Int. Jr. Rock Mech. Min. Sci. & Geomech. Abstr. Vol. 25, No. 3, pp.

117-125, 1988 0148-9062/88 $3.00 + 0.00

Printed in Great Britain Pergamon Press pie

Formulation of a Three-dimensional
Distinct Element Model Part II.
Mechanical Calculations for Motion
and Interaction of a System Composed
of Many Polyhedral Blocks
R. HART*
P. A. C U N D A L L t
J. LEMOS*
A three-dimensional formulation of the distinct element method is embodied
in computer program 3DEC, which has been adapted to run on a personal
computer. This formulation is based on a dynamic (time domain) solution
algorithm which solves the equations of motion of a three-dimensional block
system by an explicit finite-difference scheme. The scheme is well suited in rock
mechanics studies to determine if a discontinuous rock mass will fail under a
given set of applied loads. Part H of this paper describes the mechanical
calculations of rigid body motion and block interaction. Part H also presents
the technique used in 3DEC to generate a three-dimensional model for rock
mechanics analysis. Several examples are provided which demonstrate the
capabilities of 3DEC and the size of problem that can be analyzed on a
personal computer. The examples emphasize the use of microcomputer graph-
ics to assess model results.

INTRODUCTION discontinuous rock masses. The important points of the

It is often difficult to define a particular numerical
method and distinguish it from other methods which
may be similar. For example, discontinuous bodies have
been modelled numerically by Burman [1], Rodriguez-
Ortiz [2], Chappell [3], Goodman and Shi [4] and Key [5]
among others. Numerical techniques in geomechanics
tend to overlap, especially when modifications are made
that borrow features from other methods that are sup-
posedly different. The "distinct element method" is not
very different from other techniques, particularly when
all existing variations are considered, but three distinct
features arc usually associated with it. First, bodies can
undergo large rotation and large displacements relative
to one another; second, interaction forces between
bodies arise from changes in their relative geometrical
configuration; and, third, the solution scheme is explicit
in time.
Because of these features, the distinct element method
is particularly well suited to investigate problems in rock
mechanics which address the question of stability of
* Ituea Consulting Group Inc., Minneapolis, MN 55414, U.S.A.
f Department of Civil and Mineral Engineering, University of
Minnesota, Minneapolis, MN 55455-0220, U.S.A.
117
distinct element formulation as it relates to rock mass
stability are as follows.
(1) Both stability and instability are modelled. When
a net force exists on the block, it accelerates and
moves to a new position. If the forces balance, then
either the system remains at rest, or it moves with
constant velocity.
(2) Forces arise between two blocks when the blocks
intersect. Normally, the overlap is small in relation
to block dimensions.
(3) The calculation marches from one state to another
in small increments of time. The "final solution"
may be equilibrium or it may be a state of
continuing motion.
A principal criticism of the method in the rock
mechanics community is that the formulation has only
been developed for two-dimensional systems. In general,
a two-dimensional model cannot fully represent the
behaviour of a discontinuous rock.
Part II of this paper describes the formulation of the
mechanical calculations of the distinct element method
in three dimensions. This discussion is based upon work
presented previously by Cundall and Strack [6] and
t 18 HART et al.: 3-D D E M - - P A R T II

