Accepted Manuscript

Thermo-Oxidative Aging of Elastomers: A Modelling Approach Based on a

Finite Strain Theory

Michael Johlitz, Nico Diercks, Alexander Lion

PII: S0749-6419(14)00069-2
DOI: http://dx.doi.org/10.1016/j.ijplas.2014.01.012
Reference: INTPLA 1772

To appear in: International Journal of Plasticity

Please cite this article as: Johlitz, M., Diercks, N., Lion, A., Thermo-Oxidative Aging of Elastomers: A Modelling
Approach Based on a Finite Strain Theory, International Journal of Plasticity (2014), doi: http://dx.doi.org/10.1016/
j.ijplas.2014.01.012

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers
we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and
review of the resulting proof before it is published in its final form. Please note that during the production process
errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
 Thermo-Oxidative Aging of Elastomers: A Modelling
Approach Based on a Finite Strain Theory
Michael Johlitza , Nico Diercksb , Alexander Lionc
a
Institute of Mechanics, Universität der Bundeswehr München
Werner-Heisenberg-Weg 39, 85579 Neubiberg, Germany
E-Mail: michael.johlitz@unibw.de
b
Institute of Mechanics, Universität der Bundeswehr München
Werner-Heisenberg-Weg 39, 85579 Neubiberg, Germany
E-Mail: nico.diercks@unibw.de
c
Institute of Mechanics, Universität der Bundeswehr München
Werner-Heisenberg-Weg 39, 85579 Neubiberg, Germany
E-Mail: alexander.lion@unibw.de

Abstract
Polymers are highly important in industrial applications such as bearings,
seals, bonds and coatings. Since these components are used in various areas
of engineering, it is obvious that they are exposed to different environmental
influences such as mechanical stresses, temperature profiles and chemical or
biological substances. Therefore, their properties change over time which
leads to limited operating times.
In this contribution, the chemical aging of elastomers within air is exam-
ined. A representative example for this phenomenon is the thermo-oxidative
aging of elastomeric bearings in automobiles. After the state of the art is
presented and highlighted by scientific data, the experimental equipment
necessary for the thermo-oxidative aging studies is introduced. According to
that, experimental data are shown and analysed. From the theoretical point
of view, a constitutive approach is formulated by evaluating the Clausius-
Planck inequality. The model exhibits a physically-based structure such that
a clear identification of all material parameters is possible and meaningful nu-
merical simulations can be shown. The contribution is closed by a summary
and outlooks to future trends and objectives.
Keywords: chemical aging, thermo-oxidative aging, thermodynamics,
nonlinear continuum mechanics

Preprint submitted to International Journal of Plasticity April 10, 2014

1. Introduction
The application range of polymeric materials covers almost all areas of
industrial engineering. Examples for components made out of these materials
are bridge bearings, engine gaskets, coatings or the use as an adhesive for
bondings in lightweight constructions. Polymers are characterised by their
excellent workability, formability and versatility.
The manifold application of these materials attracts a worldwide interest
in the study of their mechanical and thermal properties. Thus, e.g. Bou-
vard et al. (2013) propose a viscoplastic model that both describes the time-
, temperature-and stress-dependent material behaviour. Laiarinandrasana
et al. (2009) investigate the temperature-dependent material behaviour of
Polyvinylidene Fluoride and develop interesting constitutive equations based
on the mechanics of porous media. As more high ranking activities in the
field of polymer research the group of Anand and co-authors should be men-
tioned. In the paper of Anand et al. (2009) a thermo-mechanically coupled
theory for the modelling of polymers under finite deformations is developed,
which is applied to various materials in the work of Ames et al. (2009). An
extension of this contributions on coupled problems in the glass transition
region as well as an adaptation of the model parameters and meaningful
simulations can be found in Srivastava et al. (2010a) and Srivastava et al.
(2010b). Ayoub et al. (2012) developed a damage model which considers
the high-cyclic fatigue of elastomers with respect to mutliaxial loading con-
ditions and in Ayoub et al. (2013) stress-softening, hysteresis and permanent
set of rubber-like materials were investigated and modelled by using a fi-
nite time-dependent constitutive approach. As one can see, in the area of
the thermomechnical behaviour of unaged polymers and fatigue, there is a
huge amount of interesting literature. But from the industrial point of view,
the mechanical behaviour of aged polymers is also of great interest and an
increasing area of research. The current article will focus on this.
During their use polymers are exposed to various environmental influences
which may affect their mechanical properties. These mainly include climatic
factors such as temperature, UV radiation, oxygen, ozone moisture and the
interaction with media like e.g. fuels, oils or chemicals. These mentioned
influences are some of the so-called outer aging criteria and their interac-
tions with the solid lead to measurable changes in the mechanical properties
of the material so that the operation time is limited. In contrast to this,
inner criteria are for example the incomplete polycondensation or vulcanisa-

2
tion, unstable crystallisation or internal stresses. Aging processes are usually
thermally activated, i.e. an elevated temperature may accelerate the aging
process, cf. Becker and Braun (1996).
Following the contributions of Ehrenstein and Pongratz (2007), aging is
defined as an ensemble of all physical and chemical changes of a material
over a period of time which influences the mechanical material behaviour in
such a way that the durability is limited. Besides the notation physical and
chemical aging also the items short-term and long-term behaviour are used.
This contribution is foccused on chemical aging.
Chemical aging is an irreversible process that changes the chemical struc-
ture of the molecules in the polymer network. It cannot be reversed by
heating, see for example Hutchinson (1995), Ehrenstein and Pongratz (2007).
Rather this process is rooted in the macromolecular structure of the polymers
and the associated bonding forces. Therefore, the macromolecules are degen-
erated or rebuilded due to diffusion-controlled or reaction-driven processes.
Such processes can be for example induced by exposure, biological or chem-
ical environmental conditions or the reaction with oxygen. In this case, the
solid is penetrated by the substance (fluid, gas) through its surface and after
a so-called initial time, the irreversible changes of the mechanical properties
are triggered by chemical reactions between substance and solid. In general,
several substances can simultaneously diffuse into the solid and not only react
with the solid, but also among themselves.
One of the most important aging processes is the reaction and interaction
with oxygen, the so-called thermo-oxidative aging, cf. Becker and Braun
(1996). This refers to a diffusion-reaction-driven process. Regarding the
ongoing chemical reactions in this case, both network degradation or chain
scission as well as network reformation and reorientation processes can be
observed. It has been examined by Tobolsky (1967) that the two effects can
overlap, where, depending on the system and the ambient medium, the one
or the other effect may be dominant. Already in 1944, studies on thermo-
oxidative aging were perfomed and published by Tobolsky et al. (1944). It
has been worked out that the chemical degradation process depends on both,
the oxygen concentration and the temperature. This property is exploited
in experimental aging studies to achieve a significant reduction in testing
time. At first, the material’s lifetime is examined for higher temperatures
than the operating temperature. In a second step, the observed material
behaviour is extrapolated to lower temperatures so that the operating time
can be estimated.

