See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/280086103

Fatigue Damage of Composite Structures Applying a Micromechanical Approach

Research · July 2015

DOI: 10.13140/RG.2.1.3804.5287

CITATIONS READS
0 396

3 authors, including:

Gasser F. Abdelal Antony John Robotham

Queen's University Belfast University of Plymouth
51 PUBLICATIONS   462 CITATIONS    16 PUBLICATIONS   68 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Simulating artificial lightning and composite damage View project

Inelastic behavior of composites View project

All content following this page was uploaded by Gasser F. Abdelal on 16 July 2015.

The user has requested enhancement of the downloaded file.

0 (0000) 0 – 0

ISBN 0-9780479

AES-ATEMA’ 2011International Conference on Advances and Trends in Engineering Materials

and their Applications
(Milan, Italy: July 04 - 08, 2011)
Fatigue

Fatigue Damage of Composite Structures Applying a

Micromechanical Approach
Gasser F. Abdelal 1*, E. J. Barbero 2, A. J. Robotham 3
1
The Virtual Engineering Centre , University of Liverpool, Liverpool, UK, L69 3GH
(Email: [email protected])
2
Mechanical & Aerospace Engineering, West Virginia University, Morgantown, WV 26505
(Email: [email protected])
3
The Virtual Engineering Centre , University of Liverpool, Liverpool, UK, L69 3GH
(Email: [email protected])
*Corresponding Author

Abstract damage model parameters can be determined

Fatigue damage calculations of unidirectional using available data from quasi-static loading
polymer composites is presented applying tests. Decomposition of the state variables
micromechanics theory. An orthotropic down to the constituent scale is accomplished
micromechanical damage model is integrated by micromechanics theory. Phenomenological
with an isotropic fatigue evolution model to damage evolution models are then postulated
predict the micromechanical fatigue damage of for each constituent and for interphase among
the composite structure. The orthotropic them. Comparison between model predictions
micromechanical damage model is used to and experimental data is presented.
predict the orthotropic damage evolution
within a single cycle. The isotropic fatigue Keywords
model is used to predict the magnitude of Composites, Damage, Fatigue, Finite Element,
micromechanics.
fatigue damage accumulated as a function of
the number of cycles. The advantage of using
1 Introduction
this approach is the cheap determination of Orthotropic material models are used to model
model parameters since the orthotropic the behavior of the polymer matrix composites

© AES-Advanced Engineering Solutions (Ottawa, Canada)

All rights are reserved
Gasser F. Abdelal, E. J. Barbero, A. J. Robotham
and their damage tensor [1,2,3]. Fatigue (the data required is listed in Table 1).
damage of metal matrix composites has been 2 Damage
modeled using a scalar damage measure and 2.1 Quasi-static Damage Model
simulating the material as homogeneous and The proposed approach is a damage model for
transversely isotropic [4]. Only a limited a single loading cycle. The model combines
number of tests are sufficient to determine the two continuous damage models at the
model parameters, but all tests need to be constituent level and one continuous damage
fatigue tests, which may be difficult and time model at the lamina level. Damage is
consuming. Alternatively, quasi-static-loading represented for each phase by a state variable,
damage models can be generalized for fatigue in the form of a second order damage tensor
loading by introducing cyclic hardening [5]. Dij, or by its complement the integrity tensor
That is, the hardening law is generalized to Ωij [9]. Since tension, compression, and shear
include a dependency on the number of cycles have different effects on damage, crack closure
N. Such models usually require detailed coefficients cn and cs, for compression and
computation of each and every cycle during the shear respectively, are introduced in the
life of the material, which may be definition of effective stress, as follows
computationally expensive. A more effective
σ = M ijklσ kl (1)
methodology, in which these two approaches
are combined, is discussed in this paper. where
Micromechanics is applied to predict lamina
ζ ijkl σ kl′ ζ ijkl −σ kl′
stiffness from constituent (fiber, matrix, and M ijkl = + +…
1 − Dij 1 − Dij
interphase) properties[6], and decompose the (2)
I ijkl − ζ ijkl ⎛ 1 1 ⎞
lamina state variables (e.g., stress, strain, ⎜ + ⎟
damage) into their components in each of the
2 ⎝ 1 − Cn Dii 1 − Cs Dii ⎠
constituents [7]. Therefore, damage evolution and ζijkl = 1 if i=j=k=l=1, zero otherwise, and <
models and failure criteria can be formulated at > represent the Heaviside function. Such
the constituent level. definition assures symmetry of the effective
The material parameters simulating damage stress tensor. The effect of crack closure on the
evolution in each phase and their interaction, normal and shear components of effective stress
within a loading cycle, are evaluated from can be adjusted from no effect to full effect by
quasi-static-loading tests [8]. Then, a S-N varying the coefficients cn, cs, from zero to one.
curve of a single laminate is used to adjust two The stiffness tensor C of one configuration
additional parameters governing the evolution (damaged) is obtained in terms of the stiffness
of the magnitude of fatigue damage as a C of the precursor configuration (virgin) by
function of the number of cycles. Finally, using the energy equivalence principle [9], as
predicted S-N curves for other laminates are −1 −1
Cijkl = M ijmn C mnpq M klpq (3)
compared with their corresponding test data.
The only problem encountered was to find The model combines three configurations [32-
fatigue data for the same materials systems for
35]: effective ( σ ), partially damaged ( σ ) and
which comprehensive quasi-static data exists