Cundall and Hart [7]. The description of the formulation of the discontinuity within a plane. For a persistence
is given with specific application to stability analysis in equal to 1.0, all blocks along the joint plane are split; for
rock mechanics and only addresses the mechanical cal- a persistence equal to 0.5, half of the blocks are split, on
culation for rigid body motion and block interaction. average.
This description of motion is a sufficient representation The joint generation process is conducted in 3DEC by
for stability studies in which the applied stress state is issuing the description for the joint set in the command
low compared to the intact rock strength and in which mode and viewing the results in the graphics screen
motion is concentrated along structural features. mode. The user can switch back and forth between
The three-dimensional program also accounts for command and screen modes to view the result of the
block deformability and failure of intact material. In this joint set command directly after the command is given.
formulation, each polyhedral block is subdivided into an Alternately, joints can be generated individually in the
internal finite difference mesh consisting of constant screen mode by single strokes made directly from the
strain tetrahedral elements. The solution is an explicit, keyboard. Discontinuous joints can be created manually
large-strain one, with elastic and elasto-plastic models by first "hiding" blocks, in either the command mode or
for the block material. A more complete description of screen mode, and then generating joints. Hidden blocks
the deformable block logic will be presented in a sub- are not split by the joint generator command and can be
sequent paper, but an analogous formulation for two- made visible again after generation is completed. An
dimensional blocks is given by Lemos [8]. The contact EGA graphics display of a model of a blocky system
detection scheme for three-dimensional rigid and defor- created by the joint generator is shown in Fig. 1. The
mable blocks is discussed in Part I of this paper. different colours in this figure distinguish the presence of
The three-dimensional formulation described here is the joints in the model. The menu shown to the right of
embodied in a computer program called 3DEC, which is the model displays the keystrokes operating in the screen
written specifically to run on a personal computer. mode.
3DEC makes liberal use of interactive microcomputer A simple example of the application of the joint
graphics in three dimensions to assist in generation of generator in command mode is demonstrated in Fig. 2.
the model and presentation of results. A description of A blocky system is created from the following sets of
model generation is provided to demonstrate the joint commands.
application of interactive graphics for rock mechanics
Set 1 2 3
analysis.
Dip angle
Mean 45 45 45
MODEL GENERATION
Deviation 0 0 0
A prerequisite of a three-dimensional model of discon- Dip direction
tinuous rock is the ability to generate joint patterns that Mean 220 70 70
resemble those in the natural rock. Because of the Deviation 0 0 0
variability of rock structure, joint patterns must be Spacing
described in statistical terms. Lemos [8] describes a Mean 0.25 0.5 0.5
procedure proposed by Cundall for generation of joint Deviation 0.1 0 0
sets in a two-dimensional distinct element model. The Persistence 1.0 0.5 0.5
extension to three dimensions is described herein. Number 15 15 15
The 3DEC code uses a joint set generator which Origin (0, 0, 0) (0, 0, 0) (0, 0, 0)
permits the creation and superposition of several joint
sets in order to create a blocky structure. Sets ranging Discontinuous joints can also be made by "joining"
from single faults to a system of many joints can be blocks. This technique, described in Part I ofthis paper,
created. In this model, a joint set is characterized by six can be used to join blocks in such a fashion as to create
parameters: discontinuous joints and internal cracks within the
model.
(1) dip direction; The joint generator is quite general and allows the user
(2) dip angle; to "cut up" the model to represent the defined structural
(3) spacing between joints in the set;
features. Also, joints are identified individually so that
(4) number of joints in the set;
different constitutive behaviour and material properties
(5) location point for one joint in the joint set; and
can be specified for each joint or joint set.
(6) persistence (i.e. probability that any given block
In addition to splitting blocks with the joint generator,
lying in the path of a joint will be split).
blocks can be created separately. For example, a block
The dip direction, dip angle and spacing are each representing a foundation footing can be created and
defined by a mean value and a random deviation from located on the top of the model shown in Fig. 2.
the mean, according to a given probability distribution. Finally, holes can be created anywhere in the model.
