The numerical modelling of

ceramics subject to impact

using adaptive discrete
element techniques
M.G. Cottrell
Department of Civil Engineering, University of Wales, Swansea
J. Yu and Z.J. Wei
Rockeld Software Ltd, Technium, Prince of Wales Dock, Swansea
D.R.J. Owen
Department of Civil Engineering, University of Wales, Swansea
Keywords Adaptive techniques, Ceramics
Abstract In recent years, developments in the eld of lightweight armour have been of primary
importance to the defence industry. This necessity has led to many organisations adopting
composite armours comprising both the traditional heavy armours and new lighter weight ceramic
armours. The numerical modelling of metal based armour systems has been well documented over
the years using purely continuum based methods; and also the modelling of brittle systems using
discrete element methods, therefore it is the objective of this paper to demonstrate how a coupled
nite and discrete element approach, can be used in the further understanding of the quantitative
response of ceramic systems when subjected to dynamic loadings using a combination of adaptive
continuum techniques and discrete element methods. For the class of problems encountered within
the defence industry, numerical modelling has suffered from one principal weakness; for many
applications the associated deformed nite element mesh can no longer provide an accurate
description of the deformed material, whether this is due to large ductile deformation, or for the
case of brittle materials, degradation into multiple bodies. Subsequently, two very different
approaches have been developed to combat such deciencies, namely the use of adaptive remeshing
for the ductile type materials and a discrete fracture insertion scheme for the modelling of material
degradation. Therefore, one of the primary objectives of this paper is to present examples
demonstrating the potential benets of explicitly coupling adaptive remeshing methods to the
technique of discrete fracture insertion in order to provide an adaptive discontinuous solution
strategy, which is computationally robust and efcient.
The Emerald Research Register for this journal is available at The current issue and full text archive of this journal is available at
http://www.emeraldinsight.com/researchregister http://www.emeraldinsight.com/0264-4401.htm
This work was sponsored in part by the Engineering and Physical Sciences Research Council
(EPSRC) under the support of the Defence Science and Technology Laboratories (DSTL),
Chertsey, UK.
Notation
A Johnson-Cook yield strength; initial
or current elemental area
B Johnson-Cook hardening coefcient
C Johnson-Cook strain rate constant
C
v
specic heat capacity
E elastic modulus
EC
20,1
82
Received May 2002
Accepted September
2002
Engineering Computations
Vol. 20 No. 1, 2003
pp. 82-106
q MCB UP Limited
0264-4401
DOI 10.1108/02644400310458856
Introduction
The use of ceramic material is becoming more important in the design of
lightweight vehicle armour. Combined ceramic and metal armours have the
potential to be used in highly weight efcient armour systems not only for
todays projectile velocities (1,500-1,800 m/s) but also for considerably higher
impact velocities that are inevitably likely to be used tomorrow
(2,500-3,000 m/s). Therefore, it is of paramount importance to advance the
development of robust and efcient numerical strategies that can aid in the
development and understanding of such composite ceramic and metallic
armour systems.
The purpose of this study is to demonstrate that techniques based on
continuum adaptive remeshing and discrete element fracture can be coupled to
obtain well-dened and quality time-resolved results for the impact of metallic
deformable projectiles into ceramics. Enhancement of the understanding of the
fracture processes, which occur for various impact velocities, is of primary
importance.
Continuum adaptive remeshing
Adaptive remeshing is today becoming an increasingly recognised important
feature of nite element modelling in areas of application such as cold forging,
crash worthiness tests and, of principal interest here, ballistic impact tests that
are frequently encountered within the defence industry (Camacho and Ortiz,
1997; Cottrell et al., in press). The key factors associated with any adaptive
mesh generation analysis are:
f
t
maximum tensile strength
f adaptive error assessment
frequency; adiabatic heating
conversion factor
F elemental or local failure factor
F associated nodal failure factor
G elastic shear modulus
G
f
specic fracture energy
h nite element mesh size
h

adapted nite element mesh

size
h
min
minimum global nite element
mesh size
h
max
maximum global nite element
mesh size
i, j, k elemental index
K elastic bulk modulus
m Johnson-Cook softening exponent
n Johnson-Cook strain hardening
exponent
N elemental shape function
p nodal index
q eld variable
T Johnson-Cook current temperature;
target thickness
T
melt
Johnson-Cook melting temperature
T
ref
Johnson-Cook reference temperature
V initial or current elemental volume
w weighting factor for fracture
insertion
a initial or current nite element
corner angle
1
f
inelastic fracturing strain
1
f
c
critical inelastic fracturing strain
1 total incremental strain rate
1
p
inelastic incremental strain rate
1
p
effective inelastic strain
L

adaptive state variable measure

j local elemental error
u elemental discrete fracture
orientation
u

associated nodal discrete fracture

orientation
V element
Numerical
modelling of
ceramics
83
.
the ability to base models on a geometric entity or assignment level;
.
indication of the quality associated with element distortion for a given
mesh;
.
the automated prediction of element density within the newly generated
mesh;
.
the re-generation of the mesh based upon the predicted densities, and
.
the accurate transfer of history dependent variables between the two
meshes.
A large amount of literature associated with the above listed topics can be
found elsewhere (Camacho and Ortiz, 1997; Cottrell et al., in press; Dyduch
et al., 1992; Morancay et al., 1997; Peric et al., 1994, 1996, 1999; Yu, 1999;
Zienkiewicz and Zhu, 1987, 1991) and it is the intention of this paper to
provide only a brief technical description of techniques that can be
successfully employed to provide an efcient coupled adaptive/discrete
formulation.
Mesh adaptivity is the process by which the element mesh is changed at
selected intervals, either to preserve the quality of element shape or to control
the error in the solution. Mesh adaptation is triggered by control criteria, which
monitor threshold values of specic problem parameters. Typical quantities
such as element distortion or element and nodal quantities such as velocity
gradients are often employed (Camacho and Ortiz, 1997; Yu, 1999). Aremeshing
indicator based on element distortion (Camacho and Ortiz, 1997; Dyduch et al.,
1992) is used in this work and has been proven to provide an efcient criterion
for large deformation problems. Typically element distortion is measured from
the ratio of internal angles, or the ratio of areas between the initially generated
nite element and the deformed nite element, which provides
a computationally cheaper alternative. An error estimator, for example, the
Zienkiewicz error norm type (Zienkiewicz and Zhu, 1987), is then employed to
provide the prediction of the element density distribution required in the new
nite element mesh.
The distortion of a given nite element can adversely affect the ability to
obtain a numerical result that is of the desired resolution. To this end Dyduch
Dyduch et al. (1992) proposed that a function based on the current and initial
elemental areas could provide an indicator that can be used as a trigger for
continuum adaptive remeshing.
j
k

