JOURNAL OF

COMPOSITE
Article M AT E R I A L S
Journal of Composite Materials
2017, Vol. 51(20) 2949–2960
! The Author(s) 2017
A progressive multi-scale fatigue model Reprints and permissions:
sagepub.co.uk/journalsPermissions.nav
for life prediction of laminated composites DOI: 10.1177/0021998317709610
journals.sagepub.com/home/jcm

Seyed Ali Hosseini Kordkheili, H Toozandehjani and Z Soltani

Abstract
This article presents a multi-scale progressive micro-mechanical fatigue model. The model employs fundamental equation
of the kinetic theory of fracture to calculate damage parameters of both fiber and matrix during cyclic loading. In order
to adapt the equation, required material coefficients of the constituents can be achieved from fatigue test results of
longitudinal and transverse unidirectional composites, only. Sharing stress capacities of fiber and matrix are determined
using a modified progressive micro-mechanical bridging model in the presence of damage. The damage parameters in the
constituents are calculated employing two different equivalent scalars. However, during sinusoidal load application, these
damage parameters are also updated using a first kind Bessel function of amplitude stresses in the constituents as well as
their material coefficients. The enhanced formulation is then implemented into the commercial finite element software of
ABAQUS via a developed user material (UMAT) subroutine utilizing a suitable failure criteria and an own solution
algorithm. Advantages of the proposed model are assessed and comparisons with available solutions are presented.

Keywords
Multi-scale model, progressive damage, micro-mechanical fatigue, bridging model, laminated composite

Introduction investigation of the failure mechanisms in each

Fiber-reinforced composite materials have gained constituent. These models are based on the concept of
popularity in high-performance products such as aero- the representative volume element (RVE) or repeated
space, wind turbine, automotive, and sporting goods. unit cell (RUC) which enable the simulation of com-
Fatigue is a common type of failure under variable posite materials by an equivalent fiber–matrix
amplitude cyclic loading which may causes degradation representation.
of material properties and eventually failure. A survey Reifsnider and Gao1 established a micro-mechanical
in the literature reveals that numerous numerical and fatigue criteria based on an average stress formulation
empirical models have been proposed to predict com- of composite materials derived from the Mori–Tanaka
posites fatigue responses which are all based on a great method. Subramanian et al.2 used the concentric
amount of experimentally generated data.1–5 Due to cylinders model to predict the fatigue characteristics
anisotropy and heterogeneity of fiber-reinforced com- of off-axis unidirectional (UD) laminates. The averaged
posite materials, their fatigue life prediction is still a stresses along the radial distance in each constituent
challenging task after five decades of research. phase were applied to predict the S–N curves of UD
Macro-mechanical modeling and micro-mechanical laminates. Abdelal et al.3 presented a model based on a
modeling are two major approaches that can be used combination of an orthotropic damage model and an
for composite material analysis. Macro-mechanical
approaches employ empirical factors to predict Aerospace Engineering Department, Center of Research and
damage progression and usually ignore to consider Development in Space Science and Technology, Sharif University of
any physics-based criteria. These phenomenological Technology, Tehran, Iran
models are specific to certain kinds of composite mater-
Corresponding author:
ials and are unable to be used in a general case.1 Seyed Ali Hosseini Kordkheili, Azadi Avenue, P.O. Box 11365,
Unlike macro-mechanical models, the micro- Tehran, Iran.
mechanical ones provide us with a detailed Email: [email protected]
2950 Journal of Composite Materials 51(20)

