Beruflich Dokumente
Kultur Dokumente
by
StdphaneMahdi
February 2001
--ý'
0,
CL
77
x
wx
ABSTRACT
The aim of the present work has been to model and produce recommendations for the design of
bonded repairs to generic composites structures. An extensive literature survey has shown that
the repair of aircraft structures by adhesively bonded patches has been a standard procedure for
some years now. Whereas current design methods allow a reliable prediction of behaviour of
components, the predictions of deformation and ultimate failure of repaired structures is less
well established. The development of a methodology for the design of repairs to structural
elements of CFRP aircraft structures is however seen as an essential step in ensuring that
maintenance efficiencies and costs are competitive with current metallic structures.
The performance of composite sandwich panels was assessedin four-point bending. Static and
fatigue tests were conducted on undamaged and repaired beams. A 60mm damage to one skin
was assumed and the performance of overlap and scarf patch repairs were assessed,with the
repairs either in tension or in compression. Tested statically and in fatigue, scarf repairs were
found weaker than overlap repairs when tested in compression. Conversely, scarf repairs were
stronger in tension. The finite element (FE) method was used to predict the performance of the
beams. Good correlation between calculated and measured stiffness properties ýkZ observed.
However, the prediction of the failure load was difficult due to the complex nature the beams
failed. The use of small coupons that n-dmic the behaviour of the beams was suggested as an
alternative to FE analysis for repair designs. The static and fatigue performance of coupons was
compared with that of the sandwich beams, and they were found to be in good agreement,
although more work is needed to fully characterise their behaviour under a fatigue regime.
The performance of compression loaded I-stiffened panels was assessed. One panel was
undamaged and served as a control specimen. Two damaged panels were then tested. The
damage was a 55nun cut-out to the central stiffener and attached skin on one hand, and a cut-
out to the skin only on the other. The damage resulted in a loss of 57% and 30% from the
undamaged strength, respectively. The damage was repaired with overlap patches. The repair
schemesemployed were successful and recovered 83% and 90% of the undamaged strength, for
the through-hole and the skin-hole damage, respectively. The FE method was used to predict
the performance of the structures and good correspondence between calculated and measured
stiffness properties was found. The FE method was subsequently used to predict, with some
success,the failure of the undamaged and the damaged panel. The failure load of the repaired
panel could not be predicted because the current approach does not model the interaction
between the repair plugs and the wall of the repaired cut-out, where failure is thought to have
initiated.
2
ACKNOWLEDGMENTS
The present project could not have been completed without considerable assistance
from many sides and I would like to express my gratitude to all those people who
helped me during my stay at Imperial College.
My thanks also go to the staff at the Centre for Composite Materials and at the
Mechanical Engineering Department and in particular to Ms S Grune von Stieglietz for
her invaluable assistance. Offers of technical and practical advice from Dr J
Hodgkinson, Mr W Godwin, Mr G Senior and Mr H MacGillivray are also kindly
acknowledged.
Many thanks to Dr j Chen and Professor MA Crisfield for their help with the
I would also like to thank the Room 398 'Lunch BuncW for their friendship and
Finally I would like to express my gratitude to all those people who gave their help to
this project and specially to my family for their continuous encouragement and
support.
3
TABLE OF CONTENTS
1. Introduction 20
..............................................................................................................
20
1.1 Use of composite materials ..............................................................................................
1.2 Main aims of the present research ................................................................................. 21
2. Literature Survey 22
.....................................................................................................
2.1 Introduction 22
........................................................................................................................
2.2 Composite materials and structures ..............................................................................22
2.2.1 Prediction of static failure 24
..............................................................................................25
2.2.2 Structural design configurations ...................................................................................
2 3R epa i r concep t s 28
. .................................................................................................................. 28
2.3.1 Background
.......................................................................................................................29
2.3.2 Repair procedures
............................................................................................................29
2.3.3 Repair designs
..................................................................................................................
2.4 Adhesive joints and bonded repairs 36
..............................................................................
2 4 1B ack groun d 36
. . .......................................................................................................................37
2.4.2 Analysis of adhesively bonded lap joints and repairs
...............................................
2.4.3 Design of adhesively bonded lap joints and repairs 44
..................................................
2.4.4 Analysis of adhesively bonded scarf joints and repairs 48
............................................
2.4.5 Design of adhesively bonded scarf joints and repairs 49
...............................................
2.4.6 Adhesively bonded repairs in fatigue 52
..........................................................................
2.4.7 Experimental analysis of repaired composite structures 53
...........................................
2.4.8 Prediction of the behaviour of repaired structures 57
.....................................................
2.5 Conclusions 60
........................................................................................................................
3. Experimental Procedures 62
.......................................................................................
3.1 Introduction 62
........................................................................................................................
3.2 The structural elements investigated 62
............................................................................
3.3 The composite sandwich beams 63
.....................................................................................
3.3.1 Introduction 63
......................................................................................................................
3.3.2Materials 64
...........................................................................................................................
3.3.3 Repair procedures 66
............................................................................................................
3.3.4 Experimental programme 71
..............................................................................................
73
3.3.5 Flexural test ......................................................................................................................
3.4 The composite stiffened Panels ......................................................................................82
3.4.1 Introduction 82
......................................................................................................................
3.4.2 Materials 84
...........................................................................................................................
3.4.3 Repair procedures 85
............................................................................................................
3.4.4 Experimental programme ..............................................................................................89
3.4.5 Direct compression test ...................................................................................................89
6
7.4 Conclusions on the performance of the I-stiffened panels 292
.....................................
References 307
Appendix A 318
Appendix B 321
7
LIST OF SYMBOLS
ALPHABETICAL SYMBOLS
moment to a curvature
Dn* Reduced bending stiffness of a plate
D Deflection of the beam (D2-Di) in the four-point bending test
Di Deflection at the loading point in the four-point bending test
D2
Total central deflection in the four-point bending test
E. Parent axial modulus
Ed
Overlap axial modulus
Ei Axial modulus in the i-direction
Ei-i Axial modulus of the ithply in the I-direction
E. Adherend axial modulus in a repaired joint
E(n) Axial modulus after n cycles
EI Bending stiffness (per unit width)
Gy Shear modulus in the i-j plane
G. Adhesive shear modulus
G, Transverse shear modulus of the core (21)analysis)
h. Adhesive thickness
8
hi distance from the neutral axis of a beam to the centroid of the ith ply
I Length
1. Overlap length
n Number of cycles
P Load applied to a bonded joint (per unit width)
P. Ultimate failure load of a bonded joint (per unit width)
Q Empirical variable in the residual stiffness degradation model
W Width
W Load on a beam in four-point bending (per unit width)
GREEK SYMBOLS
a Scarf angle
A Dimensionless coefficient
C Axial strain
r Shear stress
9
ABBREVIATIONS
10
LIST OF FIGURES
Chapter 2
Figure 2.1: Failure modes in long fibre composites 23
Chapter 3
Figure 3.1: The structural elements investigated 62
Figure 3.2: The BAe Airbus sandwich beams 63
Figure 3.3: 5-harness satin prepreg: a top and cross-sectionalview 64
Figure 3.4: A 2-ply overlap patch repair (dimensions in mm) 68
11
Figure 3.7: A 1/10 scarf patch repair (dimensions in mm) 69
Figure 3.8: A beam ready to be repaired (scarf patch repair) 70
Figure 3.9: Repair procedure 70
Figure 3.10: Detail of the specimen for the fractography work 73
Figure 3.11: Details of the four-point bending test rig 74
Figure 3.12: Four-point bending test 74
Figure 3.13: Relative positions of strain gauges 75
Figure 3.14: Bending moment and shear force diagram in four-point bending 76
Figure 3.15: A beam in four-point bending 78
Figure 3.16: Indentation of sandwich beams 79
Figure 3.17: As measured Dl deflections (seeFigure 3.15) 80
Figure 3.18: Corrected D1 deflections (seeFigure 3.15) 80
Figure 3.19: Bending stiffness calculation 81
Figure 3.20: The I-stiffened panels 82
Figure 3.21: Dimension of the panels (dimensions in mm) 83
Figure 3.22: Panel's lay-up sequence 83
Figure 3.23:The 'damaged' panels 85
Figure 3.24: Overlap repair to the back and stiffener face 86
Figure 3.25: Overlap repair to the back face only 86
Figure 3.26: Repair patch design 87
Figure 3.27. Repair plugs 87
Figure 3.28:The repaired panels 88
Figure 3.29: Manufacture of the stiffener repair patch 88
Figure 3.30:Schematics of a potted panel 90
Figure 3.31: Location of strain gauges on the undamaged panel 91
Figure 3.32: Location of strain gauges on the damaged panels 92
Figure 3.33: Location of strain gauges on the repaired panels 92
Chapter 4
12
Figure 4.6: Comparisons of the axial strains on the TF of the beams 98
13
Figure 4.39: The strain field in a scarf joint 150
Figure 4.40: Experimental and computed axial strain distribution: 150
Figure 4.41: Experimental and computed axial strain distribution: 152
Figure 4.42: Surface strain distribution with the TF in compression and in tension 153
Figure 4.43: Computed axial strain distribution in the TF [45/-45] ply 154
Figure 4.44: Calculated adhesive shear strain (1/30 scarf) 156
Figure 4.45: Calculated adhesive shear strain (1/10 scarf) 158
Figure 4.46: Schematics of the variation in adhesive shear strain 158
Figure 4.47: Calculated adhesive shear strain (2-ply overlap) 159
Figure 4.48: Summary of the static failure loci and loads 162
Figure 4.49: Stressstate for a repair in tension and in compression 164
Figure 4.50: Idealisation of repairs 168
Chapter 5
Figure 5.1: Failure of undamaged beams tested with the TF in compression 173
Figure 5.2: Compressive failure close to the repair (a scarf repair is shown) 174
Figure 5.3: Compressive failure at the scarf/overlapping ply junction 174
Figure 5.4: Tensile failure at the parent/repair interface (a scarf repair is shown) 178
Figure 5.5: Tensile failure in the repair (an overlap repair is shown) 178
Figure 5.6: Delamination failure of the repair patch 178
Figure 5.7: Summary of fatigue failure loci and number of cycles to failure (60%UFL)179
Figure 5.8: Sun-unaryof static failure loci and loads 179
Figure 5.9: Experimental stiffness degradation curves: TF in compression 187
Figure 5.10: Experimental stiffness degradation curves: TF in tension 190
Figure 5.11: Observation of a specimen by light microscopy 194
Figure 5.12: Compressive damage on the TF 196
Figure 5.13:Tensile damage on the TF 197
Figure 5.14: Schematicsof the damages observed in compression 198
Figure 5.15:Schematicsof the damages observed in tension 199
Figure 5.16: Onset of damage in undamaged beams tested with the TF in tension 200
Figure 5.17. Summary of fatigue failure loci and number of cycles to failure 202
14
Chapter 6
Figure 6.1: Stress and strain distribution in composite beams 207
Figure 6.2: Strain distribution through the thickness of a sandwich beam 208
Figure 6.3: Geometry of the coupons 209
Figure 6.4: Manufacture of undamaged coupons 209
Figure 6.5: Manufacture of repaired coupons 209
Figure 6.6: Schematics of the locus of failure for overlap coupons 211
Figure 6.7: Loci of failure for the overlap repaired coupons and beams 213
Figure 6.8: Loci of failure for overlap repaired coupons tested in tension fatigue 216
Figure 6.9: Definition of the parameters used in Equation 6-9 for the sandwich beams
Chapter 7
Figure 7.1: Buckling of plates 241
Figure 7.2: Representation of the locus of failure of the D1 panel 242
Figure 7.3: The performance of the undamaged panel (D1) 243
Figure 7.4: Location of the strain gauges on the D1 panel 243
Figure 7.5: Representation of the locus of failure of the D2 panel 244
Figure 7.6: The performance of the panel damaged with a through-hole (D2) 245
Figure 7.7: Location of the strain gauges on the D2 panel 245
Figure 7.8: Representation of the locus of failure of the D3 panel 246
Figure 7.9: The performance of the panel damaged with skin-hole (D3) 246
Figure 7.10:Representation of the locus of failure of the D4 panel 247
Figure 7.11: The performance of the panel damaged with a through-hole and repaired
(D4) 248
Figure 7.12:Location of the strain gauges on the D4 panel 248
Figure 7.13:Representation of the locus of failure of the D5 panel 249
Figure 7.14:The performance of the panel damaged with a skin-hole and repaired (D5)250
15
Figure 7.15:Location of the strain gauges on the DS panel 250
Figure 7.16: Load-displacement curves (Undamaged and damaged panels) 252
Figure 7.17: Load-displacement curves (Undamaged and repaired panels) 252
Figure 7.18: Edges and centre back-to-back strains (undamaged panel) 255
Figure 7.19: Cross-sectional view of the skin-stringer attachment and its equivalent
16
Chapter 8
Figure 8.1: Improved design of the overlap repair 299
Appendix A
Figure A. 1: Compressive failure outside the repair 319
Appendix B
17
LIST OF TABLES
Chatper 2
Table 2-1: A repair methodology for primary structures 34
Table 2-2: Predictions of single-lap joints strength [47] 42
Table 2-3: Experimental failure load of aluminium/epoxy singIe-lap joint [471 43
Table 2-4: Effect of adhesive toughness on joint strength [52] 47
Table 2-5: Tensile efficiencies of pre- and co-cured CFRP scarf repairs [71] 50
Chatper 3
Table 3-1: Properties of the woven materials used on the BAe Airbus beams 65
Table 3-2: Properties of the woven ACG materials 67
Table 3-3: Number of specimens tested statically 71
Table 3-4: Number of specimens tested in fatigue 72
Table 3-5: Properties of the materials used on the stiffened panels 84
Chatper 4
Table 4-1: Test results for undamaged beams tested statically 102
Table 4-2: Test results for undamaged beams tested statically 102
Table 4-3: Computed axial and transverse stressesin the TF (TF in compression) 122
Table 4-4: Tests results for 100mm repaired beams tested statically 129
Table 4-5: Tests results for 100mm repaired beams tested statically 132
Chatper 5
Table 5-1: Tests results for 100mm wide repaired beams tested in fatigue 172
Table 5-2: Tests results for 100 mm wide repaired beams tested in fatigue 176
Chatper 6
Table 6-1: Tests results for undamaged coupons tested statically 210
Table 6-2: Tests results for coupons tested statically 211
Table 6-3: Coupons vs. beams skin failure load 213
Table 6-4: Test results for undamaged coupons tested in tensile fatigue 215
Table 6-5: Tests results for 2-ply HTC overlap repairs in tensile fatigue 217
Table 6-6: Tests results for 2-ply LTC overlap repairs in tensile fatigue 217
a--
18
Table 6-7: Comparison between the fatigue performance of beams-ým flexure and
Chatper 7
Table 7-1: Tests results for the I-stiffened panels 251
Table 7-2: Strain results at failure for the I-stiffened panels 253
Table 7-3: Computed stress resultants for a node across the n-dd-length and in the skin
bay of the undamaged panel, 3.7mm.applied displacement (node 5606) 281
Table 7-4: Computed stress resultants for a node at the edge of the cut-out of the
damaged panel, 1.88mm applied displacement (node 1349) 284
Table 7-5: Computed stress resultants for a node at the edge of the cut-out of the
19
1. Introduction
The potential to reduce the weight of structures, and increase fatigue and corrosion
connection with the design, construction and testing of these materials. Hence, as a
result of an increased understanding of their behaviour, the relative weight of
composites in aircraft has increased steadily in recent years. In the early 1970s,the first
Airbus A300 used a polymer composite in its fin leading edge and other secondary
structures (i.e. a structure that would not lead to the loss of the aircraft if it were to be
damaged). Later, in 1985,Airbus was the first manufacturer to use composite materials
in commercial primary structure applications (i.e. a structure that would lead to the
partial or complete loss of the aircraft if it were to fail) with an all CFRP tail fin [1].
Today a significant amount of primary structure and nearly all moveable controls on
the wings, the empennage, the cabin floor and most of the service panels of the Airbus
are made of composites. There are two main types of structure which are employed:
monolithic stiffened or sandwich, chosen according to the detailed design
requirements. For instance, a 20% reduction in weight and 10% reduction in cost was
achieved by Airbus when the decision to build the A310's rudder in composites
sandwich materials was taken. On the other hand, stiffened monolithic composite
structures were adopted in the design of the A310-30(Ysfin torsion box.
However, the use of CFRP materials is limited due to their poor delamination
resistance which can cause severe reductions in strength and stiffness, and may lead to
catastrophic failure of the structure. This characteristic can limit their use in some
20
particular engineering applications where damage is likely to occur. Thus, repair
methods for in-service or fabrication damage need to be developed to ensure that
composite primary structures are more economically viable than their metallic
counterparts throughout the service life of the aircraft.
The present project aims to produce recommendations for the design of repairs to
static and fatigue loading on repaired structures. The predicted results will be
compared with experimental results. The objective is to calculate the stiffness and
strength of the repaired structures accurately and to estimate the structural failure
load.
repaired structural elements in the way they fail. The objective is to compare the
performance of the coupons with that of the structures. Coupons are tested under
static and fatigue loading to obtain static failure strength and fatigue degradation rates
and the results are used to predict the failure load or mode of the structural elements.
21
2. Literature Survey
2.1 Introduction
Structures are at risk of being damaged during their lifetime and there exist - two ways
matrix cracking, splitting and fibre fracture, as shown in Figure 2.1(a) [3]. The
mechanisms which dominate depend nevertheless on the composite system and the
loading modes.
22
For instance, in unidirectional composites, loaded in uniaxial tension along the fibres,
static strengths can be as high as 3GPa and fibre fracture constitutes a dominant failure
mechanism. Conversely, the strength in compression is generally lower and the fibres
fail by a localised form of buckling, also called fibre kinking (Figure 2.1(b)) [4].
Unidirectional laminates, being very anisotropic in their properties, are usuallv not
used for practical applications, where cross-ply laminates for improved transverse
properties, or angle-ply for increased shear or compression properties, are used.
Although the mechanical properties are improved by laminating, they often lead to
failure modes, such as delan-dnation (Figure 2.1(c)), that are not present in
unidirectional lay-ups, and such failure has been shown to be influenced by ply
Fib- bý
/-/-7-7
fibre pi
44
t+
Fibre failure
(a) m-plane daniage (b) microbuckling (c) delamination
structural configuration that will meet strength, stiffness and weight requirements.
Analysis techniques, either simple or complex (i. e. finite element analysis), may be
employed to determine the behaviour of the structural element under load. A failure
23
2.2.1 Prediction of static failure
combinedstress if the maximum stress in the material reaches the appropriate limiting
value. For composites, the maximum stresscriterion consists of five different sub-
criteria corresponding to the ply strength in eachof the five failure modes taken in the
principal material directions;tensilefailure along and transverseto the fibre direction,
compressivefailure along and transverseto the fibre direction or in-plane shear
failure. So, for the maxianumstresscriterion, failure occurs if one of the following
statements is true [6]:
1
where subscripts and 2 denote the directions along and transverse to the fibres
terms cri and S, are respectively the resolved stress and the ultimate stress in the
angle is close to 0" or 90(', probably because the stressesare assumed not to interact
with each other. Maxwell in the n-dd-nineteencentury, initiated the study of interactive
criteria which take into account three-dimensional states of stress. Von Mises (1913)
later derived such a criterion which was adapted to anisotropic materials by Hill
(1953).Tsai later adapted Hill's criterion to predict strength in composite materials [7].
The Tsai-Hill failure criterion is defined as (using the same notation as above):
222
al 0*1472 47 2 cr
ý2 L--
+ s2 +'>1 at failure
s2 s2
112 12
24
2.2.2 Structural design configurations
2.2.2.1 Introduction
Structural concepts available to designers and used to make structures include
monolithic and stiffened panels, sandwich- and I-beams. Timoshenko [8] suggested
that Duleau (1820) was the first to learn from experiments that the rigidity of a beam
increases with the distance of its "skins" from the neutral axis. Thin beams connected
by a web and, later, sandwich constructions, are direct consequences of these earlier
studies.
Composite sandwich structures are used to maximise the stiffness in relatively low
weight constructions where the main loads are flexural [9]. The core acts as a spacer
for the skins, thus increasing the plate thickness and hence the flexural rigidity. The
stiff skins take the in-plane stresses due to bending and the core is responsible for
shear transfer (Figure 2.2(a)). Composite sandwich panels are often used as fuselage
shells of helicopters and in light structures of civil aircraft, such as rudder, dorsal-fin,
flaps and floor panels, where they may have face sheets as thin as one ply (e.g.
0.125mm). The use of stiffeners commenced in the mid 19th century with the
development of railway engineering where failure in thin wall structures was seen to
initiate by instability and buckling [8]. The stiffeners are added to stabilise the plate
and there exists different geometries as shown in Figure 2.2(b). Blade stiffened panels
are easy to manufacture and versatile in use but are inefficient in bending and have
low torsional stability under axial loading. The J- and I-stiffeners are more difficult to
manufacture but have higher post-buckling strength. Typical wing panels are often of
the skin-stiffener type, with discrete stiffeners bonded to a flat skin [10,11].
; ion
i-h-
25
2.2.2.2 Damage tolerance
During their lifetime, structural components may experience damage or deterioration
either in service (i.e. foreign object irnpact, bird strikes, hailstones, moisture ingress,
etc.) or out of service (i.e. tool drop, foreign object impact, etc.). Structural components
may also experience damage due to cyclic loads. For instance, in fatigue loading the
evolution of damage for multidirectional laminates is shown in Figure 2.3 [3]. Matrix
cracks, leading to fibre failure, that develop in transverse plies occur from the first
cycle. Crack growth then occurs along interfaces between plies leading to crack
coupling. In the last part of the life, delamination is the dominant failure mechanism.
Damage tolerant design requires the establishment of life prediction techniques.
JJI1j::
:: U111i1
::::
000000
:::: '
nIm
0000
2- Crack Debondin&
Interfacial debonding,
4- Delarnination
Growth, Fibre
Fibre breaking Breaking Oocahsed)
LIFE
Impact damage causesa stiffness reduction in the impact area of the skin, resulting in
skinned CFRP sandwich beams (identical to those studied in this work, and described
in Chapter Three). A delarnination of 40mm diameter will effectively reduce the
strengthby a factor of two [ 12].
26
100
90
80
u 70 0
0
60
0
4.1 50
40
30
0
4.1 20
u
10
a)
CLý
0
0 10 20 30 40 50 60
The GARTEUR group (Group for Aeronautical Research in Europe) has recently been
elements [13]. DERA (UK) investigated the damage tolerance of I-stiffened panels
in the bay (i. e. between two stiffeners) led to a 51 % drop in strength [ 14].
Impact in the bay between two stiffeners Impact under the central stringer foot
27
2.3 Repair concepts
2.3.1 Background
way that they are able to sustain all the necessary static and fatigue loads, under the
appropriate ageing and environmental conditions, for their entire life, is of a great
importance. Composite repairs to secondary structures, or non-critically damaged
primary structures, seem to have initiated in the early 1970s to restore the original
mechanical properties to cracked metallic structures [15,16]. Traditional methods of
repair involved the use of bolted patches and, later, externally bonded patches.
The technique was based on the use of CFRP, or boron fibre-reinforced plastic (BFRP)
materials. Boron fibres were initially used for their superior fatigue resistance and low
electrical conductivity. Carbon fibres, on the other hand, were used because they were
cheaper, more readily available and more suitable for curved surfaces (17]. BFRP
patches were used by the RAAF in 1972 to restore strength to cracked structures of
Hercules, Mirage and F-111C military aircraft. Dassault (France) used BFRP doublers
to repair the door frames of Air France's Mercure commercial aircraft in 1974. The
same approach was used by the US Air Force in 1973 to repair the lower wing skin of
an F-111 military aircraft. An overview of repair methods to metallic structures can be
found in references [ 18,19] and the literature survey will now focus on the repair of
composite structures.
It was the National Aeronautics and Space Administration (NASA) and the US Air
Force (USAF) that started to investigate generic repair methods to CFRP structures in
the late 1970sand early 1980s[20-23]. NASA contracted Lockheed Company (USA) to
study repair techniques to CFRP structures [23]. Similarly, in the search for safe,
reliable and operationally suitable repair design for large area damage to composite
structures, the USAF funded the Large Area Composite Structure Repair (LACOSR)
programme. Northrop Corporation (USA) was contracted to undertake this work [22].
Various repair methods were investigated and the main conclusion of these
programmes was that CFRP is a material which, when used in structures, can be
satisfactorily repaired.
28
2.3.2 Repair procedures
Procedures for the restoration of an aircraft to an airworthy state are described in the
'Structural Repair Manual' (SRM) provided by the 'Original Equipment Manufacturer'
(OEM) [24,25]. All parts of the aircraft are described in it with the types of damage
removal of the damaged material, analysis and design of the repair, repair
implementation and repair inspection. The SRM is approved by the Civil Aviation
Authority (CAA) and repair beyond the levels laid down must be agreed both by the
CAA and the OEM. Full-scale tests are necessary to demonstrate the validity of the
Design criteria for permanent repairs are identical to those that are used to design the
Temporary repairs are less demanding and only full recovery of the static strength at
the firnit load is usually required.
The selection of an appropriate repair method for composites is complex and several
factors [2], which are presented next, have to be taken into consideration.
29
REPAIR STIFFNESS: In order to avoid any recalculations of the overall dynamic
behaviour of the components, the stiffness of a repair patch should match the parent
attract more than its share of load, causing adjacent areas to which it is attached to be
overloaded. Furthermore, stiffness mismatch between parent material and the patch
may generate peel stressesthat can debond the patch or delaminate the parent. Panels
loaded in compression are stability dependent and any repair on structures that are
permitted to buckle should not affect the buckling and post-buckling modes. Thus
matching of stiffness is of the utmost importance.
structural criteria the repair may have to be compatible with conditions such as the
critical zone typically includes the wing leading edge, tail and fuselage.
COST-EFFECnVENESS:
Cost and time of repair are the most important criteria
considered by airliner operators [27]. Wherever possible, repairs should be made
within reasonable time and cost limits. Aircraft downtime is costly because the aircraft
is out of service while the repair is being made. The repair procedures should be kept
as simple as possible without requiring expensive facilities or special equipment.
Many papers on repair techniques and concepts have been published in the literature
e.g. references [18,25,28-30] and more recently Schwartz thoroughly reviewed the
subject [31]. Techniques range from cosmetic to structural repair and have been
successfully applied to stiffen under-designed structures, restore strength, reduce
stress concentrations and improve damage tolerance. The various techniques are
discussed below.
2.3.3.1 Potting
A dent on a lightly loaded structurecan simply be filled by a resin compoundand the
surfacethen re-finished. A crucial step is to use appropriate NDT to ensurethat no
delamination has been generatedduring damage.However, this type of repair is no
30
more than a cosmetic repair and, for added efficiency, a thin damaged laminate that
has been resin filled can receive a bonded patch as shown in Figure 2.6(a). A
honeycomb panel may be repaired in the same way provided the core has not been
broken (Figure 2.6(b)).
pressure through the laminate thickness [32]. This method works well for small
delaminations, but has been shown to be inadequate for treating non-interconnected
consists of a scarf joint (Figure 2.6(e)). As the thickness of one side is diminished, the
thickness of the opposite side (i.e. the repair) is increased which implies the removal of
a certain amount of undamaged material. The damage is machined out and the patch
is simply bonded over the clean area. Composite patches can be pre- or co-cured and
can usually be easily formed onto the surface, rendering the concept attractive for field
repair.
and allow inspection of the structure by NDT after repair. The riveting or bolting
operation is very easy to perform and if carefully designed they can restore an
appreciable amount of strength to the parent structure. However, the fasteners
introduce a weight penalty and render inappropriate the repair of aerodynamic
surfaces. However, composite materials posses a lower bearing strength (i. e. failure
due to local in-plane compression at the pin-hole contact area) when compared with
metals, and it is sometimes difficult to design a laminate that will present a high
bearing strength as well as an adequate structural strength.
31
Cosmetic Repairs: Filled and Patched Repair of Delamination
(a) Thin
(b) Sandwich (C) Resill
laminate
structure injection
Patch Repairs
77
(d) Overlap patch repair (e) Scarf patch repair
Sandwich structures can be repaired in the same way as flat laminates, the two skins
being treated separately. However, the repair of sandwich structures with broken
skins requires the replacement of the damaged core and the repair of the skins with an
external or scarf bonded patch (Figure 2.7). The missing core can be replaced by a filler
if it is small, but is usually replaced with a slightly denser honeycomb or foam
adhesively bonded in place. Skin-stringer panels can be repaired the same way as
planar laminates with external overlap or scarf patches as shown in Figure 2.7.
T77
Double side external patch repair Single side scarf patch repair
)k erlap repair
o the stiftoner
32
2.3.3.6 Repair in practice
Historically, bolted repairs were applied to highly loaded primary structures because
they allow the possibility of NDT. Adhesively bonded repairs dizt. favoured in the
aerospace industry for repairing secondary structures, since they are lightweight,
present a smooth aerodynamic surface and do not introduce any foreign object into the
structure, as well as imparting a superior fatigue resistance to the repaired component.
Also, by using adhesive bonding, the repair time is shorter and, provided that surface
preparation, bonding quality and repair design are adequate, it is a repair method
which has shown to restore the static strength up to the undamaged level [20].
The repair of composite structures in the British Airways fleet has been reviewed by
Armstrong [33]. Primary structures, that were not critically damaged, such as
radomes, rudders or ailerons, were repaired with adhesively bonded scarf patches.
Maier [34] described the repair to a structure which is not covered by the SRM. The
vertical stabiliser of a CF-18 military aircraft suffered from damage such as scratches
and delaminations. Due to the thin skin of the structure, bolted and scarfed repairs
were impossible. An external pre-cured patch was successfully applied. Several case
histories of structural elements repaired with a bonded CFRP patch were reported in
However, repairs are not always done in ideal depot conditions but sometimes
accomplished in the field. Ideally repairs should be carried out using the original
prepreg materials. In the field however, these materiatmay not be available and, even
at the depot, the limited shelf life of resins means that the material stock has to be
renewed every 6-12 months. It is therefore sometimes not possible to use the original
materials. The problem of field repair to composite structures was addressed by
companies through several programmes e.g. the British Defect, Damage and Repair
(DDR) [35] and the US Aircraft Battle Damage Repair (ABDR) programmes [36].
Problems of availability of repair material and tooling was elin-dnated by developing
techniques and materials, such as low-energy cure prepreg or low-viscosity adhesive,
adapted to the constraints of field maintenance facilities. Research has shown that
hand-wetted fabric repairs could be applied in most repair situations to current
composites, and could be more cost effective than prepreg repair by reducing depot
stocks [35].
33
2.3.3.7 Repair of primary aircraft structures
Repairs described in the SRM are to secondary or non-critically damaged primary
structures. A critical repair is defined as one in which the static strength of the
unrepaired element would be reduced below the design ultimate if crack growth were
to occur. Repair to primary elements requires certification which, in turn, requires a
demonstration that the repaired structure is as airworthy as the original.
Baker has discussed the problems of certifying composite repairs to critically cracked
primary metallic structures [37]. The repair of an F-111C wing skin was described. A
BFRP patch was applied to the skin and was monitored during the certification
enviromnental behaviour.
GENERIC BONDED JOINT TESTS:Tests on generic joints (e.g. scarf or double lap joints)
are then performed in order to confirm failure criteria based upon the materials'
approaches for generic repair configurations and to assess damage tolerance. Static
and fatigue tests under a variety of environments would produce a structural details'
34
There are several examples in the literature that show that repair procedures are now
taken into consideration at the design stage. In order to gain experience in repair
philosophy, Deutsche Aerospace Airbus ran a series of full scale fatigue tests on
artificially damaged structures (A320 fin box) [40]. After sufficient degradation had
occurred, typical repairs (bolted) were conducted and were shown to perform well.
Similarly McDonnell Douglas (USA) recently conducted tests on an an-composite
wing box with and without damage and repair [411. CASA (Spain) designed typical
repair procedures for solid laminates, stiffened panels and sandwich structures of their
Airbus fleet [42]. Bolted metal or composite plates were used for the repair of skin,
spar and ribs, whereas adhesively bonded patches were used for the repair of
sandwich structures. Analysis (simple joint strength calculation software and FE) was
used for the sizing and the static strength validation of the repair. However it should
be noted that validations of ageing were done purely by testing because no reliable
35
2.4 Adhesive joints and bonded repairs
2.4.1 Background
The design of adhesively bonded joints and repairs is based on the same philosophy, a
repair being a joint made with adherends manufactured at different instants in time.
Bonded joints can be subdivided to several basic types as shown in Figure 2.8 [43]. A
great deal of work has been done, both experimentally and theoretically, in order to
understand the rules that govern the design of bonded joints and repairs. Matthews et
al. [44,45] and Adams [46,47] have published excellent historical reviews on strength
prediction in fibre-reinforced plastic bonded joints. The design of bonded joints and
bonded repairs is identical and guidelines recommend that joints should be designed
so that the adhesive is loaded in shear or compression (as defined in Figure 2.9) [43,
48,49]. Furthermore, for advanced polymer matrix composites, bad design can give
rise to high stresses that act in the direction of the low strength for either the adhesive
or the composite. For optimum results, cleavage, peel and tensile stresses should be
iI
Bonded doubler Scarf joint
-----------------
36
2.4.2 Analysis of adhesively bonded lap joints and repairs
A recent paper reviewed the life of Norman de Bruyne (50], whose work is linked with
the development of adhesives' technology and the first synthetic adhesive, namely the
Redux, phenolic based, system. Needless to say, as adhesive systems improved, in part
due to the work of de Bruyne and co-workers, structural bonding in aircraft found
general acceptance. 'Simple' single-lap shear joints (see Figure 2.8) were used to
evaluate the strength of bonded components. In the single-lap configuration, the
adhesive is ideally loaded in shear by the application of a tensile force in the plane of
the adherends. The shear stress in the adhesive (assumed to be constant), for a joint of
length 1.and width w, subjected to a tensile force P is:
p
Equation 2-1
I. W
It was however recognised that in practice the adherends stretch during the
application of the load giving rise to differential shear stressesat the extremities of the
It
overlap. was on this basis that the behaviour of bonded lap joints was first discussed
in the late 1930s by Volkersen [51]. The so-called shear lag analysis was developed
with the assumptions that the adhesive deformed only in shear and that the adherends
deformed only in tension. The variation of adhesive shear stress along the overlap can
be calculated for identical adherends from (e.g. [47]):
x
cosh(Nf2A
-42A 1.
I' = -ro( Equation 2-2
2 r2-A
sl -,
2
1.2
in which A=G. Equation 2-3
Est,h,
where G,,,1. and k are the adhesive shear modulus, length and thickness, and E, and G
are the adherend in-plane modulus and thickness, respectively. As an example, Figure
2.10 shows the variation of normalised elastic adhesive shear stress (i.e. stress divided
by the average shear stress lo) for an aluminium/epoxy joint (e.g. E,=70GPa, t,=4mm
37
along the overlap length and is maximum at the overlap ends. The maximum shear
stress in the adhesive layer at the ends of the joint, is related to the average shear
stress, V,,,by:
A1A,
rj )1 coth(-)1 Equation 24
In the example taken above, the stress concentration at the overlap ends is about five
times the applied shear stress, and is a function of the coefficient A (equation 2-3). The
Volkersen analysis predicts that, for instance, increasing the thickness of the adherend
or the adhesive layer will decreasethe shear stress concentration and hence increase
the joint strength (if calculated when the maximum adhesive shear stress equates to
the adhesive shear strength). However, even though increasing the stiffness of the
adherend has been reported experimentally to be beneficial an increase in adhesive
thickness can be detrimental [52] due to the likelihood of flaws being present.
E.-70GPa E. -3950INTa
t. -4mm h. -O. 17mnm
11=25mm
5
liz
-10 -5 05 10
Distance along the overlap, 1, (mm)
Another development in the theory of lap joints was the recognition that, due to
eccentricity of the load path, bending of the adherends occurs at the joint extremities
(Figure 2.11) and results in the generation of peel stresses [531. It was shown that
transverse, out-of-plane, tensile stresses (i.e. peel stresses) were maximum at the
overlap ends [52]. The inclusion of peel stresses in the analysis is important as they
38
often lead to delamination failure of composite adherends (Figure 2.11). Hart-Sn-dth
[54] derived an expression for the peel stressesdeveloped at the overlap ends:
where C.,EI and 1.are the adherend thickness, bending stiffness and overlap length, E,,
and h. are the adhesive modulus and thickness and P, the applied load. In this
equation, we can see that peel stressesare related to the applied load and as the load is
increased the severity of the stressesdrops. This is due to the fact that as the load is
increased the overlap rotates to reduce its eccentricity, hence reducing the intensity of
peel stresses.Thin adherends are easier to bend hence effectively reducing the load
eccentricity. Long and thin overlaps are therefore preferred if peel stressesare to be
minimised.
It should be noted that, in these earlier analyses, maximum shear stresses were
predicted to occur at the overlap ends, which should be a stress free surface. Other
features were also overlooked. For instance, the adhesive was usually treated as a
series of shear and tension springs, and variation in shear and peel stresses through
the adhesive thickness was therefore neglected, yielding only 'averaged-through-the-
thickness' values. In reference [46,47], results from the above analysis were compared
with results from an experimental investigation into the variation of joint strength
with overlap, and the experimental results were found to be very different from the
prediction. Later analyses were then performed, initially on metal and then on
composite joints, taking into account adherend bending, shear and peel stresses,as
well as setting the shear stress at the overlap end to zero (55-57]. However, due to an
39
solved to predict the strength of joints; namely the inclusion of the materials' non-
linearity and the effect on the stress concentrations of the presence, or absence,of an
adhesive spew fillet. The issue of the materials non-linearity was first tackled by Hart-
Smith who performed non-linear continuum mechanics analyses in the early 1970s.
Several papers of a high standard were published, for example see his review articles
[43,49,58]. The analyses were based on the Volkersen analysis (including adherend
bending) and on using an idealised elastic-plastic adhesive shear stress/shear strain
curve as shown in Figure 2.12. As the load increases, the adhesive at the ends of the
joint is plastically deformed, and plastic zones form there (Figure 2.12). To predict
failure it was assumed that the adhesive yields at a constant stress (z,, and failure
y)
occurs when the adhesive reachesits limiting shear strain ()If).
Adhesive'k tAdllesive
ray T----------- ra
y
Adhesive
L
shearr stress
str 17)
(13
No
7ay )"af
Adhesive shear strain Distance along the joint
For instance, the potential load carrying capacity of a double-lap joint for adhesiý,,e
rhfý ý+yýf)Et,
Fý -2 ý Equation 2-6
A serious limitation to this approach is that CFRP joints made with modern tough
adhesives are likely to fail by delamination of the parent or by peeling of the patch.
Moreover, for thin joints, the adhesive is often not the weak link and failure of the,
typically thin-sheet, adherends is likely to happen before the joint has developed its
40
full potential (the load capability of the adherend can be estimated by P=E,E,,t, where
Variations and inconsistencies in the strength of bonded lap joints have however often
been reported in the literature. They were reported to be due to the presence of the
adhesive fillet at the overlap ends [47]. Hence, more complicated analyses, such as FE
analysis, were needed to predict joint strength. The use of the FE method applied to
bonded joints was reviewed by Adams et al. [47]. For instance, Adams et al. [591
modelled the adhesive with a spew fillet, which has the effect of altering the stress
pattern at the end of the bondline. They found that the major stresses in the fillet are
tensile and that the maximum principal tensile stress occurs near the corner of the
adherend at approximately 45' to the axis of loading, where they are thought to initiate
Other joint geometries, such as double-lap joints have also been studied and although
their symmetry decreasesthe magnitude of the peel stressesgenerated, local bendhig
moments still cause normal tensile and compressive stressesacross the adhesive laver
(see Figure 2.8). The shear stress distribution is similar to that in a single-lap joint.
However, Adams et al. [59], using the FE technique, reported that in the adhesive
fillet, the ratio of maximum principal tensile stress in a double-lap joint to that of a
single-lap (equivalent in thickness to half the double-lap) is 0.54. The use of a double-
lap results in a decrease of the bending moment, which decreasesthe stressesin the
fillet.
a universal failure criterion for predicting the strength of a wide range of joint
41
geometries has been the subject of much research. To date there exists no widely
It seems that the most widely used criterion is the one that assumes that failure in the
adhesive will occur when the stress (strain) at the most critical point reaches the
independently measured failure (stress) strain of the adhesive. Adams and co-workers
tested aluminium single-lap joints made with four different epoxy adhesives [47]. Two
non-modified epoxy (CIBA MY750 and a plasticised but relatively brittle CIBA AY103)
However, it is difficult when using FE to calculate the value of the stress at a region
with mesh density, whereas there may not be a stress singularity in reality, with local
plastic deformation causing stress redistribution (the stiff adherend will also have a
restraining effect on the adhesive). One solution is to incorporate some degree of
rounding at the adhesive comer. Adams et al. [47] have shown experimentally that an
adhesive fiRet and a radiused adherend end can enhance the efficiency of a joint as
shown in Table 2-3 below. The problem for the modeller does however still exist as an
42
Table 2-3: Experimental failure load of aluminium/epoxy single-lap joint [471
Another method first postulated by Hart-Smith (60] and used in his continuum
analysis (Equation 2.6, Figure 2.12), assumes that failure in the adhesive occurs when
the strain energy density in the adhesive reachesa critical value. This critical value was
taken to be equal to the area under the adhesive shear stress/strain curve up to failure.
It was subsequently used by Joneset al. [6 1] and by the RAAF [62] in conjunction with
a non-linear FE model. However, it does not seem to have been of used elsewhere and
its applicability to a wide rangeof joint geometrieshasnot beenproven.
The problem posed by the singularity obtained in analysis can be overcome by using
averaging criteria. Soutis et al. [63] used an "average strength criterior-Cto predict the
adhesive shear failure strength of scarf repairs. A linear FE analysis was used to
calculate the adhesive shear stress distribution. A 1mm. distance from the singularity
was assumed, from which the adhesive shear stress was averaged. The value was then
compared with the adhesive shear strength. Good agreement was found when
compared with the experimental panel strength, but the choice of the 1mm distance is
questionable. Clark et al. [64] proposed the use of the "ultimate tensile stress over a
zone" to be more universal. Although the adhesive in a joint is largely loaded in shear,
the theoretical analyses [46] have shown that a high tensile stress concentration is
predicted in the adhesive fillet. Clark et al. stated that failure occurs if the largest
distance orthogonal to the principal stress, in the zone bounded by the maximum
principal stress, reaches a critical value. Again, this makes the criterion disputable
because the 'critical value' is not a material property. Furthermore, the criterion
It is noted that one problem with the approachesdiscussed above is that the
measurementof the tensileand/or shearstressversusstrain behaviour of adhesivesis
far from straightforward. For example, special "bulk" samples have to be made and
43
they are prone to premature failure due to their brittleness. Alternatively, special types
of joint are made but the measurement of the low strains involved is very difficult.
out that the design of lap joints is relatively easy [43], with the ratio of overlap length-
to-adherend thickness having a strong influence on the failure mode.
seen, when the overlap is increased from 20mm to 50mm (i.e. a 150% length increase)
[67]. The increase in failure strength may be thought to be as a result of the decreasein
the average shear stress -, as the overlap length increases (equation 2.1).
However, it is noted that there is no direct proportionality between the fracture load
and the overlap length since the adhesive shear stress-strain curve is non-linear, and
the adhesive yields at a more or less constant stress, as shown in Figure 2.12 [47]; the
joint failure load increases with overlap length up to an approximately constant
plateau.
25
20
ýý
15
10
0
0 10 20 30 40 50 60 70 80
Overlap length, 1.(mm)
44
Designers could be tempted to chose the smallest overlap that would achieve a
specified design load. Figure 2.15 shows a double-lap joint, with a short and a long
overlap, and with associated shear stress distributions. If the lap length is too small,
high stresses will exist at the centre of the overlap length and thus make the joint
sensitive to creep and envirorunental. attack [52]. However, if the lap length is
sufficiently long, the shear stress falls nearly to zero at the centre and the load is
carried by the plastic adhesive zones at the overlap ends. The non-uniformity in stress
is desirable and once the two zones are fully formed there is no gain in increasing
further the overlap length. The extent of the elastic and plastic regions can be
ýýerlap length, 1.
E,, t, E 2
'114 "'
P4-p
I
PI (2 r. )
T-
'r.y
Adhesive shearstress
distribution along the T. Vl 0 r-Y
overl ap f or short
1
and long overlap -k . ..........
7-
where A2=(2GA,, Et, )
Figure 2.15: Stressesin long and short overlap double-lap joints [431
composite adherends. It has been suggested that the overlap length-to-thickness ratio
of a single-lap joint is recommended to be at least 80:
1 [43]. This ratio drops to 30:1 for
double-lap joints, as the magnitude of peel stresses generated is less than that for
single-lap joints.
