Minerals Engineering 127 (2018) 48–60

Contents lists available at ScienceDirect

Minerals Engineering
journal homepage: www.elsevier.com/locate/mineng

Insights into advanced ball mill modelling through discrete element T

simulations
⁎
Victor A. Rodriguez, Rodrigo M. de Carvalho, Luís Marcelo Tavares
Department of Metallurgical and Materials Engineering, COPPE, Universidade Federal do Rio de Janeiro – UFRJ, Cx. Postal 68505, CEP 21941-972 Rio de Janeiro, RJ,
Brazil

A R T I C LE I N FO A B S T R A C T

Keywords: Important advances have been made in understanding ball milling during the last 25 years or so, a great part of it
Ball mill owing to the widespread application of the discrete element method (DEM), which is now an integral part of
Discrete element method several advanced ball mill models. These models, however, must rely on assumptions regarding the mill me-
Modeling chanical environment that can help make the problem more manageable. The paper analyzes the validity of
Advanced mill models
some assumptions that have been the basis of several advanced ball mill models by conducting DEM simulations
Simulation
of dry batch mills including both grinding media and particles. The validity of the assumption of perfect mixing
of grinding media and particles, as well as of simulating exclusively the grinding media in order to collect the
collision energy information for prediction of breakage and, thus, saving computational effort, are analyzed in
great detail. It is concluded that the assumption of perfect mixing in the radial direction is generally valid, except
for mill frequencies that are unusually high and exceedingly large ratios of mean ball and particle sizes being
ground. It is also observed that a fraction of the number of collisions inside the mill do not involve particles, so
that an empirical expression that is based on the ratio of surface areas of the ball and particle charge is proposed
to estimate such proportion of unsuccessful collisions. Finally, a model from the authorś laboratory, that assumes
that each collision in the mill will involve a monolayer bed of particles, is tested, demonstrating reasonable
agreement when compared to simulations.

1. Introduction (Cleary, 2001; Datta and Rajamani, 2002; King and Bourgeois, 1993;
Powell et al., 2008; Tavares and Carvalho, 2009; Wang et al., 2012;
Mathematical modeling of tumbling ball mills has evolved tre- Capece et al., 2014), with some similarities as well as important dif-
mendously over the last 70 years or so. From the empirical size-re- ferences among them, as recently reviewed by Tavares (2017).
duction relationships (Charles, 1957; Bond, 1952; Morrell, 2004) to the Several of the advanced models of ball mills, as well as other media
traditional population balance model (Austin et al., 1984; Herbst and mills (Beinert et al., 2015), rely on collision energy information from
Fuerstenau, 1980) and the more recent mechanistic models (King and simulations that do not include particles, only grinding media (Datta
Bourgeois, 1993; Capece et al., 2014; Datta and Rajamani, 2002; and Rajamani, 2002; King and Bourgeois, 1993; Tavares and Carvalho,
Tavares and Carvalho, 2009), several approaches have been proposed, 2009), while others include both grinding media and ore/powder par-
which vary significantly in respect to complexity, fidelity and level of ticles (Capece et al., 2014; Powell et al., 2008; Wang et al., 2012). In the
detail in describing the physical phenomena in ball milling. case of models that rely on DEM simulations that only contain the
In recent years, these so-called advanced models have been re- grinding media, as well as in every case when grinding of fine particles,
ceiving progressively more attention, given their power to open, at least called sub-DEM particles (Powell et al., 2008) is to be predicted, a
in part, the “black box” of ball milling through detailed descriptions of reasonable assumption must be made regarding the mass (or volume) of
the media motion and the corresponding energies involved in the col- material captured in each stressing event and how the available colli-
lisions. This has been made possible by the application of the discrete sion energy is split among the particles (Datta and Rajamani, 2002;
element method (DEM) to ball milling, coupled to some form of po- King and Bourgeois, 1993; Tavares and Carvalho, 2009). In nearly all of
pulation balance model formulation (Tavares, 2017; Weerasekara et al., these approaches it is assumed that every collision in a ball mill will
2013). Several different approaches have been proposed over the years involve particles (sub-DEM particles). Another underlying assumption

⁎
Corresponding author.
E-mail address: [email protected] (L.M. Tavares).

https://doi.org/10.1016/j.mineng.2018.07.018
Received 5 May 2018; Received in revised form 18 July 2018; Accepted 19 July 2018
0892-6875/ © 2018 Elsevier Ltd. All rights reserved.
V.A. Rodriguez et al. Minerals Engineering 127 (2018) 48–60

