You are on page 1of 22

Journal of Composite

Materials http://jcm.sagepub.com/

Effects of Fiber Arrangement on Mechanical Behavior of Unidirectional Composites


Yuanchen Huang, Kyo Kook Jin and Sung Kyu Ha
Journal of Composite Materials 2008 42: 1851 originally published online 17 July 2008
DOI: 10.1177/0021998308093910

The online version of this article can be found at:


http://jcm.sagepub.com/content/42/18/1851

Published by:

http://www.sagepublications.com

On behalf of:

American Society for Composites

Additional services and information for Journal of Composite Materials can be found at:

Email Alerts: http://jcm.sagepub.com/cgi/alerts

Subscriptions: http://jcm.sagepub.com/subscriptions

Reprints: http://www.sagepub.com/journalsReprints.nav

Permissions: http://www.sagepub.com/journalsPermissions.nav

Citations: http://jcm.sagepub.com/content/42/18/1851.refs.html

>> Version of Record - Aug 18, 2008

OnlineFirst Version of Record - Jul 17, 2008

What is This?
Downloaded from jcm.sagepub.com at UNIV OF VIRGINIA on July 16, 2012
Effects of Fiber Arrangement on
Mechanical Behavior of
Unidirectional Composites

YUANCHEN HUANG, KYO KOOK JIN AND SUNG KYU HA*


Hanyang Structures and Composites Laboratory, Department of Mechanical
Engineering, Hanyang University, #1271, Sa 1-dong, Sangnok-gu, Ansan-si
Gyeonggi-do, 426-791, South Korea

ABSTRACT: Micromechanical approaches are employed to investigate the


influence of different fiber arrangement on the mechanical behavior of unidirectional
composites (UD) under various loading conditions. A micromechanical model with a
random fiber array is generated and used in a finite element analysis together with
two frequently used representative volume elements (RVE), or unit cell models of
square and hexagonal arrays. The algorithm for generating the random fiber array is
verified by comparing the comprehensive performance of a unit cell based on our
random array and that of a unit cell based on a real fiber distribution in the UD
cross-section. Performance of the random and regular fiber arrays is also evaluated
through frequency distributions of stress invariants in matrix and tractions at the
fiber–matrix interface due to various loading types. The effects of different loading
angles on the overall response of regular arrays to various loading conditions are
investigated thoroughly. Finally, the Weibull distribution of the maximum normal
interfacial traction in random array is compared with the cumulative probability
distribution of transverse strength data acquired from experiment, and good
agreement is achieved.

KEY WORDS: random fiber array, square array, hexagonal array, representative
volume element (RVE), unit cell, statistics, unidirectional composites.

INTRODUCTION

OR MOST ENGINEERING applications, unidirectional composites are regarded as


F transversely isotropic homogeneous materials from a macroscopic view, and theories
based on this assumption are well tested and widely used. The advantage of a macro-
scopic approach is that, instead of dealing with a complicated internal micro structure
and interactions between constituents, it focuses on the macro-level behavior of UD
and works well for a large selection of materials in general cases. However, the

*Author to whom correspondence should be addressed. E-mail: sungha@hanyang.ac.kr


Figures 1, 9–13, 15 and 17 appear in color online: http://jcm.sagepub.com

Journal of COMPOSITE MATERIALS, Vol. 42, No. 18/2008 1851


0021-9983/08/18 1851–21 $10.00/0 DOI: 10.1177/0021998308093910
ß SAGE Publications 2008
Los Angeles, London, New Delhi and Singapore
Downloaded from jcm.sagepub.com at UNIV OF VIRGINIA on July 16, 2012
1852 Y. HUANG ET AL.

disadvantage of a macroscopic approach is obvious: since fiber and matrix have


completely different behaviors, there must be fluctuation in the micro stress/strain
distribution, which is ignored by the hypothesis of homogeneity. Therefore, important
mechanisms related to the interaction between constituents might be neglected, some of
which probably play substantial roles in failure analysis. So, micromechanics is utilized to
reveal detailed information in UD.
Micromechanical approaches have been employed for decades as powerful tools to
investigate the local stress/strain state within composites from a microscopic view. One
common problem for researchers is to establish a proper model so that the micro-
structure of the composite is well characterized. It is expected that by using such a
model in the finite element analysis, both the effective material properties and failure
strengths of a composite can be predicted with high accuracy, and all information
required is constituent properties, as well as the fiber volume fraction. A number of
analytical studies have been conducted in this area [1–7]. Recently several finite element
models, such as the representative volume element (RVE), and unit cells based on
square or hexagonal fiber arrays, have been developed [8–11]. Li gave an extensive
summary of the features of square and hexagonal arrays [12]; Whitcomb and Tang
employed both square and hexagonal arrays to study moisture diffusion in composites
with impermeable fibers [13]; Zhang et al. devised a unit cell of cross-ply laminate based
on the square array to study residual thermal stresses/strains in composites [14]; Bonora
and Ruggiero presented the development of unit cells for MMC based on regular arrays
and different failure mechanisms [15,16]; Drago and Pindera discussed appropriate
boundary conditions for a repeating unit cell (RUC) comprised of multiple square array
unit cells and used it to predict engineering moduli of unidirectional composites [17];
Cai et al. utilized unit cells of regular arrays to predict the long-term fatigue strength of
CFRP structures [18].
However, the basis of the FE models mentioned above is the postulation that the fiber
arrangement in composite follows a regular, periodic pattern, which deviates from reality.
Actually, although fibers are longitudinally unidirectional in UD, they distribute
randomly in the cross-sectional view. In other words, the implementation of a regular
fiber array is a compromise between computational cost and accuracy. Therefore, the
effect of different fiber arrangements on the mechanical behavior of unidirectional
composites needs evaluation. Everett and Chu pointed out microstructure characteristics
of non-uniform composites, which merit study in modeling [19]; Chen and Papathanasiou
generated multi-fiber RVE with different homogeneity using the Monte Carlo algorithm,
and investigated the interface stress distribution [20]; Wangsto and Li studied the effect of
different boundary conditions on the response of a unit cell of random array to external
loading and concluded that incorrect boundary conditions only affect results in regions
near boundaries [21]; the distribution of interfacial strain and thermal residual stress
in unit cell of random array were studied by Ha et al. [22,23]. Figure 1 shows a comparison
of fiber arrangement in a real case and two idealized regular fiber arrays.
In this study, micromechanical unit cell models of random fiber array and regular fiber
arrays (square and hexagonal) are generated. The validity of the random fiber array from a
statistical viewpoint is confirmed. The performance of different fiber arrays in various aspects
are compared, including the prediction of effective material properties, stress invariants,
and interfacial traction distribution, which in most cases contribute to failure initiation.

