Professional Documents
Culture Documents
The Structural Integrity of Carbon Fiber Composites: Peter W.R. Beaumont Constantinos Soutis Editors
The Structural Integrity of Carbon Fiber Composites: Peter W.R. Beaumont Constantinos Soutis Editors
Beaumont
Constantinos Soutis Editors
The Structural
Integrity of
Carbon Fiber
Composites
Fifty Years of Progress and Achievement
of the Science, Development, and
Applications
The Structural Integrity of Carbon Fiber
Composites
Peter W.R. Beaumont • Constantinos Soutis
Editors
123
Editors
Peter W.R. Beaumont Constantinos Soutis
Department of Engineering Aerospace Research Institute
University of Cambridge The University of Manchester
Cambridge, UK Manchester, UK
It may have escaped the notice of the casual observer and user of carbon fibre
products such as the tennis racket or golf club that carbon fibre has been around now
for 50 years from those ‘heady days’ of the 1960s when, as a matchbox curiosity,
the short black ‘prickly’ fibre was worth more than its ‘weight in gold’. Within a
year, the fibre was being produced as continuous fibre in bundles or tows of 10,000
individual filaments wound onto spools that could be thrown like an American
football to the back of college lecture theatres, the fibre unravelling as it sped from
the speaker’s hand over the heads of excited engineering students.
The discovery of this high-strength, high-stiffness, lightweight material that
could be woven into fabric had returned us to those factories of northern British
towns of the eighteenth and nineteenth century. Fibre in and product out through the
factory front door, we have now arrived where material processing and manufacture,
design and fabrication of the final product can now take place under one roof. We
have arrived back to the factory of the future.
It is fitting on the occasion of 50 years of carbon fibre to place on record the
historical stages of the development of structural carbon fibre composites (and
other fibre composites), as well as premonitory things that happened on the way.
An impressionistic map, if you like, of the present state of the art for readers
coming new to the field as well as those ensconced in research and development
of composite materials and large composite structures, from the emergence of the
science of composites and the evolution of a number of neighbouring disciplines.
In the ‘blink of an eye’, we have in composite materials discovered a clearly
defined and distinct discipline, which in practice doubles up as a multidiscipline
with a substantial number of independent ‘branches’, each one with its multifarious
journals and textbooks, read by practicing materials scientists and engineers alike
who communicate with each other on that basis. The discipline of the materials
science and engineering of contemporary composite materials has emerged from
those small beginnings one half century ago.
But this book is in no sense a textbook of carbon fibre composite materials
alone. In this single volume, the authors have produced an impressionistic map
of the present state of the science and mechanics of composite materials, seen as
v
vi Preface
Robert Hooke writes, in his book Micrographia (1665), an imprint of the Royal
Society of London:
And I have often thought that probably there might be a way found out to make an artificial
glutinous composition much resembling if not full as good, nay better than that excrement
or whatever substance it might be out of which the silk worm draws his clew. If such a
composition were found it were certainly an easy matter to find very quick ways of drawing
it into small wires* for use. I do not mention the use of such an invention nor the benefit that
is likely to accrue to the finder, they being fully obvious. This hint may I therefore hope to
give some inquisitive person an occasion of making some trials which if successful I have
my aim and I suppose he will have no occasion to be displeased.
• Old English wı̄r; Germanic origin, probably the base of Latin viere ‘plait, weave’
First of all, we want to acknowledge our debt to the late Professor Tony Kelly,
materials scientist and friend, who through his great depth of understanding of
composite materials gave sage advice and encouragement from the very beginning
when the idea of this book began to crystallise 2 years ago.
We thank the authors of the chapters for spending so much of their valuable
time and support throughout the process of preparing this book. They are
to be congratulated in bringing together the basic concepts of physical and
micromechanics modelling and mathematical and continuum modelling and in
reconciling them with each other to reappraise current design practice and future
design strategies and to develop and validate risk-based assessment methodologies.
We thank them for their cooperation throughout this project and their writing of
such excellent chapters contained herein.
ix
Contents
Part I
1 50 Years in Carbon Fibre, 60 Years in Composites .. . . . . . . . . . . . . . . . . . . . 3
Robert D. Adams
2 ‘But How Can We Make Something Useful Out of Black
String?’ The Development of Carbon Fibre Composites
Manufacturing (1965 –2015) . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 29
Kevin Potter
3 Boron Fiber to Carbon Fiber.. . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 59
Carl T. Herakovich
4 Serendipity in Carbon Fibres: Interfaces and Interphases
in Composites . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 71
Frank R. Jones
Part II
5 Nano-Engineered Hierarchical Carbon Fibres and Their
Composites: Preparation, Properties and Multifunctionalities.. . . . . . . 101
Han Zhang, Emiliano Bilotti, and Ton Peijs
6 Nano-engineered Carbon Fibre-Reinforced Composites:
Challenges and Opportunities.. . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 117
Larissa Gorbatikh and Stepan V. Lomov
7 A Nano-micro-macro-multiscale Model for Progressive
Failure Prediction in Advanced Composites .. . . . . . . .. . . . . . . . . . . . . . . . . . . . 137
Samit Roy, Abhishek Kumar, and Shibo Li
8 Carbon Fibre-Reinforced Polymer Laminates
with Nanofiller-Enhanced Multifunctionality . . . . . . .. . . . . . . . . . . . . . . . . . . . 171
Wei Han, Youhong Tang, and Lin Ye
xi
xii Contents
Part III
10 Microscale Characterization Techniques
of Fibre-Reinforced Polymers . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 283
M. Herráez, F. Naya, C. González, M. Monclús, J. Molina,
C.S. Lopes, and J. LLorca
11 Fibre Distribution and the Process-Property Dilemma . . . . . . . . . . . . . . . . 301
John Summerscales
12 Analysis of Defect Developments in Composite Forming . . . . . . . . . . . . . . 319
P. Boisse, N. Hamila, and A. Madeo
Part IV
13 Deformation Mechanisms of Carbon Fibres and Carbon
Fibre Composites . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 341
Robert J. Young
14 Micromechanical Evidences on Interfibre Failure of Composites . . . . 359
Federico París, Elena Correa, and Vladislav Mantič
15 Progressive Damage in Fibre-Reinforced Composites:
Towards More Accurate and Efficient Computational
Modelling and Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 391
B.Y. Chen and T.E. Tay
16 Predicting Properties of Undamaged and Damaged
Carbon Fibre Reinforced Composites .. . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 425
L. Neil McCartney
Part V
17 Composites Toughen Up! . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 471
Ivana K. Partridge
18 Slow Cracking in Composite Materials: Catastrophic
Fracture of Composite Structures . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 489
Peter W. R. Beaumont
19 Finite Fracture Mechanics: A Useful Tool to Analyze
Cracking Mechanisms in Composite Materials . . . . .. . . . . . . . . . . . . . . . . . . . 529
E. Martin, D. Leguillon, and N. Carrère
Contents xiii
Part VI
23 A Virtual Testing Approach for Laminated Composites
Based on Micromechanics . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 667
Pierre Ladevèze, David Néron, and Hadrien Bainier
24 Virtual Testing of Composite Structures: Progress
and Challenges in Predicting Damage, Residual Strength
and Crashworthiness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 699
Brian G. Falzon and Wei Tan
25 Contribution of Virtual Simulation to Industrialisation
of Carbon Fibre-Reinforced Polymer (CFRP) Composites
for Manufacturing Processes and Mechanical Performance . . . . . . . . . . 745
Anthony K. Pickett
Part VII
26 Multi-scale Progressive Failure Modeling:
From Nano-structured Carbon Fibers to Textile Composites . . . . . . . . . 767
A.E. Bogdanovich
27 Textile Structural Composites: From 3-D to 1-D Fiber Architecture . 795
Frank K. Ko and Lynn Y. Wan
28 Experimental and Multiscale Numerical Studies of Woven
Fabric Carbon Composite Cylinder Subjected to Internal
Pressure Loading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 849
Young W. Kwon, Timothy D. Ponshock, and John D. Molitoris
xiv Contents
Part VIII
29 Fatigue of 2D and 3D Carbon-Fiber-Reinforced Polymer
Matrix Composites and of a Unitized Polymer/Ceramic
Matrix Composite at Elevated Temperature . . . . . . . .. . . . . . . . . . . . . . . . . . . . 873
M.B. Ruggles-Wrenn and M.P. Wilkinson
30 Carbon Fibers in Tribo-composites.. . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 909
Klaus Friedrich and Li Chang
Part I
Chapter 1
50 Years in Carbon Fibre, 60 Years
in Composites
Robert D. Adams
Early postwar cars were, to say the least, not well made. After less than 10 years,
there were serious problems with paint defects and corrosion. Most private owners
could not afford to take the car to a garage so they chose to do their own repairs.
Paintwork was relatively easy to deal with, but where the steel was corroded
completely through, the only solution was welding and an expensive garage repair.
But then, in the mid-1950s, car accessory firms began to market glass fibre repair
kits, using woven or random glass fibre mat and a polyester resin. Thus, it was that
even schoolchildren could repair a car with GRP. The resin had a catalyst and a
hardener that had to get mixed accurately and thoroughly and then applied briskly
before the resin began to gel. First, the surfaces had to be cleaned and dried to bare
metal, any moisture providing a sure recipe for disaster. On the other hand, any form
of hand-applied rust-preventing chemicals onto the bare metal was faith in untried
chemistry and usually led to failure of the bond quite quickly. The second lesson
was to mix the polyester ingredients quickly and accurately. Third, the glass fibre
mat had to be impregnated thoroughly before the mixture began to gel. Then, after
the GRP had hardened, it could be smoothed with a grinder and wet and dry emery
paper prior to painting. And here was the fourth lesson; the system was suitable for
painting when a film of water covered the surface. This, of course, we know now as
the water break test.
In 1961, the Royal Aircraft Establishment at Farnborough, UK, was interested
in using filament wound glass fibre composites for rocket motors. At that time,
voids and a controlled volume fraction. It was noted that visually poor specimens
had a lower E modulus and higher damping than good specimens. But as only good
material was wanted, most of the defective material was discarded.
In about 1970, there was an important breakthrough. Sheets of fibre could now be
obtained which were pre-impregnated with resin which was just sufficiently cured so
as to remain holding the fibres in place between paper sheets which had been coated
with a release agent. It was now possible to make laminated plates with fibres in
any direction. Initially, there was an excess of resin which had to be squeezed out
as in the original leaky mould technique, but soon the resin/fibre proportions were
carefully controlled so that plates could be made without any squeeze out.
Of course, there had to be care in the symmetry of the laid-up laminae, or the
cured sheet would twist or bend (or both) when released from the mould. This is
because of the low coefficient of thermal expansion (CTE) in the fibre direction but
a far from zero CTE in the transverse direction. When the resin is liquid or soft, there
is a stress-free condition. If cure takes place at room temperature, only resin cure
shrinkage can cause any residual stresses. But since most cure cycles involve cure in
excess of 100 ı C and often approaching 200 ı C, significant stresses will be built in to
the laminate when it is cooled. A suitably symmetric laminate will not twist or bend,
but the stresses between the laminae are there nonetheless. These interply laminate
stresses are always present in composites as the highly cross-linked cured resins
will have little or no creep. Thermal cycling, such as will occur with an aircraft, will
induce fatigue stresses between the plies. The laws of physics do not change.
So at the end of the beginning, to paraphrase Winston Churchill, there was a
lot of experience in manufacturing composites, particularly with the new prepreg
materials. It was now possible to make mainly good material with fewer and
fewer defects, and on the original idea of measuring the damping and moduli
of composites, we could now be embarked. But another line of inquiry has also
appeared. If defective composites have lower moduli and higher damping than
good composites, there is a potential for a means of non-destructively assessing
the quality of a component ab initio and also as it is damaged by fatigue, impact,
and so on. Both lines were therefore pursued.
But the “end of the beginning” was nearly the end of it all as the exciting work
at Rolls-Royce on the RB 211 blades came to a spectacular end. Although most
of the stress and fatigue requirements had been met, it was found that the blades
could not withstand “bird strike” in which a bird the size of a chicken is fired at
several hundred miles per hour at the engine. Unfortunately, the blades shattered;
many suggestions were made as to how to deflect the bird or otherwise minimise its
devastating impact, but none succeeded and the project was cancelled. Even so, it
must be said that this CFRP blade project was far in advance of anything else at that
time, and it was many years before anything so ambitious was tried. Fortunately,
Rolls-Royce had been working on a titanium fan blade in parallel with the CFRP
blade, so the advanced three-spool RB 211 engine project continued, eventually
morphing into the world beating Trent series.
1 50 Years in Carbon Fibre, 60 Years in Composites 7
In vibration, the main problems occur when the excitation frequencies coincide
with one or more natural frequencies of the structure; this is called resonance.
Vibrational energy from the power plant is fed into the structure such that, at
resonance, large vibration amplitudes occur, giving problems with fatigue and
acoustic radiation (noise). The only satisfactory method of controlling resonance
is via some means of dissipating the energy; this dissipation mechanism is called
“damping”. There are many sources of this damping, such as the structural material
itself, transmission to the surroundings or friction between structural elements.
There are several commonly used units of damping, and, especially for materials, the
commonly used unit is the specific damping capacity (SDC) which is defined as the
ratio of the energy dissipated over a cycle of stress to the maximum stored energy.
Figure 1.2 gives the definition of the relationship between SDC and other commonly
used definitions of damping. Purpose built rigs were made and developed over the
years for measuring damping in torsion and flexure as no commercial apparatus is
available.
The aim was to measure the damping and moduli of the fibres and the resin
matrix and then to use micromechanics to predict the damping of the composite. At
first, only unidirectional material was used, made with liquid resins and dry fibres.
Tests were in bending so that the fibres were in direct tension/compression or in
longitudinal shear about the composite axis. The volume fraction could be varied
in the 40–60 % range. Flexural test specimens were excited electromagnetically at
resonance in the fundamental free-free mode of vibration. The beam was supported
at its nodes on cotton fibres or low-density rigid polyurethane foam knife edges to
minimise force transmission and hence energy loss to the surroundings. Cantilever
vibration is favoured by some authors but suffers from slipping in the clamp
which gives a reduced (and therefore incorrect) modulus and significantly increased
damping. A torsion pendulum was used for the shear tests. One end of the test
specimen was clamped to the top of a rigid frame and the other end to an inertia
bar. A dummy specimen, made from a high-strength aluminium alloy which was
known to have very low damping [5], was similarly attached to the inertia bar
and the base of the frame to restrain flexural motion. The inertia bar was driven
electromagnetically at resonance by attached to produce torsional oscillations of the
fixed-fixed system. The specimen ends were enlarged to give effective gripping. A
very preliminary investigation was published in 1969 in the Journal of Composite
Materials [6]. One of the results was the variation of the longitudinal shear modulus
with volume fraction. The experimental results fell between some rather crude limit
values of the only micromechanics theories then published. But the JCM editor,
Steve Tsai, sent the paper back with an as-yet-unpublished theory by D. F. Adams
and Doner [7] which went straight through the experimental results. This new theory
was readily incorporated!
Subsequent papers [8–11] extended and polished these initial results in an
attempt to correlate predictions of damping and modulus using our own and
others’ micromechanics theories. One problem was that although the longitudinal
E modulus (and strength) of carbon fibres was easy to measure on single fibres,
there were no directly measured values available for the transverse (radial) modulus
or the longitudinal (shear) modulus of single carbon fibres. Even to this day,
the transverse modulus can only be inferred by back calculation from some
micromechanics theory. However, the longitudinal shear problem was solved, and
direct measurements were made of the longitudinal shear modulus and damping
of a range of carbon fibres [12, 13] which had been heat-treated at a range of
temperatures (1000–2800 ı C) and supplied by RAE Farnborough. Some of the
results of this work, Fig 1.3, showed that the longitudinal shear modulus was almost
constant at about 17 GPa, whereas the longitudinal Young’s modulus doubled from
about 200 to 400 GPa. The damping (SDC) was small at about 0.13 % over the range
of heat treatment temperatures. It was assumed that the fibre damping was very
small or zero in the longitudinal direction. Again there were no directly measured
data for the transverse direction.
What Adams and Bacon showed was that while micromechanics could be used
to predict moduli, predictions of damping were less accurate and under-predicted
the measured results by a wide margin (30–50 times for some authors [14, 15]).
While uncertainty in measuring the volume fraction and the difficulty of ensuring
perfect fibre alignment were no doubt contributory factors, these could not account
for the massive discrepancy, and it was felt that there must be some other features
which were influencing the situation. Prime suspects were the interaction or possible
slippage at the fibre-matrix interface and the nagging suspicion that the matrix was
not the same when cured in the presence of fibres. There were also the mysterious
properties and dimensions of the silane surface coatings (sizing) applied to the
carbon and glass fibres. With prepreg, it was impossible to know the exact nature of
1 50 Years in Carbon Fibre, 60 Years in Composites 9
the resin, and, for commercial secrecy, no manufacturer would release any data on
the resin they used. However, it was still possible to make composites from liquid
resin and dry fibre.
Fibreglass Ltd agreed to supply glass fibres with a diameter ranging from 10
to 50µm. Using the leaky mould technique and a polyester resin, it was possible
to make specimens with volume fractions ranging from 30 to 65 % [16]. To allow
for changes in fibre diameter at constant fibre volume fraction, the surface area of
fibre per unit volume, A, was defined. It was found that the fibre diameter had no
significant effect on Young’s modulus in flexure, which is not surprising as the strain
energy is overwhelmingly stored in the fibres in tension and compression. There was
also no significant effect on the longitudinal shear modulus and damping. A typical
value for the shear modulus of glass fibres is 8.3 GPa, which is about six times
that of the resin. Increasing the volume loading increases the strain energy stored
in the fibres and decreases that stored in the matrix; it also increases the stiffness
of the composite. The fibre/matrix bond is not highly stressed, so, provided it does
not fail, there is no reason why a difference in A per se should have other than a
small effect on the damping, this being masked by the high overall level. However,
there was a pronounced variation of flexural damping with fibre diameter and fibre
volume fraction and hence with A. The smaller was the fibre diameter, and hence
the larger the surface area for a given volume fraction, the higher was the damping
as shown in Fig. 1.4. But even with the 50µm fibres, there was still a large difference
between the law of mixture prediction and the measured values. Fibre surface area
was obviously part of the damping discrepancy, and this would account for resin
changes near to the fibres and to any effects due to the silane coating which was
applied to all glass fibres at the time of manufacture. The discrepancy remained a
mystery; it was probably due to non-linear shear in the matrix, but the matter was
never clarified satisfactorily. As much as could be done had been done with trying
10 R.D. Adams
2.4
2.2
2.0
+
1.8 +
1.6 +
1.4 +
ψF (%)
1.2 +
+
1.0
0.8
0.6 Law of mixtures
d = 10µm
0.4 + d = 20µm
d = 30µm
0.2 d = 50µm
0
0.1 0.2 0.3 0.4 0.5 0.6 0.7
v
Fig. 1.4 Variation of specific damping capacity in flexure with fibre volume fraction v for different
diameter glass fibres
Fig. 1.5 Variation of flexural Young’s modulus and damping with fibre orientation for HM-S
carbon fibre in DX209 epoxy resin, v D 0.5
were sent to Bristol, some of whom joined my team, and these took up the challenge
of making, analysing and testing vibrating plates. A method was developed for
predicting the natural frequencies, mode shapes and vibration damping parameters
of laminated composite plates. Experimentally, it was a major challenge to support,
excite and measure the vibration without affecting the natural frequencies or mode
shapes and, particularly, the measured damping. The theoretical method was based
on the finite element technique using a damped element model, first proposed in
Bristol [10] and now extended to include the effects of transverse shear deformation
and rotary inertia [17–20]. The results were obtained using a fairly coarse mesh
comprising 36 elements. This was the first time that the damping values of free-
free (i.e. the edges were not clamped or simply supported) laminated composite
plates had been presented both theoretically and experimentally, and the agreement
obtained was generally excellent. The significance of the difference of damping
values from different mode shapes and fibre orientations was emphasised. The
damping values for mode shapes in which there is a lot of twisting are greater
than for those in which the large majority of the strain energy is stored in
tension/compression in the fibres and not in matrix tension or shear. The results
showed that the finite element technique using the damped element model is a
good general purpose tool for the analysis of structures fabricated from composite
materials. Figure 1.6 shows a typical set of results from this “Chinese” period.
12 R.D. Adams
2 279.14 0.93
(272.0) (0.88)
387.8 0.63
3
(372.3) (0.65)
4 432.57 1.23
(407.8) (1.26)
5 511.43 0.98
(486.1) (0.99)
800.37 0.92
6
(779)
Outer layer
Fibre direction
conjunction with a French satellite company [29]. They were keen to know if there
was any significant damping in the CFRP-skinned sandwich panels they used in the
satellites so as to limit the vibration amplitudes during the launch phase, but also
when deployed in space. The panels were extremely light and stiff. At atmospheric
pressure, acoustic radiation provided most of the damping and it was difficult even
to excite above the third mode. Fortunately, in connection with Bacon’s work, a
large vacuum chamber had been made which could be used to reduce the acoustic
radiation to a very small value. The practical answer to the satellite vibration
problem was that in the early stages of launch, damping would be quite high due
to acoustic radiation, but as the air pressure drops, so does the damping.
One important message which came out very clearly from the wide ranging work
carried on damping was that the frequency dependence was very small provided
the composite was 20 ı C or so below the main glass transition temperature, Tg.
Since composites are rarely if ever used near Tg, it can be stated unequivocally that
the damping of composites made using epoxy, polyester and even thermoplastic
matrices is NOT viscoelastic in nature as there was no significant frequency
dependence. This was quite fortunate as finite element models could be used
without time-dependent functions. It is also unfortunate for all those advocates of
viscous damping which, while the equations are easier, produce wrong answers
and predictions. Sadly, even major commercial FE programmes use this viscous
approach which is taught in theory classes in universities. While this theory works
well with the sort of dampers used in motor car suspensions and even for viscoelastic
damping treatments, it must not be used with CFRP (or metals). The viscous
equations may be nicer and easier to use, but they are wrong.
Early manufacturing attempts produced some good and some bad material. Even
material which looked good sometimes had above-average damping and below-
average modulus due to unseen imperfections (visible in GFRP but not CFRP).
This posed the possibility that these dynamic properties could be used to sort
good from bad material non-destructively. As manufacturing techniques improved,
defective specimens became fewer and fewer. However, the concept of using
dynamic measurements to reveal damage or other manufacturing defects opened up.
Such defects could include fibre misalignment, poor cure, fatigue or impact damage
or environmental attack.
Initial work with AERE Harwell and subsequently [30–33] using the torsion of
unidirectional CFRP and GFRP bars showed that:
1. The measurement of shear modulus was a clearer and more discriminating test of
quality in the sense of fibre/resin ratio, type of fibre and fibre surface treatment
than the measurement of damping alone.
14 R.D. Adams
5.1
4.9
spec. 291
virgin data
4.8 N = 2,400Δτ = 53MN/m2
12,200 ,,
,,
16,600
20 ,,
20,500
31,000 ,,
36 500 ,,
18
16
S.D.C. %
14
12
10
Fig. 1.7 Effect of torsional fatigue cycling on the dynamic modulus and damping of a unidirec-
tional specimen
fibre-matrix bonding produced high levels of acoustic emission at low shear loads
and failed prematurely. Where one shear crack was the source of emissions, the
number of counts was directly related to the area of crack surface created during the
loading until the crack becomes so large that significant relative movement between
the crack surfaces occurs. The number of emissions prior to failure depended on
the initial defect size. Nominally identical components may give widely differing
numbers of counts to failure if the initiating defects are not identical. To eliminate
the effects of emissions from randomly aligned fibres, repeated initial proof loading
should be made.
Since composites, rather like wooden pit props, creaked and groaned well before
fracture, there was a growing study of acoustic emission in the UK. Bristol (with
Plymouth Polytechnic) was not slow to get started in this because of its relationship
to non-destructive testing. The main conclusions [34] were that composite materials
vary widely in their load-bearing behaviour and acoustic emission response. In fibre
volume fraction composites, two types of acoustic signal can be observed. Low-
level emissions are generated by deformation of the matrix material and high-level
emissions which can be associated with macroscopic events such as fibre failure,
matrix cracking, interface failure and fibre pull-out. Under carefully controlled
geometrical and loading conditions, it is possible to set the threshold level of
the equipment such that one count corresponds to an individual fibre breakage.
However, in real composites, the results are far more complex since combined
modes of failure occur. With these materials, acoustic emission can be observed
at a relatively low level of loading from the breakage of non-aligned fibres and
from interface failure, prior to any major strength degradation. In addition, the rate
of increase of acoustic emission can be broadly related to material degradation,
but quantitative relationships rely to a large extent upon successful failure mode
identification. But in the mid-1970s, the UK funding was largely stopped so that a
certain government facility could sort out this problem once and for all. Sadly, the
long awaited report was something of a damp squib, so while the Americans pushed
ahead with acoustic emission, most UK researchers pursued other lines.
Later work [35, 36] at Bristol (again with Plymouth) with 0 and 90ı specimens
vibrated in flexure showed that small cracks in GFRP beams could be successfully
detected by measurement of specific damping capacity. These measurements indi-
cated an unambiguous change in material condition after acoustic emission had been
recorded and before any crack was visible to the naked eye.
So how could defects be found in components or structures as distinct from
specimens? Using vibration tests, two separate directions were possible, damping
and natural frequency.
Measuring the local energy dissipation is not easy as although the energy
dissipation is local, it can only be measured as an overall effect on the damping of the
component. Damping, however, produces heat, even with metals [37], but covering
the surface of a plate with thermocouples was not a practical solution. Fortunately,
the AGA Thermovision 680 system became available at this time. It consisted of
a camera which was sensitive to the infrared emissions from a body and produced
a signal which could be displayed to show a ‘thermal’ image of the structure. The
16 R.D. Adams
system then available could only detect temperature differences greater than 0.2 ı C
and this provided a limit to the sensitivity of the test. Nonetheless, by tuning the
vibrating component to resonance and so getting high amplitudes of vibration, it
was possible to locate damaged areas on plates and bars [38, 39]. Because of their
higher thermal conductivity, carbon composites were more of a problem than GFRP.
Although damage causes smaller changes in modulus than damping, the mea-
surement of natural frequencies can be very precise. Hence, changes in resonant
frequency were easy to measure. A finite element model of a vibrating plate was
constructed using 36 elements for a square plate so that it was possible to predict
the natural frequencies of the first six modes of vibration. If one element (or a part
thereof) is caused to be “damaged” by locally reducing the stiffness, there should
be a change of the natural frequencies which will be different depending on where
the damage is located. The natural frequencies of the plate were measured before
and after applying some sort of physical damage and the damage site located by
computing an error function at each of a series of grid points which was a measure
of the error in assuming the damage to be at that point. The position where the error
is a minimum gives approximately the position of the damage. Plates of various
shapes were modelled and tested, and excellent predictions of the actual damage
site and even some measure of the damage were obtained [40–42]. A typical result
is given in Fig. 1.8.
2 3 6 5
12 8 3 3 3 6 8 4
7 10
3 4
16 11 4 9 4 19
6 4 10 4 12
3 23
6 16
22 14 6 7 14 19 17 4 25
4 16 21
11 4
34 9 8 14 25 38 31 3 15 13
10 6 10 11
66 25 33 50 20 4 5 7
7 11 9
10 7
10 7 4 3
34 39 11 20 29 14 13 5
11 2 3
20 4
43 45 18 23 12 20 28 2 2
12
43 1
100 79 28 31 39 51 31 12 4
31 18 9 7
53 64 20 35 32 30
18 9 5
33 38 14 16 15 15
6 7
23 27 21 18 8 8
10
20 20 15 9 5 9
area of blister
Fig. 1.8 Location chart for CFRP sandwich panel with burn damage. The higher the number (up
to 100), the more likely is it the position of the damage
1 50 Years in Carbon Fibre, 60 Years in Composites 17
Although better production methods now existed, errors could still be made,
especially in filament winding which could be readily detected by measuring
natural frequencies [43, 44]. One curious artefact sent to the laboratory was a
small sandwich panel from a military aircraft which had been rejected although
the reasons for this were imprecise. A finite element model was made and the
frequencies duly predicted. Unfortunately, it was impossible to get agreement
experimentally and theoretically and certain frequencies were very “wrong”. The
only possibility was that one of the prefabricated CFRP skins was the wrong way
up. This was plugged in to the FE model and suddenly there was excellent agreement
between theory and experiment. The case was proved by gently grinding away
the CFRP skins to check the fibre alignment. It is interesting to note that this
manufacturing error could not be revealed by X-rays or by ultrasonic inspection.
But vibration worked!
In the late 1970s, the Technical Director of Bristol Composite Materials (Avon-
mouth) issued a challenge to explain how the coin tap test worked. It was common
knowledge that it was possible to locate defects, especially in sandwich panels,
by listening to the sound emitted. Both Airbus and Boeing used to specify their
own tap test tool, particularly for sandwich panels, but it is not known if they still
do this. In the late 1970s, Yu Ermolov in the USSR devised an electromagnetic
tapper with a microphone to detect the sound. This was then analysed by a rather
primitive but effective frequency analyser and the result displayed on a screen.
What he did not realise was that the primary information was contained in the
frequency analysis of the force pulse. A unit was researched and devised [45, 46]
which was called the Tapometer to produce a controlled tap with more advanced
(than Ermolov’s) frequency analysis electronics. By linking the Tapometer with
an X–Y recording table, it was possible to map defects in plates as shown
in Fig. 1.9. It had the advantage of operating dry, whereas C-scan ultrasonics
required immersion in water. The invention was taken by BTG (an arm of the
government) and licensed to Mateval and subsequently to Rolls-Royce. At the
same time, the Japanese “Woodpecker” system was introduced. The Woodpecker
only measured the force pulse magnitude (and not the frequency content) but
produced a more consistent initial tap and was marketed more vigorously. So
it won!
Structures made from composite materials contain a variety of defects. It is
important to establish a good working knowledge of the likely defects in composites
and joints and how to detect them. Adams and Cawley [47] listed the most likely
defects and the common techniques for detecting them. While it is not claimed that
this list is exhaustive, most of the likely defects in composites (and adhesive joints)
have been described.
18 R.D. Adams
Attack by aggressive environments was known about from the early days of
composites and a lot of experimental data was accumulated. This showed that [31]
while carbon ILSS specimens could recover when dried, glass did not. This was
believed to be related to the nature of the bond of the fibre to the surface and whether
it was physical or chemical. While dynamic mechanical/thermal analysis was an
accepted technique for assessing moisture effects, especially on Tg, it suffered from
the problem of drying the specimen as the test proceeded so that the results were
uncertain. Two approaches were used to overcome the drying problem. If moisture
affects the main Tg, it should also affect the beta and gamma transitions which occur
at lower temperatures. The beta transition [48] was studied and it was found that the
changes were more to the height of the damping peak rather than the temperature
at which it occurred. The second approach was to measure Tg where the specimen
could be heated [49] to 100 ı C or so and cooled to 20 ı C in about a minute so as
not to advance the cure or to dry the specimen significantly. This method used a
feedback control system [37] to maintain the specimen at resonance, measuring the
damping to indicate at which temperature was Tg.
It was found that moisture uptake was not simply by diffusion of water into the
matrix and it appeared that there was evidence of osmotic processes occurring.
Further work [50] on epoxy resins and PEEK thermoplastic confirmed our much
earlier work on ILSS specimens in that carbon fibre composites were much less
affected than glass fibre composites. But it also showed that some epoxies were
1 50 Years in Carbon Fibre, 60 Years in Composites 19
Most metallic structures contain riveted or bolted joints. Slippage at these joints
was the major, but uncertain, source of vibration damping. But it was known from
the beginning that structures built from composite materials were weakened by
point loads, so adhesive bonding must be used. However, bonded joints would not
slip, so what would be the source of damping other than the composite itself or
acoustic radiation? As adhesives are polymers, they might provide useful damping.
As damping is quantified as energy dissipation as a function of the stored energy, it
was necessary to know the stored energy and hence the stresses. It quickly became
apparent that there was little knowledge of these stresses or of the damping and
dynamic moduli of adhesives. At about this time (1970), finite element analysis
had been “invented” by O. Zienkiewicz [51], and it was enabled by the rapid
development of digital computers. It was therefore possible to predict the stresses in
bonded joints [52–54] and even to compare them with predictions (albeit elastic) by
using rubber models.
In particular, it was noted the high stress concentrations near the joint ends and
the effects of a spew fillet on these stresses. The implications for the bonding of
composites were becoming understood. An early paper [55] using, then state of
the art but crude by today’s standards, non-linear finite element analysis examined
the efficiency of single and double lap joints for bonding metals to composites.
Rectangular, scarf and bevel laps were used but the greatest gains came from
improving the ductility of the adhesive. What this preliminary work showed was
that joints involving composites need to be considered in a different way than when
metals alone were the substrates.
Using later and more effective finite element analysis with experimental results
from PERME (Waltham Abbey), it was found that, by modifying the geometry of
the double lap joint in the critical regions at the edge of the overlap, significant
increases in the strength of the joint could be achieved [56]. In particular, by a
combination of adherend tapering and including a fillet of adhesive at the edge
of the overlap, a threefold increase in joint strength was obtained, giving a joint
efficiency of approximately 73 %. The finite element model of the joint showed that
this increase in strength was achieved because the concentration in transverse tensile
stress in the CFRP was reduced as shown in Fig. 1.10. For the same applied load, the
transverse tensile stress in the CFRP has been reduced from 30 MPa in Fig. 1.10a
to 6 MPa in Fig. 1.10b. The taper and fillet prevented interlaminar failure in the
CFRP and failure in the modified joint now initiates in the adhesive. Therefore,
in principle, the observed increase in joint strength may have been predicted from
the finite element analysis of the joint geometries, requiring only knowledge of the
20 R.D. Adams
Fig. 1.10 Transverse tensile stresses on CFRP adherend showing (a) the effect of no modification
and (b) the benefit of an internal taper and an adhesive fillet
material properties of the component parts. In essence, the key to the successful
use of composites in load-bearing situations depends on how to transfer the loads
without inducing tensile stresses in the composite which are transverse to the fibres
as these cause delaminations. The simple laws of physics and mechanics tell us that
the transverse tensile strength of a composite is that of the matrix, factored down by
the stress concentration effect of the fibres. In other words, if the tensile strength of
the matrix is 80 MPa, the transverse tensile strength of the composite is going to be
less than this.
Later work [57–61] applied these principles to high-strength joints and especially
how to bond unidirectional CFRP to end tabs in a testing machine so that failure
occurred in the composite and not in the end tabs and even how to control the
direction of the crack by using a Kapton film.
When carbon fibre composites are bonded to metals, significant stresses are
caused by curing from high temperatures due to differential coefficients of thermal
expansion (CTEs) [62–66]. Further thermal stresses can also be caused by operating
conditions. Commercial aircraft cruise from 50 to 60 ı C, while supersonic
military aircraft can experience skin temperatures of the order of 200 ı C at Mach
2 and above. Operating over a wide temperature range poses particular problems
as many high temperature adhesives are brittle at 50 ı C. By using a dual
1 50 Years in Carbon Fibre, 60 Years in Composites 21
50
45
40
Failure load (kN)
35
30 Redux 326
Supreme 10HT
25
MAJ1-1
20 MAJ3
15
10
5
0
-100 -50 0 50 100 150 200 250
Temperature (ºC)
Fig. 1.11 Experimental failure load for titanium/composite DLJs. The error bars correspond to
the maximum and minimum values
adhesive concept with low temperature adhesive at the ends of the joint and a
high temperature adhesive in the middle as shown in Fig. 1.11, it is possible to
design a suitable lap joint which can withstand extremes of temperature [63, 65].
A major problem is the availability of suitable adhesives, especially to find a low
temperature adhesive which can withstand the high cure temperature of the high-
strength adhesive without charring. The viability of this dual adhesive concept was
shown both experimentally and by using finite element analysis.
In the 1970s, Steve Tsai said that carbon composites would only become viable
when they were flown in major commercial aircraft, not just in military aircraft.
And now in the very early twenty-first century, the Boeing 787 and various Airbus
projects are flying commercially with major CFRP parts such as wings. Motor sport
uses CFRP extensively for its strength and stiffness, and the crash-resistant box
22 R.D. Adams
surrounding the driver has made racing cars very much safer than those used in
earlier years. More mundane sports such as rowing, fishing and tennis have also
seen significant performance enhancement by using CFRP.
The one area where CFRP has made little commercial penetration is in mass-
produced motor vehicles. Increasing awareness about environmental issues includ-
ing energy consumption and material costs at all stages of product life is putting
industry in general and the transport industry in particular, under increased pressure
to reduce CO2 emissions. The transport industry is estimated to generate 22 %
of Europe’s CO2 , and its expected growth means that steps have to be taken to
improve technologies so that we can reach international targets of limiting CO2 .
Public service vehicles, buses, trains and trams play a major role in the transport
industry, and this is expected to increase as a shift is encouraged from private cars.
Multimaterial technology uses sandwich and composite materials. This technol-
ogy can be used to improve the specific strength, stiffness and crashworthiness
of vehicles. Some 10 years ago, there was a European project under the name of
LITEBUS (with 13 partners) which was focussed on the development of novel
technology to manufacture bus/coach bodies, so as to reduce the weight and
production costs and to produce a structure which would meet the EU regulations
on safety in the “roll-over” test. The Oxford contribution to this project was to lead
the joining work package which was essential to providing a satisfactory means
of linking the various sandwich panels and structural composite members together
and to the steel sub-chassis. Figure 1.12 shows the parts to be assembled and
the supporting frame which is part of the engine/transmission/suspension system.
Adhesives had to be selected to account for manufacturing constraints such as
temperature, assembly methods, working time (how long the adhesive remains
usable), gap filling and suitability for the different surfaces to be bonded, as well as
the expected variables of strength, ductility and stiffness. The joints were analysed
using finite element techniques to determine the stresses and strains and to locate
“hot spots” where redesign might be necessary. The coach section was built to full
scale by Caetanobus in Porto. Having chosen the best adhesive for each joint in
accordance with the substrates and the loads to be carried, it was then necessary
to compromise considerably to account for the “mundane” constraints given above.
Acrylic adhesives had excellent strength and ductility, but the reaction times were
so short that assembly of the large composite parts was impossible.
Finally, the coach structure was subjected to the roll-over test and it passed! The
structure remained largely intact and the deformations were such that the passenger
space was not intruded into. It was shown that composites could be successfully
used to make the passenger section of a coach using prefabricated components.
The question as to whether weight or manufacturing costs were saved cannot be
answered by such a limited (only two million euros) project, but the structural
feasibility and method of manufacturing was proven.
So what is happening in passenger car production? Apart from a few “supercars”
where cost is largely unimportant, there has been no significant introduction of
composites. Part of the problem is the intense rivalry and lobbying by the steel and
aluminium producers to devise multimaterial car bodies with increasingly complex
1 50 Years in Carbon Fibre, 60 Years in Composites 23
Fig. 1.12 The EU LITEBUS project. Exploded view of the composite panels to fit on the steel
sub-frame
steels and aluminium alloys. Problems of joining these different materials are not
insignificant and CFRP just adds another variable. At about the same time as the
LITEBUS project, there was the EU Super LIGHT-CAR project in which a car
similar to a VW Golf was conceived. By weight, this concept body shell contained
96 kg of aluminium, 66 kg of steel, 11 kg of magnesium and just 7 kg (less than
4 %) of glass-reinforced thermoplastic. Of CFRP, there was no mention. Recently,
some of the more advanced auto-manufacturers are beginning to incorporate CFRP
to try and reduce weight. At a recent conference, it was remarkable to see that
while there was some adhesive bonding, blind riveting was being used extensively
to make CFRP/CFRP and CFRP/steel joints. CFRP seems to be being treated as a
black metal! Apart from delamination damage done to the CFRP during riveting and
problems with corrosion, the concept of point loading of CFRP in stressed structures
seems to fly in the face of all that had been learnt with composites over the last 50
years.
So, after 50 years, we are making large aircraft with major CFRP parts, and car
manufacturers are dipping their toes in the water (not to mention the very large
numbers of glass FRP boats), but what happens at the end of life? CFRP can be
burnt, but can it be recycled? Towards the end of my career at Bristol, we looked
at the possibility of recycling composites, including carbon. The aim was to avoid
chopping the end-of-life composite material into tiny pieces and using them as filler,
or burning, or by using complex and costly methods [67–69] of trying to separate the
fibres from the matrix. Instead, a combination of thermal and crushing processes was
24 R.D. Adams
used [70], the aim being to retain as much as possible of the mechanical properties
of the original material together with the chemical feedstock. The recyclate could
then be reformed or remanufactured into another component, keeping some 50 %
of the original strength and stiffness. It was shown that initially curved components
could be made flat, as shown in Fig. 1.13, and flat ones could be remanufactured
into tubes. CFRP was more difficult to remanufacture than glass fibre composites
because of its greater stiffness, but it can be done by the methods outlined above
and given in [70].
The European automotive industry has strict regulations on the recycling of
end-of-life vehicles. Steel and aluminium are relatively easy to recycle, provided
they can be separated easily. But composites pose an entirely new and more
difficult problem. Thermoplastic matrix composites offer easier possibilities than
thermoset matrices. Perhaps this recycling problem, coupled with the lobbying of
the aluminium and steel industries, is what is delaying the large-scale introduction
of CFRP into automotive body shells. The ideas suggested above [70] might just be
the beginning of a new recycling and remanufacturing industry.
Altiora etiam petamus
Acknowledgements It is impossible to list here all the appropriate co-workers in this account,
so I have not included any; the names will be found in the papers referenced. But this work has
been made possible by the enormous contribution of two technicians who supported the team over
several decades. These are Brian Smith, who sadly died some years ago, and John Skinner. They
made and maintained much of the in-house equipment we used, as well as the specimens we tested.
1 50 Years in Carbon Fibre, 60 Years in Composites 25
References
23. R.D. Adams, M.R. Maheri, Finite element prediction of the modal response of damped layered
composite panels. Compos. Sci. Technol. 55, 13–23 (1995)
24. R.D. Adams, M.R. Maheri, The dynamic shear properties of structural honeycomb materials.
Compos. Sci. Technol. 47, 15–23 (1993)
25. R.D. Adams, M.R. Maheri, Dynamic flexural properties of anisotropic fibrous composite
beams. Compos. Sci. Technol. 50, 497–514 (1994)
26. R.D. Adams, M.R. Maheri, Steady state flexural vibration damping of honeycomb sandwich
beams. Compos. Sci. Technol. 52, 333–347 (1994)
27. M.R. Maheri, R.D. Adams, On the flexural vibration of Timoshenko beams, and the applicabil-
ity of the analysis to a sandwich configuration. J. Sound Vib. 209, 419–442 (1998). ISBN/ISSN:
0022460X/98030419
28. M.R. Maheri, R.D. Adams, Modal vibration damping of anisotropic FRP laminates using the
Rayleigh-Ritz energy minimization scheme. J. Sound Vib. 259, 17–29 (2003)
29. M.R. Maheri, R.D. Adams, J. Hugon, Vibration damping in sandwich panels. J. Mater. Sci. 43,
6604–6618 (2008)
30. R.D. Adams, J.E. Flitcroft, N.L. Hancox, W.N. Reynolds, Effects of shear damage on the
torsional behaviour of carbon fibre reinforced plastics. J. Compos. Mater. 7, 68–75 (1973)
31. R.D. Adams, The effect of steam and hot air on the shear strength of glass and carbon fibre
reinforced plastics. In Proceedings of the Second International Conference on Mechanical
Behaviour of Materials (ICM-2), Boston, MA, 1976, Library of Congress Cat. No. 76–27115,
pp. 1508–1513
32. R.D. Adams, J.E. Flitcroft, The detection of matrix cracks in fibre reinforced plastics and their
effect on shear strength. In Proceedings of the Institution of Civil Engineers Conference on
Fibre Reinforced Materials: Design and Engineering Applications, London, 1977, pp. 79–86
33. J.E. Flitcroft, R.D. Adams, A study of shear crack propagation in glass and carbon fibre
reinforced plastics using acoustic emission monitoring. J. Phys. D. Appl. Phys. 15, 991–1005
(1982)
34. F.J. Guild, D. Walton, R.D. Adams, D. Short, The application of acoustic emission to fibre
reinforced composite materials. Composites 7, 173–179 (1976)
35. F.J. Guild, R.D. Adams, A new technique for the measurement of the specific damping capacity
of beams in flexure. J. Phys. E: Sci. Instrum. 14, 355–363 (1981)
36. F.J. Guild, R.D. Adams, The detection of cracks in damaged composite materials. J. Phys. D.
Appl. Phys. 14, 1561–1573 (1981)
37. R.D. Adams, A.L. Percival, Measurement of the strain-dependent damping of metals in axial
vibration. J. Phys. D. Appl. Phys. 2, 1693–1704 (1969)
38. C.J. Pye, R.D. Adams, Heat emission from damaged composite materials and its use in non-
destructive testing. J. Phys. D. Appl. Phys. 14, 927–941 (1981)
39. C.J. Pye, R.D. Adams, Detection of damage in fibre reinforced plastics using thermal fields
generated during resonant vibration. NDT Int. 14, 111–118 (1981)
40. P. Cawley, R.D. Adams, The predicted and experimental natural modes of free-free CFRP
plates. J. Compos. Mater. 12, 336–347 (1978)
41. P. Cawley, R.D. Adams, The location of defects in structures from measurements of natural
frequencies. J. Strain Anal. 14, 49–57 (1979)
42. P. Cawley, R.D. Adams, A vibration technique for nondestructive testing of fibre composite
structures. J. Compos. Mater. 13, 161–175 (1979)
43. P. Cawley, A.M. Woolfrey, R.D. Adams, Natural frequency measurements for production
quality control of fibre composites. Composites 16, 23–27 (1985)
44. R.D. Adams, J.M.W. Brownjohn, P. Cawley, The detection of defects in GRP lattice structures
by vibration measurements. NDT&E Int. 24, 123–134 (1991)
45. P. Cawley, R.D. Adams, The mechanics of the coin-tap method of non-destructive testing. J.
Sound Vib. 122, 299–316 (1988)
46. P. Cawley, R.D. Adams, Sensitivity of the coin-tap method of non-destructive testing. Mater.
Eval. 47, 448–563 (1989)
1 50 Years in Carbon Fibre, 60 Years in Composites 27
47. R.D. Adams, P. Cawley, A review of defect types and NDT techniques for composites and
bonded joints. NDT Int. 21, 208–222 (1988)
48. R.D. Adams, M.M. Singh, Low temperature transitions in fibre reinforced polymers. Compos-
ites. Part A 32, 797–814 (2001)
49. Y. Zhang, R.D. Adams, L.F.M. da Silva, A rapid method of measuring the glass transition
temperature using a novel dynamic mechanical analysis method. J. Adhes. 89, 785–806 (2013)
50. R.D. Adams, M.M. Singh, The dynamic properties of fibre-reinforced polymers exposed to
hot, wet conditions. Compos. Sci. Technol. 56, 977–997 (1996)
51. O. Zienkiewicz, R.L. Taylor, J.Z. Zhu, The Finite Element Method: Its Basis and Fundamen-
tals, 7th edn. (Butterworth-Heinemann). Print Book ISBN: 9781856176330; eBook ISBN:
9780080951355
52. R.D. Adams, N.A. Peppiatt, The effect of adherend Poisson’s ratio strains on the stresses of an
idealized lap joint. J. Strain Anal 8, 134–139 (1973)
53. R.D. Adams, N.A. Peppiatt, Stress analysis of adhesively bonded lap joints. J. Strain Anal. 9,
185–196 (1974)
54. R.D. Adams, N.A. Peppiatt, Stress analysis of adhesively-bonded tubular lap joints. J. Adhes.
9, 1–18 (1977)
55. R.D. Adams, N.A. Peppiatt. Stress analysis of lap joints in fibre reinforced composite materials.
In Proceedings of the Institution of Civil Engineers Conference on Fibre Reinforced Materials:
Design and Engineering Applications, London, 1977, pp. 45–56
56. R.D. Adams, R.W. Atkins, J.A. Harris, A.J. Kinloch, Stress analysis and failure properties of
carbon-fibre reinforced plastic/steel double-lap joints. J. Adhes. 20, 29–53 (1986)
57. A. Towse, K. Potter, M. Wisnom, R.D. Adams, A novel comb joint concept for high strength
unidirectional carbon fibre bonded joints. In ICCM-11 International Conference on Composite
Materials, Gold Coast, QLD, 1997
58. A. Towse, R. Davies, A. Clarke, M. Wisnom, R.D. Adams, K. Potter, The design and analysis
of high load intensity adhesively bonded double lap joints. In DFC-4 Deformation and Fracture
of Composites, Manchester, 1997
59. R.D. Adams, R. Davies, Strength of joints involving composites. In Adhesion Science and
Technology: Proceedings of the International Adhesion Symposium, ed. by H. Mizumachi
(Gordon & Breach Science Publishers, Yokohama, 1994), pp. 251–262
60. K.D. Potter, F.J. Guild, J.H. Harvey, M.R. Wisnom, R.D. Adams, Understanding and control
of adhesive crack propagation in bonded joints between carbon fibre composite adherends I.
Experimental. Int. J. Adhes. Adhes. 21, 435–443 (2001)
61. F.J. Guild, K.D. Potter, J. Heinrich, R.D. Adams, M.R. Wisnom, Understanding and control
of adhesive crack propagation in bonded joints between carbon fibre composite adherends II.
Finite element analysis. Int. J. Adhes. Adhes. 21, 445–453 (2001)
62. L.F.M. Da Silva, R.D. Adams, Stress-free temperature in a mixed-adhesive joint. J. Adhes. Sci.
Technol. 20, 1705–1726 (2006)
63. D. Silva, F.M. Lucas, R.D. Adams, Adhesive joints at high and low temperatures using similar
and dissimilar adherends and dual adhesives. Int. J. Adhes. Adhes. 27, 216–226 (2007)
64. D. Silva, F.M. Lucas, R.D. Adams, Techniques to reduce the peel stresses in adhesive joints
with composites. Int. J. Adhes. Adhes. 27, 227–235 (2007)
65. L.F.M. Da Silva, R.D. Adams, Joint strength predictions for adhesive joints to be used over a
wide temperature range. Int. J. Adhes. Adhes. 27, 362–379 (2007)
66. L.F.M. Da Silva, R.D. Adams, Effect of temperature on the mechanical and bonding properties
of a carbon-reinforced bismaleimide. Proc. IMechE, Part L: J. Mater. Des. Appl. 222, 45–52
(2008)
67. J.M. Henshaw, W. Han, A.D. Owens, An overview of recycling issues for composite materials.
J. Thermoplast. Compos. Mater. 9 (1996); Technomic Publishing
68. A.K. Bledzki, K. Goracy, The use of recycled fibre composites as reinforcement for thermosets.
Mech. Compos. Mater. 29(4), 352–356 (1993)
28 R.D. Adams
69. J.R. Kennerley, R.M. Kelly, N.J. Fenwick, S.J. Pickering, C.D. Rudd, The characterisation and
reuse of glass fibres recycled from scrap composites by the action of a fluidised bed process.
Compos. Part A 29A, 839–845 (1998); Elsevier Science
70. R.D. Adams, A. Collins, D. Cooper, M. Wingfield-Digby, A. Watts-Farmer, A. Laurence, K.
Patel, M. Stevens, R. Watkins, Recycling of reinforced plastics. Appl. Compos. Mater. 21,
263–284 (2014)
Chapter 2
‘But How Can We Make Something Useful Out
of Black String?’ The Development of Carbon
Fibre Composites Manufacturing (1965–2015)
Kevin Potter
2.1 Introduction
K. Potter ()
Department of Aerospace Engineering, University of Bristol, Queen’s Building
University Walk, Bristol BS8 1TR, UK
e-mail: K.Potter@bristol.ac.uk
Automated tape laying and fibre placement are the only major processes where
development started after the 1950s, and they are really fibre collation processes
2 But How Can We Make Something Useful Out of Black String?. . . 31
Fig. 2.1 Patent drawings relating to composites manufacture from the 1950s
rather than moulding processes; even in this case the earliest known patent dates to
1967 (US3574040 filed by General Dynamics).
The 1969 paper [2] shows many articles of complex geometry being made from
glass-reinforced composites, and these shared one common factor; the glass has
been chopped into fairly short lengths and randomly arranged into a formable
sheet or sprayed directly onto a tool surface or onto a preform tool. To a first
approximation, most of those applications of glass fibre composites were semi-
structural and weight-insensitive structures, primarily taking advantage of the
material’s ability to generate the smooth, thin shells of complex, doubly curved
geometry which are very expensive to fabricate in metal. In this respect glass fibre
composites were perhaps closest to filling the same design niche as papier mâché did
in Victorian England and papyrus cartonnage did in pharaonic Egypt. A very strong
case can be made for cartonnage mummy cases as the first ‘modern’ composite
structures.
This was never the target for carbon fibre composites, which were always
intended for use in the structural applications (primarily in aircraft) for which their
high moduli were essential. McMullen [4] gives a history of the early years of
aircraft composites and [5] gives a more recent overview. To maintain the high
moduli (and avoid a waste of extremely costly material), the assumed design and
manufacturing philosophy was to use the new material in continuous form at a high-
volume fraction of fibres. The slight problem in the very early days was that the tows
of fibre were longish, but not continuous.
32 K. Potter
The early, 1-m long, Morganite carbon fibre made in the mid-1960s came in
packages, carefully wrapped in brown paper, and, when unwrapped, the individual
tows could be very carefully extracted. This dissonance between a demand to
maintain good alignment and a high-volume fraction, and only having relatively
short lengths of fibre, led to the development of the ‘leaky mould’ technique. Tows
of carbon fibre were saturated in resin and very carefully placed by hand into a
mould cavity until the right charge weight has been achieved. The second part of
the mould was then put in place, and a combination of pressure and temperature
was used to consolidate the layup, eliminate excess resin, limit voidage and cure
the resin. This is not a process where it will be easy to achieve a consistent result.
The leaky mould technique has another, probably more critical limitation: it could
only really make unidirectional (UD) specimens for test. This may have contributed
to some of the early overselling of carbon fibre laminate properties, as only UD
properties were readily available. An alternative approach based on impregnation of
tows laid side by side with a solution of epoxy resin in acetone gave an intermediate
material form (unidirectional [UD] prepreg) that was more easily handled and with
which it was possible to make multidirectional laminates. More importantly, with
UD prepreg manufacture could move out of the lab and onto the factory floor to
make objects with a bit of shape to them—but not very much because UD prepreg
is still far from easy to do much with in terms of very complex geometry.
The very first prepregs were made from the original short lengths of fibre, but as
continuous fibre became more widely available, turning it into UD prepreg was the
obvious thing to do. As noted earlier these continuous fibre UD prepregs were rather
tricky to work with, they tended to tear and wrinkle easily when being sheared and
stretched transversely to form onto a surface of double curvature, and the handling
properties were very sensitive to temperature. Far from being easy to apply they
needed very skilled staff to make anything other than flat sheet, and this is as true
today as it was in 1970. The early resin matrices were also very troublesome to
work with, falling to very low viscosities in the heat up to cure and requiring
real care to ensure that the resin flow was not excessive in cure, leading to voidy
laminates [6]. The acceptable process window in many cases was close to zero. The
obvious solutions to the problems with layup on complex shapes could be seen in
the experience with earlier classes of fibre. If the fibres are continuous like glass,
then chop and randomly disperse them or weave them; if the fibres are short such
as asbestos or whiskers, then form them into felts with a level of fibre alignment to
maintain good properties. It took a few years to work out how to weave carbon fibres
without damage, and the techniques developed earlier at ERDE to sort and align
whiskers and asbestos fibres could be applied to the production of high-performance
aligned short-carbon fibre felts.
In 1974 woven carbon fibre textiles were just becoming available; the team
at ERDE had demonstrated that short, chopped carbon fibres could be aligned
extremely well [7], and soon after that, the first of the second-generation prepreg
systems with some thermoplastic additives that stopped viscosity falling too far were
becoming available. These second-generation systems were a revelation; it was no
2 But How Can We Make Something Useful Out of Black String?. . . 33
longer necessary to be extremely careful in designing the bleed pack and exact point
of autoclave pressure application to get a decent quality laminate. They made many
intractable moulding problems disappear overnight.
Most of the author’s early work was in thinking about how to make things. A fairly
typical example was a helicopter tail rotor gearbox, essentially a flanged cone with
local deviations from a simple volume of revolution. It was very difficult to form
the multiple curvatures required from continuous UD prepreg, but much easier to
build if the reinforcement has a degree of manufacturing ductility, for example,
achieved in a highly aligned but discontinuous fibre prepreg [8]. One thing that was
becoming increasingly obvious was that how the reinforcement formed to the tool
was absolutely critical and would largely control the cost-effectiveness of composite
applications. This was the initial interest in understanding drape and reinforcement
deformation properties; as things developed it became clear that it was also critical
from a performance viewpoint—in terms of being able to avoid wrinkled plies.
This interest led to the start of a study in how to understand, model and control
the forming of composite reinforcements, both woven and unidirectional, and later
non-crimp fabrics. The study probably started in the mid-1970s and was reported
on in a couple of papers in 1979 [8, 9]. The most recent paper in this area from the
Bristol University team was published in 2015 [10], and the field is still very much
‘live’. It has taken more than 35 years to reach the current level of understanding,
but it now seems possible to close out this area of study in the not too distant future.
From the beginning there have been two strands to the study of reinforcement
deformation. The first relates to trying to understand the deformation and forming
properties of composites reinforcement from both scientific and practical view-
points. The second relates to how to use that developing understanding to deliver
manufacturing solutions and understand issues of design for manufacture, although
on the academic side, the first strand has been heavily dominant.
Automated solutions to composites manufacture are becoming much more
common, but in the first decade of carbon fibre composites, the only realistic option
was manual layup of unidirectional prepreg. Even today for the fairly small and
complex parts that make up the bulk of aircraft secondary structures, and essentially
everything in motorsport, manual layup still dominates, but generally with woven
rather than UD reinforcements (see Fig. 2.2).
The recognition that the costs and difficulty of manual layup were a major
obstacle to the development of composites was made very early on by several
groups, with the first US patents for what are recognisably automated tape laying
34 K. Potter
(ATL) or automated fibre placement (AFP) machines dating from the early 1970s
[11], and a UK-developed ATL was reported at a conference in 1980 [12].
Interestingly, an essentially identical machine was demonstrated at a major trade
show in 2015 (JEC Paris).
The early conference and journal papers on the drape of reinforcements seemed
to disappear without trace at the time, but slowly, over time, the number of papers
started to grow significantly, with pretty much an exponential growth from 1985 to
2015. As the study of reinforcement deformations became more mainstream, and the
number of papers greatly expanded and became focused almost entirely on woven
cloths, the study tended to narrow down to how to model the drape of woven cloths
over arbitrary surfaces. This is clearly very important, but it must be stressed that
it doesn’t actually directly answer the questions around how to support or improve
the manual layup processes, much less the question of how to improve designs or
automate manufacture. It has taken the whole community, the author included, far
too long to long to understand this. With regard to automation, it is really rather
surprising how little some of the early work attempting robotic layup took account
of the practicalities of how the reinforcement was actually applied to the tool by
the skilled laminators [13]. Equally, designers of composite parts have not generally
taken much account of the difficulties of manufacturing the parts that they design.
Realistically, when manufacturing engineers have been unable to tell designers
what a good, manufacturable design that makes best use of laminator skills and
reinforcement formability actually looks like, to minimise costs or maximise quality,
it is unsurprising that the designers often take little account of manufacturability.
Recent work has been investigating how the experience and expertise of laminators
can be brought into the design process by facilitating structured conversations
between designers and laminators. It has been very instructive how positive the
interactions have been, so it does seem to be a pity that it has taken so long to
engage two of the main sets of players in the composites community.
2 But How Can We Make Something Useful Out of Black String?. . . 35
5000
4000
3000
2000
1000
0
1980 1985 1990 1995 2000 2005 2010 2015
Fig. 2.3 Drape-related paper numbers 1980–2015 (Source: Google Scholar, search string: papers
including all the words “woven fabric drape modelling”)
Fig. 2.4 VFP model output (left): Woven glass cloth laid up on tool (right)
The intention was to use VFP in a design environment to debug and troubleshoot
the design virtually by simulating how each ply could be applied to the tool surface
without defects and then generate totally unambiguous manufacturing instruction
sheets to improve quality and reproducibility. However, further work showed that
although the VFP programme could be used to instruct an operator to proceed in
a specific fashion, it could not, of itself, identify whether that fashion would lead
necessarily to a cost-effective solution or identify what the best approach to layup
would be, although it could be used to avoid the generation of significant defects.
In recent years there has been much more serious work carried out on automated
manufacture along a few tracks in both academia and in industry.
Automated fibre placement machines were developed to do several things, essen-
tially to tackle the perceived limitations of manual layup and drape in manufacture.
The machines would replace manual labour and bring down labour costs, reduce
learning curve effects and lead time, improve quality and reliability, increase kg/h
production rate, reduce materials wastage and hence materials costs, reduce the
requirement for an expanding pool of skilled labour thus debottlenecking production
and lastly not be so dependent on the drape properties of the reinforcement through
the use of narrow (typically 6 mm) tape. In practice the AFP machines have
rather fallen short of their initial promise, although they do still have a considerable
amount of development potential. For example, whilst touch labour has reduced,
a significant level of higher priced machine programming and maintenance has
been experienced. Equally on many lines, whilst the AFP can manage most of the
layup, some hand layup is commonly still required. Materials wastage is certainly
reduced, although the additional cost of tape slitting from broad goods UD can
sometimes more than offset the savings in the costs of materials wastage. If the
38 K. Potter
narrow tapes that are used in AFPs are laid down following geodesic paths (such
as a meridian on a sphere), they will generally conform well to the surface without
significant wrinkling. However, for surfaces other than simple single curvatures,
the requirement to follow a geodesic path is too restrictive, and there is a need to
steer the tapes, which tends to create gaps, and/or overlaps, between adjacent tapes,
which can have structural implications (see Fig. 2.5). In addition to the gaps/laps,
each tow will have to bend in plane to follow the path dictated by the guide/drive
roller. As with any bending, this generates a compressive stress on the inside of the
radius and a tensile stress on the outside of the radius. There is then a tendency
of the fibres to lift off the tool on the outside of the radius and to wrinkle on the
inside of the radius. A complicating factor is that for a solid roller the speed of
rotation will not match the speed at which the tape is being laid down, leading to a
very complicated set of deformations. To date, these issues are primarily tackled by
carrying out experimental steering trials rather than by an attempt to understand the
details of the drape properties of the slit tape. In most cases the limiting minimum
radius will be of the order of 1 m, although the exact radius is more likely to be set
by engineering judgement than by a rigorous assessment of the impacts on cost and
structural integrity.
There are essentially two major types of AFP. The first uses a gantry to support
the layup head (Fig. 2.6a) which offers great stiffness and stability. The other
(Fig. 2.6b) uses a robot to carry the layup head which results in a more flexible
but less reproducible layup. The development and increasing use of AFP machines
for the layup of primary structures in aerospace has been a major driver of work
to understand manufacturability with significant academic work in Europe and the
USA. Figure 2.7 shows some structures manufactured using an AFP. This work has
helped to develop an understanding of the limitations of AFP [20, 21] especially
in terms of geometry and quality/defects. For example the relationship between
as-laid consolidation quality and operating conditions for a range of commercial
ATL/AFP machines was examined, but no such relationship was found. Looking
in more detail, it became clear that the actual control over the operating conditions
on the commercial machines was rather limited, so an ATL simulator was built that
could lay down short lengths of prepreg under very closely controlled conditions.
Using the simulator the relationship became very clear; as the layup temperature was
increased, the quality first declined then improved greatly at higher temperatures
[22]. The high variability seen in the commercially operated machines was driven
2 But How Can We Make Something Useful Out of Black String?. . . 39
Fig. 2.6 (a) Gantry-mounted machine. Note the head carries the tape spools in this case, and the
head itself is very large making it very limited in terms of possible geometry. (b) Robot-mounted
layup head. This is much smaller than the gantry-mounted machine, and the tape spools are held
remotely from the head in a temperature- and humidity-controlled creel cabinet. Image courtesy of
the NCC
Fig. 2.7 Winglet skins manufactured by AFP. Images courtesy of the NCC
by their operating principally in the temperature range where quality was declining.
The quality reduction was due to changes in the surface morphology (roughening)
of the prepreg as it was heated. It is becoming increasingly clear that our current
understanding of the details of the nature and processing of prepreg is still rather
incomplete. In recent years more of the very costly commercial AFP machines are
becoming available to academic researchers, either in their own labs or through
developing relationships with industry and through RTOs such as the UK’s National
Composites Centre. More capable AFP layup heads are also under development to
overcome some limitations of current AFP approaches and can, for example, offer
a tenfold reduction in defect-free minimum steering radius [23].
40 K. Potter
Focus has recently returned to how the production of components that cannot
be made on an AFP might be automated—which, as noted before, turns out to
be most of the small complex parts that abound in aerospace, motorsport and the
wider automotive industry. There had been many attempts to use robots to lay up
composite reinforcements, using either UD or woven prepreg, but no significant
progress had been made, and AFP had come to be seen as the default solution, even
though its geometrical capability is really very limited. One successful approach is
to use robotically laid small patches of prepreg as the small patches are much easier
to conform to the tool than large continuous plies. If the UD prepreg is kept very
thin, the knockdown in mechanical properties can also be minimised. The process
shows great promise in many respects, but achieving high-rate processing is going
to be very problematical.
In order to understand why robotic automation of layup has proven to be so
difficult, it is necessary to carry out a detailed study of how operators actually form
and manipulate woven prepreg onto the tool. This study started more than 30 years
after it had first been recognised that automation would be needed [24, 25]. It proved
to be necessary to develop a structured language which could be used to describe
the details of the layup process. This turned out to be one of the things needed to be
in place to make full use of the VFP software and provide instructions and feedback
to the operators. Alongside this the details of the ways in which the component
geometry and the characteristics of the woven prepreg impacted on the time taken
to complete the layup were studied in more depth. Relatively small changes in
geometry can be shown to make a very significant difference to the layup time,
and choosing the right prepreg from a manufacturing perspective could halve the
layup time and hence the direct labour cost. For the first time, the ability to define
what manufacturable looks like and identify the cost of less manufacturable designs
(or material choices) is becoming a possibility, through an improved integration of
manufacturing understanding and costing methodologies.
One other issue should be aired briefly here. The academic literature (and most
of the more commercial offerings) tends to equate composites manufacture with
composites layup and thus equates automated manufacture with automated layup.
This is far from being entirely correct, even though the layup phase is critically
important. Twelve major process steps, and a lot of paperwork in an aerospace
environment, can be identified from cleaning the mould tool through to inspection
and NDT of the finished product. For a conventional complex part layup, there is
probably only a single automated process—ply cutting. Even if AFP is used for ply
collation, there are still eight manual processes.
There are some tools available to support manual layup processes, the most
commonly used being laser ply projection. First introduced by Assembly Guidance
in 1988, and now widely available from multiple vendors, these systems project the
next ply edge position onto the tool to support the laminator to correctly position
the ply [26]. Prior to the introduction of laser ply projection, the standard industrial
practice was that the position of each ply was defined by manually scribing a set
of location features for each layer through a template or stencil. For simple parts,
the template was made of Mylar. For complex parts, the templates were made of
2 But How Can We Make Something Useful Out of Black String?. . . 41
fibreglass in order to conform to the shape of the tool. Using templates to define
the location of each ply in a laminate required an inordinate amount of process
time, which could be much greater than the amount of time required to actually
place the prepreg on the tool. Templates for parts made up of hundreds of plies had
thousands of marking points, which made the scribing process extremely confusing.
The use of laser projectors to provide the templates greatly improved the efficiency
of layup whilst reducing the scope for error and enabling an increase in the size and
complexity of composite laminates.
Although the laser projectors were developed to support manual layup processes,
they have also been developed in support of automated manufacture. For example,
laser projectors can be used with AFP/ATL where the projected patterns enable
inspectors to verify that the layers of material are in the correct locations (see
Fig. 2.8). In addition, machine vision systems can also be integrated with laser
projectors to automate the inspection of laminates, both in the case of hand layup
and automated layup. This automated imaging enables the verification of ply
boundaries as well as identifying FOD (see Fig. 2.9) and measuring fibre orientation
and shear as the part is laid up (see Fig. 2.10). This automatic inspection and
documentation can be used to provide detailed records of as-built characteristics
which will be of benefit in forming tighter linkages between design, manufacture
and performance.
Another reason for trying to understand manual layup better is that the expected
growth in the composites industry will require the training of a new generation of
both designers and laminators. Better training tools are needed alongside everything
else, as well as better control in the workplace. A few approaches to developing tools
in this area have been demonstrated. Firstly, a low-cost alternative to laser projection
systems that projects a step-by-step layup sequence onto the tool and uses a Kinect
game controller/motion-sensing device to both control the sequencing and capture
layup information on the fly [27]. Secondly, it has been shown to be possible to
42 K. Potter
Fig. 2.10 The fibre directions have been automatically identified and traced from the pattern of
warp and weft captured in this image. Image courtesy of Assembly Guidance
incorporate gaming systems and virtual reality approaches in initial training routines
[28]. Working with the laminators to identify how they approach their work, study
has been made of the simple tools they use to make their jobs easier. Most, if not
all, laminators have a pocketful of these ‘dibbers’, generally homemade from hard
plastic, but essentially uncontrolled. It is very hard to think of any other area of the
manufacture of aircraft flight-standard components where uncontrolled tools are the
norm. Based on a range of the homemade dibbers and in-depth discussions with
operators, a standardised multipurpose dibber has been developed [29]. This can
then be incorporated into training regimes to help to control the variability in the
manual layup process.
As noted earlier this work stream started with an eye to automating the layup
process, but identified a number of ways to improve, control and speed up the
2 But How Can We Make Something Useful Out of Black String?. . . 43
process by designing in the right features, choosing the right prepreg, supporting the
laminators properly and standardising their tools. Taken together a real reduction in
cost and improvement in quality is available, potentially making the development
of cost-effective automation even more difficult by reducing the effective costs of
manual layup. A very recent development has been the robotic layup of a relatively
complex honeycomb-cored sandwich panel from woven carbon fibre prepreg with a
quality indistinguishable from that made by a skilled laminator [30] (Figs. 2.11 and
2.12). To the author’s knowledge, this is the first such demonstration, and it perhaps
seems ironic that it has been enabled more by studying what laminators actually do
than by any advances in robotics, but the reality is that without that knowledge the
task would have been infeasible.
The trajectory for this development is illustrated in Fig. 2.13. The starting point
was the use of VFP software to simulate the layup appropriately. This was followed
by the development of low-cost operator support tools, step-by-step manufacturing
instructions and appropriate hand tools. The detailed understanding coming from
that work enabled the development of the automated solution.
Automating woven reinforcement layup can be thought of as the complex
manipulation of a simple material. The opposite case might then be thought of as
the simple manipulation, e.g. press forming, of complex reinforcements—which
essentially means how can a reinforcement be made that behaves more like a metal
so that it can be embossed or stamped with a clamped edge? A recent development in
44 K. Potter
Fig. 2.13 Development trajectory required prior to robotic automation of complex layup
this area is a novel fibre alignment process by which short fibres, including recycled
fibre, can be aligned to a very high degree to give a truly formable material [31].
Material made this way has just taken the record for the strongest material made to
date from aligned short fibres. These are very early days for this development, and
it may not be possible to commercialise the process, but it does very clearly indicate
that there may still be life in concepts that were first being explored more than 40
years ago.
Towards the end of the 1970s, several groups were looking at the need to meet
requirements for very complex geometries with tight geometrical tolerances, which
really could not easily be manufactured in typical, single-sided autoclave tooling.
The RTM process had essentially been invented and patented in the 1950s [32]
only to be very largely forgotten (at least in the carbon fibre composites arena),
needing to be reinvented and oddly enough repatented later. Various product
development exercises in a number of companies had very clearly identified the
advantages of RTM in terms of process robustness, dimensional reproducibility
and geometrical complexity in general industrial applications using glass fibre
reinforcements. To transition the technology for aerospace meant demonstrating that
carbon fibre/epoxy structures could be reliably manufactured, incorporating features
difficult or impossible to achieve in an autoclave. The necessary technology was
developed and rapidly matured through a series of small demonstrator contracts
(Fig. 2.14), and a contract was won to develop aircraft flight components. These
were of very complex geometry; they were assembled into a mould tool from a
number of individual preforms, most of which were stamped from woven cloth in
simple matched preform tooling, although UD elements were also used (Fig. 2.15).
The preforms generally consisted of two plies of woven cloth, each coated with
a small amount of a thermoplastic powder binder which could be heat softened
prior to forming. Once formed and cooled, the binder held the layers of formed
cloth together to give a robust, stable and handleable preform. This mechanical
preforming of reinforcements to be assembled into a mould tool represents the link
between the two major elements of the author’s work, drape and RTM. The mould
2 But How Can We Make Something Useful Out of Black String?. . . 45
Fig. 2.14 Early aerospace RTM prototypes. Left: a complex engine pylon-de-icing duct with
a major undercut feature, made using a bismaleimide matrix for >200 ı C service. Right: an
electronics cassette with essentially zero radii of curvature corners. Neither part could be made
at that time using conventional autoclave moulding techniques
Fig. 2.15 Left: front and rear views of the main moulding of an early (1989) RTM production
component (the full assembly consisted of this moulding and a second RTM moulding bonded
around a honeycomb core)
tool was then closed and resin was injected under pressure. The actual RTM was
carried out in rigid metallic tooling. Prototypes were very extensively tested and
approved for use on aircraft. A production line was built for the volume manufacture
and parts went into commercial service.
There are probably three points that can be made about this project. Firstly, more
than 25 years later, the author is not aware of any significantly more geometrically
complex parts being made today. Secondly, no flow, cure analysis or other process
simulation tools were used in the development—as none were available. The same
is true of drape models for the stamped parts, where some preforms were made
in double-acting tools—way beyond the state of the modelling art in the 1980s.
It’s not certain that any of today’s computer-based resin flow models would really
cope with the complex geometry and internal resin gating that were used within
the mould. The simple approach was taken that if the tools were tightly sealed, all
the air was evacuated from a completely dry preform, and a resin was used that
had a long open time before significant curing at the injection temperature, then the
details of resin flow would look after themselves, and this proved to be the case.
Lastly, once the line had been debugged, the production proved to be very reliable
with a much lower shop reject rate than the lines in the same factory making much
simpler honeycomb-cored sandwich panels by an approach based on manual layup
of woven prepreg and autoclave moulding. Most of that reduction could be put down
directly to the thickness control made possible by rigid tool RTM.
46 K. Potter
Too many of the papers describing the process simulation of RTM start with
some variant of the statement that ‘Without computer simulation components cannot
reliably be made by RTM’—it can clearly be stated without any hesitation that they
can. Acting rigorously on a knowledge of the fundamental quality drivers can deliver
reliable production in many cases; although process modelling can certainly help,
it is not always critical; understanding the fundamental quality drivers is however
always critical. Recent work taking into account stochastic effects to handle in-
process variability does, however, offer real promise [33]. It can also be noted that
BMW’s i3 body shell mouldings are made in essentially the same way as the parts
described here, from a number of relatively simple pressed preforms assembled in
the tool and infused with resin to deliver a complex moulding. Other approaches
to the manufacture of preforms have been developed over the years including the
use of a variety of stitching techniques, for example, to attach stiffening stringers to
skins. These methods can be aimed at simply providing a handleable preform [33,
34] or at developing an element of 3D reinforcement using stitching or tufting with
a structural thread [35].
Rigid tool RTM has become a very capable process that can be applied to a wide
range of part geometries and scales. Figure 2.16 shows the range of part sizes that
are routinely dealt with. The largest part in Fig. 2.16 is the glass fibre-reinforced
cored sandwich panel spinner of 1.2 m diameter; at the other end of the scale, parts
weighing only a few grams have successfully been made by RTM.
Alongside rigid tool RTM, a number of other variants of resin infusion processes
have been developed based on single-sided tooling and vacuum providing the
pressure to drive the resin impregnation [36]. These processes are used, for example,
in the manufacture of boat hulls and wind turbine blades, Siemens’ development of
a one-shot resin-infused blade requiring no assembly being a notable achievement.
Figure 2.17 shows a resin-infused tufted foam-cored sandwich panel which has been
subjected to a very severe edgewise impact (typical levels for automotive utility pole
impact).
Whilst very severe damage has been done to the panel in the immediate vicinity
of the impactor, associated with a very high level of energy absorption, no significant
damage has been experienced remote from the impactor. By contrast an equivalent
panel without the through thickness reinforcement would be completely destroyed.
The current emphases in the area of resin infusion processes are on the use of
single-sided approaches for ever larger structures and on high-pressure RTM for
high-volume automotive structures. The HPRTM processes are capable of cure
cycles in the range of minutes, using injection pressures up to 100 bars in very
heavy, and very expensive, presses (see Fig. 2.18).
The critical differences between conventional RTM as it has been used in
advanced applications with carbon fibre and HPRTM lie in the injection pressure
and the speed of cure of the resin. In conventional RTM the resin injection and cure
processes are essentially sequential, which makes it relatively straightforward to
model both processes. Equally, the speed of resin flow is relatively low so that there
is little interaction between the reinforcement and the flowing resin, so that for most
purposes, the reinforcement geometry can be assumed to be fixed throughout the
injection. For HPRTM neither assumption is valid, greatly increasing the complexity
of modelling [37]. In spite of the difficulties in modelling, it will be more important
to build a strong modelling capability for high-volume manufacture by HPRTM than
it has been for low-volume conventional RTM, if we are to achieve reliable mass
production. Automotive structural components manufactured by HPRTM are likely
to be the first truly mass production carbon fibre-reinforced structures (>100,000 s
parts/year of a single part type. Composite structures on aircraft such as the Boeing
737 or Airbus A320 families may require in excess of 100,000 carbon fibre parts a
year today but only require hundreds of each specific part).
48 K. Potter
Fig. 2.18 Typical large-scale HPRTM press. Image courtesy of the NCC
2.4.1 Introduction
The author first became interested in defects and variability in composites manu-
facturing as a result of working as part of a product improvement team tasked with
improving quality and right first time yield on a line manufacturing large numbers of
aircraft wing sandwich panels. The morning was spent examining defective panels
in some detail and the afternoon tracking them around the shop floor trying to
determine where the defects arose. Things started to improve when a well-structured
quality and defect database was instituted which allowed the real problems to be
identified and the defect count to be pushed down. One outcome of spending very
many days and weeks linking the activities of the Materials Review Board and
the shop floor was the recognition that very many, if not most, of the defects had
in essence been designed in. To a large extent the need to dig deeper and deeper
into defects and variability and how an understanding of design for manufacture
might be achieved has been a major feature of composites manufacturing research in
recent years. This is really the nexus at which composites manufacture and structural
integrity meet as the performance of the laminates and components are essentially
controlled by details of the micro and mesostructures and the residual stresses that
are generated in manufacture as a result of design and manufacturing decisions.
An approach to defects has been developed based on taxonomies of both defects
and sources of variability [38]. The defects that have attracted most attention
have been those relating to fibre waviness, both in and out of plane. These have
2 But How Can We Make Something Useful Out of Black String?. . . 49
been studied in a wide variety of ways, from characterising and testing incoming
materials and samples cut from production parts to devising manufacturing relevant
and reproducible ways of generating some of the defects seen on production parts to
generating highly controlled defects with very well-defined, if slightly unrealistic,
internal geometries. Both RTM and bag moulded parts have been considered, and
the many processes leading to wrinkle formation have been studied (e.g. [39–42]).
So it can be said with complete clarity that wrinkles are a bad thing and that they
cause a very significant reduction in performance in static and fatigue strength, at
least when coupons are being tested where the defects are of the same sort of scale
as the width of the sample. It might reasonably be objected that that was known
before the research was started, and that the more important question is really how
big the defect has to be, in relation to the structure’s size before it becomes a
structural concern. This is actually not a question that can realistically be asked in an
aerospace context, as the fundamental design assumption would be that the strength
of the whole structure should be written down to that of the weakest link. If that
assumption was really being taken entirely seriously the industry would probably
be even more concerned about wrinkles than it currently is.
One focus of current research is to identify the emergence of defects during the
manufacturing process. As plies are added to the tool, whatever the process used,
there are opportunities for defects to be incorporated. For example, Fig. 2.19 shows
fibre buckling in an AFP layup that would be seen as fibre waviness in that ply after
cure.
The two main defects that should be avoided are fibre waviness and fibre
bridging. Both of these defects can significantly reduce the component’s structural
strength, as well as generating dimensional errors and contributing to dimensional
Fig. 2.19 Incipient fibre waviness defects in an AFP layup. If additional plies were laid down over
these buckled tows, an internal fibre waviness defect would be formed
50 K. Potter
Fig. 2.20 Left: wrinkles in an AFP layup. Right: wrinkles in a draped woven prepreg cloth. Both
clearly visualised, automatically identified and measured using a regular array of laser dots. Image
courtesy of Assembly Guidance
stresses. It has been known for as long as there has been a carbon composites
industry that the shape of a hot-cured part will not be the same as that of the tool. A
90ı corner tool will generally produce an approximately 89ı moulding for a 180 ı C
cure, assuming a balanced and symmetrical laminate. If a part must fit to close
tolerances, the tool geometry may well have to be modified to get the final geometry
correct or at the very least within the tolerance band. We have quite a good handle
on these issues now (e.g. [43, 44]).
There are really three elements to the spring in. A simply calculated geometrical
effect due to differences between in-plane and out of plane thermal expansion; this
accounts for about half the total. An additional through thickness shrinkage due
to post-gelation cure shrinkage (pre-gelation shrinkage has no effect), which is
often of roughly the same magnitude. These two elements are readily calculable,
although measuring cure shrinkage throughout cure isn’t trivial; producing a first
order estimate based on measurements of resin density and thermal expansion
coefficients in the various cure states does, however, give a reasonable result. And
lastly, other stuff. This includes interactions between tooling and laminate if they
have different CTEs (which is true most of the time, even with Invar tooling if
accounted for properly, especially when UD prepreg is used) or direct tension in
the prepreg due to reacting the autoclave pressure in a bridged layup. These are
often taken as second-order effects, but can be dominant in some circumstances
(see Fig. 2.22).
52 K. Potter
Fig. 2.23 Resin shrinkage cracking of resin rich zones in an orthogonal 3D woven composite. The
cracks are highlighted by fluorescent dye penetrant; the blue colour arises from the fact that the
matrix resin is also fluorescent
Latterly effort has been applied to understand the issues around new product
introduction and why, despite the solid growth that carbon composites have seen,
that growth is not nearly as high as was predicted in the early days of the carbon fibre
story [45]. Even with the more automated processes such as AFP, actually getting a
factory up and running and operating as predicted is still proving to be problematical
[46], and these factory level issues have received very limited academic attention.
As the composites industry has developed, a wide range of modelling and
predictive tools have become available that, in principle, support all phases of the
composites design and development cycle. From a business perspective, the most
important of these is perhaps costing as if this is wrong the future of the company
could be at stake. Costing software is available from several vendors and may be
stand-alone or integrated into CAD/CAM or similar software design and analysis
suites. It is well known that the great bulk of total project costs are fixed by decisions
2 But How Can We Make Something Useful Out of Black String?. . . 53
on materials, process and geometry that are made at an early stage of conceptual
design. It is an unfortunate truth that at the point in any project when there is the
potential to make the biggest impact on total project costs, the tools available to
quantify that impact are at their weakest.
Parametric costing relies on having a good, reliable and critically assessed
database of past experience to work within (extrapolating in parametric costing is
very dangerous). For designs that fall within the previous experience, these models
are fit for purpose. If the target is to design for significantly reduced costs, then
more or less by definition a costing approach that relies on using the same designs
as in the past is not going to be a very effective tool. On the other hand bottom-up
costing relies on being able to define the process steps with sufficient granularity
to attach a cost to each step and is amenable to use with novel designs—but this
level of detail is only likely to be developed in detailed design, again limiting its
utility in the early stages of design. This impasse is one of the reasons that many
designs do copy previous practice, and even though the costs cannot be significantly
reduced, they can be predicted with better fidelity—which is often the critical factor
in a commercial environment. It is essential to have costing as a central part of the
design process, but it’s undeniably difficult to do it well.
Many of the CAD/CAM tools essentially focus on automating the work flow in
composites design for large or complex structures, where there may be thousands
of individual plies that must be drawn and multiple design constraints that must be
satisfied in a typical aerospace environment. The use of a wholly manual design
process would be uneconomically time consuming in this case, and the CAD/CAM
tools, such as Catia or Fibersim, are extremely successful in achieving a detailed
design time reduction and will, for example, output and link drape predictions, flat
ply geometry, nesting, ply cutting, AFP control and laser projection aids. These
tools really support the detailed design phase and the transition from that phase into
manufacturing. They are rather less effective in supporting the earliest design stages
where, as noted before, the cost of the project as a whole can be determined. In an
aerospace environment, this may not be an issue as much of the design freedom is
constrained by certification and similar requirements, and cost-efficiently meeting
that constraint set is of primary importance. There is a lack of tools that support
design for manufacture in its broadest sense (rather than design for the transition
into manufacture), largely driven by the rather poorly developed knowledge bases
of design features and their manufacturing impacts.
Beyond the integrated tool suites, there are specific stand-alone tools that can be
used for everything from the design of woven cloths (e.g. TexGen, Wisetex) to flow
analysis in RTM (e.g. LIMS, RTM-Worx, PAMRTM) to stamp forming analyses
(e.g. PAM-Form, AniForm), to cure and distortion models (e.g. Compro, LMAT),
as well as a range of FEA tools for stress analysis both static and dynamic.
The common limitation with many of the tools that are available for modelling
the composites manufacturing processes is that they are quite effective at, for
example, predicting wrinkling in forming ply stacks over an arbitrary geometry by
hot drape forming, but have very little to say about how to avoid that wrinkling by
changes to materials, geometry or process details. It would, of course, be possible
54 K. Potter
to run the software thousands of times to map out the design space, so long as
the software had been validated for that range of materials, geometry and process
parameters. The same would be true for simulations of RTM flow accommodating
race tracking due to preform fit to the tool [47]. There are also still gaps in the
modelling base. There is, for example, a lack of validated models that predict
the development of contact between tool and prepreg or the development and
distribution of tool surface voids or other imperfections.
As noted before the critical point in the development of composite products is
the earliest stages of the design process, and this is the area in which the support
from modelling and simulation is weakest. In this area the need is probably for
‘good enough’ models that solve with no time delay that allow ‘what if’ studies
with little cost or time delay, rather than for very high fidelity models. It may be
that the modelling and simulation approach may not be the best way to achieve
this and that simple look-up tables for best practices for different design types and
moulding processes would be preferable. Equally, developing better approaches to
design reviews for composites would be of great benefit, even a simple approach
based on a design checklist to capture and eliminate common errors could be of
great benefit, especially with novice designers.
In conclusion, the analytical, numerical and predictive tools associated with
composites design and manufacture have still in many instances not quite caught
up with the ability to manufacture the most complex geometries without defects,
but they are very much closer today than they were even a decade ago, let alone
50 years ago when the carbon fibre composites story started. The challenge now is
going to be how best to disseminate both the tools themselves and the understanding
derived from the intelligent application of those tools throughout the industry.
The last decade has seen a very significant increase in the use of carbon fibre
composites. Alongside that there has been a growing awareness from the UK and
other governments that manufacturing really is important, and money has started
to flow in support of that. A major step in this for the UK was the publication
in November 2009 of the National Composites Strategy. This led directly to
the funding of the National Composites Centre, led by industry and operated
autonomously within the University of Bristol’s Engineering Faculty, with the first
build phase being launched in 2010 and operational in 2011 and the second build
phase following on in 2013 and operational in 2015. Similar facilities have been put
in place in many other countries including, Japan, the USA, France, Germany and
Brazil. A real challenge for the future will be how to develop the skilled people
we will need in increasing numbers. The first generation of composites people,
the men and women that invented carbon and aramid fibre and made them into
industrial reality, are well past retirement age—even if some of them are still active.
The second generation are getting close to retirement now, and there seems to be
something of a demographic gap opening up, especially in the area of composites
manufacturing, which is presenting the community with some challenges.
Lastly, it must be very clearly recorded that the author stopped working in the lab
a decade or more ago (and in recent years has spent far too much time with architects
and builders delivering new university labs and the NCC). There is absolutely no
2 But How Can We Make Something Useful Out of Black String?. . . 55
way that progress could have been made without the support of a large number of
academic colleagues, postdoctoral researchers and an even larger number of PhD
students who were prepared to take the risk of not following a well-worn path, but
rather going places without much of a literature base. The composites community
is moving towards a real synthesis of design, manufacture and structural integrity
in carbon composites, and it is the current and future generation of researchers that
will have to be delivering it.
References
1. L.N. Phillips, Forming processes for carbon fibre and resin. Composites 1(2), 101–103 (1969)
2. D. Wildman, Forming processes for glass fibre and resin—contact moulding. Composites 1(1),
35–40 (1969)
3. J. Mountfield, Forming processes for glass fibre and resin—other methods. Composites 1(1),
41–49 (1969)
4. P. McMullen, Fibre/resin composites for aircraft primary structures: A short history,
1936–1984. Composites 15(3), 222–230 (1984)
5. C. Soutis, Recent advances in building with composites. Plast. Rubber. Compos. Macromol.
Eng. 38(9/10), 359–366 (2009)
6. D. Purslow, R. Childs, Autoclave moulding of carbon fibre-reinforced epoxies. Composites
17(2), 127–136 (1986)
7. H. Edwards, N.J. Parratt, K.D. Potter, Synthesis and application of aligned discontinuous
composites. In International Conference on Composite Materials, Number 2, Toronto, 1978
8. K.D. Potter, The influence of accurate stretch data for reinforcements on the production
of complex structural mouldings: Part 2. Deformation of random mats. Composites 10(3),
168–173 (1979)
9. K.D. Potter, The influence of accurate stretch data for reinforcements on the production of
complex structural mouldings: Part 1. Deformation of aligned sheets and fabrics. Composites
10(3), 161–167 (1979)
10. K. Potter, C. Ward, Draping processes for composites manufacture, in Advances in Composites
Manufacturing and Process Design, ed. by P. Boisse (Jai-Elsevier Science, Kidlington, 2015),
pp. 93–109, 17 p. ISBN 978-1-78242-307-2
11. W.B. Goldsworthy, Geodesic path length compensator for composite-tape placement method.
Patent US 3810,805, 14 May 1974
12. P.W. Taylor, The development and use of a tape laying machine. In Symposium on Fabrication
Techniques for Advanced Reinforced Plastics (IPC Science and Technology Press, Salford,
1980)
13. R.O. Buckingham, G.C. Newell, Automating the manufacture of composite broadgoods.
Compos. Part A 27(3), 191–200 (1996)
14. C. Mack, H.M. Taylor, The fitting of woven cloth to surfaces. J. Textile Inst. Trans. 47(8),
477–488 (1956)
15. R.E. Robertson, T.-J. Chu, R.J. Gerard, J.-H. Kim, M. Park, H.-G. Kim, R.C. Peterson, Three-
dimensional fiber reinforcement shapes obtainable from flat, bidirectional fabrics without
wrinkling or cutting. Part 1. A single four-sided pyramid. Compos. A: Appl. Sci. Manuf. 31(7),
703–715 (2000)
16. L. Dong, C. Lekakou, M.G. Bader, Solid-mechanics finite element simulations of the draping
of fabrics: A sensitivity analysis. Compos. A: Appl. Sci. Manuf. 31(7), 639–652 (2000)
17. A.C. Long, B.J. Souter, F. Robitaille, A fabric mechanics approach to draping of woven
and non-crimp reinforcements, in Proceedings-American Society for Composites (2000),
pp. 76–83
56 K. Potter
18. P. Boisse et al., Composites forming. In Advances in Material Forming (Springer, Paris, 2007),
pp. 61–79
19. K.D. Potter, S. Hancock, Development and application of the virtual fabric placement concept
for the production of complex shaped components from woven reinforcement. In SAMPE
Europe International Conference, Paris, Apr 2007
20. J. Chen et al., Impact of layup rate on the quality of fiber steering/cut-restart in automated fiber
placement processes. Sci. Eng. Compos. Mater. 22(2), 165–173 (2015)
21. R. Lichtinger et al., Simulation and experimental validation of gaps and bridging in the
automated fiber placement process. Sci. Eng. Compos. Mater. 22(2), 131–148 (2015)
22. D.H.J.A. Lukaszewicz, K.D. Potter, J. Earles, A concept for the in situ consolidation of
thermoset matrix prepreg during automated lay-up. Compos. Part B 45, 538–543 (2013)
23. B.C. Kim, P.M. Weaver, K. Potter, Manufacturing characteristics of the continuous tow
shearing method for manufacturing of variable angle tow composites. Compos. A: Appl. Sci.
Manuf. 61, 141–151, 11 pp (2014)
24. L.D. Bloom, M. Elkington, C. Ward, A.P. Chatzimichali, K.D. Potter, On prepreg properties
and manufacturability. In 19th International Conference on Composite Materials, Montreal,
24–28 July 2013, pp. 4397–4409
25. M. Elkington, L.D. Bloom, C. Ward, A.P. Chatzimichali, K.D. Potter, Understanding the lam-
ination process, in Proceedings of 19th International Conference on ‘Composite Materials’,
Montreal, Canada, 2013, pp. 4385–4396
26. S. Blake, Replacing composite hand layup templates with a simple projection system, in 34th
International SAMPE Symposium and Exhibition, 1989, pp. 362–372
27. M.C. Such, C. Ward, W. Hutabarat, A. Tiwari, Intelligent composite lay-up by the application
of discrete technologies. Procedia CIRP 25, 122–131 (2014)
28. S. Kularatna, C. Ward, Improving training in advanced composites hand layup by the
application of virtual reality and gamification techniques, in SAE 2016 Augmented and Virtual
Reality (AR/VR) Technologies Symposium, Philadelphia, Pennsylvania, USA, 14–16 November
2016
29. H.V. Jones, A.P. Chatzimichali, R.A. Middleton, K.D. Potter, C. Ward, Exploring the discrete
tools used by laminators in composites manufacturing: Application of novel concept. Adv.
Manuf.: Polym. Compos. Sci. 1(4), 185–198, 14 pp
30. M. Elkington, C. Ward, K. Potter, Automated layup of sheet prepregs on complex moulds. In
SAMPE Long Beach Conference, May 2016
31. H. Yu, K.D. Potter, M.R. Wisnom, A novel manufacturing method for aligned discontinuous
fibre composites (high performance-discontinuous fibre method). Compos. A: Appl. Sci.
Manuf. 65, 175–185, 11 pp
32. K.D. Potter, The early history of the resin transfer moulding process for aerospace applications.
Compos. Part A 30, 619–621 (1999)
33. U. Beier et al., Mechanical performance of carbon fibre-reinforced composites based on
stitched preforms. Compos. A: Appl. Sci. Manuf. 38(7), 1655–1663 (2007)
34. U. Beier et al., Evaluation of preforms stitched with a low melting-temperature thermoplastic
yarn in carbon fibre-reinforced composites. Compos. A: Appl. Sci. Manuf. 39(5), 705–711
(2008)
35. D.D.R. Cartié et al., 3D reinforcement of stiffener-to-skin T-joints by Z-pinning and tufting.
Eng. Fract. Mech. 73(16), 2532–2540 (2006)
36. J. Summerscales, T.J. Searle, Low-pressure (vacuum infusion) techniques for moulding large
composite structures. Proc. Inst. Mech. Eng. L J. Mater. Des. Appl. 219(1), 45–58 (2005)
37. M. Deléglise et al., Modeling of high speed RTM injection with highly reactive resin with
on-line mixing. Compos. A: Appl. Sci. Manuf. 42(10), 1390–1397 (2011)
38. K. Potter, B. Khan, M. Wisnom, T. Bell, J. Stevens, Variability, fibre waviness and misalign-
ment in the determination of the properties of composite materials and structures. Compos. A:
Appl. Sci. Manuf. 39(9), 1343–1354 (2008)
39. J.S. Lightfoot, M.R. Wisnom, K. Potter, Defects in woven preforms: Formation mechanisms
and the effects of laminate design and layup protocol. Compos. A: Appl. Sci. Manuf. 51,
99–107, 9 pp (2013)
2 But How Can We Make Something Useful Out of Black String?. . . 57
40. J. Wang, K. Potter, J.A. Etches, Experimental investigation and characterisation techniques of
compressive fatigue failure of composites with fibre waviness at ply drops. Compos. Struct.
100, 398–403, 6 pp (2013)
41. S.L. Lemanski, J. Wang, M.P.F. Sutcliffe, K.D. Potter, M.R. Wisnom, Modelling failure of
composite specimens with defects under compression loading. Compos. A: Appl. Sci. Manuf.
48(1), 26–36, 11 pp (2013)
42. L.D. Bloom, J. Wang, K.D. Potter, Damage progression and defect sensitivity: An experimental
study of representative wrinkles in tension. Compos. Part B 45(1), 449–458, 10 pp (2013)
43. D.W. Radford, T.S. Rennick, Separating sources of manufacturing distortion in laminated
composites. J. Reinf. Plast. Compos. 19(8), 621–641 (2000)
44. N. Ersoy, T. Garstka, K. Potter, M.R. Wisnom, D. Porter, G. Stringer, Modelling of the spring-
in phenomenon in curved parts made of a thermosetting composite. Compos. A: Appl. Sci.
Manuf. 41(3), 410–418, 9 pp (2010)
45. A.P. Chatzimichali, K.D. Potter, From composite material technologies to composite products:
a cross-sectorial reflection on technology transitions and production capabilities. Trans. Mater.
Res. 2(2), 026001 (2015)
46. D.J.J. Winter, C.D.C. Jones, C. Ward, P.M. Gibbons, C.A. McMahon, K.D. Potter, The
application of a lean philosophy during manufacturing of advanced airframe structures in a
New Product Introduction (NPI) environment. In Advances in Sustainable and Competitive
Manufacturing Systems: 23rd International Conference on Flexible Automation and Intelligent
Manufacturing (2013), pp. 1503–1513, 11 pp
47. T.S. Mesogitis, A.A. Skordos, A.C. Long, Uncertainty in the manufacturing of fibrous
thermosetting composites: A review. Compos. A: Appl. Sci. Manuf. 57, 67–75 (2014)
Chapter 3
Boron Fiber to Carbon Fiber
Carl T. Herakovich
3.1 Introduction
The author was introduced to fiber composites materials in the summer of 1969.
He was a NASA-ASEE Faculty Fellow working at NASA’s Langley Research
Center in Hampton, VA, USA. He had completed a PhD in mechanics with a
dissertation on plasticity and finite element methods at the Illinois Institute of
Technology in 1968; he knew nothing about composite materials when he started
his work at NASA. He was a young assistant professor in the Engineering Science
and Mechanics Department at Virginia Tech (at the time, the name of the department
was Engineering Mechanics). He had the good fortune to teach a course in finite
element methods to several NASA engineers who were working on graduate degrees
at Virginia Tech. They told him about the NASA-ASEE Faculty Fellow Program.
Seeing this as an opportunity to develop a research activity, he applied and was
accepted into the Program for the summer of 1971. He conducted research on
mechanics of composites at NASA Langley during the summers of 1971, 1972,
and 1975.
His first assignment at NASA was to investigate the reason for the exceptionally
low failure strength of boron–epoxy-reinforced aluminum tensile coupons. These
hybrid laminates were failing at approximately 50 % of the predicted strength—
based upon the strain to failure of the boron fibers. Research into the unusual results
eventually led to a free edge problem as will be discussed in the following. However,
before an explanation of the early failures of the boron fiber composites could be
determined, carbon (referred to as graphite at the time) fiber composites came into
vogue and were studied intensively by the author, his colleagues, and their students
over the final quarter of the twentieth century. The research on the boron–epoxy-
reinforced aluminum was discontinued when the emphasis was shifted to carbon
fiber composites.
The strong interest in carbon fiber composites played a key role in the establishment
of the NASA-Virginia Tech Composites Program. The Program started in 1974,
3 years after this author spent his first summer at NASA Langley Research Center
in Hampton, Virginia. Initially, the Program consisted of one graduate student and
one faculty member working with one NASA engineer. The Program grew, year
by year, eventually involving 13 faculty members, 91 graduate students, and 25
NASA engineers and scientists over the 22-year life of the program. Students in the
program earned MS and PhD degrees in the Engineering Science and Mechanics
Department or the Aerospace and Ocean Engineering Department at Virginia Tech.
The graduate students typically completed a research residency at NASA Langley,
6 months for MS candidates and 12 months for PhD candidates. Fifteen graduates
of the Program accepted full time positions at NASA Langley or other government
laboratories. Other graduates accepted faculty positions or industrial work, mainly
in the aerospace industry. An untold number of students, undergraduate and
graduate, were taught the fundamentals of mechanics of anisotropic, layered, fibrous
composite materials and structures through the courses established as a result of the
research activity and the program with NASA.
The research conducted in this program covered a wide variety of topics, mostly
concerned with carbon fiber composite materials; research on boron and borsic fiber
composites and carbon–carbon composites received limited attention. Experimental
methods (tension, compression, shear and Moiré interferometry), finite element
methods, analytical methods, structural optimization, composite fabrication, ther-
mal effects, damage and crack growth, stability, space environmental effects, and
nonlinear response were investigated. Static, dynamic and pressure loading of
structural composites (beams, sandwich panels, tubes, and stiffened panels) were
considered.
The NASA-Virginia Tech Composites Program was the major contributor to the
outstanding reputation that Virginia Tech gained as a premier university for the study
of mechanics of composites.
The author’s research activity with carbon fiber composite materials actually started
with research in boron fiber composites. And the differences between these two
types of advanced fibers demonstrate why carbon has become the fiber of choice for
most applications.
3 Boron Fiber to Carbon Fiber 61
Boron-Epoxy
Aluminum
taking into consideration that the boron–epoxy is in residual compression (in the
fiber direction) after co-curing and cooling to room temperature. No delamination
was evident in the failed hybrid laminates; the boron–epoxy tended fail along a line
(more or less) perpendicular to loading direction and away from the fiber glass end
tabs.
The surprising experimental results led to a multifaceted investigation in an
attempt to explain the early failures. Factors that were considered included edge
effects, residual thermal stresses from the curing process, nonuniform load introduc-
tion, damage induced when cutting coupons from plates, specimen width, co-curing
and room temperature adhesive bonding of the boron–epoxy, reversing the stacking
sequence, and aluminum tubes with boron–epoxy bonded to the tube exterior.
Acoustic emissions obtained from the specimens exhibiting early failure indicated
that extensive fiber breakage initiated at the knee of a bilinear stress–strain curve
and continued until failure. Finite element investigations of the three-dimensional
state of stress indicated stress concentrations, including high interlaminar stresses,
near the free edge, but no explanation for fiber failure.
A comparison of the experimental results from a number of different specimen
configurations is shown in Figs. 3.2 and 3.3. Figure 3.2 shows a comparison of the
experimental and predicted responses of the co-cured, BE/Al coupon cut from a
larger plate. The experimental response is bilinear with a sharp knee at a low level
of strain. Acoustic emission indicated that the knee corresponded to the initiation of
fiber breakage. The specimen failed at a stress and strain well below the expected
value.
Figure 3.3 shows examples of results from six different types of hybrid specimens
including a coupon cut from a plate, a tube, and a “modified dogbone.” The modified
dogbone had a stepped-down region of load introduction with the boron–epoxy
bonded to the aluminum in a co-curing operation; there was no machining after
co-curing and bonding to the aluminum.
The results in this figure demonstrate that there was considerable variation in
strain to failure depending upon the types of specimen tested, i.e., co-cured, room
temperature bonded, elevated temperature bonded, or tube. The ultimate strain of the
different specimens ranged from approximately 0.4 % to 0.8 %. The failure strain
of the boron fiber is approximately 0.6 %. The specimens with failure strains in
excess of 0.6 % can be explained by the fact that the boron–epoxy is in residual
compression (in the fiber direction) after co-curing and bonding to the aluminum,
resulting in extended tensile failure strain for the hybrid material.
It may be tempting to assign the low failure strains of the hybrid coupons to the
damage introduced when they were cut from a larger plate; however, all boron–
epoxy coupons also were cut from a plate and they did not exhibit the low failure
strains. Unfortunately, no definitive explanation for the low failure strains (0.4 %)
of the hybrid coupons was determined before carbon fiber composites came on
the scene. Once the interest in, and advantages of, carbon fiber composites was
established, the research on the boron–epoxy-reinforced metals was discontinued.
The low failure strains remain an unanswered question.
64 C.T. Herakovich
Simply looking at photos (Figs. 3.4 and 3.5) of boron and carbon fibers shows how
different they are. The boron fiber is a large diameter, stiff, brittle, single filament;
it can withstand little bending prior to failure. Carbon fiber is generally provided
as a “tow” of many (2000–14,000) small diameter, flexible filaments. The carbon
fiber tow can sustain extensive bending (it can be tied in a knot) without breaking.
The boron fiber can support compression loading, but is susceptible to buckling
without lateral support; the carbon fiber cannot support compression unless bonded
to a material that provides lateral support.
Tables 3.1 and 3.2 give comparison of some engineering and physical properties
of boron fiber and three different types of carbon fibers. The table also gives
properties of an aerospace grade aluminum (7075-T6) and an epoxy matrix material
Tensile Tensile
Density Tensile strength ultimate Specific Specific
g/cm3 modulus E u MPa strain " stiffness strength Axial CTE ˛
Material (lb/in.3 ) GPa (Msi) (ksi) (%) (E/)/(E/)Al ( u /)/( u /)Al µ/ı C (µ/ı F)
Aluminum 2.7 (0.097) 69 (10) 483 (70) >1.0 1.0 1.0 23.4 (13.0)
AS4 1.80 (0.065) 235 (34) 3599 (522) >1.0 5.1 11.1 0.8 (0.44)
T300 1.76 (0.064) 231 (33) 3654 (530) >1.0 5.1 11.5 0.5 (0.3)
IM8 1.8 (0.065) 310 (45) 5171 (750) >1.0 6.7 16.1 –
Boron 2.6 (0.094) 385 (55.8) 3799 (551) 0.6 5.8 8.3 8.3 (4.6)
Epoxy 1.38 (0.050) 4.6 (0.67) 58.6 (8.5) >1.0 0.08 0.4 63 (35)
65
66 C.T. Herakovich
Fibrous composite materials are used most often in a thin, laminated form. This
raises issues for developing specimens with regions of uniform stress and strain
without cutting fibers; cutting fibers introduces stress concentrations that leads to
early failure. And, while it might be possible to use cylindrical test specimens,
they are more expensive to fabricate, and because of fabrication issues, it cannot
be stated with certainty that the properties of the cylinder would be the same as
those of the flat laminate. As a result, testing for the mechanical properties of highly
anisotropic carbon–epoxy (and other polymer matrix composites) required new
methods for determining tension, compression, and shear properties. Some of the
issues encountered and the approaches to overcome these difficulties are described
in the following.
3.5.1 Tension
gages mounted. The specimen must of course be symmetric about its midplane to
eliminate bending. This specimen usually provides good experimental results for
the tensile response of a variety of carbon fiber composite laminates.
3.5.2 Compression
Carbon fiber composites are very susceptible to buckling (fiber kinking) when
subjected to compression; thus, the specimen length must be very short and the axial
alignment must be tightly controlled. A variety of specimens and loading fixtures
have been offered as viable methods for determining the compression response.
One example is shown in Fig. 3.7 [1]. This fixture has been used for cyclic tension–
compression loading. Alignment of the specimen in the fixture and the alignment of
the fixture along the line of the axial load are critically important.
68 C.T. Herakovich
3.5.3 Shear
Testing for the shear stress–strain response of carbon fiber composites is another
example of the difficulty when testing flat laminates. For a meaningful shear test,
there must be a region of known, uniform shear in the specimen. It is very difficult to
develop such a stress state in unidirectional fibrous composites without introducing
stress concentrations at points of load introduction that lead to early failure. Again,
a wide variety of specimens has been investigated. The preferred specimen is the
Iosipescu shear test originally used for metals. Figure 3.8 shows the specimen and
a loading fixture designed by Adams and Walrath [2]. The actual specimen is the
double v-notched rectangle in the center of the fixture.
One of the more interesting aspects of laminated fibrous composites is the inter-
laminar stresses present near free edges of the laminated materials. The mismatch
of material properties between layers combined with the stress-free edge results in
a three-dimensional state of stress with the potential for some interlaminar stress
components to have a singular nature near the edge. These stresses can lead to
delamination. One of the more dramatic effects is shown in Fig. 3.9 where failed
tensile coupon specimens of angle-ply (˙30ı ) carbon–epoxy laminates with dif-
ferent stacking sequences are compared. The two stacking sequences are [(˙30)2]s
3 Boron Fiber to Carbon Fiber 69
(alternating) and [(302/302 )]s (clustered). As is apparent in the figure, the clustered
laminate debonded (due to interlaminar shear stresses), but the alternating laminate
did not. For the clustered laminate, there is no fiber breakage; the laminate debonds
and fails along a line between fibers. In contrast, the alternating laminate exhibits no
debonding. The laminate fails along a 30ı line breaking fibers in half of the layers.
The alternating laminate exhibited an ultimate stress that was 48 % higher than that
of the clustered laminate.
The use of carbon fiber composite materials has experienced tremendous growth
during the approximately 40 years that I have worked with them. The extensive use
of these materials in modern commercial aircraft may be the best example of how
well they have been received. Their use will continue to expand as more designers
become familiar with their advantages and as fabrication methods improve to make
the materials more cost competitive.
Acknowledgments Much of the work described in this chapter can be found in the author’s book
Mechanics of Fibrous Composites, John Wiley & Sons, 1998 [3], and the references cited therein
Elsevier [4]. Additional references on carbon composites can be found in the authors’ publication
Mechanics of composites: A historical review, Mechanics Research Communications 41 (2012) 1–
20, Elsevier. The author is grateful to the National Aeronautics and Space Administration (NASA)
for financial support, the faculty and students in the NASA-Virginia Tech Composites Program and
to all the students and colleagues who he has the good fortune to be associated with over the years.
70 C.T. Herakovich
References
1. M.W.Y. Lin, M.-J. Pindera, Elastoplastic response of unidirectional graphite aluminum under
combined tension-compression cyclic loading. Center for Composite Materials and Structures
Report CCMS-88-01 (Virginia Tech, Blacksburg, VA, 1988)
2. D.F. Adams, D.E. Walrath, Iosipescu shear properties of SMC composite materials, ed. by I.M.
Daniel, in Composite Materials: Testing and Design, ASTM STP 787 (American Society for
Testing and Materials, Philadelphia, 1982), pp. 19–33
3. C.T. Herakovich, Mechanics of Fibrous Composites (Wiley, New York, 1958)
4. C.T. Herakovich, Mechanics of composites: a historical review. Mech. Res. Commun. 41, 1–20
(2012)
Chapter 4
Serendipity in Carbon Fibres: Interfaces
and Interphases in Composites
Frank R. Jones
4.1 Introduction
In the mid-1960s, William Watt and colleagues at the Royal Aircraft Establishment
(RAE) in Farnborough discovered the technique for converting polyacrylonitrile
(PAN) into high-strength carbon fibres (CFs), which heralded the field of composites
[1, 2]. Prior to that, these materials were referred to as fibre-reinforced plastics
(FRPs) or glass fibre-reinforced plastics (GFRPs) although the concept of using
fibre reinforcement for structural materials had triggered the work at RAE. For
example, boron fibres were being tested for reinforcing metals. These fibres were
prepared by chemical vapour deposition (CVD) of boron onto a substrate fibre,
namely, tungsten wire. William Watt was renowned for his dogged intention to
succeed in making high-strength CF. In a way it was serendipitous that he succeeded
because ‘black Orlon’ was already known [3]. It was the black product from heating
polyacrylonitrile (PAN). Shindo [4, 5] in Japan had also reported that CFs with
relatively ‘high strength’ could be obtained using thermal treatment of a PAN textile
fibre.
The UK company Courtaulds had solved the many problems associated with
manufacturing a polymer suitable for high-performance acrylic textile fibres.
Acrylonitrile (AN) is difficult to polymerise in a controlled manner because the
polymer is insoluble in the monomer, which can lead to a runaway exotherm. Highly
polar solvents such as dimethylformamide can be used, but they strongly interact
with PAN and are difficult to remove from the spun fibre. Schmidt et al. [6] had
discovered that AN could be polymerised in a 50 % aqueous solution of sodium
thiocyanate (NaSCN), which had the required properties for spinning directly into
fibres, using 10 % aqueous NaSCN as the coagulant with a water wash. Furthermore,
they used a terpolymer of AN, 1 mol% itaconic acid as the dye site and 6 mol%
methylacrylate as an internal plasticiser which facilitated the drawing of the fibre
into ‘high’-modulus textile fibre. The latter reduced the glass transition temperature
(Tg) from 100 to 90 ı C so that the copolymer fibre could be drawn in steam (at
100 ı C) to the draw ratio required to ensure the carbonisation achieves a strong
CF. The homopolymer PAN fibre could only be drawn at higher temperatures (in
a glycerol bath?) [7]. Furthermore the carboxylic acid (-COOH) groups in the
itaconic acid comonomer segment initiate the polymerisation of the nitrile side
groups into a ladder polymer, which is essential to the oxidative stabilisation of
the fibre. Buchmeiser et al. [8] have recently reviewed the use of PAN and other
precursors for preparation of carbon fibres.
As a result of the above discussion, it was natural for Watt [1, 7] to use Courtelle as
the precursor for carbon fibres since it was the UK-available acrylic textile fibre. He
found that oxidative stabilisation of the fibre while it was restrained was essential
to high-strength and high-carbon yields after carbonisation. This was the critical
discovery, which led to what we now know as high-strength CF. Moreton et al.
[1, 2, 7, 9, 10] compared the performance of the RAE and Japanese CFs obtained
from PAN precursors. Figure 4.1 gives the typical strengths and moduli of these
fibres.
The critical stage is oxidation; however, Watt and Johnson [10] showed elegantly
by comparing copolymers with and without the itaconic acid comonomer that the
thermal cyclisation (Fig. 4.2) preceded oxidation. Thus, the cyclised PAN has a
structure consisting of sequences of puckered rings which, after oxidation, become
planar and aromatic as shown in Fig. 4.3.
Figure 4.4 shows the oxidised zone of PAN fibres propagating from the outer
surface during the oxidation stage at 220 ı C, which confirms that the thermal
cyclisation precedes the oxidation.p The kinetics of this process is provided in
Fig. 4.5 where a diffusional law of t provides a linear relationship.
These sequences of planar aromatic rings in oxidised PAN represent the nuclei
for graphitisation (Fig. 4.3). Therefore, it is critical to maintain that the molecular
orientation induced in the precursor in the drawing stage is during oxidative
stabilisation. This determines the size and number of nuclei, which initiate the
growth of the graphite crystals in the carbonisation and graphitisation stages.
The PAN copolymer is atactic, which means the molecular structure contains
random short sequences of syndiotactic and isotactic nitrile (CN) side groups on the
chain. An extended syndiotactic polymer has the CN groups on alternative sides of
the molecular chain, whereas on an isotactic chain, they are all on the same side.
This explains why the degree of draw of the PAN fibre determines the modulus of
CF (Fig. 4.6) because the first stage involves self-polymerisation of the CN groups
into sequences of the ladder polymer. Polymer molecules are normally arranged in
4 Serendipity in Carbon Fibres: Interfaces and Interphases in Composites 73
Fig. 4.1 Comparison of tensile strengths ( B /GPa) and tensile moduli (E/GPa) of PAN-based
carbon fibres as a function of the heat treatment temperature reported by Moreton et al. (present
work) [2, 7, 9] and Shindo [5]
CH CH CH
C C C
O OH N N C C C
O O N NH
Fig. 4.2 Shows the carboxylic acid initiation of anionic polymerisation of nitrile side groups to
form a ladder polymer [7]
random coils so only the extended isotactic sequences, in the drawn polymer, are in
a reaction volume for thermal polymerisation of the pendant CN groups.
Since the PAN precursor is atactic with short sequences of isotactic pendant
CN groups, the cyclised sequences of the ladder polymer will be limited to 5–15
groups [11]. Thus, the nuclei for the subsequent carbonisation are of similar size
with relatively low numbers of aromatic rings. In addition, the molecular chain is
free to rotate about the skeletal bonds linking the cyclised sequences. The reaction
between syndiotactic and isotactic nitrile groups in the stabilisation stage will tend
to cause the molecular chain to permanently fold so that the molecular chains in the
74 F.R. Jones
N
H2 H2 H2 H2 C
H C H C H C H C
C C C C C
H
C C C C
N N N N
H
N N N N N
H2 C C C C C
C O2 N N N
C C C C C
H C H C H C C 200˚C
H2 H2 H2 H2
400 - 600 ˚C – H2
N N N
N N N
600 - 1300 ˚C
– N2
N N N
N N N
Fig. 4.3 Chemistry of conversion of PAN into carbon fibres [11, 12]
Zone thickness, mm 8
4
0 4 8
(Hours)1/ 2
p
Fig. 4.5 Oxidised zone thickness as a function of time for three denier Courtelle PAN fibres in
air at 220 ı C [1]
10
Draw ratio
so-called oxidised PAN will take on a curvature rather than the planar ribbon-like
form. To avoid the relaxation of the extended molecular arrangement, the oxidation
stage is carried out with the fibre under restraint. This was the essential difference
between the work at RAE and in Japan and the basis of the UK patent. The final
carbonised fibre is turbostratic in nature because of the numbers of nuclei involved
and their orientation in the graphitisation process [13].
76 F.R. Jones
The formation of carbon fibres is described in Fig. 4.7. Special acrylic fibre (SAF)
differs from textile grade in two respects: the degree of draw (spin–stretch factor)
and the careful filtration of the polymer dope and spinning in a clean room. In
this way, the number and size of strength-reducing flaws and defects in the CF
are limited. This was demonstrated by Reynolds and Sharp [15] who showed
that high strength could be achieved by carbonisation at 1100 ı C as opposed to
1500 ı C. CFs produced at 1500 ı C were originally known as high strength (HS),
whereas those produced at 1200 ı C are referred to as Type A. High-modulus (HM)
CFs are produced by graphitisation at higher temperatures approaching 2500 ı C.
Intermediate-modulus (IM) CFs are formed at a similar temperature to Type A fibres
from a PAN yarn which has received an additional drawing to a higher spin–stretch
factor. Watt [1] had showed that the modulus of the CF was directly proportional to
the spin–stretch factor of the yarn (Fig. 4.6).
Fig. 4.7 Schematic of the stages involved in the manufacture of carbon fibres from polyacryloni-
trile (PAN) precursors. HM high modulus (7 m diameter), HS high strength (7 m), A Type A
(7 m), IM intermediate modulus (5 m) [14]
4 Serendipity in Carbon Fibres: Interfaces and Interphases in Composites 77
The carbonisation stage is carried out in an inert atmosphere over the temper-
ature range of 400–600 ı C. In this stage, hydrogen (and water) is evolved in the
condensation of the aromatic ladder sequences.
The graphitisation stage is also carried out in an inert atmosphere at temperatures
between 600 and 1300 ı C. Further condensation occurs with the loss of nitrogen
(N2 ), hydrogen cyanide (HCN) and oxygen-containing molecules. The fibres
obtained after heat treatment to 1200 or 1300 ı C are usually referred to as Type
A and have high tensile strength.
Further heat treatment to temperatures as high as 2500 ı C increases the dimen-
sions of the graphitic domains and crystals because the carbon atoms become
mobile at around 1100 ı C. The increase in the degree of graphitisation leads
to a higher modulus. Thus, high-modulus (HM) fibres are usually subjected to
temperatures near 2500 ı C. Unfortunately, the fibre strength tends to decrease when
the graphitisation temperature exceeds 1200 ı C. Reynolds and Sharp [15] showed
that the critical flaw size for failure of CF is always larger than the measured
crystallite dimensions. Thus, the strength-reducing flaws arise from the catalysed
graphitisation around inclusions, which lead to sufficient continuity within crystals
with highly misorientated angles (to the fibre axis) to form a flaw of critical size.
Typical inclusions were poorly dissolved polymer precursor nodules and other
foreign particulate matter. Hence, the preferred use of special acrylic fibre (SAF)
for CF production. SAF is spun in clean room conditions from carefully filtered
PAN dope.
Fibre surface treatment is an essential stage in the production of CF. The
graphitic structure of CF differs for HM or HS versions. In the former, the
orientation of the graphene sheets to the fibre axis is more perfect than in the HS
fibre.
The theoretical modulus of graphite in the plane is 1000 GPa so that the
perfection in graphite orientation is 20 (HS) to 40 % (HM) so that the edges of
the graphene planes exit the surface to differing degrees. The surface is complicated
by the apparent core–skin structure, whereby the degree of graphite perfection is
higher in the skin. Thus, the surface of the fibres oxidise to differing degrees on
exiting the graphitisation oven. Control of the atmosphere in the process is therefore
important. The fibre surface is insufficiently polar for wetting by the matrix resins.
Therefore the surface of the fibre needs to be oxidised to ensure that a strong
interface between fibre and matrix is formed in the cured composite. Several surface
treatment techniques have been described [2], but generally electrolytic oxidation is
employed [16–19].
It is easier to treat HS fibres because the oxidation occurs at the edges of the
‘graphene’ sheets where they reach the surface as described in Fig. 4.8. Once the
fibres are surface treated, they are washed and coated with a protective size. A size
is a polymer, which is usually deposited from an aqueous emulsion, and is used
to prevent the introduction of strength-reducing surface flaws during handling. It
should also enable the impregnation of the fibres with resin during processing of
composites. Since many matrices for composites are epoxy resins, sizing polymers
have a similar epoxy-like chemical structure.
78 F.R. Jones
H2 O ! HO C HC C e
2HO ! H2 O C O (4.1)
2O ! O2
William Watt studied the surface chemistry of carbon fibres using thermal desorp-
tion coupled to evolved gas analysis. In this way, the oxygenated functional groups
could be identified from desorption of CO2 , CO and H2 O. Watt found that the
fibres exiting the carbonisation oven exhibited the presence of oxygen and was
able to quantify the concentrations. The other technique used at RAE was dye
adsorption [22]. On retirement, Watt became a research fellow at the University
of Surrey and transferred his equipment. There were inspirational discussions
with the author, which led to a study of the surfaces of CF by XPS (X-ray
photoelectron spectroscopy). In Surrey University at the time was the world-leading
surface scientist, J E Castle, so the pros and cons of XPS for CF surface analysis
were regularly discussed [23]. It was necessary to develop a labelling technique
for quantifying the surface functional groups. The most important functionality
responsible for chemical coupling to epoxy resin was the carboxyl croup, –COOH.
Denison et al. [24, 25] showed that Ba2C could be used to quantify the surface
acidity of carbon fibres. Ba2C is bidentate so a statistical analysis was needed for
the data to be corrected for the presence of isolated –COOH groups.
With Type A CF, Denison et al. [24] found that the surface acidity was such
that the number of –COOH groups present exceeded the number that could be
accommodated on a cylindrical graphitic surface of the fibre of diameter 8 m.
Therefore to accommodate the excess number of functional –COOH groups, the
surface must have a higher surface area, which can be understood by the creation of
micropores where the functional groups are located. This means that a Type A fibre
has a surface structure as shown schematically in Fig. 4.8.
HM CF surface has a more ‘graphitic’ surface because of the higher axial orientation
of the graphite planes. Since the planes run parallel to surface, the electrolytic
oxidation occurs at the tilt boundaries between the crystallites, which have a scale
of 6–8 nm [25]. Figure 4.9 illustrates this mechanism.
80 F.R. Jones
Fig. 4.9 Electrolytic oxidation of HM carbon fibre showing the formation of micropores, func-
tionalised and adsorbent sites by etching at crystal boundaries. • is an active site or functional
group; o is an adsorbate molecule such as water. (a) Before oxidation, (b) after oxidation [20, 26]
log Œ1 C N .O/
DFT D (4.2)
log Œ1 C N.O/100
where N(O *) is the number of moles of active oxygen at the fibre surface per cm of
fibre tow (containing 6000 fibres). N(O *)100 represents the 100 % surface treatment
identified for optimal composite performance. The 100 % surface treatment is
defined by the seminal work, in Fig. 4.10, of Dunford et al. [27].
Table 4.1 shows the calculated
p acidity
p as a function of the surface treatment and
the average separation ( 100/n or 10/ n) of n-labelled carboxylic acid groups. It
was also shown that the tightly bound barium label was attached to two adjacent
carboxyl groups. The acidity increased linearly with current density up to a DFT
of 49 % and thereafter remained constant. Furthermore, while the acidity levels of
the untreated fibres are comparable to those measured by direct titration [28], it is
not possible to make a direct comparison of the surface acidities of treated fibres
because Ehburger et al. [28] used nonelectrolytic oxidation. Furthermore, other
carbon–oxygen functional groups (COOH, OH, CDO, lactones, anhydrides) are
introduced into the surface of a carbon fibre during oxidation which complicates the
analysis. For example, at 43 % DFT, only 470 COOH groups account for 940 of the
3500 oxygen atoms on the 100 nm2 segment surface. In addition, adsorbed water at
active sites could explain the higher oxygen concentration.
4 Serendipity in Carbon Fibres: Interfaces and Interphases in Composites 81
Table 4.1 The calculated number of oxygen atoms and barium-labelled groups
(COOH) on a unit segment of 40,000 carbon atoms on HM carbon fibre surface,
obtained by XPS [14, 25]
Unlabelled Ba-labelled
Average
No. of separation of True no. of Fibre
No. of O No. of Ba labelled labelledcarboxyl groups
p acidity
DFT (%) atoms atoms groups (n) groups 10p nm
.n/
3
n 6252 (eq./g)
0 1700 26 52 1.4 270 3.1
2700 53 110 0.95 350 ˙0.4
1600
23 3400 69 140 0.85 380 3.9
3300 84 170 0.77 400 ˙0.1
43 3400 120 240 0.65 450 4.7
3600 150 300 0.58 490 ˙0.2
140 280 0.60 480
49 4500 200 400 0.50 540 6.3
4400 220 440 0.48 560 ˙0.8
4700 390 780 0.36 780
4300 310 620 0.40 620
73 4900 192 380 0.51 530 5.4
6300 210 420 0.49 550 ˙0.1
100 3800 120 240 0.65 450 5.4
4000 210 420 0.49 550 ˙0.9
290 580 0.42 610
Table 4.2 The extent (XPS) of barium chemisorption by HM carbon fibres (in
atomic percent) after pre-adsorption of diglycidyl ether of bisphenol A (DGEBA,
n D 0.1) epoxy resin from solution
DFT (%) Treatment As received (atomic %) Epoxy resin (atomic %)
C O N Ba C O N Ba
0 Unlabelled 95 4.8 0 – 89.5 10.2 0 –
0 Ba-labelled 92 6.9 1.2 0.09 90 9.4 0 0
49 Unlabelled 87 9.6 3.4 – 82 14.6 2.2 –
49 Ba-labelled 80 13.3 2.7 0.59 83.5 13.2 2.75 0.16
DFT is the % degree of fibre surface treatment as defined in Eq. (4.2) [14, 29]
reduced the number of the acid groups that could be labelled with Ba2C (Table 4.2),
and only partial coverage of the surface was achieved.
The model given in Fig. 4.9 shows that the micropores, which form during
electrolytic oxidation, will need to have similar dimensions to the epoxy group(s) in
the resin molecule for interaction. The dimensions of the micropores were quantified
from the adsorption of chloroform, CHCl3 , as a function of the degree surface
treatment [30]. Typical values are given in Table 4.3. In this way, it was shown
that the functional groups located in the micropores were only accessible to epoxy
molecules of similar dimensions as illustrated in Fig. 4.11.
The role of the functional groups in forming a strong chemically bonded interface
with epoxy resin as opposed to a ‘physical’ bond resulting from the microporosity
was also studied using Ba2C labelling.
Table 4.3 The effect of heat treatment of the HM carbon fibres in vacuo to remove oxygen functional groups of fibre (detreatment)
for 2 h at high temperature (1000 or 1400 ı C) on the microporosity and acidity of fibres and properties of their epoxy resin composites
[14, 31–33]
Fibre or CHCl3
TD (ı C) XPS surface atomic concentrations (%) (molecules/100 nm2 ) ILSS (MPa)
Unlabelled Ba-labelled CHCl3 -labelled TGDDM DGEBA
C O N C O N Ba C O N Cl
(HMU) 95 5.2 0 92 6.9 1.2 0.09 95 3.8 0 0.95 135 24.8 25.0
(HMT) 86.5 10.0 3.3 85 11.6 3.1 0.46 86 10.6 3.6 0.68 105 72.8 78.0
1000 97.5 2.8 0 85.5 13.5 0.85 0.49 95 5.0 0 0.95 135 70.3 –
1400 – – – – – – – 96 3.7 0 0.4 56 43.7 75.1
TD detreatment temperature, HMU untreated, HMT treated or oxidised, ILSS interlaminar shear strength, TGDDM four-functional
epoxy, DGEBA difunctional epoxy 80 phr NMA hardener, 1.5 phr BDMA accelerator. DGEBA is the diglycidyl ether of bisphenol
A of degree of polymerisation 0.1. TGDDM is tetraglycidyl-4,4-diaminodiphenylmethane
4 Serendipity in Carbon Fibres: Interfaces and Interphases in Composites
83
84 F.R. Jones
[35] pioneered the technique SIMS imaging to study interfacial effects in composite
materials. Composites with low and interlaminar shear strength (ILSS) were studied.
It was shown that CN (26 D) secondary ion was representative of the CF
surface, while Cl (35 D) was a suitable label for the epoxy resin. The Cl was
attributed to the epoxy resin matrix (residual reactants—residual epichlorohydrin—
from DGEBA synthesis).
Figure 4.13 shows that time-of-flight secondary ion mass spectroscopy provided
the opportunity for creating chemical microscopic images which demonstrated that
the failure mechanism shifted from interfacial to matrix. The CFs in Fig. 4.13a were
untreated, while in Fig. 4.13b had been surface treated electrolytically employing
230 cm2 . It was observed that the interface in the untreated CF composite was
clearly interfacial, as shown by the distinct ‘red’ bands associated with the bare
fibres. The ‘green’ bands represent the resin matrix. On surface treatment of the CF,
the locus of failure has shifted predominantly into the matrix resin. The ‘yellow’
regions arise from the spectrometer receiving secondary ions from both matrix and
fibre. Of interest is the presence of patches of bare fibres but most are covered in a
thin resin layer. Thus, the failure crack has propagated in the interphase region. The
interphase is a region of ‘matrix’ adjacent to the fibre surface (i.e. at the interface)
so it is not surprising that the crack will be located in this region since the shear
stresses are concentrated here.
Interphase
Fig. 4.14 Typical results from the fragmentation test (A) complete interfacial debonding (CD),
(B) partial interfacial debonding (PD), (C) mixed mode which includes transverse matrix cracking
and partial debonding (MM) and (D) transverse matrix cracking (TMC); fibre breaks are shown
by arrows. The variation in adhesion from (A) low to (D) high was achieved by an increasing
concentration of acrylic acid in a plasma polymer from acrylic acid and 1,6 hexane. See
Fig. 4.15 [44]
There are several methods for measuring a value of interfacial shear strength, but
this article is limited to the fragmentation test since it has been used extensibly.
The readers can find full details of interfacial quantification in [40–43]. Figure 4.14
is a compilation optical micrographs showing the principal mechanisms observed
as the interfacial strength increased. The interfacial performance was modified by
increasing the concentration of –COOH in the coating on the CF surface. This was
achieved by conformal low-pressure plasma copolymerisation of acrylic acid and
hexane.
The fragmentation test requires an extensible matrix to ensure that saturation is
achieved before the matrix fractures. Generally for CFs, this means that the matrix
failure strain ("u ) should exceed 6 %. If a lower-failure strain matrix needs to be
studied, the microdebond test or pull-out test has to be used. Alternatively, the laser
Raman measurement of fibre stress can be used to map the stress transfer from the
end of a short fibre or fracture. The Raman technique relies on the band shift under
strain.
The technique has been effectively used for high-modulus CFs, other highly
crystalline fibres (such as Kevlar) and carbon nanotubes and graphene sheets
but has limited sensitivity when applied to high-strength and Type A CFs [45].
Photoelasticity has also been used for measuring interfacial shear stresses directly,
but the diameter of CF of 7 m limits the observations in an optical microscope [46].
88 F.R. Jones
Conventionally the Kelly–Tyson (or constant shear stress) stress transfer model
is employed because fragmentation ceases when the fragments are completely
debonded [47]. In which case the interfacial shear strength is given by
fu.Lc / rf
u D (4.3)
Lc
Lc D 4Lf =3 (4.4)
where u is the interfacial shear strength, rf is the fibre radius and Lc is the critical
fibre length. fu.Lc / is the strength of a fibre of length Lc. This is obtained from
a separate experiment in which Weibull statistics of fibre strength is measured on
‘short’ fibres of a few millimetres in length so that an accurate value of fu.Lc / can
be calculated.
Since a fragment of L Lc will break into two fragments, Lc is related to the
average fragment length Lf according to Eq. (4.4).
Examination of Fig. 4.14 shows that complete debonding was not always
observed for fragments at saturation and the properties of the matrix needed to be
included in the analysis. Tripathi et al. [48–50] input the yield and debond lengths
into the variational model of Nairn [51] as shown in Fig. 4.15.
Variational
mechanics
Yielded region
Debounded region
Fibre length
Fig. 4.15 The calculated shear stress profile across a fragment illustrating the plasticity effect
model of Tripathi et al. [50] which interrupts the elastic model of Nairn [51] to account for matrix
yield and debonding
4 Serendipity in Carbon Fibres: Interfaces and Interphases in Composites 89
Fig. 4.16 The cumulative stress transfer function (CSTF) as a function of the relative concentra-
tion (XPS) functional groups (amine, hydroxyl or carboxyl) achieved using plasma copolymers.
See Fig. 4.14 for definition of CD, PD, MM and TMC [52]
iDN Z
X
L1
f .x/ dx
iD1 0
CSTF D (4.5)
X
iDN
Li
iD1
A stress transfer efficiency (STE) can also be calculated according to Eq. (4.6):
CSTF
STE D (4.6)
ŒCSTF0
where [CSTF]0 is the stress transfer function for a fragment with perfect bond in an
elastic matrix.
Figure 4.16 shows how the CSTF approach differentiates between the differing
degrees of adhesion identified in the micromechanics of the fragmentation described
in Fig. 4.14. The results in Fig. 4.16 appear to show how the chemistry influences
interface bond formation. Surface –COOH groups appear to react with the epoxy
90 F.R. Jones
resin more efficiently than –NH2 groups, while –OH groups are inefficient in
forming a good interfacial bond. The role of chemical coupling at the interface is
confirmed since plasma polymers are conformed coatings of molecular dimensions
and would block interaction in the micropores on the CF surface.
Figure 4.12 shows that polymers are strongly adsorbed onto CF surfaces, so we
can expect that sizing polymers will not be completely solubilised in the resin
during impregnation. Therefore, we can expect either complete dissolution to create
a graded interphase or in contrast poor dissolution leads to a distinct interphase.
Figure 4.17 shows that this mechanism is critical to the measured interfacial shear
strength. In this case, a thermoplastic matrix, polyethersulphone, has been studied.
Differing sizes were used with high-strength CF [34]. We see that the interfacial
shear strength obtained by the fragmentation test is higher for the sized fibres
(TBrES, TPESS) compared to the unsized fibre (TUS). Surface treatment of the
CF is still beneficial (see UTUS and TUS). We postulated that the sizing polymer
was compatible with the PES matrix. This was confirmed by Tof SIMS chemical
imaging of the fracture surface of a composite. The image in Fig. 4.18 is dominated
by the yellow colour which arises from the secondary ions sputtered from both the
sizing polymer (red label) and matrix (green label). This confirms that the sizing
polymer dissolves in the matrix to create a graded interphase.
In contrast to the data in Fig. 4.17, Fig. 4.19 shows that the interfacial shear
strength of the sized fibre (TBrES) is lower than that for the unsized (TUS) fibre.
The implication is that the sizing polymer is not compatible with the matrix. The
surface treatment of the CF is still beneficial to interfacial bond strength to the cured
epoxy resin (see UTUS and TUS).
ToF chemical image in Fig. 4.20 shows that the fracture surface is dominated by
the sizing polymer, which is labelled by a red colour, representing the sputtered Br
secondary ions from the epoxy polymer used as the fibre coating. The absence of
sulphur (S ) secondary ions from the cured epoxy resin shows that sizing has not
interacted with the matrix. We can infer that a distinct interphase had formed and
92 F.R. Jones
Fig. 4.21 The concept of the interphase structure in the preparation of carbon fibre composites.
(a) Graded interphase from brominated epoxy resin-sized fibre in thermoplastic PES matrix and
(b) distinct interphase from brominated epoxy resin-sized fibre in a thermoset epoxy resin matrix.
The locations of the fracture crack are also indicated [34]
that failure had either occurred within the interphase or at the interface between
sizing and matrix. If failure had occurred at the interface between the distinct
interphase and the matrix, matrix secondary ions should have been present.
Figure 4.21 is a schematic description of these conclusions showing how
interphase formation can differ for various combinations of sizing polymer and
matrix resin.
Generic interphase formation involves a number of mechanisms which are
summarised in Fig. 4.22.
4 Serendipity in Carbon Fibres: Interfaces and Interphases in Composites 93
Fig. 4.22 Generic interphase formation in carbon fibre composites including the contribution of
the interface [14, 36]
Figures 4.14 and 4.16 show how plasma polymers can be used to tailor the interface
of carbon fibre composites. Plasma polymerisation can provide the opportunity to
control interphasal properties. It proved experimentally difficult to handle unsized
CF in continuous coating of fibre tows for manufacture of high-Vf composites.
Therefore, glass fibres were used in subsequent studies. This article cannot cover
these aspects in great detail, but Fig. 4.23 shows how design of the interface and
interphase can lead to improvements in mechanical properties. Swait et al. [53]
showed that both composite tensile strength and toughness could be improved.
The optimum low-pressure plasma polymer coating on E-glass fibres was
observed to be from the monomer feed containing 40 % acrylic acid/60 % octadiene.
The improved strength of the composites arose from a combination of an increase in
fibre failure strain and improved interface and the introduction of matrix cracking.
The latter causes an extended stress–strain curves as shown in Fig. 4.24.
94 F.R. Jones
Fig. 4.23 Tensile properties of 0ı glass fibre epoxy composites, with 10 nm plasma polymer
interphases with increasing acrylic acid concentration in the monomer feed with 1,7 octadiene. (a)
Fracture energy estimated from area under the stress–strain curve, (b) tensile strength normalised
to measured fibre volume fraction (Vf ). The dashed lines represent the values for uncoated glass
fibres with ˙standard deviation [53]
The limitation to the plasma polymer coating discussed here is the need for low-
pressure conditions. Current research is attempting develop atmospheric plasma
polymerisation for fibre coating with retention of the intended functionality in the
coatings [54]. In principle, the functional chemistry of the coating can be readily
tuned to the matrix employed for the composite manufacture. Furthermore, an on-
line plasma coating system could replace both the electrolytic oxidation and sizing
stages and since it is a gas phase process and it is more environmentally friendly.
4 Serendipity in Carbon Fibres: Interfaces and Interphases in Composites 95
Fig. 4.24 The extended tensile stress–strain curves of the optimal composite from E-glass fibres
with a 40 % acrylic acid/60 % 1,7-octadiene plasma polymer coating interphase [53]
4.5 Conclusions
The paper has discussed the development of high-strength and high-modulus carbon
fibres from polyacrylonitrile. The serendipity of the availability of Courtelle acrylic
fibres as precursors for the work of W Watt is considered. Equally Watt’s interest
in establishing the surface functionality of electrolytically oxidised high-strength
CF inspired the author’s interest in interfacial chemistry and micromechanics.
The development of understanding of interfacial micromechanics of carbon fibre
composites is discussed logically. It is also shown how controlling the interphase
structure is critical in optimising mechanical performance of these composites for
structural applications. The use of functional plasma polymer coatings to replace
the traditional oxidation and sizing stages is considered. This represents a future
opportunity once the use of low-pressure conditions for ensuring the deposition of
conformal functional coatings becomes commercial or can be replaced by more
industrially friendly atmospheric plasma technology.
Acknowledgements The following have made a significant contribution to the work described
here over many years: G Dorey, J E Bailey, M Alexander, T Choudhury, T H Cheng (deceased),
P Denison, A Kettle, A Abu Bakar, S A Hayes, C Johnson, R Lane, S Krishnamurthy, N
Lopattananon, D Marks, D Wang, D Tripathi, F M Zhao, S Behzadi, Z Liu, X Liu, S Yumitori,
P Photjanataree, T Swait, H Sugihara, J Watts and W Watt (deceased).
96 F.R. Jones
References
1. W. Watt, Chemistry and physics of the conversion of polyacrylonitrile fibres into high modulus
carbon fibres, in Strong Fibres, ed. by W. Watt, B.V. Perov (Elsevier, Amsterdam, 1985),
pp. 327–388
2. P. Morgan, Carbon Fibres and Their Composites (Taylor & Francis, Boca Raton, 2005)
3. R.C. Houtz, J. Text. Res. 20, 786–801 (1950)
4. A. Shindo, Osaka Kogyo Gijitsu Shikenjo Kiha 110, 119 (1961)
5. A. Shindo, Osaka Kogyo Gijitsu Shikenjo Hokoku 317 (1961)
6. W.G. Schmidt, Courtaulds Ltd., British Patent, GB796294 (1958)
7. R. Moreton, in The Tensile Strengths of PAN Based Carbon Fibres in Strong Fibres, ed. by
W. Watt, B.V. Perov (Elsevier, Amsterdam, 1985), pp. 445–474
8. E. Frank, L.M. Steudle, D. Ingildeev, J.M. Spörl, M.R. Buchmeiser, Carbon fibers: precursor
systems, processing, structure, and properties. Angew. Chem. Int. Ed. 53, 5262–5298 (2014)
9. R. Moreton, W. Watt, W.E. Johnson, Nature 213, 690 (1967)
10. W. Watt, W. Johnson, Nature 257(5523), 210–212 (1975)
11. P.J. Goodhew, A.J. Clarke, J.E. Bailey, Mater. Sci. Eng. 17, 3–30 (1975)
12. F.R. Jones, Reinforced plastics composites in modern aluminium alloys, in Treatise on
Materials Science and Technology, ed. by R. Doherty, A. Vasudevan, vol 31 (Academic Press,
New York, 1989), pp. 605–469
13. S.C. Bennet, D.J. Johnson, W. Johnson, J. Mater. Sci. 18, 3337 (1983)
14. F.R. Jones, J. Adhes. Sci. Technol. 24, 171–202 (2010)
15. W.N. Reynolds, J.V. Sharp, Carbon 12, 103 (1974)
16. J. Harvey, RAE Technical Report, TR86077 (HMSO, London, 1986)
17. J. Mahy, L.W. Jenneskens, O. Grabandt, A. Venema, G.D.B. Houwdingen, Surf. Interface Anal.
21, 1–13 (1994)
18. T.R. King, D.F. Adams, D.A. Butloy, Compos. Sci. Technol. 44, 351–359 (1994)
19. C. Kozlowski, P.M.A. Sherwood, Carbon 25, 751 (1987)
20. F.R. Jones (ed.), Handbook of Polymer-Fibre Composites (Longman, Harlow, 1994)
21. M.R. Alexander, F.R. Jones, Carbon 34, 1093–1102 (1996)
22. G. Dorey, in The Fibre-Matrix Interface and Composite Strength in Composite Materials in
Aircraft Structures, ed. by D.H. Middleton (1990), pp. 51–68
23. J.E. Castle, J.F. Watts, The study of interfaces in composite materials by surface analytical
techniques, in Interfaces in Polymer, Ceramic and Metal Matrix Composites, ed. by H. Ishida
(Elsevier, New York, 1988), pp. 57–71
24. P. Denison, F.R. Jones, J.F. Watts, J. Mater. Sci. 20, 4647–4656 (1985)
25. P. Denison, F.R. Jones, J.F. Watts, Surf. Interface Anal. 9, 43–435 (1986)
26. F.R. Jones, in Handbook of Adhesion, ed. by D. Packham, 2nd edn. (Wiley, Chichester, 2004),
pp. 177–186
27. D.V. Dunford, J. Harvey, J. Hutchings, C.H. Judge, RAE Technical Report TR81096 (HMSO,
London, 1981)
28. P. Ehburger, J.J. Herque, J.B. Donnet, in Proceeding of the 5th London International Carbon
and Graphite Conference, vol. 1 (Society of Chemical Industry, London, 1978), p. 398
29. P. Denison, F.R. Jones, J.F. Watts, in Interfaces in Polymer, Ceramic and Metal Matrix
Composites, ed. by H. Ishida (Elsevier, New York, 1988), pp. 77–85
30. P. Denison, F.R. Jones, J.F. Watts, Surf. Interface Anal. 12, 455–460 (1988)
31. P. Denison, F.R. Jones, A.J. Paul, J.F. Watts, in Interfacial Phenomena in Composite Materials,
ed. by F.R. Jones (Butterworths, London, 1989), pp. 105–110
32. P. Denison, F.R. Jones, G. Dorey, L.F. Jones, in Carbon 88, ed. by B. McEnaney, T.J. Mays
(IOP, London, 1988), pp. 634–366
33. F.R. Jones, in Proceeding 28th Risoe International Symposium on Materials Science: Interface
Design of Polymer Matrix Composites—Mechanics, Chemistry, Modelling and Manufacturing,
ed. by B.F. Soerensen, L.P. Mikkelson, H. Lilholt, S. Goutianos, F.S. Abdul-Mahdi (Risoe
National Laboratory, Roskilde, 2007), pp. 21–44
4 Serendipity in Carbon Fibres: Interfaces and Interphases in Composites 97
5.1 Introduction
After the great success of carbon fibre (CF), scientists have kept exploring the
world of carbon-based materials. Carbon has a variety of allotropes and structures
of various dimensionalities due to the ability of hybridization in sp, sp2 and
sp3 , which makes it a highly versatile element. Carbon-based materials have
attracted significant attention since the mid-1980s, especially by the discovery of
fullerenes, followed by carbon nanotubes and graphene. The unique properties of
nano-carbons make them of interest for a wide range of applications including
energy, environment, biology, medicine, etc. Carbon nanotubes (CNTs) have been
of particular interest to the composite community because they are essentially
nanosized carbon fibres possessing a defect-free structure. The nanostructures
of CNTs have brought about various extraordinary properties such as electrical,
mechanical and thermal properties, making many researchers believe that CNTs
were the new wonder material that could replace carbon fibre in various industrial
applications. However, so far it has been proven difficult for CNTs to right out
replace carbon fibres, mainly because of processing issues related to fibre alignment
and loading. While typical carbon fibre composites exhibit fibre volume fractions
as high as 60 %, CNT-based composites are often limited to a few percent, limiting
the ultimate properties of these composites. For this reason, many researchers have
been exploring the potential of combining the micron- and nano-world of carbon
in order to create high-performance multifunctional composites. By combining the
Fig. 5.1 In situ growth of CNTs onto CF, with good spatial distribution of CNTs [5]
Fig. 5.2 Illustration of EPD method [14] and SEM images of (a) acid-treated and (b) ozone-PEI-
treated MWCNTs deposited on carbon fibre fabrics [15]
Fig. 5.3 (a) Illustration of spray-coating process for carbon fibre prepregs [18]; (b) and (c) SEM
images of percolated CNT networks on fabrics via spray-coating technique (image courtesy of Dr.
Han Zhang, QMUL)
functionalization on the EPD process and resulting properties, with both acid-
treated and ozone-PEI treated CNTs [15]. Good dispersion was achieved with
functionalized CNTs since the surface charge was developed to facilitate CNT
dispersion and mobility under an electric field. In general, the surface charge of
CNTs needs to be modified in order to provide attractive forces between CNTs and
CFs during the EPD process.
Very recently, spray-coating technique has also been adapted for CNT deposition
and localization in carbon fibre laminates, with good CNT distribution, spatial
control and, most importantly, feasibility to scale up [11, 16–18]. After dispersion of
CNTs, Zhang et al. utilized spray coating to deposit CNTs onto carbon fibre fabrics,
resulted in a fine percolated network on fabric surfaces (Fig. 5.3). This method
was also utilized on carbon fibre prepregs to localize CNTs at the damage-prone
5 Nano-Engineered Hierarchical Carbon Fibres and Their Composites. . . 105
interlaminar region [18]. The original high aspect ratio and intrinsic high electrical
conductivity were preserved in these CNTs since no functionalization was required
for this process.
Other methods have also been developed with the aim of coating CNTs onto
CF fabrics. Drzal et al. have developed a continuous coating process by immersing
surface oxidized CFs through a CNT suspension [19], with the CNTs being func-
tionalized to possess positive charge in order to attach onto the negatively charged
CF fabrics. These cationic polymer-treated MWCNTs showed good compatibility
with oxidized carbon fibre, resulting in uniform coatings [19]. CNT deposition
was also achieved by immersing the CFs into the suspension of functionalized
CNTs [20].
In short, various deposition methods have been developed in recent years to
produce hierarchical nano-engineered composites. Table 5.1 summarizes some of
the most common methods with their advantages and drawbacks. Clearly, no method
has the absolute edge over others to suit everything. Depending on processing and
property requirements, certain selection criteria need to be considered before these
methods can be selected to prepare the multi-scale CNT-CFRPs.
One of the main motivations of introducing CNTs into CFRPs is to improve the
mechanical performance of traditional carbon fibre laminates, especially for rela-
tively weak out-of-plane properties. A fair amount of research has been conducted
in this area so far, with improvements reported from interfacial bond strength
between CF and epoxy resins to macroscopic properties like interlaminar fracture
toughness. In this section, the mechanical performance of multi-scale CNT-CFRPs
is presented for various nano-engineering methods, with a focus on both interfacial
and macroscopic properties.
With the presence of grown or deposited CNTs at CF surfaces, interfacial
properties such as the interfacial shear strength can be improved due to an improved
interaction as well as increased surface area and roughness [3, 7, 12, 21–24]. After
the grafting of CNTs onto CFs via CVD method, the apparent interfacial shear
strength (IFSS) reportedly increased by 60 % as measured by single-fibre pull-
out tests [12]. Guo et al. utilized EPD process to deposit CNTs onto CFs with
homogeneous dispersion and obtained an improved tensile strength and Weibull
modulus from single-fibre tests, as well as an obvious increment (68 %) in interfacial
shear strength from single-fibre pull-out tests [7]. By controlling the orientation as
well as length of CNTs grown on CFs, significant improvements in IFSS (175 %)
were reported with perpendicular alignment and long nanotube lengths (47.2 µm)
compared to pristine CFs [25].
However, it is generally acknowledged that extensive oxidations under elevated
temperatures yield a reduction in the tensile strength of carbon fibre, especially
for the case of in situ growth of CNTs using high-temperature CVD methods.
106
It has been reported that after CVD processes with random and aligned CNTs,
the ultimate tensile strength of the carbon fibres decreased by 30 and 37 %,
respectively, compared to untreated reference fibres [23], while an even higher
strength loss of 55 % was reported in another study [12]. These strength reductions
in carbon fibre were attributed to thermal degradation-induced surface flaws as well
as surface oxidation. In another study performed by Qian et al., a 15 % reduction
in fibre strength from single-fibre tensile tests was observed, which was believed
to be due to the dissolution of catalyst into the carbon fibre at high processing
temperatures [22]. Similar trends were found in fragmentation tests, where the CNT-
grafted fibre showed a 14 % reduction in tensile strength compared to non-grafted
carbon fibre [21]. Therefore, the reaction temperature and time need to be controlled
in order to avoid a reduction in carbon fibre strength. Effects of CVD conditions on
fibre tensile properties have also been studied [26, 27], while attempts have been
made using water-cooled substrates during CVD processing to reduce carbon fibre
property reduction [4]. Wardle et al. also demonstrated that tensioning the carbon
fibre to 12 % of its tensile strength during CVD results in a preserved fibre strength
and stiffness even at elevated temperatures [27].
It is worth noting that with improved interfacial bonding and decreased fibre
strengths, it is possible to obtain cohesive failure modes in carbon fibre upon com-
pressive loading (i.e. single-fibre push-out). Therefore, with the aim of improving
the mechanical performance of carbon fibres and their composites, optimized rather
than maximized interfacial bonding is required. However, it should be realized that
although certain levels of interfacial shear strength improvements were achieved
by grafting CNTs onto carbon fibres, some of the commercial sizings or surface
treatments on carbon fibres provide better mechanical performances [23]. Hence,
more focus should be placed on improvement of macroscopic properties of CFRPs,
as well as added multifunctionalities by CNTs. However, after optimization of
various CNT grafting parameters, such as temperature, catalyst choices and density
for the CVD process, functional groups and surface charges for EPD process, the
improved interfacial strength together with preserved carbon fibre properties will
provide a huge potential for the enhancement of mechanical properties of CFRPs.
In the case of nano-engineered CF fabrics, the macroscopic properties, especially
matrix- and interface-dominated properties such as interlaminar shear strength
(ILSS) and interlaminar facture toughness (GIC ), have all been successfully
improved. With the presence of CNTs at CF surfaces, increased surface area as well
as enhanced interphase properties could all improve the properties at interlaminar
regions in composites, resulting in improved delamination and shear resistance.
Bekyarova et al. used EPD methods to deposit CNTs onto CF fabrics and obtained
a 30 % increment in interlaminar shear strength (ILSS). Flexural stiffness and
strength of these CFRPs were also increased by utilizing CVD-grown CNTs onto
CF fabrics [28]. Li et al. used EPD to deposit CNTs and improved the compressive
strength and modulus of CNT-modified carbon/carbon composites by 37 % and
19 %, respectively, compared to the reference system [8].
Delamination of CFRPs which is one of the most common failure modes in
carbon fibre laminates has also been investigated, and improvements in delamination
108 H. Zhang et al.
resistance via nano-modified CFRPs have been achieved. Zhang et al. obtained a
50 % increment in GIC value by spray-coating CNTs onto carbon fibre prepregs with
extremely low CNT loadings (0.05 wt.% to the composite) [18]. CVD-grown CNTs
also resulted in improved fracture toughness of composites [5]. In another study
performed by Falzon et al., CVD-grown CNTs on silicon substrates were transferred
to T700/M21 prepregs, leading to a 31 % increase in fracture toughness [13]. An et
al. utilized ozone-treated MWCNTs chemically reacting with PEI dendrimers and
coated them onto CF via an EPD process. They obtained a near doubling in GIC
value with 40 min deposition time, alongside with improved in-plane shear strength
compared to CFRPs without CNT modifications [15]. A comparison of various
interlaminar fracture toughness improvements reported for hierarchical CNT/CF
composites taken from [18] is shown in Fig. 5.4, indicating high toughening
efficiency by the localized spray-coating method and the possibility of introducing
high CNT loadings via EPD methods.
Great success has been achieved by grafting and depositing CNTs onto carbon
fibres, with improved interfacial properties for single-fibre composites as well as
improved macroscopic properties reported, especially out-of-plane properties of
CFRPs. Although some challenges remain regarding processing or scale-up aspects,
nano-modified CFRPs demonstrated a promising solution to overcome the relatively
weak out-of-plane mechanical performance of CFRP.
Fig. 5.4 Summarized interlaminar fracture toughness improvement in CFRPs through the intro-
duction of CNTs taken from [18], where solid markers are for non-functionalized MWCNTs and
hollow markers are for either DWCNTs or functionalized CNTs, while shadowed marker bars are
for specimen produced by EPD
5 Nano-Engineered Hierarchical Carbon Fibres and Their Composites. . . 109
Fig. 5.5 Improved stability in sensing signals due to a changed conduction mechanism with the
presence of a spray-coated CNT network [18]
was greatly improved, especially in the elastic deformation region [18]. This is
believed to be due to a change in conduction mechanism as shown in Fig. 5.5 [18].
Under standard delamination test conditions, the observed electrical sensing
signals can be correlated to the applied load, revealing the internal damage state
in composite laminates upon loading. Figure 5.6 shows that with increased load,
relative resistance changes remain constant before the specimen starts to yield.
Therefore, in this elastic deformation region, the electrical signals from the CNT/CF
hierarchical composites can be used as strain sensor. Once cracking is initiated,
sudden load drops can be observed, which are reflected by incremental changes
in the sensing signal. With cracking, the overall resistance of the specimen is
increasing due to the increased distance between electrodes, and each increment in
resistance value is associated with an internal breakdown of the CNT network due
to crack propagation. As shown in Fig. 5.6b, excellent correlation between internal
damage state and sensing signal can be observed which is believed to be very
important for such damage sensing and structural health monitoring applications.
Other failure modes such as interlaminar shear failure have also been successfully
monitored with CNT networks in CFRP. It is found that with different testing
manners, the obtained sensing signals can be optimized to provide progressive
signals which are believed necessary and useful for structural health monitoring
purposes [11]. Under interlaminar shear testing conditions, the through-thickness
sensing signals progressively changed upon loading with an abrupt signal change
at the point of failure, while no obvious sensing signals were obtained for in-plane
measurements (Fig. 5.7) [11].
These developed sensing methodologies based on CNTs with improved signal
stability and consistency possess advantages over traditional non-destructive testing
methods due to aspects related to materials’ integration, weight saving and preserved
or improved mechanical properties. Recent researches have shown great promises
5 Nano-Engineered Hierarchical Carbon Fibres and Their Composites. . . 111
Fig. 5.6 In situ damage sensing of CNT-modified CFRPs under mode I delamination tests,
revealing the possibility of utilizing electrical signals for internal health monitoring of composite
laminate [18]
Fig. 5.7 In situ damage sensing of interlaminar shear strength (ILSS) testing for CNT-modified
CFRP for (a) in-plane and (b) through-thickness measurements [11]
So far, recent progress on CNT-modified CFs and their composites has been
reviewed, starting from different manufacturing methods such as CVD, EPD
and spray coatings to various mechanical and functional properties. The resulted
morphologies from these different methods have been reviewed, with concluded
advantages and drawbacks for each individual method. The mechanical performance
at both single-fibre level and composite level has been discussed, while added
multifunctionalities such as strain and damage sensing have been analysed.
Enhanced mechanical properties at both interfacial and interlaminar levels have
been achieved by CNT modification onto CF surfaces. Increased surface area and
improved matrix properties near the interfacial region are believed to contribute
to the improved mechanical performance. For electrical and thermal properties,
improvements were mainly achieved in through-thickness direction where CNTs
were grown vertically on CFs. Regarding the multifunctionalities, the CNT-modified
CFRPs have also shown great promises for in situ damage sensing and structural
health monitoring with enhanced signal stability.
In order to fully utilize these multi-scale reinforcements, especially from labo-
ratory success to industrial products, a number of issues need to be considered and
addressed including (1) preservation of CF original properties, (2) simplicity and
ability to scale up for manufacturing processes and (3) alignment and orientation of
CNTs.
Firstly, it is obvious that due to the chemically inert nature of CFs, their surfaces
need to be modified in order to improve the interaction with polymer matrices.
