Chemical Engineering Science 94 (2013) 93–107

Contents lists available at SciVerse ScienceDirect

Chemical Engineering Science

journal homepage: www.elsevier.com/locate/ces

De-mixing of binary particle mixtures during unloading

of a V-blender
G.G. Pereira n, P.W. Cleary
CSIRO Mathematics, Informatics and Statistics, Private Bag 33, Clayton South, Victoria 3169, Australia

H I G H L I G H T S

c De-mixing that occurs during unloading of a V-blender is significant.

c Sloping walls of V-blender lead to shear induced segregation during unloading.
c Shear induced segregation occurs on the surface of heaped granular matter.
c For particles which differ in size percolation drives de-mixing.

a r t i c l e i n f o abstract

Article history: The V-blender is an industrial batch mixing device which is used to blend granular mixtures in the food,
Received 7 May 2012 household products, agricultural and pharmaceuticals industries. The mixing characteristics of this
Received in revised form device have been previously studied both experimentally and numerically. When emptying the device,
21 January 2013
de-mixing of this granular mixture can occur due to segregation produced by the flow. This reverses
Accepted 22 February 2013
Available online 1 March 2013
some fraction of the mixing produced by the blender. In this study we consider the de-mixing that
occurs when the particles flow out of the device. We begin with close to a perfectly mixed state and
Keywords: measure the change in particle homogeneity during discharge. In this initial study, we consider only the
Discrete element method simplest case of a binary mixture with components only differing by size. The extent of de-mixing is
Segregation
evaluated as a function of the size ratio of the mixture components. Different unloading scenarios are
De-mixing
considered and it is shown that the presence of additional transferring surfaces can greatly increase
Unloading
Percolation the segregation of the mixture. We also consider how to optimise the set-up so as to minimise any
Homogeneous de-mixing.
Crown Copyright & 2013 Published by Elsevier Ltd. All rights reserved.

1. Introduction referred to review articles by Ottino and Khakhar (2000); Meier

et al. (2007) or Poux et al. (1991).
Mixing and segregation in granular media is not only of The V-blender is one of the most common batch mixing
fundamental scientific interest but also has a myriad of industrial devices used in industry. It is composed of two arms, where each
applications that range from pharmaceuticals, food processing, arm has a cylindrical cross-section and joined at a 75–901 angle
plastics, mining, metallic products, ceramic products, cosmetics (Brone et al., 1998). There is typically an access port at the bottom
and household products (Ennis et al., 1994; Terayama et al., 2005; of the V-blender for filling and discharge of the particulates.
Simpson et al., 2001). In most applications, mixing is required and The V-blender normally rotates about a horizontal axis passing
the quality of the final product depends critically on the degree through the middle of the device and mixing of particles occurs in
of mixing of the constituent materials. However, in other cases, the ensuing shearing bulk flow. A snap-shot of a typical V-blender
separation of granular materials according to some material geometry is show in Fig. 1. Fig. 1a is in the y–z plane, with the
property is desirable. For more details regarding the various gravity vector pointing in the negative y direction. The origin of
mixer/segregation devices that have been used the reader is our system is located at the centre of the V-blender access port
(directly below the apex of the V). Fig. 1b shows the V-blender in
a different (rotated) view looking down one of the axes on one
arm of the blender.
n
Corresponding author.
The V-blender is mainly used to mix powders so that the
E-mail addresses: [email protected] (G.G. Pereira), final product will be homogenous. Before the 1990s the main
[email protected] (P.W. Cleary). method to analyse experimentally whether the final mixture was

0009-2509/$ - see front matter Crown Copyright & 2013 Published by Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.ces.2013.02.051
94 G.G. Pereira, P.W. Cleary / Chemical Engineering Science 94 (2013) 93–107

neglected while compartmental modelling considers much larger

particle systems but focuses on various regions (or ‘‘compart-
ments’’) of the flow. With the advent of faster computers it is now
possible, with DEM simulations, to focus on comparatively large
particle systems (in excess of million particles) and so this
method is now becoming the preferred method for numerical
studies.
DEM studies of the mixing flow in the V-blender have
previously been considered in some detail (Kuo et al., 2002;
Moakher et al., 2000; Alexander et al., 2003; Lemieux et al.,
2007, 2008; Cleary and Sinnott, 2008). Kuo et al. (2002) validated
their DEM models against positron emission particle tracking
(PEPT) measurements and observed the flow exhibited two
phases: a separation phase as material splits into two streams
and flows along the arms of the V-blender and a re-combination
phase where the streams come together at the bottom of the
blender. They showed that mass transfer from one arm to the
other occurs during separation of the streams. Moakher et al.
(2000) compared the mixing in the cone blender to the V-blender
and also found the flow in the V-blender consisted of two distinct
phases. Alexander et al. (2003) reported several different segrega-
tion patterns evolved with mixtures of different particle sizes.
Lemieux et al. (2007) compared the V-blender’s mixing capability
versus the bin-blender and concluded that the V-blender was the
equal (in many cases) and superior in other cases. Subsequently,
the same group (Lemieux et al., 2008) considered large-scale
numerical investigation of mixing within the V-blender (for large
numbers of particles and long times). Cleary and Sinnott (2008)
considered the detailed mixing dynamics within the blender and
the critical issue of scale up from the laboratory scale (1.864 l
Fig. 1. Snap-shots of the V-blender geometry. (a) Front on in the y–z plane (y-axis blender) to the process scale (210 l blender). They found mixing
points upwards, opposite to gravity vector, while x-axis points into the page) and
efficiency decreased as the V-blender size increased.
(b) down an axis of one of the cylindrical arms.
After the mixing process is complete, the V-blender is emptied
through the bottom exit port by removing the cover. In this
homogenous was to extract a ‘‘thief sample’’ at a number of points emptying process, de-mixing can occur due to either the shear
in the V-blender (Berman and Planchard, 1995). In principal, by during flow within the V-blender and/or due to flow into the
sampling a number of different sites within the V-blender a good collection device. It is interesting to note that as far back as 1995,
overall quantification of the degree of homogeneity could be Berman and Planchard (1995) recognised that unloading of the V-
achieved. However, it was quickly realized by extracting the thief blender could lead to significant de-mixing. Although there are
sample, one physically affects the surrounding particle distribu- many studies of mixing within the V-blender, there are no studies,
tion and hence this invasive technique was open to debate to the best of our knowledge, concerning the de-mixing that may
regarding its efficacy. Subsequently, a number of non-invasive occur during emptying. This is a vitally important question, since
experimental techniques have been implemented to analyse the if there is significant de-mixing during emptying this will negate
flow pattern and degree of homogeneity within the V-blender. some (or in the worst case all) of the mixing that has already
These include near-infrared spectroscopy (Igne et al., 2011; Beach occurred.
et al., 2010; El-Hagrasy et al., 2001, 2006; Sekulic et al., 1996), Although there have been no studies of unloading from a
radioactive particle tracking (Doucet et al., 2008), gamma-ray V-blender, there have been studies of unloading from other
detection, via PEPT (Perrault et al., 2010) and X-ray computed industrial mixing and storage devices such as hoppers and silos.
tomographic (XRCT) scans (Yang and Fu, 2004). These studies Although the V-blender’s shape is quite unique compared to these
have been able to evaluate the degree of mixing for a variety of other industrial devices (two sloping cylindrical arms joined
conditions, including solid volume fraction fill of the V-blender, at a V), it will be useful to compare and contrast this study with
rotation rate or perturbations of the rotation (Brone et al., 1998) those works. Artega and Tuzun (1990) and Tuzun and Artega
and variation in particles sizes, shapes or densities. (1992) experimentally studied the discharge of binary and ternary
Computer modelling of granular media can be advantageous granular mixtures from funnel- and mass-flow hoppers. They
in understanding the flow within a processing device without considered variations in particle size ratios, fractions of particles
perturbing the particles’ positions or velocities, and hence opti- in the mixtures and found that size segregation was related to the
mising its efficiency. Hence, in conjunction with the large number granular microstructure (i.e., whether it was ‘‘coarse continuous’’
of experimental studies, there have been a variety of numerical or ‘‘fines continuous’’). Samadani et al. (1999) were able to image
studies of the V-blender. These include continuum modelling the flow within a silo during discharge of a binary mixture of
(Khakhar, 2011), Monte Carlo simulations, compartmental mod- spherical glass particles. The bed within the silo formed a V-shape
elling (Portillo et al., 2006) and discrete element method (DEM) where the slope angle is the angle of repose of the particles.
modelling (Moakher et al., 2000; Kuo et al., 2002; Lemieux et al., Segregation occurs due to filling of voids, in the surface layers
2007; Cleary and Sinnott, 2008). Continuum models necessarily of the flow, by the smaller particles. In the literature this mech-
simplify the flow modelling, as still there is no consensus on the anism is either referred to as ‘‘percolation’’, ‘‘sifting’’ or ‘‘dynamic
rheological models of granular flow. Monte Carlo methods only sieving’’. Samadani et al. (1999) were able to rule out segregation
consider the final state and as such the intermediate dynamics are due to velocity gradients within the bulk of the particle bed in the
 G.G. Pereira, P.W. Cleary / Chemical Engineering Science 94 (2013) 93–107 95

