Composite Structures 82 (2008) 127–131

www.elsevier.com/locate/compstruct

Anisotropic damage model for woven fabric composites during

tension–tension fatigue
C. Wen, S. Yazdani *

Department of Civil Engineering, North Dakota State University, Fargo, ND 58105, United States

Available online 10 January 2007

Abstract

Inelastic deformations of woven fabric composites are attributed to the formation of multitude of cracks that develop in the material
in the fatigue environment. Cracks destroy material bonds and render composites more compliant. This paper presents an anisotropic
damage model for glass–epoxy fabric composites accounting for the strong directionality nature of damage during a tension–tension
fatigue loading. Damage is reflected through the fourth-order compliance tensor involving a damage parameter whose evolution is deter-
mined from the second invariant of the stress tensor. The model is capable of capturing permanent deformation due to damage and is
checked against experimental results.
Ó 2007 Elsevier Ltd. All rights reserved.

Keywords: Anisotropic; Woven composites; Damage; Response; Fatigue

1. Introduction environment. They used flat cruciform shaped specimen

subjected to five different sets of biaxial stress load paths
Glass–fiber woven reinforced composites are used exten- with fiber orientation at angle 0, 221/2, and 45 degrees to
sively in engineering applications due to the inherent the principal stress directions. A significant difference in
advantages that these materials offer over the conventional material response for different angled plies was observed
uni-directional lamina. Two dimensional glass fabrics are as expected. A distinct knee was observed in the material
known to provide better resistance to impact loading, pro- stress–strain relation when the loading and fiber axes where
vide high strength and dimensional stability, offer high heat the same. Pre- and post-knee behavior were quite linear.
and fire resistance, and are preferred in applications where Generally it is accepted that the modes and rate of crack
bi-axial stress states exist. growth depend on the relative moduli, strength, and ductil-
The presence of multitude of cracks at fatigue failure has ity of the constituent materials, strength of fiber–matrix
been documented by a number of investigations. Broutman interface, as well as the far field stress conditions. Equally
and Sahu [1] studied the progressive failure of the material important are the type, size, distribution, evolution, and
by monitoring the crack density at the through thickness. interaction of microdefects formed during manufacturing
Owen [2,3] reported damage initiation at fiber–matrix processes. With the nucleation and propagation of micro-
interface due to debonding while, Mandell et al. [4,5] inves- cracks and microvoids, energy is dissipated and composite
tigated fatigue damage propagation and failure modes of material becomes more compliant. It is a formidable task,
woven glass composites. Smith and Pascoe [6] investigated if not an impossible one, to monitor each crack or defect’s
the behavior of glass reinforced composites under multi- evolution and growth. Therefore, many researchers have
axial state of stress in both monotonic and fatigue loading opted to monitor changes in the material stiffness as an
indirect, yet an effective, measure of internal changes and
*
Corresponding author. Fax: +1 701 231 6185. energy dissipation within the material due to damage pro-
E-mail address: [email protected] (S. Yazdani). cesses [7,8]. Following this approach, Yoshioka and Seferis

0263-8223/$ - see front matter Ó 2007 Elsevier Ltd. All rights reserved.
doi:10.1016/j.compstruct.2007.01.003
128 C. Wen, S. Yazdani / Composite Structures 82 (2008) 127–131

