Charge Behaviour and Power Consumption I

Int. J. Miner. Process. 63 Ž2001.
79–114
www.elsevier.nlrlocaterijminpro
Charge behaviour and power consumption in ball

mills: sensitivity to mill operating conditions,
liner geometry and charge composition
Paul W. Cleary )
CSIRO DiÕision of Mathematical and Information Sciences, PriÕate Bag 10, Clayton South MDC, Clayton,
Victoria, 3169, Australia
Received 21 January 2000; received in revised form 7 November 2000; accepted 17 November 2000
Abstract
Discrete element method ŽDEM. modelling has been used to systematically study the effects of
changes in mill operating parameters and particle properties on the charge shape and power draw
of a 5-m ball mill. Specifically, changes in charge fill level, lifter shape Žeither by design or wear.
and lifter pattern are analysed. The effects of changes to the properties of the charge Žball fraction,
ball and rock shape, type of ball and rock size distributions and the lower cutoff of the rock size
distribution. can all be interpreted in terms of their effects on the shear strength of the charge.
Some changes increase the shear strength leading to higher dynamic angles of repose of the
charge, higher shoulder positions and higher power consumption for sub-critical speeds. For
super-critical speeds, they lead to lower power consumption, due to lower particle mobility as the
particles lock together better. Changes to the charge that weaken the interlocking of particles have
the opposite effect on the charge shape and power consumption. The combination of these effects
means that the speed for which peak power consumption occurs is predominantly determined by
the shear strength of the charge material and the fill level. This demonstrates the sensitivity of mill
behaviour to the charge characteristics and the critical importance of various assumptions used in
DEM modelling. q 2001 Elsevier Science B.V. All rights reserved.
Keywords: comminution; ball mill; discrete element method; power draw; lifter
)
Fax: q61-3-9545-8080.
E-mail address: [email protected] ŽP.W. Cleary..
0301-7516r01r$ - see front matter q 2001 Elsevier Science B.V. All rights reserved.
PII: S 0 3 0 1 - 7 5 1 6 Ž 0 1 . 0 0 0 3 7 - 0
80 P.W. Clearyr Int. J. Miner. Process. 63 (2001) 79–114
1. Introduction
Ball mills are used in mineral processing for particle size reduction. These consist of
a rotating cylindrical shell of up to 7 m in diameter with axial lengths greater than this.
Lifter bars and sacrifical liner plates are bolted to the inside of the mill shell. Rock
generally arrives from a crusher or perhaps a semi-autogenous ŽSAG. mill Ždepending
on the design of the milling circuit and the mineral being processed. and enters the feed
end of the mill. Grinding media consisting of steel balls, with typical diameters from 25
to 75 mm, is already present in the mill along with any rock that has not yet overflowed
through the discharge end. Typical power consumption for a 5-m diameter by 7-m-long
ball mill is between 2.5 and 3.5 MW. The actual proportion of this energy usefully used
in size reduction is thought to be very low, perhaps in the range of 1–5%. Significant
financial and environmental benefits can be obtained by improving this efficiency even
slightly. Substantial benefits are also potentially available through higher downstream
recoveries of the liberated mineral if the product particle size distributions ŽPSD. can be
tailored for subsequent flotation or other separation processes. Large saving would also
be possible if the lifespan of the mill liners could be extended by reducing peak wear
rates.
Microscale simulation of ball and other types of mills by discrete element methods
ŽDEM. offers the opportunity of better understanding the internal mill dynamics and for
developing improvements to mill design and operation that can lead to large increases in
mill efficiency and throughput. DEM modelling has been used for a very wide range of
applications in geophysics Žsuch as landslides by Cleary and Campbell, 1993; Campbell
et al., 1995 and ice flows by Hopkins et al., 1991. and industry Žball mills by Mishra
and Rajamani, 1992, 1994; Cleary, 1998a, SAG mills by Rajamani and Misha, 1996 and
Cleary, 2001, vibrational segregation by Haff and Werner, 1986; Ristow, 1994 and
Cleary, 1998c, rotating drums by Ristow, 1994; Poschel ¨ and Buchholtz, 1995 and
others, sampling by Robinson and Cleary, 1999, draglines by Cleary, 1998b, 2000,
hopper flows by Holst et al., 1999; Ristow, 1994 and many more.. Of particular
relevance is experimental validation of DEM predictions for a centrifugal mill by Cleary
and Hoyer Ž2000., which showed excellent agreement both for the charge motion and
the power draw.
This paper is the second installment of an ongoing numerical study of ball mills. The
first installment ŽCleary, 1998a. described the basic modelling of a 5-m ball mill.
Predictions of charge profile, charge segregation, power draw, torque, liner wear and
collisional force distributions were made for a range of mill conditions. The sensitivity
of the power draw to some particle properties and some operating parameters was
analysed in a preliminary way. The power draw was found to be relatively insensitive to
material properties such as the coefficients of restitution and friction, with variations
generally around 5% or less for normal operating conditions.
An important observation from this work was that the behaviour of a charge
consisting of balls and rocks was reasonably different to that of a charge consisting
purely of balls Žas frequently used in earlier DEM simulation, Mishra and Rajamani,
1992, 1994.. The toe position was found to be higher and the surface profile was more
P.W. Clearyr Int. J. Miner. Process. 63 (2001) 79–114 81
clearly defined with a more bi-linear shape when the rock fraction was included. The
power draw was also found to be noticeably lower.
A simple wear model also allowed predictions of liner wear to be made leading to
plausible estimates of lifter life spans. The spatial distributions of wear were also
consistent with anecdotal information from mill operations. Finally, two mechanisms
were also identified for producing size segregation in mills. One operates at low speed
and concentrates large particles around the outside of the charge. The other operates at
high speeds and causes large particles to accumulate in the center of the charge. At
intermediate speeds, these mechanisms cancel.
In this paper, we analyse the effect of changes in particle size and shape distribution
Žfor both balls and rocks., lifter pattern and profile, and fill level on power draw and
charge behaviour for rotation rates between 50% and 130% of the critical speed at which
particles begin to centrifuge. All simulation parameters are held constant so that the
effect of each of the above influences can be independently evaluated. This study is
intended to illustrate the functional dependences of mill power on various mill condi-
tions and charge properties. These dependences are presently still somewhat qualitative,
since detailed data for comparison is difficult to obtain and the current DEM model is
still somewhat simplified. Detailed validation and the use of large scale 3D DEM should
soon allow these to be understood quantitatively as well. The dependences analysed in
this paper are based on the relative changes of predicted power with changes in mill
design and operational parameters and not on the absolute accuracy of the power
predictions. We also evaluate the importance of various assumptions Žsuch as whether
particle shape can be neglected. that are made to render feasible the simulation of this
complex industrial system. Understanding the effects of different assumptions is critical
to developing quantitatively accurate DEM mill modelling.
2. The discrete element method
The essential feature of discrete element simulation is that collisional interactions of

