You are on page 1of 11

Engineering Fracture Mechanics 95 (2012) 2–12

Contents lists available at SciVerse ScienceDirect

Engineering Fracture Mechanics


journal homepage: www.elsevier.com/locate/engfracmech

An implicit gradient plasticity–damage theory for predicting size effects


in hardening and softening
R.H.J. Peerlings a,⇑, L.H. Poh a,b, M.G.D. Geers a
a
Department of Mechanical Engineering, Eindhoven University of Technology, P.O. Box 513, 5600 MB Eindhoven, Netherlands
b
Department of Civil Engineering, National University of Singapore, 1 Engineering Drive 2, E1A-07-03, Singapore 119260, Singapore

a r t i c l e i n f o a b s t r a c t

Keywords: An implicit gradient plasticity–damage theory is constructed, which can capture size
Plasticity effects in hardening plasticity, as well as regularise the localisation of deformation due
Damage mechanics to softening. In hardening, boundary layers of finite thickness are formed due to constraints
Ductile fracture imposed on the plastic deformation. Upon softening, a localisation band of finite thickness
Shear bands
emerges. Both thicknesses are governed by an intrinsic length scale which is incorporated
Size effects
via an implicit gradient formalism. Numerical solutions for a constrained layer in simple
shear illustrate the typical responses obtained. As the thickness of the layer is varied, size
effects are predicted in hardening as well as in softening. As the damage evolution pro-
gresses, the strain distribution progressively localises, culminating in a discrete crack.
Ó 2012 Elsevier Ltd. All rights reserved.

1. Introduction

Over the past two decades, higher-order continua, or generalised continua, have seen a tremendous development, from
exotic theoretical exercises towards viable and effective alternatives where classical continuum theories do not suffice. The-
oretical frameworks have been formulated for higher-order elasticity, plasticity, crystal plasticity, damage, fracture and com-
binations of these, see e.g. [1–6]. At the same time, computational algorithms have been developed which allow one to solve
the governing equations accurately and efficiently [7–10].
Most of these developments are motivated by one of two classes of problems which classical continuum mechanics can-
not properly deal with. The first is the pathological localisation of deformation which is predicted by classical inelastic con-
tinuum theories at the onset of material softening. Once the maximum strength of a material is reached, conventional
plasticity and damage theories predict the inelasticity to instantaneously localise into a discontinuity. However, the consti-
tutive relations used in such theories assume a continuous deformation, as well as a finite amount of energy dissipation per
unit of volume. As a consequence, no further energy dissipation occurs upon localisation and a perfectly brittle fracture is
obtained [11]. In numerical simulations, this localisation is limited by the spatial discretisation, resulting in an apparent
pathological dependence on the grid size. Furthermore, predicted crack paths exhibit a pathological dependence on the ori-
entation of the grid.
In damage mechanics, a popular remedy is the use of a nonlocal averaging operator, as proposed by Pijaudier-Cabot and
Bažant in the late 1980s [1]. This (integral) averaging operator introduces an intrinsic length scale and regularises the evo-
lution of damage such that it fluctuates smoothly at that scale. Similar concepts have been used in softening plasticity [12].
However, most softening plasticity theories traditionally adopt a different type of regularisation, by introducing an explicit
dependence of the yield stress on the gradient – usually second-order – of the plastic strain, e.g. [7]. Dimensional consistency

⇑ Corresponding author. Tel.: +31 40 2472788; fax: +31 40 2447355.


E-mail addresses: R.H.J.Peerlings@tue.nl (R.H.J. Peerlings), L.H.Poh@tue.nl (L.H. Poh), M.G.D.Geers@tue.nl (M.G.D. Geers).

0013-7944/$ - see front matter Ó 2012 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.engfracmech.2011.12.016
R.H.J. Peerlings et al. / Engineering Fracture Mechanics 95 (2012) 2–12 3

Nomenclature

e strain tensor
ee elastic strain tensor
ep plastic strain tensor
ep effective plastic strain
ep nonlocal effective plastic strain
r Cauchy stress tensor
re equivalent stress
C isotropic elasticity tensor
G shear modulus
f yield function
ry initial yield stress
h hardening modulus

h nonlocal hardening modulus
j history variable
j0 initial value for j
jc critical value for j
x damage variable
l intrinsic length scale
~
r gradient operator
H shear layer thickness
x thickness coordinate
U applied displacement
u displacement
c local shear
C average shear
s shear stress
cy shear at yield
sy shear stress at yield

