Você está na página 1de 18

Engineering Fracture Mechanics 71 (2004) 601–618


Application of discrete element methods to fracture

mechanics of rock bursts
P.P. Proch
Department of Structural Mechanics and Civil Engineering, Czech Technical University,
Thakurova 7, Prague 6 166 29, Czech Republic
Received 30 October 2002; accepted 5 November 2002

In deep mining engineering sudden release of accumulated potential energy occurs under special conditions. This
phenomenon is known as bumps or rock bursts. During the bumps the rock turns out to grain material, which bursts
into a free space. The mathematical and experimental modeling requires very attentive treatment. Two methods put
forward in this paper can serve a mathematical tool for solving such problems. The first, discrete hexagonal element
method can be considered as one of discrete element methods (DEM), which are very often used in mechanics of
granular media. They substitute the methods for solving continuum problems. The great disadvantage of the classical
DEM, such as the particle flow code––PFC (material properties are characterized by spring stiffness), is to feed them
with material properties provided from laboratory tests (YoungÕs moduli, PoissonÕs ratio, etc.), which are not quite
consistent with stiffnesses of springs, the PFC requires. This is why we utilize the principal idea of the DEM, but cover
the continuum by hexagonal elastic, or elastic–plastic elements. In order to complete the study, other DEM is discussed
and numerical results of both methods are compared with experiments in scale model.
 2003 Elsevier Ltd. All rights reserved.

Keywords: Free hexagonal element method; Statical PFC; Localized damage; Cracking and bumps occurrence during mining or

1. Introduction

For bumps to occur, the rock has to possess certain particular material properties, leading to accu-
mulation of potential energy and ability to release this energy. Such a material may be brittle, or the bumps
arise at interfacial zones of two parts of the rock, both having principally different material properties.
The experiments and numerical studies are concentrated on loading of longwall seams. The coal is
supposed to be drilled in great depths. The mathematical models are prepared in compliance with the
experiments. Both numerical methods, free hexagonal element method and statical PFC are selected to
assess the local energy concentration and forthcoming bumps.

E-mail address: webmaster@vc.cvut.cz (P.P. Prochazka).

0013-7944/$ - see front matter  2003 Elsevier Ltd. All rights reserved.
602 P.P. Prochazka / Engineering Fracture Mechanics 71 (2004) 601–618

Not many publications that were concerned with local sudden failure of stability were issued in the past.
They are concerned with various ways of the coal seam loadings, material properties of rock overburden,
extrusion of gas, disturbances in zones off-face the longwalls, etc. In [10] a study of a behavior of cubes of
coal that were loaded by concentrated load was carried out. In [22], scale model and theoretical studies on
the mechanism of coal and gas bumps are discussed. In [14] model material is used to simulate coalmine
bumps. Since determination of in situ rock strength is required for successful design of underground
structures, Bureau of Mines in Denver (there is only one Bureau of Mines in the USA now being in
Pittsburgh) issued a series of publications, where estimates for engineers can be found. A summary is
published in [12]. The publication [11] deals with reinforcement of upper part of mining (strong roof).
The principal problem of classical numerical methods, such as the finite element methods, the boundary
element methods, etc., consists in ‘‘too stiffÕ models, or too complicated simulations of the real states when
no a priori knowledge of crack initiation is available, [4,5], where local weakening due to damage is in-
troduced. This is why discrete element methods (DEM) have been introduced. Cohesive Zone Models use
classical ideas of Barenblatt, i.e. fracture mechanics tools, and are applied by Needelman, Tvergard,
Hutchison (composite materials), and others, see e.g. [9]. The fracture mechanics problems can be sub-
stituted by contact problems, which are in many respects more transparent, and lead to the same results, if
the material properties are properly stated. Moreover, it is not necessary to know in advance the crack
initiation. These models we will study in this paper.
In early 1970Õs Cundall, [2], and others, [3], introduced discrete elements starting with dynamic equi-
librium. First, brick-like elements were used (professional computer program UDEC), and later circular in
2D and spherical in 3D elements (PFC––particle flow code––both computer systems issued by ITASCA)
simulated the continuum behavior of structures. The application of such methods found their place mainly
in geotechnics, where soil is a typical grain material with the above-mentioned shape, [15,19]. If the material
parameters are well chosen, the mechanical behavior of discrete elements is very close to the reality. The
problem consists of finding such material parameters. There are plenty of attempts on how to find out these
parameters, but still there is no satisfactory output from those studies. The most prospective seems to be
coverage of the domain, defining the physical body by hexagonal elements, which are very close to disks,
and can cover the domain with a very small geometrical error. Substituting the discrete elements by elastic,
or elastic–plastic hexagons with the full contact of adjacent elements along their common boundaries yields
honeycomb-like shape of elements, [19,21], covering the structure of the continuous medium. The relations
inside the hexagonal particles are solved by a special form of the boundary element method, [1], for ex-
ample. It is necessary to note that beams create the honeycomb boundaries and there is no material inside
such particles.
The idea of hexagonal elements probably arose from honeycomb elements, their interior is filled with
the material. The free hexagons are used by Onck and van der Giessen in [16], for example, where large
scale of references on this topic can be found. In the latter publication the finite element method, e.g. [24],
is used to create the stiffness matrices of the elements, namely six finite elements are substructured to a
In applications to the geotechnical problems the disturbed state concept (DSC) issued by Desai [4,5] can
describe a wide spectrum of material states inside the elements, starting with elastic, elastic–plastic, [6], and
even damage states, [13], can be considered in DesaiÕs model. Using eigenparameters for plastic strain, or
relaxation stress, [7,8] completes the description of possible and suitable non-linear constitutive laws, which
moreover can be ‘‘tuned’’ from ‘‘in situ’’ measurements, or from results of scale modeling, [18]. Geo-
technical properties are defined on the boundaries of the elements. A typical formulation of the problem
involving generalized Mohr–Coulomb law combined with exclusion of tensile zones is proposed in [17],
where the technique using LagrangianÕs multipliers leads to mixed problem (both displacements and
stresses––element boundary tractions are iterated). In this paper penalty method is applied, and element
boundary tractions (former LagrangianÕs multipliers) are substituted by spring stiffnesses (penalty func-
P.P. Prochazka / Engineering Fracture Mechanics 71 (2004) 601–618 603