Various surface treatments have been developed, with the aim of improving adhesion
and interfacial shear strength leading to better load transfer between fibre and
matrix as well as improved composite properties. It is well known that most
commercial CFs are also been “sized” to have improved wettability with matrix
materials. Although above-discussed nano-engineered CF surfaces can induce
improved properties and new functionalities over traditional surface treatments, the
preservation of the original CF properties such as high tensile strength and modulus,
as well as compressive strength, remains a high priority. The same principles apply
to the preservation of electrical properties in order to enhance electrical properties
and multifunctionalities.
Secondly, among the listed preparation methods, no single method has absolute
advantages over others at this development stage. The costs and complexity of
processing steps like CVD and EPD need to be reduced, while solvent recovery and
environmental effects need to be considered for spray- and dip-coating processes.
An ideal process or a combination of processes needs to be simple and versatile,
suitable for different complex shapes with controlled deposition, at relatively low
cost to promote scale up and industrial usage.
Thirdly, the alignment and orientation of CNTs on CFs needs to be addressed.
Several studies have already demonstrated that vertically grown CNTs can greatly
improve the relatively weak through-thickness performance of CFRP, for both
114 H. Zhang et al.
mechanical and electrical properties. Hence, with the aim of further enhancing the
out-of-plane properties of CFRPs, the CNT alignment needs to be controlled to fully
utilize the effect of nano-modifications.
References
1. S. Zhu, C.H. Su, S.L. Lehoczky, I. Muntele, D. Ila, Carbon nanotube growth on carbon fibers.
Diamond Relat. Mater. 12, 1825–1828 (2003). doi:10.1016/s0925-9635(03)00205-x
2. Z.G. Zhao, L.J. Ci, H.M. Cheng, J.B. Bai, The growth of multi-walled carbon nan-
otubes with different morphologies on carbon fibers. Carbon 43, 663–665 (2005).
doi:10.1016/j.carbon.2004.10.013
3. E.T. Thostenson, W.Z. Li, D.Z. Wang, Z.F. Ren, T.W. Chou, Carbon nanotube/carbon fiber
hybrid multiscale composites. J. Appl. Phys. 91, 6034–6037 (2002). doi:10.1063/1.1466880
4. T.R. Pozegic, I. Hamerton, J.V. Anguita, W. Tang, P. Ballocchi, P. Jenkins, S.R.P. Silva, Low
temperature growth of carbon nanotubes on carbon fibre to create a highly networked fuzzy
fibre reinforced composite with superior electrical conductivity. Carbon 74, 319–328 (2014).
doi:10.1016/j.carbon.2014.03.038
5. K.L. Kepple, G.P. Sanborn, P.A. Lacasse, K.M. Gruenberg, W.J. Ready, Improved fracture
toughness of carbon fiber composite functionalized with multi walled carbon nanotubes.
Carbon 46, 2026–2033 (2008). doi:10.1016/j.carbon.2008.08.010
6. A.R. Boccaccini, J. Cho, J.A. Roether, B.J.C. Thomas, E. Jane Minay, M.S.P. Shaf-
fer, Electrophoretic deposition of carbon nanotubes. Carbon 44, 3149–3160 (2006).
doi:10.1016/j.carbon.2006.06.021
7. J. Guo, C. Lu, F. An, Effect of electrophoretically deposited carbon nanotubes on the
interface of carbon fiber reinforced epoxy composite. J. Mater. Sci. 47, 2831–2836 (2012).
doi:10.1007/s10853-011-6112-5
8. K.Z. Li, L. Li, H.J. Li, Q. Song, J.H. Lu, Q.G. Fu, Electrophoretic deposition
of carbon nanotubes onto carbon fiber felt for production of carbon/carbon compos-
ites with improved mechanical and thermal properties. Vacuum 104, 105–110 (2014).
doi:10.1016/j.vacuum.2014.01.024
9. T. Takeda, Y. Shindo, T. Fukuzaki, F. Narita, Short beam interlaminar shear behav-
ior and electrical resistance-based damage self-sensing of woven carbon/epoxy com-
posite laminates in a cryogenic environment. J. Compos. Mater. 48, 119–128 (2014).
doi:10.1177/0021998312469240
10. D.D.L. Chung, Carbon materials for structural self-sensing, electromagnetic shielding and
thermal interfacing. Carbon 50, 3342–3353 (2012). doi:10.1016/j.carbon.2012.01.031
11. H. Zhang, Y. Liu, E. Bilotti, T. Peijs, In-situ monitoring of interlaminar shear damage in carbon
fibre composites. Adv. Compos. Lett. 24, 92–97 (2015)
12. H. Qian, A. Bismarck, E.S. Greenhalgh, G. Kalinka, M.S.P. Shaffer, Hierarchical composites
reinforced with carbon nanotube grafted fibers: the potential assessed at the single fiber level.
Chem. Mater. 20, 1862–1869 (2008). doi:10.1021/cm702782j
13. B.G. Falzon, S.C. Hawkins, C.P. Huynh, R. Radjef, C. Brown, An investigation of Mode I and
Mode II fracture toughness enhancement using aligned carbon nanotubes forests at the crack
interface. Compos. Struct. 106, 65–73 (2013). doi:10.1016/j.compstruct.2013.05.051
14. E. Bekyarova, E.T. Thostenson, A. Yu, H. Kim, J. Gao, J. Tang, H.T. Hahn, T.W. Chou,
M.E. Itkis, R.C. Haddon, Multiscale carbon nanotube-carbon fiber reinforcement for advanced
epoxy composites. Langmuir 23, 3970–3974 (2007). doi:10.1021/la062743p
15. Q. An, A.N. Rider, E.T. Thostenson, Electrophoretic deposition of carbon nanotubes onto
carbon-fiber fabric for production of carbon/epoxy composites with improved mechanical
properties. Carbon 50, 4130–4143 (2012). doi:10.1016/j.carbon.2012.04.061
5 Nano-Engineered Hierarchical Carbon Fibres and Their Composites. . . 115
33. J.S. Im, J. Yun, J.G. Kim, T.S. Bae, Y.S. Lee, The effects of carbon nanotube addition and
oxyfluorination on the glucose-sensing capabilities of glucose oxidase-coated carbon fiber
electrodes. Appl. Surf. Sci. 258, 2219–2225 (2012). doi:10.1016/j.apsusc.2011.08.017
34. A. Khan, A.A.P. Khan, A.M. Asiri, M.A. Rub, M.M. Rahman, S.A. Ghani, In vitro studies
of carbon fiber microbiosensor for dopamine neurotransmitter supported by copper-graphene
oxide composite. Microchim. Acta 181, 1049–1057 (2014). doi:10.1007/s00604-014-1202-0
35. J. Bai, L. Wu, X. Wang, H.M. Zhang, Hemoglobin-graphene modified carbon fiber microelec-
trode for direct electrochemistry and electrochemical H2O2 sensing. Electrochim. Acta 185,
142–147 (2015). doi:10.1016/j.electacta.2015.10.100
36. J. Bai, P. Qi, X. Ding, H. Zhang, Graphene composite coated carbon fiber: electrochemical
synthesis and application in electrochemical sensing. RSC Adv. 6, 11250–11255 (2016).
doi:10.1039/c5ra26620c
Chapter 6
Nano-engineered Carbon Fibre-Reinforced
Composites: Challenges and Opportunities
6.1 Introduction
The materials science community has been excited by the discovery of carbon
materials like fullerenes, nanotubes and graphene. Their exceptional multifunctional
characteristics opened up opportunities to engineer new and improve existing
properties in materials. Driven by high expectations, significant research efforts have
been put forward, including research in the field of polymer-based composites.
One of the most studied nanoscale reinforcements of polymers is carbon
nanotube (CNT). The first reference to tubular nanosized carbon filaments appeared
more than six decades ago in Radushkevich and Lukyanovich [1], but it is not
until the article of Iijima [2] that CNTs attracted interest of the materials science
community and became a subject of intensive research. According to the web of
science, the number of articles published in 2014 in the subject category of ‘carbon
nanotube’ and ‘composite’ exceeded 8000. The evolution of journal publications has
been gradually increasing since 1991 as shown in Fig. 6.1a. The relative fraction of
articles that report on CNTs in composites is now close to 9 % of the total number
of journal publications in composites (Fig. 6.1b).
One of the reasons for this immense interest in CNTs is their unique properties.
On the mechanical side, the stiffness and strength are superior to those of carbon
fibre—the best high-performance fibre known today. CNT’s stiffness and strength
are on the order of 1 TPa and 50 GPa, respectively [3]. Moreover, a multiwall
CNT may exhibit a high failure strain (12 % in [3]) which is due to their unusual
Fig. 6.1 (a) The number of journal articles per year in carbon nanotube research for composites;
(b) the relative percentage of articles reporting on carbon nanotubes in composites to the total
number of publications in the field of composites. The data is sourced from the Web of Science
Fig. 6.2 Different strategies for integration of CNTs in composites with microscopic fibres: (a)
and (b) CNTs in the matrix in dispersed and agglomerated states, respectively; (c) and (d) CNTs
grown on fibres with random and radially aligned morphologies, respectively. The collage is
assembled with images from Romanov et al. [8–10]
fibre-reinforced composites so far has been for the increase of interlaminar fracture
toughness, interlaminar shear strength and interfacial shear strength. CNTs have
also shown to hinder matrix cracking. These improvements are important for such
critical macroscopic properties as compressive strength after impact and fatigue life.
A wide range of improvement levels have been reported: from a few percent to
hundreds of percent. It has been shown to depend on a variety of factors: the way
CNTs are integrated in the composite, the type of CNTs, their surface functional-
ization, toughness of the base matrix and many other variables. It is typically much
more difficult to achieve significant improvements at the macro-scale. In addition
to the mechanical improvements, CNTs created opportunities to develop structural
composites with better energy capacity, lightning strike protection, resistance to
aggressive environments, wear resistance and sensing capabilities [11, 14–17].
The current chapter highlights opportunities and challenges for the use of
CNTs to make structural composites of high stiffness, strength and toughness. It
revises three strategies for CNT integration in composites: dispersion in the matrix,
localization at interfaces and as reinforcing fibres.
The main benefit of adding CNTs in the matrix of carbon fibre composites
is to improve composite toughness. Composites with tougher matrices exhibit
better resistance to delaminations. The effect of CNTs on the mode I and II
6 Nano-engineered Carbon Fibre-Reinforced Composites: Challenges. . . 121
Fig. 6.3 (a) Matrix cracks in transverse plies; (b) zoom in on a matrix crack revealing debonding
at the fibre/matrix interface in a carbon fibre/epoxy composite. The composite was loaded under
tension along the 0ı fibre direction [8–10]
interlaminar fracture toughness was summarized in Qian et al. [12] and Lubineau
and Rahaman [18]. The typical level of improvement is in the range of tens
of percent [19]. In exceptional cases it can reach a hundred level mark [20].
The interlaminar shear strength can be increased by about 30 %. The improved
performance is attributed to the CNT bridging effect [21]. Theoretically the pull-
out mechanism can be exploited to lead to an up to two orders of magnitude
better toughness in comparison with carbon fibres for uniformly dispersed and
unidirectionally aligned CNTs [22, 23]. The maximum pull-out energy is achieved
when CNT’s length is just below the critical length.
CNTs integrated in the matrix of fibre-reinforced composite have also been
proven to hinder formation and propagation of matrix cracks. It is known that high-
performance polymer composites, such as epoxy-based carbon fibre composites,
have early damage initiation. The first damage occurs at about 0.2–0.4 % of strain
(under tension). It is typically in the form of microcracks originating near or at
fibre/matrix interfaces and then propagating through the ply thickness (Fig. 6.3).
According to Yokozeki et al. [24, 25], the onset strain for the matrix cracks was
shifted from 0.43 to 0.53 % for laminates with CNTs. The crack density evaluated
at different strain levels was shown to decrease. Similar observations of crack
retardation were reported in other works and detected with different technique. They
were confirmed by acoustic emission measurements in De Greef et al. [26] and by
X-ray examination of samples tested till specific strain levels in De Greef et al. [27].
122 L. Gorbatikh and S.V. Lomov
of the design criterion in the aerospace applications. The CAI property of CNT-
modified laminates was evaluated in Yokozeki et al. [24, 25], Kostopoulos et al.
[36] and Siegfried et al. [37]. These studies were conducted with carbon fibre/epoxy
laminates and the CNTs were dispersed in the matrix. Different types of CNTs were
used: single wall, cup stacked and multiwall CNTs. It was noted that while the
interlaminar fracture toughness of a composite can be considerably improved by
adding CNTs in the matrix, the effect of these improvements on the CAI strength is
less pronounced. The improvements were in the order of few percent and not always
statistically significant.
carbon fibres is that the carbon fibres may have degraded tensile strength after the
CNT growth [43, 44]. The decrease in strength can be by as much as 50 %. The
reduction is due to the damage introduced to the outer layers of fibres by catalyst
particles. Similar results were obtained in Zhang et al. [45]. The resulting properties
of carbon fibres are sensitive to the type of carbon fibre, the presence of sizing and
growth conditions. This challenge has received significant attention in the literature
and has been, to a large extent, addressed by now. Many research groups are able to
grow CNTs on carbon fibres without degradation of the fibre performance [46–48].
Other challenges are associated with processing of these ‘fuzzy’ fibres and
fabrics into composites. The promising results in the mechanical performance trig-
gered interest in manufacturability of these composites including their permeability
[49], formability [50] and compressibility [51–55]. The compressive behaviour
of the reinforcement is important in such manufacturing techniques as vacuum
infusion, light RTM and autoclave composite manufacturing. In these techniques
compressibility of the fibre reinforcement controls the fibre volume fraction in the
composite. If manufacturing of the composite is not accounted for the change in
compressibility of the reinforcement, it may lead to a much lower fibre volume
fraction in the composite. Thus, conditions of composite manufacturing require
adjustments if fuzzy fibres are used. It was established that CNTs grown on carbon
fibres increase compressive resistance [53, 54]. Fuzzy fibres also affect the internal
geometry of fabrics (Fig. 6.4). According to Aravand et al. [56], it was found that
fuzzy yarns swelled due to the CNT grafting leading. This leads to the increase of
the yarn dimensions and, consequently, to the laminate thickness. These changes in
the internal structure are expected to have a significant influence on the progressive
damage development in such hierarchical composites.
Some additional considerations have to be taken into account when using fuzzy
fibres in polymer composites. According to Lachman et al. [57], CNT-grafted fibres
may behave as fibres with a larger diameter. The CNT fuzz creates an interphase
layer of high stiffness around a fibre [58]. Modelling studies indicate that stress
concentrations due to the mismatch in properties of the fibre and matrix can move
from the fibre interface to the end of the CNT forest [10]. Since the nano-modified
interphase is likely to have a higher strength, the fracture is expected at the end of
the CNT forest and not at the fibre interface. In Romanov et al. [59], it was also
shown that CNTs grown on fibres can fundamentally change stress distribution near
the fibre/matrix interface.
Deposition of synthesized CNTs on fibre surfaces via sizing or coating technolo-
gies is another promising technique for the introduction of CNTs in composites.
This route takes advantage of available techniques for CNT dispersion. It has
clear scale-up perspectives from laboratory to industrial scales, while CNT growth
on fibres is a complex process that at present is difficult to upscale. The nano-
modified sizings have been proven to strengthen the fibre/matrix interface [60, 61]
and improve other interface-dominated properties. The approach has a number of
advantages over other methods. Unlike CNTs that are directly synthesized on fibres,
the ones in fibre sizings and coatings do not damage the fibre surface. The strength
is therefore not degraded. Moreover, CNT-reinforced sizings have shown to ‘heal’
6 Nano-engineered Carbon Fibre-Reinforced Composites: Challenges. . . 125
Fig. 6.4 Micro-CT cross-sectional image of the laminates: ‘reference’ is the material without
CNTs; ‘short’ and ‘long’ contain CNTs with short (few microns) and long (10–20 µm) lengths.
Long P is produced under an additional pressure. Reproduced from Aravand et al. [56]
defects on the fibre surface. For some fibre types like glass, this healing effect leads
to even higher fibre strength [62, 63]. The interfacial adhesion strength was also
increased in CNT-modified E-glass/PP systems. According to Godara et al. [61], the
interface strength was characterized in glass fibre/epoxy composites where CNTs
were introduced in three different ways: (1) in the fibre sizing, (2) in the matrix
and (3) combination of the two approaches. The highest increase in the interface
strength was achieved for the case of CNT sizings. CNTs in the fibre coating create
opportunities to design a multifunctional fibre/polymer interphase. A high potential
of CNT sizings for such multifunctional properties was shown by Ma et al. [64] on
the example of glass fibre composites. Recent advances on the surface modification
routes for composites can be found in a review article by Karger-Kocsis et al. [65].
Another important interface in composites is the interlaminar interface. To
strengthen and toughen this interface, CNTs can be positioned in between laminate
layers. Different improvement levels were realized using the nano-modification
of the interlaminar interfaces [66, 67]. Joining laminate plies with aligned CNTs
showed an impressive increase of fracture toughness by 2–3 times [68]. Adding
CNTs to a thermoplastic (PEK-C) interleaf of 3 g/m2 areal density provided up
to 30 % increase of the laminate strength in the CAI test [69]. The approach with
126 L. Gorbatikh and S.V. Lomov
(matrix) and the mesh of an embedded region (individual CNT). The stiffness of
each CNT is introduced into the model by constraining degrees of freedom at
CNT nodes to matrix nodes. One of the advantages of the ER technique is that
the matrix mesh geometry is not affected by the CNTs. Therefore well-structured
meshes can be created, which is beneficial for post-processing of the data. The mesh
sensitivity study showed that results are dependent on the ratio of sizes of the host
and embedded region elements. The element size in the matrix region should be set
as a function of the number of CNTs being present there. This model was applied
to predict stresses in a unidirectional fibre composite with CNTs in a large variety
of configurations: CNTs that are dispersed in the matrix (with a different degree
of agglomeration, [8]), CNTs that are grown on fibres (with a different degree of
alignment, [10]) and CNTs that are concentrated in the interphase region mimicking
fibre sizings or coatings. Some of the configurations are shown in Fig. 6.1. In this
model, CNT length, waviness, alignment and positioning can all be varied. These
configurations reflect the reality of what can be practically realized these days in
nano-engineered composites (as discussed in previous sections). Additionally, there
are possibilities to model more sophisticated configurations with spatially resolved
CNT positioning that may not yet be practically feasible to make.
Predictions of the model allowed drawing a number of conclusions. In Romanov
et al. [8–10], the authors showed that CNTs in the polymer matrix of a unidirectional
carbon fibre composite have a drastic effect on the stress distribution at the micro-
scale. Depending on the morphology, CNT assemblies lead to either suppression or
magnification of stresses (Fig. 6.5). The strongest effect has been seen when CNTs
form dense assemblies such as CNT growth on fibre surfaces or agglomerates. CNTs
grouped in agglomerates were found to behave collectively as microscopic entities
of a high stiffness. They generate high-stress concentrators in the matrix and at
the fibre/matrix interface. The magnification of stresses depends on the density and
size of the CNT agglomerates as well as the degree of CNT agglomeration (not
agglomerated, partially agglomerated or fully agglomerated). CNT agglomerates of
higher densities lead to higher microscopic stresses. As for the degree of agglom-
eration, even partial CNT agglomeration leads to vast matrix stress concentrations.
Homogeneously dispersed CNTs introduce the least disturbance in the stress fields.
CNTs grown on fibre surfaces were found to reduce radial and shear stresses
at the fibre/matrix interface [10]. However, additional stress concentrations were
observed in the narrow interfibre spaces if CNT forests were short and did not
overlap. For long overlapping CNT growth, microscopic stresses were, on the
contrary, suppressed due to the forest interpenetration. The stress was, however,
magnified in surrounding areas. CNTs deposited in fibre coatings produced quite
different stresses. The overall effect of CNT-coated fibres on stresses was less
pronounced than for the grown CNTs. This was explained by the difference in the
orientation of CNTs towards the fibre surface that was tangent and not radial like in
the case of the growth.
The advantage of having modelling tools is that they allow simulating scenarios
that are not yet possible to realize in practice. For example, with the developed
model, the authors showed that CNTs can be used to mitigate stress concentrations
128 L. Gorbatikh and S.V. Lomov
Fig. 6.5 Contour plots of the maximum principal stress in the matrix and CNTs for composites
with four CNT configurations shown in Fig. 6.1 and the reference composite without CNTs [9]
in the same way nature uses nanostructures in biological composites [75]. For this,
spatially resolved CNT networks are required [9]. Thus an intelligent design with
CNTs promises interesting effects and to be explored in the future.
In the recent years, there has been a lot of interest in the development of CNT-
based fibres. As already mentioned in the beginning of this chapter, the main
problem for not meeting the theoretical expectations for the high stiffness and
strength in CNT nanocomposites is related to not being able to practically achieve
high CNT volume fractions and unidirectional alignment. Once these difficulties
are overcome, superior properties will follow. The challenges have been related to
intrinsic limitations of the dispersion route; therefore, new routes have been sought.
An alternative technology is to produce fibres and sheets with well-aligned and
highly packed CNTs and then to infiltrate them with a polymer to make a structural
nanocomposite. The key concept is to align CNTs as much as possible and to achieve
as high packing as possible.
There are several approaches that are currently explored for making such CNT-
based fibres: (1) spinning from a CNT solution, (2) spinning from an aligned CNT
forest, (3) spinning from a CNT aerogel and (4) twisting/rolling from a CNT film.
These techniques and corresponding fibre properties are reviewed in Lu et al. [76]
and Wu and Chou [77]. The range of the tensile strength and the Young’s modulus
achieved with these approaches is shown in Fig. 6.6. The strength of the CNT fibres
depends on properties of individual CNTs such as their length, the number of walls,
6 Nano-engineered Carbon Fibre-Reinforced Composites: Challenges. . . 129
Fig. 6.6 SEM images of CNT fibres produced by (a) drawing from aligned CNT forests
(Reproduced from Zhang et al. [78]) and (b) rolling of CNT sheets [79]. (c) The strength versus
Young’s modulus of CNT fibres produced using different approaches (Wu and Chou [77], including
the references)
diameter and waviness, but also on a fibre morphology—how CNTs are arranged
and compacted. The stiffness and strength of these CNT fibres are impressive
and comparable to those of conventional fibres. They are, however, 1–2 orders of
magnitude lower than the properties of individual CNTs, indicating that there is still
room for improvement.
CNT fibres can be spun into CNT yarns with a diameter of 0.1–0.3 mm [78, 80].
The yarns can be then processed into different preforms using conventional textile
methods as knitting, braiding and weaving [80–82]. The current challenges include
130 L. Gorbatikh and S.V. Lomov
difficulties with their infusion. CNT fibres have low permeability and it is not trivial
to achieve their complete wetting [83]. High-pressure processes like pultrusion can
provide good impregnation [84].
As identified by Lu et al. [76], another challenge is related to the development
of common standards for fibre characterization. Different approaches are used in
the literature to report the stress level, and therefore properties of such fibres cannot
be easily compared. The internal structure of these CNT fibres in relation to their
mechanical performance is not yet well understood. The same holds for the key
parameters in production. With composite application in mind, it is also necessary
to better understand the interfacial behaviour of the CNT fibre with the polymer.
This is not trivial as CNT fibre has a nonhomogeneous hierarchical structure and
the interface with the matrix is not defined. This may have an important effect on
the way the fibre transfers the load.
On the other hand, CNT fibres create opportunities for structural composites.
They promise composites with high stiffness, strength and also toughness as these
fibres can achieve higher strains to failure. Traditional structural fibres used in
composites like carbon and glass are brittle and therefore their composites are also
brittle. Composites based on CNT fibres are expected to fail in a gradual manner
and have a superior toughness in comparison with the current advanced composites.
CNT fibres also have high flexibility and resistance to torsion. They can be knotted
with fracture and their knot efficiency can be as high as 100 % [85]. In addition, the
production of CNT fibres is simpler than the one for traditional high-performance
fibres [76]. Significant research efforts are needed to realize the high potential of
these novel fibres.
Acknowledgements The authors would like to acknowledge the support of Skolkovo Institute of
Science and Technology in Russian Federation through the No. 335-MRA project linked to the
Center for Design, Manufacturing and Materials.
References
1. L.V. Radushkevich, V.M. Lukyanovich, On the carbon structure developed during thermal
decomposition of carbon monoxide on an iron contact (in Russian). Zurn Fisic Chim 26, 88–95
(1952)
2. S. Iijima, Helical microtubules of graphitic carbon. Nature 354(6348), 56–58 (1991)
3. M. Yu, B.S. Files, S. Arepalli, R.S. Ruoff, Tensile loading of ropes of single wall carbon
nanotubes and their mechanical properties. Phys. Rev. Lett. 84, 5552 (2000)
4. J. Cumings, A. Zettl, Low-friction nanoscale linear bearing realized from multiwall carbon
nanotubes. Science 289(5479), 602–604 (2000)
5. W. Cho, M. Schulz, V. Shanov, Growth and characterization of vertically aligned centimeter
long CNT arrays. Carbon 72, 264–273 (2014)
6. J.N. Coleman, U. Khan, W.J. Blau, Y.K. Gun’ko, Small but strong: a review of the mechanical
properties of carbon nanotube–polymer composites. Carbon 44, 1624–1652 (2006)
7. Z. Spitalsky, D. Tasis, K. Papagelis, C. Galiotis, Carbon nanotube-polymer composites:
chemistry, processing, mechanical and electrical properties. Prog. Polym. Sci. 35(3), 357–401
(2010)
8. V.S. Romanov, S.V. Lomov, I. Verpoest, L. Gorbatikh, Stress magnification due to carbon
nanotube agglomeration in composites. Compos. Struct. 133, 246–256 (2015)
9. V. Romanov, S.V. Lomov, I. Verpoest, L. Gorbatikh, Modelling evidence of stress concentration
mitigation at the micro-scale in polymer composites by the addition of carbon nanotubes.
Carbon 82, 184–194 (2015)
10. V.S. Romanov, S.V. Lomov, I. Verpoest, L. Gorbatikh, Inter-fibre stresses in composites with
carbon nanotube grafted and coated fibres. Compos. Sci. Technol. 114, 79–86 (2015)
11. T.W. Chou, L. Gao, E.T. Thostenson, Z. Zhang, J.H. Byun, An assessment of the science and
technology of carbon nanotube-based fibres and composites. Compos. Sci. Technol. 70, 1–19
(2010)
12. H. Qian, E.S. Greenhalgh, M.S.P. Shaffer, A. Bismarck, Carbon nanotube-based hierarchical
composites: a review. J. Mater. Chem. 20(23), 4751–4762 (2011)
13. E.T. Thostenson, Z.F. Ren, T.W. Chou, Advances in the science and technology of carbon
nanotubes and their composites: a review. Compos. Sci. Technol. 61(13), 1899–1912 (2001)
14. C.R. Carpenter, P.H. Shipway, Y. Zhu, Electrodeposition of nickel-carbon nanotube nanocom-
posite coatings for enhanced wear resistance. Wear 271(9–10), 2100–2105 (2011)
15. G. Pandey, M. Wolters, E.T. Thostenson, D. Heider, Localized functionally modified glass
fibres with carbon nanotube networks for crack sensing in composites using time domain
reflectometry. Carbon 50(10), 3816–3825 (2012)
16. C.X. Wu, H.B. Lu, Y.J. Liu, J.S. Leng, Study of carbon nanotubes/short carbon fibre
nanocomposites for lightning strike protection, in ed. by Z. Ounaies, J. Li. Behavior and
Mechanics of Multifunctional Materials and Composites (Spie-Int Soc Optical Engineering,
Bellingham, 2010)
17. N. Yamamoto, R. Guzman de Villoria, B.L. Wardle, Electrical and thermal property enhance-
ment of fibre-reinforced polymer laminate composites through controlled implementation of
multi-walled carbon nanotubes. Compos. Sci. Technol. 72(16), 2009–2015 (2012)
18. G. Lubineau, A. Rahaman, A review of strategies for improving the degradation properties of
laminated continuous-fibre/epoxy composites with carbon-based nanoreinforcements. Carbon
50, 2377–2395 (2012)
132 L. Gorbatikh and S.V. Lomov
19. A. Godara, L. Mezzo, F. Luizi, A. Warrier, S.V. Lomov, A.W. VanVurre, L. Gorbatikh,
P. Moldenaers, I. Verpoest, Influence of carbon nanotubes reinforcement on the processing
and the mechanical behaviour of carbon fibre/epoxy composites. Carbon 47(12), 2914–2923
(2009)
20. S. Tsantzalis, P. Karapappas, A. Vavouliotis, P. Tsotra, V. Kostopoulos, T. Tanimoto et al.,
On the improvement of toughness of CFRPs with resin doped with CNF and PZT particles.
Compos. A: Appl. Sci. Manuf. 38, 1159–1162 (2007)
21. F.H. Gojny, M.H.G. Wichmann, U. Kopke, B. Fiedler, K. Schulte, Carbon nanotube–reinforced
epoxy-composites—enhanced stiffness and fracture toughness at low nanotube contents.
Compos. Sci. Technol. 64, 2363–2371 (2004)
22. I. Greenfeld, H.D. Wagner, Nanocomposite toughness, strength and stiffness: role of filler
geometry. Nanocomposites 1, 3–17 (2015)
23. H.D. Wagner, P.M. Ajayan, K. Schulte, Nanocomposite toughness from a pull-out mechanism.
Compos. Sci. Technol. 83, 27–31 (2013)
24. T. Yokozeki, Y. Iwahori, S. Ishiwata, Matrix cracking behaviors in carbon fibre/epoxy
laminates filled with cup-stacked carbon nanotubes (CSCNTs). Compos. A: Appl. Sci. Manuf.
38(3), 917–924 (2007)
25. T. Yokozeki, Y. Iwahori, S. Ishiwata, K. Enomoto, Mechanical properties of CFRP laminates
manufactured from unidirectional prepregs using CSNT-dispersed epoxy. Compos. A: Appl.
Sci. Manuf. 38, 2121–2130 (2007)
26. N. De Greef, L. Gorbatikh, A. Godara, L. Mezzo, S.V. Lomov, I. Verpoest, The effect of carbon
nanotubes on the damage development in carbon fibre/epoxy composites. Carbon 49(14),
4650–4664 (2011)
27. N. De Greef, L. Gorbatikh, S.V. Lomov, I. Verpoest, Damage development in woven carbon
fibre/epoxy composites modified with carbon nanotubes under tension in the bias direction.
Compos. A: Appl. Sci. Manuf. 42, 1635–1644 (2011)
28. C.S. Grimmer, C.K.H. Dharan, High-cycle fatigue of hybrid carbon nanotube/glass
fibre/polymer composites. J. Mater. Sci. 43, 4487–4492 (2008)
29. C.S. Grimmer, C.K.H. Dharan, Enhancement of delamination fatigue resistance in carbon
nanotube reinforced glass fibre/polymer composites. Compos. Sci. Technol. 70, 901–908
(2010)
30. L. Böger, J. Sumfleth, H. Hedemann, K. Schulte, Improvement of fatigue life by incorporation
of nanoparticles in glass fibre reinforced epoxy. Compos. A: Appl. Sci. Manuf. 41(10),
1419–1424 (2010)
31. L. Gorbatikh, T. Li, N. De Greef, S.V. Lomov, I. Verpoest, Effect of carbon nanotubes on
fatigue life of carbon fibre/epoxy composites. 18th International Conference on Composite
Materials, Jeju Island, Korea, August 21–26, 2011
32. L.M. Gao, T.W. Chou, E.T. Thostenson, A. Godara, Z.G. Zhang, L. Mezzo, Highly conductive
polymer composites based on controlled agglomeration of carbon nanotubes. Carbon 48(9),
2649–2651 (2010)
33. Y.S. Song, J.R. Youn, Influence of dispersion states of carbon nanotubes on physical properties
of epoxy nano—composites. Carbon 43, 1378–1385 (2005)
34. M.A. Aravand, S.V. Lomov, I. Verpoest, L. Gorbatikh, Evolution of carbon nanotube dispersion
in preparation of epoxy-based composites: from a masterbatch to a nanocomposite. eXPRESS
Polym. Lett. 8(8), 596–608 (2014)
35. A. Haesch, T. Clarkson, J. Ivens, S.V. Lomov, I. Verpoest, L. Gorbatikh. Localization of
carbon nanotubes in resin rich zones of a woven composite linked to the dispersion state.
Nanocomposites 1, 206–213 (2016)
36. V. Kostopoulos, A. Baltopoulos, P. Karapappas, A. Vavouliotis, A. Paipetis, Impact and after-
impact properties of carbon fibre reinforced composites enhanced with multi-wall carbon
nanotubes. Compos. Sci. Technol. 70, 553–563 (2010)
37. M. Siegfried, C. Tola, M. Claes, S.V. Lomov, I. Verpoest, L. Gorbatikh, Impact and residual
after impact properties of carbon fibre/epoxy composites modified with carbon nanotubes.
Compos. Struct. 111, 488–496 (2014)
6 Nano-engineered Carbon Fibre-Reinforced Composites: Challenges. . . 133
38. N. Yamamoto, A.J. Hart, E.J. Garcia, S.S. Wicks, H.M. Duong, A.H. Slocum, B.L. Wardle,
High-yield growth and morphology control of aligned carbon nanotubes on ceramic fibres for
multifunctional enhancement of structural composites. Carbon 47(3), 551–560 (2009)
39. E.J. Garcia, A.J. Hart, B.L. Wardle, A.H. Slocum, Fabrication and nanocompression testing of
aligned carbon-nanotube-polymer nanocomposites. Adv. Mater. 19(16), 2151C (2007)
40. S.S. Wicks, R.G. de Villoria, B.L. Wardle, Interlaminar and intralaminar reinforcement of
composite laminates with aligned carbon nanotubes. Compos. Sci. Technol. 70(1), 20–28
(2010)
41. S.S. Wicks, W. Wang, M.R. Williams, B.L. Wardle, Multi-scale interlaminar fracture mecha-
nisms in woven composite laminates reinforced with aligned carbon nanotubes. Compos. Sci.
Technol. 100, 128–135 (2014)
42. E.T. Thostenson, W.Z. Li et al., Carbon nanotube/carbon fibre hybrid multiscale composites. J.
Appl. Phys. 91(9), 6034 (2002)
43. H. Qian, A. Bismarck, E.S. Greenhalgh, G. Kalinka, M.S.P. Shaffer, Hierarchical composites
reinforced with carbon nanotube grafted fibres: the potential assessed at the single fibre level.
Chem. Mater. 20(5), 1862–1869 (2008)
44. R.J. Sager, P.J. Klein, D.C. Lagoudas, Q. Zhang, J. Liu, L. Dai, J.W. Baur, Effect of carbon
nanotubes on the interfacial shear strength of T650 carbon fiber in an epoxy matrix. Compos.
Sci. Technol. 69, 898–904 (2009). DOI: http://dx.doi.org/10.1016/j.compscitech.2008.12.021
45. F.H. Zhang, R.G. Wang, X.D. He, C. Wang, L.N. Ren, Interfacial shearing strength and
reinforcing mechanisms of an epoxy composite reinforced using a carbon nanotube/carbon
fibre hybrid. J. Mater. Sci. 44(13), 3574–3577 (2009)
46. N. De Greef, L. Zhang, A. Magrez, L. Forro, J.-P. Locquet, I. Verpoest, J.W. Seo, Direct growth
of carbon nanotubes on carbon fibers: effect of the CVD parameters on the degradation of
mechanical properties of carbon fibers. Diam. Relat. Mater. 51, 39–48 (2015)
47. R. Li, N. Lachman, P. Florin, H.D. Wagner, B.L. Wardle, Hierarchical carbon nanotube carbon
fibre unidirectional composites with preserved tensile and interfacial properties. Compos. Sci.
Technol. 117, 139–145 (2015)
48. K. Naito, J.M. Yang et al., Tensile properties of carbon nanotubes grown on ultrahigh strength
polyacrylonitrile-based and ultrahigh modulus pitch-based carbon fibres. Appl. Phys. Lett.
92(23) (2008). Paper 231912
49. S.V. Lomov, L. Beyers, L. Gorbatikh, I. Verpoest, V. Koissin, Z. Kotanjac, M. Karahan,
Permeability and compressibility of CNT-CNF grafted textile reinforcement. In: 10th Interna-
tional Conference on Flow Processes in Composite Materials (FPCM-10), Ascona, Switzerland
(2010), p. CD edition
50. V. Koissin, Z. Kotanjac, S.V. Lomov, L. Gorbatikh, L. Warnet, R. Akkerman, Deformability
of a textile reinforcement modified with nanofibres, in ed. by C. Binetruy, F. Boussu,
Proceedings of the 10th International Conference on Textile Composites (TexComp-10) (2010),
pp. 181–186.
51. S.V. Lomov, L. Gorbatikh, M. Houlle, Z. Kotanjac, V. Koissin, K. Vallons, I. Verpoest,
Compression resistance and hysteresis of carbon fibre yarns with grown carbon nan-
otubes/nanofibres. Compos. Sci. Technol. 71, 1746–1753 (2011)
52. S.V. Lomov, L. Gorbatikh, Z. Kotanjac, V. Koissin, M. Houlle, O. Rochez, M. Karahan, L.
Mezzo, I. Verpoest, Compressibility of carbon woven fabric with carbon nanotubes grown on
the fibres. Compos. Sci. Technol. 71(3), 315–325 (2011)
53. S.V. Lomov, L. Gorbatikh, I. Verpoest, A model for the compression of a random assembly of
carbon nanotubes. Carbon 49, 2079–2091 (2011)
54. S.V. Lomov, L. Gorbatikh, I. Verpoest, Compression behaviour of a fibre bundle with grafted
carbon nanotubes. Carbon 49, 4458–4465 (2011)
55. S.V. Lomov, S. Wicks, L. Gorbatikh, I. Verpoest, B.L. Wardle, Compressibility of nanofibre-
grafted alumina fabric and yarns: aligned carbon nanotube forests. Compos. Sci. Technol. 90,
57–66.G (2014)
134 L. Gorbatikh and S.V. Lomov
56. M.A. Aravand, O. Shishkina, I. Straumit, A.H. Liotta, S.S. Wicks, B.L. Wardle, S.V. Lomov, L.
Gorbatikh, Effect of “fuzzy” fibre morphology on the internal geometry of textile composites
characterization by micro-computed tomography. 20th International Conference on Composite
Materials, Copenhagen, 19–24 July 2015
57. N. Lachman, B.J. Carey, D.P. Hashim, P.M. Ajayan, H.D. Wagner, Application of
continuously-monitored single fibre fragmentation tests to carbon nanotube/carbon microfibre
hybrid composites. Compos. Sci. Technol. 72(14), 1711–1717 (2012)
58. C.D. Wood, M.J. Palmeri, K.W. Putz, G. Ho, R. Barto, L.C. Brinson, Nanoscale structure and
local mechanical properties of fibre-reinforced composites containing MWCNT-grafted hybrid
glass fibres. Compos. Sci. Technol. 72(14), 1705–1710 (2012)
59. V. Romanov, S.V. Lomov, I. Verpoest, L. Gorbatikh, Can carbon nanotubes grown on fibers
fundamentally change stress distribution in a composite? Composites Part A 63, 32–34 (2014).
DOI: http://dx.doi.org/10.1016/j.compositesa.2014.03.021
60. P. Drescher, M. Thomas, J. Borris, U. Riedel, C. Arlt, Strengthening fibre/matrix interphase
by fibre surface modification and nanoparticle incorporation into the matrix. Compos. Sci.
Technol. 74, 60–66 (2013)
61. A. Godara, L. Gorbatikh, G. Kalinka, A. Warrier, O. Rochez, L. Mezzo, F. Luizi, A.W. van
Vuure, S.V. Lomov, I. Verpoest, Interfacial shear strength of a glass fibre/epoxy bonding in
composites modified with carbon nanotubes. Compos. Sci. Technol. 70, 1346–1352 (2010)
62. S.-L. Gao, E. Mäder, R. Plonka, Nanocomposite coatings for healing surface defects of glass
fibres and improving interfacial adhesion. Compos. Sci. Technol. 68(14), 2892–2901 (2008)
63. J.E. Zhang, R.C. Zhuang, J.W. Liu, E. Mader, G. Heinrich, S.L. Gao, Functional interphases
with multi-walled carbon nanotubes in glass fibre/epoxy composites. Carbon 48(8), 2273–2281
(2010)
64. P.-C. Ma, J.-W. Liu, S.-L. Gao, E. Mäder, Development of functional glass fibres with
nanocomposite coating: a comparative study. Compos. A: Appl. Sci. Manuf. 44, 16–22 (2013)
65. J. Karger-Kocsis, H. Mahmood, A. Pegoretti, Recent advances in fibre/matrix interphase
engineering for polymer composites. Prog. Mater. Sci. 73, 1–43 (2015)
66. V.P. Veedu, A.Y. Cao, X.S. Li, K.G. Ma, C. Soldano, S. Kar, P.M. Ajayan, M.N. Ghasemi-
Nejhad, Multifunctional composites using reinforced laminae with carbon-nanotube forests.
Nat. Mater. 5(6), 457–462 (2006)
67. H. Qian, E.S. Greenhalgh, M.S.P. Shaffer, A. Bismarck, Carbon nanotube-based hierarchical
composites: a review. J. Mater. Chem. 20, 4751–4762 (2010)
68. E.J. Garcia, B.L. Wardle, A.J. Hart, Joining prepreg composite interfaces with aligned carbon
nanotubes. Compos. A: Appl. Sci. Manuf. 39(6), 1065–1070 (2008)
69. X. Xu, Z. Zhou, Y. Hei, B. Zhang, J. Bao, X. Chen, Improving compression-after-impact per-
formance of carbon-fiber composites by CNTs/thermoplastic hybrid film interlayer. Compos.
Sci. Technol. 95, 75–81 (2014)
70. G. Chatzigeorgiou, Y. Efendiev, D.C. Lagoudas, Homogenization of aligned “fuzzy fibre”
composites. Int. J. Solids Struct. 48(19), 2668–2680 (2011)
71. S.I. Kundalwal, M.C. Ray, Effective properties of a novel composite reinforced with short
carbon fibres and radially aligned carbon nanotubes. Mech. Mater. 53, 47–60 (2012)
72. S.I. Kundalwal, M.C. Ray, Effect of carbon nanotube waviness on the elastic properties of the
fuzzy fibre reinforced composites. J. Appl. Mech. 80(2), 1–13 (2013)
73. X. Ren, J. Burton, G.D. Seidel, K. Lafdi, Computational multiscale modeling and character-
ization of piezoresistivity in fuzzy fibre reinforced polymer composites. Int. J. Solids Struct.
54, 121–134 (2015)
74. C.Y. Li, T.W. Chou, Multiscale modeling of carbon nanotube reinforced polymer composites.
J. Nanosci. Nanotechnol. 3(5), 423–430 (2003)
75. L. Gorbatikh, S.V. Lomov, I. Verpoest, Original mechanism of damage initiation revealed
through modeling of naturally occurring microstructures. J. Mech. Phys. Solids 58, 735–750
(2010)
76. W. Lu, M. Zu, J.H. Byun, B.S. Kim, T.W. Chou, State of the art of carbon nanotube fibres:
opportunities and challenges. Adv. Mater. 24, 1805–1833 (2012)
6 Nano-engineered Carbon Fibre-Reinforced Composites: Challenges. . . 135
77. A.S. Wu, T.W. Chou, Carbon nanotube fibres for advanced composites. Mater. Today. 15(7–8)
(2012)
78. M. Zhang, K.R. Atkinson, R.H. Baughman, Multifunctional carbon nanotube yarns by
downsizing an ancient technology. Science 306(5700), 1358–1361 (2004)
79. J.M. Feng, R. Wang, Y.L. Li, X.H. Zhong, L. Cui, Q.J. Guo, F. Hou, Carbon 48, 3817 (2010)
80. A. Bogdanovich, P. Bradford, D. Mungalov, S.L. Fang, M. Zhang, R.H. Baughman, S. Hudson,
Fabrication and mechanical characterization of carbon nanotube yarns, 3-D braids, and their
composites. SAMPE J. 43(1), 6–19 (2007)
81. P.D. Bradford, A.E. Bogdanovich, Carbon nanotube yarn and 3-D braid composites. Part I:
tensile testing and mechanical properties analysis. Compos. A: Appl. Sci. Manuf. 41, 230–237
(2010)
82. P.D. Bradford, A.E. Bogdanovich, Carbon nanotube yarn and 3-D braid composites. Part II:
dynamic mechanical analysis. Compos. A: Appl. Sci. Manuf. 41, 238–246 (2010)
83. J.J. Vilatela, R. Khare, A.H. Windle, The hierarchical structure and properties of multifunc-
tional carbon nanotube fibre composites. Carbon 50(3), 1227–1234 (2012)
84. Y. Shimamura, K. Oshima, K. Tohgo, T. Fujii, K. Shirasu, G. Yamamoto, T. Hashida, K.
Goto, T. Ogasawara, K. Naito, T. Nakano, Y. Inoue, Tensile mechanical properties of carbon
nanotube/epoxy composite fabricated by pultrusion of carbon nanotube spun yarn preform.
Compos. A: Appl. Sci. Manuf. 62, 32–38 (2014)
85. J.J. Vilatela, A.H. Windle, Adv. Mater. 22, 4959 (2010)
Chapter 7
A Nano-micro-macro-multiscale Model
for Progressive Failure Prediction in Advanced
Composites
7.1 Introduction
The bottleneck for inserting new and improved polymer matrix composites (PMC)
is the overwhelming amount of testing required to incorporate new material into
the Design Allowables Database. This extensive testing is a direct consequence of
the system designer’s need to address uncertainty and reduce risk, specifically, how
variations in the constituent materials, the processing, the manufacturing process,
and end-use scenarios propagate into performance and failure of a component. The
development of the Design Allowables Database requires evaluation of multiple
batches of composites with the associated construction of very large mechanical
and other physical property databases. Also, to quantify uncertainty in the structural
performance, (1) numerous subcomponents may have to be fabricated requiring
expensive tooling, and (2) one must then perform expensive tests on these elements
to determine their long-term performance. Moreover, additional time and cost are
incurred if there are any complications (complications are typical) such as tooling
rework, large error bands in mechanical properties, or processing complications.
In this context, Integrated Computational Materials Science and Engineer-
ing (ICMSE) provides a methodology to support materials development without
recourse to expensive trial-and-error approach to materials fabrication and char-
acterization. The motivation for this chapter is that design requirements and
knockdown factors are limiting our ability to exploit advanced composite and nanos-
tructured composite materials for system-level payoffs. The challenge in achieving
these goals stems from our current inability to accurately predict end-of-life
properties, damage tolerance, and durability of multifunctional PMCs. A multiscale
physics-based modeling approach, in which individual roles of constituent materials
and their interfaces are recognized, is necessary to accurately represent the material
anisotropy, the diffusion of permeant species (such as moisture and/or oxygen), to
the level of fidelity required for predicting the response of PMC aircraft structures.
In the proposed approach, nanoscale simulations will provide input data for the
micromechanical response, which, in turn, will provide input to a three-dimensional
ply-level analysis capable of modeling the spatial variability of material properties
and multi-functionality in the material and also account for the ply-level stresses
that could ultimately initiate failure. Details regarding the implementation of this
hierarchical mechanism-based approach are presented in this chapter. Emphasis will
be on the implementation and extension of these mechanism-based models within
the framework of hierarchical levels to represent the PMC behavior/properties
as a function of the damage/degradation state. It is envisioned that the proposed
physics-based multiscale modeling will provide the framework for a revolutionary
life-prediction tool for nanoparticle-modified multifunctional PMC which will lead
to the development of reliable design guidelines for aerospace structures subjected
to aggressive environments, such as at elevated temperatures.
It is now well established that the addition of a few weight percent of nanopar-
ticles can result in significant improvements in the mechanical properties of a
fiber-reinforced polymer composite. Figure 7.1 depicts the length scales involved in
a typical nanoparticle-reinforced polymer composite, spanning at least nine orders
of magnitude in length scale and even greater orders of magnitude in interaction
time scale. In recent years, numerous efforts have been directed toward modeling
nanocomposites in order to better understand the reasons behind the enhancement
in mechanical properties, with the addition of a few weight percent of nanofillers.
In studying molecular systems, however, a multiscale and multi-physics simulation
approach is considered computationally more viable since relying on a single time or
length scale can take a huge amount of computational resources and can potentially
lead to physically inaccurate results.
The two main techniques in multiscale modeling are the hierarchical and the
concurrent simulation schemes. In the hierarchical approach, the physical system is
studied in isolation to far-field stimuli, and the results are translated to a continuum
7 A Nano-micro-macro-multiscale Model for Progressive Failure Prediction. . . 139
response using curve fits and/or statistical averaging. Valavala et al. [1] used
the energy equivalence of continuum and atomistic models of polymer systems
to characterize the nonlinear stress–strain response of polymers (in particular,
polycarbonate and polyimide). Burchyachenko et al. [2] used the Eshelby and Mori-
Tanaka methods to determine the effective properties of nanocomposite materials.
Riddick et al. [3] used equivalent modeling of carbon nanotubes in polymers to
study the fracture toughness of PMCs with carbon nanotubes (CNTs) embedded in
them. Awasthi et al. [4] employed molecular dynamics (MD) to determine the force-
displacement curves between nanoscale inclusions (CNT) and the polymer system.
In general, it was observed that the strengthening behavior observed in nanocom-
posite materials due to nanosized inclusions can only be explained using theoretical
formalisms that appeal to fundamental physics of interactions at nanometer length
scales. True material behavior (e.g., failure) cannot be accurately simulated through
effective or “smeared” continuum analysis since these techniques, by definition, do
not explicitly take into account the existence of a bi-material interface as well as
the characteristic length scales defined by the atomic bond length. Consequently,
the “smeared” analyses fail to capture local details, such as favorable surface
energy interactions for both bonded and nonbonded atoms, which are responsible
for the stiffness and strength enhancements at the nanoscale. From a future design
standpoint, it is imperative that we are able to describe the material response from
damage initiation to failure progression under all loads and temperature conditions.
Advances in computational tools have allowed us to study small-scale phe-
nomena using atomic-scale classical techniques such as molecular dynamics (MD)
or Monte Carlo (MC) methods and macroscale responses using continuum level
simulations. In MD, the simulation is conducted in a deterministic, time-dependent
manner through numerical integration of Newton’s equations of motion. In MC,
simulations are not time dependent, but it explores the conformational phase space
of the material system by accepting or rejecting proposed configurations based
on a specified algorithm [5]. The equivalence between these two approaches is
established through the ergodic principle [6]. However, the key challenge in both
hierarchical and concurrent modeling is to simultaneously process all information
available from both length scales and efficiently pass this information continually
across time as well as length scales. Concurrent simulations have the unique
advantage that it can tie local events at the nanoscale to stimuli from a larger
time and length scale during a real-time simulation. One of the earliest forays
into concurrent coupling was by Abrahams et al. [7] in their scheme, MAAD. The
scheme involved simultaneous coupling of tight binding (TB), molecular dynamics
(MD), and finite element analysis (FEA). The scheme was successfully applied to
study fracture behavior of silicon material systems. However, the method had some
shortcomings in that it was limited by the time step size of the tight-binding step (on
the order of 10–15 s) which made the procedure computationally expensive. Ogata
et al. [8] used FEA to concurrently couple with density function theory (DFT) to
study the surface oxidation of a silicon system. Some of the more recent efforts in
multiscale concurrent simulations have been the bridging-scale technique [9, 10]
and the concurrent MD-MPM (material point method) coupling technique proposed
140 S. Roy et al.
by Ma et al. [11]. In Ma et al. [11], the atoms within the handshake region of the
MD domain are directly coupled point-wise to MPM particles to simulate fracture
in a crystalline material. The continuum particles are refined down to atomic sizes
and the boundary material points are assumed to overlap with boundary atoms
of the MD zone at all times. The main drawback with this approach is that the
thermal vibrations in the MD domain induce spurious high-frequency vibrations in
the continuum domain. Recently, Saether et al. [12] at NASA Langley proposed an
embedded statistical coupling method to overcome the high-frequency vibrations in
a concurrent coupling scheme to study grain boundary de-cohesion in aluminum.
However, the application of ESCM directly to simulation of polymeric materials
results in the appearance of “ghost forces” at the coupling boundary that causes the
simulation to become unstable [13].
In summary, while there has been a large body of research conducted in atomistic
simulation of crystalline metallic materials, a large technology gap exists in the
multiscale simulation of amorphous polymeric materials and their composites. This
chapter is a description of the methodology for fundamentally addressing failure
in a polymer composite using the hierarchical multiscale modeling approach. The
nanoscale would be addressed first, to highlight elastic properties and incorporation
of damage using both strength- and fracture-based criterions and problems associ-
ated when modeling a polymer system. The nanoscale is followed by a discussion of
microscale modeling which communicates with the nanoscale using homogenized
properties, which in turn leads to a discussion on how the micro- and macroscales
interact to provide global stiffness and strength estimates of an actual carbon fiber-
reinforced thermoset composite structure.
Figure 7.2 depicts a schematic of this hierarchical multiscale modeling approach
for a block of cross-linked epoxy polymer composite containing a macroscale
crack. The figure illustrates a three-tier nano-micro-macro coupling, with the
goal of predicting progressive failure in a composite structure (e.g., a composite
aircraft wing). In this example, the FEA integration point at a “hot spot” in the
composite wing passes stress and strain information down to the microlevel for
micromechanics-based analysis, and the microlevel gets material properties data
from the nanoscale. The micromechanics model of fiber and matrix as depicted
in Fig. 7.2 uses this information as boundary condition to compute the stress
amplification due to the presence of the fiber in a matrix and uses that information to
predict failure in the fiber, in the matrix, or at the interface. The generalized method
of cells [14] (described later in this chapter) or some similar micromechanics
modeling methodology is used to incorporate micromechanical effects in the failure
analysis. The matrix properties in the microscale analysis, with and without the
presence of nanoparticles, are determined through the use of MD modeling at the
nanoscale. The nonlinear material stiffness properties for the polymer matrix are
obtained using nanoscale-informed damage mechanics (NIDM) described later in
this chapter. For the case where failure is governed by a dominant crack rather
than damage, nanographene platelets are inserted in the MD model ahead of a
crack tip in the polymer, adjacent to the carbon fiber as shown in Fig. 7.2. The
resulting changes in the critical value of atomistic J-integral (JC ), with and without
7 A Nano-micro-macro-multiscale Model for Progressive Failure Prediction. . . 141
the nanoparticle, are evaluated. This novel methodology allows for the influence of
nanographene loading (wt%), dispersion, and orientation to the mechanical loading
direction to be studied at the nanoscale in the presence of a carbon fiber, and the
results are transmitted to the microscale. The entire multiscale modeling procedure
must be carried out iteratively to allow for damage evolution and resulting changes
in mechanical load path due to “failed” elements at the macroscale. The type of
analyses used for each length scale is discussed in detail in the next sections.
The nanoscale interaction between polymer molecules and nanoparticle is a key fac-
tor in determining the macroscale strength of advanced composites. In recent years
numerous efforts have been directed toward modeling nanocomposites in order to
better understand the fundamental mechanism behind the dramatic enhancement of
fracture properties with the slight addition (<1 wt%) of nanoparticles (Fig. 7.3).
In order to better understand the local influence of nanoparticle on the fracture
properties of the polymer, a nanoscale analysis is imperative. Open-source MD
software, large-scale atomic/molecular massively parallel simulator (LAMMPS), is
a computational tool frequently used for simulation of nanoscale interactions. In
molecular dynamics, the atomistic structure at initial time is input to the algorithm,
with the required force fields chosen to describe the various atomic interactions.
A typical MD simulation is illustrated in the flowchart shown in Fig. 7.4.
Due to the complicated nature of the potential functions, the equations of
motion cannot be solved analytically. Numerical integration is used to determine
the updated positions and velocities of individual atoms in an MD system. Various
142 S. Roy et al.
Fig. 7.3 Experimental observations for (a) mode I fracture using CT test on a polymer nanocom-
posite (PNC) [15]. (b) Mode I interlaminar fracture using DCB testing on a unidirectional
laminate. (c) Mixed mode fracture using four-point bend testing on a PNC with respective wt%
of nanographene loading
interaction exists between all the particles in the system, accounting for all those
interactions could sometimes act as a bottleneck during MD simulations. Conse-
quently, a cutoff radius is often employed to reduce computation time. Because of
the predefined bonding arrangements, standard “harmonic” potential functions (e.g.,
OPLS) cannot model the chemical bond formation and breakage that occurs during
the simulation. In order to overcome this shortcoming of the empirical potential
functions, there exists another class of potential functions which are based on bond
order [17]. In this chapter, two types of potential functions, an empirical harmonic
potential (OPLS) and a bond order-based potential (ReaxFF) are employed to
highlight the difference in failure modeling using a nanographene platelet as a
benchmark study.
Fig. 7.5 (a) Cross-linked EPON 862 system. (b) EPON 862 modified with a nanographene
platelet
This approach assumes that the damage evolution is primarily due to separation
at the interface between nanoparticle and the polymer through nonbonded interac-
tions at the nanoscale; hence the moniker nanoscale-informed damage mechanics
(NIDM) is chosen for this approach. The model attempts to capture the stiffness
and strength enhancements due to nanoscale reinforcement by tracking the change
in the Helmholtz free energy over baseline as measured through the framework
of volume-averaged atomistic stresses. The proposed NIDM model correlates the
dissipation of internal energy of the system to the deterioration of interface and
bulk properties of the nanoparticle-reinforced polymer composite (see Fig. 7.6).
146 S. Roy et al.
Fig. 7.6 Development of the equivalent damage model (a) the nanoscale polymer nanocomposite,
(b) the nanoscale RVE with equivalent properties (c) interface debond and void coalescence in the
actual polymer nanocomposite, (d) damage development in the smeared model
As depicted in Fig. 7.6a, the NIDM model constructs an equivalent model with a
“smeared” representation of the nanographene–polymer system (Fig. 7.6b). As the
“smeared” model deforms, a critical value of strain is reached at which damage
initiates in the nanoscale RVE (Fig. 7.6d). Figure 7.6c depicts the actual damage
evolution at the nanoscale due to debonding and micro-void formation.
In the NIDM model, the internal state variable (ISV) approach proposed by
Talreja [21] is extended to the case of damage in a nanoscale RVE, and a three-
dimensional NIDM model is proposed for a neat thermoset polymer resin and
for a polymer reinforced with nanoparticles. The primary mode of damage is
characterized by (a) interface debond growth (opening) and (b) interfacial slip
(shear) for nanographene-reinforced polymer resin and (c) void growth in case of
the neat polymer resin. The failure in the damage model is considered to occur
when it is subjected to deformation beyond a critical strain limit, obtained from the
nonbonded Lennard-Jones potential at the atomistic scale. The constitutive law for
neat polymer and polymer nanocomposite incorporating damage at the nanoscale
can be written in the general matrix form as
f¢g D C0 C CD f©g (7.1)
The total stiffness matrix [C] of the polymer and the PNC is written as the
sum of two symmetric 6 6 matrices (i.e., [C] D [C0 ] C [CD ]) which are (a) the
elastic stiffness matrix denoted by [C0 ] and (b) the inelastic stiffness matrix, [CD ],
reflecting the change in elastic stiffness brought about by the initiation and evolution
7 A Nano-micro-macro-multiscale Model for Progressive Failure Prediction. . . 147
of damage. The solid lines in Fig. 7.7a show the NIDM estimates for the normal
(opening) stress as a function of applied strain for the example cases depicted in
Fig. 7.5, containing 0 wt% (baseline), 2:1 aspect ratio nanographene, and 4:1 aspect
ratio nanographene, respectively. It can be observed that there is good correlation
between the nonlinear constitutive law from the NIDM model and the virial stress–
strain curves obtained from MD in spite of the highly oscillatory nature of the
latter. Similarly, the solid lines in Fig. 7.7b show the NIDM estimates for the shear
stress as a function of applied strain for the example cases depicted in Fig. 7.5,
containing 0 wt% (baseline), 2:1 aspect ratio nanographene platelet, and 4:1 aspect
ratio nanographene platelet, respectively. It is evident that the NIDM model is able
to capture the increase in elastic stiffness and strength beyond elastic limit due to
nonbonded interactions at nanoscale between the graphene sheet and the polymer
molecules. Further, the influence of nanographene platelet aspect ratio on stress–
strain behavior is also captured in these plots. The vertical dashed line in Fig. 7.7a, b
represents the predicted critical strain at which the onset of damage occurs and
is fairly well corroborated by results from the MD simulations. Additional details
regarding the NIDM approach can be found in a recent paper by the authors [22].
For polymers the mechanism of deformation is strain rate controlled and the rate
of loading plays a key role in determining the mechanical response. In this context,
one of the drawbacks of MD simulations is that they involve extremely high strain
rates of the order of 1010 s1 due to the extremely small time step used because of
computational accuracy as opposed to real-world experimental strain rates which
occur at about 101 s1 . Therefore, when comparing the mechanical property values
(modulus, strength) from MD simulations to the ones obtained experimentally, it
becomes necessary to take into account the strain rate effect in the polymer using a
power-law [23].
In the event that failure is governed by the presence of a dominant discrete crack
rather than diffuse damage, the multiscale model has to accommodate fracture
mechanics at the nanoscale. Two approaches can be used while developing a coupled
fracture-based damage model:
There is a wide range of material parameters that contribute to the fracture toughness
of a material, that is, the capability of a material to resist crack initiation and growth.
Some of these parameters are only applicable to materials with relatively linear
mechanical behavior (e.g., stress intensity factor, energy release rate), and others
are difficult to measure experimentally (e.g., crack-tip opening displacement). One
fracture parameter that is relatively easy to measure experimentally and to calculate
computationally and can be applied to materials with and without significant
148 S. Roy et al.
Fig. 7.7 (a) Extension along Y-axis. (b) Shear in the X-Y plane
nonlinear behavior is the J-integral. The J-integral is defined as the amount of energy
necessary to extend a crack by an incremental length. It can be determined in test
specimens using the load–displacement response and in finite element models using
a contour integral surrounding the crack tip. In the case of a linear material, the
J-integral is the same as the energy release rate (G) and can be easily correlated
to the stress intensity factor (K). Because of these conveniences, the J-integral has
enjoyed great success as a fracture toughness parameter. Therefore, it is desirable to
apply the J-integral to polymer nanocomposite materials.
7 A Nano-micro-macro-multiscale Model for Progressive Failure Prediction. . . 149
In Eq. (7.2), S is the Eshelby stress tensor, ‰ is the Helmholtz free energy density,
H is the displacement gradient tensor, P is the first Piola-Kirchhoff stress tensor,
N is the outward normal to the surface @ along which contour the J-integral is
being evaluated, and JT , JU , and J are the total, energetic, and entropic components
of the J-integral, respectively. Here it is assumed that the ensemble average hi is
approximated by the time average of the quantity over a sufficiently long period of
time through the ergodic principle [6].
The critical value of JT at crack initiation (JTC ) is related to the fracture toughness
of the material. Therefore, the J-integral could be used as a suitable metric for
estimating the crack driving force as well as the fracture toughness of the material
as the crack begins to initiate. However, for the conventional macroscale definition
of the J-integral to be valid at the nanoscale (in terms of the continuum stress
and displacement fields and their spatial derivatives) requires the construction of
local continuum fields from discrete atomistic data and the use of these data in
the conventional contour integral expression for J, as given by Eq. (7.2) [24, 25].
One such methodology was proposed by Hardy [26] that allows for the local
averaging necessary to obtain the definition of free energy, deformation gradient,
and Piola-Kirchhoff stress as fields (and divergence of fields) and not just as total
system averages. In this manner, using the Hardy localization methodology at the
nanoscale, the continuum definition of the J-integral can be modified for atomistic
(non-continuum) applications, as validated for a metallic material in [25] and for a
nanographene platelet (Fig. 7.8), to be discussed in the next section.
It is well known from linear elastic fracture mechanics p (LEFM) that the mode
I (opening mode) stress intensity factor KI D ˇ a where is the far-field
applied stress, a is the half crack length for a center-cracked plate, and ˇ is the finite-
geometry correction factor. It can also be shown from LEFM that the J-integral is
related to the stress intensity factor given by JI D KI2 =E, where E is the Young’s
modulus of the material and plane-stress conditions are valid [25].
From thermodynamics and statistical mechanics, the Helmholtz free energy
density is given by ‰ D U T D kBVT LogZ, where U is the internal energy
density, T is the temperature, is the entropy density, V is the volume of the
ensemble, and Z is the partition function of the atoms occupying the region .
Note that our definition of free energy density (‰) available for crack propagation
is less than the internal energy (U) at sufficiently high temperatures. Ignoring the
entropic term (T) in the free energy density at elevated temperatures can result in
significant overestimation (>10 %) in atomistic J computation as discussed in Jones
150 S. Roy et al.
Fig. 7.8 Atomistic J-integral computation in mode I nanoscale crack propagation in a single
graphene platelet. (a) A graphene sheet with nanoscale center crack and contour for J-integral
computation. (b) J-integral values at T D 300 K for graphene sheet for different applied stress
intensity factor [24]
et al. [25] and corroborated by our own findings (Fig. 7.8b). Physically, this entropy
increase is manifested in thermal expansion within the bulk material, and therefore
this energy is no longer available for crack growth under isothermal conditions.
As a demonstration of the atomistic J-integral approach in our preliminary work,
we simulated mode I (opening mode) nanoscale crack propagation at T D 300 K in
a 10.5 nm-long, 10 nm-wide single-nanographene platelet with a 2.2 nm central
crack, as depicted in Fig. 7.8a and described in detail in Roy et al. [27]. In
7 A Nano-micro-macro-multiscale Model for Progressive Failure Prediction. . . 151
this example, the graphene platelet is modeled using the ReaxFF potential with
Chenoweth parameters [17] available in the open-source molecular dynamics (MD)
software, LAMMPS. The graphene platelet was subjected to a far-field uniaxial
strain in a direction transverse to the crack, and the J-integral was computed on
three concentric contours as depicted in Fig. 7.8a. In order to establish proof of
concept for atomistic J-integral computation, the MD simulations were carried out
isothermally at T D 300 K. The coarse time step for the problem ( t) was set
equal to 4 fs. Figure 7.8a shows the MD simulation of the initiation of a nanoscale
mode I crack in the graphene platelet. The atomistic J-integral results are plotted in
Fig. 7.8b as a function of applied stress intensity factor and compared with linear
elastic fracture mechanics (LEFM) predictions. Not surprisingly, the quadratic
dependence of the energetic portion of the atomistic J-integral (JU D KI2 =E)
on stress intensity factor is in good agreement with strain energy-based LEFM
predictions (JLEFM ) at 300 K. Also, the deviation of the MD-computed atomistic
J value from LEFM prediction due to nonlinear (bond-breaking) effects at the crack
tip beyond KI D 5 MPa.m1/2 can be clearly observed in the figure (refer to the red
curve). Further, we computed the total value of J-integral (JT ) for the same case
including entropic effect as shown by the green curve, using a local harmonic (LH)
approximation to the Cauchy–Born model as described in [25, 27]. As can be seen
from Fig. 7.8b, there is a significant reduction (50 %) in the computed value of
J if entropic effects are included. It is also significant that the deviation of the MD-
computed atomistic J value from LEFM prediction occurs almost exactly when the
bond order of the ReaxFF force field between carbon atoms at the crack tip goes to
zero (refer to the pink line). Our predicted critical value of JT for nanographene at
300 K (18 J/m2 ) is in good agreement in comparison with the experimental value
of JIC D15.9 J/m2 reported in [28] and indicated by the dashed line in Fig. 7.8b.
As discussed in [25], due to its use of LH approximation, this methodology is
only applicable to a defect-free crystalline material, such as graphene, and hence
cannot be applied to amorphous epoxy polymers. However, the path independence
of J-integral facilitates the use of a cohesive-contour-based technique to compute
J-integral [29]. For this purpose, a J-integral contour in the shape of a cohesive strip
encompassing the thin cohesive-contour ahead of the crack tip is constructed as
illustrated in Fig. 7.9.
Because of the way the contour is designed, it can be shown from Eq. (7.2) that
by making the contour height h sufficiently small, the Helmholtz free energy term
can be eliminated from the J-integral equation for mode I as well as for mixed mode
loading. To better understand this approach, let us consider the scalar (indicial) form
of the expression for total J-integral:
Z
.JT /K D h‰ıKJ PiJ HiK iNJ d (7.3)
I, and J D 2 along the narrow cohesive-contour in Fig. 7.9, the first term on the
RHS of Eq. (7.3) vanishes, because ı12 D 0 by definition. Therefore, the only
terms that need to be evaluated along the narrow cohesive-contour are the Cartesian
components of the first Piola-Kirchhoff stress tensor (Pi2 ) and components of the
displacement gradient tensor (Hi1 ), giving
Z
.JT /1 D hPi2 Hi1 iN2 d (7.4)
The blue curve in Fig. 7.8b for the center-cracked graphene sheet indicates
that the cohesive strip approach agrees very well with the exact solution for total
J-integral (JT ) given by the green curve, thereby validating the cohesive strip
approach for graphene sheet.
Concurrent simulations have the unique advantage that it can tie local events at
the nanoscale to stimuli from a larger time and length scale during a real-time
simulation. Concurrent coupling is particularly useful for simulating events in large-
scale structures, such as crack growth in a test specimen that is too large to model
using MD alone. Most of the proposed multiscale concurrent coupling methods in
literature have been restricted to modeling crystalline materials, such as metals.
When applied to amorphous materials like polymers which do not possess an
ordered crystalline structure, coupling of these randomly oriented molecular chains
to a continuum domain can pose serious challenges.
In our example of concurrent coupling, the continuum system is modeled
using the generalized interpolation material point method (GIMPM), which is
a particle-based Euler–Lagrangian approach that is more suitable for modeling
large deformation problems compared with finite element analysis (FEA). In the
GIMP method, smooth particle characteristic function, which defines contiguous
particles, can be implemented to overcome the numerical artifact noise problem of
7 A Nano-micro-macro-multiscale Model for Progressive Failure Prediction. . . 153
standard MPM, which occurs when material points are crossing computational grid
boundaries and results in impractical stress states and erroneous evolution of history
variables. GIMPM was developed by Bardenhagen and Kober [30], where they
generalized the traditional MPM interpolation functions using a variational form and
a Petrov–Galerkin discretization scheme, thereby mitigating problems due to cell-
crossing by material points. For fracture problems, dynamic strain energy release
rates (SERR) is calculated using virtual crack closure technique (VCCT) [31] and
the dynamic (i.e., with inertial effects included) version of the J-integral [32].
Similar to the original MPM, the continuum body is discretized into a finite
number of material points in the GIMP method, as shown in Fig. 7.10. Each material
point is initialized with original material properties, such as density and volume,
and state variables, such as position, acceleration, velocity, strain, and stress. For
each time step, information on Lagrangian material points are mapped onto the
Eulerian grid nodes of the cell containing the particles as shown in Fig. 7.10.
Smooth particle characteristic functions, which defines contiguous particles, can
be implemented to overcome the numerical artifact noise problem of MPM, which
occurs when material points are crossing computational grid boundaries and results
in impractical stress states and erroneous evolution of history variables. In the
GIMPM methodology, the nodal property (e.g., velocity) at the ith Eulerian node
is given by
X
t
i D p .xi / t
p (7.5)
p
where ti is the nodal property, tp is the material point property, and p is the particle
shape function associated with the ith node. Similarly, the continuum fields (e.g.,
154 S. Roy et al.
displacement) can also be represented using grid node information, which can be
described as
X
g .x/ D Si .x/ gi (7.6)
i
where g(x) is the continuum variable, Si is the grid shape function at ith node, and
gi is the nodal variable, analogous to FEA. The variational form for conservation of
momentum is derived in vector form as
Z Z Z Z
a wdx C ¢ W rwdx D b wdx C £ wdS (7.7)
˝ ˝ ˝ @˝
where is the mass density, a is acceleration vector, b is the body force vector, £
is the traction on the boundary, ¢ is the Cauchy stress, and w is the test function.
In GIMPM, for simplicity, the characteristic shape function for the pth particle is
commonly chosen as
p .xi / D H xi xip lip C H xi xip C lip i D 1; 2; 3
(7.8)
p .x/ D p .x1 / p .x2 / p .x3 /
where H(x) is the unit step function and xp is the particle centroid position, while
lp is the half length of the particle. It should be noted here that the GIMP approach
assumes a finite particle size, whereas the original MPM scheme assumed point
particles. By applying Eqs. (7.5) and (7.6) into Eq. (7.7), the discrete form of
momentum equation can be obtained as
:
pi D fint
i C fi C fi
b
(7.9)
where
: X : X
pi D Sip pp Ii D
fint ¢ p rSip Vp
p
X Z
p
(7.10)
fbi D mp brSip I fi D £Si dS
p @
in which
Z xp Clp
1
Sip D 2lp Si .x/dx
Zxp l p
xp Clp (7.11)
rSip D 2l1p rSi .x/dx
xp lp
After the momentum vector pi at the ith grid node at each time step is updated
using Eq. (7.9) in a manner similar to explicit FEA methodology, mapping operation
7 A Nano-micro-macro-multiscale Model for Progressive Failure Prediction. . . 155
from the grid nodes to material points are performed, so that the position of the
particles are updated for the next time step. The grid nodes are then reset to their
original state at the end of the step, and hence, by updating the position of the
moving material points and resetting the nodes to their original position, large
deformation can be simulated without mesh distortion or the need for remeshing.
For concurrent coupling between atomistic and continuum domains, a GIMPM-
MD concurrent coupling scheme is used with “anchor points” [33] acting as
transition zone to transfer information between continuum and molecular systems
via a “handshake” region as shown in Fig. 7.11. The anchor points are added to the
original MD model as extra carbon atoms as schematically depicted by the red atoms
in Fig. 7.11. The anchor point approach is implemented in the proposed scheme as
shown in Fig. 7.12. An appropriate number of anchor points (red points) are created
inside the overlapped domain of MD and GIMPM, keeping in mind not to introduce
too many “artificial” anchor atoms. Each anchor point is connected to an atom on
the polymer chains using a harmonic potential that provides a preassigned stiffness
to the anchor spring. At the end of “M” MD time steps, reaction force from the
MD domain are calculated at the GIMPM cell nodes in the transition region and
interpolated onto corresponding material points within the respective cell. Taking
this force as force boundary condition for GIMPM, nodal displacements in the
continuum domain are solved. At the end of the GIMPM time step, the computed
material point displacements are interpolated and averaged onto corresponding MD
anchor points within the cell. The process is repeated for each coupled GIMPM-MD
time step to carry out the complete crack growth simulation in a polymer system.
Two cases of fracture simulation using the proposed GIMP-MD coupling
approach were performed to verify the method in crystalline material and for
amorphous materials like polymers: (A) mode I fracture simulation in a single
sheet of crystalline graphene with an edge crack of length 5 nm and (B) mode I
fracture simulation of EPON 862 polymer system with an edge crack of length
156 S. Roy et al.
Fig. 7.13 Simulation setup (a) Case A, graphene (b) Case B, EPON 862 polymer
system (Case B). Figure 7.16 shows the crack propagation in EPON 862 system. A
diffused damage rather than clear crack propagation is observed in this case because
the crack length of 5 nm is well below the threshold crack length of 2560 nm for
EPON 862 under pure mode I loading, resulting in a failure mode switch from brittle
fracture to ductile damage. This is because the theoretical strength of the polymer
is exceeded for any crack length below the critical crack length, thereby causing
the transition in failure mode [34]. In order to accurately model crack propagation
in this polymer, a much larger coupled system greater than 2560 nm will need to
158 S. Roy et al.
Fig. 7.15 Traction-separation plot for graphene with center crack under tension (Case A)
Fig. 7.17 (a) Graphene sheet with a 0.3 nm edge crack. (b) Diffused damage under tensile loading
7 A Nano-micro-macro-multiscale Model for Progressive Failure Prediction. . . 161
Fig. 7.18 MAC/GMC RUCs (a) Doubly periodic (b) triply periodic
Such properties are used to feed into successively higher length scale RUCs
containing microscale features (i.e., continuous fibers, effective matrix, fiber/matrix
interface, as shown schematically in Fig. 7.2) and/or mesoscale structures (individ-
ual or woven fiber tows, effective matrix, etc.). Figure 7.18 shows two of the many
choices of RUCs available in MAC/GMC to model the microscale subcells, e.g.,
doubly periodic (2-D) or triply periodic (3-D).
Homogenized ply-level properties obtained from the highest length scale RUC
are then used to establish finite element stiffness properties for use in global
structural models. Analogously, finite element integration point stresses obtained
from a global structural analysis are mapped onto the highest length scale RUC
for use in progressive failure analysis using MAC/GMC. Local field quantities
obtained within a given RUC are used to establish boundary conditions for use in
progressive failure analysis at the next lower length scale RUC. This multiscale
modeling methodology involves a robust local-to-global iteration strategy, where a
hierarchy of RUCs may be used to explicitly model composite material structure,
morphology, and failure over a range of length scales. The effect of local damage or
failure at a given length scale is automatically propagated at each load or time step
to the global finite element scale through the change in homogenized microscale
properties.
In GMC for continuous (or discontinuous) fibrous composites, the repeating
unit cell can consist of an arbitrary number of phases. Hence GMC is capable
of modeling a multiphase composite. This generalization extends the modeling
capability of the original method of cells to include the following: (1) inelastic
thermomechanical response of multiphased metal matrix composite, (2) modeling
of various fiber architectures (including both shape and packing arrangements), (3)
modeling of porosities and damage, and (4) the modeling of interfacial regions
around inclusions including interfacial degradation [38].
The basic homogenization approach employed in the micromechanical analysis
consists essentially of four steps. First, the repeating volume element (RVE) of
the periodic composite structure is identified. Here, it should be recalled that the
162 S. Roy et al.
RVE represents the smallest section of the composite structure that captures the
full physics of the problem (e.g., packing, fiber-fiber interactions, etc.), whereas
the RUC is the smallest repeating unit cell that is derived purely from the
geometry and/or symmetry of the structure. Second, the macroscopic or average
stress and strain state in terms of the individual microscopic (subcell) stress
and strain states is defined. Third, the continuity of tractions and displacements
are imposed at the boundaries between the constituents. These three steps, in
conjunction with microscale equilibrium, establish the relationship between micro
(subcell) total, thermal, and inelastic strains and macro (composite) strains via the
relevant concentration tensors. In the fourth and final step, the homogenized macro-
constitutive equations of the composite RVE are determined. These four steps form
the basis of the micro-to-macro mechanics analysis which describes the behavior of
heterogeneous media.
The resulting micromechanical analysis establishes the overall (macro) behavior
of the multiphase composite and is expressed as a constitutive relation between the
volume-averaged stress ( ), kinematic strain (" ), thermal strain (" T ), and inelastic
strains (" I ), in conjunction with the effective elastic stiffness tensor, B* [38]. Hence,
D B " " I " T (7.12)
1 X ˛ X̌ X
N N N
B D d˛ hˇ l C .˛ˇ / A.˛ˇ / (7.13)
dhl ˛D1
D1
ˇD1
where d˛ , hˇ , l are the subcell depth, height, and length, respectively. The volume-
averaged composite inelastic strain vector is defined as:
B 1 X
N˛ X̌
N
XN
" D I
d˛ hˇ l C .˛ˇ /
D.˛ˇ / " I " I.˛ˇ / (7.14)
dhl ˛D1 s
D1 ˇD1
B 1 X
N˛ X̌
N
XN
"T D d˛ hˇ l C.˛ˇ / D.˛ˇ / " T " T.˛ˇ / (7.15)
dhl ˛D1 s
D1
ˇD1
where " is the uniformly applied macro level strain and " I and " T
are the inelastic
s s
and thermal subcell strains. In the above equations, matrix notation is employed,
where
7 A Nano-micro-macro-multiscale Model for Progressive Failure Prediction. . . 163
D f 11 ; 22 ; 33 ; 23 ; 13 ; 12 g
" D f"11 ; "22 ; "33 ; 2"23 ; 2"13 ; 2"12 g (7.16)
n o
" I D " I.111/ ; : : : ; " I .N˛ Nˇ N /
s
It should be noted that a key step in the construction of this macro-constitutive law
is the derivation of the appropriate elastic concentration matrices A.˛ˇ / and D.˛ˇ /
at the subcell level. The matrix C.˛ˇ / represents the elastic stiffness tensor of each
subcell.
The nanoscale modeling and the micromechanical level of modeling were developed
and implemented within a synergistically coupled framework, named MAC/GMC.
As mentioned earlier, data obtained from lower length scale calculations (i.e.,
MD) was used as an input for the hierarchical RUC-based multiscale model of
nanoparticle-reinforced composites (GMC), which in turn provides data to the
macroscale FEA model.
One example of the hierarchical coupling methodology is Abaqus/MAC/GMC
coupling for modeling fiber-reinforced polymer nanocomposites. A FORTRAN
subroutine, FEAMAC, developed by NASA acts as a gateway between macroscale
(Abaqus) and microscale (GMC) for a coupled simulation. Figure 7.19 shows the
transfer of information between the macro- and the microscale. At each time step of
a macroscale simulation, global stresses are computed at element integration point,
which is then transferred to the GMC model for the computation of local stresses
based on the constitutive definition using Eqs. (7.12), (7.13), (7.14), and (7.15). The
material properties for the resin subcells in an RUC are obtained from the nanoscale
using NIDM and atomistic J-integral (toughness) data.
Any damage arising in the microscale is propagated onto the next macroscale
time step through the degradation of element stiffness. This allows damage evolution
at micro- and nanoscales due to applied stresses at the macroscale since the
constitutive material model can be directly formulated using MD simulations
(Sect. 7.2.1).
Figure 7.20 shows a simple example of progressive failure prediction for an
open-hole tension (OHT) specimen using the methodology mentioned above. The
composite specimen is a [45/0/45/90]2s laminate with EPON 862 matrix and
IM7 fiber with nanographene as the nanoscale reinforcement. The laminate size is
64 mm 16 mm with thickness of 0.5 mm and a hole size of 3.175 mm in diameter.
Each modeling scale was defined using the following steps:
164 S. Roy et al.
1. Nanoscale stiffness and damage parameters for unreinforced (neat resin) and
graphene-reinforced EPON 862 (2 aspect ratio (AR) NGP) was obtained using
the NIDM method described in Sect. 7.2.1.
2. For incorporation of the above properties in the microscale GMC, the stiff-
ness parameters were homogenized assuming a random orientation of the
nanographene in the EPON 862 matrix.
3. The homogenized resin properties were incorporated in a 2 2 RUC model in
GMC as in Fig. 7.19 where one subcell was used to model the fiber and the
remaining three subcells were used to model the matrix. Damage initiation at
the microscale was modeled using a maximum stress failure criterion for the
matrix subcell and a maximum strain failure criterion for the fiber subcell. The
maximum stress estimates for the matrix subcell are directly obtained from the
critical stress at the initiation of damage as predicted by the NIDM-based MD
7 A Nano-micro-macro-multiscale Model for Progressive Failure Prediction. . . 165
simulations in Sect. 7.2. A subcell was assumed to lose its load-carrying capacity
at the onset of damage.
4. An OHT specimen undergoing tensile loading was modeled in the Abaqus finite
element software and was iteratively coupled with GMC to model the microscale
RUC. In addition to evolving damage within the RUC, a cohesive damage
model available in Abaqus was employed to model macroscale delamination
phenomenon outside the RUC.
Figure 7.21a shows the load–displacement plot for the OHT simulation under
tensile loading. Several cases were studied, with and without nanographene rein-
forcement and delamination modeling, to observe its influence on the predicted
failure. As is evident from Figure 7.21, the hierarchical multiscale model is able
capture small changes in stiffness and damage initiation (Figure 7.21b), due to the
presence of NGP. It is also evident that while the influence of nanographene platelets
is significant in increasing the OHT strength of the composite, the inclusion of
cohesive layers to model delamination is not as significant for this case. Figure 7.22
shows the evolution of stress contours in each individual lamina until maximum
load is reached. For this simple example, the hierarchical multiscale model predicts
a 3.7 % increase in OHT strength over baseline case with the addition of NGP
(with platelet aspect ratio D 2) if delamination was not modeled using cohesive
elements between lamina. The same model predicts a 4.1 % increase in OHT
strength over baseline case with the addition of NGP (with platelet aspect ratio D 2)
if delamination was modeled using cohesive elements between lamina. Hence,
the three-level multiscale model predicts a slight increase in OHT strength of the
laminate due to the presence of NGP in EPON 862 matrix. Experimental verification
of the numerical predictions is currently underway.
Fig. 7.21 (a) Load–displacement plot for multiscale OHT simulation for various cases. (b)
Magnified view at load drop
7 A Nano-micro-macro-multiscale Model for Progressive Failure Prediction. . . 167
Fig. 7.22 Von Mises stress contour for individual lamina in IM7/EPON 862 OHT simulation
References
1. P.K. Valavala, T.C. Clancy, G.M. Odegard, T.S. Gates, Nonlinear multiscale modeling of
polymer materials. Int. J. Solids Struct. 44, 1161–1179 (2007)
2. V.A. Buryachenko, A. Roy, K. Lafdi, K.L. Anderson, S. Chellapilla, Multi-scale mechanics of
nanocomposites including interface: Experimental and numerical investigation. Compos. Sci.
Technol. 65, 2435–2465 (2005)
3. J.C. Riddick, S.J.V. Frankland, T.S. Gates, Multiscale analysis of delamination of
carbon fiber-epoxy laminates with carbon nanotubes, in Proceedings of the 47th
AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynamics, and Materials Conference,
Newport, Rhode Island, 1–4 May 2006
4. A.P. Awasthi, D.C. Lagoudas, D.C. Hammerand, Modeling of interface behavior in carbon
nanotube composites, in Proceedings of the 47th AIAA/ASME/ASCE/AHS/ASC Structures,
Structural Dynamics, and Materials Conference, Newport, Rhode Island, 1–4 May 2006
5. K.E. Wise, State of the art assessment of simulation in advanced materials development.
NASA-Technical Memorandum (TM), TM No. 215118 (2008)
168 S. Roy et al.
6. E.B. Tadmor, R.E. Miller, Modeling Materials: Continuum, Atomistic and Multiscale Tech-
niques (Cambridge University Press, Cambridge, 2012)
7. F.F. Abraham, J.Q. Broughton, N. Bernstein, E. Kaxiras, Concurrent coupling of length scales:
methodology and application. Phys. Rev. B 60(4), 2391–2403 (1999)
8. S. Ogata, E. Lidorikis, F. Shimojo et al., Hybrid finite-element/molecular-dynamics/electronic-
density-functional approach to materials simulations on parallel computers. Comput. Phys.
Commun. 138, 143–154 (2001)
9. G.J. Wagner, W.K. Liu, Coupling of atomistic and continuum simulations using a bridging
scale decomposition. J. Comput. Phys. 190, 249–274 (2003)
10. W.K. Liu, E.G. Karpov, H.S. Park, Nano Mechanics and Materials: Theory, Multiscale
Methods and Applications (Wiley, UK, 2006)
11. J. Ma, H. Lu, B. Wang et al., Multiscale simulation using generalized interpolation material
point (GIMP) method and molecular dynamics (MD). Comput. Model. Eng. Sci. 14(2),
101–117 (2006)
12. E. Saether, V. Yamakov, E. Glaessgen, An embedded statistical method for coupling molecular
dynamics and finite element analyses. Int. J. Numer. Methods Eng. 78, 1292–1319 (2009)
13. S. Roy, M. Benjamin, Modeling of permeation and damage in graphite/epoxy laminates for
cryogenic fuel storage. Compos. Sci. Technol. 64, 2051–2065 (2004)
14. M. Paley, J. Aboudi, Micromechanical analysis of composites by the generalized cell model.
Mech. Mater. 14, 127–139 (1992)
15. A. Kumar, S. Li, S. Roy et al., Fracture properties of nanographene reinforced EPON 862
thermoset polymer system. Compos. Sci. Technol. 114, 87–93 (2015)
16. A.R. Leach, Molecular Modelling: Principles and Applications (Addison Wesley Longman
Ltd, Harlow, 1996)
17. K. Chenoweth, A.C.T. van Duin, W.A. Goddard, ReaxFF reactive force field for molecular
dynamics simulations of hydrocarbon oxidation. J. Phys. Chem. A 112, 1040–1053 (2008)
18. W.L. Jorgensen, D.S. Maxwell, J. Tirado-Rives, Development and testing of the OPLS all-atom
force field on conformational energetics and properties of organic liquids. J. Am. Chem. Soc.
118(45), 11225–11236 (1996)
19. S.J. Weiner, P.A. Kollman, D.A. Case et al., A new force field for molecular mechanical
simulation of nucleic acids and proteins. J. Am. Chem. Soc. 106(3), 765–784 (1984)
20. E.M. Duffy, P.J. Kowalczyk, W.L. Jorgensen, Do denaturants interact with aromatic hydrocar-
bons in water. J. Am. Chem. Soc. 115(20), 9271–9275 (1993)
21. R. Talreja, Continuum modelling of damage in ceramic matrix composites. Mech. Mater. 12(2),
165–180 (1991)
22. S. Roy, A. Srivastav, Multiscale modeling of progressive failure in polymer nanocom-
posites using nanaoscale informed damage mechanics. J. Mech. Mater. Struct. (2015).
doi:10.1080/15376494.2015.1101513
23. G.M. Odegard, B.D. Jensen, S. Gowtham et al., Predicting mechanical response of crosslinked
epoxy using ReaxFF. Chem. Phys. Lett. 591, 175–178 (2014)
24. R.E. Jones, J.A. Zimmerman, The construction and application of an atomistic J-integral via
Hardy estimates of continuum fields. J. Mech. Phys. Solids 58(9), 1318–1337 (2010)
25. R.E. Jones, J.A. Zimmerman, J. Oswald, T. Belytschko, An atomistic J-integral at finite
temperature based on Hardy estimates of continuum fields. J. Phys. Condens. Matter 23,
015002 (2011)
26. R.J. Hardy, Formulas for determining local properties in molecular dynamics simulations:
shock waves. J. Chem. Phys. 76(1), 622–628 (1982)
27. S. Roy, A.R. Akepati, Determination of atomistic J-integral of graphene sheet using the
molecular dynamics method. Compos. Interfaces 20(6), 431–442 (2013)
28. P. Zhang, L. Ma, F. Fan et al., Fracture toughness of graphene. Nat. Commun. 5, 3782 (2014)
29. P. Klein, H. Gao, Crack nucleation and growth as strain localization in a virtual-bond
continuum. Eng. Fract. Mech. 61(1), 21–48 (1998)
30. S.G. Bardenhagen, E.M. Kober, The generalized interpolation material point method. Comput.
Model. Eng. Sci. 5, 477–495 (2004)
7 A Nano-micro-macro-multiscale Model for Progressive Failure Prediction. . . 169
31. R. Krueger, Virtual crack closure technique: history, approach, and applications. Appl. Mech.
Rev. 57(2), 109–143 (2004)
32. Y. Guo, J.A. Nairn, Calculation of J-integral and stress intensity factors using the material point
method. CMES. 10, 1–4 (2004)
33. S. Pfaller, M. Rahimi, G. Possart et al., An arlequin-based method to couple molecular dynam-
ics and finite element simulations of amorphous polymers and nanocomposites. Comput.
Methods Appl. Mech. Eng. 260, 109–129 (2013)
34. S.H. Cheng, C.T. Sun, Size-dependent fracture toughness of nanoscale structures: crack-tip
stress approach in molecular dynamics. J Nanomech Micromech 4, 4 (2014)
35. J. Aboudi, Micromechanical analysis of thermo-inelastic multiphase short-fiber composites.
Compos. Eng. 5(7), 839–850 (1995)
36. S.M. Arnold, T.E. Wilt, A.F. Saleeb, M.G. Castelli, An investigation of macro and microme-
chanical approaches for a model MMC system. NASA CP 19117, II, 52.1–52.12 (1993)
37. S.M. Arnold, M.J. Pindera, T.E. Wilt, Influence of fiber architecture on the elastic and inelastic
response of metal matrix composite. Int. J. Plast. 12(4), 507–545 (1996)
38. S.M. Arnold, B.A. Bednarcyk, T.E. Wilt, D. Trowbridge, Micromechanics Analysis Code
with Generalized method of cells (MAC/GMC) User Guide Version 3.0. NASA/Technical
Memorandum(TM), No. 209070 (1999)
Chapter 8
Carbon Fibre-Reinforced Polymer Laminates
with Nanofiller-Enhanced Multifunctionality
8.1 Introduction
sinks in electronics and for space structures that require dimensional stability.
Conventional continuous CFRP composites have been developed over the past half-
century to provide a wide range of materials that are lightweight and combine
superior mechanical properties with chemical and environmental resistance, making
them particularly attractive for numerous applications in aerospace, automation,
marine environments, sporting goods, civil infrastructure and the energy sector [1].
Advances in producing nanostructured materials with novel material proper-
ties have stimulated research to create multifunctional macroscopic engineering
materials by designing structures at the nanometre scale. Motivated by the recent
enthusiasm in nanotechnology, the development of nanocomposites is a rapidly
evolving field in the composites research community. The interest of scientists
in applying nanofillers into matrices lies in the attainment of potentially unique
properties as a result of the nanometre scale [2, 3]. Nanocomposites are a new class
of composites that are particle-filled composites in which at least one dimension of
the dispersed particles is within the 100 nm range.
Scientists and engineers working with CFRP practise the ‘bottom-up’ approach
in processing and manufacturing nanofiller-incorporated CFRP composites. When
a composite is designed, the material properties are tailored for the desired
performance across various length scales. From the selection and processing of
nanofiller, matrix and fibre materials and the design and optimization of the
nanofiller/fibre/matrix interface/interphase at the nano-/submicron scale to the
manipulation of yarn bundles in 2D and 3D textiles to the layup of laminae
in laminated composites and finally the net-shape formation of the macroscopic
composite part, the integrated approach used in composite processing is a remark-
able example of the successful use of the bottom-up approach. The extension of
length scales from metres (finished composite parts), micrometres (fibre diame-
ter) and sub-micrometres (fibre/matrix interphase) to nanometres (nanofiller size,
filler/fibre/matrix interphase) presents tremendous opportunities for innovative
approaches in the processing, characterisation and analysis of this new type of
FRP composites. As scientists and engineers seek to build practical materials and
devices from nanostructures, understanding material behaviour across length scales
from atomistic to macroscopic levels is required. Knowledge of how the nanoscale
structure influences the bulk properties of CFRP composites will facilitate enhanced
design of the nanostructure to create multifunctional CFRP.
In this chapter, we address the state of knowledge of different nanofiller-
enabled/nanofiller-enhanced multifunctionalities of CFRP and compare the
properties and performance of nanofiller-enabled/nanofiller-enhanced CFRP with
traditional CFRP to highlight the advantages of the new generation CFRP with
nanofillers. Future trends are also explored here.
8 Carbon Fibre-Reinforced Polymer Laminates with Nanofiller-Enhanced. . . 173
Past decades have seen considerable interest in the use of nanofillers as secondary
reinforcement in addition to the primary reinforcing fibres in FRP laminates, in
a ‘hybrid’ approach. Some recent developments have been detailed in the use of
nanofillers as additional reinforcing phases in FRP laminates (especially with brittle
thermosetting resins) to enhance interlaminar fracture toughness and compression-
after-impact (CAI) strength [4]. Strategies for obtaining FRP composites with
enhanced interlaminar strength and toughness by various techniques have been
highlighted [5]. As the strength of FRP composites is mostly limited by interlaminar
properties, various ways to enhance interfacial mechanical properties have been
described by modifying the matrix and reinforcement phases to obtain high strength
and toughness. The effects of interleaving phases on mechanical characteristics
have been summarised. Various fibre architectures and their influences have been
emphasised [5]. The work of Prusty et al. provides a broad overview and technically
viable routes for obtaining particular combinations of strength and toughness.
The influence of filler size, shape, orientation and volume content of conventional
composite materials on the properties of the composite can be described by the
boundary concept and the model concept [6]. Figure 8.1 shows the surface-to-
volume ratio for a variety of nanofillers. Carbon nanotubes (CNTs) with different
aspect ratios (length/diameter) as a function of the diameter are shown here, and the
same ratios are shown for spherical nanoparticles such as fumed silica (FS) and car-
bon black (CB). For comparison, conventional reinforcements like glass balls (GB),
glass fibres (GFs) and CFs are also given. In the double logarithmic scale of Fig. 8.1,
the surface/volume ratio decreases linearly with increasing particle diameter. It can
be seen that a small volume content of nanofillers provides huge surface areas,
thereby enhancing the nucleation of polymer crystals in thermoplastic materials or
the cross-linking density in thermosets, resulting in increased mechanical properties
by changing the polymer morphology [6]. The advantage of nanofiller reinforcement
may thus be a synergistic effect of introducing the reinforcing phase with desirable
mechanical properties and enhancing the polymer morphology.
8 Carbon Fibre-Reinforced Polymer Laminates with Nanofiller-Enhanced. . . 175
8.2.2.2 Nanofibres
The use of nanofibres (NFs) can be a feasible way to enhance interlaminar properties
without compromising other mechanical properties. NFs can be applied in small
amounts and, because they are very flexible, they can take the shape of the
8 Carbon Fibre-Reinforced Polymer Laminates with Nanofiller-Enhanced. . . 177
8.2.2.3 Organoclay
Organoclays are well suited for the design of hybrid composites because they have
high in-plane strength, stiffness and a high aspect ratio. The platelet ratio exceeds
300, giving rise to a high degree of polymer-clay surface interaction which results in
barrier and mechanical properties that are far superior to those of the base material.
Some of the earliest work on inorganic toughening occurred in 1987 at Toyota.
Researchers demonstrated the possibility of functionalising the surface of layered
inorganic silicates to increase their affinity for a polymeric matrix [38]. Clays have
been shown to be effective reinforcement in neat polymeric structures, but not
much work has been done to examine CFRP composites that utilise nanocomposite
matrices [29, 39–44]. Possible reasons for this neglect are that (1) clay can only
be dispersed in matrix for enhancement. With clay added, thermoset resins have
enhanced fracture toughness in terms of Gm IC , but, as the amount of clay increases,
most of the Gc IC /Gm IC values decrease [29, 39, 42, 44], indicating that transfer
efficiency decreases, which is the result of the constraint on the development of
a larger plastic zone in the resin-rich region between plies by the fibres in the
adjacent plies [4] and (2) only intercalated nanocomposites where the polymer
chains are sandwiched between silicate layers and exfoliated nanocomposites where
the separated, individual silicate layers are more or less uniformly dispersed in the
polymer matrix have the reinforcement effect. Organoclay is a cost-effective and
commercially available nanofiller, examples being Cloisite 30B and NanomerI.30E,
but the lack of a commercially available resin that contains organoclay compromises
homogeneous dispersion of organoclay in polymeric matrices, leading to resultant
mechanical properties of CFRP that are inconsistent or contradictory.
178 W. Han et al.
Besides the nanofillers already mentioned that have been incorporated in CFRP
laminates, other nanofillers have also been used for enhancement. A product
of special interest is fullerene (nanom mix™ ST) provided by Frontier Carbon
Corporation, Japan. This fullerene is a mixture of C60 (60 %), C70 (25 %) and
other higher fullerenes (15 %). The mean particle size is smaller than 10 µm
because of aggregation, and the bulk density of packed fullerene is approximately
0.6 g/cm3 . Fullerene/bisphenol-A epoxy (Epikote 828; Japan Epoxy Resins Co.,
Ltd.) master batch (fullerene 5 wt%) was provided by Frontier Carbon Corp. in
research conducted by Ogasawara et al. [53]. Other nanofillers of interest are
graphene and graphene oxide (GO). It has been reported that both graphene and GO
showed promise as reinforcements in high-performance nanocomposites. However,
8 Carbon Fibre-Reinforced Polymer Laminates with Nanofiller-Enhanced. . . 179
it is not easy to obtain good dispersions, and there are challenges in obtaining the
full exfoliation of graphene into single- or few-layer material with reasonable lateral
dimensions or in producing graphene oxide without imparting significant damage to
the flakes [54, 55]. Those may be the reasons for the reported marginal mechanical
reinforcement of graphene and GO on CFRP [19, 56–61]. Apart from its good
prospects of mechanical reinforcement, there are also possibilities of using graphene
to control functional properties such as electrical conductivity, gas barrier behaviour
and thermal conductivity, expansion and stability [62]. Other nanofillers, including
polystyrene-block-polybutadiene-block-poly(methylmethacrylate) (SBM) copoly-
mer triblock [29], carbon black [60], halloysite nanotubes [63], silver particle and
silver wire [57] and inorganic fullerene-like tungsten disulphide (IF-WS2) [64],
have been reported to enhance mechanical properties of CFRP composites.
Interlaminar shear strength (ILSS) is an important material property for the design
of laminated composite structures subjected to transverse loads. A review of these
studies concludes that SBS test, at best, gives a qualitative value of ILSS. It has
180 W. Han et al.
a 2400
Nanofiller enhanced
Neat
800
0
CNT Nanofibre Organoclay Sillica&rubber Others
b 3000
Nanofiller enhanced
Neat
2000
GIIC (MPa)
1000
0
CNT NanoFiber Organoclay Sillica&rubber Others
Fig. 8.2 (a) GIC of CFRP composites with different types of nanofillers. (b) GIIC of CFRP
composites with different types of nanofillers
been reported that the ILSS of hierarchical composites containing nanotubes in the
interlaminar that can be increased by 2–45 % depend on the CNT type, content
and surface chemistry [13, 20, 23, 26, 29, 32, 39–41, 43, 53, 56, 57, 61, 63, 64].
Figure 8.4 shows recently reported ILSS results of nanofiller-enhanced laminates.
It can be seen that the ILSS values were all below 100 MPa without considering
filler type, filler size, filler concentration and characterisation conditions. However,
there are a few points in Fig. 8.4 showing ILSS values higher than 100 MPa when
fillers are added into CFRP composites [57]. Looking at this research carefully, the
enhanced ILSS value was due to the synergistic effect of silver wire and graphene,
and if only one type of nanofiller is added, the ILSS value will decrease compared
with that of the conventional CFRP composite itself. Nevertheless, many factors are
known to play a role in laminate ILSS, including the degree of porosity, interlaminar
layer thickness and crystallinity. These strong improvements in ILLS are attributed
to the high quality (very low void content) of the laminates. A combination of
8 Carbon Fibre-Reinforced Polymer Laminates with Nanofiller-Enhanced. . . 181
3000
Nanofiller enhanced
Neat
1000
0
0 1000 2000 3000
G (MPa)
IC
Fig. 8.3 Toughness ratio for conventional and nanofiller-enhanced resin laminates
The CAI test has become a key experiment to gather damage tolerance performance
data during the design or certification phase of a new structure or materials
involving composites. CAI performance is first of all dependent on the behaviours
of material under impact loading conditions. The Federal Aviation Administration
(FAA) identified 14 different parameters influencing the impact performance of
a composite structure [67]. Some are related to the structure being impacted,
and others are related to the impacting object. Various investigations have been
studied on many of these parameters such as the stacking sequence, the impact
velocity and the constituents [68, 69]. Cartie and Irving [70] reported that resin
toughness rather than fibre strength and stiffness was the major parameter that
influences CAI performance in quasi-isotropic CFRP laminates. It was suggested
that measurements of Pc (the critical load) obtained during impact can be used as
an easier route to determine the value of GIIC than delamination testing [70]. Davies
et al. [71] also proposed a model based on mode II interlaminar fracture toughness,
GIIC , of the composite to predict the critical force to initiate delamination damage
182
Table 8.1 The fracture toughness ratio (GIIC /GIC ) of composite laminates with different nanofillers
Fillers Filler content GIC (J/m2 ) GIIC (J/m2 ) GIIC /GIC References
Cup-stacked CNT D D 70–80 nm, AR D 10 or 100 0 86 568 6.60 Yokozeki et al. [25]
5 170 732 4.31
5 & AR10 sprinkle 148 816 5.51
5 & AR10 film 227 1753 7.72
5 & AR100 sprinkle 190 1091 5.74
5 & AR100 film 161 751 4.66
VA CNT D D 8 nm, space D 80 nm AS4/8552 210 350 1.67 Garcia et al. [14]
AS4/8552/CNT 340 1100 3.24
SWCNT D D 1.3 nm 0 343 1100 3.21 Ashrafi et al. [12]
With CNT 387 1410 3.64
MWCNT D D 50 nm 0 740 840 1.14 Joshi & Dikshit [17]
1.32 g/m2 977 2016 2.06
2.25 g/m2 644 1554 2.41
3.62 g/m2 429 1100 2.56
CNF D D 50–200 nm Neat 390 1724 4.42 Kostopoulos et al. [31]
W. Han et al.
CNF 1 % 790 2626 3.32
CNF 1 %/PZT 4 % 270 2542 9.41
VGCF D D 200 nm 0 425 750 1.76 Quaresimin and Varley [29]
7.5 120 1500 12.5
Cloisite 30B 0 425 750 1.76 Quaresimin and Varley [29]
5 125 880 7.04
Nanopox F 400; Albipox 1000 0 433 1299 3.0 Sprenger [47]
10R 688 1124 1.63
5S, 10R 613 1034 1.69
7.5S, 10R 555 1031 1.86
10S, 10R 621 1080 1.74
Nanopox F400 0 995 969 0.97 Tang et al. [48]
10S 1007 929 0.92
20S 1203 750 0.62
SBM Neat 425 750 1.76 Quaresimin and Varley [29]
10 295 285 0.97
Carbon black D D 35 nm 0 974 838 0.86 Zhang et al. [60]
3 1415 877 0.62
Halloysite nanotubes 0 400 1300 3.25 Ye et al. [63]
1 550 1620 2.95
8 Carbon Fibre-Reinforced Polymer Laminates with Nanofiller-Enhanced. . .
400
ILSS (MPa)
200
0
CNT Nanofiber Organoclay Silica&rubber Others
during transverse impact. The model predictions for the initiation of delamination
agreed well with their experimental data on quasi-isotropic laminates [71]. This
model supports the common views in the composites research community that a
large GIIC benefits damage resistance under transverse impact, even though the
current available data in the literature showed that generally CAI strength of
nanofiller-enhanced composites does not have clear relationship to GIIC [4].
As delamination can be produced by low velocity impact, large strength reduc-
tion in compression can occur. Delamination divides the laminate into sub-laminates
which have a lower bending stiffness than original laminate and are less resistant to
buckling loads. Under a compressive load, a delamination can cause bucking in one
of three modes: global instability/bucking of the laminate, local instability (bucking
of the thinner sub-laminate) or a combination of the above [72]. Figure 8.5 shows
the CAI strength values with different nanofillers added in the CFRP composites,
and those values were lower than 300 MPa irrespective of matrix, CF type, nanofiller
type and manufacture procedures. With CNT added [12, 24], the CAI strengths were
higher than those of silica and/or rubber added [47]. It had the lowest values with
clay added [41], and the CAI strength values reported by adding other nanofillers
varied significantly [30, 33, 53, 58].
350
CAI(MPa) 280
210
140
70
CNT Organoclay Sillica&rubber Others
Nf G
Nanofillers Region Matrix/hardener/CF Fabrication (105 ) (J/m2 ) References
MFC & BC; 0 Dispersed in matrix Epikote 828/JER cure-113/PW fabric PW 8 layers; 2.5 200 Shao et al. [34]
D D 50–200 nm TR3110M & UD TR50S 12 L hand layup; UD
filament wound;
Vf D 50
MFC 0.3 50 330
MFC 0.8 15 330
BC 0.3 35 330
BC 0.8 30 340
SWCNT Pristine Interlayers Epoxy/NA/woven CF 1.8 Thaker [23]
Functional 4.4
MWCNT; D D 20–50 nm CF/PP Grow on CF by CVD PAN-based short CF (5 wt%,)/PP Internal mixer 113 Rahmanian et al. [20]
and hot press
CF/CNT/PP 152
MWNT-7; D D 60 nm 0 Dispersed in matrix XNR6809/XNH6809/twill woven [0ı ]20 ; VARTM 2.0 86 Arai et al. [11]
and used as fabric C06347B
interlayers
10 g/m2 3.6
20 g/m2 3.1 96
30 g/m2 2.6
MWCNT; D D 10 nm 0; 605 MPa Dispersed on prepreg Araldite LY [0ı /90ı ]2 s; 6.42 Knoll et al. [19]
1556/Aradur1571/Accelerator Prepreg; Vf D 57
1573/Hardener XB 3403/T700S
0; 651 MPa 1.10
0; 745 MPa 0.07
0.3; 605 MPa 15.3
0.3; 651 MPa 2.10
W. Han et al.
of the energy release rate in fatigue crack progress under displacement control tests
was 96 J/m2 , compared with 86 J/m2 for the base laminate at the values of crack
propagation rate 5 109 m/cycle. Shao et al. [34] found that the capability of
resistance to the growth of delamination was enhanced with the adding of CNF
under mode I DCB cyclic loading. Gth increased by 1.6–1.7 times compared to
unmodified composites despite the type of CNF, as an evidence of resistance to
initiation of delamination in composite under the effect of nanofiller. However, more
research reports needed to further evidence the current findings.
CFRP composites have high in-plane but relatively low through-thickness thermal
conductivity. The low through-thickness properties are dominated by the poor
thermal conductivity of the polymer matrices. Therefore, despite the desirable
mechanical properties exhibited by CFRP composites, it is important to enhance
Table 8.3 Nanofiller-enhanced electrical conductivity of CFRP
188
One of the most successful examples is nanosilica, with the trademark Nanopox
F400, obtained at a concentration of 40 wt% in the diglycidyl ether of bisphenol-A
(DGEBA) epoxy resin [46–49]. The recently available fullerene/bisphenol-A epoxy
(Epikote 828; Japan Epoxy Resins Co., Ltd.) master batch (fullerene 5 wt%) pro-
vided by Frontier Carbon Corp. is another example [53]. More and more commercial
providers of nanofiller-matrix composites with consistent material properties will
appear with the improvements in processing methods and equipment, providing a
greater range of suitable nanofiller-modified matrices for CFRP composites.
Meanwhile, studies of hybrid composites using 1D fibrous and 2D platelet
nanofillers, such as CNTs, clays and graphene dispersed in polymers, have reported
only marginal mechanical property enhancement if there is no control of the
alignment of the nanofillers. Attaching nanofillers directly onto primary reinforcing
CFs has been studied, particularly incorporating the use of vertical-grown CNT
(VGCNT). With this method, CNTs can act as interlaminar nano Z-fibres/Z-pins,
termed ‘interlaminar nanostitches’. However, due to the time-consuming nature
and other cost-limiting factors of hybrid laminate fabrication, large-scale and
commercial production would be difficult in the current stage. As reported by
Kinloch et al. [86], Hsieh et al. [46], Xu et al. [87] and Sprenger [47], a combination
of silica nanoparticles and CTBN rubber microparticles in a DGEBA epoxy as the
matrix for CFRPs can achieve synergistic effects. Those materials are commercially
® ®
available, with the trademarks of Nanopox and Albipox , respectively. Cavitation
of the rubber particles and subsequent void growth promoting significant matrix
shear deformation are the main toughening mechanisms, and meanwhile, the energy
dissipated in the process of debonding of nanosilica particles from the matrix also
contributes to the improvement of fracture toughness in such ternary composites.
Based on the fracture toughness mechanisms in modified matrices, appropriate
selection of different nanofillers can be a practical way of further exploring and
utilising nanofiller-modified matrices for producing hybrid CFRPs. Meanwhile,
for thermal and electrical conductivity, Kandare et al. [57] recently reported
experiments incorporating graphene sheet, silver nanoparticle and silver nanowire
fillers into CFRP laminates, with the objective of enhancing through-thickness
thermal and electrical conductivity without adversely affecting mechanical proper-
ties. This significant increase in the thermal conductivity of laminates containing
both graphene nanoplatelets and silver nanoparticles/nanowires is greater than
theoretical estimates. The more than mere additive effect on through-thickness
thermal conductivity suggests synergistic physical interactions between graphene
nanoplatelets and silver nanoparticles/nanowires leading to further enhancement
in through-thickness thermal conductivity, which is another excellent example of
multi-filler selection for achieving multifunctionality of CFRP.
For fabricating composite laminates, hand layup with dipping, brushing and
rolling has the advantage of simplicity, but is limited to relatively low fibre volume
fractions and may always include voids and defects within a laminate of poor qual-
ity. The ILSS and transverse flexural strength of unidirectional CFRP are roughly
inversely proportional to the square of void content in the composites, a relationship
that highlights the importance of fabrication of high-quality composites [88]. A high
192 W. Han et al.
fibre volume fraction (>60 vol.%) is essential for CFRP composites to maintain
their superior performance with specified stiffness and strength in comparison
with high-performance light metallic alloys. Automated procedures such as resin
transfer moulding (RTM) and vacuum-assistant resin infusion moulding (VARIM)
are typical methods for fabricating fibre-reinforced composites of high quality; both
methods, especially the latter, are preferred for the production of large and com-
plex structural composite parts. Preparing composite prepreg using fibre filament
impregnation on a drum followed by hand layup and vacuum bagging for curing
in an autoclave is also an effective method for fabricating composite laminates of
high quality. Most automated manufacturing methods, particularly RTM, require a
matrix resin with low viscosity, in the typical order of 0.5–1 Pa s [89]. However,
the viscosity of most nanofiller-modified matrices increases dramatically with an
increase in nanoparticle content, leading to incomplete matrix impregnation and
infusion or misplacement of fibre reinforcement. To penetrate into the primary
fibre tows and avoid filtration effects, low nanoparticle content can be used, but
that approach reduces the potential for significantly enhanced fracture toughness.
A diluted matrix using an organic solution may help impregnation, but the gradual
change in matrix viscosity with evaporation of the solution or complete removal of
the solution can add new problems of quality control of the composites. The use of
different types of electrospun fibres, reported recently [30, 32] as the interleaves on
CFRP laminate, may be a good choice to achieve multifunctionalisation of CFRP
composites. Nanofillers can be incorporated with polymeric matrix first and spun
out as the nanofibre for enhancement. Further exploration of this area is warranted.
Additive manufacturing processes, such as 3D printing, offer another promising
direction in the fabrication of CFRP laminates. Silicon Valley start-up Arevo
Labs announced new CF- and CNT-reinforced high-performance materials for 3D
printing. Patent pending 3D printing technology and specialised software algorithms
exist that enable production-grade 3D printed ultra-strong polymer parts using
commercially available filament fusion 3D printers in 2014 with supported materials
including PEEK, PAEK, PPSU and SRP. In the same year, Compton and Lewis
[90] demonstrated the first 3D printed cellular composites composed of oriented
fibre-filled epoxy with exceptional mechanical properties. Tumbleston et al. [91]
subsequently demonstrated the continuous generation of monolithic polymeric parts
up to tens of centimetres in size with feature resolution below 100 µm. These print
speeds allow parts to be produced in minutes instead of hours.
Little research has been conducted into environmental and economic factors
for the development of recycling routes for the increasing amount of CFRP waste
generated and also for the future nanofiller-enhanced CFRP. Pimenta and Pinho [92]
reviewed the current status and outlook of CFRP recycling operations, focusing
on state-of-the-art fibre reclamation and remanufacturing processes and on the
commercialisation and potential applications of recycled products. The potential
of using nanofillers to recover/enhance the multifunctionality of recycled CFRPs
with competitive structural performance is another interesting direction for future
exploration. Occupational health and safety of nanofiller hybrid laminates are
8 Carbon Fibre-Reinforced Polymer Laminates with Nanofiller-Enhanced. . . 193
Acknowledgements Y Tang and L Ye are grateful for the support of the Premier’s Research and
Industry Fund (PRIF) with a Catalyst Research Grant and the Australian Research Council (ARC)
with a Discovery Project (DP) grant for the research work, respectively.
References
10. A.M. Diez-Pascual, M. Naffakh, C. Marco, M.A. Gomez-Fatou, G.J. Ellis, Multiscale fiber-
reinforced thermoplastic composites incorporating carbon nanotubes: a review. Curr. Opin.
Solid State Mater. Sci. 18, 62–80 (2014)
11. M. Arai, J. Hirokawa, Y. Hanamura, H. Ito, M. Hojo, M. Quaresimin, Characteristic of mode
I fatigue crack propagation of CFRP laminates toughened with interlayer. Compos. Part B 65,
26–33 (2014)
12. B. Ashrafi, J. Guan, V. Mirjalili, Y. Zhang, L. Chun, P. Hubert, B. Simard, C.T. Kingston,
O. Bourne, A. Johnston, Enhancement of mechanical performance of epoxy/cabob fibre
laminate composites using single-walled carbon nanotubes. Compos. Sci. Technol. 71,
1569–1578 (2011)
13. E. Bekyarova, E.T. Thostenson, A. Yu, H. Kim, J. Gao, J. Tang, H.T. Hahn, T.W. Chou,
M.E. Itkis, R.C. Haddon, Multiscale carbon nanotube carbon fiber reinforcement for advanced
epoxy composites. Langmuir 23, 3970–3974 (2007)
14. E.J. Garcia, B.L. Wardle, A.J. Hart, Joining prepreg composite interfaces with aligned carbon
nanotubes. Compos. Part A 39, 1065–1070 (2008)
15. A. Godara, L. Mezzo, F. Luizi, A. Warrier, S.V. Lomov, A.W. van Vuure, L. Gorbatikh,
P. Moldenaers, I. Verpoest, Influence of carbon nanotube reinforcement on the processing and
the mechanical behaviour of carbon fibre/epoxy composites. Carbon 47, 2914–2923 (2009)
16. L. Gorbatikh, S.V. Lomov, I. Verpoest, Nano-engineered composites: a multiscale approach for
adding toughness to fibre reinforced composites. Procedia Eng. 10, 3252–3258 (2011)
17. S.C. Joshi, V. Dikshit, Enhancing interlaminar fracture characteristics of woven CFRP prepreg
composites through CNT dispersion. J. Compos. Mater. 46, 665–675 (2011)
18. K.L. Kepple, G.P. Sanborn, P.A. Lacasse, K.M. Gruenberg, W.J. Ready, Improved fracture
toughness of carbon fiber composite functionalized with multi walled carbon nanotubes.
Carbon 46, 2026–2033 (2008)
19. J.B. Knoll, B.T. Riecken, N. Kosmann, S. Chandrasekaran, K. Schulte, B. Fiedler, The effect
of carbon nanoparticles on the fatigue performance of carbon fibre reinforced epoxy. Compos.
Part A 67, 233–240 (2014)
20. S. Rahmanian, K.S. Thean, A.R. Suraya, M.A. Shazed, M.A. Mohd Salleh, H.M. Yusoff,
Carbon and glass hierarchical fibers: influence of carbon nanotubes on tensile, flexural and
impact properties of short fiber reinforced composites. Mater. Des. 43, 10–16 (2013)
21. G. Romhany, G. Szebenyi, Interlaminar crack propagation in MWCNT/fiber reinforced hybrid
composites. Express Polym. Lett. 3, 145–151 (2009)
22. R.J. Sager, P.J. Klein, D.C. Davis, D.C. Lagoudas, G.L. Warren, H.J. Sue, Interlaminar fracture
toughness of woven fabric composite laminates with carbon nanotube/epoxy interleaf films.
J. Appl. Polym. Sci. 121, 2394–2405 (2011)
23. Thaker PR. Processing and characterisation of carbon nanotubes reinforced epoxy resin based
multi-scale multi-functional composites. Ph.D. Thesis, Texas A&M University, 2009
24. X. Xu, Z. Zhou, Y. Hei, B. Zhang, J. Bao, X. Chen, Improving compression-after-impact per-
formance of carbon-fiber composites by CNTs/thermoplastic hybrid film interlayer. Compos.
Sci. Technol. 95, 75–81 (2014)
25. T. Yokozeki, Y. Iwahori, M. Ishibashi, T. Yanagisawa, K. Imai, M. Arai, T. Takahashi,
K. Enomoto, Fracture toughness improvement of CFRP laminates by dispersion of cup-stacked
carbon nanotubes. Compos. Sci. Technol. 69, 2268–2273 (2009)
26. G.J. Zhang, The effect of carbon fibres and carbon nanotubes on the mechanical properties of
polyimide composites. Mech. Compos. Mater. 47, 447–450 (2011)
27. H. Zhang, Y. Liu, M. Kuwata, E. Bilotti, T. Peijs, Improved fracture toughness and integrated
damage sensing capability by spray coated CNTs on carbon fibre prepreg. Compos. Part A 70,
102–110 (2015)
28. V. Kostopoulos, P. Tsotra, P. Karapappas, S. Tsantzalis, A. Vavouliotis, T.H. Loutas,
A. Paipetis, K. Friedrich, T. Tanimoto, Model I interlaminar fracture of CNF or/and PZT doped
CFRPs via acoustic emission monitoring. Compos. Sci. Technol. 67, 822–828 (2007)
29. M. Quaresimin, R.J. Varley, Understanding the effect of nano-modifier addition upon the
properties of fibre reinforced laminates. Compos. Sci. Technol. 68, 718–726 (2008)
8 Carbon Fibre-Reinforced Polymer Laminates with Nanofiller-Enhanced. . . 195
75. M.C. Koecher, J.H. Pande, S. Merkley, S. Henderson, D.T. Fullwood, A.E. Bowden, Piezore-
sistive in-situ stain sensing of composite laminate structures. Compos. Part B 69, 534–541
(2015)
76. E.T. Thostenson, T.W. Chou, Carbon nanotube networks: sensing of distributed strain and
damage for life prediction and self-healing. Adv. Mater. 18, 2837–2841 (1996)
77. L. Boger, M.H.G. Wichmann, L.O. Meyer, K. Schulte, Load and health monitoring in glass
fibre reinforced composites with an electrically conductive nanocomposites epoxy matrix.
Compos. Sci. Technol. 68, 1886–1894 (2008)
78. Y. Lin, M. Gigliotti, M.C. Lafarie-Frenot, J. Bai, D. Marchand, D. Mellier, Experimental study
to assess the effect of carbon nanotube addition on the through-thickness electrical conductivity
of CFRP laminates for aircraft applications. Compos. Part B 76, 31–37 (2015)
79. E.F. Reia da Costa, A.A. Skordos, I.K. Partidge, A. Rezai, RTM processing and electrical
performance of carbon nanotube modified epoxy/fibre composites. Compos. Part A 43,
593–602 (2012)
80. S. Hida, T. Hori, T. Shiga, J. Elliott, J. Shiomi, Thermal resistance and phonon scattering at the
interface between carbon nanotube and amorphous polyethylene. Int. J. Heat Mass Transf. 67,
1024–1029 (2013)
81. D.J. Radcliffe, H.M. Rosenberg, The thermal conductivity of glass-fibre and carbon-
fibre/epoxy composites from 2 to 80 K. Cryogenics 22(5), 245–249 (1982)
82. Wikipedia.org. https://en.wikipedia.org/wiki/Rule_of_mixtures. Accessed 9/5/2016
83. J.R. Gaier, Y. YoderVandenberg, S. Berkebile, H. Stueben, F. Balagadde, The electrical and
thermal conductivity of woven pristine and intercalated graphite fibre-polymer composites.
Carbon 41, 2187–2193 (2003)
84. H. Jopek, T. Strek, Optimization of the effective thermal conductivity of a composite,
in Convection and Conduction Heat Transfer, ed. by A. Ahsan (Intech, Croatia, 2011),
pp. 197–214
85. G. Gkikas, D.D. Douka, N.M. Barkoula, A.S. Paipetis, Nano-enhanced composite materials
under thermal shock and environmental degradation: a durability study. Compos. Part B 70,
206–214 (2015)
86. A.J. Kinloch, R.D. Mohammed, A.C. Taylor, S. Sprenger, D. Egan, The interlaminar toughness
of carbon-fibre reinforced plastic composites using “hybrid-toughened” matrices. J. Mater. Sci.
41, 5043–5046 (2006)
87. S.A. Xu, G.T. Wang, Y.W. Mai, Effect of hybridization of liquid rubber and nanosilica particles
on the morphology, mechanical properties and fracture toughness of epoxy composites.
J. Mater. Sci. 48, 3546–3556 (2013)
88. K.J. Bowles, S. Frimpong, Void effects on the interlaminar shear strength of unidirectional
graphite-fibre-reinforced composites. J. Compos. Mater. 26, 1487–1509 (1992)
89. C.L. Lee, K.H. Wei, Resin transfer molding (RTM) process of a high performance epoxy
resin. II: effects of process variables on the physical, static and dynamic mechanical behaviour.
Polym. Eng. Sci. 40, 935–943 (2000)
90. B.G. Compton, J.A. Lewis, 3D printing: 3D-printing of lightweight cellular composites. Adv.
Mater. 26, 5930–5935 (2014)
91. J.R. Tumbleston, D. Shirvanyants et al., Continuous liquid interface production of 3D objects.
Science 347, 1349–1352 (2015)
92. S. Pimenta, S.T. Pinho, Recycling carbon fibre reinforced polymers for structural applications:
technology review and market outlook. Waste Manag. 31, 378–392 (2011)
93. D. Bello, B.L. Wardle, N. Yamamoto, R.G. deVilloria, E.J. Garcia, A.K. Hart, K. Ahn, M.J.
Ellenbecker, M. Hallock, Exposure to nanoscale particles and fibers during machining of
hybrid advanced composites containing carbon nanotubes. J. Nanopart. Res. 11, 231–249
(2009)
Chapter 9
Analysis Models for Polymer Composites
Across Different Length Scales
9.1 Introduction
Inelasticity, either resulting from plastic deformations or damage, has been one of
the main challenges in the prediction of the mechanical response of polymer com-
posites under general thermo-mechanical loads. The difficulties in the prediction
of the inelastic deformation of polymer composites have two main consequences:
expensive and time-consuming certification tests based on a large number of
coupons and, in some occasions, unexpected failures at advanced stages of the
product development process [1].
This chapter presents the efforts of the authors and of their colleagues to develop
improved models that are able to simulate the onset and propagation of damage
in composite materials. Taking into account the different length and temporal
scales associated with the damage mechanisms, models that range from a discrete
representation of the microstructure to the length scale of the laminate are covered.
Due to the clear separation between scales and constituents, computational micro-
mechanical models are taken as the baseline. The idea is not to link in a formal way
all the scales, but rather to use an approach similar to that suggested by Batterman
[2] “. . . bottom-up attempts will likewise induce changes and improvements in the
construction of higher scale models. Mesoscopic structures cannot be ignored and,
in fact, provide the bridges that allow us to model across scales.” Hence, the models
at the lower scales will provide the understanding that will improve the predictive
capabilities of the models developed at the higher scales.
Fig. 9.1 View of the damage mechanisms on composite laminates at different length scales [5, 6]
9 Analysis Models for Polymer Composites Across Different Length Scales 201
An RVE can be defined as the smallest volume fraction of the material whose
homogenized response is representative of the global mechanical performance of
the material system [10, 11] and virtually independent of the randomization
of the microstructure [12–18]. The critical size of an RVE depends not only on
the constituent and interface properties, but also on the spatial distribution of
reinforcements [12]. This critical size, though, cannot be too large, as it would
endanger the possibility to numerically analyze it.
In addition to the size of the RVE, other important issues include fiber distri-
bution, volume fraction, and spatial arrangement of reinforcements in the matrix.
This spatial arrangement is usually not periodic and it is highly dependent upon
the manufacturing process [13]. To accurately reproduce the onset and evolution of
damage in a composite system, it is important to properly represent the transverse
randomness of the fiber distribution (i.e., distance to first neighbors, occurrence of
clusters of fibers, etc.) [13, 19]. However, in the computational analysis of composite
materials, periodic RVEs are often employed, assuming that the material has a
deterministic and ordered distribution of fibers, mainly because this assumption
leads to lower computational costs [19]. To understand if this type of assumption
conducts to worse results than when using RVEs with randomly distributed fibers,
202 P.P. Camanho and A. Arteiro
Trias et al. [20] compared the stress and strain distributions between models with
periodic and random microstructures (e.g., Fig. 9.2). This comparison showed that
periodic models can be used for the determination of effective elastic properties
if high accuracy is not sought; nevertheless, for statistical and reliability analysis
or for the simulation of local phenomena and failure, random models must be
considered [20].
Different methodologies have been suggested to model random fiber distributions
in an RVE. Some of the most used methodologies were briefly reviewed by Melro
et al. [13], who found problems such as difficulties in generating distributions with
fiber volume fractions greater than 50 % [21–23], difficulties to guarantee non-
overlapping fibers [24], lack of statistical analyses [25], difficulties in achieving
a random pattern [26], or a very high computational effort even for moderate fiber
volume fractions (lower than 60 %) [27]. Alternatively, digital image analysis [27,
28] can provide a perfect replica of the transverse section of polymer composites,
9 Analysis Models for Polymer Composites Across Different Length Scales 203
but it can be extremely time and resource consuming as it requires specific software
and hardware for image acquisition and processing, and material to be analyzed.
To overcome these problems, a new algorithm to generate random distributions
of fibers, based on the hard-core model [24], has been proposed [13]. This algorithm,
called RAND_uSTRU_GEN (Random Microstructure Generator), introduces two
heuristics that allow the generation of RVEs with high fiber volume fractions in
a relatively short amount of time. Consequently, each iteration of the algorithm is
composed of three steps, where the first is based upon the classical hard-core model,
followed consecutively by the two new heuristics.
Because the algorithm was developed to generate fiber distributions for micro-
mechanical Finite Element Analyses (FEA), a condition of material periodicity is
imposed along opposite edges of the RVE [13] for implementation of Periodic
Boundary Conditions (PBCs). A minimum distance between fibers is also imposed
to avoid regions of poor mesh quality between the fibers when generating the Finite
Element (FE) model [13]. Figure 9.3 shows three examples of RVEs generated with
the algorithm RAND_uSTRU_GEN, with fiber volume fractions of 56, 60, and 65 %.
In [8, 13], it was shown that the algorithm RAND_uSTRU_GEN is faster than
alternative methodologies to model random fiber distributions [25, 27], even for
fiber volume fractions as high as 65 %. Moreover, a statistical characterization of the
spatial distribution of fibers was performed based on different statistical descriptors
[8, 13]. Because it is an almost purely random algorithm, the coefficients of variation
of the areas of the Voronoi polygons1 and distances to neighboring fibers2 tend to
be larger than in alternative models (e.g., [25]), in particular for high fiber volume
fractions. Moreover, the distributions generated with the proposed algorithm follow
closely the perfectly random Poisson distribution [13]. This was assessed by means
of the Ripley’s K function [30], which provides insight about the spatial pattern
at several distances [13, 24], and by means of the pair distribution function [31],
which describes the intensity of fiber distances [13]. Analysis of the Cumulative
Distribution Function of the orientation of the nearest neighbor3 shows that the
1
In a region with a set of points, it is possible to define a Voronoi polygon as the sub-region that
is closer to a given point than to any other. The set of Voronoi polygons defines a subdivision of
this region, known as Dirichlet tessellation [29]. The standard deviation of the areas of the Voronoi
polygons can be used as a measure of the periodicity of the distribution of fibers [13]—in a periodic
distribution all Voronoi polygons are equal, and the standard deviation of the areas is zero.
2
The neighboring fibers are defined as the fibers that share the sides of the Voronoi polygon of
the fiber of interest [13]. The standard deviation of the distances to neighboring fibers provide a
measure of how separate from each other the fibers are [13]. Like the standard deviation of the
areas of the Voronoi polygons, the standard deviation of the distances to neighboring fibers in a
periodic distribution is zero.
3
The Cumulative Distribution Function of the orientation of the nearest neighbor represents the
total number of fibers that have the nearest neighbor oriented along a certain direction. For a
perfectly random spatial arrangement, given by the Poisson distribution, this statistical descriptor
follows a straight diagonal line meaning that a given orientation has the same probability of
occurring as any other orientation. Deviations from this line correspond to the existence of
preferred orientations, as in periodic distributions.
204 P.P. Camanho and A. Arteiro
a b
Fig. 9.3 Examples of RVEs generated with the algorithm RAND_uSTRU_GEN (after Melro [8]).
(a) Fiber volume fraction of 56 %. (b) Fiber volume fraction of 60 %. (c) Fiber volume fraction
of 65 %
Figure 9.4 shows the transverse elastic properties obtained from 250 random
spatial fiber distributions with a requested fiber volume fraction of 60 %, generated
using the algorithm RAND_uSTRU_GEN [13], where E22 and E33 are, respectively,
the transverse in-plane and out-of-plane Young’s moduli, G23 in the transverse shear
modulus and 23 and 32 are the transverse Poisson’s ratios. A good agreement
between independently calculated material properties, analyzed by applying differ-
ent loading conditions, can be observed.
Definition of the most appropriate constitutive model for the polymer resins is a
challenging subject [18, 34]. In fact, polymer resins can exhibit a strong nonlinear
response under general loading scenarios, which must be accurately captured
to ensure that reliable analysis of its deformation can be performed. Moreover,
temperature, chemical ageing, and strain-rate are known to significantly affect
the mechanical response of these materials, and their effect on the deformation
and damage needs to be accounted for in the case of general thermo-mechanical
analyses.
The plastic deformation in polymer resins has been often modeled using either
the Drucker–Prager model or the Mohr–Coulomb model. However, experimental
evidence reveals that neither of these models are able to properly capture the
constitutive behavior of epoxy resins [8, 34]. Hence, for an accurate representation
of the nonlinear response of typical epoxy resins, an elasto-plastic constitutive
model based on a paraboloidal yield criterion has been proposed [34].
206 P.P. Camanho and A. Arteiro
Fig. 9.4 Transverse elastic properties obtained from 250 random spatial fiber distributions with
a requested fiber volume fraction of 60 %, generated using the algorithm RAND_uSTRU_GEN
[13]. Isotropic E-glass fibers in an MY750/HY917/DY063 epoxy matrix were considered [32].
For the analyses, Abaqus’ generalized plane strain three-node linear elements CPEG3 [33] were
used, and PBCs were applied [13]. The corresponding loading conditions are depicted next to each
histogram [reprinted from Composites Science and Technology, Vol 68, A.R. Melro, P.P. Camanho,
S.T. Pinho, Generation of random distribution of fibres in long-fibre reinforced composites, Pages
No. 2092–2102, © Elsevier (2008), with permission from Elsevier]
9 Analysis Models for Polymer Composites Across Different Length Scales 207
The initial elastic behavior is defined by a linear relation between the stress
tensor, , and the elastic strain, "e :
D De W " e (9.1)
where De is the standard isotropic elasticity fourth order tensor. The paraboloidal
yield criterion, defined as a function of the stress tensor and of the tensile and
compressive yield strengths, ym
t
and ym
c
, respectively, can be written as:
c
ˆ D 6J2 C 2I1 ym ym
t
2ym
t
ym
c
(9.2)
9 1 2mp
˛D (9.4)
2 1 C mp
where mp is the plastic Poisson’s ratio of the matrix. The resulting incremental form
of the flow rule is then defined, in tensorial notation, as [34]:
2
" D
p
3S C ˛I1 I (9.5)
9
where k is a constant that varies with the yield criterion, defined ensuring that the
equivalent plastic strain is equal to that obtained in a simple uniaxial test. In the case
of the paraboloidal yield criterion, k is obtained as [34]:
208 P.P. Camanho and A. Arteiro
1
kD p 2 (9.7)
1 C 2 m
where ymn
are the predicted yield stresses at a given strain-rate "Ppe and temperature
T, and ym are the yield stresses at the reference strain-rate "Pp
n
e and reference
temperature T . The material constants ˛n and ˇn are determined experimentally
by scaling back to the reference state each stress–strain curve that is obtained for
the different strain-rates and temperatures. Script n D .t; c/ refers, respectively, to
tension and compression. It is noted that the scaling laws proposed by Bai et al. [18]
are used to scale the tensile and compressive yield strengths independently so that
accurate predictions for general stress states can be obtained.
Since it is fully differentiable, the paraboloidal yield function can be integrated
by means of a general return mapping algorithm [34]. The details of the integration
of the constitutive model are presented in [34].
Damage onset is defined by a damage activation function similar to the
paraboloidal yield criterion, but using the final tensile and compressive strengths
of the epoxy resin (Xmc and Xmt , respectively) instead of the yield strengths, and
using the concept of effective stress tensor, i.e., the stress tensor calculated using
the undamaged stiffness tensor [34]:
1
Q D H0m W" (9.9)
Fmd D d
m rm 0 (9.10)
where md is the loading function and rm is an internal variable related with the
damage variable: while the material is in an undamaged condition, dm D 0 and
rm D 1; once damage is activated, 0 < dm < 1 and rm > 1; when the material is
completely damaged, dm D 1 and rm will tend to infinity. The loading function is
defined as [34]:
3JQ2 C IQ1 Xmc Xmt
d
m D (9.11)
Xmt Xmc
where the invariants JQ 2 and IQ1 are determined using the effective stress tensor Q
[Eq. (9.9)]. In the elasto-plastic regime, the damage activation function is negative,
Fmd < 0; when the damage criterion is activated, the condition Fmd D 0 must be
satisfied. The internal variable is defined as [34]:
9 Analysis Models for Polymer Composites Across Different Length Scales 209
n ˚ o
rm D max 1; max d
m;t (9.12)
t!1
The relation between the internal variable rm and the damage variable dm is given
by the damage evolution law, which must respect the two boundaries imposed by
the values of the damage variable before the damage criterion is activated (dm D 0)
and when the material is fully damaged (dm D 1). In the present damage model, the
following damage evolution law is proposed [34]:
p
A 3 7C2rm 2
e m
dm D 1 p (9.13)
7 C 2rm2 2
The parameter Am needs to be determined by means of the regularization of the
computed dissipated energy [34]. To avoid damage localization (mesh size depen-
dency), an approach based on the crack band model [35] is applied, which makes
use of the characteristic length of the finite element (l ) and the fracture toughness
of the epoxy resin (Gcm ) to regularize the computed dissipated energy [34]:
Z 1
@Gm @dm Gcm
‰m D drm D (9.14)
1 @d m @r m l
where E11f and E22f are the longitudinal and transverse Young’s moduli, G12f and
G23f are the longitudinal and transverse shear moduli, 12f and 23f are the Poisson’s
ratios, and df is the damage variable of the fibers. To ensure that the damage process
210 P.P. Camanho and A. Arteiro
is irreversible, the rate of change of the complementary free energy density must be
greater than the externally applied stresses:
@Gf @Gf P
GPf P W " D " W P C df 0 (9.16)
@ @df
@Gf
"D (9.17)
@
The strain tensor is defined using engineering shear strains. The stiffness tensor, Cf ,
@2 G f
can be obtained by inversion of the compliance tensor Hf D @ 2
[8]:
2 3
E11f .1ı 2 / E22f 12f .1df /.1Cı/ E22f 12f .1df /.1Cı/ 0 0 0
6 E22f Œ1 .1df / E22f .1df /.23f C /
7
6 0 0 0 7
6 7
1 df 6
6
E22f Œ1 .1df / 0 0 0 7
7
Cf D 6 G12f 0 0 7
6 7
6 0 7
4 G12f
E22f
5
sym:
2.1C23f /
(9.18)
with [8]:
ı D 23f 1 df (9.19)
D 12f 21f 1 df (9.20)
D .1 C ı/ 1 ı 2 1 df (9.21)
Ffd D d
f rf 0 (9.22)
d
where f and rf are, respectively, the loading function and the internal variable [8]:
Q 11
d
f D (9.23)
Xft
n ˚ o
rf D max 1; max d
f ;t (9.24)
t!1
9 Analysis Models for Polymer Composites Across Different Length Scales 211
where Xft is the tensile strength of the fibers. To account for the stochastic
nature of the tensile strength of the fibers, a random strength distribution can be
assigned to the elements that represent the fibers by generating random numbers (X)
between 0 and 1 and using the Weibull distribution to calculate the random tensile
strength [41]:
1=m
L0
Xft D 0 ln.1 X/ (9.25)
L
where L is the length of the fiber and 0 and m are, respectively, the Weibull scale
and shape parameters at the characteristic length L0 .
According to Eq. (9.23), damage is activated when the maximum stress criterion
is satisfied in the longitudinal direction only. It is also noted that the loading function
d
f [Eq. (9.23)] is defined as a function of the effective stress component along the
longitudinal direction, Q 11 . The effective stress tensor Q is defined as:
1
Q D H0f W" (9.26)
where H0f is the compliance tensor of the undamaged material (df D 0).
To avoid damage localization and to control the energy dissipated in the fracture
process, the crack band model [35] is implemented to regularize the computed
dissipated energy [8]:
Z 1
@Gf @df Gcf
‰f D drf D (9.27)
1 @df @rf l
eAf .1rf /
df D 1 (9.28)
rf
where Af must be computed for each finite element by solving Eq. (9.27).
9.2.2.3 Interface
The interface between fibers and matrix can be represented using a cohesive
formulation [12, 14, 15, 17, 38, 40, 41]. This is typically defined by a bilinear
traction–separation law, which relates the displacement jump across the interface
with the tractions acting on it (Fig. 9.5).
Before damage onset, the response of the interface is considered linear-elastic,
with a high initial stiffness (K). The onset of cohesive damage is assumed mode
dependent, and defined by the corresponding strengths in mode I (30 ) and mode II
212 P.P. Camanho and A. Arteiro
Fig. 9.5 Bilinear traction–separation laws under mode I and mode II loading [43] [reprinted from
Engineering Fracture Mechanics, Vol 74, A. Turon, C.G. Dávila, P.P. Camanho, J. Costa, An
engineering solution for mesh size effects in the simulation of delamination using cohesive zone
models, Pages No. 1665–1682, © Elsevier (2007), with permission from Elsevier]
(10 D 20 ). A quadratic nominal stress criterion can be used to predict damage
initiation. According to this criterion, damage is assumed to initiate when a
quadratic interaction function involving the nominal stress ratios reaches a value
of one, i.e., [42]:
2 2 2
1 2 h3 i
C C D1 (9.29)
10 20 30
where 1 and 2 are the tangential components and 3 is the normal component of the
traction tensor, and h i stands for the Macaulay brackets, which return the argument
if positive and zero otherwise.
Interface damage evolution is controlled by a linear softening law. A scalar
damage variable, d, is used to represent the interface damage. It evolves from 0
in the absence of damage to 1 when the interface is fully damaged. The softening
law (Fig. 9.5) can be expressed as [42]:
1 D K.1 d/!1
2 D K.1 d/!2
K.1 d/!3 if !3 0
3 D
K!3 otherwise
where !1 and !2 are the shear components and !3 is the normal component of
the displacement jump. To describe the evolution of damage under a combination
of normal and shear deformation across the interface, an effective displacement
jump, !m , can be defined as the norm of the displacement jump vector across the
interface [42]:
9 Analysis Models for Polymer Composites Across Different Length Scales 213
q
!m D !12 C !22 C h!3 i2 (9.31)
where GT D GI C GII C GIII and Gshear D GII C GIII are, respectively, the total and
the shear contribution of the Energy Release Rate (ERR), is a material parameter,
and Gc is the critical value of the total ERR. The evolution of the damage variable is
then defined as a function of the maximum value of the effective displacement jump
!mmax attained during the loading history [42]:
!mf !mmax !m0
dD (9.33)
f
!mmax !m !m0
where !m0 is the effective displacement jump at the initiation of damage and !mf is
the effective displacement jump at complete failure, given as [42]:
2Gc
!mf D 0
(9.34)
eff
0
where eff is the effective traction at damage initiation, given by the norm of the
traction tensor:
q 2 ˝ ˛2
0 2
eff D 10 C 20 C 30 (9.35)
It is noted that the values of the interface fracture toughness must be considered
rather low. This assumption is justified not only by experimental evidence [45],
but also by previous micro-mechanical numerical analyses [17, 38, 40], which have
demonstrated that the brittle behavior in transverse tension typical of composite
laminates is only captured for low values of the interface fracture toughness.
The range of stress states that can be imposed in composite materials by means of
experimental tests is limited by the complexity of the load introduction systems.
But micro-mechanical models, conceptually, allow the virtual identification and
214 P.P. Camanho and A. Arteiro
Fig. 9.6 Examples of micro-mechanical models of an FRP under different loading conditions
and respective equivalent plastic strain fields in the matrix [17] [reprinted from International
Journal of Solids and Structures, Vol 50, A.R. Melro, P.P. Camanho, F.M. Andrade Pires, S.T.
Pinho, Micromechanical analysis of polymer composites reinforced by unidirectional fibres: Part
II – Micromechanical analyses, Pages No. 1906–1915, © Elsevier (2013), with permission from
Elsevier]
Fig. 9.7 22 –33 failure envelope—micro-mechanics versus analytical failure criteria [37, 47]
Fig. 9.8 22 –12 failure envelope—micro-mechanics versus analytical failure criteria [37, 47]
Fig. 9.9 22 –13 failure envelope—micro-mechanics versus analytical failure criteria [37, 47]
Fig. 9.10 22 –23 failure envelope—micro-mechanics versus analytical failure criteria [37, 47]
Fig. 9.11 12 –23 failure envelope—micro-mechanics versus analytical failure criteria [37, 47]
using the computational micro-mechanics predictions for the pure uniaxial and
biaxial stress states. Table 9.2 shows the computed material properties of the com-
posite system predicted by computational micro-mechanics, where YT and YBT
are, respectively, the uniaxial and biaxial transverse tensile strengths, YC and
YBC are, respectively, the uniaxial and biaxial transverse compressive strengths, and
SL is the longitudinal shear strength [37, 47].
9 Analysis Models for Polymer Composites Across Different Length Scales 217
Fig. 9.12 13 –23 failure envelope—micro-mechanics versus analytical failure criteria [37, 47]
Fig. 9.13 RVE of a laminate with a discrete 0.020 mm thick 90ı ply [39] [reprinted from
Composites Part A: Applied Science and Manufacturing, Vol 79, A. Arteiro, G. Catalanotti, A.R.
Melro, P. Linde, P.P. Camanho, Micro-mechanical analysis of the effect of ply thickness on the
transverse compressive strength of polymer composites, Pages No. 127–137, © Elsevier (2015),
with permission from Elsevier]
composite systems [13, 34], in this case the full thickness of the transverse ply is
explicitly represented, so its effect in the response of the sublaminate can be taken
into account. The faces of the transverse ply perpendicular to the z-direction (top and
bottom faces) will be connected to the homogenized outer plies, and, therefore, in
the generation of the RVE, fibers are not allowed to intersect these faces. These
homogenized 0ı plies are intended to simulate the mesoscopic elastic behavior
of the surrounding laminae and its effect on the mechanical response and on the
damage initiation and growth in the mid-90ı ply.
The epoxy matrix is modeled using the elasto-plastic constitutive model with
damage presented in Sect. 9.2.2.1, and the interface between fibers and matrix
is modeled using a cohesive damage formulation defined by a bilinear traction–
separation law (Sect. 9.2.2.3). A simple transversely isotropic, linear-elastic con-
stitutive model is used to simulate the individual reinforcing carbon fibers, whose
diameter is considered constant throughout the transverse ply. The outer plies of the
cross-ply sublaminate are modeled assuming a linear-elastic transversely isotropic
material behavior; because the homogenized outer plies are introduced in the micro-
mechanical model to assess the effect of the stiffness of the adjacent plies in the
mechanical response and damage evolution of the intermediate 90ı lamina, no
nonlinear or fracture behaviors are considered. The elastic properties of a typical
carbon-epoxy system [73–75] are used to define the elastic response of the outer
plies of the proposed RVE. The interfaces between the mid-90ı lamina and the outer
homogenized plies are modeled using a cohesive damage formulation defined by a
mode dependent bilinear traction–separation damage law to predict delamination
[76]. Table 9.3 shows the material properties of all constituents of the sublaminate
RVE used in the in situ simulations.
220 P.P. Camanho and A. Arteiro
Fig. 9.14 Contour plots of the matrix damage variable on representative thin-ply RVEs at an
applied tensile remote strain of 2.0 % (only the 90ı plies are presented)
9.2.4.1 Tension
Figures 9.14 and 9.15 show the contour plots of the matrix damage variable on
representative RVEs of 90ı plies with thicknesses between 0.020 mm and 0.140 mm
subjected to transverse tensile remote strains (y-direction) of 1.2 % or 2.0 %.
Although only one random distribution of fibers is presented for each case, it is
representative of other random distributions in terms of damage localization.
As expected, a crack develops perpendicularly to the applied load. Similarly to
what was observed experimentally from in situ observations [58], damage starts
in the narrower portions of matrix between the closest adjacent fibers aligned
with the loading direction at approximately the same applied remote strain, first
9 Analysis Models for Polymer Composites Across Different Length Scales 223
It is possible now to compare the results of the micro-mechanical model with the
predictions from the analytical models for the in situ effect based on Linear Elastic
Fracture Mechanics (LEFM) [56]. Figure 9.16 shows the in situ transverse tensile
strength as function of the ply thickness determined from the micro-mechanical
models presented in this work. These in situ strengths are calculated from the
applied remote strain corresponding to the development of a damage damage
localization zone (for a value of the damage variable greater than 0.99) through
the thickness of the 90ı ply before the onset of secondary failure mechanisms (such
as interlaminar damage localization, oblique cracking, or dispersed/non-localized
matrix failure) and from the elastic properties determined from a linear-elastic
analysis of the same RVEs. Figure 9.16 also shows the predictions of the in situ
transverse tensile strength calculated using the fracture mechanics model proposed
by Camanho et al. [56]. The strength and elastic properties required by the analytical
model were determined from a separate run of the micro-mechanical model using
UD RVEs, as in Sect. 9.2.3. As can be observed, the same trends for the in situ
effect have been obtained using the analytical and computational micro-mechanics
representations.
9.2.4.2 Compression
Figures 9.17 and 9.18 show the contour plots of the equivalent plastic strain in the
matrix of representative RVEs subjected to a transverse compressive remote strain
(y-direction) of 2.5 %. For cross-ply sublaminates with thick 90ı plies (Fig. 9.18),
failure caused by transverse compressive loading is dominated by fiber-matrix
interface cracking and large localized plastic deformation of the matrix, forming
a localized band of damage in a plane not aligned with the loading direction. This
wedge shaped transverse fracture is in agreement with what has been described
9 Analysis Models for Polymer Composites Across Different Length Scales 225
Fig. 9.17 Contour plots of the equivalent plastic strain in the matrix of representative thin-ply
RVEs at an applied compressive remote strain of 2.5 % (only the 90ı plies are presented)
in the literature [17, 64]. In the case of cross-ply sublaminates with thin plies
(Fig. 9.17), a similar damage pattern can be identified, but developing at higher
applied remote strains. In fact, compared with the thick plies (Fig. 9.18), a transverse
crack suppression effect can be clearly identified. As the ply thickness decreases,
damage progression becomes more and more gradual, and through-the-thickness
transverse fracture is delayed, indicating that an in situ effect exists in transverse
compression.
Figure 9.19 shows the in situ transverse compressive strength as a function of
the ply thickness determined from the micro-mechanical models presented in this
work and the predictions from the analytical models for the in situ effect based on
phenomenological 3D failure criteria [71, 72]. Several RVEs have been analyzed,
and the respective data has been plotted in Fig. 9.19.
The in situ strengths determined from the micro-mechanical models (Fig. 9.19)
were calculated based on the applied remote strain corresponding to the devel-
opment of through-the-thickness ply failure, characterized by a localized band
of damage, and based on the elastic properties determined from a linear-elastic
analysis of the same RVE. It is noted that, in the case of the thinner sublaminates,
the development of the first through-the-thickness damage localization band is
delayed, and other regions of localized damage start propagating before failure of
226 P.P. Camanho and A. Arteiro
the thin transverse ply have occurred. Unless one of these damage bands penetrates
completely through the ply thickness, the transverse ply is not considered to have
failed. As the ply thickness decreases, the number of localized bands increases,
but one eventually will be the first to penetrate through the thickness. That is the
point that defines the strength of the thinner transverse plies. It is interesting to
note that, once the first band penetrates through the thickness, the remaining will
also penetrate progressively. The strength and elastic properties required by the
analytical models were determined from a separate analysis of the micro-mechanical
model using UD RVEs.
Comparing the predictions of the in situ effect for transverse compression with
the results obtained using computational micro-mechanics (Fig. 9.19), similar trends
are observed, demonstrating that the analytical models based on phenomenological
3D failure criteria proposed in [71, 72] are accurate.
9 Analysis Models for Polymer Composites Across Different Length Scales 227
Fig. 9.20 Failure process in an AS4 carbon fiber obtained in the analysis of a non-hybrid carbon-
epoxy system subjected to a longitudinal tensile load: (a) distribution of fiber strength, (b) stress
distribution prior to fiber failure, (c) stress redistribution immediately after fiber failure, and
(d) damage at the fiber-matrix interface after fiber failure [41] [reprinted from Computational
Mechanics, Vol 57, R.P. Tavares, A.R. Melro, M.A. Bessa, A. Turon, W.K. Liu, P.P. Camanho,
Mechanics of hybrid polymer composites: analytical and computational study, Pages No. 405–421,
© Springer-Verlag Berlin Heidelberg (2016), with permission of Springer]
9 Analysis Models for Polymer Composites Across Different Length Scales 231
Fig. 9.21 Matrix damage and stress concentrations in the surrounding fibers after fiber failure
[41] [reprinted from Computational Mechanics, Vol 57, R.P. Tavares, A.R. Melro, M.A. Bessa,
A. Turon, W.K. Liu, P.P. Camanho, Mechanics of hybrid polymer composites: analytical and com-
putational study, Pages No. 405–421, © Springer-Verlag Berlin Heidelberg (2016), with permission
of Springer.] (a) Matrix damage. (b) Stress concentrations in the intact fibers surrounding a broken
fiber (in black)
232 P.P. Camanho and A. Arteiro
Fig. 9.22 Location of fiber breaks (in black) after failure of the composite [41] [reprinted from
Computational Mechanics, Vol 57, R.P. Tavares, A.R. Melro, M.A. Bessa, A. Turon, W.K. Liu, P.P.
Camanho, Mechanics of hybrid polymer composites: analytical and computational study, Pages
No. 405–421, © Springer-Verlag Berlin Heidelberg (2016), with permission of Springer]
Fig. 9.23 Stress–strain relations of AS4-M50S hybrid carbon fiber-epoxy composites subjected
to a longitudinal tensile load [41] [reprinted from Computational Mechanics, Vol 57, R.P. Tavares,
A.R. Melro, M.A. Bessa, A. Turon, W.K. Liu, P.P. Camanho, Mechanics of hybrid polymer
composites: analytical and computational study, Pages No. 405–421, © Springer-Verlag Berlin
Heidelberg (2016), with permission of Springer]
9 Analysis Models for Polymer Composites Across Different Length Scales 233
Fig. 9.24 Stress–strain relation and microstructure of an AS4-M50S hybrid carbon fiber-epoxy
composite with a volume fraction of M50S fibers of 0.25 subjected to a longitudinal tensile
load [41] [reprinted from Computational Mechanics, Vol 57, R.P. Tavares, A.R. Melro, M.A.
Bessa, A. Turon, W.K. Liu, P.P. Camanho, Mechanics of hybrid polymer composites: analytical
and computational study, Pages No. 405–421, © Springer-Verlag Berlin Heidelberg (2016), with
permission of Springer]
This causes an increase of the applied stress after the first load drop. At the second
peak stress, the failure strain of the AS4 fibers has been reached, causing the
failure of the composite material. This tensile response is close to what is generally
described as pseudo-ductility, in spite of the moderate load drop after failure of the
M50S fibers.
To reduce the load drop observed on the AS4-M50S hybrid composite, the AS4
fibers are now combined with different volume fractions of T300 carbon fibers, as,
unlike the M50S fibers, the failure strain distribution of the T300 carbon fibers is
closer to that of the AS4 carbon fibers (Table 9.4). Figure 9.25 shows the stress–
strain curves for the AS4-T300 hybrid composite.
For the hybrid systems with volume fractions of T300 fibers of 0.5 and 0.75 there
is a delay in first fiber failure in comparison to the non-hybrid T300 composite,
which occurs at a strain of 0.7 %. Moreover, no major stress drop is observed due to
failure of the low-elongation fibers in these hybrids. This is attributed to the higher
dispersion in fiber strength of the T300 carbon fibers, which causes them to fail
at different strains, resulting in a more gradual failure. This can be observed in
the microstructures shown in Fig. 9.26 for the AS4-T300 hybrid composite with a
volume fraction of T300 fibers of 0.5. It is also possible to see that not all low-
elongation fibers (T300 fibers) fail before the AS4 (high-elongation) fibers start to
fail, resulting in a gradual failure response.
234 P.P. Camanho and A. Arteiro
Fig. 9.25 Stress–strain relations of AS4-T300 hybrid carbon fiber-epoxy composites subjected to
a longitudinal tensile load [41] [reprinted from Computational Mechanics, Vol 57, R.P. Tavares,
A.R. Melro, M.A. Bessa, A. Turon, W.K. Liu, P.P. Camanho, Mechanics of hybrid polymer
composites: analytical and computational study, Pages No. 405–421, © Springer-Verlag Berlin
Heidelberg (2016), with permission of Springer]
Similar results are obtained for the hybrid composite with a volume fraction
of T300 carbon fibers of 0.75. The corresponding stress–strain curve and the
microstructures are shown in Fig. 9.27. As before, the high-elongation AS4 fibers
fail prior to complete failure of the low-elongation T300 fibers, which is apparently
key to the achievement of pseudo-ductility.
Finally, it is noted that the proposed micro-mechanical framework, which takes
the fiber strength variability into account, is able to capture the main failure
mechanisms in UD composite materials and establish the failure sequence in UD
composites subjected to longitudinal tensile loading. Moreover, its application to
material design through hybridization demonstrates the potential of reliable compu-
tational micro-mechanics frameworks as useful tools to improve the properties and
the performance of composite materials.
Fig. 9.26 Stress–strain relation and microstructure of an AS4-T300 hybrid carbon fiber-epoxy
composite with a volume fraction of T300 fibers of 0.5 subjected to a longitudinal tensile load
[41] [reprinted from Computational Mechanics, Vol 57, R.P. Tavares, A.R. Melro, M.A. Bessa,
A. Turon, W.K. Liu, P.P. Camanho, Mechanics of hybrid polymer composites: analytical and
computational study, Pages No. 405–421, © Springer-Verlag Berlin Heidelberg (2016), with
permission of Springer]
irreversible nonlinearities, such as plasticity, damage, and fracture. At the ply level,
due to the morphology of FRP composite systems, these nonlinearities are translated
into matrix-dominated intralaminar fracture parallel to the fiber direction, either
as transverse cracking or split cracking, and fiber-dominated intralaminar fracture
perpendicular to the fibers.
But the interaction and extent of the different ply damage mechanisms depend
not only on the constituents of the composite system, but also on the reinforcement
configuration, stacking sequence, and ply thickness. In the analysis of the damage
evolution and ultimate failure of representative composite laminates used in real
coupons and structures, a fine description of the discrete damage events by means
of computational micro-mechanics invariably results in untractable, prohibitively
complex models that cannot be solved using the computational power currently
available in a time frame that serves the needs of the industry. Therefore, the
detailed representation of damage and failure of composite laminates must rely on
phenomenological, physically based models formulated at the meso-scale, which
must account for both initiation and propagation of the different ply damage
mechanisms.