Downloaded from jcm.sagepub.com at UNIV OF VIRGINIA on July 16, 2012


Effects of Fiber Arrangement on Mechanical Behavior of Unidirectional Composites 1853
z

(a)

z z

(b)

y y

: Fiber : Matrix

Figure 1. Comparison of a real fiber arrangement and two idealized fiber arrays: (a) real fiber array;
(b) square and hexagonal fiber arrays.

METHOD

Generation of Random Fiber Array

For a given fiber volume fraction, the number of fibers can be determined by the
equation:

nf ðr2f Þ
vf ¼ ð1Þ
A

where vf, nf, and rf are the fiber volume fraction, number of fibers and fiber radius,
respectively, while A is the area of cross-section of the unit cell region. Generally speaking,
vf is a known parameter, and the diameter of carbon fiber is between 5 and 8 mm. Thus,
there remain only two user-adjustable parameters: nf and A. However, nf ought to be an
integer, and a square cross-section is preferred in practice. Nevertheless, from a statistical
perspective, small fluctuation is allowed, which means vf does not need to be rigorously
equal to the given value, so that the determination of nf and A becomes quite flexible.
Another problem worthy of attention is that there must be a sufficient number of fibers to

Downloaded from jcm.sagepub.com at UNIV OF VIRGINIA on July 16, 2012


1854 Y. HUANG ET AL.

exhibit some statistical properties of a system. The algorithm for the automatic generation
of a random fiber array is briefly stated below:
1. Determine values of all three parameters on the right-hand-side of the Equation (1),
so that the calculated vf is as close as possible to the given value;
2. Select an arbitrary point within the domain of area A;
3. Take the selected point as the center of circle, and generate a circle of radius rf,
which represents the perimeter of the fiber cross-section;
4. Permanently exclude all points located in the circular region, which has the same center
as the circle generated in step 2, and with a radius of 2rf, from being selected;
5. Select another point residing in the rest domain as the center of a second fiber so that
there is no overlap between the current fiber and previously generated boundaries;
6. Repeat steps 3–5 until all fibers are generated.
A schematic presentation of this procedure is illustrated in Figure 2. It might be desir-
able to include as many fibers as possible in one unit cell to better reflect fiber configu-
ration in the micro scale, but in reality it may be limited by computational time and
capability.

Real Fiber Array

Since random fiber array is regarded as the closest representation to real cases, and is
probably used as a standard to check the accuracy of other micromechanical models, it is
necessary to validate the algorithm used for generating a random fiber array. In this case, a

rf
× ×
z z

y y

2rf 2rf
rf rf
z × z ×

y y

Figure 2. General procedures of generating a random fiber array.

Downloaded from jcm.sagepub.com at UNIV OF VIRGINIA on July 16, 2012


Effects of Fiber Arrangement on Mechanical Behavior of Unidirectional Composites 1855

real fiber array is used as a comparative example. A photograph of the cross-section of a


certain UD ply shows the arrangement of fibers in the matrix. This photograph is
duplicated and transformed into a computer-recognizable format so that a FE model, in
which all information, such as the location and diameter of each fiber, is well preserved,
and can be easily generated for further analysis. The fiber diameter and fiber volume
fraction measured from the photograph are 6.75  0.33 and 0.608  0.018 mm, respectively.
Because this photograph shows only a small region of the entire cross-section, and
statistical fluctuation is expected for those measurements. Thus, in order to compare
results from a random array and those from a real fiber array, the random array with
vf ¼ 0.6 is appropriate owing to the similar fiber diameter and vf. Figure 3 shows both the
original photograph and computer-generated model, and it can be clearly observed that
the original image is perfectly reproduced.

Finite Element Model

The evaluation of the effect of fiber arrangements on the mechanical behavior of UD


should be both qualitative and quantitative. To achieve this goal, finite element analysis
must be performed for models with different fiber configuration: real array, random array,
square array, and hexagonal array. FE models of regular arrays have already been used by
many researchers for years, but models for the former two arrays are completely new.
Unlike the unit cell model of a regular array, it is impossible to extract a similar unit cell
that contains only one fiber from either a random or real array, because such an action will
eliminate all statistical information. Hence, unit cell models of random and real fiber
arrays are built by simple extrusion of 2-D fiber array images. Perfect bonding is presumed
between fiber and matrix, meaning that there will be no separation at the fiber–matrix
interface. Compromises are made between accuracy and computational cost. For each
fiber, there are 24 nodes along its circumference with equal spacing, and then a free-mesh
is generated inside the matrix and fiber. One example of a finite element model of random
fiber array with vf ¼ 0.6 is shown in Figure 4.
Moreover, since interfacial traction is involved in this study, the definition of three
components is clearly shown in Figure 5. The cylinder represents a small section of fiber

(a) (b)

Figure 3. A computer generated model by image processing of original photograph of the real fiber array:
(a) photos of real cross section; (b) image based model.

Downloaded from jcm.sagepub.com at UNIV OF VIRGINIA on July 16, 2012


1856 Y. HUANG ET AL.

(a) (b)

z z

y y

(c) (d)

z z

y y
x x

Figure 4. FE modeling of random fiber array (vf ¼ 0.6, No. of fibers ¼ 120): (a) 2D random array; (b) mesh
generation; (c) 3D fiber and (d) 3D matrix.

z tn

tt
x
tx
θ
y
r

tn: Normal traction


tt : Tangential traction
tx: Longitudinal traction

Figure 5. Traction vectors at the fiber–matrix interface.

with three interfacial traction components at an arbitrary point situated at the fiber–
matrix interface: normal traction tn being perpendicular to the tangential plane passing
through that point; tangential traction tt being perpendicular to both tn and the central
axis of the fiber; and longitudinal traction tx being parallel to the central axis of the fiber
and perpendicular to both tn and tt. Note that those stress-based interfacial tractions are
continuous from fiber to matrix, while interfacial strains are not.

