DEM Prediction of Particle Flows in Grinding Processes: P. W. Cleary, M. D. Sinnott and R. D. Morrison

INTERNATIONAL JOURNAL FOR NUMERICAL METHODS IN FLUIDS Int. J. Numer. Meth.
Fluids 2008; 58:319353 Published online 7 January 2008 in Wiley InterScience (www.interscience.wiley.com). DOI: 10.1002/d.1728
DEM prediction of particle ows in grinding processes

P. W. Cleary1, , , M. D. Sinnott1 and R. D. Morrison2
1 CSIRO 2 Julius
Mathematical and Information Sciences, Private Bag 33, Clayton, Vic. 3168, Australia Kruttschnitt Mineral Research Center, The University of Queensland, Isles Rd, Indooroopilly, Qld. 4068, Australia
SUMMARY Particle size reduction is a critical unit process in many industries including mineral processing, cement, food processing, pigments and industrial minerals and pharmaceuticals. The aim is to take large feed material and as efciently as possible reduce the size of the particles to a target size range. Over time, a very large range of equipment has been developed to perform this for many materials and in many different conditions. Discrete element method (DEM) modelling is a computational tool that can allow detailed exploration of the particle ow and breakage processes within comminution equipment and can assist in developing a clearer and more comprehensive understanding of the detailed processes occurring within. In this paper, we examine the particle and energy ows in representative examples of the equipment used in many grinding processes. We study a 36 semi-autogenous mill used in primary grinding for mineral processing, a ball mill used for cement clinker grinding, a grinding table also used for cement grinding, a tower mill used for ne grinding both in mineral processing and for industrial minerals and nally for an Isamill, which is used for ultra-ne grinding in mineral processing. Copyright q 2008 John Wiley & Sons, Ltd.
Received 22 March 2007; Revised 12 November 2007; Accepted 20 November 2007 KEY WORDS:
comminution; milling; particles; breakage; energy utilization; DEM
1. INTRODUCTION Particle breakage is an essential process in many different industries, ranging from mineral processing, quarrying, cement to food processing and pharmaceuticals. The aim is to reduce the original feed stock particles from the sizes supplied to the much smaller sizes required for either further processing or to be a part of nal products. In some industries, particularly mineral
Correspondence
to: P. W. Cleary, CSIRO Mathematical and Information Sciences, Private Bag 33, Clayton, Vic. 3168, Australia. E-mail: paul.cleary@csiro.au
Copyright q
2008 John Wiley & Sons, Ltd.
320
P. W. CLEARY, M. D. SINNOTT AND R. D. MORRISON
processing, the intent is to do this as efciently as possible (that is, with respect to energy consumption and wear materials), producing optimal sizes for liberating target minerals at maximum mill throughput and minimizing operating costs. For a good introduction to the issues of comminution in mineral processing, see [1]. In industries such as food and pharmaceuticals, the mill sizes are much smaller and capital and energy costs are unimportant, but the nal product specications are tight and the focus is on quality and cleanliness/avoidance of contamination. Size reduction is typically accomplished in a series of stages whose design is highly dependent on the nature and size of the feed stock and the reduction ratio that is required. The rst stage in mining and mineral processing is usually some form of blasting to liberate ore and waste from the deposit. A typical mineral processing grinding circuit is tailored to the properties of the particular mine ore. Commonly, this involves two or more classes of equipment starting with crushers, followed by semi-autogenous (SAG) mills and then sometimes ball mills. Occasionally, high-pressure grinding rolls or other novel devices are substituted. For cement industries, multi-stage ball mills, roller mills or grinding tables and high-pressure grinding rolls are often used, sometimes in series. For ne and ultra-ne grinding, particles will typically pass through a number of stages of grinding, nishing with tower mills or stirred mills such as the Netzsch or Isamill. Regardless of the industry and which of the possible outcomes is required, the size reduction process needs to be better understood and optimized for those applications. Traditionally, this has been done using extensive laboratory, pilot scale and eld trials. These are expensive and time consuming and allow models to be constructed that are limited by the range of conditions tested and the complexities of the many different processes occurring within these mills. For a good introduction to data tted modelling of a broad range of comminution equipment, see [1]. Traditional experimental and modelling approaches are not well placed to create the required insights into comminution fundamentals or facilitate radical step changes in performance. Computational modelling, using the discrete element method (DEM), has been growing in capability over the past decade and is now able to provide signicant insights into the operation of existing and future mills and can be used as a tool to aid equipment and process design and optimization. DEM has been used for modelling a wide range of industrial applications, particularly in milling. Early work on ball mills by Mishra and Rajamani [2, 3] has been followed by Cleary [4, 5] and Cleary and Sawley [6]. Axial transport and grate discharge were studied in [7]. Similarly, SAG mills were modelled by Rajamani and Mishra [8] and subsequently by others [912]. Experimental validation has been performed for a centrifugal mill [13] and for charge motion and shape in a laboratory scale model SAG mill [1416] and for grinding rates [17]. Other applications include vibrational segregation [1820], ows in rotating drums [1921], hopper ows by Ristow [19] Potapov and Campbell [22] and many more. A broad range of applications simulated using DEM by Cleary et al. are reviewed in [23].
2. DISCRETE ELEMENT METHOD DEM is a simulation method that models particulate systems whose motions are dominated by collisions. This involves following the motion of every particle or object in the ow and modelling each collision between the particles and between the particles and their environment (stationary and moving objects such as mills and crushers). The discrete element algorithm has three main
Copyright q 2008 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Fluids 2008; 58:319353 DOI: 10.1002/d
DEM PREDICTION OF PARTICLE FLOWS IN GRINDING PROCESSES
321
stages: 1. A search grid is used to periodically build a near-neighbour interaction list that contains all the particle pairs and objectparticle pairs that are likely to experience collisions in the short term. Using only pairs in this list reduces the force calculation to an O(M) operation, where M is the total number of particles. Industrial simulations with 2 million particles are now possible in reasonable times on current single-processor workstations. 2. The forces on each pair of colliding particles and/or boundary objects are evaluated in their local reference frame using a suitable contact force model and then transformed into the simulation frame of reference. 3. All the forces and torques on each particle and object are summed and the resulting equations of motion are integrated to give the resulting motion of these bodies. Time integration is performed using a second-order predictorcorrector scheme and typically uses between 15 and 25 time steps to accurately integrate each collision. This leads to small time steps (typically 104 106 s depending on the controlling length and time scales for each application). 2.1. Collision model The general DEM methodology and its variants are well established and are described in review articles [2426]. The implementation used here is described in more detail in Cleary [4, 20, 23]. Briey, the particles are allowed to overlap and the amount of overlap x, and normal vn and tangential vt relative velocities determine the collisional forces via a contact force law. There are a number of possible contact force models available in the literature that approximate collision dynamics to various extents. We use a linear spring-dashpot model that is shown diagrammatically in Figure 1. For more complex models, see [26, 27]. The normal force Fn = kn x +Cn vn (1)
consists of a linear spring to provide the repulsive force and a dashpot to dissipate a proportion of the relative kinetic energy. The maximum overlap between particles is determined by the stiffness kn of the spring in the normal direction. Typically, average overlaps of 0.10.5% of the particle diameter are desirable, requiring spring constants of the order of 106 108 N/m in three dimensions.
Figure 1. Contact force law consisting of springs, dashpots and frictional sliders used for evaluating forces between interacting particles and boundaries.
322
The normal damping coefcient Cn is chosen to give the required coefcient of restitution (dened as the ratio of the post-collisional to pre-collisional normal component of the relative velocity), and is given in [20]. The tangential force is given by Ft = min Fn , kt vt dt +Ct vt (2)
where the vector force Ft and velocity vt are dened in the plane tangent to the surface at the contact point. The integral term represents an incremental spring that stores energy from the relative tangential motion and models the elastic tangential deformation of the contacting surfaces, while the dashpot dissipates energy from the tangential motion and models the tangential plastic deformation of the contact. Depending on the history of the contact, it is possible for the spring to be loading in one direction and simultaneously unloading in the orthogonal direction. The total tangential force Ft is limited by the Coulomb frictional limit Fn , at which point the surface contact shears and the particles begin to slide over each other. 2.2. Geometry representation Industrial applications place heavy demands on the geometric capabilities of DEM codes, particularly for three-dimensional geometries. In our DEM code, any two-dimensional cross section can be constructed using an appropriate number of line segments, circular segments or discs. This gives essentially unrestricted geometric capability. In three dimensions, boundary objects are constructed by either extruding two-dimensional cross sections or using triangular nite-element style surface meshes. Such meshes can be produced using any reasonable mesh generator from solid models generated in most CAD packages. This provides enormous exibility in specifying three-dimensional environments with which the particles interact. 2.3. Particle shape The particles are represented either as spheres or super-quadrics, which in a standard frame of reference are given by x a
m
y b
z c
=1
(3)
The ratios of the semi-major axes b/a and c/a are the aspect ratios of the particles and the superquadric power m determines the shape of the particle. For m = 2 elliptical particles are obtained. If additionally a = b = c, then the particles are spherical. As m increases, the shape becomes progressively more cubic, with the corners becoming sharper and the particle more blocky. By m = 10, the particle shape is essentially a cube (for aspect ratios 1). This is a very exible class of shapes, which varies continuously (m is not restricted to integer values). This allows plausible shape distributions to be represented and the shape can be made to change dynamically during simulation (for example, with abrasion as measured by the shear energy dissipation on the particle). The computational cost of simulations using super-quadrics is currently between 2 and 5 times that of using spheres. This shape description was introduced to DEM in two dimensions in [28] and in three dimensions in [23].
