You are on page 1of 8

313

A H Y B R I D P R O C E D U R E OF D I S T I N C T - B O U N D A R Y
ELEMENT FOR DISCRETE ROCK DYNAMIC ANALYSIS
Jin Feng, Wang Guanglun and Zhang Chuhan
(Tsinghua University, Beijing, China, 100084)

Abstract
A hybrid procedure of distinct element- time domain boundary element for discrete rock dynamics
analysis is presented, in which the boundary element domain is embedded into DEM procedure as a distinct
block. Combining the advantages of DEM and BEM, this coupling model not only can model the dynamic
behaviour of discrete rock medium, but also can simulate the radiation damping of far field and non-uniform
input of dynamic loading such as earthquake or blasting wave. A simple problem is studied to verify the
presented procedure.

Introduction
Distinct element method(DEM) is a useful tool to study static and dynamic response and stability of
discrete rocky tunnels or slopes. Since the emergence of the first discrete element model, i.e. distinct element
method (DEM) proposed by Cundall [1] in the 1970's, significant developments have taken place both in
numerical simulation and engineering applications. Fully deformable distinct element methods [2'3'4]for twoand three-dimensional problems have been reported, which are gaining increasingly more users in rock
mechanics and engineering. Meanwhile, some other numerical techniques for simulating discontinuous
medium have been presented. For example, a discontinuous deformation analysis method (DDA) has been
proposed by Shi and Goodman [5'6], who used strain mode superposition rather than the finite difference
scheme of DEM to account for element deformation. Another discrete element model proposed by Williams
and Mustoe [7] uses orthogonal modes to approximate the deformation patterns of block elements, in which
only a few lower modes are needed to describe complex deformation. On the other hand to simulate granular
material, Ghaboussi et al. [8]presented a special rigid discrete element. Additionally, as applications of DEM
in dynamic analysis of rocky slopes and structures, Iwashita and Hakuno [13] studied the dynamic failure
process of a cliff using the extended DEM, while Meguro et al. [14] also studied the seismic collapse of
concrete structures by modelling the structure as a system comprising numerous discrete particles.
One outstanding feature of DEM is its ability to consider the slip and separation behaviours of contact
joints as well as the deformation of solid rock, in which any constitutive law for the material can be
implemented. The second important feature of DEM is its inherent suitability for non-linear dynamic
problems including earthquake shaking and rock blasting. This is due to its explicit time marching scheme
for solving the dynamic equilibrium equations and its incremental formulation of the constitutive behaviour
in response to stress increments, where linear and static problems are viewed as special cases of the solution.
In coupling the DEM with another numerical scheme, Lorig et al. [9] developed a hybrid distinctboundary element system for jointed - intact rock interaction. Huang and Ma [l~ presented a couple procedure
of distinct element and boundary element which simulate a granular soil deposit as a two-dimensional,
circular disk assembly. They used this hybrid model to study the behaviour of granular materials under cone
penetration tests. But the above two hybrid distinct-boundary element procedures are only for static analysis.

314
In the context of dynamic problems, to simulate the effects of radiation damping of far field, hybrid
schemes have significant advantages. Lemos [~l] developed a dynamic DEM program for studying seismic
response of a dam-foundation system. Non-reflecting lateral flee-field boundaries are imposed to simulate
radiation effects. Similarly, Dowding et al. [12'131 studied the dynamic response of an underground cavern
using a model for coupling near-cavern rigid blocks with surrounding finite elements. The far-field effects
were considered by introducing a viscous boundary, and the results were compared with a physical model
test.
The dynamic boundary element method(BEM) has the advantages of simplicity to model far field, as the
radiation conditions can be satisfied in the analytical fundamental solutions. In the early of 1980s, two
approaches of two-dimensional dynamic BEM employing time domain fundamental solutions for transient
elastodynamics were developed. Niwa et al. t~61 solved two-dimensional dynamic problems utilising a
simplified three-dimensional time domain fundamental solution. Mansur and Brebbia t17'~8] presented a BEM
procedure based on two-dimensional fundamental solutions. They systematically studied the integral
techniques and singularity of each term of fundamental solutions and concluded that the accuracy and
efficiency of the time domain BEM can be enhanced by analytical integration of the time domain integral
and the singularity can be treated only at the first time step. The numerical results by these authors are very
encouraging. Rice and Sadd I191 developed a time domain BEM for problems with a regular free surface
without discretization of the surface by utilising time domain fundamental solutions of the half-plane. Jin et
al. L2~developed a simplified procedure to simulate the effects of the material damping in time domain BEM,
which employs time domain fundamental solutions of linear elastic medium. As applications of the time
domain BEM, some simple two- and three-dimensional soil-structure interaction problems have been studied
by Spyrakos and Beskos [211,and Karabalis and Beskos [221.
This paper presents a hybrid scheme of DEM and time domain BEM, which can study the effects of
radiation damping of far field and non-uniform input mechanism on discontinuous medium with continuous
far field. To the authors' knowledge, there is little work to study these influences. At first, the basic outline
of distinct element method for deformable blocks and time domain boundary element method with material
damping are reviewed. Then, the coupling procedure is presented and a simple benchmark problem is studied
to verify its validation.