A
c
2A
0
A
0
100 1
In equation (1), it is clear to see howthe proposed remeshing indicator is formed
as a function of both the initial and current element areas, which are denoted by
their appropriate subscripts.
EC
20,1
84
Due to the simplistic nature of this indicator, there are key issues that render
the above formulation problematic for many mesh discretisations. One of the
fundamental difculties, lies with the fact that when an element undergoes a
purely shear based distortion, then zero volume change will be realised, hence
the element can potentially be highly distorted yet the indicator will suggest
that there is zero deformation.
Therefore it has been proposed that supplementary checks be made if the
initial indicator, in equation (1), signals that there is zero element distortion,
whereby an additional measure is derived from the internal nodal angles of the
considered triangular area. The function that checks each corner angle in turn
is given in equation (2), and if this function yields a distortion value that is not
realised using the original triangular area estimator this value will be used as
an alternative.
j
k

a
c
2a
0
a
0
100 2
This form of remeshing indicator lends itself as an ideal tool for the automated
triggering of adaptive remeshing processes due to the fact that the entire
indicator computation is relative based.
Once the remeshing estimation process has been completed and it is deemed
that one or more elements violate the selected error criterion, then mesh
prediction must be completed immediately prior to the generation of a new
mesh. The task of the error estimator is to provide an error map for the
prediction of the element densities within the new nite element mesh.
The development of error estimator procedures for nonlinear problems is
still the subject of intense research and, as yet, no universal measure that is
applicable to completely general problems is available. Consequently, explicit
error measures are currently employed to control the mesh density for classes
of problems exhibiting specic phenomena.
For example, for problems in which the strain elds are predominantly
elastic and continuous the standard elastic energy norm indicator proposed by
Zienkiewicz and Zhu (1987) can be used to good effect. However, situations that
involve strain localisation phenomena, such as multi-fracturing quasi-brittle
solids, can benet from using strain rate criteria to develop appropriate
adaptive mesh predictions. On the other hand, error estimates for problems
involving large-scale plastic deformation, such as forming processes
or penetration of ductile materials, are more appropriately based on either
incremental or total plastic work indicators (Peric et al., 1994).
To illustrate the procedure by which element density predictions are
produced in a mesh adaptivity process, the process is summarised below for
the elastic energy norm error estimator.
The associated error for a generic element k in the form of the standard
energy norm is dened as
Numerical
modelling of
ceramics
85
_
_
e
*
s
_
_
2
k

_
V
_
s* 2s
h
_
T
D
21
_
s* 2s
h
_
dV 3
where the terms
h
represents the stress tensor obtained from the nite element
analysis and the term s* denotes the solution obtained using the process of
smoothed stress recovery (Zienkiewicz and Zhu, 1991). The associated error
from the elemental level is summed over all of the elements in the domain to
give,
_
_
e
*
s
_
_
2

n
elem
i1
_
_
e
*
s
_
_
2
i
4
Then the relative global error is given by the function, which comprises the
summed elemental error and the standard energy norm terms as
h
s
<
_
_
e
*
s
_
_
_
C
*
s

_
_
e
*
s
_
_
2
1
2
; C
*
s

_
V
_
s
T
D
21
s
_
dV 5
If we consider the relative global error h
s
with the relative target allowable
error h, then the local target error for each element is dened as
e h

C*
n
elem
_ _

6
which is then used to evaluate the element error indicator j
k
.
j
k

_
_
e*
_
_
k
e
7
If the element error indicator is less than unity the current mesh density for the
considered element may be coarsened, if however, the element error indicator is
greater than unity then the mesh density is required to be increased. Through
use of the current element size, the predicted element size may be obtained
based on the asymptotic convergence properties of elements, and restricting
attention to linear elements, as