then also requires an intrinsic length scale, which has a similar effect as in nonlocal theories. A third category of higher-order
damage and plasticity–damage theories includes gradients in a somewhat different fashion, which we term implicit [2,8,13].
They require the solution of an additional partial differential equation, next to equilibrium, in terms of an independent kine-
matic variable and its gradients. The method has been shown to be largely equivalent to the nonlocal approach, but is more
friendly from a computational standpoint since it does not require the implementation of a nonlocal operator. Apart from
predicting crack initiation, it has also been demonstrated to be able to predict crack growth, provided numerical methods
are employed which allow a discontinuity to appear in the solution. [14,15]. No additional fracture criteria are required
in such models, since the crack growth rate and direction are entirely governed by the damage evolution ahead of the crack
tip. However, the success of this approach towards fracture critically depends on the proper regularisation properties of the
underlying damage/plasticity theory.
A second class of problems in which conventional continuum theories reach their limits of applicability is the plastic
deformation of metals and alloys at the scale of microns. At this scale, heterogeneities in the plastic strain distribution
may arise which cannot be captured by conventional continuum plasticity. These heterogeneities often take the form of
boundary layers in which the plastic deformation is constrained by boundary conditions or interface conditions (e.g. at phase
or grain boundaries). In much larger specimens, the effect of these boundary layers may be neglected, but for small (or thin)
structures it may contribute significantly to the overall mechanical response. It then generally results in a ‘smaller is harder’
trend – see for instance the early experimental results of [16,17]. On the other hand, additional freedom due to e.g. free sur-
faces may result in soft boundary layers and thus a ‘smaller is softer’ trend, e.g. [18].
Higher-order continuum theories which have been formulated to capture such size effects in hardening plasticity gener-
ally introduce a dependence of the yield stress on the (second) gradient of plastic strain – see e.g. early work by Aifantis [19]
as well as more recent work in which these ideas have been extended [4,5,20]. Nonlocal (integral) plasticity theories have
also been proposed, as well as an implicit gradient formulation [10,21]. Each of these theories introduces at least one intrin-
sic length scale which governs the thickness of the boundary layers which emerge upon plastic deformation. For structural
dimensions on the order of this intrinsic length scale, a size-dependent response is predicted, whereas for larger dimensions
the conventional bulk response is obtained.
For a representative selection of theories for both classes of problems as discussed above, Engelen et al. [22] have com-
pared predictions made in terms of size effects in hardening as well as their localisation behaviour in softening. The main
4 R.H.J. Peerlings et al. / Engineering Fracture Mechanics 95 (2012) 2–12

conclusion is that theories designed to deal with one class of problems produce questionable results when applied to the
other class. This raises the question how to deal with problems in which both effects are relevant, i.e. size effects are ob-
served in hardening, as well as localised plastic strains and softening due to some degradation mechanism. Such combined
effects may for instance be expected in metallic microsystems, which contain micron-sized structures and are thus prone to
size effects in hardening, but on the other hand may also exhibit a ductile damage like failure process [23,24].
Another class of materials for which size effects occur in hardening as well as in softening are metallic foams. With this
application in mind, Dillard et al. [25] have formulated a theory to combine both effects and thus answer the above question.
It uses an existing micromorphic framework for hardening and introduces softening via a hardening law which reaches a
maximum and then softens. The material parameters were calibrated to experiments on a nickel foam and a comparison
of predictions with experimental data was made.
In this short paper we follow a different route towards a theory which can predict size effects in hardening and at the
same time regularise the localisation of deformation in softening. Our approach is based on a combination of (purely) hard-
ening plasticity and a damage influence to incorporate softening. As a basis for our development, we take the implicit gra-
dient theories for hardening proposed by Peerlings [10] and the earlier implicit gradient plasticity–damage theory by
Engelen et al. [13]. The latter has been demonstrated to exhibit favourable localisation properties. In particular, the strain
localisation associated with damage growth naturally converges to a displacement discontinuity in the limit of complete
fracture. Furthermore, compared with the nonlocal and (explicit/strain) gradient approaches discussed above, the treatment
of boundaries is natural and straightforward. In particular, the enforcement of special conditions at the internal elastic–plas-
tic boundary, as required in the explicit type of gradient theories, is avoided – see the discussion in Ref. [10].
The structure of the paper is as follows. We start by a brief summary of the existing implicit gradient theories for soft-
ening (Section 2) and hardening (Section 3). These two theories are subsequently integrated to obtain a novel theory for
hardening and softening in Section 4. The typical responses obtained are illustrated by the example of a constrained shear
layer in Section 5, before we close with a brief discussion in Section 6.