Too thin adherendswill, however,fail cohesivelyunder the action of the applied direct
stressand thereforeprevent the joint developingits maximum efficiency.A minimum
thicknessis thereforerequired, and for compositesadherendsit is also necessaryto
design the joint against delan-tinationfailure. It was shown that peel stressescan be
45
greatly reduced either with a local tapering of the ends of the adherend (e.g. a 1/10
taper) or with a local thickening of the adhesive, as shown in Figure 2.16 [54]. Adams
[46] tested steel/CFRP single-lap joints and reported that a local thickening of the
adhesive at the overlap ends resulted in a 69% increase in joint strength. This was
because the inverse tapering of the steel adherends resulted in smoother load transfer
when compared with the straight edge joints, that failed by delamination of the
composite arrn.
adherend with the lower membrane stiffness is loaded [47]. For this reason, matching
the adherends' membrane stiffness is recommended. Similarly, for double lap joints,
an optimum in strength is attained when the membrane stiffness of the outer
adherends equals half the membrane stiffness of the inner adherend [43].
Therefore, for composite adherends the lay-up and stacking sequenceis important [681
adherend bending stiffness, allowing the joint to rotate with load, but this will in turn
increase the adhesive shear stress. However, matching the membrane and bending
stiffness of the adherend would limit the choice of potential lay-up configurations.
46
2.4.3.4 Effect of adhesive thickness and adhesive properties
The analytical approaches, such as the Volkersen analysis outlined earlier, predict that
increasing the adhesive thickness should lower the adhesive shear stress
concentration, and hence increase the efficiency of a joint. The FE approach used by
Adams et al. [59] has later shown that the former theoretical analyses were wrong and
The adhesive's mechanical properties will also have an influence on joint strength.
Kinloch et al. [52] tested steel/polybismaleimide joints in tension. A polybismaleimide
adhesive, otherwise very brittle, was toughened with rubber particles which induced a
second polymeric phase in the adhesive. The results are summarised in Table 2-4 and
show that the strength of single-lap steel joints increases as the rubber concentration
increases. The fracture strain and the fracture energy of the adhesive are also seen to
increase with rubber concentration. As the adhesive gets tougher, and failure by rapid
crack propagation in the adhesive is inhibited, the development of the full strength of
the joint is permitted.
0 4 0.5 15
10 8 0.9 45
30 18 2.0 140
50 24 2.2 180
47
2.4.4 Analysis of adhesively bonded scarf joints and repairs
In general, due to the eccentricity in the load path, the efficiency of thick lap joints is
limited. Scarf joints are preferred for thick, highly loaded, adherends and can often be
as strong as the members being bonded because they result in a smoother transfer of
the load from one adherend to the other, via the adhesive layer, when compared with
lap joints.
The analysis of scarf joints is relatively complex [60], but it is often assumed that, if the
scarf length is large compared with the thickness of the adherend, the load is
transferred entirely by shear, and in-plane tensile stresses can usually be neglected.
From a force balance argument applied to the adhesive, the shear stress in the adhesive
(assumed to be constant) can be estimated from (for identical adherends) :
sin2a
ro =P Equation 2-7
2.ts
where P is the applied load, a the scarf angle and t, the adherend thickness. A
minimum scarf angle can be calculated from this simple analysis by setting the
condition for reaching an allowable strain in the adherend [69]:
T-Y
a< Equation 2-8
E, c
where E, and e are the adherend's Younes modulus and design strains, respectively.
For example, Soutis et al. [63] estimated the critical scarf angle of CFRP/epoxy repair
joints (assuming that the adhesive yields at -r.y=40MPaand that the joint will fail when
the stress in the adherend reaches the failure stress of the CFRP, e.g. c--454MPa) to be
about 5* (i.e. a thickness-to-overlap length ratio of about 1/11). In practice, a safety
factor is always used and scarf angles are usually limited to less than 3* (i. e. 1/30).
Assun-dng that the joint will fail in the parent when the parent's ultimate failure strain
is reached, or in the adhesive when the adhesive reaches its shear strength (rf), the
joint failure load, P., may be calculated by:
48
2rf
P. =E, F.t, or P. a Equation 2-9
sin(2a)
This analysis, limited to very small scarf angles (below 1/30), suffers from several
shortcomings. The method is inadequate if the adherends are dissimilar or if, as in co-
cured joints, the repair plies follow the scarf surface, giving then a variation of shear
stress along the overlap length. A simple method to account for these parameters is
developed in reference [69].
One problem with these approaches is that the scarf tip is assumed 'very sharp' which
cannot be true in practice. The effect of scarf tip geometry is therefore not taken into
account, and it was shown to lead to variation in joint strength [70]. Again,
improvement in failure prediction may be obtained by using more complicated
sensitive to scarf angle. Webber [71] tested pre- and co-cured (Le wet lay-up) scarf
repairs with increasing scarf angles. The failure load of the joints, as shown in Table 2-
5, is seen to increase with decreasing scarf angles and the failure locus is seen to
change from adhesive to adherend failure. It is remarked that the failure loads of the
co-cured joints were lower (e.g. about half for 1/80 scarf joints). Co-cured patches are
usually cured by vacuum bagging and are therefore more porous than their autoclaved
counterparts. Porosity is thought to decrease the stiffness and strength of vacuum
cured CFRP materials. On the other hand, the failure load of the plain laminate was
about 39kN, which brings the maximum efficiency achieved by the pre-cured joints to
100%.
were found to be approximately twice as strong as 9.2* (i. e. 1/6) scarf joints and again
49
the failure locus was seen to shift from adhesive to adherend failure as angle
decreased.
Table 2-5: Tensile efficiencies of pre- and co-cured CFRP scarf repairs 1711
shown in Figure 2.17. Thin specimens with small scarf angle are likely to have
damaged scarf tip, creating a stress concentration at the end of the joint [70], and can
Brokenscarftip'
An overlapping ply, as shown in Figure 2.17, is often used in real repair situations. It
seems that it then serves a double function. Firstly, it restores the loss in stiffness alld
strength associated with vacuum curing. Secondly, it protects the tip of the scarf and
enhances the load transfer in this area.
Plain scarf joints that simulated repairs were tested by Northrop Corp. [20] and NASA
[23] in the 1970s with efficiencies of respectively 77% and 60%, respectivelv. NASA
50
[23] in the search for higher efficiencies introduced an overlap with straight edges as
shown in Figure 2.18. Failures were reported to be by peeling of the extra plies, since
thick overlaps were shown to develop peel stresses.Tapering of the plies' ends was
In the Northrop programme the replacement plies were of the same stacking order as
the parent, and in addition one ply in each of three directions (0*/45"/-451) was added
as overlapping plies to each side of the joint. But because the coupons tested showed
high efficiencies their design was not investigated. Little information is available on
the effect of the design of the overlapping plies on joint strength. Robson et al. used a
laminate analysis approach to produce some guidelines for determining which extra
plies should be to
added regain the parent strength [72]. A loss of 10% in strength due
to vacuum curing was assumed in their analysis. It was found that, for instance, in
quasi-isotropic laminates one 0' (longitudinal direction) ply is needed for each set of
16 plies in order to restore longitudinal strength and stiffness.
sheets were repaired with scarf patches with identical ply count but different lay-up.
The failure path for all specimens was identical, a crack initiated close to the bondline,
and propagated in the parent following the scarf. The specimens with a patch
membrane stiffness close to that of the parent materials were found to be the strongest.
This is not surprising since Hart-smith [43] has shown that a balanced joint (i.e.
51
identical adherends' membrane stiffness) results in the most uniform adhesive shear
stress. Inasmuch as for the design of lap repairs it is essential to match the parent and
Robson [25] tested CFRP scarf specimens repaired with unidirectional and woven
prepregs. Most specimens showed the same failure path. A crack, initiating at the end
of the scarf, was found to run through the composite along the parent side. However,
coupons repaired with woven fabrics were reported to be weaker than those repaired
with tape materials. Woven fabrics are made by interlacing bundles of fibres to form a
planar mesh pattern. Woven fabrics are therefore thicker than equivalent
unidirectional tapes, and they would therefore be more liable to initiate failure and
may account for the lower strength of the woven repairs.
The static Performance of repaired joints and structures has been shown to be
excellent, with repair efficiency quoted as high as 100%. However the ageing of
aircraft, and inevitably of their joints, is a real concern to industry and little has been
reported yet on this issue. The choice of the adhesive system seems to be of primary
importance in this case and it seems that some adhesives can maintain adequate
fatigue performance, although others cannot [74].
A vast amount of work on the testing of adhesively bonded joints under fatigue has
been reported in the literature. However, only a limited amount of work has been done
to compare the fatigue performance of repaired joints with that of the parent material.
For instance, Charalambides et al. [65] compared S-N curves (i. e. plots of peak stress
versus log cycles to failure) for undamaged and repaired CFRP scarf coupons and
panels subjected to an increasing conditioning (up to 16 months in distilled water at
50"C). The adhesive used was an Hexcel Redux 319 epoxy film and it was found that
there was no major effect of the conditioning on the static properties, despite the
failure locus moving from the parent to the adhesive as the ageing time was increased.
Although an 84% static repair efficiency was quoted.. the repairs did not perform
nearly as well as the parent material in fatigue. Again, pre-conditioning of the joints
did not have any effect. The typical decreasein fatigue life is illustrated in Figure 2.19.
A similar trend was found for the scarf repaired panels. Similarly, Choi et al. [75]
52
tested CFRP coupons and double-lap repaired joints. Although the repaired joints
were said to have recovered up to 80% of the static coupon strength, their fatigue
700
601
xx
m
C. 50(
xxx
Da Rt 00
40(
0 C%O
r- CO rýCD
30(
0 dry
4 mths conditioned dowas( I
20( 0
A 8 niths conditioned
II triths conditioned
I ()( 0 16 mt hs cond itione d
x parent
ot
0.1 10 1000 1W 10,
Number of cycles to failure
2.18) and the approach proved successful. Similarly, Robson [25] investigated the
performance of scarf repairs to thick (48 and 64 ply) monolithic structures. It was
concluded that the efficiencies of the repairs drop as the scarf angle is increased; a
conclusion similar to those drawn when testing scarf coupons (e.g. [70]).
It may be expected that structural elements behave in the same way as the coupons,
but they may be proved superior in some case. For instance, Soutis et al. [76]
investigated the design of scarf repairs in compression and found that the optimum
53
be increased from 3' for coupons (from a 2D analysis) to 7' for real
scarf could
structures. A higher efficiency may be achieved because 'real repairs' (i. e. round
patches located at the centre of structures) do not suffer from edge effects and the
ligaments outside the repair carry some load as shown in Figure 2.20.
Ligament
The Northrop repair scheme as shown in Figure 2.17 (i.e. scarf repairs with overlaps)
[21,22], was subsequently used on sandwich constructions tested in four-point
bending [21]. The paper is well documented and shows photos of repair preparation.
The repair, a co-cured bonded scarf patch, was designed to replace the parent material.
Plies were added (±45/0/±45) on both faces to restore both axial and shear capability.
The panels recovered 100% of their design allowable capability and failure occurred
outside the repair area. Although the loading mode between coupons and beams in
four-point bending was different, this work shows how experience gained in testing
coupons can be applied to the repair of more complicated structures.
The external patches were of the same lay-up as the parent material (i.e. two woven
ply at [+45/-45,0/90]. The panels were tested in direct compression and all recovered
the original compressive strength. Failure was described to be by skin wrinkling in the
parent and they remarked that neither delaminations nor patch disbonds were seen. In
the same way, Westland Helicopters [78] investigated overlap repairs to sandwich
54
structures representative of the trailing edge of a helicopter main rotor blade (T300-
913C/Nomex, 6 ply at [45/-45]). They assessed701C temperature-cure pre- and co-
repairs recovered the undamaged strength, except the co-cured prepreg patch which
failed at a lower load (i. e. 75%). The high porosity in the composite and the adhesive
was said to be responsible for its lower strength. Interestingly, in the DDR programme
[35], hand-wetted 60'C cure, and 175*C co-cured patches were placed on the
compressive side of curved sandwich beams and tested in four-point bending. The co-
cured repair was found to be stronger than the wet repair (strength efficiencies of,
respectively, 83% (prepreg repair), 67% (hand-wetted repair) and 41% (damaged
beam) were quoted.
In the last three papers [35,77,78], co-, pre-cured and wet lay up patches were found,
in turn, equally as strong, stronger or weaker as the parent. It is clear that the choice of
the repair system is important. Ahn and Springer [79] tested different repair systems
on overlap and scarf coupons under different ageing conditions and concluded that
the repair system has a major effect on the results. Wet lay-up repairs were, in turn,
seen equally as good, better or worse than a prepreg repair, depending on the repair
system. For real applications a lengthy experimental programme is needed to select an
appropriate adhesive system. Such a programme was described in reference [80] to
which the is
reader referred for more information on the subject.
As for overlap coupons it seems important to use the adhesive with the highest
toughness and fracture energy [52]. For instance, Clark et al. [8 1] tested damaged and
repaired GFRP/foam beams. The behaviour of the repaired beams was found similar
in nature to that of the damaged ones, though in some casesit was worse, rendering
the repair ineffective. The adhesive used was rather brittle. A tougher adhesive was
then selected and the failure load improved to 7% higher than that of the reference
materials.
stiffener will be more challenging and both scarf and overlap repairs were attempted
55
as shown in references [82,83]. J-stiffened panels loaded by shear were repaired with
scarf and external patch repairs (the paper is not clear on the repair lay-up), as shown
in Figure 2.21. Both repairs failed above the undamaged strength (i. e. 115% and 117%).
A similar approach was used by NASA to repair a lightly-loaded and a highly-loaded
hat-stiffened structure (i.e. respectivelya fin cover and a wing cover) [23]. The sub-
Foam filler
placesdamage
)verlap repair
:) the stiffener
Stiffened panels are sometimes designed to operate at loads above buckling load, and
they are also considered for designs in which they buckle just below their limit load for
applications such as for control surfaces. It is therefore important to make sure that a
repair does not alter the stiffness of a structure, and more importantly, its buckling
capability. Zhang et al. [84] presented experimental and numerical investigations on
the performance of repaired thin-skinned (8-ply) blade-stiffened panels in the post-
buckling range. Part of a central stiffener was cut out and replaced by an overlap patch
as shown in Figure 2.22. The patch had a similar stiffness to the parent (although its
lay-up is not mentioned) and the repair recovered the initial stiffness and strength of
the structure. The buckling pattern of the undamaged and repaired structures were
analysed by the shadow Moird fringe technique. The buckling mode of the repaired
panel was similar to that of the undamaged one, although the size and position of the
buckle peaks were altered due a local increase in thickness around the repair area. The
latter effect seems not to be important, since both panels buckled at the same load of
56
20kN. The failure load, in comparison, was much higher at 60kN and close to the
Skin
Overlap patch
Central bladefilled
with resin
Figure 2.22: Repair to a thin-skinned blade stiffener [841
composite structures. However, geometric scaling problems and edge effects limit the
use of coupons. Furthermore the damage zone in structures is localised, and not
necessarily widespread to the entire volume, leading to progressive damage growth. It
is therefore of interest to develop techniques that can be used to predict the behaviour
of repaired structures. The prediction of the failure strength of structural elements,
even without repairs, is an active and challenging area, however, it is surprising that
there is little in the literature concerned with the modelling of repaired structures.
For instance, Baker et al. [85] discussed the stress distribution in 1/30 scarf repairs
(Figure 2.23) on sandwich beams tested in 4-point bending. The repair was considered
to be successful because the beam skins failure strains was over 7800listrain. A 3D FE
model in which the parent and repair were modelled ply-by-ply was analysed and the
adhesive shear stress distribution along the bondline was found to be constant. Surface
strains were measured experimentally and a good correlation was found when
compared against the calculated values (Figure 2.23). This correlation is important
since it means that the model predicted accurately the stiffness of the repaired beam. A
local increase in strain was reported in the repair adherend above the top of the scarf
(see Figure 2.23), and it was suggested that failure would initiate at this point.
However, the beams failed outside the scarf region under a loading roller. It is to be
noted that sandwich beams are sensitive to through-thickness locafised forces which
57
may have promoted a failure under the load introduction point and prevented the
Figure 2.23: Area where an increase in strain was reported in reference [85]
The analysis of Baker et al. used expensive and time consuming 3D elements which are
not practical to use on a day-to-day basis. Other approaches exist. For instance,
Oztelcan et al. [86] studied the repair of damaged rotor blades. They first used a shell
element model of the entire rotor blade to identify a potential damage site (i. e. a region
of high strain). A local 2D plane strain model, with boundary conditions from the
global model, was then set up to evaluate the adhesive shear stresses developed in an
overlap or scarf patch repair. Failure was assessed at each load increment by
comparing a maximum stress (transverse to the fibres for matrix failure or shear stress
for adhesive failure) to an allowable value. Elastic properties at the nodes for which
the stress has reached the allowable were reduced and a new stiffness matrix was
generated for the successive increments. The simulation effectively reduced
The use of shell elements that includes deformations due to membrane and bending
strain and takes into account coupling and transverse shear effects can be used to
model a stack of perfectly bonded orthotropic layers (quasi-3D). A repair can be
modelled by changing the ply stacking sequence, from element to element, accordingly
to the repair design. Such approach will be discussed later in the present work. Odi et
al. [87] compared the 2D, quasi-3D (i. e. shell elements) and 3D approach to model an
overlap joint and found that all three methods were able to predict the stiffness of
repair joints correctly. The quasi-31) method was however unable to calculate the
58
The quasi-31) method was also used by Zhang et al. [84] to predict the buckling
pattern of repaired blade-stiffened panels, and the method was able to correctly
calculate the position of buckled peaks (seesection 2.4.7.3).The panels were seen to fail
by simple fibre fracture initiating near the central stiffener. The stressesfound from the
FE analysis were used with the Hashin failure criterion [84] and the predicted failure
load for the undamaged and the repaired panels were in good agreement with
It seems, however, that the stiffness of repairs can be predicted accurately, which is
important for stiffness designed elements. However, given the plethora of numerical
techniques that exist, namely 21), shell or 31), with which the designer is free to model
the structures with a choice of more or less precise idealisation e.g' ply-by-ply
modelling, homogenised or real properties, there is a need to identify the outcome of
any assumptions made. Moreover the calculation of the failure strength of repaired
structures for modes other than direct tension in the parent or adhesive shear has yet
to be addressed. The techniques employed are usually very complex and outside of the
scope of the present work. For instance, continuum and FE analyses were used by
Soutis and al. to study the performance of overlap and scarf repaired panels tested in
compression [63,761. Complex FE analyses were used by Chen et al. to model the
debonding of stiffeners from a panel (a case similar to the debond of a repair patch)
[88].
For instance, S-N curves (see Figure 2.19) can be proven very useful but requires an
59
hence requires again an extensive and expensive experimental programme. Moreover,
the process of fatigue degradation is not simply related to the residual strength, with
many composites failing by the so-called sudden-death phenomenon. On the other
hand, it has been established that parameters such as laminate stiffnesses can be
related to the damage state and can be measured in-situ without the need of any
destructive testing. The stiffness of a laminate is seen to decrease with cycles during
testing and stiffness degradation models are then proven useful. Mechanistic
approaches require very little testing, but the models suffer from their complexity as it
is often necessary to follow the evolution of damage with cycles, which is difficult to
obtain experimentaRy.
Most methods have been used to predict the behaviour of composite coupons [89,901
and the same methods are used on bonded joints, however little has been done to
apply them to real structural elements. The issue is complicated and will be developed
in a subsequent chapter.
2.5 Conclusions
40 Composite structures can sustain a variety of damage that can considerably reduce
their mechanical properties. For instance, it was shown that a 40mm delamination
to the skin of a sandwich beams may reduce its flexural strength by a factor of two
(see section 2.2.2.2).Similarly, it was shown that a 15J impact damage to the bay
between the stiffeners of a skin-stiffened panel may reduce its compressive
placed in the structure, they present poor fatigue and aerodynamic performance
and it is difficult to design a laminate with adequate structural and bearing
strength.
Bonded repairs are light, present a smooth aerodynan-dcsurface and, it is possible
to tailor the repair to match the stiffness of the structure. Hence, they may result in
60
good fatigue performance of the repaired components. Disadvantages of bonded
repair are that NDE is difficult to perform, they are difficult to manufacture
(needing a source of heat and pressure) and a proper choice of adhesive, repair and
an adhesively-bonded joint can take. The single-lap joint is the simplest of all
joints. However, due to its intrinsic load eccentricity problem, severe peel stress
concentrations occur at both ends of the joint, hence limiting the thickness of the
adherend that can be bonded on one hand, and the joint efficiency on the other. For
the double-lap joints, and although the peel stressesare not so severe as in single-
lap joint, there is still a limit in the thickness of the adherend that can be joined.
The scarf joints produce negligible peel stresses but it requires very careful
ever increasing although no failure criterion can currently account for the variety
of failure modes in which a composite can fail.
The modelling of the static and fatigue behaviour of repaired joints and structures
is one of the major issues that needs to be addressed.
61
3. Experimental Procedures
3.1 Introduction
In this chapter, the experimental procedures employed for the study of the
Sandwich beam
all
--I
62
3.3 The composite sandwich beams
3.3.1 Introduction
The beams were 700mm long by 200mm.wide, with two-ply CFRP skins on a 15.8mm
Nomex honeycomb core. The panel geometry and lay-up are shown in Figure 3.2. The
skin material was a five-harness satin woven prepreg, and composed of Toray T300
carbon fibres and Hexcel 914 epoxy resin, manufactured by Hexcel (UK) (e.g. Hexcel
F914C).One face had an inner [45/451 ply (i.e. the fabric warp direction is at 45' and
the weft direction is at -45') and an outer [0/90] ply. This face, also called the tool face
(TF), was representative of the trailing edge of an Airbus and contained the repair if
present. The lay-up on the other face, also called the bag face, comprised two plies, of
the same material, but both at [0/90] to ensure failure occurs first on the tool face. An
Hexcel Redux 319A adhesive was used to bond the skins to the core. The skins and
N
l 7.21ntit
-00111111
Core
172..
63
3.3.2 Materials
3.3.2.1 Skins
Figure 3.3 shows a top and cross-sectional view of a 5-harness satin weave for which
one repeat tow of fibres passes over four tows and under one. As a consequence, satin
cloths are not completely symmetrical because, on one side, a high proportion of fibres
are running in the warp direction (i. e. the fabric production direction), whereas on the
other side a high proportion of fibres are running in the weft direction (i. e. the shuttle
direction), which leads to different warp and weft properties. In addition, the crimping
of the fibres introduces different tensile and compressive properties [91 ]. However, in
the present study, equal properties in the warp and weft direction, as well as equal
tensile and compressive in-plane moduli, were assumed for the modelling studies.
1750C for 1hour, plus a 4-hour post cure at 1900C. A fabric areal weight is 285g/M2 and
a cured fibre volume fraction of 56%, leads to a cured ply thickness of 0.3mm. The
tensile mechanical properties of the F914C woven fabric were conununicated by BAe
Airbus [12] and are presented in Table 3-1. They are in good agreement with results
64
3.3.2.2 Honeycomb core
A phenolic impregnated aramid honeycomb core (Hexcel HRH-10), with a cell size of
3.175mm and a density of 144kg/m3 was used. Table 3-1 lists the core elastic properties
woven nylon supported epoxy film adhesive. The adhesive mechanical properties are
listed in Table 3-1 and were taken from Charalambides et al. [65]. It is noted that
reference [65] used an unsupported adhesive film (e.g. Redux 319), and it is
anticipated that the strength data for the supported film would be higher. However,
only the stiffness properties were required for the modelling work of the sandwich
beams, as described in Chapter Four. The cured ply-thickness of the unsupported film
Table 3-1: Properties of the woven materials used on the BAe Airbus beams
F914C 10/90]$ F914C [45/-451$ Core$ Redux 3191c
Yaf - - 0.35
%-.(MPa) - - 50
y
r.f (MP a) - - 60
65
3.3.3 Repair procedures
3.3.3.1 Introduction
A series of tests, performed by BAe Airbus, has established that a 60mm delan-Lination
damage, to the tool face (TF, see Figure 3.2) only, would reduce the bending strength
of the beam used in this work by 50% [12]. Repairs to a 60mm 'damage' area, that
extend across the width of the TF only of the beams, were investigated in this work.
The literature survey has shown that overlap and scarf repairs were favoured repair
methods, and their performance was assessed here under tensile and compressive
regime.
evaluated:
prepreg was supplied by Hexcel. An Hexcel Redux 319 unsupported tape adhesive
was used to bond the patch to the parent. The patches were cured following the
recommendations of the manufacturer (see section 3.3.2.1), but under vacuum only.
The properties of the F914Cfabric and the Redux 319 adhesive were presented in Table
3-1, and no loss in strength and stiffness associatedwith the vacuum curing operation
were taken into account in the present work.
66
Table 3-2: Properties of the woven ACG materials
LTM26EL [0/90] LTM26EL[45/-45] XLTA225
E2
Ei = (GPa) 60 17.1
Ei (GPa) 9 9
V12 0.04 0.75
V13 - V23 0.3 0.076
G12
(GPa) 5 32.5
G13 = G23
(GPa) 5 4.9
S(I d 2) (MPa) 700 90
.. -
S12 (MPa) 90 25.6
-
repairs. The principles of design of bonded repairs, presented in section 2.4.3, were
followed for the sizing of the patches.
Overlap 12atchrepairs:
Two-ply overlap repairs as shown in Figure 3.4 were produced. The patch resembles a
four-ply laminate, but the two bottom plies do not play a role in the strengthening
mechanism, they just "fill the gap", and low density fillers are often used instead. The
overlap repair -to one skin of a sandwich beam can be idealised as a supported lap
joint. It was assumed that the core provides a support to the overlap. The patch lay-up
was chosen identical to the removed material (e.g. a [0/90,45/-45] woven lay-up) to
have a membrane stiffness identical to the parent TF skin.
In order to keep peel stressesto a minimum, it was recommended (see section 2.4.3)
that a ratio of at least 30 of overlap length-to-adherend thickness is used in the design
of supported lap repairs [43]. The minimum overlap length, for a 0.6mm thick overlap,
would therefore be 18nurL The thicknesses involved are however small and peel
stresses are not likely to promote an adhesive peel failure. Therefore, in the current
work it was decided to evaluate the minimum overlap length that would support the
load. Current aerospacepractice [43] requires that the plastic zone that develops under
load at the overlap ends, carry all the load at the design ultimate condition. The load
carrying potential of the repair for adhesive failure (Equation 2-6) is P. =215N/mm
-T
67
width (i. e. with the properties shown in Figure 3.5). However, thin adherends are
more likely to fail by the action of the direct stress and the load capability of the
10/901
145/ 451
in/gni
[45/ 451 Ad h-- IIt ,ýv
/'-451/10 )01
P
L CORE
Es, t,
P/(2rýy)=
2mm
G,,=1.35GPa
E,=33.4GPa li,,=0.17mni
t, =0.6mm ý,,,=50MPa
r, /10
y
)/,Y=0.35
=0.05
7,:f
-4-
7- where 22=(2G,,Ili,, E,t, )
68
Scarf patch repairs:
It was recommended, in section 2.4.5, that a 3' angle be used in the design of scarf
joints [43]. A conservative 1/30 scarf (i. e. a 1.9' angle), as shown in Figure 3.6, was
initially used, but this however implies the removal of a lot of undamaged material.
Thus, a 1/10 scarf (i. e. a 6' angle), as shown in Figure 3.7, that would remove less
material, was also assessed in the present studies for comparison. Matching the parent
and repair membrane stiffness has been shown to be important for scarf repairs and
the patches were therefore made of two woven plies identical to the parent, e.g. a
[0/90,45/451 lay-up. The potential load capability of the joint for adhesive failure
(Equation 2.9) is P,,=1086N/mm and 346N/mm for the 1/30 (i. e. 1.90) and 1/10 (i. e. 6')
ply was used to enhance the strength of the repair since, as discussed in reference [23]
(Figure 2.23), an overlap protects the tip of the scarf from being damaged. It is noted
that repairs to thin structural elements do not leave much choice of lay-up to the
designer, and a [45/-45] overlap ply was chosen, as it was felt that a [0/901 overlap
would overstiffen the repair and create a stress concentration in the vicinity of the
patch. The final repair patch lay-up was therefore [45/45,0/90,45/45]. The same
Tool face
145/ 45]/10/901
145/451 'i
-71.4
[0/901
65.7---ý Tool face
145/451 [45/ 451/10/90]
69
3.3.3.4 Manufacture of the repair patches
The 'damaged' area, on the tool face of the beams, was machined off with a diamond
coated cutter, leaving clear honeycomb cells on the surface. A beam ready to be
repaired with a scarf patch repair is shown in Figure 3.8. The repair area was then grit-
blasted and subsequently thoroughly cleaned with acetone. A patch was then applied
directly over the prepared surface.
TOOL FACE
SCARF
TRANSMON
REI;ION
AREA TO BE REPAIREO
HONEYCOM8 COQE
A perforated PTFE film (1gm perforation) was laid over the repair area, together with
a porous Melinex release film. The entire beam was covered in a breather cloth, and
placed in a vacuum bag, as shown in Figure 3.9. All repairs were cured according to
the recommendation of the manufacturer, but with vacuum pressure only.
A
Repair Patrh
Beam
70
3.3.4 Experimental programme
Sandwich constructions are often used on aircraft for applications where bending
loads are important. Hence, the specimens were tested in four-point bending. This test
has the advantage of being rather simple to perform, without the need for special
specimen preparation (i.e. no glued backing plates). A total of twenty eight, 200mm
wide, beams were supplied by BAe Airbus, and were used in the present work. It was
anticipated that a higher number of specimens would be required for the fatigue
characterisation work and the first part of the programme was concerned with the
static characterisation of undamaged beams, in four-point bending as a function of the
width of the beam. An optimum width was subsequently chosen, and the work
proceeded with the static characterisation of repaired beams and the fatigue
characterisation of undamaged and repaired beams.
or in tension. Load, ram deflection and surface strains were recorded up to failure. The
details of the specimens tested statically are presented in Table 3-3.
My overlap HTC 1 1
71
3.3.4.2 Fatigue characterisation
Fatigue loading consists of the application of fluctuating loads, on the specimen, which
are less than the static value required to fracture the specimen. Undamaged and
repaired beams, 100 mm in width, were tested in fatigue, TF in compression or tension,
at peak fatigue loads ranging from 50% to 85% of the respective static ultimate failure
load (UFL) of 200mm wide beams. The ultimate static failure load (UFL) of 200mm
wide beams was chosen as the reference load, as extensive compressive fracture data
were already available [12]. As it will be shown in Chapter Four, there does not seem
to be any significant difference between the UFL of 200mm and 100mm wide beams
but the UFL of 200mm wide beams (in compression) were statistically significant. A
load ratio R=0.1 and sinusoidal test frequency of 1Hz, were used. Ram displacement,
load and surface strains were monitored with the number of fatigue cycles. The cyclic
loading introduces a degradation of properties such as strength and stiffness [94]. The
evolution of the bending stiffness of the beams was monitored with fatigue cycles. The
details of the specimens tested in fatigue are presented in Table 3-4.
3.3.4.3 Fractography
Optical microscopy was used to assessthe degradation mode of selected undamaged
and repaired beams tested in fatigue. Beams,as detailed in Table 3-4, were cycled, with
the TF in compression or in tension, for a limited time and the fatigue test was stopped
before failure occurred. The beams were then cut along their length and used for the
fractography work. The other two were kept for future work.
72
The aim was to examine the surface using an optical microscope after damage had
occurred. The specimens were cut to a length spanning from the load introduction
point to the centre of the beam, as shown in Figure 3.10. The inner surface of the
specimens was ground using a rotary grinder and watered silicon carbide grit (from a
240 to a 1200 grit). The polishing stage was achieved by employing a synthetic cloth-
covered rotary polishing wheel with polishing fluid added. The specimens were then
examined under a light n-dcroscope which has objectives with a range of
magnifications, and which was equipped with a CCD camera, linked to a PC.
The beams were tested using an ESH servo-hydraulic bi-axial fatigue testing machine.
A rig was manufactured as shown in Figure 3.11. Load spreaders and rubber pads, 25
mm in width, were used at the loading points to prevent local indentation and failure
of the skin and the core. The span between the outer and the inner loading points was
selected by BAe Airbus to promote skin failure. Figure 3.12 shows a view from above a
beam being tested.
consecutive data sets were recorded, could be varied by the software. For instance, in
static tests, data were typically sampled twice a second, continuously, and until failure
of the specimen. In fatigue experiments, the data were recorded every thirty minutes,
for two seconds,and then they were sampled forty times a second. The individual data
points were then treated in an Excel spreadsheet. Thus, it was possible to analyse the
73
II, Iý,. I, It,,,,
it,, tIý,.
tatig., t, stmg -a, hm,
V-'ý
Inner i (, Het
Sandwich Beam
(hiter roller
700nim
74
3.3.5.2 Application of strain gauges
Strain gauges were used for the determination of surface strains. The application of
gauges required preparation of the surface to which the gauge is to be applied which
includes an abrasion stage (e.g. grit blasted with a silica grit) and a cleaning stage (e.g.
with acetone). The adhesive is then applied to the carrier surface and the gauge
positioned carefully on the surface. Pressure, with the use of a soft rubber pad, is then
applied overnight.
A variation in measured strains with gauge size occurs with woven fabrics [95]. Long
gauges average the local variations in strain due to the weave pattern, whereas small
gauges, of length less than the weave pattern, show a lot of scatter. Acceptable results
were found for 6mm long gauges, with which the axial modulus of the F914C [0/901
specimens was measured to be 59.9GPa [95]. This is very close to the 60GPa given in
Table 3-1.
MEME CEA-06-125UN-350 strain gauges were used. They had a gauge factor of 2.085
and a 3500 resistance, for a 6mm gauge length. The gauges were connected to 1/4-
Wheatstone bridge amplifier (2.5V applied voltage). Strain measurements were taken
huwrloadog points
----------------
Straingauges
at the centre
75
3.3.5.3 The four-point bending test (4PB)
The advantage of using a 4PB test is that the region between the inner supports is
subjected to a constant bending moment (VVb)while the transverse shear force is zero,
as shown in Figure 3.14. The constant bending moment means that the curvature of the
beam is constant over that region, which in turn means that the axial stresses in the
skins are constant. In the region between the inner and outer supports the transverse
force is constant (Vv), which in turn means that the shear stress in the core is constant
ww
Figure 3.14: Bending moment and shear force diagram in f our-point bending
3.3.5.4 Stiffness
The study of the deformation of sandwich beams uses the same principles as classical
engineering simple beam theory (SBT), with the exception that transverse shear
deformations are accounted for [97]. The deflection of the beam is the sum of the
ordinary bending deflection and the deflection due to the shear forces. It was assumed
76
that the core carries the shear forces and the skin carries the direct forces arising from
bending. These simplifications are largely satisfied with a so-called beam with
'antiplane core' (Le honeycomb cells core) and thin skins (see e.g. [97] for further
details). The skins and the core of the beam were assumedto be perfectly bonded
together. Each section was assumedto bend cylindrically, with the cross-sections
remaining plane and perpendicularto the neutral axis.The skins were assumedto be
very thin so that the local bending stiffnessaround their own axis could be neglected.
Furthermore, the core and the skins were assumed not to deform through their
thickness.
The beamýsdeflection, D, the ram deflection, D1, and the total central deflection, D2.,Of
a homogeneous beam in four-point bending, as defined in Figure 3.15, are given by:
[96]:
nC2
D=- Equation 3-1
2EI
2C 3)
W(3b +b
D Equation 3-2
3EI
"C2 n2C n3
D2 =-+-+. Equation 3-3
2EI El 3EI
where W is the applied load and b and c are the dimensions shown in Figure 3.15, and
EI is the bending stiffness of the bean-LThe contribution of the shear force to the
deflection can now be added to Equation 3-2 and 3 as [97]:
"te
d,
h. r 2 Equation 3-4
2G, d
where G, is the core transverse shear stiffness and the other dimensions are shown in
Figure 3.15. Later, the experimental measurement of the bending stiffness of the
undamaged and the repaired beams will be described. However, analytical expression
for undamaged beams, based on SBT, can also be sought. The bending stiffness of an
undamaged sandwich beam (Eý, composed of multiple layers (with an ith ply of axial
modulus Ei) is given by [61:
77
In
EI=- Equation 3-5
3
j=1
where Ii, is defined in Figure 3.15. As an example, the bending stiffness (per unit
width) of the undamaged beams considered in this study is EI=4.54E6Nmm (with the
material properties given in Figure 3.15 and ignoring the core's contribution). The ram
deflection, DI, of such a beam at the load (per unit width) of W=25N/mm (therefore a
load of 500ON for a 200mm wide beam), is 37.4mm (with b=190mm, c=125MM). In
comparison, the shear deflection is 1.8mm (with G,=76MPa, t, =16mm and d=16.6mm).
-I-- W-ýw A
.4%-
W*b
D2
62
4-
Layer 1: El= 67GPa ------
A h2 Layer 2: F, = 17.1 Gpa
d L0 tl, 2
-------------
0-
hl, Layer 3: E,= 67GPa
+ve PIN, thickness=03mm
Layer 4: Eý= 67GPa
A standard test will involve the application of the load to the rig (e.g. 2M, and the
measurement of corresponding deflections. Care, however, must be taken when
measuring the deflections. Indentation of the skins and deformation of the rig upon
loading means that a correction for the observed deflection should be considered as
78
Discussion of the possible errors
A calibration was attempted using a 200mm wide beam. A panel was placed under
5000N. The indentation of the skin is therefore negligible and it was ignored.
0.8
Oß
E
.50.4
0.2
The deformation of the test rig was studied next. Figure 3.17 shows the deflection of a
200mm wide beam, at the loading point (ram), as read on the machine controls,
together with the displacement of the beam as measured with a travelling microscope
set on the top skin, of one edge, of the beam. The shift between the two curves is due to
the slack in the system that is being recorded at very low load. This difference can be
corrected by plotting the values recorded from the machine controls for loads ranging
from 25% to 50% of the failure load. A regression analysis is then performed, and the
deflections are then set back to zero at zero load. In the example in Figure 3.18, the
displacement read from the machine controls were set back 0.91mm. The corrected
deflection curve is seen to be overlapping the deflection measured by the travelling
microscope. It is noted that, at failure, the recorded ram deflection is about 39mm, and
hence close to the theoretical 39.2mm (i. e. 37.4mm + 1-8mm) calculated earlier.
79
45
40
35
30
25
20
151
10
51
0
0 2,000 4,000 6,000 8,000 10,000
Load (N)
45
0 as measured with a travelling microscope
40 as read on the machine controls and corrected
- as read on the machine controle (0.25 to 0.5*load)
35
Linear fit (0.0039*load+0.91)
0
30
'0
0
25-
0
20-
15-
10-
5- 0
0-9
0 2,000 4,000 6,000 8,000 10,000
Load (N)
80
The recorded values of displacement were corrected following the aforementioned
procedures, for all tests. The bending stiffness of undamaged and repaired beams was
then back-calculated from the experimentally measured values of deflections and
loads. For example, using the data in Figure 3.18, it was possible to calculate the
bending stiffness of the beams in terms of load by rearranging the expression for the
2C
W(3b + b')
EI = Equation 3-6
3D1
ram deflection, D1, measured with a travelling microscope. Figure 3.19 shows that the
three methods are within 10%, except at very low loads, which is due to measuring
errors. The theoretical bending stiffness, calculated earlier, is seen in Figure 3.19 to
overestimates slightly the measured bending stiffness, because the SBT analysis
assumed equal axial and transverse mechanical properties, as well as equal tensile and
compressive modulus. The shear deflection was ignored in the calculation. The error
is, however, small.
A Theoretical
4.5E6 bending
stiffness
A
4.OE6 AAA
ZEIýý),
81
3.4 The composite stiffened panels
3.4.1 Introduction
The panels were manufactured to be 400mm long by 360mm wide. The 4.87mm thick
skin was supported by three I-section stringers, as shown in Figure 3.21. The stringers
were 120mm apart.
360mm
ai, f'V
400mm
The material used was T300/914 unidirectional prepreg (Toray T300 fibres and Hexcel
914 resin, from Hexcel (UK). The quasi-isotropic (+450/-450/00/90)4, skins were laid-
up first according to the axis shown in Figure 3.21. The I-section stringers were
separately assembled from four uncured Ian-dnates, comprising a tapered foot, two C-
sections back-to-back and a spar cap, all of the same lay-up (+45'/-45'/0')2,, as shown
in Figure 3.22. The triangular space between the webs and the spars were filled with
strips of unidirectional prepreg (Figure 3.21). The foot and cap of each stringer have
the web is symmetric, because a +45' become a -45' when the plies are wrapped into a
C shape. The stringers were then secondarily bonded to the skins using Redux 319
adhesive.
82
360---- -
30
1.83
4.87
40
55---
-45
-45
-45
u
45--
-- w-
45
-45
ju-
-- zu, -
ZT--
--- ö--
. 49--
---- jr-
zr
, 2. m
-45
83
3.4.2 Materials
a temperature of 1751Cfor 1hour, plus a 4hour post cure at 190*C. It has a cured fibre
volume fraction of 56% and ply thickness of 0.152mm. The mechanical properties of
this widely used aerospace material can be found in the literature, e.g. reference [98].
Hexcel Redux 319 was used to bond the stringers to the skin, and its mechanical
properties were determined by Charalarnbides et al. [65]. Table 3-5 lists the properties
of materials used on the stiffened panels. The material axis is such that 1 represents the
fibre direction, 2 is across the width, perpendicular to the fibre direction, and 3 is
through the thickness of the prepreg.
Vn 0.02 0.4*
V23 0.5 0.4*
V13 0.32 0.4*
G12
(GPa) 5.4 1.35
G13
(GPa) 5.4 1.35
G23 3.6
(GPa) 1.35
S, (Tension) 1490MPa 64.5MPa
S, (Compression) 1402MPa
S2 46 MPa 64.5MPa
(Tension)
S2 215 MPa
(Compression)
S3 46 MPa 04.5MPa
(Tension)
S3
(Compression) 215 MPa
S12 79 MPa 60MPa
* are estinzatedproperties
84
3.4.3 Repair procedures
3.4.3.1 Introduction
As stiffened panels are good candidates for primary aircraft structures it is of concern
to establish a technique for the repair of stiffener elements. The literature survey has
shown (see section 2.2.2.2) that an impact over a stringer can lead to a significant
reduction in the compressive strength of panels designed for high buckling strains
[141. The panels were not impacted in this current work, however a 'damage' was
assumed and was cut out with a diamond-coated tool. The damage simulated an
impact to an area immediately underneath the central stiffener. In one case the
damaged area was assumed to be a 55mm circular area to the central stiffener and skin
(Figure 3.23(a)).The second damage type was similar but was confined to the skin only
and the stiffener was left intact (Figure 3.23(b)). The literature survey has shown that
overlap and scarf patch repairs are good candidates for the repair of stiffened
constructions. Overlap patches were chosen as it was felt that a scarf patch would be
too complicated to manufacture for in-situ repairs.
III
IIII
I
(b) Paneldamagedby a 55mm hole through the skin undem,, it I, tit, --ntral shIfenei
Back face Stiffener face
Stiffeneq
85
3.4.3.2 Repair materials
A repair system, identical to the parent materials, was selected; the unidirectional
was used to bond the patch to the parent. The patches were cured, according to
manufacturer's recommendations, but under vacuum only (i. e. lhour at 175'C, plus a
4hour post cure at 190'C). The material properties were assumed to be unaffected by
the vacuum curing operation and were presented in Table 3-5. However, as the panels
Two overlap repair schemes were evaluated. One involved the repair of the back and
front face of the panel, including the stiffener, while the second involved the repair of
the back face only as shown in Figure 3.24 and Figure 3.25, respectively.
Fý_ III
... -
Overlall patch
Skin lay-up: [(+45/-45/0/90M, lay-up: j(+45/45/f)/90)., ts
I'he'damage'consists of a 55nim hole through the skin underneath the central stiffner
I R"r. i, plul" I
Adh'-,
Skin lay-up: [(+45/45/0/90ý]s Cl,,,, I, I, 1,.t, h ky-1,
[(, 45,/- 15/0, '90) J,
86
The repair to the back and stiffener face is shown in Figure 3.24. The patches were
treated separately, as two supported overlap repairs, as outlined in Figure 3.26. The
patches were square in plan form, and their lay-up was identical to that of the parent,
although with only half the number of plies in order to halve their membrane stiffness.
Thick overlap repairs with straight edges were reported to fail prematurely [23].
