Department of Aerospace Engineering

AENGM3100 Research Project

2011-2012

DAMAGE PROPAGATION MODEL FOR THE COMPRESSIVE STRENGTH OF FIBRE

REINFORCED LAMINATES

Ioannis Sophocleous

Department of Aerospace Engineering, University of Bristol, Queen’s Building, University Walk,

Bristol. BS8 1TR. UK.

ABSTRACT

The compressive strength of fibre reinforced laminates is much lower than the tensile strength due to
the fibre misalignment angle, a material property resulting from imperfections and limitations within
the manufacturing process. With the increased application of composites for load bearing
applications, it is essential to have an in-depth understanding of the fundamental factors which
contribute to the fibre kinking failure, and consequently, to the formation of a kink band. To capture
the failure mode, the LaRC03 matrix failure criteria and a damage propagation model based on stress
degradation have been implemented in a user subroutine in the explicit finite element code LS-Dyna.
The failure is shear dominated and is triggered by an initial fibre misalignment angle. Due to the
importance of the shear behaviour, a non-linear shear curve is implemented in the code. To avoid
mesh dependency and strain localisation during damage propagation, a smeared formulation is used.
The validation of the model against existing experimental data reveals the ability of the model to
capture the initiation, propagation and kink band broadening accurately. The predicted kink band and
fibre misalignment angles correlate well with experimental results and the model also shows good
mesh convergence characteristics.

Keywords: Finite Element Analysis (FEA), Failure, Fibre Kinking, Continuum Damage Mechanics,
Compressive Strength

INTRODUCTION

With the increased application of Fibre Reinforced (FR) composites over the last few years, it is
necessary to develop an in-depth understanding of the complex mechanisms and the various aspects of
mechanical failures in these structures. The design of composites requires methods which can predict
the different damage mechanisms which lead to complete failure.

When FR composites are subjected to loading, failure can initiate at the inter or intra-laminar level. At
the intra-laminar level, composites fail by matrix cracking and fibre debonding which leads to fibre
breakage, while inter-laminar failure consists of delamination [1]. This paper focuses on the modelling
of compressive damage in Uni Directional (UD) composites due to plastic microbuckling and fibre
kinking. Compressive failure of FR composites is very complex and of particular interest to the
research industry. The apparent low compressive strength of composites with respect to the tensile
strength is mainly caused by the fibre misalignment angle, which is a result of imperfections in the
manufacturing process. Fibre micro buckling, also known as kinking, is one of the compressive failure
mechanisms that have been extensively studied. During compression, the shear stress acting on the
fibres induces further rotation which leads to the fibre kinking failure. This leads to the formation of a
kink band which can be defined as a localised shear deformation of the matrix and depends mainly on
imperfections in the manufacture process. These include fibre misalignment or waviness, porosity and
residual stresses [2].

Dow and Gruntfest [2] were the first to suggest that the apparent lower compressive strength of
composites is due to fibre micro buckling, while Rosen [3] was the first to initiate tests to predict the
compressive strength of UD composites. Argon [4] extended Rosen’s work by incorporating the fibre
 Ioannis Sophocleous

misalignment angle and the plastic yielding of the matrix to produce the following relationship for the
compressive strength:

(1)

where is the inter-laminar shear strength and is the fibre misalignment angle. Budiansky [5]
further expanded Argon’s formula to include the shear strain , which results in more fibre rotation.

(2)

Following this, many improvements were made and many models were created to study the post
bucking behaviour of the fibres and inelastic behaviour of the matrix. These experiments provided a
greater comprehension of the factors governing the kink band formation such as kink band
propagation angle β, width w and fibre rotation θ shown in Figure 1 [6].

(a) (b)

Figure 1: (a) Micrograph form experiments [7] and (b) kink band characteristics

Moran et al [8] studied the initiation, formation and broadening of the kink band and made several
significant conclusions. He determined that the initial stages of the kink band involve fibre rotation
and matrix deformation which are confined to a narrow band. The material follows a linear elastic
response until a point is reached where the material begins to show non-linear behaviour. This is
followed by rapid fibre rotation and large matrix shearing which is terminated by fibre lock-up. At this
point, the kink band broadens laterally into unkinked material, however, the kink band angle is much
greater than it was at initiation. Berbinau’s [9] subsequent analysis on the kink band propagation
supported Moran’s conclusions and suggested that the kink band forms in three stages: elastic kinking,
plastic kinking and then collapse as a result of the fibre failure at the kink boundaries. Vogler et al [10]
also studied the steady state axial propagation of the kink band and established that the fibre rotation
angle inside the band was 36.8° and that the band was inclined at 18°. In addition, Wass et al [11]
carried experiments on notched uniply composite laminates and revealed that three dimensional
kinking patterns exist. Specifically, as well as obtaining in plane rotations of the fibre, out of plane
rotations where also obtained. The effects of various parameters studied by Zhang et al [6] indicated
that the most important parameter in addition to the fibre misalignment is the shear strength of the
matrix.