At present, a uniform probability density function is An automatic tunnel generator is used to create single or
used in 3DEC. The persistence parameter creates discon- intersecting tunnels, drifts, etc. With this generator, the
tinuous joints in the model and describes the areal extent user defines faces at the ends of the tunnel by a sequence
 HART et al.:
of vertices. Straight lines connecting the two faces define
the tunnel boundary. Any convex shape of the tunnel
face can be prescribed by an arbitrary number of
vertices. The faces may be positioned either outside the
model boundary or inside the model. Figure 2 illustrates
a circular tunnel in the block system. Blocks created to
facilitate tunnel generation, involving fictitious joints,
are joined automatically by the tunnel generator. Joined
blocks are shown in the same colour on Fig. 2.
This example contains 159 blocks and requires 800
kbytes memory on a personal computer. If a co-
processor board is used in the computer (e.g. the DSI-
780 co-processor board with 4 mbyte memory), then up
to approximately 1000 blocks can be modelled. The run
time for a model consisting of 1000 blocks, using a
DSI-780 board, is roughly 48 hr.
CALCULATION CYCLE
3DEC is based upon a dynamic (time domain) algor-
ithm which solves the equations of motion of the block
system by an explicit finite-difference method. A solution
scheme based upon the equations of motion is demon-
strated by Cundall [9] to be better suited to indicate
potential failure modes of discontinuum systems than
schemes which disregard velocities and inertial forces
(e.g. successive over-relaxation). At each time-step, the
law of motion and the constitutive equations are applied.
In the case of rigid blocks, the latter consist only of the
contact force--displacement relations. The integration of
the law of motion provides the new block positions and,
therefore, the contact-displacement increments (or ve-
locities). The contact force--displacement law is then
used to obtain the new contact forces, which are to be
applied to the blocks in the next timestep. The cycle of
mechanical calculations is illustrated below; the com-
ponents of the cycle are described in the following
sections.
CONTACT FORCE,
UPDATE
BLOCK CENTROID RELATIVE CONTACT
FORCES VELOCITIES
T contacts is straightforward, as noted in Part I, and
, BLOCK MOTION / contact which is a convex, simply-connected polygon.
UPDATE
Contact force update
Part I of this paper described the procedure for
updating the geometric parameters associated with con-
tact between blocks. The unit normal to the common-
plane is taken as the contact normal, ni (pointing from
block A to block B.) The reference point on the
common-plane, C;, defines the line of action of the
contact force and is taken as the contact location.
3-D DEM--PART II 119
The contact velocity (defined as the velocity of block
B relative to block A at the contact location) is calcu-
lated as:
V, = ~p + eukco~(Ck -- B,) - ~ - e~co~(C, - Ak), (1)
where A; and B~ are the position vectors of the centroids
of blocks A and B; ~A and .~p are the translation velocity
vectors of the centroids of blocks A and B; a~A and cop
are the corresponding angular velocity vectors; %, is the
permutation tensor; and indices i, j, k take the values 1
to 3, and denote components of a vector or tensor in the
global co-ordinate system. (The summation convention
applies for repeated indices.)
The contact displacement increment vector is
calculated as:
AU, = V, At, (2)
which can be resolved into normal and shear com-
ponents along the common-plane. The normal displace-
ment increment is then given by
A U n = AU, n,, (3)
and the shear displacement increment vector by
AU~ = A U , : AU:n,nj. (4)
The unit normal to the common-plane, ni, is updated
every timestep. In order to account for the incremental
rotation of the common-plane, the vector representing
the existing shear force (in global co-ordinates), ~ , must
be corrected as:
~ , = F ~ - e~ ek,,~n~an~, (5)
where n~ d is the old unit normal to the common-plane.
The contact displacement increments are used to
calculate the elastic force increments. The normal force
increment, taking compressive force as positive, is
AF" = - K . A U " A (6)
and the shear force vector increment is,
A ~ = - K , AU~Ao (7)
where K, and Ks are the joint stiffnesses
[stress/displacement], and Ac is the contact area. Contact
stresses are then calculated directly from the contact
forces and area.
The calculation of the contact area for face-to-face
involves only the calculation of the common area of
For contacts other than face-to-face the area of contact
is more difficult to define. At present a minimum value
of the contact area is assumed in equations (6) and (7)
to ensure a lower bound value for contact stiffness. The
minimum area is calculated as a small fraction (e.g. 1%)
of the average face area, but can be redefined by the user.
The total normal force and shear force vectors are
updated
F*,=F ~ + A P , (8)
F ~ , = ~ + AFt, (9)
120 HART et al.: 3-D D E M - - P A R T II

