR. TITEL:
HFD. GROEP PROEFSTUK 1
HFD. TEK. K. SAMEN ST. τκα
TAL
PAÎT
OPDR.GEVER WIJZ. DA T. PA R HFDT
ONDERWERP:

RD
ENERGIEONDERZOEK CENTRUM NEDERLAND
AUTEURSRECHT VOORBEHOUDEN BIJ DE WET OPDRACHT NR. FORMAAT A 4

Figure 2. Welded structural detail: cruciform joint.

163
ΙΏ
C STEEL GRAIN SIZE VELO
πι
ω ST 3 7 ­ 2 4ram B (A STM E l 12] DOUBLE BUTT
ST 3 7 ­ 2 Bram

U3
c 1000
ra
TRANSVERSE JOINT
αϊ
ια

TAPERED IN
j THICKNESS
αϊ
o 500
CONST. AMPL.

O

3" STRESS
Π)
R - 0. 1
RANGE
ω [MPa]
3
UI
<
ra


Ol
♦ ♦♦♦ o
Õ;
5' ♦ ♦
m
■□
ra
-i
ra
α

100 ι 1 MIMI ι ι ι mil 1 1 ΙΙΙΙΙΙ ι ι t u m Ι Ι MIHI 1 I miti 1 I I I lilt 1 1 MIHI


4
LOG Ν
3
ra
UI
UI
ια
c STEEL GRAIN SIZE VELO
­t
ra
J>
ST 37­2 4mm B CA STM El 12] DOUBLE BUTT
ST 37­3 50mra

LO
C 1000
ra
CRUCIFORM
m
ια
­t
ω
3 LOAD CARRYING

i VELDS
Bl
O 500
C ; CONST. AMPL.
ra
o
:
STRESS
ra RANGE R - 0. 1
o
c [MPa]
o ;
o
­t
♦>
3

100 ι ι ι mil ι ι ι ιιιιι ι ι ι mil ι­ ι­ι ι ι ι ι ι l l ιιιιιι ι ι ι ιιιιι ι ι ι mil


4
LOG Ν
Figure 5. Photographs of the fracture surface of the cruciform joint.

- 166 -
ια
c
­t 3.0
ro
σι ­

ai = : ι - "Π
3 g 3 S ­
'S. « · <5·
Ä 3 > c 2.5
c » X ra
αοι o ­

n 5.·^ ια
S'3 3 ­
■ rr ro 3 2.0
» äo ­ A v B
^ .
­ \ ^
S
a M
"cg ­ \ \ A A
Q. 1> ^ ­
οι Ό m
­ . ■ ° 3D 1.5
o ®
3 .
§ro MAX. ­v. lü·^
­­Í2. STRAIN ­ ^
CXI ­ *
"■— ä \
mra— 1.0
Ü H J ­
c s · ro
ûi Ρ
o α 2.
3 _ α ­
« | |
u?o ­
^α 3ro o=
» βι 3
ro SÌ —
I3 =. ^2 . . J ι ιιιιιι ι ι HUM ι ι ιιιιιι ι ι ιιιιιι ι ι mm 1 1 MUM ι ι ιιιιιι " "
<° 5 3 0 1 2 3 4 5 6 7 8
-κ on?
o ' o
-i o ro
— " Q­
LOG Ν ι­
m
ro fl>4»
5­5",
o m 2.
3 α 3
A O'S
β) sr a
Sa»

ια
c STEEL GRAIN SIZE VELD
ro
^ι ST 3 7 ­ 2 4mm θ (A STM El 12] DOUBLE BUTT
ST 3 7 ­ 2 8mra
>-π
UI
u
o ια 1000
3 Γ
α TRANSVERSE JOINT
o α
D) αϊ
ία
a οι
οι =? TAPERED IN
roΟ THICKNESS
-<·
Η .

3-
Ξ
ro
o (Λ 500
01 TJ ; ♦ VISPER
o m
c 33
oi J ,
MAX.
ro ra STRESS !
o. UI
UI NEN 2063
^
s
■**■

φ
[MPa] • K 45
I-f

3 3"

ro
F*
Hl 1-7
φ
oi
η ·­♦■

o PÌ
u ­ι
ά UI
3 <
ca ro
in • i <
o ro
ζ o
m s
ζ UI
ro ' ' ' '""
sr 100 1 1 1 Mill 1 1 1 Mill Ι Ι 1 Mill ι ι ι mil 1 1 1 lllll 1 1 1 lllll I 1111»

ω
■π 4
ro LOG N
ro
α.
3

ro
ui
ui
ια
c STEEL GRAIN SIZE VELD
­τ
ra
oo
ST 37-2 4mm 8 (ASTM El 123 DOUBLE BUTT
ο ­π
ST 37-3 50nni
B> Q)
",5 1000
Β· CRUCIFORM
_ B>
—ta

li
M o
V)
Ξ
LOAD CARRYING
VELDS
° «5S
n 500
o en ♦ VIS PER
§3
_^ #♦
MAX.
=► ro
o U)
z (β
STRESS
• «> NEN 2063
ζ f. • K 35
[MPa]
SU
σι 5
" Q
c
o,
o
3
3
o'
5'

in 100
O 4
5' LOG Ν
α
õ'
οι
**.
ra
α
αϊ
-t
ra
ro
3. Components

3.1 RUG, Root Parts of Glass-Polyester Rotor Blades (W. Sys)

3.1.1 General
The results presented as S-N curves in Section 2.1.4 are generally used for comparative
purposes. The design of engineering components, however, can seldom be reduced to the
degree of simplicity with which S-N curves are associated. Thus the results obtained from
small laboratory specimens are of rather limited value to the designer. Conversions of avail­
able S-N curves or fatigue data will be necessary to construct the S-N curve of the structural
element under consideration. Converting the data of fatigue tests on small specimens to the
design conditions of interest is an exacting process with a fairly low level of reliability. The
outcome depends on the estimation procedures applied, which present numerous problems.
Several conversion steps may be eliminated if fatigue data or S-N curves are established
from tests on components. In order to gain more practical experience in the fatigue behaviour
of rotor blades, full scale fatigue tests on components are vital. The results of such full scale
fatigue tests on root sections of rotor blades are presented below.

3.1.2 Description of Components


Basically, fatigue testing of full scale rotor blades can be subdivided into two main groups:
• fatigue tests on entire rotor blades by forcing the blade to oscillate at a natural frequency;
• fatigue tests on parts of rotor blades by servo-hydraulic systems in load or in displace­
ment control.
The tests described in this section are performed in servo-hydraulic setups on root parts of
rotor blades. These parts can be obtained by cutting entire blades, or by fabricating special
root parts, ready for testing. The loading consists of one concentrated load acting at a point
about 2 m from the base of the root part.
Three different types of rotor blades were tested:
• Two blades of relatively old design (test numbers are VI1 and VI2). The main construc­
tional features are as follows:
- rotor diameter: 22.5 m;
- material: glass fiber reinforced polyester;
— specific mass: 1690 kg/m3;
— glass content (%WGT): 57;
— Ε-modulus (tension II): 30.7 GPa;
- construction:
two blade halves, made by hand lay-up technique, are connected together
before the material has cured;
- flange connection:
— type:
the blades are bonded to a flange in glass fiber reinforced polyester. This
flange is sandwiched between two steel flanges (trumpet type). Pre-
stressed bolts through the complete flange connect the blade to the hub;
— bolt hole circle: 508 mm;
— number of bolts: 18;
— bolt size and grade: M20x155 - 10.9;
— washer (diameter, thickness): 49 mm, 8 mm;
— tightening torque: 400 Nm;
— flange thickness: 110 mm

170
bottom flange: 25 mm
polyester : 30 mm
upper flange : 55 mm
• Three Polymahn blades of new design (test numbers are VI3, VI4 and VI6). The main
constructional features are as follows:
rotor diameter: 26 m;
­ material: glass fiber reinforced polyester;
— specific mass: 1630 kg/m3;
— glass content (%WGT): 50
— Ε­modulus (tension II): 22.3 GPa;
construction:
the load bearing part of the blade is a D­shaped spar which runs from the root
flange to the tip. This spar is made of wound laminate which consists of 80%
unidirectional fiberglass and 20% + 45° glass;
­ flange connection:
— type: pre­stressed shape as shown in Figure 1 on page 181;
— bolt hole circle: 753 mm;
— number of bolts: 24;
— bolt size and grade: M24x260 ­ 8.8;
— washer (diameter, thickness): 26 mm, 4 mm;
— thightening torque: 700 Nm;
— flange thickness: 174 mm
bottom flange: 40 mm
polyester: 40 mm
upper flange: 94 mm
• Two Polymarin blades of new design (test numbers are VI5 and VI7). This blade has the
same constructional features as the former with the exception of the attachment to the
flange:
­ flange connection:
— type: pin­and­hole as shown in Figure 2 on page 181;
— flange thickness (full steel): 65 mm
bottom flange: 40 mm
upper flange : 25 mm
— pin­and­hole connection:
18 bolts M30 at regular intervals on the same circumference.

3.1.3 Fatigue Testing

Special equipment is needed for full scale tests. The test setup is built according to the shape
of the components, the required forces and the resulting displacements. Two different setups
are shown ¡n Figure 3 and Figure 4 on page 182. The main components, such as actuators,
electronic control equipment, load cells and displacement gauges, are identical to those used
in standard servo­hydraulic testing machines. The test setup shown in Figure 4 on page
182, e.g., is suited for tests up to about 300 kNm bending moment; the maximum displace­
ment at the top of the root part is limited to about 220 mm.

3.1.4 Defining the Test Loads

Representation in a test of the whole spectrum of loads that occur to a component in service
is the ideal. This makes it possible to use simple conversion processes from test specimen
to real structure. In the case of rotor blades, test loads must differ from service loads in two
respects:

171 ­
• The amplitude of the test load must be high enough to obtain results within a realistic
period. In its lifetime, a rotor blade is subjected to about 109 cycles. The test must be
finished in about 107 cycles, which represent a test duration of about 4 to 5 months.
• The real load spectrum, which is very complex, can never be realized. When starting the
fatigue tests described in this report, no widely accepted load spectrum was available.
This was a good reason to perform cyclic constant amplitude tests. There is also another
reason: not everybody believes in the superiority of variable amplitude tests. Indeed,
Miner's rule (or a modified form of it) can rarely be avoided in design applications.
Consequently, the results of variable amplitude tests are of limited importance when
constant amplitude test results are missing.

The question of the maximum test load was difficult. The first tests on blades V11 and V12
were performed on an exploratory basis. The maximum test load in the other tests was cal­
culated according to P.H. Jensen et al. [1] who refers in his work to the "Draft to Danish Code
for Loads and Safety". The rotor blade is considered to be a statically determinate beam of
the built­in type. It is assumed that the load at right angle to the axis is a linearly distributed
load varying from zero at the rotor center to a maximum value at the free end. The intensity
of the load depends on whether a static test or a fatigue test is envisaged.

The intensity or resultant Ps of the static proof load is defined as follows:

_ (m2swept- rotor- area) ■ 300N/m2


s
no.blades
In a static test, the blade should resist Ps without damage (according to Risø certification
criterion for static load).
The intensity P, of the fatigue proof load is defined as follows:

P,= A 3.2

In a fatigue test, the blade should endure 107 load cycles fluctuating between 0 and P, without
any apparent damage. This is a part of Risø's certification criterion for fatigue loads.
The test loads applied to the present full scale components were based on the aformentioned
fatigue proof load P, in the sense that the test load creates a bending moment M in the flange
section that is equal to the bending moment caused by P,. The shear forces, however, are
about four times higher than in the case of P, loading. For a three bladed 22.5 m rotor the
bending moment M in the flange section amounts to 141 kNm; for a three bladed 26 m rotor
to210kNm.
A simple procedure is used to reduce the test time from 107 cycles to 5 · 106 cycles. Firstly,
as in the case of the S­N relationship, a linear log(M)­log(N) relationship may be adopted:
Μ=0·/ν* 3.3
Then, it can be assumed that the constant k is equal to the value obtained in coupon testing.
Results from such tests indicate k values in the range of ­0.08 to ­0.12 (in Section 4.1.4, the
k values which are equal to ­1/b, are in the range of ­0.084 to ­0.122). The constant C can
now be determined by assuming that the point N = 107 cycles and M = 210 kNm lies on the
M­N curve of 100% survival. With M in kNm and N in number of cycles, the constant C =
1052 if k=­0.1 is adopted. Therefore, Equation (3.3) can be written as:

Μ=1052·ΛΓ°­ 1
Substituting Ν = 5 ■ 106:
M =225 kNm

172 ­
If the Mandell law is used (M-logN curve is linear) with a coefficient of 0.1, M amounts to 231
kNm instead of 225 kNm. Taking a conservative view, M = 231 kNm was adopted.

3.1.5 Test Program

The test program is summarized in Table 1 on page 177.


At regular intervals the varying magnitudes of different quantities are simultaneously
recorded:
1. The load: this is the servo-controlled parameter.
2. The deflection of the root part at the load application point. In the course of the test, the
maximum deflection may increase if noticeable damage occurs.
3. The relative motion between the GRP material and the steel flange.
4. The prestress in bolt no. 1 situated in the most stressed area. Two strain gauges on the
head of the bolt are used to measure this parameter.
5. The prestress in bolt no. 3. The bolts are numbered consecutively.
6. The strain in the GRP spar where the tensile stress is maximum in a section as shown
in Figure 5 on page 183.
7. The strain in the GRP spar where the compressive stress is maximum.
8. The strain in a representative point on the steel flange.

3.1.6 Results

Components VI6 and VI7 were subjected to static tests in order to measure the following:
• strains at selected places in the glass fiber reinforced polyester and in the steel parts;
• the stiffness of the component;
• the stress variation, due to external loading, in the preloaded bolts of the flange con­
nection.
In the course of a fatigue test of each component, periodical inspections were performed at
zero load. During these inspections, the bolted connection was always unscrewed and
tightened again to the required preload, whilst the prestress in two bolts was measured.

Static tests
For the component VI6 with flange of the prestressed shape type, the positions of the strain
gauges are shown in Figure 5 on page 183. Strain gauge K7, not shown in this figure, is
diametrically opposite to K6. Before performing measurements, the component was loaded
several times to eliminate effects of initial incompatibilities. The applied maximum loads are
equal to these applied in the fatigue tests, i.e., in one direction, taken here as the plus
direction, 113 kN creates in the flange section a bending moment of 231, kNm and in the
minus direction, a load of 56.5 kN creates a bending moment of 115.5 kNm. The side of the
component shown in Figure 5 on page 183 is in tension when the load acts in the plus
direction.
The results are summarized in Table 2 on page 177.
The loading path was as follows:
0 - 113.7 - 0 - 113.7 - (-56) - 0 - (-56) - 0 kN
The load-strain diagram for strain gauge K6 is shown in Figure 6 on page 183. A quasi-linear
diagram is obtained. At the end of the test at zero load, a small strain offset is generally
observed.

- 173
Additional external loads cause additional stresses in the bolts. These stresses, which must
be added to the prestress, are shown in Figure 7 on page 184. The load is on the abscissa,
and strain reading in μ β ^ ί η from the bolt head gauges is on the ordinate. As will be shown
later, a calibration procedure provided the following scale: 100% prestress = 700 Nm torque
= 460 MPa stress in the tensile stress area of the bolt = 528 μ β ^ ί η reading from the above
mentioned gauges. The relationship load-variation of prestress is not linear and sensibly
confused. This phenomenon was experienced throughout all strain measurements on bolts.
Table 3 on page 178 summarizes the results of Figure 7 on page 184. From this table it can
be seen that the strain gauge readings are small. For the complete cycle 113.7 kN to -56 kN,
the greatest variation in prestress amounts to about 9.7%. The theoretical prestress is about
460 MPa, hence the maximum variation is about 45 MPa.
The anomaly in the displacement at the load application point, as may be seen in Figure 8
on page 184, is due to the test set-up where the displacement at zero load is not unambig­
uously defined. It depends upon the direction in which the load is acting; when load reversals
occur, the compressive stress distribution, induced by the external load, changes completely
in the vicinity of the point where the load is applied. These compressive or bearing stresses
have an influence on the measured displacement. The first load reversal seems very normal,
but after this, an offset of about 1.6 mm is originated. Nevertheless, the stiffness for both
loading directions is about 0.15 mm per kN.
The above mentioned strain measurements were repeated a few days later, and practically
identical results were obtained.
For specimen VI7 with pin-and-hole connection, the strain gauge locations are shown on
Figure 9 on page 185. Gauges R1 to R8, K6 and K7 are attached to GRP material, whilst
B1, B2, K8 and K9 are on steel: The results of the strain measurements are summarized in
Table 4 on page 178.

Preload of bolts
The bolts with which the rotor blade is connected to the hub are preloaded with a tightening
torque of 700 Nm. In a fatigue situation, the preload is of critical importance. Hence, in the
course of a fatigue test, the preload was regularly checked and adjusted.
Two bolts were instrumented with strain gauges on the head (see Section 3.1.5). The strain
in the head of the bolt is proportional to the preload in the bolt, within the load range of
interest. Hence, a calibration must be performed that provides the desired transformation
between the strain in the head of the bolt and the tightening torque or the preload. In order
to achieve realistic calibration figures, it is necessary to measure in actual conditions. The
following approach was adapted:
• The bolts were calibrated in their real situation without external loading on the compo­
nent.
• Torque was applied with a conventional used torque spanner.
• Strain gauge readings were performed before and after the tightening of all bolts of the
connection. These measurements were repeated 40 times. Finally, a statistical analysis
was used to define the distribution function, the mean and the standard deviation.
The distribution function of the preload of bolt 1 is shown in Figure 10 on page 185. For bolt
1, the mean strain in the head, for 700 Nm torque is 528 microstrain, whereas for bolt 3, a
mean value of 492 microstrain was found. This experimental verification of the preload shows
that its magnitude is not clearly defined when commonly used tools are employed to measure
the torque.
In the course of a fatigue test, the preload has a fading tendency. The evolution of this
phenomenon is shown in Figure 11 on page 186. In this figure the number of cycles is on

174
the ordinate and the preload on the abscissa. The preload is expressed in % of the calibrated
mean preload. The "O's" in the diagrams represent the preload, just before executing a block
of about 0.5*106 cycles. In a perfect situation all these O's must be on the 100% line. The
stars * represent the measured preload after execution of a block of about 0.5*106 cycles.
The dashed lines simply connect the points representing the preload at the beginning and
the points representing the preload at the end of a block of cycles.
The figures show that a decrease of preload is likely to occur in the beginning of the loading.

