and Materials
Engineering
Abstract
In this study the discrete element method (DEM) is developed in the framework of the
constitutive law of elastic isotropic materials for two-dimensional plane stress analysis
of solids. Contact stinesses (normal and tangential spring constants) are theoretically
derived for hexagonal elements in an arbitrary arrangement as a function of thickness
and material parameters including Youngs modulus and Poissons ratio. Moreover, a
new method is presented to calculate stress components within an element. To validate
the accuracy and eciency of our discrete model several test problems are given. At
first, uniaxial tension test using a bar is performed and the convergence of the solution to a definite value in the limit of mesh refinement is validated. In addition, two
examples of stress wave propagation problems are given. Compressive wave speed
is calculated and through comparing the numerical results with the other discrete element models and also analytic solution, the accuracy of the present DEM model is then
discussed.
Key words : Discrete Element Method, Constitutive Law, Plane Stress, Tensor, Stress
Wave
1. Introduction
Numerical simulation is an eective measure for studying various mechanical phenomena in materials and structures. In this process, the continuum assumption is usually used
as a basis for the idealization of many engineering materials. Finite element (FEM), finite
dierence (FDM), and boundary element methods (BEM) are among the numerical solution
techniques, which are founded on this assumption. However, there are many situations especially in large deformation problems with void nucleation and fracture in engineering materials where continuity can not be assumed. In these cases, for the above numerical techniques,
special treatments therefore have to be added in the calculation procedures, which make the
process more complicated.
Discrete element method (DEM) is a member of the numerical analysis family, like
FEM, FDM and BEM, that solves numerically the mathematical formulation for the whole
medium constructed from a constitutive formula for each element discretizing a continuous
medium. However, the DEM is especially designed to deal with even engineering problems
that are dominated by discontinuous mechanical behaviors. The fundamental assumption used
in DEM distinguishes this method with other numerical methods. It assumes that the material consists of discrete rigid particles, which may have dierent shapes and properties. The
rigid particles interact with neighboring particles according to interaction laws that are applied at the point of contact. This numerical technique was first proposed by Cundall(1) to
solve the problems in rock mechanics where the continuity between the separate elements
does not exist. DEM is capable for modeling multiple interacting continuities, discontinuities
1478
or fracturing bodies, and also it has been widely used in geotechnical engineering and powder
technology(2) . Recently, this method has also been used to simulate the failure and damage
processes of brittle materials(3), (4) and also for some impact problems(5) . However, a main
issue in DEM as a tool for modeling continuum problems is to determine the interaction laws
governing at points of contact between elements.
This paper aims to present a new approach to establish the contact stinesses theoretically
as a function of thickness and material parameters including Youngs modulus and Poissons
ratio for an elastic isotropic material under the plane stress condition. However, most of the
past researches to develop a mathematical expression for the DEM inter-element parameters
rely on special arrangements of discrete elements as a unit cell(6) or seven-disk elements(7) .
The distinct point of the present approach is to develop a mathematical expression of spring
constants by only considering two elements in an arbitrary orientation, which also promises
a positive value of contact stinesses for all engineering materials. Moreover, an alternative
method to the average stress method(9) is proposed to calculate the stresses within element.
In addition, several test problems are given and through comparing the results with analytic
solutions, the accuracy and validity of the present DEM model and stress calculation is then
discussed.
Nomenclature
E : Youngs modulus, GPa
: Poissons ratio
: density, kg/m3
kn : normal stiness, N/m
k s : shear stiness, N/m
un : normal displacement, m
u s : shear displacement, m
fn : normal spring force, N
f s : tangential spring force, N
: thickness, mm
r : radius, mm
V : volume , mm3
t : time, s
dt : time increment, s
: coecient of mass damping, s1
p : applied stress, kPa
gi : local coordinate basis
ei : global coordinate basis
Subscripts
NoE : number of elements
el : element
Fig. 1
1479
Taking the assumption that elastic internal energy is totally stored in springs, the relation
u
u
n1i j = 2rni j and si j = 2rsi j ( j = i1, ..., i6), between normal, shear strains and corresponding
displacements the total internal energy stored in an arbitrary element eli may be given as
eli
Wint
= r2
i6
{kni j ( ni j )2 + k si j ( si j )2 }
(1)
j=i1
Further more, the strain energy density of one disk element eli , is given as follow
eli
Wint(d)
=
eli
Wint
V
V = 2 3r2
(2)
(3)
eli
Wint(d)
mn
(4)
By using Eqs. (1)(4), assuming homogenous material properties and the transformation
n1i j = cos2 (i j ){11 } + sin2 (i j ){22 } + sin(2i j ){12 }
(5)
(6)
between strain components in local and global coordinate system, the following relation between stress and strain components will be obtained.