323
2.4. Model outputs The nature of the quantitative predictions that one wants to make, using the extensive state data available in DEM simulations, depends very much on the application and the questions one is seeking to answer. For industrial applications we typically calculate: Boundary stresses (for mechanical design and fatigue prediction, including coupling to FEM for structure response prediction). Wear rates and distributions (for estimating the lifespan of equipment and wear components). Accretion rates (for predicting accretion-induced blockage, changes in open area of screens). Collisional force distributions, collision frequencies, energy absorption spectra (for understanding breakage and agglomeration). Power consumption and torques (for equipment design). Flow rates and ow statistics (for summarizing characteristics of complex ows). Sampling statistics (to assess the accuracy of various sampling processes). Mixing and segregation rates (to assess the progress of intended mixing and de-mixing processes and to understand the degree of segregation and its effects on other processes where segregation and/or mixing are not intended). Residence time distributions (to assess the range of times that particles are exposed to various environments within a process, for example in granulation and in comminution). Axial transport rates (to assess the axial ow along cylinders (rotating and stationary) that are commonly used). For more details on the simulation method and on the data analysis see Cleary [20, 23].
3. TUMBLING MILLS Tumbling mills are large cylindrical vessels that rotate about their axes at reasonable fractions of the speed required to centrifuge the particulate material that is ground within. The grinding is produced by a range of mechanisms including high-speed impact, abrasion, chipping and nipping of small particles between large ones. Grinding is often augmented by adding various types of grinding media. There are several specic variants of the tumbling mill, which are distinguished by their size, the size of material fed to the mill and the type of grinding media. These include: Autogenous grinding (AG) mills, which are fed a mixture of large and small rocks with the large rocks acting as grinding media. SAG mills are the largest tumbling mills used (up to 40 or 12.2 m diameter) and are used for primary grinding, mainly in mineral processing, using a coarse feed (typically up to 250 mm diameter rock) coming from some arrangement of crushers. There is insufcient coarse rock to produce large impacts so the charge is augmented by steel grinding balls. These can make up between 30 and 50% of the volume of the charge. However, neither large rocks nor balls produce very many large impacts as discussed later. Rod mills are mills where the grinding media consist of large rods (often almost the length of the mill). These mills typically take a relatively ne feed material of perhaps 20 mm topsize, (see [23] for an example of DEM modelling of a rod mill). Ball mills are typically used for secondary grinding in mineral processing, taking feed from a SAG mill or an equivalent device. It is typically smaller than the SAG mill (38 m diameter)
324
and has higher ball loadings using smaller balls and with a ner feed. Ball mills are also commonly used in cement clinker grinding. These mills tend to be at the smaller end of the size range and have less aggressively designed lifters. In this paper, we present results for a large-scale model of a 36 (10.86 m) diameter SAG mill used in mineral processing and for a 4 m diameter (rst compartment) ball mill used for cement grinding for coarse and ne feeds.
3.1. SAG mill The SAG mill used in this modelling is a common design 36 or 10.86 m diameter SAG mill, which would typically be used for hard rock mineral processing. This mill rotates at 10 rpm (which is 78% of the critical speed at which the particles within the mill will centrifuge). Since collisions of the rock with the mill wear the mill surface, the mill is lined with a replaceable liner, typically made of high-grade collision- and abrasion-resistant steel. The liner serves a dual purpose of protecting the expensive mill shell and gearing from damage, and locking into the charge to lift it higher and deliver more energy into the charge to increase the rate of grinding. Here, we use a traditional 72 row (typically twice the diameter of the mill measured in feet) conguration of symmetric, close packed, steep face angle lifters (7 ). The overall ll level of the charge is 30% of the mill volume. The ball loading is 10% (of the mill volume) or around 1 of the 3 charge (by volume). The ball size distribution is given in Table I. A typical rock size distribution for a hard rock ore is also used and this is also given in Table I. The top size of the balls used is 125 mm and the top size of the rock is 200 mm. The rocks are correctly resolved down to 6 mm with smaller rocks neglected. Previously [14] we have established, by detailed comparison of various DEM models with experiments for a scale model SAG mill, that two-dimensional models have signicant quantitative errors. Therefore, we need to use some form of three-dimensional model. Since the mill geometry is symmetric along its length, we are able to model the charge motion in a 0.5 m wide axial slice of mill. A key difference between a three-dimensional slice model and a two-dimensional model is that the particles are represented as spheres rather than as cylinders and lock together in three-dimensional particle structures that are stronger than the simpler structures possible in two dimensions. Hence, although the overall charge shape is invariant in the axial direction, the
Table I. Rock and ball size distributions for a 1.2 million particle SAG mill model.
Balls 125 88 63 44 to to to to +88 mm +63 mm +44 mm +31 mm 30% 40% 20% 10% 200 140 100 70 50 35 25 18 12.5 8.5 to to to to to to to to to to Rocks +140 mm +100 mm +70 mm +50 mm +35 mm +25 mm +18 mm +12.5 mm +8.5 mm +6 mm 8% 12% 15% 15% 15% 10% 10% 10% 4% 1%
Copyright q
Int. J. Numer. Meth. Fluids 2008; 58:319353 DOI: 10.1002/d
325
three-dimensional particle interactions are important to predicting the correct ow dynamics. A second key need for using a three-dimensional model is that the energy transfer mechanisms during collision depend on the masses of the interacting particles. In two dimensions, the cylindrical particle masses vary as the square of the radius, whereas in three dimensions they vary as the cube. Hence, to obtain meaningful collision energy spectra from which we make deductions about comminution performance, it is essential to perform the simulations in three-dimensions. This 0.5 m slice is sufciently wide to properly represent particle motion in a long mill. This type of model is therefore suitable for predicting breakage, power consumption, belly-lifter wear and to understand the main ow patterns of the charge. If one were interested in axial transport then the full length of the mill needs to be modelled [7]. The current SAG conguration was previously modelled [10, 12] using a similar rock size distribution but with the rock size truncated at a much higher size of 25 mm. In this paper, we present simulation predictions using the more realistic 6 mm cut-off for the rock size distribution. This involves simulating 1.2 million particles in this system. Figure 2 shows the charge distribution looking along the axis of the SAG mill. The charge surface has a very clear bi-linear shape, rising slowly from the toe to the middle of the charge and then steeply up to the shoulder. Substantial cataracting streams depart from the lifters above the shoulder and fall in well-dened coherent bands to impact on the liner above the toe at around the 3 Oclock position. There is a central slow-moving band within the charge midway between the free surface and the mill liner with strong shear on either side. The cataracting stream reaches around 7 m/s as it hits the liner above the toe of the charge.
Figure 2. Slice model of a 36 SAG mill with 1.2 million particles coloured by speed with red being 19 m/s and dark blue being stationary.
326
Figure 3. Energy spectra for 36 SAG mill: (a) frequency distribution of different collision energies and (b) rate of energy dissipation at each collision energy.
The power prediction for this 0.5 m axial slice of the mill is 796 kW. This is very close to the average power draw reported in Cleary [10] of 778 kW for the case with a 25 mm cut-off for the rock size. This suggests that the lower cut-off has little effect on the power draw prediction. For the full 6 m axial length of the mill we would therefore expect a power draw of 9.54 MW. Figure 3 shows the way in which the energy consumed by the mill is dissipated. The left plot shows the frequency distribution of the various collision energies. The peak energies, corresponding to the largest rocks and balls falling at the top of the cataracting stream, are 50 J. However, in comparison with the numbers of rocks, there are very few of these high-energy impacts. The most common collisions occur for very low energies around 0.1 mJ with frequencies exceeding 1 million collisions per second. There are enormous numbers of collisions recorded at even lower energies (down to the minimum of 1010 J). Of more interest is the energy dissipated at each of these energy levels. This is shown in the right plot and indicates that energy levels of around 100 W are being dissipated by the highest energy collisions. The strongest energy dissipation rate occurs at around 0.15 J with around 5 kW dissipated by these collisions. The energy dissipated per energy level decreases slowly until around 0.1 mJ and then drops linearly and more rapidly, so that by the 106 J level only 1 W of energy is being dissipated, despite there being a frequency of a million collisions per second. Hence, although there are enormous numbers of collisions at these small energy levels, the total energy dissipated by these very gentle collisions is minimal. Comparing the distributions of shear and normal collision energies in Figure 3, it is clear that the shear energy levels are consistently higher than the normal ones. The shear energy dissipation rate (shown in the lower plot of Figure 3) occurs at 0.2 J, which is double that of the overall maximum dissipation rate and 20 times higher than the energy level for the maximum normal energy dissipation rate (at 0.01 J). The area between the shear and normal energy curves is the difference in the total energy consumption for impact and abrasion collision modes. Remembering that the vertical scale is logarithmic, the difference between these curves is signicant. This demonstrates that a large majority of energy dissipation in a SAG mill occurs by particles sliding over and past each other and not by direct head-on impact of the particles. An alternative approach to the spectra is to consider energy levels by tracking some selected particles in each size class. This approach has demonstrated that breakage in a single impact is highly unlikely for most particles larger than a few millimeters [12]. Hence, a detailed
327
Table II. Energy usage within the SAG mill.

