Download as pdf or txt
Download as pdf or txt
You are on page 1of 18

Downloaded from http://rspa.royalsocietypublishing.

org/ on April 8, 2016

A pseudo-elastic model for the


Mullins effect in filled rubber
By R. W. O g d e n a n d D. G. Roxburgh†
Department of Mathematics, University of Glasgow,
University Gardens, Glasgow G12 8QW, UK

Received 26 August 1998; accepted 2 November 1998

When a rubber test piece is loaded in simple tension from its virgin state, unloaded
and then reloaded, the stress required on reloading is less than that on the initial
loading for stretches up to the maximum stretch achieved on the initial loading. This
stress softening phenomenon is referred to as the Mullins effect. In this paper a sim-
ple phenomenological model is proposed to account for the Mullins effect observed in
filled rubber elastomers. The model is based on the theory of incompressible isotropic
elasticity amended by the incorporation of a single continuous parameter, interpreted
as a damage parameter. This parameter controls the material properties in the sense
that it enables the material response to be governed by a strain-energy function on
unloading and subsequent submaximal loading different from that on the primary
(initial) loading path from the virgin state. For this reason the model is referred
to as pseudo-elastic and a primary loading–unloading cycle involves energy dissipa-
tion. The dissipation is measured by a damage function which depends only on the
damage parameter and on the point of the primary loading path from which unload-
ing begins. A specific form of this function with two adjustable material constants,
coupled with standard forms of the (incompressible, isotropic) strain-energy func-
tion, is used to illustrate the qualitative features of the Mullins effect in both simple
tension and pure shear. For simple tension the model is then specialized further in
order to fit Mullins effect data. It is emphasized that the model developed here is
applicable to multiaxial states of stress and strain, not just the specific uniaxial tests
highlighted.
Keywords: pseudo-elasticity; rubber elasticity; Mullins effect;
stress softening; solid mechanics

1. Introduction
The theory of isotropic elasticity, in particular the incompressible theory, has been
used extensively for modelling the mechanical response of rubber-like materials, such
as vulcanized natural rubber and synthetic rubbers (see, for example, the reviews
by Ogden (1982, 1986)). The success of the theory is reflected in its widespread use
for engineering design calculations, based mainly on finite-element software packages
such as Abaqus and Marc into which specific material models have been incorpo-
rated. Elasticity theory has also been applied to many of the rubber-like materials
used in engineering components, notably rubbers hardened by the inclusion of fillers
† Present address: Department of Engineering Mathematics, University of Newcastle, Stephenson
Building, Newcastle upon Tyne NE1 7RU, UK.

Proc. R. Soc. Lond. A (1999) 455, 2861–2877 c 1999 The Royal Society

Printed in Great Britain 2861 TEX Paper
Downloaded from http://rspa.royalsocietypublishing.org/ on April 8, 2016

2862 R. W. Ogden and D. G. Roxburgh

such as carbon black, which exhibit distinctly inelastic behaviour, exemplified by


hysteresis, permanent set, the Mullins effect and frequency-dependent response. It
is now recognized that more sophisticated theories which incorporate some or all of
these effects are needed in order to model more accurately the material response,
particularly in the light of the high-powered computing environments now widely
available. The expanding use of rubber products in engineering applications has
increased interest in developing such models for programming into software pack-
ages. A number of theories with this aim in mind have been developed in the last
few years (see, for example, Govindjee & Simo 1991, 1992a, b; Miehe 1995; Lion
1996), but these are all complex three-dimensional theoretical models involving many
parameters which, while accurately reflecting the available one-dimensional data, are
essentially untested in more general deformations.
Much of the recent work has concerned the Mullins effect. It has been known for
a long time that the initial material properties of unstretched (virgin) samples of
rubber compounds are changed after the sample has been subjected to load. This
was observed by Mullins (1947) and has subsequently become widely known as the
Mullins effect. The essence of the Mullins effect is described in § 3 a. In order to
explain this behaviour Mullins & Tobin (1957) and Mullins (1969) proposed a model
in which the rubber could exist in either one of two possible phases, the so-called
hard and soft phases. They reasoned that initially the rubber existed solely in the
hard phase and that as it was subjected to greater and greater deformation more
and more of the rubber would degrade into the soft phase. To measure the extent of
the conversion between the two phases, they introduced a damage parameter which
they related to strain amplification in the remaining hard phase of the rubber. There
was considerable interest in the Mullins effect around the time that Mullins & Tobin
formulated their theory, and Bueche (1960, 1961), Harwood et al . (1965) and Har-
wood & Payne (1966a, b) also implemented strain-amplification function approaches
in their models.
There was then a lengthy break in the study of the Mullins effect, and it was not
until fairly recently that interest was again stimulated. Govindjee & Simo (1991,
1992a) proposed a fairly complex hyperelastic-based damage model which drew on
their work in continuum damage mechanics (see Simo (1987) and Simo & Ju (1987)
for references), and then modified this to incorporate viscoelastic effects (Govindjee
& Simo 1992b). These continuum damage models work well for materials that behave
in a predominantly viscoelastic manner, such as solid rocket fuel, but not so well for
the stiffer rubber compounds. More recently, Johnson & Beatty (1993a, b) adopted a
much simpler approach for uniaxial states of stress based on the original two-phase
idea of Mullins. These authors developed a theory that is consistent with that of
Mullins & Tobin (1957) and, moreover, provides good agreement with experimental
data in simple tension. Their theory, however, is not easily adaptable to general
biaxial or triaxial deformations.
The present paper forms the first stage in a programme aimed at incorporating
various inelastic effects into the theory. We adopt a simple approach based firmly on
the successful isotropic elastic modelling mentioned in the first paragraph of § 1 and
which is applicable to multiaxial states of deformation and stress. In this paper we
focus solely on the Mullins effect, treated as a quasi-static phenomenon. Subsequent
papers will be concerned with hysteresis, residual strain (or permanent set), thermal
and viscoelastic effects.