tions, or in our case the penalty parameters). The springs enable one to simulate the interfacial constraint,
namely the exclusion of the tensile tractions and application of the Mohr–Coulomb law. The Mohr–
Coulomb law is used in two basic forms, for brittle or almost brittle materials, and for soft rocks or
The reliability of the methods put forward is tested for rock bumps in coalmines. The experimental data
and the results from these models are in a good agreement with experimental results from [23].
The paper starts with formulation of the free hexagonal element method and then statical particle flow is
described. The bumps are typical problem in mining engineering, which is not yet clear enough. There is a
lack of numerical models, and this paper should partly bridge this deficiency. Examples are focused on
bumps occurrence due to extreme depth of longwall mines. The statical PFC is also employed and both
numerical methods are compared with experimental results.

2. Free hexagonal element method

The discrete free hexagonal element method may approximately be considered as one of the DEM. The
great disadvantage of some classical DEM, however, is to feed them with material properties provided from
laboratory tests (this is the case of statical particle flow code, formulated in the next section, as the discs are
connected by springs, while laboratories provide completely different material parameters). This is over-
come here by considering the material characteristics, which are similar to continuum. The principal idea of
classical DEM is adopted, and the domain defining the structure continuum is covered by the hexagonal
elastic in our case, or other (elastic–plastic, visco-elastic–plastic, etc.) elements. This step avoids necessity to
estimate the material properties of springs, which are essential for the DEM. The free hexagonal element
method fulfills a natural requirement consisting of the fact that the elastic properties are assigned to the
particles and other geotechnical material parameters (angle of internal friction, shear strength or cohesion)
to the contacts of elements. Considering that most particles are of the same shape leads to possibility to
apply very powerful iteration procedures, because the stiffness matrix can be stored in the internal memory
of computers.
When dealing with crack problems, two principal methods are used. First, the means of fracture me-
chanics are applied, or contact problem can be formulated. The latter case is generally not suitable, because
the direction, or way of propagation of the cracks should be known in advance.
In this paper the second possibility is used and the obstacle of unknown propagation of the cracks is
avoided by creating a mesh of free hexagonal elements, which are in mutual contacts in undeformed state,
but can be disconnected because of violating contact conditions.
The computational model is described in the next paragraph, where also relations needed for numerical
computation are introduced. The interface conditions are formulated in paragraph 2.2, where the La-
grangian principle is based on the penalty method. The penalty parameters are spring stiffnesses, and
springs connect the adjacent elements. The material characteristics of springs can posses a large value to
ensure the contact constrains. On the other hand, if, say, the tensile strength condition is violated, the
spring parameters tend to zero and naturally no energy contribution in normal direction to the element
boundary appears in the energy functional in this case.
The hexagonal particles are studied under various contact (interfacial) conditions of the grain particles
(elements). In out paper basically two contact conditions are involved:

• generalized Mohr–Coulomb hypothesis with exclusion of non-admissible tensile stresses along the con-
tact (rock mass),
• limit state of shear stresses and exclusion of tensile tractions along the contact (brittle coal seam).
604 P.P. Prochazka / Engineering Fracture Mechanics 71 (2004) 601–618

The first case is generally connected with applications in geotechnics, composite materials, shotcrete,
etc., and the second case is more appropriate for applications in underground bumps, or rock bursts. So far,
two-dimensional formulation and its solution are prepared and studied in this paper.
The problem formulated in terms of hexagonal elements (which are not necessarily mutually connected
during the loading process of the body, because of arising non-linearities due to the interfacial conditions)
enables one to simulate the way of propagation of cracks. The cracking of the medium can be described in
such a way that the local damage may be derived. Local deterioration of the material can also be seen from
the pictures, drawn for particular examples. Such a movement of elements and change of stresses cannot
probably be obtained from continuous numerical methods.

2.1. Computational model

Consider now a single hexagonal element (described by domain X with its boundary C). Its connection
with the adjacent elements is shown in Fig. 1. In each hexagonal element the pseudo-elastic material
properties are taken into consideration, i.e. in every iteration step the element behaves linearly, but the
material properties can change during the process of loading and unloading. This makes it possible to
introduce only elastic material stiffness matrix, which is homogeneous and isotropic, and we get well-known
integral equations being valid along the boundary abscissas of the hexagons, [1]:
( Z Z 
X2 X
6 X2
ckl ðnÞul ðnÞ ¼ pi ðxÞuik ðx; nÞ dx  ui ðxÞpik ðx; nÞ dx
l¼1 s¼1 i¼1 Cs Cs
2 Z
þ bi ðxÞuik ðx; nÞ dx ; k ¼ 1; 2; ð1Þ
i¼1 X

where bi components of the volume weight vector, Cs are edges (abscissas) of the boundary elements, n is
the point of observation, x is the integration point, ui are components of the vector of displacements
(defined not exclusively on the boundary, but also in the domain of the hexagonal element), pi are com-
ponents of the tractions, ckl is a matrix its values depend on the position of the point of observation. The
quantities with asterisk are given kernels. The kernels can be expressed as (for example, see [1]):
 xixk  l 2xixk
uik ¼ A Mðlog rÞdik  2 ; pik ¼ 2A 2 kðnk xi nixk Þ  kdik þ 2 xj nj ;
r r r

Fig. 1. Geometry of adjacent hexagonal elements.

