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

net/publication/308304413

A non-locking composite tetrahedron element for the combined finite discrete

element method

Article  in  Engineering Computations · January 2016

DOI: 10.1108/EC-09-2015-0268

CITATIONS READS

4 104

6 authors, including:

Zhou Lei Esteban Rougier

Los Alamos National Laboratory Los Alamos National Laboratory
32 PUBLICATIONS   410 CITATIONS    83 PUBLICATIONS   604 CITATIONS   

SEE PROFILE SEE PROFILE

Luke Frash Bill Carey

Los Alamos National Laboratory Los Alamos National Laboratory
23 PUBLICATIONS   56 CITATIONS    179 PUBLICATIONS   3,185 CITATIONS   

SEE PROFILE SEE PROFILE

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

Wellbore Integrity in CO2 Sequestration and Oil and Gas Applications View project

Development of numerical algorithms for the impact fracture simulations of laminated glass View project

All content following this page was uploaded by Zhou Lei on 05 July 2017.

The user has requested enhancement of the downloaded file.

 Citation: Zhou Lei, Esteban Rougier, Earl E. Knight, Luke Frash, James William Carey, Hari Viswanathan, (2016)
"A non-locking composite tetrahedron element for the combined finite discrete element method", Engineering
Computations, Vol. 33 Issue: 7, pp.1929-1956, https://doi.org/10.1108/EC-09-2015-0268

A NON-LOCKING COMPOSITE TETRAHEDRON ELEMENT FOR THE

COMBINED FINITE DISCRETE ELEMENT METHOD

Zhou Lei*, Esteban Rougier, Earl E. Knight, Luke Frash, William Carey and
Hari Viswanathan

Los Alamos National Laboratory

*
Corresponding Author: Dr. Zhou Lei, zlei@lanl.gov

ABSTRACT
Purpose – In order to avoid the problem of volumetric locking often encountered when using constant
strain tetrahedral finite elements, a new composite tetrahedron element which is especially designed
for the combined finite-discrete element method (FDEM) is presented.
Design/methodology/approach – A ten-noded composite tetrahedral finite element, composed of
eight four-noded low order tetrahedrons, has been implemented based on Munjiza’s multiplicative
decomposition approach. This approach naturally decomposes deformation into translation, rotation,
plastic stretches, elastic stretches, volumetric stretches, shear stretches, etc. The problem of volumetric
locking is avoided via a selective integration approach that allows for different constitutive
components to be evaluated at different integration points.
Findings – A number of validation cases considering different loading and boundary conditions and
different materials for the proposed element are presented. A practical application of the use of the
composite tetrahedral finite element is presented by quantitative comparisons of numerical model
results against simple theoretical estimates and results from acrylic fracturing experiments. All of
these examples clearly show the capability of the composite element in eliminating volumetric
locking.
Originality/value – For this tetrahedral element, the combination of “composite” and “low order sub-
element” properties are good choices for FDEM dynamic fracture propagation simulations: 1) in order
to eliminate the volumetric locking, only the information from the sub-elements of the composite
element are needed which is especially convenient for cases where re-meshing is necessary, and 2)
the low order sub-elements will enable robust contact interaction algorithms, which maintains both
relatively high computational efficiency and accuracy.
Keywords: volumetric locking, incompressible elasticity, multiplicative decomposition, composite
tetrahedron element, combined finite-discrete method.
Paper type: Research paper 

1. INTRODUCTION
Since its inception the combined finite–discrete element method (FDEM) (Munjiza, 1992; Munjiza et
al., 1995; Munjiza and Andrews, 1998; Munjiza et al., 1998; Munjiza and Andrews, 2000; Munjiza et
al., 2006; Munjiza, 2004; Munjiza et al., 2012) has become a tool of choice for a diverse field of
practical engineering and scientific simulations (Lei, 2011; Munjiza et al., 2013; Xu et al., 2013; Ji et
al., 2012; Latham et al., 2008; Rougier et al., 2014; Rougier et al., 2011; Carey et al., 2015; Lisjak et
al., 2014; Trivino and Mohanty, 2015). FDEM merges a finite element based analysis of continua with
discrete element based transient dynamics, contact detection, and contact interaction solutions. In