Failure Criteria

In order to model the initiation of the kink band, accurate failure criteria are needed. As depicted by
the World Wide Failure Exercise (WWFE), it has always been particularly challenging and of great
interest to the academic research community to accurately predict the compressive strength of

2
 Ioannis Sophocleous

composite materials [12]. Even though this topic is expanding, ‘there is lack of evidence’ to show that
the criteria used provide accurate predictions [13]. The model implemented in this work uses accurate
physically-based criteria which are preferred over standard curve fitting based criteria. Curve fitting
criteria have limited suitability and are applicable only to the load combinations that were used to
derive them. It should be noted that it is also impractical to test every possible loading combination.

The new criterion used for matrix transverse compression applicable to fibre kinking is a model based
on the Mohr-Coulomb (M-C) interaction of the stresses on the fracture plane by Dávila et al [12]. This
criterion makes use of the concepts initially proposed by Hashin [14] and the fracture plane concept
proposed by Puck [15]. Puck’s fracture plane accounts for the influence of the transverse compressive
strength on the matrix shear strength as shown in the failure criteria below.

(3)

where , and are the material’s shear stresses and normal stress acting on the fracture plane as
shown in Figure 2. and are the transverse and longitudinal shear strength respectively. The
internal material friction coefficient and are found experimentally.

Figure 2: Shear traction components and normal stress on the fracture plane [15].

where α is the fracture angle. Puck’s predicted failure envelope shows good agreement with
experimental test results, however Dávila et al [12] argued that the semi-empirical approach uses
several material properties which are very difficult to quantify and thus it is difficult to apply the
criterion.

Based on the M-C criterion, Dávila et al [12] proposed new criteria for matrix failure called the
LaRC03 criteria. In their analysis, a robust criterion for fibre kinking is presented. Initially, the
stresses are rotated to the fibre misalignment plane. By considering a pure axial compression loading,
the initial misalignment angle of the fibres is calculated. The shear stresses on the fracture plane are
further calculated by a transformation of the fracture angle α, as shown in Figure 2. The new set of
fibre kinking criteria accounts for matrix compression. The prediction of failure initiation shows
excellent correlation with experimental results, something which was not possible in the past due to
the various formulations used. For the above reasons, the criteria proposed by Dávila et al [12] form
the basis of the model presented in this paper.

Continuum Damage Mechanics (CDM)

Predicting failure of FR composite materials is not only dependent on the failure criteria used. This is
because after initial damage, the material undergoes large deformations until the final failure is
reached. It is thus essential to combine failure criteria with the appropriate damage model. There are
two main approaches to the propagation of failure model; using the CDM approach and fracture
mechanics approach [16]. The fracture mechanics approach is widely used and precedes the CDM
approach but it is mesh dependent and requires continuous adaptation of the mesh. CDM based
methods do not require this continuous adaptation and are thus more suitable for this model [17].In

3
 Ioannis Sophocleous

CDM, the composite laminate is treated as a continuum where damage tensors are used to degrade the
material properties after the failure criteria has been satisfied:

(4)

where is the ‘effective stress’ based on the cross-sectional area effectively resisting the
loading, is the ‘applied stress’ based on a macroscopic cross-sectional area as if the material was
undamaged and d is the damage variable. Degradation of the material properties is often performed on
strength, stress and stiffness, depending on the model. A ply level stiffness degradation model for
intra-laminar failure mechanisms was devised by Maimí [18]. In his model, a new regularisation
scheme was developed which calculated a critical element length and lowered the strength of the
element to keep the fracture toughness constant in order to avoid mesh dependency. The model was
found to accurately predict the intra-laminar damage accumulation. Damage models on fibre kinking
were initially formulated by Pinho et al [17]. A similar model was created by Ataabadi et al [19] but
by considering strain instead of stress as considered by Pinho et al. The results obtained were as
accurate as Pinho’s model with similar kink band and fibre misalignment angles but the elements were
more distorted. Sørensen et al [20] modelled the actual fibre/matrix system using a micromechanical
based finite element model and focused on the performance of the model with respect to the
convergence rate. The results obtained were heavily mesh dependent at the initial kink band formation.
Another micromechanical model was developed by Hsu et al [21]. The model predicted the kink band
propagation stress accurately and captured the rate dependence of the material.

In this paper, the failure criteria initially developed by Dávila et al [12] and the propagation model by
Pinho et al [17] are described and implemented in Ls-Dyna [22] in a user subroutine. The
implementation is for LS-Dyna’s 8-noded solid elements with one point integration. The model is then
validated on failure of unnotched specimens under pure compression (IM7/8552) and is
experimentally validated on a notched specimen by Gutkin et al [23] (T800/924). Furthermore, a mesh
convergence study is carried out, an important inherent feature of the numerical model that will enable
its applicability into more complex specimens and its interaction with other failure modes.

