Download as pdf or txt
Download as pdf or txt
You are on page 1of 17

DYNAMIC PROPERTIES OF FILLED RUBBER.

PART II: PHYSICAL BASIS


OF CONTRIBUTIONS TO THE MODEL

H. R. AHMADI, A. H. MUHR∗
TUN ABDUL RAZAK RESEARCH CENTRE (TARRC), HERTFORD SG13 8NL, UK

ABSTRACT
A relatively simple time-domain model is proposed with the scope to capture those aspects of the uniaxial stress–
strain behavior of filled rubber that are most significant in engineering applications, and is discussed in the context of a
literature review. Its performance is investigated in simple shear using analytical expressions. Attention has been given to
assembling the model from separate physical contributions, each already established in the literature, so that not only is the
number of parameters small but also they may be at least semi-quantitatively related to the formulation of the elastomer.
The small number of parameters helps to keep tests for fitting them simple, while their connection to mechanisms also
enables a degree of utility of the model even when extrapolated to situations beyond those covered by tests.

INTRODUCTION
The basic model for the stress–strain behavior of filled rubber, consisting of hyperelastic
(HE), viscoelastic (VE), and elastoplastic (EP) contributions in parallel (Figure 1), was de-
scribed in a previous paper,1 referred to hereafter as Part I. As explained in Part I, the aim
is to synthesize very different historical approaches to modeling rubber—quasi-static (hypere-
lastic), time-dependent (viscoelastic), and dynamic (equivalent visocelastic frequency-domain
characterization with amplitude dependence).
A fit to experimental data for filled styrene-butadience rubber (SBR) was given in Part I, and
it was also shown that the model is cast in a suitably general form for 3D implementation in finite
element analysis (FEA).More details of fits to other materials, implementation in FEA, and investi-
gation in multiaxial deformations will be given in further parts. The purpose of this paper is to pro-
vide analytical equations for applying the model to simple shear, showing how the EP contribution
captures the high modulus at low strain, imparted by reinforcing filler, in a way that can connect the
apparently different treatments given by authors from the quasi-static and dynamic approaches.
A review is used to derive the contributions to stress in filled rubber from their physical origins
and previous literature, hence suggesting how temperature and formulation affect the parameters.
Attention is restricted to simple shear, having the advantages that
1. the mode of deformation in many rubber components approximates to simple shear;
2. it is a convenient deformation for cyclic testing, having the advantage of symmetry about
the zero strain configuration and hence absence of kinematic sources of even harmonics
in the dynamic properties;2
3. since neither the cross-sectional area nor the rubber thickness changes in passing to finite
strain, there are no kinematic effects to complicate the calculation of shear stress and
shear strain from the measured force and deflection;
4. for the above reasons, a lot of shear stress–strain materials data, for checking the model
against, is available in the literature.
However, it is believed that characterization in any single mode of deformation, not neces-
sarily simple shear, provides an adequate foundation for fitting the model, and the nature of the
model will result in it being a good approximation in all modes of deformation.1, 3, 4
According to the model (Figure 1) the shear stress τ isgiven by

τ = τhe + τep + τve , (1)

∗ Corresponding author. Ph: +44 1992 584966 email: amuhr@tarrc.co.uk. [Paper 39 presented at the Fall 172nd Technical

Meeting of the Rubber Division of the American Chemical Society, Inc. (Cleveland, OH), 16–18 October 2007]

24
DYNAMIC PROPERTIES OF FILLED RUBBER. PART II 25

VE HE EP

FIG. 1. — Schematic diagram of the model; note that the actual model is 3D, and the VE and EP branches each
represent an array or spectrum of similar parallel elements.

where the subscripts he, ep, and ve stand for hyperelastic, elastoplastic, and viscoelastic respec-
tively.
It will be assumed below that the rubber has been suitably preconditioned, so that the scope
of the model does not include the Mullins effect, i.e., the semi-permanent stress-softening that
occurs during the primary loading. Incorporation of the Mullins effect into the model will be
addressed in Part V.

HYPERELASTIC CONTRIBUTION
The hyperelastic model is calculated from a strain energy function, which, it is believed,1, 3, 4
can capture the essential stress–strain behavior of unfilled crosslinked rubber if expressed as a
function of only the first and third strain invariants

I1 = λ21 + λ22 + λ23


(2)
I3 = λ21 λ22 λ23 ,

where λi is the extension ratio (i.e., current length of a material segment divided by the reference,
unstrained, length) in the ith principal direction.
It is more convenient to use the volumetrically neutralized first strain invariant:

−1/3
I¯1 = I1 I3 . (3)

Here I¯1 /3 represents the average factor by which the square of√the length of embedded line
segments, having random orientations, increase, whereas J = I3 is the dilatation, i.e., the
volumetric expansion ratio.
As suggested previously,4 the essential behavior for an unfilled rubber requires only two
parameters, the low-to-moderate strain shear modulus Ghe , denoted simply G, where there is no
ambiguity, and the low strain bulk modulus K. Nonlinearity associated with higher distortional
26 RUBBER CHEMISTRY AND TECHNOLOGY VOL. 84

or volumetric strains could be catered for by choice of the functions:


¯ 
G Im − I¯1 K
W = − ( I¯m − 3) ln + [ln(J )]2 . (4)
2 ¯Im − 3 2

In the first part of eq 4, due to Gent,5 the parameters G and I¯m are not independent, but may both
be related to the crosslink density of the polymer:

G = N kT, (5)