isotropic fatigue evolution model for fatigue damage Kenik15 has employed a matrix breakage criterion, in
evolution of polymer matrix composites. Huang4 the present work both matrix and fiber damage has
established the fatigue strength of a UD fibrous com- been considered in a progressive manner. In this
posite in terms of the fatigue properties of its constitu- model fiber–matrix interface bonding assume to be per-
ent. Hung et al.5 proposed a micro-mechanics-based fect to decrease computational cost. However, authors
methodology for fatigue life prediction of polymer have previously investigated effect of interface imper-
matrix composite. Shokrieh et al.6 used combination fection on composite material properties.16 The present
of the micro-mechanical and normalized stiffness deg- model enables us to calculate stiffness degradation of
radation approaches for fatigue life prediction of UDs (meso-scale) and consequently laminated compos-
nanoparticle/fibrous polymeric composites. Qian ite materials (macro-scale) to predict their fatigue fail-
et al.7 used a multi-scale numerical fatigue model for ure by calculating average stresses in constituents
the characterization of axial tension–tension fatigue (micro-scale). The main advantage of the proposed
behavior of UD glass fiber composites. Adibnazari model is that, one only needs to perform fatigue tests
et al.8 used a bridging micro-mechanical model in on 0 and 90 UD composites to conduct unknown
order to observe the behavior of UD laminated com- parameters of kinetic theory’s formula for both fiber
posite under fatigue loading. Hansen and Baker-Jarvis9 and matrix, respectively. Then these results are
developed a theory to study fracture in polymers under employed to predict fatigue life of off-axis UDs and
variable stress histories. Later on Fertig10 employed composite laminates under multi-axial loading
this theory to estimate matrix damage parameter and conditions. In the developed model fiber and matrix
predict fatigue life of off-axis UD composites. Based on brakeage are only considered. Other phenomena like-
continuum damage mechanics, Zhang et al.11 modeled wise fiber debonding, delamination as well as hysteretic
behavior of fiber-reinforced composite laminae and localized heating effects may be added to this
under fatigue loading. They developed a non-linear model in the future works. However, in this work the
cumulative model to take into account three different effects of some mention phenomena are included due to
damage failures; fiber breakage, elastoplastic damage the usage of UD experimental test data.
under shear and transverse tension stress. Also, In order to demonstrate the accuracy of the model,
Sayyidmousavi et al.12 recently introduced a progres- some examples are executed and the results obtained
sive fatigue damage model within a multi-scale frame- within this approach solution are compared with
work to model failure in UD and multidirectional those of available in the literature. The capability of
laminates using a finite element program. Li et al.13 the model is also investigated by examining damage
extended theory of micro-mechanics of failure to ana- distribution and fatigue life prediction of laminated
lyze fatigue progressive of failure and predict strength composites with hole.
of bolted joint in composite structures. Krause14 pre-
sented a novel physically based fatigue failure criterion
for polymers under multi-axial loading conditions. He Progressive multi-scale fatigue model
also developed a micromechanical approach by using
Theory
extended failure model to evaluate damage initiation
and evolution for fiber reinforced polymer under fati- Employing the fundamental equation of the rate
gue loading condition. dependent kinetic theory of fracture, a differential
The survey in the literature reveals that, over a couple equation for damage variables of fiber as well as
of last decades, the development of suitable micro- matrix are introduced as follows9
mechanical models for fatigue analysis of laminated
composites has been one of the most important activ- dDiðtÞ  
ities. This is partially motivated by the need to reduce the ¼ D0  DiðtÞ Kib ðtÞ ; i ¼ f, m ð1Þ
dt
number of tests which are normally required when any
changes occur in stress ratio, loading direction and/or where superscript i stands for fiber (f) or matrix (m),
frequency, stacking sequence and/or number of layers. D0 ¼ e/(e  1) is a constant parameter, e is Napier’s con-
This paper aims to consider physics of damage in the stant and Kib ðtÞ are time dependent bond rupture rate in
constituents during model generation to reduce the constituents correspond to variable carrying
i
required experimentally generated data for life predic- equivalent stresses ðtÞ which are given by Hansen and
tion of laminated composites. For this purpose the kin- Baker-Jarvis9
etic theory of fracture is employed for both matrix and !
fiber together with a modified progressive micro- kT Ui   i ðtÞ
i

mechanical bridging model to obtain a particular phy- Kib ðtÞ ¼ exp  ð2Þ
h kT
sics-based multi-scale model. However, Fertig and
Hosseini Kordkheili et al. 2951

where k is Boltzman’s constant, T is absolute tempera-

Equivalent stresses in fiber and matrix
ture value, h is Planck’s constant, and Ui and  i are During development of the theory one important
activation energy of the fracture process and a material aspect may be choose of a proper equivalent sca-
coefficient in constituents, respectively. It should be lar stress for fiber and matrix. To this end, Von-
noted that DiðtÞ represents the evolution of micro- Mises equivalent stress is employed for fiber as
crack density under cyclic loading in constituents. The follows:
limiting value Dið0Þ ¼ 0 corresponds to a pristine condi-
tion and Dið i Þ ¼ 1 represents a completely failure (i.e.  i f
ðtÞ f
¼ ðI1ðtÞ
2
f
 3I2ðtÞ Þ1=2
is fiber or matrix rupture time). In order to determine f f f f
Ui and  i , equation (2) is substituted in equation (1) I1ðtÞ ¼ 1ðtÞ þ 2ðtÞ þ 3ðtÞ
ð7Þ
then yielded relation is rewritten and integrated in f
I2ðtÞ f
¼ 1ðtÞ f
2ðtÞ f
þ 2ðtÞ f
3ðtÞ
proper intervals as follows: 2 2 2
f f f f f
þ 3ðtÞ 1ðtÞ  4ðtÞ  5ðtÞ  6ðtÞ
Z Z !
1 dDiðtÞ i Ui   i ðtÞ
i
kT
 ¼ exp  dt ð3Þ f
where iðtÞ are micro-stresses in fiber that are calculated
0 D0  DiðtÞ 0 h kT
after computing load sharing capacities of fiber and
matrix using bridging model, see the section
Getting integration from both sides of the relation ‘‘Progressive micro-mechanical bridging model’’.
(3) we have In the case of matrix, since its failure mode is
! occurred via cracking parallel to the fiber, it is
    Z i
D0 kT Ui  i ðtÞ
i expected that tensile forces perpendicular to the
ln ¼ exp  exp dt ð4Þ fiber as well as shear stresses on these planes will
D0  1 h kT 0 kT
play a substantial role in equivalent matrix stress
value. Also, due to transverse isotropy the failure
Knowing that D0 ¼ e/(e  1) and by replacing criterion must be invariant under any rotation of
 i ¼ Nif =f where f is sinusoidal load frequency, from the x2, x3 axes around x1 axis. Regarding on these
equation (4) results points, in the present study an effective stress relation
! suggested by Fertig is used to calculate the equivalent
  Z Ni =f stress in matrix10
kT Ui f  i ðtÞ
i
1¼ exp  exp dt
h kT 0 kT
  Z 1=f ! ð5Þ  1  !1=2
i kT Ui  i ðtÞ
i
m Bt m 2  m m Bs2 m
¼ Nf exp  exp dt ðtÞ ¼ fI g þ 1  fIhðtÞ g Is1ðtÞ þ I
h kT 0 kT Bs1 tðtÞ 0 Bs1 s2ðtÞ
ð8Þ
Now, assuming constant stress magnitude for each
loading cycle from equation (5) yields where