3D KINKING MODEL

Various models that have been proposed for fibre kinking assume that this mode occurs only in the
plane of the lamina hence all out of plane movements are constrained. The model devised by Dávila et
al [12] exploited only the 2D nature of the failure mode. However, a lot of the undergoing research
suggests that kinking failure occurs in 3D space, where a kinking plane needs to be defined along with
the misalignment plane. The model described in this section is a 3D version of Dávila’s et al model,
proposed by Pinho et al [17].

For a unidirectional lamina with fibres running along the α direction, the kinking plane is defined at an
angle ψ with the b axis as shown in Figure 3. The kinking plane can vary between 0-180° depending
on the load, while for a constrained specimen in the b direction the angle becomes 90°. The stresses in
the fibre kinking plane can be obtained using the following transformation equations.

(5)

where denotes the stresses in the fibre kinking plane. The kinking plane angle can be obtained by:

4
 Ioannis Sophocleous

(6)

Figure 3: Fibre misalignment and fibre kinking plane [17].

Having defined the fibre kinking plane, the same is required for the misalignment plane. A
transformation of the stresses in the fibre kinking plane by an angle results in the stresses in the
misalignment plane shown below, where m denotes these stresses.

(7)

To define the misalignment angle , Dávila et al [12] assumed that the fibre angle can be decomposed
into an initial angle resulting from imperfections and manufacturing flaws as well as the rotation
caused by the shear load. To calculate , the misalignment angle and shear strain at failure under pure
compression were considered from equation 7 [12]. (
. The initial misalignment angle is the difference between the fibre angle and shear
strain at failure:

(8)
(9)

where the subscript c and m indicates failure under pure compression and the misalignment plane
respectively. The misalignment angle at failure is a material property and is given by:

(10)

5
 Ioannis Sophocleous

where is the compressive strength of the material. Having computed the initial misalignment angle,
the shear stress in the misalignment plane can be computed by iteration using equation 11.

(11)

Since matrix behaviour is very important in the fibre kinking failure, a non-linear shear response was
considered by Pinho et al [17]. The authors implied that an inherent characteristic of this model is the
possibility of failure by instability, rather than by matrix cracking. An explanation of this is viable
when considering Figure 4 [17]. The left and right hand side of equation 11 is considered in the plots
for a linear and a non-linear shear response. As shown in Figure 4a, for a specific compressive stress,
the stress rotation curve obtained on the right hand side of the equation intersects both the linear and
non-linear curves. For the specific compressive stress, a solution exists on both curves, whereas for an
increase of this stress, a solution is still obtained for the linear shear curve but not for the non-linear
curve as indicated in Figure 4b. A possible interpretation of this phenomenon is that failure can occur
by matrix cracking, but it can also occur as a result of elastic instability. The instability is experienced
when the stress rotation curve and the non-linear shear response do not coincide with each other. In
such a situation, an increase in the stress leads to fibre instability which results from the unlimited
rotation of the fibres.

Equilibrium Stress rotation

Equilibrium Nonlinear
point for linear point for linear
Linear
shear behaviour shear behaviour
Linear

Nonlinear
Stress rotation There is no equilibrium
Equilibrium point point for non-linear shear
for non-linear shear behaviour
behaviour
0 γ 0 γ
(a) (b)
Figure 4: (a) Equilibrium for both shear responses and (b) no equilibrium point for non-linear shear

Finally, the misalignment angle is given by the following equation.

(12)

Upon defining the fibre kinking and misalignment plane, the failure criterion can be used to detect
matrix failure. For matrix compression ( ), the criterion proposed by Puck and Schürmann [15,
24] is used.

(13)

6
 Ioannis Sophocleous

To apply the criterion only the knowledge of longitudinal shear strength , the transverse
compression strength and the fracture angle for pure transverse compression (typically )
are required. The remaining parameters are obtained from these, as proposed by Dávila et al [12].

(14)

For matrix tensile failure, failure occurs when the quadratic equation shown below is satisfied.

(15)

In addition, the traction components on the fracture plane are given by:

(16)

where is the fracture angle which is obtained by trying a small number of solutions in the interval
.

FINITE ELEMENT (FE) IMPLEMENTATION

LS-Dyna is one of the most widely used FE codes due to its vast capabilities for detecting composite
failures. It has failure criteria and degradation models which are embedded in the code and can be used
to predict failure. The degradation features are used to degrade the stresses linearly to ‘0’ over a fixed
amount of time steps, however the applicability of such features is rather constrained for use in
specific circumstances. Furthermore, LS-Dyna provides features for the addition of a user subroutine
for user defined materials (material cards 041-050). This section describes the propagation model
devised by Pinho [17] and the implementation of the model in LS-Dyna [22].

Elastic Behaviour

The elastic strains used by LS-Dyna and the compliance matrix are used to elastically update the
stresses as shown in the equation below.

(17)

7
 Ioannis Sophocleous

The in-plane shear is not included in these equations as the model uses a non-linear shear response
obtained experimentally.