h
k

h
k
j
k
8
where the term h
k
is the considered element current size.
For problems involving strain localisation, as is the case in multi-fracturing
solids, the current incremental strain rate provides a very effective measure for
predicting the new mesh density requirements. Such an approach has been
EC
20,1
86
successfully employed by Yu (1999) for the numerical modelling of fracturing
events including discrete ballistic impact tests similar to those described in
Cottrell et al. (in press), Strassburger and Senf (1995), Westerling et al. (2001)
and Yu, (1999). This methodology is based on dening a heuristic element size
distribution based on the current effective total incremental strain rate
distribution. Practically, this is implemented by providing information in a
piecewise linearly interpolated manner as illustrated in Table I. The
methodology provides a computationally efcient strategy in that a high
mesh density is produced in regions with high strain rates, with coarse
renements in areas of low strain rate activity.
Whatever approach is adopted for mesh density prediction, the following
steps are common to all. Firstly, mesh regeneration must be undertaken using
any of the standard techniques such as advancing front or Delauney
triangulation (Peraire et al., 1987; Sloan, 1987). In the examples presented in this
work, the advancing front formulation has been employed for performing the
remeshing process. Secondly, it is necessary to transfer the eld variables from
the original nite element mesh to the newly generated mesh. A schematic of
the technique adopted is given in Figures 1-3.
(1) The quadrature point values
h
1
P
n;G
;
h
q
n;G
_ _
are directly projected on to
the corresponding element nodes to provide the nodal point values of
h
1
P
n; N
;
h
q
n; N
_ _
. Typically, the nite element shape functions N may be
used for the projection of the quadrature values back to the
corresponding nodal positions. Upon completion of quadrature point
projection, nodal point averaging is performed which then results in the
nodal point values of
h
1
P*
n; N
;
h
q
*
n; N
_ _
.
(2) Interpolation of the nodal point values
h
1
P*
n;N
;
h
q
*
n;N
_ _
from the
quadrature point values
h
1
P*
n;G
;
h
q
*
n;G
_ _
is then required. This operation
results in the nodal components of the state variables being transferred
from the nodal points on the old nite element mesh h to the nodal points
on the newly generated nite element mesh h+1 resulting in the new
nodal state variables of
h1
1
P*
n;N
;
h1
q
*
n;N
_ _
. The completion of this
process is achieved through three distinct steps, rstly the construction
Parameter effective total incremental strain rate Updated mesh density
0.00E+000 h
k
1.00E+003 h
k+1
1.00E+004 h
k+2
1.00E+005 h
k+3
Table I.
Mesh renement
control data based
on effective total
incremental strain
rate
Numerical
modelling of
ceramics
87
of the nite element background mesh, then dening a local coordinate
system, and nally the transfer of the nodal values from the old nite
element mesh to the nodal positions on the new nite element mesh.
These three key steps are outlined in the following steps.
.
Construction of the nite element background mesh is required for
each nodal point A of the newly generated nite element mesh h+1,
which has the known nodal global coordinates
h1
x
n;A
; the
background nite element is located on the old nite element mesh
h, i.e. nite element
h
V
e
for which
h1
x
n; A
[
h
V
e
is satised.
Figure 2.
Procedural
representation of the key
stages required for the
transfer of eld variables
from the original mesh to
the new mesh, showing
the transfer of the nodal
variables
h
1
P*
n;N
;
h
q
*
n;N
_ _
from the old mesh to the
nodal points on the
newly generated mesh
h1
1
P*
n;N
;
h1
q
*
n;N
_ _
Figure 1.
Procedural
representation of the key
stages required for the
transfer of eld variables
from the original mesh to
the new mesh, showing
the projection of the
quadrature point
variables
h
1
P
n;G
;
h
q
n;G
_ _
to the nodal positions
h
1
P
n;N
;
h
q
n;N
_ _
using the
elemental shape
functions. Then nodal
point averages are then
performed resulting in
h
1
P*
n;N
;
h
q
*
n;N
_ _
EC
20,1
88
.
Evaluation of the local coordinate system for each element present in
the background nite element mesh is required. The local coordinate
system
h
r
A
;
h
s
A
_ _
is required within each background element
h
V
e
that corresponds to the global position of node A in the newly
generated nite element mesh h+1, this is obtained from the
expression
h1
x
n; A

3
b1
h
N
b
h
r
A
;
h
s
A
_ _
h
x
n;b
9
where the elemental shape functions of the element
h
V
e
are
denoted
h
N
b
: For the previous mesh, namely the background mesh
h, which is composed of linear triangular elements, the values of
the local coordinates
h
r
A
;
h
s
A
_ _
for each nodal point A on the new
nite element mesh can be obtained through solution of the
equation
h1
x
n; A

3
b1
h
N
b
h
r
A
;
h
s
A
_ _
h
x
n;b
10
.
Mapping of the nodal point values is the nal and most complex part
of the mapping transfer operation. Through use of the element shape
functions
h
N
b
h
r
A
;
h
s
A
_ _
the state variable point values
h
~
L
n;B

h
1
P*
n; B
;
h
q
*
n; B
_ _
are then directly mapped from the nodes B located on
the old nite element mesh h to the nodal points A on the new nite
element mesh h+1, through use of the mapping function
Figure 3.
Procedural
representation of the key
stages required for the
transfer of eld variables
from the original mesh to
the new mesh, with the
re-interpolation of the
Gauss point variables
back from the nodal
points
Numerical
modelling of
ceramics
89
h1
~
L
n;A

3
b1
h
N
b
r
A
; s
A

h
~
L
n; b
11
(3) The transfer of the new nodal point state variable values to the
quadrature points
h1
1
P*
n;G
;
h1
q
*
n;G
_ _
on the new nite element mesh
h+1 are obtained through further use of the element
h1
V
e
shape
functions N and the quadrature point global coordinates via the function
h1
~
L
n; A