Results of fatigue tests


The results for components V11 and VI2 (rotor diameter 22.5 m) are shown in Table 5 on
page 179.
In Table 5 on page 179, L(m) denotes the distance between the flange section and the force
application point; hence, the applied force multiplied by L gives the maximum bending
moment in the root part. The deflection is measured at maximum load at the point of force
application.
For VI1 the first damage occurred after 17,800 cycles at 200 kN maximum bending moment.
The damage was situated at the point of load application. The force was too concentrated
and caused local damage. The final fracture was a shear fracture located at the leading edge.
This fracture is not representative of a real situation. As explained in Section 3.1.4, shear
forces in the test conditions are three to four times higher than those in service conditions.
For VI2 the load was in fact too high as compared to the loads decribed in Section 3.1.4. The
final fracture after 470,000 cycles was, as in the case of the former component, a shear
fracture at the leading edge. Thus, again this fracture is not representative of a real situation.
The strain, measured about 20 mm from the upper part of the flange assembly, was about
3600 μβίΓβίη at a maximum bending moment of 225 kNm. During the tests, problems arose
with fatigue fractures in the bolts. These problems were one of the reasons to stop the fab­
rication of blades having dimensions in the root section as indicated in Section 4.1.2.
The results of fatigue testing of VI3 to VI7 are summarized in Table 6 on page 179. The
deflection in column 5 is based on measurements performed during fatigue loading; the result
is given in mm per 100 kNm bending moment. The strains given in columns 7 and 8 were
measured during the fatigue test by strain gauges K6 and K7, respectively (see Table 2 on
page 177 and Table 4 on page 178).
The quantities of columns 5, 6, 7 and 8 do not remain constant during the fatigue test. The
results in Table 6 on page 179 are the measured magnitudes after about 10,000 cycles. The
variation of these magnitudes during the fatigue test is plotted in graphs as shown in
Figure 12 on page 187 and Figure 13 on page 188. In these figures, the magnitudes are
made dimensionless by plotting the ratio of the magnitude at Ν cycles to the magnitude at
10,000 cycles.
It is reasonable to accept that the behaviour of the component is excellent if the four quan­
tities mentioned remain constant from the beginning of the test to the end. On the other hand,
it is reasonable to have doubts about the behaviour of the component when the graphs show
important deviations from the horizontal line. In this respect several comparisons can be
made:
• Looking at the results for the R = 0.1 tests (VI3, VI4 and VI5), it is clear that VI5 (pin-
and-hole) behaves better than VI3 and VI4.
• Looking at the results for the R = -0.5 tests (VI6 and VI7), here also it is quite clear that
VI7 (pin-and-hole) behaves better than VI6.

175
• Looking at the results for R = 0.1 and R = -0.5 tests, as expected indeed, the R = -0.5
tests cause more damage than the R = 0.1 tests.
The cyclic variation of prestress in two bolts owing to cyclic loading was also recorded as
mentioned in Section 3.1.5.
The variation of the prestress (maximum prestress-minimum prestress) is summarized in
Table 7 on page 180.

3.1.7 Conclusions and Discussion


Two full scale tests showed fractures which were not representative of a real situation. The
test loads were too high. Four full scale tests endured 5 · 106 cycles under modified loading
based on the 300 N/m2 criterion. These blades did not show any apparent damage. One
blade was tested under a real 300 N/m2 criterion load. That blade, which had withstood
5 · 106 cycles in a former test, was subjected to another 5 ■ 106 cycles and achieved 107
cycles without any apparent damage.
The recording of several parameters such as strains, displacements and prestress, was
aimed at providing a measure of the seriousness of the induced damage. I t is recognised that
this measure does not provide a quantitative characterisation of the damage.

3.1.8 Bibliography
[1] Jensen, P.H., Krogsgaard, J., Lundsager, P., Rasmussen, F. Fatigue Testing of Wind Turbine
Blades European Wind Energy Association Conference and Exhibition, Rome, Italy, October 1986.

176
3.1.9 Tables

Nr. Type of Mmax R f End of test


flange (kNm) (M m i n /M m a x ) Hz

VII Hütter tested on an explciratory


V12 Hütter 225 0.1 - failure
V13 Pre-stressed 231 0.1 1.5 5.106 or failure
shape
V14* Pre-stressed 210 0.05 1.5 10.106 or failure
shape
V15 Pin-and-hole 231 0.1 1.5 5.106 or failure
V16 Pre-stressed 231 -0.5 1.5 5.106 or failure
shape
V17 Pin-and-hole 231 -0.5 1.5 5.106 or failure

Table 1. Test program of full scale tests.

Bending moment in flange section (kNm) 232 -115

strain Rl (microstrain) GRP 1511 -749


R2 1559 -853
R3 1682 -885
R4 2420 -1338
K6 2074 -1115
R5 1686 -987
R6 1785 -1093
R7 1598 -1021
R8 1479 -944
K7 2446 1198

strain K9 (microstrain) STEEL 11 -8

displacement at load application point (mm) 17.6 -10.5

Table 2. Results of static tests on VI6.

- 177
External load Strain gage reading Total stress in bolt
(kN) (pstrain) (% of prestress)

0 0 100
113.7 42 108
0 25 104.7
113.7 51 109.7
0 28 105.3
-56 0 100
0 22 104.2
-56 3 100.6
0 24 104.5

Table 3. Variation of prestress in bolt.

Bending moment in flange section (kNm) 227 -116

strain Rl (microstrain) GRP 695 -447


R2 708 -593
R3 756 -535
R4 789 -547
K6 928 -631
R5 835 -639
R6 819 -823
R7 866 -826
R8 703 -551
K7 -1009 478

strain Bl (microstrain) STEEL 104 -84


K8 93 -49
B2 86 -65
K9 43 -9

displacement at load application point (mm) 15.2 -8.3

Table 4. Results of strain measurements on VI7.

178
tn Α χ
R L Defl. N Damage
(kNm) (m) (mm) cycles

Component 1

80 0.38 4 67 175,300 no
90 0.33 4 75 173,400 no
200 0.1 2.5 96 17,800 yes

After reparation and reinforcement:


200 0.1 2.5 101,200 no
225 0.1 2.5 60 313,000 fract.

Component 2

225 0.1 2.5 62 470,000 fract.

Table 5. Fatigue tests on VI1 and VI2.

1 2 3 4

Specimen Flange Stress Maximum Deflection Motion


type' ratio bending flange/
moment GRP
(R) (kNm) (mm/lOOkNm) (mm/lOOkNm)

V13 p.s. 0.1 231 7.50 0.125


V14 p.s. 0.05 210 7.25 0.240
V15 p&h 0.1 231 5.74 0.053
V16 p.s. -0.5 231 7.34 0.165
V17 p&h -0.5 231 6.47 0.069

* p.s. = prestressed shape


p&h = pin-and-hole

10 11

Specimen Maximum Maximum Cycles Fai lure Damage


strain strain
tension compres.
(ustr.) (ustr.) (N) (Yes/No)

V13 1932 1999 5.106 Ν Figure 3..12


V14 1323 1607 107 M Figure 3..12
V15 577 633 5.106 Ν Figure 3..12
V16 2116 2444 5.106 M Figure 3..13
V17 935 935 5.106 Ν Figure 3,.13

Table 6. Results of fatigue testing VI3 to VI7.

- 179
Component R % variation
Bolt 1 Bolt 2
V13 0.1 3.5 4.5
V14 0.05 4.0 4.3
V15 0.1 3.8 5.4
V16 -0.5 7.9 5.5
V17 -0.5 7.8 7.2

Table 7. Prestress variation in % of prestress-at-zero external load.

180 -
3.1.10 Figures

COMPOSITE

Figure 1. Flange and prestressed shape connection.

COMPOSITE

Pin & Hole bolts

65mm
0
φ 558 Loading d i r e c t i o n

φ 640mm

φ 753mm

Figure 2. Flange and pin-and-hole connection.

181 -
Figure 3. Test setup for fatigue tests on root parts.

Figure 4. Test setup for fatigue tests on root parts. Bending moments up to 300 kNm are
feasible.

- 182 -
Figure 5. Strain gauge location for component VI6.

E400
s

-z.
< 1600 ÍS
a:
\—
en
¿r

i 800.
Ζ
<T
OC .000
co

­800
¿?
­40.0 ­.000 40,0 80.0 120
'k N

Figure 6. Load/strain curve from strain gauge K6 of test VI6.

­ 183 ­
—~ t^A n -
34 . U
Ζ
<

ι— Ad . U -
CO
Α
ν?& y '
/
y

λy\
*, <ί?
Q 30.0 η /
< s
o
LU
tv
Ια.U /}
Áν ι
/
0_

ο . UU - ΛV r
/

Ψ /
-40.0 -.000 40.0 80.0 120.
kN

Figure 7. Variation in bolt preload with externally applied load (53 Listrain corresponds to
about 10% of initial preload)

Ί*ΐ n -
¿«J , υ

£

ε
Q
a . uu

q nn „
fifl ­


S
/

o
J . uu
X
/ /
_j J . uu ' J"
LL
UJ
Q
/
_Q
y, nn
uu ­

­40.0 ­.000 40.0 80.0 120.


kN

Figure 8. Load/displacement curve (test VI6).

184 ­
Figure 9. Strain gauge locations of component VI7.

0.3
BOLT1
Sample
s i z e = A0
m e a n : 528
Stdev= 67.3

100 300 500 700 900


Strain gage reading (microstrain)

Figure 10. Data from measurements at nominally 700 Nm torque and the probability distrib­
ution curve.

185
VL5
BOLT 1

Τ 'ΤΓΤΤΪΤΐ
Ι ΓΓ T f

.00 1.0 2.0 3.0 4.0 5.0


Ν lEB-
c cy ìes)

VL5 VL7
SOLT 3 BOLT 3

'ΓΓτν

'" 'τπιΐ'Τ ' ΙΠΐΐΓΓπη

.00 1.0 2.0 3.0 4.0 5.0 .00 1.0 2.0 3.0 4.0 5.0
Ν (E6-cycles) Ν (εε-cycles)

VL3
BOLT 1

Χ» wft

.00 1.0 2.0 3.0 4.0 5.0


Ν lEe-cycles)

VL3 VL 4
Β 3LT 3 BOLT 3

-pr-T- m

|
.00 1.0 3.0 3.0 4.0 5.0 .00 1.0 2.0 3.0 4.0 5.0
Ν (E6-cycles) Ν (E6-cycles)

VLB
SOLT 1

Γ
Jrn< "ΠΙΤΤΥ

Ι i'rntïïfl
Q JOO-
I
Q
ii 'h Ι
I!
I

u
.00 1.0
Ι Ν
2.0 3.0
(EB-cyclee)
4.0 5.0

Figure 1 1 . Measurement of preload in bolts at zero external load.

­ 186
tuima
GFRP

rn~\Jm'
S MJt
fl - o .i
­ 231«*
*V^ min · * * *

* ETMJ>

.00

FATIGUE/flOTOHBLADf

GFRP

fl ­ 0.05

Λ^ l M / ­ 210kHM

1,10 ­ «*»

'Tr ■ fUÄ/PSLT.

Ä

FATIGUE/ROTOflaLAOE

MTUUit.

GF ñP

R · 0.1

Μχ ­ 23iJt*

^^=f^ *=*«= " S ^ £ ­in ­ 2 * *

• FUMC /PO.T.
■ tenni»
• muj>
■ ÏTBUW

«J'JUHIVÖCITEIT
ten

0 5 0 7.

Figure 12. Effect of endured number of cycles on four measured parameters at R = 0.1 for
VI3, VI4 and VI5.

187 ­
FATIGUE/ROTOHBLADE

TEST: 116

lUJBIUL

GFRP

R - ­0.5

H «a, ­ 23ikNm

«,,„ — llSrM»

ι FU«E/ra.T.
» OFimiw
> 5HU1H­
r STÎUIK.

tfflr
RIJKSUNIVERSITEIT
«NT

L o g I NI

FATIGUE/ROTORBLADE

TEST: H.7 i

(U7SUIL

GFRP

fl - -0.5

«.in - " * »

ι FbWC E/PtLT.

« OEFurrow
► ÎT1UB4-

1 STMBH

fUJKSUHrVERSnETT
SEKT

Log fM

Figure 13. Effect of endured number of cycles on four measured parameters at R = -0.1 for
VI6 and VI7.

188 -
3.2 DLR, Spar Beams of Glass­Epoxy Rotor Blades (Ch.W. Kensche)

3.2.1 General

Fatigue data from coupon specimens are used for proper material selection and for a first
lifetime estimate of the structures to be designed. But in most cases they are of laboratory
quality and tested under ¡deal conditions, and hence represent optimum results.
A whole structure containing the same material may show reduced fatigue properties caused,
e.g., by
• effects due to the manufacturing process, such as inproper lay-up, flaws etc.,
• interlaminar shear stresses in the joint between the caps and the shear web of the rotor
blade spar,
• influence of the load introduction area,
• local instability effects.
Thus, investigations of whole rotor blades or components including the load attachment to
the hub are necessary! Therefore, several modified spar beams of industrially available rotor
blades were tested. The benefit was not only increased knowledge about the fatigue
behaviour of the material in the structure, but also to study inspection methods which were
in discussion for an operational wind turbine. Last but not least, only full scale fatigue tests
will produce reliable design allowables for the manufacturer and the certification authority.

3.2.2 Description of the Structural Components


General
The spar beams represent the inner load bearing part of the AEROMAN 12.5/40 rotor blades,
see Figure 1. This WEC (Wind Energy Converter) is a two bladed machine from MA N with
a diameter of 12.5 m and a rated power of 40 kW. The spar is D­shaped. The material is
similar to that of Type 2C, which was used also for the ε­Ν data in the specimen programme,
see Table 3. The resin system is the low temperature / low pressure epoxy prepreg M9 and
M10 from Ciba­Geigy. The load bearing layers of the laminate are made of UD fabric (UD940
/ M9), while the torsional and shear loads are carried by a ±45° fabric (1023 / M10). The
laminate is monolithic, i.e. without a sandwich core. The length of the component including
the steel flange is 2.8 m and it is flapwise transversely loaded at the 2.5 m position. The load
attachment to the hub is conservative and strong. The principle of the connection between
the Gl­Ep root and the steel flange is shown in Figure 2.

Modification of the Standard Spar Beams


For the investigations it was necessary to modify the series blade in order to get a well
defined test cross section. This was discovered after static tests on the first blade spar (POL),
which was taken from series production to check the test­bed for the structure investigations
[1]. It became clear that the location of maximum strain in the spanwise direction would differ
from one spar to another. Thus, a cross section with reduced wall thickness over a certain
length would enable determination of the location of maximum strain. This test section was
decided to be about 1 m distant from the wall. It had a length of 120 mm. The lay­up equaled
the stacking sequence and thickness of laminate Type 2C, see Table 5 on page 199. The
wall thickness increased over a length of 70 cm back to the original laminate thickness, see
Figure 1 and Figure 3.
This special design results from the requirement that by applying a transverse load of
7.5 kN corresponding to a bending moment of 18.75 kNm, a maximum strain of about 0.6

­ 189
per cent should occur at the test section. This transverse load was defined to be the design
load for the test spar beams.
The advantages of the special test cross section design can be summarised as follows:
• Failure is expected in the thin-walled area. Therefore, observation can focus nearly tot­
ally on this location, i.e. the strain gauges need only be applied here, and also the
inspection of the Gi-Ep parts of the spar can likewise be limited.
• The testing time could be shortened, since for producing a defined strain in the test
section, the deflection of a normal spar beam (as measured at POL, see Table 1) was
three to four times greater than one with the modification. Consequently, the testing
frequency would have had to be reduced drastically during fatigue loading, which would
have led to longer testing times.
• An additional advantage of this step was to reduce loads on the testing equipment.
Wall thickness reduction was accompanied by the disadvantage of an instability in that area,
which occurred relatively early. The first modified spar beam (P01) buckled reproduceably
already at a load of 6.9 kN and the nominal design strain was not reached. To prevent this
in future it was decided to foam out the relevant test section of the test structures.
Due to the thickness change a manufacturing problem came up, too. Before being moulded
all the prepreg layers of a spar are laminated on a flat plate and then laid down into the
female mould as shown schematically in Figure 4. However, in that procedure the single
layers of the laminate can shift slightly. Thus, in some components folds raised up transver­
sely to the span direction, especially in the test cross section. A severe manufacturing fault
which occurred at spar beam P06 is shown in Figure 16. The influence of that irregularity
was disastrous and will be described under Fatigue Tests on page 192.
Due to these and other attributes of the spar beams a classification into four categories was
made, see Table 1. The quality grades range'from very good to very bad, with the bad cat­
egories (3) and(4) given for cross folds in the laminate. Other manufacturing faults were
especially noted at the edges between the spar caps and the shear webs, where somtimes
large visible voids were produced.
As those problems were discovered only in the thin walled test cross sections and their
neighbourhood, it is supposed that they do not occur in standard spar beams. However, for
the fatigue investigation the unintended faults offered the advantage of obtaining manifold
and differentiated information about the fatigue behaviour of the various structures. Thus, a
look at the damage tolerance of the faults was possible, too.