  2Im m m m m 2
kT kT fh Ui tðtÞ ¼ 2ðtÞ þ 3ðtÞ þ ½ð2ðtÞ þ 3ðtÞ Þ
i i
ðN iÞ ¼ i log Nf þ i log þ i ð6Þ m m m 2 1=2
f  log e  log e kT   4ð2ðtÞ 3ðtÞ  ð6ðtÞ Þ Þ
 2  2
i i
Im
s1ðtÞ
m
¼ 4ðtÞ þ 5ðtÞm
ð9Þ
From equation (6) observes that in ðN i Þ vs log Nf
f
1 m 2  2
diagram the slope of linear regression is directed by Im
s2ðtÞ ¼ 2ðtÞ m
 3ðtÞ m
þ 6ðtÞ
4
kT= i log e and i
ðNi Þ -axis intercepted by Im m
¼ 2ðtÞ m
þ 3ðtÞ
f hðtÞ
kT logð fh=kTÞ= i log e þ Ui = i . Now, employing these
m
relations together with two different sets of fatigue where iðtÞ are micro-stresses in matrix and {},  and 0
test results on longitudinal and transverse specimens, are the Macaulay brackets, pressure strengthening coef-
Uf and  f as well as Um and  m are calculated, respect- ficient and the matrix shear strength with no pressure
ively. Having these data as input values of the proposed strengthening, respectively. Bt , Bs1 , and Bs2 are also
algorithm that will be discussed in the next section; determined from three composite static failure tests;
stiffness degradation of composite laminates are ana- transverse tension, transverse compression, and in-
lyzed and fatigue life under arbitrary cyclic loading plane shear, all of which involve failure of the matrix
condition are predicted. constituent.
2952 Journal of Composite Materials 51(20)

Fatigue damage parameters of constituents

under sinusoidal loading
For the applied sinusoidal loading on composite plate
with frequency f, mean stress value m , and amplitude
stress a , the equivalent stresses in fiber and matrix is
i
written as ðtÞ ¼ mi ðtÞ þ aðtÞ
i
sinð2ftÞ. Substituting this
relation together with equation (2) into equation (1)
yields

dDiðtÞ kT  
¼ D0  DiðtÞ
dt h  1
0
Ui   i mðtÞ
i i
þ aðtÞ sinð2ftÞ ð10Þ
exp@ A
kT

Knowing that N ¼ ft, an alternative relation for

damage parameters history versus load cycle (N) can
be obtained as follows:

dDiðNÞ kT  
¼ D0  DiðNÞ
dN fh
0  1
Ui   i mðNÞ
i i
þ aðNÞ sinð2NÞ
exp@ A
kT

ð11Þ
Figure 1. Flowchart of algorithm to implement the PMF model.
or PMF: progressive multi-scale fatigue.

dDiðNÞ
  Table 1. Material properties of T800H/2500EP.17
D0  DiðNÞ
0  1 E11 ðGPaÞ E22 ðGPaÞ G12 ðGPaÞ 12
i i i i
kT U    mðNÞ þ aðNÞ sin ð2N Þ
¼ exp@ AdN Fiber 245 12.8 15.0 0.276
fh kT Matrix 4.01 – – 0.46
ð12Þ Lamina 159 8.45 4.78 0.357

