Sie sind auf Seite 1von 11

Powder Technology 293 (2016) 15–25

Contents lists available at ScienceDirect

Powder Technology

journal homepage:

Characterization of granular mixing in a helical ribbon blade blender

T.A.H. Simons a,⁎, S. Bensmann b, S. Zigan c, H.J. Feise b, H. Zetzener a, A. Kwade a
TU Braunschweig, Institute for Particle Technology, Volkmaroder Str. 5, 38104 Braunschweig, Germany
BASF SE, Chemical and Process Engineering, 67056 Ludwigshafen, Germany
University of Greenwich, School of Engineering, Chatham Maritime, Kent ME4 4TB,UK

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

Article history: Experiments of bulk solid mixing in a twin ribbon blade blender have been performed in this work in order to
Received 26 August 2014 characterize mixing behavior in such a mixer for binary mixtures with different cohesionless materials. The ef-
Received in revised form 10 November 2015 fects of fill height and blade rotation speed on mixing homogeneity have been studied. Mixing homogeneity
Accepted 15 November 2015
was determined by sampling. It has been observed that mixing is relatively fast towards a final mixing state with-
Available online 18 November 2015
in approximately 100 blade rotations for different combinations of material, fill height and blade rotational speed.
Moreover, these final mixing states seemed stable within a range of fluctuations, which may prove useful in de-
Granular mixing experiments termining an optimal mixing time for binary, cohesionless particle mixtures and potentially lead to a reduction in
Ribbon blade blender the required number of blade rotations in mixers of this type in industry. Mixing homogeneity results indicated
Mixing homogeneity an increased final mixing homogeneity for increasing fill height, most clearly in the range of 30–70 vol.% in the
Torque studied twin ribbon blade mixer. The torque on one of the shafts was determined. The latter results showed
that no significant influence of rotational speed on the required torque for the tested mixtures at rotational
velocities in the range of Fr 0.17–1.1 could be determined for most combinations of the tested materials and
fill heights in this work. The quantitative characterization of mixing behavior with the two key parameters
mixing homogeneity and torque on the shaft may be used for mixer validation of DEM simulations.
© 2015 Elsevier B.V. All rights reserved.

1. Introduction mixing in a non-intrusive way and to save part of these experiments.

However, the results from these DEM simulations require experimental
Granular mixing is a unit operation encountered in many process in- results for validation. While studying batch mixers, velocity profiles
dustries and has been a topic of experimental research for many de- near walls and particle trajectories were often chosen as parameters
cades [1–8]. It plays an important role in e.g. preparing correct to validate or compare simulation results with laboratory experimental
amounts of active pharmaceutical ingredients for tablets or ensuring results by e.g. particle image velocimetry (PIV) [13,14] and positron
constant product quality in detergents or blends of spices. In contrast emission particle tracking (PEPT) [15–19] techniques respectively. PIV
with the mixing of fluids, mixing of particles is difficult to predict due can provide particle velocity profiles using high-speed camera images
to e.g. distribution of properties such as particle size and shape leading of the particles in a defined section of the system near a wall. Local
to phenomena such as segregation. This is why the study of granular mixing homogeneities near the wall may be determined from particle
mixing and determining the ‘mixedness’ of a granular mixture often positions on these images as well. PEPT offers the opportunity to track
remains mainly empirical and mixing behavior is still hard to predict one or more particles throughout the mixer and in this way establish
reliably. particle velocity profiles. These methods are very powerful as they
Determining the ‘mixedness’ or mixing homogeneity empirically offer the advantages of being non-intrusive and of providing particle
was mostly done in the experiments by sampling. Reviews on sampling velocity results that can be directly compared with simulation results
and sampling theory have been published in [1,4,7–11]. However, [16,18]. However, PIV is limited or incapable of providing direct
sampling is laborious and may be prone to errors, since this method is information on the mixing homogeneity of the inner parts of a mixer,
intrusive and may therefore disturb the mixing state. With the advent which becomes more important in larger systems as encountered in
of simulation using the Discrete Element Method (DEM), originally pro- industrial practice. PEPT is capable of studying the inner parts of an
posed by Cundall and Strack [12], as a new means of studying granular opaque system in a mixer by tracking one or a couple of particles,
mixing challenges, the topic greatly attracts the interest from academia providing information of one or some positions in the mixer at a time.
and industry. DEM simulation offers the opportunity to study granular It was demonstrated by Windows-Yule et al. that segregation, the
opposite of mixing, could be studied using PEPT in a vibrofluidized
⁎ Corresponding author. system under steady state conditions, as the authors were able to relate
E-mail address: (T.A.H. Simons). the residence time of the tracked particle to local compositions and in
0032-5910/© 2015 Elsevier B.V. All rights reserved.
16 T.A.H. Simons et al. / Powder Technology 293 (2016) 15–25