Tapering of the plies' ends was however shown to limit the development of peel
stresses, and hence increase repair efficiency. The plies were graded in thickness, with
an overlap length 30 times the ply thickness (e.g. 300.15=4.5mm). Appropriate pre-
cured plugs (i. e. a skin and/or a stiffener plug), as shown in Figure 3.27, were used for
the repairs. The extent of the repair to the back and stiffener face is given in Figure
3.28(a). The repair of the damage to the back face only is shown in Figure 3.25 and
Figure 3.28(b), and is identical to the repair to the back face shown in Figure 3.24.
0 15. m
87
(a) Repaired panel damaged by a 55mm hole through the central stiffener
-rr-64mm- 99mm ni
(b) Repaired panel damaged by a 55mm hole through skin underneath the central stiffener
Baýk face Stiffener face
III-II
IIII
plugs were then placed in the cut-out regions, enveloped by adhesive to ensure that no
clearance existed between the plug and the structure wall (a clearance of 0.5mm exists
between the plugs and the wall). The patch to the stiffener was then applied, in one
operation, with the help of wood dams to make sure that the patch would follow the
curvatures of the stiffener, as shown in Figure 3.29. The dams were subsequently
removed and the patches were vacuum bagged following the procedure shown in
Figure 3.9.
?d
Wood
d-,
88
3.4.4 Experimental programme
The major part of the structure of an aircraft takes the form of monolithic structures
that are stabilised with respect to compression loading by stringers. Five panels were
supplied by DERA and were tested in direct compression, as it is a relevant loading
The panels were loaded up to failure, and the load, cross-head displacement and back-
to-back axial surface strains were recorded during the tests.
The tests were carried out in a 25OOkNtesting machine, which has been specially
designed to have the high stiffness necessary for buckling and post-buckling
experiments [100].
The ends of the panels were first machined, parallel and flat, to ensure an even
loading. The length of the panels after machining was 398mm. As it is necessary to
protect the loaded ends from local damage and crushing, the ends were potted in
50mm deep aluminium C-channel sections. The potting compound used was supplied
by CIBA (UK), and comprised CY219 resin, 200parts; HY219 Hardener, 100parts;
DY219 Accelerator, 4parts; a slate powder, 40parts and chopped strand mat, 14parts.
The panels were left in the C-section and the compound was simply poured in and left
to cure for 24hours. The free length of the panel after potting was 308mm, as shown in
Figure 3.30.
The testing machine comprised two platens that are 1m long and 0.5m wide. The
panels were placed at the centre of the platens. It is, however, important to ensure that
the specimens are loaded uniformly across their cross-sectional area. Although the
89
panel's ends were machined flat, it cannot be guaranteed that the machine's platens
are flat and parallel. Hence, a thick adhesive paste (CIBA Aradilte 2011 epoxy paste
adhesive bulked up with slate powder) was applied to the top end of the panels, after
they were placed in the machine. A thin aluminium plate was then just laid on. The
panels were loaded up to a load of lOkN and left under load overnight. The excess
adhesive was removed and the panels were then subsequently unloaded. Once this
operation was done, the panels were not moved from the machine. The tests were
Aluminium plate
---l
Ad
41ý
seconds, until failure of the specimen. The individual data points were then treated in
an Excel spreadsheet and were corrected for zero displacement at zero load, as
described in section 3.3.5.4.
specimens were hand-abraded with a silicon carbide grit paper, and then carefulIV
cleaned using a solvent. An instant cure cyanoacrylate adhesive was used to bond the
gauges to the specimens. The gauges were positioned carefully onto the surface and
pressure was applied for a few seconds.TML (Japan) FLA-6-11-IL strain gauges were
90
used. They had a gauge factor of 2.12 and a 120Q resistance, for a 6mm gauge length.
The gauges were connected to the data acquisition system, which comprised eighteen
Sixteen strain gauges per panel were used on average. The location of gauges on the
undamaged panel is shown in Figure 3.31. Fourteen back-to-back gauges were used
and they served a double purpose. Eight back-to-back gauges were placed half way
along the panel length, whereas six of them were place at a of the length. The
uniformity in load within the panel was checked by comparing the strain readings
from the aforementioned groups of gauges, as they should be identical at low loads.
Furthermore, the extent of any out-of-plane deformation can be determined from the
calculated and can give an indication of when buckling has occurred. The strains on
The location of strain gauges on the damaged panels is shown in Figure 3.32. Some
back-to-back strains were recorded as previously, but the gauges were placed in the
vicinity of the 'damage' in order to determine the extent of any strain concentrations.
The location of strain gauges on the repaired panels were the same as on the damaged
308 .m
15
11 13
-----
----- 97
35 775
25
15 12
79 13 11
91
21 le
15 17 97
200
-- -------- _T
77.5
15 15
79 17 15
21 19
15 17
97
79 17 15
deformation during testing. The back face of the panels were painted white and placed
in front of a thick Plexiglas grating (0.4mm between each line). The grating was placed
as close as possible to the panel and was illuminated by a projector at 600 to the normal
92
3.5 Conclusions on the experimental procedures
repaired CFRP structural elements have been reported in this Chapter. Two types of
structures are being examined:
Undamaged and repaired sandwich beams were tested staticafly and in fatigue. The
'damage' was assumed to be a 60mm delan-tination damage on the centre of one face
only of the beams. Scarf and overlap patch repairs, that extend across the width of the
beams, were assessed. Two repair systems were chosen; a high-temperature cure
Undamaged, damaged and repaired skin-stringer panels were tested statically. The
'damage' simulated a ballistic impact to an area immediately underneath the central
stiffener. On one occasion, the damaged area was assumed to be a 55mm circular area
to the central stiffener. The second 'damage' was identical but it was assumed to be
contained to the skin only and the stiffener was therefore left intact. Two repairs
schemes were evaluated. One repair, involves the repair of the back and front face of
the panel, including the stiffener. The second repair involves the repair of the back face
only. The repair material and repair adhesive were identical to the parent material and
adhesive. The ram deflection, load and surface axial strains were recorded during test.
93
4. The static performance of composite sandwich beams
4.1 Introduction
In the present Chapter the behaviour of composite sandwich beams tested statically in
four-point bending is discussed. One face of the beam is representative of the rudder
of an Airbus (i.e. a [0/90, -45/45] lay-up), and is called the tool face (i.e. TF). This is the
face that will carry the repair, if any. The other face is called the bag face (i.e. BF) and
has a [0/9012lay-up to ensure that failure occurs first on the TF.
various width. The beams were tested with the TF in compression and in tension. The
first tests were done on 'as-received' 200mm wide beams. These tests were conducted
in order to firstly validate the tests rig, and secondly to understand the flexural
behaviour of the beams. More tests were then conducted on 100mm and 30mm wide
beams. The finite element method (FE) was then used to predict the flexural behaviour
of the beams. Three approaches, one two-dimensional method and two three-
dimensional methods, are evaluated and the results compared with experimental
results. The calculated stress outputs are then used with the Tsai-Hill criterion to
predict the failure load.
The performance of scarf and overlap repaired beams is then presented. The repairs
to
were subjected compressive or tensile loading. Again a comparison of the results is
presented. Two numerical methods, previously validated on the undamaged beams,
are used to predict the flexural behaviour of the repaired beams. A comparison is
made between the skin surface strains and the adhesive strains. Finally, the flexural
behaviour of the sandwich beams and the performance of the repairs are discussed.
94
4.2 Undamaged beams: experimental studies
4.2.1 Introduction
4.2.2.1 Beams tested with the TF in compression: BAe Airbus' test results
The static performance of ten undamaged beams, tested with the TF in compression,
was investigated by BAe Airbus (UK) as part of a larger experimental programme [12].
The results are shown in Figure 4.1 (ranked from the weakest to the strongest beam).
The ultimate total failure load (i. e. UFL) was between 800ON and 10,OOON.The average
UFL per unit width in compression is 43.8N/mm, for a standard deviation of
2.74N/mm. The coefficient of variation for these results is therefore about 6%. The
beams failed from a compressive failure of the TF, away from the loading points.
10,000 r7
7- 8000 40
M
6000 30
m
N 4000 20
:D 2000 10 ý-
0
456789 10
Specimen Number
95
4.2.2.2 Beams tested with the TF in compression: present results
A single 200mm wide beam (CCND-6-204-6) was tested, in-house, under conditions
similar to the BAe Airbus' beams (as described in section 3.3.5). The beam failed at a
load of 44AN/ mm (i. e. a total load of 8755N). This is only 1% higher than the average
failure load of the BAe Airbus' test results. The beam failed by a compressive failure of
the TF, in the region between the inner loading points, i. e. the region of constant
bending moment, as shown in Figure 4.2. The surface strain at failure, measured at tile
Core
Figure 4.2: Sketch of the failure of beams tested with the TF in compression
A single 200mm wide beam (CCND57) was subsequently tested with the TF in tension.
The beam failed at a load of 50N/mm (e.g. a total load of 10,OOON),by a tensile failure
of the TF, away from the loading points, as shown in Figure 4.3. The surface strain at
failure, at the centre of the TF, was 9000pistrain.
Figure 4.3: Sketch of the failure of beams tested with the TF in tension
Additionally, on this beam, the back-to-back strains (i. e. one gauge was located at the
centre of the TF, the other gauge was located at the centre of the BF) were recorded.
The results are shown are in Figure 4.4 below. The TF was subjected to in-plane tensile
stresses, whereas the BF was subjected to in-plane compressive stresses. The BF strain
96
readings were normalised by multiplying by -1 to appear positive and hence allow for
comparisons to be drawn. It may be seen in Figure 4.4 that the back-to-back strains are
significantly different. This is a consequenceof the unsymmetric lay-up of the beams
and it indicates that the neutral axic,of the beam is shifted away from its centroid [96],
as shown schematically in Figure 4.5. A calculation, assuming that the strain
distribution is linear through the thickness of the beam, shows that the neutral axis is
located some 0.72mm away from the centroid of the beam, towards the face of greater
12,000-
Strain gauge
10,000- TF
Tool face
8000-
It
6000- Bag face
Q)
U 4000-
12
LO
- 2000-
to
0
0 2ý00 40'00 6000 8000 10,000
Load (N)
Core
BF: [0/90]
97
recorded higher TF strains than the beam tested with the TF in tension, e.g. about 15%
difference at failure.
The latter observation is interesting since the assumptions made in section 3.3.2.1
cannot explain this phenomenon. We assumed there that the compressive and tensile
modulus of the materials that comprise the face, i. e. the F914C woven fabric, are equal
in tension and compression. However, in practice the crimping of the fibres almost
certainly guarantees a lower in-plane modulus in compression [91]. However, a lower
TF compressive in-plane modulus will increase the discrepancy between the faces in-
plane moduli, and hence further shift the neutral axis towards the BF. The net result is
a further increase of the TF strains.
12,000
10,000 TF in compression
(CCND6-204-6)
8000-
6000-
S
4000- .
TF in tension
(CCND57)
2000-
0
0 2000 4000 6000 8000 10,000
Load (N)
A calculation may be done using the different tensile and compressive mechanical
properties of the F914C fabric found in the open literature [93]. Although the
properties are different than the ones used in this work, they may be used to show that
the net effect of having different in-plane tensile and compressive facings' moduli
The software LAP (Laminate Analysis Program-me) [101] was used to calculate the
homogenised bending stiffness of the beam with different assumed compressive and
tensile properties. The calculations are based on Equation 3-5, but taking into account
the Poisson's ratio effect [6]. The compressive in-plane modulus of a [0/901 laminate,
98
was found 58GPa [93]. This is about a 15% reduction
to be Ei(compression)ý from the value
of the modulus in tension used in the present work (E, =67GPa, see Table 3-1). The
mechanical properties of the [45/-451 ply were calculated by the software by the
method of matrix transformation [6]. The other elastic constants were assumed
The results of the calculation for the beam with the TF in compression and in tension
are shown in Figure 4.7. The calculated bending stiffness is seen to be slightly lower
for the case of the beam loaded with the TF in compression (i. e. with the TF having a
lower assumed compressive in-plane properties and with the BF having the tensile in-
with tensile in-plane properties and the BF with lower assumed in-plane properties) of
EI=4.4lE6N. mm. The calculation of the associated strain at a load of 43.8N/mrn shows
that the TF strains are -9975ptstrain and +8782ptrain for beams with the TF in
compression and in tension, respectively. Different in-plane warp and weft material
properties, could, as well, be included in the calculations. However, it has clearly been
shown that the material properties of the facings are responsible for the increases ill
the strains on the TF, when the TF is in compression.
E,
O, (),,
-p-ston Core 1101 F914C 951 FjOj, t--,,,
E -mpression
0/901 F[45/
O, I-sioll
45/4,,
Core l, -,, n6700OMPa Core -451
E, =0.28MPa E 0/q,,
:, 'on, -p-sion
E10/90] E EjO/4 compres-n= 5810OMPa E10,90,
2=0.28MPa --pre-on
E go en"si, G 12=0' 07MPa G
12
150MPa F[0/901
10,
V
12ý0 49
. v 12ý0 .03
complex than the uniaxial-only behaviour of the materials that comprise the structure.
For instance, it was mentioned that the F914C material exhibits different in-plane and
bending may then be accurately explained by taking into account these issues, as was
99
shown by Haberle [102] for a CFRP Ian-dnatein flexure and more recently by Daniel et
al. [1031for CFRP-faced honeycomb sandwich beams. These more complicated issues
have not been considered in the current work. The arguments developed so far do
performed by BAe Airbus [12], has shown that the beams failed at an average ultimate
failure load (UFL) of 43.8N/mn-L
Two beams were tested in-house, with the TF in compression and in tension,
variation in UFL is significant, and the scatter in strength makes it difficult to interpret
the result from one test only. Consequently, it may be seen that the static strength of
the beams is the same whether the beams were tested with the TF in compression or in
tension.
Both beams failed in the region between the inner loading points, by a compressive or
a tensile failure of the TF, respectively. Failure could occur anywhere in this region.
However, and most importantly, no excessivethrough-thickness stressesseem to have
been introduced at the loading points by the test rig, since the beams failed well away,
in the central region. Failure under the load-introduction points would not have been
acceptable, since it may have prevented the repaired beams achieving their maximum
potential.
+9000ýLstrainat failure, for the beam with the TF in compression and in tension,
respectively. The failure strains of the Hexcel F914C fabric have been reported in the
open literature, e.g. reference [92,931. For instance, in reference [93] the uniaxial
compressive and tensile failure strains of a [0/90] F914C laminate are given as -
13,020gstrain and +9280gstrain, respectively. It may be seen that the tensile value
100
compares well with the failure strain recorded here, although the strain at failure for
the beam tested with the TF in compression is lower. Several reasons may be
suggested, although the subject will be further discussed in section 4.2.5.3.
Nevertheless, it may be initially concluded from these preliminary tests that it is
possible -,6 characterise the performance of repairs, since the beams failed wen away
from the loading points, and at failure strains close to the failure strains of the material
that comprise the faces.
The average undamaged failure load (UFL) of the 200mm wide undamaged beams in
subject to some scatter, i. e. 1ON/mm on ten tests. Unfortunately, the limited number of
specimens available for testing has prevented the use in this work of any detailed
statistical analyses. One 200mm beam only was tested with the TF in tension
(CCND57), and failed at a UFL of 50N/mrrL This is the value used as a reference UFL
for the beams tested with the TF in tension, although it may be argued that this value
is not statistically acceptable, and may not represent the average strength of the beams
in this loading mode. It is nevertheless the author's view that the UFL of the beams
when tested with the TF in compression should be lower than when tested with the TF
in tension. The reason is that it was shown that the difference in the compressive and
tensile mechanical properties of the facings resulted in an increase in the strain
experienced by the TF, when the TF is loaded in compression. The strain on the TF,
when the TF is loaded in compression, is higher than that if the TF was loaded in
tension. The higher UFL used for the beams tested with the TF in tension (i.e. 50N/mm
instead of 43.8N/mm) may therefore be regarded as a correction factor to account for
the non-symmetric arrangement of the beam. The fatigue testing for the beams tested
with the TF in tension should therefore be done at a higher load to be more realistic of
damage evolution, if a comparison between compressive and tensile regime is to be
made.
101
4.2.3 The performance of undamaged beams
4.2.4 Results
The results of the flexural tests are presented, in Table 4-1 and in Table 4-2, for
undamaged beams tested with the TF in compression and in tension, respectively. The
specimen designation is given for future reference. Failure loads and failure loads
divided by the width of the specimens, are given. The strains at failure, measured at
the centre of the TF (when otherwise stated), are also given.
compression.
102
The ultimate failure loads normalised by the width of the beams, of the test results
presented in Table 4-1 and in Table 4-2, are further shown in Figure 4.8. The average
failure load from the ten BAe Airbus's tests is included. A scatter bar, that ranges from
the lowest to the highest failure loads recorded in reference [121 (see Figure 4.1), is also
shown on the BAe Airbus' result (it is noted that the dashed lines in Figure 4.8
represents the highest and the lowest failure loads of the BAe Airbus' test results).
= Compression = Tension
50
30
20
10
200 100 30
Specimen width (rnm)
Figure 4.8: Ultimate failure load with respect of the width of the specimens
were tested in-house, with the TF in compression and in tension, respectively. With the
TF in compression, a beam failed at a load of 44AN/mm (see Table 4-1), but with the
TF in tension a beam failed at a higher load of 50N/mm (see Table 4-2). Both beams
failed in the region between the inner loading points, by a compressive or a tensile
failure of the TF. The TF failure strains were -10,063pistrain and +9000pstrain for the
A total of six 100mm wide beams were tested, three with the TF in compression and
three with the TF in tension, as shown in Table 4-1 and in Table 4-2. The strength in
103
compressionis seento be slightly higher than in tension,with an averagefailure load
of 47.9N/mm and 47.3N/rrun, respectively,which is contrary to the observations
rriddeon the 200mmwide beams.However, the small number of specimenstested,in
conjunction with the in
scatter strength,prevents any firm conclusionsbeing drawn.
Thus the static strength is again to be taken as independentof the loading mode. The
beamsfailed in the samemanner as the larger, 200mmwide beams,well away from
the loading points, as was shown in Figure 4.2 and Figure 4.3, respectively.
Furthermore,the TF strains at failure were closein value to the failure strains of the
200mmwide beams(seeTable4-1 and Table4-2),i.e. -10,513listrainand +9285ptrain
for the 100mmwide beamswith the TF in compressionand tension,respectively.This
is encouraging since it may mean that the 100mm wide beams would be good
candidatefor the repair and fatigue programmes,allowing a doubling in the number
of specimensavailablefor testingfrom a limited stockof materials.
also failed under a loading point, for both loading modes. The strains at failure were
also lower than expected, being -9087gstrain and +7440ptrain, respectively. The fact
that the narrow beams failed prematurely underneath a load spreader would limit
their use for repair and fatigue testing. The likely reason for this behaviour may be due
to the stress at the load-introduction points, i. e. the 'wedging stress. The existence of a
'wedging stress' is commonly accepted and may lead to significant errors in flexural
strength measurements as discussed in the next section.
were essentially the same, whether the beams were tested with their TF in compression
or in tension. Furthermore, the fracture load may be seen to be independent of the
width of the beams.
104
The narrow, 30mm wide, beams failed, however, underneath a loading point, whether
loaded with the TF in compression or in tension. The local indentation of composite
laminates or sandwich beams under load, and the associated failure modes, is an
active topic of research. The problem was experimentally studied for composite
coupons in four-point bending by, for instance, Binienda et al. [104]. The research was
motivated by the experimental observation that CFRP specimens tested in four-point
bend test were failing as a results of cracks initiating at the loading points. A local
contact analysis was later proposed [105], and it was shown that the maximum stress
location, and hence the likely location of crack initiation, may be located well away
from the central region. Furthermore, the location of the maximum stress was found to
be dependent on the geometry of the specimen, as well as on the laminate fibre angles.
A more recent experimental and numerical work on the geometry of four-point bend
test specimens [106] also highlighted that various load-span/specimen-thickness ratios
and various support-span/load-span ratios, could lead to non-uniform stressesin the
nominally pure bending moment region. From the above work, it may be suggested
that the narrow beams failed prematurely underneath a loading point, when compared
with wider ones that failed in the central region. It may be concluded that the 30mm
wide beam would be unsuitable for repair and fatigue testing.
On the other hand, the 100mm wide beams Performed as well as the original 200mm
wide ones. The locus of failure was away from the loading points, and the respective
failure strains were identical. It was therefore consideredthat testing 100mm wide
beamsis a good alternativeto the original 200mmwide beams.All 200mmwide beams
were subsequently cut in half for repair and fatigue testing, hence doubling the
number of specimensavailable. It should be noted that, since the width of the saw
blade used to cut the beamswas finite (i.e. 2mm), the specimenswere narrower than
the nominal 100mm quoted.
105
4.2.5 Flexural behaviour of the 100mm wide beams
It was discussed in the previous section that, the behaviour of 100mm wide beams was
identical to that of the original 200mm wide beams. The fatigue and repair
programmes will therefore be performed using 100mm wide beams and therefore
results of the 100mm wide beam are presented in this section. The experimental data
will be later used to compare with the results from the predictive models.
The bending stiffness of the beams, tested with the TF in compression and in tension,
are shown as a function of the applied load in Figure 4.9. General comments may first
be made regarding the inverted parabolic shape of the curves. At a low load (i.e.
between ON and 150ON)a region where the bending stiffness rapidly increases is seen.
This may be attributed to errors in the measurement of the deflections, as well as to the
indentation of the rubber pads. At higher loads, the errors in the measurements
become smaller, hence reducing the errors in the calculation of the bending stiffness.
Finally, upon approaching the failure load (i.e. 60% of the loading curve), the traces in
Figure 4.9 are seen to depart again from linearity, which may be due to damage
effectively softening the beams as the failure load is approached.
106
5.OxlO'
4.8x10'
_
46x10'
4.4x10'
Q) 4.2x10' +x
XX
WX
3.8x10'
r- TF in compression (201-4(l))
- 3ý6x106
x
"ý5
+ TF in compression (CCND5-5(l))
3.4x10' - TF in compression (CCND5-5(2))
3.2x 106
3. Oxl 0'
0 1000 2000 3000 4000 5000
Load (N)
5.0x10, -
4,8x 10' -
4.6x10' -
4.4x10'-'
IE
r, 4. Oxl 0' -
3.2x10'
3.0x10'
0 1000 2000 3000 4000 5000
Load (N)
The region between the inner loading points is supposedly subjected to a constant
bending moment, which results in a constant strain field in this region, as mav be seen
in Figure 4.10 for a beam with the TF in compression. One trace (from the gauge
located 70mm from a loading point) may be seen to depart slightly from the rest of the
readings, although at failure, the values are still within 10% of each other.
The deflections of two beams tested with the TF in compression are compared in
Figure 4.11. The deflections were measured using a travelling microscope focused oil
the edge of the beam. The deflection fields for the two beams may be seen to overlap
each other, indicating good reproducibility. The deflections, at a load of 2000N, were
15.5mm at a loading point, and 19mm at the centre of the beams' length.
107
100mm wide beam
-6 ,000 -
Q)
CCND5-5(2)
0
C TF in compression
-8,000-
X -10,000
-12,000 111,,
-,
0 1000 2000 3ý00 4000 5000
Load (N)
Figure 4.10: 100mm wide beams with the TF in compression: TF axial strains
-17.5-
'Z A
-18.0-
Cn 0 Experiment CCND5-5(2)
-18.5-
A Experiment CCND5-5(I)
I
-19.0-
-19.5-
-20.0 10210
ýO
4'0 8'0 1ý0 120
Figure 4.11: 100mm wide beam with the TF in compression: deflection measured
108
4.2.5.3 Flexural behaviour of 100mm wide beams: discussion and conclusions
Bending stiffness
In the present work, the bending stiffness (per unit width) of the beams were
calculated from the measured valued of the ram deflection and it was noted that the
beams appear softer when tested with the TF in compression. This aspect has already
been discussed in section 4.214. It may be recalled that the TF surface strains on a 200
mm wide beams tested with the TF in compression were higher than on a similar beam
but tested with the TF in tension (Figure 4.6). The complex mechanical properties of
the materialsthat comprise- the faces were identified as being responsible for the
dependence of the flexural behaviour of the beams on the loading mode (i.e. TF in
values are identical to the theoretical values calculated in section 4.2.2.4 (i. e. UE6 and
4.4E6N/mm for beams with the TF in compression and tension, respectively).
Some comments may be made on the shape of the bending stiffness curves as the
fracture load is approached. It was seen that, in Figure 4.9, the bending stiffness of the
undamaged beams departs from the plateau region at loads above about 3500N, which
corresponds to a tensile and compressive strain on the TF of about 6000ILstrain(Figure
4.10). Recent work on two-ply [0/90] CFRP 8h-woven (i. e. one tow of fibres passes
over 7 tows and under one as described in section 3.3.2.1) fabric coupons tested in
tension reported a softening of the fabric for applied tensile strains above 5000listrain
(i. e. about half the failure strain in tension) [107]. Transverse matrix cracking, in the
weft tows, and delamination, at the region where the warp and weft tows cross (i.e.
the crimp region), were observed. In addition, a loss of mechanical properties with
applied strains was reported. For example, the elastic modulus and Poisson's ratio
were seen to decreasefor strains higher than 6000gstrain (i.e. a decreaseof about 5%
and 10% at failure, respectively). The approximate threshold strain for significant
cracking was about 5000;xstrain. Also, Manger et al. [108] reported a 6000gstrain
threshold strain for initiation of matrix cracking in transverse tows. These values of
threshold strains compare well with the departure in non-linearity of the bending
stiffness value seen above 350ONin Figure 4.9 (i. e. equivalent to about 6000ptstrainin
Figure 4.10). It is noted that the departure from non-linearity may not be clearly
109
observedon a load versus deflection graph. However, bending stiffnessversus load
graphs, as in Figure 4.9, may be seen to clearly reveal such information.
respectively [109]. Errors due to the transverse sensitivity of a gauge may be larger,
especially in the case of a beam subjected to a pure bending moment. The anticlastic
curvature ensures that different degrees of correction are needed if the gauge is placed
on the concave or convex side [110]. However, modern strain gauges have very low
transverse sensitivities, e.g. 0.3% in the present case, and the error is therefore
negligible. Finally, in practice the strains are not measured at one point, but averaged
over the length of the strain gauge (i. e. 6mm in this study). Thus, errors due to the
wavy nature of woven fabric will exist. The indicated strains depend on whether the
gauge is positioned on a uniaxial fibre bundle or at a warp/weft intersection [95]. No
correction factors were investigated in the present work but it should be noted that a
variability in strain reading exists and it may explain the higher strain readings at
70mm from the loading point shown in Figure 4.10.
110
the inner loading points, away from the loading roller. The testing rig was judged
adequate for the rest of the programme.
The TF compressive failure strain was recorded at about -10,000gstrain. For the beams
tested with their tool face in tension, the tensile failure strain was +9000gstrain. The
failure stiains for the F914Cfabric were found in the literature [93] at +9280ptstrainand
compared with that of the original 200mm wide beams. From Figure 4.8 and taking the
experimental scatter into consideration, the strength of the beams whether tested with
the TF in compression or in tension were found to be not significantly different for any
width. The 100mm wide beams failed in the region between the loading points, i. e. the
region of constant bending moment, and at failure strains identical to the larger
200mm wide beams. However, it was noted that the 30mm wide beams failed under
the loading points, and at lower strains. The indentation of the skin under the roller
was thought to act as a stress concentration, resulting in a lower failure load. A width
of 100mm was subsequently selected for the static and fatigue characterisation of the
beams, allowing a significant increase in the number of specimens tested. There
very good. The bending stiffness of the beams was seen to be lower when tested with
the TF in compression, compared with the TF in tension. Similarly, the TF surface
strains measured on a beam tested with the TF in compression were higher than on a
similar beam but tested with the TF in tension. The complex mechanical properties of
the material that comprises the faces were identified as being responsible for the
dependence of the flexural behaviour on the loading mode (i.e. TF in compression or in
tension).
ill
4.3 Undamaged beams: predictive studies
4.3.1 Introduction
The flexural behaviour of undamaged beams was modelled using the commercially
available finite element (FE) package, ABAQUS. The performance of several numerical
techniques, that incorporate different degrees of idealisation, was assessed.
Simplifying assumptions were formulated for all the models and they include:
" Identical tensile and compressive elastic properties of the materials were assumed.
" The contact areas,at the load-introduction points, were not modelled.
used were presented in Table 3-1. The models were created using the PATRAN pre-
processor and they were run on a DEC workstation.
The calculated deflection and axial strain fields on the TF of the beams are compared
with the experimental results presented in section 4.2.5. The stress outputs, from the
3D models, are then used in conjunction with the Tsai-Hill criterion to predict the
failure load.
The calculated results were output using the ABAQUS post-processor (i.e. a graphic
interface). The stresses and strains were calculated at the integration points, but the
value of stressesin the output file were determined at the nodes by extrapolating the
stressesfrom the integration points.
112
4.3.2 Description of the FE models
constructed. One model for each of the 200mm, 100mm and 30mm wide beams was
created, and run with the TF in compression and in tension, respectively (i.e. six
models were run in total).
Eight-node linear brick elements with incompatible modes (C3D81) were used to
model the skin and the core. As many as 140 elements along the length and 8 elements
across the width were used, as it is important to keep these first-order elements
rectangular in shape. The narrower beams were modelled with only 4 elements across
the width as this was found to be satisfactory. One element per ply for the skins and
two elements across the thickness for the core were used. The materials were modelled
as orthotropic solids. The adhesive between the core and the faces was included and it
was modelled as an isotropic medium. Typical meshes are shown in Figure 4.12.
The boundary condition option was used to model the outer supports as 'line
supports'. One support was rigidly fixed in space (i.e. all nodes on the line of actions
were fixed in translation), whereas only translation along the length of the beam was
allowed on the other support. It is the author's present view that this represents the
bestway of modelling the externalloading points without going into complexcontact
problems,sincethis region of the model is of no interestfor the presentwork.
The load spreaders and the rubber pads were also modelled using eight-node linear
brick elements (C3D8). The Young's modulus and Poisson!s ratio of the steel blocks
were taken as 220GPa and 0.3, and values of 3.5MPa and 0.49 were assumed for the
rubber pads. A prescribed load was applied on the load spreaders. A total of 56473
nodes and 9152 elements were used for the model of the 200mm wide beam.
113
identified. 70 elements along the length and 12 across the width were used. As many
as 40 elements were used in the region between the loading points. The resulting mesh
is shown in Figure 4.13.
as boundary conditions. One support was rigidly fixed in space (i. e. all the nodes on
the line of actions were fixed in translation), whereas only translation along the length
of the beam was allowed on the other support. In addition, all rotations except the
rotation allowing the beam to bend were prevented on one node of the first support.
The loading points were modelled as a uniformly distributed load to an area
equivalent to the area of the rubber pads, as shown in Figure 4.13.
used. The main feature of this model is the plane strain assumptions (i.e. the cross wise
strain is zero) which reduces the analysis to a two-dimensional problem.
The skins were modelled with one element per ply. Eight elements across the core
were used. The adhesive layer between the core and the skins was modelled. 2720
elements were used in total. As many as 160 elements per ply were used in the
constant bending moment area. Details of the mesh and boundary conditions are
shown in Figure 4.14.
The boundary conditions were applied at nodal points; one support was rigidly fixed
(i.e. one node is prevented from translating in all directions), whereas the other
support is only allowed to move along the length of the beam. The loading was
applied to one node located at the centre of the load spreaders. Orthotropic plane
strain properties were employed for the constituent materials.
114
xiY
A nodal translation is
th. ý ... s
brick beams
Figure 4.12: 3D element models of undamaged
ý%k,
beam
Figure 4.13: Quasi-313 shell element model of an undamaged
JYx
115
4.3.3 Stiffness prediction
3D model.
DISPLACED MESH
116
The deflections, measured at an edge of a 100mm wide beam tested with the TF in
compression (e.g. seeFigure 4.11, section 4.2.5),are compared with the results from the
different FE approaches in Figure 4.16. It may be seen that the 3D and the quasi-31)
models lead to almost identical predictions. The 2D plane strain model may be seen to
calculate smaller deflections, i. e. the model is stiffer than the actual beam. The
calculated deflection are nevertheless Predicted within 10% of the measured values.
corresponding numerical predictions is shown in Figure 4.17. The axial surface strains,
at a load of 43.8N/mm, are shown from a load introduction point (Omm) to the centre
of the beams' length (125mm). It may be seen that the brick and the shell models are in
good agreement with the experimental results of the CCND5-5(2) beam, with the
calculated strains being within a few percent of the measured ones. The predictions are
poorer in the caseof the CCND5-5(l) beam, but they are still within 10%.
It may be further seen that the plane strain model under-predicts the experimental
strains and always yields results lower than those of the shell or the brick element
models. The calculated results are lower by about 10% if compared with experimental
results of the CCND5-5(2) beam. Furthermore, the predicted strain distribution is 20%
lower than the CCND5-5(1) readings. This discrepancy is not acceptable and will be
discussed in the next section.
It may be noted that the distribution of stresses along the beams' length for the 3D
model shows an increase in stress close to the loading point. The 'wedging stresses'
(see section 4.2.4.3), at the loading points were not studied in this work. Ignoring the
"wedging stress' it is noted that the axial surface strain distribution is seen to increase
slightly as we approach the centre of the beam (i.e. for the results from the 3D brick
element model in Figure 4.17). The condition for a constant bending moment in the
region between the loading points is nevertheless satisfied for the present structure
and test geometry (seesection 4.2.4.4).
117
-15.0- 100mm wide undamaged beani
. wý TF in compression
-15.5-
-16.0- +
++++.
-16.5 ++++
+'++++++
Jý -17.0 Applied load= 20N/mm
E +++++ý'
. ++,
r- -17.5- o'+*,,
0 A +++++,
+,+
-18.0-
(U
ýM - 18.5- 0 ExperimentCCND5-5(2)
A ExperimentCCND5-5(l)
-19.0- 3D brick elementmodel, Edgeof the panel
Quasi-31D shell elementmodel, Edgeof the panel
-19.5 2D PlaneStrain model
118
4.3.3.3 Stiffness prediction: discussion and conclusions
The first requirement of a model is to be able to predict the stiffness of the structural
component accurately. The calculated deflection and surface strain distributions were
compared with equivalent experimental results. The predictions are discussed in this
section.
Stiffness prediction
The calculated results, from the three different approaches, were seen to be in good
agreement with the experimental results. The predicted deflections were within 10% of
the measured values, and the 3D brick and quasi-31) shell element methods were seen
to give the best predictions. Similar results were seen for the prediction of the axial
strain distribution on the surface of the TF of the beams, when compared with the
CCNDS-5(2) strain results.
may be seen that the CCND5-5(1) trace departs from linearity at about 3000N, whereas
the onset of non-linearity is seen at a higher load for the CCND5-5(2) beam. This
behaviour may be attributed to the damage occurring in the facing materials.
Consequently, all the traces are seen to overlap at low loads, but rapidly depart from
the linear-eIastic prediction as the load is increased. Thus, a model that includes
damage progression in the facing materials would be needed to completely
characterise the behaviour of the beams. The calculated results are nevertheless seen to
be in good overall agreement with the experimental results.
119
12,000- 100mm wide beams
11,000 v TF in compression (CCND5-5(l))
10,000 (Data normalised by -1)
9,000- 0 TF in compression (CCND5-5(2))
(Data normalised by -1)
8,000-
Prediction - 3D brick element model
7,000-
vv v
Q)
-5 6,000-
0 5,000-
4,000-
3,000-
2,000-
1,000
0
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
Load (N)
3D models involve the use of continuum brick elements which allow the 3D deflection
field to be accurately characterised and the corresponding stress and strain fields to be
determined. Consequently, the 3D technique may offer the best idealisation, when
compared with other approaches. The quasi-31D shell idealisation allows the complex
way the beams deform to be accurately modelled. However, the stiffness predictions
from the 2D plane strain model were seen to be smaller than those of the 3D
approaches. This is undoubtedly due to the assumption of zero lateral strain (i. e. plane
strain) in the former model. In practice, lateral strains can take place over the width of
the beams, and the exact solution may fall between the plane strain and the plane
stress (i. e. zero lateral stress) solutions [96]. The calculated deflections were,
nevertheless, in good agreement with the measured values.
120
4.3.4 Failure prediction
4
4.3.4.1 Introduction
An attempt to calculate the failure load of the undamaged beams, with the stress
results from the FE analysis models, is presented in this section. The calculations were
based on the results from the 3D brick element models. Undamaged beams, 30mm,
100mm and 200mm wide, were modeRed with the TF in compression and in tension.
Six models were created in total (i. e. one model of each beam, with the TF in
The axial (cri) and transverse (a2) stresses, calculated at the centre of the TFof the
I
beams, were therefore used with the Tsai-Hill failure criterion to predict structural
failure. For a unidirectional ply, the Tsai-Hill criterion is [6]:
222
ý112
aý al or2 ai + ý: I at faflure Equation 4-1
S2 S2 + S2 Sý
2 1,2
where subscripts 1 and 2 denote the directions along and transverse to the fibres
respectively. The subscript 12 denotes the in-plane shear component. The terms ai and
S, are respectively the resolved stress and the ultimate stress in the principal material
direction 1, etc. Failure of the material occurs when the left-hand side of the equation
reaches unity. It may be seen that it is a quadratic criterion and if used in a linear
analysis, the failure index (i. e. the left-hand side of Equation 4-1) and the associated
failure load are therefore related by a square root factor. The failure loads of the beams
were calculated by multiplying the load applied to the 3D models by the square root of
the failure index. The total load applied to the models was 43.8N/mm.
models. Only the stresseson the TF of the beams are used for the calculation of the
failure load, as this is the face that causesstructural failure of the beams. The stresses
on the other face, i. e. the BF, are much lower. It is A&AA.that the BF is stiffer as it
contains two [0/90] plies, and is hence subjected to lower stresses than the TF. The
axial and transverse stresses,acting at the centre of the TF are presented in Table 4-3
for the TF in compression. The values were taken at the centre of the beams, since it
121
was shown in Figure 4.17that if the 'wedging stress'at the loading points is ignored,
the strains are maximum there.The stresseswere read at an integration point within
an element.The calculatedresultsfor the TF in tension were seento be exactly equal,
but opposite in sign, to those in Table 4-3 and are not presented here. However, they
will be used to calculate the failure load of the beams tested with the TF in tension.
For the beam with the TF in compression (Table 4-3), it may be seen that the axial
stress in the [0/90] and [45/-45] ply of the TF remains more or less constant, regardless
of the width of the beam. The transverse stresses in the [0/901 ply of the TF are,
however, seen to decreasewith 4-tcreasing width. It is also evident that the axial and
transverse stressesare of opposite in sign in the [0/90] ply, but are equal in sign in the
[45/45] ply, as a consequenceof stressand moment equilibrium through the thickness
of the bean-L
Table 4-3: Computed axial and transverse stresses in the TF (TF in compression)
calculated failure loads are compared with the experimental results (from Figure 4.8) in
4.19, for beamstestedwith the TF in compressionor in tension.It may be seenthat the
failure loadsare underestimatedin all cases.The [45/-45] ply is predictedto fail first, at about
30N/mm,for both loading modes.'Me [0/90] ply is then predictedto fail, at the higher load of
35NImm-The experimentalfailure loadswere all above40N/mm. It is noted that the failure
ýloads for first-ply failure, shown in Figure 4.19, were calculated using a laminate
arialysis approach. This approach calculates a first-pIy failure load for all the plies that
comprise the laminate. The calculated failure loads of the [45/-451 and the [0/90] ply
are therefore the first-pIy failure loads of each independent ply. The lowest value of
the first-ply failure loads only is relevant (i.e. the first-ply failure load of the [45/-451
ply) since progressive damage evolution is not being taken into account in the present
analysis.
122
TF in compression
50-
E]30nin,
40
E
E
30
0 20
2
10-
0
Pmdictod JO/W1 II Pred., t. JI
TF in tension
50- I- C]30..,
40
E
E
30-
20-
10-
01 --
j '' ý i., 1""d" t,
more complex analysis than could not be undertaken in the time available to complete
the programme.
Failure predictions
The predicted failure loads were in relatively poor agreement with the experimental
observations for all the beams. We may first question the validity of the strength
values used in this work. The values used in the Tsai-Hill equation, were
communicated by BAe Airbus [121. For instance, the tensile strength of tile F914C
fabric, with a [0/90] orientation, was quoted to be Si=553MPa hi Table 3-2. But values
of 539MPa and 572MPa have been quoted in references [92] and [93], respectivek.
123
Now, the calculated failure loads for the 100mm beam with the TF in tension, and with
Table 4-2). Thus, it may be seen that the difference in the inputted strength properties
do not explain the discrepancies between the experimental and predicted values of
failure loads of the beams.
macroscopic, failure criterion used. The Tsai-Hill criterion calculates a load for first ply
failure, and does not take into account the effect of microscopic damage. However,
damage development in composite laminates is complicated, and can be thought of as
approaching the failure load, was experimentally observed in Figure 4.9. The evolution
of the bending stiffness of a 100mm wide undamaged beam tested with the TF in
compression is reproduced below in Figure 4.20 for clarity. It is seen in Figure 4.20 that
the curve departs from the plateau at loads of about 3500N, which corresponds to a
compressive strain on the TF of about -6000jAstrain (Figure 4.10).
4.4006
100mm wide undamaged beam
oýe
4.30011 0. D
.........CCD
ocb
- 00 00
DO 0 GD
00 cAýlb
a) 4.2x 10" -
IE TF in compr6sion (201-4(l))
U)
4. lxlOl
4. Ox10' 4-
2000 3000 4000 5000
Load (N)
124
Damage development in woven composites is complicated, due the wavy nature of the
fabric [107]. Modelling the stiffness behaviour caused by these damage modes and
recently been developed for laminate other than cross-ply (i.e. 0' and 90* ply
orientation), although they are limited to laminates fabricated from unidirectional
layers. Again no attempts were made at predicting laminate failure.
The FE method may also be used in conjunction with a classical interactive criterion to
simulate damage evolution and ultimately predict the failure load of structures. The
analyses generally use shell elements in combination with a non-linear analysis [113].
At each load step, an analysis is performed until a converged solution is obtained,
assuming no changes in the material model. Then, using this equilibrium state, the
stresseswithin each lamina are determined from the analysis solution. These stresses
are then used to determine failure according to certain failure criteria. If lamina failure
is detected, the lamina properties are changed according to a particular degradation
" residual properties methods (based on physical microscopic model, e.g. shear-lag).
An equilibrium of the structure is then re-established utilising the modified lamina
properties for the failed laminae while maintaining the current load level. This
iterative process is used for each load step until catastrophic failure of the structure is
detected. This is a method that needs to be used in the present work if failure is to be
accurately predicted. The method has been successfully employed to predict failure in
notched composites in tension [1141 or compression [1151. Unlike the analytical
method described previously this method may be use to predict the strength of
structural elements such as stiffened panels in compression [113] and plates subjected
to transverse pressure [116].
125
Conclusions
The prediction of the failure loads of the beams has not been possible due to the
complex way CFRP materials fail on one hand, and due to the limitation of the failure
criterion used on the other. However, it is noted that it is possible to predict the failure
load of the beams if the predicted surface strains (as in Figure 4.17) are compared with
the material failure strains [93]. Based on this remark, a method involving the use of
small coupons, that would mimic the way the sandwich beams failed, will be
presented in Chapter 6.
Stiffness predictions
Good correlation between the calculated and measured stiffness properties were
observed. The 3D brick element model gave the most accurate predictions and the
predicted deflection and axial surface strains were in good agreement with the
experimentally measured values. The axial surface strain distribution on the TF of the
beams was seen to be constant in the region between the loading points, which in turn
indicates a constant bending moment does indeed exist in this region. The quasi-3D
shell element model was seen to lead to results that are comparable to the results of the
3D brick element model, although the stresses at the loading points cannot be
predicted by the former approach. The 2D plane strain model, was seen to predict a
higher stiffness than the 3D approaches. The deflection and surface strains were
under- and over-predicted by about 10%. This difference was identified as being due
to the assumption of zero lateral strains, inherent to plane strain idealisation. Beams
subjected to bending forces deform in a complex manner and Poisson's ratio effect
induces a cross-wise bending that produces the well known anticlastic curvature.
Idealisations; that assume zero strain, as in the present case,in the cross-wise direction,
cannot therefore lead to precise predictions. The overall effect is that the beam is too
stiff. The results computed from the 2D plane strain approach were, however, shown
to give good correspondence with experimental results.
