You are on page 1of 28

10.1098/ rspa.2002.

1057

Crack blunting and the strength of


soft elastic solids
By C.-Y. H u i1 , A. J ag o ta2 , S. J. B e n n is o n2 a n d J. D. L o n d o n o2
1 Departmentof Theoretical and Applied Mechanics, Cornell University,
Ithaca, NY 14853, USA (ch45@cornell.edu)
2
CR&D, The DuPont Company, Experimental Station E356/317A,
Wilmington, DE 19880-0356, USA (anand.jagota@usa.dupont.com)

Received 10 May 2002; accepted 30 July 2002; published online 24 April 2003

When a material is so soft that the cohesive strength (or adhesive strength, in the
case of interfacial fracture) exceeds the elastic modulus of the material, we show that
a crack will blunt instead of propagating. Large-deformation ­ nite-element model
(FEM) simulations of crack initiation, in which the debonding processes are quanti-
­ ed using a cohesive zone model, are used to support this hypothesis. An approximate
analytic solution, which agrees well with the FEM simulation, gives additional insight
into the blunting process. The consequence of this result on the strength of soft, rub-
bery materials is the main topic of this paper. We propose two mechanisms by which
crack growth can occur in such blunted regions. We have also performed experiments
on two di¬erent elastomers to demonstrate elastic blunting. In one system, we present
some details on a void growth mechanism for ultimate failure, post-blunting. Finally,
we demonstrate how crack blunting can shed light on some long-standing problems
in the area of adhesion and fracture of elastomers.
Keywords: blunting; fracture; adhesion; elastomer

1. Introduction
The adhesive strength of interfaces and cohesion of solids (called interface fracture
toughness or fracture toughness) is often characterized by the amount of energy Gc
required to advance an interface crack (or a planar crack in a homogeneous solid)
per unit area. For elastomers, Gc varies from 1 to 105 J m¡2 (Gent 1996). It has long
been known that Gc depends on both the amount of inelastic deformation (e.g. vis-
coelasticity) and the separation process (Gent 1996; Gent & Schultz 1972; Andrew &
Kinloch 1973; Gent & Lai 1994), although the quantitative coupling between these
two mechanisms is still not well understood in soft materials. Several recent the-
oretical e¬orts have attempted to link microstructural failure mechanisms to the
continuum ­ elds governing bulk deformations (Tvergaard & Hutchinson 1992; Xu &
Needleman 1994; Rahul Kumar et al . 1999; Knauss 1993; Hui et al . 1992). One of
the most powerful theoretical tools used to investigate the coupling between sepa-
ration and bulk deformation processes is the cohesive zone model, which was ­ rst
introduced by Barenblatt (1968) to study fracture. A cohesive zone model describes
the interfacial normal and shear forces which resist separation and relative sliding of
an interface. The forces, when integrated to complete separation, yield the fracture

Proc. R. Soc. Lond. A (2003) 459, 1489{1516 °


c 2003 The Royal Society
1489
1490 C.-Y. Hui, A. Jagota, S. J. Bennison and J. D. Londono

energy, Gc . For sti¬ materials such as metals and ceramics, the material modulus is
usually far in excess of the peak cohesive stress or strength of the interface. For soft
solids, e.g. elastomers and biological tissues, a physically based value for cohesive
strength is often far greater than the modulus of the material. As a consequence,
cracks in such materials undergo very large deformations before fracture. The ques-
tion arises: is it possible that a material will be so soft that it will blunt elastically?
Given that the cohesive or adhesive strength of materials and interfaces are so large
in comparison with the material modulus, how do soft materials fail? Blunting of
a crack in a homogeneous material or at an interface has a profound in®uence on
the fracture process, usually resulting in much greater energy dissipation. While the
onset of blunting is reasonably well understood in elastic{plastic materials (Tver-
gaard & Hutchinson 1992), conditions for its occurrence in elastic solids have not
been clearly established.y That is, it has not been established under what conditions
a crack tip would blunt, simply because the material is soft. We ­ nd that the answer
to this question, which is the main subject of this paper, is related to several open
and puzzling issues in the failure and adhesion of soft solids.
Lake & Thomas (1967) established the relationship between molecular weight
between entanglements and the fracture energy of soft elastomers, based on the
assumption that all the energy between crosslinks is lost whenever any bond between
the two is severed. However, as shown in the following sections, the accompanying
cohesive zone is found to be vanishingly small, a paradox. A related unresolved issue
emerges when one attempts to explain the rate dependence of peeling force using
bulk viscoelastic data for the polymer. It is found that the dissipation region is far
too small to account for the increase in dissipation (Gent 1996; Rahul Kumar et al .
2000). In peeling of polymers from a substrate (Gent & Petrich 1969), with increas-
ing rate, the mode of failure changes from cohesive in the elastomer to adhesive at
the interface. This is accompanied by a considerable reduction in fracture energy. A
quantitative description of the condition that controls such a change in failure mode
is lacking. The process of separating soft pressure-sensitive adhesives from a sti¬
substrate includes cavity nucleation (often at the interface), lateral growth of cracks,
followed by vertical growth of cracks and stretching of ­ brils (Crosby et al . 2000;
Creton & Lakrout 2000; Lakrout et al . 1999). The last process, which implies blunt-
ing of the laterally growing cracks, contributes a majority of the dissipated energy
(Creton & Lakrout 2000).
Our premise is that a crack in a soft material will typically blunt before failure can
occur. The region ahead of the blunted crack can be considered as a cohesive zone
since its size is small in comparison with typical specimen dimensions. The strength
of this cohesive zone, according to our theory, is limited by the elastic modulus of
the material. Failure must take place within this cohesive zone, e.g. by cavitation or
by the growth of a micro-crack initiating from the original crack. A micro-crack can
grow because the material in the blunted zone is highly stretched and therefore has
much higher modulus than the material outside the blunted zone.
We begin in x 2 with a discussion of some open problems and questions concerning
the failure of soft materials. In x 3 we develop the condition for elastic blunting using

y See, for example, Atkins & Mai (1988) for a qualitative description of elastic blunting: `In highly
extensible materials . . . \cuts" become Inglis \ellipses". . . . relatively harmless notch. . . .’. Atkins & Mai
also pose the unanswered question of why cutting of soft materials is often much easier than tearing by
remote loading.

Proc. R. Soc. Lond. A (2003)


Crack blunting and the strength of soft elastic solids 1491

an approximate analysis of stresses near an ellipse under remote tension. That section
also contains results of ­ nite-element computations that support the approximate
results and provide a picture of deformation near the crack tip. In x 4 we discuss some
mechanisms by which an elastically blunted crack would grow. Section 5 presents
experiments on blunting and crack growth in two soft polymers that support the
theoretical ­ ndings. For one, plasticized (poly)vinylbutyral, we present some details
of the mechanism by which the blunted crack eventually propagates. We conclude in
x 6 with further discussion of the implications of our results.

2. Some unresolved problems in the strength of soft solids


In this section we discuss some unresolved issues in understanding the strength of
soft solids. We begin with the model of Lake & Thomas (1967) for the dependence of
elastomer fracture toughness Gc on entanglement molecular weight dependence. This
model started with a consideration of the paradoxically high value of the intrinsic
fracture toughness of rubbers, Gc ¹= Goc , obtained at low rates or high temperatures
where viscoelastic dissipation is minimal. Based on the typical number of chains
crossing a unit area of the fracture plane § o (§ o º 108 m¡2 ) and the energy U
needed to break a single chemical bond (U º 400 kJ mol¡1 ), Goc should be only
ca. 1 J m¡2 , in contrast with experimental values which range from 10 to 100 J m¡2 .
Lake & Thomas (1967) resolved this discrepancy in fracture energy by noting that
the polymer chains at and in the vicinity of the crack tip are highly stretched. When a
bond breaks, the entire chain relaxes to zero load. It was assumed that all the stored
elastic energy in the chain from the point of breakage to the nearest crosslink or
entanglement is no longer released to the remaining body, but instead is dissipated.
Thus, the energy dissipation is proportional to the number of bonds in a polymer
chain. This explains why the actual fracture energy is much higher than the energy
needed to break one single bond. In addition, since the stored energy in a chain is
proportional to the number of bonds, Gc depends on the molecular weight between
crosslinks.
Implicit in the theory of Lake & Thomas is that there exists a region near the crack
tip where the material behaviour is not described by linear elasticity. This idea has
recently been explored by Ghatak et al . (2000) using a cohesive zone model. In their
model, crack opening is resisted by the stretching of the chains bridging the crack
faces. The resisting force f of these chains is directly proportional to the cracking
opening displacement ¯ , i.e.
f = ks ¯ ; (2.1)
where ks º 10¡4 N m¡1 is the spring constant of the polymer chains bridging the
crack faces. The energy required to propagate the crack per unit area, under steady-
state conditions, is simply the amount of work needed to stretch the chains in a unit
area (§ o º 1018 m¡2 ) from their initial state (zero length at the crack tip) to a ­ nal
length ¯ m ax , i.e.
Goc = 12 § o ks ¯ m2 ax : (2.2)
Since 12 ks ¯ m2 ax is approximately equal to nU , where n is the number of bonds in
pa
chain and U is the energy needed to break the chemical bond, Goc = nU § o / n,
which is the classical Lake{Thomas result. The last relation is due to § o / n¡1=2 .

