Powder Technology 101 1999.

6372

Simulation of grinding in a shaker ball mill

Dmitri Gavrilov, Oleg Vinogradov ) , William J.D. Shaw
Department of Mechanical and Manufacturing Engineering, Uniersity of Calgary, Calgary, Alberta, Canada T2N 1N4
Received 30 June 1997; revised 24 June 1998

Abstract
Little information currently exists on designing highly efficient ball mills for producing mechanically alloyed material. This work is an
attempt at understanding the mechanisms of alloying in order to design better equipments. The problem of efficiency of mechanical
alloying is investigated numerically by simulating the dynamics of a shaker ball mill. The model consists of two major parts: shaker ball
mill dynamics simulation and the grinding model. The dynamics simulation is used to find out how the number of collisions, the total
kinetic energy, and the rate of energy dissipation in the system depend on both the frequency and amplitude of vibrations and on the
number of balls. The model of an ideal gas is used to validate the dynamics model. Two models are developed to simulate the distribution
of radii of particles during the grinding process: a statistical model and a deterministic one, representing an average-case behavior of the
statistical model. Both models compare favorably against each other as well as against experimental results. q 1999 Elsevier Science S.A.
All rights reserved.
Keywords: Shaker ball mill; Molecular dynamics simulation; Grinding model; Optimization; Statistical model; Analytical model

1. Introduction
Mechanical alloying has been around since the late
1960s and the principles have been explained and established by Benjamin w1x. However, little information currently exists with respect to the design and efficient operation of grinding equipment for the production of this
material w2x. Commercial ball mills of a number of different types have been constructed and used, but most design
and operating information is obtained by trial and error.
Many of the attritor designs are based on the original work
by Szegvari and Yang w3x. A pilot attritor equipment has
been used to generate experimental operating information
w4x, which is a starting basis, but is specific to that equipment. Other studies in terms of applying theoretical analysis techniques are currently evolving w5,6x. Models put
forth for different processing equipment are based on
energy principles combined with kinematics w710x. However, due to the simplifying assumptions, these models are
and will always be quite limited in application. A computer simulation approach does not have the many limita-

)
Corresponding author. Tel.: q1-403-220-7187; Fax: q1-403-2828406; E-mail: vinograd@enme.ucalgary.ca

0032-5910r99r$19.00 q 1999 Elsevier Science S.A. All rights reserved.

PII: S 0 0 3 2 - 5 9 1 0 9 8 . 0 0 1 4 9 - 1

tions of simplified analytical models and as such can be a

powerful tool in studying and optimizing the grinding
processes in shaker ball mills. Such an approach is the
subject of this paper.
The process of grinding in a shaker ball mill starts when
two components in a granular form are placed in a shaker,
and ends when the final size of granules of these components becomes less than some specified level. The event of
grinding takes place when a material particle gets in
between the colliding balls. The efficiency of grinding,
defined as the time needed to reduce the size of granules
from the original to the final size, depends on the frequency of shaking, the number and size of balls, and the
geometry of the shaker. The number of balls should be
such that their relative density is small compared to the
shaker volume. The toughness of the material to be ground
is usually much less than that for the steel balls. During the
milling, the energy supplied to the balls from the moving
walls is transferred to the material particles at each collision, which results in more particles with refined sizes.
Thus, the efficiency of milling is determined by the number of collisions between the balls and the energy released
in collisions. The process of milling is best simulated by
the discrete dynamic model of the system, which is known
as a molecular dynamics model. The process of refine-

