This article appeared in a journal published by Elsevier.

The attached
copy is furnished to the author for internal non-commercial research
and education use, including for instruction at the authors institution
and sharing with colleagues.
Other uses, including reproduction and distribution, or selling or
licensing copies, or posting to personal, institutional or third party
websites are prohibited.
In most cases authors are permitted to post their version of the
article (e.g. in Word or Tex form) to their personal website or
institutional repository. Authors requiring further information
regarding Elsevier’s archiving and manuscript policies are
encouraged to visit:
http://www.elsevier.com/authorsrights
 Author's personal copy

Powder Technology 248 (2013) 3–24

Contents lists available at ScienceDirect

Powder Technology
journal homepage: www.elsevier.com/locate/powtec

Review

The contribution of DEM to the science of comminution

N.S. Weerasekara a,⁎, M.S. Powell a, P.W. Cleary b, L.M. Tavares c, M. Evertsson d, R.D. Morrison a,
J. Quist d, R.M. Carvalho c
a
University of Queensland, JKMRC, 40 Isles Road, Indooroopilly, QLD 4068, Australia
b
CSIRO Mathematics, Informatics and Statistics, Clayton, Victoria, 3168, Australia
c
Department of Metallurgical and Materials Engineering, Universidade Federal do Rio de Janeiro — COPPE/UFRJ, Rio de Janeiro, RJ, Brazil
d
Rock Processing Systems, Chalmers University of Technology, SE-41296, Göteborg, Sweden

a r t i c l e i n f o a b s t r a c t

Available online 24 May 2013 The ability to model comminution techniques has been key in the mineral processing industry and other in-
dustries where size reduction of rocks is a key focus. With the need for more accurate and sophisticated
Keywords: models, the introduction of computational models like the discrete element method (DEM) has started to
DEM solve some of the complex problems in comminution. This review article provides a general, though not
Grinding mills exhaustive, overview of the application of computational techniques to the science of comminution.
Liner wear modelling
Over the last two decades the DEM has become an important tool for understanding comminution funda-
Comminution
Mechanistic models
mentals and providing information for the design, optimisation and operation of comminution devices.
Crushers Over that time, the application of DEM in simulating fracture, breakage, crushing, milling and equipment
wear has increased in size and complexity. Information from DEM has contributed to modelling techniques
like, mechanistic modelling, the Unified Comminution Model and the Virtual Comminution Machine. These
have enabled step changes in understanding of comminution processes. DEM is now widely used in indus-
tries where comminution is critical.
Important challenges remain for the next decade including representation of unresolved fine material: predic-
tion of its influence on the overall processes and its size reduction. Similarly, slurry phase transport and rheology
have barely been touched on in using computational techniques in comminution modelling. Most importantly in
the future is the uptake of the DEM outputs into standard comminution models used in design and optimisation
of equipment and processes, as despite the progress made in understanding and in quantitative outputs the
barriers to routine uptake have not been surmounted.
© 2013 Elsevier B.V. All rights reserved.

Contents

1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2. The components needed to assemble a comminution model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
3. DEM model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
3.1. Contact model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
3.2. Bonding model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
3.3. Energy logging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
4. Breakage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
5. Ore characterisation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
5.1. Impact damage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
5.2. Surface damage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
5.3. Fine particle breakage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
6. Application to specific comminution devices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
6.1. DEM modelling of rock crushing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
6.2. Ball mill DEM modelling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
6.3. Stirred mill DEM modelling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
6.4. SAG/AG mill DEM modelling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

⁎ Corresponding author.
E-mail address: [email protected] (N.S. Weerasekara).

0032-5910/$ – see front matter © 2013 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.powtec.2013.05.032
 Author's personal copy

4 N.S. Weerasekara et al. / Powder Technology 248 (2013) 3–24

7. DEM based modelling techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

7.1. Mechanistic ball mill model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
7.2. Unified comminution model (UCM) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
7.3. Virtual Comminution Machine (VCM) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
7.4. Wear modelling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
7.4.1. Crusher liners . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
7.4.2. Mill liners . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
8. Validation of DEM predictions of comminution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
9. Future challenges for DEM modelling in comminution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
10. Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

1. Introduction of mineral processing — the SAG mill. There are dozens of papers in
the proceedings of the SAG conference series that attest to issues with
The discipline of comminution addresses the size reduction of process design, SAG11 [9] and SAG06 [10]. In general, pilot-scale testing
rocks. In mining this reduction in size is required to expose the valu- in a 6 ft diameter mill is required to validate design predictions, but
able minerals so as to enable their recovery; in the cement industry it requires up to 100 tons of ore and is only valid for the bulk sample
is used both in preparing the raw material and grinding the clinker; in used — which is never representative of the entire ore body, which is
the aggregate industry comminution is used to produce an aggregate always variable across the ore deposit.
to a strength and shape specification; in the power industry coal is However, these models have significant limitations as they cannot
ground to promote rapid and complete combustion; and in a plethora be used to determine how to best operate a comminution device
of chemical industries from paints to pharmaceuticals finely ground based on the intrinsic physical properties of the ore. These models and
particles are required to enable chemical reactivity. data were developed for specific comminution devices in specific sce-
The operation of comminution devices in mineral processing oper- narios. They therefore cannot be applied to a new comminution device,
ations is informed by a vast suite of data gathered over many years for to a device being used outside of its established operating window, to a
their operation with various ores and the semi-empirical models which device placed in a different part of a comminution circuit or if partnered
have evolved through a combination of practice, observation, research with additional devices (e.g. screens or crushers) in a circuit. Predicting
and model-fitting data. The modelling of these processes has tradition- liberation is the key to linking comminution with recovery, however,
ally been via power-based relationships and semi-empirical models — liberation cannot be genuinely incorporated into these current models.
most based on the population balance model. Being able to correctly Similarly, current models can only treat an average ore so a blending
calculate the power required to produce a specific desired size reduction factor must be applied for real ore mixtures. However, this approach
underpins machine selection. The Bond P80 approach has been the most does not produce a realistic outcome. Breakthroughs in circuit design
successful of these, covering ball and rod mills [1,2], with more limited have been severely hampered by these limitations.
application in crushing. The energy required (E) for size reduction is cal- In summary, the existing power-based and semi-empirical/
pffiffiffiffiffi pffiffiffiffiffi
culated using the so-called “Third Law”, E ¼ K ½1= x2 −1= x1 , where x1 phenomenological models suffer inadequacies in accuracy, application
is feed size and x2 is product size. The Bond technique provides a simple to multi-component ores, ability to incorporate liberation, and inability
single figure that has led to its ubiquitous use in mining projects since to predict the performance of new or novel equipment. This provides a
the 1960s. The population balance method (PBM) was introduced clear motivation to shift towards mechanistic, predictive models with a
later and became popular in the 1970s [3–6]. They incorporated a break- deeper insight, as described by the work of Powell and Morrison [11].
age rates function that was used to model the rate that material breaks It also provides the criteria by which their usefulness can be assessed.
out of each size class. Combined with an appearance (or breakage) The DEM is an outstanding technique that occupies a central position
function – the average size distribution of the progeny from each break- in mechanistic modelling, so has found a firm place in advanced commi-
age event – this allowed a product size distribution to be calculated. nution modelling.
Most researchers since then have used this technique as a basis and This overview paper presents a perspective of the authors on how
added semi-empirical relationships that are fitted to measured equip- DEM contributes to the science of comminution applied to the mineral
ment response. These relationships modify the breakage rates by size processing and aggregate industry. Rather than attempt to present a
enabling the model to respond to changes in feed and operating condi- fully comprehensive review of the area, we have focussed on where
tions, giving the models a useful predictive capability. Examples of we think some of the most promising applications are, and how this
this are the suite of models that have emanated from the JKMRC, collaborating group are applying them.
Napier-Munn et al. [7], which are a small sub-set of the plethora of models
that have been published over the years. 2. The components needed to assemble a comminution model
The power and semi-empirical modelling approaches have been
extensively used in the industry in comminution circuit design with In considering how to shift to mechanistic modelling, it is first es-
a marked degree of success, but with many instances of poor equip- sential to work out what is required to model the desired processes.
ment selection. This may not be the perception in the industry, as in The following list provides a brief summary of these requirements
general only the successes are publicised. However, if one reads when modelling a mechanical comminution process:
through the plethora of works on process optimisation, a large num-
ber of these address under-performance relative to design, with it • distribution of collision energies
being common to spend 1 to 3 years to reach full production capacity, • rate of collisions
in terms of throughput and size reduction as analysed in the work of • types of collisions
Mackey and Nessett [8]. The problem of inadequate or unreliable • residence time
equipment sizing is particularly acute in processes where the simple • discharge from the device
power relationships do not hold, such as the more recent workhorse • damage response of the rocks to the range of possible collisions
 Author's personal copy

N.S. Weerasekara et al. / Powder Technology 248 (2013) 3–24 5

Table 1 3.1. Contact model

Spring stiffness and damping coefficients used in the contact model.

Normal direction (n) Tangential direction (t) A non-linear model, combining Hertz's theory in the normal direc-
pffiffiffiffiffiffiffiffiffiffi 
pffiffiffiffiffiffiffiffiffiffi tion and the improvement to Mindlin's no-slip model [12] in the tangen-
Spring stiffness constant (K) E ffiffi R δn
K n ¼ 43 q K t ¼ 8G
qffiffi R

δn
Damping coefficient (C) pffiffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffiffiffi tial direction is employed [13–15] in modelling the contact between
C n ¼ 2 6 β Sn m
5
C t ¼ 2 6 β K t m
5

pffiffiffiffiffiffiffiffiffiffi particles. Hertz's theory of elastic contacts provides compact relations

Sn ¼ 2E R δn , β ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffi
lnε ffi
2
ln εþπ2 for the normal direction, derived from integration of the normal pres-
sure distribution over the contact area [16]. The resulting total force
(Fn) is then a sum of elastic and dissipative (damping) force components
given by [14,17];
The above parameters vary according to equipment type, design
and mode of operation. rel
F n ¼ −K n δn þ C n vn ð1Þ
Although not exhaustive, the list provides a basic framework about
which to consider the development of mechanistic models. It is notable All terms are defined in the nomenclature. This provides a spring
that this full list is not addressed by current population balance based repulsive force that pushes colliding particles apart and a dashpot to
models, collapsing to breakage, selection and transfer functions. dissipate a portion of the relative kinetic energy.
In the tangential direction the possible force–displacement config-
urations depend on both normal and tangential loading history. The
3. DEM model relative tangential velocity (vt) from the relative tangential motions
over the collision, behaves as an incremental spring that stores energy
The discrete element method (DEM), also called the distinct element and represents the elastic tangential deformation of the contacting sur-
method, is a numerical method for computing the motion and collisions faces. The dashpot dissipates energy from the tangential motion and
of particles. DEM has become widely accepted as an effective method models the tangential plastic deformation of the contact. The total tan-
of addressing engineering problems in granular and discontinuous ma- gential force (Ft) is limited by Coulomb's law of friction [14,17,18].
terials, especially in granular flows, powder mechanics, rock mechanics,
and comminution. Discrete element methods are relatively computa- n o
rel
F t ¼ min μF n ; K t δt þ C t vt ð2Þ
tionally intensive, which limits either the length of a simulation or the
number of particles. DEM codes take advantage of parallel processing
capabilities (on shared or distributed systems) to scale up the number The coefficients in Eqs. (1) and (2) are given in Table 1.
of particles or duration of the simulation. The DEM codes used for comminution modelling generally use a
Fundamentally, DEM solves Newton's equations of motion to re- linear spring and dashpot model in both the normal and tangential
solve particle motion and uses a contact law to resolve inter-particle direction with a frictional slider in the tangential direction, or a vari-
contact forces. Forces are typically integrated explicitly in time to pre- ation thereof. Due to its simplicity, this form of model is suited to
dict the time history response of the material using an appropriate tracking the motion and gross kinetics of a large number of particles.
quadrature method. The DEM includes a family of techniques that The energy dissipated in collisions is captured in the dashpot, allowing
use radically different treatments for the element geometry and the it to be accounted for and recoded for later analysis in the comminution
form of the contact forces. Various forms of DEM employed and the models. However, the idealised nature of the contact model does not
applications typically considered are discussed within the context to allow for the detailed analysis of the contact event, indeed the outputs
their application to comminution problems. should not be over-interpreted in terms of physical significance. Rather,

Fig. 1. Force–displacement behaviour of grain–cement system [26].