particles with each other and with their environment are detected and modelled using a
contact force law. Equations of motion are then solved for the particle motions and for
the motion of any boundary objects with which the particles interact. In rotating mills,
the axial migration of particles is slow compared to the circulating motion in the plane
orthogonal to the mill axis. This allows the mill to be plausibly modelled in two
dimensions. It is important to be able to model the complex boundary geometry of the
liner, including any lifters. When operating in two dimensions our code uses boundary
objects that are constructed from segments, which can be line segments, circular
segments or discs. Nearly arbitrary shaped two-dimensional mill cross-sections can be
constructed.
In this paper, the particles are modelled either as disks or super-quadrics with varying
size distributions. Super-quadrics are defined by the equation x n q Ž yrA. n s s n , where
the power n determines the blockiness of the resulting particle and A determines the
aspect ratio. Aspect ratios of up to 12:1 and blockiness factors of up to 20 can be used.
The general DEM methodology and its variants are well established and are described
in review articles by Campbell Ž1990., Barker Ž1994. and Walton Ž1994.. Here we use a
reasonably standard implementation that is described in more detail in Cleary Ž1998a,c..
Briefly, the particles are allowed to overlap and the amount of overlap D x, and normal
Õn and tangential Õt relative velocities determine the collisional forces via a contact force
law. There is a range of possible contact force models available in the literature that
approximate the collision dynamics to various extents. We use a linear spring-dashpot
model. For more complex models, see Walton Ž1994. and Schafer ¨ et al. Ž1996.. The
normal force:
Fn s yk n D x q Cn Õn , Ž 1.
consists of a spring to provide the repulsive force and a dashpot to dissipate a proportion
of the relative kinetic energy. The maximum overlap between particles is determined by
the stiffness k n of the spring in the normal direction. Typically, average overlaps of
0.1–1.0% are desirable, so a normal spring constant of 8 = 10 6 Nrm is used in all these
simulations. The normal damping coefficient Cn is chosen to give a required coefficient
of restitution ´ Ždefined as the ratio of the post-collisional to pre-collisional normal
component of the relative velocity.:
ln Ž e . mi m j
(
C n s 2g m i j k n , where g s y
(p 2
q ln2 Ž e .
and mi j s
mi q m j
,
is the reduced mass of particles i and j with masses m i and m j , respectively. This arises
from the analytic solution of the normal collision Eq. Ž1. for two such particles. Each
particle can have a different coefficient of restitution Žand corresponding damping
constant., but we generally use the same value for all the particles because little
experimental data on particle to particle variations in ´ is available.
The tangential force is given by:
½
Ft s min m Fn ,k tbiÕt d t q Ct Õt ,5 Ž 2.
where the integral of the tangential velocity Õt over the collision behaves as an
incremental spring that stores energy from the relative tangential motions and represents
the elastic tangential deformation of the contacting surfaces. The dashpot dissipates
energy from the tangential motion and models the tangential plastic deformation of the
contact. The total tangential force Žgiven by the sum of the elastic and plastic
components. is limited by the Coulomb frictional limit at which point the surface
contact shears and the particles begin to slide over each other. We choose the tangential
spring stiffness to be k t s a k n and Ct s 'a Cn . For these contact force laws ŽEqs. Ž1.
and Ž2.. the frictional behaviour in two dimensions is insensitive to the choice of a for
values in the range 0.1 F a F 10 for both oblique collisions and for sliding along
inclined slopes. The value used throughout this study is a s 10.
The contact force model requires both the coefficient of elasticity ´ and friction m
for the particles to be supplied. Such quantities are in practice very difficult to obtain
experimentally and vary from particle to particle and during the flow as the particles are
ground. Typical values are therefore chosen, ´ s 0.3 and m s 0.75. In Cleary Ž1998a., it
was shown that for normal mill operating conditions, the power draw was relatively
insensitive to the choices of these two physical quantities. The same values are used
throughout all the simulations in this paper in order to allow the effect of each changed
condition to be independently evaluated.
The three key parts of our DEM implementation are the following.
v A search grid is occasionally used to construct a particle near-neighbour list. Using
only particle pairs in this list reduces the force calculation to an O Ž N . operation, where
N is the total number of particles. Using such methods allows very large problems to be
solved. Typically, problems with up to 100,000 particles are now easily soluble on
single processor workstations.
v The collisional forces for each collision are estimated using the spring-dashpot
model for each pair of particles in the near-neighbour list.
v All the collisional and other forces acting on the particles are summed and the
resulting equations of motion are integrated:
x˙ i s u i , u˙ i s Ý Fi j q g , Ž 3.
j
u̇ i s v , v˙ i s Ý Mi j , Ž 4.
j
where x i , u i and Fi j are the position, velocity and collisional forces on particle i, and u i
and v i are the particle orientation and spin produced by the moments Mi j and g is the
gravity vector. The integration scheme is a second order predictor–corrector. Between
15 and 25 timesteps are required to accurately integrate each collision. This necessitates
very small timesteps Žtypically 10y4 to 10y6 s depending on the controlling length and
time scales of each problem.. If the integration is accurate then the coefficient of
restitution from which the damping coefficient was chosen should be recovered.
Quantitative predictions of wear rates and distributions, collision forces, dynamic
loads on boundaries, power consumption, torques and flow rates, sampling statistics,
mixing and segregation rates and many other quantities can be made from the informa-
tion available in DEM simulations. For more details on the simulation method and on
the data analysis, see Cleary Ž1998c.. In this paper, we concentrate on predictions of
power draw.
3. Calculation of torque and power
Every particle collision with the mill liner produces a torque on the mill. For each
timestep these individual torques are summed to give the instantaneous torque on the
mill. Only particles in contact with the mill contribute. If, for example, particles are in
free flight, then they do not contribute to the torque until they strike the liner at the
bottom of their flight. The product of the total mill torque with the angular mill speed Žin
radrs. gives the instantaneous power consumption required to maintain the mill moving
at the specified speed against this applied torque.
An alternative way of calculating the power draw is to sum the energy dissipation in
each timestep for all the energy dissipation mechanisms for all particle collisions. The
energy dissipation in the normal direction is the product of the normal dashpot damping
force with the normal approach speed. In the tangential direction, it is the product of the
tangential speed of the particle contact with the tangential dashpot force if the contact is
not sliding, or the product of the Coulomb friction and the tangential contact speed if the
contact is sliding. This measure of power is much more difficult to obtain accurately
because it is made up of many small contributions distributed over many timesteps. But,
if the integration of the collisions is accurate then the average power calculated in this
way should equal the power predicted from the torque. Since the torque-based power is
more accurate we use this for all the results presented here.
The instantaneous power consumption is highly variable due to impulsive nature of
the interactions. Averaging the power over several seconds gives a relatively invariant
measure of the power draw that can be meaningfully compared to laboratory and field
measurements. The choice of the averaging period is important because it determines the
variability and reproducibility of the average. In principle, the longer the averaging
period is, the less variability is present in the average. Unfortunately, whenever there is
range of particle sizes present in the charge, then it can and does segregate rapidly. The
amount of segregation increases the power consumption ŽCleary, 1998a. and it increases
strongly with rotation rate, leading to rises in power consumption as high as 15–20% for
speeds of 105–110% critical. To make valid comparisons between different mill speeds
and between cases with different particle distributions, the averaging period must be
restricted so that the averaging is completed before any substantive segregation has
occurred. The averaging period is therefore chosen to be between 3 and 10 s for almost
all cases presented here. The first 3 s are omitted since this is the approximate time
required for the slower rotating mills to reach equilibrium. This averaging period
produces several hundred instantaneous power readings, which are calculated every
1000 timesteps. From a purely statistical perspective, this means that the errors in the
power are less than "2%.
The length of the averaging period also depends on the number of particles in the
simulation. When rocks are modelled as well as the balls, then the above averaging
period is a satisfactory compromise between having enough data to get good statistical
averages, but short enough to avoid structural changes in the flow due to segregation.
When only balls are modelled, then much longer averaging periods are required to
compensate for the increased fluctuations produced by the much smaller numbers of
particles. Averaging periods of between 180 and 240 s are then required to give stable
averages. Shorter averaging periods produce significant statistical variations of the
predicted power vs. mill speed curve that do not have a physical basis. In general, a
properly averaged power vs. speed curve should be reasonably smooth. DEM simulation
results that demonstrate large irregular fluctuations should be treated with suspicion.
4. The ‘standard’ mill configuration
The ‘standard’ ball mill used throughout this study has an internal shell diameter of 5
m. There are 23 lifter bars spaced circumferentially around the mill shell. The standard
lifter is symmetric with an 858 face angle and is 210 mm wide at the top and has a
height of 200 mm above the liner. The wear plates between the lifters protect the mill
shell and reduce the internal diameter to 4.8 m. The ball mill rotates clockwise at
various constant fractions N of the critical speed of 19.5 rpm, at which the particles
begin to centrifuge.
The charge consists of both steel balls and rocks with a range of particle size
distributions ŽPSD.. The default charge consists of steel balls with diameters uniformly
distributed between 50 and 100 mm and rocks with diameters uniformly distributed
between 5 and 50 mm. The rock frequency is 10 times higher than the ball frequency,
leading to a charge consisting of around 45% Žby volume. of rock and 55% Žby volume.
of balls. The ball charge fraction actually varies between 53% and 58% due to the
inherent randomness of the particle generation process. This charge composition is used
for most of the power predictions here. Fill levels of 40% and 50% are used for most of
the comparisons in this paper.
The standard configuration contains 5265 particles when the fill level is 40% and
requires a computation time of 1.14 hrs of simulation on a 500 MHz alpha XP1000
workstation when using a conservative timestep on 9.04 = 10y6 . This is about 4.5 h per
mill revolution.
5. Effect of rotation rate for the standard ball mill
5.1. Charge behaÕiour
Fig. 1 shows typical charge shapes predicted for our ‘standard’ 5 m ball mill and
charge Ždescribed above. filled to 40% Žby volume. for four rotation rates that span the
typical range of operational speeds. For such a charge with a realistic size range, the free
surface of the charge is well defined at low and intermediate rotation rates. For
N s 60%, material moves with the mill rotation around the bottom of the charge and up
the left side of the mill to the shoulder. Here the net force on the particle declines to zero
and the charge fluidises. Most of the material then flows down the steep free surface into
the horizontal toe region. This is generally described as the cascading part of the flow. A
small amount of material in the shoulder region is trapped between the lifters and is
ballistically ejected into the mill cavity. This material drops onto the charge in the toe
region and sometimes onto the unprotected liner of the mill. This part of the flow is
generally termed as cataracting.
The particles are shaded according to their speed with light grey being low speed and
