Materials and Design 58 (2014) 65–73

Contents lists available at ScienceDirect

Materials and Design

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

Experiment and modeling of uniaxial tension fatigue performances

for filled natural rubbers
Wen-Bin Shangguan a,b,⇑, Xiao-Li Wang a,b, Jian-Xiang Deng a,b, Subhash Rakheja c,1,
Xiao-Yong Pan d, Bin Yu e
a
State Key Laboratory of Automotive Simulation and Control, Jilin University, Changchun 130012, PR China
b
School of Mechanical and Automotive Engineering, South China University of Technology, Guangzhou 510640, PR China
c
School of Mechanical & Automotive Engineering, South China University of Technology, Guangzhou, PR China2
d
Ningbo Tuopu Group Co., Ltd., Ningbo 315800, PR China
e
Pan Asia Technical Automotive Center Co., Ltd., Shanghai 201201, PR China

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

Article history: A tension fatigue model of filled natural rubbers is investigated to study the contributions of two key fac-
Received 5 August 2013 tors, namely, the damage parameter and the specimen geometry used in the fatigue experiment. The uni-
Accepted 18 January 2014 axial tension fatigue experiments were carried out for three filled natural rubber specimens with
Available online 25 January 2014
different geometry: a dumbbell simple tension specimen (STS), a dumbbell cylindrical specimen (DCS),
and a hollow cylindrical specimen (HCS). The commonly used damage parameters for fatigue life predic-
Keywords: tion are discussed. The fatigue life prediction models are formulated using the measured tension fatigue
Fatigue life prediction
life of the STS together with different damage parameters. The effectiveness of the models is established
Fatigue testing
Damage parameter
in terms of a correlation coefficient characterizing the error between the measured and predicted fatigue
Filled natural rubber lives. It is concluded that all the damage parameters considered in the study can effectively estimate the
tension fatigue life with correlation coefficients exceeding 0.9. The fatigue life model formulated for the
STS was also found to be appropriate for predicting the fatigue life of specimens with different geometry
(DCS and HCS) suggesting that the relationship between the tension fatigue life and the damage param-
eters is independent of the specimen geometry. One may thus conduct tension fatigue tests with STS
alone in order to model the tension fatigue life of rubbers with alternate geometry.
Ó 2014 Elsevier Ltd. All rights reserved.

1. Introduction
Owing to their superior ability to withstand large strains with-
out permanent deformations, rubbers are widely used in many
engineering applications, such as engine powertrain mounts, sus-
pension bushes, exhausting isolators, seals and so on [1]. Rubber
components are usually subjected to substantial static and dy-
namic loads, and often fail due to nucleation and growth of defects
or cracks. Prevention of such mechanical fatigue failures necessi-
tates thorough understanding of the deformation mechanisms of
the rubber materials during cyclic loading, so as to predict the fa-
tigue life of rubber components more accurately.
The long-term durability of rubber components is strongly
dependent on a number of material and geometry-related factors
⇑ Corresponding author at: School of Mechanical and Automotive Engineering,
South China University of Technology, Guangzhou 510640, PR China. Tel.: +86
18820072208.
E-mail address: shangguanwb99@tsinghua.org.cn (W.-B. Shangguan).
1
On leave from Concordia University, Montreal, Canada.
2 reasons for choosing a particular damage parameter are seldom
Present address.
in a relatively complex manner. Mars and Fatemi [2] and Mars
[3] have presented a comprehensive review about factors affecting
the fatigue life of rubbers, together with the key issues that need
special attention for designs of elastomeric structures. These in-
cluded factors related to mechanical loading and its history, envi-
ronment, rubber formulation effects, and dissipative aspects of
the constitutive response of rubbers. Among the various factors
influencing the fatigue behavior of rubber, the factor related to
mechanical loading history has been most widely investigated in
many reported studies [4–14].
Uniaxial fatigue [4–10] continues of significant importance
especially during the initial design stages that involve selection
of the rubber compound for optimal fatigue resistance [4–10],
although multi-axial fatigue [11–14] occurs more frequently in
rubber components during service. Characterization of fatigue
loading experienced locally by the material firstly requires deter-
mination of the damage parameters. Reported studies, however,
have employed widely different damage parameters for prediction
of fatigue life of rubbers under uniaxial loads [4–10], although