Proc. R. Soc. Lond. A (1999)


Downloaded from http://rspa.royalsocietypublishing.org/ on April 8, 2016

A pseudo-elastic model for the Mullins effect 2863

Our starting point is the (quasi-static) incompressible isotropic theory of elasticity,


which is used to model the material response on any loading path from the unstressed
configuration of the virgin material. We refer to such a path as a primary loading
path. A single continuous damage parameter is incorporated into the theory in order
to modify the strain-energy function so that the material response is then different
on unloading from a point on the primary loading path (and on subsequent submax-
imal loading) from the response on a primary loading path, and, specifically, stress
softening is accounted for. The (equilibrium) equations associated with the general
form of this damage model are summarized in § 2. We emphasize that the theory
is purely phenomenological and does not take into account the underlying physical
structure of the material, such as the presence of carbon black filler or the volume
fraction of the filler. Thus, the model can be applied to any material exhibiting the
Mullins effect; there is evidence to suggest that the Mullins effect occurs in unfilled
as well as filled rubbers (see Rigbi (1980) and Johnson & Beatty (1993a) for refer-
ences). Since the material response is governed by different forms of strain-energy
function on primary loading and unloading, we refer to the model as pseudo-elastic.
This terminology was used by Fung (1980) for a similar concept in the context of
modelling biological tissue.
A specific form of the theory appropriate for the present discussion is introduced
and justified in § 3. In particular, we introduce a damage function which is directly
related to the energy dissipated in a primary loading–unloading cycle. The govern-
ing equations show that, through the deformation function, the damage parameter is
expressible (in general, implicitly) in terms of the deformation, thus providing, when
the parameter is active, both an evolution equation for the damage and a means
of modifying the strain-energy function. A particular form of the damage function
is then chosen to illustrate the main features of the Mullins effect in pure shear
and how these features respond to changes in the two adjustable parameters (mate-
rial constants) that the function includes. Some numerical results for pure shear
are discussed in § 4 with respect to each of the neo-Hookean and Ogden forms of
the strain-energy function. Results for both the main stress and the side stress are
included to show the relative effect of stress softening in the two directions. For sim-
ple tension we then use the model to fit experimental data obtained from Mullins
& Tobin (1957). Finally, in § 5, we emphasize that the model is valid, not just for
uniaxial tests such as pure shear and simple tension, but also for multiaxial states of
stress and deformation since the damage function is controlled by the strain energy
associated with the primary deformation and not just by the specific deformation
from which it is calculated.

2. Basic equations and notation


We consider the deformation of a continuous body which in its natural (undeformed
and unstressed) configuration is taken to occupy the region Ω. Material points are
identified by their position vectors X in Ω, with Cartesian coordinates Xα (α =
1, 2, 3). After deformation the body occupies the region Ωc , say, and the point X
is deformed to the position x with coordinates xi (i = 1, 2, 3). The deformation
gradient tensor, denoted F , has Cartesian components given by
∂xi
Fiα = (i, α = 1, 2, 3). (2.1)
∂Xα

Proc. R. Soc. Lond. A (1999)


Downloaded from http://rspa.royalsocietypublishing.org/ on April 8, 2016

2864 R. W. Ogden and D. G. Roxburgh

We assume that the material response of the body can be described in terms of
an energy function of the form W(F , η), defined per unit volume in Ω. This is the
prescription used in standard nonlinear elasticity theory except that an additional
scalar variable η has been included. Because of the interpretation and influence of
this variable, discussed below, it is no longer appropriate to regard W as a stored- (or
strain-) energy function and we therefore refer to it as a pseudo-energy function and
the ensuing constitutive theory as pseudo-elasticity, terminology used by Lazopoulos
& Ogden (1999). In this paper it suffices to take η to be a continuous variable. A
theory in which η is allowed to be discontinuous was discussed by Lazopoulos &
Ogden (1998, 1999) in the context of modelling phase-change phenomena.
The pseudo-energy function is required to satisfy the usual objectivity conditions,
for details of which we refer to Ogden (1984), for example. The material is taken to
be incompressible, so that F satisfies the constraint
det F = 1. (2.2)
The nominal and Cauchy stress tensors, S and σ, respectively, are then given by
∂W ∂W
S= (F , η) − pF −1 , σ=F (F , η) − pI, (2.3)
∂F ∂F
where p is a Lagrange multiplier associated with the constraint (2.2) and I is the
identity tensor.
From Lazopoulos & Ogden (1998), it follows that in equilibrium in the absence of
body forces,
Div S = 0 in Ω, (2.4)
∂W
(F , η) = 0 in Ω, (2.5)
∂η
where Div denotes the divergence operator in Ω. Equation (2.4) is the usual equi-
librium equation, while (2.5) is an additional equation arising from inclusion of the
variable η. Note that (2.5) is easily modified to account for more than one additional
variable. Moreover, if η is allowed to be discontinuous, then further equations (jump
conditions) are needed, as discussed in Lazopoulos & Ogden (1998, 1999).
The role of the parameter η can be interpreted as follows. During a deformation
process, η may be either active or inactive and may switch from inactive to active
and conversely provided it remains continuous in so doing. When it is not active, the
material behaves as an elastic material with strain-energy function W(F , η), η being
held constant. However, when η is active it is determined implicitly in terms of F
by equation (2.5); we write η = χ(F ) in this case. The material will again behave as
an elastic material, but with a different strain-energy function, namely W(F , χ(F )).
At a transition point where the parameter η switches on or off, as we see below, the
pseudo-energy function W and the associated stresses change continuously.

