You are on page 1of 26

Available online at www.sciencedirect.

com

Particuology 6 (2008) 419–444

Assessing mixing characteristics of particle-mixing


and granulation devices
Paul W. Cleary ∗ , Matthew D. Sinnott
CSIRO Mathematical and Information Sciences, Private Bag 33, Clayton South 3168, Australia
Received 1 June 2008; accepted 15 July 2008

Abstract
The mixing of particulates such as powders is an important process in many industries including pharmaceuticals, plastics, household products
(such as detergents) and food processing. The quality of products depends on the degree of mixing of their constituent materials which in turn
depends on both geometric design and operating conditions. Unfortunately, due to lack of understanding of the interaction between mixer geometry
and the granular material, limited progress has been made in optimizing mixer design. The discrete element method (DEM) is a computational
technique that allows particle systems to be simulated and mixing to be predicted. Simulation is an effective way of acquiring information on
the performance of different mixers that is difficult and/or expensive to obtain using traditional experimental approaches. Here we demonstrate
how DEM can be used to unravel flow dynamics and assess mixing in several different types of devices. These devices used for mixing and/or
granulation of particulates, are classified broadly as gravity controlled, bladed and high shear. We also explore the role of particle shape in mixing
performance and use DEM to test whether Froude number scaling is suitable for predicting scale performance of rotating mixers.
© 2008 Chinese Society of Particuology and Institute of Process Engineering, Chinese Academy of Sciences. Published by Elsevier B.V. All
rights reserved.

Keywords: Discrete element modelling; Mixing; Mixers; Blenders; Granulation; Scale up; Particle shape

1. Introduction limited progress has been made in optimizing mixer design.


Many challenges remain including:
The mixing of granular systems such as powders is a funda-
mental operation, important for a range of industries including • scale up from laboratory scale,
pharmaceuticals, plastics, house-hold products (such as deter- • robustness of performance to changing conditions (typically
gents) and food processing. The quality of products depends on fill level and speed),
the degree of mixing of their constituent materials. However, the • how the mixer will perform for new materials or new combi-
homogeneity of the final mixture relies on the nature of the mix- nations of materials, and
ing procedure used. Any systematic mixing process is incapable • increasing the extent of mixing as quality requirements on
of creating a completely random mixture since the likelihood of products become more demanding.
it re-ordering the particles increases with the degree of mixing.
Eventually an equilibrium point is reached where the mixing To date, there has been a great deal of focus on studies of
and de-mixing mechanisms are in balance and there is a natu- granular flows but only moderate attention has been devoted
ral cessation of mixing. For different devices, the equilibrium to granular mixing. A homogeneous mixture is not a guaran-
states will depend on both geometric design and operating con- teed end-state for the operation of any mixing device, since the
ditions. Unfortunately, due to a lack of understanding of the process involves complex three-dimensional flows generating
interaction between mixer geometry and the granular material, regions of high and low mobility. Differences in particle size,
shape and density may in fact promote segregation leading to
self-organisation and pattern formation. Historically, the oppor-
∗ Corresponding author. Tel.: +61 3 95458005; fax: +61 3 95458080. tunity to study flows in granular mixing has been hampered by
E-mail address: Paul.Cleary@csiro.au (P.W. Cleary). the lack of suitable means for visualizing them. However, recent

1674-2001/$ – see inside back cover © 2008 Chinese Society of Particuology and Institute of Process Engineering, Chinese Academy of Sciences. Published by Elsevier B.V. All rights reserved.
doi:10.1016/j.partic.2008.07.014
420 P.W. Cleary, M.D. Sinnott / Particuology 6 (2008) 419–444

advances in discrete element modeling (DEM) have resulted in Mixing devices vary considerably in their design and
this method becoming a useful simulation tool that can provide intended application, and are operated in either continuous or
detailed information not easily measured during experiments. batch modes. Continuous operation requires the mixing time be
There are many challenges in developing computational matched to the residence time of particles flowing through the
models for studying granular mixing in commercial devices. device in order for the product exiting from the device to be
Production-scale rotary batch mixers typically have capacities adequately mixed. This places a limit on the throughput of such
of hundreds of litres (and can range up to 20,000 l) filled with devices. Devices intended for batch operation continue mixing
wet/cohesive, non-spherical particles of sizes ranging down to until a desired mixing state is reached. Common batch mixers
tens of microns. DEM models of granular mixing are mostly include rotary mixers like the bin and V-blenders, as well as
small-scale laboratory mixers acting on materials of dry spher- high-shear mixers. Typically the duration of a batch mixing pro-
ical particles. It has been shown that mixing rates for dry, free cess is chosen in advance based on historical experience. Mixing
flowing materials in rotating drums are independent of rota- quality can be checked by sampling after the mixing process has
tional speed (Sudah, Coffin-Beach, & Muzzio, 2002), whereas ended, but it is expensive to return a batch for additional mixing
for cohesive materials, a higher rotational speed results in more if the quality is not adequate. Often for new products or when
rapid mixing (Chaudhuri, Mehrotra, Muzzio, & Tomassone, conditions such as humidity lead to variation of the cohesiveness,
2006). Scale-up is still poorly understood for the majority the required mixing duration cannot be adequately estimated in
of mixing devices. Ding, Forster, Seville, and Parker (2001) advance.
describe scaling relationships for rotary drums, but the develop- Granulation refers to any process which results in the growth
ment of larger, industrial-scale models are needed to determine of size of particulate aggregates due to the formation of inter-
scaling laws for other devices. With the maturing of DEM simu- particle bonds (e.g. liquid cohesion, sintering or crystallization).
lation, it is now becoming possible to run simulations of millions Its use in process industries is widespread, with examples includ-
of particles with complex shapes and inter-particle cohesive ing the production of tablets in the pharmaceutical industry,
forces in tolerable times on single processor, desktop computers. the manufacture of fertilizer pellets, agglomeration of wood
Particle morphology (shape) can vary dramatically with feed fibers in the paper industry and sintered iron pelletization in
material and is known to be of critical importance in obtaining the steel industry. Other industries in which granulation has
quantitative accuracy for the modelling of mixing in rotating critical importance include the manufacture of food additives
drums (Cleary & Metcalfe, 2002; Cleary, Metcalfe, & Liffman, (especially phosphates) and food products (such a breakfast
1998). Spherical particles have low resistance to shear compared cereals), soap/detergent production, minerals processing and the
to more angular or blocky particles. This can be a major prob- production of specialty plastics.
lem for correctly predicting mixing rates, where the flow regime “Wet granulation” involves a liquid binder sprayed onto
and mixing patterns for realistic particle assemblies can depart the surface of a granular bed. If this occurs during mixing,
significantly from that of spheres. Angularity and aspect ratio then the binder is able to penetrate into the granular mate-
are important parameters for characterising the effects of shape rial. Under these circumstances liquid bridges form between
on particulate flows and microstructures. Angularity leads to particles and thus provide a cohesive force that allows the for-
tighter packing as neighbours lose their ability to freely roll mation and growth of aggregates. The equipment used to mix
over each other. Elongated particles, with high aspect ratio, the granular material and liquid binder varies between indus-
can lead to complicated interlocking of the microstructure and tries and applications, but can be broadly classified as tumbling
restrict the flow; as well as generate complex force networks agglomerators (e.g. dish pelletizers and drum granulators) or
within a granular system. Rolling and sliding friction coeffi- mixer-agglomerators (e.g. paddle mixers, plough share mixers
cients are often applied to modify the flow dynamics in such and peg mixers). From a process-engineering perspective, the
systems in order to mimic the flow behaviour of particles of control of wet granulation is focused on manipulating the oper-
more complex shape. However this simplistic approach is unable ating and design variables that influence the final properties of
to correctly predict a number of important aspects of assem- the aggregates. These include size distribution, aggregate poros-
blies of non-spherical particles such as packing fractions, and ity, shape and chemical composition. In general, these properties
the extent of dilation of a granular bed in regions of high shear will be determined by the nature of interactions between parti-
(which can affect the rate of percolation of fine particles). Under- cles and the distribution of liquid binder. Since these factors
standing the key role that shape plays in determining packing are essentially determined by the granular flow during mixing,
efficiency, strength of a microstructure, flowability and seg- it is reasonable to expect that identifying means for control-
regation is important for predicting mixer performance for a ling wet granulation will follow from an understanding of the
given granular material. Due to the computational expense in flows.
contact detection for complex shaped particles, existing three- In this paper, we present DEM models for several types of
dimensional DEM models of granular mixing generally only mixers as summarized in Fig. 1. They are classified into three
use spherical particles (Bertrand, Leclaire, & Levecque, 2005). groups:
To the best of our knowledge, only a single three-dimensional
study of granular mixing involving non-spherical particle shape 1. gravity-controlled mixers,
in DEM has been reported in the literature so far (Sinnott & 2. stirred mixers, and
Cleary, 2005). 3. high shear (or high energy) mixers.
P.W. Cleary, M.D. Sinnott / Particuology 6 (2008) 419–444 421

Fig. 1. Summary of the mixing devices presented in this paper and classified into batch/continuous operation and gravity/agitator driven mixing.