http://dx.doi.org/10.1016/j.matdes.2014.01.035
0261-3069/Ó 2014 Elsevier Ltd. All rights reserved.
66 W.-B. Shangguan et al. / Materials and Design 58 (2014) 65–73
explained. Since the nonlinearity and finite deformation character-
istics of rubber materials, several kinds of strain measures can be
used as damage parameters, such as Green–Lagrange strain,
Almansi–Euler strain, engineering strain, logarithmic strain and
the stretch ratio. While Kim et al. [10] suggested that the Green–
Lagrange strain is more appropriate to be used as a damage
parameter for estimating fatigue life of rubbers, other studies have
used alternate strain measures as damage parameters [4–9]. Only
limited information could be found in the literature on relative
merits or limitations of different damage parameters for charactering
the fatigue behaviors of rubbers, even under uniaxial loading.
Owing to the highly nonlinear properties of rubbers, the mea-
surements with simple test specimens under constant-amplitude
uniaxial loads are vital for fundamental understanding of multi-ax-
ial fatigue mechanisms. The data obtained from uniaxial fatigue
experiments are also essential for developing reliable fatigue life
prediction models. Furthermore, the effects of specimen geometry
on the fatigue properties also need to be investigated.
This study is aimed at analysis of different damage parameters
used in uniaxial fatigue of rubbers by investigating the fatigue life
estimation models using different damage parameters based on
measured fatigue life data. Furthermore, the specimen geometry
effects are investigated considering three types of specimens
of varying geometry in the fatigue experiments in order to
establish the specimen geometry-independency of the fatigue life
prediction. The material and configurations of the three specimens
are initially described together with the measurement methods,
the criterion of crack nucleation fatigue life and the test loads.
The three specimens used for the uniaxial fatigue tests included
the dumbbell simple tension specimen (STS) (ASTM: D4482-11),
the dumbbell cylindrical specimen (DCS) [10,11], and the hollow
cylindrical specimen (HCS) [15].
The different damage parameters are subsequently described,
and classified as strain-based and energy-based damage parame-
ters. The strain-based parameters considered in the study include
the peak of the maximum (1st) principal strain (Green–Lagrange
strain, Almansi–Euler strain, engineering strain, logarithmic strain
and stretch ratio), and peak of the octahedral shear strain. The en-
ergy-based damage parameters employ two different strain energy
density peaks derived from the stress–strain data during loading
and unloading.
The fatigue life prediction models are formulated using differ-
ent damage parameters and measured tension fatigue life data
for the STS specimen, while the effectiveness of the models is char-
acterized by a correlation coefficient relating the predicted fatigue
life with the measured data. The relative performance of the fati-
gue life models employing different damage parameters is subse-
quently assessed. The specimen geometry-dependence of the
prediction model is finally evaluated using the test data acquired
for the other two specimens (DCS and HCS), and applicability of
the fatigue life prediction models for rubbers of different geometry
is highlighted.
2. Experiment design
2.1. Rubber materials and specimens
Fatigue life experiments were designed to investigate the ten-
sion fatigue of a particular type of filled natural rubber. Three rub-
ber specimens of different geometry, molded from a filled natural
rubber (NR) compound, were considered for the experiments,
including a dumbbell simple tension specimen (STS), a dumbbell
cylindrical specimen (DCS), and a hollow cylindrical specimen
(HCS). Fig. 1 illustrates the geometric configurations of the selected
specimens. The STS specimen was compression molded and cured
at 150 °C for 7 min. The DCS and the HCS specimens were transfer
molded, and cured at 150 °C for 10 and 8 min, respectively. The
constituents and mechanical properties of the specimens are listed
in Tables 1 and 2, respectively.
The STS specimen was designed in accordance with the ASTM
standard (ASTM: D4482-11), while the DCS and HCS specimens
were designed as described in [10,11,15]. The STS could be used
for measurements of fatigue life only in the tension direction;
while the fatigue experiments on DCS and the HCS specimens
could be performed under tension and compression, either inde-
pendently or simultaneously. The fatigue life tests with the STS,
however, can be performed in a more efficient and cost-effective
way compared to those with DCS and HCS specimens, since rela-
tively fewer samples of DCS and HCS could be tested simulta-
neously with most of the available test machines.
2.2. Fatigue measurement methods for rubber materials
Owing to the visco-elastic behavior of the rubber materials,
considerable accumulation of ratcheting strain (cyclic creep) may
occur under load (stress) controlled fatigue tests, which is known
to be detrimental to the fatigue life [16]. The ratcheting strain ef-
fect could be circumvented through displacement (strain) con-
trolled fatigue tests, where the mean stress relaxation tends to
saturate in relatively short duration. The loading could thus be re-
tained near the same steady level during the fatigue test [16]. The
fatigue tests in this study were thus conducted under controlled
sinusoidal displacement at a frequency of 5 Hz.
Fatigue tests on the STS specimens were performed with the
EKT-2102-DF Demattia test machine [5] (Machine A), while those
for the DCS and HCS specimens were carried out on a MOOG dura-
bility test machine (Machine B) and a Bose EFL3500 test machine
(Machine C), respectively. Machine A could perform fatigue tests
with twenty samples simultaneously under identical displace-
ment, as shown in Fig. 2(a). The test machine B, however, could
employ only 3 samples of the DCS specimen simultaneously using
the specially designed clamp, as shown in Fig. 2(b), while a single
sample of the HCS specimen could be tested on machine C at a
time, as shown in Fig. 2(c). The fatigue experiments with the DCS
and HCS specimens are thus deemed inefficient and costly. All
the tests were conducted at the laboratory temperature (23 °C),
and a fan was used to cool the samples in order to minimize the
contributions due to possible thermal load. A loading cycle counter
was used to record the number of load cycles. The cycle counter,
however, continued to operate even after the total fracture of a
sample. A video recorder was thus also used to determine the in-
stant complete fracture of a sample and thus the number of load
cycles to failure.
2.3. Criterion of nucleation fatigue life of rubber materials
The total fatigue life of a specimen is the summation of the
crack nucleation life and the crack growth life. The fatigue analysis
approaches for the two stages, however, differ. The approach for
predicting crack nucleation life is based on the continuum
mechanics, while the crack growth life is predicted from the frac-
ture mechanics. In this study, the method of crack nucleation fati-
gue life prediction is used, which requires the use of a criterion to
identify the occurrence of crack nucleation. While the length of the
crack is determined arbitrarily [17], it strongly depends on the
specimen geometry and size. The occurrence of crack nucleation,
however, is always related to a significant decrease in the sample
stiffness.
The measured data obtained with different samples of a rubber
specimen under a given loading condition generally reveal some-
what consistent trends, although substantial variations in the data
 W.-B. Shangguan et al. / Materials and Design 58 (2014) 65–73 67

(a) STS (b) DCS (c) HCS

Fig. 1. Configurations of the three test specimens.

Table 1
2.4. Test loads in the fatigue experiment
Constituents of the rubber compound.