silo. There are a number of other experimental studies (e.g., Sleppy mechanism where the finer particles fall into randomly generated
and Puri, 1996; Alexander et al., 2000) of flow during discharge voids created in the layer underneath. Thomas (2000) and Felix and
with similar observations and conclusions to those above. On Thomas (2004) experimentally studied segregation of binary particle
the other hand, there have been limited numerical studies of mixtures when flowing down inclined chutes. Their observation and
discharge from hoppers or silos. Ketterhagen et al. (2007, 2008) conclusions for this case were similar to those for the formation of a
compared DEM simulation results to experiments for discharge of heap. More recently, Wiederseiner et al. (2011) have compared their
binary particle mixtures from wedge-shaped hoppers. They varied experimental results for segregation of binary granular mixtures
the angle of the wedge shaped walls, particle size ratio and the when flowing down a chute to a theory of Gray and Chugunov
fraction of smaller particles in the mixtures and found good (2006). They found that the primary segregation mechanism, which is
agreement between experiments and simulation. Subsequently, driven by percolation, competes with diffusive remixing leading to
the same group (Anand et al., 2008; Ketterhagen et al., 2009) have the boundary between large and small particles being smeared out.
used the DEM method to investigate discharge rates and flow The agreement between experiment and theory were best far away
modes from a number of different wedge-shaped hoppers. More from the inflow but there were discrepancies closer in the inflow
recently Engblom et al. (2012) developed models based on experi- region most probably due to non-steady bulk flow conditions.
mental correlations to predict the concentrations of fines close to There are two main conclusions from this survey of previous
the walls during discharge of powder mixtures. work on segregation during outflow. Firstly, segregation depends
Once the granular mixture has exited the V-blender, it is intimately on the geometry of the particular device and, secondly,
collected in a suitable container. The mixture usually forms a the initial conditions for the formation of heaps, or flow down
conical heaped surface. This can induce additional segregation to chutes, also contribute significantly to the ensuing segregation. In
the mixture. There have been studies of segregation during heap this paper, we therefore explore the degree of de-mixing that
formation in the past. The initial conditions for the formation of occurs in both parts of the discharge process. Initially, we consider
the heap will differ from case to case, but it is instructive to the case of de-mixing just due to discharge from the V-blender
consider these studies. Drahun and Bridgewater (1983) were and subsequently we consider the de-mixing due to two different
probably the first experimental study that looked at the forma- collection procedures. The first uses a cylindrical container placed
tion of two-dimensional heaps for binary particle mixtures in a directly under the exit port while the second uses a flat, sloping
controlled manner. They observed that segregation on the (upper) chute down into a cylindrical container. These two procedures are
free surface occurred by avalanching, inter-particle percolation the simplest approaches one can implement when studying the
and migration. The free fall height was also important as, above unloading of the V-blender. From these cases, we should be able to
a certain height, the small particles would bounce down the infer important characteristics of de-mixing during unloading.
free surface. They also found the initial distribution of particles Another important aspect of the de-mixing process is how the
markedly changed the final segregation pattern. Factors that degree of de-mixing varies with particle size. To understand this
influence the degree of segregation include particle size ratios, we focus only on binary sized particles and vary the size ratio
particle density ratios and fractions of one particle compared to between simulations. With the information obtained from these
the other. Thomas (2000) considered mixtures of two types of simulations we attempt to optimise the unloading set-up so as to
glass particles (of differing size but same density) in the forma- minimise de-mixing during discharge of the V-blender.
tion of a heap. Interestingly, Thomas (2000) found two kinds of
segregation, based on the particle size ratio. At low particle size
ratio (below 5 or so) the usual percolation-driven segregation 2. Model description
occurs. However, for large particle size ratios (greater than 5) a
‘‘reverse segregation’’ occurs, where the large beads segregate to We use the discrete element method (DEM) to model the flow
the interior of the heap. This is driven by the high mass difference of particles out of the V-blender and into the collection and
between particles, since now the heavier particles sink into the transfer devices. DEM models particulate systems whose motions
bed. This reverse segregation is particularly apparent when there are dominated by collisions. It follows the motion of every
is a low fraction of large particles. Later Felix and Thomas (2004) particle or object in the flow and models each collision between
extended these experiments to consider binary particle mixtures the particles and between the particles and their environment
with variations in density and size. They were able to observe (such as the blender). This study is performed using the DEM code
both segregation due to size (i.e., dynamical sieving) and segrega- described in Cleary (2004, 2009).
tion due to density (which they called a ‘‘push-away process’’, but The particles are allowed to overlap and the amount of overlap
is also referred to in the literature as a ‘‘buoyancy effect’’ Pereira Dx, and normal vn and tangential vt relative velocities determine
et al., 2011a). Smith et al. (2001) used granular dynamic simula- the collisional forces via a contact force law. There are a number
tion methods to study the formation of two-dimensional heaps. of possible contact force models available in the literature to
They were able to differentiate their results based on the impact approximate collision dynamics to various extents. We use a
of particles onto the heap (which in turn is governed by the linear spring-dashpot model.
free fall height). At low impact, segregation was dominated by The normal force
dynamic sieving of the finer particles while at high impact vertical F n ¼ kn Dx þC n vn , ð1Þ
pluming of particles occurred, i.e. periodic avalanches which
would cause the heap to partially collapse and all surface particles consists of a linear spring, of magnitude kn, to provide the repul-
to flow to the base. Rahman et al. (2011) have compared sive force and a dashpot to dissipate a proportion of the relative
experiments and DEM simulations to a theoretical model for kinetic energy. The normal damping coefficient Cn is calculated to
segregation in heaps of binary granular particle mixtures. give the required coefficient of restitution e (defined as the ratio
Another common method for emptying granular mixing and of the post-collisional to pre-collisional normal component of the
storage devices is via a chute, i.e. and inclined surface, connecting relative velocity), and is given by
the outlet of the V-blender to the entrance of the collector. Savage pffiffiffiffiffiffiffiffiffi
2 ln e mkn
and Lun (1988) developed a theoretical model for segregation of C n ¼  pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi , ð2Þ
binary granular particle mixtures when flowing along a rough- p2 þ ln2 e
ened inclined chute. The model was based on a dynamic percolation where m is the reduced particle mass.
96 G.G. Pereira, P.W. Cleary / Chemical Engineering Science 94 (2013) 93–107