2. Implicit gradient plasticity–damage for softening

An implicit gradient formulation of damage coupled with plasticity has been proposed by Engelen et al. [13], partially
based on insights obtained for an elasticity-based gradient damage theory [8]. The original small-strain theory has later been
extended to finite strain by Geers [26] and has been further refined by Mediavilla et al. [27] by incorporating a dependence of
the damage evolution on the stress triaxiality. Here, however, we limit ourselves to the original small-strain version of Enge-
len et al. [13] for simplicity.
Adopting the usual additive split of the strain tensor e into an elastic part ee and a plastic part ep, the elastic response of
the material is given by
r ¼ C : ðe  ep Þ ð1Þ
with r the Cauchy stress tensor and C the fourth-order, isotropic elasticity tensor.
The elastic limit is given by a yield criterion f = 0, where the yield function f is formulated as
f ¼ re  ð1  xÞðry þ hep Þ ð2Þ

In this expression, re and ep are the (von Mises) equivalent stress and effective plastic strain according to the usual defini-
tions. The initial yield stress ry and the hardening modulus h are assumed to be constant; in the absence of damage, linear
hardening is thus assumed. The factor (1  x), in which x is the damage variable, progressively reduces the effective yield
stress as the damage variable increases from its initial value x = 0 to the ultimate value of x = 1. This reduction in yield
stress results in a softening response and ultimately, for x = 1, a complete loss of local material strength.
The plasticity part of the model is completed by an associative flow rule, which can be written as

3r d
e_ p ¼ e_ p ð3Þ
2r e
with rd the deviatoric part of r.
The regularisation enters the constitutive response via the evolution of the damage variable x. This is natural since it is
the damage evolution which results in softening and which thus would cause pathological localisation of the deformation if a
standard, local continuum were to be used. In Engelen et al.’s early formulation, the evolution of x is driven by the effective
plastic strain ep [13]. A nonlocal counterpart to the effective plastic strain, the nonlocal effective plastic strain ep , is defined as
the solution of the partial differential equation

ep  l2 r2 ep ¼ ep ð4Þ

where r2 ep is the Laplacian of ep and l an intrinsic length scale. This intrinsic length parameter should be representative of
the microstructural scale at which the relevant micro-damage processes characterised by the evolution of the damage x take
place. In the continuum model, it governs the width of the regularised damage bands obtained, as evidenced by numerical
simulations using this model and more advanced versions of it, see e.g. [13,26,27].
R.H.J. Peerlings et al. / Engineering Fracture Mechanics 95 (2012) 2–12 5

It has been shown by Forest [6] that Eq. (4) may be interpreted as a non-standard, additional force balance in the spirit of
micromorphic media – see also [21]. Associated with it is a non-standard boundary condition, which we choose here to be
the natural (‘traction-free’) boundary condition

~ ~ ep ¼ 0
nr ð5Þ

where ~n is the unit outward normal to the boundary. It is worth mentioning that this boundary condition is always to be
applied on the exterior boundary of the body under analysis. This is unlike many other higher-order theories, which require
such additional boundary conditions at the elastic–plastic boundary. Since the position of the elastic–plastic boundary is un-
known a priori – as determining it is part of the analysis itself – imposing higher-order boundary conditions on it raises the-
oretical questions as well as practical (implementation) difficulties – see the discussion in Ref. [10].
Kuhn–Tucker conditions

ðep  jÞj_ ¼ 0 ep  j 6 0 j_ P 0 ð6Þ

together with an initial value j = j0 define the evolution of a history variable j, which in turn determines the damage var-
iable x. Here we adopt the simple bi-linear damage law
( jj0
jc j0 if j < jc
x¼ ð7Þ
1 otherwise

Damage is thus initiated once j exceeds the initial value j0, and increases linearly with j until complete failure occurs for
j = jc, where jc is a parameter which is greater than j0.
It is the dependence of damage evolution, and thus of the softening contribution in the hardening/softening law (2), on
the nonlocal effective plastic strain ep which regularises the problem. In the conventional, local model, which is obtained by
setting l = 0, any tendency of the effective plastic strain ep to grow rapidly leads to an increased damage growth, which in
turn leads to an accelerated straining. In the nonlocal model, however, rapid fluctuations of the local effective plastic strain
ep propagate only partially to the smoother nonlocal effective strain ep . As a result, damage is initiated in a region of finite
width, which scales with l [13].
As the damage process evolves, further damage growth is increasingly localised. As a consequence, the strain progres-
sively localises as well, culminating in a strong discontinuity – a crack – at complete failure (i.e. at x = 1) – see e.g. the
numerical results of Engelen et al. [13] or the spectral analysis by Di Luzio and Bažant [28]. A natural transition to a discrete
crack is thus obtained, see e.g. [11]. If the proper numerical methods are used which allow this discontinuity to evolve, the
subsequent growth of the crack may also be simulated, see e.g. [14,15]. Since as a result of the intrinsic length scale a finite
volume takes part in the damage process leading to final failure, a finite amount of energy is dissipated and no pathological
localisation is observed.
The fact that an intrinsic length scale is embedded in the theory also implies that size effects are predicted as the struc-
tural scale of the problem, L, is varied with respect to the intrinsic length scale, l. The width of the damage band is essentially
governed by l. The material outside this band starts to unload elastically once the peak in the load–displacement response is
reached. A higher ratio L/l thus results in relatively more unloading and a more brittle post-peak response. In inhomogeneous
problems, the peak load itself may also be affected by this ratio.