3.2.3 Testing Procedure


In general the testing procedure for the spar beams was the following:
Measurement of frequency response in the virgin state
Static test (strains and deflection)
Fatigue test (constant amplitude or load sequence)
Interruption of the fatigue test at defined states, inspection
Frequency response
Static test (strains and deflection)
Continuation of fatigue test
Continuation of stiffness measurements
Stop either after failure or after a specified lifetime
The structures were inspected immediately after receipt from the manufacturer. Faults were
documented, see Table 1. Then strain gauges were glued on the test section in both span-

190 -
wise and - at the position of maximum longitudinal strains - in circumferential directions in the
pattern shown in Figure 9, as applied the first time on spar beam P01 [3].
Static and fatigue bending were performed by a vertically hanging 60 kN cylinder with an
actuator movement of ±200 mm at the 2.5 m position. For the purpose of minimizing the
time for the dynamic investigation, a triple testing rig was built where three spar beams could
be loaded simultaneously, see Figure 5, [1].
For static tests of only one component, e.g. for interim inspection purposes, the wooden
clamps of the other two spar beams could be removed quite simply. The deflection was
generally measured by an inductive sensor in the piston of the cylinder.
As shown above in the scheme of the testing procedure, frequency response measurements
were carried out at certain stages of lifetime. In Table 8 to Table 13 the life history of the
six fatigue-tested spar beams is shown together with the changes in eigenfrequency, strain
and deflection. The damping measurements were carried out in the test-bed beside the
fatigue testing rig where the blade was cantilever mounted at the wall, too. Figure 6 gives
a view of the whole test rig.

3.2.4 Results
Static Tests
In general each spar beam was loaded statically up to the design load of 7.5 kN to verify the
correct design and to determine the characteristic deflection and strain distribution.
Additionally to the quality of manufacturing and the design of the test cross section,
Table 1 shows the maximum longitudinal strain in the upper surface and the deflection of the
tested spar beams with respect to the design load of 7.5 kN. Table 2 on page 197 gives a
survey of the type of loading (statically or fatigue), the maximum applied loads and the cor­
responding bending moment, maximum strain and deflection. Additionally to that, Table 4
shows the manner of failure, and at which stages of life first failure appeared.
The investigations of the components started with a standard spar beam (POL) in order to
have a basis for comparing a series product with the modified spars. The maximum com­
pression strain was -0.16 per cent at the design load of 7.5 kN, and the corresponding
deflection 55.8 mm. It failed by compression in the upper surface, at a load of 33.3 kN. The
maximum strain was -0.69 per cent and the deflection 245 mm, see Table 1 to Table 4.
The deflections of the modified components with the antibuckling foam core are about
20 % higher compared to that of POL, but do not differ remarkably from one another. Only
spar beam P01 which was built without that support showed somewhat larger deflection.
However, the reference strains, measured on strain gauges at the same prescribed locations,
show dramatic differences compared to the mean value, see also Figure 8. This is related
to the manufacturing process, because the layers, which are laminated flat, may slightly
change their position when laid down in the female mould. Thus, the thickness of the lami­
nates at the same locations of the different spars can differ, which leads in consequence to
differences in the measured strains.

Spar beams P01 and P02 were loaded only statically. P01 was investigated carefully by
experiment and by analysis [3]. At the location of experimentally determined maximum strain
in the upper cap, the strains were also measured circumferentially. Figure 10 and
Figure 12 show these strains at different loads below the design load. The calculation was
carried out to check the predicted maximum strain in the test cross section at the 1 m posi­
tion. But additionally the stress and strain of the whole structure were computed assuming
pure bending load, and also a combination of both bending and torsion. For comparison the
spar was also loaded in an experiment under the same conditions, see Figure 7 [3]. In Fig-

191
ure 13 good coincidence between calculated and measured longitudinal strains can be
noted. Figure 11 shows the shear distortion. The computed and the measured strains differ
by a factor of 2. This can be explained by uncertainty about the G­modulus value. Moreover,
the computer program for the shell design assumes stiff disks for the cross sections of the
torqued beam, which leads to higher shear distortions than exist in reality. Generally the
comparison showed satisfying results. The static failure test of P01 then revealed, as men­
tioned above, that a foam core was necessary for the following spar beams to prevent them
from buckling.
Thus, spar beam P02 was tested statically up to failure to check the usefulness of the anti­
buckling foam core. This method was more successful, as the fracture load now exceeded
30 kN and the maximum longitudinal failure strain in the compression cap was ­1.49 per
cent, see Table 2, which is more than twice the design strain of ­0.6 per cent. The smaller
deflection at reference load of P02 and also of the following spars, compared to P01, can
be assigned to the stiffening effect of the foam core.
Finally, tests on P01 and P02 were required to define the possible static failure value and to
fix the reference (limit) load level for fatigue testing.

Fatigue Tests
Two structures, P03 and P07, were tested with constant amplitude fatigue at R = ­ 1 , i.e. equal
positive and negative loads. The applied reference moments and strains are shown in
Table 4 on page 198. For both simultaneously loaded spar beams local bulges of the size
of a large coin could be observed, however, after only a few load cycles. The reason was a
separation of the laminate from the relatively rigid epoxy foam at maximum load. The surface
temperature in the buckling area rose to ca. 40°C during fatigue loading. Though the maxi­
mum measured nominal strain was relatively low (­0.38 and ­0.5 per cent), the spars had
their first failure relatively early at 0.344 ■ 106 and 0.513 ■ 106 load cycles. Similar behaviour
was observed on coupon specimens where buckling was allowed and the temperature could
increase remarkably, see also Section 4.2.1.4. The failure pattern of the spar beams showed
that in the buckled area intralaminar shear failure had occurred along the UD fabric, which
initiated the failure. The elevated temperature on P03 can be related to these additional
buckling stresses. However, there was no remarkable increase in temperature on P07. Thus,
early failure in the spar beams arose mainly due to buckling stresses. This is confirmed by
observations at P05, which was loaded by load sequence and where no temperature
increase could be observed, although buckling behaviour was present, see Table 4.

In Figure 19 results for spar beams P03 and P07 and the two specimens which were allowed
to buckle at R = ­1 (see also chapter 4.2.1.4) are contrasted with the ε­Ν data of all Gl­Ep
data of specimens tested at R = ­ 1 .
Spar beams No. 4­8 (P06, P05, P10 and P09) were tested with the load sequences WISPER
and WISPERX. It was assumed that there is no influence of the omissions in WISPERX on
the fatigue performance of the spars. One pass through the sequence represents two
months' operation of a hypothetical wind turbine. The maximum longitudinal strain level in the
upper spar cap was slightly more than 0.6 per cent, relative to the reference bending moment
of 18.75 kNm.
In general the components showed very good fatigue behaviour. P05 failed after 398 passes
through the sequence, which would be equivalent to more than 65 years of operation of the
fictive WEC. The reason for failure was also a local instability, similar to that seen in the case
of the constant amplitude loaded spars. A temperature increase could not be observed.
A peculiar case of failure occurred at spar beam P06. Originally this worst manufactured
structure was to be refused as a substandard product. However, later it was decided to
investigate the influence of the above described deep crosswards fold on fatigue behaviour.

192
This fold got a first delamination, accompanied by a loud cracking, on the occasion of loading
the beam with the negative design load. A first delamination could be observed, see
Figure 16. After some WISPER passes a growth of the delamination took place and a sec­
ond one also appeared . This state of damage propagation after 14 WISPER passes can be
seen in Figure 17. The spar beam then failed catastrophically at WISPER cycle No.17 due
to rapid delamination, see Figure 18.
In this context it should be noted that the standard spar beams had thicker walls and thus
would show another failure mechanism. Furthermore, in our case the fault was obvious, and
the structural part would have been refused if intended for a WEC.
The two spar beams P09 and P10 did not show instability effects after 258 WISPERX cycles
for P09 and 140 WISPER plus 258 WISPERX cycles for P10. Additional 125 WISPERX
cycles under a 20 per cent increased load for both components showed no visible fatigue
damage. Additional 41 WISPERX cycles were carried out for P10 with the load increased
by 40 per cent before this component failed. While the maximum strain was now 0.96 per
cent in the upper spar cap, failure occurred in the lower part, which under WISPER loading
gets only about 60 per cent of the strain in the compression area. That was surprising
because prior to this event no indication of any fatigue damage was observed. Thus, it can
be concluded that in the lower spar cap, which was not monitored as well by eye and strain
gauges as was the upper one, a local bulge might have been produced during the final
loadings, causing the failure.
As spar beams P05, P09 and P10 were fatigue-loaded with the load sequences WIS-
PER/WISPERX, they can be compared with the results of the sequence-loaded specimens,
Type 2A. In Figure 20 it is shown impressively that - though the fatigue life of the spar beams
was stated to be satisfactory - the specimens have superiour values.
In general it can be concluded that all the failures which occurred during static or fatigue tests
in the spar beams were caused by instability problems and not by fatigue due to nominal
stresses. Thus, a wind turbine blade, which should be lightweight, i.e. calculated with high
design values (e.g. -0.6 per cent), should be stably designed so that no buckling, and in
consequence early fatigue failure, can occur.

Frequency Response and Damping


In addition to continuous observation and inspection by naked eye, the fatigue test procedure
for the spar beams was interrupted from time to time, and also after the occurrence of first
failures, to check stiffness or stiffness degradation, by means of static bending tests and by
testing the damping behaviour of the component, respectively [4]. The latter method partic­
ularly was extensively applied to all spar beams to get background information from the
laboratory which could indicate an impending failure due to fatigue. As mentioned above, the
simple static tests carried out at certain stages of lifetime did not show differences in global
stiffness. However, measuring the frequency response seemed to promise more success.
Thus, the measurement and evaluation procedure will be described on the example of spar
beam P09, which is representative for the fatigue and damping behaviour also of the other
blades. Our measurements were carried out in the laboratory under ideal conditions. Field
measurements will additionally depend on, e.g., temperature, humidity, coupling to other
WEC components, etc. Thus, for P09 some parameter studies were carried out, e.g. variation
of the temperature or the torque influence of the screws clamping the blade to the hub.
Frequency response was measured by means of forced vibrations produced by electrody-
namic excitation with constant energy input. The manner of exciting can be recognized in
Figure 14 and Figure 15.

- 193
In this case spar beam P09 was mounted into a self-built temperature chamber. Figure 21
shows the influence of the temperature frequency response, the lowest natural frequency
being about 18.2 Hz at +60°C and the highest about 18.7 Hz at -40°C.
The torque influence of the flange screws on frequency response can be seen in
Figure 22. Screw torques between 60 Nm and 5 Nm resulted in natural frequencies of about
18.4 to 18.1 Hz. Figure 23 shows curves for P09 before and after the first static test, with
7.5 kN transverse force causing a maximum strain of 0.64 per cent on the upper spar cap
of the test section. The tendency of decreasing natural frequency is continued after 158
WISPERX cycles. An increase of the maximum sequence load up to 9 kN and 125 further
WISPERX cycles show nearly no change. However, in subsequent loading the structure
buckled at 10 kN on the upper cap. After this failure a new frequency response curve was
measured. Again, a large decrease in resonance frequencies could be seen.

The difference in frequencies due to service life effects of less than 0.2 Hz, corresponding
to 1.09 per cent, is relatively small, measureable only with very good equipment.
The influence of temperature and also the torque of the bolts is higher and must be consid­
ered if an estimate of damage propagation is to be made on the basis of field measurements.
A similar tendency of decreasing natural frequencies over lifetime as for P09 was observed
for all spar beams. But this difference is perhaps not a sufficient indicator of immanent fatigue
failure.
Another means to find signs of fatigue damage, however, could be the logarithmic damping
decrement Λ. This was evaluated according to the half-bandwidth method [2]:
fn-f,
Λ = π-

Frequency response was measured generally with three different values of energy input. The
calculated damping curves can be shown in a diagram of amplitude measured at the blade
tip versus the logarithmic damping decrement Λ. Figure 24 shows damping versus amplitude
for P09 as a function of temperature and life history. Due to time limits on testing the tem­
perature effects, only one energy level was taken (Nell=3N). Damping increased with
decreasing temperature. By contrast, the damping of the different torques on the flange
screws did not change remarkably and is not plotted in the graph. In Table 6 and Table 7
the eigenfrequencies and logarithmic damping decrements as well as their changes in per­
centage, depending on the temperature and torque of the screws, are listed.

Concerning the influence of service life on damping behaviour, Λ decreases - in general


nonlinearly - with lower energy input, i.e. lower amplitudes, see Figure 24. However, damp­
ing before and after the first static test show quite different figures. And shortly before failure
in the static test, this curve is much lower than all the others. This is opposite to experience
with damping measurements using small specimens, where damping increases with an
increase of fatigue damage. Large structures are composed of many elements with proper­
ties which can be compared with those of the specimens. However, all these parts have
different damage rates locally. While there are only few elements with a high damage rate,
most are not as highly loaded, i.e. their damping properties may change only a very small
amount. They carry the loads if damage in the highly loaded elements has progressed
severely. Thus, the stiffness (including damping) properties of the spar will change, but not
as much as expected from the coupon specimens.

Frequency response measurements at the above mentioned three energy levels were carried
out on all fatigue-tested spar beams. The curves for the calculated damping as a function
of amplitude are plotted in Figure 25.

194
Also these results do not seem to suggest a reliable means to detect increasing fatigue
damage. However, the approach allows us to compare at least the damping at one ampli­
tude. Thus, for further comparison possibilities these plots were used to normalise damping
at an amplitude of 5 mm, see Figure 25. Table 8 on page 200 to Table 13 on page 201
show the natural frequencies and the normalised damping decrement as well as their change
over lifetime. Also in Table 6 and Table 7 Λ is normalised to an amplitude of 5 mm. This
was done by assuming that the shape of curves with different energy inputs would be similar
to that of the measured curves.
In Figure 28 to Figure 33 the terms Logarithmic Damping A and Eigenfrequency are plotted
over lifetime. For the frequency, the scale has been distorted in order to better visualise the
change. It is obvious that the damping decrement goes up and down in size, while the fre­
quency decreases with fatigue life.
However, in a plot with normal scaling the difference in the change of the resonance fre­
quency does not appear so clearly, see Figure 34 to Figure 39. These block diagrams were
designed to compare the change of the three stiffness parameters damping, eigenfrequency
and deflection over lifetime. While also in this presentation the changes in damping behaviour
are obvious, the eye cannot discover large effects in frequency and displacement.
In principle it can be found that damping increases sometimes very early, already after the
first static test. The increase is between a factor of 1.1 and 1.5 for the different components
after the first measurement. For some spars the logarithmic damping decrement shows a
large increase, up to a factor of 1.9 after the first visible failure, and on spar beam P03 even
2.4 after continuing fatigue loading with propagation of the failure in the spar cap growing to
the web, see Figure 34.
However, a decrease of damping could also be observed when the tests were continued after
the first measurement of the frequency response. In no case could a dramatic increase of the
logarithmic damping decrement be observed just before the occurrence of a visible failure,
which could have predicted this failure. Sometimes it even happened that the damping after
the first failure was lower than at any state before, e.g. P07, see Figure 35 or P09, see
Figure 39.
Though this feature of the damping behaviour of the spars may be confusing, it can be seen
that damping spars P10 and P09, which were characterised by quality status " 1 " (best), does
not yield a result exceeding 1.2 times its original status, see Figure 38 and Figure 39.
Sometimes they even undercut the factor d/d0 = 1.0. The change in damping is - also after
an increase of the nominal load - relatively low. By contrast, the other spar beams which are
rated quality status "3" or "4" show a much larger alteration of their damping, see Figure 34
to Figure 37. This indicates that the quality of a spar beam can probably be determined by
means of the damping method. The quality marks assigned to the different spars before the
tests by normal eye inspection (see Table 1) are confirmed quite well by the damping char­
acteristics.
Unfortunately for our case, this method of determining quality may be used only in the labo­
ratory and during fatigue load tests. However, for detecting an impending failure of the
structure it does not seem to be the right method.
It may be that the natural frequencies which only show decreases during the life histories of
all tested spar beams could indicate immanent failure. However, they then should be moni­
tored on-line.

3.2.5 Conclusions
Fatigue tests on the spar beams demonstrated that, rather than the observed nominal
stresses, local instabilities, e.g. bulges, seem to be responsible for premature failure. Nev-

195
ertheless, for nominal maximum strains of more than 0.6 per cent, 500 WISPER/WISPERX
cycles were achieved, which corresponds to a lifetime of a hypothetical wind turbine of more
than 80 years. This holds promise that higher design limits for GRP rotor blades with diam­
eters larger than, e.g., 25 m could be allowed, resulting in more economical lightweight
blades.
The damping behaviour evaluated from frequency measurements of rotor blades was
investigated extensively. It could be demonstrated that this method does not seem to be a
reliable means for detecting the state of damage progression. Visual inspection will surely
remain the best and irreplaceable method for the near future.