The tangential force is given by 2.2. Simulation measures

n X o
F t ¼ min mF n , kt vt dt þC t vt , ð3Þ We use a variety of measures to quantify the degree of
segregation during the outflow of granular particles from the
where m is the coefficient of friction, kt is the tangential spring V-blender. Specifically:
stiffness, vt is the relative tangential velocity, dt is the incremental
time-step and Ct is a tangential damping coefficient. Ct is also
given by Eq. (2) with the only change being that kn is replaced by
 Qualitative pictures: The DEM technique constantly updates all
the positions and velocities of the particles. By colouring
kt. The tangential spring stiffness, kt, is set to be 0.5kn. The vector
particles according to their radius we can obtain a qualitative
force Ft and velocity vt are defined in the plane tangent to the
picture of the amount of de-mixing during the flow out of the
surface at the contact point. The sum term represents an incre-
V-blender.
mental spring that stores energy from the relative tangential
motion and models the elastic tangential deformation of the
 Sample planes: We can place data collection planes at various
positions to monitor the types, velocities and numbers of
contacting surfaces, while the dashpot dissipates energy from
particles flowing through these locations. The sample planes
the tangential motion and models the tangential plastic deforma-
are in the x–z plane and placed 1 mm and 6.1 cm below the
tion. The total tangential force Ft is limited by the Coulomb
V-blender exit.
frictional limit mFn,, at which point the surface contact shears
and the particles begin to slide over each other.
 Centroids: We measure the centre of mass (or centroid) of each
type of particle throughout each simulation. Each component
(x, y, z) of the centre-of-mass can be independently evaluated
2.1. Simulation parameters and the degree of separation of these components, for each
particle type, yields information on the amount of segregation.
We are interested here in millimetre scale particles, which are  Homogeneity mixing measure: This quantitative measure of
representative of many foods, grains, household products and segregation involves looking at the distribution of local
some plastic feed stocks. It is well known that particles segregate averages of the particle radius. This is done using a small
based on differences in size, density or shape (Rosato et al., 1988; characteristic volume at an array of locations throughout
Ottino and Khakhar, 2000). Here we will concentrate on particles the domain. If the binary mixture is perfectly mixed then the
which differ in size. Furthermore, although in general there will distribution is a uni-modal Gaussian peaked at the mass
be a continuous distribution of particle sizes, we will explore a weighted average of the particles. If the particles are comple-
simple system considering only a binary mixture of particle sizes tely de-mixed, then the distribution is bi-modal with peaks
(with equal volumes of each type of particle). The particles are centred on the radius of each particle type. The segregation
spherical in shape, have a mean size of 2.75 mm and the ratio of state that is computed falls between these two limits, where a
sizes of the two constituent varies from 1.75 up to 4.0 (see Table 1 value of 0% indicates a perfectly mixed mixture and 100%
for number of particles and size ratios used). The particles have a indicates a perfectly segregated mixture. The theory behind
small polydispersity of 0.1 mm about their mean values. The this measure can be found in Cleary (1998) and Cleary and
particle density is 1500 kg/m3 and the blender is initially filled to Sinnott (2008). It differs from the earlier measures in that it
50% of its volume. We simulate a laboratory scale V-blender does not depend on any spatial direction just measuring the
(Cleary and Sinnott, 2008) which has an arm length of 0.23 m and local homogeneity of the particle mixture (e.g. centroids are
an internal volume of 1.864 l with a vertex angle of 751. This given with respect to a particular direction x, y or z).
means the mass of particles in the mixer is close to 0.78 kg
(0.39 kg of each particle type) for each simulation. The coeffi-
cients of restitution are 0.7 for particle–particle collisions and 3. Case 1: flow out of the V-blender
0.85 for particle–wall (steel) collisions. The friction coefficient
was 0.5 for particle–particle contacts and 0.4 for particle–wall We first consider the flow out of the V-blender without any
contacts. The spring stiffness used was 1000 N/m. This corre- collection apparatus. Recently Fan and Hill (2011) and May et al.
sponds to DEM time-steps in the range of 4  10  6 –10  5 s and (2010) have shown that roughened bounding walls lead to
minimum binary collision duration of roughly 15 time steps. In increased shear of the granular mixture. This leads not only to
each of the simulation cases discussed in the following sections increased segregation but also some interesting reversal effects,
we initially have to fill the V-blender. This is done by rapid filling where the small and large particles can swap position in the flow.
(from a virtual inflow) and takes an initial 0.8 s. All our quanti- For these reasons we make the walls of the V-blender smooth
tative measures begin from the start of the filling period (which (as in experiments), to reduce shear effects close to the walls.
we consider to be the zero time) and we can check, with our As emptying progresses, the number of particles present in the
mixing measures, that the mixture is comparatively homogenous simulation domain decreases until the end when there are no
at the beginning of the outflow from the V-blender. It is not particles remaining. This configuration allows us to quantify
possible to create a perfectly mixed binary material initially filling segregation purely arising from the flow within the V-blender
the blender. and when discharging through the exit port. This is the absolutely
smallest amount of de-mixing that will be possible to achieve. For
Table 1 this case we have considered four different size ratios.
Particle diameters, number of particles, outflow time and outflow rate for the Fig. 2 displays the particle distributions for a size ratio of 1.75.
various particle size ratios simulated. Note the particles we use are distributed In this case the particles sizes are 2 mm and 3.5 mm. In all our
uniformly about the mean value with a spread of 0.1 mm. The total mass of
figures for the particle configurations, red coloured spheres
particles in each simulation is 0.78 kg.
represent the larger particles and blue coloured spheres represent
Size ratio 1.75 2 3 4 the smaller particles. According to our homogeneity measure, at
Particle diameters (mm) 2, 3.5 1.83, 3.66 1.37, 4.12 1.11, 4.44 the beginning of the outflow the mixture is perfectly mixed. After
# Particles (  10,000) 8 9.3 21 43 the first 0.7 s (Fig. 2a) of discharge the mixture appears to have
Outflow time (s) 2.8 2.6 2.2 2.0
Flow rate (kg/s) 0.32 0.33 0.38 0.45
not de-mixed to any significant extent, except that there appears
to be a small excess of larger (red) particles on the top (free)
 G.G. Pereira, P.W. Cleary / Chemical Engineering Science 94 (2013) 93–107 97

Fig. 3. (a) Variation of y-component of the centroid position (m) of large (red) and
small (blue) particles for the simulation shown in Fig. 2, and (b) mass flow rate
through the upper sample plane (1 mm below V-blender exit) as a function of
time. (For interpretation of the references to colour in this figure caption, the
reader is referred to the web version of this article.)

particles discharge at a faster rate than the larger particles for

much of the emptying process. Towards the end when the supply
of remaining finer particles is dwindling, the mass flow rate of
coarser particles then dominates. This is in keeping with Fig. 2,
which shows larger particles had collected at the top of the
particle bed during emptying. The sample planes data for the
second sample plane, which is 6 cm below the first sample plane,
shows similar characteristics. The average outflow rate for this
ratio is 0.32 kg/s and is the smallest of all the size ratios (see
Table 1). These results demonstrate that there is a small amount
of de-mixing with the smaller particles percolating vertically
downwards through the bed and exiting faster than the lagging
larger particles.
Next we consider outflow for the largest of the particle size
ratios. Fig. 4 shows snapshots of the particle distribution at two
Fig. 2. Discharge from a V-blender for a size ratio of 1.75, after (a) 0.7 s and
(b) 2.5 s of outflow. Smaller particles are coloured blue while larger particles are times during the outflow. After the first 0.5 s of outflow (Fig. 4a),
coloured red. (For interpretation of the references to colour in this figure caption, we observe that a layer of larger particles has accumulated at
the reader is referred to the web version of this article.) the top surface. During the remainder of the outflow this process
continues and by approximately 1.5 s of outflow (Fig. 4b) the
surface. After a further 1.8 s (and close to the end of the outflow, larger particles form a distinct layer at the top and in the centre of
Fig. 2b) the remaining mixture appears to be only marginally de- the outflow stream.
mixed, with larger particles predominantly in the centre of the The segregation process occurring here is related to particle
outflow stream. size and commonly occurs in granular media with differing
Fig. 3a shows the y-component of the centroid location for the particle sizes. For example see Pereira et al. (2011b) where
two particle sizes. It decreases continuously with time as the segregation due to percolation was observed in the shearing
particles discharge and the remaining particles are progressively avalanching layer of a binary granular mixture placed in a slowly
lower in the domain. There is a small but observable difference rotating drum. This percolation mechanism occurs generally
in centroid position of the two particle types, with the smaller when small particles fall through the voids created by the larger
particles (blue curve) being slightly below the larger particles (red particles (Rosato et al., 1988; Ottino and Khakhar (2000)). As a
curve), except at the very end. This indicates that the finer result, a layering of particles occurs with smaller particles found,
particles are on average lower in the system and have exited in larger concentration, lower in the bed and larger particles
the blender earlier. This means that there is a small amount of de- found higher in the bed. For percolation to occur, sufficient free
mixing of the different particle sizes during outflow. The mass volume must be available so that the particle can re-distribute.
flow rate data for the upper sample plane (1 mm under the V- Near the free surface of the bed this occurs readily but it also
blender exit) is shown in Fig. 3b. It indicates that the smaller occurs, albeit to a much smaller degree, near the V-blender exit.
98 G.G. Pereira, P.W. Cleary / Chemical Engineering Science 94 (2013) 93–107

