Sei sulla pagina 1di 7

The discrete element method for the simulation of

ball mills
B. K. Mishra and Raj K. Rajamani

Comminution Center, University of Utah, Salt Lake City, Utah, USA

The discrete element method (DEM) is a proven numerical technique for modelling the multibody
collision behavior ofparticulate systems. This method is used here to study the motion of ball charge in
tumbling mills. To get meaningful results, it is essential that the parameters involved in the model be
carefully determined. These parameters embody essentially the material properties of the system:
stiffness, damping, andfriction. In this study, all the parameters are carefully determined experimentally
for different operating mill conditions. A computer code based on DEM has been developed to model the
motion of the bails in tumbling mills. The code incorporates a scheme to calculate the applied torque, and
hence power input to the mill. A 55cm ball mill is simulated,for two different liner cross-sections-
rectangular and triangular. Results ofthe simulations pertaining to applied torque are compared with the
experiments. It is found that, with a particular modelfor the coefficient offriction, the predicted torque
agrees well with the experiments.

Keywords: ball mill, ball charge motion, simulation, discrete element method

approach to this problem offers a unique advantage in


Introduction
that the subprocesses can be quantified. Therefore the
Ball mills are the primary grinding equipment in the research work reported here is concerned with numeri-
mineral processing industry. Tonnages in the range of cal simulation followed by experimental verification.
40,000-100,000 tons/day are processed. These mills The simulation is based on Newton’s second law for
consume enormous amounts of electrical energy, and force calculation and requires some knowledge of the
consequently, the power cost can be as high as half the energy dissipated during the collisions. Since the mo-
total processing cost. For this reason, much of today’s tion of individual balls is computed, the en masse
comminution research is aimed at understanding grind- motion of the balls emerges as a result.
ing mechanisms, estimated power draft, and modelling What makes this simulation unique is that the mate-
milling circuits. As progress is made in these areas, rial properties of the colliding bodies are the only pa-
fresh problems have surfaced that require practical as rameters used in the computation. Therefore the
well as theoretical solutions. These problems deal with, scheme easily lends itself to scale-up work. In this paper
among other things, accurately predicting the power we pay special attention to accurately determining ma-
draft of the mill, the distribution of collision forces terial properties.
inside the mill, and the wear rate of balls and liners.
As the power drawn by the mill is intimately related
to the motion of the ball charge, the principal problem is
to model the motion of the balls. An analytical solution Discrete element method
to this problem, in which the collisions are inelastic, is
The discrete element method (DEM) is a special class of
not possible. Therefore the intention here is to explore
numerical schemes for simulating the behavior of dis-
some other avenues of fundamental inquiry, by means
crete, interacting bodies. Hence, by its very nature,
of which it will be possible to model the individual
DEM is highly suitable for the ball mill problem. By
collisions and hence obtain the motion of the entire ball
definition,’ ” discrete element method applies to a com-
charge.
puter program only if it allows finite displacements and
In this analysis the motion of the charge is treated as a
rotations of the discrete bodies, including complete de-
multibody collision problem. Obviously, a numerical
tachment, and recognizes new contacts as the calcula-
tion progresses.” This method was pioneered by
Cundall and Strack,’ who analyzed the behavior of soil
using two-dimensional disc elements in the numerical
Address reprint requests to Dr. Mishra at the Comminution Center,
University of Utah, Salt Lake City, UT 84112, USA.
scheme. With the increasing popularity of numerical
modelling in various problems, coupled with the avail-
Received 8 August 1991; revised 24 March 1992; accepted 2 April 1992 ability of computational time, DEM applications are