Collision type Rockrock Rockliner Rockmedia Mediamedia Medialiner % Total energy in this type 51.9 1.2 38.3 8.3 0.3
investigation into incremental breakage was carried out. This work found a reasonably well-dened minimum energy below which essentially no impact damage occurs. For impacts with larger energy, the particle accumulates incremental damage, increasing the probability of breakage with each event [29]. Table II shows the distribution of energy dissipation by the different types of collisions occurring in the mill. Around 52% of the energy is dissipated in rockrock collisions with another 38% dissipated in rockmedia collisions (leading to both rock comminution and ball wear). A further 8.3% of the energy is dissipated in ballball collisions with some fraction of this leading to ball wear. Only 1.2% of the energy is lost in rockliner collisions, with some fraction of this contributing to liner wear. A further 0.3% of the energy dissipated is in medialiner collisions and contributes to ball and liner wear. DEM can be used to explore modications to liner geometry, mill operating parameters, feed and ball size distributions to help move energy dissipation from non-useful categories (such as liner collisions) to rockrock and rockmedia collisions, which should improve grinding rates for the mill. 3.2. Ball mills for cement grinding Ball mills are often used for grinding clinker as part of the cement production process. Often two chamber mills are used with the chambers separated by a grate. Typically, ner ball charges are used in the second compartment to give a ner grind than is possible in the coarser rst compartment. Here we simulate the rst compartment conguration of such a ball mill. The liners used for cement mills tend to be less aggressive than those used in mineral processing. Here we use a common wave liner in a mill with an internal shell diameter of 4.0 m. The liner consists of 36 liner plates, each with a plate length of 340 mm and a height of 65 mm, with a single wave per plate. The overall ll level used is 35% (of the volume of the mill) with the charge consisting of half balls and half clinker. The ball size distribution is uniformly mass weighted with a smallest size 50 mm (2 in) and a top size of 95 mm (3 3 in). Two different clinker sizes are used: 4 a coarse feed with particles in the range 2580 mm and a ne feed with particles in the range 1050 mm. The ne feed case is more typical of cement clinker grinding. A periodic slice of the mill is again sufcient to predict the dynamics of interest. The periodic slice length used was 250 mm. The mill rotation rate was 16.28 rpm, which is 76% of critical speed (based on a diameter of 3.9 m corresponding to the lowest point in the wave liner). For the coarse feed, the spring stiffness used was 1.5107 N/m and for the ne feed a value of 2.0107 N/m was required to give the desired average particle overlaps of 0.50.7%. The friction coefcient was 0.5 for all materials. The coefcient of restitution was chosen to be 0.8 for steelsteel collisions, 0.5
328
for steelclinker and 0.3 for clinkerclinker collisions. The number of particles was 11 200 for the coarse charge and 75 000 for the ne charge. The speed of a DEM code can be quantied using the Cundall number (Cu), which is the number of particle time steps per cpu second. On a single processor 3 GHz Pentium, these ball mill simulations produce Cundall numbers of 450 000. Figure 4 shows the charge distribution in the ball mill for the coarse and ne size feeds. The charge shape is very similar in both cases with the familiar bi-linear shape for the free surface rising from the toe to the inexion point near the centre of the mill and then steeply up to the shoulder position. Despite the difference in mill size and the size differences in the charge, this ow pattern is fairly similar to that of the SAG mill, but with the inexion point now closer to the mill axis and the angular difference between the two surfaces of the charge being less severe. It is mainly in the energies delivered to the particles that the mills behaviour differs. For both
Figure 4. Charge distribution in the ball mill with (left) coarse size distribution and (right) ne size distribution. The top row is coloured by the particle speed (with red being 7 m/s and dark blue being stationary) and the bottom row is coloured by particle class (with dark blue showing the balls).
329
coarse and ne feeds, the shoulder position lies in the range of 210220 , as measured clockwise from the 3 Oclock position. This range reects the difference in the timing of the disengagement of the charge from different points on the wave lifter. For the high point of the lifter, which is closer to the centre of the mill, the disengagement occurs a little earlier at around 210 , but for the more distant low point of the line, the disengagement occurs around 220 . The toe position for the coarse feed lies in the range of 35 to 40 , whereas for the ne feed the toe lies consistently in the tighter range of 35 to 36 . The ner feed is more able to cushion the large balls and this damps uctuations in the avalanching ow, leading to lower variability of the stopping position at the toe. The other visible effect of the ner feed is that the modest cataracting stream is more clearly dened because of the much larger numbers of much smaller particles in this region. The peak particle speeds are consistently around 6.57.0 m/s at the bottom of the cataracting stream. The pictures in the lower row of Figure 4 show the particles coloured by their type. The dark blue particles are the balls and they are fairly uniformly distributed throughout the charge. With such a large ball load, it is difcult for the balls to concentrate around the centre of recirculation due to size segregation driven by the centrifugal force (as sometimes occurs for SAG mills). The power draw predicted for the coarse feed is 146.9 kW/m of mill length or 1.028 MW for a 7 m long ball mill. This corresponds to a torque of 86.2 kN/m average torque (averaged over 6 s of operation of the mill after steady state has been achieved). For the ne feed, the power draw rises slightly to 148.2 kW/m of mill length, corresponding to an average torque of 87.0 kN/m. This represents a 0.9% increase when moving to the ner feed. We attribute this very minor change to the fact that the ner particles are slightly better able to pack the voids between the balls and this will lead to a slightly higher bulk density for the charge whose centre of mass will therefore move slightly closer to the mill shell, thereby increasing the mill power draw. This is a real but weak effect and will continue as the feed is made ner or ner product particles are included in the simulation. Figure 5 shows the selected energy spectra for the different types of collisions for both feed size distributions. The energy dissipated in every collision is tracked and the frequency distribution of the different collision energy intensities is calculated. This is done for the normal work (as a measure of impact breakage), shear work (as a measure of abrasion/attrition damage) and total work (measuring overall breakage). A variant of the spectra that we use here involves multiplying the energy of collisions by the frequency to give the energy dissipation rate for each collision energy level in the mill. This helps us understand where the 147 kW of power is being consumed within the mill. Note that the energy dissipation rates quoted here are for just the particles in a single periodic slice. For a full mill, these need to be scaled by the ratio of the mill to slice lengths. To start, we concentrate on the coarse feed case (left column of Figure 5). All the energy spectra have a log-normal-like distribution. There is a maximum energy that any particular type of collision can dissipate, typically governed by the maximum fall height of the heaviest particle involved in that type of collision. These events are very rare with frequencies typically around 0.1 per unit length of the mill. This is the equivalent of one per 10 s or once per two to three mill revolutions. Although rare, these events are fairly strong so the total energy dissipated by these is around 0.11.0 W. For lower energy levels the frequency of collision rises rapidly to a peak level of around 105 collisions/s/m at 7 mJ total collision energy. The highest rate of energy dissipation of 200 W occurs at a higher energy level of 0.17 J. From these median peaks, the collision rates decline slowly at rst and then more rapidly. The combination of declining collision energies and frequencies means that the energy dissipation drops off rapidly from the peak. Figure 5 shows the spectra for the four most important types of collision in the mill. The strongest energy dissipation
330
Figure 5. Energy spectra for (left) coarse size distribution and (right) ne size distribution in the cement ball mill.
331
rate is experienced for the rockmedia collision at around 150 W occurring at an energy level of 0.2 J. The next most important energy dissipation is in the rockrock collisions whose peak energy dissipation is 100 W occurring at 0.1 J. The next most important contribution occurs for mediamedia collisions with peak energy dissipation of 70 W at a collision energy level of 0.36 J. This energy contributes only to ball wear and results from a much smaller frequency of much more intense impacts (due to the large weight of the balls). Rockliner collisions are more intense with their peak at 0.4 J but these are much less common so their contribution to the overall dissipation is only 6 W. As was observed earlier for the SAG mill, the energy absorption is again dominated by the shear component. For all the collision types, the distribution of shear energy dissipation is very similar to the overall energy dissipation distribution with the peak shear dissipation rate occurring at between 1.0 and 2.0 times the overall level. Conversely, the normal energy dissipation is much smaller with its peak rate of dissipation being around 40% of that of the shear and occurring at collision energy levels around 10 times smaller. This again demonstrates that the energy dissipation is predominantly produced by particles sliding over each other rather than during head-on impacts. Sliding motion of the balls is the dominant mechanism leading to size reduction in the ball mill. Comparing the energy spectra for the ne and coarse feeds (Figure 5), we observe a range of changes. The maximum energy found for rockrock collisions declines by a factor of 2, which is smaller than one might expect since the mass of the largest rocks has decreased by a factor of 4 and that of the lightest rock by a factor of 10. This indicates that the presence of the media is having a signicant effect on the rockrock collisions by increasing the pressures involved. The highest energy dissipation rate is observed to move to a lower energy (from 0.1 to 0.01 J), which is about a quarter of what might be expected on the basis of the reduction in the product of the rock masses. The actual peak energy dissipation has increased slightly indicating that the distribution of energy dissipating collisions has become tighter. The maximum energies for rockmedia collisions have dropped by a factor of 2 (similar to the change for rockrock) but the highest energy dissipation collision level has only declined by a factor of 3, which is again around half the level that one might expect based on the rock mass change. The peak energy dissipation rate is unchanged in magnitude but the location has moved to lower-energy levels. The mediamedia spectra are broadly unchanged, with a modest reduction in the energy dissipation in the low energy tail and a slight narrowing of the spectra. This demonstrates that the ball component dominates the charge behaviour and has large impacts on the rock behaviour but the rock changes have very limited effects on the media. For the rockliner collisions, the maximum shear energies have declined by about one-third, whereas the normal energy has declined by a factor of 3. This indicates that direct liner impacts by larger rocks in the toe region have been substantially reduced in line with the reduction in the rock masses. The shear energy dissipation which is controlled by the pressure of the charge above the rock particles sliding along the liner only varies slightly since the ball mass is largely unchanged. Table III shows the distribution of energy consumption by the different types of collisions in the ball mill for the two feed sizes. These values are the integrated areas under the energy spectra curves. For the coarse feed about one-third of the energy is dissipated in rockrock collisions, with a large 46% dissipated in rockmedia collisions. The mediamedia dissipation is also signicant at 17.8%. The rockliner and medialiner energy dissipations (leading to liner wear) are very small. For the ner feed, the fraction of energy dissipated in rockrock collisions has jumped sharply to 51% with matching falls in the rockmedia and mediamedia dissipation of 8 and 9%, respectively.
332
Table III. Energy usage within a cement ball mill.