Proc. R. Soc. Lond. A (2003)


1492 C.-Y. Hui, A. Jagota, S. J. Bennison and J. D. Londono

So far, the prediction of the cohesive zone model seems to be quite satisfactory,
at least from the energy perspective. Let us now estimate the length of the cohesive
zone L. Since, outside the cohesive zone, the material is linearly elastic, the stress
­ eld directly ahead of the crack tip at distances large compared with L is
K
p I ; (2.3)
2º x
where KI is the stress intensity factor. An estimate of the cohesive zone length,
L, can be obtained by equating (2.3) to the characteristic chain fracture stress,
¼ f ² ks ¯ m ax § o . This gives
KI2
L= : (2.4)
2º ¼ f2
For an incompressible material, the energy release rate G is related to the stress
intensity factor by G = 3KI2 =4E. Using this relation and (2.4), the cohesive zone
length L can be expressed in terms of the measured fracture toughness, i.e.
2EGoc E
L= 2
= ; (2.5)
3º ¼ f 3º ks § o

where E is Young’s modulus. For soft materials, using E º 106 N m¡2 , a simple
calculation using (2.5) shows that L ¹= 1 nm. Thus, the characteristic length-scale of
the zone surrounding the crack tip where energy is dissipated is of the order of the
mean square end-to-end distance between crosslinks. This means that the stress and
strain in this region must exceed all realistic values if they have to account for most
of the energy loss.
This apparent paradox is not con­ ned to polymer fracture at low rates. Indeed, this
paradox was ­ rst stated in a pioneering paper by Gent & Lai (1994), who reported
peel data on thin sheets of elastomers which were adhered together by C{C or S{S
interfacial bonds. The goal of their experiments was to elucidate the relationship
between the viscoelastic properties of the elastomer and the fracture toughness G of
the bonded elastomers strengthened by the interfacial bonding. Their test conditions
included a wide range of cracks speeds v (or peel rate), from 4 m m s¡1 to 4 mm s¡1 ,
and test temperatures, from ¡ 40 to 130 ¯ C. They found that the fracture tough-
ness G increases from its low value Goc at very low crack speeds by several orders of
magnitude as the crack speed increases. That viscoelastic losses control the increase
in fracture toughness was demonstrated by the success of time{temperature super-
position procedures to form log(G=G o ) versus log(vaT ) master curves, where aT is
the Williams{Landel{Ferry (WLF) shift factor (Ferry 1980) measured separately.
Indeed, these master curves were very similar in shape to curves of log[· 0 (!)=· 0 (0)]
versus log(va T ), where · 0 (!) is the storage shear modulus at an angular frequency !.
Gent & Lai (1994) estimated the e¬ective size of the dissipative zone by comparing
the dependence of the measured fracture toughness with the dependence of · 0 (!)
upon oscillation frequency. Speci­ cally, their procedure can be summarized as fol-
lows. Let v1 denote the crack speed corresponding to an increment ¢1 of log(G=G o )
in the log(G=G o ) versus log(vaT ) master curves. Let !1 denote the frequency corre-
sponding to the same increment ¢1 of log[· 0 (!)=· 0 (0)] in the log[· 0 (!)=· 0 (0)] versus
log(vaT ) curve. A characteristic distance d1 is de­ ned by d1 = v1 =!1 . In this way
many values of d can be determined using di¬erent increments of log(G=Go ), and it

Proc. R. Soc. Lond. A (2003)


Crack blunting and the strength of soft elastic solids 1493

was argued that these d values represent the size of the dissipative zone near the crack
tip. It should be noted that this procedure of estimating d is entirely consistent with
the scaling arguments of de Gennes (1996), which are based on linear viscoelastic
fracture mechanics. The e¬ective dissipative zone size using this procedure, however,
ranges from 0.1 to 10 A̧. The size of these dissipative zones was recently computed
by Rahul Kumar et al. (2000) using a ­ nite-element model. In these computation
models, local failure of the interface is modelled using a rate-independent cohesive
zone model. The size of the viscous dissipation zone was found to be of the order
of 10¡11 m for peel velocities that showed signi­ cant increase in fracture toughness.
Hence the paradox: how could such a small dissipation zone cause a major increase
in viscoelastic dissipation?
Indeed, theory and simulations of fracture in soft materials based on cohesive zone
model often require the use of parameters that are di¯ cult to justify in terms of physi-
cal separation processes and length-scales. For example, consider a Johnson{Kendall{
Roberts (JKR) (Johnson et al. 1971) test where two identical poly(dimethylsiloxane)
(PDMS) spheres are placed next to each other with no applied external load. It is
found that the two spheres will jump into contact. The boundary of the contact
region is a circle of radius a, which is proportional to the third power of the sur-
face energy, ® . A detailed analysis of associated contact stresses was carried out by
Maugis (1992) using a Dugdale{Barenblatt cohesive zone. In this model, the normal
traction ¼ n acting on the PDMS surfaces just outside the contact zone depends only
on the separation ¯ between the spheres. In addition, ¼ n vanishes if ¯ > ¯ c . However,
if ¯ < ¯ c , then ¼ n = ¼ o , where ¼ o is a positive material constant that is assumed
to be independent of ¯ . This simple model of surface interaction implies that the
intrinsic fracture toughness is 2Gin = ¼ o ¯ c . Thus, ¯ c can be interpreted as a char-
acteristic decay distance of the surface interaction force and ¼ o the strength of this
interaction. The size of the region, d, where these surface forces act is estimated by
Maugis (1992) to be
º E®
d= ; (2.6)
3¼ o2
where the material is assumed to be incompressible. A typical value for surface energy
for an elastomer (e.g. PDMS (Chaudhury et al . 1996)) is ® ¹= 25 mJ m¡2 . According
to Israelachvili (1992), ¼ o for van der Waals interaction is ca. 7 £ 108 N m¡2 . Since
the modulus of PDMS is ca. 4 £ 105 N m¡2 , the size of the cohesive zone estimated
using (2.6) is less than 10¡11 m. However, if we do not allow cohesive stress to
exceed the modulus, ¼ o º E, then d is tens of micrometres. Another example is the
viscoelastic sintering of acrylic beads on a rigid substrate (Lin et al . 2001). As in
the case of PDMS, the surface interaction in this case is primarily of van der Waals
type. However, the surface interaction strength required to match the experimental
data of Mazur & Plazek (1994), ¼ o , was found to be 1:24 £ 106 N m¡2 { about two
orders of magnitude lower than the strength of the van der Waals forces.
The common source of these paradoxes, we propose, is the stress needed to fail an
interface. Near the crack tip, the stress has the form given by (2.3). Crack initiation
occurs when the p energy release rate equals the intrinsic fracture toughness Gin , that
is, when KI º EGin , where we have ignored constants of order 1. For soft materials
with low surface energy or work of adhesion, KI at crack initiation is therefore very
small. On the other hand, the stress needed to fail the material or the interface
between the material and a substrate is not small since the strength of materials

Proc. R. Soc. Lond. A (2003)


1494 C.-Y. Hui, A. Jagota, S. J. Bennison and J. D. Londono

with low surface energy (such as those governed by van der Waals forces) is about
two orders of magnitude greater than the elastic modulus. To break the interface at
this stress, it is necessary to be extremely close to the crack tip (i.e. x º 10¡12 m).
However, attendant blunting due to large deformations is likely to mitigate the ability
of remote loading to increase crack tip stress, as explored in the next section.

3. Crack blunting in elastic materials


While one may accept intuitively that a crack in a soft elastic material will blunt if
the material has su¯ ciently high strength, it is not immediately clear how the con-
dition controlling blunting can be derived. Neither linear elastic fracture mechanics
(Lawn 1993) nor analyses of cracks with local large strains (Knowles & Sternberg
1983; Geubelle & Knauss 1994; Gao 1997) yield stress ­ elds that would predict blunt-
ing, except in a few degenerate cases. Blunting has been observed in computational
analyses of crack propagation in hyperelastic materials where fracture was modelled
using a cohesive zone (Jagota et al. 2000; Rahul Kumar et al . 1999) when the max-
imum cohesive stress and material modulus were of similar magnitude. However, if
one poses the question of blunting using a cohesive zone model within the context of
linear elasticity (Barenblatt 1968), one again ­ nds that cracks will always propagate
at a critical remote loading.
The foregoing observations and background point to the conjecture that crack
blunting in soft elastic materials has essentially to do with large deformations, and
occurs when the cohesive stress exceeds the elastic modulus. This conjecture is
explored in this section ­ rstly by deriving an approximate condition for blunting
in soft elastic solids, and secondly by numerical simulation of crack growth in an
hyperelastic material.

