A THREE-DIMENSIONAL CONSTITUTIVE MODEL

FOR THE LARGE STRETCH BEHAVIOR OF

RUBBER ELASTIC MATERIALS

ELIXN &I. ARKUDA and MARY C. BOYCIY

The Dcparlmcnt of Mechanical Engineering. The Massachusetrs Institute ol‘Tcchnolog~.
Cambridge. MA 02139. U.S.A.

A CXWXITI WI: model ib proposed for the deformation of rubber malerialr which is sh(lun 10 rcprescnt
successfully the response of thehe matcrrals In uniaxial cxlcnsion. biaxial exlenslon. uniaxial comprcsrion.
plane strain compression and pure shear. The dcvcloped constiturivs relation is based on an eight chain
rcprcscntation of the underlying macromolecular netuork structure of the rubber and the non-Gaussian
behavior of the individual chains in the proposed nctworh. ‘The eight chain model accurately captures IIIC
coopcrativc nature of netuork deformation while requiring only IWO matzrlal parameters. an initial
modulus and ;I lirnlting chain cxtcnsihilily. Slncc Ihcsc two parameters arc mechanistically linked IO the
physics of molecular chain orienlation involved in the deformation ofrubhcr. the proposed model reprcscnrs
a simple and accura~c COIIS~IIU~I~C’model 01 ruhbcr deli)rmatmn. The cham cxtcnsion in thlc network model
reducch to a function of the root-mean-square of the principal applied stretches as a result of cKccti\el)
sampling eight orientations uf principal stretch hpacc. The rcsulrs of the proposed tight cham model as
well as Ihosc oTse~eral prominent models arc compared with cxperimenral data of TKFLOAR (1944. 7ions
~rrru&~~ SW. 40, 59) illustrating the superiorit). simplicity and predictive ability of the proposed model.
Additionally. a new set of experimentswhich captures ths SUIL’ of deformation dependence of rubber is
described and conducted un three rubber materials. The eight chain model is found 10 rntdel and pr~d1c1
accurately chc behavior of the three ~sled materials furlhcr confirming its superiority and ctrcctivcncss
over earlier models.

I. INTRODUCTION

THE AIMOF THISWORKhas been to develop a fully three-dimensional. mechanistically

motivated constitutive relationship for nonlinear elasticity which would successfully
account for the state of deformation dependent response of rubber materials. The
need for such a rdationship is readily evident to any processor of rubber components
or purveyor of rubber goods concerned with aspects of material behavior under large
stretch deformation states which are not simple tension. A good constitutive model
should reprcscnt the three-dimensional nature of the stress -stretch behavior using a
minimal number of parameters to represent physically the deformation process.
Ideally, .the parameters should be obtainable from a small number. preferably one, of
experiments. Previous models of rubber elasticity such as WAM; and GLJTH (1952)
and FLORY and REHNIX (1943) have described or otherwise accounted for the charac-
t&tic ‘S.-shaped load versus stretch curve cxhibitcd by rubber materials in uniaxial
3x9
390 E. M. AKKUIIA and M. C. BOYCE

tension. The loaddstretch curve is highly dependent upon the state of imposed defor-
mation, shown, for example, in the data of TREL~AR (1944) depicted in Fig. 1. The
authors know of no existing model which accurately represents the behavior of such
materials in various deformation states and satisfies the criterion of requiring only a
small number of “physically based” parameters. Such material parameters or con-
stants should be independent of deformation state in order to provide a predictive
capability to the constitutive model.
The first statistical mechanics approach to describing the force on a deforming
polymeric network assumed Gaussian statistics to apply, that is the chains never
approached their fully extended length rL = IN where N is the number of statistical
links of length I in the chain between chemical crosslinks. [See TRELOAR (1975) for
instance for a more detailed description of Gaussian statistics and the corresponding
assumptions.] Gaussian statistics yields

6I

0
0
l

l 0

FIG. 1. Data fromTKELOAK (1944), plotted as force per unit unstrained area versus stretch, showing the
state of deformation dependence of a rubber material in uniaxial extension. biaxial extension and shear.
 Model for rubber elastic materials 391

W= ~G(lLf+A:+A$3), G = nk@ (1)

for the strain energy where the A, are the applied stretches and the rubbery modulus
G is a function of the chain density, n, Boltzmann’s constant, k and temperature 0.
Investigators since have built networks from chains described by Gaussian statistics
or have modified the chain statistics to allow larger stretches than are afforded by the
assumption of Gaussian statistics, then incorporated these non-Gaussian chains into
networks of three, four or an infinite number of chains. See WANG and GUTH (1952),
FLORY and REHNER (1943) and TRELOAR (1946) for the three chain non-Gaussian,
four chain Gaussian and four chain non-Gaussian networks, respectively, or TRELOAR
(1975) for an overview of network models including his own proposed method of
averaging the contribution of a single chain over a large number of orientations.?
These models have in common two physically based parameters, the rubbery modulus
G given in (1) and a chain locking stretch &_ defined as the value of the chain stretch
when the chain length reaches its fully extended state. Chain stretch is always given
by the current chain length divided by the initial chain length