Collision type Rockrock Rockliner Rockmedia Mediamedia Medialiner Coarse feed (%) 33.5 1.9 46.2 17.8 0.5 Fine feed (%) 51.4 1.6 38.2 8.5 0.2
Table IV. Distribution of energy absorption among different particle classes and the mass and area efciencies of energy absorption for the ball mill with the coarse feed.
Class 1 2 3 4 Size range (mm) +50 +60 +40 +25 to to to to 95 80 60 40 balls feed feed feed % Total energy 41.21 8.44 22.82 26.33 % Mass 74.30 5.15 10.27 10.28 % Energy/% mass 0.555 1.639 2.221 2.562 % Area 37.27 7.84 21.66 33.24 % Energy/% area 1.106 1.077 1.054 0.792
Table V. Distribution of energy absorption among different particle classes and the mass and area efciencies of energy absorption for the ball mill with the ne feed.
Class 1 2 3 4 5 Size range (mm) +50 +40 +30 +20 +10 to to to to to 95 50 40 30 20 balls feed feed feed feed % Total energy 27.74 7.86 15.86 22.98 24.65 % Mass 74.29 3.86 6.43 7.71 7.71 % Energy/% mass 0.373 2.038 2.468 2.980 3.196 % Area 24.24 5.99 12.90 21.59 35.28 % Energy/% area 1.144 1.313 1.230 1.064 0.699
This would suggest that the media used here is much better at supplying energy to the ne feed material than to the coarse and also suggests that the coarse feed will produce higher ball wear rates. The liner energy dissipation components have also dropped by 25%, which would suggest that the coarser feed will produce higher wear rates on the liner and a shorter lifespan for the liner. This comparison demonstrates that although the overall power draw for the two feeds are almost the same, the nature of the collisional environment in the charge and how this energy is dissipated is substantially different with the ner feed being better able to absorb energy with perhaps 56% absorption leading to better grinding than for the coarse feed with energy absorption of only 42%. On this fairly crude energy calculation one might anticipate a one-third increase in grinding for the ne feed (which should result in some combination of higher throughput and a ner grind). To help understand the energy absorption by the rock, we look now at the energy dissipation in each of the size classes of the charge. Tables IV and V show the distribution of energy absorption for the different particle classes and the mass and area efciencies of energy absorption for the
333
coarse and ne feeds, respectively. The actual fraction of the energy dissipated is heavily dependent on the amount of that material in the mill. We therefore look at two measures of the energy absorption efciency of each of the charge components, one based on energy absorption per unit mass and the other per unit area. From Table IV we see that the balls absorb a much lower fraction of energy per unit mass than average (around half). The coarsest fraction of rock absorbs around 1.6 times the average. This continues to increase to 2.56 for the smallest of the rock classes. Hence, the energy is dissipated disproportionately to the mass with the ner sizes absorbing more. There are a number of contributors to this including the difference in density of the rock to the steel, the different coefcients of restitution for collisions of different combinations of material and the different numbers of contacts that each particle class has with neighbours. Looking at the energy absorption per unit of surface area we nd a much better correlation. The ball and rst two rock classes have area efciencies of a little over 1, meaning that they absorb energy broadly in proportion to the surface area of each class. This behaviour was conrmed in more detail in small-scale autogenous mill testing reported by Djordjevic et al. [30]. Relating the particle wear rate to the DEM estimate of frictional energy works quite well in these energy ranges. The smallest rock fraction though has a much lower efciency of around 0.8. This probably relates to the absence of smaller rock classes that make it harder to load the surface of this smallest class. Table V shows the energy absorption for the ne feed. The trends are similar to that observed for the coarse feed. The mass efciency for the balls has dropped to only 0.37 while the largest rock class has a mass efciency of 2.0, which is similar to that of the matching size class in the coarse feed. The mass efciency continues to rise with decreasing size to a peak of over 3 for the 1020 mm class. The correlation with area is again much better with the ball efciency being around the same level of 1.1 as was found for the coarse feed. The largest rock class has an area efciency of 1.3, meaning it is 30% better at absorbing energy per unit of surface area than average. The area efciency declines very slowly with decreasing size (as it did for the coarse feed). The smallest size class has an efciency of only 0.7, which is fairly similar to the behaviour of the smallest class in the coarse feed. This further suggests that this might relate to the absence of smaller size classes.
4. GRINDING TABLE Another grinding device, commonly used for cement grinding, is a grinding table or sometimes called a roller mill. In this particular example, a 4 m diameter table rotates around its centre at 20 rpm. It has a lip around the outside to prevent particles being thrown off. Four 1.2 m diameter rollers are positioned just above the table. They counter-rotate so that the peripheral roller speed matches the table speed at the lowest point. Particles (with diameters 2575 mm and density 3000 kg/m3 ) are fed onto the centre of the table at a rate of 450 tph. Upon contacting the table, they are thrown outward. The lip on the edge of the table traps the particles, leading to the formation of a bed of particles on the table. Many of the outward owing particles enter the region in front of the rolls, are drawn in by the motion induced by the roller rotation and are trapped between the table and the rollers where they are crushed due to compression. In this case, we impose a lower limit on the fragment particle size of 5 mm to ensure that the particle numbers remain computationally reasonable. The largest particles are typically the weakest and have a breakage force limit of 15 kN rising to 60 kN for the ner particles that have less remaining aws.
334
Figure 6. Grinding table rotating at 20 rpm with four equi-spaced 1200 mm rollers and a central feed shown obliquely from above with particles coloured/shaded by speed.
Figure 6 shows the ow (where the particles are coloured by speed) in such a mill once the bed depth has stabilized and the system is in approximate steady state. The incoming speed of the central feed stream is relatively fast. The particles slow as they strike the accumulated particle bed on the surface of the grinding table. As they are pushed outward, they accelerate until they encounter the rolls, are trapped and are dragged under the rolls. Here they are crushed by compression. Some high-speed fragments y out from the rollers. Each roll generates a bow wave effect with a mass of particles building up higher in front of the roll than is found throughout the rest of the bed. Some of these particles are pushed sideways and ow around the roll forming the sides of the bow wave and some are pushed in front of the roll. The pile of material pushed along in front of the roll needs to generate enough back pressure in order to force particles under the roll. For this four-roll case, the bow wave from each roll is directed under the middle of the following roll. This would appear advantageous for particle breakage. The DEM model used here includes a simple but effective dynamic breakage model [10] where particles reaching a critical breakage force (as determined by the DEM contact force model) are replaced by a suitable assembly of daughter particles that are then free to move independently. In this case of spherical particles fracturing into spherical progeny, a geometric packing algorithm is employed. A primary fragment is chosen and placed in the parent. The next biggest particle that can t is added along the same axis as the rst. Then layers of maximally sized ner particles are added into the biggest gaps in the previous layer. This is performed repeatedly with up to ve levels of progressively ner fragments added. For more details on the progeny construction method, see [10]. For the grinding table, this means that the daughter fragment particles can then be re-trapped further under the roll and be broken again as many times as is needed for them to pass through the gap between the table and the rolls. Figure 7(a) shows the bed of particles on the table from below with the particles coloured by size (red for the largest and dark blue for the smallest fragments). The central feed and the outward ow of this coarser material (light blue to red particles) are visible in the middle sections of the table. The bow wave of coarser bed material piled up in front of and owing around each of the rolls is also clearly visible. The compression region under the roll can also be observed with the larger particles able to approach only within several diameters
335
Figure 7. Grinding table with particles coloured/shaded by size shown from: (a) underneath and (b) looking through the roll at the bow wave/zone of compression breakage under the roller.
of the middle of the roll before the rolltable distance becomes smaller than the particles size and it is fractured. There is a clear gradation of size as the particles move further under the rolls. Finally, only a layer of dark blue product particles is observed as the particles reach the minimum size enabled for this model. These ne fragments then pass under the roll and either join the concentrated band of nes trapped by the lip around the edge of the table or ow over this packed bed of nes and are discharged from the table as product. The discharge material commonly includes both unbroken feed and product and then needs to be separated by some type of air classication system with the coarse fraction returned to the feed stream in a closed circuit. At this stage, we do not seek to model either the ultra-nes or the coupled air ow. Figure 7(b) shows the breakage region and bow wave in front of a roller from a view looking through the transparent roll with the particles again coloured by size. The restriction of the ow under the roller causes a bow wave of densely packed particles to form in advance of the roller. Many of the particles in the bed are pushed sideways around the roll, some being forced over the lip of the table to be discharged and some radially inwards onto the table. Some particles are dragged under the roll and rapidly crushed since the gap between the roll and the table decreases as the particle is dragged further under the roll. The nes (coloured dark blue) are free to ow out from the back of the roll. Figure 8 shows the energy spectra predicted for the grinding table. The normal energy dissipation (typically associated with impact breakage) is shown in black and the shear energy absorption is shown in grey. It is clear that the shear energy is weighted to higher levels than the normal energy. Measured over all collisions, the normal impact energy accounts for 35% of total energy consumption while the shear dissipation accounts for 65%. The shape of the energy spectra is noticeably different to those of the ball mill (see Figure 5) even when considering the different minimum energy thresholds used in the plotting. This indicates that there are important performance differences between the two types of mill even though they are often used for the same cement