(a) Approximate condition for blunting


In this section we explore the possibility that blunting in a soft elastic solid has
essentially to do with large deformations. In the following, we analyse this conjecture
with a simple approximate model, which yields a condition for blunting.
Consider a pre-crack of length 2a in a large elastic body subjected to a remote
biaxial tension. The interface directly ahead of the crack tip is characterized by its
fracture toughness, Gc , and a maximum cohesive stress, ¼ o . This means that the
normal stress directly ahead of the crack tip cannot exceed ¼ o . Once this condition
is achieved, increase in remote loading will cause the crack to propagate, stably or
unstably, depending on whether energy release rate decreases or increases with crack
advance.
If one disregards the e¬ect of large deformations, increasing the remote loading
will always lead to crack growth, eventually. To investigate the e¬ect of ­ nite defor-
mations, we model the deformed crack pro­ le by approximating it as a long, thin,
ellipse. Further incremental loading of this ellipse results in a tendency for crack tip
stresses to increase due to an increase in remote loading. However, this increase in
crack tip stress is mitigated by an increasingly blunted shape. A comparison of the
two tendencies yields the condition for blunting.
Figure 1 shows that deformed crack modelled as an ellipse with major and minor
semi-axes a and b (a ¾ b). Let the ellipse be subjected to a remotely applied biaxial

Proc. R. Soc. Lond. A (2003)


Crack blunting and the strength of soft elastic solids 1495

b 1 s0
a

Figure 1. Opening crack modelled as an ellipse. Cohesive traction just ahead of


the crack tip is equal to or less than the maximum allowed, ¼ o .

tension, ¼ . The maximum tensile stress ¼ m ax occurs at x = a and is (Timoshenko &


Goodier 1970)
2¼ a
¼ m ax ²¼ :
22;m ax = (3.1)
b
The radius of curvature, R, at x = a is found to be
b2
R= ; (3.2)
a
using (x=a)2 + (y=b)2 = 1. Using (3.2), the maximum tensile stress directly ahead of
the ellipse is r
a
¼ m ax = 2¼ : (3.3)
R
For an incremental increase in remote loading, d¼ , the change in stress at x = a is
r p
a a dR
d¼ m ax = 2 d¼ ¡ ¼ d¼ : (3.4)
R R 3=2 d¼
Equation (3.4) is an approximation since it assumes that the incremental constitutive
response of the material is incompressible and linear elastic. Furthermore, it assumes
that Young’s modulus, E, is independent of deformation.
The solution to the problem of an ellipse subject to a remotely applied biaxial
tension also furnishes a result for the displacement of the ellipse. Let y(x) denote the
current deformed shape of the ellipse. The result in Timoshenko & Goodier (1970)
(see also Appendix A) implies that near x = a
y = yo (1 + ¬ d¼ ); (3.5)
where
2a
¬ =
: (3.6)
E¤ b
In (3.6), E ¤ = E if the deformation is carried out in plane stress (i.e. the crack
is embedded in a large thin sheet of elastic material). If the deformation is carried
out under plane strain conditions (i.e. the out-of-plane dimension of the specimen is
much greater than the crack length), then E ¤ = 43 E. yo is the deformed shape before
application of the incremental remote stress d¼ . Equation (3.6) implies that the rate
of change of the radius of curvature R is
dR
= ¬ R: (3.7)

The blunting condition, d¼ m ax = 0, can be evaluated by combining (3.4) and (3.7).
It is easy to verify that the blunting condition is satis­ ed when ¼ = 2=¬ . Since the

Proc. R. Soc. Lond. A (2003)


1496 C.-Y. Hui, A. Jagota, S. J. Bennison and J. D. Londono

maximum tensile stress is related to the remote stress ¼ by (3.1) and ¼ m ax = ¼ o,


blunting occurs when
¼ o
= 2; (3.8)

that is, blunting is predicted when the maximum cohesive stress exceeds Young’s
modulus by about a factor of two.
The above condition is approximate since it does not take into account the strain
hardening of the elastomer as it deforms. In general, the blunting condition would
depend on the deformation behaviour of the material. We note that (3.8) applies also
for a uniform tension applied in the y-direction as long as a ¾ b is satis­ ed.
Some further insight into this condition for blunting can be extracted by examining
the evolution of the maximum stress in this approximate model. The rate of change of
the maximum stress as a function of the remote stress can be obtained by substituting
(3.7) and (3.6) into (3.4). This results in
r µ ¶
d¼ m ax a ¼ a
=2 1¡ : (3.9)
d¼ R E¤ b

Using equations (3.1){(3.3) in (3.9) we get


µ ¶
d¼ m ax ¼ m ax ¼ m ax
= 1¡ : (3.10)
d¼ ¼ 2E ¤

Equation (3.10) can be readily integrated to give

¼ m ax 2¼ =E ¤
= ; (3.11)
E¤ ¼ =E ¤ + C

where C is a constant of integration. For large ¼ =E ¤ , ¼ m ax =E ¤ ! 2 as predicted


by (3.8). The constant C can be determined by noting that, for small ¼ =E ¤ , (3.11)
implies that ¼ m ax =E ¤ = 2¼ =C E ¤ . This, together with (2.6), implies that

C = (b=a)o ; (3.12)

where the subscript `o’ denotes the initial aspect ratio of the ellipse. This initial
aspect ratio can be estimated by taking a to be the crack length, and b the criti-
cal crack opening displacement from the Dugdale{Barenblatt cohesive zone solution
(Barenblatt 1968; Lawn 1993) (see also ­ gure 3). Ignoring constants of order 1,
Gc Gc
b¹ ) C¹ : (3.13)
¼ o ¼ oa

A typical value for C for elastomers, based on a = 10 mm, Gc = 50 J m¡2 and


¼ o = 105 N m¡2 would be 0.05. Figure 2 shows the maximum stress as a function of
remotely applied stress for several values of C. These ­ gures show that the maximum
stress increases rapidly with increasing applied stress. For small C (e.g. long cracks
and low fracture toughness) and ¼ o =E ¤ < 2, crack growth occurs at applied stress
substantially less than E. If ¼ o =E ¤ > 2, crack blunting due to large geometrical
changes mitigates the intensi­ cation of remote stress by the crack.

Proc. R. Soc. Lond. A (2003)


Crack blunting and the strength of soft elastic solids 1497

2.8
s0 / E * > 2
2.4

maximum stress smax /E *


2.0

1.6
s0 / E * < 2
1.2

1.8
C = 0.01
1.4 C = 0.05
C = 0.1
0 0.2 0.4 0.6 0.8 1.0
remote stress s /E *
Figure 2. Maximum stress as a function of remotely applied stress. If the cohesive stress is
greater than twice Young’ s modulus, the crack blunts before it propagates. Otherwise, crack tip
stress increases with remote stress until local stress equals the cohesive stress. After this point,
local stress is limited by cohesive stress and the crack propagates.

applied displacement:
‘K’ field

2
1 hyperelastic
material

pre-crack line of cohesive elements


Figure 3. Schematic of boundary conditions for simulation of a crack in a soft elastic material,
with cohesive forces ahead of the crack tip holding it shut.

(b) Numerical simulation


More realistic material behaviour and accurate change in geometry have been
modelled by simulating the propagation of a pre-crack in a hyperelastic material
using a commercial ­ nite-element code (ABAQUS® R ). This ­ nite-element code is
augmented by cohesive elements that model the fracture process (Jagota et al . 2000;
Rahul Kumar et al. 1999). Figure 3 shows the computational domain schematically.
A semi-in­ nite plane-strain crack was simulated by applying the small-strain `KI ’
displacement ­ eld at boundary points remote from the crack tip.
The material behaviour has been modelled using rate-independent hyperelastic
constitutive equations. Let F be the deformation gradient and B the left Cauchy{
Green strain, de­ ned in terms of the position vectors x and X of material points in

Proc. R. Soc. Lond. A (2003)


1498 C.-Y. Hui, A. Jagota, S. J. Bennison and J. D. Londono

the deformed and undeformed con­ gurations, respectively:


@x
F = ; B = F ¢ F T: (3.14)
@X
De­ ne the invariants:
I1 = Tr(B) and I2 = (I12 ¡ Tr(B ¢ B)): (3.15)
Then, the strain energy density © is represented as a polynomial series in the invari-
ants (Ogden 1984):
XN
© = Cij (I1 ¡ 3)i (I2 ¡ 3)j : (3.16)
i+ j= 1

A material de­ ned by the single constant C10 is termed neo-Hookean (Ogden 1984).
A reduced polynomial, in which the elastic energy is a function of I1 only, has been
proposed based on data on several elastomers (Yeoh 1993). It proves convenient
here to study blunting using a two-parameter model that includes sti¬ening at large
strains. That is, we consider n = 2, with corresponding constants C10 and C20 .
For the case of uniaxial tension under plane stress conditions, let Lo and L be
the initial and stretched lengths, respectively, and " and "l be the nominal and
logarithmic (or true) strains
9
L
¶ = = 1 + " is the stretch;=
Lo (3.17)
;
"l = ln(1 + "):
The strain energy density is given by
9
© = C10 (I1 ¡ 3) + C20 (I1 ¡ 3)2 ;=
2 (3.18)
I1 = ¶ 2 + ; ;

and ½ and ¼ , the nominal and Cauchy (or true) stresses, are
9
@© @© >
½ = = ; >
>
@¶ @" >
=
6C10 " 1 2 4C 20 2 3 4 5 (3.19)
½ = (1 + " + 3 " ) + (15" + 18" + 8" + " );>
(1 + ")2 (1 + ")3 >
>
>
;
2"1 ¡"1 2"1 ¡"1 2"1 ¡"1
¼ = 2C10 (e ¡ e ) + 4C20 (e ¡ e )(e + 2e ¡ 3):
The initial elastic modulus, to be identi­ ed with Young’s modulus E of previous
sections, is
E = 6C10 : (3.20)
Figure 4 shows the Cauchy stress as a function of the logarithmic strain for di¬erent
ratios of C20 =C10 . All show considerable increase in tangent modulus, Et = @¼ =@"l ,
with increasing strain. De­ ne the stretched region as that where the modulus
increases over its initial value by a factor of 2. The points in ­ gure 4 show the
strain at which this occurs for the di¬erent cases.
The interface between the polymer layer and the sti¬ substrate has been modelled
using a cohesive zone, implemented as cohesive elements that connect the surfaces.