Downloaded from jcm.sagepub.com at UNIV OF VIRGINIA on July 16, 2012


Effects of Fiber Arrangement on Mechanical Behavior of Unidirectional Composites 1857
(a) (b) (c)

Figure 6. Generated RVE of random fiber array for various fiber volume fractions: (a) Volume fraction 0.4;
(b) volume fraction 0.5 and (c) volume fraction 0.6.

RESULTS AND DISCUSSION

Random Fiber Array

Following the procedures presented in the previous section, a random fiber array is
generated. The fiber volume fraction vf varying between 0.1 and 0.6, and a value of 3.75 mm
for rf, are considered in this study. Figure 6 shows random fiber arrays with fiber volume
fractions of 0.4, 0.5, and 0.6. The number of fibers varies depending on vf, which is 120 when
vf ¼ 0.6. Two individually generated arrays with equal vf are in one column. Obvious
similarities are not noted in the two random arrays with the same vf, but these phenomena
cannot prove the validity of the algorithm. Further numerical analysis is needed.

Comparison Between Random and Real Array

To verify the algorithm for generating a random fiber array, frequency distributions of
interfacial traction components in both the random array and real array are plotted. For
one fiber in random fiber array, according to the FE model, there are 24 nodes at the fiber–
matrix interface, and at each point the value of local interfacial traction can be obtained.
Therefore, a total of 2880 nodes are available to collect interfacial traction data in a
random array with vf ¼ 0.6. In the real fiber array, the number of nodes for data collection
is slightly different because the number of fibers cannot be arbitrarily designated though
the mesh density can be controlled. Material properties used in FEA are listed in Table 1.

Downloaded from jcm.sagepub.com at UNIV OF VIRGINIA on July 16, 2012


1858 Y. HUANG ET AL.

Table 1. Material properties of fiber and matrix [24].

Material properties Carbon fiber Material properties Epoxy


E11 (GPa) 303 Em (GPa) 3.31
E22 ¼ E33 (GPa) 15.2 vm 0.35
G12 ¼ G12 (GPa) 9.65 m (106/8C) 57.6
v21 0.2
v32 ¼ vTT 0.2
1 (106/8C) 0.0
2 ¼ 3 (106/8C) 8.3

(a) 0.1 (b) 0.2 (c) 0.08


σ2=1Pa σ6=1Pa ∆T = −100°C

0.08 0.16
0.06 Real
Normalized frequency

Random
0.06 0.12
0.04
0.04 0.08

0.02
0.02 0.04

0 0 0
−1 −0.5 0 0.5 1 1.5 2 −4 −3 −2 −1 0 1 2 3 −60 −40 −20 0 20 40
Normal traction, tn(Pa) Longitudinal traction, tx(Pa) Normal traction, tn(Pa)

Figure 7. Frequency distribution of tractions at the fiber–matrix interface in random and real fiber arrays due to
various loading: (a) normal traction due to transverse loading; (b) longitudinal traction due to longitudinal
shear loading and (c) normal traction due to thermal loading.

In Figure 7, there are three subplots showing the frequency distributions of interfacial
traction components under different loading conditions. Three typical loads, transverse
tensile stress, longitudinal shear stress, and temperature increment, are applied to unit cell
models of random and real arrays. Since transverse tension and temperature increment will
greatly affect the magnitude of normal traction, and likewise, longitudinal traction is also
greatly influenced by longitudinal shear, the statistical distribution of normal interfacial
traction due to transverse tension and temperature increment, together with longitudinal
interfacial traction due to longitudinal shear, are plotted.
In all three subplots, the results from random and real fiber arrays are in excellent
accordance, much better than the authors originally expected. Clearly, the validity of the
algorithm used for generating random fiber array is validated.

Effective Material Properties

Since the performance of the random fiber array has been tested, it can be set as a
standard to judge the accuracy of other micromechanical models. First, the effective
material properties predicted by unit cell models of two regular arrays are examined.
Figures 8(a)–(h) show plots of different effective material properties versus vf, which

Downloaded from jcm.sagepub.com at UNIV OF VIRGINIA on July 16, 2012


Effects of Fiber Arrangement on Mechanical Behavior of Unidirectional Composites 1859
(a) 200 (b) 10

150 8
E1(GPa)

E2(GPa)
6
100
Sqr. 4 Sqr.
50 Hex. Hex.
SROM 2 SROM
Random Random
0 0
0 0.1 0.2 0.3 0.4 0.5 0.6 (vf) 0 0.1 0.2 0.3 0.4 0.5 0.6 (vf)

(c) 0.4 (d) 4

0.3 3

G12(GPa)
ν21

0.2 2
Sqr. Sqr.
Hex. Hex.
0.1 1
SROM SROM
Random Random
0 0
0 0.1 0.2 0.3 0.4 0.5 0.6 (vf) 0 0.1 0.2 0.3 0.4 0.5 0.6 (vf)

(e) 0.6 (f) 3.5


0.5 3
2.5
0.4
G23(GPa)

2
ν32

0.3
1.5
0.2 Sqr. Sqr.
1
Hex. Hex.
0.1 Random 0.5 Random
0 0
0 0.1 0.2 0.3 0.4 0.5 0.6 (vf) 0 0.1 0.2 0.3 0.4 0.5 0.6 (vf)

(g) 6 (h) 80
5 Sqr. Sqr.
Hex. 60 Hex.
ag (10−6/°C)

ag (10−6/°C)

4 Random Random
3 40
2
20
1
0 0
0 0.1 0.2 0.3 0.4 0.5 0.6 (vf) 0 0.1 0.2 0.3 0.4 0.5 0.6 (vf)

Figure 8. Comparison of effective material property predictions according to fiber volume fraction:
(a) longitudinal modulus, E 1 (GPa); (b) transverse modulus, E 2 (GPa); (c) major Poisson’s ratio, 21 ;
(d) longitudinal shear modulus, G  12 (GPa); (e) transverse-transverse Poisson’s ratio, 32 ; (f) transverse-
 23 (GPa); (g) longitudinal CTE,  1 (106/8C) and (h) transverse CTE,  3(106/8C).
transverse shear modulus, G

ranges from 0.1 to 0.6. In each plot, the results of square, hexagonal and random arrays
are presented for comparison.
Through an overview, in every subplot, predictions of three models with different fiber
arrangement agree fairly well, but some clear differences still exist in particular cases,
which reflect the effect of fiber arrangement. For effective transverse Young’s modulus E 2
and longitudinal shear modulus G 12 , the values predicted by the hexagonal array are in