3
3
11 =
(7)
(8 3kn1 + kn2 + 3k s2 )11 +
(kn2 k s2 )22
12
4
3
3
22 =
(kn2 k s2 )11 +
(3kn2 + k s2 )22
(8)
4
4
3
12 =
(3kn2 + 2k s1 + k s2 )12
(9)
6
1480
Considering Eqs. (7)(9), the spring constants for an elastic isotropic material under plane
stress condition will be obtained as
3
3 E
kn1 = kn2 = kn =
(c12 + c22 ) =
(10)
3
3 1
3
3 (1 3)E
(c22 3c12 ) =
(11)
k s1 = k s2 = k s =
3
3
1 2
However, the above spring constants are valid only when each element is surrounded
by six disk elements. It is clear that the elements on the boundary, which have less than six
neighbors can not be modeled by seven-disk model. To overcome the above disadvantage, a
two-disk model with hexagon shape boundaries is developed as depicted in Fig. 2 (b). At the
interface of two elements, a local coordinate system is defined along and perpendicular to the
edge with unit normal basis g1 , g2 and g3 given by
g1 = cos i j e1 + sin i j e2
(12)
g2 = sin i j e1 + cos i j e2
(13)
g3 = e3
(14)
According to the cauchys theorem the traction force tni j on the each edge of the element
ni j in the local coordinate system by
eli is denoted by the unit normal vector
tni j =
ni j
ni j
(15)
where
ni j = g1 and stress tensor
ni j is expressed by
ni j = (
ni j )mn gm gn ,
for
(m, n = 1, 2, 3)
(16)
Taking the assumption that internal energy is stored only in springs, the traction force is given
by
tni j = (
fni j
f si j
)g1 + ( )g2
l
l
(17)
where fni j = kni j uni j , f si j = k si j u si j and l = 2 3r/3. By considering Eqs. (15)(17) and
symmetry of the stress tensor, the following relations for a plane stress problem are given
m3 =
3m = 0
for m = 1, 2, 3
t1i j =
11i j =
n1i j =
kni j uni j
fni j
=
=
l
l
t2i j =
12i j =
21i j =
si j
(18)
3kni j uni j
2r
3k si j u si j
f si j
k si j u si j
=
=
=
l
l
2r
(19)
(20)
Furthermore, the normal and shear strains on the edge may be given by
n1i j =
uni j
2r
u si j
2
si j =
=
2l
(21)
3u si j
4r
(22)
where they are expressed in the local coordinate system. To achieve the total strain energy at
a point on the edge, stress and strain vectors may be expressed in the local coordinate system
as
11
n1i j
=
(23)
22 =
n2i j
12
si j
1481
n1i j
=
n2
ij
2 si j
(24)
and
= c
or
n1i j
n2i j
si j
(25)
c11
= c21
c31
c12
c22
c32
c13
c23
c33
n1i j
n2i j
si j
2
(26)
si j are only a
However, as the stress components are given by Eqs. (19)(22),
n1i j and
function of
n1i j and
si j respectively. Considering the above fact, symmetry of the constitutive
law, and Eq. (26), it can be deduced
c12 = c21 = c13 = c31 = c23 = c32 = 0
On the other hand, c has a diagonal form and is given by
c11 0
0
c = 0 c22 0
0
0 c33
(27)
(28)
For an elastic isotropic material under the plane stress condition, the constitutive matrix in the
global coordinate system is given by
E
1 2
(30)
c12 = c21 =
E
1 2
(31)
c33 =
E
2(1 + )
(32)
(33)
However, the constitutive matrix should be independent of the coordinate system. On the
other hand, to keep the equivalence between c and c and considering a diagonal form for c ,
the components of c may be identified by the eigenvalues of c and c11 , c22 and c33 can be
determined eventually as
c11 = c11 + c12
(34)
(35)
c33 = c33 =
1
(c11 c12 )
2
(36)
3
3 E
(c12 + c22 ) =
(37)
kn =
3
3 1
1482
E
(c11 c12 ) =
4
4(1 + )
(38)
which are independent of the DEM element radius. In addition, assuming non-negative values
for contact stinesses, equations given here for normal and tangential spring constants can be
applied to all engineering materials whereas shear stiness given by Liu(7) limits the maximum
value of Poissons ratio to = 13 .