dark grey being high. The fastest particles are the cataracting ones as they impact on the
toe. The next fastest group of particles is the cascading or avalanching ones making up
the free surface. They reach a peak speed just below the center of the mill and slow
before arrival at the toe. The next fastest group of particles are the ones rotating with the
mill near the shell. There is a semi-circular band Žindicated by the light grey. of particles
that move quite slowly. At the very center of this band is the center of the circulation
pattern for which the particle speed is zero. Changes in the shading indicate shear that
leads to particle size reduction by both abrasion and attrition. Such shear regions are
clearly visible between the faster moving particles around the outside of the mill and the
Fig. 1. Charge shape predicted for our ‘standard’ ball mill filled to V s 40% with rocks Ž5–50 mm. and steel
balls Ž50–100 mm. for different rotation rates: Ža. N s60%, Žb. N s 70%, Žc. N s80% and Žd. N s90%.
The particles are shaded according to their speed.
slowly moving interior particles and also between the cascading surface particles and the
slow moving interior. The largest amounts of shear Žindicated by the sharpest changes in
grey scale. occur in toe region where material abruptly changes its motion from
cascading down the surface to being pulled under the bulk of the charge by the lifters
and swept to the left.
At N s 70%, ŽFig. 1b., the shoulder position moves slightly higher and the toe
moves slightly lower. The inflection point on the free surface moves much closer to the
center of the mill as the free surface profile becomes almost bi-linear. The amount of
cataracting material has increased significantly and it possesses a clear banded structure.
Each band is slightly curved, is almost vertical but tipped to the right a bit at the top and
leads to the front face of one of the lifters above. Each lifter leaving the shoulder region
acts somewhat like a filled bucket. As the lifter rises and its angle changes, the particles
trapped inside gradually pour out with the last particles departing as the lifter reaches
vertical. Each band corresponds to the material that poured from the space in front of the
lifter to which it leads. The impacts of cataracting material with the charge and mill liner
in the toe region are therefore not continuous in time, but are bunched together and
produce a cyclic variation in the impact forces in this region. For some rotation rates,
spectral analysis of the instantaneous power draw detects a signal with a frequency
given by the product of the mill frequency and the number of lifters that corresponds to
the one that such cyclic impacts should produce. The distribution of particle speeds, and
therefore shear, within the charge is little changed from the lower speed. The relative
speed of the particles moving around the shell with the lifters compared to the speed of
the cascading free surface particles has increased slightly.
Similar changes continue to occur when the mill speed is increased to N s 80% Žsee
Fig. 1c.. The amount of cataracting material has again increased strongly and their
trajectories take particles close to the lifters above. The banded structure of these
particles is more clearly visible. The shoulder again moves slightly higher and the toe
marginally lower. The speed of the particles moving with the lifters and shell at the
bottom again increases relative to the speed of the cascading particles. The semi-circular
band of slow moving particles has become better-defined as the shear to either side
becomes more concentrated. Near the shoulder region, the centrifugal force causes the
free surface to become less sharply defined as this balances the gravitational force.
By N s 90%, ŽFig. 1d., the outermost cataracting particles barely move away from
the shell and collide with the backs of the lifters slightly above the centerline of the mill.
This would produce high wear on the lifters and is very undesirable. The fluidisation of
the upper cascading free surface by the centrifugal force is now strong and the
concentration of the shear into two semi-circular bands from the toe to the shoulder
continues. For higher rotation rates particles become permanently trapped between the
lifters and are then centrifuged for N ) 100%. The depth of the centrifuged layer
increases with the mill speed N.
5.2. Variation of torque and power with mill speed
Fig. 2 shows the variation of the average torque and power with mill speed for the
ball mill described above with a fill level of 50%. The torque increases slowly until the
peak is attained around N s 80%. As the amount of cataracting increases, the balance of
the charge improves and the torque required to maintain the asymmetric charge position
declines steadily. At N s 100%, the charge begins to centrifuge and the rate of decline
of the torque with speed increases.
The power, which is given by the product of this torque curve with the mill speed,
has a much rounder peak centered on N s 100% before dropping sharply for higher
rotation rates. For a mill operating at a typical 80% of critical, the power predicted by
the model is 3 = 10 5 Wrm of mill length. For a typical 7-m-long mill, the power
required is therefore predicted to be 2.1 MW. This is consistent with a peak motor
power rating of 3.3 MW that would commonly be used for a mill of this size.
Field data is very difficult to obtain for comparison with the DEM results so we
instead compare with some of laboratory observations. Liddell and Moys Ž1988. used a
0.545 m mill Žwith smaller and more rounded lifters than the ones used here. and found
that the torque peaked at 72% of critical speed for the 50% fill level used in the previous
Fig. 2. Average torque and power as a function of mill speed N for the standard ball mill filled to 50% by
volume. This is the base case for the following comparisons.
figure. This is moderately lower than the 80% speed predicted here by the DEM.
Similarly, Liddell and Moys Ž1988. found the peak power occurred at N s 82% for this
fill level, compared to 100% for the DEM prediction. This quantitative discrepancy
arises because of differences in the two mill liners Žthe experimental lifters were smaller
and more rounded and therefore less able to lift the charge. and from some of the
simplifying assumptions made in the DEM model. It is therefore very important to
ascertain the effects of various modelling assumptions on mill power predictions and to
incrementally improve the DEM model so that these predictions become quantitatively
better. For example, it will be shown later in this paper that the inclusion of rock shape
is able to move the peak power from around 110% critical down to around 95%
compared to 88% for these experiments. The inclusion of particle shape is able to
explain most of this difference and is critically important for improving the quantitative
accuracy of DEM mill predictions. Three-dimensional effects and the inclusion of the
pulp slurry are also expected to influence this.
6. The effect of mill fill level
6.1. Charge shape and flow behaÕiour
Fig. 3 shows the changes in charge structure with decreasing fill level for our
standard mill. When decreasing the fill level from 50% to 40%, the main change in the
Fig. 3. Charge shape predicted for a ball mill rotating at 80% of critical speed and filled with rocks and balls
with differing fill levels: Ža. V s 50%, Žb. V s 40%, Žc. V s 30%, Žd. V s 20% and Že. V s10%. The
particles are shaded according to their speed.
charge structure is the lowering of the cascading free surface. There is also a marginal
increase in the amount cataracting material. Reducing the fill level from 40% to 30%
again lowers the free surface. The shoulder and toe locations retreat only slightly so that
the smaller amount of charge is spread out more thinly over only a slightly smaller
fraction of the mill circumference. This behaviour continues with each subsequent
reduction of the fill level, with the COM of the remaining charge approaching a shell of
the mill. The trends observed with decreasing fill level can be summarised as follows.
vThe free surface profile changes from being bi-linear to being increasingly curved.
The shoulder position moves slowly lower and the toe moves slowly to the left, so
v
the proportion of the mill circumference covered by the charge decreases much more
slowly than the charge volume, leading to: Ž1. A charge mass that is progressively
thinner. Ž2. The particle velocity near the shell remains that of the mill shell and the
velocity of the cascading particles decreases only gradually with the gradual lowering of
the shoulder. Ž3. The combination of decreasing charge depth with relatively constant
velocity difference across the charge leads to steadily increasing shear rates within the
charge and should therefore lead to more efficient grinding. The increase in the shear
rates can be seen in Fig. 3 by the increasing gradients in the shading of the particles. In
frame d, the fast moving particles at the bottom and top of the charge are very close to
the curved central region of slow moving or stationary particles.
For fill levels of - 10%, there are almost no cascading particles. The charge depth
v
is barely deeper than the depth of the lifter and almost all the particles move in the
cataracting regime.
6.2. Power draw
Fig. 4a shows the power draw as a function of mill speed for fill levels between 10%
and 50% for our standard mill. This clearly shows that a decrease in the fill level leads
to a decrease in the power consumption. This general lowering of the power curve is
expected, since all the material in the mill contributes to power consumption to one
degree or another, so any reduction in fill level leads to reduced power consumption.
The torque required to support the charge in an off center position is essentially the
product of the weight of the charge and the horizontal distance between the COM of the
charge and the center of the mill. The power consumed is the product of this torque and
the rotation rate. When the fill level is reduced from 50% to 40% the COM of the charge
moves only marginally further to the left from its position somewhat below and to left of
the center of the mill. This is because the material removed in lowering the fill level
from 50% to 40% is very close to the center of the mill and has made very little
contribution to the charge being off center and therefore very little contribution to the
overall power consumption.
Comparing the material removed when lowering the fill level from 40% to 30% with
that removed when going from 50% to 40%, we find that horizontal distance of the
COM of this removed material from the center of the mill is more than double the size
for the first removal. The amount of material removed is the same in both cases, so the
more than doubling of the moment arm of the removed material more than doubles the
reduction in power consumption. This behaviour continues with each subsequent
Fig. 4. Effect of mill fill level on power draw as a function of mill speed, Ža. power draw and Žb. power
consumptionrtonne.
reduction of the fill level, with the COM of the remaining charge approaching a shell of
the mill in increasingly large increments. Each removal therefore makes a proportionally
larger reduction in the power consumption. These changes are reflected in the relative
movement of the power curve in Fig. 4a.
Of equal importance is the change in the structure of the power curve. This is best
analysed by considering the power consumption per tonne of charge Žknown either as
power density or specific power draw.. It represents the rate at which energy is supplied
to each unit of charge and controls the intensity of local grinding activity. Its variation
with mill fill level is shown in Fig. 4b. When the fill level declines from 50% to 40%
specific power consumption increases strongly Žby between 15% and 20%.. So although
the total power consumed declines, the amount of power consumed per unit mass
actually increases sharply. This suggests that more efficient grinding will occur for a
lower total energy consumption at a 40% fill level rather than one jammed to near or
above capacity at 50%. This is consistent with operational experience.
Decreasing the fill level to 30% again increases the specific power consumption, Žbut
by a smaller percentage than that of the last increase.. The peak specific power
consumption remains around 105% but continues to broaden. For the 25% fill level the
power curve is again translated higher by a smaller increment and again the peak
broadens. This curve rises strongly and essentially linearly for speeds up to N s 80% at
which point the gradient of the power curve drops abruptly. The specific power
continues to rise much more slowly and essentially linearly to a peak at N s 110% after
which it falls off rapidly as the charge centrifuges.
For fill levels between 25% and 20%, there is a strong qualitative change in the
behaviour of the mill with the power curve developing two peaks. The dominant one
moves to a significantly lower rotation rate between N s 85% and 90% and a broad flat