sampling, and the torque on one of the shafts are presented. Two
model systems were tested; the first consisting of two size overlapping
components, i.e. of similar size and size distribution, and the second
Roman symbols
consisting of two clearly differently sized materials, i.e. their particle
ffc flowability [−]
sizes were completely separated. The influences of blade rotational
Fr dimensionless Froude number [−]
speed and fill height on the development of mixing homogeneity and
g gravitational acceleration [m/s2]
on torque in this mixer were studied.
M mixing index [−]
P overall fraction of a mixture component [−]
R radius of mixing blade [m]
2. Experimental method
s measured standard deviation from sampling [−]
s2 measured variance from sampling [−]
2.1. Material and equipment
t time [s]
VA volume of particle type A [m3]
The materials used in this work for mixing homogeneity studies
VS sample volume [m3]
were differently colored plastic granules as the size overlapping model
x10 particle size at 10% of the weighted particle size distri-
system and limestone particles in different size ranges as the differently
bution, volume-based and, assuming constant particle
sized model system, as shown in Fig. 1. The images in Fig. 1 are not at the
density, also mass-based [μm]
same scale. Fig. 1a and b show microscopic images of the plastic
x50 weighted average particle size, volume-based and, as-
granules (BASF, Germany) consisting of the same plastic material, but
suming constant particle density, also mass-based [μm]
different in color. The differently colored plastic granules had
x90 particle size at 90% of the weighted particle size distri-
overlapping particle size distributions, both distributions being roughly
bution, volume-based and, assuming constant particle
200–1000 μm in size. The blue and orange plastic granules therefore
density, also mass-based [μm]
each represented one of the components with particle size overlap in
the first model system. Fig. 1c and d show limestone particles in two
Greek symbols
different size ranges, ordered from KSL Staubtechnik (Germany). Both
ρb bulk density [kg/m3]
limestone fractions were sieved into narrower fractions of 100–
σ true standard deviation from sampling [−]
400 μm (fine) (1d) and 500–800 μm (coarse) (1c) respectively to
σM standard deviation due to measurement error [−]
produce different and separated fractions, using each fraction as one
σ2 true variance [−]
of the components of the second model system. For the torque measure-
σ20 initial true variance [−]
ments a third system of 3 mm spherical polyethylene particles was used
σ2R true variance for a completely random mixture [−]
as well (BASF, Germany), as shown in Fig. 1e and f. Size and shape
φeff effective angle of internal friction [°]
distributions for all of these particle types were determined by a com-
φsf angle of internal friction at steady flow [°]
mercial image analysis setup (CamSizer, Retsch, Germany). All particles
ω rotational velocity of mixing blades [rad/s]
of the two model systems in the mixing homogeneity research were
non-spherical and free flowing (ffc N 10) for both limestone fractions
and the plastic granules according to Jenike's classification for
flowability [23,24]. Further frictional properties were determined
that way to determine mixing homogeneity [20,21]. However, in considering the potential use of the experimental work here to be corre-
unsteady-state systems this relation can no longer be used [21], and in lated with DEM studies. Frictional properties of the limestone fractions
batch high-shear mixers it is therefore difficult to determine the overall and plastic granules were measured in a Schulze type ring shear tester
state of mixing in the entire mixer i.e. the mixing homogeneity for the and are expressed as internal friction angle of the bulk instead of that
complete mixture, from these results. As mixing homogeneity is the of a single particle. In many DEM calibration studies frictional parame-
dominant property determining the quality of most mixtures, they are ters are calibrated using a shear tester and bulk internal friction param-
evaluated based on this property, describing the quality of the mixture. eters or simply bulk shear stress at a given normal load, as may be found
Therefore, sampling is still a powerful and relevant tool as it can provide in e.g. [25–30]. The internal friction angles at 3· 103 Pa normal load are
the information on the mixing homogeneity by calculating the differ- reported in Table 1 and expressed as both effective angle of internal
ences in composition in each sample. Although mixing homogeneity, friction as is commonly used in bulk solids engineering [31] and as
expressed in some form of mixing or segregation index, is studied angle of internal friction at steady flow as this is the mode in a shear
using DEM as well, investigations in which DEM mixing homogeneity experiment that is often used for calibration of DEM parameters in liter-
results are directly compared with experiments are scarcely found ature, assuming no cohesion [25–29]. A detailed description of different
[22]. Sampling was therefore chosen in this work as the method to char- friction angles may be found in the work by Schulze [24]. Details on the
acterize mixing behavior experimentally by determining mixing homo- particle size distribution of each component, expressed in x10, x50 and
geneity over mixing time for the entire mixer and to serve as validation x90 values as well as further physical data on all materials discussed
data for DEM simulations in later papers. above are summarized in Table 1.
Additionally torque measurements on a ribbon blade blender were Mixing behavior was characterized in a 10 L vertical twin helical
conducted, using the required torque for mixing a specific granular mix- ribbon blade mixer (type HM 10 by Amixon, Germany), of which a top
ture as a further parameter to characterize the effectiveness of granular view is shown in Fig. 2a. and as briefly introduced in Simons et al.
mixing. Since the DEM method is based on the calculation of contacts [32]. The ribbon blade mixer has 10 cm diameter blades; the distance
forces for interactions between particles and between particle and the between its shafts is 14.2 cm, allowing an overlap of the blades in the
wall, the overall required torque can also be used as a further validation mixer center. A side view schematic in Fig. 2b shows the flow profiles
parameter for simulation results. of the material in the mixer. The material in the mixer is rotated due
This work presents an experimental study of the characterization of to normal and shear forces between the blades and the material. On
mixing behavior in a twin helical ribbon blade blender using different top of the rotating mixing motion, the material to be mixed is conveyed
cohesionless mixtures. These mixtures were chosen as a basis for later up along the helical blades, indicated with the red arrows, and flows
comparison with simple DEM simulations, which can later be extended down near the shafts in the centers of the two helices, indicated by
for cohesive materials. Mixing homogeneity over time, determined by the blue arrows. An exchange of material takes place in the region
T.A.H. Simons et al. / Powder Technology 293 (2016) 15–25 17