3
a1
h1
N
a
r
G
; s
G

h1
~
L
n; A
12
The mapping operator detailed above ( Peric et al., 1994, 1996, 1999 ) has proven
to be very successful in the transfer of eld variables between nite element
meshes for both small and large deformation cases. However, experience has
found that simulations, which involve signicant changes in geometry, coupled
to adiabatic heat generation, can inherently suffer from excessive amounts of
diffusion of the eld variables such as inelastic strain. The diffusion process
itself is recognisable through several salient features with regard to high and
hyper velocity ballistic impact situations. Firstly, the inelastic state variables
are often propagated or diffused into regions in which inelastic activity should
be minimal. In addition, interpolation-mapping operators smooth the numerical
solution hence the region located at the initial impact interface rarely retains
the high gradients associated with the penetration process. Consequently, the
development of new mesh transfer techniques, specically weighted least
squares mapping operators (Morancay et al., 1997), has received much research
attention.
Discrete fracture insertion
Traditionally, the nite element method has focused on continuum modelling,
however for many applications it is necessary to degrade a single body into a
set of independent physical entities (Camacho and Ortiz, 1996; Klerck, 2000;
Yu, 1999). The numerical capability to degrade a continuum body is paramount
to the successful numerical modelling of brittle systems subject to
deformations arising from both low and high velocity impact situations. The
transition of a brittle body from a continuum description into a discrete
(or multi-body) system is developed from the dispersed micro cracks coalescing
into macroscopic fractures. In an attempt to provide an accurate description of
the material failure processes experienced within quasi-brittle materials, a
number of constitutive model have been developed. Typically these models
have adopted the common softening plasticity and damage theories so often
presented in non-linear nite element analysis.
EC
20,1
90
The failure mechanisms in brittle materials are generally associated with
anisotropic fracture phenomena. The coalescence and development of micro-
cracks within a brittle solid occur in directions that attempt to maximise
the subsequent energy release rate and simultaneously minimise the strain
energy density. The manifestation of a discrete fracture within the material
results in the realisation of localised inelastic strains and the associated
unloading of the surrounding material. In an impact situation, these
macroscopic fractures typically align themselves with the direction of the
maximum principal strain. The subsequent localisation of micro cracking
into effective crack bands then results in the softening process occurring
normal to the crack directions.
The onset of discrete fracture is directly considered from the anisotropic
rotating smeared crack plasticity model presented by Cope et al. (1980). This
model utilises a tensile strain softening formulation to represent the
degradation of strength in the tensile regime. The smeared crack model
provides a mechanism for directional softening within a continuum framework
by envisaging a cracked solid as an equivalent anisotropic continuum with
degraded properties in directions normal to crack band orientation. After initial
yield the rotating crack formulation introduces anisotropic damage by
degrading the elastic modulus in the direction of the current principal stress
invariant. The model enforces coincident rotation of the principal axes of
orthotropy and the principal strain axes. It provides an effective mechanism for
eliminating stress locking and excess shear stress and has been shown to yield
a more reliable lower bound response compared to xed crack models.
However, damage rotation is only realised in a smeared or average sense with
respect to the material response at the macroscopic level.
Although energy dissipation in the crack band model is rendered objective
by normalising the softening curve with respect to the specic fracture
energy, the spatial localisation is necessarily arbitrary. Localisation occurs
in individual elements, resulting in the width of localisation and the crack band
spacing depending on the mesh discretisation. Furthermore, the mesh
orientation gives rise to a directional bias of propagating crack bands due to
the fact that strain discontinuities exist at the element boundaries. Optional
formulations of regularisation techniques, which render the mesh discretisation
objective, include non-local damage models, Cosserat continuum approaches,
gradient constitutive models, viscous regularisation and fracture energy
releasing/strain softening approaches. All models effectively result in the
introduction of a length scale and have specic advantages depending on the
model of fracture and loading rate. The current approach adopts a non-local
averaging of the damage measure in each orthotropic direction to ensure
discretisation objectivity by introducing a length scale to govern the width of
the localisation zone. Acomprehensive overviewof the various techniques used
to obtain mesh objectivity is given in Klerck (2000).
Numerical
modelling of
ceramics
91
Within a numerical simulation environment, the insertion of discrete
fractures into a continuum domain introduces additional degrees of freedom
that are associated with the nodes introduced to dene the fracture geometry.
Furthermore, the introduction of discrete fractures also introduces additional
contact surfaces that must be included in the nite element solution procedure.
The process of inserting a discrete fracture into a continuum based nite
element mesh follows a set of three key steps. Firstly the creation of a non-local
failure map, which is based upon the weighted nodal averages of the failure
parameters within individual elements, is required. Then the failure map is
used to determine the onset of local fracture within the domain. Once the onset
of fracture has been determined, then the nal task of a numerical code is to
perform the topological update whereby a discrete fracture is inserted into the
domain, any additional nodes must be inserted and the necessary elemental
connectivities are updated. The so-called failure fracture indicator is typically
dened as the ratio of the inelastic fracturing strain 1
f
to the critical fracturing
strain 1
f
c
. The elemental or local failure factor F
k
that is associated with the
Gauss point of a considered element k is given by
F
k

1
f
1
f
c
_ _
k
13
where F
k
is also associated with the elemental local fracture direction u
k
which
is dened as being normal to the local failure softening direction. Discrete
fracture is realised through the failure factor reaching a value of unity. Use of a
nodal basis for fracture leads to a simpler and more efcient methodology for
the insertion and creation of discrete fractures. The associated failure factor F

p
and the direction of failure for the considered node p are given by the weighted
expressions,

F
p

ngauss
adj
k1
F
k
w
k

ngauss
adj
k1
w
k
;

u
k

ngauss
adj
k1
u
k
w
k

ngauss
adj
k1
w
k
14
where the summation is taken over the number of element Gauss integration
points that are immediately adjacent, and w
k
is the elemental weighting factor,
normally taken as the elemental area or volume.
When the associated failure factor F