64

D. Garilo et al.r Powder Technology 101 (1999) 6372

ment, in terms of number of particles with a reduced size,

can then be deduced based on the rate of collisions. Both
the rate of collisions and the rate of refinement are the
subject of this paper.
A molecular dynamics approach is accurate and efficient for simulating granular materials in cases when the
volumetric density of particles is relatively small. The
concept of molecular dynamics was originally introduced
by Alder and Wainwright w11x to simulate the dynamics of
molecules. Later, it found many applications in the field of
computational physics, where the method became a standard simulation tool w12,13x.
The molecular dynamics approach in computer simulations of granular media remains methodologically the same
as in molecular physics except that the laws of motion and
interaction between the particles and the geometry of
particles are application-specific. For example, the particle
size can vary from 0.1 mm to up to 10 m w14x, the
collisions may be inelastic resulting in a loss of energy
this is usually taken into account by introducing restitution coefficients w15x., and also, particles may have different shapes w16x.
This paper presents the simulation methodology of
grinding in shaker ball mills based on a molecular dynamics sometimes called granular gas in granular materials
applications. approach. Therefore, the event-driven simulation scheme was used for these simulations. The simulation procedure comprises two steps. The first step deals
with the determination of the number of collisions in a
specified time for a given set of parameters, which are:
shaker geometry radius, height, and lid configuration:
plain or spherical., number, diameters and materials of
balls, and frequency and amplitude of shaking, as well as
some other parameters.
The second step deals with the investigation of the
grinding process in terms of the distribution of particle
radii during the milling process. Two approaches are developed: one based on the random sampling from all
possible system states, and another based on the conservation of mass after each event of collision. The first approach results in a discrete Monte Carlo simulation model,
while the second one results in a continuous model. Both
approaches give a similar distribution of particle sizes after
a specified number of ball collisions, and compare favorably with the corresponding experimental results.
An object-oriented software was developed to simulate
the dynamics of grinding. The validity of simulation was
confirmed when, based on numerical experiments, the
optimum parameters were identified so that the efficiency
of grinding was increased almost 10-fold. The experimental results presented in the paper concern a composite
material made out of two components: polyamide and
polyethylene milled together from the initial 4-mm size to
the final 1-mm size. The powder produced in this way is
then compressed at some temperature below the melting
point to obtain a composite material.

2. Shaker ball mill dynamics simulation

The motion of balls inside the shaker ball mill is
simulated in a 3D space. The simulation system comprises
the objects representing the cylinder itself and balls moving inside the cylinder. The balls are moving along the
parabolic trajectories in a gravitational field:
u t . s u 0 q z0 t q

gt 2
2

,z t . s z0 q gt ,

2.1 .

where u s u x , u y , u z .T is the position vector, z s x ,

y , z .T is the velocity vector and g s 0,0, g . is the
gravitation vector.
The motion of the cylinder is described as being harmonic in a vertical direction with the amplitude A and
frequency f. For simplicity of calculations, it is assumed
that the cylinder moves with constant acceleration during
each half-period of oscillations, i.e., along a parabolic
trajectory which corresponds to the second order spline
approximation:
z t . s 0 t q

at 2
2

2.2 .

where for the first half-period, 0 G t - Tr2, T s 1rf is

the period of vibrations., a s y32 Af 2 , s 8 Af, and for
the second half-period, Tr2 F t - T ., a s 32 Af 2 , s
y8 Af. The assumption that the periodicity of the cylinder
is of a parabolic form allows one to calculate collision
times directly by solving only second-order algebraic equations, while in the case of a sinusoidal form, the numerical
solution would be required.
The collisions between the balls and between the balls
and the cylinder are assumed to be central and frictionless,
i.e., the tangential velocity component of colliding balls is
conserved. The cylinder is assumed to have an infinite
mass, i.e., its velocity is not affected by the collisions. The
collisions are presumed to be instant and one-to-one only,
i.e., collisions between three or more objects are not
allowed. We also assume that all energy lost during the
collisions is transformed into heat.
Due to collisions, the normal component of velocity of
colliding balls changes according to the law < 1 y 2 < s
< 10 y 20 <, where 1 and 2 are the normal velocities of
balls after the collision, 10 and 20 , are the normal
velocities of balls before the collision, and is the
coefficient of restitution. From the laws of conservation of
energy and impulse, the following formula can be obtained:
1
ay
s
2
a

1qya
1ya

n 10
,
20

2.3 .

where a s 1 q . m1 .r m1 q m 2 ., m1 and m 2 are masses

of two colliding balls.
When a ball collides with the side wall of the cylinder,
its normal velocity changes according to the relationship

