Minerals Engineering 21 (2008) 770781

Contents lists available at ScienceDirect

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

Towards a virtual comminution machine

R.D. Morrison a, P.W. Cleary b,*
a b

JKMRC, The University of Queensland, Isles Road, Indooroopilly, Queensland 4068, Australia CSIRO Mathematical and Information Sciences, Private Bag 33, Clayton South, Victoria 3168, Australia

a r t i c l e

i n f o

a b s t r a c t
Towards the end of the 1990s readily available personal computers became sufciently powerful when combined with an efcient numerical code to use discrete element modelling (DEM) in two dimensions for models involving a few hundred to a few thousand particles in commercially available packages. Some proprietary codes reported up to 200,000 particles [Herbst, J.A., Nordell, L., 2001. Optimization of the design of sag mill internals using high delity simulation. In: Vancouver, B.C., Barratt, D.J., Allan, M.J., Mular, A.L. (Eds.), Proceedings of the SAG Conference, University of British Columbia, IV, 150164; Cleary, P.W., 2001a. Charge behaviour and power consumption in ball mills: Sensitivity to mill operating conditions, liner geometry and charge composition. Int. J. Min. Process. 63, 79114 and Cleary, P.W., 2001b. Recent advances in DEM modelling of tumbling mills. Minerals Eng. 14, 12951319]. In early 2000, JKMRC and CSIRO-MIS agreed to an informal collaboration with the objective of testing various DEM approaches against detailed process measurements. The initial collaboration demonstrated that 3D-DEM using spheres was sufciently realistic for ow patterns and power estimation within tumbling mills. The results were reported in papers which were presented at SAG 2001 and in the technical literature [Morrison, R.D., Cleary, P.W., Valery, W., 2001. Comparing power and performance trends from DEM and JK modelling. SAG 2001, Department of Mining and Minerals Process Engineering. University of British Columbia, Vancouver, 284300; Cleary, P.W., Morrison, R., Morrell, S., 2003. Comparison of DEM and experiment for a scale model SAG mill. Int. J. Min. Process. 68, 129165]. The commencement of the CRC for Sustainable Resource Processing in 2003 provided an opportunity to formalize the collaboration and bring increased resources to bear. The objective of this collaboration is to develop a virtual comminution machine (VCM). The VCM will allow a comminution machine design which exists as a suitably detailed design in a 3D Computer aided design le (CAD) to simulate processing an ore (which has been characterised by suitable test work) to predict progeny, power consumption, wear and even machine component loadings. This paper reports on the substantial progress made to date towards a practical Virtual Comminution Machine. 2008 Published by Elsevier Ltd.

Article history: Received 30 December 2007 Accepted 5 June 2008 Available online 15 August 2008 Keywords: Comminution Discrete element modelling Grinding SAG milling

1. Introduction 1.1. Preamble A new comminution device typically requires 1525 years for development and to gain a sufcient degree of industrial acceptance to have any chance of commercial survival. Unsurprisingly, many promising devices suffer from interruptions to their development schedule or fail altogether for lack of development nance as much as any technical shortcomings. There are also broader reasons for seeking more efcient ways to carry out comminution. It has been estimated that comminution
* Corresponding author. Tel.: +61 3 9545 8005; fax: +61 3 9545 8080. E-mail addresses: R.Morrison@uq.edu.au (R.D. Morrison), Paul.Cleary@csiro.au (P.W. Cleary). 0892-6875/$ - see front matter 2008 Published by Elsevier Ltd. doi:10.1016/j.mineng.2008.06.005

processes world wide consume about 3% of all electricity generated (Schoenert, 1986; La Nauze and Temos, 2002; Fuerstenau and Abouzeid, 2002). This compares with 1.3% by ore milling in North America (DOE report, 1981). Consumption of grinding media and wear resistant liners consumes about the same order of energy in terms of green house gas production (Musa and Morrison, 2007). Hence the total energy equivalent is more like 6%. In Australian terms, 6% and 12%, respectively are reasonable estimates given the intensity of resource production. The current substantial pressure to reduce greenhouse gas production therefore provides a strong incentive to devise more energy efcient comminution devices, and to reduce the calendar time required for their development. The most serious obstacle in the traditional approach to developing a new comminution device is the time and cost of building successive prototypes. Each prototype represents a substantial

R.D. Morrison, P.W. Cleary / Minerals Engineering 21 (2008) 770781

771

investment in development time and capital cost. Mechanical shortcomings such as component rigidity may cause rejection of a design with potentially good process performance. If at least some of the required prototypes can be constructed and assessed by simulation, then a large reduction in development time and cost should be possible. In many ways worse still are machines which work well at lab and pilot scale but cannot be scaled up to commercially viable sizes. Each prototype cycle can easily cost 612 months of development time and tens to hundreds of thousands of development dollars. Hence, a system which can allow a good deal of the development work to reside in a simulation model has much to recommend it. However, to be credible as a Virtual Comminution Machine, both the simulation program and the ore characterisation process must be thoroughly veried using existing comminution devices. Hence the VCM should also be useful for modelling and further development of existing comminution devices. A VCM has the further advantage that variations on a theme can more easily be considered allowing estimation of the total energy footprint in terms of electrical energy and wear whilst treating identical simulated feed materials. 1.2. The underlying premise The Virtual Comminution Machine (VCM) must be able to relate each collision type and energy within a DEM simulation to the probable degree of breakage of that particle based on characterisation tests of a particular ore. As several modes of comminution occur within tumbling mills, relating collision spectra to comminution represents a particular challenge. The greater part of comminution energy is expended in a slurry environment in wet autogenous and semi-autogenous (AG and SAG) mills and ball mills. The proportion of ne (37l) material in the slurry has a strong effect on its viscous properties. A reasonably viscous slurry which can retain particles on the surface of the grinding media helps to ensure selection of those particles for breakage. Hence the uid and particle interactions are also important for particle breakage and transport. Models based on computational uid dynamics (CFD) are not well suited to the high shear, high viscosity environment with complex free surface behaviour found in wet tumbling mills. DEM can be adapted to uid modelling using Smoothed Particle Hydrodynamics (SPH), see Cleary (1998a) and Cleary et al. (2007). The shear dependant viscosity model developed by Shi and Napier-Munn (1996, 2002) has been implemented in the CSIRO Mathematics and Information Sciences (CMIS) code using SPH as a way to model mineral slurries. In a DEM simulation, the forces exerted by each particle on the equipment surfaces are estimated. Therefore the relative wear rates can also be estimated using an appropriate wear model. Similarly, the total energy requirements for the machine can be estimated by summing the energies consumed by particle movement. In short, the VCM objective is to model each process within any comminution machine to an acceptable level of accuracy so that it can be used as development environment for existing and new comminution machines. 2. A simple example The vertical shaft impactor (VSI) turns out to be very well suited to the VCM approach. Particles are fed into the center of a rapidly spinning rotor. The rotor accelerates each particle to a high velocity and throws them into a circle of steel anvils or a bed of particles. The resulting kinetic energy of each particle of mass m is mV2/2. Therefore the specic energy or energy per unit mass is simply

