Sei sulla pagina 1di 14

JOURNAL OF

COMPOSITE
Article M AT E R I A L S
Journal of Composite Materials
0(0) 1–14
! The Author(s) 2015
Micromechanical damage analysis in Reprints and permissions:
sagepub.co.uk/journalsPermissions.nav
laminated composites with randomly DOI: 10.1177/0021998315614992
jcm.sagepub.com
distributed fibers

Nithin K Parambil and Suhasini Gururaja

Abstract
Damage mechanisms in uni-directional fiber-reinforced plastics have been studied at the micro-scale by developing three-
dimensional-repeating unit cells with randomly distributed fibers. Three damage mechanisms have been considered, viz.,
matrix damage, fiber failure and fiber-matrix interface debonding. The development of these damage modes and their
effect on the overall stress–strain response of the micro-structure due to varying fiber volume fractions, loading con-
ditions (longitudinal, transverse and combined transverse tension and shear), spatial distribution of fibers and fiber
strength distributions (deterministic and Weibull) are of interest. A numerical framework has been developed that
allows for conducting such studies for the chosen parameters. Microscopic images of real micro-structure of uni-
directional fiber-reinforced plastics have been used to develop the random micro-structure for the three-dimensional-
repeating unit cell. In the longitudinal loading scenario, fibers with Weibull strength distribution are closer to reality.
While the three-dimensional-repeating unit cell under longitudinal loading fails in predominantly fiber failure mode,
debonding and matrix damage play a major role in determining the average response of the micro-structure under
transverse and combined transverse and shear loading.

Keywords
Damage analysis, uni-directional fiber-reinforced plastics, Weibull strength, random fiber distribution

property of FRPs.3,8–11 However, the assumptions


Introduction
used for analytical model development prevent them
In recent times, conventional structural materials such from predicting material behavior by considering
as steel, aluminum, etc. have been replaced by fiber- actual micro-structure of FRPs in addition to account-
reinforced plastics (FRPs) in aerospace, marine engin- ing for different damage mechanisms.12 On the other
eering, defense and other sectors due to their better hand, empirical models are inadequate in terms of cap-
performance, viz., higher specific strength, stiffness, turing the physics of the damage mechanisms and often
fatigue, corrosion, tailorability, etc.1 Anisotropy and times, fail to yield accurate predictions.13 Additionally,
heterogeneity in FRPs result in a multitude of unlike macro-scale experiments,14,15 conducting micro-
damage mechanisms such as matrix cracking, fiber- scale experiments to capture micro-scale damage mech-
matrix debonding, fiber breakage, fiber pull-outs, anisms in FRPs is quite complex and difficult to carry
delaminations, etc. that occur at multiple length out.16 Among the numerical approaches, generalized
scales. Therefore, design and failure analysis of struc- method of cells,17–20 variational asymptotic method
tures made from FRPs is more complex than trad- for unit cell homogenization21 and finite element
itional metals and necessitates advanced
computational techniques for accurate predictions of Aerospace Engineering, Indian Institute of Science (IISc), Bangalore,
material behavior across multiple length scales.2 Karnataka, India
In the past, experimental, analytical and numerical
Corresponding author:
methods have often been employed to characterize the Suhasini Gururaja, Aerospace Engineering, Indian Institute of Science
material behavior of FRPs.3–7 Well-established analyt- (IISc), C V Raman Avenue, Bangalore 560012, Karnataka, India.
ical models are available to predict the effective Email: suhasini@aero.iisc.ernet.in

Downloaded from jcm.sagepub.com at University of Otago Library on December 11, 2015