Distinct element method


As shown in Figure 1(a), the system
is divided into blocks by contact joints
and the interior region of the blocks are
discretized into triangular elements
which are assumed to be fully
deformable. The geometry of the joints
can be specified individually or
stochastically generated according to
statistical data from field measurements.
It is assumed that normal and shear
springs with stiffness kn and ks,
respectively,
represent the
forcedisplacement relationship of the joints.
The latter are assumed to exhibit zero
tension in the normal direction and
follow Coulomb's friction law for shear,
i.e.

Prescribed
force Fh ~ "

Ar

~_._ %
(a) System

-2v 1
Prescribed ~ ~ ' - ~
[~__.motionub, ub
(b) Detail A

Fig. 1 Deformable DEM system

315

Fe, = k,u,,
Fe,=O
Fc.~.= k,u,.

if
if

u, < O
u,>O
if [Fe.,.[ < f [F~,[+ eL

Fc~, = sign(u,.)(f ]Fcn[+ cL)

if

(1)

[Fc, > f [Fc,[+ cL

where Fcn and Fcs are, respectively, the normal and tangential components of the contact force vector
Fc; kn and ks represent the normal and shear stiffness of the joints; Un and Us are the corresponding
components of the relative joint displacement;f c are friction and cohesion coefficients of the joint material;
and L denotes the length of the contact boundary.
The technique by Cundall to determine the deformation of the interior region of blocks employs
triangular finite difference elements of the constant-strain type. As shown in Figure 1(b), the grid point N is
a common node surrounded by elements B1 B2 ... Bn. The inertial mass of each element is assumed to be
equally distributed to the three vertices of the element yielding a mass of the polygon area confined by F1 be
lumped at node N. In order to maintain consistency in computing inertial and elastic forces, the boundary F1
should also be used as the integration route for elastic stresses. Since constant strain elements are assumed,
the following equation defines the force vector Fe at node N due to elastic deformation of the blocks:

Fe : Icron.,
F,

ds:

Icron.,ds=s
F

i,]

1,2
(2)

k=l

where cr/j" is the stress tensor in the element; nj denotes the unit normal vector of the integration
boundary; F1 and F are, respectively, the polygon boundaries denoted by dashed and solid lines; and n
represents the number of boundary segments surrounding node N.
In DEM solutions, damping is regarded as the dissipater of vibration energy for both static and dynamic
loading. In both cases, mass proportional damping and stiffness proportional damping may be used
separately or in combination. In the dynamic case, Rayleigh damping for continuous media yields the
following equations:
F

=- crmit
dm

Fd~ ~ . v - At

where Fdm, Fdk are damping


forces due to mass and stiffness,
respectively; m represents the
lumped mass of the node; a, fl
represent corresponding damping
constants; and AF denotes the
incremental total force vector
including joint contact and
internal deformation. Empirical
values of damping ratios can be
assumed to determine a and ft.
The solution scheme of DEM
is illustrated briefly in Figure 2.
For each time step At, the
incremental forces acting at each
node due to deformation of the
surrounding contact joints and the
internal deformation of the
material are determined from
incremental joint displacements

(3)