D. Garilo et al.r Powder Technology 101 (1999) 6372

s y 0 . For the collision between a ball and the lid of

the cylinder which is moving., the normal velocity of the
ball changes according to the relationship s c q c y
0 ., where c is the normal component of the lid velocity
at the time of collision.
The assumption of instantaneous-type collisions is justified because the duration of a collision is short compared
with the time scales of other events such as times between
collisions and the overall simulation time. The above
results are based on the theory of collision dynamics see,
for example, Johnson w17x.. The Hertz theory w18x is used
to estimate parameters of collision, such as collision duration and the volume of deformed area.
All collisions are assumed to be one-to-one. In the case
when more than two objects collide simultaneously, the
collision is handled consequently in pairs. For example, if
balls A, B, and C collide simultaneously, then the collision
between A and B is handled first, independently of C, then
the new velocities of A and B are calculated, and the
collision between C and either A or B or both A and B is
handled in the same manner. Such a collision resolution
scheme provides reasonably accurate results and significantly improves the efficiency of computations. The statistics of the simulations shows that the cases of simultaneous collisions between three or more objects are very rare.
A one-to-one assumption of collisions is in accordance
with the assumption of instantaneous-type collisions, since,
if the collision is assumed to be instant, the probability of a
simultaneous collision of more than two objects is zero.
But even if the duration of collision is taken into account,
this probability is still very small.
The purpose of the simulations was to investigate the
energy-transfer properties of the system, i.e., how efficiently the external energy of shaker vibrations is transformed into the internal energy of balls inside the shaker.
With each collision, the lost energy is transformed into
heating the balls as well as grinding. For a ball-to-ball
collision, this energy is
H1 s

m1 q m 2 nn2
m1 m 2

1y 2 .

2.4 .

65

contribution can be negative in the case when the lid and

the ball move in the same direction and the ball moves
faster than the lid.

3. Statistical model of grinding process

The results of the simulation obtained using the above
model are used to simulate the grinding process occurring
in the shaker ball mill. The objective is to investigate how
the parameters of ground particles, namely, their radii, are
changing with time.
In a straightforward approach, one could try to simulate
the system of material particles and steel balls as one
system in which the number of material particles modeled
by spheres. grows over time. The split of a material
particle takes place every time it is caught between the two
colliding balls. However, in this scenario, the number of
particles becomes so large with time i.e., one 1-mm
particle can be split into 1 million 0.01-mm pieces during
first seconds of shaking., that it becomes computationally
inefficient to keep track of the position and velocity of
each particle. In this respect, a statistical model offers an
alternative way of describing the distribution of particles
radii during the milling process.
The random process of grinding can be represented by a
directed graph with system states represented by the vertices, and transition states by the edges. Each edge is
weighted with the probability of transition to another state,
and the sum of weights of all edges going out of a vertex
is always equal to 1. This is demonstrated on a simple
example shown in Fig. 1, where the numbers indicating the
probabilities are given as illustrative.
In this example, a single particle in the state S0 is
transformed into a two-particle system in the state S1, and
then into a three-particle system different particle sizes. in
the state S2, and eventually into a four-particle system in
the state S3 final particle size in this example.. Note that
different particle sizes are placed in different baskets. In
this example there are only three baskets: B 0 , B 1 and B 2 .
Thus, the initial state S0 has one particle in the basket B 0 ,

and for a ball-to-cylinder collision, correspondingly,

H2 s

m1 nn2
2

1y 2 . ,

2.5 .

where n is the relative normal velocity of approach.

When a ball collides with the moving lid of the cylinder, the energy is contributed to the system of moving
balls. One part of this energy is spent on heating the ball
and grinding H2 . and another part on changing the kinetic
energy of the ball. The energy contributed to the system as
a result of a ball-to-cylinder collision is
C s 2 m1 nc nc y nn . ,

2.6 .

where c is the normal component of the cylinder lid

velocity at the time of collision. As we can see, the energy

Fig. 1. An illustrative example of the graph describing the grinding

process.