Nlinks k Nlinks 0.044


I¯m − 3 ≈ n = = ≈ , (6)
N G/T G/T

where N is the number of network chains per unit volume, Nlinks = Nn is the total number of
effective chain links in unit volume, n being the average number per chain, and the constant
0.044 MPa K−1 having been calculated from the data of Yeoh and Fleming.6 They changed the
concentration of sulfur and accelerator (CBS), keeping the ratio of these two ingredients fixed
at unity, to give four semi-EV NR vulcanizates with different crosslink densities. Note that the
statistical theory predicts proportionality of G with kT; in contrast, I¯m would be expected to
depend only on the network geometry and not on T. However, I¯m would in general depend on the
polymer as well as on the crosslink density, so the numerical coefficient in eq 6 should be adjusted
if the polymer is not NR. The bulk modulus K would be expected to decrease with increase in
temperature, since the free volume increases. It should be pointed out that Yeoh and Fleming6
found that while eq 4 can fit the experimental data at moderate to large strains, it was necessary
to introduce an additional term, with two parameters, to account for the extra high modulus of the
unfilled material at small strains. They tentatively attributed this enhanced low-strain stiffness to
entanglements between polymer chains.
Following Bergström and Boyce7 and Klüppel,8 the effect of volume fraction φ of a rigid
filler on the hyperelastic contribution may be included by modifying eq 4 in the following way:
 ¯ 
G ( Im − 3) − ( I¯1 − 3)X K /(1 − φ)
W = −(1 − φ) (Im − 3) ln + [ln (J )]2 , (7)
2 I¯m − 3 2

where the factor of (1 – φ) is the volume fraction of the rubber, being the phase in which all
strain energy is considered to be stored and to which the moduli G and K refer, and X is the
so-called strain amplification factor, accounting for the fact that only the polymer phase deforms
significantly, so that the strain in it exceeds the macroscopic value, i.e., that averaged over both
phases. The hyperelastic contribution to shear stress may be found from eq 7 using

∂W
τhe = 2γ , (8)
∂ I¯1

where γ is the shear strain, equal to I¯1 − 3, and the derivative with respect to the second strain
invariant has been neglected, since W is taken as dependent only on the first and third invariants.4
At moderate strains, eq 7 reduces to the neo-Hookean form, but with G replaced by
(1 − ϕ)X G, so that the simple shear stress–strain behavior plot [Eq 8] is linear, with a gradient of
(1 − ϕ)X G. The strain amplification factor X may be estimated from the “hydrodynamic” theory,
and knowledge of the shape of the filler particles. Unfortunately the theory has been developed
DYNAMIC PROPERTIES OF FILLED RUBBER. PART II 27

rigorously only as far as isolated ellipsoids9 and has been most thoroughly studied in the context
of enhancement of the viscosity of a fluid. Although the equations for particles embedded in an
elastic medium are in many ways analogous, the issue of orientation of the particles with the
flow lines, a preoccupation in the fluid viscosity literature,9–11 is quite different, and further work
seems to be needed to identify the correct equation for enhancement of elastic modulus of an
elastic medium in which ellipsoidal particles are embedded. For noninteracting spheres this issue
does not arise, and Einstein’s12 equation may be recast, for small strain and small φ, as

G app /G = (1 − φ)X = 1 + aφ
(9)
a = 2.5 for spheres,

where Gapp is the apparent shear modulus of the filled rubber, G being that of the polymer matrix
on its own. Efforts have been made to extend eq 9 to higher order terms to take into account
interaction between the perturbed stress fields in the matrix surrounding the particles, but it is
difficult to make good progress, as is also apparent from numerical analyses.7, 13 Brinkman14 has
suggested a simpler way forward: if we imagine the process of adding successive particles, each
one is added to a medium of higher viscosity, so application of eq 9 at each step results in a
differential equation which needs to be integrated to find the resultant (1 – φ)X. Krieger and Do-
herty (e.g., Krieger15 ) further pointed out that the process must deliver an infinite value of
(1 – φ)X at the value of φ such that the particles are as closely packed as possible in
space, termed the packing ratio, p. For spheres, for example, p = 0.64–0.78 depending on
random16 or close packing,17 respectively. Brinkman’s process may then be replaced by one
in which at each stage a particle plus (1 – p)/p times its volume of the matrix medium is
added, resulting in a differential equation which on integration yields the Krieger–Doherty
equation:15

 −ap
φ
(1 − φ)X = 1 − , (10)
p

where the value of a, the first order coefficient of eq 9, gives the limiting behavior at low φ and,
like p, depends on particle shape. In conclusion, (1 – φ)X may be taken to depend on the volume
fraction of filler and the two parameters a and p which depend on the particle geometry. These
parameters are, in general, difficult to derive mathematically but could be fitted empirically. The
volume fraction (1 – p) is akin to the “occluded rubber” concept of Medalia,18 and relates also to
the volume fraction of DBP absorbed by the filler at the point of formation of a paste, taken as a
measure of the “structure” of the filler aggregates.
Figure 2 shows the effect of temperature on the storage shear modulus for a series of natural
rubber-based materials, including unfilled. The formulations differ only in content of N330, for
each ten parts of which one part of process oil was also included; further details are available in
the data source.19
It is apparent that the storage shear modulus at 0.1 Hz is not simply proportional to absolute
temperature, but even for the simplest case, unfilled NR, shows more complex behavior. Ideally
our model should predict the temperature dependence, without introducing extra stress–strain
parameters, so care will be taken in the following sections to examine how the parameters would
be expected to vary with temperature.
28 RUBBER CHEMISTRY AND TECHNOLOGY VOL. 84

1.5
EDS19 unfilled NR
1.4 EDS14 NR+15pphrN330
EDS15 NR+30N330
1.3
EDS16 NR+45pphrN330
1.2 G'(23)*(T+273)/(23+273) for EDS19

1.1
G' /MPa

0.9

0.8

0.7

0.6

0.5

0.4
-40 -20 0 20 40 60 80 100 120
T /degC
FIG. 2. — Effect of temperature and N330 content on large strain (50%) storage shear modulus of natural rubber at
0.1 Hz.