considered when only grinding media is included in the simulations is advantage of the convenience and reduced computational demand as-
that particle to particle and particle to liner collisions are not sufficient sociated to this simple shape and relying on the limited influence of
to cause breakage and may be considered negligible in ball milling. particle shape in mixing in a ball mill observed by Höhner et al. (2015).
Nearly all workable advanced mill models assume that mills are A summary of properties of steel grinding media (and mill shell) and
perfectly mixed in the radial direction. The validity of this assumption particles simulated is given in Table 4.
has been analyzed using both experiments and simulations. Cleary Simulations were conducted at different values of voids or powder
(1998) inferred, with the aid of DEM simulations, that there are two filling, which is defined as percentage of the voids left in the ball charge
mechanisms responsible for size segregation in mills: one that operates that are filled with particles, including the voids between them, when
at low speed and concentrates coarse particles around the outside the mill is at rest. This is estimated assuming a nominal porosity (voi-
portion of the charge, and the other that operates at high speeds and dage) of the ball charge of 40%.
that is responsible for accumulation of coarse particles in the center of The no-slip Hertz-Mindlin contact model was used in the simula-
the charge. He suggested that, at intermediate rotation speeds, the mill tions using the discrete element method. The values of the parameters
charge behavior could be considered perfectly mixed. A later study for steel-particle and particle-particle contacts were collected from a
(Cleary and Morrison, 2011) analyzed the behavior of the charge in- previous study in the authorś laboratory (Ramos et al., 2011), having
fluenced by the presence of fine particles. The authors concluded that, allowed to match the measured mill power with the simulations.
even with 100% of powder filling, the toe region is not yet packed with Parameters for steel-steel contacts were gathered from the literature
powder, with that occurring only at higher percentages. In addition, it (EDEM, 2016) and are also summarized in Table 5. In the case of DEM
was observed that the particles were not uniformly distributed and simulations with no ore particles, in which the ball charge must de-
neither well mixed. However, these authors’ research relied on simu- scribe the movement as if both particles and media are present, the
lations on some particular conditions, that is, a 435 mm diameter mill parameters used were 0.6 for the coefficient of restitution, 0.35 for the
equipped with six small (8 × 8 mm) lifters, which directly influenced coefficient of static friction and 0.2 for the coefficient of rolling friction.
their outcome. In some cases the whole length of the mill (Table 1) was included in
The present work used DEM to conduct virtual experiments to assess the simulations. However, in order to reduce computational effort, only
the validity of several important assumptions commonly used in ad- a section of the mill length (slice) was simulated in most cases.
vanced ball mill modeling, by including not only grinding media (balls) Whenever that was the case, periodic boundaries were used, which
but also particles (ore/powder) in the simulations. As such, it allowed allowed both grinding media and particles leaving from one boundary
investigating directly the validity of the assumptions of perfect mixing appear immediately on the opposite boundary of the simulation. In
in a dry batch ball mill as well as the effectiveness of the grinding media these cases, the length of the slice was chosen to be equal to at least
when nipping particles. three times the maximum diameter of the grinding media (Carvalho,
2013; Carvalho and Tavares, 2013; Cleary, 2001; Weerasekara et al.,
2. Methodology 2016).
Simulations were initially conducted for up to 15 s, guaranteeing
2.1. Setup of DEM simulations that steady-state conditions were reached. After this, information on
particles and grinding media positions and collision energies (energy
The simulations were performed using the discrete element method. loss values) were recorded during two complete revolutions of each mill
This method has been successfully used in different approaches in the studied.
field of particulate materials (Barrios and Tavares, 2016; Cleary and
Morrison, 2011; Powell et al., 2008; Segura-Salazar et al., 2017), and a 2.2. Post-processing of DEM simulations
detailed description of the method may be found elsewhere
(Weerasekara et al., 2013). Simulations were carried out using the 2.2.1. Mixing
EDEM software version 2.7 by DEM Solutions (Edinburgh, UK). Two approaches were used in the present work to analyze mixing in
Different mill geometries and operation modes were selected and a the mills. First, an analysis following a microscopic approach using the
summary of the mills and operating conditions simulated is presented in relative standard deviation (RSD) of the number of particles contained
Table 1, whereas a snapshot of the mills simulated is given in Fig. 1. The in grids inside the mills was carried out. This method, which has been
rectangular shape and the significant height of the lifters used in the successfully used in the past (Cleary, 1998; Pantaleev et al., 2017;
30 cm-diameter mill resulted in onset of cataracting motion with mill Seyed-Alian, 2011), basically consisted in dividing the cross section of
frequencies as low as 50% of the critical speed. Nevertheless, simula- the mill in a preset number of grids, resulting in a number of bins
tions were also conducted for shallower lifters in order to assess their (Fig. 2). After that, the mean number (x) of ore/powder particles in
effect. In most cases, particles were simulated as monosize. In selected each bin (whenever any particles were present), as well as the standard
simulations, however, particles were simulated following the size dis- deviation (σ) of the simulated data, are calculated. This relative stan-
tribution given in Table 2. In addition to the small-scale mills simu- dard deviation is then given by:
lated, pilot and industrial-scale semi-autogenous (SAG) mill were also n
∑i = 1
σ
xi
simulated in DEM. A summary of their characteristics is given in RSD =
n (1)
Table 3. Although the work was not focused on SAG milling, simulation
data from these mills allowed to demonstrate the validity of some of the where n is the number of bins.
approaches to larger diameter mills and to other types of media mills. The lower the value of RSD, the more uniformly distributed parti-
Both grinding media and particles were simulated as spheres, taking cles are and, thus, the better the mixing of the charge. The spacing of

Table 1
Summary of the small-scale mills and conditions used in the simulations.
Mill Diameter (cm) Length (cm) Filling (%) Ball size (mm) % critical speed Particle size (mm)

A 30 30 20/30 15/25/40 40/50/60/80 1.5–15.9

B 58 24 20/30 15/25/40 40/50/60/80 1.5–15.9

49
V.A. Rodriguez et al. Minerals Engineering 127 (2018) 48–60

Fig. 1. Schematics of the batch mills simulated in DEM: mill measuring 30 × 30 cm with eight 1.25-cm height lifters (left); mill measuring 58 × 24 cm with eight 1-
cm height lifters (right).

Table 2 Table 5
Particle size distributions used in selected simulations. Contact parameters used in the DEM simulations.
Particle size (mm) Retained (%) Type of contact Coefficient

Fine Coarse Restitution Static friction Rolling friction

9.5 0 47.94 Steel – steel 0.70 0.20 0.10

6.2 0.18 21.20 Steel – particle 0.37 0.34 0.25
4.4 0.31 17.92 Particle – particle 0.35 0.34 0.20
3.1 67.12 12.94
2.2 23.50 0
1.6 8.89 0

Table 3
Summary of conditions used in the SAG mill simulations.
Mill diameter (m) 1.7 12
Mill speed (% of critical) 75 70

Ball charge Size (m) Weight (%) Size (m) Weight (%)
0.14 34.13
0.16 0.50 0.13 25.48
0.11 0.32 0.11 18.43
0.096 0.18 0.10 13.20

Particle size distribution (DEM) Size (m) Weight (%) Size (m) Weight (%)
0.21 44.15
0.14 0.12 0.19 35.67
0.11 0.15 0.13 10.65
0.088 0.23 0.10 5.52
0.062 0.31 0.076 2.28
0.036 0.14 0.064 1.36
0.018 0.05 0.038 0.29
0.025 0.08

Table 4
Materials properties.
Material Poisson’s ratio Shear modulus (Pa) Density (g/cm3)

Steel 0.30 7 × 109 7.8 Fig. 2. Grid and bins in the 30 cm-diameter mill used for computing RSD.
Powder/ore 0.25 1 × 108 2.7

used to avoid the disproportional effect of bins with a small number of

particles.
the grid adopted was related to the size of the particle analyzed, being
An alternative to this is the bulk or macroscopic approach. It con-
equal to three times the top size of the particle simulated. In analogy to
sists of calculating the distance between the centers of mass (CoM) of
other studies (Pantaleev et al., 2017), the average number of particles
grinding media and particles in the charge:
in each bin was computed and only those containing more particles
than the average were used in computing Eq. (1). This procedure was ΔCoM = (CoMx , b−CoMx , p)2 + (CoMz, b−CoMz, p)2 (2)

50
V.A. Rodriguez et al. Minerals Engineering 127 (2018) 48–60

where CoM are the x and z coordinates of the center of mass of the balls 10000
(b) and particles (p). If the particles and grinding media are well mixed, Ternary ball-particle-ball collisions
Leftover binary collisions after procedure

Frequency of collisions (1/s)

their CoM would be located at the same position, so that the distance
Binary ball-particle collisions
between the two would be zero. As such, the ratio of this distance to the
mill diameter decreases with an increase in the degree of mixing. 1000