(a) Isotropy
Henceforth we restrict attention to isotropic materials. The pseudo-energy function
then depends on the deformation only through the (positive) principal stretches
λ1 , λ2 and λ3 , where λ2i (i = 1, 2, 3) are the eigenvalues of the right Cauchy–Green

Proc. R. Soc. Lond. A (1999)


Downloaded from http://rspa.royalsocietypublishing.org/ on April 8, 2016

A pseudo-elastic model for the Mullins effect 2865

stretch tensor C = F T F . With this dependence we write the pseudo-energy function


as W̄(λ1 , λ2 , λ3 , η). The principal values σ1 , σ2 , σ3 of the Cauchy stress are then given
by
∂ W̄
σi = λi (λ1 , λ2 , λ3 , η) − p (i = 1, 2, 3). (2.6)
∂λi
In terms of the stretches, the incompressibility constraint (2.2) takes the form
λ1 λ2 λ3 = 1. (2.7)
This may be used to eliminate λ3 in terms of λ1 and λ2 , which are then adopted
as the independent deformation measures. Accordingly, the pseudo-energy function
may be expressed as
W (λ1 , λ2 , η) ≡ W̄(λ1 , λ2 , λ−1 −1
1 λ2 , η), (2.8)
which defines the notation W . Substitution of (2.8) into (2.6) and elimination of p
yields
σ1 − σ3 = λ1 W1 , σ2 − σ3 = λ2 W2 , (2.9)
where Wβ = ∂W /∂λβ (β = 1, 2), while (2.5) simplifies to
∂W
(λ1 , λ2 , η) = 0. (2.10)
∂η
The latter equation determines η implicitly in terms of λ1 and λ2 .
Henceforth, we consider only homogeneous deformations, so that (2.4) is satisfied
identically.

(b) Simple tension and pure shear


For simple tension we may take σ2 = σ3 = 0 and we write σ1 = σ. We also write
λ1 = λ, so that λ2 = λ3 = λ−1/2 , and define Ŵ by
Ŵ (λ, η) ≡ W (λ, λ−1/2 , η). (2.11)
Equations (2.9) and (2.10) then simplify to
σ = λŴλ (λ, η) ≡ λt, Ŵη (λ, η) = 0, (2.12)
wherein the principal Biot stress t ( = t1 ) is defined and the subscripts signify partial
derivatives.
From the second equation in (2.12), η is determined in principle in terms of λ, and
from equations (2.12) it follows that
dt
= Ŵλλ − (Ŵλη )2 /Ŵηη . (2.13)

Equation (2.13) shows how the inclusion of the variable η modifies the stiffness Ŵλλ
of the material appropriate to inactive η.
For pure shear, λ1 = λ, λ2 = 1, λ3 = λ−1 and the first equation in (2.12) applies
for the main stress but with Ŵ (λ, η) defined as W (λ, 1, η) instead of by (2.11), while
the side stress is given by
σ2 = W2 (λ, 1, η), (2.14)
which is equal to the corresponding principal Biot stress t2 .

Proc. R. Soc. Lond. A (1999)


Downloaded from http://rspa.royalsocietypublishing.org/ on April 8, 2016

2866 R. W. Ogden and D. G. Roxburgh

c'
c

stress (t)
b'

b C
B

λ
Figure 1. Schematic loading–unloading curves in simple tension (Mullins effect).

3. A model for the Mullins effect


(a) Description of the Mullins effect
Figure 1 shows the main features of the Mullins effect in simple tension in schematic
form, with the stress t plotted against λ. This is an ideal representation of the Mullins
effect since in practice there is some permanent set (residual strain) and hysteresis.
Consider the primary loading path abb0 from the virgin state with loading ter-
minating at an arbitrary point b0 . On unloading from b0 the path b0 Ba is followed.
When the material is loaded again the latter path is retraced as aBb0 , and if fur-
ther loading is then applied the path b0 c is followed, this being a continuation of
the primary loading path abb0 cc0 d (which is the path that would be traced if there
were no unloading). If loading is now stopped at c0 then the path c0 Ca is followed
on unloading and then retraced back to c0 on reloading. If no further loading beyond
c0 is applied then the curve aCc0 represents the subsequent material response, which
is then elastic. For loading beyond c0 , the primary path is again followed and the
pattern described is repeated. Clearly, there is stress softening on unloading relative
to the primary loading path, that is, the value of t on aBb0 or aCc0 is less than that
on abb0 cc0 for the same value of λ.
At the microscopic level, the stress softening is generally interpreted as being due
to damage caused by the loading, for example the severing of bonds between the filler
particles and the rubber molecular chains. Since the lengths of these chain links vary,
they break at different extensions and as the (macroscopic) deformation proceeds
the damage can be regarded as taking place continuously. Hence, stress softening
is evident on unloading from any point on the primary loading path, and damage
increases with the magnitude of the strain (or stress) on that path. When unloading
takes place, only part of the stress that was required to effect the damage is needed
to maintain any given submaximal strain and this is reflected retrospectively on the
unloading path and on the same path during reloading. An equivalent interpretation

Proc. R. Soc. Lond. A (1999)


Downloaded from http://rspa.royalsocietypublishing.org/ on April 8, 2016

A pseudo-elastic model for the Mullins effect 2867

is that the energy required to cause the damage is not recoverable, and the area
between, for example, the curves abb0 and aBb0 represents this energy for primary
loading up to b0 . These interpretations are embodied in the model discussed below.