V2/2. The actual exit velocity is a good approximation to the rotor tip speed divided by the square root of two. The JKMRC Drop Weight Test (Napier-Munn et al., 1996) calibrates an ore for breakage at moderate input energies. Fig. 1 shows a comparison of measured and predicted VSI product size distributions based on DWT characterisation (Djordjevic et al., 2003). The predictions are precise over a wide range of operating conditions. The onset of particle interactions at higher throughputs is also well predicted. However, the accuracy of prediction of the progeny size distribution decreases as the particle interactions become more complex. A better prediction would require simulation of all the particles in the breakage region of VSI or at least in a radial slice with periodic boundaries. 3. Tumbling mills 3.1. Background As tumbling mills dominate comminution world wide, any performance improvements gained from a better understanding of these mills can potentially impact on an enormous asset base. A tumbling mill is a simple piece of equipment. However, a wide range of breakage mechanisms are active in tumbling mills making the processes occurring within them far from simple and making them very challenging to understand and to model. The VCM project has made substantial progress but considerable further development is still required. Perhaps the earliest claimed success for 2D DEM modelling was prediction of which lifter designs when used in large SAG mills would cause media overthrow and consequent liner damage and ball breakage by Rajamani and Mishra, 1996. However, simple single particle trajectory models (Powell, 1991; McIvor, 1983) are quite adequate for avoiding seriously awed lifter design. While this application can help to avoid a damaging problem, it offers limited insight into mill operation or comminution performance. However, the single particle trajectory models essentially ignore multi-particle interactions. The 2D DEM simulations do take some account of multi-particle interactions (where combinations of several rocks and/or balls can still cause ball overthrow and consequent liner damage) and can provide useful qualitative information (Cleary, 1998b, 2001a). But they do not provide even reasonable quantitative estimates of 3D behaviour except at the operating point where they are adjusted to match perceived mill behaviour as closely as possible. 3D models of tumbling mills (see Herbst and Nordell, 2001; Cleary, 2001b) capture much more

Fig. 1. Comparison of measured and predicted VSI product size distributions based on DWT characterisation (after Djordjevic et al., 2003).

772

R.D. Morrison, P.W. Cleary / Minerals Engineering 21 (2008) 770781

of the dynamics of real particle motion. A detailed comparison of a wide range of conditions in a 600 mm diameter scaled AG mill (Cleary et al., 2003) demonstrated that 3D simulation with spheres provided a realistic trajectory map over a wide range of conditions. 2D simulations with shaped particles were at least plausible but 2D discs were very poor estimators of 3D behaviour. This is perhaps not surprising as about two thirds of the 3D interactions between particles are ignored in a 2D simulation. Similarly, the tendency of large mass particles (rocks or balls) to congregate towards the center of the charge as mill speed is increased above 72% of critical is not realistically estimated by 2D simulation. Both of these issues have been considered in the detailed comparison of lab scale measurement with 2D and 3D predictions (Cleary et al., 2003). DEM simulations of tumbling mills have for a long time focussed on dry processes or simply ignored the slurry content that is, water and particles ner than a millimetre or two. As the slurry coating of grinding media or larger pebbles (Shi and Napier-Munn, 1996) ensures that at least some ne particles are selected for breakage in every collision, this behaviour is critical to prediction of ne progeny. As a 50 mm ore particle has a mass about one million times more than a 0.5 mm particle, even a low energy collision between two 50 mm particles can easily cause maximum breakage to an adhering layer of 0.5 mm particles around the point of collision. Even with todays computers in multi-processor arrays, computation of more than a very small sample of collisions at this level of detail is still a numerical impossibility and empirical approximations are required. In short, there are two major continuing challenges for DEM simulations of tumbling mills ne progeny and the behaviour of mineral slurry. The remainder of Section 3 reports on progress to date in simulation of tumbling mills. 3.2. Progress to date To explore a full scale SAG mill in 3D would require many millions of particles. Instead we have typically chosen to use slice models of full scale units (Morrison et al., 2001) and developed a full 3D simulation of a Hardinge style AG/SAG pilot mill (Morrison and Cleary, 2004). This mill is 1.8 m in diameter with conical ends and an internal volume equivalent to a cylinder 0.6 m long. The Hardinge mill is the industry standard for pilot scale test work. It requires 350450 thousand elements to simulate mill charge down to 6 mm particles. Typically, particles greater than 6 mm would constitute more than 9095% of the mill charge. Fig. 2 shows a cut away view of the simulated mill including discharge grates and pulp lifters. This is a complete mill model and includes the feed end, the main/belly/circumferential lifters, the end lifters, the grate, the pulp lifters, the discharge cone and the external mill shell. This conguration has 16 belly lifters with 10 face angles, 16 lifters on the feed end cone and 8 lifters on the discharge end. A representative SAG mill charge was used, with a rock top size of 122 mm and the bottom size resolved in the DEM model chosen to be 6 mm, so that >90% of the charge was correctly represented in the model. The ll level was 39% with a ball load of 5% giving around 425,000 particles. The mill rotation rate was 24 rpm corresponding to 76% critical. Fig. 2 also shows a view of the mill, which has been sectioned by an axial plane, and viewed along the axis looking towards the discharge end of the mill. The particles are shaded by size. A typical mill ow pattern is visible, with the particles carried up clockwise by the motion of the mill until they reach the shoulder position and then ow down the cascading free surface of the charge. For these lifters and this mill speed, there are signicant cataracting streams that are composed almost entirely of nes (due to strong radial segregation) leading to signicant numbers of impacts on the liner above the toe position.