Devices that use gravity to drive mixing include rotary forces from knowledge of the current positions, orientations,
devices such as drum blenders (Finnie, Kruyt, Ye, Zeilstra, & velocities and spins of the particles. Here we used the simplest
Kuipers, 2005; Kwapinska, Saage, & Tsotsas, 2006; Metcalfe, of the force laws which is the linear spring-dashpot model. The
Shinbrot, McCarthy, & Ottino, 1995), bin blenders (Arratia, overlap x scaled by the spring constant k provides a repul-
Duong, Muzzio, Godbole, Lange, et al. 2006; Arratia, Duong, sive force. The dashpot contributes an inelastic component to
Muzzio, Godbole, & Reynolds, 2006; Lemieux, Bertrand, the collision. The damping coefficient C is determined from the
Chaouki, & Gosselin, 2007), pan granulators (Morton & Cleary, specified coefficient of restitution for each combination of mate-
2002), and V-blenders (Kuo et al., 2002; Kuo, Knight, Parker, rials colliding. The spring and dashpot together define the normal
& Seville, 2005; Lemieux et al., 2007, 2008). Stirred mixers force:
make use of a horizontally or vertically aligned impeller to agi-
tate the granular bed. There are many agitator types used in Fn = −k x + Cvn , (1)
industrial practice with only a small number of these having where vn is the normal component of the relative velocity at the
been investigated using DEM simulation. The variants of stirred contact point. In similar way, the tangential force has an incre-
mixer examined so far include plough share mixers (Cleary, mental spring based on the integrated tangential displacement
Laurent, & Bridgewater, 2002; Laurent & Cleary, submitted and a dashpot for inelastic dissipation. The tangential force is
for publication), peg mixers (Morton & Cleary, 2002), pad- then subject to the sliding Coulomb friction limit to give:
dle mixers, and ribbon blenders (Bertrand et al., 2005). High   
shear mixers are a special class of stirred mixer that use a very Ft = min μFn , kvt dt + Cvt (2)
high speed impeller to work the granular material (Kuo, Knight,
Parker, Adams, & Seville, 2004; Sinnott & Cleary, 2003, 2005; where vt is the tangential component of the relative velocity at
Stewart, Bridgewater, & Parker, 2001; Stewart, Bridgewater, the contact point and μ is the coefficient of friction. A general
Zhou, & Yu, 2001). Stewart, Bridgewater, Zhou, et al. (2001) review of DEM and its variants can be found in Barker (1994).
have made good use of PEPT measurements to validate their The essential elements of the algorithm for the simulation
DEM models. Zhou, Yu, Stewart, and Bridgewater (2004) have procedure are
also identified micro-dynamic variables related to flow struc-
tures for high shear mixers. 1. A search grid regularly maintains a near-neighbour interac-
tion list for all particle pairs and particle-boundary pairs that
2. Simulation technique might participate in a collision in that given period.
2. For each timestep, this list is used to identify all collisions
2.1. DEM method involving particles and boundary elements. The forces on
particle pairs and boundaries are evaluated using the contact
The discrete element method is a numerical technique used to force model.
predict the behaviour of collision dominated particle flows. Each 3. All the pairwise collision forces and torques are summed to
particle in the flow is tracked and all collisions between parti- give net forces and torques. Forces from other forms of inter-
cles and between particles and boundaries are modelled. The action can also be added in this step. Newton’s equations of
particles are allowed to overlap and the extent of overlap is used motion and the matching kinematic equations are then inte-
in conjunction with a contact force law to give instantaneous grated to give the changes in the position, speed, orientation
422 P.W. Cleary, M.D. Sinnott / Particuology 6 (2008) 419–444

and spin of the particles in response to these net forces and sinusoid with the exponent characterising the rate of mixing and
torques. the period of oscillation relating to some form of characteristic
turnover of the particles in the mixer which is very dependent
Steps 2 and 3 are repeated sequentially stepping the system on the nature of the flow pattern. This method was originally
state forward in time until the search information is no longer used by Metcalfe et al. (1995) to experimentally measure mixing
valid. Then a new search (step 1) is performed to update the in a rotating drum. It was then used to assess mixing in DEM
interaction list. This DEM has been used to model many indus- simulations of the same system in Cleary et al. (1998).
trial granular flows (see Cleary, 2004; Cleary & Sawley, 2002
for examples). More details of the implementation can be found 2.2.2. Particle mixing measure
in Cleary (1998). The method is flexible and can be used with both continuous
DEM simulations of real materials require more complex par- and discrete particle distributions. A three-dimensional grid is
ticle shape descriptions. A flexible class of shapes that balance constructed over all the particles. Local averages of a selected
increased realism with acceptable loss of computational speed is property (e.g. mass, density, diameter, colour) are calculated
the super-quadric (SQ). This enables the angularity and varying using all particles in the region centered at each grid point. The
principle axis aspect ratios to be included in the modelling allow- size of the averaging cell is typically chosen to be fives times
ing more realistic flow predictions. Cleary and Metcalfe (2002) the average particle diameter in order to ensure that the local
showed that the inclusion of shape using SQ particles was able averages are statistically meaningful. These local averages at
to give DEM quantitative accuracy in predicting mixing in rotat- each of the grid points are then collected to form a probability
ing drums. These shapes were first used in two dimensions by distribution for the chosen property. The choice of a multiple
Williams and Pentland (1992) and in three dimensions by Cleary of five means that in 3D there are typically 50–100 particles
(2004) which contains a number of three-dimensional examples. used in the local averages, giving an estimate of the standard
They are described by deviation of the noise in the local means of 0.3 of these means.
 x m  y m  z m Therefore the estimate of the local mean is, to a 95% confidence
+ + =1 (3) level, within ±0.9x true local mean of the true mean. This is
a b c
sufficiently accurate for the current purpose and balances the
The resultant particles are somewhere between an ellipsoid need to have good local estimates of the mean with the need to
and a rectangular box with the angularity/smoothness of the cor- have sufficient independent data points to accurately characterise
ners controlled by the choice of m (with m = 2 being ellipsoidal the distribution of local means.
and large m being a box) and the aspect ratios being controlled The mean, standard deviation, and coefficient of variation η,
by the ratios of the semi-major axes (b/a and c/a). (defined as mean divided by standard deviation) are then cal-
culated for the entire distribution of local means at each time.
2.2. Measuring mixing in DEM Ideal limits are calculated for the coefficient of variation for the
fully segregated and fully mixed states, ηs and ηr respectively
Mixing of particulates can be measured in a number of ways. (see Cleary, 1998 for details). This allows η to be normalised
Two methods that are very suitable for use in DEM for predicting and provide a single mixing measure in the range of 0–1. Since
mixing and segregation in granular systems are described in this quantity evolves with time (reflecting mixing and segrega-
Cleary et al. (1998) and Cleary (1998). They are tion processes) it is not only a good indicator of the quality of
the mixture but also a natural measure of the mixing rate. This
• Colour centroids and mixing/segregation measurement process is described in more
• Mixing measure. detail in Cleary (1998).

Both are used in this paper and are described next. 3. Gravity controlled mixers

2.2.1. Colour centroids Gravity controlled mixers are devices whose primary mode
Particles can be coloured into groups reflecting their initial of flow is avalanching down free surfaces of a bed of particles.
positions and the nature of the mixing that is to be measured. Mixing is due to local mixing in this avalanching layer and is
The distance of the centroids for each of the different colours restricted by the rate at which material can be supplied to the
from the overall center of mass then characterises the degree source of the avalanches. Different devices use different machine
of mixing of that group of particles. Initially, all the particles geometries and different motions to periodically or continuously
belonging to each colour lie in separate regions and the centroid create opportunities for surface avalanche mixing.
locations are well separated. As the particles mix, some cross
from one region to another and the centroids move closer to the 3.1. Drum mixer
overall centroid. The differences in the centroid locations from
the overall centroid can be normalised by their initial values Drum mixers can be configured to operate in both contin-
to be unity. If the particles become perfectly mixed then all uous and batch modes. Here we predict the particle flow in a
the normalised centroid differences become zero. The centroid laboratory scale continuous drum mixer. It has a 250-mm inter-
components can often be fitted by an exponentially declining nal diameter and is 500 mm long. The feed end on this mixer
P.W. Cleary, M.D. Sinnott / Particuology 6 (2008) 419–444 423

the shoulder (highest) position to the toe (lowest) position. For


granular motion in tumbling cylinders, such motion is commonly
referred to as the “rolling” regime. Given that the rotation speed
simulated corresponds to 40% of the critical speed at which cen-
trifuging occurs, this relatively gentle motion is to be expected.
In addition to indicating the gentle rolling nature of the sur-
face flow, the cross-sectional view shown in Fig. 2 also shows
a slow moving interior core around which the granular material
rotates. The regions at the periphery of this core correspond to
regions where fast moving particles are able to slide past layers
of slower moving particles. Such conditions produce significant
shear. In mixing and granulation processes, this shear will lead
to attrition of particles or aggregates or the re-distribution of
inter-particle bonds. Hence, it is expected that the outer limits
of the slow moving core will be a region in which some attrition
occurs and aggregate density (and porosity) is modified.
Fig. 3 shows the way in which new feed material flows
through the existing particle bed in such a continuously fed
mixer. The new feed particles are coloured red and the exist-
ing bed is coloured light blue. The feed stream falls into the
bed a small distance from the end of the drum trapping a small
length of existing material behind the stream. Once the stream
contacts the surface it avalanches down the free surface and is
dragged under the surface by the drum rotation. The red particles
then become visible again at the shoulder a small axial distance
further down the drum. They then avalanche down the surface
reaching and then blending into the original feed stream on the
surface. The interface between the coloured materials in Fig. 3b
and c is almost planar with red material to on the feed side and
blue predominantly on the discharge side. The interface plane
is modestly angled with the red material at the toe moderately
in advance of the red material at the shoulder. The feed material
Fig. 2. Steady-state motion in a laboratory scale continuous drum mixer with forms a plug that moves at nearly uniform speed along the drum.
particles colored by speed (top) oblique view of the entire drum contents, and There is a modest degree of mixing across the interface which is
(bottom) an axial sectioned view (at z = 500 mm which is immediately before
the conical outlet).
purely diffusive in nature. After about 8 s (or 4.5 revolutions of
the drum) the new feed material reaches the discharge and starts
to leave the drum. In the last frame of Fig. 3, the red particles
consists of a flat plate and a horizontal feed port with an inner densely fill the entire drum except for the small band of blue
diameter of 55 mm. The discharge was modeled as a convergent trapped against the feed end and a scattering of blue particles
conical section, reducing from 250 mm to 155 mm. The drum mixed with the currently discharging red particles.
was inclined at 7.5◦ and rotates in an anti-clockwise direction This uniform progression of feed particles is clearly identifi-
at 34 rpm. Simulations were run for a time corresponding to able when examining the residence time distribution of the feed
approximately 15 revolutions to allow material introduced at which is shown in Fig. 4a. This distribution is calculated over
the feed end to be transported to the discharge end. The feed 15 revolutions of the drum. The distribution is essentially uni-
rate was chosen to ensure that the mass of granular material was modal, with a dominant peak at around 7.5 s with some spread
consistently around 8.0 kg corresponding to a fill level of about from 6 s to 10 s. This shows that the motion in the drum mixer
30%. The particle diameters in the system ranged from 2.51 mm approaches that of plug-flow. An important contributor to the
to 8.82 mm with density 2000 kg/m3 . The particle coefficient modest variation in residence time is the size of the particles.
of restitution was 0.3 and the friction coefficient was 0.75. A Smaller particles percolate quickly through the surface layers
spring stiffness of 5000 N/m was sufficient to produce the target and migrate towards the center of the bed. The amount of axial
average overlap of less than 1%. dispersion depends on the distance down the avalanching sur-
A steady flow pattern was established after about four revo- face that a particle can travel (Cleary, 2006). So smaller particles
lutions and is shown in Fig. 2. As might be expected the axial can be expected to travel with lower axial speeds compared to
flow is from the inlet towards the conical discharge. The cross- larger particles which can avalanche all the way down to the
sectional view indicates that the granular material is lifted to a toe region and therefore produce the largest axial migrations per
shoulder position with a steady angle of approximately 32◦ to cycle. Close inspection of the animations of these drum simu-
the horizontal. Material at the free surface then tumbles from lations show that the leading particles are all larger ones. The
424 P.W. Cleary, M.D. Sinnott / Particuology 6 (2008) 419–444