Fig. 5. Mass flow rate as a function of time for the simulation shown in Fig. 4
through the (a) upper sample plane (1 mm below exit), and (b) lower sample
plane (6 cm below upper plane).

have pointed out that above a particle size ratio of 5, the bigger
particles are much more massive than the smaller particles and
hence will sink into the particle bed. This process will begin to
override the percolation mechanism. In this case, big particles will
tend to embed deep in the particle bed. This is most pronounced
for low fractions of the large particle. So we may expect the
segregation to plateau out (or possibly diminish) for larger size
ratios (i.e. greater than 5).
The mass flow rates for this case are shown in Fig. 5. The
smaller particles (blue curve) again pass through the sample
planes faster in the early stages of the outflow. The larger particle
flow rate (red curve) is again only highest once the fine particles
are substantially removed from the blender and the large particle
rich region at the top of the bed reaches the discharge port and
flows out. This agrees with the percolation mechanism for de-
mixing, since the smaller particles are found lower in the particle
bed and hence closer to the exit. The flow rate difference between
Fig. 4. Discharge from a V-blender for a size ratio of 4, after (a) 0.5 s and (b) 1.5 s
the two particle sizes has increased from (on average) 0.32 kg/s,
of outflow. Smaller particles are coloured blue while larger particles are coloured
red. (For interpretation of the references to colour in this figure caption, the reader for particle size ratio of 1.75 to 0.45 kg/s for the simulation for a
is referred to the web version of this article.) particle size ratio of 4. The discharge time is also shorter at 2.0 s
for the size ratio of 4 than it was at 2.7 s for the size ratio of 1.75.
Table 1 results indicate a gradual decrease in the discharge time
As the emptying proceeds the top surface becomes inclined (and correspondingly an increase in flow rate) as the particle size
towards the discharge port. As particles roll down this sloped ratio increases. This is likely to be due to the ability of particles to
surface, the smaller (blue) particles can easily percolate down rearrange within the entrance to the outflow stream and to flow
through the gaps (interstitial voids) between the larger particles. out through the exit. For very small particles this is much easier
As a result, larger particles segregate more rapidly to the top of than for large particles. In general, the discharge process for large
the mixture while smaller particles become more concentrated size ratio leads to greater de-mixing.
underneath. This leads to a high concentration of large particles The homogeneity measure quantifies the degree of de-mixing
on the free surface directly above the exit hole. Hence the larger for all the particles in the simulation domain. The homogeneity
particles are funnelled through the middle of the outflow stream, measure for the four particles size ratios is given in Table 2. This
as observed in Fig. 4b. This mechanism becomes stronger and data is taken near the end of the simulation (in the last 0.1 s of
faster as the particle size-ratio increases, which explains what we outflow), when there are not many particles left (about 1000), so
have observed in these two cases. Note, that the largest particle there are large statistical fluctuations, but to be consistent with
size ratio we use is 4. Thomas (2000) and Felix and Thomas (2004) other cases, we have decided this is the best time to compare this
 G.G. Pereira, P.W. Cleary / Chemical Engineering Science 94 (2013) 93–107 99

measure. There is an increase in the segregation percentage with In this case, where only the flow within the blender is consi-
increasing size ratio (from 1.75 to 4), although the decrease from dered, de-mixing is pre-dominantly due to percolation of small
3 to 4 is probably not indicative of a systematic trend but more particles through the shear flow that occurs along the sloping
due to the small number of particles sampled. However, as upper surfaces of the bed. This effect is small for small size ratios
will become apparent later when we consider the segregation but significantly increases (non-linearly) with increasing particle
measure in the presence of a collection device, these segregation size ratio.
percentages are quite low.

4. Case 2: flow into a cylindrical vessel

Table 2 We next consider flow from the V-blender into a cylindrical

Homogeneity segregation measure, as a percentage where 0% corresponds to well container directly under the exit. The radius of cylindrical con-
mixed and 100% corresponds to completely segregated, for Cases 1, 2 and 3. tainer is 15 cm and its bottom is 12 cm below the exit of the
V-blender. Four particle size ratios from 1.75 to 4 were again
Particle size No collector Cylindrical collector Chute into cylinder
ratio (Case 1) (Case 2) (Case 3) considered. Fig. 6 shows the particle distributions at different
times for the size ratio of 2. During the first 0.9 s of outflow
1.75 0.5 14 17 (Fig. 6b) a small heap of particles forms below the V-blender exit.
2 1 22 50 While the heap is small it does not show a significant amount of
3 45 76 99
4 26 86 97
de-mixing. The initial formation of the heap occurs with a large
amount of energy being dissipated (due to particles rebounding of

Fig. 6. Flow out of a V-blender into a cylindrical container for a particle size ratio of 2 at three different stages during the emptying process, (a) initially and then after
outflow of (b) 0.9 s, (c) 1.8 s, and (d) 3.5 s. Smaller particles are coloured blue while larger particles are coloured red. (For interpretation of the references to colour in this
figure caption, the reader is referred to the web version of this article.)
100 G.G. Pereira, P.W. Cleary / Chemical Engineering Science 94 (2013) 93–107

the flat base of the cylinder) which randomly re-arranges the The maximum amount of separation between the two curves is
particle distribution. Hence, the particles in the heap remain about 11 mm or about 4 average particle diameters. After this
relatively well mixed in Fig. 6b. time the two curves converge and the final separation between
Once the particles in the base of the heap have covered the the two curves is 3 mm or about 1 average particle diameter. The
cylinder base area the heap has been established (Fig. 6c). reason that the two curves converge after the maximum separa-
Particles no longer collide and rebound from the cylinder base, tion is achieved is that the smaller particles flow out of the V-
but rather fall onto the heap, where they collide with other blender (on average) before the larger particles. Thus earlier in
particles, dissipate their incoming kinetic energy and push other the outflow, they are found at a lower y-coordinate. Once the
particles down the free surface slope. The heap builds until the outflow has completed both smaller and larger particles are in the
particles in the apex of the heap can no longer be supported by same cylindrical collector and hence the scale of the separation is
those underneath. At this point the apex collapses and small smaller.
particles in the avalanching flow then percolate down through the Fig. 7b shows the homogeneity measure as a function of time.
avalanching layer, concentrating larger particles (red) along or It correlates quite well with the centroid measure with an
near the upper surface. As discharge progresses and the heap increase in degree of segregation (de-mixing) once the outflow
grows in size, the avalanching region that produces the segrega- begins. The amount of segregation becomes constant once the
tion increases in length leading to increasing amounts of segrega- flow out of the V-blender has ceased. The amount of segregation
tion. So the de-mixing in the collection container increases once all flow ceased is 22% which represents a significant amount
strongly once the heap has been established in the collector. of de-mixing that clearly degrades the mixture quality reversing
When the blender is empty (Fig. 6d) the final heap shows that a much of the hard work done by the blender.
reasonable amount of de-mixing has occurred (compared to Fig. 8 shows the particle distribution at three different times
having no collection device), with significantly more large parti- during the outflow for a size ratio of 4. The initial distribution of
cles on the upper surface. particles is partially de-mixed (from our homogeneity mixing
Fig. 7a shows the relative y-component of the centroid for both measure in Fig. 9 it is around 15% at the beginning of outflow).
particle sizes over time. The overall y-component of centre-of- This comes about from the fact that the two particle size classes
mass of all particles was subtracted to give a relative displace- have a large difference and hence almost any flow process
ment. The relative centroids both start at zero and have a involving these granular particles will result in some de-mixing.
zero difference since the material is initially well mixed. Once Due to the large difference in the size of the two particle classes,
the outflow begins, the difference between the relative centroids percolation of the small particles through the surface shear layer
increases. This is due to the segregation produced by the percola- results in a layer of large particles forming on the top surface
tion mechanism occurring in the flow within the V-blender. Once (Fig. 8b) as it did for the previous case. Particles which flow into
the particles have formed a heap in the cylindrical container, the cylindrical container initially rebound of the bottom surface,
further de-mixing occurs also due to percolation of particles collide with other incoming particles and then come to rest on the
flowing down the heap surface. The general shape of this centroid base of the cylinder. In this way, a layer of particles form on the
curve suggests at around 2.7 s the maximum separation occurs. base of the cylindrical container. Once a heap has begun to form
in the cylinder, particles roll along the sloped surface of the
heap. Periodically, the apex of the conical heap collapses and an
avalanche of particles cascades down the heap surface. Percola-
tion of the smaller particles results in a layer of larger particles on
the upper surface of the heap (Fig. 8c). The larger particles are
more concentrated in the middle of the outflow stream. This is
because the surface layer of larger particles flows radially inwards
to the vertical centre-line of the blender and then flows vertically
downwards towards the discharge port. By the time the flow has
ended (Fig. 8d) there is a reasonable amount of de-mixing of the
heap visible, i.e., in comparison to Fig. 6d, there are fewer blue
particles visible. The larger particles form a thick surface layer and
smaller particles concentrated in the middle of the bed.
The centroid data for this size ratio is not shown here as it is
qualitatively similar to Fig. 7b. In comparison to the case with size
ratio 2 (Fig. 7b), the maximum difference between the two curves
has increased (roughly) by a factor of two. The difference rises to
nearly 19 mm (about 7 average particle diameters) compared
to 11 mm (4 average particles) for size ratio 2. The homogeneity
mixing measure (Fig. 9) for this case also shows an increase
(compared to simulation for a particle size ratio of 2). However, in
this case the segregation measure rises more rapidly up to 86% by
the completion of the outflow. This is close to the maximum value
for this measure and indicates that the mixture is now close to
completely de-mixed. Clearly, the outflow, for a particle size ratio
of 4, has almost completely undone the work done by the
V-blender and hence is extremely harmful insofar as obtaining a
Fig. 7. (a) Relative y-centroid position of large and small particles for the outflow homogenous mixture.
into a cylinder for a particle size ratio of 2. Larger particles are shown by the red These results show that when particles flowing out from a
curve and blue corresponds to the smaller particles. (b) Homogeneity measure for
size ratio of 2. Initially the segregation is small and increases to a final value of
V-blender are collected in a container directly below the exit,
around 22%. (For interpretation of the references to colour in this figure caption, the amount of de-mixing is greatly increased, primarily due to
the reader is referred to the web version of this article.) percolation of smaller particles within surface shearing layers of
 G.G. Pereira, P.W. Cleary / Chemical Engineering Science 94 (2013) 93–107 101