 Author's personal copy

6 N.S. Weerasekara et al. / Powder Technology 248 (2013) 3–24

the outputs should be utilised in a pragmatic manner to provide the These two reviews observed that the application of the simple contact
rates and frequencies, distributions and energies of collision events. models for future DEM simulation of comminution systems, as the
Recently, Thornton et al. [19,20] compiled a comprehensive review contact mechanics are inadequate for detailed analysis of the collision
of all known elastic and inelastic DEM contact models and made an as- and breakage events. With the massive improvements in computa-
sessment of their relative performance for oblique single particle–wall tional capability the time may be ripe to take up a new level of more
collisions. It was observed, that the elastic models, for appropriately physically meaningful contact models.
calibrated normal spring stiffness and a realistic ratio of the tangential
to normal spring stiffness, excellent results could be obtained by using 3.2. Bonding model
a simple linear spring model. The inelastic models were found to be
quite consistent for high normal coefficients of restitution but signifi- Early work of Cundall and Strack [13] and Cundall [22] used particles
cant differences are noted between the models as the normal coeffi- consisting of dense packing's of arbitrarily sized circular particles that
cient of restitution reduces. Zhu et al.'s [21] detailed review on other are bonded together at their contact points and simulated by the DEM.
major theoretical developments in DEM observed that the theories Similar approaches to modelling a solid material with bonded disks
underlying the DEM and models for calculating the contact forces have been used by Meguro et al. [23], Morikawa et al. [24] and others.
between particles are more sound, although not perfect yet. Secondly, This model has been used to model rock as a direct, particle-model
more contact points have been used in DEM simulation, which makes approach to modelling inelastic deformation and fracture of rock [25].
the DEM model more applicable to particulate research. The most The bonded-particle model (BPM) for rock (referred to by many
important development in this direction is the so-called CFD–DEM authors as BPM) directly mimics the rock behaving like a cemented
coupling approach which can handle the particle–fluid flow widely granular material of complex-shaped grains in which both the grains
encountered in process engineering. Thirdly, the theory to link the and the cement are deformable and may break [26]. The BPM con-
discrete simulation to continuum modelling is gradually established. ceptual model can, in principle, explain all aspects of the mechanical

Fig. 2. Distribution of collision frequencies vs collision energy components in the normal and tangential directions for the three mill sizes [28].
 Author's personal copy

N.S. Weerasekara et al. / Powder Technology 248 (2013) 3–24 7

behaviour. This system exhibits a rich set of emergent behaviours that both in the normal and tangential (shear) directions. Spectra can
correspond well with those of real rock. It provides the ability in DEM then be calculated for each type of collision event and for each ball
to investigate the micro-mechanisms that combine to produce complex and rock size class. These collision energy spectra provide the oppor-
macroscopic behaviours and predict these macroscopic behaviours [26]. tunity to better understand the various contributions to the overall
The grain-based portion of the force–displacement behaviour at each energy dissipation within the mill and how the energy supplied by
contact is described by six parameters (Fig. 1). Potyondy and Cundall the mill to the particles is consumed. For example Weerasekara
[26] and others provide detailed descriptions of the contact force– et al. [28] presented the technique applied to different sized SAG
displacement calculation. The maximum tensile and shear stresses acting mills (Fig. 2) and ball mills [28].
on the parallel-bond periphery are calculated from beam theory to be: A DEM simulation for a typical mill section with several thousand
particles generates massive amounts of output data, running to several
n s
max −F jM jR Gigabytes. Even with good sampling techniques [29] this generates files
σ ¼ þ ð3Þ
A I of close to or bigger than a Gigabyte in size. Although some DEM soft-
 n
max −F
n M R ware provides a GUI based data analysis environment, their techniques
τ ¼ þ
A I either require a considerable amount of processing power or are not
capable of delivering the required inputs for a fully mechanistic commi-
If the maximum tensile stress exceeds the tensile strength nution model. Therefore, a data logging system is required to fulfil
 max 
σmax ≥σ c or the maximum shear stress exceeds the shear strength the following objectives: provide flexibility in handling huge data sets
τ ≥τ c then the parallel bond breaks, and it is removed from the generated by typical DEM code; extract particle collision information
model along with its accompanying force, moment and stiffness. In from the DEM output database; and summarise data to provide more
general, the BPM is characterised by the grain density, grain shape, useful information for further comminution modelling processes. To
grain size distribution, grain packing and grain-cement microproperties. this end, considerable effort has been directed into data extraction
Each of these items influences the model behaviour. For more details, and analysis dedicated to providing the outputs required for comminu-
collections of publications are contained in the volume of Shimizu tion modelling, as in the work of Weerasekara et al. [30,31] and Powell
et al. [27]. et al. [32].

3.3. Energy logging

4. Breakage
The total energy dissipation associated with particle interactions
can be calculated using the DEM contact law, with components The DEM can be used to directly predict breakage. The BPM model,
being normal and tangential. It should be noted that the results are which was discussed in the previous section, was developed by
dependent upon the law that is utilised and the simplifications that Potyondy and Cundall [26] for the purpose of simulating rock break-
are applied. The losses from all the individual collision events can age. The concept is based on bonding or gluing a packed distribution
be collected to form frequency distributions of these energy losses, of spheres together forming a breakable body. By controlling the strength

Fig. 3. (a) 3D fracture network (DFN), (b) the corresponding 3D synthetic rock mass sample, and (c) synthetic rock mass basic components. The colours in (b) and (c) denote intact
rock blocks bounded by joints. Notice the internal non-through-going jointing in the “intact” rock blocks. [39].
 Author's personal copy

8 N.S. Weerasekara et al. / Powder Technology 248 (2013) 3–24

and mechanical properties of each bond it is possible to emulate the properties were not prescribed directly, rather the rock mass response
mechanical properties of rock material and breakage. The approach has arises from the combined behaviour of intact material and joints
been applied and further developed by Cho et al. [33]. (which includes fracturing, joint shearing and opening, and joint propa-
Potyondy et al. [25] calibrated the PFC 2D model of rock to capture gation and coalescence). This method was used in Ivars' work for rock in
the following characteristic properties of granite: elastic pre-failure large-scale continuum models of cave mining; large-scale open pit slopes
microcracking (in the form of distributed bond breakage), peak failure mechanisms investigations; charactering geo-mechanical behav-
strength under both confined and unconfined compression conditions iour of coal seam reservoirs; and to study the influence of veining in
and softening stress-strain relation in the post-peak region, plus in sub- intact rock strength [27,41–45].
sequent work by Potyondy and Cundall [26], Tan et al. [34] acoustic For modelling the brittle fracture of particles during impact, Potapov
emissions — in the form of kinetic energy release arising from bond and Campbell [25] used assemblies of glued polygonal particles with
breakage. Thornton and Antony [35] explored quasi-static shear defor- and without internal finite element style calculation for polygonal par-
mation of a polydisperse system of soft elastic spheres with interface ticle deformation. The dynamic breaking of glued bonds, in a manner
friction and adhesion. Kafui and Thornton [36] presented numerical similar that of the BPM model, allows breakage to be predicted. Both
simulations of spherical, crystalline face centred cubic agglomerate this and the BPM model have the disadvantage of needing to model
impacting orthogonally with a target wall. The effects of impact velocity all the sub-particle elements of each macro-particle, significantly in-
and bond strength on the evolution of the wall force, kinetic energy of creasing the simulation cost. To avoid this cost penalty, Cleary et al.
the agglomerate and the proportion of bonds broken were demonstrat- [46] introduced a replacement strategy. The collisional energy and/or
ed. They concluded from the observations of bond breakage evolution, stress are monitored for each particle and when breakage thresholds
that fracture is shear-induced and, in effect, the agglomerate creates are exceeded then the parent particle is instantly replaced by a collec-
its own flaw population during loading. It was also observed from tion of progeny that are densely packed into the space of the parent
DEM simulations that fracture occurs as a result of strong force trans- particle. This assumes that the fracture event is fast compared to the
mission into the agglomerate that creates a heterogeneous velocity other particle scale processes, which is the case for comminution appli-
field. This produces shear weakening along velocity discontinuities cations. This initial method allowed spherical particles to be broken into
that subsequently become the potential fracture planes [37]. Diametri- spherical progeny. It used an optimal Apollonian packing approach to
cal compression of agglomerate modelled within DEM [38] showed a specify the progeny size distribution and the progeny arrangement.
progressive fracture process. During loading, bond breaking develops Recently, this has been extended by Delaney et al. [47] to allow non-
along potential fracture planes and, immediately the applied force spherical particles to be broken and replaced by non-spherical progeny
drops, there is a rapid readjustment to break further contacts so as to whose size distribution is given by empirical measurement.
create a compatible set of broken bonds in the upper and lower parts The main advantage of using the PBM approach is computational
of the agglomerate in order to produce a ‘smooth’ fracture plane. economy as fewer particles are needed for modelling a rock population
Ivars et al. [39] adapted a technique using the BPM model to repre- than the BPM approach where every rock particle is built up by hundreds
sent intact material and smooth-joint contact model (SJM) [40] to or thousands of bonded spheres.
represent the in situ joint network, which was defined as a synthetic In addition to the PBM and BPM approach FEM-DEM based 2D
rock mass (SRM) (Fig. 3), to simulate standard laboratory tests (UCS, codes have been used for modelling rock breakage. The primary use
tri-axial loading, and direct tension tests). The joint contact was de- of this approach has been to capture the influence of microstructure,
scribed as smooth because the particle pairs jointed by a smooth-joint such as cracks and mineral grain distribution, on progeny composi-
contact may overlap and slide past each other instead of move around tion. Until these models are extended to the third dimension, their
one another. The rock mass behaviour was dependent on the in-situ contribution to modelling rock breakage will be limited.
joint network. The weakening and softening effect produced by rock
joints depended on joint orientation, density, size, strength, and stiff- 5. Ore characterisation
ness. Rock mass brittleness and primary fragmentation also depended
on the character of the in-situ jointing. The SRM methodology allows The application of a mechanistic model is of no use without the ability
for consideration of a complex non-persistent joint network as well as to mathematically describe the response of the rock to the applied colli-
block breakage that includes the impact of incomplete joints (non- sion forces. This is possibly the weakest aspect of the development of the
block-defining) on block strength and deformability. Rock mass application of the DEM in comminution modelling. Powell and Morrison

Table 2
List of breakage modes with their associated testing technique and response.

Mode of breakage Device Mode of operation Particle size Energy range Comments
range (mm) (kWh/t)

Single impact body breakage JKDWT Single particle 12–70 0.1–2.5 Good accuracy and reproducibility.
JKBT Single particle 5–45 0.1–3.5 Average rock response. Good accuracy.
ILC Single particle 0.5–20 0.1–2 Individual rock response. Good accuracy.
ILC Mono-layer of 0.1–2 0.01–1 Average rock response. Good accuracy.
many particles
Schönert rotary impactor Many particles 0.025–3 0.1–3 Average rock response. Good accuracy.
Incremental impact body JKRBT Many particles 5–45 0.002–0.2 Average rock response. Good accuracy.
damage and breakage Schönert rotary impactor Many particles 0.025–3 0.001–0.2 Average rock response. Good accuracy.
Impact surface breakage JKRBT Many particles 5–45 0.002–0.05 Average rock response. Fair accuracy
Schönert rotary impactor Many particles 0.025–3 0.0005–0.05 Average rock response. Good accuracy
Impact surface incremental Schönert rotary impactor Many particles 0.025–3 0.0001–0.01 Average rock response. Fair accuracy
damage and breakage Banini mill Many particles 5–20 0.0002–0.001 Average rock response, smeared energy input uncontrolled over
input range [58].
Abrasion surface damage Powell abrasion mill Many particles 20–250 0.004–0.1 Average rock response, smeared energy input uncontrolled over
input range, energy distribution extracted via DEM simulations.
Bed breakage Piston and Die Many particles 2–20 0.1–2.0 Average rock response, smeared energy input over the particles.
Good reproducibility.
 Author's personal copy

N.S. Weerasekara et al. / Powder Technology 248 (2013) 3–24 9

a. Normal energy distribution These are further broken down by the mode of application. Table 2
provides a succinct list of these modes, with the associated testing
technique, particle size range, energy range and a comment on the
quality of the test. Based on specific comminution energy Ecs (kWh/t)
input, the amount of breakage or breakage index t10 can be defined
as t10 = A[1 − e(−b. Ecs)], where A and b are ore impact breakage
parameters [51].
All the impact breakage tests use a steel anvil to apply the energy
to the rock. For example, the JK drop weight test (JKDWT) employs steel
weight dropped from a known height to apply energy to rock particles
that are resting on a steel anvil. The JK Rotary breakage tester (JKRBT)
and Schönert rotary impactor use a central spinning rotor to accelerate
a particle to an accurate velocity, using controlled kinetic energy to
break the rock particle against one of the ring of steel anvils. The impact
load cell (ILC) is essentially a hybrid between the traditional drop
weight tester and the Hopkinson pressure bar [52], providing more
effective and convenient measurement of the properties of materials
during dynamic loading. The distribution of energy between impacting
bodies is dependent upon their relative material properties. This is gen-
erally not accounted for, except in the work of Tavares [53], in which
he assumes the validity of the Hertz contact theory to account for
material-dependent energy-split between bodies in contact.
b. Tangential energy distribution Some comments on the ore characterisation testing techniques
are described below:

5.1. Impact damage

Extensive testing of ores subjected to a wide range of impact damage

have been conducted using the JKRBT and at the fine end (sub 2 mm)
on the Schönert breakage tester housed at the University of Erlangen.
This data is being used to test and extend the application of the theories
of Vogel and Peukert [54] and methods of Shi and Kojovic [55] to model-
ling the size dependency and incremental damage of particles subjected
to impact damage. As highlighted in the work of Powell et al. [49],
and shown by the large number of very low impact energies below
0.1 kWh/t ( in Fig. 4), the majority of breakage for particles in the
DEM size range will be via incremental damage rather than single
impact breakage. Therefore considerable effort has been applied to
measuring and modelling this mode of damage.
Using an alternative approach, the incremental damage has been
investigated using results from measurements of distribution of particle
fracture energies, measured using the impact load cell, and repeated im-
pacts of particles in a drop weight test apparatus. Along with a model
based on continuum damage mechanics these have been successfully
Fig. 4. Truncated cumulative energy distributions per particle per second [32].
used by Tavares to describe breakage due to low-energy impacts [56].

[48] provided an introduction to these requirements, and Powell has 5.2. Surface damage
provided further insights, Powell and Weerasekara [32], Powell et al.
[49]. For impacts that do not cause bulk body breakage, an abrasion or
The degree of breakage of rocks is strongly dependent upon the surface damage model is required. As in the impact body damage
collision force, the energy dissipated in the collision and the mode model, this requires a prediction of the probability of damage, the
in which it is applied. In order to develop a mechanistic comminution amount of damage (mass removed) and the appearance of the progeny
model the contribution of the modes and forces needs to be calculated for abrasion events as a function of the impact energy or force. To this
as a function of equipment set-up and operation, the DEM is well suited end, the promising results of earlier abrasion studies, Powell and
to this task. Morrison [48], have been further tested. The Powell Abrasion mill [57]
It is proposed by Powell and Weerasekara [32] that the principal is being developed to test this mode of breakage. A simple batch mill
modes of breakage or damage are: of 300 mm long with a diameter from 0.8 to 1.8 m is loaded to 50%
filling with discrete rock sizes and run at a slow speed to bias the rock
• body — major damage or fracture of the particle
interactions to sliding abrasion surface damage. A water sprayer is
• surface — only the surface layer is damaged or removed. This
used to rapidly remove the progeny, and the product collected over
includes smoothing off of the initial surface angularity and subse-
timed intervals. DEM simulations of this simple system with only a
quent steady rate of damage of the smoother surface.
few hundred particles are used to log the collision energies over the
The terms breakage versus damage can be used to differentiate duration of the test, allowing a rock-specific correlation to be derived
between the loss of integrity of the particle versus the introduction between applied energy and the resultant wear rate and progeny size
of some type of fractures in the particle, but it remains intact [50]. distribution. Larbi Bram [58] used low energy repeated impacts to
 Author's personal copy

10 N.S. Weerasekara et al. / Powder Technology 248 (2013) 3–24

study surface, developing a relationship between degree of damage and A selection of comminution devices that the authors are working
input energy. are presented as follows.