3. Implicit gradient plasticity for hardening

An implicit gradient plasticity theory for predicting size effects in hardening has been discussed by Peerlings [10]. It em-
ploys the fact that in regions with heterogeneous plastic strain the difference between the local effective plastic strain ep and
the nonlocal effective plastic strain ep as defined by (4) may become large, whereas in regions of uniform plastic strain it
vanishes. This allows one to approximate the second gradient of ep featuring in many (explicit) gradient plasticity theories
by ep  ep . A thermodynamical foundation for this theory, based on the concept of generalised micromorphic continua as
coined by Forest [6], has been recently formulated by Poh et al. [21] for a scalar as well as a tensorial version of the theory.
Here, we adopt the simple scalar formulation as discussed in [10]. In this theory, the stress–strain law, flow rule and addi-
tional balance equation are identical to those of the softening model discussed in Section 2, i.e. they are given respectively by
(1), (3) and (4). However, the yield function (2) is replaced by one that does not have a damage influence and instead has an
additional hardening term:
 ep  ep Þ
f ¼ re  ½ry þ hep þ hð ð8Þ

This term, hð ep  ep Þ, enhances the hardening in regions where the local plastic strain exceeds the nonlocal strain, e.g. due to
boundary constraints imposed on ep . On the other hand, the amount of hardening is reduced where ep is lower than the aver-
age plastic strain ep . The strength of this additional hardening depends on the parameter h.  Note that indirectly the length
scale l is also of influence, since it determines, via (4), the smoothness of ep and thus the magnitude of the difference ep  ep
and the additional hardening contribution resulting from it.
6 R.H.J. Peerlings et al. / Engineering Fracture Mechanics 95 (2012) 2–12

The ‘traction-free’ boundary condition (5) is a natural companion for the partial differential Eq. (4) at boundaries at which
the plastic deformation is unrestricted. For uniform plastic straining this results in ep ¼ ep and thus no additional hardening.
Where the plastic straining is severely constrained, e.g. by a passivation layer, the essential boundary condition ep ¼ 0 may
be more appropriate. Enforcing this condition results in a boundary layer with a harder response and thus less plastic strain.
If the structural size, L, is of the same order as the thickness of this boundary layer, which scales with l, the overall response is
observed to be harder than predicted by conventional elasto-plasticity. The magnitude of this additional hardening depends
on the ratio L/l and it thus gives rise to a size effect.
Note that, irrespective of which of the above two boundary conditions is used and like in the softening plasticity theory of
Section 2, it must be enforced on the external boundary of the problem and never at the (internal) elastic–plastic boundary.
Implementation difficulties associated with the interface conditions required at the elastic–plastic boundary are thus
avoided – see the discussion in [10].

4. Combined theory

A simple combined theory for hardening and softening is obtained from the gradient plasticity–damage theory discussed
in Section 2 by replacing the hardening part of yield function (2), i.e. the factor (ry + hep) by that of the size-dependent yield
function (8). This results in
 ep  ep Þ
f ¼ re  ð1  xÞ½ry þ hep þ hð ð9Þ

Combining this yield function with Eqs. (1), (3)–(7), a generalised theory for hardening and softening is obtained. The soft-
ening theory of Section 2, without any influence of gradients in hardening, is recovered by setting h  ¼ 0. The gradient-hard-
ening model of Section 3, on the other hand, follows by setting j0 to a sufficiently high value, so that no damage is ever
initiated.
A spectral analysis of the linear comparison solid fully parallel to that of [28] (and omitted for brevity here) predicts that
upon complete failure (x = 1) the deformation is fully localised. The regularised softening and localisation thus naturally cul-
minates in a strain discontinuity – a feature which is desirable when modelling crack initiation and, possibly, growth – see
e.g. [11].
In the simple form presented here, the same governing Eq. (4) is used in both scale-dependent parts of the model. This
implies two assumptions which may not always be justified. First, strengthening due to strain gradients and the regularisa-
tion of plastic strain localisation are assumed to be controlled by the same variable, the (nonlocal) effective plastic strain. In
reality, the evolution of ductile damage is known to depend on the stress triaxiality, whereas there is no reason to assume
that the hydrostatic stress would have a significant effect on strengthening due to heterogeneous plasticity. Second, the same
intrinsic length scale applies to the two processes. As the physical processes underlying hardening and softening are differ-
ent, one could imagine that the intrinsic length scales associated with them should also be different. To deal with this lim-
itation, more involved versions of the theory may be imagined in which two equations like (4) are used for two different
variables – possibly with two different length scales.
A finite element formulation of the theory may be derived by casting the usual equilibrium equation (or balance of
momentum) as well as Eq. (4) in a weak form. Discretisation by finite element shape functions according to Galerkin’s meth-
od then results in a system of nonlinear equations in terms of the nodal displacements and nonlocal effective plastic strains,
which may be solved e.g. by Newton–Raphson iterations. The implementation of these equations is fully parallel to that of
the softening plasticity–damage theory as discussed in Engelen et al. [13] and is therefore not detailed any further here.