The initial chain length is given from random walk statistics as r,, = aI. Thus for
the fully extended or locking chain length, rL = IN, the locking stretch becomes

Other statistical models have investigated affine versus phantom deformation of

junctions in networks and these have been discussed by MARK and ERMAN (1988). In
many cases models such as these require a third or fourth parameter to describe some
measure of chain interaction in the network.
Excellent historical perspectives of the phenomenological invariant-based models
of rubber deformation such as those of MOONEY ( 1940), RIVLIN (1948), VALANIS and
LANDEL (1967) and OGDEN (1972) have been given by TRELOAR (1975, 1976). The
theories of this genre are aimed at obtaining an expression for the elastic strain
energy which is of the form dictated by continuum mechanics as concerns an initially
isotropic, incompressible, hyperelastic solid. The phenomenological theories lack a
direct physical connection to the underlying mechanisms of deformation. The most
sophisticated of these is that of OGDEN (1972) which proposes the following form for
the strain energy function

with ,u~ and the x,, as experimentally fitted constants. Ogden makes no attempt to link
the adjustable parameters to any physical deformation mechanism. The Ogden model
is essentially empirical and generally requires more than one experiment to obtain the
number of constants required to capture state of deformation dependence.

1-The averaging was later published in TKELOAK

and RJLXSG(1979)
392 E. M. ARRUDAand M. C. B~YCF

2. PROPOSED MODEL

2.1. Requirements ofa model

The existing models referenced above have merit and are frequently used in treating
rubber deformation problems. Many of the early models capture the ‘S’-shaped load
versus stretch curve exhibited by rubber materials in uniaxial tension. A relative dearth
of experimental evidence exists in the literature for the response of rubbers in other
deformation states up to very large stretches save for the work of TRELOAR (1944),
in uniaxial extension, biaxial extension and shear, considered by many to be the
quintessential rubber data,t and the extensive biaxial experiments of JONES and
TRELOAR (1975). Consequently, many of the existing models fail in the task of describ-
ing the response of a rubber material under different states of deformation without
changing the model parameters. In other cases the models are prohibitive in the
number of parameters required to fit the data or the mathematical complexity they
represent. The need for a constitutive relationship which possesses mathematical
simplicity, requires one test to characterize the material and has a limited number of
parameters, has prompted the authors’ development of the following model.

2.2. Langevin chain statistics

The statistical mechanics approach to rubber elasticity models the rubber chain
segment between chemical crosslinks as a number N of rigid links of equal length 1.
The rigid link is that segment length of the actual chain which undergoes rigid body
motion in response to an imposed strain, depending on the actual rubber material the
statistical rigid link may span one or several repeat molecular units. The initial chain
length is taken from a random walk consideration of N steps of length 1, and is denoted
by ro,

r(, = JNl. (5)

The fully extended chain has approximate length IN so that the limiting extensibility
(or chain locking stretch), defined as the final length divided by initial length, is given
in terms of the statistical parameters as & = J/z. At any value of chain length the
most probable angular distribution of rigid links about the chain vector length may
be found. Following the use of Langevin statistics by KUHN and GRUN (1942), the
probability of the most probable link angle distribution about a given vector length
is taken to be equal to the probability of the vector length. The chain vector length is
denoted by rCha,“.In this way Kuhn and Griin obtain an expression for the probability
density function for chain lengths and subsequently the configurational entropy of a
stretched chain of current length rchaln

t Some debate has persisted over whether the characteristic upturn in the observed stress-stretch behavior
is solely due to crystallization and therefore, not present in all rubbers. In the following section we offer
additional experimental results on three rubber materials which exhibit the same basic, characteristic
behavior as that of Treloar’s data.
 Model for rubber elastic materials 393

where c is a constant, k is Boltzmann’s constant and /I is the inverse Langevin function,

B = pm- ’ [~&al”/N/l, for the Langevin function defined as _Y[p] = coth b-(1//I). The
use of Langevin statistics properly accounts for the limiting chain extensibility. The
work of deformation is proportional to the entropy change on stretching the chains
from the unstretched state and may be written in terms of the chain length as

W = nk@N -Oc’,
>

where n is the chain density, c’ is a combination of constants and the remaining terms
are as defined previously. The material is considered to be incompressible and the
principal stresses can be determined from the work of deformation to within an
arbitrary pressure, p,

where the ,I, are the principal stretches and the pressure may be determined from the
boundary conditions. The stress-stretch relations are frequently written in terms of
the difference in two principal stresses in order to eliminate the pressure term.