2
© AES-Advanced Engineering Solutions (Ottawa, Canada)
All rights are reserved
Gasser F. Abdelal, E. J. Barbero, A. J. Robotham
damaged (σ ). In the effective configuration, the where σr indicates the average stress in the
undamaged portion of fiber and matrix carry the phase r=f,m,L, and σ is the average stress in the
load. In the partially damaged configuration, homogenized material. Eq. (4) accounts for the
the fiber and matrix are damaged but the stress redistribution between the fiber and the
interaction damage is not present. In the matrix that must take place when both phases
damaged configuration, all three damages are undergo damage at different rates. Stress
included. The three configurations are redistribution also takes place at the macro-
illustrated in Fig. 1, starting with damaged on level (among laminate) as a result of updating
the right, partially damaged at the center, and of the lamina stiffness tensor C according to
effective on the left. Eq.(3).
The stress and strain concentration tensors are
obtained using [8] in the effective
configuration. Then, the concentration tensors
in the partially damaged configuration are
computed as

= ( c m M ijpq + c f I ijkl )
f −m m −f −f −1
Aijkl Apqrs Arsmn M mnkl

Aijkl =
m
( )
I ijkl − c f Aijkl
f (6)
cm

where I ijkl is a fourth-order identity tensor,

1
Figure 1. Micromechanics damage I ijkl = (δ ik δ jl + δ il δ jk )
2
configuration. δ ij = 1 if i = j (7)
δ ij = 0 if i ≠ j
ML transforms from damaged to partially
damaged (due to interaction damage effects), Similar equations are used to compute the
Mf and Mm transform from partially damaged to concentration tensors in the damaged
effective for fiber and matrix phases configuration in terms of the same in the
respectively. The total damage-effect tensor M partially damaged one. Also for each
that accounts for the combined effect of fiber, configuration, the stiffness tensor of the
matrix, and interaction damage is given by homogenized material is computed by
t
M ijrs (
= c f M ijkl
f f
Bkluv + c m M ijkl
m m
Bkluv ) d
M uvrs (4)
micromechanics
as
At each configuration, mapping between phases
Cijkl = ν f Cijmn
f f
Amnkl +ν m Cijmn
m m
Amnkl (8)
(fiber, matrix, and lamina) is modeled by
micromechanics using the stress and strain where vf and vm are the fiber and matrix volume
concentration tensors, B and A respectively, fractions of the current configuration. The
according to volume fractions in the damaged configuration
are those determined during fabrication of the
σ ijf = Bijklf σ kl , σ ijm = Bijkl
m
σ kl
(5) composite. The volume fractions in the
ε ijf = Aijklf ε kl , ε = Aijkl ε kl
m m

3
© AES-Advanced Engineering Solutions (Ottawa, Canada)
All rights are reserved
Gasser F. Abdelal, E. J. Barbero, A. J. Robotham
effective configuration are different from those Using these explicit equations, the damage
in the damaged configuration because the surface can be written in stress space, and its
effective configuration deals with the volume of shape is the same as that of the Tsai-Wu
undamaged fiber and matrix, both of which are quadratic failure criterion, but its size is
different from the original volumes of fiber and variable as controlled by the magnitudes of the
matrix of the as-produced composite. The fiber damage threshold and damage parameter.
and matrix volume fractions in the effective In summary, Eq. (10) reduces to
configuration are computed by taking into
g = f ijσ iσ j + f ijσ i − ( γ + γ 0 ) (12)
account the amount of damaged volume in the
fiber and matrix phases [8], as follows Eq.(12) coincides with the Tsai-Wu criterion