5. Application to constrained shear

In order to study the response predicted by the combined theory as proposed above we model a simple shear problem
which has been used before e.g. by Bittencourt et al. [29] to study size effects in hardening. The geometry and boundary con-
ditions of the problem are illustrated in Fig. 1. An infinitely long and wide strip of thickness H is sandwiched between two
rigid supports. The top support is displaced parallel to the strip, in x1-direction, by a monotonically increasing positive dis-
tance U, resulting in an overall (average) shear C = U/H. We assume that a simple shear deformation exists throughout the
thickness of the strip and that this deformation is uniform in x1 and x3. No displacement in x2 and x3 direction is assumed.
The local deformation of the strip thus is fully characterised by the horizontal displacement u = u1 and its gradient c = c12 =
@u1/@x2 = du/dx, with x = x2 the through-thickness coordinate.
Under the assumptions made above, conventional equilibrium requires the shear stress s = r12 to be constant in x and
equal to the reaction force (traction) on the supports, T. The conventional boundary conditions for u read u(0) = 0 and
u(H) = U. The additional governing equation, Eq. (4), reduces to the ordinary differential equation
2
d ep
ep  l2 ¼ ep ð10Þ
dx2
pffiffiffi
where for the assumed simple shear deformation ep ¼ cp = 3 with cp ¼ cp12 ¼ 2ep12 . We consider here only the essential
boundary condition, which fully constrains ep at both supports: ep ð0Þ ¼ ep ðHÞ ¼ 0.
R.H.J. Peerlings et al. / Engineering Fracture Mechanics 95 (2012) 2–12 7

Fig. 1. Geometry and boundary conditions of the constrained shear problem considered.

The elastic response of the strip is governed by the shear modulus G = 100ry. For the hardening modulus h, a value of h =
20ry is used throughout this section. The higher-order modulus h  equals h
 ¼ G. The damage parameters j and j , where
0 c
relevant, are respectively j0 = 0 and jc = 0.1. Unless stated otherwise, the thickness of the shear layer equals H = 10l, where
l is the intrinsic length scale of the theory.
The one-dimensional boundary value problem thus obtained is solved using the finite element method. Quadratic shape
functions are used for the displacement and linear shape functions for the nonlocal effective plastic strain. Time is discretised
by the implicit Euler method and full Newton–Raphson iterations are used to solve the resulting equations. In all simulations
a spatial discretisation of 100 elements and a time discretisation of approximately 100 increments are used.
We first study the case without damage evolution, i.e. the theory of Section 3 or, equivalently, the combined theory of
Section 4 with a value of j0 which is sufficiently high to ensure x = 0 at all times. For uniform shear, or for an infinitely thick
shear layer, the stress–strain response for the parameters as defined above is given by the dashed curve in Fig. 2. Note that
pffiffiffi
the two axes in this diagram are scaled by the initial yield point (cy, sy), where sy ¼ ry = 3 and cy = sy/G.
For a finite thickness of the layer, H = 10l, the evolution of the shear profile across the thickness, c(x), is illustrated in Fig. 3.
The constraint imposed on ep at both surfaces of the layer results in a reduced plastic slip near these surfaces. These hard
boundary layers extend well into the thickness of the layer. Fig. 4 shows the effect of increasing the thickness of the layer
to H = 20l and H = 50l. In terms of the absolute coordinate x, the thickness of the boundary layers remains approximately
constant. This implies that a smaller fraction of the layer’s thickness (H) is occupied by them – as is clearly visible in the
diagram. Indeed, for H/l = 50 a plateau of more or less constant shear is observed at the centre of the layer and boundary
effects are observed across approximately 50% of the thickness only.
In terms of the overall mechanical response of the shear layer, the presence of boundary layers may have a pronounced
effect if they occupy a significant fraction of the cross-section. This is illustrated in Fig. 5, which shows the applied shear
stress T versus shear deformation C for the three thicknesses considered above. In the reference case of H/l = 10, a signifi-
cantly increased hardening response is observed due to the constrained plastic deformation at both surfaces of the layer.
For larger H/l, this effect is less pronounced and in the limit H/l ? 1 the computed response approaches that of the uniform
case (dashed curve in the diagram; cf. Fig. 2).