Meso-models take the laminate layers or sublaminates as the basic homogeneous
building blocks and use the orthotropic material properties related to the directions
236 P.P. Camanho and A. Arteiro
Fig. 9.27 Stress–strain relation and microstructure of an AS4-T300 hybrid carbon fiber-epoxy
composite with a volume fraction of T300 fibers of 0.75 subjected to a longitudinal tensile
load [41] [reprinted from Computational Mechanics, Vol 57, R.P. Tavares, A.R. Melro, M.A.
Bessa, A. Turon, W.K. Liu, P.P. Camanho, Mechanics of hybrid polymer composites: analytical
and computational study, Pages No. 405–421, © Springer-Verlag Berlin Heidelberg (2016), with
permission of Springer]
of the fibers to predict the behavior of the whole laminate. The material properties
can be determined from characterization tests at the lamina level, or from micro-
mechanical models based on adequate homogenization techniques.
Here, a Smeared Crack Model (SCM) is proposed to model transverse fracture
of polymer composites [70]. With a formulation based on Continuum Damage
Mechanics (CDM), the proposed SCM assumes that each ply is composed of a
homogeneous material. Transverse damage localization is modeled explicitly by
means of a cohesive law, taking into account the orientation of the transverse crack,
which depends on the 3D stress state imposed to individual ply. Transverse cracking
is therefore represented as a “soft discontinuity” [4], modeled as a gradual, local loss
of material integrity. The proposed SCM can be easily implemented in commercial
FE packages as a user-defined material since no intricate coding is required to
take into account any additional degrees-of-freedom, special discretization, or
re-meshing techniques. Here the proposed SCM is implemented in an explicit
dynamics analysis procedure in the FE package Abaqus/Explicit [42] as a user
material subroutine.
Because the SCM is based on a local CDM formulation, the lack of numerical
objectivity needs to be addressed. This is done ensuring proper energy dissipation,
following the crack band approach [35]. The characteristic length of the finite
element, which represents the crack band width, is used to regularize the computed
9 Analysis Models for Polymer Composites Across Different Length Scales 237
The onset of transverse fracture and the orientation of the fracture plane, which
are functions of the stress state and material properties, are predicted using a
3D invariant-based failure criterion for transverse damage mechanisms [37, 72]. For
transverse failure of UD fiber-reinforced laminates, the transversely isotropic failure
criterion reads
with:
The set of transversely isotropic invariants used to formulate the failure criterion
represented by Eq. (9.36) reads [84]:
1
I1 D tr . p /2 a . p /2 a (9.38a)
2
I2 D a . p /2 a (9.38b)
I3 D tr a a (9.38c)
D p C r (9.39)
9 Analysis Models for Polymer Composites Across Different Length Scales 239
with:
1 1
r D .tr a a/I .tr 3a a/A (9.40a)
2 2
p D r (9.40b)
1
˛1 D (9.41a)
ST2
1
˛2 D (9.41b)
SL2
YT Y2
1 ˛1 T
2YBT 4
˛32
t
D (9.41c)
YT2 2YBT YT
1
˛3t D 2˛32
t
YBT (9.41d)
2YBT
YC Y2
1 ˛1 C
2YBC 4
˛32
c
D (9.41e)
YC2 2YBC YC
1
˛3c D 2˛32
c
YBC (9.41f)
2YBC
240 P.P. Camanho and A. Arteiro
where ST and SL are, respectively, the transverse and longitudinal shear strengths, YT
and YC are, respectively, the transverse tensile and compressive strengths, and YBT
and YBC are, respectively, the biaxial transverse tensile and compressive strengths.
Once the onset of transverse damage is predicted, i.e., when the condition
fM . / > 1 is satisfied [Eq. (9.36)], the inelastic deformation associated with damage
localization must be determined.
First, the intersection with the failure surface is defined, to ensure that transverse
cracking initiates when the condition fM . / D 1 holds. This is done employing
an iterative scheme based on the Newton–Raphson method [99]. Having defined
the conditions that trigger transverse failure mechanisms, a constitutive model that
simulates the propagation of transverse matrix cracks under general loading condi-
tions must be defined. Taking into account that, in such conditions, the orientation
of the fracture plane is not known in advance, it is not possible to use cohesive zone
models placed in predefined locations of the FE domain. Therefore, the use of an
SCM, based on a “fixed crack” formulation [82], is proposed [70]. The SCM uses
an additive decomposition of the strain tensor, which is divided into an elastic strain
and a cracking strain, and it is based on a cohesive law that relates the tractions
acting on the fracture plane with the corresponding displacement jumps across the
crack faces.
When the onset of transverse damage is predicted, fM . / > 1, the stress state
for such condition is unfeasible, because damage is assumed to initiate at
the intersection with the fracture surface, as shown in Fig. 9.28. For the sake of
simplicity, a biaxial stress state composed of nonzero 12 and 22 stress components
is considered. If, at a certain time increment, the stress state yields a point outside
the failure surface, and if such stress state is taken as the condition for the onset of
transverse failure mechanisms without returning to the original failure surface, the
latter (solid line) is replaced by a new failure surface (dashed line) that overpredicts
the onset of fracture. This may have a significant effect on the cohesive law
governing the propagation of matrix cracks (Fig. 9.29). Because the condition that
triggers the onset of transverse failure is overpredicted, and because the transverse
fracture energy is constant (material property), the equivalent displacement jump at
failure and, consequently, the inelastic deformation at failure, are underpredicted.
To represent accurately the conditions for the initiation of transverse cracks, it is
therefore necessary to locate the intersection with the failure surface, i.e., the stress
state at which fM D 1.
For a given elastic stress increment i at time increment i satisfying the failure
criterion (fM > 1), assuming that the intersection with the fracture surface can be
given by a linear interpolation between the previous ( i1 ) and the new (purely
elastic) stress state ( i D i1 C i ), valid for sufficiently small time (or stress)
increments (typically the case in FEA with explicit integration rules), the location
9 Analysis Models for Polymer Composites Across Different Length Scales 241
Fig. 9.29 Effect of determining the location of the intersection with the fracture surface on the
inelastic response
of the intersection of the elastic stress vector with the fracture surface can be
determined imposing the following condition [99]:
and solving it for c. In Eq. (9.42), the stress states i1 and i1 C i are such
that the following two conditions must be satisfied:
242 P.P. Camanho and A. Arteiro
with:
It is noted that Eq. (9.44) can also be expressed in terms of strain increments
(particularly useful in explicit FEA) as:
FjcD0
c.0/ D (9.47)
FjcD0 FjcD1
where FjcD0 and FjcD1 are, respectively, F . i1 / and F . i1 C i /. The scalar c
can be updated as:
FjcDc.k1/
c.k/ D c.k1/ ˇ ; k D 1; 2; : : : (9.48)
dF ˇˇ
dc ˇcDc.k1/
ˇ
dF ˇ
dc cDc.k1/ is the derivative of F with respect to c for c D c.k1/ . The derivative of
F, dF
dc , is given as:
Because in FRPs the fracture plane is always parallel to the fiber direction
(Fig. 9.30), the unit vector normal to the fracture plane, n2 , reads [71]:
where sgn f23 g is the sign of the transverse shear stress component 23 (i.e.,
sgn f23 g D 1 if 23 0, and sgn f23 g D 1 otherwise).
Observing Fig. 9.31, it becomes clear that Eq. (9.54) provides reasonable esti-
mates for the orientation of the fracture plane, particularly if taking into account
that the fracture angle for pure transverse compression [65, 71, 96] is never used to
obtain such estimates.
Because the proposed SCM is formulated for general 3D stress states, this prag-
matic approach should now be extended to other stress combinations. For instance,
when subjected to through-the-thickness transverse compression (33 < 0), a similar
equation to that presented for in-plane transverse compression (22 < 0) can be
deduced. In this case, assuming a transverse compression-dominated failure mode
(33 < 0 and j33 j ST ), the orientation of the fracture plane can be estimated as:
s
ST
˛ D sgn f23 g arcsin (9.55)
j33 j
In the case of biaxial transverse compression, it is assumed that, when j22 j > jYC j
or j33 j > jYC j, the orientation of the fracture angle remains constant. Thus,
Eqs. (9.54) and (9.55) can be rewritten, respectively, as:
s
ST
˛ D sgn f23 g arccos (9.56)
minfj22 j ; jYC jg
9 Analysis Models for Polymer Composites Across Different Length Scales 245
Fig. 9.31 Orientation of the fracture plane as a function of the off-axis angle of off-axis
compression tests performed by Koerber et al. [75], and a comparison with the results from the
proposed pragmatic approach used to estimate the fracture angle
and
s
ST
˛ D sgn f23 g arcsin (9.57)
minfj33 j ; jYC jg
For general transverse compressive failure (I3 < 0), it is assumed that the highest
transverse stress (in absolute value) will define the orientation of the fracture plane.
In other words, when I3 < 0, j22 j > j33 j and j22 j ST , Eq. (9.56) holds; on the
other hand, if I3 < 0, j33 j > j22 j and j33 j ST , Eq. (9.57) must be used.
When I3 < 0 but j22 j < ST and j33 j < ST , or when I3 > 0, failure is either
dominated by a transverse state of stress or by longitudinal shear. To assess which
failure mode is dominating, the ˇ contribution of each ˇ type of stressing to the failure
criterion can be used. Thus, if ˇ˛1 I1 C ˛3 I3 C ˛32 I32 ˇ > j˛2 I2 j, the fracture angle can
be estimated by the orientation of the plane of the secondary principal stresses:
1 22 33
˛D arctan (9.58)
2 223
otherwise, the orientation of the fracture angle can be estimated by the in-plane
shear effort:
22 33
˛ D arctan (9.59)
223
246 P.P. Camanho and A. Arteiro
Having defined the fracture angle, the traction tensor acting on the fracture plane
can now be calculated using the stress tensor and the unit vector n2 , normal to the
fracture plane [71]:
t D n2 (9.60)
The normal component, tN , and shear components, tL and tT , of the traction tensor t
are [71]:
tN D t n2 ; tL D t n1 ; tT D t .n1 n2 / (9.61)
The tractions can now be used to define the cohesive law established in terms of the
displacement jumps across the crack faces (!cr ) [70].
The SCM uses an additive decomposition of the strain tensor, where the total strain,
", is calculated as a function of the elastic strain, "e , and of the cracking strain,
"c , as:
where "cr cr cr cr
c is the cracking strain projected onto the crack coordinate system x1 x2 x3
(Fig. 9.30) and R is the rotation matrix that relates the global coordinate system
to that associated with the fracture plane, whose components are functions of the
fracture angle ˛.
The stress–strain relation reads
D De W ." "c / D De W " De W R "cr
c R
T
(9.66)
and the tractions acting on the fracture plane, projected onto the crack frame, can be
expressed as:
tcr D RT t D RT . n2 / (9.67)
9 Analysis Models for Polymer Composites Across Different Length Scales 247
The tractions tcr can also be obtained by an appropriate cohesive law, which can
be written as tcr D g.!cr /, or, using Eq. (9.67):
RT . n2 / D g .!cr / (9.68)
(9.69)
where ncri are the unit vectors acting on the fracture plane written in the crack
coordinate system. In the FE implementation of the SCM, the characteristic length
of the finite element, l , is taken here as the square root of the finite element’s in-
plane area.
To solve the system of nonlinear equations expressed by Eq. (9.68), a cohesive
law can be defined in the context of the SCM using a modification of the constitutive
model originally proposed by Turon et al. [102] to address delamination growth
under general loading conditions. The proposed constitutive model imposes a linear
softening cohesive law, defining the tractions ticr .d/ acting on the fracture plane as:
ˇ cr ˇ ˝ ˛ " ˇ cr ˇ #
1d ˇ! ˇ ! cr 1d ˇ! ˇ cr
ticr .d/ D i
Nticr ıi2 ˇ cr2ˇ i
Nticr E22 "22 "c22
cr
d !fm ˇ! ˇ d !fm
2
(9.70)
where d is a scalar damage variable, ıij is the Kroenecker delta, !fm is the equivalent
displacement jump at failure under mixed-mode loading conditions, !icr are the
scalar components of the displacement jump vector !cr , E22 is the transverse
Young’s modulus, "cr 22 and "c22 are, respectively, the scalar components of the total
cr
and cracking strain tensors "cr and "crc projected onto the crack coordinate system,
and h i are the Macaulay brackets.
The damage variable d is obtained from the loading function L .!cr / proposed
by Turon et al. [102] as:
with:
( )
L .! / D min
cr
;1 (9.72)
!fm
248 P.P. Camanho and A. Arteiro
with:
q
cr 2 cr 2
!scr D !1 C !3 (9.74)
2 .GIc C AB /
!fm D (9.75)
Ntcr
with A D GIIc GIc and B D Gshear =GT , where GIc and GIIc are, respectively,
the mode I and mode II intralaminar fracture toughness associated with transverse
failure mechanisms and Ntcr is the norm of the traction tensor at the onset of transverse
failure:
q
2 ˝ cr ˛2
Ntcr D Ntscr C Nt2 (9.76)
with:
q
cr 2 cr 2
Ntscr D Nt1 C Nt3 (9.77)
The mode ratio B can be calculated extracting the mode partitions of the total energy
R !cr
release rate using G D 0 tcr d!cr , which results in:
Ntscr ˇ
BD ˝ ˛ ˝ ˛ (9.78)
ˇ Ntscr Nt2cr C Nt2cr
with:
!scr
ˇD ˝ ˛ (9.79)
!scr C !2cr
The nonlinear system of equations defined by Eq. (9.68) can now be solved
for the cracking strain using a multi-variable Newton–Raphson method. Rewriting
Eq. (9.68), the residual for this system of nonlinear equations is given as:
cr cr
r "cr
c D f "c g " c (9.80)
with:
c D R De W " De W R "c R
f "cr T cr T
n2 (9.81)
ˇ
ˇ ˇ @r "cr ˇ 2
"cr ˇ Dr "cr ˇ C
c n
"cr
r c CO "cr D0 (9.82)
c nC1 c n
@"cr
c
c
with:
ˇ
cr ˇ
ˇ
cr ˇ
"cr
c D "c nC1 "c n (9.83)
@r. "cr j /
To calculate the Jacobian matrix of the system of nonlinear equations, @"ccr n ,
c
the derivatives of the first and second terms of the right-hand side of Eq. (9.80)
need to be determined. While the derivatives of the first term are straightforward,
the derivatives of the second term depend on the cohesive law [Eq. (9.70)], which
depends on the damage variable d and on the equivalent displacement jump at failure
!fm , which in turn depend on the equivalent displacement jump and on the mode
ratio B, both functions of the cracking strain tensor "cr c.
To begin the iterative procedure, and since damage occurs after the failure
envelope is reached, a suitable value for the initial cracking strain, "trialc , can be
defined by the current total strain rotated to the fracture plane, "cr , minus the strain
at damage initiation rotated to the fracture plane, "cr
init , i.e.:
"trial
c D "cr "cr
init (9.84)
After calculating the residual and the Jacobian matrix for each iteration, the
increments to the cracking strains can be obtained as:
ˇ !1
ˇ @r "cr ˇ
"cr D r "cr ˇ c n
(9.85)
c c n
@"cr
c
The new value for the cracking strains is obtained by adding the increment to the
cracking strains obtained in the previous iteration:
250 P.P. Camanho and A. Arteiro
ˇ ˇ
"cr ˇ cr ˇ
c nC1 D "c n C "cr
c (9.86)
The procedure is repeated until the convergence criterion is met. Having determined
the cracking strain, the stress tensor is easily obtained using Eq. (9.66).
where "11 and "T1 are, respectively, the strain component and the tensile strain-to-
failure in the fiber direction. If the tensile strain-to-failure in the fiber direction,
"T1 , is not known, assuming that the tensile behavior of the composite in the fiber
direction is linear-elastic until failure, Eq. (9.87) can also be written as:
E11
fF ."11 / D "11 1 (9.88)
XT
where E11 is the longitudinal Young’s modulus of the composite system and XT is
the longitudinal tensile strength of the UD ply.
For the onset of fiber kinking, it is assumed that longitudinal compressive failure
of FRPs occurs from the collapse of the fibers as a result of damage of the supporting
matrix [65, 95]. Because 3D stress states greatly affect the kinking failure mode
[71, 96], a 3D kinking model relying on an invariant failure criterion for transverse
fracture has been recently proposed [37].
This 3D kinking model assumes that fiber kinking is originated by local micro-
structural defects that trigger a kink band when local matrix cracking occurs in
the vicinity of the misaligned, micro-buckled fibers. Figure 9.32 shows the kinking
plane in the 3D kinking model, where the coordinate system x01 x02 x03 is aligned with
the material axes of the composite (x01 is the fiber direction and x02 and x03 are the
transverse directions). The kinking plane is associated with the coordinate system
9 Analysis Models for Polymer Composites Across Different Length Scales 251
x11 x12 x13 , obtained after rotating by an angle (angle of the kinking plane) the x01 x02 x03
frame around the x01 axis. The misaligned fibers are associated with the rotated
xR1 xR2 xR3 coordinate system, originated from the rotation of the x11 x12 x13 frame by an
angle ' (kinking-angle) around the x13 axis.
Based on the assumption that a kink band is triggered by local matrix cracking in
the vicinity of the misaligned fibers, the onset of fiber kinking can be predicted using
the invariant-based failure criterion for transverse failure mechanisms, formulated
in the misalignment frame:
where ˛3 and ˛32 depend on the sign of the linear invariant I3 [see Eq. (9.37)]. The
sets of invariants and failure parameters used in Eq. (9.89) are given, respectively, by
Eqs. (9.38) and (9.41). The former are based on a decomposition of the stress tensor
in plasticity inducing stresses p and reaction stresses r [Eq. (9.40)] with respect
to the preferred direction of the transversely isotropic material. In the formation of
a kink band, this preferred direction is the direction of the misaligned fibers (a.R/),
which coincides with the xR1 -direction (Fig. 9.32):
To express the preferred direction in the material coordinate system (a.0/ ), two
successive transformations are necessary, corresponding to the rotation of the x11 x12 x13
252 P.P. Camanho and A. Arteiro
Fig. 9.33 Kinking plane and involved shear stresses (after Catalanotti et al. [71])
frame by the angle ' around the x13 axis and to the rotation by an angle of the
x01 x02 x03 frame around the x01 axis (Fig. 9.32), yielding:
2 3
cos '
a.0/ D 4 cos sin ' 5 (9.91)
sin sin '
The angle of the kinking plane, , is the angle that maximizes the fiber kinking
failure index. It can be obtained applying the kinking failure criterion [Eq. (9.89)]
to a range of tentative angles i in the range 0 i < [96]. However, because
this is a numerically expensive solution, an alternative (pragmatic) expression for
has been derived based on the works of Pinho et al. [96] and Catalanotti et al. [71].
Assuming that is a function of the stress state, in particular of the shear stresses
that act on the 2–3 plane (Fig. 9.33), the angle of the kinking plane can be calculated
as [71]:
13
D arctan (9.92)
12
If both shear stress components 12 and 13 are equal to zero, the angle of
the kinking plane can be calculated by the maximum principal stress that acts on
the transversely isotropic plane [71]. Following the reasoning presented by Pinho
.1/
et al. [96], if the shear stress 23 in the kinking plane is nonzero, an out-of-plane
movement of the kinked fibers would occur, contradicting the evidence that fibers
.1/
kink on this plane (see Figs. 9.32 and 9.33). Therefore, imposing 23 D 0, the angle
of the kinking plane can be obtained as:
9 Analysis Models for Polymer Composites Across Different Length Scales 253
1 223
D arctan (9.93)
2 22 33
Following Catalanotti et al. [71], the kinking-angle ' can be obtained as:
n o
.1/
' D sgn 12 R (9.94)
n o
.1/
where sgn 12 is the sign of the shear stress component in the frame of the kinking
n o n o
.1/ .1/ .1/
plane (i.e., sgn 12 D 1 if 12 0, and sgn 12 D 1 otherwise) and R is
the rotation originated by the applied shear loading, which depends on the shear
constitutive law. Assuming linear shear behavior, Catalanotti et al. [71] suggested
the following relation:
.R/
12 D R (9.95)
.R/
where 12 is the shear stress in the misalignment frame and is a micro-mechanical
parameter associated with the creation of a kink band, which accounts for the
imperfections that trigger fiber kinking, such as local micro-structural defects from
manufacture or the intrinsic fiber volume fraction oscillations in the material. The
.R/
shear stress 12 in the misalignment frame is easily obtained by a transformation of
coordinate systems (Fig. 9.32), yielding:
.1/
11 D 11 (9.97a)
.1/
22 D 22 cos2 C 223 cos sin C 33 sin2 (9.97b)
.1/
12 D 12 cos C 13 sin (9.97c)
sin.2'C / XC
D (9.98)
2'C
where 'C is the misalignment angle at failure when pure longitudinal compression is
applied and XC is the longitudinal compressive strength. The former can be obtained
imposing a pure longitudinal compressive stress state (11 < 0 and ij D 0) and
254 P.P. Camanho and A. Arteiro
applying the failure criterion for fiber kinking [Eq. (9.89)]. At failure, 11 D XC ,
' D 'C and fK .11 D XC ; ij D 0/ D 1, and solving Eq. (9.89) for 'C yields
(" q
1 2
'C D arccos 4 ˛1 4˛2 C ˛22 XC2 C ˛3c C 2˛2 ˛3c XC C 4˛32
c
2
# )
1
C ˛1 C 4˛32 XC C 4˛3 ˛1 4˛2 C 4˛32 XC
c c c
(9.99)
1 .1/ ˇ ˇ
.1/ ˇ .1/ ˇ
F.R / D R C 11 22 sin .2R / ˇ12 ˇ cos .2R / (9.100)
2
whose derivative with respect to R is given as:
ˇ ˇ
dF .1/ .1/ ˇ .1/ ˇ
D C 11 22 cos .2R / C 2 ˇ12 ˇ sin .2R / (9.101)
dR
Using Eqs. (9.100)–(9.101), the angle R can be computed using the following
recursive formula:
F.Ri /
RiC1 D Ri ˇ (9.102)
dF ˇˇ
d ˇ
R R DRi
the R-curve are the lengths of the fracture process zone, lfpz , and the steady state
propagation values of the fracture toughness, G1C ss
for tensile fracture propagation
and G1 for kink band formation.
ss
To establish the constitutive law for longitudinal damage, the following definition
of the complementary free energy density in the material is proposed:
2
11 2 C 33
2
12 23
G D C 22 .11 22 C 11 33 / 22 33
2 .1 d1 / E11 2E22 E11 E22
(9.103)
2 C 13
2
2
C 12 C 23
2G12 2G23
where E11 and E22 are the longitudinal and transverse Young’s moduli, G12 and G23
are the longitudinal and transverse shear moduli and 12 and 23 are the Poisson’s
ratios of the transversely isotropic UD lamina, and d1 is the damage variable
associated with longitudinal (fiber) failure. To ensure the thermodynamically irre-
versibility of the damage process, the difference between the rate of change of the
complementary free energy density GP and the externally supplied work to the solid
at constant strains, P W ", must not be negative:
GP P W " 0 (9.104)
which, expanding in terms of the stress tensor and damage variable d1 , gives
@G @G P
" W P C d1 0 (9.105)
@ @d1
256 P.P. Camanho and A. Arteiro
@G
"D DHW (9.106)
@
The lamina compliance tensor H can be represented, in Voigt notation, as:
2 3
1 21 31
6 .1 d1 / E11 0 0 0 7
6 E22 E33 7
6 12 1 32 7
6 0 0 0 7
6 7
6 E11 E22 E33 7
6 23 1 7
6 13
0 0 0 7
@2 G 6 7
6 E33 E22 E33 7
HD D6 7 (9.107)
@ 2 6 1 7
6 0 0 0 0 0 7
6 G12 7
6 7
6 1 7
6 0 0 0 0 0 7
6 G12 7
6 7
4 1 5
0 0 0 0 0
G23
Closure effects under load reversal are taken into account by defining two damage
variables associated with longitudinal damage as follows [69]:
h11 i h11 i
d1 D d1C C d1 (9.108)
j11 j j11 j
where h i stands for the Macaulay brackets. The present model for longitudinal
failure tracks damage caused by tension loads (d1C ) separately from damage caused
by compression loads (d1 ), depending on the sign of the longitudinal stress
component 11 .
The damage activation functions associated with longitudinal failure in tension
and in compression are defined as [69]:
where the loading functions 1C and 1 are given, respectively, by the fail-
ure indices for longitudinal tension [Eq. (9.87)] and longitudinal compression
[Eq. (9.89)] calculated using the effective stress tensor Q D H1
0 W ", where H0
is the undamaged compliance tensor obtained from Eq. (9.107) using d1 D 0. The
internal variables r1C and r1 , which are related to the damage variables d1C and
d1 by the damage evolution laws, take an initial value of 1 when the material is
9 Analysis Models for Polymer Composites Across Different Length Scales 257
undamaged, and they increase with damage; in other words, they define the level of
elastic strains that can be attained before the accumulation of additional damage.
The evolution of the internal variables r1C and r1 is expressed by the Kuhn–
Tucker conditions [69]:
Neglecting viscous effects, the damage activation functions, Eqs. (9.109), always
have to be non-positive; while the damage activation functions F1C and F1 are
negative, the material response is elastic. When the strain state activates a criterion,
the corresponding damage activation function (F1C or F1 ) is equal to zero, and it is
necessary to evaluate the gradient of the loading function ( P1C or P 1 , respectively).
If the gradient is not positive, the state is one of unloading or neutral loading. If the
gradient is positive, there is damage evolution, and the consistency condition has to
be satisfied [69]:
FP 1C D P 1C rP1C ; or (9.111a)
FP 1 D P 1 rP1 (9.111b)
Under longitudinal tensile stresses, the fracture plane is perpendicular to the fiber
direction, and when reversing the loading direction, the cracks close and can still
transfer load [69]. However, under longitudinal compression, the damaged material
consisting of broken fibers and matrix cracks forms a kink band. Upon load reversal,
the broken and misaligned fibers do not carry any additional load [69]. Therefore,
the evolution of damage thresholds for longitudinal damage are defined as [69]:
n o
r1C D max 1; max f 1C g ; max f 1 g (9.112a)
t!1 t!1
n o
r1 D max 1; max f 1 g (9.112b)
t!1
As explained before, when the material is undamaged, the internal variables r1C
and r1 take the initial value of 1, and d1C and d1 are equal to zero. The condition
for positive dissipation is satisfied if dP 1C 0 and dP 1 0 [69]. The condition for
positive dissipation is automatically fulfilled if the damage evolution laws satisfy
the conditions @d1C =@r1C 0 and @d1 =@r1 0 [69].
The solution normally used to ensure the correct computation of the energy
dissipated regardless of the refinement of the mesh is to adjust the damage evolution
laws using a characteristic dimension of the finite element and the fracture toughness
[69], in a procedure based on the crack band model proposed by Bažant and Oh [35]:
G1C
g1C D (9.113a)
l
G1
g1 D (9.113b)
l
258 P.P. Camanho and A. Arteiro
where g1C and g1 is the energy dissipated per unit volume associated, respectively,
with tensile and compressive longitudinal fracture. For consistency, in the current
meso-mechanics framework, the same characteristic length, l , is used for both
transverse (Sect. 9.3.1) and longitudinal damage models [70]. The energy dissipated
per unit volume for uniaxial stress conditions is obtained by integrating the rate of
dissipation:
Z 1 Z 1
@G1C @d1C G1C
g1C D Y1C dP 1C dt D dr1C D (9.114a)
0 1 @d1C @r1C l
Z 1 Z 1
@G1 @d1 G1
g1 D Y1 dP 1 dt D dr1 D (9.114b)
0 1 @d1 @r1 l
Equation (9.114) can be used to calculate the damage law parameters that ensure
that the dissipated energy computed by the numerical model is independent of mesh
refinement [85].
To avoid the occurrence of a local snap-back in the stress–strain relation, the
elastic energy of a given finite element at the onset of localization must be lower
than or equal to the fracture energy. Therefore, the maximum size for the finite
element is given as [85]:
2E11 G1C 2E11 G1
l min ; (9.115)
XT2 XC2
If a finite element model consists of elements larger than the maximum size
prescribed by Eq. (9.115) and a mesh refinement is unfeasible, the snap-back in
the constitutive model can be avoided by reducing the corresponding strength [85].
Nevertheless, it is important to stress that, although the modification of the strengths
ensures the correct calculation of the energies dissipated, this modification should
not be performed in the elements representing the region where crack initiation
controlled by the stress tensor takes place. For instance, in the regions of stress
concentrations, where crack initiation is likely to take place, the mesh should be
sufficiently refined to avoid any adjustment of the material properties [85].
Two preliminary verifications were performed to ensure that the results of the
proposed model are numerically consistent. First, a verification based on single-
element tests, subjected to simple stress states, is performed. Then, following
Camanho et al. [70], simple models of a single UD ply subjected to uniaxial
transverse tension are used to demonstrate that the predictions do not depend on
the level of mesh refinement.
9 Analysis Models for Polymer Composites Across Different Length Scales 259
To show that the predictions are independent of the finite element size, following
Camanho et al. [70], simple models of an UD IM7/8552 carbon-epoxy Œ90 laminate,
with a thickness of 0.125 mm, are used. A rectangular plate, 12 mm wide and 40 mm
long, subjected to a remote tensile load is simulated. Three different levels of mesh
refinement were generated. The nominal in-plane sizes of the finite elements are
0:3 0:3 mm2 , 0:5 0:5 mm2 , and 0:8 0:8 mm2 . One element along the thickness
of the UD laminate is used. The SCM is applied to the elements placed on the
fracture plane to simulate the onset and propagation of transverse cracking.
Figures 9.36 and 9.37 show, respectively, the scalar damage variable that
represents transverse matrix cracking at the peak loads and the relation between
the remote stress and the applied displacement for each FE discretization. As it can
be observed, the predicted peak load is independent of the mesh refinement.
The previous preliminary verification shows that the implementation of the
SCM [70] is consistent with the proposed damage idealization and that it does
not suffer from lack of objectivity. However, it is now necessary to assess the
accuracy of the proposed damage model, and understand how well it represents the
inelastic response and failure behavior of composite laminates and structural details
subjected to different loading conditions. With this purpose, a comparison between
the predictions of the proposed damage model and data obtained from experimental
9 Analysis Models for Polymer Composites Across Different Length Scales 261
Fig. 9.36 Scalar damage variable that represents transverse matrix cracking at the peak load for
different FE discretizations: (left) 0:3 0:3 mm2 , (center) 0:5 0:5 mm2 , and (right) 0:8 0:8 mm2
Fig. 9.37 Relation between the remote stress and the applied displacement for different FE
discretizations
tests carried out on an IM7/8552 Œ90=0= ˙ 453S laminate is presented. Test data
includes results on unnotched and open-hole specimens loaded in tension and in
compression.
The FE models were constructed using C3D8R finite elements [42]. Each ply
of the IM7/8552 Œ90=0= ˙ 453S laminate was modeled as a homogeneous material
(Table 9.5), with one element through the thickness. The constraining effect of the
adjacent plies on the initiation of transverse damage growth in the embedded and
outer plies was taken into account using as ply strengths the in situ strengths.
The in situ transverse tensile and in-plane shear strengths, YTis and SLis ,
respectively, can be calculated using the analytical models proposed by
262 P.P. Camanho and A. Arteiro
a b
Fig. 9.38 Parameters L and T . (a) 22 –12 UD and in situ failure envelopes. (b) 22 –23 UD and
in situ failure envelopes
Camanho et al. [56]. The remaining transverse strength properties, though, should
be defined in the framework of the corresponding failure criterion [71, 72]. For
instance, the in situ transverse shear strength, STis , and the in situ biaxial transverse
is
tensile strength, YBT , are calculated imposing [71]:
8
.C/ .C/
ˆ
< L D L;is
(9.116)
:̂ .C/ D .C/
T T;is
where L and T are, respectively, the slopes in the 22 –12 and 22 –23 failure
envelopes when 22 D 0 [64] (Fig. 9.38).
In compression, it is assumed that the biaxial transverse compressive strength
is
is not an in situ property, i.e., YBC D YBC
ud
In this case, the in situ transverse
is
compressive strength, YC , can be determined imposing one of the following relations
(both giving the same result): result):
./ ./
L D L;is (9.117a)
./ ./
T D T;is (9.117b)
Figure 9.39 shows the relation between the ply thickness and the in situ strengths
for an IM7/8552 inner ply embedded in a multidirectional laminate.
Ideally, the characteristic length l should be calculated as a function of the
fracture angle and of the direction of crack propagation. However, in the present
models, l is taken as the square root of the finite element’s in-plane area (the
element’s thickness is constant and equal to the ply thickness). This means that
the most accurate predictions are obtained with elements with in-plane aspect ratios
close to one, with the element edges aligned with the crack propagation direction,
and for cracks that are perpendicular to the symmetry plane of the laminate (e.g.,
uniaxial transverse tensile failure).
It is noted that the last condition is dependent on the stress state and on
the material response (e.g., wedge fracture due to transverse compression), and
9 Analysis Models for Polymer Composites Across Different Length Scales 263
Fig. 9.39 Relation between ply thickness and in situ strengths for the IM7/8552 carbon-epoxy
therefore cannot be easily addressed when constructing the mesh for the FE models.
On the other hand, due to the morphology of FRPs, the transverse crack propagation
direction is always parallel to the fibers, whereas longitudinal failure typically
occurs perpendicularly to the fiber direction. Hence, to ensure that the element edges
are aligned with the crack propagation direction, a structured mesh aligned with the
fiber direction was employed [107, 108]. Due to meshing difficulties in the off-axis
˙45ı plies of the open-hole specimens, six-node linear wedge (C3D6) elements
were included close to the edges of the open holes.
Cohesive surfaces, defined as surface interactions [42], were used to simulate
delamination between plies. An uncoupled traction–separation behavior, a maxi-
mum stress criterion for damage initiation, and an energy-based damage evolution
criterion were used [42]. The mixed-mode behavior (based on the BK law [44]) was
provided in tabular form [108]. Because delamination is a matrix-dominated failure
mechanism, the transverse tensile and shear strengths, YT and SL , respectively, and
the fracture toughness GIc and GIIc (Table 9.5) were used in the definition of the
interlaminar properties used in the cohesive surfaces.
Abaqus/Explicit [42] was used to simulate the quasi-static behavior of the
IM7/8552 Œ90=0= ˙ 453S laminate, while ensuring a low kinetic energy throughout
the simulations. Loading was applied by means of appropriate boundary conditions,
imposing a velocity along the longitudinal direction of the specimens using a smooth
step amplitude [108].
In the FE models of the open-hole specimens, taking advantage of laminate
symmetry, only one-half of the laminate was simulated to reduce computing time.
In addition, the damage model was applied to the elements in the central region,
where damage is expected to grow due to the presence of the geometric discon-
tinuity. A linear-elastic material model, defined using only the elastic properties of
IM7/8552 carbon-epoxy (Table 9.5), was applied to the finite elements in the regions
far from the center.
264 P.P. Camanho and A. Arteiro
Table 9.6 Comparison between experiments [107, 109] and predictions [108] of
the proposed implementation of the meso-models
Specimen Width Hole diameter Experiments Predictions Relative
type (mm) (mm) (MPa) (MPa) error (%)
Unnotched specimens
Tension 12 – 845:1 871:4 3:1
Compression 12 – 532:6 540:5 1:5
Open-hole specimens
Tension 12 2 555:7 611:4 10:0
Tension 24 4 480:6 529:6 10:2
Tension 36 6 438:7 473:9 8:0
Tension 48 8 375:7 438:9 16:8
Tension 60 10 373:7 407:8 9:1
Compression 12 2 383:1 414:5 8:2
Compression 18 3 372:9 421:1 12:9
Compression 24 4 365:2 387:8 6:2
Compression 30 5 353:7 382:5 8:1
Experimental results for unnotched and open-hole specimens of an IM7/8552
Œ90=0= ˙ 453S laminate
Table 9.6 shows a comparison between the ultimate remote stresses obtained
from experimental testing and from the predictions of the proposed damage
model for both unnotched and open-hole specimens. The relative errors between
experiments and predictions are also presented.
This preliminary validation study (Table 9.6) shows that, in the case of the
unnotched specimens, the predictions of the proposed implementation of the meso-
models are only slightly above the experimental results in both loading scenarios,
correlating very well with the corresponding ultimate remote stresses. However, in
the case of the open-hole specimens, the numerical results overpredict the ultimate
remote stresses obtained experimentally, with relative errors around 10 %.
This larger overprediction can be attributed to the role of delamination in
the failure behavior of the notched specimens, since current implementations of
cohesive surface interactions [42] are known to lack accuracy in the prediction of
interlaminar damage growth [107]. Cohesive elements (e.g., [110]), on the other
hand, are currently mature enough to accurately predict delamination in composite
laminates. Unfortunately, the choice for an oriented, structured mesh, intended
to improve the accuracy of the proposed damage models, does not allow the
implementation of cohesive elements between the solid elements that represent each
ply in the laminate. Nonetheless, the predictions of the unnotched specimens of
the IM7/8552 Œ90=0= ˙ 453S laminate, whose response is reasonably independent
of delamination growth before ultimate failure, show that the proposed meso-
modeling framework can be seen as a valid tool to predict failure of multidirectional
laminates using only properties determined from tests carried out at the lamina level
(Table 9.6).
9 Analysis Models for Polymer Composites Across Different Length Scales 265
Structural components are typically designed taking into account the effect of stress
concentrations, which can be originated from several sources, such as geometrical
discontinuities or discrete through-penetration damage. The replacement of metallic
structures by composite materials in such applications is therefore partially depen-
dent on the availability of accurate, physically based and fast strength prediction
methods for the design of composite structures with stress concentrations. Accurate,
physically based FE models for the strength prediction of composite laminates
were described in Sect. 9.3. However, they normally result in long computing times
that are not acceptable for preliminary sizing and for the optimization of structural
details.
The most widely used analytical models, such as the Point Stress and Average
Stress models [111], use empirical “characteristic distances” that, in practice,
are determined from notched strength data. They are often thought as correction
factors that account for apparent changes in the stress distribution or in the
fracture toughness with varying notch size, with limited physical meaning [112].
Although analysis models that include characteristic distances provide reasonable
predictions for notch dimensions close to that used for calibration, they fail to
predict the notched response of laminated plates with different lay-ups and different
geometries.
To eliminate the need for non-physical calibration parameters, the strength
prediction methods should be based on independently measured material properties.
To this end, Finite Fracture Mechanics (FFMs) [113, 114] provides a sound physical
basis to predict the failure stress of notched laminates [115].
The FFMs model assumes that crack propagation results from the simultaneous
fulfillment of a stress-based criterion and an energy-based criterion. For the
particular case of a laminated plate with an uncracked central circular hole subjected
to a remote stress parallel to the x-axis (Fig. 9.40a), the coupled stress and energy
FFMs criterion yields [115]:
8 Z RCl
ˆ
ˆ 1
< xx .0; y/dy D X L
l ZR (9.118)
ˆ 1 RCl 2 2
:̂ KI .a/da D KIc
l R
where R is the hole radius, xx .0; y/ is the stress distribution along the ligament
section perpendicular to the loading direction (along the y-axis) [116], X L is the
laminate unnotched strength, KI .a/ is the mode I Stress Intensity Factor (SIF)
of a laminated plate with a central circular hole of radius R and two symmetric
cracks emanating from the hole edge (Fig. 9.40a) [117], KIc is the laminate mode I
fracture toughness, and l is the crack extension at failure. According to the proposed
model, failure is the result of the propagation of kinematically admissible cracks
with finite sizes. Taking into account that the system of Eq. (9.118) results in two
266 P.P. Camanho and A. Arteiro
a b
Fig. 9.40 Notched configurations. (a) Central circular hole. (b) Central notch
unknowns (the remote notched strength N 1 and the crack extension at failure l),
there is no need to use empirical characteristic distances identified from notched
test specimens. Only two independently measured material properties are required:
the laminate unnotched strength, X L , and the laminate mode I fracture toughness,
KIc . The first can be measured testing an unnotched specimen configuration of the
laminate. The second can be obtained, for instance, testing a notched specimen
configuration of the laminate and applying the FFMs analysis. For a laminated
plate with a central notch of length 2a, for example (Fig. 9.40b), the FFMs criterion
reads [118]:
8 Z aCl
ˆ
ˆ 1
< xx .0; y/dy D X L
l Za (9.119)
ˆ 1 aCl 2 2
:̂ KI .a/da D KIc
l a
where KI .a/ is the mode I SIF of a plate with a central crack of length 2a [119–121].
Knowing the laminate unnotched strength, X L , and the laminate remote notched
strength obtained from the test, N 1 , it is possible to determine the laminate fracture
toughness KIc .
Figures 9.41 and 9.42 show comparisons between the experiments carried out
on open-hole specimens and the predictions of the FFMs model [Eq. (9.118)].
The comparisons also include the predictions obtained using alternative analytical
models based on characteristic distances. In general, Figs. 9.41 and 9.42 indicate
9 Analysis Models for Polymer Composites Across Different Length Scales 267
Fig. 9.41 Comparison between predictions [111, 115] and experiments [109, 122] for open-hole
tension tests on different IM7/8552 carbon-epoxy laminates [reprinted from Composites Part
A: Applied Science and Manufacturing, Vol 43, P.P. Camanho, G.H. Erçin, G. Catalanotti, S.
Mahdi, P. Linde, A finite fracture mechanics model for the prediction of the open-hole strength
of composite laminates, Pages No. 1219–1225, © Elsevier (2012), with permission from Elsevier].
(a) [90/0/˙45]3S lay-up. (b) [45/90/45/0]4S lay-up
268 P.P. Camanho and A. Arteiro
Fig. 9.42 Comparison between predictions [111, 118] and experiments [123] for open-hole
tension and open-hole compression tests on different non-crimp fabric T700/AR-2527 carbon-
epoxy laminates. (a) [(0/45)/(90/45)]6T lay-up. (b) [(0/45)/(45/0)/(90/45)/(45/90)]S lay-up
9 Analysis Models for Polymer Composites Across Different Length Scales 269
that the FFMs model provides more accurate predictions than the alternative models.
In addition, it should be stressed that, unlike the alternative models, the FFMs model
does not require any inverse identification from one of the open-hole tests, resulting
in higher versatility and generality and making it ideal to evaluate macroscopic size
effects on composite laminates.
Taking into account that the FFMs model provides predictions of the notched
strength in a few seconds, it can be used to generate design charts for notched
laminates [115]. These relate the predicted normalized notched strength of the
laminate, defined as N N D N 1 =X L , with the inverse of the width-to-hole diameter
ratio, 2R=W, for different hole diameters 2R. These design charts also include
the predictions for the limit cases of a notch-sensitive material, whose normalized
strength is a function of the stress concentration factor KT (N N D 1=KT ) [116], and a
notch-insensitive material, whose normalized strength is a function of the geometry
(N N D 1 2R=W). Figure 9.43 shows the design charts for open-hole tension of
different non-crimp fabric T700/AR-2527 carbon-epoxy laminates. As expected,
for a constant 2R=W ratio, the FFMs model predicts that the mechanical response
of a notched laminate moves from a notch sensitive to notch insensitive material
behavior for decreasing hole diameters, in agreement with the experimental results
[123]. It should be noted that the design charts presented in Fig. 9.43 can be obtained
in a fraction of a minute, without the need of extensive experimental programs
to determine calibration factors (or characteristic distances) for each geometry,
making it very useful for preliminary sizing and optimization of notched composite
structures. It is also noted that calculating these design charts using any of the
alternative models would not be effective, since, to do so, a set of open-hole tests
would be necessary for inverse identification of the characteristic distances for the
different 2R=W ratios.
To predict the notched response of laminates with sharp notches (Fig. 9.40b),
Eq. (9.119) can be used. In fact, motivated by design and certification requirements
[124], in the aircraft industry it is common to perform analyses of the residual
strength of composite laminates with large through-penetration damage, usually
designated as Large Damage Capability (LDC) or Large-Notch Capability analyses.
Experimental characterization of several composite materials and different lay-
ups has shown that composite laminates exhibit a fracture resistance behavior
resembling an R-curve type of response, characterized by an increasing critical
ERR with crack growth. This damage growth resistance can be attributed to a
load redistribution resulting from microcracking, splitting, and/or delamination,
which relieve the stress concentration and delays fracture to higher applied loads
[112, 125–128], and to a bridging effect promoted by the intact fibers of the plies
that are adjacent to the principal load carrying plies with broken fibers [125, 126].
To take these mechanisms into account in the macro-mechanical model, the
analysis methods should be reformulated to include the effect of the R-curve. This
is possible using the proposed FFMs model by rewriting the system of Eq. (9.119)
as [128]:
270 P.P. Camanho and A. Arteiro
Fig. 9.43 Design charts for open-hole tension of different non-crimp fabric T700/AR-2527
carbon-epoxy laminates. (a) [(0/45)/(90/45)]6T lay-up. (b) [(0/45)/(45/0)/(90/45)/(45/90)]S
lay-up
8 Z aCl
ˆ
ˆ 1
< xx .0; y/ dy D X L
Zl aCl
a Z l (9.120)
ˆ
:̂ GI .a/ da D R. a/ d a
a 0
where GI .a/ is the mode I ERR [119–121] and R. a/ is the R-curve, expressed in
terms of a critical ERR, of the laminate [105, 106]. Figure 9.44 shows a comparison
between the FFMs predictions that do not take and that take into account the
R-curve of the laminate and the experimental results obtained from tests on center-
notched T700/AR-2527 carbon-epoxy coupons with notch lengths of 4 and 36 mm.
The predictions of alternative analytical models [111] based on characteristic
9 Analysis Models for Polymer Composites Across Different Length Scales 271
Fig. 9.44 Comparison between predictions [111, 128] and experiments [123, 128]
for center-notched tension tests on a non-crimp fabric T700/AR-2527 carbon-epoxy
[(0/45)/(45/0)/(90/45)/(45/90)]S laminate [reprinted from Composites Part A: Applied
Science and Manufacturing, Vol 63, A. Arteiro, G. Catalanotti, J. Xavier, P.P. Camanho, Large
damage capability of non-crimp fabric thin-ply laminates, Pages No. 110–122, © Elsevier (2014),
with permission from Elsevier]
where XTL is the laminate tensile unnotched strength, xx .0; / and GI ./ are written
in terms of the dimensionless coordinate (Fig. 9.45) [129]:
2
D .y R/ (9.122)
W 2R
272 P.P. Camanho and A. Arteiro
2
D a (9.123)
W 2R
Figure 9.46 shows the predictions of the FFMs model and a comparison with
the experimental results for the net-tension failure of a [90/0/˙45]3S IM7/8552
carbon-epoxy laminate [129]. The bearing cut-off strength [73] is also shown. The
predictions obtained are in good agreement with the experimental results.
The information presented in the previous sections leads to the conclusion that,
despite the complexities of the fracture process of polymer composites, significant
advances were reached during the last decade.
The micro-mechanical models not only provide relevant information on the
effects of the microstructure on the response of the composite material, but are also
important to support and validate the models developed taking the composite ply as
a homogeneous material.
The meso-mechanical models are able to predict with good accuracy the
failure load and failure mode of composite laminates. Such models do not require
9 Analysis Models for Polymer Composites Across Different Length Scales 273
Fig. 9.46 Comparison between predictions and experiments [129] for the net-tension failure
of a [90/0/˙45]3S IM7/8552 carbon-epoxy laminate [reprinted from Composites Science and
Technology, Vol 76, G. Catalanotti, P.P. Camanho, A semi-analytical method to predict net-tension
failure of mechanically fastened joints in composite laminates, Pages No. 69–76, © Elsevier
(2013), with permission from Elsevier]
pre-existing information of the fracture planes and can integrate different failure
mechanisms. In addition, these models can be used to simulate different structural
details without any modification of the underlying constitutive relations.
The macro-mechanical models based on FFMs are ideal tools to perform quick
sizing and optimization of some composite structures. These models reduce the level
of empiricism of traditional analysis models such as the point- or average-stress
criteria and result in improved prediction when compared with these methods.
Despite the advances that took place, some challenges remain. For example, the
identification of cohesive laws for fiber-dominated failure modes needs to be further
addressed. While there has been some progress in the identification of cohesive
laws for multidirectional laminates, there is still the need to identify cohesive laws
for 0ı plies. Furthermore, recent advances in Computational Mechanics, such as the
two-scale Finite Element Method, can be used to formally link some of the scales
addressed in this chapter. Finally, the in situ characterization of the microstructure
of composite materials, combined with the methods described here, will enable the
much sought “in silico” material design.
References
20. D. Trias, J. Costa, J.A. Mayugo, J.E. Hurtado, Random models versus periodic models for
fiber reinforced composites. Comput. Mater. Sci. 38, 316–324 (2006)
21. J. Feder, Random sequential adsorption. J. Theor. Biol. 87, 237–254 (1980)
22. V.A. Buryachenko, N.J. Pagano, R.Y. Kim, J.E. Spowart, Quantitative description and
numerical simulation of random microstructures of composites and their effective elastic
moduli. Int. J. Solids Struct. 40, 47–72 (2003)
23. J.H. Oh, K.K. Jin, S.K. Ha, Interfacial strain distribution of a unidirectional composite with
randomly distributed fibers under transverse loading. J. Compos. Mater. 40(9), 759–778
(2006)
24. R. Pyrz, Quantitative description of the microstructure of composites. Part I: morphology of
unidirectional composite systems. Compos. Sci. Technol. 50(2), 197–208 (1994)
25. A. Wongsto, S. Li, Micromechanical FE analysis of UD fibre-reinforced composites with
fibres distributed at random over the transverse cross-section. Compos. Part A Appl. Sci. 36,
1246–1266 (2005)
26. W.S. Jodrey, E.M. Tory, Computer simulation of close random packing of equal spheres.
Phys. Rev. A 32(4), 2347 (1985)
27. D. Trias, Analysis and simulation of transverse random fracture of long fibre reinforced
composites. Ph.D. thesis, Universitat de Girona, Escola Politècnica Superior, Girona (2005)
28. S. Yang, A. Tewari, A.M. Gokhale, Modeling of non-uniform spatial arrangement of fibers in
a ceramic matrix composite. Acta Mater. 45(7), 3059–3069 (1997)
29. A. Bowyer, Computing Dirichlet tessellations. Comput. J. 24(2), 162–166 (1981)
30. P.M. Dixon, Ripley’s K function, in Encyclopedia of Environmetrics, vol. 3, ed. by A.H.
El-Shaarawi, W.W. Piegorsch (Wiley, New York, 2002), pp. 1796–1803
31. T. Matsuda, N. Ohno, H. Tanaka, T. Shimizu, Effects of fiber distribution on elastic-
viscoplastic behavior of long fiber-reinforced laminates. Int. J. Mech. Sci. 45, 1583–1598
(2003)
32. P.D. Soden, M.J. Hinton, A.S. Kaddour, Lamina properties, lay-up configurations and loading
conditions for a range of fibre-reinforced composite laminates. Compos. Sci. Technol. 58,
1011–1022 (1998)
33. ABAQUS 6.6/Standard User’s Manual. Hibbit, Karlsson & Sorensen, Inc. (2006)
34. A.R. Melro, P.P. Camanho, F.M.A. Pires, S.T. Pinho, Micromechanical analysis of polymer
composites reinforced by unidirectional fibres: Part I - constitutive modelling. Int. J. Solids
Struct. 50, 1897–1905 (2013)
35. Z.P. Bažant, B.H. Oh, Crack band theory for fracture of concrete. Mater. Struct. 16(93),
155–177 (1983)
36. A. Arteiro, G. Catalanotti, A.R. Melro, P. Linde, P.P. Camanho, Micro-mechanical analysis
of the in situ effect in polymer composite laminates. Compos. Struct. 116, 827–840 (2014)
37. P.P. Camanho, A. Arteiro, A.R. Melro, G. Catalanotti, M. Vogler, Three-dimensional
invariant-based failure criteria for fibre-reinforced composites. Int. J. Solids Struct. 55,
92–107 (2015)
38. T.J. Vaughan, C.T. McCarthy, Micromechanical modelling of the transverse damage
behaviour in fibre reinforced composites. Compos. Sci. Technol. 71, 388–396 (2011)
39. A. Arteiro, G. Catalanotti, A.R. Melro, P. Linde, P.P. Camanho, Micro-mechanical analysis
of the effect of ply thickness on the transverse compressive strength of polymer composites.
Compos. Part A Appl. Sci. 79, 127–137 (2015)
40. T.J. Vaughan, C.T. McCarthy, A micromechanical study on the effect of intra-ply properties
on transverse shear fracture in fibre reinforced composites. Compos. Part A Appl. S. 42,
1217–1228 (2011)
41. R.P. Tavares, A.R. Melro, M.A. Bessa, A. Turon, W.K. Liu, P.P. Camanho, Mechanics of
hybrid polymer composites: analytical and computational study. Comput. Mech. 57, 405–421
(2016)
42. Abaqus 6.12 Documentation, Dassault Systèmes Simulia Corp., Providence (2012)
43. A. Turon, C.G. Dávila, P.P. Camanho, J. Costa, An engineering solution for mesh size
effects in the simulation of delamination using cohesive zone models. Eng. Fract. Mech.
74, 1665–1682 (2007)
276 P.P. Camanho and A. Arteiro
68. J.A. Mayugo, P.P. Camanho, P. Maimí, C.G. Dávila, Analytical modelling of transverse matrix
cracking of f˙=90n gs composite laminates under multiaxial loading. Mech. Adv. Mater.
Struct. 17, 237–245 (2010)
69. P. Maimí, P.P. Camanho, J.A. Mayugo, C.G. Dávila, A continuum damage model for
composite laminates: Part I — constitutive model. Mech. Mater. 39, 897–908 (2007)
70. P.P. Camanho, M.A. Bessa, G. Catalanotti, M. Vogler, R. Rolfes, Modeling the inelastic
deformation and fracture of polymer composites — Part II: smeared crack model. Mech
Mater 59, 36–49 (2013)
71. G. Catalanotti, P.P. Camanho, A.T. Marques, Three-dimensional failure criteria for fiber-
reinforced laminates. Compos. Struct. 95, 63–79 (2013)
72. P.P. Camanho, A. Arteiro, G. Catalanotti, A.R. Melro, M. Vogler, Three-dimensional
invariant-based failure criteria for transversely isotropic fibre-reinforced composites, in
Numerical Modelling of Failure in Advanced Composite Materials, Chap. 5, ed. by P.P.
Camanho, S.R. Hallett (Woodhead Publishing, Cambridge, 2015), pp. 111–150
73. P.P. Camanho, M. Lambert, A design methodology for mechanically fastened joints in
laminated composite materials. Compos. Sci. Technol. 66, 3004–3020 (2006)
74. M.R. Wisnom, B. Khan, S.R. Hallett, Size effects in unnotched tensile strength of
unidirectional and quasi-isotropic carbon/epoxy composites. Compos. Struct. 84, 21–28
(2008)
75. H. Koerber, J. Xavier, P.P. Camanho, High strain rate characterisation of unidirectional
carbon-epoxy IM7-8552 in transverse compression and in-plane shear using digital image
correlation. Mech. Mater. 42, 1004–1019 (2010)
76. P.P. Camanho, C.G. Dávila, M.F. de Moura, Numerical simulation of mixed-mode progressive
delamination in composite materials. J. Compos. Mater. 37(16), 1415–1438 (2003)
77. J.M. Berthelot, Transverse cracking and delamination in cross-ply glass-fiber and carbon-
fiber reinforced plastic laminates: static and fatigue loading. Appl. Mech. Rev. 56(1), 111–147
(2003)
78. W.A. Curtin, N. Takeda, Tensile strength of fiber-reinforced composites: II. Application to
polymer matrix composites. J. Compos. Mater. 32(22), 2060–2081 (1998)
79. F. Tanaka, T. Okabe, H. Okuda, I.A. Kinloch, R.J. Young, Factors controlling the strength of
carbon fibres in tension. Compos. Part A Appl. Sci. 57, 88–94 (2014)
80. A.E. Scott, I. Sinclair, S.M. Spearing, A. Thionnet, A.R. Bunsell, Damage accumulation in a
carbon/epoxy composite: comparison between a multiscale model and computed tomography
experimental results. Compos. Part A Appl. Sci. 43, 1514–1522 (2012)
81. Y. Swolfs, H. Morton, A.E. Scott, L. Gorbatikh, P.A.S. Reed, I. Sinclair, et al. Synchrotron
radiation computed tomography for experimental validation of a tensile strength model for
unidirectional fibre-reinforced composites. Compos. Part A Appl. Sci. 77, 106–113 (2015)
82. M. Jirásek, P. Grassl, Evaluation of directional mesh bias in concrete fracture simulations
using continuum damage models. Eng. Fract. Mech. 75, 1921–1943 (2008)
83. E.V. Iarve, M.R. Gurvich, D.H. Mollenhauer, C.A. Rose, C.G. Dávila, Mesh-independent
matrix cracking and delamination modeling in laminated composites. Int. J. Numer. Methods
Eng. 88, 749–773 (2011)
84. M. Vogler, R. Rolfes, P.P. Camanho, Modeling the inelastic deformation and fracture of
polymer composites — Part I: plasticity model. Mech. Mater. 59, 50–64 (2013)
85. P. Maimí, P.P. Camanho, J.A. Mayugo, C.G. Dávila, A continuum damage model for
composite laminates: Part II — computational implementation and validation. Mech. Mater.
39, 909–919 (2007)
86. M.J. Hinton, P.D. Soden, Predicting failure in composite laminates: the background to the
exercise. Compos. Sci. Technol. 58, 1001–1010 (1998)
87. P.D. Soden, M.J. Hinton, A.S. Kaddour, A comparison of the predictive capabilities of current
failure theories for composite laminates. Compos. Sci. Technol. 58, 1225–1254 (1998)
88. M.J. Hinton, A.S. Kaddour, P.D. Soden, Evaluation of failure prediction in composite
laminates: background to ‘part B’ of the exercise. Compos. Sci. Technol. 62, 1481–1488
(2002)
278 P.P. Camanho and A. Arteiro
89. M.J. Hinton, A.S. Kaddour, P.D. Soden, A comparison of the predictive capabilities of current
failure theories for composite laminates, judged against experimental evidence. Compos. Sci.
Technol. 62, 1725–1797 (2002)
90. M.J. Hinton, A.S. Kaddour, P.D. Soden, Evaluation of failure prediction in composite
laminates: background to ‘part C’ of the exercise. Compos. Sci. Technol. 64, 321–327 (2004)
91. A.S. Kaddour, M.J. Hinton, P.D. Soden, A comparison of the predictive capabilities of current
failure theories for composite laminates: additional contributions. Compos. Sci. Technol. 64,
449–476 (2004)
92. M.J. Hinton, A.S. Kaddour, P.D. Soden, A further assessment of the predictive capabilities
of current failure theories for composite laminates: comparison with experimental evidence.
Compos. Sci. Technol. 64, 549–558 (2004)
93. S.W. Tsai, E.M. Wu, A general theory of strength for anisotropic materials. J. Compos. Mater.
5(1), 58–80 (1971)
94. A. Puck, H. Schürmann, Failure analysis of FRP laminates by means of physically based
phenomenological models. Compos. Sci. Technol. 58, 1045–1067 (1998)
95. C.G. Dávila, P.P. Camanho, Failure criteria for FRP laminates in plane stress. Tech. Rep.
NASA/TM-2003-212663, NASA Langley Research Center, Hampton, Virginia (2003)
96. S.T. Pinho, L. Iannucci, P. Robinson, Physically-based failure models and criteria for
laminated fibre-reinforced composites with emphasis on fibre kinking: Part I: development.
Compos. Part A Appl. Sci. 37, 63–73 (2006)
97. P.J. Hine, R.A Duckett, A.S. Kaddour, M.J. Hinton, G.M. Wells, The effect of hydrostatic
pressure on the mechanical properties of glass fibre/epoxy unidirectional composites. Com-
pos. Part A Appl. Sci. 36, 279–289 (2005)
98. A.J.M. Spencer, Kinematic constraints, constitutive equations and failure rules for anisotropic
materials, in Applications of Tensor Functions in Solid Mechanics, Chap. 10, ed. by J.P.
Boehler. CISM Courses and Lectures, vol. 292 (Springer, Wien, 1987), pp. 187–201
99. M.A. Crisfield, Non-Linear Finite Element Analysis of Solids and Structures, vol. 1 (Wiley,
Chichester, 1991)
100. S.W. Tsai, Strength and Life of Composites (Composites Design Group, Department of
Aeronautics and Astronautics, Stanford University, 2008)
101. H. Thom, A review of the biaxial strength of fibre-reinforced plastics. Compos. Part A Appl.
Sci. 29A, 869–886 (1998)
102. A. Turon, P.P. Camanho, J. Costa, C.G. Dávila, A damage model for the simulation
of delamination in advanced composites under variable-mode loading. Mech. Mater. 38,
1072–1089 (2006)
103. C.G. Dávila, C.A. Rose, P.P. Camanho, A procedure for superposing linear cohesive laws
to represent multiple damage mechanisms in the fracture of composites. Int. J. Fract. 158,
211–223 (2009)
104. S.T. Pinho, P. Robinson, L. Iannucci, Fracture toughness of the tensile and compressive fiber
failure modes in laminated composites. Compos. Sci. Technol. 66, 2069–2079 (2006)
105. G. Catalanotti, J. Xavier, P.P. Camanho, Measurement of the compressive crack resistance
curve of composites using the size effect law. Compos. Part A Appl. Sci. 56, 300–307 (2014)
106. G. Catalanotti, A. Arteiro, M. Hayati, P.P. Camanho, Determination of the mode I crack
resistance curve of polymer composites using the size-effect law. Eng. Fract. Mech. 118,
49–65 (2014)
107. G.H. Erçin, Stress gradient effects in laminated composites. Ph.D. thesis, Faculdade de
Engenharia, Universidade do Porto, Porto (2013)
108. A.A.P.D. Branco, Virtual testing of advanced polymer composite materials. Master’s thesis,
Faculdade de Engenharia, Universidade do Porto, Porto (2015)
109. P.P. Camanho, P. Maimí, C.G. Dávila, Prediction of size effects in notched laminates using
continuum damage mechanics. Compos. Sci. Technol. 67, 2715–2727 (2007)
110. E.V. González, P. Maimí, A. Turon, P.P. Camanho, J. Costa, Simulation of delamination by
means of cohesive elements using an explicit finite element code. CMC Comput. Mater.
Continua 9(1), 51–92 (20096)
9 Analysis Models for Polymer Composites Across Different Length Scales 279
111. J.M. Whitney, R.J. Nuismer, Stress fracture criteria for laminated composites containing
stress concentrations. J Compos. Mater 8, 253–265 (1974)
112. T.H. Walker, W.B. Avery, L.B. Ilcewicz, C.C. Poe Jr., C.E. Harris, Tension fracture of
laminates for transport fuselage. Part I: material screening, in, Proceedings of the 9th
DoD/NASA/FAA Conference on Fibrous Composites in Structural Design, vol. II, ed. by J.R.
Soderquist, L.M. Neri, H.L. Bohon (Lake Tahoe, Nevada, 1991), pp. 747–787
113. D. Leguillon, Strength or toughness? A criterion for crack onset at a notch. Eur. J. Mech. A
Solid 21, 61–72 (2002)
114. P. Cornetti, N. Pugno, A. Carpinteri, D. Taylor, Finite fracture mechanics: a coupled stress
and energy failure criterion. Eng. Fract. Mech. 73, 2021–2033 (2006)
115. P.P. Camanho, G.H. Erçin, G. Catalanotti, S. Mahdi, P. Linde, A finite fracture mechanics
model for the prediction of the open-hole strength of composite laminates. Compos. Part A
Appl. Sci. 43, 1219–1225 (2012)
116. S.C. Tan, Stress Concentrations in Laminated Composites (Technomic, Lancaster, 1994)
117. J.C. Newman Jr., A nonlinear fracture mechanics approach to the growth of small cracks, in
Proceedings of the AGARD Conference, vol. 328(6) (1983), pp. 1–26
118. A. Arteiro, G. Catalanotti, J. Xavier, P.P. Camanho, Notched response of non-crimp fabric
thin-ply laminates: analysis methods. Compos. Sci. Technol. 88, 165–171 (2013)
119. D.L. Chen, B. Weiss, R. Stickler, A new geometric factor formula for a center cracked plate
tensile specimen of finite width. Int. J. Fract. 55, R3–R8 (1992)
120. Z. Suo, G. Bao, B. Fan, T.C. Wang, Orthotropy rescaling and implications for fracture in
composites. Int. J. Solids Struct. 28(2), 235–248 (1991)
121. G. Bao, S. Ho, Z. Suo, B. Fan, The role of material orthotropy in fracture specimens for
composites. Int. J. Solids Struct. 29(9), 1105–1116 (1992)
122. B.G. Green, M.R. Wisnom, S.R. Hallett, An experimental investigation into the tensile
strength scaling of notched composites. Compos. Part A Appl. Sci. 38, 867–878 (2007)
123. A. Arteiro, G. Catalanotti, J. Xavier, P.P. Camanho, Notched response of non-crimp fabric
thin-ply laminates. Compos. Sci. Technol. 79, 97–114 (2013)
124. MIL-HDBK-17-3F, Composite Materials Handbook, volume 3. Polymer Matrix Composites
Materials Usage, Design and Analysis. Department of Defense Handbook (2002)
125. T.W. Coats, C.E. Harris, A progressive damage methodology for residual strength predictions
of notched composite panels. Tech. Rep. NASA/TM-1998-207646, NASA Langley Research
Center, Hampton, Virginia (1998)
126. T.W. Coats, C.E. Harris, A progressive damage methodology for residual strength predictions
of notched composite panels. J. Compos. Mater. 33(23), 2193–2224 (1999)
127. M.A. Caminero, M. Lopez-Pedrosa, C. Pinna, C. Soutis, Damage monitoring and analysis
of composite laminates with an open hole and adhesively bonded repairs using digital image
correlation. Compos. Part B Eng. 53, 76–91 (2013)
128. A. Arteiro, G. Catalanotti, J. Xavier, P.P. Camanho, Large damage capability of non-crimp
fabric thin-ply laminates. Compos. Part A Appl. Sci. 63, 110–122 (2014)
129. G. Catalanotti, P.P. Camanho, A semi-analytical method to predict net-tension failure of
mechanically fastened joints in composite laminates. Compos. Sci. Technol. 76, 69–76 (2013)
Part III
Chapter 10
Microscale Characterization Techniques
of Fibre-Reinforced Polymers
Fig. 10.1 Bottom-up multiscale simulation strategy for the analysis of structural composites
by means of a global finite element analysis. Then, the displacement field is used
to model regions with large strain gradients, identified as “hot spots”, including
enhanced kinematic descriptions to account for localized failure (e.g. enriched
X-FEM formulations) or non-linear models to account for diffuse damage and plas-
ticity. Top-down approaches rely on the determination of the mechanical behaviour
of materials as model inputs using vast and costly experimental campaigns. As a
result, material innovations/modifications are hindered to some extent due to the
lack of experimental data to address damage and failure.
Bottom-up approaches can be used to tackle such problems and to carry out
virtual tests from simple assumptions of the mechanical behaviour of the material
at the low scale level. A bottom-up simulation strategy presented by LLorca et al.
[1, 2] is shown in Fig. 10.1. Such approach takes the advantage of the hierarchical
structure of composite laminates obtained by stacking individual plies according to
some design sequence. Three different entities can be identified, namely, the plies,
the laminate and the final component, and the mechanical response at each level
is related with specific features at its scale. Individual ply strength is traditionally
endorsed to intrinsic physical failure mechanisms occurring at the micro level, e.g.
fibre/matrix debonding, matrix cracking or fibre failure. In the case of laminates, the
mechanical behaviour depends on the competition of different failure mechanisms
10 Microscale Characterization Techniques of Fibre-Reinforced Polymers 285
taking place at the ply and interply level. This hierarchical length scale separation
between micro, meso and macro allows to perform realistic multiscale simulations
based on information obtained at one scale, homogenized and passed to the next
level.
In the first scale level, computational micromechanics can be used to deter-
mine the intrinsic ply behaviour from the thermo-mechanical properties of the
constituents, fibres, matrix and fibre/matrix interfaces, in combination with the
definition of the architecture of the reinforcement, fibre shape, volume fraction and
spatial distribution within the ply (González and LLorca [3–5], Canal et al. [6, 7]).
Computational micromechanics assumes that the overall behaviour of the structural
composite can be studied through the finite element simulation of a representative
volume element (RVE) of the material containing information of the microstructure.
For unidirectional reinforcements (UD), the RVE can be formed by a periodic
dispersion of parallel fibres distributed homogeneously in a parallelepiped-shaped
domain. The fibre centres can be generated sequentially following either a random
dispersion [4, 8] or specific statistical descriptors based on nearest neighbours
obtained from optical inspections of the current microstructure of the material
[9–11]. The planar distribution of fibres can be extruded in the fibre direction
to create the final three-dimensional representation of the UD composite material
(Fig. 10.1). The RVE geometry can be discretized with finite elements to obtain
the stress and strain microfields when it is submitted to homogeneous stress states
(tension, compression and shear) in the different directions (in-plane and out-of-
plane). The relevant strength parameters of the UD composite can be obtained
then [4] as well as the failure envelopes for combination or interaction among the
different stress components [12–16].
The behaviour of the different constituents and interfaces in the RVE is a critical
factor to determine the overall properties and should reflect the real physics of
deformation and failure of the UD composite material. Failure of UD composites
in the fibre direction is controlled by the tensile strength and fracture of individual
fibres and the interaction with the matrix. In compression along the fibre direction,
failure is usually governed by local fibre kinking, which is driven by the initial
fibre misalignments resulting from the manufacturing process. On the other hand,
tensile failure perpendicular to the fibres is essentially controlled by fibre/matrix
interface debonding, while in transverse compression and shear, ductile matrix shear
is often accompanied by fibre/matrix debonding. It should be mentioned that the
final strength of the composite is a result of the competition of the different failure
mechanisms, and, hence, accurate models for fracture of FRP have to include such
physical description of the mechanisms as well and the complex interaction among
them.
As it was mentioned above, accurate models for material behaviour always
rely on the accurate and independent determination of the physical parameters of
the constitutive equations used to represent the actual deformation and fracture
mechanisms at the microscopic level. This chapter summarizes a set of advanced
experimental techniques for the determination of the relevant parameters governing
the mechanical behaviour of the UD composite constituents at the micro level:
fibre, matrix and fibre/matrix interfaces, in Sects. 10.2–10.4, respectively. The work
286 M. Herráez et al.
is specifically carried out using the AS4/8552 prepreg system (Hexcel), one of
the most used in structural applications of composites in aerospace, although
the methodology can be easily expanded to other material systems. Finally, the
conclusions and the future developments in this field are briefly described in
Sect. 10.5.
Small diameter fibres such as carbon, glass, aramid or polyethylene stand among
the strongest man-made materials in the present state of technological development.
Applications as reinforcements in composites emerged some decades ago driven
by weight reductions due to their excellent strength and stiffness. The strength and
stiffness of fibres are usually well characterized by the manufacturer. Straining up
to failure individual fibres or strands/tows containing thousand of fibres is a normal
operation to optimize the manufacturing process of the fibres or to monitor the final
quality of the products. However, other important fibre mechanical properties, such
as the fibre elastic anisotropy in carbon and polymeric fibres or the fibre toughness,
are less known due to the enormous experimental difficulties associated with testing
small diameter coupons. It should be noticed that the fibre strength and toughness
as well as the fibre/matrix interface properties play a crucial role in the strength and
energy dissipation mechanisms in brittle unidirectional composites.
A crude estimation of the fracture toughness in small diameter fibres can be
carried out based on the direct observation of the mirror, mist and hackle structures
visualized in post-mortem coupons of the composite material subjected to tensile
loading in the fibre direction. The size of the defect in combination with the
fracture stress provides an approximate value of the fibre toughness. However,
the methodology obviously introduces a large experimental scatter due to the
uncertainty in the determination of the exact defect size that led to the failure.
Focused ion beam (FIB) milling techniques open revolutionary opportunities
for testing materials at the micro level. The FIB probe with GaC ions is able
to remove/carve regions in the fibre to generate complex geometries for material
testing at the micro level (micropillars, dog bone-shaped specimens, holes, etc.). The
basic concept of FIB is similar to SEM but using ions instead of electrons. A fine
tungsten pin covered with liquid gallium (Ga) is used as ion source from which
Ga atoms are extracted and ionized via high voltage. GaC ions are accelerated in
the range of 0.5–50 keV and focused on the sample surface by electrostatic lenses.
Depending on the ion beam current, its interaction with the sample can be used both
for imaging, by means of different electron/ion detectors, and for micromachining
material by sputtering in selected locations. Almost all materials can be machined
by FIB, ranging from soft matter (polymers) to ultra-hard materials (diamond), with
smallest milled features in the range of 10–15 nm, depending on beam size and
beam/sample interactions. However, it should be noticed that this ion milling leads
to ion implantation into the top few nanometres of the surface and to the introduction
of local thermal stresses/defects.
10 Microscale Characterization Techniques of Fibre-Reinforced Polymers 287
The focused ion beam technique was successfully used to produce notched
fibres for the direct measurement of fibre fracture toughness by tensile testing. This
methodology controls precisely the geometry of the notch in terms of depth and tip
radius (Herráez et al. [17], Kant and Penumadu [18], Ogihara et al. [19]). The mode
I apparent fracture toughness, KIC , is then inferred from the maximum strength of
the notched fibre and the geometry of the notch.
AS4 is a high-strength carbon fibre manufactured from PAN (polyacrylonitrile)
precursor commonly used as reinforcements in prepreg sheets and dry fabrics
for structural applications. The fibre strength and toughness was determined by
means of individual tests carried out on plain and notched fibres, respectively. The
standard methodology to characterize filaments is based on the use of the well-
established cardboard technique. Firstly, fibres were carefully extracted from tows
supplied by the manufacturer using mechanical tweezers. The edges are then bonded
with adhesive (e.g. cyanoacrylate) to rectangular cardboards containing an internal
window with the desired fibre gauge length. The cardboard is directly connected by
mechanical grips to a universal testing frame or to a dedicated fibre tensile tester.
After cutting the cardboard bridges, the load is transmitted through the fibre, which
is strained up to failure. The load is recorded by the load cell of the testing machine
and the stress computed from the individual cross section of the filament that can
be measured individually using scanning electron microscopy images or by means
of the frequency method. In this latter methodology, the fibre is first prestressed to
obtain the natural frequencies from which the linear density and the corresponding
average cross-sectional area of the filament are derived. The average fibre diameter
found was D D 7.8 ˙ 0.2 m.
The statistical distribution of the strength of AS4 carbon fibres is obtained from
the failure stress of a set of 50 individual fibres tested with a gauge length of 20 mm.
The Weibull stress and modulus were 0 D 3959 MPa and m D 6.5, respectively.
The toughness of individual AS4 carbon fibres was evaluated from the tensile
strength of fibres with artificial notches introduced with a focused ion beam.
To this end, straight and sharp notches, perpendicular to the fibre axis, were
introduced with a depth a0 . The notch depth to the fibre diameter D ratio was set to
a0 /D 0.12 ˙ 0.03, as shown in Fig. 10.2a. The notch radius at the tip was 50 nm.
Assuming the validity of linear elastic fracture mechanics (LEFM), the failure stress
is dictated by the condition
p
KI .a; a=D; / D Y .a=D/ a D KIC (10.1)
where Y(a/D) is the shape function for the stress intensity factor of the cracked fibre,
a the crack depth and the far-field applied stress. The factor Y(a/D) for a straight
crack in a cylinder subjected to uniaxial tension along its axis was determined
numerically using the finite element method (Herráez et al. [17]).
The methodology assumes that neither the small crack tip radius ( 50 nm)
nor the possible material modifications induced during FIB milling influenced
significantly the fracture properties at the notch tip, and, therefore, the results
obtained can be interpreted as good approximations to the actual fibre toughness.
288 M. Herráez et al.
Fig. 10.2 (a) Scanning electron microscopy micrograph of a notched AS4 carbon fibre, (b) geom-
etry of the straight notch, (c) scanning electron microscopy micrograph of the fracture surface of a
notched AS4 carbon fibre
Table 10.1 Experimental results obtained in the tensile characterization of AS4 carbon fibre
p
Material Diameter (m) E (GPa) 0 (MPa) m L0 (mm) KIC (MPa m) GIC (J/m2 )
AS4 7.8 ˙ 0.2 229 ˙ 14 3959 6.5 20 2.12 ˙ 0.45 52 ˙ 2
p
The fracture toughness of the AS4 carbon fibre was KIC D 2.12 ˙ 0.45 MPa m,
which corresponds to a fracture energy of GIC D 52 ˙ 2 m2 . These results were
independent of the notch depth used in the experiments a0 /D D 0.12 ˙ 0.03. It
should be mentioned that carbon fibres usually exhibit a sheath/core structure
resulting from the manufacturing process. The skin layer thickness was estimated
around 300 nm for high-strength T700 carbon fibres by Kant and Penumadu [18]
which is much lower than the average notch depth used in this study (a0 0.9 m).
The relative notch depth used in these experiments exceeds the skin structure
of the fibre, and the toughness can be interpreted as a homogeneous property
corresponding to the central core of the fibre. The fracture surface of one AS4 carbon
fibre (see Fig. 10.2b) was flat and presented a granular morphology as also noticed
by other authors (Kant and Penumadu [18], Ogihara et al. [19]). The size of the
damage/plastic process zone ahead the notch tip lp can be estimated according to
the well-known Irwin relation as
2
1 KIC
lp D (10.2)
ave
where KIC stands for the fracture toughness and ave is the average tensile strength.
lp was equal to 0.105 m for AS4 carbon fibres, which is around 1/10 of the
notch depth. All the experimental data from micromechanical tests carried out on
AS4 carbon fibres are summarized in Table 10.1 for the sake of clarity.
10 Microscale Characterization Techniques of Fibre-Reinforced Polymers 289
Fig. 10.3 Typical load-displacement curve obtained in a nanoindentation test. (a) AS4/8552 cross
section showing a matrix pocket to perform nanoindentation, (b) atomic force microscope (AFM)
image showing a Berkovich pyramidal indenter footprint
indentation process were carried out to ascertain the effect of contact area on the
apparent hardness for a wide range of material properties, elastic modulus, yield
strength and pressure dependence parameter. The correction factor applied to the
apparent hardness Hap to compute the actual hardness, without pile-up effects, can
be obtained from the ratio of the elastic to total work of indentation, We /Wt measured
from the experimental load-displacement curves P• (see Figure 10.3) and a set
of master curves obtained with the FE simulations aforementioned (Fig. 10.4).
Typically, the frictional angle is assumed to be ˇ 22–36ı, a normal value for many
glassy polymers (Quinson et al. [27]), so the compressive strength of the polymer
can be determined directly from the indentation curves.
The methodology was used to determine the elastic modulus and yield strength of
the 8552 epoxy resin by in situ measurements performed on cross sections of the
AS4/8552 cured material (Fig. 10.3a). The cross sections of the material were first
carefully polished with diamond slurry down to 1 m particle size to produce a
suitable surface finish for nanoindentation tests. Indentations were performed on
selected resin pockets sufficiently far away from the fibre edge in order to minimize
the known constrain effect produced by the stiff fibres. In general, it is not difficult to
find such resin locations because of the inhomogeneous reinforcement arrangement
at the microscopic level even for plies containing large volume fractions of
reinforcement (60–70 %).
Nanoindentation tests were performed using a Hysitron TI950 TriboIndenter
(Hysitron, Minneapolis, USA) equipped with a Berkovich (pyramidal) indenter.
At least 30 indentations were carried out under load control using an equivalent
10 Microscale Characterization Techniques of Fibre-Reinforced Polymers 291
quasi-static strain rate of d"/dt D 0.07 s1 . Tests were performed under load control
for maximum load values ranging between 500 and 5000 N. Nanoindentation test
results obtained on matrix pockets are summarized in Table 10.2 for the AS4/8552
composite material. The average ratio between the elastic work (We ) and the total
work (Wt ) performed during the indentation obtained from the load-displacement
curves is 0.544, and the relationship between polymer matrix apparent hardness
and yield strength in compression is Hap 1.95¢ yc as estimated from the master
curve in Fig. 10.4 for the typical friction angles of 22–36 ı (Quinson et al. [27]).
Fig. 10.5 (a) Polymer resin micropillars carved on the AS4/8852 ply transverse cross section,
(b) FIB-induced skin on a polymer resin micropillar after compression
Fei et al. [31]). Micropillars were carved using FIB in the resin pockets of the
AS4/8552 cured composite transversal section (see Fig. 10.5a), as well as in neat
resin plates supplied by the manufacturer.
In order to investigate the influence of micropillar size on the strength, micropil-
lars with diameters ranging between 2 and 8 m were milled, with an aspect ratio
3:1. This length-to-diameter ratio ( 3) is necessary to ensure a uniaxial stress state
and avoid pillar buckling while minimizing the constraint effects at the contact with
the flat punch. The pillars were produced using annular milling which induced
slightly tapered cylinders (taper angle 4ı ). Milling steps were carried out with
decreasing currents down to 0.43 nA, using a final (MnSO4 )-7H2 O gas-assisted
milling step in order to produce suitable pillar geometries while keeping reasonable
milling times that minimize Ga C ion surface damage. After carving, the composite
samples were mounted in a special holder and transferred to the nanoindenter
system (Hysitron TI950 TriboIndenter). The load was applied in the direction of
the pillar axis using a 10 m diameter flat punch under displacement control
at a constant displacement rate of dh/dt D 40 nm/s, ensuring quasi-static loading
conditions similar to the ones used for the nanoindentation tests. Prior to each set
of tests, the composite samples were kept in the nanoindenter system during 1 h to
minimize the thermal drift.
During the compression loading, the behaviour of the pillar was initially linear
and elastic, until a point when the response curve starts to deviate from linearity
as the material undergoes plastic deformation. The initial slope of the load-
displacement curve, dP/dh, cannot be used directly to obtain the elastic modulus
of the polymer matrix, and its determination requires additional analysis for pillar
geometry corrections due to taper and curvature at the pillar base. Yang et al.
[32] determined the effect of micropillar geometry on the elastic modulus of the
material as
H .1 / D0 4H dP
Epillar D ; ˇ; 1C
D0 D0 8H D0 .D0 C 2H tan ˇ/ dh
(10.4)
10 Microscale Characterization Techniques of Fibre-Reinforced Polymers 293
Fig. 10.6 (a) Elastic modulus of 8552 epoxy resin carved pillars of different diameters, (b) yield
strength of 8552 epoxy resin carved pillars of different diameters
294 M. Herráez et al.
where the effect of the skin/core structure can be neglected. As a result, the elastic
modulus and yield strength of the pillars can be taken as the corresponding values of
the unaffected polymer, and the corresponding values are summarized in Table 10.3
for this situation. The results agree well with the data obtained from the indentation
tests presented in Table 10.2 in the previous section.
Fig. 10.7 Typical load-displacement push-in test curve. (a) AS4/8552 cross section showing
highly packed hexagonal fibre clusters, (b) atomic force microscope (AFM) image showing the
flat punch indenter footprint
curve is observed, Pc in Fig. 10.7). In this virtual push-in test, the interface
strength is introduced arbitrarily as an input of the cohesive traction-separation
law. Therefore, the virtual experiment has to be repeated for a range of interface
strength values, obtaining the corresponding critical loads from the virtual load-
displacement curves. At this point, using the shear-lag theory, the critical normal
load is transformed onto critical shear stress acting on the interface crit , and,
hence, a simple relationship between the critical shear load required to initiate the
debonding and the interface shear strength crit D f(S) is obtained. Finally, using
this relationship and the critical shear load measured in real experiments, the real
interface strength is obtained.
It should be noticed that neither push-in nor push-out tests provide the actual
value of the normal interface strength that may differ from the shear strength.
Ogihara and Koyanagi [40] studied the interface failure under combined shear and
normal stress states in a glass fibre/epoxy composite and concluded that the shear
strength of the interface was 1.5 times higher than the normal strength and proposed
a quadratic stress criterion to estimate the interface strength under multiaxial stress
states.
The fibre/matrix interface characterization was carried out using small coupons
extracted from the unidirectional AS4/8552 cured composite laminate. Samples
were first cut using a diamond wire and the cross section perpendicular to the fibres
polished using diamond slurry down to 1 m grain size. A typical cross section after
polishing is shown in Fig. 10.7a. All push-in tests were carried out using a Hysitron
TI950 TriboIndenter equipped with a 5 m diameter flat punch tip that was used to
target the centre of the 7 m AS4 carbon fibres with the aid of an optical microscope.
In order to achieve good reproducibility, fibre push-in tests were performed on the
central fibres of highly packed clusters with hexagonal symmetry, a feature easily
found in unidirectional AS4/8552 plies (Fig. 10.7b). A total of fifteen push-in tests
were carried out at room temperature and dry conditions.
The corresponding values for the interface shear strength of the material as well
as the estimation of the normal shear strength according to the criterion proposed
by Ogihara and Koyanagi [40] are summarized in Table 10.4 for the AS4/8552
composite material.
taking into account microstructural features. The book chapter presents a set of
experimental techniques that were applied to determine the relevant mechanical
parameters of fibres, matrix and fibre/matrix interface, governing the mechanical
behaviour of the AS4/8552 unidirectional carbon epoxy composite at the micro
scale. The strength and toughness of the individual AS4 carbon fibres was first
measured by means of tensile tests carried out on plain and notched individual
fibres, respectively (Table 10.1). To this end, FIB carving was used to introduce
a notch perpendicular to axis of the fibre in order to obtain the residual strength
and the corresponding fibre toughness. Matrix elastoplastic behaviour was also
characterized by means of in situ nanoindentation tests carried out in the resin
pockets of the AS4/8552 material or by uniaxial compression tests on micropillars
carved using FIB milling (Tables 10.2 and 10.3). Finally, interface shear strength in
the AS4/8552 composite was obtained through detailed push-in tests carried out on
individual fibres of the material (Table 10.4).
Further developments in the area of experimental micromechanics are expected
to continue in several directions. Firstly, carbon fibres are strongly anisotropic
solids, and their elastic modulus and strength in the fibre direction are normally well
documented in the literature. However, the transversal properties are rather difficult
to measure directly due to the obvious experimental difficulties. Secondly, although
the elastoplastic behaviour of the matrix is well characterized, the response under
dominated tensile loads is markedly dependent on the defect distribution, residual
stresses in the material, etc. that may vary locally throughout the microstructure.
Finally, interface shear strength can be measured using the push-in technique, but
the determination of the normal interface strength and toughness are still remaining
challenges to solve. Taking into account the aforementioned limitations, Focused
ion beam (FIB) carving technique opens revolutionary opportunities to generate the
appropriate test geometry at the micro level.
References
1. J. LLorca et al., Multiscale modeling of composite materials: a roadmap towards virtual testing.
Adv. Mater. 23, 5130–5147 (2011)
2. J. LLorca, C. González, J.M. Molina-Aldareguía, C.S. Lopes, Multiscale modeling of compos-
ites. Towards virtual testing : : : beyond. JOM 65, 215–225 (2013)
3. C. González, J. LLorca, Multiscale modeling of fracture in fiber-reinforced composites. Acta
Mater. 54, 4171–4181 (2006)
4. C. González, J. LLorca, Mechanical behavior of unidirectional fiber-reinforced polymers under
transverse compression: microscopic mechanisms and modeling. Compos. Sci. Technol. 67,
2795–2806 (2007)
298 M. Herráez et al.
29. J.H. Kim, S.J. Jeong, H.J. Lee, S.W. Han, B.I. Choi, S.H. Park et al., Linear analysis of the
viscoelastic response of polymer micro-pillars using the open-loop flat punch indentation test.
Philos. Mag. 86, 5679–5690 (2006)
30. D.R.P. Singh, N. Chawla, G. Tang, Y.L. Shen, Micropillar compression of Al/SiC nanolami-
nates. Acta Mater. 58, 6628–6636 (2010)
31. H. Fei, A. Abraham, N. Chawla, H. Jiang, Evaluation of micro-pillar compression tests for
accurate determination of elastic-plastic constitutive relations. J. Appl. Mech. 79, 061011
(2012)
32. Y. Yang, J.C. Ye, J. Lu, F.X. Liu, P.K. Liaw, Effects of specimen geometry and base material
on the mechanical behavior of focused-ion-beam-fabricated metallic-glass micropillars. Acta
Mater. 57, 1613–1623 (2009)
33. S. Wang, Y. Yang, L.M. Zhou, Y.W. Mai, Size effect in microcompression of epoxy micropil-
lars. J. Mater. Sci. 47, 6047–6055 (2012)
34. A. Kelly, W.R. Tyson, Tensile properties of fibre-reinforced metals: copper/tungsten and
copper/molybdenum. J. Mech. Phys. Solids 13, 329–338 (1965)
35. L.J. Broutmari, Measurement of the fiber–polymer matrix interfacial strength Interfaces in
composites. ASTM Int. 452, 27 (1969)
36. B. Miller, P. Muri, L. Rebenfeld, A microbond method for determination of the shear strength
of a fiber/resin interface. Compos. Sci. Technol. 28, 17–32 (1987)
37. M. Kharrat, A. Chateauminois, L. Carpentier, P. Kapsa, On the interfacial behaviour of a
glass/epoxy composite during a micro-indentation test: assessment of interfacial shear strength
using reduced indentation curves. Compos. A Appl. Sci. Manuf. 28, 39–46 (1997)
38. J.M. Molina-Aldareguía, M. Rodríguez, C. González, J. LLorca, An experimental and numer-
ical study of the influence of local effects on the application of the fibre push-in test. Philos.
Mag. 91, 1293–1307 (2011)
39. M. Rodríguez, J.M. Molina-Aldareguía, C. González, J. LLorca, A methodology to measure
the interface shear strength by means of the fiber push-in test. Compos. Sci. Technol. 72,
1924–1932 (2012)
40. S. Ogihara, J. Koyanagi, Investigation of combined stress state failure criterion for glass
fiber/epoxy interface by the cruciform specimen method’. Compos. Sci. Technol. 70, 143–150
(2010)
Chapter 11
Fibre Distribution and the Process-Property
Dilemma
John Summerscales
11.1 Introduction
11.1.1 Process
J. Summerscales ()
Advanced Composites Manufacturing Centre, School of Engineering, Plymouth University,
Drake Circus, Plymouth, PL4 8AA, UK
e-mail: jsummerscales@plymouth.ac.uk
viscosity term. Equation 11.1 is for the isotropic case, although the anisotropy
within composites normally requires the tensor relationships:
Q D kA P=L (11.1)
where " is the porosity (1 Vf ), Vf is the fibre volume fraction, m is the mean
hydraulic radius and k is the Kozeny constant.
Blake [16] defined the hydraulic radius, m (Eq. 11.3), as the volume in which
fluid actually flows, "V (where V D AL), divided by the wetted surface area (S):
m D ":V=S (11.3)
Until fibres touch, the increase in surface area [17] will be linear with volume
fraction (Vf 2 is substituted for S2 ) as in expression 11.5:
3
∴ K ˛ 1 Vf =Vf2 or "3 = 1 "2 (11.5)
Summerscales [18] showed that as fibres gather into increasingly large clusters with
the interfibre spaces closed off from flow, the permeability increases as the wetted
area decreases despite the reduction in effective flow area due to the inaccessible
interfibre volumes.
11 Fibre Distribution and the Process-Property Dilemma 303
Wang and Grove [19] and Wang et al. [20] used a dual-scale resin infusion
model to show a characteristic relationship between tow impregnation speed, the
macro-scale resin pressure surrounding the tow and the degree of tow saturation.
They introduced a sink and a source term into the two different scaled flows (intra-
tow and inter-tow) to couple them together.
The estimation of the permeability of “real” reinforcements is now a feature of a
number of 3D flow modelling and textile design software packages, e.g. FiberSIM,
TexGen and WiseTex (Celper and FlowTex).
11.1.2 Properties
Ec D d l o Vf Ef C Vm Em (11.6)
In Eqs. 11.6 and 11.7, the total volume fraction for a monolithic fibre composite will
be given by Eq. 11.8:
Vf C Vm C Vv D 1 (11.8)
where subscript v indicates the void content. In an ideal composite before resin
impregnation, Vv is the volume available for flow (" D 1 Vf with Vm D 0). How-
ever, few processes produce perfect materials, so porosity (connected pores) and
voids (separate bubbles) do occur in composites. The causes may include (1) air
304 J. Summerscales
Table 11.1 Fibre volume fraction and void content of RCF/VER composites (extended
from Alhuthali and Low [28] data)
Fibres Fibre Theoretical Experimental [By JS]
weight volume density density Void %void
Sample fraction fraction (kg/m3 ) (kg/m3 ) content (%) in resin
20 % RCF/VER 0:2 0:16 1210 1180 ˙ 30 5:55 6.61 %
30 % RCF/VER 0:3 0:24 1240 1210 ˙ 20 4:75 6.25 %
40 % RCF/VER 0:4 0:33 1270 1240 ˙ 50 3:28 4.89 %
50 % RCF/VER 0:5 0:43 1310 1270 ˙ 20 2:74 4.81 %
in the resin mix, (2) volatiles in the resin “boiling”, (3) the degree of impregnation
of pre-impregnated materials, (4) trapping air between prepreg layers or (5) race-
tracking in LCM processes.
Judd and Wright [25], Ghiorse [26] and Baley et al. [27] have reviewed
porosity/voids in composites. Judd and Wright concluded that “although there is
a considerable scatter in results (reflecting in part the difficulties of accurate void
content determination) the available data show that the inter-laminar shear strength
of composites decreases by about 7 per cent for each 1 per cent voids up to at least
the 4 per cent void content level, beyond which the rate of decrease diminishes.
Other mechanical properties may be affected to a similar extent. This is true for
all composites regardless of the resin, fibre or fibre surface treatment used in their
fabrication”. See Table 1 of Judd and Wright [25] for a comprehensive analysis of
the data. It may be time to consider whether the above findings remain relevant to
modern toughened resin systems.
Stone and Clarke [28] reported that (a) below Vv D 1.5 %, voids tend to be volatile
induced and hence spherical with diameters in the range 5–20 m, while (b) above
Vv D 1.5 %, the voids are flattened and elongated in the in-plane direction due to
the limitation of space between the fibre bundles and are also significantly larger
than those voids at a lower Vv . Mayr et al. [29] reported that small pores in CFRP
with porosity levels <1.8 % often have roughly circular cross-sections and found an
abrupt increase in the out-of-plane shape factors at this percentage porosity.
Purslow [30] proposed a novel classification system for voids and considered
the earlier system was only significant for fairly uniformly distributed voids. For
example, to quote a Vv of 0.5 % for a composite of generally high quality (voids
<0.2 %) but with an occasional very large void could be very misleading and
potentially dangerous. It is difficult to measure void contents to such low values. He
suggested that the void content should be quoted as “0 < voids < 0.2 %; infrequent
local voids >0.5 %”. His studies have suggested that when Vv < 0.5 %, the voids are
spherical with a diameter of 10 m and are due to trapped volatiles. As Vv increases,
the voids due to trapped volatiles decrease in number and are replaced by large intra-
tow/intra-laminar voids. The results suggested a linear relationship between Vv and
void thickness, where the thickness is related to fibre diameter.
Alhuthali and Low [31] reported densities, fibre volume fractions and void
contents for recycled cellulose fibres (RCF, density 1540 kg/m3 ) in vinyl-ester
11 Fibre Distribution and the Process-Property Dilemma 305
resin (VER, density 1140 kg/m3 ) (Table 11.1). As fibre weight fraction increased,
the theoretical and experimental composite densities rose, but there was a clear
reduction in void content at a greater rate than predicted from the reduced matrix
volume fraction (column 7 of the Table).
Close [32] has published an interesting account of what remains when you take
all the matter away.
The use of microscopy to reveal fine detail in structures has a long history [33]. The
distribution of features within a plane or volume may be quantified in a variety of
ways. The classification of structured populations can be achieved using a variety
of parameters, e.g. (a) nearest-neighbour index [34], (b) chi-squared analysis [35],
(c) quadrat analysis [36], (d) mean free path and mean random spacing [37], (e)
space autocorrelograms [38], (f) area fraction variance analysis and mean intercept
length [39] and (g) contiguity index [40]. In recent years, the quantification of
microstructures has been achieved using tessellation techniques or, latterly, fractal
dimensions (FD) which are reviewed in the following sections.
The nature of fibre-reinforced composites is such that there is generally dual-
scale structure with clustering of fibres in bundles (tows) and larger features dictated
by the reinforcement architecture (e.g. chopped strand mat, woven fabrics or stitched
non-crimp fabrics (NCF)). The microstructure of fibre-reinforced composites is
normally defined using the parameters in Eq. 11.6. This data may be insufficient
where clustering of fibres occurs. The use of image processing and analysis for
the characterisation of composite micro-/meso-structures has been the subject of a
number of publications (e.g. [41–45]).
The morphological description of the dispersion of phases at the micro-/meso-
structural level is an important factor for the determination of the overall behaviour
of the macro-composites. The micro-failure threshold is dominated by fluctuations
in the local stress field. These “hot spots” are a function of the local disorder of the
reinforcement [42].
The analysis of a plane with a number of nonoverlapping objects (e.g. fibre cross-
sections in a micrograph) can be achieved by identifying the centroid of each object
and using that set of points. The Delaunay [46] tessellation simply joins adjacent
points to produce a series of contiguous triangles where no point is inside the
circumcircle of any triangle. The Voronoi diagram (also known as the Dirichlet
tessellation, which is the complement (dual) of the Delaunay tessellation) partitions
306 J. Summerscales
Fig. 11.1 Delaunay (left) and Voronoi (right) tessellations of a set of points
the plane into convex polygons where each polygon contains just one generating
point and every point in each individual polygon is closer to its own generating
point than to any of the other points. The cells are called Dirichlet regions, Thiessen
polytopes or Voronoi polygons [47–49]. For each Delaunay tessellation, there will
be a unique Voronoi diagram and vice versa (Fig. 11.1). The lines between the
generating points in the Delaunay tessellation will be orthogonal to the edge of
the Voronoi polygons [50].
Summerscales et al. [51] studied unidirectional carbon-fibre reinforced-epoxy-
composites manufactured using vacuum-bag processing with different “dwell
times” before pressure was applied to consolidate the laminates. The cross-section
microstructure was described using the size of the Voronoi cell around each fibre
and found to be dependent on the manufacturing parameters. Samples with a 90-
min (rather than 0- or 180-min) dwell time were found to have the lowest thickness
(highest fibre volume fraction) and to be the least clustered (i.e. the most uniform
fibre distribution). The plate was believed to be processed within the optimum
viscosity limits (7500–16,500 mPa.s) identified by Stringer [52].
Griffin et al. [53] conducted permeability experiments on a series of five 2 2
twill weave carbon-fibre fabrics containing a variable number of twisted/bound
flow-enhancing tows (FET) in the weft direction. The permeability was found to
be dependent on the flow fluid used, but values for a single fluid showed the
expected increase in permeability with number of FET. Quantitative microscopy
revealed that flow enhancement was accompanied by the presence of large flow
areas adjacent to the FET and increased in line with the average perimeter of these
areas. The transverse section composite microstructure was characterised using
maximum feature height and width (vertical and horizontal ferets, i.e. the two
orthogonal pairs of parallel tangents on the outer boundaries of a feature), x and
y centres of gravity and perimeter. The recorded perimeters were broadly aligned
with the expectation of the Blake-Kozeny-Carman equations (Eqs. 11.2, 11.3 and
11.4) although it was not practical to determine the effective penetration of the
flow front into the tows during wetting. Griffin et al. [54] used three of the above
reinforcements to measure the total number of areas, and values of total area, for
the three laminates (Table 11.2) and used the zones of influence (Voronoi cells) to
11 Fibre Distribution and the Process-Property Dilemma 307
Table 11.2 Pore areas and for the 2 2 twill weave laminates
Total area of Normalised flow
Number of pore space rate (centipoise
Fabric/laminate % FET flow areas (mm2 ) mm/s) (sic)
Twill 0:0 210 5:47 796
156 12:5 115 8:54 5200
126 50:0 310 11:21 7982
compare fibre separations within either the normal tows (Vf D 40 %) or the FET
(Vf D 57 %) in the orthogonal horizontal/vertical directions.
Summerscales et al. [55] used a radial flow permeameter to measure the perme-
abilities of the above five fabrics and found increasing permeability with increasing
number of FET (Table 11.3). Basford et al. [56] measured the mechanical properties
of laminates manufactured with these fabrics and found that both compression and
apparent inter-laminar shear strengths (ILSS) decreased with increasing proportion
of FET (Table 11.3).
Everett [57] used Radial Distribution Functions (RDF) to describe composite
material microstructures and reported that the proximity of near-neighbours has
important ramifications for the micromechanical modelling of tow-based compos-
ites. Bertoncelj et al. [58] used the RDF approach in conjunction with Voronoi
diagrams to quantify random fibre arrangements.
The use of tessellation normally produces data as histograms. Selection of
specific subsets of that data may correlate to measured material parameters, but
the choice of subset is normally subjective.
Fig. 11.2 Representative images showing selected stages of fractal data generation for the
resin/void volume fraction
the figure has been decomposed [59]. The total number of objects, N, is a function
of the magnification factor, r, and the dimension, D, as given by N D rD .
In Euclidean geometry, the exponent will be an integer, but in fractal geometry,
it can be a real number. Fractal geometry is the complement of Euclidean geometry
and crystal symmetry [42]. Richardson [60] pioneered a process for calculating the
length of a coastline using rulers of varying lengths which was popularised as fractal
geometry by Mandelbrot [61, 62].
The fractal dimension, FD, measures the complexity of a self-similar object.
The box-counting method (BCM) extends the perimeter measuring method used
for coastlines by covering the image with a grid. The numbers of elements of the
grid which cover features within the image are counted. As increasingly fine grids
(smaller boxes) are used, the structure of the pattern is more accurately captured,
but now N is the number of boxes that cover the features (Fig. 11.2). The BCM has
been widely used in fractal dimension research [63–66].
The slope of the Richardson plot (measured length plotted against the size of
the measure or, for the box-counting method, detected area plotted against box size
with both plots on a log-log scale) gives the fractal dimension. The definition of the
fractal dimension (FD) for a self-similar object is given by Eq. 11.9:
Both the Richardson plot and BCM appear to be applicable for patterns with or
without self-similarity [63, 65, 67, 68].
11 Fibre Distribution and the Process-Property Dilemma 309
Worrall and Wells [69] used fractal variance analysis to characterise differences
in filamentisation (i.e. fibre separation) between bundled and filamentised press-
moulded long discontinuous glass-fibre/polyester resin composites. Changes in
the Richardson plot were used to identify changes in the optical micrographs of
composite structures.
Dzenis et al. [70] used an atomic force microscope to analyse the “self-affinity” of
surface topography for several reinforcement fibres. The fractal dimensions of the
®
surfaces were determined as 2.09 (“graphite”), 2.37 (Kevlar 149), 2.52 (undrawn
polyimide) and 2.18 (polyimide with a draw ratio of 8). Dzenis [71] then defined the
interfacial fractal dimension as smooth interface (2.0), moderate irregularity (2.4)
and very strong irregularity (2.8). He further presented analytical formulations for
a variety of elastic, viscoelastic and thermal properties of unidirectional composites
in both the axial and transverse directions.
Pearce et al. [72] investigated the effects of fabric architecture on the processing
and properties of 6 K carbon-fibre reinforced-composites using three different
weave styles with the same areal weight (290 gsm) and woven from the same batch
of fibre. Composites were produced by resin transfer moulding in an unsaturated
radial flow permeameter using 9 (Vf D 49 %), 10 (Vf D 54 %) or 11 (Vf D 60 %)
layers. Pore space was determined from transverse sections. Inter-laminar shear
strength was measured in accordance with CRAG standard method 100 [73].
The permeabilities for the three fabrics were ranked in the same sequence as the
proportion of larger pore spaces (>0.5 mm2 ) and ranked in the same order as
descending ILSS (Table 11.4).
In further work on the same fabrics, Pearce et al. [74] added tensile and com-
pressive tests and fractal analysis of transverse sections to obtain fractal dimensions
for the three fabrics. Variations in permeabilities and mechanical properties were
related to observed differences in the microstructure. The ranking of ILSS and both
tensile and compressive moduli was satin > twill > Injectex. For both the tensile and
compressive ultimate strengths, the satin fabric again had the highest properties,
with twill generally having marginally better properties than the Injectex. The
compressive strengths were lower (around one-half) than the respective tensile
strengths for all fabrics as expected due to the pre-buckling of fibres arising from
the crimp of the fabrics. Table 11.5 shows the relative magnitudes of the respective
properties and the fractal dimension for each type of fabric.
Pearce et al. [75] sought to correlate variations in permeability and in the
laminate mechanical properties, to differences in microstructure using a series of
experimental carbon-fibre fabrics woven to incorporate a novel flow enhancement
concept (use of different proportions of 3 K weft tows in a 6 K fabric). Permeability
data was derived from measurements made during manufacture of the composite
plates in a transparent radial flow RTM mould. The manufactured plates were
subsequently sectioned for mechanical testing (moduli and strengths in tension and
compression) and for automated image analysis to determine fractal dimensions
from polished sections. The increase in permeability was not consistent with the
proportion of flow-enhancing (3 K) tows, but could be ranked in the same sequence
as the fractal dimension derived from polished sections (Fig. 11.3). Note that
the software used to generate the data returns FD values reduced by one. The
compression strengths were unaffected by the change in tow size. The weft direction
had lower moduli in both tension and compression. The rate at which modulus
decreases with increasing proportion of 3 K tows was broadly consistent with the
change in volume fraction of fibres in the test direction. The weft tensile strength
was ranked in the same sequence as the fractal dimension (Fig. 11.4).
Using the single average data set in Fig. 11.4, we might speculate that extrap-
olation of the weft tensile strength Excel Trendline to integer values could yield
important information. At FD D 0, the intercept at 97 MPa is of a similar magnitude
to that expected by extrapolation of the cast resin tensile strengths for the Ampreg 26
epoxy/26SL slow hardener system (the Gurit Ampreg 26 epoxy laminating system
manufacturers data sheet PDS-Ampreg 26-12-0515 gives room temperature cure
58 MPa, 50 ı C post-cure 80.3 MPa, 80 ı C post-cure (as used) strength not given).
At FD D 1, the prediction of 1540 MPa is of the order of the strength of a carbon
fibre that has been through textile processes (the original paper/thesis that provided
11 Fibre Distribution and the Process-Property Dilemma 311
Fig. 11.3 Permeability plotted against fractal dimension (from resin) for Brochier and Carr
Reinforcements fabrics
the data does not specify the specific grade of fibre used). Subsequent work by
different researchers has not reproduced the above finding.
Summerscales et al. [76] quantified 150 X-ray CT cross-sectional slices of plain
woven E-glass-fibre-reinforced epoxy resin using fractal dimension. The analysis
returned a consistent numerical value for each of the slices in the two similar
orthogonal planes.
Aniszewska and Rybaczuk [77] presented simulated quasi-static fracture pro-
cesses for parallel-fibre-reinforced composites using cellular automata and BCM
fractal characteristics of defect growth. Defect evolution in composites was treated
as a dynamical system depending on external and internal conditions and properties
of fibres. The simulations were intended to inform further work on the evolution of
fractal characteristics for 3D textile-reinforced aluminium matrix composites.
Pimenta and Pinho [78] have presented a model for the translaminar tensile
toughness of fibre-reinforced polymer matrix composites. The model is based on
fibre and interfacial properties and assumes a hierarchical failure process with the
formation of stochastic variations of quasi-fractal fracture surfaces. The model could
reproduce the effect of different types of fibre and matrix and revealed a marked
increase in toughness for thicker plies.
312 J. Summerscales
Fig. 11.4 Tensile and compressive strengths for Carr twill fabrics with differing proportions
of flow-enhancing tows. Warp tensile and compressive and weft compressive strengths sensibly
constant but weft tensile strength rises with fractal dimension (not with fibre volume fraction)
Acknowledgements The author acknowledges with sincere thanks the collaborations with Pro-
fessor Felicity Guild as the primary stimulus for his interest in microstructural characterisation.
Further acknowledgement is due to Paul Russell who did much of the image analysis in the early
studies. The various co-authors of referenced papers were also critical to our achievements in these
studies. Finally, I very much appreciate the review by Stephen Grove of a late-draft manuscript.
Any remaining issues are the responsibility of the chapter author.
References
1. C.D. Rudd, A.C. Long, K.N. Kendall, C.G.E. Mangin, Liquid Moulding Technologies (Wood-
head Publishing, Cambridge, 1997). ISBN 1-85573-242-4
2. T.M. Kruckenberg, R. Paton, Resin Transfer Moulding for Aerospace Structures (Kluwer
Academic, Dordrecht NL, 1998). ISBN 0-412-73150-9
3. R.S. Parnas, Liquid Composite Moulding (Hanser Gardner Publications, 2000). ISBN 1-569-
90287-9
4. K. Potter, Resin Transfer Moulding (Chapman & Hall, London, 1997). ISBN 0-412-72570-3
5. D. Cripps, T.J. Searle, J. Summerscales, in Open Mould Techniques for Thermoset Composites,
eds. by R. Talreja, J.-A. Månson. Comprehensive Composite Materials Encyclopædia, vol 2
Polymer Matrix Composites (Elsevier Science, Oxford, 2000), Chapter 21, pp. 737–761. ISBN
0-08-043725-7. doi:10.1016/B0-08-042993-9/00188-1
6. J. Summerscales, Resin Infusion Under Flexible Tooling (RIFT). Encyclopedia of Composites,
2nd edn. (Wiley, 2012), pp. 2648–2658. doi:10.1002/9781118097298.weoc216
7. J. Summerscales, T.J. Searle, Review: low pressure (vacuum infusion) techniques for moulding
large composite structures. Proc IMechE Part L J. Mater. Design Appl. L219(1), 45–58 (2005).
doi:10.1243/146442005X10238
8. C.D. Williams, J. Summerscales, S.M. Grove, Resin infusion under flexible tool-
ing (RIFT): a review. Compos. Part A Appl. Sci. Manuf. 27A(7), 517–524 (1996).
doi:10.1016/1359-835X(96)00008-5
9. H.P.G. Darcy, Les fontaines publiques de la ville de Dijon (Dalmont, Paris, 1856)
10. R. Arbter, J.M. Beraud, C. Binetruy, L. Bizet, J. Bréard, S. Comas-Cardona, C. Demaria,
A. Endruweit, P. Ermanni, F. Gommer, S. Hasanovic, P. Henrat, F. Klunker, B. Laine, S.
Lavanchy, S.V. Lomov, A. Long, V. Michaud, G. Morren, E. Ruiz, H. Sol, F. Trochu, B.
Verleye, M. Wietgrefe, W. Wu, G. Ziegmann, Experimental determination of the permeability
of textiles: a benchmark exercise, Compos. Part A Appl. Sci. Manuf. 42(9), 1157–1168 (2011).
doi:10.1016/j.compositesa.2011.04.021
11. N. Vernet, E. Ruiz, S. Advani, J.B. Alms, M. Aubert, M. Barburski, B. Barari, J.M. Beraud,
D.C. Berg, N. Correia, M. Danzi, T. Delavière, M. Dickert, C. Di Fratta, A. Endruweit, P.
Ermanni, G. Francucci, J.A. Garcia, A. George, C. Hahn, F. Klunker, S.V. Lomov, A. Long,
B. Louis, J. Maldonado, R. Meier, V. Michaud, H. Perrin, K. Pillai, E. Rodriguez, F. Trochu, S.
Verheyden, M. Wietgrefe, W. Xiong, S. Zaremba, G. Ziegmann, Experimental determination
of the permeability of engineering textiles: Benchmark II. Compos. Part A Appl. Sci. Manuf.
61, 172–184 (2014). doi:10.1016/j.compositesa.2014.02.010
12. F.N. Cogswell, Flow processes in composite materials. Adv. Compos. Bull. 1(10), 1–5 (1988)
13. J. Summerscales, The effect of permeant on the measured permeability of reinforce-
ment. 7th international conference on flow processes in composite materials (FPCM-7),
Newark DE, 7–9 July 2004, 471–476. https://www.fose1.plymouth.ac.uk/sme/fpcm/fpcm07/
Extended_abstracts/EA471.pdf
14. P.C. Carman, Fluid flow through a granular bed. Trans. Inst. Chem. Eng. (Lond.) 15, 150–166
(1937)
15. J. Kozeny, Uber die kapillare Leitung des Wassers in Boden (Sitzungsberichte Akademie der
Wissenschaft Wien Math-naturw, 1927) 139(Kl.abt.IIa), 271–306
314 J. Summerscales
16. F.C. Blake, The resistance of packing to fluid flow. Trans. Am. Inst. Chem. Eng. 14, 415–421
(1922)
17. J. Summerscales, in Appendix B: Fabric Permeability, eds. by D. Cripps, T.J. Searle, J.
Summerscales. Chapter 21: Open Mould Techniques for Thermoset Composites. in R. Talreja,
J.-A. Månson (editors): Comprehensive Composite Materials, volume 2: Polymer Matrix
Composites, (Elsevier Science, Oxford, 2000), pp. 737–761. Set ISBN: 0-08-043720-6.
doi:10.1016/B0-08-042993-9/00188-1
18. J. Summerscales, A model for the effect of fibre clustering on the flow rate in resin transfer
moulding. Compos. Manuf. 4(1), 27–31 (1993). doi:10.1016/0956-7143(93)90013-X
19. Y. Wang, S.M. Grove, Continuum dual-scale modelling of liquid composite moulding pro-
cesses. J. Reinf. Plast. Compos. 28(12), 1469–1484 (2009). doi:10.1177/0731684408089533
20. Y. Wang, M. Motamedi, S.M. Grove, Modelling microscopic flow in woven fabric reinforce-
ments and its application in dual-scale resin infusion modelling. Compos. Part A Appl. Sci.
Manuf. 39(5), 843–855 (2008). doi:10.1016/j.compositesa.2008.01.014
21. A.S. Virk, W. Hall, J. Summerscales, Modulus and strength prediction for natural fibre com-
posites. Mater. Sci. Technol. 28(7), 864–871 (2012). doi:10.1179/1743284712Y.0000000022
22. J. Summerscales, W. Hall, A.S. Virk, A fibre diameter distribution factor (FDDF) for natural
fibre composites. J. Mater. Sci. 46(17), 5876–5880 (2011). doi:10.1007/s10853-011-5569-6.
23. H.L. Cox, The elasticity and strength of paper and other fibrous materials. Br. J. Appl. Phys.
3(3), 72–79 (1952). doi:10.1088/0508-3443/3/3/302
24. H. Krenchel, Fibre Reinforcement (Akademisk Forlag, Copenhagen, 1964)
25. N.C.W. Judd, W.W. Wright, Voids and their effects on the mechanical properties of composites
- an appraisal. SAMPE J. 14(1), 10–14 (1978)
26. S.R. Ghiorse, Effect of void content on the mechanical properties of carbon/epoxy laminates.
SAMPE Q. 24(2), 54–59 (1993)
27. C. Baley, P. Davies, D. Cartié, Porosity in ocean racing yacht composites: a review. Appl.
Compos. Mater. 22(1), 13–28 (2015). doi:10.1007/s10443-014-9393-4
28. D.E.W. Stone, B. Clarke, Ultrasonic attenuation as a measure of void content
in carbon-fibre reinforced plastics. Non-Destr. Test. 8(3), 137–145 (1975).
doi:10.1016/0029-1021(75)90023-7
29. G. Mayr, B. Plank, J. Sekelja, G. Hendorfer, Active thermography as a quantitative method for
non-destructive evaluation of porous carbon fibre reinforced polymers. NDT & E Int. 44(7),
537–543 (2011)
30. D. Purslow, On the optical assessment of the void content in composite materials. Composites
15(3), 207–210 (1984). doi:10.1016/0010-4361(84)90276-3
31. A. Alhuthali, I.M. Low, Effect of prolonged water absorption on mechanical properties in
cellulose fibre reinforced vinyl-ester composites. J. Mater. Sci. 48(18), 6331–6340 (2013).
doi:10.1007/s10853-013-7432-4
32. F. Close, Nothing: A Very Short Introduction (Oxford University Press, Oxford, 2009). ISBN
978-0-19-922586-6
33. T. Allen, Microscopy: A Very Short Introduction (Oxford University Press, Oxford, 2015).
ISBN: 978-0-19-870126-2
34. P.J. Clark, F.C. Evans, Distance to the nearest neighbor as a measure of spatial relationships in
populations. Ecology 35(4), 445–453 (1954). doi:10.2307/1931034
35. P. Davis, Describing point patterns, in Science in Geography Book 3: Data Description and
Presentation (Oxford University Press, Oxford, 1974), pp. 29–35. ISBN 978-0-19-913067-2
36. P. Greig-Smith, The use of random and contiguous quadrats in the study of the structure of
plant communities, Ann. Bot. NS16(2), 293–316 (1952)
37. W.R. Cribb, Quantitative metallography of polyphase microstructures. Scripta Metallurgica
12(10), 893–898 (1978). doi:10.1016/0036-9748(78)90177-1
38. F.S.M.A. Mirza, A statistical study of the structure of mixtures in particulate solids, PhD thesis,
University of Exeter, 1970
39. Q.F. Li, R. Smith, D.G. McCartney, Quantitative evaluation of fiber distributions in a
continuously reinforced aluminium alloy using automatic image analysis. Mater. Charact.
28(4), 189–203 (1992). doi:10.1016/1044-5803(92)90081-R
11 Fibre Distribution and the Process-Property Dilemma 315
40. D. Short, J. Summerscales, The definition of microstructures in hybrid reinforced plastics. 5th
International Conference of the European Chapter of SAMPE, Montreux Switzerland, 12–14
June 1984, vol. 2, paper 19
41. F.J. Guild, J. Summerscales, Microstructural image analysis applied to fibre composite
materials: a review. Composites 24(5), 383–394 (1993). doi:10.1016/0010-4361(93)90246-5
42. R. Pyrz, Chapter 1.16: Morphological characterization of microstructures, in Comprehensive
Composite Materials, vol. 1, Fiber Reinforcements and General Theory of Composites
(Elsevier, 2000a), pp 465–478. doi:10.1016/B0-08-042993-9/00053-X
43. R. Pyrz, Chapter 2.15: The Application of Morphological Methods to Composite Materials,
in Comprehensive Composite Materials, Volume 2: Polymer Matrix Composites (Elsevier,
2000b), pp. 553–576. doi:10.1016/B0-08-042993-9/00071-1
44. J. Summerscales (ed.), Microstructural Characterisation of Fibre-Reinforced
Composites (Woodhead Publishing, Cambridge, 1998). ISBN 1-85573-240-8.
doi:10.1533/9781855737563
45. J. Summerscales, N.R.L. Pearce, P. Russell, F.J. Guild, Voronoi cells, fractal dimensions and
fibre composites. J. Microsc. 201(2), 153–162 (2001). doi:10.1046/j.1365-2818.2001.00841.x
46. B. Delaunay, Sur la sphère vide. A la mémoire de Georges Voronoï. Bulletin de l’Académie
des Sciences de l’URSS. Classe des sciences mathématiques et naturelles 1934(6), 793–800
(1934)
47. G.L. Dirichlet, Über die Reduktion der positiven quadratischen Formen mit drei unbestimmten
ganzen Zahlen. Journal für die reine und angewandte Mathematik 40, 209–227 (1850).
doi:10.1515/crll.1850.40.209
48. G. Voronoi, Nouvelles applications des paramètres continus à la théorie des formes quadra-
tiques. Journal für die Reine und Angewandte Mathematik 133, 97–178 (1907)
49. E.W. Weisstein, Voronoi Diagram, MathWorld - A Wolfram Web Resource (2015), http://
mathworld.wolfram.com/VoronoiDiagram.html. Accessed 07 Nov 2015
50. T. Ringler, in Introduction to Voronoi diagrams and Delaunay Triangulations, Summer
Colloquium on Numerical Techniques for Global Atmospheric Models, National Center
for Atmospheric Research (NCAR), Boulder CO, 02–13 June 2008, paper 25. http://
www.cgd.ucar.edu/cms/pel/asp2008/25-Ringler-VoronoiDelaunay.pdf. Accessed 07 Nov 2015
51. J. Summerscales, D. Green, F.J. Guild, Effect of processing dwell-time on the
microstructure of a fibre reinforced composite. J. Microsc. 169(2), 173–182 (1993).
doi:10.1111/j.1365-2818.1993.tb03292.x
52. L.G. Stringer, Optimization of the wet lay-up/vacuum bag process for the fabrication of carbon
fibre epoxy composites with high fibre fraction and low void content. Composites 20(5), 441–
452 (1989). doi:10.1016/0010-4361(89)90213-9
53. P.R. Griffin, S.M. Grove, P. Russell, D. Short, J. Summerscales, F.J. Guild, E. Taylor, The
effect of reinforcement architecture on the long range flow in fibrous reinforcements. Compos.
Manuf. 6(3/4), 221–228 (1995a). doi:10.1016/0956-7143(95)95015-Q
54. P.R. Griffin, S.M. Grove, F.J. Guild, P. Russell, J. Summerscales, The effect of microstructure
on flow promotion in resin transfer moulding reinforcement fabrics. J. Microsc. 177(3), 207–
217 (1995b). doi:10.1111/j.1365-2818.1995.tb03552.x
55. J. Summerscales, P.R. Griffin, S.M. Grove, F.J. Guild, Quantitative microstructural examina-
tion of RTM fabrics designed for enhanced flow. Compos. Struct. 32(1–4), 519–529 (1995).
doi:10.1016/0263-8223(95)00025-9
56. D.M. Basford, P.R. Griffin, S.M. Grove, J. Summerscales, Research report: the relationship
between mechanical performance and microstructure in composites fabricated with flow-
enhancing fabrics. Composites 26(9), 675–679 (1995). doi:10.1016/0010-4361(95)98917-A
57. R.K. Everett, Quantification of random fiber arrangements using a radial distribution function
approach. J. Compos. Mater. 30(6), 748–758 (1996). doi:10.1177/002199839603000606
58. B. Bertoncelj, K. Vojisavljević, J. Rihtaršič, G. Trefalt, M. Huskić, E. Žagar, B. Malič,
A Voronoi-diagram analysis of the microstructures in bulk-molding compounds and its
correlation with the mechanical properties. Expr. Polym. Lett. 10(6), 493–505 (2016)
316 J. Summerscales
79. D.M. Shotton, Confocal scanning optical microscopy and its applications for biological
specimens. J. Cell Sci. 94, 175–206 (1989)
80. M. Ferrando, W.E.L. Spiess, Review: Confocal scanning laser microscopy. A
powerful tool in food science. Food Sci. Technol. Int. 6(4), 267–284 (2000).
doi:10.1177/108201320000600402
81. J.R. Helliwell, C.J. Sturrock, K.M. Grayling, S.R. Tracy, R.J. Flavel, I.M. Young, W.R.
Whalley, S.J. Mooney, Applications of X-ray computed tomography for examining biophysical
interactions and structural development in soil systems: a review. Eur. J. Soil Sci. 64(3), 279–
297 (2013). doi:10.1111/ejss.12028
82. E. Maire, P.J. Withers, Quantitative X-ray tomography. Int. Mater. Rev. 59(1), 1–43 (2014).
doi:10.1179/1743280413Y.0000000023
83. J.D. Miller, C.L. Lin, A.B. Cortes, A review of x-ray computed tomography and its
applications in mineral processing. Miner. Process. Ext. Metall. Rev. 7(1), 1–18 (2007).
doi:10.1080/08827509008952663
Chapter 12
Analysis of Defect Developments in Composite
Forming
12.1 Introduction
Three types of defects that may appear during forming of textile composite rein-
forcement or prepreg are analyzed and modeled in the present chapter. Wrinkling is
one of the flaws that occur most often. Wrinkles are frequent because of the weak
textile bending stiffness due to possible slippage between fibers. It is necessary to
simulate their onset but also their growth and their shape in order to verify that they
don’t extend to the useful part of the preform. The influence of the different rigidities
of the textile reinforcement is analyzed in Sect. 12.2. The concept of “locking angle”
is questioned. If in-plane shear stiffness plays a main role in the onset of wrinkles,
it will be shown that there is no direct relation between shear angle and wrinkling.
The second type of defect analyzed in this chapter is the slippage that can
occur between warp and weft yarns of a woven reinforcement. This slippage can
happen during a forming process and leads to a lack of cohesion of the textile
reinforcement. The modeling and the simulation of this phenomenon must be done
at the mesoscopic scale in order to be able to describe the relative motions of
the yarns and the friction loads. In this situation the continuity assumption of the
macroscopic macroscale approach is not valid.
A third type of unusual behavior of fibrous composite reinforcements that can be
seen as a flaw during their forming process is the onset of peculiar “transition zones”
that are directly related to the bending stiffness of the fibers. The standard continuum
mechanics of Cauchy is not sufficient to model these defects. A second-gradient
approach is presented that allows to account for such behaviors and to master their
onset and development during forming process simulations. This approach also
allows to understand and to simulate the S shape obtained when an unbalanced
woven reinforcement is submitted to a bias extension test.
global compressive state [17]. More frequently one ply (or some plies) in the stack
is submitted to a compressive stress in the fiber direction leading to wrinkling of the
ply and of its neighbors.
The mechanical problem is complex. The laminate is made of UD plies with very
anisotropic mechanical behavior. The tensile stiffness in fiber direction is large, but
the in-plane shear stiffness and the tensile stiffness in the direction transverse to the
fibers are low. In addition compression in the fiber direction can lead to buckling
because of the small thickness and bending stiffness of the ply.
The different plies are in contact with friction with their neighbors. The friction
influences the stresses in the plies. The geometry of the shaped part associated
to the direction of the plies, to the mechanical properties of the plies, and to the
friction between plies can lead, in some cases, to compressive stresses in the fiber
direction of some plies and to wrinkling. The stacking sequence plays a major role in
wrinkling onset. The change of this sequence can reduce the out-of-plane wrinkling
[17]. Nevertheless this is not without consequence on the cured composite part
mechanical properties.
A mechanism for the formation of wrinkles in UD prepreg draping has been
shown in [18]. For example, a set of two plies, respectively, at 90ı and 0ı are laid
down, in that order, in a U-shaped mold. The first 90ı ply correctly fits the shape and
has a good contact even in the corner because of its weak property in the direction
perpendicular to the fibers. On the other hand, the 0ı ply is much stiffer, and the
contact in the radius is not assured (bridged ply). The laminate is consolidated under
autoclave pressure which ensures that the 0ı ply is in contact with the radius. The
resulting 0ı ply movement and the friction between the plies lead the 90ı ply to
form a wrinkle [18].
Dynamic explicit approaches [23–25] are well suited to both forming and
wrinkling simulations. The principle of virtual work associated to a finite element
approximation leads to
:
MuR C Cu D Fext Fint (12.1)
M and C are the mass and damping matrices. Fext and Fint are the exterior and
interior nodal loads, and u is the nodal displacement vector. In explicit codes, this
equation is solved on each time step ti D tiC1 ti using the central difference
scheme:
: i1=2 1 i1 : i1=2
u iC1
Du C u
i
C t C ti M1 Fiext Fiint Cu ti
2 D
(12.2)
The masse matrix is made diagonal in MD [26]. The stability of the scheme needs
to use sufficiently small time increments [27]. In composite reinforcement forming
or prepreg thermoforming, shell finite elements are usually used. The specific
mechanical behavior is included in the interior nodal loads Fiint . A simulation based
on Eqs. (12.1) and (12.2) taking into account the shell kinematics and the specific
mechanical behavior describes the onset and development of eventual wrinkles
during forming. A simplified shell finite element made of woven unit cells is briefly
presented below. It is used to analyze the reason of wrinkling developments.
A triangular finite element made of woven cells is shown in Fig. 12.2a. Neighboring
elements are used to calculate the curvatures in the elements and consequently to
take bending into account [28]. The element is composed of ncell textile woven
unit cells. The internal virtual work associated to a virtual displacement field ˜ is
assumed to be the sum of the virtual works of tension, in-plane shear, and bending:
where
X
ncell
t
Wint .˜/ D p
©11 .˜/p T1 p L1 Cp ©22 .˜/p T2 p L2 .Virtual work of tension/
pD1
(12.4)
12 Analysis of Defect Developments in Composite Forming 323
Fig. 12.2 (a) Triangular shell finite element 1, 2, and 3 made of woven cells and its neighbors. (b)
Load resultants on a woven cell
p
X
ncell
s
Wint .˜/ D .˜/ p M s .Virtual work of in-plane shear/ (12.5)
pD1
X
ncell
p p
b
Wint .˜/ D p
11 .˜/p M 11 L1 Cp 22 .˜/p M 22 L2 .Virtual work of bending/
pD1
(12.6)
p
"11 (˜) , p "22 (˜), p (˜), p 11 (˜), p 22 (˜) are the virtual warp and weft axial
strains, in-plane shear angle and warp, and weft curvatures in the cell p, respectively.
L1 and L2 are the length of the woven cell in warp and weft directions. T11 and T22
are the tensions on the unit woven cell in warp and weft directions. M11 and M22 are
the bending moments on the woven cell, respectively, in warp and weft directions.
Ms is the in-plane shear moment (Fig. 12.2b).
The shell finite element made of textile woven cells (Fig. 12.2a) is used to
analyze wrinkle onset and development. As shown in Fig. 12.2b, each woven cell
is submitted to tension, in-plane shear, and bending. The interior nodal loads of
tension, in-plane shear, and bending can be obtained from the load resultants T11 ,
T22 , Ms , M11 , and M22 and the corresponding interpolation matrices of the element
[29–31]. For bending, the curvatures are computed using the neighboring elements.
324 P. Boisse et al.
Fig. 12.3 (a) Draping simulation with tensile stiffness only, (b) with tensile and in-plane shear
stiffnesses and (c) with tensile, in-plane shear, and bending stiffnesses
Fig. 12.5 Forming of a tetrahedral shape. (a) Tetrahedral punch with six blank holders. (b) Shaped
preform. (c) Forming simulation. (d) 60ı shear angle in a corner (e) Forming with a low blank older
load. (f) Forming with a low blank older load simulation
12 Analysis of Defect Developments in Composite Forming 327
the minimization of potential energy (i.e., the solution of dynamic Eq. 12.1) leads to
out-of-plane solutions. To determine if wrinkles will develop, all the strain energies
(tension, in-plane shear, and bending) must be taken into account.
Figure 12.5e, f shows the result of the forming experiment and simulation when
the blank holder pressure is decreased from 1 to 0.2 bar. In this case wrinkles
propagate to the tetrahedral part which is the useful part of the preform [44].
Large slippage between yarns can appear during a reinforcement forming process.
Figure 12.6b shows such slippage in the vertical face of a prismatic preform made
of glass plain weave fabric [45]. Figure 12.6a presents a hemispherical forming test
for which the blank holder was tightened and created such slippage between the
warp and weft yarns. The objective of this forming experiment was the validation of
forming simulation in the case of such slippages.
The slippage highlighted above leads to a loss of continuity of the textile reinforce-
ment. The simulation of their development can hardly be considered at macroscopic
scale. Many composite analyses are performed at the mesoscopic scale, i.e., the
scale of the fibrous yarns. For periodic structures, homogenized properties of the
composite can be computed by mesoscopic analyses [46, 47] as damage initiation
and crack propagation [48–51]. The simulation of the resin flow within deformed
woven cells allows to compute the permeability of the reinforcement for different
Fig. 12.6 Slippage between yarns of textile reinforcements. (a) Carbon twill fabric hemispherical
forming. (b) Glass plain weave prismatic preform [45]
328 P. Boisse et al.
Fig. 12.7 (a) Simplified FE model of the woven unit cell, (b) meso-analysis of the bias extension
test, (c) comparison of the shear angles, (d) meso-FE simulation of a hemispherical forming with
slippages, and (e) meso-FE simulation of a prismatic forming
12 Analysis of Defect Developments in Composite Forming 329
allows rotation at the crossover of yarns. When the shear angle becomes large, the
lateral contact between yams increases the in-plane shear stiffness.
The mechanical behavior of each yarn is that of a fibrous tow. It is described by a
hypoelastic model with an objective derivative based on the fiber rotation [54, 59].
An important feature is the possible slippage between the yarns which is allowed
by the proposed mesoscopic model. The theoretical analysis of the bias extension
test is based on the assumption that there is no slip at the yarn crossover. This
assumption is correct for shear angles smaller than 35ı or 40ı . For larger angles,
slippage occurs, and the shear angle (measured by optical method) is smaller than
the theoretical one (Fig. 12.7c). It is shown in Fig. 12.7b, c that the meso-FE analysis
of the bias extension test gives results consistent with the experiments even for large
angles when the test theory is no longer verified. This agreement is possible because
of the possible slippage between the yarns in the meso-FE model. Figure 12.7d, e
shows the results of meso-FE analyses of the hemispherical shaping and prismatic
shaping simulations [60, 61]. The corresponding experimental forming is shown in
Fig. 12.6a, b, respectively. In both cases the forming process leads to strong defects
due to large slippage between the yarns. These slippages are correctly described
by the meso-FE simulations. They are in correct agreement for the hemispherical
forming (Figs. 12.6a and 12.7d), but the slippages are more important in the
simulation of the prismatic forming (Fig. 12.6b and 12.7e). Actually, in the case
of the hemispherical forming, the material parameter of the carbon twill has been
measured in the laboratory, in particular the friction between the yarns. This is not
the case for the prismatic forming. Furthermore, the boundary conditions and more
precisely the loads at the edge of the preform are very important for slippage. They
are difficult to be assessed in many cases.
Nevertheless, the presented meso-FE simulations of a bias extension test and of
preform stamping processes have shown that this approach is a possible way for
the simulation of slippages that can occur during textile composite reinforcement
forming.
Recent works have highlighted another type of defect that can develop when the
geometry of the preform leads to adjacent areas with constant but different shear
angle. “Transition zones” are observed between these areas. For instance, curvatures
of the yarns are observed on a face of the tetrahedral shape analyzed in Sect. 12.2
(Fig. 12.8d) and on the face of a prismatic preform (Fig. 12.8c) [45, 62]. These yarn
curvatures can lead to buckles (Fig. 12.8d). Similar transition zones have also been
highlighted in bias extension tests [63]. The standard analysis of the test is based
on two assumptions: inextensibility of the fibers and rotations at the yarn crossovers
330 P. Boisse et al.
Fig. 12.8 (a, b) Transition zones in a bias extension test, (c) transition zones on the face of
prismatic shape, and (d) buckles on one edge of a tetrahedron shape
I4 D M1 C M1 I6 D M2 C M2 I8 D M1 C M2 (12.7)
M1 and M2 are orthonormal vectors in the warp and weft directions in the initial
configuration. The invariants I4 and I6 measure the changes of length in warp and
weft. I8 is related to the angle variation between warp and weft yarns. A possible
strain energy density is of the form
K4 K6 K8 A8 ˛
WD .I4 1/2 C .I6 1/2 C .I8 /2 C .I8 /n C rI8 rI8 (12.8)
2 2 2 2 2
This strain energy density includes a second-gradient term that takes into account
of the gradients of shear angle. ˛ is the associated material parameter. K4 and K6 are
the warp and weft tensile rigidities. The tensile behaviors are assumed to be linear
for simplicity purpose. K8 and A8 account for linear and nonlinear in-plane shear
strains, respectively.
The bias extension test is simulated both with a standard approach (first gradient)
and with a simulation based on the strain energy density presented in Eq. (12.8)
[63, 64]. The finite element mesh has been chosen in order to avoid locking that
can occur in this case [65, 66]. The simulation based on a first-gradient approach
gives the solution close to the “fishnet” theory of the test. There are three zones with
strictly constant in-plane shear in each zone (Fig. 12.9a). This is globally correct,
but the transition zones shown in Fig. 12.8b are not described. The simulation based
on the strain energy density with a second-gradient term in in-plane shear (Eq. 12.8)
gives in-plane shear patterns that are close to the real solution in the specimen.
But in addition, it correctly describes the transition zone corresponding to the fiber
bending between two areas with constant in-plane shear (Fig. 12.9b). This approach
has been used to give solutions to other cases in which the fiber bending stiffness is
important and where the standard simulations fail to give correct results [67]. The
use of second-gradient models for the simulation of forming cases where transition
zones develop such as in Fig. 12.8c or Fig. 8d is in progress.
332 P. Boisse et al.
Fig. 12.9 Simulation of the bias extension test using (a) a standard approach and (b) simulation
based on a second-gradient approach
12.5 Conclusions
Fig. 12.10 S shape in a bias extension test on an unbalanced woven reinforcement(c), (a)
unbalanced fabric, (b) deformed experimental shape, and (c) computed deformed shape (second
gradient)
The large shear angles that are necessary for double-curved preforms can lead to
wrinkling. But there is no direct relation between shear angle and wrinkling. In
particular, large tensions due to blank holder can avoid wrinkling although shear
angles are very large. Consequently the “locking angle” is a questionable concept
for the determination of the wrinkling onset and must be used cautiously. The size
of wrinkles depends on bending stiffness. Membrane approaches are frequent in
draping simulations, but they cannot describe wrinkle shapes.
Mesoscopic approaches for the simulation of woven preform forming allow to
describe the slippage that can occur between the yarns of a reinforcement during
forming. Mesoscopic modeling is mandatory in this case where the continuity is no
more valid at macroscopic scale. A meso-FE simulation of the complete preform
using a coarse mesh of shell elements for each yarn showed that it was possible
to simulate large slippage between the yarns during forming. Macroscopic forming
(i.e., forming of the whole preform) based on mesoscopic models of the reinforce-
ment is probably a promising way in composite preform draping simulations. More
efficient numerical methods will be necessary for the expansion of this approach
and to be able to develop meso-FE models as efficient as macroscopic simulations.
“Transition zones” have been highlighted in several preform forming processes
and in the bias extension test. First- and second-gradient approaches were compared
in the simulation of the bias extension test. Only the second gradient is able to
describe the “transition zones” and to simulate the S shape obtained in a bias
extension test on an unbalanced woven composite reinforcement.
Acknowledgments This research was supported by the National Scientific Research Centre
(CNRS) in the scope of the project PEPS INSIS and by the European Commission in the scope of
the project ITOOL.
334 P. Boisse et al.
References
1. E. Phil, C. Soutis (eds.), Polymer Composites in the Aerospace Industry (Elsevier, Amsterdam,
2014)
2. E.R. Fuchs, F.R. Field, R. Roth, R.E. Kirchain, Strategic materials selection in the automobile
body: Economic opportunities for polymer composite design. Compos. Sci. Technol. 68(9),
1989–2002 (2008)
3. K. Friedrich, A.A. Almajid, Manufacturing aspects of advanced polymer composites for
automotive applications. Appl. Compos. Mater. 20(2), 107–128 (2013)
4. X.Q. Peng, J. Cao, A continuum mechanics-based non-orthogonal constitutive model for
woven composite fabrics. Compos. A: Appl. Sci. Manuf. 36(6), 859–874 (2005)
5. W.R. Yu, P. Harrison, A. Long, Finite element forming simulation for non-crimp fabrics using a
non-orthogonal constitutive equation. Compos. A: Appl. Sci. Manuf. 36(8), 1079–1093 (2005)
6. R.H.W. Ten Thije, R. Akkerman, J. Huétink, Large deformation simulation of anisotropic
material using an updated Lagrangian finite element method. Comput. Methods Appl. Mech.
Eng. 196(33), 3141–3150 (2007)
7. Y. Aimene, E. Vidal-Sallé, B. Hagège, F. Sidoroff, P. Boisse, A hyperelastic approach for
composite reinforcement large deformation analysis. J. Compos. Mater. 44, 5–26 (2010)
8. A. Charmetant, J.G. Orliac, E. Vidal-Sallé, P. Boisse, Hyperelastic model for large deformation
analyses of 3D interlock composite preforms. Compos. Sci. Technol. 72(12), 1352–1360
(2012)
9. Q.T. Nguyen, E. Vidal-Sallé, P. Boisse, C.H. Park, A. Saouab, J. Bréard, G. Hivet, Mesoscopic
scale analyses of textile composite reinforcement compaction. Compos. Part B 44(1), 231–241
(2013)
10. K.A. Fetfatsidis, D. Jauffrès, J.A. Sherwood, J. Chen, Characterization of the tool/fabric and
fabric/fabric friction for woven-fabric composites during the thermostamping process. Int. J.
Mater. Form. 6(2), 209–221 (2013)
11. S. Allaoui, G. Hivet, A. Wendling, P. Ouagne, D. Soulat, Influence of the dry woven fabrics
meso-structure on fabric/fabric contact behaviour. J. Compos. Mater. 46(6), 627–639 (2012)
12. B. Cornelissen, U. Sachs, B. Rietman, R. Akkerman, Dry friction characterisation of carbon
fibre tow and satin weave fabric for composite applications. Compos. A: Appl. Sci. Manuf. 56,
127–135 (2014)
13. H.J.L. Dirk, C. Ward, K.D. Potter, The engineering aspects of automated prepreg layup:
History, present and future. Compos. Part B 43(3), 997–1009 (2012)
14. D. Leutz, M. Vermilyea, S. Bel, R. Hinterhölzl, Forming simulation of thick AFP laminates
and comparison with live CT imaging. Appl. Compos. Mater. 23(4), 583–600 (2016)
15. P. Hallander, M. Akermo, C. Mattei, M. Petersson, T. Nyman, An experimental study of
mechanisms behind wrinkle development during forming of composite laminates. Compos.
A: Appl. Sci. Manuf. 50, 54–64 (2013)
16. P. Hallander, J. Sjölander, M. Åkermo, Forming induced wrinkling of composite laminates
with mixed ply material properties; an experimental study. Compos. A: Appl. Sci. Manuf. 78,
234–245 (2015)
17. J. Sjölander, P. Hallander, M. Åkermo, Forming induced wrinkling of composite laminates: A
numerical study on wrinkling mechanisms. Compos. A: Appl. Sci. Manuf. 81, 41–51 (2016)
18. J.S. Lightfoot, M.R. Wisnom, K. Potter, A new mechanism for the formation of ply wrinkles
due to shear between plies. Compos. A: Appl. Sci. Manuf. 49, 139–147 (2013)
19. S.P. Haanappel, R.H.W. Ten Thije, U. Sachs, B. Rietman, R. Akkerman, Formability analyses
of uni-directional and textile reinforced thermoplastics. Compos. A: Appl. Sci. Manuf. 56,
80–92 (2014)
20. T.G. Gutowski, G. Dillon, S. Chey, H. Li, Laminate wrinkling scaling laws for ideal
composites. Compos. Manuf. 6(3), 123–134 (1995)
21. R.K. Pandey, C.T. Sun, Mechanisms of wrinkle formation during the processing of composite
laminates. Compos. Sci. Technol. 59(3), 405–417 (1999)
12 Analysis of Defect Developments in Composite Forming 335
22. K. Potter, B. Khan, M. Wisnom, T. Bell, J. Stevens, Variability, fibre waviness and
misalignment in the determination of the properties of composite materials and structures.
Compos. A: Appl. Sci. Manuf. 39(9), 1343–1354 (2008)
23. T. Belytschko, An overview of semi-discretisation and time integration procedures, in Compu-
tation Methods for Transient Analysis, ed. by T. Belytschko, T.J.R. Hughes (Elsevier Science,
Amsterdam, 1983), pp. 1–65
24. T.J.R. Hughes, T. Belytschko, A precise of developments in computational methods for
transient analysis. J. Appl. Mech. 50, 1033–1041 (1983)
25. M.A. Crisfield, Non-Linear Finite Element Analysis of Solids and Structures, vol 2 (Wiley,
New York, 1997)
26. O.C. Zienkiewicz, R.L. Taylor, The finite element method for solid and structural mechanics
(Butterworth-Heinemann, Oxford, 2005)
27. T. Belytschko, W.K. Liu, B. Moran, Nonlinear Finite Elements for Continua and Structures
(Wiley, Chichester, 2000)
28. E. Onate, F. Zarate, Rotation-free triangular plate and shell elements. Int. J. Numer. Methods
Eng. 47, 557–603 (2000)
29. N. Hamila, P. Boisse, A meso–macro three node finite element for draping of textile composite
preforms. Appl. Compos. Mater. 14(4), 235–250 (2007)
30. P. Boisse, Finite element analysis of composite forming, in Composite Forming Technologies,
ed. by A.C. Long (Woodhead Publishing, Cambridge, 2007), pp. 46–79
31. N. Hamila, P. Boisse, F. Sabourin, M. Brunet, A semi-discrete shell finite element for textile
composite reinforcement forming simulation. Int. J. Numer. Methods Eng. 79(12), 1443–1466
(2009)
32. R. Hill, A general theory of uniqueness and stability in elastic–plastic solids. J. Mech. Phys.
Solids 8, 236–249 (1958)
33. N. Friedl, F.G. Rammerstorfer, F.D. Fischer, Buckling of stretched strips. Comput. Struct.
78(1–3), 185–190 (2000)
34. A.G. Prodromou, J. Chen, On the relationship between shear angle and wrinkling of textile
composite preforms. Compos. Part A 28A, 491–503 (1997)
35. G. Lebrun, M.N. Bureau, J. Denault, Evaluation of bias-extension and picture frame test for the
measurement of shear properties of PP/glass commingled fabrics. Compos. Struct. 61, 52–341
(2003)
36. S.B. Sharma, M.P.F. Sutcliffe, S.H. Chang, Characterisation of material properties for draping
of dry woven composite material. Compos. Part A 34, 1167–1175 (2003)
37. B. Zhu, T.X. Yu, J. Teng, X.M. Tao, Theoretical modeling of large shear deformation and
wrinkling of plain woven composite. J. Compos. Mater. 43, 125–138 (2009)
38. J.S. Lightfoot, M.R. Wisnom, K. Potter, Defects in woven preforms: Formation mechanisms
and the effects of laminate design and layup protocol. Compos. A: Appl. Sci. Manuf. 51, 99–
107 (2013)
39. A.A. Skordos, C. Monroy Aceves, M.P.F. Sutcliffe, A simplified rate dependent model of
forming and wrinkling of pre-impregnated woven composites. Compos. Part A 38, 1318–1330
(2007)
40. J. Lee, S. Hong, W. Yu, T. Kang, The effect of blank holder force on the stamp forming
behaviour of non-crimp fabric with a chain stitch. Compos. Sci. Technol. 67(3–4), 357–366
(2007)
41. P. Boisse, N. Hamila, E. Vidal-Sallé, F. Dumont, Simulation of wrinkling during textile
composite reinforcement forming. Influence of tensile, in-plane shear and bending stiffnesses.
Compos. Sci. Technol. 71(5), 683–692 (2011)
42. L.M. Dangora, C.J. Mitchell, J.A. Sherwood, Predictive model for the detection of out-of-
plane defects formed during textile-composite manufacture. Compos. A: Appl. Sci. Manuf. 78,
102–112 (2015)
43. ITOOL ‘Integrated Tool for Simulation of Textile Composites’. European specific targeted,
research project, sixth framework programme, aeronautics and space. http://www.itool.eu
44. S. Allaoui, P. Boisse, S. Chatel, N. Hamila, G. Hivet, D. Soulat, E. Vidal-Salle, Experimental
and numerical analyses of textile reinforcement forming of a tetrahedral shape. Compos. A:
Appl. Sci. Manuf. 42(6), 612–622 (2011)
336 P. Boisse et al.
65. R.H.W. Ten Thije, R. Akkerman, Solutions to intra-ply shear locking in finite element analyses
of fibre reinforced materials. Compos. A: Appl. Sci. Manuf. 39(7), 1167–1176 (2008)
66. N. Hamila, P. Boisse, Locking in simulation of composite reinforcement deformations.
Analysis and treatment. Compos. A: Appl. Sci. Manuf. 53, 109–117 (2013)
67. A. Madeo, M. Ferretti, F. Dell’Isola, P. Boisse, Thick fibrous composite reinforcements behave
as special second-gradient materials: Three-point bending of 3D interlocks. Z. Angew. Math.
Phys. 66(4), 2041–2060 (2015)
68. A. Madeo, G. Barbagallo, M.V. D’Agostino, P. Boisse, Continuum and discrete models for
unbalanced woven fabrics. Int. J. Solids Struct. (2016). doi:10.1016/j.ijsolstr.2016.02.005
69. G. Barbagallo, A. Madeo, I. Azehaf, I. Giorgio, F. Morestin, P. Boisse, Bias extension test on an
unbalanced woven composite reinforcement: experiments and modeling via a second gradient
continuum approach. J. Compos. Mater. (2016). doi:10.1177/0021998316643577
Part IV
Chapter 13
Deformation Mechanisms of Carbon Fibres
and Carbon Fibre Composites
Robert J. Young
13.1 Introduction
Raman spectroscopy is now widely used to characterise the structure and defor-
mation of a number of many materials, including different forms of carbon. Being
a black material, it was thought originally that it would not be possible to obtain
Raman scattering from graphitic carbons until the groundbreaking work of Tuinstra
and Koenig [1]. Their original studies were then extended to carbon fibres [2–6].
It was then found that the Raman technique could be used to characterise both
carbon nanotubes [7–10] and graphene [11–15] as interest upon these materials
has developed rapidly over more recent years. Raman spectroscopy is able to
provide unique information upon the understanding of the relationship between
microstructure and deformation in carbon fibres and allows the direct measurement
of fibre stress or strain in a number of systems. It relies on the phenomenon that
when high-performance fibres are deformed, the bands in their spectra shift in a
well-defined way. It is possible to calibrate the behaviour of a particular band against
the stress or strain in the fibre [7]. This then enables the direct determination of stress
or strain at a point in a fibre under any general state of deformation.
For the analysis of composite micromechanics, it is necessary to have measure-
ment techniques that are capable of quantifying the level of interfacial adhesion
between the fibres and matrix. A number of different micromechanical test methods
such as pull-out, push-out, microbond and fragmentation tests have been developed
in order to determine the strength of the interface in high-performance composites
[16]. Unfortunately, however, the tests often give different results even with the same
fibre/matrix systems prepared under identical conditions [17]. One reason for this
is that the state of stress at the interface may be different in the different tests, but
a more fundamental problem is that the determination of fibre stress or strain using
these test methods is invariably indirect. A number of assumptions usually have to
be made about material properties such as the distribution of fibre strength [16].
Also data reduction techniques have to rely upon the use of analytical equations
that may or may not give an accurate picture of the state of fibre and matrix stress
or strain in the system [16]. In this presentation, it will be demonstrated that the
use of Raman spectroscopy enables hence an unambiguous interpretation of the
micromechanical test results and opens up the possibility of obtaining a unique
insight into the mechanics of reinforcement by carbon fibres.
Raman spectroscopy is now widely used to characterise both the structure and
deformation of carbon fibres. The experimental procedure is shown schematically
in Fig. 13.1 where a sample such as a carbon fibre is irradiated with a laser beam
of frequency, 0 . Most of the light is scattered at the same frequency (or energy)
with a small amount (the Raman scattered light) scattered at different frequencies
of 0 ˙ . It has been found over recent years that many different types of high-
performance fibres show well-defined Raman spectra [18].
Original Raman studies upon carbon fibres reported that there are normally
two major first-order bands found, D and G [1, 4]. Robinson et al. [5] were the
first to show that these two Raman bands shift to lower wave number upon the
application of stress or strain and this was later confirmed independently by Sakata
et al. [6]. Light is very strongly absorbed by sp2 -hybridised carbon materials. Since
Raman spectroscopy can therefore only detect signals from a few tens of nm in
depth in carbon materials and the Raman spectra are characteristic of the state of
graphitisation of the material, it can also be a very useful technique for studying
the variation of microstructure across carbon fibres [4]. For example, by embedding
the fibres in an epoxy resin and then polishing to different depths into the fibre,
differences between the skin and core microstructures can be observed using a
Raman microprobe system [4]. These differences can then be correlated with the
results obtained from transmission electron microscopy or other techniques.
a PAN-Based Fibres
T300
T40
T800
Intensity
T650
HMS4
T50
b Pitch-Based Fibres
P25
P55
Intensity
P75
P100
P120
Fig. 13.2 Raman spectra for series of (a) PAN-based carbon fibres and (b) pitch-based carbon
fibres
Raman spectra in the region 1000 cm1 –3000 cm1 are shown in Fig. 13.2a for
a series of polyacrylonitrile (PAN)-based carbon fibres and in Fig. 13.2b for a range
of mesophase pitch-based carbon fibres. In each case, the fibres at the top have the
lowest values of Young’s modulus and those at the bottom the highest values. Hence,
there is a systematic decrease in the Young’s modulus of the fibres from the bottom
to the top spectra for the two sets of fibres.
It can be seen that there are, in fact, three main Raman hands for carbon fibres,
with an additional band (D’) sometimes found at around 1600 cm1 . The G band
at about 1580 cm1 has been attributed to the C-C in-plane stretching mode of the
graphene planes of an infinite crystal. The band at 1360 cm1 is thought to be due
344 R.J. Young
to the crystal boundaries in the graphite. The ratio of intensities of two bands ID /IG
has been shown to be related to the crystal size, La [1], and the bandwidth reflects
the orientation of the graphite layer plane with respect to the fibre axis [19]. The
Raman band at around 2700 cm1 , known as the 2D band, is a second-order hand
since it is found at about twice the wave number of the 1360 cm1 band [20]. In
the case of the PAN-based carbon fibres, it is only seen in the spectra of the two
highest modulus fibres HMS4 and T50. The band is present in all of the spectra for
the pitch-based fibres in Fig. 13.2b although it is most well defined in the case of the
P75 fibre. It can be seen from Fig. 13.2 that the Raman spectra can be essentially
used as fingerprints to identify carbon fibres. They also tend to become sharper and
more well defined as the fibre moduli increase. This is because of the presence of
larger, better-oriented crystals.
Raman spectra are presented in Fig. 13.3 for PAN- and pitch-based carbon
fibres with different levels of surface treatment. In all cases, the spectra exhibit the
same appearance, that is, four well-resolved bands, namely, D (1330 cm1 ), G
( 1580 cm1 ), D’ ( 1620 cm1 ) and 2D ( 2660 cm1 ), along with additional weaker
features. Figure 13.3 shows spectra for the two types of fibres both untreated and
treated with oxygen plasma to modify the level of fibre-matrix adhesion [21]. No
differences were observed in the Raman band positions and widths after plasma
treatment of the fibres. The intensity ratio of the two first-order bands D and
G, however, is always higher after plasma treatment. This shows that the plasma
treatment introduces defects and functionality into the fibre surface which will be
shown to result in better fibre-matrix adhesion.
Raman spectroscopy has also been employed to follow the development of
graphitisation in carbon fibres. An example of this is given in Fig. 13.4 for the
production of carbon fibres from a cellulose fibre precursor [22]. Typical Raman
spectra of progressively carbonised cellulose, at temperatures ranging from 400 ı C
to 1500 ı C, are presented. The first Raman spectrum indicative of the presence of
carbon is detected at a temperature of 400 ı C, although it is difficult to identify
due to the low intensity of the G (IG ) and D (ID ) bands [22]. As the carbonisation
temperature increases, IG and ID also increase, making the carbon structure more
easily identifiable. It can be seen from Fig. 13.4 that ID is still greater than IG after
pyrolysation at a temperature of 1500 ı C. A graphitic structure had not formed at
this stage as the 2D band, which should be seen at a position of 2700 cm1 for a
graphitic structure, is not present. The 2D band, however, becomes apparent only
after the heat treatment temperature reaches 2500 ı C [22].
The positions of all the Raman bands are found to shift when the carbon fibres are
subjected to tensile deformation, a phenomenon that can be employed to follow
the deformation micromechanics of carbon fibres. Figure 13.5 shows the 2D band
of a pitch-based P100 fibre in the undeformed state and two levels of tensile and
13 Deformation Mechanisms of Carbon Fibres and Carbon Fibre Composites 345
a D T50
G T50-O
2D
D'
Intensity
b P100
P100-O
Intensity
Fig. 13.3 Raman spectra of (a) a T50 PAN-based carbon fibres and (b) a P100 pitch-based carbon
fibres (untreated (top) and plasma treated (bottom)). (Adapted from [23] with permission from
Elsevier)
compressive strain (0.6 % and 0.2 %, respectively). The band shifts towards lower
wave number (red shift) when the fibre is subjected to tension, whereas the shift is
in the opposite direction (blue shift) when the fibre is in compression. A significant
broadening of the Raman band when the fibre is deformed is also found as can be
seen from Fig. 13.5. Similar stress-induced Raman band shifts are also found for
PAN-based carbon fibres such as T50.
346 R.J. Young
D band G band
Precursor cellulose
o
400 C
o
600 C
o
800 C
Intensity
o
1000 C
o
1200 C
o
1300 C
o
1400 C
o
1500 C
Fig. 13.4 Typical Raman spectra obtained from cellulose fibres during the carbonisation process
at temperatures in the range 400–1500 ı C and the cellulose precursor material (Adapted from [22]
with permission of Springer)
ef= 0.0%
ef= 0.6%
ef= -0.2%
Intensity
Fig. 13.5 The shift of the 2D band peaks on the application of tensile and compressive strain to a
P100 carbon fibre. (Adapted from [23] with permission from Elsevier)
13 Deformation Mechanisms of Carbon Fibres and Carbon Fibre Composites 347
2670
T50
P100
Raman wavenumber (cm )
-1
2660 P120
2650
2640
2630
0.0 0.2 0.4 0.6 0.8
Fibre strain, ef (%)
Fig. 13.6 Variation of the 2D Raman band maximum with strain in as-received T50, P100 and
P120 carbon fibres
It is found that there is an approximately linear shift of the band position with
tensile strain as shown in Fig. 13.6 for the 2D band of the T50 and P100 fibres, along
with P120. Since the fibres deform in an approximately linearly elastic manner, the
shift with stress is also linear. It will be shown later that Fig. 13.5 can be used as a
simple calibration to determine the distribution of stress along a fibre using a Raman
laser beam focused onto individual fibres inside the matrix resin. It is also important
to note that the slope of the lines in Fig. 13.6 is found to depend upon the Young’s
modulus of the carbon fibres [7]. This is shown in Fig. 13.7 for a number of pitch-
based carbon fibres. The Raman band shifts of PAN-based carbon fibres are also
found to fall upon the same line [18].
It can be seen that there is an approximately linear dependence of the band shift
rate upon the fibre Young’s modulus and the slope of the dashed line is of the order
of 50 to 60 cm1 /TPa. This is found to be a universal relationship for the 2D band
and is valid for all forms of graphitic sp2 carbon materials [7]. It is now used widely
to follow the deformation behaviour of other types of sp2 carbon materials such as
carbon nanotubes [8] and graphene [15], and it has revolutionised our understanding
of deformation micromechanics in these materials. The rates of Raman band shift
(per unit strain) for sp2 carbon materials are found to be proportional to the effective
values of their Young’s moduli in nanocomposites.
348 R.J. Young
Fig. 13.7 Shift rate of the 2D Raman band per unit strain as a function of the tensile Young’s
modulus for a number of different pitch-based carbon fibres
The fibre stress or strain in an individual fibre can be determined from the stress-
induced shift of the Raman bands obtained from the fibre using a laser beam focused
onto the fibre inside the matrix resin as shown in Fig. 13.8. The laser beam diameter
in a microscope-based Raman spectrometer is typically 1 µm. This is significantly
smaller than the usual carbon fibre diameter (5–10 µm) which means that fibre
stress and strain mapping can be undertaken with high precision along individual
fibres [23].