dW dW
(7, -cr‘z = 2, ~~- -A--.
d1. , - d/l,

2.3. Use of’ Langevin chain networks

The use of Langevin chain statistics in networks has been considered by WANG and
GUTH (1952) in their three chain model by TRELOAR (1946) in his extension of the
tetrahedron model of FLORY and REHNER (1943). TRELOAR (1975) and later TRELOAR
and RIDING (1979) also considered a model of a large assembly of chains which were
averaged by integrating over many spatial orientations. Analytical results of non-
Gaussian three chain, tetrahedron and total assembly of chains models are presented
in TRELOAR (1975) for uniaxial tension. The three chain and tetrahedron models have
been considered by the present authors for use as constitutive models of rubber
deformation, results of simulations of different deformation states using both models
will be presented later and compared to the proposed model. The three chain and
tetrahedron model systems are sketched in Figs 2 and 3 respectively for undeformed,
uniaxial extension and biaxial extension load geometries.
These models were also previously considered for their general predictive capabilities
of the characteristic ‘S’-shaped uniaxial response by TRELOAR (1975). He observed
that the response of the tetrahedron model was slightly dependent upon the orientation
of the volume element with respect to the extension direction. In their 1952 paper,
Wang and Guth stipulated that the faces of the three chain cube element be aligned
with the principal stretch space during deformation. TRELOAR (1975) concluded that
394 E. M. ARRUDA and M. C. BOYCE

2. Three chain rubber elasticity model for (a) undeformed, (b) uniaxial extension and (c) biaxial
extension configurations.

these models possess an inherent anisotropy which could be eliminated by a system

of a large assembIy of chains, each of Icngth equal to the root-mean-square chain
length, yO = JNI, which occupy random orientations in space. In this manner the
contributions from chains in several orientations may be averaged to give the overall
response. His proposed total assembly of chains model requires integration over
the distribution of chains for each subjected stretch state, a procedure which was
mathematically prohibitive 15 years ago and remains cumbersome today.
The authors agree with TreIoar’s assessments of the existing network models except
for his reluctance to restrict the orientation of a model to some specific relationship
with respect to principal stretch space as Wang and Guth had done. Motivated by
Treloar’s assertion to present a model which averages several spatial orientations, but
recognizant of the need for a mathematically concise representation, we have proposed
a model which possesses the cubic symmetry of principal stretch space as it averages
eight orientations of that space in determining the network response. The clear
advantage of this technique is its ability to simulate a true network response of
cooperative chain stretching which presents a clear picture of the deformation process
with mathematical ease.

2.4. Proposed model geometry

The proposed model considers eight orientations of chains in space which may be
envisioned by the eight chain network system sketched in Fig. 4 for undeformed,

FIG. 3. Four chain rubber elasticity model for (a) undeformed, (b) uniaxial extension and (c) biaxial
extension configurations.
 Model for rubber elastic materials 395

FIG. 4. Eight chain rubber elasticity model for (a) undeformed, (b) uniaxial extension and (c) biaxial
extension configurations.

uniaxial extension and biaxial extension loaded configurations. The chain length,
Y,,,~~,,,is sought for each of the chains in the eight chain model as a function of the
imposed principal global stretches A,, AZ, ,I3 in order that (7) may be differentiated to
obtain the stress-stretch relations (9). Consider the cube containing the eight chain
configuration in Fig. 5. The cube edges are taken to remain aligned with principal
stretch space during deformation, chains linked at the center of the cube extend to
the eight corners. In principal space the cube is allowed expansion along each principal
direction subject only to incompressibility which may be expressed as

/2,&A, = 1. (10)
Note that the restriction to principal stretch space does not limit the usefulness of
the model to axisymmetric deformations. For example, in pure shear the principal
stretches are 1, = I., A2 = 1 and I, = A- ‘. The directions in which these stretches act
rotate continuously with the deformation and this rotation is monitored by the
standard kinematics of finite strain deformation making use of the polar decompo-
sition and subsequent extraction of the stretch tensor eigenvalues from the general

FIG. 5. The unstretched network for the proposed eight chain model
396 E. M. ARRUDA and M. C. BOYCE

deformation gradient tensor. See for example the discussion by Hopkins in TRELOAR
(1976) or FARDSHISHEH and ONAT (1972). The eight chain model is always stretched
in the principal frame and is in general oriented differently from the laboratory frame.
The rationale for use of the model in this manner follows from the consideration that
in response to any deformation a principal stretch frame exists and the chains in that
reference frame will undergo stretches describable by the principal values of stretch,
i,,, ;1? and j_,.
The unstretched network includes eight chains of length Y(,= JNI inside a cube of
dimension a,,. From this geometry

2
N,, = --/I Yg. (11)

In Fig. 6 the cube is stretched by A,f, AZ/and i.,k^ so that the cube edges measure i., a,,,
izao and Lia, in the i^,,yand k^directions, respectively. A chain vector from the center
of the cube to a corner may be written for one chain as

(12)

This chain has vector length

~ch:U”= (13)

as do all remaining chains in the given network geometry, regardless of deformation

state. Substitution from (11) and (5) into (13) gives the chain length in terms of the
statistical parameters and the principal stretches

FIG. 6. The tight chain network in a stretched configuration.