Proc. R. Soc. Lond. A (2003)


Crack blunting and the strength of soft elastic solids 1499

10
C20 /C10 = 0
C20 /C10 = 0.25
C20 /C10 = 0.50
8
C20 /C10 = 0.75
C20 /C10 = 1.00
linear
6 Et /E = 2
stress s /E

0 0.2 0.4 0.6 0.8 1.0 1.2


strain ln (L /L 0 )

Figure 4. Cauchy stress as a function of logarithmic strain in hyperelastic materials. In this series,
the circle represents the strain at which tangent modulus equals twice the initial, small-strain
elastic modulus. We use this to de¯ne a stretched region near the crack tip.

Let T be the tractions across a cohesive zone, with opening and shearing components
Tu and Tv , respectively. Let u and v be the jump in opening and sliding displacements
across the cohesive zone. We assume that cohesive tractions can be derived from a
potential, ª (u; v), as
@ª @ª
Tu = ; Tv = : (3.21)
@u @v
The following phenomenological form has been used for the potential, ª (u; v):
¤ ¤ 2
ª (u; v) = Gin [1 ¡ (1 + u=¯ ¤ )e¡u=¯ ¡(v=¯ )
] + (1 ¡ H(u))Au3 : (3.22)
where ¯ ¤ is a characteristic opening displacement, H(u) is the Heaviside step func-
tion, and A is a penalty parameter. The ­ rst term in (3.22) is a simpli­ ed version
of the cohesive model of Xu & Needleman (1994). According to (3.21) and (3.22),
cohesive tractions vanish for u; v ! 0 and u; v ! 1. For v = 0, the normal tractions
increase with increasing u, reaching a maximum value of
Gin
¼ o = (3.23)
e¯ ¤
and decrease thereafter. The second term in equation (3.22) is introduced to restrict
mesh interpenetration under compression, and does not a¬ect the work of separating
the interface.
A series of simulations was conducted with ­ xed Gin but varying maximum cohe-
sive stress, ¼ o . Remote boundary conditions were increased incrementally until the

Proc. R. Soc. Lond. A (2003)


1500 C.-Y. Hui, A. Jagota, S. J. Bennison and J. D. Londono

s /E = 1/3
C20 /C10 = 0

cohesive traction
-0.4

Ty /E
s /E = 2/3 -0.8
C20 /C10 = 0 -1.2
-1.6

0.50
0.46
0.42
0.38
0.33
0.29 0
cohesive traction

0.25
0.21
-0.4
Ty /E

0.17 -0.8
0.13 -1.2
0.08 s /E = 1 -1.6
0.04
0
C20 /C10 = 0

0
cohesive traction

-0.4
Ty /E

-0.8
-1.2
s /E = 4/3 -1.6
C20 /C10 = 0

0
cohesive traction

-0.4
Ty /E

-0.8
-1.2
-1.6
Figure 5. Contours of maximum principal logarithmic strain near the crack tip as it begins to
propagate for a neo-Hookean material. Fracture toughness is the same in all cases. Remotely
applied boundary conditions are increased to just over the fracture toughness to initiate crack
propagation. The four cases di® er in the level of cohesive stress. About when cohesive stress
exceeds the modulus, the crack blunts. Note that there is little appearance of a stretched region
before elastic blunting for this material. Also shown below each contour plot is the distribution
of cohesive tractions.

Proc. R. Soc. Lond. A (2003)


Crack blunting and the strength of soft elastic solids 1501

s /E = 2/3
C20 /C10 = 1.0

cohesive traction
-0.4
-0.8

Ty /E
s /E = 1
C20 /C10 = 1.0 -1.2
-1.6

0.4575
0.1800
0.1650
0.1500
0.1350
0.1200
0.1050 0
cohesive traction

0.0900
0.0750 -0.4
0.0600 -0.8
Ty /E

0.0450 s /E = 4/3
0.0300 C20 /C10 = 1.0 -1.2
0.0150 -1.6
0

0
cohesive traction

-0.4
-0.8
Ty /E

s /E = 5/3
C20 /C10 = 1.0 -1.2
-1.6
cohesive traction

-0.4
-0.8
Ty /E

-1.2
-1.6
Figure 6. Contours of maximum principal logarithmic strain near the crack tip as it begins to
propagate for a `hardening’ hyperelastic material. (C 20 =C10 = 1). Fracture toughness is the
same in all cases. Remotely applied boundary conditions are increased to just over the fracture
toughness to initiate crack propagation. The four cases di® er in the level of cohesive stress. Note
that before blunting a considerable stretched region (white) appears near the crack tip. Also
shown below each contour plot is the distribution of cohesive tractions.

Proc. R. Soc. Lond. A (2003)


1502 C.-Y. Hui, A. Jagota, S. J. Bennison and J. D. Londono

Figure 7. Drawing of stretched region as calculated using an isotropic hyperelastic model (in
white), compared with the expected Dugdale-like stretched region (uniform grey region) due to
developing anisotropy near the crack tip.

applied energy release rate `G’ just exceeded the intrinsic value. Figure 5 shows the
results for a neo-Hookean material. It contains contour plots of principal logarithmic
strain at the end of the simulation. The range of contours is limited by the value of
logarithmic strain at which tangent modulus (in uniaxial plane stress) exceeds the
small-strain value by a factor of 2. For each case, we also show the distribution of
cohesive tractions ahead of the crack tip. When ¼ o is signi­ cantly smaller than E, a
cohesive zone develops from the pre-crack. When the applied stress intensity factor
KI reaches its critical value, the crack begins to propagate. The cohesive zone size
scales inversely with (¼ o )2 , as predicted by fracture mechanics. Increasing the maxi-
mum cohesive stress ¼ o decreases the cohesive zone size and large deformations near
the crack tip are evident. At a critical value of ¼ o =E ¹ 1, the crack no longer propa-
gates in the simulations but blunts instead. Note that there is little appearance of a
stretched region before elastic blunting for this material. Figure 6 shows the results
for a sti¬ening `Yeoh’ material with C20 =C10 = 1. Blunting is delayed to a larger
ratio of ¼ o =E. Also, there is a considerable stretched region prior to blunting, shown
by the white elliptical patch near the crack tip at higher cohesive stresses. It should
be noted that the stretch regions in ­ gure 6 are oriented ca. 90¯ counterclockwise
from the cohesive zone. Physically, we expect that these stretch regions should be
oriented more or less in the same direction as the cohesive zone (­ gure 7). A possible
explanation for this discrepancy is that the constitutive model for the elastomer has
been assumed to be isotropic. This assumption is likely to be incorrect near the crack
tip since the material there is subjected to a multi-axial state of stress and very large
strains.
Based on linear fracture mechanics, the cohesive zone scales as K I2 =¼ o2 ; the stretch-
ed zone scales as (KI =E)2 . The stretched region becomes much larger than the cohe-
sive zone if E ½ ¼ o . For a neo-Hookean material, a prototype for a soft solid that
does not sti¬en much, this does not occur before onset of blunting. For a sti¬ening
elastic solid, a stretched region develops along with blunting as shown in ­ gure 6.

4. How do blunted cracks grow?


Of course, the elastic blunting condition only captures the fact that the increase
in remote loading now causes very large deformations, destroying the incremental
increase in stress at the crack tip. Essentially, despite local sti¬ening, there is some

Proc. R. Soc. Lond. A (2003)


Crack blunting and the strength of soft elastic solids 1503

region away from the crack tip that will not transmit stress much greater than the
elastic modulus to the crack tip. We propose that when the elastic blunting condition
is met, one of the following will occur.
(1) The blunted region can be viewed as a cohesive zone with a modi­ ed cohesive
strength ¼ c º E. It is likely that this modi­ ed cohesive strength varies from
material to material. However, it is considerably lower than the actual cohesive
strength or intrinsic strength of the material (e.g. ¼ o ).
(2) Finite specimen size e¬ects may come into play. For example, if one is cutting
with a sharp knife, the elastic blunting e¬ect may be limited by the fact that
the crack opening displacement is determined by the knife geometry. For exam-
ple, if the knife has radius of curvature Rkn ife that is smaller than the cohesive
thickness or critical opening displacement, it will prevent blunting. Further-
more, a knife with a very small radius of curvature can give rise to very large
stresses near its edge. This may o¬er an explanation for why cutting is ener-
getically easier than peeling, even after accounting for intervening viscoelastic
losses.
(3) If no failure mechanism is available that operates at a stress of about the
modulus, the material will eventually sti¬en enough to exceed the intrinsic
strength and the crack will propagate.
An important implicit assumption of our argument is that the behaviour of material
in the blunted region or the modi­ ed cohesive zone is sharply di¬erent from the
material outside the blunted zone. This means that all the stored elastic energy
in the blunted region is lost during crack propagation. This will be the case, for
example, if the material in the blunted zone is highly anisotropic so that it cannot
support lateral stresses. This loss of lateral constraint can be due to cavitation or
­ brillation. Another possibility is that the material in the blunted region undergoes a
phase transformation. Clearly, the validity of our hypothesis depends on the nonlinear
elastic behaviour of the material at large strains. Thus, although this assumption is
entirely consistent with the theory of Lake & Thomas, it clearly cannot be obeyed
by all elastomers; counterexamples will be given in the discussion.
The most general possibility, (1), explains why, when cohesive zone models are
used to simulate experiments, agreement is possible if the cohesive strength is much
less than that suggested by physically based mechanisms. Because the blunted zone
can be treated as a cohesive zone, the fracture toughness Gc is simply
Gc º ¼ c¯ c º E¯ c ; (4.1)
where ¯ c is the critical opening of the blunted crack at the onset of crack growth {
so far an unknown quantity that must determined by experiment. For example, if
the measured toughness of the material is 50 J m¡2 , then ¯ c º 10 m m for a material
with E = 5 £ 106 N m¡2 .
Since many soft materials do fail by crack growth, the blunted crack in these
materials must eventually grow. The questions are as follows.
(i) How does a blunted crack provide enough stresses to break bonds or to over-
come the high strength of van der Waals forces?
(ii) What is the connection between ¯ c and the local failure mechanics?