Fig. 3. Progression of feed particles (red/dark grey) from entry to discharge in a laboratory scale continuous drum mixer through a pre-existing bed (shown as light
blue/light grey particles).

smaller particles have longer residence times (up to around 10 s) In addition to the residence time distributions discussed
and the largest particles have smallest residence times (down to above, it is useful to collect information about the energy dissi-
6 s). It is clear from these results that the drum is actually quite pation associated with collisions. This is particularly important
poor as a mixer with the particles flowing as a fairly uniform when the drum is used as a granulator. Then the intensity of
plug. The small compositional change is actually produced by collisions is responsible for breaking up granules and allow-
size segregation in the avalanching zone, which in turn mod- ing binder to be transferred between particles. Fig. 4b shows the
erately influences the residence time. Segregation in mixers is energy spectra for all collisions and the contribution for particles
typically not an attractive attribute. within the diameter range 5.3–6.2 mm. This spectra shows the

Fig. 4. (Left) Residence time distribution of red particles over 15 revolutions of the drum mixer and (right) collision energy spectra for all collisions over the same
period.
P.W. Cleary, M.D. Sinnott / Particuology 6 (2008) 419–444 425

Fig. 5. Particle motion in a laboratory dish granulator, (left) particles colored by velocity (with blue being slow and red being fast) and (right) distribution of the red
feed material in the original blue bed of particles.

most frequent collisions occur at the lower energy levels, in the occur at roughly the 8 o’clock position where the bed is most
range 10−6 − 10−5 J. At such energies, it is reasonable to expect dilute and most mobile. In a granulation application, this is the
that aggregate breakage will be minimal and that the primary region in which liquid binder would typically be sprayed. From
effect of collisions will be to bring particles into contact. Hence, an operational sense, this would seem appropriate since there
for the conditions considered here, the main factors affecting will be a high frequency of high-energy particle–particle colli-
aggregate size are expected to be the particles’ residence time sions. This is likely to lead to efficient binder transport as fluid is
and the distribution of liquid binder. The weak collisional inten- exchanged during the formation and breakage of liquid bridges.
sities will also limit the ability of this device to operate as a Fig. 5b shows the particles coloured by whether they are new
granulator. feed (red) or existing bed material (blue). The red material has
filled a coherent band from the feed location down to around
the 7 o’clock location. This indicates that the particle motion is
3.2. Pan mixer
essentially a uniform plug flow down the middle and left side of
the pan. The flow behaviour for the pan mixer does not suggest
A more complex mixer, often used for granulation, is a pan
that this device does a particularly large amount of mixing.
mixer. This consists of a shallow pan or dish with a circular flat
bottom and a low surrounding wall. It is held at a constant angle
and moved with a swirling motion, in much the manner used 3.3. V-blender
in ‘panning for gold’. Feed material is poured onto the mixer
near the top. The particles follow complex swirling trajectories A V-blender uses a V shaped mixing vessel for batch mixing.
and form a bed built up against the lower lip of the pan. New Each arm of the V has a cylindrical cross-sectional shape and
feed mixes with the old feed in the bed. This raises the bed level there is typically an access port at the bottom of the V for filling
and leads to discharge by overflowing near the lowest point of and discharge of the particulates. This is sealed during mixing
the pan. In this case, the pan is 500 mm diameter and has a operation. The blender has a horizontal shaft attached to the V
wall height of 100 mm. The pan has an inclination of 45◦ and and the blender rotates around this shaft. The configuration is
is rotated at 45 rpm in an anti-clockwise direction. A steady- symmetric to reflection in a vertical plane through the V. The
stream of material is dropped onto the pan around the one to particulates start in the two ends of the blender and during the
two o’clock position. The feed rate is dynamic and matches the first half revolution the particles flow down towards the vertex
discharge rate leading to a constant bed mass of 9.0 kg. The feed of the V with mixing occurring in the shearing bulk flow. In the
particle diameter varied from 2.5 mm to 8.8 mm with density second half of a revolution the particles flow back away from
2000 kg/m3 . The particle coefficient of restitution was 0.3 and the vertex towards the ends of the arms of the V producing a
the friction coefficient was 0.75. The spring stiffness used was further increment of mixing. This batch mixing process contin-
106 N/m. ues for a set number of revolutions with increments of mixing
Fig. 5 shows the particle distribution inside the pan, coloured occurring for each half revolution as particles flow alternatively
both by velocity and by whether it is feed material. They are away and then towards the ends. The basic flow in a V-blender
consistent with the generally observed behavior of industrial has previously been studied using DEM by Kuo et al. (2002)
dish granulators. Material in the region of the 4 o’clock position and is therefore not described here. They validated their DEM
passes over the granulator wall, forming a discharge stream. models against positron emission particle tracking (PEPT) mea-
Apart from the incoming feed stream, the highest velocities surements and observed that the flow exhibited two phases: a
426 P.W. Cleary, M.D. Sinnott / Particuology 6 (2008) 419–444

separation phase as material splits into two streams and flows Table 1
along the arms of the V-blender; and a recombination phase Model parameters for the scaled up V-blender.
where the return streams come together at the bottom of the Blender Particle Length of Rotation speed for
blender. In particular, they showed that mass transfer from one volume (L) numbers blender (m) Froude scaling (rpm)
arm to the other is more likely to occur during separation of the 1.86 48,500 0.230 30.00
streams. From a pure experimental study, Kuo et al. (2005) con- 5.88 153,000 0.338 24.76
sidered the similarity between V-blenders and tumbling drums 18.6 485,000 0.500 20.44
in that they both have a surface flow regime and a regime of 58.8 1,530,000 0.727 16.87
186.0 4,850,000 1.067 13.93
solid body motion. Circulation in each arm occurs with an inten-
sity that increases proportionally with the square of the angular
speed, and decreases with increasing fill level from 10% to 46
%. Mixing has also previously been studied for small-scale lab- to failure and avalanching down the free surface of the particle
oratory V-blenders by Lemieux et al. (2007, 2008) who showed bed. In order to compare the mixing at different scales, a blender
the evolution of mixing changes for different loading profiles speed scale up rule needs to be applied. The most common one is
and for different rotational speeds. For their mixing measure, to keep the Froude number constant. Here we scale the blender
they have used the relative standard deviation, a metric com- speeds in this way so that we can evaluate the appropriateness
monly used in the pharmaceutical industry. This is the standard of this scale up law. The rotational speeds at each scale are also
deviation for a completely random binary mixture. given in Table 1. A spring stiffness of 1000 N/m was used in all
Now that DEM simulations can be performed for particles these simulations giving average overlaps of 0.16% of the parti-
numbers well beyond one million on single CPU desktop com- cle diameter for the laboratory scale and 0.4% for the industrial
puters, a critical issue that can be explored with DEM is the scale.
question of how mixing scales up from the laboratory scale to Fig. 6 shows the progress of mixing in the laboratory scale
process scale. Here we start to explore how DEM can provide V-blender. The particles start in a settled pile above the vertex
information about this for the V-blender. The laboratory scale of the blender which is oriented downwards. The particles are
blender used here has an internal volume of 1.864 l with a vertex coloured into four vertical strata according to their axial position
angle of 75◦ . It starts with the discharge port at the bottom of across the width of the blender. This colouring scheme allows
the V pointing down and the cylindrical arms pointing upward. assessment of mixing within each of the two cylindrical sec-
It rotates around a horizontal axis that is mid-way between the tions and across the central plane of symmetry. The strata each
lower vertex and the tops of the arms. The blender is scaled contain 25% of the particles so they are different widths (due
through two orders of magnitude in volume and the mixing is to the difference in the bed depth along the sloping lower sur-
assessed at each scale. The largest of the blenders considered face). As the blender rotates slowly around its horizontal axis
here has a volume of 210 l which is suitable for 100 kg batch the free surface changes its angle until it reaches the angle of
blending and is of genuine operational size. DEM modelling of failure of the material which then avalanches down the slope set-
industrial scale mixers for coarser materials (mm scale) is now tling at the angle of repose. Each increment of rotation causes
feasible, but mixing of finer powders (10 ␮m to 1 mm) is still some particles to exceed the angle of failure leading to contin-
not feasible at the industrial scale. uous avalanching from across the high side of the blender. In
Here we are interested in the mixing of coarser mm scale non- the first half revolution the avalanching flow is directed down
cohesive particulates. These are representative of particle sizes into the angled cylindrical sections of the blender shell. These
in foods, some household products and plastics feed stocks. The progressively fill and the charge mass is almost separated into
particles used here have a mean size of 2.75 mm with a range of two masses with only a thin tongue of material across the V
0.25 mm above or below this size. The same particle sizes are shaped boundary between sections of the chamber. After 1 s, the
used in all the different scale mixers. The solid packing fraction mixer has become completely inverted. In the second half of the
is 0.5564. The particle density was 1500 kg/m3 and the blenders revolution particles flow down the slope in the direction of rota-
were filled to 50.1% by volume. For the laboratory scale mixer tion but also converge back towards the middle of the blender
the mass of particles was therefore 0.78 kg and for the largest as the particles pack into the apex region. The mixer continues
scale blender it was 78 kg. The physical length of blender and this flow pattern and each revolution produces an increment of
the number of particles in the model for each blender scale are mixing.
given in Table 1. The coefficient of restitution was 0.7 for the The distribution of particles is shown after one revolution
particle–particle collisions and 0.85 for particle–wall (steel) col- (2 s) in Fig. 6b. The particle bed does not return to the initial
lisions, while the friction coefficient was 0.5 for particle–particle position because of the angle of repose free surface with par-
contacts and 0.4 for the particle–wall ones. ticles higher at the back than at the front. The first revolution
The Froude number is the ratio of the centrifugal force result- has also substantially re-arranged the strata. The combination of
ing from the rotation to gravity. The laboratory scale blender is the free surface avalanching in the direction of rotation with the
operated at 30 rpm. This corresponds to a Froude number of cyclic axial flow induced by the angles of the cylindrical cham-
about 0.1 which corresponds to gentle avalanching flow. The ber sections produces a complex circulatory flow within each
gradually changing angle of the blender causes the particles to section. On the left, both the blue and red particles now stretch
exceed the angle of failure of the particulate material leading from the center to the far left with the red occupying the rear of
P.W. Cleary, M.D. Sinnott / Particuology 6 (2008) 419–444 427

the blender and the blue material at the front. This is in sharp centerline. After four revolutions the blue material is towards the
contrast to the initial state where the particles were coloured front with the red predominantly at the back. This state is very
purely in axial strata. There is only modest diffusive mixing of similar to that observed after one revolution. This continues for
the colours along the interfaces. After two revolutions (4 s) the very large numbers of revolutions but becomes less distinct as
particles occupy the same position, but the red and blue sets of the colours intermix. The slow circulation of the colour groups
particles have changed their position from that after one revolu- within each cylindrical section means that the mixing flow has
tion. The red material is predominantly on the top and the blue a longer-term convective component which is quite complex in
is now underneath. There is clearly a circulatory component to nature.
the flow with periodicity that is slower than the rotation speed The continuous avalanching flow down the free surface and
of the blender. After three revolutions, the blue is more towards the cyclic axial flow generate diffusive mixing. This can be
the front and is concentrated towards the outside and near the clearly discerned by the steady blurring of the interfaces between