 Model for rubberelasticmaterials 397

= ~Jzl(~;+~:+;~)‘? (14)
rcha’” Ih
In this form the expression for the chain vector length is suitable for substitution
into (7) which with (9) yields the following stress-stretch relation for the proposed
model

where the chain stretch, ,&,,,, = rchain/rO,is the same in each chain and is given by

AchaIn= &:+1.:+ng)‘? (16)

J?
Because each chain in the system undergoes a stretch equivalent to that in every other
network chain, the model is likened to averaging the contributions of a single chain
over eight spatial orientations.
The three stretch invariants are given by

I, = n:+ll:+1*:, (17)
2^2 “2”Z
12 = II,A2+A:A:+A,A3, (18)

z 3 = /12/12A2.
I 2 3 (19)
The expression for chain stretch is seen to reduce to a function of the first stretch
invariant, I,, and (16) may be rewritten as

(20)

The strain energy of the proposed model may be found from integration of (15) using
the series expansion form for the inverse Langevin function given for example in
TRELOAR (1954). The first five terms for the strain energy of this model are

W= nk@ f(Z, -3)+ 2&1:--Y)+ &+Zi-27)

1
+nk@ J&(1:-81)+
7000N
&&gox”(‘:-24”)
I
+ .‘.. (21)

Note that the strain energy in (21) exhibits a nonlinear I, dependence as a result of
the chain stretch being defined in terms of I, only.

3. EXPERIMENTAL PROCEDURE

The uniaxial compression test and plane strain compression test represent near
extremes in the behavior of polymeric networks under biaxial deformation states.
398 E. M. ARRUDA and M. C. BOWI<

FIG. 7. Sketch of plane strain compression testing apparatus. Specimen (A) is compressed by punch (B)
in the Y-direction and expands in the X-direction. The die (C) constrains the specimen along the Z-axis.

During uniaxial compression the polymer chains in the material stretch freely in all
directions within a plane perpendicular to the load axis. Under plane strain com-
pression the specimen is held within a channel die which completely constrains the
material in one direction while allowing expansion in a second direction perpendicular
to the first. The material flows along the expansion direction as a result of the
compressive load applied perpendicular to the plane containing the constrained and
expansion directions. In the plane strain compression die apparatus sketched in Fig.
7, expansion occurs along the X-axis, the material is constrained along the Z-axis and
Yis the load direction. Polymer network stretching occurs only along the axis of free
expansion in plane strain. The resulting stress versus stretch curve in plane strain
compression differs from that in uniaxial compression because of the vastly different
chain stretch behaviors in these two compression geometries.
Three commonly available rubber materials were chosen for this study: silicone
rubber (40 durometer), neoprene rubber (60 durometer) and gum rubber (hardness
unknown).? Uniaxial cubes of length 12 mm and plane strain specimens measuring
I2 mm x 9 mm x 12 mm were cut from these materials. In the plane strain compression
tests the 9 mm dimension was constrained. Uniform deformations of the order of
-200% strain were achieved in uniaxial compression and - 150% strain in plane
strain compression. The specimen (and die, in the case of plane strain) was placed
between flat plates on the ends of the load strain on an lnstron model 1350 servo-
hydraulic testing machine. Liberal lubrication was applied on all specimen surfaces
in contact with plates or die walls to insure uniform deformations; a MoS, based
paste made by Dow Corning was deemed sufficient for lubricating these materials.
Linear ramp tests were performed on all three materials for both deformation states
at displacement rates of approximately I rnrn s ‘. The load and displacement output
responses of the testing machine were converted to digital signals by a Kiethly A/D

+ These three materials were obtained at the Greene Rubber Company of Cambridge. MA
 Model for rubber elastic materials 399

+
0

+
0

0 +
0
0
+
+
t

1 .o 0.6 0.6 0.4 0.2 0.0

Stretch
FE. 8. Ramp tests in uniaxial compression and plane strain compression for silicone rubber.

board then transferred to a Macintosh personal computer for storage. The raw load
versus displacement data were corrected for the testing machine compliance before
being converted to nominal load (that is, load divided by initial area) versus stretch
information. The results of ramp tests in uniaxial compression and plane strain
compression are presented in Figs 8,9 and 10 for silicone, gum and neoprene rubber,
respectively. Each of the three materials shows the basic characteristic response of a
limiting stretch which is dependent upon the state of deformation. The three materials
differ significantly in their initial moduli and/or locking stretch values.