1
FDEM the solid domains (called discrete elements) are discretized into finite elements, where finite
rotations and finite displacements are assumed a priori. Through failure, fracture and fragmentation,
single domains represented by separate finite element meshes are transformed into a number of
interacting domains. The finite element discretization of solid domains is also conveniently used to
resolve the contact between discrete elements (Munjiza, 1992; Munjiza et al., 1995; Munjiza and
Andrews, 1998; Munjiza et al., 1998; Munjiza and Andrews, 2000; Munjiza et al., 2006; Munjiza,
2004; Munjiza et al., 2012).
Within this framework the domain discretization in 3D has usually been conducted by
implementing the constant strain tetrahedral (CSTet) finite element. This four-noded tetrahedral
element has some very advantageous features in that, with these type of elements, the mesh can be
generated easily and even automatically for complicated geometries. Moreover, the contact interaction
between CSTets is relatively easy to resolve while maintaining a high computational efficiency – this
is very important in large scale FDEM simulations where a large number (ranging from a few thousand
to several millions) of separate interacting solids domains, each one with its own finite element
discretization, are usually present. Although in many cases the CSTet finite element can produce good
results, it has been recognized that “volumetric locking” problems associated with the CSTet
formulation (in situations where the material behaves as an incompressible or near-incompressible
medium) can seriously degrade the accuracy of the results.
Locking is a situation wherein two or more physical mechanisms compete for the available degrees
of freedom within each finite element especially in incompressibility conditions (Munjiza et al., 2015).
Several approaches have been proposed to resolve this problem, such as the mixed finite element
methods (Malkus, 1978), reduced and selective integration techniques (Zienkiewicz, 1971; Hughes,
1980), B-bar method (Moran et al., 1990), F-bar method (de Souza Neto et al., 1996) and the smoothed
finite element methods (Liu and Trung, 2010). Zienkiewicz et al.’s (Zienkiewicz et al., 1998)
tetrahedron finite element approach was based on the operator splitting approach, which separated the
volumetric and deviatoric parts of the deformation. Bonet and Burton (Bonet and Burton, 1998)
proposed a linear tetrahedron based on an average nodal pressure algorithm in which nodal pressures
are evaluated based on the ratio between current and initial nodal volumes which are calculated taking
into account the finite elements surrounding the nodes. Similarly, Dohrmann et al. (Dohrmann et al.,
2000) introduced the node-based uniform strain triangle/tetrahedron elements for small strain
elasticity where the nodal averaging process was applied to the strain tensor. By extending some of
the ideas proposed by Dohrmann et al., Bonet et al. (Bonet et al., 2001) developed a linear tetrahedral
element which is based on average nodal deformation gradient. Following this line, Joldes et al.
(Joldes et al., 2009) proposed a method to improve the solution for multiple material interfaces in
which the deformation gradient is modified through a nodal averaged pressure. De Micheli and
Mocellin (De Micheli and Mocellin, 2009) presented a linear tetrahedral element in which the
volumetric term of strain tensor is replaced by a nodal averaged volume ratio which is calculated in
an incremental form. Taylor (Taylor, 2000) presented a formulation based on a mixed-enhanced
treatment involving displacement, pressure and volume effects. The displacement and pressure are
used as nodal quantities while volume effects and enhanced modes belong to individual elements. de
Souza Neto et al. (de Souza Neto et al., 2005) proposed an F-bar based tetrahedral for finite strain
analysis. In their method, they used a modified deformation gradient whose volumetric component
was defined as the volume change ratio of a pre-defined patch of elements. Following the seminal
work done by Camacho and Ortiz (Camacho and Ortiz, 1996), Thoutireddy et al. (Thoutireddy et al.,
2002) developed a composite tetrahedron that was composed of twelve four-noded linear sub-
tetrahedrons in each of which the material deformation was described using a linear field. The
integration over the domain of the element was done using a five point quadrature rule.

2
 These noteworthy research efforts have shown that by properly relaxing the constraints on the
volumetric term of the material deformation a reduction or even elimination of the volumetric locking
problem can be achieved. Recently, Munjiza et al. (Munjiza et al., 2015) proposed a multiplicative
decomposition approach which naturally decomposes deformation into translation, rotation, plastic
stretches, elastic stretches, volumetric stretches, shear stretches, etc. In essence, the total deformation
description is obtained from the displacement field via a composition of the respective deformation
functions; which, when derivation is applied, results in multiplication - thus the term multiplicative
decomposition. This approach opens doors for a selective integration scheme that allows for a
relaxation of the volumetric terms, since the volumetric stretches are calculated separately from the
rest of the stretches.
In this paper, a ten-noded composite tetrahedral (COMPTet) finite element is constructed based
on Munjiza’s multiplicative decomposition approach (Munjiza et al., 2015). For this tetrahedral
element, the combination of “composite” and “low order sub-element” properties are good choices
for FDEM simulations, especially for cases where re-meshing is necessary and also when robust
contact interaction algorithms are needed for dynamic fracture propagation simulations. First, the
computational implementation, as well as the formulations of the proposed finite element, are
described. Second, the convergence of the numerical solutions with a decreasing element size for the
proposed finite element is demonstrated for a number of different cases. Finally, the proposed
composite element formulation is used to analyse the stress distribution for a specific engineering
problem.

2. ELEMENT DEFINITION
Geometry. The geometry of the ten-noded COMPTet and its solid-embedded material axes are defined
as shown in Figure 1-a. Each COMPTet contains eight four-noded sub-tetrahedrons, as shown in
Figure 1-b. A Gauss point is placed at the center of each sub-tetrahedron, where a generalized material
element is defined. This generalized material element is used for the calculation of the deformation
field, from which the corresponding stretches are calculated.


8 9 10
4 9 10
8 9
9

7 7 6
10 7
5 5 7 6
8  4 8 9 10
9
3

10 7 6 3
7 6 8
 9
1
5
5
2 7 6
1
5 2
a) b)
Figure 1. a) Geometry of the ten-noded COMPTet; b) Decomposition of the ten-noded COMPTet
into sub-tetrahedrons.

Generalized Material element. An infinitesimal generalized material element, defined by the

α β γ  material base (see Figure 2) is placed at the center of each sub-tetrahedron.
3
 a) b)

 

 F-1

Figure 2. a) Initial position of the material axis; b) Current position of the material element. The
connection between the initial position and current position of the material element is given by the
deformation gradient F.

The initial position of the material axis,

 i i i 
α β 

γ   j j  j
 (1)
 k k  k 

is provided by the user as an input. The geometry of the material base at any stage during the
simulation is calculated from deformation kinematics (Munjiza et al., 2015), which will be explained
in the following section.

3. DEFORMATION KINEMATICS
Shape functions. For each sub-tetrahedron the following shape functions are employed

N1  , ,    1      ; N 2  , ,     ; N 3  , ,     ; N 4  , ,     (2)

where  ,  and  are the solid-embedded coordinates for each sub-tetrahedron.