Figure 13.9 shows how strain along a carbon fibre within a matrix can be
determined experimentally from the positions of the Raman 2D band obtained from
the fibre. Well-defined Raman bands can be obtained from the fibres since the
Raman scattering from the fibres is usually much stronger than that of the resin
matrix. The band shift needs to be calibrated first of all by deforming single fibres
in air to produce data such as those shown in Fig. 13.6. The fibre strain can then be
easily determined since the bands tend to shift wave number approximately linear
with strain.
13 Deformation Mechanisms of Carbon Fibres and Carbon Fibre Composites 349
Fig. 13.8 A model composite specimen containing a single fibre embedded within a transparent
polymer resin. The matrix strain is determined using a resistance strain gauge and fibre strain from
the Raman spectra obtained along the fibre length
The nature of the surface of carbon fibres generally leads to poor stress transfer from
the matrix to the fibres. This lack of fibre/matrix adhesion can be overcome partially
through the development of surface treatments that are usually implemented fully
in the carbon fibre fabrication process. The conventional surface treatment used
by carbon fibre manufacturers is generally an electrochemical oxidation process.
Apart from the relative success in the improvement of the interfacial behaviour
of composites, the increasing concern about environmental pollution has limited
wide industrial application of such chemical surface treatments. The development of
alternative environmentally friendly methods is an attractive proposition from both
scientific and technological viewpoints. The surface modification of fibres by cold
plasma is becoming popular among these new methods, and it is now an important
branch of plasma technology.
High-modulus (HM) carbon fibres are one of the most impressive reinforcements
used in composites in terms of their specific tensile properties. Such properties
are related principally to the high degree of orientation of the crystallites in the
fibres, and this highly graphitic characteristic also leads to high levels of thermal
and electric conductivity. All these properties make HM carbon fibre candidates for
wider use in the aerospace industry. The enhanced crystallinity of this type of fibre,
on the other hand, is often reflected in a lower efficiency of techniques for increasing
the carbon surface activity, in comparison with the high-strength (HT) carbon fibres.
This is because HM carbon fibres are generally more resistant to electrochemical
oxidation than HT ones.
The behaviour of untreated and unsized T50 PAN-based HM carbon fibres has
been studied by Montes-Morán and Young [23]. Plasma-treated samples (-O series)
were produced by surface treatment of the as-received fibres. Microwave (2.45 GHz)
plasma treatments were carried out using a cylindrical chamber in which the
fibres were placed for 3 min of residence at 75 W. Oxygen (99.999 %) was used
350 R.J. Young
2660
2640
MATRIX
2620
2600
CARBON FIBRE
2580
0 100 200 300 400 500
1.5
1.0
Fibre strain, ef (%)
0.5
0.0
-0.5
Fig. 13.9 The method of determination of strain from the Raman spectra obtained along a carbon
fibre. (Adapted from [23] with permission from Elsevier)
as the activation gas at a chamber pressure of 1.0 ˙ 0.1 mbar throughout the
treatment time. Such an arrangement gave rise to a very homogeneous surface
treatment confirmed by several fibre surface characterisation techniques. Scanning
electron micrographs of the two types of fibres, untreated and treated, are shown in
Fig. 13.10.
The variation of fibre strain ef D f =Ef along a T50-O plasma-treated carbon
fibre in an epoxy resin subjected to different levels of matrix strain, em , is shown
in Fig. 13.11a. The detailed analysis of this situation is elucidated in a number of
composite textbooks [24–26]. The deformation of composites containing fibres of
finite length deformed axially has been considered by a number of authors dating
back as early as Krenchel [27].
13 Deformation Mechanisms of Carbon Fibres and Carbon Fibre Composites 351
The problem of transfer of stress from a matrix to a fibre and the subsequent
variation of stress along a fibre of finite length in a matrix was first tackled properly
by Kelly [28] in his classical text, ‘Strong Solids’. This groundbreaking work
involved the revival of the shear-lag concept of Cox [29]. The equation for the
distribution of fibre stress, f , as a function of distance, x along the fibre is given
by [30]
cosh .nx=r/
f D Ef em 1 (13.1)
cosh .nl=2r/
where Ef is the Young’s modulus of the fibre, em is the matrix strain, l is the fibre
length, r is the fibre radius and n can be thought of as a fitting parameter related to
the efficiency of stress transfer [30].
The data in Fig. 13.11a have been fitted to Eq. (13.1) by choosing appropriate
values of n, and it can be seen that there is a close correlation between the theoretical
curves and experimental data points. Additionally, it can be seen that the strain in
the fibre is that same as the matrix strain along the centre of the fibres as is assumed
in the theoretical analyses [28, 30].
352 R.J. Young
em=0.0%
em=0.35%
1.0 T50-O
em=0.7%
a full-bonding model
0.8
0.6
Fibre strain, ef (%)
0.4
0.2
0.0
-0.2
0 1000 2000 3000
Distance along fibre (μm)
50
b em=0.35%
40
em=0.7%
Interfacial shear stress, τ i (MPa)
30
20
10
0
-10
-20
-30
-40
-50
0 1000 2000 3000
Distance along fibre (μm)
Fig. 13.11 (a) Experimental fibre strain distributions determined for a plasma-treated T50-O fibre
from strain-induced Raman band shifts at different levels of matrix strain in an epoxy resin up to
0.7 % strain. (b) Derived distribution of interfacial shear stress along the same fibre. (Adapted from
[23] with permission from Elsevier)
n sinh .nx=r/
i D E f em (13.2)
2 cosh .nl=2r/
13 Deformation Mechanisms of Carbon Fibres and Carbon Fibre Composites 353
0.6
0.4
0.2
0.0
-0.2
0 1000 2000 3000
Distance along fibre (μm)
Fig. 13.12 The fibre strain distributions determined from strain-induced Raman band shifts at
different levels of matrix strain (up to 1.1 %) for a plasma-treated T50-O fibre in an epoxy resin,
showing the effect of fibre fragmentation. (Adapted from [23] with permission from Elsevier)
It can be seen that the interfacial shear stress is highest at the fibre ends (where there
is a gradient of axial fibre strain or stress). For a matrix strain of 0.7 %, the value
increases to around 45 MPa which approaches the shear yield stress of the resin.
The effect of increasing the matrix strain further up to 0.9 % is shown in
Fig. 13.12. At this matrix strain level, the T50-O fibre undergoes fragmentation
(since the strain falls to zero at the fibre breaks) which saturates at a matrix strain
of 1.1 %. In this case, the interfacial adhesion has been lost, and stress transfer at
the interface is essentially frictional such that the variation of strain along the length
of each fragment is approximately triangular in shape [31]. Moreover, it can be
seen that the peak fragment strains are all well below the matrix strain, indicating
that once the fibres have undergone fragmentation, the level of reinforcement is
diminished.
The stress transfer behaviour of a T50 fibre that had not been modified by plasma
treatment is shown in Fig. 13.13. The distribution of strain in the fibre at 0.35 %
matrix strain follows the shear-lag model given by Eq. (13.1). When the matrix
strain is increased to 0.7 %, however, the distribution of strain is somewhat different
with an approximately linear rise in fibre strain at the two ends. This is an indication
of the fibre undergoing debonding that starts at the fibre ends and progresses along
the fibre as the level of matrix strain is increased [23]. This should be compared with
the behaviour for the T50-O fibre at 0.7 % strain shown in Fig. 13.10a.
Figure 13.13b shows that the fibre-matrix interface has failed completely at high
strain as the fibre had broken into two fragments producing approximately triangular
strain distributions. The length of the two fragments is around 1000 µm (1 mm).
This is significantly longer than that of the fragments of the T50-O fibre shown in
354 R.J. Young
Fig. 13.13 Fibre strain distributions determined from strain-induced Raman band shifts at dif-
ferent levels of matrix strain for an untreated T50 fibre in an epoxy resin, showing the effect
of debonding at the fibre-matrix interface at the higher strain level. (a) Low-strain region and
(b) high-strain region. (Adapted from [23] with permission from Elsevier)
Fig. 13.12 which were found to be only around 300 µm long. This is a clear simple
demonstration of the effect of fibre surface treatment upon the micromechanics of
deformation in carbon fibre composites.
13 Deformation Mechanisms of Carbon Fibres and Carbon Fibre Composites 355
60
T50-O
Interfacial shear stress, τ (MPa)
40
20
Fig. 13.14 Variation of maximum interfacial shear stress i,max with applied matrix strain for
T50-O fibre/epoxy composite. (Adapted from [23] with permission from Elsevier)
The maximum value of the interfacial shear stress, i,max , can be determined directly
from plots of the variation of i (x) with distance along the fibre (e.g. Fig. 13.11b).
Figure 13.14 compares, for a T50-O single-fibre/epoxy composite, the evolution of
i,max with applied matrix strain. It can be seen that max increases initially with
increasing matrix strain and then reaches a maximum value of 45 MPa at 0.9 %
matrix strain. There are basically two possible routes that can lead to failure of
the interface in fibre-reinforced composites. Firstly, interfacial failure may occur
when i,max reaches the interfacial shear strength (IFSS), i.e. a parameter used
conventionally to quantify the level of adhesion between a fibre and a matrix.
Secondly, the interfacial failure can also occur when i,max reaches the shear yield
stress of the matrix, y . The shear yield stress of the epoxy resin matrix used in
this study was thought to be around 45 MPa, as indicated in Fig. 13.13. It appears
therefore that the strength of the fibre-matrix interface in the case of the T50-O
plasma-treated fibre was limited by the shear yield stress of the matrix. This should
be contrasted with similar measurements upon the equivalent untreated T50 fibre in
the same matrix, where i,max was found to be only around 20 MPa and with value
controlled in this case by cohesive failure of the fibre-matrix interface [23].
In conclusion, it is clear that the fibre surface treatment can increase the value of
i,max significantly compared with that of the untreated fibre. Nevertheless, even with
an improved level of interfacial adhesion, it is found that the strength of the fibre-
matrix interface is eventually limited by shear yielding of the epoxy resin matrix.
356 R.J. Young
13.4 Conclusions
It has been shown that Raman spectroscopy can be employed to characterise the
structure of carbon fibres, to evaluate differences between different fibres and
to follow changes in structure of the fibres during carbonisation. Moreover, the
technique can also be used to follow deformation of different types of fibres
from stress-induced Raman band shifts. The deformation of carbon fibres within
a polymer resin can also be followed with a high level of precision. It is found that
a relatively simple shear-lag analysis can be used to predict and model the local
distributions of fibre stress and strain in a composite, and it has been demonstrated
that there generally is a good correlation between these theoretical and experimental
approaches.
References
1. F. Tuinstra, J.L. Koenig, Raman spectrum of graphite. J. Chem. Phys. 53, 1126–1130 (1970)
2. Y. Huang, R.J. Young, Nonlinear elasticity in carbon-fibers. J. Mater. Sci. Lett. 12, 92–95
(1993)
3. Y. Huang, R.J. Young, Effect of fibre microstructure upon the mechanical properties of
PAN—and pitch-based carbon fibres. Carbon 33, 97–107 (1995)
4. K. Morita, Y. Murata, A. Ishitani, K. Murayama, T. Ono, A. Nakajima, Characterization of
commercially available PAN (polyacrylonitrile)-based carbon-fibers. Pure Appl. Chem. 58,
455–468 (1986)
5. I.M. Robinson, M. Zakikhani, R.J. Day, R.J. Young, C. Galiotis, Strain dependence of the
Raman frequencies for different types of carbon-fibers. J. Mater. Sci. Lett. 6, 1212–1214
(1987)
6. H. Sakata, G. Dresselhaus, M.S. Dresselhaus, M. Endo, Effect of uniaxial-stress on the Raman-
spectra of graphite fibers. J. Appl. Phys. 63, 2769–2772 (1988)
7. C.A. Cooper, R.J. Young, M. Halsall, Investigation into the deformation of carbon nanotubes
and their composites through the use of Raman spectroscopy. Compos. Pt. A 32, 401–411
(2001)
8. L.B. Deng, S.J. Eichhorn, C.C. Kao, R.J. Young, The effective Young’s modulus of carbon
nanotubes in composites. ACS Appl. Mater. Interfaces 3, 433–440 (2011)
9. M. Lucas, R.J. Young, Raman spectroscopic study of the strain effects on the breathing modes
of carbon nanotubes in epoxy/SWNT composites. Compos. Sci. Tech. 64, 2297–2302 (2004)
10. M. Lucas, R.J. Young, Unique identification of single-walled carbon nanotubes in composites.
Compos. Sci. Tech. 67, 2135–2149 (2007)
11. L. Gong, I.A. Kinloch, R.J. Young, I. Riaz, R. Jalil, K.S. Novoselov, Interfacial stress transfer
in a graphene monolayer nanocomposite. Adv. Mater. 22, 2694–2697 (2010)
12. L. Gong, R.J. Young, I.A. Kinloch, I. Riaz, R. Jalil, K.S. Novoselov, Optimizing the reinforce-
ment of polymer-based nanocomposites by graphene. ACS Nano 6, 2086–2095 (2012)
13. L. Gong, R.J. Young, I.A. Kinloch, S.J. Haigh, J.H. Warner, J.A. Hinks, Z.W. Xu, L. Li, F. Ding,
I. Riaz, R. Jalil, K.S. Novoselov, Reversible loss of Bernal stacking during the deformation of
few-layer graphene in nanocomposites. ACS Nano 7, 7287–7294 (2013)
14. R.J. Young, L. Gong, I.A. Kinloch, I. Riaz, R. Jalil, K.S. Novoselov, Strain mapping in a
graphene monolayer nanocomposite. ACS Nano 5, 3079–3084 (2011)
15. R.J. Young, I.A. Kinloch, L. Gong, K.S. Novoselov, The mechanics of graphene nanocompos-
ites: a review. Compos. Sci. Tech. 72, 1459–1476 (2012)
13 Deformation Mechanisms of Carbon Fibres and Carbon Fibre Composites 357
16. F.J. Herrera-Franco, L.T. Drzal, Comparison of methods for the measurement of fibre/matrix
adhesion in composites. Composites 23, 2–27 (1992)
17. M.J. Pitkethly, J.P. Favre, U. Gaur, J. Jakubowski, S.F. Mudrich, D.L. Caldwell, L.T. Drzal, M.
Nardin, H.D. Wagner, L. Di Landro, A. Hampe, J.P. Armistead, M. Desaeger, I. Verpoest, A
round-robin programme on interfacial test methods. Compos. Sci. Tech. 48, 205–214 (1993)
18. R.J. Young, Monitoring deformation processes in high-performance fibres using Raman
spectroscopy. J. Text. Inst. 86, 360–381 (1995)
19. G. Katagiri, H. Ishida, A. Ishitani, Raman-spectra of graphite edge planes. Carbon 26, 565–571
(1988)
20. C. Galiotis, D.N. Batchelder, Strain dependences of the 1st-order and 2nd-order Raman-spectra
of carbon-fibers. J. Mater. Sci. Lett. 7, 545–547 (1988)
21. M.A. Montes-Morán, R.J. Young, Raman spectroscopy study of HM carbon fibres: effect of
plasma treatment on the interfacial properties of single fibre/epoxy composites, Part I: fibre
characterization. Carbon 40, 845–855 (2002)
22. K. Kong, L.B. Deng, I.A. Kinloch, R.J. Young, S.J. Eichhorn, Production of carbon fibres
from a pyrolysed and graphitised liquid crystalline cellulose fibre precursor. J. Mater. Sci. 47,
5402–5410 (2012)
23. M.A. Montes-Morán, R.J. Young, Raman spectroscopy study of HM carbon fibres: effect
of plasma treatment on the interfacial properties of single fibre/epoxy composites, Part II:
characterisation of the fibre/matrix interface. Carbon 40, 857–875 (2002)
24. R.F. Gibson, Principles of Composite Material Mechanics, 3rd edn. (CRC Press, Boca Raton,
2012)
25. D. Hull, T.W. Clyne, An Introduction to Composites Materials (Cambridge University Press,
Cambridge, 1996)
26. A. Kelly, N.H. Macmillan, Strong Solids, 3rd edn. (Clarendon, Oxford, 1986)
27. H. Krenchel, Fibre Reinforcement (Akademisk Forlag, Copenhagen, 1964)
28. A. Kelly, Strong Solids (Clarendon, Oxford, 1966)
29. H.L. Cox, The elasticity and strength of paper and other fibrous materials. Br. J. Appl. Phys. 3,
72–79 (1952)
30. R.J. Young, P.A. Lovell, Introduction to Polymers (CRC Press, Boca Raton, 2011). Chapter 24
31. A. Kelly, W.R. Tyson, Tensile properties of fibre-reinforced metals—copper/tungsten and
copper/molybdenum. J. Mech. Phys. Solids 13, 329–350 (1965)
Chapter 14
Micromechanical Evidences on Interfibre
Failure of Composites
14.1 Introduction
Composite materials have been presented from the very beginning as non-isotropic,
being this its main characteristic. This fact certainly plays a key role from a
structural point of view when designing in terms of a required level of stiffness
or when evaluating the distributions of stresses with a stacking sequence involving
plies with different orientations of the fibres.
However, the non-isotropic character of a lamina is irrelevant from a failure point
of view, the important question when damage or failure is under consideration is the
non-homogeneous character of composite materials.
When revising classical books on composites, see Jones [1], Tsai and Hahn
[2], Agarwal and Broutman [3] or Halpin [4], among many others, there is a
vast part of them impregnated of the non-isotropic character of the material,
whereas only a small part is associated with the non-homogeneous constitution of
composites. Micromechanics has been the word traditionally used to recognize the
non-homogeneous character of a composite. However, the micromechanic parcel
has been usually associated with the predictions of stiffness or strength properties
of the composite as a function of the properties of the constituents, fibre and matrix.
It is also remarkable that when generating the failure criteria of composites, a big
emphasis was put originally in adapting existing failure criteria to the non-isotropic
character of the material, rather than recognizing the internal structure, the non-
homogeneous character of composites.
14.2 Tools
In the open model, Williams [6], the interface crack is assumed to be open (with
traction-free crack faces), whereas in the contact model, Comninou [7], the faces
near the crack tip are assumed to be in contact under the load application.
None of the two models is free of inconsistencies and/or difficulties in its
application. Thus, in the open model, oscillations and interpenetrations appear near
the crack tip indicating the necessity of having a contact zone. On the contrary, when
the contact model is used, although the solution obtained with this model is strictly
speaking the only one physically correct, the size of the contact zone may be so
small that is not coherent with a continuum representation of the material.
Having these limitations in mind, there are situations where only one of these
models can be considered adequate, but there are also cases, like those appearing
in the studies carried out in this chapter, where both models are required along
the fracture process. Thus, a switch from the open model to the contact model is
required when analysing the growth of the fibre-matrix debonding.
An excellent example to clarify these questions is the one represented in
Fig. 14.1, where it is shown a crack between dissimilar materials, one stiff, that can
be associated with a fibre, and another compliant, that can be associated with the
matrix. On the one side where the natural relative displacements coincide with the
shear applied (left in the figure), a large contact zone will appear, whereas on the
other side where the relative displacements of the lips are opposed to the shear
applied (right in the figure), a bubble appears. Notice that a very small contact zone
exists at this tip of the bubble, being represented in Fig. 14.1. Thus, the left-hand
side of the crack would be controlled by a contact model approach whereas the
right-hand side would be controlled by an open model approach.
A more detailed explanation on the application of these concepts into the study
of interfibre failure of composites may be found in Mantič et al. [8], París et al. [9]
and in Correa et al. [10], among others.
The debonding between fibre and matrix already presented is only a first phase
of interfibre damage. As situations where a debonding crack abandons the interface
and enters the matrix may appear in the study carried out, a criterion for the
competition between an interface crack (the debonding crack between fibre and
matrix) controlled by IFM and a regular crack (that penetrating into the matrix)
controlled by LEFM is required.
The criterion used, represented in expression (14.1), is formulated in energetic
basis, He and Hutchinson [11], where Gint is compared with Gkink , in terms of their
ratios with respect to the corresponding critical values.
The values of the toughness associated with a bimaterial system strongly depend
on the mixity k . In this study, the critical value of the energy associated with the
interfacial crack will be evaluated, when required, by means of expression (14.2),
362 F. París et al.
External load
“Bubble”
Fig. 14.1 Morphology of a crack between dissimilar materials under shear external load
Hutchinson and Suo [12], with reference to the critical value of G associated with
mode I.
2
C .
Gint k/ D Gint
1C Œ1 C tan .1 / k (14.2)
The problem under consideration can be clearly illustrated by means of Fig. 14.2,
Gamstedt [14], which corresponds to a glass fibre system.
The objective of this section is to illustrate how a micromechanical analysis
of fibre-matrix debonding can explain a question associated with interfibre failure
prediction. Many criteria for this failure take as reference the stress vector associated
with the plane of failure, see, for instance, Hashin and Rotem [15], Puck and
14 Micromechanical Evidences on Interfibre Failure of Composites 363
Schürmann [16] and Sun et al. [17]. The original idea of Hashin would have led,
for a 3D stress state, to the following criterion:
n 2 nl 2 nt 2
C C D1 (14.3)
Yc S ST
where n , nl and nt are, respectively, the components of the stress vector, normal to
the plane .n /, tangent to the plane in the direction of the fibre .nl / and tangent to
the plane transverse to the fibre .nt /. The denominators correspond to the allowables
of the components of the stress vector.
This equation was maintained for tension and modified for the failure under
compression. The important thing in any case is that all this family of interfibre
failure criteria did not involve other components of the stress state different from
those included in the stress vector associated with the plane of failure. Thus, with
reference to the two-dimensional case represented in Fig. 14.3a, only components
22 and 23 would be involved in the failure, whereas no role on it would be assigned
to the component 33 . However, a detailed view of the failure described in Fig. 14.3b
indicates that the debonding between fibre and matrix might be affected by the
presence of the component 33 .
To analyse this question, a BEM model of a single fibre embedded in a matrix
under a bidirectional stress state is going to be used, Fig. 14.4. This model allows
a debonding crack to be considered, avoiding interpenetrations between fibre and
matrix and giving the possibility of generating a contact zone between fibre and
matrix, as indicated in Fig. 14.4. Notice that the model developed includes one
symmetry plane, the angle d represented in the figure being associated with one
half of the total debonding crack. The properties of the materials used are specified
in Table 14.1.
The ERR associated with the debonding crack is evaluated using the crack
closure technique, expression (14.4), the crack growing from an angle ˛ to ˛ C
ı.ı << ˛/
Z
1 ˛Cı
G.˛/ D f.rr /˛ .ur /˛Cı C .r /˛ .u /˛Cı gd (14.4)
2ı ˛
364 F. París et al.
σ23 σ23
σ22 σ22
σ22
σ23 σ22
σ23
crack
σ23 σ23
σ33 σ33
Fig. 14.3 Micromechanical implications of associating the failure at a plane with the components
of the stress vector associated with such plane: (a) a general view, (b) single fibre view
MATRIX
s0
FIBRE qd
2
a
s0
Fig. 14.4 Single fibre configuration with damage modelled by a debonding crack between fibre
and matrix
The evolutions of the ERR and their counterparts corresponding to modes I and
II are shown, as a function of the debonding angle, in Fig. 14.5, for the particular
case of having a single nominal stress. The values of G are calculated for different
debonding angles (starting by a total 10ı debonding) and are normalized by dividing
the values of G by G0 D .1 C km =8/02 ˛ (Toya [18] and Murakami [19]), where
km D 3 4 m , is the shear modulus of the matrix and 0 denotes the value of the
applied tension.
14 Micromechanical Evidences on Interfibre Failure of Composites 365
0,80
GI/G0
GI
0,70 GII
GII/G0
GG/G
TOTAL
) σ0
0
0,60
0,50
G/G0
0,40
0,30
0,20
0,10
0,00
0 20 40 60 80 100 120 140 160
Semidebonding angle, qd (º)
Fig. 14.5 Normalized total, G, and corresponding fracture modes parts, GI and GII , under tension
normal to the plane of failure (case T 0)
As can be clearly observed, mode I controls initial small debondings, as the crack
is oriented almost perpendicular to the nominal stress. A mixed mode appears, with
more preponderance of mode II as the crack progresses, reaching a pure mode II,
with the generation of a contact zone, when the semidebonding angle reaches a
value in the model of about 60ı –65ı .
If a second nominal stress is applied, parallel to the plane of failure, then a clear
alteration in the values of G is appreciated, see Fig. 14.6, where only the values
of the total ERR are represented for simplicity. The values represented correspond
to the case of a single tension .T 0/ versus the same tension with an additional
compression parallel to the assumed plane of failure .T C/.
The presence of a compression parallel to the plane of failure produces a
noticeable increment of G, which would lead to think that it produces an acceleration
in the breaking procedure, the value of the breaking load decreasing in presence of a
lateral compression. To check these predictions, experimental tests were conducted,
París et al. [20], applying a tension normal to the fibres in a 90ı specimen,
with a lateral precompression. Notice that although the tension and the transverse
compression are applied in the model at the same time, as the application of the
compression involves a conforming contact problem, the two loads, compression
and tension can be sequentially applied in the test for simplicity reasons, without
altering the problem (assuming no friction is involved).
If it is assumed that the components of the stress vector control the interfibre
failure, the ultimate strength of the specimen would not be affected by the presence
of a lateral precompression. On the contrary, if the numerical BEM predictions are
correct, then a significant decrease in the ultimate strength ought to be observed
when increasing the lateral precompression.
366 F. París et al.
1,4
1,2
1,0
0,8
G/G0
0,6
0,4
T-C
0,2
T-0
0,0
0 20 40 60 80 100 120 140
Semidebonding angle (qd)
Fig. 14.6 Micromechanical implications of associating the failure at a plane with the components
of the stress vector associated with such plane
The results, shown in Fig. 14.7, regardless of their noticeable scattering associ-
ated with a test performed under a load normal to the fibres and with the in-house
design of the rig, clearly indicate the adequacy of the following assumption: due
to the non-homogeneous character of the composite, the damage consisting on the
debonding between fibre and matrix is clearly affected by components of the stress
state not included in the stress vector.
The agreement between predictions based on micromechanics and the macroex-
perimental results makes us feel confident on the role that Micromechanics may
play in helping to understand different aspects of the behaviour of composites, as
will be shown in what follows.
Once a particular question affecting the interfibre failure criteria of composites has
been studied in the previous section, the complete morphology of the interfibre
failure under tension and under compression is now going to be studied in detail,
covering not only the debonding between fibre and matrix but also the complete
failure of the lamina, always contemplated from a micromechanical point of
view.
14 Micromechanical Evidences on Interfibre Failure of Composites 367
1,1
1
0,9
Ultimate Strength/YT
0,8
0,7
0,6
0,5
0,4
0,3
0,2
0,1
0
0 0,1 0,2 0,3 0,4 0,5 0,6 0,7 0,8 0,9 1 1,1
Lateral Compression/YT
Fig. 14.7 Ultimate strength of the 90ı specimen as a function of the lateral precompression
applied
Kinking Interface
crack
s0
FIBRE qd
2
a
s0
MATRIX
Fig. 14.8 Model of the damage in a single fibre under tension once the debonding crack kinks and
penetrates into the matrix
1,2
G/G0
G/G0
1,0 GC/G(l=0,2)
Gc/G0 0 (λ=0.2)
GC/G(l=0,25)
Gc/G0 0 (λ=0.25)
0,8 GC/G(l=0,3)
Gc/G0 0 (λ=0.3)
G /G0
0,6
0,4
0,2
0,0
0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150
Semidebonding angle, qd (º)
Fig. 14.9 Evolution of G and GC as a function of the semidebonding angle
of the ERR G such as those associated with the toughness of the material, GC , need
to be known. As the mixity of the crack changes with the size of the debonding,
it will be necessary to represent along with G the evolution of the values of GC
with the growth of the debonding, following the indications given in Sect. 14.2, as
a function of a material parameter and the mixity . An interesting discussion
about the different options to represent the mixity can be found in Mantič and París
[21] and París et al. [9], having chosen the mixity associated with G, G , in this
study.
Figure 14.9 shows the evolution of G and GC for different values of the parameter
. It has been considered that the failure criterion .G D GC ) holds for a small
semidebonding angle (10ı).
The evolution appearing in Fig. 14.9 clearly shows that starting at a semidebond-
ing of 10ı , an unstable growth takes place, the crack stopping .GC > G/ for an
interval between 55ı and 65ı , depending on the value of taken. Notice that the
horizontal evolution of GC observed is due to the appearance of a pure mode II,
once a contact zone is generated.
Following on from the development of damage, once the crack stops, a compe-
tition between continuing the growth along the interface under a pure mode II in a
stable way, and kinking and penetrating into the matrix, appears.
The possibility of kinking must be investigated under the light of a criterion,
which in this case will be taken as the Maximum Circumferential Stress , Erdogan
and Sih [22]. Figure 14.10 reproduces the orientations of the directions associated
with the maximum values of for different positions of the debonding crack tip. A
range is represented to cover the predictions for different values of the radius from
the tip.
14 Micromechanical Evidences on Interfibre Failure of Composites 369
Range of θ(σθmax)
90º
θd
Fig. 14.10 Evolution, in the neighbourhood of the debonding crack tip, of the angle of maximum
circumferential stress in the matrix
It can be clearly observed how in the range under interest (in the neighbourhood
of 60ı ) the orientations of the maximum value of correspond to an almost vertical
direction, i.e., almost normal to the external applied load.
To elucidate now the question under consideration, this is, the progression of the
damage, the values of the ERR G for both a debonding crack and a potential kinking
crack (values corresponding to the vertical direction and those corresponding to
the direction of maximum are represented for completeness) are represented in
Fig. 14.11.
The values of G associated with the kinking crack are, in the interval of interest,
significantly greater than those associated with the debonding crack. Additionally,
if the values of G associated with a kinked crack progressing into the matrix
are represented, as shown in Fig. 14.12, it is clearly observed that, for obvious
geometrical reasons, this growth will take place under mode I, which makes even
more plausible the abandon of the crack of the interface to kink and penetrate into
the matrix.
All of the above lead to consider that the growth of the damage could be
visualized by the scheme shown in Fig. 14.13, under the acronym T 0. The
influence of the presence of another transverse load is also represented in Fig. 14.13
under the acronym T 0:5T for the case of the superposition of a transverse tension
with a value of one half of the reference tension, and under the acronym T C for
the case of the superposition of a transverse compression with the same value than
of the reference tension.
In general terms, it can be observed that the main alterations when a secondary
transverse load appears in a failure dominated by tension occur in the interface
crack growth, in particular, this growth could be larger for the T nT cases and
slightly smaller for the T nC situations. Kinking towards the matrix seems not
370 F. París et al.
1,0
0,9
0,8
0,7
0,6
G/G0
0,5
0,4
0,3 crackthrough
throughthethe matrix following
crack matrix following the the maximum
maximum
0,2 circumferentialstress
circumferential stress direction
vertical crack in the matrix
vertical crack in the matrix
0,1 crackalong
crack along
thethe interface
interface
0,0
0 10 20 30 40 50 60 70 80 90 100
2,5
GI/G0
GI
GII/G0
GII
2,0 GG/G
0
1,5
G/G0
1,0
0,5
0,0
0,0 0,2 0,4 0,6 0,8 1,0
Length of the kinked crack/Fibre radius
Fig. 14.12 Values of G for a kinked crack progressing into the matrix in a vertical direction
T-0 a qd
T-0.5T
>a >qd
T-C <qd
<a
Fig. 14.13 Qualitative comparison of the phases of the mechanism of damage for different
combinations of bidimensional loading under a reference tension
A parallel scheme to that employed in Sect. 14.4.1 for tension will now be developed
for compression. Further details of the study summarized in this section can be
found in Correa et al. [10, 24, 25].
In this section the configuration under investigation that is going to be analysed
by BEM is shown in Fig. 14.14, which includes, similarly to what was considered
in Fig. 14.8, not only the debonding between fibre and matrix but also a potential
abandon of the fibre–matrix interface, which is kinking into the matrix.
The first question is to identify the location at which the damage may start. In the
case of tension studied in Sect. 14.4.1 this location was immediate, whereas in this
one is more complicated as there are no significant tension stresses normal to the
interface, the shear stresses being the most significant stresses along the interface.
As these stresses reach their maximum value for angles ˛ D 45ı , 135ı, 225ı and
315ı, none of them seems to be the most plausible location of the beginning of the
damage. In this study the location ˛ D 135 has been selected.
Once the onset of the damage has been identified, then a similar procedure than
in Sect. 14.4.1 is going to be followed. Thus, an initial debonding of 10ı centred at
˛ D 135ı (a debonding between 130ı and 140ı) is going to be studied by a BEM
model. Figure 14.15 shows the morphology of the crack under the application of the
compressive load.
372 F. París et al.
α =130º
MATRIX FIBRE
δ ur
Undeformed position
of the interface
σ =140º
Fig. 14.15 Morphology of the incipient debonding crack under compression represented by the
radial displacements of the lips of the debonding crack
14 Micromechanical Evidences on Interfibre Failure of Composites 373
0,050
GG/G0
0,045
Glambda=0.2
Gc C/G0 (λ=0.2)
0,040
Glambda=0.25
Gc C/G0 (λ=0.25)
0,035 Glambda=0.3
Gc C/G0 (λ=0.3)
0,030
G/G0
0,025
0,020
0,015
0,010
0,005
0,000
130 140 150 160 170 180 190 200 210 220
Position of the crack tip, a(º)
Fig. 14.16 Evolution of G and GC for the debonding crack under compression as a function of the
semidebonding angle
The BEM predictions of the morphology of the crack indicate that, in accordance
with what was explained in Fig. 14.1, there is a closure of the lips at the top tip,
indicating that the fracture growth is at this tip controlled by the contact model. On
the contrary, the presence of a bubble is detected at the bottom tip of the incipient
debonding crack, indicating that the fracture growth is in this case controlled by the
open model.
This information leads to consider that the crack will propagate from the tip
placed at 140ı (under a mixed mode with significant participation of mode I) instead
of propagating at the tip placed at 130ı (under a pure mode II). In accordance with
this idea the evolution of G is now investigated, and for the sake of brevity, only the
total values of G will be shown, see Fig. 14.16.
The inspection of the predictions shown in Fig. 14.16 leads to conclude that the
incipient debonding crack would grow in an unstable manner until an angle ˛ in
between 200ı and 210ı, slightly depending on the material parameter , stopping
the growth at this position.
The next step now is to investigate the stress state in the neighbourhood of the
tip of the debonding crack once it has stop. As in Sect. 14.4.1, the criterion taken
to establish the direction of kinking is that based on the maximum circumferential
stress. Figure 14.17 represents in this case the direction of maximum circumferential
stress in accordance with a BEM model for two physically relevant values of the
radius at which the calculations are performed. The analytical predictions based on
the Comninou contact model are also included for comparison.
It is now relevant to mention that in Sect. 14.4.1 the key point connecting the
analysis of the evolution of the damage at micromechanical level, as well as the
observed damage at mesomechanical level, i.e., the apparent plane of failure of the
374 F. París et al.
130º
206º 2
r -->0 Comninou
r=0.001a (BEM)
r =0.01a (BEM)
48.4º
50º 58º
Fig. 14.17 Comparison between prediction of maximum circumferential stress for physically
relevant radii based on BEM and analytical predictions based on contact fracture model
lamina, was the direction of maximum circumferential stress at which the debonding
crack abandoned the fibre-matrix interface, kinking into the matrix. In the case
studied in Sect. 14.4.1 this direction was that normal to the load applied. Following
now a similar reasoning, the expected apparent angle of failure would be one of an
order in between 50ı and 58ı .
Figure 14.18 shows a picture of the final configuration of the broken specimen,
indicating that the apparent plane of failure found in practice is in the indicated
order, 55ı in the case shown, which is clearly within the range found in the
predictions. This evidence is a clear confirmation of the role that micromechanics
can play in predicting the behaviour of composites as an equivalent homogeneous
material.
All of this leads to consider that the growth of the damage could be visualized
by the scheme shown in Fig. 14.19, under the reference C 0. The influence of the
presence of another transversal loading is also represented in Fig. 14.19 under the
reference C 0:1T for the case of the superposition of a transversal tension with a
value of one tenth of the reference compression, and under the reference C 0:5C
for the case of the superposition of a transversal compression with a value of one
half of the reference compression.
In general terms, it can be observed that the main alterations when a transverse
loading appears in a failure dominated by compression are an accelerative effect
of the external secondary tensile load on failure versus an inhibitive effect of
the external secondary compressive one. In particular, larger lengths of interfacial
growth are found as the secondary tension increases, whereas shorter lengths are
predicted for the compressive case. In fact, these results justify the selection of the
particular cases in the figure, as a sufficient secondary tension increase could turn
14 Micromechanical Evidences on Interfibre Failure of Composites 375
Fig. 14.18 Scheme of the failure of a specimen under compression normal to the direction of the
fibre
135º α α
C-0
θkink
C-0.1T
≈θkink
C-0.5C
Fig. 14.19 Qualitative comparison of the phases of the mechanism of damage for different
combinations of bidimensional loading under a reference compression
376 F. París et al.
the propagation into the mechanism of damage under tension, whereas a sufficient
secondary compression could limit the crack growth just to the interface. A detailed
description of these alterations can be found in Correa et al. [25].
Another parcel where Micromechanics has been able to play a clarifying role is
fatigue loading in the direction transverse to the fibres.
Gamstedt and Sjogren [26] had found that the debonding between a single
fibre and its surrounding matrix grew under certain combinations of tension and
compression cycles, a fact not easy to explain.
These experimental observations seemed to indicate, Fig. 14.20, that after several
T T cycles, the debonding angle maintained an almost constant value of 60ı 70ı .
However, entering into T C cycles led to increase the debonding angle.
To give an explanation of this fact, again a comparative study between the
evolution of G and GC with the debonding angle was carried out. Figure 14.21 shows
this evolution using in this case two values of the parameter and starting the study
from the debonding observed after the T T cycles, assuming that the conditions
for any additional debondings are fulfilled.
The evolutions shown in Fig. 14.21 support the observed phenomenon that
excursions into T C cycles produce an increase in the debonding angle (reaching
a position closed to 100ı ), which can lead to the appearing of a generalized level of
damage at macromechinal level in a laminate.
The role of residual stresses associated with the curing process is very well known
and its effects understood at mesomechanical level, these stresses being involved
in the design of a structural component. The reason for the appearance of these
14 Micromechanical Evidences on Interfibre Failure of Composites 377
0,12
0,10
0,08
G/G0
0,06
0,04
G/G0(Compressive
Gtotal (T-0) Case)
0,02 GC/G0(λ=0.2)
Series1
GC/G0(λ=0.25)
Series3
0,00
0 10 20 30 40 50 60 70 80 90 100 110 120
Semidebonding angle, qd (º)
Fig. 14.21 Evolution of G and GC along the T C cycle as a function of the semidebonding angle
Having in mind the different phases of the appearance of damage transverse to the
fibres under tension (nucleation of an incipient debonding crack, propagation of this
debonding crack, kinking and penetration into the matrix), it is believed that the
one being mainly affected by the generation of residual stresses at micromechanical
level has to be necessarily the debonding between the fibre and the matrix, as there is
a stress state originated by curing acting on the interface along which the debonding
will take place.
378 F. París et al.
0,7
GI/G0 (ΔT=-80ºC)
Series5
G/G0
Series7 (ΔT=-80ºC)
0,5 GI/G0
Series1 (ΔT=0ºC)
GII/G0 (ΔT=0ºC)
Series2
G/G0 (ΔT=0ºC)
G/G0
0,4 Series3
0,3
0,2
0,1
0,0
0 20 40 60 80 100 120 140 160
Semidebonding angle, qd (º)
Fig. 14.22 G evolutions versus the debonding angle of a glass fibre system under tension for two
curing conditions
The distribution of G versus the debonding angle has been studied for two
different situations of curing, one considering a cooling of T D 80 ıC and the
other one without considering it, i.e., T D 0 ı C, see Fig. 14.22.
The evolutions of G and corresponding parts GI and GII for the two values of T
studied are very similar qualitatively speaking, but it can be easily noticed how the
presence of residual stresses .T D 80 ı C) produces a protective effect on the
interface, against the onset and progression of a debonding crack along it.
No significant variations were observed by the presence of curing stresses in the
other phases of the damage, Correa et al. [28].
Experimental tests were designed to check the adequacy of the predictions. To
this end, two cycles of curing were designed for a carbon composite, Cycle 1 being
the recommended by the supplier of the pre-preg and Cycle 2 having a significant
decrease in the temperature of curing, both with reference to a 12-ply unidirectional
graphite-epoxy laminate:
• Cycle 1: heat to 121 ı C in 300 , hold at 121 ı C for 1 h, heat to 177 ı C in 300 , hold
at 177 ıC for 3 h and cool to room temperature.
• Cycle 2: heat to 121 ı C in 300 , hold at 121 ıC for 48 h and cool to room
temperature.
With reference to the results found in terms of the strength of the laminates, a
mean value of 38:4 MPa, with standard deviation of 4:9 MPa, was found for the
composite corresponding to cycle 2, with a curing temperature of 121 ı C, whereas
14 Micromechanical Evidences on Interfibre Failure of Composites 379
0,050
G I/G0 ΔT=-80ºC
GI
0,045 GII/G0 ΔT=-80ºC
GII
0,025
0,020
0,015
0,010
0,005
0,000
130 140 150 160 170 180 190 200 210 220
Position of the lower crack tip, a(º)
Fig. 14.23 G evolutions versus the debonding angle of a glass fibre system under compression for
two curing conditions
a strength of 48:6 MPa was the mean strength found, with a standard deviation
of 5 MPa, for the composite corresponding to Cycle 1, with a curing temperature
of 177 ı C. These results confirm the numerical predictions already obtained, i.e.,
the presence of higher residual stresses has a protective effect under conditions of
transverse tension.
Similar predictions as those made in Sect. 14.6.1 for the case under tension are now
made for the case under compression. Thus, Fig. 14.23 shows the evolutions of G,
GI and GII for two cases of curing, one at room temperature T D 0ı and another
with a T equal to 80 ı C.
It can be seen that although there are differences between the two cases studied,
they are not quite significant, as the shape of the curves are practically equal and
there are only small differences from a quantitative point of view. Thus, based on
these predictions, not too many differences are expected in the implementation of
experimental tests. Further details can be found in Correa et al. [29].
As in the case of tension previously studied, two curing cycles were designed to
elucidate the effect on residual stresses.
• Cycle 1: heat to 177 ı C in 500 , hold at 177 ıC for 2 h and cool to room
temperature.
• Cycle 2: heat to 121 ı C in 300 , hold at 121 ıC for 54 h and cool to room
temperature.
380 F. París et al.
In this case the value of the mean strength found for the 12-ply unidirectional
graphite-epoxy laminate corresponding to Cycle 1 was 220 MPa, with a standard
deviation of 22 MPa, whereas for the laminate cured with Cycle 2 the mean strength
was 212 MPa, with a standard deviation of 11 MPa. The proximity of these two
results confirms the numerical predictions in the sense that the micromechanical
residual curing stresses do not play a significant role under a compression loading.
All the results presented till this moment correspond to the case of a single fibre
embedded in a matrix under different types of loading. Two questions of different
nature are now going to be studied. The first refers to the case of considering a
secondary fibre in the neighbourhood of the damaged fibre, while the second refers
to the micromechanical view of the scale effect in composites.
All the studies shown so far in this chapter correspond to a single fibre embedded
in a matrix. In this section the study will be extended to the presence of a second
fibre to see if the results associated with the new situation considered may offer
information about the local weakness of the composite due to its heterogeneous
nature.
To this end, the new configuration shown in Fig. 14.24 is going to be studied
again using a BEM model. As shown in the figure, there is a central fibre presenting
a debonding crack and a secondary fibre which orientation with respect to the
damaged fibre is defined by the distance r2 and the angle 2 .
The interest is again focused on studying the evolutions of the ERR by the
debonding crack, yet now with an emphasis in how the position of the secondary
fibre alters these evolutions. A detailed description of this question can be found in
Sandino et al. [30]. Here, to summarize, a particular attention will be devoted to a
couple of questions that are representative of the damage under consideration, that
corresponding to the debonding between fibre and matrix.
First of all, the extension of the damage versus the position of the secondary fibre
in terms of the angle 2 is studied as seen in Fig. 14.25 for a fixed position of the
secondary fibre .r2 = 2:4 r1 /. The end of unstable growth corresponding to the single
fibre case is taken as a reference, representing the same feature, the end of unstable
growth, for different positions of the secondary fibre.
The information obtained from Fig. 14.25 is represented by means of a graph, to
facilitate its comprehension, in Fig. 14.26.
Thus, it can be seen from Fig. 14.26 that the positions of the secondary fibre
marked in red correspond to configurations to which the extension of the damage is
14 Micromechanical Evidences on Interfibre Failure of Composites 381
Fig. 14.24 Model including a primary fibre with an interface crack and an undamaged secondary
fibre
120
End of unstable growth (single-fibre)
Semidebonding angle,θd
110
End of unstable growth (two-fibres)
100
90
80
70
60
50
0 20 40 60 80 100 120 140 160 180
Position of the secondary fibre, θ2
Fig. 14.25 End of unstable growth of the debonding crack versus the position of the secondary
fibre
greater, see quantitatively this effect in Fig. 14.25. On the contrary, those positions
marked in green correspond to configurations in which the extension of the unstable
growth is smaller, see quantitatively this effect in Fig. 14.25. Those positions marked
in black correspond to configurations where the end of unstable growth is not
affected by the position of the secondary fibre.
The second question studied is the effect of the position of the secondary fibre
in the amount of energy released by the debonding crack. In this sense, the value
of G associated with the first debond considered, i.e., d D 10ı in this case,
could be chosen as a representative parameter of the mechanism of damage at
micromechanical level, in order to establish a connection between the micro- and
382 F. París et al.
Primary fibre
The unstable growth of the interface crack stops at larger lengths of the crack than in the reference case
The unstable growth of the interface crack stops at smaller lengths of the crack than in the reference case
The unstable growth of the interface crack is similar to the reference case one
Fig. 14.26 Graph showing the effects of the presence of the secondary fibre on the extension of
the unstable growth
macro-mechanical level, Correa et al. [28]. Thus, a comparison between the value
of the ERR G for a fixed initial semidebonding of 10ı associated with different
values of 2 is presented in Fig. 14.27.
It can be clearly seen how positions of the secondary fibre aligned with the
primary fibre produce higher values of G. A particular attention is required to
understand the case of 2 equal to 0ı (the secondary fibre is aligned with that under
consideration and ahead of the location where the debonding is going to take place),
as the high value of G obtained has required the use of a different scale near 2 equal
to 0ı , see Fig. 14.27.
The explanation of this fact is that the opening mode of the debonding crack
(i.e. the mode I associated with the growth of the debonding) is dominant for small
values of the debonding and is conditioned by the stiffness of the material occupying
the area ahead of the zone where the debonding is going to take place. Thus, the area
14 Micromechanical Evidences on Interfibre Failure of Composites 383
40
0
20 40 60 80 100 120 140 160
-20
200
-40 100
0
-60
-100
0 20 40
-80
Position of the secondary fibre, θ2(º)
Fig. 14.27 Variation of G.d D 10ı / versus 2 in relation to the reference case with detail for
2 D 0ı
occupied by the stiff material (glass in this case) represented by the fibre forces the
crack to open much more than in the case in which the matrix, less stiff, occupies
the same area, then giving rise to a greater value of G.
A summary of the results shown in Fig. 14.27 is represented in Fig. 14.28 where
positions of the fibres in red would correspond to positions where a higher level of
load is required for an incipient crack to propagate, whereas those in green would
correspond to positions where a smaller level of load is required. Thus, in simple
terms, positions in red would represent an accelerative effect on the appearance of
debonding, whereas those in green would represent a retardant effect in terms of
appearance of the debonding.
With regards to the distance between the fibres, it is observed that most effects
disappear for r2 =r1 D 2:5, although the disappearance of the totality occurs for
r2 =r1 D 7.
Primary fibre
Higher level of load needed to initiate the propagation
Lower level of load needed to initiate the propagation
Mechanics (FFM), Mantič [34], this view of the problem gives a minimum size of
the expected damage in a weak lamina (typically 90ı lamina). As the predictions are
based on a mesomechanical view of the problem, it will be reviewed in this chapter
from a micromechanical point of view.
The configuration now studied, represented in Fig. 14.29, Velasco et al. [35],
corresponds to a Œ0; 90nS laminate, and a double scale model, based again in
BEM, is going to be carried out. The outer 0ı lamina .t0 D t90 D 170 m)
is modelled at a mesoscale level with an orthotropic behaviour in the transversal
longitudinal plane under consideration, plane xy in Fig. 14.29. The central 90ı
lamina is also considered at a mesoscale level having isotropic behaviour in the
plane under consideration. This 90ı lamina includes a micromechanical cell in
which a fibre (7:5 m of diameter) surrounded by matrix is modelled, maintaining
the fibre volume (55 %) of the composite.
The properties of the four materials involved in the model are given in Table 14.2.
Table 14.2 Elastic properties of the materials involved in the multiscale model
Material Properties
0ı ply (orthotropic) E11 D 135 GPa, E22 D 8:75 GPa, E33 D 8:75 GPa, v12 D 0:3, v13 D 0:3, v23 D 0:4, G12 D 4:75 GPa
90ı ply (isotropic) E.E22 D E33 / D 8:75 GPa, v D 0:3
Matrix (epoxy, isotropic) E D 4:2 GPa, v D 0:32
Fibre (carbon, isotropic) E.E22 / D 15 GPa, D 0:2
14 Micromechanical Evidences on Interfibre Failure of Composites
385
386 F. París et al.
y u
u
0 degrees lamina t 0°
6
GG(N=3)
(n=3)
5 GG(N=1)
(n=1)
GG(N=0.5)
(n=0.5)
4
G(J/m2)
0
0 20 40 60 80 100 120 140 160
Semidebonding angle, qd
Fig. 14.30 Scale effect on the values of the ERR G corresponding to a debonding crack between
the fibre and the matrix
Now the question is to see the influence that the scale effect (the value of n
in the laminate) has on the main mechanism of interfibre damage: the debonding
between the fibre and the matrix. To this end, the evolution of G as a function of the
debonding angle is going to be studied for different values of n. Figure 14.30 shows
such evolution for n D 0:5, 1, 3.
Figure 14.30 clearly shows, due to the coincidence of the evolutions of G for the
three values of n studied, that the damage under consideration is not affected at all
by the scale effect (for the sake of brevity only the total value of G is represented, not
having found any difference neither in GI nor in GII ). The appearance of this damage
would then be connected to the ‘quality’ of the interface/interphases between the
fibre and the matrix. Thus, a similar level of external loading ought to lead to a
similar level of damage, if any, of this type.
First of all the appearance of this damage is widely observed in the literature and
has been also detected by the authors, see Fig. 14.31.
14 Micromechanical Evidences on Interfibre Failure of Composites 387
Looking for a specific correlation with the scale effect, Saito et al. [36] show
the presence of this type of damage, isolated debonding between the fibre and the
matrix with equivalent morphology, for two laminates with different values of the
parameter n, and at the same level of loading.
14.8 Conclusions
A study on the interfibre failure based on Micromechanics has been carried out in
this chapter. A single fibre model embedded in matrix has been used to investigate
the onset and growth of the phase of the mechanism considered to be of major
importance in the interfibre failure, the debonding between the fibre and the matrix.
The problem has been studied using the approaches associated with Interfacial
Fracture Mechanics, and using BEM as the most appropriate tool to deal with a
problem involving singular stress states and non-linear features derived from the
contact between debonded surfaces of fibre and matrix.
First of all, a micromechanical study has been conducted to clarify the question
of adopting the stress vector at mesomechanical level associated with a plane as
the entity responsible for the interfibre failure at this plane. A micromechanical
explanation has been given, showing the involvement in the failure of stress compo-
nents non-associated with this stress vector. The numerical predictions have been
supported by experimental tests involving lateral precompressions on specimen
under tension transverse to the fibres.
388 F. París et al.
The morphologies of the damage found under tension and compression have
been explained by means of micromechanical analyses, establishing the correlation
between the predictions and the experimental observations. The effect of superim-
posing a lateral transverse load to the nominal transversal one has also been studied.
The role of micromechanical curing residual stresses has been investigated
using again a BEM model. The predictions showed a significant influence of these
residual stresses under transverse tension and almost no influence under transverse
compression. Different manufacturing procedures were tried to obtain materials
with similar properties but curing with different temperatures. Transverse testing
under tension and compression corroborates the predictions, showing that residual
stresses play a protective effect under tension, not having any relevant role under
compression.
An excursion to a more complicated model involving a secondary fibre has
also been carried out, identifying the configurations that precipitate or delay the
appearance of failure, under tension and under compression.
Finally, the well-known size effect has been visited from a micromechanical
point of view. It has been proven that there is no size effect for the appearance of an
isolated debonding of a fibre from the matrix. In other words, similar damage at this
micromechanical level ought to be observed for a certain level of load, regardless
the origin of the scale effect, the thickness of the lamina that affects the failure,
which has given rise to the in-situ strength concept.
It has to be said that the model used in most of the cases is a very simple one
involving a single fibre. It could be argued that it is quite simple to model all
the possibilities one fibre may find in an actual laminate. However, in the light
of the information obtained with this simple model versus others that are more
complicated, involving a complete generation and propagation of damage, one may
say that only physical properties associated with foundations of Fracture Mechanics
are used, fitting properties of a model not being required. It is also important to
mention that the purpose of the analysis carried out is not to establish predictions
on strength but to understand the mechanisms of failure that may help to design
composites with improved performance, as well as to be able to propose physically
based failure criteria.
Acknowledgements This study was supported by the Spanish Ministry of Education and
Science/Economy and Competitiveness and the Junta de Andalucía (Projects MAT2013-45069-
P, DPI 2012-37187 and P11-TEP-7093).
References
30. C. Sandino, E. Correa, F. París, Numerical analysis of the influence of a nearby fibre on the
interface crack growth under transverse tensile load. Eng. Fract. Mech. (2016). http://dx.doi.
org/10.1016/j.engfracmech.2016.01.022
31. A. Parvizi, K.W. Garret, J.E. Bailey, Constrained cracking in glass fibre-reinforced epoxy
cross-ply laminates. J. Mater. Sci. 13, 195–201 (1978)
32. D.L. Flaggs, M.H. Kural, Experimental determination of the in situ transverse lamina strength
in graphite/epoxy laminates. J. Compos. Mater. 16(2), 103–116 (1982)
33. I.G. García, V. Mantič, A. Blázquez, F. París, Transverse crack onset and growth in cross ply
Œ0=90S laminates under tension. Application of a coupled stress and energy criterion. Int. J.
Solids Struct. 51, 3844–3856 (2014)
34. V. Mantič, Interface crack onset at a circular cylindrical inclusion under a remote transverse
tension. Application of a coupled stress and energy criterion. Int. J. Solids Struct. 46, 1287–304
(2009)
35. M.L. Velasco, F. París, J.C. Marín, J. Justo, A. Barroso, E. Graciani, Numerical and experimen-
tal study on the failure of non-conventional laminates, in ECCM17—European Conference on
Composite Materials, Munich (2016)
36. H. Saito, H. Takeuchi, I. Kimpara, Experimental evaluation of the damage growth restraining
in 90 layer of thin-ply CFRP cross-ply laminates. Adv. Compos. Mater. 21, 57–66 (2012)
Chapter 15
Progressive Damage in Fibre-Reinforced
Composites: Towards More Accurate
and Efficient Computational Modelling
and Analysis
15.1 Introduction
Fig. 15.2 Open-hole tension (OHT). (a) Illustration of damage mechanisms [2]; (b) cross-
sectional view of the fracture paths [1]
parameters. Recently, such models have had some success in modelling size effects
in open-hole specimens of various layup sequences under tension and compression
loading. However, the interactions between delamination and in-plane damage are
not explicitly modelled; indeed, in cases where such interactions are especially
influential, there is evidence to suggest that such models may not be adequate.
Moreover, it has been found that for the results to be mesh-independent and free
of modelling artefacts, the elements must at least be of a minimum size, which
means that the required number of degrees of freedom, and hence computational
effort, is substantial even for coupon-level laminates. These shortcomings have led
to methods that aim to model matrix cracks explicitly in composites, such as the
extended finite element method (XFEM), phantom node, augmented finite element
and floating node methods. Yet another proposed method is to integrate the cohesive
element with XFEM (i.e. the XFEM-CE). While these methods show promise,
their effectiveness and accuracy have yet to be extensively demonstrated because
results of the analysis need to be carefully compared with experiments. Nonetheless,
they provide useful computational platforms on which to test the effects of various
material properties such as strength and fracture toughness, which are sometimes
unavailable or difficult to obtain. However, explicitly modelling large numbers of
cracks is unlikely to reduce computational effort.
Arguably the most straightforward way of modelling cracks with the standard finite
element method is through remeshing, i.e. the mesh is regenerated with respect to
crack propagations to conform to the new crack surfaces. However, in the case
of composites, the large count of matrix cracks, often at least in the thousands,
render this approach impractical. An efficient method, where matrix cracks can be
represented without changing the existing mesh of the problem, is preferred for
the modelling of composites progressive damage and failure. The simplest way of
achieving this is through element stiffness degradation. If a crack is determined to
propagate through a certain element, the stiffness of this element will be degraded
to approximate the effect of this crack on the load-bearing capability of the element.
Earlier stiffness degradation methods typically use stress- or strain-based failure
criteria to determine the onset of failure, and once it is satisfied, the stiffness of the
element is degraded to a fixed small value close to zero (it is usually not assigned
zero stiffness to avoid numerical ill-conditioning). Such an approach suffers from
the mesh dependency on strength predictions, as there is no consistency enforced
between energy dissipation and element failure [10]. In addition, as the crack tip is
effectively modelled as a blunt notch in the mesh (Fig. 15.4), the stress field around
the crack tip is poorly reproduced. As a result, the directionality of the mesh may
dictate the next element to fail, and the method would then also suffer from the mesh
dependency on the prediction of fracture paths [11].
15 Progressive Damage in Fibre-Reinforced Composites: Towards More. . . 395
Crack Failed
Fig. 15.6 Different scaling methods used in [1]. (a) In-plane scaling: scaling w.r.t hole diameter
d; (b) thickness scaling [45m /90m /-45m /0m ]ns
The size effects imply that using the coupon test results directly on the design of
actual composite structures can be non-conservative. A good understanding and the
accurate prediction of the size effects are important prerequisites for the safe and
optimal design of composite structures. While the in-plane size effects are known
398 B.Y. Chen and T.E. Tay
0 10mm
to exhibit in materials with notches [23], the effects of sublaminate scaling and ply
scaling on the laminate strength and failure mode are particular characteristics of
composite laminates. The experiments revealed that the laminate strength decreases
with respect to both sublaminate scaling and ply scaling (Fig. 15.10). The laminate
strength quickly reaches a plateau with respect to sublaminate scaling, while it keeps
on decreasing with respect to ply scaling. In addition, sublaminate scaling causes the
laminate to fail in a more brittle manner, while ply scaling leads to the delamination
failure mode (Fig. 15.7). The ply-scaling effect can be explained by the fact that
thicker ply blocks have larger amounts of transverse shear strain energy available to
drive the delamination [12, 15, 18]. The physics of the sublaminate-scaling effect
is however not that evident. While the weakest link theory can explain the initial
decreasing trend of the strength, it cannot predict the plateau in Fig. 15.10 as the
laminate gets thicker. One explanation is that the cracks in surface plies of the
thinner laminate propagate until the innermost 0 plies and then cause delamination,
which effectively release the stress concentration and delay the fibre failure [1, 22].
However, such a mechanism was found to be insufficient to explain the sublaminate-
scaling effect [15].
The authors have performed numerical modelling and analysis of OHT and OHC
specimens with the aim of studying size effects in [24–26], using the smeared crack
method for ply damages, i.e. fibre breaking and matrix cracking. The smeared crack
method assumes that a cohesive crack develops with the element, once the matrix
and fibre failure criteria are met in an element. The cohesive crack is not explicitly
represented. Instead, the element stiffness terms gradually degrade according to a set
of cohesive laws corresponding to matrix cracking and fibre breaking, respectively.
15 Progressive Damage in Fibre-Reinforced Composites: Towards More. . . 399
500 500
Strength (MPa)
Strength (MPa)
400 400
300 300
200 200
100 100
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9
Laminate thickness t (mm) Laminate thickness t (mm)
The failure criteria are typically stress-based and take the following general form:
f .¢/ 1 D 0 (15.1)
where ¢ is the stress at a material point. Various failure criteria, such as Tsai-
Wu, Hashin and Pinho, have been used [24–26]. The cohesive law is energy based,
which ensures that the energy dissipation in an element upon final failure (i.e. zero
remaining stiffness) equals the surface energy required to propagate the fictitious
cohesive crack within the element. In this way, the energy dissipation within
elements are regularised against the fracture toughness of the fictitious cohesive
crack. Different cohesive laws exist for the smeared crack approach. A linear
cohesive softening law was used in [24], while a zigzag softening law was proposed
in [25, 26]. Figure 15.11 demonstrates a linear cohesive law for fibre tensile failure.
Here, u is the fibre-direction displacement, 1 is the fibre-direction stress; Xt is the
fibre tensile strength; Gf is the fibre tensile fracture toughness (which is the area
under the 1 u curve); u0 is the u at failure onset; uf is the u at final failure; and
E1 is the intact material modulus in the fibre direction. After the onset of failure, the
material behaviour is allowed to soften. In smeared crack models, the softening is
represented by a degradation factor applied to the intact modulus:
1 D E1 .1 d/ 1 (15.2)
It may be verified that given the cohesive law of Fig. 15.11, the degradation factor
may be calculated as
uf .u u0 /
dD (15.3)
u uf u0
400 B.Y. Chen and T.E. Tay
1800
linear
1600 zigzag
1400
1200
(MPa)
1000
800
1
600
400
E1
200
0
0 0.05 0.1
strain
1
The element stiffness reduces with respect to loading until all the energy needed
to propagate a fictitious cohesive crack has been dissipated. The zigzag cohesive
law in [25, 26] modifies the straight softening line to a zigzag curve, as shown in
Fig. 15.12. In this way, for small-enough increments of u, a positive tangent modulus
exists to facilitate convergence.
When using cohesive laws to model fracture, the fracture toughness is an
important parameter as it determines the area under the stress-displacement curve in
Fig. 15.11. The fracture toughness is a material property whose value is determined
15 Progressive Damage in Fibre-Reinforced Composites: Towards More. . . 401
Fig. 15.13 Length of fibre-bundle pull-outs in blocked (left) and single (right) 0 plies [27]
from so-called standard experiments; it is used to define the cohesive law for the
respective mode of fracture. Experiments have been developed to measure the
toughness of the different damage mechanisms in composites. Of particular interest
is the toughness measured for fibre breaking. Fibre breaking in composites rarely
appears as a well-defined crack with smooth surfaces. It usually presents itself as
a rugged fracture band, full of pulled-out fibre bundles. This toughness, termed
translaminar fracture toughness, is then measured as the energy dissipation per unit
area of fracture projected onto the midplane of the fracture band [28]. It is found
that the translaminar fracture toughness scales with respect to the thickness of the
ply block, because thicker ply blocks have larger extents of fibre-bundle pull-out,
i.e. larger widths of the fracture band (Fig. 15.13) [27]. Consideration of the ply-
blocking effect on translaminar fracture toughness was proposed in the numerical
modelling in [24]. In the FE model, the assigned value of the fibre fracture toughness
for the elements in m -blocked plies, Gm fc , is related to that for the elements in single
plies, Gsfc , through the following equation (see also Fig. 15.14):
fc D m Gfc
Gm s
(15.4)
If the above equation is used to define the cohesive laws for fibre breaking
in different ply blocks of the laminate, then the sublaminate-scaling effect of the
laminate strength in Fig. 15.10 can be explained by the average fibre-breaking
toughness of the 0 plies in the laminate. For the sublaminate-scaling cases, the
middle 0 plies are doubly stacked together; hence, their toughness (denoted by Gdfc )
is twice of that of a single ply according to Eq. 15.4, i.e. Gdfc D 2Gsfc . Let Gafc
represent the average toughness of all the 0 plies in the laminate. For a laminate of
layup [45/90/-45/0]ns, Gafc is
From the above equation, it may be verified that as the number of sublaminates
(i.e. n) increases, Gafc decreases but converges to Gsfc . This trend corresponds to the
sublaminate-scaling effect of the laminate strength in Fig. 15.10. Indeed, our work
has shown that models which take into account of this ply-blocking effect of the
translaminar fracture toughness can accurately predict the OHT strengths and the
size effects, especially the sublaminate-scaling effect [24]. The ply-blocking effect
of the translaminar fracture toughness has been used in [25] for a different set of
OHT problems and in [26] for a set of OHC problems. These work repeatedly
confirmed that considering this ply-blocking effect in the modelling was critical
for the prediction of the laminate strengths and the size effects. Recently, analytical
solutions of the translaminar fracture toughness were derived in [29], shedding light
on the detailed mechanics of the ply-blocking effect.
The importance of delamination modelling has also been examined in cases
where the laminates are not failed by delamination [24]. The objective was to
see if delamination modelling was always needed, even when the failure mode
was predominantly by ply failure and pull-out. It was found that models without
considering delamination, such as the classical lamination theory (CLT)-based
models, gave rise to a spurious mesh dependency which caused increasing over-
prediction with respect to mesh refinement [24]. Therefore, unless the failure mode
could be determined to be brittle, delamination (even though occurring only locally)
should always be considered in the model.
The thickness-direction discretisation of the numerical model for the open-hole
problem is illustrated in Fig. 15.15. A layer of continuum shell element is used
for each ply. A user-material subroutine is coded to implement the smeared crack
model for ply damages described above. A layer of cohesive element is used to
model delamination. Details of the cohesive properties can be found in [24].
15 Progressive Damage in Fibre-Reinforced Composites: Towards More. . . 403
A layer of continuum
shell element per ply
Sublaminate-scaling, t = 4 mm Ply-scaling, t = 4 mm
600 600
Strength (MPa)
Strength (MPa)
500 500
400 400
300 300
200 200
0 2 4 6 8 10 12 14 0 2 4 6 8 10 12 14
d (mm) d (mm)
500 500
400 400
300 300
200 200
0 1 2 3 4 5 0 1 2 3 4 5
t (mm) t (mm)
Fig. 15.16 OHT strength, in-plane and thickness size effect predictions [24]
The predictions of the OHT strengths on the two sets of experiments are summarised
in Fig. 15.16 (forthe set in [1]) and Fig. 15.17 (forthe set in [10]) and those of
the OHC strengths in Fig. 15.18. It can be seen that the smeared crack model can
accurately predict the strengths of most of the test cases. In addition, the predicted
failure patterns are also in good agreement with the experimental X-ray images (see
404 B.Y. Chen and T.E. Tay
700
Experiment
600 Simulation
Strength (MPa)
500
400
300
200
0 2 4 6 8 10 12
Hole diameter (mm)
Fig. 15.17 OHT strength, in-plane size effect predictions [25]
Experiment t : laminate thickness
Simulation d : hole diameter
Sublaminate-scaling, t = 4 mm Ply-scaling, t = 4 mm
450 450
Strength (MPa)
Strength (MPa)
400 400
350 350
300 300
250 250
5 7 9 11 13 15 5 10 15 20 25 30
d (mm) d (mm)
400 400
350 350
300 300
250 250
1 2 3 4 5 1 2 3 4 5
t (mm) t (mm)
Fig. 15.18 OHC strength, in-plane and thickness size effect predictions [26]
Figs. 15.19 and 15.20). The matrix splitting and subsequent delamination are well
predicted. In the case of OHC, the local bulging of surface plies around the hole is
also captured in the simulation, as shown in Fig. 15.21.
15 Progressive Damage in Fibre-Reinforced Composites: Towards More. . . 405
a Total delam 90 45
Partial delam
Intact Loading 0
Fig. 15.19 OHT delamination predictions (a) in [24] vs. experimental X-ray images (b) in [30]
a b c
Fig. 15.20 OHC damage predictions (b–c) vs. experiment (a) [26]: (a) experimental X-ray image,
(b) ply damage patterns, (c) delamination patterns
The largest discrepancies between experimental data and simulation results come
from the OHT cases of the 4 mm-thick ply-scaling laminates (top-right plot of
Fig. 15.16). The dominant failure mode of these cases is delamination. The first
significant load drop in these cases is not caused by fibre failure in the 0 plies
but by delamination at the -45/0 interface. The failure patterns appear to be well
captured by the model, as shown in Figs. 15.19 and 15.22. The delamination
follows the boundaries defined by the adjacent matrix cracks, which indicate that
the delamination is induced by the preceding matrix cracks (see Fig. 15.22).
However, prediction of the full extent of the delamination is more difficult. The -
406 B.Y. Chen and T.E. Tay
Fig. 15.21 OHC out-of-plane displacement contour shows local buckling around the hole [26]
45/0 delamination observed in the experiment covers almost the entire area of the
laminate, and it occurs before the fibre failure in the 0 plies. In the simulation,
the delamination area is much smaller and fibre failure initiated around the notch
(Fig. 15.22).
In the case of the delamination failure mode, the prediction of the laminate failure
is equivalent to the prediction of the delamination propagating from the existing
matrix cracks. The accuracy of the latter depends on how well the deformation and
stress concentrations are captured, especially at the intersections between matrix
cracks and interfaces. In the smeared crack method, a matrix crack is not explicitly
15 Progressive Damage in Fibre-Reinforced Composites: Towards More. . . 407
represented. In the event of matrix cracking, the related stiffness terms of the
failed ply element are reduced to almost zero. As this is equivalent to modelling
a sharp crack tip on the interface as a blunt notch (see Fig. 15.23), the local
stress concentration, which is the driving mechanism for subsequent delamination
growth, is therefore lost. Consequently, when this crack-induced delamination is
the dominant failure mode of the laminate, the accuracy of the strength prediction
cannot be guaranteed by employing the smeared crack method. Similar findings
have also been reported in [15].
The preceding section shows that approaches combining the smeared crack method
for modelling ply damages (taking into account the effect of ply blocking on
translaminar fracture toughness) and the use of cohesive elements for delamination
can accurately predict the size effects of most open-hole laminates cases. Exceptions
are for cases where the matrix crack-induced delamination plays a critical role in
the failure process. In these cases, the approach gives less satisfactory results due
to its inherent inability to represent sharp matrix crack tips. This limitation has led
to the development of alternative novel numerical methods, namely, the integrated
XFEM-cohesive element (XFEM-CE) method [31] and the floating node method
(FNM) [32–34]. These new methods aim to explicitly represent the sharp matrix
cracks and the stress concentrations on the interfaces to more accurately capture the
interactions between matrix cracks and delamination.
408 B.Y. Chen and T.E. Tay
The XFEM-CE method is the integration of XFEM and cohesive element tech-
nologies in a single model to account for both matrix cracks and delamination in
composites. The standard XFEM is used to model matrix cracks within the existing
mesh through Heaviside enrichment. The displacement of point x in a cracked ply
element, u(x), is approximated by the standard XFEM as
where N is the standard shape function matrix, uN is the nodal displacement vector,
H is the Heaviside enrichment function and aH is the enriched nodal degree of
freedom (DoF) vector. The location and orientation of the matrix crack are described
by the Heaviside enrichment function. It is assumed that the matrix crack within an
element is always planar; hence, it can be fully characterised by a point on the crack
interface, xc , and the interface normal vector, nc . These two variables are usually
determined by the failure criterion or propagation criterion used in the analysis [31].
H is then defined as [31]
(
nc .xxc /
jnc .xxc /j
; nc .x xc / ¤ 0
H .x/ D (15.7)
0; nc .x xc / D 0
It is readily seen that H equals to C1, 0 and 1 for points above, on and below
the crack interface, respectively. Figure 15.24 illustrates the enriched nodes and
elements with respect to a matrix crack in a ply mesh. The standard subdomain
integration technique is used to integrate each subdomain of the cracked element
[35]. Note that a basis-shifted H function can be used to avoid additional treatment
of the affected elements (i.e. blending elements) [35]. With XFEM, the matrix
crack is no longer represented in a smeared fashion, since the sharp discontinuity
across the crack interface is explicitly represented. In addition, the direction of
the matrix crack can be enforced by the definition of H. The mesh dependency of the
smeared crack method on the prediction of crack propagation directions is therefore
eliminated.
As in the smeared crack method, in the XFEM-CE, cohesive elements are also
inserted between plies of different fibre angles to model possible delamination.
However, the cohesive elements in the XFEM-CE method are formulated differ-
ently. In order to capture accurately the stress field in the cohesive elements, the
Heaviside enrichment is also included in the solution space of the cohesive element
(Fig. 15.25). In a standard cohesive element, the separation vector • at a point x on
the crack interface is calculated from the nodal displacement vector uN through the
standard cohesive element shape function matrix Nu :
Fig. 15.24 Nodal enrichment and element partition in XFEM: (a) finite elements with enriched
nodes and (b) partitioned elements [31]
where D is the constitutive matrix and is the domain of the cohesive interface.
On the other hand, in the XFEM-CE method, the Heaviside enrichment in the ply
element, H, is included in the calculation of •:
Kuu Kua
KCE D (15.11)
Kau Kaa
where
Z
’“
K D NT’ DN“ d ; ’; “ 2 fu; ag (15.12)
Therefore, the enriched cohesive element takes into account the additional DoF
with respect to the Heaviside enrichment, such that a discontinuous solution is
reproduced on the interface [31].
410 B.Y. Chen and T.E. Tay
Fig. 15.26 Delamination migration from the top interface (0/90) to the bottom interface
(90/0) [31]
The XFEM-CE method has been shown to be able to capture the mechanisms
of delamination migration [31]. Figure 15.26 shows the migration of delamination
in a cross-ply laminate. The migration event is clearly captured. Figure 15.27
shows the delamination migration in an angle-ply laminate of layup [0/45]s. The
delamination, which propagates along the top, 0/45 interface, subsequently branches
onto the lower, 45/0 interface. However, the propagation also continues on the
top interface. In an experiment conducted by Tao and Sun [3], the end notch
flexure (ENF) specimen is used to investigate delamination migration. The ENF
specimen is a quasi-isotropic laminate with a pre-delamination located on the lowest
interface (Fig. 15.28a). The failure patterns predicted by XFEM-CE are shown in
Fig. 15.28b–g. A close agreement is reached between experimental delamination
patterns and prediction, except for the delamination at the 45/90 interface, where
experimental observation shows that it is of a triangular shape bounded within the
90 and 45 matrix cracks, while the predicted shape is rectangular. This suggests that
in the simulation, delamination continues to propagate along the 45/90 interface
despite the occurrence of the migration. This could lead to numerical difficulties
such as excessive iteration or non-convergence of solution, as finding the dominant
path of fracture is not always easily attainable by the solver in the presence of
multiple competing delaminations.
Recently, the XFEM-CE method has been applied to model the delamination
migration in a double cantilever beam test of a [0/30/45]s laminate (Fig. 15.29) [36].
The computational results show reasonable agreement with experiment [37, 38].
Recently, a new method named the floating node method (FNM) is developed for
the modelling of complex crack networks in composites in an integrated manner,
where all the cracks (including both matrix cracks and delamination) are modelled
within the same element [32–34]. It is, in essence, a new way of defining elements.
The FNM offers a powerful new approach of modelling composites. It is capable
15 Progressive Damage in Fibre-Reinforced Composites: Towards More. . . 411
Fig. 15.27 Delamination migration in a [0/45]s laminate: (a) deformation of the laminate (b)
matrix crack in the 45 plies [31]
Fig. 15.28 Failure patterns: (a) experimental patterns in (b) matrix crack in the -45 ply,
(c) delamination at the 90/-45 interface, (d) matrix crack in the 90 ply, (e) delamination at the
45/90 interface, (f) matrix crack in the 45 ply and (g) delamination at the 0/45 interface [31]
412 B.Y. Chen and T.E. Tay
Fig. 15.29 Delamination migration study in a [0,30,45]s laminate specimen: (a) experiment,
(b) simulation
15 Progressive Damage in Fibre-Reinforced Composites: Towards More. . . 413
a b
Element A
Nodes : 1, 2, 5, 4
: Node
Element B
Nodes : 2, 3, 6, 5
Fig. 15.30 Standard FEM definition and mesh [34]: (a) element definition, (b) mesh
where u is the solution field, N is the shape function matrix of the element and
uN is the element nodal solution vector. This equation is only valid when the
element domain remains continuous throughout the analysis. When there is a
discontinuity, the XFEM approach is to introduce the Heaviside enrichment function
and additional nodal DoFs to represent the discontinuity, as detailed in the previous
section.
In the FNM, a different approach is taken. First, the element definition is enriched
to include not only the nodal connectivity but also the edge connectivity (Fig. 15.31).
Supposing that a planar crack is to propagate across an element domain during
analysis (Fig. 15.32), the locations of the crack nodes (i.e. the intersection points
of the crack and the element edges) are determined from the failure criterion or
the propagation criterion used in the analysis. The crack node locations are used to
define the Heaviside enrichment function in XFEM (see Eq. 15.7). However, they
also define the subdomains A and B in Fig. 15.32. The nodal coordinate matrices
of the two subdomains are given by
414 B.Y. Chen and T.E. Tay
a b Element A
Nodes : 1, 2, 5, 4
Edges : 1, 2, 3, 4
: Node
: Edge i Element B
Nodes : 2, 3, 6, 5
Edges : 5, 6, 7,-2
Fig. 15.31 Element definition and mesh in FNM [34]: (a) element definition, (b) mesh
: Crack node
: Nodal DoF
Fig. 15.32 A crack creates four new nodes on the element boundaries
n o
x A D x cC ; x cC ; x 3 ; x 4 (15.14)
IV II
˚
x B D x 1 ; x 2 ; x c
IV
; x c
II
(15.15)
These two matrices effectively define the domains of two sub-elements (SEs). For
isoparametric linear finite elements, their shape functions in the natural coordinates
are NA and NB . Assuming small deformations and linear elasticity apply within each
SE, the stiffness matrices, KA and KB , are fully defined as
Z
KA D BTA DBA det .JA / d„A (15.16)
„A
Z
KB D BTB DBB det .JB / d„B (15.17)
„B
where „A and „B are the integration domains in the natural space of the two SEs. In
the case of Fig. 15.32, they both refer to the standard square integration domain of a
quadrilateral element. BA and BB are their respective strain-displacement matrices,
JA and JB are their Jacobian matrices, and D is the constitutive matrix. KA and
KB can be integrated using the standard Gauss quadrature rule. Unlike in XFEM,
no subdomain integration is needed here. This not only simplifies the calculation
15 Progressive Damage in Fibre-Reinforced Composites: Towards More. . . 415
but also improves the integration accuracy [33]. The stiffness matrix of the whole
element is simply the assembly of KA and KB :
KA
KD (15.18)
KB
Similarly, the force vectors of the SEs, QA and QB , are also defined (ignoring the
body force):
Z
QA D NTA t det .JA / d A (15.19)
A
Z
QB D NTB t det .JB / d B (15.20)
B
where t is the external traction and A and B are the integration boundaries in
the natural space of the two SEs. The force vector of the whole element is thus the
assembly of those of the two SEs:
QA
QD (15.21)
QB
where
UTA D uTC ; uTC ; uT3 ; uT4 (15.23)
cIV cII
n o
UTB D uT1 ; uT2 ; uTc
IV
; uT
c
II
(15.24)
In the above equation, ui; iD14 are the nodal DoFs of the existing nodes of the
element which are already available. What are missing, however, are the DoFs of the
crack nodes, i.e. ucC ; ucC ; uc
IV
; and uc
II
. As these crack nodes do not exist before
IV II
analysis, these nodal DoFs are not defined, and they are hence not available to form
the complete DoF vector in Eq. (15.22).
Clearly, in order to form the complete equilibrium Eq. (15.22), we require the
four DoF vectors. In conventional FEM, DoF vectors are always associated with
nodes and hence understood to be nodal DoF vectors. However, mathematically,
they are vectors which may be used to describe the DoFs of other entities. Let us
now also define DoFs for the edges of the element (see Fig. 15.33). These DoFs are
not associated to any particular node a priori. Hence, they are named floating DoFs,
416 B.Y. Chen and T.E. Tay
: Nodal DoF
: Floating DoF
: Crack node
Fig. 15.33 Floating DoFs are defined for two edges of the element
in contrast with nodal DoFs. We may use the floating DoFs of the cracked edges
to complete the DoF vector in Eq. (15.22). The floating DoFs then have a direct
physical meaning, i.e. the solution vectors at the crack nodes.
In existing FEM packages, the DoFs generally cannot be defined without being
associated with nodes. In order to allocate the floating DoFs to edges, extra nodes
are defined, and they are associated with the edges of the element, i.e. they share
the same connectivity with the edges. However, the coordinates of these nodes are
not used in the analysis. Only their DoFs are used to form the DoF vector in Eq.
(15.22). Therefore, these extra nodes can have arbitrary coordinates, hence the name
“floating” node.
The two-dimensional (2D) version of the FNM has been used to predict the
classical matrix crack-delamination interaction phenomenon in the crack density
evolution problem of cross-ply laminates under tension [33]. The FNM predicts
accurately the matrix crack density evolution with respect to the applied stress
(Fig. 15.34) and captures the experimentally observed inverse relationship between
the saturation crack density and the thickness of the 90 plies (Fig. 15.35) [39]. When
modelling the self-similar propagation of a crack, the virtual crack closure technique
(VCCT) may be used to calculate the energy release rates at crack tips [40, 41]. In
the FNM, information such as the nodal forces at the crack tip and the crack opening
at the crack wake are available to the element ahead of the crack tip. As a result,
the VCCT can be naturally incorporated into the FNM, and failure criteria based
on energy release rates can be employed to determine the propagation of cracks.
The combined FNM and VCCT method has predicted accurately the delamination
migration in cross-ply tape laminates [32]. More details of the 2D applications of
the FNM can be found in [32–34, 39].
Recently, the three-dimensional (3D) version of FNM has been developed for
modelling progressive failure in angle-ply laminates [34]. Figure 15.36a shows
the definition of a FNM ply element. It is composed of 24 nodes in total. Nodes
1–8 represent the vertices of the brick. Two floating nodes are allocated to each
horizontal edge of the element (for simplicity, it is assumed that matrix cracks only
traverse the horizontal edges). The floating nodes are shared with neighbouring
elements through their common edges according to the edge connectivity of the
element. When a matrix crack needs to be modelled within its domain, the FNM
15 Progressive Damage in Fibre-Reinforced Composites: Towards More. . . 417
2
1.8
1.6
Crack density (mm-1)
1.4
Experiment data
1.2
Predictions
1
0.8
0.6
0.4
0.2
0
0 200 400 600 800
Applied stress (MPa)
Fig. 15.34 Matrix crack density evolution prediction by the FNM [39]
[0/90]s
[0/902]s
[0/904]s
Fig. 15.35 Predicted saturation crack density with respect to the thickness of 90 plies (red dots
are failed integration points) [39]
ply element may be partitioned into SEs as shown in Fig. 15.36b–c, by using the
associated floating DoFs to represent the DoFs of the crack nodes on the cracked
edges.
The definition of a 3D FNM cohesive element is shown in Fig. 15.37a. The
element is composed of 24 nodes. Eight real nodes are the standard cohesive element
vertices and 16 floating nodes are associated with the eight breakable edges which
are parallel to the interface; these breakable edges are shared with the upper and
418 B.Y. Chen and T.E. Tay
a 8 22 21
7
23
24
14 13 20
4 19 3
5 15 6
17 18
16 12
11
1 2
9 10
Floating node
Crack position SE3
c
b Xl c
c
Xj
c
Xk
SE2
c
Xi
SE1
lower FNM ply elements. Based on the partitions of the upper and lower FNM
ply elements, the FNM cohesive element is partitioned accordingly to represent the
matrix crack edges explicitly on the interface. If the top ply has a matrix crack, then
the FNM cohesive element is partitioned to SE1 as shown in Fig. 15.37c. Similarly,
if the bottom ply has a matrix crack, then the partition is SE2 in Fig. 15.37d. If both
plies are cracked, then SE1 and SE2 are superimposed, each bearing half the weight
in integration. Each of the two SEs has 16 nodes, i.e. eight real nodes and eight
floating nodes on one of the two surfaces. SE1 has the eight floating nodes of the
upper cohesive surface, and SE2 has those of the lower surface.
With the FNM ply and cohesive elements defined above, an FNM laminate
element may be defined such that the FNM ply and cohesive elements are SEs
of the laminate element. Hence, multiple plies and interfaces may be represented
within such a laminate element, together with matrix cracks, delamination and
their intersections, represented, respectively, by the ply SEs and cohesive SEs, as
described above. The information of one SE is directly available to another. For
example, the matrix crack coordinates of ply elements are available to the interface
cohesive SEs, so that the latter may be partitioned to explicitly represent the matrix
crack edges on the interface. Furthermore, a planar mesh can be used for laminates
with general layups. The preprocessing of the model for different layups is much
simplified, and parametric studies on the layup are greatly facilitated (Fig. 15.38).
15 Progressive Damage in Fibre-Reinforced Composites: Towards More. . . 419
a 8 22 21 7
4 Floating node
23 14 13 20 3
19 Crack position
24 Interpolated node
15 12
5 16 17 18 6 11
1 2 26
9 10
c
SE1
Top ply matrix crack edge SE1b
SE1a
b
25
d
27
SE2 SE2b
Bottom ply matrix crack edge 28
SE2a
FNM
FNM ply cohesive
elements elements
Fig. 15.40 “Staircase” delamination patterns bounded by the matrix cracks and splits [34]
The FNM laminate element is applied to model the problem in Fig. 15.22,
i.e. the OHT of ply-scaled quasi-isotropic laminate [1]. The layup is [454 /904
/-454 /04 ]S . The laminate is failed by the delamination failure mode, where full-
scale delamination at the -45/0 interface disintegrated the laminate prior to the fibre
breaking in the 0 plies. Such a scenario was not captured by modelling in Fig. 15.22,
where the smeared crack method was used. Figure 15.39 shows the predicted failure
patterns by the FNM instead. It can be seen that the matrix cracks and longitudinal
splits, the triangular-shaped delamination in the 45/90 interface and the widespread
delamination on the -45/0 interface are all well predicted. The typical staircase
delamination patterns, where delamination on different interfaces are bounded by
the matrix cracks, are clearly captured (Fig. 15.40).
15 Progressive Damage in Fibre-Reinforced Composites: Towards More. . . 421
While the FNM shows powerful capabilities in modelling the physics and
mechanics of matrix cracks and delamination, it however comes with a price. With
much of the details of multiple cracks modelled explicitly, it is computationally
more costly than the smeared crack method. At this point in time, directly applying
the FNM to the modelling of relatively large composite structures is computationally
challenging because it is resource intensive.
15.5 Conclusions
Acknowledgement The authors are grateful for the support of the National University of
Singapore through the Competitive Strategic Grant No. WBS R265000523646.
422 B.Y. Chen and T.E. Tay
References
1. B.G. Green, M.R. Wisnom, S.R. Hallett, An experimental investigation into the tensile strength
scaling of notched composites. Compos. Part A Appl. Sci. Manuf. 38(3), 867–878 (2007)
2. M.R. Wisnom, S.R. Hallett, The role of delamination in strength, failure mechanism and hole
size effect in open hole tensile tests on quasi-isotropic laminates. Compos. Part A Appl. Sci.
Manuf. 40(4), 335–342 (2009)
3. J.X. Tao, C.T. Sun, Influence of ply orientation on delamination in composite laminates.
J. Compos. Mater. 32(21), 1933–1947 (1998)
4. E.S. Greenhalgh, C. Rogers, P. Robinson, Fractographic observations on delamination growth
and the subsequent migration through the laminate. Compos. Sci. Technol. 69(14), 2345–2351
(2009)
5. C. Canturri, E.S. Greenhalgh, S.T. Pinho, The relationship between mixed-mode II/III
delamination and delamination migration in composite laminates. Compos. Sci. Technol. 105,
102–109 (2014)
6. M.F. Pernice et al., Experimental study on delamination migration in composite laminates.
Compos. Part A Appl. Sci. Manuf. 73, 20–34 (2015)
7. J.A. Nairn, S.F. Hu, J.S. Bark, A critical-evaluation of theories for predicting microcracking in
composite laminates. J. Mater. Sci. 28(18), 5099–5111 (1993)
8. X.Q. Li et al., Experimental study of damage propagation in over-height compact tension tests.
Compos. Part A Appl. Sci. Manuf. 40(12), 1891–1899 (2009)
9. M.R. Wisnom, B. Khan, S.R. Hallett, Size effects in unnotched tensile strength of unidirec-
tional and quasi-isotropic carbon/epoxy composites. Compos. Struct. 84(1), 21–28 (2008)
10. P.P. Camanho, P. Maimi, C.G. Davila, Prediction of size effects in notched laminates using
continuum damage mechanics. Compos. Sci. Technol. 67(13), 2715–2727 (2007)
11. F.P. Van Der Meer, L.J. Sluys, Continuum models for the analysis of progressive failure in
composite laminates. J. Compos. Mater. 43(20), 2131–2156 (2009)
12. S.R. Hallett et al., An experimental and numerical investigation into the damage mechanisms
in notched composites. Compos. Part A Appl. Sci. Manuf. 40(5), 613–624 (2009)
13. S.T. Pinho, L. Iannucci, P. Robinson, Physically-based failure models and criteria for laminated
fibre-reinforced composites with emphasis on fibre kinking: Part I: development. Compos. Part
A Appl. Sci. Manuf. 37(1), 63–73 (2006)
14. S.T. Pinho, L. Iannucci, P. Robinson, Physically based failure models and criteria for laminated
fibre-reinforced composites with emphasis on fibre kinking. Part II: FE implementation.
Compos. Part A Appl. Sci. Manuf. 37(5), 766–777 (2006)
15. F.P. van der Meer et al., Computational modeling of complex failure mechanisms in laminates.
J. Compos. Mater. 46(5), 603–623 (2012)
16. T. Belytschko, T. Black, Elastic crack growth in finite elements with minimal remeshing. Int.
J. Numer. Methods Eng. 45(5), 601–620 (1999)
17. F.P. van der Meer, L.J. Sluys, A phantom node formulation with mixed mode cohesive law for
splitting in laminates. Int. J. Fract. 158(2), 107–124 (2009)
18. M.J. Swindeman et al., Strength prediction in open hole composite laminates by using discrete
damage modeling. AIAA J. 51(4), 936–945 (2013)
19. D. Ling, Q. Yang, B. Cox, An augmented finite element method for modeling arbitrary
discontinuities in composite materials. Int. J. Fract. 156(1), 53–73 (2009)
20. L.F. Kawashita, A. Bedos, S.R. Hallett, Modelling mesh independent transverse cracks in
laminated composites with a simplified cohesive segment method. CMC Comput. Mater.
Continua 32(2), 133–158 (2012)
21. J. Lee, C. Soutis, Measuring the notched compressive strength of composite laminates:
specimen size effects. Compos. Sci. Technol. 68(12), 2359–2366 (2008)
22. M.R. Wisnom, S.R. Hallett, C. Soutis, Scaling effects in notched composites. J. Compos.
Mater. 44(2), 195–210 (2010)
15 Progressive Damage in Fibre-Reinforced Composites: Towards More. . . 423
23. Z.P. Bazant, Size effect in blunt fracture: concrete, rock, metal. J. Eng. Mech. 110(4), 518–535
(1984)
24. B.Y. Chen et al., Numerical analysis of size effects on open-hole tensile composite laminates.
Compos. Part A Appl. Sci. Manuf. 47, 52–62 (2013)
25. M. Ridha et al., Modelling complex progressive failure in notched composite laminates with
varying sizes and stacking sequences. Compos. Part A Appl. Sci. Manuf. 58, 16–23 (2014)
26. Z.C. Su et al., Progressive damage modeling of open-hole composite laminates under
compression. Compos. Struct. 122, 507–517 (2015)
27. M.J. Laffan et al., Measurement of the in situ ply fracture toughness associated with mode
I fibre tensile failure in FRP. Part II: Size and lay-up effects. Compos. Sci. Technol. 70(4),
614–621 (2010)
28. S.T. Pinho, P. Robinson, L. Iannucci, Fracture toughness of the tensile and compressive fibre
failure modes in laminated composites. Compos. Sci. Technol. 66(13), 2069–2079 (2006)
29. S. Pimenta, S.T. Pinho, An analytical model for the translaminar fracture toughness of fibre
composites with stochastic quasi-fractal fracture surfaces. J. Mech. Phys. Solids 66, 78–102
(2014)
30. K. Song, Y. Li, C.A. Rose, Continuum damage mechanics models for the analysis of
progressive failure in open-hole tension laminates. AIAA 1861, 1–18 (2011)
31. T.E. Tay, X.S. Sun, V.B.C. Tan, Recent efforts toward modeling interactions of matrix cracks
and delaminations: an integrated XFEM-CE approach. Adv. Compos. Mater. 23(5–6), 391–408
(2014)
32. N.V. De Carvalho et al., Modeling delamination migration in cross-ply tape laminates.
Compos. Part A Appl. Sci. Manuf. 71, 192–203 (2015)
33. B.Y. Chen et al., A floating node method for the modelling of discontinuities in composites.
Eng. Fract. Mech. 127, 104–134 (2014)
34. B.Y. Chen, T.E. Tay, S.T. Pinho, V.B.C. Tan, Modelling the tensile failure of composites
with the floating node method. Comput. Methods Appl. Mech. Eng. 308, 414–442 (2016).
doi:10.1016/j.cma.2016.05.027
35. T.-P. Fries, T. Belytschko, The extended/generalized finite element method: an overview of the
method and its applications. Int. J. Numer. Methods Eng. 84(3), 253–304 (2010)
36. M. Tirvaudey, ARPE Final Report: Experimental and modelling analysis of delamination and
damage propagation in multidirectional composite material. École Normale Supérieure de
Cachan and National University of Singapore, 2015
37. X.F. Hu, B.Y. Chen, M. Tirvaudey, V.B.C. Tan, T.E. Tay, Integrated XFEM-CE analysis
of delamination migration in multi-directional composite laminates. Compos. A: Appl. Sci.
Manuf. 90, 161–173 (2016). doi:10.1016/j.compositesa.2016.07.007
38. T.E. Tay, X.F. Hu, M. Tirvaudey, V.B.C. Tan, Modelling delamination migration in multi-
directional composite laminates, in 17th European Conference on Composite Materials
(ECCM17), Munich, Germany, 26–30 June 2016
39. B.Y. Chen, PhD Thesis, Numerical modelling of scale-dependent damage and failure of
composites, in Department of Mechanical Engineering. National University of Singapore,
2013
40. E.F. Rybicki, M.F. Kanninen, A finite element calculation of stress intensity factors by a
modified crack closure integral. Eng. Fract. Mech. 9(4), 931–938 (1977)
41. R. Krueger, Virtual crack closure technique: history, approach, and applications. Appl. Mech.
Rev. 57(2), 109–143 (2004)
Chapter 16
Predicting Properties of Undamaged
and Damaged Carbon Fibre Reinforced
Composites
L. Neil McCartney
16.1 Introduction
(e.g. woven) and damage modes (e.g. delamination) can arise but these are usually
beyond the capability of analytical methods, and they will not be considered in this
chapter.
It is of interest to note that the National Physical Laboratory (NPL) has been
associated for over 60 years with the development of stress transfer theory. The
pioneering model of stress transfer for fibre systems (for paper in fact) was
developed by H L Cox at NPL in 1952 (see reference list) and is the first shear-
lag model for composite materials having perfectly bonded fibre/matrix interfaces.
In the early 1970s, John Aveston, George Cooper and Tony Kelly developed a shear-
lag model that applied to fibre reinforced composites having matrix cracks (or fibre
breaks) where fibre/matrix debonding occurs near the fractures. This model led on
to the well-known concept of stress transfer lengths and was used in an energy
balance equation to predict the matrix cracking stress leading to the very well-known
ACK theory. These pioneering papers continue today to be cited extensively in the
literature. Over the years, stress transfer has been a key research area at NPL by
the author that led to several other papers (see reference list) for both unidirectional
composites and laminates, the most recent being written in 2013.
The approach of this chapter will be to quote the analytical results that have been
derived in the literature, briefly describing the basis of their derivation, and giving
references where the full details are provided. Specific predictive examples will be
given mainly for the case of composites that are made using carbon fibres.
This section presents recommended analytical methods for calculating the thermo-
elastic constants of undamaged UD composites, some of which are almost impossi-
ble to measure. UD composites reinforced with carbon fibres are more realistically
modelled if the fibres are treated as transverse isotropic (i.e. axial properties differ
from properties in directions normal to the fibre axes). Both carbon fibres and
UD composites are characterised by seven independent thermo-elastic constants:
(1) axial Young’s modulus (EA ); (2) axial shear modulus (A ); (3) transverse
Young’s modulus (ET ); (4) axial Poisson’s ratio ( A ); (5) transverse Poisson’s
ratio ( T ); (6) axial thermal expansion coefficient (˛ A ) and (7) transverse thermal
expansion coefficient (˛ T ). Because of the assumption of transverse isotropy, the
transverse shear modulus (T ) may be calculated using the well-known relation
ET D 2T .1 C T /. The thermo-elastic properties appear in stress–strain relations
of the form:
1 T A
"rr D rr zz C ˛T T; (16.1)
ET ET EA
16 Predicting Properties of Undamaged and Damaged Carbon Fibre. . . 427
T 1 A
" D rr C zz C ˛T T; (16.2)
ET ET EA
A A 1
"zz D rr C zz C ˛A T; (16.3)
EA EA EA
rz z r
"rz D ; " z D ; "r D : (16.4)
2A 2A 2T
The stress and strain " components refer to cylindrical polar coordinates (r, ,
z), with the z-axis oriented parallel to the fibres. In laminate calculations it is often
useful to be able to estimate lamina properties from those of the reinforcements and
matrix that have been used to make the composite. It should be noted that if the
transverse isotropic properties of a UD composite have been measured, then models
can be used to estimate properties of the transverse isotropic fibres that are difficult
to measure such as Poisson’s ratios and transverse thermal expansion coefficients,
assuming that the matrix is well characterised and isotropic.
Additional composite properties that are beyond the scope of this chapter concern
transport coefficients such as the effective thermal and electrical conductivity.
However, it is of interest to mention that the first analytical formula predicting an
effective property of a two-phase composite is thought to be Maxwell’s [1] relation
for the effective electrical conductivity of an assembly of uniformly distributed
spherical particles having a different conductivity to that of the matrix in which
they are embedded. The key contribution of Maxwell was to develop a simple
very powerful methodology that enables estimates to be made of the effective
properties of many different types of composite. Results obtained using Maxwell’s
methodology do in fact lead to many results for particulate and fibre composites that
have been derived using alternative methods such as concentric cylinder models and
variational methods (see review by Hashin [2], and more recent publications by
McCartney and Kelly [44] and McCartney [3]).
For a multi-phase fibre reinforced composite, the isotropic matrix is reinforced
with N different types of fibre such that the fibre volume fraction of the ith fibre is
denoted by V if and that of the matrix by Vm such that
X
N
Vm C Vfi D 1:
iD1
composites, denoted with the superscript eff, may be estimated using Maxwell’s
methodology,
1 XN
Vfi Vm
D C m ; (16.5)
kT C m
eff
k C m
i
iD1 T
kT C m
! !
X
N
Vfi kTi Ai V m k m m X
N
Vfi kTi Vm k m
Aeff D C m T = C m T ; (16.6)
iD1
kT C m kT C m
i
iD1
kT C m
i kT C m
0 1
2 f .i/ f .i/ 2 !
4kTeff Aeff m X
N 4kT A m 2
4kTm m
B f .i/ C m
EAeff C D Vfi @ EA C f .i/ A C Vm Em C m ;
kTeff C m kT C m kT C m
iD1
(16.7)
1 X Vf Vm
N i
D C ; (16.8)
A C m
eff
iD1
A C m
i m C m
1 XN
Vfi Vm kTm m
D C ; where m D ; (16.9)
eff
T C
m iD1
i
T C
m m C m kTmC 2m
4Aeff ˛Teff C Aeff ˛Aeff X
N
f .i/ f .i/
EAeff ˛Aeff C D
Vfi EA ˛A C Vm Em ˛m
1=kTeff
C 1=m
iD1
(16.10)
f .i/ f .i/ f .i/ f .i/
XN 4A ˛T C A ˛A 4m .˛m C m ˛m /
C i
Vf f .i/
C Vm :
iD1 1=kT C 1=m 1=kTm C 1=m
X
N
Vfi kTi ˛Ti C Ai ˛Ai Vm kTm .1 C m / ˛m
C
iD1
kTi C m kTm C m
˛Teff C Aeff ˛Aeff D ; (16.11)
X
N
Vfi kTi Vm k m
C m T
iD1
kTi C m kT C m
where kT is the plane strain bulk modulus for a transverse isotropic solid defined by
2
1 2 .1 T / 4 A
D ; (16.12)
kT ET EA
eff 2 2
1 1 1 Teff 1 1 4 Aeff
D C eff C Aeff ; D eff ; (16.13)
ETeff 4eff
T 4k T EA ETeff 4eff
T 4kT EAeff
T and T .
it is possible to use the relations (16.5)–(16.13) to estimate values of Eeff eff
While the result (16.11) arises from an extension of the McCartney and Kelly [44]
analysis to fibre reinforcement, the result (16.10) is based on observations of the
structure of the relationship of multi-phase composite results to their two-phase
counterparts. There has, therefore, to be some doubt over the validity of (16.10),
which is regarded as a conjecture, even though it can be shown to predict the well-
known result for two-phase composites.
For an undamaged two-phase composite having fibre and matrix volume frac-
tions Vf and Vm (D1 Vf ), respectively, the multi-phase relations (16.5)–(16.11)
may be written in the form of a mixtures estimate of the effective property plus a
correction term as follows:
!2
1 Vf Vm 1 1
D f C m m Vf Vm ; (16.14)
kTeff kT kT 2 kTf kT
!
f 1 1 f
Aeff D V f A C V m m m A Vf Vm ; (16.15)
2 kTm kTf
2
f
A m Vf Vm
f
A D Vf A C Vm m
eff f
; (16.16)
1 C Vf m C Vm A
2
f
T m Vf Vm
f
T D Vf T C Vm m
eff f kTm m
; (16.17)
Vm T C Vf m C km C2 m T
2
f f
EAeff D Vf EA C Vm Em C 2 m A Vf Vm ; (16.18)
f f f f f f
EAeff ˛Aeff DVf EA ˛A C Vm Em ˛m C 2 m A ˛m Cm ˛m ˛T A ˛A Vf Vm ;
(16.19)
f f f
˛Teff C Aeff ˛Aeff D Vf ˛T C A ˛A C Vm .˛m C m ˛m /
(16.20)
C 2 1f
k1m
f f f
˛m C m ˛m ˛T A ˛A Vf Vm ;
kT T
Suffices f and m denote that a property refers to the fibre and matrix, respectively.
430 L.N. McCartney
Table 16.1 Properties of AS4 carbon fibre and Hercules 3501-6 epoxy
matrix
Matrix Young’s modulus (GPa) Em 4.2
Matrix Poisson’s ratio m 0.34
f
Fibre longitudinal Young’s modulus (GPa) EA 225
f
Fibre transverse Young’s modulus (GPa) ET 15
f
Fibre longitudinal Poisson’s ratio A 0.2
f
Fibre longitudinal shear modulus (GPa) A 15
f
Fibre transverse shear modulus (GPa) T 7
Matrix thermal expansion coefficient (106 /ı C) ˛ m 45
f
Fibre longitudinal thermal expansion coefficient (106 /ı C) ˛ A 0.5
6 ı f
Fibre transverse thermal expansion coefficient (10 / C) ˛ T 15
Fibre volume fraction Vf 0.6
f f f f
Note that as ET D 2T 1 C T then T D 0:07143
16 Predicting Properties of Undamaged and Damaged Carbon Fibre. . . 431
Having shown in Sect. 16.2 how to estimate the effective thermo-elastic properties
of undamaged fibre reinforced unidirectional composites, the corresponding effec-
tive properties of undamaged laminates will now be considered. The approach to be
adopted will provide explicit analytical formulae for the effective thermo-elastic
constants of laminates that can be recommended for application to composites
reinforced with transverse isotropic fibres which are embedded in an isotropic
matrix. This explicit approach avoids having to set up matrices and having to invert
them numerically.
First of all it is required to consider the stress/strain relations for a ply that is
angled with respect to the axial direction. One very important parameter that must
be clearly defined is the angle defining the fibre directions in an angled lamina
as shown in Fig. 16.1. The angle is measured from the global x1 - axis in a
clockwise direction when viewing from a point situated on the negative part of
Fig. 16.1 Diagram showing method of defining principal directions and coordinate system for an
angled lamina, and the angle specifying the fibre direction
432 L.N. McCartney
the global x3 - axis. Following rotation of the lamina the stress/strain equations,
including thermal expansion terms, relating global strain components to global
stress components are given, using the well-known Voigt matrix notation, by
2 3 2 3 2 3 2 3 2 3
"A "11 S11 S12 S13 0 0 S16 A V1
6" 7 6 " 7 6S 0 0 S26 7 6 7 6V 7
6 T7 6 22 7 6 12 S22 S23 7 6 T7 6 27
6 7 6 7 6 7 6 7 6 7
6 "t 7 6 "33 7 6 S13 S23 S33 0 0 S36 7 6 t 7 6 V3 7
6 7 6 7D6 7 6 7 C 6 7 T; (16.22)
6 t 7 6 2"23 7 6 0 0 0 S44 S45 0 7 6 t 7 6 0 7
6 7 6 7 6 7 6 7 6 7
4 a 5 4 2"13 5 4 0 0 0 S45 S55 0 5 4 a 5 4 0 5
A 2"12 S16 S26 S36 0 0 S66 A V6
m2 n2
S44 D C ; (16.32)
t a
16 Predicting Properties of Undamaged and Damaged Carbon Fibre. . . 433
1 1
S45 D mn ; (16.33)
a t
m2 n2
S55 D C ; (16.34)
a t
2 2
2 2 1 1 2A m n2
S66 D 4m n C C C ; (16.35)
EA ET EA A
V1 D m2 ˛A C n2 ˛T ;
V2 D m2 ˛T C n2 ˛A ;
(16.36)
V3 D ˛t ; V4 D 0; V5 D 0;
V6 D 2mn .˛A ˛T / :
The inverted form for the stress/strain relations of an angled ply is given by
2 3 2 3 2 3 2 3 2 3
A 11 C11 C12 C13 0 0 C16 "A U1
6 7 6 7 6C C23 0 0 C26 7 6" 7 6U 7
6 T7 6 22 7 6 12 C22 7 6 T7 6 27
6 7 6 7 6 7 6 7 6 7
6 t 7 6 33 7 6 C13 C23 C33 0 0 C36 7 6 " t 7 6 U3 7
6 7 6 7D6 7 6 7 6 7 T;
6 t 7 6 23 7 6 0 0 0 C44 C45 07 6 t 7 6 0 7
6 7 6 7 6 7 6 7 6 7
4 a 5 4 13 5 4 0 0 0 C45 C55 05 4 a 5 4 0 5
A 12 C16 C26 C36 0 0 C66 A U6
(16.37)
where
4 4 2 2
C11 Dm E A C n E T C 2m n A ET C 2A ;
C12 D m4 C n4 A ET C m2 n2 EA C ET 4A ;
C13 D m2 a C n2 t Et ; C16 D mn m2 EA n2 ET m2 n2 A E T C 2A ;
C22 D m4 ET C n4 EA C 2m2 n2 A ET C 2A ; C23 Dm2 t C n2 a Et ;
C26 D mn n2 EA m2 ET C m2 n2 A ET C 2A ; C33 D Et ;
C36 D mn . a t / Et ; C44 D m2 t C n2 a ; C45 D mn .a t / ;
2
C55 D m2 a C n2 t ; C66 D m2 n2 EA C ET 2 A ET C m2 n2 A ;
(16.38)
and
U 1 D m2 E A ˛ A C n 2 E T ˛ T ;
U2 D n2 EA ˛ A C m2 ET˛ T ;
(16.39)
U3 D Et ˛ t ; U6 D mn EA ˛ A ET ˛ T :
434 L.N. McCartney
In the relations (16.38) and (16.39) the following quantities are used:
EA D EƒA 1t2 EETt ; ET D EƒT 1a2 EEAt ; Et D Eƒt 1 A2 EETA ;
A ET D EƒT A Ca t EETt ; a Et D Eƒt .a Ct A / ; t Et D Eƒt t Ca A EETA ;
(16.40)
EA ˛ A D EA ˛A C A ET ˛T C a Et ˛t ;
ET ˛ T D A ET ˛A C ET ˛T C t Et ˛t ; (16.41)
Et ˛ t D a Et ˛A C t Et ˛T C Et ˛t ;
Et Et ET Et
ƒ D 1 a2 t2 A2 2a t A : (16.42)
EA ET EA EA
The concept of shear coupling is now introduced where the stress/strain relations
(16.22) are written
.P/ .P/ .P/
1 A a A .P/
"A D .P/ A .P/ T .P/ t .P/ A C ˛A T;
EA EA EA A
.P/ .P/ .P/
A 1 t .P/ T
"T D .P/ A C .P/ T t .P/ A C ˛T .P/ T;
EA ET ET A
.P/ .P/ .P/ (16.43)
.P/
"t D a.P/ A t.P/ T C 1.P/ t A C ˛t t
.P/ T;
EA ET Et A
.P/ .P/ .P/
.P/
A D A.P/ A T.P/ T .P/ t
t C 1.P/ A C ˛S T;
A A A A
where the superscript (P) denotes that the thermo-elastic constant refers to a specific
angled ply. By comparing (16.22) and (16.43) it is clear that
.P/ .P/ .P/
A A .P/
S11 D 1
.P/ ; S12 D .P/ ; S13 D a.P/ ; S16 D .P/ ; V1 D ˛A ;
EA EA EA A
.P/ .P/
T .P/
S22 D 1
.P/ ; S23 D t.P/ ; S26 D .P/ ; V2 D ˛T ;
ET ET A (16.44)
.P/
1 t .P/
S33 D .P/ ; S36 D .P/ ; V3 D ˛t ;
Et A
1 .P/
S66 D .P/ ; V6 D ˛S :
A
Consider now a laminate that is made by perfectly bonding together various laminae
(i.e. plies) having different orientations so that there are no defects (i.e. the laminate
is undamaged). The situation under consideration concerns the deformation of a
symmetric multi-layered laminate of total thickness 2h constructed of 2n perfectly
bonded plies that can have any combination of orientations, such that symmetry
about the mid-plane of the laminate is preserved. The plies in each half of the
laminate can be made of different materials and each can have a different thickness.
As laminate symmetry is assumed, it is necessary to consider only a set of n layers.
A global right-handed set of Cartesian coordinates is chosen having the origin at
the centre of the mid-plane of the laminate. The locations of the n 1 interfaces in
.i/
one half of the laminate (x3 > 0) are specified by x3 D x3 , i D 1, : : : , n 1. The
.0/
mid-plane of the laminate is specified by x3 D x3 D 0 and the external surface by
.n/
x3 D x3 D h where h is the half-thickness of the laminate. The thickness of the ith
.i/ .i1/
layer is denoted by hi D x3 x3 such that
X
n
hD h i: (16.45)
iD1
Stress, strain and displacement components, and material properties associated with
the ith layer are denoted by a superscript (i) . The orientation of the ith layer is
specified by the angle i defined in Sect. 16.3.1 and illustrated in Fig. 16.1.
It is now required to specify the effective laminate properties S and V in terms
of the Young’s and in-plane shear moduli, Poisson’s ratios and thermal expansion
coefficients of the laminate. This is achieved by modifying the relations (16.22)
derived for a single angled lamina so that they apply to a laminate rather than to
an individual angled ply. When the laminate is considered as a homogenised plate,
the effective stress/strain relations must be of the following form analogous to the
corresponding single ply relations:
2 3 2 3 2 .L/ .L/ .L/ .L/ 3 2 3 2 .L/ 3
"A "11 S11 S12 S13 0 0 S16 A V1
6" 7 6 " 7 6 S.L/ .L/ .L/
S23 0 0 S26 7
.L/ 6 7 6 V .L/ 7
6 T7 6 22 7 6 12 S22 7 6 T7 6 2 7
6 7 6 7 6 .L/ .L/ .L/ .L/ 7 6 7 6 .L/ 7
6 "t 7 6 "33 7 6 S13 S23 S33 0 0 S36 7 6 t 7 6 V3 7
6 7 6 7D6 .L/ .L/ 7 6 7C6 7 T;
6 t 7 6 2"23 7 6 0 0 0 S44 S45 07 6 t 7 6 0 7
6 7 6 7 6 7 6 7 6 7
4 a 5 4 2"13 5 4 0 0
.L/ .L/
0 S45 S55 05 4 a 5 4 0 5
.L/ .L/ .L/ .L/ .L/
A 2"12 S S26 S36 0 0 S66 A V6
16
(16.46)
where the superscript (L) is used to denote effective thermo-elastic constants of the
laminate, and where
436 L.N. McCartney
where the various laminate properties will be specified below. By applying the
following expressions for effective stresses and strains:
1X 1X 1X 1X
n n n n
.i/ .i/ .i/ .i/
A D h i A ; T D h i T ; A D h i A ; " t D h i "t ;
h iD1 h iD1 h iD1 h iD1
(16.48)
it can be shown that the laminate properties appearing in (16.47) are calculated using
the following relations:
.L/ 2 .L/ 2
1 1 A 1 1 T
.L/ D Q A.L/
C .L/ ; .L/ D Q T.L/
C .L/ ;
EA E A ET E A
.L/
2 .L/ .L/
1 1 t .L/ .L/ Qt
.L/
t T
.L/ D Q t.L/
C .L/ ; t D ET Q T.L/
.L/ ;
A
Et E
.L/ .L/ .L/
E
A .L/ .L/
QA A T .L/ t A
.L/ .L/ .L/ .L/ Qa
A D E A Q A.L/
.L/ ; a D E A Q A.L/
.L/ ;
E A E A
.L/ .L/ .L/ .L/ .L/ .L/ .L/ .L/ .L/ .L/ .L/ .L/
˛A D ˛Q A A ˛S ; ˛T D ˛Q T T ˛S ; ˛t D ˛Q t t ˛S ;
(16.49)
where
.L/ 2 .L/ .L/ 2
.L/ .L/ 12 .L/ 12 .L/ .L/ 12
EQ A D 11 .L/ ; Q A D .L/ ; EQ T D 22 .L/ ;
22 22 11
.L/ .L/ .L/ .L/ .L/ .L/ Q .L/ .L/
.L/ E
Q a D 13 Q A 23 ; Q t D 23 Q A Q T.L/ 13 ; (16.50)
EA
2 2
.L/ .L/ .L/ .L/
1 Qa .L/ Qa Qt .L/ Qt .L/ .L/
Q t.L/
D Q .L/ 11 C 2 Q .L/ .L/ 12 C
Q Q .L/ 22 C 33 ;
E E A E E E A T T
.L/ .L/
.L/ 1 .L/ Q A .L/ .L/ 1 .L/ Q A .L/
A D
.L/ 16
;
.L/ 26
T D
.L/ 26
;
.L/ 16
(16.51)
EQ A EQ A EQ T EQ A
.L/ .L/ .L/ .L/ .L/ .L/
t D 36 13 A 23 T ; (16.52)
16 Predicting Properties of Undamaged and Damaged Carbon Fibre. . . 437
2 2
.L/ .L/ .L/ .L/ .L/ .L/ .L/ .L/ .L/
A D 66 11 A 212 A T 22 T : (16.53)
.L/ .L/
.L/ 1 .L/ Q A .L/ .L/ 1 .L/ Q A .L/
˛Q A D !
.L/ 1
! ;
.L/ 2
˛Q T D !
.L/ 2
! ;
.L/ 1
(16.54)
EQ A EQ A EQ T EQ A
.L/ .L/ .L/ .L/ .L/ .L/ .L/ 1 .L/ .L/ .L/ .L/ .L/
˛Q t D !3 13 ˛Q A 23 ˛Q T ; ˛S D .L/
!6 16 ˛Q A 26 ˛Q T :
A
(16.55)
.L/
X
n
.i/ .L/
X
n
.i/ .L/
X
n
.i/ .L/
X
n
.i/
11 D 1h h i 11 ; 12 D 1h h i 12 ; 22 D 1h h i 22 ; 33 D 1h h i 33 ;
iD1 iD1 iD1 iD1
.L/
X
n
.i/ .L/
X
n
.i/ .L/
Xn
.i/
16 D 1h h i 16 ; 26 D 1h h i 26 ; 66 D 1h hi 66 ;
iD1 iD1 iD1
(16.56)
.L/ 1 X
n
.i/ .L/ 1 X
n
.i/ .L/ 1 X
n
.i/
13 D h i 13 ; 23 D h i 23 ; 36 D h i 36 ; (16.57)
h iD1 h iD1 h iD1
1X X X X
n n n n
.L/ .i/ .L/ 1 .i/ .L/ 1 .i/ .L/ 1 .i/
!1 D h i !1 ; ! 2 D h i !2 ; ! 3 D h i !3 ; ! 6 D h i !6 :
h h h h
iD1 iD1 iD1 iD1
(16.58)
where
.i/ Q A.i/
E .i/
.i/ .i/
QA E QT .i/ b
a
.i/
.i/ EQ A.i/ .i/ .i/ .i/
QA EQ T .i/
11 D ‰ .i/
; 12 D ‰ .i/
; 13 D ‰ .i/ ; 16 D ‰ .i/ A C ‰ .i/
T ;
.i/ Q T.i/
E .i/ b
.i/
t .i/
.i/ Q .i/
QA E .i/
.i/
EQ T .i/
22 D ‰ .i/
; 23 D ‰ .i/
; 26 D ‰ .i/
T
A C ‰.i/ T ;
.i/ .i/ .i/
.i/ ƒ .i/ b
a .i/ b
t .i/ .i/
33 D Q t.i/
; 36 D ‰ .i/ A
C‰ .i/ T C t ;
‰ .i/ E
Q A.i/ .i/ .i/ Q .i/ .i/ .i/
b C ‰T.i/ T b
.i/ E E .i/
66 D ‰ .i/ A A
T C A ;
(16.59)
QA .i/ .i/ .i/
QT QT .i/
.i/ E .i/ QA E .i/ .i/ E .i/ .i/ .i/
!1 D ˛Q
‰ .i/ A
C ‰ .i/
˛Q T ; !2 D ‰ .i/
˛Q T C Q A ˛Q A ;
.i/ .i/ Q .i/ .i/ .i/ QT.i/
!6 D ‰A.i/ b b
.i/ b
a .i/ b
t .i/ .i/ .i/ E E .i/ .i/ .i/ .i/
!3 D ˛Q
‰ .i/ A
C ˛Q
‰ .i/ T
C ˛Q t ; A ˛Q A C ˛Q
‰ .i/ T T
C A ˛S ;
(16.60)
438 L.N. McCartney
and where
2 EQ .i/ Q .i/
.i/ .i/ .i/ .i/ .i/ .i/ ET
‰ .i/ D 1 Q A T
; a.i/ D Q a.i/ C Q t Q A ;
b t D Q t C Q a.i/ Q A
b
QEA.i/ .i/
EQ A
(16.61)
.i/ .i/
.i/ Q .i/ .i/ Q .i/
ƒ .i/ D ‰ .i/ Q a EQ t.i/ b
a Q t EQ t.i/ b
t ;
2 .i/EA 2 E.i/T 2 Q .i/ .i/ (16.62)
.i/ Qt
E .i/ EQt .i/ ET .i/ .i/ .i/ Q
D 1 Q a Q .i/
Q t Q .i/
Q A Q .i/
2Q a Q t Q A EQ t.i/ ;
EA ET EA EA
Q .i/
.i/ .i/ ET
b.i/ .i/
A D A C T Q A .i/ ; b.i/ .i/ .i/ .i/
T D T C A Q A : (16.63)
EQ A
The effective properties characterising the out-of-plane shear of the laminate are
given by
!
.L/
X
n
m2i n2i
1
S44 D h hi .i/
C D 1.L/ ;
.i/
t t
a
iD1 !
.L/
Xn
1 1
S45 D 1h h i mi n i .i/
.i/ Dˆ.L/ ; (16.64)
iD1 a t
!
.L/
Xn
m 2
n 2
1
S55 D 1h hi i
.i/
C .i/ i
D .L/ ;
iD1 a t a
where
! ! !
1 X 1 X 1 X
n n n
.L/ .i/ .L/ .i/ .L/ .i/
S44 D h i S44 ; S45 D h i S45 ; S55 D h i S55 :
h iD1 h iD1 h iD1
(16.65)
It useful to use the set of carbon fibre and epoxy ply properties given in Table 16.2 to
calculate values for the effective thermo-elastic constants of a [0/45/-45/90]s sym-
metric laminate having plies of equal thickness. The predicted laminate properties
are shown in Table 16.3.
It should be noted that the through-thickness thermal expansion coefficient is
more than ten times larger than the in-plane values.
To conclude this section it is useful to emphasise the nature of the analytical
results given above that apply to the undamaged unidirectional and laminated com-
posites. The derivations of formulae for the effective thermal expansion coefficients
have taken full account of the effects on deformation of thermal residual stresses. It
has been shown how to use the various analytical formulae to predict the effective
properties of symmetric laminates (possibly unbalanced) from the properties of the
fibre and matrix, a specific example being given. It is emphasised that the explicit
nature of the results for laminates enables the identification of the dependencies of
the many effective laminate properties on the properties of the individual plies. The
remainder of this chapter will focus on some important damage modes that can be
investigated using analytical methods.
For unidirectional polymer composites having very strong perfectly bonded inter-
faces, fibre fracture is perhaps the most important damage mode, as it can be the
precursor to progressive damage growth that leads on to the catastrophic failure
440 L.N. McCartney
only for the special case where the fibre/matrix interface remains perfectly bonded.
He applied the model to the analysis of single-fibre fibre-pull-out tests and micro-
drop debond tests.
McCartney [16] has recently explored the use of the concentric cylinder model
for a unidirectional composite to determine the stress distributions based on
Nairn’s [15] approach and corresponding displacement distributions, for a variety
of states of the fibre/matrix interface. For our purposes in this chapter, relations are
required describing the normal traction distribution and crack opening displacement
distribution of either an array of broken fibres all fractured in the same plane, or
of a matrix crack that leaves all fibres intact. The normal traction distribution is
calculated from the axial stresses in the fibre and matrix given by
Vf p
zzm .z/ D m C C.z/; R < r < a D R= Vf : (16.67)
Vm
where
Z z
C .z/ D C z0 dz0 ; (16.70)
0
and where the average of any quantities f (r, z) and m(r, z) associated with the fibre
and matrix, respectively, are defined by
Z Z
1 R
1 a
f .z/ D 2 rf .r; z/ dr; m.z/ D 2 rm .r; z/ dr: (16.71)
R2 0 .a R2 /
2
R
In the relations (16.66)–(16.70), the parameters f and m are the uniform axial fibre
and matrix stresses and "A is the uniform axial strain for an undamaged composite.
The function C(z) is a stress transfer function that must satisfy the following fourth
order ordinary differential equation:
000 0
FR4 C0 .z/ C G R2 C0 .z/ C H C.z/ D 0; (16.72)
442 L.N. McCartney
where the constant coefficients F, G and H are specified [16] in terms of fibre and
matrix properties, and volume fraction, as are the parameters ˛, ˇ and that appear
in the relations (16.68) and (16.69). The boundary conditions are given by
.1C/ A
rzf .r; ˙L/ D 0; zzf .r; ˙L/ D 2 Vf ; 0 r R;
.1/ A (16.73)
rzm .r; ˙L/ D 0; zzm .r; ˙L/ D 2 Vm
; R r a;
where on using (16.66) and (16.67) the effective applied stress A is given by
On setting D 1, the boundary conditions (16.73) are valid for fibre fractures, and
on setting D 1 these boundary conditions are valid for matrix cracks. It has been
shown [16] how to solve the differential equation (16.72) subject to the boundary
conditions (16.73).
f
Em C 2m Vf m A
1 4Rp Vm f Vm
D 2 1 2ˇA ; (16.76)
ƒ p q2 EAf Vf Vf EA
where the parameters p and q are defined in terms of the coefficients F, G and H of
the differential equation (16.72) by
r r r
1 1 G H
pD .r C s/; qD .r s/ < p; rD > 0; sD ; (16.77)
2 2 2F F
and where
!
1 1 1 Vf Vm m
D C C f ; where km D : (16.78)
2 m km kT 1 2m
If the laminate is isotropic, the dimensionless parameter has the value 1 for
conditions of generalised plane stress. On defining
x 4ƒ c p .x; c/ th
XD ; D ; P .X; / D ; (16.80)
c EA A th
it has been shown [18] for embedded matrix cracks that it is necessary to solve the
following linear integral equation for the function P(X, ):
Z 1
P .X; / D Œ1 P .t; /k .t; X/ dt; 0 X 1; (16.81)
0
where
ˇp p ˇ
1 ˇ 1 t2 C 1 X 2 ˇ
ˇ ˇ
k .t; X/ D ln ˇ p p ˇ D k .X; t/ : (16.82)
ˇ 1 t2 1 X 2 ˇ
The corresponding expression for the mode I stress intensity factor K may be written
Z 1
p 2 1 P .X; /
K D c .A th / Y ./ where Y ./ D p dX:
0 1 X2
(16.83)
444 L.N. McCartney
The integral equation (16.81) can be solved using numerical methods to find the
fibre stress distribution P(X, ). The value of Amc th for which matrix crack growth
is initiated is found using (16.83) by imposing the following fracture criterion
derived from energy balance considerations [18]:
2EA
K2 D ; (16.84)
where 2 is the mode I effective fracture energy for matrix cracking for the
homogenised material. For matrix cracks in a unidirectional composite where
the fibres are normal to the crack surfaces it follows that Vm m where 2 m
is the fracture energy of a homogeneous sample of matrix material. At the point of
the initiation of crack propagation it follows from (16.80), (16.83) and (16.84) that
Amc th 1
Dp ; (16.85)
0 Y ./
where
8ƒ
02 D : (16.86)
EA
c0 D ; (16.87)
4ƒ
so that D c/c0 . The fracture condition (16.85) may then be expressed in the
dimensionless form
Amc th 1
Dp : (16.88)
0 c=c0 Y .c=c0 /
3
Normalised matrix or ply cracking stress
Cracking stress
Long crack limit
Griffith crack
0
0 5 10 15 20 25 30 35 40
Normalised crack length
Fig. 16.2 Dependence of the critical axial applied stress for matrix or ply cracking on
the length
of the pre-existing matrix crack. The normalised matrix cracking stress is Amc th =0 and the
normalised crack length is c/c0
independent of the length 2c of the matrix crack. The asymptotic value defined by
(16.89) is shown in Fig. 16.2 as a broken line and is described as the long crack
limit. On using (16.86) and (16.87)
2 EA
c0 02 D D KIc2 ; (16.90)
where KIc is regarded as the mode I effective fracture toughness of the homogenised
material.
In the long crack limit the stress intensity factor may be expressed in the
following dimensionless form independent of crack length:
r
K 2 A th
D : (16.91)
KIc 0
446 L.N. McCartney
For cracks of any length, the corresponding expression for the normalised stress
intensity factor is given by
K A th p c
D Y ./ ; where D : (16.92)
KIc 0 c0
On using (16.90) the well-known Griffith fracture criterion K D KIc leads to the
result
r
Amc th c0
D : (16.94)
0 c
This relationship is plotted in Fig. 16.2 using the label ‘Griffith crack’. As to be
expected it approaches the bridged crack solution when c=c0 ! 0 but reduces to a
zero value as c=c0 ! 1.
To complete the discussion it is worth mentioning that, in the absence of an
accurate numerical solution of the bridged crack problem, use can be made of the
long crack limit and the prediction for a Griffith crack to construct an approximate
approach. With reference to Fig. 16.2, the long ply crack limiting value given by
(16.89) is used in the Griffith relation (16.94) to determine the crack length c1 ,
denoting the point at which the long crack asymptote crosses the Griffith curve, that
is given by
c1 2
D D 0:6366: (16.95)
c0
Thus the approximate estimate of the matrix cracking stress is obtained using
(16.94) when c < c1 and using (16.89) when c > c1 . It is noted from Fig. 16.2
that the approximation always underestimates the ply cracking stress thus providing
a relatively simple conservative estimate that could be used when designing
composites that must be resistant to matrix cracking damage.
For some typical materials systems, it is useful to provide estimates in Table 16.4
of the normalising parameters 0 , c0 that have been used when generating the master
1
curve in Fig. 16.2, and the value of the long crack limit mc and the value of c1 .
The results for carbon-epoxy have assumed that the fibre and matrix property values
are given in Table 16.1. Poisson’s ratios are, however, assumed to be zero. The
predicted matrix cracking stresses that are seen to be very large, clearly indicating
that other damage modes, such as fibre failure, are inevitable before the matrix is
16 Predicting Properties of Undamaged and Damaged Carbon Fibre. . . 447
able to crack. Also shown (in brackets) for this material are predictions using the
more accurate expression for the axial modulus EA and the non-zero Poisson ratios
for fibre and matrix which have been calculated using the properties or values given
in Table 16.1.
It will become clear in Sect. 16.9 that the analysis presented in this section is
almost identical to that which arises when considering embedded ply cracks in
cross-ply laminates.
Having considered a linear crack bridging law that arises when the fibre/matrix
interfaces are always perfectly bonded, it is now appropriate to consider a non-
linear law that arises when matrix cracking occurs in the presence of fibre/matrix
debonding. Aveston et al. [10, 19] developed the very well-known ACK theory for
the study of matrix cracking phenomena. Their model applies to the consideration
of very long matrix cracks where matrix cracking is considered in the absence of the
effect of thermal residual stresses. This model was the basis of methods developed to
investigate crack bridging for much shorter cracks [20, 21], and for the case where
thermal stresses are present [17, 18]. A summary of the key equations, extended
here to include the effects of thermal stresses, and results will be presented here.
It is useful to discuss crack bridging for unidirectional composites by referring to
Fig. 16.3. In Fig. 16.3a is shown a unidirectional composite having a single matrix
crack with frictionally slipping fibre/matrix interfaces and subject to a uniform
uniaxial applied stress A . Figure 16.3b shows the corresponding representative
volume element for a single fibre. The simplified analysis described above indicates
that stress transfer between fibre and matrix is restricted to the regions A which are
adjacent to the matrix crack. For the other regions B stress transfer is negligible
so that the local stress and displacement fields in the fibre and matrix correspond
to those found in an undamaged composite made of the same material and subject
to the same applied stress and temperature. In regions A the stress carried by the
matrix is transferred to the fibres by action of the shear stresses at the fibre/matrix
interfaces.
448 L.N. McCartney
Fig. 16.3 Schematic diagram of a unidirectional composite containing a matrix crack for which
interfacial frictional slip has occurred: (a) composite as a whole, (b) discrete model for a single
fibre and (c) corresponding continuum model
The top half of the composite shown in Fig. 16.3 is now considered where the
undeformed separation of the lines PQ and RS is denoted by 2L. A z-coordinate is
now introduced where the origin lies on the line PQ and the location of the cracked
plane before deformation is on z D L. The debonded interface is assumed to have
length L and exhibit a uniform interfacial shear stress . It has formed because of
the presence of a matrix crack on the plane z D L that is normal to the fibre direction.
The axial stresses in the fibre and matrix are
2
zzf .z/ D f C z; 0 < z < L; (16.96)
R
2 Vf
zzm .z/ D m z; 0 < z < L; (16.97)
R Vm
where f and m are the axial stresses in fibre and matrix, respectively, for an
undamaged composite. For a matrix crack, the normal traction in the matrix on
the crack surface is zero so that zzm .L/ D 0 and it then follows from (16.97) that the
debond length L is given by
R Vm
LD m : (16.98)
2 Vf
16 Predicting Properties of Undamaged and Damaged Carbon Fibre. . . 449
For an undamaged composite having a uniaxial strain "A the fibre and matrix stresses
are
f f
f D EA "A ˛A T ; m D Em "A ˛Am T : (16.99)
A D Vf f C Vm m : (16.102)
On using (16.99)
A D EA ."A ˛A T/ ; (16.103)
where
f f f
EA D Vf EA C Vm Em ; EA ˛A D Vf EA ˛A C Vm Em ˛Am : (16.104)
The parameters EA and ˛ A are mixtures estimates of the effective axial Young’s
modulus and effective axial thermal expansion coefficient of the composite.
It follows from (16.96) that the maximum value of the fibre stress occurs when
z D L and its value is given by
2 Vm A
fmax D f C L D f C m D : (16.105)
R Vf Vf
It can be shown that the matrix crack opening displacement has the value 2mc
where
R Vm Em f
f
2
mc D 2
A C Vf EA ˛A ˛Am T : (16.106)
4 Vf EA EA
f
When developing the equivalent continuum model of a bridged crack problem, use
is made of the relation (see [17, 18] for details)
Vm Em
D mc ; (16.107)
EA
450 L.N. McCartney
indicating that the continuum crack opening displacement 2 differs from the value
2mc obtained using the discrete model. On using (16.106) the crack bridging
relation for the continuum model of a matrix crack is written
1
vD .p th /2 ; (16.108)
ƒ
where
f 2
2EA Vf EA f f
v 2 ; ƒD ; th D Vf EA ˛Am ˛A T: (16.109)
R Vm Em
It is assumed that all fibres remain intact during loading of the composite, so that
the discontinuous tractions associated with the discrete fibre reinforcement can be
approximated by a continuous distribution p(x, c) such that
On defining
it can be shown [21] that it is necessary to solve the following non-linear integral
equation for the function P(X, ):
p Z 1
.P .X; //2 D 1 X2 P .t; /k .t; X/ dt ; 0 < X < 1: (16.112)
0
where
ˇp p ˇ
1 ˇ 1 t2 C 1 X 2 ˇ
ˇ ˇ
k .t; X/ D ln ˇ p p ˇ D k .X; t/ : (16.113)
ˇ 1 t2 1 X 2 ˇ
The integral equation (16.112) can be solved using numerical methods to calculate
the distribution P(X, ) and the function Y(). On using (16.114)1 to eliminate c or
A th it follows that
16 Predicting Properties of Undamaged and Damaged Carbon Fibre. . . 451
r
EA p
K D .A th /3=2 Y ./ ; (16.115)
4ƒ
or
p ƒ 3=2 Y ./
K D 4 c : (16.116)
EA
The value of Amc th for which crack growth is initiated is found by imposing the
following fracture criterion derived from energy balance considerations:
2EA
K2 D ; (16.117)
where 2 is again the mode I effective fracture energy for matrix cracking for the
homogenised material such that Vm m where 2 m is the fracture energy of
a homogeneous sample of matrix material. At the point of the initiation of crack
propagation, such that (16.117) is satisfied A D Amc and D c , it follows from
(16.115) and (16.117) that
1=3
Amc th 1
D ; (16.118)
0 c Y 2 .c /
where
1=3
8ƒ
0 D : (16.119)
From (16.116) it follows at the point of crack growth initiation, for which c D cmc ,
2=3
cmc c
D ; (16.120)
c0 Y .c /
where
EA 1=3
c0 D : (16.121)
8 ƒ2
It is of interest to note that on setting K D KIc it follows from (16.117), (16.119)
and (16.121) that
2EA
c0 02 D D KIc2 ; (16.122)
where KIc is regarded as the effective fracture toughness of the homogenised
material. By solving the integral equation (16.112) for a range of values of and
452 L.N. McCartney
Bridged crack
Long crack limit
4
Normalised matrix cracking stress
Griffith crack
0
0 5 10 15 20 25 30 35 40
Normalised crack length
Fig. 16.4 Dependence of the critical axial applied stress for matrix cracking
on the length of
the pre-existing matrix crack. The normalised matrix cracking stress is Amc th =0 and the
normalised crack length is c/c0
1 1=3
mc th 3
D D 1:33067: (16.123)
0 4
When the thermal stress th is zero this result corresponds exactly to the well-known
relation of ACK theory derived by Aveston et al. [10].
It is useful to consider the behaviour of matrix cracking when ! 0
corresponding to matrix cracks of decreasing length. For this special case it follows
from (16.112) that P .X; / ! 0 and from (16.114)2 that Y ./ ! 1 leading to the
following expression for the stress intensity factor for a stress-free embedded crack
in an infinite plate:
p
K D c .A th / : (16.125)
This relationship is shown in Fig. 16.4 using the label ‘Griffith crack’. As to be
expected it approaches the bridged crack solution when c=c0 ! 0 but reduces to a
zero value as c=c0 ! 1.
It is emphasised that the bridged crack solution shown in Fig. 16.4 is a master
curve that applies to all materials whose deformation is governed by the simplified
model. The effects of all material constants, including fracture parameters, are
included in the normalising constants 0 and c0 defined by (16.119) and (16.121). If
an alternative stress transfer model is used, such as that described earlier for the case
of perfect fibre/matrix bonding, then a different crack bridging curve to that shown
in Fig. 16.4 would result (see Fig. 16.2). Such a curve is easily generated using very
similar numerical methods.
To complete the discussion of the simplified model it is also worth mentioning
that, in the absence of an accurate numerical solution of the bridged crack problem,
use can be made of the long crack limit and the prediction for a Griffith crack to
construct an approximate approach. With reference to Fig. 16.4, the long crack
limiting value given by (16.123) is used in the Griffith relation (16.126) to determine
the crack length c2 , denoting the point at which the long crack asymptote crosses
the Griffith curve, that is given by
2
/3
c2 4
D D 0:5648: (16.127)
c0 3
Thus the approximate estimate of the matrix cracking stress is obtained using
(16.126) when c < c2 and using (16.123) when c > c2 . It is noted from
Fig. 16.4 that the approximation always underestimates the matrix cracking stress
thus providing a relatively simple conservative estimate that could be used when
designing composites that must be resistant to matrix damage.
454 L.N. McCartney
1
Table 16.5 Example predictions of values of 0 , c0 , c1 and mc for
debonded interfaces
Property Carbon-cement Carbon-glass Carbon-epoxy
f
EA (GPa) 214 380 225
Em (GPa) 19 70 4.2
Vf 0.12 0.4 0.6
Km 1/2
Ic (MPa . m ) 0.39 0.75 0.6
R (m) – 4 4
(MPa) 5.2 105 R 10 10
0 (MPa) 56.966 341.47 4891.3 (4891.8)
c0 (m) 29.299 2.5534 0.06235 (0.06234)
c2 (m) 16.548 1.442 0.03521
1 .MPa/
mc 75.803 454.38 6508.7 (6509.4)
The first damage mode encountered when laminates are subject to multi-axial
loading is the formation of ply cracks in plies whose fibre direction is transverse
to the principal loading direction. Ply cracks usually form suddenly and completely
crack the ply cross-section as described in the pioneering work of Parvizi et al.
[23]. Increasing loads lead to the progressive formation of ply crack arrays. When
a ply crack forms in an undamaged laminate the load it carried is transferred to
neighbouring plies which are intact as the fibres usually have a different direction
to those in the cracked ply. The stress transfer occurs through the action of shear
tractions on the interfaces between the cracked ply and its neighbours. The non-
uniform stress field that results has a significant effect on the values of the effective
thermo-elastic constants of the laminate.
Much of the analytical modelling work on the prediction of ply crack formation
in laminates has focused on [0m /90n ]s cross-ply laminates because of the relative
simplicity of the geometry. Again because of relative simplicity, much modelling
work has made use of shear-lag models, extending the approach developed for fibres
in unidirectional composites to laminates. While there are a variety of shear-lag
models in the literature, the best approach, as used by McCartney [24] and confirmed
by Nairn and Mendels [25], is the method first developed by Nuissmer and Tan
[26]. It should be noted that a shear-lag model for cracks in layered materials seems
to have been derived independently by Hobbs [27] when considering geological
applications (see also more recent work by Bai and Pollard [28]).
Hashin [29, 30] developed a variational method for solving ply cracking prob-
lems for [0m /90n ]s cross-ply laminates which is valid only for plane stress conditions
and for cases where thermal residual stresses are negligible. However, Hashin [31]
later considered the thermal expansion of cracked laminates. A variational analysis
of cross-ply laminates has also been provided by Nairn [32] and Nairn and Hu [33].
McCartney [34] developed an analytical model that applied to the generalised plane
strain conditions normally encountered when ply cracking occurs, and included
the effects of thermal residual stresses. It has been shown [35] that the solution
of the generalised plane strain model provides both the stress and displacement
distributions that minimise the Reissner [14] energy functional when using a
variational method. Very recently, this approach has been significantly extended [36]
to consider more general loading conditions, and to compare predictions of effective
thermo-mechanical properties with those calculated using an optimal shear-lag
approximation, thus confirming the limitations of shear-lag models. It should be
noted that both analytical methods [35] and variational methods [37] of solving ply
crack problems for cross-ply laminates have been developed for cases where the
laminate is subjected to combined in-plane and loading out-of-plane bending.
Ply crack formation in quasi-isotropic laminates has been studied by McCartney
[38] using analytical methods and by Crocker et al. [39] and Tong et al. [40] using
finite element analysis of cracked laminates for the case of generalised plane strain
456 L.N. McCartney
D A ; D 0A : (16.128)
McCartney [38, 42, 43] has considered more general laminates where shear coupling
terms are included in the analysis. For the special case when in-plane shear
deformation is absent, the effective stress/strain relations of the damaged laminate
may be expressed in the form
t a t
"t D A T C ˛t T; (16.129)
Et EA ET
16 Predicting Properties of Undamaged and Damaged Carbon Fibre. . . 457
Fig. 16.5 Schematic diagram of a 0o /90o /0o laminate containing a ply crack: (a) laminate as a
whole, (b) discrete model and (c) corresponding continuum model
a A A
"A D t C T C ˛A T; (16.130)
EA EA EA
t A T
"T D t A C C ˛T T : (16.131)
ET EA ET
t0 t A0 A 1 1
ET0
ET EA0
EA ET
ET0 ˛T ˛T0
D 1 1
D D D k; (16.133)
a0
a
EA EA0
A0
A ˛A ˛A0
EA0 EA EA0 EA
1 1 a0 a t0 t
EA0
ET0
˛t ˛t0
Et Et0 EA ET
D 1 1
D D D k0 ; (16.134)
a0
a
EA EA0
A0
A ˛A ˛A0
EA0 EA EA0 EA
458 L.N. McCartney
where
0 T E0 E0
EA0 A a0 t0 EA0 B
EA0 ˛A0 C B˛T0 C EA0 B A EA0
k1 D ; kD 0 ; k0 D T
;
1 A0 B ET 1 A0 B 1 A0 B
(16.135)
and where A, B and C are constants calculated from unidirectional undamaged ply
properties (having a superscript ‘eff’ for effective property) as follows:
The parameters k1 , k and k0 are laminate constants that are independent of the state
of damage. If the undamaged ply properties can be assumed transverse isotropic,
then use can be made of the results given in Sect. 16.2 so that the constants A, B and
C can be calculated from fibre and matrix properties and volume fractions.
where all the effective thermo-elastic constants of a damaged laminate can be cal-
culated in terms of the single damage dependent parameter ˆ that characterises the
change in the values of all thermo-elastic constants resulting from the presence of
ply cracking damage. They enable an exact and simple means of fully characterising
the effects of microstructural damage on the effective thermo-elastic constants of
laminates, requiring only the value of the parameter ˆ which depends upon the
axial modulus, a property that is frequently measured in practice or can be estimated
from a suitable model. It is emphasised that the relations (16.137) can be derived
and used without having to solve for the stress and displacement distributions in a
cracked laminate. Such solutions are needed only to derive values for the parameter
ˆ that determines the values of the effective thermo-elastic constants. Nairn [9]
has reached a similar conclusion for a limited set of elastic constants in the case
of cross-ply laminates. The results are in fact valid for any damage in laminates
16 Predicting Properties of Undamaged and Damaged Carbon Fibre. . . 459
Consider a multiple-ply laminate that already contains ply cracks in some (but
not necessarily all) 90o plies. This situation is to be known as damage state 1.
Then consider the simultaneous formation of new ply cracks in the 90o plies
leading to damage state 2. The locations of the ply cracks are assumed to be
such that the overall deformation of the laminate is governed by the effective
stress/strain equations (16.129)–(16.131). This means that the crack distribution
is regarded as being sufficiently uniformly distributed for shear and out-of-plane
bending deformations to be negligible. The new crack surfaces are assumed to
form quasi-statically under conditions of fixed effective applied tractions and fixed
temperature. From energy balance considerations and the fact that kinetic energy
is never negative, the criterion for crack formation under these conditions has the
form [45]
C G < 0; (16.138)
where the energy absorbed in volume V of laminate by the formation of the new
cracks is given by
D. 2 1 / V: (16.139)
In (16.139) the parameter i denotes the energy absorbed in unit volume of laminate
during the formation of new crack surfaces in the 90o plies that have led to the
damage state i. The energy absorbed is taken as the area of crack surface formed
multiplied by the effective fracture energy for transverse cracking in the 90o plies.
The corresponding change of Gibbs free energy (equivalent to complementary
460 L.N. McCartney
energy which must be used when crack growth occurs under fixed applied loads)
in the region V of the laminate is
Z
G D .g2 g1 / dV; (16.140)
V
where the suffices 1 and 2 applied to the Gibbs energy per unit volume g denote the
damage state. It has been shown [38, 42, 43] that
Z
1 1
.g g0 / dV D ˆ .s sc /2 ; (16.141)
V V 2
where g0 is the Gibbs free energy per unit volume for an undamaged laminate, s is
the effective applied stress for multi-axial loading defined by
s D k0 t C A C kT ; (16.142)
and where sc D k1 T is the ply crack closure stress when the laminate is subject
only to uniaxial loading in the axial direction normal to the ply cracks in the 90o
plies.
The absorbed energy function appearing in (16.139) is calculated for length
2L of laminate using
h1 X
m
D ıj j ; (16.143)
h L jD1
where 2h1 is the total thickness of all 90o plies in the laminate having total thickness
2h, and where m is the number of potential cracking sites in 90o plies which are
ordered in a regular way, e.g., from top to bottom in the plies which are taken in
order from the centre of the laminate to the outside, symmetry about the mid-plane
of the laminate being assumed. The quantity 2 j is the effective fracture energy for
the growth of a ply crack in the jth potential cracking site of the 90ı plies that are
allowed to crack in a simulation. The parameters ı j describe the crack pattern in the
laminate such that
It is of great practical interest to know the level of applied loading that is likely
to cause the occurrence of ply cracking. Ply crack formation can lead to a loss
of stiffness, an important issue for stiffness critical components, or to leakage in
the case of liquid containers and pressure vessels. It follows from (16.137) and
(16.138)–(16.141) that the formation of an initial set of ply cracks occurs when
16 Predicting Properties of Undamaged and Damaged Carbon Fibre. . . 461
v
u
u 2
s>t 1 1
C sc : (16.145)
EA EA0
The result (16.145) is consistent with fracture mechanics where the energy balance
associated with a long ply crack propagating across the laminate is considered (see,
for example, Sect. 16.9). It is also consistent with the results of Pagano et al. [46]
who applied the three dimensional J-integral to the tip region of a long ply crack in
a cross-ply laminate.
It is useful to discuss progressive cracking in a length 2L of laminate where the
damage state n is defined by fL1 , L2 , : : : Ln g where n is the number of cracks in
length 2L of the 90o plies and where 2Li , i D 1, : : : n denote the distances between
neighbouring cracked planes. The criterion for progressive crack formation may be
written in the form
v
u
u 2 Œ i i1
s> t 1 1
C sc ; i D 1; : : : ; n; (16.146)
.i/ .i1/
EA EA
where the superscript (i) denotes that the effective axial modulus EA is to be
calculated for the damage state i. It is emphasised that the exceedingly simple
results (16.145) and (16.146) are exact, and that they take full account of the
effects of anisotropy, of multi-axial loading and of the effect of thermal residual
stresses. The results involve only macroscopic properties defined at the laminate
level even though they have been derived from a detailed consideration of the stress
and deformation distributions in a cracked laminate at the ply level. The results are
based on the single assumption of damage homogeneity where damage formation
in the laminate is such that the stress/strain relationships (16.129)–(16.131) are
always satisfied. When using (16.146) to investigate progressive crack formation
(where cracks form one at a time) the fracture energies 2 j are taken at random
from a statistical distribution of fracture energies (assumed to be normal) and then
allocated at random to the various potential fracture sites in the 90o plies. The value
of the applied stresses are then regarded as slowly increasing until the condition
(16.146) can be met at just one potential fracture site in the laminate. This means
that crack formation must be investigated at all available sites in order to determine
the location of the next crack to form. This procedure has to be repeated for every
new crack that forms during loading. If the fracture energies at each site have the
same value, then the criterion (16.146) predicts that the crack densities are always
uniform having values 0 , 20 , 40 , etc., where 0 is the initial crack density. Thus
new crack formation for this case occurs as a progressive doubling of the crack
density at a discrete set of applied stress levels.
By introducing the concept of fracture energies for ply cracking that are
statistically distributed, it is possible to consider progressive ply crack formation
for the case where ply cracks form one at a time in locations that lead to non-
uniform ply crack distributions. McCartney and Schoeppner [47] have analysed this
462 L.N. McCartney
more realistic situation and validated two distinct analytical modelling approaches
for non-uniformly cracked laminates. The methodology has been incorporated in a
software module known as PREDICT that forms part of the NPL Component and
Composite Design Analysis system [48]. PREDICT has been used recently as one
contribution (see McCartney [42, 43]) to the third phase of the Worldwide Failure
Exercise (WWFE III) (see [49, 50]), where many of the test cases involve carbon
fibre composites.
So far, the discussion of ply cracking in laminates has referred to the special
case where any ply crack that forms in a 90o ply is regarded as traversing the
entire section of the 90o ply. While this is a reasonable assumption for models of
progressive ply crack formation during monotonic loading, it is certainly inadequate
when modelling ply damage growth due to cyclic loading where fatigue mechanisms
operate in the regions of crack tips. For the latter loading case, which is of significant
practical importance, ply cracks fully embedded within a 90o ply have lengths that
slowly increase during cyclic loading. To model this situation, as described by
McCartney and Kelly [44], it is required to consider ply cracks of total length 2c
embedded in a laminate of total width 2W such that c < W. The following approach
will assume that c << W so that ply cracks do not interact with the free edges of
the laminate.
It is useful to discuss crack bridging for cross-ply composites by referring to
Fig. 16.5. In Fig. 16.5a is shown a face view of a 0o /90o /0o laminate having a single
ply crack with perfectly bonded plies and subject to a uniform uniaxial applied
stress A . Figure 16.5b shows the corresponding edge view of the laminate. Stress
transfer between plies is restricted to the regions A which are adjacent to the ply
crack. For the other regions B stress transfer is negligible so that the local stress and
displacement fields in the plies correspond to those found in an identical undamaged
laminate subject to the same applied stress and temperature. In regions A the stress
carried by the 90o ply is transferred to the adjacent 0o plies by the action of shear
stresses at the ply interfaces. As the ply crack is approached the stress in the 90o ply
reduces to zero while the stress in the 0o plies increases to a maximum value max 0 .
It is assumed that the discontinuous tractions on the ply crack plane associated with
the ply geometry can be approximated by a continuous distribution p(x) such that
undamaged parts of the laminate, is uniform and has been extended to the ply crack
surfaces. The opening of the ply crack, when the continuum traction applied to the
crack faces is p, is denoted by 2. The value of is determined by the axial
displacements of the lines PQ and RS shown in Fig. 16.5. By defining the axial
displacement V so that it is zero on the mid-plane it is clear that
where "A is the axial strain for an undamaged laminate and 2L is the separation of
neighbouring ply cracks in a uniformly cracked laminate. The total crack opening
displacement is denoted by v D 2 . As the problem is linear it follows from the
continuum model shown in Fig. 16.5c that the crack bridging relation has the form
where the parameter and the thermal stress term th are determined by a
stress transfer model (see McCartney [36], for example). It is observed that the
crack bridging relation (16.149) is identical in form to (16.75) which arises when
considering matrix cracking in a unidirectional composite. This means that the
analysis presented in Sect. 16.5 applies also to the embedded ply crack problem. The
only changes that need to be made are that the matrix cracking stress mc
A is replaced
pc
by the ply cracking stress A , and that the parameters , th D k1 T and have
different interpretations in terms of material properties and geometrical parameters.
This means that the normalising parameters 0 and c0 defined by (16.86) and (16.87)
must also have different interpretations. The key conclusion is that the master curve
shown in Fig. 16.2 applies to both matrix cracks in UD composites where the fibres
and matrix are perfectly bonded and to embedded ply cracks in cross-ply laminates,
provided that the normalising parameters 0 and c0 are interpreted correctly.
To conclude this section it is useful to provide values for the parameters 0 ,
1
c0 , c1 and the asymptotic value pc for the ply cracking stress valid as ply crack
length increases to infinity. The effects of thermal residual stresses are ignored
so that th k1 T D 0. Values are shown in Table 16.6 that have been
calculated using the fibre and matrix properties given in Table 16.1 which lead to
the undamaged ply properties given in Table 16.2. Table 16.6 also shows the values
assumed for the ply thicknesses a and b of the 90o and 0o plies, respectively, the
value assumed for the effective fracture energy of the matrix 2 m , and the values
of the parameters , ˝, E0A and EA . The value of appearing in the relation
(16.149) and the value of the modulus EA for a damaged laminate (for a ply crack
density of 0.01/mm) are calculated using results derived by McCartney [36] for
the generalised plane strain analysis that is valid for a simple cross-ply laminate.
The values of the undamaged modulus E0A and the parameter ˝ defined by (16.79),
which is the anisotropy factor or an undamaged laminate, have been calculated using
the undamaged effective properties of the laminate that are calculated using the
1
formulae described in Sect. 16.3. The value of pc shown in Table 16.6 is obtained
using the asymptotic limit (16.89) which applies also to ply cracking in laminates.
464 L.N. McCartney
1
It is useful to determine another estimate of the first ply cracking stress pc for
the laminate by making use of the relation (16.145) and values for the effective
axial modulus of the damaged laminate EA predicted using the formula derived by
1
McCartney [36]. It is found that pc D 0:4625 GPa which, as to be expected, is
slightly above the asymptotic limiting value 0.4608 given in Table 16.6. Use can
also be made of the software tool PREDICT (see http://www.npl.co.uk/science-
technology/composites-and-polymeric-materials/services/predict-software) which
also leads to a first ply cracking stress having the value 0.4625 GPa. This agreement
between the three different predictions of first ply cracking stress indicates
consistency, thus validating the various approaches that have been made. It should be
noted that PREDICT has been designed to predict progressive ply crack formation
in general symmetric laminates (possibly unbalanced) in the presence of thermal
residual stresses for general in-plane loading conditions. It should also be noted that
for cross-ply laminates a model of ply crack formation has been developed for non-
symmetric laminates subject to combined in-plane biaxial and through-thickness
loading, and out-of-plane bending, again in the presence of thermal residual stresses
(see, for example, McCartney [35, 51]).
This chapter has been concerned with the analytical modelling of carbon fibres
either as part of a unidirectionally fibre reinforced composite or as part of a laminate.
A multi-scale approach (at the continuum level) has been made where the results of
applying models at the fibre/matrix length scale (i.e. the micro-scale) have been
used as homogenised effective properties for input into laminate calculations (both
undamaged and damaged by ply crack formation). Two principal characteristics
of carbon fibres, their transverse isotropy and low thermal expansion, have been
taken into account. The latter has been recognised by ensuring that thermal residual
16 Predicting Properties of Undamaged and Damaged Carbon Fibre. . . 465
stresses are included in all models both at the micro and laminate levels. However,
it must be emphasised that carbon fibres have a third characteristic that cannot
be modelled using analytical models for composite materials. Their non-linear
behaviour, especially at low strains, can have an important effect on deformation
and damage development but this can only be modelled using numerical methods
such as the finite element method. While this aspect has been neglected, the results
presented in this chapter provide considerable insight into the behaviour of fibre
reinforced materials, and analytical formulae that have been given can be used
to help design damage resistant composite materials, and predict the progressive
development of damage when fibre reinforced materials are used in laminate form
during the construction of engineering components.
A final remark that must be made is that the carbon fibre story has much further
to develop. The relatively recent advent of carbon nanotubes (both single- and multi-
walled) is leading to hopes that even more damage resistant composite materials can
be developed and used in commercial applications. The advent of ‘split’ nanotubes
in the form of graphene sheets is also the subject of intense study. While the
principal practical problems are the dispersion of nanoscaled reinforcements in
polymer matrices, and the scaling-up of the size of batches of composite material
samples, there are additional modelling challenges that will need to be addressed.
For example, the inclusion of ‘surface tension’ effects associated with the use of
very small diameter fibres has been considered and will continue to be needed.
Also, the problem of representing, using continuum models, reinforcements that are
only one atom thick needs careful consideration. The current concept of multi-scale
modelling of composite materials clearly needs to be extended from the continuum
level to encompass macro, meso, micro and atomistic length scales in a consistent
physically acceptable way.
References
1. J.C. Maxwell, A treatise on electricity and magnetism, Chapter 9, (vol. 1, Art. 310–314, pp.
435–441), (Clarendon Press, Oxford, 1873), 1st edn. (3rd edn., 1892). See also Art. 322, pp.