Assuming that damage parameters are known for up

to cycle N1 and should be determined for cycle N2, Table 2. Ui and  i values for T800H/2500EP.
taking integral from both sides of equation (12) at
i Ui ðkJ=molÞ  i ðJ=MPa molÞ
interval [N1, N2] we obtain
! ! Fiber (f) 146.13 29.7
D0  DiðN1 Þ kT  i mðN
i
1Þ
 Ui Matrix (m) 121 1487.7
ln ¼ exp
D0  DiðN2 Þ fh kT
Z N2 !
i i
 aðN1 Þ sinð2NÞ
exp dN
N1 kT
matrix are updated after estimating a new value for
ð13Þ damage parameters. Thus identified mean and ampli-
tude values of stress at last cycle, i.e. N1, are employed
It should be noted that during applying progressive to provide a solution for equation (13). This consider-
multi-scale fatigue (PMF) model stresses in fiber and ation leads us to observe that, the remained integral in
Hosseini Kordkheili et al. 2953

Londitudinal UD
x x x x x

x θ=10o x
x
θ=15o x
x x
θ=20
o x
x x x
x x
θ=30
o
x
σmax(MPa)

x x x
θ=45 o x x x
x
x
Transverse
UD x
x x
x x
x x
x
Sympol data, Kawai et al. 2001(Experimental) x
0.0 deg
10.0 deg
15.0 deg
20.0 deg
30.0 deg
45.0 deg
90.0 deg
x Present work

Log(Nf)
Figure 2. Comparison between fatigue lives predicted by the PMF model and experimental test data17 for T800H/2500EP UDs.
PMF: progressive multi-scale fatigue; UD: unidirectional.

RHS of equation (13), is an expression for the Bessel 1

function of the first kind of order zero i.e. 0.9

Z !
N2  i aðN
i
sinð2NÞ 0.8
σ max=

1Þ
σ max=

σ max=
σ =
exp dN ¼ ðN2  N1 Þ
σ max=

kT max
1300

1100
1500

N1 0.7 1200
1400

!
MP

M Pa
M

MP
M

 i aðN
i
a
Pa

0.6
Pa

a
E /E0

1Þ
f

BesslI 0,
f

kT 0.5

ð14Þ 0.4

0.3
Substituting equation (14) into equation (13) yields
0.2
! !
D0  DiðN1 Þ kT  i mðN
i
1Þ
U i
0.1
ln ¼ ðN2  N1 Þ exp 0 1 2 3 4 5 6
D0  DiðN2 Þ fh kT Log(N)
!
 i aðN
i
1Þ
Figure 3. Normalized fiber stiffness degradation for longitu-
BesslI 0, dinal UD (lamina).
kT
UD: unidirectional.
ð15Þ
where
Thus ! !
kT  i mðN
i
1Þ
 Ui  i aðN
i
1Þ
   ðN1 Þ ¼ exp BesslI 0,  N1
DiðN2 Þ ¼ D0  D0  DiðN1 Þ exp N2  ðN1 Þ ð16Þ fh kT kT
ð17Þ
2954 Journal of Composite Materials 51(20)

f f
E /E0 for off-axis UDs
m m
It should be noted that assuming constant stress
E /E0 for Longitudinal UD
G
A
F
D
E A G
F A FA AG A
magnitude for each loading cycle during derivation
C
B E
D
C
B G
E
D
C
B F G A A A A A
equation (6) may causes a little error in Ui and  i
E F
D
BC G
E

C
D F
G
values. These errors can be eliminated using a try and
B
E
F
error procedure employing relation (16).
D
m
E /E0 ; E /E0

C G
B
m

E
F
D
G Progressive micro-mechanical bridging model
f

C
B
E
f

Ef/Ef0; σmax=1100MPa, θ=0o D

F
G
For an RVE of a UD with fiber volume fractions Vf ,
E /E0; σmax=1100MPa, θ=0
m m o
A C
B Em/Em0; σmax=300MPa, θ=10o
B
E the volume averaged stress components may defined
E /E0; σmax=190MPa, θ=15
m m o F
C
D Em/Em0; σmax=130MPa, θ=20o
C
D
G as fr g ¼ Vf r f þ ð1  Vf Þ rm . However, the
E /E0; σmax=75MPa, θ=30
m m o B
E E
F Em/Em0; σmax=40MPa, θ=45o
D
F matrix average stress rm can be related to the fiber
E /E0; σmax=20MPa, θ=90
m m o
G G
Ef/Ef0; σmax=20MPa, θ=90o B
C
E average stress s f using a bridging matrix ½Ars  as
F
follows4
Log(N)

Figure 4. Normalized fiber and matrix stiffness degradation in rm ¼ ½Ars  s f ð18Þ
laminae.
Indeed the bridging matrix represents the load shar-
ing capacity of matrix material with respect to fiber
material and its proper definition is the most important
aspect of micro-mechanical bridging model. According
to Huang’s work, there are only eight nonzero elements
in the bridging matrix, in which for an axisymmetric
cylindrical RVE these elements are4
σmax (MPa)