5.3. Fine particle breakage 6.1. DEM modelling of rock crushing

As outlined in previous papers on the UCM and modelling The incoming feed material to a cone crusher is subjected to a
[49,59–61], the DEM simulations only cover particles above a cut-off number of consecutive compression events. In each event the material
size that limits the particles to a number that the simulation can is compressed with a specific compression ratio depending on the
cope with, along with a ratio of particle diameters – generally below position in the chamber, chamber design and crusher dynamics. The
20 – that the contact overlap algorithm can account for with a feasible compression ratio is the main machine parameter influencing the
time-step interval. Breakage and damage of particles with sizes below amount of breakage in each event. The properties of the feed material
this cut-off size, called sub-DEM particles have to be separately in terms of particle size, particle shape and strength also influence
accounted for. This is currently poorly or not accounted for. Breakage the outcome of compression. These insights regarding the crushing
testing is being addressed by fine particle breakage techniques, such operation have generally been formulated based on experimental
as the mono-layer bed breakage [62], the Schönert rotary tester and observations and mechanistic analytical derivations based on first prin-
adapting vertically loaded ball breakage tests [63]. However, this still ciple equations.
has to be transferred to the breakage environment within the commi- However, it is a challenging task to directly monitor what happens
nution device. An approach is proposed in Section 7.1 with the mecha- inside a cone crusher or other crushers such as e.g. the vertical shaft
nistic ball mill model. impact crusher. DEM has opened up the possibility to model and sim-
ulate crushers in order to investigate how particles are flowing and
6. Application to specific comminution devices interacting with the machine components. The simulations give in-
sights regarding where and how rock particles are subjected to loads
The process of comminution requires staged size reduction, with a exceeding the strength of the particle, resulting in crushing. Hence,
series of devices each targeting a range of size reduction. These differ- the previously stated analytical models are now possible to evaluate
ent comminution devices present quite different mechanical environ- and validate.
ments to the rocks, so require a range of modelling approaches and For cone crushers DEM has contributed in the following way:
application-specific techniques:
• Understanding and insight regarding particle dynamics, types and
• Crushers capture rocks in a compression chamber that squeezes the rates of interaction, compressive forces and flow behaviour in the
rocks, either individually or in a bed, forcing them to break if their crushing chamber.
(small) elastic deformation limits are exceeded. • Validation of hypotheses forming the conditions for analytical
• Mills are large rotating horizontal cylinders that use balls and or mechanistic flow models derived from first principle mechanical
rocks to form tumbling grinding media. equations.
• Impacting machines either strike the rocks with anvils, such as in a • Deeper understanding regarding the influence of segregated and
hammer mill, or throw the rocks at an impacting surface, such as in miss-aligned feeding conditions on crusher performance.
a vertical shaft impact crusher.
A few publications exist on the topic of using DEM for rock crushers
These different environments require different modelling approaches. in general and cone crushers in particular. In the case of impact crushers
For example: Djordjevic et al. [64] as well as Schubert [65] have simulated a horizon-
tal impact crusher using the BPM approach. However in both cases rel-
• For crushers the breakage must be simulated within the DEM environ- atively few particles have been used and the geometries are simplified.
ment as the compressive chamber contracts to below the rock dimen- Cunha et al. [66] have investigated the distribution of collision energies
sions so the DEM rocks would protrude through the walls, rendering and the residence time distribution inside the vertical shaft impact
the simulation meaningless. crusher. Khanal [67] investigated impact breakage, single particle com-
• In mills the rate and energy of collisions can be recoded in a steady- pressive breakage and inter-particle breakage of different types of ag-
state environment, to be later applied to a comminution model. It is glomerates and rock types using DEM. A DEM model for simulating
a reasonable assumption to ignore breakage during the simulations. rock breakage in cone crushers has been presented by Lichter et al.

Fig. 5. DEM simulation of a cone crusher in secondary application in which the BPM is used. Left: Rock particles are displayed by the bonds between sub-particles. Right: Rock par-
ticles are displayed with sub-particles [66].
 Author's personal copy

N.S. Weerasekara et al. / Powder Technology 248 (2013) 3–24 11

and liner wear were higher for a coarse feed, even though the power
draw and flow pattern were indistinguishable. Cleary [69] extended
the DEM ball mill model from a 3D slice to the entire ball mill and
modelled both full chambers of a two chamber cement ball mill. In
the second chamber, where axially stepped classifying liners were
used, complex ball segregation patterns were identified.

6.3. Stirred mill DEM modelling

The first detailed exploration and comparison of tower and a pin mill
were by Sinnott et al. [82] and Cleary et al. [83]. Power draw, flow struc-
ture, energy utilisation, wear and particle transport efficiency were all
evaluated. The tower mill was found to perform well on all measures
whereas the pin mill was found to suffer from excessive localisation of
the high energy impacts close to the pins and from poor charge trans-
port. Fig. 7 shows an example of the flow in an industrial scale tower
Fig. 6. DEM simulation of a double roll crusher using the particle replacement mill (with around 10 million particles) simulated using the Cleary
technique.
DEM code. The particles are coloured by axial speed which clearly
shows the upward stream of media within the screw and the down-
[68]. DEM models of breakage in a jaw crusher and vertical shaft impac- ward flow of media in the outer annular region. Morrison et al. [84]
tor were given by Cleary [69] using the replacement strategy. Cleary later compared, using DEM, a matched lab scale ball mill and pilot mill
et al. [46] presented a model of a choke fed cone crusher in which prod- that produced equivalent product to establish that the tower mill was
uct size distribution and throughput were predicted directly from the able to break particles more efficiently because it generated a sharper
DEM model using the same breakage modelling approach. The BPM ap- and more focused collision energy spectra. The ball mill in contrast pro-
proach has also been used by Quist [70] to simulate a secondary duces a wide flat spectrum which is advantageous for the robustness of
crushing operation, see Fig. 5. The BPM rock material model was cali- the process to changes in feed size but at the cost of being fairly uni-
brated using single particle compressive breakage experiments. The formly inefficient for all feed sizes. Sinnott et al. [85] considered the ef-
DEM output results in terms of mass flow, power draw and pressure fect of slurry in the tower mill using the 1-way coupled DEM-SPH model
were compared to and validated by industrial scale experiments. The first proposed by Cleary et al. [86]. Finally, Sinnott et al. [87] have ex-
same approach was used for investigating the influence of liner wear plored the effect of non-round media on stirred mill performance and
on crushing performance for a primary gyratory crusher [71]. Fig. 6 showed that high levels of non-sphericity produced strong and detri-
shows an example of flow and breakage prediction in a double roll mental changes in the flow structure and collisional environment in
crusher using the DEM code of Cleary. Fracturing of feed material the narrow grinding shear zone between the upward stream
(coal in this instance) of up to 250 progeny of mm into as small as transported by the screw and the downward gravity driven outer annu-
10 mm is resolved. lar flow.
When modelling compressive breakage mechanisms in e.g. a cone Another important type of stirred mill that has also been studied
crusher the rock material model needs to respond to effects and phe- using DEM is the Isamill. Yang et al. [88] studied the flow of grinding
nomena such as incremental damage, particle loading condition and media in a simplified IsaMill using DEM in terms of velocity, force and
breakage mode. Generally, a rock particle subjected to a load will either power draw under dry conditions. It was reported that the increase in
be; undamaged, weakened, abraded, chipped, split or broken. In the stirrer rotation speed or media filling rate increased the flow velocity,
general application of the PBM approach only the last effect is consid- compressive force and the mill power draw. This work was later
ered as the particle only responds to loads above the set threshold. extended to multiple discs for studying the micro-dynamic properties
Hence it is not capable of capturing particle weakening, chipping or related to the grinding performance [89]. A similar investigation
abrasion as the particle is subjected to multiple particle–geometry and studied the influence of grinding media properties (i.e. media sliding
particle–particle interactions at loads below the threshold level for
complete breakage. However, Morrison et al. [72] have modelled abra-
sion independently and implemented the capability to accumulate a
stress and damage history. Another exception is the work by Tavares
and Carvalho [73], who used the PBM in addition to the fracture energy
convolution model to describe the damage and weakening of particles
subjected to stresses, in addition to size reduction in a grinding mill.
What is lacking here is the detailed experimental work against which
to calibrate and validate this approach. In the future the PBM breakage
function may be able to react to these effects.

6.2. Ball mill DEM modelling

The earliest DEM model of a ball mill was in 2D by Mishra and

Rajamani [74,75]. This was followed by Cleary [17,76] and then Cleary
and Sawley [77]. Then Mishra [78,79] provided a review on DEM, its
use of contact models and application to tumbling mills in general.
Cleary [80] considered axial transport and grate discharge in a dry ball
mill and showed that in the absence of air flow that this is a slow diffu-
sion process. Cleary et al. [81] examined the effect of feed size on the en-
ergy utilisation in cement ball mills. They showed that the fraction of Fig. 7. Prediction of media flow in a full scale tower mill with particles coloured by
energy used in grinding was higher for a finer feed and that media their axial speed.
 Author's personal copy

12 N.S. Weerasekara et al. / Powder Technology 248 (2013) 3–24

6m mill

Fig. 8. Analysis of the breakage zones in a mill through impact energy spectra.

friction and restitution coefficients, media density and size, slurry discharge end that applies a significant reverse flow to the system.
properties) on the media flow and grinding performance in a sim- This pushes the media up against the feed end, rendering the simula-
plified Isa Mill by Jayasundara et al. [90,91]. Jayasundara et al. [92] tion of different fillings to be of little practical significance. Rather, it is
showed that the grinding process in a stirred mill follows first-order necessary to establish the realistic media packing in the grinding zone
kinetics and the rate constant can be correlated to the impact energy. in order to conduct realistic simulations. Cleary et al. [81] used DEM
Jayasundara et al. [93] extended their work by incorporating a wear to explore the distribution of media and much finer feed material in a
model into DEM and studied the wear pattern of the disc holes in a periodic section of an Isa Mill and have included the fluid flow in simu-
model Isa Mill. This range of work suffers from the limitation of lations. These outputs should be far more useful in understanding the
modelling a batch segment. The real mill has an expeller at the milling process.

Fig. 9. The DEM calculated energy spectrum for 35 mm rocks in a SAG mill [101].
 Author's personal copy

N.S. Weerasekara et al. / Powder Technology 248 (2013) 3–24 13

20 1.0
No particle
0.425 - 0.300 mm
1.70 - 1.40 mm
6.30 - 4.75 mm 0.8
15
Capture radius rc (mm)

Model

Cumulative distribution
0.6
10

0.4
5

0.2

0.62

1.10
0
0 2 4 6 8 10
Impact energy (J) 0.0
0.1 1 10 100
Fig. 10. Comparison between measured and fitted values of radii of capture as a func- Normal collision energy -E (mJ)
tion of impact energy and particle size for impacts with 88-mm steel ball of particles
sitting on a flat surface [96]. Fig. 12. Normal collision energy spectra for DEM simulations presented in Fig. 11.
Collision frequencies were 479,957 and 340,249 s−1 and median collision energies
were 0.62 mJ and 1.10 mJ for the mill with smooth and square lifters, respectively [73].

6.4. SAG/AG mill DEM modelling It is evident from energy distributions for a range of particle sizes
that the smaller particles receive considerably more impacts than larger
DEM can be used to provide the collision environment in each particles and are subjected to higher specific energies of impact.
distinct grinding zone in a mill, as illustrated in Fig. 8. This analysis This simulated energy environment inside a mill, where low energy
allows both qualitative insights into the influence of changing operat- impacts are dominant (causing abrasion rather than particle breakage)
ing conditions and quantification of the contribution of different is dramatically different to the traditionally understood SAG mill mode
grinding zones to the overall milling process. The simulations for of breakage. Therefore, rounding and abrasion are likely to be the dom-
SAG/AG mills are revealing that the conditions for breakage are differ- inant comminution mechanisms for particles larger than approximately
ent to those traditionally assumed and reported by existing breakage 30 mm and incremental breakage for particles in the size range 10–
tests. This is evident from the primary output of the DEM simulations, 30 mm. This also indicates that SAG/AG mills are not very effective at
the calculated particle energy spectra shown in Fig. 9. The DEM rapidly comminuting this size range of particles.
calculated energy spectrum for a SAG mill indicates that most of the
particle impacts are occurring at very low energy [28], in fact this 7. DEM based modelling techniques
energy level is well below that required for a typical ore to undergo
breakage. For single impact breakage to occur, the energy would need For most comminution devices detailed use of DEM to inform re-
to be above 0.5 kWh/t, however the figure indicates that less than searchers of the underlying mechanisms driving the comminution
0.1% of impacts are in this range. There are further 2% of impacts that process has come increasingly important. These represent some of the
are sufficient to cause incremental damage and breakage through major areas of contribution of the DEM to the science of comminution.
repeated impact events. The majority of the impacts in a SAG mill do
not cause particle breakage for the rocks (plus 15 mm) in the mill. 7.1. Mechanistic ball mill model
Most impacts are of low energy and are more likely to cause attrition
than fragmentation. Therefore, comminution models based on the con- Following the footsteps of the work originally carried out in the
cept of high energy single impact breakage events do not reflect the University of Utah during the early 1990s [94], the research group
collision environment within the mill and are likely to be misleading. at the Universidade Federal do Rio de Janeiro have proposed a novel

Fig. 11. DEM simulations on the effect of liner design in a 0.6 m diameter mill operating at 68% of critical speed, with 30% ball filling and 25 mm steel balls [73].
 Author's personal copy