In addition, it should be noted that Eqs. (19) to (22) are dependent on element special
geometry, but by introducing the special coordinate system, geometry parameters vanish from
the both sides of Eq.( 26) and consequently the derived spring constants are independent of
element special geometry. However, because the strain components are derived by use of the
hexagonal geometry property, the application of the proposed DEM model must be limited to
the hexagonal elements whereas there is no limitation of the type of loading as no special type
of loading is assumed to derive Eqs. (37) and (38).
2.3. Stress Calculation (cross sectional method)
In the present method to calculate stresses the hexagonal discrete element is approximated by a circle which is virtually cut into two halves. Stress components are then calculated
from the vertical and horizontal static equilibrium principle. For clarity, the method to calculate the normal stress in e1 direction is presented here. At first a particle is cut vertically
L
into two halves, as depicted in Fig. 3. R11 and 11
are then calculated by force equilibrium
Fig. 3
for right and left half circles respectively. To decide normal stress, 11 in e1 direction, the
following conditions are given, where the unbalance force will go to displace the element:
L
L
L
If {R11 > 0 & 11
> 0 & (R11 > 11
)} 11 = 11
R
L
L
R
R
If {11 > 0 & 11 > 0 & (11 > 11 )} 11 = 11
L
L
If {R11 < 0 & 11
< 0 & (R11 > 11
)} 11 = R11
L
L
L
If {R11 < 0 & 11
< 0 & (11
> R11 )} 11 = 11
R
L
If {(11 11 ) <= 0} 11 = 0
L
} 11 = R11
If {R11 = 11
Similar technique is also applied to calculate 22 and 12 .
1483
Fig. 4
(a) The uniaxial tension test on a bar under a force P per unit volume, (b) stress
across the cross section for the finest mesh using the presented method and
average stress method(9) , (c) stress distribution across the cross section of the
bar for dierent DEM element size and (d) average strain energy versus time
measured at the center of the DEM numerical specimen.
for element radii r of 0.05, 0.1 and 0.2 mm with total number of elements NoE of 9125, 2243
and 542 respectively. In order to minimize unrealistic element oscillation in the model, the
coecient of mass damping, of 100 s1 is used in the calculation, which is obtained through
trial and error process. The time increment dt of 0.01 s is chosen to ensure the stability of
the calculation.
Theoretically, engineering stresses for a bar under uniaxial tension across an arbitrary
cross section of the bar perpendicular to the longitudinal axis should be equal to the applied
stress. To validate the accuracy of the present DEM method, stresses across a cross section
10 mm distant from the end are measured by the present method and then compared to the
average stress method(9) as shown in Fig. 4 (b). Figure 4 (c) shows the convergence of the
solution to the applied stress as expected by the mesh refinement. Furthermore, it should be
noted that stress calculation in the present DEM model is only a post-processing calculation
and in fact displacements are primary variables to be used to calculate other parameters like
stresses or strains. To show the equivalence between the present stress calculation and corresponding displacements, the strain energy per unit volume is measured for the element at
center of the bar. The average strain energy(eint ) is calculated by two dierent approaches as
1
eint = 4V
(kn u2n j + ks u2s j ) for j = 1 to 6 and alternatively as eint = 12 (11 11 + 22 22 + 212 12 )
where stresses are calculated by cross sectional method presented in the previous section.
Consequently strains are calculated by stresses and the constitutive law. Results for dierent
element size are plotted in Fig. 4 (d). A good agreement between stresses and corresponding
displacements can be easily seen, which further confirms the accuracy of the present method
to calculate stresses.
1484
Fig. 5
E
2.98
210
0.3
0.27
1440
7800
r
0.5
0.2
NoE
1112
3204
pmax (MPa)
1.65
250
t1 (s)
20
0.5
dt(s)
0.1
0.005
(a) Example1: Bar specimen geometry (b) Example2: Plate specimen geometry
(c) 11 history at the center of the bar (d) 22 history at the center of the plate
Spring constants are calculated by Eqs. (37) and (38) for the present model and Eqs.