598 Appl. Math. Modelling, 1992, Vol. 16, November 0 1992 Butterworth-Heinemann
Discrete element method in modelling ball mills: B. K. Mishra and R. K. Rajamani

growing in a diverse range of fields, from rock mechan-


ics to computer graphics.
In particulate material research, the problems gener-
ally involve a large number of distinct bodies. It follows,
therefore, that a large number of particles need to be
modelled for many time steps in DEM simulations to
obtain meaningful results. Consequently, most imple-
mentations of the DEM in this area have not progressed
past the two-dimensional stage. We too have used a
two-dimensional code in which the real ball mill is
represented by a thin cylindrical section in which an
assembly of discs, instead of balls, collide; therefore
only the X- and y-coordinates need to be considered in
the calculations. Even with such a drastic simplifica-
tion, the computations require a supercomputer. How-
ever, it will be shown that this two-dimensional approx-
imation is adequate for the estimation of power draw
and motion analysis.
Fs(max)=p l Fn
The DEM ulgorithm
The numerical algorithm is conceptually simple. Ev-
ery disc in the assembly is identified separately by its
radius, mass, moment of inertia, and collision proper-
ties. For all discs, a list of the discs that are in immediate
contact and in the near-neighborhood is maintained. Figure 1. Representation of a contact
Thus to find which other discs are in contact with a
given disc, only that disc’s list is checked instead of
checking every disc. This procedure drastically reduces and are given by
the number of searching steps at the expense of com- m.f; + C,i, + K,.q = F, i= I,2 (I)
puter memory space. Once there is overlap between
two discs, then the normal and shear forces at the
16 + i (K,X, + Ci-f,)s, = A4
contact are determined from the magnitude and rate of
i= I
overlap. Finally, the contact forces on a disc are
summed to get the net out-of-balance force acting on it. where F, is the force acting in direction i, A4 is the
From this force the acceleration, velocity, and displace- moment about the centroid of the disc, and si is the
ment of the disc are calculated. As this procedure is perpendicular distance from the line of action of the
carried out, for every disc an entire set of the disc force (K,xi + C;..$) to the centroid of the ball. The force
positions emerges at the next instant of time, and these F, is the net force acting on the disc at the previous
calculations are repeated over a number of time steps. instant of time. The parameters K and C are the spring
The DEM algorithm applies Newton’s second law to and dashpot constants, respectively. The directions i =
the discs in motion and a force displacement model to 1 and 2 refer to the orthogonal coordinate axis. The
the discs in collision. Newton’s second law gives the above two equations are a special case of the second-
motion of the discs resulting from the forces acting on order nonlinear differential equation of the following
them. Figure I shows a cluster of discs where disc i is in form:
contact with other discs such as k, I, m, and n. At every
contact the forces developing due to collision are 1+ai=b(n_,i)
m
(3)
modelled by a pair of normal and tangential spring
dashpots. The spring-dashpot arrangement for the con- The quantity F(x, .a!_)is the force generated at the con-
tact made by disc i and n is shown in Figure 1. The tact. Of course, the gravitational force is acting through
provision of viscous contact damping models the sys- the center of the disc at all times, and in addition a drag
tem realistically, because for different types of contacts force term can be added if the discs move through a
different damping constants can be specified in the viscous fluid. During a small time step, t to t + St, this
algorithm. Also, friction damping is implemented dur- equation can be integrated by finite-difference approxi-
ing sliding of one disc over another. mations of the derivatives:
The equations for computing the incremental contact
forces between discs were developed by Cundall and f = i(t + St) - a(t)
Strack’; the mode1 equations are presented below in a (4)
St
very genera1 way for ease of discussion. The dynamic
equilibrium equations can be written for a disc charac- ~ = i(t + St) + i(t)
(5)
terized by position X, mass m, and moment of inertia I 2