Proc. R. Soc. Lond. A (2003)


1504 C.-Y. Hui, A. Jagota, S. J. Bennison and J. D. Londono

y
uniform vertical
displacement

d /2
x

Figure 8. Geometry of the micromechanics problem: the blunted region is modelled by an in¯nite
strip of incompressible elastic material. The e® ective Young’ s modulus of the strip material can
be several times that of E, the Young’ s modulus of the elastomer. The strip thickness is ¯ c , the
opening of the blunted crack. The strip is loaded by applying a uniform vertical displacement on
its boundary so that the normal stress is equal to ¼ c ¼ E at distances su± ciently far away from
the crack tip. For the case of fracture by cavitation, the strip material is highly anisotropic.

There are at least two ways in which a blunted crack can grow. The ­ rst is by
the nucleation and coalescence of cavities inside the blunted region. The criterion
of cavity nucleation has been explored by Gent & Wang (1991), who showed that
cavities can form in soft materials when the hydrostatic tension is of the order of
the elastic modulus. This is precisely the stress state inside the blunted zone (at
least for material points with distances greater than the radius of curvature of the
blunted crack tip). The mechanics of cavity growth and failure will not be analysed
in this work. However, we expect this process to be similar to the separation of
a soft pressure-sensitive adhesive from a rigid substrate (Creton & Lakrout 2000).
Typically, this process starts with cavity nucleation at the interface between the
adhesive and the substrate, followed by vertical growth of these cavities. The material
between these elongated cavities appears as ­ brils that eventually break down leading
to the failure. In the following section we show some experiments where cavity growth
is the main damage mechanism.
The second way a blunted crack can grow is by the initiation and growth of a
micro-crack near the blunted crack front. To see this, consider the schematic blunted
region as shown in ­ gure 7. For materials with su¯ ciently large strain hardening,
the e¬ective modulus of the material close to the blunted crack tip is much sti¬er
than the material outside the blunted region as illustrated by our previous numerical
results. This being the case, a micro-crack (or micro-cracks) can initiate from the
blunted crack tip. Because of strain hardening, this micro-crack can propagate with
little or no blunting. The spirit of the analysis below is similar to the theory of
craze failure at a crack tip in glassy polymers (Brown 1991). There is, however, an
important di¬erence between a craze and the blunted region. In the case of craze
breakdown, most the energy dissipation occurs near the craze{bulk interface, where
­ bril drawing occurs. On the other hand, it is unlikely that such an interfacial region
exists for elastomers, so that most of the energy loss is due to the elastic unloading
of the material in blunted region, as proposed by Lake & Thomas (1967).
Let us consider a scenario where we can work out the details of how such blunted
cracks fail. Recall that ¯ c in (4.1) is an unknown quantity that, as will be shown
below, is determined by the local failure process (e.g. growth and coalescence of
cavities, breaking of C{C bonds in a homogeneous elastomer or the adhesive failure
of an interface reinforced by van der Waals type surface forces). Since the length-scale
of such local processes (say lf ) is much smaller than the length of the cohesive zone
(or the blunted zone), we can approximate the cohesive zone as a semi-in­ nite strip of
thickness ¯ c occupying the positive x-axis (see ­ gure 8). In this ­ gure, the unloaded

Proc. R. Soc. Lond. A (2003)


Crack blunting and the strength of soft elastic solids 1505

pre-crack lies along x < 0 (see ­ gure 8). The micro-crack directly ahead of the pre-
crack is assumed to be much shorter than the pre-crack. Since the chains inside the
cohesive zone are highly stretched (especially those that are very close to crack tip),
incremental deformations are small and small strain theory is applicable. To simplify
analysis, we model this region as a linearly isotropic incompressible elastic material
having an e¬ective Young’s elastic modulus, Ee¬ , which is likely to be several times
E. The usage of isotropy is not necessary but it simpli­ es the calculations. The
normal stress directly ahead of the micro-crack can be estimated using the standard
fracture mechanics result
Klocal
¼ local º p ; (4.2)
2º x
where Klocal is the local stress intensity factor. Assuming plane stress deformation,
Klocal can be estimated using
p
Klocal º Ee¬ (SE1 ); (4.3)
where (SE)1 is the elastic strain energy per unit area stored at distances x ¾ ¯ c in
the strip. (SE)1 is approximately given by
2
(SE)1 ¹= ¼ c¯ c
: (4.4)
Ee¬
Therefore, p
Klocal º ¼ c ¯ c: (4.5)
Equation (4.5) predicts that the local stress in the stretched region near the blunted
crack tip is directly proportional to the square root of the blunted crack opening. It
shows how crack blunting leads to crack propagation. Indeed, crack growth occurs
when the local stress reaches the cohesive strength of the material, that is
K
p local ¹= ¼ o : (4.6)
2º lf
Substituting (4.5) into (4.6) gives the relation
p p
¼ c ¯ c = ¼ o 2º lf : (4.7)
That is, the critical opening of the blunted crack obeys the following scaling law,
µ ¶2 µ ¶2
¯ c ¹ ¼ o ¼ o
= º ; (4.8)
lf ¼ c E
where we have ignored the numerical p factor of 2º . Another way of interpreting this
expression is to multiply (4.7) by ¯ c on both sides, resulting in
p p p
¼ c ¯ c = ¼ o 2º lf ¯ c = ¼ o lf 2º ¯ c =lf : (4.9)
The left-hand side of (4.9) can be interpreted as the applied critical energy release
rate
p (or fracture toughness Gc ). If we take lf to be the characteristic length such that
2º ¼ o lf is the intrinsic fracture toughness Gin , then fracture toughness scales with
the intrinsic fracture toughness as
r
Gc ¯ c ¼ o
= º : (4.10)
Gin lf E
In addition, the critical opening scales as ¯ c º (Gin =E)(¼ o =E).

Proc. R. Soc. Lond. A (2003)


1506 C.-Y. Hui, A. Jagota, S. J. Bennison and J. D. Londono

The above argument can be modi­ ed to account for the case of fracture by cavity
growth. Following Brown (1991), we assume that the presence of large cavities inside
the cohesive zone results in a material that is highly anisotropic. Speci­ cally, the strip
material is strong in tension and weak in shear and cannot support normal stress in
the direction of parallel to the cohesive zone. Therefore, the cavitated material can
be modelled as a homogeneous linearly orthotropic elastic material. As before, we
shall assume the cohesive zone length is long in comparison with its thickness so that
it can be approximated as a long thin strip. For an orthotropic material under plane
stress deformation, the non-vanishing components of strain and stress are related by
¼ 11 = C11 "11 + C12 "22 ; (4.11)
¼ 22 = C12 "11 + C22 "22 ; (4.12)
¼ 12 = 2C66 "12 ; (4.13)
where the Cij are material constants. Our assumption is that the strip material
is highly anisotropic so that C66 =C22 ½ 1, C11 =C66 ½ 1 and C12 º C66 . These
conditions imply that the strip material is strong in tension and weak in shear (C12 º
C66 , C66 =C22 ½ 1) and cannot support normal stress in the direction of parallel
to the strip (C11 =C66 ½ 1). For an anisotropic material with this behaviour, the
loading on the strip boundary y = §¯ c =2 can be approximated by a uniform normal
displacement v = §¼ c ¯ c =2C22 . The normal stress directly ahead of the crack can be
estimated using the fracture mechanics result (4.2). The local stress intensity factor,
Klocal , can be estimated using a result from Sha et al. (1995) as
p
Klocal ¹ ¼ c ¬ ¯ c ; (4.14)
p
where ¬ ² C66 =C22 ½ 1. Equation (4.14) predicts that the local stress near the
blunted crack tip is directly proportional to the square root of the blunted crack open-
ing. It shows how crack blunting progresses into crack propagation when damage is
by void growth ahead of the crack tip. Indeed, failure of the material occurs when
K
p local ¹= » ¼ o; (4.15)
2º lf
where (1 ¡ » ) is the volume fraction of voids. By (4.15), crack growth occurs when
p
¼ c ¬ ¯ c ¹
p = » ¼ o; (4.16)
2º lf
p
Since the measured toughness Gc ¹= ¼ o ¯ c , we have, using 2º ¼ o lf = Gin ,
r µ ¶
¹ ¯ c »
Gc = ¼ c ¯ c = Gin p : (4.17)
lf ¬
Comparing (4.17) with (4.10), wep note that the fracture toughness due to the void
growth is much higher, since » = 2¬ ¾ 1. Equation (4.17) implies the following scal-
ing relation between ¯ c and the intrinsic toughness, i.e.
µ 2 ¶µ ¶µ ¶ µ 2 ¶µ ¶µ ¶
» Gin ¼ o » Gin ¼ o
¯ cº º ; (4.18)
¬ ¼ c ¼ c ¬ E E
where we have neglected factors of order 1.