(10) and (11) for seven-disk model. The normal stress history in the direction of applied load
at the center of plates can be successfully analyzed. The stress wave propagation results are
shown in Figs. 5 (c) and (d). From the results, the wave propagation inside the plates can be
clearly indicated. At first the incident wave travels as the compressive stress wave and after
the reflection from the back boundary, it is inverted as a tensile stress wave.
To verify the accuracy of the calculation, stress wave velocity in the model is compared
with the theoretical one. The stress wave velocity in the model can be calculated by dividing
the traveling distance by the traveling time (the interval between the negative and positive
peaks). The result for both examples are given in Table 2.
Table 2
Specimen
Theoretical wave speed (m/s)
Presented DEM model (m/s)
Seven-disk model
Bar
1508
1505
1437
Plate
5800
5568
5446
Deviations of 0.2 % and 4 % from the theoretical longitudinal wave velocity are obtained
by the presented model, for the bar and the plate respectively. In contrast, the seven-disk
1485
element model gives a deviation of 4.7 % and 6.1 %. In other words, these results show the
eect of spring constants on wave velocity. As it is given by Eqs. (10), (11), (37) and (38)
both models calculate the same value for normal stiness whereas the shear stiness is dierent.The presented model gives the shear stiness values of 0.573 KN/mm and 41 KN/mm for
plaster and steel respectively, whereas the other model gives 0.189 KN/mm and 24 KN/mm
for both the materials.
4. Conclusion
Summarizing the obtained results can deduce conclusion that the discrete element model
proposed here can be validly used for the numerical analysis of stress wave propagation problems. By comparing the numerical results from the present model with the results from the
other model such as seven-disk model(7) , the validity and the accuracy of the presented algorithm were clearly confirmed. In particular, the spring constants of the present approach induce more acceptable wave velocity in the problems of the stress wave propagation. Moreover,
the present expressions of contact stinesses can be applied to a broader range of materials
whereas corresponding equations given by other researchers(6) (8) limit the value of Poissons
ratio to = 13 to ensure a positive value of shear stiness. However, application of the present
approach is limited to the problems with small deformation. Further development is needed
for large deformation problem.
References
( 1 ) Cundall, P.A., A Computer Model for Simulating Progressive Large Scale Movement in
Block Rock System, Symp ISRM Proc 2, (1971), pp. 129-136.
( 2 ) Thomas, P.A. and Bray, J.D., Capturing Nonspherical Shape of Granular Materials with
Disk Clusters. Journal of Geotechnical and Geoenvironmental Engineering, Vol. 125,
No. 3 (1999), pp. 169-178.
( 3 ) Camborde, F., Mariotty, C. and Donze, F. V., Numerical Study of Rock and Concrete
Behaviour by Discrete Element Modeling, Computers and Geotechnics, Vol. 27 (2000),
pp. 225-247.
( 4 ) Mihradi, S., and Homma, H., Fracture Simulation of Brittle Material under Impact Loading by Discrete Element Method, Proceeding of The 13th International Symposium on
Plasticity and Its Current Application, Alaska (2007), pp. 307-309.
( 5 ) Hentz, S., Donze, F. V., and Daudeville, L., Discrete Element Modeling of Concrete
Submitted to Dynamic Loading at High Strain Rates, Computers and Structures, Vol. 82
(2004), pp. 2509-2524.
( 6 ) Tavarez, A.F, and Plesha, M.E., Discrete Element Method for Modelling Solid and Particulate Materials, Int. J. Numer. Meth. Eng., Vol. 70 (2007), pp. 379-404.
( 7 ) Liu, K., Gao, L. and Tanimura, S., Application of Discrete Element Method in Impact
Problems, JSME International Journal, Series A, Vol. 47, No. 2 (2004), pp. 138-145.
( 8 ) Sawamoto, Y., Tsubota, H., Kasai, Y., Koshika, N. and Morikawa, H., Analytical studies
on Local Damage to Reinforced Concrete Structures under Impact Loading by Discrete
Element method, Nuclear Engineering and Design, Vol.179 (1998), pp.157-177.
( 9 ) Drescher, A. and Jong, G., Photoelastic Verification of a Mechanical Model for the Flow
of a Granular Material, J. Mech. Phys. Solids, Vol. 20 (1972), pp. 337-351.
1486