Chemical Engineering and Processing 57–58 (2012) 42–50

Contents lists available at SciVerse ScienceDirect

Chemical Engineering and Processing:

Process Intensification
journal homepage: www.elsevier.com/locate/cep

Solving granular segregation problems using a biaxial rotary mixer

Janet Cho a , Yunfeng Zhu a , Karol Lewkowicz a , SungHee Lee a , Theodore Bergman b ,
Bodhisattwa Chaudhuri a,c,∗
a
Department of Pharmaceutical Sciences, University of Connecticut, Storrs, CT 06269, United States
b
Department of Mechanical Engineering, University of Connecticut, Storrs, CT 06269, United States
c
Institute of Material Sciences, University of Connecticut, Storrs, CT 06269, United States

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

Article history: Granular mixing is a critical but poorly understood aspect in the manufacture of many industrial products,
Received 27 November 2011 for example, pharmaceuticals, food, cosmetics, ceramics, fertilizer and polymers. The mixing and segrega-
Received in revised form 18 March 2012 tion phenomenon occur in most systems of granular solids and have a significant influence on the quality
Accepted 23 April 2012
and outcome of the final product. The usual approach to mix the powders is by using a tumbling blender
Available online 28 April 2012
which rotates around one axis, where, the radial convection is reported to be faster than axial dispersion
transport, hindering the mixing performance. A double cone mixer is fabricated which rotates around
Keywords:
two axes, causing axial mixing competitive to its radial counterpart. Samples are collected intrusively
Granular mixing
Powder flow
using the discrete pocket samplers to quantify the characteristics of mixing for millimeter sized glass
Rotary mixer beads. Digital video recording and MATLAB based image analysis techniques are used for non-intrusive
Sampling characterization of mixing in micron sized art sands. A parametric study of the effect of particle size,
Image analysis vessel speeds on the granular mixing is accomplished. Incorporation of dual axis rotation enhances axial
mixing by 70–90% in comparison to single axis rotation. Particles of smaller sizes (art sand) tend to mix
quicker than the bigger particles (glass beads) due to mild cohesive effects.
© 2012 Elsevier B.V. All rights reserved.

1. Introduction in the geometry of the vessel. One of the most common batch type
rotary mixers in industrial use is double cone mixer, where gran-
The significance of granular mixing to the world economy is ular materials are caused to flow by a combination of the action
paramount as the products from industries spanning from pharma- of gravity and the rotating motion of the mixer. The double cone
ceuticals, minerals, agriculture and food to chemical and ceramics, mixer consists of a vertical cylindrical shell with the conical top
all increasingly depend on reliable granular flow and uniform and bottom. Despite the common usage of the double cone mixer
granular mixing. As for instance, the annual cost of inefficient in batch mixing operations, relatively less research on it has been
industrial mixing in the US has been estimated to be US$10 billion published. In several preliminary studies on standard double cone
[1]. The efficiency of all pharmaceutical solid dosage formulations mixer, experiment based parametric investigations on mixer per-
(tablets/capsules) depends on their blend homogeneity; therefore, formance were done for a given set of materials and operating
inconsistency in the mixture can be detrimental for the patients. conditions [6–9]. Studies on the mixing of solids in a double cone
The routinely used powder mixers can be categorized into two blender [10,11] investigate axial and radial mixing separately using
particular types: rotary blenders [2,3] and convective blenders sampling ports located parallel to the rotation axis for axial mix-
[4,5]. Despite the rotary blenders rely upon the action of gravity ing and perpendicular to the rotation axis for radial mixing. The
to cause the powder to cascade and mix within a rotating ves- effect of particle size and flowability on sampling and quantita-
sel, the convective blenders employs an paddle, impeller, blade, tive characterizations of mixing performance as a function of the
or screw which stirs the powder inside a static vessel. Convective most basic parameters, such as vessel speed or filling level, are also
blenders show variations in both impeller and vessel geometries, scarce in the literature [12–14]. Experimentally validated model-
while rotary blenders which rotate around one axis, differ mainly ing of granular flow was first attempted by [1] to analyze robust
segregation tendencies persistent in double cone blenders. In most
of the rotary mixers, the radial convection is faster than the axial
∗ Corresponding author at: Department of Pharmaceutical Sciences, University of
dispersion transport [1,11,15,16]. This slow dispersive process hin-
Connecticut, Storrs, CT 06269, United States. Tel.: +1 860 486 4861;
ders mixing performance. In order to address this mixing problem,
fax: +1 860 486 2072. previous researcher [9,13,17] performed experiments comparing a
E-mail address: [email protected] (B. Chaudhuri). standard double cone to one with a baffle attached to the shell of the