minor peak appears between N s 105% and 120%. None of the mill conditions or
parameters previously examined had had such an effect in changing the structure of the
power consumption. Previously, the shape of the power curve and the location of its
peak have been found to be insensitive to material properties of the charge the charge
particle size distribution ŽCleary, 1998a.. This suggests that for speeds of N G 93% the
mill efficiency may be higher for a 25% fill level, whereas for lower speeds it may be
more efficient at the 20% fill level.
Decreasing the fill level to 15% continues this pattern of change. The dominant peak
moves down to 80% critical and again increases in magnitude. The point of inflection in
the curve moves from N s 100% to 105% and the minor peak moves to higher speeds
and declines rapidly in magnitude. Decreasing the fill level to 10% continues these
trends, with the peak in the specific power curve again increasing in magnitude and
moving down to a speed of N s 70% and the power consumption at higher speeds
collapsing. Moys Ž1999. reported that for a charge consisting of a single particle Žthe
limit of decreasing fill level. that the peak power consumption occurs at a speed of
N s 50%. This is consistent with the trends identified using these DEM simulations.
In order to understand the effect of the changing of the fill level for a given mill
operating at a single fixed speed, we re-plot Žin Fig. 5a. the power as a function of the
fill level for four common mill speeds. For each rotation rate the lowest power
consumption occurs for the lowest fill level of 10%. The power then increases with
Fig. 5. Effect of mill fill level for four fixed mill rotation rates N s 70%, 80%, 90% and 100% on Ža. power
consumption and Žb. specific power consumption.
increasing fill level, but with steadily decreasing gradients. This occurs because each
new mass added to the mill charge is physically closer to the center of the mill and
therefore contributes progressively less torque and power than did the last addition of
mass. Of particular interest is the observation that the gradients of these curves increase
with the speed of the mill. The curve is much flatter at N s 70% than for the higher
speeds. This suggests that power draw becomes more sensitive to the fill level as the
mill speed increases. Also, at low fill levels, the highest power is used for the lowest
speed, whereas at high fill levels the highest power is used for the highest speed.
Fig. 5b shows the corresponding specific power draw curves. For a mill speed of
N s 70%, the highest power intensity occurs for the lowest fill level of 10%. The power
intensity then decreases almost linearly with increasing the fill level. For a mill speed of
N s 80%, the highest intensity to give occurs at an intermediate fill level of 15%. The
intensity again decreases almost linearly as the fill level increases above this. The
intensity is also lower for a 10% fill level, since for this speed much of this very small
charge is almost being centrifuged by the lifters. For a speed of 90%, the power intensity
peaks for a 20% fill level. For a mill speed of N s 100%, the peak intensity occurs at a
fill level of 25% and declines non-linearly with increasing fill level. The increasingly
thick centrifuged layer causes the intensity to decline even more for the lower fill levels.
7. Effect of lifter shape
Sacrifical liners are used because of the high wear rates occurring within the mills.
This wear occurs most strongly on the protruding lifter bars, reducing the lifter heights
and their face angles over time. In order to evaluate the effect of the lifter life cycle on
mill performance, a series of simulations were run for lifters with progressively
shallower face angles. The width of the lifter at the base was kept constant while the
height of the lifter and its width at the top were suitably reduced to maintain plausible
aspect ratios. The wear on the lifters was assumed to be symmetric, as would be the case
for mills that are periodically reversed. Table 1 shows the geometric details of the four
lifter sets corresponding to lifter face angles of 858 Žnewly installed., 608 Žmoderately
worn., 458 Žwell worn. and 22.58 Žjust before replacement..
Table 1
Geometric details of the four lifters used to represent the wear-induced changes in lifter shape throughout the
lifter life cycle
Lifter face Lifter top Lifter
angle Ž8. width Žmm. height Žmm.
85 200 200
60 110 142
45 76.25 102
22.5 76.25 52
Fig. 6 shows the charge behaviour in the mill for these four lifter shapes for a 40%
fill level. The 858 face angle case ŽFig. 6a. is the same as the ‘standard’ configuration
ŽFig. 1a., but with particles now shaded by size. The main characteristics of the charge
are a bi-linear profile with the free surface inclined slightly down towards the toe and a
moderate amount of cataracting material falling in bands through the upper part of the
mill. When the lifter is moderately worn with a face angle of 608 ŽFig. 6b., there are
only subtle changes to the charge behaviour. The bi-linear free surface is largely
unchanged, but the toe location has risen slightly. The main change is that the
cataracting material is thrown on much lower trajectories by the shallower face angle of
the lifters. Similar changes again occur for the more worn 458 face angle case in Fig. 6c.
The shoulder location is a little lower and the toe is marginally higher. The amount of
cataracting material is much reduced and even the highest trajectory material lands on
the charge surface about half way between the mill center and the toe. When the lifter
Fig. 6. Charge shape predicted for a ball mill rotating at 80% of critical and filled with rocks and balls for
lifters with face angles: Ža. 858, Žb. 608, Žc. 458 and Žd. 22.58.
has been reduced to a 22.58 face angle, there is almost no cataracting material at all and
it barely moves beyond the avalanching surface of the charge. The toe location has
moved noticeably higher and is now higher than the surface at the bottom of the
avalanching region in the middle of the mill. The change in charge behaviour as the liner
wears is gradual, but the cumulative effect is quite large.
Fig. 7a shows the specific power as a function of mill speed for the four lifter
geometries. Each power curve is essentially parabolic in shape. At low speeds Že.g.
N s 50%., the highest power draw occurs for new 858 lifters and decreases slowly with
each reduction in face angle. At high speeds, the opposite is true, with the highest
specific power consumption occurring for the shallowest lifter and the lowest power
draw occurring for the steepest lifter. This occurs because the steep lifters are able to
trap more of the near centrifuging material. As the lifter face angle and the height
declines then more of the charge remains mobile, flowing across increasingly gentle
lifter undulations. More specifically, for speeds N - 70%, the steepest Žnew. lifters
produce the highest power draw. At this speed, the 608 lifter takes over as having the
highest power draw. At a speed of N s 82%, the 458 takes over having the highest
power draw and finally for N ) 94%, the 22.58 has the highest specific power draw.
This demonstrates that intensity of collisional energy dissipation changes significantly
throughout the lifter life cycle.
Fig. 7b shows the specific power draw as a function of the declining lifter face angle
for four common mill speeds.
vFor a mill speed of 70%, the power consumption varies only slightly throughout the
lifter evolution. The specific power is essentially unchanged from 858 to 608 and then
declines gradually and linearly. The total decrease in energy supplied to the charge by
worn lifters is only 4.7% lower than that produced by a new liner.
vFor a mill speed of 80%, the specific power draw actually increases as the face
angle wears from 858 to 608, then declines slightly at 458 and more rapidly at 22.5%.
This means that peak energy consumption does not occur for new liners when operating
at this speed, but increases with some wear, peaking for a 608 face angle and then
declining.
vFor a mill speed of 90%, the lowest specific power consumption occurs with the
new lifters. As the lifter wears, the power increases with the peak occurring for a face
angle of 458. The mill performance then drops off as the lifter wears further. The change
in power consumption between the best and worst lifter shapes is now 8.4%.
vFor a mill speed of 100%, the lowest specific power occurs for new lifters. Energy
consumption increases as the lifters wear and is still increasing even at 22.58, by which
point it has increased by 16.5%.
The lifter face angle for which the peak power consumption is expected decreases
with the mill speed. For high speed mills operating at between 90% and 100% Žas is
common in South Africa., the best face angles would appear Žon this basis. to be
between 22.58 and 458. The amount of change in the power draw with decreasing face
angles increases with mill speed. For a speed of 70%, the power is relatively insensitive
to the lifter face angle, whereas for a speed of 100%, the power varies strongly
throughout the lifter life cycle. This suggests that mill control will be more difficult as
the speed increases.
Fig. 7. Ža. Dependence of the specific power consumption vs. mill speed curve on the lifter shape for face
angles between 858 and 22.58. The fill level here is 40%. Žb. Change in specific power consumption with face
lifter angle throughout the lifter life cycle for the rotation rates N s 70%, 80%, 90% and 100%.
Fig. 8. Charge shape predicted for a ball mill with Ža. a conventional hi–hi lifter pattern, Žb. a hi–lo pattern
and Žc. with half the number of lifters Ž12.. The mills in the left column have a speed of N s 70% and the
mills on the right rotate at N s80%. The fill level is 40%.
8. Effect of lifter pattern
The radial distribution of lifters around the mill shell is also important. Here we
compare the performance of three important lifter patterns.
v The standard hi–hi lifter pattern used throughout this paper.
v A hi–lo pattern consisting of alternating high and low lifters. This is achieved by
changing only half the lifters at each reline, so that half are older and half are newer
lifters at any particular time. This is modelled using 12 standard 858 lifters alternating
with moderately worn 608 lifters.
v A hi–blank pattern where every second lifter is omitted, thereby substantially
increasing the spacing between lifters.
Fig. 8 shows the charge behaviour for the three liner patterns for two mill speeds in
the common operational range, 70% and 80% critical. Changing from a hi–hi to a hi–lo
pattern produces only subtle changes to the charge distribution. The lower face angles on
the low lifters means that the corresponding cataracting material is thrown on lower
trajectories. This produces modestly less cataracting material and also smears its banded
structure. The toe location also moves slightly higher. Removing the low lifters
eliminates every second band of the cataracting material. The toe location moves slightly
higher again and the bi-linear profile of the charge surface becomes much more rounded.
Fig. 9. Specific power consumption for three lifter bar patterns; hi–hi, hi–lo and with half the number of
lifters.
The variation in the toe location with the lifter pattern is somewhat larger for the 80%
mill speed than for the 70% speed.
Fig. 9 shows the corresponding predictions for the specific power consumption. For
speeds up to and including 70% of critical, there is no dependence of the power on the
lifter pattern. Above that, the power consumption becomes increasing sensitive to lifter
pattern. The hi–lo lifter configuration has consistently higher power draw than the
standard pattern, until the centrifuged layer produced at very high speeds reaches the
tops of the lifters Ž N s 120%.. The lifter pattern consisting of half the standard number
of lifters produces an even greater increase in specific power for this range of mill
speeds. This suggests that halving the number of lifters should, for higher speed mills,
increase energy input into the charge.
9. Effect of ball and rock size distribution
One of the largest sources of variation in mill performance is in the feed material and
the proportion of the charge that is made up of steel balls. There are several attributes of
the rock distribution whose effect on mill performance need to be explored, both to
assist in understanding mill behaviour better and in understanding the effects of various
modelling assumptions on the DEM predictions. Changes in the material properties such
as the friction coefficient and the coefficient of restitution have previously been shown
to be relatively unimportant ŽCleary, 1998a. and will not be examined here. Here we
will concentrate on the effect of changes in PSD. From a modelling perspective, this is
very important because the entire size range cannot be simulated and some degree of
approximation is required.
9.1. Effect of the shape of the particle size distribution
9.1.1. Discrete Õs. uniform continuous ball sizes