2 Journal of Composite Materials 0(0)

analysis (FEA) with continuum damage modeling22,23 Several algorithms are currently available for produ-
are a few that have received considerable attention in cing random fiber distributions, the most common one
recent past. Thus, in the last decade, numerical multi- being the hard-core model. In the hard-core model, the
scale damage accumulation models have been devel- fibers were modeled as non-overlapping circular disks.
oped with the aim of capturing damage initiation and The maximum Vf produced using this algorithm was
progression at multiple length scales replicating the 50%; above this, Vf overlapping takes place (this is
actual physics of damage in FRPs.16 called jamming limit).38,39 A statistically equivalent
Traditionally, micromechanics principles have been RVE developed by Swaminathan et al.40 using a two-
used for effective property estimations of compos- point correlation function based on the parent hard-
ites.12,22,24,25 Correlating the micro-structure to the core model was capable of generating Vf up to 60%.
effective property is done in two ways, namely, repre- A new algorithm named random sequential expansion
sentative volume element (RVE) approach and repeat- overcame the jamming limit of hard-core model and
ing unit cell (RUC) approach. In RVE method, the efficiently produced Vf greater than 60%.41 Other
smallest volume element that replicates the composite approaches using the principle of nearest neighbor dis-
macro-behavior is first chosen and analyzed using tances have also been developed.42,43 Besides these
‘‘homogenization’’ theory (based on a Micro–Meso– methods, digital imaging techniques have also been
Macro principle) to yield the effective property of the employed wherein the micro-structure of the UD-
composite.9,22 On the other hand, RUC assumes that FRP laminate is directly modeled by measuring the
the constituents of the composite (matrix, fiber, voids, centroid coordinates of the fibers from the image and
interface, etc.) follow a periodic distribution known a reproducing it.44
priori and the domain of analysis is chosen to be this In addition to spatial randomness of fibers, tensile
smallest repeating periodic volume element.26–29 Drago test results of single fiber and fiber bundles indicate
and Pindera30 have compared and distinguished these that fiber strength is not a single deterministic
two approaches. Uni-directional FRPs (UD-FRPs) are strength value but follows a Weibull distribution.45–48
often modeled as individual fibers embedded in matrix Incorporating such a probabilistic strength for fibers
with a uniform square or hexagonal distribution. It is within a micro-scale progressive damage framework,
recognized that the micro-structure for a UD-FRP is, in considering random distribution of fibers in matrix
fact, characterized by random distribution of fibers. has received little attention. In this work, effective
For a micro-structure with an evolving damage state, properties of FRPs have been estimated using a
it becomes necessary to model a realistic micro- three-dimensional (3D)-RUC of randomly distributed
structure and explicitly account for the spatial random- fibers under evolving damage using non-deterministic
ness in fiber distributions.31,32 It has been observed that fiber strength. A framework for damage initiation
although longitudinal properties (without considering and progression has been adopted considering three
any damage and failure) do not vary across the two major damage mechanisms, viz., matrix damage,
RUCs (uniform and random arrangement of fibers), fiber-matrix debonding and fiber failure. The effective
transverse properties do depend on spatial distribution constitutive response of the composite has been
of fibers.33 Trias et al.34 carried out a comparative study obtained through non-linear homogenization proced-
on RUCs with uniformly distributed fibers and com- ure in the presence of the aforementioned damage
pared with an RUC with randomly distributed fibers. It processes. The present work aims at developing a
was observed that matrix cracking and damage initi- framework for damage development and progression
ation were also influenced by the spatial arrangement in FRPs at the micro-scale. It is envisaged that the
of fibers. In the longitudinal tensile loading scenario, current framework can be extended to higher length
the failure of fibers was not found to be unique.35 In scales paving way toward multi-scale progressive
recent times, attempts have been made to model the damage development and virtual mechanical testing
micro-structure as an RUC with randomly distributed of FRPs.
fibers with an evolving damage state wherein the
response of the micro-structure is evaluated using
standard homogenization techniques.36,37 The micro-
Approach
mechanical model, such as RVE or RUC model, 3D-RUC (with randomly distributed fibers) has been
which is coupled with a macro-model used to evaluate modeled with three different phases, namely: matrix,
overall behavior of a composite laminate is often fiber and fiber-matrix interface as shown in Figure 1.
known as multi-scale modeling.16 Thus, in order to To understand the damage mechanism of 3D-RUC,
achieve a more efficient multi-scale modeling paradigm, three major damage mechanisms, viz., matrix damage,
it becomes important to capture damage evolution real- fiber-matrix debonding and fiber failure have been
istically at the micro level. adopted.

Downloaded from jcm.sagepub.com at University of Otago Library on December 11, 2015


Parambil and Gururaja 3

3D-RUC with random fibers MATLAB code. From the distribution fittings, it is
A polished sample of UD-FRP laminate cross-section concluded that the distance from the center of the
has been used for microscopic study. To capture the fibers to origin follows a normal distribution. Thus, in
position of fibers in the matrix, different images were this work, the fibers are distributed using normal dis-
obtained using an optical microscope (Zeiss Axiovert) tribution within the 3D-RUC.
as shown in Figure 2. Central coordinates of fibers have
been obtained by using inbuilt option in Mathematica.
By fixing the origin, distance between the center of
Constitutive modeling
fibers and origin has been calculated. Probability dens- Fiber damage. Fiber has been modeled as an orthotropic
ity function of these data was calculated using elastic material, and its material constants are shown in
Table 1. Fiber breakage is modeled using the maximum
stress criterion. Two cases of fiber damage have been
studied in this work, viz., deterministic fiber strength
and probabilistic fiber strength. In the deterministic
fiber strength case, 2500 MPa has been taken as fiber
strength.49 As mentioned earlier, tensile testing of fibers
and fiber bundles reveals a Weibull strength distribu-
tion.47 Historically, there are two types of Weibull dis-
tributions, namely, two-parameter and three-parameter
Weibull distribution.45 For the fixed fiber diameter case
as in the present analysis, two-parameter Weibull dis-
tribution fits the failure stress curves. For studying the
variation in strength of fibers, two-parameter Weibull
distribution as shown in equation (1) has been imple-
mented using scale ( o) and shape (m) factors as
2500 MPa and 6, respectively.50
  m 
x
PðxÞ ¼ 1  exp  ð1Þ
Figure 1. 3D-RUC with randomly distributed fibers. o
3D-RUC: three-dimensional-repeating unit cell.

Figure 2. Microscopic images of UD-FRPs.


UD-FRPs: Uni-directional fiber-reinforced plastics.

Table 1. Material properties for C-fiber and epoxy matrix.49

Elastic constants E11 (GPa) E22 (GPa) 12 23 G12 (GPa) G23 (GPa) T (MPa) Y (MPa) C (MPa)

Carbon fiber 235 14 0.2 0.25 28 5.6 2500


Matrix 4.8 – 0.34 – – – – 60 120

Downloaded from jcm.sagepub.com at University of Otago Library on December 11, 2015