2.5

1.5

0.5

0
0 5 10 15 20

Fig. 2. Stress–strain response for uniform straining (no gradient effects) for the linear hardening model without damage (dashed curve) and the combined
model with damage evolution (solid curve).
8 R.H.J. Peerlings et al. / Engineering Fracture Mechanics 95 (2012) 2–12

0.8

0.6

0.4

0.2

0
0 10 20 30 40 50

Fig. 3. Evolution of the shear strain distribution c(x) across the thickness of the shear layer for the size-dependent hardening plasticity theory (i.e. without
damage).

0.8

0.6

0.4

0.2

0
0 10 20 30 40 50

Fig. 4. Influence of the layer thickness H on the shear strain distribution c(x) for the hardening plasticity theory (no damage). Note that the vertical axis is
scaled by the layer thickness H.

2.5

1.5

0.5

0
0 5 10 15 20

Fig. 5. Influence of the layer thickness H on the shear traction–average shear strain response for the hardening plasticity theory (no damage). The dashed
curve represents the uniform case (cf. Fig. 2).
R.H.J. Peerlings et al. / Engineering Fracture Mechanics 95 (2012) 2–12 9

We now consider the response obtained for the combined plasticity–damage theory. For the material parameters as de-
fined above, the uniform shear response is given by the solid curve in Fig. 2. Since we have set j0 = 0 the damage starts to
grow at the yield point. Nevertheless, some hardening is still observed, albeit at a lower rate than for the undamaged mate-
rial (dashed curve). At c/cy  8 the damage influence starts to dominate the hardening, resulting in an overall softening. The
rate of softening appears to be quite low in the diagram because of the assumption of uniform straining. As we will see be-
low, in nonuniform cases the deformation and damage growth immediately localise once the peak stress is reached, resulting
in a significantly less ductile overall response.
Fig. 6 shows the evolution of the shear distribution c(x) for the finite shear layer and the combined gradient plasticity–
damage theory; see ahead to Fig. 8 for the different stages of loading from which these distributions were taken. Initially, a
similar distribution is observed as for the theory without damage – cf. Fig. 3. However, as soon as a significant amount of
damage is generated, a more localised deformation pattern appears. In a region at the centre of the shear layer the strain
continues to increase, whereas near the boundaries elastic unloading may be observed. This process results in an increas-
ingly sharp shear strain peak, which is consistent with earlier results obtained for implicit gradient models for softening only
[8,13]. In these studies, the strain was observed to converge towards a Dirac function (and thus the displacement field to a
jump) in the limit x ? 1. Such a convergence to a displacement discontinuity provides a natural transition to a crack. A
bifurcation analysis (not shown here) indeed suggests that for the present model the localisation width also converges to
zero in the fracture limit.
The damage evolution, which is shown in Fig. 7, shows a similar trend. Initially, when the material is still hardening, a
smooth distribution of damage is observed across the thickness of the shear layer. The damage growth is somewhat faster
at the centre of the layer because more plastic deformation occurs there. Note that as a result, no imperfection is necessary to

0.8

0.6

0.4

0.2

0
0 10 20 30 40 50

Fig. 6. Evolution of the shear strain distribution c across the thickness of the shear layer for the combined plasticity–damage theory. See Fig. 8 for the
average strain levels at which these snapshots were taken.

0.8

0.6

0.4

0.2

0
0 0.2 0.4 0.6 0.8 1

Fig. 7. Evolution of the damage distribution across the thickness of the shear layer for the combined plasticity–damage theory. See Fig. 8 for the average
strain levels at which these snapshots were taken.
10 R.H.J. Peerlings et al. / Engineering Fracture Mechanics 95 (2012) 2–12