ELASTOPLASTIC CONTRIBUTION
It is well known that, in addition to the general stiffening effect of all fillers on the rubber
discussed in the previous section reinforcing fillers impart a thixotropic effect, whereby small
changes in the state of strain are associated with an enhanced stiffness, and changes in the
direction of straining are accompanied by hysteresis.20 The effect correlates with the specific
surface area of the filler, and is also influenced by the strength of the rubber-filler interaction and
the dispersion of the filler particles.
Many authors have modeled the effect using an elastoplastic or triboelastic contribution to
the stress, and such an approach was adopted in part I. We start here with a more general scheme
than presented in part I,1 in order to bring out the connection to the earlier literature, particularly
that related to Kraus’s21 work on dynamic properties.
The elastoplastic stress contribution in simple shear is considered to be governed by two
smooth continuous curves, illustrated in Figure 3 and described in a generic way in eqs 11 and
12.

THE LOADING CURVE τ epL

The loading of the preconditioned material from a stress and strain free state is given by a
function:

dτepL d 2 τepL
τep = τepL (γ ) : for γ > 0,τepL ≥ 0; ≥ 0 and → 0 as γ → ∞; ≤ 0, (11)
dγ dγ 2

where L stands for loading. For loading in the negative direction τ epL (–γ ) = –τ epL (γ ). The
positive first derivative ensures that the elastoplastic contribution always opposes change in strain
DYNAMIC PROPERTIES OF FILLED RUBBER. PART II 29

0.5

stress
0.3
Disallowed by
continued τepL
loading rule
0.1
strain

-1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1


-0.1
ΔτepR

-0.3
Disallowed by
continued
loading rule
-0.5

FIG. 3. — Schematic diagram showing hysteresis loop for an elastoplastic material with the loading curve given in
eq 13 [with A = 0.5, b = 0.6, and c = 0.005].

so that the behavior is stable and, with the retraction behavior next described, dissipative. The
negative or zero curvature, together with the asymptotic limit of the gradient at infinite strain, is
intended to ensure a monotonic decline in the contribution to the storage modulus as the strain
amplitude is increased, in accord with experimental data at small and moderate strains. Any
upturn at very high strains due to finite chain extensibility can be modeled using the hyperelastic
part. Upturn and hysteresis due to strain crystallization at high strain, e.g., for NR and CR, could
not be accommodated by this model. An upturn in τ epL would, according to the retraction rule
given below, be associated with a decrease in hysteresis increment – even to a negative dissipation
if the tangent of τ epL at high strain exceeds that at zero strain.

THE RETRACTION CURVE τ epR

The change in stress τ epR resulting from a strain increment γ following a change in
direction of straining is calculated from an incremental relationship:
 

τ ≡ τepR (γ ) = 2τepL , (12)
2

where R stands for retraction. Equation 12 is valid until the next reversal in direction of straining,
when the increments are reset to zero provided the stress-strain trajectory does not pass through a
previous point of retraction, termed a turning point; eq 12 ensures that closed loops are generated
for all strain cycles but is not necessarily valid beyond such a cycle.

THE CONTINUED LOADING RULE

To avoid “ratcheting” of the stress magnitude upwards by a small retraction followed by


continued loading, the stress–strain coordinates of each turning point are logged together with the
maximum and minimum strains from which unloading and loading trajectories respectively have
30 RUBBER CHEMISTRY AND TECHNOLOGY VOL. 84

approached it. In the event that the trajectory returns to the turning point and proceeds without
retraction, the strain increment in eq 12 is redefined to be relative to the maximum or minimum
logged strain for continued unloading or loading respectively. In the special case of a trajectory
returning to the primary loading curve, the continued loading trajectory defaults to eq 11.
It is easily confirmed that if such a trajectory initiates from a turning point on the primary
loading curve, then its gradient is equal to that of the primary loading curve at the meeting
point and the stress and strain values of meeting and turning points are equal in magnitude
and opposite in sign. It follows from the above rules that the elastoplastic modulus contribution
will be independent of static strain, as illustrated in Figure 3 and in reasonable accord with
experiment.
In Part I1 the loading function was taken to be
 2 −b/2
γ
τepL (γ ) = Aγ (γ 2 + c2 )−b/2 ≡ G
ep (0)γ + 1 (13)
c2

where G
ep (0) ≡ Ac−b is the elastoplastic shear modulus contribution in the small strain limit and
0 < b < 1 to ensure a monotonically increasing loading function with monotonically decreasing
gradient as proposed by Davies et al.22 for fitting the high stiffness, taken in their case to be
fully elastic, at low strain, and the rapid fall with increasing strain, characteristic of rubbers with
reinforcing filler. However, alternative functional forms may be used. The model proposed by
Coveney et al.34 is equivalent to τ epL ∝ (αγ + 1)0.5 – 1, which they showed arises in the limit
of a large number of identical frictional sliders and springs connected in series. However, their
model would need more parameters to provide a phenomenological fit to experimental data. The
Kraus equation,21 designed to fit the strain-dependence of the dynamic modulus, can be used to
construct an equivalent elastoplastic loading curve. To see how this may be done, the extraction
of the equivalent linearized viscoelastic dynamic storage G
and loss G

moduli will be discussed.


The harmonic method may be used to express G
and G

in terms of Fourier integrals of the


stress response to a forced sinusoidal strain history:23
γ = γ̃ sin ωt

ω
G
(ω) ≡ τ (t) sin(ωt)dt (14)
π γ̃

ω
G

(ω) ≡ τ (t) cos(ωt)dt.