0255-2701/$ – see front matter © 2012 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.cep.2012.04.002
 J. Cho et al. / Chemical Engineering and Processing 57–58 (2012) 42–50 43

vessel. The baffle was same as the length of the vessel, and was at an 2.1. Intrusive sampling of glass beads
angle with the rotation axis, apparently in an effort to cause axial
motion within the system. The presence of the baffle, however, did After the required number of rotations, the experiment is tem-
not always affect the rate of mixing [9]. The reason behind the enor- porarily stopped to open up the top half of the cone, place the
mous loss of resources (energy, materials and manpower) [18] is sampling template (shown in Fig. 1c) on the top surface of the gran-
the lack of knowledge and unpredictability of mixing in rotational ular bed, and then to draw samples using discrete pocket sampler
mixing devices. In a situation where a final product is particularly (shown in Fig. 1d, manufacturer: GlobalPharma, NJ) from the same
expensive to produce or sensitive to small fluctuations in the final area of the bed through the holes of the sampling template. The
form, efficient and reproducible mixing carries key importance in holes in the template (ash shown in Fig. 1c) are strategically made
the process. With that goal in mind, in a previously published paper to draw samples along the axis of the double cone mixer to quantify
[19], we introduced a novel concept of multi-dimensional rotation mixing in the axial direction. No attempt of sampling in the radial
to the traditional double cone blender and perform experiments direction is made for quantifying radial mixing as it is known to be
with spherical glass beads and particle dynamics based numerical very rapid as compared to its axial counterpart [10,11]. The mix-
modeling to study granular flow and mixing. While traditional mix- ing states are quantified in each batch run, by hand counting the
ers (drum, double cone, bin, V, bohle-bin) rotate about a single axis, number of beads of different colors present in each sample to be a
the multi-dimensional rotary mixer rotates simultaneously about function of space and time.
two perpendicular axes using two separate step motors, mounted
on a frame built to support rotation about one axis, while being 2.2. Non-intrusive sampling of art sand
rotated itself about another. An experimental study with more real-
istic micron sized powder systems and subsequent characterization The main obstacles in the characterization of mixing states for
using image analysis technique based non-intrusive is presented art sands or powder blends for uniform distribution of blend com-
in this paper. The organization of the article is as follows. The brief ponents depend in the collection of representative samples and
descriptions of the experimental method and experimental results their accurate analysis. Conventionally, samples are obtained from
are presented in Sections 2 and 3, respectively. Section 4 highlights the blender by inserting the thief sampler or discrete pocket sam-
the major conclusions of the research. pler at defined time intervals. Despite the simplicity in operating
these sampler, the detrimental effects of intrusive probes such as
tendency to disturb the powder bed, withdrawal of non-uniform
2. Experimental method, sampling and materials samples in terms of composition and quantity, has long been
recognized [20–22]. Several noninvasive methods have been inves-
A vessel of 3 liters volume is fabricated out of two Pyrex made tigated by other researchers, such as use of Raman spectroscopy
long stem funnels where the stems are plugged to the outer alu- [23], radioactive tracers [20], image analysis based on color dif-
minum made frame (as shown in Fig. 1) to create the appropriate ference [24], thermal effusivity [25], light induced fluorescence
geometry. The photograph of the actual apparatus is illustrated in [20,26], fluorescence microscopy [27], and near infrared (NIR) spec-
Fig. 1. The vessel is 5.6 in. in diameter and 12 in. high, with a 3 in. troscopy [28–30]. Among these methods, image analysis method is
straight side. The glass blower is used to attach external hooks to being chosen due to its simplicity and low cost. Image analysis tech-
be used in fastening the two funnels together by rubber bands. The nique allows a full experiment to continue without interruption or
vessel frame, which surrounds the vessel, consists of an aluminum disturbance to the granular bed. It also offers the ability to per-
ring (marked C in Fig. 1a) of 14-in. outer diameter and 13-in. inner form multicomponent analysis in a fast, non-destructive manner,
diameter and 2.25 in. of thickness. Two step motors (Lin Engineer- requiring little or no sample preparation. The continuous runs of
ing, CA) are used to separately rotate the frame and the vessel tied the mixing experiments are digitally recorded using a camcorder at
up to the frame using a flanged ball bearing respectively. The bot- the rate of 40 frames per second for non-intrusive sampling purpose
tom portion of the frame is machined so that the one of the stepper and also to qualitatively validate of numerical simulation. As shown
motor (marked A in Fig. 1a) can be mounted. The vessel attaches in the Fig. 2a, a frame corresponding to particular time, is acquired
to the ball bearing and the motor shaft so that it can be rotated from the digitally recorded movie of our mixing experiment with
about the vertical axis. The base frame is a sheet of 5/8 in. plywood. art sands. The movie of the side view of the mixing experiment
Steel rods are bolted and supported to build the structure to sup- is captured keeping the digital camcorder outside the apparatus.
port the vessel frame. The other stepper motor (marked B in Fig. 1a) This image is loaded in to Adobe Photoshop and as all the particles
is mounted on the outside of one of the steel supports on the base are concentrated in the triangular part of the double cone vessel,
frame to rotate the vessel frame via axels about the horizontal axis. that part of the image is zoomed and saved into another image file
The motors are connected to the respective drivers, which are con- (as shown in Fig. 2b). The triangular domain is discretized into 8
trolled by a PC. A slip ring is placed outside the frame and across by 8 matrix rectangular space and a previously developed MATLAB
the higher powered motor to connect the driver and the low pow- based custom designed pixel-counting program [31,32] is used to
ered motor, thus removing the possibility of tangled wiring caused determine the number of red and blue pixels in each of the squares.
by simultaneous rotations by two different motors. To perform the Thereafter, intensity of segregation (I) as a function of time is cal-
rotation experiments of radial mixing equal volume (and mass) of culated from the concentration of each of species in each of the
glass beads of two different colors, but of same size, are loaded up discretized squares at the corresponding time instants. The non-
to the fill level of 40%, in a side-by-side (as shown in Fig. 1b) initial intrusive characterization of mixing is also accomplished by same
configuration, then rotated at definite speeds using computer con- digital image analysis approach (as depicted in Fig. 3a–c) to the top
trolled motors. Experiments with systems homogenous in size but surface of the granular bed by analyzing the top view image. The
comprising different colors are being performed using two different horizontal and vertical axes of the double cone are named X and Y
sizes of spherical glass beads (1 mm and 3 mm; Ceroglass Technolo- axes for convenience, throughout the rest of the paper. In our study
gies, TN) and art sand (250 ␮m, Catskill Mountain Industries, NY) of we have checked the effect of initial loading (front–back, side–side),
two different colors. A typical run includes 450 ml of each species. fill level, rotational speeds at orthogonal directions (about X and Y)
A computer program interfacing the drivers of the motors is devel- on the mixing of initially segregated granular systems. The evolu-
oped to control the rotation rates, direction of rotation and duration tion of mixing is quantified by estimating variation of concentration
of rotation of the two different step motors. and intensity of segregation (I) as a function of time for all batch
44 J. Cho et al. / Chemical Engineering and Processing 57–58 (2012) 42–50