(b) The model


Our aim is to model the Mullins effect in a general biaxial setting, embracing the
simple tension phenomenon described above as a special case. Consider the pseudo-
energy function W (λ1 , λ2 , η) appropriate for biaxial deformations, so that equations
(2.9) and (2.10) apply. When η is inactive, as indicated in § 2, it is set to a constant
value. Without loss of generality we set this constant to be unity and define the
function W̃ (λ1 , λ2 ) via

W̃ (λ1 , λ2 ) ≡ W (λ1 , λ2 , 1) (3.1)

in this case. This is the energy function of the perfectly elastic material for which
the primary loading path is also the unloading path. We therefore take W̃ (λ1 , λ2 )
to characterize any primary loading path in (λ1 , λ2 )-space, exemplified by the path
abb0 cc0 d in figure 1 in the simple tension specialization. Standard forms of the strain-
energy function, such as the neo-Hookean or Ogden forms (see, for example, Ogden
1984), may then be used as representative of W̃ (λ1 , λ2 ).
From (2.9) the specialization (3.1) yields the stresses

σ̃β − σ̃3 = λβ W̃ β (β = 1, 2), (3.2)

where a superposed tilde refers to a primary loading path and (2.10) is not operative.
We impose the usual requirements

W̃ (1, 1) = 0, W̃ β (1, 1) = 0 (β = 1, 2), (3.3)


W̃ 11 (1, 1) = W̃ 22 (1, 1) = 2W̃ 12 (1, 1) = 4µ (3.4)

on W̃ , where µ(> 0) is the shear modulus of the material in Ω and W̃ βγ denotes


∂ 2 W̃ /∂λβ ∂λγ (β, γ ∈ {1, 2}).
In addition to (3.3) and (3.4) we impose the key property of W̃ that it has a
global minimum of 0 at (1, 1) and that it has no other stationary points in (λ1 , λ2 )-
space. This is reasonable since the vanishing of both stresses in (3.2) (in the biaxial
deformation of a thin sheet with σ̃3 = 0, for example) is unlikely at any other point,
except possibly for pathological forms of W̃ .
It is therefore appropriate to define a primary loading path in (λ1 , λ2 )-space as
a path starting from (1, 1) on which W̃ is increasing. This point will be discussed
further in § 5 but we remark here that, for many standard forms of the strain-energy
function, W̃ increases along any straight-line path from (1, 1).
Unloading may take place from any point on a primary loading path, and the start
of unloading is taken as the signal for η to be activated. For unloading and subsequent
submaximal loading and unloading paths, the value of η then varies in accordance
with equation (2.10). We suppose that equation (2.10) can be solved explicitly for η
and write

η = χ(λ1 , λ2 ) = χ(λ2 , λ1 ). (3.5)

Proc. R. Soc. Lond. A (1999)


Downloaded from http://rspa.royalsocietypublishing.org/ on April 8, 2016

2868 R. W. Ogden and D. G. Roxburgh

Then, an energy function for unloading, symmetrical in (λ1 , λ2 ) and denoted by


w(λ1 , λ2 ), may be defined by
w(λ1 , λ2 ) ≡ W (λ1 , λ2 , χ(λ1 , λ2 )). (3.6)
From equations (2.9) and (2.10) it follows that
∂w ∂W
σβ − σ3 = λβ = λβ (β = 1, 2). (3.7)
∂λβ ∂λβ
Let (λ1m , λ2m ) be the values of (λ1 , λ2 ) at a point at which unloading begins.
Then, χ(λ1m , λ2m ) = 1. This implies that the function χ, and hence w, depends on
the point from which unloading starts. This dependence is made more explicit in
what follows.
We now specialize the form of constitutive law and propose the model
W (λ1 , λ2 , η) = η W̃ (λ1 , λ2 ) + φ(η), (3.8)
where φ(η), referred to as the damage function, is a smooth function of its argument
and, for consistency with (3.1), is subject to
φ(1) = 0. (3.9)
From (3.7) and (3.2) the stresses are calculated as
∂ W̃
σβ − σ3 = ηλβ = η(σ̃β − σ̃3 ) (β = 1, 2). (3.10)
∂λβ
The simple tension specialization of this, obtained on the basis of equations (2.11)
and (2.12), is, in terms of the Biot stress,

t = η Ŵλ (λ, 1) = η t̃, (3.11)

where t̃ is the Biot stress on the primary loading path at the same value of λ. For
(3.11) to predict stress softening it is clear that we must have η 6 1 on the unloading
path. We also take η > 0, so that t remains positive on unloading until λ = 1 is
reached.
On substitution of (3.8) into equation (2.10), we obtain
−φ0 (η) = W̃ (λ1 , λ2 ), (3.12)
which, implicitly, defines the damage parameter η in terms of the deformation. The
simple tension specialization of (3.12) is

−φ0 (η) = Ŵ (λ, 1). (3.13)


We emphasize that, in general, the value of η derived from (3.12) will depend on
the values of the principal stretches λ1m and λ2m attained on a primary loading
path, as well as on the specific forms of W̃ (λ1 , λ2 ) and φ(η) used. Since η = 1 at any
point on the primary loading path from which unloading is initiated, it follows from
equation (3.12) that
−φ0 (1) = W̃ (λ1m , λ2m ) ≡ Wm , (3.14)

Proc. R. Soc. Lond. A (1999)


Downloaded from http://rspa.royalsocietypublishing.org/ on April 8, 2016

A pseudo-elastic model for the Mullins effect 2869

wherein the notation Wm is defined. In accordance with the properties of W̃ , Wm


increases along a primary loading path.
On differentiation of (3.13) we obtain

−φ00 (η) = Ŵλ (λ, 1). (3.15)