Fig. 1. Microscopic images of colored plastic granules in blue (a) and orange (b), both in the size range of approx. 200–800 μm; photographic image of coarse (c) and fine (d) limestone
particles (500–800 μm and 100–400 μm resp.); microscopic images of red (e) and green (f) polyethylene spheres, both in the size range of approx. 3.0–3.5 mm; pictures (e) and
(f) courtesy by BASF SE.

where the blades overlap in the center of the mixer, shown by the green metal tube with a 5 cm slit through which an approximately 5–6 mL
arrow. sample can be taken by turning the tubes along each other. The bottom
of the thief was rounded to be able to also sample the lowermost parts
2.2. Sampling method of the mixer.
The mixer volume was divided into 32 possible sample locations
To study mixing homogeneity in the helical ribbon blade mixer, allowing for sampling from these volumes, as illustrated in Fig. 3 in
samples were taken with a custom designed sampling thief, shown in the left. Applying representative sampling, after each mixing time
Fig. 3 in the right. This sampling thief consisted of an inner and outer studied in this work 10 random samples from these 32 locations were

Table 1
Physical data of colored plastic granules, polyethylene spheres and natural limestone particles.

Material parameter Symbol Unit Plastic granules, Plastic granules, Limestone, fine Limestone, coarse Polyethylene Polyethylene
blue orange fraction fraction spheres, red spheres, green

Size distribution (Q3) x10 μm 259 203 118 513 3.0 · 103 2.9 · 103
Mean size (Q3) x50 μm 494 365 210 676 3.43 · 103 3.23 · 103
Size distribution (Q3) x90 μm 994 629 368 825 3.6 · 103 3.5 · 103
Sphericity – 0.50–0.85 size 0.43–0.82 size 0.74–0.94 size 0.61–0.91 size 0.95–0.99 size 0.94–0.99 size
dependent dependent dependent dependent dependent dependent
Bulk density ρb kg 315 ± 6 319 ± 5 1345 ± 9 1280 ± 4 543 ± 10 543 ± 10
Flowability ffc – N10 N10 N10 N10 N10a N10a
Effective internal friction φeff ° 39.5 ± 1 39 ± 1 36 ± 1 40.5 ± 1 – –
Internal friction angle at φsf ° 38 ± 1 37 ± 1 33.5 ± 1 38 ± 0.5 – –
steady flow
Based on visual observation, compared with the other materials in this work.
18 T.A.H. Simons et al. / Powder Technology 293 (2016) 15–25