Failure predictions
An attempt to calculate the structural failure load of the undamaged beams was made
using the stress outputs from the 3D brick element model. The axial and transverse
stresses,at the centre of the TF of the beams, were read at the integration points and
126
were used in conjunction with the Tsai-Hill failure criterion. The calculated failure
loads, for failure in the [45/-45] or the [0/90] ply, respectively, were seen to under-
predict the structural failure load by more than 10%. It was suggested that this
discrepancy arose becausecomposite materials fail by progressive damage, rather than
by sudden failure. The prediction of the progressive damage in CFRP materials and
structures is a subject of active research. A world-wide failure exercise, with the airn of
establishing the current status of theories for predicting the strength of CFRP
materials, has been initiated recently [117]. Twelve different theories were compared
and they gave a wide spread of predictions of initial and final failure stresses and
deformations. Final recommendations on the use of the failure criteria have not yet
been released. An alternative experimental approach, using small, inexpensive,
coupons will be used in a later section to predict structural failure (seeChapter 6).
Type of models
The 3D models lead to the more precise predictions. However, brick elements are
computationally expensive. The brick elements used in the present work are linear and
contain 8 nodes, i. e. 24 degrees of freedom. Although they are a good alternative to the
20-noded, 60 degrees of freedom, quadratic brick elements, they are still expensive and
impractical to run on a day-to-day design exercise.The two-dimensional and the quasi
three-dimensional models were seen as good alternative approaches.
The quasi-31) model was simple to set up and was best suited to predict the overall
flexural stiffness of the beams. Shell elements offer a better cost-to-performance ratio.
Very few papers have applied the method described above to the study of repairs to
composite structures [87] and the technique will be evaluated to model the flexural
behaviour of the repaired sandwich beams in the following sections.
Despite the fact that the 2D plane strain model was seen to predict a higher stiffness
than the 31) approaches, it was shown adequate to obtain reliable results. The main
feature of the 21) plane strain models is their ability to describe a planar geometry
accurately without increasing much the computational costs. Adams et al. [47] used
the FE technique to calculate the stress distribution of bonded lap joints in tension. The
method will be in
used section 4.5 to model the flexural behaviour of the repaired
beamsand the resultswill be comparedwith thosefrom the shell elementsapproach.
127
4.4 Repaired beams: experimental studies
4.4.1 Introduction
The repairs simulated a repair to a 60mm long 'delan-dnation damage' area on the TF
of the beams. It was shown in the literature survey that damage of such extent would
reduce the bending strength of the beams by 50% [12]. It may be recalled that three
repair schemes were evaluated: 1/30 scarf, 1/10 scarf or 2-ply overlap patches that
extend across the width of the beams (100mm wide). The repaired beams were tested
in four-point bending, with the repaired TF either in compression or in tension. The
overlap and scarf patches were made of two plies that are identical in orientation to
the removed material. The tool face parent lay-up and overlap patch lay-up is [0/90,
45/-45]. An additional overlapping ply was used on the scarf patch to protect the tip of
the scarf from damage [851.The length of the overlapping ply was 15mm for all three
patches. The repairs were vacuum cured to be representative of typical aerospace
practice.
Two repair systems were evaluated. A high-temperature cure (HTC) system identical
to the parent materials was first used. The repair material, supplied by 'Hexcel', was a
5-harness F914C prepreg and the repair adhesive used was 'Redux 319' epoxy film
adhesive. Both were cured at 175'C for 1-hour. A 4-hour post cure at 190'C was also
performed to obtain a repair material with optimum material properties. The low-
temperature cure (LTC) prepreg repair system which was supplied by 'The Advanced
Composite Group' (ACG) comprises a 5-harness T300 woven fabric with LTM25EL
resin. The repair adhesive was an ACG XLTA225 epoxy film adhesive. The LTC repair
system was cured for 4 hours at 100*C.
In this section, the performance of the repaired beams is compared with that of
undamaged beams. It is noted that the failure load of the repaired beams (i.e. 100mm
in width) are compared with the undamaged failure load (UFL) of 200mm wide
beams. The reasons being that the UFL of the beams was seen to be the same for all
width (Figure 4.8), and that the UFL of the 200mm wide beams was the more statically
accurate, as opposed to the UFL of the 100mm wide beams (section 4.2.2.4). It was
therefore decided that the former value be used as a reference UFL.
128
4.4.2 The performance of repaired beams
Results are presented, in Table 4-4 and in Figure 4.21, for the repaired beams tested
with their TF in compression. The dashed line, in Figure 4.21 below, represents the
average ultimate failure load (UFL) of the 200mm wide undamaged beams tested with
the TF in compression, i. e. 43.8N/mm.
Table 4-4: Tests results for 100mm repaired beams tested statically
point. The strains were measuredoutside the repair patch, at the centre of the beam (farfield strams).
TF in compression
11TCHigh-temperaturt, curt
6000 UFL I. TC: Lm-temperalun, 60
(wre
(200mm)
5000 50
.............. .............. A,
4000 40
3000 30
Q)
2000 20
1000 10
0 0
Undamaged 1/30 scarf 1/30 scarf 1/10 scarf 2-plY ovvrlap 2-ply overlap
(10011ull wide) I FIC 1, FC IFIC HIC 1,1 (-
129
In Figure 4.21 and Table 4-4 it may be seen that all repairs recovered more than 90% of
the undamaged failure strength, with failure loads above about 40N/mn-L
Furthermore, it may be seen that the repair system, whether high-temperature cure or
low-temperature cure, did not have any effect on the static strength.
It may be seen that the 1/30 scarf repaired beams did not perform as wen as the
overlap repaired beams, with failure loads below the undamaged UFL of 43.8N/mm.
On the other hand, all the overlap repaired beams failed at loads higher than the UFL.
These results go against the popular belief that scarf repairs are always stronger than
overlap repairs [431.However, the few specimens tested in the present work prevent
the use of statistical tools, making it difficult to draw any firm conclusions.
From Table 44 it may be seen that the 1/30 scarf repaired beams, whether fabricated
repairs. The HTC scarf repaired beams failed outside the repair patch, in the region
between the loading points as shown in Figure 4.22, whereas the LTC scarf repair
beam failed at the overlapping ply/ scarf junction, as shown in Figure 4.23.
The 1/10 HTC scarf repaired beam performed slightly better than the 1/30 scarf
repaired beams. The failure strength was above the control strength (i.e. 43.8N/mm),
and the repair therefore recovered the control strength. The beams failed outside the
repair patch, as shown in Figure 4.22.
Overlap repaired beams, whether with HTC or LTC patches, performed very well and
failed above the failure load measured from the undamaged beam, therefore
recovering more than 100% in strength. The failure loci were however different for the
two repairs. The HTC repair failed at the end of the overlap as shown in Figure 4.24,
whereas the LTC repair failed outside the repairs, as in Figure 4.22.
130
Cmli p". ", ,It ho vu, Adhos"'.
Figure 4.22: Compressive failure outside the repair (an overlap repair is shown)
R, , Adhm,
Pamnt Skin ý:-Tau
Figure 4.23: Compressive failure inside the repair (a scarf repair is shown)
(see also Figure A. 2, Appendix A)
Compressive failure
at the varent/overlap
The far field strains (i. e. strains in the parent, away from the repair patch) at failure,
shown in Table 4-4, may be seen to vary from beam to beam, and are generally lower
than the strains at failure recorded on the undamaged beams (Table 4-1 and Table 4-4).
It is noted that the far field strains are only an indication of the strain on the surface of
the beams, away from the locus of failure. A low far field strain at failure may be due
to the somewhat lower failure load of a repaired beam, or due to the stiffening effect of
131
4.4.2.2 Repair intension
The performance of repaired beams tested with the TF in tension is shown in Table 4-5
and in Figure 4.25. Again, the dashed line in Figure 4.25 represents the average
ultimate failure load (UFL) of the 200mm wide undamaged beams tested with the TF
in tension, i. e. 50N/mm (see section 4.2.2.5).
Table 4-5: Tests results for 100mm repaired beams tested statically
point. The strains Weremeasured outside the repair patch, at the centre of the beam (farfield strains).
TF in tension
HIC HiSh-temp-nitur, urc
UFL I. I C: 10711-h-111POWWT Ool' 60
(200mni)
... ............................. 50
40
C-
30
20
10
0
1/30 scarf 1/10 scarf 2-ply overlap 2-ply overlap
LTC IF YC I ITC LIC
132
Again, most repaired beams performed well and the repairs restored more than 90% of
the control strength (i.e. the 200mm undamaged UFL in tension: 50N/mm). An
exception was the 2-ply HTC overlap repaired beam that failed at a much lower load,
and in the repair, recovering only about 70% of the control strength.
The scarf repaired beams may now be seen to perform as well if not better than the
overlap repaired beams. The 1/30 HTC and LTC scarf repaired beams failed at a load
of 45.3 and 49.5N/mm therefore recovering 95% and 100% of the control strength,
respectively. The beams failed outside the repair area, by a tensile failure of the parent,
as shown in Figure 4.26. The 1/10 HTC scarf repaired beam performed as well as the
1/30 scarf repaired beams and recovered 100%strength.
On the other hand, the overlap repaired beams did not perform as well, with failure
loads 27% and 10% below the control strength, for the HTC and the LTC repaired
beams, respectively. The 2-ply LTC overlap repaired beam recovered 90% of tile
control strength and failed outside the repair. However, the 2-ply HTC overlap repair
failed at the parent/ repair patch interface, as shown in Figure 4.27, and only recovered
73% of the control strength. A possible reason may be the increase in strain resulting
from the stress transfer at the end of the overlap [47]. This theme will be discussed next
in section 4.5.3.8.
Figure 4.26: Tensile failure outside the repair (an overlap repair is shown)
(see also Figure A. 5, Appendix A)
133
I ension failure at the
R Adl
parent/ repdir interface
Figure 4.27: Tensile failure at the parent/repair interface (an overlap repair is shown)
compared with that of an undamaged beam in Figure 4.28. It is noted that the
undamaged beam used for the comparison of the results is 100mm in width, since a
200n-Lm wide beam would invariably be stiffer due to cross-wise bending nionielit
arising from Poison's contraction, and hence make the results not comparable.
It may be seen that the repaired beams are stiffer than an equivalent undamaged
beam. The bending stiffness of the repaired beams lies between 4.4E6-4.6E6Nmm,
while that of an undamaged beam is 4.3E6Nmm. The repaired beam may be expected
to be stiffer than the undamaged one, since the overlapping plies make tile beams
locally thicker and stiffer than the parent material.
Similar results may be seen for the beams tested with the repairs in tension ill Figure
4.29. The repaired beams are generally stiffer than an equivalent undamaged beam
(4.7E6Nmrn). However, the bending stiffness of the 1/30 scarf repaired beams are now
relatively low and similar to that of the undamaged beam, at about 4AE6Nmm. Again,
the repaired beams may be expected to be stiffer than the undamaged beam. 1-lowever,
it is not clear as to why the behaviour of the scarf repaired beams is different with the
134
5. OxlO6
TF in compression
06
4.8xl
4.6xl 06
4.,15F6N.nin
4.4xlO6 k 4.3E6N. mm
06 +
4.2xl
06
4. Oxl
06
Undamaged 100mm wide bearn (201-4(l))
E 3.8xl 1/30 scarf HTC (204-4(2))
0
06
Ip 3.6xl 0 1/30 scarf LTC (CCND5-4(2))
- 1/10 scarf HTC (203-5(2))
3.4xlO6
+ 2-ply overlap HTC (CCND67(2))
3.2xl 06 16 2-ply overlap LTC (CCND5-2(l))
3. OxlO"
0 1000 2000 3000 4000 5000 6DOO
Load (N)
FigUre 4.28: Repaired beams tested with the TF in compression: bending stiffness
5. OX106 TF in tension
3. Ox 10 6 - III -r --
I
0 1000 2000 3000 4000 5000
Load (N)
Figure 4.29: Repaired beams tested with the TF in tension: bending stiffness
135
4.4.2.4 The performance of repaired beams: discussion and conclusions
The strength of the repaired beams
Figure 4.30 shows the failure loci and loads for the repaired beams tested with tile TF
in compression and tension. Considering the results, then only a few specimens were
tested statically which makes the discussion difficult, since the strength data may not
be entirely representative. However, it may be concluded that most repairs were
successful. It is recalled that the present repairs were designed following the design
rules described in the open literature [118]. The strength restored was in excessof 90%
of the undamaged strength in most cases and many repairs restored more than 100%,
whether the repairs were placed on the face in compression or in tension. It may be
noted that the repair system, being either high- or low-temperature cure, did I-lot have
Approximate
1/30 scarf 4530N
400ON .5 position of tht-
HTC
loading point
I
2TIN
360ON
5190N
2-ply WOO\
475ON overlap
LTC
The lowest failure load was recorded for the 2-ply HTC overlap repaired beam tested
with the repaired TF in tension, and the repair recovered only 73% of the strength.
However, it may be noted that 73% of the undamaged failure load is still well above
the operational load that would be seen by the beam in service, where the straiiis are
136
typically limited to ±4000gstrain. This is equivalent to about 200ON (i.e. as shown in
Figure 4.10), whereas the beams failed at loads of about 4500-5000N. It is further seen
in Table 4-4 and in Table 4-5 that the far field failure strains of the repaired beams were
for the beams tested with the repair in compression and in
well above ±90OOpLstrain,
tension, respectively. The repair schemes used in the present study therefore
successfully restored the full load-capability of the structure.
method. Secondly, most repaired beams recovered more than 90% of their respective
control strength, with the beams failing outside the repair patch. Only the 1/30 LTC
scarf repaired beam tested with the repair in compression failed in the repair.
It is of interest to compare the loci of failure of HTC and LTC repair systems. It may be
seen in Figure 4.30 that the repair system has undoubtedly had an effect on the failure
mode of the 2-ply overlap repairs. The HTC overlap repaired beam failed at the
parent/overlap interface, whereas the LTC overlap repaired beam failed in the parent.
Furthermore, the repair system may also be seen to have an effect on the loci of failure
of the 1/30 scarf repaired beams tested with the TF in compression. The 1/30 HTC
scarf repaired beams failed in the parent, whereas the 1/30 LTC scarf repairs failed at
the overlapping ply/scarf junction. Although the repair system, being either high- or
low-temperature cure, did not have any major effect on the static strength of the
repairs, it may be seen that it had an effect on the locus of failure of the repaired
beams. The reason is unclear, but it may be thought that the HTC and LTC repair
material have different material and geometric characteristics that may have an effect
on the repair strength and this theme is discussed next.
Repair systems
The end of the overlapping ply of the HTC and the LTC repair systems, are shown in
Figure 4.31 and in Figure 4.32, respectively. The adhesive fillet may be seen in each
picture. It may be seen that the fillet is about the thickness of the overlap and extends
for about 1.5mrn from the end of the overlap. The parent TF layer and the overlap
patch are also shown. However, whereas the adhesive layer can be clearly seen in
137
Figure 4.31, it has blended between the TF and the overlap patch in Figure 4.32. It may
be seen that the LTC adhesive layer is non-existent, and this may have an influence of
Figure 4.31: A high-temperature cure repair (the end of the overlap is shovvii)
Figure 4.32: A low-temperature cure repair (the end of the overlap is shown)
Thus, it may be that the different properties of the LTC repair system (i. e. materials
repairs, when compared with HTC repairs, and this may explain the differences iii the
failure modes of the HTC and LTC repairs seen in Figure 4.30. For instance, it was
shown in the literature survey (section 2.4.2) that the adhesive shear stress increases as
the adhesive thickness is decreased. The lower adhesive thickness of the LTC repairs
may therefore explain the failure in the repair of the 1/30 LTC scarf repaired beam.
the loading mode, i. e. the repair loaded in compression or in tension. For instance,
when tested with the TF in compression, overlap repaired beams were found to be
stronger than scarf repaired beams. Indeed, the scarf repaired beams were found to
recover only 90% of the control strength, in comparison with the full, original,
unrepaired strength which was recovered by the overlap repaired beams. Since, the
beams mostly failed outside the repair patch, it is uncertain if the lower failure loads
138
are caused by the presence of the repair or if it is just an artefact of the scatter in
(i.
strength e. the basic beams are weaker, not the scarf repairs). However, it may be
noted that the failure load, of the beams tested with the TF in compression, increases
as the scarf angle increases,and therefore as the repair area decreases.
Conversely, the scarf repaired beams were found to be slightly stronger in tension.
Scarf repairs are usually found to be stronger thaAequivalent lap repairs [431,since the
scarf results in a smoother load transfer when compared with the abrupt, step-like,
change in geometry introduced by the lap repairs on one hand, and reduces the
eccentricity in the load path in the repair on the other.
Thus, the superior strength of the lap repairs compared with the scarf repairs for the
beams tested with the repaired TF in compression is not entirely clear. A difference in
the behaviour of the scarf repairs in compression and in tension was further observed
in Figure 4.28 and Figure 4.29, as discussed next.
with that of equivalent undamaged beams. It was found that the repaired beams,
The overlap repaired beams were all seen stiffer than equivalent undamaged beams,
whether tested with the TF in compression or in tension. The overlap patches are made
of two plies, with an orientation identical to the plies removed on the TF. The excessof
material in the overlap, which transfers the load from the parent to the patch, and the
eccentricity of the patch from the parent surface, contribute to the increase in stiffness
of the beams, whether the repairs are loaded in tension or in compression.
The scarf repaired beams Were, however, seen to be stiffer or have the same value of
stiffness as their corresponding undamaged beams. Indeed, the behaviour of the 1/30
scarf repaired beams was seen to be different whether the repaired TF was in
compression or in tension (see Figure 4.28 and Figure 4.29). Again, it may be seen that
139
the behaviour of the scarf repaired beams is different whether the TF is in compression
or tension. The reason is unclear as it may be thought that the effect of the repair patch
on the stiffness of the beam should be the same, whether the beam is loaded with the
TF in compression or in tension.
It may, however, be observed that the repair patches are located on the same side as
the inner loading points when the TF is in compression. They are, however, on tile
opposite side when the beams are tested with the TF in tension, as shown in Figure
4.33. The edge of the 1/30 scarf, 1/10 scarf and 2-ply overlap repairs are located 62mm,
74.3mm, and 80mm, respectively, from a loading point. It may be possible that tile
proximity of the load spreaders and the repair patches influences the flexural
behaviour of the beam, when the TF is in compression. It is not possible at the present
time to confirm this suggestion. It may also be noted that the anticlastic curvatures are
in the opposite direction whether the repair is in compression or in tension. A study of
the interaction between the patch and the load spreaders would be complicated and it
would involve doing 3D non-linear FE analyses of repaired beams, and include the
contact area between the beam and the spreaders. This has not been attempted in tile
present work.
ner
loading pointý
Repair in tension
I/ ýo ,rI w1mir 1/ 10 Sca If lel"I II 2-ply overlap icli. m
140
Conclusions
From the results, and on the strength of the above discussions, it may be concluded
that in the present work the scarf repairs were weaker than the overlap repairs when
tested with the repaired TF in compression. However, for the repaired TF loaded in
tension, the scarf repairs were found to be stronger than the overlap repairs.
The static behaviour of scarf and overlap repaired beams was studied. The beams were
tested with the repaired TF loaded in compression or in tension. The testing method
was found to be a success with the repaired beams failing away from the loading
points. The failure load of most repaired beams was above 90% of the undamaged
failure load. The far field failure strains of the repaired beams was over ±9000ýtstrain
for aH beams, and therefore weU above the beams' design ultimate of ±4000;istrain.
The performance of the repairs was, however, seen to be dependent on the loading
mode. Tested in compression, scarf repairs were found to be weaker than overlap
repairs. On the contrary, tested in tension, scarf repairs were found to be stronger than
overlap repairs. On one other hand, it was found that the repair system being either
high-temperature cure or low-temperature cure did not have an effect on the static
strength of the repairs, but it was found to have an effect on the failure mode on the
other. The difference in the geometric and materials properties of the repair systems
was thought to lead to different stress concentration in the vicinity of the repairs, and
so explains these observations. Furthermore, the bending stiffness of the scarf repaired
beams was different whether the scarf repaired TF was loaded in compression or in
tension. With the repair loaded in compression, scarf repaired beams were seen to be
stiffer than an undamaged beam. However, with the repair loaded in tension, the scarf
repaired beams were found to have the same value of stiffness as the undamaged
beam. It was considered that the loading pads may interact with the scarf repairs when
the repairs are in compression, since the repairs are located on the same side as the
loading points when the TF is in compression, and are located on the opposite side
when the TF is in tension. It was however not possible to confirm this suggestion.
141
4.5 Repaired beams: predictive studies
4.5.1 Introduction
The FE method is used in this section to predict the stiffness of the repaired beams. The
load, ram deflection and axial TF surface strains recorded experimentally on the
repaired beams (see section 4.4) are compared with the predicted results. The
commercially available, PC-run and windows-based, LUSAS (FEA Ltd. ) FE package is
now used, instead of the ABAQUS package used in section 4.3. The only reason was
that modelling support was offered by FEA Ltd. The software incorporates a graphics
pre-processor that allows the efficient definition of the geometry and the mesh. A
graphics post-processor may also be used for the reduction of the results. The two
modelling approaches that were validated on the undamaged beams in section 4.3, are
now used to model the repaired beams.
The quasi-3D shell elements method was seen as a good alternative to the more
complicated 3D method in section 4.3 and it is used here to model a 1/30 HTC scarf
repaired bearn. The analysis was linear elastic. The computed surface strains and
deflection field of the 1/30HTC scarf repaired beam are compared with equivalent
experimentalresults.
The 2D plane strain method was also seen as a good method to study the behaviour of
the beams. Plane strain models of the scarf and overlap repaired beams, with the TF
loaded in compression and in tension, were constructed. The CFRP materials were
assumed to be linear elastic in these analyses. However, the repair adhesive was
assumed to behave in a linear elastic/plastic manner. The computed axial surface
strain distributions are compared with experiments. Since, this modelling approach
allows the accurate definition of the geometry and the models included the repair
adhesive and the repair patches, the distribution of the adhesive shear strain in the
different type of repairs is compared and the results discussed.
The elastic properties of the parent materialswere presentedin Table 3-1. The high-
temperaturecure repairs only were modelled in this work, and the compositethat
comprisesthe repair patch was assumedto have properties identical to the parent
materials.
142
4.5.2 Description of the FE models
method described in reference [871. The model is essentially identical to the shell
model of the undamaged beam presented in section 4.3.2.2. However, the scarf repair
was modelled by changing the ply properties of the elements along the beam's length
as shown in Figure 4.34. General purpose, eight-node, thick shell elements (QTS8)
were used. The mesh was identical to the one used in the modelling of the undamaged
beams, i. e. 70 elements along the length and 12 across the width were used. Composite
elements, that identify ply orientation and orthotropic properties, were defined. The
material properties were assigned to the elements as showii in Figure 4.34. The 1/30
scarf resembles a series of overlap joints. An appropriate number of 'dummy' layers
are added in the parent zone to make up the maximum number of layers; the numbers
of layers staying the same across the entire model. The 'dummy' material layer had
very low stiffness properties so that they would not influence the behaviour of the
structure. The load and boundary conditions were identical to the model of tile
undamaged beam, i. e. on one support the beam is rigidly fixed, but allowed to bend
horizontally. On the other support the beam is allowed to translate along its leiigth
and is allowed to rotate in the plane of the beam only.
Zone I
x
Zone 2
Zone 3
Parent -----
CORE
LU/9()i
[0/901
Figure 4.34: Modified quasi-3D shell element model: 1/30 1ITC scarf repair
143
4.5.2.2 2-D plane strain model
Scarf and overlap repaired beams were modelled. Due to the symmetry of the
problem, and in order to save on computer resources, only one half of the beam was
elements (QPN8), with a few six-node triangular elements used at the adhesive conler
and along the scarf length (i. e. for the model of scarf repaired beams). The mesh
density was finer than previously used on the model of the undamaged beams because
of the various singularities present in the repair geometry. Two elements per ply were
used to model the facing materials and the repair patch. Four rows of elements were
used through the thickness of the core. In the repair patch, the adhesive was modelled
with three rows of elements.
The mesh used for the model of the 1/30 scarf repaired beam is shown in Figure 4.35.
A total of 177 elements was used along the length of the beam. The total nulliber of
elements used was 2367. The mesh used for the model of the 1/ 10 scarf repaired bea iii
is not shown here as it is identical in principle to the one shown in Figure 4.35.0111ý
144
The mesh used for the model of the 2-ply overlap repaired beam is identical to the
mesh used for the scarf repaired beams, although on this model there is an uncertainty
regarding the area where the parent and the repair meet (Figure 4.36). The parent aild
the repair patch are separated in this region by a pocket of adhesive. The distalice
between the faces was arbitrarily chosen at 0.5mm, as was done in reference [47].
Pocket of adhesive
5mm ý
.
Core
The boundary conditions, for all models, are as shown in Figure 4.35. A symmetry line
is used on one side of the model. On the other side, the external support is idealised by
a node that is only allowed to move along the length of the bearn. The loading was
applied to a nodal point located at the centre of the load spreaders.
A linear elastic/plastic model was used for the adhesive. A modified von Mises yield
criterion was chosen. This criterion is based on the classic von Mises yield criterioll bLIt
it allows the definition of the different uniaxial yield stresses, in tension and ill
compression, that are shown by polymeric materials. The values of 3.78GPa alld 0.4
were used for the Young's modulus and the Poisson's ratio, respectively (Table 3-1). A
value of 34.7MPa was used for the uniaxial yield stress in tension [651. For the epoxy-
film adhesive used in this study, the ratio of yield stress in compression to tensiorl was
assumed to be equal to 1.3 [65]; A value of 45.1MPa was therefore adopted for tile
uniaxial yield stress in compression.
145
4.5.3 Finite element models of the repaired beams: results
4.5.3.1 Introduction
The results from the FE models are compared below with experimental data that were
edge of the beams and the strain distributions were determined via strain gauges
placed at the centre of the TF.
may be seen that the general trends from both the 2D and quasi-3D models are in
agreement with the experimental data. As expected from the results of tile modelling
of the undamaged beams, the shell model leads to more accurate predictions than that
of the 2D plane strain model. The plane strain assumption leads to deflection that are
smaller, i. e. the model is stiffer. The predictions, are however, still reasonable and tile
difference is only about 10%.
-17.0-
17.5-
M Experiment (204-4(2))
x 2D plane strain model
-18.5- Quasi-3D shell element model
-19.0 ýO ýO ,
0 4'0 8'0 1ý0 120
146
A comparison between experimentally measured and calculated surface strains, on the
TF of a 1/30 scarf repaired beam in compression, is shown in Figure 4.38. It may be
seen that the shell model leads to predictions than are in better agreement with the
experiment. Again, the 2D model leads to predictions 10% below that of the shell
model, as it was seen in the modelling of the behaviour of the undamaged beams
(section 4.3).
Comments may however be made regarding the shape of the strain distribution, as a
variation may be seen over the beamýshalf-length. The two first plies of the patch have
the same orientation as the parent material (i.e. [45/-45], [0/90]), which should
guarantee, because of the low scarf angle used, a smooth load transfer between the
parent and the repair. However, the repair patch contains an overlapping ply (of [45/-
45] orientation) making the patch one ply thicker than the parent. This has the effect of
stiffening the repair area, with the strain in the repair (i. e. between 60 to 125mm from
the loading point) being lower than that in the parent (which is 0 to 60mm from the
loading point).
From the 2D plane strain model, a strain concentration may be seen at 60mm from the
loading point. The increase in strain coincides with the start of the overlapping ply,
and it is located right underneath the adhesive fillet (i.e. point A in Figure 4.38). In this
region the load is transferred from the parent, through the adhesive, to the repair patch
(point B). The resulting shear lag generates a direct stress concentration in the parent,
in the vicinity of the edge of the overlapping ply, as was shown by Adams et al. [119].
In the present study, the increase in strain is 1.35 times the far field strain in the parent.
A value of 1.3 times the far field strain was quoted in reference [119] for CFRP double-
lap joints in tension, the value being taken at the surface of the central adherend.
147
125 mm
10
:
80mm
1
70 mm
-111
25
,1
-->
Core Adhesive
-9,000
-8,000 Quasi-3D model
F-
-7,000
-6,000
Applied load= 40N/mrn /' .......
-5,000
. 21) Plane strain model
.
Z: 0 Experiment (2044(2)) x,
-4,000 xx
x
x
-3,000 xx
-2,000 nB
-x
-1,000
0 , ,
,
2'0 4'0 610 8,0 100 1210
148
4.5.3.3 TF surface strain distribution: 1/10 scarf repaired beam
Although the quasi-3D shell element model may be expected to be in better agreement
with experiment than the 2D plane strain model (seesection 4.3.5), the 1/10 HTC scarf
repaired beam was modelled using the latter approach only. The reason being that,
although the plane strain model leads to predictions of stiffness that are somewhat
higher than the expected values, 2D models were studied so that the stresses and
In Figure 4.40, the measured and calculated surface strain distributions on a 1/10 scarf
repaired beam with the TF in compression are compared. The experimental results are
in good agreement with the calculated strain field. Again, the calculated values are
somewhat lower in the repair than the measured values, due to the plane strain
assumption. It may be seen that the shape of the calculated strain distribution is
identical to that of the 1/30 scarf repaired beam (Figure 4.38), although it has shifted to
the right. The start of the overlapping ply (point A) is now located 74mm,away from a
loading point; i. e. 1/10 repairs are smaller than 1/30 repairs. The stress concentration
at the end of the overlapping ply (point A) is identical to the one calculated earlier,
which is expected since the geometry of the 1/30 and the 1/10 scarf are similar.
However, an increase in strain may be seen at about 90mm,from the loading point, at
the region where the scarf angle terminates (point C in Figure 4.40). A similar increase
in strain was observed at the same location for a 1/20 scarf repairs of thick sandwich
beams by Baker et al. [85]. The origin of this strain increase was suggested [851to be
related to the shear deformation of the adhesive within the scarf. The adhesive shear
strain in the adhesive is uniform along the scarf, which means that the displacements
of the adherends are uniform. However, the adhesive shear strain in the overlap is
zero away from the end of the overlapping ply, which in turn means that the relative
displacement between the overlapping ply and the adherend is negligible. The net
effect is an incompatibly in the strain at the scarf end, at the region where the scarf and
the overlapping ply meet (i.e. point C). This may result in an increase in tensile strain
in the repair at this location (Figure 4.39) and so explain the results seen in Figure 4.40.
149
The displacement between the The relative displacement between
parent and the overlap is nil the parent and the overlap is finite
125 mm
75mm
25mm
A, B
I he surface IT strain S Repair patch
were read on the dash lines 1111ý19W
- T--
Adhesive
-2,000-
1,000
0
0 20 40 60 80 100 120
150
4.5.3.4 TF surface strain distribution: 2-ply overlap repaired beam
For the reasons explained in section 4.5.3.3the 2-ply HTC overlap repair was modelled
with the 2D plane strain model only. The surface axial strain distribution on a 2-pIy
overlap repaired beam in compression is shown in Figure 4.41. The calculated values
are in good agreement with the experimental results. The distribution of strains
follows the now expected trend, although the overlap end is located at 80mm from a
loading point and the entire curve is again shifted to the right, when compared with
Figure 4.38.
stepped-lap joints, and it has been shown that the joint geometry in this area has an
influence on the load transfer mechanism. It was, for example, discussed by Adams et
al. [47] that the maximum shear stress at the end of the step increases when the butt
spacing increases. This is because the proportion of the load transferred by shear,
rather than by the butt faces, is increased when the thickness of the adhesive between
the steps increases. Increasing the butt spacing increases the adhesive shear stress at
the end of the step, but it reduces the stress concentration between the butt faces [47].
A butt spacing 0.5mm was assumed in this work. The effect on the TF surface strain
distribution is however small.
It may be noted that the far field strain is predicted as -8000gstrain, a value identical to
the far field strain of the undamaged beam in compression, as was shown in Figure
4.17. The models were computed at the same load, which means that the beams were
subjected to an identical bending moment, which in turn means that the parent surface
strains were calculated to be equal. Only the strain in the region where the cross-
sectional properties or the materials have changed (i.e. the repair patch) are different.
151
125 mm
ý 95mm
75 mm
25
,111
Core
Adhesive
-7,000
FD]
-6,000
-5,000 AA-00000000R
-4,000 Experiment (CCND67(2)) om
0
0
2D plane strain model
-3,000
-2,000 0
0
- 1,000
0
0 20 40 60 80 100 120
152
4.5.3.5 Summary on the prediction of the TF surface strain distributions
The calculated, normalised (e.g. divided by the far field strain), surface strain
distribution on the TF of the repaired beams, is summarised in Figure 4.42, for the
The results for the repaired beam tested with the TF in tension were not presented
previously, as they are largely identical (apart from the minus sign) to the results for
the repaired beams tested with the TF in compression. However, it mav be noted tilat
the increase in strain at the overlap end is lower when the repairs are under tensile
loads. The normalised value of 1.2 the far-field strain in tension may be compared with
the value of 1.35 in compression. This feature has not been investigated the present
work and will be discussed in a later section. It may also be noted that the strain in tile
repair region is 70% of the far field strain for the scarf repaired beams, while it is onk,
50% of the far field strain for the overlap repaired beams; the overlap repaired beams
are expected to be stiffer than the scarf repaired beams. The present repair schemes
restored the load-carrying capability, and protected the damaged area frorn further
damage.
Figure 4.42: Surface strain distribution with the TF in compression and in tension
153
4.5.3.6 TF [45/451 ply strain distribution
It is of interest to look at the calculated strain distribution in the [45/-451 ply of tile '1'1-.
The graphs were observed largely identical for the beams with the TF in compressioll
and in tension. Consequently, only one graph (of the normalised strain values) is
plotted in Figure 4.43. The values were normalised by the far field strain (i. e.
+7814ýistrain at 43.8N/mm). It is seen that the effect of the load transfer betweeii the
parent and the repair patch has little influence on the [45/-45] ply, and it is therefore
an effect localised to the vicinity of the adhesive fillet.
I ioygbj ----
strains were 145/-451
ý dash lines
---------------------
Ln 1.0-
0.8- 00
C?
0 C?
0.6- 0
CZ 2-ply overlap repair
0
1/30 scarf repair
cc 0.4 - I/ 10 scarf repair
-0
0.2
z0 20 40 60 80 100 120
154
4.5.3.7 Theoretical adhesive shear strain distributions
In the 2D plane strain approach the detailed repair patches and adhesive geometry can
be modelled, allowing the distribution of the adhesive shear strain in the different type
The strain distribution in the adhesive is much as expected from the theoretical
.,
analyses presented in the literature survey (section 2.4.2). The shear lag at end of the
overlapping ply (Omm) causes a peak in the strain, which progressively decreases
towards zero at the centre of the overlap (elastic trough). It may however be seen that
the maximum strain is dependent on the loading mode, i. e. whether the repair is on
the compressive or tensile side. The maximum strain at the end of the overlapping ply
is about -25,000ýtstrainin compression whereas it is 60,000ýLstrainin tension. It may be
recalled that the ultimate shear strain of the present adhesive is 350,000gstrain (i. e. 35%
failure strain) [65], and the adhesive would therefore not be expected to fail in shear,
i. e. if we assume that the adhesive will fail when the shear strain reaches the maximum
adhesive shear strain. The overlap terminates at point C and the adhesive shear strain
distribution is then seen to be more or less constant in the scarf region. This may have
been expected owing to the low scarf angle of 1.9* used. A small step in strain is
furthermore seen half-way along the scarf (point Cl), and this corresponds to the
region where the parent/repair ply orientation changes, i. e. from [0/90] to [45/-451. A
[45/-45] ply has a lower axial stiffness than a [0/90] ply and would therefore attract a
lower load which, as a result, decreasesthe adhesive shear stress in the zone adjacent
155
It may be noted that the values given in Figure 4.44 were read at the centre of the
adhesive layer, which may or may not be fully representative of the current straill
distribution in the adhesive, since a singularity exists at the overlap edge and the straill
increases as we approach it.
I he,
15mm 18mni
-60,000 TF in compression
Loading= 43.88N/nim
-50,000- 1/30 scarf
---
=L
Z -40,000-
-30.000-
Q)
X
-20,000-
nc
10,000-1
Icil
klý
0
-l'
0 10 20 30 40 50 6
60,000
50,000
40,000
li
tt 30,000
> 20,000
10,000
0! ýO IA
0 1,0 30 4'0 5'0 60
156
1/10 scarf repaired beams
The calculated shear strain distribution in the 1/10 scarf repair is shown in Figure 4.45.
The peak in shear strain at the adhesive comers (point A) in Figure 4.45 may be
compared with the one in Figure 4.44, and they may be seen to be identical (i.e. with
respectively identical loading mode). This is not a surprise since the geometry of the
overlapping ply is identical, only the scarf angle is different and this does not affect the
stress transfer at the end of the repair. The shear strain distribution at the start of the
scarf (point C) is seen, however, to be very different. A high shear strain is reached at
that location, i. e. 25% for the repair in compression and 10% for the repair in tension.
The peak in strain may have been caused by the sharper 6* angle of the scarf.
However, we may recall at this stage that the values were read at the centre of the
adhesive layer, and it is noted that the adhesive shear strains were seen to vary
significantly through the thickness of the adhesive layer, and along the length of the
scarf for this model. The region of high strains was seen to move from the top of the
adhesive layer at the start of the scarf, to the bottom of the adhesive layer at the end of
the scarf, as shown schematically in Figure 4.46. The origin of this peak in strain is
questionable, and has not been investigated in the present work. If this peak is
ignored, the strain distribution in the scarf region may then be seen to be
approximately constant over the scarf length, although higher than for the 1/30 scarf
repairs as a result of the higher scarf angle.
Conclusions
2D plane strain FE models were used to calculate the strain distribution in the
adhesive layer of scarf and overlap repaired beams. The trends in the strain
distribution calculated in the present work are in good agreement with the analyses
previously undertaken on lap and scarf joints, and presented in the literature survey
(see section 2.4.2). It is now of interest to predict the strength of the repairs, based on
the results from the FE analyses, and this theme is discussed in the next section.
157
A
ic, _4; I
145/45]
Core
-250,000- TF in compression
-150,000-
.c
125,000-
100.000- F--
-75,000-
-50,000-
x -25,000
0
25 000
,
0 10 20 30 40 50 60
Distance from the overlap end (rmn)
100,000-
90,000-
80,000-
70,000-
60,000-
50,000-
40,000-
30,000-
20,0001
10,000
0
0 10 20 30 40 50 60
Distance from the overlap end (min)
[0/901
10/90] ........
[45/451
[45/-451
158
sII II
A
Core
-80,000 TF in compression
-70,000 Loading= 43.8N/mm
60,000 2-ply overlap
-50,000
FDý
-40,000 El
-30,000
-20,000
.< -10,000
'
123 lO 11 12 13 14 15
80,000 TF in tonsion
70,000
60,000
50.000
40,000
30,000
20,000
10,000
01 23456789101112131415
159
4.5.3.8 Finite elements models of the repaired beams: discussion and conclusions
The FE method was successfully used to model the flexural behaviour of scarf and
overlap repaired beams. First, the calculated deflections and surface strain
distributions were compared with the experimental data and the results were seen to
be in very good agreement. Secondly, the calculated surface strain and adhesive shear
strain distributions from the 2D plane strain models were compared for the different
type of repair. The use of these data to predict the failure load and failure mode of the
repaired beams is discussed in this section.
Stiffness predictions
The deflection field at an edge of a 1/30 HTC scarf repaired beam tested with the TF in
compression was compared with calculated FE results in Figure 4.37. The results from
the 2D plane strain and the quasi-31) shell element models were in good agreement
with the experimental data. The shell model was however seen to lead to more
accurate predictions than that of the 2D plane strain model. Indeed, the deflections
calculated by the 2D plane strain model were seen to be only about 10% smaller than
the measured deflections. Furthermore, the same trend in results was seen when the
calculated surface strain distributions of the scarf and overlap repaired beams were
compared with experimental results. Thus, the calculated FE results were in good
agreement with the experiment, although the 2D plane strain model predicted a
response that is somewhat stiffer than the response calculated by the shell element
model.
The disagreement between the results calculated by the 2D plane strain model and the
experiments may be seen as a consequenceof the plane strain assumption used in the
formulation of the 2D element. Indeed, beams subjected to a bending load deform in a
complex manner and Poisson's ratio effects induce a cross-wise bending that produces
the well known anticlastic curvature. Idealisations that assume zero strain (i.e. plane
strain) in the cross-wise direction cannot, therefore, lead to precise predictions. The
overall effect is that the beam is too stiff. On the other hand, the shell element model
takes into account the quasi-3D deformation field and as a result, calculated
deflectionswere seenin very good agreementwith experimentallymeasuredvalues.
160
Type of models
The results from the quasi-31) shell and 2D plane strain FE models of a 1/30 HTC scarf
repaired beam were compared, and they showed good correspondence with
experiments, although the 2D plane strain idealisation led to slightly stiffer
For instance, it is noted that quasi-3D shell models do not cut through a slice of the
structure. However, even though composite shell elements can identify layer
orientations and take into account mechanical couplings, their nature leads to a loss of
the details being modelled. As a result, and although the calculated deflections and
strains were seen to be in good agreement with their corresponding measured values,
the accurate prediction of the adhesive shear stress in the repairs may be seen to be
impossible with this type of modelling approach. If this type of information is of
interest in designing the repair patches, then a 2D or a full 3D model is needed.
The main feature of 2D plane strain models is their ability to describe a planar
scarf, repair patch and adhesive geometry can be studied precisely since the model
allows the predictions of surface and adhesive stress/strain distributions. It was
however seen that the reduction of the analysis to a 2D problem led to predictions that,
in the present study, were overstiff when compared with experimental results.
We have only shown so far that the modelling approaches that were used to
characterise the behaviour of the undamaged beams can equally be applied to model
the behaviour of the repaired beams. The 2D and quasi-3D models were seen to be
complementary in this study. The two-dimensional model allows the careful
evaluation of several repair designs (e.g. it allows the investigation of the effect of
repair stacking sequence, adhesive thickness and properties on the strain/stress
distribution). The quasi-31) model was however simple to set up (i. e. no complex
geometry in the repair area) and best suited at predicting the overall deformation
161
Failure predictions
It is also of interest to use the calculated results to predict the failure load of the
repaired structures. However, we may recall that the aforementioned tools were
partially inadequate at predicting the structural failure load of the undamaged bean-is.
The progression and evolution of damage needs to be taken into account if ail accurate
calculation of the structural failure load is to be made. The sarne cornmeiit is valid for
the prediction of the failure load of the repaired beams. Moreover, the exercise is n1ore
complicated here, as other type of failure are possible. These are:
i) Failure in the parent due to the action of the in-plane direct stress in the
iv) Failure in the adhesive due to the action of shear or peel stresses.
The calculations for those failure modes were not attempted in the present work.
However, the calculated results may be analysed to explain the failure mode of tlie
repaired beams and this theme is discussed next. For clarity the summary of tilt, static
failure loci and loads is given in Figure 4.48.
Appr-imal,
1/30 scdri 4530N
400ON
HTC I-iti- ot th,
I
-rf 495ON
45DON I/10
1 1,rc
519ON ý160ON
I
2-ply
4750N 440ON
. -I.
LTC
(For comparison the undamaged beams failed at 438ON and 500ON for the beams
162
The calculated surface strain distribution
The calculated surface strain distribution shown in Figure 4.38, Figure 4.40 and Figure
4.41 may be used to predict the failure in the parent or the repair material.
repairs (point A in Figure 4.38, Figure 4.40 and Figure 4.41). We identified that this was
due to the load transfer from the parent to the repair patch. The increase in strain was
seen to be localised to the ply immediately underneath the adhesive fillet (i.e. the
[0/90] ply of the TF), as the strain distribution in the [45/-45] ply of the TF was seen to
be constant over the beams' half length (Figure 4.43). A strain concentration of 1.35
and 1.2 was quoted for the repairs in compression and in tension, respectively. It is
noted that a higher peak in adherend direct strain was calculated for the repair on the
compressive side and the is it
reason not clear, although may have been caused by the
difference in stress state when the repair is in compression or in tension, as shown in
Figure 4.49. The strain concentration at the end of the overlapping ply corresponds to a
35% and 20% increase in the direct strain. Failure in the adherend is then likely to be
expected in the vicinity of the repair, for all repairs. However, most repaired beams
failed in the parent, and away from the end of the overlap. Only the 2-ply HTC overlap
A possible reason for this is that it is well known that the evolution of damage in CFRP
materials may help to reduce the high stresses present at stress raisers. Moreover,
recent research on the effect of ply drop-off in laminates and sandwich beams [120] has
shown that a ply-drop (which is identical to the overlapping ply in a repair joint) leads
to local bending effects extending a short distance to either side of the ply-drop
position. In bending, the overlapping ply will either push through or pull the core,
whether the repair is in compression or in tension, as shown in Figure 4.49. The
overlap repairs, being thicker and stiffer that the scarf repairs, would be more critical
in producing this effect, and this may cause the failure of the overlap repairs at the
parent/overlap interface. Indeed, in Figure 4.48 the HTC overlap repairs may be seen
to have failed away at the end of the overlapping ply. A question then arise, since, as
seen in Figure 4.48, only the HTC overlap failed at the parent/overlap junction; the 2-
ply LTC overlap repairs failed in the parent. However it has been argued, in section
163
4.4.2.4, that the different properties (i. e. geometric and material) of FITC and LTC
repair systems may led to a different stress state in the repairs, and thus may explain
the difference in the failure modes of the HTC and LTC repairs seen in Figure 4.48.