If we associate unloading with decreasing η, for consistency with stress softening (see
figure 1), it follows from (3.15) that
φ00 (η) < 0, (3.16)
and we therefore assume that this inequality holds. It follows from (3.16) that φ0 (η)
is a monotonic decreasing function of η and hence that η is uniquely determined
from (3.12) as a function of W̃ (λ1 , λ2 ).
Note that on specializing (2.13) with respect to (3.8) we obtain
dt dt̃ t̃2
=η − 00 , (3.17)
dλ dλ φ (η)
and it follows from (3.16) and the assumption η > 0 that this is positive if
dt̃
≡ Ŵλλ (λ, 1) > 0. (3.18)

The inequality (3.18) ensures that the material response is stable in simple tension
on the primary loading path. With reference to (3.17) it then follows from (3.16) that
the unloading–reloading paths are also stable.
When the material is fully unloaded, with λ1 = λ2 = 1, η attains its minimum
value, ηm say. This is determined by inserting these values into equation (3.12) to
give, using the first equation in (3.3),
φ0 (ηm ) = −W̃ (1, 1) = 0. (3.19)
From (3.14) we may deduce that the function φ depends on the point where unloading
started and hence, from (3.19), so does ηm ; that is it depends, through Wm , on the
values of λ1m and λ2m . In view of the assumed continuity of η, when the material is
reloaded from any point on an unloading path its properties are again governed by
(3.6) until η reaches the value unity. At this point, either unloading is repeated or
further primary loading is initiated, governed by (3.1).

(c) Dissipation
When the damaged material is in a fully unloaded state the pseudo-energy function
(3.8) has the residual value
w(1, 1) = W (1, 1, ηm ) = φ(ηm ). (3.20)
Thus, the residual (non-recoverable) energy φ(ηm ) may be interpreted as a measure
of the energy required to cause the damage in the material. In a uniaxial test such as
simple tension, φ(ηm ) is the area between the primary loading curve and the relevant
unloading curve.

Proc. R. Soc. Lond. A (1999)


Downloaded from http://rspa.royalsocietypublishing.org/ on April 8, 2016

2870 R. W. Ogden and D. G. Roxburgh

We recall that in view of the condition (3.14) the function φ depends, through
Wm , on the point at which unloading starts. We accommodate this by defining the
function f , independent of Wm , by
φ0 (η) + Wm = φ0 (η) − φ0 (1) = f (η). (3.21)
Specification of the form of the function f is then equivalent to specifying the con-
stitutive function φ subject to (3.9), (3.14) and (3.16). Note that we must have
f (1) = 0, f (ηm ) = Wm , (3.22)
and therefore, by the monotonicity of φ0 and hence of f , the second equation in (3.22)
gives ηm uniquely in terms of Wm .
Integration of equation (3.21) with respect to η followed by rearrangement leads
to
Z η
φ(η) = f (η) dη + (1 − η)Wm ≡ Φ(η, Wm ), (3.23)
1

in which the function Φ is defined to reflect the dependence of φ on Wm . Thus,


φ(ηm ) = Φ(ηm , Wm ), (3.24)
which, through (3.22), gives φ(ηm ) in terms of Wm .
Since, from the definitions (3.23) and (3.19), we have ∂Φ(ηm , Wm )/∂η = φ0 (ηm ) =
0, we obtain, on differentiation of (3.23), the dissipation ‘rate’
dΦ ∂Φ
φ̇(ηm ) = (ηm , Wm )Ẇm = (ηm , Wm )Ẇm = (1 − ηm )Ẇm , (3.25)
dWm ∂Wm
where the superposed dot represents, for example, the time derivative or the deriva-
tive with respect to any deformation parameter that increases with primary loading.
Since we have ηm 6 1, with equality only at the beginning of the primary loading
process, and since the stored energy on a primary loading path must be increas-
ing, that is Ẇm > 0, we see from (3.25) that the dissipation rate is non-negative,
which is consistent with the second law of thermodynamics and the Clausius–Duhem
inequality.
We note finally in this section that Wm − φ(ηm ) is the recoverable part of the
energy and that this increases at the rate ηm Ẇm .

(d ) The damage function φ(η)


The function φ serves to determine the damage parameter in terms of the state
of deformation through equation (3.12). The choice of φ(η) is arbitrary subject to
(3.9), (3.14) and the inequality (3.16) with η satisfying 0 < η 6 1.
Let λ1m and λ2m be the values of the principal stretches corresponding to the
point at which unloading has most recently been initiated from a primary loading
path and W̃ (λ1m , λ2m ) = Wm . Then we choose φ(η) to be such that
−φ0 (η) = m erf −1 (r(η − 1)) + Wm , (3.26)

where m and r are positive parameters (material constants) and erf −1 (·) is the inverse
of the error function, which has properties particularly well suited to our aims. It is

Proc. R. Soc. Lond. A (1999)


Downloaded from http://rspa.royalsocietypublishing.org/ on April 8, 2016

A pseudo-elastic model for the Mullins effect 2871

not necessary to integrate this equation in order to satisfy (3.9) explicitly, but we
note that (3.14) follows immediately from (3.26). The properties of the error function
ensure that the inequality (3.16) holds. On substitution of equation (3.26) into (3.12)
and after a little algebra the expression
 
1 1
η = 1 − erf (Wm − W̃ (λ1 , λ2 )) (3.27)
r m
for η is obtained.
Since m and r were defined to be positive and, by definition, W̃ (λ1 , λ2 ) 6 Wm ,
it follows that η 6 1, with equality only when W̃ (λ1 , λ2 ) = Wm . Moreover, taking
r > 1 ensures that η is positive. The minimum value ηm of η, corresponding to the
undeformed configuration, is given by
 