Non-linear Shear Strain Behaviour

Due to the importance of correctly modelling the in-plane shear behaviour in the fibre kinking failure,
a non-linear shear curve is implemented in the FE analysis through the coefficients of a set of
polynomials. In addition, the shear response after damage is initiated can be very complex with
different loading, unloading and reloading paths. Many experimental results suggest that when
unloading the shear modulus is fairly constant with some hysteresis [15]. The current model requires
only the shear modulus which remains constant when unloading and reloading, hence the small
hysteresis has been neglected.

Damage Variable

To model the behaviour of the material after damage initiation, a damage variable is used to degrade
the stresses. Many degradation models exist, however, this model assumes that the material behaves as
in Figure 5a, demonstrating a bilinear softening constitutive law. The damage variable d is ‘0’ as long
as the material is undamaged and the failure criteria are not satisfied and ‘1’ when the material has
failed completely. As soon as failure is detected, the damage variable is used to degrade the stresses
linearly to zero. The instantaneous damage variable is defined as:

(18)

In fibre kinking, the failure process is associated with the rotation of the fibres under the action of the
shear stress in the misalignment plane until complete failure is detected. As the driving stress
is , the damage variable is formulated with the driving strain , which is defined as:

(19)

In addition, the strain at onset of failure is given by:

(20)

In CDM, when the material exhibits a softening behaviour, strain localisation occurs, which results in
mesh dependency. This is because the energy dissipated decreases as the mesh is refined. To avoid
this, a strength limit is used in the implementation of the model which is not kept constant
throughout the material but is dependent on the characteristic length .

To define the strain at final failure , Pinho [17] employed a smeared formulation which avoids strain
localisation [24]. Strain localisation is a consequence of highly strained zones in the model before the
detection of damage. Damage models developed particularly for kink bands suffer from strain
localisation and are therefore very mesh dependent. The method presented by Pinho [17] is based on
the one initially devised by Bažant et al [25]. In the methodology used, fracture is characterised by
only three parameters: the fracture energy, the uniaxial strength and the width of the crack band. The
fracture energy is distributed over the entire volume of the element and can be obtained as a

8
 Ioannis Sophocleous

function of the fracture energy per unit area of the surfaces created by damage, Γ, the material strength
and the characteristic length .
(21)

where is the shear strength of the material. A method to obtain the fracture energy using standard
tests is given by Miami et al [26].

(a) (b)
Figure 5: (a) Material in transverse loading and (b) material behaviour after damage [17]

Considering an element of dimensions and as shown above in Figure 5b, the characteristic
length is defined as:

(22)

where is the fracture area for matrix failure. The exact calculation of can be
found in Pinho’s paper [17]. For the calculation of , the nodal coordinates of the elements are
automatically obtained within the material subroutine.

Stress Degradation

Once the material is damaged, the damage variable is initiated and used to degrade shear stresses on
the kink band, as well as the normal stress to the kink band depending on its sign.

(23)

where is the Mc-Cauley operator defined as . In the stress degradation

equations, the stress on the right is being degraded, while the stress on the left is considered to be the
new stress. The stresses which are degraded correspond to the driving stresses responsible for the
failure at a local level. In this case the shear stress in the misalignment and fibre kinking plane are
degraded. In addition, the Mc-Cauley operator is used on the stress in the misalignment plane.
When the stress is compressive, it tends to close any cracks formed in the matrix hence degradation is

9
 Ioannis Sophocleous

only performed when the stress is tensile. Since the driving strain is dependant on , as shown
in equation 19, the degradation of can cause problems in the calculation of the damage variable,
as it will be lower than the damage variable from the previous time step. To enforce irreversibility, a
constraint is imposed on the damage variable as shown:

(24)

As long as the damage variable is lower than the maximum damage variable, the stress degradation on
the particular element is stopped. The element is still considered to be damaged and further loading is
applied until the damage variable increases. Following the degradation of the stresses in the local
coordinate plane, the stresses in the global coordinate frame are updated. Since the degradation used in
this model is stress based there is no stiffness degradation, hence the global stresses need to be
updated. The new degraded stresses are used to compute the equivalent global stresses used by LS-
Dyna [22]. This is done by considering a back transformation of the fibre misalignment angles (– )
and fibre kinking angle (– ) respectively. This effectively changes the load carrying capacity of the
failed element and LS-Dyna distributes the load to the rest of the specimen. This forms the basis of the
damage propagation model. The initiation and damage propagation model is implemented in Ls-Dyna
as shown in Figure 6.

Development of the 3D model (Patran)

Compute stresses in the fibre kinking and

misalignment plane

Continue
Apply matrix failure criteria (LaRC03)
loading

NO Damage
detected

YES
Compute damage variable d

NO

YES

YES
STOP

NO

Degrade local stresses - update global stresses

Figure 6: Implementation of the model in LS-Dyna.

10
 Ioannis Sophocleous

RESULTS

The numerical model was implemented to assess its ability to capture the initiation, propagation and
broadening of the kink band. The model was applied to both notched and unnotched specimens. To
generate the compressive stress, a linear displacement with respect to time was implemented. All the
specimens were created in Patran.