Appl. Math. Modelling, 1992, Vol. 16, November 599


Discrete element method in modelling ball mills: B. K. Mishra and R. K. Rajamani

Substitution of equations (4) and (5) into equation (3) numerical results; these were very helpful to com-
results in a simple equation that may be solved for the pare with the experimental data.
velocity during the time step t to t + 6t. This velocity is l The original code (DISC) connected the rigid ele-
then integrated to find displacements: ments or walls in such a way that they formed only a
x(t + 6t) = x(t) + _%(t+ &) x 6t topologically convex region. This fact was seminal to
(6) the contact searching scheme in the code. However,
In a similar manner, equation (2) is solved to get the ball-mill lifter-bar geometry implies a topologically
rotation of the disc. Thus by repeating these calcula- concave space, and hence the code was modified to
tions for each disc, the new position of all the discs is accommodate this feature.
determined, and the time evolution of charge motion in l A serious numerical problem occurred, especially
the X, y-coordinate space is given by repeating the en- after simulations over thousands of time steps. At all
tire set of calculations at successive time steps. times the positions of the mill walls are kept in the
The stability of the numerical algorithm is important memory. Since the mill walls rotate about the center,
in assuring dependably accurate results. Since the inte- the wall coordinates change incrementally at each
gration scheme outlined is the central difference time step. However, in evaluating the wall position
method, the numerical stability depends on the time coordinates, some precision is lost because of the
step chosen. In the spring-dashpot model, a linear elas- very small numbers used in the calculations. It was
tic behavior of the spring (F = K 6x) is assumed, and in noticed that in a typical simulation using a time step
the dashpot the force is proportional to the relative ve- of 10d4 second, the polygonal geometry of the mill got
locity of the colliding bodies (F = C 6~). Since the ac- visibly distorted after just two revolutions. There-
celeration is related to the force acting on the disc, the fore at the end of each revolution the wall positions
value of the time step that assures numerical stability is were corrected to assume their original position at
related to the parameters involved. This time step is the start.
given in terms of the mass and the stiffness of the small- l The code identifies midair collisions and assigns a
est disc as different coefficient of restitution than those used in
ball-to-wall collisions.
6t < 2flK (7) l An important addition to the computer code was the
Failure to ensure this condition results in unrealistic implementation of the energy and torque calcula-
motion of the ball charge. tions. As a result, the distribution of energy among all
collisions and the distribution of fall heights of the
balls can be computed.
Applied torque
l Many utility codes were added to create data files
The torque applied to the mill can be calculated from and result files. For example, one utility uses the
the shear forces acting on the mill shell. Assuming that input data to generate the mill shell.
the applied torque is the retarding torque, the net
torque applied to the mill is determined from The results of the simulation were verified with
actual experiments. Recorded video motion pictures of
T = i ri x Fshear.; (8) a 25-cm-diameter mill compared very favorably with
i=l simulation results.’ Also, the torque, measured in the
where, for any disc i in contact with the wall, ri is the same mill at a different speed, compared well with the
position vector and Fshear,i is the shear force acting at predictions. In the following, we present the impor-
the contact. In this manner the torque value is cal- tance of accurately determining the parameters of the
culated at each time step, and an average value is simulation model.
reported at the end of the simulation.

Simulation parameters
Computer code Specialized but rather simple experiments can be
The algorithm was coded into a computer program carried out to determine simulation parameters. Since
called 2DMILL. This code is based on Corkum’s DISC the model deals with individual disc-to-disc contacts, it
code3.4 and Cundall’s BALL code.’ While the basic is necessary to choose realistic values of contact prop-
structure of the parent code was kept the same, a erties; this applies to disc-to-wall contacts also. These
number of features unique to the ball mill were intro- parameters are material stiffness, coefficient of restitu-
duced. While the details of the 2DMILL code can be tion, and coefficient of friction. Material stiffness is
found in Mishra’s thesis,6 the major points are summa- required to correctly establish the forces generated in
rized below: the springs. The coefficient of restitution is a measure of
the damping property of the material, and hence it
l The code generates x- and y-coordinates of each disc figures into the damping parameters, which in turn
as a function of time. Hence an animation program establish the forces developed in the dashpot. Finally,
was written to examine the results. The snapshots of the coefficient of friction is used to check whether
the animation were the only way of analyzing the sliding occurs at any contact.