336
Figure 8. Energy spectra for the grinding table: (a) over all collisions; (b) from rock-grinding table and rolls; and (c) from rockrock collisions.
applications. The energy spectra for the collisions between the particles and the grinding surfaces (tables and rolls) are very similar to the spectra for all collisions. The spectra for the particle particle collisions though are quite different. The collision rates are much lower as are the average levels of energy dissipation (around 10 times lower) but there is also a long tail of high-energy events. These occur between particles in the compression zone as they are loaded under the rolls. The rate of these collisions is low but they are an important part of the breakage and energy consumption process. Note that the smallest particle size can become jammed under the rolls in an unphysical manner because of its inability to break and so can absorb unrealistic amounts of energy. Hence, ideally the minimum particle size used in the model should be less than the separation of the rolls from the table (i.e. less than the bed depth under the rolls). This will allow more accurate prediction of the energy absorption by the smallest class and therefore more accurate predictions of the power draw. Figure 9 shows the distribution of energy absorption across all the breakable size classes. The largest particle sizes absorb only around 5% of the energy (despite being around 10% of the particle mass). The rate of absorption rises almost linearly for particle classes down to 40 mm diameter. At this size, the particles absorb around 10% of the energy and represent 10% of the particle mass. For smaller sizes, the energy absorption rises sharply, with 27% of the energy absorbed for the smallest of these size classes. The particle sizes used in this grinding mill simulation are
337
30.0 25.0 Energy absorption % 20.0 15.0 10.0 5.0 0.0 30.0
40.0
50.0
60.0
70.0
Particle class diameter (mm )
Figure 9. Distribution of energy absorption across all the breakable size classes.
very similar to the ones used in the coarse cement ball mill discussed earlier. The distribution of energy absorption with particle size is remarkably similar for the two machines (see Table IV for the ball mill results) with the % energy/% mass being around 0.5 for the largest particle sizes and increasing to around 2.5 for the nest. This suggests that the rate of energy absorption for the grinding table is again controlled by the total surface area of each size class. Figure 10 shows the rate of fracture of the particles and the rate of product creation by the mill. The overall rate of fracture is quite variable with high-frequency uctuations, depending sensitively on the specic particles being trapped under the rolls. The overall rate varies between 250 and 450 fracture events per second. The fracture rates are smallest for the largest particles (since there are relatively few of these per unit mass). Figure 10(c) shows the fracture rate for the largest size with an average rate of 2 fractures/s. The breakage rate increases with decreasing size. For class 7 (5562 mm) the rate has increased to 5 fractures/s. For class 4 (3036 mm) the much larger numbers of particles available mean that breakage is more continuous, albeit still with wide uctuations and with an average of around 50 fractures/s. For the second smallest class, the fracture rate has risen to 130 fractures/s. The fracture rates are clearly dependent on the number of particles in each size class and also on their breakage strength. DEM is able to predict not only the energy consumption, the pattern of usage of that energy, but also the breakage outcomes. This work demonstrates that DEM models can, in principle, make predictions of the resident size distributions and product size distributions of mills. Determining how realistic these predictions are and rening the dynamic breakage models to improve them are the subject of ongoing work. For the current grinding table conguration, with its relatively light bed loading and for the material hardnesses chosen, the product generation rate predicted is about 11 kg/s or about 40 tonnes/h. The power draw measured was 65 kW with 25 kW being supplied by the grinding table and 10 kW supplied by each of the four rolls. This size of table can be run with much higher bed loads so as to generate product at rates up to 400 tonnes/h. Scaling up the power similarly would lead us to expect a power draw of around 650 kW. This is around 6 times lower than the typically installed power. This leads us to expect that truncating the product size distribution may lead to under-prediction of power draw for these types of compressive mills since much of the ner breakage occurring in
338
Figure 10. Particle breakage predicted for the grinding table: (a) overall rate of fracture events for all particles; (b) the rate of product generation (given by particles in the smallest class); (c) biggest class of particles (class 10); (d) class 7; (e) class 4; and (f) class 2 which are the smallest particles allowed to fracture by the model.
the real mill is not well represented in the model with this choice of minimum particle size. This suggests that quantitatively accurate prediction of all mill outputs may require resolving more of the ner end of the size distribution. This is becoming increasingly feasible as computer power increases. More investigation is needed to clarify these issues.
339
5. STIRRED MEDIA MILLS Stirred mills are becoming increasingly used for ne and ultra-ne grinding. This technology is still poorly understood when used in the mineral processing context. This makes process optimization of such devices problematic. Three-dimensional DEM can again play a part in understanding the fundamentals of particle size reduction and media behaviour inside these mills. In this section, we present examples of DEM simulation for two ne-grinding mills. 5.1. Tower mill A tower mill has a vertical cylindrical shell and employs a vertical double helical steel screw agitator located along the axis of the mill shell. The mill is substantially lled with grinding media (typically steel or ceramic balls). The screw agitator stirs the media while simultaneously lifting and circulating it throughout the mill. Fine feed material is introduced at the top surface of the charge and is circulated throughout the mill where it is ground by being crushed or sheared between media particles. The rotation of the screw creates centrifugal forces that compress the charge radially outwards. This is balanced by gravity. The screw rotational speed is limited by the need for the vertical force resulting from the centrifugal compression to remain smaller than gravity so that the charge is not lifted out of the mill, (see [31] for more details). The rotational speeds and therefore the energy intensity produced within the tower mill is much higher than for a ball mill (with a speed differential here of around a factor 6) and so a tower mill can typically produce a much ner product. The nature of the ow pattern of the media and manner of energy dissipation within the mill though is not well understood. Here we model a tower mill lled with dry ceramic media. A coefcient of restitution for the media of 0.3 is used (to take account of the ne powder being ground that will occupy the interstices between the balls in the mill). The friction coefcient was chosen to be 0.5 for mediamedia and medialiner collisions. The ne feed material which is intended to be ground by the mill is too small to be able to be included directly in the DEM model and is omitted. The grinding media, which are typically 12 orders of magnitude larger, can be fully included. A spring constant of 10 000 N/m was used giving average overlaps of about 1%. Key geometric details and operating parameters for the tower mill are given in Table VI. The details of the ball charge are given in Table VII. The mill is lled nearly to the top and contains around 30 000 media particles. Figure 11 shows the distribution of the media in the tower mill once it has reached the steady state. The mill and charge have been cut away to allow us to see inside. Figure 11(a) shows the media coloured by speed. There is a solid core of media, occupying a large fraction of the mill vessel, which moves with the screw agitator and which is coloured green. Near the outer edge of the screw (which is the most rapidly moving part of the screw and the part that encounters the highest collision intensity) the media particles are driven faster, with particles achieving double the average core speed (and coloured yellow to red). Around the central moving core is a fairly clearly dened layer of slow-moving particles that are adjacent to the cylindrical mill shell. Frictional drag slows these particles. The sharp change in speed from the slow outer layer to the inner core means that there is signicant shear in the narrow transition region between the blue and the green particles, which will lead to signicant abrasive wear. Figure 11(b) shows the same particle distribution but now coloured by the amount of energy that each particle has absorbed from the normal component (head-on impact) of the collisions that they have experienced. There is a clear gradient in the energy absorption with the highest levels (which
340
Table VI. Key details of the tower mill charge.