( ) when γ+γ0 =1. Since the Tsai-Wu criterion

0.5
Deqr = Dijr Dijr , r = f ,m
predicts lamina failure in terms of available
f ν (1 − D
r f
) (9)
ν = strength data, it is possible to determine all the
eq

ν f (1− Deqf ) +ν m (1− D )

m
eq coefficients of the tensors J and H in Eq.(10) in
2.2 Damage Evolution terms of available strength data for each phase
A damage surface is assumed to limit the space [11].
of thermodynamic forces Y for which no The size of the damage surface in Y-space
damage occurs. In this work, an off-centered evolves according to the following equation,
surface is used [10] to account for different ⎡ (
−δ
)⎤
behavior in tension and compression of each γ = c1 ⎢1 + e c2 ⎥ (13)
⎢⎣ ⎥⎦
phase, as follows,
in terms of two empirical parameters c1 and c2
g (Y , γ ) = Y ⋅ J d ⋅ Y + H d . Y − (γ + γ o )
to be determined from experiments, with the
(10) damage parameter changing according to
where J and H are second order tensors of
∂g
material coefficients which are related to the dY
material properties of each phase, γ is the d δ = ∂Y 2 (14)
⎛ ∂g ⎞ ∂γ
damage parameter in Y-space, and γ0 is the ⎜ ∂γ ⎟ ∂δ
⎝ ⎠
damage threshold in Y-space [2]. Assuming
that the damage principal directions coincide The damage multiplier is found interactively so
with the principal material directions, in which that the consistency conditions g=0 and dg=0
case the damage and integrity tensors become are met. That is, after an increment of strain that
diagonal. It is possible to derive explicit causes damage, the Y-state must remain on the
equations relating the thermodynamic forces to g=0 surface with no further change of that
the stress components [6], surface (dg=0). Once the damage surface is
reached, damage accumulates along the normal
1 ⎛ ∂M pqij ∂M rskl ⎞ −1
Ymn = σ ij ⎜ M rskl + M pqij ⎟ E pqrsσ kl to the damage-flow surface defined as,
2 ⎝ ∂Ω mn ∂Ω mn ⎠
Ωij = δ ij − Dij F (Y , γ ) = Y ⋅ J d ⋅ Y − (γ + γ o ) (15)

(11)

4
© AES-Advanced Engineering Solutions (Ottawa, Canada)
All rights are reserved
Gasser F. Abdelal, E. J. Barbero, A. J. Robotham
The magnitude of additional damage is where α and β are parameters adjusted using an
controlled by the damage multiplier so that experimental S-N curve of a laminate, Ωn is the
scalar magnitude of the integrity tensor,
∂F
dDij = μ (16)
Ω n = ( Ωij Ωij )
0.5
∂Yij (18)

The quasi-static-loading model described in this and the normalized stress amplitude Fm is
section is capable of predicting damage defined in [11] in terms of experimental
accumulation for one cycle only. Since the parameters. the scalar magnitude of the damage
evolution of the damage surface, given by tensor can be calculated as
Eq.(13), is independent of the number of cycles, 1
⎡ ⎛ N ⎞ 1(1−α ) ⎤ (1+ β )
no further damage can occur during cycling
Dn = 1 − ⎢1 − ⎜ ⎟ ⎥ (19)
loading at constant strain amplitude. A ⎢ ⎝ NF ⎠ ⎥
⎣ ⎦
phenomenological model is therefore necessary
to predict the accumulation of damage during The magnitude of fatigue damage Dn is used to
cyclic loading, as it is described in the next update the magnitude of the orthotropic damage
section. tensor as