3
 From the experimental point of view, there exist established tests by
means of which the chemical aging behaviour of materials can be detected
as a function of the temperature and the exposure in a variety of media.
The standard experiments can be devided into continuous and intermittent
investigation of the mechanical properties of aged specimen. The basic idea
and the physical motivation of these tests can be found in the contributions
of Andrews et al. (1946), Scalan and Watson (1957), Dunn et al. (1959), Ore
(1959), Tobolsky (1967) and Smith (1993).
Continuous relaxation tests are also known as chemical stress relaxation
tests. In these experiments, substance samples are stored at constant tem-
perature and strain in an ambient medium over a period of several weeks or
months and the stress behaviour is recorded over time, cf. Tobolsky (1967).
With this test it is possible to capture the temporal behaviour of network
degradation at constant temperatures and subsequently to extrapolate the
results to lower temperatures by using the well-known Arrhenius function,
cf. Tobolsky (1967), Shaw et al. (2005), Duarte and Achenbach (2007). In
general, the stress decreases monotonically during testing, unless the network
reformation process causes shrinkage, Andrews et al. (1946). The temporal
decrease in stress may reach the value zero, which is equivalent to a complete
degradation of the original network of the virgin specimen with respect to
the deformed configuration.
In additon to this test, the intermittent experiments have to be men-
tioned. In this case, substance samples are stored (usually unencumbered)
under constant temperature in a selected medium. At predetermined time
intervals, the aged samples are taken out from the medium and subjected to
a short-term test at room temperature. This ensures that the aging state of
the sample is frozen during the short-term tests. At this point, there is a
large variety of experiments that can be performed. Examples are the short-
term relaxation test, the compression set test, monotonic tests or all kinds
of dynamic tests.
Further studies with respect to the influence of temperature profiles,
weathering, radiation and first modelling approaches have been proposed
by Blum et al. (1951), Shaw et al. (2005), Duarte and Achenbach (2007)
and Ehrenstein and Pongratz (2007). Compared to the problems related to
the viscoelastic stress relaxation, known as a physical, reversible aging effect,
chemical aging usually occurs on much larger time scales, cf. Budzien et al.
(2008).
However, all experiments described have one thing in common: They

4
all investigate substance samples with small dimensions so that any diffusion
processes are negligible and a homogeneous oxygen distribution in the sample
can be assumed. However, especially for components with larger dimensions,
this is not the case. There, the oxygen distributions are generally inhomoge-
neous. Furthermore, oxidation-reaction regions can be formed, see Pochiraju
and Tandon (2006). A review article dealing with this subject is the paper
of Audouin et al. (1994). Based on these knowledges, it is on the one hand
possible to investigate sufficiently thin samples where a homogeneous oxygen
concentration and saturation can be assumed as described in Blum et al.
(1951). On the other hand, the diffusion and the reaction with the solid have
to be taken into account, cf. Johlitz and Lion (2012), Shaw et al. (2005).
Therefore, measurements with respect to the oxygen concentration in the
solid have to be provided as described in Steinke et al. (2011).
From the modelling point of view, there is a certain number of papers
in which the authors deal with chemical aging processes and their influences
on the material. In the work of Pochiraju and Tandon (2006) the modelling
and simulation of the degradation process of polymers as a result of thermo-
oxidative aging is investigated. Thereby, the temperature properties and
oxygen concentration as well as the weight loss are taken into account. Fur-
ther modelling approaches can be found in Duarte and Achenbach (2007),
Johlitz (2012), Lion and Johlitz (2012) where the basic effects of the ag-
ing behavior of elastomers is described. With regard to diffusion processes,
which are related in the context of chemically reacting substances, the work
of Dunwoody (1970), Samohýl and Šı́pek (1985) and Quang et al. (1988)
should be mentioned. While Dunwoody developed a theoretical concept for
modelling diffusion processes in solid-fluid mixtures, Quang et al. worked on
the concepts of reversible and irreversible thermodynamics in reacting and
non-reacting fluid-solid mixtures. A similar, thermomechanically consistent
formulated contribution was delivered by Lustig et al. (1992) who described
Fickian and non-Fickian diffusion processes in polymers by using multiphase
continuum mechanics and thermodynamics. In a recent paper by Loeffel and
Anand (2011) a chemo-thermomechanically coupled theory with respect to
thermal barrier coatings is developed which considers the mechanical proper-
ties as well as the diffusion and swelling processes due to chemical reactions.

5
2. Material and experimental methods
For the experimental investigation of the phenomena described above two
different kinds of tests have been carried out:

• continuous relaxation tests

• intermittent monotonic tension tests

All tests have been performed in uniaxial tension. The samples are
rectangular-shaped, the cross section is 2.5 mm x 5 mm and the initial-length
L0 is 50 mm. All samples are fabricated out of natural rubber, filled with
60 phr of carbon-black. During aging phases the samples were exposed to
air. The stress-values P11 are 1st Piola-Kirchhoff-stresses thus meaning that
they are referenced to the undeformed configuration.