The discharge grates can be seen on the end of the mill. Small collections of nes are also observed to collect on the end wall radial lifters. Impact breakage on the toe of the charge has long been considered to be a dominant contributor to throughput for SAG mills. Fig. 3 plots DEM estimates of impacts against time for a 26 mm particle. As expected, the particle receives about one major impact (say > 0.05 J) per revolution of the charge (Morrison and Cleary, 2004). This conrms the common wisdom but also shows that

Fig. 2. Cut away views of the Hardinge pilot mill including discharge grates and pulp lifters. Particles of lighter shades are larger in size.

Impact Energy J

0.25 0.2 0.15 0.1 0.05 0 0 1 2 3 4 Time s 5 6 7 8

Fig. 3. Impact history for a 26 mm particle circulating through the toe of the charge of the simulated pilot mill. The mill rotates at 30 rpm (after Morrison and Cleary, 2004).

R.D. Morrison, P.W. Cleary / Minerals Engineering 21 (2008) 770781

773

the energy levels are much too small to cause breakage in a single impact. Indeed about 25 J would be required to achieve a t10 of 10 in an average (A x b = 50) ore. The denition of t10 is the percentage of the progeny which will pass through an aperture one tenth of the size of the original particle. Napier-Munn et al. (Napier-Munn et al., 1996 Chapter 5) provide a detailed description of how A and b can be measured. Hence, comminution models based on the concept of high energy single impact breakage events in the toe region do not reect the reality of the collision environment actually occurring within the mill. Therefore we rst consider some low energy modes of breakage and then return to impact breakage. 3.3. Particle rounding Almost all freshly broken ore particles are angular in shape. Applying a force transverse to a sharp corner can remove the corner through a shear fracture in tension. As this is a low energy mode of breakage, there will be many shear events of sufcient energy and rounding is likely to happen quickly. Hence a particle will be well rounded soon after it enters the mill charge. A useful working denition of the rounding mechanism is simply the removal of corners from freshly broken, angular particles. The recent addition of parameterised super-quadric particles to the CMIS code (Cleary, 2004) allows such shape evolution of the particles based on the energy dissipation of these interactions. Fig. 4 shows a DEM simulation of 1.19 m diameter by 0.31 m long rubber-lined pilot mill. The mill was tted with 14 square metal lifters with heights of 40 mm. The rotational velocity of the mill was 3.14 rad/s (77% of critical speed or 30 rpm). When the experimental mill charge contained fresh ore, very rapid rounding occurred as the sharp corners of the particles were removed. This process is simulated using super-quadric particles in Fig. 4. To evolve their shapes, the contact locations on the non-spherical particles are used to analyse the fraction of the energy that is trans-

mitted into the centers of the particles and how much is absorbed in their corners and edges. Both the normal and the shear components contribute to the rounding mass loss mechanism. The mass loss rate is proportional to the magnitude of the rounding energy absorption and the particle shape is constantly re-calculated to match the reduced mass. This produces preferential mass loss from corners and from protruding ends leading to rounding of the particle. DEM prediction of the rounding process for initially angular and elongated initial feed material using this methodology are shown for the 1.19 m diameter batch pilot mill in Fig. 4. The behaviour produced matches closely to that observed in the experimental pilot mill (quantitative comparisons will be presented later). Banini (2002) carried out a detailed experimental investigation of this mode of breakage and concluded that the volume of the largest fragment produced by rounding was related to the initial geometry of the particle and the sphere which would just t into it. For an initial cube, this is a chip of linear dimension 0.21 of the edge of the cube or about 0.9% of the volume of the original particle. This corresponds to a gap in the size distribution of the progeny between the remainder of the original particle and the chips and ner progeny (or four fourth root of two sieve sizes). To reduce the size of the original particle sufciently to fall into the next lower size fraction requires the loss of about 40% of its mass. Therefore, many or most of the rounded particles will remain in their original size fractions. Banini further concluded that the rate of production of fragments by rounding varied with input energy and that the cumulative size distribution of the progeny could be normalised by dividing by the 50% passing size. This approach provides a relatively simple way to estimate the progeny of the rounding process. Fig. 5 shows a set of typical progeny produced by tumbling 90 + 75 mm particles of a siliceous copper ore in a series of small mills of different diameters. The energy estimates are based on the estimated mean lift heights in each small mill.

Fig. 4. Elongated angular particles are rounded in a few mill revolutions with the largest rate of shape change occurring early.

774

R.D. Morrison, P.W. Cleary / Minerals Engineering 21 (2008) 770781

0.1

Cumulative % Passing

0.01

0.39 J 1.24 J 3.17 J 6.83 J

0.001

0.0001 0.01

0.1

10

100

Size (mm)
Fig. 5. Progeny of 90 + 75 mm copper ore particles tumbled in a series of small mills at low energies, (after Banini, 2002).