trigger localisation. The later stages of damage growth are more localised, with hardly any growth near the surfaces and a
rapid increase at the centre of the layer. Note that the damage distribution is significantly smoother than that of the shear
strain, cf. [8,13]. No pathological localisation is observed, i.e. the damage band has a finite width and a finite amount of en-
ergy is therefore dissipated in the fracture process.
The overall traction–shear response is given by the solid curve in Fig. 8. The response without damage is also shown
(dashed curve; cf. Fig. 5) for comparison. At the onset of plastic yielding a strong hardening is observed, which is partially
due to the constraint imposed on the plastic deformation at both surfaces. As the damage evolves, a peak load is reached,
followed by softening. Compared to the uniform response of Fig. 2, this softening part is much steeper (less ductile) due
to the localisation of damage and shear strain at the centre of the shear layer. A snap-back response is observed in the final
stages of the loading process, i.e. the drop in stress (traction) is accompanied by a decreasing average shear. This instability is
due to partial elastic unloading near the constrained boundaries, as evidenced by the shear strain evolution in Fig. 6, which
starts to overwhelm the continued forward deformation in the shear band.
Fig. 9 shows the final damage distribution for the three different ratios H/l considered before. Note that the vertical axis of
the diagram is scaled by the thickness H. The thickest shear layer, for which H/l = 50, clearly shows the most pronounced
localisation of damage, in a band of approximately H/5. For the five times thinner layer considered as a reference case above
(i.e. H/l = 10) this implies that the damage band occupies virtually the entire thickness, which explains the relatively smooth
damage distribution observed for it.
In terms of the traction–average shear response, a more pronounced localisation (for large H/l) results in a more brittle
post-peak response. This is shown in Fig. 10, in which the response for the three different thicknesses is given. The diagram
highlights the two different size effects which are captured by the proposed theory: for smaller (thinner) samples, a stronger
hardening is observed, as well as a (relatively) more ductile post-peak response.

2.5

1.5

0.5

0
0 5 10 15 20

Fig. 8. Shear traction–average shear strain response for the combined plasticity–damage theory (solid curve). The markers indicate the increments from
which the shear strain and damage distributions of Figs. 6 and 7 were taken. The dashed curve is for the model without damage evolution.

0.8

0.6

0.4

0.2

0
0 0.2 0.4 0.6 0.8 1

Fig. 9. Influence of the layer thickness H on the damage distribution for the combined plasticity–damage theory.
R.H.J. Peerlings et al. / Engineering Fracture Mechanics 95 (2012) 2–12 11

2.5

1.5

0.5

0
0 5 10 15 20

Fig. 10. Influence of the layer thickness H on the shear traction–average shear strain response for the combined plasticity–damage theory.

6. Summary and concluding remarks

By combining earlier work on implicit gradient plasticity and damage theories for hardening and softening [13,10], we
have constructed a theory which is able to capture size effects observed in both regimes. In hardening, local constraints
on the plastic deformation may result in boundary layers which are harder than the bulk material. The overall response
is therefore also harder, particularly for small (or thin) structures, in which the boundary layers comprise a significant frac-
tion of the material: ‘smaller is harder’. Softening, for instance due to the nucleation and growth of damage, leads to a local-
isation of deformation in a narrow band, whereas the remaining material unloads elastically. In the limit of damage equal to
unity (fracture), a displacement discontinuity (crack) is obtained. Again, the size of the structure relative to that of the local-
isation band matters: ‘smaller is more ductile’.
Both trends are predicted by the implicit gradient theory which proposed here. The effect of boundary layers in hardening
is captured by comparing the local effective strain with a smoother nonlocal effective strain. Differences between these
quantities are an indication of heterogeneous plastic yielding and result in an increased hardening. Localisation in softening
is controlled by the same nonlocal effective plastic strain, via a damage variable. The intrinsic length scale introduced via the
nonlocality governs the localisation bandwidth in softening, as well as the boundary layer thickness in hardening. It should
be noted, however, that the thickness of the boundary layers and localisation band also depend on the other material param-
eters. In particular, the thickness of boundary layers in hardening depends on h,  whereas the localisation process in softening
is influenced by j0 and jc.
The fact that a single intrinsic length scale is used for both phenomena, and in fact that the same variable controls both
phenomena, is a limitation of the theory in the simple form presented here. The extension to two, or multiple, separate con-
trolling variables, each with their own length scale, is relatively straightforward. However, it comes at the price of having to
solve two partial differential equations akin to Eq. (4) rather than just one. In some cases, finally, it may also be necessary to
consider tensorial variables, cf. Ref. [21].
Similar combined theories for hardening and softening may be imagined in the context of (integral) nonlocal or explicit
strain gradient plasticity. The combined implicit gradient theory proposed here, however, inherits a major benefit from the
underlying theories for hardening and softening: complete (mathematical) clarity on the necessary non-standard boundary
conditions, which are always defined on the external boundary of the problem. As a result, efficient finite element solution
methods can be constructed, which can be implemented without much difficulty.