D. Garilo et al.r Powder Technology 101 (1999) 6372

66

and no particles in baskets B 1 and B 2 . When a collision

occurs, the probability is 0.2 that the particle in B 0 may
break up into two smaller ones, and then the system will
move into the state S1 with 0 particles in basket B 0 , two
particles in basket B 1 , and none in B 2 . The other possibility is to stay in the same state S0, i.e., the particle was not
broken in the collision. The probability of the latter event
is 0.8.
When the system is in the state S1 with two particles in
basket B1 , there are three possible outcomes of a collision.
In the first scenario, both particles are broken and moved
into the basket B 2 . The probability of this is 0.01. With
this transition, the system is moved into the final state S3,
when all particles are in the last basket B 2 . The second
possible outcome is when only one of two particles in
basket B 1 is broken. The probability of this is 0.18, and
with this transition, the system moves into the state S2
with one particle in basket B 1 and two particles in basket
B 2 . The third outcome is when no particles are broken due
to a collision, and the system stays in the state S1. The
probability of this is the remaining 0.81. Finally, when the
system is in the state S2, with one particle in B 1 , there are
two possible outcomes: first, when the particle is broken
with probability 0.1. and the system is moved into the
final state S3, or, second, when the particle is not broken
with probability 0.9., and the system stays in the state S2.
The above example represents all possible outcomes
during the grinding process from the initial to the final
state. It is clear that if the initial particle is to be reduced
by many orders of magnitude e.g., from 4 mm to 1 mm.,
the graph becomes very complicated, and the determination of all intermediate states becomes time consuming.
Instead of this, a random sampling technique can be used
to obtain a statistically representative sample of outcomes
after a specified number of transitions from the initial to
the final state. A single simulation run then represents a
random path from the initial S0. to the final state SN..
The probabilities of transition from state to state can be
found using the Monte Carlo simulation technique.
The statistical approach is based on the assumption that
the distribution of particles is uniform in space for each
particle size. It is also assumed that the condition for a
particle to be split is associated with the possibility of its
location between the colliding balls within the so-called
grinding volume see Fig. 2, where the grinding volume
is shaded.. It follows then from these two assumptions that
the probability of a particle of a specific size rA to be
inside the grinding volume is equal to the ratio of this
volume Vgr to the total volume occupied by the particles in
the shaker Vavail s Vshaker y V balls , where Vshaker is the volume of the shaker, and V balls is the total volume of balls
moving inside the shaker. The grinding volume can be
calculated approximately by the following formula
Vgr f 4p

R1 R 2
R1 q R 2

rA2 ,

3.1 .

Fig. 2. The grinding volume.

where rA is the radius of the particle A, and R 1 and R 2

are the radii of two collided balls.
Then, since it is assumed that the particle can be located
at any point of the available volume Vavail with equal
probability, the probability that the particle A is located
inside the grinding volume is
Pgr s

Vgr
Vavail

3.2 .

Eq. 3.2. is used to calculate the probability of grinding

for each particle in the system. One can see that the
probability of grinding decreases with the square of the
particle radius. This conforms to the experimental fact that
the milling rate is greatly reduced when particles become
small see Section 6 below..
We will assume that if a particle with radius rA participates in a collision, then, it is split into two equal particles
rA
so that the total volume is conserved..
with radii 3
'2
Assuming that all particles have equal radii r 0 initially, it
is sufficient to keep track only of the number of particles
Ni for each of the radii
ri s r 0 2yi r3 , i s 0 . . . k

3.3 .

Here, r k is the minimum particle radius, at which the

particles no longer break into smaller particles. Usually,
this radius is about 1 mm w19x.
As in the example of Fig. 1, the distribution of particles
according to their radii can be viewed as their placement in
the corresponding baskets so that each basket B i holds
particles of radius ri . Initially, all particles have radius r 0
and are placed in the basket B 0 . Let us denote the number
of particles in a basket B i as Ni . If a particle from basket
B i was involved in a collision, then one particle is removed from basket B i and two particles are added into
basket B iq1 Fig. 3..

D. Garilo et al.r Powder Technology 101 (1999) 6372

67

Fig. 3. Effect of collision on redistribution of particles in baskets.

Since all particles in a basket have the same radius,

their probabilities of grinding are calculated by the same
formula Eq. 3.2... Then, the number of particles from the
same basket involved in the collision is described by the
binomial distribution, which for the number Nic of particles with radius ri participating in a collision is as follows:
x
N yx
P Nic s x . s N pix 1 y pi . i , x s 0 . . . Ni ,
i

4.

where pi is the grinding probability Pgr ri . see Eq. 3.2..

for a particular particle radius ri . Note, that the volume
required for the number of crashed particles Nic can
exceed the grinding volume Vgr . However, for the binomial
distribution, the probability of this event decreases with the
number of particles.
In summary, when a collision between two balls takes
place, the particles are redistributed in baskets. For each
basket B i , i s 0 . . . k y 1, a random number of particles
0 F Nic F Ni involved in collision is determined, and then
Nic is deducted from Ni , and 2 Nic is added to Niq1. Nic is
a discrete random variable distributed by the binomial law
with parameters pi , Ni .. The process is stopped when all
particles are in the last basket B k , i.e., all material is
ground. The number of collisions, as well as the distribution of particle sizes during the grinding, are the objectives
of the simulation.