A11 ¼ Em
11 =E11 f ð19  1Þ

A22 ¼ A33 ¼ A44 ¼ 0:5 1 þ Em
11 =E22 f ð19  2Þ
 
A12 ¼ A13 ¼ S12 f  Sm m
12 ðA11  A22 Þ= S11 f  S11
Present work,[0, 90]3s
Kawai et al. 2007 (Experimental), [0, 90]3s ð19  3Þ
Present work,[0, 60,-60]2s
Kawai et al. 2007 (Experimental), [0, 60,-60]2s

A55 ¼ A66 ¼ 0:5 1 þ Gm
12 =G12 f ð19  4Þ
Log(Nf)
where Si11 ¼ 1=Ei11 and Si12 ¼ i12 =Ei11 . Decreasing of
Figure 5. Fatigue life of T800H/2500EP laminated composite material properties of constituents under cyclic loading
under tension–tension fatigue load with R ¼ 0.1. condition dictates modification of Huang’s bridging

σmax=600 MPa σmax=700 MPa

E /E0
f

E f/E0f
f

[15/0/-15]s [15/0/-15]s
[30/0/-30]s [30/0/-30]s
[45/0/-45]s [45/0/-45]s
[60/0/-60]s [60/0/-60]s
[75/0/-75]s [75/0/-75]s
[90/0/-90]s [90/0/-90]s

Log(N) Log(N)
o
Figure 6. Degradation in 0 layer for the plates with different laminations.
Hosseini Kordkheili et al. 2955

1600
1400
flowchart in Figure 1, stresses in meso-scale (in each
1200 integration point) are determined firstly.
1000 Subsequently, the modified micro-mechanical bridging
800 model, presented in the section ‘‘Progressive micro-
600
mechanical bridging model’’, is employed to determine
contribution of fiber and matrix in tolerating of stress.
σmax(MPa)

400
Now using effective stress values in the constituents,
calculated by equations (7) and (8), damage parameters
of matrix and fiber are determined using equation (16)
in the current cycle of loading. Degradation in fiber and
[15/0/-15]s
200
[30/0/-30]s matrix material properties employing their related
[45/0/-45]s
[60/0/-60]s damage parameters causes degradation in overall prop-
[75/0/-75]s
[90/0/-90]s erties of each layer. The algorithm will continue until
one failure mode; matrix or fiber, happens in each
2 3 4 5 6 7
Log(Nf) layer. A sudden drop in equivalent stiffness matrix of
the structure is the selected criteria to examine lami-
Figure 7.  xx versus log Nf curves for different laminates. nated composite failure.
Hahn and Tsai’s model,17 i.e. 12 ¼ 12 =G12 þ 123
,
matrix coefficients in presence of damage. For this pur- is also used to take into account the nonlinear shear
pose material properties Ei11 , Ei22 , and Gi12 are degraded response of composite lamina. In order to apply this
using ð1  DiðNÞ ÞEi11 , ð1  DiðNÞ ÞEi22 , and ð1  DiðNÞ ÞGi12 , model in the UMAT subroutine nonlinear shear mod-
respectively. Therefore relations (19) are rewritten in ules G 12 ¼ ð1=G12 þ 312
2 1
Þ is derived by getting par-
modified form as follows: tial differential from Hahn and Tsai’s model with
respect to 12 .
A11 ¼ Dmf mf
ðNÞ E11 ð20  1Þ

A22 ¼ A33 ¼ A44 ¼ 1 þ Dmf mf

ð20  2Þ
Results and discussion
ðNÞ E11
PMF model verification for UD plates
Dmf
ðNÞ 12 fDmf mf m
ðNÞ E11  12 Here to assess the validity of the proposed theory and
A12 ¼ A13 ¼ 
2 Dmf mf
ðNÞ E11  1
algorithm for UD plates (laminae), T800H/2500EP UD
  ð20  3Þ
composites with 100 mm gauge length and 2 mm thick-
Em
 1  2Emf
11 þ 11 ness are considered. Kawai et al.18 reported experimen-
E22 f tal results for the same UD plates with seven different
off-axis angle ( ¼ 0, 10, 15, 20, 30, 45, and 90 ) and
Gm
A55 ¼ A66 ¼ 1 þ Dmf
ðNÞ
12
ð20  4Þ material properties of Table 1. The fiber volume frac-
G12 f tion is about 0.64 and the specimens’ width are 10 mm
for ¼ 0 , and 25 mm for the others. The tensile fatigue
where Dmf m
ðNÞ ¼ ð1  DðNÞ Þ=ð1  DðNÞ f Þ and Emf 11 ¼ analysis is performed with a constant stress ratio
Em
11 =E11 f. Once the bridging matrix is determined, aver- R ¼ 0.1 and sinusoidal fatigue with 10 Hz frequency
aged stresses in constituents can be obtained using same as those reported in Kawai et al.18
ri ¼ Birs s . Where Birs are averaged stresses In Table 2, Um ,  m , U f, and  f are listed for con-
amplification matrices in constituents and can be deter- stituents of T800H/2500EP which are determined using
mined as follows4 equation (6) and ðtÞi
versus log Nif curves for longitu-
dinal and transverse UDs. Using material properties of
 1
Brsf ¼ Vf ½I þ ð1  Vf Þ½Ars  , the composite constituents together with algorithm
 1
ð21Þ Figure 1 fatigue life of UDs with different off-axis
Bm
rs ¼ ½Ars  Vf ½I þ ð1  Vf Þ½Ars  angles are predicted and depicted in Figure 2. This
figure also compares the results from PMF model
with those experimentally extracted and reported in
Numerical implementation of PMF model Kawai et al.18 This figure shows a very good agreement
A suitable algorithm is required to implement the devel- between the results and reveals that the employed
oped PMF model to analyze laminated composites theory and algorithm appropriately predicts fatigue
using the ABAQUS user material subroutine life of UDs with arbitrary off-axis angle using simply
(UMAT). Regarding to the algorithm with a given 0 and 90 fatigue test results. In this figure, for
2956 Journal of Composite Materials 51(20)