Kinematics of a given sub-tetrahedron. For any point inside a sub-tetrahedron, the initial and
current global coordinates can be obtained by interpolating from the global nodal coordinates with the
help of the shape functions:

 N  , ,  
 x   x1 x2 x3 x4   1
 y   y N  , ,  
y2 y3 y 4   2 (3)
   1  N  , ,  
 z   z1 z2 z3 z 4   3 
 N 4  , ,  

4
  N  , ,  
~ x  ~ x1 ~
x2 ~
x3 ~
x4   1
~   N  , ,  
~
 y    y1
~
y2 ~
y3 y 4   2
~ (4)
 N  , ,  
 ~
z   ~
z1 ~
z ~
z ~
z 4   3 
 N 4  , ,  
2 3

where  xi , yi , zi  and ~
xi , ~ z i  are the initial and the current coordinates of node i. In equation (3)
yi , ~
and (4) the “¯” and the “~” on top of the variables indicates that the quantities are based on the initial
and on the current coordinates of the system respectively. The derivatives of the initial and current
global coordinates with respect to the solid-embedded coordinates are obtained by (Munjiza et al.,
2015)

 x x x 
      x  x 
 
2 1 x3  x1  x 4  x1 
 y y y  
  y 2  y1   y 3  y1   y 4  y1  (5)
    
 z z z    z 2  z1  z 3  z1  z 4  z1 
 
    

 ~x ~
x ~
x
     ~
 ~  x2  ~x1  ~x3  ~x1  ~x 4  ~x1 
 y ~
y ~
y  ~ ~
  y 2  y1   ~y 3  ~y1   ~y 4  ~y1  (6)
      ~ ~
 ~z ~
z z    z 2  z1 
~ ~z 3  ~z1  ~z 4  ~z1 
 
    
Kinematics of the material element. For a given material point in a sub-tetrahedron, the initial and
current material-embedded vector bases can be expressed in terms of the solid-embedded vector base
as follows (Munjiza et al., 2015):

 x x x 
   
 i i  i    
   
y y y  
α β 

γ   j j

 j  
     
    
 (7)
 k k  k   z     
 z z   
 
    

5
  ~
x ~
x ~
x
~      
~i i ~
i   ~   
~ ~



α
~ ~ ~
~ β 
γ   j
~
j
~

~ j   

y

y

y 
  
    
 (8)
~k k ~k   ~z    
 ~
z z   
~ 
 
    
The matrix

 i i i 
 
 j j  j (9)
 k k  k 

defines the initial orientation of the material axes of the solid. In the current implementation the
 
material base α β γ is specified for each finite element, thus allowing the orientation of the
material axes to change over the domain on an element by element basis. At the beginning of the finite
element analysis, equation (7) is used to calculate solid-embedded coordinates (Munjiza et al., 2015):

1
 x x x 
  
          i i i 
  y y y   
       j j  j (10)
    
       z  k k  k 
 z z  
 
    
By substituting equation (10) into equation (8), the geometry of the current material embedded base
is obtained, and it is used to calculate the deformation of the material point, as explained in the next
section.

4. CONSTITUTIVE LAW
The ten-noded COMPTet was implemented in conjunction with the unified constitutive approach
proposed by Munjiza et al. (Munjiza et al., 2015). In order to calculate the stresses the deformation
field is decomposed into three different parts: volumetric, linear and angular stretches. The procedure
to obtain these stretches are described in the following paragraphs.
The current geometry of the material axes is given by (Munjiza et al., 2015)
 ~x ~ x ~ x
 x y z 
~  ~ ~ ~   i  i  i 

α
~
~ β ~ 
  x 
γ  α β γ   
y y
 y
x y z  
    
j j j (11)
 x   
 ~z ~ z ~ z   k  k  k 
 
 x y z 
The volumetric stretch is obtained as follows (Munjiza et al., 2015)

6
 v~
sv  (12)
v
where v and v~ are the initial and current volumes of the generalized material element, which are
given by
~
 i  i  i  ~i  i ~i 
   ~ 
v  det  j  j  j  ; v~  det ~ j  j ~ j  (13)
 k  k  k  ~k ~k ~k 
   
In order to avoid volumetric locking, the volumetric stretch for all Gauss points contained in a
given COMPTet are calculated from the average of the volumes as follows
8

 v~ i
sv  i 1
8 (14)
 vi
i 1

where vi and v~i are the initial and current volume of the material element at Gauss point i.
The linear stretches of the edges of the material element are defined by (Munjiza et al., 2015)
~
~  ~
s  ; s  ; s  (15)
  
where the initial and the current lengths of the edges of the material element are given by (Munjiza et
al., 2015)
 α ;  β ;  γ (16)
~ ; ~  ~
~  α β ; ~  ~
γ (17)

The angular stretches are defined by (Munjiza et al., 2015)

~ ~ ~
s    ; s    ; s    (18)
     
where the initial and the current angles of the material element are given by (Munjiza et al., 2015)
 αβ   β γ  γ α 
   arcos  ;    arcos  ;    arcos 
 αβ   β   α 
~
~β ~ (19)
α   β ~γ ~ ~
γ α
~  arcos  ~ ~  ;    arcos  ~~  ; ~  arcos  ~~ 
  ~ 
 αβ   β   α 
The logarithmic strains are obtained from the generalized stretches as follows (Munjiza et al.,
2015),
ev  ln sv ; e  ln s ; e  ln s ; e  ln s
(20)
e   ln s  ; e   ln s  ; e   ln s 

7
 The components of the Munjiza stress tensor matrix are given by (Munjiza et al., 2015)
mv  M v ev ; m  M α e ; m  M β e ; m  M  e
(21)
m   M ψ e  ; m   M ψ  e  ; m   M  e 

where M v , M α , M β , M ψ , M ψ  and M  are the Munjiza elastic constants. The Cauchy stress
tensor can then be calculated from the Munjiza stress tensor (Munjiza et al., 2015).
In the current approach, stretches generated due to deviatoric deformation are calculated at the
eight integration points for each sub-tetrahedron while stretches generated due to volumetric
deformation are calculated for the whole composite tetrahedron (which has some conceptual
similarities with the F-bar-Patch method proposed by de Souza Neto et al. (de Souza Neto et al., 2005)
and the composite element proposed by Thoutireddy et al. (Thoutireddy et al., 2002)).