Fig. 1. (a) The double cone is set inside the aluminum made frame (C). The frame is rotated around horizontal axis by motor (B) and the vessel is rotated around its vertical
axis by another motor (A). (b) Top view of the double cone showing the initial configuration of side-by-side loading of same material but of two different colors to check axial
loading. A divider is used to load the beads. (c) The template with four holes to ensure sampling from same space of the granular bed. (d) A discrete pocket sampler used to
collect a mixture of red and blue glass beads. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Fig. 2. (a) A snapshot (side view) of the mixer from the digitally recorded movie. (b) A zoomed view of the mixer containing the mixture of powders. (c) The area of the
image is discretized into a 8 by 8 square grid for image analysis purpose.

runs. The fill ratio for all the experiments with glass beads and art within each half of the blender [13], creating convective flow on
sand was maintained to be 40%. each side of the plane of symmetry, but not across the center-
line of the vessel in uniaxial rotation (around the horizontal line
3. Results and discussion of symmetry of the vessel). A simultaneous secondary rotation
around the vertical axis is incorporated along with the usual rotary
3.1. Effect of vessel speed motion around the horizontal axis. The effect of vessel speeds on
the mixing patterns is studied experimentally. The effect is stud-
An important feature of a double cone blender is that the cross ied separately with glass beads of two sizes: 1 mm diameter and
sectional area available for powder flow varies with time as the 3 mm diameter and with art sand of size of 250 ␮. Red and white
vessel rotates. The change in cross sectional area creates axial flow colored glass beads of 3 mm size are used for experiments using

Fig. 3. (a) A snapshot (top view) of the mixer. (b) A zoomed view of the mixer containing the mixture of powders. (c) The area of the image is discretized into a 8 by 8 square
grid for image analysis purpose.
 J. Cho et al. / Chemical Engineering and Processing 57–58 (2012) 42–50 45