1 Wm
ηm = 1 − erf , (3.28)
r m
and the dissipation rate (3.25) is given explicitly by
 
1 Wm
φ̇(ηm ) = erf Ẇm . (3.29)
r m
The parameters r and m have the following physical interpretations. The parame-
ter r is a measure of the extent of the damage relative to the virgin state. In particular,
from (3.27) it follows that the larger the value of r the less the damage parameter
η can depart from unity and the less damage can occur. Note that if r < 1 were
allowed then η, which enters as a multiplicative factor in the pseudo-energy func-
tion (3.8), and hence the stress, would vanish before the undeformed configuration
is recovered. This possibility has been ruled out in our considerations in this paper.
In contrast, the parameter m controls the dependence of the damage on the extent
of deformation. For small values of m, significant damage is caused for small strains,
and the material response in the small strain region is not markedly affected by fur-
ther primary loading. For larger values of m there is relatively little damage for small
strains but the material response changes significantly in the small strain region after
subsequent primary loading. Figure 2 illustrates these observations in respect of a
neo-Hookean material undergoing pure shear with r and m varied independently.
The dashed curve in each of the figures in the right-hand column is, for fixed r, the
asymptotic unloading–reloading curve for large primary deformation.

4. Numerical results
In this section we consider first the example of pure shear to highlight the theory
developed heretofore, in particular to show the relative behaviour of the main and side
stresses with respect to stress softening. The theory is then used to fit the (somewhat
limited) experimental data obtained by Mullins from simple tension tests.

(a) Pure shear


When σ3 = 0, equations (3.10) reduce to
∂ W̃ ∂ W̃
σ1 = ηλ1 , σ2 = ηλ2 . (4.1)
∂λ1 ∂λ2

Proc. R. Soc. Lond. A (1999)


Downloaded from http://rspa.royalsocietypublishing.org/ on April 8, 2016

2872 R. W. Ogden and D. G. Roxburgh

4
r = 1.2, m = 2.0 r = 2.0, m = 0.5
3

Biot stress (t)


2

0
4
r = 2.0, m = 2.0 r = 2.0, m = 2.0
3
Biot stress (t)

0
4
r = 3.0, m = 2.0 r = 2.0, m = 4.0

3
Biot stress (t)

0
1 2 3 4 1 2 3 4
stretch (λ) stretch (λ)
Figure 2. Plots of the main Biot stress t in pure shear against the corresponding stretch λ for
the neo-Hookean material, showing the effect of changing the parameters r (left-hand column)
and m (right-hand column).

In pure shear, equations (4.1) are specialized by setting λ1 = λ, λ2 = 1 and λ3 = λ−1 .


For a neo-Hookean material, defined by
W̃ (λ1 , λ2 ) = 12 µ(λ21 + λ22 + λ−2 −2
1 λ2 − 3), (4.2)
where µ is the shear modulus identified in (3.4), the pure shear specialization of (4.1)
yields
σ1 = λt1 = ηµ(λ2 − λ−2 ), σ2 = t2 = ηµ(1 − λ−2 ), (4.3)
t1 and t2 being the relevant principal Biot stresses.

Proc. R. Soc. Lond. A (1999)


Downloaded from http://rspa.royalsocietypublishing.org/ on April 8, 2016

A pseudo-elastic model for the Mullins effect 2873

4 3.5
(a) (b)
3
3
Biot stress (t) 2.5

Biot stress (t)


2
2
1.5

1 1
0.5
0 0
1 2 3 4 1 2 3 4
λ λ
Figure 3. Pure shear loading–unloading curves for (a) the neo-Hookean material and (b) the
Ogden material. The main Biot stresses t1 (continuous curves) and the side stresses t2 (broken
curves) are plotted against λ.

Similar expressions may be written down for the Ogden form of strain-energy
function
X3
µi αi
W̃ (λ1 , λ2 ) = µ (λ1 + λα −αi −αi
2 + λ1
i
λ2 − 3), (4.4)
i=1
αi

with material constants given by


α1 = 1.3, α2 = 5.0, α3 = −2.0,
µ1 = 1.491, µ2 = 0.003, µ3 = −0.0237
(see, for example, Ogden 1972, 1984), but they are not given explicitly here.
We now simulate a pure shear experiment in which the material is subjected to
a series of loadings with the sample returned to the undeformed configuration after
each loading and then reloaded beyond the previous maximum. We use the damage
model developed in § 3 together with (3.27) and m = r = 2.0. The resulting values of
the Biot stresses t1 and t2 for the constitutive equations (4.2) and (4.4) are displayed
in parts (a) and (b) of figure 3, respectively. In each case the solid curves correspond to
the main stress t1 , and the dashed curves to the side stresses t2 required to maintain
the pure shear deformation.
At present there is no comprehensive set of Mullins-effect data available for pure
shear with which to compare the theory. Some preliminary pure shear tests in which
both the main and side stresses are measured have recently been carried out by Julia
Gough at the Tun Abdul Razak Research Centre. The results are consistent with
the behaviour depicted in figure 3, but more data are needed before a meaningful
correlation with the theory can be conducted.