Ingredient PHRa Weight % The fatigue experiments were performed under the loading
Natural rubber 100 68.45 condition corresponding to minimum strain, being equal to zero.
Zinc oxide 6 4.11 Table 3 summarizes different loading conditions used in the exper-
N330 carbon black 10 6.84
iments with the three types of specimens considered in the study
N550 carbon black 10 6.84
Naphthenic oil 7 4.79 together with the minimum (dmin) and peak (dmax) measured dis-
Others 13.1 8.96 placements. The load levels are selected such that the strain is
Total parts per hundred rubber 146.1 100 within the range observed for an engine mount. The table also pre-
a
Parts per hundred rubber, by weight.
sents the initial (L0) and maximum (Lmax) gauge length of the STS
specimen under different loading cases, where the maximum
gauge length corresponds to peak displacement dmax.
Table 2 It should be noted that the gauge lengths of DCS and HCS spec-
Mechanical properties of the studied natural rubber. imens were not measured due to varying cross-section of these
Density (g/ Shore efail UTS 300% Elongation modulus
specimens along the axial direction. The measured gauge length
cm3) hardness (%) (MPa) (MPa) of the STS specimen is used to calculate the engineering strain,
1.05 48 620 26 4.4
while the strains of the DCS and HCS specimens at the critical loca-
tion are evaluated through finite element analysis (FEA). The gauge
length of the STS is regarded as the distance between the two
could also be observed. Taking the DCS as an example, the axial markers placed at each end of the STS. The positions of the markers
stiffness evolution of five samples subjected to identical loading are selected to ensure that the stress state within the gauge length
is illustrated in Fig. 3 as a function of the number of load cycles. is under simple tension. The engineering strain is subsequently cal-
It is seen that the axial stiffness of the DCS samples decreases con- culated from the measured gauge length before and after the load-
siderably during the initial few loading cycles due to the Mullins ing. The measured data revealed slight differences between the
effect. Following this initial transient period, the axial stiffness given and the measured displacements for some of the loading
tends to decrease slightly until the sample is completely broken, cases. The amplitude (da) and mean (dm) values of the displace-
which is observed from the sharp drop in the stiffness. Similar ment corresponding to different loading cases, calculated from dmin
trends were also observed for the other two types of specimens and dmax, are also summarized in Table 3.
considered in the study under all the loading conditions. From The experiments were conducted on multiple samples of each
the measured data, it may thus be concluded that the crack nucle- rubber specimen so as to examine repeatability of the measure-
ation time closely corresponds with total test duration to complete ments across the samples and to reduce the scatter in the mea-
fracture, and is greater than 90% of the total test time. In the fatigue sured fatigue life data. The fatigue experiments for the STS
experiments, the nucleation fatigue life of a sample is thus consid- specimen were conducted considering 10 different loading cases
ered as the number of cycles to complete fracture, which is consis- and a total of twenty samples for each loading case. The tests on
tent with that reported in [14]. the 20 samples were conducted simultaneously under each loading

(a) STS [5] (b) DCS (c) HCS

Fig. 2. Fatigue experiment spots for the three kinds of specimens.
68 W.-B. Shangguan et al. / Materials and Design 58 (2014) 65–73

30 fatigue life estimation models using different strain-based damage

28 parameters, however, may lead to different fatigue life estimations.
In this section, the definitions of different principal strain mea-
26
Axial stiffness (N/mm)

sures for finite deformation are briefly discussed in view of their

24 applications as damage parameters and effectiveness in estimating
22 the fatigue life. The octahedral shear strain and energy-based dam-
age parameters are also described.
20
Sample 1
18 Sample 2
Sample 3 3.1. Definition of the principal strain measures
16 Sample 4
Sample 5
14 The maximum (or 1st) principal Green–Lagrange strain eG and
12 the 1st principal Almansi–Euler strain eA are, respectively, given
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 by [18]:
5
Number of load cycles N (cycle) x 10
k2  1
Fig. 3. Axial stiffness peak evolution with number of cycles for five DCS samples eG ¼ ð1Þ
under one identical loading.
2

k2  1
condition. The fatigue experiments with the DCS specimen in- eA ¼ ð2Þ
volved measurements on 12 samples under each of the seven load- 2k2
ing conditions listed in Table 3. Three samples of this specimen where k ¼ l=L is the 1st principal stretch ratio, the ratio of deformed
could be tested simultaneously using the specially designed clamp length l to undeformed length L of a bar.
shown in Fig. 2(b). The experiment with the HCS specimen in- The 1st principal engineering strain eE is evaluated from the 1st
volved 3 samples for each of the five load cases listed in Table 3. principal stretch ratio k, such that [18]:

3. Damage parameters eE ¼ k  1 ð3Þ

The 1st principal logarithmic strain eL is obtained from [18]:

A damage parameter for predicting crack nucleation life should
Z  
be formulated in terms of continuum mechanics quantities in or- l
dl l
der to be combined with standard finite element method in engi- eL ¼ ¼ ln ¼ lnðkÞ ð4Þ
L l L
neering applications [12]. Among the continuum mechanics
quantities, the strain-based parameters are often taken as the dam- The strain measures described in Eqs. (1)–(4) are related as shown
age parameters for rubber materials to correlate with the crack in Fig. 4. It is seen that all of the strain measures (eG, eA, eE and eL) are
nucleation life. Considering the nonlinear and finite deformation identical for stretch ratio, 1:0 6 k 6 1:2, where all of the measures
characteristics of filled natural rubbers, a number of different converge to the small strain definition, eE = Dl/L. A simple Taylor
strain measures have been defined to describe finite deformation series expansion validates this result for l  L [18]. Considerable dif-
of the rubber. These include the Green–Lagrange strain, Almansi– ferences among all the strain measures, however, are apparent un-
Euler strain, engineering strain, logarithmic strain, the stretch der large deformation, k > 1:2. The results in Fig. 4 further show
ratio, and more [18]. Theoretically, all of the strain measures could that the relationships yield eG > eE > eL > eA. Different strain mea-
serve as damage parameters for estimating the fatigue life. The sures would thus yield diverse fatigue life predictions. It is thus vital

Table 3
The loading cases and control parameters used in the uniaxial tension fatigue tests.