Number of particles Particle density Total particle mass Spherical particle size 29 870 2700 kg/m3 58.4 kg 14 to +9 mm
Table VII. Specications of the tower mill used in the DEM simulations.
Rated power Angular speed Chamber volume Stirrer Chamber diameter Chamber height Stirrer diameter Pitch 1.5 kW 100 rpm 39 l Double helical steel screw 0.24 m 1.2 m 0.14 m 0.09 m
Figure 11. Three-dimensional ow pictures showing a clipped section of the tower mill. Media particles are coloured by: (a) speed with red representing a speed of 0.5 m/s; (b) normal energy absorption rate with peak values (red) of 7 mW; and (c) shear energy absorption rate with peak values (red) of 20 mW.
341
will correspond to regions of high-impact damage for the ne feed material that occupies the space between the balls in the real mill), with the highest levels found at the top of the mill and the lowest levels at the bottom. This occurs because the screw agitator is actually a very effective transport device and lifts the media in the core of the mill upwards, producing higher pressures at the top of the mill and therefore more intensive impacts. Figure 11(b) shows the distribution of the shear energy absorption in the mill, which will correspond to the intensity of abrasive size reduction of the feed material. There is still a vertical gradient present with the highest values found at the top of the mill, but it is much weaker than for the impact (normal) energy absorption. The reason for this is that the Coulomb limit on the frictional force is proportional to the normal load. The shear energy absorption is a product of the shear force and the shear velocity. The vertical gradient in the normal force therefore creates a matching gradient in the shear component. However the shear velocity also contributes. Looking at the speed distribution (Figure 11(c)) we see that the transition from slow-moving blue material around the shell to fast-moving core material is much sharper at the bottom of the mill than at the top. Therefore, the shear rate is higher at the bottom than at the top. This increases the shear energy dissipation at the bottom of the mill. The mild vertical gradient in the shear energy absorption is therefore a balance between the normal force, which increases with distance from the bottom and the shear rate which decreases. One way of characterizing the ow in this type of mill is to calculate averaged one-dimensional proles of the velocity components, as a function of mill height and radius. These are shown in Figure 12. The velocities are collected in radial and vertical slices of the mill to calculate a mean velocity and standard deviation for each radius and height. The mean radial velocity is statistically zero over most of the mill height. Near the base of the mill, it has a small but meaningful negative value, indicating that the particles are owing from near the outer mill chamber radially inwards towards the central screw. Near the free surface, the radial velocity becomes signicantly positive as particles ow back from around the screw, outwards towards the outer mill shell. The standard deviation is a measure of the variability of the velocity from its means and of the extent of uctuations in the ow. The standard deviation is much larger than the peak values of the radial velocity and is constant with height. The bounding envelope, dened by the limits of the vertical lines, indicates a region of phase space which is symmetric around zero that the particle radial velocities are likely to occupy. There is no meaningful mean radial velocity at any radius. This means that in any vertical column of particles there is equal radial movement in both positive and negative directions. The outside of the screw makes the largest contribution to driving the ow and also drives the uctuations in the ow. The mean tangential or swirling velocity is low at the base of the mill and increases steadily over the bottom quarter of the mill. The swirling motion is then essentially constant over most of the height of the screw. There is a strong drop in the particle swirl in the top surface layers of the charge where the particles are more mobile and move out towards the mill shell. The variability of the swirl speed is also quite constant over most of the mill height, except in the surface layers. The tangential speed variation with radius is very strong. It is zero for radii less than or equal to 0.02 m since this is the radius of the core of the screw and particles cannot be found at these positions. The swirl is low near the core of the screw (r >0.03) and increases strongly and monotonically with mill radius up to a peak at the edge of the screw (r = 0.07). It then decreases almost linearly with radial distance out towards the mill chamber wall. There is a reasonable degree of slip experienced between the charge and the wall at about a third of the peak tangential speed. The mean axial velocity is statistically zero at each height and the amount of variability in the instantaneous axial velocities is also constant with height. Since the screw and chamber
342
0.1 0.05 Vr (m/s) 0 0 -0.05 -0.1
0.1 0.05 Vr (m/s) 0 0 -0.05 -0.1 Height (m) Mill Radius (m)
0 Vtang (m/s) -0.1 -0.2 -0.3 -0.4 -0.5 Height (m) Mill Radius (m)
0.2 0.1 Vz (m/s) 0 0 -0.1 -0.2 Height (m) Mill Radius (m) 0.05 0.1 0.15
0.5
0.05
0.1
0.15
0.2 Vtang (m/s)
0.05
0.1
0.15
0 0 -0.2 0.5 1
-0.4
0.2 0.1 Vz (m/s) 0 0 -0.1 -0.2 0.5 1
Figure 12. One-dimensional proles of media velocity (Vr, Vtang, Vz) versus mill height and radius for the tower mill at t = 10 s. Condence limits of one standard deviation show the variance in velocity. The radius of the screw is 0.07 m and the chamber radius is 0.12 m.
cross sections do not vary with height, this means that in any axial slice there is, on average, as much material moving down the mill as there is moving up the mill with the screw motion. In contrast, the mean axial velocity varies strongly with radius. It has intermediate values within the screw, which increase moderately with radius to a peak transport rate at the outer edge of the screw. The mean axial velocity then drops almost linearly to a point just inside the mill shell. From a radius of around 0.09 m, the mean axial speed is negative, indicating that there is strong coherent down ow in an outer annular region that is about 25% of the radius of the mill chamber. The average speed of the outer down ow is much slower than that of the inner up ow because the mass ow rates must be equal (since this ow is in equilibrium) and the outer annulus has a much larger large cross-sectional area. The downward mass ux through this outer cylinder therefore matches the upward ux in the inner core.
343
Combining the interpretations of the radial and axial mean velocity elds, it is clear that there is a strong global recirculation pattern within the tower mill driven by the screw motion. Particles are transported upwards by the screw in a central cylindrical core. As they approach the free surface, they begin to ow outward towards the mill shell. Once they are further out than 75% of the mill radius, they begin to ow somewhat erratically downwards in an annular down ow region around the outside of the charge. At the bottom of the mill, the particles ow back in towards the screw where they are again picked up and transported upwards by the screw. To understand the nature of the energy utilization in the tower mill, we calculate one-dimensional proles of the energy absorption rate as functions of mill height and radius. The distribution of energy absorption by the media indicates where in the mill the particle size reduction of feed material is occurring. Figure 13 shows that mean energy absorption rates for normal, shear and total energies are largely insensitive to axial position in the mill, except right at the bottom of the mill
Rate of Norm E Abs(mW)
4.0
Rate of Norm E Abs(mW)
4.0
2.0
2.0
0.0 0 0.5 Height (m) 1
0.0 0 0.05 0.1 0.15 Mill Radius (m)
Rate of Shear E Abs (mW)
20.0 15.0 10.0 5.0 0.0 0 -5.0 0.5 Height (m) 1
Rate of Shear E Abs (mW)
20.0 15.0 10.0 5.0 0.0 0 0.05 0.1 Mill Radius (m) 0.15
20.0 15.0 10.0 5.0 0.0 0 0.5 Height (m) 1
Rate of Total E Abs (mW)
Rate of Total E Abs (mW)
20.0 15.0 10.0 5.0 0.0 0 0.05 0.1 0.15 Mill Radius (m)
Figure 13. One-dimensional proles of media energy absorption rates versus mill height and radius for the tower mill collected over only two revolutions when in steady state. Condence limits of one standard deviation show the variance in velocity. The radius of the screw is 0.07 m and the chamber radius is 0.12 m.
344
where they are lower because the screw impeller does not reach down to the oor of the mill. The energy absorption by the media declines reasonably strongly over the top third of the mill. The shear energy absorption in the bulk of the mill also shows a modest increasing trend with increasing height. The mean energy absorption rates all have strong radial dependence. They are relatively low at small radii near the shaft of the screw since the centrifugal forces are smaller here. The energy absorption rates increase rapidly with radii and reach peak values at about 0.085 m, which is moderately outside the outer diameter of the screw and well beyond the location of the peak axial velocity. This is the region of highest shear between the upward- and downward-directed streams of particles. Energy absorption then decreases with increasing radius but is still signicant all the way out to the mill chamber wall. The average power draw for the tower mill was predicted to be 0.27 kW. For more details of the media motion and energy consumption in this tower mill see [31] and for details of mixing and transport [32]. 5.2. Isamill The Isamill is a horizontal stirred mill with multiple grinding discs on a central rotating shaft. Since it has a closed grinding vessel and does not rely on gravity to keep the mill charge within the mill, signicantly higher rotational speeds can be used for this mill. Speed increases of factors of 1020 are possible. The much higher speeds produce centrifugal forces that are 100400 times higher, leading to much higher pressures, much higher shear rates and much higher levels of energy consumption for breakage. This type of mill can be used for ultra-ne grinding with product sizes down to a few microns possible. These very ne products can often be produced at favourable energy consumption rates if well tuned to the material being ground. The stirring agitator used is a shaft with several discs attached at regular intervals along the shaft. These extend most of the way to the outer cylindrical shell. Typically the discs have 46 holes (passing completely through the disc) with circular or elliptical shapes regularly spaced around the discs at about half the radius of the disc. The mill is lled to 75% volume with a charge composed mostly of grinding media (which could be steel or ceramic balls, sand or ne slag). The discs partition the charge into distinct zones or compartments and the disc rotation generates recirculating motion of the grinding media within and between the compartments. Fine feed material is added at one end under high pressure and is transported along the mill, being progressively ground by the media and then ideally discharged at the other end. The media is retained in the mill by a centrifugal separator. Here, we model just one compartment around one of the discs of a laboratory scale Isamill using periodic boundary conditions in the axial direction. The mill details are given in Table VIII. The mill has a diameter of 337.5 mm and the periodic slice is 200 mm (being the centre to centre spacing of the discs along the shaft). Details of the grinding media and the feed ore are also given in Table VIII. We wish to understand the distribution of ore within the charge so feed particles need to be included in the simulation. The feed ore is represented as 1 mm particles which are over-sized compared with the actual 0.2 mm feed of interest. Since these particles are small compared with the 6 mm grinding media, this is expected to give reasonable predictions for the spatial distribution of feed. We would not expect this approach to be reasonable for predicting the breakage of the feed particles by the media. Figure 14 shows a side view of the periodic slice of the Isamill. The mill is sectioned to allow viewing inside the charge. The 6 mm media are shown as blue and the 1 mm ore particles are shown
345
Table VIII. Isamill conguration, charge and operating conditions.