Dn ( n + δ n )
3 Cyclic loading Dij ( n + δ n ) = Dij ( n ) (20)
Dn
A scalar fatigue damage evolution reduces the
amount of fatigue data required, since only two The next step is to decompose the damage into
parameters are need to be adjusted in the fiber and matrix damage. Experimental data
isotropic damage model. Stress redistribution suggest that the rate of damage accumulation
that results from orthotropic damage is dD/dn is approximately constant for most of the
evaluated by recomputing the relative fatigue life, with faster accumulation near
magnitude of damage among the three principal fracture [13]. Therefore, it is possible to assume
material directions. Stress redistribution among that the total increment of damage dD
the phases is also accomplished by distributing decomposes into fiber and matrix damage (r=f,
the scalar damage into the contributions of m) as
fiber, matrix, and interaction effects.
dDklr = α ijkl
r
dDij (21)
The scalar model of Arnold and Kruch [11] was
used to track the magnitude of damage where αr is the damage decomposition tensor
accumulated as a function of the number of for phase r=f,m,L, which is derived as follows.
cycles n. The number of cycles to failure N is
∂g ∂Ymn
predicted as dσ pq
∂Ymn ∂σ pq ∂f
dDij =
⎛ ∂g ⎞ ∂γ
2
∂Yij (22)
⎛⎡ β +1 1−α β +1 1−α ⎞ ⎜ ⎟
⎝ ∂H ⎠ ∂δ
⎜ ⎣1 − (1 − Ω k ) − ⎡1 − (1 − Ω k −1 )
⎤ ⎤
⎦ ⎣ ⎦ ⎟
N=⎝ ⎠ dDij = X ijpq dσ pq
Fmβ (1 − α )( β + 1)

(17)

5
© AES-Advanced Engineering Solutions (Ottawa, Canada)
All rights are reserved
Gasser F. Abdelal, E. J. Barbero, A. J. Robotham
in terms of the damage-stress tensor X. Using with eq. (20). Then, the damage tensor is
eq.(22) and the average stress theorem at the divided to the phases using Eq. (21).
lamina level, Then, the quasi-static model is used again to
refine the damage tensor Dr of each phase. The
dDijd = X ijkl
d
(ν f dσ klf + ν m dσ klm ) (23)
stress is redistributed using eq. (3) at the
from eq.(22) laminate level and using Eq. (5) at the
constituent level. The procedure is repeated
dDijd = X ijkl
d
(ν f X klrs− f d Ωrsf + ν m X klrs− m d Ωmrs ) (24) until the remaining life is equal to zero. The
Using the chain rule pair (S, N) obtained in this way is used to plot
one point in the predicted S-N curve. The
∂Dijd
= ν f X ijkl
d −f
X klrs ( r = f , m) (25) procedure is repeated for several values of
∂D r
rs
strain amplitude S until the complete S-N curve
Then, using eq. (25) into (24) we get the is generated. The predicted curve is then
damage decomposition tensor for phase r=f, m compared with the experimental S-N curve and
as the parameters α and β in eq. (17) are adjusted
to minimize the difference.
∂Drst ∂Drst ∂D pq
d

= = α rsto
r
( no summation ) Nine model parameters (three per phase) are
∂Dtor ∂D pq
d
∂Dtor (26)
needed to track the evolution of damage within
r = f,m
one loading cycle. For each phase (fiber,
which can be used to decompose the total matrix, and lamina), there are three parameters:
damage into the damage in each phase c1r, c2r, in the evolution law eq. (13) and the
according to Eq. (21) damage threshold γor in eq. (10). The nine
parameters are determined by modeling quasi-
3 Computational Procedure static-loading tests for which data are available.
Process starts with a fixed strain amplitude S, The procedure is illustrated using available
one loading cycle is simulated using the quasi- data, shown in Table 1 for T300/5208
static model described in Sect. 2 to determine carbon/epoxy composite [8].
the orthotropic damage tensor Dr of each phase, Table 1. Material properties T300-5208
the damage decomposition tensor αr of each Property Fiber Matrix Lamina
phase (Eq. (26), and the estimated life N of the Modulus, E (GPa) 230 2.6
lamina (Eq. (17). Poisson’s ratio 0.22 0.38 0.268
A finite increment of the number of cycles δn is Initial volume 0.7 0.3
then selected. This can be selected as the lowest fraction
remaining life in the structure, as the number of Ft, (GPa) (trans) 3.654 0.058 0.058
cycles to the next stress amplitude change, or Fc, (GPa) (long.) 1.096 0.1876 1.096
any other user selected number of cycles. The Critical Dt 0.105 0.5 0.105
magnitude of the damage tensor is then
Critical Dc 0.11 0.5 0.11
computed using Eq. (19). The orthotropic
F6, (GPa) 0.086
damage tensor of the lamina is then scaled up
G12, (GPa) 104.5 0.97 5.09
Weibull disper. 8.9
6
© AES-Advanced Engineering Solutions (Ottawa, Canada)
All rights are reserved
 Gasser F. Abdelal, E. J. Barbero, A. J. Robotham
First, a longitudinal tensile test (ASTM D3039
[4]) is simulated. The fiber parameters (c1f, c2f, Table 3. Damage characterises-2 of T300-5208
γof) are adjusted, so that at failure the fiber Parameter Fiber Matrix Lamina
stress equals the fiber strength Fft and the fiber C1 1 1 1
damage equals the known value Dft=1-exp(- C2 -1.1E5 -4.2E6 -1.5
1/m), where m is the Weibull modulus. Second, γ0 -6.5 2.0 0
a transverse tensile test is simulated [4]. The cn 1 1 1
matrix parameters (c1m, c2m, γom) are adjusted,
so that at failure the transverse stress equals the 80