P.P. Prochazka / Engineering Fracture Mechanics 71 (2004) 601–618 605

where A ¼ ðk þ lÞ=4plðk þ 2lÞ, M ¼ ðk þ 3lÞ=ðk þ lÞ, k ¼ l=ðk þ lÞ, xi ¼ xi  ni , r2 ¼ x21 þ x22 , and k
and l, are LameÕs material constants.
Assuming uniform distribution of boundary quantities (displacements ui ðxÞ and tractions pi ðxÞ; i ¼ 1; 2Þ,
and volume weight forces bi to be uniform in the domain X, and positioning the points of observation n
successively at the points ns , which are the centers of boundary abscissas of the hexagonal elements, a
simplified version of (1) is written as:
( )
X6 X2  Z Z  X 2 Z
1 s  s  s 
uk ðns Þ ¼ pi uik ðx; ns Þ dx  ui pik ðx; ns Þ dx þ bi uik ðx; ns Þ dx ; k ¼ 1; 2;
2 s¼1 i¼1 Cs Cs i¼1 X

usi pis
where and are values of the relevant quantities positioned at the ns , s ¼ 1; . . . ; 6, i.e. ¼ ui ðns Þ andusi
pis ¼ pi ðns Þ. Moreover, vector of influences of the volume weight forces on the boundary abscissas is
bs ¼ fcs1 ; cs2 g; s ¼ 1; . . . ; 6, and
X 2 Z
csk ¼ bsi uik ðx; ns Þ dCðxÞ; k ¼ 1; 2:
i¼1 X

For better and convenient computation the most important integrals are given in Appendix. In this way,
the integrals in (2) may be calculated directly, without numerical integration.
Let us introduce vectors as , bs , s ¼ 1; . . . ; 6, and also u and p as:
0 1 0 1 0 1
a1 b1 b1
B a2 C B b2 C B b2 C
 s  s B C B C B C
u1 p1 B a3 C B b3 C B b3 C
as ¼ ; bs ¼ C
; u ¼ B C; p ¼ B C; b ¼ B
B b4 C ;
us2 p2s B 4Ca b
B 4C B C
@ a5 A @ b5 A @ b5 A
a6 b6 b6

X Z 
ask ¼ usi pik ðx; ns Þ dx ;
i¼1 Cs

2 Z
bsk ¼ pis uik ðx; ns Þ dx:
i¼1 Cs

Using this notation, the relations on the elements (2) can be recorded as:
Au ¼ Bp þ b; ð3Þ
where A and B are (12 12) matrices, their components are singular integrals over the boundary abscissas.
Matrix A is generally singular, while matrix B is regular. This fact enables one to rearrange equations (3)
into the form:
Ku ¼ p þ V; K ¼ B1 A; V ¼ B1 b; ð4Þ
where the stiffness matrix K is different from that arising in applications of finite elements (it is prevailingly
non-symmetric), V is the vector of volume weight forces concentrated on the boundary abscissas (more
precisely at the point ns ). In this way, the discretized problem turns to the problem similar to the FEM.
Along the adjacent boundary abscissas it should hold (pil are EshelbyÕs forces):

pi þ piþ ¼ ðpil Þ þ ðpil Þ ; ð5Þ
606 P.P. Prochazka / Engineering Fracture Mechanics 71 (2004) 601–618

where superscript plus means from the right and minus from the left (at most two particles can be in a
Now using the relation (4) and (5), we get two times more unknowns then equations, because no con-
nection between elements has been introduced, yet. Equation (5) have to be accomplished by the constraint
of the type
ki ðu þ
i  ui Þ ¼ p i : ð6Þ
The latter conditions are penalty-like conditions, since if ki is great enough, the distribution of dis-
placements is continuous, displacement from the right is equal to displacement from the left. These con-
ditions can locally be violated, because of contact conditions, which are discussed later in this text.
Introducing boundary conditions and assuming ki remains great enough leads us to the stable system of
equations delivering unique solution. Even in the case when local disturbances occur, the solution can be
stable. It can happen that there are too many disturbances, such as cracks density, or localized damage
along a path (earth slope stability violation). Then the solution is unstable, there is a failure of the structure.
This is also the case of rock burst.
The discretization in the previous sense leads the non-linear system of algebraic equations, which are
solved by over-relaxation iterative procedure. This method is sufficient for study reasons. For larger extend
of equations the conjugate gradient method is prepared.
For displacements in the element domain X it holds:
( )
6 X 2  Z Z  X2 Z
s  s  s 
uk ðns Þ ¼ pi uik ðx; ns Þ dx  ui pik ðx; ns Þ dx þ bi uik ðx; ns Þ dCðxÞ ; k ¼ 1; 2;
s¼1 i¼1 Cs Cs i¼1 X

where the element boundary displacements and tractions are known from the previous computation,
providing the solution is stable. Using kinematical equations and HookeÕs law, internal stresses can be
calculated from (7). There is no danger of singularities, as the points x and ns never meet (point x lies inside
domain X and ns on boundary C for every admissible s).

2.2. Formulation of the contact problem

Recall that the displacements are described by a vector function u ¼ fu1 ; u2 g of the variable x ¼ fx1 ; x2 g.
The traction field on the particle boundaries is denoted either as p ¼ fp1 ; p2 g, or after projections to normal
and tangential directions as p ¼ fpn ; pt g. Similar result is valid for projections of displacements, u ¼ fun ; ut g.
Assuming the ‘‘small deformation’’ theory, it may be satisfactory to formulate the essential contact
conditions on the interface as follows (no penetration conditions):
½u kn ¼ uk;c k;a k
n  un 6 0 on CC ; ð8Þ
where CkC , k ¼ 1; . . . ; n are boundaries between adjacent particles, uk;a
is the normal displacement of current
element (a ¼ c) and a ¼ a belongs to the adjacent element, both on the current common boundary CkC , k
runs all common sides of the particles, n is the number of common sides of hexagons (having exactly two
adjacent particles inside the domain, one or none on the external boundary).
Let knk is the spring stiffness in normal direction and ktk is the spring stiffness in tangential direction on the
boundary between particles with common boundary CkC . Then in elastic region pnk ¼ knk ½u kn and ptk ¼ ktk ½u tn .