2.2.2. Capture of particles between grinding media

In the mechanistic mill model developed in the authors’ laboratory 100
(UFRJ model) (Tavares and Carvalho, 2009; Carvalho and Tavares,
2013), the number of particles captured in each collision is estimated
on the basis of expressions proposed by Barrios et al. (2011), which are
10
summarized in Appendix A. This model uses as a basis the collision
energy spectra obtained in a DEM simulation considering solely the
grinding media, giving the mass of material captured in each stressing
event as a function of grinding media diameter, impact energy, median
1
fracture energy and size of particles stressed, as well as elastic prop- 0.00001 0.0001 0.001
erties of both media and particles. By assuming that particles are
Normal energy loss (J)
stressed as monolayer particle beds and that the stressing energy is
equally split among the particles that are captured (Barrios et al., Fig. 4. Comparison of the collision energy spectra directly from the DEM si-
2011), it allows calculating the collision energies per particle. mulations from ball-particle (binary) energy loss values and collision spectra
In order to assess the validity of this model, data from selected si- after applying the proposed procedure, yielding ball-particle-ball (ternary)
mulations, including both grinding media and particles, at an inter- energy loss values. Simulation for 30-cm diameter mill, with 40-mm balls and
stitial filling of 100%, were post-processed using a novel procedure. 3-mm particles, with 30% filling and 100% of voids filling.
Such procedure became necessary since the DEM software records only
collisions involving two bodies in contact, whereas at least three are Table 6
involved when a particle is nipped between grinding media. Whenever Information registered in the collision event represented in Fig. 3 and illus-
particles are included in the DEM simulations, collisions that lead to tration of processing the data.
breakage would most likely involve three, rather than two bodies in
contact. As such, in order to make valid comparisons between the ID element ID element Normal energy Ternary ball-particle-
Collision end (s)
1 2 loss (J) ball energy loss (J)
collision energy spectra of simulations in the absence of particles and
those including particles, an exercise must be made of condensing, in a 1 3 0.4918 5.52E-05
8.62E-05
5 3 0.4920 3.10E-05
single event, two collisions involving the same particle but different
5 4 0.4924 0.00096
grinding media. 0.0019
1 4 0.4925 0.00097
In order to illustrate this approach, a collision event between two 1 3 0.4927 0.00142
grinding media and three particles is shown in Fig. 3. The grinding 0.0029
5 3 0.4927 0.00144
media are identified as elements #1 and #5, while particles are #2, #3 1 2 0.4927 0.00019
0.0027
and #4. In this procedure, only particle-ball collisions are extracted. In 5 2 0.4927 0.00247
each one, the ID element #1 always corresponds to a grinding medium,
while ID element #2 corresponds to a particle in the collision event.
Fig. 4 summarizes an example of information collected from a collision from application of the proposed routine (ternary ball-particle-ball
event. The routine initially identifies a ball-particle collision, then collisions). It shows a reduction in the number of collisions by con-
checks if that particle is involved in another particle-ball collision over sidering ternary contacts and an increase in their magnitude. It also
the course of the first collision. If that is the case, the collision energies shows that the procedure mainly converted binary collisions with
involved are added (ternary ball-particle-ball energy loss row in higher magnitude into ternary collisions, with limited role on the col-
Table 6), and the number of ternary collisions is incremented by one. lisions of lower magnitude, which were mostly irreconcilable into
Using this approach, the routine enables computing cases of sequential ternary collisions.
collisions involving multiple particles.
An example of application of the routine is presented in Fig. 4, 3. Results and discussion
which compares the original collision energy spectra (binary ball-par-
ticle collisions) collected from the DEM simulation to the one resulting 3.1. Mixing and segregation of media and particles in mills

In order to verify the observation by Cleary and Morrison (2011)

that the relative proportion of particles and grinding media influences
directly the degree of mixing in mills, a preliminary analysis of mixing
was conducted for simulations in the 30 cm-diameter mill (25 mm balls)
with 30% of ball filling as a function of mill interstitial filling. Results
from analyzes using both the macroscopic and microscopic approaches
are summarized in Fig. 5. They show that, as voids filling is increased,
the degree of mixing in the mill also increases towards perfect mixing.
As such, segregation becomes significant at low powder filling levels.
This may be explained by the fact that, under these conditions, there are
not enough particles to fill all the voids contained among the charge,
leading to particles trickling down into the ball charge (Fig. 6). As a
Fig. 3. Representation of two grinding media impacting three ore/powder result, in the RSD analyzes several bins are identified with small vo-
particles. The numbers represent the ID of each element. lumes of particles, generating deviations in respect to the mean.

51
V.A. Rodriguez et al. Minerals Engineering 127 (2018) 48–60

0.6 0.10 First, the bulk charge behavior varied as a function of the ball size, with
the lifters being able to raise to higher shoulder angles both balls and
0.5 particles in the case of the 40 mm balls in comparison to the 25 mm
0.08
balls. Whenever the difference in size between balls and particles was
0.4 smaller (Fig. 8 left) the fine particles followed the trajectories of the
0.06

ΔCoM/D
grinding media, as has already been observed in previous simulations,
RSD

0.3 reaching the high-energy impact zone. However, in simulations with

0.04 40 mm-diameter balls (Fig. 8 right) the particles trajectories do not
0.2 follow those of the balls. Under these conditions, a significant portion of
RSD
0.02
ore particles is just located in the shoulder region and are raised again
0.1 ΔCoM/D without being released to the toe region, resulting in a much greater
difference between the centers of mass of the balls and particles. In this
0 0.00 case, the assumption of perfect mixing would not be valid, primarily
40 80 120 160 200
due to the fact that, in the zone with the most powerful collisions, only
Voids filling (%)
few particles would be present. This behavior may be explained by the
Fig. 5. Results of the microscopic (RSD) and macroscopic (ratio of the distance fact that the charge containing larger balls have larger voids, facil-
between the CoM and mill diameter (D)) mixing indices for the 30 cm-diameter itating powder particles to trickle through them. As such, the sizes of
mill simulated with 25 mm steel balls, 3 mm particles at 30% of ball filling and particles and balls have an important influence on the degree of seg-
50% of critical speed. regation. This can also be observed in Fig. 9, where the differences in
the centers of mass are illustrated and presented as a function of the
The analysis of the centers of mass showed similar behavior. In si- normalized difference in ball and particle sizes. It shows that the larger
mulations with voids fillings as low as 40% the centers of mass of balls this ratio the greater the difference between the centers of mass of the
and particles differ, while they approach each other as voids filling balls and particles. Indeed, others authors have identified the im-
increases. Cleary and Morrison (2011) concluded that this phenomenon portance of these variables on the basis of experimental testing and
is caused by the positioning of particles on top of the lifters just below empirical models (Alonzos et al., 1991).
the shoulder region, without following the trajectory of the balls. Another important topic studied was segregation of the particles
The trajectories of both grinding media and particles are affected by according to size inside the mill, which was identified by Geng et al.
the relative proportion of particles, or voids filling level. Fig. 7 illus- (2016) to occur in a mill. Although it is implicit in the RSD index,
trates the trajectories of the ore particles and balls with 40% of voids Fig. 10 represents the particle charge discriminated by mass, in which
filling. It should be noted that even with the low voids filling level the the particle size distribution simulated is given in Table 4. At first, it
particles are thrown in the same direction as the balls, arriving at the appears that there is no size segregation, with all particles homo-
toe of the charge, zone where impacts of highest magnitude occur. In geneously distributed inside the charge. However, the size segregation
this moment, particles trickle down through the voids left by the balls in normal rotational speeds in tumbling mills observed by Geng et al.
and concentrate near the walls of the mill. Finally, the lifters then work (2016) could have been caused by the short times simulated (2 s),
as buckets that raise particles to higher positions until they are, once during which the mill might not have reached steady-state conditions,
again, thrown against the toe of the charge. as others author also found experimentally (Van Puyvelde et al., 1999).
Some DEM simulations were carried out considering a size dis- The effect of lifter height in mixing behavior was also studied, and it
tribution containing particles from 1.5 mm up to 9.5 mm (Table 2). has been shown that neither the RSD nor the relative distance between
Likewise, even those with finer sizes also follow the same trajectories of the center of mass varied significantly for the 30-cm diameter mill when
the balls, supporting the observation that the ore particles are not rectangular lifters with heights ranging from 5 to 35 mm and 25 mm
concentrated on the shoulder of the charge. From these analyzes, it balls were used (Fig. 11). In spite of the lifter height effect on the falling
could be concluded that under normal operating conditions, the as- trajectory of particles found by Bian et al. (2017) and Kalala et al.
sumption of perfect mixing is reasonably valid. (2008), as well as the present work (Fig. 8), that influence is similar to
Further, particle trajectories of selected simulations with monosize both, the media and the particles. As a result, there was no significant
powder particles (1.5 mm in diameter) were also analyzed (Fig. 8). difference in the relative distance of the CoM and no increase in the

Fig. 6. Snapshots of DEM simulations showing different voids filling levels: 60% (left) and 200% (right). Mill with 30 cm of diameter, running at 50% of critical
speed, 30% of mill filling, 25 mm balls and 3 mm particles.