π γ̃

As the Fourier modulus functionals are linear it follows that


G
= G
he + G
ve + G
ep
(15)
G

= G

he + G

ve + G

ep .

Because the viscoelastic stress will in general be out of phase with the rate-independent hy-
perelastic and elastoplastic contributions, this advantage of additivity of the dynamic modulus
contributions does not strictly apply for alternative methods of defining an equivalent linear
viscoelastic storage modulus—being the secant and skeleton methods.23 However, provided the
viscoelastic contribution is small, Equation 15 will be a close approximation for the secant method

|G ∗ | ≡ τpeak /γ̃
(16)
sin(δ) ≡ W L /(π γ̃ 2 |G ∗ |).
DYNAMIC PROPERTIES OF FILLED RUBBER. PART II 31

Similarly, for the kind of elastoplastic contribution envisaged, with τ ep a decreasing function of
γ , the same should apply to an estimate of loss modulus from the stress at zero strain:

ep ≈ τep (0)/γ̃ . (17)

Equations 16 and 17 are convenient because they enable simple analytical approximations
for the moduli to be derived from candidate analytical expressions for eqs 11 describing the
elastoplastic contribution, such as eq 13. Experimental determinations of the moduli usually
use the harmonic method, so this is compared below to the simple approximations, eqs 16
and 17.
From eq 13 the elastoplastic contribution to the secant shear modulus, equal to the contribution
to the dynamic storage modulus because there is no rate dependence in the elastoplastic model,
is given approximately by

τepL
G
ep (γ̃ ) ≈ G ∗ep (γ̃ ) = = A(γ̃ 2 + c2 )−b/2 . (18)
γ̃

Differentiating eq 18 with respect to γ̃ reveals that −dG


ep /d γ̃ and −dG
ep /d(ln γ̃ ) have maxima
√ √
at strains given by γ̃ /c ≈ 1/(1 + b) and 2/b respectively.
Since only the contribution G
ep to G / depends on strain it follows that R, the ratio of the
excess of storage modulus at strain amplitude γ̃ (over that at high strain) to the maximum value
of the excess is given by
  2 −b/2
G
(γ̃ ) − G
(∞) G
ep (γ̃ ) γ̃
R(γ̃ ) ≡


≈ 1+ . (19)
G (0) − G (∞) G ep (0) c

Note that R(0) = 1 by definition. Equation 19 now has a similar form to that of Kraus:21
  2m −1
G
− G
∞ γ̃
R(γ̃ ) ≡

= 1 + (20a)
G0 − G∞ γc

suggesting the loading function:


  2m −1
γ
τepL ≈ G
ep (0)γ 1+ for Kraus-type behavior. (20b)
γc

Equations 19 and 20a describe similar behavior. Choosing parameters in such a way that they
both give R(γc ) = 1/2 and equal values for R (1), leads to

 2 1
γc = 2 b − 1 2 c
(21)
 1 b 
− log c2
+1 2 −1
m= .
2 log(γc )

Hysteresis loops generated from eqs 11 and 12 with either 13 or 20b for the loading curve, using
comparable parameters according to eq 21, are shown in Figures 4 and 5. Equations 19 and 20a,
with parameters from eq 21, are compared in Figure 6. The level of agreement may be taken as
32 RUBBER CHEMISTRY AND TECHNOLOGY VOL. 84

0.05

0.04

0.03

0.02
shear stress

0.01

0
-0.1 -0.08 -0.06 -0.04 -0.02 0 0.02 0.04 0.06 0.08 0.1
-0.01

-0.02

-0.03

-0.04

-0.05
shear strain

FIG. 4. — Elastoplastic hysteresis loops at small strains. ——— Equation 12 (based on Davies et al.22 ), A = 0.0746, b
= 0.8, c = 0.04; · · · · · eq 20b (based on Kraus21 ), 2m = 1.016, γ c = 0.086.

satisfactory for the current purposes, although future work may indicate an advantage in one or
the other form.
Turning attention to G

, the elastoplastic contribution may be calculated from integration


of eqs 12 and 13 or 20b, according to the secant or harmonic method for example.23 A quick

0.08

0.06

0.04

0.02
shear stress

0
-1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1
-0.02

-0.04

-0.06

-0.08
shear strain

FIG. 5. — Elastoplastic hysteresis loops at large strains. Same key and equations as Figure 4.
DYNAMIC PROPERTIES OF FILLED RUBBER. PART II 33

0.9

0.8
R storage modulus ratio

yc=0.086, 2m=1.018 0.7


b=0.8, c=0.04
0.6
yc=0.015, 2m=0.746
0.5
b=0.6, c=0.005
0.4
b=0.8; c=0.04
0.3
b=0.6; c=0.005
0.2
yc=0.086, 2m=1.018

yc=0.015, 2m=0.746 0.1

0
0.0001 0.001 0.01 0.1 1 10
shear strain amplitude

FIG. 6. — Comparison of amplitude dependencies of storage modulus for candidate ep functions. Lines—harmonic
linearization (eqs 14); points, approximate secant linearization (eq 16). Full lines and round symbols—eqs 13 and 19
based on Davies et al.22 Dashed lines and square symbols—eq 20a based on Kraus.21

estimate can, however, be made from eq 17 with eq 13 or 20b:


 
γ̃
γ̃ .G

ep ≈ τep (0) = −τepL (γ̃ ) + 2τepL (22)


2

⎧ − b2  ⎫
⎨ 2 − b2 ⎬
τep γ̃ γ̃
G

ep ≈ = G
ep (0) 1+ 2 − 1+ 2 fit to Davies et al., (23)
γ̃ ⎩ 4c c ⎭

which is readily shown to reach a maximum value at γ̃ /c = 2 (λ − 1)/(4 − λ) where
λ = 41/(1+b/2) , and
    