Mill shell radius Grinding discs 168.75 mm 145 mm radius disk 25 mm thick 30% cross-sectional area open to ow (provided by 4 circular holes) 25 mm 6 mm balls with density 2700 kg/m3 80% v/v 1 mm and 5100 kg/m3 20% v/v 75% 1315 rpm 20 m/s
Shaft diameter Grinding media Media volume of the charge Ore feed Ore fraction of the charge Charge volume ll Shaft/disc speed Disc tip speed
Figure 14. Media (blue) and ne feed material (red) distributions at steady state in an Isamill. The feed material segregates rapidly to the outside lling the interstitial gaps between the media particles.
346
Figure 15. Grinding media (blue) and feed ore (red) distributions in different slices of the Isamill, showing: (a) slice at the feed end just before the grinding disk; (b) in a slice aligned with the grinding disk; and (c) in a slice on the discharge side of the grinding disk.
as red. The pressure gradient on the charge (implemented as an axial body force) pushes particles against the left side of the disc and leaves a matching void on the right side. The media densely packs the mill in all locations except in this void. The feed material distribution is more complex. The centrifugal force is extremely high and pushes these particles outwards. Since the feed is so small compared with the media, it is relatively free to migrate through the void spaces between the close-packed media particles until all the pores are densely lled, starting from the mill shell and moving inwards. Once in steady state, the distance that the ore particles penetrate in towards the axis of the mill is then determined by the feed ore volume fraction. A full mill can be pictured by placing several copies of this periodic slice along the axial direction. The periodic nature of the computational domain does not require the spatial distribution of charge to be uniform in the axial direction within an individual periodic region. The distribution along the mill is periodic (not
347
uniform) with successful dense zones on the feed sides and void zones on the discharge sides of each of the grinding discs. Figure 15 shows axial slices of the mill charge at three locations along the axis of the mill. The rst one lies in the feed side of the mill compartment and shows densely packed media reaching in to the location of the shaft. The ore densely lls the voids around the outside of the mill and extends inwards for a reasonable distance. The interface is not smooth at the edge of the ore since the recirculation of the media (which is comparatively slow compared with the speed of the nes migration) causes some mixing in the interface region. The second slice is at the location of the disc. The media and ore are both conned to the narrow annulus around the outside of the disk. The locations of the four circular holes in the disc can be identied by the presence of some media and ore particles that are currently located in the openings, being moved around circumferentially by the disk. The third slice shows the charge distribution part way along the discharge side of the periodic volume. The media still densely lls the space between the shaft and the shell, but the amount of feed material is much reduced because of the restriction to axial transport that the disc represents. The ore particles are again densely packed from the shell inwards but now extend only about half the distance towards the shaft. The interface is also more diffuse than on the feed side. It is also useful to understand the nature of the particle motion in different regions of the mill charge. To do this, Figure 16 shows individual trajectories for selected media and rock particles. The trajectory thickness is drawn to match the diameter of the particle (hence, thick curves are for media particles and thin curves are for the ne rock particles). Figure 16(a) shows a media particle between discs at a moderate radius from the shaft. This particle executes close to circular motion with some axial motion (which appears to be of the random walk type rather than a coherent axial ow). Figure 16(c) shows a similar trajectory for a particle at a large radius near the outer shell. It again follows an almost circular path. Figure 16(e) shows a media particle at a similar initial radius but starting closer to the disc. The particle follows a clear spiral pattern, moving with a reasonably steady axial speed past the grinding disc and into the next compartment between discs. Figure 16(b) shows a feed particle starting in a similar location to the media particle in Figure 16(e). It also moves in a coherent spiral pattern past the outside of the grinding disc into the next compartment. This shows that there is a coherent axial ow along the mill at radii larger than the discs. Figure 16(d) shows another trajectory for a feed particle that starts at a smaller radius. This particle moves in a spiral pattern but with increasing radius. This demonstrates that the axial ow extends some distance into the space between discs and that particles are able to migrate outward and then spiral past the grinding disk. Figure 16(f) shows a feed particle that starts at a similar radius but further from the disk. This particle travels on an essentially closed circular path with small random walk variations in the axial and radial directions. Many cycles of the rotation are shown to establish that the path is a stable closed trajectory. Figure 17 shows multiple trajectories of some other media and rock particles within the Isamill. In the top picture, one media and one feed particle are shown together. The media particle lies on a stable closed trajectory, whereas the feed particle trajectory spirals in the axial direction with the radius of the circulation decreasing once the particle has passed the centre disc and nally passing through one of the holes in the next grinding disk to the left. This demonstrates that small changes in the particle position can lead to signicant differences in the trajectories and also demonstrates the ow of particles through the grinding disc holes. Figure 17(b) shows some additional media trajectories that spiral past the outside of the grinding disk. The range of variation of the particle trajectories is substantial and occurs for only small changes in the particle starting positions. This demonstrates that the velocity eld for the charge is spatially complex and presents signicant
348
Figure 16. Selected individual trajectories of particles within the Isamill (left) media trajectories and (right) rock trajectories.
challenges for visualization and interpretation. Trajectories allow the phase space to be sampled but they are not suitable for detailed analysis of the entire velocity eld. It is not clear at this point if the velocity eld is steady or if there are important transient variations associated with these various types of motion. If the velocity eld is steady then spatially averaging the velocities onto an Eulerian grid is one approach that can be taken to help in analysing the velocity eld. Figure 18 shows the energy spectra for this Isamill conguration. They appear to be skewed normal distributions. The collision energies range from 107 up to 2000 J. Compared with the ball mill (in Figure 5) the peak energy levels are increased by a factor 200, demonstrating the signicant increase in energy intensity obtainable for such high super-critical mill rotation speeds. The minimum collision energy tail also extends down to much lower energies (107 J), indicating
349
Figure 17. Multiple selected trajectories of media and rock particles within the Isamill.
that in regions of the mill the conditions are relatively benign with many very weak collisions occurring. Considering the rockrock energy spectra, we nd that for the shear energy, dissipation has a similar shape and energy dissipation level to that of the overall spectra, but signicantly reduced normal impact energy levels (by a factor of around 200). In contrast, the normal and shear energy absorption spectra are quite similar for the mediarock collisions. Looking carefully at the normal energy spectra for all collisions, we observe a clear double peak. From the sub-spectra, it is obvious that the rst peak arises from the rockrock collisions and the second higher energy peak arises from the mediarock collisions. The rockliner spectra cover a similar range of energies but the distributions are more symmetric with the modal peak moving from around 1 J down to around 0.5 J. The frequency of the highest energy collisions also decreases from around 600 collisions/s to only 10 collisions/s, leading to more than a factor of 100 lower overall energy dissipation for rockliner collisions. This indicates that rock collisions with the mill chamber and grinding discs are unimportant to breakage. The mediamedia spectra are very similar in structure to the mediarock spectra with the normal and shear energy dissipation being reasonably similar. The modal peak energy level is around 4 times higher for the mediamedia collisions but the collision rates are around 50 times lower. This means that the total mediamedia energy dissipation is around 10 times lower than that of the rock media. The higher energy levels of these collisions are understandable given the larger masses of the media particles. The signicantly reduced collision rates for the mediamedia contacts reects the effectiveness of the ner feed particles in separating the media particles. Each media particle is typically densely surrounded by a collection of small rock particles that limit direct mediamedia contact. This is a substantial part of the reason why such a stirred mill is able to be effective as a grinding device. The small interstitial particles are very effective at isolating the media particles and ensuring that the vast bulk of the energy is dissipated in collisions involving feed particles leading to their comminution and thereby reducing the media wear rates as well. If this was not true then the majority of the energy would be expended grinding the media rather than the feed. For this mill conguration, the energy wasted in direct mediamedia grinding appears to be around 10 times lower than the energy expended in collisions involving feed material. This also demonstrates
350
Figure 18. Energy spectra for the Isamill.
the importance of including the feed material in such simulations as they can radically alter the pattern of energy utilization in the mill model. Finally, we observe that the medialiner collisions have a symmetric prole much like the rockliner ones but with a higher model energy peak (due to the larger particle mass) and with a much lower collision rate (due to lower numbers of these particles and their insulation from the liner by interstitial feed material).
351