K ¼ fu 2 V ; ðpnþ Þk P pnk ¼ knk ½u kn ; if ðpnþ Þk < pnk then pnk ¼ 0;

ktk j½u kt j 6 ck on CkC ; k ¼ 1; . . . ; ng; ½u kt ¼ uk;c k;a
t  ut ;
P.P. Prochazka / Engineering Fracture Mechanics 71 (2004) 601–618 607
where uk;a t is the tangential displacement on the side k, ðpnþ Þ denotes the tensile strength, ck is the shear
strength, V is the set of displacements which fulfill the kinematical boundary conditions and the condition
(8). If pnk ¼ 0, the admissible displacements satisfying the essential boundary and contact conditions create a
cone and K valid for brittle or almost brittle material. If the material exhibits elasto-plastic behavior, then K
turns to:
k k k
K ¼ fu 2 V ; ðpnþ Þ P pnk ¼ knk ½u n ; if ðpnþ Þ < Pnk then pnk ¼ 0;
k k
ktk j½u t j 6 ck jðpnk Þ  pnk tan / on CkC ; k ¼ 1; . . . ; ng; ½u t ¼ uk;c k;a
t  ut ;

where / is the angle of internal friction, and pnk normal traction on the side k, j is generalized HeavisideÕs
function being equal to zero for positive argument and one otherwise. Here the sign convention is im-
portant: positive normal traction is tension.
k k
From the above defined spaces one can deduce that pnk , ½u n and ptk , ½u t behave linearly in between certain
limits, which are given by material nature of the body.
The total energy J of the system reads:
Xn Z Z Xn Z
1 k 2 k 2 k k k
J ðuÞ ¼ aðu; uÞ  fknk ð½u n Þ þ ktk ð½u t Þ g dC  bT u dX þ fðpnþ Þ ½u n þ ck ½u t g dC ð9Þ
2 k¼1 CC X k¼1 CC

T oun out 1 out oun
aðu; uÞ ¼ ðeÞ Ce dX0 ; e¼ ; ; þ ;
C0 oxn oxt 2 oxn oxt
where e is the strain tensor, C is the stiffness matrix of the particle, T denotes transposition, X0 is the sum of
sub domains X, i.e. of hexagonal elements, b is the volume weight vector.
Note that the spring stiffnesses knk and ktk act out a role of penalties. Recall that the problem can also be
formulated in terms of LagrangianÕs multipliers, [17], and then leads to mixed formulation. The latter case
is more suitable for small amount of boundary variables; the problem discussed in this paper decreases the
number of unknowns introducing the penalty parameters.

3. Statical PFC

In this section an idea of modeling the structure, say the earth body, by using statical version of PFC is
dealt with. Recall that the PFC is based on dynamical equilibrium; for slow movements of structures, which
appears, for example, in most of geotechnical problems, it seems to be better to employ statical equilibrium.
The earth mass is modeled in statical version by balls in 3D or disks in 2D in a similar way such as in the
dynamical version. The balls are connected by springs that relate forces and the appropriate difference of
displacements in the direction of the springs. The springs are considered either in normal and tangential
directions to the boundary of particles, or in the direction of coordinate axes x and y. In our case the statical
equilibrium has to be fulfilled all over every ball, and, moreover, at the contact points between adjacent
balls. The balls (disks) are considered to be rigid. Introduce coordinate system 0xyz in 3D. Then each disk
has six degrees of freedom (displacement ux in direction of x, displacement uy in direction y and uz in di-
rection z, and three rotations with respect to the three axes x, y and z). In what follows we restrict our
considerations to 2D for simplicity, generalization to 3D is straightforward. The movement of each disk is
described by two displacements ux , uy and rotation u.
The forces concentrated at one contact point of the adjacent disks obey contact conditions, which are
typical for soil in our study. The plastic behavior provided rigid disks are considered is imposed only by
forces brought about by spring stiffnesses and eigenparameters, e.g. plastic strain (displacement), or re-
laxation stresses (forces). When introducing some spring (of the shape of straight line, or more precise of
608 P.P. Prochazka / Engineering Fracture Mechanics 71 (2004) 601–618

abscissa) spanned between two points, its stiffness is k, the relation force F -displacement u in elastic case
may generally be written as (considering one-dimensional case):
F ¼ ku þ k; or F ¼ kðu  lÞ; ð10Þ

where respectively k is eigenstress (eigenforce), l is eigenstrain (eigendisplacement). Another possibility is to

drop out the eigenparameters and impose the non-linear conditions exceptionally on the springs. The ei-
genparameters enable one larger scale of physical laws, which can be used in the material models. This is
why they are mostly considered in the mechanical models.

3.1. Relations between disks

Let us have a set of disks describing the discontinuous medium positioned in coordinate system 0xy, see
Fig. 2. At the contact points (nodal points), the springs in the tangential and normal directions are in-
troduced, which have the stiffnesses kt in tangential direction to the boundary of the disk and kn in the
normal direction to the boundary at the interface nodal point connecting two adjacent disks.
Such a connection is described in more detail in Fig. 3 for three disks in mutual contact. In Fig. 3,
external (volume weight) forces Fi ; i ¼ 1; . . . ; n (n is the number of all disks), contact forces in normal and

Fig. 2. Structure of the disks.