2.1. Continuous relaxation test

Figure 1: Device for testing 3 samples at a time

The continuous relaxation test is very useful to investigate the network

degradation process. The samples are stretched to a certain strain which is
held constant for a sufficient long duration (e.g. 500 h) at higher temperature
(e.g. 80◦ C). After an initial phase of physical relaxation (due to viscoelastic
effects) a decrease of stress caused by chemical aging is expected. This test is
independent from the influence of network reformation as the strain is kept
constant for the whole duration while the new network is formed stress-free
in the deformed configuration.

6
 2

1.8

1.6

1.4
P11 [MPa]

1.2

0.8

0.6
60°C
0.4 80°C
100°C
0.2 3 4 5 6
10 10 10 10
Time [s]
Figure 2: Relaxation-test at ε= 20 %

This test has been carried out at 3 different levels of deformation (ε=
20 %, 35 % and 50 %) and 3 different temperatures (θ = 60◦ C, 80◦ C and
100◦ C). Furthermore each test has been performed with 3 samples using
a device shown on the picture in figure 1 to ensure the reliability of the data
obtained.
Figure 2 shows the stress-over-time signals of samples that have been
stretched to 20 %. As expected, the physical relaxation of viscoelastic over-
stresses is visible in the beginning (t < 104 s for 100◦ C, resp. t < 105 s for
60◦ C). This phase is followed by a turning point resulting in a significant
decrease in stress. This decrease is supposed to be caused by the network
degradation process.
The curve at 100◦ C shows an increase of stress in the final phase (t >
7 ∗ 105 s), what was not expected. This increase is supposed to be caused
by a shrinkage process: After such a long duration at higher temperatures,
shrinkage of material is not uncommon. Due to the constant length during
the test this results in an increase of effective strain leading to an increase of
stress. This is an important side-effect of chemical aging, but not the topic
of this paper. Therefore, the stress-signal after the minimum is neglected in
this context.

7
 3 3.5

2.5 3

2.5
2
P11 [MPa]

P11 [MPa]
2
1.5
1.5
1
1

0.5 60°C 60°C

80°C 0.5 80°C
100°C 100°C
0 3 4 5 6
0 3 4 5 6
10 10 10 10 10 10 10 10
Time [s] Time [s]
Figure 3: Relaxation-test at ε= 35 % (left) and ε= 50 % (right)

Figure 3 shows the stress-over-time curves for strains of 35 % and 50 %.

The behaviour is very similar to the observation at 20 %.
The observation of the tests presented in this section could lead to the
assumption that the stiffness of the material is significantly reduced during
the aging process, but the contrary was observed. During unclamping at
the end of the experiments, it was noticed that the samples became very
hard and brittle. The quantification of this effect is investigated in the next
section.

2.2. Intermittent monotonic tension test

This test is used to investigate the reformation of a new network. In this
case, samples are stored at higher temperatures without any deformation.
After certain durations they are cooled down to room-temperature (aging
is intermitted) and a mechanical test with monotonic increasing strain is
carried out (0 % to 50 %). The rate of deformation is kept small (0.1 %/s) to
avoid significant viscoelastic effects. For each test, a new sample is taken to
ensure that no other influences (e.g. Mullins-Effect) lead to a change in the
material behaviour.
Figure 4 shows the stress-strain-relation for samples that endured certain
aging phases at 100◦ C. It is observed that even short aging phases result in
a significant increase in stiffness (up to + 70 %). Furthermore, it can be seen
that sample fracture occurrs after aging phases with θage = 100◦ C and dage ≥
3 days.

8
 6

P11 [MPa]
3

2 unaged
aged 12 hours at 100°C
aged 24 hours at 100°C
1 aged 48 hours at 100°C
aged 72 hours at 100°C
aged 6 days at 100°C
0
0 10 20 30 40 50
Strain [%]
Figure 4: Intermittent tension-test, θage = 100◦ C

The graphs in figure 5 (θage = 60◦ C and 80◦ C) show a similar behaviour to
the graphs for 100◦ C. While the network degradation process observed during
the relaxation-tests (figure 2 and 3) is less pronounced at lower temperatures,
the increase of stiffness caused by the network reformation becomes visible
after a quite short duration (only some days) even at lower temperatures.

6 6

5 5

4 4
P11 [MPa]

P11 [MPa]

3 3

2 2 unaged
unaged aged 36 hours at 80°C
1 aged 72 hours at 60°C 1 aged 4 days at 80°C
aged 12 days at 60°C aged 6 days at 80°C
aged 21 days at 60°C aged 8 days at 80°C
0 0
0 10 20 30 40 50 0 10 20 30 40 50
Strain [%] Strain [%]
Figure 5: Intermittent tension-test, θage = 60◦ C (left) and θage = 80◦ C (right)

For a more detailed investigation of the network reformation process,

some more monotonic tension tests with different aging durations have been

9
carried out, but are not shown in figures 4 and 5 for better visibility. To
quantify the influence of the aging phase, the stress at a certain strain (here
ε = 25 %) is analysed as a function of duration dage and temperature θage and
plotted in figure 6.

5
60°C
80°C
100°C
4.5
P11 (25 %) [MPa]

3.5

2.5 4 5 6
10 10 10
dage [s]

Figure 6: Intermittent tension-test, stress at ε = 25 %

10
3. Theory and calculation
On the basis of the given experimental results from section 2, a consti-
tutive approach is motivated and prepared which allows the simulation of
thermo-oxidative aging on the basis of network degradation and reformation
processes.
For this purpose, the following assumptions and restrictions are made:
the material is saturated with oxygen, i.e. the diffusion process inside the
laboratory samples plays no role and the chemical reaction between the solid
and oxygen runs homogeneously. Furthermore, no local temperature profiles
are taken into account. It is assumed that the material behaves incompress-
ible and that the aging process is not depending on the strain which is applied
on the samples. It is expected from a phenomenological point of view that
thermo-oxidative aging can be described by two competing mechanisms, the
network degradation and the network reformation process. As process vari-
ables, the right Cauchy-Green tensor C and the internal variables qd and qr
are taken into account. While the internal variable qd describes the degra-
dation of the existing polymer network, the internal variable qr maps the
network reformation process. The development of these introduced variables
is described via evolution equations. A very good review article concerning
the use of internal state variables can be found for example in Horstemeyer
and Bammann (2010). The formulation of the material model provides the
possibility to separate isochoric and volumetric deformation parts.
This motivates the isochoric-volumetric split of the √ deformation gradient
1/3
F = FV · FI (Flory, 1961) with FV = J I and J = det C. Therefore the
well-known relations between the isochoric right Cauchy-Green tensor CI , its
first and second invariants ICI , IICI as well as their derivatives with respect
to C can be derived:
CI = J −2/3 C, ICI = J −2/3 IC , IICI = J −4/3 IIC
∂ ICI  
= J −2/3 I − 31 IC C−1 (1)
∂C
∂ IICI 2
= − J −4/3 IIC C−1 + J −4/3 (IC I − C)
∂C 3
The second law of thermodynamics in the form of the Clausius-Planck in-
equality reads
−ρ0 ψ̇ + T̄ : Ė ≥ 0 , (2)