known transverse strength of the composite Fmt

and the matrix damage equals Dmt=0.5 as 75
* Experimental
estimated by [14]. Next, adjust the lamina Theoretical
parameters (c1L, c2L, γoL) to minimize any 70

discrepancies in the shape of the shear stress-

100 (S/Su)
strain plot of a unidirectional lamina, as seen in 65

Fig. 2. The error is measured by the χ2

statistical measure of the difference between the 60

predicted values pi and the experimental values

ei. The values obtained are shown in Table 2-3. 55

7
x 10
8
50
2 2.5 3 3.5 4 4.5 5 5.5 6 6.5
Log(N)
6
shear Stress, Pa

4 * Experimental Figure 3. S-N Relationship for Fatigue Test of

2
Theoretical [±45] 2S T300-5208
Keeping all the parameters fixed, the S-N curve
0
of T300/5208 [0/30/-30]6S laminate [16], also at
-2
0 0.005 0.01 0.015 0.02 0.025 0.03 R=0.1, was predicted and compared with
Shear Strain-part b
experimental data in Fig. 4.
Figure 2. Predicted and experimental, shear
75
stress-strain curve for T300/5208.
* Experimental
The parameters (α=0.2) and (β=12.6) were 70

Theoretical
adjusted to fit the S-N data of [45/-45]2s 65
100(S/Su)

Carbon/Epoxy T300/5208 with R=0.1 [15], as 60

shown in Fig. 3. 55

Table 2. Damage characterises-1 of T300-5208 50

Parameter Fiber Matrix Lamina 45

3.5 4 4.5 5 5.5 6 6.5 7
H1 0.263E-13 0.214E-13 0.147E-10 Log(N)

H2 = H3 0.263E-13 0.214E-13 0.73E-13 Figure 4. S-N Relationship for Fatigue Test of

J11 -0.14E-5 -0.616E-6 -0.0000313 [0/±30] 6S T300-5208.
J22 = J33 -0.14E-5 -0.619E-6 0.129E-8
7
© AES-Advanced Engineering Solutions (Ottawa, Canada)
All rights are reserved
 Gasser F. Abdelal, E. J. Barbero, A. J. Robotham
Using the same parameters, the predicted and estimate orthotropic damage. The nine
experimental S-N curves of T300/5208 constants of the quasi-static model are adjusted
[0/90/45/-45]s laminate [17], also at R=0.1, are with available lamina strength data. Further
shown in Fig. 5. validation with experimental data for other
The effect of the stress ratio R on the number of composite structures would be desirable as the
cycles to failure of the 90-deg lamina of the quasi-static and fatigue data for new material
[0/90/45/-45]s laminate is shown in Fig. 6. systems becomes available. Going forward this
approach will be integrated with process
1.05
models being devised to simulate composite
1 * Experimental
curing during manufacture.
Normalized Secant Modulus

Theoretical
0.95

0.9
Acknowledgments
This work was first started as part of a doctoral
0.85
research program at West Virginia University and
0.8
0 10 20 30 40 50 60 70 80
is currently supported through the Virtual
Cycles in Thousands
Engineering Centre (VEC), which is a University
Figure 5. Stiffness Reduction Curve for a of Liverpool initiative in partnership with the
[0/90/±45] S laminate. Northwest Aerospace Alliance, the Science and
Technology Facilities Council (Daresbury
32
Laboratory), BAE Systems, Morson Projects and
31
Airbus (UK). The VEC is funded by the
Cycles in Thousands