52
V.A. Rodriguez et al. Minerals Engineering 127 (2018) 48–60

Fig. 7. Trajectories of particles and balls with 30% of mill filling, 40% voids filling levels, 50% of critical speed, 3 mm particles, 25 mm balls and 30 cm-diameter mill.
Larger arrows correspond to the balls, while smaller to the powder particles.

RSD owing to the variation in lifter height within the range of condi- 0.6 0.10
tions studied. This suggests that the segregation observed in Fig. 8 for RSD
40 mm balls may be more appropriately explained by the trickling of 0.5
0.08
ΔCoM/D
the finer powder particles through the media, rather than by direct
0.4
lifter effects.
0.06

ΔCoM/D
The effect of mill speed on the degree of mixing was then studied. In
RSD

0.3
Fig. 12, it is evident that the trends are opposite for the indices, with the
0.04
relative distance between the CoM decreasing as mill speed increases
0.2
and the opposite observed for the RSD. That may be explained by the
fact that the center of mass of the balls is moving only in the vertical 0.02
0.1
direction, while the CoM of the particles is moving towards the axis of
rotation of the mill (Fig. 13). However, that does not mean that the balls 0.0 0.00
and powder particles are well mixed. As the mill speed increases, par- 0.3 0.5 0.7 0.9
ticles concentrate near the wall of the mill and are thrown higher from | − |/
the shoulder than the balls (Fig. 14). On the other hand, the RSD cap-
Fig. 9. Effect of the relative difference between ball (db) and particle size (dp)
tures the real phenomenon, with higher values, which suggests larger
on the charge behavior. Simulations of the 30 cm-diameter mill, at 20% and
dispersion as the rotational speed increases (Fig. 12). Such segregation
30% of ball fillings, 100% of voids filling and 50% of critical speed.
at high fractions of critical speed has also been observed previously by
Cleary (1998). Therefore, good mixing would require small values for
ΔCoM
both indices. Is = RSD +
D (3)
In order to identify the onset of segregation of balls and powder in
the mill charge, an index (Is) is hereby proposed, being given by the
From the previous discussion, the variables with the greater influ-
average of the RSD and the relative difference in centers of mass of
ence on the degree of mixing in the mill were the normalized difference
particles/powder and balls:
in mean ball and particle sizes, the fraction of critical speed and the

Fig. 8. Simulations with 1.5 mm particles at 100% voids filling: 25 mm balls (left) and 40 mm balls (right), 30 cm-diameter mill with 20% mill filling, running at 50%
of critical speed.

53
V.A. Rodriguez et al. Minerals Engineering 127 (2018) 48–60

Fig. 10. Particle distribution by size in ball mill simulations with a particle size distribution. 30 cm-diameter mill, running at 50% of critical speed, 30% of mill filling
and 100% of voids filling. Simulations with 25 mm balls and a finer ore size distribution (Table 2) (left), and 40 mm balls and a coarser ore size distribution (Table 2)
(right). Particles identified in red are the coarsest and in blue, the finest. (For interpretation of the references to colour in this figure legend, the reader is referred to
the web version of this article.)

voids filling level. These have been condensed in the form of a char-
0.6 0.12 acteristic group, given by the dimensionless group f
RSD
0.5 0.10 |db−dp | Nc
ΔCoM/D f=
db U (4)
0.4 0.08
where db and dp are the sizes of balls and particles, respectively, and U is
ΔCoM/D

the voids filling level. In the case of simulations in which particles and/
RSD

0.3 0.06
or balls follow a size distribution, db and dp were given by the weighted
average values. The relationship between this factor and the segrega-
0.2 0.04
tion index Is given by Eq. (3) is illustrated in Fig. 15, which demon-
strates the good correspondence between the two. As such, if the result
0.1 0.02
from Eq. (4) is large the charge could not be sufficiently mixed for the
purpose of advanced ball mill modeling which assumes perfect mixing.
0.0 0.00 According to analyzes of snapshots of simulations, the charge could be
0 10 20 30
Lifter Height (mm) considered perfectly mixed when the value of the index of mixing is
below about 0.5, which corresponds to values lower than about 0.7 of
Fig. 11. Effect of lifter height on the degree of mixing. Simulations for 30 cm- the factor f given by Eq. (4), given Fig. 15. This expression was also
diameter mill, with 25 mm diameter balls, 3 mm particles, 50% of critical
tested for larger scale mills (12-m and 1.8-m diameter semi-autogenous
speed, 30% of mill filling and 100% voids filling level.
grinding mills – Table 3), which generally follow the same trend ob-
served in the simulations of laboratory-scale mills.
0.6 0.10 From the range of variation of values of RSD and the relative dis-
tance of CoM, it is evident in Figs. 5, 9, 11 and 12 that the degree of
segregation index given by Eq. (3) is more strongly influenced by RSD.
0.5
0.08 Nevertheless, the inclusion of the relative distance of CoM guarantees
that cases in which significant displacements of the centers of mass of
0.4
balls and particles are identified.
0.06
ΔCoM/D
RSD

0.3
3.2. Particle collisions not involving grinding media
0.04
0.2
The collision energy spectra of the batch mills operating under
RSD 0.02 different conditions is analyzed as follows. The relative energy con-
0.1 tribution of the different types of collisions with magnitude greater than
ΔCoM/D
10−4 J was analyzed. Initially, the simulations were carried out for the
0.0 0.00 30 cm-diameter mill and only the normal component of each collision
30 50 70 90
was taken into consideration in Fig. 16. It is evident that the total
Percentage of critical speed normal energy dissipated in collisions not involving the grinding media
Fig. 12. Effect of mill rotational speed on the degree of mixing. Simulations for (particle-particle and particle-liners) is significantly lower than that
30 cm-diameter mill, with 25 mm diameter balls, 3 mm particles, 30% of mill involving balls. This is also evident for the 58 cm-diameter mill
filling and 100% voids filling level. (Fig. 17). In this later, the energies dissipated in particle-particle col-
lisions are higher but may still be neglected in comparison to collisions
involving balls. In particular, the simulation with the coarsest particles