4. Analytical model of grinding process

Although the model described above is much more
efficient compared with the analysis of the complete graph
of grinding events, it is still quite slow. In a usual run, the
number of collisions required to grind the material can be
as large as 10 12 . Since each collision must be handled
separately, the run times could be quite long even on a
supercomputer. To solve this problem, a continuous analytical model is introduced, which represents the average-case
behavior of the statistical model.
The analytical model also considers baskets containing
balls with similar radii. However, instead of recording the

number of particles in a basket, the total mass of particles

in each basket is considered. Then the incremental mass
balance equation for a single basket is derived, which leads
to a differential form of the mass flow equation.
Let us denote the mass in a single basket by X i ,
i s 0 . . . k. Clearly, if a mass of one particle with radius ri
is m i which is proportional to ri3 ., then X i s Ni m i . Let us
consider how masses of particles in baskets are redistributed after a collision. Since Nic is a discrete random
variable distributed according to the binomial law with
parameters pi , Ni ., its mean value is pi Ni , i.e., on average, that many particles from basket B i are broken due to
collision. Then, the mass transferred from basket B i into
basket B iq1 is
pi Ni m i s pi X i , i s 0 . . . k y 1.

4.1 .

The mass transferred from the last basket B k is always

zero, since particles in this basket do not break. Now, we
will obtain a differential equation describing the change of
mass of the material in baskets over time. Let us assume
that collisions occur in the system with frequency l. Then,
during a time interval d t there will be ld t collisions. The
mass transferred from basket B i into basket B iq1 can be
represented as l pi X i d t. At the same time, the mass
l piy1 X iy1d t is transferred into basket B i from basket
i s 1 . . . k for all i s 1 . . . k no mass is transferred into the
first basket B 0 .. Then, the total change of mass in basket
B i during the time d t is
d X i s l piy1 X iy1 y pi X i . d t , i s 1 . . . k y 1.

4.2 .

Basket B 0 does not have any mass transferred into it,

and therefore, d X 0 s yl p 0 X 0 d t. The last basket B k does
not have any mass floating out, thus, d X k s l p ky1 X ky1d t.
The resulting differential system describing the averagecase behavior of the grinding process is
X s l PX ,

4.3 .

D. Garilo et al.r Powder Technology 101 (1999) 6372

68

where X s  X i 4T, i s 0 . . . k, and

Ps

y p0

...

p0

y p1

...

0
0

y p1

p2

...

...
0

...
0

...
0

...
...

...
y pky 2

...
0

...
0

0
0

0
0

0
0

...
...

pky 2
0

y pky1
pky 1

0
0

4.4 .

The initial conditions are set as

X 0 0 . s M , and X i 0 . s 0, i s 1 . . . k.

4.5 .

This linear differential system has the following analytic solution:

my1

Xm t . s M

is0

eyl p i t

pi

is0

, m s 0 . . . k.

p j y pi .

js0, j/i

4.6 .
The numerical results Section 6. confirm that the continuous analytical model provides a very good approximation of the discrete statistical model. Moreover, by the law
of large numbers, the average values of statistical model
parameters calculated for a large number of runs converge
to the analytical solution. Eq. 4.6. gives an explicit dependence of the outcome of the grinding process on the
system parameters.

5. Implementation
An object-oriented library developed to simulate the
dynamics of granular-type materials w20x was used to
simulate the shaker ball mill. The above library allowed
the creation of a needed application very quickly by
defining only application-specific properties of objects,
such as laws of motion and collision rules. The object-oriented approach also allowed flexibility in changing and
updating the simulation system. For example, when the
shape of the shaker needed to be changed, we had to create
a new class SphericalLid from the existing prototype class
Boundary and define the specific properties of the new
class, such as methods for determination of collision times
with balls, motion laws, and collision laws how a ball
changes its velocity when the collision occurs.. Then, the
new object was included into the simulation system without changing anything else.
The simulation system consists of objects representing
balls and the cylinder. The main algorithm is based on the
discrete event simulation scheme. The types of events
handled in the simulation include events representing prediction of the trajectory of an object, collision events, and
auxiliary events such as output of statistical parameters at
a specified moment of time to the file of results.