and adjusted according to the contact constitutive re-
lations. At present, 3DEC employs a Coulomb friction
law with a limited tensile strength. If the ioint tensile
strength, Tis exceeded (i.e. F" < - TAt), then both shear
and normal forces are set to zero. Otherwise, the max-
imum shear force is calculated as
s -_ cAc + F n tan 4),
Fma~
where c is the joint cohesion [stress], and ~bis the friction
angle.
If the absolute value of the shear force, F', given by
F' = (F~F~) '/2
is greater than F~,~ (i.e. if F ' > F~a,), then the shear
force is reduced to the limiting value--i.e.
S.__ S S term, the equations become:
F,..-F, [(Fmax)/F S ].
The contact forces are then added to the forces and
moments acting on the centroids of both blocks. The
contact force vector, which represents the action of
block A on block B, is given by
F , = - (F~n, + F,~).
The force and moment sums of block A are therefore
updated as
FA,=F A - F~,
M A , = M A -- e ~ ( C / - A/)Fk.
Similarly, for block B,
F~'.=F~ + F~.
M ~ . = M ~ + et, (C~- E)F,.
previously, blocks can be discretized into a tetrahedral
zone mesh, and the inertial terms are then consistently
represented.
Rigid block models are more appropriate for quasi-
static analyses and, in these cases, the rotational equa-
tions of motion can be simplified. Because velocities are
small, the non-linear term in the above equations can be
(10)
dropped, uncoupling the equations. Also, because the
inertial forces are small compared with the total forces
applied to the blocks, an accurate representation of the
inertia tensor is not essential. In 3DEC, therefore, only
(I 1) an approximate moment of inertia, L is calculated based
upon the average distance from the centroid to vertices
of the block. This allows the above equations to be
referred to the global axes. Inserting a viscous damping
(12)
M,
h, + ~co,.= --I' (20)
where the velocities co~and the total torque M; are now
referred to the global axis.
A centred finite-difference procedure is used to inte-
(13)
grate the equations of motion. The follo~ng expressions
describe the translational and rotational velocities at
time t in terms of the values at mid-intervals.
(14) 1 [ (,-~)(t +At)]
(15)
(16)
co 0
(17)