Specimen type Loading case Mean measured displacement d (mm) Gauge length (mm)
Valley dmin Peak dmax Amplitude da Mean dm The initial length L0 The maximum length Lmax
STS 1 0 45.76 22.88 22.88 25.00 70.76
2 0 40.20 20.10 20.10 25.00 65.20
3 0 33.82 16.91 16.91 25.00 58.82
4 0 30.67 15.34 15.34 25.00 55.67
5 0 29.71 14.86 14.86 25.00 54.71
6 0 24.94 12.47 12.47 25.00 49.94
7 0 23.26 11.63 11.63 25.00 48.26
8 0 20.51 10.26 10.26 25.00 45.51
9 0 17.77 8.89 8.89 25.00 42.77
10 0 15.45 7.73 7.73 25.00 40.45
DCS 11 0 38.76 19.38 19.38 – –
12 0 36.07 18.04 18.04 – –
13 0.36 31.64 15.64 16.00 – –
14 0 26.41 13.21 13.21 – –
15 0.36 21.62 10.63 10.99 – –
16 0 18.64 9.32 9.32 – –
17 0 16.28 8.14 8.14 – –
HCS 18 0 6.80 3.40 3.40 – –
19 0 5.45 2.73 2.73 – –
20 0 4.26 2.13 2.13 – –
21 0 3.70 1.85 1.85 – –
22 0 3.10 1.55 1.55 – –
 W.-B. Shangguan et al. / Materials and Design 58 (2014) 65–73 69

to determine, which strain measure is most effective for applica- 5

tions in the fatigue life models. εE
4
εG
3.2. Definition of the octahedral shear strain
εA
3

Strain ε
εL
The octahedral shear strain is a strain-based invariant, obtained
from the Green–Lagrange strain tensor E using the following rela- 2
tion [19]:
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 1
2
eo ¼ ðE11  E22 Þ2 þ ðE22  E33 Þ2 þ ðE33  E11 Þ2 þ 6ðE212 þ E223 þ E231 Þ
3pffiffiffi
  0
2  2 1 1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 2.8 3
¼ k  Stretch ratio λ
3  k
ð5Þ Fig. 4. Relationships between different strain measures and the stretch ratio.

where Eij (i, j = 1, 2, 3) are the strain components of the Green–

Lagrange strain tensor E. It should be noted that the octahedral
shear strain given in Eq. (5) differs from that by Mars [13]. The value suggested a highly correlated power-law relationship between
of octahedral shear strain based on Eq. (5) is exactly twice of that the peak SED and peak loading strain, eE,max (see Fig. 6), given by:
given by Mars [13]. W L;max ¼ 0:6521ðeE;max Þ1:5789 ; r 2 ¼ 0:9999 ð8Þ

3.3. Strain energy density and its calculation

W U;max ¼ 0:5889ðeE;max Þ1:5233 ; r2 ¼ 0:9966 ð9Þ
Strain energy density (SED) represents the energy stored in a High correlations observed in Fig. 6 and above equations sug-
material due to its deformation. In the context of fracture mecha- gest that the peak SED can be predicted from the power laws with
nism, the SED is a measure of the energy release rate of naturally a high degree of confidence for both loading as well as unloading
existing defects leading to crack growth under some conditions cycles, which may serve as two different damage parameters.
[7]. The SED is thus another commonly used parameter in rubber The results further suggest that for a prescribed engineering strain,
fatigue analysis. The incremental definition of SED is expressed the SED can be directly obtained from Eqs. (8), (9) in a highly effi-
as [18]: cient manner unlike the relatively demanding integration and FEA
methods.
dW ¼ r : de ð6Þ
where the stress tensor r and the strain tensor e are conjugates in 4. Experimental results and discussions of fatigue life models
view of the energy. The SED can be obtained from integration of Eq. with different damage parameters
(6). For a uniaxial stress state, the SED can be obtained as:
Z eE 4.1. Experimental results and calculations of damage parameters
W¼ rE de ð7Þ
0 The average measured fatigue lives of the STS samples corre-
where eE and rE are the engineering strain and stress, respectively. sponding to the 10 loading cases considered in the experiments
(Table 3) are summarized in Table 4 together with the different
3.3.1. Two methods for calculating strain energy density damage parameters obtained from the relations defined in Sec-
Mars [13] measured engineering stress and engineering strain tion 3. The average value of the ratio of maximum and minimum
of a simple tension specimen corresponding to different peak engi- number of loading cycles to failure, Nmax and Nmin, are also pre-
neering strains, and proposed a power-law relation between the sented in the table for each loading case. This ratio provides a mea-
peak SED and the peak engineering strain. The SED was calculated sure of the scatter in the test data. From the results, it is evident
through numerical integration of the engineering stress–strain that the average ratio ranges from 1.43 to 2.85 for the 10 load cases
curves along the loading direction. Dizon et al. [20], on the other considered in the study, which is smaller than the range reported
hand, suggested that numerical integration of the unloading paths by Mars [13].
of the engineering stress–strain curves are appropriate for obtain- It has been shown that the fatigue life data generally follow a
ing the SED. It is thus necessary to compare the differences be- log-normal distribution [21]. This assumption has also been rec-
tween the two methods for calculating SED, and their ommended in the ASTM D4482 standard (ASTM: D4482-11). The
correlations with the estimated fatigue life. average life data listed in Table 4 represents the geometric mean
Filled natural rubbers, invariably, exhibit Mullins effect [13], assuming the log-normal distribution, such that:
which is observed as the initial transient softening effect in the X
n

stress–strain curve. The Mullins effect greatly affects the rubber fa- log10 ðNave Þ ¼ log10 ðNi Þ=n ð10Þ
tigue properties. In case of cyclic loading, nearly steady engineer- i¼1