Fig. 3. Forces between disks.

P.P. Prochazka / Engineering Fracture Mechanics 71 (2004) 601–618 609

tangential directions Nij , Qij , respectively, where i; j ¼ 1; 2, and reactions at the supports (created in this
case by a flat plate in 3D, or by a straight line in 2D) A, B, HA , HB are depicted. The first index denotes the
number of the current disk and the second index stands for the number of the adjacent disk. This notation
is kept in the following text. The main objective is to formulate the equations of equilibrium in each disk
i; i ¼ 1; . . . ; n and from this equilibrium to determine displacements of center ui , vi and rotations ui of each
disk. The connection with the adjacent disks is created by the quantities with indices i (the current disk) and
j (the adjacent disk).
In the sense of (10) the physical equations at each nodal point are formulated as:
   ij  ij   ij 
Nij k 0 Dn kn
¼ n ij þ ; ð11Þ
Qij 0 kt Dijt kijt

where kijn and kijt are, respectively, eigenforces in normal, and tangential directions. i, j are numbers of disks
in mutual contact and Dijn , Dijt are differences of displacements in normal and tangential directions, re-
spectively, between disk i and j, i.e. Dijn ¼ uijn  ujin , Dijt ¼ uijt  ujit .
Let the nodal point ij under consideration be deviated from x-direction by angle xij . Then the trans-
formation of forces to the 0xy coordinate system is written by:
Nx cos xij  sin xij Nij T Nij
ij ¼ ¼ T ; ð12Þ
Ny sin xij cos xij Qij ij Q

where Nxij , Nyij are forces in x, and y directions, Tij is transformation matrix and superscript T denotes
Recall that Tij is unitary, it means that T1 T
ij  Tij . Since the same equations hold for displacements, the
following forces–displacements relation holds valid as:
 ij   ij  ij   ij   ij ij
 ij   i 
Nx T kn 0 Dx kx kxx kxy Dx kx
¼ T þ ij ¼ þ ; ð13Þ
Nyij ij
0 ktij ij
Dy ky ij
kxy ij
kyy Dijy kiy

kxx ¼ knij cos2 xij þ ktij sin2 xij ; ij
kyy ¼ ktij cos2 xij þ knij sin2 xij ;
kxy ¼ ðknij  ktij Þ sin 2xij ; kix ¼ kijn cos xij  kijt sin xij ; ð14Þ
kiy ¼ kijn sin xij þ kijt cos xij ;

and Dijx , Dijy are differences of displacements in x and y directions, respectively in disks i and j, i.e.
Dijx ¼ uix  ujx , Dijy ¼ uiy  ujy .
A typical disk with springs introduced at nodes in x and in y directions and induced forces in the normal
and tangential directions is illustrated in Fig. 4.
If no rotations were considered, the above formulas would be valid without improvement and the
computation may start with (13). For each disk two degrees of freedom in 2D (two independent dis-
placements are unknown), and three DOF in 3D (three independent displacements are sought).
In the case of admitted rotations of disks, additional unknown angles ui describing the rotations of disks
have to be introduced. Recall that three DOF (two displacements ui , vi and one angle of rotation ui ) in 2D
are to be sought.
Let us focus on one typical disk. Its basic movements and their denotations are clear from Fig. 5.
The total displacement at any point on the boundary of disk in x direction, or y direction consists of two
kinds of displacements that are related to rotation (subscript rot), and translation (subscript tran), i.e.
uix ¼ uix;rot þ uix;tran ; uijy ¼ uiy;rot þ uiy;tran , see Fig. 5.
610 P.P. Prochazka / Engineering Fracture Mechanics 71 (2004) 601–618

Fig. 4. Starting model of one disk for next computation.

Fig. 5. Three basic movements of a disk.

Since 2D case is considered, the unknown quantities in each disk i are uix;tran , uiy;tran and ui . It only remains
to express the influence of rotation of each node.
From Fig. 5 it obviously holds:

uix;rot ¼ ri ½cosðxij þ ui Þ  cos xij ;

uiy;rot ¼ ri ½sinðxij þ ui Þ  sin xij ;

where xij is given for each node and ri is the radius of disk i.
The forces Nxij , and Nyij acting between disk i and disk j are then given by (13).
P.P. Prochazka / Engineering Fracture Mechanics 71 (2004) 601–618 611

3.2. Governing equations

Equation (13) can be rearranged in a more suitable way for program coding:
0 1 2 ij 30 i 1 0 i 1
Nxij kxx ij
kxy kxxij
ux kx
B Nyij C 6 kxy
ij ij
kyy kxyij
kyyij 7B i C
u B k C
B ji C ¼ 6 ij 7 B yC
þ B yC
@ Nx A 4 kxx kxyij ji
kxx ji 5@ j A
kxy ux @ kj A ;
Nyji kxyij
kyyij ji
kxy ji
kyy ujy kjy

and from (15):

0 i1 0 i 1 0 1
ux ux;tran ri ½cosðxij þ ui Þ  cos xij
B uiy C B uiy;tran C B ri ½sinðxij þ ui Þ  sin xij C
B jC¼B j C B C
@ ux A @ ux;tran A þ @ rj ½cosðxji þ uj Þ  cos xji A: ð17Þ
ujy ujy;tran rj ½sinðxji þ uj Þ  sin xji

The third unknown ui appears in the conditions of equilibrium in non-linear terms, namely in cosines
and sinus (recall that xij is the angle of deviation from x-axis of the point ij on the boundary of the disk i
being in contact with the disk j). In order to avoid very complicated and unreliable non-linear computation,
the load (for example volume weight) will be divided into increments, and in each increment the small
displacement (or more precisely small rotation) theory will be considered. Assuming a small enough in-
crements, small enough angle u also results and (17) is substituted by:
0 1 0 i 1 0 1
uix ux;tran ri ui sin xij
B uiy C B uiy;tran C B ri ui cos xij C
B jC¼B j C B C
@ ux A @ ux;tran A þ @ rj uj sin xji A: ð17aÞ
ujy ujy;tran rj uj cos xji