14 N.S. Weerasekara et al. / Powder Technology 248 (2013) 3–24

10 damage them internally, so that their strengths (fracture energies)

Square, Dm=0.6m can vary as milling progresses;
• It explicitly deals with the grinding media (balls) trapping and
Specific breakage rate(1/min)

Smooth, Dm=0.6m crushing a bed of finer ore particles;

• It is able to model the transient domain, making it possible to de-
scribe non-first-order breakage rates that are commonly observed.
Therefore, whereas most DEM-based approaches actually account
1 for a single state variable to describe comminution processes – particle
size – the mechanistic model describes an additional variable, the particle
strength (fracture energy). A third variable, namely, particle component
has also been included, although the current version of the model can
only handle independent components, not transfer of particles from
one component class to another, which is what is required to describe
liberation.
Different approaches need to be used to model different commi-
0.1 nution devices. For instance, the approach that is used when model-
0.1 1 10
Particle size (mm) ling machines such as ball mills is essentially a one-way coupling of
DEM and the population balance model, that is, DEM is simulated to
Fig. 13. Effect of liner profile, depicted in Fig. 11, on breakage rates (30% mill filling and provide the collision energy spectrum, which is then used to predict
68% of critical speed). breakage using the microscale population balance model. In this
case, DEM is simulated considering only the grinding media, and the
simulations are conducted accounting for the fact that contact param-
DEM-based approach to model comminution machines [73,95]. The
eters have to consider that powder exists between colliding grinding
model is based on the combination of DEM that is used to describe
media [73,96–98]. In the case of other mill types, which include au-
the mechanical environment in the comminution machine, to a myriad
togenous and semi-autogenous mills, as well as crushers, ore particles
of empirical and semi-empirical models that describe the different
have to be included in the simulations, besides the grinding media,
breakage modes and that are fitted to particle breakage data, and to
whenever present. For these cases, a two-way coupling becomes nec-
models that describe how “sub-DEM” particles are captured and how
essary, since the size distribution and composition of solids contained
the collision energy is split amongst particles when they are assembled
in the mill will vary with time, representing a significantly greater
in beds. These pieces of the puzzle are then stitched together using two
challenge in applying the model framework [97]. A more complex
models: the microscale population balance model, which keeps track of
case is the vertical impact shaft crusher (VSI), which has been modelled
the mass balance of particles in each size and ore class, and the fracture
considering that particles suffer different collision events and remain
energy convolution model, which describes the variation of the distri-
for different time periods in different regions in the crusher [66], but
bution of particle fracture energies inside the machine as comminution
work is on-going to describe breakage in this machine.
progresses. This latter is based on the continuum damage model of par-
In the mechanistic model, the collision energy spectrum is repre-
ticle breakage [56] and the distribution of particle fracture energies.
sented by the energy loss, either total or normal, rather than the
Both equations will vary according to the mode of operation of the
kinetic energy of the collision, which has been used by a number of
comminution device, if batch or continuous, and also according to the
researchers. This is consistent with the use of fracture energy infor-
mixing conditions that prevail.
mation, which is the energy that is dissipated when a particle is com-
In essence, the model distinguishes itself from other attempts to
pressed between steel grinding bodies and loses its integrity [52]. In
use DEM information in comminution modelling because:
this case it becomes essential that, when characterising the response
• It discriminates between the different breakage mechanisms, of a particle to breakage, the energy absorbed, rather than the kinetic
namely body and surface breakage; input energy is described. Whilst this could earlier only be measured
• It accounts for different components in the feed, although as inde- using the twin pendulum device [99], it has been demonstrated that it
pendent from one another; can be measured in a drop weight using the impact load cell [100].
• It considers the distribution of strengths (fracture energies) in the The use of energy loss, however, creates a special challenge in DEM
ore, that is, it considers that particles contained in a given size and parameter calibration, since it is more sensitive to the choice of
component class do not have the same breakage strengths; parameters in the simulation than the kinetic energies. As such, it is
• It considers that the distributions of strengths vary as a result of recognised that DEM parameter calibration in mills is an area that
low-energy impacts that are unable to break the particles, only requires much greater attention in the future.
One good example of the mechanistic model approach is in de-
scribing breakage in a dry batch grinding mill [73,101]. The equation
that describes the rates of changes in mass of material contained in
Table 3
Definition of terms in Eq. (3) [73].
size class i in a well-mixed mill is

dwi ðt Þ ω h i
Breakage mode Rate of appearance (A)

Body (b) X
i−1 ∞ 1 ¼ −Di;b ðt Þ−Di;s ðt Þ þ Ai;b ðt Þ þ Ai;s ðt Þ ð4Þ
Ai;b ðt Þ ¼ wj ðt Þ∫ mj ðEÞpðEÞ∫ bij ðeE; t ÞF j ðeE; t ÞpðeÞdedE dt M
j¼1 0 0
X
i−1 ∞ 1
 
Surface (s) Ai;s ðt Þ ¼ wj ðt Þκ j ∫ mj ðEÞpðEÞ∫ aij ðeE; t Þ 1−F j ðeE; t Þ pðeÞdedE Functions A and D represent the rate of appearance and disappear-
j¼1 0 0
ance of material in class i due to fracture, being defined in Table 3,
Rate of disappearance
∞ (D)1 whereas subscripts b and s stand for body and surface breakage
Body (b) Di;b ðt Þ ¼ wi ðt Þ∫ mi ðEÞpðEÞ∫ ½1−bii ðeE; t ÞF i ðeE; t ÞpðeÞdedE mechanisms, respectively.
0 0
∞ 1 The density distribution of stressing energies in the mill p(E) is
Surface (s) Di;s ðt Þ ¼ wi ðt Þκ i ∫ mi ðEÞpðEÞ∫ ½1−F i ðeE; t ÞpðeÞdedE calculated from the cumulative distribution, from p(E) = dP(E)/dE.
0 0
The product eE is the fraction of the impact energy that is absorbed
 Author's personal copy

N.S. Weerasekara et al. / Powder Technology 248 (2013) 3–24 15

by each particle captured in an impact event. Details of the models 7.2. Unified comminution model (UCM)
that describe the different breakage mechanisms and solution tech-
niques can be found elsewhere [52,102]. The concept of a unified comminution model (UCM) was formulated
As part of the mechanistic modeling approach, models that sit in with the vision of bringing all comminution models onto a common
the mesoscale, that is, neither in the particle level nor in the mill level, base [103]. The structure of the model relies on knowledge of the
are used to predict what happens to sub-DEM particles that are assem- mechanical environment in the comminution devices — DEM has
bled in beds. It is then relevant to predict the mass of material captured been used as the tool to simulate this [49,60,104].
in each collision event, as well as the energy split amongst particles The UCM model is developed [32,61] as a mechanistic model
contained in the bed. applicable to all comminution devices and novel equipment designs,
The energy distribution derived for the colliding ball charge needs to model the comminution process rather than the comminution
to be applied to the ore contents of the mill. The allocation of ball equipment outcomes, and to guide the development of improved
impact energy to beds of particles in a slurry is a highly complex scenario breakage testing techniques. The aim is to develop a fully predictive
to model. Tavares and Carvalho [73,101] have begun to address this for mechanistic model of breakage occurring in devices, that can handle
dry powder beds. all known modes of breakage — low energy impact surface damage,
The mass of material of size class i captured between two grinding chipping, sliding abrasion, cumulative impact body breakage and
media elements (mi) may be estimated from the product of the number transition of rock between modes. This aims to enable the true inte-
of particles caught between the media elements (Ncap,i) and the average grated plant simulation that is required for circuit design and optimi-
 
weight of the particles m p;i , by mi ¼ m p;i Ncap;i . Assuming that, in sation, especially when incorporating novel equipment or processes.
unconfined bed, particles normally only break when they are typically The model has reached the point of delivering test simulations of a
squeezed down to a monolayer between grinding media and that they pilot SAG mill and the next stage is application to full production mills.
are arranged in a dense hexagonal packing, then the number of particles The approach taken to develop the UCM was to develop a particle-
captured as a function of radius could be estimated by the method based model that describes the particle collision environment in
proposed by Barrios et al. [96]. detail. Discrete element modelling (DEM) is used to simulate the
Barrios et al. [96] proposed that the radius of capture can be calculated mechanical environment of any given device. From these calculations,
by adding the radius of contact due to geometry and the radius of contact the rate and probability functions for the various types of collisions
due to the elastic deformation in the vicinity of contact, which are de- occurring are extracted [28]. The UCM tracks the mechanical collision
scribed individually using appropriate expressions that account for the environment experienced by all particle classes. The response of real
influence of the geometry and stiffness of the bodies in contact, the par- particles to applied breakage energies in the same range as the simu-
ticle fracture energies and the particle sizes. Fig. 10 compares the mea- lated energy range is measured. These two sets of information are
sured results and those fitted to the model, showing good agreement. then combined to predict the breakage rate and the size distribution
In the description of the energy split function in the bed p(e), it is of the particles produced. Material is moved through the device in
considered that the stressing energy is split equally amongst particles the simulation by adding a transport and discharge function and the
positioned within the active breakage zone in the mill. In this case the outcomes are to be distilled into a simulation model and framework
energy distribution function may be given by that can be run rapidly on a desktop computer.
The input data the model needs is therefore the actual breakage

mechanisms occurring in a given comminution device and the ore
pðeÞ ¼ δ e−1=Ncap;i ð5Þ
characterisation data measured under the energy conditions which
would exist in that particular device. These enable the calculation of
where δ is the Dirac delta function. the collision energies and collision rates for each particle size in the
Comparisons between measured and predicted size distributions comminution device. This physical information is provided by the
from batch grinding data have been presented [101], whereas a para- breakage testing program and the energy conditions are provided
metric investigation on the influence of selected design and operating by DEM simulations (Fig. 8). A major lesson from this work is that
variables in batch grinding has been given by Carvalho and Tavares current breakage techniques are inadequate for mechanistic modelling.
[98]. Simulations on the effect of mill liner on a laboratory mill are As yet, within the industry there has been poor uptake of learnings
presented in Figs. 11 to 12, which demonstrate that the different col- about the physical breakage environment into models. For example,
lision energy spectra from DEM translate into differences in breakage the assumption that energy is transferred to particles once per mill
rates calculated using the mechanistic ball mill model. revolution is still widespread. In actual fact, the charge circulation
Finally, given the time-dependent variation of particle fracture rate is 1.3 to 1.7 times per mill revolution, varying with mill speed
energies with comminution environment, then the model becomes and filling [105,106].
intrinsically dynamic. Given the complexity and the effort required As mentioned earlier, the DEM simulations are all truncated at a
in describing the dynamics of continuous full-scale tumbling mills, it lower size determined in principle by the limit to the number of par-
is proposed to use this model framework to simulate full-scale tum- ticles that can reasonably be simulated and by the relative sizes of the
bling mills operating in batch mode, predict the product size distribu- particles — with a 10-fold size range being the general limit before
tions as a function of time, and then fit the traditional population encountering issues with excessive particle overlap. The material
balance model to this data, determining parameters of breakage rate below this size is ignored in the simulation. In general this does not
(as in Fig. 13) and appearance (breakage distribution) functions. represent much mass of material (maybe less than 5% of the mill con-
Assuming appropriate residence time distributions and internal clas- tents), but it represents the majority of the product from the mill. The
sification functions, the model can then be used to predict grinding efficient removal of this material from the comminution device ensures
using the steady-state solution of the traditional population balance that there is only a small mass in the device. Typically for a full size SAG
model of the ball mill, which is available in a number of mineral process- mill, with a feed size range from 250 mm down to sub-mm size, the
ing plant simulators. This has the obvious advantage of not requiring the DEM simulations are truncated at 15 mm, and for a pilot mill at a
use of empirical scale-up relationships, which have been the limitation 3–5 mm, typically excluding about 10–15% of the material by mass.
of traditional population balance model-based approaches. It benefits The DEM can be used as input to predict the production of fine material
from the fact that the key design and process parameters such as mill from the breakage of the coarser particles, but the re-breakage of the
speed, mill filling and grinding media type and size, are directly sub-DEM material is critical to the correct predictions of the model, so
accounted for in the DEM simulations. cannot be ignored [32,49,61]. The solution to this issue has not been
 Author's personal copy

16 N.S. Weerasekara et al. / Powder Technology 248 (2013) 3–24

addressed in any of the literature, and not yet adequately investigated particularly in SAG mills where there are many weak collisions.
and presented in the published UCM work. Experimental support for the existence of this mechanism is given
by Tavares [56]. This has been implemented in the DEM code of
7.3. Virtual Comminution Machine (VCM) Cleary [69,107] and demonstrated in Morrison et al. [72].