Fig. 6. Mixing in a laboratory scale V-blender. Vertical strata show the nature of the mixing. Longer term mixing in a laboratory scale V-blender.
428 P.W. Cleary, M.D. Sinnott / Particuology 6 (2008) 419–444

Fig. 6. (Continued)

the colours. The migration of individual particles from one the opportunity to become more mixed declines as the mate-
region to an adjacent one is essentially a random walk type rial becomes more uniform and because the mixing on each side
behaviour and leads to steadily increasing mixing. The convec- becomes complete by 40 revolutions leaving only the slower dif-
tive transport accelerates the mixing by stretching and folding fusive mixing across the plane of symmetry after that. By 120
the interface and bringing new unmixed material close to other revolutions the material is 80% mixed. It is clear that to become
unmixed material allowing the diffusive mixing (which is inher- fully mixed (the asymptotic state in this mixer) will take a sig-
ently a short scale phenomena) to be more effective. nificant amount more time. Therefore for batch processing, a
By 40 revolutions (80 s) the red and blue material on the left target level of mixture quality is selected and the required mix-
are very close to perfectly mixed and the green and white in the ing time to reach the desired state can then be determined from
right are also almost perfectly mixed. But there is only mild mix- this graph. The very slow mixing in the later stages can poten-
ing of materials from one side of the blender into the other. The
symmetric nature of the blender geometry creates a symmetric
flow pattern which makes it difficult to transfer particles across
the plane of symmetry. The mixing across this interface is purely
by diffusion as the material avalanches down the free surface of
the rotating bed. Diffusive mixing is much slower, so the mixing
across the two sides is very slow in comparison to the mixing
within each side. By 120 revolutions (240 s), the mixture has
finally attained a reasonably good degree of mixing. This means
that the axial mixing between sides takes more than four times
longer to become well mixed than is required for material on
each side of the blender.
Fig. 7 shows the evolution of the state of mixing for the
laboratory scale V-blender. This mixing rate (given by the gradi-
ent of this curve) is initially reasonably rapid and then declines
steadily over 120 revolutions. The mixing rate declines because Fig. 7. Evolution of the state of mixing for the laboratory scale V-blender.
P.W. Cleary, M.D. Sinnott / Particuology 6 (2008) 419–444 429

Fig. 8. Mixing in V-blenders as the size of the blenders is increased: (a) laboratory scale, (b) scaled up by 3.16, (c) scaled up by 10, (d) scaled up by 10 in a different
phase of rotation, and (e) scaled up by 31.6, and (f) initial strata scaled up by 100.

tially be improved by removing the symmetry of the flow. Some particles are proportionally smaller. For the 186 l chamber, it
form of geometrical change that generates exchange of particles has become hard to distinguish individual particles amongst the
(even mild) between the sides of the blender will significantly 4.85 million that are present. The cost of the computation rises
increase the mixing rate since the cross-chamber mixing will with increasing scale both because the particle numbers increase
then have a convective component which will be much faster but also because the speed of revolution decreases in order to
than the currently used diffusive mechanism. maintain the Froude number.
Fig. 8 shows the particles in V-blenders at different stages The particles in the 5.88 l mixer are shown in Fig. 8b after
of mixing and different orientations for the different scales of 34 revolutions (83 s). The mixing is substantially advanced for
mixers considered here. At the laboratory scale the particles are materials on either side of the mixer but there is only weak
relatively large and visible. As the mixer scale increases the mixing across the plane of symmetry. The number of revolu-
430 P.W. Cleary, M.D. Sinnott / Particuology 6 (2008) 419–444

Table 2 4. Bladed mixers


Mixing states for the different size mixers measured after three revolutions and
the ratio of the mixing rate to that of the laboratory scale.
The gravity driven mixers examined in the previous sec-
Blender volume (L) Mixing measure after Ratio of mixing rate tion have at best a slow rate of mixing limited by the speed
three revolutions to the lab scale of avalanching down the free surface and the rate at which par-
1.86 7.2% 1.00 ticles are transported to the source location of these avalanches.
5.88 5.5% 0.76 At worst these devices behave more like transport equipment
18.6 4.2% 0.58 with a plug flow pattern, uniform residence time distribution
58.8 3.4% 0.47
and poor mixing between different regions and different mate-
rials. To improve both the rate and extent of mixing, one of the
tions is similar to that of the laboratory scale mixer in Fig. 8a options is to use some form of mechanical agitation of the bed.
(37 revolutions) and the degree of homogeneity of red/blue and With bladed mixers, the speed of agitation is moderate. It is suf-
white/green are similar but there is clearly less diffusive mixing ficient to generate more complex flow patterns and to increase
across the mid-plane. The mixing state is calculated in each case the collisional intensities but gravity remains an important con-
using the mixing measure described in Section 2.2. To facilitate tributor to the flow pattern. In most bladed mixers, the blade
comparison of the different scales we characterize the mixing dominates the flow behaviour near the blade while gravity con-
by the magnitude of the mixing after three revolutions for each trols the motion of the bed over large areas of the mixer away
case. The extent of mixing and the ratio of this measure to that from the agitator. Here we examine the behaviour of two bladed
of the laboratory scale mixer are given in Table 2. The rate of mixers.
mixing for the 5.88 l mixer is only 0.76 of the laboratory scale.
The particles are shown after 9.2 and 9.3 revolutions for the 4.1. Plough share mixer
18.6 l mixer in Fig. 8c and d. The bed is around 10% mixed with
the original colour sets still appearing to be reasonably coherent The laboratory scale plough share mixer used here consists
but with moderate diffusive mixing. The modest volume of red of a horizontal cylindrical shell of diameter 250 mm and length
particles that have crossed the mid-plane (Fig. 8d) are concen- 450 mm with flat end walls. One or more bent chevron shaped
trated along the angled lower wall of the right chamber of the plough blades (shown in Fig. 9) are mounted on a shaft of
mixer. The blue particles are heavily concentrated on the outside diameter 30 mm located along the centerline of the shell. The
of the left side and the white on the outside of the right side. The blade(s) extend outwards to have minimal clearance between
mixing percentage after three revolutions is 4.2% which is only them and the inside of the mixing chamber. The plough blade(s)
0.58 of the laboratory scale. The particles are shown after four and shaft rotate stirring a bed of particles. In this case the bed
revolutions in Fig. 8e for the 58.8 l mixer. As for the previous consists of rice grains of approximately 2 mm × 5 mm. These
case the blue and white particles are located at the outside of the are approximated as spherical particles with sizes distributed
left and right sides of the mixer and almost entirely at the top. between 2.5 mm and 5 mm. The DEM plough share blade and
It is interesting to note a narrow band of green and red leading containment vessel were constructed to exactly match the spec-
from the main groups of these colours down along the plane of ifications of the laboratory mixer used in the PEPT experiments
symmetry into the bottom of the access port. A reasonable frac- of Laurent, Bridgwater, and Parker (2000). The particle density
tion of this port is covered by these two colours. This shows that used was 1400 kg/m3 , the particle coefficient of restitution was
the internal convective flow patterns in this type of mixer are 0.3 and the friction coefficient was 0.75. A spring stiffness of
quite complex (compared to the simple planar recirculation pat- 4 × 106 N/m was used.
tern in drum mixer). The mixing degree for this case after three The granular flow induced by a single plough in a lab scale
revolutions is only 3.4% which is less than half of the laboratory plough share mixer has been investigated using both DEM
scale mixing rate. (Cleary et al., 2002) and PEPT (Laurent & Bridgwater, 2002;
From the mixing rates summarised in Table 2 it is clear that Laurent et al., 2000). They both showed the development of two
the rate of mixing does change with increasing scale when the loops of circulation one on either side of the plough. The parti-
speed of the mixer is scaled to maintain the same Froude num- cle displacements and therefore the mixing are controlled by the
ber. Although it is often customary to scale rotating devices to number of blade passes. Preliminary comparison of the PEPT
maintain the Froude number, there is absolutely no guarantee and DEM results has shown reasonable agreement (Laurent &
that this is the appropriate dimensionless quantity to preserve Cleary, submitted for publication).
during scale up to give consistent rates of mixing. For this V-
blender, scaling up the volume by 31.6 times has led to a halving 4.1.1. One blade mixer
of the mixing rate when measured in revolutions. The situation Fig. 10a shows the initial set up of the mixer with the particles
is much worse if the mixing rate is measured in seconds since coloured according to their initial position, so that we can see
the period of the rotation has also increased by a factor of 1.778 the extent and nature of the mixing. In each radial cross section
for the 58.8 l mixer. So the increase in the actual mixing period, the particles in the left half and right half have different colours.
which is of critical interest for industrial application, leads to a These colours are also different on either side of the blade in the
mixing rate that is actually 3.78 times slower at this scale than axial direction. This simulation has around 103,000 particles.
it was for the laboratory scale. The remaining frames of Fig. 10 show the system after 0.5, 1.0
P.W. Cleary, M.D. Sinnott / Particuology 6 (2008) 419–444 431

Fig. 9. Sketches of the top view and side view of the plough (dimensions in mm) on the left and two views of the CAD model used in the DEM on the right.

and 1.5 revolutions for a blade speed of 4 Hz. After half a rev- side of the blade path have flowed back down to largely fill the
olution, the blade has risen above the initial surface level of the trench left by the passage of the blade.
particles and is lifting a large number of particles and throwing At 1.5 revolutions the blade has again cleared a trench behind
them sideways leaving an empty trench behind it. Blue and green it and is throwing another group of particles. For this and sub-
particles are being pushed into the region initially occupied by sequent blade passages the mass of particles lifted is smaller
the red and yellow particles. After one revolution, the particles than the initial one since the surface level in the region through
lifted by the blade are travelling towards both ends of the mixer which the blade passes never returns to the original height. In the
in a coherent wave like structure. Meanwhile, particles to either axial direction, particles pile up near the ends to form a surface

Fig. 10. Particle distribution in the plough share mixer with blade revolving at 4 Hz during the first 1.5 revolutions of blade motion.
432 P.W. Cleary, M.D. Sinnott / Particuology 6 (2008) 419–444

Fig. 11. Bed motion in the 1 blade plough share mixer for a rotation rate of 4 Hz
with particles coloured by their axial speed (with blue being high speed to the
left and red being high speed to the right and green being zero axial motion).

that slopes down to a low point/trench where the blade passes.