the larger size in glass beads. In the trial with the smaller size in where  2 was the variances of the concentration of red particle
glass beads, red and blue colored glass beads are used. The red and calculated at any time instant and at time = 0 s for the numerator
blue colored art sand of 250 ␮ size are used for mixing experi- and denominator of I. Variance ( 2 ) is calculated using the following
ments. A fill level of 40% and side–side initial loading is used for formula:
all the experiments. This side–side initial loading is chosen due  2
to the well-known segregation problem for this particular loading (Ci − C̄)
2 = (2)
[11,16]. Twelve different experiments (X = 10, 20, and 30 rpm and N
Y = 0, 10, 20, and 30 rpm) for all combinations of orthogonal rota- where Ci is the concentration of red beads at point i (‘i’
tion speeds are being performed separately with smaller, bigger corresponds to the sample points, i = 1, 2, 3, 4), C̄ the average con-
beads and art sands. Digital video recording of the experiments is centration of red beads, N is the number of sample points (N = 4).
done for the qualitative analysis. In each of the experiments with all The intensity of segregation is estimated for sampling data col-
sizes of glass beads, to ascertain the temporal and spatial concen- lected at every time interval of stoppage during the experiments.
tration of each of the species, a discrete pocket sampler is used to Fig. 6a illustrates the temporal variations of intensity of segrega-
intrusively draw samples at definite time intervals by stopping the tion (measured of the concentration of red beads of big red/white
experiments. Non-intrusive characterization of mixing is done for bead system) corresponding to three rotary systems ((i) X rota-
art sand, using image analysis techniques to the time based snap- tional speed: 10 rpm; (ii) X rotational speed: 10 rpm, Y rotational
shots of the digital movies (and photos) recorded from different speed: 10 rpm; (iii) X rotational speed: 10 rpm, Y rotational speed:
view points (side and top views). 30 rpm). In all the cases, the granular bed becomes less segre-
Fig. 4a and b shows the time sequence of experimen- gated as time proceeds. Mixing rate is determined by the slope of
tal snapshots with smaller beads (blue and red) rotated at the intensity of segregation plots. The curve reaching the asymp-
X = 10 rpm; Y = 0 rpm and X = 10 rpm; Y = 30 rpm respectively. The totic state earliest compared to other curves is considered to be
time sequences of mixing states of bigger particles (white and red mixing in the quickest rate. Though the uniaxially rotating case
glass beads) under similar experimental conditions are depicted (X rotation = 10 rpm) decreases steadily but the dual axes rotat-
in Fig. 4c and d. Fig. 4a and c corresponds to single axis rotation ing cases (X rotation = 10 rpm, Y rotations = 10 rpm, 30 rpm) have
and in both cases very nominal mixing is observed, in the course a higher slope of decline. Moreover, the case with highest rota-
of time. However, with the addition of second rotation around tional speed around Y axis (X rotation = 10 rpm, Y rotation = 30 rpm)
the vertical (Y) axis, the mixing is quickly achieved as evident in reaches asymptotic value of intensity at around 50 s, whereas uni-
Fig. 4b and d for smaller and bigger beads respectively. The mixing axial rotating case is yet to reach the asymptote even at 300 s. The
states from experiments using different sizes of beads but under reduction in segregation (increase in mixing) occurs faster with
same rotational conditions look very similar to each other. Thus, increase in the Y rotational speed. The value of intensity of segre-
no prominent effect can be attributable to the size of the beads and gation (Ln(I) = −3) reached in 300 s by the case of X rotation = 10 rpm
further experimentation is done with the micron sized art sands. In and Y rotation = 0 rpm is achieved in less than 30 s by the case with
Fig. 5 the time series of snapshots gathered from the experiments X rotation = 10 rpm, Y rotation = 30 rpm. Incorporation of dual axis
with art sand are illustrated. The time series of experimental snap- rotation enhances the time of mixing by almost 90% in compari-
shots are presented for the cases where the blender is rotated at son to single axis rotation. Fig. 6b depicts the temporal variations
10 rpm around X axis and the Y axis rotation speed is set at: 0 rpm, of intensity of segregation corresponding to three rotary systems
10 rpm and 30 rpm (shown in Fig. 5a–c respectively). Similar to ((i) X rotational speed: 20 rpm; (ii) X rotational speed: 20 rpm, Y
the experiments with the glass beads, a sluggish axial mixing is rotational speed: 10 rpm; (iii) X rotational speed: 20 rpm, Y rota-
observed in Fig. 5a, corresponding to the single axis rotation. From tional speed: 30 rpm). Similar to the series of X rotation = 10 rpm,
the snapshots in Fig. 5b and c, it is evident that rotating the blender in the series of X rotation of 20 rpm, the reduction in segrega-
around two axes manifests quicker mixing conditions. The second tion occurs faster with increase in the Y rotational speed and the
rotation (about Y axis) promotes axial transport of the particles mixing is quickest for X rotation = 20 rpm and Y rotation = 30 rpm.
along X axis, causing particles to change sides from one longitudinal However, the mixing patterns for X rotation = 30 rpm series show
half of the blender to the other half. distinct difference than the counterparts of X = 10 and X = 20 series.
After qualitative observation of the importance of dual rotation In Fig. 6c, it is observed that with the introduction of Y axis rota-
to granular mixing, mixing pattern is quantified using the experi- tion, the mixing is achieved quicker, but mixing with Y rotational
mental sampling technique and image analysis for glass beads and speed = 10 rpm is quicker in comparison to counterpart with Y rota-
art sands respectively, as mentioned in Section 2. The double cone is tional speed = 30 rpm. Moreover, the case showing quickest mixing
rotated at a set speed (s) till the mixed state is achieved. In the intru- (X rotation = 30 rpm, Y rotation = 10 rpm) reaches asymptotic value
sive characterization of mixing of glass beads, the vessel is stopped of intensity at around 25 s, whereas uniaxial rotating case is yet
during the experiment and samples are collected at each of the to reach the asymptote even at 200 s. Thus the incorporation of
four sampling locations of the template. The sample size drawn by dual axis rotation enhances the time of mixing by almost 88% in
only one pocket of the discrete pocket sampler is 20 ml compris- comparison to single axis rotation.
ing almost 100 beads. For each sampling location, the number of The mixing pattern in the art sand is non-intrusively estimated
beads of each color was recorded. The beads are then placed back by analyzing several snapshot images (frames) of various time
into the original sampling location and the rotation of the vessel is instants, from the digitally recorded movies of the continuous runs
resumed. The concentration of red beads for four of the locations of the mixing experiments. As described in the Section 2, the image
(in the sampling template) of the granular bed is estimated as a is discretized into small square areas and MATLAB is used to esti-
function of time for mixing experiment where the vessel is rotated mate the number of red and blue pixels present in each of the
around horizontal axis (X) at 10 rpm with no Y axis rotation (0 rpm). squares. Colors can be always described as vectors belonging to a 3D
The evolution of mixing is also quantified using I, defined as Eq. (1). space, where each color can be defined by three co-ordinates. Col-
The intensity of segregation, I was calculated using the formula ors images are categorically digitized as 24 bit RGB files by MATLAB,
[4]: where 8 bits for each red, green and blue are used. The concentra-
tion of red particles in each of the square area is ascertained by
2
I= (1) calculating the ratio of red pixels to total number of pixel present
02 in the space. The average concentration of red particles is then
46 J. Cho et al. / Chemical Engineering and Processing 57–58 (2012) 42–50