10
+

0
0.6 0.4 0.2 0.0
Stretch

FE. 9. Ramp tests in uniaxial compression and plane strain compression for gum rubber.
400 E. M. AKRUDAand M. C. BOYCE

0
~ 30--
2 0 +

0 +
2 0 --
0
+
0
0 +
+
1 o-- 00
00 ++
OOooO ++++++

0 -r+ +_@@+Q@@FY+++++

1 .o 0.6 0.6 0.4 0.2 0.0

Stretch

FIG. 10. Ramp tests in uniaxial compression and plane strain compression for neoprene rubber.

4. RESULTS

4.1. Explanation of the experimental data

As shown earlier in Fig. 1, the data on Treloar’s vulcanized rubber include tests in
uniaxial extension, biaxial extension and pure shear to very large extension ratios and
are plotted as load normalized by initial area versus stretch. The strong state of
deformation dependence of the large and limiting stretch behavior of the polymeric
network materials is shown in the data of Figs 1, 8, 9 and 10 and remains a primary
obstacle to true characterization of rubber elasticity. Additionally, ARRUDA and
BOYCE (1991) have also reported a similar state of deformation dependent strain
hardening response in glassy polymeric materials. These data on several different
polymeric materials, both above and below the glass transition temperature, clearly
demonstrate a certain level of generality to the presence of limiting stretch states in
polymeric materials whether rubbery or glassy.
The equi-biaxial extension data in Fig. 1 varies significantly from either the uniaxial
extension or pure shear data. The divergence of the biaxial data is due to the nature
of the molecular chain stretching in equi-biaxial extension versus uniaxial extension.
Biaxial extension offers two venues of principal tensile stretch (2, = ;1, 2: = 2,
’ ‘) ; an initially isotropic network of chains will reach limiting chain extension
1”q = 1”
due to stretching in both directions providing a planar state of orientation. In uniaxial
tensile deformation the chains extend along one direction only (A, = L, IU2= E ’ ‘.
a
A3 = A’ I”), additional stretch is thus allotted through the drawing of material from
the transverse directions, and the onset of limiting chain extension is delayed with
respect to the biaxial deformation locking. The pure shear data are plotted in terms
of the maximum principal stretch, I,, vs the corresponding normal force (i.e. not the
shear force) which acts in the direction of 2,. Pure shear deformation is more closely
related to uniaxial extension than to equi-biaxial extension because chain stretch
 Model for rubber elastic materials 401

occurs due to stretching along one principal direction (2, = L, I, = 1, L3 = Z ‘) with

chains being drawn from one direction transverse to the extension direction in pure
shear. Thus a pure shear experiment yields a limiting stretch value which is similar to
that obtained in uniaxial extension. A good model should be able to capture the
observed differences between biaxial and uniaxial tensile behavior as well as the
similarities in behavior between uniaxial and shear tests.
The plane strain compression and uniaxial compression data in Figs 8, 9 and 10
diverge for reasons similar to those used to explain the biaxial extension divergence
from the uniaxial extension and pure shear data above. Uniaxial compression offers
two directions of equal principal stretch (2, = 2, AZ = Iti I,‘*, A3 = Z “I), whereas in
plane strain compression chains extend along one direction only. The plane strain
compression stretch state is equivalent to that of pure shear (i, = 1, A2 = 1, A3 = 2. ‘).

4.2. Explanation qf’simulutions

Equation (15) has been used together with the appropriate boundary conditions to
simulate the uniaxial extension, biaxial extension and shear reponses of the proposed
model. Characterization of a particular material requires determination of two par-
ameters, the initial modulus and limiting extensibility. which is accomplished by fitting
the model to the data for any one deformation state. The authors chose to fit Treloar’s
uniaxial extension data which yielded the following parameters for this material:
CR = 0.09 and N = 26.5. These parameters were used in the simulations of biaxial
extension and pure shear tests. The results appear in Fig. 11. Using constants deter-
mined from this single set of uniaxial extension data the model quite accurately predicts
the biaxial extension and pure shear data. These results demonstrate the predictive
and fully three-dimensional aspects of our proposed eight chain model.
This exercise was repeated for both the three chain model of WANG and GUTH
(19.52) and the non-Gaussian tetrahedron model as modified by TRELOAR (1946,
1954). The stress-stretch relations for the Wang and Guth model are of the type