In the radial plane of the plough blade, the surface is lower
on the side where the blade enters and higher on the exit side
for obvious reasons. The passage of the plough blade generates
a three-dimensional recirculation pattern on either side of the
blade, with particles rising with the blade, being flung sideways
onto the surface and then flowing back down the slope towards
the middle of the mixer and the blade entry point. This circu-
latory flow causes the mixing. After only 1.5 revolutions there Fig. 12. 2 Isosurfaces of speed for (a) 1.25 m/s, and (b) 1.0 m/s showing the
are already clear signs of mixing of the colours. Of particular structure of the flow pattern generated by the passage of one blade in the mixer.
importance is the presence of unmixed stagnant zones in the end These are coloured by axial velocity where red is 0.8 m/s directed to the left,
quarters of the mixer. This occurs because the perturbing effect and blue is 0.8 m/s to the right.
of the mixer blade is unable to extend this far into the particle
mass. region of very high speed arising from particles trapped on and
Fig. 11 shows the particles coloured by their axial speed (with flying away from the plough blade as it rotates. As the blade
blue being to the left and red to the right) when the blade has moves along this upper trajectory the volume of particles being
just reached vertical. The large mass of particles picked up by dragged along by the blade becomes smaller and so the high-
the blade are now travelling ballistically in the air above the speed region becomes narrower. It eventually disappears when
bed. The particles on the left of the blade center are moving there are no further particles moving with the blade just after
strongly to the left and the ones on the right are moving to the the blade reaches the forward horizontal position. Once the par-
right. The particles closer to the blade are moving more slowly ticles leave this high-speed region they are moving on rising
in the axial direction whilst the particles near the surface of the trajectories and so their vertical velocity declines due to gravity.
ballistic mass are travelling fastest. The avalanching return flow This means that their speed falls under the 1.25 m/s isosurface
where particles flow back into the trench behind the blade is also cut off and a large void region is created surrounding the long
clearly visible with particles on the left flowing to the right into slender neck. The particles reach the tops of their parabolic tra-
the trench and vice versa on the right. This shows the two key jectories or impact the mixer shell and then begin to fall. After
elements of the flow in a bladed mixer with particle behaviour a suitable period of fall they have accelerated to speeds again
dominated by the blade when near the blade but controlled by exceeding 1.25 m/s. This produces the large symmetric cloud
gravity elsewhere. structures at the ends of the mixer. There is a scattering of
Fig. 12 shows the structure of the complete flow pattern. The structure just above the bed produced by particles hitting the
instantaneous speed of all particles is averaged into a fixed Eule- bed before slowing. These slowed particles then fall below the
rian grid placed over the computational domain. This allows isosurface cut-off and are then not visible. The isosurfaces are
the calculation of a steady average velocity field for this com- coloured by their axial speed. This shows that all the material
plex cyclic flow. Isosurfaces of constant speed are then plotted. on the left is moving to the left and all the material on the right
They show regions for which the particles are travelling at or is moving to the right. The material in the slender neck region is
above this cut-off speed and allow the coherent structure of coloured close to green indicating that it has a low axial speed
the flow to be visualized. Fig. 12a shows the regions of the but a very high circumferential speed. The 1.0 m/s contour in
flow with a high speed of 1.25 m/s and above. There is a large Fig. 12b has a similar structure but the surfaces lie further from
region towards each end of the mixer and a long slender neck the blade and the outside the high-speed surface. Contours for
like structure rising from the bed at the back of the middle of progressively lower speeds form a sequence of growing nested
the mixer and extending up and forwards. This represents the shells.
P.W. Cleary, M.D. Sinnott / Particuology 6 (2008) 419–444 433

The rotation rate of the plough blade strongly affects the


nature of the flow in the mixer. At 4 Hz the blade is able to
throw a significant mass of particles. At the lower rotation rates
of 2.25 and 1 Hz the slower blade is much less able to throw
particles. At 4 Hz, the thrown particles reach both the top and
the ends of the mixer whilst at 2.25 Hz they just rise higher than
the shaft and barely reach the ends. Finally, for 1 Hz there is little
if any throw at all. After four revolutions at 4 Hz there is clearly
significant mixing already visible. At 2.25 Hz there is some mix-
ing but not quite as much. At 1 Hz there is clearly significantly
less mixing. The flow pattern is still a circulatory one, but it is
much more like a bulldozer pushing a wave of particles in front
of the blade until the slope is high enough for back flow to fill
the trench behind the blade. This is a gentle flow with mixing
being essentially by diffusion in the shearing regions rather than
convective mixing obtained when throwing masses of particles.
The percentage mixed (using the mixing measure) is shown
for the three rotation rates in Fig. 13. In all cases the mixing rate
is highest early and progressively declines as the bed becomes
progressively more mixed. The mixing rate is clearly rotation
rate dependent with 4 Hz providing the best mixing and 1 Hz
producing the worst. After 200 revolutions the % mixed is about
42%, 36% and 25%, respectively, for the 4 Hz, 2.25 Hz and 1 Hz
speeds. It is clear that the mixing states are trending to asymptotic
limits that are far from zero. This means that there is a limit to
how mixed this device can make the bed and this limit is strongly
rotation rate dependent. Based on the curves in Fig. 13 one might
expect the 4 Hz case to reach a 60–70% mixed state, whilst in
contrast one would expect the 1 Hz case be less than 30% mixed.
Fig. 14 shows the radial and axial centroid mixing measures
for the three rotation rates for very long times. Again, the radial
mixing is initially quite fast and tends to an asymptotic limit that
is rotation rate dependent. The extent of radial mixing increases
slowly over very long times. This is due to the axial mixing
limiting the extent of radial mixing. The slow transport of parti-
cles in the axial direction in the bed (towards the blade) enables
them to eventually reach the region where the blade can pick
up these particles and rapidly mix them in the radial direction.
So the slow axial flow also leads to slow increases in the radial
mixing measure. The radial mixing rates are again clearly blade
speed dependent, with 4 Hz producing about 80% mixed by 500
revolutions, whereas a 1 Hz rotation is struggling to reach 50%.
The axial mixing is shown in the right column of Fig. 14.
The early passages of the blade actually lead to permanent net
Fig. 13. Progress of the mixing for blade rotation rates of: (a) 4 Hz, (b) 2.25 Hz
migration of particles away from the blade path as the blade and (c) 1 Hz.
throws/pushes particles towards the ends (see Fig. 10). This con-
tinues until the slope produced is sufficient to generate a return
avalanching flow back towards the blade path that balances the 4.1.2. Four blade mixer
amount of material thrown towards the ends by the passage of Here we examine the flow in a more complex arrangement
the blade. This means that the axial centroids actually increase with four plough share blades arranged in a 90◦ offset spiral
their separation until this equilibrium is achieved and the magni- pattern. Fig. 15a shows the initial set up of the four blade mixer
tude is rotation rate dependent. These increases are around 20%, with the particles coloured according to the part of the bed that
10% and 5% for the three speeds respectively. There is strong they were originally located in with the colours changing on
symmetry in the axial mixing for the materials on either side of either side of each blade and on front and back halves of the
the blade. It is clear from Fig. 14 that the radial mixing is much mixer. This allows the extent and nature of the mixing to be
faster than the axial mixing and that the extent of axial mixing observed. This simulation contains around 250,000 particles for
is strongly dependent on the blade rotation rate. the 25% fill level. Fig. 15b shows the mixer after one revolution
434 P.W. Cleary, M.D. Sinnott / Particuology 6 (2008) 419–444

Fig. 14. The centroid locations of the four different colours/quadrants for rotation rates in the 1 blade plough share mixer for rotation rates: (a) 4 Hz, (b) 2.25 Hz and
(c) 1 Hz. (Left) the x-centroid component shows the radial mixing and (right) the axial mixing state.

at a 4 Hz blade speed. Each of the four blades is at a different is approaching the bed and the airborne particles from its last
phase of the mixing cycle. The leading blade (on the right) has pass are now falling back towards the bed surface. Each blade
just reached its highest point and is throwing a large stream of produces a similar flow pattern to the one found for the single
particles to either side. The next blade (lagging by 90◦ ) is just blade mixer.
emerging from the bed. The third blade is pointing down and After 1 s (four revolutions) as shown in Fig. 15c, there is
is pushing part of the bed ahead of it, whilst the fourth blade clearly significant radial mixing but only moderate axial mixing.
P.W. Cleary, M.D. Sinnott / Particuology 6 (2008) 419–444 435