447–448, for a composite laminate analysis
2. Z. Hashin, Analysis of composite materials—a survey. J. Appl. Mech. 50, 481–505 (1983)
3. L.N. McCartney, A. Kelly, Maxwell’s far-field methodology applied to the prediction of
properties of multi-phase isotropic particulate composites. Proc. R. Soc. Lond. A464, 423–446
(2008)
4. L.N. McCartney, Maxwell’s far-field methodology predicting elastic properties of multiphase
composites reinforced with aligned transversely isotropic spheroids. Philos. Mag. 90, 4175–
4207 (2010)
5. W.B. Rosen, Z. Hashin, Effective thermal expansion coefficients and specific heats of
composite materials. Int. J. Eng. Sci. 8, 157–173 (1970)
6. R.M. Christensen, K.H. Lo, Solutions for effective shear properties in three phase sphere and
cylinder models. J. Mech. Phys. Solids 27, 315–330 (1979)
7. H.L. Cox, The elasticity and strength of paper and other fibrous materials. Br. J. Appl. Phys. 3,
72–79 (1952)
466 L.N. McCartney
34. L.N. McCartney, Theory of stress transfer in a 0-90-0 cross-ply laminate containing a parallel
array of transverse cracks. J. Mech. Phys. Solids 40, 27–68 (1992)
35. L.N. McCartney, Physically based damage models for laminated composites. Proc. Inst. Mech.
Eng. Part L. J. Mater. Des. Appl. 217, 163–199 (2003)
36. L.N. McCartney, Microlevel approaches to the modelling of damage in composite materials:
generalised plane strain analysis, Chapter 13, ed. by R. Talreja, J. Varna, in Modelling Damage,
Fatigue and Failure of Composite Materials (Woodhead Publishing, Cambridge, 2016)
37. M. Hajikazemi, M.H. Sadr, R. Talreja, Variational analysis of cracked general cross-ply
laminates under bending and biaxial extension. Int. J. Damage Mech. 24, 582–624 (2015)
38. L.N. McCartney, Model to predict effects of triaxial loading on ply cracking in general
symmetric laminates. Compos. Sci. Technol. 60(2000), 2255–2279 (2000)
39. L.E. Crocker, S.L. Ogin, P.A. Smith, P.S. Hill, Intra-laminar fracture in angle-ply laminates.
Compos. Part A 28A, 839–846 (1997)
40. J. Tong, F.J. Guild, S.L. Ogin, P.A. Smith, On matrix crack growth in quasi-isotropic laminates
- I. Experimental investigation, - II Finite element analysis. Compos. Sci. Technol. 57,
1527–1545 (1997)
41. M. Hajikazemi, M.H. Sadr, A variational model for stress analysis in cracked laminates with
arbitrary symmetric lay-up under general in-plane loading. Int. J. Solids Struct. 51, 516–529
(2014)
42. L.N. McCartney, Energy methods for modelling damage in laminates. J. Compos. Mater.
47(20–21), 2613–2640 (2013)
43. L.N. McCartney, Derivations of energy-based modelling for ply cracking in general symmetric
laminates. J. Compos. Mater. 47(20–21), 2641–2673 (2013)
44. L.N. McCartney, Energy methods for fatigue damage modelling of laminates. Compos. Sci.
Technol. 68, 2601–2615 (2008)
45. L.N. McCartney, Predicting transverse crack formation in cross-ply laminates resulting from
micro-cracking. Compos. Sci. Technol. 58, 1069–1081 (1998)
46. N.J. Pagano, G.A. Schoeppner, R. Kim, F. Abrams, Steady state cracking and edge effects
in thermo-mechanical transverse cracking of cross-ply laminates. Compos. Sci. Technol. 58,
1811–1825 (1998)
47. L.N. McCartney, G.A. Schoeppner, Predicting the effect of non-uniform ply cracking on
the thermo-elastic properties of cross-ply laminates’. Compos. Sci. Technol. 62, 1841–1856
(2002)
48. CoDA (2016) http://www.anaglyph.co.uk/coda.htm
49. A.S. Kaddour, M.J. Hinton, P.A. Smith, S. Li, The background to the third world-wide failure
exercise. J. Compos. Mater. 47(20–21), 2417–2426 (2013)
50. A.S. Kaddour, M.J. Hinton, P.A. Smith, S. Li, A comparison between the predictive capability
of matrix cracking, damage and failure criteria for fibre reinforced laminates: part A of the
third world-wide failure exercise. J. Compos. Mater. 47(20-21), 2749–2779 (2013)
51. L.N. McCartney, in Stress Transfer Mechanics for Multiple Ply Cross-Ply Laminates Subject
to Bending, Summary in Proceedings of 6th International Conference on the Deformation and
Fracture of Composites, Manchester, April 2001, pp. 57–66
Part V
Chapter 17
Composites Toughen Up!
Ivana K. Partridge
17.1 Introduction
Toughening of DGEBA epoxies with reactive liquid rubber was already established.
It was known that at least two reactive end groups were required on the epoxy and
that the rubber had to be carboxyl or amine terminated, in order to initially dissolve
into the epoxy monomer. Increasing molecular weight of the epoxy through cure
would then lead to phase separation and the creation of distinct spherical rubber-rich
particles, in size range of microns. Typical rubber concentrations of under 10 % by
weight led to an order of magnitude increase in the GIC of the cured resin, but only
about a 20 % reduction in the elastic modulus. The near complete phase separation
of the rubber from the epoxy ensures a minimal effect of the modifier upon the Tg
of the continuous epoxy phase. This was the start of the extremely effective use of
such toughened epoxies in the engineering adhesives context. At the time, it was
not really known what made the particular type of microstructure so efficient in
17 Composites Toughen Up! 473
toughening. Early work by Sultan and McGarry [2], Meeks [3] and Bucknall [4]
and later more detailed studies [5–7] established that the phase-separated rubber-
rich discrete particles fail by internal cavitation upon loading, initiating energy
absorption in the matrix by extensive shear yielding. Parametric studies on the
effects of the rubber particle size (typically between 0.1 and 5 m in diameter)
were never clear-cut as the particle size changes are accompanied by changes in
the internal structure of the particles and in the particle volume fraction. It was
generally agreed that particles much below 0.5 m in diameter were ineffective in
triggering the important shear yielding in the surrounding matrix, and there were
suggestions that a bimodal size distribution of rubber particles might be efficient in
toughening. Similar rubber particles employed in engineering thermoplastics, such
as polystyrene, can also bring about multiple crazing, but this particular energy
absorption mechanism has never proved to be significant in highly cross-linked
networks.
Indeed the high crosslink density also inhibits shear yielding; cured difunctional
epoxies are able to exhibit a distinct yield drop in compression but not so the
higher functionality epoxies. Hence, there are two reasons why reactive rubbers
are not usable for toughening purposes in TGDDM—firstly they cannot withstand
the cure temperatures required, and secondly, even if successfully distributed phase
separation could be achieved (not easy!), there is no shear yielding to be exploited.
This is an important point, underpinning the next stage of development of these
matrix resins.
Fig. 17.1 SEM micrographs of fracture surfaces of PEI-modified tetrafunctional epoxy resins
5 m. (a) Tetrafunctional epoxy containing sufficient amount of PEI to achieve phase
inversion upon cure; the continuous phase is the thermoplastic. (b) Another fracture surface sample
from the same specimen, after etching with methyl-ethyl-ketone, dissolving away the thermoplastic
little variation to be achieved, the Dicy dictating its own cure path. The resins
remained resistant to solvents and other chemicals and maintained the intended
high Tg of above 180 ı C. The cured composites were uniform, generally of
excellent manufacturing quality, but not ‘tough’. Change of hardener to the more
controllable diamino-diphenyl-sulphone (DDS) varieties also entailed a qualitative
change in the phase separation between the multifunctional epoxies and the PES,
with phase separation indicated to occur on a submicron scale [8]. Among the other
potential thermoplastic modifiers investigated at the time, polyetherimide (PEI)
looked promising [9, 10].
First indications of significant increase in toughness of such cured blends came
from compositions containing sufficient amount of the thermoplastic to arrive at
phase inversion, a state in which the thermoplastic becomes the continuous phase
(see Fig. 17.1a). This makes good sense in light of the above comments on the need
for a shear-yielding mechanism in the matrix to make it toughenable. However, there
is an obvious drawback to having a continuous phase of amorphous thermoplastic,
namely, the collapse of solvent resistance (Fig. 17.1b). More chemical know-how
and targeted functionalisation of the modifiers were needed.
The next step in the understanding and some very successful commercial
offerings came from the ICI/Fiberite powerhouse of McGrail and Street [11]. The
Fiberite 977 series exemplifies the families of the so-called second-generation
thermoplastic-modified high-temperature resins. The phase separation is in the form
of submicron co-continuous networks of functionalised aromatic thermoplastics and
the multifunctional epoxies, creating toughenability in the matrix without sacrificing
any of the desired chemical and thermomechanical properties (Fig. 17.2). As to
toughness as such, this was roughly doubled in comparison to the fracture toughness
of the base resins. Although nice to have, it is a sobering thought that the GIC of
PMMA is 500 J/m2 , and we do not often think of that polymer as ‘tough’.
17 Composites Toughen Up! 475
Fig. 17.2 ‘Cusps’ of F977-1 in a resin-rich layer of a UD carbon fibre laminate, fractured under
shear loading. At resolution levels offered by SEM, the resin is featureless
temperatures, and typical aerospace components were not ready to be made by hot
stamping. Commercial version of automated fibre placement (AFP) was some way
in the future. Aerospace composites were back to rely on thermosets, with both
epoxies and bismaleimides [14] well into their ‘second-generation’ developments.
The above stated resin GIc values are relevant in the sense that cracks will almost
invariably start in the matrix resin or at the fibre-resin interface. Thus, the resistance
of the matrix to crack initiation can be seen as a critical attribute, as determined
by the fracture mechanical test. Mostly these tests are carried out at ambient tem-
perature and under quasi-static loading conditions, both for the unreinforced resin
and for its composite. However, when new resins are being formulated, the down
selection of the promising versions is carried out on the basis of the Compression-
after-impact test (CAI) on the laminate, more representative of the in-use condition.
This is an elaborate and costly test, usually not shedding any additional light on
the reasons for any improvement or otherwise in the performance of the composite.
Furthermore, the driving mode for delamination under impact is shear (i.e. Mode
II), not a value easily determined in the composite and arguably undefined in the
unreinforced resin. Considerable effort therefore came to be expended in looking
into the relationships between the fundamental fracture mechanical values in resins,
how these translate into the corresponding values in the composites and whether any
of these correlate with the CAI results.
In a seminal paper from the National Bureau of Standards [15], a plot of GIC in
composite versus GIC of its resin matrix over the widest range of resins available
suggested that a 1:1 correspondence was only achieved in the fairly brittle resins,
with the effectiveness of transfer of any additional toughening from the matrix
to the composite being severely limited thereafter. Whilst the exact values on
the graph could be questioned (exact states of cure of the resin in the different
forms, measurement validity at high toughness levels, etc.), its main message
corresponded to experience. The rationalisation followed fairly swiftly—if the resin
crack-resisting performance relies on the creation of a large process zone to blunt
the crack tip and this formation is constrained by the proximity of stiff fibre plies
[16], then toughening of the resin becomes a law of diminishing returns. The theory
can be tested by self-same resin interleaving [17, 18], whereby additional layers of
the matrix resin film are placed in the crack path of the fracture test specimen in
the lay-up stage and co-cured with the same prepreg matrix. This avoids potentially
weak interfaces between dissimilar resins, which plagued earlier versions of this
type of study. It can be demonstrated that the GIC in the composite initially rises
quite rapidly as the thickness of the cured resin interlayer increases but asymptotes
to a maximum at the resin GIC value. In contrast, the potential for improvement
of the GIIc values in the composite is seemingly unlimited, showing a linear trend
17 Composites Toughen Up! 477
Fig. 17.3 SEM of cryofracture surface of cured (Fibredux) 920 resin matrix and enlargement of
central section. The largest particle dimension is 20 m and will define the minimum spacing
between fibre plies in the cured composite
with interleaf resin thickness of up to 300 m [19]. In practical terms, such thick
interleaves are undesirable as they begin to be a significant detriment to the stiffness
of the laminate, but a way forward to increasing the CAI test values is fairly clear.
With the aerospace industry worrying about brittle thermosets and competition with
new high-performance thermoplastics, the F1 industry commissioned its own tough
resin matrix, namely, Fibredux 920. The big advantage of the application is the
fact that the Tg requirement can be relaxed to below 120 ı C, making the use of
the more flexible (and hence toughenable) difunctional epoxy as possible as the
matrix basis. Closer examination of fracture surfaces of this resin (see Fig. 17.3)
reveals the presence of particles that are larger than those observed in phase-
separated thermoset/thermoplastic blends and nonspherical. The shape indicates that
the particles cannot have arisen from phase separation but rather been shaped by the
milling of the resin insoluble thermoplastic additive.
Whatever the detailed mechanisms of toughening might be in this resin, the
presence and the size of the added particles during cure of the composite ensure
a fairly fixed separation between the fibre plies—the above mentioned studies
suggest that this in itself would ensure an improvement in the CAI performance
compared to composites without such particle-rich interlayers. A number of so-
called interlayer-toughened aerospace grade systems followed [20] forming a new
family of ‘third-generation’ matrices. Their variants find increasing use in prepreg-
based structural applications at present (Fig. 17.4).
478 I.K. Partridge
Fig. 17.4 Polished cross-sections of conventional carbon fibre UD laminate on the left; interlayer
particle-toughened commercial system on the right. Average interlayer thickness 25 m
By the mid-1990s it had become fairly clear that existing thermosetting resin
chemistry and formulation had reached some kind of a limit in terms of the
resistance to crack initiation in the composite matrix, somewhere near 400 J/m2
under quasi-static crack opening loading at ambient temperature. Not even the
toughness of PMMA : : : .. The early promise of PEEK and its variants to provide
significantly tougher composites to the aerospace industry had receded in face of
processing difficulties and implications of a wholesale change to thermoplastics on
manufacturing costs.
Wood has always been known to delaminate easily and solutions to construction
of highly loaded structures from it by the use of woodworking design features and
metallic nails have stood the test of time. Why not ‘nail’ laminate plies together
then?
®
Fig. 17.5 From left to right: Commercial Z-Fiber double-foam preform; detail of chamfered
0.5 mm diameter Z-pin; early hand-held ultrasonic insertion hammer and power source
matrix sufficiently to allow some sideways movement of the in-plane ply fibres,
accommodating the z-direction inserted pin with much less damage to the rein-
forcing fibres than would have been incurred by drilling a suitable hole for the
pin in a cured laminate. During subsequent cure the matrix has a chance to flow
around and seal onto the Z-pins, making them more difficult to disbond from the
cured laminate under eventual loading. The early Z-pins were made of titanium,
compatible with application to carbon fibre prepreg. There were some difficulties
with consolidation of the prepreg around the metallic pins, and the mismatch in the
coefficient of thermal expansion meant that often they were not very well bonded
to the laminate, pulling out easily under Mode I loading. The currently preferred
Z-pins are made by a pultrusion process, from fine carbon fibre tows impregnated
with a high-temperature resistant bismaleimide resin. Because the cured Z-pins are
slender (diameter 0.28 mm or 0.5 mm for pins up to 40 mm long), it is important
that their matrix resin does not soften on the localised heating that occurs in the
ultrasonic insertion. Fuller descriptions of the process of manufacture of a Z-pinned
composites can be found elsewhere [22]. It is worth noting that shaped metallic pins
are again being considered, in the context of metal-to-composite and composite-to-
composite joints [23].
The role of the Z-pins in a laminate under load is for the rows of Z-pins to slow
down and stop an approaching delamination crack. This they do very successfully,
particularly under Mode I loading [24] (Fig. 17.6). The ‘toughness’ improvement
can be of an order of magnitude, much higher than can be envisaged as achievable
through the use of a toughened matrix. The pins effectively shield the crack tip by
bridging across the crack opening faces, to an extent given by the local topology
(pin block length, pinning pattern, fibre ply lay-up and laminate thickness) and by
the crack-driving forces. Friction between the debonded Z-pin and its surrounding
composite is the main energy-absorbing mechanism.
Delamination-limiting performance of Z-pins under shear loading can be less
convincing, especially if the Z-pin sideways displacement is constrained by the
prepreg fibre architecture (e.g. tight cross-ply lay-up). In such cases the Z-pins
behave in a fairly brittle manner, absorbing little energy in fracturing. Nevertheless,
the CAI performance of Z-pinned test specimens has been shown to be attractive,
and no detrimental effects on fatigue crack growth rates have been found [25].
480 I.K. Partridge
Fig. 17.6 Carbon fibre/BMI Z-pins bridging the delamination plane in a Mode I DCB test
specimen. The energy-absorbing process zone extends behind the visible crack tip
It is self-evident that dry fibre preforms can be held together very effectively
by judicious placement of z-direction threads, fixed in position by either loops
(‘stitches’) or by friction against the thread within the preform itself and/or a
manufacturing aid such as a foam bed placed underneath the preform (‘tufts’).
Various forms of stitching have been investigated extensively both by academic
studies and in the industry, but somehow the technology failed to gain widespread
use. Possibly, as with the earlier case of thermoplastic matrices, there was little
appetite for investment into large-scale new production machinery, complicated by
the requirement of two-sided access to the preform workpiece. In technical context,
the high level of crimp, imposed onto the preform fabric by locked seams and loops
of the stitching process, may have resulted in unacceptably high detriment to the
in-plane properties of stitched composites [26].
A much more recently introduced technology is the robotic application of the
age-old tufting process [27]. This is a single-sided access process, completely
analogous to the manufacture of carpets, but the tufts are placed selectively and
much less densely spaced, into the dry fibre preform. Figure 17.7 shows a modern
KSL tufting head fixed on a robot, which enables it to reach a large working space,
with the potential of tufting simple or curved structures of metre scale dimensions.
The minimum radius of curvature that can be tufted is defined by the dimensions of
the tufting head but also by the depth of preform that needs to be tufted. Because the
needle usually needs to penetrate the preform (see Fig. 17.7 top right) if the loops
on the underside are to be held in place by the foam bed, it is currently not possible
to tuft directly on a solid tool. This is a disadvantage of the technology and may
be resolved in the future by designing tufting patterns that terminate within the dry
preform bed. However, this will require us to develop more detailed understanding
of the mechanism of failure and hence the performance of the tufts upon subsequent
loading of the cured structure. Design for manufacture procedures will then have to
be fully adapted to such an approach. This is some way away in terms of current
developments. For the time being, attention is focused on creating the shortest tufts
17 Composites Toughen Up! 481
Fig. 17.7 The KSL tufting head on robot arm; schematic of the manufacturing process and a tuft
block of 0.5 mm aramid thread loops on the underside of a tufted carbon fibre preform, lifted off
its foam support block
possible consistent with them remaining in place once the tufted preform is lifted
off the supporting bed (see Fig. 17.7 bottom right). When they flatten out on the
tool during subsequent resin infusion and cure, the short loops give rise to a surface
resin-rich region of a limited thickness, which is preferred.
The tufts bridge the growing delaminations in much the same way as the solid rod
Z-pins in the case of prepreg-based laminates, but their internal structure is much
more complicated, as is the multiplicity of possible energy-absorbing mechanisms
of failure in the tufted composites. Under Mode I loading, the tufts stretch and may
debond to an extent but do not pull out (Fig. 17.8). Final failure is tuft failure, the
friction between the tufts and its environment never being fully exploited. It is thus
likely that the balance between the resistance to crack opening and to shear will be
different to that noted in Z-pinned composites, at least where the carbon fibre/BMI
Z-pins are used. As in the case of Z-pinning, tufting offers a greater improvement in
the yard-stick CAI test than is achievable by matrix toughening [28].
The localised reinforcement described in Sects. 17.6.1 and 17.6.2 cannot be applied
all over any engineering structure. Not only would it make the structure manufacture
unsustainably costly, it would also add too much weight. Furthermore, ‘too much
strengthening’ simply shifts the location of the eventual structural failure to a
different place and often to a different mode. This can create a more sinister
problem that the one the localised reinforcement was intended to prevent (albeit at a
482 I.K. Partridge
Fig. 17.8 (a) Glass thread tuft path and loop in cured carbon fibre fabric/epoxy composite, shown
by a CT scan. The infused and cured tuft diameter is 0.5 mm. (b) Stretched and partially debonded
tuft in a composite sample subjected to crack opening loading
Fig. 17.9 Left—general modelling scheme developed for Z-pinning of laminates; right—design
of the single Z-pin test specimen, which provides the basic bridging law for the particular geometry
and environment of the Z-pin
significantly higher load level). Hence, the design of highly damage resistant struc-
tures requires intelligent placement of limited through-the-thickness reinforcement
(TTR). Figure 17.9 summarises the current modelling philosophy being developed
for the case of delamination performance of Z-pinned composites [29].
There is a tight interplay between the fundamental experiments (left-hand side of
the scheme) and the largely finite element model development (right-hand side
of scheme), with independent coupon and structural element tests to establish the
17 Composites Toughen Up! 483
Fig. 17.10 Glass fibre/epoxy T-joint, highly reinforced with carbon thread tufts, was loaded in
4-point bending up to ultimate flexural failure in the skin section. A permanent deflection of the
sample is evident. The control sample (on the bottom) was not loaded up to complete separation
and shows the unconstrained and undamaged ‘noodle’ region and no permanent deflection
validity of the models. The single Z-pin test specimen, shown in Fig. 17.9 for
the specific case of an 8 mm thick laminate, is essentially the calibration step in
the scheme, fixing the basic bridging law to a specific combination of pin type
and size and the environment in which it exists [30]. A semi-analytical treatment
of this bridging law relates a small number of experimentally determined critical
parameters to a wider range of loading situations and hence to the development of
numerically efficient cohesive elements which can then be used in scaled up versions
of the structural models.
Within the range of experimental error, and the currently observed manufacturing
variability, the modelling scheme has been found to perform very satisfactorily
for structural elements such a T-joints [31]. Work is underway to determine if
loading rate dependency can be captured within the same or modified modelling
scheme. It is also appreciated that in realistic structures single delaminations are
rare, and hence any future design schemes need to address the question of multiple
delaminations [32] as well as the question of the extent of changes in the in-plane
properties of the composite as a result of the through-the-thickness reinforcement.
This is to avoid shifting the failure mode of a locally reinforced structure to an
undesirable scenario, such as depicted in Fig. 17.10. A typical unreinforced T-joint
structural element would be expected to delaminate completely between the skin and
the stiffener, when tested in 4-point bending. It has been shown that placing several
microfastener rows on the edge of the skin-spar joint greatly increases the load-
bearing capacity of the structure, in the sense that the delaminations are stopped by
the microfastener block. This prevents total separation of the skin and the stiffener,
and energy is absorbed by multiple delaminations within the T-stiffener, triggered by
the stress concentration around the ‘noodle’. As the central damage region around
the ‘noodle’ becomes more constricted by the placement of additional reinforcement
rows, the mode of failure changes from multiple delaminations between the skin and
484 I.K. Partridge
the stiffener and within the stiffener to the outer skin flexural failure. The highly
constrained specimen shown in Fig. 17.10 represents an extreme case of the use of
localised reinforcement, in this case carbon thread tufts.
The general modelling scheme outlined in Fig. 17.9 should apply equally well
to tuft-reinforced structures, but early work indicates that the complex geometry of
the embedded tufts (in comparison to the Z-pins) requires subtly different treatment,
on multiple scales. Equally, the question of the relative mechanical effectiveness of
the two different reinforcement types remains unresolved, as rigorous experimental
comparisons are hampered by the inevitable differences between the fibre forms and
resins used in the two manufacturing routes [33].
would be too high. There have been attempts to co-locate dissolvable thermoplastic
®
threads with the reinforcement threads in the dry preform (Priform ), but it is
difficult to incorporate sufficient thermoplastic additive this way and other physical
forms continue to be investigated [37, 38]. Nanosilica additive in certain epoxies
appears to increase toughness without increasing the viscosity of the resin [39].
Despite such promising cases, it seems that a new wave of inventive effort is due, to
produce well-toughened resin infusion systems, particularly of the new ‘snap-cure’
varieties. The new generations of composites still need to ‘toughen up’.
References
1. C.B. Bucknall, I.K. Partridge, Phase separation in epoxy resins containing polyethersulphone.
Polymer 24, 639–644 (1983). doi:10.1016/0032-3861(83)90120-9
2. J.N. Sultan, F.J. McGarry, Effect of rubber particle size on deformation mechanisms in glassy
epoxy. Polym. Eng. Sci. 13, 29 (1973)
3. A.C. Meeks, Fracture and mechanical properties of epoxy resins and rubber-modified epoxy
resins. Polymer 15, 675 (1974)
4. C.B. Bucknall, T. Yoshii, Relationship between structure and mechanical properties in rubber-
toughened epoxy resins. Br. Polym. J. 10, 53 (1978)
5. A.J. Kinloch, S.J. Shaw, D.L. Hunston, Deformation and fracture behaviour of a rubber
toughened epoxy. Polymer 24, 1341–1355 (1983)
6. D. Verchère, H. Sautereau, J.P. Pascault, C.C. Riccardi, S.M. Moschiar, R.J.J. Williams,
Rubber-Modified Epoxies. I. Influence of carboxyl-terminated butadiene-acrylonitrile random
copolymers (CTBN) on the polymerisation and phase separation processes. J. Appl. Polym.
Sci. 41, 467 (1990)
7. A.F. Yee, R.A. Pearson, Toughening mechanisms in elastomer modified epoxies Pt I—
Mechanical studies. J. Mater. Sci. 31, 2462–2474 (1986)
8. C.B. Bucknall, I.K. Partridge, Phase separation in crosslinked resins containing polymeric
modifiers. Polym. Eng. Sci. 26, 54–62 (1986)
9. C.B. Bucknall, A.H. Gilbert, Toughening tetrafunctional epoxy resin using polyetherimide.
Polymer 30, 213 (1989)
10. A.J. Kinloch, M.L. Yuen, S.D. Jenkins, Thermoplastic-toughened epoxy polymers. J. Mater.
Sci. 29, 3781–3790 (1994). doi:10.1007/BF00357349
11. P.T. McGrail, A.C. Street, Structure-property relationships in high performance
thermoset-thermoplastic blends. Macromol. Chem. Macromol. Symp. 64, 75–84 (1992).
doi:10.1002/masy.199220640110
486 I.K. Partridge
12. P. Davies, H.H. Kausch, J.G. Williams, A.J. Kinloch, M.N. Charalambides, A. Pavan, D.R.
Moore, R. Prediger, I. Robinson, N. Burgoyne, K. Friedrich, H. Wittich, C.A. Rebelo, A. Torres
Marques, F. Ramsteiner, B. Melve, M. Fischer, N. Roux, D. Martin, P. Czarnocki, D. Neville,
I. Verpoest, B. Goffaux, R. Lee, K. Walls, N. Trigwell, I.K. Partridge, J. Jaussaud, S. Andersen,
Y. Giraud, G. Hale, G. McGrath, Round-robin interlaminar fracture testing of carbon-fibre-
reinforced epoxy and PEEK composites. Compos. Sci. Technol. 43, 129–136 (1992)
13. P.T. Curtis, An investigation of the mechanical properties of improved carbon fibre composite
materials. RAE Technical Report TR86021, 1986
14. H.D. Stenzenberger, W. Römer, M. Herzog, P. König, Toughened bismaleimide: modification
with thermoplastic, in Proceedings 33rd International SAMPE Symposium, Anaheim, 1988,
pp. 1546–1561
15. D.L. Hunston, R.J. Moulton, N.J. Johston, W.D. Bascom, Matrix Resin Effects In Composite
Delamination: Mode I Fracture Aspects, in Toughened Composites, ASTM STP 937, ed. by
N.J. Johnston (ASTM, Philadelphia, 1987), pp. 74–94
16. W.L. Bradley, Relationship of Matrix Toughness to Interlaminar Fracture Toughness, in
Application of Fracture Mechanics to Composite Materials, ed. by K. Friedrich (Elsevier, New
York, 1989)
17. I.K. Partridge, J.A.M. Jaussaud, P.E.L. Corberand, Toughness transfer from matrix to compos-
ite under static and dynamic loadings, in Proceedings 1st International on Deformation and
Fracture of Composites (DFC1), IOM3, Manchester, 25–27 March 1991
18. S. Singh, I.K. Partridge, Mixed-mode fracture in an interleaved carbon-fibre/epoxy composite.
Compos. Sci. Technol. 55, 319–327 (1995)
19. I.K. Partridge, D.D.R. Cartié, Suppression of Initiation of Delamination Cracking in Unidi-
rectional Composites, in The Application of Fracture Mechanics to Polymers, Adhesives and
Composites, ed. by D.R. Moore (Elsevier, New York, 2004), pp. 265–271. ESIS Publication
33
20. N. Odagiri, T. Muraki, K. Tobukuro, Toughness improved high performance prepreg
T800/3900 series, in Proceedings 33rd International SAMPE Symposium, Anaheim, 1988,
p. 272
21. G. Freitas, C. Magee, P. Dardzinski, T. Fusco, Fiber insertion process for improved damage
tolerance in aircraft laminates. J. Adv. Mater. 24, 36–43 (1994)
22. I.K. Partridge, D.D.R. Cartié, T. Bonnington, Manufacture and Performance of Z-Pinned
Composites, in Advanced Polymeric Materials: Structure—Property Relationships, ed. by
S. Advani, G. Shonaike (CRC Press, Boca Raton, FL, 2003)
23. S. Stelzer, S. Ucsnik, G. Pinter, Strength and damage tolerance of composite-composite joints
with steel and titanium through the thickness reinforcements. Composites A88, 39–47 (2016)
24. D.D.R. Cartié, I.K. Partridge, Suppression of Propagation of Delamination Cracking in Unidi-
rectional Composites, in The Application of Fracture Mechanics to Polymers, Adhesives and
Composites, ed. by D.R. Moore (Elsevier, New York, 2004), pp. 273–278. ESIS Publication
33
25. A. Rezai, D. Cartié, I. Partridge, P. Irving, T. Ashton, P. Negre, J. Langer, Interlaminar damage
®
resistance of Z-Fiber reinforced structural CFRP, in Proceedings of ICCM13, Beijing, 2001
26. A.P. Mouritz, K.H. Leong, I. Herszberg, A review of the effects of stitching on the in-plane
mechanical properties of fibre-reinforced polymer composites. Composites A28, 979–991
(1999)
27. G. Dell’Anno, J.W.G. Treiber, I.K. Partridge, Manufacturing of composite parts rein-
forced through-thickness by tufting. Robot. Comput. Integr. Manuf. 37, 262–272 (2015).
doi:10.1016/j.rcim.2015.04.004
28. G. Dell’Anno, D.D.R. Cartié, I.K. Partridge, A. Rezai, Exploring mechanical property balance
in tufted carbon fabric/epoxy composites. Composites A38, 2366–2373 (2007)
29. B. Zhang, G. Allegri, M. Yasaee, S.R. Hallett, Micro-mechanical finite element analysis of
Z-pins under mixed-mode loading. Composites A78, 424–435 (2015)
17 Composites Toughen Up! 487
30. M. Yasaee, J. Lander, G. Allegri, R.S. Hallett, Experimental characterisation of mixed mode
traction-displacement relationship for a single carbon composite Z-pin. Compos. Sci. Technol.
94, 123–131 (2015)
31. F. Bianchi, T.M. Koh, X. Zhang, I.K. Partridge, A.P. Mouritz, Finite element modelling of
z-pinned composite T-joints. Compos. Sci. Technol. 73, 48–56 (2012)
32. M. Yasaee, G. Mohamed, S.R. Hallett, Interaction of Z-pins with multiple mode II delamina-
tions in composite laminates. Exp. Mech. (2016). doi:10.1007/s11340-016-0175-9
33. D.D.R. Cartié, G. Dell’Anno, E. Poulin, I.K. Partridge, 3D reinforcement of stiffener-to-skin
T-joints by Z-pinning and tufting. Eng. Fract. Mech. 73, 2532–2540 (2006)
34. D.J.-P. Turmel, I.K. Partridge, Heterogeneous phase separation around fibres in epoxy/PEI
blends and its effect on composites delamination resistance. Compos. Sci. Technol. 57,
1001–1007 (1997)
35. R.W. Venderbosch, H.E.H. Meijer, P.J. Lemstra, Processing of intractable polymers using
reactive solvents: 2 Poly(2,6-dimethyl-1,4-phenylene ether) as a matrix materials for high
performance composites. Polymer 36, 1167–1178 (1995)
36. K. Potter, But how can we make something useful out of black string? The development of
carbon fibre composites manufacturing 1965–2015, in The Structural Integrity of Carbon Fiber
Composites, ed. by P. Beaumont, C. Soutis (Springer International Publishing Switzerland,
2017). doi:10.1007/978-3-319-46120-5_2
37. R.W. Hillermeier, J.C. Seferis, Interlayer toughening of resin transfer molding composites.
Composites A32, 721–729 (2001)
38. V.A. Ramirez, P.J. Hogg, W.W. Sampson, The influence of the nonwoven veil architectures
on interlaminar fracture toughness of interleaved composites. Compos. Sci. Technol. 110,
103–110 (2015)
39. A.J. Kinloch, R.D. Mohammed, A.C. Taylor, C. Eger, D. Egan, The effect of silica nanopar-
ticles and rubber particles on the toughness of multiphase thermosetting epoxy polymers.
J. Mater. Sci. 40, 5083–5086 (2005). doi:10.1007/s10835-005-1716-2
Chapter 18
Slow Cracking in Composite Materials:
Catastrophic Fracture of Composite Structures
Peter W. R. Beaumont
Glare has better corrosion and impact resistance than conventional aluminium
alloys and can be repaired using conventional repair techniques. Laser beam
welding, for example, has the added advantage of eliminating riveting in much of its
construction. Modern protective clothing in the British Army consists of alternate
layers Kevlar/ceramic material. In Glare there is favourable resistance to fatigue
crack growth around large wing joints. This is because any crack formed grows in a
single layer only, the remaining intact layers effectively bridging the crack.
This effect is observed in wood where a split exhibits fibre bridging and carries
traction, leading to crack arrest. On the Islands of Micronesia, we observe the
traditional method of making sailboats by glueing and stitching together hewn
wooden planks with natural fibre to inhibit splitting and cracking (Fig. 18.2).
Today, we call that z-pinning or tufting [2]. More said later about crack bridging
mechanisms and toughness.
On BBC television (during Wimbledon’s international tennis tournament, June
2009), a commentator characterised the ideal composite player by combining the
best attributes selected from the most outstanding tennis players in the history of
the game. But already composite man exists with advances in implant surgery.
However, the first use of the term composites in engineering refers to the composite
construction of the nineteenth-century clipper ships—ocean sailing vessels that
brought wool from Australia and tea from China to England—constructed from
conventional wooden planking on novel iron frames (Fig. 18.3).
Most of the 170,000 airplanes built during World War I were constructed of
wooden frames using adhesively bonded joints of spruce, birch plywood or balsa
wood across which were stretched fabric coverings impregnated with a resin for
improved lightness (Fig. 18.4). Today, in the wings of the Airbus A380, we
have familiar aluminium skins on composite ribs and frames. For Boeing’s 787
18 Slow Cracking in Composite Materials: Catastrophic Fracture. . . 491
Fig. 18.3 The first use of the term composites in engineering refers to the composite construction
of the clipper ships—of wooden planking on iron frames. The iron frame was the novel element at
the time on the conventional wooden plank
Dreamliner, the upper and lower surface panels and spars of the wing are made of
carbon fibre composite on ribs of aluminium machined from a single plate. But the
first application of carbon fibre in aerospace did not begin with a success story.
In 1971, the jet engine manufacturer Rolls-Royce came unstuck with their new
RB211 engine (Fig. 18.5). Disaster struck this first generation of big turbofan jet
engines as a consequence of the brittleness of the carbon fibre-epoxy composite
®
compressor fan blades (known by the name HyFil ) under foreign object impact
(FOI). The novel composite, from which these large blades were made, shattered
492 P.W.R. Beaumont
Fig. 18.4 Early construction of airplane wings consisting of wooden struts upon which is stretched
cloth impregnated with dope
into pieces at fast rotation when a frozen chicken was fired into them simulating FOI.
Titanium was used instead. This change of material set back the RR Company in
supplying the engine to Lockheed for its L1011 TriStar, the consequences of which
led to both their bankruptcy: a bleak period in the firm’s history. Originally, this
engine was developed for the Lockheed L1011 TriStar, but later variants including
the RB211-524 went on to power the Boeing 747.
Those early composite blades made of an extremely strong bond between carbon
fibre and epoxy matrix (due to sodium hypochlorite surface treatment of the
fibre [4]) exhibited notch brittleness due to the absence of crack blunting. If only the
process of stitching (z-pinning and tufting) as practised by the South Sea islanders
had been adopted by the aerospace engineers at the Derby plant some 50 years ago.
In the meantime, competitor GE with its GE90 (launched in 1995) and latest
GEnx engines carved its own niche with composite blade technology. Its GE90
engine, equipped with composite blades, won it a monopoly on the 777-200LR and
777-300ER versions of Boeing’s airliner, shutting out competition.
Likewise, the all-carbon fibre composite Lear Fan 2100 exhibited notch brittle-
ness and cracked at the wing root joint on flexing in a full-scale test and, just like
Rolls-Royce, also went bust (Fig. 18.6). There is a more recent example. In the wake
of a JAL Boeing 747 aircraft flying out of New York’s JFK International Airport,
the American Airlines Airbus A300-600 had its vertical carbon fibre fin blown away
with disastrous consequences (Fig. 18.7).
A carbon fibre composite pressure vessel filled with natural gas (200 bars
(20 MPa)) exploded, totally destroying the bus and surrounding cars nearby
18 Slow Cracking in Composite Materials: Catastrophic Fracture. . . 493
®
Fig. 18.5 The HyFil first
stage compressor fan blades
of the RB211 jet engine
shattered as a consequence of
the brittleness of the carbon
fibre-epoxy composite
®
(known by the name HyFil )
under foreign object impact
(FOI). This photograph shows
the 33 blades subsequently
made out of solid titanium.
http://jetpropulsion.co.uk/jet-
engine-projects/rolls-royce-
rb-211-22b [3]
(Fig. 18.8). In the filament winding of a pressure vessel, the fibres are positioned
on geodesic paths to provide maximum strength. Now carbon fibre exhibits a
distribution in tensile strength along its length. In this example, the dominant
failure mechanism is fibre rupture over time. Damage is related to the accumulation
of broken fibres at their weakest points, particularly those fibre breaks in close
proximity, which increases the damage rate. At the point where the vessel’s residual
strength equals the applied stress, the joining up of fibre fractures results in crack
instability (fast fracture) and the vessel explodes. Where the storage of hydrogen is
concerned (700 bars (70 MPa)), absolute reliability is vital. Periodic verification of
structural integrity is essential [5].
494 P.W.R. Beaumont
Fig. 18.8 The NG pressure vessels are carried on the roof of the bus (top and bottom left). A gas
tank explodes catastrophically with the remains shown top right and the consequences shown
bottom right (Auto Fire with Compressed Natural Gas Fuel Tank Explosion, F. Department, Editor.
2007, Seattle. Courtesy of Prof. Tony Bunsell [5])
Afghanistan, the British Army suffered loss of personnel from surface mine blast
impact damage of armoured (composite) fighting vehicles (AFVs). In these few
examples, we see structure suffering a critical loss of function.
In spite of early failures, where high stiffness and low weight, lack of corrosion
combined with durability against fatigue is required of an engineering material,
carbon and other fibre composites are becoming common place. Examples of
applications are legion microlight (man-powered) aircraft, Formula 1 car chassis,
unstayed masts, high-performance racing sails, vaulting poles, squash and tennis
racquets, skis and golf clubs (each of which has raised the modern standards of the
game and sport to new heights), helicopter rotor blades, lightweight construction
and repair in civil engineering. There are many others. There is the B-2 bomber.
496 P.W.R. Beaumont
Military aircraft have incorporated composite components for many years, but
now civil aircraft do so in increasing quantity. Boeing now manufactures the large
civil airliner, the Dreamliner 787 with more than 50 % of the airplane made of
composite. However, in January 2016, an engine on the Japan Airlines 787 flight
from Vancouver to Tokyo had to be shut down due to the accumulation of ice on
the fan blade in GE’s most advanced GEnx engine. Consequently the blades’ tips
rub against the engine casing with possible “damage” resulting in an in-flight non-
restartable power loss. The Dreamliner has been dogged by a series of problems
during development and production as well as since its first commercial flight in the
late 2011.
It is interesting that Boeing is aiming to take the lead since Airbus may
legitimately claim to have pioneered the use of advanced composite materials with
the A300B incorporating them in secondary structures such as tail fin leading edges.
The A340-600 saw the first use of composites in crucial primary structures such
as the rear pressure bulkhead and the keel beam. Other components made from
composites on this aircraft include the fin and rudder, horizontal tail plane and wing
trailing-edge moving surfaces as well as the floor panels in the passenger deck.
The airframe materials by weight of Boeing’s 787 are 50 % composite, 20 %
aluminium, 15 % titanium, 10 % steel and 5 % others. This might be contrasted
with the same breakdown for the Airbus A380, the world’s largest civil airliner
now flying in service since the beginning of 2008, of 22 % composite (or 25 %
if the 3 % of Glare is included), 10 % steel and titanium and 61 % aluminium. In
contrast, Boeing’s 777 contains 50 % aluminium and only 12 % composite. By
volume, the 787 will comprise 80 % composite. Each 787 contains 35 tonnes of
CFRP made from 23 tonnes of carbon fibre. The lighter weight provides greatly
reduced fuel burn, and a side advantage is that high humidity in the passenger cabin
is possible because composites do not corrode like aluminium. Other innovations
are an automatic active gust alleviation system developed for the B-2 bomber.
Boeing has an agreement with Toray Industries to purchase $6B worth of carbon
fibre.
The most striking innovation of the 787 is its all-composite fuselage made by
filament winding four barrel sections joined end to end. This eliminates the use of
50,000 fasteners and allows a higher cabin pressure during flight compared to that
attainable using aluminium. The Airbus riposte, the A350 as a direct competitor,
is constructed of an airframe of 52 % composite, 20 % Al-Li, 14 % titanium, 7 %
steel and 7 % miscellaneous. Once again, an all-composite fuselage but with the
fuselage made of curved longitudinal sections. A composite fuselage offers the
greatest advantage for highly integrated design concepts and greater reduction in
production costs. Much potential is seen in thin-walled sandwich structures leading
to higher bending stiffness than single-skin designs.
The A380 is the first commercial airliner with a central wing box been made of
carbon fibre-epoxy and the first to have a wing cross section smoothly contoured—
which allows maximum aerodynamic efficiency—where thermoplastics are used
in the leading edges of the slats. Glare (GLAss-REinforced fibre metal laminate)
is used in the upper fuselage and in the stabiliser leading edges. Glare has better
18 Slow Cracking in Composite Materials: Catastrophic Fracture. . . 497
fatigue, corrosion and impact resistance than conventional aluminium alloys and
can be repaired using conventional repair techniques. Laser beam welding is used
to eliminate riveting in much of the construction.
The avoidance of the failure of large composite-structured assemblies such as
these and their complexity of design represents a major challenge. Anticipating
possible material and structural problems is just too complicated and too interactive
to admit of solution by practical testing, and in any case, the structures are too large.
It follows that judicious testing coupled with well-substantiated computer modelling
is the only way forwards at present. Bird strike, the fatal nemesis of the brave RB211
fan project, can now be adequately modelled so that the EASA (European Aviation
Safety Agency) will accept bird strike simulation. And some forms of virtual testing
have been accepted. Virtual testing procedure for drop weight impact tests has been
developed using finite element methods. Recent advances allow the inclusion of
complex constitutive equations and their manipulation with fast computers. But the
difference from design and testing with metal structures is still vast. Data sheets
used for metal structures in the certification process are not suitable for composites
because there is too little data, hence no data sheets.
A feature of modern composite materials engineering with large or very impor-
tant new structures is the manner in which university groups and research institutes,
ex-house to the main contractor, are involved both in design, particularly for the
last stages, and in testing activities. As an illustration, the production project for
the end section of the A350 involves the following specific requests for work on
the production method:
– Getting it right the first time and quality control of the manufacturing process.
– Effects of impact damage—with composite impact due to falling or struck objects
(at up to 900 km/h)—do not immediately give visual evidence of the event on
the exterior surface (delamination may be produced and/or peeling at the inner
surface). It turns out that curved sections appear to be more susceptible to damage
than are planar ones.
– Damage tolerance is a most important concept, and much consideration must be
given to coming up with a reliable quantitative statement of how much may be
allowed with safety.
– For both of the last two, NDI method for checking the adhesive bond failure is
important.
Carbon is an unsung but vital engineering material. It has long been manufactured
in many different forms to suit various purposes, but in recent years, with the
development of the nuclear and aerospace technologies, it has acquired a high
degree of sophistication [4].
498 P.W.R. Beaumont
More than 50 years ago, fibre composite materials were available, and their
advantage in terms of lightweight and great strength to conventional materials was
familiar to engineers. Glass-reinforced plastic was available and in terms of strength
per unit weight was superior to aluminium by a considerable factor. Indeed, in
unidirectional applications coupled with one-off uses, such as a rocket motor case,
it was superb and was used in the Polaris missile programme demonstrating that a
real military interest was aroused.
However, there was lacking a truly stiff material, one capable of being woven
and of lightweight. There arose a quickened research interest in composites because
of the discovery at Bell Telephone Laboratories of whisker crystals of the lighter
materials such as alumina, beryllia and silicon nitride. These, whilst still denser
(density of 3–4 Mg m3 ) than aluminium (density 2.6 Mg m3 ), were very much
stiffer with values of Young’s modulus of up to 700 GPa—much larger of that of
aluminium 70 GPa—and hence much stiffer specifically. Asbestos, particularly
chrysotile, had been found also to be stiff, but toxicity prohibited its use and
continues to inhibit it.
In 1958, Talley, working at the Texaco laboratories, deposited boron on tungsten
wires and produced a fibre of stiffness 400 GPa, certainly stiff but too thick to
weave. In the early 1960s, Roger Bacon at Union Carbide grew some whisker
crystals of graphite. We knew that stiff graphite could be produced by hot working
at high temperature and the elastic moduli of graphite crystals were being measured
accurately. Following a Royal Society meeting on new materials (1963), Dr William
(Willie) Watt of the Royal Aircraft Establishment, Farnborough, attempted to make
carbon fibre by carbonising a polymer.
On the advice of Mr Leslie Phillips (coinventor of the RAE carbon fibre), Watt
selected to carbonise the polymer PAN (polyacrylonitrile (CH2 CHCN)n ), which
forms a ladder polymer on heating. The PAN fibre was heated in an oven before
carbonising at high temperature. During this initial (slight) oxidisation stage, the
fibres were wound on a frame to prevent their shrinkage. This is the most significant
step. Bacon had hot stretched a cellulosic fibre and produced a stiff carbon fibre,
which he called Thornel. Akio Shindo in Japan had previously made a fibre a good
deal stiffer than normal textile fibre. Since Japanese scientists had heard of the
efforts in the UK and USA, Shindo also chose to carbonise PAN. Bacon was unlucky
in choosing cellulose. PAN is a much better characterised fibre and is produced in
quantity. Shindo was lucky in getting the company Toray, a major producer of PAN
fibre, to make available the best grade for carbonising.
At about the same time, the Rolls-Royce company developed independently their
®
own carbon fibre composite (HyFil ). Stiff carbon fibre of small thickness and of
“continuous” length could now be made and was easy to weave. In effect that was
the end of whiskers, or was it? Carbon nanotubes are the present day example. The
fact is that the PAN-based carbon fibre fabric was easy to shape and drape was
almost as important as its stiffness and its strength. Shortly afterwards, Stephanie
Kwolek and colleagues at the DuPont Research Station, Wilmington, Delaware,
invented Kevlar® fibre, also easy to weave as a fabric. Fibres were stiff and the
composites very stiff, but would the material be tough?
18 Slow Cracking in Composite Materials: Catastrophic Fracture. . . 499
But by the turn of the decade, within a matter of only 5 years since the
discovery of the modern carbon fibre, disaster struck. The newly designed carbon
fibre composite compressor blades for the Rolls-Royce RB211 jet engine suffer
catastrophic disintegration as a consequence of composite brittleness. The essential
“trick” by which a composite of brittle carbon fibre in a brittle epoxy resin can be
made tough is due to the interaction between the two components. Provided they
break at quite different strains and then, if the matrix fails first, it may be traversed
by a set of innocuous cracks bridged by the fibre. When the composite finally parts,
fibres are then pulled from their sockets in the matrix against restraining forces
acting at the interface. The first of these processes requires the provision of the
surface energy of the cracks and the second the work to be done in overcoming the
sliding friction between fibre and matrix and when both act together provide much
larger values of energy per unit of weight than is provided by a tough metal. In some
cases, the work of fracture (crack propagation resistance) of composite materials can
approach the values shown by steel based on energy per unit volume.
So why do we still encounter materials that crack and structures that fail? It is
because our knowledge is based (almost) entirely on this store of information being
empirical in nature. Basic human instincts, thinking, perceptions and judgements
became misdirected which became the compulsory basis of creative endeavours in
building large structures: empirical understanding of structural behaviour became
representations of a distorted reality. For decades, an “invisible” college of contin-
uum mechanicians studied mechanical behaviour based on an idealisation of what
behaviour is, without any reference to microstructure. The route less travelled by
some practitioners but becoming more familiar today to solving this dilemma has led
to the development of the principles of damage mechanics (constitutive equations
of material behaviour based on physical models of the deformation and cracking
processes). A requirement is that extensive (and time-consuming) test programmes
must reveal all those failure mechanisms that are likely to operate in service. And
for composite materials, the components of damage are complex, and, ideally, they
must be observed directly and not inferred indirectly.
The main damage mechanisms known by experiment involve intra-laminar and
interlaminar cracking, and they must explicitly be taken into account. The behaviour
of the interface elements may be controlled using a simple cohesive crack model,
and the maximum load at failure and absorbed energy can be accurately predicted.
But life is just not that simple. Damage by cracking involves complex non-linear
processes: at the micro-scale; it includes microcracking of the matrix and fibre
breakage, and at the macro level of size, we see the appearance of cracks in large
components. There are just too many possible interacting mechanisms of damage to
understand let alone to build into a single realistic predictive model of catastrophic
structural failure. Prediction really is a problem.
500 P.W.R. Beaumont
Various design methodologies exist, all dealing with critical issues of structure
and all common to the overall design process of production, maintenance and repair.
Superimposed are two items—assessment by nondestructive inspection (NDI) and
safety. Is safety compromised where the fatal flaw(s) in the structure is (are) smaller
than the NDI detection limit? What initial flaw (damage) size (content) is acceptable
in the final structure as a result of the manufacturing process and then the service
conditions? And what is an appropriate inspection period? Fitness considerations of
large structures require though-life monitoring of damage growth.
Whilst many ancient temples and cathedrals have stood the test of time, many
others collapsed prematurely without warning leaving the builder not knowing why.
History is littered with structural disasters where the crucial failure event eluded the
experimentalist. A few of the more recent ones with respect to composite materials
were illustrated above. Yet despite an acquisition of vast collections of experimental
data, information and compelling evidence and an engineer’s intuition based on
“feel” (experience coupled with intelligent observation)—a phenomenology—our
ability to fully understand that long-standing problem of structural failure remains
unresolved.
Traditional empirical design formulations have not done well dealing with these
challenges. Empirical “laws” simply do not have the power of prediction. Successful
prediction of mechanical behaviour of material and design life of a structure requires
detailed information of all possible failure mechanisms across the widest spectrum
of size scale under all sorts of operational conditions. To set up an experimental
programme that covers all eventualities, the scope of the test programme would
502 P.W.R. Beaumont
be immense and unaffordable. Not only would the findings be complicated to sort
out because of the many different test and material variables and the complicated
mechanisms of fracture and fatigue involved, the question of their interaction would
have to be resolved. And material characterisation over one range of temperature
could not safely be extrapolated into another range: a new characterisation would
be needed for every set of composite lay-up, selection of fibre and matrix system,
operating conditions and environment.
To follow the traditional route of engineering design is merely an attempt to
ensure that a highly stressed critical component could not possibly fail within its
design lifetime. This exercise involves deriving over the entire life of the structure;
the spectrum of loads experienced in service and to compare with material test
data; the strength, fatigue and creep data; the stress corrosion cracking rate; etc.—
that phenomenology mentioned above. The next step would be to determine the
component’s dimensions to maintain the design loads and to guarantee the design
life within a margin of safety, allowing for some acceptable risk of disaster. In this
way, we build up to testing larger and larger items until the complete structure
undergoes that multimillion dollar test to destruction. Determining the ultimate
strength of the structure experimentally has it benefits; it helps verify the analytical
methods used to calculate the loads and deformations the structure will have to carry
in service. The reality is there are still many tests to complete from elements to
components to substructures where internal stress fields are too intricate to evaluate
by analysis [7].
Traditionally, practical design methods followed time-consuming, expensive
test programmes to establish damage tolerance certification of the large structure
(Fig. 18.13). The consequences of changing laminate design (fibre orientation,
stacking sequence or ply thickness) or material system along the way would be
disastrous. A heavy price is paid for making mistakes in this approach. Quite simply,
the entire test matrix for material qualification has to be repeated. Furthermore, there
are complications as the result of laminates being heterogeneous elastic bodies con-
taining sites of stress singularity; and secondly, the mechanisms controlling damage
initiation and propagation are non-linear. The likelihood is those mechanisms would
not be known. Direct identification is best.
In contrast, for a metal undergoing fatigue, we know that cracks grow and we
can control life by monitoring the growth of the longest crack. The behaviour under
multiaxial stresses can be quantified by the use of Goodman diagrams. However,
these diagrams are not feasible for a composite structure, and though it can be said
that composite structures are more resistant to fatigue than are metal ones, this is
only strictly the case in a tensile stress—tensile strain situation. In most cases,
in contrast to the metals, when polymer matrix composites are fatigued, damage
takes the form of numerous microcracks predominantly in the matrix material or
at the fibre-matrix interface and often most importantly towards the end of life by
fibre fracture. To begin with, damage can be sustained, but it spreads over time
through significant parts of the bulk material and structure. There is no dominant
crack, and so it is not feasible with present knowledge to access the nature of the
damage simply by microscopic examination. Detection by acoustic emission may be
18 Slow Cracking in Composite Materials: Catastrophic Fracture. . . 503
2 2
(1) (2)
80 20
thousands of 1 or 2 tests
coupon 160 40 complete aircraft,
specimens to or major
1040 300
establish component to
basic laminate 2500 2500 ultimate load or
properties. Current Aspirational fatigue life.
Current
(Based on EF2000)
Future
Fig. 18.13 The experimental challenge [8] (Courtesy of Davies and Ankersen)
possible but sorting out the different cracking mechanisms not so straightforward
[5, 9]. Some suggestions have been made to use the measurement of Poisson’s
ratio of the fatigued structure as a monitor of the damage [Tony Kelly, private
communication]. To overcome this uncertainty, the designer reduces the allowable
stress on the material so that it then becomes overweight or over-cost.
For half a century or longer, those factors that influence the endurance boundary of
the composite material on the one hand and performance limit of the composite
structure on the other have been the subject of a great number of analytical
investigations, validated by precise measurement of critical property data. But
predicting precisely where a crack will develop in a material under stress and exactly
when in time catastrophic failure of the structure will occur remains an unsolved
mystery. Our comprehension of the sustainable damage of a composite material and
the mechanical stability of a composite structure over a wide range of operating
conditions remains restricted.
Selecting the right material system at the very beginning of the design process is
a sensitive issue and requires careful material property profiling. Dimensions must
be consistent with the overall function including minimum weight, and there are
databases for material properties to which designers can refer, for example, the
Cambridge Materials Selector [10]. When it comes to material property profiling,
frequently Young’s modulus E and density are the performance drivers in which
case the material engineer consults a chart having the axes E- with superimposed
504 P.W.R. Beaumont
lines of constant design or merit indices, (a) E/ for a strut, (b) E1/2/ for a beam
and (c) E1/3/ for a plate, where the index is maximised for optimum design.
Additional input in selecting the correct combination of fibre and matrix
might include thermal conductivity and thermal expansion coefficient. For thermal
property profiling, an expansion coefficient mismatch between ceramic constituents
leads to thermal fatigue cracking or ratchetting in metal-matrix composites (MMC).
This requirement constrains selection of fibre reinforcement and protective coatings
for high-temperature application; Ti requires Al2O3 fibres, not SiC fibres. Ceramic-
ceramic composite (CMC) preference is for SiC/SiC or mullite (aluminosilicate
ceramic)/Al2O3 over SiC/Al2O3.
Where toughness is a critical requirement of a material, experience indicates
a practical minimum level of fracture toughness of 10–15 MPa m1/2 in fracture
mechanics units. Toughness, however, is not a unique property of composites, which
complicates things. For example, “blunting” mechanisms stabilise damage: multiple
matrix cracking, fibre bridging of delamination cracks, fibre buckling zones around
notches or holes in compression, etc. Furthermore, the stress concentration factor
around holes diminishes under increasing (and repeated) load because inelastic
(damage) zones develop with an elevation in the local tensile strength. Multiple
fibre fracture and matrix-dominated cracking below ultimate strength allows other
inelastic mechanisms to activate in the matrix and stabilise the effect of damage, and
the failure probability distribution is dramatically modified. Since notch strength
scales with fracture toughness, notch sensitivity is a more robust, useful measure of
material performance.
In service, composites can undergo combined attack from stress and envi-
ronment. The result is the activation of a complexity of atomistic defects and
microscopic flaws, and their accumulation over time will be felt at the component
level of size. Corrosion fatigue degradation of glass fibres in epoxy, for example,
occurs by two rate-limiting phenomena. Hostile species penetrate the composite
through matrix cracks. Reaction with the fibres reduces their strength, and they fail
at the matrix crack front. This is a reaction-controlled stress corrosion cracking
process. On the other hand, for a narrow matrix crack opening, concentration
gradients develop along the crack, and the stress corrosion cracking process
becomes diffusion controlled. The chemically activated kinetics of the processes
are thermally sensitive, so models based on statistical mechanics lead to a rate that
depends upon temperature: In solving this particular problem, the difficulty is that
pure atomistic models on their own breakdown because certain structural variables
(diffusion rates, jump frequencies, chemical activation energies, etc) are not known
or neither easily measured.
In the twentieth century, modern mechanical design evolved with the development
of the continuum theories of mechanics (mathematical and continuum models of
elasticity and plasticity), diffusion and reaction rates. With the advent of computer
18 Slow Cracking in Composite Materials: Catastrophic Fracture. . . 505
power, this resulted in finite element (FE) modelling, numerical analyses and high-
fidelity simulation. This led to an optimisation to minimise cost or to maximise
performance or safety. Thus, the modern designer shows dexterity in the use of
two boxes of tools: mathematics and continuum modelling, from which those
continuum theories and constitutive modelling have evolved (a sort of “distilled
empiricism”), and micromechanical modelling (or physical modelling). Using these
tools to determine constitutive equations relies on knowledge of the rules of material
behaviour. The idea is that the response at one size level is described by one or more
parameters and passed to the next level up (or down).
Hierarchy of structural scales ranges and discrete methods of analysis in
design ranging from micromechanics to the higher structural levels of modelling,
continuum mechanics, etc. (Fig. 18.14). Structural integrity embraces contributions
from materials science and engineering, fabrication and processing technology,
nondestructive inspection (NDI), fracture mechanics and probabilistic assessment
of failure, across a broad spectrum of size scale.
autoclave moulding
design
methodology structural
integrity
system complete
design structure
structural component
interactions testing
element structural
design element
laminate coupon testing
properties ply testing
thickness
weakest
102m 10 om 10 -2m 10-4m 10-6m fibre 10 -6m 10-4m 10 -2m 10 om 10 2m
link statistics
micromechanics
fracture mechanics
damage mechanics
continuum mechanics
filament compression
structural mechanics
winding moulding
modelling
Fig. 18.14 Hierarchy of structural scales ranges; discrete methods of analysis in design ranging
from micromechanics to the higher structural levels of modelling
506 P.W.R. Beaumont
Fig. 18.15 Structural integrity is affected from the micro- to meso- to macro-size levels (Courtesy
of Jančář [12]). Testing and analysis across a size spectrum reflect responses at all structural levels.
This is illustrated with Boeing’s 787 multi-scale composite structure. Macroscopic response of
material and component reflects responses at all levels beneath
of the physical model are not known with any certainty and they are too difficult to
determine or measure with any confidence. They can be estimated only by empirical
means. Knowledge of matrix crack density, for instance, would characterise the state
of damage in a composite under fatigue, but there is no straightforward way of
quantifying it in a large structure.
Such interplay of materials science and engineering is of crucial importance
where composite material properties vary continuously with some internal
parameter that relates to composite architecture in some way. Optimum material
microstructure (and nanostructure) can be forecast and designed rather than found
by trial and error (with the possibility of calamity), whilst maximising structural
high performance and sustainable safe life. Then, when a set of properties is
specified, it should be possible to select a particular lay-up or weave of an
appropriate composite material system, and set of processing conditions, to
meet that specification and provide structural integrity. We require mathematical
formulations that link TD and BU approaches in single design codes that
represent fine-scale phenomena embedded in calculations representing larger-scale
phenomena, tracing damage mechanisms through all size scales.
Beware of premature claims of success: those top-downers who fit engineering
data with a large number of material and experimental parameters and infer their
model are unique and those bottom-uppers who project their mechanism as the
dominant one amongst all others in order to control engineering behaviour. In
reality, is there a mechanism that has not been observed and is there a significant
crack that has gone undetected?
Fig. 18.16 Crack propagation in alumina fibre-alumina matrix ceramic composite showing a
bridged crack and delamination and splitting modes of failure [14, 15]
At the heart of the matter, then, is the matrix or interface crack between
layers bridged by fibre that requires de-bonding to occur in preference to fibre
fracture. In the absence of de-bonding or when the sliding (shear) resistance along
the de-bonded interface is high, crack tip stresses are concentrated in the fibre,
and they decay rapidly with distance from the matrix crack plane. Consequently,
fibres are more likely to snap at or near to the crack plane rather than pulling
out, thus diminishing their vital role in bridging matrix cracks and delamination
cracks. Under these circumstances, the composite would exhibit notch sensitivity
(brittleness). Recall the fate of the original carbon fibre compressor blades of the
RB211 jet engine.
Thus, the critical issue concerning structural integrity of the composite centres on
the extent of this de-bonding mechanism and its dependence on interface properties
and its effect on crack opening and fibre fracture. These contributions on the
dissipation of energy in a stable manner can be derived in terms of the constituent
properties of the fibre and the shear resistance or shear toughness of the interface.
Furthermore, the question of structural integrity concerns the definition of optimum
surface treatment of fibre and optimum properties of any coating or interphase
between the fibre and matrix.
Vital, then, is the nature of the bond and integrity of the interface and possible
thermal stresses and shrinkage effects of the matrix during processing and ageing
in wet and dry environments. Thus, questions surrounding the mechanisms of
mixed modes of splitting and delamination cracking require resolution. Another
consideration is how to include in a physical model the probabilistic nature of
the failure behaviour of composite materials. So, whilst our understanding of the
deformation and fracture behaviour of materials based on defect theory and crack
mechanics has advanced considerably, failure prediction of composite structures on
a macro-scale becomes problematic. At the heart of the problem lies those failure
mechanism(s) best identified by direct observation. But that’s not straightforward
to undertake by any means. It is dangerous to assume a mechanism without direct
evidence or that it is dominant.
18 Slow Cracking in Composite Materials: Catastrophic Fracture. . . 511
Fig. 18.17 Failure mechanisms of interest (Courtesy of Brian Cox, Constantinos Soutis, Mark
Kortschot [e.g. 1–5])
At the heart of the problem lies those failure mechanism(s) best identified by
direct observation. But that’s not straightforward to undertake (Figs. 18.17 and
18.18). The next best thing is by indirect observations; mechanisms can be inferred
by C-scan or by changes in modulus or Poisson’s ratio.
The value of Poisson’s ratio is a more sensitive indicator of the presence of cracks
than is a direct measure of the other elastic constants. It has been argued against this
in that the principal Poisson’s ratio of an aligned carbon fibre composite is very
small and hence changes in its value difficult to detect. One answer to this objection
is that a composite has a number of Poisson’s ratios, and so whilst a very small
one may occur between one particular pair of directions, it will be accompanied
by a much larger value shown when another pair of directions is taken for the
measurements. The latter pair should be chosen. An angle-ply laminate has a very
large value of the principal Poisson’s ratio.
512 P.W.R. Beaumont
(902 /02)2s Ds
(90/0)4s Ds
Fig. 18.18 The specimen (upper picture) has sustained greater damage emanating from the notch
tip in the form of delamination, splitting and matrix cracking than the specimen shown in the lower
picture. Consequently, the local tensile stress gradient is much reduced compared to the specimen
(lower picture) where less damage has been sustained. The result is in an increase in residual tensile
notch strength of the upper specimen over the lower one
dynamics, etc. that we learnt at college. Exploit previously validated models of the
problem using those tools. If parts of the process cannot be modelled, introduce an
empirical fit to the data (e.g. a power law) that can be replaced later by a better
model when it becomes available.
The point is this: physical models suggest forms for constitutive equations
(laws) and for the significant groupings of the variables that enter them. Empirical
methods can then be used to establish the precise functional relations between these
groups. Finally, we finish up with a constitutive equation that contains the predictive
power of micromechanical modelling with the precision of ordinary curve fitting of
experimental data. In other words, input variables, like maximum stress, stress range
or stress amplitude, frequency, etc., temperature, concentration of chemical species
and damage state, are all embedded in the physical model. An example on modelling
structure that evolves with time below may help to illustrate this.
A structure under stress suffers damage with time: cracks form, which in turn
increase the rate of damage formation. There is positive feedback. Ageing weakens
the material, its structure changes and constitutive models (equations) describe
this sort of behaviour, best derived using the internal state variable method.
(Recall: a constitutive model is a set of mathematical equations that describe the
behaviour of a material element subjected to external influence: stress, temperature,
etc.) Constitutive equations take on many forms: algebraic, differential or integral
equations, which may be embedded in a computation (e.g. finite element analysis).
The response equation describes the relationship of (say) current modulus, Ec , of
the laminate (a measure of the effect of damage), to the applied stress, or stress
range, load cycles, N, and to the current value of the internal state variable, D.
We call the internal state variable damage because it describes a change in
the state of a material, brought about by an applied stress or by load cycling. It
(meaning D) uniquely defines the current level of damage in the material, for a
given set of test variables.
In this example, we observe changes in the composite modulus (stiffness) with
the accumulation of fatigue damage. The response equation describes this change
of (damaged) modulus, Ec , to the stress magnitude, temperature and time (number
of load cycles) and to the current value of the internal state variable D:
hand, can be defined as total (meaning actual or measured) delamination crack area
normalised with respect to the total area available for delamination, i.e. D D A/Ao
[16, 17].
Or it might be useful to normalise matrix crack spacing, s, with delamina-
tion crack length ld (i.e. s/ld ), because of coupling effects between these two
mechanisms [18]. But problems arise from a lack of detail of these microscopic
parameters including number (or density) of fibre breaks, which can only be
determined microscopically, and this is not practical in real structures. Instead of
characterising a material property as a function of the independent variables, (which
experimentally is very time-consuming and expensive), fit data to a coupled set
of differential equations, one for the modulus E0 and two (or more) depending on
damaging mechanisms, D1 0 and D2 0 .
Since the internal state variable, D, evolves over time with the progressive nature
of the damaging processes, its rate of change can be described by:
Ds
a 90° 0° 90°
b Modulus, E
Eo
transverse ply
cracking dominant
de-lamination
cracking
fibre
dominant
fracture
dominant
Fig. 18.19 The state variable approach (a) leads to the ideas of a mechanism map or damage
diagram (b), a versatile method of presenting both results of the model and experimental data
[16, 17]. The map displays the response (loss of stiffness), identifies the mechanisms and shows
the extent of the damage for a given loading history. Various failure modes spread, reducing the
section until the fracture load exceeds the residual strength culminating in a cascade of breaking
fibres
appears when stress and temperature or other field variables are nonuniform. Whilst
simple geometries can be treated analytically, using, for example, the modelling
tools of fracture (damage) mechanics, more complex geometries require discrete
methods.
The finite element method of modelling is an example. Here the material
is divided into cells, which respond to temperature, body forces and stress via
constitutive equations, with the constraints of equilibrium, compatibility and con-
tinuity imposed at their boundaries. Internal material state variable formulations
for constitutive laws are embedded in the finite element computations to give
an accurate description of spatially varying behaviour. Ultimately, the aim is to
develop a design tool that incorporates an initial material variability and operating
environment to provide a “knockdown” factor f that corresponds to a specified
probability of failure Pf.
Critical aspects include:
1. Understanding of the expected load and environment for a particular structure
based on a statistical description via a Monte Carlo simulation
2. Development of a database of initial strength based on a Weibull distribution
and residual (fatigue) strength evolution curves (the input) based on a stress
analysis and structure evolution to support the “informed empiricism” of a
residual strength model
Local failure
analysis
Microstructure
Effective properties
Macro elements
Structural analysis
Fig. 18.20 A typical multi-scale challenge: a stringer-stiffened panel fabricated from a textile
composite (Courtesy of John Whitcomb)
Fig. 18.21 All of the pictures are synthesised from the same basic microstructure. The difference
is the degree of repetition of the basic unit and also how closely the material must be modelled
(Courtesy of John Whitcomb)
Fig. 18.22 The need for finite elements smaller than the process zone at a crack tip means less than
1 mm for typical resins. A mesh this fine is unrealistic for a realistic structure. Most codes contain
a local/global strategy whereby an existing coarse (global) finite element model has embedded a
refined (local) region. (Courtesy of G A O Davies [8])
This route leads to a quantitative prediction that represents the actual performance of
the material or component or full-size structure across a broad spectrum of size (or
multi-scale) without the cost, time, accident risk, effort and repeatability, problems
normally associated with real testing. But how much detail of failure mechanisms do
we need to know in order that a successful physical model can be incorporated into
a simulated virtual test that reproduces the outcome of a real structural situation?
Creating physically sound damage simulations is difficult because damage initiation
must first be predicted for parts containing no cracks.
Advances in computer power have made it possible to simulate materials by
describing the motion of each atom, rather than making the approximation that
matter is continuous. Most simulations of cracks, however, ignore the quantum-
mechanical nature of the bonds between the atoms. This limitation is overcome
by using a new technique called “Learn-on-the-fly” (LOTF). This method uses
a quantum-mechanical description of bonding near the tip of the crack, where
essentially it is coupled almost seamlessly to a large (on the atomic scale) region
described with an interatomic potential.
Developments in non-linear elements in computational mechanics have led to
damage simulations of sufficient fidelity in engineering design. A key feature is
the incorporation into a finite element formulation of elements that can explicitly
520 P.W.R. Beaumont
Fig. 18.23 A realistic simulation of a delamination crack modelled by a cohesive zone idealisation
based on the finite element method which computes stress distributions for generic geometry and
loads. (Courtesy of Brian Cox [19])
18 Slow Cracking in Composite Materials: Catastrophic Fracture. . . 521
the fold geometry to enhance the absorption of impact energy. Fold-core has an open
cell structure, a major advantage for an aircraft fuselage compared with conventional
closed cell honeycomb cores, since the core can be ventilated to prevent moisture
uptake.
An example of a new structural fold-core concept for fuselage panels is the
Z-crimp core made from a resin-impregnated aramid or carbon paper, felt or
fabrics. Novel composite sandwich structures with energy-absorbing cores protect
load-bearing composite structures from impact damage. Key issues include the
development of models for folding and collapse in the composite cores, the
development of constitutive laws based on the modelling of the failure mechanisms
and the implementation of the material models into FE codes. Multi-scale modelling
arises because impact damage is localised and requires fine-scale modelling of
failure processes at the micro level, whilst the structural length scales are much
larger. The solution is to use mesoscale models based on continuum damage
mechanics (CDM) in explicit FE codes providing a framework within which failure
processes may be modelled.
There is an increasing amount of work on the cost-effective fabrication of large
composite structures. The driver here is the new “black” large transport aircraft
with wings and fuselage in carbon fibre composite. These materials will only be
competitive if they can be manufactured into very large shapes through automation
production. For example, the one-piece barrel section of Boeing’s 787 Dreamliner
fuselage includes stringers, which require a computerised lay-down of carbon fibre
composite pre-preg tape in its manufacture. The structure is wrapped and cured in
one of the largest autoclaves currently available. This composite fuselage section,
7 m long and 6 m wide, offers potential for integrated design concepts.
Current fuselage structures of the Dreamliner incorporate composites in the
fuselage barrel with conventional frame/stringer concept monolithic skins. But
the potential is for the application of highly integrated design concepts with
complete pressurised fuselages made of carbon fibre composite or hybrid alu-
minium/composite, where production costs are 40 % less and a weight saving of
30 %. And projecting forward, the aircraft industry sees twin-walled sandwich struc-
tures due to their much higher shell bending stiffness and higher strength/weight
ratio. This would allow novel highly efficient fuselage concepts without stringers
and larger frame spacing such as would be found in the blended wing aircraft of the
future.
The Airbus A380 is the first aircraft with a carbon fibre composite centre wing
box made by automatic tape laying (ATL). The wing box is a cantilevered beam
that carries the wing to the fuselage and supports leading-edge and trailing-edge
devices, control surfaces, engines and landing gear. The test piece represents a
portion of the wing section that begins at about the centre of the airplane and stops
at approximately one-half of the span of the wing. It weighs 8.8 tonnes, with 5.3
tonnes of composite, saving over 1.5 tonnes compared to the metallic equivalent.
The main challenge here is the wing root joint, with composites up to 45 mm thick,
which links to carbon fibre composite keel beams, each one 16 m long and 23 mm
thick carrying 450 tonnes. The A380 wing is the largest ever made for a civilian
aircraft. It weighs 35 tonnes but is flexible enough to bend 7 m at its tip.
18 Slow Cracking in Composite Materials: Catastrophic Fracture. . . 523
With the advent of powerful computers and software that can be purchased
at reasonable cost, this means that many of the physical models and computer
simulations that would be cumbersome for design engineers to use could be
implemented as user-friendly computer applications or integrated within com-
mercial finite element design systems. In this respect, mathematical challenges
include hierarchical meshing strategies, which must be coarse enough at the largest
scales (entire structure), whilst cascading down through finer and finer meshes to
atomic scale (if necessary). The real challenge is to formulate design equations
that combine continuum (spatially averaged) and discrete damage representations
through physical (mechanism) models in a single calculation. And at the heart of
the model lies those mechanism(s) best identified by direct observation.
Successful implementation of physical models or simulations requires knowl-
edge of appropriate phenomena such as impact. Multi-scale modelling will arise
because impact damage is localised and requires fine-scale modelling of failure
processes at the micro level, whilst the structural length scales are much larger.
The solution is to use mesoscale models based on continuum damage mechanics
(CDM) in explicit FE codes providing a framework within which failure processes
may be modelled. Key issues currently under investigation include the development
and implementation of constitutive laws for modelling failure mechanisms, models
for folding and collapse in composite sandwich cores and material models into FE
codes.
But how much detail of failure mechanisms do we need to know to come up
with a successful model for incorporating into a simulated virtual test in order to
reproduce the outcome of a real test on an engineering structure? How frequently
do we use NDT or NDE to inspect critical components for cracks? And then, what
decision do we make should we find a crack that might, after all, turn out to have
been innocuous?
A word of warning: whilst carbon fibre is strong, stiff and lightweight, recall the
RB211 story, it does have an Achilles heel. Recently, Professor Nicola Pugno of
the Laboratory of Bio-Inspired & Graphene Nanomechanics, University of Trento,
Italy, fractured his right shoulder due to the sudden breakage of the carbon fibre
frame of his bicycle due to the poor toughness [20]. “In principle you can negate that
poor toughness with hybrid composites like carbon/Kevlar”, said Pugno; “However,
Kevlar is not as strong as carbon fiber, so if Kevlar is used to increase toughness
then strength is compromised”.
Frequently, compromise is the price paid by engineers.
Acknowledgements This paper raises some important issues and topics on structural integrity of
composite materials and structures presented and discussed at two residential Research Discussion
Meetings sponsored by the Engineering and Physical Sciences Research Council (EPSRC) and
the National Science Foundation (NSF). The generous financial support of the EPSRC; the NSF;
the European Office of Aerospace Research and Development (EOARD), London; the Institute
of Materials, Minerals and Mining (IOM3 ), UK; and the American Institute of Aeronautics and
Astronautics (AIAA) in these two meetings is acknowledged.
18 Slow Cracking in Composite Materials: Catastrophic Fracture. . . 527
More recently (January 2016), the support of the Royal Society is acknowledged for its
generous support of speakers and for providing the location of a 2-day residential Research
Discussion Meeting at Chicheley Hall, Buckinghamshire, England, and the publication of papers
in a special issue Phil. Trans. Part A (2016) (to be published).
I have drawn upon material presented at these meetings and benefited from numerous
conversations with the authors attending. In particular, I would like to acknowledge Dr Alastair
Johnson and Dr Brian Cox and Professors John Whitcomb, Josef Jančář, Carlos González,
Constantinos Soutis, Tony Bunsell, Scott Case and Glyn Davies for providing me with figures.
Numerous conversations with Professor Michael Ashby on solving problems of material
behaviour have influenced greatly the research of my group at Cambridge, and some of this work
is included in this chapter. Also, I acknowledge the many valuable conversations with Professor
Tony Kelly who from the beginning encouraged the writing of this book.
References
1. R.A. Smith, I.A. Veres, V.J. Pinfield, Toward the 3D characterisation of GLARE and other
fibre-metal laminate composites, in NDT 2014—53rd Annual Conference of the British Institute
of Non-Destructive Testing, British Institute of Non-Destructive Testing, 2014
®
2. I.K. Partridge, D. Cartie, Delamination resistant laminates by Z-Fiber pinning: manufacture
and fracture performance. Compos. A Appl. Sci. Manuf. 36(1), 54–64 (2005)
3. T. Robinson, Blades of Glory (Royal Aeronautical Society News and Publications, 2012)
4. R.W. Cahn, B. Harris, Newer forms of carbon and their uses. Nature 221, 132–141 (1969)
5. A.R. Bunsell, Composite pressure vessels supply an answer to transport problems. Reinf. Plast.
50(2), 38–41 (2006)
6. H. Sekine, P.W.R. Beaumont, A physically based micromechanical theory of macroscopic
stress-corrosion cracking in aligned continuous glass fibre-reinforced polymer laminates.
Compos. Sci. Technol. 58(10), 1659–1665 (1998)
7. K.T. Kedward, P.W.R. Beaumont, The treatment of fatigue and damage accumulation in
composite design. Intl. J. Fatigue 14(5), 283–294 (1992)
8. G.A.O. Davies, J. Ankersen, Virtual testing of realistic aerospace composite structures. J.
Mater. Sci. 43(20), 6586 (2008)
9. H.-Y. Chou, Damage analysis of composite pressure vessels using acoustic emission monitor-
ing. Ph.D. Thesis, School of Aerospace, Mechanical & Manufacturing Engineering College of
Science, Engineering and Health, RMIT University, 2011
10. M.F. Ashby, Materials selection in mechanical design, 3rd edn. (Butterworth-Heinemann,
Oxford, 2003)
11. S.M. Spearing, P.W.R. Beaumont, Towards a predictive design methodology of fibre composite
materials. Appl. Compos. Mater. 5(2), 69–94 (1998)
12. J. Jančář, Review of the role of the interphase in the control of composite performance on
micro- and nano-length scales. J. Mater. Sci. 43(20), 6747 (2008)
13. A.H. Cottrell, Discussion Meeting on New Materials held at the Royal Society, 6th and 7th
June, 1963, and published in Proceedings of the Royal Society of London. A282, 20th October
1964
14. G. Vekinis, H.R. Shercliff, P.W.R. Beaumont, Dynamic testing of ceramics and ceramic
composites in the SEM. Met. Mater. 7, 279–284 (1991)
15. G. Vekinis, M.F. Ashby, P.W.R. Beaumont, The micromechanisms of fracture of alumina and
a ceramic-based fibre composite: modelling the failure processes. Compos. Sci. Technol. 48,
325 (1993)
16. A.P. Poursartip, M.F. Ashby, P.W.R. Beaumont, Damage accumulation during fatigue of
composites. Scr. Metall. 16, 601–606 (1982)
528 P.W.R. Beaumont
17. A.P. Poursartip, M.F. Ashby, P.W.R. Beaumont, Fatigue Damage mechanics of a carbon fibre
composite laminate: Parts 1 and 2. Compos. Sci. Technol. 25, 193–218 (1986). ibid 25,
283–299
18. R. Dimant, H. Shercliff, P.W.R. Beaumont, Evaluation of a damage-mechanics approach to the
modelling of notched strength in KFRP and GRP. Comput. Sci. Technol. 62, 255–263 (2002)
19. B. Cox, The role of cohesive models and length scales in high-fidelity damage simulations.
Paper presented at the meeting multi-scale modelling of composite material systems, Monterey,
CA, 25–30 September 2005
20. D. Mathijsen, Beyond carbon fiber: what will be the fibers of choice for future composites?
Reinf. Plast. 60(1), 38–44 (2016)
Further Reading
20. P. Beaumont, C. Soutis, Multiscale modelling of the structural integrity of composite materials.
Philos. Transact. A Math. Phys. Eng. Sci. 374(2071), 20150284 (2016)
20. P.W.R. Beaumont, C. Soutis, A. Hodzic (eds.), Structural integrity and durability of advanced
composites (Elsevier/Woodhead Publication, Cambridge, UK, 2015)
20. P.W.R. Beaumont, Advances in multi-scale modelling of composite material systems and
components. Special issue of the J. Mater. Sci. 41(20), ISSN 0022-2461 (2006)
20. P.W.R. Beaumont, Stretching the endurance boundary of composite materials: pushing the
performance limit of composite structures. Special issue of the J. Mater. Sci. 43(20), ISSN
0022-2461 (2008)
20. C. Soutis, P. Beaumont, Multi-Scale Modelling of Composite Material Systems
(CRC/Woodhead Publishing Limited, 2005)
20. B. Harris, Fatigue in Composites (CRC/Woodhead Publishing Limited, 2004)
Chapter 19
Finite Fracture Mechanics: A Useful Tool
to Analyze Cracking Mechanisms
in Composite Materials
19.1 Introduction
In laminated composite materials, damage mainly develops in three forms [1, 2]:
matrix microcracking (intralamina damage), delamination (interlamina damage),
and fiber failures. As reported by Nairn [3], a strength-based approach predicts
that the strain to initiate microcracking in transverse plies is independent of ply
thickness. This result contradicts experimental data which show that a thinnest ply
increases the strain at onset of microcracking [4]. To predict the accumulation
of matrix microcracks within a transverse ply, an energy-based approach which
reproduces this thickness dependence (the thinnest the ply, the lowest the energy
released) must be preferred [5, 6]. Hashin [7] was the first to propose the term “finite
fracture mechanics” (FFM) to analyze the spontaneous formation of a new crack
with a finite surface. It is assumed that the crack develops in a very short time, and
no attempt is made to accurately follow the history of crack development. The crack
formation is governed by the following relation which derives from a basic energy
balance equation:
ıWp
Ginc D D Gc ; (19.1)
ıS
E. Martin ()
LCTS, CNRS UMR 5801, Univ. Bordeaux, Pessac, France
e-mail: martin@lcts.u-bordeaux.fr
D. Leguillon
IJLRA, CNRS UMR 7190, Univ. P. et M. Curie, Paris, France
N. Carrère
LBMS, ENSTA Bretagne, Univ. Brest, Brest, France
where ıWp is the change in potential energy after the onset of a new crack with
a surface ıS and Gc is the fracture toughness. The incremental energy release rate
Ginc thus involves a finite amount of fracture area and not an infinitesimal one like
in usual fracture mechanics [8]. Using FFM requires the evaluation of the crack
increment surface ıS which is usually taken as the ply section when transverse
cracking is analyzed. A different choice relies on the theory of critical distance
[9] which states that the crack increment is a material constant given by 2 Lc which
introduces the characteristic fracture length:
EGc
Lc D ; (19.2)
. c /2
where E is the Young’s modulus, c is the tensile strength, and Gc is the fracture
toughness. To get rid of this strong assumption, Leguillon [10] has proposed to
combine the energy condition with a stress condition: crack onset is predicted if the
energy condition (19.1) is fulfilled and if an additional stress condition (the opening
stress over the expected crack surface exceeds c ) is satisfied. Both conditions are
necessary, and their combination forms a sufficient condition. It will be shown
in the next section how this coupled criterion (CC) links the crack increment to
the characteristic length Lc . Since its introduction, the CC has been demonstrated
efficient to analyze various situations of crack initiation in homogeneous and brittle
materials like V and U notches [11–13]. Prediction of notched strength of composite
laminates is also reported [14–16]. Bonded strength has been analyzed [17–20]
and a recent review can be found in [21]. The required fracture properties can be
obtained from independent tests. A comparable efficiency can also be achieved with
the help of cohesive zone models or other techniques [22], but such approaches may
reveal computationally intensive as they need solving nonlinear problems.
This paper focuses on the use of CC to model failure mechanisms in composite
materials at different scales and is organized as follows: Sect. 2 outlines the coupled
criterion and two formulations are presented. Section 3 considers edge delamination
at the mesoscale, while Sect. 4 concentrates on crack nucleation at the microscale
in the vicinity of an interface.
y
Lh a
x
H
w
Fig. 19.1 A notched sample submitted to a three-point bending test: crack initiation from the
notch tip
where ıWp and ıWk are the changes, respectively, in potential and kinetic energy.
The inequality ıWk 0 provides the relation
ıWp
Gc (19.4)
ıS
ıW
which introduces the incremental energy release rate defined by Ginc D ıSp . It
is worthy of note that the limit ıS ! 0 leads to the differential energy release rate
dW
G D dSp which enters the formulation of the classical Griffith criterion.
The stress condition states that the opening normal stress op along the expected
crack increment ıS must exceed the tensile strength c with
In order to apply the CC to predict crack onset at the V notch (Fig. 19.1), we will
first refer to the full field (FF) approach within the plane elasticity framework. Finite
element computations are thus performed to estimate the incremental energy release
rate and the opening normal stress. Using a two-dimensional model, a refined mesh
is generated at the notch tip. Nodes are unbuttoned to introduce a crack increment of
ıW
length a. The incremental energy release rate is then obtained with Ginc .a/ D a p
where ıWp D Wp .0/ Wp .a/ is the change in potential energy induced by crack
initiation. The energy and stress conditions are formulated with
a σ xx/σ c
inc
G /G
c b σ xx/σ c inc
G /G c
1 1
Fig. 19.2 Schematic illustration of the determination of the crack length at initiation: the ratios
/ c and G/Gc are plotted versus the crack length. (a) The applied load is low and the two
conditions of the CC are incompatible with amin amax > 0. Increasing the applied load
reduces amin amax . (b) The applied load reaches the value and crack initiation occurs with
amin D amax D a
where the dimensionless coefficients (Ā, kxx ) are introduced. The applied stress
depends on the applied load F with D bh FS
2 where (S, b, h), respectively, define
the length, the width, and the thickness of the sample. Numerical results indicate
that kxx (y) is a monotonically decreasing function (as a consequence of the presence
of the stress singularity at the notch tip), while Ā(a) is monotonically increasing.
As schematically plotted in Fig. 19.2a, the energy condition (19.6a) provides a
lower bound amin of the crack increment for a given value of the applied loading.
Conversely, the stress condition (19.6b) supplies an upper bound amax of the crack
increment. Increasing the applied load reduces the lower bound and enhances the
upper bound until the initiation stress * is reached for which amin D amax D a
(Fig. 19.2b). This leads to
A .a / 1 EGc Lc
D D : (19.8)
Œkxx .a /2 h . c / 2 h
As plotted in Fig. 19.3a, the left hand side of (19.8) is an increasing function of
a* vanishing for a D 0. Thus, Eq. (19.8) can always be solved. This relationship
demonstrates that the crack increment a* depends both on the characteristic fracture
length Lc and on the structural geometry. It is to be noted that the fracture length Lc
19 Finite Fracture Mechanics: A Useful Tool to Analyze Cracking. . . 533
12 16
a w=120° b
10 14
A (a)
w=90°
kxx (a)
2
8
σ* (MPa)
12
6 w=60° H/h=0.3
10
4 H/h=0.4
8
2
H/h=0.3 6
0
0 0.5 1 1.5 2 2.5 3 60 70 80 90 100 110 120
3
10 a/h w (degree)
Fig. 19.3 (a) The ratio (19.8) versus the crack length for different values of the notch angle. This
ratio is required to determine the crack length at initiation. (b) The applied stress at crack initiation
versus the notch angle. The material data [67] used for this computation are E D 2:3 GPa, D
0:36, Gc D 393 Jm2 , c D 120 MPa, h D 17:8 mm, b D 12 mm, S D 76:2 mm
q
can also be written as Lc D hs2 where s D 1c EG
c
h
is the brittleness number
introduced by Carpinteri [23]. Once the initiation length a* is determined, the
initiation stress * is given by
s
EGc s c
D Dq c D : (19.9)
hA .a / kxx .a /
A .a /
The studied sample (Fig. 19.1) is a positive geometry [24] which means that the
nucleated crack is unstable: the crack length a* is not the final crack length after
crack onset but corresponds to the first state after nucleation. The propagation of the
initiated crack must be studied with the help of the Griffith criterion which writes
G.a/ Gc where G(a) is the energy release rate. Similarly to relation (19.6a), this
condition can be written as
h
G.a/ D A.a/ 2 Gc (19.10)
E
Z a
1
where A(a) is a dimensionless coefficient. Reminding that Ginc .a/ D a
G.y/dy
0
implies that A(a) and Ā(a) are linked by
Z
1 a
A.a/ D A.y/dy (19.11)
a 0
Z a
1
The energy condition (19.6) can thus be written as a G.y/dy Gc , but this
0
formulation is misleading because it suggests an empirical form (based on an
534 E. Martin et al.
Within the plane elasticity framework, the tensile stress in the vicinity of the notch
tip can be expressed as
y 1
xx .y/ D k0 (19.12)
h
where the singularity exponent depends on the angle ! such that 0:5 1. The
value D 0:5 corresponds to a crack .! D 0/, and the no more singular value D 1
holds for a straight edge .! D /. The dimensionless coefficient k0 is related to the
generalized stress intensity factor and can be computed using a path-independent
integral [25]. Expression (19.12) relies on the first term of a Williams’ series, and
the expected crack length must be small enough to be sure that the higher-order
terms are negligible. The incremental energy release rate is given by
h a 21 2
Ginc .a/ D A0 (19.13)
E h
results obtained from three-point bending test on notched ceramic samples (with a
low value of Lc ) is reported by Yosibash et al. [27]. For a mixed mode stress state in
the vicinity of the notch tip, the crack initiates at an angle which must be determined.
The corresponding MA formulation is given by Yosibash et al. [28], while Hebel
et al. [29] have proposed a FF approach combined with an optimization procedure.
y
z z
+10
y -10
x -10
+10
0.6 1.5
a b
0.5
0.4 1
kxz
0.3
A (a)
0.2 0.5
0.1
0 0
0 0.5 1 1.5 2 0 0.5 1 1.5 2
y/h a/h
Fig. 19.5 (a) The normalized shear stress. (b) The normalized incremental energy release rate
along the (C10/10) interface for a G947/M18 laminate
0.3
0.2
0.1
0
0 50 100 150 200
Lcd /h
where Exx is the longitudinal modulus of the laminate, "xx is the tensile loading
strain, (Ā, kxz ) are dimensionless coefficients, and (Gcd , cd ) are, respectively, the
fracture energy and the shear strength of the interface. As depicted in Fig. 19.5,
kxz and Ā are, respectively, decreasing and increasing functions of the distance to
the free edge.
Applying the CC is thus straightforward. The crack increment at nucleation a* is
given by
A .a / Lcd
D ; (19.16)
Œkxz .a /2 h
Gc
where Lcd D Exx . cd/2 is the characteristic length of the interface. Solving (19.16)
d
shows that the length a* may reach the ply thickness for high values of Lcd as shown
in Fig. 19.6.
The initiation strain "* is derived from
s
"0 Gcd
" Dq with "0 D (19.17)
hExx
A .a /
19 Finite Fracture Mechanics: A Useful Tool to Analyze Cracking. . . 537
0,8 0.8
a b
0,7 0.7
c
s =100MPa
0,6 d 0.6
εe (%)
εe (%)
0,5 0.5
*
*
h
0,4 c
s =50MPa 0.4
d 3h
0,3 0.3
c
s =50MPa
d
0,2 0.2
200 400 600 800 1000 200 400 600 800 1000
-2 -2
Gc (Jm ) Gc (Jm )
d
d
Fig. 19.7 Predictions of the CC for a G947/M18 laminate. (a) Applied strain at onset of delam-
ination versus the interfacial toughness for two values of the interfacial strength. (b) Influence of
the ply thickness on the initiation strain versus the interfacial toughness for dc D 50 MPa and
h D 0.19 mm
650
600
0 1 2 3 4 5
Ply thickness / h
As could be expected, "* increases with Gcd and cd (Fig. 19.7a). Figure 19.7b shows
that an increase of the ply thickness reduces "* in accordance with experimental
results. As plotted in Fig. 19.8, it is possible to identify the values ( cd , Gcd ) giving
the best agreement with the experimental data providing the applied stress at
delamination versus the ply thickness.
It is worthy of note that the first transverse crack in a cross ply laminate under
tension which is also a common cracking mechanism at the mesoscale has been
recently analyzed with the CC [39].
a 2d
Matrix Fiber
Matrix Fiber Matrix Fiber
p
Fig. 19.9 Mechanisms of crack nucleation in the vicinity of the tip of a stationary matrix crack:
a. crack debonding (length 2d) and crack deflection at the fiber/matrix interface and b. fiber
penetration (length p) and crack penetration at the interface
separates before the matrix crack reaches the interface [40, 41]. This is consistent
with another experimental results obtained with model laminates and bimaterials
[42–44]. Description of crack deflection at an interface must take into account the
initiation of an interfacial crack induced by the stress concentration of the primary
matrix crack. This mechanism is depicted in Fig. 19.9a. The rupture of the ligament
` leads to the deviated geometry. This mechanism is also called the Cook and
Gordon mechanism in reference to their paper considering the interaction between
an advancing crack and a weak plane [45]. In order to analyze the competition
between penetration and crack deflection, the fiber penetration in the vicinity of
the matrix crack is also investigated (Fig. 19.9b).
2L
2Rf
2Rm
where (krr , kzz ) are dimensionless factors and Ed is the effective modulus of the
2 1f2 2
1m
fiber/matrix interface defined by Ed D Ef C Em .The incremental energy release
rates (Ginc inc
d , Gp ) for decohesion and penetration are written as
(
2
d .`; d/ D Ad .`; d/ Rf Ed " ;
Ginc
2 (19.19)
Gp .`; p/ D Ap .`; p/ Rf Ef " ;
inc
where (Ād (`, d), Āp (`, p)) are dimensionless coefficients.
Using a finite element model [46], the coefficients (krr , kzz , Ād , Āp ) can be easily
estimated. The CC is now used to describe crack initiation in the vicinity of
the matrix crack. Considering the fiber cracking mechanism, the two conditions
required by the CC are
(
zz .`; r/ D kzz .`; r/ Ef " fc for r p;
2 (19.20)
p .`; p/ D Ap .`; p/ Ef Rf " Gf ;
c
Ginc
where cf and Gcf denote, respectively, the fiber tensile strength and the fiber fracture
toughness. In the case of interfacial debonding, one obtains
where cd and Gcd are, respectively, the interfacial tensile strength and the interfacial
fracture toughness. Plotting (krr (z, `), kzz (r, `)) for a given value of ` reveals
540 E. Martin et al.
3 0.2
a b
2.5
0.15
A (l,d) Ap(l,p)
2
k (l,r) k (l,z)
A (d)
rr
1.5 0.1 d
k (r)
zz
1
d
zz
0.05
k (z) A (p)
0.5 rr p
max
d
0 0
0 2 4 6 8 10 0 2 4 6 8 10
r/R (%) z/R (%) d/R (%) p/R (%)
f f f f
Fig. 19.11 (a) Dimensionless radial stress along the fiber/matrix interface and axial stress in
the fiber. (b) Dimensionless incremental energy release rate for interfacial debonding and fiber
penetration. The data are computed selecting Rf D 7:5 m, Ef D 100 GPa, Em D 10 GPa,
2
m D f D 0:2, Rf =Rm D 0:4, `=Rf D 1:33 102
a b
w
1
d
, e /e
max
*
d
p /Rf , d /d
*
w
p
e /e
*
p
*
1
Lcf Lcd Lcf Lcd
max
, max max
,
L p Ld L
p Lmax
d
1 1
Fig. 19.12 Schematic illustration for debonding and penetration mechanisms. (a) The initiation
length versus the ratio Lc /Lmax . (b) The initiation strain versus the ratio Lc /Lmax
100
PENETRATION
10 s dc
= 0.33
(Jm )
-2
s cf
DECOHESION
c
d
G
1
s dc
=5
s cf
0.1
0.1 1 10 100 1000
c -2
G (Jm )
f
Fig. 19.13 Competition between fiber penetration and interfacial debonding in the vicinity of the
tip of a stationary matrix crack: the penetration/debonding domain is plotted in the plane (Gcf , Gcd ).
The data are computed selecting Rf D 7:5 m, Ef D Em D 100 GPa, m D f D 0:2,
2
Rf =Rm D 0:4, `=Rf D 1:33102 , dc D 500 MPa
s s
Gcf Gcd
"W
p D ; "W
d D (19.22)
Ap Rf ; ` E f Rf Ad .d max ; `/ Ed Rf
Matrix Fiber
Matrix Fiber
The FF approach is used here but the MA formulation for the Cook and Gordon
mechanism is also reported [47]. It is worthy of note that the presence of a fiber
coating can be taken into account to analyze the deflection of the matrix crack [48].
A more realistic description of the crack deflection mechanism should take into
account the propagation of the matrix crack which is here considered as stationary.
Such an analysis is proposed for the penetration mechanism in recent papers
[49, 50]. The CC also proves its ability to predict crack bifurcation in a compressive
layer within a layered ceramic [51]. The debonding at the fiber/matrix interface
under transverse load has been studied with the CC by Mantič [52].
Previous studies [53, 54] analyze the interaction of a matrix crack with the interface
by considering a stationary matrix crack impinging upon the fiber/matrix interface
(Fig. 19.14). Using now the MA formulation, the theory of singularities provides
the asymptotic expression of the stress field in the vicinity of the crack tip:
8 1
ˆ
< rr .z/ D Ed krr z
R
";
f 1 (19.24)
:̂ zz .r/ D Ef kzz r ";
Rf
The matched asymptotic approach also gives the incremental energy release rate
for fiber penetration and interface debonding:
8 21
ˆ
< Ginc .d/ D Ad Ed d
d Rf
"2 ;
21 (19.25)
:̂ Ginc
p .p/ D Ap Ef Rf
p
"2 ;
The singular exponent depends on the elastic contrast between the fiber and the
matrix. Assume isotropic components lead to > 1=2 if Ef > Em and < 1=2
if Ef < Em . In the first case, relations (19.24) and (19.25) show that the CC can
be applied for both mechanisms because ( rr (z), zz (r)) and (Ginc inc
d (d), Gp (p)) are,
respectively, decreasing and increasing functions. The energy conditions read
(
d .d/ Gd ;
Ginc c
(19.26)
Gp .p/ Gf :
inc c
As already mentioned, this result which is obtained within the framework of the
MA approach which is valid for small values of the fracture lengths. It is interesting
to make a comparison with the previous analysis of He and Hutchinson [53]. This
simplified approach considers the differential energy release rates (Gd (d), Gp (p)) at
the tips of the penetrated and deflected cracks with the assumption d D p. The He
and Hutchinson condition for crack deflection is thus
Gcd Gd
: (19.30)
Gcf Gf
544 E. Martin et al.
As discussed by previous authors [55], the two approaches only provide similar
results for D 1=2 and thus for Ef D Em . The criterion proposed by He and
Hutchinson does not consider the influence of the tensile strength ratio which can
play a crucial role as also reported by Parmigiani and Thouless [56] with the use
of cohesive zone models. The whole analysis is carried out assuming a symmetric
debond. However, a recent analysis demonstrates that an asymmetric debond could
be preferential as the mean released energy per unit debond surface is higher for this
configuration [57].
It is to be noted that only the tensile stress component enters the stress condition
(19.24). This assumption may be questionable for the debonding geometry: although
the normal opening stress is prevailing, shear stress is also present along the
interface. To take into account the influence of mode mix [58], the stress energy
conditions are written in the following form for the Cook and Gordon mechanism:
8 r 2 2
ˆ
ˆ rr .`;z/ rz .`;z/
< dc C dc 1; 8z d
Z d (19.31)
ˆ inc
:̂ Gd .d/ d1 Gcd . .z// dz:
0
where rz (z, `) and cd are, respectively, the shear stresses and the shear strength.
The mode mix is obtained with the help of the stress state prior to the crack onset
with
1 rz .z/
.z/ D tan : (19.32)
rr .z/
The definition of Gcd ( (z)) relies on experimental data allowing the identification
of the interfacial fracture energy under various mode mixes [59]. The influence of
mode mix on the initiation of interfacial debonding is expected to be weak at least
for moderate values of the fracture length Lcd which will lead to a small crack length
at initiation.
19.5 Conclusion
parameters (a critical radius and a critical value), and the equivalence with the
strength and the toughness needed by the CC can be established. The CZM
produces a nonlinear response for large values of the interfacial fracture length.
A good agreement between CC and CZM is thus only obtained for small values
of the fracture length as already shown by [63, 64]. The coupled criterion can be
implemented within a periodic representative cell in order to process an increasing
number of cracks. This was done in a recent paper in order to analyze four-point
bending tests performed on samples of an epoxy-type thermoset polymer oxidized
under several conditions [65]. The presented results were obtained assuming a plane
strain state or an axisymmetric stress state. Further work should extend the CC to a
3D analysis. A first attempt to predict interface delamination starting from a wedge
is reported by Leguillon [66].
References
14. J. Andersons, S. Tarasovs, E. Spārniņš, Finite fracture mechanics analysis of crack onset at a
stress concentration in a UD glass/epoxy composite in off-axis tension. Compos. Sci. Technol.
70, 1380–1385 (2010). doi:10.1016/j.compscitech.2010.04.017
15. P.P. Camanho, G.H. Erçin, G. Catalanotti, S. Mahdi, P. Linde, A finite fracture mechanics
model for the prediction of the open-hole strength of composite laminates. Compos. Part A
Appl. Sci. Manuf. 43, 1219–1225 (2012). doi:10.1016/j.compositesa.2012.03.004
16. E. Martin, D. Leguillon, N. Carrère, A coupled strength and toughness criterion for the
prediction of the open hole tensile strength of a composite plate. Int. J. Solids Struct. 49,
3915–3922 (2012). doi:10.1016/j.ijsolstr.2012.08.020
17. N. Carrère, E. Martin, D. Leguillon, Comparison between models based on a coupled criterion
for the prediction of the failure of adhesively bonded joints. Eng. Fract. Mech. 138, 185–201
(2015). doi:10.1016/j.engfracmech.2015.03.004
18. S. Hell, P. Weißgraeber, J. Felger, W. Becker, A coupled stress and energy criterion for the
assessment of crack initiation in single lap joints: a numerical approach. Eng. Fract. Mech.
117, 112–126 (2014). doi:10.1016/j.engfracmech.2014.01.012
19. A. Moradi, N. Carrère, D. Leguillon, E. Martin, J.-Y. Cognard, Strength prediction of bonded
assemblies using a coupled criterion under elastic assumptions: effect of material and geomet-
rical parameters. Int. J. Adh. Adhes. 47, 73–82 (2013). doi:10.1016/j.ijadhadh.2013.09.044
20. P. Weißgraeber, W. Becker, Finite Fracture Mechanics model for mixed mode fracture in
adhesive joints. Int. J. Solids Struct. 50, 2383–2394 (2013). doi:10.1016/j.ijsolstr.2013.03.012
21. P. Weißgraeber, D. Leguillon, W. Becker, A review of finite fracture mechanics:
crack initiation at singular and non-singular stress raisers. Arch. Appl. Mech. (2015).
doi:10.1007/s00419-015-1091-7
22. V. Mantič, Prediction of initiation and growth of cracks in composites. Coupled stress
and energy criterion of the finite fracture mechanics. ECCM-16th European Conference on
Composite Materials, Seville, Spain, 22–26 June 2014
23. A. Carpinteri, Notch sensitivity in fracture testing of aggregative materials. Eng. Fract. Mech.
16, 467–481 (1982)
24. P. Weißgraeber, S. Hell, W. Becker, Crack nucleation in negative geometries. Eng. Fract. Mech.
(2016). doi:10.1016/j.engfracmech.2016.02.045
25. D. Leguillon, E. Sanchez-Palencia, Computation of Singular Solutions in Elliptic Problems
and Elasticity (Wiley, 1987)
26. D. Leguillon, E. Sanchez-Palencia, in Fracture in Heterogeneous Materials, Weak and Strong
Singularities. ed. by P. Ladevèze, O. Zienkiewicz. Proceedings of the European Conference
on New Advances in Computational Structural Mechanics (Elsevier, Amsterdam, 1992),
pp. 229–236
27. Z. Yosibash, A. Bussiba, I. Gilad, Failure criteria for brittle elastic materials. Int. J. Fract. 125,
307–333 (2004)
28. Z. Yosibash, E. Priel, D. Leguillon, A failure criterion for brittle elastic materials under mixed-
mode loading. Int. J. Fract. 141, 291–312 (2006). doi:10.1007/s10704-006-0083-6
29. J. Hebel, R. Dieringer, W. Becker, Modelling brittle crack formation at geometrical and
material discontinuities using a finite fracture mechanics approach. Eng. Fract. Mech. 77,
3558–3572 (2010). doi:10.1016/j.engfracmech.2010.07.005
30. C. Mittelstedt, W. Becker, Interlaminar stress concentrations in layered structures - part I: a
selective literature survey on the free-edge effect since 1967. J. Compos. Mater. 38, 1037–1062
(2004)
31. N.J. Pagano, G.A. Schoeppner, in Delamination of Polymer Matrix Composites, Problems and
Assessment. ed. by A. Kelly, C. Zweben. Comprehensive Composite Materials, vol 2 (Elsevier,
2000), pp. 423–434
32. J.M. Whitney, R.J. Nuismer, Stress fracture criteria for laminated composites containing stress
concentrations. J. Compos. Mater. 8, 253–265 (1974). doi:10.1177/002199837400800303
33. A.S.D. Wang, in Fracture Analysis of Interlaminar Cracking. ed. by N.J. Pagano. Interlaminar
Response of Composite Materials (Elsevier, 1989), pp. 69–109.
19 Finite Fracture Mechanics: A Useful Tool to Analyze Cracking. . . 547
34. R.Y. Kim, S.R. Soni, Experimental and analytical studies on the onset of delamination in lam-
inated composites. J. Compos. Mater. 18, 70–80 (1984). doi:10.1177/002199838401800106
35. E. Martin, D. Leguillon, N. Carrère, A twofold strength and toughness criterion for the onset
of free-edge shear delamination in angle-ply laminates. Int. J. Solids Struct. 47, 1297–1305
(2010). doi:10.1016/j.ijsolstr.2010.01.018
36. B.R. Pipes, N.J. Pagano, Interlaminar stresses in composite laminates under uniform axial
extension. J. Compos. Mater. 4, 538–548 (1970)
37. L. Lagunegrand, T. Lorriot, R. Harry, H. Wargnier, J. Quenisset, Initiation of free-
edge delamination in composite laminates. Compos. Sci. Technol. 66, 1315–1327 (2006).
doi:10.1016/j.compscitech.2005.10.010
38. D. Leguillon, A method based on singularity theory to predict edge delamination of laminates.
Int. J. Fract. 100, 538–548 (1999)
39. I.G. García, V. Mantič, A. Blázquez, F. París, Transverse crack onset and growth in cross-ply
laminates under tension. Application of a coupled stress and energy criterion. Int. J. Solids
Struct. 51, 3844–3856 (2014). doi:10.1016/j.ijsolstr.2014.06.015
40. B.S. Majumdar, D.B. Gundel, R.E. Dutton, S.G. Warrier, N.J. Pagano, Evaluation of the tensile
interface strength in brittle matrix composite systems. J. Am. Ceram. Soc. 81, 1600–1610
(1998)
41. N.J. Pagano, On the micromechanical failure modes in a class of ideal brittle matrix
composites, Part 1. Coated-fiber composites. Compos. Part B 29B, 93–119 (1998)
42. J.A. Bennett, R.J. Young, Micromechanical aspects of fibre/crack interactions in an
aramid/epoxy composite. Compos. Sci. Technol. 57, 945–956 (1997)
43. Y. Kagawa, K. Goto, Direct observation and modelling of the crack fibre interaction process
in continuous fibre-reinforced ceramics: model experiments. Mater. Sci. Eng. A250, 285–290
(1998)
44. L.R. Xu, Y.Y. Huang, A.J. Rosakis, Dynamic crack deflection and penetration at interfaces in
homogeneous materials: experimental studies and model predictions. J. Mech. Phys. Solids 51,
461–486 (2003)
45. J. Cook, J.E. Gordon, A mechanism for the control of crack propagation in all-brittle systems.
Proc. R. Soc. 282A, 508–520 (1964)
46. E. Martin, B. Poitou, D. Leguillon, J.M. Gatt, Competition between deflection and pene-
tration at an interface in the vicinity of a main crack. Int. J. Fract. 151, 247–268 (2008).
doi:10.1007/s10704-008-9228-0
47. D. Leguillon, E. Martin, in Crack Nucleation at Stress Concentration Points in Composite
Materials – Application to the Crack Deflection by an Interface, Chapter 10, ed. by V. Mantic.
Mathematical Methods and Models in Composites, Décembre 2013, vol 5, Computational and
Experimental Methods in Structures (Imperial College Press, 2012), pp. 401–424, e ISBN:
978-1-84816-784-1
48. C. Lacroix, D. Leguillon, E. Martin, The influence of an interphase on the deflection of a matrix
crack in a ceramic matrix. Compos. Sci. Technol. 62, 519–523 (2002)
49. D. Leguillon, E Martin, The strengthening effect caused by an elastic contrast—part I: the
bimaterial case. Int. J. Fract. 179, 157–167 (2012). doi:10.1007/s10704-012-9787-y
50. D. Leguillon, E. Martin, The strengthening effect caused by an elastic contrast—part II:
stratification by a thin stiff layer. Int. J. Fract. 179, 169–178 (2012). doi:10.1007/s10704-012-
9785-0
51. D. Leguillon, E. Martin, O. Ševeček, R. Bermejo, Application of the coupled stress-energy cri-
terion to predict the fracture behaviour of layered ceramics designed with internal compressive
stresses. Eur. J. Mech. A/Solids 54, 94–104 (2015). doi:10.1016/j.euromechsol.2015.06.008
52. V. Mantič, Interface crack onset at a circular cylindrical inclusion under a remote transverse
tension. Application of a coupled stress and energy criterion. Int. J. Solids Struct. 46,
1287–1304 (2009). doi:10.1016/j.ijsolstr.2008.10.036
53. M.Y. He, J.W. Hutchinson, Crack deflection at an interface between dissimilar elastic materials.
Int. J. Solids Struct. 25, 1053–1067 (1989)
548 E. Martin et al.
54. D. Martinez, V. Gupta, Energy criterion for crack deflection at an interface between two
orthotropic media. J. Mech. Phys. Solids 42, 1247–1271 (1994)
55. E. Martin, D. Leguillon, C. Lacroix, A revisited criterion for crack deflection at an interface in
a brittle bimaterial. Compos. Sci. Technol. 61, 1671–1679 (2001)
56. J. Parmigiani, M. Thouless, The roles of toughness and cohesive strength on crack deflection
at interfaces. J. Mech. Phys. Solids 54, 266–287 (2006). doi:10.1016/j.jmps.2005.09.002
57. I.G. García, V. Mantič, E. Graciani, Debonding at the fibre–matrix interface under remote
transverse tension. One debond or two symmetric debonds? Eur. J. Mech. A/Solids 53, 75–88
(2015). doi:10.1016/j.euromechsol.2015.02.007
58. I.G. García, D. Leguillon, Mixed-mode crack initiation at a v-notch in presence of an adhesive
joint. Int. J. Solids Struct. 49, 2138–2149 (2012). doi:10.1016/j.ijsolstr.2012.04.018
59. L. Banks-Sills, Interface fracture mechanics: theory and experiment. Int. J. Fract. 191, 131–146
(2015). doi:10.1007/s10704-015-9997-1
60. E. Martin, D. Leguillon, A strain energy density criterion for the initiation of edge debonding.
Theor. Appl. Fract. Mech. 79, 58–61 (2015). doi:10.1016/j.tafmec.2015.06.011
61. E. Martin, T. Vandellos, D. Leguillon, N. Carrère, Initiation of edge debonding: coupled
criterion versus cohesive zone model. Int. J. Fract. (2016). doi:10.1007/s10704-016-0101-2
62. F. Berto, P. Lazzarin, A review of the volume-based strain energy density approach
applied to V-notches and welded structures. Theor. Appl. Fract. Mech. 52, 183–194 (2009).
doi:10.1016/j.tafmec.2009.10.001
63. C. Henninger, D. Leguillon, E. Martin, Crack initiation at a V-notch—comparison between
a brittle fracture criterion and the Dugdale cohesive model. ComptesRendusMécanique 335,
388–393 (2007). doi:10.1016/j.crme.2007.05.018
64. S. Murer, D. Leguillon, Static and fatigue failure of quasi-brittle materials at a
V-notch using a Dugdale model. Eur. J. Mech. A/Solids 29, 109–118 (2010).
doi:10.1016/j.euromechsol.2009.10.005
65. D. Leguillon, M.C. Lafarie-Frenot, Y. Pannier, E. Martin, Prediction of the surface cracking
pattern of an oxidized polymer induced by residual and bending stresses. Int. J. Solids Struct.
91, 89–101 (2016)
66. D. Leguillon, An attempt to extend the 2D coupled criterion for crack nucleation
in brittle materials to the 3D case. Theor. Appl. Fract. Mech. 74, 7–17 (2014).
doi:10.1016/j.tafmec.2014.05.004
67. M.L. Dunn, W. Suwito, S. Cunningham, Fracture initiation at sharp notches: correlation using
critical stress intensities. Int. J. Solids Struct. 34, 3873–3883 (1997)
Chapter 20
Traction-Separation Relations in Delamination
of Layered Carbon-Epoxy Composites Under
Monotonic Loads: Experiments and Modeling
Nomenclature
20.1 Introduction
The requirements of today’s structural design call for the development of materials
that are lightweight and strong and possess relatively high toughness. Continuous
fiber-reinforced polymers (FRPs) are an important category of materials that attempt
to meet some of these challenging material needs. These materials are formed by
staking plies of different orientations in a tailored manner to obtain the desired
properties in different directions [1].
While these materials demonstrate the best qualities of their constituents and
are stronger and many times tougher than the matrix itself, they tend to fail by
delamination. By definition, delamination is a surface formed between adjacent
plies or within a single ply which may be of the same or different orientation. The
most common source of delamination is the material and/or structural discontinuity
which creates normal and/or tangential stress concentrations between plies. Thus,
delamination can originate from stress-free edges (straight or curved), due to
property mismatch of the individual layers, at ply drops due to thickness reduc-
tion, etc. [2]. Internal delamination also appears in composites after low-energy
20 Traction-Separation Relations in Delamination of Layered Carbon-Epoxy. . . 551
® ®
codes (i.e., Abaqus , MSC Nastran , etc). These methods are well established and
can be used effectively in a variety of configurations when the process zone ahead
of the crack tip is very limited. However, when extensive bridging is present, the
corresponding tractions should be known, and their contribution to the reduction of
the stress intensity at crack tip should be taken into account. Such data on bridging
are not always easy to obtain and should come from experimental observations and
physical arguments. While important progress has been achieved in the experimental
evaluation and numerical modeling of the critical ERR at delamination initiation,
where bridging is absent, damage tolerance design requires the characterization of
ERR during delamination growth accompanied by bridging.
(a) (b)
c
GI,i
c
GI,i GI,b
max
d
c max
c max
Fig. 20.2 Cohesive laws, (a) without bridging and (b) with bridging (not to scale)
the crack tip, so-called cohesive elements, with typical stress-opening displacement
relationship (or traction-separation relations) as shown in Fig. 20.2. Since cohesive
zone models offer tools for exploring the effects of damage on fracture, the form of
the traction-separation relation is very important in modeling realistically a fracture
process using FE simulations. While damage initiates when a peak stress is reached,
the decaying part of the curve indicates a softening behavior and is normally
described in terms of a damage parameter. The area under the curve or part of it
is associated with the ERR for crack initiation GIC D GI;i (Fig. 20.2a). If bridging is
added to the cohesive law, the area under that part of the curve is taken equal to the
ERR due to bridging GI,b (Fig. 20.2b) [14, 29], and the element fails at maximum
opening. Thus, the method with cohesive elements is relatively easy to implement
as long as the traction-separation relation to characterize delamination is known and
no damage develops elsewhere in the composite.
A method to evaluate the traction-separation relation, conventionally called direct
method, uses measurements of crack opening displacement (COD), ı, at the end of
the bridging zone, which are combined with J-integral calculations to determine the
bridging tractions [14, 30, 31]. In this method, however, the length of the bridging
zone should be known for a precise measurement of the CODs. Moreover, the
resulting traction-separation relation, b b .ı/, can be significantly influenced by the
fit of the J - ı data, especially near the crack tip [32, 33]. Another method to deduce
the form of b b .ı/ is by an iterative scheme where strain distribution along the entire
bridging zone direction is combined with parametric FE modeling and optimization
[14]. The strain data are obtained by FBG sensors. Due to the wavelength, decoding
these measurements is absolute and highly precise along the entire bridging zone
length [34].
Independently of the adopted methods to characterize bridging, only limited
works have reported traction-separation relations based on detailed experimental
data in delamination of polymer composites. Thus, in several reported modeling
works with cohesive elements, the implemented traction-separation relations do
554 J. Botsis et al.
not always rely on solid experimental evidence and, for convenience, simple
linear, or bilinear, traction-separation expressions are often considered [32, 35–39].
Moreover, in large-scale bridging (LSB), the specimen size has a strong influence of
the resistance to delamination. While works to account for the effects of specimen
thickness on delamination and bridging have been reported, these important effects
have not been well understood [30, 31, 40–46]. Thus, in terms of delamination
modeling, using cohesive elements, the appended part on the traction-separation
relation (Fig. 20.2b) is geometry dependent [25, 26, 47], and a unified approach to
model scale effects has not emerged.
In this chapter, a summary of results on characterization of LSB in layered
carbon-epoxy composites is presented using an iterative approach to obtain bridging
tractions, as a function of distance from the crack tip. This method relies on strain
data from fiber Bragg gratings (FBGs) and parametric FE simulations. The evaluated
tractions are combined with CODs to obtain traction-separation relations, which
are subsequently used in cohesive zone modeling to predict the load-displacement
response, crack length, and resistance curves. It is shown that the specimen thickness
has a strong effect on LSB and the resistance curve. These effects are also captured
by a micromechanics model using the embedded-cell approach.
The chapter is organized as follows: In Sect. 3, the background on strain
measurements using fiber Bragg gratings inscribed on optical fibers is described.
Section 4 outlines the experimental methods and Sect. 5 the parametric FE and
optimization methods used to the bridging tractions during delamination and
fracture. The cohesive element formulation is summarized in Sect. 6. The results for
interlaminar, intralaminar, and cross ply fracture are presented in Sect. 7. Finally a
micromechanics approach, based on an embedded-cell model of the DCB specimen,
and pertinent results are summarized in Sect. 8.
Recently, Botsis and coworkers [14, 29, 47] reported experimental and numerical
studies on delamination, accompanied by LSB in layered composite materials.
Using embedded fiber Bragg grating (FBG) sensors, on planes parallel to the
delamination, they have evaluated bridging tractions in mode I delamination, using
the strains from the embedded sensors and inverse FE parametric modeling. These
tractions have been appended to a linear softening cohesive model (Fig. 20.2b)
to predict very well the load-displacement curve of the DCB specimen [14, 29].
In addition, using SIF due to a dipole force on the delamination faces [29], the
contribution of the bridging tractions to the delamination toughness of the composite
can be determined. The results of both analyses agree very well and demonstrate that
the proposed semi-experimental method provides an alternative to the existing ones
for bridging tractions identification [29].
20 Traction-Separation Relations in Delamination of Layered Carbon-Epoxy. . . 555
Fig. 20.3 Spectra from an FBG sensor subjected to (a) homogeneous and (b) nonhomogeneous
strain
A schematic of an optical fiber with an FBG and reflection spectra are shown
in Fig. 20.3. The refractive index modulation of the core of a standard optical fiber
acts as a wavelength selector reflecting a narrow wavelength band and transmitting
unaltered all other wavelengths when a broadband light passes though it [48, 49].
The reflected signal presents a peak centered on the Bragg wavelength B0 that
is linked to the induced refractive index modification by B0 D 2neff ƒ0 where
neff is the mean core index of refraction and 0 is the period of the induced
index modulation. When 0 is constant, the grating is said to be uniform. When
such a grating is subjected to a homogeneous axial strain, "z , both neff and 0
are modified resulting in a shift of the Bragg wavelength to B . Assuming that
the two lateral strains on the fiber "x , "y are related to "z by "x D "y D f "z
(where f is the Poisson’s ratio of the fiber), the wavelength difference B D
B B0 , obtained from the peak shift of the spectra before and after loading, is
B =B0 D .1 pe / "z , where pe is a grating gage factor measured experimentally.
If a temperature difference exists, a compensation to B D B B0 should be
added to the above relation [34]. In such cases, the resulting spectrum is simply
shifted to a lower or higher wavelength B if strains are compressive or tensile
in nature (Fig. 20.3a) [50]. Thus, a measure of the peak shift B is enough to
obtain the strain. Note that such measurements can also be carried out at very high
frequencies, i.e., under impact loads [51].
When the sensor is embedded in a nonhomogeneous strain field, the interpre-
tation of the sensor response, as a function of the position along the sensor, is
complicated. Typically the reflection spectrum becomes broader and several peaks
appear (Fig. 20.3b). Note that such spectra modifications may also be due to
lateral strain induced polarization. This form of the signal complicates identifying
556 J. Botsis et al.
Fig. 20.4 Schematic of a wavelength-multiplexed fiber and corresponding spectra when each FBG
is subjected to different homogeneous strain
the appropriate strain distribution on the fiber, especially in cases of high strain
gradients. To analyze such spectra modifications and obtain the strains along a long
FBG sensor, a method has been proposed in the literature [52].
One of the major advantages of the FBG sensors is that it can be multiplexed
[53]. This means that several short FBGs, each one with a slightly different Bragg
wavelength, can be inscribed on the same fiber and be interrogated at the same time.
A typical schematic of an optical fiber with a few FBGs is displayed in Fig. 20.4
[54]. Thus, we can obtain strain data simultaneously in several points of the structure
by monitoring the wavelength changes of each FBG sensor and using: œBi =œB0;i D
.1 pe / ©z;i (with T D 0 for simplicity) where i D 1, 2 : : : , n is the FBG number
along the optical sensor (Fig. 20.4). Using the multiplexed sensors, quasi-continuous
strain distributions can be extracted and used in iterative procedures to study damage
and fracture. However, it should be noted that strain measurements are made on the
fiber itself, while the interest is on the strains in the host material. Thus, in cases
where iso-strain conditions between the sensor and the host structure do not prevail,
numerical models can be developed to analyze and interpret the strain data [50].
In this work, two types of composite plates, namely, unidirectional and cross ply
laminates, have been fabricated. The material used for fabrication of the composite
laminates is a low-temperature cure, high-toughness carbon-epoxy prepreg system
with the commercial name SE 70 from Gurit ST™. The nominal cured thickness of
the prepreg layer is 0.2 mm. In all cases, composite plates are prepared by stacking
20 Traction-Separation Relations in Delamination of Layered Carbon-Epoxy. . . 557
Fig. 20.6 Schematic of the interlaminar DCB specimen with embedded optical fiber (not to scale)
prepreg layers and cured in an autoclave using the standard procedure recommended
by the prepreg manufacturer. In this study, three different types of specimens are
tested:
(a) Unidirectional composite plates for interlaminar fracture testing, with a width
of 25 mm and nominal thicknesses of H D 2, 4, 8, and 10 mm, are produced
(Figs. 20.5 and 20.6). The number of layers is proportionally increased with
the plate thickness to preserve the same fiber volume fraction in the composite
plates of different thicknesses.
(b) Cross ply composite plates are fabricated with an asymmetric lay-up of [0/90]10
and thickness of 4 mm (Fig. 20.8). A 60 mm-long release film A6000 from
®
Aerovac with a 13 m thickness is introduced in the midplane of each plate to
create an initial crack.
(c) The specimens for the intralaminar fracture are prepared from a 10 mm-
thick plate that represents the width of the specimens (Figs. 20.5 and 20.7).
Specimens with thicknesses (i.e, heights) of 6, 10, and 14 mm are cut from such
plates. Afterward, an intralaminar pre-crack (60 mm in length) is introduced in
the cured specimens using a diamond wire saw with a diameter of 130 m.
558 J. Botsis et al.
Fig. 20.7 Schematic of the intralaminar DCB specimen with the glued optical fiber on the upper
surface (not to scale)
Fig. 20.8 Schematic of the cross ply layout for the DCB specimen with embedded optical fiber
(not to scale)
In selected composite plates for the interlaminar and cross ply fracture testing,
single-mode optical fibers (SM28, 125 m in diameter) are embedded. The optical
fibers, embedded in the unidirectional laminates, are aligned parallel to the carbon
fiber direction as illustrated in Fig. 20.6. In the cross ply laminates, the optical
fiber is embedded along the 0ı carbon fiber direction (z-direction in Fig. 20.8),
three layers from the fracture surface. In both the cases, heat-shrinkable PTFE
tubes are used to protect the optical fibers at the two ends of the composite plate.
To ensure an intact interface between the sensors and the surrounding composite
material, the polyimide coating of the optical fibers is removed from the sensor
zone (approximately 30 mm) using sulfuric acid.
20 Traction-Separation Relations in Delamination of Layered Carbon-Epoxy. . . 559
The optical fibers employed in this work contain arrays of ten wavelength-
multiplexed FBGs, each one of 1 mm in gauge length and all spaced at approx-
imately 3–3.6 mm apart. Each sensor has a bandwidth (FWHM) of 1.5 nm and
a minimum reflectivity of 50 %. The corresponding Bragg wavelengths are in the
range of 1520–1565 nm spaced by 5 nm. The cured plates equipped with the optical
sensors are cut into beams using a diamond saw so that the optical fiber is centered
in the specimen width. The exact location of FBGs along the longitudinal axis of
the beam is determined using the optical low-coherence reflectometry OLCR (HP
8504B reflectometer). In this method, the distance of the FBGs to the cut fiber end is
measured and subsequently converted into global coordinates of the DCB specimen.
Note that in intralaminar fracture, similar optical fibers are used and are glued
in the middle of the upper exterior surface of the specimens only before testing, in
order to avoid damaging it during specimen preparation [26] (Fig. 20.7).
With the composite beams, DCB specimens are prepared by bonding steel
®
loading blocks using an epoxy adhesive from Araldite . These blocks have sizes
of 10 25 10 mm (interlaminar and cross ply specimens) and 10 10 10 mm
(intralaminar specimens) with the loading pin inserted in their middle. One side face
of each specimen, normal to the pre-crack plane, is painted white and marked with
black thin lines one millimeter apart. The latter serves to provide a clear image of
the crack-tip position during the fracture tests. The material properties are measured
using standardized procedures [26, 27].
P2 @C
Gtotal D GI;i C Gi;b D (20.1)
2B @a
560 J. Botsis et al.
where P is the applied load, B is the specimen’s width, C is the compliance, and a is
the crack length. GI,i refers to the ERR at initiation and Gi,b is the contribution
of the toughening mechanisms in the bridging zone. Use of Eq. 20.1 requires
experimentally acquired values of C and a in a fitting scheme to produce the
compliance vs. crack length function. Conventionally, two empirical approaches
are implemented to express the compliance vs. crack length. The first is referred
to as the compliance calibration (CC) method [55], where C D Kan where K and
n are fitting parameters, and the second as the modified compliance calibration
(MCC) [55], where the compliance is fitted by C D .a=HA1 C A2 /3 where H is
the specimen thickness and A1 ; A2 are fitting parameters.
After the fracture tests, transverse and longitudinal sections of the specimens
are prepared by standard polishing techniques. These sections are examined under
an optical microscope to reveal the morphology of the crack growth path and its
surrounding microstructure. Fracture surface of selected specimens are carbon-
coated and examined using scanning electron microscopy.
bridging parameters. Thus, a linear elastic analysis can be used to model LSB in
the carbon-epoxy system used in the present studies. It should be also noted that
the only experimentally obtained ERR values used in the identification are the
initiation values GI,i . The experimental values of Gtotal using Eq. 20.1 are used only
for comparison with the ERR obtained with the identified tractions using linear and
nonlinear geometric effects.
In the reported studies, the distributed strains along the embedded FBG are used
to characterize the bridging tractions by an iterative numerical identification model
including a parametric form of the bridging tractions. It is assumed here that the
FE model provides an accurate and unique correlation, for a given crack length
a between imposed displacement, , the bridging tractions b (z), and the strain
distribution "z (z) along the position of the FBG sensor.
It is mentioned earlier that during delamination the spectrum of each of the FBG
sensors is displaced without changes in its form (Fig. 20.3a) [50]. This indicates that
the strain over its length of 1 mm is homogeneous. Moreover, due to the decoding
in the wavelength domain, there is no need for a reference and the precision of all
sensors is the same over the length of the FBG array. This feature is very useful
since the high accuracy and precision of strain data is not limited to the crack-tip
zone but all along the bridging zone.
Based on experimental observations of the fiber bridging in DCB specimens
and physical considerations, the following simple parametric form of the bridging
function can be selected [14]:
where ’ denotes the vector of unknown parameters, b (z) is the bridging stress
normal to the delamination surface, z is the position from the crack tip, and
is the exponential parameter defining the degree of nonlinearity of the bridging
traction profile (Fig. 20.9). Parameter A1 defines the maximum bridging stress at the
crack tip, max , and the ratio of – A1 /A2 the maximum length zmax , of the bridging
zone. Since these parameters have a physical meaning, limits have been imposed
to their values as follows: 0 A1 D max m ( m is an effective strength of
the composite material); 0 –A1 =A2 D zmax a –a0 (where a0 is the initial
crack length). For convenience, the vector of unknown parameters is rewritten as
’ D Œmax ; ; zmax . Thus, the bridging tractions take the form:
Fig. 20.9 Schematic of the numerical model used in the optimization scheme for the symmetric
case of the DCB geometry (not to scale)
The parametric form of the normal bridging tractions has been introduced in
the FE model as a user-defined pressure distribution applied to the portion of the
delamination surface between the crack tip and the pre-crack tip as illustrated in
Fig. 20.9.
An error norm describing the discrepancies between the simulated and measured
strain distributions, interpolated on a linear grid of points zi (i D 1, n) denoted,
respectively, "Qz .zi ; ’/ and "z (zi ), is defined by the following least-squares objective
function F(’):
1
F.’/ D kf .z; ’/k2 with f.’/ D Œf"1 .z1 ; ’/ ; : : : ; f"n .zn ; ’/ ; wP fP .’/ ; wJ fJ .’/
2
(20.3)
.z .z /
in which f"i .zi ; ’/ D "Qz i ;’/"
"z .z/
z i
(i D 1, 2, : : : ,n) represents the relative error in
local strain distribution and "z .z/ a reference strain value for normalization which
is set equal to either the mean value of the strains h"z (z)i or to an individual strain
Q J .’/G
measurement "z (zi ). The terms fP .’/ D P.’/P P
and fJ .’/ D tip GI;i I;i represent
optional regularization terms measuring the relative error between the experimental
applied load P and the reaction force at the loading point in the model PQ .’/ and the
error between the experimentally determined critical energy release rate in mode I
at initiation GI,i and the J-integral evaluated at the crack tip of the model Jtip (’).
These regularization terms all represent measures of the static stress and strain field
at the critical state corresponding to the identification condition and can optionally
be used to ensure a unique and well-defined minimum in the minimization of the
error norm F(’). The associated weight factors wP and wJ are either pset to zero to
disable the regularization or set typically to a value in the range of n (where n is
the dimension of the strain data vector) to have the same relative weight as the strain
error terms f"i (zi , ’) in the global error norm F(’).
Given these definitions, the identification of the bridging tractions can be reduced
to the following optimization problem:
20 Traction-Separation Relations in Delamination of Layered Carbon-Epoxy. . . 563
(e) The calculated value of JI,b is added to GI,i and the results are compared with
the Gtotal obtained experimentally (Eq. 20.1). Note here that the experimental
value of Gtotal is not used in the iterative method for the calculation of b
b .ı/.
The method is summarized in Fig. 20.11.
To predict the load-displacement response during the entire fracture process, sim-
ulations using cohesive elements are carried out. The traction-separation relation,
b
b .ı/, obtained from the optimization in Sect. 5, is appended to a standard cohesive
element zone as with properties shown schematically represented in Fig. 20.10 and
summarized in the following equation:
8
ˆ
< 0ı .ıı
ˆ for 0 < ı ıc
c/
.ı/ D .1 D .ı// K0 ı with D .ı/ D ı.ı1 ıc / ; for ıc < ı ı1 ;
1
ˆ (20.4)
:̂ 1 bb .ı/ ; for ı < ı ı
1
K0 ı max
and ı1 D ıc C 1 max c
ı d ıc
The cohesive law described in Eq. 20.4 consists of three discrete regions: (a) an
initial linear response with a stiffness K0 , until a specified threshold, c , where the
damage D(ı) is introduced; (b) a linear softening part as described by Camanho
and Dávila [59] with a small modification to adapt the maximum opening, ı 1 , to
correspond to the maximum calculated bridging traction max ; and (c) a third part
built by the identified function b
b .ı/ which is shifted from zero to ı 1 .
20 Traction-Separation Relations in Delamination of Layered Carbon-Epoxy. . . 565
In all created models, the geometry of the DCBs is discretized with quadratic
plane strain elements in the regions of the composite beams, while the fracture
region is modeled using linear four-node cohesive elements. An Abaqus “Tie”
constraint is applied to tether the cohesive element zone with the elements of the
DCB arms. This technique allows for a very fine mesh for the cohesive zone, while
the element size of the rest of the specimen remains within reasonable range.
The model size varies, depending on the fracture mechanism and the different
thicknesses since an increase in size is necessary due to longer bridging zone
lengths in relation to thickness. In all cases, enough elements ahead of the crack
tip are implemented to allow for precise decay of the strains. This region can reach
30–60 mm of DCB length depending on the thickness of the tested specimens.
20.7 Results
Resistance curves for three thicknesses (H D 6, 10, and 14 mm) are shown in
Fig. 20.12 [26]. The data in Fig. 20.12 are obtained using the MCC method and
are the averages of at least 4 specimens per thickness. The error bars indicate
the standard deviation of experimental data per specimen thickness. Note here the
fluctuations in the R-curves which are common in intralaminar fracture.
The strain’s profile vs. the crack length for the thin specimen (H D 6 mm) is
depicted in Fig. 20.13. Note that, contrary to the case of interlaminar fracture
(Sect. 7.2), the recorded strains are compressive and the crack tip, located on the
rising part of the strain data at about 2/3 of the maximum absolute strain, is not
discernible (as verified by the numerical simulations). This is because the optical
fiber is placed on the top surface of the DCB specimen (compressive strains) and far
from the crack tip.
The axial strain profile of the specimen with H D 10 mm is non-smooth and
discontinuous [26], originating from fluctuations on the crack’s surface, which
causes a variation in beam thickness of about 2.5 %. This behavior may appear
in some specimens of all three investigated thicknesses, with negligible effects on
the load-displacement response and the R-curves. The aforementioned change in
thickness translates into a stiffness variation of 7 %, taking into account the area
moment of inertia of the beam’s section, proportional to (H/2)3 . This behavior is
attributed to the generation and the failure of bridging bundles which are in general
greater that those in interlaminar fracture (Sect. 7.2).
To account for uncertainties in the crack-tip position, the crack length is
allowed to vary by D ˙ 2 mm in the optimization scheme for the bridging
566 J. Botsis et al.
Fig. 20.12 Intralaminar experimental R-curves compared to numerically predicted ones, from the
cohesive models. Typical fractured specimens at the end of the experiments are shown in the insert
93
97
-2000 101
105
-2500 109
-3000
-3500
-4000
0 20 40 60 80 100 120 140
Crack length (mm)
tractions. This parameter can also absorb the difference between the actual 3D crack
propagation and the 2D model originating from a bowing crack front. To enhance
the robustness of the analysis, the fracture toughness at initiation, GI,i , is employed.
Hence, the optimization algorithm minimizes the norm of the residual error of
the experimental strains vs. the numerical ones, plus the error of the numerically
calculated J-integral [60] at the crack tip, Jtip , and the experimentally evaluated GI,i .
®
The Jtip is calculated using the Abaqus v6.12 contour integral tool [61], which
uses the numerical definition of the domain integral formulation in absence of body
forces, thermal strains, and crack face tractions [62].
The formed error vector, optimized by the trust-region reflective Newton’s
algorithm to solve the constrained nonlinear least-squares problem, is given by:
" #
"Q z "z Jtip GI;i
f .max ; zmax ; ; / D ; (20.5)
"z GI;i rankŒ"z
To construct the traction-separation relation, bb .ı/, the CODs acquired from
the numerical model, ı(z), are correlated with the identified b (z) (Eq. 20.2b) to
obtain b b .ı/. The traction-separation may also be obtained with the use of b (z)
and the appropriate weight function [63]. Since in the latter case, a numerical
integration is necessary, ı(z) from the optimized solution is used for simplicity.
Selected experimentally measured values of CODs are compared with computed
values and the differences are negligible.
With the traction-separation relation known, the total ERR, for the generic case
Jtotal , due to crack growth in the presence of bridging is calculated using the
following relation [60]:
Z ımax
Jtotal D Jtip C JI;b D GI;i C b
b .ı/dı (20.6)
0
Here Jtip is the fracture toughness at initiation, taken equal to GI,i measured
experimentally, and JI,b is the contribution of bridging to Gtotal , expressed by the
integral of the traction stresses, b
b .ı/, on the crack plane with ı max representing the
COD at the end of the bridging zone z D zmax . The ERR in Eq. 20.6 corresponds to
a general elastic material model without neglecting geometrical nonlinearities, the
effect of which is assessed at the end of the analyses, as described later in Fig. 20.16
([26]) and is considered insignificant.
For the optimization procedure of the specimen with H D 6 mm, a symmetric
plain strain numerical model is adopted as is schematically shown in Fig. 20.9.
However, for the H D 10 mm specimen, a symmetric model is insufficient, due to
variations of the fracture surface during propagation [26]. Therefore, the entire DCB
specimen is modeled, with appropriate adaptations on the beams’ thicknesses, using
discrete thickness measurements in order to follow the actual fracture plane. The
specimen with H D 14 mm is not instrumented with FBGs and thus no optimization
is carried out. However, a numerical model is created to extract the COD profile,
568 J. Botsis et al.
necessary for the evaluation of the traction-separation relation and the energetic
contribution of bridging. For this purpose, a symmetric model is sufficient. In all
numerical simulations the initial height loss due to the diamond wire pre-crack
is taken into account in order to reduce the error between the numerical and the
experimental specimens’ compliance.
The corresponding calculated bridging traction parameters (Eq. 20.2b) for the
two instrumented specimens are max D 8.19 MPa, zmax D 25.5 mm, D 0.12 mm1
(H D 6 mm), and max D 8.39 MPa, zmax D 40.5 mm, D 0.082 mm1
(H D 10 mm). Interestingly, a practically thickness-independent maximum bridging
traction max is identified. It is worth noticing that a thickness-independent max is
also evaluated in interlaminar fracture (Sect. 7.2).
On the basis of these results and earlier studies [25, 45], max can be considered
as a material constant for the particular fracture mechanism, reflecting an average
characteristic strength of the microstructure around the crack tip. When compared
with the corresponding parameter in interlaminar fracture on the same material, it is
about 6 times greater reflecting the differences on the created bundle cross-sectional
areas. The bridging tractions in the H D 14 mm specimen are established by utilizing
the identification results of the H D 6 and 10 mm specimens. The procedure is
based on the thickness dependence of bridging tractions in interlaminar fracture
of glass polyester [45] and the following results for the carbon-epoxy system for the
interlaminar fracture, and it is described as follows:
(a) The maximum bridging traction max is considered as a characteristic parameter
of the fracture mechanism at the crack tip and independent of thickness. Thus,
it is taken as the average value of the other two identified thicknesses.
(b) The relationship between zmax and H is considered linear [25, 26].
(c) The exponential softening parameter and the corresponding thickness satisfy
the relation H D const for each specimen thickness. Therefore, based
on the above considerations, the bridging parameters for H D 14 mm are
max D 8.29 MPa, zmax D55.5 mm, D 0.059 mm1 . These parameters are used
in a numerical model to obtain the corresponding b b .ı/. The calculated traction-
separation relations bb .ı/ for all thicknesses are illustrated in Fig. 20.14 [26]
based on the bridging profiles b (z) coupled with the CODs, acquired from the
optimized numerical models to construct.
The traction-separation relations for the three thicknesses identified earlier,
shown in Fig. 20.14, are used in a cohesive zone model to predict the load-
displacement response and R-curves from the modeled specimens. In this frame-
work, the DCB arms are modeled by quadrilateral, quadratic plain strain elements
with reduced integration (Abaqus CPE8R) and the cohesive layer by modified-for-
zero thickness quadrilateral, linear, cohesive elements (Abaqus COH2D4).
The predicted and experimental load-displacement curves are illustrated in
Fig. 20.15 [26]. It can be observed that the maximum load is moderately overes-
timated in all three thicknesses, but the post-peak response is very well predicted
by the adopted cohesive model. The aforementioned overestimation around the
20 Traction-Separation Relations in Delamination of Layered Carbon-Epoxy. . . 569
8
H=6mm, Idenfied
H=10mm, Idenfied
Bridging Tractions, σˆb(MPa)
H=14mm, Esmated
6
0
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5
Crack opening, δ(mm)
350
H=6mm, Experimental
H=6mm, Numerical
300
H=10mm, Experimental
H=10mm, Numerical
250 H=14mm, Experimental
H=14mm, Numerical
Load (N)
200
150
100
50
0
0 5 10 15 20 25 30 35
Displacement (mm)
Fig. 20.15 Experimental and numerically obtained load-displacement response for intralaminar
fracture
peak load may be attributed to changes in the growth pattern of the bridging zone,
i.e., from transitional growth to a steady propagation, while in the model a single
traction-separation, identified at the steady state, is used throughout the fracture
process.
570 J. Botsis et al.
The experimental R-curves, using the MCC method, for the three different
thicknesses, are represented in Fig. 20.12. These curves are the averages of at least
four specimens per thickness, while the error bars represent the standard deviation
of the total number of experiments per thickness. In Fig. 20.12, the experimental
results at the steady state are compared with the R-curves calculated based on the
aforementioned load history prediction from the cohesive models and by defining
the crack advance tracing the crack tip in the numerical models by the last intact
cohesive element (i.e., D D 0), using the MCC method. The predicted values show
a good agreement with the experimental results of the corresponding specimens.
Similar to the load-displacement curves (Fig. 20.15), the transitional part is slightly
overestimated by the simulations as already explained earlier.
As it has been described, the inverse characterization technique is a generic
tool allowing nonlinear effects to be included. In Fig. 20.16 the ERR calculated
using the identified traction-separation relations and the initial fracture toughness
as described in Eq. 20.6 is compared with R-curves formed using the experimental
data for segmented CC framework, which has better accuracy on the fitted equation,
as described in Pappas and Botsis [26]. As it can be observed, the two approaches
are very close indicating that any presence of nonlinearities is of insignificant effect.
These experimental R-curves compared with the corresponding ones for delami-
nation display a tendency for fluctuations. This behavior is connected with the size
Fig. 20.16 Intralaminar segmented experimental R-curves compared to the ERR calculated from
the numerical integration for the optimized models. Fracture surfaces and transverse section
(H D 6 mm) shown in the insert
20 Traction-Separation Relations in Delamination of Layered Carbon-Epoxy. . . 571
of the generated bundles (insert in Fig. 20.16), which are generally greater than those
of the interlaminar fracture of the cross ply and definitely greater than the ones of
the unidirectional laminate [25, 27].
A comparison of the ERR results at steady-state propagation, with the corre-
sponding interlaminar values, indicates that it is about 2.2–2.3 times higher than the
interlaminar values for the same material and similar beam thicknesses (Sect. 7.2
and Fig. 20.17a). Analogous trends have also been described in other recent works
[64, 65]. However, the initiation toughness is practically the same: GI,i D 285 J/m2
(intralaminar) and 300 J/m2 (interlaminar) using the CC approach in both cases and
260 J/m2 for the MCC. The resulting R-curves in Fig. 20.12 illustrate an increasing
trend of the ERR at the steady state with steps of 20 % per thickness, suggesting a
strong thickness effect in maximum fracture toughness, similar to the interlaminar
results and previous work of Manshadi et al. [45]. The overall shape of the R-curve
also varies over the different specimen thicknesses with a steeper initial slope for the
thinnest specimen. This behavior has also been previously reported and is explained
by the increase in bridging zone length along the thickness [30, 46].
ERR (J/m2)
1000
800 600
Specimen thickness (mm)
600 As produced 4mm
10 400
8 10 mm milled down to 4mm
400 Gic
Gic 4 8 mm milled down to 4 mm
2 200
200
0 0
0 20 40 60 80 100 120 140 160 180 0 10 20 30 40 50 60 70 80
Crack advance (mm) Crack advance (mm)
Fig. 20.17 (a) Representative R-curves obtained from the interlaminar DCB specimens of
different thickness (b) R-curves obtained from the 4 mm-thick specimens produced from the
laminates with other original thicknesses. The results are compared with the as-processed 4 mm-
thick specimen
572 J. Botsis et al.
Fig. 20.18 Strain monitoring with 19 multiplexed FBGs during the DCB test of the 8 mm-thick
specimen. The arrow indicates the crack-tip position registered by the sensor
Fig. 20.19 Identified traction-separation relation representing fiber bridging, for each specimen
thickness. The insert shows representative photographs of the tested DCB specimens
With the bridging tractions distribution b (z) known, the corresponding crack
opening displacement profiles ı(z) are extracted from the optimized FE models.
The identified traction-separation relations bb .ı/ representing large-scale bridging
in the specimens of different thicknesses are shown in Fig. 20.19 [25]. Note that the
surface area under each curve in Fig. 20.19 reflects the ERR due to the bridging
in each specimen (Eq. 20.6). The identified traction-separation relations clearly
demonstrate a more pronounced bridging effect in delamination of the thicker
specimens. The curves begin from practically the same traction at the crack tip while
decaying to zero with different rates. Therefore, the effect of the specimen thickness
can be explained by the change in the rate of traction decay due to the change of
the beam’s curvature in the bridging zone. The latter also implies that a so-called
bridging law cannot be considered a material property.
The identified traction-separation relations for the specimens of different thick-
nesses are included in the cohesive zone model shown in Fig. 20.10 to simulate the
loading response of the specimens in delamination tests. To this end, the geometry of
the DCB specimens are reproduced in 2D plane strain models. Eight-node quadratic
®
elements with reduced integration (Abaqus CPE8R) are employed to discretize
®
the DCB arms. A single layer of cohesive elements (Abaqus COH2D4) with
a thickness of 20 m is positioned along the physically observed delamination
path. The cohesive layer is distanced from the loading point by the pre-crack
length plus the initial crack pop-in. The load-displacement curves simulated by
20 Traction-Separation Relations in Delamination of Layered Carbon-Epoxy. . . 575
Fig. 20.20 Experimental load-displacement curves and corresponding simulations with the iden-
tified traction-separation relations
cohesive zone models with the identified traction-separation relations are shown
in Fig. 20.20. Also shown in Fig. 20.20 are the load-displacement curves measured
from the DCB specimens of different thicknesses. The good agreement between the
simulation results and the experimental data confirms that the identified traction-
separation relations can address the geometry dependence of large-scale bridging
and consequently predict well the loading response to delamination in the specimens
of different thicknesses.
Fig. 20.21 Photographs of (a) a delaminated cross ply specimen showing bridging ligaments and
polished (b) transverse and (c) longitudinal sections
The array of FBG sensors are centered in the specimen width and placed at three
layers from the pre-crack plane along the longitudinal axis of the specimen (i.e.,
z-direction in Fig. 20.8). In total six specimens are tested with the same conditions
under monotonic displacement-controlled loading. A representative photograph of
a loaded cross ply specimen is shown in Fig. 20.21a [27]. After testing, transverse
and longitudinal sections are polished and examined under an optical microscope,
to reveal the morphology of the crack path and the surrounding microstructure (see
Fig. 20.21b, c).
Delamination in the present cross ply specimen is analyzed in three steps:
(1) in the first step, crack initiation in the cross ply specimen is considered.
Fracture mode mixity at crack initiation is computed using the VCCT method.
The crack propagation direction at the onset of delamination is analyzed with
the XFEM method. (2) In the next step, a parametric FE model, representing
the effect of the steady-state bridging tractions on strains sensed by FBGs, is
used in an inverse identification scheme to determine bridging traction parameters
from the experimental data. (3) Finally, the identified traction-separation relation is
implemented in a cohesive zone model to simulate crack propagation and compare
the results with the experimental data. In all the cases, plane strain FE models of
the DCB testing configuration are built using the geometry and boundary conditions
taken from the experiments. Due to the asymmetry of the laminate layout, both arms
of the specimen are modeled, and each composite ply is considered separately to
obtain an accurate distribution of strain/stress through the thickness of the specimen.
To this end, the specimen thickness is partitioned in 20 sections (corresponding to
the number of plies) with alternating material properties of the 0ı and 90ı layers.
Experimental observations indicate that crack direction in the cross ply specimen
deviates from the direction of the crack starter at the 0ı /90ı interface into the
20 Traction-Separation Relations in Delamination of Layered Carbon-Epoxy. . . 577
Fig. 20.22 Strain monitoring during the delamination of the cross ply specimens
neighboring 90ı layer where it propagates and is confined by the two adjacent 0ı
layers. It is observed that crack propagation in the cross ply specimen only occurs
in the 90ı layer, adjacent to the pre-crack plane, and is accompanied by fluctuations
from a planar growth direction (see Fig. 20.21c).
The strain measurements with the array of ten sensors during the fracture test
are depicted in Fig. 20.22 [27]. Each curve represents the response of an individual
FBG to crack growth during the test. Note that the steep rise of each strain curve
reflects the crack approaching an individual sensor and the local jumps are due to the
stick-slip crack propagation in the cross ply laminate. After specimen testing, the
measured Bragg wavelengths are compared with the corresponding values before
testing in order to assess the release of the residual strains as a result of fracture.
The average measured R-curve from the cross ply DCB specimens with
H D 4 mm is shown in Fig. 20.23 [27]. For comparison the R-curve of unidirectional
specimens with the same dimensions is also shown. The obtained experimental
results for the unidirectional and cross ply layouts indicate practically identical
crack initiation ERR values (GI,i D 310 ˙ 10 J/m2 for cross ply and 320 ˙ 20 J/m2
for unidirectional specimens of the same thickness). In contrast, the shape of the
R-curve and the plateau level are strongly influenced by the ply orientation. The
steady-state ERR associated with crack growth in the cross ply specimen is almost
70 % higher than the one in delamination of the unidirectional specimen.
The crack starter a0 in the cross ply specimen separates two dissimilar layers and
thus is inherently subjected to mixed-mode condition. Yet, a scant contribution of
mode II at crack initiation (GII,i /GI,i D5 %) is found with the VCCT analysis. For
a long crack at the 0ı /90ı interface, a similarly small contribution of mode II is
assumed. However, when the crack plane is considered in the middle of the 90ı ply,
578 J. Botsis et al.
1600
1400
1200
ERR (J/m2)
1000
800
600
Specimen lay-up
400
Unidirectional
200
Cross-ply
0 10 20 30 40 50
Crack advance (mm)
Fig. 20.23 Total energy release rate in terms of crack advance for the cross ply and unidirectional
specimens with thickness H D 4 mm
the simulation indicates a clear mode I dominance. Employing the XFEM analysis,
crack deviation from the initial crack at 0ı /90ı interface is simulated. The analysis
results indicate that the local distribution of maximum principle stress has a major
influence on the crack extension direction. The latter leads to the oscillating crack
growth path confined by the adjacent 0ı layers that cannot be penetrated by the
crack.
The resulting wavy crack path in the cross ply specimen is associated with a
larger fracture surface as compared to the one obtained from the nominal straight
crack length recorded during the experiment. To evaluate the resulting differences
in the ERR, a projected crack growth length is defined as an D as cos®, where
as is the actual length of the crack advance and ® is an average deviation angle
measured from the polished longitudinal sections of the specimens (Fig. 20.21c).
Crack deviation angle ® is measured at approximately 30 locations (including
initiation point and several points along the propagation direction) in different
micrographs obtained from the tested cross ply specimens. The averaged value
of ® is found equal to 22.4ı and used for the calculation of the actual crack
length defined as as D a0 C as [27]. Such a variation in crack length results in
approximately 10 % reduction in the steady-state ERR values that cannot account for
the observed difference of the resistance curves in the cross ply and unidirectional
specimens (Fig. 20.23). Alternatively, the observed difference between the R-curve
behavior of cross ply and unidirectional laminates beyond the crack initiation can be
attributed to the different bridging mechanisms involved in the fracture of the two
specimen types. The fracture surface analysis of the cross ply specimen indicates
20 Traction-Separation Relations in Delamination of Layered Carbon-Epoxy. . . 579
crack deviation and growth into the adjacent 90ı layer and bundles of bridging fibers
oriented along the width direction.
In order to identify the tractions representing large-scale bridging in the cross
ply specimen, a parametric FE model is built and the strains sensed by FBGs are
simulated. Identification of bridging tractions is performed at the steady-state crack
propagation regime using the strain data expressed in the local crack-tip coordinate
system at the position of the last FBG in the sensors array. The computed values of
longitudinal strain are extracted from the 0ı layer at the position of the nodes of the
FE model corresponding to the FBG sensor location (Fig. 20.21b).
An error vector f" is calculated from the experimental and numerical strain data
"z and "Q z as follows:
"Q z "z
f" .max ; zmax ; / D (20.8)
h"z i
where h"z i represents the mean value of the measured strain distribution. A penalty
p Q
value of the reaction force fP .max ; zmax ; / D n PP P
, where PQ and P denote
the simulated and measured reaction forces at the load point, is appended to the
global error vector f D Œf" I fP in order to ensure the correct global response and
numerical convergence of the model. The difference between the numerical and
the experimental data is iteratively minimized to identify the bridging tractions
representing large-scale bridging in delamination of the cross ply specimens. The
identified parameters of the bridging tractions (Eq. 20.2b) in delamination of the
cross ply specimen are found as follows: max D 4.3 MPa, zmax D 23 mm,
D 0.16 mm1 . Regarding the corresponding values of delamination in the
unidirectional specimen, they are: max D 1.4 MPa, zmax D 43 mm, D 0.13 mm1
(Sect. 7.2). The obtained results indicate that the fracture of the cross ply specimen
involves a larger traction at the crack tip max and a smaller bridging zone length
compared with the unidirectional specimen.
The calculated COD profile from the FE model with the identified tractions is
related to the corresponding tractions to obtain the traction-separation relation. The
latter is displayed in Fig. 20.24 [27]. For comparison the corresponding relation for
the unidirectional specimen is also shown. Such results suggest that the toughening
mechanism in delamination of the cross ply laminate is more efficient and the
transverse bridging ligaments in the 90ı layer carry greater tractions as compared
to the longitudinal bridging fibers in delamination of the unidirectional composite.
Thus, the more enhanced R-curve behavior of the cross ply specimens compared to
the unidirectional specimens can be attributed primarily to the greater load bearing
efficiency of the 90ı ligaments bridging the crack faces.
The identified traction-separation relation, bb .ı/, serves in the cohesive zone
formulation, described in Fig. 20.10, to simulate the loading response and predict
crack growth in the cross ply DCB specimen. The numerical model, employed in
®
this study, consists of two layer-wise DCB arms (discretized by Abaqus CPE8R
®
elements) that are joined by a layer of cohesive elements (Abaqus COH2D4) with
a geometric thickness of 20 m and a constitutive thickness equal to unity. The
580 J. Botsis et al.
Fig. 20.24 Traction-separation relation obtained from the iterative analysis of the cross ply
specimen delamination. For comparison the corresponding relation is shown for a unidirectional
specimen with the same geometry. The inserted photographs show delamination of corresponding
specimens
cohesive elements are placed in the middle of the 90o layer adjacent to the pre-crack
plane where damage progression is experimentally observed. The discretization of
the cohesive zone model is shown in the insert at Fig. 20.25 [27].
Figure 20.25 shows the simulated load-displacement curve (solid black line)
obtained by implementing the identified traction-separation relation in the cohesive
element model. The experimental load-displacement data from the cross ply DCB
specimens are also shown. As demonstrated in Fig. 20.25, the simulated loading
response of the specimen is in a good agreement with the experimental data
indicating that the adopted identification approach yields realistic results regarding
the traction-separation relation in delamination of the cross ply laminate analyzed
herein.
It is worth noticing that the bridging ligaments in cross ply fracture are transverse
to the crack direction and constrained by the specimen width B. To shed some
light on the effects of B on the R-curve, DCB specimens with B D 12.5, 25, and
50 mm are tested. The experimental data from these tests are shown in Fig. 20.26.
Interestingly the specimen width sharply influences bridging and consequently the
R-curve response: the steady-state ERR is decreased by more than 25 % when the
width of the specimen is reduced from 25 mm to 12.5 mm and increased by 20 %
20 Traction-Separation Relations in Delamination of Layered Carbon-Epoxy. . . 581
100
Experiments Cohesive zone simulation
80
60
Load (N)
40 90°
0°
Pre-crack
90°
20 Cohesive elements
0°
90°
0 10 20 30 40 50
Applied opening displacement (mm)
Fig. 20.25 Experimental load-displacement curves of several cross ply specimens and prediction
of the cohesive zone modeling using the identified traction-separation relation
1800
1600
1400
1200
ERR (J/m2)
1000
800
0
0 10 20 30 40 50
Crack advance (mm)
Fig. 20.26 Energy release rate due to delamination of cross ply specimens with different widths
582 J. Botsis et al.
when the width is increased to 50 mm. Such size effects are very important and
certainly need to be addressed in details for a complete characterization of fracture
in cross ply layouts.
Fig. 20.27 (a) Schematic of a bridging bundle and connectors linking it to the DCB specimens.
(b) Schematic of the connection between the bridging bundles and the DCB specimen (not to scale)
584 J. Botsis et al.
Load (N)
75
50
25 H=6mm distribuon 1
H=6mm distribuon 2
H=6mm distribuon 3
0
0 2 4 6 8 10 12 14
Displacement (mm)
Fig. 20.29 Detail of the crack propagation and the bridging region
The bridging ligaments from the simulation of the embedded-cell model are
represented in Fig. 20.29. The crack initiates and propagates by breaking the weak
connectors. The progressive opening of the specimen results in the rotation of the
bridging bundles which also extend due to failure of intact connectors. Finally,
the bridging bundles fail when the maximum stress in the beam elements reaches
the bundle strength ( bf,max ). The competition between failure and generation of
bridging bundles determines the extension of the bridging region at steady state.
Numerical simulations are able to accurately reproduce the stages of the crack
propagation and the development of bridging in the DCB test.
Figure 20.30 [56] shows a sequence of photographs captured during a DCB test
together with the deformed shape obtained from the simulations for three applied
displacements. Evidently, the position of the crack tip and the extension of the
bridging region are very similar at the different loading stages.
20 Traction-Separation Relations in Delamination of Layered Carbon-Epoxy. . . 585
Fig. 20.30 Experimental photographs and numerical results showing the crack propagation. The
applied displacement of each figure is (a) 6.56 mm, (b) 12.1 mm, and (c) 16.10 mm
(a) (b)
350 H=6mm 3.2
H=10mm
300 H=14mm
250 2.4
ERR (kJ/m2)
Load (N)
200
1.6
150
100 H=6mm
0.8 H=10mm
H=14mm
50
0 0.0
0 2 4 6 8 10 12 14 16 18 20 0 10 20 30 40 50 60 70 80
Displacement (mm) Crack advance (mm)
Fig. 20.31 (a) Load displacement and (b) R-curves computed form the simulations of the mode I
intralaminar DCB test for three different thicknesses
The embedded-cell model is applied to study the effect of the specimen thickness
on the resistance curve obtained from the DCB tests. The same distribution of bun-
dles is embedded in three models of specimens with different thicknesses (H D 6,
10, and 14 mm), and numerical simulations of the mode I DCB tests are carried out
to evaluate the responses. The load-displacement curves obtained from the virtual
tests are shown in Fig. 20.31a [56]. Numerical simulations reproduce the effect of
the specimen thickness on the initial stiffness and the maximum load achieved. The
position of the crack tip during the simulations is measured by locating the point
of stress concentration in the iso-stress contour plots; this information allows the
determination of the R-curves using the MCC method (Fig. 20.31b, [56]). The ERR
586 J. Botsis et al.
increases upon crack propagation due to the development of the bridging zone until
it reaches a plateau when the bridging is fully developed and the crack propagates in
a stable manner. In good agreement with the experimental observations, the level of
the plateau achieved in the simulations shows a clear dependency with the thickness
of the specimen.