Fig. 4. Snapshots from the experiment with glass beads of different sizes for single and dual axes rotations. (a) 1 mm glass beads: corresponds to X rotational speed = 10 rpm;
no rotation about the Y axis (i.e. 0 rpm); (b) 1 mm glass beads: corresponds to X rotational speed = 10 rpm; the Y rotational speed = 10 rpm; (c) 3 mm glass beads: corresponds to
X rotational speed = 10 rpm; no rotation about the Y axis (i.e. 0 rpm); and (d) 3 mm glass beads: corresponds to X rotational speed = 10 rpm; the Y axis rotational speed = 30 rpm.
Dual rotation enhances mixing. (For interpretation of the references to colour in this text, the reader is referred to the web version of this article.)

calculated using Eq. (3) and thereafter calculates the variance where Cpi is the concentration of red particles in the i-th square;
(shown in Eq. (4)) of the pertinent triangular space of the image Np is the total number of squares which fully or partially contains
of the mixer. the mixture of materials.
 2
(Cpi − C̄i )
p2 = (4)
N
 Fig. 7a illustrates the temporal variations of intensity of seg-
Cpi
C̄i = (3) regation (of red art sand particles) corresponding to three rotary
Np systems ((i) X rotational speed: 10 rpm; (ii) X rotational speed:

Fig. 5. Variation of intensity of segregation (for art sands) with time for the cases with (i) X rotational speed 10 rpm; (ii) X rotational speed: 10 rpm, Y rotational speed:
10 rpm; (iii) X rotational speed: 10 rpm, Y rotational speed: 30 rpm. For higher Y rotation speeds, the beads quickly become less segregated with time. Quickest mixing at Y
rotation speed = 30 rpm.
 J. Cho et al. / Chemical Engineering and Processing 57–58 (2012) 42–50 47

Fig. 6. Variation of intensity of segregation (with glass beads) with time for the cases with (a) X rotational speed 10 rpm, Y rotation = 0, 10, 30 rpm; (b) X rotational speed:
20 rpm; Y rotational speed: 0, 10, 30 rpm; (c) X rotational speed: 30 rpm; Y rotational speed: 0, 10, 30 rpm. For higher Y rotation speeds, the beads quickly become less
segregated with time.

10 rpm, Y rotational speed: 10 rpm; (iii) X rotational speed: 10 rpm, axis rotation enhances the time of mixing in art sands by almost
Y rotational speed: 30 rpm), all measured from the images gath- 75% in comparison to single axis rotation. The intensity of segre-
ered from the side view of the mixer. The same experiments with gation plot shown for the similar cases, made from the top view
art sands are repeated to gather the digital images of the evolution of the experiments, show essentially the same trend as observed
of mixing in the granular bed from the top view. Images pertinent to for side view results (Fig. 7a). The intensity of segregation curves in
specific time intervals are acquired by stopping the experiment. The Fig. 7b shows some fluctuations (unlike the corresponding ones in
images from the viewpoint at the top look similar to image shown in Fig. 7a) and the time of mixing vary from the side view results. The
Fig. 3a. The discretization technique described in Section 2 is used to intrusive techniques reveal the evolution of mixing on the surfaces
estimate the intensity of segregation of all the three cases of series of the granular bed and the mixing on the top surface (top view)
X rotation = 10 rpm (shown in Fig. 7b). In all the cases, the granular exhibits slower progress in mixing in contrast to the slant longitudi-
bed becomes less segregated as time proceeds. Though the uniax- nal surfaces (side view). The characterization of mixing using image
ially rotating case (X rotation = 10 rpm) decreases steadily but the analysis technique of images acquired from different viewpoints
dual axes rotating cases (X rotation = 10 rpm, Y rotations = 10 rpm, (side and top) reveals almost the same trend and thus results esti-
30 rpm) have a higher slope of decline. Moreover, the case with mated from the side view will be presented in the rest of the paper.
highest rotational speed around Y axis (X rotation = 10 rpm, Y rota- The image analysis work performed on the side views of the mixer is
tion = 30 rpm) reaches its lower values of intensity (mixed state) at also preferred as it can be done without stopping (and restarting)
around 22 s, whereas, uniaxially rotating case is yet to reach the the experiments at different time intervals necessary for procur-
asymptote even at 90 s in Fig. 7a. Thus the incorporation of dual ing the top view images. The reduction in segregation (increase in