However, the FE models were based on the HTC repair system material properties and
geometry. The LTC repairs were not modelled in the present work and a discussion oil
the effect of repair material properties on stress concentrations is therefore not
possible.
than the parent. This has the effect of protecting the repair area, with the strains iii the
repair being lower than in the parent material. The surface strains ill the repair patch
were seen to be 70% and 50% of the far field strain for the scarf and overlap repairs,
irrespective of the loading mode. As a result, the overlap and scarf repaired beams are
predicted to be stiffer than the undamaged beams, and this was verified
experimentally, except for the 1/30 scarf repaired beams in tension. The strains iii the
repair are much lower than that of the parent and failure in the repair is therefore not
expected. This is in agreement with the experimental observations, see Figure 4.48.
An exception, however, was seen for the 1/30 LTC scarf repair which failed iii the
repair (Figure 4.48). The 1/30 LTC overlap beam failed at the scarf/ overlappiiig pk
junction where an incompatibility in adhesive shear strain was shown to exist (see
section 4.5.3.3), although only on the results of the 1/10 scarf repair. Again, a possible
reason for the failure in the repair patch may be as a result of the difference het", een
the HTC and LTC repair systems (section 4.4.2.4). However, it was seen that tile saille
repair, but tested with the repaired TF in tension, failed in the parent. The discussion is
164
further complicated by the fact that failure may also be caused by voids or defects in
the repaired material. Whatever the reasons, in the present work, the results from the
FE models do not seem to be able to predict the performance of the LTC repairs, since
the FE models were based on the HTC repair system material properties and
(iii) Delaminationfailure
In Figure 4.49 it may be seen that when the TF repair is tested in compression the
overlap tends to push through the core of the beam whereas it peels it in tension. An
opposite sign in through-thickness stresses is thus expected. The local bending effect
associated with the ply-drop is important since it may initiate failure modes such as
delamination (in tension), or core crushing (in compression), in addition to the direct
bending failure of the parent (in compression and in tension). Of the aforementioned
failure modes, delarnination failure would be the most difficult to predict. Several
methods, involving the use of advanced FE analyses, exist and were recently described
in the literature. Charalambides et al. [66] used a numerical fracture mechanics
approach based on the virtual crack closure method to predict the failure load of scarf
repairs that fail by delan-dnation of the overlapping ply. The calculations were in good
agreement with results, however a crack of known length, at the supposed failure
location, has to be included in the geometry of the model. Newer methods, that use
cohesive zone models, do not have this problem, and were recently successful at
predicting the delarrdnation behaviour of CFRP materials [88,121].
shear strain distributions for the three types of repair were shown in Figure 4.44,
Figure 4.45 and Figure 4.47. The calculated shear strain distributions along the overlap
length may be seen to be in good agreement with the theoretical shear stress
distribution of lap joints discussed in section 2.4.1; the shear stress is predicted to vary
along the overlap length and is maximum at the end of the overlap. We may be
tempted to use the adhesive shear strain distributions to predict the failure load of the
repaired beam (i.e. for adhesive failure). However, several problems then arise, as
developedbelow.
165
(iv)Failure of the adhesiveby theaction of shearforcesat theend of the overlappingply
The peak values in strain at the end of the overlapping plies were seen to be dependent
on the repair design (i.e. scarf or overlap), although they were identical for the scarf
repairs: maximum strains of -2.5% and 6% for the repairs in compression and in
tension,respectively,were calculated.The maximum strainscalculatedfor the overlap
repairswere, however,seento be higher at -4%and 8%,for the repairs in compression
and in tension, respectively.For eachrepair, the highest value of peak in strain was
thus calculatedfor the repair in tension.However, it is noted that thesestrains are all
well below the value of the maximum shear strain of the adhesive, i. e. 45% [65,66].
Thus, the adhesive is not predicted to fail at the end of the overlap.
It is recalled that the adhesive shear strain values were read at the centre of the
adhesive layer. It is recognised that the calculated shear strain increases as we move
closer to the edge of the overlapping ply, since it represents a numerical singularity. As
a results, it is impossible to say what the true maximum shear strains are in the
adhesive. However, with the current meshes, the maximum value of adhesive shear
strain calculated (i.e. at the Gauss point of the element the closest to the singularity)
was 26% for the overlap repair in tension, this value is still well below the maximum
adhesive shear strain. Thus, it may be concluded that in the present work the
theoretical analyses agree with experimental observations, since the repairs did not fail
by adhesive shear failure.
of the 1/10 scarf repaired beams (Figure 4.45 and Figure 4.40, respectively) at the
region where the scarf angle tern-tinates.The origin of the strain increase in surface
strain was suggested to be related to the shear deformation of the adhesive within the
scarf (see section 4.5.3.3) [851. The maximum value of the strain increase in the
adhesive is questionable, but it may be seen in Figure 4.45 that a peak in the shear
strain of 25% is predicted at the scarf/ overlapping ply junction. This value of strain is
nevertheless still below the maximum shear strain for the adhesive (i. e. 45%) and the
adhesive is therefore not expected to fail at this location. The FE results agree with the
experimental results since the 1/10 scarf repaired beams failed in the parent.
166
Conclusions on the prediction of failure
The use of the FE results to predict structural failure was discussed. The complex stress
transfer mechanism between the parent and the repair patch was shown to result in a
peak in the surface strain close to the end of the overlapping ply of the repairs. The
repaired beamswere consequentlyall expectedto fail at this location. However, only
the 2-ply HTC overlap repairs did, see Figure 4.48. This difference between the
predicted and experimentalresults was suggestedto be a result of the evolution of
damagein the materials(e.g. parent and repair) that was not taken into accountin the
presentmodels.The differencein failure modesof the beamswas alsoconsideredto be
a result of the difference in the properties (material and geometry) of the high- and
low-temperaturecure repair systems,which were not taken into account.
The use of the adhesive shear strain distribution to predict failure was also discussed.
Assuming that the adhesive fails when its reaches its maximum shear strain value, it
was seen that with the current mesh adhesive failure was not predicted to occur.
However, the adhesive at this location is under a complicated state of stress, for which
the in-plane shear stress failure criterion may not be representative. The choice of a
failure criterion to predict adhesive failure is an active subject of research; the study of
the variation of strain energy density, equivalent shear stressesand the use of fracture
mechanics approach has been suggested as appropriate failure criterion amongst
others. However, none of them has been proven to be reliable for an types of
joint/repair designs [122]. As a results, due the lack of knowledge of a working failure
criterion, these graphs may be used to understand the stress transfer mechanism
through the adhesive, but it is the author's view that they should not be used to
predict failure.
materials that are far from being perfect. On the other hand, the corresponding
idealisation (right hand side) is defined by precise geometric entities, with no defects.
There exists is no guidelines at present as to whether 'defects' should or not be taken
167
into consideration to be able to predict accurately the failure load of the repairs. As a
result it may be seen that at the present time the FE method can only be used for
particularly suitable at studying the stress concentration arising in the parent and
repaired materials. This method is therefore suitable to design qualitatively the repair
patches.
1. Ah
A . ', .1
rI repr
,
Broken fibre
. -I. gried
Core Conditionnnig
ply, r-,. -. &,
p...
-
i-k, n
ad h-on, etc.
performance of the repaired beams. The flexural behaviour of the beams was
successfully modelled, although the prediction of the failure load has been more
problematic. This was identified as being largely due to the complex nature of CFRP
materials and to the inadequacy of the existing method to predict the failure load of
adhesive joints. An alternative experimental approach, using small, inexpensive,
coupons will be used in Chapter 6 to predict structural failure.
Quasi-3D and 2D models were seen to be complementary in this study. The quasi-31)
shell element model was simple to set up (i. e. no complex geometry in the repaii- area)
and best suited to predicting the overall deformation pattem of a chosezi repaired
structure. The 2D plane strain model allows the careful evaluation of several repair
designs (e.g. effect of repair stacking sequence, adhesive thickness and properties, etc. ).
168
4.6 Conclusions on the static performance of composite sandwich beams
The performance of scarf and overlap repaired beams (see Figures 3.4,3.6,3.7) with
their tool face (TF) loaded either in compression or in tension was assessed.From the
present work it is concluded that:
Experimental studies
The static strength of the undamaged tool face was essentially the same whether
loaded in compression or in tension. The repair schemes used in the present work
were found to be very successful. Most repairs failed at a load above 90% of the
undamaged failure load, therefore recovering the load-carrying capability of the
structure. In compression, scarf repairs were, however, found to be weaker (i.e. failed
at a lower load) than overlap repairs; the former achieved only about 90%, whereas the
latter achieved 100% of the undamaged strength. Conversely scarf repairs were
stronger in tension, achieving 100% of the undamaged strength, with now the overlap
repairs only showing about 80% of the undamaged strength.
The repair system, being either high- or low-temperature cure did not have any major
effect on the strength of the beam in this study, although a difference in loci of failure
was identified.
The loci of failure for the undamaged beams was in the region between the inner
loading points and away from the load spreaders. Most specimens failed outside the
repair, except for the 1/30 LTC scarf repair that failed in the repair in compression,
and the 2-ply HTC overlap repairs that failed at the parent/repair interface (seeFigure
4.48).
Values of the load, ram deflection and axial TF surface strains were recorded during
the tests. The relevant test results were analysed and compared. A good
reproducibility in the measured stiffness values of the beams were recorded.
Predictive studies
The FE analysis method has been shown to be a good candidate to evaluate the
deflections and the stiffnesses of the repaired beams. Good correlation between the
169
in the complexity of the model improves the quality of the prediction. The 3D brick
element model gave the most accurate results but brick elements are computationally
expensive and repairs could be evaluated on a day-to-day basis, rendering brick
models impractical. The 2D and quasi-31) models were seen to be complementary in
this study. The 2D plane strain model allows the careful evaluation of several repair
designs (e.g. effect of repair stacking sequence,adhesive thickness and properties, etc.).
The quasi-3D model was simple to set up (i.e. no complex geometry in the repair area)
and best suited to predicting the overall deformation pattern of a chosen repaired
structure. The prediction of the failure load of the beams was found to be very
difficult. This was identified as being largely due to the complex nature of the failure
of CFRP materials and to the inadequacy of the existing methods to predict the failure
load of adhesive joints.
170
5. The fatigue performance of composite sandwich beams
5.1 Introduction
and provide a basis to explain the behaviour of the beams under the different loading
modes.
In section 4.2.2.5, it was found that undamaged beams tested with the TF in
compression failed at a load of 43.8N/mm width. This value was chosen as the
reference for beams tested with the TF in compression and was called the undamaged
failure load (UFL) of the beams. The UFL of beams tested with the TF in tension was
similarly found to be 50N/mm width. The fatigue programme involved 100mm wide
beams only. These values of the two UFLs were used to determine the maximum
fatigue load to employ in the tests for the beams with the TF in compression and in
tension, respectively. For instance, a maximum fatigue load of 50%UFL in compression
would mean that the maximum value of the cyclic load applied to the beam was
0.5x43.8=21.9N/mm. Likewise, in tension a 50%UFL means that the maximum applied
load was 0.5x5O=25N/mm. The fatigue tests were performed under load control, with
a load ratio of R=0.1 and at a frequency of 1 Hz. Undamaged beams were tested at
maximum fatigue loads ranging from 60% to 85% of their respective ultimate static
failure load (UFL). The performance of the undamaged beams was determined and
then used as a reference to assessthe performance of the repaired beams tested with
the TF either in compression or in tension, at 60% of their respective UFL.
171
5.2 Undamaged and repaired beams: Experimental results
Table 5-1 sun-unarises the test results for undamaged and repaired beams tested in
fatigue with the tool facein compression.It may be seenin Table5-1 that the scatterin
the resultsis relatively high, which is typical of fatigue tests.
Table 5-1: Tests results for 100mm wide repaired beams tested in fatigue
(TF in compression)
Specimen II Type Fatigue load I Cycles to failure I Failure details
(%UFL) as in
studies.
172
5.2.1.1 Undamaged beams in compression
In Table 5-1, it may be seen that an undamaged beam tested with the TF iii
LCIII,
maximum fatigue load of 60% of the UFL. The beams failed outside the repair, as
shown in Figure 5.2. However, it should be noted that failure occurred within 5min of
the overlap end, and hence close to the repair patch. It may be recalled that the
modelling studies (section 4.5.3.5) identified an area of increased strain close to the
overlap end due to the load transfer from the parent to the repair overlap. Altilough
the static locus of failure was located away from this zone, it is suggested that the
The 1/30 scarf LTC (low-temperature cure) repaired beams did not perforin very well
when compared with the 1/30 scarf HTC repaired bearn. The 1/30 scarf LTC repaired
beams lasted an average of only 55,000 cycles and failed in the repair, as shown in
Figure 5.3. The possible reasons for the premature failure are not clear as the calculated
strain concentrations in the overlap and adhesive at the failure location were seen to he
very low in section 4.5.3.8. It may be also recalled that, tested statically, the 1/30 scarf
LTC repaired beam was the only beam to fail in the repair, although the possible cause
of failure were not found. It was however suggested, iii sectioii 4.5.3.8, that the
different material and geometric properties of the HTC and I-TC repair systenis rnav
173
lead to different stress concentration in the repairs, which would then deteriniiie the
strength and locus of failure. In the present study the LTC scarf repair was shown to
fail prematurely in the repair when tested in fatigue, and it was shown to be the only
repair to fail in the repair patch when tested statically. Thus, it is clear that the LTC
Figure 5.2: Compressive failure close to the repair (a scarf repair is shown)
(see also Figure A. 4, Appendix A)
significant effect on the fatigue life. The 1/10 scarf HTC repaired beam tested in
compression lasted about 160,000 cycles, similar to the 1/30 HTC repaired bearlis. The
locus of failure was, however, in the repair, as shown in Figure, 5.3. The patch was also
seen to have delarninated, along the scarf region and through the adhesive layer. It
may be recalled that the modelling studies had shown, in section 4.5.3.7, that a strain
concentration in the adhesive existed all along the scarf length, althouF,11the validity in
174
the actual peak values of strain was questioned. It is suggested that the high
alternating stress has promoted failure at the scarf/ overlapping ply junction, where
the high strains were calculated.
compression was found to be as good as, and even sometimes superior to, the
performance of the scarf repaired beams. One HTC overlap repaired beam lasted
250,000 cycles before failing close to the repair patch, as shown in Figure 5.2. This
beam lasted slightly longer that the 1/30 scarf HTC repaired beam.
On the other hand, the LTC repaired beam failed in the repair, as shown in Figure 5.3,
but still lasted 175,000cycles. The LTC repaired beams repeatedly failed in the repair
patch. However, the scatter in results makes it difficult to draw any firm conclusions,
and although the LTC overlap repaired beam did not last as long as the HTC overlap
repaired beam, it may be seen that their performances are still comparable.
175
5.2.2 Beams tested with the TF in tension
The results for the undamaged and repaired beams, tested in fatigue, and with the tool
face in tension, are shown in Table 5-2. Again the scatter in results is relatively high.
Table 5-2: Tests results for 100 mm wide repaired beams tested in fatigue
(Tool Face in tension)
Specimen Type I Fatigue load Cycles to failure Failure details
(%UFL)
*Thetest was stoppedbeforethe beamfailed and the beamswere then usedfor microscopy
studies.
176
5.2.2.1 Undamaged beams intension
It may be seen in Table 5-2 that undamaged beams tested with their tool face in tension
had not failed up to 400,000cycles, irrespective of the applied fatigue load. It may be
noted that it was not possible to test an undamaged beams at 85%UFL, as was done for
the beams tested with the TF in compression, making the comparison between tensile
and compressive results more difficult.
when tested in compression, with an average of 306,000 cycles being recorded. The
beams failed at the end of the overlapping ply, as shown in Figure 5.4, a failure mode
not significantly different to that seen when the beam was tested in compression (see
also Figure A. 3 and Figure A. Z Appendix A).
On the other hand, it may be seen in Table 5-2 that the 1/30 LTC scarf repaired beam
had not failed up to 800,000 cycles. This is rather surprising, since the same beam
tested in compression had shown the lowest fatigue resistance.
shown in Figure 5.5. It may be recalled that this mode of failure is identical to the one
manufacturing damage, although the exact reasons for this observation were not
estabhshed.
cycles and failed by delamination of the patches, and through the adhesive layer, as
shown in Figure 5.6. This type of premature failure may have been caused by the high
177
len'lon fallulcat the
parent/repair interface
Figure 5.4: Tensile failure at the parent/repair interface (a scarf repair is showii)
Figure 5.5: Tensile failure in the repair (an overlap repair is shown)
5.2.3 Undamaged and repaired beams tested in fatigue: discussion and conclusions
The fatigue behaviour of undamaged and repaired beams, tested with die '11 either in
summary of the fatigue loci of failure and life of the repaired beams is shown in Figure
5.7. A summary of the static failure loci and loads is re-printer from Chapter Four and
178
Aý CIIý, I- I) UI] I IILT 01 Fatigue compression , I., ý,., I ...... 1" ,, 'I
cycles to failure Y, k-s t" lallow
182,000 306, ý) O
1/30 scarf
±50,000 ±55, )ý;
111C
55,000
1/30 scarf
±20,000 (no failure)
LXC
2-ply Delamination
250,000 6800
overlap -4 through the
II FC adh VSIVV
Figure 5.7: Summary of fatigue failure loci and number of cycles to failure (60"/,,Ul-'I, )
(For comparison the undamaged beams did not fail after 400,000 cycles.. )
ppr-n-W
1/30 , A, f 453ON
400ON
ITC
E2,1
1,11,,
%WN
519ON
-]
475ON 4400N
'TC
_a
(For comparison the undamaged beams failed at 4380N and 500ON for the beams
179
The fatigge performance of the undamaged beams
It has been shown in Table 5-1 that an undamaged beam tested with the TF in
above, it is apparent, in Figure 5.7, that the fatigue performance of the repaired beams
is inferior; the repaired beams having a significantly lower fatigue life than the
undamaged beams. Overall, it may be seen in Figure 5.7 that the performance of all
repaired beams was affected by the repairs, whether tested with the repaired TF in
180
(ii) 1/30 LTC scarfrepairedbeams
Tested with the TF in compression, it may be seen in Figure 5.7 and Figure 5.8 that the
1/30 LTC scarf repaired beams have failed in the repairs. It may be noted that the LTC
scarf repaired beams had the lowest fatigue life of all beams (Figure 5.7), as wen as the
lowest static failure load of all the beams, tested with the TF in compression (Figure
5.8). However, in contrast when tested with the TF in tension the 1/30 LTC scarf
repaired beam had not failed up to 800,000 cycles. The exceptional fatigue Iffe of the
1/30 scarf repaired beam in tension is unclear but it may also be noted that the 1/30
LTC scarf repaired beams performed very well when tested statically in tension
(Figure 4.30). Tbus, the fatigue performance of the 1/30 LTC scarf repaired beams may
be seen to be related to the static performance of the repairs, with respect to the
loading mode. Further, the LTC repair system may be thought to have induced a
The fatigue locus of failure of the 1/10 HTC scarf repaired beams whether tested with
the TF in compression or in tension was at the end of the scarf region, where the
predictive modelling has shown that high adhesive stresses may be present at this
location (section 4.5.3.7).Although the actual peak values of strain were considered to
be unreliable, it was considered that a strain concentration exists at this location, and
that it may have promoted the failure of the adhesive at the end of the scarf region
when tested in fatigue.
181
number of cycles to failure of all the repaired beams. The performance of the 2-ply
HTC overlap repaired beams is comparable in this case to the performance of the 1/30
HTC scarf repaired beam. Furthermore, it is noted that the locus of failure was
identical whether the beams were tested in fatigue or statically (Figure 5.7 and Figure
5.8). Moreover, this locus of failure is identical to the other HTC repaired beams (i.e.
the 1/30 HTC scarf repaired beams) tested in fatigue. The failure may have been
caused by the high direct stressesthat develop at the end of the overlapping ply, as
was discussed above for the 1/30 HTC scarf repairs. However, the 2-ply HTC overlap
repaired beams tested with the TF in tension were not only seen to have failed at the
end of the overlap, but they also delaminated through the adhesive layer. The high-
through thickness stresses occurring at the end of the two-ply overlap would have
helped to initiate the delarnination failure, when tested with the TF in tension. It may
be seen that the fatigue life of the 2-ply overlap repaired beams was the lowest of all
the repaired beams, as a result of the delan-dnation failure of the patch. It may be noted
that since the sign of the through-thickness stresses is opposite when the beams are
tested with the TF in compression; the overlap repaired beams were seen to have
delaminated only when tested with the TF in tension.
equivalent to the one of the HTC scarf repaired beams. The locus of failure was,
however, in the repair. Again, the LTC repair system may be thought to be responsible
for the failure of the repair, when tested in compression. Tested with the TF in tension,
the 2-pIy HTC overlap repaired beams were also seen to have failed at the overlap end
and have delaminated through the adhesive layer. In Figure 5.7 it may also be seen
that the fatigue life of the 2-ply overlap repaired beams is the lowest of all the repaired
beams, when tested with the TF in tension
and the type of repairs and repair systems. In Figure 5.7, it may be seen that the 1/30
HTC scarf repaired beams performed better than any other repaired beams. Tested
182
either with the TF in compression or tension, the 1/30 HTC scarf repaired beams lasted
an average of about 200,000and 300,000cycles, respectively.
On the other hand, the fatigue performance of the 1/30 LTC scarf repaired beams was
poorer since the beams failed prematurely, in the repair, when tested with the TF in
compression. The LTC repair system is suggested to have promoted a premature
failure of the LTC repaired beams when tested in compression.
Increasing the scarf angle from 1/30 to 1/10 did not have any significant effect on the
life of beams when tested with the TF in compression. However, when tested with the
TF in tension, they performed better than the overlap repaired beams, but not as well
as the 1/30 repaired beams. Consequently, the fatigue performance of the 1/10 HTC
scarf repaired beams may be seen to be intermediate when ranked with the other
designsof repairs.
The fatigue performance of the 2-ply overlap repaired beams was the poorest of all the
repaired beams. Although the beams lasted as long as, if not better than, the scarf
repaired beams when tested with the TF in compression, they failed prematurely from
a delamination failure of the repair patch when tested with the TF in tension. The
choice of the repair system, being HTC or LTC did not have any effect on the fatigue
life of the overlap repaired beams
seen by the structure in service. Indeed, as it was discussed in section 4.4.2.4, the
maximum service strains are typically limited to ±4000ptstrain(i. e. the maximum strain
that may exceptionally be seen by the beams during its service Iffe). This is equivalent
to a load of about 200ON for the present beams (i.e. as shown in Figure 4.10). The
repaired beams were tested at maximum fatigue loads of 2633N and 300ON in
compression and tension, respectively and those loads correspond to a TF strains of
about ±6000gstrain (Figure 4.10) which is likely to trigger some damage mechanisms in
the TF of the beams (the static threshold strain for damage initiation was identified to
be about 6000listrain in section 4.3.4.4).
183
The maximum fatigue loads used in the present study are therefore higher than a
structure would normally see in daily operation. The number of fatigue cycles to
failure of a repaired beam may be expected to increase when tested under a lower
maximum fatigue load and it may then be thought that the fatigue life of the overlap
repaired beams be regarded as acceptable. However, a problem which still arises is the
need to assessthe performance of the repaired beams when subjected to a wide range
of loading conditions; this being very costly and time consuming. A method involving
small, inexpensive, coupons, will be discussed in Chapter Six to assess the
Conclusions
When tested with the TF in compression, the overlap repaired beams were seen in the
present study to have performed as well as, if not better than, the scarf repaired beams
(a conclusion identical to that made when the beams were tested statically). Increasing
the scarf angle did not have any significant effect of the fatigue life. This conclusion is
however contrary to the usual belief that a scarf repair is better than an overlap repair.
However, when tested with the TF in tension the fatigue performance of the scarf
repaired beams was seen to be far better than the overlap repaired beams. Indeed, with
the TF in tension increasing the scarf angle had a negative effect on the life of the
repaired beams, with the overlap repairs having the lowest fatigue fife as a result of
the delan-dnation of the patch.
The static and fatigue loci of failure were seen very different in some cases. For
instance, tested statically most repaired beams failed outside the repairs. However,
they failed in the repair when tested in fatigue. It is further noted that the loci of failure
in fatigue is located at the region of increase in stresses predicted by the theoretical
analysis. Although the theoretical analyses could not be used for the quantitative
prediction of the failure loads, they may be used to understand the mechanism of
stress transfer and explain the loci of failure of the repaired beams in fatigue. Also, the
FE method is considered to be very useful to design qualitatively the repair patches.
184
5.3 Fatigue degradation of beams in flexure
5.3.1 Introduction
Further experimental data on of the behaviour of the repaired beams tested in fatigue
are presented in this section. Fatigue damage in CFRP materials is complicated and
involves phenomena such as matrix cracking, fibre failure and deIamination. Extensive
work has been carried out on the effect of individual damage modes on the behaviour
of the material [123,124]. It is well known however, that the various damage modes
take place simultaneously and interact with each other. The total, cumulative, fatigue
damage may then be evaluated by the reduction in residual strength and stiffness [941.
VVThi1st
the residual strength is a meaningful measure, the remaining life of structures
cannot be assessed by non-destructive evaluation, and the process requires an
extensive and expensive experimental programme. Moreover, the process of fatigue
degradation is not simply related to the residual strength, with many composites
failing by the so-called sudden-death phenomenon. On the other hand, it has been
established that parameters such as laminate stiffness can be related to the damage
state and can be measured in-situ without the need of any destructive testing. For
example, the stiffness of a laminate is seen to decreasewith cycles during testing [941.
In this work, the reduction in bending stiffness with fatigue cycling of the beams was
used as a means to monitor damage development. The literature has shown that the
softening of the CFRP laminates may be directly related to the internal damage state of
the material [891.The reduction in bending stiffness of the repaired beams tested at
60%UFL is compared with the reduction in bending stiffness of undamaged beams
cycled from 60% to 75%UFL.
The softening curves were constructed from measurements of the evolution of the ram
deflection with time. The calculations were presented in section 3.3.5.4.The individual
points on the curves were obtained from the value of ram deflection at the maximum
applied load. It may be recalled that the fatigue test were performed under load
control. The data were then normalised by the first value of bending stiffness that was
calculated (i. e. after a few cycles).
185
5.3.2 Beams tested with the TF in compression
The softening curves of undamaged and repaired beams tested with the TF in
compression are presented in Figure 5.9. On the top graph the reduction in bending
stiffness of undamaged beams is presented. The same data are compared with the
reduction in bending stiffness of scarf repaired and overlap repaired beams in the
centre and bottom graphs, respectively.
A general comment may be made regarding the shape of the curves. The initial
reduction in bending stiffness in the early fife of the beams is followed by a region of
gradual reduction. A third phase has been identified in the literature which
undamaged beams. The repaired beams accumulate damage in the same way as an
undamaged beam but tested at the higher maximum fatigue load of 75%UFL. The
'knee' that was identified for the undamaged beams at 100,000cycles has disappeared,
with the curve now running straight down to failure at about 150,000cycles.
186
0.99
ti)
0.98
c:
0.97
-Undamaged Compression 60%
r:jUndamaged Compression 75%
0 96
ca . I OUndamaged Compression 85%
F-
s- I
0 95
.
0 50,000 100,000 150,000 200,000 250,000 300,000 350,000 400,000 450,000
Number of cycles
1
IE
ýt 0.99
0
0 0
0.97 + Undamaged Compression 60%
0 (:1Undamaged Compression 75%
01 /30 scarf HTC - Compression 60%
0.96
E X 1/30,; carf LTC - Compression 601/6
01 /10 scarf HTC - Compression 60%
z
0.95
0 50,000 100,000 150,000 200,000 250,000 300,000 350,000 400,000 450.000
Number of cycles
cu
0.99
-
0.98
0.95
0 50,000 100,000 150,000 200,000 250,000 300,000 350,000 400,000 450,000
Number of cycles
187
5.3.2.3 Overlap repaired beams
On the other hand, in the third graph of Figure 5.9, overlap repairs are shown to soften
in the same way as an undamaged beam tested at the same load ratio of 60%UFL. It
was commented previously that the overlap repaired beams performed as wen, if not
better, than the scarf repaired beams, when tested with the repair in compression. The
overlap repaired beams lasted significantly longer than the scarf repaired beams and
failed after about 250,000cycles. It may now be seen that the overlap repaired beams
behave in much the same way as the undamaged beams.
The reduction in bending stiffness for the beams tested with the TF in tension are
shown in Figure 5.10. In the same way as in Figure 5.9, the reduction in bending
stiffness of the undamaged beams is presented in the top graph. The data are then
compared with the reduction in bending stiffness of scarf and overlap repaired beams
in the centre and the bottom graphs, respectively.
curve is also shown. The reproducibly in stiffness results is good, with the two
duplicatecurvesoverlapping eachother.
The shape of the curves is similar to those presented for the beams tested with the TF
in compression. Two phases may be seen, the first phase consists of a sharp drop in
bending stiffness early in the life of the specimen, whilst the second phase consists of a
linear reduction in bending stiffness.
The bending stiffness of undamaged beams tested at 60% and 65%UFL may be
compared. Although the maximum fatigue load applied to the beams is only 5%
higher, the degradation with cycles is much more rapid for the beam tested at the
higher maximum load. However, it may be recalled that the beams did not fail up to
400,000cycles.
188
It should be noted that the softening of the beams tested with the TF in tension is
greater than the softening of the beams tested with the TF in compression. The
bending stiffness of an undamaged beam tested with the TF in tension appears to have
decreased by 5% after 400,000 cycles at 60%UFL in Figure 5.10. A similar beam, but
tested with the TF in compression, appears to have softened by only 3% after 400,000
cycles at 60%UFL (Figure 5.9). This is interesting since the compression fatigue
behaviour of the beams was shown to be inferior compared with the tension fatigue
behaviour. It will be shown in the fractography studies that the mechanisms of
damage are different whether the TF is subjected to tensile or compressive loads and
that this may be responsible for the greater degree of softening of the beams in
compression fatigue.
reduction of bending stiffness of undamaged beams in the second graph of Figure 5.10.
It may be seen that the 1/30 scarf repaired beams soften as much as an undamaged
beam tested at the higher maximum fatigue load of 65%UFL. Again the repair system,
being either low- or high-temperature cure, is seen to have no influence on the
accumulation of damage. However, increasing the scarf angle may be seen to have a
significant effect on the rate of softening of the beams, with the 1/10 scarf repaired
beams accumulating damage at a higher rate than the 1/30 scarf repaired beams.
compared with the softening curves of scarf repaired beams. It may be recalled that the
overlap repaired beams lasted only about 10,000 cycles, and failed by delamination of
the repair patch from the parent. It may be observed that the higher the scarf angle, the
steeper the softening (i. e. an overlap is effectively a 90' scarf repair).
189
k Undamaged - Tension 601/o
+Undamaged -Tension 601%
0.99 ýý%F-qe jL &Undamaged -Tension65%
a
C:
w
0
-. 0.97 --
"a
111%Y6
A
%P. A
44
4
+++
A
7; V.M
E
0 0.95
z0
50,000 100,000 150,000 200,000 250,000 300,000 350,000 400,000 450,000
Number of cycles
Undamaged - Tension RM
'A [:I Undamaged - Tension 65%,
x 1/30 scarf FITC - Tension 60%
ýt 0.99
12--jr X1 /30 scarf LTC - Tension 60%
0 1/10scarf HTC -Tension 601%
0,
,a ___- ar -,:,.,! .
C:
0.97 0
(U
ch xx
.
_; 0.96
E
0X
z 0.95
0 50,000 100,000 150,000200.000 250,000 300,000 350,000400,000 450,000
Number of cycles
11
ap - Undamaged - Tension 6(YY.
15P
jr E] Undamaged - Tension 65%
A 2-ply overlap HTC - Tension 60%
0.99
o 2-ply overlap LTC - Tension 601%
%Cb- jrý
0.98
13
0.97
-0 q&
"2
0.96
E
0
z 095
0 50,000 100,000 150,000 200,000 250,000 300,000350,000 400,000 450,000
Number of cycles
190
5.3.4 The fatigue degradation of beams in flexure: discussion and conclusions
Introduction
The evolution of the bending stiffness with the number of fatigue cycles of undamaged
and repaired beams was studied. The bending stiffness was seen to decrease steadily
with fatigue cycles. This softening may be related to a decrease in the elastic properties
of the materials that comprise the facings, the core, and the adhesive. Whilst fatigue
damage to the core material may not be relevant for the 'hard' aramid honeycomb
cores (i.e. if compared with 'soft' foam-based cores), a reduction of the bending
stiffness of the beams may be directly attributed to a reduction in the in-plane
mechanical properties of the TF and BF materials.
softening. The shape of the softening curve results from a relative increase of the ram
displacement with the number of fatigue cycles and is consistent with the evolution of
deflection (although inversely proportional) with the number of fatigue cycles of CFRP
It may be recalled that the data in Figure 5.9 and in Figure 5.10 were normalised by the
initial value of bending stiffness (i.e. calculated after a few hundred cycles). The
bending stiffness decreased by as much as 5% for the beams tested with the TF in
tension. However, it is not possible to comment on this value since the data at the
beginrung of the life of the beams were not recorded due to the time needed to set up
the experiment. It is possible to use the static value of the beams as a normalisation
variable. However, a small difference exists between the value 'back-calculated' from
the static and fatigue valuesof ram deflection,sincethe fatigue valuesare subjectedto
inertial effectsthat are not presentin the static tests,i.e. the beamsappearsstiffer in
fatigue loading.
191
lyl2es of repairs and repair systems
It was observed that the repair system had no significant effect on the softening rate,
although a difference in fracture behaviour was observed in the preceding section. The
undamaged and repaired beams were however shown to accumulate damage in
different ways, depending on the type of repair and the mode of loading.
Tested with the TF in compression at 60%UFL, scarf repairs were seen to accumulate
damage in the same way as an undamaged beam, but tested at 75%UFL. On the other
hand, overlap repairs in compression were shown to soften in the same way as an
undamaged beam, tested at the same load ratio of 60%UFL. It should be noted that in
compression the overlap repairs lasted longer than the scarf repairs. Thus, it seemsthat
a premature failure is initiated by the scarf patch when tested in compression. As was
discussed in Chapter Four, the 1/30 repairs give the longest bonded repair patch, and
it may be that the repair interacts with the load spreaders when tested with the TF in
compression. The load spreaders are on the other side of the beam when it is tested
with the TF in tension.
When tested with the TF in tension, the 1/30 scarf repaired beams are seen to soften as
much as an undamaged beam tested at 65%UFL. The 1/10 scarf repaired beams,
however, show a higher rate of softening, which increases again for the overlap
repaired beams. It was noted that the higher the scarf angle, the steeper was the
softening. Increasing the scarf angle appeared to be detrimental under a tensile regime,
whereas it did not significantly affect the compressive response.
Again, the overall conclusions may be seen identical to those made in section 5.2.3.The
study of the evolution of the bending stiffness with the number of fatigue cycles of the
beams has shown that when tested with the TF in compression the overlap repaired
beams were seen to have performed better than the scarf repaired beams. However,
tested with the TF in tension the fatigue performance of the scarf repaired beams was
better than that of the overlap repaired beams; a conclusion identical to that made
192
Effect of the loading mode
The extent of the degradation for the beams tested with the TF in compression and in
tension is now compared. It may be seen in Figure 5.9 that the bending stiffness of the
beams tested with the TF in compression has reduced by no more than 3%. In Figure
5.10, however, the reduction in bending stiffness of the beams tested with the TF in
tension may be seen to drop by as much as 5%.
As it was discussed above, the drop-off in stiffness might have been larger if the data
were normalised by the 'true' value of the stiffness corresponding to the beam without
any damage. However the value of the drop-off of the beams tested with the TF in
compression may be readily compared with the value of the beams tested with the TF
in tension. The tests were conducted in much the same way and the start of the
recording of data was done at approximately the same time, i. e. we may assume that
the initial drop in bending stiffness is identical for all beams.
The beams have therefore appeared to soften more when tested with the TF in tension,
which is in contradiction with the fact that they lasted longer when tested with the TF
in tension. It will be shown in the next section that the exact nature of damage may be
193
5.4 Fractography studies
5.4.1 Introduction
The surface of undamaged and repaired beams, that were tested for over 50,000 cycles,
with the TF in compression and in tension (as seen in Table 5-1 and in Table 5-2,
respectively) were observed using optical microscopy. The surfaces observed spanned
an area from the loading point to the centre of the beams. The grinding and polishing
operations, that were required in order to view the features of the specimen, were
undertaken twice for each specimen. The second grinding operation was perforliled
deeper into the specimen in order to indicate any depth-profile effect.
The specimens were observed along three different directions, as shown in Figure 5.11.
The first direction was along the warp bundles of the [0/901 ply of the 'I'l, (i. e. Ole
longitudinal direction of the beams). The second direction followed the welt bundles
of the [0/901 ply of the TF (i. e. across the width). The third direction was dinied at
observing the [45/-45] ply of the TF. The specimens were hence cut at a 45' an))Ie in(-]
Observation of the
[45/45] ply 45-
-- --- - Observation
TF
along the
T-J-T-T-T- weft bundles
of the [0/90]
ply
reductions in bending stiffness were observed experimentally, as reported iii Hie last
section, very little damage was observed in the current exercise. This iiiav be
understood by considering the large volume of the beams under stress relative to die
small area observed under the microscope. Only damage occurring in tli(, [()/()ol pIN, of
the TF was seen. No damage was seen when looking across the widtli or at a 143/451
orientation. It may be noted that this does not mean that no daniage existed in tlie
194
5.4.2 Beams tested with the TF in compression
compressive loading may be seen in Figure 5.12. The pictures were taken from an
undamaged beam tested with the TF in compression at 60%UFL (CCND6-6(l)). Matrix
cracks were seen to develop along the weft (transverse) bundles, and to run at an angle
of about 451. The cracks were isolated (as may be seen in the bottom picture) or
coupled with others (as may be seen in the top picture) and ran along fibre/matrix
interfaces. It may be further seen that the matrix cracks lead to the partial debonding of
the weft bundle. However, only a few signs of damage were observed, and they were
exclusively located on the surface of 900weft bundles.
The damage observed in the TF of the beams as a consequenceof tensile cyclic loading
is shown in Figure 5.13 (the pictures were taken from an undamaged beam tested at
60%UFL, (CCND6-1(2))). Matrix cracks developed along the weft (transverse) bundles.
The cracks were perpendicular to the applied in-plane load and have initiated at the
fibre/matrix interface and run on surface 90* bundles, as well as inside 90* bundles (as
may be seen in the bottom picture). The cracks arrested when they reached the 0* warp
bundles and then separated the warp and weft bundles.
A limited amount of damage was seen in the repair area. The damage was identical to
that described above (i. e. for the repaired beams tested with the TF in compression and
in tension, respectively) and was mostly contained in the parent material.
195
the weft tows '\ latrix crack
Debonding of warp and
196
Figure 5.13: Tensile damage on thcTF
197
5.4.5 Fractography studies: discussion and conclusions
Introduction
The damage observed in the undamaged and repaired beams was exclusiveIN, secii oii
the [0/901 ply of the TF and was observed when looking along the warp buildles (ix.
in Figure 5.14. Transverse bundles failed by matrix cracks running at 45' froill the
loading axis. The cracks appear to have propagated at the fibre/matrix interface, aiid
materials tested in transverse compression. It should be noted that only a few cracks
were observed and that they were located exclusively on surface transverse bundles,
i. e. underneath an undulating warp bundle. It is suggested that the undulation results
in a local decrease in axial stiffness of the fabric, which in turn results in all inClVdSOill
strain in the weft surface bundles. The fibre/matrix interface may then be expected to
fail by shear under the high compressive strains.
The separation of the warp and the surface weft bundles was also clear. The separatioll
may have initiated at the tip of the transverse shear cracks, L,ffectivelv Ljeta(-jjijjý' tll(,
warp and weft bundles. The separation may then develop progressively in the iliterior
of the undulations. It would be expected that the separation of the weft and \\, arp
bundles would locally reduce the stiffness of the fabric, and effectively OW
i1-1C1VdSL'
strain at the centre of the undulation of the warp bundle, which in turn may pro\.,oke a
10/901 ply
198
Tensile damage on the TF
The damage observed as a consequence of tensile loading on the TF is sketched ill
Figure 5.15. Transverse bundles have separated across their width, with rnatrix cracks
running across the weft bundles, perpendicular to the loadmg directiorl. The cracks
The development of these transverse cracks may be due to the relatively low stif-hiess,
around the weft bundle, resulting from the undulatioii of the warp buildle 11251.This
lower stiffness causes a local softening, aml hence an increase iii straiii, with the
maximum occurring at the centre of the weft bundle. Similar damage inecila 11is ills
have been observed on woven fabric CFRP Ian-finates under static tensile loading 11071,
and they are very similar to matrix cracking in unidirectional cross-ply laminates [ 1231.
Gao et a]. [1071 have reported that the cracks appear to grow instantaneously across
the buridle thickness, and that they span at least one fibre bundle.
[0/901 Ply
Weft bundle
---------- 10.
(,ft buildles
Compressive shear-induced cracks or tensile cracks were observed in the V%,
of the beams tested with the TF in compression or tension, respectively. Although the
observation of the shear-induced cracks has not been widely reported in Ow literature,
the formation of tensile cracks in transverse ply (i. e. a transverse fibre bundle) has beell
discussed by several workers. For instance, Gao et al. [1071 studied 2-pIN, CFRII 811-
woven laminates in tension and reported that the approximate strain applied to the
specimen for significant cracking was about 0.52% (although the applied strain at
which the first transverse crack was observed was 0.3%). It may be recalled that the
bending stiffness versus load curves of beams tested statically with the 'IT in tonsion
199
bundles. The sudden decrease observed at the beginning of the life of the beams maý
also be attributed to the increase in the crack density in the weft bundles.
Such phenomena have been also observed in unidirectional cross-ply lainiiiates ill
tension, and it has been reported that a saturation stage exists, wherebN, crack
formation stops, and the density remains unchanged for the remaining life of the
laminate [89]. Such phenomena are likely in woven fabric composites, although they
have not been observed in the present work.
12.000
o 100mmwide(CCND6-2(1» 11.000
10,000
9,000
8.000
%e !ýý, 7.000
w 0 199 -
0- OW . -,
00 ýer ,-
.1 0,00 ý 0 6,000
0, W,
6, *Z 5,000
4.000
3,000
Z 2.000
19 1.000
4 2K1 0
M0
2000 2ý00 3 3ý00 4M0 4500 5(M
Figure 5.16: Onset of damage in undamaged beams tested with the'll'in tensioil
tested with the TF in compression or in tension. Agaiii the separatioii Inay Ilave
initiated at the tip of matrix cracks. Whereas matrix cracks are fliouglit to iiiitiate at
low working strains (i. e. 0.3% [1071), this separation would initiate at liiglier vý.,
orkhig
strains, i. e. above 0.6% and then propagate with fatigue cycles. Also, as soon as a
delarnination initiates, the stiffness of the facing materials would decrease
considerably [1251. After delamination has occurred the load is takeii by Hie %varp
bundle only. Since the undulation of the weft bundles creates a higher straill at the
crimp region, and since the local constraint has been reduced at the crimp region bv
the delamination, the crimped bundles can therefore elongate more under load, and
thus produce a lower axial modulus [1071. The gradual reduction M beiidillg stiffiless
in this region with the number of cycles may be suggested as being a I-OSUltOf all
increase in the number of separation of the warp and weft fibre bundles.
200
Effect of the loading modes on the formation of damage
The fatigue performance of the beams was inferior when tested with the TF in
compression, compared with the beams tested with the TF in tension. However, the
degradation of the beams' stiffness with the number of cycles was relatively higher
From a damage point of view, the separation of the warp and weft bundles and the
of matrix cracks observed in tension may reduce the axial modulus of the TF more
than the shear-induced cracks observed in compression. This may explain the
contradictory observations that the beams have appeared to soften more and have also
lasted longer when tested with the TF in tension.
The fatigue performance of the beams tested with the tool face (TF) in compression
was inferior compared with the beams tested with the TF in tension. In compression,
an undamaged beam tested at 60%UFL did not fail up to 400,000 cycles. An increase in
the maximum fatigue load to 75% and 85%UFL led to failure occurring at 75,000 cycles
and 35,000 cycles, respectively. On the other hand, undamaged beams tested with their
tool face in tension did not fail after up to 400,000 cycles, irrespective of the magnitude
whatever the loading mode. The fatigue loci of failure and the number of cycles to
failure of the repaired beams are shown in Figure 5.17.