Downloaded from jcm.sagepub.com at UNIV OF VIRGINIA on July 16, 2012


1860 Y. HUANG ET AL.

excellent agreement with those by the random array, while slightly larger values are
obtained in the square array. For effective transverse–transverse Poisson’s ratio 32 and
shear modulus G 23 , the values noted by the hexagonal array are also very similar to those
by the random array, while the prediction of the square array is the smallest. Note that the
predictions from the simple rule of mixture, denoted by SROM in Figures 8(a) to 8(d),
are far below other predictions in the transverse and longitudinal shear stiffness. The
coefficients of thermal expansion are in excellent agreement between the three arrays.

Distribution of Stress Invariants

Stress invariant distributions in the random fiber array and two regular arrays with
vf ¼ 0.6 under typical loading conditions were also investigated. Since a number of failure
theories for the composite materials have emphasized the importance of stress invariants,
especially von Mises stress  VM and the volumetric stress invariant I1, contour plots of the
stress invariants in three fiber arrays subject to transverse tension, longitudinal shear, and
temperature increment are compared in Figure 9 [24,25]. Three models in each subplot are
subject to the same loading and contours have the same scale. Contours due to transverse
tension, longitudinal shear, and temperature increment are arranged from top to bottom,
respectively, while the left column contains contours of  VM and the right column contains
contours of I1. Unlike in the regular array, invariant distribution in the random array does
not follow any specific periodic pattern.
By observation, a high stress concentration appears wherever fibers congregate,
whereas a low stress concentration appears in matrix-rich area. Regions with maximum
concentration, whose value exceeds the maximum in the regular array in most cases,
scatter over the entire cross-section of the unit cell of random fiber array. In this sense, ply
strength is always overestimated if the analysis is based on the unit cell model of a regular
fiber array, because local failure due to random fiber arrangement might be omitted.
However, it is expected in real composites that such local stress concentrations might be
reduced due to highly concentrated local plastic deformation and some interfacial stiffness.
Therefore, the statistical distribution of stresses needs to be obtained and evaluated to
judge the characteristics of the random fiber array.

EFFECT OF LOADING ANGLE FOR RANDOM AND REGULAR ARRAYS

Although the macroscopic behavior of unidirectional composites can be recognized as


transversely isotropic, due to the effect of fiber arrangement at the micro level, the failure
strength of the unit cell cannot be regarded as transversely isotropic. Many researchers
take the normal direction of a certain unit cell side boundary face as the default loading
direction for transverse tension/compression or longitudinal shear. However, in the
macroscopic view, so-called transverse loading could be applied along any direction that
is perpendicular to the fiber direction; likewise, longitudinal shear can also be applied
in any plane whose normal direction is perpendicular to the fiber direction. Therefore,
by intuition, it is necessary to evaluate the effect of loading angle on the response of a
unit cell to macro transverse tension, transverse compression and longitudinal shear.
Taking the unit cell of hexagonal array as an example, as shown in Figure 10(f), the
procedure of testing is as follows: apply macro stress (transverse tension/compression or

Downloaded from jcm.sagepub.com at UNIV OF VIRGINIA on July 16, 2012


Effects of Fiber Arrangement on Mechanical Behavior of Unidirectional Composites 1861
(a) Random Sqr. (b) Random Sqr.

Hex. Hex.

z z

y y

(c) Random (d)


Sqr.

In the longitudinal shear loading,


Hex. the first stress invariant I1 is zero
over the entire matrix region.
z

(e) Random Sqr. (f) Random Sqr.

Hex. Hex.

z z

y y

Figure 9. Contours of stress invariants in regular and random fiber arrays due to unit mechanical and thermal
loading (vf ¼ 0.6, unit: Pa): (a) von Mises stress,  VM, due to transverse tensile loading (1 Pa); (b) first stress
invariant, l1, due to transverse tensile loading (1 Pa); (c) von Mises stress,  VM, due to longitudinal shear
loading (1 Pa); (d) first stress invariant, l1, due to longitudinal shear loading (1 Pa); (e) von Mises stress,  VM,
due to thermal loading (T¼18C) and (f) first stress invariant, l1, due to thermal loading (T¼18C).

longitudinal shear) in the ply coordinate system y0 –o–z0 and maintain this coordinate
system in a steady state; rotate the unit cell gradually around fiber direction, which is
perpendicular to the y0 –o–z0 plane, and there will be an angular difference  noted as the
‘loading angle’ or ‘unit cell direction angle’, between the unit cell coordinate system y–o–z
and the ply coordinate system y0 –o–z0 ; multiply a stress transformation matrix by macro
stress, and the stress components applied to the unit cell at  can be readily obtained;
calculate micro stresses at every node in the fiber region and matrix region in the unit cell,
as well as interfacial tractions at every node on the fiber–matrix interface, through either
finite element analysis or the stress amplification factor, a quicker and simpler alternative
of FEA introduced in the authors’ previous work [25]; give another increment to  and
repeat previous steps, until 1808 is covered. Due to the periodical variation of the distance

Downloaded from jcm.sagepub.com at UNIV OF VIRGINIA on July 16, 2012


1862 Y. HUANG ET AL.

(a) (b)
1.4 4
Max. von Mises stress, σVM,max (Pa) Transverse tensile loading (1Pa) Transverse tensile loading (1Pa)
1.3 3.5

Max. first invariant, I1 (Pa)


97%
1.2
97% 3
1.1
2.5
1
2
0.9
86%
89% 1.5
0.8 cumulative
Sqr. frequency
0.7 Hex. 1
for random
0.6 0.5
0 20 40 60 80 100 120 140 160 180 0 20 40 60 80 100 120 140 160 180
Unit cell direction angle, Θ Unit cell direction angle, Θ