The Virtual Comminution Machine (VCM) is a concept that brings The chipping mechanism is quite important because almost all
together DEM modelling [17,29,59,69,76,80,107–109], augmented by freshly broken ore particles, including feed, are angular in shape.
other models such as SPH models of slurry flow in mills [86,110,111] Applying an oblique force to a sharp corner is able to easily remove
and ore characterisation targeted at making particle breakage predic- the corner via a shear fracture in tension. This is a low energy mode
tion in such mill models sufficiently accurate [48,112–115]. The VCM of breakage, of which there are many shear events with sufficient
framework is motivated by early DEM modelling work, that showed energy so that rounding typically occurs quickly. Hence a particle
that breakage in SAG mills is not dominated by the cataracting stream, will be well rounded soon after it enters the mill charge or soon
and by advances in ore characterisation (see Section 5) and in under- after it is created by some body breakage event. A useful working def-
standing the fundamental breakage mechanisms occurring in different inition of the rounding mechanism is simply the removal of corners
comminution devices. The concept was introduced by Morrison and from freshly broken, angular particles. Parameterised super-quadric
Cleary [116] and summarised in Cleary and Morrison [117]. The VCM particles (see [107] for details) as used in the VCM allow particle
objective is to model each process within any comminution machine shape evolution, based on the energy dissipation by these mecha-
at the particle scale to an acceptable level of accuracy so that it can be nisms. This is demonstrated by Morrison et al. [72] where predictions
used as a development environment for existing and new comminution for the time evolution of a well characterised charge in a 1200 mm
machines. Such tools are critical for reducing the development period pilot mill are compared favourably against measurements.
for new generations of existing or novel machines, the capital required The predictions of the VCM are demonstrated here for a full 3D
for such development, and the risk involved in such developments. simulation of a Hardinge style AG/SAG pilot mill (first published by
Based on the energy distributions such as given in Fig. 9, commi- Cleary and Morrison [112], and then in Cleary [69,107,117]). This
nution models based on the concept of single high energy impact mill is 1.8 m in diameter with conical ends and an internal volume
breakage events in the toe region do not reflect the reality of the col- equivalent to a cylinder 0.6 m long. The Hardinge mill is the industry
lision environment that actually exists within the mill. A broader standard for pilot scale test work. It requires 350–450 k particles to
range of breakage mechanisms, including ones that operate at very simulate the mill charge down to 6 mm particles which means that
low impact energies, are involved in comminution. Five breakage around 90–95% of the mill charge is accurately represented in the
mechanisms are identified and implemented within the VCM: DEM model.
Fig. 14 shows a cut away view of the simulated mill including
1. Abrasion: reduction in diameter by shear interactions. Morrison discharge grates and pulp lifters. This is a complete mill model and
et al. [118] demonstrated the shrinking of ore particles due to includes the feed end, the main/belly/circumferential lifters, the end
abrasion in a 1200 mm pilot mill using mass loss rates calibrated lifters, the grate, the pulp lifters, the discharge cone and the external
with data from a 300 mm Bond mill test; mill shell. This configuration has 16 belly lifters with 10 degree face
2. Rounding: change in shape (reduction in aspect ratio and angularity) angles, 16 lifters on the feed end cone and 8 lifters on the discharge
by preferential frictional sliding on corners and high curvature parts end. A representative SAG mill charge was used, with a rock top
of the particle surfaces (demonstrated by Cleary and Morrison [109] size of 122 mm and the bottom resolved size of 6 mm. The fill level
in a 1200 mm pilot mill); was 39% with a ball load of 5%. The mill rotation rate was 24 rpm cor-
3. Chipping: preferential removal of corners and edges by normal responding to 76% of critical speed.
impacts (leading to shape change). This was demonstrated again Fig. 14 also shows the mill, sectioned by an axial plane and viewed
in a 1200 mm pilot mill by Morrison et al. [72]; along the axis looking towards the discharge end, with the particles
4. Impact breakage: single impact breakage due to large energy normal shaded by size. A typical mill flow pattern is visible, with the particles
impacts (first demonstrated by Cleary [76]); carried up clockwise by the motion of the mill until they reach the
5. Incremental damage: cumulative damage from normal impacts shoulder position and then flow down the cascading free surface of
leading to body breakage. This is a critically important mechanism, the charge. For these lifters and this mill speed, there are significant

Fig. 14. Full 3D model of flow in a Hardinge pilot mill; a) looking towards the feed end with particles coloured by size (red is large and blue is small) and b) looking towards the
discharge end with particles coloured by speed (with red being fast moving and blue being close to stationary).
 Author's personal copy

N.S. Weerasekara et al. / Powder Technology 248 (2013) 3–24 17

Fig. 15. Slurry flow within the: a) grinding chamber, and b) pulp chamber, for a viscosity of 0.1 Pa s and a fluid resolution of 5 mm. The fluid is coloured by its speed. The mill rotates
clockwise in this view.

cataracting streams that are composed almost entirely of fines (due to and slurry swirls around to fill the spaces between the lifters. In the
strong radial segregation) leading to significant numbers of impacts pulp chamber, flow is initiated along the pulp lifters at around the
on the liner above the toe position. The discharge grates can be seen 2 o'clock position. The fragmented stream falling from one lifter can be
on the end of the mill. Small collections of fines are also observed to seen striking the discharge cone and being deflected towards the trom-
collect on the end wall radial lifters. mel location.
Many mills, particularly SAG, ball and tower mill feed are operated The outputs from the VCM model include all the traditional DEM
with a wet collisional environment by introducing water into the mill ones mentioned more broadly in this review including energy spectra,
feed and removing slurry from the discharge. The usual purpose of power draw and liner wear. But with the inclusion of dynamic breakage
operating in such a wet mode is to significantly improve the transport models for five identified mechanisms, now also allows prediction of
of fine particles (product) through the mill. The slurry, consisting of resident and product PSD and throughput. With the inclusion of the
high solid loads of fine sub-millimetre particles is the medium in slurry component, axial transport of slurry and its discharge can be pre-
which the fine particles are broken and is where much of the mill dicted along with flow and discharge characteristics of the discharge
energy is expended. The proportion of fine (b 37 μm) material in the grates and pulp chamber. These are the majority of outputs required
slurry has a strong effect on its viscosity. A viscous slurry which can for complete balanced optimisation and testing of machine designs,
retain particles on the surface of the grinding media can help to evaluation of responses to changes in operating conditions, scale-up
ensure selection of those particles for breakage. The fluid and particle and testing of hypothetical new designs. Some developments remain
interactions are important for particle breakage and transport. There- to be done, including dealing with the presence of fine material for
fore, a model of slurry behaviour and transport is a critical element of ball and stirred mills, air sweeping of dry ball mills, and two-way cou-
the VCM for wet mills. pling of the fluid and solid components (which is needed for predicting
Models based on conventional grid based computational fluid pebble discharge from SAG mills).
dynamics (CFD) are not well suited to the high shear, high viscosity en- Many of these capabilities are relatively recent and significant
vironment with complex free surface behaviour found in wet tumbling testing and validation are still required to prove the predictive capa-
mills. SPH which is a Lagrangian mesh-less method (see [111,119]) is bilities of the VCM. But, in principle, it is a nearly complete particle
particularly well suited to predicting such complex free surface flows scale model with enormous potential for comminution equipment
and interaction with moving complex shaped equipment [120]. Cleary design. However, in order to make quantitative predictions from the
et al. [86] proposed a one-way coupled DEM-SPH model for the VCM. VCM, it needs to be remembered that good quality characterisation
In this approach, the behaviour of the coarse particle charge (media of the breakage behaviour of the specific rocks being used by all the
and coarse rock) is predicted using DEM and then averaged to give a breakage mechanisms is required. The VCM is only as good as both
spatially varying average velocity and porosity field within the mill. its numerical model and ore characterisation components.
Then in a second step, a transient fluid simulation is performed using
the SPH method with the solid charge represented as a dynamic porous
50
media. This model was able to elucidate the fluid mechanics of slurry
flow in a cross-section of the mill, including predicting dry regions
under the shoulder when the slurry volume was low and a slurry pool Moving liner
when it was high. This model has since been used for a tower mill 0
[85]. This has been extended to 3D and demonstrated on the Hardinge
pilot mill by Cleary et al. [110], Fig. 15. This was able to show slurry
flow within the grinding chamber, slurry flow into and through the Wear
discharge grate and into the pulp chamber. The importance of the dis- 1/100 mm -50
charge end lifters in pumping slurry and the shadowing of some grate
panels was identified. The flow dynamics in the pulp chamber, with Fixed liner
slurry levels building when a pulp chamber is near the bottom and -100
then flowing down along the pulp lifters as they rise, were demonstrated.
Finally, the model can make quantitative predictions of flow rates
through each grate panel, including capturing backflow into the mill.
Fig. 15 shows the slurry distribution within the grinding and pulp cham- -150
0 50 100 150
bers of the Hardinge mill for a slurry viscosity of 0.1 Pa s. The slurry dis-
Position on liner [mm] 0=top
tribution in the grinding chamber is similar to that of the solid part of the
charge but with a fluid shoulder that is below that of the solid charge. Fig. 16. Wear rate on fixed and moving liner. 3 mm closed side setting, after 40 min of
Complex flow patterns arise where the end wall lifters enter the charge crushing.
 Author's personal copy

18 N.S. Weerasekara et al. / Powder Technology 248 (2013) 3–24

Fig. 17. Left: Photo of surface of the moving liner. Indentation marks are clearly observed whilst no ploughing grooves can be seen. Right: Stationary liner with ploughing grooves.

7.4. Wear modelling hardening, composition of the alloy and the heat treatment. The wear
rate is for these reasons in reality far from a simple function. It has
Slightly different approaches have been taken, depending on the been demonstrated that for a specific steel alloy the wear rate can be lin-
application and appropriateness of the simplifications. Two are ear with the relative hardness of the mineral up to a given ratio where it
highlighted in this review. suddenly increases dramatically in a step change manner [123].
In an earlier work Lindquist and Evertsson [124] demonstrated
7.4.1. Crusher liners how a mechanistic model for predicting cone crusher performance
Crushers that utilize compression are commonly equipped with [125] can be combined with a modified Archard type of wear model
liners made of manganese alloyed cast steel. This group of manganese in order to predict the wear of the liners. The original wear model
steels are often referred to as Hadfield steels [121]. Due to the com- proposed by Archard [126] only accounted for sliding wear. Lindqvist
pression of either single particles or particle beds a stress field required and Evertsson introduced an additional term to the Archard model
to disintegrate the rock particles is applied. Normal and tangential con- and thus made it sensitive to pressure even if there was no relative
tact forces occur in the contact points between the particles and the motion in the contact according to Eq. (6):
crusher liners. The Hadfield type manganese alloy steel has a unique
ability to work harden when it is deformed plastically. The Vickers hard- 1
t
p
ness can typically increase from 200 to 600 HV1. The work hardening Δw ¼ ∫ pvdt þ ð6Þ
W1 0 W2
appears as a hardness profile which decreases with the depth from
the surface [122].
Wear of the crusher liners appears when rock particles are in contact Evertsson and Lindqvist demonstrated that the wear mechanisms
with the crusher liners. Depending on the condition in the contact are different for the fixed and the moving liners in a jaw crusher. The
between a rock particle and the steel surface different types of wear wear rate is significantly higher on the stationary liner compared to
modes may occur. Predicting the wear of the geometry is therefore a the moving liner as shown in Fig. 16. The worn surfaces are shown
complex task. The resulting wear rate at a given point is a result of the in Fig. 17. Ploughing grooves can clearly be seen on the fixed liner.
contact conditions, the wear resistance of the material and the mineral This is an indication that sliding occurs on the fixed liner. In contrast,
composition of the rock material. The wear resistance of the manganese on the moving liner there are no such grooves and only indentation
steel is in turn dependent on the degree and depth of the work marks can be seen.

Fig. 18. Abrasive wear distribution predicted for the: a) inside of the grinding chamber of a Hardinge pilot mill, and b) a secondary cone crusher after the passage. In each case, red is
high wear, green moderate and blue is low.
 Author's personal copy

N.S. Weerasekara et al. / Powder Technology 248 (2013) 3–24 19

Replacing the mechanistic cone crusher model with a DEM-based mills can be performed using a range of techniques including
model that can provide more detailed information about the pressure photographing particle trajectories and charge shape in laboratory
distributions and the contact conditions, the accuracy of the wear mills with perspex ends, by Positron Particle Emission Tracking (PEPT)
prediction can be expected to increase. The local pressure and the rel- and by comparing the power draw for lab, pilot and full scale mills.
ative motion are the components needed in the Archard's wear model Often the validation is treated in a superficial manner with pictorial
in order to predict the worn geometry. snapshot comparisons being presented as ‘validation’ data. Clearly,
these do not form quantitative validation. Additionally, the end-walls
7.4.2. Mill liners have a marked influence on the trajectories, presenting a view that is
The numerical methodology for calculating wear in comminution not representative of the bulk charge motion. This was analysed by
devices has been presented by Cleary [76] and used to demonstrate Powell and Nurick [131]. In order to minimise this effect Powell used
liner stress and impact and abrasion wear on a SAG mill for a 3D filming of rods in a glass-ended mill to study the motion of the grinding
slice DEM model. Fig. 18 shows examples of the wear predicted by media. Powell later used bi-planar x-ray filming to track the motion of
the DEM code of Cleary for the inside of the grinding chamber of the media within the bulk of the charge [106,131,132]. The accuracy
the Hardinge pilot mill and for wear on both concave and mantle in of this technique was extensively improved upon the continuation of
a primary gyratory crusher after one pass of ROM (run of mine) ore. this work by Govender, McBride and Powell [133] and Govender and
This approach was used to make wear predictions for pilot tower Powell [134], providing a 3D spatial accuracy of 0.3 mm at an interval
and pin mills by Cleary et al. [83]. A similar approach was used by of 50 time steps per second. This technique has been further extended
Weerasekara et al. [127] to model liner wear in SAG mills, and the cal- to utilising the power of radioactive tracing in PEPT, allowing media to
culation procedure is described in more detail in Powell, et al. [128]. A be tracked in a realistic grinding environment for equipment up to
simple pressure and relative velocity based liner wear model were 400 mm in size [135]. Steel balls, rocks and small particles can be
also employed in the work of Rezaeizadeh et al. [129]. Recently, Cleary tracked dry or in a slurry-filled environment to accuracies of 0.5 mm
et al. [130] used this approach for wear modelling in a case study of a spatial resolution. This data provides the most rigorous technique avail-
Hicom mill to predict the full liner life cycle. New and end of life liners able for detailed validation of the DEM outputs.
were used to calculate the actual wear at all points in the liner and A perspex ended centrifugal mill was used by Cleary and Hoyer
this was used to determine the best combination of wear mechanisms [136] to compare DEM predictions of the structure of the charge
to explain the observed wear. For this mill, abrasion was found to be re- and its motion for different fill levels, with good agreement obtained.
sponsible for all the observed liner wear with no observable contribu- Fig. 19 qualitatively shows the experiment and matching full 3D DEM
tion from impact. The wear model was then used to predict the entire prediction from Cleary et al. [137] for a lab scale SAG mill with a fill
life cycle of the liner including the coupled performance variation of level of 35% and a speed of 75% critical. Visually, the charge distribu-
the mill using the liner evolution approach introduced by Cleary [76]. tion and flow patterns match, with good agreement on the shoulder,
vortex centre and free surface shapes and reasonable agreement
8. Validation of DEM predictions of comminution around the toe region. The DEM shoulder angles lie in the range 45–
50° (measured anti-clockwise from the 3 pm location) and are close
Since the DEM method offers such strong advantages in modelling to the experimentally observed values of 50–56°. Similarly the toe
and understanding comminution, it is essential that both the DEM angles from the full 3D DEM simulation of 226–231° are close to the
method and the DEM codes that are used be properly and adequately experimental range of 232–238°. Further comparisons by Cleary et
validated. Validation is typically limited both by the availability of al. [137] for mill speeds of 65% and 95% and for several types of
suitable good quality data and by the preparedness of groups and DEM model showed that 2D DEM using disc particles under-predicted
software developers to take responsibility for validating their pre- the shoulder and toe positions by at least 10°. The inclusion of shape
dictions. It is important to differentiate between validation of the effects improved the shoulder predictions by 5°. Moving from 2D to
methods and of the software implementation. It is easily possible 3D slice models gave further improvement. The best match to the
for problems to occur in either area. That one DEM code is validated experiments was obtained using a full three dimensional models,
indicates that it is possible for such a prediction to be made accurately including the end walls.
and that the test code can do this. But this does not inherently mean that Dong and Moys [138], using an experimental mill with chrome
all DEM software will reproduce this level of accuracy. Each one needs to steel balls, tried to calibrate DEM predictions of ball trajectories at dif-
be independently validated. Validation of the predictions of DEM for ferent simulation conditions. This work was supported though their