201
Aý crage number of Fatigue compression I atigue tension 1-111h, IM
cycles to failure cycles to tailurt,
182,000 306, WO
1130 scarf
±50,000 ±55, (X)()
HTC
-r- --
55,000
1/30 scarf
±20,000 (no failure)
LTC
160,000 38,000
1/10 scarf
±50,000 HTC ±30,M)
2-ply Delamination
250,000 680(
overlap 4---- through the
HTC adhesive
Delamination
2-ply
175,000 through the 1.5,000
overlap a dh es ive an d
LTC
re air overlap
Figure 5.17: Summary of fatigue failure loci and number of cycles to failure
(For comparison the undamaged beams did not fail over 400,000 cycles. )
average 182,000 cycles. Tested in tension, the repairs lasted longer: on average 300,000
cycles. All beams failed just outside the repair area, at the end of the overlapping 1)1\,,
as shown in Figure 5.17. The beams may have failed due to the increase in strain at t1le
overlap end, that resulted from the load transfer. This increase in strain was calcLilated
to be higher in compression than in tension, and this may explain why the hCallls
lasted longer when tested with the TF in tension.
occurring in the repair. This is rather surprising, since tile adliesive shear stress ill a
1/30 scarf repair was shown to be relatively low. When loaded iii tensimi, hmý,(,vel-,
the repair had not failed up to 800,000cycles. Furthermore, it may he recalled that the
same beams failed prematurely in the repairs when tested statically III ('0111pression,
but performed far better when tested in tension (see section 4.4.2.4). It may therefore be
202
concluded that the performance of the LTC repair system is significantly worse in
compression than in tension, although the exact reasons for this observation were not
established.
when tested in compression. The beams failed in the repair and it is anticipated that
the fatigue life would shorten if the scarf angle was increased further. The beams,
whether tested with the TF in compression or in tension, were seen to fail at the end of
the scarf region, where an increase in strain was numerically calculated.
scarf repaired beams tested in compression. It should be noted that, similarly, the
compressive static performance of the overlap repairs was found to be greater than for
the scarf repairs. The overlap repaired beams tested in compression failed at the
overlap end for the HTC repaired beam, and in the repair for the LTC repaired beams.
This agrees with the earlier conclusion that the LTC repair system may be responsible
for the loci of failure occurring in the repairs.
When testedin tension,the fatigue performanceof the overlap repairs dropped to the
lowest of all the specimens.The beamsfailed by delamination of the repair patchesas
shown in Figure 5.17.The scarf repairs did not fail by delan-dnation,becausethey are
only one ply thick and hencelessprone to delaminationfailure. The overlap repaired
beamsdid not fail by delamination of the patch when tested in compressionbecause
the direction of the through-thicknessstresseswould not promote delan-tinationfailure
in this case.
and weft bundles were seen to be responsible for a gradual softening of the TF, for the
beams tested with the TF in compression. For the beams tested with the TF in tension,
203
transverse cracks, leading to the separation of the warp/weft bundles were similarly
observed. It is suggested that a threshold for significant fatigue damage may exist and
that this threshold is associated with the initiation of delamination damage, since only
the beams tested at loads higher than 60% in compression showed a finite fatigue life.
Tested in tension, 1/30 scarf repairs softened as much as an undamaged beam tested at
65%UFL. The 1/10 scarf repaired beams, however, show a higher rate of softening,
which increased again for the overlap repaired beams. It is noted that, the higher the
scarf angle, the steeper was the softening. Increasing the scarf angle was detrimental
under a tensile regime, whereas it did not significantly affect the compressive
response.
Conclusions
The number of cycles to failure and the evolution of the bending stiffness with the
number of cycles of scarf and overlap repaired beams were analysed in this Chapter.
The conclusions from the fatigue studies may be seen to correspond with the
conclusions from the static studies (section 4.6). In the present work, the overlap
repaired beams were seen to have performed as well as, if not better than, the scarf
repaired beams when tested with the TF in compression. However, when tested with
the TF in tension, the fatigue performance of the scarf repaired beams was seen to be
superior than the overlap repaired beams.
204
6. The performance of tensile coupons representative of the tool
face of the sandwich beams
6.1 Introduction
The static and fatigue performance of undamaged and repaired sandwich beams was
discussed in Chapter Four and in Chapter Five. The experimental testing of such
beams is relatively costly and time consuming and the prediction of the failure load of
the sandwich beams has been shown to be difficult. This was identified as being
largely due to the complex nature of the failure of CFRP materials and to the
inadequacy of the existing methods to predict the failure load of adhesive joints.
Therefore in this Chapter, the static and fatigue performance of small coupons, that
mimic the behaviour of the tool face (TF) of the beams, is presented.
The stresses and strains that develop in a sandwich beam can be calculated by the
application of the engineering theory of composite beams. It will be shown that, in its
simpler form, the theory predicts that the core carries transverse forces as shear
stresses and the skins carry the bending moment by axial stresses. Therefore, for a
large radius of curvature (i.e. small deformations) the skins may be idealised as
laminates loaded axially (i.e. in simple tension or compression). From this basis it is
studied. Furthermore, becauseof the difficulties associated with the testing of slender
laminates in compression, the coupons were subjected to tensile forces only.
205
6.1.1 Engineering theory of composite beams
ME
Equation 6-1
Iy=Ry
MyjEj-j
a, = 1: Equation 6-2
EI
MY,
6, = ZEI Equation 6-3
where a, (s;) is the uniaxial stress (strain) in the ith layer of elastic modulus Ei-i, LEI is
the sum of the bending stiffness of each layer, yi is the distance from the centroid of the
considered material to the beam's neutral axis and M is the applied bending moment.
In Figure 6.1, a sandwich beam with skins of respective axial modulus E1.1and Ei-2and
respective thickness ti and t2, is shown. It may be seen that the neutral axis is not
necessarily located at the centre of the beam but shifts away towards the stiffer skin.
The position of the neutral axis can be found from the condition that the resultant axial
force acting on the cross-section is zero, which after simplification reduces to (with Ai
being the cross sectional area of the ithply):
ZEI-I. yI. Ai
- Equation 64
-ý ýEj-j. Aj
206
E (T
y
Ec, tc
Ivith
El-I < 1-1
2
------------ -- ----- --- --------- - ---
E, z
It may be seen from this simple analysis that the skins carry the bending moment as in-
plane tensile and compressive stresses, depending on which side of the neutral axis
they are located. We may now derive an expression to calculate the magnitude of the
equivalent in-plane load developing in the faces, as a functioii of the applied beii(litig
moment.
axes. Given the strain distribution shown in Figure 6.2, Equation 6-2 and Elquation 6-3
may be used to calculate the stress (strain) distribution in each layOr that coniprises the
skins. The equivalent force (/unit width) that will induce a siniiiar stress (strain) is:
2
f(T(ýJ, ME, f
-i ydy El-ic, t, j:,quation 6-5
Y EI
2
In bending, the strains are assumed to vary linearly through the thickness of tlle be, 1111,
but an equivalent constant strain (stress) may be taken as the mean value of t1le
maximum and minimum strain (stress) for the skin considered. It should be wted that
The load per unit width, either acting on each layer that comprises tile skills, or actilig
as a whole on each skin, may be calculated by the application of Equation h-5. III thL'
present work, the equivalent force on the TF of the beams at failure was calculated bN,
using the surface strains at failure measured with strain gauges (see section 4.4.2).
207
i:
0 0,
C- ---------------
Flat coupons, representative of the TF of the beams, were manufactured and tested in
tension, statically and in fatigue. The geometry of the coupons is shown in (0.
The thickness of the parent coupons was twice that of the TF skins Of tll(' b0cIIIIS, USing
a lay-up as shown in Figure 6.3. This was done in order to have a symmetric lay-up
and so be able to manufacture flat, unwarped, laminate sheets. The material ývas a
woven F914C prepreg supplied by Hexcel (UK) (see Table 3-1). The plate was aired
according to the manufacturer's recommendation in an autoclave. GRP eiid-t, ibs wert,
used on the coupons to reduce the stress concentratimi at the loaded emis an(] to
ensure uniform gripping and load transfer. The end tabs were 50inni long bv 2.5111111
thick. The end tabs were glued onto the plate with an overnight-cure epoxv adliesive
paste. The tensile coupons were then cut, 25mm wide, along the fabric warp direction,
Undamaged, 2-ply overlap HTC and LTC repaired coupons were manufactured. The
sheet into two halves, as shown in Figure 6.5. These two pieces were theii used to
manufacture the repaired coupons and constituted the parent material. A repair 11111g,
60mm long, was used to separate the two parents. I lexcel F914C alid ACC
T300/LTM26EL were used as HTC and LTC repair materials, respectivek, (see sectioli
3.3.3.2). Hexcel Redux 319 (HTC) and ACG XI-TA225 (1,TC) adhesives were us(, (-] to
bond the overlap plies to the parent. The repaired plate was then bagged and vacullill
cured as described in section 3.3.3.4. The coupons were tested using a servo-hydraulic
MAYES fatigue testing machine. Data, such as the evolution of load alld fdtigUO C)'cleS
were logged using a MACLABdata acquisition systern as described iii sectioii 3.3.5.1.
208
'flu
[0/901
145/451 [0/901
[45/451 [45/-451_
10/90]
-------------
11 iaged coupon
--------- -----
i()/90.4511: 9945141(ý
W. rp cl-tion
Rep- pat, 1,
Ad h
Each pact are 1-d I" the Undanýagod
rnanufact. - f he .. P""
coupon
repaired (oupons
Rep- plug
209
6.3 The static performance of undamaged and repaired coupons
Results for the undamaged coupons tested statically in tension are presented in Table
6-1 below. The ultimate failure load of the coupons is given, as well as the TF skin
failure load (i. e. the ultimate failure load divided by the specimen width, 25mm). The
ultimate failure stress is then calculated by dividing the TF skin failure load by the
specimen thickness (i.e. 1.2mm on average).
It may be seen from Table 6-1 that the average ultimate tensile stress of the undamaged
The majority of the specimensfailed by a simple tensile failure away from the end-
tabs.The coefficientof variation on theseresultsis rather high and this may have been
causedby manufacturing defects.It should be noted that the edgesof the specimens
were not polished prior to testing,although this has beenshown to reducethe scatter
in results.
210
6.3.2 Repaired coupons
Results for the repaired coupons tested statically in tension are preseMed in Tahle 6-2.
A total of four high-temperature cure (HTC) and three low-temperatilre ('Llre
specimens were tested. From a statistical point of view the two types of repaired
standard deviation calculated are moreover comparable to the ones of the undamaged
coupons (Table 6-1). Thus, the repaired couporis may be seen to be as stroiig as the
undamaged coupons. The locus of failure for the HTC and IJC ox,erlap repaired
coupons are shown schematically in Figure 6.6. The HTC repaired coupons failed at
the end of the overlap, in the parent material and underneath the adhesive fillet. 011
the other hand, the LTC repaired coupons failed away froin the repair area, aild awaý,
from the end tabs.
2-plv H'lCoverlap
1ra ===r-
14
2-ply 1,TC.,, -, I. V
v
ý:
211
6.3.3 A comparison of the static performance of coupons in tension and beams in
flexure
The tensile performance of the coupons is now compared with the flexural
performance of the beams. In Table 6-3, the experimentally measured skin failure load
of the coupons is compared with the calculated skin failure load of the TF of the
beams.
The skin failure loads of the TF of the beams were calculated using Equation 6-5 (i. e.
the skin failure load is the product of the tool face axial stiffness * its thickness * the
measured strain at failure). The assumption taken here is that the strain is uniform
through the thickness of the TF, and is acceptable since the TF is only 0.6n-unthick. The
axial stiffness of the TF has been taken as an 'homogenised' value. Homogenised
material properties can be calculated by using a laminate analysis approach [6]. The
commercially available LAP software (i.e. laminate analysis programme) [101] was
used to calculate the TF homogenised axial stiffness.
A value of E=33.4GPa was found (according to the material properties in Table 3-1,
and for a 0.6mm laminate made of one 0.3mm [0/90] and one 0.3mm [45/-45] ply).
The values of strain at failure were taken from Table 4-5 and Table 4-2 for the
undamaged and the repaired beams tested in tension, respectively. These were
experimentally measured values, as described in section 4.2.3.
The results are presented in Table 6-3, where the skin failure load of the coupons is
given as half the value shown in Table 6-2, since the coupons were made symmetric
(and hence twice the thickness) when compared with the TF of the actual sandwich
beams. It may be seen in Table 6-3 that the skin failure loads for the coupons and the
beams are in good agreement. Considering the results shown in Table 6-3 in detail,
then the skin failure load (/unit width) of undamaged and 2-ply LTC materials are
very close. However, the skin failure load of 2-ply HTC overlap beams is about 15%
lower than that of equivalent coupons, and may be attributed to the scatter in the
results.
212
Table 6-3: Coupons vs. beams skin failure load
Beams I Coupons
ffie TF in
The loci of failure of the overlap repaired beams tested statically, with
tension, are compared with the loci of failure of equivalent coupons in Figure 6.7. The
manner in which the beams and coupons fail may also be seen to be in very goo(I
agreement. The HTC repairs failed at the parent/repair iiiterface, underileatil tll(,
adhesive fillet, whereas, the LTC repairs failed away from the repair patch and away
from the loading area.
2- ply
over laap
i
F------ -1
I I'l c
Coupon
I-ply 'I
overlap
IAC
Figure 6.7: Loci of failure for the overlap repaired coupons and beams
6.3.4 The static performance of undamaged and repaired coupons: discussion and
conclusions
Undamaged coupons were seen to fail at an ultimate failure stress of 314N,11'.) k%'itll a
standard deviation of ±29MPa. The HTC and LTC overlap repaired coupons failed at
an ultimate stress of 331 and 313MPa, with standard deviations of i22 and 27NIla,
strong as the overlap repaired coupons. Moreover, the repair system did not Ilave
'111Y
effect on the failure loads, although the loci of failure were differeiit. 'I'lle reasoiis for
the difference in the loci of failure of the HTC and L,TC repairs were not clear. Oil mie
hand, the HTC repairs (i. e. beams and coupons) were seen to fail at the ond of the
overlap, at the adhesive fillet, whereas on the other hand, the LTC repairs were seen to
213
fail away from the repair. These conclusions may be seen largely identical to the
conclusions drawn in the analysis of overlap repaired beams tested with the TF in
tension (seesection 4.4.3).Thus, the coupons were able to min-dc the static behaviour of
the TF of the beams, and consequently the static behaviour of the beams.
Furthermore, it may be recalled that in Chapter Four the results from 3D FE analyses,
in conjunction with the Tsai-Hill failure criterion, were examined to predict failure of
undamaged and repaired beams. The analyse was unsuccessful due to the Tsai-Hill
criterion calculating a load for first ply failure, and not taking into account the effect of
n-dcroscopic damage. The use of a more complicated analysis, that does take into
account the evolution of damage, was suggested. However, the coupons may be seen
to be a good alternative to FE analysis methods as a means for proving different repair
designs.
6.4.1 Introduction
presented in this section. The coupons were cycled at maximum fatigue loads ranging
from 50% to 80% of the failure load of the undamaged coupons. The undamaged
failure load of the coupons was shown to be 9436N (±873) in Table 6-1. However, it
was conveniently assumed to be 1OkN for the fatigue testing programme. The
specimens were tested under load control with a load ratio of 0.1. The test frequency
was 5Hz.
Results for the undamaged coupons tested in tensile fatigue are presented in Table 64,
below. It may be seen that undamaged coupons tested at maximum fatigue loads
ranging from 50% to 65% of the coupons' control strength have not failed up to 3.5
It
million cycles. may also be seen that increasing the maximum load to 75% of the
control strength, leads to failure occurring at 1 million and 320,000 cycles for
specimens S4 and S6, respectively. At a higher maximum fatigue load, the life of the
214
undamaged coupons is seen to decreaserapidly, and the coupons sustained only about
6000cycles when tested at 80% of the control strength.
The undamaged coupons were seen to fail (if they did fail) by a simple tensile failure
across the cross-section of the specimens. The failure mode was identical to that
observed when tested staticaRy in tension.
Table 6-4: Test results for undamaged coupons tested in tensile fatigue
si 50 over 3,500,000
I S3 1 60 1 over 3,500,000 -1
I S2 1 65 1 over 3,500,000
S4 75 1,000,000
S6 75 320,000
S13 80 7000
S15 80 6000
Test results for the HTC and LTC overlap repaired coupons are presented in Table 6-5.
It may be seen in Table 6-5 that HTC overlap repaired coupons have not failed up to
2.5 million cycles when tested at 50% of the control load. Increasing the maximum
fatigue load to 60% and 70% of the control load promoted failure after about 400,000
cycles and 90,000 cycles, respectively. However, the LTC overlap repaired coupons
were weaker (i.e. failed at a lower number of cycles) than the HTC overlap repaired
coupons. In Table 6-6, the LTC overlap repaired coupons may be seen to have failed
after 400,000 cycles when tested at 50% of the control load, whereas the HTC overlap
215
repaired coupons have not failed even after 2.5 million cycles. Again, an increase iii the
maximum fatigue load promoted a rapid decrease in the fatiFU(' life Of the SPL'CiIII011S.
The specimens lasted about 100,000 cycles and 10,000 cycles when tested at the
maximum fatigue loads of 60% and 70% of the control load, respectively.
The locus of failure for the two repaired coupon types were ditterent, as shown in
Figure 6.8. The HTC repaired coupons failed by a single delamination that initiated at
one end of the overlap, on one side of the specimen only. The delarnination initiated iii
the adhesive fillet and it was observed, by eye, to progress steadily in the adhOSiVO
layer, with increasing number of fatigue cycles, towards the butt-face regioll of the
overlap. The specimens were seen to fail when the crack was close to the butt-faces. 011
the other hand, the LTC repaired coupons failed by the propagatioii of two
delaminations located on the opposite sides of one end of the overlap, as showii iii
Figure 6.8. In this case, the progression of the delarniiiations was not socil
suggested that they initiated the premature failure of the LTC repaired coupotis.
delamination was seen to have propagated betweeti tile parent aml the overlap plies.
Delarnination
2 -ply HTC overlap through the
adhesive layer
Fast fracturv
occurred when the
Delarnination crack reached the
through the but joints
2 -ply L:FC overlap
adhesive layer
F---
on each side of
thecoupons
Figure 6.8: Loci of failure for overlap repaired coupons tested in tension fatigue
216
Table 6-5: Tests results for 2-ply HTC overlap repairs in tensile fatigue
HTC5 60 412,000
HTC6 60 457,000
HTC7 60 273,000
I HTC10 1 70 1 90,000
Table 6-6: Tests results for 2-ply LTC overlap repairs in tensile fatigue
LTC5 50 465,000
LTC6 50 483,500
LTC9 50 365,000
LTC7 60 89,000
LTC8 60 137,000
LTclo 60 85,000
LTC12 70 13,000
LTC13 70 25,000
217
6.4.4 A comparison of the fatigue performance of coupons in tension and beams in
flexure
undamaged beams tested at maximum fatigue loads ranging from 50% to 75% UFL are
Table 6-7. Comparison between the fatigue performance of beams in flexure and
It may be seen in Table 6-7 that although the static results did look pron-dsing, the
performance of the repaired beams and repaired coupons were different. The overlap
repaired coupons have a better fatigue performance than the equivalent repaired
beams. It may be recalled that the coupons are symmetric double-lap joints tested in
tension, whereas the beam repairs are single strap joints, supported by a sandwich
core. It would be surprising if the life of the beams was identical to that of the coupons.
Nevertheless, the large differences in the fatigue behaviour is disappointing.
stressesis likely to increase. This will increase the load on the TF skin, while it is being
reduced on the other. Therefore, although the test is being done under load control, the
skin load on the TF increases due to the evolution of damage during fatigue cycling.
On the other hand, the coupons were tested under a constant load, and consequently
218
Secondly, as the coupons were manufactured with a symmetric lay-up, their in-plane
axial elastic properties are therefore higher than those of the tool face of the beams,
since coupling effectively reduces the tool face axial stiffness [111,i. e. the coupons are
stiffer than the actual TF of the sandwich beams.
Thirdly, it may be noted that a small change in tool face axial stiffness (such as the
presence of a repair) would not change significantly the overall stiffness of the beams,
since the bag face is stiffer than the tool face. The repair is, however, likely to change
significantly the overall stiffness of the coupons; and hence change the parent axial
strain for a fixed load.
Fourthly, a variation in strain may be expected between the undamaged, the HTC
Table 6-8: A comparison of the far field strain in coupons and beams
In Table 6-8, the coupons may be seen to be generally stiffer than the equivalent
beams,i.e. the strains are lower. The ratio of axial strains in the undamagedbeamsto
that in the undamagedcouponsmay be seento be 1.18.Similarly, this ratio may be
219
seen to be equal to 1.17 for the LTC overlap repairs. The ratio increases to 1.37 for the
HTC overlap repairs.
Furthermore, since the coupons are stiffer than the equivalent beams, it is of interest to
compare the fatigue results under equivalent strains. The load applied to the
undamaged, HTC and LTC repaired coupons has to be multiplied by 1.18,1.37 and
1.17, respectively, to reach a strain equivalent to that seen by the tool faces of the
beams in flexure. For instance, a 60%UFL on the undamaged beams is equivalent to
70%UFL (60*1.17) on the undamaged coupons. Similarly, a 60%UFL on the beams is
The results presented in Table 6-7 may now be factored to take into account the higher
stiffness of the coupons. In Table 6-9 the number of cycles to failure of the beams are
now compared to the number of cycles to failure of the coupons, but tested at the
factored maximum fatigue loads.
Table 6-9: Comparison between the fatigue life of sandwich beams and tensile
coupons
(The beams and coupons are now compared at equivalent strains. )
Beams Coupons
Undamaged over 400,000cycles over 400,000cycles (average)
2-ply HTC overlap 6800cycles 1000cycles
2-pIy LTC Overlap 10,000cycles 19,000cycles (average)
The number of cycles to failure of coupons and sandwich beams are now seen to agree
better (compare Table 6-7 and Table 6-9). The undamaged beams and coupons are seen
to last over 400,000 cycles (on average). The 2-ply HTC overlap beams and coupons
lasted less than 10,000cycles. Similarly, the 2-ply LTC repaired beams and coupons are
220
6.4.5 The fatigue performance of undamaged and repaired coupons: discussion and
conclusions
The fatigue performance of HTC and LTC overlap repaired coupons was presented in
the previous section. The LTC overlap repaired coupons were seen to be weaker (i. e.
failed at a lower number of cycles) than the HTC overlap repaired coupons. Indeed,
the LTC overlap repaired coupons failed after 400,000cycles when tested at 50% of the
control load, whereas the HTC overlap repaired coupons did not fail even after 2.5
n-dMon cycles. Again, an increase in the maximum fatigue load promoted a rapid
decreasein the fatigue life of the specimens.
Although coupons and beams showed similar static strength, the fatigue performance
of the coupons was superior to that of beams tested at an equivalent load. It was
further seen that, for an equivalent load, the coupons were stiffer than the beams.
Coupons tested at a higher load (corresponding to the coupon-to-beam strain ratio)
were seen to present similar fatigue lives than the sandwich beams.
The overlap repaired coupons failed by delamination of the overlap away from the
parent, and through the adhesive layer. The HTC repair failed by the propagation of a
delamination that has initiated at one end of the overlap. The delamination ran
towards the centre of the joints. The LTC repairs failed from the complete separation of
the parent from the repair patches, i. e. both sides of the overlap delaminated. It may be
expected that the LTC repairs failed very shortly after the delarnination initiated.
The LTC overlap repairs behaved differently compared with the HTC overlap repairs.
Both HTC and LTC repair adhesive are known to be toughened by a second polymeric
phase. Although, their static properties were similar, their fatigue performance are
different. It has been shown in the literature [126] that low-temperature cure adhesives
may possess lower fatigue damage threshold values when compared with high-
temperature cure adhesives, although their static fracture toughness may be identical.
This would mean that, under the same applied load, the LTC adhesives are likely to
fail prematurely when compared with HTC adhesives. Although no explanations were
given in the reference [126] it is noteworthy that the same observation was made in the
present study.
221
6.5 Modelling of the sandwich beams stiffness degradation
6.5.1 Introduction
It was concluded in the previous section that the coupons are a good alternative to FE
analysis for helping to develop sound designs for repairs. Indeed, they appear to have
a good potential for the designing of repairs under static and fatigue loading. It may
now be of interest to look at the fatigue degradation of the coupons and compare this
with the fatigue degradation of the sandwich beams. It is reminded that becauseof the
difficulties associated with the testing of slender laminates in compression, the
coupons were subjected to tensile forces only. Consequently, the modelling work is
undertaken for the beams tested with the TF in tension only.
It has been established that parameters such as laminate stiffnesses can be related to
the damage state and can be measured in-situ without the need for any destructive
testing [94]. This was the approach used in Chapter Five to study the performance of
the repaired beams and it is the approach being used in this section to study the
A stiffness degradation model [127] will be used to fit the stiffness degradation curves
of the undamaged and repaired coupons. A unique equation relating the evolution of
axial stiffness with cycles under an arbitrary load will be determined. This empirical
residual stiffness law will then be used, in conjunction with an analytical strip model
of a beam in four-point bending, to predict the evolution of the beam's deflection, and
hence the bending stiffness, with respect to the fatigue cycles.
6.5.2.1 Introduction
It was recently noted by Clark et al. [128] that, although the modelling of the
222
fatigue modulus concept A similar method, based on a model by Yang et al. [127] is
The Yang et al. model [127] assumesthat the stiffness degradation rate is a power law
function of the number of cycles. Although such an approach is know to be
appropriate for displacement controlled tests [127], it will be applied here to load
cycles. This empirical model will be used in the present work to predict the stiffness
behaviour of the beams in flexure via the modelling of the stiffness behaviour of the
coupons in tension.
where E(n) and E(O) are the axial stiffness after n cycles and the initial stiffness,
respectively. Q and v are random variables that depend on the applied stress level,
loading frequency, etc. The random variables Q and v are to be detern-dned
Q=ai+a2V
v=a3+BS
where al, a2, a3 and B are random variables, that are to be determined from two
different set of experimental curves, and S is the applied load level. It was shown that
223
This equation defines the stiffness reduction with number of cycles under an arbitrary
applied load level S. The development of Equation 6-7 requires only that the stiffness
degradation rate of a material under two different load levels (Equation 6-6) be
measured. Equation 6-7 may then be used to predict the materials degradation rate
under any load ratio.
section (Equation 6-7) to calculate the evolution of the beamýsdeflection under a fixed
load and an increasing number of cycles.
The model must allow the input of the mechanical properties of the skins and the core.
A plethora of techniques have been identified in the literature [97,1301. However, a
simple 'strip' analysis of an infinitely wide beam in circular bending, and under four-
point loading, will be used in the present work [130]. This approach was used for its
relative simplicity and is presentednext.
The model used assumes the same principles as classical engineering beam theory,
with the exception that Poissonýscontractions are taken into account. It is assumed that
the core carries the shear forces and the skin carries the direct forces arising from
bending. The skins and the core of the beam were assumed to be perfectly bonded
together and each section was assumed to bend cylindrically, with the cross-sections
remaining plane and perpendicular to the neutral axis. The skins were assumed to be
very thin so that the local bending stiffness around their own centroidal axis could be
neglected. Furthermore, the core and the skins are assumed not to deform through
their thickness. It is assumed in the present analysis that the mechanical properties of
the TF only will degrade with the number of cycles. The BF, being stiffer and stronger
than the TF, is assumed not to degrade with the number of cycles. Transverse shear
deformations of the core are ignored.
224
4W
d
EI j7 = q(x)
where w is the transverse deflection of the beam of bending stiffness EI. The applied
transverseload, q(x), is a function of the length of the beam only. The sameequation
may be derived for a strip of a laminate composed of an arbitrary number of
orthotropic layers [11] with now:
4W
D, -d Equation 6-8
1 2; 7 = q(x)
where the D, I* term is an element of the plate constitutive equations [6]. The term plate
is now being used because the Poisson's ratio terms are taken into account in the
analyses. The coefficient Dii relates the bending moment to the plate curvatures (i. e.
equivalent to EI in an isotropic plate). The coefficient D,, * may however be seen as a
reduced bending stiffness that allows an unsymmetric Ian-dnate to be treated as
orthotropic, using only a modified bending stiffness in the solution [11]. The
D, 0B2
where the coefficient All and Bil are again elements of the plate constitutive equation.
The coefficient Ail relates the in-plane direct force to the in-plane direct strain (i.e. a
function of the plate axial material properties, Ei, vi2 and G12),and Bli relates an in-
plane force to a curvature and may be seen as a coupling term. It should be noted that
if the coupling increases, the overall bending stiffness decreases.The elements of the
stiffness matrix are calculated, for a laminate composed of n plies, as follows [6]:
Z (-)k(h, El ho)k
Al -'ý - Equation 6-9
I k=l
I-vl2V21
IR
El )k(h, 22 )k
B1,=-'(
21k--f 1- -h0
VI V21
2
1"
El )k(h, 3 3)
DI, 3k-I 1-V12Y21 - ho
k
225
where the kth ply has a Young's modulus E, and Poisson's ratios v,2 and The k", ply
is located within distances Ii, and lio from the centroid of the beam, as showii iii Figwv
6.9 for the sandwich beams studied in this work.
0.6mm
w
+ve
Figure 6.9: Definition of the parameters used in Equation 6-9 for the saiidwicli
beams used in this work
Because every load can be expressed in the form of a Fourier series, aii expressioil for
the four-point loading with respect to the position in the beam, x, may be sought. The
Fourier transform for two line loads, W, located at distances 17and (1;+2(-), as showii iii
Figure 6.10 is [130]:
2N
--------------------
--------------
The equation for the deflection of a beam (at a distance x aloiig a beam of total 1)
is found by successive integration of Equation 6-8. The beam is aSSU1110dSiIIII)I\'
supported on both ends (i.e. at a distance x=O and x=I). The solution to Equation 0-8
with the simply supported boundary conditions was shown to take the forni 11311:
226
q(x)l' M71X
W(X) .4. 2) sin
AIIDII
II
The coupons will be used to obtain a relation for the evolution of the axial stiffness of
the TF with the applied number of cycles (Equation 6-7). The stiffness reduction curves
are then back-calculated from the evolution of the cross-head displacement with
fatigue cycles and hence will include a reduction of all the material properties of the
It is assumed in the present analysis that the TF only undergoes damage. Equation 6-7
will therefore only be applied to the TF layer of the beam, and it will affect the
constitutive equations of the laminate and hence the elements of the stiffness matrix
All, Bil and Dii (Equations 6-9). The elements of the stiffness matrix are now a function
of the number of cycles, n, and of the applied load S. For instance, the expression for
All now becomes:
EI (n, S) EI-
A (n, S) = TF 2 (h ho)TF+ (hl - h), +
- 2
1- v12-TF V12-core
EI-,., h0 )BF
+ 2 (h1-
- V12-BF
227
The expression for the Bli and Dii terms are modified accordingly. The expression for
the evolution of the ram deflection with fatigue cycles is now:
23
wýx,n, W)
A, (n, S) 1 'T(c + S) T=
, W [sin(m7'rE)+sin(m'
-I )]sin
A, (n, S)D, (n, S) - B, (n, 17) II
m , , ,
Equation 6-11
where it is recalled that W is the applied load on the beam and S is the coupon's (i.e.
TF) load ratio. Equations 6-9 and 6-11 were programmed into the windows-based
Results for undamaged and repaired coupons subjected to fatigue tensile loads
ranging from 50% to 80% of the coupons' control load were presented in section 6.4.
The tests were performed under load control and the evolution of the cross-head
displacement with time, and hence with the number of cycles, were recorded during
the tests.
such as the axial stiffness and the Poissonýs ratio [107,1111, which results in a
progressive increase in cross-head displacement. Under load control, an increase in
cross-head displacement may be expected from a decreasein axial stiffness. It may be
shown that, under load control, an increase in cross-head displacement is inversely
proportional to a decrease in axial stiffness. Residual stiffness curves were therefore
calculated by inversion of the cross-head displacement data. The data presented were
normalised to allow comparisons to be made. However, for fatigue experiments in
flexure, the value of the maximum displacement of the cross-head displacement at
zero cycles (the one that needs to be used for the data normalisation) is difficult to
obtain. Consequently, the data were divided by the first recorded value of the
maximum ram displacement.
228
6.5.4 Results
applied load, are presented in Figure 6.11. General comments may be first made
regarding the shape of the curves in Figure 6.11. Two zones may be identified. Early in
the life of the coupons, the coupons' stiffness decreasesrapidly, as a result of a sudden
increase of cross-head displacement, which in turn results from an increase in the
nucleation and progression of internal damage. The rate of decrease in stiffness may
then be seen to decreasesteadily. It should be noted that these remarks are consistent
with the observations made when discussing the degradation of the beams in fatigue
(section 5.3).
Residual stiffness curves of undamaged coupons may be seen in the top graph in
Figure 6.11. The specimens were tested at maximum fatigue load ratios ranging from
50% to 75% of the undamaged ultimate load of the coupons. It may be seen that the
higher the applied load level, the greater the softening, a higher load level induces a
greater amount of damage in the material. At failure, the stiffness has decreased by as
much as 18%.
The second and third graphs, in Figure 6.11, present the same data, but for the 2-ply
HTC and LTC overlap repaired coupons, respectively (it should be noted that the X-
and Y-axes range has been changed accordingly to be able to view the data at their
optimum). The coupons were tested at maximum ultimate load ratios ranging from
50% to 70% of the undamaged ultimate load. Again, it may be seen that a higher load
level induces a greater amount of damage in the material. However, the maximum
amount of softening is different for each group. The undamaged coupons soften the
most, followed by the HTC overlap coupons. The LTC overlap coupons have softened
the least at failure. The residual stiffness curves may be seen to be power functions of
the number of cycles.
229
1.00 Undamaged coupons
0.98
C
0.96 - 60% (S3)
0.94 65% (S2)
7 75%(S4)
x
M
0.92 jF
n
"a 0.90 D0 [7
Q)
(h 14b 13 IT) CP. CIp
13C3 ýP
.
; u.68 -
E
0.86 -
Z H4
00 ** ** * **0100 **
cbý)
0.82 1 ,, *0
00
0.80
111
0.0 5.Ox1O I. Oxio , 1.50 062.000' 2.5x1O' 3.OxlO6 3.500'
Number of cycles
1.00 HTC overlap repaired coupons
0.98
El
cl
0.96 49
-0
w
0.94 50% (HTC4)
m6o 60% (HTC7)
E .0o 70% (HTC10)
z
0% 0.92 -
0
0
0.90 iI I
0.0 5.000 1.0xio 1.500 2.Ox1O' 2.5xl 3,Ox1O 3.500
Number of cycles
1.00 LTC overlap repaired coupons
1.0
UII
0 OU UOUU
013 110 11
E 0.99 U13
,
0
0.98 50% (LTC5)
bo 60% (LTC10)
w o 70% (LTC13)
0.97
E
0 0.96
Z
0.95 11111
0 100,000 200,000 300.000 400.000 500,000
Number of cycles
230
6.5.4.2 Prediction of the residual stiffness curves
It is now proposed to fit the residual stiffness curves presented in Figure 6.11 to the
power-law function presented in section 6.5.2.2 (Equation 6-6 [127]). It has been
established that two fits only are needed to detern-dne the random variables in
Equation 6-7 (i.e. two fits of Equation 6-6 are needed to establish Equation 6-7). The
residual stiffness curves under the minimum and maximum applied load were used.
Equation 6-7 will subsequently be used to predict the residual stiffness curves of the
coupons and the beams under the minimum, maximum and intermediate load ratios.
Yang et al. [127] proposed a method to estimate the best set of random variables Q and
Table 6-10: Empirical parameters used to fit the stiffness degradation curves of the
The experimental stiffness degradation curves, and the resulting fits are shown in
Figure 6.12 for the undamaged, HTC overlap and LTC overlap repaired coupons,
respectively. It may be noted that the curve fitting exercise is purely empirical and
variables other than the ones used could have been employed. However, it is
considered that they describe the behaviour of the coupons reasonably well. It is noted
that better fits may be obtained (i.e. the fit on 50%UFL LTC repairs can be improved)
by changing the corresponding Q and v variables. However, in the present work the
first acceptable fit found was kept for the rest of the analysis.
231
1.00 Undamaged coupons
0.98
0.96
00YO
60%
6 (SX,
(S3)
1E 0.94
0.24)
E(n,
E(n,660)=EO(1-0.003n
0.92
ý1
0.90 F1
"a "- W
'1ý 0.88 0 Lj r) Li ! I-,,,
(1) -0 [1
0.86 0
0
0 75%,(S4)
0 0.84 -E(n, 75)=EO(1-0.02nO-'5)
z Cý
-
0.82 (9
0.80
0.0 5.000 5 l. OxO 1.500 , 2.000 ' 2.5x1O 3.Ox1O 3.5xl06
Number of cycles
1.00 HTC overlap coupons
0.98
,E
Ll r
n r1r.
1
0.96
"0
W 50% (HTC4)
4 0.94
E(n, 50)=EO(1-0.006nO-15)
0
Z 0.92
o 70% (HTC1O)
1)
E(n, 70)=Eo(l -0.02nO-1
0.90
0.0 5.0X10 1.0xio 1.5X10 2.OXIO 6 2.5x1O 3.Ox1O
Number of cycles
0.99 0 El I I:
U
X
ft (1 50% (LTC5)
73 0
0) E=EO(1-0.00lnO-")
Ln -
. 0.98
*0 o 70% (LTC 13)
0 14)
0 E=EO(1-0.0046nO-
z
0.97 11111
0 100,000 200,000 300,000 400,000 500.000
Number of cycles
Figure 6.12:A fit is forced through two coupons degradation curves (Equation 6-6)
232
6.5.4.3 Prediction of the residual stiffness curves
Equation 6-6 was used, in the previous section, to fit experimental stiffness
degradation curves and hence obtain the values of the empirical variables Q and v. A
system of two equations (i. e. the random variables Q and v are unique for each load
level S) was worked out for each group of coupons. The expression that relates the
residual stiffness curve under an arbitrary applied load of each type of coupon may
now be derived (Equation 6-7).
Yang et al. [127] have shown that the variables in Equations 6-6 and 6-7 were linked by
the expressions Q=ai+a2vand v=a3+BS.The variables al, a2,a3and B are to be used in
Equation 6-7 and may be found by solving the systems of four unknowns and four
equations for each group of coupons. For example, the fit through the residual stiffness
data of the undamaged coupons has led to the set of variables Q(60)=0.003 and
Q(75)=0.02,for an applied load level 60% and 75% of the coupons' control load (hence
S=60 and S=75), respectively (Table 6-10). Similarly, values of v(60)=0.24 and
v(75)=0.15 were calculated. The al, a2,a3 and B variables are found by solving the four
equationsin Q and v.
The calculated variables for each group of coupons are presented in Table 6-11 below.
Equation 6-7 may now be used to predict the residual stiffness under any arbitrary
applied load and the results are presented in Figure 6.13. The top graph in Figure 6.13
Table 6-11: Empirical parameters calculated from two residual stiffness fit
Undamaged 2-ply overlap HTC 2-ply overlap LTC
a2 -0.097 -0.067
-0.189
a3 0.6 0.3 0.29
233
Undamaged coupons
E(n, S)=I - [0.048-0.189xO.6+(-0.189) (-0.006)SI
1,00
60% (S3)
0.98 Prediction, S=60
65% (S2)
0.96
,E Prediction, S=65
0.94 0 75%(S4)
0.92 Prediction, S=75
0.90 ýp
'? [fF01rPq: F LIC,
r3,
-4 0.88
M
E 0.86
0 00
z 0.84
% 0 040
0.82 ý9 0*0 * 0* 0 * **
*00*
0.80
0.0 5.Ox1O 1.Ox1O ' 1.5xIO 6 2.Ox1O 2.5x1O 6 IWO ' 3.5X10
Number of cycles
HTC overlap coupons
(n, S)=l-[0.021+0.097xO. 3+(-0.097)(-0.002)S]fn(l). 3-0.
tX)2S)
1.00
0.98
,E
0.96
50% (HTC4)
0.94 Prediction, S= 50
-
60% (HTC7)
E .0
%- Prediction, S= 60
z
0 0.92 -
o 70% (HTC10)
0
Prediction, S= 70
. 0
0.90 !I
0 051.0xio 1 1.5xio 2.000 2.5X10 3.000
.05.00
Number of cycles
w DD mum CM M0 CM Ma
0 013
00 00
0 r
IE 00 0 C)
ýn 0.99 0 01:1 1300
*40,00
x . w
4z
c 50% (LTC5)
0 Prediction, S= 50
R0o 609%(LTC10)
0.98 -
E 0- Prediction, S=60
160 70% (LTC13)
z o
Prediction, S= 70
0.97
f
0 100,000 200,000 300,000 40600 500,000
Number of cycles
Figure 6.13: Calculated residual stiffness degradation curves (using Equation 6-7)
234
It is noteworthy that only one equation is now being used to predict the residual
stiffness of the coupons. Equation 6-7 was derived by only using the stiffness
degradation rate under two different load levels (Equation 6-6). The prediction of the
materials degradation rate was good for each group of specimens, but Equation 6-7
may lead to a better or worse fit than Equation 6-6. For instance, in Figure 6.12
(Equation 6-6) the fit to the residual stiffness data of the undamaged coupons tested at
60%UFL may be seen very good. However, in Figure 6.13 (Equation 6-7) the prediction
to the same data looks a somewhat worse fit, although it is still acceptable.
It was assumed in the present analysis that the residual stiffness curves are power
functions of the number of cycles. However, close to failure the residual stiffness
curves may depart from the fits (see for example the fit of the undamaged coupons
tested at 75%UFL in Figure 6.13). This is due to a rapid evolution of damage in the
skins that are not modelled in the present analysis.
6.5.4.4 Prediction of the fatigue degradation of the sandwich beams tested in flexure
It is now possible to calculate the evolution of bending stiffness of undamaged and
repaired sandwich beams tested in fatigue in four-point bending, by using the stiffness
degradation laws defined by Equation 6-7. Equation 6-7 is used to define the
behaviour, with the increasing number of cycles, of the axial properties of the TF. A
fixed load is prescribed on the beam, and the resulting deflections are calculated. The
same calculation is done for increasing number of cycles. The bending stiffness is then
back-calculated from the deflections (as shown in section 3.3.5.4).
Figure 6.14 presents the evolution of the bending stiffness of undamaged beams tested
at 60 at 65%UFL with the TF in tension, against the calculated ones (i. e. using Equation
6-11 and with S=60 and S=65, respectively). It may be seen that the measured and
in
calculated curves are good agreement. Figure 6.15 shows the data for the 2-ply HTC
and LTC overlap repaired beams. The agreement is reasonably good, at least at
relatively low cycles. In Figure 6.15, the agreement is less good at a high number of
cycles, usually close to failure, where the measured bending stiffness decreases more
than the calculated value. This is due to damage evolution close to failure that
rapidly
is not being modelled.
235
1.00 Undamaged beam in four-point bending
-="
Tý TF in tension
0.9 8-
rh
r/I v
'a , --
_f==P"___
w - :. ý -: 04
-Z _
-
'Z;
0.96
_C 0.94
Number of cyles
0.99 X--
Prediction,
HTC overlap repair- I
C:
-0
CU
0.98
0.97 11111
0 2,500 5,000 7,500 10,000 12,500 15,000
Number of cycles
236
6.5.5 The modelling of the sandwich beams stiffness degradation: discussion and
conclusions
The empirical approach used in the present work has been shown to successfully
predict the fatigue degradation curves of repaired beams from fatigue data obtained
from representative tensile coupons.
However, the applied load on the beam and the load applied to calculate the stiffness
degradation of the TF with increasing number of cycles, were identical. This is
incompatible with the earlier comment that stated that the fatigue lives of the coupons
and the beams were identical only if a higher load was applied to the coupons (section
6.3.3). This was because the coupons were slightly stiffer than the TF of the beams.
From Figure 6.14,it is clear that the predictions would be less sound if the load applied
in the coupons stiffness degradation law were higher than the one applied on the
beam. For example, if a residual stiffness law with S=70%of the coupons ultimate load
is used, in conjunction with a beam loaded at W=60%UFL. The net effect would be an
overestimate of the beards residual stiffness curves, i. e. the beams would be calculated
to soften more than observed.