Fig. 2. (a) Top view of laboratory ribbon blade blender, (b) schematic of side view showing material flow patterns.

taken respectively as representative for the entire mixer volume after Lacey [1] as a measure for mixing homogeneity, as shown in Eq. (1).
the respective mixing time. Although taking samples from all the 32 However, to be able to calculate the mixing index, both the initial
sample volumes in the entire mixer would produce statistically more variance of the mixture σ20 and the variance for a completely random
accurate results, removal of 32 samples would lead to complete destruc- mixture σR2 are required, as defined in Eqs. (2) and (3).
tion of the mixing state by removing too much material. Samples were
taken in 3 measurement series as a compromise to avoid factors σ 0 2 −σ 2
M¼ ð1Þ
influencing the mixture, such as removing too much material from the σ 0 2 −σ R 2
mixer on the one hand, and to save time and material costs by not
having to reload the mixer for each experiment on the other hand. σ 0 2 ¼ P ð1−P Þ ð2Þ
The sampling method of 3 measurement series is shown graphically in
Fig. 4 for better understanding. Fig. 4 shows the three series, all starting VA
σ R 2 ¼ P ð1−P Þ ð3Þ
with the same initial conditions at 0 s of mixing time and they are VS
indicated by the blue bars. Samples are taken at three points of mixing
time (not to scale) per series without reloading the mixer at the mo- Eq. (1) shows Lacey's mixing index, calculated from initial variance
ments, indicated by the red dots. In order to check the reproducibility σ20, variance during mixing σ2 and variance at complete random mixing
of this method, the mixing time last studied in a series was repeated σ2R. As shown in Eq. (2), the initial variance for a binary mixture can be
in the next series. This led to a series of 15 s, 30 s and 1 min; a series calculated from the known fraction P of one the components. Assuming
of 1 min, 2 min and 3 min; and a series of 3 min, 5 min and 10 min of constant sample volume, Eq. (3) can be applied in which VA represents
total mixing time respectively, totaling 90 samples at 7 mixing times the largest particle's volume and VS represents the sample volume [33,
for a complete run of 3 series. The mixing times of 1 min and 3 min 34]. The Lacey mixing index thus expresses the mixing achieved at a
are therefore sampled twice and this will be reflected in the graphs point of time, divided by the maximum achievable mixing homogenei-
presenting the results (see Section 3.2). ty. The index offers the advantage of being easily understandable as it is
The composition of each sample was determined with the methods 0 at complete segregation and 1 at maximum mixing homogeneity
presented in the next subsection. In both analysis methods average achieved. A possible disadvantage of this index may be that results are
volume fractions could be calculated for each sample for a key compo- found in a narrow range between 0.9 and 1.0 if only interested in the
nent, being one color of the colored plastic granules or one size fraction final mixing state.
of limestone. The standard deviation s and its square the variance s2 in Mixing behavior was studied at different blade rotational speeds (40
composition were determined from a set of ten samples taken at one and 80 rpm; Fr 0.17 and Fr 0.70 resp.) and different fill heights of the
point in mixing time. Applying a χ2-distribution as shown e.g. by Stiess, mixer (30 vol.%–50 vol.%–70 vol.%–90 vol.%). These fill heights represent
it was obtained that the measured variance s2 is less than twice the true mixture volumes of 3 L, 5 L, 7 L and 9 L respectively. Additionally, mixing
variance σ2 with 95% confidence [33]. The measured standard devia- behavior was studied by manually moving the blades to a fixed number
tions were then used to calculate the mixing index as defined by of complete rotations. This was done by pushing the upper end of the

Fig. 3. Schematic of sampling positions and image of sampling thief with a slit length of 5 cm.
T.A.H. Simons et al. / Powder Technology 293 (2016) 15–25 19

Fig. 4. Graphical overview of sampling method indicating the 3 measurement series by blue bars and the moments the samples were taken in time (not to scale) by red dots. Each mea-
surement series started with the same initial conditions, for which the mixer was reloaded.