The simulation starts with assigning the balls random

positions and velocities inside the cylinder and scheduling
the predict trajectory events for each ball, as well as the
stop simulation event at a desired time.
The handling of a predict trajectory event includes
setting internal parameters of a ball x 0 ,z0 . and scheduling
the next predict trajectory event. Also all possible collision events for a ball are calculated and scheduled.
The handling of a collision event includes the following operations: all events scheduled for the currently colliding bodies are cancelled, the new parameters of collided
bodies are calculated, and the predict trajectory events
are scheduled at the present time for both colliding bodies.
With the increase of the number of balls, the collision
detection becomes a very time-consuming operation if
each pair of balls is to be checked for a collision. It is
noted w21x that the collision detection problem is the
efficiency bottleneck for most molecular dynamics problems. To optimize the solution of this problem, a special
data structurerectangular map, is introduced. The whole
simulation space is divided into a set of cells. Each ball
knows in which cell it is currently moving, and collisions
are detected only with balls in the neighboring cells of the
map. When a ball moves from one cell to another, only the
balls in the new neighboring cells must be checked for
collisions. The introduction of the collision detection optimization algorithm allowed achievement of very high event
processing rates up to 20,000 events per second without
visualization.. This efficiency is comparable with those
achieved on professional workstations in similar simulations w22x, which report processing efficiency about 25,000
on a Sparc-10 Sun workstation.
The software system consists of two parts: the first part
is used to simulate the dynamics of ball motion in the
shaker ball mill and to measure system parameters, such as
frequency of collisions. A special energy dissipation cutoff value was used to exclude low-velocity collisions
from consideration. This parameter becomes particularly
important for the simulations with high filling ratios, when
velocities of balls are comparatively low and many lowvelocity collisions occur. The point is that when two balls
collide with a low relative velocity, the energy dissipated
in a collision may not be enough to break any particles.
The value of the energy dissipation cut-off parameter must
be determined experimentally for the materials to be
ground.
The second part of the program deals with the simulation of grinding. It inputs the results of the shaker simulations and uses them in subsequent runs of the grinding
process model, then outputs particles radii distributions at
specified moments of time. The program implements both
statistical and analytical models for the grinding simulation.
The software runs on a PC with Pentium-166 and 32
MB of RAM running Windows95. The program was implemented in Borland Delphi 2.0 programming environ-

D. Garilo et al.r Powder Technology 101 (1999) 6372

69

Fig. 4. Parameters screen.

ment. The parameters of the system are specified in a

specialized dialog box Fig. 4..
After the parameters have been entered, the simulation
is run. If the user specified that hershe desires to see the
simulation process visualized, then a 2D projection of the
simulation space is periodically displayed on the screen.

The parameters displayed during simulation include

virtual simulation time i.e., internal time of the simulated
system., how much real time is left until the end of
simulation, as well as virtual and actual event rates.
After the shaker simulation is finished, the user can
switch to the grinding simulation mode, where the particle

Fig. 5. Visualization of the grinding process.

70

D. Garilo et al.r Powder Technology 101 (1999) 6372

Fig. 6. Energy dissipation vs. amplitude frequencys 52.6 Hz..

radii distribution is plotted against the number of collisions

Fig. 5..

6. Numerical results
A number of numerical experiments have been made.
The first series of experiments investigated the effect of
the shakers external parameters, such as the frequency and
amplitude of vibrations, on the internal parameters, such as
rate of energy dissipation due to heating of the balls,
number of collisions, kinetic energy of the balls, and the
rate of energy flow into the system. One of the interesting
results was that the dissipation of energy inside the system

H is proportional to the amplitude of shaker vibrations A

and to the third degree of frequency f, H ; Af 3 see Figs.
6 and 7; the following data was used in experiments: 100
steel balls, each of radius 3.2 mm; cylinder: height s 103
mm, radius s 31.5 mm..
The finding that internal milling energy is proportional
to the third power of operating frequency is in keeping
with the results of Streletskii w8x; however, the linear
behavior of energy with amplitude is not. The effect of the
density of ball packing and the shape of the shaker cylinder on the efficiency of the shaker is currently being
investigated.
A model of an ideal gas was constructed to validate the
simulation software. The ideal gas was simulated by a

Fig. 7. Energy dissipation vs. frequency of vibrations amplitudes 10.9 mm..