11
with the densitiy of the reference configuration ρ0 , the 2nd Piola-Kirchhoff
stress tensor T̄ and the time derivative of the Green strain tensor Ė = 1/2 Ċ.
The colon indicates the double scalar product of the basis vectors. Starting
from this considerations an additive split of the Helmholtz free energy func-
tion ψ is motivated into a volumetric part ψvol , into a degradative part ψd
and into a part ψr which contains the network reformation process. The
volumetric part is considered as (Simo and Taylor, 1982)
1
ρ0 ψvol = K (J − 1)2 + (ln J)2 ,
 
(3)
2
by introducing the bulk modulus K, which, from the numerical point of
view, has to be chosen approximately three orders of magnitude larger than
the other mechanical parameters in order to allow a nearly incompressible
simulation of the material’s behaviour. The network degradation part of the
energy function is described via a Mooney (Mooney, 1940) approach,
ρ0 ψd = c10 (qd ) (ICI − 3) + c20 (qd ) (ICI − 3)2 + c30 (qd ) (ICI − 3)3
(4)
+ c11 (qd ) (ICI − 3) (IICI − 3) + c01 (qd ) (IICI − 3) ,
in which the material parameters cij have to be formulated in a suitable
manner as a function of the internal variable qd . To ensure that the network
reformation process occurs stressless, a hypoelastic formulation regarding to
the papers of Hossain et al. (2009) and Lion et al. (2011) is considered. For
this purpose, the corresponding part of the free energy has to be placed in
integral form as a history functional
1 t 4
  
ρ0 ψr = Γ (s) : [E(t) − E(s)] : [E(t) − E(s)] ds . (5)
2 0
4
In doing so the time-dependent fourth order tensor Γ (t) is introduced with
the following mathematical relationship:
4 ∂2w ∂2w
Γ (t) = qr (t) = 2 q r (t) (6)
∂ E2 ∂ C2
In this equation, the variable w represents the strain energy density of the
network reformation process. Similar to equation (4), a Mooney approach is
chosen
w = d10 (ICI − 3) + d20 (ICI − 3)2 + d30 (ICI − 3)3
(7)
+ d11 (ICI − 3) (IICI − 3) + d01 (IICI − 3) ,

12
Based on this definitions, the time derivatives of the parts of the free energy
function are calculated to
 
J 1
ρ0 ψ̇vol = K (J − 1) + ln J C−1 : Ċ ,
2 J
ρ0 ψ̇d = c10 (qd ) + 2 c20 (qd ) (ICI − 3) + 3 c30 (qd ) (ICI − 3)2


 
− 23 1 −1
+ c11 (qd ) (IICI − 3)] J I − IC C : Ċ
3
 
− 34 2 −1
+ [c11 (qd ) (ICI − 3) + c01 (qd )] J IC I − C − IIC C : Ċ
3
∂ρ0 ψd
+ q̇d ,
∂qd
 t 4 

ρ0 ψ̇r = Γ (s) : [E(t) − E(s)] ds : Ė .
0
(8)
Inserting these expressions into the isothermal form of the dissipation in-
equality (2), the standard argumentation leads to an additive split of the
stress tensor into three parts

T̄ = T̄vol + T̄d + T̄r (9)

with
 
1
T̄vol = J K (J − 1) + ln J C−1 ,
J
T̄d = 2 c10 (qd ) + 2 c20 (qd ) (ICI − 3) + 3 c30 (qd ) (ICI − 3)2


 
− 32 1 −1
+ c11 (qd ) (IICI − 3)] J I − IC C
3
 
− 34 2 −1
+ 2 [c11 (qd ) (ICI − 3) + c01 (qd )] J IC I − C − IIC C ,
3
 t 4
T̄r = Γ (s) : [E(t) − E(s)] ds .
0
(10)
The integral form of equation (10)3 is reformulated by using the time differ-

13
entiation  •
˙ 4 4 1 ∂w
T̄r = Γ (t) : Ė = Γ (t) : Ċ = qr (t) . (11)
2 ∂C
Thus, it is guaranteed that the reformation process runs stressless. This ap-
proach also shows that the network reformation process does not contribute
to the dissipation inequality. Moreover, it is possible to avoid the calcula-
tion of the fourth order tensor by using some clever mathematical relations.
Taking equation (7) into account and using the time derivatives

I˙C = tr Ċ ,
 
− 23 1 −1
I˙CI = J I− C : Ċ ,
3
˙C
II = (IC I − C) : Ċ ,
(12)
 
˙C − 43 2 −1
II I
= J IC I − C − IIC C : Ċ ,
3
1
J˙ = J C−1 : Ċ ,
2
the final expression reads
  
˙ ˙ ˙

− 23 1 −1
T̄r = qr (t) (2 d20 + 6 d30 (ICI − 3)) IC + d11 II C J I − IC C
3

+ d10 + 2 d20 (ICI − 3) + 3 d30 (ICI − 3)2 + d11 (IICI − 3)

 

 2 
1
∂ J −3
I − 3 IC C−1

∂t
 
4 2
+ d11 I˙CI J − 3 IC I − C − IIC C −1
3
 4  
∂ J − 3 IC I − C − 23 IIC C−1
+ (d11 (ICI − 3) + d01 ) .
∂t
(13)
Lastly the dissipation inequality has to be evaluated
∂ρ0 ψd
+ q̇d ≥ 0 . (14)
∂qd