blades in the direction of rotation for fill heights of 50% and 70%. In these The torque on shaft was recorded with 25 measurements per
cases the blades were not completely covered with mixture material. second, i.e. with 40 ms intervals, during 2 min, consisting of one minute
Mixtures of 50–50 vol.% orange with blue plastic particles were of measurements while mixing and one minute while the mixer was
prepared as well as mixtures of 50–50 vol.% limestone fine and coarse halted. The purpose of the latter was to determine the drift in the
fractions. No mixtures of plastic granules with limestone fractions measurement, which subsequently was subtracted from the torque
were studied. The mixing index was studied as a function of mixing while mixing to produce a net torque. This procedure was followed
time, blade rotation speed and fill height. for the mixer being loaded, as a function of rotational speed and fill
height, and for the mixer being empty, as a function of rotational
speed. Subsequently, the net torques for the empty mixer were
2.3. Sample analysis methods
subtracted from the net torques for the loaded mixer.
All torque measurements were conducted at rotational speeds
Sample analysis was carried out with two methods, depending on
ranging from 30 to 100 rpm in intervals of 10 rpm for the plastic granule
the difference in particle properties.
mixtures and the polyethylene spheres mixtures. The measurements
The colored plastic granules were analyzed with color analysis of
with the limestone mixtures were conducted within this range, in a
pictures of these samples. Samples were spread out on a sheet of
range from 40 to 100 rpm in intervals of 20 rpm for the 70% fill height
paper after subdividing it into 4 subsamples. Each subsample was
case and at 40 and 80 rpm for the cases with 50 and 90% fill height.
then photographed and analyzed. Pictures were taken in a photo tent,
This means that all measurements were obtained for values of a Froude
lit by three flood lights, to prevent shadows and reflections. In addition,
number in the range Fr 0.1–1.1. The dimensionless Froude number ex-
the analysis software for detecting colors was calibrated with known
presses the ratio of inertial forces exerted on the material by the blades
samples of orange and blue particles to allow for calculating volume
over the gravitational forces, based on a characteristic velocity. The
fraction by color in the mixer from area fraction by color from the
Froude number for mixing is shown in Eq. (4).
The limestone particles, differing in particle size, were analyzed by a
ω2 R
commercial image analysis setup (CamSizer, Retsch, Germany). Sam- Fr ¼ ð4Þ
ples were analyzed by video detection and subsequent size analysis of
particles falling down a vibrating chute, using a camera and a backlight.
In Eq. (4) ω represents the rotational velocity in rad/s, R the diameter
The particles featuring different size fractions, a cut at 424 μm was
of the blade in m and g the gravitational acceleration in m/s2. A Froude
defined to classify measured fractions into fine and coarse fractions
number below 1 indicates that gravitational effects dominate over
per sample. It was found that abrasion of the limestone mixtures during
centrifugal effects. The corresponding Froude numbers for the studies
the mixing procedure was negligible until 800 rotations in this type of
mixer and for the studied rotational speeds.

2.4. Torque analysis

Torque was measured directly on one of the shafts of the ribbon

blade mixer. This was done to produce a measurement of the required
torque of only the mixing of particles as accurate as possible while e.g.
not having to take into account the losses due to friction in the gear
box or (frequency) converter when measuring torque indirectly or
measuring the energy input instead of torque.
The setup is shown schematically in Fig. 5. The torque transducer
(type T5 by HBM, Germany) was placed on one of the shafts between
the gear box and the mixer container. The shafts were extended for
this purpose. The torque transducer had a measurement range between
0 and 20 Nm with a maximum measurement deviation of b0.2% and a
standard deviation of the reproducibility of 0.05% [35]. It was protected
against torque overloads by mounting safety couplings (R + W,
Germany) on both shafts. Measurements were recorded with a PC Fig. 5. Schematic setup of the twin ribbon blade mixer with the torque transducer placed
using a measurement amplifier (type MVD 2555 by HBM, Germany). directly at the shaft.
20 T.A.H. Simons et al. / Powder Technology 293 (2016) 15–25