(22)

where the parameters have the same meaning as in (15). The numerical treatment
used for the tetrahedron model followed the early method of TRELOAR (1954) which
allowed the central junction point to seek an equilibrium position for affine dis-
placements of the tetrahedron corners. Equilibrium was found using an iterative
Newton scheme after each stretch increment for the condition of no net force on the
junction point. Stretch-stress relations are a function of the junction point equilibrium
position and must be determined numerically. Each of these models requires one set
of data to determine an initial modulus and locking stretch. The constants were found
by fitting the models to the uniaxial extension data then used in predicting the biaxial
extension and pure shear responses. The results for the three chain model are given
in Fig. 12 and for the tetrahedron model in Fig. 13 ; each figure includes the values
used for the model parameters. As these figures clearly show, the early models do not
402 E. M. ARRUDA and M. C. BOYCE

B
3
s

1 0 biaxial data
+ shear data
- uniaxial (eight chain)
.“....... (eight chain)

0
1
1 2 3 4 5 6 7 a

FIG. Il. Results of simulations using the eight chain model versus data from TKEL~AK (1944) in uniaxial
extension, biaxial extension and shear.

predict much of a state of deformation dependence beyond the initial modulus and
cannot be used as constitutive indicators of rubber elastic deformation.
The three chain model shows no state of deformation dependence beyond a stretch
of 4.0 for the three stretch states examined; the results of the simulations for three
states of deformation converge at large stretches whereas the data diverge. The failure
of the three chain model to capture the state of deformation dependence lies in the
absence of a cooperative nature of network chain deformation in this model except
through the imposition of incompressibility. In uniaxial extension, depicted in Fig. 2,
the chain parallel to the extension direction stretches until it reaches the locking
stretch. The remaining two chain stretches are determined by incompressibility and
do not contribute significantly to the force. Under equi-biaxial extension, two coplanar
chains extend independently until simultaneously reaching the locking stretch and the
third chain stretch is determined by consideration of incompressibility. Therefore the
 Model for rubber elastic materials 403

0 uniaxial data

L---l
+ shear data
l biaxial data
- uniaxial (three chain)
.......... biaxial (three chain)
- - - shear (three chain)

; ; 3 4 5 6 7 6
1

FIG. 12. Results of simulations using the Wang and Guth model versus data from TRELOAR (1944)in
uniaxial extension. biaxial extension and shear.

limiting stretch in equi-biaxial extension as predicted by the three chain model has
the same value as that obtained in uniaxial extension and in pure shear, the result is
convergence of the load vs stretch predictions for the three different stretch states
rather than the actual divergence of these curves which the data show.
The tetrahedron model shows a slight state of deformation dependence at small
stretches which is retained at large stretches. This model has some success in predicting
a state of deformation dependence because the chains in this network do respond
cooperatively to the deformation, see Fig. 3. Four chains stretch along the extension
direction in uniaxial extension resulting in an overall network limiting tensile stretch
which exceeds the locking stretch of a single chain (AL = ,6). In biaxial extension
perpendicular to one chain in the tetrahedron the remaining three chains undergo
stretching. The result is again increased network extensibility over the single chain
limiting stretch to a value which differs from the tensile locking stretch. The different
values for locking in uniaxial and biaxial extension are manifested in the load-stretch
404 E. M. AKRUDA and M. C. Bovcr

6 -L

I”

FIG. 13.Results of simulations using the four chain model versus data from TKELOAR (1944) in uniaxial
extension, biaxial extension and shear.

predictions of the four chain model which reach asymptotes at different values of
stretch for different states of deformation.
The tetrahedron model does not possess symmetry with respect to the principal
stretch space. The load versus stretch predictions depend upon the orientation of the
four chain tetrahedron with respect to i. ,, i, and i, in addition to the magnitudes of
the A,.
The eight chain model is clearly superior to the other statistical models in the state
of deformation dependence predicted. The advantages of the eight chain model arc
its proximity in kind to a system which is initially isotropic, the use of Langevin chains
which capture the effects of limiting chain extensibility and a network configuration
which responds cooperatively to an imposed deformation. These features allow for
mechanically simulating the state of deformation dependence of the behavior during
large deformation while retaining mathematical simplicity. The features of the eight
chain model as seen in Fig. 4 combine the favorable aspects of the previous three
 Model for rubber elastic materials 405

WANG and GUTH (1952) in their

discussion. All eight chains stretch uniaxially in response to an imposed uniaxial
extension deformation as is true for the four chains in the FLORY-REHNER (1943)
model. Again the limiting network stretch of the eight chain network exceeds the
extensibility of a single chain in extension. In biaxial extension all eight chains extend
identically due to stretching in two principal directions. The resulting limiting stretch
is something other than either the individual chain locking stretch or the limiting
stretch in uniaxial extension. A significant feature of the eight chain model is that all
chains stretch equally under biaxial extension as well as uniaxial extension. The chain
stretch is always the root-mean-square of the global principal stretch state. This
unique feature of the eight chain model is true for all other deformation states; all
chains stretch equally in response to any given deformation state.
Additional existing models were similarly tested against Treloar’s data. These results
are discussed in the Appendix.

30.