14
From a physical point of view, it makes sense to require the following assump-
tions for the internal variable qd which describes the degradation process:
∂cij (qd )
q̇d ≥ 0, 0 ≤ qd ≤ 1, ≤ 0. (15)
∂qd
These requirements are fullfilled by using the simple approach

cij (qd ) = cij (1 − qd ) , qd (0) = 0, cij (1) = 0 . (16)

Since the network degradation is depending on the temperature θ of the solid,

the corresponding evolution equation is formulated as follows:
Ed
q̇d = νd e− R θ (1 − qd ) . (17)

In (17), the model parameters νd and Ed are introduced. R = 8.314 J/molK

represents the universal gas constant and a deformation dependence is om-
mited. Concerning the network reformation that is represented by the evo-
lution equation for the internal variable qr , it can be argued in a physically
similar manner, so that the following conditions must be hold

q̇r ≥ 0, 0 ≤ qr ≤ 1, qr (0) = 0 . (18)

The associated temperature-dependent evolution equation is formulated in

analogy to equation (17)
Er
q̇r = νr e− R θ (1 − qr ) , (19)

with two additional model parameters νr and Er . This model is implemented

into the robust finite element code PANDAS (Ehlers and Ellsiepen, 1998) for
the prospective simulation of complex structures. Equation (101 ) is used as
a penalty term in order to ensure that the incompressibility condition is full-
filled. All the time dependent equations, i.e. (13),(17),(19) are numerically
treated by the Euler-backward integration procedure.

4. Results and discussion

Applying the theory described in chapter 3, the experiments presented
in chapter 2 have been simulated. Therefore the constitutive equations have
been adopted for the use in uniaxial tension with aging at contant strain and
isothermal conditions, which leads to a significant simplification.

15
4.1. Continuous relaxation test
As mentioned before, the network reformation process has no influence
during this test (P11 = P−11 ) as the strain is kept constant. This allows the
separation of the effects which enables a better identification of the model
parameters. The time depending stress-function can be written as

P11 (t, λ1 , θ) = [1 − qd (t, θ)] · P−

11,0 (λ1 , θ) (20)

where P−11,0 (λ1 , θ) denotes the equilibrium stress of the unaged material at the
chosen stretch λ1 = 1 + ε1 and temperature θ. Using isothermal conditions
(θ = const.), the evolution equation for qd (eq. 17) can be solved analytically
to
− t
qd (t, θ) = 1 − e τd (θ) , (21)
Ed

−1
where τd (θ) = νd e− R θ represents a chemical relaxation time for the net-
work degradation process. The task of parameter identification is performed
in two steps:
1. Identification of P−
11,0 and τd (θ) independently for each relaxation-tests.
As seen in the experimental data, P− 11,0 cannot be derived directly as
the processes of physical relaxation in the beginning of the test and
the chemical aging are overlapping. Therefore P− 11,0 is used as a fitting
parameter for each curve.
2. Identification of the model parameters Ed and νd by using the previ-
ously identified relaxation-times τd (θ).

ε [%] θ=60◦ C θ=80◦ C θ=100◦ C

20 1.070 × 107 1.095 × 106 1.830 × 105
35 9.962 × 106 1.031 × 106 1.656 × 105
50 9.819 × 106 7.327 × 105 1.625 × 105

Table 1: τd for relaxation-tests

The identified values for τd are listed in table 1. At a first view, it is a little
surprising that the relaxation-times depend on the applied strain (decreasing
relaxation-times with increasing strain). On the one hand, one might think
of material spread, but on the other hand, when the chemical processes of
the aging phenomena are considered, this tendency makes sense. Due to

16
the thickness of the samples (2.5 mm) it can be expected that the process of
diffusion of oxygen into the bulk of the material has an influence on the aging
behaviour. When tension samples are loaded to higher strains the thickness of
the samples is reduced due to the incompressibility of the material. This leads
to the effect that the surface-to-volume-ratio increases with increasing strain
which allows oxygen to diffuse into the bulk more easily (or in fact: reducing
the amount of bulk material) and by this speeding up aging processes.
But nevertheless, in the context of this paper, a phenomenological ap-
proach is followed which is open to be extended to include further processes
(like diffusion of oxygen) in the future. But within this work a purely
temperature-dependend aging process is simulated. So in the next step, the
aging at ε= 20 % is focussed using the relation:
Ed

−1
τd (θ) = νd e− R θ (22)

while the result for τd (θ) at ε= 20 % is shown in figure 7. A look at the model

7
10
τd [s]

6
10

5
10
50 60 70 80 90 100 110
θ [ C]
◦

Figure 7: τd at ε= 20 %

parameters displayed in table 2 yields that the parameter νd is independent

from the applied strain while the activation energy Ed differs.
Using these parameters, the relaxation-tests at ε= 20 % are simulated
and shown in figure 8. The dotted lines denote the experimental data while

17
 ε [%] Ed [J mol−1 ] νd [s−1 ]
20 1.0528 × 105 3.1508 × 109
35 1.0506 × 105 3.1508 × 109
50 1.0451 × 105 3.1508 × 109

Table 2: Model parameters for network degradation

the solid lines denote the simulation. For obtaining the model parameters,
only the marked points of the experiment are used.
Since the model is created to predict the evolution of the equilibrium
stress without considering viscoelastic effects like physical relaxation, the
beginning phase of the experiment (where the physical relaxation is domi-
nant) is not fitted. Furthermore, the upturn at the end of the test at 100◦ C
is ignored, as shrinkage is not taken into account.

1.8

1.6
11 (λ = 1.2) [MPa]

1.4

1.2

0.8
P−

0.6

0.4
60°C
0.2 80°C
100°C
0 3 4 5 6
10 10 10 10
Time [s]
Figure 8: Simulation of the relaxation-tests at ε= 20 %

The simulation shows good agreement with the experimental data. The
simulation for 60◦ C and 80◦ C differ a little bit more than for 100◦ C, but
taking into account that only 2 model parameters are needed to describe this
behaviour, this agreement is still very satisfying.