τep γ̃ 2m γ̃ 2m
G

ep ≈ ≡ G
ep (0) 1+ − 1+ fit to Kraus. (24)
γ̃ 2γc γc

Figure 7 shows a plot of these predictions, for the same parameters as given in Figure 6. The
approximations (points) agree fairly well with the harmonic integration method (lines) except at
strains in the region of the maximum in loss modulus, as might be expected. Taking Figures 6
and 7 together, it is apparent that the “Kraus” loading function differs primarily from the “Davies
et al.” loading function in giving higher energy loss and sensitivity of dynamic properties to
changes in amplitude at very small strains. Detailed modeling at such strains is not of primary
interest in many applications.
The key point is that the amplitude dependence of both the storage modulus (Figure 6) and
the loss modulus (Figure 7) is derived from the same set of hysteresis loops generated by the
model, and hence from the application of the elastoplastic model to a single equation for the
constitutive properties (eqs 13 or 20b).
34 RUBBER CHEMISTRY AND TECHNOLOGY VOL. 84

0.25
Eq23 c=0.04; b=0.8
Eq 23 c=0.005; b=0.6
Eq 24 yc=0.015; 2m=0.746 0.2
Eq 24 yc=0.086; 2m=1.018
c=0.04; b=0.8
G'' /G'ep(0)

0.15
c=0.005; b=0.6
yc=0.086; 2m=1.018
yc= 0.015; 2m=0.746 0.1

0.05

0
0.00001 0.0001 0.001 0.01 0.1 1 10
shear strain amplitude

FIG. 7. — Comparison of amplitude dependencies of storage modulus for candidate ep functions. Lines—harmonic
linearization (eq 14); points, approximate secant linearization (eq 22). Full lines and round symbols—eq 13 based on
Davies et al.,22 dashed lines and square symbols—eq 20b based on Kraus.21

Kraus21 had noticed a similar connection between the amplitude dependences of storage
and loss moduli, and proposed a mathematical connection based on his agglomeration theory of
reinforcement. Although this predicted the right shape for the amplitude dependence of the loss
modulus from that of the storage modulus, he had to introduce a new parameter for the magnitude
scale.21 It is important to note that the elastoplastic model predicts a direct connection, without
introduction of a new parameter.
The linkage of the elastoplastic contribution with the Kraus equation enables the considerable
body of experimental work on the latter to be carried over into the present model. It would thus
be expected that the temperature dependence would be given by the Arrhenius equation:

 
E
G ep ∝ exp , (25)
RT

where E is an activation energy and R is the gas constant. Equation 25 follows from the notion
that the breakdown of the “structure” responsible for the extra reinforcing effect is an activated
process, and results in a reduction in the contribution to the storage modulus. In other words,
Brownian motion tends to disrupt the “jamming” of the reinforcing particle population—i.e.,
their ability to provide load paths spanning the system.
Klüppel8 gives an alternative explanation of the filler networking, believed to be responsible
for the amplitude effect, to that of Kraus.21 Klüppel and Meier24 suggest the apparent elastoplastic
contribution arises from the bending stiffness of linkages in the network formed by the filler
clusters, and this in turn arises from the tightly bound polymer forming a “glue” between the filler
particles. This material is relatively immobile, because of the constraining influence of the bonds
to the filler surface, thus having a higher effective Tg than the polymer in the absence of the filler,
and softening as the temperature is raised. We return to these considerations in the Discussion.
DYNAMIC PROPERTIES OF FILLED RUBBER. PART II 35

VISCOELASTIC CONTRIBUTION
Generally, even well above their Tg , polymers are found to have a significantly higher shear
modulus than given by eq 5, and to exhibit creep, time-dependent recovery after subjection to
stresses and strains, and stress relaxation. This behavior is believed to arise from time-dependent
effective crosslinks, typically being entanglements of the polymer chains. Such effects were
included in part 1 by adding to the hyperelastic contribution, corresponding to a shear modulus
of NkT, a spectrum of Maxwell elements such that the relaxation modulus is linear with log
(t). The temperature dependence of such a viscoelastic contribution is, in general, a function of
details of the polymer in question, although it should be related to the frequency dependence
through the WLF equation.25 Thus the viscoelastic model of part I1 will give rise to a prediction
of a particular temperature dependence of the shear modulus. The viscoelastic contribution is
captured by a relaxation shear modulus, which may be written in the form of an integral over the
full range of relaxation times τ

∞
h(τ )e− λ dτ
t
G ve (t) = ρT (26)
0

where the dependence on density ρ and absolute temperature T has been made explicit, following
Ferry.26 We assume here that G ve (∞) = 0, since the rubbery plateau has been assigned to Ghe .
The dynamic properties may be calculated from the relaxation modulus using Boltzmann’s
superposition principle27 :

t
τve (t) = G ve (t − s)γ̇ (s)ds, (27)
0

whence (e.g., Ferry26 )

∞
G
ve (ω) = ω G ve (s) sin(ωs)ds (28)
0

∞
G

ve (ω) =ω G ve (s) cos(ωs)ds (29)


0

According to Boltzmann’s superposition principle,27 we can calculate all viscoelastic properties


from knowledge of the relaxation modulus G(t) over a broad enough time range. If also we
know the temperature dependence of G(t) then we can predict viscoelastic properties at any
temperature and frequency or time using the time-temperature correspondence principle. Ferry26
(Ch.11) asserts that a suitable form for G(t, T) is
   