The embedded-cell approach of the DCB specimen is able to predict the initiation
and propagation of the failure, as well as the development of fiber bridging.
Furthermore, the numerical model is able to reproduce the thickness dependency
experimentally observed. These results show the ability of the multiscale approach
to study complex failure processes in composite materials as shown by the good
agreement with experimental data on load displacement (Figs. 20.15 and 20.31a)
and R-curves (Figs. 20.12 and 20.31b).
20.9 Conclusions
• The implemented iterative method based on accurate strain data along the
delamination direction, parametric FE models, and optimization yields very
reliable traction-separation relations in delamination and fracture. This is attested
by the cohesive element model predictions of the load-displacement curves and
the crack lengths. Moreover, the calculated energetic contribution of bridging
tractions coincides with the experimentally measured R-curves at the steady state.
• The identification method is based on material linear response but does not rely
on measurements of the ERR at the steady state. Thus, the bridging tractions can
easily be evaluated using linear and nonlinear geometric effects. Comparison of
the results from linear and nonlinear geometric effects is found very small in the
present studies.
• The adopted methodology also helps to elucidate the effects of the thickness on
the recorded R-curve in inter- and intralaminar fracture. While the maximum
bridging stress at the crack tip is independent of the specimen thickness, in each
case, the maximum bridging zone linearly increases with thickness. The effect
of thickness is also manifested in the decay rate of the tractions in a way that
H D const: for each series of specimens.
• In cross ply delamination fracture, the R-curve is strongly dependent on the
specimen width. A thorough 3D numerical approach accompanied by detailed
experimental studies is necessary to understand and model the effects of speci-
men width in delamination of cross ply specimens.
• Numerical simulations by means of an embedded-cell model of the unidirectional
DCB specimen predict the initiation and propagation of the failure, as well
as the development of fiber bridging. These results demonstrate the ability of
the multiscale approach to study complex failure processes in heterogeneous
materials. The model reproduced the thickness effects observed in the DCB
experiments and indicates that in thicker specimens, longer bridging bundles are
present which lead to noticeable increased ERR.
20 Traction-Separation Relations in Delamination of Layered Carbon-Epoxy. . . 587
Acknowledgements The authors acknowledge the partial financial support from the Swiss
National Science Foundation under Grant 200020_149721.
References
20. ASTM Standard, D6671-01, Standard Test Method for Mixed Mode I—Mode II Interlaminar
Fracture Touhness of Unidirectional Fiber Reinforced Polymers (ASTM International, West
Conshohocken, 2001)
21. ASTM Standard, D7905/D7905M – 14, Standard Test Method for Determination of the
Mode II Interlaminar Fracture Toughness of Unidirectional Fiber-Reinforced Polymer Matrix
Composites (ASTM International, West Conshohocken, 2014)
22. P. Davies, Review of standard procedures for delamination resistance testing. Delamination
behavior of composites, in ed. by S. Sridharan. Delamination Behavior of Composites
(Woodhead Publishing, CRC, Boca Raton, 2008), pp. 65–86
23. M. De Moura, Interlaminar Mode II fracture characterization, in ed. by S. Sridharan.
Delamination Behavior of Composites (CRC, 2008), pp. 310–325
24. S. Lee, An edge crack torsion method for mode III delamination fracture testing. J. Compos.
Technol. Res. 15(3), 193–201 (1993)
25. E. Farmand-Ashtiani, D. Alanis, J. Cugnoni, J. Botsis, Delamination in cross-ply laminates:
Identification of traction-separation relations and cohesive zone modeling. Compos. Sci.
Technol. 119, 85–92 (2015)
26. G. Pappas, J. Botsis, Intralaminar fracture of unidirectional carbon/epoxy composite: experi-
mental results and numerical analysis. Int. J. Solids Struct. 85–86, 114–124 (2016)
27. E. Farmand-Ashtiani, J. Cugnoni, J. Botsis, Specimen thickness dependence of large scale fiber
bridging in mode I interlaminar fracture of carbon epoxy composite. Int. J. Solids Struct. 55,
58–65 (2015)
28. A. Turon, P. Camanho, J. Costa, Delamination in Composites: Simulation of Delamination in
Composites Under Static and Fatigue Loading Using Cohesive Zone Models. VDM Verlag,
D-66123 (VDM Verlag Dr. Müller, Saarbrücken, 2008)
29. S. Stutz, J. Cugnoni, J. Botsis, Studies of mode I delamination in monotonic and fatigue loading
using FBG wavelength multiplexing and numerical analysis. Compos. Sci. Technol. 71(4),
443–449 (2011)
30. B.F. Sørensen, T.K. Jacobsen, Large-scale bridging in composites: R-curves and bridging laws.
Compos. A: Appl. Sci. Manuf. 29(11), 1443–1451 (1998)
31. V. Tamuzs, S. Tarasovs, U. Vilks, Progressive delamination and fiber bridging modeling in
double cantilever beam composite specimens. Eng. Fract. Mech. 68(5), 513–525 (2001)
32. M. Heidari-Rarani, M. Shokrieh, P. Camanho, Finite element modeling of mode I delamination
growth in laminated DCB specimens with R-curve effects. Compos. Part B 45(1), 897–903
(2013)
33. L. Sorensen, J. Botsis, T. Gmur, L. Humbert, Bridging tractions in mode I delamination:
measurements and simulations. Compos. Sci. Technol. 68(12), 2350–2358 (2008)
34. J. Botsis, Fiber Bragg grating applied to in-situ characterization of composites, in Wiley
Encyclopedia of Composites (Wiley, New Jersey, 2012), pp. 1–15
35. A. Airoldi, C. Dávila, Identification of material parameters for modelling delamination in the
presence of fibre bridging. Compos. Struct. 94(11), 3240–3249 (2012)
36. A.B. De Morais, A new fibre bridging based analysis of the Double Cantilever Beam (DCB)
test. Compos. A: Appl. Sci. Manuf. 42(10), 1361–1368 (2011)
37. R. Gutkin et al., Modelling the R-curve effect and its specimen-dependence. Int. J. Solids
Struct. 48(11), 1767–1777 (2011)
38. V. Shanmugam, R. Penmetsa, E. Tuegel, S. Clay, Stochastic modeling of delamination growth
in unidirectional composite DCB specimens using cohesive zone models. Compos. Struct. 102,
38–60 (2013)
39. C. Sun, Z. Jin, Modeling of composite fracture using cohesive zone and bridging models.
Compos. Sci. Technol. 66(10), 1297–1302 (2006)
40. A. Agrawal, P. Jar, Analysis of specimen thickness effect on interlaminar fracture toughness
of fibre composites using finite element models. Compos. Sci. Technol. 63(10), 1393–1402
(2003)
41. P. Davies, W. Cantwell, C. Moulin, H.H. Kausch, A study of the delamination resistance of
IM6/PEEK composites. Compos. Sci. Technol. 36(2), 153–166 (1989)
20 Traction-Separation Relations in Delamination of Layered Carbon-Epoxy. . . 589
66. D.A.W. Kaute, H.R. Shercliff, M.F. Ashby, Delamination, fiber bridging and toughness of
ceramic matrix composites. Acta Metall. Mater. 41, 1959–1970 (1993)
67. S.M. Spearing, A.G. Evans, The role of fiber bridging in the delamination resistance of fiber-
reinforced composites. Acta Metall. Mater. 40(9), 2191–2199 (1992)
68. B. Cox, Q. Yang, In quest of virtual test for structural composites. Science 314(314),
1102–1107 (2006)
69. J. LLorca et al., Multiscale modeling of composite materials: a roadmap towards virtual testing.
Adv. Mater. 23, 5130–5147 (2011)
70. C. González, J. LLorca, Multiscale modeling of fracture in fiber-reinforced composites. Acta
Mater. 54(16), 4171–4181 (2006)
71. V. Šmilauer et al., Multiscale simulation of fracture of braided composites via repetitive unit
cells. Eng. Fract. Mech. 78(6), 901–918 (2011)
Chapter 21
Damage and Failure Analysis of Bolted Joints
in Composite Laminates
21.1 Introduction
A. Ataş
University of Balıkesir, Balıkesir, Turkey
C. Soutis ()
Aerospace Research Institute, University of Manchester, Manchester, UK
e-mail: constantinos.soutis@manchester.ac.uk
Fig. 21.1 Comparison of joint efficiency (joint strength/base laminate strength) of brittle materi-
als, CFRP composites and ductile metals as a function of hole diameter-to-width ratio (d/w) [6–8]
21 Damage and Failure Analysis of Bolted Joints in Composite Laminates 593
Typical geometry of a pin joint (instead of a bolted joint for clarity) in a composite
laminate is shown in Fig. 21.2 with definitions of the width (w), free edge distance
(e), hole diameter (d) and thickness (t). The x-y and 1-2 coordinate systems define
the global laminate and local material coordinate systems, respectively. The angles
® and define the layer orientation angle with respect to the x (loading) axis and
the circumferential coordinate direction around the hole boundary, respectively.
Stress concentrations around the fastener hole under applied loads are the
primary reason for the failure of a joint. Typical in-plane critical stress distributions
around a pin-loaded hole are the normal stresses at the net-tension plane, shear
stresses at the shear-out plane and the radial bearing stresses at the bearing plane
as illustrated in Fig. 21.3. The stresses decay rapidly away from the hole boundary
at the net-tension and shear-out planes with increased w and e, respectively.
Maximum radial compression (bearing) stress develops at the bearing plane due
to the fastener/hole interaction.
In this chapter, failure of a mechanical joint describes failure of the fastener
hole. Three common in-plane failure modes (net-tension, shear-out and bearing)
observed in single pin joints in CFRP composite laminates are illustrated in Fig. 21.4
[9, 10]. Failure modes can be developed in isolation or in combination to each other,
including out-of-plane delamination depending upon the stress state.
There are various factors which may affect the stress distributions and, thus, the
mode of failure. Specimen geometrical ratios such as w/d and e/d (see Fig. 21.2),
laminate lay-up and material system are the most important parameters among
others [8, 11–15]. In the geometrical aspect, net-tension mode is observed when
the specimen width is not wide enough to relieve the high normal stress gradients
e t
594 A. Ataş and C. Soutis
Shear-out
Decreasing e planes
Decreasing w sx
Bearing
srr plane
t xy
Net-tension
plane
y
a) b) c)
+φ
x
P P P
Fig. 21.3 Typical stress distributions around a pin-loaded hole: (a) net-tension, (b) shear-out and
(c) bearing failure plane
P P P P
(low w/d ratio). Shear-out mode is observed in laminates with adequate width
against the net-tension mode but insufficient free edge distance to relieve the high
shear stresses developed at the shear-out plane (low e/d ratio). In the net-tension
21 Damage and Failure Analysis of Bolted Joints in Composite Laminates 595
and shear-out failure modes, the compressive stresses developed at the bearing
plane are not high enough to cause significant fibre failure before those net-tension
or shear-out modes occur. Bearing mode is caused by the compressive stresses
and includes matrix cracking, fibre microbuckling and kinking with a significant
amount of delamination. The above-mentioned failure modes are observed for the
pin joints, and applied clamping conditions in bolted joints can change the failure
mode significantly.
Stress analysis is the first step of any stress-based failure analysis in which the
predicted stresses are compared to the corresponding strength or allowable stresses.
Stress analysis is much more complex in composites than in isotropic materials
due to their anisotropy and layered nature. Furthermore, in mechanical joints, it is
not possible to obtain closed form solutions for the contact behaviour between the
fastener and boundary of the hole due to the changing contact arc as a function of
the applied load. Therefore, the finite element (FE) method is extensively used to
model and analyse the mechanical joints in composite laminates.
However, the accuracy of an FE analysis should be checked against experimen-
tally and/or analytically determined data since the FE method is an approximation
technique. Errors can occur especially due to the inappropriate mesh density
and/or element selection. Analytical and experimental stress analysis methods are
introduced in the next sections followed by numerical two-dimensional (2-D) and
three-dimensional (3-D) FE stress analyses.
the stresses are prescribed on one part and displacements on the remaining part of
the boundary. In a pin joint, the unloaded part (the part which is not in contact
with the pin) of the hole was considered to be stress-free region, and corresponding
boundary conditions were directly substituted into the stress functions to simplify
the calculations. In the constitutive stress function, therefore, the only boundary
condition to be enforced is the displacement or stress on the loaded part of the
hole boundary. By this approach, the circumference of the hole boundary subjected
to boundary conditions was reduced, thereby decreasing the solution time of the
problem.
Load distribution in multi-bolt joints is of great importance since it is the first
step in designing the practical structural joints. Therefore, analytical models were
developed for in order to avoid the high computational complexity of the FE models
including many bolts [23].
Strain gauges, photoelasticity and moire techniques have been used to obtain
stresses around the fastener holes in composite laminates. These experimental data
are of paramount importance for FE analysts in order to verify model predictions as
well as analytical solutions.
Transmission photoelasticity was used by Prabhakaran [24] to observe
the stress distribution around a pin-loaded hole in a unidirectional and
[0ı 4 /C45ı 4 /45ı 4 /90ı 4 ]s quasi-isotropic glass/epoxy laminate of 4 mm nominal
thickness. The effect of material anisotropy was also qualitatively highlighted;
higher local stresses develop in the unidirectional laminate where the axial modulus
is several times higher (16:1) than the transverse value.
Hyer and Liu [25] have conducted research on pin-loaded isotropic (poly-
carbonate, PSM-1), [0ı 4 /C45ı 4 /45ı 4 /90ı 4 ]s quasi-isotropic, unidirectional and
[C45ı 4 /45ı 4 ]2s angle-plied glass/epoxy laminates (w/d D 4, e/d D 2) using trans-
mission photoelasticity. Stresses were calculated by using a finite difference rep-
resentation of the plane stress equilibrium equations and the stress-optic equations.
Stress distributions were calculated around the hole boundaries and at three different
failure planes (net-tension, shear-out and bearing) for the isotropic plate and the
other laminates based on the fringe patterns.
Double-lap pin-loaded unidirectional and woven fabric graphite fibre/epoxy
specimens were investigated by Koshide [26] using a technique that combines
the moire method and strain gauge measurements. The surface of the composite
laminate was coated with a thin epoxy resin/aluminium powder mixture layer for
higher frequency. The influence of fibre orientation on strain distributions as well as
on the shear deformation caused by the pin load was demonstrated.
21 Damage and Failure Analysis of Bolted Joints in Composite Laminates 597
The FE method is usually implemented to analyse the stresses around the fastener
holes in composite laminates due to the complexity of the problem including the
contact phenomena, pin/bolt elasticity, clearance, variable loading conditions, etc.
Limited performances of computer processors have directed researchers to 2-D FE
stress analyses in the past decades where plane stress approximation and equivalent
orthotropic laminate properties were assumed. Although providing a great deal
of information, these models suffer from many significant drawbacks such as the
homogenisation of the laminate which neglects the interface of consecutive layers
and stacking sequence which can lead to significant errors in predicting the joint
failure strength.
Eriksson [27] conducted FE analyses for [0ı (25 %) /90ı (25 %) /˙45ı (50 %) ]s quasi-
isotropic, [0ı (69 %) /90ı (6 %) /˙45ı (25 %) ]s and [0ı (6 %) /90ı (69 %) /˙45ı (25 %) ]s laminates
for T300/914C CFRP material system. The influence of laminate lay-up, clearance
(gap between the bolt and plate), friction and bolt stiffness on the radial stress
distributions ( rr ) around the hole boundary was investigated. The effect of the bolt
stiffness was shown to be negligible, whereas the effects of the other parameters
were important.
Crews et al. [28] have made extensive stress analyses on pin joints of [0ı ],
[90ı], [0ı /90ı ]s , [˙45ı ]s , [0ı /˙45ı ]s and [0ı /˙45ı /90ı ]s quasi-isotropic laminates
in order to determine the influence of the laminate orthotropy and geometrical
parameters on the tangential ( ), radial ( rr ) and shear ( xy ) stress distributions
around the pin holes. Stress distributions around the hole boundary were calculated
for a constant e/d ratio to show the influence of the w/d ratio in a quasi-isotropic
laminate. The decrease in the specimen width resulted in tangential stresses, ,
that were approximately two to three times higher at the end of the contact arc
( 90ı ), a trend observed for almost all laminate patterns. was associated
with the net-tension failure mode, and it was concluded that the specimen width
should have a significant effect on the net-tension strength. The effect of the width
on the rr and xy was relatively small.
Continuous advances in computer processors and improved FE formulations
have allowed the development of relatively more complex 3-D models to be solved
with improved efficiency. These models incorporate essential interlaminar (through-
the-thickness) normal and shear stress components (¢ 13 , ¢ 23 , ¢ 33 ). These stresses
are important when studying the stacking sequence and clamping force effects on
bolted joints of composite laminates particularly in single shear specimens where
the fastener body may exert non-uniform contact stresses to the hole boundary. Also,
countersunk fastener holes cannot be accounted for in 2-D models.
Furthermore, delamination onset and growth are governed by the interlaminar
stress components which necessitate the use of 3-D models. Widely used commer-
cial FE codes offer layered structural solid elements which can be used to model
laminated composites. However, the use of more than one composite layer within
598 A. Ataş and C. Soutis
bearing plane was compensated with an increase in the tangential stresses at the net-
tension plane. The interlaminar normal stress at the 0ı /90ı interface was shown to
be higher than that of the 90ı /0ı interface which was related to the experimentally
obtained higher strength of the [90ı /0ı ]s laminate. Compressive interlaminar normal
stresses were developed under the washer area with a tendency to approach tensile
values at the outer edge with an applied clamping force. Interlaminar shear stresses
were increased at the washer outer edge with an increase in clamping force, and
this was related to the restriction of remote bearing failure to the outer edge of the
washer.
Chen et al. [38] have conducted parametric numerical studies on 3-D models
of double-lap graphite/epoxy and GFRP specimens including stacking sequence,
friction, clearance, clamping torque and bolt elasticity. The effect of friction and
stacking sequence on the delamination onset was investigated based on interlaminar
stresses. However, no attempt was made to investigate the delamination growth.
The first nonlinearity, the first or the global peak point (maximum load sustained by
the joint) in the load-displacement curve and the amount of hole deformation have
been generally used to define experimental failure load [12, 14, 29, 39–46]. Strength
prediction, on the other hand, is the estimation of the load-carrying capability of a
mechanical joint with the data obtained from the analytical and numerical stress
analyses techniques introduced earlier. Analytical and FE stress predictions along
the fastener hole boundary have been used to define the failure load based on
various theories such as maximum stress/strain criteria [47–49]. These criteria can
be applied to a layer of the laminate or to the whole laminate depending upon the
methodology used. Although simple, these methods lead to underestimation of the
joint strength because of omitting the stress redistribution due to the local subcritical
damage mechanisms, e.g. matrix cracks, splitting and delamination.
Accordingly, methods were developed to account for subcritical damage mech-
anisms caused by the local high stress concentrations [50–54]. Stresses were
calculated at a so-called characteristic distance to determine the failure load in
some models where only two experimentally defined parameters required: the
characteristic distance and the laminate unnotched strength. They have been widely
used owing to their simplicity, although the need to determine the unnotched
strength and characteristic distance of each different laminate system are the
drawbacks of these methods. Other methods were also developed to take into
account the subcritical damage mechanisms based on the relation between elastic
isotropic stress concentration factors and the stress concentration factors at failure
of the specific material system of the same joint geometry [8, 55].
Progressive failure analysis (PFA) is a methodology which simulates the
initiation and growth of the damage in mechanical joints based on stress
analysis, implementing failure criteria and material property degradation rules
600 A. Ataş and C. Soutis
[11, 14, 30–33, 39, 56–65]. The effect of the subcritical damage on the global
response was simulated by reducing the material properties of the failed elements.
Special attention must be paid to the material degradation rules which are key
parameters representing the subcritical damage. Generally, a parametric study
is necessary to determine the individual degradation factors for the best fit to
experimental data of a certain laminate configuration. Then, the same factors can
be used to predict the joint strengths of various joint geometries and laminate
configurations of the same material system with varied success.
Fracture mechanics-based DZM has been used to predict the strength of the
composite joints as an alternative to the characteristic distance and the PFA
methods [66]. The DZM can be thought as a PFA methodology since the initiation
and the growth of damage are taken into consideration. This method requires the
laminate strength and fracture energy properties to be determined experimentally
for each laminate type as input parameters. Although providing good predictions,
this method has limited use compared with the PFA methodology.
Strength prediction based on hole boundary stresses, semiempirical techniques,
progressive failure analysis (PFA) and the damage zone model (DZM) are intro-
duced in the following sections.
Waszczak and Cruse [47] have used the maximum stress, maximum strain and
distortional energy failure theories to predict both the failure load and failure modes
of pin joints in graphite/epoxy and boron/epoxy laminates by using a 2-D FE
method. The FE contour plots were used to predict the failure modes. A ply discount
method was used to predict the failure loads where the stresses at the first row of the
elements around the hole periphery for each layer were checked against the failure
criterion. A layer was assumed to have failed when any of its elements failed, and
the layer was then removed from the laminate. After the initial failure, the load was
redistributed to the remaining layers, and the failure criterion was reapplied in a
successive manner until all plies of the laminate have failed. Conservative failure
load predictions were obtained up to 50 % since the damaged plies can still carry
extra load. The distortional energy criterion provided improved failure load and
mode predictions in comparison with maximum stress and maximum strain criteria,
when compared to the measurements.
A 3-D analytical strength prediction method was developed by Smith et al. [49]
based on calculation of bearing stress at the onset of failure for a quasi-isotropic
laminate in terms of material properties and the specimen geometry. The net bearing
load on the hole boundary was obtained by subtracting the total clamping torque
effect from the applied load by considering the friction between the laminate and the
washers. The total effect of the clamping torque was simulated by considering the
restraints under the washer area due to the initial clamping torque and through-the-
thickness expansion restriction under simultaneously applied in-plane bolt loading.
21 Damage and Failure Analysis of Bolted Joints in Composite Laminates 601
Radial and tangential in-plane stress components were calculated from the 2-D
analysis of Bickley [67]. The cosine stress distribution was used to represent the bolt
loading. Due to the complexity of the model, the constitutive equation is affected by
many variables that influence the predicted stresses. For higher levels of clamping
pressure, the experimental failure onset stress is increased with increasing friction.
This was attributed to a change in failure from a local bearing to a tensile mode.
General qualitative explanations were also provided on varying parameters such as
the coefficient of friction between the washers and the laminate, the washer size and
different lap material configurations.
This simple formula makes it possible to predict the strength of a wide range of joint
geometries with a limited number of tests.
The PFA was developed in order to avoid the drawbacks of the semiempirical
characteristic curve methods such as experimentally determined (material, laminate
lay-up and specimen geometry dependent) characteristic distances and to capture
damage evolution. The PFA uses the stresses obtained from the FE analysis to detect
the damage by means of the failure criteria used. It implements material property
degradation factors, which are functions of the damage type detected, to simulate
the damage growth up to final failure of the joint.
Early 2-D PFA analyses that consider only the in-plane stresses [39, 58, 60, 61,
63] were followed by more detailed 3-D analyses for single-lap single-fastener [32,
59, 73, 74], double-lap single-fastener [9, 14] and single-lap multi-fastener joints
[31] taking into account of out-of-plane stresses in the vicinity of the hole boundary.
PFA is also used to predict the behaviour of bolted joints in large-scale aircraft
structures [75, 76].
Lessard and Shokrieh [63] have used a PFA in double-lap pin joints in CFRP
laminates to investigate the applicability and limitations of 2-D analyses both in lin-
ear and nonlinear cases. The nonlinear definition includes both the shear stress/shear
strain material nonlinearity and the large deformation theory. Hashin failure criteria
[77] were used to detect matrix tension/compression, fibre tension/compression
and fibre/matrix shearing failure modes. The effect of the damage was introduced
by reducing the related material elastic properties to zero. This sudden material
property degradation led to numerical instabilities which forced the FE programme
to stop before reaching the final failure of the joint due to the highly distorted
elements. Nevertheless, the strength of the joints was predicted within an accuracy
of 10 % despite the unsuccessful failure mode predictions.
In order to improve the prediction of the failure mode, an approach was proposed
[63] in which the elastic material properties of the failed elements were increased
instead of reducing them. This allows damage propagation up to laminate edges
or to a predefined point without numerical errors. Nonlinear models with shear
21 Damage and Failure Analysis of Bolted Joints in Composite Laminates 603
stress/shear strain nonlinearity including this approach were also developed for
comparison with the linear model predictions. It was shown that predictions by the
nonlinear model were in better agreement with experimental data than the linear
model, especially for the bearing mode.
Choice of the failure criteria and material property degradation rules used to
detect the failure initiation and growth has a significant influence on the PFA
predictions. Dano et al. [39] have applied different failure criteria combinations
including nonlinear shear stress/shear strain behaviour and large deformation
theory in 2-D double-lap pin joint models to investigate this effect. Hashin failure
criteria [77] were selected with four (matrix tensile, matrix compression, fibre
tensile, fibre compression) and five components (matrix tensile, matrix compression,
fibre tensile, fibre compression, fibre/matrix shear-out) which were denoted as
Hashin (IV) and Hashin (V), respectively. Hashin fibre compression criterion is
identical to the maximum stress criterion, whereas a shear stress component is
included in the fibre tension criterion. This shear component leads to significantly
conservative predictions in the case of mechanically fastened joints since high shear
stresses are developed around the loaded hole boundary. The authors therefore
introduced mixed-mode criteria, which merely substitute maximum stress criterion
for both fibre tensile and fibre compressive failures in the Hashin criteria [77].
The mixed-mode criteria provided higher strength predictions than either Hashin
criteria because the latter predicted the final failure earlier due to the higher
shear stress components. The inclusion of the large deformation theory was shown
to be insignificant. However, predictions for all combinations were increased
considerably owing to decreased shear stress components by including both the
large deformation theory and nonlinear shear stress/shear strain behaviour together.
Therefore, predictions tended to converge due to the decreased influence of the shear
stress component in the Hashin criteria. The effect of the nonlinear shear stress/shear
strain behaviour with mixed-mode criteria was also shown to be negligible for quasi-
isotropic laminates as opposed to the [(0ı /90ı )6 ]s and [(˙45ı)6 ]s laminates.
A 3-D PFA model of double-lap finger-tightened T300/914 CFRP bolted
joints including out-of-plane stress components was developed by Camanho and
Matthews [15, 29, 30]. Eight-node solid elements were used to model the laminate
with one element per layer approach. Hashin failure criteria [77] were used to
predict the failure initiation, and elastic properties of the failed elements were
reduced by the extended approach originally proposed by Tan et al. [78] in order
to model the growth of the damage. Failure of the specimens was assumed to
occur when the first load drop in the numerical load-bolt displacement curves was
observed [30]. For the net-tension and shear-out specimens, the damage in the fibres
reached to the free edge of the laminate with the first load drop. The bearing tests
showed that the specimens were able to sustain some additional load after the first
drop-off with a significant stiffness reduction. Accordingly, it was stated that the
first load drop in the experiments could be used as the final failure load for the
bearing specimens, which led to slight over-prediction. The model predictions were
in agreement with the experiments within an accuracy of 6.2–13 % [30].
604 A. Ataş and C. Soutis
The effect of lateral constraint type, washer size and clamping torque on the
strength predictions of metal-composite-metal double-lap joints (w/d D 8, e/d D 6)
has been investigated by Chang et al. [79] using PFA. Pin and bolted joint (with or
without washers between the metal side plates and the laminate) configurations were
examined. Six failure modes were considered, namely, matrix tension, fibre-matrix
shearing, matrix compression, fibre tension, fibre compression and fibre-matrix
compression. Material degradation rules proposed by Shahid and Chang [80] for
2-D models based on the theory of elasticity and continuum damage mechanics
have been extended for the 3-D case. Transverse tensile strength and shear strength
were updated throughout the analyses depending on the matrix crack density. One
solid eight-node layered brick element was used for half of the laminate thickness
in all cases. The load-displacement curve of a joint may follow quite different paths
depending on the lateral constraint type and clamping force level. A clear load drop
was generally observed in pin joints and bolted joints with low washer diameters,
and the maximum load before the sudden drop was taken as the failure load.
However, the load drop was not clear in bolted joints with high washer diameters
and without washers. In that case, the proportional load limit as a certain percentage
of the bolt diameter was used to define the failure load [79].
The method offers many advantages, but still many assumptions are made in
relation to the degradation algorithms used. A detailed investigation about the
current strategies in PFA can be obtained from Tay et al. [81].
The DZM has been successfully used for open hole tensile strength predictions
of composite laminates [82]. Hollmann [66] concluded that the DZM model was
expected to predict in-plane net-tension and shear-out failure strengths of bolted
joints accurately since the in-plane failure mechanisms of open hole tensile and
bolted joint specimens were expected to operate in a similar fashion. Consequently,
this model was applied to the bolted joints in T300/914C graphite/epoxy laminates
in order to predict the shear-out strength [66]. A double-lap, finger-tightened, single-
bolted joint configuration was selected with the geometrical ratios enhancing the
shear-out failure mode (w/d D 6, e/d D 1.67).
The DZM takes into account the failure initiation and growth by treating the
complicated failure mechanisms (matrix cracks, delamination, fibre cracks) in
the highly stressed area around the notch as a through-the-thickness equivalent
crack [66]. Failure initiation is predicted when the stress at the edge of the net-
tension or shear-out planes reaches the corresponding strength value. A linear
softening behaviour is adopted following the failure initiation, and the critical
fracture energies need to be known associated with the tensile and shear failures.
This process provides a load-displacement curve on which the maximum load
21 Damage and Failure Analysis of Bolted Joints in Composite Laminates 605
represents the joint ultimate strength. The strength predictions obtained by the DZM
were shown to provide good correlation with the experimental data within 10 %
accuracy.
The strength prediction method which will be presented in the next section is
developed based on modelling subcritical damage mechanisms. The penetrant-
enhanced X-ray radiography technique was used to identify the locations of these
subcritical damage mechanisms following the double-lap shear tensile tests for
various joint geometries and laminate lay-ups. Those locations were then used to
model the subcritical damage planes. This section summarises the material system
used and specimen design, measurement of the joint strength and determination of
the damage locations.
The material system used was HTS40/977-2 carbon fibre-epoxy in the form of pre-
impregnated (prepreg) tape. The prepreg tapes were made of unidirectional high
®
tensile strength/standard modulus aerospace grade carbon fibres (Toho Tenax ,
ı
HTS40-F13-12 K-800tex) pre-impregnated with 177 C curing toughened epoxy
®
resin (Cycom 977-2). The nominal thickness of the prepreg tape is 0.25 mm with
a fibre volume fraction of 58 % [83]. The composite laminates were fabricated
by hand lay-up technique in an autoclave according to the manufacturer’s rec-
ommended curing procedure. A diamond tip saw was used to cut the laminates
to dimensions with special care given to the precise alignment of the laminates.
Fastener holes were drilled with a backing plate in order to prevent drilling
induced delamination failure. The laminates were inspected by X-ray radiography
to establish specimen quality.
Double-lap pin and bolted joint loading fixtures were manufactured in accor-
dance with the ASTM standard D5961/D5961 M-01 [84]. This standard originally
proposes a loading fixture only for bolted joints which contains a machined section
representing a 13 mm washer diameter (dw). This washer section was machined to
12 mm in order to obtain a washer-to-bolt diameter (dw/d) ratio of 2.0 for the bolted
joint tests, whereas that section was completely removed for the pin joint tests [85].
Stainless steel pins of 5.98 mm and 12.9 grade steel bolts of 5.95 mm diameter were
used to load the pin and bolted joint specimens, respectively.
606 A. Ataş and C. Soutis
Cross-ply specimens are often not used in real life structural applications. They
were used in this work, however, due to their relatively apparent damage locations
to facilitate comparison with the FE predictions. Three cross-ply lay-ups, as shown
in Table 21.1, were tested considering the effect of sublaminate-level ([90ı/0ı ]2s )
Table 21.1 Test parameters Laminate pattern Specimen code w/d e/d
for cross-ply pin and
finger-tight bolted joint [90ı /0ı ]s P1W (pin) 6 3
specimens B1W (bolt) 6 3
[90ı 2/0ı 2]s P2W (pin) 6 3
B2W (bolt) 6 3
[90ı /0ı ]2s B3W (bolt) 6 3
B3N (bolt) 3 3
21 Damage and Failure Analysis of Bolted Joints in Composite Laminates 607
3000
2000
Load (N)
Fig. 21.5 Typical load-displacement curves of pin and finger-tight bolted joints in [90ı /0ı ]s lay-
up (e/d D 3, w/d D 6)
and ply-level ([902 ı /02 ı ]s ) scaled laminates on the bolted joint strength [88–91]
(W and N stand for the wide and narrow specimens, respectively). Laminates with
90ı outer layers were used to provide the same degree of constraint for the inner 0ı
layers [92]. The width-to-hole diameter and edge distance-to-hole diameter ratios
were kept constant (w/d D 6 and e/d D 3) for most joint configurations.
Typical load-displacement curves of the [90ı /0ı ]s lay-up specimens are shown
in Fig. 21.5. The pin joints in [90ı/0ı ]s lay-up initially exhibit predominantly linear
behaviour. This is followed by a nonlinear portion with a considerable stiffness loss
and consequently a sudden load drop due to the extensive bending of the laminate
under compressive pin contact stresses without any lateral support. Finger-tight
lateral support prevents the sudden failure of the [90ı /0ı ]s lay-up and provides a
strength increase of 16 %. Similar results were reported by Smith et al. [93] for the
[0ı /90ı ]s lay-up (the bolted joint appears softer in Fig. 21.5 due to some slippage
that occurred during the tests).
The X-ray images of pin (P1W) and finger-tight bolted (B1W) joints in [90ı /0ı ]s
lay-up are shown in Figs. 21.6 (i) and (ii), respectively.
Transverse matrix cracks, compressive fibre failure and delamination damage are
observed in the pin joint specimens. The local fibre compressive failure initiated at
the highly loaded bearing area (indicated by d in Fig. 21.6 (i)), and it triggered
delamination between the 90ı and 0ı layers. Increased loading, following the
delamination, caused excessive bending of the layers and resulted in a sudden
load drop in the load-displacement curve, as shown in Fig. 21.5. This extensive
local damage prevented the development of any axial splitting due to increased pin
displacement.
The finger-tight lateral constraint changed the damage mechanisms drastically
as shown in Fig. 21.6 (ii). Axial splits running from the hole edge (approximately
tangent to D 90ı ) to the free edge were observed in addition to failure modes
observed in the pin joints. The washers prevented any premature bending of the
608 A. Ataş and C. Soutis
Fig. 21.6 X-ray radiographs showing damage in pin and finger-tight bolted joints in [90˚/0˚]s lay-
up at failure: (a) transverse matrix cracks, (b) axial splitting, (c) delamination and (d) compressive
fibre failure in 0ı layers
layers due to the local fibre compression failure. The increased bolt displacement
caused the bundles of 0ı layers to be sheared out after delaminating from the 90ı
layers as reported by Hart-Smith [8]. These splits were accompanied by widespread
90ı transverse matrix cracks.
The load drop is more gradual for the pin joints of relatively thicker ply-
level scaled [902 ı /02 ı ]s lay-up as shown in Fig. 21.7. This is probably due to the
increased bending stiffness of the central 0ı layers which allows more progressive
damage to occur before any catastrophic bending of the layers. It is clear from the
X-ray radiograph in Fig. 21.8 (i) that some degree of splitting develops. However,
eventual bending (out-of-plane deflection) of the central 0ı layers, accompanied
with extensive delamination, causes final failure before the splits reach the free edge.
The finger-tight washers limited the bending of the 0ı layers which resulted
in less delamination damage that was similar to that observed for the [90ı /0ı ]s
lay-up. Consequently the splits that were accompanied by extensive transverse
matrix cracks grew to the free edge with increased bolt displacement as seen in
Fig. 21.8 (ii).
Additional ply-level scaled [90ı2 /0ı 2 ]s specimens were tested to 95 %, 90 %
and 80 % of the average ultimate load. The specimens were intact up to 95 %
of the average failure load where transverse matrix cracks, longitudinal splitting,
compressive fibre failure and delamination appeared as seen in Fig. 21.9.
Figure 21.10 shows that the w/d ratio does not have a significant effect on the
failure response of bolted joints in the sublaminate-level scaled [90ı /0ı ]2s lay-up
with a constant edge distance (e/d D 3).
Similar observations were reported elsewhere [94] for the [0ı /90ı ]2s CFRP
laminates. This trend is expected since it has been shown that [28] the w/d ratio
has an insignificant effect on the shear stress distributions at the shear-out damage
planes for cross-ply laminates.
21 Damage and Failure Analysis of Bolted Joints in Composite Laminates 609
5000
3000
0
0 0.5 1 1.5
Pin/bolt displacement (mm)
Fig. 21.7 Typical load-displacement curves of pin and finger-tight bolted joints in [902 ı /02 ı ]s
lay-up (e/d D 3, w/d D 6)
b
d b d
c
c
a
a
P
P
d=6mm washer edge d=6mm
The bearing strength of the cross-ply specimens (Smax ) at failure are given in
Fig. 21.11. The bearing strength is defined as
where Pmax is the maximum load sustained by the joint and d and t are the hole
diameter and laminate thickness, respectively.
610 A. Ataş and C. Soutis
Fig. 21.9 X-ray radiographs showing subcritical damage development in pin joints of the
[90ı 2 /0ı 2 ]s laminates at 80, 90 and 95 % of the average ultimate load (100 %): (a) transverse
matrix cracks, (b) axial splitting, (c) delamination and (d) compressive fibre failure in 0ı layers
The increased thickness reduced the coefficient of variation (Cv ) in the pin joints
due to the increased specimen stability without a significant strength increase.
Comparison of X-ray radiographs, Figs. 21.6 (i) and 21.8(i), clearly shows that
the thinner [90ı /0ı ]s lay-up experienced much less local damage than the thicker
[902 ı /02 ı ]s lay-up due to the premature bending of the material beneath the pin.
Any physical comparison regarding the failure mechanisms will therefore not be
reasonable for the pinned joints.
For the bolted joints, the sublaminate-level scaled [90ı/0ı ]2s lay-up showed a
strength increase of approximately 19 % over the [90ı/0ı ]s lay-up whereas the ply-
level scaled [90ı2 /0ı 2 ]s lay-up only showed a 4 % strength increase. The significant
strength increase of sublaminate-level scaled specimens is due to the interspersion
of the layers instead of grouping them together as in ply-level scaled specimens.
The key features of the global shear-out failure mode observed in the bolted joints
of cross-ply laminates are the splitting within the 0ı layers and the delaminations
between the consecutive layers. Grouping the layers together has a twofold effect
21 Damage and Failure Analysis of Bolted Joints in Composite Laminates 611
6000
5000
4000
Load (N)
3000
Fig. 21.10 Typical load-displacement curves of finger-tight bolted joints in [90ı /0ı ]2s lay-up as a
function of w/d (e/d D 3)
600
400
Bolt (B3N)
Bolt (B3W)
200
[90/0]s [902/02]s [90/0]2s
on the strength of those joints. First, the splits initiate at the hole edge and
grow through-the-thickness of the grouped layers without an interruption from the
adjacent layers. Second, grouping the layers reduces the number of the interfaces
and therefore increases the magnitude of the shear stress to be carried by each
interface [95]. As a result, greater delamination was developed under lower load
levels for the ply-level scaled [90ı2 /0ı 2 ]s lay-up with a broader extent of transverse
matrix cracks. Although this could result in a stronger joint in a quasi- or near
quasi-isotropic lay-up that fails in net-tension mode by providing better stress
concentration relief [7], that greater delamination size has facilitated the shear-out
mode for the cross-ply lay-up due to poorer shear properties.
612 A. Ataş and C. Soutis
Figure 21.12 shows the schematic of the subcritical in-plane damage planes
observed from the X-ray radiographs of the cross-ply lay-ups. The extent of
the transverse matrix cracks dispersed along the axial splits as shown in X-ray
radiographs. However, it has been shown that those transverse matrix cracks can be
represented as a single crack where the maximum stress concentration occurs for the
centre-notched specimens [96]. Based on this assumption, transverse matrix cracks
were characterised as single cracks at the maximum stress concentration locations
( D ˙90ı ). Axial split planes within the 0ı layers were extended from the hole
edge at D ˙90ı to the free edge as observed from the radiographs.
8000
6000
Load (N)
4000
Finger-tight bolt, w/d=3 (Q1N)
2000 Finger-tight bolt, w/d=6 (Q1W)
[+45/0/-45/90]s lay-up (Q1)
0
0 1 2 3
Displacement (mm)
Fig. 21.13 Typical load-displacement curves of finger-tight bolted joints in [45ı /0ı /45ı /90ı ]s
lay-up as a function of w/d (e/d D 3). (a) Transverse matrix cracks, (b) axial splitting, (c)
delamination and (d) compressive fibre failure in 0ı layers
b b
d
c d
d
c
a d a
d
P d
P
washer edge washer edge d=6mm
Fig. 21.14 Damage in bolted [45ı /0ı /45ı /90ı ]s lay-up at ultimate load: (a) transverse matrix
cracks, (b) 0ı axial splits, (c) delamination and (d) ˙45 axial splits
Fig. 21.15 Photographs of failed bolted joint specimens in [45ı /0ı /45ı /90]s lay-up as a function
of w/d (e/d D 3)
8000
6000
Load (N)
4000
Fig. 21.16 Typical load-displacement curves of finger-tight bolted joints in [90ı /45ı /45ı /0ı ]s
lay-up as a function of w/d (e/d D 3)
The wider specimens (Q2W and Q3W) exhibited bearing failure modes as
expected. The narrow specimens for both lay-ups (Q2N and Q3N) experienced
mixed-mode failure which resulted in a sudden load drop and reduced strength
values (see Figs. 21.18, 21.19 and 21.20). Mixed-mode failures generally include
transverse tensile cracks in 90ı layers, axial splits in 0ı and ˙45ı layers and
extensive delamination between all the interfaces.
Therefore, it is clear that the modes of failure are not only a function of the
specimen geometry. The high strength of the fibres (HTS40) and high toughness of
the resin matrix (977-2) used in the current material system is the main reason for
change in the mode of failure. In comparison, HTS40 fibres and the similarly widely
used T300 fibres have tensile strengths of 4275 MPa and 3530 MPa, respectively.
21 Damage and Failure Analysis of Bolted Joints in Composite Laminates 615
8000
6000
Load (N)
4000
Fig. 21.17 Typical load-displacement curves of finger-tight bolted joints in [0ı /90ı /45ı /45ı ]s
lay-up as a function of w/d (e/d D 3)
b b c
c a d
c
d d a
d d
P P
washer edge d=6mm washer edge d=6mm
Fig. 21.18 Damage in bolted quasi-isotropic lay-ups at ultimate load: (a) transverse matrix cracks,
(b) 0ı axial splits, (c) delamination and (d) ˙45 axial splits
Accordingly, the high stress concentration developed at the hole edge (at
D 90ı ) may not reach the strength value of the fibres in the present specimens.
This results in some other matrix-dominated failure modes with increased bolt
displacement. This is the reason for the increased transverse matrix cracks, axial
splits both in 0ı and ˙45ı layers and broader delamination. Smith and Pascoe [93]
have reported similar mixed-mode failures for the XAS/914 CFRP material system,
which has high tensile strength fibres that are comparable to those of the currently
discussed system.
However, the presence of high-strength fibres in the current material system alone
does not explain why narrow specimens of different lay-ups failed in different modes
(Q1N failed in bearing mode). A possible explanation may be the dependency of
thermal stresses induced by the production of the CFRP laminates on the stacking
sequence. It has been shown that the interlaminar normal stresses (ILNS) developed
616 A. Ataş and C. Soutis
Fig. 21.19 Photographs of failed bolted joint specimens in [90ı /45ı /45ı /0]s lay-up as a function
of w/d (e/d D 3): (a) transverse matrix cracks and (d) ˙45 axial splits
Fig. 21.20 Photographs of failed bolted joint specimens in [0ı /90ı /45ı /45ı ]s lay-up as a
function of w/d (e/d D 3): (b) 0ı axial splits
during the cool-down process after curing may cause important changes in the local
failure mechanisms of bolted joints in quasi-isotropic CFRP laminates [97], and
those local changes may result in global changes to the failure mode. Additionally,
the subcritical damage mechanisms for the bolted joint specimens were shown to be
shear dominated (Mode-II) without any consideration given to the thermally induced
ILNS stresses (Mode-I). The cohesive zone length or the process zone ahead of a
crack tip for Mode-II damage is considerable longer than that of Mode-I which
is proportional to the fracture toughness values of each mode. For example, the
Mode-I fracture toughness of the HTS40/977-2 system is approximately 25 % of its
Mode-II fracture toughness. Therefore, Mode-I ILNS may be influential in causing
mixed-mode behaviour (Mode-I and II) by facilitating the gradual development
of subcritical damage in Q1 specimens. Additional work is required in order to
21 Damage and Failure Analysis of Bolted Joints in Composite Laminates 617
700
679.2
Bearing stress (MPa)
640.82
500
Q1 [45/0/-45/90]s Q2 [90/45/-45/0]s Q3 [0/90/45/-45]s
examine the contribution of the ILNS and interlaminar shear stresses (ILSS) to the
subcritical damage mechanisms.
Although it has been shown that the shear stress distribution is not affected by
the width of the specimen for quasi-isotropic laminates [28], a specific layer should
be considered separate from its surrounding layers due to interlaminar cracking
(delamination). When re-examined as separate layers, the width effect may become
more important with respect to the observed mixed-mode failure.
The bearing stresses at failure (average of three specimens) are given in
Fig. 21.21. The Q2W specimens show approximately 14 % and 5.6 % increased
strength over Q1W and Q3W specimens, respectively. The same trend was observed
for GFRP specimens by Quinn and Matthews [98] which was associated with the
90ı layers placed at or close to the outer surface. According to classical lamination
theory (CLT) [99], those 90ı layers produce compressive ILNS at the hole boundary
and increase the strength of the joint.
The difference in strength of the Q1W and Q1N specimens is insignificant. This
is because of the identical bearing mode of failure observed for both specimens.
The lower strengths of the Q2N and Q3N specimens in contrast to their wider
counterparts Q2W and Q3W are associated with their sudden mixed-mode failures.
The schematic of the subcritical in-plane damage planes observed from the X-
ray radiographs of the quasi-isotropic lay-ups is given in Fig 21.22. The transverse
matrix crack and the axial split planes were determined as for the cross-ply lay-ups
explained earlier. Axial split planes within the ˙45ı layers were located tangent to
hole boundary at D ˙45ı across the width of the specimen as observed from the
failed specimens.
618 A. Ataş and C. Soutis
It was shown in Sect. 4 that the success of the widely used strength prediction meth-
ods was strongly dependent on several parameters related to the laminate, geometry
and material system. A method based on subcritical damage mechanisms will be
presented in this part which is reducing the number of analytically/experimentally
determined correlation factors required for a successful strength prediction.
3-D FE models were developed with one element per layer approach using
ANSYS software [100]. Cohesive zone elements (CZEs) were embedded into
the subcritical (matrix-based) damage locations that were determined from the
X-ray radiographs in the previous section. CZEs use a strength-based failure
criterion to predict the onset of damage and a fracture mechanics-based approach to
predict its growth. The only material properties required to simulate the subcritical
damage modes were the interfacial strength and fracture energies of the particular
material system used. The strength of the joints was determined through the load-
displacement curves of the cross-ply laminates and by a simple maximum stress
criterion for the quasi-isotropic specimens. It has been shown that the effect of
various joint geometries and laminate lay-ups on the joint strength was accounted
for by the method developed.
Before attempting to predict the strength of bolted joints in CFRP laminates based
on the subcritical damage modes, it is necessary to verify the FE model predictions
using experimental measurements such as split length as a function of applied stress.
However, damage in pin-/bolt-loaded laminates initiates at 90–95 % of the ultimate
load which makes it quite difficult to measure the split/delamination growth as a
function of the applied load.
A previous experimental study [101] showed that, in a cross-ply [90ı /0ı ]s
centre-notched laminate under tensile loading, axial splitting within the 0ı layers
21 Damage and Failure Analysis of Bolted Joints in Composite Laminates 619
P b P
delamination
split
initiated at the notch tips at relatively low applied loads. Splits grew along the fibre
direction accompanied by delamination between adjacent layers. The axial splitting
and delamination were the dominant failure modes, and they generally formed
self-similar triangular shapes. An X-ray radiograph of the laminate subjected to
monotonic loading at 350 MPa (Š80 % of the failure strength) and a schematic
representation are shown in Fig. 21.23a, b, respectively [101]. The dark areas at the
notch tips extending along the x-axis, Fig. 21.23a, represent the zinc iodide solution
(ZnI2 ) that filled damaged area including axial splits (0ı splits), transverse ply
matrix cracks (90ı cracks) and delaminations between 0ı and 90ı adjacent layers.
Owing to progressive subcritical damage evolution, the experimentally observed
split length and delamination angle as a function of the applied far field stress in
[101] have been used as a metric for the verification of the FE models implementing
the subcritical damage modelling by Wisnom and Chang [96]. Similarly, a cross-ply
[90ı/0ı ]s laminate with a centre notch was modelled in order to validate the current
FE modelling approach.
The geometry, FE mesh and boundary conditions of the [90ı /0ı ]s laminate are
shown in Fig. 21.24 [96, 101]. Only one-eighth of the laminate was modelled due
to the symmetry conditions in length, width and thickness directions (thickness
symmetry is not shown in Fig. 21.24). Each composite layer, of 0.125 mm thickness,
was modelled separately using fully integrated SOLID185 linear eight-node brick
elements and was discretised with one element in the thickness direction. The elastic
material properties used are shown in Table 21.3.
The stress distribution around the notch tip was shown to be insensitive to the
exact shape of the notch in double-end-notched (DEN) CFRP specimens [102].
Therefore, the centre notch was modelled as a sharp slit of length a in the y-direction
by setting all nodes of both the 0ı and 90ı layers free to displace in all directions.
Additionally, an initial transverse crack was modelled running through the width of
the specimen in the 90ı layer starting from the notch tip due to the low transverse
tensile strength of the 90ı layer [96]. It was achieved by only applying symmetric
620 A. Ataş and C. Soutis
L/2 (mm) 45
w/2 (mm) 12
a (mm) 4
L/2
Notch tip
a
y
Symmetry in y direction
x
Fig. 21.24 The geometry, FE mesh and boundary conditions of the centre-notched [90ı /0ı ]s
laminate [96, 101]
boundary conditions to the nodes of 0ı layer and setting the nodes of the 90ı layer
free to displace as shown in Fig. 21.24.
The axial splitting can be regarded as delamination or debonding damage
between bunches of the fibres extending through-the-thickness of the unidirectional
layers. Thus, both of the subcritical damage modes were modelled using CZEs
[103]. CZEs were inserted between the 90ı and 0ı layers (all over the interface)
and within the 0ı layer (throughout its length starting from the notch tip) where
delamination and splitting damage modes have been observed experimentally [101].
The interfacial properties of the T300/914 CFRP material system, as shown in
Table 21.4, were used to model these matrix-based subcritical damage modes.
It has been shown that the inclusion of the nonlinear in-plane shear stress-strain
behaviour was necessary for accurate prediction of subcritical damage modes [96].
The experimentally determined nonlinear shear stress-strain behaviour of XAS/914
[100] was assumed to be valid for T300/914 [96] due to the similarities in polymer
resin material, volume fraction and the properties of the high-strength carbon fibres
21 Damage and Failure Analysis of Bolted Joints in Composite Laminates 621
Fig. 21.25 Comparison of FE predictions and test results for split length as a function of
applied stress: (a) linear and (b) nonlinear shear stress-strain behaviour that results into a smaller
delamination width at the notch tip (wd , delamination width; ld , delamination length)
used in both material systems. A subroutine was written [85] in order to model the
nonlinear shear stress-strain behaviour of the T300/914 material system based on
the work of Hahn and Tsai [106].
Figure 21.25 compares the extent of linear (a) and nonlinear (b) delamination
damage predictions as a function of applied stress. The progressive nature of the
damage was captured clearly. The delamination angle, under an applied stress of
400 MPa, was computed to be approximately 18ı and 6ı for linear and nonlinear
shear stress-strain behaviours, respectively. The nonlinear prediction is in a good
agreement with the experimentally observed angle of about 4ı at a failure stress of
426 MPa [101].
Delamination is driven by the splitting damage, and it is extended at the notch
tip with increased split length [92, 101]. The linear model leads to significantly
wider delamination damage than the nonlinear model as a result of the shorter split
length. High shear stresses in the linear analysis, developed across the width of the
specimen starting from the centre notch tip, resulted in the localised failure of the
622 A. Ataş and C. Soutis
not influenced by the model size. One element per laminate layer was used to model
the composite laminates and each of the 0.25 mm thick layers were modelled using
SOLID185 linear eight-node brick elements. The edge distance (e) of the specimens
was 18 mm, and the width (w) was set to 18 mm and 36 mm depending upon the
specimen design as detailed in the following sections.
In conventional 3-D FE models of strength prediction methods, such as progres-
sive failure analysis (PFA) [10], all the laminate layers are modelled with the same
mesh structure and bonded together at their interface since the subcritical damage
planes are ignored. However, when the subcritical damage modes are considered,
each layer has to be meshed with a special structure due to the different locations
of subcritical damage modes that are dependent on the orientation angle of the
layers. Commonly used composite laminates are generally composed of 0ı , 90ı
and ˙45ı layers. Therefore, the subcritical damage locations of each of these layers
were determined by the X-ray radiography for the purpose of the present modelling
approach. The same subcritical damage locations obtained for HTS40/977-2 can be
used for other CFRP and GFRP material systems and various joint configurations
when the respective interlaminar strength, critical fracture energies and nonlinear
shear stress-strain behaviour are provided.
In the bolted joint experiments, the clamping force was transferred by means of
the washers between the composite laminate and the steel plates [84, 85]. Therefore,
only the washer and bolt shank were considered in the models, i.e. the steel plates
and bolt head/nut were omitted. The bolt shank and the washer (will be referred to
as the bolt hereafter) were modelled as a single steel solid deformable body (see
Fig. 21.27) using SOLID185 linear eight-node brick elements [107, 108]. The steel
properties were E D 210 GPa and D 0.3. The contact algorithms were defined
between (a) the bolt (TARGE170) and the laminate hole boundary (CONTA174)
and (b) the bolt (TARGE170) and the CFRP laminate top surface (CONTA174) [85,
100, 107]. An incremental displacement was applied to the bolt along the (negative)
x-direction in order to load the specimen.
The clearance between the bolt and the laminate was set to the experimentally
measured value of 0.05 mm. The coefficient of friction was assumed to be 0.2 in
all contact surfaces [34]. An augmented Lagrangian contact algorithm was used to
check the amount of penetration against the allowable tolerance. It is worth noting
that the CONTA174 element is a higher-order element, and its mid-nodes were
dropped to comply with the eight-node SOLID185 elements of the laminate.
Equivalent linear elastic orthotropic material properties [109] (see Table 21.5)
were assigned to each unidirectional composite layer according to predefined local
coordinate systems which represent the individual orientation angles. The fibre
direction of the 0ı layer coincides with the loading direction. Strengths of the
HTS40/977-2 CFRP unidirectional layer are given in Table 21.6.
The interfacial properties of HTS40/977-2 CFRP laminates are given in
Table 21.7 which were adopted from the T300/977-2 CFRP material system.
These systems exhibit similar elastic material properties since the high-strength
HTS40 and T300 carbon fibres have similar properties, and both use the same
624 A. Ataş and C. Soutis
Table 21.7 Interfacial properties used for HTS40/977-2 adopted from T300/977-2
[104, 110]
GIC (N/mm) GIIC (N/mm) max (MPa) max (MPa) Kn (N/mm3 ) Kt (N/mm3 )
0.352 1.45 60 80 1 105 1 105
toughened epoxy matrix [110]. These properties were used to model both the axial
splits (within the 0ı and ˙45ı layers) and subcritical delamination damage modes
through the CZEs.
The importance of experiments for modelling the subcritical damage modes
was seen in our earlier modelling efforts on pin/bolt joints in [0ı /90ı ]s cross-
ply laminates loaded in tension [9, 111–113]. The only subcritical damage mode
considered in those models was delamination. In-plane mesh structures created
without considering the transverse matrix cracks in 90ı layers and axial splits within
the 0ı layers are shown in Fig. 21.27.
21 Damage and Failure Analysis of Bolted Joints in Composite Laminates 625
2.45% 2.6%
P P
3.0% 5.0%
P P
Damaged
Undamaged
Fig. 21.28 Predicted delamination onset and growth at the 0ı /90ı interface of the [0ı /90ı ]s
laminate as a function of pin/bolt displacement [9, 85, 111, 112]
Merged nodes
Merged nodes
CZEs Duplicate
areas
created along the white vertical dashed lines in Fig. 21.32a in order to model the
transverse cracks. Transverse cracks occur at early stages of the loading process,
and it was assumed that they were fully developed at the beginning of the analyses
due to the weak matrix-dominated transverse strength of the 90ı layers [96]. Due to
these duplicate areas, the 90ı layer would not resist any loading if the layer alone
were to be loaded by a frictionless pin. However, when located in a multidirectional
laminate, part of the total load of the joint is transferred to the 90ı layers by adjacent
layers through shear stresses at the interfaces. In this manner, 90ı layers contribute
to the load-carrying capability of the joints.
Figure 21.33 shows the FE volume discretisation and mesh structure of 45ı layer
including the subcritical damage planes. A similar FE structure was constructed
628 A. Ataş and C. Soutis
for 45ı layer. Based on the observations on the X-ray radiographs, it was assumed
that the splitting damage in the ˙45ı layers developed tangentially to the hole
edge at angles of ˙45ı with respect to the loading direction. As for the 0ı layers,
the volumes were discretised 0.25 mm away from the hole edge in order to avoid
irregular element shapes. The lighter and darker areas represent the areas where
fully integrated and reduced integration elements were used, respectively.
The in-plane subcritical damage planes have been the focus thus far. Additionally,
delamination damage planes were defined between all adjacent layers throughout
the whole interface. It is noted that each individual layer has a specific mesh
21 Damage and Failure Analysis of Bolted Joints in Composite Laminates 629
structure which does not conform to the adjacent layers. The use of CZEs based
on the standard contact algorithms enabled the modelling of such damage planes
with different mesh structures [100].
It was possible to confine the delamination damage planes to the areas where the
delamination damage was actually observed by merging the nodes at the interfaces
of the adjacent layers outside of those areas. However, such a modelling procedure
would render the general applicability of the method impractical since a peculiar
interface is created whenever two different layers come into contact. Furthermore,
merging between thin laminate layers is a complex operation that is prone to
erroneous definitions. Therefore, a compromise between the modelling complexity
and the solution time was required. Based on the experiences developed over the
course of the present study, the chosen approach was to model the delamination
damage planes all over the interface.
It is practical to model bolted joints of various laminate lay-ups (consisting of
0ı , 90ı and ˙45ı layers), stacking sequences and joint geometries after having
modelled the individual layers, including the subcritical damage planes. In order
630 A. Ataş and C. Soutis
to facilitate the modelling procedure, a template file consisting of the models of all
the four individual layers was created in the ANSYS software. This file was then
manipulated according to the specifications of the modelled joints.
The global mode of failure in bolted joints of cross-ply laminates was shown to
be shear out in Sect. 5.2.1, which is a matrix-dominated mode of failure. In this
failure mode, a section of the specimen ahead of the fastener, generally as wide as
the bolt diameter, is sheared out with increased displacement (see Fig. 21.6). The
shear-out mode is realised with the accumulation of subcritical damage modes (axial
splits in 0ı layers, transverse matrix cracks and delamination). Transverse matrix
cracks develop at early loading stages in the 90ı layers around the hole boundary
due to high stress concentrations. Splits then start to form within the 0ı layers
with increased loading. Finally, those splits drive the delamination between the
consecutive laminate layers resulting in the global shear-out mode of failure. CZEs
simulate all of those entirely shear-dominated subcritical damage modes. Therefore,
predicted load-displacement curves were directly used to define the strength of the
joints in cross-ply specimens.
Figure 21.34 shows the predicted load-displacement curve of the [90ı /0ı ]s
specimen (B1W) as an example. The trend of the curve is quite similar to its
experimental counterpart which was shown in Fig. 21.5. The load increases in a
nonlinear fashion up to point A, due to the initiation of the subcritical damage modes
at early stages of loading, where the maximum load of the joint is reached. After that
point, the load decreases with increased displacement since the subcritical damage
modes become unstable and grow without an increase in the applied load.
The bearing strength of the bolted joints in cross-ply specimens was calculated
using Eq. 21.2 and was based on the maximum load carried by the joints (Pmax )
which was obtained from the numerically predicted load-displacement curve.
Figures 21.35 and 21.36 show the predicted delamination damage at the 90ı /0ı
interface and the splitting damage within the 0ı layers of the B1W ([90ı/0ı ]s )
specimen at maximum load, around point A, in Fig. 21.34. The approximate length
of the delamination (ld ) and the splitting damages (ls ) are 6 and 10 mm, respectively.
Thus, it is obvious that the splitting drives the delamination damage as suggested in
earlier works [92, 96, 102, 114].
The experimental bearing strength of the joints was determined from the load-
displacement curves after a significant load drop was observed according to ASTM
standard D 5961 [84]. Therefore, the X-ray radiographs of the failed specimens
correspond to a load level after the maximum peak in which the splits and other
632 A. Ataş and C. Soutis
subcritical damage modes were extended up to the free edges of the specimens.
This is the reason for the relatively short length of the delamination in Fig. 21.35,
which was captured at the maximum load but not the ultimate failure load. In order
to demonstrate the suitability of the modelling approach, Fig. 21.37 was captured
when the delamination reached the free edge as the bolt displacement increased.
This agrees well with the prediction from the X-ray radiograph in Fig. 21.6 (ii).
Figure 21.38 shows the splitting damage within the 0ı layers of the B3W
specimen. The length of the split within the first 0ı layer is significantly shorter
than that of the second 0ı layer (the second 0ı layer is at the specimen midplane).
This difference may be explained by the number of delamination planes associated
with each splitting plane as the delamination is driven by the splitting damage. The
second 0ı layer is connected to only one delamination plane, whereas the first 0ı
21 Damage and Failure Analysis of Bolted Joints in Composite Laminates 633
600
553.92
Bearing Strength (MPa)
516.46
500 482.53
445.15
451.18
412.3 402.54 407.14
390.7
400
344.14
319.47
300 263
200
B1W B2W B3W
Fig. 21.39 Predicted and experimental bearing strengths for the cross-ply lay-ups (e/d D 3,
w/d D 6)
Fig. 21.40 xx stress distribution (MPa) for the 0ı layer in the Q1W ([45ı /0ı /45ı /90ı ]s )
specimen at the load step in which the ultimate failure load, Pmax D 7331.4 N, was predicted
It is well known that the width of the specimens has a significant effect on the
strength and mode of failure in bolted joints of FRP laminates [28]. This is mainly
due to the sensitivity of the tensile stresses at the net-tension plane ( D ˙90ı ) to
decreased width. The longitudinal stress ( xx ) distribution in the 0ı layer of the
Q1N specimen (e/d D 3, w/d D 3) at the load step in which the ultimate load was
predicted is shown in Fig. 21.45. The mode of failure was predicted as bearing
since the compressive stress at the loaded part of the hole boundary exceeded the
compressive strength.
Although the maximum tensile stress around D ˙90ı is below the tensile
strength value, it is significantly higher than that of the Q1W specimen (see
636 A. Ataş and C. Soutis
Fig. 21.42 Bilinear softening response of the completely failed node at the tip of the splitting
plane, depicted by A, in Fig. 21.41
Fig. 21.43 Shear stresses (MPa) at the 45ı /90ı interface of the Q1W specimen corresponding
to the load step in Fig. 21.42, Pmax D 7331.4 N
Fig. 21.40) (e/d D 3, w/d D 6). This is consistent with the development of the net-
tension mode of failure with the reduced width of the specimens. However, the
high tensile strength of the HTS40/977-2 prevents a net-tension mode of failure. It
is worth noting that the stress value ( 2370 MPa) might cause a pure net-tension
mode of failure in some other CFRP material systems, such as T300/914C which
has a tensile strength of approximately 1700 MPa [101].
Figures 21.46 and 21.47 show the current model predictions, linear and nonlinear
PFA predictions [85] and experimental (average of three specimens) bearing
strengths of the wide (e/d D 3, w/d D 6) and narrow (e/d D 3, w/d D 3) quasi-
isotropic specimens, respectively.
21 Damage and Failure Analysis of Bolted Joints in Composite Laminates 637
-1000
-1500
-2000
-2500
0.021
0.042
0.063
0.084
0.105
0.126
0.147
0.168
0.189
0.231
0.252
0.273
0.294
0.315
0.336
0
0.21
Bolt displacement (mm)
Fig. 21.44 Nonlinear nature of the compressive stress at the 0ı layer of the Q1W specimen at
the bearing plane ( D 0˚) as a function of the bolt displacement (dotted line represents the linear
straight line)
Fig. 21.45 xx stress distribution (MPa) for the 0ı layer in the Q1N ([45ı /0ı /45ı /90ı ]s )
specimen at the load step in which the ultimate failure load, Pmax D 7437.9 N, was predicted
The current model predictions generally provide better correlation with the
experimental data in comparison to the PFA predictions. Inclusion of a thermal
analysis into the current modelling approach is possible and could provide better
strength predictions by modifying the initiation stress of the matrix-dominated
damage modes.
21.7 Conclusions
The success of currently used strength prediction methods for bolted joints in
composite laminates is strongly dependent on several laminate, geometry and
material system-related parameters. This dependency stems from the variations in
the extent of the subcritical damage modes and is generally compensated with
638 A. Ataş and C. Soutis
800
763.5 752.31
500
421
400
Q1W [45/0/-45/90]s Q2W [90/45/-45/0]s Q3W [0/90/45/-45]s
Fig. 21.46 Predicted and experimental bearing strengths of the wide quasi-isotropic specimens
(e/d D 3, w/d D 6)
700
638.58
Bearing Strength (MPa)
523.25
495.78 506.68
500
421.84
392.84
400
300
Q1N [45/0/-45/90]s Q2N [90/45/-45/0]s Q3N [0/90/45/-45]s
Fig. 21.47 Predicted and experimental bearing strengths of the narrow quasi-isotropic specimens
(e/d D 3, w/d D 3)
References
49. P.A. Smith, M.F. Ashby, K.J. Pascoe, Modeling clamp-up effects in composite bolted joints.
J. Compos. Mater. 21(10), 878–897 (1987)
50. F.K. Chang, R.A. Scott, G.S. Springer, Strength of mechanically fastened composite joints.
J. Compos. Mater. 16, 470–494 (1982)
51. F.K. Chang, R.A. Scott, G.S. Springer, Failure of composite laminates containing pin loaded
holes—method of solution. J. Compos. Mater. 18(3), 255–278 (1984)
52. H.A. Whitworth, O. Aluko, N.A. Tomlinson, Application of the point stress criterion to the
failure of composite pinned joints. Eng. Fract. Mech. 75(7), 1829–1839 (2008)
53. H.A. Whitworth, M. Othieno, O. Barton, Failure analysis of composite pin loaded joints.
Compos. Struct. 59(2), 261–266 (2003)
54. J.M. Whitney, R.J. Nuismer, Stress fracture criteria for laminated composites containing
stress-concentrations. J. Compos. Mater. 8, 253–265 (1974)
55. L.J. Hart-Smith, Bolted joint analyses for composite structures—current empirical methods
and future scientific prospects, in Proceedings of the Joining and Repair of Composite
Structures, eds. by K.T. Kedward, H. Kim (ASTM International, West Conshohocken, PA,
2004), pp. 127–160
56. J.M. Hundley et al., Three-dimensional progressive failure analysis of bolted titanium-
graphite fiber metal laminate joints. J. Compos. Mater. 45(7), 751–769 (2010)
57. A. Evcil, Progressive failure analysis of pin joints in composite laminates, in Computational
Methods and Applied Computing, 2008, pp. 404–411
58. J.H. Kweon, S.Y. Shin, J.H. Choi, A two-dimensional progressive failure analysis of pinned
joints in unidirectional-fabric laminated composites. J. Compos. Mater. 41(17), 2083–2104
(2007)
59. A. Riccio, Effects of geometrical and material features on damage onset and propagation in
single-lap bolted composite joints under tensile load: Part II—numerical studies. J. Compos.
Mater. 39(23), 2091–2112 (2005)
60. Y. Xiao, T. Ishikawa, Bearing strength and failure behavior of bolted composite joints (part
II: modeling, and simulation). Compos. Sci. Technol. 65(7-8), 1032–1043 (2005)
61. B.M. Icten, R. Karakuzu, Progressive failure analysis of pin-loaded carbon-epoxy woven
composite plates. Compos. Sci. Technol. 62(9), 1259–1271 (2002)
62. X.L. Qing et al., Damage-tolerance-based design of bolted composite joints. Compos Struct
Theory Pract 1383, 243–272 (2001)
63. L.B. Lessard, M.M. Shokrieh, 2-dimensional modeling of composite pinned-joint failure.
J. Compos. Mater. 29(5), 671–697 (1995)
64. C.L. Hung, F.K. Chang, Bearing failure of bolted composite joints. Part II: model and
verification. J. Compos. Mater. 30(12), 1359–1400 (1996)
65. Z.Q. Wang et al., Progressive failure analysis of bolted single-lap composite joint based on
extended finite element method. Mater. Des. 37, 582–588 (2012)
66. K. Hollmann, Failure analysis of bolted composite joints exhibiting in-plane failure modes.
J. Compos. Mater. 30(3), 358–383 (1996)
67. W.G. Bickley, The distribution of stress round a circular hole in a plate. Philos. Trans. R. Soc.
Lond. A 227, 383–415 (1928)
68. J.H. Kweon, H.S. Ahn, J.H. Choi, A new method to determine the characteristic lengths of
composite joints without testing. Compos. Struct. 66(1-4), 305–315 (2004)
69. C.T. Sun, S.E. Yamada, Strength distribution of a unidirectional fiber composite. J. Compos.
Mater. 12, 169–176 (1978)
70. S.E. Yamada, C.T. Sun, Analysis of laminate strength and its distribution. J. Compos. Mater.
12, 275–284 (1978)
71. F.K. Chang, The effect of pin load distribution on the strength of pin loaded holes in laminated
composites. J. Compos. Mater. 20(4), 401–408 (1986)
72. B.L. Agarwal, Static strength prediction of bolted joint in composite-material. AIAA
J. 18(11), 1371–1375 (1980)
73. A. Olmedo, C. Santiuste, On the prediction of bolted single-lap composite joints. Compos.
Struct. 94(6), 2110–2117 (2012)
21 Damage and Failure Analysis of Bolted Joints in Composite Laminates 643
74. C. Huhne et al., Progressive damage analysis of composite bolted joints with liquid shim
layers using constant and continuous degradation models. Compos. Struct. 92(2), 189–200
(2010)
75. G.N. Labeas et al., Adaptative progressive damage modeling for large-scale composite
structures. Int. J. Damage Mech. 21(3), 441–462 (2012)
76. F.X. Irisarri et al., Progressive damage and failure of mechanically fastened joints in
CFRP laminates. Part II: failure prediction of an industrial junction. Compos. Struct. 94(8),
2278–2284 (2012)
77. Z. Hashin, Failure criteria for unidirectional fiber composites. J. Appl. Mech. 47(2), 329–334
(1980)
78. S.C. Tan, A progressive failure model for composite laminates containing openings. J. Com-
pos. Mater. 25(5), 556–577 (1991)
79. H.T. Sun, F.K. Chang, X.L. Qing, The response of composite joints with bolt-clamping loads,
Part II: Model verification. J. Compos. Mater. 36(1), 69–92 (2002)
80. I. Shahid, F.K. Chang, An Accumulative damage model for tensile and shear failures of
laminated composite plates. J. Compos. Mater. 29(7), 926–981 (1995)
81. T.E. Tay et al., Progressive failure analysis of composites. J. Compos. Mater. 42(18),
1921–1966 (2008)
82. I. Eriksson, C.G. Aronsson, Strength of tensile loaded graphite epoxy laminates containing
cracks, open and filled holes. J. Compos. Mater. 24(5), 456–482 (1990)
83. A. Jumahat et al., Fracture mechanisms and failure analysis of carbon fibre/toughened epoxy
composites subjected to compressive loading. Compos. Struct. 92(2), 295–305 (2010)
84. ASTM, D 5961/D 5961M—01 Standard Test Method for Bearing Response of Polymer Matrix
Composite Laminates, United States, 2001
85. A. Atas, Strength Prediction of Mechanical Joints in Composite Laminates Based on
Subcritical Damage Modelling, Ph.D. Thesis, Department of Mechanical Engineering, The
University of Sheffield, Sheffield, 2012, p. 196
86. C. Soutis, N.A. Fleck, Static compression failure of carbon-fiber T800/924c composite plate
with a single hole. J. Compos. Mater. 24(5), 536–558 (1990)
87. C. Soutis, N.A. Fleck, P.T. Curtis, Hole–hole interaction in carbon-fiber epoxy laminates
under uniaxial compression. Composites 22(1), 31–38 (1991)
88. C. Soutis, J. Lee, Scaling effects in notched carbon fibre/epoxy composites loaded in
compression. J. Mater. Sci. 43(20), 6593–6598 (2008)
89. J. Lee, C. Soutis, Measuring the notched compressive strength of composite laminates:
specimen size effects. Compos. Sci. Technol. 68(12), 2359–2366 (2008)
90. J. Lee, C. Soutis, A study on the compressive strength of thick carbon fibre-epoxy laminates.
Compos. Sci. Technol. 67(10), 2015–2026 (2007)
91. J. Lee, C. Soutis, Thickness effect on the compressive strength of T800/924C carbon fibre-
epoxy laminates. Compos. Part A Appl. Sci. Manuf. 36(2), 213–227 (2005)
92. M.T. Kortschot, P.W.R. Beaumont, Damage mechanics of composite-materials. Part I:
measurements of damage and strength. Compos. Sci. Technol. 39(4), 289–301 (1990)
93. P.A. Smith, K.J. Pascoe, Behaviour of Bolted Joints in [0/90]ns Laminates, Cambridge
University, CUED/C-mat./TR.121, 1985
94. P.A. Smith et al., The behavior of single-lap bolted joints in Cfrp laminates. Compos. Struct.
6(1-3), 41–55 (1986)
95. J.F. Mandell, S.S. Wang, F.J. Mcgarry, Extension of crack tip damage zones in fiber reinforced
plastic laminates. J. Compos. Mater. 9, 266–287 (1975)
96. M.R. Wisnom, F.K. Chang, Modelling of splitting and delamination in notched cross-ply
laminates. Compos. Sci. Technol. 60(15), 2849–2856 (2000)
97. P.A. Smith, K.J. Pascoe, The effect of stacking-sequence on the bearing strengths of quasi-
isotropic composite laminates. Compos. Struct. 6(1-3), 1–20 (1986)
98. W.J. Quinn, F.L. Matthews, Effect of stacking sequence on pin-bearing strength in glass-fiber
reinforced plastic. J. Compos. Mater. 11, 139–145 (1977)
644 A. Ataş and C. Soutis
99. R.M. Jones, Mechanics of Composite Materials, 2nd edn. (Taylor & Francis, Philadelphia,
PA, 1999), p. 519
®
100. ANSYS Academic Research, Release 12.1, 2009
101. S.M. Spearing, P.W.R. Beaumont, Fatigue damage mechanics of composite-materials. Part
I: experimental-measurement of damage and post-fatigue properties. Compos. Sci. Technol.
44(2), 159–168 (1992)
102. M.T. Kortschot, P.W.R. Beaumont, Damage mechanics of composite-materials. II: a
damaged-based notched strength model. Compos. Sci. Technol. 39(4), 303–326 (1990)
103. A. Ataş, C. Soutis, Strength prediction of bolted joints in CFRP composite laminates using
cohesive zone elements. Compos. Part B Eng. 58, 25–34 (2014)
104. P.W. Harper, S.R. Hallett, Cohesive zone length in numerical simulations of composite
delamination. Eng. Fract. Mech. 75(16), 4774–4792 (2008)
105. M.R. Wisnom, J. Haberle, Prediction of buckling and failure of unidirectional carbon-fiber
epoxy struts. Compos. Struct. 28(3), 229–239 (1994)
106. H.T. Hahn, S.W. Tsai, Nonlinear elastic behavior of unidirectional composite laminae.
J. Compos. Mater. 7, 102–118 (1973)
107. R.H. Oskouei, M. Keikhosravy, C. Soutis, Estimating clamping pressure distribution and
stiffness in aircraft bolted joints by finite-element analysis. Proc. Inst. Mech. Eng. G J. Aerosp.
Eng. 223(G7), 863–871 (2009)
108. J. Montgomery, Methods for modeling bolts in the bolted joint, in ANSYS User’s Conference,
2002.
109. S. Heimbs et al., Low velocity impact on CFRP plates with compressive preload: test and
modelling. Int. J. Impact Eng. 36(10–11), 1182–1193 (2009)
110. V.K. Goyal, E.R. Johnson, C.G. Davila, Irreversible constitutive law for modeling the
delamination process using interfacial surface discontinuities. Compos. Struct. 65(3-4),
289–305 (2004)
111. A. Atas, G.F. Mohamed, C. Soutis, Effect of clamping force on the delamination onset and
growth in bolted composite laminates. Compos. Struct. 94(2), 548–552 (2012)
112. A. Atas, C. Soutis, Modelling of interlaminar delamination around a pin loaded composite
joint, in Joint conference of Deformation and Fracture of Composites/5th Structural Integrity
of Composite Materials (DFC11/SI5), Cambridge, UK, 2011
113. A. Atas, C. Soutis, Effect of clamping force on the delamination onset and growth in bolted
composite laminates, in 16th International Conference on Composite Structures (ICCS16),
Porto, 2011
114. M.T. Kortschot, P.W.R. Beaumont, M.F. Ashby, Damage mechanics of composite-materials.
III: prediction of damage growth and notched strength. Compos. Sci. Technol. 40(2), 147–165
(1991)
115. S.R. Hallett et al., An experimental and numerical investigation into the damage mechanisms
in notched composites. Compos. Part A Appl. Sci. Manuf. 40(5), 613–624 (2009)
116. W.G. Jiang, S.R. Hallett, M.R. Wisnom, Modelling of damage in composite materials using
interface elements, in 5th Europian LS-DYNA Users Conference, Birmingham, 2005
Chapter 22
Interfaces, Cracks and Toughness: City Cars
Made from Composites
The city car has major problems: the fuel consumption over the years has not
dropped as fast as might be expected from technical progress on engines and
improved control systems. Emissions, both of carbon dioxide and of toxic exhaust
(SOx, COx, NOx, VOCs and nanoparticles), have been pushed down by legislation
dependent on emission catalysts since the 1970s, but the problem has been the car
manufacturers who have subverted the tests in various ways which mean that life-
threatening fumes are increasing in most cities. Recent reports (Fig. 22.1) suggest
that about 8 % of cars exceeded the emission regulations in 2001, whereas more
than 40 % infringed in 2014 [5–7]. This problem became difficult to ignore when
on-road NOx data for Volkswagen cars were found in 2015 to be 40 times higher
than official test figures [5].
This problem suggests that end-of-pipe solutions to emissions are not reliable
and sustainable because loopholes through regulations will eventually be found.
The conclusion is that fundamental improvements in car emissions must be sought,
for example, by using lightweight composites.
Another interesting problem is that cars appear to be getting heavier as models
evolve as shown by the rising weight curve for the Honda Accord shown in Fig. 22.2
[8, 9]. The car rose from 1 tonne in 1980 to 1.5 tonne in 2005, a 50 % rise. This rise
J. Jostins
Coventry University, Coventry, UK
Microcab, Coventry, UK
K. Kendall ()
University of Birmingham, Birmingham B15 2TT, UK
Adelan, Birmingham, UK
e-mail: K.Kendall@bham.ac.uk
10%
0%
2000 2005 2010 2015
year
in consumer product is contrasted with the Formula 1 racing car decrease during the
same period. The graph suggests that the market trend is to heavier cars, whereas
the performance trend is to lighter vehicles. Later, we discuss in detail the need
for lighter, stronger, stiffer materials such as carbon fibre composites in racing car
competition.
The Honda Accord is not the worst example of auto-bloat with time; the Mini is
perhaps the most famous, starting at 616 kg in 1958, close to F1 weight, but rising
by almost a factor three to 1465 kg in 2010.
There is a movement to electric drive and also to fuel cells with hydrogen to cut
emissions. But what is less well known is the effect that lightweight materials can
have on energy usage and pollution output. This chapter explains the background
to these issues and then goes on to show how fracture mechanics, composite
moulding experimentation and green hydrogen fuel cell electric drivetrains have
come together over the last 40 years to create a new direction in city car concepts.
First consider the effect of weight on emissions.
The curve and points in Fig. 22.3 show the fuel consumption for 3800 different
types of gasoline combustion cars in China within years 2000–2010. The energy
consumption in MJ/km, essentially the average friction force in kN, was plotted
against the weight in kN to reveal a linear dependence with an intercept. It is
clear that there is some scatter, especially for heavier cars where fuel costs are less
important for luxury buyers. But surprisingly, the results prove that the resistance to
movement, therefore emissions, depends only on drivetrain and weight [1–6]. This
is essentially Coulomb’s friction law. A Jaguar car is shown as an example point.
22 Interfaces, Cracks and Toughness: City Cars Made from Composites 647
Fig. 22.3 Effect of weight on energy consumption for 3800 petrol cars in China 2000–2010 [1]
Weight is the key parameter. At the same design, changing from steel to carbon
fibre should produce a fuel and emissions saving almost linear with the density ratio
steel/composite D 7700/1900 D 4. Therefore it is important to consider the source
of benefits due to composite microstructure. Elastic modulus and interface fracture
energy are shown to be very significant. Then it is essential to consider the benefits
demonstrated in racing experiments before showing the present novel applications
to prototype city cars, especially those with zero-emission drivetrains.
The benefits of composite materials are substantial: the moulding and shaping can be
simple, the components can be lightweight and economic and the resulting solid can
be stiff, strong and tough, that is, resistant to cracks, under certain circumstances.
With the right composition and processing, composites can exhibit very high energy
to fracture. This is shown below to stem from the many interfaces which a crack
must break through to destroy the component. In fact, the main feature visible in
composite materials is the pattern of interfaces between fibres and matrix, stemming
from the mixing and subsequent adhesion of the components to give a fully dense
and coherent material. First we consider the background to composites and then
move on to assess two causes of crack resistance: (1) the elastic modulus mismatch
and (2) the crack defection caused by brittle interfaces. It is essential that these
mechanisms are considered in the final composite car components to be described
afterwards.
648 J. Jostins and K. Kendall
Fig. 22.4 Safety glass fracture showing how the laminate holds together despite crack formation
Composites in cars started with French chemist Edouard Benedictus who in 1903
found that a glass flask used for boiling cellulose nitrate polymer solution would
crack but not shatter when dropped on the floor (Fig. 22.4). He patented his Triplex
invention in 1909 [11], but Le Carbone corporation had filed a patent on plastic-
coated glass in 1902 [12]. Later development of better polymers like polyvinyl
butyral and ethylene vinyl acetate allowed such laminated safety glass to be used
on all automobile windscreens by 1930 when Britain introduced legal requirements
for cars to have composite glass screens [13]. There was no doubt that the interface
between glass and polymer was the cause of this ‘toughening’ effect. The cracks
propagated along and through the glass sheet but could not cross the interface into
the tough polymer layer, typically 0.38 mm thick, from the 2.5 mm thick glass outer
sheets. Thus, cracks in composites exist but are inhibited by the interfaces.
The use of fibres to reinforce plastics, either glass fibre-reinforced plastic (GFRP)
or later carbon fibre-reinforced plastic (CFRP), began in the 1930s after fibreglass
wool became available on an industrial scale from blowing air into a glass melt,
originally discovered by Slayter at Owens-Illinois [14]. Later, suitable resins were
found to glue the fibres into a solid sheet. Various products including boats were
prototyped, and the first car to have a GFRP body was a 1946 version of the Stout
Scarab, but the model did not enter production [15]. A more recent description
of glass fibres in vehicles is found in reference [16]. The mechanics of how such
composite materials functioned started with Griffith in 1920 who showed that melt-
drawn glass fibres tended to become very strong as the fibres were pulled down to
finer diameters. Griffith thought of the cracking process in a sheet of ideal material
as propagation of an elliptical hole shown in Fig. 22.5.
From an energy-based theory of extending the surface of this ellipse in the sheet
of material, Griffith created his famous equation for strength of a cracked material
([17] equation corrected later)
D .ER= c/1=2 (22.1)
22 Interfaces, Cracks and Toughness: City Cars Made from Composites 649
elliptical crack
a b c
s, E, R, c E1 E2 R Ri R
Fig. 22.6 (a) Griffith edge crack; (b) edge crack hitting high modulus material; (c) edge crack
meeting interface with low fracture energy Ri
where E is the elastic modulus, R is the fracture surface energy and 2c is the length
of the central crack in an infinite plane sheet. His argument was that large cracks
could not exist in fine fibres so that strength must therefore be large according to
Eq. (22.1).
Griffith’s leap of theoretical analysis, showing that strength was not a funda-
mental material property but limited by defects c, elasticity E and fracture energy R
(sometimes called fracture toughness [also used to describe (ER)1/2 ]), had a defining
influence on the understanding of materials from that date. There was really no limit
to strength; you could make materials more perfect, stiffer or tougher to raise the
stress to failure. What was then needed was an extension of this model to show the
effect of interfaces on failure. This was not revealed until 1975–1976 when Griffith’s
energy balance theory was applied successfully to a crack crossing a single interface
[18–22]. The fundamental problem is shown in Fig. 22.6.
Figure 22.6a shows the Griffith model slightly modified to apply to an edge
crack, which essentially follows Eq. (22.1). This crack propagates rapidly and
catastrophically through the brittle sheet once the critical applied tensile stress is
reached. Figure 22.6b shows a simplified picture of a composite in which the crack
is stopped at an interface into a material of higher elastic modulus E2 . Figure 22.6c
shows a composite in which the two materials on each side are identical, but the
interface fracture energy is reduced to Ri , making the interface more brittle than the
bulk material, diverting the crack sideways.
650 J. Jostins and K. Kendall
a interface TV camera
b
glass plate
interface
material 1 material 2 material 2
material 1
F
Fig. 22.7 (a) Situation to be modelled; (b) peel crack model experiment observing crack through
the glass to measure crack speed
where F is the peel force, b the width of the strip and R the fracture surface energy.
Equation (22.2) shows that the peel crack travels at constant speed because the
driving force does not depend on crack length as in the Griffith case. Consequently, it
was easy to show experimentally that the increase in elastic modulus at the interface
slowed down and even stopped the crack. Mathematical analysis [18] confirmed the
experimental result that, to keep the crack going at constant speed, the peeling force
F needed to increase at the interface to
F D bRE2 =E1 (22.3)
22 Interfaces, Cracks and Toughness: City Cars Made from Composites 651
showing that, for a crack moving from a plastic resin of E1 D 4 GPa into glass of
E2 D 70 GPa, 17 times more force was needed or, better yet, 100 times more force
for high modulus carbon fibres, the first time that the elastic crack stopping effect
had ever been fully demonstrated.
The second dramatic effect of an interface (Fig. 22.6c) is that it can deflect
the crack if it is more brittle than the bulk material. An experiment using brittle
rubber was set up to demonstrate this as shown in Fig. 22.8 (top). The crack travels
through a rubber block to meet an interface at distance c. It is travelling in bulk
material of cohesive fracture energy Rc , but when it experiences the interface of
lower fracture energy Ri , it tends to move sideways as well as forwards. The question
to be answered is ‘What is the criterion for this critical deflection condition’? This
was the question posed initially by Cook and Gordon [24] in 1964, but they thought
that the interface must be five times weaker than the bulk, which is not a rational
statement according to the Griffith theory. By carrying out the experiment shown
in Fig. 22.8 (bottom), an interface of controlled fracture energy could be prepared,
and then the edge crack could be started with a razor blade and observed crossing
or deflecting at the interface. Ethylene propylene rubber was an ideal material to
crack transition
adhere
razor slit
stress
Fig. 22.8 Top crack meeting an interface which is just so brittle that it travels equally forwards and
sideways along the interface; bottom experimental procedure for making and testing the interface
652 J. Jostins and K. Kendall
model this effect because of its brittleness, its transparency making the crack easily
visible and its hot cross-linking moulding process. By putting together a sheet and
a block of raw rubber, the sheet could be adhered to the side of the block (Fig. 22.8,
bottom) and then cross-linked in a press mould at 140 ı C for 1 h. The interface had
then disappeared, and the block and strip had interdiffused completely, so the razor
crack went straight through. But when the block and strip were separately cross-
linked fully and then adhered together and tested, the crack was observed to deflect
because the adhesion was lower along the interface than through the bulk rubber. By
doing a number of tests in which the block and strip were partly cross-linked before
adhering and then fully cross-linked before cracking, the adhesive fracture energy
was varied systematically, and the point at which crack deflection just occurred was
found [19].
Both the cohesive and adhesive fracture energies were measured in standard tear
and peel tests. The measurements of the cohesive fracture energy are shown in
Fig. 22.9 as the dark line on the plot of fracture surface energy vertically against
crack speed on the horizontal axis, both on logarithmic scales. It can be seen that
the rubber was quite brittle, with a fracture energy of 1000 Jm2 at a crack speed
of 100 ms1 , with toughness rising as crack speed increased. Therefore, the crack
speed had to be maintained constant to standardise the results. 34 ms1 was chosen
as a suitable experimental speed, and this was controlled by adjusting the uniform
tension within the rubber test block using the tensile test machine.
The interface adhesion is shown as the circular experimental points, starting with
the top point on the cohesive line where the adhesion fracture energy was equal to
Interface fracture
10
20 200 2000
crack speed mm/s
Fig. 22.9 Diagram showing how the transition to interface fracture occurred as the interface
fracture energy was decreased at constant crack speed. The transition from cohesive to adhesive
fracture was clearly defined experimentally
22 Interfaces, Cracks and Toughness: City Cars Made from Composites 653
the cohesive fracture energy. In this case the crack travelled straight through the
interface as indicated by the diagram on the right of Fig. 22.9. Then the adhesion
was gradually reduced, as shown by the circular points dropping to lower values.
At an adhesion fracture energy value of about 100 Jm2 , the first indication was
seen that the crack was tending to travel along the interface as well as straight
through, illustrated by the diagram on the right of Fig. 22.9. This was the transition
point at which the cohesive and adhesive cracks travelled equally. Lower values of
interface adhesion gave crack deflection along the interface and no penetration of the
interface. Figure 22.9 shows this as black triangular points and the bottom diagram
on the right.
The experimental ratio of Rc /Ri was measured as 9.1 giving good agreement with
the theoretical prediction of 9.1 from reference [19] of
Rc =Ri D 4 1 2 (22.4)
where is Poisson’s ratio required for the plane strain case. Equation (22.4) was
derived by dividing the two criteria for crack propagation in the two directions [27]
1=2
D ERc = 1 – 2 c and D .4ERi =c/1=2
Through the 1980s, motor sport teams such as Williams F1, Benetton, Toleman F2,
Spice Engineering, March Engineering, Theodore Racing and Gatoil Cheetah in
Switzerland were working on novel composite components for Le Mans, IndyCar
and F1. Systems and equipment for developing composite processes were quite
experimental as teams became more adept at designing and managing the composite
construction process.
At Williams in 1980, the team constructed a composite curing oven similar to
that in Fig. 22.10, large enough to hold car-sized tooling, in particular to produce
the ground-effect underwing. This oven resembled a kind of indoor garage, with a
large door at one end for wheeling in the tool containing a component laid into it
and vacuum bagged down. Electric elements were spaced out around the walls with
a simple thermostat controller to provide controlled heating.
Special epoxy tooling for composite ground-effect underwings had to be con-
structed. Underwings were standard at the time in F1 and IndyCar series with full
654 J. Jostins and K. Kendall
Fig. 22.10 Modern walk-in oven for curing automobile composites (ACS)
ground effect giving immense downforce. However, these were later disallowed and
removed from regulation as they made cars too fast and raised safety concerns. The
ground-effect underwing extended almost the full length and width of the car and
was an inverted wing-shape section. The force generated from the upturned wing
operated in reverse to the aeroplane wing at speeds over 50–60 mph and forced the
car down against the track, to give increased cornering performance.
Because these inverted downforce wings had to be accurately designed and
built to give a continuous complex curve, a moulded solution was inevitable.
High stiffness was essential to maintain integrity through the speed range, but the
large area of the part coupled with its essential curved surface meant that carbon
composite was the best choice (perhaps the only choice) to give the required strength
and stiffness. The composite structure was constructed of two skins of carbon, epoxy
resin and a thin (5 mm) core of Nomex honeycomb, this slim box section providing
the necessary stiffness with very low mass.
Prior to 1983, motorsports chassis construction tended to comprise flat alu-
minium honeycomb panels, cut to shape and then bonded with Redox 410 adhesive
and riveted along the seams. Over the top of this ran the single skin bodywork in
a combination of glass/resin, carbon/resin or Kevlar/resin panels and underneath,
the underwing. As the decade proceeded, further parts of the chassis and body were
designed in carbon composite materials for the twin benefits of low weight and high
stiffness.
In 1983 at Theodore Racing in Birmingham, UK, their IndyCar chassis was a half
and half construction (Fig. 22.11). The bottom was the traditional flat aluminium
honeycomb, but the top was now moulded carbon composite—the two pieces glued
and riveted as before. The top carbon fibre moulding was still a separate item but
was moulded in a single large body panel, almost 4 m long. It was so lightweight
due to its carbon and Kevlar double weave that it could be lifted off by one person
standing alongside the middle of the car.
22 Interfaces, Cracks and Toughness: City Cars Made from Composites 655
Fig. 22.11 Theodore Racing IndyCar from 1983 showing left lower assembly; right carbon Kevlar
composite top
At this time, some of these weaves were also finding their way into the bottom
flat honeycomb panels (aluminium, honeycomb and two skins), and these ready-
made boards were used to construct the chassis from flat pieces joined along their
edges. This low-cost option required no tooling but showed the weight and strength
benefits of added carbon composite structures.
At Gatoil Cheetah team in Lausanne in 1985/1986, this mix of top composite
moulding, usually glass roving/polyester for reasons of cost, attached to the bottom
carbon reinforced aluminium panel was indeed the construction method. As time
went on, more and more areas of the race car were being made in carbon
composite—from whole chassis to brake ducts. Ultimately, the whole structural
racing car shell in Formula 1 was made in a single moulding with the added
composite and metal parts attached to it, the first team to achieve it being McClaren
in 1981. Meanwhile, road-going sports cars were also beginning to use composites
because the customers expected to see exotic materials as part of the alluring
mix of technology and speed. Ultimately, the composite materials could then be
adapted economically to the city car, which was also changing radically owing to
zero-emission vehicle (ZEV) legislation, which meant the end of combustion and
the introduction of battery-electric drive with hydrogen and fuel cells for range
extension.
John Jostins took all these factors into account when he designed his first eco-
car around 1997 with focus on efficiency—getting the most out of energy used to
propel the vehicle while at the same time reducing emissions in polluted cities. The
two key lines of development were lightweight structures and hybrid drivetrains.
These concepts were not fully understood at that time because the data of Fig. 22.3
were not available to allow reasonable predictive capacity. John Jostins met Kevin
656 J. Jostins and K. Kendall
Fig. 22.13 Photo; Microcab H4 driven by Shang [29] on Stratford raceway; graph result for
the final design shown as circular point on the line for all prototype hydrogen fuel cell vehicles
published (diamond points). The full line is that from Fig. 22.3, while the dashed line shows the
effect of the fuel cell electric drivetrain
under the EU ‘L7e’ class regulations) was calibrated and then modified to include
regenerative electric braking, improved lithium battery storage, more powerful PEM
fuel cell and better matching to remove the lossy DC-DC converter for final testing.
Two UK government-funded projects allowed the collaboration to proceed with
five Microcabs running on campus, together with the first UK hydrogen filling
station and several PhD students supported by EPSRC and by the Science City
project from Advantage West Midlands [1–3, 28]. Several students studied the
Microcab H4 to obtain PhDs [28], and the most significant was that by Shang
[29] who later joined John Jostins at Coventry University. His result is shown in
Fig. 22.13, giving the lowest energy consumption of all the known hydrogen fuel
cell prototypes. In addition, the graph with diamond points showed for the first time
that the Coulomb curve is also followed by hydrogen fuel cell electric cars, but with
a factor 2 improvement on the combustion line from Fig. 22.3. There was large
scatter compared to Fig. 22.3, but this may be explained by the relative immature
fuel cell technologies, where several lossy components still remain to be optimised.
At the same time, John Jostins was designing the next model of Microcab,
the H2EV (Fig. 22.14) which benefitted from the H4 results; eight vehicles were
constructed ([30]; www.swarm-project.eu). The H2EV was designed under M1
regulations, and the fleet was fully road legal in order to play a major role in the EU
SWARM trials, the large collaborative project initiated by Kendall and Jostins. A
new vehicle design named ‘HyLite’ was scheduled for release in 2016, a single-seat
commuter/light freight vehicle with even lower weight than H2EV (see Table 22.1
for Microcab model weights compared to selected production cars). Microcab’s
658 J. Jostins and K. Kendall
Fig. 22.14 Microcab H2EV at Brussels Air Liquide hydrogen station opened on 22 April 2016
Table 22.1 Mass of composite city cars to compare with conventional vehicles
Manufacturer Model Drivetrain type Unladen weight kg Format
Microcab H3 (2001) Fuel cell electric 250 2 seat (tricycle)
Microcab H4 (2004) Fuel cell electric 680 4 seat
Microcab H2EV (2010) Fuel cell electric 750 4 seat
Microcab HyLite (2015) Fuel cell electric 475 1 seat
Toyota Prius ICE hybrid 1380 4/5 seat
Ford Focus ICE 1330 4/5 seat
Mini Cooper ICE 1150 4 seat
Fiat 500 ICE 970 4 seat
Smart For two ICE 820 2 seat
heaviest vehicle in 2016 is the four-seat H2EV, and the nearest production car is
the Smart two seater which is still 70 kg heavier. Other production vehicles can be
seen to be considerably greater, the Prius being 630 kg heavier.
The problem for the cost-effective city car is that a complete carbon fibre composite
body chassis is too expensive at present. Therefore, the intermediate racing car
design of aluminium flat chassis bolted to a moulded composite shell was used for
Microcab. The H2EV comprised a bonded aluminium chassis built by Lotus with
22 Interfaces, Cracks and Toughness: City Cars Made from Composites 659
The Microcab is an economic halfway house in the movement towards full carbon
composite city cars with zero emissions. As in the F1 field, it should be possible
ultimately to produce a full mechanical design of integrated body and chassis using
carbon fibre composites, with other composite parts attached later such as seats,
hydrogen cylinders, etc. Here we discuss how this ultimate vehicle design could be
attained.
22 Interfaces, Cracks and Toughness: City Cars Made from Composites 661
Table 22.2 Well-to-wheel greenhouse gas emissions for various vehicles, depending on electricity
and H2 sources
WTW GHG emissions (g/CO2 per km)
22.8 Conclusions
Three concepts have converged over the last 40 years to raise the level of
understanding of how composites can contribute to the emerging ZEV city car.
First, the problems of emissions have been highlighted by the major car
manufacturers fudging of emission results and by their marketing of heavy vehicles
to the public. Zero-emission legislation first appeared in California in the 1990s,
and, although this mandate failed to be enforced, the ZEV is still the target that city
inhabitants demand. Consequently the drivetrain must change. Combustion must be
eliminated and electric drive brought to domination.
Secondly, the principles of composite materials, their fundamental stiffening and
toughening mechanisms, have been established such that the benefits in terms of
safety and weight have been quantified, with pragmatic advances demonstrated
through racing, especially Formula 1.
Finally, the interaction of drivetrain and vehicle weight has been proved as shown
in Fig. 22.13. The move to CFRP cars can give a drive cycle energy reduction of a
factor 4 (equivalent to emission reduction), whereas the move to hydrogen fuel cell
battery drive gives a factor 2. The total improvement using both modifications is a
factor 8. This is a major leap of understanding which will take some years to push
through to full commercialisation.
Acknowledgements The authors wish to thank the UK for EPSRC and Science City funding and
EU for support in projects SWARM and SAFARI funded by the Fuel Cells and Hydrogen 2 Joint
Undertaking.
References
1. I. Staffell, J.L. Shang, K. Kendall, Atomic models of strong solids interfaces viewed as com-
posite structures. Appl. Compos. Mater. 21(1), 45–55 (2014). doi:10.1007/s10443-013-9330-y
2. J.E. Jostins, S. Glaser, Microcab on Show, and Presented to HRH Prince Charles, ‘Revolve
Eco Vehicle Event and Rally’, Held June 2007 in London, UK, 2007
3. K. Kendall, Hydrogen and fuel cells in city transport. Int. J. Energy Res. 40(1), 30–35 (2016).
doi:10.1002/er.3290
4. D. Sperling, D. Gordon, Two Billion Cars (Oxford University Press, New York, 2010)
5. G.J. Thompson, D.K. Carder, M.C. Besch, A. Thiruvengadam, H.K. Kappanna, In-use
emissions testing of light-duty diesel vehicles in the U.S. Report from West Virginia University,
30 May 2015
6. A. Bandivadekar, J. German, P. Mock, Policy solutions to reduce vehicle exhaust emissions
under real-world driving conditions, ICCT Report, Washington, 1 October 2015
7. U. Tietge, N. Zacharof, P. Mock, V. Franco, J. German, A. Bandivadekar, N. Ligterink,
U. Lambrecht, From laboratory to road, ICCT Report, Washington, 24 September 2015
8. C.R. Knittel, Automobiles on steroids: Product attribute trade-offs and technological progress
in the automobile sector, NBER Paper, 2009
9. C.R. Knittel, Automobiles on steroids: product attribute trade-offs and technological progress
in the automobile sector. Am. Econ. Rev. 101, 3368 (2012)
22 Interfaces, Cracks and Toughness: City Cars Made from Composites 663
10. P. Wright, Ferrari Formula 1-Under the Skin of the Championship-Winning F1-2000 (David
Bull Publishing, Phoenix, AZ, 2004)
11. French Patent 405,881, registered 25 Nov 1909, Benedictus
12. French Patent no. 321,651, registered 31 May 1902, Le Carbon, SA
13. A. Irwin, Risk and the Control of Technology: Public Policies for Road Traffic Safety in Britain
and the United States (Manchester University Press, Manchester, 1985), p. 197
14. U.S. Patent, Method & Apparatus for Making Glass Wool, Published 11 Nov 1933, Issued 11
Oct 1938
15. T. Hobart, Car of the Future in Plastics, Article in the Mercury (The Mercury, Hobart, 1946),
p. 16
16. F. Aird, Fiberglass & Composite Materials: An Enthusiast’s Guide to High Performance Non-
metallic Materials for Automotive Racing and Marine Use (HP Books, New York, 1996),
p. 153
17. A.A. Griffith, The phenomena of rupture and flow in solids. Phil. Trans. R. Soc. A 221,
163–198 (1920)
18. K. Kendall, Control of cracks by interfaces in composites. Proc. R. Soc. Lond. A 341, 409–428
(1975)
19. K. Kendall, Transition between cohesive and interfacial failure in a laminate. Proc. R. Soc.
Lond. A 344, 287–302 (1975)
20. K. Kendall, Interfacial cracking of a composite Part 1: interlaminar shear and tension. J. Mater.
Sci. 11, 638–644 (1976)
21. K. Kendall, Interfacial cracking of a composite Part 2: bending. J. Mater. Sci. 11, 1263–1266
(1976)
22. K. Kendall, Interfacial cracking of a composite Part 3: compression. J. Mater. Sci. 11,
1267–1269 (1976)
23. A. Kelly, Strong Solids (Clarendon, Oxford, 1966)
24. J. Cook, J.E. Gordon, A mechanism for the control of crack propagation in all-brittle systems.
Proc. R. Soc. Lond. A 282, 508–520 (1964)
25. A.R. Zak, M.L. Williams, Trans. ASME J. Appl. Mech. 34, 967–974 (1963)
26. J.R. Rice, G.C. Sih, Trans. ASME J. Appl. Mech. 87, 418–423 (1965)
27. K. Kendall, Molecular Adhesion and Its Applications (Kluwer Academic/Plenum, New York,
2001). ISBN 0-306-46520-5
28. K. Kendall, B.G. Pollet, A. Dhir, I. Staffell, B. Millington, J. Jostins, Hydrogen fuel cell hybrid
vehicles for Birmingham campus. J. Power Sources 196, 325–330 (2011)
29. N.J.L. Shang, Ph.D. Thesis, University of Birmingham, 2014
30. J. Jostins, Performance and energy efficiency testing of a lightweight FCEV Hybrid Vehicle,
in European Electric Vehicle Congress Brussels, Belgium, 3–5 Dec 2014
31. Composites Research Summary Report 2009, www.composites.co.uk
Part VI
Chapter 23
A Virtual Testing Approach for Laminated
Composites Based on Micromechanics
23.1 Introduction
The last quarter-century has witnessed considerable research efforts in the mechan-
ics of composites in order to understand and predict the behavior of these materials.
The ultimate goal is the design of the materials/structures/manufacturing processes.
The scientific literature on composites is immense: numerous books and a dozen of
international reviews have dealt with the understanding, modeling, and identification
of the thermomechanical behavior of composites, with particular emphasis on
long-fiber laminated composites, which are the most important composites for the
aeronautical and space industries and now automotive industries. Even in the case
of laminated composites, the prediction of the evolution of damage up to, and
including, final fracture remains a major challenge which is at the heart of today’s
virtual structural testing revolution. Virtual structural testing consists, whenever
possible, in replacing the numerous experimental tests used today by virtual tests.
In the early 2000s, there were two competing approaches: the micromechanics of
stratified composites, see [1–4], in which one counts cracks. Another very active
area was that of the structural analysis itself, but it is unfortunately limited to
elasticity, especially delamination analysis [5–8] and optimization [9]. Over the
past few years, numerous computational models have been introduced outside of
these two families to simulate and interpret experimental tests but very few could
be seen today as predictive ones; among them, one can cite [10, 11] and the LMT1 -
Cachan’s damage-fracture mesomodel. This last answer to Virtual Structural Testing
is in-between the “fracture criterion” vision, which is rather coarse, and the “micro”
vision, which is very satisfactory in terms of physics, but beyond the capabilities of
today’s computation means.
More, LMT had been working to create a micro-meso bridge [12–16] in order
to develop a complete synergy between micromechanics and macromechanics.
The first application was the Computational Damage Micromodel introduced in
[17–19], which is compatible with all the preexisting knowledge on the micro-
and mesolevels and, contrary to its “micro” rival models, capable of simulating
complex structures. A major difference compared to existing models is that all
cracks are modeled in extenso regarding both their initiation and their propagation
thanks to what one calls “Finite Fracture Mechanics” [3, 20]. Therefore, the
reproduction of experiments does not require the introduction of predefined sets
of cracks based on knowledge derived from tests. However such a model leads
to considerable computation costs and then cannot be used now in structural
engineering calculations [21]. So, in the LMT-approach, this micromodel constitutes
a foundation that can be seen as a Reference Virtual Material of a whole family
of materials and not just of a single material. Using the micro-meso bridge, the
LMT-mesomodel appears as an homogenized version of the micromodel and is
sound in terms of physics [22]. This version has been implemented or is pending
implementation into industrial codes for Virtual Testing. It constitutes a significant
improvement over the standard version developed over some 20 years and available
in many industrial structural analysis programs [23–26]. A last step has been done
recently in [27] with the mesomodeling of the interaction between microcracking in
plies and delamination which is a quite important observed phenomenon [28–37].
The paper is a revisit of our multiscale modeling approach. We will emphasize
on the practical aspects related to the different stones:
• a unified micromodel also involving fatigue, high-velocity, and oxidation, i.e., a
Virtual Reference Material;
• a general bridge between micro- and meso-mechanics;
• the damage mesomodel used for structure computations as a homogenized
model.
Some complements regarding identification and also kinking and crack initiation
are given.
An another main question in Virtual Structural Testing is to compute engineering
composite structures with such a mesomodel. This is not so easy because in the case
of composites, damage and fracture involve many local instabilities; there are many
cracks in the different plies and interfaces in particular splits, which are initiated
and then propagated. For quasi-static loadings, to master the used localization
1
LMT is the Laboratory of Mechanics and Technology of Ecole normale supérieure de Cachan,
France.
23 A Virtual Testing Approach for Laminated Composites Based on Micromechanics 669
limiters is not a simple task. The last part of paper deals with this main question;
we will give our recent results to maximize the robustness of the computation
strategy; in particular, we will describe a new technique to handle splits. Finally,
the current capabilities and limits of this multiscale approach are pointed out as
well as computational challenges that accompany Virtual Structural Testing.
Fig. 23.3 The Reference Virtual Material for a family of composite materials
The working scale in micromechanics is between the dimension of the structure and
the diameter of a fiber (Fig. 23.3). The state of the structure is in fact described as
an assembly of cracked interfaces and cracked layers made with a “fiber-matrix”
material assumed to be homogeneous or quasi-homogeneous. The difficulty comes
from the great number of cracks. One starts with the initial state, where residual
stresses occur and that can be calculated from the process simulation. A more
pragmatic and standard approach simply consists in introducing a uniform negative
variation of temperature and then the corresponding residual stresses are computed
in elasticity.
23 A Virtual Testing Approach for Laminated Composites Based on Micromechanics 671
1
2ED D h11 i2 C ˚.h11 i/ 212
0 0
11 22 213 11 33
.1 df /E10
" ! #
h22 i2 h33 i2 0
23 0
32
C C C 0 22 33
E20 E30 E20 E3
" #
h22 i2 h33 i2
C C
E20 .1 d0 / E30 .1 d0 /
" #
2 2 2
12 23 13
C C C (23.1)
G012 .1 d/ G023 .1 d23 / G013 .1 d/
with ˚.x/ D 2=3rc x3 and rc is a material constant which takes into account
the nonlinear elastic response in compression [39]. All internal damage variables
remain constant through the thickness of each single layer. d0 , d, and d23 serve to
describe the diffuse damage mechanisms for which we suppose a transverse isotropy
behavior related to the fiber direction; the unilateral aspect of microcracks associated
with diffuse damage is taken into account by splitting the energy into a “tension”
energy and a “compression” energy. df is the damage variable which describes fiber
breakage. h i denotes the positive part. The thermodynamic forces associated with
the mechanical dissipation are
672 P. Ladevèze et al.
ˇ * +
2 2
@ h ED i ˇ 1 12 13
Yd D ˇ D C
@d ˇ Dcst 2 .1 d/2 *G012 G013
ˇ
@ h ED i ˇ 1 h22 i2 h33 i2
Yd0 D ˇ D C
@d0 ˇ Dcst 2 .1 d 0 /2 E20 E30
+
2
1 d23 23 (23.2)
C
0 G0
.1 d0 / 1 C .1 d0 / 23 23
ˇ
@ hhED i ˇ 1 ˝˝
Ydf D ˇ D 2 h11 i2 C ˚.h11 i/
@df ˇ Dcst 2 1 df E10
0 0
˛˛
212 11 22 213 11 33
where h i denotes the mean value through the thickness. The material parameters
to identify this law are the initial elasticity Hooke tensor as well as the constant rc .
Diffuse Damage Evolution Law From experimental results, it follows that the
governing forces of damage evolution are
where b2 is a material constant which balances the influence of the transverse energy
and the shear energy. For small damage rates and quasi-static loadings, we get
p p
YN Y0
d Dp p for d < 1 and otherwise d D 1
Yc Y0 (23.4)
d 0 D b3 d
experimental data
identification
0.6 IM7/977-2
identified from [45,-45]2s
0.5
Shear damage d
0.4
0.3
0.2
0.1
0
0 0.5 1 1.5 2 2.5
Shear damage forces Y1/2 (MPa1/2)
Fig. 23.4 Shear damage material function for the “fiber-matrix” material (room temperature,
IM7/977-2)
h22 i h33 i
Q 11 D 11 Q 22 D h22 i C Q 33 D h33 i C
.1 d0 / .1 d0 /
12 23 31
Q 12 D Q 23 D Q 31 D
.1 d/ .1 d23 / .1 d/
"PQ11p D "P11p "PQ22p D h"P22p i.1 d 0 / hP"22p i "PQ33p D h"P33p i .1 d0 / hP"33p i
"PQ12p D "P12 .1 d/ "PQ23p D "P23p .1 d23 / "PQ31p D "P31p .1 d/
(23.5)
where "ij for i; j 2 f1; 2; 3g denotes the usual inelastic strain. It has been shown
that very simple high fidelity material models can be obtained using such effective
quantities. A very simple plasticity model describing inelastic strains is defined by
the following elastic domain:
q 2
2 2 2 2
f .Q ; R/ D Q 12 C Q 23 C Q 31 C a2 Q 22 C Q 33 R R0 (23.6)
pf D 0 with pP 0 and f 0
1 Q ij
"PQijp D pP for i ¤ j and i; j 2 f1; 2; 3g
2 R C R0 (23.8)
a2 Q ii
"PQiip D pP for i 2 f2; 3g
R C R0
One introduces minimum cracked surfaces, the characteristic length being the
thickness of the elementary ply (see Fig. 23.6). All the finite energy release rates
are associated with these surfaces.
140 120
120
100
Threshold (R+R0) (MPa)
100
80
Shear stress
80
60
60
40
40
20
20
0 0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 0 0.01 0.02 0.03 0.04 0.05 0.06
~ Shear strain
Cumulated plastic effective strain p (x10-3)
Fig. 23.5 Hardening curve for the “fiber-matrix” material (room temperature, IM6/914)
Fig. 23.6 Minimum cracked surface for microcracking in plies and delamination cracking
23 A Virtual Testing Approach for Laminated Composites Based on Micromechanics 675
GI GII GIII
C C III D 1 (23.11)
GIc GIIc Gc
GI GII GIII
C C 1 (23.12)
GcI GcII GcIII
Identification The material constants to identify herein are the classical critical
energy release rates and the “transition” thickness hN which is between 2 and 5 times
the thickness of the elementary ply. For delamination cracking, only one interface
676 P. Ladevèze et al.
GIIc GIII
c 3:5GIc
(23.14)
Gc I GcII 3:5GcI
where k is a material constant. Denoting the mean value over the cube of height h
by h i , the criteria for traction and compression are
h Ydfc i
with rNC D sup and where Ydt f c and Ydcf c are two material constants.
t Ydcf
Function `, which depends on the parameters ˛C and mC , is detailed in Appendix.
It describes the kinking damage in compression. Parameters ˛C and mC can
be identified, thanks to experimental and theoretical results given in [42, 43].
Approximated as a brittle mechanism, the second relation of (23.16) could be
written:
A least, one has to identify the two thresholds Ydt f c , Ydcf c and the coupling
coefficient k.
Structure computations have been done in [21, 44, 45] using the LATIN multi-
scale computational approach and parallel computing. Results are very close to
experiment and Fig. 23.7 shows such a comparison for a composite plate Œ02 =902s
23 A Virtual Testing Approach for Laminated Composites Based on Micromechanics 677
Fig. 23.7 Comparison experiment and computation for a holed composite plate Œ02 ; 902 in tension
[45]
in tension [45]. In particular, one reproduces perfectly the splits in the 0ı -ply.
Unfortunately, the computation cost is today much too prohibitive for its use for
structure computation.
Extension to fatigue has been done in [46, 47] for fatigue impacts the fiber-matrix
material. As the diffuse damage, although it exists, appears to be small, we have
preferred to introduce in [46] a new internal damage variable f such that for cyclic
loading, the critical energy release rates:
where f follows classical fatigue laws in term of damage forces over the “minimal”
surfaces introduced in the last paragraph. That is the only change in the model. An
illustration is given in Fig. 23.8.
Several items are less mature. Viscoelasticity has been added considering
effective quantities [45] and in relation with impact problems in [48, 49]. Numerous
experimental results can be founded in [50]. Oxidation has been studied in
[46, 51, 52]. Electrical behavior of laminated composites is studied in [53].
23.2.4 Extension
0.5
[0 /90 ] ; 296.5Mpa
0.45 2 2s
[0 /90 ] ; 262Mpa
2 2s
ηf (relative decrease of tenacity)
[0 /90 ] ; 262Mpa
0.4 2 3s
[0 /90 ] ; 206.8Mpa
2 3s
0.35
0.3
0.25
A
0.2 B
C
0.15
D
0.1
0.05
0
0 1 2 3 4 5 6
Number of cycles, log N
Fig. 23.8 Evolution of the damage indicator f during cycling for fatigue tests at room tempera-
ture [46]
composites, the additional scale of the yarn is present. This work extends the
multiscale computational approach previously developed for laminates made of
unidirectional plies to woven composites.
A rather complete bridge has been built in [12, 13, 15, 16] and Fig. 23.9 describes
the used two-scale computational scheme. At the left, the real structure submitted
to a given loading is defined at the microscale. It is made with the fiber-matrix
material and there are cracks, delamination, and transverse ones. The problem to
solve is clearly a two-scale one, and then the solution consists in two parts: the large
wavelength part for which the characteristic length is the structure dimension and the
small wavelength part which has a characteristic length equal to the ply thickness.
A classical scheme to solve this two-scale problem is to separate the calculation
of the two parts. In a first step, one determines the large wavelength part solving
the so-called homogenized problem where the real structure is replaced by the
homogenized structure. Its solution defines the mesoquantities. In a second step, the
microquantities that make the small wavelength part of the solution are determined
23 A Virtual Testing Approach for Laminated Composites Based on Micromechanics 679
in term of the mesoquantities. This approach has been applied to the basic problem
described in Fig. 23.10 where one considers one layer between two interfaces and
two layers. The upper and lower parts of the studied cell are homogenized. Periodic
conditions and elastic behavior are prescribed.
The equivalence should hold for any value of the residual which can be written
in term of mesoquantities. The fundamental micro-meso link which defines the so-
called homogenized structure holds exactly for the two basic problems. It could be
written:
8 "meso D h "micro i
(23.19)
meso N 3 D h micro i N 3
1
R
where h i D mes. / dS.
is any cross-section orthogonal to N 3 and compatible to the periodicity
associated with the layer or interface containing . Practically for large crack
densities, could be replaced by any large cross-section with respect to the plate
thickness. is the projector on the plane orthogonal to N 3 .
A very large amount of calculations have been performed considering the
practical ranges of the parameters and in particular of the microdamage variables.
We have proved that this cell represents all engineering situations and that it can
be approximated by two basic problems: the ply basic problem and the interface
basic problem. The homogenized operators are quasi-intrinsic. They do not depend
practically on the parameters of the upper and lower parts.
680 P. Ladevèze et al.
The mesomodel of the fiber-matrix material should take into account the matrix
transverse microcracking through the additional mesodamage variables dN 22 , dN 12 , and
dN 23 . It follows that the diagonal terms of the energy written in (23.1) are modified
considering the new moduli:
E20 .1 d0 /.1 dN 22 /
G012 .1 d/.1 dN 12 / (23.20)
E20 .1 d23 /.1 dN 23 /
RC R h
C D ; D and D (23.21)
h h D
Illustrations of this approach are given in [12, 13]. In practice, one can neglect
the influence of C and . Then, we get the intrinsic material functions:
dN 12 D f12 ./
dN 22 D f22 ./ (23.22)
dN 23 D f23 ./
23 A Virtual Testing Approach for Laminated Composites Based on Micromechanics 681
<<micro>> <<meso>>
R+
h 3
D R- 2
1
Residual loading:
Same << ply >> energy
s N 2 : periodic
s N 3 : periodic
Fig. 23.12 The microcracking damage functions for a T700 M21 carbon-epoxy composite at room
temperature [55]
that are plotted in Fig. 23.12 for a T700 M21 carbon-epoxy composite. One has to
use these functions and their derivatives. So here, we propose to simplify their use
considering approximations over the practical range [0,0.8]:
dN 22 D AL .1 exp.=L //
dN 12 D A12 dN 22 (23.23)
dN 23 D A23 dN 22
where A, A12 , A22 , and L are coefficients computed from the original functions.
682 P. Ladevèze et al.
Let us go further with the evolution law at the mesoscale. First, for a given , one
associates the following damage force:
"" # ! # !
Y22 Y12 Y23
Y D C II A12 C III A23 A exp N Q
min.h; h/ D Y exp
GIc Gc Gc L L
(23.24)
with
1 h h22 i2 i
Y22 D
2E20 .1 d0 / 1 dN 22 2
1 h 12 i
Y12 D (23.25)
2G012 .1 d/ 1 dN 12 2
1 h 23 i
Y23 D
2G23 .1 d23 / 1 dN 23 2
0
D lnŒsup YQ (23.26)
L t
with
" #
Y22 Y12 Y23
YQ D N
C II A12 C III A23 A min.h; h/ (23.27)
GIc Gc Gc
and with an added coefficient . The role of this coefficient is to take into account
the “not quite” periodical nature of the transverse cracking network. This is a
shifting coefficient comparable to that used in [12, 15] and very close to that
introduced by Nairn and Hu [38]. Its identification could be performed considering
the comparison between random and periodic behaviors done in [17–19], Another
technique, proposed in [55], is based on a multiple cracking tests on cross-ply
laminates. The identified value for classical laminates is
1:4 (23.28)
Remark The external plies, and also the parts of plies adjacent to completely
delaminated interfaces, behave regarding microcracking, as half a ply [15, 38].
23 A Virtual Testing Approach for Laminated Composites Based on Micromechanics 683
D
N3 N2
1
1'
θ/2 θ/2
N1
(23.29)
where hI is the interface thickness, E and G the Young and shear moduli of the
matrix.
684 P. Ladevèze et al.
d33 C .1 d33 / . C 0 / sin2
d13 D 2
1 C .1 d33 / . C 0 / sin2
2
(23.30)
0
2
d33 C .1 d33 / . C / cos
d23 D 2
1 C .1 d33 / . C 0 / cos2
2
where ./ is a material function computed with an interface with D 90ı . One
can approximate it by:
./ I (23.31)
hI h33 i2
Y33 D
2E .1 d33 /2
2
hI 13
Y13 D (23.32)
2G .1 d13 /2
2
hI 23
Y23 D
2G .1 d23 /2
The effective damage force, which is responsible for the increase of the inter-
face’s damage, is
2 !2 !2 3
Y33 Y13 1 d13 Y23 1 d23
YN I D sup 4 C C 5 (23.33)
t Gc I GcII 1 d33 GcIII 1 d33
where and 0 are the dimensionless microcracking densities of the adjacent plies
to the interface. S is the saturation value which is around 0.8.
1
2ED D h11 i2 C ˚.h11 i/ 212
0 0
11 22 213 11 33
.1 df /E10
" ! #
h22 i2 h33 i2 0
23 0
32
C C C 0 22 33
E20 E30 E20 E3
" #
h22 i2 h33 i2
C C 0
E20 .1 d0 / 1 dN 22 E3 .1 d0 /
" #
2 2 2
12 23 13
C C 0 C 0
G012 .1 d/ 1 dN 12 G23 .1 d23 / 1 dN 23 G13 .1 d/
(23.36)
0
1 d
with ˚. / D 2=3rc 3 and .1 d23 / D .
0
1 1C230 d0
23
Rh
with h i D 1h 2 h dz. The damage evolution law and the modeling of inelastic
2
phenomena are given in Sect. 23.2.2.2 through the formula (23.2)–(23.8).
Microcracking The -damage evolution law is defined by Sect. 23.3.2.1 through
the formula (23.20)–(23.27).
1 2 0 0
Ydf D 2 h h11 i C ˚ .h11 i/ 212 11 22 213 11 33 i
2 1 df E10
Ydt f D Ydf
Rh
with h i D 1h 2 h dz. The damage evolution law is defined thanks to a damage
2
model with delay effect. One has
1
dP f D .1 exp.ac hqi// if df < 1; otherwise df D 1 (23.39)
c
" # 12 " # 12
Ydt f Ydcf
rT D and rC D (23.40)
Ydt f c Ydcf c
Parameters ˛C , mC , Ydt f c , and Ydcf c have already been introduced in Sect. 23.2.2.4.
Let us note that the compression behavior could be approximated by a brittle
mechanism. The localization limiter parameters ac and c will be discussed in the
next paragraph.
1
dP D .1 exp .ac hqi// if d < 1; otherwise d D 1 (23.41)
c
Force
0° 1
90°
Displacement
Ply Inter. Ply Inter.
y y
1) x
2) x
0 ρ dI
0.7 1
0
0 0
90 90
Transverse Delamination Transverse Delamination
cracking (ρ) (dI) cracking (ρ) (dI)
Fig. 23.15 Experimental damage mechanisms, stress/strain curve, and damage prediction in a
cross-ply tensile test Œ0=904 S with the proposed interface model [27]
Another approach, implemented in SAMCEF (the finite element code used here),
is the well-known gradient regularization introduced in [61] and developed for
laminates in [11]. The scalar “regularized force” YQ is defined from the initial one
Y by the equation:
The damage mesomodel could be too much conservative in some cases such as
composite structures involving extensive splitting. Here, the mesomodel has been
found to present a shortcoming: due to numerical constants, the calculated splits,
i.e., the completely damaged zones due to matrix microcracking, are too thick. To
model and simulate splits is a big problem.
690 P. Ladevèze et al.
·104
3
r=0.1
2.5 r=0.3
r=1
2
1.5
load(N)
0.5
−0.5
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
displacement(mm)
Fig. 23.16 Simulation of a Œ02 =902 S tension test: curve load/displacement for different values of
the localization parameter r D c =T
A first approach is the high fidelity micromechanic model described in Sect. 23.2
where every single discontinuity in the plies and in the interfaces is described
and therefore, one has to solve numerically a very large scale problem involving
thousands of cracks. Even with a high performance computational tools, such a
micromechanic model leads to prohibitive computational efforts and, thus, is far
from meeting the virtual structural testing requirements. However, fundamentally, a
split appears to be as a particular microcrack relatively isolated and mainly due
to shear [44]. From purely microscopic and mesoscopic approaches, intermedi-
ate approaches have recently been proposed in the literature. Classical cohesive
interfaces are a priori introduced for transverse microcracking and combined with
delamination interfaces. Two ways can be distinguished. The first one uses a priori
experimental information about the cracking pattern (e.g., the position of the splits)
and then is not a predictive approach [62]. The second one considers a set of
equally space potential cohesive interfaces in the different plies. To avoid prohibitive
computations, the distance between two cohesive interfaces cannot be too small and
then is not necessarily compatible with the micromechanics [37, 63]. Anyway these
approaches do not give the “continuous” part of the microcracking density.
Our understanding is quite different. The difficulty to get “thin” splits with
the mesomodel is only due to numerical problems associated with localization
phenomena involving very small time and space discretization. The mesomodel,
where microcracking is described by microcracking density, is correct and a split
23 A Virtual Testing Approach for Laminated Composites Based on Micromechanics 691
·104
3
lc =0.1mm
2.5 lc =0.5mm
lc =1mm
2
load (N)
1.5
0.5
−0.5
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
displacement(mm)
Fig. 23.17 Simulation of a Œ02 =902 S tension test: curve load/displacement for different values of
the localization parameter lc
-104
2
without split
with splits
1.5
1
load (N)
0.5
–0.5
0 0.2 0.4 0.6 0.8
displacement (mm)
Fig. 23.18 Simulation of a Œ02 =902 S tension test: computed load/displacement curve (with and
without cohesive elements)
Fig. 23.19 Simulation of a Œ02 =902 S tension test: computed microcracking density maps closed
to final fracture (with and without splits)
23 A Virtual Testing Approach for Laminated Composites Based on Micromechanics 693
23.5.3 Applications
Previous versions of the damage mesomodel are today used in a number of industrial
codes. Numerous academic validations have been done during the last 25 years.
These include classical delaminations tests in [25, 65–67], WWFE-III in [68], open
hole tensile tests on quasi-isotropic laminates [60], low velocity impacts [69], and
impact problems [48, 49, 70, 71].
23.5.4 Limits
23.6 Conclusion
The multiscale material model described in this paper for laminate composites is
quite general and should be extended to other composite materials. The version
of the mesomodel presented here is pending implementation into industrial codes
for Virtual Testing. It constitutes a significant improvement over the standard
version developed over some 20 years and available in many industrial structural
analysis programs, especially regarding the interaction between microcracks and
delamination and split modeling and computation.
Taking into account all Lack-of-Knowledge (LoK) sources, essentially here the
defects, constitutes one of the scientific challenges involved by Virtual Testing [74].
Not only does one have to simulate reality, but one must also position the result
of the model quantitatively with respect to that reality which, itself, presents some
variability, both at the micro-, meso-, and macroscales. Moreover, computational
694 P. Ladevèze et al.
x
α α’ 1
costs being prohibitive for designers, another challenge—probably the main one—
has to be tackled using the damage mesomodel presented here: the building of
virtual charts, i.e., reduced models including the description of uncertainties [75].
Let us introduce z D .1 d/x with d D `.x/. Function ` is plotted in Fig. 23.20 and
is defined as follows:
!m
0
1 x
• for x 2 Œ0; ˛ , z D .1 d/ x with d D `.x/ D .m 1/
1Cm ˛
1 1x
• for x 2 Œ˛ 0 ; 1, z is linear and d D `.x/ D 1 z.˛ 0 /
x 1 ˛0
So, function ` over Œ0; 1 is maximum for x D ˛. The first derivative of ` is
continue over the complete interval Œ0; 1. z can be interpreted as a stress and x as
a strain. Let us note that the slope of the secant is: 1 d. Another main property
is related to the maximum of ` over Œ0; 1. This point can also be interpreted as an
instability point or an initiation point. Let us consider that the strain x is the square
root of the damage force; one has at the point:
.1 d/ xP d;x xP x D 0 (23.45)
1
and then, using d D 1Cm . ˛ / ,
x m
one has
1
.1 d/ D d;x x D md and dI D (23.46)
1Cm
The parameters are ˛ and m with .1 C m/˛ m < 1, then one has:
!
0
˛
˛ ˛ 1C (23.47)
mC1
23 A Virtual Testing Approach for Laminated Composites Based on Micromechanics 695
and
Z !
Gc 1 m m
2 2
D x ıd ˛ C˛ (23.48)
Yc 0 mC1 .m C 1/.m C 2/
for defining Yc .
References
20. Z. Hashin. Finite thermoelastic fracture criterion with application to laminate cracking
analysis. J. Mech. Phys. Solids 44(7), 1129–1145 (1996)
21. P. Ladeveze, G. Lubineau, D. Violeau, A computational damage micromodel of laminated
composites. Int. J. Fract. 137(1–4), 139–150 (2006)
22. G. Lubineau, P. Ladeveze, Construction of a micromechanics-based intralaminar mesomodel
and illustrations in ABAQUS/Standard. Comput. Mater. Sci. 43, 137–145 (2008)
23. P. Ladeveze, E. L. Dantec. Damage modelling of the elementary ply for laminated composites.
Compos. Sci. Technol. 43, 257–267 (1992)
24. O. Allix, P. Ladeveze, Interlaminar interface modelling for the prediction of delamination.
Compos. Struct. 22, 235–242 (1992)
25. P. Ladeveze, A damage computational method for composite structures. Comput. Struct.
44(1–2), 79–87 (1992)
26. P. Ladeveze, A damage computational approach for composites: basic aspects and microme-
chanical relations. Comput. Mech. 17(1–2), 142–150 (1995)
27. P. Ladeveze, F. Daghia, E. Abisset, C. Le Mauff, On the intra/interlaminar coupling of
laminated composites: modeling and identification, in Advanced Modeling and Simulation
in Engineering Sciences (Springer, Berlin, 2014)
28. C.E. Harris, D.H. Morris, Role of delamination and damage development on the strength of
thick notched laminates, in Delamination and Debonding of Materials. ASTM STP, vol. 876
(Philadelphia, 1985), pp. 424–447
29. J.A. Nairn, S. Hu, The initiation and growth of delaminations induced by matrix microcracks
in laminated composites. Int. J. Fract. 57(1), 1–24 (1992)
30. S.R. Finn, Y.F. He, G.S. Springer, Delaminations in composite plates under transverse impact
loads - experimental results. Compos. Struct. 23(3), 191–204 (1993)
31. H. Eggers, H.C. Goetting, H. Bäuml, Synergism between layer cracking and delaminations
in multidirectional laminates of carbon-fibre-reinforced epoxy. Compos. Sci. Technol. 50(3),
343–354 (1994)
32. W.H. Chen, S.H. Yang, Multilayer hybrid-stress finite element analysis of composite laminates
with delamination cracks originating from transverse cracking. Eng. Fract. Mech. 54(5),
713–729 (1996)
33. P. Johnson, F.K. Chang, Characterisation of matrix crack-induced laminate failure - Part II:
analysis and verifications. J. Compos. Mater. 35(22), 2037–2074 (2001)
34. S. Li, S.R. Reid, Z. Zou, Modelling damage of multiple delaminations and transverse matrix
cracking in laminated composites due to low velocity lateral impact. Compos. Sci. Technol.
66(6), 827–836 (2006)
35. H. Zhang, L. Minnetyan, Variational analysis of transverse cracking and local delamination in
Œm =90n s laminates. Int. J. Solids Struct. 43(22–23), 7061–7081 (2006)
36. S.R. Hallett, W.G. Jiang, B. Khan, M.R. Wisnom, Modelling the interaction between matrix
cracks and delamination damage in scaled quasi-isotropic specimens. Compos. Sci. Technol.
68(1), 80–89 (2008)
37. F. Van der Meer, C. Olivier, L. Sluys, Computational analysis of progressive failure in a notched
laminate including shear nonlinearity and fiber failure. Compos. Sci. Technol. 70, 692–700
(2010)
38. J.A. Nairn, S. Hu, Matrix microcracking, in Damage Mechanics of Composite Materials
(Elsevier, Amsterdam, 1994), pp. 187–243
39. O. Allix, P. Ladeveze, E. Vittecoq, Modelling and identification of the mechanical behaviour
of composite laminates in compression. Compos. Sci. Technol. 51, 35–42 (1994)
40. O. Allix, N. Bahlouli, Ch. Cluzel, L. Perret, Interlaminar Modeling and identification of
temperature-dependent mechanical behavior of the elementary ply in carbon/epoxy laminates.
Compos. Sci. Technol. 56, 883–888 (1996)
41. O. Allix, D. Leveque, L. Perret, Interlaminar interface model identification and forecast of
delamination in composite laminates. Compos. Sci. Technol. 56, 671–678 (1998)
23 A Virtual Testing Approach for Laminated Composites Based on Micromechanics 697
66. L. Daudeville, P. Ladeveze, A damage mechanics tool for laminate delamination. Compos.
Struct. 25, 547–555 (1993)
67. N.D. Flesher, C.T. Herakovich, Predicting delamination in composite structures. Compos. Sci.
Technol. 66(6), 745–754 (2006)
68. F. Daghia, P. Ladeveze, Identification and validation of an enhanced mesomodel for laminated
composites within the WWFE-III. J. Compos. Mater. 47(20–21), 2675–2694 (2013)
69. S. Guinard, O. Allix, D. Guedra-Degeorges, A. Vinet, A 3D damage analysis of low-velocity
impacts on laminated composites. Compos. Sci. Technol. 62(4), 585–589 (2002)
70. L. Greve, A.K. Pickett, Delamination testing and modelling for composite crash simulation.
Compos. Sci. Technol. 66(6), 816–826 (2006)
71. J.M. Guimard, O. Allix, N. Pechnik P. Thévenet, Characterization and modeling of rate effects
in the dynamic propagation of Mode-II delamination in composite laminates. Int. J. Fract. 160,
55–71 (2009)
72. Ch. Hochard, S. Miot, N. Lahellec, F. Mazerolles, M. Herman, J. Charles, Behaviour up to
rupture of woven ply laminate structures under static loading conditions. Compos. Part A 40,
1017–1023 (2009)
73. P. Camanho, G. Ercin, G. Catalanotti, S. Mahdi, P. Linde, A finite fracture mechanics model
for the prediction of the open-hole strength of composite laminates. Compos. Part A 43,
1219–1225 (2012)
74. N. Carrere, Y. Rollet, V. Retel, L. Boubakar, J.F. Maire, Composites structural modelling with
uncertain data. Compos. Sci. Technol. 69, 60–66 (2009)
75. F. Chinesta, P. Ladeveze (eds.) Separated Representations and PGD-Based Model Reduction-
CISM, vol. 554 (Springer, Berlin, 2015)
Chapter 24
Virtual Testing of Composite Structures:
Progress and Challenges in Predicting Damage,
Residual Strength and Crashworthiness
Nomenclature
24.1 Introduction
Each year the two major airframe manufacturers, Airbus and Boeing, release market
forecasts in air traffic and new aircraft requirements for the next 20 years. At the time
of writing, Airbus predicts that air traffic will continue to double every 15 years,
necessitating the need for 32,600 new passenger and freighter aircraft with a
market value of $4.9 trillion [1]. Boeing’s forecasts are even higher, predicting the
requirement of 38,050 aircraft at a cost of over $5.5 trillion [2]. These aircraft will
need to meet increasingly demanding environmental targets. The Advisory Council
for Aeronautical Research in Europe (ACARE), for example, has set an aspirational
target of a 75 % reduction in CO2 emissions and a 90 % reduction in NOx emissions,
by 2050, with respect to the year 2000 baseline [3]. Weight reduction will continue
to play a key role in delivering these objectives. The aerospace industry has
already invested heavily in developing the nascent knowledge base, manufacturing
infrastructure and certification protocols for advanced carbon fibre composite
aerostructures and is therefore likely to continue to build on this foundation for
the further utilisation and exploitation of these materials in the foreseeable future.
Moreover, airframe manufacturers are fully aware that to maintain their
competitiveness, they must not only be in a cycle of incessant innovation but
24 Virtual Testing of Composite Structures: Progress and Challenges. . . 701
must also deliver new products with reduced development and certification costs.
The certification of composite aerostructures entails extensive experimental testing
which increases costs and development times, consequently restricting the design
space that can be explored within viable timescales. This conservative approach
stems from limited in-service experience of composite materials. Indeed, the FAA
and EASA each applied a number of additional conditions towards the certification
of the Boeing 787 (and the Airbus A350) where it was deemed that the current
airworthiness standards were not adequate to ensure that composite primary
aerostructures provided the same level of safety as their metallic counterpart. One
such condition concerned the crashworthiness of the composite fuselage. This lack
of confidence is articulated in a report which was released by the US Government
Accountability Office (GAO), a month before the Boeing 787 entered service with
All Nippon Airways [4]. This report examined the actions of the FAA (and by
association, EASA’s) in assessing the safety of composite airplanes. The GAO
identified a number of concerns arising from limited information on the behaviour,
detection and characterisation of damage in composite structures and the challenge
in ensuring the structural integrity of a bonded repair. While the report stated that it
was ‘too early to fully assess the adequacy of FAA and industry efforts to address
safety-related concerns’, it acknowledged the ongoing efforts of the FAA to address
the concerns relating to maintenance and repair.
One approach for reducing development time and cost is to increase the use
of modelling and simulation at all levels of the product development cycle. The
aerospace and automotive industries have been at the forefront of integrating
modelling and simulation tools into their product development, but the shift
towards the increased use of composite materials has highlighted the inadequacy
of existing simulation tools to reliably predict the structural response of composites
under damage-inducing loads. The exhaustive experimental programme, currently
required as part of the development and certification of a new composite airframe,
follows a building block approach which was first represented by Rouchon [5]
as a test pyramid. The purpose of adopting this approach is to mitigate risk by
progressing through a sequence of testing, starting from simple (generic) coupons,
for basic material characterisation, and moving to more complex (non-generic)
structural details as shown in Fig. 24.1.
In reality, a number of tests, across the height of the pyramid, are often conducted
simultaneously. This is partly the result of the links between damage across scales
being poorly understood and consequently undermines the risk mitigation strategy
to some extent. Hence, while modelling and simulation are currently used as part
of the design and development cycle, certification is still based on physical testing.
Indeed, the challenge of certification by simulation, which is the ultimate objective
of virtual testing, primarily boils down to the ability to reliably predict damage.
This challenge is twofold, even if the two are intrinsically linked; (1) there is a
need to improve our understanding of the underlying physics which governs the
complex evolution of composite damage to enable the development of robust and
representative mathematical models, and (2) it is necessary to provide a reliable
framework which will facilitate the exploitation of these tools by product developers
and lead to their acceptance, by the certification authorities, as a valid substitute for
some of the physical testing. This aspirational target is represented by the right side
702 B.G. Falzon and W. Tan
of the Rouchon pyramid, shown in Fig. 24.1, where a ‘waist’ is formed to represent
the reduction of physical testing and replaced with simulation. The unchanged size
of the base of the pyramid suggests that the extent of the experimental programme
to establish basic material data is unlikely to change, but the red dotted line proposes
that a high-fidelity computational model may be able to also reduce the degree of
testing required at this level.
This chapter will focus on recent developments in addressing the challenge
of developing a mesoscale damage model for predicting impact damage, residual
strength and energy absorption capacity of carbon fibre composite structures. It
is worth noting that Szabo and Actis [6] introduced the concept of ‘simulation
governance’ as a vital requirement towards certification by simulation. This is an
open-ended framework which provides a systematic approach towards verification
procedures, updating mathematical models as new experimental data becomes
available, and the collection, management and interpretation of empirical data
required for validation. Implicit in simulation governance is the aim to reduce
epistemic uncertainties arising from, for example, limitations in available data or
fidelity of the mathematical models. It should also lead to a progressive reduction in
aleatoric uncertainties, as more relevant empirical data is made available.
Ply-level physically based failure models are proposed for each failure mode in
laminated fibre-reinforced composites. The failure modes may be classified as
intralaminar (matrix cracks and fibre pull-out/breakage) and interlaminar (delam-
ination) damage illustrated in Fig. 24.2. Intralaminar damage may be classified
as (1) fibre dominated, arising from loading which is predominately in the fibre
24 Virtual Testing of Composite Structures: Progress and Challenges. . . 703
Fig. 24.2 (a) Interlaminar and intralaminar damage in a composite laminate, (b) damage modes
within a composite ply
direction ( T11 / C11 ), or (2) matrix dominated, arising from transverse and shear
loading ( T22 / C22 , 12 , 23 , 31 ), where the superscripts ‘T’ and ‘C’ denote ‘tension’
and ‘compression’, respectively.
where Gc is the mixed-mode fracture toughness and B is the local mixed-mode ratio
defined as B D Gshear = .GI C Gshear /. is the mixed-mode interaction coefficient
determined from experimental measurements. This mixed-mode softening law [9]
may be illustrated by the interaction diagram shown in Fig. 24.3.
704 B.G. Falzon and W. Tan
Fig. 24.4 (a) Schematic of fibre tension or compression (Poisson contractions not shown), (b)
T(C)
bilinear law (shaded area is volumetric strain energy density, g11 )
Damage Initiation
A bilinear law (Fig. 24.4) was used, for simplicity, to model the material response
in the fibre direction where damage is characterised by fibre pull-out, fibre–matrix
debonding and fibre breakage. The maximum stress criterion is widely used to
predict the initiation of fibre tensile failure. For fibre under compression, failure
mostly occurs through elastic instability (so-called micro-buckling or kinking) of
the fibres embedded in the matrix. While sophisticated damage models have been
proposed for fibre compressive failure (e.g. [11]), strain-based initiation functions
are adopted to model the material response in the longitudinal direction as the
longitudinal tensile and compressive strengths can be measured directly. Therefore,
the failure index defined for both tensile FIT11 and compressive FIC11 loading is
given by
2
"11
FIT11 D 1; (24.5)
"OT
11
2
"11
FIC11 D 1; (24.6)
"OC
11
where the failure initiation strains ("OT 11 and "11 for tension and compression,
OC
respectively, are determined by the strengths in the respective directions, i.e. "OT
11 D
0
X T =E11 , where XT is the longitudinal tensile strength, etc.
The area under the traction-separation curve is the volumetric strain energy
density, and the damage parameter associated with a reduction in secant stiffness,
under loading in the fibre direction, is given by
706 B.G. Falzon and W. Tan
!
FT.C/
"11 "11
OT.C/
T.C/ T.C/ OT.C/
d11 ."11 / D FT.C/ OT.C/
1 T.C/
; "11 > "11 : (24.7)
"11 "11 "11
T.C/ T.C/
FT.C/ 2g11 2 11
"11 D T.C/ D T.C/ : (24.8)
X X lfib
V
lfib D ; (24.9)
A
where A is calculated using an approach proposed in [13] and described in
Sect. 24.2.2.
Prior to damage initiation, the respective initial tensile and compressive moduli
are maintained upon load reversal (i.e. traversing paths 1 and 4 in Fig. 24.5).
Beyond damage initiation, the damage caused by tensile and compressive loading
interacts when unloading and load reversal are introduced. It is assumed that the
growth of damage in the tensile mode does not significantly affect the response in
compression when the loading is reversed. Even though fibre breakage has occurred,
it is assumed that the entrapped fibre and matrix debris under compression can
still sustain loading. While some modulus degradation would be expected, no such
degradation is accounted for as this is deemed to be of minimal influence on overall
behaviour.
Consequently, unloading along paths 3 and reversing the load such that the
material is now in compression will result in an initial elastic stiffness response
represented by path 4. However, the reverse is not true. In compressive loading-
induced damage, some fibre breakage occurs under kink-band formation. Hence,
the stiffness is reduced when the material is subsequently loaded in tension.
Therefore, the growth in the compressive damage parameter will also cause the
tensile damage parameter to grow and the stiffness to decrease (paths 6–7). To
achieve the interaction shown in Fig. 24.5, the modulus is reduced according to
the longitudinal damage parameter defined as
24 Virtual Testing of Composite Structures: Progress and Challenges. . . 707
Fig. 24.5 Stress–strain response during fibre direction loading–unloading: paths 1–2, tensile
loading; 3, unloading; 4–5, loading in compression; 6–8, compressive unloading and tensile
reloading until failure
8 C
< d11 "11 < 0
dfib D d11
T
"11 0 and d11T C :
> d11 (24.10)
: C
d11 "11 0 and d11T
< d11
C
Applying the damage parameter to the longitudinal component, the stress in the
longitudinal direction is determined by
T.C/ T.C/
11 D .1 dfib / Q 11 : (24.11)
A non-linear damage model that accounts for inelastic shear deformation, stiffness
degradation and load reversal forms part of the composite damage model. Prior to
damage initiation, some plasticity is evident such that shear loading and unloading
occur along gradients defined by the initial shear modulus Gini ij and plasticity-
reduced shear modulus G*ij as shown in Fig. 24.6b. The plastic strain represents all
irreversible deformations including those caused by microcracks. The shear strain
ij is further decomposed into an elastic part, ij, el , and a plastic part ij, in ,
a 3 3
ε13, d13 → τ13, d13
3
ε23, d23 → τ23, d23
ε12, d12 → τ12, d12
2 2 2
1 1 1
Gij
ini
Gij* Gij* (1- dmat)Gij* Dgij
Gij*,t
ini
gij Gij
gij,in gij,ef gijo gijf
Gij*,t+Dt
t
gij gijt+Dt
gij,in
t
gij,el
t gij
-t0
gij,in
t+Dt
gij,el
t+Dt
Fig. 24.6 (a) In-plane and out-of-plane shear, (b) non-linear shear model and (c) calculating shear
stress from shear strain using the elastic predictor method
Since the in-plane and out-of-plane behaviour (Fig. 24.6a) do not differ sig-
nificantly under shear loading and due to the lack of experimental results for
out-of-plane shear behaviour, similar constitutive curves have been used to model
out-of-plane shear non-linearities. The in-plane and out-of-plane stress–strain con-
stitutive laws were fitted using an exponential model:
(
ijY exp ˛ij exp ˇij ; ij 0
ij D (24.14)
ijY exp ˛ij C exp ˇij ; ij < 0
where Yij is the initial yield strength, determined by the 2 % offset strain point,
˛ is a strain hardening coefficient and ˇ controls the initial shear modulus and
elastic–plastic transition region. To characterise the degradation of the secant shear
modulus, a degraded shear modulus, G*, t
ij , was introduced, and a strain-degraded
modulus curve is shown in Fig. 24.17d. The degraded modulus was coupled with
the plastic deformation in the constitutive relation, making it convenient to obtain
the fitting parameters from experiment results:
ˇ ˇ ˇ ˇ
ˇ ˇ ˇ ˇ
G;
ij
tC t
D p1 exp p2 ˇijtC t ˇ C p3 exp p4 ˇijtC t ˇ : (24.15)
determining the final load state. The elastic stress predictor is given by ij;el tC t
D
; tC t
ij;el C Gij
t
ij . The nominal stress is updated based on an explicit integration
scheme. In the case of unloading and reloading, an initial stress state ( tij , tij ) is
reached after partial unloading along the in situ shear modulus (G*ij ). The stress
state after subsequent reloading to ijtC t depends on whether plastic yielding
has occurred. At a stress state below the yield stress state, the material behaves
elastically. Once yielding occurs, stress is updated according to the yield surface.
The implementation of this method is illustrated by the flow charts in Fig. 24.12b.
For a material under a reversed loading condition, the subsequent yield stress
is determined by the isotopic hardening approach, which assumes the reversed
compressive/shear yield stress is equal to the tensile/original yield stress (jABj D
jBCj). Isotropic hardening only applies while the loading remains below the
threshold of matrix damage initiation. Once macroscale matrix cracking initiates,
unloading occurs along the reduced secant shear modulus to the permanent plastic
strain oij,in at damage initiation. Typical cyclic loading response with increasing
amplitude and corresponding damage parameter is shown in Fig. 24.7.
fracture plane will be normal to the load, but as suggested by Puck and Schürmann
[14], under compressive and/or shear loading, the failure plane orientation is
determined by the ability of the matrix to withstand shear loading. The damage
initiation and the subsequent progression of damage are calculated on the fracture
plane as shown in Fig. 24.8. The stress tensor is rotated onto the fracture plane using
the standard transformation matrix T():
2 3
1 0 0
ŒT ./ D 4 0 cos ./ sin ./ 5 (24.16)
0 sin ./ cos ./
and if the normal stress in the fracture plane is tensile, NN > 0, then
2 2 2 2
NN LN NT NN LN NN
F ./ D is
C is
C is
C is is
C is
S23 S12 S23 S23 S12 S23
(24.19)
2 2
.Sis12 / .Y T;is / 2LN Sis
where parameters and are given by D Sis T;is and D Sis
23
, YT,is
23 Y 12
is the in situ transverse shear strength and Sis12 and Sis23 are the in situ shear strengths.
The transverse friction coefficients, defined in [16], are based on Mohr–Coulomb
theory where NT D tan 12 and S23 is
D 2 tanY C;is
, YC,is is the in situ transverse
. f/ . f /
Sis
compressive strength and LN D 12
Sis
NT . The fracture plane orientation, f , is
23
ı
typically found to be approximately 53 for unidirectional composites [16] under
uniaxial transverse compressive loading. For a general 3D load state, the orientation
is not known a priori and is determined by the angle which maximises the failure
criteria functions. Brent’s algorithm was used for this purpose which combines a
golden section search with parabolic interpolation [17].
Two parameters are introduced to describe the damage propagation under compres-
sion and shear loading (Fig. 24.9): ① shear damage in the strain hardening part, dIij ,
and ② shear damage in the strain softening part, dIIij ,
(
dijI ij < ijo
dmat D ; (24.20)
dijII ij < ijo
f !
ij ij;in
o
ij ijo
dijII D dijI C 1 dijI (24.22)
f
ij ijo ij ij;in
o
where ij,in is the plastic strain at the onset of failure, l* is the characteristic crack
o
length described in the next section and oij is the shear strength. gij is the volume
energy release rate associated with the elastic fracture energy part (shaded red part in
712 B.G. Falzon and W. Tan
Fig. 24.6b) and ij is the shear fracture toughness. A simple quadratic interpolation
function was used for the fracture energy in the mixed-mode case given in the next
section.
with corresponding strain, "r , acting on the fracture plane [18, 19], defined as the
vector sum of the elastic and inelastic components, "r D "r;el C "r;in , where
q
el 2 el 2
"r;el D h"NN i2 C NT C NL ; (24.24)
q
2 in 2
"r;in D NT
in
C NL (24.25)
24 Virtual Testing of Composite Structures: Progress and Challenges. . . 713
and hxi D max .0; x/ is the McCauley operator. Figure 24.10 shows the overall
damage propagation for mixed-mode matrix damage where the damage in stage I is
the maximum damage parameter from dijI .ij D 22; 12; 23/ given by
!
"fr "0r;in "0r "r
dmat D max dijI C 1 max dijI (24.26)
f
"r "0r "r "0r;in
The shear stresses on the fracture plane are degraded by the matrix damage
parameter dmat :
LN D .1 dmat / Q LN ;
NT D .1 dmat / Q NT ; (24.27)
NN D e
NN dmat hQ NN i :
These stresses are then transformed back to the material coordinate system to form
the complete stress tensor of the damaged element by 123
d
D ŒT ./T LNT ŒT ./
T. The overall shear loading and unloading, prior to damage initiation (Fig. 24.10),
occur along gradients defined by the initial shear modulus Gr , given by
r0
Gr D : (24.28)
"0r;el
2 rC
"fr D "r;in C (24.29)
r0 lmat
714 B.G. Falzon and W. Tan
The critical mixed-mode strain energy release rate, Cr , governing the energy
dissipation in the elastic fracture part (red shade area in Fig. 24.10), is then given by
˝ 0 ˛ !2 0 2 0 2
NN LN NT
C
D C
22 C C
12 C C
23 ; (24.30)
r
r0 r0 r0
where ijC .ij D 22; 12; 23/ are the corresponding critical strain energy release rates
for each stress component. The combined volumetric strain energy release rates,
g0r,el , at damage initiation, associated with each stress component on the fracture
plane, are combined using a quadratic relationship, which can be represented by the
red-shaded region in Fig. 24.10:
˝ 0
˛ !2 0 2 0 2
NN LN NT
g0r;el D g0NN C g0LN C g0NT ; (24.31)
r0 r0 r0
The volumetric strain energy associated with each stress component, g0i , where i
denotes NN, LN and NT, is given by
Z "0i
g0i D i d"i : (24.32)
0
The shaded grey area, in Fig. 24.10, is the inelastic volumetric strain energy
release rate, gr,in , given by
gr;el D r =lmat ;
C
(24.34)
where gr,el is the combined elastic volumetric strain energy release rate
The characteristic length given by ABAQUS is the cubic root of the volume of
a finite element, which is only accurate when elements have aspect ratios close
to unity and crack planes are assumed to evolve perpendicular to the midplane of
the element. Therefore, a more accurate measure of characteristic length should be
calculated as a function of the crack plane orientation [13, 20].
_
The fracture surface is defined by a unit normal vector ( n) in an arbitrary
hexahedral element (Fig. 24.11). This normal vector contains information on the
material coordinate system as well as the fracture plane rotation. The characteristic
length algorithm determines the points, pi , where the fracture plane intersects the
24 Virtual Testing of Composite Structures: Progress and Challenges. . . 715
1 ! p
p !
Ai D i iC1 (24.35)
2
X
AD Ai (24.36)
V 2
lc D (24.37)
A X"
8
ˆ
ˆ
T.C/
lfib ; . ; X; "/ D 11 ; X T.C/ ; "11
0T.C/
ˆ
<
T.C/ 0T.C/
lc D lmat ; . ; X; "/ D 22 ; Y T.C/ ; "22 (24.38)
ˆ
ˆ
:̂ . ; X; "/ D ij ; Sij ; ij0 ij D 12; 23; 32
The final element size was determined by taking into account both intralaminar and
interlaminar characteristic lengths, i.e. Le min flc ; le g.
716 B.G. Falzon and W. Tan
V
det F D ; (24.39)
V0
The limits on det F are user defined and the quoted values were found to yield
good results.
Fig. 24.12 Flow charts of (a) overall subroutine, (b) non-linear shear behaviour subroutine,
(c) fibre-dominated damage subroutine and (d) matrix-dominated damage subroutine
718 B.G. Falzon and W. Tan
thermoplastic composite was selected for this study. There are several existing
standard test methods to characterise the interlaminar fracture toughness. For mode
I tests, the double-cantilever-beam (DCB) test is the most widely used approach and
is the method adopted by the internationally recognised standard, ASTM D5528-01
[21], Fig. 24.13a. The ASTM standard D7905/D7905M is a recently announced
test method for determination of the mode II interlaminar fracture toughness of
unidirectional fibre-reinforced polymer matrix composites using the ENF test [4].
Four-point end-notched flexure (4ENF) [22], Fig. 24.13b, which can deliver a
stable crack propagation to yield a resistance curve, may be used to determine
the propagation fracture toughness. For mixed-mode fracture toughness, the mixed-
mode bending (MMB) configuration, Fig. 24.13c, is often used for the wide range of
mode mixity it can create (ASTM D6671-01) [23]. Experiments with three different
mode mixities (25 %, 50 %, 80 %) were carried out using the MMB apparatus with
corresponding loading lever length, c, of 75.49 mm, 41.40 mm and 27.10 mm.
The thickness and width of all the tested specimens were 3.36 mm and 20 mm,
respectively. The length, outer/inner span distance, initial delamination length and
constant loading rate are given in Table 24.1. All the specimens were tested under
displacement control, while the load and displacement were recorded by the load
cell.
A resistance curve (R-curve) depicting GIC calculated by modified compliance
calibration (MCC), as a function of delamination length, a, was generated to
characterise the initiation and propagation fracture toughness, shown in Fig. 24.14a.