18
4.2. Intermittent monotonic tests
On the basis of the identification of the network degradation process, the
parameters for the network reformation process are identified in the next
step.
In a very similar manner, at first the chemical relaxation-times τr are de-
+
termined. For this purpose, experimental data for P11 (λ, dage , θage ) is needed
that contains only the evolution of the reformed network qr without the
stress-strain relation or the influence of the degenerating network. This is
obtained by using the stress at a certain strain ε1 (like shown in figure 6)
and subtracting the contribution of the degrading network. Here it has to be
kept in mind that this value is reduced by the network degradation process:

+ −
P11 (λ1 , dage , θage ) = P11 (λ1 , dage , θage ) − [1 − qd (dage , θage )] · P11,0 (λ1 ) (23)
−
while the values for P11,0 (λ1 ) are taken directly from the curve of the
unaged sample and qd is calculated using the parameters Ed and νd that
have been identified in the previous step. Using the relation

+ +
P11 (λ1 , dage , θage ) = qr (dage , θage ) · P11,∞ (λ1 ) (24)
with dage
−τ
qr (dage , θage ) = 1 − e r (θage ) (25)
and
−1
Er
τr (θage ) = νr e− R θ , (26)
nearly the same process as applied in chapter 4.1 is performed. By us-
+
ing the stress P11,∞ (λ1 ) of the fully reformed network (qr = 1) and the
relaxation-times τr (θ) as fitting parameters, the model parameters Er and
νr are obtained in a second step. Figure 9 shows the modified experimen-
+
tal data (triangles) for P11 as well as the simulation with the parameters
obtained (solid lines).
This procedure has been performed for different stretches λi , while every
set of curves at a certain strain yields a value for the stress of the fully
+
reformed network P11,∞ (λi ). This allows the construction of the stress-strain-
+
relation P11,∞ (λ) which has been fitted by use of a Mooney-Rivlin-model with
5 parameters. Assuming incompressible material behaviour the stress-strain-
relation for uniaxial tension reads as:

19
 5
60°C
4.5 80°C
100°C
4
11 (λ = 1.25) [MPa]
3.5

2.5

2
P+

1.5

0.5

0 4 5 6
10 10 10
dage [s]
Figure 9: Intermittent tension-test, stress at ε1 = 25 %, simulation

+ ∂IC ∂IC ∂IC

P11,∞ (λ) = d10 + d20 2(IC − 3) + d30 3(IC − 3)2
∂λ ∂λ ∂λ
  (27)
∂IC ∂IIC ∂IIC
+d11 (IIC − 3) + (IC − 3) + d01 .
∂λ ∂λ ∂λ
Figure 10 shows the result of the fit.
To obtain the stress-strain-relations of aged samples like presented in
figures 4 and 5 both networks are superimposed to:

− +
P11 (λ, dage , θage ) = [1 − qd (dage θage )] · P11,0 (λ) +qr (dage , θage ) · P11,∞ (λ) . (28)
The fact that this equation is rather simple compared to the equations
in chapter 3 arises from the special feature that the aging of these samples
is performed in the undeformed configuration (ε = 0 resp. λ = 1) and at
isothermal conditions (θ = const.).
−
The stress contribution of the degenerating network P11,0 (λ) is identified
from the monotonic tension test of the unaged sample by use of a Mooney-
Rivlin-model with the parameters c10 , c20 , c30 , c11 and c01 . Taking into

20
 8

11,∞ [MPa]
6

4
P+

0
0 10 20 30 40 50
strain ε [%]
Figure 10: Stress-strain relation of reformed network P+
11,∞

account that the rate of deformation has been kept rather small (ε̇ =0.1 %/s),
viscoelastic effects are neglected.
Finally the parameters are obtained as shown in table 3. Note that some
of the parameters for the Mooney-Rivlin-model are negative, but as the stress
is monotonously increasing with increasing strain in the whole regime of
deformation, this does not violate any physical laws.
The simulation of the intermittent tension-tests are shown in figure 11
and 12. It is observed that the stiffening of the material is simulated in
good relation to the experimental data. The quality of the fit for short aging
duration at high temperature (12 h at 100◦ C) as well as certain durations
at low temperature still leave some room for improvement. Furthermore
the simulation shows a better agreement at higher than at lower strains.
But besides this, it can be seen that the model is capable of simulating the
stiffening due to chemical aging pretty well.
When assessing the deviation of the simulation from the experiment, it
is important to remark that the aging in this simulation is assumed to be
driven only by temperature and that viscoelastic effects are not taken into
account. It is expected that an extension of the evolution-equations for qd
and qr to include other dependencies (e.g. diffusion, oxidation etc.) will lead

21
 6

4
P11 [MPa]

2
unaged
aged 12 hours at 100°C
1 aged 24 hours at 100°C
aged 48 hours at 100°C
aged 72 hours at 100°C
aged 6 days at 100°C
0
0 5 10 15 20 25 30 35 40 45 50
Strain [%]
Figure 11: Simulation of intermittent tension-tests, θage = 100◦ C

to an improvement.

c10 c20 c30 c11 c01 Ed [J mol−1 ] νd [s−1 ]

0.86 -0.47 0.74 -0.73 2.05 1.05 × 105 3.15 × 109

d10 d20 d30 d11 d01 Er [J mol−1 ] νr [s−1 ]

1.08 -0.46 1.58 -1.77 4.04 7.37 × 104 8.36 × 104

Table 3: Complete set of model parameters (cij and dij in [MPa] )

5. Conclusions
Concerning the previous chapters, the illustrated and explained procedure
for the detection and modelling of thermo-oxidative aging of elastomers shows
that the modelling approaches and methods can map the basic effects in an
appropriate manner.
Especially the experimental results have pointed out that both network
degradation and reformation processes have to be taken into account. Al-
though the continuous chemical relaxation experiment showed a softening of

22
 6 6

5 5

4 4
P11 [MPa]