Fig. 19. Comparison of a) full 3D DEM model, with b) the experiment, for a lab scale model SAG mill for a fill level of 35% (by volume) and a speed of 75% critical.
 Author's personal copy

20 N.S. Weerasekara et al. / Powder Technology 248 (2013) 3–24

coefficient of friction and coefficient of restitution measurements which they modelled the location and behaviour of the powder. This
[139]. This work was extended by validating properties of impact has revealed that the powder is preferentially located at the shoulder
between balls and walls [140]. They then tried to validate the DEM and fills back towards the toe and that powder density in the primary
predictions of mill power and load behaviour affected by two kinds of grinding region under the toe is quite low unless the powder fill level
lifter profiles over a wide range of rotational speed of the mill [141]. is at 125% or higher of the inter-ball pore space. To many operators
The same group also attempted validation by comparing experimental and users this behaviour is not intuitive and is an example of how com-
and DEM simulated forces exerted on liners using an instrumented lifter plex mill behaviour can be and how difficult it can be to optimise mill
bar [142,143]. Good agreement has been found in terms of amplitude of performance based on intuition. This has highlighted the complexity
forces and positions of shoulder and toe at low speed, and power draw of the powder behaviour and the importance of including it in the
at sub-critical speeds. DEM model. However, direct inclusion will not be possible for a signifi-
Comparisons were made with experiments for a pilot AG mill by cant time and so innovative alternative approaches are required to deal
Morrison et al. [144]. The measurements were performed using a with powder and fines.
rubber-lined mill with an inside diameter of 1.19 m, length of 0.31 m The PBM, coupled to appropriate empirical and semi-empirical
and fitted with 14 square metal lifters with height of 40 mm. The rota- models of particle capture, that is, models that allow estimating the
tional velocity of the mill was 3.14 rad/s or 77% of Critical Speed. Eight volume of material captured between colliding media, represent and
different sets of operating conditions were considered and the average important alternative, for the time being, for accounting for breakage
error in the power draw was found to be 3.1%. Power predictions have of particles that are too fine to appear explicitly in DEM simulations
also been validated for a centrifugal mill by Cleary and Hoyer [136]. [95,97,98]. Evidently, the challenge associated to non-perfectly-mixed
The prediction of power by DEM was found to be within the experimen- conditions identified in the work by Cleary and Morrison mentioned
tal range for a wide range of operating speeds. above will have to be taken into account for its realistic application.
Recently, Govendar et al. [145] have used comparisons of PEPT Another area that represents a challenge for the future is in describing
with DEM to assess the accuracy of the DEM implementation of Cleary realistically comminution in highly confined (packed bed) conditions,
[107,108] for mill modelling. A small lab scale ball mill was used with such as those found inside a high-pressure roll crusher (HPGR). This ma-
closely matched experiments and simulations. These revealed a high chine poses the additional challenge of requiring a mechanical coupling
degree of agreement with typically more than 70–80% of voxels in that will allow predicting the response of the movable rolls to the condi-
the grids of spatially averaged velocity and porosity having an effec- tions existing inside the packed bed. Although some progress has already
tive zero error. been made [153], this is an area that requires continuing efforts.
DEM predictions have also been validated in related geomechanical Most important in the future is the uptake of the DEM outputs into
applications. Holt et al. [146] used assemblies of glass beads under con- standard comminution models used in design and optimisation of
trolled laboratory experiments to validate DEM mechanical behaviour equipment and processes, as despite the progress made in under-
of rock, and then looked into fundamental mechanisms of rock de- standing and in quantitative outputs the barriers to routine uptake
formation and failure. Tan et al. [34] have modelled the dynamic process have not been surmounted. A primary barrier is the gap between
of initiation and propagation of the micro-cracks using the bonded- the detailed and complex outputs of the DEM simulations and the
particle model (BPM). Microscopic material properties for particles simplistic, or even non-existent, collision and energy profiles used
and contacts were calibrated by comparing the results of numerical ex- in the industrial semi-empirical models.
periments such as unconfined compressive test, Brazilian test, three- It is considered worthwhile to review the application of the simple
point bending test and fracture toughness test with the equivalent contact models for future DEM simulation of comminution systems,
experimental results. Tan et al. [34] showed the numerical results as the contact mechanics are inadequate for detailed analysis of the
from DEM modelling agreed well with the experimental observations. collision and breakage events. With the massive improvements in
Similar work has being carried out by Wang and Tonon [147] using computational capability the time may be ripe to take up a new level
Lac du Bonnet granite. The work by O'Sullivan et al. [147–149], Cho et of more physically meaningful contact models. As an illustration of the
al. [33,150], Cavarretta et al. [151] and Ergenzinger et al. [152] are inadequacies, the energy transmitted to a rock particle will change
some validation work on brittle rocks, breakage strength, material re- from 10 to 15% to around 30% of the applied energy depending wether
sponse, fracture and damage. Further validation in this area has been the rock is left intact or broken in a collision. Most investigators use the
conducted by Shimizu et al. [27]. convenient dashpot dissipated energy to equate to the energy available
In general the status of validation is rather patchy, with most for breakage, yet a fixed value underestimates by at least a factor of two
being superficial and at best partial, that is choosing to validate one the absorbed energy. This is clearly an area to address.
or two factors and then assuming that this validates the full suite of There is an overarching need to overcome unrealistic, indeed mis-
predictions. In order to ensure the integrity of the application of DEM leading, simplifications used in most current DEM applications. These
techniques to the science of comminution (and other areas) the quality include the use of only coarse grinding media, no slurry in wet appli-
of validation should be improved and directed at the outputs being used cations, influence of particle shape, lacking fine particles, batch in-
in the modelling. stead of continuous operation — which includes the assumption of
perfect mixing, limited ability to simulate transport through devices,
9. Future challenges for DEM modelling in comminution ignoring the influence of the powder or slurry between the colliding
elements, to mention some of the lingering issues.
One of the key challenges remaining for DEM, highlighted in a Validation is still inadequate, so there should be an on-going focus
number of places in this review, is the need to adequately take ac- of attention in this area. This includes direct validation of DEM simu-
count of the fine particles in milling applications: SAG, ball, tower lation outputs, especially in realistic environments. PEPT shows the
and stirred mills. These particles represent a reasonable volume of best promise here. Validation against controlled pilot experiments
the charge, modify the collisional and transport behaviour of the coarser to test the breakage predictions is also essential. This must include
mill charge components and are the key component about which the widely varying conditions so as to preclude parameter fitting that
user wishes to make predictions in terms of their size reduction and does not generally hold.
transport. So the question is: how to predict the effect of powder that Improved understanding of the meso-level breakage mechanisms
is and will remain for the foreseeable future, too fine to include directly is required, so as to improve the application of mechanistic breakage
in the model. Recently Cleary and Morrison have explored the effect of models that are based on DEM outputs. The work in this area is prom-
including powder in a DEM model on a small laboratory scale mill, in ising, but not yet widely utilised.
 Author's personal copy

N.S. Weerasekara et al. / Powder Technology 248 (2013) 3–24 21

The application to comminution is now ripe for the inclusion of m p;i Average weight of the particles in the bed
mineral liberation – the separation of bonded components in a rock – Ncap,i Number of particles caught between the media elements
which is the goal of comminution in mineral processing. p Pressure
p(E) Distribution of stressing energies E
10. Conclusions p(e) Function that represents the energy split amongst particles
R⁎ Effective radius
DEM now plays an important role in understanding comminution V Volume loss
fundamentals and in the design and optimisation of comminution v Velocity
equipment. Over the past three decades model sizes have increased w Wear
by several orders of magnitudes and can be expected to increase at wi(t) Mass fraction of particles contained in size class i
that same exponential rate. This provides an unprecedented and grow- W1, W2 Wear resistance coefficient
ing ability to look deep inside comminution equipment and understand ΔHel Height loss in a selected element
the breakage mechanics, liberation, transport and classification. ΔVi Wear track volume
One of the key successes of DEM in the last decade has been reve- δ Dirac delta function
lations of the nature of the collision energy spectra and their domi- δ Overlap
nance by vast numbers of weak collisions. This clearly demonstrated ε Coefficient of restitution
that the impact of cataracting material on the toe of a SAG mill was κi first-order surface breakage rate
a small contributor to particle size reduction and that the dominant μ Coefficient of static friction
mechanisms were of low energy progressive damage. This has led ω Frequency of stressing events in the comminution machine
to development of the concept of incremental breakage, where a par-
ticle fails by accumulation of damage over many weak collisions. This Subscripts
has been experimentally verified and shown to consume more energy n Normal components
than the same degree of breakage produced by a single high energy t Tangential components
impact. This is in part why mills are inherently energy inefficient.
These insights have also contributed to motivating the development Superscripts
of new generations of ore breakage characterisation, including inves- b Body breakage mechanisms
tigation of much lower energies and of repeated collisions. el Element
DEM is now able to predict the nature of the particle flows for i Contact event
coarse and mid-sized particles within most comminution machines. rel Relative
It can predict and give insight into the nature of how the energy sup- s Surface breakage mechanisms
plied by the machine is used for particle breakage. Liner stress and
wear can be predicted as can the evolution of the liner and the change
in performance over the liner life cycle. However, there are still many Acknowledgements
questions to be resolved about how to relate specific wear patterns to
operating conditions. Transport and classification can also be studied Portions of this work were carried out under the auspices and
effectively when the finer particulate material can be included in the with partial financial support of the Centre for Sustainable Resource
DEM models. Major challenges though remain in relation to the inclu- Processing, which is established and supported under the Australian
sion of fine particles and slurry in mills. For crushers, the emergence Government's Cooperative Research Centres Program.
of effective dynamic particle breakage models is now making simula-
tion of crusher performance viable. In all comminution areas, valida-
tion of both the underlying DEM method and of the DEM software References
being used is an on-going requirement. Although there is a useful
[1] F.C. Bond, The third theory of comminution, Transactions of AIME 193 (1952)
amount of validation that encourages the view that DEM predictions 484–494.
are reasonable, there is still much that should be tested. [2] C.A. Rowlands, D.M. Kjos, Rod and ball mills, in: A.L. Mular, R.B. Bhappu (Eds.),
Mineral Processing Plant Design AIME, 1980, pp. 239–278.
In closing, it may be said that the DEM and accompanying computa-
[3] L.G. Austin, R.R. Klimpel, P.T. Luckie, Process Engineering of Size Reduction: Ball
tional techniques are providing an essential contribution to elevating Milling, Society Mining Engineers/AIME Pub, 1984.
the study of the comminution process from a semi-empirical basis to an [4] L.G. Austin, P.T. Luckie, Methods for determination of breakage distribution
engineering science, based on the underlying mechanics of the processes. parameters, Powder Technology 5 (1972) 215–222.
[5] J.A. Herbst, D.W. Fuerstenau, Scale-up procedure for continuous grinding
mill design using population balance models, International Journal of Mineral
Processing 7 (1980) 1–31.
Nomenclature [6] W.J. Whiten, A matrix theory of comminution machines, Chemical Engineering
Science 29 (1974) 589–599.
A Rate of appearance of material in class i due to fracture [7] T.J. Napier-Munn, S. Morrell, R.D. Morrison, T. Kojovic, Mineral Comminution
Ael Area of a liner element Circuits: Their Operation and Optimisation, JKMRC, Queensland, Australia, 1996.
aij Surface breakage density functions [8] P.J. Mackey, J.E. Nessett, The impact of commissioning and start-up performance
on a mining/metallurgical project, 35th Annual Meeting of the Canadian Mineral
bij Body breakage density functions Processors, Ottawa, 2003.
D Rate of disappearance of material in class i due to fracture [9] SAG11, in: K. Major, B. Flintoff, B. Klein, K. McLeod (Eds.), Proceedings of the
d Wear track width International Autogenous and Semiautogenous Grinding Technology 2011,
E⁎ Effective Young's modulus CIM, 2011.
[10] SAG06, in: Mular, et al., (Eds.), Proceedings of the International Autogenous and
F Force Semiautogenous Grinding Technology 2006, CIM, 2006.
G⁎ Effective Shear modulus [11] M.S. Powell, R.D. Morrison, The future of comminution modelling, International
Journal of Mineral Processing 84 (2007) 228–239.
HB Material hardness
[12] R.D. Mindlin, Compliance of elastic bodies in contact, Journal of Applied
h Wear track depth Mechanics 16 (1949) 259–268.
M Powder load [13] P.A. Cundall, O.D.L. Strack, A discrete numerical model for granular assemblies,
m⁎ Effective mass Geotechnique 29 (1979) 47–65.
[14] EDEM, EDEM user manual, 2006.
mj Mass of particles contained in size class j captured in each [15] PFC3D, Particle Flow Code in 3 Dimension, Ver 2.0 User's Manual, Minneapolis,
stressing event 1999.
 Author's personal copy