Single and Multiple Element Failure

To verify the applicability and robustness of the model it was applied to unnotched UD specimens.
Eight-noded solid elements were used for meshing the specimen, while the mesh size was varied to
assess the mesh dependency of the model. To produce an initial defect on the specimen, a single
element in the middle and multiple elements in the kink band direction were given lower mechanical
properties with respect to the rest of the specimen. The main reason for this was to create a stress
concentration in order to capture the propagation of the kink band. The mechanical properties of
IM7/8552 are shown in Table 1.

Table 1: IM7/8552 mechanical properties

160 11 5 3.8 0.32 0.43 250 1690 174 170

where the subscripts 1 denote the fibre direction,2 and 3 denote the transverse direction, E is the
Young’s Modulus, G is the shear modulus and ν is the Poisson’s ratio. and are the transverse
tensile and compression strengths and is the longitudinal compressive strength .

The specimen and its dimensions are shown in Figure 7a. The failure is very abrupt and catastrophic,
with misalignment angles around 10° at initiation of failure, leading to complete instability of the
fibres after one time step. At initial damage the compressive strength is predicted to be 1610MPa.
Kink band propagation angles are also very low, as shown in Figure 7c.

Damage
Variable

17

3 α 10
(a) (b) (c)
Figure 7: (a) Unnotched specimen loading conditions and dimensions in mm; (b) initial failure of
element with degraded properties and (c) kink band formation.

The failure propagation with multiple elements having degraded properties in the kink band direction
is shown in Figure 8.

11
 Ioannis Sophocleous

(a) (b) (c) (d)

Figure 8: (a) Elements with degraded properties; (b) initiation of failure; (c) and (d) kink band
propagation.

The model was tested for mesh convergence by doubling the in-plane and through thickness elements.
The resulting force displacement graphs for each mesh density are shown in Figure 9.

Figure 9: Force displacement curves for different mesh densities.

Experimental Validation

To validate the model against experimental data, experimental data by Gutkin et al [23] was compared
against the numerical model. Gutkin et al investigated the longitudinal compressive failure of notched
UD specimens and observed the failure process. The material used was T800/924 and its mechanical
properties are shown in Table 2.

Table 2: T800/924 mechanical properties

148 9.25 6 4.9 0.36 0.49 250 1615 135 105

12
 Ioannis Sophocleous

The UD specimen and the boundary conditions used are shown in Figure 10a with fibres running
along the α direction. The specimen was constrained in the b direction as shown in Figure 10b, and the
lower surface of the specimen was also constrained in the c direction. For the mesh, elements of length
0.016mm were used for the tip of the crack (3476 elements mesh) and the kink band propagation area
with an aspect ratio of 1, while for the remaining specimen a coarse mesh was used to reduce the
computation time.

α
b

b
α
(a) (b)

Figure 10: (a) Dimensions (mm) of specimen used [23] and (b) specimen created in Patran.

Figure 11 shows the experimental and numerical kink band formation at the crack tip of the specimen.
The experimental kink band is divided into 3 different regions for comparison of the experimental
data. The highly damaged zone at the crack tip has a damage variable of 0.9. The results obtained from
the experiment and the numerical model are shown in Table 3.

Damage
Variable

(a) (b)

Figure 11: (a) Experimental kink band [23] and (b) formation of kink band in simulation.

13
 Ioannis Sophocleous

Table 3: Experimental and numerical results

Experimental [23] ~ 20° >40° ~ 30° ~ 18°

Numerical 5-7° 18-20° Failure due to instability 23-35° 12-18°

The kink band initiation, propagation and broadening at different points on the force displacement
curve is shown in Figure 12.

Figure 12: Force displacement with kink band initiation, propagation and broadening

Mesh Convergence

Figure 13: Force displacement curves for different mesh densities

14
 Ioannis Sophocleous

The force displacement curves shown in Figure 13 do not represent the actual loads that would be
experienced, as the specimen used for the numerical model was different in thickness to save
computation time.

For completeness, since no experimental data is available for validation, the compressive stress and
shear stress at initial failure were found to be 904 and 102 respectively. The initial
fibre misalignment angle (material property) was also calculated as 3°.

DISCUSSION

The results presented previously show some interesting progress in this research area. However, an in
depth-analysis of the results is essential in order to investigate their retrospective credibility and
conclusions made. Moreover, there are further areas to be considered for future investigation.
Experimental data based solely on fibre kinking are scarcely found due to the difficulty of capturing
the failure process in the absence of other failure modes. Consequently, a quantitative validation has
not been possible for the purposes of this paper. In the case of unnotched specimens, due to the lack of
experimental data the findings are only compared with the literature review presented in the
introduction. Two conditions are discussed below, these being, notched and unnotched specimens.
Furthermore, the mesh convergence studies will also be considered.