In each increment uix;tran , uiy;tran and ui are parts of the total values, so are the forces computed from these
movements. Another advantage of this incremental process is a possibility to test contact conditions as
described in the following section.
An additional condition is necessary to complete the equations for three unknowns in each disk. This is
the moment condition with respect, say, to the center of disk under study:

Qij ri ¼ 0; ð18Þ

where Mi , is the number of nodes on the boundary of disk i. Since ri is constant in each disk i, it may be
dropped out and three conditions in the disk i are obtained as:

Mi X
Nxij ¼ Fxi ; Nxij ¼ Fyi ;
j¼1 j¼1
Mi X
Qijt ¼ Nxij sin x þ ij
Nyij ij
cos x ¼ 0;
j¼1 j¼1

where Fxi and Fyi are volume weight forces. If only gravitation is considered, the denotation Fi  Fyi , and
assumption Fxi  0 can be used, such as in Fig. 3.
612 P.P. Prochazka / Engineering Fracture Mechanics 71 (2004) 601–618

After introducing boundary conditions and eigenforces, the Eq. (19) create an algebraic system for
unknown displacements uix;trans , uiy;trans , and rotations ui of the disks. Solving it, the forces can be determined
from (16). Note that when properly stated, the system (19) has unique elastic solution, providing the in-
cremental formulation (17a) is assumed, i.e. small displacement theory may be employed.

3.3. Interfacial conditions

In geotechnics and mining engineering it is well known that the material behavior of the soil mass is not
elastic, but exhibits either plastic behavior, or localized damage, or both of them. In contrary to the case of
free hexagons, the plastic behavior have to be employed only for spring properties. For the sake of com-
pleteness an example of such properties is discussed here.
Consider two disks being connected by the above-described system of springs. For the sake of simplicity
we concentrate our attention only on 2D case again.
Using the formula (11), the normal forces Nij and tangential forces Qij can be obtained for each linear
state. Vector transformation of coordinates applied to (17) provides differences of displacements Dijn and Dijt .
In wide range of problems classical plastic laws of deformation and stress state, e.g. elasto-plastic law, is
used as:

2 2
Nij2 þ 4Q2ij 6 k02 ; i:e: ðknij Dijn Þ þ 4ðktij Dijt Þ 6 k02 ; ð20Þ

where k0 is given positive number. Here Dijn , Dijt are given from previous step of iteration. Dijn is mostly
bounded by some value, which cannot be exceeded. Moreover, knij changes with Dijn . Mostly parabolic rule is
applied as:

knij ¼ kn0 ð1 þ Dijn Þs ; ð21Þ

where s is an exponent to be stated from laboratory tests, as well as the starting value of kn0 . Other rules can
be applied, such as arctan relation. In the case of violating conditions (20), new kt may be determined.
While the value of knij is taken from (21), ktij is restricted by (20). The value of Nij increases non-linearly and
when the tensile strength is reached, the crack is assumed at the point ij and the spring is suddenly removed.
In practical examples the spring in normal direction, which is in tension is removed gradually to stabilize
the convergence and speed up the iteration process.
On the other hand, at this point a ‘‘penetration’’ of one disk into the adjacent is fully permitted. The
damage is attained, when violating Mohr–Coulomb hypothesis or tensile strength is reached. In this case it
means that

(a) jQij j 6  Nij tan / þ cjðNij Þ, where c is the shear strength (cohesion), j is Heaviside function, and / is
the angle of internal friction,
(b) Nij 6 Nijþ , where Nijþ is the tensile strength.

When the condition a) is not fulfilled, then ‘‘cut’’ of Qij is supposed: Qij ¼ ½Nij tan / þ cjðNij Þ sign Duijt .
Note that more complicated rules may be imposed. For example, both internal parameters, angle of in-
ternal friction and shear strength may change with the values of Qij .
In the case of violation of the condition (b), a local disconnection (debond) occurs and the spring is
removed again. In this case it is not due to local cracking but because of disconnecting of the disks having
originally been in contact at the point ij.
P.P. Prochazka / Engineering Fracture Mechanics 71 (2004) 601–618 613

4. Examples

Fig. 6 shows the mesh of particles in both free hexagonal element method with internal radius of 0.25 m.
This mesh is also used for disks in such a way that centers coincide with centers of hexagons and the radius
is also 0.25 m. The number of particles is 1532. The lower part with the free face represents the coal seam,
the upper part is the rock. The coal seam is 4.75 m high, such as in the experimental model, [23]. Along the
vertical and bottom external boundaries normal displacements are restricted and there is no friction. The
upper part of the body is subject to the loading of overburden given by the depth of the coalmine.
The numerical procedure for solving the problems is over-relaxation method with improvement of
unknowns in the sense of the contact conditions between, adjacent disks. First, fixed contacts are con-
sidered, i.e. high penalty coefficients are introduced. After 1000 iteration steps the contact conditions are
taken into account and approximately after another 1500 iteration steps the procedure terminates. The
termination obeys the rule of Euklidean norm error imposed to the change of displacements of centers of
the particles (in both methods). All these values are considered for the examples discussed in what follows.
Note that when using statical PFC, the final load is divided into 10 equal increments of volume weight to
suppress the influence of possibly large rotations. Since the elastic–plastic law is applied to free hexagons
solution, the structure is loaded from the beginning by full loading.
Results from both hexagonal method and statical PFC are compared and the experimental measure-
ments that are fully described in [23] are also taken into consideration.
The examples are focused on the problem of extrusion of rock (or coal) mass from the rock body.
Material properties of the hexagons are: E ¼ 500 GPa, G ¼ 150 GPa, the shear strength c ¼ 1 MPa and
the tensile strength pnþ ¼ 100 kPa. These values are valid for the rock. The lower part with the free face
describes the coal; its material properties E and G are 10 times lower than that of the rock, but the shear
strength and the tensile strength change. The load is given by the overburden. The volume weight c ¼ 25
kN/m3 , the depth of the opening is 1000 m.
The disks are connected by springs with the following material properties: knij ¼ 500 GN/m and ktij ¼ 150
GN/m, the shear strength c ¼ 1 MN/m and the tensile strength pnþ ¼ 100 kN/m for the rock. For the coal it
holds: knij ¼ 50 GN/m and ktij ¼ 15 GN/m, the shear strength and the tensile strength change in compliance
with the hexagons.
In Fig. 7 the extrusion of particles is observed in the front part, while next to the face (approximately 3
m) the particles move backwards and stabilize the coal seam. The movements of the rock and coal with
these material properties cause the opening stability relatively safe. Some particles at the face of seam are
disconnected, but the direction of vectors shows still stabilized influence of the overburden. Fig. 8 shows the
same arrangement of the system shown in Fig. 7, but statical PFC is used in this case. The displacements at