(b) Simple tension


In order to compare the theory with available experimental results, we consider
the special case of simple tension, for which λ1 = λ and λ2 = λ3 = λ−1/2 . With this
specialization, σ2 = 0, and the remaining non-vanishing principal stress is given by

Proc. R. Soc. Lond. A (1999)


Downloaded from http://rspa.royalsocietypublishing.org/ on April 8, 2016

2874 R. W. Ogden and D. G. Roxburgh

70
(a) (b)
60
50
Biot stress (t) 40
30
20
10
0
0 0.5 1 1.5 2 2.5 3 3.5 1 1.5 2 2.5 3 3.5 4 4.5
strain (ε ) λ
Figure 4. (a) Data obtained from Mullins & Tobin (1957); (b) best-fit curves from the model
developed here.

the first equation in (4.1). For the neo-Hookean material, we therefore have
σ1 = λt1 = ηµ(λ2 − λ−1 ), (4.5)
and similarly for (4.4).
In figure 4a we display data taken from Mullins & Tobin (1957, fig. 4) for a simple
tension experiment in which the stress t = t1 is plotted against the strain measure
 = λ − 1. All the measurements were taken during loading. It should be noted that
the material used by Mullins & Tobin (1957) exhibited permanent set on removal of
the stress, so that a given sample was longer after the initial stretch than it was in
its virgin state. This meant that after the initial loading the raw data curves did not
pass through the origin, but intercepted the horizontal axis at some positive value.
In their original paper Mullins & Tobin (1957) adjusted their data by subtracting
the permanent set from length measurements to ensure that the stress–strain curves
passed through the origin. Consequently, the data points on the reloading curves
have been shifted to the left of their true positions by the amount of the ‘set’ strain
(which is different for each reloading). There must therefore be some uncertainty
about the precise interpretation of the data.
We do not discuss the nature of the permanent set here nor the extent to which it
is recoverable, but clearly it affects the ability of the theory discussed here to fit the
data. We note that previous models designed to describe the Mullins effect do not
take account of the permanent set either. The theory discussed here will be adapted
to incorporate permanent set in a subsequent paper.
Figure 4b shows the result of a nonlinear least-squares best-fit solution to the data
obtained from Mullins & Tobin (1957) based on a strain-energy function of the form
(4.4). Here, t is plotted against λ rather than . First, a best fit was carried out for
the primary loading curve to obtain the appropriate values for the material constants
in (4.4). These values are
µ1 = 8.00, µ2 = 0.76, µ3 = −4.50,
α1 = 1.25, α2 = 4.0, α3 = −2.0.
With the primary loading curve fitted, it was then a simple matter to find suitable
values of the parameters r and m in (3.27) to fit the reloading curves. The values

Proc. R. Soc. Lond. A (1999)


Downloaded from http://rspa.royalsocietypublishing.org/ on April 8, 2016

A pseudo-elastic model for the Mullins effect 2875


5
(a) (b)
3
4

2 3
λ2
2
1
1

0 1 2 3 0 1 2 3 4 5
λ1 λ1

Figure 5. Constant energy contour curves in (λ1 , λ2 )-space for (a) the neo-Hookean
strain-energy function (4.2) and (b) the strain-energy function (5.2).

obtained are r = 2.104 and m = 22.45. The results are displayed as continuous curves
in figure 4b. The comparison between the theory and data is very good, especially
bearing in mind the reservations expressed above. Indeed, the fit of the theory to
the data shown in figure 4b is somewhat better than that achieved by Govindjee &
Simo (1992a), who, it should be noted, included only the second and third reloading
paths in their numerical simulation.

5. Discussion
In the examples of pure shear and simple tension considered above, the point at which
unloading began was identified by the value of the stretch at that point. However, it is
clear from the theory developed in § 3 that it is the associated energy on the primary
path, rather than the deformation, that is the key ingredient. In the case of general
biaxial deformations it is the value of the energy maximum Wm = W̃ (λ1m , λ2m )
on the primary loading path, rather than the specific (λ1m , λ2m ) pair, that governs
the unloading response. Thus, any other pair of (λ1 , λ2 ) values corresponding to the
same value of Wm could equally be taken as the starting point for unloading. The
collection of all such pairs satisfies the equation

W̃ (λ1 , λ2 ) = Wm , (5.1)

and, for given (constant) Wm , forms a closed contour in (λ1 , λ2 )-space. This is
depicted in figure 5a in respect of the neo-Hookean strain-energy function (4.2),
illustrated for Wm /µ = 1, 2, 3, 4.
The contour defined by the current maximum value Wm represents the current
damage threshold, and η = 1 at any point on this contour. For any deformation path
within this contour η < 1 and no further damage occurs. The energy required on
the initial loading path to cause the damage is not required on subsequent loading
up to the contour boundary. However, if the deformation path crosses the contour,
primary loading is again activated, η = 1 and further damage will occur. The value
of W̃ (λ1 , λ2 ) will increase until the next maximum value Wm is reached, at which

Proc. R. Soc. Lond. A (1999)


Downloaded from http://rspa.royalsocietypublishing.org/ on April 8, 2016

2876 R. W. Ogden and D. G. Roxburgh

point loading terminates and a new threshold contour is established (which encloses
the previous one).
Note that if σ3 = 0, then at any point (λ1 , λ2 ) on the contour (5.1) the vector
(t̃1 , t̃2 ), being the gradient of W̃ , is the outward pointing normal. For the incremental
deformation vector (δλ1 , δλ2 ), the corresponding increment in W̃ is given by the
scalar product (t̃1 , t̃2 ) · (δλ1 , δλ2 ). This is positive if (δλ1 , δλ2 ) points out of the
region enclosed by the contour (primary loading) and negative if it is inward pointing
(unloading).
Note that the regions enclosed by the contours shown in figure 5a are convex and
hence W̃ increases on any straight-line path from (1, 1), as intimated in § 3 b. This
is also the case for many commonly used forms of W̃ including those described by
(4.4) with µi αi > 0 and |αi | > 1 for each i. By contrast, we note in passing that
for a single-term form of (4.4) with |α1 | < 1 and Wm sufficiently large the contours
are not convex but, nevertheless, W̃ increases on any straight-line path from (1, 1).
The non-convexity is illustrated in figure 5b for α1 = 12 , that is, for the strain-energy
function
1/2 1/2 −1/2 −1/2
W̃ (λ1 , λ2 ) = 8µ(λ1 + λ2 + λ1 λ2 − 3). (5.2)
It is important to emphasize that for the model developed in this paper the extent
of the damage sustained by the material is controlled by the maximum energy state
Wm attained. We therefore have an energy-based damage model, as distinct from the
strain-based models of Johnson & Beatty (1993a, b) and Govindjee & Simo (1991,
1992a, b). For this reason the model is readily applicable to three-dimensional (multi-
axial) states of deformation. Specifically, for computational purposes, the value Wm
may be stored and the current energy state may be compared with it to determine
if further damage is being caused. Within a finite-element code this comparison can
be carried out on an element-by-element basis.
D.G.R. was supported by a research grant awarded by the UK Engineering and Physical Sciences
Research Council. The authors thank the Tun Abdul Razak Research Centre for their technical
advice and support.