[9] presented a model to predict the fatigue process by the model’s inability to capture permanent deformation
observing modulus deterioration. Chou and Ko [10] pro- during fatigue damage.
posed a model through the prediction of elastic stiffness This paper aims at improving on the perceived short-
based on the lamination theory. Huang [11] applied the comings of Hansen’s model by proposing a mathematical
bridging micromechanics model to simulate the fatigue framework of damage mechanics whereby anisotropic
strength and the S–N curve of a plain-woven fabric com- behavior is modeled. Damage is reflected through the
posite under multi-axial fatigue loading. Degrieck and Pae- material compliance tensor involving a damage parameter
pegem [12] have summarized the major fatigue models and whose increment is obtained from the second invariant of
life time prediction methodologies for reinforced polymer the stress tensor. The formulation is also capable of pre-
composites under fatigue loading. dicting permanent deformations that may arise due to
Since fatigue damage and failure is caused by multitude imperfect fracturing processes.
of cracks and microcracks, it is therefore logical to make
use of a class of theories that is structured to describe the 2. Formulation
weakening of solids in the presence of multiple defects,
namely damage mechanics. One such approach was pre- With the assumption of small deformations which is
sented by Hansen [13] to model fatigue damage growth valid for brittle materials and for low frequency fatigues
of woven glass reinforced composites. His model consid- where thermal effects could be ignored, we consider the
ered the material compliance tensor as an internal variable state of composite material to be given as
evolving with damage. The approach of reflecting damage
through the fourth-order compliance tensor has gained a oG
e¼ ¼ CðkÞ : r þ ei ; ð1Þ
wide acceptance and popularity and has also been applied or
to other materials such as ceramics [14,15] and concrete where, the strain and average stress tensors are given by e,
[16,17]. and r, respectively, G denotes the Gibbs free energy, C des-
Hansen [13] investigated damage nucleation and growth ignates the material compliance tensor, and k is the cumu-
in glass-reinforced woven composites in both quasi-static lative fatigue damage parameter. Here, the stress tensor r
and fatigue environment. Three categories of specimen represents the far-filed stress condition and therefore
were studied: undamaged, barely visible impact damaged should be interpreted as the average stress over the full
(BVID), and penetrated damaged. By using infrared ther- laminate. The tensor contraction operation is denoted by
mography, as a non-destructive detection method, it was ‘‘:’’ in Eq. (1) and the inelastic component of deformation
found that this method could detect damage development due to damage processes is given by ei. Implied in the con-
in early stages. Based on his experimental testing, Hansen stitutive form of Eq. (1) is that damage is recorded in the
[13] divided the damage processes into three phases. Phase fourth-order material compliance tensor. Designating the
I referred to the early part of the fatigue life when the mod- fatigue cycle number as N, the rate form of Eq. (1) is given
ulus dropped steeply, due to the phenomenon referred to as below where differentiation is taken with respect to N:
the ‘‘knee effect’’. This clear drop in the stiffness has been _
e_ ¼ CðkÞ : r_ þ CðkÞ : r þ e_ i : ð2Þ
attributed to the breakdown of the glass/resin bond in
the roving as well as the cracking within the resin rich areas To account for induced anisotropy, we adopt an addi-
[6]. The knee occurs at about the same stress level for both tive decomposition of the compliance tensor as
weft and warp directions. When the breakdown of the
CðkÞ ¼ C0 þ Cc ðkÞ ð3Þ
bond and cracking saturate, the phase I ends and phase
c
II begins. where, C ðkÞ denotes the added flexibility due to damage
In phase II a different mechanism dominates the damage and C0 corresponds to the initial undamaged state of the
evolution, such as friction between fibers and some delam- material. The rates of the added flexibility and the inelastic
ination. Continuum Damage Mechanics (CDM) is strain tensors are regarded as fluxes in the thermodynamic
assumed to be still valid to simulate the phase II as cracks state sense and are given below with respect to response
to a large extent remain distributed, although a more sig- tensors R and M:
nificant heat production and loss is expected to occur com-
C_ c ¼ kR
_ and _
e_ i ¼ kM: ð4Þ
pared to Phase I. Phase III is considered by Hanson to
signify the onset of localized deformation and failure of To progress further specific forms of the response ten-
the specimen. Such localized patterns of deformation can sors R and M must be specified. One should note that if
not be modeled by CDM alone. R is chosen to be an isotropic tensor, such as the formula-
Hansen [13] subsequently proposed a constitutive model tion implied in Hansen’s work [13], the model will be iso-
for the fatigue damage relation and growth in tensile stress tropic. To predict anisotropic behavior due to damage, R
load paths for phases I and II. His model was multi-axial should be formulated such that anisotropy is achieved.
and produced reasonable results. However, the proposed To do this, we assume that damage takes place in direction
model was isotropic and therefore could not capture the of applied stresses and in tension regimes only. To expand
induced anisotropy effects of cracking. Also lacking was this to other situations is feasible and has been reported
 C. Wen, S. Yazdani / Composite Structures 82 (2008) 127–131 129

elsewhere [14–17]. An elegant mathematical way is to 21

decompose the stress tensor into its positive and negative

cones. The mathematical formulation to achieve this has

longitudinal stiffness (GPa)

proposed model
20
been published in [14,15]. Let the positive cone of the stress Hansen model
experimental results
tensor by given as r+ whereby all the negative eigen-stres-
ses are removed from the stress tensor r. Similarly, let r 19