Fig. 7. Variation of intensity of segregation (with art sand) with time for the cases with (a) side view: X rotational speed 10 rpm, Y rotation = 0, 10, 30 rpm; (b) top view: X
rotational speed: 10 rpm, Y rotational speed: 0, 10, 30 rpm.
48 J. Cho et al. / Chemical Engineering and Processing 57–58 (2012) 42–50

Fig. 8. The time series of snapshots from the experiments (with art sands) for single and dual axes rotations. (a) corresponds to X rotational speed = 30 rpm, no rotation about
the Y axis (i.e. 0 rpm); (b) corresponds to X rotational speed = 30 rpm, the Y rotational speed = 10 rpm; and (c) corresponds to X rotational speed = 30 rpm; the Y axis rotational
speed = 30 rpm.

mixing) occurs faster with increase in the Y rotational speed for art As a general observation, we see that when the blender is rotated
sand, similar to what is been observed for binary systems of glass only around the horizontal axis (X), i.e. for no vertical rotation, mix-
beads. ing is faster for higher speeds. Mixing is also faster for dual rotations
Thereafter, in Fig. 8 the time series of snapshots gathered from in the systems of glass beads or art sands, with any combination
the experiments with art sand are illustrated, where the blender of vertical and horizontal rotational speeds, if compared with the
is rotated at 30 rpm around X axis and the Y axis rotation speed same undergoing single axis rotation.
is set at: 0 rpm, 10 rpm and 30 rpm (shown in Fig. 8a–c respec-
tively). Similar to the experiments with glass beads, sluggish axial
mixing is observed in Fig. 8a, corresponding to the case with only
one axis rotation. From the snapshots in Fig. 8b and c, it is evi-
dent that rotating the blender around two axes manifests quicker
mixing conditions but mixing is quicker for Y rotation = 10 rpm
as compared to Y rotation = 30 rpm. This observation is similar to
what observed with glass beads. Irrespective of the materials used,
quicker mixing is always achieved when there is phase lag between
the rotations around two orthogonal axes (X and Y). The phase lag
between the rotational speeds is responsible to break the symmetry
of movement of the granular bed, leading to better mixing. Better
mixing is evident in the cases of X = 10 rpm, and Y = 30 rpm, as com-
pared to X = 10 rpm, and Y = 10 rpm in the series of X = 10 rpm; and
in the cases of X = 30 rpm, and Y = 10 rpm, as compared to X = 30 rpm,
and Y = 30 rpm in the series of X = 30 rpm. Moreover, the high rota-
tional speed (30 rpm) creates more centrifugal force to the particles
causing them to stick to the wall, manifesting lack of axial flow. This
is again reinforced with the intensity of segregation plots shown in
Fig. 9, estimated from the side view images of the experiments elu- Fig. 9. Variation of intensity of segregation (for art sands) with time for the cases
with (i) X rotational speed 30 rpm; (ii) X rotational speed: 30 rpm, Y rotational speed:
cidated in Fig. 8. The mixing is fastest in dual axis rotation with the
10 rpm; (iii) X rotational speed: 30 rpm, Y rotational speed: 30 rpm. For higher Y rota-
secondary rotational speed around Y = 10 rpm, while the rotational tion speeds, the beads quickly become less segregated with time. Quickest mixing
speed around X = 30 rpm. at Y rotation speed = 10 rpm.
 J. Cho et al. / Chemical Engineering and Processing 57–58 (2012) 42–50 49

Fig. 10. The effect of size of the particles on mixing. The time series of snapshots from the experiments with dual axes rotations (X rotational speed = 10 rpm; rotation about
the Y axis = 30 rpm). (a) corresponds to experiments with 3 mm sized glass beads, (b) corresponds to experiments with 1 mm sized glassbeads, (c) corresponds to experiments
with 250 ␮ art sands. Art sands mix faster than the glass beads.