600 Appl. Math. Modelling, 1992, Vol. 16, November


Discrete element method in modelling ball mills: B. K. Mishra and R. K. Rajamani
Material stiffness known from the Hertzian contact theory for spheres
The material stiffness was determined in an appara- that the tangential stiffness K, may vary from 213to 1 of
tus known as the “ultrafast load cell,” an apparatus the normal stiffness KS. In this study, KS is taken to be
used by Hofler and Herbst’ for particle bed breakage 2/3 of the value of K,. Cundall and Strack5 have shown
studies. It consists of a 5-m-long steel rod in which fast- that for a low friction angle the contact forces did not
responding strain gauges are embedded. These gauges change significantly when the KS/K,, ratio varied from
record the strain waves when a steel ball is dropped onto 2/3 to 1.
the rod. As soon as the ball strikes the rod, a primary
wave is recorded, and as the wave reaches the bottom Coefficient of restitution
end of the rod, it is reflected; this wave too is caught by Coefficient of restitution for colliding bodies by deli-
the strain gauges. From these recordings the force ver- nition is
sus displacement curve can be computed, which is of
primary use to the ball motion study. Simply, the slope
of the force versus displacement curve is the required
(9)
stiffness. Consequently, the stiffness of ball-to-ball col-
lision involving layers of particles in between can also
be determined with this apparatus.
The results of the experiments in which the ball is where JR dt and JP dt are the deformation and restitu-
dropped onto the anvil (metal-metal collision) are pre- tion impulses, respectively. Using this definition, for
sented in Table I. It is seen from the table that, for a drop two discs moving in the same direction before and after
height of 0.3 m, the value of the stiffness for different impact, the coefficient of restitution is given by
ball weights is of the order of 100 MN/m. As per e = (4 - (vu)2
equation (7), for this value of stiffness the computa- (10)
tional time step is 8 x lo-’ second. Such a small time (vu),- (Jd,
step entails calculations of the order of lo6 for each ball where v0 and v,, are the velocities of discs a and b and 1
in 1 second of real time. Furthermore, in a ball mill and 2 refer to the situation just before and after impact,
grinding particles, metal-metal impacts are less likely to respectively. By measuring these relative velocities ex-
occur. Therefore a series of experiments was done with perimentally, the coefficient of restitution can be deter-
a 250-gram ball dropping onto a layer of particles situ- mined. However, it is known that the coefticient of
ated on the anvil. The results of the slope of the force- restitution is not a material property; it depends primar-
displacement curves are presented in Table 2. In these ily on the size, shape, and impact velocities of the
experiments, for the same ball weight, the drop heights colliding bodies. For these reasons the coefficient of
are varied. In comparison to the data presented in Table restitution must be measured under conditions that are
1, the stiffness values obtained are quite small. The known to exist in the ball mill.
maximum force recorded when the ball touches the Depending on the nature of the motion of the ball
anvil is also comparatively small. charge, generally two types of collisions are observed in
The tangential stiffness is determined from a theoret- the ball mill. The balls that flow down the free surface of
ical standpoint: For a Poisson’s ratio below 0.5 it is the charge depict “cascading motion,” whereas those

Table 1. Result of the ultrafast load cell experiments to determine the


material stiffness; metal-metal collision

Maximum
Ball mass Fall height Maximum displacement Stiffness
(kg) (m) force (Nl (mm) (MN/m)
0.0926 0.3 8109 0.075 140
0.2523 0.3 11066 0.119 135
0.6468 0.3 24872 0.138 126

Table 2. Result of the ultrafast load cell experiments to determine the material stiffness; layers of quartz particles used between
the colliding bodies; ball mass = 0.252 kg

Maximum
Fall height No. of Particle size Maximum displacement Stiffness
fm) layers (mm) force (N) (mm) (N/m)
0.15 1 3.35-4.0 638 1.41 466,829
0.25 1 3.35-4.0 834 1.93 395,052
0.50 1 3.35-4.0 1212 2.69 429,448
0.25 3 1.14-1.70 370 2.89 111,445
0.50 3 1.14-1.70 601 4.51 190,189
0.75 3 1.14-1.70 1065 5.08 341,346