represent the negative cone of the stress tensor. Note that

r+ + r = r. As it was stated, damage is assumed to occur 18
in the cleavage mode of cracking which is shown schemat-
ically below in Fig. 1. To generalize this and with a further
17
assumption of no coupling between cleavage type cracks in
orthogonal directions, the following response tensors are
postulated for R and M: 16
0 10000 20000 30000 40000 50000 60000 70000
ðrþ  rþ Þ
R¼ þ þ ; ð5Þ number of cyclic
ðr : r Þ
arþ Fig. 2. Comparison of modulus reduction prediction between models and
M¼ 1=2
; ð6Þ experimental results from Hansen’s work [13].
ðrþ : rþ Þ
 þ þ m þ þ
where, the symbol ‘‘’’ signifies the tensor product opera- r :r B ðr  r Þ
e_ ¼ C : r_ þ A N
tion. It should be noted that the material in Hansen’s work r20 ðrþ : rþ Þ
was assumed to be quasi-isotropic laminate, with the lay-  þ þ m
r :r rþ
up sequence of [(+45° # 45°)/(90° # 0°)]s, made up of : r þ aA 2
NB 1=2
: ð10Þ
r0 ðrþ : rþ Þ
plain woven glass/epoxy prepreg fabrics. With this assump-
tion, the strength values in different directions are consid- For computational purposes, the rate of stress tensor is
ered equal. The next step is to establish an evolutionary needed and can be described as:
expression for damage. One particular form proposed by   þ þ m þ þ
 þ þ m
Hansen [13] is given below which represents the damage r :r B ðr  r Þ r :r
r_ ¼ E : e_  A 2
N : r  aA
rate in terms of second invariant of strain tensor. r0 þ þ
ðr : r Þ r20
 n #
ee rþ
k_ ¼ A ; ð7Þ  NB 1=2
: ð11Þ
e0 ðrþ : rþ Þ
with e0 being the reference strain level,
pffiffiffiffiffiffiffiffiand ee being an
effective strain measure given as ee ¼ e : e. Although, this 3. Numerical simulation
form is simple to use, the results predicted were not quite
satisfactory as shown in Fig. 2. In this paper, we propose The model proposed has five material parameters. The
an alternate and new damage evolution law based on the uniaxial tensile stress is used as the reference stress and
second invariant of the stress tensor. Let hence the result of a conventional uniaxial tensile test is
Z N  þ þ m needed to establish r0. The other four parameters are ide-
r :r
k¼A N B dN k 6 k lim ; ð8Þ ally obtained as follows. Since damage, k, in the proposed
0 r20
model is identified as the added flexibility (inverse of stiff-
where A, B, and m are material constants, N is the cyclic ness), therefore the measurements of stiffness and hence
number, and r0 is a reference stress. By differentiating flexibility at three different stages of loading cycle is needed
Eq. (8) with respect to N, the increment of damage in to compute k and hence material parameters, A, B, and m.
one cycle is given as The kinematic parameter, a, is determined by measuring
 þ þ m the permanent deformation during one of the cyclic load-
_k ¼ A r : r N B: ð9Þ
r20 ing stated above. As these measurements are scarce in the
literature, for the numerical simulations, judgments were
Substituting Eqs. (5), (6) and (9) into Eq. (4), and then used to obtained numerical results.
substituting Eqs. (3) and (4) into Eq. (2) yields: The model predictions are compared with Hansen’s work
[13] in Figs. 2 and 3. The experimental data showing the
modulus reduction correspond to a woven glass/epoxy pre-
σ+ σ+ preg fabric composite laminate with the lay-up sequence
[(+45° # 45°)/(90° # 0°)]. For the numerical simulation
the following constants were used. A ¼ 0:85  106 , B ¼
0:7, m ¼ 1:4, rþ ¼ ½155 0 0 MPa, and r0 ¼ 330 MPa.
Fig. 1. Schematic representation of crack opening under tension–tension Parameters A, B, a, and m are estimated by comparing pre-
loading. dicted results and experimental results over a range of
130 C. Wen, S. Yazdani / Composite Structures 82 (2008) 127–131

350 200
applied farfield stress (MPa)