3.2. Effect of size of the particle 4. Conclusions

The effect of size of particles on the granular mixing patterns A multi-dimensional rotary mixer is fabricated to successfully
is also studied in the multi axis mixer. The effect is studied sep- resolve the conventional segregation problem of axial mixing in
arately with glass beads of two sizes: 1 mm diameter and 3 mm rotary double cone blender. The effect of rotational vessel speed and
diameter and 250 ␮ sized art sands. As mentioned in the Section particle size on mixing performance is evaluated by experiments
2, red and blue art sand are used for mixing experiments. In the and numerical modeling. The time of axial mixing is reduced in dual
trial with the smaller sized glass beads, red and blue colored beads axis rotation as compared to single axis rotation. Incorporation of
are used. Red and white colored glass beads are used for experi- dual axis rotation enhances the time of mixing by 70–90% in com-
ments using the larger size. A fill level of 40% and side–side initial parison to single axis rotation. Higher rotational speed (30 rpm)
loading is used for all the experiments. Fig. 10a–c shows the time around both the axes results in sluggish mixing. Micron sized art
sequence of experimental snapshots with bigger beads (white and sand particles mix significantly quicker than the bigger glass beads
red), smaller beads (blue and red) and art sands (red and blue) of size 1–3 mm due to the mild cohesive behavior in art sands. Irre-
respectively, all rotated at X = 10 rpm; and Y = 30 rpm. The three spective of the materials used, quicker mixing is always achieved
different time points (0, 15, 30 s) show the progression of mixing when there is phase lag between the rotations around two orthogo-
and it appear qualitatively different where art sand has quickest nal axes. Mixing of very fine particles is a challenge due to cohesion
mixing rate and nearly complete mixing at t = 30 s. The end times and van der waals forces. Though the mixer was successful to effi-
of complete mixing for all three different materials are ascertained ciently mix glass beads and art sands, but a proof of concept of
from the intensity of segregation graphs and also qualitatively from mixing the real pharmaceutical drug powder and excipients is war-
experimental movies. Though there is not much difference in the ranted as future work to bolster the applicability of the blender in
time of mixing for glass beads of 1 mm and 3 mm sizes but signifi- reality. Intrusive analytical technique with discrete pocket sam-
cant difference is observed between the beads and art sand. Micron pler was used when mixing was quantified by drawing samples
sized art sand particles may have experienced mild cohesive effects from four fixed areas of bed through the holes of the card board
due to electrostatics and Van der Waals forces causing rapid mixing template. Sample size of 10 ml was drawn from regions of layer
compared to the free flowing (non cohesive) glass bead materials, thickness 10–15 particle size diameters (for the bed with 3 mm
a phenomena previously observed by Shinbrot [33] and Chaudhuri size). On the contrary, image analysis based non-intrusive tech-
[34]. nique was used for smaller particles (art sand) where the mixing
The rotating wall imparts shear or frictional force on the parti- was only characterized on the outer surface of the granular bed.
cles processed in the tumbling blender dictating the shear rate and
the linear response of the particles participating in the 2-D flow Acknowledgments
in the cascading layer situated at the top surface of the granular
bed. During the biaxial motion, the resultant axis of rotation of the We thank the University of Connecticut Research Foundation
vessel changes with time, also changing the direction of normal for supporting the work, and the undergraduate students Melissa
vector of the cascading layer surface. The particles at the surface Kuhn, Roshan Shah, Mi Hye Kim, Varunkumar Bhattaram, Michael
in contact with the moving walls thus acquired time-dependent Saito, Sweta Vachani, Rayhan Saikh, Jason Tomei, Seetha Manickam,
variation in linear velocity to produce a chaotic convective mix- and Eesha Desai for helping in the experiments. We are grateful to
ing in the cascading layer and eventually the whole granular Dr. Robin Bogner and Dr. Michael Pikal for their support in pursuing
bed. Mechanical Engineering Senior Design Project.
50 J. Cho et al. / Chemical Engineering and Processing 57–58 (2012) 42–50

References [19] S. Manickam, R. Shah, J. Tomei, T. Bergman, B. Chaudhuri, Investigating mixing