IM7/8552 Unnotched Specimen

In order to verify the accuracy of the model and to assess its applicability, an unnotched specimen was
used which contained an initial defect. Understanding the failure process in an unnotched specimen
under compression is very important due to such specimen’s increased applicability in structural
elements. As it can be seen in Figures 7 and 8, the kink band initiation and formation is captured with
single and multiple element failures. Observing these images shows that the angle of the kink band is
much lower when compared to theoretical values which range between 20-30°. As the damage is very
high, the degradation does not change principal stresses direction as much, thus the damage propagates
at a lower angle than expected. More explanation regarding the propagation angle can be found in the
notched specimen explanation. The unnotched specimen is a good validation of the findings by Moran
et al as explained in the Introduction. Specifically, in Figure 8, the damage initiation, kink band
propagation and broadening can be seen. As the shear stress in the misalignment plane is degraded,
there is a distribution of stress to the undamaged material. This causes the fibres surrounding the
matrix to rotate and the kink band to be broadened. Fibre rotation in the kink band remains constant as
the kink band broadens something which Moran et al described as fibre interlocking.

T800/924 Notched Specimen

Once the model capturing the kinking failure mode had been verified, it was subsequently applied to a
more complex notched specimen. The experimental results by Gutkin et al [23] provide very good
graphical representation of the kink band formation at the crack tip, as shown in Figure 11a. Figure
11b provides a line drawing of the main features of the kink band as these have been characterised by
the author, while Figure 11c shows the resulting kink band obtained numerically. The numerical
model depicts the formation of the kink band when compared to the experimental results. The kink
band formation can be divided into three different regions, as shown in Figure 11a. Region 1 is at the
crack tip, region 2 is at the middle of the kink band and region 3 is at the kink band tip.

In region 1, the numerical model captures the complex failure mechanisms which are responsible for
the damage. Due to the high shear stress at the tip, the fibres fail as a result of the instability
mechanism which is presented in the Implementation section. At the region where failure occurs due
to instability, unlimited fibre rotation is obtained which mimics the failure obtained experimentally.
As already explained, failure can occur due to instability but also as a result of matrix cracking. In this
case, due to the high combination of shear stress and longitudinal compressive stress at the crack tip,

15
 Ioannis Sophocleous

the non-linear shear response and the shear calculated in the misalignment plane do not intersect. This
result can be seen in Figure 14, which shows the fibre misalignment angle for the specimen. The
region shown in blue denotes the area of undamaged material and initial fibre misalignment angle of
3°. At the region near the crack tip, the pink colour denotes failure by instability as the fibres have
unlimited rotation in excess of 50°.

θ /rad

Figure 14: Fibre instability at crack tip.

In region 2, the numerical model captures the kink band propagation and the fibre misalignment which
match the experimental results. The kink band propagation angle is also shown to match the
experimental result. The kink band propagates at an angle of 18-20° which also matches the
experimental result and the findings by Vogler et al [10]. Since this failure is shear driven, the kink
band is expected to propagate at the plane of maximum shear stress, which lies at 45° to the loading
axis. The ability to capture the propagation correctly lies in the implementation of the damage model
in Ls-Dyna. The damage calculated is used to degrade the shear stress in the misalignment and fibre
kinking plane, as well as the stress normal to the kink band depending on its sign. Regarding the shear
stress in the misalignment plane , an explanation for the kink band propagation angle is best
explained using the Mohr’s circle in the misalignment plane as shown in Figure 15 below.

The circle corresponds to the stress state for a single element. The continuous circle represents the
element when undamaged. The dotted circle represents the stress state of the damaged element. Point
1 denotes the stresses in the global coordinate frame which are used by LS-Dyna. These stresses are
rotated to the fibre misalignment plane, as shown by equation 7 in the 3D Fibre Kinking section, by an
angle θ (denoted as 2θ on the Mohr’s circle). The new stress state at the local misalignment plane is
denoted by 2. When the failure criterion is satisfied, a damage variable d is used to degrade the shear
stress in the misalignment plane, as explained by equation 18 in the Implementation section, and the
new stress state becomes point 3. The final step in the degradation model is to transform the new
degraded stresses back to the global stresses at point 4. The transformation is identical to the initial
transformation but with a –θ angle. The purpose of the degradation model used is to increase the
principal angle denoted by as shown in the Figure. At stress state 1, the principal angle is lower
than at state 4. This difference denotes the direction of the maximum shear in the element. As the
element is damaged and degraded, LS-Dyna distributes the load to other elements. Since this failure
mode is shear dependent and the failure criteria depend on changes in shear on the fracture surface,
nearby elements will fail according to the maximum shear direction. Consequently, the failed elements
will produce the kink band shown in region 2.

16
 Ioannis Sophocleous

1
4

2
-2θ
2θ σ
3

Figure 15: Mohr’s circle.

In region 3, a large percentage of the material has been damaged. The shear stress is much smaller as
most of the stress has been redistributed to the rest of the material. Consequently, the rotation of the
fibres is small and the kink band angle decreases. Further loading after the kink band has propagated
up to this point, leads to kink band broadening and loss of a large percentage of the load carrying
capability of the material.