5. NUMERICAL TESTS FOR CONVERGENCE

In this section three different test cases are presented. In each test case the numerical solution is
compared against the analytical solution and also convergence with a decreasing element size is
demonstrated. It is worth noting that all of the numerical results presented in this paper correspond to
the static solution of the problems. In order to obtain these results the proper damping was included
in the numerical model.
Case 1: Compressive load on opposite surface in both x and y directions. In this case a
compressive load p  1000 Pa is applied on the opposite surfaces of a 1.0 m cube in both x and y
directions as shown in Figure 3. The analytical expressions for the displacements at point A are given
by:

1 ν 1 ν 2ν (22)
x   p ; y   p ; z  p
E E E
This calculated displacement is compared against the analytical displacement and the error in
displacement is obtained:

 (23)
e 


where   x 2  y 2  z 2 and   x 2  y 2  z 2 are the analytical and the numerical solutions

respectively.
Four element sizes were used in conjunction with three different Poisson’s ratios. The evolutions
of the error in the displacement as a function of the element size for different values of Poisson ratios
are shown in Figure 4. As can be observed from the figure, the error e nicely converges to zero as
the element size is decreased, the convergence for the case with ν  0.499 being the fastest one.

8
 p=1000Pa

p=1000Pa
A

p=1000Pa
y

x
z
p=1000Pa
Figure 3. Model setup: a 1m cube loaded with compressive load on four surfaces.

Figure 4. Relative error in displacement ( e ) as a function of element size – quadratic convergence

with element size is observed.

Case 2: Stress due to quadratic displacement field. In this case, the following quadratic
displacement field is supplied to the 1m cube shown in Figure 5,

~
x  x  kx x 2
~
yy (24)
~
z z
where ~x , ~y , ~z  and  x , y , z  are the current and the initial coordinates of the nodes, and kx is a
constant parameter which is set to 1e-6.

9
 A

y 1m

x
1m
z
1m
Figure 5. Model setup.

The analytical expression for the small strain formulation corresponding to the displacement field
shown in equation (24) is given by

 x  2k x x 
    0  (25)
 y  
 z   0 

where  x ,  y , and  z are the strains in the x, y, and z directions respectively. These strains produce
the following stresses

 x  (1  ν) ν ν   x 
   E  ν
 y  1  ν  1  2ν   (1  ν) ν   y  (26)
 z   ν ν (1  ν)   z 

where E and ν are the Young’s module and the Poisson’s ratio respectively.
The stresses obtained from the numerical calculations are compared against the analytical stresses
and for each Cartesian direction the error in stress is obtained by

  (27)
e 

where  and  are the values of the stress obtained analytically and numerically.
The material used in this case had a Young’s modulus E  1 . 0 e 9 and a Poisson’s ratio ν  0 . 0 .
Different mesh sizes were used in order to study the convergence rate of the ten-noded COMPTet
formulation. The four coarsest meshes used in this analysis are shown in Figure 6 while the results
obtained are shown in Figure 7, from where it can be concluded that for the quadratic displacement
field, the convergence rate in stress is linear, with respect to the element size.

10
 a) b) c) d)

Figure 6. Different meshes used: a) 1.0 m, b) 0.5 m, c) 0.25 m and d) 0.125 m.

Figure 7. Relative error in stress ( e ) as a function of element size – Linear convergence with
element size is observed.

6. NUMERICAL TESTS FOR VOLUMETRIC LOCKING

In this section the behaviour of the composite element COMPTet in nearly incompressible conditions
is demonstrated. Comparisons between the results obtained with COMPTet and with CSTet are also
presented.
A 1.0 m cube with its back-, left- and bottom-faces fixed (shaded faces shown in Figure 8) is one
of the cases used to test the volumetric locking behaviour of the composite element formulation. A
force

 fx  1
f   f y   1.0e6 1 N
  (28)
 f z  1

is applied at point A of a cube, as shown in Figure 8. The obtained results for the stress map are shown
in Figure 9 and in Figure 10. The stress distribution obtained for the CSTet (Figure 9-b and -d) shows
clear signs of volumetric locking behaviour, evidenced by a quite marked checkerboard pattern and
large values of stresses near the point of application of the force. The COMPTet does not suffer from
these ailments, as is shown in Figure 9-a and -c, i.e., the stress distribution is smooth without

11
checkerboard patterns and the values of the stresses are not excessively large close to the point where
the force is applied.

f
y
A
x
z

Figure 8. Model setup: 1m cube loaded with a concentrated load at point A.

σxx / Pa σxx / Pa

a) b)
p / Pa p / Pa

c) d)
Figure 9. Stress distribution: a) xx for the composite element, b) xx for the constant strain element,
c) Pressure for the composite element and d) Pressure for the constant strain element (ν=0.499).

Other evidence that the derived composite element formulation does not exhibit volumetric
locking problems is given by Figure 10, where the magnitude of the displacements is plotted on top
of the un-deformed mesh for both cases. Due to the volumetric locking effects, a very small
displacement is observed for the case of the CSTet (in the order of 0.05 m), while for the case of the
COMPTet a much larger displacement is obtained (in the order of 0.18 m).

12
a) b)
Figure 10. Displacement distribution on undeformed mesh: a) composite element and b) constant
strain element (ν=0.499).

The second example that was utilized to test the volumetric locking behaviour is shown in Figure
11, where an infinite plate with a circular hole of radius a  0.1 m in the middle is subjected to tensile
stresses at its ends in plane strain.

z L
y
r
a θ L
x
p p
FEM Domain

Figure 11. Infinite plate with a circular hole.