54
V.A. Rodriguez et al. Minerals Engineering 127 (2018) 48–60

-0.15 -0.1 -0.05 0 0.05 0.1 0.15 0.8

0.15 Industrial 12 m diameter mill
0.7 Pilot-scale 1.8 m diameter mill
Mill wall Laboratory 0.58 m diameter mill
0.1
40% Nc balls 0.6 Laboratory 0.3 m diameter mill

0.05 50% Nc balls

0.5
70% Nc balls
0 80% Nc balls 0.4

Is
40% Nc particles
0.3
-0.05 50% Nc particles
70% Nc particles 0.2
-0.1 80% Nc particles
0.1
-0.15
0
Fig. 13. Center of mass of particles and balls as a function of rotation speed. 0.2 0.4 0.6 0.8 1
Simulations with 25 mm balls, 3 mm particles, with 30% ball filling and 100% | − |/ ( / )0.5
voids filling. %Nc is the percent of critical speed. The scale in the graph is given
Fig. 15. Degree of segregation index (Is) as a function of particle and ball size,
in meters.
percent of critical speed and voids filling level. The dashed line gives the linear
regression of the data.
(12 mm) in Fig. 18 presents the most significant contribution, due to the
high kinetic energy reached by particles when projected down from the 1

Fraction of normal energy dissipated

shoulder of the charge.
It may be argued that more important than analyzing the total en-
ergy dissipated in each type of collision is assessing the magnitudes of
Ball-Ball
collisions only involving particles. As such, a more detailed analysis of 0.1
these simulations is possible through Fig. 18, which compares the Ore-Ore
maximum energy involved in particle collisions (En,max) in the different Ball-Ore
mills with the minimum energy required to fracture particles in a single
Ore-Liner
impact. This is made possible using models and data that describe the
0.01
amount of energy required to break individual particles by impact Ball-Liner
(Tavares and King, 1998; Tavares, 2007). As such, Fig. 18 compares the
median (E50) values as well as the 10th percentile (E10) of the fracture
energy distribution for a granulite rock and for a soft iron ore (Carvalho
0.001
and Tavares, 2013), this later corresponding to the energy required to 0 50 100 150 200
break 10% of the particles in a given impact. Indeed, even when Voids filling level (%)
comparing to the coarser particles, which are not normally fed to a ball
mill, any collision involving only particles will not be of sufficient Fig. 16. Fraction of energy loss for each type of collision for different voids
magnitude to cause their body breakage. filling levels in a 30 cm-diameter mill, with 30% ball filling, running at 50% of
critical speed, with 25 mm balls and 3 mm particles. Only collisions with
Tavares and King (2002) demonstrated that when particles are
magnitude larger than 10-4 J were considered in the computations.
subjected to low-energy impacts they may break after repeated loading.

Fig. 14. Charge behavior of the 30-cm diameter mill operating to 80% of critical speed, 3 mm particles, 25 mm balls, 30% of mill filling and 100% voids filling.

55
V.A. Rodriguez et al. Minerals Engineering 127 (2018) 48–60

1 400000
Fraction of normal energy dissipated

Number of collisions per second

Ball-Ball
Ball-Liner
Ball-Ball 300000
0.1
Ore-Ore

Ball-Ore 200000
0.01
Ore-Liner

Ball-Liner 100000

0.001
0 5 10 15 0
Particle size (mm) 0 50 100 150 200 250
Fig. 17. Fraction of energy loss for each type of collision as a function of par- Voids filling level (%)
ticle size in a 58 cm-diameter mill, running at 60% of critical speed, with 30% Fig. 19. Frequency of ball-ball and ball-liner collisions as a function of voids
of ball filling and with 100% of voids filling with 25 mm balls. Only collisions filling level (30 cm-diameter mill and 30 cm-length mill, with 30% ball filling,
with magnitude larger than 10-4 J were considered in the computations. running at 50% of critical speed, 25 mm balls and 3 mm particles).

1
1

0.1 0.8

Cumulative distribution
Em,max (30-cm mill)
0%
Energy (J)

Em,max (58-cm mill)

0.6
0.01 E50 granulite 40%
E10 granulite 80%
E50 iron ore
0.4 100%
0.001 E10 iron ore 120%
0.2 150%
200%
0.0001
0 4 8 12 16 0
1E-08 0.000001 0.0001 0.01
Particle size (mm)
Normal energy loss (J)
Fig. 18. Comparison between maximum particle-particle normal energy loss
(En,max) and median (E50) and 10th percentile specific fracture energies (E10) of Fig. 20. Cumulative distributions of ball-ball collisions in simulations involving
granulite and iron particles given by Tavares and Carvalho (2013). 30 cm- particles at different voids filling levels (U). 30 cm-diameter mill, 30% of ball
diameter mill (D), running at 50% of critical speed and 58 cm-diameter at 60% filling, 50% of critical speed, 25 mm ball size and 3 mm particles. Distributions
critical speed. Both mills operating with 30% of ball filling and 100% of voids truncated at 1 × 10-8 J.
filling.

diameter mills and with fine particles, cases in which the number of
Although possible, that is unlikely to be significant as a result of par- particles required for a complete simulation would become un-
ticle-particle and particle-liner impacts in a ball mill, given the low manageable (Tavares, 2017). One of the most important aspects that
frequency of the impacts of sufficient magnitude involving only parti- can be observed in Figs. 16 and 17 is the fact that, whenever relatively
cles. On the other hand, surface breakage requires significantly less coarse particles are being ground, a significant portion of the energy
energy and this mechanism would be important in ball mills as well for dissipated in the mills is associated to ball-ball collisions, thus de-
autogenous (particle-particle and particle-liner) impacts. Evidently, as monstrating the fragility of this assumption that is conveniently used in
the mill diameter and particle sizes increase particle collisions not in- several of the advanced mill models (King and Bourgeois, 1993; Datta
cluding grinding media could reach an energetic level in which auto- and Rajamani, 2002; Tavares and Carvalho, 2009). Evidently, the fre-
genous breakage becomes possible and even significant. That reaches a quency of both ball-ball and ball-liner collisions decreases significantly
critical level when modeling other tumbling mills, namely autogenous as the voids filling level increases, as is illustrated in Fig. 19.
and semi-autogenous mills (Carvalho and Tavares, 2011; Carvalho and A first question that arises is whether the presence of particles in the
Tavares, 2014). simulations affects the magnitude of the collisions that do not involve
In summary, unless the material being ground is extraordinarily soft particles, that is, ball-ball collisions. As such, ball-ball collision spectra
and feed particles very coarse, no bias is introduced in advanced ball from simulations at different voids filling levels are presented in Fig. 20,
mill models by neglecting particle-particle and particle-liner collisions which shows that such ball-ball collision spectra remain approximately
from analysis of DEM simulations for advanced ball mill modeling. constant. This suggests that, in principle, the same collision spectra that
is predicted when no particles are included in the simulations could be
used to describe what happens when different void fillings are used. As
3.3. Grinding media collisions not involving particles such, all that would be necessary is to consider that the frequency of
collision events that involve only grinding media is reduced. In the case
As mentioned previously, several mechanistic ball mill models as- of the ball-liner collisions, such comparisons suggest that the presence
sume that, whenever two grinding media collide, particles will be in- of particles in the simulations had also the effect of reducing the fre-
volved in the event (King and Bourgeois, 1993; Datta and Rajamani, quency of high energy collisions, but those results are omitted for
2002; Tavares and Carvalho, 2009), since they rely on DEM simulations brevity.
of ball mills only containing grinding media. This is an important as- In order to simulate the frequency of these unsuccessful collision
sumption, which makes the models usable in simulations of large-