The average initiation fracture toughness was measured at GIC D 0:924 kJ=m2 with
24 Virtual Testing of Composite Structures: Progress and Challenges. . . 719
Fig. 24.13 Interlaminar fracture toughness characterisation tests and simulations of (a) DCB, (b)
4ENF and (c) MMB
Table 24.1 Dimensions of interlaminar fracture toughness test specimens (dimension, mm)
Outer span
Tests Length (l) (2L) Inner span (d) Precrack length (a0 ) Loading rate (mm/min)
DCB 150 – – 50 2
4ENF 180 140 70 70 1
MMB 160 100 50 25 0.5
GIIc, (kJ/m2)
GIc (kJ/m2)
2
1.5
1.5
1
1
0.5 0.5
0 0
45 55 65 75 85 95 105 65 75 85 95
Crack length, a (mm) Crack length (mm)
c 2.5 4ENF d
1000
Exp-Solid line
80% Num-Dash line 4ENF
2 50% 800
25%
GT (kJ/m2)
Laod (N)
DCB 600
1.5 MMB-80%
400
MMB-50%
1
200 MMB-25%
Initiation Propagation Predicted DCB
0.5 0
0% 25% 50% 75% 100% 0 5 10 15
GII/GT (%) Displacement (mm)
Fig. 24.14 (a) Mode I, (b) mode II and (c) mixed-mode fracture toughness; (d) experimental and
numerical results
a 5 b
CT-1 CT-2 CT-3 500
CT-1 CT-2
Δa=8.87
2 200
Δa=13.28
1 100 Γ11 =243.9kJ/
2
Δa=15.31
0 0
0 2 4 6 26 28 30 32 34 36
Displacement (mm) Crack length, a (mm)
c 7
Area method d 160
CC-1
Δa=15.23
4
Δa=20.25 80
3 Γ11 =108.3kJ/
60 2
2 40
CC-1
1 20 CC-2
0 0
0 2 4 6 20 30 40
Displacement (mm) Crack length, a (mm)
Fig. 24.15 (a) Load–displacement and (b) R-curves from CT tests, (c) load–displacement and (d)
R-curves from CC tests
Fig. 24.16 Experimental and numerical results of (a) CT and (b) CC tests
Fig. 24.17 (a) Test setup, (b) strain field from DIC and FEA, (c) stress–strain curves, (d)
modulus–strain, (e) load–displacement
laminates [29, 30], as well as the formation of permanent indentations after impact
events [20, 31]. Therefore, a model dealing with shear non-linearities is required
to accurately predict the failure under multiaxial loading states, and appropriate
material characterisation tests are required. The V-notched rail shear test method
(ASTM standard D7078/D7078M-12) [32] is essentially a combination of the best
features of two commonly used methods, the Iosipescu shear [33] and the two-rail
shear test method [34], generating a relatively uniform shear stress state within a
larger gauge section between the V-notches as well as eliminating edge crushing
and the need of multiple loading holes as shown in Fig. 24.17a. The applied load
was measured continuously with a 50 kN load cell. DIC was used to measure the
strain field under shear loading (Fig. 24.17b).
Unidirectional laminate specimens loaded in shear, parallel to the fibres, failed at
very low displacements with a crack propagating through the matrix in the central
notch area with little plastic deformation. Fast fracture was triggered by the stress
concentration at the notch tip, as expected. The mode II intralaminar matrix fracture
toughness can be obtained using the area method by dividing the total fracture
energy by the central notch area to yield a value of 12 D 34:48 kJ=m2 .
The loading rates for both static and cyclic tests were 1 mm/min, while the
unloading rate in cyclic tests was 5 mm/min. Monotonic and cyclic loading curves
were coincident up to a displacement of approximately 6 mm. Afterwards cyclic
tests show a certain extent of damage recovery in the fracture process (displacement
range from 10 to 15 mm). The area under the load–displacement curve is the total
energy dissipated in the process. The corresponding in-plane shear stress–strain
curve, until final failure, for a cross-ply laminate specimen, is shown in Fig. 24.17c.
The specimen shows initial elastic behaviour, giving a shear modulus of elasticity of
G12 D = D 5:02˙0:2 GPa. The offset in-plane shear strength was determined
724 B.G. Falzon and W. Tan
by translating the shear chord modulus of elasticity line along the strain axis from
the origin by 2 % and extending this line until it intersected the stress–strain curve,
to give a shear strength of S12 D 80:81 ˙ 1:5 MPa. The initial linear behaviour was
followed by the onset of matrix plastic deformation (developed from 2 to 10 %).
Several consecutive loading–unloading cycles were applied to specimens, and
considerable permanent shear plastic strain was observed. This inelastic strain is
irreversible due to the presence of plasticity and/or extensive matrix cracking.
To characterise the stiffness degradation, the in situ stiffness was defined as the
secant shear modulus of each loading–unloading cycle as shown in Fig. 24.17c.
The relationship between shear modulus and applied shear strain is plotted in
Fig. 24.17d. It can be seen that the shear modulus degraded quickly in the initial
matrix yielding regime and then reached a constant level of 2.5 GPa during the fibre-
dominated load-bearing process, indicating that the reduced stiffness was mainly
associated with matrix plastic deformation.
The numerically predicted strain contours, at different loading stages, were
compared with experimental results (Fig. 24.17b) and showed good qualitative and
quantitative agreement. The predicted structural deformation of V-notched speci-
mens was similar to those observed experimentally. The overall load–displacement
response of the numerical test correlated well with experimental tests (Fig. 24.17e)
in the elastic region as well as the fibre rotation region and global fracture regime.
The permanent deformation and degraded stiffness were well represented in the
model when unloading was applied. The predicted peak load was shown to be in
good agreement with the test response, indicating the failure criteria were able to
predict the onset of fracture of cross-ply laminates accurately under shear loading.
This section describes the application of the developed material model to simulate
a low-velocity impact event follow by a compression-after-impact test. In order
to demonstrate the effectiveness of the material model as a virtual testing tool,
its performance in predicting the response of impact and residual strength was
24 Virtual Testing of Composite Structures: Progress and Challenges. . . 725
Fig. 24.18 Finite element, (a) impact and (b) CAI test setup models
Once the impact simulation was completed, the damaged specimen was first
stabilised by replacing the impact boundary conditions with those representing a
picture frame clamped around the specimen to yield a 90 130 mm test section.
The out-of-plane displacements of the nodes, in contact with the picture frame,
were constrained to represent a fixed boundary condition (Fig. 24.18b). Although
the CAI test is essentially quasi-static (0.5 mm/min), it was simulated using
ABAQUS/Explicit to avoid the severe convergence difficulties encountered with
implicit analysis when modelling highly non-linear behaviour. The load rate was
chosen at 3.75 m/min to reduce CPU time. Selective mass scaling, which only scaled
elements whose stable time increment was below 1e-07 s (controlled by the contact
algorithm of cohesive surfaces due to the zero thickness), was only employed in the
CAI process to achieve a reasonable run time.
Material properties for T700/M21 in Table 24.3 were obtained from [35]. dir jj
denotes the longitudinal (jj D 11) and transverse (jj D 22) intralaminar fracture
toughness in tension (dir D T) or compression (dir D C). GI and GII are the
interlaminar fracture toughness for mode I and mode II. The mode mixity parameter,
, was determined from experimental measurements provided by Prombut et al. [37]
using the method of least squares.
24.4.2 Results
The intralaminar damage model allowed the various forms of intralaminar failure
to be investigated as the impact event progressed. As shown in Fig. 24.19a, b, the
impactor load/applied stress vs. displacement history shows very good agreement.
Figure 24.19c shows matrix cracking, delamination (modelled using cohesive
contact laws) and a small amount of fibre failure, for each double-ply and interface,
respectively. Fibre damage is found to be perpendicular to the ply orientation, while
matrix damage is parallel to the fibre direction. The delamination contours shown
in Fig. 24.19c correlate well with results obtained from C-scan studies, which are
parallel to the lower ply fibre orientation. CAI intralaminar damage plots for each
ply pair and delaminations are shown in Fig. 24.19d. During the CAI process, new
delamination and intralaminar matrix damage developed from the impacted region.
Fibre damage was primarily observed in the top and bottom plies. Local buckling
induces delamination growth and further provokes the instability. The predicted
damage correlated well with experimental findings [35].
Table 24.3 Material properties for T700/M21 [35, 38]
Property Values
Elastic properties E1 D 130 GPa; E2 D E3 D 7:7 GPa; G23 D 3:8 GPaI G12 D G13 D 4:8 GPa; 12 D 13 D 0:3I 23 D 0:35
Strength X T D 2080 MPa; X C D 1250 MPa; Y T D 60 MPaI Y C D 120 MPaI S12 D 110 MPa
T 2 C 2 T C 2 2
Intralaminar fracture toughness 11 D 133 kJ=m ; 11 D 40 kJ=m ; 22 D 0:5 N=mm; 22 D 2:1 kJ=m ; 12 D 23 D 13 D 2:1 kJ=m ;
2 2 0 0
Interface properties GIC D 0:5 kJ=m ; GIIC D 1:6 kJ=m ; D 1:45; 3 D 20 MPa; sh D 36 MPa
24 Virtual Testing of Composite Structures: Progress and Challenges. . .
727
728 B.G. Falzon and W. Tan
a 10 b
Exp 400 Undamaged
8 Num Exp
300
Stress(Mpa)
Load (kN)
Num
6
200
4
2 100
0 0
0 2 4 6 0 0.5 1 1.5
Displacement (mm) Displacement (mm)
Fig. 24.19 (a) Impact response, (b) CAI response, (c) predicted impact damage contours and (d)
predicted CAI damage contours
24 Virtual Testing of Composite Structures: Progress and Challenges. . . 729
Fig. 24.20 (a) Failure modes and (b) finite element model of wedge-shaped specimen
At the beginning of the crushing process (stage 1), damage was primarily in the
form of local fragmentation due to intralaminar damage. From stage 1 to stage 2, the
reaction force increased gradually to the peak load at which point the entire uniform
cross-section of the crush element came in contact with the platen (displacement,
S D 1.5 mm). Afterwards, extensive delamination initiated with a splaying mode
(stage 3), leading to a sudden drop in the reaction force. The specimen underwent
some bending as the chamfer was being consumed. From stages 4 to 6, mixed-
mode damage with fragmentation in the middle plies and splaying in the outer
plies was observed. Internal debris was created and acted like a ‘wedge’ in driving
delamination.
The evolution of energy dissipated through various mechanisms during crushing
is illustrated in Fig. 24.21c for all specimens tested, verifying the energy balance
relationship between external work done and energy absorbed. In all of these cases,
the majority of energy was dissipated through intralaminar damage, followed by
friction between the crushing platen and specimen, and delamination. The small
amount of viscous energy dissipated is due to the use of bulk viscosity [36] to damp
out spurious oscillations in explicit dynamic simulations. Most of the energy in this
crushing event was dissipated at the initial stage (i.e. displacement from 0 to 2 mm in
Fig. 24.21c) of the crushing process, when the reaction force increased dramatically
as the chamfer was gradually consumed. Hence, the energy dissipated was primarily
due to fragmentation rather than splaying.
Corrugated webs manufactured with carbon fibre/epoxy have been used for several
energy-absorbing applications such as helicopter subfloor [41] and aircraft fuselage
24 Virtual Testing of Composite Structures: Progress and Challenges. . . 731
Fig. 24.21 (a) Experimental morphologies [24] and numerical matrix damage contours, (b) force–
displacement curve of crushing test on [(0ı /90ı )4 ]s specimen, (c) energy dissipation–displacement
curves
Fig. 24.22 Specimen geometries, (a) chamfer and loading direction and (b) cross-section dimen-
sions (mm), (c) optical microscopy of cross-sectional area
®
samples with 12 plies were fabricated using a Collin compression moulding
machine in a consolidation cycle according to the manufacturer’s specifications,
where these specimens are shown in Fig. 24.22a. The orientation of the 0ı ply is
also highlighted. The specimen denoted as ‘Semi-3p’ has semi-circular corrugations
where ‘3p’ denotes the number of semi-circles or ‘half-waves’. The ‘Hat-3p’
specimens have three semi-hexagonal segments. All these shapes have end-lips on
each side for additional stability.
Figure 24.22b shows the two different specimens with a single-sided 45ı
chamfer. The non-chamfered loaded edge of each specimen was polished and
24 Virtual Testing of Composite Structures: Progress and Challenges. . . 733
Damage at the mesoscale can be described with reference to the axial (Fig. 24.23a)
and circumferential (Fig. 24.23b) directions. Damage mechanisms in the axial
direction can be generally categorised by two primary failure modes: (1) splaying
mode where bending loads lead to the formation of continuous fronds, longitudinal
fibre fracture (0ı ply) and pull-out/kinking, transverse splitting/matrix cracking (90ı
ply) and mixed-mode interlaminar delamination. Mode I dominant delamination
was caused by debris, acting like a ‘wedge’ in driving delamination. (2) A
fragmentation mode where compressive loads result in the sequential damage of
material, involving fibre fragmentation of 0ı plies near the contact surface between
plies and the metallic base and the localised damage of 90ı plies due to matrix
crushing and multiple shear cracks at the crushing front of the plies. The final
intralaminar ply failure consists of fibre breakage and matrix cracking due to the
combination of bending, compression and shear (Fig. 24.23c).
The complex crushing damage mechanisms in the circumferential direction of
the tested corrugated structures are evident in Fig. 24.23b. In the 0ı plies, the main
forms of failure are ply splitting and brittle fibre fracture via bending or shearing.
Failure in the 90ı plies was predominantly shear-induced matrix cracking, triggered
by compressive loading. Fibre tensile and compressive breakage can be seen in
Fig. 24.23d on both sides of the corrugation, when the fronds are forced to move
outwards and inwards, respectively (red arrows indicate the trends). Although the
splits occur extensively during crushing, the adjacent plies formed the continuous
fronds or ‘semi-circular rings’.
Fig. 24.23 (a) Damage mechanisms in the axial direction, (b) damage mechanisms in the
circumferential direction, (c) side view of crushed specimens and (d) top view of crushing results
of three specimen types
24 Virtual Testing of Composite Structures: Progress and Challenges. . . 735
Fig. 24.24 (a) Typical FE model of Semi-3p specimen under crushing load and (b) DIC results
Load Response
At the beginning of the crushing process (stage 1), damage was primarily in the
form of local fragmentation as the trigger region was consumed under displacement
control. The reaction force increased gradually to the peak load at which point the
entire uniform cross-section of the crush element came in contact with the platen
(displacement, S D 2.4 2.7 mm for the three specimen types). In the second stage,
delamination initiated with a splaying mode accompanied by extensive fibre tensile
736
a 35 b 1000 Exp
A1 A2 A3 Sim Semi-3p
30 A4 3 A5 Sim 800
Fibre
2 Matrix
25 5
Delamination
Energy (J)
Load (kN)
1 600 Friction
20 4 6 Viscous effect
15 400
10 Semi-3p
200
5
0 0
0 10 20 30 40 50 0 10 20 30 40 50
Displacement (mm) Displacement (mm)
c 30 A1 A2 A3 d 1000 Exp
A4 A5 Sim Sim Hat-3p
25 2 3 4 5 6 800 Fibre
Matrix
20 1 Friction
Energy (J)
Load (kN)
600 Delamination
15
Hat-3p 400
10
5 200
0 0
0 10 20 30 40 50 0 10 20 30 40 50
Displacement (mm) Displacement (mm)
Fig. 24.25 (a) Force–displacement curves and (b) energy dissipation mechanisms of Semi-3p
specimens, (c) force–displacement curves and (d) energy dissipation mechanisms of Hat-3p
specimens
Fig. 24.26 Side view of failure process: (a) Semi-3p specimens, (b) Hat-3p specimens
Fig. 24.27 Semi-3p specimens (left) and Hat-3p specimens (right): experimental and numerical
matrix damage results (a) front and (b) top view
Crushing Morphology
The numerical results from Figs. 24.25, 24.26 and 24.27 showing the crushing
morphology confirm the qualitative accuracy of the present damage model. Matrix
cracking, shear fracture, delamination, lamina bending and fragmentation were well
predicted. The virtual formation of fronds shows excellent qualitative similarity
when compared with experimental results.
24 Virtual Testing of Composite Structures: Progress and Challenges. . . 739
Crashworthiness Assessment
SEA (J/g)
0.6
60
0.4
40
20 0.2
0 0
Semi-3p Hat-3p Semi-3p Hat-3p
Num Exp Num Exp
Fig. 24.28 Comparison of experimental and numerical results of (a) SEA and (b) crush efficiency
References
1. J. Leahy, Airbus’ Global Market forecast 2015–2034 (2015). Available from: http://
www.airbus.com/company/market/forecast/
2. W. Tan et al., Predicting low velocity impact damage and Compression-After-Impact (CAI)
behaviour of composite laminates. Compos. A Appl. Sci. Manuf. 7, 212–226 (2015)
3. Beyond Vision 2020 (Towards 2050), Aeronautics and Air Transport (2013). Available from:
ftp://ftp.cordis.europa.eu/pub/technology-platforms/docs/acare-background-2010_en.pdf
4. GAO, Status of FAA’s Actions to Oversee the Safety of Composite Airplanes (Aviation Safety:
US Government Accounting Office, Washington, DC, 2011), p. 50
5. J. Rouchon, Certification of large airplane composite structures, in Recent Progress and New
Trends in Compliance Philosophy (ICAS, 1990), pp. 1439–1447
6. B. Szabo, R. Actis, Simulation governance: technical requirements for mechanical design.
Comput. Methods Appl. Mech. Eng. 249, 158–168 (2012)
7. D. Systems, ABAQUS Documentation 6.12. SIMULIA (2012)
8. M.L. Benzeggagh, M. Kenane, Measurement of mixed-mode delamination fracture toughness
of unidirectional glass/epoxy composites with mixed-mode bending apparatus. Compos. Sci.
Technol. 56, 439–449 (1996)
9. P.P. Camanho, C.G. Davila, Mixed-mode decohesion finite elements for the simulation of
delamination in composite materials. NASA-Technical Paper 211737(1) (2002), p. 33
10. J. Lemaitre, J.-L. Chaboche, Mechanics of Solid Materials (Cambridge University Press,
Cambridge, MA, 1994)
11. S. Pinho, L. Iannucci, P. Robinson, Physically-based failure models and criteria for laminated
fibre-reinforced composites with emphasis on fibre kinking: Part I: Development. Compos.
A Appl. Sci. Manuf. 37(1), 63–73 (2006)
12. D.C. Fleming, Modelling composite crushing initiation using a cohesive element formulation.
Int. J. Crash. 16(5), 475–485 (2011)
13. L.N. Chiu, B.G. Falzon, R. Boman, A continuum damage mechanics model for the analysis
of the crashworthiness of composite structures: a work in progress, in Proceedings of the 15th
Australian Aeronautical Conference, Melbourne (2013)
14. A. Puck, H. Schürmann, Failure analysis of FRP laminates by means of physically based
phenomenological models. Compos. Sci. Technol. 58(7), 1045–1067 (1998)
15. G. Catalanotti, P.P. Camanho, A.T. Marques, Three-dimensional failure criteria for fiber-
reinforced laminates. Compos. Struct. 95, 63–79 (2013)
16. M.V. Donadon et al., A progressive failure model for composite laminates subjected to low
velocity impact damage. Comput. Struct. 86(11–12), 1232–1252 (2008)
17. W.H. Press, Numerical Recipes in Fortran 77: The Art of Scientific Computing, vol 1
(Cambridge University Press, Cambridge, 1992)
18. A. Faggiani, B.G. Falzon, Predicting low-velocity impact damage on a stiffened composite
panel. Compos. Part A 41(6), 737–749 (2010)
19. L. Raimondo et al., A progressive failure model for mesh-size-independent FE analysis of
composite laminates subject to low-velocity impact damage. Compos. Sci. Technol. 72(5),
624–632 (2012)
20. W. Tan et al., Predicting low velocity impact damage and Compression-After-Impact (CAI)
behaviour of composite laminates. Compos. Part A 71, 212–226 (2015)
742 B.G. Falzon and W. Tan
21. ASTM, Standard Test Method for Mode I Interlaminar Fracture Toughness of Unidirectional
Fiber-Reinforced Polymer Matrix Composites, in D5528-13, West Conshohocken, PA (2013)
22. A. Offringa, et al., Fiber reinforced thermoplastic butt joint development, in Proceedings of
International SAMPE Symposium and Exhibition (2008)
23. ASTM, Mixed Mode I-Mode II Interlaminar Fracture Toughness of Unidirectional Fibre
Reinforced Polymer Matrix Composites, in D6671/D6671M-13e1, West Conshohocken, PA
(2013)
24. S.T. Pinho, P. Robinson, L. Iannucci, Fracture toughness of the tensile and compressive fibre
failure modes in laminated composites. Compos. Sci. Technol. 66(13), 2069–2079 (2006)
25. ASTM, Standard test method for plane strain fracture toughness of metallic materials, in E399-
90, Philadelphia, PA (1993)
26. ASTM, Standard test method for measurement of fracture toughness, in ASTM E1820-15, West
Conshohocken, PA (2015)
27. S. Jose et al., Intralaminar fracture toughness of a cross-ply laminate and its constituent sub-
laminates. Compos. Sci. Technol. 61(8), 1115–1122 (2001)
28. M.W. Czabaj, J. Ratcliffe, Comparison of intralaminar and interlaminar mode-I fracture
toughness of unidirectional IM7/8552 graphite/epoxy composite. Compos. Sci. Technol. 89,
15–23 (2012)
29. L. Feo, G. Marra, A.S. Mosallam, Stress analysis of multi-bolted joints for FRP pultruded
composite structures. Compos. Struct. 94(12), 3769–3780 (2012)
30. F.P. van der Meer, C. Oliver, L.J. Sluys, Computational analysis of progressive failure in a
notched laminate including shear nonlinearity and fiber failure. Compos. Sci. Technol. 70(4),
692–700 (2010)
31. D. Fanteria, G. Longo, E. Panettieri, A non-linear shear damage model to reproduce permanent
indentation caused by impacts in composite laminates. Compos. Struct. 111, 111–121 (2014)
32. ASTM, Standard test method for shear properties of composite materials by V-notched rail
shear method, in D7078/D7078M-12, West Conshohocken, PA (2012)
33. ASTM, Standard test method for shear properties of composite materials by the V-notched
beam method, in D5379/D5379M-12, West Conshohocken, PA (2012)
34. ASTM, Standard test method for in-plane shear properties of polymer matrix composite
materials by the rail shear method, in D4255/D4255M-15a, West Conshohocken, PA (2015)
35. S. Rivallant, C. Bouvet, N. Hongkarnjanakul, Failure analysis of CFRP laminates subjected to
compression after impact: FE simulation using discrete interface elements. Compos. Part A 55,
83–93 (2013)
36. A.I.T. Method, Determination of Compression Strength after Impact, AITM 1-0010 (2010)
37. P. Prombut et al., Delamination of multidirectional composite laminates at 0ı /™ı ply interfaces.
Eng. Fract. Mech. 73(16), 2427–2442 (2006)
38. N. Hongkarnjanakul, C. Bouvet, S. Rivallant, Validation of low velocity impact modelling on
different stacking sequences of CFRP laminates and influence of fibre failure. Compos. Struct.
106, 549–559 (2013)
39. H.A. Israr, S. Rivallant, J.J. Barrau, Experimental investigation on mean crushing stress
characterization of carbon–epoxy plies under compressive crushing mode. Compos. Struct.
96, 357–364 (2013)
40. W. Tan, B.G. Falzon, M. Price, Predicting the crushing behaviour of composite material using
high-fidelity finite element modelling. Int. J. Crashworthiness 20(1), 60–77 (2015)
41. M. Waimer et al., Experimental study of CFRP components subjected to dynamic crash loads.
Compos. Struct. 105, 288–299 (2013)
42. S. Heimbs, F. Strobl, P. Middendorf, Integration of a composite crash absorber in aircraft
fuselage vertical struts. Int. J. Veh. Struct. Syst. 3(2), 87–95 (2011)
43. P. Feraboli, Development of a corrugated test specimen for composite materials energy
absorption. J. Compos. Mater. 42(3), 229–256 (2008)
44. W. Tan, B.G. Falzon, Modelling the nonlinear behaviour and fracture process of AS4/PEKK
thermoplastic composite under shear loading. Compos. Sci. Technol. 126(1), 60–77 (2016)
24 Virtual Testing of Composite Structures: Progress and Challenges. . . 743
45. Cytec, APC-2 PEKK Thermoplastic Polymer Technical Data Sheet, Cytec Engineering Mate-
rials (2012)
46. M. Dao, R.J. Asaro, On the critical conditions of kink band formation in fiber composites with
ductile matrix. Scr. Mater. 34(11), 1771–1777 (1996)
47. W. Tan, Modelling the Behaviour of Composite Structures Under Impact and Crush Loading
(Queen’s University Belfast, Belfast, 2016)
48. G.C. Jacob et al., Energy absorption in polymer composites for automotive crashworthiness. J.
Compos. Mater. 36(7), 813–850 (2002)
Chapter 25
Contribution of Virtual Simulation
to Industrialisation of Carbon Fibre-Reinforced
Polymer (CFRP) Composites for Manufacturing
Processes and Mechanical Performance
Anthony K. Pickett
25.1 Introduction
Many early analysis methods for advanced composites were developed as CFRP
materials became commercially available during the 1950s and 1960s. For manu-
facturing analytical kinematic draping methods originally developed for the clothing
industry were available to predict fabric draping, and classical laminate theory was
developed by numerous authors for mechanical analysis of laminates. Early failure
criteria were based on the ‘Rankine’ maximum stress and Saint-Venant maximum
strain criteria or other simple phenomenological models that described a failure
envelope from principle composite failure strengths.
The finite element method developed during the 1950s opened many new
possibilities for composites analysis. The first general-purpose commercial FE
codes appeared during the 1960s and could be used for linear stress analysis of
laminate composite structures. During the following 20 years, advances in these
codes were evolutionary with better elements for laminate analysis, improved ply
failure criteria and stress analysis for coupled mechanical-hygrothermal loading
conditions. During this period constant progress in computer hardware, numerical
models and software allowed the method to tackle ever-larger structures with greater
accuracy and undertake linear buckling and frequency analysis.
During the 1970s, FE techniques based on dynamic explicit analysis were applied
to very high-velocity short-duration military impact problems. The availability of
highly vectorised Cray-1 and later X-MP supercomputers in the 1980s opened
new opportunities to apply these methods to a lower-velocity (longer duration)
manufacturing and crash problems. The first successful full car crash simulation
was undertaken as a German consortium project in 1985 [1], followed by a
European project to develop metal stamping simulation in 1989 [2]. In both fields
full industrialisation of the technology took 5–10 years as necessary software
capabilities were developed, hardware CPU performance improved and companies
gained confidence to support existing design practices with simulation results. The
ability to virtually design crash and manufacturing has been a major reason for
significant reduction in time to design a new car, from typically 4 to 2 years, and
substantial cost savings. Simulation has been largely responsible to enable optimal
designs and has made a real contribution to improved vehicle safety. Today, it is an
indispensable part of the design process, with prototype testing either avoided or
kept to a minimum.
Application of these simulation techniques to composite manufacturing and
structural analysis also took place, albeit later and at a slower pace due to
fewer industrial applications and greater complexity of the materials and their
processing. Concerning manufacturing simulation the first notable attempt involved
thermoforming of continuous fibre-reinforced thermoplastic sheets for aerospace
and automotive applications in 1992 in a European project [3]. Subsequently, the
technology was extended to dry and hot forming of prepregs, allowing accurate
FE draping simulation that overcame many limitations of analytical geometric
methods. Other applications of process simulation in recent years have included
braiding, filament winding and tape laying, all of which have used explicit FE
analysis. Implicit FE methods are usually used for consolidation and residual stress
deformation analysis. Also, FE resin infusion simulation of dry fabrics has become
an important tool to support industrial design and optimisation of the process.
Failure modelling of composites has made considerable advances over the past
20 years with much improved criteria that represent physically observed composite
failure. The worldwide exercise [4] has helped encourage developments on failure
modelling and has provided a unique database of failure information from which
researchers could develop new validated models. Most of this work to date has
focussed on unidirectional composites, using concepts first proposed by Hashin [5]
and by Puck [6] based on the Mohr-Coulomb criteria. Some of these developments
are now finding their way into commercial codes for implicit failure analysis.
Unfortunately, the focus to date has been unidirectional composites, and future
work is necessary for failure of woven and general 2D/3D textile composites. In
parallel advances in composite impact and crash simulation have been made with
new models for ply damage and inter-ply delamination which are now implemented
in most commercial FE codes.
This chapter will first overview manufacturing methods and areas where simula-
tion can support process and composites mechanical analysis. A brief description of
the main FE methods used is given, followed by a number of example applications
demonstrating the evolution of these methods, the state-of-the-art and some possible
future directions for research and developments. Others chapters in this book
focus on research aspects of process and mechanical analysis; the emphasis here
is developments and capabilities of commercial FE codes since these are used
by industry by industry today and contribute most to industrialisation of CFRP
composites.
25 Contribution of Virtual Simulation to Industrialisation of Carbon Fibre: : : 747
Many different methods are available to manufacture composites, for which dif-
ferent software tools have emerged. Figure 25.1 shows some methods that are
separated into those using dry reinforcements that must be infused and those
using a fibre-resin mix that must be consolidated. Generally all methods must first
place fibres and reinforcement, followed by a subsequent processing method, either
infusion or consolidation, to convert these to a successful cured composite. The final
composite structure is then analysed depending on its application and structural
loading requirements; these can range from simple laminate analysis through to
damage propagation under dynamic loading conditions. In each of these areas,
simple analytical tools and more sophisticated FE-based simulation methods have
evolved to support industry design and development.
Finite element techniques for most manufacturing and structural analysis are either
based on an ‘implicit’ or ‘explicit’ solution. Implicit methods were developed first
and are mainly used for linear analysis of statically loaded composite structures
for stress, failure and deformation and for linear buckling, frequency, thermal and
similar field analysis problems. Basically, stiffness matrices for each element in
the structure are computed and assembled to give the structure stiffness [K]. From
inversion of [K] and application of applied nodal loads fFext g, it is possible to
determine nodal displacements fug, Eq. (25.1). From these nodal displacements,
element strains are found and element stresses are computed using the material
constitutive law:
For geometric and material non-linear problems, [K] is not a constant and varies
with loading. Implicit methods can tackle such problems, but iterative solution
schemes with regular updating of structural stiffness and convergence criteria are
Fig. 25.1 Possible manufacture, process and analysis options to realise a composites part
748 A.K. Pickett
needed. For highly non-linear problems, or problems involving contact, the implicit
method can be particularly difficult to converge.
For many manufacturing problems involving large deformations, material non-
linearity and contact such as between composite sheets and tools, the explicit
FE method has proven the most efficient and robust solution method. Essentially,
conventional finite elements are used to discretize the problem, but in this case, no
structure stiffness is assembled, and the problem is solved as a dynamic one using
the linearised dynamic equations of motion, Eq. (25.2):
˚:
ŒM fRugn C ŒC u n
C ŒK fugn D fFext gn (25.2)
where [M]fügn are nodal inertia forces due to nodal masses and accelerations.
˚:
For short-duration problems, the velocity-dependent damping term ŒC u n is
neglected. A solution is obtained by rearranging Eq. (25.2) and using a time-
stepping scheme with central finite difference Eq. (25.3):
where n is the number of timesteps and Tn is the timestep that depends on the
Courant criteria .D Lmin =c/, which is the smallest element in the mesh divided by
the material sound speed. Provided the integration timestep is below the critical
value, a solution is ˚stable and can be done on a node-by-node basis. At each cycle
:
updated fügn and u nC1=2 and fugnC1 are found from which updated element
strains and stresses are computed. At the end of each cycle, new displacements
fugnC1 replace fugn in Eq. (25.3) and are used to compute new internal forces;
the procedure then repeats with an updated timestep. The advantage is nodal
displacement increments, and element strains are very small within one timestep
allowing a pseudo-linear solution. Also, contact is easily treated by applying penalty
forces to resist intrusion of contacting elements; these are applied as temporary
additional external nodal forces that modify nodal accelerations.
b c
Fig. 25.2 Examples of principle deformation mechanisms for drapeable fabrics. (a) Interfibre
shear. (b) Interfibre sliding. (c) Fibre buckling. (d) Fibre straightening (for woven-type fabrics)
essentially prevents fabric shearing so these fabric types are limited to flat shapes
or simple bending. Friction sliding of fabric layers in a stacked system is another
possible deformation mode.
Geometric methods are still popular today for fabric draping analysis and
are widely used in CAD packages. However, these have many limitations, and
consequently, starting in the 1990s, industry and software developers have worked
towards improved methods based on FE techniques. This work is ongoing, and
further improvements in constitutive models are still under development. One
difficulty of this approach is that shell-type elements are not able to capture any local
deformation mechanisms that may occur at the individual yarn level. As a result new
research is investigating mesoscale fabric models for better representation of yarn
architecture for drape simulation. An overview of these methods, the state of the art
for industry simulation and future research, is presented.
L1
P
b L2
i-1 j
a
a surface f(x, y, z)
i j-1
b
i j
2 2 2
xi;j xi1;j C yi;j yi1;j C zi;j zi1;j D a2i
2 2 2 (25.4)
xi;j xi;j1 C yi;j yi;j1 C zi;j zi;j1 D b2j
f xi;j ; yi;j ; zi;j D 0 (25.5)
Some basic assumptions of the mathematical model have been discussed by Rudd
et al. [8]. Furthermore it should be noted that the starting point P and projected fibre
directions are unlikely to be correct in real draping problems, and important effects
such as fabric stiffness, friction and process features such as blankholders are all
ignored.
Early work on composite forming was started in the 1990s, for example [9],
which used an implicit FE code to form thermoplastic sheets. At about this time,
metal stamping simulation using explicit FE codes was proving successful in the
industry. In order to build on this success, a major research effort was started with
a Brite-Euram consortium project [3] for thermoforming simulation of continuous
fibre-reinforced thermoplastics using an explicit code [10]. The main steps of the
process are shown in Fig. 25.4; these involve preheating the sheets to a forming
temperature, stamping in hot (or cool) tooling, cooling, extraction and trimming.
At the time thermoforming was considered a particularly promising technology for
fast manufacturing, but was found difficult to control, especially residual thermal
stresses which often caused part distortions and satisfactory consolidation due to
sufficient pressure, temperature and time could be unpredictable. For these reasons
there was considerable industrial interest to develop new analysis tools.
The Brite-Euram project developed a dedicated software to analyse preheating,
plus various new options for a commercial explicit FE including a viscous fabric
models coupled to thermal analysis, viscous frictions laws and a methodology to
model stacked plies that could slide relative to each other. For post-processing
temperature and forming, defects such as buckling and prediction of consolidation
25 Contribution of Virtual Simulation to Industrialisation of Carbon Fibre: : : 751
Fig. 25.4 Heating, stamping, cooling/trimming and extraction operations in thermoforming. (a)
Sheet heating. (b) Transfer to tooling. (c) Stamping. (d) Extraction and trimming
were some new options. A large and essential part of the project undertook industrial
validation using progressively complex parts from simple shapes to industrially
relevant components. For the project carbon fibre-reinforced thermoplastic, PEI
(Cetex), PEEK (APC2-AS4) and a low-cost glass-reinforced PA (Vestopreg) were
all characterised.
One example simulation is shown in Fig. 25.5 for thermoforming eight plies
of woven 50/50 T300 carbon fabric PEI-Cetex with layup [45,45,0,90]s. The
matched metal tools had a dimension 875 mm 500 mm with a central 300 mm
long by 50 mm deep ‘sikken’, which is intended as a geometric shape for stiffening.
The study was specifically designed to validate prediction of ply buckling and
consolidation using a specifically developed model based on the theoretical work
of Lee and Springer [11]. Calibration of the consolidation model used experimental
tests with various processing limits (cooling rates and pressure) which gave varying
degrees of consolidation [12], from which the following governing equation and
model parameters were determined for PEI-Cetex:
Z 15
a b 1
t
Papp . / .bo C wo /
Dic D w oCob 5 C 5 d Parameters CETEX
o o ao o ŒT . / .ao bo /3
ao (m) 224:0
bo (m) 603:0
wo (m) 1219:0
0:73
Fig. 25.6 Draping simulation of a biaxial non-crimp fabric showing (a) simulation setup, (b)
manufacture, (c) contour of fabric shear and (d) contour of fabric thickness
are shown which indicate the fabric has reached its limit of 60ı , and contours of
fabric thickness are shown in Fig. 25.6d showing the initial thickness 0.1875 mm has
exceeded 0.5 mm in some regions. The fabric deformations do show some buckling,
but only outside the region of interest. For such simulations fabric shear stiffness is
most important, and this is calibrated against a pure shear ‘picture frame’ test or
can be estimated from a bias extension shear test. For thickness either a constant
volume is assumed or more correctly thickness is measured as a function of fabric
shear angle and provided as input for the constitutive model.
Some software products and ongoing research work are developing ‘chaining’ to
link process steps in composite manufacturing. For example, residual stresses from
curing could be brought through as residual stresses for mechanical analysis, but
the most developed areas to date are transfer of fibre direction information from
draping to infusion analysis for modification of permeabilities and transfer of fiber
information to mechanical analysis for modification of stiffness. A further important
result from this simulation is prediction of correct fabric thickness distribution as
shown for the draped part in Fig. 25.6d. For the proper design of matched tooling
for resin transfer moulding (RTM) infusion, this thickness variation should be taken
754 A.K. Pickett
Fig. 25.7 Fabric architecture, unit cell model and a mesoscale drape simulation
into account. Although such ‘chaining’ of information is possible and now starting
to be applied in industry, it should be mentioned that further work is needed in this
area. For example, composite failure behaviour will require new constitutive models
that properly account for effects of shear; a simple transfer of fibre directions and
thickness is insufficient.
Fig. 25.8 Mesoscale drape simulation with coupling to infusion and impact simulation. (a)
Mesoscale draping simulation. (b) Mesoscale infusion simulation. (c) Mesoscale impact simulation
25.10 Braiding
where ı y is the spacing between axial yarns for Ny yarns. These simplified kinematic
relations are based on simple geometric considerations that ignore many important
effects in real braiding processes such as yarn stiffness, yarn friction and yarn
interactions.
A first attempt at braiding simulation used a dynamic explicit FE code for the
more complex case of 3D braiding which was developed through two German
756 A.K. Pickett
Fig. 25.9 Example rotary braider and a braided preform. (a) A modern braider with 176 bobbins
and robot control for the mandrel. (b) Typical biaxial braided perform
national projects in the 1990s [13]. The prototype 3D braider is shown in Fig. 25.11,
where bobbins are held on rotating horngears and can be transferred to adjacent
horngears via switching points, so arbitrary movement across the baseplate is
possible. The unique feature is that braids can be produced with different types of
sections in the length direction. Unfortunately there were many technical difficulties
and the technology proved too complicated for practical manufacture. However, the
project did provide a basis to develop simulation techniques for 2D braiding.
A simulation setup for 2D braiding is shown in Fig. 25.12, where bar and beam
elements are used for the yarns, contacts are defined between yarns and a shell
model for the braiding ring and mandrel is defined. Bobbins are represented with
special spring elements that control tension and are given serpentine velocities for
their motion. Such a simulation gives accurate prediction of yarn trajectories and
proper modelling of yarn interlacing leading to a good estimate of braid architecture.
Figure 25.13 shows test and simulation results for braiding a generic part with
several different section shapes and flanges [14]. A reasonable agreement between
simulation and analytical results for yarn paths is found, but some differences
are seen, especially after the first flange and along the larger cylindrical section,
Fig. 25.13c. In these regions the analytical path starts to diverge from test and
25 Contribution of Virtual Simulation to Industrialisation of Carbon Fibre: : : 757
Fig. 25.11 Early (ca. 1992) prototype 3D braider with an example 3D braid
simulation results. The simulation model was able to predict observed test imper-
fections including yarn bridging over the flange and over-compaction on the square
section causing a very loose fit to the mandrel.
Future work will have to try to develop techniques that use braid simulation
results as a basis for other possible process steps such as estimation of permeability
for infusion simulation or for stiffness and failure analysis. Some first attempts in
this direction are shown in Fig. 25.14 where the simulated braid architecture is
extracted and overlaid with a solid FE model for the matrix. The two models are
joined by modifying and merging nodes, but constraint functions could be another
possibility. Figure 25.14c shows analysis results of matrix principle stresses and
758 A.K. Pickett
Fig. 25.13 Simulation of a generic braided part and comparison to analytical yarn paths. (a)
Mandrel geometry with overlaid comparison of analytical and FE solutions for a typical yarn
path. (b) FE braid architecture. (c) Braided preform with overlaid comparison of analytical and
FE solutions for a typical yarn path
Fig. 25.14 Coupling of braid simulation with mechanical analysis for stiffness. (a) Braid archi-
tecture. (b) Merging the yarn and matrix models. (c) Load application and internal loadings
yarn forces from an applied static torsion loading. This approach has been shown
to give a good prediction of stiffness, but is unlikely to be a sound basis for failure
prediction as internal stresses will be approximate. Another possibility for stiffness
prediction could be to use the yarn direction and spacing information as input to
dedicated micromechanical codes WiseTex [15] or TexGen [16].
Manufacture of composites that infuse liquid resin into a dry preform at the part
manufacturing stage is usually referred to as liquid composite moulding (LCM) [8].
The main processes are either resin transfer moulding (RTM) or liquid resin infusion
25 Contribution of Virtual Simulation to Industrialisation of Carbon Fibre: : : 759
Fig. 25.15 The two main filling methods, (a) RTM and (b) VARI
Fig. 25.16 Experimental setup, filling and FE infusion model with example results. (a) Cross
section of infusion setup and experiment. (b) Infusion and numerical model results
setup. For the simulation model shown in Fig. 25.16b, one quarter symmetry with
appropriate boundary conditions was used with 580,000 tetrahedral elements. Plies
were oriented with one element layer per ply, and pressure boundary conditions
were specified that conform to the test setup. In this case a good agreement was
found at an intermediate filling state with test 224 s and simulation 237 s; for final
filling the agreement was less good with a 20 % difference. Generally, infusion
filling simulation is a challenge; often simple imperfection in the preform can
lead to internal race tracking, and obtaining good permeability data for the fabric,
particularly if it has been deformed or compressed in the tooling, is difficult.
Nevertheless, these analyses are a valuable aid to help design and optimise the
injection process.
Other chapters in this book focus on details of failure and damage modelling of
carbon composites, and only some general comments concerning status and recent
progress in commercial FE codes are briefly discussed. Currently composite damage
constitutive models are based on macro-mechanical phenomenological approaches.
That is, a homogenised state of stress, or strain, over the element volume is tested
against a postulated failure envelope that describes experimentally the observed
composite strengths. This failure envelope is described via discrete points from
coupon tests, and interpolation between points is ‘hopefully’ correctly represented
by the chosen envelope; for damage a post-failure degradation of strength is imposed
depending on the type of fibre or matrix failure.
25 Contribution of Virtual Simulation to Industrialisation of Carbon Fibre: : : 761
25.13 Conclusions
fragmentation and failure and damage models for deformed fabric composites such
as draped fabrics.
Acknowledgements Special thanks are due to former colleagues at ESI GmbH. Also, to past and
present researchers at Cranfield University, UK, and the Institute for Aircraft Design, Stuttgart
University, Germany, for their collaboration on many aspects of this work.
References
20. C. Davila, P. Camanho, C.A. Rose, Failure criteria for FRP laminates. J. Compos. Mater. 39,
323–345 (2005)
21. PAM-CRASH (Paris, France, 2016), https://www.esi-group.com/software-solutions/virtual-
performance/virtual-performance-solution
22. P. Ladeveze, E. LeDantec, Damage modelling of the elementary ply for laminated composites.
Compos. Sci. Technol. 43, 257–267 (1992)
23. J. Lemaitre, J. Chaboche, Mechanics of Solid Materials (Cambridge University Press, Cam-
bridge, 2000)
24. C. Davila, P. Camanho, A. Turon, Effective simulation of delamination in aeronautical
structures using shells and cohesive elements. J. Aircr. 45(2), 663–672 (2008)
25. K. Park, G. Paulino, Cohesive zone models: a critical review of traction-separation relation-
ships across fracture surfaces. Appl. Mech. Rev. 64, 1–20 (2011)
26. J. LLorca et al., Multiscale modeling of composite materials: a roadmap towards virtual testing.
Adv. Mater. 23, 5130–5147 (2011)
Part VII
Chapter 26
Multi-scale Progressive Failure Modeling:
From Nano-structured Carbon Fibers
to Textile Composites
A. E. Bogdanovich
26.1 Introduction
Very diverse materials and structures that can be analyzed using generic 3-D mosaic
chain models are illustrated in Fig. 26.1; the chains are built up of homogeneous
material bricks (HMBs). It is assumed that each HMB has assigned (generally
anisotropic) effective elastic and strength properties. These models may represent
individual fibers (e.g., carbon, aramid, glass, polymer), unidirectional composites,
laminated composites, textile composites, and textile composite laminates. The
elastic and strength property variation from one brick to the other within the chain
may be caused by any possible factors, including different orientations of the
principal material axes of symmetry (like in the case of curved fibers/tows), different
fiber volume fractions, different strength characteristics, etc. As always, when the
analysis approach and computational model are relatively simple while general in
nature, there are inevitable limitations. In the case of 3DMC model, the following
two can be noticed: (I) the chain cross sections are assumed rectangular and (II) the
mechanical property variation along the chain is stepwise; therefore, the material
property transition from one brick to the other cannot be made continuous.
The model construction starts with connecting HMBs into a single chain as
shown in Fig. 26.1a. After that, two individual chains can be connected in parallel
either directly (as in Fig. 26.1b) or via some intermediate chains (as in Fig. 26.1c);
the latter kind of a chain may be viewed either as a “matrix” chain or as an
“interphase” chain. Of course, any number of matrix and/or interface chains can
be connected in parallel within a block of chains. The next steps of the composite
material model construction are to assemble such elementary blocks of chains
770 A.E. Bogdanovich
Fig. 26.1 Examples of 3DMC models: one chain (a), two chains (b), and three chains (c).
Different brick colors relate to their different material properties
into more and more complex structures. The assemblies may expand in all three
directions, i.e., along coordinates x, y, and z, and any resulting chain assembly can
be viewed as the specific case of general 3-D mosaic model.
It is worth pointing out that arbitrary curvature/misalignment of an individual
fiber/tow can be accounted in this type of a model by imposing respective
stepwise stiffness and strength variation along the chain, while arbitrary cur-
vature/misalignment of different fibers/tows can be accounted by the property
variation from one chain to the other. When two chains joined directly (as in
Fig. 26.1b), they can be either “perfectly bonded” (by imposing the three displace-
ment continuity conditions along the entire interface between the chains) or may
have “local disbonds” (by removing the displacement continuity conditions between
adjacent bricks). As illustrated in Fig. 26.1c, the third material can be inserted
between the chains with imposing perfect bonding or allowing for local disbonds.
The fiber volume fraction in 3DMC models can be easily varied by changing the
relative dimensions of the reinforcement chains vs. matrix/interphase chains.
Fig. 26.2 Illustration of the RVE (a) discretization into EMBs (b) and further homogenization of
the EMBs
element (aka unit cell) is periodically repeated in one, two, or all three dimensions
throughout entire structure, however, as many distinct RVEs as necessary can be
incorporated in the model. Second, the RVE is discretized into some number of
elementary material bricks (EMBs) using three sets of planes perpendicular to x, y,
and z coordinate axes. As illustrated in Fig. 26.2, each of the resultant EMBs may
be either homogeneous (if it only contains a matrix material or one fiber material) or
inhomogeneous (if it contains both matrix and fiber or more than one fiber). Third,
elastic properties of fiber(s) and matrix occupying each EMB are homogenized
within its volume that results in the model containing only homogeneous material
bricks (HMBs). The computational procedures include determination of the matrix
volume and the volumes of each distinct fiber material present within the EMB.
If the fiber is anisotropic, the characteristic spatial orientation of each fiber
volume within the EMB has to be determined. For the latter purpose, the original
discretization into EMBs (illustrated in Fig. 26.2) should be sufficiently fine as the
fiber segment centerline could be accurately represented by a straight line within
each EMB. Conventional computational procedures, such as voxel sorting technique
and orientational averaging method (see Pastore et al. [8] and Bogdanovich and
Pastore [6]), can then be applied for the computation of effective elastic properties
of each individual EMB.
Further on, for the purpose of failure analysis, effective strength properties
(which are typically expressed in terms of ultimate strains or stresses) of each
EMB have to be determined. This task is simple for all homogeneous EMBs that
contain only a matrix or only one fiber material, but it gets more complicated for the
EMBs that contain one fiber segment (notably inclined!) surrounded by the matrix.
In this case using some failure criteria for respective unidirectional composite
can be attempted, but finding experimental strength data for a given fiber volume
fraction may be challenging. Inevitably, different EMBs would have different (and
772 A.E. Bogdanovich
a priori unknown) fiber volume fractions, so the only practical alternative could
be performing a supplementary set of the failure analysis runs for each distinct
EMB to generate the necessary strength input data for the whole RVE analysis. It is
worth noting that even if the simplest maximum strain or maximum stress criteria
is used for an EMB, in order to generate the full set of ultimate strains or stresses,
nine failure analysis runs should be conducted under different loading conditions
(e.g., three in tension, three in compression, and three in shear). Therefore, even the
preliminary step of generating strength input data for the constituent EMBs may be
very time-consuming.
The above-described problem becomes even more challenging if two or more
different fiber segments are contained within the same EMB (such possibility
is obvious from Fig. 26.2). Those different fiber segments would likely have
different volume fractions and different orientations and, possibly, also different
material properties. In this case there is no doubt that the only feasible option
would be to determine all strength input data theoretically. Similarly to the above-
suggested approach for one fiber segment in each EMB, in the case of two or
more fiber segments, all effective strength properties of each distinct EMB can
be determined by solving a number of separate failure analysis cases for different
loading directions. The output of those analysis cases can be then used as the input
strength data for the RVE failure analysis and strength predictions. Keeping in mind
that there may be tens or even hundreds of distinct EMBs in a real-life composite
RVE, and that at least nine ultimate stress/strain values have to be determined for
each EMB, the set task could already become enormously cumbersome and time-
consuming at the stage of input data preparation for the RVE analysis.
Although the above-described analysis approach if feasible in principle and there
are all the necessary tools in currently available 3-D mosaic code to realize it,
the amount of the analyst’s work and computational time required for conducting
full hierarchical analysis, from an individual fiber to a complex laminate or textile
composite, may be prohibitive. Due to this reason, it was assumed in several initial
demonstrations of this analysis approach reported in Bogdanovich [2, 9] that (a)
there is no more than one fiber segment within each EMB, (b) all EMBs contain the
same fiber type, and (c) all EMBs have the same fiber volume fraction. Also it was
assumed that the required nine ultimate strain values (due to the maximum strain
failure criterion was used for each HMB) are known; hence, the step of determining
those values from separate analysis runs was skipped. Future work required to
demonstrate how to solve specific problems without assumptions (b) and (c). The
most challenging analysis cases would be when assumption (a) is not applicable.
Either the chains composed of HMBs in Fig. 26.1 or the RVE composed of HMBs in
Fig. 26.2b can be viewed as specific cases of general 3DMM illustrated in Fig. 26.3.
Here, each (distinctly colored) HMB is treated as elastic anisotropic material
26 Multi-scale Progressive Failure Modeling: From Nano-structured Carbon. . . 773
z
a z
b
y y
x x
zN+1 zN+1
zN zN
... ...
z3 z3
z2 yM+1 z2 yM+1
yM yM
... ...
z1 y1 y2 z1 y1 y2
x1 x2 x3 ... xL xL+1 x1 x2 x3 ... xL xL+1
Fig. 26.3 Generic 3DMM composed from homogeneous anisotropic materials bricks (a) and
an example of its discretization into hexahedral elements using three sets of (dashed) mutually
orthogonal planes (b)
with already determined (by the use of either experimental or theoretical means)
effective elastic and strength properties. The HMB assembly includes imposing
the displacement continuity conditions between the bricks. If necessary, additional
strain continuity conditions can be selectively imposed. Further on, the boundary
value problem formulation requires to apply either kinematic (e.g., formulated in
terms of displacements) or static (e.g., formulated in terms of stresses) external
boundary conditions. The displacement boundary conditions are imposed uniformly
along each HMB exterior surface element. The stress boundary conditions are also
applied at each HMB exterior surface element, but in a “soft” variational sense;
details can be found in Bogdanovich [7].
Further on, a computation element mesh (CEM) has to be chosen for each interval
fx1 , x2 g, : : : , fxL , xLC1 g in x direction, fy1 , y2 g, : : : , fyM , yMC1 g in y direction,
and fz1 , z2 g, : : : , fzN , zNC1 g in z direction as illustrated in Fig. 26.3b. For each
interval, the CEM can be chosen individually. Either uniform or nonuniform user-
assigned meshes are allowed; the nonuniform meshes are defined by the ratios of
computational element dimensions within the HMB along the three coordinates.
In addition to that, user can select the degree of Bernstein polynomial basis
functions (1, 2, 3, : : : ) by entering respective number from the code interface.
This is particularly useful for the convergence studies, because a sequence of
finer analysis cases can be solved with increasing accuracy using the same CEM.
As was shown in Bogdanovich [7], in a broader sense a CEM described here is
analogous to a conventional finite element mesh, and each computational element
enclosed between the dashed planes in Fig. 26.3b can be viewed as an 8-noded
3-D hexahedral finite element with Bernstein approximation polynomials serving
the same purpose as shape functions in conventional FEA. However, contrary to the
case of Lagrange interpolation polynomials, Bernstein approximation polynomials
(those are generated by special recursive procedure incorporated in 3-D mosaic
code) do not add “nodes” when the polynomial degree is increased; instead they
add degrees of freedom to the element via increasing the number of Bernstein basis
functions involved in the analysis. The computational element always remains 8
noded, and the “shape functions” do not oscillate; as was previously shown, the
latter feature allows one to avoid computational instabilities even when using 5th,
10th, or even 15th degree Bernstein polynomials.
774 A.E. Bogdanovich
Also importantly, the procedures implemented in this analysis approach for the 3-
D strain-stress computations are different from the ones used in conventional FEA.
The displacements, strains, and stresses are computed point-wise for each given
triad of fx, y, zg coordinate values and that computation is performed by directly
using the displacement approximations; details can be found in Bogdanovich [7].
This feature is especially valuable, because it enables to perform a point-wise failure
analysis at the user-assigned grid of points within 3DMM using those point-wise
computed strains or stresses, as described in Bogdanovich [2, 9]. Generally, any
desirable combination of the computed strain and/or stress components can be incor-
porated in a point-wise failure criterion. The aforementioned papers illustrated some
applications of the described progressive failure analysis methodology and strength
predictions of 3-D woven composite unit cells. Comparisons with experimental
strength properties showed very good agreement for the in-plane tensile loading
cases.
The new cases of 3-D stress-strain and progressive failure analysis solved for the
3DMC models shown in Fig. 26.1, and other similar models revealed interesting
effects of the chain curvature variation, assembly of chains having different
curvatures, matrix volume content, and mechanical properties of the matrix material.
A 10-brick chain shown in Fig. 26.1a can be interpreted as a single curved fiber
embedded in some matrix or a resin-impregnated fiber tow with curved fibers. In
that case all fibers in a tow are oriented within the same brick at identical angle
with respect to axis z (different brick colors correspond to different inclination
angles). Under tensile loading in z direction, all bricks carry the same load (i.e.,
are exposed to the same stress level); therefore, the brick having lowest stiffness
in z direction develops the highest z-directional strain and, naturally, that brick
fails first. For a single-chain model, the first brick failure means also the ultimate
chain failure, because the whole chain is as strong as its weakest brick. In the
case of two perfectly bonded chains in Fig. 26.3b, the longitudinal stress and strain
distributions among the bricks become much more complex and cannot be predicted
by elementary means. Intuition says that the weakest link in this case would be at
the location of two bricks having lowest total z-directional stiffness. However, the
stiffness dependence on the fiber inclination angle is highly nonlinear; hence it is
not obvious from a glance which is the weakest brick in this case. Also worth noting
that the entire progressive failure process may include many (tens or even hundreds)
consecutive failure events before the ultimate failure occurs. Further on, in the
case of a 3-chain model of Fig. 26.3c, the strains and stresses in each composite
chain and their individual bricks become strongly dependent on the “chain width”
(i.e., on the volume fraction) and the mechanical properties of the “intermediate”
26 Multi-scale Progressive Failure Modeling: From Nano-structured Carbon. . . 775
It is well known that carbon fibers are composite structures which are usually
considered homogeneous at the microscale but are highly inhomogeneous at the
nanoscale. In general terms, they are composed from 2-D layered graphite crystals
joined together edge to edge via mainly amorphous carbonized matter. Any specific
carbon fiber product inherits the principal chemical features of the raw materials
and traces the morphological transformation path experienced in the course of heat
treatment and mechanical drawing [10]. Specifically, as described in Guigon et al.
[11], carbon fibers in general are made of aromatic basic structural units (of about
1 nm size) forming layered wrinkled sheets. The sheets are crimped parallel to the
fiber axis, and the folds are entangled. Their thickness ranges from 6 to 24 nm
[12], and their length and width according to different sources can be from tens to
100–200 nm. The turbostratic structure of such layered aromatic sheets allows for
some moderate misorientation with respect to the fiber axis while it is in complete
rotational disorder. This results in a porous texture with pores elongated parallel
to the fiber axis. Strong lateral cohesive bonding (cross-linking atoms, tetrahedral
bonds, etc.) exists between those aromatic layers within the sheets, and, according
to Guigon et al. [11], about half of the fiber volume is amorphous in nature; the
typical size of amorphous regions is less than 100–200 nm.
Failure of a carbon fiber, according to Guigon et al. [11], is mainly due to the
propagation of transverse cracks between aromatic layers of basic structural units.
Such nanoscale cracks are impeded by lateral bonds, which are even stronger than
bonding, and the failure stress increases with growing density of these bonds.
As pointed out in Endo [12], the fibers with high strength and high elongation to
failure have a turbostratic carbon structure, which is different from the nearly 3-D
graphite structure in ultrahigh modulus carbon fibers. In summary, carbon fibers are
heterogeneous nanostructures composed of graphitized crystallites and carbonized
amorphous matrix. There are no obvious interfaces with sharp property variation.
The length-to-thickness aspect ratio of such crystallites is typically 10–30, so it
would be absolutely impossible to efficiently transfer internal stresses among them
if the matrix was something like a low-modulus polymer and if there were weak
interfaces separating “hard” and “soft” phases.
776 A.E. Bogdanovich
E1 DE2 D 200 GPa; E3 D1000 GPa; G12 D80 GPa; G13 DG23 D100 GPa;
31 D32 D0:25; 21 D0:2
(26.1)
Elastic properties of the isotropic matrix material were varied in a rather broad range
in order to determine which ones best match the “target” modulus of the fiber, i.e.,
240 GPa, in the longitudinal (z) direction. The following values were found most
satisfactory:
26 Multi-scale Progressive Failure Modeling: From Nano-structured Carbon. . . 777
E1 DE2 DE3 D45 GPa; G12 DG13 DG23 D18 GPa; 21 D31 D32 D0:25 (26.2)
The four inclined bricks in the upper part of Fig. 26.4a model got their orientation
by the following respective two consecutive rotations about axis y then axis x:
C15ı ; C15ı ; C15ı ; 15ı ; 15ı ; C15ı ; 15ı ; 15ı (26.3)
This results in 22ı angle between the local principal axis of symmetry 3 of the
hard material and the global fiber axis z. Note that accordingly with the literature,
typical misorientation angle for PAN-type carbon fibers has been measured in the
range of 20–25ı.
The structure of the lower part of the model in Fig. 26.4a repeats exactly the
described structure of the upper part. The middle part is all carbonized matrix.
Another two important input values are the volume fraction of hard bricks (taken
here 0.729) and the length-to-thickness aspect ratio of a hard brick (taken here 20).
With the use of above-listed input data, we have first determined, following
the methodology described in Bogdanovich [2], effective elastic properties of the
homogenized carbon fiber model. Of our particular interest was the longitudinal
modulus which was obtained as Ez D 240 GPa, exactly as targeted. Of course, this
is not a unique solution—the same Ez value can be obtained for numerous other
combinations of the elastic properties in Eq. (26.1) and Eq. (26.2), misorientation
angle Eq. (26.3), volume fraction of hard bricks, and aspect ratio of the hard brick.
The computed effective transverse elastic and shear moduli would be different
though for any other set of input data, and matching them closely to available
experimental data has been used as additional model calibration criteria; we skip
description of that effort here. Anyway, our main goal here was to validate that the
proposed model with some reasonable set of input data can give good predictions
of the longitudinal ultimate stresses and strains in a typical PAN carbon fiber. From
this perspective, successfully matching a 240 GPa longitudinal fiber modulus should
be viewed as the first model calibration step.
Turning to the progressive failure modeling and strength predictions of a
PAN-based carbon fiber, we assumed the following ultimate strains for the hard
graphitized orthotropic brick in its principal axes of symmetry (superscripts “t,”
“c,” and “s” indicate tension, compression, and shear):
"t1 D"t2 D0:01; "t3 D0:015; "c1 D"c2 D0:02; "c3 D0:025; "s12 D"s13 D"s23 D0:03
(26.4)
"t1 D "t2 D "t3 D 0:02; "c1 D "c2 D "c3 D 0:03; "s12 D "s13 D "s23 D 0:03 (26.5)
The stiffness reduction factors for both hard and soft materials were arbitrarily taken
0.1 in tension while 0.5 in compression and shear. With this set of input data, the
progressive failure simulations have been performed under uniformly distributed
778 A.E. Bogdanovich
displacement applied to both ends of the Fig. 26.4a model in the z direction. They
showed that the failure process is initiated in the middle part of the structure, where
the “weak” carbonized matrix is located, as the formation of cracks perpendicular
to the loading direction; this is what one would expect intuitively. The failure then
propagates across the structure until all matrix bricks fail in the middle section;
that constitutes ultimate failure of the fiber model. The z stress, averaged over any
cross section (ideally it should be constant along z coordinate), remains very low;
its maximum reaches only about 10 % of the experimental tensile strength value
(the target was 4200–4400 MPa) by the end of progressive failure process, and there
was no way to significantly increase it with this model. The conclusion was that
the model of Fig. 26.4a is too simplistic for adequate strength modeling of carbon
fibers.
Further model development went in the direction of staggering hard graphitized
bricks as to avoid the situation where weak soft material occupies entire central
region. One example of a staggered brick model is shown in Fig. 26.4b. In this
particular case, there is one segment of a hard material separating matrix bricks
in the longitudinal direction; hence, no more weak cross sections of the soft
matrix are present. Progressive failure analysis performed with the same input
data Eqs. (26.1)–(26.5) showed remarkable difference between the results provided
by Fig. 26.4b model vs. Fig. 26.4a model. The dominating failure mode in the
case of Fig. 26.4b model is a shear failure along longitudinal planes x–z and y–z,
and the classical shear-lag-type stress transfer mechanism is engaged in this case.
Accordingly, the failure process became highly dependent on the length of that part
of hard brick which is “inserted” between two nearest soft bricks. That parameter
is, of course, variable in a staggered brick model and can be used for the further
model calibration. With increase of that length, the ultimate stress z also increases,
but at some point certain asymptotic level of the failure stress is reached; this result
is sensible from the shear-lag theory viewpoint. Additionally, the staggered brick
model showed much higher sensitivity to the assumed elastic properties of the
matrix material (unfortunately, its properties is hard to define with certainty).
With the input data listed in Eqs. (26.1)–(26.5), the progressive failure sim-
ulations using Fig. 26.4b model showed that ultimate longitudinal fiber stress
(determined as the threshold for the hard brick tensile failure initiation) reached
the values of 4200–4300 MPa, thus getting into experimental data range for
PAN-type carbon fibers such as Toho Tenax HTA or HTS. The modeling results
showed sensitive to several input parameters: to the ultimate longitudinal tensile
strain of the hard brick material in Eq. (26.4), to the elastic and shear moduli of
the soft brick in Eq. (26.2), to the ultimate tensile and shear strains of the soft
brick in Eq. (26.5), to the hard material misorientation angle in Eq. (26.3), and
to the hard material volume fraction. Varying each of these parameters makes
certain effect (though not equal) on the longitudinal modulus and tensile strength
output data. For example, increasing the misorientation angle lowers the predicted
fiber modulus; increasing ultimate longitudinal strain of hard bricks moves the
fiber strength up; a decrease of hard brick volume fraction reduces fiber modulus
26 Multi-scale Progressive Failure Modeling: From Nano-structured Carbon. . . 779
and, in the studied range 50–70 %, also reduces tensile strength of the fiber.
A detailed parametric study (not included here) provided useful information on how
the 3DMC model nanoscale structural parameters influence effective elastic and
strength characteristics of the simulated carbon fibers.
Fig. 26.5 Three-chain models of a unidirectional composite with one brick (a), two bricks (c),
and three bricks (d) in longitudinal direction; five-chain model with one brick in longitudinal
direction (b)
780 A.E. Bogdanovich
E1 D E2 D E3 D 3:2 GPa; G12 D G13 D G23 D 1:01 GPa; 21 D 31 D 32 D 0:35
(26.7)
"t1 D "t2 D 0:012; "t3 D 0:018; "c1 D "c2 D 0:025; "c3 D 0:03; "s12 D "s13 D "s23 D 0:04
(26.8)
"t1 D"t2 D"t3 D0:025; "c1 D"c2 D"c3 D0:04; "s12 D"s13 D"s23 D0:05 (26.9)
The stiffness reduction factors (see Bogdanovich [2, 15] for their definition) are
taken 0.1 for the fiber and 0.5 for the matrix materials.
The volumetrically averaged initial failure stress and strain results (i.e., those
obtained in Step 1 of progressive failure analysis using 3DMC model of Fig. 26.5a)
are 2900 MPa and 1.8 %, respectively. Note that both fibers are in identical stress-
strain state in this case; the volumetrically averaged longitudinal stress in both fiber
chains is 4320 MPa. The volumetrically averaged longitudinal strains in both fiber
chains and in the matrix chain are all identical, equal to 1.8 %. Due to the detected
initial failure mode is longitudinal fiber breakage, in this model the first failure also
means ultimate failure of the composite because after the first failure the load it
can carry drops instantaneously by approximately ten times, in accordance with
the assumed stiffness reduction factors. The progressive failure analysis run was
terminated accordingly after Step 1. The ultimate failure stress provided by this
trivial analysis case is significantly higher than the experimental data for this kind
of composite (typically in the range of 2300–2500 MPa).
The question now is how to sophisticate this basic model in order to allow for
some reduction of the peak stress but, at the same time, to prolong progressive
failure process? In other words, how to add some damage tolerance to the composite
without involving usual statistical considerations such as fiber strength distribution
and/or random fiber waviness? One reasonable approach is to increase the number
of parallel chains in the model from 3 to 5, 7, etc. The 5-chain model is shown in
Fig. 26.5b (the width of the matrix chain has been reduced to keep the same fiber
volume fraction as in Fig. 26.5a). One could expect the results from this model to
be different, because the inner fiber chain is under different boundary conditions
than the outer fiber chains (no lateral constraint is applied to the latter ones) and,
accordingly, the stress-strain states of the inner and outer chains are not identical
anymore. However, the analysis performed for the model of Fig. 26.5b showed very
26 Multi-scale Progressive Failure Modeling: From Nano-structured Carbon. . . 781
close to the results obtained with the model of Fig. 26.5a. All three fibers failed
simultaneously at the same stress level of 4320 MPa and strain level of 1.8 %, and
the computed ultimate stress for the composite appeared to be the same 2900 MPa.
Therefore, adding chains connected in parallel does not solve the issue.
Another approach is to split each fiber and matrix chain of Fig. 26.5a model into
two chains as shown in Fig 26.5c, into three chains as shown in Fig. 26.5d, etc.
We can expect again that some differences in the boundary conditions between the
top and bottom bricks on one side and the middle brick on the other may result in
different progressive failure processes and give significant change to the ultimate
failure characteristics. However, the performed analysis showed that all four fiber
bricks in the model of Fig. 26.5c and all six fiber bricks in the model of Fig. 26.5d
are exactly in the same stress-strain state and all of them fail simultaneously when
the strain level of 1.8 % and stress level of 2900 MPa are reached in the composite.
The conclusion from the above-described preliminary numerical experiments is
that, independently of the number of identical parallel fiber chains and the number of
identical material bricks within each chain, all fiber bricks fail simultaneously when
longitudinal strain and stress reach certain characteristic levels. Of course, one could
anticipate this conclusion a priori, but for the model validation purpose, it was useful
to verify that it holds. Involving more realistic progressive failure processes can be
attempted in three ways: (1) assigning some ultimate strain variation to the fiber
bricks, (2) assigning some stiffness variation to the fiber bricks, and/or (3) using
“short” (fragmented) fiber models.
Consider first the effect of ultimate fiber strain variation. The simplest model
of this kind is shown in Fig. 26.6a; this is a direct extension of Fig. 26.5a model.
We leave all ultimate strains for the fiber on the left as in Eq. (26.8) but reduce "t3
for the fiber on the right from 0.018 to 0.017 (the latter fiber is painted differently
in Fig. 26.6a because its material is now different). All other input data remain
the same. Results of the analysis showed that, naturally, the right fiber fails first,
when strain in the composite reaches 1.7 % and stress reaches 2740 MPa (compare
to 2900 MPa obtained with Fig. 26.5a model). After that Step 1 of the progressive
failure analysis, the applied displacement has to be increased as to reach 1.8 % strain
in the composite, and the left fiber fails under much lower longitudinal stress of
1605 MPa. After that Step 2, the stress carried by the model drops dramatically, and
the analysis run is terminated.
Therefore, if the number of parallel chains increases and there is some variation
of the longitudinal ultimate strain among the fiber chains, the chain having lowest
ultimate strain will fail first, followed by the second lowest ultimate strain chain,
etc., until all of the chains get broken. The initial failure stress decreases with
decrease of the weakest chain ultimate strain. However, if there is sufficiently large
number of chains and if some of them are significantly stronger than the weakest
one, it is possible that the peak ultimate stress for the composite is reached in
some intermediate progressive failure step, and the entire failure process is less
“catastrophic.”
Next, the models of Fig. 26.6b, c show two examples of chains having ultimate
strain variations along their length. Results of the analysis performed for the
782 A.E. Bogdanovich
Fig. 26.6 Three-chain models of a unidirectional composite with one (a), two (b), and three
bricks (c) in the longitudinal direction; fiber bricks with ultimate strain of 1.8 % and 1.7 % are
distinguished by color
Fig. 26.6b model showed that the fiber bricks having 1.7 % ultimate strain failed
first, when composite strain and stress reached 1.7 % and 2740 MPa values
(analogously to the Fig. 26.6a model). However, the next failure steps got less
trivial: the matrix bricks (having 2.5 % ultimate strain) got overstressed and failed
before the fiber bricks having 1.8 % ultimate strain. Also interestingly, considerable
transverse shear strain "xz and stress xz were recognized prior to the failure of those
stronger fiber bricks. The stress transfer between the bricks in Fig. 26.6b model was
more complex than in case of Fig. 26.6a model, and it got even more complex in
case of Fig. 26.6c model. These results show that increasing number of bricks with
distinct ultimate failure strains in the model makes progressive failure process more
prolonged, and the internal stress redistribution after each consecutive failure event
is getting less predictable intuitively.
Next consider the effect of stiffness variation from one fiber chain to the other
and within each chain. Of a particular interest are the cases of the stiffness variations
caused by a global fiber misalignment and by a local fiber waviness. The first
example, illustrated in Fig. 26.7a, shows a three-chain model where fiber in the
left chain is inclined at some angle Cˇ with respect to axis z (the fiber is rotated in
counterclockwise direction about axis y), while fiber in the right chain is inclined
at ˇ angle to axis z (the fiber is rotated in clockwise direction about axis y).
To mark this distinction, the respective fiber chains are painted differently. This
model is “balanced” but nonsymmetric; it can be viewed as one specific case of
general three-chain model shown in Fig. 26.1c. Further, Fig. 26.7b shows a more
general, five-chain model where three chains have the same fiber inclined at angles
Cˇ, ˇ, and Cˇ (from left to right) and are connected by the matrix chains. The
fiber and matrix chain widths were adjusted here to keep the same fiber volume
26 Multi-scale Progressive Failure Modeling: From Nano-structured Carbon. . . 783
Fig. 26.7 Three-chain (a, c, d) and five-chain (b) models of a unidirectional composite with one
brick (a, b), four bricks (c), and five bricks (d) in the longitudinal direction; red bricks are aligned
with z-axis; pink bricks are inclined at Cˇ angle; brown bricks are inclined at ˇ angle
fraction of 0.67 and the same fiber length-to-width aspect ratio 30 as before. This
model is symmetric (hence, the global tension-bending coupling is not present)
but unbalanced. Fig. 26.7c shows seven-chain model, which is both symmetric and
balanced.
Next, Fig. 26.7d, e show more general three-chain models, where there are
distinct material segments along each chain serving the purpose of simulating local
fiber waviness. In case of Fig. 26.7d, the top and bottom segments are aligned with
axis z, while the middle two segments are inclined at angles Cˇ (lower segment)
and ˇ (upper segment) in the left chain and at angles ˇ (lower segment) and
Cˇ (upper segment) in the right chain. In case of Fig. 26.7e, analogous inclined
segments are the top two in the left chain and the bottom two in the right chain.
These models enable to study a variety of important fiber misalignment and local
waviness effects on the progressive failure process and strength of unidirectional
composites. In the performed simulations, all material input data are kept as before,
defined by Eqs. (26.6), (26.7), (26.8), and (26.9).
Principal results of the first four steps of progressive failure simulations per-
formed with the model of Fig. 26.7a for different values of misalignment angle
ˇ are summarized in Table 26.1 (note that the case of ˇ D 0 corresponds to the
earlier studied model of Fig. 26.5a). There are four principal failure modes revealed:
fiber failure under longitudinal tension (further called Mode 1), matrix failure under
tension (Mode 2), fiber failure under longitudinal shear (Mode 3), and fiber failure
under transverse tension (Mode 4). The first interesting effect to be noted is the
change in the sequence of these failure modes when the misalignment angle varies
from 0ı to 45ı . For the cases of 0ı and 5ı , the initial failure is due to Mode 1
followed by Mode 2. In fact, the initial failure in these two cases also signifies the
ultimate failure, because the stress z in which the composite is able to bear drops by
almost seven times after initial failure occurrence. The situation is different for 10ı ,
15ı , and 20ı misalignment angle cases—initial failure in all of them is due to Mode
3 (which can be also interpreted as interfacial failure between fiber and matrix,
784 A.E. Bogdanovich
Table 26.1 Computed progressive failure characteristics for the Fig. 26.7a model with fiber
misalignment angle varying between 0ı and 45ı
Step 1 Step 2 Step 3 Step 4
"z , z, "z , z, "z , z, "z , z,
Angle ˇ Averaging % MPa % MPa % MPa % MPa
0ı /0ı (i) 1.80(1) 2900 2.50(2) 428
(ii) 1.80 4320 2.50 600
C5ı /5ı (i) 1.90(1) 2796 2.50(2) 428
(ii) 1.90 4157 2.50 601
C10ı /10ı (i) 2.02(3) 2351 2.06(2) 1824 2.41(1) 2121
(ii) 2.02 3490 2.06 2703 2.41 3162
C15ı /15ı (i) 1.81(3) 1592 2.10(2) 1218 3.39(1) 1948
(ii) 1.81 2356 2.10 1792 3.39 2895
C20ı /20ı (i) 1.85(3) 1231 2.16(2) 878 4.83(1) 1933
(ii) 1.85 1816 2.16 1281 4.83 2861
C25ı /25ı (i) 1.99(3) 1037 2.33(2) 682 4.28(4) 1283 9.58(1) 2214
(ii) 1.99 1520 2.33 987 4.28 1890 9.58 3244
C30ı /30ı (i) 2.15(3) 921 2.30(2) 566 3.47(4) 836 15.8(1) 2601
(ii) 2.15 1347 2.30 812 3.47 1226 15.8 3744
C35ı /35ı (i) 2.34(2) 855 2.35(3) 844 2.98(4) 611 26.0(1) 3177
(ii) 2.34 1245 2.35 1247 2.98 892 26.0 4556
C40ı /40ı (i) 2.39(2) 771 2.55(3) 812 2.63(4) 482 42.7(1) 4029
(ii) 2.38 1118 2.55 1197 2.63 702 42.7 5700
C45ı /45ı (i) 2.33(4) 689 2.32(2) 223 2.37(3) 229 71.0(1) 5346
(ii) 2.33 995 2.32 297 2.36 305 71.0 7447
(a) Averaging: (i) over entire three-chain volume; (ii) over fiber chain volume only
(b) Failure modes: (1) fiber in longitudinal tension "t3 ; (2) matrix in tension "t3 ; (3)
fiber in
longitudinal shear "s13 ; (4) fiber in transverse tension "t1
though such a failure mode is not explicitly incorporated in the Fig. 26.7a model). In
the second failure step, Mode 2 is realized. And only in the third step Mode 1 takes
place. The z drop between the first and second failure steps in these cases is only
about 40–50 %, much smaller than for the 0ı and 5ı misalignment cases. Then, z
increases (note that the increase is nearly proportional to the failure strain increase),
but after Step 3, it drops dramatically due to Mode 1 is realized. The failure mode
sequence changes again for 25ı and 30ı misalignment angles. Although the initial
failure is again due to Mode 3, the next is Mode 2, and then Mode 4 is realized (this
can also be interpreted as interfacial failure between fiber and matrix). Notably,
this failure mode precedes Mode 1 which is realized only in Step 4. Further on, in
cases of 35ı and 40ı , the situation is different again—the first failure step is due
to Mode 2. This is followed by Mode 3, and then Mode 4 is realized in the third
step. And, in the case of 45ı angle, failure starts with Mode 4, followed by Mode
2, and finally Mode 3 takes place. Notably, for misalignment angles 20ı , 25ı , 30ı ,
35ı , 40ı , and 45ı , Mode 1 (which is, of course, the most critical failure mode of
a unidirectional composite) is realized at respectively increasing levels of "z , all of
them above 4.8 %, while the other three principal failure modes take place at "z
26 Multi-scale Progressive Failure Modeling: From Nano-structured Carbon. . . 785
levels lower than 4.3 %. This indicates that tensile fiber failure will likely not be
observed for these misalignment angles prior to a total composite failure. Moreover,
accordingly with Table 26.1 data, the chance to see a tensile fiber failure would
be small even for the misalignment angles of 10ı and 15ı , because respective "z
levels for Mode 1 are significantly higher than those for Modes 3 and 2. Of course,
these results are illustrative and the input data change would affect them to some
extent. Yet, the revealed effect of significant change in the sequence of realized
failure modes with the misalignment angle variation may be of a general nature,
particularly when the angle increases in the range between 5ı and 10ı .
More detailed progressive failure simulations for the misalignment angles 6ı , 7ı ,
8 , and 9ı are summarized in Table 26.2. It is seen that at 6ı , 7ı , and 8ı angles, the
ı
initial failure mode is still tensile fiber failure. Interestingly, at 8ı angle the second
failure occurs due to Mode 3 while it was due to Mode 2 for 7ı angle. And at
9ı angle, the fundamental alteration of the initial failure mode occurs—instead of
Mode 1 it becomes Mode 3. This is followed by Mode 2 and only after that Mode 3
is realized. As Table 26.1 shows, this sequence of failure modes holds for the angles
10ı , 15ı , and 20ı . Also, Mode 1 and Mode 2 remain the first two failure modes for
the angles 25ı and 30ı .
Most important conclusion made from these theoretical simulations is that
efficient loading of carbon fibers within their unidirectional composite is possible
in some range of small fiber misalignment angles; with the input data used here, the
range extends to 8ı . At the larger angles, the load transferred to the fibers gradually
decreases, and the fiber breakage becomes less and less likely. The averaged stress
value in fibers z at initial failure has its maximum of 4320 MPa for the perfectly
aligned fiber case; it retains rather high values for the angles up to 8ı and drops
sharply when initial failure mode changes from Mode 1 to Mode 3. Analogous trend
is observed for the averaged stress values in composite z .
Conventional 2-D woven fabric composites and, particularly, plain weave com-
posites have been the subjects of extensive theoretical studies in the past three
decades. A variety of elementary analytical models and, later, various 2-D and 3-D
finite element analyses have been reported. Our goal here is to illustrate how some
principal aspects of progressive failure analysis and strength prediction of this textile
composite type can be captured with the use of 3DMC approach. To accomplish this
goal, the mosaic chain models of a representative plain weave composite unit cell
have been generated, 3-D strain-stress states computed for the quasi-static in-plane
tensile loading cases, progressive failure analysis performed, and the ultimate failure
stress/strain values predicted.
26 Multi-scale Progressive Failure Modeling: From Nano-structured Carbon. . . 787
Fig. 26.8 Three representative 3DMC models of a plain weave composite unit cell with increasing
finesse of the construction detail
the same fiber volume fraction as the red and blue ones but is rotated at some angle
about either coordinate axis y (e.g., the warp-oriented brick) or about axis x (e.g., the
weft-oriented brick). The other brick kind has twice smaller fiber volume fraction;
again, one such brick is rotated at some angle about axis y while the other is rotated
about axis x. The purpose of these bricks is to “blend” mechanical properties of
the unidirectional warp and weft composites with the properties of pure matrix in
the proportion of 50/50 percent. The inclination angle has to be determined from
specific unit cell geometry in order to reflect the tow undulation as close as possible
in the framework of this 3DMC model.
Further model refinement, shown in Fig. 26.8c, illustrates the possibility of
adding bricks in all three dimensions x, y, and z. In this case each warp and weft
composite tow is modeled by three through-thickness bricks; thus more bricks are
incorporated in the undulated tow regions. In this model, one can use more than two
types of inclined composite bricks and, what is of a particular importance, one can
represent the undulated tow trajectory not as a straight line (like it was in Fig. 26.8b
model) but as a three-segment broken line. Using the same principle, further model
refinement can be achieved, if necessary, for a higher accuracy of the analysis.
Numerical simulations of progressive failure processes in carbon-epoxy plain
weave composites under in-plane tensile loading were performed with the use
of Fig. 26.8c model, assuming various tow path geometries; the tow inclination
angle was varied in the range from 1.4ı to the 14ı . Progressive failure simula-
tions typically included 150–200 consecutive failure events with many different
failure modes realized. The results showed mechanistically reasonable trends but
significantly underestimated (by 30–40 %) ultimate failure stress values both in
the warp and weft directions in comparison with typical experimental data. Of a
particular interest was the finding that failure has always started with transverse
cracking in the weft-directional composite bricks under warp-directional loading
and with warp-directional composite bricks under weft-directional loading. This
initial failure mode was most commonly reported in experimental studies. After
many intermediate failure events, the critical failure stage showed transverse
splitting of undulated tows in the transitional region mixed with the matrix cracking.
That part of progressive failure process is mostly controlled by the tow inclination
angle, and it typically results in the formation of distinct “weak cross section” that
grows from one side edge of the unit cell to the other. Local strains in that region
exceed the overall strains averaged over entire unit cell volume by many times.
Such high local strains in the regions of undulated tows (notably they increase with
increasing inclination angle) are reached prior to the longitudinal failure is initiated
in composite tows oriented in the respective loading direction and facilitate the
latter failure mode. This result explains the well-known experimentally observed
phenomena that textile composites with interlaced tows usually show much lower
in-plane tensile strength than those made with non-crimp reinforcements; see Brandt
et al. (1996), for example.
The other important result revealed in this study showed that considering only
one plain weave composite unit cell with idealized fiber architecture is insufficient
for accurate ultimate strength predictions. The reason is similar to the cases of the
carbon fiber modeling and the unidirectional composite modeling cases that have
26 Multi-scale Progressive Failure Modeling: From Nano-structured Carbon. . . 789
been discussed earlier in the chapter. Indeed, like in the cases of Fig. 26.4a model for
the carbon fiber and Fig. 26.5c, d and other models for the unidirectional composite,
the plain weave model of the type shown in Fig. 26.8c allows the failure process to
be fully localized within transitional regions of the undulated tows. As the analysis
results showed, as soon as the failure starts there, it remains localized there for the
rest of the failure process forming progressively weaker cross section of the unit
cell. And this leads, in turn, to significantly underestimated ultimate failure stress
values.
As is well known, predicting properties of a material by considering a single-
unit cell model assume that the material is perfectly periodic, meaning that all unit
cells have to be perfectly aligned and all of them should fail simultaneously and in
exactly the same fashion. When producing a plain weave fabric preform, handling it
and then manufacturing composite, the preform gets inevitably distorted. The rows
and columns of assumingly perfect and identical unit cells may shift with respect to
each other and get misaligned. Hence, after the microcracks initiate in different loci
of the composite specimen, they would not develop in a self-similar manner into
the macrocrack contained within some preferred cross-sectional region—contrary
to what is predicted by the single-unit cell model of Fig. 26.8c. The damage
development in plain weave composites is always dispersed over a large number
of unit cells, and the microcrack coalescence into the final catastrophic macrocrack
always shows very tortuous path. This dispersed failure phenomenon obviously
results in a higher composite’s strength, and in order to capture it in the model,
several staggered unit cells have to be considered. Respective 3DMC models have
been constructed (by generalizing the Fig. 26.8c model), and their implementation
fully confirmed the above-expressed expectation. The predicted ultimate stresses
got within 10 % discrepancy of the experimental data.
The final examples presented here illustrate application of the developed 3DMC
methodology to non-crimp 3-D orthogonal weave composites which have been
under extensive theoretical and experimental studies in the past two decades; see
Brandt et al. [16], Mohamed et al. [17], Bogdanovich and Mohamed [15], Karahan
et al. [1], and Bogdanovich et al. [18] as representative publications on this subject.
Three different unit cell models of one specific composite are shown in Fig. 26.9.
They include four warp layers (red) and five fill layers (blue); matrix is painted
yellow. Notably, the warp and fill tows are not interlaced; they follow straight
trajectories, and, as the experimental study of Karahan et al. [1] showed, their
incidental waviness is very small. The slight in-plane tow waviness can be accounted
in 3DMC models similarly to the cases of unidirectional composite and plain weave
composite considered earlier in the chapter. However, for simplicity it is assumed in
790 A.E. Bogdanovich
Fig. 26.9 Representative 3DMC models of non-crimp 3-D orthogonal weave composite unit cell
the models of Fig. 26.9 that all warp and fill composite tows are straight, and their
elastic and strength properties do not vary along the length. The model of Fig. 26.9a
is most basic, because it neglects “Z-crowns” and respective matrix pockets in the
outermost layers; it also assumes that the Z-tow path is straight and vertical; all
the above are strong idealizations of the real composite construction. The model
of Fig. 26.9b takes into account Z-crowns and respective matrix pockets but still
assumes that the Z-tows are straight and vertical. In order to make a smoother
transition between the vertical Z-tow segment (green) and the horizontal segment
(brown), two “transitional” material bricks are incorporated here. The fibers in those
bricks are inclined at 45ı angle (by rotation about y axis). Finally, the model shown
in Fig. 26.9c adds the Z-tow inclination and curvature observed in this kind of 3-D
woven composites. The latter model is most realistic among the class of unit cell
models of these composite materials, and its different versions have been used for
numerous progressive failure simulations.
The performed simulations of progressive failure processes for some carbon-
epoxy composites with the use of Fig. 26.9c model typically showed between 30
and 100 individual failure events prior to the ultimate failure. Those simulations
had elucidated on the failure onset and progression, dominating failure modes, and
enabled to predict ultimate failure loads. The obtained theoretical results were found
26 Multi-scale Progressive Failure Modeling: From Nano-structured Carbon. . . 791
in a close agreement with experimental strength data and with general trends of
progressive failure development reported in Bogdanovich et al. [18]. Particularly,
those theoretical results revealed a clear distinction between progressive failure
processes in the non-crimp 3-D woven composite unit cell model of Fig. 26.9c
on one side and in the plain weave composite model of Fig. 26.8c on the other.
As pointed out earlier in the chapter, the idealized plain weave composite unit cell
experiences disperse failure within transitional region of undulated in-plane tows,
with internal tow splitting being the principal predecessor of the ultimate failure. As
described in the previous section, the entire progressive failure process is very long,
gradual and incorporates numerous failure modes. Contrary to that, the non-crimp
3-D woven composite experiences first transverse tow splitting, matrix cracking,
and then sudden tensile breakage of the tows oriented in the loading direction. The
latter failure occurrences lead to a quick catastrophic failure of the composite. The
predicted ultimate failure stresses for the studied non-crimp 3-D woven composites
were found substantially higher than respective values for comparable plain weave
composites. This result is in agreement with known experimental studies. For
example, Brandt et al. [16] showed that in-plane tensile strengths of non-crimp
3-D woven carbon-epoxy composites are 1.2–1.8 times higher than their plain weave
counterparts.
26.10 Conclusions
experimental data. We have learned from the case of considered plain weave
composite (characteristic with undulated tows caused by their interlacing) that
using a single idealized unit cell yields significantly underestimated in-plane tensile
strength results. In order to get close agreement with experimental data, it was
necessary to consider several staggered plain weave unit cells. However, this appears
not to be the case for non-crimp 3-D orthogonal weave composite (characteristic
with practically straight and well-aligned in-plane tows). Even a single-unit cell
model gave in this case excellent agreement with experimental strength data. This is
attributed to the absence of predetermined “weak cross section” within the unit cell
that would allow for a premature critical damage accumulation, propagation, and
ultimate failure. Contrary to the above, such weak cross section is present in the case
of a single idealized plain weave unit cell model. This emphasizes our major lesson
learned—adequate failure modeling and strength predictions of complex composite
materials require close attention to the details of reinforcement architecture, such as
the presence of fiber/tow misalignment, undulations, and waviness.
References
13. D.L. Chung, Carbon Fiber Composites (Butterworth-Heinemann, Newton, MA, 1994)
14. J.G. Lavin, Fracture of carbon fibers, in Fiber Fracture, ed. by M. Elices, J. Llorca (Elsevier,
2002), pp. 157–179
15. A.E. Bogdanovich, M.H. Mohamed, Three-dimensional reinforcements for composites.
SAMPE J. 45(6), 8–28 (2009)
16. J. Brandt, K. Drechsler, F.-J. Arendts, Mechanical performance of composites based on various
three-dimensional woven-fibre preforms. Compos. Sci. Technol. 56, 381–386 (1996)
17. M.H. Mohamed, A.E. Bogdanovich, L.C. Dickinson, J.N. Singletary, R.B. Lienhart, A new
generation of 3D woven fabric preforms an composites. SAMPE J. 37(3), 8–17 (2001)
18. A.E. Bogdanovich, M. Karahan, S.V. Lomov, I. Verpoest, Quasi-static tensile behavior and
damage of carbon/epoxy composite reinforced with 3D non-crimp orthogonal woven fabric.
Mech. Mater. 62, 14–31 (2013)
Chapter 27
Textile Structural Composites: From 3-D to 1-D
Fiber Architecture
27.1 Introduction
With its outstanding specific strength and modulus, over half a century of rapid
development, carbon fiber has evolved from a laboratory curiosity to a material
of choice for the structural reinforcement of aerospace structures, aircraft, high-
performance sporting goods such as formula racing cars, and giant windmill blades.
There is a global renewal of interest in the manufacturing technology for
lightweight materials, with a major focus on automotive composites. This is
evident in the national initiatives embarked in recent years, notably the Institute for
Advanced Composites Manufacturing Innovation (IACMI) (http://iacmi.org/) led by
the University of Tennessee in the USA, the Institute for Carbon Composites (LCC)
(http://www.lcc.mw.tum.de/en/home/) in Germany, and the National Composite
Center (http://ncc.engg.nagoya-u.ac.jp/) led by Nagoya University in Japan. It is
of interest to note that carbon fibers and textile structural composites (TSCs) are
the key focuses in these manufacturing initiatives. TSCs are composite materials
reinforced by textile structures for primary structural applications. Although textile
structures have long been used for composite reinforcement, the serious use of
textiles for structural composites did not occur until the entering of the Space Age
with the development of carbon fibers and multiaxial 3-D textile preforms. The
need for affordable composites that have significantly improved through-thickness
strength and damage tolerance intensified the development of a wide variety of
textile preforms, which established the foundation for the use of TSCs for aircraft
primary structures. During the same period, we also witnessed the development of
a large family of sporting carbon composites such as mountain bikes and booms
for sails. Future growth and widespread usage of carbon fibers in automobiles and
consumer goods depend on the availability of low-cost carbon fibers and affordable
composite manufacturing processes. As we enter the nano-era in the 2000s, the
need for super strong and multifunctional carbon fibers led us to the development
of carbon nanotube-reinforced 1-D composite carbon fibers. Through nanofiber
architecture tailoring, these 1-D composite fibers enable the development of a new
generation of multifunctional smart composites. At the occasion of commemorating
the 50-year anniversary of the development of carbon fiber-reinforced composites,
it would be timely to reflect on the role TSC plays in the structure, processing, and
properties of carbon fiber composites. We will begin by reviewing three areas of
structural applications including aerospace, aircraft, and automotive wherein textile
structures have been used extensively as reinforcements. Chronologically developed
over three decades, these three families of applications command different perfor-
mance/cost requirements. Invariably a majority of these programs are the result
of industry/academia collaborations supported financially by government agencies.
Using three specific products that we have participated in developing as examples,
we will examine the lessons learned from these TSC development programs. We
will specifically examine the role textile fiber architecture plays in the dynamics
of structure/processing/property interaction. The analytical framework that governs
the engineering design for manufacturing of the TSC is presented through the Fabric
Geometry Model (FGM) quantifying the translation of fiber material properties to
composite structures via specific textile fiber architectures. Building on the FGM
platform, we look into the potential of new carbon fibers such as low-cost carbon
nanofibers and multifunctional and strong carbon nanofibers through 1-D nanofiber
architecture design and continuous progress toward a new generation of textile
preforming technology such as hexagonal 3-D braiding to face the challenges we
will encounter in the next 50 years of carbon fiber composite development.
materials organized by the Ko Lab, Benson Dexter and John Buckley of NASA
sponsored a series of workshops on 3-D TSCs providing a forum for academia–
government and industry interaction and contributed much to the establishment of
the knowledge base that we have today.
During this period there were significant activities on high-temperature engine
components such as turbine rotors, blades, and rocket nozzles using high-
temperature ceramic matrix composites and high-temperature polymer matrix
composites (e.g., PRD). For example, a composite compression turbine rotor was
developed in collaboration with Teledyne by integrated design and combining
weaving, braiding, and winding technology using a hybrid of glass and carbon
fibers in the Ko Lab in 1987. A picture and CAD drawing of the rotor are shown in
Fig. 27.1.
In this work the development of the compressor rotor was developed according to
an integrated design methodology for three-dimensional fabric composite structural
components. In designing these fabric composite structures, the design criteria were
first established based on the mechanical and geometrical requirements. Appropriate
material systems and fabrication techniques were then selected accordingly. Fiber
architectures resulting from these material systems and fabrication techniques were
quantified and modeled to yield the mechanical properties of fabric preforms.
These properties were subsequently input into finite element models, and structural
analysis of the composite component was performed. The analysis results were
then used to modify the fabrication technique, and eventually an optimal fiber
architecture with a proven structural behavior was identified. As a demonstrating
example of this methodology, the process of design and analysis of the 3-D fabric
composite rotor was presented in the paper by Tan et al. [3].
798 F.K. Ko and L.Y. Wan
Building on the confidence developed during the past decade mostly on composite
structural components, the aircraft industry was ready to develop the technology for
commercial composite aircraft structures with the support of two NASA contracts—
Advanced Composite Technology (ACT) initiative and the Advanced Technology
Composite Aircraft Structures (ATCAS). An additional contract was initiated to
verify this technology at a large scale. The goal of the ACT initiative was to
develop composite primary structure for commercial transport aircraft with 20–25 %
less cost and 30–50 % less weight than equivalent metallic structures. It is worth
noting that these programs demonstrated another example of government, industry,
and academia collaboration—twenty-five faculty members (including the author
Frank K.) from 12 North American universities were part of the design-built team.
For technology verification Boeing selected a section of the fuselage, which has
all the representative structural features of an aircraft (door and window cutouts,
stiffeners, etc., as shown in Fig. 27.2). Material selection and composite structure
design and fabrication were carried out starting from coupon level to sub-element
and structure level. For the stiffeners, carbon fiber braided structures were designed
and fabricated into composite parts using resin transfer process. It took over 10 years
before these structures were certified to be flight worthy!
In July 8, 2007, the first 787 Dreamliner was unveiled. As a pioneering airliner
with the use of composite materials as the primary material in the construction of
its airframe, Boeing 787 was designed to consist of 50 % by weight and 80 % by
volume carbon fiber-reinforced plastic (CFRP). As reported by Boeing, the 787 is
approximately 20 % more fuel efficient than the 767 with approximately 40 % of the
efficiency gains from the engines plus gains from aerodynamic improvements due
to the increased use of lighter-weight composite materials and advanced systems.
Boeing 787 is the first production airliner with the fuselage assembled with one-
piece composite barrel sections instead of the multiple aluminum sheets and some
50,000 fasteners used on existing aircraft. Composites are also used on wings, tail,
doors, and interior.
In 1990s, the through-thickness stitching of graphite preforms was selected to
fabricate a12.8 m-long full-scale composite wing box. Braided fuselage frames and
window-belt reinforcements, woven/stitched lower fuselage side panels, stitched
multiaxial warp knit wing skins, and braided wing stiffeners were also fabricated.
2-D and 3-D braids were used to create stiffeners, frames, and beams with complex
cross sections. Figure 27.3 shows the all-composite aircraft wing by stitching of
braided stiffeners on multiaxial warp knit skin structure which fully demonstrated
textile preforming as a pathway to low-cost structural composites.
This cost-effectiveness of the TSCs was demonstrated by Ray Palmer of McDon-
nell Douglas in a trade study of a stiffened panel, as shown in Fig. 27.4. It was shown
that the use of multiaxial warp-knitted reinforced composite resulted in lower cost
($30 vs. $32Al) at 25 % weight reduction.
27 Textile Structural Composites: From 3-D to 1-D Fiber Architecture 799
Fig. 27.2 (a) Boeing baseline vehicle and study section, (b) baseline crown panel configuration,
(c) fuselage quadrants, and (d) images of the textile composite fuselage [4]
A major driving force for the development of TSCs is the automobile industry.
In addition to earlier technology demonstration projects such as Ford Motor’s
million-dollar-all-composite car, we have witnessed at least three generations of
national (government/industry) programs aiming at advancing the technology of
composites for automobiles starting from the Automotive Composite Consortium
(ACC) wherein the major auto companies worked closely with their industry
suppliers and academic researcher to target selected structural components such
800 F.K. Ko and L.Y. Wan
Fig. 27.3 All composite aircraft wing developed by McDonnell Douglas [5]
as the cross members and the crush rails. While the earlier program proved
the feasibility of building a composite car, the ACC program demonstrated the
feasibility of cost reduction by significant part consolidation for selected structural
components. Following ACC was the partnership for the new generation vehicles
(PNGV) promoted by then Vice President Al Gore. The goal of PNGV was to
achieve 3 the fuel efficiency or 80 miles/gallon (at the 1993 level) by reducing
the weight of the car body and chassis by 50 %. This has to be accomplished
with reduction in manufacturing cost and emission level. Composite materials were
identified as key components to achieve structural weight reduction of body and
chassis components by at least 60 % and overall vehicle weight by at least 50 %
comparing to 1997 passenger vehicles.
The solution to the weight-reduction issue was demonstrated in the Sunrise™
electric vehicle (EV) program sponsored by the Department of Commerce (NIST) in
1997. The Sunrise team, through extensive and strategic use of TSCs, demonstrated
that a full-size electric vehicle equivalent to a Ford Taurus in interior room but
weighs 40 % less at 159 kg (350 lb) can be built for a market-competitive selling
price of $20,000 at an annual production of 20,000 units. The Sunrise EV has a
combined city/highway range of 290 km (179 miles) and achieved 38.5 km per liter
(90.5 miles per gallon) equivalent of gasoline with an average energy efficiency of
57 watt-hours per km (91 watt-hours per miles) proving that the Sunrise Technology
already exceeded the PNGV target of 34 km per liter (80 miles per gallon) of
gasoline. Figure 27.5 illustrates the extensive use of textile preforming technology to
create cost-effective fiber architectures that meet the specific structural requirement
corresponding to the specific part of the vehicle. In order to meet the cost target,
hybrids of carbon fiber and glass fiber were used. This was done through a strategic
placement of the carbon fibers as guided by the structural analysis. Figure 27.6
shows a photo of the final Sunrise™ electric vehicle (EV).
Top shell
MWK
Pilars SCRIM
braid P
Crossmember
braid
Battery box Pultrusion
braid
Battery box cover
woven
Floorpan Composite
Thermoforming 1.
MWK 0.91667
0.83333
0.75
0.66667
Wheel well 0.58333
0.5
net shape knit 0.41667
0.33333
0.25
Rocker panel RTM 0.16667
0.08333
0.
braid
Kick panel
woven
Thermoforming
Firewall
woven/nonwoven
Fig. 27.5 Design and structural analysis of the Sunrise™ electric vehicle (EV) [6]
F.K. Ko and L.Y. Wan
27 Textile Structural Composites: From 3-D to 1-D Fiber Architecture 803
Fig. 27.6 Photo of the final Sunrise™ electric vehicle (EV) [6]
A discrete fiber system such as a whisker or fiber mat has no material continuity.
The fibers are usually randomly oriented, although there are some aligned discrete
fiber systems that exist. The structural integrity of a discrete fibrous preform is
derived mainly from interfiber friction. The strength translation efficiency, or the
fraction of fiber strength translated to the nonaligned fibrous assembly of the
reinforcement system, is quite low.
A continuous filament or unidirectional (0ı ) system has the highest level of fiber
continuity and linearity and consequently has the highest level of strength translation
efficiency. This system is very suitable for filament wound and lay-up structures.
The drawback of this fiber architecture is its lack of in-plane and out-of-plane yarn
interlacing induced intra- and interlaminar weakness.
A third category of fiber reinforcement is the planar interlaced and interlooped
systems. The intralaminar failure problem associated with the continuous filament
system is addressed with this fiber architecture, but the interlaminar strength is still
weak owing to the lack of through-thickness fiber reinforcement.
The fully integrated or 3-D system forms the fourth category of fiber architec-
ture. 3-D systems for structural composites are fully integrated continuous fiber
assemblies having multiaxial in-plane and out-of-plane fiber orientation. The most
attractive feature of the integrated structure is the additional reinforcement in the
through-thickness direction which makes the composite virtually delamination-free.
Another advantage of the fully integrated structures is their ability to assume
complex structural shapes.
2-D and 3-D fabrics are distinguished by yarn orientation distribution and the
number of yarn diameters in the thickness direction. A 2-D fabric consists of two
to three yarn diameters in the thickness direction with fibers oriented in the x–y
plane. A 3-D fabric, consisting of three or more yarns in the thickness direction,
is a fibrous network wherein yarns pass from surface to surface of the fabric in all
three directions. While weaving, braiding, and knitting can produce planar or 3-D
structures, nonwoven fabrics can be a 2-D planar system with random or organized
fiber orientation, as well as the orthogonal 3-D system.
The efficient translation of fiber properties to the composite depends on the judi-
cious selection of fiber architecture that is governed by the directional concentration
of fibers. This directional fiber concentration can be quantified by fiber volume
fraction Vf and fiber orientation . Depending upon the textile manufacturing
process used and the type of fabric construction, families of Vf – functions can
be generated. These Vf – functions can be developed by geometrical modeling
as detailed by Ko and Du [9]. Accordingly, the structure–property relationship of
TSCs is a result of the dynamic interaction of microstructural and macrostructural
geometries. By proper selection of the geometry of the fibrous structures and
architecture and the method of placement or geometric arrangement of the fibers,
the structural performance of the resulting composite can be tailored. Preform fiber
architecture plays a key role in composite manufacturing by facilitating processing
steps including forming and resin infiltration. The properties of the composite will
also vary depending on the fiber architecture. Fiber orientation () and volume
fraction (Vf ) are key engineering parameters for TSC formability, permeability, and
performance. Table 27.1 summarizes the engineering and processing parameters for
textile preforms [9].
Of the large family of textile structures, 3-D fabrics have attracted the most
serious interest in the aerospace industry and served as a catalyst in stimulating
the revival of interest in TSCs [7]. The expansion of global interest in recent years
in 3-D fabrics for resin, metal, and ceramic matrix composites is a direct result
of the current trend in the expansion of the use of composites from secondary to
primary load-bearing applications in aerospace to automobiles, building infrastruc-
tures, surgical implants, aircraft, and space structures. These industries require a
huge reduction of fabrication cost and a substantial improvement in the through-
thickness strength, damage tolerance, and reliability of composites. In order to
improve the damage tolerance of composites, a high level of through-thickness
and interlaminar strength is required. The reliability of a composite depends on the
uniform distribution of the materials and consistency of interfacial properties. The
structural integrity and handleability of the reinforcing material for the composite
is critical for a large-scale, automated production. With the experience gained in
the 3-D carbon–carbon composites and the recent progress in fiber technology and
computer-aided textile design and liquid-molding technology, the class of 3-D fabric
structures is hence increasingly being recognized as serious candidates for structural
composites.
The introduction of a third system of yarns in the through-thickness direction
necessitates the reduction of yarn packing density (therefore fiber volume fraction)
in the planar direction. As shown in Table 27.1, the range of orientation of yarns in
the three principal planes and the packing density (Vf ) and structural integrity vary
depending on the method of fabric formation. The structural integrity of the fabric
preform depends on the integrity of yarn interlacing. Everything being equal, the
3-D woven and 3-D braid should have higher structural integrity than the nonwoven
fabric and the MWK fabric. While the width of the 3-D woven and MWK fabrics
can be over 60 inches, the thickness of these fabrics is limited by the machine gauge.
On the other hand, it is easier to produce thickness in the nonwoven and 3-D braided
fabrics but at a narrower width. Although the nonwoven and 3-D woven fabrics
provide the most direct method to introduce a large quantity of through-thickness
reinforcement for thick and thin composite preforms, respectively, it should be
noted that it has been demonstrated that less than 10 % of the total fiber volume
is required to produce a sufficient level of through-thickness reinforcement for
structural composites. From the point of view of conformability and the possibility
of near-net shape manufacturing, it is evident that the 3-D braid is suitable for solid
net shapes as well as formed shapes. The 3-D knitted structures also have a high
level of conformability and are quite suitable for molded composites.
On the basis of the qualitative assessment, it is difficult, if not impossible,
to identify a uniquely superior 3-D system, for it depends a great deal on the
specific matrix system to be used and most importantly the end-use requirements.
In order to provide a basis for the objective assessment of the performance of
the various composite systems, it would be desirable to compare their mechanical
properties. The fundamental question is what properties and how these properties
are characterized such that they can bring out the uniqueness of the toughened
composites by through-thickness reinforcement.
808 F.K. Ko and L.Y. Wan
27.2.3.1 Formability
Fig. 27.10 Formability comparison between plain woven and triaxial basic fabric of similar area
density [6]
810 F.K. Ko and L.Y. Wan
Fig. 27.11 The effect of fiber volume fraction and fiber orientation on permeability [6]
27.2.3.2 Permeability
Textile fluid flow permeability is an indication of how easily and uniformly a matrix
can be infiltrated into the fibrous assembly. McCarthy and Kim [14] concluded that
the permeability of textile preforms is mainly affected by fiber volume fraction and
fiber orientation as shown in Fig. 27.11. In the same figure, it can be noted that, for
the same fiber volume fraction, ordered structure such as 0/90ı woven fabrics have
higher permeability than disordered structures such as discrete chopped fibers. The
dependence of permeability on fiber volume fraction was also observed by Loos and
Weldemen [15]. Their study of carbon multiaxial warp knit preforms found that the
introduction of through-thickness fibers significantly increased the permeability of
the preforms, especially for preforms with high fiber volume fraction. The Kozeny–
Carman equation was found to be adequate to provide a quantitative relationship
between permeability and preform porosity.
27.2.3.3 Properties
Fig. 27.12 Dynamic interaction of fiber, fiber architecture, and processing [16]
The engineering design and analysis of 3-D composites requires knowledge of the
geometric arrangement of the fabric reinforcement in addition to the contributions
of fiber properties, matrix properties, and fiber-matrix interface. The geometry
of the fabric preforms dictates the integrity of the composite structure. Fiber
orientation and fiber volume fraction affect not only the performance but also
the processability of the composite system. A typical procedure for the design of
3-D fabric composites is to establish unit cell geometry and define the geometric
parameters and their interrelationships quantitatively. An excellent example of this
procedure is illustrated in the classic paper on fabric geometry by Pierce [17].
Knowing the fiber density and linear density, the yarn bundle size and yarn packing
density are determined to produce the fiber volume fraction for a given fiber
orientation. One of the few articles published on the design of 3-D fabrics for
composites was presented by McAllister and Lachman [18].
Paul [19] used energy theorems of elasticity theory to find upper and lower
bounds on the elastic moduli in tension and shear for two-phase materials and
derived “strength of materials” type approximate solutions. The laminate analog
has been used by Halpin et al. [20] to predict the elastic constants of 3-D fiber-
reinforced composites. Employing the mini-mechanical analysis of a unit cell,
Aboudi proposed a method to predict the effective elastic moduli and coefficient
812 F.K. Ko and L.Y. Wan
thermal expansion of 3-D composites [21]. Chatterjee and Kibler [22] used the
variational bounding technique for the determination of the upper and lower bounds
on the elastic properties of 3-D composite. In Rosen et al.’s approach [23], the
constant strain technique of Voigt was used to derive the overall constants of the
tri-directionally reinforced composite.
Other approaches to the mechanistic analysis of 3-D-reinforced composite
include the finite element method [24–28] and the stiffness averaging methods
[29, 30]. Ma and coworkers used two different analytical models to investigate the
elastic properties of 3-D braided fiber-reinforced composites. The “fiber interlock
model” [31] is based on the maximum strain energy principle, while the “fiber
inclination model” [32, 33] utilizes the modified classical laminate theory to analyze
the effective elastic properties of the composite. The classical laminate theory has
also been employed by Crane and Camponeschi [34] to determine the extensional
stiffness of the 3-D braid reinforced composite in the three principal geometric
directions.
Given a large family of fiber architectures that can be generated by an impressive
array of textile preforming techniques, it is quite evident that one can tailor
composite properties to meet various end-use requirements. In order to facilitate
the rational selection and to stimulate the creative design of fiber architecture
by textile preforming, a science-based design framework must be established to
bridge the communication gap between textile technologists, composite materials
engineers, and structural design engineers. This framework must be capable of
relating preform-manufacturing parameters to fiber architectural geometry as well
as material properties. This design framework can be constructed through three
levels of modeling including topological, geometrical, and mechanical models.
The topological model is a quantitative description of the preforming process.
The geometrical model is the heart of the design framework that quantifies fiber
orientation and fiber volume distribution in terms of fiber bundle (yarn) and fabric
structural geometry created by the preforming process. The mechanical model
provides the link between the mechanical properties of the material system and
the fiber architecture. The product of the mechanical model is the stress–strain
response of the textile-reinforced composite for a given boundary condition as well
as a stiffness matrix reflecting the material and fiber architecture contribution to the
properties of the composite system. With the stiffness matrix, the structural designer
can perform finite element analysis with meaningful information, thus facilitating
the integration of material design concepts, manufacturing processes, and structural
design to product engineering.
The mechanical properties of textile-reinforced composites can be predicted with
a knowledge of the fiber properties, matrix properties, and textile preform fiber
architecture through a modified laminate theory approach. Geometric unit cells
defining the fabric structure (or textile preform) can be identified and quantified
to form a basis for the analysis. Using 3-D braided composites as an example, the
following illustrates an approach that focuses on the integration of fiber architecture
design for manufacturing. The unit cell geometry of two-step braids has been
reported by Du et al. [35]. In this unit cell, pitch length and percentage of braider
27 Textile Structural Composites: From 3-D to 1-D Fiber Architecture 813
yarns were identified as key process parameters that control the braid microstructure
and the jamming criterion for the two-step braid given. The unit cell geometry of
the track-and-column braid has also been investigated by many researchers since
the early 1980s [36–38]. A common assumption made in most of the analyses
is that the braider yarns are oriented along the four diagonals in the unit cell.
However, the fiber volume fraction of 3-D braid is normally over 0.5, so the yarns
cannot be treated as dimensionless lines to cross each other at the center of the
unit cell. Li et al. [37, 38] described a more realistic unit cell geometry, assuming a
cylindrical shape for yarns. From a “preform processing science” point of view, Ko
and coworkers [39–45] developed a “Fabric Geometry Model” (FGM) based on the
unit cell geometry. The stiffness of a 3-D braided composite was considered to be the
sum of stiffness of all its laminate. The unit cell for the 3-D braid can be represented
by several yarns running parallel to the body diagonal of the cell. However, in
some instances, yarns are placed in longitudinal (0ı ) and transverse (90ı) directions
of the fabric and are referred to as longitudinal and transverse reinforcements (or
lay-ins), respectively. The preform processing parameters are specifically related to
corresponding unit cell geometries. The geometric descriptions form the basis for
an FGM that models a characteristic volume. Accordingly, the generation of the
stiffness matrix through the FGM provides a link between microstructural design
and macrostructural analysis. In this section, we will use a four-step Cartesian
braiding process as an example to explain how the unit cell is built and how the
FGM can work for microstructural design and macrostructural analysis.
Fig. 27.13 The spatial traces of yarn moment (red), idealized yarn traces (blue), and unit cells
(green)
variety, k equals 1. As shown in Fig. 27.14, the angle the yarns in the interior of the
preform make with the Z-axis, , is defined as “interior braiding angle” (it is evident
that the “interior braiding angle” is preserved). Setting the yarn length in a unit cell
as L, the geometric relations for the unit cell can be described with the equations
shown below.
W
D cos (27.1)
L
V
tan ˛ D (27.2)
W
Since k D U
V
, we have
p p
U2 C V 2 k2 V 2 C V 2 p V p
tan D D D 1 C k2 D 1 C k2 tan ˛ (27.3)
W W W
i.e.,
p
D arctan 1 C k2 tan ˛ (27.4)
27 Textile Structural Composites: From 3-D to 1-D Fiber Architecture 815
where
Here, the subscripts i and s denote the interior yarn and surface yarn, respectively;
Ny is the total number of yarns in the composite preform; Nf is the total number
of fibers in the yarn; df is the diameter of the fibers; L represents the fiber’s total
length; W is the height of the composite preform; Ac describes the cross-sectional
area of the composite preform (cm2 ); f denotes the fiber density (g/cm3 ); and Dy is
the yarn linear density (denier).
According to the definition of yarn linear density (denier):
df2
Dy D Nf f cd (27.8)
4
Nyi Dy
vfi D (27.9)
cd Aci f cos
Nys Dy
vfs D (27.10)
cd Acs f cos
Substitute Eqs. (27.4), (27.5), (27.6), (27.7), (27.8), (27.9), (27.10), and (27.11),
p
cd f Aci vfi cos ar tan 1 C k2 tan ˛
Nyi D (27.13)
Dy
where
2 2
dc 2dy 2dy
vfi D vf D 1 vf (27.14)
dc2 dc
Figure 27.15 shows the portion of the surface yarn and interior yarn volume fraction
vary in composite preforms with different sizes. When a large section is considered,
the proportion of interior cells is much larger than surface cells, the volume and
the properties of the interior cells will dominate the composite properties, and the
contribution of surface yarns can be neglected, i.e., vf Š vfi . Therefore, the total
yarn numbers composes a composite preform that can be approximately written as
p
cd f Ac vf cos ar tan 1 C k2 tan ˛
Ny D (27.15)
Dy
Fig. 27.15 Volume fraction of interior yarn and surface yarn at different composite diameter
Once the fabric geometry has been quantified, the result can be used together with
the fiber and matrix properties to predict the mechanical properties of the composite
system through a modified lamination theory. Assuming each system of yarn can be
represented by a comparable unidirectional lamina, then the elasticity matrix of each
lamina in material coordinate is determined by fiber properties, matrix properties,
and fiber volume fraction, the elasticity matrix of the lamina in a unit cell coordinate
system is determined by coordinate transformation, and the elasticity matrix of unit
cell can be obtained by the summation of individual elasticity matrices in the unit
cell. The lamina with an elastic stiffness matrix is defined as follows:
2 3
C11 C12 C13 0 0 0
6 C22 C23 0 0 0 7
6 7
6 7
6 C33 0 0 0 7
ŒC D 6 7 (27.17)
6 C44 0 0 7
6 7
4 C55 0 5
C66
where
1 v23 v32 v21 v31 v23 1 v13 v31 v32 v21 v31
C11 D ; C12 D ; C22 D ; C23 D ;
E22 E33 E22 E33 E11 E33 E11 E33
818 F.K. Ko and L.Y. Wan
and
1 v12 v21 v23 v32 v31 v13 2v21 v32 v13
D I
E11 E22 E33
then the elastic stiffness matrix of this yarn system in the longitudinal direction of
the panel can be expressed as
where [Ci ] is the stiffness of the ith system of yarns, [T" ] is the geometric strain
transformation for the ith system of yarns, and [C] is the stiffness matrix for a
comparable unidirectional composite, wherein the transformation matrix [T" ] is a
Hamiltonian tensor transformation matrix and can be defined in the following forms:
2 3
l21 m21 n21 2m1 n1 2l1 n1 2l1 m1
6 l2 m2 n2 2m2 n2 2l2 n2 2l2 m2 7
6 2 2 2 7
6 2 2 2 7
6 l3 m3 n3 2m3 n3 2l3 n3 2l3 m3 7
ŒT" D 6 7 (27.19)
6 l2 l3 m2 m3 n2 n3 m2 n3 C m3 n2 n2 l3 C n3 l2 l2 m3 C l3 m2 7
6 7
4 l1 l3 m1 m3 n1 n3 m3 n1 C m1 n3 n3 l1 C n1 l3 l3 m1 C l1 m3 5
l1 l2 m1 m2 n1 n2 m1 n2 C m2 n1 n1 l2 C n2 l1 l1 m2 C l2 m1
where
l1 D cos m1 D 0 n1 D sin
l2 D sin cos ˇ m2 D sin ˇ n2 D cos cos ˇ
l3 D sin sin ˇ m3 D cos ˇ n3 D cos sin ˇ
E11 D Ef Vf C Em Vm (27.20)
E22 D E33 D Em Ef = Ef Vf C Em Vm (27.21)
G12 D G13 D Gm Gf = Gm Vf C Gf Vm (27.22)
h p i
G23 D Gm = 1 Vf 1 Gm =Gf 23 (27.23)
27 Textile Structural Composites: From 3-D to 1-D Fiber Architecture 819
The FGM can be used to predict the stress–strain behavior of a TSC. In order
to determine the stress–strain behavior of the fabric-reinforced composites, it is
necessary to utilize each of the yarn systems. A model for yarn system interaction
has been chosen wherein the stiffness matrices for each system of yarns are
superimposed proportionately according to contributing volume to determine the
fabric-reinforced composite system stiffness:
X
ŒCs D ki ŒCi (27.27)
where [Cs ] is the total stiffness of unit cell and ki is the volume fraction of the
constituent.
The elastic constants of composites can be calculated from the responding
elements of the compliance matrix. From Eq. (27.28), the elastic constants are given
as follows:
A D 1= ŒSs (27.29)
A11
v12 D (27.33)
A12
A11
v13 D (27.34)
A13
A22
v23 D (27.35)
A32
820 F.K. Ko and L.Y. Wan
In order to account for the potentially nonlinear behavior of the materials, the
system stiffness matrix should be calculated anew at each strain level. Thus, the
stress–strain behavior of the composite can be expressed as:
DC (27.37)
Uci Um (27.38)
ˇ ˇ
ˇ Z ˇ
ˇ ˇ
ˇ
where Um D ˇvf fu "fu =2 C 1 Vf u ."/ "d"ˇˇ ;
ˇ ˇ
0
Using these maximum energy criteria, a failure point for each system of yarns can
be found. When a system of yarns fails, its contribution to the total system stiffness
is removed. When all systems have failed, the composite is said to have failed. In
this way, the entire stress–strain curve for the composite can be generated up to the
point of composite failure.
The FGM has been employed with satisfactory results to predict the stress–strain
properties of polymer, metal, and ceramic matrix components. Shown in Fig. 27.16
are the theoretical and experimental tensile stress–strain relatively FP-reinforced
Al-Li composites, respectively.
By using the FGM analysis, one can obtain the elasticity matrix [C] of a braided
composite. In a displacement-based finite element formulation, the stiffness matrix
[K] can be obtained from the equation
Z
KD BTCBdy (27.39)
27 Textile Structural Composites: From 3-D to 1-D Fiber Architecture 821
Fig. 27.16 Comparison of experimental and theoretical frystress–strain behavior of a 3-D braided
fiber FP-reinforced Al-Li composite with fiber volume fraction D 17 % and braid angle D 20ı [46]
where [B] is the strain-displacement transformation matrix. Therefore, the FGM can
be incorporated into general finite element analysis programs for solving braided
composite structures with complex shapes. Data input to this program include
fiber/matrix properties and braiding parameters, i.e., surface angle, inclined angle,
fiber volume fraction, and braiding percentage. In order to reflect the contribution of
local fiber architecture effect in the structural analysis of the vehicle, an integrated
design for manufacturing methodology was used. Figure 27.17 shows the logic
flow of the methodology. This integrated design for manufacturing methodology
(IDM) has been applied to the design and analysis of various complex-shaped engine
components [7, 41, 47], the braided composites of the aircraft structures [48], and
automotive composites [6]. Here we use a braided automotive pillar design as an
example.
The pillars are the structural backbones of the automobile top shell. The
B-pillar/seat–rail structure experiences high seat-belt loads in a crash situation.
The laminate in that area was designed to sustain seat-belt loading resulting from
frontal and rear impact based on the FEA analysis shown in Fig. 27.18. The
laminate schedule calls for a combination of 2-mm carbon uni-tape with 1-mm bias
fiberglass multiaxial warp knit which yields the following composite properties:
E11 D 88.3 GPa (12.8 Msi), E22 D 8.3 GPa (1.2 Msi), E22 (flange) D 26.2 GPa
(3.8 Msi), and G12 D 4.8 GPa (0.7 Msi). The braided B-pillar is shown in Fig. 27.19.
To simplify the assembly time, triaxial braided structure was designed and built
by mapping the laminate properties to the braided composite according to the FGM.
The following braided composite properties were generated by the FGM analysis:
822 F.K. Ko and L.Y. Wan
Plot Stress,
Displacement,
Contour
Does
Design
meet
Criteria ? No
Yes
Complex Shape
Verification
E11 D 85:6 GPa .12:4 Msi/, E22 D 14:5 GPa .2:1 Msi/, E22 .flange/ D 29 GPa
(4.2 Msi), and G12 D 9:7 GPa .1:4 Msi/. Figure 27.20 shows the predicted and
experimental tensile moduli for the various pillars, indicating reasonable agreements
between theoretical predictions and experimental results. This provided confidence
in the selection of braided fiber architecture according to the processing window
prescribed in the volume fraction-braiding angle relation.
Judging from the phenomenal growth of carbon fiber in the last 50 years as
a reinforcement material for composite products, it is anticipated that carbon
fibers in different size/forms will sustain the growth in energy-/environment-related
applications. Because of its impact on energy consumption and environment,
automotive application is expected to be the next major growth area for carbon
27 Textile Structural Composites: From 3-D to 1-D Fiber Architecture 823
70
60
Modulus (GPa)
50
40
30
20
10
0
E11 E22 E11 E22 E11 E22
Design/Predicted Measured
Fig. 27.20 Predicted and experimental tensile moduli for the various pillars [6]
It is anticipated that the next major growth in the carbon fiber market will come
from automotive and construction industries. The key obstacle to the entry of carbon
fiber into the auto composite market is the high cost of PAN-based carbon fiber.
The manufacturing cost of a suitable grade of PAN-based carbon fiber is highly
dependent on petroleum price, with around 51 % being precursor cost, 18 % utilities,
12 % depreciation, 10 % labor, and 9 % represented by other fixed costs [49].
Current price of carbon fiber is over US$30/kg, while the automotive industry
requires suitable carbon fiber cost less than $11–$15.40/kg with a tensile strength of
1.72 GPa (250 ksi) and a modulus of 172 GPa (25 Msi) [49]. Therefore, the cost of
carbon fiber must be significantly reduced for it to be attractive for wider utilization
for automotive.
One of the candidate polymer precursors that has attracted a lot of attention
is lignin, which is arguably the second most abundant biopolymer on Earth. It is
nature’s most abundant source of aromatic compounds and represents 30 % of all
nonfossil organic carbon on Earth (http://iacmi.org/). Lignin is produced in large
quantities as a by-product of the pulp and paper industry wherein it is primarily
burnt as a low-efficiency fuel to power the mill. With the depletion of fossil-derived
chemicals and energy, using a renewable source for energy and chemicals will
become crucial in the near future. As a result, many governments have passed
legislation to increase the production of energy and chemicals from renewable
resources (http://www.lcc.mw.tum.de/en/home/). For example, the US Department
of Agriculture and US Department of Energy have set goals to produce over 20 %
27 Textile Structural Composites: From 3-D to 1-D Fiber Architecture 825
7
Carbon fiber tensile strength (GPa)
6
Commercial PAN CF
Strength:3 -7GPa, Modulus: 200-600 GPa
5
3
Automotive CF target :
Strength: 1.72 GPa, Modulus: 172 GPa
2
Typical lignin CF :
1 Strength: 0.5-1 GPa, Modulus: 50-70 GPa
0
0 200 400 600 800
Carbon fiber Modulus (GPa)
in the relative intensity and position of Raman bands in the tubes embedded in
the fibers was observed due to the interaction of SWNTs with the polymer or
carbon matrix, debundling, and other effects. The distribution and orientation of
SWNTs in PAN fibers were studied by transmission electron microscope (TEM)
analysis. According to TEM analysis, SWCNTs maintain their straight shape and
are parallel to the axis direction of the PAN fibers indicating that a good alignment
of SWNTs has been achieved, Fig. 27.23. The improved orientation also results
in a better distribution – nearly every investigated section of the polymer fibers
contained at least some SWCNTs. Figure 27.24 shows the results of the mechanical
property measurements of the stabilized CNT/PAN and neat PAN fibers placed on a
mica substrate. Experiments show a nonlinear load–deformation relationship for the
SWNT/PAN composite nanofibrils. This may be attributed to the finite deformation
of the fiber under large load. The elastic modulus of the fiber is calculated to be
around 207 GPa using the linear portion of the load–deformation curve at small
deformation and under low forces (<100 nN).
To avoid the agglomeration of CNT, emulsion electrospinning was used to pre-
pare CNT-reinforced lignin-based carbon nanofibers. The 30 wt% SKLF4 /SWNTs
(99.5/0.5 w/w) as-spun nanofibers were collected for single nanofiber tensile testing,
830 F.K. Ko and L.Y. Wan
500 Mica
PAN fiber
400 PAN-CNT fiber
Force (nN)
300
200
Δd
100 Δz
0
0 10 20 30 40
Δz (nm)
60
50
Engineering Stress (MPa)
SKL/SWNTs
40
30
SKL
20
0
0 0.01 0.02 0.03 0.04 0.05
Engineering Strain (mm/mm)
and the average fiber diameter was 886 ˙ 248 nm. A representative stress–strain
curve of single as-spun nanofiber and a random nanofiber mat is shown in Fig. 27.25.
The average strength and Young’s modulus of the 30 wt% SKLF4 /SWNTs as-spun
nanofiber were 46.60 ˙ 7.84 MPa and 3.73 ˙ 0.52 GPa, respectively, which were
higher than that of the pure SKLF4 as-spun nanofiber with strength and Young’s
modulus of 34.36 ˙ 11.53 MPa and 2.91 ˙ 1.34 GPa, respectively.
27 Textile Structural Composites: From 3-D to 1-D Fiber Architecture 831
Structural health monitoring (SHM) aims to give, at every moment during the
life of a structure, a diagnosis of the “state” of the constituent materials, of the
different parts, and of the full assembly of these parts constituting the structure as
a whole [63]. It involves the integration of sensors, possibly smart materials, data
transmission, computational power, and processing ability inside the structures. By
integrating of smart or intelligent materials and structures for SHM, vehicles will
become more reliable and require less operation and maintenance cost. Therefore,
the concept of smart or intelligent composite has occupied the minds of engineers.
The component technologies of smart or intelligent composite are mainly grouped
into two categories: multifunctional materials and structures and smart materials and
structures.
One pioneer work is the project titled “Bio-Inspired Intelligent Sensing Materials
for Fly-By-Feel Autonomous Vehicles” as part of “Multidisciplinary Research Pro-
gram of the University Research Initiative” (MURI) project, which was a scientific
initiative sponsored by the Air Force Office of Scientific Research (AFOSR). As
shown in Fig. 27.26, the objective of this project is to lay a basis for the use of
sensory materials as an integral part of the total structural system rather than the
sensors and actuators add-on, which adds weight and introduces defects to the
structure. As one of the project section, UBC undertook the tasks of design and
creating the next generation of lightweight, intelligent sensing aerospace materials
by taking advantage of nano-functionalized fibers that have desirable sensing and
actuation properties engineered into them. Accordingly, the UBC developed the
enabling materials for sensor/actuators by multiscale integration of these functional
Fig. 27.26 (a) Biosensor network and (b) bioinspired sensor network
832 F.K. Ko and L.Y. Wan
Stretchable and conformal strain sensors which can emulate the tactile sense are
indispensable in realizing wearable and skin-like electronics. Such sensors can be
used in a wide range of applications including direct measurement of mechanical
strain and cracks for monitoring of aging infrastructure, sensing mechanical forces
27 Textile Structural Composites: From 3-D to 1-D Fiber Architecture 833
Fig. 27.27 Fabrication process of core-shell NF webs. (a) Schematic of the fabrication steps of
the encapsulated core-shell NF mesh sensors: (I) electrospinning of PAN NFs, (II) sputtering a thin
gold shell using NF mesh as template, (III) cutting active ribbons of NF mesh, (IV) transferring
the ribbons to a PDMS substrate, (V) deposition of electrical contacts, (VI) encapsulation with
the top PDMS layer, and (VII) curing the sensor. (b, c) Photographs of typical samples showing
dimensions and flexibility of piezoresistive NF sensors. The inset SEM micrographs show gold-
coated core-shell NF mesh before transferring to PDMS; a thick layer consists of multiple levels
of NFs as well as a high-magnification cross-sectional SEM image of the core-shell NF mesh,
demonstrating a tubular continuous gold shell NFs
834 F.K. Ko and L.Y. Wan
a vacuum infiltration and curing process, which varies depending on the polymer
matrix, to obtain sensors at desired size as shown in the Fig. 27.27b. The sensors
can be laminated on different surfaces, as shown a typical sensor laminated on a
curved carbon fiber-reinforced composite structure. Scanning electron microscopy
(SEM) images of the typical NF mesh (Fig. 27.27) demonstrate that a uniform
conformal coating is achieved by gold coating on the surface of the NFs as well as
formation of inter-NF junctions between overlapping fibers. The measured electrical
resistance of both ribbons and single Au-coated NFs shows that the resistance of the
metallized NFs depends on the density of NFs and the thickness of the gold layer.
The resistance of segments of single metallized NFs is found to be proportional to
the length of the segment, which confirms good uniformity of the diameter of NFs
and the metallic shell.
Figure 27.28a illustrates the measured resistance of the NF-based strain sensor
in response to an increasing uniaxial tensile strain from 0 to 2 %. For strains
below 1 %, the gauge factor increases linearly and stabilizes for strains above 1 %.
Figure 27.28b depicts the response of the sensor to over 1000 stretching cycles,
demonstrating durability of the sensors. The sensor shows a remarkably stable
response after 1000 cycles, which can be attributed to the plastic relaxation of the
PDMS matrix. The measured response of the sensor at different applied strain rates
shows no significant differences between the responses at different strain rates.
The dynamic response of an NF sensor laminated on a steel cantilever (clamped
on one side) is demonstrated in Figure 27.29. When the free end of the blade
is released from a deflection point, the blade oscillates applying both tensile and
compressive strains, depending on the deflection way and the surface of the blade
that the sensor is laminated on. It can be seen that the sensor provides a very accurate
response to low compressive and tensile strains applied during this experiment,
Fig. 27.28 (a) Measured resistance of the NF sensor as a function of time in response to the
changing strain (dashed line) from 0 to 2 % at a constant speed of 0.2 % per second. (b) The
repeatability of the strain sensor response over 1000 uniaxial stretching cycles (the insets show
the close-ups of the curve at two different intervals near the beginning and close to the end of the
experiment)
27 Textile Structural Composites: From 3-D to 1-D Fiber Architecture 835
Fig. 27.29 (a) The measured change in the resistance of the sensor showing damped oscillations
of the sensor laminated on the surface of the vibrating beam. A close-up of the graph is shown in the
inset. (b) The response of the sensor laminated on the blade to a desired applied strain at different
temperatures. The response is measured for both compressive and tensile applied deflections. Inset
shows a schematic of the placement of sensor on one side of a vibrating clamped beam for testing
the dynamic response of the sensor as well as the electromechanical response of the sensor at
different temperatures
with the increase in resistance associated with tensile strains. The oscillation of
the free end of the cantilever is detected by measuring the changes in resistance of
the sensor. Figure shows the response of the sensor to damped natural vibrations
of the blade. By looking at the FFT of the measured signal, the extracted natural
frequency of oscillation is very well inconsistent with the expected first bending
mode natural frequency of the steel blade calculated using the dimensions and
physical properties of the blade. Figure 27.29b illustrates the relative changes in
the resistance versus temperature in response to static deflections of the blade (as
pictured in the experimental setup) for applying tensile and compressive strains
" in the range of C0.12 % to 0.12 %. The values of the relative change in the
electrical resistance under the maximum tensile and compressive strains, R/R0max ,
are plotted against temperature (25–240 ı C) in Fig. 27.30. It is observed that
upon heating, the R/R0max increases slightly from 0.66 % at room temperature
to 0.77 % at 240 ı C, in tensile mode, and from 0.48 to 0.59 %, in compressive
mode. This slight increase in the sensitivity of the sensor can be attributed to the
increase in elastic modulus of the PDMS elastomer, due to further cross-linking at
higher temperatures, which decreases the compliance of PDMS matrix, causing the
disjoining of NF junctions to happen at lower strain levels.
To examine the capability of NF strain sensors for embedding in composite
structures, strain sensors were fabricated with epoxy resin matrix instead of PDMS
elastomer. Figure 27.30 illustrates the typical NF strain sensors made with epoxy
resin (Epon Resin 828 and Epicure 3055). This figure also shows the response of NF
sensors made with epoxy resin (with two different tensile strengths) laminated on a
metallic blade to successive compressive and tensile deflections. As shown sensors
836 F.K. Ko and L.Y. Wan
Fig. 27.30 Photograph of typical NF strain sensors made with epoxy resin as well as the response
of similar NF sensors (with two different tensile strengths) laminated on a metallic blade to
successive compressive and tensile deflections
show clear response to the applied strain. The weaker response of the sensor denoted
with 100-40 can probably be attributed to lower deformation of the laminated sensor
due to the high tensile strength of the matrix.
Fig. 27.31 (a) SEM micrograph of CFs bundle coated with ZnO NWs, (b) cross-sectional SEM
micrograph of a single ZnO NW, (c) XRD pattern of ZnO NWs
(Fig. 27.33). The effective piezoelectric coefficient of the woven fabric was found to
be 2–3 pC/N, which is lower than bulk ZnO (10–12 pC/N), but met the expectation
as a result that the normal forces were not fully transferred to wires, and the
measurement was subject to fiber reconfiguration reducing repeatability. Within
the woven fabric, it was demonstrated that there was no cross talk through the
fabric (Fig. 27.33c), which is expected to enable individual sensor addressing and
measurements throughout the network with limited interference.
These encouraging results demonstrated the feasibility of the woven piezoelectric
sensor device that is capable of both reinforcing and sensing functions. With further
optimization and careful verification, these hybrid woven carbon-ZnO systems may
be integrated as a smart sensing device for a wide range of applications such as real-
time damage sensing in critical infrastructures (e.g., airplanes, bridges, etc) as well
as for human health monitoring applications.
While the success has been achieved in electrospun carbon nanofiber mats on a
laboratory scale, there are strong demands for continuous production of carbon
nanofiber yarns to facilitate the fabrication of hierarchical fiber structures that would
benefit numerous composite applications.
838 F.K. Ko and L.Y. Wan
Fig. 27.32 (a) Schematic depicting the various components and process for the woven sensor,
including hydrothermal growth of ZnO NWs on carbon fibers, dipcoating of PDMS protection
layers on ZnO NWs, and dipcoating of conductive PEDOT: PSS layers on polymer fibers to form
charge conduction paths, interweaving fibers to form fabric, and making contacts; (b) SEM image
of a single carbon fiber functionalized with ZnO nanowires and coated with PDMS
Fig. 27.33
Electromechanical tests on
ZnO nanowire woven
network; (a) force applied at
region A, charge collected at
region A (effective
piezoelectric coefficient of
2–3 pC/N); (b) force applied
at region B, charge collected
at region A (no cross talk
between regions)
Fig. 27.35 Composite ESEM image of continuous yarn [65]. (a) 700 untwisted yarn and (b) 350
Denier 3.48 tpi Yarn
cams, and therefore six different directions of planar carrier movement. The foot
part of the carrier is designed as a glider, whereby the bobbin is gliding during
movement from one position to the other by using the adjacent carriers as a bearing
surface. Compared to traditional four bobbin carriers that are capable of moving
carriers in four different planar directions, this unique hexagonal cam arrangement
adds two more planar movement directions to the process.
The first prototype machine following the hexagonal concept was designed and
manufactured at the Advanced Fibrous Materials Laboratory (University of British
Columbia, Vancouver, Canada) in cooperation with the Institut für Textiltechnik
(RWTH Aachen University, Aachen, Germany) [67], as shown in Fig. 27.37. This
machine can accommodate 132 carriers plus 37 axial yarns. This new braiding
27 Textile Structural Composites: From 3-D to 1-D Fiber Architecture 841
Fig. 27.36 Hexagonal braiding unit [66]. (a) Single unit, (b) basic unit, and (c) hexagonal cam
packing
Fig. 27.39 Simulated structures that can be realized by hexagonal braiding [66]
27 Textile Structural Composites: From 3-D to 1-D Fiber Architecture 843
References
1. D.R. Tenney, et al., NASA composite materials development: lessons learned and future
challenges. NASA Report LF99-9370, 2009
2. L. Gause, J. Alper, Braided to net section graphite/epoxy composite shapes. J. Compos.
Technol. Res. 10(2), 33–46 (1988)
3. T. Tan, et al., An integrated design methodology for fabric composite structural components,
in Proceedings of the Annual Meeting of the ASME—Advanced Composites and Processing
Technology, vol. 5, 1988, pp. 37–43
4. T.H. Walker et al., Advanced Technology Composite Fuselage—Structural Performance (Lan-
ley Research Center, Hampton, VA, 1997)
5. R. Palmer, Techno-economic requirements for composite aircraft components, in Innovative
Composite Aircraft Primary Structure, 1994
6. F. Ko, Textiles Composites for Automotive Structural Components, in Textile Advances in the
Automotive Industry, ed. by R. Shishoo (Woodhead Publishing, Cambridge, 2008)
7. F. Ko, 3-D Textile Reinforcements in Composite Materials, in 3-D Textile Reinforcements in
Composite Materials, ed. by A. Miravete (Woodhead Publishing, Cambridge, 1999)
8. F. Scardino, An introduction to textile structures and their behavior, in Textile Structural
Composites, ed. by T.W. Chou, F.K. Ko. Composite Materials Series, vol 3 (Elsevier, Oxford,
1989)
9. F.K. Ko, G. Du, Textile Preforming, in Handbook of Composites, ed. by S.T. Peters (Springer,
New York, 1998), pp. 397–424
10. F. Ko, G. Du, Processing of Textile Preforms, in Advanced Composites Manufacturing, ed. by
T.G. Gutowski (Wiley, New York, 1997)
11. K. Potter, The influence of accurate stretch data for reinforcements on the production of
complex structural mouldings: Part 1. Deformation of aligned sheets and fabrics. Composites
10(3), 161–167 (1979)
12. F.L. Scardino, F.K. Ko, Triaxial woven fabrics Part I: behavior under tensile, shear, and burst
deformation. Text. Res. J. 51(2), 80–89 (1981)
27 Textile Structural Composites: From 3-D to 1-D Fiber Architecture 845
13. N.F. Dow, Woven Fabric Reinforced Composites for Automotive Applications. NSF Grant No.
DMR-8212867, MSC TFR 1605/81021985 (1985)
14. S. Mccarthy, Y.R. Kim, Resin flow through fibre reinforcement during composite processing,
in the 5th Textile Structural Composites Symposium, Drexel University, Philadelphia, PA, 1991
15. A. Loos, M. Weldemen, RTM process modeling for advanced fiber architectures, in Advanced
Composite Materials: New Developments and Applications, Seventh Annual ASM/ESD
Advanced Composite, Detroit, MI, 1991, pp. 209–216
16. T.G.P. Gutowski, Advanced Composites Manufacturing (John Wiley & Sons, New York, 1997)
17. F.T. Peirce, The geometry of cloth structure. J. Text. Inst. Trans. 28(3), T45–T96 (1937)
18. L.E. McAllister, W.L. Lachman, Fabrication of Composites, in Handbook of Composites, ed.
by A. Kelly, S.T. Mileiko (Elsevier, New York, 1983), p. 109
19. B. Paul, Prediction of the elastic constants of multiphase materials. Trans. AIME 218, 36–41
(1960)
20. J. Halpin, K. Jerine, J. Whitney, The laminate analogy for 2 and 3 dimensional composite
materials. J. Compos. Mater. 5(1), 36–49 (1971)
21. J. Aboudi, Minimechanics of tri-orthogonally fibre-reinforced composites: overall elastic and
thermal properties. Fibre Sci. Technol. 21(4), 277–293 (1984)
22. S. Chatterjee, J. Kibler, An Analytical Model for Three-Dimensionally Reinforced Graphite
Composites, in Modern Developments in Composite Materials and Structures, ed. by
J.R. Vinson (ASME, New York, 1979), pp. 269–287
23. B. Rosen, S. Chatterjee, J. Kibler, An analysis model for spatially oriented fiber composites, in
Composite Materials: Testing and Design, ASTM STP, 1977, vol. 617, pp. 243–254
24. A. Ross, Designing with three dimensional composites. ASME Paper (74-DE), 1975, p. 25
25. P. Paramasivam, J.I. Curiskis, S. Valliappan, Micromechanics analysis of fibre reinforced
cement composites. Fibre Sci. Technol. 20(2), 99–120 (1984)
26. M.S. Aly-Hassan et al., Comparison of 2D and 3D carbon/carbon composites with respect to
damage and fracture resistance. Carbon 41(5), 1069–1078 (2003)
27. T. Zeng, L.-Z. Wu, L.-C. Guo, A finite element model for failure analysis of 3D braided
composites. Mater. Sci. Eng. A 366(1), 144–151 (2004)
28. L. Chen, X. Tao, C. Choy, Mechanical analysis of 3-D braided composites by the finite
multiphase element method. Compos. Sci. Technol. 59(16), 2383–2391 (1999)
29. A. Kregers, Y.G. Melbardis, Determination of the deformability of three-dimensionally
reinforced composites by the stiffness averaging method. Mech. Compos. Mater. 14(1), 1–5
(1978)
30. A. Kreger, G. Teters, Use of averaging methods to determine the viscoelastic properties of
spatially reinforced composites. Mech. Compos. Mater. 15(4), 377–383 (1980)
31. C.-L. Ma, J.-M. Yang, T.-W. Chou, Elastic stiffness of three-dimensional braided textile
structural composites, ed. by J.M. Whitney. in Composite Materials: Testing Design (Seventh
Conference) ASTM STP 893, (American Society for Testing and Materials, Philadelphia, PA,
1986), pp. 404–421
32. J.-M. Yang, C.-L. Ma, T.-W. Chou, Fiber inclination model of three-dimensional textile
structural composites. J. Compos. Mater. 20(5), 472–484 (1986)
33. T.J. Whitney, T.-W. Chou, Modeling of 3-D angle-interlock textile structural composites. J.
Compos. Mater. 23(9), 890–911 (1989)
34. R. Crane, E. Camponeschi Jr., Experimental and analytical characterization of multidimension-
ally braided graphite/epoxy composites. Exp. Mech. 26(3), 259–266 (1986)
35. G.-W. Du, T.-W. Chou, P. Popper, Analysis of three-dimensional textile preforms for multidi-
rectional reinforcement of composites. J. Mater. Sci. 26(13), 3438–3448 (1991)
36. F.K. Ko, Three-Dimensional Fabrics for Composites, in Textile Structural Composites, ed. by
T.W. Chou, F. Wo (Elsevier, Amsterdam, 1989), pp. 129–171
37. W. Li, M. Hammad, A. El-Shiekh, Structural analysis of 3-D braided preforms for composites
Part I: the four-step preforms. J. Text. Inst. 81(4), 491–514 (1990)
38. W. Li, M. Hammad, A. El-Shiekh, Structural analysis of 3-D braided preforms for composites
part II: the two-step preforms. J. Text. Inst. 81(4), 515–537 (1990)
846 F.K. Ko and L.Y. Wan
39. C. Pastore, F. Ko, Modelling of textile structural composites Part I: Processing-science model
for three-dimensional braiding. J. Text. Inst. 81(4), 480–490 (1990)
40. F. Ko, D. Whyte, C. Pastore, Control of fibre architecture for tough net-shaped structural
composites, in ASTM STP, 1988, vol. 979, pp. 290–298
41. T. Tan, C. Pastore, F. Ko, Engineering design of tough ceramic matrix composites for turbine
components. J. Eng. Gas Turbines Power 113(2), 312–317 (1991)
42. G. Du, F. Ko, Geometric modeling of 3-D braided preforms for composites, in Textile Structural
Composites Symposium, Drexel Laboratory, Philadelphia, PA, 1991
43. G.-W. Du, F.K. Ko, Unit cell geometry of 3-D braided structures. J. Reinf. Plast. Compos.
12(7), 752–768 (1993)
44. C. Lei, Y.-J. Cai, F. Ko, Finite element analysis of 3-D braided composites. Adv. Eng. Softw.
14(3), 187–194 (1992)
45. Y.T. Gao, F.K. Ko, H. Hu, Integrated design for manufacturing of braided preforms for
advanced composites Part II: 3D braiding. Appl. Compos. Mater. 20(6), 1065–1075 (2013)
46. T.-W. Chou, F.K. Ko, Textile Structural Composites (Elsevier, Amsterdam, 1989)
47. T. Tan, C.M. Pastore, F.K. Ko, Engineering design of tough ceramic matrix composites for
turbine components, in ASME 1989 International Gas Turbine and Aeroengine Congress and
Exposition, American Society of Mechanical Engineers, 1989
48. M.J. Fedro, C. Gunther, F.K. Ko, Mechanical and Analytical Screening of Braided Composites
for Transport Fuselage Applications, 1991.
49. D.A. Baker, T.G. Rials, Recent advances in low-cost carbon fiber manufacture from lignin. J.
Appl. Polym. Sci. 130(2), 713–728 (2013)
50. J.F. Kadla et al., Lignin-Based Carbon Fibers, in Chemical Modification, Properties, and Usage
of Lignin (Springer, New York, 2002), pp. 121–137
51. J.F. Kadla et al., Lignin-based carbon fibers for composite fiber applications. Carbon 40(15),
2913–2920 (2002)
52. C.D. Warren, 2011 Annual Progress Report for Lightweighting Materials, in 3. Polymer
Composites, 2011, pp. 3–41
53. F. Ko et al., Lignin-Based Composite Carbon Nanofibers, in Lignin in Polymer Composites, ed.
by O. Faruk, M. Sain (William Andrew, Norwich, NY, 2015)
54. R. Ruiz-Rosas et al., The production of submicron diameter carbon fibers by the electrospin-
ning of lignin. Carbon 48(3), 696–705 (2010)
55. M. Lallave et al., Filled and hollow carbon nanofibers by coaxial electrospinning of alcell lignin
without binder polymers. Adv. Mater. 19(23), 4292 (2007)
56. I. Dallmeyer et al., Preparation and characterization of interconnected, kraft lignin-based
carbon fibrous materials by electrospinning. Macromol. Mater. Eng. 299(5), 540–551 (2014)
57. A. Goudarzi, L.-T. Lin, F.K. Ko, X-Ray diffraction analysis of kraft lignins and lignin-derived
carbon nanofibers. J. Nanotechnol. Engineer. Med. 5(2), 021006 (2014)
58. L. Lin, Y. Li, F.K. Ko, Fabrication and properties of lignin based carbon nanofiber. J. Fiber
Bioeng. Inform. 6(4), 335–347 (2013)
59. S.-X. Wang et al., Lignin-derived fused electrospun carbon fibrous mats as high performance
anode materials for lithium ion batteries. ACS Appl. Mater. Interfaces 5(23), 12275–12282
(2013)
60. F. Ko et al., Electrospinning of continuous carbon nanotube-filled nanofiber yarns. Adv. Mater.
15(14), 1161–1165 (2003)
61. W.H. Weber, R. Merlin, Raman Scattering in Materials Science (Springer, New York, 2000)
62. C.N.R. Rao, A.K. Cheetham, Science and technology of nanomaterials: current status and
future prospects. J. Mater. Chem. 11(12), 2887–2894 (2001)
63. D. Balageas, Introduction to Structural Health Monitoring, in Structural Health Monitoring,
ed. by D. Balageas, C.-P. Fritzen, A. Güemes (Wiley-ISTE, London, 2006), pp. 13–44
64. S. Soltanian et al., Highly piezoresistive compliant nanofibrous sensors for tactile and
epidermal electronic applications. J. Mater. Res. 30(01), 121–129 (2015)
27 Textile Structural Composites: From 3-D to 1-D Fiber Architecture 847
65. R. Foedinger, et al., High strength nanomaterials fiber for lightweight composite missile cases,
in 230th ACS National Meeting, Washington, DC, 2005
66. F. Schreiber, et al., Novel three-dimensional braiding approach and its products, in 17th
International Conference on Composite Materials, Edinburgh, 2009
67. F. Ko, et al., Recent advancements in three-dimensional braiding, in Proceedings of the 1st
World Conference on 3D-Fabrics and their Applications, 2008
68. R. Dow, New concept for multiple directional fabric formation, in Advanced Materials: the Big
Payoff, 1989, pp. 558–569
69. A. Bogdanovich, D. Mungalov,Recent advancements in manufacturing 3-D braided preforms
and composites, in Proceedings of ACUN-4 Composite Systems-Macro Composites, Micro
Composites, Nanocomposites, University of New South Wales, Sydney, NSW, 2002
70. Staff, The markets: automotive (2015). CompositesWorld. Retrieved on 12 October 2015. http:/
/www.compositesworld.com/articles/the-markets-automotive-2015
71. T. Johnson, History of Carbon Fiber: The Growth and Future of Carbon Fiber. Retrieved on 1
September 2015. http://composite.about.com/od/Industry/a/History-Of-Carbon-Fiber.htm
Chapter 28
Experimental and Multiscale Numerical Studies
of Woven Fabric Carbon Composite Cylinder
Subjected to Internal Pressure Loading
28.1 Introduction
Composite materials have been used increasingly for various applications from
sporting equipment to engineering components. Initially, fiber composites were
mostly used for applications where stiffness played an important role. As more
research continued with better understanding of the material behavior, their appli-
cations have been expanded for structural components for which strength is
critical. Especially, carbon fiber composites were the major player for the strength
applications because of their high strength property. There are too many examples
to be mentioned. Examples of large-scale composite structures are commercial
and military aircraft and ships. More recently, more extreme loading applications
have been studied [1–5] where composite cylinders were used to contain explosive
materials.
As composite materials are designed and analyzed for strength application,
reliable damage and failure criteria should be used. First of all, failure criteria
for fibrous composite materials were developed by modifying those for metallic
materials. Because metallic materials are assumed to be isotropic with equal strength
in every direction, the failure criteria for metallic materials are based on stress
invariants, especially of deviatoric stress components. However, fibrous composite
materials have strength which depends on the orientation of the loading relative
to the fiber direction. Because fibers are the major load-carrying materials, the
strength in the fiber direction is much greater than that in the perpendicular direction.
Additionally, the tensile strength in the fiber direction is generally greater than the
compressive strength in the same direction.
Many failure criteria were proposed for fibrous composite materials [6–10].
All those criteria are at the macroscale of the fibrous composite. One failure
criterion proposed in Ref. [11] was a second-order polynomial theory for anisotropic
materials, and that is expressed as
Fi i C Fij i j D 1 i; j D 1; 2; : : : ; 6 (28.1)
where Fi and Fij are the strength parameters to be determined experimentally and
i is the stress in the vector format. For an orthotropic lamina under the plane stress
condition, Eq. (28.1) is rewritten as
in which the strength parameters can be determined from the tensile and compres-
sive strength of the fibrous composite in both the fiber direction called Xt and Xc and
the transverse orientation called Yt and Yc , respectively. In addition, a biaxial test is
also required with the stresses 1 D and 2 D . Then, the strength parameters
are expressed as
1 1
F1 D C (28.3a)
Xt Xc
1 1
F2 D C (28.3b)
Yt Yc
1
F11 D (28.3c)
Xt Xc
1
F22 D (28.3d)
Yt Yc
1
F12 D 2
(28.3e)
S12
1 1 1 1 1 1 1
F12 D 2
1 C C C C C 2 (28.3f)
2 Xt Xc Yt Yc Xt Xc Yt Yc
where S12 is the in-plane shear strength. Later, a revision for the F12 was proposed
[11] such that
.F11 F22 /2
F12 D (28.4)
2
28 Experimental and Multiscale Numerical Studies of Woven Fabric Carbon. . . 851
1 1 1
2
.22 C33 /2 C 2 23 2 22 33 C 2 12 2 13 2 D1 for tensile matrix failure
Yt St Sc
(28.5c)
where 13 and 23 are the out-of-plane shear stresses of the lamina, 33 is the normal
stress in the thickness direction, and St is the transverse shear strength.
The previously mentioned failure criteria are based on the stress components.
Alternatively, strain-based failure criteria were also proposed [12]. The stress-based
and strain-based criteria can be converted from one to the other, and the equivalence
can be shown between them. All those failure criteria required experimental tests at
the laminar level. Even though the same fiber and matrix materials were used, the
experimental tests would be conducted again if there were a change in the fiber
volume fraction. In addition, thermal stresses cannot be properly represented in
the failure criteria. For example, let us consider a unidirectional fibrous composite
sample without any constraint. The fibrous composite sample has a constant
temperature initially. Then, the composite sample is subjected to a uniform heat
flux such that the composite reaches another uniform temperature. Because there
is no constraint, the composite sample will expand or contract freely depending on
the temperature change, and there is no stress at the macroscale level, i.e., at the
composite lamina level. However, there are significant stresses at the microscale
level because of the mismatch in the coefficients of thermal expansion of the fiber
and matrix materials. The stress in one material is in tension, while that in the other
material is in compression of the same magnitude such that they cancel each other
at the macroscale level. Therefore, such stresses at the constituent material level
influence failure loads of the composite sample.
In order to overcome such limitations, new failure criteria were proposed based
on multiscale models. In the new criteria, stresses and strains at the constituent
852 Y.W. Kwon et al.
material level are computed, and the failure criteria are applied to those stresses
or strains. As a result, more physics-based failure criteria can be expressed with
simplified failure modes. The failure criteria based on the constituent material level
are described in the next section.
One of composite applications is a pressure vessel or a pipe which can hold
internal pressure loading. There has been extensive research for composite pipes
under different loading conditions [13–20]. The eventual goal of the present study
is to design carbon fiber composite cylinders to contain high explosive materials.
In order to achieve the goal, it is necessary to be able to test carbon composite
cylinders under internal pressure loading as well as to model and simulate the
behavior accurately. Because explosive testing is expensive and risky, it is desirable
to minimize live fire testing. Therefore, static internal pressure tests were conducted,
and the multiscale modeling was also undertaken to validate the model as the
starting point.
In this study, a simple mechanical device was designed and fabricated so that
internal pressure loading could be applied using a uniaxial loading frame without
using any fluid. The loading device was validated using the finite element analysis
(FEA) of the device as well as experiments using aluminum alloy cylinders. Then,
the device was used to test carbon fiber composite cylinders for their burst pressure.
Furthermore, the multiscale analysis technique developed in Ref. [21–25] was
used to predict the burst strain of the same composite cylinders. The multiscale
model begins with the constituent materials such as fibers and the matrix material.
Failure criteria were also applied at the constituent materials such as fiber failure in
compression or tension, matrix cracking, and fiber/matrix debonding. A composite
cylinder subjected to internal pressure experiences tensile loading along the hoop
direction. Therefore, the fiber tensile strength was used for the present study.
The next section describes how composite cylinders were fabricated using plain
weave composite fabrics. Then, the multiscale analysis model is presented followed
by the detailed description of the new mechanical device for internal pressure
loading. Subsequently, experimental and numerical studies were conducted for both
composite cylinders and tensile test coupons. Finally, conclusions are provided.
Composite cylinders are usually fabricated using the filament winding technique.
Another simple technique is to wrap a woven fabric composite sheet over a
cylindrical object. The latter technique was used in this study. Some detailed
description of the fabrication process is described below.
The mold was made of a hollow aluminum tube with the outer diameter 7.62 cm
(3 in). Two standard metal shelf supports were mounted to a sheet of wood to create
a support base for the cylinder form. A large wood dowel was then inserted into the
aluminum tube, and screws were used to connect the cylinder form to the support
base. A layer of release ply paper and a layer of Teflon sheet were wrapped around
28 Experimental and Multiscale Numerical Studies of Woven Fabric Carbon. . . 853
the aluminum cylinder in order to ensure an easy release after the fabrication. Then,
two rubber couplers were inserted into the aluminum tube so that composite layers
would be placed between the two couplers. Figure 28.1 shows the mold prepared
for the composite cylinder.
Before wrapping the composite fabric, the resin and hardener were thoroughly
mixed in a ratio 4.17 by weight. Following the mixture, the epoxy was applied to
either side of the composite fabric for the portion to be used for the first layer as
well as the mold rig. This initial application ensures complete epoxy coverage for
the first wrap of the fiber strip to minimize voids during the curing process. Next, the
composite fabric was wrapped around the cylindrical mold. Before every layer of
wrapping, epoxy was applied evenly. Maintaining tension on the fiber strip during
the wrapping process is a key to fabricating a uniformly thick cylinder with minimal
voids.
Then, a strip of the perforated release ply and the breather cloth were applied to
the wrapped composites. The breather cloth absorbs any excess epoxy that is pressed
through the perforated release ply. The entire mold cylinder was then removed from
the base, and a spiral-wound line was connected between the mold cylinder and a
vacuum pump. Then, the mold cylinder was inserted into a vacuum bag to take out
air from the wrapped composite.
The partially cured composite cylinder was inspected. In order to ensure a
uniform thickness and the smooth outer surface, a roller was used and a cylindrical
rubber was wrapped tightly around the composite cylinder. The perforated release
ply and breather cloth were again used along with a vacuum bag for final curing.
Following the final cure, the composite cylinder and the mold were removed from
the vacuum bag, breather cloth, and perforated peel ply. Following removal from the
mold, the top and bottom guide and the outer surface form were removed from the
composite cylinder, which resulted in the rough cylinder in its both edges as shown
in Fig. 28.2. The rough edges were trimmed using a rotary cutting tool to make the
final composite cylinders with length of 7.62 cm (3 in).