Fig. 10 shows the specific power draw as a function of mill speed for three different
types of size distribution with a fill level of 50%:
1. The circles correspond to a charge consisting entirely of balls in three discrete

sizes, 5% Žby number. of 200 mm diameter, 47.5% of 100 mm and 47.5% of 75
mm.
2. The triangles correspond to a charge consisting entirely of balls with a continuous
uniform distribution between 50 and 150 mm.
3. The squares correspond to a combination of a continuous ball distribution between
50 and 200 mm and rocks with sizes uniformly distributed between 10 and 50 mm.
For N F 75%, there is negligible difference in the specific power draw for the first
two ball only cases. Between 80% and 100% critical there is a small difference of 2% to
5% with the continuous ball size distribution being lower. For super-critical speeds, the
Fig. 10. The effect of the shape of the ball size distribution on the specific power consumption.
continuous ball case shows a significant reduction in power consumption of 20–25%.

This demonstrates that the power draw is sensitive to even small changes in the ball
distribution.
Comparing the ball only cases to the ball and rock case, we find that the inclusion of
rocks leads consistently to a 0.5-kWrtonne lower specific power draw prediction, for
speeds below 80%. This is a decrease of 5% to 8% and occurs in the speed range where
many mills are operated. This results from a higher toe location in the rock and ball case
producing a more balanced charge and therefore requiring a lower torque and power
consumption to support it. At higher speeds the ball only curves have different shapes
and different maxima, with the ball and rock specific power exceeding both the ball only
cases at around 94% of critical speed. For higher speeds, the predictions diverge
substantially, indicating that the omission of the rock component from such simulations
is likely to lead to substantially erroneous power predictions.
9.1.2. Uniform number distribution of rock size Õs. uniform mass distribution
In the majority of the simulations used in this paper, the particle distribution has been
a binary mixture of balls and rocks each with a continuous uniform distribution between
upper and lower limits. The ball size distribution in a real mill is generally somewhat
controllable and depends on the replacement strategy used for the mill. The uniform size
distribution is a plausible approximation for this and corresponds to a strategy of
regularly adding new balls of the largest size to the charge and occasional removal of
balls smaller than 50 mm Žeither deliberately or by breakage..
The choice of a uniform distribution for the rock sizes does not necessarily represent
the actual size distributions occurring in real mills. In general, there are more particles in
the smaller size ranges than in the large. When a large particle breaks, it produces many
daughter fragments of smaller sizes. Partially counter-balancing this is the removal of
some smaller particles at the discharge end of the mill. A plausible generic particle size
distribution is one where there are equal masses of particles in each part of the size
range. In two dimensions, this means that there are four times as many particles of a
given size as there are of particles with twice that size. This much higher frequency of
smaller particles substantially increases the computational problem size. Nevertheless, it
is important to understand the effect that using different types of size distributions has
on the charge behaviour and the power predictions.
Fig. 11 shows the charge distribution in the mill for the standard configuration with a
fill level of V s 40% and a speed of N s 80% for the two different frequency
distributions for the rock size. The particles are shaded by size. The regions between the
lighter shaded balls are much darker for the mass weighted size distribution, demonstrat-
ing the reduction in the average size of the rock particles. The change in the size
distribution has a number of observable effects.
v The cataracting streams falling from the spaces between the lifters are much more
clearly defined because of the greater number of particles in each stream.
v The toe position is slightly higher and the shoulder position slightly lower. This
shows that the charge is weaker when there are more smaller particles and the charge
slope fails and flows at a slightly lower angle.
v The shape of the free surface of the charge is noticeably different for the mass
weighted case. It is slightly lower at all points indicating that the charge packs slightly
better and so occupies a smaller volume. In both cases, the surface is bi-linear with an
inflection point slightly to the left of the mill center. For a uniform number density the
cascading region is more dilated and the surface always slopes down from the inflection
Fig. 11. Charge behaviour for a mill with a fill level of V s 40% and a speed of N s80% for Ža. uniform
number frequency for the rock size distribution and Žb. uniform mass frequency.
point to the toe. For the mass weighted case, this lower section is essentially horizontal.
This indicates that the charge containing more fines is able to flow better and more of it
is able to reach the toe causing it to move higher.
Fig. 12 shows the specific power consumption for these two types of rock distribu-
tions. The power curve for the mass weighted case is lower across the entire sub-critical
range of speeds and the peak power has moved from N s 110% to 120% critical. The
power is much higher at super-critical speeds than it was for the original size distribu-
tion. The lower power consumption at N s 80% is consistent with the changes observed
above in the charge distribution. That is, the lowering of the shoulder and raising of the
toe decreases slightly the amount by which the charge is out of balance and thereby
reduces both the torque and the power draw.
The changes observed in the power curve when increasing the proportion of smaller
rocks are a continuation of the same trends exhibited earlier. These changes can be
interpreted in terms of the strength of the charge structure.
v For sub-critical speeds the main part of the charge occupies the region below the
characteristic bi-linear free surface. If the charge structure is weakened, then the
cascading slope fails more easily and has a lower dynamic angle of repose leading to a
lower shoulder angle. The failed material is also more able to flow in the direction of the
toe and causes the toe to move slightly higher. This leads to a reduction in the power
consumption. If the charge structure is strengthened, then the opposite effects occur and
the shoulder position moves higher and the toe lower leading to higher power consump-
tion.
Fig. 12. The effect of the shape of the rock size distribution on the specific power consumption. The circles
show the power prediction for a uniform number distribution for the particle size while the triangles show the
power for a uniform mass distribution, which has substantially larger numbers of smaller particles.
v For super-critical speeds the particles are distributed in a centrifuged layer around
the inside of the mills. If the charge structure is weakened, then the particles near the
free surface of this centrifuged region become more mobile. They therefore dissipate
more energy in collisions and are able to re-distribute themselves leading to the charge
being more out of balance. This increases the power consumption. Conversely, if the
charge structure is strengthened, then particles are trapped more efficiently and held
immobile at the free surface leading to less motion and less energy dissipation and
therefore lower power consumption.
The combination of these effects necessarily means the peak power consumption
moves to higher speeds as the charge structure is weakened. The addition of the rock
fraction to the ball only case weakens the charge and pushes the power curve down and
to the right. Similarly, the addition of more fines ‘lubricates’ the flow leading to a
weaker charge and a further reduction in power for sub-critical speeds and an increase
for higher speeds. This demonstrates:
v the effect of increasing fines within the mill on the power draw, and
v the overall importance of having the correct structural strength for the charge in
DEM simulations of mills in order to correctly predict the power draw.
9.2. Effect of the ball fraction