The analytical solution for the radial and tangential displacements is given by (Timoshenko and
Goodier, 1970)

p1    1  a2  a4 4a 2  
r   r   r  3   cos 2 
2 E 1   r  r 1  r   (29)
p1     a 4 21  a 2 
    r  3   sin 2
2E  r 1  r 
where    /(1   ) . The displacements in the global coordinate system are given by:

13
   x  cos   sin     r 
    (30)
 y   sin  cos     

Due to the symmetry of the problem, only one quadrant was simulated. A finite analysis domain
in x-y plane of size L=1.0 m=10a is taken from the infinite plate (see Figure 11). The exact analytical
displacements calculated from equation (29) are applied to the nodes on the remote sides of the
analysis domain. The nodes of the finite element mesh that are on the other sides of the plate are
allowed to move only in the x-z plane, while the nodes that are on the top and bottom faces of the
plate are allowed to move only in the x-y plane. The material properties used for this case are Young’s
modulus E  30 MPa and Poisson’s ratio ν  0.25, 0.4999 . This latter value of the Poisson’s ratio
(  0 . 4999 ) tests the performance of the proposed COMPTet in a near-incompressible conditions.
Four different mesh sizes (see Figure 12) were used in order to verify the convergence of the results
with the decreasing element size.

a) b)

c) d)

Figure 12. Finite element model: a) mesh size h/L = 1/12, b) mesh size h/L = 1/24,
c) mesh size h/L = 1/48, d) mesh size h/L = 1/96.

The relative error (based on the displacement field) of the numerical results when compared
against the analytical ones is calculated as follows

 (31)
e


where    x 2  y 2  z 2 d and    x 2  y 2  z 2 d are the sum of all the displacements

over the domain for the analytical and the numerical solutions respectively. The evolution of the error
with the element size is shown in Figure 13. Again, this composite element formulation clearly shows
its capability in eliminating volumetric locking.

14
Figure 13. Relative error in Norm-displacement as a function of element size – Linear convergence
with element size is observed.

7. PRACTICAL APPLICATION OF THE COMPTet

Early analysis results for a Los Alamos National Laboratory directed research program experiment
(Frash et al., 2015) utilized a FDEM CSTet element to estimate the stress state at the bottom of a notch
manufactured inside of an acrylic cylindrical sample. The low order element’s limitations were
evident, as it is demonstrated by the stress checkerboard pattern observed in Figure 14. These
limitations were due to volumetric locking problems inherent to the CSTet formulation for the
particular conditions being tested. As is demonstrated in the rest of this section, the composite
tetrahedral element (COMPTet) simulation results show remarkable improvement on the stress field
and on the convergence with decreasing element size over the CSTet results. The numerical results
obtained with the COMPTet are also compared against experimental results.

z z
y y
x x
a) b)
Figure 14. Checkerboard problems encountered with the constant strain tetrahedron: a) map of  xx .
b) map of  yy .

The geometry of the notched samples under consideration is shown in Figure 15. Two different
models were taken into account: Model 1 contains a smaller notch with a radius of 5.0 mm, a depth of
3.81 mm and aperture of 0.15 mm, while Model 2 contains a larger notch which radius, depth and
aperture are 11.0 mm, 9.4 mm and 0.17 mm respectively (Figure 15).
15
 These notched samples contain a novel geometry that was utilized in hydrofracture experiments
under triaxial stress conditions. As such, the samples were subjected to three different stress fields (or
loading cases) (Frash et al., 2015): 1) an axial confining stress ( p1 ), 2) a lateral confining stress ( p2 )
and 3) a notch opening stress ( p3 ).

R δ

A
H

Figure 15. Geometry of the sample. D and H are the diameter and the height of the sample, R, d and
 are the radius, the depth and the aperture of the notch respectively. D  H  25.4 mm . Model 1:
R  5.0 mm , d  3.81 mm ,   0.15 mm ; Model 2: R  11.0 mm , d  9.4 mm ,   0.17 mm .

p1
p2 p3

Case 1 Case 2 Case 3

Figure 16. The general loading of the sample can be obtained as the sum of individual loading cases:
Case 1: axial load, Case 2: confining pressure load, Case 3: pressure in the notch load.

For the purpose of the estimation of the stress state at the bottom of the notch (point A in Figure
15) as a function of the stress fields, it is assumed that the material behaves as a linearly elastic
medium. Therefore, the stress state at point A can be determined by superimposing the stress states
due to the three stress fields described before. Because of this, the three different loading cases were
considered: Case 1: axial load with a pressure p1 , Case 2: confining load with a pressure p2 , and Case
3: notch opening load with a pressure p3 (see Figure 16). In this way, the stresses at point A are given
by,

16
  xxA  c1xx p1  c2 xx p2  c3 xx p3
(32)
 yyA  c1 yy p1  c2 yy p2  c3 yy p3

where c1xx , c 2 xx , and c3 xx are the contributions of the stress fields p1 , p2 , and p3 to the stress  xxA at
point A. In a similar way, c1 yy , c2 yy , and c3 yy are the contributions of the stress fields p1 , p2 , and p3
to the stress  yy
A
at point A. Since the material is assumed elastic, the loads for each one of the stress
fields can be set to unity and the coefficients shown in equation (32) can be determined, which allows
for the estimation of the stress state at point A for any general combination of p1 , p2 , and p3 . The
material of the sample is acrylic with the following properties: Young’s modulus E  3.3 GPa ,
Poisson’s ratio ν  0 . 38 and density ρ  1200 kg m 3 .
Model 1: The dimensions for this model are shown in Figure 15. In order to verify the convergence of
the stress values at the bottom of the notch with decreasing element sizes three different mesh sizes
(see Figure 17) were tested.
a) b) c)

z z z

y x y x y x

Figure 17. Details of the finite element discretization at the bottom of the notch for model 1: a) L1 –
mesh size 0.3r, b) L2 – mesh size 0.2r, c) L3 – mesh size 0.15r; where r = δ/2 (see Figure 15).