3.2.6 Bibliography
[1] L. Lampe Konstruktion, Bauaufsicht und Test einer Mehrfach-Einspannvorrichtung für schlanke,
biegebelastete Bauteilkomponenten DFVLR Interner Bericht IB-88-435/19 (1988)
[2] H. Georgi Damping Effects in Aerospace Structures AGARD Conference Proceedings No.277,
1979
[3] C. Schloz Experimentelle Untersuchung und Berechnung eines Windkraft-Rotorblattholms aus
Glasfaserverstärktem Kunststoff (GFK) unter Biegebelastung DLR IB 425-89/08 (1989)
[4] T. Kalkuhl Dämpfungscharakteristiken von GFK-Rotorblättern unter verschiedenen Randbedin-
gungen Studienarbeit, Universität Stuttgart 1990

196
3.2.7 Tables

Spar Code Manuf.- Deflection, mm I Max. Strain, % Test Cross


No. Quality * for Design Load (7.5 kN) Section
1 POL 3 55.8 -0.16 unsupp.,thick
2 P01 2 70.4 -0.49 unsupported
3 P02 2 67.0 -0.47 foam core
4 P03 3 66.0 -0.49 foam core
5 P07 3 65.2 -0.75 foam core
6 P06 4 67.2 -0.64 foam core
7 P05 3 67.7 -0.63 foam core
8 P10 1 65.8 -0.64 foam core
9 P09 1 66.3 -0.67 foam core

* Manufacturing Quality: (1) very good (2) good (3) bad (4) very bad

Table 1. Survey of the design quality of spar beams.

Spar Code Type of Failure/ Bending Reference Reference


No. Loading Applied Moment Strain Deflection
Load [ kN ] [ kNm] [%] [ mm ]
1 POL statically 33.3 83.25 -0.69 245.0
2 P01 statically 6.892 17.23 -0.45 64.7
3 P02 statically 13.512 33.78 -1.49 120.6
4 P03 R=-1 5 12.50 -0.38 66.0
5 P07 R=-1 5 12.50 -0.50 65.2
6 P06 WISPER 7.5 18.75 -0.64 67.2
7 P05 WISPER/WISPERX 7.5 18.75 -0.63 67.7
8 P10 WISPER/WISPERX 7.5 18.75 -0.64 65.8
20%-incr. load 9.0 22.50 -0.76 79.0
40%-incr. load 10.5 26.25 -0.89 92.1
9 P09 WISPERX 7.5 18.75 -0.67 66.3
20%-incr. load 9.0 22.50 -0.80 79.5
40%-incr. load 10.0 25.00 -0.89 88.54

Table 2. Survey of strains and deflections of spar beams related to the failure / applied loads.

197
Style Producer Weight Weave Thickness
g/m2 Pattern mm
92146 Interglas 425 UD (10/1) Plain 0.47
92125 Interglas 280 2/2 Twill 0.40
UD 940 Brochier 920 UD(ca.19.1/1) 0.83
(with M9-Prepreg) (Ciba) (1483) Plain
1023 Brochier 390 2/2 Twill 0.45
(with M10-Prepreg) (Ciba) (650)

Table 3. Glass reinforcement data.

Spar Code First Failure after Type of Loading Referenc Manner of


e
No Load Cycles / Strain Failure
Passes Through [%]
Load Sequence
1 POL — statically -0.69 Compression
2 P01 — statically -0.47 Buckling
3 P02 — statically -1.49 Compression
4 P03 3.44E5 LC R=-1 -0.38 Buckling Fatigue
5 P07 5.13E5 LC R=-1 -0.51 Buckling Fatigue
6 P06 17 WISPER -0.64 Manuf. Fault
7 P05 140/258 WISPER / -0.63 Buckling Fatigue
WISPERX
8 P10 120/258 WISPER / -0.64 Buckling Fatigue
WISPERX
+125 WISPERX -0.77 in Lower Cap
+41 WISPERX -0.90 after 2x incr. Load
9 P09 258 WISPERX -0.67 Buckle in Upper
+125 WISPERX -0.80 Cap at Static
+41 WISPERX -0.94 Test after 2x
increased Load

Table 4. Survey of loading and failure modes of spar beams.

198 -
Resin Mean Fibre Mean Thick­
Type Lay-Up System Content E-modulus ness
vol% GPa mm
1A 5x92146/0° L20/SL 36.9 29.6 (R=-1) 2.0 - 2.4
31.9 (R=.1)
2A (1x92125/±45°; L20/SL 38.6 24.6 (R=-1) 2.0 - 2.4
2x92146/0°) sym 22.7 (R=10)
2C (1x1023/+45°; M9/M10 53.0 31.7 (R=-1) 2.6 - 3.0
1.5xUD940/0°)sym (Prepreg) 34.0 (R=10)

Table 5. Lay-Up, fibre content, modulus and thickness of specimens.

Temperature f, Change Λ Change


°C Hz % %
60 18.19 -1.30 0,0274 -4.53
40 18.33 -0.54 0.0278 -3.14
RT 18.43 0 0.0287 0
20 18.43 0 0.0307 +6.97
0 18.56 +0.71 0.0340 +18.47
-20 18.68 +1.36 0.0415 +44.60

Table 6. P09, influence of temperature on eigenfrequency and damping.

Torque f, Change Λ Change


Nm Hz %
80.NL 18.38 0 0.0255 0
80 18.38 0 0.0246 -3.53
70 18.38 0 0.0260 +1.96
60 18.38 0 0.0235 -7.84
50 18.36 -0.11 0.0238 -3.92
40 18.34 -0.22 0.0256 +5.49
30 18.29 -0.49 0.0258 +5.49
10 18.20 -0.99 0.0248 -2.35
5 18.08 -1.63 0.0253 -7.06

Table 7. P09, influence of torque of the flange screws on eigenfrequency and damping.

199
Load f, Change Λ Change Deflection Change
Cycles Hz % % mm %
OLC 18.19 0 0.0168 0 66.54 0
1E3LC 18.23 +0.22 0.0250 +48.81 66.12 -0.63
1E4 LC 18.22 +0.16 0.0225 +33.93 - -
1E5 LC 18.10 -0.49 0.0250 +48.81 67.17 +0.95
3.4E5 LC 17.77 -2.31 0.0325 +93.45 66.35 -0.28
5.13E5 LC 17.42 -4.23 0.0400 +138.0 68.88 +3.52

Table 8. P03, stiffness parameters depending on lifetime.

Load f, Change Λ Change Deflection Change


Cycles Hz % % mm %
0 LC 18.23 0 0.0175 0 65.73 0
2LC 18.20 -0.16 0.0200 +14.29 - -
1E3LC 18.25 +0.11 0.0296 +69.14 66.12 +0.60
1E4 LC 18.14 -0.49 0.0206 +17.71 - -
1E5LC 18.05 -0.99 0.0243 +38.86 66.27 +0.83
5.13E5 LC 17.90 -1.81 . 0.0256 +46.29 65.26 -0.71

Table 9. P07, stiffness parameters depending on lifetime.

Load fr Change Λ Change Deflection Change


Cycles Hz % % mm %
OLC 18.15 0 0.0188 0 67.73 0
6.71 E3 LC 18.15 0 0.0192 +2.13 66.27 -2.15
after 150 days 18.14 -0.06 0.0235 +25.0 - -
14 WISPER 18.06 -0.50 0.0338 +79.79 67.64 -0.12

Table 10. P06, stiffness parameters depending on lifetime.

- 200
Load f, Change Λ Change Deflection Change
Cycles Hz % % mm %
OLC 17.92 0 0.0185 0 68.29 0
6.71 E3 LC 17.73 -1.06 0.0225 +21.62 68.31 +0.03
after 150 days 17.95 +0.17 0.0215 +16.22 66.12 -3.18
34 WISPER 17.70 -1.23 0.0285 +54.05 65.95 -3.43
140 WISPER 17.65 -1.51 0.0195 +5.41 65.76 -3.70
240 WISPERX 17.63 -1.62 0.0244 +27.75 - -
258 WISPERX 17.54 -2.12 0.0269 +40.84 - -

Table 11. P05, stiffness parameters depending on lifetime.

Load fr Change Λ Change Deflection Change


Cycles Hz % % mm %
OLC 18.22 0 0.0240 0 66.35 0
after 120 days 18.00 -1.21 0.0247 +2.92 - -
14 WISPER 17.93 -1.59 0.0233 -2.92 66.02 -0.50
120 WISPER 17.81 -2.25 0.0213 -11.2 66.12 -0.35
240 WISPERX 17.81 -2.25 0.0249 +3.75 65.72 -0.95
258 WISPERX 17.82 -2.20 0.0250 +4.17 65.62 -1.10
+125 WISPERX 17.78 -2.41 0.0229 -4.58 66.61 +0.39
aft. stat. test 17.77 -2.47 0.0275 +14.58 - -

Table 12. P10, stiffness parameters depending on lifetime.

Load f, Change Λ Change Deflection Change


Cycles Hz % % mm %
OLC 18.37 0 0.0240 0 66.81 0
1 LC 18.35 -0.11 0.0283 +17.92 - -
240 WISPERX 18.29 -0.44 0.0257 +7.08 66.51 -0.45
258 WISPERX 18.30 -0.38 0.0250 +4.17 66.66 -0.22
+125 WISPERX 18.27 -0.54 0.0227 -5.42 68.01 +1.80
aft. buckl. 18.22 -0.82 0.0267 +11.25 - -

Table 13. P09, stiffness parameters depending on lifetime.

- 201
3.2.8 Figures

tapered area

thin measurement
cross section

Figure 1. D-formed spar, derived from AEROMAN 12.5/40 rotor blade.

/ / / \ \

ssSvi
_-
\ ^

rotor blade hub

Figure 2. Principle of blade attachment.

202 -
co
c
■ι
Φ
ω Spar Shape

C
Thin­Walled A rea
φ
-<
O
­*. Test Cross Section
C/l
T3
0)
­ι Calculated Cross Sections
IO
φ
o 400
3
Φ

ο_
200 ■
C
Q.
5'
O ca
ω Φ 500 1000 1500 2000 2500 3000
Ui

O
UI
Coordinates
UI
ω
φ
o
õ'
3
\/ / 7 / / / / / / / 7 ZZZ]

1. Lamination on a 2. Preforming 3. Curing in a Female


over a core Mould with Internal
plane board
Pressure

Figure 4. Blade laminate moulding procedure.

Figure 5. Triple testing rig for static and dynamic investigations.

- 204
Figure 6. Test bed for fatigue and frequency response tests.

Figure 7. Torsional loading of spar beam P01.

- 205
100­1 Deflection
90 = | Strain in Spar Cap

80
o
o

I
70 Η
II æa
χ
c
'cö 60

II
ι-
ω
E
50
il
ε
c
o
'­t—·
o
40­

30­ II li
M—

Q
ω 20
ί II
10
II
Ρ 01 Ρ 02 Ρ 03 Ρ 05 Ρ 06 Ρ 07 Ρ 09 Ρ 10
Figure 8. Deflection and maximum strain of spar beams at design load.

997,
Test C ross Section
1022,5
1047,5
1072,5

¿ι Y­ Coordinates

Plane of Maximal Longitudinal Strain

Figure 9. Location of strain gauges in the test cross section.

206 -
0,4

Δ 3.2 kN
V 4.0 kN
0,2­ Test Section + 5.1 kN
# X 5.1 kN
Lower Cap
I o Upper Cap

­0,2

0,4 ­ï 1 Γ

800 900 1000


Locations of Strain Gauges

Figure 10. Longitudinal strain distribution in test cross section (P01).

Web Upper Cap Lower Cap Web

C
o
fi
o
</>
b

0 100 200 300 400 500 600


Locations of Measured and Calculated Shear Distortion

—e— Calculated Shear Distortion


[ j Measurement
X Linearized Measurement

Figure 11. Shear distortions at test cross section, comparison of calculation and test at a
load of 5.1 kN.

207
0.6
RI 4 R6 R7 R8 RIO R11 R12 /* \ R14
0,4 ' 4.11 kN ' 'χ
ι 5.04 kN ι _^-Τ y
1 I t i

c 0.2 ι— ε.46 ™>^^^-~*y^


'5
♦^ _ 1 ι I Ι s ^ ^ ι ι
<n 0 1 1 ^j^^f ι ι
o

°­0,2 ι ^X.

--) , , ,-U-, , 1 , ¿—, ,— , — , — [ — , , ^-^—,—^-^—r 1 ■ ν-


­0,4 ­200 ­100 0 100 200
Evolved Shape

­200 ­100 0 100 200


Evolved Shape

0,6

£ °.4

Web
­0,4

­200 ­100 0 100 200

Evolved Shape

Figure 12. Circumferentially measured strains at the position of maximum longitudinal strain.

­ 208 ­
Web Upper Cap Lower Cap Web

c
1

0 100 200 300 400 500 600


Locations of Measured and Calculated Strains

—β— Calculated Strain

D Measurement
X Linearized Measurement

Figure 13. Longitudinal strains at test cross section, comparison between calculation and
test at a load of 5.1 kN.
Figure 14. Test bed for frequency response investigations, temperature chamber.

Figure 15. Spar beam in temperature chamber.

210 -
Figure 16. P06, delamination after static loading.

Figure 17. P06, delamination propagation after 14 WISPER sequences.

"'* PfT
■ ,""J' <■ ■'.*»

i * stó»I , .
Í

&£■<?

Figure 18. P06, fatigue failure due to delamination propagation.

­ 211
­π 3.0
co'
Φ
—L
CO
C)
33 =TO
2.5
C)
"s O Gl.­Ep. Specimens of Types 1A , 2A, 2C
¡a
«o 2.0
"D n
o '
CI
ωo
Τ3-Ό

as
O

ωa
3 _ •I 1­5
o. ST (ID
■Ses­ (7)
O Φ
3 α
O O o
Φ »
3 s?
0) 1.0 ­ R = ­ 1 ­θ
s. o
S3
? 0 0 C C DODOC ID
S°°
V Spar Beam P03 O OC D O
si Δ: /
c 2.
S Φ
Δ Spar Beam P07 o c»
2 3
S" O Specimens Type 2A with Buckling and Temperature Increase
öS. .0 I I I I I L_
Φ -,
ο. Φ 1E0 1E1 1E2 1E3 1E4 1E5 1E6 1E7 1E8
«SL
Load Cycles η
. 5sr

co' 2.5
ro
o
"0 3 Ό
moo O Compression ) Specimens
53
S
3
Ό
2.0 X Tension ) (Type 2A)
< (Λ ω
m"S
Xgo
C

Î '
'ce OX> o
1 5
0 o>
01 φ


è · )<X ο <κ o
s¡ c
_L
o a:
φ
ω
X o
ω o
|.S X­
^ 1.0
o
-s-
°;sr
£0
3 3 + Spar Beam P05
r'
<° o
«c
.5 D Spar Beam P09
il O Spar Beam P10
S »
Si? (1 χ 92125/±45° / 2 χ 92146/0°)sym / L20/SL (Type 2A)
°i
S.S
φ « 1E0 3 1E1 3 1E2 3 1E3 3 1E4
ië Passes through Load Sequence WISPER/WISPERX
7.5
7.0 Spar Beam Ρ - 09 Temperature Effects
6.5
6.0
5.5
5.0
4.5
4.0
3.5
3.0
2.5
2.0
1.5
1.0
X 40-C, A = .0261 Q 60'C, A = .0258
0.5
Δ -20"C, A = .0362 V O'C, A = .0305 + 20*C, A = .0283
0.0
18.0 18.2 18.4 18.6 18.8 19.0
Frequency , Hz

Figure 2 1 . P09, frequency response at different temperatures.

7.5

Spar Beam Ρ - 09 Influence of Torque of Hub Screws

2.5
2.0
1.5
1.0
X 40 Nm, A = .0256 D 50 Nm, A = .0238 O 60 Nm, A = .0231
0.5
Δ 5 Nm, A = .0253 7 10 Nm, A = .0248 + 30 Nm, A = .0258
0.0
17.9 18.0 18.1 18.2 18.3 18.4 18.5 18.6
Frequency, Hz

Figure 22. P09, frequency response at different torques of flange screws.

214
7.5
7.0 SPAR BEA M P­09 Influence of Service Life
6.5 on the Frequency Response
6.0
5.5
5.0
4.5
4.0
3.5
3.0
2.5
2.0
1.5
1.0
X + 1 2 5 WISPX, A = .0236 D Buckl Fail, A = .0284
0.5
Δ Unloaded, A = .0254 V Stat. Test, A = .0282 + 258 WISPX, A = .0264
0.0
18.0 18.1 18.2 18.3 18.4 18.5 18.6
FREQUENCY, Hz

Figure 23. P09, frequency response depending on fatigue history.

.040 I I
* RT, Unloaded
o A fter first Stat Test
V ­20°C
.035 o 940 WISPERX

+ 258 WISPERX
< Δ +125 WISPERX
φ
ε .030
v0<'C /? Load 20% Incr.

20°C V c X +125 WISPERX


o RT S/ /
/ζ/ >
ω
Q
4 0 o ' c ^ ­~~«í
60°C ■
7 \
in Static Test
(Load 33% incr.)
.025 '"*/
CD s
O

*
4
M
í-^
Sj£s ν Variation of
Temperature

.020

3 4 5 6 7 8 9
Amplitude, mm
Figure 24. P09, damping versus amplitude, influence of life history and temperature.

215
.040
* P­()9

O P­(53

.035 Δ P­()5
Amplitude or
+ P­()6
c o Ρ­ IO
ω .030
E α P­()7
Φ /C, *
1—
ϋ
d)
Q .025 S ¿> s /
eb ^>-
■ * %

A λ /
o

£ψ
/
m^
.020
>
S
*Z ^
.015
3 4 5 6 7 8 9
Amplitude, mm
Figure 25. Damping versus amplitude, original state of the spar beams

.04 19
&­­­­­ s­
­Q Β
.03 E] 18 I
r
φ
E
^ A ­ ­ ­ Λ if
c
a φ
u .02
ω 17 ir
Q Δ Log. Decrement ω
k_
CD M—
O
I .01 D Eigenfrequency ­ C
Φ
16 .S?
POS
LU
.00 ! I I I I I I 15
-20 -10 10 20 30 40 50 60
Temperature , ° C

Figure 26. P09, influence of temperature on damping and eigenfrequency.