D. Garilo et al.r Powder Technology 101 (1999) 6372

71

Fig. 8. Distribution of velocities in an ideal gas model.

system of perfectly elastic small balls representing

molecules of ideal gas moving inside the motionless cylinder. All collisions were supposed to be elastic, i.e., there
was no energy loss in the system. Initially, all balls were
assigned equal velocities in random directions. The relationship between the kinetic energy of the balls which
remains constant., pressure on the walls of the cylinder,
and the volume of the cylinder was investigated. Numerical experiments have shown that the PV s RT gas law is
satisfied. Here P is the pressure of the gas, V is the
volume of the cylinder, T is the temperature of the gas

which is assumed to be proportional to the internal energy

of the gas, i.e., kinetic energy of balls., and R is a
constant. This result is of a particular interest, because this
relationship does not depend on the shape of the cylinder,
i.e., the ratio between the height and the radius.
Another validating result is that the distribution of
velocities of molecules, which are all assigned equal values at the beginning of the simulation, soon comes to the
Maxwell distribution of velocities Fig. 8.. The initial
magnitude of velocities in this experiment was set to 1000
mmrs for all balls.

Fig. 9. Particle radii distribution after 24 h of milling.

72

D. Garilo et al.r Powder Technology 101 (1999) 6372

In the second series of experiments, we investigated the

particle radii distribution during the milling process. The
results presented in the following figures were obtained for
the following shaker configuration: ball-to-ball coefficient
of restitutions 0.980, ball-to-wall coefficient of restitution
s 0.950, shaker radius s 31.5 mm, shaker height s 103.0
mm, shaker frequencys 52.6 Hz, shaker amplitudes 10.9
mm, productive collision cut-offs 0.0001 J, ball radius s
3.200 mm, ball number s 125, ball density s 7850.0
kgrm3 steel.. For this set of parameters, the rate of
approximately 100,000 collisions per second was obtained.
The collision rate becomes quite stable after 0.1 s of
shaking. For this reason, the simulation time to determine
the number of collisions was 1 s, and the resulting effective collision rate was determined as l s 100,000 collisions per second.
The distribution of radii of particles after 24 h of
milling are presented in Fig. 9. Since the effective rate of
collisions for the simulation was set to 100,000 collisions
per second, the resulting distribution is obtained after
approximately 8.64 = 10 9 collisions. The experimental data
in Fig. 9 is taken from Ref. w19x. The close correspondence
between both simulation and experimental results is clearly
seen. The solutions obtained from statistical and analytical
models are almost equivalent.
Another validating result is that the distribution of radii
is almost independent of the initial particle radius, i.e., the
milling rate is significantly higher for larger particles, and
they are very quickly ground to smaller ones. Another very
interesting result, which is also in correspondence with the
experiments w19x, is that the milling rate is independent of
the amount of material put into the mill.

7. Conclusions
A methodology and a software tool were developed to
design more efficient shaker ball mills.
A discrete-event molecular dynamics simulation engine
was developed to investigate the collision rate and energy
transfer parameters of shaker ball mills. The input parameters of the model included geometry of the cylinder, filling
ratios and ball characteristics.
Statistical and deterministic models of the grinding
process compared favorably against each other and against
experimental results given in Ref. w19x.
The developed methodology allows creation of a complete design and analysis tool for simulation of shaker ball
mills, which allows optimization of the milling efficiency.