The values of the stresses in the x and y directions for the three cases analyzed are listed in Table
1. For all the cases, the results obtained with the finest mesh were taken as a reference. From the table
it can be seen that, as the element size decreases, the difference in the numerical results is smaller (in
the order of a couple of percentage points).
Table 1. Results for cases 1, 2 and 3 in model 1.
Mesh  xx /MPa error  yy /MPa error
L1 0.61 5.7% 0.65 13.5%
Case 1 L2 0.66 1.5% 0.72 4.5%
L3 0.65 Ref. 0.75 Ref.

L1 -4.94 1.6% -10.7 5.3%

Case 2 L2 -4.99 0.6% -11.0 2.7%
L3 -5.02 Ref. -11.3 Ref.

L1 3.25 2.2% 8.99 4.4%

Case 3 L2 3.27 2.8% 9.23 1.8%
L3 3.18 Ref. 9.40 Ref.

17
 Therefore, the stresses at point A for this case are given by,

 xxA1  0.65 p1  5.02 p2  3.18 p3

(33)
 yyA1  0.75 p1  11.30 p2  9.40 p3
It is worth noting that Frash et al. (Frash et al., 2015) also obtained an estimate of the stress values
at the bottom of the notch by making a number of simplifying assumptions regarding the geometry of
the sample and by considering that the material behaved elastically. Their results for the stresses at
point A (see Figure 15) are given by

  
 yyt  1.00 p1  1  2 2 d  p 2  1  2 2d  p3  (34)

where d is the notch depth and  is the aperture of the notch. For this particular sample geometry
2d   50.8 , therefore, the stress values at the bottom of the notch according to Frash et al. (Frash et
al., 2015) are given by

 yyt 1  1.00 p1  15.25 p2  13.25 p3 (35)

The coefficients for p1 , p2 , and p3 obtained from the theoretical approximation are in quite good
agreement with the ones obtained from the finite element analysis. This occurs because both
approaches use similar assumptions regarding the material behaviour (elastic). It is known that, as the
material around point A approaches the breaking point (fracture initiation) plastic deformations can
occur. This will be addressed in future work by the finite element analysis team via the incorporation
of the appropriate plasticity material model.
For illustration purposes, the stress fields for all three cases are shown in Figure 18 thru Figure 20.
In this case, the map of stresses obtained with the COMPTet is very smooth, without any sign of
checkerboard pattern issues (Figure 18–Figure 20).
a) b) c)
σxx / GPa σyy / GPa p / GPa

Figure 18. Model 1 – Case 1 – Mesh L3 – a) detail of Cauchy stress  xx , b) detail of the Cauchy
stress  yy and c) detail of the pressure p .

18
 a) b) c)
σxx / GPa σyy / GPa p / GPa

Figure 19. Model 1 – Case 2 – Mesh L3 – a) detail of Cauchy stress  xx , b) detail of the Cauchy
stress  yy and c) detail of the pressure p .

a) b) c)
σxx / GPa σyy / GPa p / GPa

Figure 20. Model 1 – Case 3 – Mesh L3 – a) detail of Cauchy stress  xx , b) detail of the Cauchy
stress  yy and c) detail of the pressure p .

Model 2: The dimensions for this model are shown in Figure 15. In order to verify the convergence
of the stress values at the bottom of the notch with decreasing element sizes three different mesh sizes
(see Figure 21) were tested.
a) b) c)

z z z

y x y x y x

Figure 21. Details of the finite element discretization at the bottom of the notch for model 2: a) L1 –
mesh size 0.3r, b) L2 – mesh size 0.2r, c) L3 – mesh size 0.15r; where r is the radius of the notch
(see Figure 15).

The values of the stresses in the x and y directions for the three cases analyzed are listed in Table
2. For all the cases, the results obtained with the finest mesh were taken as a reference. Again, from
the table it can be seen that, as the element size decreases, the difference in the numerical results is
smaller (in the order of a couple of percentage points).

19
 Table 2. Results for cases 1, 2 and 3 in model 2.
Mesh  xx /MPa error  yy /MPa error
L1 0.69 4.0% 0.85 6.7%
Case 1 L2 0.70 1.9% 0.88 2.7%
L3 0.72 Ref. 0.91 Ref.

L1 -9.37 1.1% -24.3 2.0%

Case 2 L2 -9.67 2.1% -24.5 1.2%
L3 -9.47 Ref. -24.8 Ref.

L1 7.58 1.2% 22.4 2.2%

Case 3 L2 7.86 2.4% 22.6 1.3%
L3 7.67 Ref. 22.9 Ref.
Therefore, the stresses at the bottom of the notch are then given by,

 xxA 2  0.72 p1  9.47 p2  7.67 p3

(36)
 yyA 2  0.91 p1  24.8 p2  22.9 p3
while, the stress values at the bottom of the notch according to Frash et al. are given by

 yyt 2  1.00 p1  22.0 p2  20.0 p3 (37)

Again, the coefficients for p1 , p2 , and p3 obtained from the theoretical approximation are in quite
good agreement with the ones obtained from the finite element analysis.
For illustration purposes, the stress fields for all three cases are shown in Figure 22 – Figure 24.
In this case, the map of stresses is very smooth, without any sign of checkerboard pattern issues (Figure
22 – Figure 24).
a) b) c)
σxx / GPa σyy / GPa p / GPa

Figure 22. Model 2 – Case 1 – Mesh L3 – a) detail of Cauchy stress  xx , b) detail of the Cauchy
stress  yy and c) detail of the pressure p .