Fig. 6. Setting of the hexagonal elements and disks.

614 P.P. Prochazka / Engineering Fracture Mechanics 71 (2004) 601–618

Fig. 7. Movements for pnþ ¼ 10 kPa, c ¼ 150 kPa.

Fig. 8. Disk movements for pnþ ¼ 10 kPa, c ¼ 150 kPa.

the face are principally smaller, but the triangular wedge is created in a similar manner, such as by
hexagons. As the process of cracking is influenced rather by contact conditions than by the springs, the
statical PFC is obviously less reliable then the hexagon method.
The most dangerous case is depicted in Fig. 9 for hexagons and in Fig. 10 for disks. The triangular wedge
is disconnected into two parts, extrusion of particles is now obvious. Similar pictures are shown in Figs. 11
and 12.
It can be proved that for the tensile strength of 3 kPa, the shear strength higher than 300 kPa does not
principally change the movement configuration. Similar conclusion is valid also for other relations pþ –c.
A wide scale of computations has been carried out to get the relation pnþ –c at bumps occurrence, which is
determined as singular solution of the problem. The relation is shown in Fig. 13. The numerical results are

Fig. 9. Movements for pnþ ¼ 5 kPa, c ¼ 150 kPa.

P.P. Prochazka / Engineering Fracture Mechanics 71 (2004) 601–618 615

Fig. 10. Disk movements for pnþ ¼ 5 kPa, c ¼ 150 kPa.

Fig. 11. Movements for pnþ ¼ 3 kPa, c ¼ 300 kPa.

Fig. 12. Disk movements for pnþ ¼ 3 kPa, c ¼ 300 kPa.

Fig. 13. Relation pnþ –c for bumps occurrence, fixed E a G.

616 P.P. Prochazka / Engineering Fracture Mechanics 71 (2004) 601–618

Fig. 14. Comparison experiment––computation for loading 2.5 MPa.

in a reasonable agreement with experiments that has been carried out in KloknerÕs Institute of CTU Prague
on a scale model, [23]. Material properties of scale experiments obeyed BuckinghamÕs theorem, see [20]. The
scale was approximately 1:250, it meant that the opening was 4.75 m high. This height was also applied in
the mathematical models for comparison studies.
The comparison follows from Fig. 14. The deviations at the left hand side of the picture are probably
caused by relatively rough mesh of particles, right hand side can be influenced by the boundary conditions
(in the numerical model obvious cracking occurs, i.e. drop of stresses has to be expected).

5. Conclusions

The paper is devoted to the numerical description of the behavior of an open rock that occurs during
longwall mining in coalmines, for example. The main problem is longwall instability and extrusion of rock
mass into an open space. This effect is mostly referred to as bumps, or rock bursts. In order for bumps to
occur, the rock has to posses certain special properties. In this study we consider brittle, or almost brittle
coal seam, which is treated theoretically, and experiments that have been carried out in the past posses the
same quality. The results from both models seem to be reasonable. Combination of experimental and
mathematical models appears very prospective for study of similar problems. Both methods allow studying
the problems as time dependent. They make it possible to develop cracks during bump initiation, and
therefore describe the problem, which is very closely to the reality.
As a numerical tool the free hexagonal element method and the statical particle flow code are numerical
method serving for study of bumps occurrence and comparison with results from experiments. The stiffness
of the hexagons is created by the BEM. The generalized HookeÕs law is used, involving eigenparameters,
which can represent various material phenomena. Each element is considered to behave elastically (or rigid
in statical PFC) and contact conditions turn the problem to non-linear. Using iterative procedure, a very
fast solution is obtained. If the solution converges, process of deterioration of the coal seam and the rock
mass can be observed and evaluated.
One example is presented starting with accumulation of energy due to external loading in a great depth
of mine. The system of extrusion of coal particles describes the way it behaves and reasons which locally
affect the movements of coal, or sometimes also rock particles.
As the process of cracking is rather influenced by contact conditions then by the springs, the statical PFC
is less reliable then the hexagon method. This assertion is verified from comparative figures.
Because of its simplicity in comparison to the free hexagonal element method, statical PFC can be more
applicable in dynamical problems than the hexagons. From experience with classical PFC and its statical
P.P. Prochazka / Engineering Fracture Mechanics 71 (2004) 601–618 617

version, dynamic equilibrium applied to the statical PFC concept is more reliable then that of the classical
PFC, where only short range of material parameters delivers reasonable results. If one deviates from this
range, the results are very unpredictable, a concrete structure can burst out in the space, for example.