22 N.S. Weerasekara et al. / Powder Technology 248 (2013) 3–24

[16] F.P. Di Maio, A. Di Renzo, Modelling particle contacts in distinct element simulations: [45] M. Pierce, D.M. Ivars, P. Cundall, D. Potyondy, A synthetic rock mass model for
linear and non-linear approach, Chemical Engineering Research and Design 83 jointed rock, in: E. Eberhardt (Ed.), 1st Canada–US Rock Mechanics Symposium,
(2005) 1287–1297. Taylor & Francis, London Vancouver, 2007.
[17] P.W. Cleary, Predicting charge motion, power draw, segregation and wear in [46] P.W. Cleary, M. Prakash, M.D. Sinnott, M. Rudman, R. Das, Large scale simulation
ball mills using Discrete Element Methods, Minerals Engineering 11 (1998) of industrial, engineering and geophysical flows using particle methods, in: E.
1061–1080. Oñate, R. Owen (Eds.), Particle-Based Methods, Computational Methods, Springer
[18] A. Di Renzo, F.P. Di Maio, Comparison of contact-force models for the simulation Science + Business Media B.V. in Applied Sciences, 2011, pp. 89–111.
of collisions in DEM-based granular flow codes, Chemical Engineering Science [47] G.W. Delaney, P.W. Cleary, M.D. Sinnott, R.D. Morrison, Novel application of
59 (2004) 525–541. DEM to modelling comminution processes, 9th World Congress on Computa-
[19] C. Thornton, S.J. Cummins, P.W. Cleary, An investigation of the comparative tional Mechanics Sydney, Australia, 2010.
behaviour of alternative contact force models during elastic collisions, Powder [48] M.S. Powell, R.D. Morrison, A new look at breakage testing, 11th European
Technology 210 (2011) 189–197. Symposium on Comminution, European Federation of Chemical Engineers,
[20] C. Thornton, S.J. Cummins, P.W. Cleary, An investigation of the comparative be- Budapest, Hungary, 2006, p. 6.
haviour of alternative contact force models during inelastic collisions, Powder [49] M.S. Powell, I. Govender, A.T. McBride, Challenges in applying the unified com-
Technology 233 (2013) 30–46. minution model, in: W.D. Zou, et al., (Eds.), XXIV IMPC Beijing, China, 2008,
[21] H.P. Zhu, Z.Y. Zhou, R.Y. Yang, A.B. Yu, Discrete particle simulation of particulate pp. 367–376.
systems: theoretical developments, Chemical Engineering Science 62 (2007) [50] L.M. Tavares, L.G. Austin, R.P. King, Breakage and damage of particles by impact, in:
3378–3396. S.K. Kawatra (Ed.), Advances in Comminution, SM, Colorado, USA, 2006,
[22] P.A. Cundall, Formulation of a three-dimensional distinct element mode — part pp. 205–223.
I: a scheme to detect and represent contacts in a system composed of many [51] S.S. Narayanan, W.J. Whiten, Breakage Characteristics for Ores for Ball Mill
polyhedral blocks, International Journal of Rock Mechanics and Mining Sciences Modelling, AusIMM, 1983. 31–39.
and Geomechanics Abstracts 25 (1988) 107–116. [52] L.M. Tavares, Chapter 1 Breakage of single particles: quasi-static, in: A.D. Salman,
[23] K. Meguro, K. Iwashita, M. Hakuno, Fracture analyses of media composed of ir- M.J. Hounslow, M. Ghadiri (Eds.), Handbook of Powder Technology, Elsevier
regularly shaped regions by the extended distinct element method, Structural Science B.V., 2007, pp. 3–68.
Engineering/Earthquake Engineering 8 (1991) 131–142. [53] L.M. Tavares, Optimum routes for particle breakage by impact, Powder Technology
[24] H. Morikawa, Y. Sawamoto, N. Kobayashi, Local fracture analysis of a reinforced 142 (2004) 81–91.
concrete slab by the discrete element method, 2nd Int. Conf. Discrete Element [54] L. Vogel, W. Peukert, Breakage behaviour of different materials — construction of
Methods, IESL Publications, Cambridge Massachusetts, 1993. a mastercurve for the breakage probability, Powder Technology 129 (2003)
[25] D.O. Potyondy, P.A. Cundall, C.A. Lee, Modelling rock using bonded assemblies of 101–110.
circular particle, in: H.M. Aubertin (Ed.), Rock Mechamics, Balkema, Rotterdam, [55] F. Shi, T. Kojovic, Modelling of particle size effect on breakage and its application
1996. in AG/SAG simulations, Comminution 2008 Conference,Falmouth, UK, 2008.
[26] D.O. Potyondy, P.A. Cundall, A bonded-particle model for rock, International [56] L.M. Tavares, Analysis of particle fracture by repeated stressing as damage accu-
Journal of Rock Mechanics and Mining Sciences 41 (2004) 1329–1364. mulation, Powder Technology 190 (2009) 327–339.
[27] Y. Shimizu, R.D. Hart, P.A. Cundall, Numerical modeling in micromechanics via [57] M. Powell, R. Morrison, N. Djordjevic, M. Hilden, P. Cleary, P. Owen, I. Govender,
particle methods — 2004, 2nd International PFC Symposium, Taylor & Francis, N. Weerasekara, S. Michaux, T. Kojovic, Z. Pokrajcic, F. Musa, M. Sinnott, A.
Kyoto, Japan, 2004. Mainza, L. Bbosa, Eco-efficient liberation outcomes and benefits, CSRP final
[28] N.S. Weerasekara, M.S. Powell, S. Cole, J. Favier, Modelling breakage environ- report, CRC for Sustainable Resource Processing, 2010.
ment in tumbling mills using DEM and analyzing the outputs, The Fifth Interna- [58] S. Larbi-Bram, A Study of Ore Breakage Characterization for AG/SAG Mill Modelling,
tional Conference on Discrete Element Methods, Queen Mary, University of Julius Kruttschnitt Mineral Research Centre, Sustainable Minerals Institute, The
London, London, UK, 2010. University of Queensland, Australia, 2009.
[29] P.W. Cleary, Recent advances in DEM modelling of tumbling mills, Minerals [59] P.W. Cleary, M.S. Powell, R.D. Morrison, Applying computationally intensive
Engineering 14 (2001) 1295–1319. techniques to modelling of comminution devices, in: D.Z. Wang, C.Y. Sun, F.L.
[30] N.S. Weerasekara, M.S. Powell, The New energy logging from the discrete element Wang, L.C. Zhang, L. Han (Eds.), XXIV International Mineral Processing Congress.
method, CSRP'08 2nd Annual Conference, Brisbane, Queensland, Australia, 2008. XXIV IMPC, Beijing, China, 2008, pp. 1955–1963.
[31] N.S. Weerasekara, M.S. Powell, Exploring the breakage environment in mills [60] M.S. Powell, I. Govender, A.T. McBride, Applying DEM outputs to the unified
with DEM, XXV International Mineral Processing Congress, Brisbane Australia, comminution model, Minerals Engineering 21 (2008) 744–750.
2010, pp. 1357–1366. [61] M.S. Powell, N. Weerasekara, Challenges in developing a mechanistic breakage
[32] M.S. Powell, N.S. Weerasekara, Building the unified comminution model, model such as the UCM, European Symposium on Comminution and
XXV International Mineral Processing Congress Brisbane Australia, 2010, classificationEspoo Finland, 2009, p. 20.
pp. 1133–1142. [62] G.K.P. Barrios, R.M. de Carvalho, L.M. Tavares, Extending breakage characterisation
[33] N. Cho, C.D. Martin, D.C. Sego, A clumped particle model for rock, International to fine sizes by impact on particle beds, Mineral Processing and Extractive Metal-
Journal of Rock Mechanics and Mining Sciences 44 (2007) 997–1010. lurgy 120 (2011) 37–44.
[34] Y. Tan, D. Yang, Y. Sheng, Discrete element method (DEM) modeling of fracture [63] M. Marktscheffel, K. Schönert, Liberation of composite particles by single parti-
and damage in the machining process of polycrystalline SiC, Journal of the Euro- cle compression, shear and impact loading, 6th European Symposium on
pean Ceramic Society 29 (2009) 1029–1037. ComminutionNuremberg, Germany, 1986.
[35] C. Thornton, S.J. Antony, Quasi-static shear deformation of a soft particle system, [64] N. Djordjevic, F.N. Shi, R.D. Morrison, Applying discrete element modelling to
Powder Technology 109 (2000) 179–191. vertical and horizontal shaft impact crushers, Minerals Engineering 16 (2003)
[36] K.D. Kafui, C. Thornton, Numerical simulations of impact breakage of a spherical 983–991.
crystalline agglomerate, Powder Technology 109 (2000) 113–132. [65] W. Schubert, H. Jeschke, DEM-simulation of the Breakage Process in an Impact
[37] C. Thornton, L. Liu, How do agglomerates break? Powder Technology 143–144 Crusher, New Orders of the Comminution, Otto-von-Guericke-universität,
(2004) 110–116. Magdeburg, 2005.
[38] C. Thornton, M.T. Ciomocos, M.J. Adams, Numerical simulations of diametri- [66] E.R. Cunha, R.M. Carvalho, L.M. Tavares, Simulation of solids flow and energy
cal compression tests on agglomerates, Powder Technology 140 (2004) transfer in a vertical shaft impact crusher using DEM, Minerals Engineering
258–267. 43–44 (2013) 85–90.
[39] M.D. Ivars, M.E. Pierce, C. Darcel, J. Reyes-Montes, D.O. Potyondy, R. Paul Young, [67] M. Khanal, Simulation of Crushing Dynamics of an Aggregate-Matrix Composite
P.A. Cundall, The synthetic rock mass approach for jointed rock mass modelling, by Compression and Impact Stressings, Process and System Engineering,
International Journal of Rock Mechanics and Mining Sciences 48 (2011) Otto-von-Guericke University, Magdeburg, 2005.
219–244. [68] J. Lichter, K. Lim, A. Potapov, D. Kaja, New developments in cone crusher perfor-
[40] D.M. Ivars, D.O. Potyondy, M. Pierce, P.A. Cundall, The smooth-joint contact mance optimization, Minerals Engineering 22 (2009) 613–617.
model, 8th World Congress on Computational Mechanics/5th European Con- [69] P.W. Cleary, Ball motion, axial segregation and power consumption in a full scale
gress on Computational Mechanics and Applied Science and Engineering Venice, two chamber cement mill, Minerals Engineering 22 (2009) 809–820.
2008, p. a2735. [70] J.C.E. Quist, Cone Crusher Modelling and Simulation, Product and Production
[41] D.M. Ivars, Bonded Particle Model for Jointed Rock Mass, Royal Institute of Tech- Development, Chalmers University of Technology, Göteborg, 2012.
nology (KTH) Stockholm, Sweden, 2010. [71] J.C.E. Quist, C.M. Evertsson, J. Franke, The effect of liner wear on gyratory
[42] D.M. Ivars, N. Deisman, M. Pierce, C. Fairhurst, The synthetic rock mass approach — crushing — a DEM case study, Computational Modeling 11'Falmouth, 2011.
a step forward in the characterization of jointed rock masses, in: R.S. Lisbon, G. [72] R. Morrison, P.W. Cleary, B. Loveday, G.W. Delaney, S. Cummins, Predicting the
Olalla (Eds.), 11th Congress on International Society of Rock Mechanics, Taylor & evolution of rock size and shape distributions using DEM based on different
Francis, London, 2007, pp. 485–490. modes of breakage in AG and SAG mills, in: Flintoff, et al., (Eds.), International
[43] D.M. Ivars, M. Pierce, D. DeGagné, C. Darcel, Anisotropy and scale dependency autogenous and semiautogenous grinding technology 2011, SAG11CIM, Canada,
in jointed rock-mass strength — a synthetic rock mass study, in: R. Hart, C. 2011.
Detournay, P. Cundall (Eds.), 1st International FLAC/DEM Symposium [73] L.M. Tavares, R.M. de Carvalho, Modeling breakage rates of coarse particles in
on Numerical Modeling, Itasca Consulting Group, Minneapolis, 2008, ball mills, Minerals Engineering 22 (2009) 650–659.
pp. 231–239. [74] B.K. Mishra, R.K. Rajamani, The discrete element method for the simulation of
[44] H. Konietzky, Numerical Modeling in Micromechanics via Particle Methods 1st ball mills, Applied Mathematical Modelling 16 (1992) 598–604.
International PFC Symposium, Swets & Zeitlinger B.V., Gelsenkirchen, Germany, [75] B.K. Mishra, R.K. Rajamani, Simulation of charge motion in ball mills. Part 1: experi-
2003. mental verifications, International Journal of Mineral Processing 40 (1994) 171–186.
 Author's personal copy