1 1

110
120
130

0M
0M
0M

Pa
0.8 0.8

Pa
Pa

750 MPa
E/E0

E/E0

700 M P

650 M

600 MP
550 M
0.6 0.6

115

P
a

Pa
a
0M

a
Pa
1250
0.4 0.4

M
Pa
[15/0/-15]s [30/0/-30]s
0.2 0.2
0 1 2 3 4 5 6 0 1 2 3 4 5 6
Log(N) Log(N)

1 1
700 M

650 M

650 M
0.8 0.8
600 MP

700 MP
500 MP
P

550 MPa
a

500 MP
550 M
a

600 MP
Pa
E/E0
E/E0

a
a

0.6 0.6

Pa

a
a
0.4 0.4

[45/0/-45]s [60/0/-60]s
0.2 0.2
0 1 2 3 4 5 0 1 2 3 4 5 6
Log(N) Log(N)

1 1

0.8 0.8
700 MP
550 M

600 M

500 M
650 M

600 MP
E/E0

E/E0
700 M

550 M
500 MP

650 M

0.6 0.6
P

Pa
Pa
P
Pa

P
a

a
a

a
a

P
a

0.4 0.4

[75/0/-75]s [90/0/-90]s
0.2 0.2
0 1 2 3 4 5 6 0 1 2 3 4 5
Log(N) Log(N)

Figure 8. Degradation in normalized effective modulus of different laminates.

specimen with ¼ 30 , one can observe that Kawai loading. This figure reveals that the value of stiffness
et al.18 reported different lives for the same loading decreases as the number of cycles increases in a wear-
and we depicted all data to maintain their integrity. out manner. It should be noted that different research-
Figure 3 shows normalized fiber stiffness degradation ers have used different values of E=E0 for failure. For
(or 1  DðNÞ f) for longitudinal UD under various example, Mandell and Samborsky19 used a value
Hosseini Kordkheili et al. 2957

Figure 9. An open hole [452/902/452/02]s laminated composite made of IM7/8552.

0.9
Table 3. Material properties for IM7/8552.20
0.8 X
Carbon fiber (IM7) Epoxy (8552) IM7/8552
0.7 X
E1 (GPa) 303 E (GPa) 3.31 E1 (GPa) 161
E2 ¼ E3 (GPa) 15.2  (GPa) 0.35 E2 ¼ E3 (GPa) 11.4 0.6 X

G12 ¼ G13 (GPa) 9.65 G12 ¼ G13 (GPa) 5.17

σmax/σUTS

0.5 X
G23 (GPa) 6.32 G23 (GPa) 3.98
12 ¼ 13 0.2 12 ¼ 13 0.0226 0.4

23 0.2 23 0.436 0.3

0.2
Present work
around 0.5 for this purpose. Figure 4 displays normal- X
Nixon-Pearson et al. 2013 (Experimental)
0.1
ized fiber and matrix stiffness degradation in different Nixon-Pearson et al. 2013 (FE model)
UDs versus number of cycle. From this figure one can 0
1 2 3 4 5 6
observe that in contrast with longitudinal, fiber in the Log(Nf)
other laminae has no significant contribution in to resist
against fatigue loads. Therefore, in these laminae life Figure 10. Fatigue life of IM7/8552 laminated composite with
time highly decreases by increasing in applied load level. open hole predicted by PMF model.
PMF: progressive multi-scale fatigue.