P11 [MPa]
3 3

2 2 unaged
unaged aged 36 hours at 80°C
1 aged 72 hours at 60°C 1 aged 4 days at 80°C
aged 12 days at 60°C aged 6 days at 80°C
aged 21 days at 60°C aged 8 days at 80°C
0 0
0 10 20 30 40 50 0 10 20 30 40 50
Strain [%] Strain [%]
Figure 12: Simulation of intermittent tests, θage = 60◦ C (left) and θage = 80◦ C (right)

the material, it was experimentally observed that the network reformation

process is the dominating one. Hence, intermittent relaxation tests are of
essential importance.
The concept presented within this paper offers great opportunities for
developers of high-technology polymeric products. When thinking of elas-
tomeric products (e.g. tires, bearings or engine-mounts), it is highly im-
portant to simulate the dynamic behaviour, as these products are usually
optimised for well-defined working conditions. A change in stiffness can lead
to large variations of different kinds. The most important thing could be
a change in the eigenfrequencies. Especially when thinking about vibro-
acoustic properties, the evolution of the material behaviour can lead to se-
vere miscalculations during the design of the product. One can imagine that
all requirements are fulfilled directly after fabrication, but chemical aging
processes during operation cause the product to run out of specification.
The theoretical concept formulated in chapter 3 offers tools not only to
estimate the lifetime of the product, but to consider changes in the material
behaviour during the operation time. By this the designer is enabled to
optimise the properties in a way that the specifications of the product are
fullfilled over a large domain of time and thus improving the design and the
development in a significant way.
Of course, especially in the view of industrial application, the concept has
to be expanded. Concerning the results of the paper, the following steps are
planned to be done:

23
 In a further study, the influence of aging on the viscoelastic material
behaviour has to be investigated. If this influence is negligible, the short-
term relaxation or creep behaviour of the elastomers can be modelled with the
concept of finite viscoelasticity. If thermo-oxidative aging strongly affects the
viscoelasticity, further internal variables have to be introduced, with which
the aging mechanisms of the short-time behaviour can be represented.
Another important point is the capture and simulation of components
with finite dimensions, in which an inhomogeneous oxygen concentration is
available. From the experimental side, the experimental techniques of ana-
lytical chemistry have to be taken into account. By using thermogravimetric
analysis in combination with infrared spectroscopy (TG-IR), it is possible to
analyse the composition of materials in a single measurement.
Furthermore, studies of elastomers by using the micro-Attenuated Total
Reflection Fourier Transform Infrared Spectroscopy (µ-ATR-FTIR) provide
information on the oxidation of the polymer and thus on the progression of
aging effects. The oxygen uptake itself has to be determined on laboratory
samples by using a specially constructed measuring apparatus. In combi-
nation, µ-ATR-FTIR analysis has to be performed in order to analyse the
consumed oxygen which reacted with the polymer.
Ultimately, it will be necessary to investigate the inhomogeneous aging on
samples of finite thickness. For this purpose, sufficiently thick elastomer sam-
ples are artificially aged under laboratory conditions in the ambient medium
and under elevated temperatures. At predetermined times, one of the sam-
ples is removed from the medium, cooled down to room temperature and
cutted into slices by using a microtome. The individual slices are then me-
chanically, calorically and chemically tested. Thus, it will be possible to
detect and model the inhomogeneous aging of the material.
Based on the mentioned experiments a reaction-diffusion equation has to
be formulated,
ρ0 ċ − α Div Grad c − ĉ = 0 , (29)
and coupled with the existing model. This can be done by using the concept
of multiphase continuum mechanics (Johlitz and Lion, 2012). Therefore the
main focus is based on the formulation of the reaction term ĉ(θ, qd , qr , ...) as
well as on the diffusion coefficient α(ĉ, θ, qd , qr , ...). Furthermore, the two pre-
viously established evolution equations for the degeneration and reformation
processes have to be formulated in an appropriate way as a function of the
oxygen concentration, i.e. q̇d = f (c, θ, qd , qr , ...), q̇r = f (c, θ, qd , qr , ...), in or-

24
der to detect the inhomogeneous aging processes as they occur in structures
of finite thickness.
Last but not least, it has to be emphasised that the presented concept
allows the extension to include the phenomena mentioned above in an easy
and modular manner. Thus, the concept can serve as a reliable basis for
highly sophisticated approaches to model the manifold aspects of chemical
aging.

References
Ames, N.M., Srivastava, V., Chester, S.A., Anand, L., 2009. A thermo-
mechanically coupled theory for large deformations of amorphous poly-
mers. part ii: Applications. Int. J. Plasticity 25, 1495–1539.
Anand, L., Ames, N.M., Srivastava, V., Chester, S.A., 2009. A thermo-
mechanically coupled theory for large deformations of amorphous poly-
mers. part i: Formulation. Int. J. Plasticity 25, 1474–1494.
Andrews, R., Tobolsky, A., Hanson, E., 1946. The theory of permanent set
at elevated temperatures in natural and synthetic rubber vulcanizates. J.
Appl. Phys. 17, 352–361.
Audouin, L., Langlois, V., Verdu, J., de Bruijn, J., 1994. Review: Role of
oxygen diffusion in polymer ageing: kinetic and mechanical aspects. J.
Mat. Sci. 29, 569–583.
Ayoub, G., Naı̈t-Abdelaziz, M., Zaı̈ri, F., Gloaguenb, J., Charriere, P., 2012.
Fatigue life prediction of rubber-like materials under multiaxial loading us-
ing a continuum damage mechanics approach: Effects of two-blocks loading
and r ratio. Mech. Mater. 52, 87–102.
Ayoub, G., Zaı̈ri, F., Naı̈t-Abdelaziz, M., Gloaguen, J., Kridli, G., 2013.
A visco-hyperelastic damage model for cyclic stress-softening, hysteresis
and permanent set in rubber using the network alteration theory. Int. J.
Plasticity , in press.
Becker, G., Braun, D., 1996. Kunststoff Handbuch 1, Die Kunststoffe-
Chemie, Physik, Technologie. Carl Hanser Verlag, München.
Blum, G., Shelton, J., Winn, H., 1951. Rubber oxidation and ageing studies.
Ind. Eng. Chem. 43.