It is also noted that in the present analysis (Equation 6-11) the reduction of the axial
stiffness of the TF with number of cycles is applied to the entire length of the beam,
whereas in the experiment only the region between the inner loading points damages
significantly. Indeed, the portion of the beam located between the inner and outer
loading points is under a decreasing bending moment, which as a result produces a
decreasing in-plane stress in the skins, that vanishes at the outer loading point (see
Figure 3.14). The present analysis consequently tends to overestimate the reduction in
bending stiffness of the beams with number of cycles.
Furthermore, it was argued in Chapter Five that it was difficult to obtain information
regarding the softening of the beam at the very beginning of the fatigue test.
Consequently and significantly, the initial decrease in residual stiffness is unknown.
Therefore the 'real' shape of the curve is unknown. It is not possible to draw firm
conclusions from the present study, and the exercise should be regarded as a first
attempt. Nevertheless, the method was seen to give good agreement with the
experimental results although further work is clearly needed.
237
6.6 Conclusions on the performance of tensile coupons representative of the
tool face of the sandwich beams
Static test
Flat coupons, representative of the tool face of the beams, were manufactured and
tested in tension, statically and in fatigue. Two-ply HTC and LTC overlap repaired
coupons were also manufactured and tested.
The static failure strength of undamaged and repaired coupons (HTC and LTC) was
parent away from the repair. It was suggested that the different materials properties of
the HTC and LTC systems may be responsible for these differences in the loci of
failure. The static performance of undamaged and repaired coupons was compared
with that of the sandwich beams and they were found to be in good agreement.
Fatigge tests
When the fatigue performance of undamaged and repaired coupons was compared it
was found that the repaired coupons possessed an inferior fatigue performance
compared with the undamaged coupons. A similar conclusion was reached for
undamaged and repaired sandwich beams tested in fatigue (Chapter Five).
The fatigue performance of HTC repaired coupons was seen to be superior to that of
LTC repaired coupons. All overlap repaired coupons were seen to fail by delan-dnation
of the overlap away from the parent, and through the adhesive layer. However, the
HTC repair failed by the initiation and propagation of a delamination that has initiated
at one end of the overlap. The LTC repairs failed from the complete separation of the
parent from the repair patches, i. e. both sides of the overlap delaminated. It was
suggested that the LTC repairs were weaker due to a lower resistance to fatigue crack
growth of the relatively low-temperature cured adhesive system.
The fatigue performance of undamaged and repaired coupons was compared with that
of the sandwich beams and, although they showed similar static behaviour, the fatigue
performance of the coupons was significantly superior to that of the beams. One of the
238
possible reasons was suggested to be due to the symmetric lay-up of the coupons
when compared with that of the lay-up of the TF of the beams. The coupons were seen
stiffer than their equivalent TF skins. Indeed, when the results reanalysed to give
comparable loads, the coupons and the beams showed similar fatigue performance.
Stiffness degladatio
The fatigue degradation curves of the undamaged and repaired coupons were studied.
It was shown that the higher the applied load level, the greater the reduction in
stiffness. A stiffness degradation model was used to fit the stiffness degradation curves
of the coupons. An equation relating the evolution of the axial stiffness with cycles
under an arbitrary load was then determined. This empirical residual stiffness law was
then used, in conjunction with an analytical strip model of a beam in four-point
bending, to predict the evolution of stiffness of the sandwich beams studied in the
present work. The empirical approach was shown to successfully predict the fatigue
degradation curves of repaired beams from the fatigue data obtained from the
Overall conclusions
The static failure load and failure mode of undamaged and repaired coupons,
representative of the TF skin of the sandwich beams, were compared with those of
equivalent sandwich beams, and were found to be in very good agreement. The
fatigue locus of failure of the TF coupons was furthermore seen to be identical to that
of equivalent beams, although the fatigue performance of the coupons was superior to
that of the beams. Overall, the tensile coupons were seen to be a good alternative for
helping to design repairs on thin-skinned sandwich beams.
239
7. The static performance of composite stiffened panels
7.1 Introduction
In the present Chapter the behaviour of undamaged, damaged and repaired I-stiffened
non-buckling at operational loads. The buckling strain of the panels was determined to
be -6000ptrain by Greenhalgh et al. [141 (although the panels tested in reference [14]
were made from a slightly stiffer Hexcel unidirectional prepreg (T800/924) than that
tested here (T300/914)).
central stiffener. In one casethe damaged area was taken to be a 55mm circular area to
the central stiffener and skin. The second type of damage was of the same size but
confined to the skin only, the stiffener being left intact. Two repairs schemes were
evaluated. One involved the repair of the back and front face of the panel, including
the stiffener. The second repair involved the repair of the skin face only.
undertaken using the quasi-3D FE method validated in Chapter Four. The calculated
FE results are compared with experiments. Finally, the behaviour of the stiffened panel
240
7.2 The performance of composite stiffened panels: experimental studies
7.2.1 Introduction
applied strain is c,--A117(with I being the panel length). As soon as the critical load is
reached, the flat form becomes unstable and two shapes are possible; the flat forill arid
a buckled form. Since small imperfections in the panel geometry or aji eccetitric load
path always exist in reality, the unstable form is always preferred, and tlic pailel
deforms in a two-dimensional wavy nature as shown in Figure 7.1. Tlie stabilit), (A tlie
panel has been shown to be dependent on parameters sucli as geoinetry, inaterial's
properties, edge condition and loading [96]. For plates, the buckling load iiidicates a
bifurcation in the load-deflection behaviour, and a change iii stit'hiess caii be
witnessed. However the presence of a 'knee' in the load-deflection curve is depellLjolit
on initial imperfections, and becomes rounded as imperfections increase, as slio%ýýiiiii
Figure 7.1.
L.. d A. initiallY zz
ýt, ight plMý bu, kle,
in a t-din-m. n. 1ýa, v-tu, v
Bifurcation UI,
, UP
UP
till 1)
Dmpl-n-t
N,
r
Five panels were tested in pure compression in the present work (114er tO SOCtioii 3.4.3):
01 panel with a hole through the central stiffener and skin (D2).
01 panel with a hole though the skin underneath the central stiffener (M).
241
7.2.2 The behaviour of the undamaged panel (D1)
The failure load of the undamaged panel was 1200kN for an applied displaceincot It
failure of 3.70mm. The panel collapse resulted from a sudden, explosive, failLire
involving in-plane compression fracture of the skin and the stiffeiiers, and
delamination of the stiffeners away from the skin, as sketched hi Fiý,Uft, 7.2 (cmci
shown in Figure B.1, Appendix B). It is noted that extensive delamitiatioii iii the
stiffeners' spar caps and at the unsupported edges also occurred. From iiispectioii of
the panel after testing, it may be argued that the panel collapse may have beell
initiated by an in-plane skin compressive failure (originating either at the unsupported
edges or at the centre of the panel), which then resulted in the separatioli of' Ole
stiffeners away from the skin. The stiffeners were then left to support all ii1creasilig
load and they ultimately collapsed. However, it is difficult to comment oti t1le process
of fracture, which was so violent that the evidence of the failure process were lost iii
the explosive failure.
In Figure 7.3, the load-displacement curve of the undamaged panel inay be seeti to be
linear up to a displacement of 2.25mm, after which the trace seems to deviate 11.
()11i
linearity. Back-to-back strains from gauges located at the edge of the paiiel, as sll()%N,
ll
in Figure 7.4, were recorded at various locations in the panels and were foulid to be
uniform except near the edges where bending took place. It maN, be seeii iii I-i)'ul-(, 7.3
that the strains on the skin and the stiffener sides of the panels are ideiitical for low
242
1400-
1300-
1200-
1100
1000-
900-
800-
700-
600-
500-
400
300
200
100
0
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0
-9,000
-8,000
-7,000
-6,000
-5,000
4,000
-3,000
-2,000
-1,000
o0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4. C
-f-
SGIý 77.5mm
ý 1
-- 15-
243
7.2.3 The behaviour of the damaged panels (132 and 133)
The failure load of the panel damaged by a hole through the central stiffeiiei- 111dtilt,
1.88mm. The locus of failure for this panel was simpler than that of the widainag0d
panel, since the panel's collapse resulted from an in-plane compressive failiire (it Hit,
skin that originated at the edge of the cut-out, as sketched in Figure 7.5 (aiid slimN,I) ill
Figure B.2, Appendix B). The stiffeners were also seen to have delainiiiated froin Hit,
skin.
compared with far-field strains (SG1 and SG2, as showii iii Figure 7.7) and it lnaý' be
noted that bending of the panel occurred around the damage, whereas it "'as ii1ol-0
limited at the edges of the panel. Indeed, the traces of the back-to-back SG 17 aiid 1;G IS
strains separate to a greater extent than those of the back-to-back SG I and SG2 straiiis.
In Figure 7.6 it may be noted that the strains recorded at 15min froin Ole danw),,c an,
about 2.4 times greater than the far field strains (SG2). It seerns clear that tll('
of the cut-out resulted in an increase in strains in its vicinitv. 'I'lle Strain colicelitt"Itioll
associated with the cut-out may be further seen to decrease as we rnove aw, vv froin it;
the SG16 gauge located 30mm away from the damage recorded strains lo%ver dian tl), It
of the SG18 gauge. This may suggest that failure resulted froni tile liigli str, lills
occurring at the edge of the hole.
1, Iv -p., t :Ith.
In-pldne compressive failure Dd--t... ,, thý M, ff--
I
---
H
Figure 7.5: Representation of the locus of failure of the D2 panel
"44
700- /k, -d- h\
-
600-
500-
400-
300-
200-
'00
0
0.0 0.5 1.0 1.5 2.0 2ý5
-8,000
-6,000
-4,000
-2,000
Figure 7.6: The performance of the panel damaged with a through-hole (1)2)
------------- -
77
SGI
15
15
' -4ý
On the other hand, the panel damaged by a hole through the skill undenicatli tll(,
central stiffener only, (M), was seen to fail at a load of 849kN aiid at aii applie(j
displacement of 2.72mm. The collapse of the panel seems to have resulted from all in-
plane compressive failure that has initiated at the edge of the damage, possibly clue to
high strain concentration there. The stiffeners may then have separated froln t1le skill,
and failed by in-plane compressive failure, as sketched in Figure 7.8 (and sliowii iii
Figure B.3, Appendix B). The panel's load-displacement curve is showti ill Figure 71.9,
and it may be seen to be linear up to failure. It is noted that the test was stoppeLl at a
load of 700kN (therefore about 82% of the failure load) and a ultrasonic hatid-scaii (A-
scan) was used to asses the extent of delaminations around the dainaged area.
Unfortunately, no damage was observed and the test was subSL,quently c(mtkiLled.
However, the process resulted in the loss of the strain readings.
Compressive in-
plane failure - I )eI. luIlITlIu, Il III III! sill
600-
500
0 400-
300-
200.
100-
0-
0 1.5 2-0 2.5 30
0.0
.51.0
Cross-head displacement (inm)
Figure 7.9: The performance of the panel damaged with skin-hole (1)3)
240
7.2.4 The behaviour of the repaired panels (134 and D5)
The repaired panel that was damaged by a hole through the ceritral stiffeiwi- ami tile
underneath skin, D4, failed at a load of 10OOkN for an applied displaceriwi-it oI'
2.91mm. It is noted that the sequence of failure is difficult to obtaiii shice dic repair
separated away from the skin, which resulted in the fracture of the repair patk-11oil tilt,
stiffener side on one hand, and its separation from the panel ori the other.
In Figure 7.11 the load-displacement curve may be seen to deviate from Iiiiearity for
displacement over 1.5mm. Also, in Figure 7.11, the strains in the repair patch and close
to the damage (SG17 and SG18, as shown in Figure 7.12) are compared witli far-field
strains (SG1 and SG2). It may be seen that while bending of the panel has occiii-red, it
was not as pronounced as for the undamaged panel (i. e. the SG'I and SG2 traces iiia).
be compared in Figure 7.3 and in Figure 7.11). It may be further noted that the repair
patch has had a beneficial stiffening effect, since the strains in the repair (i. e. SG 17 aild
SG18) are lower than the back-to-back far field strains (i. e. SG I and SG2), illthOL11,111
they were 2.4 time greater than the far field strains (SG2) for the damaged paliel
(Figure 7.6). The repair patch allowed the panel to fail at a load of 10OOkN, which may
be compared with the failure load of 515kN for the equivalent dainaged paiiel. 'I'lle
present overlap repair scheme therefore recovered an appreciable proporti0ii ()I the
pristine panel strength.
I I.. ' ........... ý -, . 1". 11-- . 1- "
Delamin4tion in the
-----------
-. P, -. -
247
I1 11 1. II\
1200- --.
1100-
1000-
900-
800-
700-
600-
0 500-
400
300-
200
100
0 40-
0.0 0.5 1.0 1.5 2.0 2.5 3.0
Cross-head displacement (mm)
-5000-
1000-
.t -3000-
15mm from the damage
X -2000- SG17 Skin side
- -
- SG18 Stiffener side
-
-1000
0111
0.0 0.5 1.0 1.5 2.0 2.5 3.0
Cross-head displacement (mm)
Figure 7.11: The performance of the panel damaged with a through-hole and
repaired (D4)
'Gr
"1
4-
-k
Figure 7.12: Location of the strain gauges on the D4 panel
248
The repaired panel damaged by only a hole to the skin uiidenwatli the ceiitral
stiffener, D5, failed at 1074kN, for an applied displacement of 3.31inii). Agaiii it is
difficult to comment on the sequenceof failure. However, it seems that the collapse (A
the panel resulted from a compressive failure that initiated at the centre of the repair
patch, as shown in Figure 7.13 (and shown in Figure B.6 and Figure B.7, Appendix B).
The load-displacement and strain-displacement curves of the D5 panel are sliowii iii
Figure 7.14. As for the undamaged panel, the load-displacement curve may be seeii to
deviate from linearity for displacement over 2.25mm. Also, iii Figure 7.14, back-to-back
strains in the repair patch are compared with strains close to the damage (i. e. SG
are compared with SG17 and SG18, see Figure 7.15). Again, and as for tile repaired D4
panel, the repair patch resulted in a decrease in the strain at the vicinity of tile Liamage,
since the strains in the repair (i. e. SG17 and SG18) are lower than tile far fieki strains
(i. e. SG1 and SG2). It may be seen in Figure 7.14 that the SG17 and SG18 traces "jump,,
suddenly down at an applied displacement of about 3mm. This was as a result of all
in-plane compressive failure of the repair patch at that displacement. The crack was
seen visually during the test and was estimated to span the central stiffeller width. 'I'lle
crack then stabilised and the panel carried on supporting the load until total collapse
of the panel.
Again, the repair scheme used has been shown to be successful imd restored ail
appreciable fraction of the original strength of the pariel. The panel's failure load %vas
1074kN, whereas the corresponding damaged panel failed at a load of 849kN.
-WINI
In-planecompressive failure ý Delarninati Ionof the repair patch I )el ", n, n"t, o", ý, tIý, -I, II 'u
249
iiý
- -- Repaired panel (hole through the skin onl.v)
1200-
1100-
1000-
900-
800-
700-
600-
500-
400-
300-
2001
100
0
0.0 0.5 1.0 1.5 2.0 2.5 30 3,5
Cross-head displacement (mrn)
-7,000-
Far field
X SGI Skin side
-6,000- -
___+_ SG2 - Stiffener side
.
13
-5,000-
-4,000-
-3,000-
15mm from the damage
X -2,000- SG17 Skin side
-
SG18 Stiffener side
-
-1,000-
0.1
0.0 0.5 1 2.0 2.5 3,0 35
.01.5
Cross-head displacement (rrim)
Figure 7.14: The performance of the panel damaged with a skin-hole and repaired
(D5)
",G 17:
2K
7.2.5 Comparison of the performance of the stiffened panels
The failure loads of undamaged, damaged and repaired panels are shown in Table 7-1.
The drop-off in failure load from the undamaged failure load is also shown. The
undamaged panel failed at a load of 1200kN. The failure load of the damaged panels
was greatly reduced; the panels D2 and D3 failed at 515kN and 849kN, respectively.
As expected the damage that consisted of a hole through the central stiffener led to the
biggest reduction in strength; the D2 panel lost 57% of the undamaged strength. In
comparison, the damage that consisted of a hole through the skin only resulted in only
a 30% drop-off in strength. It may be seen in Table 7-1 that the present repair schemes
have been successful. The two repaired panels failed at loads above NOUN,
recovering 9">%and 89.5%of the strength for the D4 and D5 panels, respectively.
The load-displacement curves of the undamaged and damaged panels are compared in
Figure 7.16. The load-deflection curve of the pristine panel may be seen to be the
steepest. For the damaged panels, the cut-out has resulted in a change in the panels'
load path that ultimately resulted in a loss in structural stiffness.
Figure 7.17 presents similar curves, but for the repaired panels. The load-displacement
curves of the undamaged and the repaired panels may be seen to overlap. The overlap
repairs successfully restored the stiffness of the panels.
251
Undamaged panel
0 -- Damaged panel (through-hole)
1200- I]--- Damaged panel (hole through the skin onlv)
1100-
1000-
900
800-
700
600-
500
-
400-
300-
200
100
0 jr-
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0
Cross-head displacement (inni)
-- Undamaged panel
0- Repaired panel (through-hole)
1200-
Repaired panel (hole through the skin only)
1100-
1000
900
800-
700-
600-
500-
400
300
200-
100-
0
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0
252
Strains at various locations in the panels were measuredby the application of strain
gauges.It was describedin section3.4.5.2that asmany as 18 gaugeswere usedon each
panel. However, for brevity, only far field strains and strains around the damagearea
are presented here.
The back-to-back far field strains at failure, recorded on one edge of the panels, are
sununarised in Table 7-2 below. The strains SG1 and SG2 were read on the skin and
the stiffener side, respectively. In addition, the average strain (i. e. average of SG1 and
SG2) and the percentage difference in strain (i. e. [ABS(SG2-SG1)/ average strain]*100)
are given. The difference in strain was calculated to give an indication of the out-of
plane bending occurring at the edges.
It may be seen in Table 7-2 that the percentage difference in back-to-back strains
reaches 23.2% for the undamaged panel, whereas it is lower than 10% for the damaged
and the repaired panels. In other words, significant out-of-plane deformations
occurred at the edges of the undamaged panel, but not on the others and this theme
will be discussed in section 7.2.6.
253
7.2.6 Experimental studies: discussion and conclusions
The unsupported edges of the panel were shown to deform out-of-plane and this may
not be surprising since the load path on these panels is complicated. The back-to-back
strains at the unsupported edge of the D1 panel, that were shown in Figure 7.3, are
now compared with back-to-back strain measured at the centre of the panel in Figure
7.18. It may be seen that while some bending took place at the centre of the panel (i.e.
the SG3 and SG4 traces diverge), out-of-plane deformations were maximum at the
panel's edge. This is because the unsupported edges are relatively wide and
unrestrained from bending. It should be noted that it has been reported in the
literature [132] that the unsupported edge are sometimes made narrower, and
consequently more stable to out-of-pIane deformation, so that the region around the
central stiffeners that is under study is under higher stress than the panel's edges.
It is noted that the shadow Moird fringe technique was used to assessthe extent of any
out-of-plane deformation (buckling pattern). However, no major buckling pattern was
seen on any of the five panels. It is therefore not clear if buckling of the panel occurred
in the experiment.However, it is noted that the grid spacing used in this study was
coarse(i.e. 0.4mm.betweeneachline) when comparedwith grid spacingsreported in
the literature (i.e. 0.125mm.betweeneachline in reference[14]), and it hencemay not
have picked up the smallestout-of-planedeformations.It is however clear that some
254
bending took place. In the present case, the panel will typically deform iii two-
dimensional sine waves and, consequently, the strains will vary periodically liaxing
There may be a number of locations in the panel where failure could initiate, and ill a
number of possible ways. It is noted that since the collapse of the panel was explosive,
the location of crack initiation is not known. However, due to the wide 1-111SUPPOIACd
edges, the strains were higher at that location, and failure may have initiated t1lere.
This is consistent with the earlier observation that failure may have resulted froill it
compressive failure of the skin. Failure may have otherwise initiated bet%N,
eell two
stiffeners. Extensive delamination was also observed at the unsupported lwre
ed),,es, ýN,
through-thickness effects may have helped the process of fracture initiatioll. ()tli(, I-
failure modes such as skin-stiffener separation are possible since the stiffeiiers wert,
completely debonded from the skin. In the present case, failure may have initiated at a
crack in the skin, with the initial crack growing perpendicularly to the applied stress
and leading to stiffener debonds, and ultimately collapse of the skin and the stiffeiiers.
_10,000-
-8,000- 3ý 42
-6,000-
-4,000-
Ul)
25 s
7.2.6.2 Damaged panels (D2 and D3)
In Table 7-2, the difference in back-to-back strains at the unsupported edges of the
damaged panels were seen to be about 9%, and hence much less than that of the
undamaged panel. Several reasons may be sought. First, it is noted that the failure load
of the D2 and D3 panel is much less than that of the D1 panel. Secondly, it may be
thought that the presence of the damage resulted in the load being carried mainly by
the two side stiffeners and by the skin in the area adjacent to the damage. The presence
of the damage may have resulted in a change of the load path in the panels.
Consequently, the area near the damage is under a higher stress than the unsupported
edges.
presence of the damage, but to a lesser extent since the central stiffener was left intact.
The load was carried by the three stiffeners, as for the undamaged panel. However, the
area of the skin adjacent to the damage will support a higher load, resulting in an
increase in stress in this area. The stress increase may be understood to be not as high
as for the D2 panel, since the stiffener takes some of the load. Consequently, the panel
fails at a lower load than the undamaged panels, but at a load higher than the panel
damage by a cut-out to its n-dd-section.
256
7.2.6.3 Repaired panels (D4 and D5)
The behaviour of the repaired panels is more complicated. In Table 7-2 the percentage
difference in the back-to-back strains at the unsupported edges was seen to be the
lowest for the repaired panels. Although the stiffness of the panels was comparable to
the stiffness of the undamaged panel in Figure 7.17,it may be seen in Table 7-2 that the
However the stiffening effect of the patches is beneficial since the increased stiffness
helped reduce the strain concentration at the edge of the damage. The patches have
recovered the load-carrying capability of the central stiffener as well as the area of the
skin around the damage. The excessof load, previously carried by the side stiffeners
and the skin around the damage, is now being carried by the repair patch, through the
adhesive layer. The strains around the damaged area were hence seen to be
comparable to far field strains in Figure 7.11 and Figure 7.14. Consequently, the repair
patches recovered the load path in the panels, reducing the excessof stress in the skin
around the damage, and ultimately increasing the failure load. However, the local
increase in material offered by the overlap patches has resulted in a local increase in
While the sequence of failure may be complicated for the D4 and D5 panels, it was
observed that the panels may have failed by an in-plane compressive failure of the
skin that was initiated in the repair, possibly at the edge of the damage. Therefore, and
inasmuch as for the damaged panels, it may be advanced that failure initiated when
the strain in the vicinity of the damage reached the failure strain of the material.
Although the repair was seen to reduce the strain concentration around the damage, it
may be thought that the interaction of the repair plugs used in the present repair
scheme and the wall of the cut-out, resulted in a local increase in strains around the
cut-out. The failure load was then in proportion to the strain concentration at the edge
of the damage.
257
7.2.7 The performance of stiffened panels: conclusions on the experimental studies
The static strength of the undamaged, damaged and repaired panels were much as
expected. The repair schemes employed recovered 83% and 90% of the undamaged
strength for the through-hole and the skin-hole damage, respectively.
The load path for the undamaged panel is such that the load is carried by the skin as
well as by the three stiffeners. Large out-of-plane deformation were seen to occur at
the wide unsupported edges of the undamaged panel, whereas they were limited in
the skin bay. There is a number of locations in the panel where failure could have
initiated, and in a number of possible ways. Identification of the precise failure
mechanism was, however, difficult since the collapse of the panel was explosive.
However, from visual observation of the panel after testing and analysis of the
experimental data, the panel collapse has initiated from a skin in-pIane compressive
failure that led to stiffener debonds, and ultimately to the collapse of the skin and the
stiffeners.
The panel damaged by a through-hole lost 57% of its strength. By comparison, the
damage that consisted of a hole through the skin only resulted in just a 30% drop-off in
strength. The presence of the cut-out changed the load path in the panels significantly,
resulting in the load being carried mainly by the two side stiffeners and by the skin in
the area adjacent to the damage. Failure may have resulted from the high strains
occurring at the edge of the cut-out.
The repair patches were shown to have successfully recovered the load-carrying
capability of the damaged panels. The present repair scheme recovered the load path
in the panels, reducing the excess of stress in the skin around the damage and
ultimately increasing the failure load of the panels. However, it was also discussed
that while the stiffening effect of the patch is beneficial, since it reduces the strain
around the damage, it also increases locally the bending stiffness of the panel. A repair
patch that maximises the membrane stiffness and minimises the bending stiffness may
be better than one where both stiffnesses are maximised.
258
7.3 I-stiffened panels: predictive modelling
7.3.1 Introduction
The quasi-31) FE approach, that uses composite shell elements, was established in
The undamaged panel (D1) was first modelled, and two different analyses were done.
In one case the response of the material and the panel was assumed linear elastic,
whereas in the other the panel's geometric response only was assumed non-linear
elastic. One damaged and one repaired panel were subsequently modelled, using the
same approaches. The panel damaged by the through-hole only (D2) was modelled in
the present work. The equivalent repaired panel (M) was also modelled. The overlap
repairs were modelled using the modified shell element method presented in section
4.5.2.1and described in reference [87].
are compared with experimental results. The stress resultants in the panels are then
subsequently used in conjunction with the Tsai-Hill criterion to estimate the failure
load of the panels.
7.3.2.1 Introduction
The FE meshes were created using the LUSAS graphic pre-processor. The models
included the skin and the three stiffeners, although no efforts were made to model the
aluminium. C-sections in which the ends of the panels were potted. The total length of
the panels was however included in the geometry and the effect of the C-section on the
loaded edges of the panels was modelled by preventing appropriate displacement or
The skin and the stiffeners were modelled using quadrilateral, linear, thick shell
elements (QT54). The QTS4 elements are capable of modelling composite properties as
259
well as geometric non-linearity. For these elements five degrees of freedom (tilree
translations and two in-plane rotations) are usually associated with each 11ode.
However, these elements may be used for modelling intersecting shells (i.e. stiffeiled
shell structures) where the shells may be connected to 3D beam elenieiits and in this
instance six degrees of freedom are automatically defined. Some simplificatioll have,
however, to be taken into account in order to model the skin-stringer attachment.
Figure 7.19 shows a cross-section of the skin-stringer, as well as its FE idealisation.
Zone 2:
Dummy Layer
Tapered foot lay-up
Skin lay-up
Figure 7.19: Cross-sectional view of the skin-stringer attachment and its equivalent
The panel's cross-section was separated into three zones. The first zone c0iisisted of
the skin of the panel, whereas the second consisted of the skin and the tapered foot Of
the stiffener. Finally, a third zone that consisted of the skin, the tapered foot and th('
base of the stiffener's I-section was defined. Becauseof the different thickness and lay-
260
up of the sections an appropriate number of 'dummy' layers was added in zone 1 and
zone 2 to make up the maximum number of layers; the numbers of layers staying the
same across the entire model. The 'dummy' material layer had very small stiffness
properties so that they would not influence the behaviour of the structure.
It was shown in section 7.2.5that significant bending of the skin may occur under load.
However, since in a shell element the nodal plane may not correspond with the
bending plane for the structural element being modelled, an eccentricity may be
defined. In the present work it was assumed that the centroid of the section defined in
zone 3 is the bending plane of the structure; an eccentricity ei and e2was defined for
the sections in zone 1 and 2, respectively (i.e. ei=0.915mm and e2=1.83mm),as shown
in Figure 7.19.
The two triangular sections in the stiffeners that are seen in Figure 7.19, and made of
strips of rolled unidirectional CFRP prepregs, commonly called 'noodles', were also
modelled. The noodles were modelled as triangular sections using 3D beam elements
(BTS3). It is noted that the BTS3 elements do not support composite properties, and
they were therefore modelled as isotropic solids. The beams were placed so that they
connect the skin and the top of the stringers to the web of the stiffeners as shown in
Figure 7.19.
The skin and noodles are made from a T300/914 fabric with the mechanical properties
The boundary conditions on the potted ends were idealised so as to prevent all
displacements, except the vertical applied displacement. Rotations at the potted ends
were also prevented. An applied displacement was applied to one end of the panel,
whereas the other end was fully constrained.
A linear elastic and geometrically non-linear elastic analysis were performed. The
261
7.3.2.2 Undamaged panels
The mesh used for the undamaged panel is shown in Figure 7.20. Twenty four
elements along the length and twenty four elements across the width of the panels
were necessary.The stiffeners were modelled with two elements along the depth of the
web and two elements across the width of the spar cap. It is noted that the area
highlighted at the edges of the mesh shown in Figure 7.20 represents the area to which
the boundary conditions defined above were applied.
geometry and the mesh spacing has been altered appropriately to take into account the
cut-out. The majority of elements for the skin and the stiffeners were quadrilateral
elements (QTS4), with a few three-node triangular elements around the hole. The mesh
used for the damaged panel is shown in Figure 7.21.
undamaged panel. The two side stiffeners were unaffected by the damage and the
repair, and they were therefore modelled as was shown in Figure 7.19.
However, the lay-up of the panel around the central stiffener had to take into account
the repair patches. The panel was again separated into zones as shown in Figure 7.22.
Becauseof the different thickness and lay-up of the sections an appropriate number of
'dummy' layers was added in the different zones to make up the maximum number of
layers. The offset in the local neutral axis introduced by the patch was modelled. The
material that comprises the patch was assumed to have properties identical to the
parent materials, a defined in Table 3-5.
It is noted that, in Figure 7.22, the area affected by the repair patch (on the stiffener
is
side) shown highlighted. The mesh for the repaired panel was identical to the mesh
for the undamaged panel.
262
boundmv
dpplied
263
Spar Cap:
--- [(+45/45/0)2, j
I-section lay-up:
[(+45/45/0)2_, ] Repin, to the Miffen, r
Figure 7.22: Cross-sectional view of the skin-stringer attachment oil the repaired
264
7.3.3 Stiffness predictions
geometrically non-linear elastic analyses are shown in Figure 7.23. The deforilled
meshes were magnified at an applied displacement shown in brackets for eacli of t1w
panels. For instance, the deformed mesh from the linear elastic analvsis of tlie
seen in Figure 7.23 that the linear elastic analysis predicts that the initially straiglit
undamaged panel deforms in a complex manner and, whereas the central part of Hie
panel remains fairly flat, deformation naturally occurs at the unsupported edges Llue
to the complex load path in the panel. On the other hand it may be seen that Hie non-
linear analysis of the undamaged panel resulted in a different deformed inesli ffiaii for
the linear analysis. The unsupported edges and the skin bays (i. e. the skin betweeii t1le
stiffeners) are seen to deform in a two-dimensional wavy pattern. A total of foLir 11,111-
wavelength may be seen to have developed in the panel. The wisupported edges aiid
the skin bay are predicted to buckle between the stiffeners, which have relilailled
essentially undistorted.
ti(
Ani)
ist iv
III)
265
The load-displacement curves obtained from the linear and non-linear elastic FE
analyses are compared in Figure 7.24 with experiment. It may be seen in Figure 7.24
that the non-linear results show good correspondence with the experimental load-
displacement curve. Only when the applied displacement approaches the failure
displacement are the curves seen to diverge slightly, and this may be due to
progressive damage developing in the panel that was not modelled in the present
analysis. On the other hand, it may be seen that the linear elastic analysis yields a
stiffer prediction. The experimental and linear elastic curves correspond for small
applied displacement, i. e. below 0.5mm, after which they separate. The
correspondence of all the curves at low applied displacement was to be expected since
non-linear geometries and damage evolution are only expected to have an effect at
higher applied displacements.
The experimental back-to-back far field strains at the edge of the panel (SG1and SG2),
that were shown in Figure 7.4, are compared with strains obtained from the FE
analyses in Figure 7.25. The strains on the skin and stiffener sides are shown for
comparison. It is noted that the averaged FE strains were for an element located
130mm from a loaded end, and 15mm from an unsupported edge (i. e. the location of
the SG1 and SG2 strain gauges). The prediction from the linear and the non-linear
elastic analysis are compared.
It may be seen in Figure 7.25 that the linear elastic analysis led to predictions that are
in good agreement with the corresponding experimental strains. The strains computed
on the skin side are however seen to diverge from the SG1 strain when approaching
the failure load. The predictions are nevertheless still very good. On the other hand,
the non-linear elastic analysis leads to predictions that diverge significantly from the
experimental results. Although the experimental and computed strains are comparable
at low applied displacement, the predictions diverge above 2.6mm applied
displacement, the non-linear elastic analysis indicates that the panel buckled at about
2.6mm applied displacement. Such deviation in strains were however not observed
266
Undamaged panel
1400-
0- Experimental (D1)
1300- .
FE- Linear elastic analysis
1200- 0,
. - -, 0--- FE- Non-linear geometric analysis
1100- 0
00
1000-
900-
800-
700
600
500-
400-
300-
200 -
100
0
......
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0
Cross-head displacement (min)
Undamaged panel
-12,000-
(D
Far field strain (DI) CD
_10,000- x- SG1 (Skin side) (D
CD,
SG2 (Stiffener side) CD
CD
-0--- Linear elastic (SGI) 9i
-8,000- (D
Linear elastic (SG2)
0 Non-linear elastic (SG1) xx
-6,000- ---CD Non-linear elastic (SG2)
-4,000-
-2,000-1
0 *---1 I.,
0.0 1.......1.0
0.5 1.5 2.0 2.5 3.0 3.5 4.0
267
ln Figure 7.25, the computed back-to-back far field strains from the non-linear analysis
were seen to diverge significantly from the equivalent experimental results. It 111aý,
be
interesting to look further at the strain distribution predicted by the jioii-litwar
analysis. In Figure 7.26, the calculated axial strain distribution, froin tli(, stiffeiier side
and the skin side, along the length of the unsupported edge of the undaiiiaged pailel is
shown. The axial strains may be seen to vary periodically along the unsupported
length of the panel, and in sympathy with the buckle mode shape seen in Figure 7.23;
the strain being maximum at the buckle peaks. It is noted that the strains oil tile
stiffener and the skin side are asymmetric, and can be compressive or tensile. 'I'llese
may be seen as a result of the local curvature induced by the buckle mode sbape.
Indeed, the periodic variation of the deformation leads to a periodic variatioll of tile
curvature of the skin. As a result, the strains on the stiffener and the skiii faces (i. (,. at
each extremity of the shell) will either be increased or decreased from the avei-aged
applied strain (i. e. about -9000ýtstrain at the potted end).
2,500-
0-
Cu -2,500-
-5.000-
Cu
-7,500-
-10,000 -ý
Cu
x -12,500-
-15,000
-17,500
-20.000-.
0 50 100 150 200 250 300 350 400 450
Distance from a loaded end (mun)
Figure 7.26: Computed axial strain distribution along one unsupported edge of the
It is noted that, since strain gauges measure strains averaged along the imlin leii,ý,jli ot
the gauge, there will be a limit in the use of strairi gauges to represent accLirately tlie
strain field. Indeed, the gradient in strains along the buckled edý,e inav he secti Inip,,
and this may lead to differences when comparing calculated FE straiii ()utl)Llts %vitll
experimental strains, as was shown in Figure 7.25. The accurate location of the stram
268
7.3.3.2 Damaged panel
The deformed meshes from the linear elastic and non-linear elastic analyses of tile
damaged panel are shown in Figure 7.27. First, it may be seen that the deforined
meshes, whether from a linear or a non-linear elastic analysis, are ideiitical. I lic
damaged panel is, however, not predicted to deform in the same way as the
undamaged panel did. As might be expected, having been cut, the central stiffeller
may be seen in Figure 7.27 to be of a limited use in supporting the bay skill. The colitral
part of the panel is seen to have deformed out of the plane. This behaviour inaN,be
thought to account for the marked drop in stiffness and in failure load recorded ill tllL'
It
experiment. may be thought that the load path for this panel is directed aroullLi the
cut-out and towards the side stiffeners, which are nevertheless predicted to remaill I-lilt
and not significantly distorted. The only visible difference betweeii the two deforilled
meshes in Figure 7.27 is the magnitude of the deformation; the dainaged pallel beilig
predicted to deform to a greater extent in the non-linear analysis.
t ic
2o9
The linear elastic and the non-linear elastic responses for the damage panel are
still very good between the linear elastic FE and experimental responses, although the
curves diverge when approaching the failure displacement. It is recalled that, in Figure
7.27 the deformed meshes of the damaged panel showed that only minimal
deformation occurred at the unsupported edges.The panel was not predicted to buckle
and this may explain the good agreement between the FE and experimental results.
It was shown in Figure 7.6 that the experimental strains recorded at 15mm from the
cut-out (SG18) were 2.4 times the far field strain (SG2). For clarity, the experimental
and FE strains on the stiffener side only are compared in Figure 7.29.The calculated far
field strains were read unaveraged at a node at 130mm from a loaded end and at
15mm from an unsupported edge. Similarly, the calculated strains that correspond to
SG18 (i. e. the gauge close to the cut-out) were read unaveraged at a node located
15mm from the edge of the cut-out.
It may be seen in Figure 7.29 that the linear and non-linear FE and the experimental far
field strains (SG2) are in very good agreement. However, the choice of analysis (i.e.
linear or geometric non-linear) has had an influence on the prediction of strains near
the cut-out. The experimental (SG18) and linear FE strains seem to agree for low
applied displacement, however the curves diverge significantly above 0.5mm
displacement. The damaged panel was seen in Figure 7.27 to deform significantly at its
centre, which resulted in the increase in strain around the damage. It may be thought
that the linear analysis is inadequate in this case because of the great deformation
occurring at the centre of the panel. Indeed, the non-linear FE analysis may be seen, in
this case, to lead to the best predictions, although it is noted that the strains at failure
are over-predicted significantly.
270
Damaged panel
700
Experimental (D2)
600 FE - Linear elastic
0-- FE - Non-linear elastic
500
400
300
200
100
off
0 III __T__ I
0.0 0.5 1.0 1.5 2.0 2.5
Cross-head displacement (nim)
Damaged panel
-12,000- A
Strains 15mm from the damage
.
Stiffener side
-10,500
- SGI 8 (D2) A
-9,000 -- Linear elastic A
- Non-linear elastic
-7,500 -
Far field strains - Stiffener side
-6,000 SG2 (D2)
0 Linear elastic
1)
-4,500 -0- Non-linear elastic -0-10
x
-3,000-
-1,500 -
0,
II
0.0 0 2.0 2.5
.51 .01.5
Cross-head displacement (nim)
1/11
7.3.3.3 Repaired panel
The deformed meshes for the repaired panel may be seen in Figure 7.30. For clai-ity Ow
position of the repair patch has been highlighted (darker mesh). 'I'lie dC101'111,16011
i"'
more complex than for the undamaged and damaged panels. The portioil of'tlic p, iiiel
covered by the overlap repair patches (i. e. the region highlighted by the darker iiiesli )
may be seen to remain fairly flat. The extra thickness provided by die patclies (i. e. olle
patch was used on the stiffener and the skin sides) may be seeii to SttlbiliSO Ole L-OlItIV
of the panel. The portion of skin bay between the potted end an(] die repair patcli Illay
be seen, however, to deform significantly out of the plane; four 'pockets' inav be seell
to have formed at these regions. However, these 'pockets' do not result froin [-)Ucklillý',
deformation, but rather from the transition between the thick, patched area, to t1w
edge of the repair to the skin and leads to severe out-of-plalle deforinatioll tilere. Ille
only noticeable difference between the deformed meshes from the linear d1ld 11011-
linear analysis may be seen at the unsupported edges of the panel. 'I'lle wisupported
noted however that only three half-wavelengths are now pret I icted.
I ., war elast ic
(05 at Minn)
'72
The linear elastic and the non-linear elastic FE responses for the repaired panel are
result crosses the calculated FE responses for applied displacement between 0.5 and
1mm. It may be thought that events, such as the cracking of the potting compounds
(i. e. at the loaded ends) or the release of internal stresses are responsible for the s-
It was shown in Figure 7.11 that the strains in the repair patch (e.g. SG18) are lower
than the far field strains (e.g. SG2) since the excess material provided by the patch
protects the damaged area and results in a decreasein strain in the vicinity of the cut-
out. The calculated FE and experimental strains, in the far field region and over the
repair patch, are now compared in Figure 7.32. For clarity, the strains on the stiffener
face only are represented. The strains were read as indicated in section 7.3.3.2.
In Figure 7.32 the prediction of the far field strain of the linear elastic analysis may be
seen in good agreement with experiment. The prediction for the non-linear elastic
analysis, however, is not so good. The curve may be seen to diverge from the
experimental results. It is recalled that the deformed mesh of the repaired panels
(Figure 7.30) showed the unsupported edges of the panel to have buckled. The
difference in strain may be seen as a result of this prediction, as was discussed in
section 7.3.3.1.
Also, in Figure 7.32 it may be seen that the predictions for the strains in the repair are
calculated as lower than the measured strains. This may be as a result of the modelling
approach. In section 3.4.3 it was shown that the repair patch was graded in size to
avoid an excessive development of peel stressesat the overlap edge. The patch was
however ideaIised in the FE model with a constant thickness. As a result, the model
leads to prediction stiffer than experiment. It is noted that a finer idealisation of the
is
repair patch possible, although at the expense of complexity of the model.
273
1400-
Repaired patiel
1300
1200
1100 9 Experimental (D4)
1000- *-- FE-Linear
0-- FE - Non-linear 0
900 - 0
'ea-0-
800-
700
600-
500-
400
300
200-
100
0
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5
Repaired panel
-7000-
Far field strains - Stiffener side
-6000 - SG2 (D4)
Linear-elastic - Far field #10
field V
-0- - Non-linear elastic - Far
-5000 -
X
-4000 -
0
-3000-
Strains 15nini from the damage
x
(over the repair patch)
-2000- in
m SG 18 (1)4)
V Linear elastic
_1000-
V Non-linear elastic
0
111
0.0 0.5 1 2.0 2.5 3.0 3.5
.01.5
Cross-head displacement (nim)
274
7.3.3.4 Stiffness predictions: discussion and conclusions
In this section the deformed meshes, load versus displacement and displacement
versus strain responses of the FE panels are compared with experiments. The purpose
of the section is to compare the output from a linear elastic FE analysis with that from
a geometrically non-finear elastic FE analysis.
Deformed meshes
It has been shown in Figure 7.23, Figure 7.27 and Figure 7.30 that the initially flat
panels deformed in a complex manner that depends on the internal load path, and on
the type of analysis. For instance, the deformed meshes of the linear FE model of the
undamaged and repaired panels were predicted to remain fairly flat although limited
out-of-plane deformation were seen to occur at the unsupported edges. On the other
hand, the geometrically non-linear analyses predicted the undamaged panel to buckle
for applied displacements over 2.6mm. It is noted that these results agree with the
edges, with the skin bay remaining flat, perhaps as a result of the excess material
introduced by the overlap patches that locally provide a support against buckling. The
deformation of the damaged panel was more complicated than that of the undamaged
panel and it was not predicted to buckle. Instead, the deformed linear and non-linear
FE meshes showed that the central part of the panel deform out of the plane.
The main difference between the linear elastic and the geometrically non-linear
275
is mainly as a result that the linear analysisassumesthat the deformation of the panel
is small enoughto not significantly affect the overall behaviourof the structure.On the
other hand the geometrically non-linear analysis models the changing effect of
structural deformation on the structural stiffnessand it may be thought to provide a
complete responseof the structure at all stagesof the analysis. Local and global
buckling modescan be predicted
The panels studied in the present work are subject to significant change in geometry
under load as shown by the deformed meshes of the panels, and it therefore seems
necessary to take into account the associated geometric non-linearity, as discussed
next.
could be equally modelled by both linear elastic and non-linear FE analysis, it has been
shown that the former in
could some cases lead to poor predictions of the strain field.
For instance, in Figure 7.25 (i.e. the undamaged panel) the non-linear FE model led to
far field strain predictions in slightly superior agreement with experiment than those
not predicted by the linear analysis used in the present work. In that particular case,
the agreement between the linear FE and experiment was however acceptable.
However, it was furthermore shown in Figure 7.29, that the use of the geometric non-
linear analysis was absolutely necessary to accurately predict the strains around the
cut-out of the damaged panel. Indeed, the magnitude of the strains predicted by the
linear FE model of the damaged panel were severely under-predicted near the cut-out
(see Figure 7.29). Only the non-linear FE model could calculate the correct strain
distribution, because geometric non-finearity needed to be taken into account in this
case.