216 ­
.04 19

3
­­­B ­B
­­El·—
­­□­­ — B Ν
t . 0 3_
­ B " 18 Χ
c >;
φ o
E ¿3 ¿3" —­Δ­ c
2 φ
3
Ö .02 17 α­
Φ φ
Q Δ Log. Decrement
L_
Η­
C

? 01 ■

Q Eigenfrequency ­ 16
Φ
CD
LU


.00 I I I I I I l l l l l I ι 15
0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80
Torque of Flange Bolts , Nm

Figure 27. P09, Influence of torque of the flange screws on damping and eigenfrequency.

.04­ A r 19
/
< 3 B ­ —­Β π Δ Ν
.03^ 18 Ι
c
φ
E .­­­­Δ­­­ — * · — ­ * b c
ω φ
o .02­ .—.­­""" ­ 17 g-
Φ ¿i-"'" ω
k.
Q Δ Log. Decrement Μ—
CD
C
.01 ­ D Eigenfrequency φ
3 16
üf
P03
.00^ Ι ' Ι ' Ι ' 15
1Ε0 1Ε1 1Ε2 1Ε3 1Ε4 1Ε5 1Ε6 1Ε7
Load Cycles

Figure 28. Ρ03, influence of fatigue life on damping and eigenfrequency.

217
.04 19

[3e-- --Β- - — Β — -
Ζc .03- _-Δ — Β-- - Β 18 I
φ if
E
Φ ,.-Δ'_.—·Δ c
φ
k_
o .02- ^r' 'Δ- 17 o-
φ
Q 2
Δ Log. Decrement
c
ν>-

α> D Eigenfrequency
φ
q .01 16 .^
LU

.00 15
1Ε0 1E1 1E2 1E3 1E4 1E5 1E6 1E7
Load Cycles

Figure 29. P07, influence of fatigue life on damping and eigenfrequency.

.04 19

Ν
Ζ .03 B -Β 18 I
c ><
φ
E o
2
o .02 Λ — " 1
' σ
Φ Φ
ι-
Q Δ Log. Decrement Μ—
C
D) φ
G Eigenfrequency
9 .01 Η 16 .°>
UJ

.00 15
1E0 3 1E1
Passes Through Load Sequence WISPER

Figure 30. P06, influence of fatigue life on damping and eigenfrequency.

218
.04
18
Ν
Ζc ­03­ I
φ
E
17 f
2 \ ^ Φ
Ö .02 "Α""
ZJ
σ­
Φ
Q
ο
Δ Log. Decrement
eb 16 I
9 .01 0 Eigenfrequency Φ
çp
Lu
PM
15
.00 I ' ι Ι , Ι | , |

1Ε0 3 1Ε1 3 1Ε2 3

Passes through Load Sequence WISPER/WISPERX

Figure 31. P05, influence of fatigue life on damping and eigenfrequency.


.04 19

B
t .03 —& 18 I
c -Β-- —Β
Φ
E ^ .­Λ&­ if
Λ -Δ- Δ c
φ
Ö .02
Δ" Φ

φ 17 g-
Ω
eb
Δ Log. Decrement 2
*♦­
S .01 D Eigenfrequency c
φ
16 .2>
αι

.00 15
1E0 3 1E1 3 1E2 3
Passes through Load Sequence WISPER/WISPERX
Figure 32. P10, influence of fatigue life on damping and eigenfrequency.

.04 19
[j El—Β Ν
I ; .03 18 *
c A
φ
-&.. vf
c
ε ^Δ φ
2 17 ã-
o .02 Φ
φ Δ Log. Decrement i-
Q "c
eb D Eigenfrequency
φ
16
9 .01 Η iE

.00 15
1E0 3 1E1 3 1E2 3
Passes through Load Sequence WISPER/WISPERX
Figure 33. P09, influence of fatigue life on damping and eigenfrequency.

219
2.5
ra F^3
log. Decrement Λ.

Eigenfrequency f
2.0
Deflection d
TD
"O
<^
1.5

Lo­
MS

2aL\ ^
0 LC 1E3 LC 1E5 LC 3.44E5 LC 5.13E5 LC
1 * failure in propagation of crack
lower spar cap up to sheaiweb

Figure 34. Damping and stiffness parameters of P03 versus lifetime.

2.5
P07
log. Decrement Λ.

LYX Eigenfrequency f
2.0
Deflection d
■a
­a
1.5

. 1.0
^ ^

.5­

■2L·
Iv
0 LC 1E3 LC 1E4 LC 1E5 LC 5.1E5 LC
failure in upper
spar cap

Figure 35. Damping and stiffness parameters of P07 versus lifetime.

­ 220 ­
2.5
P06
3 log. Decrement Λ.

Eigenfrequency f
2.0
Deflection d
TD
TD
1.5
I
1.0-

I I
r^

¡
^

OLC
■κ
6.7E3LC
!

after 150 days


¡ 14 WISPER
delamination in fold of
laminate, failure at 17 WISPER

Figure 36. Damping and stiffness parameters of P06 versus lifetime.

2.5 Ί
Ρ05
^ log.
log. Decrement Λ.

Eigenfrequency f
2.0
Deflection d
TD
TD
1.5-

1.0

$3 ζπ
<
.5

.0
OLC 6.71E3LC
L Κ
after 150days 34WSPER 140WSPER
2L·
240WISPX
¿L·
258W1SPX

1* failure
in spar cap

Figure 37. Damping and stiffness parameters of P05 versus lifetime.

221 -
2.5 P10
log. Decrement Λ.

W I Eigenfrequency f
2.0­
Deflection d
TD
TD
1.5

LO r^im

KV
•V
mm Ν
^

.5­ :
\
Ν
tij i\ ^
OIC àrIMdajs 14VHSPER 1 2 0 W S ? 240WISPERX 8 8 W S P 0 B +125WBPERX+125WÎOTRX

(stattestat (load 20% after stat test


167 WISPERX increased) with, load 40%
increased

Figure 38. Damping and stiffness parameters of P10 versus lifetime.

2.5
P09
log. Decrement Λ.

Eigenfrequency f
2.0
Deflection d
TD
TD
1.5

. 1.0
^

2± -£ΓΛΚΊ K^frANJ VA,


OLC 1 LC 240 WISPERX 258 WSPERX +125 WISPERX +125 WISPERX

(load 20% after buckling


167 WISPXERX increased) failure in
static test
load 33% incr.
Figure 39. Damping and stiffness parameters of P09 versus lifetime.

- 222 -
4. Statistical Evaluation and Comparison of GFRP Fatigue Data

(Ch.W. Kensche, W. Sys)

4.1 General

Most of the GFRP fatigue data presented in Section 2 are not yet ¡n a form suitable for
selecting the best material for a specific design. Either they show only single­point data sets,
or different methods of statistical evaluation were applied, i.e. the curves cannot be com­
pared. For certification purposes, however, the application of statistically approved fatigue
data is necessary. (Certification rules for the design of rotor blades require characteristic
fatigue strength data for 95% survivability with 95% lower confidence limit, see, e.g., [1]).

The method used should also be able to handle data sets with only a few fatigue data points.
Statistical evaluation according to an approach described by Sendeckyj [2] was found to fit
the necessary requirements. It can handle runouts and tab failures through progressive
censoring, and it is also well defined when small fatigue data samples are available. The data
fitting procedure is based on the method described below.

4.2 Data Fitting

The relationship for fatigue lifetimes may be expressed by a two­parameter ε­Ν curve which
is a particular form of the so­called wearout model by Halpln at al [3]. One parameter S
determines at high cycles the slope of the ε­Ν curve in a log­log plot. It is equivalent to the
reciprocal value of k in [1 ] in the expression C2b = (Ν)1Λ<. The other parameter, C, allows at low
cycles for flattening or steepening the curve in a log­log plot. This model neglects the pos­
sibility that for stresses below a certain level no failure will occur. Hence, in this study one
cannot expect to find a so­called fatigue limit.

One of the characteristic features of fatigue is the large scatter of the test results. For this
reason a probabilistic approach is needed. Therefore, a consistent set of data must be pro­
duced. This can be realized as follows. Each test result in the log­log ε­Ν plane is trans­
formed into an equivalent static strength by using the S and C parameters of the ε­Ν curve.
Hence, a complete set of fatigue data yields a new set of equivalent static strength data. It
is assumed that this new data set is two­parameter Weibull distributed.

The data fitting approach proceeds along different steps. With a good guess of the S and C
parameters, the fatigue data is transformed into equivalent static strength data. By using the
maximum­likelihood method a two­parameter Weibull distribution is fitted to the new data set.
This procedure is repeated in an iterative process, until the largest value for the shape
parameter α is obtained.

The ε­Ν curve takes the following form:

(­ΙηΡ(Λ/))­5­
('N-A).QS
where

223
εΒ maximum applied strain
β scale parameter of the Weibull distribution
α shape parameter of the Weibull distribution
S and C parameters defining the ε-Ν curve
A =-(1-C)/C
Ν number of cycles
P(N) probability of survival

This is the mean fatigue curve for 95% survivability. A fatigue curve considering 95% confi­
dence can be calculated by assuming a two-parameter Weibull distribution. The lower con­
fidence limit is described in [4] as follows:

(-InWr o
((N-A).QS
where

LL(P(N)) percentage point for survivability (here P=95%) with a confidence limit (γ=95%),
see [4], excerpt tabulated in Appendix A in Table 4 on page 249
η number of tests

The percentage point for survivability UY(P(N)) in [4] is tabulated to a minimum number of
10 tests, whereas in this study some fatigue data sets show only nine or even eight test
results. However, due to the high cycle tests, these curves are valuable and should not be
ignored in the comparison. In this case, the percentage points for the survivability UT were
estimated by extrapolation.
35 data sets were evaluated according to this data fitting procedure. 31 of them are described
in this book, 4 are results from an FFA investigation [5].
Table 1 on page 231 gives a survey of the origin of the data, a description of the material
lay-up, the loading ratio, and the parameters of the Weibull distribution α and β as well as
of the wearout model C and S and k=1/S. For the statistical evaluation, some data were
examined for stress instead for strain. For comparison purposes stress was converted into
strain via the Ε-Modulus derived from reference Section 2 in this book.
The resulting fatigue curves including the data points used for calculating the curves are
plotted in Figures 1 through 6. The fatigue lines are represented by the (mean) curves of 50%
and the curves of 95% probability of survival with the 95% lower confidence limit (in the fol­
lowing called 95/95 confidence limit). Each figure represents the fatigue lines of the reference
section tabulated in Table 1.
When failure points already existed at a certain strain level, runouts at the same level with
higher cycle numbers but without a residual strength value were also handled like failures.
For runouts at low stress levels in data sets with very few results (e.g. Figures 1c, 1d, 3d)
no censoring was applied to keep the scatter (and thus, the α parameter) in a normal stage.
Progressive censoring was used In handling tab failures, see, e.g., in Figure 5 the diagram
for type 2A laminate tested at R = - 1 .
For samples with no data at low cycles, the C parameter was set to 1 (constant slope in
^ ε - ^ Λ / space). Extrapolation is not without its problems and therefore the fatigue lines are

- 224 -
plotted only one order of magnitude beyond the measured data. With this conservative
method, however, it Is possible to predict a lower strain limit at failure up to 109 load cycles
for those data sets with failures or runouts at 108 cycles, see e.g. Figures 5b-d.
The ε-Ν curve at low cycles is strongly influenced by the static strength data. The loading rate
of the static strength test should be comparable to the cyclic loading rate. The information
about loading rate encountered in the static test was rather scarce. Nevertheless, the static
data was used in this analysis. Consequently, the survival lines at low stresses will be con­
servative for cases with lower loading rate.
The fatigue curves of the low-high and the high-low tests reported in Sections 2.1.1 and 2.1.2
are not presented since the basis for a comparison of the results is not yet clear. However,
for studying the load sequence effects, the WISPER/WISPERX standard test results were
statistically evaluated, see Figures 1, 2 and 5.
In the case of the WISPER data in Figure 2, a large scatter of data points is obvious. Nev­
ertheless, the ε-Ν curve is plotted to compare it with the other load sequence results. In this
case no censoring was performed since no relation was found between the scatter and fail­
ures occurring due to gripping or instability problems.

4.3 Comparison of the GFRP Fatigue Curves

4.3.1 General
Different materials and load combinations of the fatigue curves for the GFRP material were
selected to compare
• mainly material properties dominated by 0° orientated load bearing glass fibres
• properties of glass-polyester (GI-UP) with those of glass-epoxy (Gl-Ep)
• cross-plied material with angles varying between ±10° and ±60° to the loading direction
by considering
• the R-ratios,
• different temperatures,
• different states of conditioning.
By presenting the results in the form of 95/95 confidence limit lines, the designer of a rotor
blade has an objective means to make a material choice, since the data are presented in a
form following the certification requirements.
Table 2 on page 232 shows the available data sets in the form of a matrix. A survey of the
realized combinations of the fatigue curves is given in Table 3 on page 232 The combina­
tions of the selected curves are plotted in Figures 7 through 25 on pages 239 to 248.
In the following some comments will be given on fatigue curves plotted in the appropriate
combinations.

4.3.2 0° Orientation, R = 0.1 (Figure 7H-8)

For GI-UP the purely 0° orientated material shows over a wide range (up to 106 cycles) lower
properties compared to 0° GFRP with randomised or +45° cross-plied layers (Figure 7 on
page 239).
Pure UD material (Gl-Ep) with plain fabric and about 90% of the fibres orientated in the
loading direction shows a nearly identical but a little lower curve than the randomised 0°

225 -
laminate of GI­UP. The 0° roving material (GI­UP) is superior to the UD material (Gl­Ep) in
the high cycle range (Figure 8 on page 239).
However, both plots show an equalizing tendency for all laminates in the high cycle range
at a strain of about 0.5%.

4.3.3 0° Orientation, R = ­1 (Figure 9­11)

The ε­Ν curves for GI­UP of 07±45° (manufactured by Risø, tested by NLR) and 07random
orientation are nearly identical (Figure 9 on page 240).
In Figure 10 on page 240, the fatigue curves of three different Gl­Ep and two GI­UP lay­ups
are plotted. For Gl­Ep, the pure UD material (type 1A ) behaves nearly identically to the pre­
preg system with ±457UD lay­up of type 2C, whereas the ±457UD material of type 2A is
superior to both.
In the lower cycle range, the curve of the dry GI­UP material with pure UD lay­up is higher;
that of the wet is lower than the curves of the 1A and the 2C material with Gl­Ep. However,
at the high cycle range, a clear tendency to converge can be seen for all curves.
In Figure 11 on page 241, the GI­UP with 07+45° lay­up of Figure 9 is compared with the
±457UD Gl­Ep material shown in Figure 10. Both Gl­Ep types show better results than the
GI­UP material. But at the high cycle range, a converging trend is obvious, too.
The ε­Ν curves at R = ­1 with 0° dominated fibre orientation end at a strain level of about
±0.25% in the high cycle range.

4.3.4 0° Orientation, R = 10 (Figure 12^13)


Three curves are available, 2 Gl­Ep curves with ±45° lay­up (type 2A and 2C), see
Figure 12 on page 241, and one GI­UP curve with the 07±45° lay­up manufactured by Risø
and tested at NLR.
The comparison of Gl­Ep shows that the 2C line is significantly lower than the 2A line, pre­
sumably due to a manufacturing fault described in Section 2.2.1.
Comparison of the Gl­Ep (2A ) and the GI­UP curve shows better characteristics for the GI­UP
in the low cycle, but lower ones in the high cycle range.
However, for the ε­Ν curves at R = 10 the strain in the high cycle area is always higher than
for the R = ­1 curves. At least for the Gl­Ep material it is over 0.5%.

4.3.5 0° Orientation, different R­Ratios (Figure 14H­17)


In Figure 14 on page 242 GI­UP of 07randomlsed lay­up is compared with Gl­Ep of UD
lay­up (type 1 A) at R = 0.1 and R = ­ 1 . The curvature of the fatigue lines at the single R­ratios
is very similar. But there are differences in the strain level. In the high cycle range, the ten­
sion­tension loaded specimens show a strain level nearly twice as high as the tension­com­
pression loaded ones.
Figure 15 on page 243 shows the fatigue lines at R = ­1 and R = 10 for Gl­Ep material with
±457UD lay­up in order to compare cold curing resin systems (type 2A) with prepreg systems
(2C). In the low cycle as well as in the high cycle range, the 2A material is superior. For R
= ­1 the difference is not very significant. However, at R = 10 there is a steep decrease of
the curve in the low cycle area, expressed also in the C parameter being about 15, see
Table 1 on page 231 and Figure 5 on page 237. In general, for composite materials, the C
parameter is not higher than 1. As already explained in Section 2.2.1, the reason seems to

­ 226
be in the inappropriate manufacturing technology. In the high cycle range, the curves for both
materials meet again at the same level which is about 0.5 % for the R = 10, and 0.25 % for
the R = -1 stress ratio.
In Figure 16 on page 243, it is also shown that the tension-compression loaded material
compared with the tension-tension and compression-compression loaded material of similar
lay-up has only about half the strain level in the high-cycle range. In this case, the same
material (manufactured by Risø) was tested at R = 0.1 by Risø and at R = -1 and 10 by NLR.
Although the specimens were tested in different institutes, which may cause differences
especially in the low cycle range due to different shapes of specimens, grippings, etc., those
influences diminish at the high cycle range, I.e. low strains.
Similar results can be seen in Figure 17 on page 244. In this plot, for R = 0.1 a pure UD
material (type 1A) was taken, whereas for R = -1 and 10 specimens of type 2A (±457UD)
were used. In this case, the R=10 curve is also superior to the R = 0.1 curve.