300 experimental results

150
m= 3

stress (MPa)
m= 2.8
250
m= 2.5
100
200

150 50

100
0 20000 40000 60000 80000 100000 120000 140000 160000 0
0 0.003 0.006 0.009 0.012 0.015 0.018
number of cycle to fatigue
strain
Fig. 3. Comparison between experimental S–N curve with predicted S–N
Fig. 5. Stress strain behavior during inelastic damage accumulation.
curves for different values of m (Experimental results from Hansen’s work
[13]).
4. Conclusion
applied stresses. As shown in Fig. 2, the initial damage due
An anisotropic inelastic damage model for woven fabric
to knee effect is captured by the proposed theory quite well.
composites during a low frequency tension–tension fatigue
The model also shows its usefulness in the prediction of
is established. Since the fatigue damage in most fatigue life
phase II of fatigue damage processes.
is mainly due to development of a multitude of matrix
The predicted S–N curve of the BVID composite is com-
cracking and interfacial debonding, a class of damage
pared with the experimental S–N curve in Fig. 3. The effect
mechanics theories is used for modeling fatigue processes.
of material parameter, m, on the S–N curve is also shown on
An expression for the evolution of fatigue damage is pro-
the same figure. It is seen that the general trends of the S–N
posed based on the second invariant of the stress tensor.
curve are replicated satisfactorily by the proposed model.
This form plus the proposed expression for the response
The stress–strain cyclical behavior for the composite
tensors were used to demonstrate the model’s capability
material is further illustrated in Figs. 4 and 5. Fig. 4 repre-
in capturing the essential features of composite material
sents a behavior where no permanent deformation occurs
inelasticity subjected to fatigue loading environment. The
even though damage is progressively accumulating as evi-
comparison with experimental data is shown in the paper
denced by degradation of elastic modulus. The behavior
with good agreement.
corresponds to an idealized case whereby crack faces close
perfectly upon unloading and is achieved in the model by
References
letting M ¼ 0. In most heterogeneous materials, such as
glass-fiber composites, permanent deformations do take [1] Boutman LJ, Sahu S. Mechanical properties of particulate compos-
place. The versatility of the model is shown in Fig. 5 where ites. Polym. Eng. Sci. 1972;12(March):91–100.
the process of elastic degradation and permanent deforma- [2] Owen MJ. Fatigue processes in fibre reinforced plastics. Philos.
tion is illustrated. Trans. Ser. A 1980;294(1411):535–43 [January 21].
[3] Owen MJ. Fatigue testing of fiber reinforced plastics. Composites
1970;1:346–55.
200
[4] Mandell J, Cairns D, Samborsky D, Morehead R, Haugen D.
Prediction of delamination in wind turbine blade structural details.
In: Proc of 41st AIAA Aerospace Sciences Meeting and Exhibit,
150
Reno, NV; UNITED STATES; 6–9 January 2003. pp. 202–13.
[5] Mandell JF, Grande DH, Tsiang T-H, McGarry FJ. Modified mirco-
stress (MPa)

debonding test for direct in situ fiber/matrix bond strength determi-

nation in fiber composites. In: Proc of Composite materials: testing
100 and design, Seventh Conference; Philadelphia, Pennsylvania; USA; 2–
4 April 1984. pp. 87–108.
[6] Smith EW, Pascoe KJ. Biaxial fatigue of a glass-fiber reinforced
composite I. Fatigue and Fracture Behavior. Biaxial and Multiaxial
50 Fatigue: Mechanical Engineering Publications; pp. 367–96.
[7] Highsmith AL, Reifsnider KL. Stiffness-reduction mechanisms in
composite laminates Damage in composite materials: Basic mecha-
nisms, accumulation, tolerance, and characterization. (A83-14551 03-
0
0 0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 24) Philadelphia, PA, Am Soc Test Mater 1982:103–17.
strain [8] Degrieck J, Van Paepegem W. Fatigue damage modeling of fiber-
reinforced composite materials: Review. Appl Mech Rev (USA). July
Fig. 4. Stress strain behavior for elastic-damaging process. 2001; Vol. 54, no. 4, pp. 279–300.
 C. Wen, S. Yazdani / Composite Structures 82 (2008) 127–131 131

[9] Yoshioka K, Seferis JC. Modeling of tensile fatigue damage in resin [13] Hansen U. Damage development in woven fabric composites during
transfer molded woven carbon fabric composites. Compos A tension–tension fatigue. J Compos Mater 1999;33:614–39.
2002:1593–601. [14] Ortiz M, Giannakopoulos AE. Crack propagation in monolithic
[10] Chou TW, Ko FK. Textile structural composites. Amsterdam, the ceramics under mixed mode loading. Int J Fracture
Netherlands: Elsevier Science Publishers B.V; 1989. (Sara Burger- 1990;44(4):233–58. August 15.
hartstraat 25, P.O. Box 211, 1000 AE). [15] Ortiz M, Giannakopoulos AE. Mixed mode crack-tip fields in
[11] Huang ZM. Fatigue life prediction of a woven fabric composite monolithic ceramics. Int J Solids Struct 1990;26(7):705–23.
subjected to biaxial cyclic loads. Compos A 2002:253–66. [16] Yazdani S. On a class of continuum damage mechanics theories. Int J
[12] Degrieck J, Paepegem WV. Coupled residual stiffness and strength Damage Mech 1993;2:162–76.
model for fatigue of fiber-reinforced composite materials. Compos Sci [17] Karnawat S, Yazdani S. Effects of preloading on brittle solids. J Eng
Technol 2002;62:687–96. Mech 2001;127(1):11–7.