ing stress–strain data obtained after a few initial cycles are where n is the number of samples tested under each load case, and
considered for the fatigue analyses. In this study, the stress–strain Ni is the measured fatigue life of the ith sample.
data obtained after the initial five cycles are used for calculating
the SED. 4.2. Relative assessments of fatigue life models with different damage
The measured steady stress–strain curves under loading and parameters
unloading corresponding to four different peak strains are illus-
trated in Fig. 5. The peak SED associated with the loading (WL,max) The relation between a damage parameter and the fatigue life is
and unloading (WU,max) portions of the stress–strain curves are described by the following power law [21]:
subsequently computed for each loading case through numerical
integration using the Trapz function in Matlab. The results Pd ¼ KðNf Þb ð11Þ
70 W.-B. Shangguan et al. / Materials and Design 58 (2014) 65–73
Engineering strain σE (MPa)
2.5
ε E,max =0.5
2 ε E,max =1.0
ε E,max =1.5
1.5
ε E,max =2.0
1 1
0.5
0 0
0 0.5 1 1.5
Engineering strain εE
Fig. 5. Steady stress–strain curves corresponding to different peak strain levels.
where Pd is the damage parameter that may be expressed by the
peak of the 1st principal strain, the peak octahedral shear strain,
or the peak SED, etc. The constant K and the exponent b relate to
the material parameters, and are identified through minimization
of error between the computed damage parameter and that ob-
tained from the measured data. Non-linear optimization techniques
are often used to minimize the sum of squared residuals between
the experimental data and the power law relation in Eq. (11) with
respect to the constants K and b. The correlation coefficient r2 is
widely used to measure the degree of fit between the calculated
and the measured fatigue lives of rubbers [4–14].
In this study, the least-square error minimization method is
used to identify the constants K and b, using different damage
parameters. The relative effectiveness of the fatigue life estimation
models based on different damage parameters is subsequently as-
sessed in terms of the estimated and the average measured data.
Table 5 summarizes the material constants identified using differ-
ent damage parameters together with the corresponding r2 values.
The results are discussed in the following sub-sections to highlight
relative merits and limitations of the different fatigue life estima-
tion models.
4.2.1. Fatigue life models based on peak of the 1st principal strains
From the measured data obtained with STS samples, it is evi-
dent that the peak of 1st principal stretch ratio, kmax , ranges from
1.62 to 2.83 for the 10 loading cases considered. In this range, sub-
stantial differences exist in the different strain measures (eG, eA, eE
and eL), as seen in Fig. 5. Notable differences would thus be ex-
pected in the fatigue life estimations of the models based on differ-
ent strain measures as the damage parameters. The fatigue life
models employing five different strain measures are formulated
for evaluating the relative prediction errors of the models. Consid-
ering the strain measure as the damage parameter, Eq. (11) is re-
written as:
e1;max ¼ KðNf Þb
where e1,max is the peak of 1st principle strain representing kmax ,
eG,max, eA,max, eE,max or eL,max. The material parameters, K and b, for
each model are subsequently identified using the least squares
method.Fig. 7 illustrates the correlations of the fatigue life predic-
tion models using different strain measures with the measured fa-
tigue life data for the STS samples. The results in Fig. 7 and
Table 5 suggest highest correlation of the fatigue life model based
on the peak of 1st principle Green–Lagrange strain (r2 = 0.9966), fol-
lowed by that based on the peak of 1st principle engineering strain
(r2 = 0.9952). The lowest correlation is obtained for the fatigue life
model employing the peak of 1st principle Almansi–Euler strain
as the damage parameter (r2 = 0.9576). These results are reasonable
considering that the Green–Lagrange strain is preferred to describe
finite deformations in theory, and are consistent with those
2.5
Measured WL,max
SED peak Wmax (MPa)
2 Measured WU,max
r 2=0.9999
Eq. (8), WL,max
1.5 Eq. (9), WU,max
r 2=0.9966
0.5
2 2.5 0.5 1 1.5 2 2.5
Engineering strain peak ε E,max
Fig. 6. The relationship between the peak strain energy density and peak
engineering strain.
reported in [10]. The results also show generally good correlations
for the models based on the five strain measures with r2 greater
than 0.9. All of these models would thus be acceptable from the
point of engineering applications. Therefore, all of the five strain
measures (kmax , eG,max, eA,max, eE,max or eL,max) can be effectively used
as damage parameters for predicting fatigue lives of rubber materi-
als under uniaxial tension.
4.2.2. Fatigue life model based on peak of the octahedral shear strain
Considering peak of the octahedral shear strain, eO,max, as the
damage parameter Eq. (11) is re-written as:
eO;max ¼ KðNf Þb ð13Þ
where the constants K and b are determined through curve fitting
the measured fatigue life data (Nf). Fig. 8 illustrates the correlation
between the computed damage parameter, eO,max, and the mea-
sured fatigue life data. The model parameters together with the cor-
relation coefficient r2 are presented in Table 5. The results show
high degree of correlation of the octahedral shear strain with the fa-
tigue life data (r2 = 0.9963).
4.2.3. Fatigue life models based on peak strain energy density
The fatigue models are formulated considering the strain en-
ergy density, SED, corresponding to both the loading [13] as well
as unloading [20] stress–strain curves, as described in Section 3.3.1.
Eq. (11) is subsequently re-written as:
W max ¼ KðNf Þb ð14Þ
where Wmax may represent the SED during loading (WL,max) or
unloading (WU,max), while the constants K and b are determined
using a curve fitting technique.
The correlations of the measured fatigue lives with those esti-
mated from the two fatigue life prediction models using WL,max
and WU,max as damage parameters are shown in Fig. 8. The model
ð12Þ
parameters and the correlation coefficients are given in Table 5. It
is seen that the correlation coefficients of both the models are great-
er than 0.99. The model based on WU,max, however, provides a
slightly larger r2 value than that based on WL,max. This demonstrates
that the unloading path exhibits more ideally hyper-elastic proper-
ties, and has a direct relationship with the energy release rate [20].
Furthermore, the correlation coefficients of both the models are
nearly identical to that of the model employing peak of the 1st prin-
cipal Green–Lagrange strain as the damage parameter.
4.3. Discussions of damage parameters for fatigue life prediction of
rubbers
Reported studies have employed widely different damage
parameters, described above, for prediction of fatigue lives of
 W.-B. Shangguan et al. / Materials and Design 58 (2014) 65–73 71

Table 4
The average measured fatigue lives and damage parameters for the dumbbell simple tension specimen (STS).