Block motion update The accelerations are calculated as

The equations of translational motion for a single
block can be expressed as 1 '+7) '-

y, + ~Yc, = (F,)/m + g,, (18)

where gi denotes the velocity of the block centroid; = is
the viscous (mass-proportional) damping constant; Fl is 1 co ' + 7 ) _co ' - 7
obj.(t) = ~-~ (22)
the sum of forces acting on the block (from the block
contacts and applied external forces), m is the block
mass; and g, is the gravity acceleration vector. Inserting these expressions in the equations of trans-
The rotational motion of an undamped rigid body is lational and rotational motion, equations (I8) and (20),
described by Euler's equations, in which the motion is respectively, and solving for the velocities at time
referred to the principal axes of inertia of the body: [t + (At/2)] results in
I, o5, + (13 --/2)0-)3('02 = M,,
= +/-~+g~ At D2,
I2(./)2 @ (I, --/3)(.O1(.O 3 m M 2 , g!'+~) [D,.~!'-~)Lm[-A ] ]
/3o)3 + (I2 - I, )co2co, = M3, (19)
where I,,/2 and 13 are the principal moments of inertia (,:-')
of the block; co,, co2, co3 are the angular velocities about coi 2 = D, oJ + At D2, (23)
the principal axes; and M,, M2, M3 are the components
of the torque applied to the block referred to the
principal axes. An accurate dynamic analysis of a blocky where
system involves the solution of the above equations. D, = 1 - (aAt/2),
However, dynamic analysis of rock mechanics systems
requires the use of deformable blocks [8]. As mentioned D 2 = 1/[1 + (aAt/2)].
 Fig. 1. Example of a system of rock blocks. (Menu is displayed in screen mode.)
Fig. 2. Blocky system is created with three joint set commands; then, circular tunnel is "excavated" with the TUNNEL command.
Fig. 3. Sliding wedge model.
Fig. 4. Falling wedge model.

121
 .':J

Fig. 5. Tunnel in faulted rock. (Front blocks are hidden for better viewing of white sliding wedge.)
Fig. 6. Rock slope model.
Fig. 7. Rock slope failure in progress.
Fig. 8. Vertical section through wedge in slope showing displacement vectors.
Fig. 9. Vertical section through right side of slope showing displacement vectors.