PMF model verification for laminated plates of these laminated composites predicted by PMF model
Two different T800H/2500EP laminated composite are also in a good agreement with experimental results
plate with [0, 90]3s and [0, 60, 60]2s sequences, gauge in a cautious and conservative manner.
length 100 mm, width 20 mm, and thickness 2 mm are
considered. A tension–tension sinusoidal cyclic load Fatigue analysis of carbon/epoxy laminated
with f ¼ 10 Hz and constant stress ratio R ¼ 0.1 is also
composite
applied to predict fatigue life of the plate. Kawai and
Koizumi20 reported experimental results for the same In order to investigate the effect of lamination angle on
laminated plate. Rendering to Figure 5 the fatigue life fatigue behavior of laminated composites, in this
2958 Journal of Composite Materials 51(20)

Figure 11. Progress in fiber and matrix damage due to applied load in x direction, predicted by PMF model.
PMF: progressive multi-scale fatigue.

example fatigue analysis of [ , 0,  ]s carbon/epoxy occurred for all layers (i.e. for layers 0 , þ15 and
laminated plates are considered. Dimension of the spe- 15 ) at the same applied load around log N ¼ 3.
cimens are assumed to be same as the previous case However for other plates matrix failure happened for
one. Figure 6 shows stiffness degradation in 0 layer layer 0 firstly and after some cycles other layers (i.e.
for the plates with different laminations under max- layers þy and y ) fail. Failure in matrices at the same
imum loading 600 and 700 MPa. These figures reveal time caused sudden increase in fiber loading of layer 0 .
that stiffness degradation in the layer 0 nonlinearly Therefore the curve has been broken at log N ¼ 3.
depends on off-axis angle for < 45 , and for other One may expect that decreasing in lamination angle
plates the degradation behaviors are almost the same. causes increasing in fatigue life of composite plates.
According to these results it can be also noted that for Regarding to Figure 7 which shows  xx versus log Nf
[15/0/15] laminated plate sudden matrix failure for different laminates, this expectancy is true for those
Hosseini Kordkheili et al. 2959

composite both numerically and experimentally. The

PMF model together with the proposed UMAT algo-
rithm (Figure 1) is used to solve this problem under
80%, 70%, 60%, and 50% of static failure load of
the specimen which is 437 MPa. Figure 10 compares
predicted fatigue life for this composite from PMF
model with those from numerical22 and experimental21
ones. According to this figure there is a good agreement
between the results. As can be seen, the presented
model predicts fatigue life in a conservative manner
because the required coefficients used in the model are
extracted using fatigue experimental data tests which
are included all damage mechanisms.
The progress in fiber and matrix damage parameters
are also compared in Figure 11 for different layers at
loading 80% and two different cycles, i.e. Nf/2 and Nf.
Regarding to these results for layer 0 matrix cracking
happens in transverse direction, however, layers 45 ,
45 , and 90 encounter with matrix cracking in direc-
tion of 45 . In addition of these failures, for layer 45o
and 90o matrix cracking also may happen around the
Figure 12. Matrix damage parameter distribution at failure hole and the free edges of the laminate in which may
instant (Nf) due to applied load in x direction. also cause other internal defects such as interfacial
delamination. For layer 0o and at load cycle Nf/2,
fiber damage parameter exceeds from its critical value
just for a small group of integral points around the
plates with < 45 . But for 45 < < 90 , the lamin- hole. But finally the laminate fails at load cycle Nf
ation angle effects on fatigue life is not considerable. when fiber breakage happens in transverse direction.
Normally, to characterize the effect of damage A three-dimensional view for matrix damage parameter
propagation under fatigue, an effective modules, E, is distribution is also depicted in Figure 12 for all layers at
calculated by using relation E ¼ FL= A, where F is the failure instant.
maximum force, A is the cross-sectional area of the
specimen, L is the gage length, and is the elongation.
Damage propagation in constituent, i.e. fiber and
Conclusions
matrix causes variation in mechanical properties of In this paper a so-called PMF model as an applicable
each layer and subsequently effective modules of lamin- tool was explained in detail to analyze stiffness degrad-
ate will be changed. For different laminates in Figure 8, ation of fibrous composite laminates. This physics-
the normalized effective modulus, i.e. E/E0, are plotted based model expresses fatigue failure in laminates in
versus number of cycles. Where, E and E0 are the resi- terms of fatigue properties of its constituents. The
dual and initial effective laminate moduli, respectively. main advantage of this micro-mechanical model is to
According to these figures a gradual reduction in nor- decrease amount of required experiments during fatigue
malized effective modulus occurs for specimen [45/0/ life prediction of laminates. A progressive bridging
45]. But for other lamination sequences due to the model supports the PMF model to relate macro-stress
failure of the off axis layers at the beginning of cyclic to micro-stress in constituents by employing two vari-
load, a sudden reduction happens in normalized effect- able damage parameters for constituents in which are
ive modulus, followed by decreasing in a custom updated by a Bessel function of the first kind of micro-
manner. mechanical stresses in fiber and matrix. A UMAT
subroutine is also developed to numerically implemen-
Fatigue analysis of carbon/epoxy laminated tation of the proposed model in the ABAQUS. For
accurate performance evaluation of the model and
composite with a hole algorithm, some examples of UD and laminated
An open hole [452/902/452/02]s laminated composite plates were performed. A reasonable agreement was
made of IM7/8552 is considered with dimensions shown between the results from the PMF model and
(Figure 9) and material properties (Table 3). Nixon- those experimentally extracted in the literature. Fatigue
Pearson et al.21,22 studied the same laminated analysis of carbon/epoxy laminated composites with
2960 Journal of Composite Materials 51(20)