Previously, the ball fraction of the charge has been around 55% Žby volume.. This is
a reasonable generic value, but the ball fraction is one of the charge attributes Žalong
Fig. 13. Effect of ball fraction on the specific power consumption.

Fig. 14. Effect of the lower cutoff of the size distribution on the specific power consumption, Ža. 5 vs. 10 mm
for an AG mill, and Žb. 1, 2 and 5 mm cutoffs for the rock size in a ball mill.
with ball size distribution. that can be varied by operators, so it is useful to understand
its effect on mill behaviour. Fig. 13 shows the specific power draw for ball fractions
varying from 0% Žfully autogenous. to 100% Žball only or fully ground out.. The
specific power curves for all the cases with ball fractions - 80% are very close
together. The differences are very small for all sub-critical mill speeds with larger but
still modest differences at super-critical speeds. Cases with ball fractions as high as 80%
behave essentially the same as the autogenous case. However, for cases with either no
rocks or very few rocks, the interlocking of the balls becomes stronger, leading to an
increase in specific power at lower speeds and a reduction at super-critical speeds. This
change in behaviour occurs only for very high ball fractions between 80% and 98.5%
and indicates that the rocks are a significant contributor to determining the shear
strength of the charge even when the amount of rock is very low. This again
demonstrates the danger of using ball only DEM models to predict power draw for cases
where there are actually rocks present.
9.3. Effect of lower cutoff of rock size distribution
The computational cost of DEM simulation is proportional to the number of particles.