References

[1] Pijaudier-Cabot G, Bažant ZP. Nonlocal damage theory. J Engng Mech 1987;113:1512–33.
[2] Frémond M, Nedjar B. Damage, gradient of damage and principle of virtual power. Int J Solids Struct 1996;33:1083–103.
[3] Fleck NA, Hutchinson JW. Strain gradient plasticity. Adv Appl Mech 1997;33:295–361.
[4] Gurtin ME, Anand L. A theory of strain-gradient plasticity for isotropic, plastically irrotational materials. Part I: small deformations. J Mech Phy Solids
2005;53:1624–49.
[5] Gudmundson P. A unified treatment of strain gradient plasticity. J Mech Phy Solids 2004;52:1379–406.
[6] Forest S. Micromorphic approach for gradient elasticity, viscoplasticity, and damage. J Engng Mech 2009;135:117–31.
[7] de Borst R, Mühlhaus H-B. Gradient-dependent plasticity: formulation and algorithmic aspects. Int J Numer Methods Engng 1992;35:521–39.
[8] Peerlings RHJ, de Borst R, Brekelmans WAM, de Vree JHP. Gradient-enhanced damage for quasi-brittle materials. Int J Numer Methods Engng
1996;39:3391–403.
[9] Niordson CF. Strain gradient plasticity effects in whisker-reinforced metals. J Mech Phy Solids 2003;51:1863–83.
[10] Peerlings RHJ. On the role of moving elastic–plastic boundaries in strain gradient plasticity. Modell Simul Mater Sci Engng 2007;15:109–20.
12 R.H.J. Peerlings et al. / Engineering Fracture Mechanics 95 (2012) 2–12

[11] Peerlings RHJ, de Borst R, Brekelmans WAM, Geers MGD. Localisation issues in local and nonlocal continuum approaches to fracture. Eur J Mech A/
Solids 2002;21:175–89.
[12] Leblond JB, Perrin G, Devaux J. Bifurcation effects in ductile metals with damage delocalisation. J Appl Mech 1994;61:236–42.
[13] Engelen RAB, Geers MGD, Baaijens FTP. Nonlocal implicit gradient-enhanced elasto-plasticity for the modelling of softening behaviour. Int J Plast
2003;19:403–33.
[14] Simone A, Wells GN, Sluys LJ. From continuous to discontinuous failure in a gradient-enhanced continuum damage model. Comput Methods Appl
Mech Engng 2003;192:4581–607.
[15] Mediavilla J, Peerlings RHJ, Geers MGD. Discrete crack modelling of ductile fracture driven by non-local softening plasticity. Int J Numer Methods
Engng 2006;66:661–88.
[16] Fleck NA, Muller GM, Ashby MF, Hutchinson JW. Strain gradient plasticity: theory and experiment. Acta Metall et Mater 1994;42:475–87.
[17] Stölken JS, Evans AG. A microbend test method for measuring the plasticity length scale. Acta Materiala 1998;46:5109–15.
[18] Janssen PJM, de Keijser TH, Geers MGD. An experimental assessment of grain size effects in the uniaxial straining of thin Al sheet with a few grains
across the thickness. Mater Sci Engng A 2006;419:238–48.
[19] Aifantis EC. On the microstructural origin of certain inelastic models. J Engng Mater Technol 1984;106:326–30.
[20] Fleck NA, Hutchinson JW. A reformulation of strain gradient plasticity. J Mech Phy Solids 2001;49:2245–71.
[21] Poh LH, Peerlings RHJ, Geers MGD, Swaddiwudhipong S. An implicit tensorial gradient plasticity model – formulation and comparison with a scalar
gradient model. Int J Solids Struct 2011;48:2595–604.
[22] Engelen RAB, Fleck NA, Peerlings RHJ, Geers MGD. An evaluation of higher-order plasticity theories for predicting size effects and localisation. Int J
Solids Struct 2006;43:1857–77.
[23] Coulombier M, Boé A, Brugger C, Raskin JP, Pardoen T. Imperfection-sensitive ductility of aluminium thin films. Scripta Mater 2010;62:742–5.
[24] Espinosa HD, Prorok BC, Peng B. Plasticity size effects in free-standing submicron polycrystalline FCC films subjected to pure tension. J Mech Phy Solids
2004;52:667–89.
[25] Dillard T, Forest S, Ienny P. Micromorphic continuum modelling of the deformation and fracture behaviour of nickel foams. Eur J Mech A/Solids
2006;25:526–49.
[26] Geers MGD. Finite strain logarithmic hyperelasto-plasticity with softening: a strongly nonlocal implicit gradient framework. Comput Methods Appl
Mech Engng 2004;193:3377–401.
[27] Mediavilla J, Peerlings RHJ, Geers MGD. A nonlocal triaxiality-dependent ductile damage model for finite strain plasticity. Comput Methods Appl Mech
Engng 2006;195:4617–34.
[28] Di Luzio G, Bažant ZP. Spectral analysis of localization in nonlocal and over-nonlocal materials with softening plasticity or damage. Int J Solids Struct
2005;42:6071–100.
[29] Bittencourt E, Needleman A, Gurtin ME, van der Giessen E. A comparison of nonlocal continuum and discrete dislocation plasticity predictions. J Mech
Phy Solids 2003;51:281–310.

You might also like