56
V.A. Rodriguez et al. Minerals Engineering 127 (2018) 48–60

1.0 1.00
30-cm mill (different height lifters)

Ratio of ball-ball collisions - R (-)

Ratio of ball-ball collisions - R (-)

1.8 m Pilot-scale SAG mill

0.8
58-cm mill 0.80
30-cm mill
0.6
0.60
Coarser feed
(15.8 mm,
0.4 12.7 mm,
10.7 mm) 0.40
0.2
100%
0.20 60%
0.0
0.1 1.0
Ab /Ap
0.00
Fig. 21. Ratio of number of ball-ball collisions in a mill simulation containing 40 50 60 70 80
particles and number of ball-ball collisions in a simulation only with grinding Percent of critical speed
media, plotted as a function of ratio of total surface areas of balls and particles.
Fig. 22. Effect of mill frequency on the fraction of ball-ball collisions with
different void filling levels (U). Simulations with 30 cm-diameter mill, 30% of
events, that is, the proportion of collisions in a simulation that do not ball charge, 25 mm balls and 3 mm particles.
include particles, a number, called ratio of ball-ball collisions (R), is
calculated. It is given by the ratio between the number of ball-ball 0.3.
collisions in a simulation in the presence of particles at a given voids A simple empirical model that can describe well the relationship
filling level and the number of ball-ball collisions in a simulation that between the ratio of surface areas and the ratio of ball-ball collisions R
only contains grinding media (i.e., 0% voids filling level). It is im- is given by:
portant to recall that in the DEM simulation without ore/powder par-
Ab
ticles, the contact parameters used in the DEM simulations must re- R = 0.0 <a
Ap
present the movement of grinding media as if powder particles are Ab c
present (Tavares, 2017). ⎡ b − Ap ⎤ Ab
R = 1.0−⎢ ⎥ a< <b
It would be of interest to be able to predict this proportion of un- ⎢
b−a
⎥
Ap

successful collision events on the basis of data from a simulation in- ⎣ ⎦

Ab
volving only grinding media prior to its use in an advanced mill model, R = 1.0 >b
Ap (6)
since these collisions would not be responsible for size reduction of the
particles. As such, a simple model that would allow predicting R as a With the optimal parameters given by a = 0.07, b = 3 and c = 5.5
function of the characteristics of particles and grinding media would be for ball-ball collisions (Fig. 21). The same expression has been used to
useful. It is proposed that the ratio of ball-ball collisions would vary describe the fraction of ball-liner collisions, with good agreement be-
according to the ratio of the total surface areas of the grinding media tween the simulations results with a = 0.08, b = 4 and c = 1.6, but the
and particles. This ratio of the total surface areas of grinding media (Ab) results are omitted here for brevity.
and particles (Ap) could be estimated by assuming that both particles In order to assess whether or not the relationship observed in Fig. 21
and media are spherical, giving: remains valid for other modes of motion of the charge, additional si-
m
mulations were conducted for a number of lifter heights and mill fre-
Ab Hb/ ρb ∑ j = 1 wb, j / db, j quencies. The effect of lifter height has been analyzed and is also pre-
= n
Ap H / ρ ∑i = 1 wi/ di (5) sented in Fig. 21, for lifters with heights ranging from 5 to 25 mm,
showing that its effect was only marginal. The effect of mill speed is
where n and m are the number of particle size (di) and grinding media analyzed in greater detail in Fig. 22, which shows that it has an effect
size (db,j) classes, respectively, wi and wb,i are the mass fractions of on the ratio of collisions that do not involve ore particles. The results
particles and media contained in classes i and j, respectively. H is the indicated that as the rotational speed is increased the number of ball-
mass hold up of particles, while Hb is the mass of grinding media in the ball collisions also increase. It could be explained by the fact that
simulation, which could either be of the entire mill or the slice simu- particles are more sensitive to the increase in centrifugal force, locating
lated. even closer to the mill wall and also being thrown further from the toe
Data on the fraction of ball-ball collisions from a total of 26 DEM of the charge, as was previously observed. However, the observed effect
simulations, including those of SAG mills (Table 3), are plotted as a of speed was not included in Eq. (6). Nevertheless, since very high
function of the ratio of total surface areas of grinding media and par- speeds are also associated to departure from perfect mixing, the model
ticles (Eq. (5)) in Fig. 21. It is worth noting that all simulations included maintains its original validity to a great extent under normal grinding
in the plot were for cases in which balls and particles are sufficiently conditions. In the ball-liner collision type the effect of the mill speed
well mixed, that is, with Is values in Eq.(4) lower than 0.5. was not observed, so that as the mill speed is increased the autogenous
For instance, in a standard test such as Bond́ s ball mill grindability bed in the liner does not allow an increase of the number of ball-liner
test (Bond, 1952), assuming that the feed size distribution follows a collisions, with Eq. (6) remaining valid.
Gaudin-Schumann so that only 0.75% of material finer than 75 µm is It should be mentioned that in high rotational speeds (> 80% Nc),
present in the beginning of the test, this ratio could be estimated as simulations with only grinding media were found, in the present work,
0.15, corresponding to only 19% of the collisions being of the ball-ball not being able to describe the motion of the charge with ore particles.
type. As grinding progresses, however, this number drops further, since This is due to the synergistic action between the centrifugal force and
finer particles are progressively produced. In semi-autogenous grinding the lifters. Even after setting up the contact parameters, the influence of
(SAG), the ratio of collisions not involving particles would be particu- that synergy remains significant. As a consequence, the motion of the
larly large as simulation starts, as the material fed is coarse and the balls in simulations without particles could not resemble that of simu-
ratio between the total superficial areas (Ab/Ap) may be higher than lations when particles are also present. This is a clear limitation of the

57
V.A. Rodriguez et al. Minerals Engineering 127 (2018) 48–60

advanced mill models that rely on collision energy spectra from simu-

Cumulative number of collisions per second

140000
lations only containing grinding media in predicting the performance of
mills at high speeds. 120000