References
w1x J.S. Benjamin, Dispersion strengthened superalloys by mechanical
alloying, Metallurgical Transactions 1 1970. 29432951.

w2x Milling of Brittle and Ductile Materials, Metals Handbook, 9th edn.,
Powder Metallurgy, Vol. 7, ASM, 1984, pp. 5770.
w3x A. Szegvari, M. Yang, Attritor grinding and dispersing equipment,
Proceedings of Seminar on Dispersion of Pigments and Resins in
Fluid Media, Kent State Univ., OH, May 9, 1984, pp. 17.
w4x H. Kimura, M. Kimura, F. Takada, Development of an extremely
high energy ball mill for solid state amorphizing transformations,
Journal of the Less-Common Metals 140 1988. 113118.
w5x B.N. Babich, V.A. Djatlenko, A mathematical model for the process
of mechanical alloying, in: F.H. Froes, J.J. deBarbadillo Eds..,
Structural Application of Mechanical Alloying, ASM, 1990, pp.
287292.
w6x T.H. Courtney, D.R. Maurice, Some fundamental physics of the
mechanical alloying process, in: A.H. Clauer, J.J. deBarbadillo
Eds.., Solid State Powder Processing, TMS, 1990, pp. 319.
w7x A.S. Lazarev, A.A. Kolesnikov, V.A. Korol, Mass and energy
transfer during the mechanical alloying of powder in an attritor,
Soviet Powder Metallurgy and Metal Ceramics 25 8. 1986. 613
617.
w8x A.N. Streletskii, Measurements and calculation of main parameters
of powder mechanical treatment in different mills, Proceedings of
the 2nd International Conference on Structural Application of Mechanical Alloying, Vancouver, B.C., September 2022, 1993, pp.
5158.
w9x P.G. McCormick, H. Huang, M.P. Dallimore, J. Ding, J. Pan, The
dynamics of mechanical alloying, Proceedings of the 2nd International Conference on Structural Application of Mechanical Alloying,
Vancouver, B.C., September 2022, 1993, pp. 4550.
w10x T.H. Courtney, B.J.M. Aikin, D. Maurice, R.W. Rydin, T. Kosmac,
Local and global modeling of mechanical alloying, in: J.J. deBarbadillo, F.H. Froes, R. Schwarz Eds.., Proceedings of the 2nd
International Conference on Structural Application of Mechanical
Alloying, ASM, 1993, pp. 113.
w11x B.J. Alder, T.E. Wainwright, Phase transition for a hard sphere
system, Journal of Chemical Physics 27 1957. 1208.
w12x P.S. Lomdahl, D.M. Beazley, P. Tamayo, N. Grnbech-Jensen,
Multi-million particle molecular dynamics on the CM-5, International Journal of Modern Physics C Physics and Computers. 4 6.
1993. 10751084.
w13x D.M. Beazley, P.S. Lomdahl, N. Grnbech-Jensen, P. Tamayo, A
high performance communications and memory caching scheme for
molecular dynamics on the CM-5, Proceedings of the 8th International Parallel Processing Symposium, Cancun, Mexico, IEEE Computer Society, Los Alamos, 1994, pp. 800809.
w14x H. Salo, Numerical simulations of dense collisional systems: II.
Extended distribution of particle sizes, Icarus 96 1992. 85106.
w15x S. Luding, E. Clement, A. Blumen, J. Rajchenbach, J. Duran,
Studies of columns of beads under external vibrations, Physical
Review E 49 1994. 16341646.
w16x T. Poschel,
V. Buchholtz, Molecular dynamics of arbitrarily shaped

granular particles, Journal de Physique I France. 5 1995. 1431

1455.
w17x K.L. Johnson, Contact Mechanics, Cambridge Univ. Press, 1985.
w18x W. Goldsmith, Impact, Edward Arnold, London, 1960.
w19x W.J.D., Shaw, J. Pan, M.A. Gowler, Property relationships of some
new MA polymers, Proceedings of the 2nd International Conference
on Structural Applications of Mechanical Alloying, Vancouver, B.C.,
Canada, September 1993, pp. 431437.
w20x D. Gavrilov, O. Vinogradov, Object-oriented library for simulation
of granular-type materials, Proceedings of the 1994 Summer Computer Simulation Conference, San Diego, CA, 1994, pp. 5156.
w21x B.D. Lubachevsky, How to simulate billiards and similar systems,
Journal of Computational Physics 94 1991. 255283.
w22x G.A. Kohring, Studies of diffusional mixing in rotating drums via
computer simulations, Journal de Physique I France. 5 1995.
15511561.