276
Conclusions
The choice of the correct method of analysis is seen to depend on the kind of
information that is being sought. The linear FE analysis is powerful, because it is
simple and relatively inexpensive to run. Moreover, the calculated load versus
displacement and displacement versus strain responses showed good agreement with
experiments. The method, however, has its limitations. It cannot predict buckling and
strains in regions of great out-of-plane deformation. The linear FE strains at the edge of
the cut-out of the damaged panel were severely under predicted, the geometrically
non-linear analysis is necessary then. It may be concluded that both approaches are
able to accurately predict the stiffness of the panels although limitations exist for the
linear FE model. The second requirement of a model is however to be able to predict
the failure load and this theme is developed next.
277
7.3.4 Prediction of the failure displacement
7.3.4.1 Introduction
There exists a number of locations in the panels where failure could have initiated. For
instance, failure could have resulted from the separation of the stiffeners from the skin
The present panels are not therefore considered to buckle to a great extent, and in-
plane material failure, such as the failure of the skin or of the stiffeners, may be
thought to precede the separation of the stiffeners from the panel. It is proposed in this
section to estimate the failure displacement of the undamaged, damaged and repaired
panels from the FE results. The shell elements used in these models can identify layer
orientations and they can therefore calculate results such as strain and stress fields for
each layer that composes the shell. However, the runs were carried out on a PC with a
limited amount of memory and it was not possible to output the results for each layer.
It was nevertheless possible to read the stress resultants for each node. Stress resultants
were therefore used, in conjunction with laminate analysis software, to calculate the
equivalent stresses in the structures. The calculated stresses were then used in
conjunction with the Tsai-Hill criterion already used in the study of the sandwich
beams (Equation 4-1) to predict failure. The Tsai-Hill criterion is:
222
al al C2 ci (712
' S2 , S2 >I at failure Equation 7-1
S2 S2
112 12
278
7.3.4.2 Undamaged panel
Contour plots of stress and moment resultant in the loading direction for the
undamaged panel are shown in Figure 7.33 and Figure 7.34, respectively. The contour
plots were computed from the geometrically non-linear analysis at the failure
displacement.
It may be seen in Figure 7.33 that the in-plane load in the undamaged panel is
predominantly carried by the stiffeners and the skin bays (i. e. the yellow and green
contours). It is noted that the stiffener foots seem to carry the highest load (i.e. the blue
contour in Figure 7.33). However, fl-ds may be seen as a result from the increase in
thickness provided by the stiffener foot over the skin. The foot and the skin carries
more load than the skin alone, and this is shown as a higher axial resultant at that
location. The unsupported edges may be seen to carry little load (i.e. red contours),
since buckling reduces locally the axial stiffness of the skin. It is also noted that the in-
plane stress resultant carried by the skin bays is a minimum at their centre as a result
of buckling (i.e. orange contours).
It may be seen in Figure 7.34 that the stiffeners carry virtually no bending moment in
the axial direction, the stiffeners were not predicted to deform out of the plane. The
contour plot of the moment resultant in the axial direction may however be seen to
vary in sympathy with the buckle mode shape. Extrema in moment resultant may be
seen at buckling crests; the moment resultant being either maximum (i.e. the blue
contours) or nil (i.e. the red contours).
The stress resultant and moment resultant alone cannot determine failure, and it is
necessary to examine all resultants carefully (i.e. in-plane shear and twisting moment).
However, it has been argued in the experimental studies (e.g. see section 7.2) that the
collapse of the undamaged panel may have resulted from an in-plane compressive
failure of the skin. The latter seems to correlate with the calculated results since it may
be seen, in Figure 7.33 and Figure 7.34, that the non-linear analysis predicts that the
maximum in-plane stress and moment resultants are carried by the skin at a buckle
crest. Therefore, the possibility of in-plane skin failure at a buckle crest will be assessed
in the present work.
279
CC141*CURSOT N.ý
-3231 28
-3(XI -2
-2651 13
-2661 03
-2cr %
-22909
-2(RO
-1TO 73
.471068
-1 MOA
-4330 m
4140.45
-9M3n
a
-7to 301
4M226
-%w13
Figure 7.33: Contour plot of the stress resultant (N/mm) in the loading direction for the
CONTOURS Ctr Mx
409.11
-111774
111501
-IL1365
-Q1' 202
ý810917
-7m 533
.609188
-50b w
-403439
M 094
-202729
401 365
0
101365
Figure 7.34: Contour plot of the moment resultant (N) in the loading direction for the
280
The stress resultants, at an applied displacement of 3.7mm (i.e. the experimental
displacement at failure), taken in the skin bay are shown in Table 7-3. Results from the
linear elastic and non-linear elastic models are compared. The stress resultants were
read at the centre of a skin bay and at a buckle crest where the stressesare likely to be
maximum due to the large bending deformation. For comparison the stress resultant in
the linear elastic analysis were read at the same location.
The laminate analysis software LAP [101] was then used to calculate the stressesin a
laminate, having the lay-up of the panels' skin, and subjected to the stress resultant
shown in Table 7-3. The skin lay-up and materials' properties used were defined in
Table 3-5. The stress field, calculated with LAP for each ply, was subsequently used in
conjunction with the Tsai-Hill criterion (as defined by Equation 7.1) to predict failure.
When stressesand strength (i.e. seeTable 3-5) where used in Equation 7.1, the criterion
predicted failure of the +45* outermost ply for both type of analysis, as shown in Table
7-3. The results accord with the experimental observation. Furthermore, the
displacement at failure may be seen on good agreement whether calculated from the
linear elastic or the non-finear elastic analysis.
Table 7-3: Computed stress resultants for a node across the mid-length and in the
skin bay of the undamaged panel, 3.7mm applied displacement (node 5606)
281
7.3.4.3 Damaged panel
Contour plots of stress and moment resultant in the loading direction for the damaged
panel are shown in Figure 7.35 and Figure 7.36, respectively. The contour plots were
computed from the geometrically non-linear analysis at a displacement of 2mm.
In Figure 7.35it may be seenthat the presenceof the cut-out significantly changesthe
load path in the panel. The central stiffener may be seento be ineffective in carrying
the axial load (i.e. the red contours),which is being carried by the side stiffenersand
the skin around the damage(i.e. the green and yellow contours).The in-plane stress
resultantis maximum at the edgeof the damage(i.e. the blue contours).
Furthermore, in Figure 7.36 it may be seen that the moment resultant carried by the
damaged panel is maximum or minimum at the edge of the cut-out. A positive
maximum value may be seen at the edge of the cut-out and at the mid-length of the
panel. Similarly, a negative maximum value may be seen at the edge of the cut-out, but
at approximately 450 from the loading direction. It is further remarked that the
presence of the cut-out results in an increase in moment resultant carried by the side
stiffeners, whereas they were shown not to carry any in the undamaged panel.
Again, the axial in-plane stress and moment resultant alone cannot be used to estimate
the locus of failure of the damaged panel. However, it was shown in the experimental
studies (e.g. see section 7.2.3 and Figure B.2 in Appendix B) that the collapse of the
damaged panel may have resulted from an in-plane compressive failure of the skin at
the edge of the cut-out. Failure may be expected to occur at the edge of the damage
due to the high stresses present (i.e. the red contours in Figure 7.35). The failure
displacement assuming the aforementioned locus of failure is calculated next.
282
40NTOURSOF Ný
-3159.5
-2W. 82
-2)08.14
-249246
-2256.79
-2(131.11
-1605ý43
-1379 73
-1354,07
-1128,39
-9C2.714
-677.036
-41 357
-225679
0
225679
Figure 7.35: Contour plot of the stress resultant (N/mm) in the loading direction for the
damaged panel (non-linear elastic analysis, applied displacement 2mm)
CONTOURS OF Mý
-136769
-1ýl 5 72
-1063.76
N-I
-911.792
-759,627
.6C7.861
455.896
-aco.931
0
151.965
303.931
455.8%
607.861
Ný 739.8Zr
\_
s \ '\
Figure 7.36: Contour plot of the moment resultant (N) in the loading direction for the
damaged panel (non-linear elastic analysis, applied displacement 2mm)
283
The method used to estimate the failure displacement of the undamaged panel is now
used to estimate the failure displacement of the damaged panel. Results from the
geometrically non-linear analysis only are used since the agreement in linear FE axial
strain near the cut-out was poor when compared with experiment (see Figure 7.29).
The stress resultants at the edge of the cut-out at an applied displacement of 1.88mm
(i. e. the applied displacement at failure) are shown in Table 7-4.
The stress field for each ply of a laminate having the skin's lay-up and subjected to the
stress resultant shown in Table 74 was calculated using LAP [101]. The calculated
stresses were then used with the Tsai-Hill criterion to predict failure. Failure of the
outermost 0* ply was predicted to occur first, although at a much lower displacement
than that seen in the experiment. The calculated Tsai-Hill index is 1.93 and assuming a
linear elastic relationship in the load-displacement response (which agrees with
experiment, see Figure 7.6) the predicted failure displacement is 1.35mm (i.e.
1.88mm/SQRT(l. 93)). It is recalled that the experiment failure displacement is 1.88mm,
the FE failure displacement is about 28% lower than experiment.
One possible reason for the difference may be the coarse mesh used to model the cut-
out. It was seen in Figure 7.26 that the FE model over-predicted the strains at failure at
the edge of the cut-out. A better agreement between FE and experiment may have been
found if mesh refining had been done or if higher-order elements are used around the
cut-out. The subject has not been investigated further in the present work and will be
discussed in a later section.
Table 7-4: Computed stress resultants for a node at the edge of the cut-out of the
damaged panel, 1.88mm applied displacement (node 1349)
284
7.3.4.4 Repaired panel
Finally, contour plots of stress and moment resultant in the loading direction for the
repaired panel are shown in Figure 7.37 and Figure 7.38, respectively. The contour
plots were computed from the geometrically non-linear analysis at a displacement of
3mm.
In Figure 7.37 the load path in the panel may be seen to have been restored, although
the present repair scheme may be seen to have slightly over-stiffened the skin bays, as
a result from the increase in stiffness provided by the repair patches. The central
stiffener is also shown to support more load than the side stiffeners and this may be
seen as a result of the patch to the stiffener.
In Figure 7.38 the moment resultants in the loading direction are shown. It may be
seen that the presence of the repairs has prevented the stiffeners from carrying any
moment resultant, as for the undamaged panel (e.g. see Figure 7.33). The repair patch
may be seen to stabilise the central stiffener as well as the skin around the cut-out.
However, the moment resultant at the edges of the overlap patch is high when
compared with that of the rest of the repair patch. This is a result of the geometric
transition between the thick patch to the unsupported skin of the panel.
As for the undamaged and damaged panels, the stress and moment resultants alone
cannot be used to predict failure. The identification of the locus of failure of the
repaired panels was further complicated, since failure of the repair plugs and patches
may be thought to have preceded the collapse of the panel (see Figure 7.13, Figure 7.14,
Figure A. 3 and Figure A. 4). The method used to estimate the failure displacement of
the undamaged and damaged panel is now used to estimate the failure displacement
of the repaired panel. However, in the present case it is of interest to just give an
estimate of the failure displacement, assuming that the repaired panel failed in a
manner comparable to the undamaged panel. The collapse of the panel is therefore
assumed to have resulted from an in-plane material failure in the skin bay (hence now
comprising the overlap patch on the skin and the stiffener sides). For simplicity, the
results from the geometrically non-linear analysis only were used.
285
DNTOURS Or Nx
ýý3.55
-2583,"
-2411,11
-22*89
-2066.67
-1894.44
-1722,22
-15M
-1377-78
-IM536
403333
-861111
.688899
-516666
-344444
472.222
Figure 7.37: Contour plot of the stress resultant (N/mm) in the loading direction for the
2ONTOURSOF Mx
1ý1733
ý7N 44
'488
, 1153
-193511
-165666
-138222
-110578
40302
432888
-M 444
0
76.444
"2. NO
829302
Figure 7.38: Contour plot of the moment resultant (N) in the loading direction for the
286
The calculated stress resultants at an applied displacement of 2.91mm (i.e. the applied
displacement at failure) are shown in Table 7-5. Again, the LAP [101] software was
used to calculate the stress field in a laminate with a lay-up that comprised the skin
and the overlap repair patches, and subjected to the values of stress resultants shown
in Table 7-5. The calculated stresses were then used with the Tsai-Hill criterion to
predict failure. Failure of the outermost 01 ply (i.e. in the overlap patch on the stiffener
side) was predicted to occur first.
Table 7-5: Computed stress resultants for a node at the edge of the cut-out of the
undamaged failure displacement. However, failure occurred much earlier at the actual
displacement of 2.91mm. Possible reasons such as the interaction of the repair plugs
with the wall of the cut-out may cause the disagreement between FE and experiment,
and these themes are developed in the following section.
The stress resultants calculated at a potential site for failure initiation were used in
287
Undamaged panel
It has been shown that the load in the undamaged panel is carried by the stiffeners as
well as by the skin bays. The unsupported edges were shown to support less load than
the skin bays. The moment resultant in the direction of loading was furthermore seen
to vary periodically in the skin bay, having extrema at the buckle crests. The skin bay
was therefore shown to have high in-plane stress resultant as well as high moment
resultant at buckle crests; thus it may be expected that the collapse of the undamaged
panel occurs at one of the peaks. And, indeed, the stress resultant at a buckle peak was
used to estimate failure and it was shown that the Tsai-Hill criterion predicts failure of
the skin bay at the experimental failure displacement. The results from the linear
elastic and the non-finear elastic analysis were used to estimate failure, and both
methods were seen to show good correspondence with the experimental displacement
at failure. This may be as a result of the limited geometrically non-linear deformation
occurring in the undamaged panel. The only difference was in the prediction of the
buckling deformation. However, the linear elastic analysis was shown to predict the
undamaged strains accurately (Figure 7.25), and as a result it is now seen to be able to
estimate the panel's failure displacement accordingly.
Damaged pand
It was seen for the damaged panel that the presence of the cut-out changes the load
path significantly. The load was seen to be carried mainly by the side stiffeners as well
as by the skin around the cut-out. Furthermore, the stress and moment resultants were
shown to be maximum around the cut-out. A peak in stress resultant was identified at
the mid-length of the panel and at the edge of the cut-out and was in accord with the
experimental observation that the damaged panel failed by an in-plane compressive
failure of the skin at this precise location. However, the calculated displacement at
failure was shown to be about 28% lower than experiment.
The discrepancy between experimental and calculated results may be due to the coarse
mesh used in the FE model. The stresses at the edge of a cut-out are strongly
dependent on the mesh size and type. Another factor, other than the size of element,
may be the type of elements used. It may be seen in Figure 7.21 that some triangular
elements were used around the cut-out, as well as some distorted quadrilateral
elements. Both may lead to poor results when used in a rapidly varying stress field, as
288
around the cut-out The use of higher order elements may help in these cases,although
the distorted shape of the elements may cause inaccuracies in the calculations. Thus, a
better agreement between calculated and experimental failure displacement may be
found if the mesh around the cut-out is refined.
Repaired panel
It was seen that the load path in the repaired panel has been restored using the present
repair scheme. The repair has recovered the load-carrying capability of the central
stiffener as well as that of the skin around the cut-out. However, the study of the
contour plots did not reveal any peaks in value, nor did they suggest a potential
position for failure initiation. It was then assumed that the repaired panel failed in the
skin bays and in the repair patch. The calculated displacement at failure was 38%
higher than in the experiment.
The poor agreement between the estimated and experimental failure displacement
may be a result of the modelling approach used. Indeed, the repair method used in the
present work consisted of two overlap patches applied over the damaged area. The
cut-out was 'filled' with two repair plugs (i. e. one for the skin and one for the stiffener)
enveloped in adhesive to ensure that no clearance existed between the plug and the
structure wall (see section 3.4.3). However, it is clear that the repair plug and its
interaction with the wall of the panel is not taken into account in the present modelling
approach, and it may be expected that the presence of the plugs will have an influence
on the strength of the panels. Although the stress concentrations at the edges of the
cut-out may be greatly reduced by the stress redistribution due to the repairs, a stress
concentration at the edge of the damage will still exist. The clearance between the plug
and the wall of the panel was 0.5mm and an interaction between the plug and the wall,
when the panel is under load, will give rise to stress concentrations around the cut-out.
As a result, material in-plane failure is expected, but at a lower load than that
calculated by the present FE model. It is noted that the interaction between the plugs
and the wall of the panel cannot be modelled using the present idealisation. It may be
considered that the modelling approach used in the present work is not well adapted
at estimating the failure load of the panel, if failure results from the interaction
between the plugs and the wall of the structure.
289
7.3.5 Conclusions on the predictive studies
show good agreement with the experimental stiffness behaviour of the panels.
However, the linear elastic and the geometrically non-linear elastic approaches were
both seen to present advantages and disadvantages.
The linear elastic analysis was simple and it may be inexpensive to run (i.e. when
The geometrically non-linear analysis was seen to lead to the most accurate results as it
was able to account for large deformation and predict buckling deformation. The
method was shown to be most accurate for predicting the stiffness of the panels,
although with an increase in computing resources needed.
of failure initiation was lost in the explosive failure. The possible sites for failure
initiation in the panels and the possible mode of damage are numerous, hence
complicating the determination of the locus of failure.
The contour plots of stress and moment resultants in the loading direction of the
undamaged, repaired and damaged panels were compared. It was shown that in the
undamaged panel the load is mainly carried by the skin and the stiffeners. The cut-out
in the damaged panel was seen to render the central stiffener ineffective, therefore
redistributing the load to the side stiffeners and the skin around the cut-out. The
present repair scheme was shown to restore the original load path in the panel,
although it led to an over-stiffening of the skin bays. Potential failure locations were
identified in the undamaged and damaged panels from the contour plots.
290
The stress and moment resultants were shown to be maximum at buckle crests in the
skin bays, and the Tsai-Hill criterion successfully predicted the failure of the panel. It
is noted that the results from both the linear elastic and non-linear elastic analyses
were used to predict failure, and the results were identical. The results from the non-
linear elastic analysis only were used to estimate the failure displacement of the
damaged and repaired panels.
For the damaged panel, the stress resultants were shown to be the highest around the
cut-out. With the present mesh the failure displacement was calculated to be 28%
lower than experiment. The disagreement was shown to be as a result of the coarse
mesh used around the cut-out. It was suggested that mesh refining is necessary to
predict accurately the failure displacement.
In contrast to the two previous analyses it was not possible to identify any potential
failure site for the repaired panel. Instead it was assumed that the panel fails by a
compressive failure of the skin bay, as for the undamaged panel. The failure was in
this case significantly over-predicted, and this was as a result of the simplicity of the
modelling approach. The current FE models do no take into account the interaction of
the repair plugs and the wall of the cut-out, where failure is thought to have initiated.
291
7.4 Conclusions on the performance of the I-stiffened panels
Experimental studies
The static strength of the undamaged, damaged and repaired panels were much as
comparison, the damage that consisted of a hole through the skin only resulted in just
a 30% drop-off in strength. The repair schemes employed recovered 83% and 90% of
the undamaged strength for the through-hole and the skin-hole damage, respectively.
The overlap repair patches used in this study were shown to have successfully
recovered the load-carrying capability of the damaged panels since the far field strains
of the repaired panels at failure were above -6000gstrain.
There is a number of locations in the panels where failure could have initiated, and in
panels' collapse initiated from a skin in-plane compressive failure (rather than by
buckling) that led to stiffener debonds, and ultimately to the collapse of the skin and
the stiffeners. The failure of the damaged panels resulted from the high strains
occurring at the edge of the cut-out. The present repair scheme recovered the load path
in the panels, reducing the excess of stress in the skin around the damage and
ultimately increasing the failure load of the panels.
Predictive studies
TI-LeFE analysis method has been shown to be a good candidate to evaluate the
behaviour of the panels. The quasi-31) shell element technique was employed to model
one undamaged, one damaged and one repaired panel. Two analyses were run for
each model, one linear elastic and one geometrically non-linear elastic. Good
correlation between calculated and measured stiffness properties was observed for
both type of analysis, although the linear elastic analysis was inappropriate for the
292
The load path for the undamaged panel was such that load is carried by the skin as
well as by the three stiffeners. Large out-of-plane deformations were seen to occur at
the wide unsupported edges and in the skin bay of the undamaged panel. The
presence of the cut-out changed the load path in the panels significantly, resulting in
the load being carried mainly by the two. side stiffeners and by the skin in the area
adjacent to the damage. The present repair scheme was shown to restore the original
load path in the panel, although it led to an over-stiffening of the skin bay. Potential
failure locations were identified from the contour plots of the stress resultant
_-
and moment resultant.
For the undamaged panel, the stress and moment resultants were shown to be
maximum at buckle crests in the skin bays, and the Tsai-Hill criterion successfully
predicted the failure of the panel. For the damaged panel, the stress resultants were
shown to be the highest around the cut-out. With the present mesh the failure
displacement was calculated to be 28% lower than experiment. The disagreement was
thought to be as a result of the coarse mesh used around the cut-out. In contrast to the
two other analyses it was not possible to identify any potential failure site for the
repaired panel. The current FE models do no take into account the interaction of the
repair plugs and the wall of the cut-out, where failure is thought to have initiated.
293
8. Conclusions and suggestions for future work
8.1 Conclusions
8.1.1 Introduction
It is widely accepted that in-service or fabrication damage can greatly reduce the
The main conclusions from the present work will be presented in this Chapter. Section
8.1.2 will present the results from the structural element testing programme. Design
guidelines, based on the results presented in the section, will be drawn. The results
from the modelling studies will be presented in section 8.1.3. The aspects of stiffness
and failure predictions will be reviewed. The development of a coupon, that mimics
the behaviour of the sandwich beams will be reviewed in section 8.1.4. The
performance of the coupons will be compared with that of the structures.
294
8.1.2 Structural element testing
The first phase of this work addressed the design of repairs to generic structural
extremely fruitful in drawing design guidelines that can be used in the design of
repairs to generic CFRP structural elements.
statically and in fatigue (see Chapter Four and Five, respectively). Overlap and scarf
repairs were designed and placed on one face only (i. e. the tool face, TF) of the beams.
The repairs were subjected to either compressive or tensile load. The behaviour of a
high-temperature cure (HTC) repair system, identical to the parent materials, was
compared with that of a low-temperature cure (LTC) system that would be normally
used for in-the-field repair situations.
repair schemes used in the present work were found to be successful, with most
repairs failing at a load in excess of 90% of the undamaged failure load, therefore
The repair system, being either high- or low-temperature cure did not have any major
effect on the strength of the beams in this study, although the type of repair did.
Indeed, 2-ply overlap repairs were better than 1/30 scarf and 1/10 scarf repairs when
tested in compression; the former achieved 100%, whereas the latter only achieved
about 90% of the undamaged strength. When tested in tension, however, the 1/30 scarf
and 1/10 scarf repairs were stronger, achieving 100% of the undamaged strength,
with, now, the 2-ply overlap repairs showing about 20% reduction. Most specimens
failed in the parent laminate, away from the repair patch.
295
Fatigge stu ies;(refer to section 5.2.3 and 5.3.4)
Undamaged beams tested at a maximum fatigue load of 60% the undamaged failure
load (60%UFL) had not failed within one million cycles whether tested with the TF in
although the beams tested with the TF in tension had not failed within 400,000cycles.
The fatigue performance of the beams was worse when the beams were tested with the
TF in compression. This conclusion corresponds to the usual belief that the behaviour
of CFRP materials is worse when loaded in compression fatigue than when loaded in
tension fatigue.
The fatigue performance of the repaired beams was inferior to that of the undamaged
beams, whatever the loading mode. All the beams were seen to accumulate damage
with fatigue cycles. The softening was identified as a result of the development of
compressive, shear-induced cracks, or tensile, transverse cracks that led to the
separation of the warp and weft bundles, for the beams tested with the TF in
compression and tension, respectively (see section 5.4.5). It was, however, observed
that a trend in softening existed, depending on the repair type, and on the loading
mode. For instance, scarf repaired beams tested with the TF in compression at 60%UFL
were seen to accumulate damage in the same way as an undamaged beam, but tested
at 75%UFL. On the other hand, overlap repairs were shown to soften in the same way
as an undamaged beam, tested at the same load ratio of 60%UFL. It did seem that a
premature failure is initiated by the scarf patch when tested in compression. However
tested in tension at 60%UFL, 1/30 scarf repairs softened as much as an undamaged
beam tested at 65%UFL, but the 1/10 scarf repaired beams now showed a higher rate
of softenin& which increased again for the overlap repaired beams. The higher the
scarf angle, the steeper is the softening. Increasing the scarf angle was detrimental
under a tensile regime, whereas it did not much affect the compressive response.
296
Conclusions (refer to section 4.6 and 5.5)
The conclusions from the static studies corresponds with the conclusions from the
fatigue studies. The overlap repairs have performed as wen as, if not better than, the
scarf repairs when tested under compressive loading. The scarf repairs were, however,
better than the overlap repairs under a tensile regime. However, in order to conclude it
is important to review the overall performance of the repairs when subjected to
From the above results, the best overall perfom-iance is obtained with the 1/30 HTC
scarf repairs (i.e. about 20 angle). This conclusion is in total agreement with the
literature, since it was established that, of all the repair types, a scarf joint with a scarf
angle smaller than 1/20 (i. e. about 3*) results in a smooth load transfer, produces
negligible peel stresses,and hence allows the full potential of a joint to be achieved.
The performance of the 1/30 scarf low-temperature cure (LTC) repair was, however,
inferior when compared with the performance of the equivalent HTC scarf repair.
Although the behaviour of the repair was excellent when tested in tension, the LTC
repair system promoted a premature failure of the repair when tested in compression
(static and fatigue). The LTC repair system tj; ýr thought to be responsible for the
lower performance.
Increasing the scarf angle from 1/30 to 1/10 did not have much effect on the
performance of beams tested with the TF in compression. However, when tested with
the TF in tension, the fatigue performance of 1/10 scarf repairs was better than that of
overlap repaired beams, but not as good as that of 1/30 repaired beams. Consequently,
the performance of the 1/10 HTC scarf repairs was intermediate. This work has shown
that non-standard scarf repairs, such as a 1/10 scarf (i.e. about 6") can also be used for
the design of repairs to sandwich panels. This is important as it means that
significantly less undamaged material has to be removed compared with a standard
1/30 scarf.
The fatigue performance of the 2-ply overlap repaired beams was the poorest of all the
repaired beams. Although the beams lasted as long as the scarf repaired beams when
297
tested with the TF in compression, they failed prematurely from a delamination failure
of the repair patch when tested with the TF in tension.
was assessedin compression. The work on the stiffened panels was restricted to static
testing only since insufficient panels were available for a full fatigue study. Two
overlap repair schemes were designed; one involved the repair of the back (skin) and
front (central stiffener) faces of the panel, whereas the other involved the repair of the
back face only. The patches were graded in thickness to reduce the development of
The undamaged panel failed, explosively, at the impressive load of 1200kN. No major
buckling deformation was seen by the Moird fringe technique, although deformation
of the unsupported edges was shown to occur. However, since the collapse of the
panels was explosive, identification of the precise failure mechanism was difficult.
From visual observation of the panels after testing and analysis of the experimental
data, it was shown that the panel's collapse initiated from a skin in-plane compressive
failure (rather than by buckling) that led to stiffener debonds, and ultimately to the
The panel damaged by a through-hole lost 57% of its strength. The failure resulted
from the high strains occurring at the edge of the cut-out. By comparison, the damage
that consisted of a skin-hole only resulted in just a 30% drop-off in strength. The
repaired panels failed at 83% and 90% of the undamaged strength for the through-hole
and the skin-hole damage, respectively. The present overlap repairs restored the load
path in the panels, reducing the excess of stress in the skin around the damage and
ultimately increased the failure load of the panels. The chosen repairs were hence
successful, and allowed an almost complete recovery of the load-carrying capability of
the damaged panels.
298
The design of repairs to thin-skinned composite sandwich beams and thick I-
stiffened panels was straightforward. The repairs were designed following the
rules for double-lap joints [431. In static loading, the repairs to the sandwich beams
failed at a load not significantly different from the undamaged strength, whatever
the loading mode. The repairs to the I-stiffened panels were also successful and
their load-eccentricity. Although their static performance was good, the overlap
length-to-thickness ratio of 25:1 used in the present work led to delaillinatioll
failure when tested in tensile fatigue. A longer overlap or a tapered overlap (i. e. is
off of the repair patch on the stiffened panel was not observed; delamination
failure was not a problem. Failure was shown to be dependent on the interaction
between the repair plug and the wall of the cut-out.
energy and a cured thickness of 0.1-0.2 mm should be used in design to achieve the
highest joint strength.
The 1/30 scarf repairs provided the best overall performance in the present work.
The 1/10 scarf repair is a good alternative to the conservative 1/30 scarf repair,
although this design offers limited performance under fatigue loading. The overlap
PRES17N I DING N
299
8.13 Modelling studies
computationally, the behaviour of the repaired structures. The predicted results were
compared with experimental results. The objectives were to accurately predict the
stiffness of the repaired structures and to estimate the structural failure load and
failure mode.
calculated and measured stiffness (i.e. deflection and surface strain) were observed.
However, the different approaches led to different results. As expected the 3D brick
element method led to the best prediction, but it is an approach that is not viable to use
for day-to-day design. The repaired beams were therefore only modelled with the
The results from the quasi-31) shell and 2D plane strain FE models of a 1/30 HTC scarf
repaired beam were compared, and they showed good correspondence with
experiments, although the 2D plane strain idealisation led to slightly stiffer
predictions. However, the 2D plane strain approach allows the accurate prediction of
parent, repair and adhesive shear stresses. The quasi-31) approach predicted the
complex deformation of the beams accurately. Parent and repair stresses were also
calculated. However, the method is inappropriate if a prediction of the repair adhesive
shear strain is required.
The calculation of the failure load of the repaired sandwich beams, using the results
from the 2D plane strain model, was discussed. It was not possible to predict the
structural strength of the beams because the evolution of damage in the materials (i. e.
parent and repair) was not taken into account in the present analyses. The calculated
results were, however, used to understand the stress transfer mechanism between the
parent and the repair patch. The adhesive shear strain distribution was calculated and
assuming that the adhesive fails when it reaches its ultimate shear strain, it was seen
that with the mesh employed, adhesive failure was not predicted to occur.
300
8.1.3.2 Composite stiffened panel (refer to section 7.3.5)
The quasi-31) shell element technique was employed to model one undamaged, one
damaged and one repaired panel. Two analyses were run for each model; one linear
elastic and one geometrically non-linear elastic. Good correlation between calculated
and measured stiffness was observed for both type of analysis, although the linear
elastic analysis was inappropriate at the evaluation of strains in areas where large out-
of-plane deformation occurred. The geometrically non-linear analyses led to the most
accurate results.
The load distribution in the undamaged panel was such that the load is carried by the
skin as well as by the three stiffeners. Large out-of-plane deformation was seen to
occur at the wide unsupported edges of the undamaged panel, whereas they were
limited in the skin bay. The presence of the cut-out changed the load distribution in the
panels significantly, resulting in the load being carried mainly by the two side
stiffeners and by the skin in the area adjacent to the damage. The present repair
scheme was shown to restore the original load distribution in the panel, although it led
to an over-stiffening of the skin bay. Potential failure locations were identified from
the contour plots of the stress resultants.
For the undamaged panel, the stress and moment resultants were shown to be
maximum at buckle crests in the skin bays, and the Tsai-Hill criterion successfully
predicted the failure of the panel. For the damaged panel, the stress resultants were
shown to be the highest around the cut-out. With the present mesh the failure
displacement was calculated to be 28% lower than experiment. The disagreement was
shown to be a result of the coarse mesh used around the cut-out. In contrast to the
analysis for the other two panels it was not possible to identify any potential failure
site for the repaired panel. The current FE models do no take into account the
interaction of the repair plugs and the wall of the cut-out, where failure is thought to
have initiated.
301
8.1.3.3 Conclusions
The FE analysis method has been shown to be a good candidate to evaluate the
stiffness of the composite structures. Failure prediction was more difficult. The main
conclusionsfrom the presentstudy are:
As expected, an increase in the complexity of the model improves the quality of the
prediction. The 3D brick element model gave the most accurate results. 2D and
quasi-3D models were seen to be complementary in this study.
The quasi-31) shell element model was simple to set up and best suited to
predicting the overall deformation pattern of the structures. This method was
successfully used to predict the behaviour of the sandwich beams and the stiffened
panels, although the calculation of adhesive stressesis impossible.
The 2D plane strain model allows the careful evaluation of several repair designs
(e.g. effect of repair stacking sequence,adhesive thickness and properties, etc.). The
method was successfully used to calculate the stress distribution in the repairs to
the sandwich panels.
The prediction of the failure load of the sandwich beams was difficult due to the
complex failure of CFRP materials, and to the inadequacy of the existing method to
predict the failure load of adhesive joints. The Tsai-Hfll criterion was shown
particularly inappropriate and a progressive damage methodology should be used
instead. With the current meshes, the calculated maximum values of shear strain
were well below the uldrnate shear strain of the adhesive and the theoretical
analyses agree with experiment since the high-temperature cure repairs did not fail
when tested statically.
The quasi-3D shell element method successfully predicted the failure load of the
undamaged and damaged stiffened panels since the panels' collapse resulted from
an in-plane skin compressive failure. However, the modelling approach used in the
present work was not well adapted to estimating the failure load of the repaired
panel, since failure results from the interaction between the plugs and the wall of
the structure and is not taken into account in the present model.
302
8.1.4 Development of a generic approach
The third phase of the present research concentrated on developing cost effective
coupons, designed to mimic repaired structural elements in the way they fail. The
objective was to compare the performance of the coupons with that of the structures.
Coupons were tested under static and fatigue loading to obtain failure strength and
fatigue degradation rates and the results were used to predict the failure load or mode
In fatigue, the repaired coupons were weaker than the undamaged ones (refer to
section 6.4.5). Also, the fatigue performance of HTC repaired coupons was superior to
that of LTC repaired coupons. It was already suggested earlier that the LTC repairs
were weaker. These conclusions are again similar to those made for the undamaged
and repaired sandwich beams tested in fatigue. Thus, the coupons may be seen as a
good alternative to the more expensive exercise of structural repair testing. However,
the number of cycles to failure of undamaged and repaired coupons was compared
with that of the sandwich beams and, although they showed similar static behaviour,
the fatigue performance of the coupons was superior; the repaired coupons lasted
longer than the repaired beams. One possible reason was that the coupons were stiffer
than their equivalent TF skins. Under comparable loads, the coupons and the beams
showed similar fatigue performance. Overall, the tensile coupons were seen to be a
good alternative for designing repairs on thin-skinned sandwich beams.
Following the positive results shown above, the fatigue degradation curves of the
undamaged and repaired coupons were also studied (refer to section 6.5.5). It was
shown that the higher the applied load level, the greater the reduction in stiffness. A
stiffness degradation model was used to fit the stiffness degradation curves of the
coupons. An empirical residual stiffness law was then used, in conjunction with an
303
analytical strip model of a beam in four-point bending, to predict the evolution of
stiffness of the sandwich beams. The empirical approach was shown to successfully
predict the fatigue degradation curves of repaired beams from the fatigue data
obtained from representative coupons, although the difference in stiffness of the
coupons and the TF of the beams was not taken into account. Further work is
necessary to draw definite conclusions.
available. Since the panels' collapse resulted from in-plane compressive failure of the
skin, rather than by buckling, the development of a coupon tested in compression may,
nevertheless, be envisaged.
Conclusions
A representative coupon for the sandwich beams was successfuuy tested. The
coupon has shown that the TF skin of the sandwich beams can be idealised as if it
with the failure load and mode of the equivalent sandwich beams. However, the
fatigue performance of the coupons was superior to that of the beams.
An attempt was made to predict the stiffness reduction behaviour of the beams by
modelling the stiffness behaviour of the coupons. An empirical approach has been
used to predict the reduction in bending stiffness with number of cycles of the
beams, from the fatigue data obtained from the reduction in axial stiffness with
The work done in the present studies has proven to be successful and conclusions
were made to improve repair design guidelines and select efficient design analysis
methods for the design of repairs to composite structures. Limited work was also
successfullyundertakento develop simple but relevant test methods that can be used
in the designof repairs.However, there is room for further investigation,as presented
below.
304
8.2.1 Sandwich beams
Although no difference was found in the static strength of the repairs investigated in
the present work, more tests should be carried out to statistically characterise the
The fatigue performance of the repaired beams was found to be dependent on the
loading mode, i. e. whether tested with the repaired TF in compression or tension.
Hence, the fatigue performance of the undamaged and repaired beams should be
investigated in a combined compressive-tensile loading; the strains being taken from
compressive to tensile in each cycle.
The current 2-pIy overlap repairs schemes showed good performance under
compressive loading, although the delamination of the patch led to a premature failure
The FE modelling approaches used in the present work successfully predicted the
stiffness behaviour of the beams. The prediction of failure load was however more
difficult since composite materials can develop complex local failures. A progressive
failure analysis method that takes into account the evolution of the damage should be
to
used predict the response of the structures from initial loading to final failure.
The tensile coupons designed to min-dc the behaviour of the TF of the sandwich beams
305
8.2.2 Stiffened panels
The repair patches designed for the I-stiffened panels recovered very high proportions
of the undamaged strength of the panels. However, the design of the present patch
was not optin-dsed, and it should be investigated, perhaps using the quasi-31) FE
approach used in the present work. The design of a patch for which the axial stiffness
is maxin-dsed,but the bending stiffness minimised should be studied.
The modelling approaches used in the present work allowed a successful prediction of
the stiffness behaviour of the panels, whether undamaged, damaged or repaired. The
prediction of the failure displacement was only partially a success.The strength of the
undamaged panel was correctly predicted but the strength of the damaged panel was
over-predicted since it is dependent on the mesh size around the cut-out region. A new
model, with a refined mesh, should be constructed and the strength of the damaged
panel should be re-calculated. Also, the interaction between the repair plugs and the
wall of cut-out, where failure was thought to have initiated should be modelled with
damage progression included. It is anticipated that a full 3D approach will be needed.
Finally, the modelling of the behaviour of the panel damaged by a skin-hole and the
associated repaired panels was not undertaken in the present work, and should be
investigated.
306
REFERENCES
[10] M. C. Y. Niu, CompositeAirframe Structures, 2nd. ed: Hong Kong Comilit Press
Ltd., 1996.
307
[14] E. Greenhalgh, S. Singh, and D. Roberts, "Impact Damage Growth and Failure
[25] J. E. Robson, "The Repair of Composite Materials" : PhD Thesis: Centre for
Composite Materials, Imperial College of Science,Technology and Medicine, London,
1993.
308
[27] 1. Fidler, "British Airways Experience with Repairs" presented at 3rd
Imperial/DERA Workshop on Repair and Maintenance using FRP: Centre for
Composite Materials, Imperial College, London, 2000.
309
[391 A. A. Baker, "On the Certification of Bonded Composite Repairs to Primary
Aircraft Structures" presented at 11th International Conference on Composite
Materials. Gold Coast, Australia, pp. 1-24, vol. 1/6,1997.
[411 D. C. Jegley and H. G. Bush, "Structural Test Documentation and Results for
the McDonnel Douglas All-Composite Wing Stub Box" NASA, Technical
Memorandum TM 110204,1997.
[501 J. A. Bishop, "The History of Redux@ and the Redux Bonding Process",
International JournalofAdhesionand Adhesives,vol. 17, pp. 287-301,1997.
310
[52] A. J. Kinloch, Adhesion and Adhesives: Scienceand Technology:Chapman and
Hall, 1987.
[56] D. J. Allman, "A Theory for Elastic Stressesin Adhesively Bonded Lap Joints",
Q. 1.Mech.Appl. Maths., vol. 30, pp. 415-436,1977.
pp. 1-44,1981.
[59] R. D. Adams and N. A. Peppiatt, "Stress Analysis of Adhesive-Bonded Lap
Joints", Journalof Strain Analysis, vol. 9, pp. 185-196,1974.
[62] M. J. Davis, "'How Relevant are Load Rate Effect to Realistic Bonded Joints"
[641 J. D. Clark and I. J. McGregor, "Ultimate Tensile Stress over a Zone: A New
Failure Criterion for Adhesive Joints", JournalofAdhesion, vol. 42, pp. 227-245,1993.
311
[661 M. N. Charalambides, A. J. Kinloch, and F. L. Matthews, "Adhesively Bonded
Repairs to Fibre Composite Materials - IL Finite Element Modelling", CompositePart A,
pp. 193-216,1986.
[701 D. W. Adkins and R. B. Pipes, "Tensile Behaviour of Bonded Scarf Joints
between Composite Adherends" presented at Proceedings of the 4th Japan-US
Conference on Composite Materials. Washigton D.C., pp. 845-854,1988.
[71] J. P. H. Webber, "Scarf Repair Joints in Carbon Fibre Reinforced Plastic Strips",
JournalofAdhesion, vol. 12, pp. 257-281,1981.
[74] R. I. Mackie and N. Su, "The Effect of Ageing and Environment on the Static
and Fatigue Strength of Adhesive Joints" presented at Structural Adhesives in
Engineering III. Bristol, UK: Gordon and Breach SciencePublishers, pp. 191-207,1992.
[75] J. W. Choi, W. Hwang, H. C. Park, and K. S. Han, "Observation of Static
Strength and Fatigue Life of Repaired Graphite/Epoxy Using Tensile Coupon"
presented at 11th International Conference on Composite Materials. Gold Coast,
Australia, pp. 365-373,vol. 6/6,1997.
[761 C. Soutis and F. Z. Hu, "Design and Performance of Bonded Patch Repairs of
Composite Structures.", Proceedingsof the Institution of Mechanical Engineers,vol. 211,
pp. 263-271,1997.
[77] K. Wolf and R. Schindler, "External Patch Repair of CFRP/Honeycomb
Sandwich" presented at Composite Repair of Military Aircraft Structures
-AGARD CP
550. Seville, Spain: NATO, pp. 18-1/18-11,1994.
312
[78] A. D. Dew, "To Evaluate Repair Techniques for Use on Composite Main Rotor
Blade Trailing Edge Assemblies" Westland Helicopters Ltd., Yeovil, UK, LR 95-146,
1995.
[821 G. Gunther, "Composite Repair of Co-cured J-Stiffened Panels, Design and Test
Verification"presented at AGARD CP-402, The Repair of Aircraft Structures
involving Composite Materials. Oslo, Norway, pp. 8-1/8-19,1986.
313
in Engineering) Conference. 16-19 April 2000. University of Greenwich, London, UK,
2000.
[93] E. H. I. Ltd, "Woven Carbon Fibre Pre-Preg WHMS 488 Type 3, Composite
Materials Data Sheet" : E.H. Industries Ltd, 1997.
[95] B. Cartwright, "MSc Thesis: The Use of Strain Gauges on Woven Composite
Materials" in Centrefor CompositeMaterials, Imperial Collegeof Science,Technologyand
Medicine, London, UK, 1993.
314
[1011 "LAP, Laminate Analysis Program", Centre jbr CompositeMaterials, Imperial
Collegeof Science,Tecnologyand Medicine, London UK.
aligning Four-point Bend Testing Ring and Sample Geometry Effect in Four-point
Bend Fatigue", International Journalof Fatigue,vol. 21, pp. 889-894,1999.
pp. 123-136,1999.
[1081 C. I. C. Manger, S. L. Ogin, P. A. Sn-dth, and R. P. Greaves,
"Damage Development in Plain Weave GFRP" presented at 11th International
Conference on Composite Materials. Gold Coast, Australia,. Gold Coast, Australia, pp.
58-66,1997.
[1101 D. Zubin, "Error Analysis and Method of Accuracy Evaluation of Strain Gauge
Factor. Testing Apparatus based on Pure Bending Moment Beams", Journal of the
British Societyjbr Strain Measurement,vol. 35, pp. 87-96,1999.
315
Modelling the Effect of Damage on the Macro-mechanical Properties", Composites
Scienceand Technology,vol. 59, pp. 137-145,1999.
[114] F. K. Chang and K. Y. Chang, "'A Progressive Damage Model for Laminated
Composite Stress Concentrations", Journal of CompositeMaterials, vol. 21, pp. 834-855,
1987.
pp. 64,1978.
[120]F. Mortensen and 0. T. Thomsen, "A Simple Approach for the Analysis of
Embedded Ply Drops in Composite and Sandwich Laminates", CompositeiScience
and
Technology,vol. 59, pp. 1213-1226,1999.
316
[123] J. A. Nairm. and S. Hu, "Matrix Microcracking" in Damage Mechanics of
CompositeMaterials, vol. 9, R. Talreja, Ed.: Elsevier, pp. 187-243,1994.
317
APPENDIX A
318
Repair patch
319
Figure A. 4: Compressive failure close to the end of the overlapping ply
320
APPENDIX B
321
Figure B.I: Undamaged panel (D 1)
322
Le- I
323
Figure B.3: Damaged panel (D3)
324
;7z
325
Figure B.5: Repaired panel (D4): side views
326
Figure B.6: Repaired panel (D5): stiffener and skin views
327
L
Figure B.7: Repaired panel (D5): side views
w 328