25
Bouvard, J., Francis, D., Tschopp, M., Marin, E., Bammann, D., Horste-
meyer, M., 2013. An internal state variable material model for predicting
the time, thermomechanical, and stress state dependence of amorphous
glassy polymers under large deformation. Int. J. Plasticity 42, 168–193.

Budzien, J., Rottach, D., Curro, J., Lo, C., Thompson, A., 2008. A new
constitutive model for the chemical ageing of rubber networks in deformed
states. Macromolecules 41, 9896–9903.

Duarte, J., Achenbach, M., 2007. On the modelling of rubber ageing and
performance changes in rubbery components. Kaut. Gummi Kunstst. 60,
172–175.

Dunn, J., Scalan, J., Watson, W., 1959. Stress relaxation during the thermal
oxidation of vulcanized natural rubber. T. Faraday Soc. 55, 667–675.

Dunwoody, R., 1970. A thermomechanical theory of diffusion in solid-fluid

mixtures. Arch. ration. Mech. An. 38, 348–371.

Ehlers, W., Ellsiepen, P., 1998. PANDAS: Ein FE-System zur Simulation
von Sonderproblemen der Bodenmechanik, in: Wriggers, P., Meißner, U.,
Stein, E., Wunderlich, W. (Eds.), Finite Elemente in der Baupraxis: Mod-
ellierung, Berechnung und Konstruktion, Beiträge zur Tagung FEM ’98
an der TU Darmstadt am 5. und 6. März 1998. Ernst & Sohn, Berlin, pp.
431–400.

Ehrenstein, G., Pongratz, S., 2007. Beständigkeit von Kunststoffen. Carl

Hanser Verlag.

Flory, P.J., 1961. Thermodynamic relations for hight elastic materials. T.

Faraday Soc. 57, 829–838.

Horstemeyer, M.F., Bammann, D.J., 2010. Historical review of internal state

variable theory for inelasticity. Int. J. Plasticity 26, 1310–1334.

Hossain, M., Possart, G., Steinmann, P., 2009. A finite strain framework for
the simulation of polymer curing. Part I: elasticity. Comput. Mech. 44,
621–630.

Hutchinson, J., 1995. Physical ageing of polymers. Prog. Polym. Sci. 20,
703–760.

26
Johlitz, M., 2012. On the representation of ageing phenomena. J. Adhesion
88, 620–648.

Johlitz, M., Lion, A., 2012. Chemo-thermomechanical ageing of elastomers

based on multiphase continuum mechanics. Continuum Mech. Thermodyn.
, published online, DOI: 10.1007/s00161–012–0255–8.

Laiarinandrasana, L., Besson, J., Lafarge, M., Hochstetter, G., 2009. Tem-
perature dependent mechanical behaviour of pvdf: Experiments and nu-
merical modelling. Int. J. Plasticity 25, 1301–1324.

Lion, A., Johlitz, M., 2012. On the representation of chemical ageing of

rubber in continuum mechanics. Int. J. Solids Struct. 49, 1227–1240.

Lion, A., Peters, J., Kolmeder, S., 2011. Simulation of temperature history-
dependent phenomena of glass-forming materials based on thermodynam-
ics with internal state variables. Thermochim. Acta , in press.

Loeffel, K., Anand, L., 2011. A chemo-thermo-mechanically coupled theory

for elastic–viscoplastic deformation, diffusion, and volumetric swelling due
to a chemical reaction. Int. J. Plasticity 27, 1409–1431.

Lustig, S., Caruthers, J., Peppas, N., 1992. Continuum thermodynamics and
transport theory for polymer-fluid mixtures. Chem. Eng. Sci. 47, 3037–
3057.

Mooney, M., 1940. A theory of large elastic deformation. J. Appl. Phys. 11,
582–592.

Ore, S., 1959. A modification of the method of intermittent stress relaxation

measurements on rubber vulcanisates. J. Appl. Polym. Sci. 2, 318–321.

Pochiraju, K., Tandon, G., 2006. Modeling thermo-oxidative layer growth in

high-temperature resins. J. Eng. Mater-T ASME 128, 107–116.

Quang, N., Samohýl, I., Thoang, H., 1988. Irreversible (rational) thermo-
dynamics of mixtures of a solid substance with chemical reacting fluids.
Collect. Czech. Chem. Commun. 53, 1620–1635.

Samohýl, I., Šı́pek, X.N.M., 1985. Irreversible (rational) thermodynamics of

fluid-solid mixtures. Collect. Czech. Chem. Commun. 50, 2346–2363.

27
Scalan, J., Watson, W., 1957. The interpretation of stress relaxation mea-
surements made on rubber during ageing. T. Faraday Soc. 54, 740–750.

Shaw, J., Jones, S., Wineman, A., 2005. Chemorheological response of elas-
tomers at elevated temperatures: experiments and simulations. J. Mech.
Phys. Solids 53, 2758–2793.

Simo, J.C., Taylor, R.L., 1982. Penalty function formulations for incompress-
ible nonlinear elastostatics. Comp. Meth. Appl. Mech. Eng. 35, 107–118.

Smith, L., 1993. The language of rubber: an introduction to the specification

and testing of elastomers. Butterworth-Heinemann publication house.

Srivastava, V., Chester, S.A., Ames, N.M., Anand, L., 2010a. A thermo-
mechanically-coupled large-deformation theory for amorphous polymers in
a temperature range which spans their glass transition. Int. J. Plasticity
26, 1138–1182.

Srivastava, V., Chester, S.A., Anand, L., 2010b. Thermallyactuatedshape-

memorypolymers:experiments,theory, and numericalsimulations. J. Mech.
Phys. Solids 58, 1100–1124.

Steinke, L., Veltin, U., Flamm, M., Lion, A., Celina, M., 2011. Numerical
analysis of the heterogeneous ageing of rubber products, in: Jerrams, S.,
Murphy, N. (Eds.), Constitutive Models for Rubber VII, pp. 155–160.

Tobolsky, A.V., 1967. Mechanische Eigenschaften und Struktur von Poly-

meren. Berliner Union Stuttgart.

Tobolsky, A.V., Prettyman, I.B., Dillon, J.H., 1944. Stress relaxation of

natural and synthetic rubber stocks. J. Appl. Phys. 15, 380–395.

28