Specimen type Loading case Average measured life Nave (cycle) Nmax/Nmin kmax eG,max eA,max eE,max eL,max eO,max WL,max (MPa) WU,max (MPa)
STS 1 25,181 2.05 2.83 3.51 0.44 1.83 1.04 3.61 1.7 1.53
2 38,967 1.53 2.61 2.9 0.43 1.61 0.96 3.03 1.38 1.23
3 62,206 1.43 2.35 2.27 0.41 1.35 0.86 2.40 1.05 0.92
4 82,049 2.01 2.23 1.98 0.4 1.23 0.8 2.13 0.9 0.78
5 100,290 2.85 2.19 1.89 0.4 1.19 0.78 2.05 0.85 0.73
6 139,460 1.84 2.00 1.50 0.37 1.00 0.69 1.65 0.64 0.54
7 185,560 1.37 1.93 1.36 0.37 0.93 0.66 1.51 0.57 0.48
8 312,590 1.77 1.82 1.16 0.35 0.82 0.6 1.30 0.47 0.39
9 398,730 1.68 1.71 0.96 0.33 0.71 0.54 1.10 0.37 0.31
10 522,450 2.09 1.62 0.81 0.31 0.62 0.48 0.95 0.3 0.24

rubbers subjected to uniaxial loadings [4–10]. Among the different
strain measures, Kim et al. [10] suggested that Green–Lagrange
strain is more appropriate to be used as a damage parameter for fa-
tigue analyses of rubber. Alternate strain measures, however, have
been frequently used as damage parameters in various studies on
rubber fatigue [4–9].
From the results presented in Figs. 7 and 8, and Table 5, it is
concluded that different fatigue life prediction models yield corre-
lation coefficients in excess of 0.9, irrespective of the damage
parameter considered. The peaks of the 1st principal strain mea-
sures (Green–Lagrange strain, Almansi-Euler strain, engineering
strain, logarithmic strain, and stretch ratio), the peak of octahedral
shear strain, and the peak strain energy density computed from
either the loading or the unloading paths of the stress–strain
curves, could thus serve as effective damage parameters for predic-
tion of fatigue lives of rubbers under uniaxial loading. The relative
analysis justifies the use of different damage parameters in the re-
ported studies [4–14].
5.1. Fatigue life prediction model for evaluating the lives of DCS and
HCS specimens
Considering the peak of the 1st principal Green–Lagrange strain,
eG,max, as the damage parameter, the fatigue life prediction model
for the STS (dumbbell simple tension specimen) was obtained as:
eG;max ¼ 418:29ðNf Þ0:47 ð15Þ
which yields the fatigue life as:
Nf ¼ 358; 080ðeG;max Þ2:12 ð16Þ
The above relationship could be considered valid for predicting
fatigue lives of DCS and HCS specimens, provided it is geometry-
independent. The correlation between the measured and predicted
tension fatigue lives of the specimens is investigated to establish
the geometry-dependence or – independence of the model.
5.2. Calculation of damage parameters for DCS and HCS specimens

5. Discussions of the relationship between the tension fatigue

The estimation of fatigue lives of DCS and HCS specimens neces-
life and types of rubber specimens
sitates determinations of the damage parameters corresponding to
the global loading. Owing to variations in the cross-section areas of
The tension fatigue lives of other rubber specimens used in the
the DCS and the HCS specimens along the axial direction, the strain
experiments is further investigated to explore the applicability of
along the loading direction could not be evaluated analytically. Dif-
models presented in the above section. Since all the models pro-
ferent strain measures for each specimen were thus evaluated
vided effective prediction of fatigue lives of the STS specimen,
through FEA using the ABAQUS software. The mesh and boundary
the model based on peak of the 1st principal Green–Lagrange
strain, eG,max, alone is used.
1
1 10
10 Measured ε
o,max
Measured λ
Damage parameter

max
r2=0.9966 r2=0.9963 Measured W
Measured ε L,max
Peak of Max. principal

G,max
Measured W
2 Measured ε U,max
r =0.9929 A,max
r2=0.9962 Eq.(13), ε
Measured ε L,max o,max
strain ε1,max

r2=0.9952 Measured ε 0 Eq.(14), W

max
=W
U,max
10
0 E,max 10
2 Eq.(12), ε =λ r2=0.9963 Eq.(14), W
max
=W
L,max
r =0.9904 1,max max
Eq.(12), ε =ε
1,max G,max

r2=0.9576 Eq.(12), ε =ε
1,max E,max
Eq.(12), ε =ε
1,max L,max
Eq.(12), ε =ε
-1 1,max A,max
-1
10 10 4 5 6 7
4 5 6 7
10 10 10 10 10 10 10 10

Fatigue life Nf (cycle) Fatigue life Nf (cycle)

Fig. 7. Correlations of different strain-based damage parameters with the average Fig. 8. Correlations of the peak octahedral shear strain and peak strain energy
measured fatigue lives of the STS samples. density damage parameters with the measured fatigue lives.

Table 5
Material constants and correlation coefficients of the fatigue life models based on different damage parameters.