4.2. Peg mixer

A peg mixer is an example of a moderate-speed mixer that


uses an agitator to mix the bed of particles. This type of mixer is
commonly used to reduce residence times and thus equipment
dimensions. Here we study a laboratory-scale version. It has a
cylindrical shell that is 300 mm in diameter and 600 mm long.
Agitation was achieved using an axially mounted peg-agitator
consisting of 18 pins that are mounted on a 50 mm shaft and
arranged at equal intervals on a helical path. Each pin is 15 mm
in diameter and 125 mm in length with ends that move just
inside the surrounding shell. Simulations were performed with
the mixer rotating anti-clockwise at 60 rpm for a period of 30 s.
Feed material was introduced through a 60 mm inclined port at
the left end. The feed rate is chosen to match the discharge rate
at the other end and ensure that the bed mass remains constant
at 22.0 kg. The particles are spherical with density 2000 kg/m3
and have a size range of 2.5–8.8 mm. The particle coefficient of
restitution was 0.3 and the friction coefficient was 0.75. A spring
stiffness of 5000 N/m was used.
Fig. 16 shows the steady-state flow behavior inside the peg
granulator. The particle motion is driven by the pegs digging into
the bed. Apart from the obvious radial mixing induced by this
action, the pegs also generate a degree of axial movement. As a
peg enters the bed, it pushes material to either side, leading to the
formation of a mound in front of the peg. When the peg rotates
out of the bed this mound collapses creating flow in the axial
direction. The helical distribution of the pegs means that there is
a consistent bias in the motion generated towards the discharge
end. The particles near the pegs clearly move more rapidly than
Fig. 15. Particle distribution in a four blade plough share mixer rotating at 4 Hz. those in quiescent regions between the lines of pegs. The peg
spacing is sufficiently close that the higher speed (green) regions
produced by successive pegs are connected. In the side view, it
The axial mixing is faster than for the single blade case because can be seen that each line of pegs currently buried in the bed
the blade spacing is reduced allowing blades to throw particles creates a weak jet of material that escapes from the bed to the
directly over adjacent blade paths. Each blade leaves a trench left. In between these jets of particles moving to the left are slow
empty behind it that fills gradually by a surface avalanching flow. return flows of particles moving back to the opposite side of the
Between these trenches are relatively stable stagnant zones that mixer. So the bed has a complex flow pattern induced by the
cannot be mixed even by blades with such a close spacing. This pegs with particles near the lines of pegs moving with the pegs
demonstrates the difficulties in obtaining complete mixing in and a counter flow between the pegs of particles returning. This
industrial mixers. creates three recirculation zones which migrate axially along the

Fig. 16. Steady-state motion in the laboratory scale peg granulator, (left) view of the entire peg granulator contents and (right) an end view of a slice through the bed
at 500 mm (immediately before the conical outlet). Particles are colored by velocity magnitude (blue is slow and red is fast).
436 P.W. Cleary, M.D. Sinnott / Particuology 6 (2008) 419–444

mixer length towards the feed end. The cross-section view also
shows that the particle speeds are much higher near the surface
than they are deep inside the bed. So the pegs have strong local
effects on the particles, but when the pegs are not nearby the
particle motion slows considerably to near rest.
Fig. 17 shows a view along the axis of the mixer, but with the
particles in the discharge half of the mixer not shown. The parti-
cles are coloured by their size (with blue being fine particles and
red being coarse). This shows that there is very strong size seg-
regation being generated inside the bed by the motion induced
by the passage of the pegs. There is a clear concentration of fine
material on the right where the pegs first enter the bed and there
are only coarse particles on the left side where the pegs leave the
bed. The fine particles are more easily able to avoid the pegs and
percolate between the large particles which are much less able
to avoid being pushed forward by the pegs. Such strong segre-
gation effects are very undesirable in a device whose design aim
is to mix particles. This is potentially a problem with any device Fig. 17. Particles coloured by size (with blue being fine particles and red being
coarse) viewed along the axis of the mixer showing significant size segregation
using an agitator since the high shear induced near the blades
within the bed.
can lead to size and also density segregation.
Fig. 18 shows the progress of feed material (red) through recirculation zones set up by the peg movement. Direct interac-
the original bed of particles (blue). The feed material falls onto tion with the pegs gives the particles high velocities leading to
the rotating shaft and is distributed over the bed near the feed strong local mixing, particularly close to the free surface where
opening. Particles are moved around by the axially migrating the impacts with the pegs can move individual particles a signif-

Fig. 18. Motion of feed particles (colored red/dark grey) into an existing bed of particles (light blue/light grey) in a laboratory scale peg mixer.
P.W. Cleary, M.D. Sinnott / Particuology 6 (2008) 419–444 437

Fig. 19. (a) Residence time distribution of feed particles over 15 revolutions of the peg mixer and, (b) collision energy spectra for all collisions over the same period.

icant distance. In the second frame, the feed material dominates Lennartsson, Rasmuson, Bjorn, & Folestad, 2007) and are
the first quarter of the mixer with some feed particles having commonly used for granulation, pelletisation, emulsification,
already passed the middle of the mixer. In the third frame, the agitation and dispersion. In these mixers, gravity plays a less
feed material is around two thirds of the way along the mixer but important role than in the lower speed bladed mixers. DEM
the leading feed material is already discharging. There are also models of flow in laboratory scale high shear mixers fitted with
original blue particles visible that are confined close to the feed a blade impeller have been developed by Stewart, Bridgewater,
end of the mixer. The flow behaviour is clearly quite different and Parker (2001), and Stewart, Bridgewater, Zhou, et al. (2001)
to the plug flow observed in the drum mixer and the interface and validated against PEPT data for spherical particle assem-
between the blue and red particles is much less sharp. blies. The flow patterns, shape of the bed and velocity fields
The residence time distribution of feed material is shown within the packed bed have been well described. Other lab
in Fig. 19a. The mean residence time is significantly less than scale high shear mixers have been studied by Zhou, Yu, and
for a plug flow in a drum mixer and the distribution is broad Bridgewater (2003), Kuo et al. (2004) and Bertrand et al. (2005).
with a number of distinct modes. The breadth of the distribution To date there are no DEM models for more complex, production
occurs because the pegs push particles towards both the feed and scale high shear mixers due to issues of scale and the complexity
discharge ends. Therefore many particles move much faster than of geometry and operation of such devices. Some experimental
the plug flow speed, whilst others move much more slowly (after studies of more novel devices do exist (Ng, Kwan, Ding, Ghadiri,
being pushed backwards repeatedly). Such a broad residence & Fan, 2007; Ng et al., 2008).
time distribution will not be desirable for granulation, where it Here we study a laboratory scale high shear mixer. It con-
will contribute to broader aggregate size distributions. However sists of a vertical cylindrical shell fitted with a disc impeller
it will be positive for mixing of non-cohesive or weakly cohesive or with a rectangular blade impeller. These impellers rotate
particles. around the center of the mixer at high speed. This has previously
The collision energy spectra for the peg-mixer are shown been used in Sinnott and Cleary (2003). The key dimensions
in Fig. 19b. In addition to the spectra for all particle colli- of these mixers are given in Fig. 20. The ends and the bot-
sions, the contributions of collisions with the pegs and walls tom of blade impeller are almost in contact with the floor and
are also shown. The higher energy collisions (i.e. 10−4 –10−3 J) walls of the mixer so there is a negligible clearance. The sim-
are dominated by interactions between particles and the pegs. ulations used a spring constant of 1000 N/m. Particle–particle
The impacts with the pegs are quite strong and are likely and particle–boundary collisions were both modelled using a
to cause aggregate breakage when this device is used for coefficient of restitution of 0.9 since the material being mixed
agglomeration. This will have a strong influence on its final is quite elastic. The particle density was 2500 kg/m3 . Stewart,
product size. In contrast to the drum mixer,where the impacts Bridgewater, and Parker (2001) and Stewart, Bridgewater, Zhou,
are 5–6 times weaker, there is much more energy available et al. (2001) found the structure of particle flows in this type of
for control of aggregate size. In terms of mixing, this device mixer was relatively insensitive to specific coefficients of fric-
appears to have reasonable mixing characteristics despite the tion in the range 0.3–0.5. So a coefficient of friction of 0.4 was
tendency to segregate fine material towards one side of the chosen for both particle–particle and particle–boundary inter-
bed. actions. Particles were represented as spheres with diameters
ranging from 1.45 to 1.55 mm. A friction coefficient of 0.4 was
5. High shear mixers used for all particle collisions. Based on the cylindrical sym-
metry of the mixer, particles were classified into distinct spatial
High shear mixers are batch mixers that employ an impeller regions (quadrants, annuli, and horizontal strata) and coloured
rotating at high speed to strongly agitate the granular material. in order to separately study the different mixing regimes: radial,
Such mixers can be used for dry or wet materials (Darelius, azimuthal and vertical mixing respectively.
438 P.W. Cleary, M.D. Sinnott / Particuology 6 (2008) 419–444

Fig. 20. High shear mixer geometries fitted with the blade and disc impellers.

5.1. Disc impeller expression for constant angular velocity.



7g
Azimuthal mixing patterns for the disc impeller show simple ωD = (4)
flow behaviour. A single quadrant is displayed in Fig. 21 for 9R
various times during the mixing process. The other quadrants The fluctuations seen near the centre of the bed are due to data
are present but not shown. Particles are coloured by their height sparsity. This shows that the inner sections of the bed rotate as a
in the bed with blue particles in the lowest layer adjacent to the rigid body with an angular velocity of ωD out to approximately
disc and red particles in the surface of the heap formed near the half the mixer radius. Near the wall, frictional forces from the
wall due to centrifugal forces. The motion of the particle bed container wall dominate and the angular velocity of particles is
is driven by the drag force from the rotation of the disc below. reduced substantially. This leads to cumulative shear strain in
Particles near the wall experience a retarding drag force due to the circumferential direction creating the vortex patterns. This
particle–wall friction. The central core of the bed rotates as a is the sole mechanism for mixing with this impeller.
near rigid body and the outer bands wrap around this central
region in a spiralling pattern as shown in Fig. 21b. 5.2. Blade impeller
Fig. 21c shows that, even after three revolutions, the spiralling
band structure remains coherent and clearly visible, and there Flow behaviour for the mixer using the blade impeller is more
appears to be no significant radial migration across the spiral complex. Fig. 23 shows the interaction between the blade and
band structure. The radial mixing pattern is shown in Fig. 21d particle bed where only particles in a single quadrant are visible
for the full bed. Here particles have been divided into annular and coloured by their height in the bed. In Fig. 23a, particles
regions and coloured respectively. After 13.4 revolutions, there are pushed up into a heap in front of the approaching blade. The
has been almost no exchange of particles between these groups blade sweeps through the bed initially setting up three distinct
indicating that no radial mixing results from the effect of the disc vertical layers (at least for the outer radial region of the bed).
impeller on the bed. This occurs because, for the disc impeller, Particles flowing over the blade are deposited into the empty
the only direction that the bed can experience shear is in the space left behind by the impeller and fall towards the floor of
direction of rotation. This leads to cumulative strain of the bed the mixer as shown in Fig. 23b. Particles in the next quadrant
by the wall and by the floor, thus preventing the powder from begin to flow over the blade and accumulate on top of the bottom
rotating fully as a rigid body. layer of particles from the first quadrant. This stratifies particles
The average angular velocity of the particles in the disc in neighbouring quadrants into horizontal layers that are offset
impeller, as a function of radial position, is plotted in Fig. 22. The in the azimuthal direction as a result of the shearing by the blade.
dashed line represents the powder mechanics analytical solution Fig. 23c shows the surface bed layer and the layer near the
of Knight, Seville, Wellm, and Instone (2001) for this system. floor for the single quadrant after 1.5 revolutions, at which time
They assumed rigid body motion for the powder, deriving an the selected particles are distributed around the entire circumfer-
P.W. Cleary, M.D. Sinnott / Particuology 6 (2008) 419–444 439