20
 a) b) c)
σxx / GPa σyy / GPa p / GPa

Figure 23. Model 2 – Case 2 – Mesh L3 – a) detail of Cauchy stress  xx , b) detail of the Cauchy
stress  yy and c) detail of the pressure p .

a) b) c)
σxx / GPa σyy / GPa p / GPa

Figure 24. Model 2 – Case 3 – Mesh L3 – a) detail of Cauchy stress  xx , b) detail of the Cauchy
stress  yy and c) detail of the pressure p .

Experimental Results: Physical samples of acrylic cut to the dimensions of Models 1 and 2 were
fractured under triaxial stress conditions with hydraulic pressure inside the notches (Frash et al., 2015).
The stress fields at fracture initiation were recorded for each experiment giving the results shown in
Table 3. Substituting these stress fields into equations (33), (35), (36) and (37) gives estimates of  yy
at fracture initiation for each respective model, representing the estimated notch tensile strength of the
acrylic (Table 3). The bulk tensile yield strength of acrylic measured by conventional means is in the
range of 56 to 76 MPa (McMaster-Carr Supply Company, 2015). The error in tensile strength
estimation using the numerical approach was less than that of the theoretical approach (Frash et al.,
2015). These results emphasize the importance of incorporating plasticity in future work to more
accurately model fracturing processes.
Table 3. Experimental fracture initiation stress fields for Models 1 and 2.
Model p1 (MPa) p2 (MPa) p3 (MPa)  yyA (MPa)  yyt (MPa)
Model 1 – L3 28 2 12.5 298 414
Model 2 – L3 116 2 24.8 263 235

8. CONCLUSIONS
In this paper a new formulation for a composite tetrahedron finite element was presented. The new
element was implemented into an in-house combined finite-discrete element code and it was tested
thoroughly. The tests results demonstrate that the formulation exhibits a very good convergence with
21
decreasing element size, while also avoiding the common problems of volumetric locking inherent to
the lower order tetrahedron finite elements.
This new finite element was used in a number of simulations with the aim of estimating the stress
state at the bottom of a notch for hydraulic fracture initiation. The results obtained were in quite good
agreement with previously reported theoretically derived estimates. However, a relatively large
discrepancy is noted when comparing the numerical results to the experimental observations. The
main reason for this discrepancy is due to the assumption that the material behaves elastically the
whole time, i.e., no plasticity is considered in the finite element modeling. Because of this, further
work is needed in this regard in order to incorporate plasticity mechanisms in the material model.

9. REFERENCES
1. Munjiza, A. (1992), Discrete elements in transient dynamics of fractured media, PhD thesis,
University College Swansea, University of Wales, Swansea, UK.
2. Munjiza, A., Owen D.R.J. and Bicanic N. (1995), “A combined finite-discrete element method in
transient dynamics of fracturing solids”, Engineering Computations, Vol. 12, No. 2, pp.145–174.
3. Munjiza, A. and Andrews K.R.F. (1998), “NBS contact detection algorithm for bodies of similar
size”, International Journal for Numerical Methods in Engineering, Vol. 43, No. 1, pp. 131–149.
4. Munjiza, A., Andrews K.R.F. and White J.K. (1998), “Combined single and smeared crack model
in combined finite-discrete element method”, International Journal for Numerical Methods in
Engineering, Vol. 44, No. 1, pp. 41–57.
5. Munjiza A. and Andrews K.R.F. (2000), “Penalty function method for combined finite-discrete
element systems comprising large number of separate bodies”, International Journal for
Numerical Methods in Engineering, Vol. 49, No. 11, pp. 1377–1396.
6. Munjiza A., Rougier E. and John N.W.M. (2006), “MR linear contact detection algorithm”,
International Journal for Numerical Methods in Engineering, Vol. 66, No. 1, pp. 46–71.
7. Munjiza A. (2004), The Combined Finite-Discrete Element Method, John Wiley and Sons Ltd,
London.
8. Munjiza A., Knight E.E. and Rougier E. (2012), Computational Mechanics of Discontinua, John
Wiley and Sons Ltd, London.
9. Lei Z. (2011), Combined finite-discrete element methods and its application on impact fracture
mechanism of automobile glass, Ph.D thesis, South China University of Technology, Guangzhou,
China.
10. Munjiza A., Lei Z., Divic V. and Peros B. (2013), “Fracture and fragmentation of thin shells using
the combined finite–discrete element method”, International Journal for Numerical Methods in
Engineering, Vol. 95, No. 6, pp. 478–498.
11. Xu D., Kaliviotis E., Munjiza A., Avital E., Ji C. and Williams J.J.R. (2013), “Large scale
simulation of red blood cell aggregation in shear flows”, Journal of Biomechanics, Vol. 46, No.
11, pp. 1810–1817.
12. Ji C., Munjiza A. and Williams J.J.R. (2012), “A novel iterative direct-forcing immersed boundary
method and its finite volume applications”, Journal of Computational Physics, Vol. 231, No. 4,
pp. 1797–1821.
13. Latham J., Munjiza A., Mindel J., Xiang J., Guises R., Garcia X., Pain X., Gorman G. and Piggott
M. (2008), “Modelling of massive particulates for breakwater engineering using coupled
FEMDEM and CFD”, Particuology, Vol. 6, No. 6, pp. 572–583.
14. Rougier E., Knight E.E., Broome S.T., Sussman A.J. and Munjiza A. (2014), “Validation of a
three-dimensional Finite-Discrete Element Method using experimental results of the Split