and without hole was also carried out. These examples prediction of composites. In: SAMPE ’09, Baltimore,
reveal the capability and merit of the proposed model. Maryland, 18–21 May 2009.
11. Zhang W, Zhou Z, Zheng P, et al. The fatigue damage
mesomodel for fiber-reinforced polymer composite
Declaration of Conflicting Interests lamina. J Reinforced Plastics Compos 2014; 33:
The author(s) declared no potential conflicts of interest with 1783–1793.
respect to the research, authorship, and/or publication of this 12. Sayyidmousavi A, Bougherara H and Fawaz Z. A multi-
article. scale approach for fatigue life prediction of polymer
matrix composite laminates. J Reinforced Plast Compos
Funding 2015; 34: 1099–1109.
13. Li W, Cai H, Li C, et al. Micro-mechanics of failure for
The author(s) received no financial support for the research, fatigue strength prediction of bolted joint structures of
authorship, and/or publication of this article. carbon fiber reinforced polymer composite. Compos
Struct 2015; 124: 345–356.
References 14. Krause D. A physically based micromechanical approach
1. Reifsnider K and Gao Z. A micromechanics model for to model damage initiation and evolution of fiber rein-
composites under fatigue loading. Int J Fatigue 1991; 13: forced polymers under fatigue loading conditions.
149–156. Compos B: Eng 2016; 87: 176–195.
2. Subramanian S, Reifsnider K and Stinchcomb W. A 15. Fertig R and Kenik D. Physics-based fatigue life predic-
micromechanics model for the prediction of fatigue char- tion of composite structures. In: NAFEMS World
acteristics of off-axis unidirectional laminates. Appl Congress 2011. NAFEMS Boston, vol. 400, 2011, p.12.
Compos Mater 1994; 1: 69–74. 16. Kordkheili SH and Toozandehjani H. Effective mechan-
3. Abdelal G, Caceres A and Barbero E. A micro-mechanics ical properties of unidirectional composites in the pres-
damage approach for fatigue of composite materials. ence of imperfect interface. Arch Appl Mech 2014; 84:
Compos Struct 2002; 56: 413–422. 807–819.
4. Huang ZM. Micromechanical modeling of fatigue 17. Hahn HT and Tsai SW. Nonlinear elastic behavior of
unidirectional composite laminae. J Compos Mater
strength of unidirectional fibrous composites. Int
1973; 7: 102–118.
J Fatigue 2002; 24: 659–670.
18. Kawai M, Yajima S, Hachinohe A, et al. Off-axis fatigue
5. Huang Y, Jin K, Xu L, et al. A micromechanical meth-
behavior of unidirectional carbon fiber-reinforced com-
odology for fatigue life prediction of polymeric matrix
posites at room and high temperatures. J Compos Mater
composites. In: ICCM18, Jeju Island, Korea, 21–26
2001; 35: 545–576.
August 2011.
19. Mandell JF and Samborsky DD. DOE/MSU composite
6. Shokrieh MM and Esmkhani M. Fatigue life prediction
material fatigue database: test methods, materials, and
of nanoparticle/fibrous polymeric composites based on
analysis. Albuquerque: Sandia National Laboratories,
the micromechanical and normalized stiffness degrad-
1997.
ation approaches. J Mater Sci 2012; 48: 1027–1034.
20. Kawai M and Koizumi M. Nonlinear constant fatigue
7. Qian C, Westphal T and Nijssen RPL. Micro-mechanical
life diagrams for carbon/epoxy laminates at room tem-
fatigue modelling of unidirectional glass fibre reinforced
perature. Compos A: Appl Sci Manufact 2007; 38:
polymer composites. Comput Mater Sci 2013; 69: 62–72.
2342–2353.
8. Adibnazari S, Farsadi M, Koochi A, et al. New approach
21. Nixon-Pearson OJ, Hallett SR, Withers PJ, et al. Damage
for fatigue life prediction of composite plates using
development in open-hole composite specimens in fati-
micromechanical bridging model. J Compos Mater
gue. Part 1: Experimental investigation. Compos Struct
2014; 49: 309–319.
2013; 106: 882–889.
9. Hansen AC and Baker-Jarvis J. A rate dependent kinetic
22. Nixon-Pearson OJ, Hallett SR, Harper PW, et al.
theory of fracture for polymers. Int J Fract 1990; 44:
Damage development in open-hole composite specimens
221–231.
in fatigue. Part 2: Numerical modelling. Compos Struct
10. Fertig R. Bridging the gap between physics and large-
2013; 106: 890–898.
scale structural analysis: a novel method for fatigue life