122
 HART et al.:
The increments of translation and rotation are given
by
Axl = .21[t + (At/2)] At,
A0, = co,[t + (At/2)] At.
The position of the block centroid is updated as
xl '+~') = xl')+ Axe.
The new locations of the block vertices are given by
X v(t+t~t) ---- X v(`) "4" A N i q'- eukAOj[xVk (t) - - x~t)].
For groups of joined blocks, the motion calculations
are only performed for the master block, whose mass,
moment of inertia and centroid position are modified to
represent the group of blocks. Once the motion of the
master block is determined, the new position of the
centroid and vertices of the slave blocks are calculated
by expressions similar to equation (26).
The force and moment sums, F~ and M~, for all blocks
are reset to zero every cycle after the block motion
update is completed.
Numerical stability
The central difference algorithm is only numerically
stable if the timestep At is less than the critical timestep.
An estimate of the critical timestep is calculated in the
program by analogy to a single degree-of-freedom sys-
tem. The smallest block mass in the problem, Mini,, and
the largest normal or shear contact stiffness, Kraal, are
used to calculate the timestep as
At = F R A C " 2 [2--~max
. M m 1i n~/2
F R A C is a user-defined factor which accounts for the
fact that a block may be in contact with several blocks.
A value of FRAC equal to 0.1 is usually sufficient to
ensure numerical stability.
For quasi-static processes, inertial forces usually have
no effect on the solution as long as they remain small
compared with the other forces in the system. Therefore,
inertial masses can be scaled to improve the convergence
time (gravitational masses are not affected). A procedure
that is effective in many cases is to assign the same mass
to all blocks.
Damping
For quasi-static analysis viscous damping is com-
monly used in the equations of motion to achieve
convergence to a steady-state solution--i.e, either
equilibrium or steady-state failure (collapse). Viscous
damping introduces body forces that retard steady-state
collapse and, in extreme cases may influence the mode
of failure. Cundall [9] and [10] describes the use of
adaptive damping as an effective method to overcome
this difficulty. This approach is used in 3DEC. Adaptive
damping continuously adjusts the viscosity such that the
power absorbed by damping is a constant proportion of
the rate of change of kinetic energy in the system.
Therefore, as the kinetic energy approaches zero, the
damping power also tends to zero.
3-D DEM--PART II 123
EXAMPLES
Four examples are presented to demonstrate the appli-
cation of 3DEC to rock mechanics problems. The first
(24) two examples indicate the accuracy of the code by
comparing model results for single wedge failure to
closed-form solutions. The third example, the failure of
(25) a wedge in the roof of a square tunnel, illustrates the
model generation facilities in 3DEC. The final example,
a rock slope collapse, indicates the potential for multiple
(26) modes of failure influencing the collapse. All the
examples illustrate the use of computer graphics to
present the three-dimensional results.
Example I
The first test for the code 3DEC is the analysis of
wedge failure resulting from sliding along two inter-
secting planar surfaces. A single wedge is formed by two
structural features defined by:
A: dip = 120 dip direction = 130
B: dip = 60 dip direction = 220
The same mechanical properties are assumed for both
planes. Classical wedge stability analysis presented in
Hock and Bray [11] gives a friction coefficient required
for stability equal to 0.66 for this wedge geometry.
The 3DEC model shown in Fig. 3 consists of four rigid
blocks. Three of the blocks are fixed, and only the
isolated wedge is allowed to move under gravity. The
mechanical properties selected for this model are:
(27) density: p = 2000 kg/m 3
joint shear and normal stiffness: K, = Ks = 10 MPa/m
gravity acceleration: 10 m/sec 2
The wedge is first allowed to consolidate under its weight
by setting the friction on both planes to a high value
and applying adaptive damping. Then, friction is
reduced until failure occurs. Figure 3 shows the failure
mechanism in progress.
Two of the wedge faces contact with faces of adjacent
blocks. The edge at the back of the wedge contacts with
an edge of the rear block. In a first run, this edge--edge
contact is assumed to have zero stiffness and therefore
to carry no load. This assumption is implied in the
analytical solution. A friction coefficient of 0.658 is
calculated for stability in this case (within 1% of the
analytical solution).
In a second run, the edge-edge contact is assumed to
have the same stiffness as the face-face contacts. Joint
shear and normal stiffness of 10 MN/m are assigned to
all three contacts. In this case, the friction coefficient
required for stability is 0.70.
Example 2
The second problem for 3DEC concerns the stability
of a rock wedge in the roof of an excavation. This
problem takes into account the stabilizing effect of the
in situ stresses and also shows the influence of joint
stiffness. A closed-form solution for the friction
124 HART et al.: 3-D DEM--PART II
coefficient required for stability of the wedge is given by
Goodman et al. [12].
The wedge shown in Fig. 4 is defined by three planes
dipping at 60 and with dip directions of 30, 150 and
270 . The height of the wedge is 1 m. The mechanical
properties are the same as those in Example 1. An in situ
state of stress with the two principal stresses parallel to
the excavation roof equal to -0.05 MPa and a zero
vertical principal stress is applied. For these property
conditions, the closed-form solution yields a friction
coefficient required for wedge stability of 0.685.
The numerical simulation is performed by fixing all
the blocks surrounding the wedge and setting the contact
forces between the three wedge faces and the adjacent
blocks in accordance with the in situ stress state. Dis-
placement of the wedge under gravity leads to the
relaxation of these contact stresses. The friction
coefficient is reduced in the model until failure occurs.
The 3DEC simulation calculates a coefficient for sta-
bility of 0.684 (within 0.2%). The failure of the wedge
is indicated in Fig. 4.
Example 3
This example extends the wedge models described in
the first two examples to a more realistic problem of
tunnel stability. The problem is defined as a single
excavation in a faulted rock. The rock contains three