ρT t
G(t, T ) = G ∞ (Tref ) + G i (Tref ) exp − , (30)
ρref Tref aT τi

i.e., G ∞ and the G i are proportional to absolute temperature T and all the relaxation times τi are
subject to the same temperature dependent shift factor aT , which could for example be estimated
36 RUBBER CHEMISTRY AND TECHNOLOGY VOL. 84

from the empirical WLF25 equation

8.86(T − Ts )
log10 (aT ) ≈ −
101.6 + T − Ts
(31)
Ts ≈ Tg + 50K,

where Tg is the glass transition temperature. If Tref is some value other than Ts , then aT in equation
30 needs to be replaced by aT /aTref .
It follows from equation 30 and the Boltzmann superposition principle that the storage and
loss shear moduli are given by
  

ρT (ωaT τi )2
G (ω) = G∞ + Gi (32)
ρref Tref 1 + (ωaT τi )2

ρT  ωaT τi
G

(ω) = Gi . (33)
ρref Tref 1 + (ωaT τi )2

In Part I1 we proposed that the integral in eq 26 be limited to the time range of experimental
interest, τ min < t < τ max , where τ min and τ max are taken as the lower and upper bounds of the
integral respectively. Contributions to Gve for longer relaxation times would in effect be added
to Ghe . Contributions to Gve for shorter relaxation times would never be noticed, since they
would relax away during loading; however, if the temperature is lowered they may contribute to
noticeable stiffening of the material, since, by the time-temperature equivalence principle, their
effective time constants would be increased into the experimentally observed time range. This
could also obscure the proportionality with absolute temperature of the apparent Ghe expected
from eq 5 and not seen in the experimental results of Figure 2.
The temperature dependence made explicit in eq 30 is in fact not the primary effect of
temperature. The relaxation times τ i appearing in the distribution function are strongly influenced
through the shift factor aT . Thus on changing the temperature, log(G ve ) will not only be shifted
vertically on a log(G ve ) versus log(t) plot, but the whole distribution function will be shifted to
shorter times (for a rise in temperature) or to longer times (for a decrease in temperature). The
physical relaxation rate, expressed as percent per tenfold increase in time, should decrease with
increase in temperature, assuming the material is well above its Tg . This can be appreciated by
considering the shift, on the log time axis, of the relaxation modulus master curve to the left if
the temperature is increased, and to the right if it is decreased (Figure 8). Provided the initial
conditions of time and temperature are such that the relaxation behavior is well towards the
rubbery region, where the curvature of Gve (t) versus log(t) tends to be concave upwards, lowering
temperature will increase the relaxation rate (and the loss tangent) and raising the temperature will
decrease it. This might seem paradoxical, since raising temperature usually results in increased
rates, but in fact such higher rates, and hence more relaxation, have indeed occurred at the higher
temperature, and hence Gve (t) will already have fallen further along the master curve and so the
rate is lower, if the comparison be made at the same elapsed time from loading.
In part I it was suggested that the distribution H (τ ) ≡ ρT.h(τ ) of relaxation times be taken
as a constant H0 for τ min < t < τ max , and zero otherwise. This assumption will probably need to
be modified, at the expense of one or two more parameters needed to introduce curvature into
the relaxation plot versus log(t), if it is required to include the effect of temperature or a very
extended range of rates on the dynamic properties.
DYNAMIC PROPERTIES OF FILLED RUBBER. PART II 37

G(t) decreasing
stress
relaxation
rate

increasing
temp

Log (t)

FIG. 8. — Schematic diagram of stress relaxation master curves at three reference temperatures, showing how the
gradient (i.e., the stress relaxation rate) at a given elapsed time after loading depends on the test temperature. Vertical
shifts do not affect the gradient, and are ignored here.

DISCUSSION
The enhancement of stiffness imparted by reinforcing fillers, particularly at low strain, cannot
be reconciled with hydrodynamic theory unless (i) the proximity of the filler surface somehow
stiffens a substantial volume fraction of the rubber matrix or (ii) the filler particles somehow
cooperate in forming a load-bearing structure with a shape that permeates the rubber matrix
much more effectively than the independent filler aggregates would, or “occludes”, i.e., shields,
a substantial volume fraction of the rubber from bearing stress, which amounts to much the
same thing. Reconciliation would require effective filler elements of very high elongation, or
low packing volume (Equation 10), or, if interpreted on the basis of a Takayanagi28 model, a
geometry conducive of the load path passing primarily through the rigid filler particles. This
load bearing mechanism was thought of as a “structure” by Payne and Watson,29 which could
break down if subjected to sufficient load, accounting for the increase in mechanical hysteresis
and the softening as the amplitude is increased. However, as the stress–strain behavior is largely
recoverable on returning to low strains, there must be a mechanism for the “structure” to reform.
Extensive literature exists aiming to elucidate this mechanism, and the nature of the “structure.”
Although filler can form a structure in other media, such as oil or even gas,29 the persistence
of its load-bearing capacity to relatively high strains is characteristic only of filled rubber. This
suggests that the filler structure in reinforced rubber consists not just of filler, but incorporates
some polymer too, despite the suggestion to the contrary by Payne and Watson.29 Corroborating
evidence would be as follows.
(i) The profound effect of surface activity on reinforcement by fillers. For example, de-
creasing the filler—filler interaction of silica by a chemical coupling agent reduces the
reinforcement, but increasing the coupling to the polymer through covalent bonds partially
restores the reinforcement.30
(ii) The time-dependence of “structure” recovery from breakdown. It is well known that
recovery, although rapid, is not instantaneous. The storage modulus typically recovers a
substantial way towards the low strain value in the first measurement made after subjection
to a large amplitude, but then slowly recovers towards the initial value over many cycles
at the low strain.31 Similarly, Sternstein et al.32 found that the reinforcing effect itself is
time-dependent and suggested it extrapolates toward zero at a sufficiently low frequency
(∼10−5 Hz for a PVA material).
38 RUBBER CHEMISTRY AND TECHNOLOGY VOL. 84