Damage parameter Pd kmax eG,max eA,max eE,max eL,max eO,max WL,max WU,max
K 17.85 418.29 1.35 59.24 12.05 289.41 431.26 542.45
b 0.18 0.47 0.108 0.34 0.24 0.43 0.55 0.58
r2 0.9929 0.9966 0.9576 0.9952 0.9904 0.9963 0.9962 0.9963
72 W.-B. Shangguan et al. / Materials and Design 58 (2014) 65–73

Table 6
The average measured and predicted fatigue lives and computed damage parameters for the DCS and HCS specimens.

Specimen type Loading case Measured life Nexp (cycles) Nmax/Nmin kmax eG,max eL,max Predicted life Npre (cycle) Relative error (%) Npre/Nexp
DCS 11 83,414 1.00 2.23 1.98 0.80 84,545 1.36 1.01
12 124,680 1.53 2.14 1.79 0.76 104,782 15.96 0.84
13 144,100 2.02 1.99 1.49 0.69 154,399 7.15 1.07
14 262,410 1.28 1.82 1.16 0.60 261,441 0.37 1.00
15 483,340 1.38 1.63 0.83 0.49 528,390 9.32 1.09
16 810,740 1.69 1.57 0.73 0.45 697,902 13.92 0.86
17 1,395,000 8.15 1.49 0.61 0.40 1,010,674 27.55 0.72
HCS 18 220,570 1.19 2.01 1.53 0.7 145,841 33.88 0.66
19 258,430 1.36 1.82 1.16 0.6 261,386 1.14 1.01
20 548,150 1.50 1.65 0.86 0.5 494,042 9.87 0.90
21 1,177,700 1.34 1.57 0.73 0.45 698,210 40.71 0.59
22 1,097,200 1.54 1.49 0.61 0.4 1,011,367 7.82 0.92

conditions for the FEA model [5] are shown in Fig. 9. The FEA model
employed the conventional three-dimensional elements (C3D8IH),
while the nodes at the bottom surface of the specimen were fixed.
The prescribed harmonic displacement was applied to all nodes at
the top surface along the axial direction (Z-axis), while the dis-
placements along the other directions were set to zero.
Taking the DCS as the example, the history of the 1st principal
logarithmic strain at the critical location was obtained when sub-
jected to a prescribed fluctuating displacement at the free end of
the specimen. The history of the 1st principal Green–Lagrange
strain and the stretch ratio at the critical location were subse-
quently evaluated using the relations defined in Section 3.
The Mooney–Rivlin constitutive model was used for obtaining
the 1st principal logarithmic strain at the critical location [18]:
W ¼ C 10 ðI1  3Þ þ C 01 ðI2  3Þ ð17Þ
where C10 and C01 are the material constants, I1 and I2 are the first
and second strain invariants of the right Green deformation tensor.
The material constants, C10 and C01, were identified through
minimization of the deviations between the measured and com-
puted stress–strain responses using the least squares method.
Fig. 9. The FEA model of the dumbbell cylindrical specimen (DCS).
The measured stress–strain curves are typically computed from
the three independent strain states, namely, the simple tension,
general, show reasonably good agreements with the measured
planar tension, and equi-biaxial tension. Using the measured data
data for both the specimens, within a factor of 2 (Fig. 10). The fati-
in the three strain states under several peak strains, the material
gue life prediction models may thus be considered applicable for
constants for the Mooney–Rivlin model were obtained as:
the STS, the DCS and the HCS specimens.
C10 = 0.2897 MPa and C01 = 0.0599 MPa.
Similar degrees of correlations were also found between the
measured tension fatigue lives and those estimated from the mod-
5.3. Comparisons of the predicted and measured fatigue lives for DCS
els based on other damage parameters, discussed above. It is thus
and HCS specimens
concluded that the relationships between the tension fatigue life
and the damage parameters used in this study are geometry-inde-
The fatigue lives of the DCS and HCS specimens were evaluated
pendent for the three rubber specimens considered. This further
using the computed damage parameters described in Section 5.2
suggests that the fatigue life is primarily dependent upon the
and the fatigue life prediction model in Eq. (16). The results are
material properties. This finding has some similarity with the
presented in Table 6 together with the average measured fatigue
lives of both the specimens corresponding to the different loading 7
10
Predicted fatiuge life Npre (cycle)

cases presented in Table 3.

Comparisons of the measured and predicted lives show reason- DCS
HCS
ably good agreements under some of the loading cases and sub- 6
10
stantial deviations under others. The relative errors between the
predicted and the measured fatigue lives for the DCS specimen
range from a low of 0.37% to a maximum of 15.96% for all the 5
10
loading cases, except for the loading case 17, where the error is
27.55%. This was most likely caused by the excessive scatter in
the measured data acquired for this loading case compared to all 4
10
the other loading cases. The scatter index for this loading case, 10
4
10
5
10
6
10
7

Nmax/Nmin, is 8.15, as seen in Table 6. Measured fatigue life Nexp (cycle)