Appl. Math. Modelling, 1992, Vol. 16, November 601


Discrete element method in modelling ball mills: B. K. Mishra and R. K. Rajamani

that are thrown out of the surface and follow a parabolic equation (1.5), then slip is presumed to occur. In this
path depict “cataracting motion.” Cataracting balls hit situation, during the computation the dashpot in the
the mill shell, causing ball-to-wall collision. It is difficult shear direction is omitted, and the F,,,,, value is used
to determine the coefficient of restitution for the instead. Rose and Sullivan’ have determined the
cascading balls because there is no impact. Finally, coefficient of friction for different conditions that are
there is another type of impact that occurs in the mill, applicable to ball mills.
that is, balls colliding in midair, resulting in ball-to-ball
collision. Remarks about the parameters
Simple experiments were done to determine the
The ball charge motion is a mix of low-energy and
coefficient of restitution for ball-to-wall collisions. In
high-energy collisions. Therefore the parameters deter-
these experiments, balls of different sizes held at a fixed
mined from specialized experiments point out the order
height were dropped onto a thick metal plate. Usually,
of magnitude of the constants involved; however, they
the drop height was about 30 cm. The impacts were
need to be adjusted to obtain meaningful results. For
captured on a video camera, and the video frames were
instance, Mishra and Rajamani’ tracked the position of
analyzed to calculate the rebound heights. It is easy to
two arbitrary discs in a 25cm-diameter laboratory mill
show that the coefficient of restitution is given by
by means of video image analysis. The shoulder and toe
e = vh,lh,
. . (11) nosition of the charge was determined at various mill
speeds. In the simulations the stiffness value was fixed,
where h, and h2 are the rebound and fall heights, respec-
but other parameters were adjusted to obtain good
tively. For different ball sizes and drop heights, the
agreement with experimental results. In this manner,
value ofthe coefficient fell in the range of0.4-0.6. Next,
the correct values of the coefficients of restitution and
to determine the coefficient of restitution for ball-ball
friction are established.
collisions, a pendulum experiment was devised. Two
The material stiffness value used in the model must
spherical balls were hung by threads. One of the pendu-
represent the physical situations that are encountered in
lums was stationary, while the other was released from
a ball mill. A single stiffness value cannot represent all
different distances. Again the heights of the balls before
forms of impact. In the ultrafast load cell experiments, it
and after impact were measured from video frames. In
is noted that the contact stiffness depends on the num-
this case the coefficient of restitution is
ber of layers of particles between colliding balls, the
e=2m-1 - (12) mass of the ball, and the fall height or the velocity of
impact. However, within a range, a stiffness value can
where h, is the rebound height of the ball that was
be chosen such that (1) the time step is minimized, (2)
originally at rest and h, is the fall height of the impacting
the contact overlap is reduced, and (3) the over-
ball. In these experiments it was observed that there
all motion of the charge is preserved. For the simula-
was very little variation in the measured coefticient.
tions reported here, the best value of stiffness is
The average coefficient of restitution is found to be 0.97,
400,000 N/m.
which pertains to midair collisions only.

Damping constant Numerical results and comparison with


The value of the damping constant is estimated from experimental data
the coefficient of restitution, since the latter is a mea-
sure of the energy loss upon collision. Corkurn showed Liddel and Moys” and recently Moys” have published
that for a given coefficient of restitution e the damping torque data on a 0.545-m x 0.304-m ball mill. Since
constant C is these authors claim an accuracy of 0.5 percent, their
data are chosen for comparison.
2 In (e)VG Experiments conducted on the 0.55-m-diameter mill
C= - (13)
%W(e) + 7r2 utilized four different lifter bar profiles. For the purpose
of comparison, two different configurations of lifter
where m = m,m,l(m, + mz) and m, and m, are the bars were chosen: square and triangular. The square
masses of the colliding discs. If m, is very large in lifters were 0.02 m x 0.02 m in cross-section, and the
comparison to m,, as in the case of a moving wall, then triangular lifters were 0.02 m high and slanted in the
m is the mass of the disc only. direction of the mill rotation. Thus 18 square lifters and
18 triangular lifters were used in the experiment. The
Coefficient of friction grinding media were made up of graded charge of
In light of the spring-dashpot model of collisions, the 0.012-m, 0.018-m, and 0.025-m balls with a bulk density
shear force due to the dashpot is limited by the maxi- of 4700 kg/m3.
mum that can exist at the contact, which is given as For comparing the results of the discrete element
code with experimental data, a scheme is developed to
F s,max =pxF,, (14) extrapolate the results of the two-dimensional model to
where p is the coefficient of friction and F,, is the normal a real mill. To do this, the mill can be considered as
force at the contact. If the absolute value of the force in individual slices in which the ball charge motion is
the shear-spring and dashpot exceeds the value given by confined to the radial direction only. After simulating