30 Northwest Regional Development Agency

29
(NWDA) and European Regional Development
Fund (ERDF) to provide a focal point for virtual
28
engineering research, education and skills
27
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
development, best practice demonstration, and
knowledge transfer to the aerospace sector.
Stress Ratio

Figure 6. Number of Cycles to Failure of the

90-degree Layers of [0/90/±45] S laminate. References
[1] Liu B. and Lessard L. B., Fatigue and
3 Conclusion Damage Tolerance Analysis of Composite
Eleven parameters define the proposed model, Laminates: Stiffness Loss, Damage
and need to be adjusted with experimental data. Modeling and Life Prediction, Comp. Sci.
Nine parameters are adjusted with quasi-static Techn., 51, 43-51, 1994.
lamina strength data, which is easier and [2] Liu Y. M., Mitchell, T. E., and Wadley N.
cheaper to obtain than fatigue data. Only two G., Anisotropic Damage Evolution in
parameters are adjusted with fatigue data from a Unidirectional Fiber Reinforced Ceramics,
laminate. A fatigue damage evolution model Acta Materialia, 45 (10), 3981-3992, 1997.
using only two adjustable parameters was [3] Mittelman A. and Roman I. (1990) Tensile
integrated with a quasi-static damage model to Properties of Real Unidirectional

8
© AES-Advanced Engineering Solutions (Ottawa, Canada)
All rights are reserved
 Gasser F. Abdelal, E. J. Barbero, A. J. Robotham
Kevlar/Epoxy Composites, J. Composites, [14] Janson J. and Hult J. (1977), Damage
21 (1), 63-69. Mechanics and Fracture Mechanics: A
[4] ASTM Standards (2000) High Modulus Combined Approach, J. Mec. Appl., 1, 69-
Fibers and Composites, v15.03. 84.
[5] Voyiadjis G. Z. and Park T. (1995), [15] Wu, W-F, Lee, L. J. and Choi, S. T. (1996)
Anisotropic Damage of Fiber-Reinforced A Study of Fatigue Damage and Fatigue
MMC Using Overall Damage Analysis, J. Life of Composite Laminates, Journal of
Eng. Mech., 121 (11), 1209-1217. Composite Materials, Vol. 30, No. 1
[6] Barbero, E. J. (1999) Introduction to (1996).
Composite Materials Design, Taylor & [16] Ramani, V. R. and Williams, P. L. (1979)
Francis, Philadelphia, PA. Axial Fatigue of [0/±30] 6S

[7] Aboudi J. (1991) Mechanics of Composite Graphite/Epoxy" Ames Research Center,

Materials: A Unified Approach, Elsevier. NASA.
[8] Herakovich, C. T. (1998) Mechanics of [17] Jamison, R. D., Schulte, K., Reifsnider, K.
Fibrous Composites, John Wiley, N.Y. L., and Stinchcomb, W. W. (1984)
[9] Krajcinovic D. (1996) Damage Mechanics, Characterization and Analysis of Damage
North Holland, Amsterdam. Mechanisms in Tension-Tension Fatigue
[10] Barbero E. J. and DeVivo L. (2001) A of Graphite/Epoxy Laminates, Effects of
Constitutive Model for Elastic Damage in Defects in Composite Materials, ASTM
Fiber-Reinforced PMC Laminae, J. of STP 836, American Society for Testing
Damage Mechanics, 10(1) 73-93. and Materials, 21-55.
[11] Abdelal, G. (2000) A Three-Phase
Constitutive Model for Microbrittle
Fatigue Damage of Composites,
Dissertation, Mechanical and Aerospace
Engineering, West Virginia University.
[12] Arnold, S. M. and Kruch, S. (1991)
Differential Continuum Damage
Mechanics Models for Creep and Fatigue
of Unidirectional Metal Matrix
Composites, NASA Technical
Memorandum TM-105213.
[13] Jamison, R. D., Schulte, K., Reifsnider, K.
L., and Stinchcomb, W. W. (1984)
Characterization and Analysis of Damage
Mechanisms in Tension-Tension Fatigue
of Graphite/Epoxy Laminates, Effects of
Defects in Composite Materials, ASTM
STP 836, American Society for Testing
and Materials, pp. 21-55.

9
© AES-Advanced Engineering Solutions (Ottawa, Canada)
All rights are reserved

View publication stats