on mixing homogeneity at 40 and 80 rpm are Fr 0.17 and 0.7 respective- Although this may lead to lower reproducibility of the results during
ly. The calculation of the Froude number was based on the blade speed the first rotations, it is generally the final mixing state and how fast
in rotational direction and in the plane normal to the shafts only, mean- this state may be reached in which industry is interested.
ing that no adjustment to the characteristic rotational velocity due to As a complete mixing run of 10 min for a given fill height and rota-
the inclination angle of the helical blade was applied. tional speed was divided into three series of three measurements in
time, it was verified that the results of the three series within one run
3. Results overlap in order for them to be comparable and be treated as one run.
With the exception of the case with 40 rpm and 70 vol.% fill height
3.1. Qualitative results after 1 min of mixing time, there was at least a one-sided overlap for
all cases, i.e. at least one measurement of one of the series is within
Mixing behavior may be readily observed qualitatively from top the confidence interval of the other series after the same mixing time.
view images from the ribbon blade blender after several mixing times. This overlap implies that results from different series in one run can
Fig. 6 shows the top view images for the mixture of differently colored be directly compared and therefore be regarded as one run. This may
plastic granules in the initial state and after 1, 2, 4 and 8 rotations be more clearly seen from Fig. 7b and c, in which results are shown
respectively for 50% fill height in the first line and for 0, 20, 40 and 80 separately for different rotational speeds with focus on the first 400
rotations for 50% fill height and 80 rpm in the second line. rotations. One run of three series is each time shown in one color in
From Fig. 6 it may be noted that during the first blade rotations these figures. Double data markers and corresponding confidence
orange and blue particles seem to switch sides and are then mixed intervals show whether an overlap between series is present.
more thoroughly in the next rotations. The results after 20 and 40 Fig. 7b and c also show a final mixing homogeneity significantly lower
rotations show small ribbons of blue particles in the mixing patterns. for 30% fill height than for those at 50% and 70% fill height. This is shown
These were typically also found for 70% fill height mixtures. After 80 more pronounced in the case of 80 rpm. One cause for this deviation
rotations, the mixture approaches a stochastic mixing state and top could be that sample sizes for 30% fill height mixtures were approximate-
views do not change significantly any more. ly 60% of the size of those for 50% and 70% fill height, as sampling was im-
peded at this low fill height. Samples could only be taken from 16
3.2. Sampling results subvolumes instead of 32 for the other fill heights. Additionally, as this
type of mixers is mostly run at high filling degrees [36], mixing perfor-
Quantitative results were obtained by sampling the mixture after mance, i.e. the final mixing homogeneity and at which rate this was
fixed mixing times in the manner described in Section 2.2. Mixing achieved, may be significantly lower at 30% filling degree, as mixing in
homogeneity could be expressed in Lacey's mixing index over mixing axial direction is almost completely hindered. Further focusing on differ-
time, as presented in Section 2.2. ences in fill height it was observed that, with the exceptions at 10 and 40
Fig. 7 shows Lacey's mixing index for experiments with plastic rotations for the case with 40 rpm and 70% fill height, all measured
granules versus number of blade rotations. Data markers in Fig. 7 mixing indices for 70% fill height were higher than the results for 50%
represent the measured mixing index, based on the measured standard fill height after the same number of rotations for both rotational speeds
deviation s2, whereas error bars in Fig. 7 represent 95% confidence (see Fig. 7b and c). Although these results strongly suggest that higher
intervals, within which the true mixing index, based on the theoretical final mixing homogeneities may be achieved with higher fill heights,
standard deviation σ2, is. The 95% confidence intervals were determined this could not be determined unambiguously from these results, as con-
by applying the χ2-distribution [33]. For presentation purposes, the fidence intervals overlapped for the 50 and 70% fill height cases, meaning
y-axes in Fig. 7 are displayed at different scales. Double data markers that their true mixing indices are in the same range.
after the same number of rotations indicate results from different series Fig. 7d shows results for the mixing index of the mixer at 50% fill
in Figs. 7 and 8. height for short mixing times, for different blade rotational speeds as
Fig. 7a shows the complete mixing performance over a mixing time well as for rotating the blades manually. Comparing the data for the
of 10 min for two different blade rotational speeds (40 and 80 rpm) and results at 40 and 80 rpm in this graph, based on the results after 20
three different fill heights (30 vol.%, 50 vol.% and 70 vol.%). Based on the and 80 rotations, the results might suggest that an increase in rotational
increase in Lacey's mixing index, it may be readily seen that mixing is speed to 80 rpm might lead to a higher mixing index. However, the
very fast and seems to reach its final state within 100 rotations for all differences between the two rotational speeds were not as big as
six cases. Considering the confidence intervals, it was found difficult to those between different fill heights in e.g. Fig. 7c and confidence
determine mixing homogeneity in this mixer for the first 20 rotations. intervals were clearly overlapping. Therefore, no significant differences