Proc. R. Soc. Lond. A (2003)


Crack blunting and the strength of soft elastic solids 1507

20 s 60 s

Figure 9. Tensile loading of a crack in two transparent elastomers showing the di® erence between
crack propagation and blunting. The experiments have been viewed using transmitted light
through cross-polarizers that help visualize stress development. The material to the left is a
photoelastic elastomer. With increasing loading, the stress near the crack tip increases mono-
tonically. At a critical point, the crack starts to propagate. The material to the right is plasticized
polyvinylbutyral (PVB), an elastomeric material used widely in glass{polymer laminates. It has
similar modulus to the ¯rst material, but is tougher. Note that in the case of PVB the pre-crack
blunts dramatically. Failure occurs by void formation and coalescence.

5. Experiments on crack blunting


In this section we present experiments on crack growth in two soft elastomeric
materials. They give a qualitative demonstration of the di¬erence between signif-
icant blunting and `normal’ crack growth. For the one that blunts signi­ cantly, we
present further details on the void growth mechanism by which the crack even-
tually grows. Both samples are transparent elastomers with an internal pre-crack
created by cutting using a knife. The experiments have been viewed using transmit-
ted light through cross-polarizers to help visualize stress development. A series of

Proc. R. Soc. Lond. A (2003)


1508 C.-Y. Hui, A. Jagota, S. J. Bennison and J. D. Londono

photographs showing stress development are shown in ­ gure 9. In all cases, the
white ellipses denoting the cracks have been added to the photographs to indi-
cate the crack pro­ le. The sample to the left is a photoelastic elastomery. With
increasing load, the stress near the crack tip increases monotonically. At a criti-
cal load, the crack starts to propagate. Presumably, in this case the cohesive stress
is small compared with the modulus for such crack propagation, or the material
strain hardens considerably so that a micro-crack can grow, as discussed above.
The material to the right is plasticized polyvinylbutyralz (PVB), an elastomeric
material used widely in glass{polymer laminates. It has a similar modulus to the
­ rst material, but is tougher, meaning higher cohesive stress or less strain hard-
ening. The time interval between pictures on the left is 20 s, and between those
on the right is 60 s. Note that in the case of PVB, the pre-crack blunts dramati-
cally. Of course, PVB does eventually fail, but it does so only when a new failure
mechanism, operating at a stress of the order of the modulus, takes over. Figure 10
shows an optical micrograph of a propagating crack in PVB. Note the large defor-
mation of the crack ®anks. The whitening ahead of the crack tip, relatively uniform
across the intact section, is due to scattering from strain-generated voids. It is inter-
esting to note the clarity of the polymer behind the crack tip, in a region that
was voided and whitened when ahead of the crack tip. As mentioned below, we
found that void growth in this material is strain dependent, and vanishes upon its
removal.
Direct evidence of a void growth mechanism in PVB has been obtained by mea-
suring small-angle scattering of X-rays during tensile stretching. Figure 11 shows
the measured tensile stress{strain behaviour in plane stress. Details of these mea-
surements are given in Appendix B. The material is nearly incompressible. At these
strain rates it shows rate-dependent rubber-like elastic behaviour. Failure strains
appear to be approximately constant at a value of about 1.2. Using small-angle X-
ray scattering (see Appendix B for details), we found that voids nucleate well before
failure. Figure 12 shows scattered X-ray intensities at two di¬erent angle ranges.
The vertical orientation is the tensile direction. The meridional scattering to the
right represents ellipsoidal shaped voids, up to hundreds of nanometres in size. The
equatorial scattering to the left represents ­ bril features less than 10 nm in diame-
ter, presumably bridging the larger microvoids. More detailed analyses of these data
will be presented elsewhere. Here, approximately, by assuming that the voids are
spherical in shape and polydisperse in size, we determine from the scattering data
how void size changes with strain (Appendix B). These results are summarized in
­ gure 13.
In the context of the crack blunting results developed in this paper, the study of
damage development in PVB shows how void formation emerges as a damage and
failure mechanism when elastic crack blunting does not permit simple fracture. This
leads to the emergence of a cohesive zone that has larger openings and is softer in
the sense that it has much lower peak stress. However, the increase in separation
distances means that it is usually accompanied by an increase in fracture toughness
of the material, as suggested by (4.17).

y Photoelastic sheet PS-4C, 1 mm nominal thickness; Measurements Group, Raleigh, NC 27611, USA
(www.measurementsgroup.com).
z Butacite ® R , DuPont Company.

Proc. R. Soc. Lond. A (2003)


Crack blunting and the strength of soft elastic solids 1509

1 mm

Figure 10. Void growth ahead of blunted crack tip in Butacite® R . Note the formation of a cusp
at the crack tip due to hardening of the material at large strains. Void damage appears to be
strain controlled and independent of rate as measured by SAXS experiments. Voids collapse
upon unloading, as can be seen in the unloaded crack ° anks.

120
0.01 mm s-1

100 0.10 mm s-1


1.00 mm s-1
true tensile stress (MPa)

10.0 mm s-1
80

60

40

20

0 0.2 0.4 0.6 0.8 1.0 1.2


true tensile strain
Figure 11. True stress{strain data for plasticized polyvinylbutyral from plane stress tensile tests.
The material is nearly incompressible. At these strain rates it shows rate-dependent rubber-like
elastic behaviour. Failure strains appear to be approximately constant at a value of about 1.2.
Extensional and lateral strains were measured by following ¯ducial marks on the specimen, and
were used to determine strain in the gauge section and the true Cauchy stress.

Proc. R. Soc. Lond. A (2003)


1510 C.-Y. Hui, A. Jagota, S. J. Bennison and J. D. Londono

SDD = 727 mm SDD = 8500 mm

strain
log(I )

–0.4 0 0.4 –0.03 0 0.03


q = 4p sin q /l
Figure 12. Scattering intensities of stretched PVB indicating structure at two length-scales. The
equatorial scattering to the left represents scattering from ¯brils a few nanometres in dimension
aligned with the tensile strain. The meridional scattering shown on the right represents oblong
defects 100s of nanometres in size with the long side oriented normal to the tensile axis.

6. Discussion and conclusions


We have argued that in soft materials crack blunting due to large deformations
often prevents crack tip stresses from reaching a magnitude su¯ cient to decohere
the material or interface. Deprived of this direct mode of propagation, failure can
now proceed only by another mechanism, for example, void growth or extension in a
new stretched region ahead of the crack tip. The blunting condition, approximately
that the cohesive strength exceeds the elastic modulus, results in this changed failure
mode, which is often accompanied by an increased fracture energy. An example is the
transition from cohesive to adhesive failure, which was observed by Gent & Petrich
(1969). They found that the toughness of an adhesive bonded to a rigid substrate
­ rst increases rapidly with crack growth rate a, _ reaches a maximum at a_ m ax , then
decreases rapidly to a very small fraction of its maximum value. Further increase in
crack growth rate results in much smaller peak in toughness. Gent & Petrich noticed
that cohesive failure of the adhesive occurs when a_ 6 a_ m ax , whereas interfacial or
adhesive failure occurs when a_ > a_ m ax . While it is expected that viscoelastic dissipa-
tion is responsible for the large increase in toughness, it is not clear what causes the
transition from cohesive to adhesive failure. We propose that this transition is due
to crack blunting. At low rates of crack growth, the material near the crack tip is
fully relaxed, so that the elastic modulus is approximately Eo , the rubbery modulus
or the modulus at zero rate. Since this modulus is ca. 106 Pa, crack blunting is likely.
With increasing rate of crack growth the material near the crack tip is loaded at a
much higher rate, so that the deformation there is governed by the instantaneous
modulus E1 . Since E1 can be several orders of magnitude greater than Eo , it will
eventually be su¯ cient to allow the crack to decohere the interface without blunt-

Proc. R. Soc. Lond. A (2003)


Crack blunting and the strength of soft elastic solids 1511

100 140

effective void diameter (nm)


75 120

true stress (MPa) 50 100

25 80

60
0 0.4 0.8 1.2
true logarithmic strain
Figure 13. Evolution of e® ective polydisperse spherical void size with strain.