3.4. Particle abrasion and chipping If the surface forces are relatively low, the rounded particles then undergo a more or less constant rate of mass loss. This phase of operation has been extensively investigated by Loveday and coworkers (Loveday and Hinde, 2002; Loveday and Naidoo, 1997). A constant rate of mass loss leads to a single parameter model which can make a reasonable estimate of an AG mill load based on the feed size distribution and that parameter (Loveday and Whiten, 2004). DEM modelling in this regime (Djordjevic et al., 2006) revealed that inter-particle friction provided a good guide to particle mass loss in both single particle size and mixed particle sizes in this pilot scale autogenous mill for a tough quartzite ore. This approach worked well for both the total charges and for small groups of individual particles. Hence, at low interaction energies a single parameter model works quite well for this mode of low energy mass loss. Therefore, a working denition of abrasion is the reduction in diameter by frictional forces. The progeny of this mode of breakage consists of ne particles whose sizes have little dependence on the size of the original particles. The particles are essentially being worn down. This mode seems to be important for autogenous mills where the Loveday and Whiten model (2004) can make a reasonable prediction of mill load based on mill feed and a single wear rate parameter. However, at higher energies, particle interactions produce chips from even smoothed surfaces. This mode is more common in SAG mills and probably represents local failure at a point of impact perhaps between a steel ball and an ore particle. An interesting model proposed by Zhang (1994) hypothesised that the ratio of the volume of a chip removed to the original particle size was a function of the energy of the collision and the ratio of the Youngs modulus of the ore to its Fracture Toughness. There is a good correlation between the Fracture Toughness and the A x b value (Briggs and Bearman, 1994). The model can be tested for steel ball/ore particle collisions but would need to be extended for collisions between ore particles. 3.5. Incremental impact breakage While the properties of impact breakage at high energies are reasonably well understood (as shown in Section 2), breakage of ore at the particle interaction energies available in AG/SAG mills is poorly understood. The frictional and rounding modes of breakage produces well rounded pebbles which apparently suffer very

little internal damage. However, at higher interaction energies the rate of breakage increases rapidly as particles begin to break by what appears to be impact breakage even though the interaction energies are much too small to cause impact breakage in a single collision. This may seem like a contradiction in terms. However, as noted in Section 3.2 the impact energies are much too low to achieve breakage in a single impact. The idea of incremental breakage is not considered by traditional methods to relate to rock strength. As exposure to a low level of impact means that a particle may or may not be broken in a single impact, the traditional tests are unable to provide reproducible outputs. Hence, essentially all tests use sufcient energy to achieve breakage in a single impact to improve the reproducibility of the test. The currently accepted evidence for apparent impact breakage in AG and SAG mills is the presence of freshly broken particles in the mill charge as well as the expected rounded ones. However, as shown in Fig. 3, there are essentially no suitable impacts available to break even quite small particles in a single hit. This suggests that there exists a threshold energy for particle interactions above which cracks begin to propagate within the core of the particle even though the particles may still appear to be completely intact. This leads to a useful denition of body breakage. This term includes both impact and incremental breakage. The distinguishing feature that the core of the particle fails producing a progeny top size which is much smaller than the parent particle and does not leave any particles within the original fourth root of two size fraction. The effect of repeated small impacts has now been investigated in some detail (Morrison et al., 2007). A satisfactory model for the probability of breakage, degree of breakage and likely progeny size distribution was developed based on the standard JKMRC impact breakage model (Napier-Munn et al., 1996) and the work of Vogel and Peukert (2004) with modications suggested by Shi and Kojovic (2006). In summary, the probability of survival after a series of impacts is predicted by

S 1 expffmat x kW m;kin W m;min g

where S = breakage portion (fraction); Wm,kin = single impact energy (J); Wm,min = energy which causes no impact damage (J); x = diameter of particles (m); k = number of impacts; fmat = material parameter (1/m).The likely severity of breakage can be estimated in terms of the net energy absorbed by the particle.

R.D. Morrison, P.W. Cleary / Minerals Engineering 21 (2008) 770781

775

S 1 exp b

X
i

! Ei Eo 2

Cumulative Fraction Broken

The Vogel and Peukert (2004) equation provides a useful model to test. In slightly modied form, the probability of selection for S breakage after i events is

Note that Ei Eo may not be smaller than zero and b0 is a material parameter which may contain a particle size component. The probability of survival after i events is simply (1 S).

PSurvivalji exp b0

X Ei Eo
i

! 3

100 90 80 70 60 50 40 30 20 10 0 1 2 3 4 5 6 7 8 9 Number of Impacts

30% 40% 50% 10

lnPSurvivalji b

X Ei Eo
i

While Fig. 9 shows a good deal of scatter, the trend lines for 30% and 50% are reasonably linear and coincident. As might be expected, the lower energy inputs will show more scatter as a single soft or hard particle will make a large difference. As this is a relatively new idea in mineral processing, it is worth considering a little further in the context of DEM. Powell and Morrison (2006) suggested a

Ln(Probability of Survival)

Whyte (2005) carried out a series of preliminary tests on 16 19 mm and 22 26 mm size fractions at four energy levels. These provide a manageable sub-set to test the survival model. Figs. 6 and 7 show the fractions of survivors after 110 hits at 20%, 30%, 40% and 50% of the specic energy required to cause breakage in a single impact. These gures provide a useful way of describing incremental impact breakage. The percentages of input energy refer to the fraction of the energy required to achieve high probability of breakage in a single impact. Some scatter is apparent but this is not surprising as the DWT assumes that all particles within a fourth root of two sieve intervals have the same mass. In fact there can be a 40% difference between the masses of the largest and smallest particles. Hence the particles can receive similarly different levels of input energy. Further, there are variations of particle shape and probably composition. A new breakage device is being developed which provides equal levels of input energy to each particle. Eq. (3) provides a plausible model for the results shown in Figs. 6 and 7. The dependence is plotted below in Fig. 8 and is quite reasonable but does not fully test the form of the model. A more traditional approach is to transform the model into a form where it will produce a linear plot if the form is correct. Taking the logarithm of each side of Eq. (3) provides a suitable linear form. Fig. 9 shows the data of Fig. 6 plotted in the format of Eq. (4). While there is some scatter, most of the survival probabilities are approximately linear.