in a multi-dimensional rotary mixer: experiments and simulations, Powder
[1] M. Moakher, T. Shinbrot, F.J. Muzzio, Experimentally validated computations of Technology 201 (2009) 83–92.
flow, mixing, segration of noncohesive grains in 3D tumbling blenders, Powder [20] C. Harwood, Powder segregation due to vibration, Powder Technology 16
Technology 109 (2000) 58–71. (1977) 51–57.
[2] L. Fan, Y. Chan, F. Lai, Recent developments in solids mixing, Powder Technology [21] R. Chang, S. Badway, M. Hussain, J. Beuler, A comparison of free flowing, seg-
61 (1991) 255–287. regating and non-free-flowing, cohesive mixing systems in assembling the
[3] A. Alexandar, A.T. Shinbrot, F. Muzzio, Scaling surface velocities in rotating performance of a modified v-shaped solid mixer, Drug Development and Indus-
cylinders as a function of vessel radius, rotation rate, and particle size, Powder trial Pharmacy 21 (1995) 361–368.
Technology 126 (2002) 174–190. [22] F. Muzzio, M. Roddy, D. Brone, A. Alexander, O. Sudah, An improved powder
[4] M. Poux, P. Fayolle, J. Bertrand, D. Bridoux, J. Bousquet, Powder mixing: sampling tool, Pharmaceutical Technology (1999) 92–110.
some practical applied to agitated systems, Powder Technology 68 (1991) [23] G. Vergote, T. De Beer, C. Vervaet, J. Remon, W. Baeyens, N. Diericx, F. Ver-
213–234. poort, In-line monitoring of a pharmaceutical blending process using FT-Raman
[5] P. Portillo, M. Ierapetritou, F. Muzzio, Characterization of continuous convective spectroscopy, European Journal of Pharmaceutical Sciences 21 (2004) 479–485.
powder mixing processes, Powder Technology 182 (2008) 368–378. [24] K. Lim, V. Gururajan, P. Agarwal, Mixing of homogeneous solids in bubbling flu-
[6] M. Ashton, F.H.H. Valentin, The mixing of powders and particles in industrial idized beds: theoretical modelling and experimental investigation using digital
mixers, Transactions of the Institution of Chemical Engineers 44 (1966) T166. image analysis, Chemical Engineering Science 48 (1993) 2251–2265.
[7] J. Gray, Performance of dry solids mixing equipment, Chemical Engineering [25] G. Leonard, F. Bertrand, J. Chaouki, P. Gosselin, An experimental investigation
Progress 53 (1957) 25. of effusivity as an indicator of powder blend uniformity, Powder Technology
[8] K. Carley-Macauly, M. Donald, The mixing of solids in tumbling mixers—I, 181 (2008) 149–159.
Chemical Engineering Science 17 (1962) 493–506. [26] C. Lai, D. Holt, J. Leung, C. Cooney, G. Raju, P. Hansen, Real time and noninva-
[9] K. Carley-Macauly, M. Donald, The mixing of solids in tumbling mixers—II, sive monitoring of dry powder blend homogeneity, AIChE Journal 47 (2001)
Chemical Engineering Science 19 (1964) 191–199. 2618–2622.
[10] T. Ramanujam, S. Davies, D. Venketaswarlu, Studies on the mixing of solids in [27] J. Staniforth, S. Iveson, An investigation of the use of UV fluorescence
a double cone blender, Indian Journal of Technology 7 (1969) 218. microscopy as a method for quantifying the homogeneity of powder mixes,
[11] A. Alexander, T. Shinbrot, F.J. Muzzio, Granular segregation in the double cone International Journal of Pharmaceutics 31 (1986) 145–150.
blender: transitions and mechanism, Physics of Fluids 13 (2001) 578–587. [28] S. Sekulic, J. Wakeman, P. Doherty, P. Hailey, Automated system for the on-line
[12] A. Chester, J. Kowalski, M. Coles, E. Muegge, F. Muzzio, D. Brone, Mixing dynam- monitoring of powder blending processes using near-infrared spectroscopy.
ics in catalyst impregnation in double cone blenders, Powder Technology 102 Part II. Qualitative approaches to blend evaluation, Journal of Pharmaceutical
(1999) 85–94. and Biomedical Analysis 17 (1998) 1285–1309.
[13] D. Brone, F. Muzzio, Enhanced mixing in double cone blenders, Powder Tech- [29] A. Patel, P. Luner, M. Kemper, Quantitative analysis of polymorphs in binary and
nology 110 (1998) 179–189. multi-component powder mixtures by near-infrared reflectance spectroscopy,
[14] R. Chang, J. Shukla, J. Buehler, An evaluation of a unit dose compacting sample International Journal of Pharmaceutics 206 (2000) 63–74.
thief and a discussion of content uniformity testing and blending validation [30] C. Ufret, K. Morris, Modeling of powder blending using on-line near-
issues, Drug Development and Industrial Pharmacy 22 (1996) 1031–1035. infrared measurements, Drug Development and Industrial Pharmacy 27 (2001)
[15] A. Alexander, T. Shinbrot, F.J. Muzzio, Segregation patterns in V-blenders, 719–721.
Chemical Engineering Science 58 (2003) 487–496. [31] B. Chaudhuri, A. Alexander, A. Faqih, F. Muzzio, C. Davies, M. Tomassone,
[16] K. Hill, J. Kakalios, Reversible axial segregation of rotating granular media, Avalanching flow of cohesive powders, Powder Technology 164 (2006) 13–21.
Physics Review E 52 (1995) 4393. [32] A. Faqih, B. Chaudhuri, A. Alexander, S. Hammond, F. Muzzio, M. Tomassone,
[17] D. Brone, C. Wightman, K. Connor, A. Alexander, F. Muzzio, P. Robinson, Using Dilation effects of cohesive granular material in a rotating drum: experiments
flow perturbations to enhance mixing of dry powders in V-blenders, Powder and simulations, AICHE Journal 52 (2006) 4124–4132.
Technology 91 (3) (1997) 165–172. [33] T. Shinbrot, A. Alexander, F. Muzzio, Spontaneous chaotic granular mixing,
[18] J. Vernon, W. Hugen, A. Trujillo, Pharmaceutical manufacturing efficiency, drug Nature 397 (1999) 675–678.
prices and public health: examining the casual links, Drug Information Journal [34] B. Chaudhuri, A. Mehrotra, F. Muzzio, M. Tomassone, Cohesive effects in powder
(2007). mixing in a tumbling blender, Powder Technology 165 (2006) 105–114.