Kink Band Initiation, Propagation and Broadening

The various stages of the kink band formation along the force displacement curve are depicted in
Figure 12. Under the action of the compressive stress, shear in the misalignment plane rotates the
fibres causing matrix failure as the shear strength of the matrix is exceeded. The initial force
displacement response is linear elastic and as the specimen fails by matrix damage at the tip of the
crack, small oscillations are experienced. As the load increases the kink band becomes more profound
with the load distribution and the fibres are further rotated under the shear stresses. The load
displacement curve shows some non-linearity before the ultimate load is reached. At ultimate load, a
sudden increase in the fibre misalignment angle is seen along with fibre lock-up. At this point, the
loading drops dramatically and the kink band broadens into unkinked material in the longitudinal
direction. The numerical kink band formation shows a very good correlation with the findings of
Moran et al [8].

Mesh Convergence Study

Various models for fibre kinking presented in the Introduction, suffered from strain localisation and
mesh dependency of the results due to the high strain regions in the fibre kinking failure. The mesh
dependency of models based on CDM has been accounted for in this model by using a smeared
formulation as described in the Implementation section.

Mesh convergence is an important aspect of a numerical model which ensures that the results of the
analysis are not affected by changing the mesh size. To assess the convergence characteristics of the
model, the in-plane elements were doubled each time. For the unnotched specimen, the elements
through the thickness were doubled as well, but for the notched specimen, the elements were chosen
with respect to the element size at the crack tip. For areas of high stress concentration such as the

17
 Ioannis Sophocleous

crack tip, the aspect ratio of the elements was kept close to 1. A similar strategy had to be applied to
the elements through the thickness, as shown in Figure 16.

Crack tip

Figure 16: Mesh at the crack tip.

This was necessary because if the elements through the thickness had an aspect ratio much greater
than 1, it would not be possible to capture the kink band formation and other areas of the specimen
would fail instead. The specimens that were considered for the mesh convergence study were able to
predict the kink band formation. Any models that did not capture the failure were not included in the
study.

The effect of mesh refinement on the load displacement curves is shown in Figures 9 and 13. It is
evident from these images that mesh convergence is obtained for both models. In the unnotched case,
the same values are obtained when the mesh is refined as there are not any high stress concentration
regions that would affect the convergence and the failure criteria to any great extent. However, for the
notched case, the convergence is not as good but is still adequate for the complexity of the particular
specimen. Upon review, the final values obtained are shown to be within a 3% difference. This allows
the use of coarser meshes for the prediction of the failure without much loss of accuracy and with
lower computation times. The convergence seen and obtained on both specimens is due to the correct
calculation of the fracture energy and its distribution over the volume of each element.

Further Work

Due to the importance of the shear behaviour in fibre kinking, a non-linear irreversible shear
behaviour is implemented in the model. Consequently, various assumptions have to be made in the
implementation of the shear behaviour. One such assumption is that the shear reloads upon unloading
with a constant shear modulus and with no hysteresis. Assumptions of this kind can possibly influence
the accuracy of the model. On a similar basis, the shear response implemented also needs further
development in the case where the shear reverses direction.

The model is also able to predict areas of high damage based on the intersection of the non linear shear
and the rotation of the stresses. This method of predicting instability requires further development in
view of the fact that it is not based on the physical behaviour of the material. This also results to high
stresses in the damaged zone which will influence the numerical results.

The numerical model correlates well with experimental findings based on a qualitative analysis of the
results. For a full validation of the model, more experimental work is needed. Due to the complexity of
capturing the failure mode, existing numerical results cannot be used to validate this model
quantitatively. Although some extensive experimental work has been performed on the compressive
failure of unnotched and notched specimens, in most cases the resulting failure is not based
exclusively on fibre kinking but is also brought about by the complex interactions of other failure

18
 Ioannis Sophocleous

modes as well. Once validation is complete, the model can be applied to notched specimens to capture
the interaction of the fibre kinking failure with other failure modes, and more specifically the
interaction with delamination.

CONCLUSION

In the present work, it is demonstrated that the numerical model implemented in LS-Dyna reproduces
the main characteristics for the formation of the kink band. The results show a very good correlation
with existing experimental work, capturing the kink band initiation, formation and broadening. The
mesh dependency of CDM models and the strain localisation problems faced with the kink band
formation are avoided by employing a smeared formulation based on the fracture energy. Due to the
importance of shear, the non-linear shear response implemented in the model captures correctly areas
of fibre instability, these having also been reported by experimental findings. All the parameters used
in the model have a clear physical meaning based on the physics of the failure and can be easily
obtained from experimental tests. On the basis of the qualitative analysis performed and the mesh
convergence obtained, the model may potentially have significant application in the context of more
complex failure mechanisms.

ACKNOWLEDGEMENTS

I would like to express my gratitude to my advisor, Dr Giuliano Allegri, whose encouragement,

understanding and support throughout the entire research project has enabled me to succeed in the
completion of this project.

This project would also not have been possible without the invaluable assistance and continuous
support from Dr Xiangqian Li.