4.3.6 0° Orientation, Load Sequence Effects (Figure 18)

Figure 18 on page 244 shows fatigue curves of load sequence-tested materials for WISPER,
WISPERX and reverse WISPERX. The tests were performed at ECN, NLR and DLR. The
evaluated fatigue curves represent the 95/95 confidence limit for the maximum applied strain
versus cycles (passes) of the sequences. Some remarks will be made about the differences
of the curves.
Comparing the materials tested with the WISPER sequence (07random by ECN and 0745°
by NLR, both with polyester), a drastic decrease of the NLR material must be noted. This
appears very strange since the constant amplitude lines of the same material tested at the
same institutes are absolutely identical, see Figure 9 on page 240. An explanation may be
the large scatter in the WISPER results of NLR, see also Figure 2f on page 234.
The WISPERX line of NLR is only little higher than the plotted WISPER line. Due to the lower
scatter of the WISPERX curve, the mean curves are nearly identical; only the slope differs
by a small amount. From this point of view it might be concluded that the omission of 16%
applied for WISPERX would not influence the results obtained with WISPER. Since however
the WISPER curve of ECN is significantly higher than the described NLR curves, this con­
clusion must be confirmed by further tests.
Also, the ECN WISPER and the DLR WISPERX curves have similar slopes and curvatures.
That might be expected by assuming that the omission of WISPERX would not have signif­
icant effects, since the curves of the constant amplitude tests at R = 0.1 and -1 with the same
or similar lay-up are comparable, too, see Figures 14 and 10. However, also this assumption
must be comfirmed by further tests.
The similarity of the DLR WISPERX and Reverse WISPERX curves confirms the statement
in Section 2.2.1 that the data of Reverse WISPERX are influenced by instability effects of the
test specimens. Analysing the results of the compression-compression tests (Figure 15), the
Reverse WISPERX curve should be higher.

4.3.7 ±10° Orientation, different R-Ratios and Temperatures (Figure 19+21)

In these diagrams, test results from three different institutes are obtained for the same
material with ±10° fibre orientation and polyester. It was manufactured by Risø in a filament
winding process, see Section 2.1.3, and sent to NLR and RUG in the form of plates. The
specimens were prepared for the institutes own requirements. Thus, the width of the speci­
mens tested at Risø was more than 25mm, while it was 12mm for the NLR and RUG speci­
mens.

227
In Figure 19 on page 245, the fatigue data for two different R­values are plotted. For com­
parison purposes, the R = 0.1 data were tested at Risø and at RUG. Surprisingly the RUG
curve is ­ with similar slope ­ significantly lower than the Risø curve. This might be due to the
small width of the specimens tested at RUG, which causes the higher risk of earlier damage
initiated at the free edges. The R = 10 curve tested at NLR lies between the others. The slope
is relatively flat. Since the NLR specimens were also relatively slender it appears logical that
the curve would be higher with a larger width.

In Figure 20 on page 245 fatigue results at ­20°C are plotted for stress ratios 0.1 and ­ 1 . The
tests were conducted by RUG. As expected, the fatigue line for R = ­1 is lower than for R =
0.1.
Figure 21 on page 246 shows all results for R = 0.1 at ­20°, RT and +50°. There is a flat­
tening tendency for the slope of the curves from the higher to the lower temperatures. How­
ever, the curves are very close together except for Risø, which is superior as explained
above, see also Figure 19.

4.3.8 ±30° Orientation, different R­Ratios and dry and wet Conditions (Figure 22)

These data were made available for comparison purposes within this section of the book by
courtesy of FFA [5]. An epoxy resin system was used for the matrix. A short description of
the material is given in Appendix Β of this section.
The specimens were tested at R = 0 and R = ­1 in dry and wet condition.
A significant influence of humidity must be noted. The fatigue properties of the dry material
are for both R­values superior to those of wet material. The R = ­1 curves are each below
the R = 0.1 curves for both conditions.

4.3.9 ±45° Orientation, different R­Ratios (Figure 23+24)

In Figure 23 and Figure 24 on page 247 fatigue curves are presented for glass­polyester
with ±45° fibre orientation tested at Risø and at ECN. The specimens by Risø are manufac­
tured in a filament winding process, those of ECN in a more industrial manner with random
mats between the 45° layers.

Figure 23 shows results from ECN and Risø at R = 0.1 and R = ­ 1 . A t R = 0.1, the material
with random ¡nterlayers is superior to the filament wound one. The reason might be due to
the random mats, which smooth the ¡nterlaminar shear stresses. The R=­1 curve of ECN is
only a little below the R = 0.1 curve of Risø.
In Figure 24, the Risø results are plotted for R = 0.1 and R = 10. The compression­com­
pression line is significantly higher than the tension­tension line especially in the lower cycle
range. This tendency of better results at R = 10 compared to R = 0.1 is also obvious in other
tests, see, e.g., Figure 15 on page 243 for the 2A material of DLR. It is however in contrast
to corresponding results obtained at different testing sites. Slightly different test methods,
specimen geometries, etc. may cause the different behaviour. A n explanation for the better
R = 10 values is that under repeated tensional loading cracks propagate faster than in
compression loading.

4.3.10 Comparison of Laminates with different Ply Angles at R = 0.1 (Figure 25)

In Figure 25 on page 248 tension­tension results obtained with glass­epoxy and glass­
polyester are compared for ply­angles of ±10° (GI­UP), ±30° (Gl­Ep) and ±45° (GI­UP).

228
Surprisingly, the differences between the ±10° and ±30° material is relatively small over the
whole cycle range. This might be due to the better fatigue properties of the epoxy resin used
in the ±30° material. Compared to that, the fatigue curve for the ±45° material is significantly
lower. However, in this case the more matrix-controlled ±45° lay-up seems to be influenced
by creep effects due to the load control principle of the hydraulic pulsator. Also, the E-moduli
and thus, the static strength values for the differently orientated laminates differ by an order
of magnitude.

4.4 Summary

In this section a comparison Is made of fatigue results for glass fibre reinforced plastics which
were manufactured and tested at different institutes under various conditions, see the rele­
vant sections. The data obtained were statistically evaluated by means of the Sendeckyj
method [2|. The results are presented in the form of mean and 95/95 confidence limit curves.
Various kinds of materials and test specimens were used which differed in lay-up, matrix
system, specimen geometry, tabs, etc.. There is no doubt that significant deviations in the
testing methods also influenced the results, e.g. by the use of different clampings on
hydraulic cylinders, antibuckling guides, loading rates, etc..
Nevertheless, a comparison of fatigue curves allows some statements about the material
properties themselves:
At tension-tension (R = 0.1), materials with four different lay-ups were tested, three of them
with polyester and one with epoxy resin (Figure 7, Figure 8 on page 239). While in the low
cycle range the purely 0° material shows lower values than 0° glass mixed with random mats
and UD glass combined with ±45° fabric, all curves converge in the high cycle range to a
strain level of about 0.5%. No significant difference can be detected between polyester and
epoxy.
At tension-compression (R = -1), materials with six different lay-ups were tested, three of
them with polyester and three with epoxy resin (Figure 9 on page 240 through Figure 11
on page 241). One of the polyester lay-ups was tested in both dry and in wet condition. There
seems to be a tendency for superior behaviour of the epoxy laminates in both the low and
high cycle ranges. The curves end at a strain level of about 0.25% at 109 load cycles. While
the dry polyester laminate lies relatively close to the best curve, the line for the wet condi­
tioned shows the poorest properties.
At compression-compression (R = 10), only three different materials were tested, one with
polyester, two with epoxy laminates (Figure 12 on page 241 and Figure 13 on page 242).
Although the curve with the epoxy prepreg system (type 2C) is not representative due to bad
manufacturing quality as explained in Section 2.2.1, it shows a strain level at the high cycle
range similar to the other epoxy curve (type 2A), namely more than 0.5% at 109 load cycles.
Also at this stress ratio, the epoxy laminate shows slightly better fatigue properties than the
polyester, especially at high cycles.
A comparison of the fatigue curves at the different R-ratios 0.1, -1 and 10 is made in
Figure 14 on page 242 through Figure 17 on page 244. For both resin types, it is obvious
that the R = -1 curves show the most conservative properties while the R = 10 are superior
to the R = 0.1 curves, see Figure 16 and 17. Figure 14 shows the interesting fact that in
tension-tension, a polyester laminate is slightly superior to an epoxy one while the same
materials behaves vice versa in tension-compression.
The influence of the WISPER/WISPERX standard on fatigue properties (Figure 18 on page
244) has not yet been satisfactorily explained. However, except for the two curves being
based on results with a high scatter (see Figure 2f and 2g), it can be stated that between
103 and 104 passes through the sequence may be obtained at strain levels of about 0.8%.

- 229
The fatigue curves for laminates with ply-angles of ±10°, ±30° and ±45° are plotted in
Figure 19 on page 245 through Figure 25 on page 248. In general, the slope of these lines
is more flat than that of the 0°-dominated material, and the curves are situated at a lower
level.
The ±10° material is a polyester laminate. The curves plotted in Figures 19 through 21 show,
with the exception of the influence of the specimen geometry, that
• the R = -1 curve has the lowest values,
• the R = 10 curve is superior to the R = 0.1 curve,
• the influence of temperature at R = 0.1 is relatively low,
• the fatigue properties at low temparatures are better than at high ones.
The curves for the ±30° material in dry and wet conditions of an epoxy-laminate are plotted
in Figure 22. The influence of humidity is more significant than that of the applied R-values
of 0 and - 1 .
The curves for ±45° polyester material plotted in Figures 23 and 24 show very flat and low
curvature, except for the stress ratio of 10. The low strains obtained can be referred to the
load control mechanism of the testing machines and the strong influence of the relatively
weak matrix.
A comparison for the three different angle-ply laminates is possible for the tensile loading
plotted in Figure 25. As can be expected, fatigue properties drop as ply angles increase.
The negative influence of humidity on the properties of the reported materials is relatively
high. It must be doubted whether exposure of the specimens to pure water or 90% relative
humidity is comparable to reality. The conditions seem to be too severe. There are research
programs under way with the aim of studying the behaviour under more realistic conditions.

4.5 Bibliography

[1] Richtlinie für die Zertifizierung von Winkraftanlagen Germanischer Lloyd, Ausgabe 1993
[2] Sendeckyj, G.P. Fitting Models to Composite Materials Fatigue Data, in: Test Methods and Design
Allowables for Fibrous Composites ASTM STP 734, C.C.Chamis, Ed. 1981, pp.245-260.
[3] Halpin, J.C., Jerina, K.L., Johnson, T.A. Analysis of Test Methods for High Modulus Fibers and
Composites ASTM STP 521, American Society for Testing and Materials, 1973, pp. 5-64
[4] Bain, Lee J. Statistical Analysis of Reliability and Life-Testing Models, Theory and Methods Marcel
Dekker Inc., New York and Basel
[5] Blom, Anders F. Fatigue Evaluation of WTS-3 Glassfibre Blade Material FFA, TN 1982-26, The
Aeronautical Research Institute of Sweden, 1982

- 230
Section, Institute Lay­l'p Resin Thickness Stress ratio Cond. Static Fatigue Weibull Parameters Model P a r a m e t e r s Figure
Reference Type mm R Spec. Tests Tests α Ρ C s k=l/S
2.1.1. ECN 0° Random Polyester 5 0.1 dn 1 26 15.186 2.548 0.003 0.095 10.526 la
2.1.1. ECN 0° Random Polyester 5 ­1 dry 1 26 9.542 2.525 0.625 0.1 10.000 lb
2.1.1. ECN ±45° Random Polyester 7 01
0.1 dry 0 8 26.41 0.703 1 0,033 30.303 1c
C/5 2.1.1. ECN r45" Random Polyester 7 ­1 dry 0 9 16.194 0.902 1 0.0854 11,710 Id
σ
Ό Φ
2.1.1. ECN 0­ Ran dorn Polyester 5 WISPER dry 0 12 22.095 2.450 1 0.107 9.346 le en
(D
η
-5»
Õ' 2.1.2. NLR IT) Polyester 4.8 ­1 dry 0 14 13.716 4.283 1 0.128 7.813 2a
Ol 2.1.2. NLR I'D Polyester 4.8 ­1 wet 0 15 12.460 3.180 1 0.115 8.696 2b
c*
2.1.2. NLR 0°/±45° (from Riso) Polyester 1.8­2.2 10 diy 1 11 21.501 2.539 0.046 0.09 11.111 2c
δ' 2.1.2. NLR 0°'±45° (from Riso) Polyester 1.8­2.2 ­1
3 dry­ 1 12 15.230 2.498 1.61 0.1 10.000 2d
2.1.2. NLR ±10° (from Riso) Polyester 2 10 dry 0 8 36.738 1.560 1 0.072 13.889 2c
2.1.2. NLR 0°,±45° (from Riso) Polyester 1.8­2.2 WISPER drv 2 20 10.049 1.569 0.9424 0.0695 14.388 2f
2.1.2. NLR 0° ±45° (from Riso) Polyester 1.8­2.2 WISPERX dry 1 10 16.611 1.490 1 0.058 17.241 2g
α
DJ
2.1.3. Riso 0= Polyester 1.9 0.1 dry­ 2 13 14.758 2.462 0.132 0.0684 14.620 3a
UI
2.1.3. Riso 0= ±45° (2 χ 0°/±45° O°)sym. Polyester 2.3 0.1 dry 2 11 30.183 2.657 0.00968 0.106 9.434 3b
ra 2.1.3. Riso ±10= Polyester 2 0.1 dry 3 15 20.666 1.512 0.00608 0.0844 11.848 3c
c 2.1.3. Riso ±45° Polyester 2 0.1 dry­ 0 13 14.481 1,092 1 0.08428 11.865 3d
UI
ra 2.1.3. Riso ±60° Polyester 2 0.1 dry 3 11 23.667 0.518 20 0.0718 13.928 3e
α 2.1.3. Riso ­45° Polyester 2 10 0 9
dry 22.719 3.354 1 0.1452 6.887 3f

n- 2.1.4. Rl'G ±10° (from Riso) Polyester 2 0.1 dry. ­20° 0 16 27.980 1.510 1 0.0796 12.563 4a
ra 2.1.4. Rl'G ±10° (from Riso) Polyester 2 0.1 dry. RT 0 16 24.963 2.110 1 0.111 9.009 4b
2.1.4. RUG ±10° (from Riso) Polyester 2 0.1 dry. +50° 0 16 14.706 2.688 1 0.123 8.130 4c
5Γ 2.1.4. Rl'G ±10° (from Riso) Polyester 2 ­1 dry. ­20° 0 16 24.594 1.365 1 0.099 10.101 4d
2.1.4. Rl'G ±10° (from Riso) Polyester 2 ­1 dry. +50° 0 16 9.728 1.022 1 0.102 9.804 4e
o
ω 2.2.1. DLR UD
ra Epoxy 2.0 ­ 2.4 0.1 dry. 1A 5 32 16.482 2.250 0.00146 0,1105 9.050 5a
< 2.2.1. DLR I'D Epoxy 2,0 ­ 2.4 ­1 dry. 1A 12 22 13.405 2.447 0.196 0.1 10.000 Sb
2L 2.2.1. DLR ±45°/UD (1 χ i45°/2 χ 0°)sym. Epoxy 2.0 ­ 2.4
c ­1 dry. 2A 11 15 20.109 2.236 0.01744 0.11756 8.506 Sc
m 2.2.1. DLR ±45°/ΙΤ3 (Ι χ ±45°/2 χ 0°)sym. Epoxy 2.0 ­ 2.4 10 dry. 2A 7 7 17.429 2.190 0.00018 0.092 10.870 Sd
o' 2.2.1. DLR ±45°'UD(1 x ± 4 5 ° , 1 . 5 x l ' D ) s y m . Ep­Prepreg 2,6­3.0 ­1 dry. 2C 2 17 13.702 2.110 0.042 0.11 9.091 Se
3 2.2.1. DLR ±45 C .UD (1 χ ±45°/1.5 χ UD)sym. Ep­Prepreg 2,6­3.0 10 dry.2C 2 17 20.223 2.140 15.32 0.0539 18.553 5f
2.2.1. DLR ±45°/l'D (1 χ ±45°/2 χ 0°)sym. Epoxy 2.0 ­ 2.4 WISPERX dry. 2A 4 7 25.653 2.264 0.387 0.105 9.524 5s
2.2.1. DLR ±45°/UD (1 χ ±45°/2 χ 0°)sym. Epoxy 2.0 ­ 2,4 Rev. W'.X dry. 2A 7 7 21.637 2,208 0.49 0.0846 11.820 Sh

|5] FFA ±30° Epoxy­ 3.8 0 dry­ 0 15 19,275 1.804 1.845 0.076 13.158 6a
|5| FFA ±30° Epoxy 3.8 ­1 dry 0 13 14.100 1.307 1 0.108 9.259 6b
|5| FFA ±30° Epoxy 3.8 0 wet 0 15 12.350 1,520 1 0.078 12.821 6c
[5] FFA ±30° Epoxy 3,8 ­1 wet 2 16 8.599 1.085 1 0.103 9.709 6d
Lay-Up, 1 3 5 7 8 9
Conditions R=0.1 R=-l R=10 WISPER WISPERX Rev. W .X

il 0° Risø
b ±10° Risø/RUC, NLR
C 07±45° Risø NLR NLR NLR NLR
d 0°/Random ECN ECN ECN
e UD(1A) DLR DLR
I ±45°/UD (2A) DLR DLR DLR DLR
S ±45°/UD (2C) DLR DLR
h + 10° (-20°C) RUG RUG
■ + 10° ( + 50°C) RUG RUG
.j ±45 "/Random ECN ECN
k ±45° Risø Risø
1 ±60° Risø
m UD (dry) NLR
n UD (wet) NLR
0 ±30° (dry) FFA(R=0) FFA
P + 30° (wet) FFA(R = 0) FFA

Table 2. Matrix of available data sets for comparison.