major faults: one dipping at 50 with a dip direction of
320; the second dipping at 40 with a dip direction of
230; and the third dipping at 65 with a dip direction
of 270 . A square opening is made in the model using the
tunnel generator. The intersection of the faults with the
tunnel is shown in Fig. 5.
The three faults and the excavation form an isolated
wedge in the roof of the excavation, which is potentially
unstable and can slide along the fault plane dipping at
65 . Blocks in the front of the wedge are hidden from
view in Fig. 5 for better viewing of the wedge. (Hidden
blocks are still present for mechanical calculations.) The
model is subjected to gravity loading, and the
progressive failure of the roof wedge is illustrated in Fig.
5.
This model consists of 21 blocks and requires
480 kbytes of memory on the personal computer. A run
time of approximately 30 min is sufficient to identify
failure in this case.
Example 4
The final example is a three-dimensional represent-
ation of a rock slope in sedimentary rock. The example
is an extension of the problem presented by Starfield and
Cundall [13] and involves a cut slope in rock with steeply
dipping foliation planes. A rotational failure was found
to occur with simultaneous sliding along both the foli-
ation planes and shallow-dipping fracture planes. The
rotation failure mode was identified by two-dimensional
distinct element analysis as the principal mechanism for
the slope collapse.
This problem was reproduced using 3DEC, with
similar results to those of the two-dimensional analysis.
A three-dimensional variation was then performed by
introducing two intersecting discontinuities in the slope,
forming a wedge similar to that in Example 1. The
problem geometry is illustrated in Fig. 6.
The failure mode for this problem combines the
rotational failure with sliding of the wedge. The develop-
ment of collapse is depicted in Fig. 7. The sliding wedge
dominates the failure for this problem setting, but the
rotational mechanism contributes to the collapse,
particularly on the left portion of the slope face, as seen
in Fig. 7. Vertical sections taken through the wedge
(Fig. 8) and through the right side of the face (Fig. 9)
show the distinguishing response of the two types of
failure. Displacement vectors at block vertices are
plotted on these two figures, indicating the relative block
movement. Cross-sectional plots can be oriented at any
angle through the model, and various parameters can be
presented on these sections.
This analysis demonstrates the application of the code
to study conditions under which failure can occur. The
present model investigated the effect of two joint sets and
two intersecting discontinuities. The model contained 49
blocks and the run time, to the state of Fig. 7, was
210 min, using a DSI-780 co-processor board. This state
is further than required to identify that the slope has
failed. A run time of approximately 120 min is sufficient
to identify failure in this problem. By making several
runs with different properties, we can define the condi-
tions leading to collapse of the slope. We can also
introduce more discontinuities, if required, and still
perform analyses in a reasonable time.
CONCLUDING REMARKS
Stability analysis with 3DEC emphasizes the use of the
code on a personal computer. It is contended that rock
mechanics analyses of problems of moderate size (i.e.
requiring a model consisting of up to 1000 rigid blocks)
can be performed in a reasonable time frame on a
personal computer. This approach takes advantage o f
the state-of-the-art technology available in micro-
computers and standard screens (EGA) to present
three-dimensional graphics. This allows the user to enter
input and assess results interactively from both a typed-
command mode and a graphics-screen mode. Con-
sequently, this greatly reduces the possibility for error
when creating a three-dimensional block structure
and facilitates the interpretation of results of the
analysis. The instability of a system can be recognized
immediately in the graphics mode.
Acknowledgements--The development of the three-dimensional
distinct element model is sponsored by Faloanbridge Nickel Mines
Ltd, Toronto, Canada. A portion of the work is also fundedthrough
an Applied Research Associates contract with the Defense Nuclear
Agency.
REFERENCES
1. Burman B. C. A numerical approach to the mechanics of discon-
tinua. Ph.D Thesis, James Cook University of N, Queensland,
Australia (I 971 ).
 HART et al.: 3-D DEM--PART II
2. Rodriguez-Ortiz J. M. Estudia del comportamiento de medios
granulates heteroteneos mediante modelos discontinuos analogi-
cos y matematicos. Ph.D. Thesis, Universidad Politecnica de
Madrid (1974).
3. Chappell B. A. Load distribution and redistribution in discon-
tinua. Int. J. Rock. Mech. Min. Sci. & Geomech. Abstr. 16, 391-399
(1979).
4. Goodman R. E. and Shi G. Block Theory and Its Application to
Rock Engineering. Prentice-Hall, New Jersey (1985).
5. Key S. W. A data structure for three-dimensional sliding
interfaces. Int. Conf. on Computational Mechanics, Tokyo
(1986).
6. Cundall P. A. and Strack O. D. L. A discrete numerical model for
granular assemblies. Geotechnique 29, 47-65 (1979).
7. Cundall P. A. and Hart R. D. Development of generalized 2-D and
3-D distinct element programs for modeling jointed rock. Itasca
Consulting Group; Misc. Paper SL-85-I, U.S. Army Corps of
Engineers (1985).
8. Lemos J. A distinct element model for dynamic analysis of jointed
R.M.M.S, 25/~--C
125
rock with application to dam foundations and fault motion. Ph.D.
Thesis, University of Minnesota (June, 1987).
9. Cundall P. A. Distinct element models of rock and soil structure.
Analytical and Computational Methods in Engineering Rock
Mechanics (Edited by E. T. Brown), Chap. 4, pp. 129--163.
Allen & Unwin, London (1987).
I0. Cundall P. A. Adaptive density-scaling for time-explicit
calculations. Proc. 4th Int. Conf. on Numerical Methods in
Geomechanics, Edmonton, pp. 23-26 (1982).
11. Hoek E. and Bray J. W. Rock Slope Engineering, 2nd edn. IMM,
London (1977).
12. Goodman R. E., Shi G. and Boyle W. Calculations of support for
hard, jointed rock using the keyblock principal. In Issues in Rock
Mechanics (Proc. 23rd Symp. on Rock Mechanics, Univ. of
California, Berkeley, August 1982), pp. 883-898. SME, New York
(1982).
13. Starlield A. M. and Cundall P. A. Towards a methodology
for rock mechanics modelling. Int. J. Rock Mech. Min. Sci. &
Geomech. Abstr. 25, 99-106 (1988).