4 Journal of Composite Materials 0(0)

where x represents the stress in the fiber direction. traction-separation curve before damage evolution, and
damage (D) is calculated using equation (5). Prior to
Matrix damage. Matrix is modeled as an elastic isotropic damage onset (i.e.,   0 ), n  tn and D ¼ 0. As
material that has different strengths in tension and n 4 tn , damage evolution begins according to equation
compression with the properties as shown in Table 1. (5). D ¼ 1 represents complete debonding of the inter-
Raghava et al.51 suggested a modified von Mises failure face. A mixed-mode damage propagation criterion
criterion for the matrix wherein yielding depended on based on equation (6) is adopted.53 Table 2 shows the
both deviatoric stress invariant (von Mises equivalent properties of fiber-matrix interface used in the present
stress (vm Þ) and volumetric stress invariant (I1) as analysis.
shown in equation (2).
n ¼ ð1  DÞKn  ð4Þ
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ð  1ÞI1 þ ð  1Þ2 I21 þ 4vm 2 8
> 0   0
eq ¼ ð2Þ >
< ð   Þ
2 0 f
D ¼ max 0 5  5 f ð5Þ
>
> ðf  0 Þ
:
where  is the ratio of tensile strength and compressive 1  4 f
strength of matrix. The  eq is compared with the tensile
     
strength of the matrix  Y. When eq  Y , matrix GI GII  GIII 
damage starts to develop resulting in stiffness degrad- þ þ ¼1 ð6Þ
GIc GIIc GIIIc
ation of the matrix.52 After damage, the constitutive
equation of the matrix and damage factor (D) is where  ¼ 1
shown in equation (3).
Numerical implementation
ijm ¼ ð1  DÞCm m
ijkl kl ð3Þ
3D-RUC with randomly distributed fibers has been
where superscript ‘m’ indicates the matrix properties modeled with periodic boundary conditions (PBCs)
and D ranges from 0 to 1. using python scripting interface in the commercially
available FE software Abaqus. An algorithm has
Fiber-matrix interface. Cohesive zone modeling is used to been developed to generate randomly distributed
simulate the fiber-matrix interface damage. A bi-linear fibers (non-overlapping) in a matrix to study the
traction-separation law is adopted wherein the cohesive damage mechanisms in the 3D-RUC. Center of fibers
surface exhibits linear elastic behavior till damage onset has been generated using the inbuilt random number
beyond which linear softening law delineates damage generator function (using normal distribution) in
evolution (cf. Figure 3). Equation (4) describes this con- Python. It should be noted that the fibers do not inter-
stitutive interface behavior where Kn is the slope of the sect with the boundaries of the 3D-RUC and are always
located in the central region of the 3D-RUC for easy
implementation of PBCs (cf. Appendix 2). Such a
restriction has led to the presence of a matrix rich
band around the RUC that might result in strain local-
ization or shear band formation. However, from the
micro-structural images (cf. Figure 2), it can be inferred
that there are matrix-rich regions present abundantly.
Thus, the restriction of having non-intersecting fibers
should still be a valid representation of the micro-
structure. Required volume fraction of fibers has been
generated with the help of input parameters (radius of
the fiber, fiber volume fraction and total number of

Table 2. Properties for the C-fiber/epoxy interface.53

Kn (MPa/m) tn (MPa) ts (MPa) GIc (J/m2) GIIc ¼ GIIIc (J/m2)


Figure 3. Bi-linear traction-separation law for interface 1012 60 60 2 6
debonding.

Downloaded from jcm.sagepub.com at University of Otago Library on December 11, 2015


Parambil and Gururaja 5

fibers). Meshing has been done for both matrix and Homogenized stress–strain curves are obtained using
fiber using 3D eight-noded linear isoparametric elem- equation (7) and implemented via python script in
ents (C3D8). PBCs have been applied as multi-point post-processing mode in Abaqus.
constraint equations for each nodal pair. User material
subroutine (UMAT) has been implemented in Z I
FORTRAN within Abaqus framework to implement 1 1X p

 ij ¼ ij dV ¼  k Vk ð7Þ
the yield criteria and damage development in the V V k¼1 ij
matrix. Fiber failure has been modeled as a UMAT
based on maximum stress criterion. Fibers are assumed
to ‘‘fail’’ when the longitudinal stress in the fiber where  ¼ homogenized stress, V ¼ total volume of
exceeds the tensile strength of the fiber. Fiber ‘‘failure’’ 3D-RUC, ijk ¼ stress component in the kth integration
has been simulated by assigning a degradation factor point, Vk ¼ volume of the kth element and Ip ¼ total
(1%) to the appropriate stiffness coefficient of the number of integration points in the 3D-RUC FE
fiber.54 Flow chart for the implementation of damage model.
mechanism in UD-FRPs is shown in Figure 4. Three
different loading conditions are considered in the pre-
sent work, longitudinal (loading in the fiber direction),
Results and discussion
transverse (loading perpendicular to the fiber direction) FE models have been developed to study the differ-
and transverse shear (combination of transverse and ent loading condition (longitudinal, transverse
shear). As in traditional micromechanics analysis, load- and transverse shear) with Vf ¼ 20%, 30%, 40%
ing is applied in the form of far-field displacement and 50%. Far-field strains have been applied to
fields.29 For all cases, homogenized stress–strain the model for studying damage mechanisms for all
curves are plotted to understand the damage behavior. cases.

Figure 4. Implementation of damage for the 3D-RUC with random fiber distribution (cf. Appendix 3).
PBC: periodic boundary condition; 3D-RUC: three-dimensional-repeating unit cell.