(c) (d)
1.4 −0.5
Max. von Mises stress, σVM,max (Pa)

Transverse compressive loading (−1Pa) Transverse compressive loading (−1Pa)


1.3 −1

Min. first invariant, I1 (Pa)


1.2 97% −1.5
86%
1.1
−2
1
−2.5
0.9
89% −3
0.8
97%
0.7 −3.5

0.6 −4
0 20 40 60 80 100 120 140 160 180 0 20 40 60 80 100 120 140 160 180
Unit cell direction angle, Θ Unit cell direction angle, Θ

(e)
4 (f)
Max. von Mises stress, σVM,max (Pa)

Longitudinal shear loading (1Pa) z′


3.5 z
θ
3 95%
y
2.5
θ
2
89% y′

1.5

1
0 20 40 60 80 100 120 140 160 180
Unit cell direction angle, Θ

Figure 10. Variation of maximum and minimum matrix stress invariants in regular arrays along the unit cell
direction angle due to various loading (vf ¼ 0.6): (a) maximum von Mises stress; (b) maximum first invariant;
(c) maximum von Mises stress; (d) minimum first invariant; (e) maximum von Mises stress and (f) unit cell and
ply directions.

between two adjacent fibers in two regular arrays, periodical results are expected. Actually,
the effect of different loading angles is also checked for the random array.
Figures 10(a)–(e) illustrate variations of the maximum von Mises stress  VM,max and
first stress invariant I1,max with , in the matrix of two regular arrays under unit macro
stress (1 Pa). All results show periodicity as expected: results of the square array have

Downloaded from jcm.sagepub.com at UNIV OF VIRGINIA on July 16, 2012


Effects of Fiber Arrangement on Mechanical Behavior of Unidirectional Composites 1863

a period of 908 while results of the hexagonal array have a period of 608. Those two angles
directly reflect the characteristics of the fiber arrangement in the square and hexagonal
arrays.
There are some interesting phenomena worthy of attention. In Figure 10(a), it can be
observed that at  ¼ 08, which is the most commonly used loading (unit cell) direction,
 VM,max of the square array reaches its global minimum in this domain, while  VM,max of
the hexagonal array reaches its local (not global) minimum in the same domain, which is
about 0.1 Pa greater than the previous value, when the macro loading is transverse
tensile stress. By rough observation, averages of  VM,max in both regular arrays are close to
the value of macro loading, i.e. 1 Pa. Although the macro loading is changed to transverse
compressive stress in Figure 10(c), the variations of  VM,max in both regular arrays
are identical to those in Figure 10(a), because macro loadings have the same absolute
value, and the negative sign of compressive stress does not affect the value of  VM,max.
In both figures, the range of  VM,max variation in the square array is much wider than that
in the hexagonal array.
Figures 10(b) and (d) show variations of I1,max in regular arrays due to macro loading of
transverse tensile and compressive stress, respectively. In Figure 10(b), I1,max of regular
arrays reach their respective global maxima when  ¼ 08, but the hexagonal array shows
about a 0.2 Pa lower value than the square array. Moreover, I1,max of regular arrays also
have similar average values: about 2.5 Pa. Though the only difference between macro
loadings in Figures 10(b) and (c) is the sign, by definition of I1,max, it is easy to predict that
curves in those two figures are symmetric about the axis I1,max ¼ 0. Therefore, I1,max in
Figure 10(d) actually denotes the minimum first stress invariant in the matrix if the
negative sign is taken into consideration. As seen in Figures 10(a) and (c), the variation
I1,max in the square array is also wider than that in the hexagonal array.
Under macro longitudinal shear stress, by definition, I1,max is zero in the two regular
arrays at all possible loading angles. Judging from Figure 10(e),  VM,max of the two regular
arrays reach their respective global maxima when  ¼ 08. Due to its closer fiber
arrangement, the square array tends to predict higher  VM,max than the hexagonal array.
As mentioned previously,  ¼ 08 is always taken as the default loading angle, but it gives
only one possible result, which cannot represent the overall response of regular arrays to
external loading. Therefore, it is highly recommended to consider all possible loading
angles so that a clear picture of all possible results can be acquired.
From Figures 10(a)–(e), the upper and lower bounds of  VM,max or I1,max predicted by
regular arrays are marked with horizontal dashed lines, and the cumulative frequency of
 VM or I1 predicted by random array at each bound value is indicated. No matter at the
lower bound or upper bound in each figure, the cumulative frequency is higher than 0.85
and lower than 0.97, which means that at the point where the cumulative frequency of  VM
or I1 predicted by regular arrays reaches 1, the cumulative frequency of  VM or I1 predicted
by the random array ranges between 0.85 and 0.97. This result indicates that the
performances of regular and random arrays are quite similar.
The values of  VM,max and I1, max due to hydrostatic pressure (2 ¼ 3 ) are obtained
along with the loading angles as shown in Figure 11. Such values are constant over the
loading angles as expected, but are different in the square and hexagonal arrays. Figure 12
shows the effective transverse Young’s modulus E 2 and transverse–transverse shear
modulus G 23 calculated in square and hexagonal arrays at different . Apparently, E 2 and
G 23 in the square array vary with  in a period of 908, while E 2 and G 23 in the hexagonal
array do not vary with . When  ¼ 08, the angle used to calculate the effective material

Downloaded from jcm.sagepub.com at UNIV OF VIRGINIA on July 16, 2012


1864 Y. HUANG ET AL.

(a) 0.7 (b) 3


Hydrostatic pressure (1Pa) Hydrostatic pressure (1Pa)
Max. von Mises stress, σVM,max (Pa)

0.6

Max. first invariant, I1 (Pa)


2.8
0.5
2.6
0.4

0.3 2.4
Sqr. Sqr.
0.2
Hex. Hex.
2.2
0.1

0 2
0 20 40 60 80 100 120 140 160 180 0 20 40 60 80 100 120 140 160 180
Unit cell direction angle, Θ Unit cell direction angle, Θ

Figure 11. Variation of maximum stress invariants of matrix in regular arrays along the unit cell direction angle
due to hydrostatic pressure (vf ¼ 0.6): (a) maximum von Mises stress; (b) maximum first invariant.