Fig. 21. (a)–(c) Mixing of a single quadrant by the disc impeller with particles coloured by height and (d) top view of four annular regions showing the lack of radial
mixing in the bed after 8 s.

ence of the mixer. Particles at the bottom of the mixer are pushed 5.3. Comparison of mixing for disc and blade impellers
in front of the blade around to the next blade further enhancing
the displacement of the vertical layers. Near the surface of the Mixing measurements have been made for each impeller type
fast moving heap that precedes the blade, particles continue to in order to assess the relative mixing performance. The evolution
avalanche down the front of this incline and are trapped for long
periods within this recirculating heap.
To better observe the presence of these layers, Fig. 23d shows
all four quadrants visible at the same instant of time as that in
Fig. 23b. These four quadrants are coloured yellow, red, white
and blue respectively as a visual aid to the mixing. The layer
of red particles visible beneath the bulk of the white quadrant
is the result of a previous blade pass. This bottom layer is still
relatively unmixed with the white particles. Subsequent blade
crossings will begin to mix across this boundary as particles flow
over the blade. At the same time, red particles have been spread
across the white quadrant in a thin layer. Behind the blade there
is a radial surface flow down the heap toward the centre of the
mixer. The motion of the blade causes surface particles in front
of the heap to be thrown radially outward diffusing the already
thin surface layer near the walls. Mixing continues rapidly in Fig. 22. Predicted average angular velocity versus normalised radial position
these top and bottom layers. for particles for the high shear mixer with a disc impeller.
440 P.W. Cleary, M.D. Sinnott / Particuology 6 (2008) 419–444

Fig. 23. Progress of mixing of a single quadrant by the blade impeller: (a)–(c) particles are coloured by height to show the motion in the top and bottom layers and
(d) mixing of the four particle quadrants after 0.6 revolutions.

of the mixing state with time is shown in Figs. 24 and 25, for measure of 65% after 13 revolutions. The rate of decline sug-
the disc and blade impellers respectively. Fig. 24a shows strong gests that the asymptotic limit for azimuthal mixing will be only
azimuthal mixing for the disc impeller with most of the mixing slightly larger than this. The mixing is far from complete due
completed within five revolutions. The curve shows an expo- to the inability of the rigid rotating core to mix. From Fig. 22,
nentially decreasing mixing rate with time leading to a mixing the radial size of the central core is half the mixer radius which

Fig. 24. The mixing state for a high shear mixer with a disc impeller as a function of time for: (a) azimuthal and (b) radial mixing.
P.W. Cleary, M.D. Sinnott / Particuology 6 (2008) 419–444 441

Fig. 25. The mixing state for a high shear mixer with a blade impeller as a function of time for: (a) azimuthal and (b) radial mixing.

equates to 25% of the bed volume that will not mix. Fig. 24b blade shearing the bed. As the blade rotates through one period
shows very clearly that there is no radial mixing at all for the it continually encounters new material. Thus most particles in
disc impeller. As mentioned previously, due to the geometry the bed have a similar chance of experiencing radial migration
of the disc impeller and its cylindrical symmetry there is no when they encounter the blade regardless of radial position. This
mechanism for inducing radial transport. The combination of produces the observed linear increase in radial mixing with time
both mixing measures shows that this impeller is incapable of and after 14 revolutions. The simulation was not run for enough
producing complete mixing of the particulates. revolutions to predict the final asymptotic state for the blade
The rate of azimuthal mixing is very rapid for the blade case, impeller, but based on the degree of mixing shown in Fig. 25
leading to almost complete mixing in only 13 revolutions. Unlike there is a strong probability that this mixer will mix the parti-
the disc impeller, mixing with the blade shows no evidence of cles completely. This is a highly desirable attribute of a mixer
large coherent non-mixing regions and the final mixing state and one not possessed by a number of the mixers studied in this
asymptotes towards a near homogeneous mixture. The azimuthal paper.
mixing is faster due to the complex flow pattern over the blade. The lack of radial transport of particles when using the disc
This results from a combination of rapid circumferential flow impeller marks the greatest difference in the mechanism driv-
over the blade and the three-dimensional recirculation within ing the mixing for the two configurations. Azimuthal mixing is
the heap of material which is pushed in front of the blade. The slower for the disc impeller than the case using the blade, and
shape of the curves in Figs. 24 and 25a arises because material leads to an incompletely mixed final state due to the existence of
in the outer radial band mixes more rapidly in the azimuthal a rigidly rotating core. Complex flow patterns in the azimuthal
direction and the change in mixing state is almost linear up until direction for the blade impeller arise from particles constrained
the time when a single quadrant is able to wrap all the way to move with the blade, and lead to more thorough mixing.
around the mixer and start interacting with itself. The inner core Superimposed on the mixing measures for the blade impeller
mixes more slowly, and in the case of the disc impeller not at high shear mixer (see Fig. 25a) are small regular oscillations
all. As the outer radial band becomes well mixed, the possibility
of further mixing of this region is then reduced and the mixing
state tends towards an asymptotic limit.
The motion of the blade impeller causes strong radial mix-
ing (see Fig. 25b). Initially, this is slower than the azimuthal
mixing but the rate is almost constant and so the radial mixing
state catches up with the azimuthal mixing after about 14 revo-
lutions. Particle migration occurs along the length of the blade
into and out from the center of the mixer. Radial migration is
localized to a narrow band surrounding the blade due to the

Table 3
Extent of azimuthal and radial mixing for the disc and blade impellers after 13
revolutions.
Disc Blade

Azimuthal Radial Azimuthal Radial

Amount of mixed 65% 7% 87% 75% Fig. 26. Azimuthal mixing rates in a high shear mixer using the blade and disc
impellers.
442 P.W. Cleary, M.D. Sinnott / Particuology 6 (2008) 419–444

Fig. 27. Progress of mixing of four equal volume annular regions in a high shear mixer using a blade agitator for: (a) spherical particles and (b) super-quadric
particles.

due to the passage of the blade. The frequency is 2.5 Hz, which curves are similar in shape but the blade impeller has a consis-
is between 1 and 2 times the blade rotation rate. This suggests tently higher mixing rate. There is an initial linear increase in
that the bulk circulation rate of the particles is less than the mixing rate up until around three revolutions when the peak-
blade rotation rate due to the retarding effects of the walls. The mixing rate is achieved. This is the amount of time required
mixture quality for the azimuthal and radial mixing regimes after for significant material in each quadrant to travel the full dis-
13 revolutions for each impeller case is given in Table 3. The tance around the circumference of the mixer shell and begin
blade impeller is clearly a much more efficient mixing impeller, mixing with itself again. The mixing rates then begin to drop.
both in azimuthal and radial mixing. By four revolutions the rates have begun to decrease logarithmi-
The variation of the azimuthal mixing rate throughout the cally. As time progresses and the amount of material remaining
mixing process is shown in Fig. 26 for each impeller. The mix- to mix declines the mixing rate declines proportionally trending
ing rates are calculated by differentiating the mixing state. Both asymptotically towards zero.
P.W. Cleary, M.D. Sinnott / Particuology 6 (2008) 419–444 443

Table 4 shows strong azimuthal and vertical mixing, but only moderate
Extent of mixing for spherical and superquadric cases and for each mixing radial mixing within 10 blade revolutions. Super-quadric parti-
regime after 10 blade revolutions.
cles show a strong 14% reduction in radial mixing relative to the
Particle shape Azimuthal Radial Vertical spherical case. The azimuthal mixing is just 4% lower, and the
Spherical 87% 69% 73% vertical mixing state is actually 3% greater than for the spheri-
Super-quadric 83% 55% 76% cal particles. The reduced radial mixing is a consequence of the
strengthened microstructure of the blocky, elongated particles
and the resultant increase in resistance to shear. Flow parallel
5.4. Effect of particle shape on mixing for a blade impeller to the blade therefore proceeds very slowly (Fig. 27b). Particle
shape appears to quantitatively affect particle mixing and impor-
To evaluate the effect that particle shape has on the flow tantly its effect is different for different directions of mixing. The
and mixing in a bladed high shear mixer we compare closely inclusion of particle shape augments vertical mixing, moderately
matched simulations with spherical and super-quadric particles. slows azimuthal mixing and sharply reduces radial mixing.
The super-quadric particles were chosen with a maximum char-
acteristic length of 1.45–1.55 mm so that they had the same 6. Conclusions
particle volume and number of particles as used for the spherical
particle case. The intermediate and minor axis aspect ratios were Mixing of particulates is a critical industrial process which
randomly chosen to be in the range 0.7–1.0. The blockiness was has been traditionally hard to improve by empirical methods
chosen to be in the range 2.5–6.0 giving particle shapes vary- because of the difficulty in making measurements of particulates
ing from fairly rounded to moderately angular. These shapes are in enclosed process equipment. DEM is increasingly viable as
reasonable representations for grains and many household prod- an alternative method for investigating and optimizing mixing
uct components. The coordination numbers were 13.6 and 9.2 and granulation equipment design and operation. Simulation
for the super-quadric and spherical particles respectively. The is effective at predicting information on the performance of
50% increase in the connectivity of the particles when they have different mixers which is difficult and expensive to obtain using
plausible shape attributes is one of the contributors to the much direct experiment. In this paper we have used DEM to better
greater strength of these materials and their much greater abil- understand flow dynamics and assess mixing in several different
ity to resist shear. This is demonstrated by the increase in the types of devices. We have classified mixers broadly as gravity
power consumption of the high shear mixer which increases controlled, bladed and high shear depending on the relative con-
from 0.45 W for spherical particles up to 0.61 W for super- tribution of gravity and internal agitation. These classifications
quadric particles. This almost 50% increase in power arises depend on the generation of the flow patterns that promote mix-
because the impeller needs to do much more work to overcome ing. In the area of gravity controlled mixing we have been able
the increased bed resistance created by the interlocking of the to analyse the performance of continuously fed drum and pan
angular and elongated super-quadric particles. mixers and batch V-blenders. For the bladed mixers, where both
The mixing of annular regions in Fig. 27 shows a marked agitation and gravity play a critical role, we have analysed the
difference in the radial transport of SQ and spherical particles. flow performance for a batch plough share mixer and a contin-
The SQ annuli remain fairly coherent with limited diffusion uously fed peg mixer. For high intensity agitation where gravity
between annular regions after a single blade pass, whereas the is a minor contributor to the flow behaviour we have explored
spherical particles show significant stretching of the central the effect of the agitator type for batch high shear mixers.
annular regions along the length of the blade and a contrac- Two methods have been used to quantify mixing in all the
tion in the direction normal to the blade. Surface particles in mixer examples explored in this paper. One is based on the
the inner annuli rapidly diffuse outward, completely covering change in location of centroids of groups of particle chosen to
the outer radial regions. This change in flow behaviour results measure mixing in specific directions. The other is an effective
from the much higher flowability of the spheres relative to the method for determining the overall degree of homogeneity of
stronger microstructure of the super-quadrics. On this basis, the the mixtures. Both methods have allowed us to quantitatively
timescales for radial mixing might reasonably be expected to be explore the degree, speed and quality of mixing across a broad
faster for spheres than for SQs. range of mixing devices.
Mixing measurements have again been made for each mixing Using the single blade high shear mixer as the example we
regime: azimuthal, radial and vertical. The shape of the mixing have explored the effect of particle shape on mixing perfor-
curves is not greatly different for the spherical and super-quadric mance and found that it has an extremely strong impact. The
cases for each mixing regime. Thus we will only describe the shape of the particles, in terms of their angularity and elongation,
final mixing states for each regime after 10 blade revolutions. affects the degree of connectivity of particles to their neighbours
The percentage of material mixed after 10 blade revolutions and the degree to which they can lock together. This increases
is given for each case in Table 4. For the spheres case, mix- the strength of the material and its ability to resist shear. This
ing occurs most rapidly in the azimuthal direction and is 87% changes the internal balance of forces and the places where the
mixed at this time. The radial and vertical mixing regimes are particle microstructure fails and flows. All these impact on the
markedly slower than the azimuthal direction with both showing nature of the mixing performance. For the high shear mixer,
about 70% mixing at the same end time. The super-quadric case moderate non-sphericity leads to a sharp reduction in the radial
444 P.W. Cleary, M.D. Sinnott / Particuology 6 (2008) 419–444