Downloaded from jcm.sagepub.com at University of Otago Library on December 11, 2015


6 Journal of Composite Materials 0(0)

Convergence studies studied by considering several 3D-RUC models for


To establish a rigorous model for FE analysis, conver- Vf ¼ 30% and N ¼ 30. The stress–strain response in all
gence studies were performed. There were two main cases was found converge within 1%.
aspects that were investigated, viz., the number of
fibers in the 3D-RUC model and the mesh size. To
Longitudinal tensile case
begin with, FE analysis was carried out by varying
the number of fibers (N) from 1 to 60 within the In addition to incorporating random distribution of
matrix for Vf ¼ 30% with far-field strain applied in fibers, variation in strength of fibers has also been
the longitudinal (along the fiber) direction. Analysis accounted for. Thus, two separate cases have been stu-
has been carried out considering all three damage died, viz., (i) strength of the fiber is taken to be a con-
mechanisms (matrix damage, fiber-matrix debonding stant and (ii) strength of the fiber has been taken to
and fiber failure). Beyond 30 fibers, the stress–strain follow Weibull distribution.
response was found to converge (within 1% variation).
It should be noted that the dimensions of the 3D-RUC 3D-RUC with deterministic fiber strength. The homogenized
model were varied as N varied from 1 to 60 such that stress–strain curves for Vf ¼ 20%, 30%, 40% and 50%
Vf ¼ 30% for all cases. Following this analysis, mesh for far-field strain applied in the longitudinal (along the
sensitivity was conducted by decreasing the element fiber) direction are shown in Figure 5(a). As expected,
length from 2.66 to 1.18 microns for Vf ¼ 30% and the linear part showed an increase in stiffness with
N ¼ 30. Solution was found to converge (within 1% increase in the Vf. An abrupt drop in the homogenized
variation) when the element length was lesser than stress has been observed as the applied strain reaches a
1.78 microns. Thus, for further analysis, 30 fibers with value around 1.2%. This is attributed to all fibers fail-
mesh size of 1.78 microns have been used for all cases. ing. As the far-field strain increases, debond softening
Additionally, effect of spatial randomness of fibers was initiation of fiber-matrix interface occurs prior to fiber

(a)

(b) (c) (d)

Figure 5. (a) Homogenized stress–strain curves for Vf ¼ 20%, 30%, 40% and 50% with deterministic fiber strength, (b) and (c) von
Mises stress distribution of RUC for Vf ¼ 30% with deterministic fiber strength and (d) matrix damage (not applicable in the fiber
region).

Downloaded from jcm.sagepub.com at University of Otago Library on December 11, 2015


Parambil and Gururaja 7

failure. Once all fibers fail, the load-carrying capability Since the longitudinal strength of the fiber is not
of the 3D-RUC drastically reduces due to reduction in constant, all fibers do not fail at the same applied load-
fiber stiffness to 1% of its original stiffness. Thus, fur- ing. Thus, a sudden load drop is not observed as in
ther displacement loading results in matrix yielding and Figure 5(a).55,56 Instead, a series of smaller load drops
eventually matrix damage. von Mises stress before fiber are observed, each load drop being indicative of either
breakage and after fiber breakage are shown in Figure an individual fiber failure or failure of a collection of
5(b) and (c), respectively. Matrix damage (SDV15 pre- fibers as shown in Figure 6. These fiber failures are
assigned solution-dependent state variables for matrix found to occur in a progressive manner with load redis-
damage) has taken place at B in Figure 5(a) and is tribution occurring after each fiber(s) failure (or
shown in Figure 5(d). load drop) event. An additional observation from
Figure 6(a) is that the RUC continues to carry load
3D-RUC with Weibull distribution for fiber strength. Figure 6(a) even after the first few fibers have failed. Once a signifi-
shows the homogenized stress–strain curves for cant number of fibers have failed, the homogenized
Vf ¼ 20%, 30%, 40% and 50% with far-field stress–strain curve gradually slopes down indicating
strain applied in the longitudinal (along the fiber) dir- reduction in load-carrying capability. Eventually, a
ection with Weibull distribution for fiber strength. final load drop occurs as all the fibers fail.

(a) 800

700 V = 20%
f
A
Vf = 30%
600
V = 40% C
f
Stress, σ (MPa)

B
500 Vf = 50%
11

400
D
300

200

100

0
0 0.2 0.4 0.6 0.8 1 1.2 1.4
Strain, ε (%)
11
(b) (c) (d) (e)

(f) (g) (h) (i)

Figure 6. (a) Homogenized stress–strain curves for Vf ¼ 20%, 30%, 40% and 50% with Weibull distribution for fiber strength, (b) to
(e) von Mises stress distribution of RUC for Vf ¼ 30% with Weibull distribution for fiber strength and (f) to (i) matrix damage of RUC
for Vf ¼ 30% with Weibull distribution for fiber strength.
RUC: repeating unit cell.

Downloaded from jcm.sagepub.com at University of Otago Library on December 11, 2015


8 Journal of Composite Materials 0(0)

The debond softening initiation of fiber-matrix inter- transverse load case are fiber-matrix debonding and
face occurs prior to fiber failure. Matrix damage was matrix damage. Fiber failure does not occur under
found to initiate just after the first load drop event due this loading case. Due to the random spatial distribu-
to stress redistribution that occurs in the vicinity of tion of fibers in the 3D-RUC, the stress state is not
failed fibers (cf. Figure 6). uniform in the matrix. Fiber clustering typically leads
to increased stress concentration in the surrounding
matrix. Debonding—one of the first damage modes—is
Transverse tensile case found to occur either in a single fiber-matrix interface
Figure 7(a) shows the homogenized stress–strain curves or in a collection of fibers. Debonding in some fibers
for Vf ¼ 30% with far-field strain applied in the trans- starts from one side of fiber, whereas in others, it starts
verse (perpendicular to the fiber) direction. Damage from both sides depending on the stress state in the
propagation is shown in Figure 7(b) to (g). Unlike lon- vicinity of an interface (cf. Figure 8). Upon further
gitudinal case, the significant damage modes in loading, debonding is found to trigger damage

(a)

(b) (c) (d)

(e) (f) (g)

Figure 7. (a) Homogenized stress–strain curves for Vf ¼ 30% with far-field strain applied in the transverse direction and (b) to (g)
matrix damage propagation with different far-field strains for Vf ¼ 30%.