p
and the direction of failure for the
considered node p have been determined, as soon as the limiting value is
reached a discrete fracture in the given orientation will be inserted into the
nite element mesh passing through the associated nodal point. It is not
realistic to make the assumption that just a single nodal point will fail during
any given time step. It is often more likely that a series of nodal points of
EC
20,1
92
adjoining elements will fail simultaneously during the same time increment. It
is for this reason that it is most necessary to provide a systematic method of
inserting a discrete fracture sequence prior to performing the nal topological
update for the considered time step.
As previously stated, the process by which a discrete fracture is inserted into
a nite element mesh is associated with the denition of a failure plane which
passes through a failed nodal point and is orientated in the direction of the
weighted average fracture direction of the adjoining elements, as shown in
Figure 4(a). Having identied that a nodal point is to undergo the fracture
insertion process, it is necessary to identify the actual discrete crack orientation
within the nite element mesh.
Generally, there is one of two choices to be made. Firstly the fracture plane
can be aligned in the exact same orientation as the weighted average nodal
failure direction, thereby following a process known as intra-element fracturing
where a series of new nodal points and elements are systematically created, as
shown in Figure 4(b). Or as is computationally less demanding, the discrete
fracture orientation is aligned to the best orientated element boundary attached
to the node considered, thereby following a process known as inter-element
fracturing where a series of new nodal points are systematically created but no
new elements are created, as shown in Figure 4(c).
The process of inter-element fracturing, provided a ne mesh is utilised, will
often yield satisfactory results for most numerical simulations. The reasoning
behind the preference lies in the ability to effectively maintain control over the
critical time step used in the explicit time integration algorithm.
Post fracture insertion, all adjoining nite elements that have failed are
generally reinitialised assuming that all of the accrued element damage has been
coalesced into the establishment of the discrete crack. However, with the likely
creation of new nite elements, from intra-element fracturing, mapping of all of
the required state variables is very computationally expensive for little
improvement in numerical resolution, hence it is assumed that all failed elements
effectively heal themselves and all state variables are accordingly reinitialised.
Figure 4.
(a) The weighted average
nodal failure direction,
(b) intra-element
fractures description and
(c) inter-element fracture
description
Numerical
modelling of
ceramics
93
Numerical simulations
Numerical simulations, which can demonstrate a coupled adaptive discrete
formulation, are based upon the experimental work presented by Riou et al.
(1998), in which silicon carbide ceramic targets of varying degrees of
connement are subjected to impacts from a steel impactor of low L=D 1:81
ratio. It should be noted that as in the previous experimental work, the current
numerical work also adopts a normal impact technique, as opposed to the
reverse ballistic techniques that are often employed (Westerling et al., 2001).
The impacting penetrator is 20 mm in length and 11 mm in diameter, which
impacts a conned or unconned silicon carbide beam with a velocity ranging
from 100 to 350 m/s. The beam is 100 mm in length, 10 mm in width, and the
thickness T is 15 or 20 mm. A schematic of the impact conguration for an
unconnedandfor aRHAsteel backedsiliconcarbidebeam is giveninFigure5.
The ceramic element of the system is numerically modelled using the
rotating crack tensile softening plasticity model followed by discrete crack
insertion (Cope et al., 1980; Klerck, 2000; Yu, 1999). The corresponding material
data are given in Table II.
Figure 5.
Geometric model
description (Riou et al.,
1998)
Parameter Notation Value used
Initial density (kg/m
3
) r
0
3,150
Elastic bulk modulus (MPa) K 192,000
Elastic shear modulus (MPa) G 175,000
Specic fracture energy (J/m) G
f
24.70
Maximum tensile strength (MPa) f
t
370
Coulomb coefcient of friction m 0.25
Table II.
Material parameters
for silicon carbide
ceramic (Riou et al.,
1998)
EC
20,1
94
In addition, the metallic elements including the RHA impactor and the
connement are numerically modelled using the Johnson-Cook viscoplastic
constitutive model ( Johnson and Cook, 1983). The governing equation for the
yield stress associated with the Johnson-Cook viscoplastic constitutive model is
given as
s A B 1
pn
1 C ln _ 1
p
1 2
T 2T
ref
T
melt
2T
ref
_ _
m
_ _
15
where 1
p
is the accrued inelastic strain, 1
p
is the inelastic strain rate, T is the
material temperature which is accounted for through use of adiabatic heating
with given thermal conversion coefcients, and A, B, C, m, n, T
ref
and T
melt
are
user dened material parameters.
Unlike the simulations presented by Riou et al. (1998), the impactor and
connement are modelled with material non-linearity so as to enable greater
user control over the adaptive remeshing algorithms. The non-linear material
parameters for the RHA steel components are chosen as those referenced from
Westerling et al. (2001), these parameters are provided in Table III.
The domain is initially discretised with 3-node triangular elements. The use
of triangular elements as opposed to the 4 node quadrilateral elements is due to
the fracture insertion technology currently available for this type of element.
The adopted error estimation parameters selected for the adaptive discrete
calculations are given in Table IV.
In the present work the inelastic strain has been used as a measure for
controlling the mesh densities associated with the steel impactor and the steel
connement block. For the ceramic material mesh density prediction is based
on the total incremental strain rate. The data controlling the mesh-density
algorithms are given in Table V, with linear interpolation being used between
the points provided.
Parameter Notation Value used
Density (kg/m
3
) r
0
7,830
Elastic bulk modulus (MPa) K 159,000
Elastic shear modulus (MPa) G 81,800
Static yield limit (MPa) A 792
Strain hardening modulus (MPa) B 510
Strain hardening exponent n 0.26
Strain rate coefcient C 0.014
Thermal softening exponent m 1.03
Initial reference temperature (K) T
ref
300
Specic heat capacity (J/kg/K) C
v
477
Melting temperature (K) T
melt
1,793