References
Bueche, F. 1960 Molecular basis of the Mullins effect. J. Appl. Polymer Sci. 4, 107–114.
Bueche, F. 1961 Mullins effect and rubber–filler interaction. J. Appl. Polymer Sci. 5, 271–281.
Fung, Y. C. 1980 On the pseudo-elasticity of living tissues. In Mechanics today (ed. S. N. Nasser),
ch. 4, pp. 49–66. New York: Pergamon.
Govindjee, S. & Simo, J. C. 1991 A micro-mechanically based continuum damage model for
carbon black-filled rubbers incorporating the Mullins effect. J. Mech. Phys. Solids 39, 87–
112.
Govindjee, S. & Simo, J. C. 1992a Transition from micro-mechanics to computationally efficient
phenomenology: carbon black filled rubbers incorporating Mullins’ effect. J. Mech. Phys.
Solids 40, 213–233.
Govindjee, S. & Simo, J. C. 1992b Mullins’ effect and the strain amplitude dependence of the
storage modulus. Int. J. Solids Struct. 29, 1737–1751.
Harwood, J. A. C. & Payne, A. R. 1966a Stress softening in natural rubber vulcanizates. III.
Carbon black-filled vulcanizates. J. Appl. Polymer Sci. 10, 315–324.
Harwood, J. A. C. & Payne, A. R. 1966b Stress softening in natural rubber vulcanizates. IV.
Unfilled vulcanizates. J. Appl. Polymer Sci. 10, 1203–1211.

Proc. R. Soc. Lond. A (1999)


Downloaded from http://rspa.royalsocietypublishing.org/ on April 8, 2016

A pseudo-elastic model for the Mullins effect 2877

Harwood, J. A. C., Mullins, L. & Payne, A. R. 1965 Stress softening in natural rubber vulcan-
izates. II. Stress softening in pure gum and filler loaded rubbers. J. Appl. Polymer Sci. 9,
3011–3021.
Johnson, M. A. & Beatty, M. F. 1993a The Mullins effect in uniaxial extension and its influence
on the transverse vibration of a rubber string. Continuum Mech. Thermodyn. 5, 83–115.
Johnson, M. A. & Beatty, M. F. 1993b A constitutive equation for the Mullins effect in stress
controlled uniaxial extension experiments. Continuum Mech. Thermodyn. 5, 301–318.
Lazopoulos, K. A. & Ogden, R. W. 1998 Nonlinear elasticity theory with discontinuous internal
variables. Math. Mech. Solids 3, 29–51.
Lazopoulos, K. A. & Ogden, R. W. 1999 Spherically symmetric solutions for a spherical shell in
finite pseudo-elasticity. Int. J. Nonlinear Mech. (In the press.)
Lion, A. 1996 A constitutive model for carbon black filled rubber: experimental investigations
and mathematical representation. Continuum Mech. Thermodyn. 8, 153–169.
Miehe, C. 1995 Discontinuous and continuous damage evolution in Ogden-type large-strain
elastic materials. Eur. Jl Mech A 14, 697–720.
Mullins, L. 1947 Effect of stretching on the properties of rubber. J. Rubber Res. 16, 275–289.
Mullins, L. 1969 Softening of rubber by deformation. Rubber Chem. Technol. 42, 339–362.
Mullins, L. & Tobin, N. R. 1957 Theoretical model for the elastic behaviour of filler-reinforced
vulcanized rubbers. Rubber Chem. Technol. 30, 551–571.
Ogden, R. W. 1972 Large deformation isotropic elasticity: on the correlation of theory and
experiment for incompressible rubberlike solids. Proc. R. Soc. Lond. A 326, 565–584.
Ogden, R. W. 1982 Elastic deformations of rubberlike solids. In Mechanics of solids (ed. H. G.
Hopkins & M. J. Sewell), pp. 499–537. New York: Pergamon.
Ogden, R. W. 1984 Non-linear elastic deformations. Chichester: Ellis Horwood.
Ogden, R. W. 1986 Recent advances in the phenomenological theory of rubber elasticity. Rubber
Chem. Technol. 59, 361–383.
Rigbi, Z. 1980 Reinforcement of rubber by carbon black. Adv. Polymer Sci. 36, 21–68.
Simo, J. C. 1987 On a fully three-dimensional finite-strain viscoelastic damage model: formula-
tion and computational aspects. Comp. Meth. Appl. Mech. Engng 60, 153–173.
Simo, J. C. & Ju, J. W. 1987 Strain- and stress-based continuum damage models. I. Formulation.
Int. J. Solids Struct. 23, 821–840.

Proc. R. Soc. Lond. A (1999)


Downloaded from http://rspa.royalsocietypublishing.org/ on April 8, 2016

You might also like