BEM
/

~ /" Urescrioed
, --"N
{ force ]
F.b J

a
/

Eqs (15) [
\

/ -[Force-motion
-_--:7_-:_-~.-~
]\
[ relationship[ \ ~
[ Eqs (1)(2)(3) [ k

'
//"]Prescribed
.."
{ m,ouon )
~ ~.b,~'b j

Di=

] Dynamicequilibrium ]
Equation Eq. (5)
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Fig. 2 Schemetic flow chart of DE-BE hybrid 'procrdure

316
and intemal strains by constitutive relationships shown in the upper box of Figure 2. The summation of the
forces F goes to the lower box, where the dynamic equilibrium equations are applied, namely:
mfi + a mil = F + m g
(4)
where, mg is gravity force.
By employing the central difference method for the acceleration on the left hand side of equation (4),
one obtains
u(t + Ar / 2) = ( z~(t + Ar / 2 ) ( 1 - a A r / 2) + (F / m + g) At) / (1 + a Ar / 2)
(5)
The new velocities and displacements, hence the new coordinates, are updated at each time step by
equation (5). Note that the damping force in equation (5) is centred at time r.

Time domain boundary element method with damping


Under zero initial conditions, the time domain boundary integral equation for a two-dimensional linear
elastic body without body force is given by

where a;fl= 1, 2; up,pz are, respectively, the displacement and traction vectors; u :z, P:z represent the
corresponding displacement and traction impulse fundamental solutions and c~p is a constant related to the
boundary geometry at source point S. Q represents the field point and F denotes the boundary of the twodimensional body.
Integrating by parts and considering zero initial conditions lead to the following alternative form of
boundary integral equation Eq. (6):

%zUz(S,t) = ~ [~U*~,(Q,t;S,r)p,(Q,t)dF]dr+ ~ [Ivs


9

(7)

where vp is the velocity vector and U~p,Vap, Ps denote fundamental solutions whose expressions can
be found in Reference [18].
Discretizing the boundary into boundary elements and assuming that velocities and tractions remain
constant over time increment At transform the integrals of Eq. (7) into the summations given below:

~ [ ~ P~fl(Q,t;S,r)ufl(Q,t)elF] dr= ~ [D]M-m{u} m


m=O

(8)

m=O
M

~ [ IU*~,(Q't;S'r) P,(O,t)dF]dr=m~_~ [G]M-'{u}m


where t = MAt; D, E, G are coefficient matrices and superscripts m and M-m indicate that the variable
relates to the corresponding time step.
A time domain boundary element matrix equation for time step M can established as follows:
[H]~ u} M : [G]0 {p} M + {B} M
(9)
where
[H] ~ = [C] + [H'] ~
M-1

{B} M : Z ([GIM-m{p}m--[ITIM-'{U} m)
m=l

[H']" = [D]"+ [El"

n =0,1.... M

(10)

317
in which C is a matrix related to the shape of the boundary and vector B is known from previous
calculation.
When proportional decay damping model applied, boundary integral equation Eq. (2) takes the
following modified form: t20j

capup(S,t)- ~[ ~U*~p(p,t;S,r)pp(p,t)dF' (1-rl)(t-O/aTdr


+ ~[ ~ V:p(p,t;S,r)vp(O,t)dF]

(1-

rl)('-O/zTdr

Where T is a characteristic time and the damping coefficient of proportional decay damping t2~
r/= 1 - exp(-4~r~:)

(11)

(12)

Where ~ is the damping ratio of BEM domain.


When material damPing is sufficiently small, the assumption that (1m

q)~t-o/2r remains constant over a

m
Accordingly, Uflj,pflj should be modified to upj(1-77)
(M-m)at/zr and

time increment At is justified.

pamj(1-rl)(M-m)a'/2~,respectively.