602 Appl. Math. Modelling, 1992, Vol. 16, November


Discrete element method in modeling ball mills: 6. K. Mishra and R. K. Rajamani

one such slice, the results for the entire mill are cal-
culated, the number of slices that make up the real mill
being known. In doing so, the component of velocity in
the axial direction is neglected. In other words, the
torque component due to charge movement in the axial
direction will not be accounted for in the two-dimen-
sional model. This component could be significant, es-
pecially in the feed end and discharge end of the mill.
The procedure for determining the number of slices
that is equivalent to the actual mill is as follows: Since
the code represents the spherical balls by discs, the area Mill Diomter =55cm
of the mill occupied by the discs is known. The disc is 70 _ Mill Length = 30cm
assigned the same weight as the spherical ball it repre- Mill Filling=38.5%
sents. Then the number of discs in the slice is adjusted I I I I I

so that the percentage of the area occupied by the discs, 0.6 0.7 0.9 0.9 1.0

including the void area, is equal to the percentage Fraction of Critical Speed I-l
charge filling used in the real mill. It should be noted that
the charge filling refers to the percent of the volume Figure 2. Comparison between experimentally determined
torque and the simulation
including the void volume occupied by the balls. Now,
with the number of discs in a given slice, the mass of the
slice is computed. Finally, the total number of slices is
the actual ball weight divided by the weight of discs in a earlier, while simulating the swivel lifter profile, a linear
single slice. Therefore the power draft or torque of the variation of the coefficient of friction with speed was
actual mill is the corresponding value for the slice assumed. Because of the excessive amount of sliding at
multiplied by the number of slices. high speed, this change in the contact model was incor-
The simulations were done in the fractional critical porated.
speed range of 0.6- 1.Oat a constant ball tilling of 38.5%.
From the ball filling, the ball weights, and the bulk
Conclusion
density data of the charge, the number of slices equiva-
lent to the mill was calculated to be nine. Other perti- The processes that take place inside a ball mill are not
nent data relating to the simulations are shown in known because of a lack of sensors that can function in
Table 3. the harsh environment. Instrumented balls are used to
It is known that the friction of most materials de- gather information with regard to motion and collision
creases with an increase in the speed of sliding.” of balls’3; however, the instrumentation itself is quite
Excessive speed of sliding was observed in simulating expensive and still under development. The mathemati-
the mill with swivel lifters. This situation required a cal model employed here involves the material proper-
change in the coefficient of friction for every mill speed ties of the colliding bodies as parameters. It is found
considered for simulation. Since DEM allows for that, unless these parameters are correctly used, the
changes in material parameters during simulations, a ball mill problem cannot be simulated correctly. With
linear variation of coefficient of friction with mill speed the correct choice of the parameters, accurate trends in
was used in simulating the motion of discs in the mill the variation of power with mill speed and filling are
fitted with triangular lifters. observed. A scheme is devised to utilize the two-dimen-
Figure 2 compares the results of simulation with sional simulation results to predict the power of a real
those of the experiment. Here, the variation of torque mill. The torque required to drive the mill at different
with the rotational speed of the mill is shown for two speeds is predicted, and the predicted power draft is
different lifter bar profiles. The overall agreement is compared with actual mill data. It is shown that the
found to be satisfactory. However, as was mentioned 2DMILL code can predict the power draft of actual
mills.
An analytical description of the multibody collision
Table 3. Data used for the simulation problem is difficult, but the discrete element method
Normal stiffness
enables the numerical simulation of ball charge motion.
400,000 N im
Shear stiffness 300,000 N / m
Currently, the simulation requires hours of super-
Coefficient of friction 0.1 computer time, depending on the number of balls to be
Coefficient of restitution: simulated; a 0.55-m-diameter mill consisting of 42 walls
Ball-ball impact 0.8 and filled with 148 balls, when run at 40% of its critical
Ball-wall impact 0.4
Critical time step 0.1 x 10-3 s
speed, requires 7.84 hours of CPU time on an IBM 3090
Mill diameter 0.545 m supercomputer.
Disc radius 0.012, 0.018, 0.025 m Realistic simulation of ball motion in three dimen-
Disc weight 0.06, 0.195, 0.525 kg sions is immensely difficult when the number of balls
Critical speed 57.8 rpm
Mill filling
exceeds a thousand. The difficulty is primarily due to
38.5
the enormous amount of computer time involved. As a