Comb. Nr. Institute Lay-Up Matrix Stress ratio Cond. Figure


of Table 2 R
al/cl/dl Risø, ECN 0°, 0 ° / ± 4 5 ° , 0°/Rndm UP 0.1 7
al/dl/el Risø, ECN, DLR 0°, 0°/Rndm, UD UP, Ep 0.1 8
c3/d3 NLR. ECN 0°/Rndm, 0°/ + 45° UP -1 9
f3/g3/e3/m3/n3 DLR, NLR UD. ±45°/UD Ep, UP -1 dry, wet 10
c3/f3/g3 DLR, NLR ±45°/UD, 0 ° / ± 4 5 o Ep, UP -1 11
t'5/g5 DLR + 45°/UD Ep 10 12
c5/f5/g5 DLR, NLR ±45°/UD, 0 ° / ± 4 5 ° Ep, UP 10 13
dl/d3/el/e3 ECN, DLR 0°/Rndm, UD UP, Ep 0.1,-1 14
f3/f5/g3/g5 DLR ±45°/UD Ep -1, 10 15
ci/c3/c5 Risø, NLR 0°/ + 45° UP 0.1, - 1 , 10 16
el/O/fS DLR UD. ±45°/UD Ep 0 . 1 , - 1 . 10 17
c7/cS7d7/f8/f9 DLR, ECN, NLR ±45°/UD, 0°/Rndm Ep, UP WISPER. W.X, 18
0°/±45° reverse W.X
bl/b5 Risø, RUG, NLR + 10° UP 0.1, 10 19
hl/h3 RUG ±10° UP 0.1, -1 -20° C 20
bl/hl/il Risø, RUG ±10° UP 0.1 -20/20/50°C 21
ol/pI/o3/p3 FFA ±30° Ep 0, -I dry, wet 22
,jl/kl/j3 ECN, Risø ±45°/Rndm, ±45° UP 0.1,-1 23
kl/k5 Risø ±45° UP 0.1, 10 24
bI(Risø)/kl/(>] Risø, FFA ±10°, ± 3 0 ° , ±45° UP, Ep 0.1, 0 25

Table 3. Survey of the data set combinations.

232
4.7 Figures

I I t
TEST RESULTS
I 1 I 1 1 1
GI­UP, 0°/Random ■ TEST RESULTS
mean curve
\ R—1, ECN
■ 95/95 confidence limil

...^ (i = 15.1B6, ß=2.548S.


C=0.003. S = 0.095 α=9.542.β=2.525%

>­ tsr I 1 _
GI­UP, 0°/Random
­—
i ' ... ­ *­i^
R=0.1,ECN

Figure 1a Figure 1b

TEST RESULTS
1 ! 1 I I
­ GI­UP. ±45°/Random " GI­UP, ±45°/Random ­ TEST RESULTS

R = 0 . 1 . ECN ­ moan curvo


R=­1,ECN maan curvo

00 * 95/95 confidence limit ­


a =26.413. cc=16.194,0=0.90 2%
R = 0.701% so
C ­ 1 . S = 0.0327

00

so

Figure 1c Figure 1d

I I
GI­UP, 0°/Random TEST RESULTS

WISPER, ECN

­
^"""­^i a=22.095. 0=2.45%
C s l , 5=0.107
­­. ° **"^­*
I, ■­­7?~~~~ ­ ­ . L
ΤΓ
1
F ¡CI I I f f * 1 P LOC Passes through Sequence

Figure 1. ε­Ν curves for glass­polyester at different R­ratios and WISPER (Ref. ECN, Section
2.1.1).

233
I ! I
■ TEST RESULTS
GI-UP, UD, Batch4 (dry)
R=-1,NLR
95/95 confidence limit
- a = 13.716,B=4.283%
C=1, S=0.128

I i
- ---τ.
ir*" ;
SU
j'7^
Figure 2a Figure 2b

! ! i ' '
^ G I - U P , 0°/±45°, R=10, N L R | I ■ TEST RESULTS 2.5 |GI­UP, O°/±45°, R = ­ I , NLR| ­I ■ TEST RESULTS

~~\J j I
95/95 confidence limit
* 2 ° ­X^­­—­j­­­ \ —-.—\- 95f95 confidence limit

i 0*321.501, 6=2.539%
C=0.046, S=O.09 ­ 1.5
0=15.230,6=2.498%
C=1.61. S=0.1

... Ί ..
"F^Ss<I
-L. _
3
ijo ­ ­^­Λ­^^^—
I

■ ! i
J ­ 0.5

"i τ ^ !
Figure 2c Figure 2d

I 1
TEST RESULTS
|GI­UP, ±io°, R=IO, NLR| • TEST RESULTS GI­UP, 0°/±45°
WISPER, NLR
I 95/95 confidence limit — 1 95/95 confidence limit

i α =36,738, 6=1.56% ^^ ! a = 10.049, 6=1.569%

'■-- i C=1, S=0.072


I ~T~­^__
l ;

I i
„ i ­
Í
LOG Passes through Sequi
Figure 2e Figure 2f

■ TEST RESULTS
GI­UP, 07+45°
WISPERX, NLR
dcnce limit

a=16.611. 1=1.49%

—­ ­^
:­­­ ~-.r\

FiCIUTG 2d LOG Passes through Sequence

Figure 2. ε­Ν curves for glass­polyester of different lay­up and conditions of humidity at dif­
ferent R­ratios, WISPER and WISPERX (Ref. NLR, Section 2.1.2).

234 ­
I I
TEST RESULTS

¡GI­UP, 0°, R=0.1, Rlse| 2.5
mean cuive

— 95795 confidence limil


r"

. T~\. α = 14.75β, (1=2.462%


C=0.132. S=0.0684
i O-30.182, 6=2.657%

i IS
3 - >J ; —
—­ — ­ — = 10 -
—ΓΓΤ77—
■ lei-U?, 0°/±45°, R=0.1,

Figure 3a Figure 3b

t 1
I I ¡ ι ι
|GI­UP, ±10°, R=0.1,Rlse| ■ TEST RESULTS ϊ.5 . |GI-UP ±45°, R=0.1, Rise| ­­Ι­ ­ TEST RESULTS

ι
_
--L
mean curve
I I I i
7
95/95 confidence (imit 95795 confidence limit
I j I i
ι ¡ ! «=20.666. 6=1.512% i α =15.387. [1=1.11%
0 0 . 0 0 6 0 6 , 5=0.0844 m C=1, S=0.0866
~*~~-. i,
. L ·: I
* 10
i I I
ι ¡ """·■■;­.

I I
Γ ¡ ....... 4_ ­■■■·*

Figure 3c Figure 3d

! I '
■ TEST RESULTS
| G I ­ U P , ±60°, R=0.1, Rise|
I ■

■ |—r~h·­T~ ­ 95/95 confidence limit

0=23.667.11=0.518%
i i !
"1" :
C=20, S=0.07t8

— i i— 1I —_ i—. } +. . . _—
__
d_±d
Figure 3e Figure 3f

Figure 3. ε­Ν curves for glass­polyester of different lay­up at different R­ratios (Ref. Risø,
Section 2.1.3).

235 ­
I I 1 I I I Ι ι I
GI­UP, ±10°, R=0.1
T=­20°C, RUG
" TEST RESULTS
2.5 GI­UP, ±10°, R=0.1
T=RT, RUG
­ ■ TEST RESULTS

mean curve

95/95 confidence limit


,. I
­ ­ 9SV9S confidence limit

OS27.980. 6=1.51% a=24.963. 6=2.11%


C=1, S=0.0796

—­
­ "Τ" >,n

*—­—­L__

Figure 4a Figure 4b

GI­UP, ±10°, R=­1 " TEST RESULTS

T=­20°C, RUG
95/95 confidence limit

a=24.594. 6=1.365%
C=1.S=0.099

_L . J. ί
­ f""j~ "" |
ί ^^r^*T~^
Figure 4c Figure 4d

Figure 4e

Figure 4. ε-Ν curves for glass-polyester of ±10° orientation at different R-ratios and temper­
atures (Ref. RUG, Section 2.1.4).

- 236
Ι ι 1

Gl­Ep, UD, Typel A TEST RESULTS Gl­Ep, UD, Type 1A


R=0.1, DLR mean curve R=­1,DLR Ι­ TEST RESULTS

.. ­ ~~j —­^ 95/95 confidence limit . ι 95(95 confidence limit

. \ u=16.482.ft=2.250%
C ¡0.00146, S^O 110S
"=13.405. ft=2.447%
C=0.196, S=0.1

v ; ^*V^^
r
"r^>­^ "
C^~~­Î­
— ­ ~ i

Figure 5a Figure 5b

i i I
Gl­Ep, ±45°/UD, Type 2A • TEST RESULTS
R=­1,DLR

——. ! L 95795 confi d on c e limit

ü =20.109, 6=2.236%
C=0.01744. S=0.11756

i .'"­!.\^
^^"­­^
' 'i ! - 4­ ^ΓΓ­>»»_____^

1 ι<
Figure 5c Figure 5d

I Ι
Gl­Ep, ±45°/UD, Type 2C TEST RESULTS : Gl-Ep, ±45°/UD, Type 2C __.[. • TEST RESULTS
2.5
R=­1, DLR R=10, DLR

' I I

I,, ^ \ 0 = 13,702.6=2.11%
C=0.O42. S=0.11
.:~_
—I " Γ
95/95 confidenco limit

α=20.223. 6=2.14%
C=15.32,S=0.0539
"'■i. ^*^\.
5,
1 ....i ¡ "'"•ί-
."'" —­:­­"Γ

i
..";*

:
^—~­
^ d z±-
Ί
! : " ■"[" ; !
i
]
1 ' ί· ¡
Figure 5e Figure 5f

Gl­Ep, ±45°/UD, Type 2A • TEST RESULTS

WISPERX, DLR mean curve


­
­ ^ ι 95/95 confidence limit

n=25.653. 6=2.264%
C=0387. S=0.105
^ ^ ¿ ^
3, ..._.. .._ ^T^­— ..

F j CI U T O 5C1 LOG Passes through Sequence F Í O U TG 5 h LOG Passes through Sequence

Figure 5. ε­Ν curves for glass­epoxy of different lay­up at different R­ratios, WISPERX and
Reverse WISPERX (Ref. DLR, Section 2.2.1).

237
i ! I I i ! !
¡Gl­Ep, ±30°, R=0, FFA¡ • TEST RESULTS ¡Gl­Ep, ±30°, R=­1,FFA ¡ ■ TEST RESULTS

mean curve
I
1 ■ 95/95 confidence limit ­ .. ­.ι­ 95*95 confidence limit

I a=19.275. 6=1,304%
C=1.845, S=0.076
ι 0 = 14.100. 6=1.307%
C=1,S=0.108
­\^
^~Λ^ ^ I ! :. __ i ¡

L I ! ­­
­~t_
!
...

Figure 6a Figure 6b

| I : I Ι ι 1 I i i
Gl­Ep, ±30°, R=0 ■ TEST RESULTS Gl­Ep, ±30°, R=­1 ■
• TEST RESULTS

RH 90%, FFA RH 90%, FFA mean curve

­ 95/95 confidence Hmit 95/95 confidence limit

a=12.350. 6=1.52% a =8.599. 6=1.085%


C=1, S=0.103
..

~ — Ί ι î­i­»­
­■"
^ T 7

Figure 6c Figure 6d

Figure 6. ε­Ν curves for glass­epoxy of ±30° orientation and different conditions of humidity
at different R­ratios (Ref. FFA [5]).

238 ­
!
2,5 ­ ­
tal­UH, 0°/±45°, Risø

.* 2 GI­UP, 0°/Random. ECN


"~^>.
­... Nv. Gl­Up, 0 , KlSØ
S
S 1·5
><
'"­­..
­­.._
ro

""P
S 4 ­"­­— ­^;. · · ­

0,5 ­ ­
95/95 conf dence limit
­ ■

Log N

Figure 7. Comparison of ε­Ν curves (95/95 confidence limit) for mainly 0° oriented glass­
polyester at R = 0.1.

2,5

c
ε
S 1.5
><
ro

0,5

0 1 2 3 4 5 6 7 8 9
Log Ν

Figure 8. Comparison of ε­Ν curves (95/95 confidence limit) for 0° oriented glass­polyester
and glass­epoxy at R = 0.1.

239
I
2,5

GI­UP 0°/±45°, NLR


2 —
GI­UP 0°/Random, ECN
c
's
55 1.5
χ:
ro
­ ^
*^^·­ *"*""——w.

0,5
95/95 conf dence limit
i —

0 1 2 3 4 5 6 7 8 S
Log Ν

Figure 9. Comparison of ε­Ν curves (95/95 confidence limit) for mainly 0° orientated glass­
polyester at R = ­ 1 .

Gl­fcp ±45 /UU \¿n¡ υι_ι\


2,5
Gl­Ep, UD(1A), DLR

Gl­Ep ±45°/UD (2C) DLR ­


c GI­UP UD (dry), NLR
'5
~ 1 5
χ
GI­UP UD (wet), NLR
ro

^^>~—
0,5
— ­. _^r*^"" ! >?ί7?ττ~
95/95 conf dence limit

0 1 2 3 4 5 6 7 8 Í
Log Ν

Figure 10. Comparison of ε­Ν curves (95/95 confidence limit) for mainly 0° orientated glass­
polyester (partly in wet condition) and glass­epoxy at R = ­ 1 .

­ 240
2,5 - — —-
Gl-Ep, ±45°/UD (2A), DLR

Gl-Ep, ±45°/UD (2C), DLR


c GI-UP, 0°/±45°, NLR

S5 LS
χ
ro

" : : "" '* ^^ t

0,5
^ΤΤΤΤΓ^^τ.
95/95 con dence limit

0 1 2 3 4 5 6 7 8 5
Log Ν

Figure 11. Comparison of ε-Ν curves (95/95 confidence limit) for mainly 0° orientated glass-
polyester and glass-epoxy at R = - 1 .

2,5

ui-rzp, ±45°/UD (2A), DLR


2
* Gl-Ep, ±45°/UD (2C), DLR
c
'ε \ -
fi l i
χ
ro
■ - - .

0,5 ..·._.
95/95 conf dence limit

0 1 2 3 4 5 6 7 8 S
Log Ν

Figure 12. Comparison of ε-Ν curves (95/95 confidence limit) for mainly 0° orientated glass-
epoxy at R = 10.

241
2,5
( i . - h p , ±**v / u u \¿MJ, U'l_r\

GI­UP, 0°/±45°, NLR


c
'ίδ
fi 1 ζ
w '·° ^ ~
χ
.„,
ro —
— — ^

0,5 -
95/95 conf dence limit

ι |
0 1 2 3 4 5 6 7 8 5
Log Ν

Figure 13. Comparison of ε­Ν curves (95/95 confidence limit) for mainly 0° orientated glass­
polyester and glass­epoxy at R = 10.

GI­UP , 0°/Rndm, R=0.1,


2,5 ECN

Gl­Ep, 1A , R=0.1,DLR

, Gl­Ep 1A, R=­1,DLR


­^ΓΤ\.
GI­UP , 0°/Rndm, R=­1,
* · 1 *i
CO ' · 0 ECN
* "ν.
χ
ra
■ — _
^ Γ ^

—·.._ *' '


0,5
­ —­.•^­­....
■ ' ­ · ' ­ ■
95/95 confidence limit

0 1 2 3 4 5 6 7 8 9
Log Ν

Figure 14. Comparison of ε­Ν curves (95/95 confidence limit) for 0° orientated glass­polyestér
and glass­epoxy at different R­ratios.

­ 242
3
.

2,5 _ ZA , Κ—TU, üi.r\

2/ . , R = ­ 1 , DLR
_
c ZQ: , R = ­ 1 , DLR
'ra or. ι in ί ii F
5 l i

CO '·° ' ._;­·­ ... ^^­^^^
— ­*­iv
­ ­­ — ■"­« ; ~V I
""" ""**·**.—„
- - — -_. """"
0,5 _ ■ ­ . . : :

95/95 con idence limit I

0 1 2 3 4 5 6 7 8 5
Log Ν

Figure 15. Comparison of ε­Ν curves (95/95 confidence limit) for mainly 0° orientated glass­
epoxy at different R­ratios.

2,5 o
R ­ 0 .1,0 /±45°, Risø

2 R=10. 0°/+45°. NLR


*
c
MLK
ε R—' 1, 0 /±45 ,
cõ 1.5
'·— P\^
χ

ra
':
­­­­­·­...
2
1 ~~~rr^~­t^­.
­ " ­ JÌ.^
­·­., ­­­
0,5 95/95 con idence limit
η

0 1 2 3 4 5 6 7 8 9
Log N

Figure 16. Comparison of ε­Ν curves (95/95 confidence limit) for mainly 0° orientated glass­
polyester at different R­ratios.

243
2,5 Gl­Ep UD (1A), R=0.1,
DLR

Gl­Ep ±45°/UD (2A ),


R=10, DLR

Gl­Ep ±457UD(2A),R=­1,
DLR

Figure 17. Comparison of ε­Ν curves (95/95 confidence limit) for 0° and ±4570° orientated
glass­epoxy at different R­ratios.