22
 Hopkinson Pressure Bar test”, International Journal of Rock Mechanics and Mining Sciences,
Vol. 70, pp. 101–108.
15. Rougier, E., Knight, E.E., Munjiza, A., Sussman, A.J., Broome, S.T., Swift R.P. and Bradley, C.R.
(2011), “The Combined Finite-Discrete Element Method applied to the Study of Rock Fracturing
Behavior in 3D”, in Proceedings of the 45th US Rock Mechanics / Geo-mechanics Symposium,
San Francisco, CA, June 26–29.
16. Carey, J.W., Lei, Z., Rougier, E., Mori, H. and Viswanathan, H. (2015). “Fracture-permeability
behavior of shale”, The Journal of Unconventional Oil and Gas Resources, Vol. 11, Pages 27–43.
17. Lisjak A., Figi D. and Grasselli G. (2014), “Fracture development around deep underground
excavations: insights from FDEM modelling”, Journal of Rock Mechanics and Geotechnical
Engineering, Vol. 6, No. 6, pp. 493–505.
18. Trivino L.F. and Mohanty B. (2015), “Assessment of crack initiation and propagation in rock from
explosion-induced stress waves and gas expansion by cross-hole seismometry and FEM–DEM
method”, International Journal of Rock Mechanics and Mining Sciences, Vol. 77, pp. 287–299.
19. Munjiza A., Rougier E. and Knight E.E. (2015), Large strain finite element method: a practical
course, John Wiley and Sons Ltd, London.
20. Malkus D.S. (1978), “Mixed finite element methods – reduced and selective integration
techniques: a unification of concepts”, Computer Methods in Applied Mechanics and Engineering,
Vol. 15, No. 1, pp. 63–81.
21. Zienkiewicz O.C., Taylor R.L. and Too J.M. (1971), “Reduced integration technique in general
analysis of plates and shells”, International Journal for Numerical Methods in Engineering, Vol.
3, No. 2, pp. 275–290.
22. Hughes T.J.R. (1980), “Generalizing of selective integration procedures to anisotropic and
nonlinear media”, International Journal for Numerical Methods in Engineering, Vol. 15, No. 9,
pp. 1413–1418.
23. Moran B., Ortiz M. and Shih C.F. (1990), “Formulation of implicit finite element methods for
multiplicative finite deformation plasticity”, International Journal for Numerical Methods in
Engineering, Vol. 29, No. 3, pp. 483–514.
24. de Souza Neto E.A., Peric D., Dutko M. and Owen D.R.J. (1996), “Design of simple low order
finite elements for large strain analysis of nearly incompressible solids”, International Journal of
Solids and Structures, Vol. 33, No. 20–22, pp. 3277–3296.
25. Liu G.R. and Trung N.T. (2010), Smoothed Finite Element Methods, CRC Press, Boca Raton.
26. Zienkiewicz O.C., Rojek J., Taylor R.L. and Pastor M. (1998), “Triangles and tetrahedra in explicit
dynamics codes for solids”, International Journal for Numerical Methods in Engineering, Vol.
43, No. 3, pp. 565–583.
27. Bonet J. and Burton A. (1998), “A simple average nodal pressure tetrahedral element for
incompressible and nearly incompressible dynamic explicit applications”, Communications in
Numerical Methods in Engineering, Vol. 14, No. 5, pp. 437–449.
28. Dohrmann C.R., Heinstein M.W., Jung J., Key S.W. and Witkowski W.R. (2000), “Node-based
uniform strain elements for three-node triangular and four-node tetrahedral meshes”, International
Journal for Numerical Methods in Engineering, Vol. 47, No. 9, pp. 1549–1568.
29. Bonet J., Marriott H., Hassan O. (2001), “An averaged nodal deformation gradient linear
tetrahedral element for large strain explicit dynamic applications”, Communications in Numerical
Methods in Engineering, Vol. 17, No. 8, pp. 551–561.
30. Joldes G.R., Wittek A. and Miller K. (2009), “Non-locking tetrahedral finite element for surgical
simulation”, Communications in Numerical Methods in Engineering, Vol. 25, No. 7, pp. 827–836.
31. De Micheli P.O. and Mocellin K. (2009), “A new efficient explicit formulation for linear
tetrahedral elements non-sensitive to volumetric locking for infinitesimal elasticity and

23
 inelasticity”, International Journal for Numerical Methods in Engineering, Vol. 79, No. 1, pp. 45–
68.
32. Taylor R. (2000), “A mixed-enhanced formulation tetrahedral finite elements”, International
Journal for Numerical Methods in Engineering, Vol. 47, No. 1–3, pp. 205–227.
33. de Souza Neto E.A., Andrade Pires F.M. and Owen D.R.J. (2005), “F-bar-based linear triangles
and tetrahedra for finite strain analysis of nearly incompressible solids. Part I: formulation and
benchmarking”, International Journal for Numerical Methods in Engineering, Vol. 62, No. 3, pp.
353–383.
34. Camacho G.T. and Ortiz M. (1996), “Computational modelling of impact damage in brittle
materials”, International Journal of Solids and Structures, Vol. 33, No. 20–22, pp. 2899–2938.
35. Thoutireddy P., Molinari J., Repetto E. and Ortiz M. (2002), “Tetrahedral composite finite
elements”, International Journal for Numerical Methods in Engineering, Vol. 53, No. 6, pp. 1337–
1351.
36. Frash L.P., Carey J.W. and Viswanathan S. (2015), “Method for in situ tensile-hydraulic-
fracturing in triaxial-coreflood systems”, International Journal of Rock Mechanics and Mining.
(under review).
37. Timoshenko S.P. and Goodier J.N. (1970), Theory of Elasticity, 3rd edn, McGraw-Hill, New York.
38. McMaster-Carr Supply Company (2015), “More about plastics: Document 8574KAC”, available
at: http://www.mcmaster.com/#8574kac/=wu44pt (accessed 20 May 2015)

24

View publication stats