(a) 10 (b) 3.5


9
3
8
7 2.5
G23 (GPa)

6
E2 (GPa)

2
5
1.5
4
3 1
Sqr. Sqr.
2 Hex. Hex.
0.5
1
0 0
0 20 40 60 80 100 120 140 160 180 0 20 40 60 80 100 120 140 160 180
Unit cell direction angle, Θ Unit cell direction angle, Θ

Figure 12. Variation of effective material property predictions in regular arrays along the unit cell direction
 23(GPa).
angle (vf ¼ 0.6): (a) transverse modulus, E 2 (GPa); (b) transverse–transverse shear modulus, G

properties shown in Figure 8, E 2 predicted by the square array reaches its maximum and is
larger than the value of E 2 predicted by hexagonal array; conversely, G 23 predicted by the
square array reaches its minimum and is smaller than G 23 predicted by the hexagonal
array. When  ¼ 458, at which angle the square array is equivalent to the so-called
diamond array, E 2 predicted by the square array reaches its minimum. Obviously, results
from the hexagonal array yield an average value different from that of the square array.
The cumulative frequency distributions of  VM and I1 in matrix due to transverse tensile
and longitudinal shear stresses are also computed using the random and regular arrays and
are shown in Figures 13(a)–(d). The cumulative frequency distributions of regular arrays
generally reach one earlier than the distribution of the random array does, but this small
difference does not change the high resemblance between those distributions. Figures 13(e)
and (f) show the cumulative frequency distributions of  VM and I1 in the matrix of random
and regular arrays due to the temperature increment, respectively. The data taken
from regular arrays distribute in a narrower range than data from random arrays.
In particular, distributions representing results from the unit cell of hexagonal array are

Downloaded from jcm.sagepub.com at UNIV OF VIRGINIA on July 16, 2012


Effects of Fiber Arrangement on Mechanical Behavior of Unidirectional Composites 1865
(a) 1.2 (b)
Transverse tensile loading (1Pa) Transverse tensile loading (1Pa)
1
Cumulative frequency

Sqr. Sqr.
0.8
Hex. Hex.
Random Random
0.6

0.4

0.2

0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
von Mises stress, σVM (Pa) First invariant, I1 (Pa)

(c) 1.2 (d)


Longitudinal shear loading (1Pa)

1
Cumulative frequency

0.8 Sqr.
Hex. In the longitudinal shear
Random loading, the first stress
0.6
invariant I1 is zero over the
entire matrix region.
0.4

0.2

0
0 1 2 3 4 5 6
von Mises stress, σVM (Pa)

(e) 1.2 (f)


Thermal loading (∆T=1°C) Thermal loading (∆T=1°C)
1
Cumulative frequency

0.8

0.6 Sqr. Sqr.


Hex. Hex.
0.4 Random Random

0.2

0
0 0.1 0.2 0.3 0.4 0.5 0.6 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6
von Mises stress, σVM (Pa) First invariant, I1 (MPa)

Figure 13. Cumulative frequency distribution of matrix stress invariants of matrix in regular and random arrays
due to various loadings (vf ¼ 0.6): (a) von Mises stress distribution; (b) first invariant distribution; (c) von Mises
stress distribution; (d) first invariant distribution; (e) von Mises stress distribution and (f) first invariant
distribution.

very steep in the rising stage, and have sharp rather than smooth transitions when the
cumulative frequency is near 0 or 1.
Figure 14 supports the validity of regular arrays. Mode stress is defined as the stress
that occurs most frequently. Therefore, mode stress invariants in square, hexagonal,

Downloaded from jcm.sagepub.com at UNIV OF VIRGINIA on July 16, 2012


1866 Y. HUANG ET AL.

2
sqr
1.5 hex
random

Mode stress invariants (Pa) 1

0.5 I1 due to
transverse
compression
0
σVM due to I1 due to σVM due to σVM due to
transverse transverse transverse longitudinal
−0.5 tension tension compression shear

−1

−1.5

Figure 14. Comparison between mode stress invariants of matrix in regular and random arrays due to various
loadings (vf ¼ 0.6).

and random arrays due to different loadings are obtained from the histogram. All results
agree well for each loading case. In each array, the mode values of  VM in matrix due to
transverse tension and compression must be equal because the tensile and compressive
stress have the same magnitudes; similarly, in each array, mode values of I1 due to
transverse tension and compression must have the same magnitude but opposite signs,
because tensile and compressive stress also have the same magnitude but opposite signs,
and by definition, the calculation of I1 does not eliminate the information of signs. Note
that the mode value of  VM due to transverse loading in the hexagonal array is very close
to that in the random array, whereas the mode value of I1 in the square array is closer to
that in the random array. Pay special attention to the case of longitudinal shear loading:
both arrays show higher mode stress invariants of  VM than the random array. Therefore,
as far as mode stresses are concerned, the regular arrays tend to overestimate the micro-
stresses more than the random array.

Interfacial Traction Distribution

Through a comparison of interfacial traction distribution between the random fiber


array and two regular fiber arrays under various loading conditions with all possible
loading angles, the validity of regular fiber arrays is shown. In Figure 15, cumulative
frequency distributions of tractions in square, hexagonal, and random arrays (vf ¼ 0.6)
due to transverse tension, longitudinal shear and temperature increment are plotted and
compared. Cumulative frequency distributions of normal traction tn due to transverse
tensile stress in the three arrays show good agreement, except that regular arrays show
slightly lower values than those of the random array. Differences between the
distributions of regular arrays and the random array become more obvious in
Figures 15(b) and (c), which correspond to tangential traction tt due to transverse
tension and longitudinal shear traction tx due to longitudinal shear, respectively.