ing. Since adhesive failure is governed by van der Waals type forces, the adhesive
toughness is much less than that of cohesive failure.
It is interesting to note that (4.10) or (4.17) predict that the fracture toughness
should scale with the work of adhesion. Furthermore, the proportional factor ¼ o =E, a
large number, depends only on the elastic constitutive behaviour of the material. This
equation is the rate-independent version of the well-known empirical relation Gc =
Wad (1 + ¿ (aT a)),
_ where aT is the Williams{Landel{Ferry (WLF) shift factor, Wad is
the work of adhesion and ¿ is a dimensionless function representing the contribution
of crack tip inelastic deformation to the energy dissipation (Gent 1996). Furthermore,
(4.10) implies that 1 + ¿ (aT a_ ! 0) ! ¼ o =E for materials capable of large amounts of
strain hardening. Since the analysis in this work assumes that the material is elastic,
dissipation due to bulk viscoelasticity is not explicitly accounted for. Extension of
the results in this work to include viscoelastic material behaviour will be done in a
future work. However, we believe the empirical relation Gc = W ad (1 + ¿ (aT a)) _ can
be derived from a theoretical model that couples crack blunting and viscoelasticity.
Blunting of cracks in a soft elastic material or at the interface between one and
a rigid substrate leads to modes of separation that involve signi­ cant increases in
energy dissipation. Without crack blunting, it is di¯ cult to rationalize how soft
materials can fail since a linear analysis would indicate that the region of energy
dissipation has dimensions less than 1 A̧. In an elastic material, blunting appears
to be contingent upon large changes in geometry, resulting from the modulus being
smaller than maximum interfacial cohesive stress. The simple approximate model,
and numerical simulations using a neo-Hookean constitutive description are in rea-
sonably good agreement. For elastic materials, most of the energy release to the crack
tip is dissipated by the blunting process, only a very small fraction of this energy goes
into the actual failure process. Therefore, our analysis is consistent with the theory
of Lake & Thomas (1967), that is, most of the elastic stored energy in the blunted
region is lost during unloading. Of course, when the elastomer fails by cavity growth,
some of the stored energy is used to create new surfaces. For materials capable of
large amounts of strain hardening, the measured fracture toughness scales with the

Proc. R. Soc. Lond. A (2003)


1512 C.-Y. Hui, A. Jagota, S. J. Bennison and J. D. Londono

work of adhesion, that is, Gc º (¼ o =Ee¬ )Wad , where E is the small strain elastic
modulus and ¼ o is the intrinsic cohesive strength of the material.
Thus, crack blunting explains qualitatively the paradox presented by Gent & Lai
that there is signi­ cant viscoelastic dissipation even when the dissipative region is
only of the order of angstroms in size. That is, crack blunting introduces an additional
length-scale that is not in Gent & Lai’s analysis. Indeed, one anticipates that the
size of the blunted zone is at least of the order of micrometres for soft materials.
Therefore, it is likely that the amount of viscoelastic dissipation is controlled by
the size of the blunted zone. It is possible that the length-scale introduced by Gent
& Lai is a measure of the size of the actual chain breaking zone at the crack tip.
Exactly how a blunted zone interacts with a local failure mechanism when viscoelastic
deformation is important is the blunted region is still unclear and is the subject of
further research.
Finally, we pointed out that some of the scaling relations we have proposed in
this work depend on the implicit assumption that all the stored elastic energy in the
blunted zone is lost during crack growth. While this assumption is consistent with the
theory of Lake & Thomas, it clearly is not valid for all elastomers. In other words, a
crack can blunt with smooth changes in material behaviour everywhere. If this were
the case, then the stored elastic energy of the material in the blunted region will not
be dissipated and the fracture toughness of the material will be close to the intrinsic
work of adhesion. This is indeed the case for the fracture of low-modulus gels, which
are obviously in the crack blunting regime, yet the adhesion is essentially reversible.
The overall energy required to peel these materials from a surface is determined by
the thermodynamic work of adhesion (Mowery et al. 1997).
We acknowledge the DND-CAT facility, which is supported through E. I. DuPont de Nemours &
Company, Northwestern University, the Dow Chemical Company, the State of Illinois through
the Department of Commerce and the Board of Higher Education (HECA), the US Department
of Energy O± ce of Energy Research, and the US National Science Foundation Division of
Materials Research. C.Y.H. acknowledges support from E. I. DuPont de Nemours & Company
through a gift to Cornell University. The authors are grateful to C. Creton, E. J. Kramer and
K. Shull for valuable suggestions and informative discussions.

Appendix A.
Let x and y denote Cartesian coordinates and z be de­ ned as the complex number
z = X + iY: (A 1)
De­ ne elliptic coordinates ¹ and ² by
z = c cosh ± ; ± = ¹ + i² : (A 2)
Lines of constant ¹ are ellipses, and for ¹ = ¹ o ,
a = c cosh ¹ o ; b = c cosh ¹ o ; a2 ¡ b2 = c2 : (A 3)
The solution we seek, the displacement ­ eld, can be calculated in terms of complex
potentials Á(z) and À (z). If u(x; y) and v(x; y) denote the two Cartesian components
of displacement, then, for incompressible plane strain,
2G(u + iv) = ª (z) ¡ z ª · 0 (·
z) ¡ À ·0 (·
z ); (A 4)

Proc. R. Soc. Lond. A (2003)


Crack blunting and the strength of soft elastic solids 1513

where z· denotes the complex conjugate of z, whether of a variable or a function. For


an elliptical hole the complex potentials are
)
ª (z) = 14 ¼ c[¡ e2¹ o cosh ± + (1 ¡ e2¹ o cos 2(º =2 ¡ ² ))];
(A 5)
À (z) = ¡ 14 ¼ c2 [(cosh 2¹ o + 1)± + 12 e2¹ o cosh 2(± ¡ ¹ o ¡ iº =2)]:
This results in
3¡ ¸ ¼ a ¼ (x2 + y2 )ay
2Gv = y¡ : (A 6)
1+¸ 2 b 2b((bx=a)2 + (ay=b)2 )
Near the tip of the ellipse, y ¹ 0, this simpli­ es to
3¡ ¸ ¼ a a
2Gv ¹ yo ¡ yo ; (A 7)
1+¸ 2 b 2b
where yo is the initial shape and
y = yo + v = yo (1 + ¬ d¼ ); (A 8)
which is the result (3.5). The radius of curvature,
1 y (1 + ¬ d¼ )yo
= = ; (A 9)
R 02
(1 + y ) 3=2 (1 + (1 + ¬ d¼ )2 y22 )3=2
which implies (3.7).

Appendix B.
Extensional and lateral strains were measured by following ­ ducial marks on the
specimen, and were used to determine strain in the gauge section and the true Cauchy
stress. Tensile test specimens, 13 £ 75 mm2 , were prepared from nominally 0.75 mm
(30 mil) PVB using a die-punching method. Gripping tabs of PVB itself were cut from
the same ­ lm and attached to the specimen by auto-adhesion. Five reference marks
were placed by hand on the specimen using a ­ ne-point marker pen. The specimens
were conditioned (overnight) before testing by placing in a humidity chamber at
22 § 0:5% relative humidity at 22 ¯ C. Tensile testing was performed using a universal
closed-loop control servo-hydraulic testing frame (MTS Corporation) in constant
displacement control. The specimens were gripped using hydraulic wedge grips with a
SurfAlloy® R ­ nish at a gripping pressure of 5 MPa. A range of displacement rates from
10¡2 to 10 mm s¡1 were studied yielding a range of strain rates from 2:3 £ 10¡4 to
2:3 £ 10¡1 s¡1 for the 25 mm specimen gauge section employed. The relative motions
of the markers during loading were monitored by a video camera (30 frames per
second) at a magni­ cation of approximately 5£. A video overlay device was used to
simultaneously display elapsed time and a voltage corresponding to the applied load.
A time-lapse movie of the deformation was made by recording images at speci­ ed
time intervals using a digital frame grabber (SCION Corp.) housed in a Macintosh
personal computer (Mac IIfx). The movie was analysed, frame by frame, using an
image-analysis package (NIH Image). The axial and longitudinal strains were then
determined from the relative displacements (motions) of the markers.
Small-angle X-ray measurements were performed at the DND-CAT Synchrotron
Research Center, Advanced Photon Source, Argonne National Laboratory. The

Proc. R. Soc. Lond. A (2003)


1514 C.-Y. Hui, A. Jagota, S. J. Bennison and J. D. Londono

energy of the X-ray beam from an insertion device (ID) was tunable from 7 to
18 keV. The ID, the double-crystal monochromator, the ­ rst, second and third sets
of adjustable slits and the sample were located at 0, 30, 35, 54, 66 and 68 m, respec-
tively, along the X-ray beam path from the synchrotron orbit. The size of the square
beam was de­ ned at the ­ rst and second sets of slits, which were both set to 100 m m.
A parasitic scattering slit, having the shape of a round pinhole only large enough
to circumscribe the square beam, was placed 1{2 mm in front of the sample. The
two-dimensional CCD (Mar) detector had 2048 £ 2048 pixels with a 16-bit intensity
scale, and a circular active area of 133 mm diameter. In all cases the detector was
used in a 4 £ 4 binning mode at a resolution of 512 £ 512, with e¬ective pixel size
of 26 m m per pixel. The detector was placed at the end of an evacuated 8 inch pipe
­ tted with Kapton® R windows on both ends. The sample-to-detector distance (SDD)
was adjustable from a few centimetres to 8.5 m. PVB samples were stretched using
the custom mechanical loadframe (Instron) installed at DND-CAT.
Figure 12 shows the observed scattering at two di¬erent angles, showing that
the voids have structure in two distinct length-scale ranges. These were studied
in separate experiments by choosing detector distance to be 8.5 m for meridional
scattering and 0.72 m for equatorial scattering.
For meridional scattering, the ID was tuned to an energy of 7.95 keV, corre-
sponding to a wavelength ¶ = 1:560 A̧. At this distance the detector covered scat-
tering angles 2³ corresponding to the range in the scattering vector magnitude
8 £ 10¡4 < q (= 4º sin ³ =¶ ) < 3 £ 10¡2 A̧¡1 . This regime of scattering angles is
termed the ultra-SAXS or USAXS range, which is achieved with di¯ culty in the
laboratory. Samples were stretched at the rate of 1 mm s¡1 while load versus strain
information in addition to scattering data were collected simultaneously. The bot-
tom actuator moved at an equal rate but in the opposite direction to that of the top
actuator. This mode of operation enabled the central portion of the tensile bar to
remain in the beam while the sample was stretched. The frame formed by the vertical
columns and the crossheads, pivoted on a trundle mount set on a sturdy external
support. Remote lateral positioning of the sample, with an accuracy of 250 m m, was
achieved by the use of a third actuator, which displaced the frame horizontally in
the plane perpendicular to the beam. Scattered intensity (I) and load/strain data
were collected simultaneously as the sample was deformed. A video extensiometer
was used for local strain measurements. The maximum deformation was 406 mm.
The exposure times at the short and long SDD were 1 and 0.2 s. Readout of the
pixel array limited the data rate to a minimum of ca. 0.3 Hz. Frames were collected
at the rate of 0.43 Hz and the exposure time for each frame was 200 ms. Frames
1{10 were collected on the as-received sample, without the application of strain.
The initial pattern was isotropic. Frames 11{58 were collected as the sample was
stretched, and anisotropy developed with strain. In this con­ guration the meridional
streak was observed. Data were corrected for electronic noise from the detector, for
air/instrumental scattering, and for thinning of the sample. For equatorial scatter-
ing the ID was tuned to an energy of 8.048 keV, corresponding to a wavelength
¶ = 1:5405 A̧. A range 0:02 < q < 0:35 A̧¡1 was covered in this con­ guration. This
is a typical SAXS range covered by slit and pinhole collimated instruments in the
laboratory. The equatorial streak was weaker in comparison with the meridional
streak. An interesting observation was that in all experiments, scattering and hence