2.5

20,

z o uniaxial data
2 ‘5 + plane strain data
- uniaxial simulation
...........plane strain simulation

10,

5,

0,
3 0:a 016 0:4 0:2 0:o
Stretch

FIG. 14. Results of simulations using the eight chain model and data on silicone rubber
406 E. M. ARRUDA and M. C. BCIY~E

20

15

i
+;

z
z ‘0 -1

i’
‘...‘...... plane strain simulation

I+
ii
5

1 .o 0.6 0.6 0.4 0.2 0.0

Stretch

F~ti. 15. Results of simulations using the eight chain model and data on gum rubber

The proposed model was used to predict the plane strain compression responses of
the three materials examined in this study based on the constants derived from their
corresponding uniaxial compression responses. Results of these simulations appear
in Figs 14, 15 and 16 for the silicone, gum and neoprene rubber materials previously
described. The model parameters used to fit the uniaxial compression data are listed
for each material on the corresponding figure. The model accurately captures the state
of deformation dependence of all three materials. Each of these materials differs in
the modulus and/or locking stretch value needed by the model to fit the uniaxial
compression response. A direct comparison of the uniaxial compression responses of
the three materials has been made in Fig. 17, here the differences in actual material
moduli and limiting extensibilities are easily seen. Including Treloar’s data the model
has been shown to characterize four materials representing a range of material prop-
erties. Previous network models considered contained the same basic parameters of
an initial modulus and a measure of finite extensibility, but were unable to predict a
 Model for rubber elastic materials 407

60

50

40

-1 j
z
z 30 ...-....’ plane strain simulation

20

10

0
T I I I I r
1 .o 0.8 0.6 0.4 0.2 0.0
Stretch
FIG. 16. Results of simulations using the eight chain model and data on neoprene rubber.

state of deformation dependence because of inabilities of the models to effectively

represent a network response. In the case of the eight chain model the successful
prediction is a result of correct usage of the finite extensibility parameter as that of an
averaged measure of applied stretches over eight spatial orientations. The math-
ematically tractable eight chain model of (15) represents a true constitutive relation-
ship for rubber deformation which is able to correctly account for the large stretch
deformation response in one state of deformation, then with that characterization,
predict other states of deformation.
A polymer below its glass transition temperature exhibits the same network response
in strain hardening as rubber materials do during large stretch deformations. ARRUDA
and BOYCE(199 1) have shown the eight chain model to be successful in predicting the
state of deformation dependence of the strain hardening response of glassy polymers
whereas the three chain and tetrahedron models again fail.
408 E. M. AKRUDAand M. C. BOYU

30

25

0.6 0.4 0.2 0.0

Stretch
FIG. 17. Direct comparison of the uhiaxial compression responses of silicone, gum and neoprene rubbers.

5. SUMMARY

A physically based constitutive model for large stretch rubber deformation has been
proposed which has been specifically designed to account for the three-dimensional
state of deformation dependence in networked solids. The eight chain model presented
here is formulated such that the nature of the state of deformation dependence is
clearly seen to be the result of a network of chains reaching the individual chain
extensibility hmit at different imposed global stretch levels for different stretch states.
The eight,chain model successfully accounts for the state of deformation dependence
using a rubbery modulus and a locking stretch as its only two parameters, both of
which can be determined from a single experiment. Present in all statistical models,
these parameters are physically linked to the polymeric network and therefore provide
a basis for including other aspects of rubber elastic behavior such as temperature
dependence, swelling and Mullin’s effect. Indeed,the rubbery modulus term explicitly
includes temperature.
 Modelforrubber elastic materials 4OY

Experiments were conducted on three rubber materials to illustrate the state of

deformation dependence of these materials and to determine the predictive capability
and effectiveness of the proposed model. In addition the model was compared to
existing rubber elasticity models in the ability to capture the response of data in the
litcraturc (TKELOAR, 1944) in three states of deformation. The eight chain model was
shown to bc superior in its overall ability to successfully account for the three-
dimensional nature to the underlying mechanics of network solid deformation for a
total of four materials representing a range in material hardness and extensibilities.
The eight chain model contains many attractive features of concern in modelling
the complicated deformation proccdurcs involved in the finite straining ofamorphous
polymeric solids. It retains mathematical feasibility as a two parameter model and
also contains the ability to respond to an imposed deformation state in a manner
which simulates the actual mechanism governing the state of deformation response
of rubber materials.

This work is sponsored by The National Science Foundation (MSM-88 18233) and the MIT
Bradley Foundation.