Fig. 8. Flow out of a V-blender into a cylindrical container for a particle size ratio of 4 at three different times during outflow, (a) initially and then after outflow of
(b) 0.95 s, (c) 1.95 s, and (d) 2.95 s. Smaller particles are coloured light blue while larger particles are coloured red. (For interpretation of the references to colour in this
figure caption, the reader is referred to the web version of this article.)

Table 3
Normalised maximum difference (with respect to average particle diameter)
between the y-component of the large and small particle centroids for Cases 2
and 3.

Particle size Cylindrical collector Chute into cylinder

ratio normalised max diff normalised max diff

1.75 3.6 5.5

2 4.0 6.2
3 6.2 9.5
4 6.9 10.2

doubling of the size ratio. Table 3 gives the maximum difference

Fig. 9. Homogeneity measure for size ratio of 4. At the beginning of outflow the in the y-component of the particle centroids for each size ratio
homogeneity measure is around 15% while at the end of outflow it is approxi- simulated. For convenience, we also plot these results in Fig. 10
mately 86%. where it can be seen that the increase is non-linear (with particle
size ratio) with a limiting behaviour at higher particle size ratios.
the heap that forms in the collection device. Table 2 shows the The homogeneity measure also follows a non-linear relationship
homogeneity measure increases from 14% (for particle size ratio with particle size ratio (asymptoting to the maximum value at
of 1.75) up to 86% segregated (for particle size ratio 4), which large particle size ratio). The limiting behaviour is much stronger
leads to a six-fold increase in segregation with an approximate for the homogeneity measure, since there is a strict upper bound
102 G.G. Pereira, P.W. Cleary / Chemical Engineering Science 94 (2013) 93–107

(2007) found increased shear length (e.g. along the chute) led to
increased segregation. However, this arrangement also signifi-
cantly reduces the vertical free fall heights and therefore the
kinetic energy available for driving percolation in avalanching
surface layers of the heap. One might a priori imagine that this
would reduce the strong segregation occurring in the heap
component of the flow. So the question is which of these is the
more important effect?
Fig. 11 displays snapshots of the particle distribution for this
case with a particle size ratio of 2. The flow in and from the
V-blender is similar to the previous cases. Once out of the V-blender,
the particles now flow down along the flat chute surface. They
accelerate and there is shear induced by the frictional resistance
of the chute surface. The larger (red) particles, in comparison to
the smaller (blue) particles, strongly segregate to the top of the
flowing layer and come to rest closer to the walls of the cylinder
(Fig. 11b). Due to the horizontal component of the velocity
generated by the chute, the heap that forms in the receiving
container is asymmetrical in shape (Fig. 11c). The avalanching
flow to the left is much stronger leading to strong percolation of
finer particles down into the heap. By the completion of the
discharge, the heap that forms is significantly de-mixed (even for
this modest size ratio), with many more of the larger, red particles
on the upper surface and on the left side of the heap (Fig. 11d).
This is in contrast to the previous case (flow into a cylinder
without a chute) where the heap forms quite a symmetrical
particle distribution (about the centre of the cylinder). The final
homogeneity measure is 50% for this case, compared to 22% for
Case 2 (see Table 2), at the same particle size ratio.
Fig. 10. (a) Plot of normalised maximum difference of y-centroid (between small The shape of the y-centroid curves is qualitatively similar to
and large particles) for Case 2 (black circles) and Case 3 (red squares) as a function
of particle size ratio. (b) Homogeneity measure for Cases 2 and 3. Both measures
Case 2, with larger particles segregating to higher positions in the
are non-linear and reach limiting values. (For interpretation of the references particle beds while the smaller particles tend to flow earlier and
to colour in this figure caption, the reader is referred to the web version of this are located lower in the system. From Table 3 we can see the
article.) maximum relative vertical separation of the centroids is about
17 mm (about 6 average particle diameters). This is roughly
50% higher than for Case 2 (and for that matter also Case 1)
(of 100%) for this measure. Both these measures indicate, for this indicating that the presence of the chute has increased the
case, that the extent of de-mixing increases significantly for larger amount of segregation. The final separation of the y-centroids is
size differences between the particles which make up the gran- around 5 mm (2.5 particle diameters). Table 4 gives the separa-
ular mixture. The qualitative description of segregation during tion between the z-component of the centroid (between the two
heap formation given here is generally similar to those given particle types) for all particle ratios (recall, the chute is aligned
previously (Thomas, 2000; Felix & Thomas, 2004). However, the with the z-axis). For Case 2 these differences are negligible for all
various measures, we give, yield a quantitative evaluation of this particles size ratios meaning the amount of horizontal segrega-
segregation. Savage and Lun (1988) have predicted, from a tion in the heap is negligible. But for this collection geometry, the
continuum theoretical model, that segregation increases non- transverse segregation in the chute direction is large—around
linearly with particle size ratio, which is in agreement with 19 mm or 6–7 particle diameters. Note, the x-centroid shows no
Fig. 10. One could argue that the levelling of our segregation significant variation for particle types since their flow to either
measures, at high particle size ratio, is due to the reversal side of the chute are symmetric. From all these measures (homo-
mechanism (Thomas, 2000) but since we have equal amounts, geneity measure, y-centroid and z-centroid) we conclude that the
by volume, of large and small particles, we believe this is not the presence of the chute induces a large increase in segregation
case here. Rather, we believe that at particle size ratio of 4 we (compared to Case 2) for a particle size ratio of 2.
have almost maximised the amount of segregation. The temporal evolution of the homogeneity measure (see
Fig. 12) has a different functionality to that obtained for the
previous cases (such as Fig. 7c). For the flow directly into the
5. Case 3: flow via a chute cylinder the increase in segregation was quite gradual, with
respect to time, whereas for the chute case the segregation
We now consider the case where the particles flow down an measure increases dramatically once particles begin flowing
intermediate chute from the V-blender into a cylindrical con- along the chute (at 1 s). The peak degree of segregation is around
tainer. The end of the chute is slightly offset from the centre of the 70% occurring at 2.5 s, which corresponds to when most of the
container which has the same dimensions and properties as in the particles have completed their travel down the chute. This
previous case. We offset the end of the chute, so that most of indicates that segregation due to flow along the chute surface is
the particles remain within the cylinder once they rebound of significant and rapid. Later the degree of segregation decreases
the cylinder base. The chute is angled at 451 to the plane of the back to around 50% when flow is complete suggesting that the
cylinder base and aligned in the y–z-plane (see Fig. 11). The avalanche flow in the heap is actually helping to partially re-mix
particles travel along the chute for about 15 cm before dropping a the particles. This final degree of segregation compares to about
short distance vertically into the container. Previously Shi et al. 20% for Case 2 where particles just fall directly into the container.
 G.G. Pereira, P.W. Cleary / Chemical Engineering Science 94 (2013) 93–107 103

Fig. 11. Flow out of the V-blender via a chute into a cylindrical container for a particle size ratio of 2 at different times, (a) initially and then (b) after 1.0 s of outflow,
(c) after 2.0 s and (d) after 3.5 s of outflow. Smaller particles are coloured blue while larger particles are coloured red. (For interpretation of the references to colour in this
figure caption, the reader is referred to the web version of this article.)