Energy conversion coefcient (%) f 90
Table III.
Material parameters
for RHA steel
(Westerling et al.,
2001)
Numerical
modelling of
ceramics
95
In the experimental and numerical works of Riou et al. (1998), four impact test
cases are presented. In three of the test cases, an unconned conguration is
used at impact velocities 203, 250 and 320 m/s, and a single conned sample is
used at an impact velocity of 260 m/s. For the presented simulations, a single
unconned example will be presented with the corresponding impact velocity
of 203 m/s, and the single conned case will additionally be considered.
Figure 6 shows the experimental ndings of Riou et al. (1998) for the
unconned target conguration at an impact velocity of 203 m/s. The rst
Silicon carbide ceramic RHA steel impactor RHA steel connement
Strain rate
Mesh size
(mm) Inelastic strain
Mesh size
(mm) Inelastic strain
Mesh size
(mm)
0.00 5.00 0.00 5.00 0.00 5.00
1.00E+006 0.75 0.50 1.00 0.50 0.75
5.00E+006 0.50 1.00 0.50 1.00 0.50
1.00E+008 0.50 10.00 0.50 10.00 0.50
Table V.
Adaptive mesh
coarsening control
data
Parameter Notation Value used
Error assessment frequency (steps) f 50
Mesh density prediction type
Inelastic strain / Total strain rate
renement
Allowable element distortion error (%) j 2.50
Minimum global bound element size (mm) h
min
0.50
Maximum global bound element size (mm) h
max
5.00
Table IV.
Adaptive error
estimation control
parameters
Figure 6.
Experimental discrete
fracture evolution in an
unconned 20 mm thick
silicon carbide target
impacted at 203 m/s, at
1.00 ms intervals between
pictures ( Riou et al.,
1998)
EC
20,1
96
image is approximately 0.90 ms post impact and indicates that there is minimal
cracking present in the ceramic specimen. However the next image,
corresponding to 1.90 ms post impact, indicates the initial formation of
conical fragmentations. As indicated by Riou et al. (1998), the sample surface
exhibits a cracked zone with a high crack density.
Under the assumption that the crack front proceeds at a constant velocity,
then a crack front velocity corresponding to 8; 200 ^500 m=s can be estimated.
In addition, for the special case of an isolated crack, the experimental ndings
indicate an approximate velocity of 4,000 m/s. Figure 6 presents the
experimental images (Riou et al., 1998).
The estimate of crack front velocity conrms that the damage evolution does
not directly depend on the compressive waves, which move at a longitudinal
wave velocity equal to 11,800 m/s. Figure 6 suggests that much of the
fragmentation is initiated near the distal face of the ceramic, in addition to
those originating at the impact interface, which is caused by the reection of
the compressive waves as tensile components.
Figure 7 presents the simulated fracture patterns attained using the
proposed methodology of explicitly coupling continuum adaptive remeshing
with discrete fracture insertion processes. Upon comparison with Figure 6, it is
clear to see that the proposed technology is capable of capturing all of the
salient features present within the experimental tests. The numerical
simulations correctly identify key features, such as conical type fractures
dominating the fragmentation process. In addition, a degree of spalling type
fracture is also accurately predicted to occur on the distal face of the ceramic as
well as at the impact interface.
In addition to the fracture proles presented in Figure 7, an important
feature is the evolution of the nite element mesh associated with the adaptive
remeshing scheme. The driving motivation behind the proposed technology is
the ability to realise a numerical solution with a high level of accuracy with
efcient computational costs. Figure 8 demonstrates that the evolving nite
element mesh can be effectively governed to provide a high resolution in
regions of high strain rate activity and coarsened elsewhere.
The means of controlling the nite element mesh size using a criterion based
upon the total incremental strain rate term appears to function well for the
fragmenting silicon carbide. In addition when this is combined with the
inelastic strain criterion (Yu, 1999) it provides the numerical modeller with a
exible strategy for the efcient modelling of both metallic and ceramic
systems using coupled continuum adaptive remeshing and discrete fracture
insertion processes.
The methodology clearly demonstrates that the nite element mesh
evolution can be efciently controlled by an approach that is exible enough
to localise around regions of high strain rate. It should be noted that all of the
test congurations presented by Riou et al. (1998) have been numerically
Numerical
modelling of
ceramics
97
simulated in this work using the proposed adaptive discrete methodology and
in all cases the technology has been successful in capturing the prominent
features present in the original experimental works.
The second numerical example considers the experiments performed by
Field (1988), in which spherical pellets of varying materials, including steel,
tungsten carbide and lead, were impacted at low velocity into ceramic and
glass target plates. Field (1988) made detailed examinations of the fracture
patterns experienced within the ceramic plates, which included large conical
fragments and also monitored the failure mechanisms of the impacting pellets.
In the proposed axi-symmetric nite element model of the experimental
system shown in Figure 9, an initial prescribed velocity in the range
300-500 m/s is applied to the hardened steel pellet. In the experiments
considered, a 5 mm diameter hardened steel pellet is impacted against a Sintox
Alumina plate of 8.6 mm in thickness and 50 mm in diameter. The pellets are
Figure 7.
Simulated discrete
fracture evolution in an
unconned 20 mm thick
silicon carbide target
impacted at 203 m/s, at
1.00 ms intervals between
pictures, with (a) 0.90,
(b) 1.90, (c) 2.90, (d) 3.90,
(e) 4.90, (f) 5.90, (g) 6.90
and (h) 7.90 ms
EC
20,1
98
red at the ceramic plates at velocities of 300 m/s and 500 m/s. The xity
conditions applied to the system consist of fully restraining the outer perimeter
of the target plate against any form of translational movement. The material
properties utilised as presented by Field (1988) and Camacho and Ortiz (1996)
are collated in Tables VI and VII.
Unlike the previously presented simulations, the hardened steel pellet is also
modelled using a tensile rotating crack fracture model as opposed to a
continuum based plasticity model, the data for which are given in Table VII.
In a similar fashion to the previous simulations, the initial discretisation
consists of a uniform distribution of 3-node linear triangular elements. The