Downloaded from jcm.sagepub.com at University of Otago Library on December 11, 2015


Parambil and Gururaja 9

Figure 8. Fiber-matrix interface debonding, one-sided and two-sided cases.

initiation in matrix accompanied by further propaga- propagation purely depends on either shear or trans-
tion of interfacial debonding. Matrix damage also verse strain, whichever is predominant as is clear from
propagates triggering additional fiber-matrix debonds Figure 9(b) to (e). In transverse strain predominant case
in other interfaces. The proximity of fibers or the pres- (point D in Figure 9(a)), matrix damage initiation
ence of fiber clusters plays an important role in govern- occurs prior to interface debonding. On the other
ing the damage evolution. In regions with a prevalence hand, in shear strain predominant case (point B in
of fiber clustering, debonding/matrix damage occurs at Figure 9(a)), some fiber-matrix interface debonding
a lower applied far-field strain as compared to regions occurs prior to matrix damage initiation. As the applied
that are matrix rich (cf. Figure 7) due to the increased far-field strain increases, matrix damage propagates,
stress concentration in the vicinity of fibers. The matrix and this causes more interfaces to debond. Points A
damage and debonding develop a failure-path that and E in Figure 9(a) show pure shear and transverse
ultimately causes failure of the 3D-RUC. For an cases, respectively. At point C in Figure 9(a), applied
applied 22 far-field strain, damage initiation is far-field strains are equal, and in this case, interface
observed at 22 ¼ 0.8% accompanied by a reduction in debonding occurs prior to damage initiation. The plot
the global stiffness value. As the damage propagates, shown in Figure 9(a) can be used as failure onset enve-
further reduction in stiffness is noted. This is shown in lope since further increase in the far-field strain reduces
Figure 7(a). From Figure 7, it is clear that the damage the stiffness of matrix. A similar approach can be used
propagation seems to occur along the load direction. to develop failure envelopes for other Vf’s correspond-
This could be due to the usage of a stress-based damage ing to various damage parameters. As we discussed in
onset criterion; since maximum principal stress is along the transverse tensile case, the damage propagation is
the load direction, damage onset and propagation along the maximum principal stress direction as is clear
occurs along the same direction. from Figure 9. It should be noted that similar results
have also been obtained in Melro et al.32
Transverse shear case
In this case, a combination of transverse (22) and shear
Conclusions and future work
(23) far-field strains are applied to the 3D-RUC as In the present study, a unified framework for micro-
shown in Figure 9(a). Homogenized transverse shear scale damage evolution has been developed considering
stress envelopes for Vf ¼ 30% is shown in Figure 9(a). randomly distributed fibers in the matrix. Real micro-
This envelope is obtained by changing the combination scopic structure of UD-FRP has been studied using
of transverse and shear far-field strains to get a matrix optical microscopy for fitting the fiber distribution in
damage parameter of approximately 0.5 in any single the matrix, and an algorithm has been developed for
element in the 3D-RUC. The resultant damage modeling 3D-RUC. Damage mechanisms under static

Downloaded from jcm.sagepub.com at University of Otago Library on December 11, 2015


10 Journal of Composite Materials 0(0)

(a)

(b) (c)

(d) (e)

Figure 9. (a) Homogenized transverse shear stress curves for Vf ¼ 30% and (b) to (e) matrix damage propagation under different
combination of applied transverse and shear strains for Vf ¼ 30%.

loading have been studied by considering matrix Fibers with deterministic strength and Weibull strength
damage, debonding of fiber-matrix interface and fiber distributions have been studied to understand the
failure. Non-linear homogenized stress–strain curves damage mechanisms. It is noted that fibers with
for 20%, 30%, 40% and 50% volume fractions of Weibull strength distribution describe a more realistic
fibers were obtained for the longitudinal load case. damage scenario in UD-FRPs. In addition to the

Downloaded from jcm.sagepub.com at University of Otago Library on December 11, 2015