Table 4
Maximum difference (mm) between the z-component of the particle centroids for
Cases 2 and 3.

Particle size ratio Cylindrical collector Chute into cylinder

max diff (mm) max diff (mm)

1.75 0.01 19
2 0.4 21
3 1 27
4 0.6 25

The inclusion of any chute in the system appears to be extremely

deleterious to preserving the mixed state of the particles.
Fig. 12. Homogeneity measure for size ratio of 2. The final value of this measure is
Fig. 13 shows the particle distribution during discharge for size around 50%.
ratio of 4. The behaviour is qualitatively similar to that described
for the lower size ratio shown in Fig. 11. However, now the
vertical stratification along the chute surface is much stronger for particle in the surface layer of the heap and more on the left side
this size ratio than for a size ratio of 2. Also here there seems to of the picture (Fig. 13d). The smaller particles (blue) are not only
much more rebound of particles off the cylinder base, when underneath the larger particles, but also have re-bounded further
flowing into the cylinder (Fig. 13b and c). from the far wall of the cylindrical container and now appear in
For the flow shown in Fig. 13, the larger particles come to rest higher concentration in the right half of the collection cylinder
closer to the container wall (Fig. 13c), than for the larger particles (underneath the chute). This asymmetry in the distribution
in Fig. 11. When the particles come to rest the heap again has an of particles in the heap is shown up quantitatively in the relative
asymmetrical shape with significant concentrations of larger (red) z-component of the centroids in Fig. 14 (where the overall
104 G.G. Pereira, P.W. Cleary / Chemical Engineering Science 94 (2013) 93–107

Fig. 13. Flow out of the V-blender via a chute into a cylindrical container for a particle size ratio of 4 at different times during the outflow. (a) Initially and then (b) after
0.5 s of outflow, (c) after 1.35 s and (d) after 2.75 s of outflow. Smaller particles are coloured blue while larger particles are coloured red. (For interpretation of the
references to colour in this figure caption, the reader is referred to the web version of this article.)

z-centroid has been subtracted from that of each component so increasing particle size ratio. Hence, the segregation that occurs
that the centroid positions vary about the value zero). must be due to percolation.
Table 4 gives the maximum difference in the z-component of The homogeneity measure for this case (Fig. 14b) starts at
the centroid, between the two particle types, for both Cases 2 and about 15% segregated at the beginning of the outflow. Once
3. Flow without a chute shows negligible differences but with a particles roll along the chute (after 1 s) the segregation increases
chute the difference is large varying between 6 particle diameters rapidly up to around 90% by 2.6 s. It then increases slowly to an
(size ratio of 1.75) up to 9 particle diameters (for a size ratio of 3). asymptotic value of about 97%. The slowdown in segregation in
One other interesting aspect of the z-centroid curves (Fig. 14a) is the later stages can be attributed to the fact that the outflow is
that the average z-positions of the large and small particles highly energetic, which can re-mix some of the particles (as
reverse during the simulation. Initially more small particles are observed in the later stages of the flows shown in Figs. 11 and 13).
located near the far wall of the cylinder, but they then are shifted, Table 2 gives the final value of the homogeneity measure for all
due to more large particles entering the cylinder and displacing the particle size ratios simulated. In general, for this Case (as was
the smaller particles which were adjacent to the far cylinder found for Case 2), we see a rapid increase of this measure with
wall. Hence, as seen in Fig. 13 the final average z-position of the increasing particle size ratio. Comparing this measure, for each
large particles is close to the far wall and vice versa for the particle size ratio, we see Case 3 always yields higher values
smaller particles. The chute has produced horizontal segregation compared to Case 2 (and thus more de-mixing).
in the bed. In summary, the flow of particles into a cylinder via a chute
The y-centroid data has a similar form to those already leads to significantly higher amounts of de-mixing than for flow
discussed. The peak centroid separation increases to 28 mm or into a collection container that is without a chute. This might have
around 10 average particle diameters which is significantly higher been predicted from work of Shi et al. (2007) and Wiederseiner
than for the size ratio of 2. Table 3 shows the maximum difference et al. (2011) who suggested increased shear length leads to
in the y-component of the centroid for all the particle size ratios increased segregation. However, by reducing the free fall height
simulated. For this Case, as with Case 2, we see a gradual increase onto the heap, we hoped to reduce segregation on the heap
(again non-linearly with a limiting behaviour at high particle surface. This turned out to be not the case. The chute produces
size ratio, see Fig. 10) in the amount of vertical segregation with additional overall and vertical segregation and also induces
 G.G. Pereira, P.W. Cleary / Chemical Engineering Science 94 (2013) 93–107 105