Downloaded from jcm.sagepub.com at UNIV OF VIRGINIA on July 16, 2012


Effects of Fiber Arrangement on Mechanical Behavior of Unidirectional Composites 1867
(a) 1.2 (b)
Transverse tensile loading (1Pa) Transverse tensile loading (1Pa)
1
Cumulative frequency

0.8

0.6 Sqr.
Hex. Sqr.
0.4 Random Hex.
Random
0.2

0
−0.6 −0.3 0 0.3 0.6 0.9 1.2 1.5 1.8 2.1 −1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 1
Normal traction, tn (Pa) Tangential traction, tt (Pa)

(c) 1.2 (d)


Longitudinal shear loading (1Pa)
1
Cumulative frequency

0.8 Sqr.
In the longitudinal shear
Hex.
loading, the normal and
0.6 Random tangential tractions are zero
over the entire fiber-matrix
interface.
0.4

0.2

0
−3 −2.5 −2 −1.5 −1 −0.5 0 0.5 1 1.5 2
Longitudinal traction, tx (Pa)

(e) 1.2 (f)


Thermal loading (∆T=1°C) Thermal loading (∆T=1°C)
1
Cumulative frequency

0.8

0.6 Sqr. Sqr.


Hex. Hex.
0.4 Random Random

0.2

0
−0.3 −0.2 −0.1 0 0.1 0.2 0.3 0.4 0.5 −0.4 −0.3 −0.2 −0.1 0 0.1 0.2 0.3
Normal traction, tn (MPa) Tangential traction, tt (MPa)

Figure 15. Cumulative frequency distribution of tractions at the fiber–matrix interface in regular and random
arrays due to various loadings (vf ¼ 0.6): (a) normal traction distribution; (b) tangential traction distribution;
(c) longitudinal traction distribution; (d) normal and tangential traction distribution; (e) normal traction
distribution and (f) tangential traction distribution.

The common feature shared by these two figures is that there are more tractions of
smaller values in regular arrays, whereas more tractions of greater values exist in the
random array. Since in the random fiber array the distance between two adjacent fibers
could be smaller than that in regular arrays, a higher local stress concentration can be

Downloaded from jcm.sagepub.com at UNIV OF VIRGINIA on July 16, 2012


1868 Y. HUANG ET AL.

attained, and consequently larger interfacial traction might appear in the random array.
That is why curves of random array and regular arrays entangle. Under thermal loading,
the features of regular array curves of appearing in Figures 10(e) and (f), such as steep
rising and sharp turning, are also present in the cumulative frequency curves of tn and tt,
because in regular arrays the distance between two adjacent fibers is constant. Despite
this difference, the agreement between the results obtained from random array and
regular arrays is fairly good, at least under transverse and longitudinal shear loading.

Comparison between Transverse Strength and Interfacial Traction Distribution

Sun et al. [26] provided valuable data regarding the cumulative frequency distribution of
normalized transverse tensile strength at various temperatures in the composites. It is
assumed that normal tractions at the fiber matrix interface contribute to failure under
transverse tension. Since it is possible only for relatively large interfacial tractions to cause
failure, the cumulative frequency distribution of normal interfacial tractions above a
certain threshold value in the random array due to transverse tension is supposed to follow
the same distributions as those experimental data. In this case, all parameters in the
Weibull distribution can be determined as follows: maximum normal traction threshold
tn, max ¼ 0:9, shape parameter m ¼ 15, and scale parameter  ¼ 1.0, i.e.:
   
 tn, max  tn, max m
F tn, max ¼ 1  exp  : ð2Þ


The cumulative frequency curve based on these parameters is plotted in Figure 16,
together with the cumulative frequency distribution of experimental transverse

weibull(tn,max)
Experiment
0.8
Probability of failure

0.6

0.4

0.2

0
0.8 0.85 0.9 0.95 1 1.05 1.1 1.15 1.2
Normalized maximum normal traction, tn,max (Pa)

Figure 16. Comparison between the Weibull distribution of maximum normal interfacial traction in the random
array (vf ¼ 0.6) due to transverse tension and the cumulative probability distribution of transverse strength data
from the experiment.

Downloaded from jcm.sagepub.com at UNIV OF VIRGINIA on July 16, 2012


Effects of Fiber Arrangement on Mechanical Behavior of Unidirectional Composites 1869

strength data. Obviously, good agreement is achieved, which verifies the algorithm used
for generating random fiber array again.

DISCUSSION AND CONCLUSION

Since in reality, material properties of constituents (fiber and matrix) vary in a certain
range and the fiber arrangement in real composite is irregular, micro stresses and inter-
facial tractions within composite due to external loads follow the Weibull distribution. If
micromechanics of failure (MMF) [27] is applied to predict the strength of the lamina and
laminate, the results must follow the Weibull distribution. This relationship is clearly
shown in Figure 17. The good accordance between normal interfacial traction distribution
and transverse strength distribution verifies the predictive capability of the random array.
Because it has been shown in previous text that the response of square and hexagonal
arrays to various loadings is close to that of the random array, we can conclude that the
strength predictions based on the regular arrays are similar to strength predictions based
on the random array, and therefore also similar to real values.
The following conclusions are drawn in this article:
1. The algorithm for generating a random fiber array is effective, and the mechanical
behavior of the generated random fiber array is verified to be similar to real fiber array.

Matrix Structural Failure Loads, Pf


Pf
Material
Properties

Fiber

Lamina strength

1
weibull(tn,max)
Random Fiber Array
Probability of failure

Experiment
0.8

Micromechanics 0.6
of Failure (MMF)
0.4

0.2

0
0.8 0.85 0.9 0.95 1 1.05 1.1 1.15 1.2
Micro-stresses
Laminar strengths

Figure 17. Relationship between distribution of macroscopic strength and distribution of microstructure
properties for composites.

Downloaded from jcm.sagepub.com at UNIV OF VIRGINIA on July 16, 2012


1870 Y. HUANG ET AL.

2. The predictions of effective material properties from the random array and two regular
arrays (square and hexagonal) are in good agreement.
3. Stress invariant distributions in random array, square array and hexagonal array due to
mechanical and thermal loading are compared. The range of stress invariant values in
random fiber array is wider than in both the square and hexagonal arrays due to
irregularity in distance between adjacent fibers, which results in lower predicted
strength because of a higher probability for local failure.
4. The cumulative frequency distributions of stress invariants in random and regular
arrays due to various loading types and various loading angles show good agreement,
which validates the effectiveness of regular arrays.
5. In terms of the cumulative frequency distribution of interfacial traction, the
performance of regular arrays is acceptable at least when subject to transverse tension
and longitudinal shear with all possible loading angles.