Fig. 7. 26.5 + 22.4 mm preliminary test results for breakage caused by successive impacts at various fractions of the minimum energy required to cause breakage in a single impact (after Morrison et al., 2007).

100.00 90.00 80.00 70.00 60.00 50.00 40.00 30.00 20.00 10.00 0.00 1 2 3 4 5 6 7 8 9 10
20% 30% 40% 50%

Fig. 8. Data from Fig. 6 tted to Eq. (3).

1.000 20% 30% 40% 50% 0.100

100 90 80 70 60 50 40 30 20 10 0 1 2 3 4 5 6 7 8 Number of Impacts

Cumulative Fraction Broken

0.010 0.000

0.200

0.400

0.600

0.800

1.000

Cumulative net energy (kWH/t)

20% 30% 40% 50%

graphical approach to describing a very wide range of behaviour as shown in Fig. 10. This approach allows damage as a function of impact input energy to be divided into three main categories: 1. Single hit: E > Ecrit: Energies at which single impact breakage occurs; 2. Multiple hit: Eo < E < Ecrit: Energies at which cumulative damage occurs, resulting in particle breakage after multiple impacts;
Fig. 9. Data from Fig. 6 plotted as log of the probability of survival as vs. net accumulated comminution energy (Eq. (4)).

10

Fig. 6. 19 + 16 mm test results for breakage caused by successive impacts at various percentages of the minimum energy required to cause breakage in a single impact, (after Morrison et al., 2007).

776

R.D. Morrison, P.W. Cleary / Minerals Engineering 21 (2008) 770781

Fig. 10. DEM impact energy spectrum for a 38 mm sphere with rock like properties in an 8 m diameter mill (after Powell and Morrison, 2006).

3. Sub-damage: E < Eo: Energies that result in no damage and do not contribute to particle breakage. where  Ecrit = critical input energy. Dened by the 95% condence limit of the energy at which breakage occurs in a single hit.  Eo = threshold damage energy. Dened by the energy at which the probability of breakage in an innite number of hits is below 5%. Given that Eq. (4) appears to provide a realistic model of the process, Eo is a function of the material which is strongly related to the minimum stress level required to extend a crack. Ecrit will depend to some degree on the size of the particle if the b0 factor does indeed contain the particle size. Powell et al. (2007) are extending this concept to a grinding a mill model parameterized on the basis of multiple DEM simulations. 3.6. Progeny produced by incremental breakage The size distribution of the progeny appears to be quite similar to that achieved in a single impact suggesting a similar process of

crack extension in one event or in several. The commonly used JKMRC relationship for severity of breakage can be modied in the same way (Shi and Kojovic, 2006).

t10 A1 exp b

X
i

! Ei Eo 5

where t10 = is the fraction of the mass of the original particle which will pass through an aperture of 1/10 of the original particle size after the impact event; A = represents the maximum t10 achievable in a single breakage event; Eo = is a threshold energy per unit of particle mass below which the particle essentially does not accumulate any impact damage. Ei = is the applied energy per unit mass for each of i impact events. Note that the contribution from (Ei Eo) is zero if Ei is less than Eo. By its nature, incremental breakage also has the interesting feature that very few particles are likely to be severely broken when they do nally fail. Fig. 11 shows the size distributions of progeny produced by incremental breakage compared with progeny from single impacts. Progeny from one quite severe impact is included for comparison. It is clear that the size distributions produced by the two processes fall within the same progressively ner pattern with increasing severity of breakage and are unlikely to be distin-

100.00 90.00 Cumulative Percent Passing 80.00 70.00 7 hits - total 0.28 kWh/t 60.00 50.00 40.00 30.00 20.00 10.00 0.00 0.10 1.00 10.00 100.00 8 hits - total 0.32 kWh/t 9 hits - total 0.36 kWh/t One hit at 1 kWh/t One hit at 0.25 kWh/t

Size (mm)
Fig. 11. Comparison of progeny size distributions produced by single and incremental breakage for 31.5 26.5 mm particles at 50% of the minimum for breakage in a single impact (after Morrison et al., 2007).

R.D. Morrison, P.W. Cleary / Minerals Engineering 21 (2008) 770781

777

% probability of breakage

13.2 - 16.0 mm 19.0 - 22.4 mm 26.5 - 31.5 mm Calculated

100 90 80 70 60 50 40 30 20 10 0 0 0.1 0.2 0.3 0.4 Energy input (kWh/t)

t10 (%)

Rounded Angular

0.5

0.6

0 0.0 1.0 2.0 3.0 4.0

Fig. 14. Cumulative breakage probability curves for angular and rounded particles (after Bbosa et al., 2006).

fm at.x.k.(Ecs -Em in)

Mass of each progeny particle (g)
Fig. 12. The re-grouped Whyte (2005) data tted to Eq. (5) including size dependence (after Morrison et al., 2007).

4000 3500 3000 2500 2000 1500 1000 500 0 0 1000 2000 3000 4000 5000 Mass of each feed particle (g) Mass (f) DEM at 720s

guishable on this basis. The coarse progeny produced in this way will be rapidly rounded in the mill and may be the underlying cause of the so called critical size problem common to AG and SAG mills. Eq. (5) (Shi and Kojovic, 2006) can be rearranged in the same manner as Eq. (3) to generate Eq. (6). Eq. (6) provides a similar test for linearity of the model for progeny generated by incremental breakage. Intuitively, it seems quite reasonable that the probability of breaking at all and the probability of a certain degree of relative breakage should have quite similar structures. Clearly, particles which fail to break should not be considered as progeny of incremental breakage.

ln1 t 10 =A b

X Ei Eo
i

Fig. 15. Comparison between measured and predicted particle masses after 12 min of tumbling in a 1.2 m diameter mill in an autogenous environment.