that the mean particle diameter is 2.75 mm, while the cylinder
radius is 15 cm). We also considered a circular motion of the
cylindrical container for this case with a frequency of 1.25 Hz (for
both x and z components) and radius of motion of 5 mm. In the
following we give results for the following scenarios (which gave
the lowest segregation values): an oscillation in the x-direction of
frequency 1.25 Hz and amplitude of 5 mm, an oscillation in the
x-direction of frequency 2.5 Hz and amplitude of 5 mm and
circular motion with frequency of 1.25 Hz and radius of 5 mm.
The outflows for these simulations follow a similar pattern to
that already described in Case 2. The main visible difference now
is that when the particles hit the base they tend to form a flatter
layer rather than a heap. Eventually, however there is not enough
surface area on the base to accommodate all the particles and a
heap begins to form. However, this heap is more flattened than for
Case 2, as we would expect, due to the vibration of the cylinder.
We are now most interested in the final segregation values of the
homogeneity measure since it is this value that gives us the best
quantitative estimate of the degree of de-mixing. Table 5 gives the
final values for all the cases simulated and also, as a comparison,
the case where the cylinder has no motion. Overall the optimal
conditions from the set of scenarios we investigated result from
circular motion (see Table 5). It has the lowest segregation value
for three of the four particle size ratios simulated and only slightly
larger for the simulation where the particle size ratio is 2. One
should note that for this particle size ratio (of 2) the segregation
value for all set-ups is low (around 20%). The oscillatory motion
with a frequency of 1.25 Hz resulted in the lowest value for the
Fig. 14. (a) Relative z-centroid position (m) of the small and larger particles for
case where the particle size ratio was 2.
the outflow into a container for a particle size ratio of 4. Larger particles are shown
by the red curve and blue corresponds to the small particles. (b) Homogeneity Next we consider a cylindrical collector with a smaller dia-
measure for size ratio of 4. The final value of this measure is 97%. (For inter- meter but also taller (so that all the granular mixture from the
pretation of the references to colour in this figure caption, the reader is referred to V-blender can be contained in the collector). The diameter we use
the web version of this article.)
is 16 cm (roughly half of before) and a height of 10 cm. The
resulting homogeneity segregation measure for the four particle
a horizontal segregation in the particle beds. As with the other size ratios is shown in Table 5 (last column). These values are in
cases the extent of de-mixing increases strongly with the size fact lower than all previous cases (except for a particle size ratio
ratio of particles. of 1.75) simulated. The normalised maximum difference in the
y-component of the centroid values is 3.6, 3.8, 5.7 and 5.8 (for size
ratios 1.75, 2, 3 and 4). These values are slightly smaller than for
6. Minimisation of de-mixing the 30 cm diameter cylindrical collector (see Table 3) and also the
centroid values for circular motion of the cylinder (corresponding
It is clear from the above findings that any transfer schemes centroid values are 3.5, 4.0, 5.8 and 6.4). Overall, this collection
which introduce a shearing flow into the system will increase procedure appears to give the least amount of segregation,
de-mixing. Moreover, the presence of the sloped heap surface although at the highest particle size ratio (of 4) the segregation
produces more de-mixing. One possible strategy, to inhibit is still significant.
de-mixing, is to flatten out the heap, via a gentle shaking of the In summary, we can reduce de-mixing of the binary particle
cylindrical container. Actually it is relative motion between the mixture by diminishing the area for percolation on the heap’s
V-blender and collector that is important so that the conical heap surface. This can be achieved by introducing a small amount of
is not formed. In practice one could shake the outlet of the motion into the collection device, so as to de-stabilise the heap
V-blender (or a flexible discharge tube attached to the outlet). and hence inhibit segregation on the heap surface. However, the
Another possibility is to reduce the size of the heap and so reduce reduction is not large and certainly when the particle size ratio is
the surface area over which percolation can occur. This can
be done by having a smaller but taller cylindrical collector. This Table 5
second possibility may not always be desirable (insofar as Final homogeneity measure values (%) for flow into a cylinder placed directly
processing may be concerned) but we still consider it here. We below the exit port of the V-blender. Five scenarios are considered: (i) no motion
of the cylinder, (ii) and (iii) oscillatory motion of the cylinder along the x-axis, (iv)
evaluate both these possibilities now. circular motion of the cylinder and (v) a 16 cm diameter cylinder at rest with
Firstly we consider shaking the cylinder. To evaluate this respect to the V-blender.
possibility we introduce a horizontal oscillation to the cylinder
(along the x-axis) or a circular motion of the cylinder (in the x–z Size No Sinusoidal Sinusoidal Circular motion 16 cm
ratio motion motion motion (freq ¼1.25 Hz, diameter
plane). We consider amplitudes and frequencies of oscillation
(freq ¼ 1.25 Hz, (freq ¼2.5 Hz, radius¼ 5 mm) cylindrical
that are quite small so as to introduce just enough motion to amp ¼ 5 mm) amp ¼5 mm) collector
destabilise the heap but not enough motion so as to introduce
additional shearing over the cylinder base. Since the time for 1.75 14 13 12 10 19
outflow is around 2–3 s, we need a frequency of at least 1 Hz. We 2 22 21 23 27 24
3 76 70 60 55 33
therefore tried frequencies of 1.25 Hz, 2.5 Hz and 5 Hz. The 4 86 93 92 79 65
amplitudes considered were 5 mm, 10 mm and 50 mm (recalling
106 G.G. Pereira, P.W. Cleary / Chemical Engineering Science 94 (2013) 93–107

3 or 4, there is still significant de-mixing in the final particle bed. that it generates. The amplitude and frequency of the shaking
The motion of the collection device has to be tuned so as to not need to be sufficient to flatten the pile but not produce extra
introduce (too much) additional shear of the forming particle bed. shearing flow which would then result in segregation of the
We can also reduce the surface area of the heap, by confining mixture.
the heap to a cylinder of small diameter. In fact, this gives the
smallest amount of segregation (according to our measures).

References
7. Conclusions
Alexander, A., Muzzio, F.J., Shinbrot, T., 2003. Segregation patterns in V-blenders.
In this study we have considered the de-mixing due to outflow Chem. Eng. Sci, 58, 487–496.
from a V-blender, using the DEM method. The first point that Alexander, A., Roddy, M., Brone, D., Michaels, J., Muzzio, F.J., 2000. A method to
quantitatively describe powder segregation during discharge from vessels.
these simulations have clearly shown is that a large amount of
Pharmaceut. Technol. Yearbook, 6–21.
segregation may occur during the unloading process, so this Anand, A., Curtis, J.S., Wassgren, C.R., Hancock, B.C., Ketterhagen, W.R., 2008.
process must be carefully understood and monitored. We have Predicting discharge dynamics from a rectangular hopper using the discrete
considered three unloading configurations and for each four element method (DEM). Chem. Eng. Sci. 63, 5821–5830.
Artega, P., Tuzun, U., 1990. Flow of binary mixtures of equal-density granules in
particle size ratios, so as to clearly understand which parts of hoppers—size segregation, flowing density and discharge rates. Chem. Eng.
the outflow contribute to de-mixing and to what extent. The Sci. 45, 205–223.
V-blender, itself, is made up of sloping walls which lead to the Beach, L., Ropero, J., Mujumdar, A., Alcala, M., Romanach, R.J., Dave, R.N., 2010.
Near-infrared spectroscopy for in-line characterization of powder voiding. Part
exit port (see Fig. 1). The outflow that results from this geometry II: quantification of enhanced flow properties of surface modified active
is predominantly shear flow. The top (free) surface of the particle pharmaceutical ingredients. J. Pharmaceut. Innov. 5, 1–13.
bed becomes inwardly sloped towards a point vertically above the Berman, J., Planchard, J.A., 1995. Blend uniformity and unit dose sampling. Drug
Dev. Ind. Pharm. 21, 1257–1283.
exit port. Shearing flow then leads to smaller particles percolating Brone, D., Alexander, A., Muzzio, F.J., 1998. Quantitative characterization of mixing
through the voids created by the larger particles. As a result, of dry powders in V-blenders. Am. Inst. Chem. Eng. 44, 272–278.
smaller particles concentrate lower in the particle bed and hence Cleary, P.W., 1998. Predicting charge motion, power draw, segregation, wear and
particle breakage in ball mills using discrete element methods. Miner. Eng. 11,
tend to exit the V-blender earlier than the larger particles. This
1061–1080.
percolation driven de-mixing has been seen many times before in Cleary, P.W., 2004. Large scale industrial DEM modelling. Eng. Comput. 21,
a variety of granular devices. While, the geometry of the V- 169–204.
Cleary, P.W., 2009. Industrial particle flow modelling using discrete element
blender (i.e., the sloping walls) might be advantageous during the
method. Eng. Comput. 26, 698–743.
mixing process, it clearly increases de-mixing during discharge of Cleary, P.W., Sinnott, M.D., 2008. Assessing mixing characteristics of particle-
the V-blender. The amount of de-mixing is not so important for mixing and granulation devices. Particuology 6, 419–444.
the lower particle size ratios (1.75 and 2) but becomes significant Doucet, J., Bertrand., F., Chaouki, J., 2008. An extended radioactive particle tracking
method for systems with irregular moving boundaries. Powder Technol. 181,
for the larger size ratios (3 and 4). 195–204.
Once the granular mixture has exited the V-blender, it needs Drahun, J.A., Bridgewater, J., 1983. The mechanisms of free surface segregation.
to be collected in a suitable container. We investigated two Powder Technol. 36, 39–53.
El-Hagrasy, A.S., Morris, H.R., D’Amico, F., Lodder, R.A., Drennen, J.K., 2001. Near-
collection procedures, firstly by simply dropping into a cylindrical infrared spectroscopy and imaging for the monitoring of powder blend
container and secondly by being transferred via a sloping chute homogeneity. J. Pharmaceut. Sci. 90, 1298–1307.
into a cylindrical container. In each case, both flows introduce El-Hagrasy, A.S., D’Amico, F., Drennen, J.K., 2006. A process analytical technology
approach to near-infrared process control of pharmaceutical powder blending.
additional de-mixing of the granular mixture. When falling Part 1: D-optimal design for characterization of powder mixing and prelimin-
directly into a cylindrical container, de-mixing occurs due to ary spectral data evaluation. J. Pharmaceut. Sci. 95, 392–406.
percolation along the sloped surfaces of the collecting heap. Engblom, N., Saxen, H., Zevenhoven, R., Nylander, H., Enstad, G.G., 2012. Segrega-
tion of powder mixtures at filling and complete discharge of silos. Powder
The amount of segregation due to the collection device increases Technol. 215-216, 104–116.
markedly compared to the segregation solely within the V-blender. Ennis, B.J., Green, J., Davis, R., 1994. The legacy of neglect in the United States.
In fact, for the larger particle size ratios of 3 and 4 the final Chem. Eng. Prog. 90, 32–43.
Fan, Y., Hill, K.M., 2011. Phase transitions in shear induced segregation of granular
segregation is close to maximum and indicates the unloading
materials. Phys. Rev. Lett. 106 218301(1-4).
process has almost completely de-mixed the particles. When an Felix, G., Thomas, N., 2004. Evidence of two effects in the size segregation process
additional chute is introduced into the transfer process, quite a in dry granular media. Phys. Rev. E 70 051307(1-16).
violent flow may result, with particles being sheared along the flat Gray, J., Chugunov, V., 2006. Particle-size segregation and diffusive remixing in
shallow granular avalanches. J. Fluid Mech. 569, 365–398.
surface of the chute as well as in the gradually forming heap. Igne, B., Zacour, B.M., Shi, Z., Talwar, S., Anderson, C.A., Drennen, J.K., 2011. Online
The chute flow increases the segregation values for all particle monitoring of pharmaceutical materials using multiple NIR sensors. Part 1:
size ratios and is especially harmful for the particle size ratio of 2 blend homogeneity. J. Pharmaceut. Innov. 6, 47–59.
Ketterhagen, W.R., Curtis, J.S., Wassgren, C.R., Kong, A., Narayan, P.J., Hancock, B.C.,
(in comparison to no chute being present). Furthermore, the chute 2007. Granular segregation in discharging cylindrical hoppers: a discrete
produces horizontal segregation within the particle bed, which element and experimental study. Chem. Eng. Sci. 62, 6423–6439.
was not present in the other cases. Ketterhagen, W.R., Curtis, J.S., Wassgren, C.R., Hancock, B.C., 2008. Modelling
granular segregation in flow from quasi-three-dimensional, wedge shaped
All our segregation measures demonstrate that there is an hoppers. Powder Technol. 179, 126–143.
increase in the amount of segregation with particle size ratio. This Ketterhagen, W.R., Curtis, J.S., Wassgren, C.R., Hancock, B.C., 2009. Predicting the
agrees with our hypothesis that the mechanism behind the de- flow mode from hoppers using the discrete element method. Powder Technol.
195, 1–10.
mixing is percolation. It is clear that any transfer schemes which Khakhar, D.V., 2011. Rheology and mixing of granular materials. Macromol. Mater.
introduce a shearing flow into the system will increase de-mixing. Eng. 296, 278–289.
In the light of this observation, it appears the best transfer 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 behaviour in a
strategy is to allow the mixture to directly flow into the collection
V-mixer. Chem. Eng. Sci. 57, 3621–3638.
device. In addition, it is clear the presence of the sloped heap Lemieux, M., Bertrand, F., Chaouki, J., Gosselin, P., 2007. Comparative study of the
surface also produces more de-mixing. By reducing the size of this mixing of free-flowing particles in a V-blender and bin-blender. Chem. Eng.
surface we can reduce final segregation. As a proof of concept, a Sci. 62, 1783–1802.
Lemieux, M., Leonard, G., Doucet, J., Leclaire, L.A., Viens, F., Chaouki, J., Bertrand, F.,
gentle shaking of the cylindrical container was introduced to de- 2008. Large-scale numerical investigation of solids mixing in a V-blender
stabilise and flatten this heap in an effort to reduce the de-mixing using the discrete element method. Powder Technol. 181, 205–216.
 G.G. Pereira, P.W. Cleary / Chemical Engineering Science 94 (2013) 93–107 107