initial mesh density for the simulation is prescribed as 0.75 mm, however
through the adaptive remeshing and discrete fracture algorithms, elements
ranging from a minimum 0.10 mm to a maximum 1.50 mm will be introduced
into the domain. A summary of the adaptivity control data adopted for the
simulations presented is given in Table VIII.
Figure 8.
Simulated nite element
mesh evolution in an
unconned 20 mm thick
silicon carbide target
impacted at 203 m/s, at
1.00 ms intervals between
pictures, with (a) 0.90,
(b) 1.90, (c) 2.90, (d) 3.90,
(e) 4.90, (f) 5.90, (g) 6.90
and (h) 7.90 ms
Numerical
modelling of
ceramics
99
The total incremental strain rate measure has been used for controlling the
mesh densities associated with both the Sintox Alumina target and the
hardened steel pellet. The data controlling the mesh density prediction
algorithm for the materials in the simulation are given in Table IX, with linear
interpolation being again employed between the quoted values.
As can be seen in Table IX identical mesh prediction data is prescribed for
both the target and pellet components. The use of the total incremental strain
rate measure, as a means for controlling the predicted mesh densities is
preferred over the inelastic strain measure, for the case of a discrete description,
since inelastic straining is typically minimal at the point of fracture in quasi-
brittle materials.
Figure 9.
Model description for
(a) the experimental
geometry (Field, 1988),
and (b) the 2D
axi-symmetric geometry
EC
20,1
100
Figure 10 presents the simulated fracture patterns attained using the
proposed methodology of explicitly coupling continuum adaptive remeshing
with discrete fracture insertion processes. Upon comparison with Camacho and
Ortiz (1996) and Field (1988), it is clear that the developed technology is capable
of capturing all of the prominent features present within the experimental
ndings.
The numerical simulations, presented in Figure 10, correctly identify the key
feature of conical fracturing dominating the fragmentation process.
Furthermore, it is clearly evident that the use of a strain rate mesh
Parameter Notation Value used
Initial density (kg/m
3
) r
0
3,690
Elastic bulk modulus (MPa) K 149,425
Elastic shear modulus (MPa) G 107,438
Specic fracture energy (J/m) G
f
34.00
Maximum tensile strength (MPa) f
t
500
Coulomb coefcient of friction m 0.25
Table VI.
Material parameters
for Sintox Alumina
ceramic (Camacho
and Ortiz, 1996;
Field, 1988 )
Parameter Notation Value used
Initial density (kg/m
3
) r
0
7,800
Elastic bulk modulus (MPa) K 164,000
Elastic shear modulus (MPa) G 78,000
Specic fracture energy (J/m) G
f
4074
Maximum tensile strength (MPa) f
t
400
Coulomb coefcient of friction m 0.25
Table VII.
Material parameters
for hardened steel
pellet (Camacho and
Ortiz, 1996;
Field, 1988)
Parameter Notation Value used
Error assessment frequency (steps) f 50
Mesh density prediction type Total strain rate renement
Allowable element distortion error (%) j 2.50
Minimum global bound element size (mm) h
min
0.10
Maximum global bound element size (mm) h
max
1.50
Table VIII.
Adaptive error
estimation control
parameters
Sintox Alumina ceramic target plate Hardened steel pellet
Incremental strain rate Mesh size (mm) Incremental strain rate Mesh size (mm)
0.00E+000 1.50 0.00E+000 1.50
1.00E+005 0.10 1.00E+005 0.10
1.00E+008 0.10 1.00E+008 0.10
Table IX.
Adaptive mesh
coarsening control
data
Numerical
modelling of
ceramics
101
Figure 10.
Finite element mesh
evolution and simulated
discrete fracture
development in an
unconned 8 mm thick
Sintox Alumina target
impacted at 300 m/s,
with times post impact of
(a) + (b) 0.00, (c) + (d)
1.00, (e) +(f) 2.00, (g) +(h)
3.00 and (i) + (j) 4.00 ms
EC
20,1
102
controlling algorithm enables the efcient localisation of the nite element
mesh around signicant features such as the propagating wave front and
generated crack tip. Upon comparison with Figure 11(a) it is clear that the
proposed technology correctly identies the correct form of Hertzian crack. In
addition, the numerical simulations provide good correlation with the failure
mechanism that is observed in the hardened steel pellet, as depicted in
Figure 11(b).
The numerical results demonstrate that the proposed adaptive discrete
formulation can accurately model fragmentation induced by impact within
brittle solids. Furthermore, techniques that are available enable this to be done
with a high level of computational efciency.
Conclusions
The use of individual discrete element techniques or adaptive techniques
within nite element formulations has been well established in the multi-
fracturing numerical modelling of brittle type materials (Camacho and Ortiz,
1996; Cottrell et al., in press; Klerck, 2000; Yu, 1999). In the present work, it has
been shown that the use of adaptive remeshing and discrete element techniques
can be effectively coupled and employed in the complex numerical modelling of
ceramic armoured systems when subjected to normal low velocity metallic
impacts.
The implemented mesh transfer operators (Peric et al., 1996, 1999), which
perform the transfer of internal state variables between the previous distorted
mesh and the newly generated mesh can, for many large deformation
Figure 11.
Plots obtained from the
experimental report
(Field, 1988)
Note:
These show (a) the impact damage produced in sintox alumina impacted by a hardened steel
pellet (at 225 m/s). The upper portion of the cone corresponds to the propagation of the Hertzian
crack, and the lower rougher cone is from the interaction with the reected tensile wave, and
(b) a schematic diagramof the failure observed in a hardened steel pellet after impact on alumina.
An inverted cone, n, is formed in the sphere together with fracture planes, r. The arrows indicate
the radial movement of material observed in such type of impacts.
Numerical
modelling of
ceramics
103
applications, suffer from numerical diffusion of eld variables. In an attempt to
resolve this deciency, a weighted least squares mapping algorithm has been
proposed (Morancay et al., 1997) that is superior to the standard nodal
interpolation operators. The use of this type of transfer operator should provide
signicantly reduced diffusion of variables and subsequently provide more
pronounced gradients within the numerical solution, such as those that would
be of importance for hypervelocity depth of penetration tests.
At present, appropriate error measures for use in mesh size prediction are
dependent on the class of problem involved. For cases in which the strain elds
are predominantly elastic and continuous the standard elastic energy norm