REFERENCES

[1] D. Hull and T.W. Clyne, An introduction to composite materials, Cambridge Solid State Science
Series, Cambridge University Press, Cambridge, 1996.
[2] N.F. Dow, T.S. Gruntfest, Determination of most needed, potentially possible improvements in
materials for ballistic and space vehicles, General Electric Co, Space Science Laboratory, 1960,
pp. 389
[3] V.W. Rosen, Mechanics of Composite strengthening, Fibre Composite Materials American
Society for Metals, Metals Park, Ohio, 1965, pp. 37-75.
[4] A. Argon, Fracture of composites. Treatise on Materials Science and Technology, Vol. 1, 1972,
pp. 79-114.
[5] B. Budiansky, Micromechanics, Computers and Structures, 16(1983), 1983, pp. 3-12.
[6] Z. Yongbo, f. Huimin, On the longitudinal compressive strength prediction of unidirectional
laminated composites based on an improved model, Polymer Composites, 32(11), 2011, pp.
1817-1826.
[7] S. Pimenta, R. Gutkin, S.T. Pinho, P. Robinson, A micromechanical model for kink-band
formation: Part I - Experimental study and numerical modelling. Composites Science and
Technology, 69(2009), 2009, pp. 948-955.
[8] P. M. Moran, X. H. Liu, C. F. Shih, Kink band formation and band broadening in fiber
composites under compressive loading, Acta Metallurgica et Materialia, 43(8), 1995, pp. 2943-
2958.

19
 Ioannis Sophocleous

[9] P. Berbinau, C. Soutis, I.A. Guz, Compressive failure of 0° unidirectional carbon-fibre-

reinforced plastic (CFRP) laminates by fibre microbuckling, Composites Science and
Technology, 59(9), 1999, pp. 1451-1455.
[10] T. J. Vogler, S. Kyriakides, On the initiation and growth of kink bands in fiber composites: Part
I. experiments, International Journal of Solids and Structures, 38(15), 2001, pp. 2639-2651.
[11] A. M. Waas, A. JungHyun, A.R. Khamseh, Compressive failure of notched uniply composite
laminates, Composites Part B: Engineering, 29(1), 1998, pp. 75-80.
[12] C. G. Dávila, P.P. Camanho, C.A. Rose, Failure criteria for FRP laminates. Journal of
Composite Materials, 39(4), 2005, pp. 323-345.
[13] M. J. Hinton, P. D. Soden, Predicting failure in composite laminates: the background to the
exercise. Composites Science and Technology, 58(7), 1998, pp. 1001-1010.
[14] Z. Hashin, Failure criteria for unidirectional fibre composites, Journal of Applied Mechanics,
47(2), 1980, pp. 329-334.
[15] A. Puck, H. Schürmann, Failure Analysis of FRP laminates by means of physically based
phenomenological models, Composites Science and Technology, 58(7), 1998, pp. 1045-1067.
[16] R. Talreja, A continuum-mechanics characterization of damage in composite-materials,
Proceedings of the Royal Society of London Series a-Mathematical Physical and Engineering
Sciences, 399(1817), 1985, pp. 195-216.
[17] S. T. Pinho, L. Iannucci, P. Robinson, Physically based failure models and criteria for laminated
fibre-reinforced composites with emphasis on fibre kinking. Part II: FE implementation,
Composites Part A: Applied Science and Manufacturing, 37(5), 2006, pp. 766-777.
[18] P. Maimí, P. Camanho, J. Mayugo, C. Dávila, A continuum damage model for composite
laminates: Part I- Constitutive model, Mechanics of Materials, 39(10), 2007, pp. 897-908.
[19] K. Ataabadi, Z.R. Saeed, H.T. Hossein, An improved model for fibre kinking analysis of
unidirectional laminated composites, Applied Composite Materials, 18(3), 2010, pp. 175-196.
[20] K. D. Sørensen, H. M. Jensen, L. P. Mikkelsen, User subroutine for compressive failure of
composites, 2009 Simulia Customer Conference, Simulia, 2009, London, UK, pp. 618-632.
[21] S. Y. Hsu, T.J. Vogler, S. Kyriakides, On the axial propagation of kink bands in fiber
composites: Part ii analysis, International Journal of Solids and Structures, 36(4), 1999, pp. 575-
595.
[22] Livermore Software Technology Corporation. California, USA, LS-Dyna 970; 2003.
[23] R. Gutkin, S.T. Pinho, P. Robinson, P.T. Curtis, On the transition from shear-driven fibre
compressive failure to fibre kinking in notched CFRP laminates under longitudinal compression,
Composites Science and Technology, 70(8), 2010, pp. 1223-1231.
[24] A. Puck, H. Schürmann, Failure Analysis of FRP laminates by means of physically based
phenomenological models. Composites Science and Technology, 62, 2002, pp. 1633-1662.
[25] Z. P. Bažant, B. H. Oh, Crack band theory for fracture of concrete. Materials and Structures
(RILEM, Paris), 16(93), 1983, pp. 155-177.
[26] P. Maimi, P. P. Camanho, J. Mayugo, C.G. Dávila, A thermodynamically consistent damage
model for advanced composites, Tech. Rep, NASA/TM-2006-214282, 2006.

20