6
which were signicant fractions of the energy required to break them in a single hit. The absorbed energy during each collision was measured using a vertical version of a Hopkinson Pressure Bar (Bourgeois and Banini, 2002). The sample cylinders were examined in a Cone Beam Micro Micro-Tomograph. Successive reconstructions are shown in Fig. 13. They clearly show crack

However, a more detailed set of data from Whyte (2005) was a satisfactory t to Eq. (5) as shown in Fig. 12. Two further pieces of experimental work conrm the process of incremental damage without obvious physical damage. Hocking (2006) subjected small cylinders of rock to successive impacts

Fig. 13. Two CBT reconstructions of a small cylinder of granite which had been subjected to repeated impacts with a steel ball, (left) the top of the sample and (right) a cross Section 2 mm below. The indentation is visible to the eye but the cracks are not (after Hocking, 2006).

778

R.D. Morrison, P.W. Cleary / Minerals Engineering 21 (2008) 770781

4000 3500 3000 2500 2000 1500 1000 500 0 0 1000 2000 3000 4000 Mass of each feed particle (g) Mass (f) DEM at 720s

extension with increasing progressive damage which is not visible to the naked eye. Bbosa et al. (2006) used a rotary breakage tester to compare the probability of breakage of angular (freshly broken) particles with particles of the same material which had been rounded by tumbling in a small wet mill at very low input energies. As shown in Fig. 14, the measured particle strength is reduced by more than 10% suggesting that surface roughness does protect angular particles to a signicant degree but the Eo behaviour is clearly still present. Hence, in a tumbling mill, the initial particle rounding will likely reduce the energy barrier to impact damage quite signicantly. 3.7. Integrating the breakage mechanisms (putting it all together) A combined rounding, chipping and abrasion breakage model has been embedded within the CMIS DEM Code and used to simulate the 1.2 m diameter pilot mill (Morrison et al., 2006). The experimental set up included precise power measurement and the DEM simulations provide an excellent prediction of the measured power draw as might be expected when the mass of each particle has been measured. The mass loss predictions are encouraging for the AG mill conguration but cannot yet predict the expected transition from abrasion only to abrasion plus incremental impact breakage until the incremental damage based breakage model development is completed. However, the predicted and measured results are quite close. The difference in apparent impact strength between angular and smooth particles as shown in Fig. 15 is of the right order to explain the difference for the AG mill. Fig. 15 shows a comparison between measured and predicted particle masses after 12 min of tumbling in a 1.2 m diameter mill in an autogenous environment. This gure represents an unusual way to present comminution data. However, as in this case, we have access to experimental measurements of individual particle masses and to the simulated masses of each individual particle taking account of abrasion and impact losses, it seems reasonable to plot individual measured and simulated particle masses against their original measured particle masses. A perfect simulation would overlay the diamonds on the squares. Allowing for rounded particles and some experimental scatter, Fig. 15 seems to be a very

Fig. 16. Comparison between measured and predicted particle masses after 12 min of tumbling in a 1.2 m diameter mill in a SAG environment.

Fig. 17. Discharge of coarse sub-grate particles though the grate and ow along the pulp lifters in the Hardinge pilot SAG mill.

Mass of each progeny particle (g)

Fig. 18. Slurry ow in a SAG mill, (left) ow within the grinding chamber, and (right) ow through the discharge grate into the pulp lifter chamber.

R.D. Morrison, P.W. Cleary / Minerals Engineering 21 (2008) 770781

779

reasonable direct prediction of progeny sizes even though the mass changes are quite small as they were derived from rates of mass loss in a 300 mm diameter mill. No attempt has been made to t them to the larger mill. Fig. 16 shows a comparison between measured and predicted particle masses after 12 min of tumbling in a 1.2 m diameter mill in a SAG environment. At the higher impact energies of this milling environment, it is clear that incremental impact breakage is having a strong effect at smaller particle sizes and is pushing the experimental data well below the linear distribution found for the AG case. The absence of the incremental breakage in the DEM model at the time of making these predictions means that the DEM is not yet able to predict this change in behaviour. The observed experimental change indicates the magnitude of effect that the

incremental breakage mechanism can have on mill performance and the importance of including this previously unknown mechanism in the modelling. 3.8. Adding slurry to the VCM Many of the key processes in wet grinding mills are controlled by the ow of slurry within the mill and out of it. Fully coupled SPH and DEM models should allow transport within a mill, through the grate and within the pulp lifters to be accurately modelled in the next year or two. However, even considering dry milling and slurry behaviour on its own offers some interesting insights. Fig. 17 shows the discharge ow of ne particles only through the grate of the Hardinge pilot mill (from Cleary, 2004). Pressure

Fig. 19. Equilibrium distribution of slurry for (a) half the amount of slurry 3.6 m2, (b) base case amount of slurry 7.2 m2, and (c) double the amount of slurry 14.4 m2. The slurry is coloured by (left) its uid fraction with red being pure slurry and dark blue being 0.35 and (right) by the slurry speed with red being 7 m/s and dark blue being stationary.