(iii) The effect of temperature on reinforcement. Equation 25 is suggestive of a role for a fluid
or polymer rather than of a solid with high thermal stability.
(iv) The effect of polymer on reinforcement. The polarity of the polymer affects its interaction
with filler particles, with effects parallel to those described in (i); in addition, Zhu and
Sternstein31 found a pronounced effect of polymer molecular weight on the reinforcement.
It thus appears that viscoelasticity of the polymer directly influences the reinforcement, in
addition to affecting the underlying viscoelasticity of the filled rubber through that of the matrix.30
This effect of the polymer on the rate and temperature sensitivities of reinforcement appears not
to reflect the properties of the bulk polymer matrix. Instead, the properties of the bound polymer
film could be modified8 by the filler proximity to have a raised Tg or modified viscoelasticity due
to trapped entanglements.31
The picture thus emerges of a geometrically complicated and labile structure of filler particles
glued together by a polymeric film, the viscoelastic properties of which are unknown so can only
be inferred indirectly. The detailed behavior of this structure for representative polymer-filler
systems remains a very active area in the contemporary literature, so it is unlikely that a fully
predictive micromechanical theory for the stress–strain behavior will emerge for some time to
come. For engineering purposes it is sensible therefore to adopt a phenomenological approach to
describe the consequent behavior of filled rubber, whether we term it triboelastic,33, 34 equilibrium
hysteresis,35 jamming36 or simply viscoplastic.37 Such terminology conveys the broad nature of
the phenomenon and connects to a broader literature wherein might be found some insight and
tools for mathematical description, but as yet no practical quantitative micromechanical model.
The approach adopted for our purposes, being to provide a practical 3D model that can
be easily implemented in FEA packages,1, 38 is to treat the effect of fillers on the stress–strain
behavior as independent of rate. This effect is considered to have two parts: an elastic stiffening
effect (e.g., eq 10) due to the inclusion of rigid particles in the matrix, and a stiffening and
hysteretic effect (rules 1 to 3, and eq 13 for example) associated particularly with reinforcing
fillers. Alternatively, the parameters a and p in equation 10 could have been considered to be
functions of strain history but such an approach would not be amenable to using material models
already extant in FEA packages. Thus in addition to the volume fraction φ of filler, a set of five
other parameters—one choice being for example a, p, A, b, c—have been raised to describe its
effect on stress–strain behavior. The contribution to elastic stiffening is imparted by a and p,
which are not strictly independent but in principle are both derivable from an assumption of the
geometry of primary filler aggregates, i.e., those aggregates that do not break down on straining.
The magnitude of the hysteretic contribution to stiffening is governed by A, which, for a given
filler, would be expected to increase with the loading φ. Parameters b, c relate to the effectiveness
of strain in breaking down the filler structure, and like a, p would be expected to be sensitive to
the relative strength of filler-filler and filler-polymer interactions and to the filler structure, as will
also be the relative magnitudes of the elastic and hysteretic contributions to stiffening, imparted
by a and by A respectively. A simple default model might assume the elastic stiffness contribution
is imparted by spherical particles, so that a = 2.5 and p ≈ 0.7, that A is a function of the filler
loading and b, c are functions of the filler type, the coupling, and the dispersion. Finally, the
activation energy E is needed to model the effect of temperature on the elastoplastic contribution.
As advances in description of the material stress–strain behavior and in modeling techniques
are made, it would be expected that models capturing the behavior in more detail can be built and
would serve for a practical and more refined material simulation. Earlier, we proposed a model
for simple shear of rubber that used more sophisticated rules for retraction, designed to achieve
continuity of tangential stiffness for monotonic loading,23 but it was not clear how to generalize
it to multiaxial deformations. The phenomenological model of Lion et al.39 may well have
suitable versatility to capture the behavior for general deformation. However, the view followed
DYNAMIC PROPERTIES OF FILLED RUBBER. PART II 39

in this series of papers is to explore the successes and limitations of modeling the reinforcing
effect through use of rate-independent plasticity (part III38 ) for the filler contribution, and the
development of eq 25 for its temperature sensitivity and of finite-strain linear viscoelasticity for the
matrix contribution (part IV40 ). It is acknowledged that neglect of the sensitivity of reinforcement
to rate will mean that parameters relating to it will need to be re-fitted if application of the model
to very different timescales is required.
Quasi-irreversible stress softening (Mullins effect or damage) and the stress–strain upturn
and hysteresis seen at very high strain for strain-crystallizing polymers are left for future work.

CONCLUSIONS
Based on the hypothesis of additive hyperelastic, elastoplastic and viscoelastic contributions
to the stress, equations have been collated, and, in the case of the elastoplastic contribution,
provided, which describe the stress–strain behavior of filled rubber. Evidence that the model
is consistent with 3D application in FEA is given in other parts for this series of papers.1, 38
Excluding the effect of temperature, the minimum number of parameters needed to describe the
stress–strain behavior in shear (or any other deviatoric deformation) is five: two for the finite-strain
viscoelasticity of the rubber matrix and three for the hysteretic stiffening imparted by the filler.
Phenomena captured are: quasi-static stress–strain behavior, Kraus-type amplitude dependence
of the complex modulus, and creep, relaxation and recovery. It is apparent form the equations
and rules proposed here for the elastoplastic contribution that the small strain dynamic modulus
will be substantially independent of static prestrain, in accord with experiment.
Expected dependency of the parameters with filler loading and temperature has been dis-
cussed. It is suggested that capturing the effect of temperature will require extra parameters, in
particular an activation temperature for the reinforcing effect of filler, and parameters describing
the curvature of the plot of relaxation modulus versus log(t).