May, L.B.H., Golick, L.A., Phillips, K.C., Shearer, M., Daniels, K.E., 2010. Shear-driven Samadani, A., Pradhan A., Kudrolli, A., 1999. Size segregation of granular matter in
size segregation of granular materials: modelling and experiment, Phys. Rev. E, silo discharges. Phys. Rev. E 60, 7203–7209.
81, 051301(1-8). Savage, S.B., Lun, C.K.K., 1988. Particle size segregation in inclined chute flow of
Meier, S.W., Lueptow, R.M., Ottino, J.M., 2007. A dynamical systems approach to dry cohesionless granular solids. J. Fluid Mech. 189, 311–335.
mixing and segregation of granular mixtures in tumblers. Adv. Phys. 56, Shi, D., Abatan, A.A., Vargas, W.L., McCarthy, J.L., 2007. Eliminating segregation in
757–827. free-surface flows of particles. Phys. Rev. Lett. 99 148001(1-4).
Moakher, M., Shinbrot, T., Muzzio, F.J., 2000. Experimentally validated computa- Sekulic, S.S., Ward, H.W., Brannegan, D.R., Stanley, E.D., Evans, C.L., Sciavolino, S.T.,
tions of flow, mixing and segregation of non-cohesive grains in 3D tumbling Hailey, P.A., Aldridge, P.K., 1996. On-line monitoring of powder blend homo-
blenders. Powder Technol. 109, 58–71. geneity by near-infrared spectroscopy. Anal. Chem. 68, 509–513.
Ottino, J.M., Khakhar, D.V., 2000. Mixing and segregation of granular materials. Simpson, M.F., Goff, K.M., Johnson, S.G., Bateman, K.J., Battisti, T.J., Toews, K.L.,
Annu. Rev. Fluid Mech. 32, 55–91. Frank, S.M., Moschetti, T.L., O’Holleran, T.P., Sinkler, W., 2001. A description of
Perrault, M., Bertrand, F., Chaouki, J., 2010. An investigation of magnesium stearate the ceramic waste form production process from the demonstration phase of
mixing in a V-blender through gamma-ray detection. Powder Technol. 200, the electrometallurgical treatment of EBR-11 spent fuel. Nucl. Technol. 134,
263–277.
234–245.
Sleppy, J.A., Puri, V.M., 1996. Size segregation of granulated sugar during flow.
Pereira, G.G., Sinnott, M.D., Cleary, P.W., Liffman, K., Metcalfe, G., Sutalo, I.D.,
Trans. ASAE 39, 1433–1439.
2011a. Insights from simulations into density segregation of granular mixtures
Smith, L., Baxter, J., Tuzun, U., Heyes, D.M., 2001. Granular dynamics simulations of
in rotating cylinders. Granular Matter 13, 53–74.
heap formation: effects of feed rate on segregation patterns in binary granular
Pereira, G.G., Pucilowski, S., Liffman, K., Cleary, P.W., 2011b. Streak patterns
heaps. J. Eng. Mech. 127, 1000–1006.
in binary granular material in a rotating drum. Appl. Math. Model. 35,
Terayama, A., Kyigoku, H., Sakamura, M., Komatsu, S., 2005. Fabrication of TiNi
1638–1641. powder by mechanical alloying and its shape memory characteristics of
Portillo, P.M., Muzzio, F.J., Ierapetritou, M.G., 2006. Characterizing powder mixing sintered alloy. J. Jpn. Inst. Metals 69, 523–529.
processes utilizing compartment models. Int. J. Pharm. 320, 14–22. Thomas, N., 2000. Reverse and intermediate segregation of large beads in a dry
Poux, M., Fayolle, P., Bertrand., J., Bridoux, D., Bousquet, J., 1991. Powder mixing: granular media. Phys. Rev. E 62, 961–974.
some practical rules applied to agitated systems. Powder Technol. 68, Tuzun, U., Artega, P., 1992. A microstructural model of flowing ternary mixtures of
213–245. equal density granules in hoppers. Chem. Eng. Sci. 47, 1619–1634.
Rahman, M., Shinohara, K., Zhu, H.P., Yu, A.B., Zulli, P., 2011. Size segregation Wiederseiner, S., Andreini, N., Epely-Chauvin, G., Moser, G., Monnereau, M., Gray,
mechanism of binary particle mixture in forming a conical pile. Chem. Eng. Sci. J.M.N.T., Ancey, C., 2011. Experimental investigation into segregating granular
66, 6089–6098. flows down chutes. Phys. Fluids 23 013301(1-10).
Rosato, A., Strandbourg, K.J., Prinz, F., Swendsen, R.H., 1988. Why the Brazil nuts Yang, C.Y., Fu, X.Y., 2004. Development and validation of material labelling method
are on top—size segregation of particulate matter by shaking. Phys. Rev. Lett. for powder process characterization using X-ray computed tomography.
58, 1038–1040. Powder Technol. 146, 10–19.