6. CONCLUSIONS DEM modelling has now reached a level of maturity where simulations of the order of 100 0002.0 million particles, in realistic geometries, are feasible and can provide quantitative predictions of power draw, energy utilization, fracture and wear for screens, crushers and a broad range of mills. Previously, it has also demonstrated that two-dimensional modelling (of discs) is of almost no use since it cannot quantitatively capture the charge behaviour. Using spheres to model particles captures a great deal of the process in tumbling mills. Detailed DEM analysis of AG and SAG mills has very clearly demonstrated that breakage in these devices is almost always the result of multiple collisions. Energy absorption by each size class of the charge is broadly proportional to the surface energy of the particles. Energy dissipation is predominantly from the shear components of collisions as particles mostly slide over each other rather than having head-on impacts. The energy spectra for the cement mill show that the use of ner feed strongly changes the way energy is consumed with sharp rises in the energy consumed by direct rockrock collision and matching sharp reductions in rockmedia and mediamedia collisions. Research into modes of breakage within these mills is continuing [29, 30, 33]. Adjusting operating conditions to reduce waste of energy (in collisions which cause only elastic particle loading) offers an important avenue for process development. Liner wear evolution within AG/SAG mills can also be predicted with reasonable success using DEM. Modes of breakage and transport within stirred mills is now also an area of active research. The addition of realistic slurry behaviour via Smoothed Particle Hydrodynamics (SPH) offers substantial opportunities for the future as many aspects of wet comminution depend very strongly on transport behaviourwhich is largely controlled by slurry ow. Modelling of a grinding table has demonstrated that DEM when using a dynamic breakage model can predict reasonable particle ow patterns, fracture rates, resident and product sizes as well as help to understand the way in which the mill utilizes energy. The energy spectra obtained for the grinding table show important differences to the ball mill. Conversely, the overall energy absorption by each size class is shown also to be broadly dependent on the area of particles in each class. Stirred mills use super-critical rotation speeds to generate high-particle pressures and high shear in order to increase the energy intensity of the grinding. A tower mill operates at speeds of around 6 times that of the comparable ball mill. This mill appears to be an effective design with good transport behaviour and with broad participation of the charge in the breakage. In particular, the energy dissipation rates are very uniform along the height of the mill. The Isamill is a fully enclosed mill that operates at much higher speeds and when optimized for a particular ore can grind down to ultra-ne sizes. Here we have demonstrated the importance of including the ne feed ore in the model as it tends to isolate the media particles and ensures that the energy consumption is typically expended in contacts involving rocks with only limited mediamedia and medialiner collisions. The feed material is quite mobile and densely lls the interstices between the media particles starting from the mill shell. The distance that the feed reaching radially inwards therefore depends on the feed ll level. The energy dissipation is again dominated by the shear component. The normal energy spectra is shown to be bi-model with the rst peak arising from the weaker rockrock collisions and the higher energy second peak arising from the much stronger rockmedia collisions. Finally, particle trajectories were used to explore the complex nature of the particle motion within the compartments between discs, spiralling axially between compartments and also owing through the holes in the grinding discs.
352
DEM is increasingly useful as a tool that can help provide fundamental insights into comminution processes and into the behaviour of specic machines. It can contribute to design of new equipment, to improvement of existing equipment and to assist in increasing the operational efciency of all comminution unit processes. Here we have used DEM to gain detailed insight into ve different mills: SAG mill, a ball mill, a grinding table, a tower mill and an Isamill. Differences in the nature of their energy utilization can be deduced from their energy spectra and reveal both differences and similarities of operation.
ACKNOWLEDGEMENTS
The authors would like to acknowledge the contribution of Dave Morton to the modelling of the Isamill. The DEM tower mill modelling work reported in this paper has been partially funded by the Centre for Sustainable Resource Processing.
REFERENCES 1. Napier-Munn TJ, Morrell S, Morrison RD, Kojovic T. Mineral Comminution Circuits: Their Operation and Optimization. Julius Kruttschnitt Mineral Research Center: Indooroopilly, Qld., 1999. ISBN: 0 646 28861 X. 2. Mishra BK, Rajamani RJ. The discrete element method for the simulation of ball mills. Applied Mathematical Modelling 1992; 16:598604. 3. Mishra BK, Rajamani RK. Simulation of charge motion in ball mills. Part 1: experimental verications. International Journal of Mineral Processing 1994; 40:171186. 4. Cleary PW. Predicting charge motion, power draw, segregation, wear and particle breakage in ball mills using discrete element methods. Minerals Engineering 1998; 11:10611080. 5. Cleary PW. Charge behaviour and power consumption in ball mills: sensitivity to mill operating conditions, liner geometry and charge composition. International Journal of Mineral Processing 2001; 63:79114. 6. Cleary PW, Sawley ML. DEM modelling of industrial granular ows: 3D case studies and the effect of particle shape on hopper discharge. Applied Mathematical Modelling 2002; 26:89111. 7. Cleary PW. Axial transport in dry ball mills. Applied Mathematical Modelling 2006; 19:15171527. 8. Rajamani RK, Mishra BK. Dynamics of ball and rock charge in sag mills. Proceedings of the SAG 1996, Department of Mining and Mineral Process Engineering, University of British Columbia, British Columbia, 1996. 9. Cleary PW. Modelling comminution devices using DEM. International Journal for Numerical and Analytical Methods in Geomechanics 2001; 25:83105. 10. Cleary PW. Recent advances in DEM modelling of tumbling mills. Minerals Engineering 2001; 14:12951319. 11. Herbst JA, Nordell L. Optimization of the design of sag mill internals using high delity simulation. In Proceedings of the SAG Conference, Vancouver, BC, vol. IV, Barratt DJ, Allan MJ, Mular AL (eds). University of British Columbia: British Columbia, 2001; 150164. 12. Morrison R, Cleary PW, Valery W. Comparing power and performance trends from DEM and JK modelling. Proceedings of the SAG 2001, vol. IV. University of British Columbia: British Columbia, 2001; 284. 13. Cleary PW, Hoyer D. Centrifugal mill charge motion and power draw: comparison of DEM predictions with experiment. International Journal of Mineral Processing 2000; 59:131148. 14. Cleary PW, Morrison R, Morrell S. Comparison of DEM and experiment for a scale model SAG mill. International Journal of Mineral Processing 2003; 68:129165. 15. Powell MS, McBride AT, Govender I. Application of DEM outputs to rening applied sag mill models. In Proceedings: XXII International Mineral Processing Congress, Lorenzen L, Bradshaw DJ (eds), Cape Town, South Africa, 2003; 307316. 16. Hlungwani O, Rikhotso J, Dong H, Moyes MH. Further validation of DEM modelling of milling effects of liner prole and mill speed. Minerals Engineering 2003; 16:993998. 17. Bwalya MM, Moys MH. The use of DEM in predicting grinding rate. In Proceedings: XXII International Mineral Processing Congress, Lorenzen L, Bradshaw DJ (eds), Cape Town, South Africa, 2003; 16121617. 18. Haff PK, Werner BT. Computer simulation of the mechanical sorting of grains. Powder Technology 1986; 48:239245. Copyright q 2008 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Fluids 2008; 58:319353 DOI: 10.1002/d
353
19. Ristow GH. Granular dynamics: a review about recent molecular dynamics simulations. Annual Reviews of Computational Physics 1994; 1:275308. 20. Cleary PW. Discrete element modelling of industrial granular ow applications, TASK. QuarterlyScientic Bulletin 1998; 2:385416. 21. P schel T, Buchholtz V. Complex ow of granular material in a rotating cylinder. Chaos, Solitons and Fractals o 1995; 5:19011912. 22. Potapov AV, Campbell CS. Computer simulation of Hopper ows. Physics of Fluids 1996; 8:28842894. 23. Cleary PW. Large scale industrial DEM modelling. Engineering Computations 2004; 21:169204. 24. Campbell CS. Rapid granular ows. Annual Review of Fluid Mechanics 1990; 22:5792. 25. Barker GC. Computer simulations of granular materials. In Granular Matter: An Interdisciplinary Approach, Anita M (ed.). Springer: New York, NY, 1994. 26. Walton OR. Numerical simulation of inelastic frictional particleparticle interaction. In Particulate Two-phase Flow, Chapter 25, Roco MC (ed.). 1994; 884911. 27. Sch fer J, Dippel S, Wolf DE. Force schemes in simulation of granular material. Journal de Physique I (France) a 1996; 6:5. 28. Williams JR, Pentland A. Super-quadrics and modal dynamics for discrete elements in interactive design. International Journal of Computer-Aided EngineeringEngineering Computations 1992; 9:115127. 29. Morrison R, Shi F, Whyte R. Modelling of incremental rock breakage by impactfor use in DEM models. Minerals Engineering 2007; 20:303309. 30. Djordjevic N, Morrison R, Loveday B, Cleary PW. Modelling comminution patterns within a pilot scale AG/SAG mill. Minerals Engineering 2006; 19:15051516. 31. Sinnott MD, Cleary PW, Morrison RD. Analysis of stirred mill performance using DEM simulation. Part 1media motion, energy consumption and collisional environment. Minerals Engineering 2006; 19:15371550. 32. Cleary PW, Sinnott MD, Morrison RD. Analysis of stirred mill performance using DEM simulation. Part 2 coherent ow structures, liner stress and wear, mixing and transport. Minerals Engineering 2006; 19:15511572. 33. Morrison R, Loveday B, Djordjevic N, Cleary PW, Owen P. Linking discrete element modeling to breakage in a pilot-scale AG/SAG mill. In Advances in Comminution, Kawatra SK (ed.). SME: Littleton, Colorado, 2006; 269283.
Copyright q
Int. J. Numer. Meth. Fluids 2008; 58:319353 DOI: 10.1002/d

DEM Prediction of Particle Flows in Grinding Processes: P. W. Cleary, M. D. Sinnott and R. D. Morrison

Uploaded by

Document Informationclick to expand document information

Copyright:

Available Formats

DEM Prediction of Particle Flows in Grinding Processes: P. W. Cleary, M. D. Sinnott and R. D. Morrison

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

DEM Prediction of Particle Flows in Grinding Processes: P. W. Cleary, M. D. Sinnott and R. D. Morrison

Uploaded by

Copyright:

Available Formats

INTERNATIONAL JOURNAL FOR NUMERICAL METHODS IN FLUIDS Int. J. Numer. Meth.

DEM prediction of particle ows in grinding processes

comminution; milling; particles; breakage; energy utilization; DEM

2008 John Wiley & Sons, Ltd.

P. W. CLEARY, M. D. SINNOTT AND R. D. MORRISON