780

R.D. Morrison, P.W. Cleary / Minerals Engineering 21 (2008) 770781

from the charge on the inside of the grate forces particles smaller than the 12 mm grate opening to move through the grates. They fall and collect on the liner and on the radial pulp lifters in the lower and right sides of the mill. As the pulp lifters rotate, they lift and throw the discharged nes. These move on parabolic trajectories due to the sideways motion imparted by the lifters and the downward motion produced by gravity. The intention of pulp lifters is that the nes should ow down along the pulp lifters and onto the discharge cone, which should push them out through the trommel of the mill. What is actually observed is that they are thrown from the front face of one pulp lifter onto the back face of the preceding lifter (generating wear) and then fall down onto the outside liner (generating more wear) and are trapped at the bottom of the mill, destined to be lifted and to repeat the same behaviour forever trapped in the discharge region. Of the falling particles that avoid hitting the back of the previous lifter, most are deected by the discharge cone, not into the trommel but down into the bottom of the mill again. The dry discharge performance of this mill is very poor indeed. However, in industrial practise, some degree of air ow would be used in a dry mill to transport the nes or else they might ow over a weir onto a screen and the screen oversize would be recycled. The more common case is transport by slurry which is discussed next. Fig. 18 shows the ow of slurry within the grinding chamber of the SAG mill (in the absence of the rock and grinding media) (from Cleary et al., 2007). This is a challenging modelling problem since the speeds are high and the geometry is complex. The mill rotation means that the slurry is distributed over a wide part of the mill shell from the toe (lowest part) to the shoulder, where the slurry can be seen owing off the rising lifter bars. The lifters crashing into the slurry pool at the toe generate signicant wave motion and splashing. The end wall lifters generate signicant recirculation against the end walls which could lead to accelerated wear of these surfaces. SPH is easily able to model this system because of both its free surface capabilities and the absence of advection problems (because of its Lagrangian nature). The right frame of Fig. 18 shows the discharge of slurry through the ne grate that separates the grinding chamber from the pulp lifters that are responsible for pouring the slurry out through the discharge port around the axis of the mill on the discharge side. As mentioned earlier, the slurry ow is much cohesive than that of ne particles and the deector cone directs much of the ow towards the trammel which would be attached to the end of the mill. The pilot scale work has been extended to simulation using one way coupled ow between a slurry simulated by a ow of SHP uid in a 36 ft SAG mill charge with dynamic porosity estimated from a DEM simulation. This simulation method offers some interesting insights into slurry behaviour within the mill charge, (Cleary et al., 2006). Fig. 19 illustrates the likely effects of different levels of pulp lling within the SAG mill charge using this sequential DEM-SPH model. In case Fig. 19b, just enough slurry is added to completely ll the voids within the charge. In case a), half as much slurry is added and in case c), twice the volume of slurry is added. The left hand column of Fig. 19 shows the uid fraction (with blue being the lowest and red being 100% uid) and the right hand column shows the uid velocity (with hotter colours indicating higher velocities). It is clear that the charge acts in a similar manner to a positive displacement pump. That is, slurry ows into the loose charge at the toe. Next, cascaded material covers the slurry lled charge and seals it to some degree. Then the motion of the mill compresses the charge and carries it upwards with the contained slurry. Hence in case a), the slurry is swept around the mill shell more quickly than it can diffuse into the center of the charge. A large elliptical dry region appears below the shoulder. The slurry being pumped upwards against the mill shell by the charge is much thin-

ner (due to the lower volume of slurry available) and turns over just below the shoulder producing a matching stream owing down through the cascading region. The permeability is not sufciently high for this thinner stream to be able to ow down into the region directly below the shoulder, so this region remains dry. By the time the charge has owed down to the tail, the slurry that was in its upper regions has had plenty of time to ow down into the middle and lower sections of the charge. This leads to a sharp reduction of slurry in the toe region which is now much drier. Hence, the charge lls from the outside and leaves a void in the middle at lower levels of volumetric ll. If we instead double the amount of slurry available, then it lls all the pores available in the charge. This leads to a higher shoulder for the slurry and to the formation of a large slurry pool. This is readily identiable as the thick red layer of slurry on top of the charge in the region leading from the toe. Of interest is the observation that the slurry toe is higher than the level of the uid closer to the center of the mill. This means that the slurry free surface is angled upwards towards the toe. This results from the momentum of the slurry coursing down the cascading free surface which pushes the slurry towards the toe. The strong dynamic pressures in the uid produced by the surface uid ow from the shoulder continually maintain the slurry pool prole near the toe. Full two-way coupling between the DEM and SPH models should reveal just how much this pool can interfere with the magnitude of the one good hit per charge revolution with industry experience suggesting a serious decrease in throughput is a likely result.

4. Conclusions The collaboration to develop a VCM has been running for almost three years. As well as the work reported in this paper, crushers, dry vibratory screening, tower and stirred mills have been modelled, (Sinnott et al., 2006). Hence, progress to date is very encouraging. For Autogenous and SAG mills, some more specic conclusions can be drawn:  Mill power draw can be predicted with high precision  Particle rounding can be considered in terms of DEM interactions between particles with realistic shapes.  Mass loss of rounded particles is proportional to the DEM estimates of frictional energy loss between the particles.  Typically, particles larger than 2030 mm are not broken by single impacts but acquire internal damage by many impacts whose magnitude is greater than an ore dependant, specic threshold energy.  The required specic energy depends on degree of smoothing of the particle by the action of the mill charge.  Contrary to some historical views, the mill charge does not behave like a partially lled tank of slurry. The mill volume lls with slurry from the outer edge and a slurry pool forms only when the charge is completely full. Within the next three years the project team hope to complete comprehensive models of tumbling mills and other types of crushers. The combination of DEM and detailed experimental verication for SAG mills and for ore characterisation has exposed nonobvious opportunities for process improvement. Acknowledgements Some parts of the VCM development have been carried out under the auspices and with partial nancial support of the Centre for