3.4. Capture of particles between grinding media 100000

One key component of advanced ball mill models is related to the 80000
estimation of the mass of material involved in each collision. This mass
is often related to the active stressing volume defined by Schönert 60000
(1988). Different approaches have been used in the past. King and
Bourgeois (1993) proposed an expression that is based on particle and 40000
grinding media size to describe capture of a monolayer particle bed,
whereas Datta and Rajamani (2002) proposed that all collisions in a 20000
mill would involve a particle bed made up of four layers. Barrios et al.
(2011) proposed a comprehensive model that accounts for the effect of 0
ball and particle size, impact energy and median fracture energy of the 1E-08 0.000001 0.0001 0.01 1
particles on the mass of material nipped at each impact in a mill. The Energy loss (J)
model, described briefly in Appendix A, assumes that particles are Fig. 24. Particle-based collision energy spectra estimated on the basis of DEM
stressed as monolayer beds, which was assumed to be a valid approx- simulations with only grinding media from Barrios et al. (2011) model and with
imation whenever the mill operates at void fillings of about 100% DEM simulations with both particles and media. Particle of 10.8 mm in size and
(Carvalho and Tavares, 2013). 25 mm balls in a mill operating at 50% of critical speed, 30% of mill filling and
The model by Barrios et al. (2011) has been used to estimate the 100% of voids filling.
collision energy spectra from simulations only including grinding media
on the basis of each individual particle (Carvalho and Tavares, 2013). It collision in the mill. The discussion in 3.3 demonstrated that it is not a
would be useful to compare the predictions using this model to results valid assumption in most cases, so that predictions are also given in
from DEM simulations not only involving grinding media, but also Figs. 23 and 24 assuming that a fraction of the ball-ball collisions R (Eq.
particles. The procedure described in 2.3 has been used to combine two (6)), irrespective of collision energy, did not nip particles.
ball-particle collisions to represent a single ternary (ball-particle-ball) It can be also observed that there is reasonable agreement for the
collision, with an example of application of the routine already given in case when the ratio of ball and particle size is higher (Fig. 23) between
Fig. 4. Two simulations served as the basis for comparison: one with predictions of DEM collision spectra on the basis of simulations invol-
3 mm particles and 40 mm balls, and the other with 10.8 mm particles ving particles and those using only grinding media and the model by
and 25 mm balls, both assuming granulite as the material being ground Barrios et al. (2011), provided that the proportion of unsuccessful
(Barrios et al., 2011). In the first case, the model by Barrios et al. (2011) collision events R is accounted for. For the case when the ratio between
predicts that between one and six particles would be involved in each ball and particle size was small (Fig. 24) only a fair agreement was
collision in the mill, depending on the impact energy, while in the first obtained. It shows that, with caution, the model by Barrios et al. (2011)
case collisions would involve only one particle at a time. may be used as a first estimate of the material captured between
The results of those comparisons are shown in Figs. 23 and 24. In grinding media, but work is still needed to further identify its limita-
both cases simulated, the approach by Barrios’ overestimated the en- tions.
ergy received by the particles when compared to the DEM simulations
including both particles and grinding media. In the model by Barrios 4. Conclusions
et al. (2011), it is assumed that particles will be involved in every
The influence of several operating and design variables has been
35000
Cumulative number of collisions per second

Barrios et al. (2011) with 1-R collisions

investigated in the degree of mixing of ball mills using DEM simula-
tions. It was found that the relative difference in mean ball and particle
30000 Ternary ball-particle-ball collisions size has the greatest influence on the degree of segregation, so that the
Barrios et al. (2011) approach smaller the difference, the greater the degree of mixing. Other vari-
25000 ables, such as the voids filling level and the fraction of critical speed
were also found to directly influence the degree of mixing in the mill.
20000 An index was then proposed to identify the expected degree of segre-
gation of particles and grinding media in DEM simulations of ball mills,
15000 which can serve as a guide to identify conditions when advanced mill
models cannot assume perfect mixing of grinding media and particles,
10000 for the particular case of rectangular lifters. This index could also be
used to analyze the degree of mixing in other media mills, such as SAG
mills.
5000
DEM simulations including particles and grinding media allowed to
identify that, under normal conditions in batch mills, collision energies
0
involving events with only particles (particle-particle and particle-liner)
1E-08 0.000001 0.0001 0.01 1
were of insufficient magnitude to cause breakage of particles, at least in
Energy loss (J)
a single impact.
Fig. 23. Particle-based collision energy spectra estimated on the basis of DEM The DEM simulations also allowed to establish that a fraction of the
simulations with only grinding media from Barrios et al. (2011) model and with collisions inside the ball mill do not involve particles. An empirical
DEM simulations with both particles and media. Particles of 3 mm in size and model that is based on the ratio of total surface areas of the particles
40 mm balls in a mill operating at 50% of critical speed, 30% of mill filling and and the media has been proposed, demonstrating good ability to de-
100% of voids filling. scribe the simulated outcomes.

58
V.A. Rodriguez et al. Minerals Engineering 127 (2018) 48–60

Simulations have been also used to analyze critically the model Acknowledgements
developed previously in the authorś laboratory, which aims at pre-
dicting the amount of material captured in DEM simulations involving The authors would like to acknowledge the financial support from
only grinding media. It was found that the model was in reasonable the Brazilian agencies CNPq (grant number 310293/2017-0) and
agreement with simulation results, in particular when the ratio between CAPES, as well as to DEM Solutions for the support through the EDEM
ball and particle was higher. Academic Program.

Appendix A

Given the evidence from earlier work (Höfler, 1990), that particles in an unconfined bed normally only break when they are typically squeezed
down to a monolayer between grinding media, Barrios et al. (2011) assumed that particles can be modeled as a packed monolayer bed. If particles
had spherical shapes, and if they were arranged according to a dense hexagonal packing, then the number of particles captured as a function of
radius could be estimated by (Barrios et al., 2011)
2rc 2
Ncap, i =
1
4
+
3
4 ( )
di
for rc ⩾ di /2
=1 for rc < di /2 (A.1)
where rc is the radius of the bed, also called radius of capture, and di is the mean size of the particles caught by the colliding steel balls, estimated
from the geometric mean of sieves containing the narrow size fraction.
Barrios et al. (2011) proposed that the radius of capture can be calculated by adding the radius of contact due to geometry of both particles and
media, and the radius of contact due to the elastic deformation in the vicinity of contact, giving
1/5 1/2
15E ⎞ Δ 2
rc = ⎛
⎜ ⎟ K g2/5 + ⎡K g2−⎛K g−h ⎞ ⎤
⎝ 8K e ⎠ ⎢
⎣ ⎝ h⎠ ⎥⎦ (A.2)
where Kg is the geometric constant of the contact, given by
rc1 rc 2
Kg =
rc1 + rc 2 (A.3)
and the elastic constant of the contact is given by (Tavares and King, 2004)
K1 K2
Ke =
K1 +K2 (A.4)
where rci and Ki are the radius of curvature and the elastic stiffness of each of the bodies in contact, respectively. Δ is the maximum deformation of
the bed during impact. In the case of a monolayer bed, the initial bed height (h) may be considered equal to di. The term Δ/h, called “maximum
relative deformation of the bed” has been described by Barrios et al. (2011) using the expression
0.413
Δ ⎡ E ⎞ ⎞⎤
= 0.362 ⎢1−exp ⎛⎜−0.152 ⎛ ⎜ ⎟
⎟⎥
h ⎝ ⎝ E50b ⎠ ⎠⎦ (A.5)
⎣