ARRUDA, E. M. and BOYCE, M. C. 1991 Anisotropy and localization of plastic defor-

mation, Proc. Phficify ‘9I, p. 483.
FARD~I~ISHEH.F. and ONAT, E. T. 1972 In Prohkms in Plaslicil) (edited by A. SAW(X!K),
p. 8Y. Noordhoff, Lcyden.
GLORY. P. J. and ERMAN. B. 1982 Macromol. 15, 800.
FLORY. P. J. and REHSER, J., JR. I943 J. Chem. Plys. II, 512.
JONI~S,D. F. and TREI.OAR,L. R. G. I975 .I. P/t_w. D : A/y/. Phjx 8, 1285.
Ktirrs, W. and GK~~N, F. 1942 Kolloid Z. 101, 248.
MARK. J. E. and EKMAX. B. I988 Ruhhrlike Elasticity A Mokrular Primtv. John
Wiley, New York.
M~XINEY. M. 1940 J. uppl. Phjx 11, 582.
&DEN. R. W. I972 Proc. R. Sot. Land. A 326, 565.
RIVI.IN. R. S. 1948 Phil. Truns. R. Sot. Land. A. 241, 470.
TRI.LOAR. L. R. G. 1944 Trans. Furuday Sot. 40, 59.
TRFLOAR, I,. R. G. 1946 Truns. Faraday Sot. 42, 83.
TRI:LOAR. L. R. G. 1994 Trans. Furada~~Soc. 50, 88 I.
TREI.OAR, L. R. G. lY75 The P/t~xics of Ruhhw E/usricir_y. Oxford Uni-
versity Press, Oxford.
TRELOAK, L. R. G. 1976 Proc. R. Sot. Land. A. 351, 301.
TKEI.OAR, L. R. G. and Rmrsc;, G. 1979 Proc. R. SW. I.ond. A. 369, 261.
VALANIS, K. C. and LANLEI., R. F. 1967 J. appt. Ph~*s. 38, 2997.
WAX;C;.&I. C. and Gur~r, E. 1952 J. Chwtz. Phys. 20, I 144.

Additional models were considered for comparison with the eight chain model in their ability
IO reproduce the state of deformation dcpcndence of TRXOAR’S (1944) data. Two phenom-
410 E. M. ARRUUA and M. C. BOYCE

enological models were considered, the well known Mooney-Rivlin (1948) relation and the
Ogden (1972) model. The strain energy in the Mooney-Rivhn relation is given as
W= C,(l,-3)+C,(I,-3% (Al)
where C, and C, are constants. The strain energy expression for the Ogden model was given
in the text of this manuscript. The results of simulations with both of these models appear in
Fig. Al along with the constants needed for each model. The simulations with the Ogden
model were taken directly from OGDEN (1972). A minimum of six independently adjustable
parameters are required by the Ogden model to fit these three deformation states shown in
Fig. Al ; more than one deformation state may have been considered in determining the six
constants as only the first four constants are necessary to produce the uniaxial extension result.
An additional statistical mechanics model of FL~RY and ERMAN (1982) which accounts for
chain interactions has been considered. The elastic strain energy of the network is found from
the sum of phantom and constraint contributions
W= wP,+wL, (A2)
where IV,,, of phantom Gaussian chains is

6i

-1
I

5,--
4l --

z
g 3

-- Ogden (shear)
1 -#- M-R(uniawal)
+ M-R (biaxial)
-e M-R (shear)
l biaxial data
unlaxial data

0
I 2 3 4 5 6 7 8
h

FIG. Al. Results of simulations using the Mooney--Rivlin and Ogden models versus data from TKELOAK
(1944) in uniaxial extension, biaxial extension and shear.
 Model for rubber elastic materials 411

0
l
0 / /
1

a 0

e + /

l /

2 3 4 5 6 7 6
A.
FIG. A2. Results of simulations using the Flory and Erman model versus data from TRELOAR(1944) in
uniaxial extension, biaxial extension and shear.

W,, = jtk@(I, -3). 643)

Note that (A3) differs from (I) in the text by the parameter 5 where

The parameter n is the number of network chains and 4 is the number of chains meeting at a
junction. When C$= 4 the junction is tetrafunctional and the strain energy of the phantom
network in (A3) is one-half of the affine strain energy in (1). The contribution of constraints
to the free energy is given as
WC = inkoC[B,+D,-In(B,+l)-In(D,+l)] (A9

for
B, = K’(~,Z-I)(;L~+K)-~, (A6)
Di = /~:K-‘B, (A7)
412 E. M. ARKUDA and M. C. BOYCE

and where ICis a measure of the strengths of the constraints which depends on the relative sizes
of free (phantom) fluctuations and actual constrained fluctuations. This model depends on the
parameters n, 4 and IC. In it ti + m for completely constrained junctions and K + 0 in the
phantom chain limit, n can take on any large, positive value and 4 must be greater than two.
Results of simulations with the Flory and Erman model appear in Fig. A2.
The model results shown in Figs Al and A2 reveal that of these existing models, only the
Ogden model captures the state of deformation dependence of deformation. However, in order
to capture this behavior, the Ogden model required six parameters.