N.S. Weerasekara et al. / Powder Technology 248 (2013) 3–24 23

[76] P.W. Cleary, Charge behaviour and power consumption in ball mills: sensitivity [109] P.W. Cleary, R.D. Morrison, Understanding fine ore breakage in a laboratory scale
to mill operating conditions, liner geometry and charge composition, Interna- ball mill using DEM, Minerals Engineering 24 (2011) 352–366.
tional Journal of Mineral Processing 63 (2001). [110] P.W. Cleary, R.D. Morrison, Prediction of 3D slurry flow within the grinding
[77] P.W. Cleary, M.L. Sawley, DEM modelling of industrial granular flows: 3D case chamber and discharge from a pilot scale SAG mill, Minerals Engineering 39
studies and the effect of particle shape on hopper discharge, Applied Mathematical (2012) 184–195.
Modelling 26 (2002). [111] P.W. Cleary, M. Prakash, J. Ha, N. Stokes, C. Scott, Smooth particle hydrodynamics;
[78] B.K. Mishra, A review of computer simulation of tumbling mills by the discrete status and future potential, Progress in Computational Fluid Dynamics 7 (2007)
element method: Part I — contact mechanics, International Journal of Mineral 70–90.
Processing 71 (2003) 73–93. [112] P.W. Cleary, R.D. Morrison, Using DEM to model ore breakage within a pilot scale
[79] B.K. Mishra, A review of computer simulation of tumbling mills by the discrete SAG mill, Minerals Engineering 17 (2004) 1117–1124.
element method: Part II — practical applications, International Journal of Mineral [113] N. Djordjevic, R. Morrison, B. Loveday, P. Cleary, Modelling comminution patterns
Processing 71 (2003) 95–112. within a pilot scale AG/SAG mill, Minerals Engineering 19 (2006) 1505–1516.
[80] P.W. Cleary, Axial transport in dry ball mills, Applied Mathematical Modelling 30 [114] R.D. Morrison, F. Shi, R. Whyte, Modelling of incremental rock breakage by
(2006) 1343–1355. impact — for use in DEM models, Minerals Engineering 20 (2007) 303–309.
[81] P.W. Cleary, M.D. Sinnott, R.D. Morrison, DEM prediction of particle flows in [115] R. Whyte, Measuring incremental damage in rock breakage by impact, The
grinding processes, International Journal for Numerical Methods in Fluids 58 University of Queensland, 2005.
(2008). [116] R.D. Morrison, P.W. Cleary, Towards a virtual comminution machine, Minerals
[82] M. Sinnott, P.W. Cleary, R. Morrison, Analysis of stirred mill performance using Engineering 21 (2008) 770–781.
DEM simulation: Part 1 — media motion, energy consumption and collisional [117] P.W. Cleary, R.D. Morrison, Particle methods for modelling in mineral processing,
environment, Minerals Engineering 19 (2006) 1537–1550. International Journal Computational Fluid Dynamics 23 (2009) 137–146.
[83] P.W. Cleary, M. Sinnott, R. Morrison, Analysis of stirred mill performance using [118] R.D. Morrison, B. Loveday, M. Powell, N. Djordjevic, P.W. Cleary, Applying
DEM simulation: Part 2 — coherent flow structures, liner stress and wear, Discrete Element Modelling to different modes of breakage in AG and SAG
mixing and transport, Minerals Engineering 19 (2006) 1551–1572. mills, in: Mular, et al., (Eds.), International autogenous and semiautogenous
[84] R.D. Morrison, P.W. Cleary, M.D. Sinnott, Using DEM to compare the energy grinding technology, SAG 2006, CIM, 2006.
efficiency of pilot scale ball and tower mills, Minerals Engineering 22 (2009) [119] J.J. Monaghan, Simulating free surface flows with SPH, Journal of Computational
665–672. Physics 110 (1994) 399–406.
[85] M. Sinnott, P.W. Cleary, R.D. Morrison, Slurry flow in a tower mill, Minerals [120] P.W. Cleary, R.D. Morrison, Predicting patterns of slurry flow within and
Engineering 24 (2011) 152–159. discharge from a 3D pilot SAG mill, in: K. Major, B. Flintoff, B. Klein, K. McLeod
[86] P.W. Cleary, M. Sinnott, R. Morrison, Prediction of slurry transport in SAG mills (Eds.), SAG2011 Conference,Vancouver, BC, Canada, 2011.
using SPH fluid flow in a dynamic DEM based porous media, Minerals Engineering [121] Y.N. Dastur, W.C. Leslie, Mechanism of work hardening in Hadfield manganese
19 (2006) 1517–1527. steel, Metallurgical Transactions A 12 A (1981).
[87] M.D. Sinnott, P.W. Cleary, R.D. Morrison, Is media shape important for grinding [122] M. Lindqvist, P. Sotkovszki, Work Hardening in Cone Crusher Liners, CAMI,
performance in stirred mills? Minerals Engineering 24 (2011) 138–151. Calgary, Canada, 2003.
[88] R.Y. Yang, C.T. Jayasundara, A.B. Yu, D. Curry, DEM simulation of the flow of [123] SSAB, Software for prediction of wear rate for Hardox steel, WearCalc 2.0, 2012.
grinding media in IsaMill, Minerals Engineering 19 (2006) 984–994. [124] M. Lindqvist, C.M. Evertsson, Liner wear in jaw crushers, Minerals Engineering
[89] C.T. Jayasundara, R.Y. Yang, A.B. Yu, D. Curry, Discrete particle simulation of par- 16 (2003) 1–12.
ticle flow in IsaMill, Industrial and Engineering Chemistry Research 19 (2006) [125] C.M. Evertsson, Cone Crusher Performance, Dept. of Machine and Vehicle
6349–6359. Design, Chalmers University of Technology, Sweden, 2000.
[90] C.T. Jayasundara, R.Y. Yang, B.Y. Guo, A.B. Yu, J. Rubenstein, Effect of slurry prop- [126] J.F. Archard, Contact and rubbing of flat surfaces, Journal of Applied Physics 24
erties on particle motion in IsaMills, Minerals Engineering 22 (2009) 886–892. (1953).
[91] C.T. Jayasundara, R.Y. Yang, A.B. Yu, D. Curry, Discrete particle simulation of particle [127] N.S. Weerasekara, M. Powell, J. Franke, J. Favier, Integrating liner profile evolu-
flow in IsaMill — effect of grinding medium properties, Chemical Engineering tion and mill performance with 3D DEM modelling, in: K. Major, B. Flintoff, B.
Journal 135 (2008) 103–112. Klein, K. McLeod (Eds.), SAG 2011, Vancouver, Canada, 2011, pp. 1–15.
[92] C.T. Jayasundara, R.Y. Yang, A.B. Yu, J. Rubenstein, Effects of disc rotation speed [128] M.S. Powell, N.S. Weerasekara, S. Cole, R.D. LaRoche, J. Favier, DEM modelling of
and media loading on particle flow and grinding performance in a horizontal liner evolution and its influence on grinding rate in ball mills, Minerals Engi-
stirred mill, International Journal of Mineral Processing 96 (2010) 27–35. neering 24 (2011) 341–351.
[93] C.T. Jayasundara, R.Y. Yang, A.B. Yu, D. Curry, Prediction of the disc wear in a [129] M. Rezaeizadeh, M. Fooladi, M.S. Powell, S.H. Mansouri, N.S. Weerasekara, A new
model IsaMill and its effect on the flow of grinding media, Minerals Engineering predictive model of lifter bar wear in mills, Minerals Engineering 23 (2010)
24 (2011) 1586–1594. 1174–1181.
[94] R.P. King, F. Bourgeois, A new conceptual model for ball milling, in Proc. XVIII [130] P.W. Cleary, P.J. Owen, D.I. Hoyer, S. Marshall, Applying computationally inten-
International Mineral Processing Congress, vol. 1, (AUSIMM), 1993, pp. 81–86. sive techniques to modelling of comminution devices, International Journal for
[95] R.M. Carvalho, L.M. Tavares, Dynamic modeling of comminution using a general Numerical Methods in Engineering 81 (2010) 1157–1179.
microscale breakage model, Computer Aided Chemical Engineering 27 (2009) [131] M.S. Powell, G.N. Nurick, A study of charge motion in rotary mills Part 1 — extension
519–524. of the theory, Minerals Engineering 9 (1996) 259–268.
[96] G.K.P. Barrios, R.M. de Carvalho, L.M. Tavares, Modeling breakage [132] M.S. Powell, G.N. Nurick, A study of charge motion in rotary mills part 3 — analysis
ofmonodispersed particles in unconfined beds, Minerals Engineering 24 of results, Minerals Engineering 9 (1996) 399–418.
(2011) 308–318. [133] I. Govender, A. McBride, M.S. Powell, Improved experimental tracking tech-
[97] R.M. Carvalho, L.M. Tavares, Leaping forward in SAG and AG Mill simulation niques for validating discrete element method simulations of tumbling mills,
using a mechanistic model framework, in: K. Major, B. Flintoff, B. Klein, K. Journal of Experimental Mechanics 44 (2004) 593–607.
McLeod (Eds.), SAG2011 Conference, Vancouver, BC, Canada, 2011. [134] I. Govender, M.S. Powell, A 3D X-ray vision system for studying discrete partic-
[98] R.M. Carvalho, L.M. Tavares, Predicting the effect of operating and design variables ulate behaviour, Discrete Element Methods 07Brisbane, Australia, 2007.
on breakage rates using the mechanistic ball mill model, Minerals Engineering [135] I. Govender, N. Mangesana, A.N. Mainza, J.P. Franzidis, Measurement of shear
43–44 (2012) 91–101. rates in a laboratory tumbling mill, Minerals Engineering 24 (2011) 225–229.
[99] S.S. Narayanan, W.J. Whiten, Determination of comminution characteristics [136] P.W. Cleary, D. Hoyer, Centrifugal mill charge motion and power draw: comparison
from single particle breakage tests and its application to ball mill scale-up, of DEM predictions with experiment, International Journal of Mineral Processing
Transactions of the Institution of Mining and Metallurgy 97 (1988) C115–C124. 59 (2000) 131–148.
[100] L.M. Tavares, Energy absorbed in breakage of single particles in drop weight [137] P.W. Cleary, R.D. Morrison, S. Morrell, Comparison of DEM and experiment for a
testing, Minerals Engineering 12 (1999) 43–50. scale model SAG mill, International Journal of Mineral Processing 68 (2003)
[101] L.M. Tavares, R.M. Carvalho, A mechanistic model of batch grinding in ball mills, 129–165.
XXV International Mineral Processing Congress Brisbane, Australia, 2010. [138] H. Dong, M.H. Moys, Assessment of discrete element method for one ball
[102] L.M. Tavares, R.M. de Carvalho, Modeling ore degradation during handling using bouncing in a grinding mill, International Journal of Mineral Processing 65
continuum damage mechanics, International Journal of Mineral Processing 101 (2002) 213–226.
(2011) 21–27. [139] H. Dong, M.H. Moys, A technique to measure velocities of a ball moving in a tum-
[103] M.S. Powell, The Unified Comminution Model — a Conceptually New Model, bling mill and its applications, Minerals Engineering 14 (2001) 841–850.
XXIII IMPC, Istanbul, Turkey, 2006, pp. 1783–1788. [140] H. Dong, M.H. Moys, Measurement of impact behaviour between balls and walls
[104] M.S. Powell, A.T. McBride, I. Govender, Application of DEM Outputs to Refining in grinding mills, Minerals Engineering 16 (2003) 543–550.
Applied SAG Mill Models, IMPC 2003, Cape Town. South Africa, 2003. 307–316. [141] O. Hlungwani, J. Rikhotso, H. Dong, M.H. Moys, Further validation of DEM
[105] I. Govender, M.S. Powell, Investigating SAG model assumptions using laboratory modeling of milling: effects of liner profile and mill speed, Minerals Engineering
3D trajectory data, International Autogenous and Semiautogenous Grinding 16 (2003) 993–998.
Technology 2006, SAG06, CIM, 2006. 436–450. [142] J.T. Kalala, M. Bwalya, M.H. Moys, Discrete element method (DEM) modelling of
[106] M.S. Powell, G.N. Nurick, A study of charge motion in rotary mills part 2 — evolving mill liner profiles due to wear. Part II. Industrial case study, Minerals
experimental work, Minerals Engineering 9 (1996) 343–350. Engineering 18 (2005) 1392–1397.
[107] P.W. Cleary, Large scale industrial DEM modelling, Engineering Computations 21 [143] A.B. Makokha, M.H. Moys, M.M. Bwalya, K. Kimera, A new approach to
(2004) 169–204. optimising the life and performance of worn liners in ball mills: experimental
[108] P.W. Cleary, Industrial particle flow modelling using DEM, Engineering Compu- study and DEM simulation, International Journal of Mineral Processing 84
tations 26 (2009) 698–743. (2007) 221–227.
 Author's personal copy

24 N.S. Weerasekara et al. / Powder Technology 248 (2013) 3–24

[144] R. Morrison, B. Loveday, N. Djordjevic, P.W. Cleary, P. Owen, Linking discrete el- [149] C. O'Sullivan, L. Cui, Micromechanics of granular material response during load
ement modelling to breakage in a pilot scale AG/SAG mill, in: Kawatra (Ed.), Ad- reversals: combined DEM and experimental study, Powder Technology 193
vances in Comminution, SME2006, 2006, pp. 495–512. (2009) 289–302.
[145] I. Govender, P.W. Cleary, A.N. Mainza, Comparisons of PEPT derived charge features [150] N. Cho, C.D. Martin, D.C. Sego, Development of a shear zone in brittle rock
in wet milling environments with a friction-adjusted DEM mode, Chemical Engi- subjected to direct shear, International Journal of Rock Mechanics and Mining
neering Science 97 (2013) 162–175. Sciences 45 (2008) 1335–1346.
[146] R.M. Holt, J. Kjølaas, I. Larsen, L. Li, A. Gotusso Pillitteri, E.F. Sønstebø, Comparison [151] I. Cavarretta, C. O'Sullivan, E. Ibraim, M. Lings, S. Hamlin, D.M. Wood, Character-
between controlled laboratory experiments and discrete particle simulations of ization of artificial spherical particles for DEM validation studies, Particuology
the mechanical behaviour of rock, International Journal of Rock Mechanics and 10 (2012) 209–220.
Mining Sciences 42 (2005) 985–995. [152] C. Ergenzinger, R. Seifried, P. Eberhard, A discrete element model predicting the
[147] Y. Wang, F. Tonon, Modeling Lac du Bonnet granite using a discrete element strength of ballast stones, Computers and Structures 108–109 (2012) 3–13.
model, International Journal of Rock Mechanics and Mining Sciences 46 (2009) [153] J.A. Herbst, M.A. Mular, W.T. Pate, X. Qui, Detailed modeling of an HPGR/HRC for
1124–1135. prediction of plant scale unit performance, in: K. Major, B. Flintoff, B. Klein, K.
[148] C. O'Sullivan, J. Bray, M. Riemer, Examination of the response of regularly packed McLeod (Eds.), SAG2011 Conference,Vancouver, BC, Canada, 2011.
specimens of spherical particles using physical tests and discrete element simu-
lations, Journal of Engineering Mechanics 130 (2004) 1140–1150.