References Charles, R., 1957. Energy-size reduction relationships in comminution. Trans. AIME 208,
80–88.
Cleary, P.W., 1998. Predicting charge motion, power draw, segregation and wear in ball
Alonzos, M., Satoh, M., Miyanami, K., 1991. Optimum combination of size-ratio, density mills using discrete element methods. Miner. Eng. 11, 1061–1080.
ratio and concentration to minimize free surface segregation. Powder Technol. 68, Cleary, P., 2001. Recent advances in DEM modelling of tumbling mills. Miner. Eng.
145–152. 1295–1319.
Austin, L., Klimpel, R., Luckie, P.T., 1984. Process engineering of size reduction: ball Cleary, P.W., Morrison, R.D., 2011. Understanding fine ore breakage in a laboratory scale
milling. AIME 561, p. ball mill using DEM. Miner. Eng. 24, 352–366.
Barrios, G.K.P., Carvalho, R.M., Tavares, L.M., 2011. Modeling breakage of mono- Datta, A., Rajamani, R.K., 2002. A direct approach of modeling batch grinding in ball
dispersed particles in unconfined beds. Miner. Eng. 24, 308–318. mills using population balance principles and impact energy distribution. Int. J.
Barrios, G.K.P., Tavares, L.M., 2016. A preliminary model of high pressure roll grinding Miner. Process. 64, 181–200.
using the discrete element method and multi-body dynamics coupling. Int. J. Miner. EDEM, 2016. EDEM Users Manual, DEM Solutions. Edinburgh.
Process. 156, 32–42. Geng, F., Gang, L., Wang, Y., Li, Y., Yuan, Z., 2016. Numerical investigation on particle
Beinert, S., Fragnière, G., Schilde, C., Kwade, A., 2015. Analysis and modelling of bead mixing in a ball mill. Powder Technol. 292, 64–73.
contacts in wet-operating stirred media and planetary ball mills with CFD-DEM si- Herbst, J.A., Fuerstenau, D.W., 1980. Scale-up procedure for continuous grinding mill
mulations. Chem. Eng. Sci. 134, 648–662. design using population balance models. Int. J. Miner. Process. 7, 1–31.
Bian, X., Wang, G., Wang, H., Wang, S., Lv, W., 2017. Effect of lifters and mill speed on Höfler, A., 1990. Fundamental Breakage Studies of Mineral Particles with an Ultrafast
particle behaviour, torque, and power consumption of a tumbling ball mill: experi- Load Cell Device. Ph.D. thesis. Department of Metallurgical Engineering, University
mental study and DEM simulation. Miner. Eng. 105, 22–35. of Utah, Salt Lake City.
Bond, 1952. The third theory of comminution. Trans. AIME Min. Eng. 193, 484–494. Höhner, D., Wirtz, S., Scherer, V., 2015. A study on the in fl uence of particle shape on the
Capece, M., Bilgili, E., Davé, R., 2014. Insight into first-order breakage kinetics using a mechanical interactions of granular media in a hopper using the Discrete Element
particle-scale breakage rate constant. Chem. Eng. Sci. 117, 318–330. Method. Powder Technol. 278, 286–305.
Carvalho, R.M., Tavares, L.M., 2011. Leaping forward in SAG and AG mill simulation Kalala, J.T., Breetzke, M., Moys, M.H., 2008. Study of the influence of liner wear on the
using a mechanistic model framework. In: Proc. Int. Autogenous Grinding, Semi- load behaviour of an industrial dry tumbling mill using the Discrete Element Method
Autogenous Grinding and High-Pressure Roll Technology Conference, Vancouver, p. (DEM). Int. J. Miner. Process. 86, 33–39.
1–26. King, R.P., Bourgeois, F., 1993. A new conceptual model for ball milling. In: 18th Int.
Carvalho, R.M., Tavares, L.M., 2014. A mechanistic model of SAG mills. In: Proc. 27th Int. Miner. Process. Congr., Sydney, vol. 1, p. 81–86.
Miner. Process. Congr., Santiago, p. 1–12. Morrell, S., 2004. An alternative energy–size relationship to that proposed by Bond for the
Carvalho, R.M., 2013. Mechanistic Modelling of Semi-Autogeneous Grinding. Ph.D. design and optimisation of grinding circuits. Int. J. Miner. Process. 74, 133–141.
thesis, Universidade Federal do Rio de Janeiro, Rio de Janeiro. Pantaleev, S., Yordanova, S., Janda, A., Marigo, M., Ooi, J.Y., 2017. An experimentally
Carvalho, R.M., Tavares, L.M., 2013. Predicting the effect of operating and design vari- validated DEM study of powder mixing in a paddle blade mixer. Powder Technol.
ables on breakage rates using the mechanistic ball mill model. Miner. Eng. 43–44, 311, 287–302.
91–101. Powell, M.S., Govender, I., McBride, A.T., 2008. Applying DEM outputs to the unified

59
V.A. Rodriguez et al. Minerals Engineering 127 (2018) 48–60

comminution model. Miner. Eng. 21, 744–750. Tavares, L.M., Carvalho, R.M., 2009. Modeling breakage rates of coarse particles in ball
Ramos, E.S., Carvalho, R.M., Tavares, L.M., 2011. Simulação do movimento da em mills. Miner. Eng. 22, 650–659.
moinhos de bolas utilizando o método dos elementos discretos, Proc. 24th Brazilian Tavares, L.M., King, R.P., 1998. Single particle fracture under impact loading. Int. J.
Meeting on Mineral Processing and Extractive Metallurgy. Salvador 1, 440–447. Miner. Process. 54, 1–28.
Schönert, K., 1988. Stressing mechanisms in mills. In: Course notes, University of the Tavares, L.M., King, R.P., 2002. Modeling of particle breakage by repeated impacts using
Witwatersrand, Division of Continuing Engineering Education, Johannesburg, continuum damage mechanics. Powder Technol. 123, 138–146.
Section F3. Tavares, L.M., King, R.P., 2004. Measurement of the load–deformation response from
Segura-Salazar, J., Barrios, G.P., Rodriguez, V., Tavares, L.M., 2017. Mathematical impact-breakage of particles. Int. J. Miner. Process. 74, S267–S277.
modeling of a vertical shaft impact crusher using the Whiten model. Miner. Eng. 111, Van Puyvelde, D.R., Young, B.R., Wilson, M.A., Schmidt, S.J., 1999. Experimental de-
222–228. termination of transverse mixing kinetics in a rolling drum by image analysis. Powder
Seyed-Alian, S.-M., 2011. Analysis of the mixing of solid particles in the slant cone and Technol. 106, 183–191.
ploughshare mixers via discrete element method (DEM). Powder Technol. 274, Wang, M.H., Yang, R.Y., Yu, A.B., 2012. DEM investigation of energy distribution and
77–87. particle breakage in tumbling ball mills. Powder Technol. 223, 83–91.
Tavares, L.M., 2007. Breakage of single particles: quasi-static. In: In: Salman, A.D., Weerasekara, N.S., Liu, L.X., Powell, M.S., 2016. Estimating energy in grinding using DEM
Ghadiri, M., Hounslow, M.J. (Eds.), Handbook of Particle Breakage, vol. 12. Elsevier, modelling. Miner. Eng. 85, 23–33.
pp. 3–68. Weerasekara, N.S., Powell, M.S., Cleary, P.W., Tavares, L.M., Evertsson, M., Morrison,
Tavares, L.M., 2017. A review of advanced ball mill modelling. KONA Powder Particle 34, R.D., Quist, J., Carvalho, R.M., 2013. The contribution of DEM to the science of
106–124. comminution. Powder Technol. 248, 3–24.

60