2,5

c

fi 1 5
CO I . =

0,5

0 1 2 3 4 ί
LOG Passes through Sequence

Figure 18. Comparison of ε-Ν curves (95/95 confidence limit) for mainly 0° orientated glass-
polyester and glass-epoxy under WISPER and its derivations.

244
2,5
GI-UP, ±10°, R=0.1, Risø

GI-UP, ±10°, R=10, NLR


c GI-UP,+10°, R=0.1, RUG

z LS
χ
ro

0,5

Log Ν

Figure 19. Comparison of ε-Ν curves (95/95 confidence limit) for ±10° orientated glass-
polyester at different R-ratios.

2,5

GI-UP, ±10°, R=0.1, RUG

GI-UP, ±10°, R=-1, RUG


c

CO 1' .5°
fi
χ
ra

0,5

95/95 confidence limit

0 1 2 3 4 5 6 7 8 9
Log Ν
Figure 20. Comparison of ε-Ν curves (95/95 confidence limit) for ±10° orientated glass-
polyester at -20 C and different R-ratios.

245
3
_. I .
2,5 — . GI-UP, ±10°, RT, Risø

GI-UP, ±10°,-20°C, RUG


2
SS
- GI-UP, ±10°, RT, RUG
c
ε
fi 1'·°
ς
• - GI-UP, ±10°, +50°C, RUG -
co
χ
ro

S
1 "~~~~———^^_^
i
95/95 conf dence limit °'- '-"-*-::··
; ;
0,5
1 I

4 5
Log Ν

Figure 21. Comparison of ε-Ν curves (95/95 confidence limit) for ±10° orientated glass-
polyester at different temperatures and R = 0.1.

95/95 confidence limit I


dry, R=0, FFA
2,5
dry, R=-1,FFA

wet, R=0, FFA


c
'ε wet, R=-1,FFA
co '>°
χ
ro

0,5

0 1 2 3 4 5 6 7 8 9
Log Ν

Figure 22. Comparison of ε-Ν curves (95/95 confidence limit) for ±30° orientated glass-epoxy
at different R-ratios, dry/wet condition and 35°C.

- 246
3
95/95 confidence limit GI-UP, ±45°/Random
2,5 R = 0 . 1 , ECN

GI-UP, ± 4 5 °
R = 0 . 1 , Risø
ç
'ro GI-UP, ±45°/Random
ö
CO 1' '5° R = - 1 , ECN
χ
<0
Έ 1

0,5

0
3 4 5
Log Ν

Figure 23. Comparison of ε-Ν curves (95/95 confidence limit) for ±45° orientated glass-
polyester at different R-ratios.

95/95 conf dence limit


2,5 - — -— - - -- - - . rJ L 1
GI-UP, ±45°, R=10, Risø

SS 2 GI-UP, ±45°, R=0.1, Risø


Strai

-_
αϊ

χ
S ro
1 - _ -

0,5 - - ' - - ■ - - . —-__^___


" · - - ■ - .

-··- ^ ^ ^ .
0
0 1 2 3 4 5 6 7 8 !
Log Ν

Figure 24. Comparison of ε-Ν curves (95/95 confidence limit) for randomised and pure ±45°
orientated glass-polyester at different R-ratios.

247
95/95 con idence limit
2,5
_. , ,_ ±10°, Risø, R=0.1
GI­UP
, Gl­Ep, ±30°, FFA, R=0
c
'ε GI­UP ±45°, Risø, R=0.1
co '·°
χ
ro
τ ­ ·­ ­
0,5 ­­ »­ ­­· ~^­^~ΓΤ:"T:~^~r­^­r

0 1 2 3 4 5 6 7 8 5
Log Ν

Figure 25. Comparison of ε­Ν curves (95/95 confidence limit) for differently orientated glass­
polyester at R = 0.1 and +30° orientated glass­epoxy at R = 0.

248 ­
4.8 Appendix A: Table for Calculating Confidence Limits

No of Tests n U7(P(N))

10 8.468
11 8.191
12 7.949
13 7.744
14 7.565
15 7.404
16 7.267
18 7.038
20 6.846
22 6.688
24 6.554
26 6.444
28 6.343
30 6.254
32 6.176
34 6.110
36 6.041
38 5.984
40 5.931
42 5.883
44 5.836
46 5.798
48 5.756
50 5.719
52 5.688
54 5.650
56 5.626
58 5.596
60 5.560
64 5.518
68 5.474
72 5.429
76 5.395
80 5.356

Table 4. Percentage points U I P I N D for survivability Ρ = 95% with a confidence limit of


γ = 95% (excerpt from [4]).

- 249
4.9 A ppendix B: Description of FFA Material and Specimens [5]

The material was a laminate consisting of E­glass filaments with an Epon 826 /Jeffamine
D­230 epoxy resin. The test specimens were taken from a filament wound cylinder made by
Hamilton Standard. The fibre orientation was ±30°. The geometry of the specimens is shown
in Figure 26.

50 [mm]
/ γ -?'
Epoxy Tabs

1
~7

¡¡I 50

m Ili ~7V.
/ 150 Radius R = 450
A λ Laminate thickness t = 3.8

Figure 26. Geometry of FFA specimen

The free length was short enough to prevent buckling, which is why no antibuckling guides
were used. This material is used in the spar of the blade of the Maglarp wind turbine in
southern Sweden. It is a 3 MW machine with a rotor diameter of 80 meters.
Some basic properties of the material are listed in Table 5 below.

Tensile Ultimate Weight per cent


Modulus Strength of fibres
(GPa) (MPa) (%)

GFRP 14 296.2 69

Table 5. Properties of FFA specimen.

250
tr \
The Communities research and development
information service
CORDIS
A vital part of your programme's
dissemination strategy

CORDIS is the information service set up under the VALUE programme to give quick and easy access
to information on European Community research programmes. It is available free-of-charge online via
the European Commission host organization (ECHO), and now also on a newly released CD-ROM.

CORDIS offers the European R&D community:


— a comprehensive up-to-date view of EC R&TD activities, through a set of databases and related
services,
— quick and easy access to information on EC research programmes and results,
— a continuously evolving Commission service tailored to the needs of the research community and
industry,
— full user support, including documentation, training and the CORDIS help desk.

The CORDIS Databases are:


R&TD-programmes - R&TD-projects - R&TD-partners - R&TD-results
R&TD-publications - R&TD-comdocuments - R&TD-acronyms - R&TD-news

Make sure your programme gains the maximum benefit from CORDIS
— Inform the CORDIS unit of your programme initiatives,
— contribute information regularly to CORDIS databases such as R&TD-news, R&TD-publications and
R&TD-programmes,
— use CORDIS databases, such as R&TD-partners, in the implementation of your programme,
— consult CORDIS for up-to-date information on other programmes relevant to your activities,
— inform your programme participants about CORDIS and the importance of their contribution to the
service as well as the benefits which they will derive from it,
— contribute to the evolution of CORDIS by sending your comments on the service to the CORDIS
Unit.

For more information about contributing to CORDIS,


contact the DG XIII CORDIS Unit
Brussels Luxembourg
Ms I. Vounakis M. B. Niessen
Tel. +(32) 2 299 0464 Tel. +(352) 4301 33638
Fax +(32) 2 299 0467 Fax +(352) 4301 34989 '

To register for online access to CORDfS, contact:

ECHO Customer Service


BP 2373
L-1023 Luxembourg
Tel.+(352)3498 1240
Fax+(352) 3498 1248

If you are already an ECHO user, please mention your customer number.
European Commission

EUR 16684 — Fatigue of materials and components for wind turbine rotor blades

Edited by C. W. Kensche

Luxembourg: Office for Official Publications of the European Communities

1996 - 250 pp. - 17.6 χ 25.0 cm

ISBN 92-827-4361-6

Price (excluding VAT) in Luxem bourg: ECU 30

The book is based on work done in an R&D project in the framework of an ongoing
programme of the European Commission. This programme, called JOULE, sup­
ports research and development into renewable energy technologies. It is mana­
ged by Directorate-General XII — Science, Research and Development. The re­
port, as published in this book, is supported by the Commission under EC contract
JOUR-CT90-0071.
Venta * Salg * Verkauf - Πωλήσεις ■ Sales · Vente · Vendita · Verkoop · Venda · Myynti · Försäljning
BELGIQUE/BELGIË IRELAND NORGE

Moniteur belge/ Govern m e ni Supplies A gency NIC Info a/s Roy International
B e i g l i e l i Staatsblad 4­5 Harcourt Road Boks 6512 Etterstad 17, Shimon Haiarssi Street
Rue de Louvain 42/Leuvenseweg 42 Dublin 2 0606 Oslo P.O.B. 13056 *
B 1000 Bruxelles/B­1000 Brussel Tel ( 1 ) 6 6 13 111 . Tel. (22) 57 33 34 61130 Tel A viv
Tel (02)512 00 26 Fax (1)47 52 760 Fax ¡22) 68 19 01 Tel. ( 3 ) 5 4 6 1 4 23
F a i ( 0 2 ) S t i 01 84 Fax (3) 546 14 42
ITALIA SCHWEIZ/SUISSE/SVIZZERA
Jean De Lennoy Sub­agenl l o i the Palestinian A uthority:
Avenue du Roi 202/Koningslaan 202 Licose SpA OS EC
B­1060 Bfuxeiles/B­1060 Brussel Slamplenbachstraße 85 INDEX Information Services
Via Duca di Calabria 1/1
Tel (02) 538 51 69 Casella posiate 552 CH ­8035 Zürich PO Box 19502
Fax (02) 538 0β 41 1­50125 Firenze Tel. (01)365 54 49 Jerusalem
Tel (055)64 54 15 Fax (01) 365 54 11 Tel. ( 2 ) 2 7 16 34
Autres distributeurs/ Fax 64 12 57 Fax (2) 27 12 19
Overige verkooppunten; BALG AR UA
Librairie européenne/ GRAND­DUCHÉ DE LUXEMBOURG
Europress Klasslca BK Ltd
Europeia boekhandel Messageries d u livre 66. bd Vitosha
Rue de la Loi 244/Welslraal 244 5. rue R all leisen BG­1463Solia Middle East Observer
B­1040 BruxeUes/B­1040 Brussel L­2411 Luxembourg TeL/Fax (2) 52 74 75
Tel (02)23104 35 Tél. 40 10 20 41 Sherif St.
Fax (02) 735 08 60 Cairo
Fax 49 06 61 CESKA REPUBLIKA Tel/Fax (2) 393 97 32
Document delivery:
NEDERLAND
Credoc Havelkova 22
SDU Servicecentrum Uitgeverijen CZ­130O0Praha3
Rue de 1a Montagne 34/8ergslraal 34 Tel/Fax (2) 24 22 94 33
Boite 11/Bus 11 Postbus 20014
B­1000 Bruxelles/B­1000 Brussel 2500 EA 's­Gravnhage
Tél. (02)511 69 41 Tel. (070) 37 Zj 880 HRVÄTSKA 4611­F A ssembly Drive
Fax (02) 513 31 95 Fax (070) 3'. 89 783 U n h a m . MD 20706­4391
Mediatrade Tel. Toll Free (800) 274 48 88
Ρ Hatia 1 Fax (301) 459 00 56
ÖSTERREICH
H R 4 1 0 0 Zagreb
Maru'ache Verlags­ TeL/Fax (041) 43 03 E2
J . H. Schuilt Information A /S und Unlversltåtsbuchhandlung
Herstedvang 10­12 Kohlmarkl 16 MAGYARORSZÄG
DK­2620 A lbertslund A­1014Wien Subscriptions only
TB. 43 63 23 00 Tel (1)531 610 E u ro­l n I o­ S erv Ice Uniquement abonnements
Fax (Sales) 43 63 19 69 F a x ( l ) 5 3 1 61­181 Europé Hez Renout Publishing Co. Lid
Fax(Manaoemenl}43 63 1949 Margits ζ ige I 1294 A kjomaRoad
Document delivery: H­1138 Budapest Ottawa, Ontario K1B 3W8
Wirtschafts kammer TeL/Fax M) 111 60 61,(1) 111 62 16 Tel. (613)741 43 33
DEUTSCHLAND
Wiedner Hauplslraßo Fax (613) 741 54 39
Bundesanielger Verlag Α­1045 Wien POLSKA
Postfach 10 05 34 Tel.(0222)50105­4356
Fax (0222) 50206­297 Business Foundation
D­50445 Köln
Tel (02 21120 29­0 ui Krucza 36/42
PL­00­512 Warszawa Hunter Publications
Fax (02 21) 2 02 92 78 PORTUGAL 58A Gipps Streel
Tel. (2)621 99 93.628 28 82
International Fax&Phone (0­39) 12 00 77 Co I ling wood
Imprensa Nacional — Casa da Moeda, EP Victoria 3066
GREECE/ΕΛΛΑΔΑ Rua Marqués Sá da Bandeira, 16­A Tel. (3)9417 53 61
Ρ­1099 Lisboa Codex ROMANIA Fax (3) 9419 71 54
G.C.EteflheroudaklsSA
Tel (01)353 03 99 Euromedia
lnlirrn.ilKjN.il Book Sl 0(0 Fax (01) 353 02 94/384 01 32
N'kis Slieel 4 65. S Ira da Dionisie Lupu JAPAN
GR­10563 A thens RO­70184 Bucuresti
Tel (01)322 63 23 Tel­/Fax 1­31 29 646
Procurement Services Int.(PSJ­Japan)
Fax 323 98 21
Grupo Bertrand, SA Kyoku Dome Postal Code 102
RUSSIA Tokyo Kojimachi Post Otlice
Rua das Terras dos Vales. 4­A
Apartado 37 CCEC Tel. (03)32 34 69 21
P­2700 Amadora Codex Fax (03) 32 34 69 15
9,60­lel rya Oktyabrya A venue
Mundi­Prensa Libros, SA Tel. (01)49 59 050 117312 Moscow
Fax 49 60 255 Sub­agent:
Castellò, 37 TelJFax(095) 135 52 27
E­28001 Madrid Klnokuniya Company Ltd
Tel (91)431 33 99 (Libros) SUOMI/FINLAND Journal Department
431 32 22 (Suscripciones) SLOVAKIA
PO Box 55 Chitóse
435 36 37 (Dirección) Akateeminen Kirjakauppa Slovak Technical Tokyo 156
Fax (91) 575 39 98 Akademiska Bokhandeln Library Tel (03)34 39­0124
Pohjoisesplanadi 39 / Norra esplanaden 39 Nam. elobody 19
Bolelm Oficial dol Ealado PL / P B 128
Tfalalgar. 27­29 SLO­812 23 Bratislava 1
FIN­00101 Helsinki / Helsingfors Tel. (7) 52 204 52 SOUTH and EA ST A SIA
E­28071 Madrid Tel (90) 121 4322 Fax (7) 52 957 85
Tol (91)538 22 95 Fax (90) 121 44 35 Legal Library Services Ltd
Fax (91) 538 23 49
CYPRUS Orchard
SVERIGE PO Box 0523
Sucursal:
BTJAB Cyprus Chamber of Commerce Singapore 9123
Libreria Internacional A EDOS Traktorvagen 11 and Industry Tel. 243 24 98
Box 200 Chamber Building Fax 243 24 79
Consejo do Ciento, 39 I
E­08009 Barcelona S­221 OOLund 38 G riva s Dh ige n is A ve
Tel (93) 488 34 92 Tel (046) 18 00 00 3 Deligiorgis Streel
PO Box 1455 SOUTH A FRICA
Fax (93) 487 76 59 Fax (046) 18 01 25 Nicosia
Tel (2)44 95 00.46 23 12 Satlo
Libreria de la Generalität
de Catalunya UNITED KINGDOM Fax (2) 36 10 44 5th Floor, Export House
Cnr Maude & Wesl Streets
Rambla deis Estudis, 118 (Patau Moja) HMSO Books (A gency section) Sandten 2146
E 08002 Barcolona MALTA Tel. (011)883­3737
HMSO Publications Centre
Tel (93) 302 68 35 51 Nine Elms Lane Miller Distributors Ltd Fax (011)883­6569
Tel (93) 302 64 62
London SW8 SDR PO Box 25
Fax (93) 302 12 99
Tet (0171)873 9090
Matta International Airport LOA 05 Malta ANDERE LÄNDER
Fax (0171) 873 8463
OTHER COUNTRIES
FRANCE AUTRES PA YS
Journal offici«! Ollice des publications officielles
Service des publications des Communautés européennes
des Communautés européennes
2, rue Mercier
26. rue De sa ix Skólavôrdustlg. 2 Dû nya Infoici L­2985 Luxembourg
F­75727 Pans Cedex 15 15­101 Reykjavik TR­80050 Túnel­lstanbul Tel 29 29­1
Tel (1)40 58 77 01/31 Tel. 551 56 50 Tel (1)251 91 90/251 96 96 Télex PUBOFLU 1324 b
Fax (1)40 58 77 00 Fax 552 55 60 Fax (1)251 91 97 Fax 48 85 73, 48 68 17
NOTICE TO THE READER

All scientific and technical reports published by the European


Commission are announced in the monthly periodical 'euro abstracts'.
For subscription (1 year: ECU 63) please write to the address below. o
CD

CD
CT>
5.

c
Price (excluding VAT) in Luxembourg: ECU 30 ISBN i2-flS7-43l3l-b

* " * OFFICE FOR OFFICIAL PUBLI CATI ONS


* cSp * OF HE EUROPEAN COMMUNI TI ES
* •
L-2985 Luxembourg 7 8 9 2 8 2 " 7 4 3 6 U ■>

You might also like