This research was supported by Grant Agency of the CAS (grant number A2/19001/00).

Appendix A

In Fig. 15, basic local coordinate system is shown to enable one the calculation of integrals needed. The
local coordinate system can always be introduced in such a way that the boundary element is a function.
This can be applied even to higher order polynomials describing the boundary element geometry.
Z xB Z xB qffiffiffiffiffiffiffiffiffiffiffiffiffiffi  
1 b
log r dx1 ¼ log x21 þ a2 dx1 ¼ b logða2 þ b2 Þ  2b þ 2a arctan ; ðA:1Þ
xA xA 2 a
Z xB Z xB h x   x i
xi xj x21 A B
dx1 ¼ dx 1 ¼ a b þ arctan  arctan
xA r2 xA x21 þ a2 a a
Z xB
x1 a x2B þ a2
¼ dx 1 ¼ a log ðA:2Þ
xA x21 þ a2 x2A þ a2
Z xB   x 2  x 2 
a2 B A
¼ dx1 ¼ a arctan  arctan ;
xA x21 þ a2 a a
Z xB Z xB
xi x1 x 2 þ a2
dx1 ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffi dx1 ¼ log B2
xA r2 xA x21 þ a2 x A þ a2
Z xB x  x  ðA:3Þ
a B A
¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffi dx1 ¼ arctan  arctan ;
xA x21 þ a2 a a
Z xB x  x 
dx1 1 B A
¼ arctan  arctan ; ðA:4Þ
xA r a a a

Fig. 15. Geometry for calculating boundary integrals.

618 P.P. Prochazka / Engineering Fracture Mechanics 71 (2004) 601–618
Z xB Z xB
xi xj xk nj xi xk a
dx1 ¼ dx1 ¼ sec ðA:2Þ: ðA:5Þ
xA r2 xA r2


[1] Brebbia CA, Telles JFC, Wrobel LC. Boundary element techniques. Berlin: Springer-Verlag; 1984.
[2] Cundall PA. A computer model for simulation progressive large scale movements of blocky rock systems. In: Symposium of the
International Society of Rock Mechanics. 1971. p. 132–50.
[3] Cundall PA, Strack ODL. A discrete numerical model for granular assemblies. Geotechnique 1979:47–65.
[4] Desai CS, A consistent finite element technique for work-softening behavior. In: Oden JT, editor. International Conference on
Computational Methods in Nonlinear Mechanics, University of Texas, Austin. 1974.
[5] Desai CS. Constitutive modeling using the disturbed state concept [chapter 8] In: Mulhaus H, editor. Continuum models for
materials with microstructure. UK: John Wiley and Sons; 1994.
[6] Duvaut GJ, Lions J-L. Les inequations en mecanique et en physique. Paris: Dunod; 1972.
[7] Dvorak GJ, Proch azka P. Thick-walled composite cylinders with optimal fiber prestress. Composites, Part B 1996;27B:643–9.
[8] Dvorak GJ, Proch azka P, Srinivas S. Design and fabrication of submerged cylindrical laminates Part I and Part II. Int J Solids
Struct 1999;1:1248–95.
[9] Elices M, Guinea GV, G omez J, Planas J. The cohesive zone model: advantages, limitations and challenges. Engng Fract Mech
[10] Haramy KY, Morgan TA, DeWaele RE. A method for estimating western coal strengths from point load tests on irregular bumps.
In: Second conference on ground control in mining, West Virginia University, July 19–21, 1982. p. 123–36.
[11] Haramy KY, Magers JA, McDonnell JP. Mining under strong roof. In: 7th International Conference on Ground Control in
Mining Bureau of Mines, Denver, USA. 1992. p. 179–94.
[12] Harami KY, Brady BT. A methodology to determine in situ rock mass failure. Internal report of Bureau of Mines, Denver CO,
USA, 1995.
[13] Kachanov LM. Introduction to continuum damage mechanics. Dordrecht, Netherlands: Martinus Nijhoff Publishers; 1986.
[14] Kuch R, Lippmann H, Zhang J. Simulating coal mine bumps with model material. In: Gibowitz, Lasocki, editors. Rockbursts and
seismicity in mines. Rotterdam: Balkema; 1997. p. 23–5.
[15] Moreau JJ. Some numerical methods in multibody dynamics: Application to granular materials. Eur J Mech Solids 1994;13(4):93–
[16] Onck P, van der Giessen E. Growth of an initially sharp crack by grain boundary cavitation. JMPS 1999;47:99–139.
[17] Proch azka P, S ejnoha M. Development of debond region of lag model. Comput Struct 1995;55(2):249–60.
[18] Proch azka P, Trckova J. Coupled modeling of concrete tunnel lining. Our World in Concrete and Structures, Singapore, 2000.
p. 215–24.
[19] Silva MJ, Gibson LJ. The effect of non-periodic microstructure and defects on the compressive strength of two-dimensional
cellular solids. Int J Mech Sci 1997;39(5):549–63.
[20] Stilborg B, Stephensson O, Swan G. Three-dimensional physical model technology applicable to the scaling of underground
structures. In: 40th International Conference on Rock Mechanics, vol. 2. Moiitreux. 1979. p. 655–62.
[21] Triantafillidis N, Schraad MW. Failure in aluminium honeycombs. JMPS 1999;47:1093–124.
[22] Ujihara M, Higuchi K, Nabeya H. Scale model studies and theoretical considerations on the mechanism of coal and gas outbursts.
In: International Conference on OUTBURSTS, Sydney. 1986. p. 121–28.
[23] Vacek J, Proch azka P. Rock bumps occurrence during mining. In: Brebbia CA, editor. CMEM. Spain: Alicante; 2001. p. 125–34.
[24] Zienkiewicz OC. The Finite Element Method. New York, NY: McGraw-Hill Book Company; 1977.