Proc. R. Soc. Lond. A (2003)


Crack blunting and the strength of soft elastic solids 1515

presumably the defect structure depended only on strain, and not on rate of loading
nor on time.
The meridional component of the scattering increases in intensity with deforma-
tion. Kratky plots (Iq 2 versus q) of meridional cuts obtained from the 2D data show
a peak that increases in intensity and moves to lower q as elongation proceeds. These
data have been ­ tted to a model of polydisperse spherical voids in a dense matrix
to obtain an e¬ective void size. More detailed analysis of these data to extract other
features of the structure will be presented elsewhere. The shape of the distribution
is assumed to be log-normal. Results from the ­ ts, shown in ­ gure 13, are given
in terms of the most probable particle diameter 2ro . The stress versus strain data,
obtained simultaneously with the scattered intensity, are also shown in ­ gure 13. At
a logarithmic strain of about 0.5 the material starts to sti¬en as the stress versus
strain curve becomes progressively steeper. As ­ gure 13 indicates, the SAXS inten-
sity is observed concomitantly with the sti¬ening of the material, but not until this
point. Subsequently, the change in shape and magnitude of the SAXS intensity is
coincident with an increase in the size of the voids.

References
Andrew, E. H. & Kinloch, A. J. 1973 Mechanics of adhesive failure. I, II. Proc. R. Soc. Lond.
A 332, 385{399, 401{414.
Atkins, A. G. & Mai, Y. W. 1988 Elastic and plastic fracture. Chichester: Ellis Horwood.
Barenblatt, G. I. 1968 The mathematical theory of equilibrium cracks in brittle fracture. Adv.
Appl. Mech. 7, 55{129.
Brown, H. R. 1991 A molecular interpretation of the toughness of glassy-polymers. Macro-
molecules 24, 2752{2756.
Chaudhury, M. K., Weaver, T., Hui, C. Y. & Kramer, E. J. 1996 Adhesive contact of cylindrical
lens and a ° at sheet. J. Appl. Phys. 80, 30{37.
Creton, C. & Lakrout, H. 2000 Micromechanics of ° at-probe adhesion tests of soft viscoelastic
polymer ¯lms. J. Polymer Sci. B 38, 965{979.
Crosby, A. J., Shull, K. R., Lakrout, H. & Creton, C. 2000 Deformation and failure modes of
adhesively bonded elastic layers. J. Appl. Phys. 88, 2956{2966.
de Gennes, P.-G. 1996 Soft adhesives. Langmuir 12, 4497{4500.
Ferry, J. D. 1980 Viscoelastic properties of polymers, 3rd edn. Wiley.
Gao, Y. C. 1997 Large deformation ¯eld near a crack tip in a rubber-like material. Theor. Appl.
Fract. Mech. 26, 155{162.
Gent, A. N. 1996 Adhesion and strength of viscoelastic solids. Is there a relationship between
adhesion and bulk properties? Langmuir 12, 4492{4496.
Gent, A. N. & Lai, S. M. 1994 Interfacial bonding, energy dissipation and adhesion. J. Polymer
Sci. C 32, 1543{1555.
Gent, A. N. & Petrich, R. 1969 Adhesion of viscoelastic materials to rigid substrates. Proc. R.
Soc. Lond. A 310, 433{448.
Gent, A. N. & Schultz, J. 1972 E® ect of wetting liquids on strength of adhesion of viscoelastic
materials. J. Adhes. 3, 281.
Gent, A. N. & Wang, C. 1991 Fracture Mechanics and cavitation in rubber-like materials. J.
Mater. Sci. 26, 3392{3395.
Geubelle, P. H. & Knauss, W. G. 1994 Finite strains at the tip of a crack in a sheet of hyperelastic
materials. I. Homogeneous case. J. Elastic. 35, 61{98.
Ghatak, A., Vorvolakos, K., She, H. Q., Malotky, D. L. & Chaudhury, M. K. 2000 Interfacial
rate processes in adhesion and fracture. J. Phys. Chem. B 104, 4018{4030.

Proc. R. Soc. Lond. A (2003)


1516 C.-Y. Hui, A. Jagota, S. J. Bennison and J. D. Londono

Hui, C. Y., Ruina, A., Creton, C. & Kramer, E. J. 1992 Micromechanics of crack growth into a
craze in a polymer glass. Macromolecules 25, 3948{3955.
Israelachvili, J. 1992 Intermolecular and surface forces, 2nd edn. Academic.
Jagota, A., Bennison, S. J. & Smith, C. A. 2000 Analysis of a compressive shear test for adhesion
between elastomeric polymers and rigid substrates. Int. J. Fract. 104, 105{130.
Johnson, K. L., Kendall, K. & Roberts, A. D. 1971 Surface energy and the contact of elastic
solids. Proc. R. Soc. Lond. A 342, 301.
Knauss, W. G. 1993 Time dependent fracture and cohesive zones. J. Engng Mater. Technol.
115, 262{267.
Knowles, J. K. & Sternberg, E. 1983 Large deformations near a tip of an interface-crack between
two neo-Hookean sheets. J. Elastic. 13, 257{293.
Lake, G. J. & Thomas, A. G. 1967 The strength of highly elastic materials. Proc. R. Soc. Lond.
A 300, 108.
Lakrout, H., Sergot, P. & Creton, C. 1999 Direct observation of cavitation and ¯brillation in a
probe tack experiment on model acrylic pressure-sensitive adhesives. J. Adhes. 69, 307{359.
Lawn, B. R. 1993 Fracture of brittle solids, 2nd edn. Cambridge University Press.
Lin, Y. Y., Hui, C. Y. & Jagota, A. 2001 The role of viscoelastic adhesive contact on the sintering
of polymeric particles. J. Colloid Interface Sci. 237, 267{282.
Maugis, D. 1992 Adhesion of spheres: the JKR{DMT transition using a Dugdale Model. J.
Colloid Interface Sci. 150, 243.
Mazur, S. & Plazek, D. J. 1994 Viscoelastic e® ects in the coalescence of polymer particles. Prog.
Org. Coat. 24, 225.
Mowery, C. L., Crosby, A. J., Ahn, D. & Shull, K. R. 1997 Adhesion of thermally reversible gels
to solid surfaces. Langmuir 23, 6101{6107.
Ogden, R. W. 1984 Non-linear elastic deformations. New York: Dover.
Rahul Kumar, P., Jagota, A., Bennison, S. J., Saigal, S. & Muralidhar, S. 1999 Polymer inter-
facial fracture simulations using cohesive elements. Acta Mater. 47, 4161{4169.
Rahul Kumar, P., Jagota, A., Bennison, S. J. & Saigal, S. 2000 Cohesive element modeling
of viscoelastic fracture: application to peel testing of polymers. Int. J. Solids Struct. 37,
1873{1897.
Sha, Y., Hui, C. Y., Ruina, A. & Kramer, E. J. 1995 Continuum and discrete modeling of craze
breakdown in glassy polymers. Macromolecules 28, 2450{2459.
Timoshenko, S. P. & Goodier, J. N. 1970 Theory of elasticity, 3rd edn. McGraw-Hill.
Tvergaard, V. & Hutchinson, J. W. 1992 The relation between crack growth resistance and
fracture process parameters in elastic{plastic solids. J. Mech. Phys. Solids 40, 1377.
Xu, X. P. & Needleman, A. 1994 Numerical simulations of fast crack growth in brittle solids. J.
Mech. Phys. Solids 42, 1397{1434.
Yeoh, O. H. 1993 Some forms of the strain energy function for rubber. Rubber Chem. Technol.
66, 754{771.

As this paper exceeds the maximum length normally permitted,


the authors have agreed to contribute to production costs.

Proc. R. Soc. Lond. A (2003)

You might also like