Appl. Math. Modelling, 1992, Vol. 16, November 603


Discrete element method in modelling ball mills: B. K. Mishra and R. K. Rajamani

result of this constraint-which is implicit in the nature Cundall, P. A. and Strack, 0. D. L. A discrete numerical model
of the discrete element method-DEM-based codes for granular assemblies. Georechnique 1979, 29, 47-65
Corkum, B. T. The discrete element method in geotechnical
have not developed significantly past the two-dimen- engineering. Master’s Thesis, Univ. of Toronto, Canada, 1989
sional stage. For ball mill simulations in which the Corkum, B. T. and Ting, J. M. The discrete element method in
problems involve a large number of distinct bodies to be geotechnical engineering. Tech. Rept., Dept. Civil Engrg.,
modelled, it is felt that a two-dimensional computer Univ. of Toronto, Canada, 1986
5 Cundall, P. A. and Strack, 0. D. L. The distinct element
code employing the slicing procedure outlined here is method as a tool for research in granular media, Part I. Tech.
most efficient. Rept., Dept. Civil and Mineral Engrg., Univ. of Minnesota,
USA, 1979
6 Mishra, B. K. Study of media mechanics in tumbling mills by
the discrete element method. Ph.D. Thesis, Univ. of Utah,
Acknowledgments USA, 1991
I Hofler, A. and Herbst, J. A. Ball mill modeling through micro-
The authors would like to thank the Utah Super Com- scale fragmentation studies: Fully monitored particle bed com-
puting Institute, which is funded by the state of Utah, minution versus single particle impact tests. Preprint of the 7th
European Symposium on Comminution, 1990
and the IBM Corporation for their support in this en- 8 Rose, H. E. and Sullivan, R. M. E. Ball, Tube and Rod Mills.
deavor . Chemical Publishing Co., New York, 1958
This research has been supported by the Department 9 Mishra. B. K. and Raiamani. R. K. Motion analvsis in tumbling
of the Interior’s Mineral Institute program administered mills by the discrete element method. KONA, Powder anj
by the Bureau of Mines through the Generic Mineral Particle 1990, 8, 92-98
Liddell, K. S. and Moys, M. H. The effects of mill speed and
Technology Center for Comminution under grant num- tilling on the behavior of the load in a rotary grinding mill. J.
ber G1175149. South African Inst. Min. Metall. 1988, 8(2), 49
Movs. M. H. A model for mill Dower as affected bv mill speed,
load volume and liner design.&Preprints of the 7th European
Symposium on Comminution, 1990
References Moore, D. F. Principles and Application of Triboloby.
Pergamon Press, Oxford, 1975
1 Cundall, P. and Hart, R. Numerical modeling of dis- Rolf, L. and Simonis, W. Energy distributions in ball mills.
continuation. DEM 1st U.S. Conference, Denver, Colo., USA, Preprint of the 7th European Symposium on Comminution,
1989 (proceedings in press) 1990, pp. 543-554

604 Appl. Math. Modelling, 1992, Vol. 16, November

Potrebbero piacerti anche