R.D. Morrison, P.W. Cleary / Minerals Engineering 21 (2008) 770781

781

Sustainable Resource Processing, which is established and supported under the Australian Governments Cooperative Research Centres Program. Input from collaborators at the University of Cape Town, the University of Kwa-Zulu Natal and McGill University and also of colleagues at CSIRO Mathematical and Information Sciences and the University of Queensland is gratefully acknowledged. References
Banini, G.A., 2002. An integrated description of rock breakage in comminution machines. Ph.D. Thesis. University of Queensland (Unpublished). Bbosa, L., Powell, M.S., Cloete, T.J., 2006. An investigation of impact breakage of rocks using the Split Hopkinson Pressure Bar. J. SAIMM 106, 291296. Bourgeois, F.S., Banini, G.A., 2002. A portable load cell for in situ ore impact breakage testing. Int. J. Mineral Process. 65 (1), 3154. Briggs and Bearman, 1994. Cleary, P.W., 1998a. Modelling conned multi-material heat and mass ows using SPH. Appl. Math. Model. 22, 981993. Cleary, P.W., 1998b. Predicting charge motion, power draw, segregation, wear and particle breakage in ball mills using discrete element methods. Min. Eng. 11, 10611080. Cleary, P.W., 2001a. Charge behaviour and power consumption in ball mills: Sensitivity to mill operating conditions, liner geometry and charge composition. Int. J. Min. Process. 63, 79114. Cleary, P.W., 2001b. Recent advances in DEM modelling of tumbling mills. Min. Eng. 14, 12951319. Cleary, P.W., 2004. Large scale industrial DEM modelling. Eng. Comput. 21, 169 204. Cleary, P.W., Morrison, R., Morrell, S., 2003. Comparison of DEM and experiment for a scale model SAG mill. Int. J. Min. Process. 68, 129165. Cleary, P.W., Sinnott, M.D., Morrison, R.D., 2006. Prediction of slurry transport in SAG mills using SPH uid ow in a dynamic DEM based porous media. Min. Eng. 19 (15), 15171527. Cleary, P.W., Prakash, M., Ha, J., Stokes, N., Scott, C., 2007. Smooth particle hydrodynamics; status and future potential. Progr. Comput. Fluid Dyn. 7, 7090. Djordjevic, N., Shi, F.N., Morrison, R.D., 2003. Applying discrete element modelling to vertical and horizontal shaft impact crushers. Min. Eng. (10), 983999. Djordjevic, N., Morrison, R., Loveday, B., Cleary, P.W., 2006. Modelling comminution patterns within a pilot scale AG/SAG mill. Min. Eng. 19 (15), 15051516. DOE, 1981. Committee on Comminution and Energy Consumption, NMAB-364. Fuerstenau, D.W., Abouzeid, A.-Z.M., 2002. The energy efciency of ball mill in comminution. Int. J. Miner. Process. 67, 161185. Herbst, J.A., Nordell, L., 2001. Optimization of the design of sag mill internals using high delity simulation. In: Vancouver, B.C., Barratt, D.J., Allan, M.J., Mular, A.L. (Eds.), Proceedings of the SAG Conference, University of British Columbia, IV, 150164. Hocking, R., 2006. Private communication. La Nauze, R.D., Temos, J., 2002. Technologies for sustainable operation. CMMI Congress 2002, Cairns Queensland, May published AusIMM, 2734.

Loveday, B.K., Hinde, H.A., 2002. Application of a rock abrasion model to pilot-plant and plant data for fully and semi-autogenous grinding. Min. Process. Extract. Metall. 111, C39C43. Loveday, B.K., Naidoo, D., 1997. Rock abrasion in autogenous milling. Min. Eng. 10, 603612. Loveday, B.K., Whiten, W.J., 2004. Application of a rock abrasion model to pilotplant and plant data for fully and semi-autogenous grinding. Trans. IMMt C 111, C39C43. McIvor, 1983. Effects of speed and liner conguration on ball mill performance. Mining Eng. June, 617622. Morrison, R.D., Cleary, P.W., Valery, W., 2001. Comparing Power and Performance Trends from DEM and JK Modelling. SAG 2001. Department of Mining and Minerals Process Engineering, University of British Columbia, Vancouver, pp. 284300. Morrison, R.D., Cleary, P.W., 2004. Using DEM to model ore breakage within a pilot scale SAG mill. Min. Eng. 17, 11171124. Morrison, R., Loveday, B., Djordjevic, N., Cleary P., Owen, P., 2006. Linking Discrete Element Modelling to Breakage in a Pilot Scale AG/SAG Mill. Advances in Comminution, ed Kawatra, published SME, 495512. Morrison, R.D., Shi, F., Whyte, R., 2007. Modelling of incremental rock breakage by impact for use in DEM models. Min. Eng. 20, 303309. Musa, F., Morrison, R.D., 2007. Assessing the eco-efciency of comminution processes, XXII ENTMME Southern Hemisphere Conference International Mineral Processing Congress, November, Ouro Preto, Brazil, pp. 141149. Napier-Munn, T.J., Morrell, S., Morrison, R.D., Kojovic, T., 1996. Mineral Comminution Circuits: Their Operation and Optimisation, pp. 4992 (Julius Mineral Kruttschnitt Mineral Research Centre: Brisbane). Powell, M.S., 1991. The effects of liner design on the motion of the outer grinding elements in a rotary mill. Int. J. Min. Process. 31, 163193. Powell M.S., Morrison, R.D. 2006. A new approach to breakage testing. Eurocom 06. Powell, M.S., Govender, I., Kulya, C., McBride, A.T., 2007. Applying DEM outputs to the Unied Comminution Model the SAG mill. DEM07 Brisbane. Min. Eng., in press. Rajamani, R.K., Mishra, B.K., 1996. Dynamics of ball and rock charge in sag mills. In: Proceedings of SAG 1996. Department of Mining and Mineral Process Engineering, University of British Columbia. Schoenert, K., 1986. On the limitation of energy saving in milling. 1. World Congress Particle Technology, Part II, Comminution, Nurnberg, April 1619, p. 1. Shi, F., Kojovic, T., 2006. Validation of a model for impact breakage incorporating particle size effect. Int. J. Min. Process. 82 (3), 156163. Shi, F., Napier-Munn, T.J., 1996. A model for slurry rheology. Int. J. Min. Process. 47 (12), 103123. Shi, F.N., Napier-Munn, T.J., 2002. Effects of slurry rheology on industrial grinding performance. Int. J. Min. Process. 65 (34), 125140. Sinnott, M.D., Cleary, P.W., Morrison, R.D., 2006. Analysis of stirred mill performance using DEM simulation: Part 1 media motion, energy consumption and collisional environment. Min. Eng. 19, 15371550. Vogel, L., Peukert, W., 2004. Determination of material properties relevant to grinding by practicable labscale milling tests. Int. J. Min. Process. 74S, S329 S338. Whyte, R., 2005. Measuring incremental damage in rock breakage by impact. BE (honours) Thesis. The University of Queensland (Unpublished). Zhang, Z. 1994. Impact attrition of particulate solids. Ph.D. Thesis, University of Surrey, Guildford, Surrey, UK.