The current limit for reasonable computational times is around 100,000 particles for a
single processor workstation. It is therefore not possible Žand unlikely to be possible for
many years. to simulate the entire size range in a real mill. The particle size distribution
must therefore be truncated at some specified diameter and the smaller particles either
omitted or treated in some other manner. It is very important to understand the
sensitivity of DEM predictions to this cutoff.
Fig. 14a shows the specific power consumption for two AG mills with lower particle
size limits of 10 and 5 mm, respectively. For mill speeds N F 80%, the specific power
draw predictions are identical, but for speeds between 85% and 110% critical, there is a
small difference of 2–3%. For higher speeds, the difference is moderately higher.
Next we consider the ball mill Žrock and ball. cases with smaller cutoffs. The base
case is our standard mill with a cutoff of 5 mm for the rock size. This is compared to
cases that are same except for the cutoff being reduced to 2 and 1 mm, respectively. The
specific power draw curves are shown in Fig. 14b. The predicted power is indistinguish-
able for the three cutoffs for the speeds up to N F 85%. Very small differences occur for
speeds up to 100% of critical speed. For the super-critical speeds there are some small
differences up to 3.5% at N s 115%. As the speed increases further, the charge
increasingly centrifuges and the power differences decrease again. This suggests that
power predictions are insensitive to lower cutoff sizes in the range of 1–5 mm for low
and medium speeds and that 5 mm is a reasonable cutoff for higher speeds.
10. Effect of particle shape
So far, the particles have been modelled as disks, but particle shape has previously
been shown to be important in a range of industrial granular flow applications,
particularly in mixing during avalanche flows in rotating drums ŽCleary et al., 1998;
Cleary, 2000. and hoppers ŽCleary, 1999.. The majority of rocks in mills are not really
circular, particularly those resulting from the fracturing of hard rock ores. It is therefore
important to understand the effect of variations in the rock shape on the power
predictions. This is important because variation in shape distribution of the feed and its
fracturing characteristics are likely to lead to variability in mill performance and because
the complete shape distribution can never measured and must be approximated for
modelling.
As described in Section 2, we will model non-circular particles using super-quadrics
with various amounts of blockiness and differing aspect ratios.
10.1. Effect of ball shape
The effect of non-circularity on both the flow and power draw is evaluated for the
case where the mill contains only non-circular steel balls. We use the same configuration
as described in Section 9.1 with three discrete ball sizes and allow the balls to have
aspect ratios distributed between 1 and 2:1 and super-quadric powers between 2 and 5.
The balls vary in shape from round to squarish.
Fig. 15 shows the ball charge behaviour with circular and non-circular balls for three
mill speeds. At N s 70%, subtle differences appear in the charge distribution. The free
surface of the charge composed of non-circular balls is higher than for circular balls.
This occurs because of the dilation of the cascading flow by the non-circular particles
that require more space in which to rotate. The shoulder is also raised higher causing the
cataracting stream to be thrown on a higher trajectory. Similar changes are observed at
N s 85% with the dilation near the top of the cascading region being particularly strong.
At N s 95%, the charge distribution is quite different. For circular balls, the bulk of the
charge lies in the lower left side of the mill with a modest amount of cataracting
material passing through the middle. For non-circular particles, the charge is signifi-
cantly dilated and is spread throughout a much larger area of the mill.
The specific power consumption for these two ball shapes is shown in Fig. 16. For
N F 75% there is no change in the mill power draw. This occurs because the center of
mass of the charge is little changed by the particle shape at these speeds. At intermediate
speeds, the power draw for the non-circular balls falls off rapidly. This reflects the
increasingly well spread out and dilated charge for the non-circular particles. At
N s 95%, the charge is much more evenly spread throughout the mill than for the
circular case, leading to a less out of balance charge, a lower torque requirement and
therefore lower power consumption. This difference in behaviour can again be related to
the charge strength. Circular particle structures are much weaker than non-circular ones
and yield much more easily. In this case, the slope of circular particle is able to fail and
flow easily maintaining the charge in the lower left side. The non-circular particle
structure is stronger, delaying slope failure so that particles are raised much higher and
are thrown throughout the mill leading to a more even particle distribution and therefore
lower torque and power draw.
The inclusion of particle shape has caused the peak power draw to move to much
lower speeds, from 105% to 90% critical. This brings the DEM predictions much closer
to experimental and field observations where the peak generally lies in range 80–90% of
Fig. 15. Charge behaviour predicted for non-circular balls in a mill rotating at Ža. N s 70%, Žb. N s85% and
Žc. N s95%. The left column shows the flow with circular balls and the right column shows the flow at the
same time using non-circular balls.
Fig. 16. Effect of the ball shape on the specific power consumption.
critical speed. The remaining differences are due to the use of two dimensions, since
particles are expected to interlock better in 3D leading to further increases in charge
strength and further movement of the peak power to lower speeds.
10.2. Effect of rock shape
Next we consider non-circular rocks in a charge with circular steel balls. The rock
aspect ratio varies from 0.4 to 1.0 and the blockiness from 2.1 to 5.0. Two charge types
are considered: Ž1. The particle shape is independent of particle size. Ž2. The particle
shape varies with the size of the particles. The largest particles are given the largest
variations in shape and the range of shapes decreases linearly with the particle diameter,
so that they become progressively rounder as they become smaller. This attempts to
mimic the real changes in particle shape that occur in mills due to the rounding
produced by chipping of corners and attrition.
Fig. 17 shows close ups of the charge structure for the two sets of rock shapes. Fig.
18 shows the specific power consumption for the two non-circular rock cases and for the
standard circular rock case for a fill level of 40%. At sub-critical speeds, the circular
rocks have the lowest power consumption, while at super-critical speeds they have the
highest power consumption. The power consumption for non-circular rocks Žwith shapes
becoming rounder with size. is modestly higher than the circular particle power for all
sub-critical speeds and quite a bit lower for the super-critical speeds. The non-circular
Fig. 17. Close up of particles with non-circular shape when Ža. shape is independent of particle size and Žb.
when the particles become rounder as they become smaller. The particles are shaded according to their size.
rocks Žwith shape independent of size. have the highest power consumption at low
speeds and the lowest power draw at high speeds. The speed at which the peak power
consumption occurs varies from 110% critical to 100% and then down to 95% as the
particle shape variation increases. For this 40% fill level, Liddell and Moys Ž1988.
experimentally found the peak power at N s 88%. These results are now very close,
particularly when considering the smaller rounder lifters used in these experiments. This
demonstrates the critical importance of particle shape in DEM modelling.
All these variations in the power curve can again be explained in terms of the shear
strength of the charge structure. The non-circular rocks with independent shapes have
Fig. 18. Effect of the rock shape distribution on the specific power consumption.
the largest variation from circular allowing the rocks to lock together the most,
producing the strongest structure. This leads to the highest dynamic angle of repose, the
highest shoulder position and therefore the highest sub-critical power draw. It also leads
to the highest locking of the particles in the centrifuged layer and therefore the lowest
power draw for the super-critical speeds. The non-circular particles with shape depen-
dent on size lock together better than the circular rocks but not as well as the other
non-circular case giving an in-between power curve.
11. Conclusion
Two-dimensional DEM modelling has been used to systematically study the effects
of a wide range of mill and particle properties on the charge shape and the power draw
of a ball mill. This is important for helping to understand the effects of operational, liner
and feed variations on the performance of real ball mills, and to evaluate the sensitivity
of DEM predictions to various modelling assumptions.
An important observation is that changes in the charge composition lead to changes
in the charge distribution inside the mill and the power draw via their effect on the shear
strength of the charge. Changes to the PSD Žboth the lower cutoff of the size range and
the shape of the probability distribution., the ball fraction and the rock shape distribution
all alter the shear strength of the charge that is produced by the interlocking of these
particles. Increasing the angularity or aspect ratios of the rock or decreasing the
proportion of smaller particles strengthens the charge and makes slope failure more
difficult. For sub-critical speeds, this leads to higher dynamic angles of repose for the
cascading free surface, to higher shoulder positions and to higher power consumption.
For super-critical speeds, the stronger interlocking of particles reduces their mobility
near the free surface of the centrifuged layer leading to significantly lower power
consumption. Combined, these move the peak in the power curve to much lower and
more realistic speeds.
As the particles become rounder Žon average. or the proportion of smaller particles
increases then the shear strength of the charge declines and slope failure occurs more
easily leading to lower sub-critical power consumption. At super-critical speeds, the
particles near the surface of the centrifuged layer flow more easily to positions lower in
the mill leading to higher power consumption. This causes the peak power to move to
higher speeds. This means that the location of the peak power is quite sensitive to the
shear strength of the charge and therefore to the charge composition. This also means
that predictions made using ball only DEM models differ substantially from those
obtained when more realistic charge models are used.
The effect of the lifter wear cycle on the charge behaviour and power draw were
studied. At sub-critical speeds, the wear induced increases in lifter face angle and
reductions in lifter height lead to steadily decreasing amounts of cataracting material
thrown shorter distances and higher toe positions. At low speeds, the steepest lifters
produce the highest power draw. With increasing mill speed, the lower and shallower
worn lifters produce higher power draw. For low mill speeds, the highest power draw is
expected in the first half of the lifter life and then declines slowly. At higher speeds, the
power draw initially increases as the lifters wear and then decreases. The face angle
producing the highest power draw decreases steadily with increasing mill speed.
The effect of the fill level on mill behaviour was also analysed. For low mill speeds
Ž N F 70%. the highest specific power draw occurred for a fill level of only 10%. At a
speed of 80% critical, the fill level producing the highest power input was 15% which
then increases to 20% at N s 90% critical. The simulations also demonstrated that
filling above 25–30% leads to significantly lower specific power consumption, since the
additional material is closer to the axis of the mill and contributes little to the torque
required to support the charge.
For mill speeds of N F 70%, the power draw was found to be highly insensitive to
the lifter pattern. For higher speeds, reasonable differences were observed with the hi–lo
pattern having higher specific power draw than the standard hi–hi case and half the
number of lifter having even higher specific power draws.
Changes to the ball fraction were found to have little effect on mill behaviour for
rock fractions of 20% by volume or more. The power draw increases proportionally with
the weight of the charge, but the specific power draw is insensitive to the ball fraction.
For very small rock fractions, the specific power draw is consistently higher at
sub-critical speeds. The ball density itself has almost no effect on the charge behaviour
or on the specific power draw. For sub-critical speeds, particles below 5–10 mm seem to
have little effect on the charge behaviour or the power draw. This suggests that
truncating the PSD in DEM simulations of mills at these sizes is a reasonable modelling
assumption. For higher speeds, some sensitivity to the cutoff size is observed, but this is
still small, being less than 2.5% for cutoffs of 5 mm or lower.
DEM modelling appears to be able to qualitatively capture the behaviour of the
charge and to make sensible predictions of the power draw. The results given here
demonstrate the effect of many of assumptions used in this type of modelling. Quantita-
tively accurate predictions require still better agreement with observations of such
quantities as the speed at which the peak power occurs. Already we have demonstrated
that the inclusion of particle shape plays an important role in improving quantitative
agreement. Further work is required to validate these DEM models and to identify
further improvements that will allow highly accurate predictions to be made. Future
developments will include the direct prediction of particle size reduction, axial flows
within the mill and the effect of the pulp.
References
Barker, G.C., 1994. Computer simulations of granular materials. In: Mehta, Anita ŽEd.., Granular Matter: An
Interdisciplinary Approach. Springer-Verlag, New York, pp. 35–83.
Campbell, C.S., 1990. Rapid granular flows. Annu. Rev. Fluid Mech. 22, 57–92.
Campbell, C.S., Cleary, P.W., Hopkins, M.A., 1995. Large scale landslide simulations: global deformation,
velocities and basal friction. J. Geophys. Res. 100 ŽB5., 8267–8283.
Cleary, P.W., 1998a. Predicting charge motion, power draw, segregation, wear and particle breakage in ball
mills using discrete element methods. Miner. Eng. 11, 1061–1080.
Cleary, P.W., 1998b. The filling of dragline buckets. Math. Eng. Ind. 7, 1–24.
Cleary, P.W., 1998c. Discrete element modelling of industrial granular flow applications, TASK. Q. Sci. Bull.
2, 385–416.
Cleary, P.W., 1999. The effect of particle shape on hopper discharge. In: Schwarz, M.P., Davidson, M.R.,
Easton, A.K., Witt, P.J., Sawley, M.L. ŽEds.., Proc. 2nd Int. Conf. on CFD in the Minerals and Process
Industries, pp. 71–76.
Cleary, P.W., 2000. DEM simulation of industrial particle flows: case studies of dragline excavators, mixing in
tumblers and centrifugal mills. Powder Technol. 109, 83–104.
Cleary, P.W., 2001. Modelling communition devices using DEM. Int. J. Numer. Anal. Meth. Geomech. 25,
83–105.
Cleary, P.W., Campbell, C.S., 1993. Self-lubrication for long run-out landslides: examination by computer
simulation. J. Geophys. Res. 98 ŽB12., 21911–21924.
Cleary, P.W., Hoyer, D., 2000. Centrifugal mill charge motion and power draw: comparison of DEM
predictions with experiment. Int. J. Miner. Process. 59, 131–148.
Cleary, P.W., Metcalfe, G., Liffman, K., 1998. How well do discrete element granular flow models capture the
essentials of mixing and segregation processes? Appl. Math. Mod. 22, 995–1008.
Haff, P.K., Werner, B.T., 1986. Powder Technol. 48, 239.
Holst, J.M., Rotter, J.M., Ooi, J.Y., Rong, G.H., 1999. Numerical modelling of silo filling: II. Discrete element
analysis. J. Eng. Mech. 125, 94–110.
Hopkins, M.A., Hibler, W.D., Flato, G.M., 1991. On the numerical simulation of the sea ice ridging process. J.
Geophys. Res. 96 ŽC3., 4809–4820.
Liddell, K.S., Moys, M.H., 1988. The effects of mill speed and filling on the behaviour of the load in a rotary
grinding mill. J. S. Afr. Inst. Min. Metall. 88, 49–57.
Mishra, B.K., Rajamani, R.J., 1992. The discrete element method for the simulation of ball mills. Appl. Math.
Mod. 16, 598–604.
Mishra, B.K., Rajamani, R.K., 1994. Simulation of charge motion in ball mills: Part 1. Experimental
verifications. Int. J. Miner. Process. 40, 171–186.
Moys, M.H., Private communication, 1999.
¨
Poschel, T., Buchholtz, V., 1995. Complex flow of granular material in a rotating cylinder. Chaos, Solitons
Fractals 5, 1901–1912.
Rajamani, R.K., Mishra, B.K., 1996. Dynamics of ball and rock charge in sag mills. Proc. SAG 1996,
Department of Mining and Mineral Process Engineering, University of British Columbia. .
Ristow, G.H., 1994. Granular dynamics: a review about recent molecular dynamics simulations. Annu. Rev.
Comp. Phys. 1, 275–308.
Robinson, G.K., Cleary, P.W., 1999. The conditions for sampling of particulate materials to be unbiased—in-
vestigation using granular flow modelling. Miner. Eng. 12, 1101–1118.
¨ J., Dippel, S., Wolf, D.E., 1996. Force schemes in simulation of granular material. J. Phys. I France 6,
Schafer,
5.
Walton, O.R., 1994. Numerical simulation of inelastic frictional particle–particle interaction. In: Roco, M.C.
ŽEd.., Particulate Two-phase Flow, pp. 884–911, Chap. 25.