Parambil and Gururaja 11

longitudinal loading case, 3D-RUC response under 10. Camanho PP. Failure criteria for fibre-reinforced polymer
combined transverse and shear loads has been composites. Secçăo de Mecânica Aplicada, Departamento
obtained. Under transverse and shear loads, interface de Engenharia Mecânica e Gestăo Industrial, Faculdade de
debonding and matrix damage play a vital role in deter- Engenharia da Universidade do Porto, 2002.
11. Tucker CL and Liang E. Stiffness predictions for unidir-
mining the damage state of the 3D-RUC. A progressive
ectional short-fiber composites: review and evaluation.
damage development with matrix damage triggering Compos Sci Technol 1999; 59: 655–671.
interface disbonds and vice versa has been captured 12. Sudhir A, Nagaraja AM and Gururaja S. Effective mech-
with increased far-field loading. Failure envelopes for anical properties of carbon-carbon composites. In:
transverse shear loading cases have also been obtained. ASME 2014 international mechanical engineering congress
Based on this analysis, it can be inferred that for a and exposition, 14–20 November 2014, IMECE2014-
certain predefined damage state, one can obtain the 36583. Canada: ASME.
failure envelopes that could be coupled with a multi- 13. Talreja R. Assessment of the fundamentals of failure the-
scale damage assessment strategy in order to ascertain ories for composite materials. Compos Sci Technol 2014;
the load-carrying capability of a composite structure. 105: 190–201.
Further investigations need to be carried out to extend 14. Vaughan TJ and McCarthy CT. A combined experimen-
tal–numerical approach for generating statistically
the current formulation to include cure stresses, com-
equivalent fibre distributions for high strength laminated
pressive and fatigue loads. Effect of defects such as fiber composite materials. Compos Sci Technol 2010; 70:
misalignments can also be addressed within the current 291–297.
framework. The results obtained can be integrated with 15. Fiedler B, Hojo M, Ochiai S, et al. Failure behavior of an
a meso-scale model paving way toward multi-scale epoxy matrix under different kinds of static loading.
damage modeling of FRPs. Compos Sci Technol 2001; 61: 1615–1624.
16. Llorca J, González C, Molina-Aldareguı́a JM, et al.
Declaration of Conflicting Interests Multiscale modeling of composite materials: a roadmap
towards virtual testing. Adv Mater 2011; 23: 5130–5147.
The author(s) declared no potential conflicts of interest with 17. Paley M and Aboudi J. Micromechanical analysis of
respect to the research, authorship, and/or publication of this composites by the generalized cells model. Mech Mater
article. 1992; 14: 127–139.
18. Aboudi J, Arnold SM and Bednarcyk BA.
Funding Micromechanics of composite materials: a generalized mul-
The author(s) received no financial support for the research, tiscale analysis approach. Waltham, MA, USA:
authorship, and/or publication of this article. Butterworth-Heinemann, 2012.
19. Pineda EJ and Waas AM. Modelling progressive failure
of fibre reinforced laminated composites: mesh objective
References calculations. Aeronaut J 2012; 116: 1221–1246.
1. Renton WJ, Olcott D, Roeseler W, et al. Future of flight 20. Pineda EJ, Bednarcyk BA, Waas AM, et al. Progressive
vehicle structures (2000 to 2023). J Aircraft 2004; 41: failure of a unidirectional fiber-reinforced composite
986–998. using the method of cells: discretization objective compu-
2. Kaddour AS, Hinton MJ, Smith PA, et al. The back- tational results. Int J Solids Str 2013; 50: 1203–1216.
ground to the third world-wide failure exercise. 21. Yu W and Tang T. Variational asymptotic method for
J Compos Mater 2013; 47: 2417–2426. unit cell homogenization. In: Gilat R and Banks-Sills L
3. Hashin Z. Failure criteria for unidirectional fiber compos- (Eds.) Advances in mathematical modeling and experimen-
ites. J Appl Mech 1980; 47: 329–334. tal methods for materials and structures. Netherlands:
4. Sun C and Jun AW. Compressive strength of unidirec- Springer, 2009, pp.117–130.
tional fiber composites with matrix non-linearity. 22. Sun CT and Vaidya RS. Prediction of composite proper-
Compos Sci Technol 1994; 52: 577–587. ties from a representative volume element. Compos Sci
5. Puck A and Schürmann H. Failure analysis of FRP lamin- Technol 1996; 56: 171–179.
ates by means of physically based phenomenological 23. Shojaei A, Li G, Fish J, et al. Multi-scale constitutive
models. Compos Sci Technol 1998; 58: 1045–1067. modeling of ceramic matrix composites by continuum
6. Ibnabdeljalil M and Curtin W. Strength and reliability of damage mechanics. Int J Solids Str 2014; 51: 4068–4081.
fiber-reinforced composites: localized load-sharing and 24. Li S. On the unit cell for micromechanical analysis of
associated size effects. Int J Solids Str 1997; 34: 2649–2668. fibre-reinforced composites. Proc R Soc London, Ser A:
7. Hart-Smith L. Comparison between theories and test data Math Phys Eng Sci 1999; 455: 815–838.
concerning the strength of various fibre–polymer compos- 25. Gururaja S, Taya M and Kang Y. Design of ferromag-
ites. Compos Sci Technol 2002; 62: 1591–1618. netic shape memory alloy composite made of Fe and TiNi
8. Hill R. Elastic properties of reinforced solids: some theor- particles. J Appl Phys 2005; 102: 064910–1–6.
etical principles. J Mech Phys Solids 1963; 11: 357–372. 26. Li S. General unit cells for micromechanical analyses of
9. Hashin Z. Analysis of composite materials–a survey. unidirectional composites. Compos Part A: Appl Sci
J Appl Mech 1983; 50: 481–505. Manuf 2001; 32: 815–826.

Downloaded from jcm.sagepub.com at University of Otago Library on December 11, 2015