This results in the boundary element equation with proportional decay


damping possessing the same form as Eq.(9), but with one term changed in order to incorporate proportional
decay damping, namely the term BM is modified to
M-1

{B} M = Z

([G]M-m{P}m--[H']M-m{u}m) (1- 17)(M-m)At/2T

(13)

m=l

H y b r i d procedure of D E M - B E M
The domain shown in Fig. 3 can be
separated into two sub-domain: (1) a discrete
elastic blocky domain modelled by DEM and
(2) an infinite or semi-infinite homogeneous
continuous elastic domain modelled by BEM,
where the vibration energy in the DEM domain
will transmit to infinite or the earthquake
energy will propagate and input into the
omain
system.
Interface Block~
For assurance of compatibility, a block,
named interface block, is set along the interface
Fig. 3 DE-BE hybrid system
between the two domains. The boundary grid
points along the interface of the interface block is called interface grid points. After the length of boundary
element chosen, some interface grid points, whose distance between each other is approximate the chosen
length of boundary element, are set to be the nodes of boundary element. So, the displacement of these
interface grid points is same as that of boundary element nodes. Other interface grid points will be
approximate consistent with BEM domain. But the interface force is distributed to all interface grid points
linearly.
Employing the boundary conditions of free surface, condense BEM Eq. (9) to the interface, one can
obtain
[HI ]{u, )M = [G, ]{p, }M * {B, }M
(14)
where, ui, PI denote the displacement and traction vector on the interface, respectively; Blis a vector whose
components are all known; HI and GI are coefficient matrixes. Eq. (13) can be transformed into

{FI}=[AII{uI}M+{CI} M
where, FI is the interface force vector of interface grid points and

(15)

318
[A, ]= [RI[G, ]-' [HI l
{C, }a4 = -[R][G, ]-1 {B I }M
(16)
where, R is the transform matrix from traction of BEM to interface force.
Let the time increment At of BEM to be Ktimes of the time increment ArofDEM, namely, At = KAr. In
a time increment At, The relation between interface force FI and the displacement uI as shown in Eq. 14 can
be assumed as identity. Then, the whole BEM domain can be treated as a DEM block in a time increment At.
Under the assumption of constant velocity in a At, the displacements of interface nodes
M
M-1
M-I
K
{u,} ={u,}
+({uk}-{u,}
)~
k=1,2 .... K
(17)
where, u k is the displacement vector of grid point connected with corresponding BEM nodes.
When k = K, the traction vector of BEM pM will be updated as follow
K

~-"~[R]-' {F,' }
{p,}M=k=,

(18>

Then, B M of BEM is updated.


As shown in Fig. 2, for the interface grid points, the interface force F1 can determined by Eq. (15). So,
the BEM domain can be embedded into the DEM procedure as a distinct block.

Verification of coupling technique


To verify the validation of proposed hybrid procedure, a column comprised of 4 rock blocks, as shown
in Fig. 4, is studied and the dynamic displacement histories of point A, B, C and D are compared with that
obtained by time domain BEM. The top block is discretized by 16 boundary elements. The bottom one is
modelled by a rigid block and is fixed. The other two blocks are simulated by discrete deformable blocks. A
uniform distributed step load with amplitude ofp = 1MN/m2 is applied at the top boundary. The Young's
module E = 3 10~~ density p = 2000 kg/m3,Poisson's ratio v = 0, damping ratio ~ = 0.1, kn = ks =
10~N/m,f= 1.18 and c = 0.1MPa. The displacement histories of point A, B, C and D are calculated by time
domain BEM and the hybrid scheme of distinct element and time domain boundary element, as shown in Fig.
5.
Because in the case of analysis the interfaces between blocks keeps contact and the stiffness of contact
springs is big enough, the dynamic behaviour of block system which is simulated by the hybrid scheme is
just as same as that of a continuous column modelled by TDBEM. The displacements of all nodes reach their
corresponding static one after a few cycles of oscillation. The displacement histories of points calculated by
the two method concord with each other very well. It means that the deformable distinct element method can
simulate the wave propagation behaviour of blocks and the hybrid scheme works well.

Conclusion
A hybrid procedure of dynamic boundary element in time domain and deformable distinct element is
presented. Comparison of the results of time domain boundary element and that of the presented hybrid
procedure shows that the coupling procedure works well.

319

20[

BE

'

Domain

"

DE

1.5

~,

1.0

Time domain BEM


Hybrid procedure

"=-'-

--

- A

0.5

0%

'

'

11o

'

'

2:0

'

Time (s)

Fig. 4 Column studied

Fig. 5 Comparison of TDBEM and hybrid procedure

Reference
[1 ] P.A. Cundall, A computer model for simulating progressive large scale movement in blocky systems,
Proc. of syrup, of int. soci. rock mech., Vol. 1, Nancy, France, 1971, paper no. II-8.