Charge Behaviour and Power Consumption I

Uploaded by

Copyright:

Available Formats

Charge Behaviour and Power Consumption I

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Charge Behaviour and Power Consumption I

Uploaded by

Copyright:

Available Formats

Int. J. Miner. Process. 63 Ž2001.

Charge behaviour and power consumption in ball

2. The discrete element method

The essential feature of discrete element simulation is that collisional interactions of

3. Calculation of torque and power

4. The ‘standard’ mill configuration

5. Effect of rotation rate for the standard ball mill

5.1. Charge behaÕiour

5.2. Variation of torque and power with mill speed

6. The effect of mill fill level

6.1. Charge shape and flow behaÕiour

6.2. Power draw

7. Effect of lifter shape

8. Effect of lifter pattern

9. Effect of ball and rock size distribution

9.1. Effect of the shape of the particle size distribution

9.1.1. Discrete Õs. uniform continuous ball sizes

1. The circles correspond to a charge consisting entirely of balls in three discrete

continuous ball case shows a significant reduction in power consumption of 20–25%.

9.2. Effect of the ball fraction

Fig. 13. Effect of ball fraction on the specific power consumption.

9.3. Effect of lower cutoff of rock size distribution

The computational cost of DEM simulation is proportional to the number of particles.

10. Effect of particle shape

10.1. Effect of ball shape

10.2. Effect of rock shape

You might also like