12 Journal of Composite Materials 0(0)

27. Li S and Wongsto A. Unit cells for micromechanical ana- 44. Buryachenko VA, Pagano NJ, Kim RY, et al.
lyses of particle-reinforced composites. Mech Mater 2004; Quantitative description and numerical simulation of
36: 543–572. random microstructures of composites and their effective
28. Li S. Boundary conditions for unit cells from periodic elastic moduli. Int J Solids Str 2003; 40: 47–72.
microstructures and their implications. Compos Sci 45. Weibull W. A statistical distribution function of wide
Technol 2008; 68: 1962–1974. applicability. J Appl Mech 1951; 16: 293–307.
29. Arabatti T, Sudhir A and Gururaja S. Micromechanics of 46. Van der Zwaag S. The concept of filament strength and
damage development in polymer composites. In: ASME the Weibull modulus. J Test Eval 1989; 17: 292–298.
2014 international mechanical engineering congress and 47. Lee S, Nguyen T, Chin J, et al. Analysis of the single-fiber
exposition, 14–20 November 2014, IMECE2014-36580. fragmentation test. J Mater Sci 1998; 33: 5221–5228.
Canada: ASME. 48. Zhang Y, Wang X, Pan N, et al. Weibull analysis of the
30. Drago A and Pindera MJ. Micro-macromechanical ana- tensile behavior of fibers with geometrical irregularities.
lysis of heterogeneous materials: macroscopically homo- J Mater Sci 2002; 37: 1401–1406.
geneous vs periodic microstructures. Compos Sci Technol 49. Wang X, Zhang J, Wang Z, et al. Finite element simula-
2007; 67: 1243–1263. tion of the failure process of single fiber composites con-
31. Vaughan TJ and McCarthy CT. Micromechanical mod- sidering interface properties. Compos Part B: Eng 2013;
elling of the transverse damage behaviour in fibre rein- 45: 573–580.
forced composites. Compos Sci Technol 2011; 71: 50. Andersons J, Joffe R, Hojo M, et al. Glass fibre strength
388–396. distribution determined by common experimental meth-
32. Melro AR, Camanho PP, Pires FMA, et al. ods. Compos Sci Technol 2002; 62: 131–145.
Micromechanical analysis of polymer composites rein- 51. Raghava R, Caddell RM and Yeh GS. The macroscopic
forced by unidirectional fibres: part II–micromechanical yield behaviour of polymers. J Mater Sci 1973; 8: 225–232.
analyses. Int J Solids Str 2013; 50: 1906–1915. 52. Huang Y, Xu L and Ha SK. Prediction of three-dimen-
33. Wongsto A and Li S. Micromechanical FE analysis of sional composite laminate response using micromecha-
UD fibre-reinforced composites with fibres distributed nics of failure. J Compos Mater 2012; 46: 2431–2442.
at random over the transverse cross-section. Compos 53. Zhang H, Ericson ML, Varna J, et al. Transverse single-
Part A: Appl Sci Manuf 2005; 36: 1246–1266. fibre test for interfacial debonding in composites: 1.
34. Trias D, Costa J, Mayugo J, et al. Random models versus Experimental observations. Compos Part A: Appl Sci
periodic models for fibre reinforced composites. Comput Manuf 1997; 28: 309–315.
Mater Sci 2006; 38: 316–324. 54. Chamis CC, Abdi F, Garg M, et al. Micromechanics-
35. Beetz CP. A self-consistent Weibull analysis of carbon fibre based progressive failure analysis prediction for
strength distributions. Fibre Sci Technol 1982; 16: 81–94. WWFE-III composite coupon test cases. J Compos
36. Mishnaevsky L and Brøndsted P. Micromechanical mod- Mater 2013; 47: 2695–2712.
eling of damage and fracture of unidirectional fiber rein- 55. Parambil NK and Gururaja S. Damage development in
forced composites: a review. Comput Mater Sci 2009; 44: 3D-RUC of polymer matrix with randomly distributed
1351–1359.
fiber. In: 56th AIAA/ASCE/AHS/ASC structures, struc-
37. Mishnaevsky L and BrÃndsted P. Micromechanisms of
tural dynamics, and materials conference, 5–9 January
damage in unidirectional fiber reinforced composites: 3D
2015, AIAA-0971, Florida: AIAA.
computational analysis. Compos Sci Technol 2019; 69:
56. Wang H, Qin QH, Ji H, et al. Comparison among differ-
1036–1044.
ent modeling techniques of 3D micromechanical model-
38. Hinrichsen EL, Feder J and Jøssang T. Geometry of
ing of damage in unidirectional composites. Adv Sci Lett
random sequential adsorption. J Stat Phys 1986; 44:
2011; 4: 400–407.
793–827.
39. Pyrz R. Correlation of microstructure variability and
local stress field in two-phase materials. Mater Sci Eng:
A 1994; 177: 253–259.
40. Swaminathan S, Ghosh S and Pagano NJ. Statistically Appendix 1
equivalent representative volume elements for unidirec-
Notation
tional composite microstructures: part I – without
damage. J Compos Mater 2006; 40: 583–604. Eii Young’s modulus in ith direction [MPa]
41. Yang L, Yan Y, Ran Z, et al. A new method for generat- GI strain energy release rate (Mode I) [J/m2]
ing random fibre distributions for fibre reinforced com- GII strain energy release rate (Mode II) [J/m2]
posites. Compos Sci Technol 2013; 76: 14–20. GIII strain energy release rate (Mode III) [J/m2]
42. Wang Z, Wang X, Zhang J, et al. Automatic generation
GIc critical strain energy release rate (Mode I)
of random distribution of fibers in long-fiber-reinforced
composites and mesomechanical simulation. Mater Des [J/m2]
2011; 32: 885–891. GIIc critical strain energy release rate (Mode II)
43. Melro AR, Camanho PP and Pinho ST. Generation of [J/m2]
random distribution of fibres in long-fibre reinforced GIIIc critical strain energy release rate (Mode III)
composites. Compos Sci Technol 2008; 68: 2092–2102. [J/m2]

Downloaded from jcm.sagepub.com at University of Otago Library on December 11, 2015