ACKNOWLEDGMENT

The authors would like to thank Prof. Stephen W. Tsai for his valuable guidance.

REFERENCES

1. Bednarcyk, B.A. and Arnold, S.M. (2001). Micromechanics-based Deformation and Failure
Prediction for Longitudinally Reinforced Titanium Composites, Composites Science and
Technology, 61: 705–729.
2. Kok, de J.M.M. and Meijer, H.E.H. (1999). Deformation, Yield and Fracture of
Unidirectional Composites in Transverse Loading – I: Influence of Fiber Volume
Fraction and Test-temperature, Composites Part A: Applied Science and Manufacturing, 30:
905–916.
3. Laird II, G. and Kennedy, T.C. (1995). Micromechanics of Composite Materials under
Compressive Loading, Engineering Fracture Mechanics, 51: 417–430.
4. Berryman, J.G. and Berge, P.A. (1996). Critique of Two Explicit Schemes for Estimating Elastic
Properties of Multiphase Composites, Mechanics of Materials, 22: 149–164.
5. Lamon, J. (2001). A Micromechanics-Based Approach to the Mechanical Behavior of
Brittle-Matrix Composites, Composites Science and Technology, 61: 2259–2272.
6. Kwon, Y.W. and Berner, J.M. (1995). Micromechanics Model for Damage and
Failure Analyses of Laminated Fibrous Composites, Engineering Fracture Mechanics, 52:
231–242.
7. Brinson, L.C. and Lin, W.S. (1998). Comparison of Micromechanics Methods for Effective
Properties of Multiphase Viscoelastic Composites, Composite Structures, 41: 353–367.
8. Sun, C.T. and Vaidya, R.S. (1996). Prediction of Composite Properties from a Representative
Volume Element, Composites Science and Technology, 56: 171–179.
9. Caiazzo, A.A. and Costanzo, F. (2000). On the Effective Elastic Properties of Composites
with Evolving Microcracking, Journal of Reinforced Plastics and Composites, 19: 152–163.
10. Dong, M. and Schmauder, S. (1996). Modeling of Metal Matrix Composites by a Self-
Consistent Embedded Cell Model, Acta Materialia, 44: 2465–2478.
11. Sun, H., Di, S., Zhang, N. and Wu, C. (2001). Micromechanics of Composite Materials using
Multivariable Finite Element Method and Homogenization Theory, International Journal of
Solids and Structures, 38: 3007–3020.
12. Li, S. (2001). General Unit Cells for Micromechanical Analyses of Unidirectional Composites,
Composites Part A: Applied Science and Manufacturing, 32: 815–826.

Downloaded from jcm.sagepub.com at UNIV OF VIRGINIA on July 16, 2012


Effects of Fiber Arrangement on Mechanical Behavior of Unidirectional Composites 1871

13. Whitcomb, J. and Tang, X. (2002). Micromechanics of Moisture Diffusion in Composites with
Impermeable Fibers, Journal of Composite Materials, 36: 1093–1101.
14. Zhang, Y., Xia, Z. and Ellyin, F. (2004). Evolution and Influence of Residual Stresses/Strains of
Fiber Reinforced Laminates, Composites Science and Technology, 64: 1613–1621.
15. Bonora, N. and Ruggiero, A. (2006). Micromechanical Modeling of Composites with
Mechanical Interface – Part I: Unit Cell Model Development and Manufacturing Process
Effects, Composites Science and Technology, 66: 314–322.
16. Bonora, N. and Ruggiero, A. (2006). Micromechanical Modeling of Composites with
Mechanical Interface – Part II: Damage Mechanics Assessment, Composites Science and
Technology, 66: 323–332.
17. Drago, A. and Pindera, M.-J. (2007). Micro-Macromechanical Analysis of Heterogeneous
Materials: Macroscopically Homogeneous vs Periodic Microstructures, Composites Science and
Technology, 67: 1243–1263.
18. Cai, H., Miyano, Y., Nakada, M. and Ha, S.K. (2008). Long-Term Fatigue Strength Prediction
of CFRP Structure Based on Micromechanics of Failure, Journal of Composite Materials,
42: 825–844.
19. Everett, R.K. and Chu, J.H. (1993). Modeling of Non-Uniform Composite Microstructures,
Journal of Composite Materials, 27: 1128–1144.
20. Chen, X. and Papathanasiou, T.D. (2004). Interface Stress Distributions in Transversely Loaded
Continuous Fiber Composites: Parallel Computation in Multi-Fiber RVEs Using the Boundary
Element Method, Composites Science and Technology, 64: 1101–1114.
21. Wongsto, A. and Li, S. (2005). Micromechanical FE Analysis of UD Fiber-Reinforced
Composites with Fibers Distributed at Random over the Transverse Cross-Section, Composites
Part A: Applied Science and Manufacturing, 36: 1246–1266.
22. Oh, J.H., Jin, K.K. and Ha, S.K. (2006). Interfacial Strain Distribution of a Unidirectional
Composite with Randomly Distributed Fibers under Transverse Loading, Journal of Composite
Materials, 40: 759–778.
23. Jin, K.K., Oh, J.H. and Ha, S.K. (2007). Effect of Fiber Arrangement on Residual
Thermal Stress Distributions in a Unidirectional Composite, Journal of Composite Materials,
41: 591–611.
24. Ha, S.K. (2002). Micromechanics in the Analysis of Composite Structures, In: 6th Composites
Durability Workshop (CDW-6), Tokyo, Japan, November 14–15.
25. Jin, K.K., Huang, Y., Lee, Y.H. and Ha, S.K. (2008). Distribution of Micro Stresses
and Interfacial Tractions in Unidirectional Composites, Journal of Composite Materials
(In print).
26. Sun, Z., Daniel, I.M. and Luo J.J. (2003). Statistical Damage Analysis of Transverse Cracking
in High Temperature Composite Laminates, Materials Science and Engineering A, 341: 49–56.
27. Ha, S.K., Jin, K.K. and Huang, Y. (2008). Micro-Mechanics of Failure (MMF) for Continuous
Fiber Reinforced Composites, Journal of Composite Materials (In print).

Downloaded from jcm.sagepub.com at UNIV OF VIRGINIA on July 16, 2012

You might also like