[2] P.A. Cundall, UDEC-A generalized distinct element program for modelling jointed rock, Final technical
report to European research office, US. Army, 1980.

[3] P.A. Cundall, Formulation of three-dimensional distinct element model, Part I, A scheme to detect and
represent contact in system composed of many polyhedral blocks, Int. J. rock mech. rain. sci. & geomech.
abstr., 25(3), 1988, 107-116.
[4] R. Hart, P. A. Cundall, and J. Lemos, Formulation of three-dimensional distinct element model, Part II,
mechanical calculation of a system composed of many polyhedral blocks, Int. J. rock mech. rain. sci. &
geomech, abstr, 25(3), 1988, 117-125.
[5] G.H. Shi and R.E. Goodman, Two dimensional discontinuous analysis, Int. J. numer, anal. methods
geomech., 9, 1985, 541-556.
[6] G.H. Shi and R.E. Goodman, Generalization of two-dimensional discontinuous deformation analysis for
forward modelling, Int. J. numer, anal. methods geomech., 13, 1989, 359-380.
[7] J. R. Williams and G. G. W. Mustoe, Modal methods for the analysis of discrete systems, Computers and
Geotechnics, 4, 1987, 1-19.
[8] J. Ghaboussi and R. Barbosa, Three-dimensional and discrete element method for granular materials, Int.
J. numer, anal. methods geomech., 14, 1990, 451-472o
[9] L. J. Lorig, B. H. Brady and P. A. Cundall, Hybrid distinct element-boundary element analysis ofjointed
rock, Int. J. rock mech. rain. sci & geomech, abstr., 23(4), 1986, 303-312.
[10] A. B. Huang and M. Y. Ma, An analytical study of cone penetration tests in granular material, Can.
Geotech. J., 31, 1994, 91-103.
[11] J. Lemos, A distinct element model for dynamic analysis of jointed rock with application to dam
foundations and fault motion, Ph.D. thesis, University of Minnesota, 1987
[12] C. H. Dowding, T. B. Belytschko and H. J. Yen, A coupled finite element-rigid block method for
transient analysis of rock caverns, Int. J. numer, anal. methods geomech., 7, 1983, 117-127.
[ 13] C. H. Dowding, T. B. Belyschko and H. J. Yen, Dynamic computational analysis of opening in jointed
rock, J. geotech, eng., ASCE, 109, 1983, 1551-1566.

320
[14] K. Iwashita and M. Hakuno, Granular assembly simulation for dynamic cliff collapse due to earthquake,
Proc. 9th worm conf. earthquake eng., 3, Tokyo-Kyoto, 1988, 175-180.
[15] K. Meguro and M. Hakuno, Simulation of collapse of structures due to earthquake using extended
distinct element method, Proc. lOth world conf. earthquake eng. 7, Madrid, 1992, 3793-3796.
[16] Y. Niwa, T. Fukui, S. Kato and K. Fujiki, An Application of the Integral Equation Method to twodimensional Elastodynamics, Theor. Appl. Mech., 28,Univ. Tokyo Press, 1980, 282 - 290.
[17] W. J. Mansur and C. A. Brebbia, Transient Elastodynamics Using a Time-stepping Technique,
Boundary Elements, Springer-Verlag, 1983
[ 18] W. J. Mansur and C. A. Brebbia, Transient Elastodynamics, Topics in Boundary Element Research, 2,
Springer-Verlag, 1984, 124-155.
[ 19] J. M. Rice and M. H. Sadd, Propagation and Scattering of SH Waves in Semi-Infinite Domain- Using a
Time-Dependent Boundary Element Method, J. Appl. Mech. 51, 1984, 641-645.
[20] F. Jin, O. A. Pekau and Ch. H. Zhang, A 2-D time domain boundary element method with damping,
submitted lnt. J. Num. Methods Eng.
[21] C. C. Spyrakos and D. E. Beskos, Dynamic response of rigid strip-foundations by time domain
boundary Element Method, lnt. J. Num. Methods Eng., 23, 1986, 1547-1565.
[22] D. L. Karabalis and D. E. Beskos, Dynamic response of 3-D flexible foundations by time domain BEM
and FEM, Soil Dyn. Earthquake Eng., 4, 1985, 91-101.

You might also like