REFERENCES
1 H. R. Ahmadi, J. G. R. Kingston, and A. H. Muhr, RUBBER CHEM. TECHNOL. 81, 1 (2008).
2 J. A. Harris, RUBBER CHEM. TECHNOL. 60, 870 (1987).
3 J. Gough, I. H. Gregory, and A. H. Muhr; “Determination of Constitutive Equations for Vulcanised Rubber,” in “Finite
Element Analysis of Elastomers,” D. Boast and V. A. Coveney, Eds., Professional Engineering Publishing, London,
1999.
4 A. H. Muhr, RUBBER CHEM. TECHNOL. 78, 391 (2005).
5 A. N. Gent, RUBBER CHEM. TECHNOL. 69, 59 (1996).
6 O. H. Yeoh and P. D. Fleming, J. Polym. Sci. B35, 1919 (1997).
7 J. S. Bergström and M. C. Boyce, RUBBER CHEM. TECHNOL. 72, 633 (1999).
8 M. Klüppel, Adv. Polym. Sci. 164, 1 (2003).
9 G. B. Jeffery, Proc. Roy. Soc. (London) A102, 161 (1923).
10 G. I. Taylor, Proc. Roy. Soc. (London) A138, 41 (1932).
11 G. K. Batchelor, J. Fluid. Mech. 83, 1, 97–117 (1977).
12 A. Einstein, “Eine neue bestimmung der molekuldimension”, Annln. Phys. 19, 289 (1906); with correction in 34, 591
(1911).
13 A. Hon, “Modelling Filler Reinforcement in Elastomers” Ph.D. Thesis, Queen Mary, University of London (2005).
14 H. C. Brinkman, J. Chem. Phys. 20, 571 (1952).
15 I. M. Krieger, Advances in Colloid and Interface Science 3, 111 (1972).
16 J. M. Jaeger and S. R. Nagel, Science 255, 1524 (1992).
40 RUBBER CHEMISTRY AND TECHNOLOGY VOL. 84

17 C. A. Rogers, Proc. London Math. Soc. 8, 609 (1958).


18 A. I. Medalia. RUBBER CHEM. TECHNOL. 51, 437 (1978).
19 TARRC Natural Rubber Engineering Data Sheets, 1979–1986.
20 H. R. Ahmadi, J. Gough, A. H. Muhr, and A. G. Thomas, in Constitutive Models for Rubber, A. Dorfmann and A.
H. Muhr, Eds., Balkema, Rotterdam, ECCMRI, 1999.
21 G. Kraus, J. Appl. Polym. Symp. 39, 75 (1984).
22 C. K. L. Davies, D. K. De, and A.G. Thomas, Rubber Chem. Technol. 67, 716 (1994).
23 H. R. Ahmadi and A. H. Muhr, Plastics Rubber, & Compos. 26, 451 (1997).
24 M. Klüppel and J. Meier, in Constitutive Models for Rubber IV, P.-E. Austrell and L. Kari, Eds., Balkema, Lisse,
ECCMRIV, 101 (2005).
25 M. L. Williams, R. G. Landel, and J. D. Ferry, J. Am. Chem. Soc. 77, 3701 (1955).
26 J. D. Ferry, “Viscoelastic Properties of Polymers,” 2nd ed., Wiley, New York, 1970.
27 L. Boltzmann, “Zur theorie der elastischen nachwirkung”, Pogg. Ann. Phys., Erganzungsband 7, 624 (1878).
28 M. Takayanagi, S. Uemura and S. Minami, J. Polymer Science: Part C, 5, 113 (1964).
29 A. R. Payne and W. F. Watson, Trans. Inst. Rubber Ind. 39, T125 (1963).
30 C. Wrana and V. Härtel, Kautschuk Gummi Kunststoffe 61, 647 (2008).
31 A. Zhu and S. S. Sternstein, Compos. Sci. Technol. 63, 1113 (2003).
32 S. S. Sternstein, S. Amanuel, and M. Schofner, 172nd Technical Meeting of the Rubber Division of the American
Chemical Society, Inc. Cleveland, OH, October 16–18, 2007.
33 D. M. Turner, Plastic and Rubber, Process. Appl. 14, 197 (1988).
34 V. A. Coveney, D. E. Johnson, and D. M. Turner, RUBBER CHEM. TECHNOL. 68, 660 (1995)
35 C. Miehe and J. Keck, J. Mech. Phys. Solids 48, 323 (2000).
36 A. J. Liu and S. R. Nagel, in Jamming and Rheology—Constrained Dynamics on Microscopic and Macroscopic Scales,
A. J. Liu and S. R. Nagel, Eds., Taylor & Francis, London and New York, pp. 1–5, 2001.
37 P. Haupt, Continuum Mechanics and Theory of Materials, Springer-Verlag, Berlin, 2000.
38 H. Ahmadi, T. Dalrymple, J. Kingston, and A. Muhr, paper 78, 174th Technical Meeting of the Rubber Division of
the American Chemical Society, Inc. Louisville, Kentucky, October 14–16, 2008 to be submitted to RUBBER CHEM.
TECHNOL.
39 A. Lion, C. Kardelky and P. Haupt, RUBBER CHEM. TECHNOL. 76, 53 (2003).
40 A. Muhr, Proc. Inernational Rubber Conference, Kuala Lumpur, October 2008, to be submitted to Journal of Rubber
Research.

[ Received November 2010, Revised November 2010 ]

You might also like