indicator can be used to good effect. However, as it has been shown, for
situations that involve strain localisation phenomena, such as multi-fracturing
quasi-brittle solids, strain rate criteria can be used to develop appropriate
adaptive mesh predictions. On the other hand, error estimates for problems
involving large-scale plastic ows, such as forming processes or penetration of
ductile materials, are more appropriately based on either incremental or total
plastic work indicators. The ultimate goal, naturally, is the development of a
mesh size predictor that is applicable to problems in which all or some of these
features may be present. However, such a theoretical framework will require
long-term numerical research.
A potential benet of using such a exible technology is that for many
advanced constitutive models and problem geometries a numerical model
requires a suitably ne nite element mesh in local regions in order to resolve a
solution of desired accuracy. Therefore, the technology presented here offers a
means of localising the nite element mesh density into regions of high strain
or strain rate and providing fewer elements in low activity regions.
An important issue that arises from the present works is the ability to
correctly couple the nite element mesh size with the fracture insertion process.
Currently the length of the fracture is dictated by the size of the adjoining nite
element. However, it is more preferable to be able to dene the nite element
mesh size from the size of fracture that should be inserted into the system.
Local mesh renement is useful for resolving crack tip elds, and will
undoubtedly aid the prevention of premature crack arrest (Camacho and Ortiz,
1996). Furthermore increased mesh density at the crack tip enables the number
of possible crack orientations to increase. Therefore it is essential that
computational strategies be developed, which can be used to determine the size
and location of a aw completely independently of the surrounding nite
element mesh. It is also note worthy that remeshing is a means of departing
from directional bias associated with a xed nite element mesh employed in
the case of inter-element fracturing.
Naturally, the presented simulations represent only the initial steps in
providing a robust and computationally efcient tool for fully coupled discrete
EC
20,1
104
adaptive technology. Current work involves extension into 3D, which provides
many challenges in relation to the mesh regeneration process.
References
Camacho, G.T. and Ortiz, M.L. (1996), Computational modelling of impact damage in brittle
materials, International Journal of Solids Structures, Vol. 33 No. 20-22, pp. 2899-938.
Camacho, G.T. and Ortiz, M.L. (1997), Adaptive lagrangian modelling of ballistic penetration of
metallic targets, Computer Methods Applied Mechanics And Engineering, Vol. 142,
pp. 269-301.
Cope, R.J., Rao, P.V., Clark, L.A. and Norris, P. (1980), Modelling of reinforced concrete
behaviour for nite element analysis of bridge slabs, Numerical Methods for Nonlinear
Problems, Pineridge Press, Swansea, pp. 457-70.
Cottrell, M.G., Yu, J. and Owen, D.R.J. (in press), The adaptive and erosive numerical modelling
of boron carbide subjected to large-scale dynamic loadings with element conversion to
undeformable meshless particles, International Journal of Impact Engineering.
Dyduch, M., Habraken, A.M. and Cescotto, S. (1992), Automatic adaptive remeshing for
numerical simulations of metal forming, Computer Methods in Applied Mechanics and
Engineering, Vol. 101, pp. 238-98.
Field, J.E. (1988), Investigation of the impact response of various glass and ceramic systems,
USARDSG-UK, ARO-E, DAJA45-85-C-0021, University of Cambridge.
Johnson, G.R. and Cook, W.H. (1983), Aconstitutive model and data for metals subjected to large
strains, high strain rates and high temperatures, Proceedings of the 7th International
Symposium on Ballistics, The Hague, The Netherlands.
Klerck, P.A. (2000), The nite element modelling of discrete fracture in quasi-brittle materials,
PhD thesis, University of Wales Swansea, Swansea.
Morancay, L., Homsi, H. and Roelandt, J.M. (1997), Application of remeshing techniques to the
simulation of metal cutting by punching, in Owen, D.R.J., Onate, E. and Hinton, E. (Eds),
CIMNE, Barcelona.
Peric, D., Vaz, M. Jnr and Owen, D.R.J. (1999), On adaptive strategies for large deformations of
elasto-plastic solids at nite strains: computational issues and industrial applications,
Computer Methods in Applied Mechanics and Engineering, Vol. 176, pp. 279-312.
Peric, D., Yu, J. and Owen, D.R.J. (1994), On error estimates and adaptivity in elasto-plastic
solids: applications to the numerical simulation of strain localisation in classical and
consserat continua, International Journal of Numerical Methods in Engineering, Vol. 37,
pp. 1351-79.
Peric, D., Hochard, C.H., Dutko, M. and Owen, D.R.J. (1996), Transfer operators for evolving
meshes in small strain elasto-plasticity, Computer Methods in Applied Mechanics and
Engineering, Vol. 137, pp. 331-44.
Peraire, J., Vahdati, M., Morgan, K. and Zienkiewicz, O.C. (1987), Adaptive remeshing for
compressible ow computations, Journal of Computational Physics, Vol. 72, pp. 449-66.
Riou, P., Denoul, C. and Cottenot, C.E. (1998), Visualisation of the damage evolution in impacted
silicon carbide ceramics, International Journal of Impact Engineering, Vol. 21 No. 4,
pp. 225-35.
Sloan, S.W. (1987), A fast algorithm for constructing Delauney triangulations in plane,
Advances in Engineering Software, Vol. 9, pp. 34-55.
Strassburger, E. and Senf, H. (1995), Experimental investigations of wave and fracture
phenomena in impacted ceramics and glasses, Report ARL-CR-214.
Numerical
modelling of
ceramics
105
Westerling, L., Lundberg, P. and Lundberg, B. (2001), Tungsten long-rod penetration into
conned cylinders of boron carbide at and above ordnance velocities, International
Journal of Impact Engineering, Vol. 25, pp. 703-14.
Yu, J. (1999), A contact interaction framework for numerical simulation of multi-body problems
and aspects of damage and fracture for brittle materials, PhD thesis, University of Wales
Swansea, Swansea.
Zienkiewicz, O.C. and Zhu, J.Z. (1987), A simple error estimator and adaptive procedure for
practical engineering analysis, International Journal for Numerical Methods Engineering,
Vol. 24, pp. 337-57.
Zienkiewicz, O.C. and Zhu, J.Z. (1991), Super convergent derivative techniques and a posteriori
error estimation in the nite element method, Part I A general super convergent recovery
technique; Part II The Zienkiewicz Zhu error estimator, report of INME, University of
Wales Swansea, CR/671/91 and CR/672/91.
EC
20,1
106