Parambil and Gururaja 13

Gij shear modulus [MPa] where x, y, z are the coordinates of a point R while u, v
Kn penalty stiffness [MPa/m] and w are the displacements at this point. 011 , 022 , 033 ,
m Weibull shape parameter 023 , 013 and 012 are the macroscopic strain components.
N number of fibers Quantities with a prime are associated with R0 . The
tn normal traction strength for interface [MPa] displacement components (u, v, w) in the respective
ts first shear strength for interface [MPa] (x, y, z) directions obey the constraint equations
Vf fiber volume fraction (faces) as follows12,27
ij Poisson’s ratio
ðujx¼L  ujx¼0 Þjy,z ¼ L011
C compressive yield strength of matrix [MPa]
 eq equivalent stress [MPa] ðvjx¼L  vjx¼0 Þjy,z ¼ 0
n normal traction [MPa] ðwjx¼L  wjx¼0 Þjy,z ¼ 0
o Weibull scale parameter [MPa]
T longitudinal tensile strength of fiber [MPa] abbreviated as UA  UB ¼ FAB
 vm von Mises stress [MPa]
Y tensile yield strength of matrix [MPa]
ðujy¼B  ujy¼0 Þjx,z ¼ B012
ðvjy¼B  vjy¼0 Þjx,z ¼ B022
Appendix 2: Implementation of PBCs for ð9Þ
ðwjy¼B  wjy¼0 Þjx,z ¼ 0
the 3D-RUC
abbreviated as UC  UD ¼ FCD
Translational symmetry transformation has been used
to establish the 3D-RUC for UD-FRPs. Stresses and
strains are transformed identically from one unit cell to ðujz¼D  ujz¼0 Þjx,y ¼ D013
another under translational symmetry transforma-
tion.27 Figure 10 depicts the face, edge and vertices ðvjz¼D  vjz¼0 Þjx,y ¼ D023
identified for a 3D-RUC of L  B  D dimensions as ðwjz¼D  wjz¼0 Þjx,y ¼ D033
chosen in the current analysis. The set of all nodes on
each face are delineated into internal face nodes, edge abbreviated as UE  UF ¼ FEF
nodes and vertex nodes (cf. Figure 10(a) and (b)).
Since displacements are derived from strains, the In equation (9) (first three equations), the subscripts
relative displacements are transformed in the same ‘‘x ¼ L’’ and ‘‘x ¼ 0’’ indicate the opposite faces,
way as the strains. This leads to the following relative whereas subscripts y, z indicate the common coordi-
displacements at point R in a unit cell to those R0 nates shared by the corresponding points on the pair
(which is the image of R in another cell) of faces. To incorporate PBCs for edges and vertices,
faces have been divided into vertex nodes, edge nodes
u0  u ¼ ðx0  xÞ011 þ ð y0  yÞ012 þ ðz0  zÞ013 and internal face nodes (all the internal nodes excluding
the vertices and edges) as shown in Figure 10. Since
v0  v ¼ ð y0  yÞ022 þ ðz0  zÞ023 ð8Þ
each edge is shared by two faces and each vertex is
0
w  w ¼ ðz  0
zÞ033 shared by three faces, equation (9) is further simplified

Figure 10. Cubic unit cell.

Downloaded from jcm.sagepub.com at University of Otago Library on December 11, 2015


14 Journal of Composite Materials 0(0)

for edge node pairs (equation (10)) and vertex node


Appendix 3
pairs (equation (11)) as follows: The following tables describe the constitutive behavior
9 of matrix and fiber implemented in the current work.
UII  UI ¼ FAB >
>
>
>
UIII  UI ¼ FAB þ FCD >
>
>
>
UIV  UI ¼ FCD >
>
>
>
>
>
UVI  UV ¼ FAB >
> Table 3. Algorithm for implementation of the damage consti-
=
tutive model for the epoxy matrix.52
UVII  UV ¼ FAB þ FEF Edge Equations ð10Þ
>
>
UVIII  UV ¼ FEF >
> Step 1 Compute stress tensor at each increment
>
>
>
> rm ¼ Cm m
UX  UIX ¼ FCD >
>
>
> rm m
iþ1 ¼ ri þ ri
m
UXI  UIX ¼ FCD þ FEF >
>
> Step 2 Check for yield condition
>
;
UXII  UIX ¼ FEF eq  Y
9 if yes, go to step 3 else go to step 1
U2  U1 ¼ FAB >
> Step 3 Compute the damage variable

D
>
>
U3  U1 ¼ FAB þ FCD >
> D ¼ 1  exp 1:5 1  mY
>
>
>
> m is a scalar history variable recording
U4  U1 ¼ FCD >
= the highest matrix equivalent stress ever attained
U5  U1 ¼ FEF Vertex Equations Step 4 Update stress tensor
>
>
U6  U1 ¼ FAB þ FEF >
> rm ¼ ð1  DÞCm m
>
>
> rm m
iþ1 ¼ ri þ ri
m
U7  U1 ¼ FEF þ FAB þ FCD >
>
>
>
;
U8  U1 ¼ FCD þ FEF
ð11Þ
Table 4. Algorithm for implementation of the damage
constitutive model for the carbon fiber.49

Step 1 Compute stress tensor at each increment


 f ¼ Cf f
 fiþ1 ¼  fi þ  fi
Step 2 Check for strength criterion
11  T
if yes, go to step 3 else go to step 1
Step 3 Compute the damage variable D
D ¼ 0.99
Step 4 Update stress tensor
 f ¼ ð1  DÞCf f
 fiþ1 ¼  fi þ  fi
The tensors are denoted as bold letter.

Downloaded from jcm.sagepub.com at University of Otago Library on December 11, 2015

Potrebbero piacerti anche