transport of particles and a matching strong reduction in mixing Ding, Y. L., Forster, R. N., Seville, J. P. K., & Parker, D. J. (2001). Scaling relati-
efficiency in that direction. Mixing in vertical and circumferen- onships for rotating drums. Chemical Engineering Science, 56, 3737–3750.
Finnie, G. J., Kruyt, N. P., Ye, M., Zeilstra, C., & Kuipers, J. A. M. (2005). Lon-
tial directions was only mildly affected. Shape can be expected
gitudinal and transverse mixing in rotary kilns: A discrete element method
to have similarly strong effects in other mixers, but the specific approach. Chemical Engineering Science, 60, 4083–4091.
nature of the changes depends sensitively on the nature of the Knight, P. C., Seville, J. P. K., Wellm, A. B., & Instone, T. (2001). Prediction of
particulate flow pattern. impeller torque in high shear powder mixers. Chemical Engineering Science,
With DEM modelling now possible for millions of particles 56, 4457–4471.
more complex process issues such as how to scale up mixer Kuo, H. P., Knight, P. C., Parker, D. J., Adams, M. J., & Seville, J. P. K. (2004).
Discrete element simulations of a high-shear mixer. Advanced Powder Tech-
designs from laboratory to operating scales can now be investi- nology, 15, 297–309.
gated. Here we used models at five scales to help understand the Kuo, H. P., Knight, P. C., Parker, D. J., & Seville, J. P. K. (2005). Solids circu-
speed scaling of a V-blender with the largest model having 4.85 lation and axial dispersion of cohesionless particles in a V-mixer. Powder
million particles for a 186 l mixing vessel. The often used Froude Technology, 152, 133–140.
number scaling method is not found to be suitable for predicting Kuo, H. P., Knight, P. C., Parker, D. J., Tsuji, Y., Adams, M. J., & Seville, J.
P. K. (2002). The influence of DEM simulation parameters on the particle
mixing rates at the larger scale. A reduction of 50% in mixing behaviour in a V-mixer. Chemical Engineering Science, 57, 3621–3638.
rate per revolution was found in scaling up from laboratory scale Kwapinska, M., Saage, G., & Tsotsas, E. (2006). Mixing of particles in rotary
to the 58.8 l mixer. This represents an almost fourfold increase in drums: A comparison of discrete element simulations with experimental
the mixing period to reach a specific target level of homogeneity. results and penetration models for thermal processes. Powder Technology,
161, 69–78.
Laurent, B. F. C., & Bridgwater, J. (2002). Dispersive granular flow in a hori-
Acknowledgment zontal drum stirred by a single blade. AIChE Journal, 48, 50–58.
Laurent, B. F. C., Bridgwater, J., & Parker, D. J. (2000). Motion in a particle bed
The authors would like to acknowledge the contribution of stirred by a single blade. AIChE Journal, 46, 1723–1734.
Dave Morton to the modelling of the drum, peg and pan mixers. Laurent, B. F. C., & Cleary, P. W. (submitted for publication). Comparative study
of DEM and experimental results of flow patterns in a ploughshare mixer.
Lemieux, M., Bertrand, F., Chaouki, J., & Gosselin, P. (2007). Comparative
References study of the mixing of free-flowing particles in a V-blender and a bin-blender.
Chemical Engineering Science, 62, 1783–1802.
Arratia, P. E., Duong, N., Muzzio, F. J., Godbole, P., Lange, A., & Reynolds, Lemieux, M., Leonard, G., Doucet, J., Leclaire, L. A., Viens, F., Chaouki, J., et al.
S. (2006). Characterizing mixing and lubrication in the Bohle Bin blender. (2008). Large-scale numerical investigation of solids mixing in a V-blender
Powder Technology, 161, 202–208. using the discrete element method. Powder Technology, 181, 205–216.
Arratia, P. E., Duong, N., Muzzio, F. J., Godbole, P., & Reynolds, S. (2006). A Metcalfe, G., Shinbrot, T., McCarthy, J. J., & Ottino, J. M. (1995). Avalanche
study of the mixing and segregation mechanisms in the Bohle Tote blender mixing of granular solids. Nature, 374, 39–41.
via DEM simulations. Powder Technology, 164, 50–57. Morton, D., & Cleary, P. W. (2002). Discrete element modeling of three lab-
Barker, G. C. (1994). Computer simulations of granular materials. In A. Mehta oratory scale granulators. In Proceedings of the World congress particle
(Ed.), Granular matter: An inter-disciplinary approach. Berlin: Springer. technology 4 (CD paper 543) 21–25 July, Sydney, Australia.
Bertrand, F., Leclaire, L. A., & Levecque, G. (2005). DEM based models Ng, B. H., Kwan, C. C., Ding, Y. L., Ghadiri, M., & Fan, X. F. (2007). Solids
for the mixing of granular materials. Chemical Engineering Science, 60, motion in a high shear mixer granulator: A comparison between dry and wet
2517–2531. conditions. In International conference on particle technology, Partec2007
Chaudhuri, B., Mehrotra, A., Muzzio, F. J., & Tomassone, M. S. (2006). Cohesive Nuremberg, Germany.
effects in powder mixing in a tumbling blender. Powder Technology, 165, Ng, B. H., Kwan, C. C., Ding, Y. L., Ghadiri, M., Fan, X. F., & Parker, D. J.
105–114. (2008). Granular flow fields in vertical high shear mixer granulators. AIChE
Cleary, P. W. (1998). Discrete element modelling of industrial granular flow Journal, 54, 415–426.
applications. TASK Quarterly—Scientific Bulletin of the Academic Computer Sinnott, M. D., & Cleary, P. W. (2003). 3D DEM simulations of a high shear
Centre in Gdansk, 2, 385–416. mixer. In Third international conference on CFD in the minerals and process
Cleary, P. W. (2004). Large scale industrial DEM modeling. Engineering Com- industries 10–12 December, Melbourne, Australia.
putations, 21, 169–204. Sinnott, M. D., & Cleary, P. W. (2005). Mixing of dry powders with non-spherical
Cleary, P. W. (2006). Axial transport in dry ball mills. Applied Mathematical shapes. In Seventh World congress on chemical engineering 10–14 July,
Modelling, 30, 1343–1355. Glasgow, Scotland.
Cleary, P. W., Laurent, B., & Bridgewater, J. (2002). DEM prediction of flow Stewart, R. L., Bridgewater, J., & Parker, D. J. (2001). Granular flow over a
patterns and mixing rates in a ploughshare mixer. In Proceedings of the flat-bladed stirrer. Chemical Engineering Science, 56, 4257–4271.
World congress of particle technology 4 (CD paper 715) 21–25 July, Sydney, Stewart, R. L., Bridgewater, J., Zhou, Y. C., & Yu, A. B. (2001). Simulated
Australia. and measured flow of granules in a bladed mixer—A detailed comparison.
Cleary, P. W., & Metcalfe, G. (2002). Quantitative comparison of mixing rates Chemical Engineering Science, 56, 5457–5471.
between DEM and experiment in a slowly rotating cylinder. In Proceedings Sudah, O. S., Coffin-Beach, D., & Muzzio, F. J. (2002). Effects of blender
of the World congress of particle technology 4 (CD paper 550) 21–25 July, rotational speed and discharge on the homogeneity of cohesive and free
Sydney, Australia. flowing mixtures. International Journal of Pharmaceutics, 247, 57–68.
Cleary, P. W., Metcalfe, G., & Liffman, K. (1998). How well do discrete element Williams, J. R., & Pentland, A. (1992). Super-quadrics and modal dynamics for
granular flow models capture the essentials of mixing processes? Applied discrete elements in interactive design. International Journal of Computer-
Mathematical Modelling, 22, 995–1008. Aided Engineering—Engineering Computations, 9, 115–127.
Cleary, P. W., & Sawley, M. (2002). DEM modelling of industrial granular flows: Zhou, Y. C., Yu, A. B., & Bridgewater, J. (2003). Segregation of binary mix-
3D case studies and the effect of particle shape on hopper discharge. Applied ture of particles in a bladed mixer. Journal of Chemical Technology and
Mathematical Modelling, 26, 89–111. Biotechnology, 78, 187–193.
Darelius, A., Lennartsson, E., Rasmuson, A., Bjorn, I. N., & Folestad, S. (2007). Zhou, Y. C., Yu, A. B., Stewart, R. L., & Bridgewater, J. (2004). Micrody-
Measurement of the velocity field and frictional properties of wet masses in namic analysis of the particle flow in a cylindrical bladed mixer. Chemical
a high shear mixer. Chemical Engineering Science, 62, 2366–2374. Engineering Science, 59, 1343–1364.

You might also like