The results in Table 6 show relatively higher errors between the
predicted and the measured fatigue lives for the HCS specimen Fig. 10. Comparisons of the measured and the predicted fatigue lives of the DCS and
compared to those of the DCS specimen. However, the results, in HCS specimens.
 W.-B. Shangguan et al. / Materials and Design 58 (2014) 65–73
results reported in an early study by Gent [22] with regards to the
crack growth fatigue characteristics of natural rubbers. Gent [22]
investigated three different types of cracked test specimens and
observed a consistent relationship between the crack growth rate
and energy release rate for all the specimens, which is a material
property related to the rupture of rubber.
6. Conclusions
This study investigated the relationships between the tension
fatigue lives and various commonly used damage parameters,
and the specimen geometry-dependence of the fatigue life predic-
tion models. The tension fatigue lives of three rubber specimens of
different geometry (a dumbbell simple tension specimen, STS; a
dumbbell cylindrical specimen, DCS; and a hollow cylindrical spec-
imen, HCS), were measured and estimated from the models based
on different damage parameters.
It is shown that different strain-based (principal strain and
octahedral shear strain) and energy-based (strain energy density
corresponding to loading and unloading paths) measures could
serve as effective damage parameters for predicting uniaxial ten-
sion fatigue lives of filled natural rubbers. All of the damage
parameters considered in the study, namely, the peaks of the 1st
principal Green–Lagrange strain, Almansi–Euler strain, engineering
strain, logarithmic strain, stretch ratio; peak octahedral shear
strain and peaks of strain energy density corresponding to the
loading and unloading paths of the stress–strain curves, resulted
in very strong correlations between the measured and predicted
lives (r2 > 0.95). The fatigue life model with peak of the 1st princi-
pal Green–Lagrange strain as the damage parameter, however, pro-
vided the largest correlation coefficient.
It is further shown that the relationship between the tension fa-
tigue life and the damage parameters can be considered indepen-
dent on the geometry, which provides a material property of the
rubber fatigue. The tension fatigue lives of the DCS and the HCS
specimens can thus be effectively estimated from the fatigue mod-
el derived from the test data of the STS specimen, since the fatigue
experiments with a STS specimen are highly efficient and cost-
effective.
Acknowledgements
The authors gratefully acknowledge the financial support of
Foundation of State Key Laboratory of Automotive Simulation
and Control (Project No. 20120103), and the Natural Science Foun-
dation of China (Project No. 51275175). The authors also thank the
postgraduate students T.K. Liu, X.X. Xie and B.H. Yao from South
China University of Technology for their work in experiments in
the present work.
73
Appendix A. Supplementary material
Supplementary data associated with this article can be found, in
the online version, at http://dx.doi.org/10.1016/j.matdes.2014.01.
035.
References
[1] Lewitzke C, Lee P. Application of elastomeric components for noise and
vibration isolation in the automotive industry. In: SAE 2001 noise & vibration
conference & exposition, Michigan, USA, SAE Paper No. 2001-01-1447; 2001.
[2] Mars WV, Fatemi A. Factors that affect the fatigue life of rubber: a literature
survey. Rubber Chem Technol 2004;76:391–412.
[3] Mars WV. Fatigue life prediction for elastomeric structures. Rubber Chem
Technol 2007;80(3):481–503.
[4] Abraham F, Alshuth T, Jerrams S. The effect of minimum stress and stress
amplitude on the fatigue life of non strain crystallizing elastomers. Mater Des
2005;26:239–45.
[5] Wang XL, Shangguan WB, Liu TK, et al. Experiment of uniaxial tension fatigue
and modeling of fatigue life for filled natural rubbers. Chin J Mech Eng
2013;49(14):65–73 [Chin. Ed.].
[6] Cadwell SM, Merrill RA, Sloman CM, Yost FL. Dynamic fatigue life of rubber. Ind
Eng Chem 1940;12:19–23.
[7] André N, Gailletaud G, Piques R. Haigh diagram for fatigue crack initiation
prediction of natural rubber components. Kautschuk Gummi Kunstoffe
1999;52:120–3.
[8] Oshima H, Aono Y, Noguchi H. Fatigue characteristics of vulcanized natural
rubber for automotive engine mounting. Mem Fac Eng 2007;67(2):75–83.
[9] Zarrin-Ghalami T, Fatemi A. Cumulative fatigue damage and life prediction of
elastomeric components. Fatigue Fract Eng Mater Struct 2013;36:270–9.
[10] Kim WD, Lee HJ, Kim JY, Koh SK. Fatigue life estimation of an engine rubber
mount. Int J Fatigue 2004;26:553–60.
[11] Saintier N, Cailletaud G, Piques R. Multiaxial fatigue life prediction for a natural
rubber. Int J Fatigue 2006;28(5–6):530–9.
[12] Verron E. Configurational mechanics: a tool to investigate fracture and fatigue
of rubber. Rubber Chem Technol 2010;83(3):270–81.
[13] Mars WV. Multiaxial fatigue of rubber. The University of Toledo, USA; 2001. p.
08.
[14] Zine A, Benseddiq N, Abdelaziz MN. Rubber fatigue life under multiaxial
loading: numerical and experimental investigations. Int J Fatigue 2011;
33:1360–8.
[15] Mars WV, Fatemi A. A novel specimen for investigating the mechanical
behavior of elastomers under multiaxial loading conditions. Exp Mech 2004;
44(2):136–46.
[16] Tao G, Xia Z. A non-contact real-time strain measurement and control system
for multiaxial cyclic/fatigue tests of polymer materials by digital image
correlation method. Polym Testing 2005;24:844–55.
[17] Le Cam J-B, Huneau B, Verron E. Description of fatigue damage in carbon black
filled natural rubber. Fatigue Fract Eng Mater Struct 2008;31:1031–8.
[18] Bonet J, Wood RD. Nonlinear continuum mechanics for finite element
analysis. Cambridge: Cambridge University Press; 1997.
[19] McGarry MDJ, Van Houten EEW, Perrinez PR, et al. An octahedral shear strain-
based measure of SNR for 3D MR elastography. Phys Med Biol 2011;56:
153–64.
[20] Dizon ES, Hicks AE, Chirico VE. The effect of carbon black parameters on the
fatigue life of filled rubber compounds. Rubber Chem Technol 1974;47:
231–49.
[21] Sors L. Fatigue design of machine components. Oxford: Pergamon Press; 1971.
[22] Gent AN, Lindley PB, Thomas AG. Cut growth and fatigue of rubbers. I. The
relationship between cut growth and fatigue. J Appl Polym Sci 1964;8:455–66.