854 Y.W. Kwon et al.
Fig. 28.2 Rough carbon fiber composite cylinder after removal from mold
Forward Loop
fiber and
finite element
matrix
analysis for a
fiber laminated
Fiber-Strand Strand-Fabric cylinder
Model Model
matrix
Backward Loop
Once the effective material properties of the carbon composite cylinder are
determined, a finite element analysis is conducted for the composite cylinder which
is subjected to given loading and boundary conditions. Even though the eventual
analysis is for explosive loading inside the carbon fiber composite cylinder, the
current study was focused on static internal pressure loading as described later.
There are a couple of different ways to model laminated composite structures.
One is to model the whole layers as a single element. When this is undertaken,
there are two choices. The whole layers are smeared into an equivalent single layer
using the lamination theory [11, 30–32]. The other alternative is to use multiple
numerical integration points through the thickness of the whole layer such that
different material properties at different layers can be considered at those numerical
integration points. The latter is more effective than the former. One drawback of
the both techniques is their difficulty with modeling any delamination between two
neighboring layers, which is a very common damage mode. In order to overcome
this, each layer can be modeled using a separate finite element. However, modeling
each layer using 3-D solid elements is not numerically plausible because the element
aspect ratio would be so huge. Instead, shell elements are used. In order to have shell
856 Y.W. Kwon et al.
elements layered on top of another, 3-D solid-like shell elements would be useful
[33–38]. In other words, the 3-D-like shell element uses the shell formulation but has
displacements as nodal degrees of freedom without rotational degrees of freedom.
Because the nodes are located at the bottom and top surfaces of the shell element,
those shell elements can be stacked one another like 3-D solid elements. However,
the shell element can represent a large aspect ratio of the element. If necessary,
resin layers between two composite layers can be modeled explicitly to represent
potential delamination. Previous studies [39, 40] showed that modeling those resin
layers explicitly was important to predict delamination damage accurately.
The finite element analysis provides deformations, strains, and stresses of the
composite structure. Those strains and stresses are the effective strains and stresses
at each lamina level. If a laminar level failure criterion, as expressed in Eqs. (28.2)
and (28.5), is used, these stresses components are used directly. However, in the
present multiscale analysis technique, those strains and stresses are decomposed to
the strains and stresses at the constituent material level such as fiber and matrix. In
order to accomplish this, the same analytical unit-cell models used in the forward
loop are used and this process is called “backward loop” as shown in Fig. 28.3. The
backward loop is also called the strength loop.
The two analytical models are called fiber-strand and strand-fabric whose
unit-cells are sketched in Fig. 28.4. As stated, the both models utilize unit-cell
models. They compute the effective material properties in the forward loop. In
other words, the fiber-strand model determines the effective material properties
of the unidirectional composites made of fiber and matrix materials, while the
strand-fabric model computes the effective material properties of the woven fabric
composite using the geometric data of the woven fabric and the unidirectional
composite materials. The effective properties of the woven fabric are used for the
finite element analysis of a composite cylinder subjected to a given applied load.
a
3 b
8 6
2 1
7 5 12 8 13
6
4 5 6 9
5
c2 7 5
10 2 11
2
1
c 3 1
1
1 3
a
2
a
b b1 1
2
Fig. 28.4 Unit-cell models for (a) fiber-strand and (b) strand-fabric models
28 Experimental and Multiscale Numerical Studies of Woven Fabric Carbon. . . 857
If an analytical solution is available, it can be used instead. Either the finite element
analysis or analytical solution gives macro-level stresses and strains in the composite
cylinder. These macro-level stresses and strains are decomposed to the microlevel
stresses and strains at the fiber and matrix level using the same two models through
the backward loop.
The failure of the cylinder is determined when the fiber reaches the tensile
strength of the material. The failure criteria at the constituent material level are
expressed as below. The carbon fiber failure is tested using the following equation:
f D ft if f > 0 (28.6a)
ˇ ˇ
ˇf ˇ D fc if f < 0 (28.6b)
where f is the stress which the fiber carries along the fiber direction and ft and fc
are the fiber strength in tension and compression, respectively. The compressional
strength also considers fiber micro-buckling [41, 42]. For the resin material, the
principal strain criterion is used as shown below:
in which ("m )max and ("m )min are the maximum and minimum principal strains of
the matrix material and et and ec are the failure strain in tension and compression,
respectively.
To save space, only some details of the fiber-strand model are presented
below, while the strand-fabric model is omitted because the basic assumptions
and formulations are similar between the two modules. Details of the modules are
presented in Ref. [21–25].
The unit-cell model for the fiber-strand model is comprised of a rectangular prism
divided into eight quadrants or subcells as shown in Fig. 28.4. Each subcell can be
identified as a different material, and the unit-cell model can solve for equivalent
composite elastic moduli, Poisson’s ratio, and the coefficient of thermal expansion.
The following equations define the equilibrium stress state at the interfaces of the
subcells:
1 2 3 4 5 6 7 8
11 D 11 11 D 11 11 D 11 11 D 11 (28.7a)
1 3 2 4 5 7 6 8
22 D 22 22 D 22 22 D 22 22 D 22 (28.7b)
1 5 2 6 3 7 4 8
33 D 33 33 D 33 33 D 33 33 D 33 (28.7c)
858 Y.W. Kwon et al.
1 2 3 4 5 6 7 8
12 D 12 D 12 D 12 12 D 12 D 12 D 12 (28.8a)
1 2 5 6 3 4 7 8
13 D 13 D 13 D 13 13 D 13 D 13 D 13 (28.8b)
1 3 5 7 2 4 6 8
23 D 23 D 23 D 23 23 D 23 D 23 D 23 (28.8c)
Here, the first three sets of equations are for the normal stresses, and the other three
sets are for the shear stresses.
The subscripts denote stress components according to the axis in Fig. 28.4,
while the superscripts indicate subcell number. In addition to stress equilibrium,
deformation compatibility must be met for the unit cell. Deformation compatibility
defines the following set of equations:
"nij D Cijkl
n
kln C ˛ijn (28.11)
Because most of fiber and matrix materials are either isotropic or orthotropic, the
normal and shear components are decoupled. As a result, the two components can be
worked out independently. The mathematical development of the two components is
very similar. Therefore, only the normal components are discussed here. Arranging
above equations for the normal components yields the following equation:
where [T] is a 24 24 matrix with the combination of three vectors containing stress
relationships, strain relationships, and constitutive equations:
The inverse of [T] can be further broken down into three submatrices. These can
be expressed like the following:
As Eqs. (28.11) and (28.19) are substituted into Eq. (28.20), the following unit-
cell stress expression is found:
where [E] is a matrix of the inverse of the subcell compliance tensors stated in
Eq. (28.11). Then, the unit-cell stiffness can be calculated:
E D ŒV ŒE ŒR3 (28.22)
860 Y.W. Kwon et al.
The [Ē] matrix found through Eq. (28.22) is the 3 3 matrix of the unit-cell
stiffness. With these values, Poisson’s ratio of the unit cell can also be found. A
similar method can be utilized to solve for the shear properties of the composite
material.
For the strength loop, Eq. (28.17) is used to compute constituent material level
strains from the composite level strains. Then, Eq. (28.11) is also used to compute
the stresses in the fiber and matrix materials. Hence, this module can complete both
forward and backward loops.
Before conducting explosive testing for carbon composite cylinders, static internal
pressure loading was considered for the composite cylinders. As a result, a
mechanical device was developed to apply internal pressure loading to a composite
cylinder instead of using pressurized fluids. The device is sketched in Fig. 28.5, and
it consists of multiple parts. The two upper and bottom rams are used to apply the
axial loading using a uniaxial mechanical testing machine. The eight equal sizes of
wedges are utilized to transfer the axial loading to the radial loading like internal
pressure loading. The guide pin is used to maintain the rams in the collinear axis. A
shim may be used between the composite cylinder and the wedges if there is a small
gap in the beginning before applying the load.
Figures 28.6 and 28.7 show the free body diagrams for the ram and wedge.
After applying the equilibrium equations and assuming Coulomb’s friction law, the
internal pressure pi is related to the axial load F as below:
F tan
pi D (28.23)
aL 1 C tan
Pc
where is the coefficient of friction, is the angle shown in Figs. 28.6 and 28.7,
and a and L are the outer radius and the length of the wedge. The present wedge has
the angle D 80ı
For an axisymmetric cylinder, the hoop strains at the inner and outer surfaces of
the cylinder are written as
a2 pi b2
"ih D .1 / C .1 C / (28.24)
E .b2 a2 / a2
2a2 pi
"oh D (28.25)
E .b2 a2 /
862 Y.W. Kwon et al.
in which a, b, and E are the inner and outer radii and the elastic modulus of the
cylinder. From Eqs. (28.23) to (28.25), we obtain the following expression:
2aF tan
"oh D (28.26)
E .b2 a2 / L 1 C tan
and
aF b2 tan
"ih D .1 / C .1 C / (28.27)
E .b2 a2 / L a2 1 C tan
As the thickness of the cylinder becomes much smaller than the cylinder radius, Eq.
(28.28) is simplified as
F tan
hi D (28.29)
EtL 1 C tan
The material used for the device should be very stiff and strong such that the
device has a minimal deformation and does not yield during the test. Of course, this
requirement is relative to the cylinder material. To this end, the 1704 PH stainless
steel was selected for the device material. In order to validate the application
of the device for internal pressure loading, a finite element analysis (FEA) was
conducted for the device with a 6061 T6 aluminum alloy cylinder which has the
height 7.62 cm (3 in), the inside diameter 7.62 cm (3 in), and the wall thickness
0.3175 cm (0.125 in). In the finite element model, the bottom ram was fixed, while
the top ram was pushed down for a specific distance. The distance was determined
using the expected hoop strain in the cylinder assuming the rams and wedges were
rigid. This expected strain was called the target hoop strain which was ranging from
0.001 to 0.01. Because the rams and wedges are not rigid materials, there should
be some difference between the target strain and the FEA strain. For this study, the
coefficient of fiction was assumed 0.01. The stainless steel has the elastic modulus
200 GPa and Poisson’s ratio 0.28, while the aluminum alloy has the elastic modulus
68.9 GPa and Poisson’s ratio 0.33. Since elastic analysis was conducted for FEA, the
stress level was lower than yield strength of both materials. Table 28.1 compares the
target strain and the FEA strain at different strain levels. The difference is about 7 %.
Because of the deformation in the rams and wedges, the FEA strains were smaller
than the target strain. If there is no measurement of strain, the target strain may be
used as a conservative representation of the actual strain. However, in the following
studies, a hoop strain was measured using a strain gauge (Tables 28.2 and 28.3).
28 Experimental and Multiscale Numerical Studies of Woven Fabric Carbon. . . 863
The multiscale analysis technique has been validated against many experimental
data, especially for the effective material properties at different length scales. For
example, effective material properties of the fibrous and woven fabric composites
were studied. Table 28.2 shows the material properties of the fiber and matrix mate-
rials [43]. Then, the effective material properties of the unidirectional composite
made of the fiber and matrix materials were computed using the fiber-strand unit-cell
model and compared to the data in Ref. [43]. Both effective stiffness and strength
results agreed very well each other [23]. The next example examined a plain weave
composite made of carbon/epoxy materials as given in Table 28.4 [44] using the
strand-fabric unit-cell model. The computed effective stiffness of the plain weave
fabric was compared to the experimental data [23, 44, 45]. The predicted solution
agreed well with the published data.
Then, the whole multiscale analysis technique was applied to the present carbon
composite cylinders. Before applying the multiscale analysis technique to carbon
composite cylinders, uniaxial tensile coupons were considered. The test coupons
were fabricated using the same plain weave composites made of carbon fibers. The
purpose of these coupons and tensile tests was to check the material properties
of the composite. Both tensile coupons and cylinders had woven layers 0ı /90ı so
that the laminated composites were quasi-isotropic. Figure 28.8 shows the carbon
composite tensile test coupons after failure. A set of strain gauges were attached to
each test coupon, one in the axial direction and the other in the transverse direction.
Figure 28.9 shows a typical stress-strain plot of the carbon composite coupon.
864 Y.W. Kwon et al.
The elastic modulus was determined from the stress-strain curve and it is 40.0 GPa
and the Poisson’s ratio is 0.14. The stress at failure is 500 MPa, and the specimen
fractured at the maximum strain of 0.0135 m/m (Table 28.5).
The multiscale analysis model depicted in Fig. 28.3 was used to predict the
material behavior of the tensile test coupon. Both fiber-strand and strand-fabric
models were used one after the other sequentially. The multiscale analysis requires
fiber and matrix material properties as well as their volume fractions and geometric
data of the plain weave architecture. The elastic modulus of the carbon fiber is
230 GPa and Poisson’s ratio is 0.2. The carbon fiber has the tensile strength
28 Experimental and Multiscale Numerical Studies of Woven Fabric Carbon. . . 865
3530 MPa and the fracture strain 0.015 m/m. On the other hand, the elastic modulus
of the matrix material is 1.64 GPa, and Poisson’s ratio is 0.33. The fiber volume
fraction for the unidirectional model used for the fiber-strand model was assumed to
be 0.35. The plain weave fabric has the geometric data a D 4 mm, b D 1.5 mm, and
t D 0.25 mm as sketched for a single tow strand in Fig. 28.10. Some of these data
were measured, while others were obtained from the manufacture specifications.
The multiscale analysis computed the effective elastic modulus and the fracture
strain of the carbon composite tensile coupon. First of all, the fiber-strand model
resulted in the effective modulus of the unidirectional fibrous composite equal
to 81.6 GPa, and the effective modulus of the woven fabric was 39.9 GPa. The
Poisson’s ratio of the fabric was 0.014. Both of these values agreed well with
the tensile test data. The fracture strain was 0.013 m/m. The fracture strain was
computed when the carbon fibers reached their failure strength as provided above.
The predicted values were in good agreement with the measured data. This gave a
good confirmation that the constituent material properties as well as other geometric
data were reasonable. Then, the same data were used for the subsequent study for
composite cylinder subjected to internal pressure loading.
Carbon composite cylinders were tested using the mechanical device described
previously. In order to correlate the applied axial force F to internal pressure,
the coefficient of friction is needed. To determine the coefficient, an aluminum
alloy cylinder was used, while strain gauges were attached to the aluminum alloy
cylinder. The applied force and the hoop strain at the outside surface of the cylinder
were substituted to Eq. (28.26) to determine the coefficient of friction between the
rams and wedges. The coefficient of friction was 0.12 from the test. Figure 28.11
compares the experimental data from the aluminum cylinder and the result from Eq.
(28.26) with the coefficient of friction 0.12. They agreed very well as long as the
cylinder deformed in the linear elastic range. Once there was plastic deformation,
Eq. (28.26) is not valid any more. However, this is not a concern for the carbon fiber
composite cylinder because the carbon composite behaves linearly until fracture.
Furthermore, it is a reasonable assumption that the coefficient of friction remains
constant independent of the loading magnitude as long as the same lubricant is used.
866 Y.W. Kwon et al.
Experiments were undertaken for three composite cylinders whose inner radius
is 7.62 cm (3 in). There is some variation in thickness with the average wall
thickness 2.5 mm. A typical load vs. strain curves are plotted for a carbon composite
cylinder in Fig. 28.12. Three strain gauges were attached to the cylinder along
the circumferential direction. The whole set of wedges consists of eight pieces.
As a result, the wedges are piecewise continuous along the circumference of the
cylinder. As the wedges push out the cylinder, the cylinder locations corresponding
to gaps between two neighboring wedges experience less strains compared to other
28 Experimental and Multiscale Numerical Studies of Woven Fabric Carbon. . . 867
locations. Figure 28.12 confirms this observation. In the figure, the hoop strain
gauge #1 was placed at the cylinder location between two wedges, while the other
two hoop strain gauges were placed at the cylinder locations corresponding to the
middle of the wedges. Therefore, strain gauge #1 yielded a lower strain than the
others which were very close.
Because strain gauges #2 and 3 represent the response of the cylinder under
internal pressure loading better than gauge #1, the failure load and strain were
obtained from the two strain gauges. The failure load was 70 kN and the fracture
hoop strain was 0.010. Those values were compared to the results from the
multiscale analysis. The initial results showed that the predicted failure strain was
about 25 % lower than the experimental value. The initial model assumed a uniform
wall thickness. However, the actual measurement of the cylinders showed that the
wall thickness varied 6 % around the cylinder. Because the cylinder was fabricated
as shown in Fig. 28.1, resins flew down during the curing process. As a result, the
bottom side of the cylinder was thicker than the top side in Fig. 28.1. Considering
this, a new FEA was conducted with a gradual change in thickness along the
hoop direction, and the multiscale analysis was also conducted along with the new
FEA model. The predicted failure hoop strain was 0.0098 m/m, and the predicted
maximum force was 76 kN which was equivalent to the internal pressure 13 MPa.
These numerical values were in good agreement with the experimental results of the
composite cylinders.
28.6 Conclusions
A new mechanical device consisting of rams and wedges were designed and
manufactured so that internal pressure loading can be applied to the inside of a
cylinder without using any fluid. Then, carbon composite cylinders were tested
using the new mechanical device to determine their failure load/pressure and the
fracture strain. The experimentally measured values were compared to the values
predicted using the multiscale analysis model. The two results agreed very well. One
parameter is required in the mechanical device, which is the coefficient of friction
between the rams and wedges. The coefficient was obtained from an experimental
study on an aluminum alloy cylinder by measuring the applied axial force and the
hoop strain on the cylinder. For the current device, the coefficient was 0.12.
The success of the developed mechanical device for internal pressure loading
as well as the reliable multiscale analysis technique will lead to initial designs of
carbon composite cylinders containing high explosive materials. Then, limited tests
will be conducted for the composite cylinders with the high explosive loading to
refine the design. Eventually, carbon fiber composites will be designed and used
for extreme loading environment because of high strength and stiffness with low
density.
868 Y.W. Kwon et al.
References
1. A.M. Clayton, A simplified method to determine initial estimates of peak strains in composite
explosive containment vessels. ASME Paper No. PVP2013-97068, 2013 ASME Pressure
Vessels and Piping Conference, Paris, France, 14–18 July 2013
2. A.G. Fedorenko, M.A. Syrunin, A.G. Ivanov, Criterion for selecting composite materials for
explosion containment structures (review). Combust. Explos. Shock Waves 41(5), 487–495
(2005)
3. J.W. Pastrnak, C.D. Henning, W. Grundler, Composite vessels for containment of extreme
blast loading. Lawrence Livermore National Laboratory, Technical Report No. UCRL-CONF-
205423 (2004)
4. Q. Dong, Q.M. Li, J.Y. Zheng, B.Y. Hu, Effects of structural perturbations on strain growth in
containment vessels. ASME J. Press. Vessel Technol. 132, 011203 (2010)
5. Q. Dong, B. Hu, Dynamic behavior of carbon fiber explosion containment vessels. ASME J.
Press. Vessel Technol. 138, 011292 (2016)
6. Z. Hashin, Failure criteria for unidirectional fiber composites. J. Appl. Mech. 47, 329–334
(1980)
7. G.P. Sendeckyi, A brief survey of empirical multiaxial strength criteria for composites.
Composite Materials: Testing and Design, ASTM STP 497 (1972), pp. 41–51
8. E.M. Wu, Phenomenological anisotropic failure criterion, ed. by G.P. Sendeckyj, in Composite
Materials Volume 2, Mechanics of Composite Materials (Academic Press, New York, 1974),
pp. 253–431
9. C.C. Chamis, Failure criteria for filamentatry composites. Composite Materials: Testing and
Design, ASTM STP 460 (1969), pp. 336–351
10. B.E. Kaminski, R.B. Lantz, Strength theories of failure for anisotropic materials. Composite
Materials: Testing and Design, ASTM STP 460 (1969), pp. 160–169
11. S.W. Tsai, H.T. Hahn, Introduction to Composite Materials (Technomic Publishing Co.,
Lancaster, 1980)
12. M.E. Waddoups, Advanced composite material mechanics for the design and stress analyst.
General Dynamics, Fort Worth Division Report FZM-4763 (1967)
13. M. Xia, K. Kemmochi, H. Takayanagi, Analysis of filament-wound fiber-reinforced sandwich
pipe under combined internal pressure and thermomechanical loading. Compos. Struct. 5,
273–283 (2001)
14. P.D. Soden, R. Kitching, P.C. Tse, Experimental failure stresses for ˙55ı filament wound glass
fiber reinforced plastic tubes under biaxial loads. Composites 20, 125–135 (1989)
15. J. Mistry, A.G. Gibson, Y.-S. Wu, Failure of composite cylinders under combined external
pressure and axial loading. Compos. Struct. 22, 193–200 (1992)
16. M. Uemura, H. Fukunaga, Probabilistic burst strength of filament-wound cylinders under
internal pressure. J. Compos. Mater. 15, 462–480 (1981)
17. A. Onder, O. Sayman, T. Dogan, N. Tarakcioglu, Burst failure load of composite pressure
vessels. Compos. Struct. 89, 159–166 (2009)
18. M. Xia, H. Takayanagi, K. Kemmochi, Analysis of multi-layered filament-wound composite
pipes under internal pressure. Compos. Struct. 53, 483–491 (2001)
19. J. Xing, P. Geng, T. Yang, Stress and deformation of multiple winding angle hybrid filament-
wound thick cylinder under axial loading and internal and external pressure. Compos. Struct.
131(4), 866–877 (2015)
20. T. Hwang, J. Park, H. Kim, Evaluation of fiber material properties in filament-wound composite
pressure vessels. Compos. Part A 43, 1467–1475 (2012)
21. Y.W. Kwon, Multiphysics and Multiscale Modeling: Techniques and Applications (CRC Press,
Boca Raton, 2016)
22. Y.W. Kwon, Multi-level approach for failure in woven fabric composites. Adv. Eng. Mater.
3(9), 713–717 (2001) (Invited Paper)
28 Experimental and Multiscale Numerical Studies of Woven Fabric Carbon. . . 869
23. Y.W. Kwon, A. Altekin, Multi-level, micro-macro approach for analysis of woven fabric
composites. J. Compos. Mater. 36(8), 1005–1022 (2002)
24. Y.W. Kwon, M.S. Park, Versatile micromechanics model for multiscale analysis of composite
structures. Appl. Compos. Mater. 20(4), 673–692 (2013)
25. M.S. Park, Y.W. Kwon, Elastoplastic micromechanics model for multiscale analysis of metal
matrix composite structures. Comput. Struct. 123, 28–38 (2013)
26. Y.W. Kwon, J.M. Berner, Micromechanics model for damage and failure analyses of laminated
fibrous composites. Eng. Fract. Mech. 52(2), 231–242 (1995)
27. Y.W. Kwon, C.T. Liu, Study of damage evolution in composites using damage mechanics and
micromechanics. Compos. Struct. 38(1–4), 133–139 (1997)
28. Y.W. Kwon, C. Kim, Micromechanical model for thermal analysis of particulate and fibrous
composites. J. Therm. Stresses 21, 21–39 (1998)
29. Y.W. Kwon, J.H. Lee, C.T. Liu, Modeling and simulation of crack initiation and growth in
particulate composites. Trans. ASME J. Press. Vessel Technol. 119(3), 319–324 (1997)
30. R.F. Gibson, Principles of Composite Material Mechanics (McGraw-Hill, New York, 1994)
31. B.D. Agarwal, L.J. Broutman, Analysis and Performance of Fiber Composites (Wiley, New
York, 1980)
32. R.M. Jones, Mechanics of Composite Materials, 2nd edn. (Taylor & Francis, Philiadelphia,
1999)
33. D.R.J. Owen, Z.H. Li, A refined analysis of laminated plates by finite element displacement
methods—I: fundamentals and static analysis. Comput. Struct. 26, 907–914 (1987)
34. Y.W. Kwon, Finite element analysis of crack closure in plate bending. Comput. Struct. 32,
1439–1445 (1989)
35. G.M. Kulikov, S.V. Plotnikova, Finite rotation geometrically exact four-node solid-shell
element with seven displacement degrees of freedom. Comput. Model. Eng. Sci. 28, 15–38
(2008)
36. S. Klinkel, F. Gruttmann, W. Wagner, A continuum based three-dimensional shell element for
laminated structures. Compos. Struct. 71, 43–62 (1999)
37. R. Hauptmann, K. Schweizerhof, A systematic development of ‘sold-shell’ element formula-
tions for linear and non-linear analyses employing only displacement degrees of freedom. Int.
J. Numer. Meth. Eng. 42, 49–69 (1998)
38. Y.W. Kwon, Analysis of laminated and sandwich composite structures using solid-like shell
elements. Appl. Compos. Mater. 20(4), 355–373 (2013)
39. M.F. Boseman, Y.W. Kwon, D. Loup, E. Rasmussen, Interface fracture of hybrid joint of glass-
/steel-fiber composites. Eng. Comput. 29(5), 504–527 (2012)
40. Y.W. Kwon, L.E. Craugh, Progressive failure modeling in notched cross-ply fibrous compos-
ites. Appl. Compos. Mater. 8(1), 63–74 (2001)
41. B.W. Rosen (ed.), Fiber Composite Materials (American Society for Metals, Metals Park,
1965)
42. C. Soutis, P.T. Curtis, A method for predicting the fracture toughness of CFRP laminates failing
by fibre microbuckling. Compos. Part A Appl. Sci. Manuf. 31, 733–740 (2000)
43. D.M. Blackketter, D.E. Walrath, A.C. Hansen, Modeling damage in a plain weave fabric-
reinforced composite material. J. Compos. Technol. Res. 15(2), 136–142 (1993)
44. T. Ishikawa, M. Matsushima, Y. Hayashi, T.W. Chou, Experimental confirmation of the theory
of elastic moduli of fabric composites. J. Compos. Mater. 19(5), 443–458 (1985)
45. V.R. Aitharaju, R.C. Averill, Three-dimensional properties of woven-fabric composites. Com-
pos. Sci. Technol. 59, 1901–1911 (1999)
Part VIII
Chapter 29
Fatigue of 2D and 3D Carbon-Fiber-Reinforced
Polymer Matrix Composites and of a Unitized
Polymer/Ceramic Matrix Composite at Elevated
Temperature
29.1 Introduction
Airframe structures and components on many existing and future aerospace systems
require operation in elevated temperature environments. Examples include hyper-
sonic vehicle airframes, engine-related components (such as engine ducts, engine
vanes, and exhaust flaps), and hot trailing edges of B-2 and C-17 wings. Material
systems that show improved fatigue performance, excellent thermal resistance and
damage tolerance, as well as resistance to corrosion are prime candidates for air
vehicle structural components.
Laminated composite materials consisting of high-temperature thermosetting
polyimide resins reinforced with carbon fibers have emerged as an important class
of aerospace materials which combine light weight with damage tolerance. Polymer
matrix composites (PMCs) with advanced polyimide matrices that offer improved
high-temperature capability are currently being considered for applications at
temperatures previously meant exclusively for metals [1]. PMR-15 thermosetting
polyimide developed by NASA [2] is recognized as a state-of-the-art resin used
as the matrix for high-temperature carbon-fiber-reinforced composites. However,
replacement polyimide resins are being developed partly due to the carcinogenic
elements in PMR-15. One such polyimide is NRPE resin developed by Performance
Polymer Solutions Inc. (P2 SI/PROOF Research, Moraine, OH, USA). For success-
ful application of NRPE/carbon-fiber composites in advanced aerospace systems, it
The views expressed are those of the authors and do not reflect the official policy or position of the
US Air Force, the Department of Defense, or the US Government.
M.B. Ruggles-Wrenn () • M.P. Wilkinson
Air Force Institute of Technology, Wright-Patterson Air Force Base, OH 45433-7765, USA
e-mail: marina.ruggles-wrenn@afit.edu
reinforced with carbon fibers and of the unitized PMC/CMC. Additionally, this
effort aims to investigate the tension–tension fatigue performance of the aforemen-
tioned material systems at elevated temperature.
Three material systems were studied in this effort: a 3D PMC, a 2D PMC, and a
unitized composite consisting of a PMC and a CMC.
®
Material system 1 (MS1) is a 3D PMC consisting of a P2 SI NRPE polyimide
®
matrix reinforced with carbon fibers in a non-crimp 3D orthogonal weave. P2 SI
2
NRPE, developed by Performance Polymer Solutions Inc. (P SI/PROOF Research,
Moraine, OH, USA), is a high-temperature, structural thermosetting polyimide
®
with a low-melt viscosity. P2 SI NRPE resin is designed to maintain its struc-
tural integrity after continuous exposures at temperatures up to 343 ı C. The 3D
fiber weave structure consists of Z-yarns that interlace warp yarns with multiple
insertions of fill yarns in the cross direction. A schematic of a non-crimp 3D
orthogonal weave and an optical micrograph of the as-processed 3D PMC are
shown in Fig. 29.1a, b, respectively. The Grafil 34-700WD carbon fibers were used
as warp and fill fibers and AS4 carbon fibers, as the Z-fibers. Details of the 3D
carbon fiber fabric are summarized in Table 29.1. Material system 2 (MS2) is a
®
2D PMC consisting of a P2 SI NRPE polyimide matrix reinforced with 15 plies
Fig. 29.1 Material system 1—3D PMC. (a) Schematic of a non-crimp 3D orthogonal weave. Warp
yarns are shown in red, fill yarns in yellow, and Z-yarns in blue. (b) Optical micrograph of the as-
processed 3D PMC with 0/90 fiber orientation
876 M.B. Ruggles-Wrenn and M.P. Wilkinson
Table 29.2 Physical properties of the 3D PMC (MS1) and 2D PMC (MS2) panels
Fiber Resin
volume volume Void volume Density
Panel ID Resin content (%) fraction (%) fraction (%) fraction (%) (g/cc)
3D PMC
MS1-1 39.10 52.87 45.26 1.87 1.563
MS1-2 43.35 49.18 50.17 0.64 1.563
2D PMC
MS2-1 36.44 55.29 42.26 2.45 1.566
MS2-2 36.52 55.23 42.36 2.41 1.566
Data provided by Performance Polymer Solutions Inc. (P2 SI/PROOF Research,
Moraine, OH, USA)
Fig. 29.2 Material system 3—2D PMC/CMC: (a) PMC face, (b) CMC face, and (c) side view
a thermal barrier for the PMC. Both the PMC and the CMC parts are reinforced
with a 2D fabric woven in an 8HSW. However, the matrix and reinforcement
®
materials differ. The PMC part consists of a P2 SI NRPE matrix reinforced with
12 plies of de-sized Cytec T650-35 carbon fiber fabric. The CMC part consists of
®
a ceramic matrix reinforced with 3 plies of 1059 HT-sized JPS Astroquartz III
4581 fabric. The C5 ceramic matrix, developed by Performance Polymer Solutions
Inc. (P2 SI/PROOF Research, Moraine, OH, USA), was produced by blending KDT
HTT 1800 polysilazane-based preceramic resin with yttria-stabilized zirconia and
silica additives. The co-curing process used to fabricate the unitized PMC/CMC
composite is proprietary. Because the 2D PMC/CMC consists of two dissimilar
materials, physical properties of PMC/CMC panels such as constituent content
percentages could not be readily measured. The overall thickness of the PMC/CMC
specimens was approximately 5 mm. The average thickness of the PMC part
was approximately 4 mm, while the average thickness of the CMC part was
approximately 1 mm. Hence the plane of PMC/CMC co-curing is not located at
the midplane of the 2D PMC/CMC composite panel.
6.0
3.0
1.0569
0.80 0.30
0.35
+ R = 1.1244
with distilled water to remove contaminants from the machining process. Specimens
were then dried in a vacuum oven at 105 ı C at 1.1 in. of mercury for 10 days and
subsequently stored in a dry air-purged desiccator until testing.
A servo-controlled MTS mechanical testing machine equipped with hydraulic
water-cooled wedge grips, a compact resistance-heated furnace, and a temperature
controller was used in all tests. An MTS FlexTest 40 digital controller was
employed for input signal generation and data acquisition. Strain measurement was
accomplished with an MTS high-temperature air-cooled uniaxial extensometer of
12.5–mm gage length. In all elevated temperature tests, elevated temperature was
defined as 329 ı C on the right side of the specimen (Tright ) with the other side open
to ambient laboratory air. Additional insulation inserts were designed consistent
with the specimen geometry in order to ensure that only one side (the right side)
of the test specimen was exposed to elevated temperature with the other side (the
left side) being exposed to ambient air. Figure 29.4 shows furnace insulation setup
in preparation for testing. For elevated temperature testing, K-type thermocouples
were attached to the specimen gage section to calibrate the furnace on a periodic
basis. The furnace controller (using a noncontacting thermocouple exposed to the
ambient environment near the test specimen) was adjusted to determine the setting
needed to achieve the desired temperature of the test specimen. The calibration
procedure was performed and the power setting was determined for each material
system. The determined power setting was then used in actual tests. In all elevated
temperature tests, a specimen was heated to test temperature at a rate of 10 ı C/min
and held at temperature for additional 45 min prior to testing. Tested specimens
were examined using an optical microscope (Zeiss Discovery V12).
Fig. 29.4 Furnace insulation setup: (a) without insulation inserts, (b) with insulation inserts, (c)
with test specimen gripped, and (d) with added insulation blocks—specimen ready for testing
29 Fatigue of 2D and 3D Carbon-Fiber-Reinforced Polymer Matrix. . . 879
1000
MS2, 2D PMC, 0/90
T = 23ºC
800
600
0
0.0 1.0 2.0 3.0 4.0 5.0 6.0 7.0
Strain (%)
Fig. 29.5 Tensile stress–strain curves obtained at 23 ı C for the 3D PMC, 2D PMC, and 2D
PMC/CMC with 0/90 and ˙45 fiber orientations
The elastic modulus (stiffness) of the 0/90 3D PMC is about 20 % lower than that
of the 0/90 2D PMC. The 0/90 ultimate tensile strength (UTS) of the 3D PMC is
some 8 % lower than the 0/90 UTS of the 2D PMC. The stiffness and strength of
the 3D PMC are likely reduced by the presence of the Z-fibers [5]. The reductions
in stiffness and strength of the 3D PMC may be attributed to microstructural
defects introduced along with the Z-fibers during processing, with stiffness being
particularly affected by geometric defects (such as fiber waviness) and lower tensile
strength being caused by clusters of broken and damaged fibers. The tensile stress–
strain behavior of the 3D PMC and of the 2D PMC with 0/90 fiber orientation is fiber
dominated and nearly linear to failure. Sudden failures occur at strains near 1.5 %.
The deformation of the 3D PMC and of the 2D PMC with 0/90 fiber orientation
under tensile loading is mainly elastic; the final composite failure is most likely
triggered by sudden fiber failure.
The elastic modulus of the unitized composite 2D PMC/CMC with 0/90 fiber
orientation is nearly the same as that of the 0/90 2D PMC. However, the 0/90
UTS of the unitized 2D PMC/CMC is nearly 23 % lower than the 0/90 UTS
of the 2D PMC. In order to understand the reduction in tensile strength of the
unitized 2D PMC/CMC, we consider the stress–strain behavior and progressive
delamination of this material during tension test. The stress–strain behavior of the
0/90 2D PMC/CMC becomes markedly nonlinear as the stress exceeds 300 MPa.
Approximately at this point in tension test, delamination in the gage section of
the test specimen first becomes visible. Delamination of the 0/90 2D PMC/CMC
continues and becomes severe as the tensile load increases; the composite plies
pull apart. The PMC plies and the CMC plies bow out on their respective sides
(Fig. 29.6). The CMC ply closest to the plane of co-curing with the PMC remains
29 Fatigue of 2D and 3D Carbon-Fiber-Reinforced Polymer Matrix. . . 881
Fig. 29.6 Optical micrograph of a 2D PMC/CMC specimen with 0/90 fiber orientation showing
nonhomogeneous deformation during tension test at 23 ı C. Note severe delamination and bowing
out of the PMC and the CMC plies. The CMC ply closest to the plane of co-curing remains
relatively planar
relatively planar, suggesting that co-curing may have increased the stiffness of
that particular CMC ply. Because the combined thickness of the PMC plies
considerably exceeds that of the CMC plies, the plane of co-curing is not located
at the midplane of the specimen. Hence the bowing out of the PMC and the
CMC plies is not symmetric with respect to the specimen midplane. It is likely
that this nonhomogeneous deformation causes bending stresses in addition to the
applied tensile stress. The reduction in tensile strength of the 2D PMC/CMC is
attributed to severe delamination observed during tension test. Note that extensive
delamination and nonhomogeneous deformation were observed during all tests of
the 2D PMC/CMC with 0/90 fiber orientation performed in this work.
For the ˙45 fiber orientation, the stress–strain behavior of the three composites
is matrix dominated. Delayed failures are observed, resulting in significantly higher
failure strains. Stress–strain curves show early departure from linearity and gradual
flattening with increasing stress levels. The failure behavior of the ˙45 specimens
is likely dominated by failure processes within the matrix and by the bonding
strength of the fiber–matrix interface; early mixed-mode ply failures occur, followed
by tow “scissoring” (i.e., considerable change in the orientation of the fiber tows
as the loading continues). The mechanical response is nonlinear and likely time
dependent. The 2D PMC/CMC specimens with ˙45 fiber orientation exhibit
extensive delamination and nonhomogeneous deformation during tension test as
did the 0/90 specimens. The plies bow out as the tensile load increases (Fig. 29.7).
However, in the case of the ˙45 specimens, the CMC ply closest to the plane of co-
curing does not remain planar, but bows out as well. Furthermore, the delamination
and the nonhomogeneous deformation are generally confined to the specimen gage
section.
For the 3D PMC and the 2D PMC/CMC, the 0/90 UTS values are an order of
magnitude higher than the corresponding ˙45 UTS values. In the case of the 2D
PMC, the 0/90 UTS value is approximately five times the ˙45 UTS value. The
0/90 stiffness values obtained for all three composites are 4–5 times higher than the
corresponding ˙45 stiffness values. Low off-axis strength and stiffness values are
not surprising considering that the three composites do not contain off-axis fibers in
their reinforcement architecture. The 2D PMC produced the highest ˙45 UTS value
882 M.B. Ruggles-Wrenn and M.P. Wilkinson
Fig. 29.7 Optical micrograph of a 2D PMC/CMC specimen with ˙45 fiber orientation showing
nonhomogeneous deformation during tension test at 23 ı C. Note delamination and bowing out of
the PMC and the CMC plies, including the CMC ply closest to the plane of co-curing
1000
0/90
T = 23°C
MS1, 3D PMC
800
Stress (MPa)
600
0/90
Tright = 329°C
400
±45 ±45
T = 23°C Tright = 329°C
200
0
0.0 1.0 2.0 3.0 4.0 5.0 6.0
Strain (%)
Fig. 29.8 Tensile stress–strain curves obtained for the 3D PMC with 0/90 and ˙45 fiber
orientations at 23 ı C and at elevated temperature
followed by those produced by the 3D PMC and the 2D PMC/CMC. For the ˙45
fiber orientation, the UTS of the 2D PMC is more than two times the UTS of the 3D
PMC and three times the UTS of the 2D PMC/CMC. The 2D PMC also produced
the highest ˙45 stiffness value, while the ˙45 stiffness values obtained for the 3D
PMC and 2D PMC/CMC were close. The average failure strain obtained for the 2D
PMC is approximately three times that obtained for the 3D PMC and nearly four
times that obtained for the 2D PMC/CMC.
As seen in Figs. 29.8, 29.9, and 29.10, the stress–strain response of the three
composites with 0/90 fiber orientation is little affected by the test temperature.
The 0/90 stress–strain curves obtained at elevated temperature (Tright D 329 ı C) are
nearly the same as those produced at 23 ı C. Likewise the 0/90 UTS and stiffness
values obtained at elevated temperature are no more than 2–3 % lower than the
room-temperature values. The elevated temperature also had little influence on the
29 Fatigue of 2D and 3D Carbon-Fiber-Reinforced Polymer Matrix. . . 883
1000
0/90 MS2, 2D PMC
T = 23°C
800
0/90
Stress (MPa)
Tright = 329°C
600
±45 ±45
400 Tright = 329°C
T = 23°C
200
0
0.0 2.0 4.0 6.0 8.0 10.0 12.0 14.0
Strain (%)
Fig. 29.9 Tensile stress–strain curves obtained for the 2D PMC with 0/90 and ˙45 fiber
orientations at 23 ı C and at elevated temperature
1000
MS3, 2D PMC/CMC
0/90
800 Tright = 329°C
Stress (MPa)
600
0/90
T = 23°C
400
±45
Tright = 329°C
±45
200 T = 23°C
0
0.0 1.0 2.0 3.0 4.0
Strain (%)
Fig. 29.10 Tensile stress–strain curves obtained for the 2D PMC/CMC with 0/90 and ˙45 fiber
orientations at 23 ı C and at elevated temperature (Tright D 329 ı C)
tensile stress–strain response of the ˙45 fiber orientation; the stress–strain curves
obtained at elevated temperature are similar to those produced at 23 ı C. However,
elevated temperature caused a 20–23 % decrease in the ˙45 UTS of the 3D PMC
and of the 2D PMC. Such decrease in tensile strength of the MS1 and MS2 is hardly
surprising, since 329 ı C approaches the maximum use temperature for most high-
temperature polymer matrix composites. Notably, elevated temperature resulted in
884 M.B. Ruggles-Wrenn and M.P. Wilkinson
only a 3 % drop in the ˙45 UTS of the 2D PMC/CMC. This result suggests that the
CMC layer is acting as a thermal barrier for the PMC as intended. The ˙45 stiffness
values of all three composites also decreased somewhat due to elevated temperature.
Comparing the tensile properties produced by the three material systems at
elevated temperature (Tright D 329 ı C) and at 23 ı C, we note that the 2D PMC offers
the highest strength and stiffness under both temperature regimes. Changing the
fiber architecture from 2D to 3D did not improve tensile properties for either 0/90
or ˙45 fiber orientation. Likewise, adding a CMC thermal barrier did not improve
tensile properties of the 2D PMC for either the 0/90 or ˙45 fiber orientation.
a 1000
Tright = 329°C
UTS, 2D PMC/CMC
600
400
b 150
Tright = 329°C
125 UTS, 2D PMC
Max Stress (MPa)
100
75
UTS, 2D PMC/CMC
UTS, 3D PMC
50
3D PMC, ±45
25 2D PMC, ±45
2D PMC/CMC, ±45
0
1.E+00 1.E+01 1.E+02 1.E+03 1.E+04 1.E+05 1.E+06
Cycles (N)
Fig. 29.11 Fatigue S–N curves obtained at elevated temperature (Tright D 329 ı C) in laboratory air
for the 3D PMC, 2D PMC, and 2D PMC/CMC material systems with (a) 0/90 and (b) ˙45 fiber
orientations. Arrow indicates that failure of specimen did not occur when the test was terminated
a 100
0/90
± 45
80
Max Stress (% UTS)
60
40
20
MS1, 3D PMC
Tright = 329°C
0
1.E+00 1.E+01 1.E+02 1.E+03 1.E+04 1.E+05 1.E+06
Cycles (N)
b 100
0/90
± 45
80
Max Stress (% UTS)
60
40
20
MS2, 2D PMC
Tright = 329°C
0
1.E+00 1.E+01 1.E+02 1.E+03 1.E+04 1.E+05 1.E+06
Cycles (N)
Fig. 29.12 Fatigue S–N curves at elevated temperature (Tright D 329 ı C) in laboratory air for (a)
3D PMC, (b) 2D PMC, and (c) 2D PMC/CMC. Arrow indicates that failure of specimen did not
occur when the test was terminated. Maximum stress is shown as % UTS
fatigue performance in both 0/90 and ˙45 fiber orientations. Not surprisingly, the
tension–tension fatigue life of the 3D PMC is reduced compared to that of the 2D
PMC. Similar observations regarding reduction in fatigue life due to the presence
of through-thickness reinforcement were reported previously [26, 28, 29, 37–39].
Note that the fatigue S–N curves obtained for the 3D PMC are not parallel to those
obtained for the 2D PMC. Hence the loss in fatigue performance is not solely due to
29 Fatigue of 2D and 3D Carbon-Fiber-Reinforced Polymer Matrix. . . 887
c 100
0/90
80 ± 45
Max Stress (% UTS)
60
40
20
MS3, 2D PMC/CMC
Tright = 329°C
0
1.E+00 1.E+01 1.E+02 1.E+03 1.E+04 1.E+05 1.E+06
Cycles (N)
the lower UTS of the 3D PMC caused by microstructural defects. The Z-fibers are
affecting the fatigue mechanism and accelerating the degradation of fatigue strength
with cycles [5]. It is noteworthy that adding a CMC layer to the 2D PMC also gained
little improvement in tension–tension fatigue performance at elevated temperature.
Evolution of hysteresis stress–strain response of the 3D PMC with fatigue cycles
is typified in Fig. 29.13. For the 0/90 fiber orientation (Fig. 29.13a), we observe little
strain ratchetting (progressive strain accumulation with cycles) and only a slight
decrease in stiffness with cycles. Tensile strains accumulated in fatigue tests do
not exceed 0.5 %. The stress–strain hysteresis loops are very narrow; the overall
response of the composite is nearly linear elastic. Conversely, results in Fig. 29.13b
reveal noticeable strain ratchetting for the ˙45 specimens. Appreciable tensile
strains (1.5–3.5 %) are accumulated. Furthermore, we observe a greater decrease
in stiffness with cycles. The hysteresis loops become wider; the stress–strain
response of the composite is nonlinear. Figure 29.14 shows typical progression
of the hysteresis stress–strain response during fatigue tests for the 2D PMC. For
the 0/90 fiber orientation (Fig. 29.14a), strain ratchetting is limited; most of the
strain is accumulated during the first cycle. However, a considerable decrease
in stiffness with fatigue cycles is observed. Yet, with the exception of the first
cycle, the stress–strain loops remain narrow and the overall composite response,
nearly linear elastic. As in the case of the 3D PMC, the ˙45 2D PMC specimens
(Fig. 29.14b) accumulate considerable tensile strains during the fatigue test. Strain
ratchetting is evident. The hysteresis loops become wider as the cycling progresses;
the stress–strain behavior is markedly nonlinear. However, unlike the ˙45 3D PMC
specimens, 2D PMC specimens with ˙45 fiber orientation show increasing stiffness
with cycling early in fatigue test. Consider the ˙45 2D PMC specimen tested in
888 M.B. Ruggles-Wrenn and M.P. Wilkinson
a 600
Cycle 1
Cycle 100
Stress (MPa)
200
40
30
Cycle 20,000
20 Cycle 50,000
10
0
0.0 1.0 2.0 3.0 4.0
Strain (%)
Fig. 29.13 Typical evolution at elevated temperature (Tright D 329 ı C) of stress–strain hysteresis
response with fatigue cycles of the 3D PMC with (a) 0/90 fiber orientation and (b) ˙45 fiber
orientation
fatigue with the maximum stress of 66 MPa (Fig. 29.14b). The specimen stiffness
increased by about 12 % between cycle 10 and cycle 10,000. During tension–tension
fatigue, the ˙45 2D PMC specimen is subject to in-plane normal and shear stresses
that cause fibers to realign in the direction of applied load, resulting in increasing
stiffness. However, the stiffness drops later in fatigue test as the cycling progresses
toward failure.
29 Fatigue of 2D and 3D Carbon-Fiber-Reinforced Polymer Matrix. . . 889
a 800
Cycle 10,000 Cycle 100,000
600 Cycle 1
Stress (MPa)
Cycle 200,000
200
MS2, 2D PMC, 0/90
σmax = 585 MPa
0
0.0 0.5 1.0 1.5
Strain (%)
b 100
Cycle 1,000
Cycle 100 Cycle 10,000
80 Cycle 10 Cycle 100,000
Stress (MPa)
Cycle 1
60
Cycle 200,000
40
20
MS2, 2D PMC, ±45
σmax = 66 MPa
0
0.0 0.5 1.0 1.5
Strain (%)
Fig. 29.14 Typical evolution at elevated temperature (Tright D 329 ı C) of stress–strain hysteresis
response with fatigue cycles of the 2D PMC with (a) 0/90 fiber orientation and (b) ˙45 fiber
orientation
a 800
Cycle 100
400
Cycle 20,000
200
MS3, 2D PMC/CMC, 0/90
σ max = 530 MPa
0
0.0 0.5 1.0 1.5 2.0
Strain (%)
b 60
Cycle 10
Cycle 1,000
50
Cycle 100 Cycle 10,000
Cycle 60,000
Stress (MPa)
40 Cycle 1
30
20
Fig. 29.15 Typical evolution at elevated temperature (Tright D 329 ı C) of stress–strain hysteresis
response with fatigue cycles of the 2D PMC/CMC with (a) 0/90 fiber orientation and (b) ˙45 fiber
orientation
(see schematic in Fig. 29.16). As the cycling progresses, noticeable tensile strain is
accumulated and a decrease in stiffness is observed. Figure 29.15b shows typical
evolution of the hysteresis stress–strain response for the ˙45 fiber orientation. The
nonlinear nature of the stress–strain response is evident. As the cycling progresses,
the hysteresis modulus (determined from the maximum and minimum stress–
strain data points during a load cycle) decreases. Furthermore, the shape of the
hysteresis loops becomes increasingly irregular. Consider, for example, the stress–
strain response during cycle 60,000. As the stress increases from the minimum to
maximum value, the slope of the stress–strain curve first decreases, then increases.
We attribute the increasing irregularity of the stress–strain response to progressive
delamination and nonhomogeneous deformation (bowing out of the specimen sides)
in the specimen gage section. Strain ratchetting and a decrease in composite stiffness
are also observed.
Figures 29.17, 29.18, and 29.19 show strain accumulation with fatigue cycles for
the 3D PMC, the 2D PMC, and the 2D PMC/CMC, respectively. In the case of the
0/90 fiber orientation (Figs. 29.17a, 29.18a, and 29.19a), little strain is accumulated
throughout the cyclic life of all three composites. Strains accumulated by the 3D
PMC and the 2D PMC/CMC do not exceed 0.3 %, while strains accumulated by the
2D PMC remain below 0.4 %. In contrast, considerably larger strains are accumu-
lated in fatigue tests of all three composite with ˙45 fiber orientation (Figs. 29.17b,
29.18b, and 29.19b). The 2D PMC accumulated the largest strains (approaching
6.5 %), followed by the 3D PMC with strain accumulations approaching 4 %, and
892 M.B. Ruggles-Wrenn and M.P. Wilkinson
a 0.5
Max Stress = 425 MPa Max Stress = 483 MPa
Max Stress = 518 MPa Max Stress = 530 MPa
0.4 Max Stress = 560 MPa Max Stress = 610 MPa
Max Stress = 660 MPa
Strain (%)
0.3
0.2
0.1
0
1.E+00 1.E+01 1.E+02 1.E+03 1.E+04 1.E+05 1.E+06
Cycles (N)
Fig. 29.17 Maximum strain vs. fatigue cycles at elevated temperature (Tright D 329 ı C) for the 3D
PMC with (a) 0/90 fiber orientation and (b) ˙45 fiber orientation
a 0.5
Max Stress = 503 MPa Max Stress = 585 MPa
Max Stress = 610 MPa Max Stress = 663 MPa
0.4 Max Stress = 684 MPa Max Stress = 760 MPa
Strain (%)
0.3
0.2
0.1
0
1.E+00 1.E+01 1.E+02 1.E+03 1.E+04 1.E+05 1.E+06
Cycles (N)
Fig. 29.18 Maximum strain vs. fatigue cycles at elevated temperature (Tright D 329 ı C) for the 2D
PMC with (a) 0/90 fiber orientation and (b) ˙45 fiber orientation
a 0.5
Max Stress = 470 MPa Max Stress = 500 MPa
Max Stress = 520 MPa Max Stress = 530 MPa
0.4 Max Stress = 540 MPa Max Stress = 560 MPa
Max Stress = 580 MPa
Strain (%)
0.3
0.2
0.1
1.0
0.0
1.E+00 1.E+01 1.E+02 1.E+03 1.E+04 1.E+05 1.E+06
Cycles (N)
Fig. 29.19 Maximum strain vs. fatigue cycles at elevated temperature (Tright D 329 ı C) for the 2D
PMC/CMC with (a) 0/90 fiber orientation and (b) ˙45 fiber orientation
curves obtained for the 3D PMC and the 2D PMC/CMC are qualitatively similar to
those obtained for the 2D PMC, keeping to the classic pattern of a gradual reduction
followed by a rapid loss of normalized modulus as the test specimen approaches
failure.
As expected, the stiffness loss for the 0/90 fiber orientation was much smaller
than that for the ˙45 fiber orientation for all composites studied in this work.
In all tests, stiffness loss of the 0/90 specimens did not exceed 40 %. Note that
stiffness appears to increase slightly with cycling in tests of the 0/90 2D PMC
29 Fatigue of 2D and 3D Carbon-Fiber-Reinforced Polymer Matrix. . . 895
a 1.4
MS1, 3D PMC, 0/90
1.2
Normalized Modulus
1.0
0.8
0.6
0.0
1.E+00 1.E+01 1.E+02 1.E+03 1.E+04 1.E+05 1.E+06
Cycles (N)
b 1.50
MS1, 3D PMC, ±45
1.25
Normalized Modulus
1.00
0.75
0.00
1.E+00 1.E+01 1.E+02 1.E+03 1.E+04 1.E+05 1.E+06
Cycles (N)
Fig. 29.20 Normalized modulus vs. fatigue cycles at elevated temperature (Tright D 329 ı C) for
the 3D PMC with (a) 0/90 fiber orientation and (b) ˙45 fiber orientation
and 0/90 3D PMC specimens performed with lower maximum cyclic stress. This
phenomenon may be attributed to straightening of carbon fibers during the test.
Conversely, the ˙45 specimens of all composites exhibit considerable stiffness loss
during fatigue cycling. Stiffness loss reaches 50 % for the 3D PMC, 60 % for the
2D PMC/CMC, and 65 % for the 2D PMC. In some tests, the stiffness initially
increases, then decreases with cycles. For the ˙45 specimens, initial increase in
stiffness is attributed to the fibers realigning in the direction of applied load (“tow
scissoring”). Such realignment is most pronounced in the case of the 2D PMC.
896 M.B. Ruggles-Wrenn and M.P. Wilkinson
a 1.4
MS2, 2D PMC, 0/90
Normalized Modulus 1.2
1.0
0.8
0.6
0.4
Max Stress = 503 MPa Max Stress = 585 MPa
Max Stress = 610 MPa Max Stress = 663 MPa
0.2
Max Stress = 684 MPa Max Stress = 760 MPa
0.0
1.E+00 1.E+01 1.E+02 1.E+03 1.E+04 1.E+05 1.E+06
Cycles (N)
b 2.0
Max Stress = 66 MPa Max Stress = 70 MPa
Max Stress = 76 MPa Max Stress = 83 MPa
Normalized Modulus
1.0
0.5
Fig. 29.21 Normalized modulus vs. fatigue cycles at elevated temperature (Tright D 329 ı C) for
the 2D PMC with (a) 0/90 fiber orientation and (b) ˙45 fiber orientation
Retained strength and stiffness of all specimens that achieved fatigue runout of
2 105 cycles were evaluated in tensile tests at elevated temperature. Results in
Fig. 29.23a, b reveal that 2 105 cycles of prior fatigue caused limited loss of
strength and stiffness for the three composites studied in this work. Addition of
the through-thickness reinforcement (i.e., Z-fibers) had little effect of the retention
of tensile strength. The 0/90 2D PMC and the 0/90 3D PMC retained 85 and
83 % of their tensile strength, respectively. In the case of the ˙45 fiber orientation,
the 2D PMC retained 90 % of its tensile strength and the 3D PMC, 89 % of its
29 Fatigue of 2D and 3D Carbon-Fiber-Reinforced Polymer Matrix. . . 897
a 1.4
MS3, 2D PMC/CMC, 0/90
1.2
Normalized Modulus (E/E2)
1
0.8
0.6
Max Stress = 472 MPa Max Stress = 500 MPa
0.4 Max Stress = 520 MPa Max Stress = 530 MPa
Max Stress = 540 MPa Max Stress = 560 MPa
0.2
Max Stress = 580 MPa
0
1.E+00 1.E+01 1.E+02 1.E+03 1.E+04 1.E+05 1.E+06
Cycles (N)
b 1.50
Max Stress = 32 MPa Max Stress = 36 MPa
Max Stress = 39 MPa Max Stress = 39 MPa
1.25 Max Stress = 41 MPa Max Stress = 42 MPa
Normalized Modulus
0.75
0.50
0.25
MS3, 2D PMC/CMC, ±45
0.00
1.E+00 1.E+01 1.E+02 1.E+03 1.E+04 1.E+05 1.E+06
Cycles (N)
Fig. 29.22 Normalized modulus vs. fatigue cycles at elevated temperature (Tright D 329 ı C) for
the 2D PMC/CMC with (a) 0/90 fiber orientation and (b) ˙45 fiber orientation
tensile strength. Conversely, the presence of the Z-fibers had a beneficial effect on
the modulus retention. The 0/90 3D PMC retained 93 % of its tensile modulus,
while the 0/90 2D PMC retained only 83 % of its tensile modulus. Improved
modulus retention may be attributed to improved delamination resistance due to
the Z-fibers. In the case of the off-axis loading, the 2D PMC and the 3D PMC
exhibited approximately the same degree of modulus retention (84 %). Notably,
898 M.B. Ruggles-Wrenn and M.P. Wilkinson
a 100
60 3D PMC
smax = 483 MPa 2D PMC
smax = 585 MPa
smax = 33 MPa
smax = 69 MPa
smax = 32 MPa
2D PMC/CMC
40
20
0
0/90 ±45
Tension-Tension Tension-Tension
Fatigue Fatigue
b 100
80
Stiffnes Retention (%)
60 3D PMC
2D PMC
smax = 585 MPa
smax = 483 MPa
smax = 33 MPa
smax = 69 MPa
smax = 32 MPa
2D PMC/CMC
40
20
0
0/90 ±45
Tension-Tension Tension-Tension
Fatigue Fatigue
Fig. 29.23 Retention of (a) tensile strength and (b) stiffness of the 3D PMC, 2D PMC, and
2D PMC/CMC specimens subjected to 200,000 cycles of prior fatigue at elevated temperature
(Tright D 329 ı C)
the 2D PMC/CMC exhibited somewhat better strength retention than the other two
material systems. The 0/90 specimens retained 89 % and the ˙45 specimens, 97 %
of their tensile strength. The 0/90 2D PMC/CMC retained 90 % and the ˙45 2D
PMC/CMC, 68 % of its tensile modulus.
29 Fatigue of 2D and 3D Carbon-Fiber-Reinforced Polymer Matrix. . . 899
Failure of the 3D PMC specimens with 0/90 fiber orientation tested in this work
is typified in Fig. 29.24. Note that fracture of the 0/90 3D PMC specimens
occurred predominantly along a plane where one or more Z-fiber tows protrude
through the thickness of the specimen. Numerous matrix cracks are seen on the
surface of the specimen. Extensive pullout of the warp fiber tows is evident.
Virtually no ply delamination is seen. These observations suggest that the principal
damage events are matrix cracking, tow rupture, and tow pullout. The Z-fibers have
effectively increased the delamination resistance of the composite, with the main
toughening mechanisms likely being crack bridging and frictional resistance against
pullout [29].
Typical failure of the 3D PMC specimens with ˙45 fiber orientation tested in
this work is shown in Fig. 29.25. Matrix cracking is evident. Note that there is
little “scissoring” effect (i.e., the warp and fill fibers do not realign in the direction
of applied load, but rather stay at their respective ˙45ı angles to the direction of
applied load). In the case of the 0/90 3D PMC specimens, the warp tows rupture
at random positions along the entire specimen gage section. In contrast, damage
in the ˙45 specimens is more localized; fiber tows rupture close to the failure site
(typically the middle of the specimen gage section). Because Z-fibers are present at
every intersection of warp and fill fiber tows, one or more Z-fiber tows are exposed
upon fracture of the specimen. The Z-fiber tow and the rigid warp and fill fiber tows
are clearly visible in the fracture surface in Fig. 29.25.
Fig. 29.24 Optical micrographs showing typical failure of 3D PMC specimens with 0/90 fiber
orientation
900 M.B. Ruggles-Wrenn and M.P. Wilkinson
Fig. 29.25 Optical micrographs showing typical failure of 3D PMC specimens with ˙45 fiber
orientation
Typical failure mechanisms noted for the 2D PMC specimens with 0/90 fiber
orientation include ply delamination, fiber tow rupture, and fiber tow pullout
(Fig. 29.26). Extensile matrix cracking and ply delamination propagating through
the length of the specimen are clearly discernible. Failure of the 2D PMC specimens
with ˙45 fiber orientation is typified in Fig. 29.27. In this case, failure is more
localized. Most of the failure events are confined to the specimen gage section and
its immediate vicinity. Substantial matrix cracking and the “scissoring” effect are
evident.
Recall that the 2D PMC/CMC specimens exhibit nonhomogeneous deformation
under tensile load. The PMC plies and the CMC plies tend to bow out when
large tensile loads are applied (Figs. 29.6 and 29.7). We now examine the 2D
PMC/CMC specimens that separated into two parts upon failure. Figure 29.28
shows typical failure of the 2D PMC/CMC specimens with 0/90 fiber orientation.
Severe delamination of both the PMC and the CMC parts is clearly visible. Matrix
cracks often extend from the gage section into the gripping sections of the specimen.
29 Fatigue of 2D and 3D Carbon-Fiber-Reinforced Polymer Matrix. . . 901
Fig. 29.26 Optical micrographs showing typical failure of 2D PMC specimens with 0/90 fiber
orientation
In the case of specimens tested in tension to failure, both PMC and CMC parts
fractured in the specimen gage section. Conversely, in the case of specimens tested
in fatigue, the CMC parts fractured near top or bottom of the gage section, while the
PMC parts showed extensive damage throughout the gage section. It is noteworthy
that the CMC part did not fracture in the gage section exposed to direct heating.
Evidently the 329 ı C temperature does not have a significant damaging effect on the
CMC part. We believe that the location of the CMC fractures is caused by additional
bending stresses resulting from the PMC and CMC plies repeatedly bowing out and
in during cyclic loading.
As in the case of the 2D PMC with ˙45 fiber orientation, failure of the ˙45
2D PMC/CMC is localized in the specimen gage section. Extensive matrix cracking
and ply delamination are evident. Note the “scissoring” effect—fibers detach from
the matrix material and align in the direction of the applied load. Recall that the
˙45 specimens of the co-cured 2D PMC/CMC also exhibited nonhomogeneous
deformation when loaded in tension (Fig. 29.29). However, ply delamination and
bowing out of plies were confined to the specimen gage section.
902 M.B. Ruggles-Wrenn and M.P. Wilkinson
Fig. 29.27 Optical micrographs showing typical failure of 2D PMC specimens with ˙45 fiber
orientation
The experimental study presented in this work focused on tensile properties and
some aspects of mechanical behavior of (1) a single-ply non-crimp 3D orthogonal
weave carbon-fiber composite, (2) a laminated composite reinforced with 15 plies
of carbon fibers woven in an 8HSW, and (3) a co-cured PMC/CMC reinforced with
2D fabric woven in an 8HSW.
The Z-fibers have dramatically increased the delamination resistance of the 3D
PMC composite. Examination of the failed 3D PMC specimens reveals extensive
matrix cracking, tow rupture, and tow pullout, but virtually no ply delamination.
The 2D PMC/CMC, produced by co-curing of the two dissimilar materials,
exhibited severe delamination of both PMC and CMC plies under tensile loading.
Furthermore, nonhomogeneous deformation (i.e., bowing out of the PMC plies and
the CMC plies) was observed.
The tensile stress–strain behavior and tensile properties of the three composites
were investigated for both 0/90 and ˙45 fiber orientations at room and elevated
temperature. At 23 ı C the elastic modulus of the 0/90 unitized 2D PMC/CMC is
nearly the same as that of the 2D PMC. In contrast, the tensile strength of the
0/90 2D PMC/CMC is considerably lower than that of the 2D PMC. Progressive
delamination of the 2D PMC/CMC during tension test is likely behind the reduction
in tensile strength. The elastic modulus and the tensile strength of the 0/90
3D PMC are lower than the corresponding values obtained for the 2D PMC.
29 Fatigue of 2D and 3D Carbon-Fiber-Reinforced Polymer Matrix. . . 903
Fig. 29.28 Optical micrographs showing typical failure of 2D PMC/CMC specimens with 0/90
fiber orientation
The reductions in elastic modulus and tensile strength of the 0/90 3D PMC are
attributed to the microstructural defects introduced along with the Z-fibers during
processing. As expected, the tensile stress–strain behavior of the 2D PMC and
the 3D PMC with 0/90 fiber orientations is nearly linear to failure. Conversely,
the stress–strain behavior of the 0/90 2D PMC/CMC becomes nonlinear as the
stress exceeds 300 MPa. We believe that the nonlinear stress–strain response is
caused by the nonhomogeneous deformation and progressive delamination in the
specimen gage section. The ˙45 stress–strain behavior of the three composites
becomes nonlinear early in tension test. Strength and stiffness are much lower
while failure strains are much higher than the corresponding 0/90 values. The low
values of ˙45 strength and stiffness are attributed to the absence of the off-axis
fibers in the reinforcement architecture of the three composites. Large failure strains
are due to the tow “scissoring” typically observed in textile composites. Notably,
the ˙45 2D PMC/CMC exhibits extensive delamination and nonhomogeneous
deformation in tension test. Overall, the 3D PMC offers little improvement in tensile
properties compared to the 2D PMC at room or elevated temperature. However,
the 3D PMC exhibits increased delamination resistance. It may be possible to
904 M.B. Ruggles-Wrenn and M.P. Wilkinson
Fig. 29.29 Optical micrographs showing typical failure of 2D PMC/CMC specimens with ˙45
fiber orientation
Overall, the 2D PMC offers better fatigue performance than the other two
material systems, especially at higher cyclic stress levels. The fatigue durability
of the 3D PMC is reduced by the presence of the Z-fibers. The reduction in
fatigue performance of the 3D PMC is attributed to the loss of the UTS due to
microstructural defects caused by insertion of the Z-fibers as well as to the degrading
influence of the Z-fibers on the fatigue process. The fatigue performance of the
unitized 2D PMC/CMC is also significantly worse than that of the 2D PMC. We
believe that extensive delamination and nonhomogeneous deformation accelerate
the loss of fatigue strength of the 2D PMC/CMC with increasing number of load
cycles. The repeated bowing out and in of the PMC and CMC plies during tension–
tension fatigue cycling gives rise to additional bending stresses. Thus during
tension–tension fatigue, the 2D PMC/CMC is effectively subjected to a combination
of tension and bending. However, we recognize that fatigue damage mechanisms in
the 3D PMC and in the unitized 2D PMC/CMC are complex and require further
investigation.
References
14. L.K. Jain, Y.-W. Mai, Determination of mode II delamination toughness of stitched laminated
composites. Compos. Sci. Technol. 55, 241–253 (1995)
15. B.N. Cox, Simple, conservative criteria for buckling and delamination propagation in the
presence of stitching. J. Comp. Mater. 34, 1136–1147 (2000)
16. B.N. Cox, R. Massabò, K.T. Kedward, Suppression of delaminations in curved structures by
stitching. Composites 27A, 1133–1138 (1996)
17. B. Yu, R.S. Bradley, C. Soutis, P.J. Hogg, P.J. Withers, 2D and 3D imaging of fatigue failure
mechanisms of 3D woven composites. Compos. Part A 77, 37–49 (2015)
18. A.P. Mouritz, M.K. Bannister, P.J. Falzon, K.H. Leong, Review of applications for advanced
three-dimensional fibre textile composites. Compos. Part A 30, 1445–1461 (1999)
19. A.E. Bogdanovich, M.H. Mohamed, Three-dimensional reinforcements for composites.
SAMPE J. 45(6), 8–28 (2009)
20. S.V. Lomov, A.E. Bogdanovich, D.S. Ivanov, D.D. Mungalov, M. Karahan, I. Verpoest, A
comparative study of tensile properties of non-crimp 3D orthogonal weave and multi-layer
plain weave E-glass composites. Part 1: materials, methods and principle results. Compos. Part
A 40, 1134–1143 (2009)
21. D.S. Ivanov, S.V. Lomov, A.E. Bogdanovich, M. Karahan, I. Verpoest, A comparative study
of tensile properties of non-crimp 3D orthogonal weave and multi-layer plain weave E-glass
composites. Part 2: comprehensive experimental results. Compos. Part A 40, 1144–1157
(2009)
22. M. Karahan, S.V. Lomov, A.E. Bogdanovich, D. Mungalov, I. Verpoest, Internal geometry
evaluation of non-crimp 3D orthogonal woven carbon fabric composite. Compos. Part A 41(9),
1301–1311 (2010)
23. A.E. Bogdanovich, M. Karahan, S.V. Lomov, I. Verpoest, Quasi-static tensile behavior and
damage of carbon/epoxy composite reinforced with 3D non-crimp orthogonal woven fabric.
Mech. Mater. 62, 14–31 (2013)
24. W.-S. Kuo, T.-H. Ko, C.-P. Chen, Effect of weaving processes on compressive behaviour of 3D
woven composites. Compos. Part A 38, 555–565 (2007)
25. A.E. Bogdanovich, Multi-scale modelling, stress and failure analyses of 3-D woven compos-
ites. J. Mater. Sci. 41(20), 6547–6590 (2006)
26. A.P. Mouritz, Tensile fatigue properties of 3D composites with through-thickness reinforce-
ment. Compos. Sci. Technol. 68(12), 2503–2510 (2008)
27. V. Carvelli, G. Gramellini, S.V. Lomov, A.E. Bogdanovich, D.D. Mungalov, I. Verpoest,
Fatigue behaviour of non-crimp 3D orthogonal weave and multi-layer plain weave E-glass
reinforced composites. Compos. Sci. Technol. 70(14), 2068–2076 (2010)
28. M. Karahan, S.V. Lomov, A.E. Bogdanovich, I. Verpoest, Fatigue tensile behavior of car-
bon/epoxy composite reinforced with non-crimp 3D orthogonal woven fabric. Compos. Sci.
Technol. 71(16), 1961–1972 (2011)
29. S. Rudov-Clark, A.P. Mouritz, Tensile fatigue properties of a 3D orthogonal woven composite.
Compos. Part A 39(6), 1018–1024 (2008)
30. M.S. Dadkhah, B.N. Cox, W.L. Morris, Compression–compression fatigue of 3D woven
composites. Acta Metall. Mater. 43(12), 4235–4245 (1995)
31. K.-H. Tsai, C.-H. Chiu, T.-H. Wu, Fatigue behaviour of 3D multi-layer angle interlock woven
composite plates. Compos. Sci. Technol. 60(2), 241–248 (2000)
32. M. Gude, W. Hufenbach, I. Koch, Damage evolution of novel 3D textile-reinforced composites
under fatigue loading conditions. Compos. Sci. Technol. 70(1), 186–192 (2010)
33. G. Stegschuster, K. Pingkarawat, B. Wendland, A.P. Mouritz, Experimental determination of
the mode I delamination fracture and fatigue properties of thin 3D woven composites. Compos.
Part A 84, 308–315 (2016)
34. D.E. Glass, H.K. Rivers, Advances in hot-structure development. Technical Report 01/2006
(NASA Langley Research Center, Hampton, VA, 2006)
35. D.E. Glass, Ceramic matrix composite (CMC) thermal protection systems (TPS) and hot
structures for hypersonic vehicles. Paper AIAA-2008-2682, 15th AIAA Space Planes and
Hypersonic Systems and Technologies Conference. AIAA 2008
29 Fatigue of 2D and 3D Carbon-Fiber-Reinforced Polymer Matrix. . . 907
36. F. Breede, M. Frieß, Development of advanced CMC materials for dual-bell rocket nozzles.
Annual Report, Transregional Collaborative Research Center, German Aerospace Center,
Institute for Design and Construction Research, Stuttgart, Germany (2009)
37. F. Aumerich, P. Priolo, C.Y. Sun, Static and fatigue behaviour of stitched graphite/epoxy
composite laminates. Compos. Sci. Technol. 63, 907–917 (2003)
38. P. Chang, A.P. Mouritz, B.N. Cox, Properties and failure mechanisms of z-pinned laminates in
monotonic and cyclic tension. Compos. Part A 37, 1501–1513 (2006)
39. A.P. Mouritz, P. Chang, Tension fatigue of fibre-dominated and matrix-dominated laminates
reinforced with z-pins. Int. J. Fatigue 32(4), 650–658 (2008)
Chapter 30
Carbon Fibers in Tribo-composites
Abbreviations
AF Aramid fibers
AFM Atomic force microscopy
AFRP Aramid fiber-reinforced polymers
ANN Artificial neural network
AP Antiparallel (orientation)
BAM Bundesanstalt für Materialprüfung
CF Carbon fibers
CFRP Carbon fiber-reinforced polymers
CFRPF Carbon fiber-reinforced paper-based friction material
EP Epoxy resin
ETFE Ethylenetetrafluoroethylene
GF Glass fibers
GFRP Glass fiber-reinforced polymers
Gr Graphite
HM High modulus
HS High strength
LCP Liquid crystal polymer
MWNTs Multiwall carbon nanotubes
N Normal (orientation)
K. Friedrich ()
Institute for Composite Materials (IVW GmbH), Technical University of Kaiserslautern,
67663 Kaiserslautern, Germany
e-mail: klaus.friedrich@ivw.uni-kl.de
L. Chang
Centre for Advanced Materials Technology, School of Aerospace, Mechanical and Mechatronic
Engineering, The University of Sydney, Sydney, NSW, 2006 Australia
NP Nanoparticles
NT No fiber surface treatment
P Parallel (orientation)
PA Polyamide
PA 66 Polyamide 66
PAI Polyamide-imide
PAN Polyacrylonitrile
PC Polycarbonate
PE Polyethylene
PEEK Polyetheretherketone
PEN Polyethernitrile
PES Polyethersulfone
PET Polyethylene terephthalate
PI Polyimide
PMMA Polymethyl methacrylate
POM Polyoxymethylene (polyacetal)
PP Polypropylene
PPN Polyphenylene
PPO Polyphenylene oxide
PPS Polyphenylene sulfide
PTFE Polytetrafluoroethylene
RP Reinforced polymers
RT Room temperature
SCF Short carbon fiber
SEM Scanning electron microscope
SF Steel fiber
SFRP Short-fiber-reinforced polymers
TFL Transfer film layer
TP Thermoplastics
UHMWPE Ultrahigh-molecular-weight polyethylene
UP Unsaturated polyester resin
XPS X-ray photoelectron spectroscopy
30.1 Introduction
In the past five decades, there has been a remarkable growth in the large-scale
industrial application of fiber-reinforced polymer composites. The features that
make them so promising as industrial and engineering materials are their high
specific strength (strength/density), high specific stiffness (modulus/density), and
the opportunities to tailor material properties through the control of fiber and
matrix composition and fiber orientation states through processing technology [1].
In addition, parts made from these materials have the ability to dampen shock and
vibration, provide corrosion protection, and operate with little or no maintenance,
30 Carbon Fibers in Tribo-composites 911
making them excellent materials, e.g., for the aircraft and aerospace, chemical, and
automotive industry. Applications of polymer composites are found in components
such as aircraft wings and fuselage, solar cell panels of space stations, automotive
passenger cells, antennae in arctic or desert regions, wind energy blades, drive
shafts, or lamellae for bridge repair.
In most of these services, the materials are subjected to different kinds of
mechanical stresses and various corrosive environments. Additional problems can
arise from possible tribological loading conditions. The discipline called tribology
covers the study of friction and wear of materials in all their aspects. Friction is
the force which tends to prevent the relative motion of two surfaces in contact.
The term lubrication stands for the interposing of a substance between the two
interacting surfaces for the purpose of reducing friction. The loss of material from
a solid surface as a result of pressure and sliding exerted by one body on another is
called wear [2].
In certain applications the coefficient of friction is of importance as in
braking materials (high ) or gravity feed wear strips (low ), but largely it is
the mechanical property and the wear life of the component that determines its
acceptability in industrial use. Usually, wear is undesirable not only because it
makes necessary frequent inspections and replacements of parts but it will also lead
to deterioration of accuracy (e.g., of machine parts). It can induce vibrations, fatigue,
and consequently failure by rupture. Wear can also produce debris, which in turn
will cause immobilization of close-fitting precision mechanisms, orifice occlusions,
and foul electrical contacts [3]. Often these composite machine components are
relatively small, such as gears, cams, wheels, impellers, brakes, clutches, seals,
rollers, transmission belts, bushes, and bearings, but in case of their failure, they can
often cause a breakdown of larger machine parts or even complete manufacturing
lines. On chemical engineering plants, reinforced plastics are used in the field of
piping, duct work, pumps, agitated vessels, separators, and thermo-compressors
[4–9]. Other important applications are conveyor aides and chute liners which
are used for agricultural, mining, and earthmoving equipment and in industries
processing coal, coke, mineral ores, and glass. Erosive wear plays an important role
in many technical systems, which range from pipes and pumps handling slurries and
other corrosive abrasives to aircraft air foils passing through zones of rain and hail
[10].
Friction and wear of polymer composites (i.e., of tribo-composites) have been
the subject of many studies over the last 50 years. Empirical equations have
been derived to help in predicting the likely performance of a material in a
particular application or to aid the selection of the most suitable polymers and fiber
reinforcements. Nevertheless, it is widely accepted that the study of friction and
wear processes in polymeric composites is in many areas still not well understood.
In particular, there is no systematic account of the role played by microstructure
and material properties on the wearing behavior of the material. To illustrate,
although polymers such as ethylenetetrafluoroethylene (ETFE) reinforced with
carbon fibers (CF) [11], polymethylmethacrylate (PMMA) filled with quartz and
glass particles [12], and epoxy resin (EP) reinforced with graphite fibers [13] are
used in different commercial “wearing” applications, a better knowledge about their
intrinsic physical wear mechanisms would be preferable.
912 K. Friedrich and L. Chang
It is the purpose of this chapter to outline a few of the topics of interest in the field
of friction and wear of advanced polymer composites, especially those reinforced
with CF, and to present a survey of the findings of present and past workers. The
major focus is concentrated on their use under sliding wear conditions, primarily
against metallic counterparts (with the objective of low friction and low wear). At
first, a general overview is given about important findings regarding the friction
and wear of carbon fiber-reinforced polymers (CFRP). In the following, recent
developments regarding the simultaneous combination of carbon fibers and other
fillers for achieving not only further improvements in the tribological behavior of
these hybrid composites but also a higher degree of multifunctionality are discussed.
Finally, a few remarks about high-friction composites as well as metal, glass, and
ceramic matrix composites are made.
Fig. 30.1 External loadings and resulting changes of a polymeric structural component (after an
idea of CCM, University of Delaware, USA, 1988)
Table 30.1 Main polymers and fillers of interest for bearings [14]
Thermoplastics Thermosets
Ultrahigh molecular weight Phenolics
Polyethylene (UHMWPE) Polyesters
Acetal homo- and copolymers Epoxies
Polyamide 6, 66, and 11 Silicones
PTFE Polyimides
Polyphenyleneoxide
Polyetheretherketone
Fillers and reinforcements
To improve9mechanical properties To reduce friction To improve thermal properties
Aramide > >
>
=
Glass
Fibers Graphite Bronze
Carbon > >
>
;
Textile
MoS2 Silver
PTFE Carbon/graphite
(Particles or fibers)
Mica
Metals and oxides
914 K. Friedrich and L. Chang
and tribological properties extending the range of operation of the base polymer
[17–20]. As an example, Fig. 30.2 shows values of the specific wear rate ws (or wear
factor k*) for a wide range of commercially available composite materials derived
from thrust-bearing tests.
The main reason for adding glass fibers (GF) to a resin is to improve the short-
and long-term mechanical properties. By improving the creep resistance, thermal
conductivity, and heat distortion characteristics, glass-reinforced resins demonstrate
marked improvements in the limiting pressure (p) x velocity (v) values when
compared to unmodified base polymer. Wear resistance is improved by the addition
of glass fibers [4] to a polymer, but glass fibers increase the wear of the mating
surface [21, 22] and the coefficient of friction. The use of glass beads compared
to glass fibers gives a little bit higher values of the composite’s wear rate and a
lower load-bearing capacity but yields sometimes to a reduction of the frictional
coefficient and to lower wear of the mating surface [23].
The addition of CF to polymer resins results in compounds with the high-
est values of strength, modulus, heat distortion temperature, creep, and fatigue
endurance currently commercially available. These property improvements coupled
with greatly increased thermal conductivity and low coefficients of friction make
CF the ideal reinforcement for wear and friction applications [24, 25]. Unlike the
reinforcement with GF, the coefficient of friction of the CF compound is lower than
that of the base polymer. The amount of this reduction is determined by the degree of
graphitization of the CF. Since the limiting (pv) factor of a polymer is directly related
to its thermal conductivity and creep resistance, the increase in the operational limits
obtained with GF can be substituted and the wear rate of the mating surface sharply
reduced [4].
30 Carbon Fibers in Tribo-composites 915
Table 30.2 Influence of fibers and lubricants on wear factor, friction coefficient, and
limiting pv value of PA66 composites [4]
Wear Friction Limiting
Polymera Filler factor k* coefficient pv value
PA66 – 3.95 0.28 0.083
PA66 30 % GF 1.48 0.31 0.250
PA66 30 % CF 0.40 0.20 0.270
PA66 30 % GF C 15 % PTFE 0.32 0.26 0.583
PA66 30 % CF C 15 % PTFE 0.20 0.15 0.633
PA66 30 % CF C 13 % PTFE C 2 % silicon 0.12 0.11 0.666
k* in 106 mm3 /(Nm)
dynamic at 0.28 MPa 0.254 m/s
pv in MPa m/s at 5 m/s
Counterpart: cold rolled steel, Rt D 0.3 m; HRC 20
a
Date from LNP Engineering Plastics
The use of internal lubricants will, in many cases, offset the negative effects that
the GF have on surface characteristics. Polytetrafluoroethylene (PTFE) lubricants
pre-dispersed into a base resin greatly improve the surface wear characteristics of the
compound. During the initial running-in period, the PTFE particles embedded in the
matrix shear to form a high lubricity film over the mating surface. In addition, PTFE
serves to cushion asperities, for example, at glass fiber ends, from shock, subsequent
fracture, and the resultant enhanced wear. A further improvement of wear, friction
and (pv) properties can be achieved by the use of migratory internal lubricants
(silicone). The silicone moves to the surface of a molded part by two mechanisms:
diffusion by random molecular movement, and exclusion from the matrix polymer
(migration) because of its limited compatibility. The result of this migratory action
is continuous generation of a silicone film, which serves as a boundary, or mixed
film-lubricant. In combination with PTFE at the wear surface, a high-temperature
grease is formed [4]. A comparison of the effect of fiber reinforcement and internal
lubricants on wear factor k*, frictional coefficient , and limiting (pv) value of
polyamide 66 (PA66) compositions is given in Table 30.2.
The question of why fibers usually improve the wear resistance of a polymer
matrix has been the subject of intense study in the 1970s and 1980s. Lancaster
[26] investigated the effect of randomly oriented chopped carbon fibers on the
wear of various polymers and pointed out that the mechanism of wear reduction
in polymers by the fibers appears to involve two factors. First, the fibers are exposed
at the sliding surface and support a great part of the applied load; second, the
fibers smooth the surface of the counterface and thus reduce the localized stresses
at the asperity contacts. The first of these factors is probably largely responsible
916 K. Friedrich and L. Chang
for the observation that the friction and wear rates of the CF-reinforced materials
are almost independent of the matrix material. The second factor, abrasion of the
steel counterface, probably explains why glass fibers are less effective than carbon
in reducing wear rates. Glass fibers cause appreciable roughening of the steel,
and hence the localized stresses and the rate of wear of the composite remain
high. In addition, the coefficient of friction of glass on steel is much greater
than that of carbon, and the shear stresses on individual fibers will, in turn, be
correspondingly higher. If these stresses become high enough to fracture exposed
fibers, the subsequent escape of glass fiber debris within the contact zone is likely
to exacerbate wear still further.
The results presented by Tanaka [27] indicate that additional effects related to
the testing speed have to be considered. At very low speeds, unreinforced polyacetal
(Dpolyoxymethylene) (POM) displays a much lower wear rate than the POM filled
with short glass or carbon fibers; while in a speed range higher than a critical speed
(as a function of load), the fiber/polyacetal composites expressed lower wear rates
(Fig. 30.3). The interpretation is based on two observations: at very low speeds, the
wear rate of POM is very low compared to glass or carbon fibers. Some authors tried
to interpret the results by using the rule of mixtures, which should also be applicable
to the wear resistance of coarse two-phase materials [12, 28]:
w1 1 1
sC D .1 AF =A/ wsM C .AF =A/ wsF (30.1)
where AF /A is the ratio of exposed area of filler to the apparent contact area and
wsC 1 , wsM 1 , and wsF 1 are the specific wear resistance of the composite (C), the
matrix (M), and the filler (F), respectively. The composite wear rate, which was
actually found for these materials, increases with increasing content of fibers in this
state. At very high speeds, on the other hand, the unfilled POM wears by melting
and outflow of molten material through the rear edge of the specimen pin, so that its
wear rate significantly exceeds that of the fiber materials. It is reasonable to consider
that such an outflow of molten polymer cannot occur if the frictional surface layer
includes fibers. This explains the finding that fibers in filled POM considerably
reduce wear in the higher energy speed range. In addition, fibers are laid down along
the surface layer during this process, producing a very dense fiber-rich surface layer
(increase in value AF /A of Eq. (30.1) and, hence, additional reduction in wear rate).
Other papers on the friction and wear of unfilled and glass- or carbon fiber-filled
POM come to similar results. Clerico [29] found for low-speed wear conditions that
the wear of CF-POM against steel cylinders was at least 40 times greater than that
of the unfilled material, whereas the values of the frictional coefficient remained
nearly unaffected by the glass fibers. The author also mentioned that there was
a considerable amount of wear of the steel counterface due to the glass fibers.
The results obtained by Jain [30] with carbon fiber-reinforced polyacetal are in
agreement with those of Tanaka [27], i.e., carbon fibers are superior to glass fibers
in reducing the wear rate and the frictional coefficient. Especially at high sliding
speeds and high loads, they clearly improve the tribological properties of the base
polymer.
Tanaka and Kawakami [31] also pointed out that the wear-reducing action of
fillers is dependent on factors such as shape and size as well as the composition
of the filler material. For the friction and wear of PTFE-based composites, the
authors showed that very small fillers (TiO2 particles) on the frictional surface
cannot prevent large-scale destruction of the matrix polymer, and thus very small
fillers are easily removed from the frictional surface together with the PTFE film
transferred onto the counterface. In contrast, larger fillers such as fibers greatly
reduce the wear of PTFE.
In a lot of studies on short-fiber-reinforced thermoplastics by Friedrich et al.
[32–37], it was also shown that short carbon fibers are usually more effective in
reducing the composites’ wear rates than glass fibers. In addition, the wear was
dominated by the polymer matrix material in the range of low fiber contents (up
to about 10 %), but was dominated by the fiber material and their adhesion to the
polymer matrix when a higher amount of fibers was incorporated (Fig. 30.4).
The corresponding wear mechanisms which control the final specific wear rate
of the composite, ws,C , can be summarized as (a) fiber sliding wear (ws,Fs), (b) fiber
cracking, and (c) wear by fiber/matrix separation at the interface, and (d) matrix
wear (ws,M ). It can be assumed that mechanisms (b) and (c) occur sequentially
and they can therefore be considered as a combined process of fiber cracking and
interfacial separation (ws,Fci ) (Fig. 30.5) [33].
The partial value of this process, ws,Fci , is difficult to determine, but it is obvious
that this mechanism becomes more important (at the cost of fiber sliding), the higher
the fiber volume fraction Vf (DAF /A), the lower the interfacial (fiber/matrix) shear
strength, and the higher the (pv) product is. Under these circumstances, more broken
918 K. Friedrich and L. Chang
Fig. 30.4 Effect of short-fiber reinforcement on wear of thermoplastic materials: open symbols
refer to neat matrices (wear rates within four orders of magnitude); black-filled symbols D GF
composites; dotted symbols D CF composites (wear rates within two orders of magnitude) [35]
fiber particles, which are dug out of the surface, can act as third-body abrasives, thus
inducing more sites of fiber breakage from which subsequent interfacial separation
takes place. This means, the simple rule of mixtures approach mentioned above
(Eq. (30.1)) should be modified in the following way:
˚ 1
ws;C D 1 Vf w1 1
s;M C ˛Vf ws;Fs C ˇVf ws;Fci (30.2)
where ws,C and ws,M are the specific wear rates of the composite and matrix,
respectively. The factor ˛ accounts for the relative surface of the reinforcement worn
by sliding, whereas ˇ accounts for the relative proportions of the reinforcement
surface removed by the mechanisms excluding sliding (i.e., cracking and interfacial
removal); hence ˛ C ˇ D 1. Both factors (˛ and ˇ) are presumed to be functions
of Vf :
˛ D 0:5 1 C Vf (30.3a)
ˇ D 0:5 1 Vf (30.3b)
The exponent in Eq. (30.3) indicates how fast the transition from sliding to the
dominance of other wear mechanisms occurs. Therefore, ¦ should depend on the
fiber geometry and type, and, in particular, on the fiber-matrix bond quality. Since
in Eqs. (30.2) and (30.3), , ws,Fs , and ws,Fci are unknown, the following iterative
way was chosen to estimate them:
1. Assuming that D 0, a curve was fitted to the experimental values and ws,Fs read
at Vf D 1.
2. ws,Fci values, taken from the range given in Refs. [33, 34], were used to compute
ws,C values according to Eq. (30.2). Through an iterative process, these ws,C
values were compared with the measured values until good agreement was
reached in this step, and was kept constant as 1.
3. Final fitting of the measured data points through the model (given by Eq. (30.2))
was achieved by considering Eq. (30.3) via slight variations in .
In fact, using this modeling approach to describe the specific wear rate of
polyethernitrile (PEN) as a function of short glass and two different short carbon
fiber (PAN based vs. Idemitsu pitch based) contents resulted in a good agreement
between the experimental data and the theoretical predictions (Fig. 30.6) [38]. The
better sliding wear resistance of the Idemitsu pitch-based CF was attributed to their
better strand friction properties, as tested according to a special friction test method
(Fig. 30.7) [39]. The longer time to resist friction before breaking in this test seems
to postpone also the occurrence of the ws,Fci mechanism in the sliding wear test,
which in turn relates to a reduction of the exponent in Eq. (30.3).
920 K. Friedrich and L. Chang
Giltrow and Lancaster from the Royal Aircraft Establishment in Farnborough, UK,
emphasized already in a paper on “Carbon Fibers in Tribology” back in 1970
the different friction and wear behavior of high-modulus (HM, type I) and high-
strength (HS, type II) fibers in various polymer matrices. Figure 30.8 shows a
comparison of the coefficients of friction and steady-state wear rates (after about
5 105 revolutions) for composites containing either type I or type II fibers. The
fiber content for thermoplastic composites was about 30 % by wt., while that for
composites with thermosetting resin matrices was about 50 % by wt. The pattern of
results obtained with the thermosetting matrices is clear and unambiguous; type II
fiber reinforcement gives both higher friction and higher wear rates than type I. For
the thermoplastics, the friction with type II fiber composites is again higher than
30 Carbon Fibers in Tribo-composites 921
with type I, but the wear rates do not follow a wholly consistent pattern. With four
materials, the type II fibers give lowest wear, whereas for two others, type I fiber
composites are marginally superior [40].
To understand the reasons for the above behavior, it is necessary to review briefly
some of the earlier conclusions on the mechanisms by which carbon fibers are
believed to reduce the wear of polymers. It has been suggested [41] that when
equilibrium conditions of sliding are attained, the carbon fibers in a composite
preferentially support the greater part of the applied load, and the contact conditions
approximate to those of carbon sliding against the counterface. This is an idealized
picture because localized polymer/counterface contacts must inevitably play some
part in the rubbing process from time to time. A further factor entering into the
wear mechanism arises from the effect of the composite on the counterface. High-
modulus fiber composites, in general, produce a transfer film on a metal counterface
which appears to consist of physically degraded material from the carbon fibers and
polymer from the matrix. This transfer film reduces the surface roughness of the
counterface, and since the wear rates of polymers, and carbons, are very dependent
on counterface roughness [26, 42], the wear rate of the composite itself is, in turn,
reduced [40].
Transfer films do not form with high-strength fiber composites; the wear track
on a metal counterface is always bright and polished, and small amounts of loose
oxidized metal debris are sometimes produced. The absence of a transfer film with
these composites appears to be due to the fact that type II fibers are considerably
more abrasive than type I, and so remove any wear debris which attempts to persist
as transfer on the counterface. Relative values for the abrasiveness of type I and type
922 K. Friedrich and L. Chang
Table 30.3 Relative abrasiveness of polymers with various fillers (VPN D Vickers Pyramid
Number D Vickers Hardness HV) [43]
Relative abrasiveness of polymer composites (brass ball 200 VPN)
Relative abrasiveness
106 mm3 /(Nm) vs. brass
Class No. Composites type ball
Filled PTFE 1 PTFE/mineral fibers 29
2 PTFE/40 % glass fibers 16
3 PTFE/15 % glass fibers 6
Carbon fiber composites 4 Polyimide/type II fibers 8.1
5 Polyimide/type I fibers 5
6 PTFE/type I fibers 0.13
7 Polyamide/type I fibers 0.1
Steel/thin layer materials 8 Woven PTFE fiber/glass fiber 8
9 PTFE flock/Nomex fabric 0.17
Reinforced thermosets 10 Polyester/textile laminate 2.7
11 Phenolic/cotton fabric/PTFE 0.04
Fig. 30.9 Influence of counterface roughness on the wear of carbon fiber-reinforced PTFE:
(a) high-modulus fibers, (b) high-strength fibers. Counterface: low carbon steel; roughness:
1. Ra D 0.04 m, 2. Ra D 0.18 m, 3. Ra D 0.5 m [42]
At this point it has to be taken into consideration what effect can arise from the
orientation of fibers in a polymer material with a given fiber content. Several groups
who tested this effect with unidirectional oriented continuous fiber composites
showed that, in some cases, the lowest wear rates were obtained when the fibers were
oriented perpendicular (i.e., normal) to the sliding surface, although the coefficient
of friction was sometimes higher as compared to the parallel or antiparallel fiber
orientation in the sliding surface [44–49]. This is especially true for the tougher
aramid fibers (AF) (as shown, e.g., by Sung and Suh [45], Tzukizoe and Ohmae
[49], and Cirino, Friedrich, and Pipes [50, 51]. But also Lancaster published this
tendency for CF-reinforced epoxy composites [42]. The explanation of this effect is
based on the fact that fiber/matrix debonding can occur during the wear procedure.
It is expected that when fibers are oriented normal to the surface, debonding will
occur at the surface and will propagate down along the length of the fibers only to
a finite distance, because the tensile stress component which causes debonding will
diminish as the distance from the surface increases. The fracture of the fibers will
require a tensile loading due either to buckling or to a mechanical pulling action
at the interface. Both of these occur with difficulty since the lateral deformation
of normally oriented fibers will be constrained by the presence of neighboring
fibers. When the fibers are oriented in plane of the sliding surface, as in parallel
or antiparallel orientations, on the other hand, initiation of cracks may occur either
924 K. Friedrich and L. Chang
at the surface or at the fiber-resin interface at a finite depth from the surface where
the tensile stress component perpendicular to the surface has a maximum. Once
cracks are nucleated, they will propagate under cyclic loading and the debonding
length will increase. By this process large-scale fiber separation can occur. These
separated individual fibers will fracture more readily owing to bending, increasing
the wear rate. The differences between antiparallel and parallel orientation is less
clear, although the debonding may occur more in the transverse case [45].
Different tendencies for carbon fibers were published by other authors. Cirino
et al. [51] investigated the dry wear behavior of several polymer composite
materials (CF-EP, AF-EP, GF-EP, CF-PEEK, AF-PEEK, GF-PEEK) sliding against
smooth steel counterparts. Wear mechanisms were categorized after microscopic
observation of the worn surfaces into four main areas ((a)–(d), as mentioned already
in context with the modified rule of mixtures), and their contribution toward the
overall wear rate of a composite material was accounted for in a wear model.
The contribution from the post-sliding wear mechanisms is not directly measur-
able and was estimated by employing the wear model. A general comparison was
made between the sliding wear rates and previously [52] ascertained severe abrasive
wear rates of the materials under consideration. While for the aramid fibers, the N-
orientation resulted in the best wear resistance under both abrasion and sliding; in
case of the carbon fibers, the P-orientation was clearly superior to the N- and AP-
orientation under sliding wear conditions against smooth steel. Under abrasion, on
the other hand, N and P were almost equal and slightly superior to AP. A greater
increase in the wear resistances of the matrix materials due to fiber reinforcement
was evident under the sliding wear conditions, where PEEK resulted in much better
values than EP (Table 30.4). The authors referenced the wear rate results to the
wear rate of the unfilled epoxy, and utilized this database for the design of an
optimum wear-resistant composite material for specified wear conditions (i.e., a 3D
hybrid composite with AF in the normal direction and CF interwoven in AP- and
P-direction, Fig. 30.10) [32, 51].
Sliding wear of unidirectional CF-PEEK (XC-2, a special tribology grade of ICI
Fiberite [53]) was also the subject of a comprehensive study by Flöck [54, 55].
For most of his testing conditions, especially at room temperature, the P- and AP-
orientations were superior to the N-orientation, as far as the specific wear rate and
the coefficient of friction were concerned (Fig. 30.11).
For the same material, Varadi et al. [56, 57] analyzed the single asperity
sliding contact by finite element micro-models and scratch tribometer tests using a
diamond indenter. Under N-orientation, the equivalent stresses during the frictional
movement of the indenter (using a friction coefficient D 0.28 for P, D 0.3 for AP,
and D 0.45 for N) were much higher than under AP- and P-orientation. Thus, the
probability for an earlier failure of the composite surface elements (especially the
load-bearing fibers) was much greater for the N-orientations, so that also a higher
amount of wear under realistic conditions could be assumed (Fig. 30.12).
Tsukizoe and Ohmae [58–61] summarized the wear performance of seven kinds
of unidirectional oriented fiber-reinforced polymers (RP) with a volume fraction of
approximately 70 % in a diagram showing the relationships between the specific
30 Carbon Fibers in Tribo-composites 925
wear rate and the friction coefficient. As for the tribological anisotropy, it became
evident that every system had a good wear resistance in the parallel, but a poor wear
resistance in the antiparallel sliding direction. From the results in Fig. 30.13, it is
evident that HS-CFRP and HM-CFRP have a small specific wear rate of the order
of 107 mm3 /(Nm) and a low friction coefficient of 0.2. In contrast, steel (SF)-
and glass (GF)-reinforced polymers show a large specific wear rate of the order of
104 mm3 /(Nm) and a high-friction coefficient of 0.4.
The lowest friction coefficient, ca. 0.1, was obtained for CFRTP, having PTFE
as matrix. The good tribological properties of the carbon fiber-reinforced polymer
groups (CFRTP, HS- and HM-CFRP) may be caused by their good mechanical
properties, for instance, high Young’s modulus and high interlaminar shear strength,
as well as by the good tribological properties of the fibers, e.g., self-lubricating
926 K. Friedrich and L. Chang
Fig. 30.11 Specific wear rate of a continuous carbon fiber/PEEK composite sliding against
smooth steel [54]
Fig. 30.12 Distribution of equivalent (von Mises) stresses in a CF/PEEK composite when sliding
against steel under N- vs. P-orientation [56, 57]
ability and high strength. High-strength CF-reinforced epoxy with no fiber surface
treatment (NT-CFRP, with an assumed lower interfacial shear strength), on the other
hand, resulted in higher values of the specific wear rate.
For a further analysis, the authors [49] tried to correlate the wear behavior
of different kinds of composites against carbon steel with various mechanical
properties of the composite materials. The wear rate of their AF, GF and CF
composites increased with increasing coefficient of friction, , and decreased with
increasing interlaminar shear strength, Is , and increasing elastic modulus, E:
wsC D ˛.p=E/ˇ Is1 (30.4)
30 Carbon Fibers in Tribo-composites 927
The dimensionless constants ˛ and ˇ depend on the external conditions of the wear
system.
With respect to the frictional coefficient, , the authors derived a modified rule of
mixture by which they were able to calculate the value of the composite, C , from
the frictional coefficients of the fibers F and the matrix M [49]:
1 1 1
C D .1 VF / M C VF F (30.5)
Finally, these authors also looked at the results obtained with the carbon/glass
hybrid system, and they showed that there was a good agreement between the
calculated frictional coefficient and the experimental data. It became clear that
carbon fibers were the best reinforcement as far as the friction of fiber-reinforced
plastics was concerned. When utilizing a medium hybrid composite (VCF D 35 %;
VGF D 35 %) instead of a carbon fiber-reinforced material (VCF D 70 %, VGF D 0 %),
there was only a slight increase in the frictional coefficient. Thus, for a practical
application of composites where friction becomes an important problem, the use of
a hybrid material can be recommended in regard to the performance/cost ratio.
A similar conclusion can be drawn from wear results obtained by Hawthorne
[62] with glass/carbon-epoxy matrix hybrids. His CF specimen wear rates, at
107 mm3 /(Nm), agreed well with results reported for CFRP sliding against various
steel counterfaces [63–65]; the friction values were also compatible with previous
results. His GF composite wear rates, however, were an order of magnitude smaller
than the values obtained for sliding against a rougher mild steel counterface [65].
928 K. Friedrich and L. Chang
There were also small wear rate differences between the three fiber orientations,
but their ranking is not the same as those reported for sliding on other counterfaces
[63, 65].
For both single-fiber and hybrid-fiber composites, the decrease from high initial
wear rates to limiting values was consistent [66] with the evidence for alteration
of the counterface topography along the wear tracks by asperity removal and
transferred layers. For the hybrids, those with fibers oriented parallel to the sliding
direction exhibited the least amount of both “run-in” and total wear, and their steady-
state wear rates were closer to the CFRP rates than those of GFRP. This could be
explained by an early formation of the wear pattern where the CF laminae became
the effective bearing surface and thus mainly determined the hybrid sliding wear
characteristics. When, or precisely how, this duplex pattern developed has not been
ascertained, but the known superior fatigue resistance and less abrasive nature of
CFRP compared with GFRP were probably the cause.
After “run-in” the antiparallel oriented 1/1 hybrid-fiber specimens had the lowest
equilibrium wear rates, not much larger than those of the 100 % carbon composites.
Together with the fact that the relatively smooth wear surfaces resembled closely
those typical of worn CFRP specimens, this pointed to the dominating influence of
the CF layers once again. Since the low wear rate of CFRP is closely associated with
the formation of lubricating transferred films, it was suggested that this also applies
to the transverse-fiber hybrids. The small areas of attrition of these hybrid surfaces
are most likely the result of sliding fatigue wear [67]. The poorer fatigue resistance
of GFRP would then account for the greater incidence of these areas on the glass
fiber laminae.
The hybrids with fibers oriented normal to the counterface exhibited the largest
wear rates, but their worn surface appearance suggested only that wear might have
involved mild abrasion or surface fatigue processes [66, 68]. Transfer films from the
CF laminae may play a role here also, since many smooth areas were found, with
no back-transferred layers, which is characteristic of thin transfer film formation in
CFRP [66].
Frictional coefficients and wear rates of metals, polymers, and composite materials
are strongly influenced by the topography of the counterface against which they
are sliding. When the material slides against continuously fresh counterface (i.e.,
when no opportunity exists for sliding to modify the topography), the rates of wear
of the materials are approximately proportional to the square of the counterface
roughness (often expressed as a center line average (c.l.a.) value in Great Britain, an
AA value in the USA, but it is now to be known as the arithmetic average roughness
(Ra ) value to line up with ISO terminology) [69, 70]. The wear occurring under
this situation, which is, for example, relevant to the first revolutions against steel
counterparts (initial wear part of Fig. 30.14), is often abrasive in type and similar
30 Carbon Fibers in Tribo-composites 929
to that caused by hard abrasive particles. In the steady-state wear condition, on the
other hand, abrasive wear can also become the dominant mechanism, if the surface
of the wearing material has been modified during previous passages [71]. If such a
modification of the counterface topography results in an increase in the roughness,
often induced through the abrasive action of highly abrasive fillers (glass fibers) on
a soft counterface (aluminum), a transition from sliding to abrasive wear can occur
and thus increase the composite wear rate (see also some comparative figures for the
abrasiveness of different fillers in Table 30.3). It should be mentioned that the effect
of fillers in modifying the counterface surface may influence the wear behavior of
a composite in some cases far more than the improved mechanical properties of
the filled material. Hence, a suitable selection of filler and counterface material is
needed so that the wear rate can be reduced by the reinforcement of the polymer
matrix [14, 18].
Lancaster [26] has investigated possible transitions in wear behavior against
steel surfaces using different polymers with and without CF reinforcement. By
comparison of the wear data obtained in steady-state conditions on smooth steel
rings with those measured during single transversals over a roughened steel surface,
the author recognized that not only were the wear rates much greater than those
occurring during repetitive sliding, but also the CF reinforcement became almost
completely ineffective in reducing wear. In fact, for some polymers the wear
rates of the reinforced composites slightly exceeded those of the unfilled materials
(Fig. 30.15). Under steady-state conditions, the wear mechanism was presumably a
consequence of localized adhesion at the asperity contacts between the polymer
composite and either the steel or a transferred layer of polymer and fiber dust
on the steel. In the other case, the wear mechanism became primarily one of
penetration and microplowing of the hard steel asperities through the softer surfaces
of the polymers and composites, producing abrasive wear debris by microcutting,
microcracking of fiber bundles, or by low cycle fatigue. According to Briscoe et al.
[72], there is an inverse proportionality between the wear rate of polymers and the
product B "F , the breaking strength ( B ) times the elongation to fracture ("F ).
930 K. Friedrich and L. Chang
When fillers or reinforcing fibers are added to a polymer, the strength increases
by a factor of 1.5, but the elongation to fracture may decrease by a factor of up to
100. Following reinforcement of a polymer therefore, the product B "F is frequently
reduced, and this, in turn, leads to a reduction in its resistance to abrasive wear
[14]. Lancaster [26] could show that this reduction became especially deleterious
in the composites with a more ductile (e.g., PTFE, PP, PC, etc.) instead of a
relatively brittle matrix (e.g., EP, UP, PMMA, etc.). To illustrate, the wear rates were
correlated with data of fracture toughness and fracture energy [73, 74], respectively.
It was suggested that the main reason for the increase in wear rate of the polymers in
the first group is the relatively high reduction in ductility, and therefore in fracture
energy, caused by the presence of the fibers. The results of similar experiments
during single transversals over steel discs with different roughness confirm the
tendencies discussed more convincingly. The curves shown on Fig. 30.15 exhibit
for the ductile polymers a crossover which indicates that at very low roughness,
the carbon fibers improve the wear resistance while being deleterious at very rough
counterface [71].
Further studies on the transition from adhesive to abrasive wear mechanisms for
polymers and polymer composites against different counterparts were discussed in
[75–77]. Briefly, the resulting minimum in the wear rate of ultrahigh-molecular-
weight PE against the roughness of stainless steel counterparts was assumed to be
due to a change in wear mechanism, from adhesion in the range of very smooth
surfaces (average roughness Ra < 0.2 m) to mainly abrasion for Ra above 0.2 m
[75]. Czichos and Feinle [76, 77] found that the adhesive component at the interface
mainly governed the frictional behavior of polymer-polymer sliding pairs, while
the wear of polymers against hardened steel discs with a total surface roughness of
Rt 1 m is controlled by abrasive wear mechanisms. With respect to the influence
30 Carbon Fibers in Tribo-composites 931
of glass fibers on the friction and wear behavior of polyamide 6.6 against steel, the
authors showed that a normal fiber orientation to the sliding direction yields both
higher values of the resulting frictional coefficient and a more severe damage of the
steel counterpart. Additional analysis of the composite surfaces with EDAX gave
evidence for the fact that iron from the sliding steel counterface is transferred to the
polymer surface.
Previous work by Lancaster and coworkers on the friction and wear properties
of CF-reinforced thermosetting resins has been confined almost exclusively to
assessing their behavior during sliding against relatively smooth (< 0.25 m c.l.a.)
mild steel counterfaces [26, 41]. In these conditions, composites with type I and type
II fibers behave quite differently. The former exhibit an initially high rate of wear,
which subsequently decreases to a limiting value as a transfer film develops on the
steel counterface. In contrast, for composites with type II fibers, the initial high wear
period is much less pronounced, and transfer films do not form on the steel surface,
which becomes polished or worn. Both the coefficient of friction and the limiting
rate of wear of composites with type II fibers are greater than those obtained with
type I fiber composites.
It is known that the formation of a transfer film of wear debris on the surface
of the counterface plays a decisive role in determining the sliding performance
of PTFE-based composites [78], and it would appear that this is also true of
composites with thermosetting resin matrices. Other work, on the wear of carbon
brush materials [79] and on the lubricating ability of thin films of lamellar solid
lubricants [80], suggests that the formation of transfer films on a counterface is
dependent on the type of counterface (its hardness and chemical reactivity) and on
its surface roughness. Therefore, Giltrow and Lancaster [81] considered it important
to examine the role of these factors in the friction and wear processes of CFRPs and,
in particular, to attempt to clarify the part played by the transfer film. Attention has
been restricted to thermosetting resin matrices with high CF contents (>40 vol.%)
because these materials, according to their previous findings, exhibited superior
wear resistance to composites with thermoplastic matrices [40].
Various counterface materials were chosen to examine as wide a range of
hardness and chemical type as possible; the hardness values are listed in Table 30.5.
Attempts were made to prepare all the materials to a surface roughness within the
range 0.10–0.20 mm c.l.a., by abrasion on an appropriate grade of carborundum
paper. The roughness of the softer metals was usually close to the upper limit of this
range, whereas that of the harder ones was nearer the lower limit.
932 K. Friedrich and L. Chang
low (0.13/0.20), a transfer film developed, and wear of the counterfaces was
negligible. On other metals, the coefficient of friction was high (0.4/0.7), no
transfer film developed, and counterface wear could be extremely severe.
4. Wear of type I fiber composites was critically dependent on the surface roughness
of the counterface, whereas that of composites with type II fibers was not. On
smooth surfaces, type I composites gave lowest wear, while on rough surfaces,
type II composites should be preferred.
5. No systematic trend with hardness could be discerned.
6. Stainless steels (containing >12 % Cr) should be considered as excellent coun-
terfaces for composites with either type of CF.
Later, Jacobs et al. studied [82] the effect of various counterpart materials (i.e.,
hardened steel, austenitic steel, and Al2 O3 ), internal lubricants (PTFE, graphite,
MoS2 , and SnS2 ), and fiber reinforcements (glass and carbon fibers) on the
wear of epoxy-based composites. Using a ball-on-prism test system, the high
chromium-containing austenitic steel (German standard X5CrNi18-10) led (under
dry conditions) to a lower composite wear than the hardened ball bearing steel
(German standard 100Cr6). Alumina counterparts produced results similar to the
austenitic steel. Only PTFE provided a remarkable wear reduction, while all other
fillers had no significant effect regardless of the counterpart material. Under aqueous
conditions, the corrosion resistant and electrically isolating Al2 O3 ceramic seemed
to be the most promising counterpart. Here, the CF-reinforced version resulted in
the best wear performance.
Previous investigations into the friction and wear properties of CFRPs have been
mainly concerned with assessing their behavior during unlubricated sliding [40,
42, 80]. In these conditions, the performance of polymers and polymer-based
composites is limited either by the magnitude of the wear rate or by frictional
heating leading to softening and/or degradation of the polymers [83]. Both of these
factors are sensitive to the presence of fluids. For example, some polymers, such
as nylon and polyacetal, show marked reductions in wear when used as bearings
in conditions of marginal lubrication by oils and greases [84]. Frictional heating is
affected by fluids either indirectly, by reducing the coefficient of friction, or directly
by acting as heat-transfer media. In the latter respect, water is a fluid of particular
interest, provided that the friction and wear properties of the sliding materials
in water remain satisfactory. This was, in fact, well realized by Lancaster et al.
[85] who therefore decided to survey generally the effects of fluids on the friction
and wear of polymers and polymer-based composites, with particular reference to
CFRPs. In the first part, results obtained with water and aqueous solutions were
described; a subsequent part covered organic fluids.
There are numerous sliding applications in which water is either deliberately
introduced as a coolant, e.g., in rolling mill bearings, or is present as the working
934 K. Friedrich and L. Chang
fluid, e.g., in pumps. In the latter case, a marine environment can pose special
problems because of abrasive contamination and the corrosiveness of seawater
toward many metals [86]. Previous work on the performance of different mate-
rials sliding in water has been largely concerned with the operation of specific
components—mainly bearings—operating in conditions which were at least par-
tially hydrodynamic [84, 87]. Many carbons and polymer-based composites often
give an acceptable performance in these conditions, and it is therefore reasonable
to suppose that CFRPs would do likewise. However, from the data reported in the
literature, it is difficult to discern any significant pattern in results to indicate why
one material is able to exhibit lower wear in water than another. The only general
conclusion which can be drawn is that the wear rates of carbons and polymers in
water can sometimes be appreciably greater than those obtained during dry sliding
[84, 88]. Therefore Lancaster and his team restricted their testing conditions of
sliding, as far as possible, to the boundary regime. In this way, by minimizing
hydrodynamic effects, a clearer picture could be obtained of the major factors
affecting the wear of reinforced polymers in water. The corresponding rates of
wear of a number of CFRPs in water, together with those of other composites and
unreinforced polymers, are shown in Fig. 30.17. Results are also given for the wear
rates of the materials in dry conditions at similar loads and speeds. The following
points should be noted here:
1. The wear rates of all the materials in water are greater, and often very much
greater, than those in dry conditions.
2. Composites containing type II fibers are generally superior to those containing
type I.
3. The most important factor affecting the rate of wear of CFRPs during sliding in
water and other aqueous solutions is the topography of the counterface.
4. The counterface topography generated during sliding depends on the abrasive-
ness of the composite, the counterface hardness, and the corrosiveness of the
surrounding fluid toward the counterface.
5. The wear rates of CFRPs on one particular stainless steel can be significantly
lower in seawater and other aqueous solutions than in pure water. This was also
found by other researchers [89] for CF/PEEK composites.
6. Water and all aqueous solutions inhibit the formation of transfer films on a
metal counterface, and the wear rates of CFRPs are generally greater than those
occurring during dry sliding.
7. The addition of small amounts of abrasive to CF-reinforced polymers can greatly
reduce their rates of wear in water. Abrasive additions tend to increase the
composite wear rate under dry conditions, however, and may also increase wear
of the counterface [85].
The work described by Lancaster et al. in the following part extended the
investigation to organic fluids, including mineral oils, synthetic lubricants, and
hydraulic fluids. As before, the major objectives were to determine the most
important factors involved in the lubricated wear process and to compare the
performance of CFRPs with that of other materials [90]. Figure 30.18 shows that
30 Carbon Fibers in Tribo-composites 935
Fig. 30.17 Wear rates of different materials sliding on stainless steel of 0.15 m c.l.a. roughness.
Full lines water lubricated, dashed lines unlubricated [85]
a variety of organic fluids, including mineral oils and synthetic lubricants, are
effective boundary lubricants for CFRPs sliding against stainless steel. The wear
rates of the composites in some fluids were extremely low, and scanning electron
microscopy as well as pyrolysis-gas chromatography gave evidence that low wear
was associated with the development of films on the steel counterface. These films
appear to be mixtures of lubricant and resin from the composite matrix. The wear
rates of epoxy/type II CF composites in aircraft hydraulic fluids were generally
much less than those of many metal combinations, and the composites therefore
showed promise as replacements for metals in lubricated systems [90].
In a recent study by Ga Zhang et al. [91], the tribology behavior of polymer
composites was studied under boundary lubrication conditions with diesel as lubri-
cant. In particular, the effects of short fibers and filler combinations, i.e., fibers/solid
lubricants and fibers/nanoparticles, on the sliding behavior of EP against a 100Cr6
counterbody were investigated with the presence of limited diesel (controlled flow
rate 20 l/h). The following conclusions were drawn by the authors:
936 K. Friedrich and L. Chang
Fig. 30.18 Wear rates of epoxy/type I fiber composites on 0.15 m c.l.a. Stainless steel in various
fluids [90]
Fig. 30.19 Effect of diesel injection on friction behavior of an optimal PEEK composite: friction
coefficient vs. time (diagram above); transfer film built-up on counterface under dry sliding (lower
left); removal of transfer film after injection of diesel (lower right); contact pressure, 5 MPa, speed
0.5 m/s, counterbody temperature kept constant at 70 ı C [91]
1. Even a very small quantity of diesel decreased obviously the friction coefficient
and wear rate of pure PEEK. It was identified that the formation of a transfer film
was of importance for getting good boundary lubricity.
2. The addition of 2 L/h diesel significantly increased the friction and wear of the
PEEK composites due to the fact that diesel hindered the formation of a high-
performance transfer film. Nevertheless, an increase of diesel quantity reduced
gradually the friction and wear.
3. With engine oil lubrication, a stepwise decrease of sliding speed increased the
friction coefficients of PEEK materials. During the running-in phase of each
speed step, the friction coefficients of PEEK materials decreased due to further
development of a transfer film structure.
4. Being lubricated with engine oil, PEEK-based materials, in particular the PEEK
composites, exhibited much higher wear resistance than St50-2 steel.
Vacuum as well as elevated or lower temperature environments were investi-
gated by other authors. For example, Sawyer and his team [93] investigated the
tribological behavior of polymer composites in space. Hanchi and Eiss [94] studied
the sliding wear of CF/PEEK at elevated temperatures. Researchers at the BAM—
Institute in Berlin, Germany—studied together with the group of Friedrich [95, 96]
the friction and wear of CF-reinforced PTFE composites against steel at cryogenic
temperature conditions. They showed that the friction coefficient decreased with
temperature down to 77 K, but did not follow a linear evolution further down to
938 K. Friedrich and L. Chang
extremely low temperatures. It could be stated that the cryogenic environment had a
significant influence on the tribological performance of the polymer composites.
The effect of low temperatures was more clearly detected at low sliding speed,
where friction heat was reduced. A change in wear mechanism from adhesive to
abrasive was observed in this case. SEM and AFM analyses showed that the PTFE
matrix composites investigated under these experimental conditions had transferred
material onto the disc down to very low temperatures. Chemical analyses indicated
the presence of iron fluorides (Fig. 30.20). Similar results were achieved with PEEK
matrix composites containing PTFE particles and SCF as fillers/reinforcements
(Fig. 30.21).
Theiler and Gradt [97] recently investigated the friction and wear behavior of
PEEK composites filled with carbon fibers, PTFE, and graphite or MoS2 in rotary
Fig. 30.20 (a) Coefficient of friction of three different PTFE composites at RT and 77 K [96];
(b) specific wear rate of three different PTFE composites at RT and 77 K [96]
30 Carbon Fibers in Tribo-composites 939
Fig. 30.21 Specific wear rate of a PEEK composite under various low-temperature testing
environments [96]
motion against CrNi steel, using vacuum and a temperature range between 80 ı C
and C20 ı C as the testing environment. It turned out that the sliding behavior
of MoS2 -filled PEEK showed a better tribological performance compared to the
composite with graphite, but only at low temperatures. Surface analyses show that
the lower friction coefficient and wear rate of the MoS2 -filled composite at 80 ı C
was associated with a thin polymer transfer film on the counterface, with a higher
concentration of MoS2 at the surface of the composite. Furthermore, the sliding
speed had a significant influence on the friction and wear performance of MoS2 -
filled PEEK. Results indicated that there was a transition from severe wear to mild
wear between 0.01 and 0.1 MPa m/s. At room temperature in vacuum, the optimal
performance was obtained at pv D 1 MPa m/s.
That the previously mentioned corrosion effects can not only occur with metals
[86] was more intensively studied by Gebhard et al. [98, 99] for polyacrylnitrile
(PAN) and pitch-based short carbon fiber (SCF)-reinforced PEEK composites under
conditions with and without sliding operation. When an electric bias voltage is
applied to a PEEK/SCF composite with a fiber volume content of 30 vol.% being
intimately contacted with a X10CrNiMoTi1810 stainless steel counter electrode,
corrosion of the composite’s carbon fibers is observed (Fig. 30.22). Although this
effect also takes place without an aqueous medium, fiber corrosion is significantly
increased by up to one order of magnitude in the presence of an electrically
conductive medium. At the same time, there is no evidence found for an increase
of the corrosion phenomenon by the presence of potential oxidation agents like
hydrogen peroxide or nitrate. It is therefore concluded that the observed corrosion
is mainly an electrochemical degradation reaction that occurs at the fiber/steel
interface. An additional indication for the corrosion to be a purely electrochemical
phenomenon is that if the corrosion tests are repeated with a PEEK/pitch-SCF
940 K. Friedrich and L. Chang
10
20
30
Digital Instruments NanoScope
40 Scan size 50.00 µM
µM Scan rate 1.001 Hz
Number of samples 256
Image Data Height
Data scale 2.500 uM
Fig. 30.23 Steady-state corrosion depths of PAN-based carbon fibers in different aqueous media
of different electrical conductivity [98]
the steel/fiber contact to take place. In contrast to this pronounced impact of fiber
corrosion on wear, no influence on the coefficient of friction was detected.
In the absence of fiber corrosion, the increase of the composite’s fiber volume
content from 10 to 30 % yielded an improvement of the specific wear rate. This
however also increased the material’s electric conductivity which rendered them
vulnerable to fiber corrosion. Consequently when an electric bias voltage was
applied to the 30 vol.% fiber materials, they exhibited extensive fiber corrosion
which significantly increased the wear rate relative to corrosion-free testing condi-
tions. Since the materials which were filled with only 10 vol.% of fibers are electric
insulators, they were found to be immune to galvanic fiber corrosion and had under
corrosive conditions a lower wear rate than their higher fiber counterparts.
In technical applications where CF corrosion is considered to occur, the first
option for avoiding an increase in wear of a metal/CFRP slide pairing should
be the prevention of residual currents that flow through the sliding interface by
standard means of electrical engineering. If this cannot be guaranteed, there are
options of using a composite with an intrinsically higher resistance to galvanic
corrosion, i.e., using PAN instead of pitch fibers or using fibers with electrically
insulating finishing. Finally, if the application’s requirement concerning mechanical
performance can still be met, the same fiber/matrix combination could be used but
with a lowered fiber volume content.
942 K. Friedrich and L. Chang
Fig. 30.24 (a) Corrosion depth of carbon fibers under normal sliding, under electro-corrosion, and
under sliding wear test with an applied electrical current [99]; (b) PAN-based short carbon fibers of
a 30 % CF/PEEK composite subjected simultaneously to galvanic corrosion and tribological stress
[99]
30 Carbon Fibers in Tribo-composites 943
Fig. 30.25 Schematic design of a CF/polymer composite with low friction and high sliding wear
resistance, especially by additional use of ceramic nanoparticles [100]
944 K. Friedrich and L. Chang
Fig. 30.26 Specific wear rate of composites with and without 300 nm TiO2 particles [108]
Chang et al. [105, 108] systematically studied the effect of nano-TiO2 on short-
fiber-reinforced epoxy under different loading conditions. It was found that the
addition of 5 vol.% nano-TiO2 could significantly reduce the friction coefficient and
the wear rate of epoxy composites filled only with traditional fillers (Fig. 30.26).
Guo et al. [109] used surface-modified nano-SiO2 filler to enhance the tribo-
properties of epoxy composites filled with short carbon fibers. In order to improve
the interface bonding between nanoparticles and the polymeric matrix, the nanopar-
ticles were pretreated by graft polymerization. Again, the additional nanoparticles
(4 wt.%) proved to be useful in enhancing the wear resistance and reducing the
friction of the SFRPs. The authors indicated that the improvement mechanisms of
nanoparticles could be caused by the increased strength of the matrix and better
properties of the transfer film. These and other results, e.g., from the Lanzhou
Institute of Chemical Physics in Lanzhou, China [110–115], have clearly shown
that the addition of nanoparticles is potentially useful to improve the tribological
performance of SFRPs, even at a relatively low content.
In a recent study by Chang and Friedrich [116], the wear mechanisms of short
pitch-based carbon fiber-reinforced polymers (Epoxy, Polyamide 66) filled with
TiO2 nanoparticles were investigated in more detail. It was found that the load-
carrying capacity of these materials is also mainly determined by the properties of
the fibers. However, their tribological performance can be significantly improved
by using nanoparticles due to their friction-reducing abilities, especially under
30 Carbon Fibers in Tribo-composites 945
Fig. 30.27 Wear mechanisms in short carbon fiber-reinforced polymer composites: deep scratches
and fiber/interface damage without the use of additional nanoparticles (upper row). Nano-
polishing, topographic smoothening, and stabilization of the transfer film with ceramic nanoparti-
cles (lower row) [116]
nanoparticles and traditional tribo-fillers were more effective to form durable TFLs
on the steel counterpart, associated with desirable tribological properties of the
sliding system, i.e., a low friction coefficient and a low wear rate. However, it
was noticed that the distribution of the TFLs on the steel counterpart was not
always uniform. By studying the load-displacement behavior of polymeric TFLs
on metallic substrates, the thickness of TFLs could be estimated, thus, allowing the
comparison of TFLs formed under different sliding conditions in a quantitative way.
Based on the experimental data, the effects of TFLs on the tribological performance
of polymer composites were further discussed in terms of a “transfer film efficiency
factor” œ, which was calculated by the ratio of the average thickness of the TFL
to the surface roughness of the steel counterpart. The factor mainly considered
the relative contributions of the TFL and the metallic counterface to the wear
process of the polymer-on-metal system. Accordingly, the wear rate and the friction
coefficient of the sliding system could be analyzed as a function of the transfer film
efficiency factor, resulting in a Stribeck type diagram (Fig. 30.29), similar to as it
was suggested some time ago by Czichos for metallic materials [118].
As a practical example, the development of a high-temperature resistant slide
coating on the inside of steel-backed hybrid bushings (as used in automotive diesel
fuel injection pumps) can be mentioned [119, 120]. These coatings consisted of
a high-temperature PEEK matrix and short carbon fibers, graphite, as well as two
different kinds of nanoparticles as reinforcing fillers (Fig. 30.30).
Their friction and wear properties exceeded those of the previous commercial
PTFE-based reference. The latter had caused failure and a recall action of the diesel
30 Carbon Fibers in Tribo-composites 947
Fig. 30.29 Specific wear rate and friction coefficient of EP and PA66 based SCF composites as a
function of transfer film efficiency factor [117]
fuel injection pumps used by highly ranked German automobile companies, e.g.,
BMW (http://www.t-online.de/auto/technik/id_12826488/diesel-einspritzpumpen-
probleme-bei-zulieferer-bosch.html).
Fig. 30.31 Experimental (black points with error bars) vs. predicted (3D mesh) values of the
coefficient of friction (average value in steady state) as a function of SCF- and submicroTiO2 -
volume contents: (a) training dataset, (b) validation dataset. Test conditions: p D 1 MPa, v D 1 m/s
[129]
950 K. Friedrich and L. Chang
Fig. 30.32 Experimental (black points with error bars) vs. predicted (3D mesh) values of the
specific wear rate as a function of SCF- and submicroTiO2 -volume contents: (a) training dataset,
(b) validation dataset. Test conditions: p D 1 MPa, v D 1 m/s [129]
30 Carbon Fibers in Tribo-composites 951
the continuous range of carbon fiber loading fraction, namely, 0–15 vol.% using an
increment value of 1 vol.% prior to applying the “meshgrid” command in Matlab.
Using a smaller increment of fiber volume fraction avoids the sudden drop in the
predicted specific wear rate data since the wear rate decreases rapidly as the SCF
volume fraction increases. As for the concentration range of TiO2 particles, for all
the cases, this was kept constant, namely, 0–8 vol.% with 1 vol.% increment value.
For the entire training test data, the predicted results are either correct or very close
to the actual measured values. At this stage it can be concluded that the training
of the network is satisfactory, i.e., the network has learnt the complex relationship
between the selected input and the corresponding output for the examples presented
in the training dataset.
The developed and trained ANN model until now successfully mapped the
relationship between various input parameters and the output specific wear rate
values. Still, for its practical use, it must be checked if it is also able to generalize
this relationship. The validation of the network model for the specific wear rate is
presented in Fig. 30.32b. All validation test points were predicted exactly (or within
the error band limits). These results indicate that the developed ANN model for
predicting the specific wear rate is both reasonably correct and able to generalize
well.
At this stage, it is possible to use the excellent generalization capability of
the optimized ANN models to analyze the impact of the two fillers, namely, SCF
and submicro TiO2 , on the sliding friction and wear properties of PPS composites
over a much broader range without performing exhaustive experimental work. It
is obvious from Fig. 30.31 that the combination of the two fillers results in a
lower coefficient of friction. Although the mechanisms of reduction in coefficient
of friction contributed by each of the two fillers are different, it is clear that they
can work together simultaneously. As shown already experimentally [128], short
carbon fibers play a key role in improving the wear resistance of PPS, as a single
or second reinforcing phase. It is predicted that the specific wear rate decreases
with increasing volume fraction of SCF, but however less steeply at larger volume
fractions (7 vol.%). With the incorporation of both short carbon fibers and TiO2
submicro particles in PPS matrix, a hybrid reinforcement effect can be found. For
higher SCF concentrations (8 vol.%), the specific wear rate exhibited a drop-off
tendency with the increase of TiO2 content in the range between 1 and 8 vol.%.
This type of relationship is in agreement with the experimental results of Zhang and
Friedrich [132] for epoxy composites. Such favorable synergistic effects between
carbon fibers and submicro TiO2 have also been foreseen by ANNs in PTFE
matrix composites [133]. According to both the experimental results and the ANN
prediction, the composition of PPS with 15 vol.% SCF and 8 vol.% TiO2 (300 nm)
gives the best wear resistance (i.e., the lowest specific wear rate).
952 K. Friedrich and L. Chang
Fig. 30.34 Mechanical and functional values of various PPS composites in comparison [135]
toughness, strength, and stiffness of the composites, whereas MWNTs lead only
to marginal improvements. On the other hand, a very good dispersion of MWNT
agglomerates resulted in higher values of electrical conductivity and an early
percolation threshold Vth (Vth,MWNT D 1.8 wt.%) in comparison to the use of SCF in
PPS only (Vth,SCF D 6.4 wt.%). A combination of MWNT and SCF gave, in addition,
a synergistic effect in the volume conductivity, because the electrical pathways
through the system were overlapping now. The best performance regarding low
friction and low wear properties were found for a multimodal PPS composite
containing 3 wt.% MWNT, 10 wt.% Gr, and 10 wt.% SCF. Figure 30.34 depicts
the property values in a spider diagram. It demonstrates that a multimodal filler
combination of MWNT, Gr, and SCF results in the best overall property profile,
in particular concerning electrical and tribological performance. With the new
findings, it is possible to tailor the overall property profile of PPS composites by
using optimized filler combinations, whereby the multifunctional property profile
can be enhanced beyond that of single filler systems.
Quite recently, Golchin et al. [136] described the tribological behavior of these
PPS composites in water-lubricated sliding contact against a metallic (Inconel)
counterpart. The results showed that MWNT, Gr, and SCF were all effective in
reduction of friction and wear of PPS. However, SCF was the most efficient filler
as a reduction in wear of more than three orders of magnitude and a reduction in
friction of more than 60 % was obtained, relative to the neat PPS.
Examination of the worn Inconel discs revealed the occurrence of polymer
transfer onto the counter surfaces, whereby this mechanism was more pronounced
954 K. Friedrich and L. Chang
for the short carbon fiber-filled PPS composites in comparison to the other polymer
composites. X-ray photoelectron spectroscopy (XPS) of the “interface” between
the transferred material and the basic Inconel surface gave evidence for a chemical
reaction of the mating partners, as confirmed by the formation of inorganic sulfide
at this “interface.” However, mechanical interlocking is believed to be the more
dominant mechanism for polymer transfer onto the Inconel surfaces.
Further investigations utilizing scanning electron microscope (SEM) revealed the
formation of smooth nano-tribo-layers on load-bearing areas of Inconel surfaces
when sliding against SCF-reinforced PPS composites.
It is believed that a combination of the abovementioned mechanisms, namely,
the exposure of high wear-resistant carbon fibers, the formation of groove-filling
polymer transfer spots, and the deposition of smooth nano-tribo-layers on the
Inconel counterpart contribute to the low friction and wear characteristics of SCF-
reinforced PPS composites under water-lubricated conditions.
Further studies on the effect of counterface roughness [137] gave evidence that
the friction and wear behavior of SCF-filled PPS composites showed marginal
variation with change in topography of the Inconel counter surfaces with regard
to roughness parameters and/or lay orientation. On the other hand, the composites
without SCF exhibited up to two orders of magnitude increase in wear rate with
sliding against rough counterfaces.
Lancaster [42] stated that composites for high-energy brakes comprise a group
of materials whose requirements differ in several ways from those intended for
bearings, e.g., stability of the friction coefficient with varying load, speed and
temperature, high thermal capacity, resistance to thermal shock, etc. Although low
wear rates are, of course, important, the values which can usually be tolerated
in service tend to be somewhat greater than those acceptable for dry-bearing
applications. One of the most significant advances of recent years has been the
development of carbon-carbon composites for aircraft brakes [138], and this has
prompted a number of investigations into the ways in which performance is
influenced by the structure and composition of the materials [139–141]. The two
main problems appear to be the tendency for graphitic materials to undergo rapid
wear at elevated temperatures [142] and for non-graphitic materials to exhibit
large fluctuations in friction. Both of these problems can largely be overcome by
the introduction of additives. Rapid wear at high temperatures is associated with
the disruption of the third-body films, and additions of phosphorous-containing
additives assist in maintaining these films [143]. These additives also have the
30 Carbon Fibers in Tribo-composites 955
additional merit of reducing the rate of oxidation. The stability of the friction
coefficient with non-graphitic composites is to some extent dependent on their
structure, and a comparison showed that composites based on carbon fiber cloth
were less susceptible to friction fluctuations than those based on random mats.
Additives were again likely to be beneficial. It has been suggested [144] that many of
the inorganic additives which prevent the build-up of high friction in non-graphitic
carbons do so because they are mildly abrasive and maintain the real contact area in
a uniformly subdivided state. Only very small amounts of abrasive material can be
tolerated; however, too much leads to increased composite wear [140].
Bijwe [145] explained quite recently that the performance requirements from
the friction materials (FM) are so complex and conflicting that its design and
formulation has been well accepted as a multi-criteria optimization problem. It has
to fulfill more than a dozen of major performance parameters, which are most of
the times conflicting and so sensitive to the selected type and amount of ingredients
that a small change in one of the ingredients, even in small amount, may affect other
parameters adversely. The performance of such multifunctional FM is achieved by a
judicious choice of right ingredients (generally around 15–20) in right combination
and amounts with right shapes, orientations, and distribution in the matrix with right
manufacturing process. Hence, industrial formulations are proprietary and designing
is still considered as an art rather than science.
A model friction material was formulated by Gopal et al. [146], containing
a cashew-modified phenolic resin matrix, short carbon fibers, phenolic particles,
barytes, and steel fibers. The friction, wear, and fade characteristics of this material
were determined using a Chase friction material testing machine. The coefficient
of friction was found to vary between 0.2 and 0.5 with lower values associated
with higher loads, speeds, and drum temperatures and vice versa. Conditioning the
specimens with several fade-recovery test cycles resulted in steady friction during
subsequent fade tests followed by excellent recovery characteristics. The specific
wear rate (volume loss per unit load and sliding distance) decreased with increasing
loads, but increased with increasing drum speeds and temperatures due to thermal
degradation of the resin. The carbon fiber-reinforced friction material showed lower
specific wear rates than that of a milled-glass-fiber-based friction material at low
speeds and temperatures over a wide load range.
Recently a carbon fiber-reinforced paper-based friction material (CFRPF), as a
new type of wet friction material for automatic transmission, was prepared by a
paper-making process [147]. The frictional response of CFRPF was highly complex
under a set of dynamically variable operating conditions. To better understand the
effect of operating factors (braking pressure, rotating speed, oil temperature, and
oil flow rate) on friction stability of the material, tests were carried out using a
single ingredient experiment and the Taguchi method. Experimental results showed
that the braking stability and the dynamic friction coefficient (d ) decreased as
braking pressure, rotating speed, oil temperature, and oil flow rate increased. The
influence of braking pressure on d was largest among the four operating factors. d
declined gradually during the first 3000 repeated braking cycles and changed very
little subsequently due to the surface topography change of the friction material.
956 K. Friedrich and L. Chang
A new composite material has been developed by Gibson et al. [148] for brake
and clutch friction applications. This material was comprised of carbon fiber-
reinforced carbon matrix composite which has been subjected to high processing
temperatures. The key attributes of this material were high coefficient of friction,
low wear, low density, excellent stability at high temperatures, and consistency of
performance when using the same material against itself. Carbon/carbon friction
materials have now progressed beyond aircraft brake rotors/stators to automotive
racing car brake rotors and pads and clutch plates to possible production automotive
and truck applications.
Other studies [149–152] gave also evidence that the research on high-friction
materials is still a topic of interest and further developments are underway. As
an example, Xiong et al. [153] investigated the friction behavior of carbon/carbon
(C/C) composites for aircrafts under normal and overload landing (NL, OL) and
rejected take-off (RTO) conditions. Three kinds of composites were made from a
three-dimensional needled carbon fiber preform, which was fabricated by needling
the stack of alternating layers of unidirectional nonwoven carbon cloth and carbon
felt made from chopped fibers. Preforms of all the samples were densified by
CVI in C3 H6 C N2 atmosphere. The densities of the C/C composites were 1.82–
1.85 g/cm3 , and volume fraction of carbon fibers in the materials was 33–35 %.
The specimens were finally heat treated at 2300 ı C. The authors showed that C/C
composites with a high- and medium-textured rough laminar (RL) pyrolytic carbon
had stable friction coefficients, friction curves (Fig. 30.35a), and reasonable wear
loss at various braking energy levels, while friction coefficient of C/C composites
with smooth laminar (SL) pyrolytic carbon dropped sharply and the oxidation loss
increased greatly at overload landing and rejected take-off level. Therefore, only
C/C composites with RL pyrolytic carbon could satisfy the requirements of aircraft
brakes.
The good friction and wear properties of C/C composites with a high- and
medium-textured RL pyrolytic carbon were due to the characteristics of RL, which
lead to a uniform friction film formed on the friction surfaces of the C/C composites
during different braking levels (Fig. 30.35b).
Glass and glass-ceramic systems reinforced with ceramic fibers are possible replace-
ments [154] for metallic materials in gas turbines because of their good specific
strength and stiffness, high-temperature stability, and corrosion resistance, but,
surprisingly for materials envisaged for use in an engine with moving parts, only
limited wear studies have hitherto been carried out. Some data are available on glass
reinforced with carbon fibers [155, 156], showing that the material exhibited the
good wear resistance of glass while retaining the lubricating capability of the carbon
fiber, that its friction coefficient depended upon the modulus of the CF, HM fibers
giving lower values of friction coefficient, and that fiber orientation affected wear
30 Carbon Fibers in Tribo-composites 957
Fig. 30.35 (a) Friction curves of a C/C composite with RL structure under different braking
conditions (NL, OL, RTO). The typical noise vibration in each curve was suppressed for better
clarity [153]; (b) SEM morphologies of worn surfaces under different braking conditions: NL
(left) and RTO (right) of a high-textured RL C/C composite [153]
rate [157]. The latter is shown on Fig. 30.36, demonstrating that the AP-orientation
resulted in the best wear performance. The corresponding worn surface, observed at
a nominal pressure of p D 1 MPa and a sliding speed of v D 1 m/s, is illustrated on
Fig. 30.37 [156].
Other tribology works, carried out by Reinicke et al. [158, 159], were focused
on the dry sliding wear behavior of discontinuous, randomly orientated SiC- and C-
fiber-reinforced glass matrix composites against cast iron at high sliding speeds.
A special objective was to show the influence of (a) different fiber proportions
in SiC/C-hybrid composites (from 100 % SiC- to 100 % C-fibers) and (b) various
incorporated fillers (such as graphite Gr and ZrO2 powder) on the tribological
properties of the materials tested. Their findings suggest that the addition of
SiC-fibers takes a positive influence on the wear performance of glass matrix
composites. However, the hard ceramic fibers caused severe damage to the cast-iron
counterpart. For the hybrid composites, a correlation between friction and wear was
observed. Increasing the coefficient of friction always resulted in higher wear rates.
958 K. Friedrich and L. Chang
Fig. 30.37 SEM micrograph of a worn CF/glass sample tested at p D 1 MPa and v D 1 m/s in AP
direction [156]
The C-fiber-reinforced glass exhibited the highest coefficient of friction and wear
rate. The importance of the impregnation quality was clearly evident for the
graphite-modified 80 % C/20 % SiC composite. A remarkable reduction of friction
and wear was observed for a better impregnation quality of these composites. The
Gr- and ZrO2 -modified composites also showed a correlation between friction and
wear. For the glass matrix composites modified with ZrO2 powder, coefficients of
friction slightly above 0.4 were measured, whereas the composite with graphite
powder had a higher coefficient of friction (0.58) than the unmodified 80 % C/20 %
SiC system (Fig. 30.38). Such materials could be considered as replacements for
those currently used in automotive clutches.
30 Carbon Fibers in Tribo-composites 959
Fig. 30.38 (a) Manufacturing process of the glass matrix composites [159]; (b) tribological
behavior of modified glass matrix composites [159]
lubrication, were appreciably lower than those of the parent metals (Fig. 30.38).
The coefficients of friction were not reduced, however, nor were the wear rates at
elevated temperatures (470 ı C). It could be concluded that the carbon fibers function
by facilitating the development of a strengthened surface layer during sliding. The
fibers did not appear to improve the bulk mechanical properties (Fig. 30.39).
The friction and wear properties of copper matrix composites reinforced with
short carbon fibers (CFs) were studied by Tang et al. [161], since Cu/CFs composites
are promising functional materials used as electrical packing, electrical devices, heat
exchangers, etc., due to the good properties of their components. The influence of
short CFs contents, load, and rotating speed on tribological behavior of composites
was discussed. The results indicated that the wear resistance of Cu/short CFs
composites was greatly improved (compared to the pure copper). With a higher
CF content, the microhardness increased, and the friction coefficient as well as the
wear mass loss decreased. Both friction coefficient and wear mass loss increased
with increasing load and rotating speed, respectively. For pure copper and Cu/short
CFs composites, the wear mechanisms were adhesive wear and delaminating
wear, respectively. Oxidative wear also occurred slightly for Cu and Cu/short CFs
composites. CFs improved the tribological behavior of the Cu/short CFs composites
by hindering deformation, impeding the softness of copper, and forming a graphite
film on the worn surface.
30 Carbon Fibers in Tribo-composites 961
Acknowledgments One of the authors, Dr. L. Chang, wishes to thank the Australian Research
Council for the financial support of his research works.
References
68. Z. Eliezer, V.D. Khanna, On the effect of fiber orientation on the wear of composite materials.
Wear 53, 387–389 (1979)
69. V.K. Jain, S. Bahadur, Surface topography changes in polymer-metal sliding. J. Lubr. Technol.
10, 520–525 (1980)
70. A.E. Hollander, J.K. Lancaster, An application of topographical analysis to the wear of
polymers. Wear 25, 155–170 (1973)
71. K. Friedrich, Friction and wear of polymer composites. Fortschr.-Ber. VDI Z., Reihe 18, Nr.
15, (1984)
72. B. Briscoe, Wear of polymers: an essay on fundamental aspects. Tribol. Int. 14(4), 231–243
(1981)
73. K. Friedrich, The effect of morphology on the strength and fracture processes of polypropy-
lene. Kunststoffe-German Plast. 69(11), 17–19 (1979)
74. H.H. Kausch, Polymer Fracture (Springer, New York, 1978)
75. J.C. Anderson, High density and ultra-high molecular weight polyethylenes: their wear
properties and bearing applications. Tribology 15, 43–47 (1982)
76. H. Czichos, Influence of adhesive and abrasive mechanisms on the tribological behavior of
thermoplastic polymers. Wear 88, 27–43 (1983)
77. P. Feinle, Tribologisches Verhalten von gefüllten Kunststoffen und Verbundwerkstoffen. Z.
Werkstofftechn. 14, 342–350 (1983)
78. J.P. Giltrow, J.K. Lancaster, The Friction and Wear of Carbon Fiber Reinforced PTFE. In:
Proc. USAF-MWRL Conf. on Solid Lubrication, AFML-TR-70–127, (1970), pp. 305–331
79. J.K. Lancaster, The influence of the conditions of sliding on the wear of electrographitic
brushes. Br. J. Appl. Phys. 13, 468–477 (1962)
80. J.K. Lancaster, Lubrication by transferred films of solid lubricants. ASLE Trans. 8, 146–155
(1965)
81. J.P. Giltrow, J.K. Lancaster, The role of the counterface in the friction and wear of carbon
fiber reinforced thermosetting resins. Wear 16, 359–374 (1970)
82. O. Jacobs, R. Jaskulka, F. Yang, W. Wu, Sliding wear of epoxy compounds against different
counterparts under dry and aqueous conditions. Wear 256, 9–15 (2004)
83. NN. Dry Rubbing Bearings - A Guide to Design and Material Selection, Inst. Mech. Eng.
Eng. Sci. Data Item 68018 (1968)
84. G.C. Pratt, in Plastic-Based Bearings, Chap. 8 ed. by E.R. Braithwaite. Lubrication and
Lubricants (Elsevier, Amsterdam, 1967)
85. J.K. Lancaster, Lubrication of carbon fiber-reinforced polymers, part I - water and aqueous
solutions. Wear 20, 315–333 (1972)
86. J.-C. Caubet, C. Amsallem, G. Cros, Technologie du frottement dans I‘eau de mer. Sciences
et Techniques 11, 27–30 (1967)
87. S. Hother-Lushington, P. Sellers, Water-Lubricated Bearings - Initial Studies and Future
Prospects in the Power-Generation Industry. In: Proc. Inst. Mech. Engrs., Lubrication and
Wear Group Convention, Bournemouth, Paper 13, (1963), pp. 139–146
88. W.D. Craig, Operation of PTFE bearings in seawater. Lubr. Eng. 20, 456–462 (1964)
89. B.B. Chen, J.Z. Wang, F.Y. Yan, Comparative investigation on the tribological behaviors of
CF/PEEK composites under Seawater lubrication. Tribol. Int. 52, 170–177 (2012)
90. J.K. Lancaster, Lubrication of carbon fiber-reinforced polymers, Part II - organic fluids. Wear
20, 335–351 (1972)
91. G. Zhang, T. Burkhart, B. Wetzel, Tribological behavior of epoxy composites under diesel-
lubricated conditions. Wear 307(1–2), 174–181 (2013)
92. G. Zhang, B. Wetzel, Q. Wang, Tribological behavior of PEEK-based materials under mixed
and boundary lubrication conditions. Tribol. Int. 88, 153–161 (2015)
93. B.A. Krick, W.G. Sawyer, Space tribometers: design for exposed experiments on orbit. Tribol.
Lett 41(1), 303–311 (2011)
94. J. Hanchi, N.S. Eiss, Dry sliding friction and wear of short carbon-fiber-reinforced
polyetheretherketone (PEEK) at elevated temperatures. Wear 203, 380–386 (1997)
966 K. Friedrich and L. Chang
95. G. Theiler, W. Hübner, T. Gradt, P. Klein, K. Friedrich, Friction and wear of PTFE composites
at cryogenic temperatures. Tribol. Int. 35(7), 449–458 (2002)
96. K. Friedrich, G. Theiler, P. Klein, in Polymer Composites for Tribological Applications in
a Range between Liquid Helium and Room Temperature, ed. by S.K. Sinha, B.J. Briscoe.
Polymer Tribology (Imperial College Press, London, 2009), Chapter 11, pp. 375–415
97. G. Theiler, T. Gradt, Friction and wear of PEEK composites in vacuum environment. Wear
269(3–4), 278–284 (2010)
98. A. Gebhard, T. Bayerl, A-K. Schlarb, K. Friedrich, Galvanic corrosion of polyacrylnitrile
(PAN) and pitch based short carbon fibers in polyetheretherketone (PEEK) composites.
Corros. Sci. 51, 2524–2528 (2009)
99. A. Gebhard, T. Bayerl, A.K. Schlarb, K. Friedrich, Increased wear of aqueous lubricated
short carbon fiber reinforced polyetheretherketone (PEEK/SCF) composites due to galvanic
corrosion. Wear 268, 871–876 (2010)
100. K. Friedrich, Z. Zhang, P. Klein, in Wear of Polymer Composites, ed. by G.W. Stachowiak.
Wear- Materials, Mechanisms and Practice, as part of “Handbook of Measuring System
Design” edited by P. Sydenham, R. Thorn (Wiley, 2005), pp. 269–290
101. J. Bijwe, J.J. Rajesh, A. Jeyakumar, A. Ghosh, U.S. Tewari, Influence of solid lubricants and
fiber reinforcement on wear behavior of polyethersulphone. Tribol. Int. 33, 697–706 (2000)
102. A.M. Häger, M. Davies, Short-fiber Reinforced, High-Temperature Resistant Polymers for
a Wide Field of Tribological Applications, in Advances in Composites Tribology, ed. by K.
Friedrich (Elsevier Science Publishers, Amsterdam, 1993), pp. 104–157
103. J. Wang, M. Gua, S. Bai, S. Ge, Investigation of the influence of MoS2 filler on the tribological
properties of carbon fiber reinforced nylon 1010 composites. Wear 255, 774–779 (2003)
104. X.R. Zhang, X.Q. Pei, Q.H. Wang, Effect of solid lubricant on the tribological properties of
polyimide composites reinforced with carbon fibers. J. Reinf. Plast. Compos. 27, 2005–2012
(2009)
105. L. Chang, Z. Zhang, Tribological properties of epoxy nanocomposites: II. A combinative
effect of short carbon fiber and Nano-TiO2 . Wear 206, 869–878 (2006)
106. L. Chang, Z. Zhang, H. Zhang, A.K. Schlarb, On the sliding wear of nanoparticles filled
polyamide. Comput. Sci. Technol. 66, 3188–3198 (2006)
107. M.H. Cho, S. Bahadur, Study of the tribological synergistic effects in CuO-filled and fiber-
reinforced polyphenylenesulfide composites. Wear 258, 835–845 (2005)
108. L. Chang, Z. Zhang, C. Breidt, K. Friedrich, Tribological properties of epoxy nanocom-
posites: I. Enhancement of the wear resistance by Nano-TiO2 particles. Wear 258, 141–148
(2005)
109. Q. Guo, M.Z. Rong, G.L. Jia, K.T. Lau, M.Q. Zhang, Sliding wear performance of Nano-
SiO2 /Short carbon fiber/epoxy hybrid composites. Wear 266, 658–665 (2009)
110. Z.Z. Zhang, F.H. Su, K. Wang, W. Jiang, X.H. Men, W.M. Liu, Study on the friction and wear
properties of carbon fabric composites reinforced with micro- and nanoparticles. Mater. Sci.
Eng. A 404, 251–258 (2005)
111. F.H. Su, Z.Z. Zhang, K. Wang, W. Jiang, X.H. Men, W.M. Liu, Friction and wear properties of
carbon fabric composites filled with Nano-Al2 O3 and Nano Si3 N4 . Compos. A 37, 1351–1357
(2006)
112. F.H. Su, Z.Z. Zhang, K. Wang, W. Jiang, X.H. Men, W.M. Liu, Friction and wear properties
of carbon fabric composites filled with several nano-particulates. Wear 260, 861–868 (2006)
113. Q. Wang, X. Zhang, X.-Q. Pei, T. Wang, Friction and wear properties of solid lubricants
filled/carbon fabric reinforced phenolic composites. J. Appl. Polym. Sci. 117, 2480–2485
(2010)
114. F.H. Su, Z.Z. Zhang, W.M. Liu, Mechanical and tribological properties of carbon fabric
composites filled with several nano-particulates. Wear 260, 861–868 (2006)
115. Q. Wang, X.R. Zhang, X.Q. Pei, Study on the synergistic effect of carbon fiber and graphite
and nanoparticle on the friction and wear behavior of polyimide composites. Mater. Des. 31,
3761–3768 (2010)
30 Carbon Fibers in Tribo-composites 967
116. L. Chang, K. Friedrich, Enhancement effect of nanoparticles on the sliding wear of short
fiber-reinforced polymer composites: a critical discussion of wear mechanisms. Tribol. Int.
43(12), 2355–2364 (2010)
117. L. Chang, K. Friedrich, L. Ye, Study on the transfer film layer in sliding contact between
polymer composites and steel disks using nanoindentation, J. Tribol. 136, 021602-1–021602-
12 (2014). doi:10.1115/1.4026174
118. H. Czichos, Tribology and its many facets: from macroscopic to microscopic to nano-scale
phenomena. Meccanica 36, 605–615 (2001)
119. F. Oster, F. Haupert, K. Friedrich, W. Bickle, M. Müller, Tribologische Hochleistungs-
beschichtungen aus neuartigen Polyetheretherketon (PEEK)- Compounds. Tribologie und
Schmierungstechnik 51(3), 17–24 (2004)
120. K. Friedrich, L. Chang, F. Haupert, in Current and Future Applications of Polymer Compos-
ites in the Field of Tribology, ed. by L. Nicolais, M. Meo, E. Miletta. Composite Materials
(Springer, New York, 2011), pp. 129–167, http://www.springer.com/materials/special+types/
book/978-0-85729-165-3
121. S. Haykin, Neural Networks: A Comprehensive Foundation, 2nd edn. (Prentice Hall, New
Jersey, 1999)
122. L. Fausett, Fundamentals of Neural Networks, Architectures, Algorithms and Applications
(Prentice Hall, New Jersey, 1994)
123. T. Tchaban, J.P. Griffin, M.J. Taylor, A Comparison between single and combined backprop-
agation neural networks in prediction of turnover. Eng. Appl. Artif. Intell. 11, 41–47 (1998)
124. V. Velten, R. Reinicke, K. Friedrich, Wear volume prediction with artificial neural networks.
Tribol. Int. 33, 731–736 (2000)
125. Z. Zhang, K. Friedrich, Artificial neural networks applied to polymer composites: a review.
Comput. Sci. Technol. 63, 2029–2044 (2003)
126. Z. Jiang, Z. Zhang, K. Friedrich, Prediction on wear properties of polymer composites with
artificial neural network. Comput. Sci. Technol. 67, 168–176 (2007)
127. K. Friedrich, R. Reinicke, Z. Zhang, Wear of polymer composites. Proc. Inst. Mech. Eng. Part
J. Eng. Tribol. 216, 415–426 (2002)
128. L.A. Gyurova, Sliding Friction and Wear of Polyphenylene Sulfide Matrix Composites:
Experimental and Artificial Neural Network Approach, PhD Thesis, TU Kaiserslautern,
Germany, Institut für Verbundwerkstoffe GmbH; IVW Schriftenreihe, Band 92, 2010
129. L.A. Gyurova, K. Friedrich, Artificial neural networks for predicting sliding friction and wear
properties of polyphenylene sulfide composites. Tribol. Int. 44, 603–609 (2011)
130. L.A. Gyurova, Z. Jiang, A.K. Schlarb, K. Friedrich, Z. Zhang, in Study on the Wear
and Friction of Short Carbon Fiber and/or nano-TiO2 Reinforced Polyphenylene Sulfide
Composites Using Artificial Neural Networks, ed. by A. Fischer, K. Bobzin. Friction, Wear
and Wear Protection, Proc. International Symposium on Friction, Wear and Wear Protection,
2008 Aachen, VCH, Weinheim, Germany (2009), pp. 417–422
131. L.A. Gyurova, P. Miniño-Justel, A.K. Schlarb, Modeling the sliding wear and friction
properties of polyphenylene sulfide composites using artificial neural networks. Wear 268,
708–714 (2010)
132. Z. Zhang, K. Friedrich, Tribological characteristics of micro- and nanoparticle filled polymer
composites, in Polymer Composites: From Nano- to Macro-Scale, ed. by K. Friedrich, S.
Fakirov, Z. Zhang (Springer, New York, 2005), pp. 169–185
133. J. Zhu, Y. Shi, X. Feng, H. Wang, X. Lu, Prediction on tribological properties of carbon fiber
and TiO2 synergistic reinforced polytertrafluoroethylene composites with artificial neural
networks. Mater. Des. 30, 1042–1049 (2009)
134. K. Friedrich, in Routes for Achieving Multifunctionality in Reinforced Polymers and Com-
posite Structures, ed. by K. Friedrich, U. Breuer. Multifunctionality of Polymer Composites:
Challenges and New Solutions (Elsevier, Oxford, 2015), Chapter 1, pp. 3–41
135. A. Noll, K. Friedrich, T. Burkhart, U. Breuer, Effective multifunctionality of poly(p- pheny-
lene sulfide) nanocomposites filled with different amounts of carbon nanotubes, graphite and
short carbon fibers. Polym. Compos. 34(9), 1405–1412 (2013)
968 K. Friedrich and L. Chang
160. J.P. Giltrow, J.K. Lancaster, Friction and wear properties of carbon fibre-reinforced metals.
Wear 12, 91–105 (1968)
161. Y.P. Tang, H.Z. Liu, H.J. Zhao, L. Liu, Y.T. Wu, Friction and wear properties of copper matrix
composites reinforced with short carbon fibers. Mater. Des. 29, 257–261 (2008)
162. F.B. Bowden, D. Tabor, Friction and Lubrication of Solids (Claredon Press, Oxford, 1964)
163. W.A. Glaeser, Materials for Tribology (Elsevier, Amsterdam, 1992)
164. G.W. Stachowiak, A.W. Batchelor, Engineering Tribology, 3rd edn. (Elsevier Butterworth-
Heinemann, Oxford, 2005)
165. P.K. Rohatgi, P.J. Blau, C.S. Yust (eds.), Tribology of Composite Materials (ASM Interna-
tional, Materials Park, 1990)
166. G.W. Stachowiak, A.W. Batchelor, G.B. Stachowiak, in Experimental Methods in Tribology,
Tribology Series, vol 42 (series editor: D. Dowson), (Elsevier Science B.V., Amsterdam,
2004)
167. S.K. Sinha, B.J. Briscoe, Polymer Tribology (Imperial College Press, London, 2009)
168. L.H. Lee, Advances in Polymer Friction and Wear (Plenum Press, New York, 1974)
169. G.M. Bartenev, V.V. Lavrentev, Friction and Wear of Polymers (Elsevier, Amsterdam, 1981)
170. H. Uetz, J. Wiedemeyer, Tribologie der Polymere (Carl Hanser Verlag, Munich, 1985)
171. Y. Yamaguchi, Tribology of Plastic Materials, Tribology Series, vol 16 (Elsevier, Amsterdam,
1986)
172. K. Friedrich (ed.), Friction and Wear of Polymer Composites (Elsevier, Amsterdam, 1986)
173. K. Friedrich, in Advances in Composite Tribology, Composite Materials Series, vol. 8 (series
editor: R.B. Pipes) (Elsevier Science Publishers, Amsterdam 1993)
174. K. Friedrich, A.K. Schlarb (eds.), Tribology of Polymeric Nano-Composites, 2nd edn.
(Elsevier Butterworth-Heinemann, Amsterdam, 2013)
175. A. Abdelbary, Wear of Polymers and Composites (Woodhead, Cambridge, 2014)