Fig. 6. First line: top view results from manual rotation (0, 1, 2, 4 and 8 rotations); second line: top view results from automated rotation at 80 rpm and 50% fill height (0, 20, 40 and 80
T.A.H. Simons et al. / Powder Technology 293 (2016) 15–25 21

Fig. 7. (a) Complete mixing experiment over 10 min for 6 combinations of fill height and blade rotational speed for experiments with plastic granules; b. mixing index for experiments at
40 rpm; c. mixing index for experiments at 80 rpm; d. comparison of automatic and manual results for mixing index at 50% fill height.

in mixing index could be determined between 40 and 80 rpm experi- With increasing fill height, the mixing index after the same number of
ments for this mixture. Exactly 1, 2, 4 or 8 rotations were studied man- rotations was observed to increase as well, a similar trend as found for
ually to study mixing homogeneity during the very first blade rotations, the experiments with plastic granules. As for the results for plastic gran-
which was not possible automatically due to the short start-up and ules, the results in Fig. 8a, with an exception after 20 rotations, suggest
slow-down times of the mixing blades in which the number of rotations that higher mixing homogeneity can be achieved with higher loading.
could not exactly be determined. Results from this study are shown in However, this cannot be determined significantly due to the overlap
the embedded diagram in Fig. 7d. The large confidence intervals in of confidence intervals (see e.g. Fig. 8a), meaning that their true mixing
this embedded diagram show that determining mixing homogeneity indices may be in the same range.
exactly after a low number of rotations is difficult. Although mixtures of particles of different size distributions are
Results of mixing experiments with the two limestone fractions are known to segregate, no clear segregation in terms of a decreasing
shown in Fig. 8. Fig. 8a shows the results for the mixing index as a func- mixing index could be observed compared with the results for plastic
tion of the number of rotations for different fill heights at 80 rpm and granules. However, it may be observed that measured mixing indices
Fig. 8b shows the results for the combinations of two rotational speeds for the same number of rotations and the same fill height are lower
and two fill heights on the same scale, whereas Fig. 8c and d show re- and confidence intervals are larger for mixtures consisting of the
sults for 50% and 70% fill heights respectively in the first 100 rotations different limestone fractions than those of plastic granules in the same
on a wider scale including the data from manual rotation experiments. size fractions (compare e.g. the cases for 80 rpm, 50 and 70% respective-
As observed in the case for mixing two plastic granule fractions, in the ly, in Fig. 7c and d with Fig. 8b and d). This implies that final mixing re-
case for two limestone fractions mixing also seemed to be very fast sults for limestone, of different size fractions, may be not as well-mixed
and a final stage was reached within 100 rotations here as well. as a mixture of plastic granules, of similar and overlapping size fractions.
Comparing the overlap of sequential series within a run resulted in an This may be an indication that equilibrium between mixing and
overlap with the exception of 80 rpm at 70% fill height after 80 rotations. segregation has formed. Fig. 8a also suggested this, considering the
Fig. 8c and d show mixing index results for the first 80 rotations fluctuation in the data for mixing index for 50% fill height after 160
while the mixer was operated at different rotational velocities with rotations with decreases in mixing index to 0.95. This fluctuation is
70% fill height (8c) and 50% fill height (8d) respectively. It may also be not as pronounced in the same case at 50% fill height for plastic granules
observed from Fig. 8c and d that large confidence intervals were found in Fig. 7b and c, the latter case itself showing these fluctuations in the
in the first 10 or 20 rotations, indicating that mixing index is hard to cases for 30% fill height only. Future research may focus on studying
determine in the beginning of the mixing event. Phenomena are similar the mixing of binary mixtures of different size ratios in order to
to those observed for plastic granules in overlapping size distribution. determine whether a possible equilibrium between size segregation
22 T.A.H. Simons et al. / Powder Technology 293 (2016) 15–25

Fig. 8. (a) Complete mixing experiment over 10 min for 3 fill heights for experiments with limestone;(b) comparison of mixing index for a combination of two fill heights and two rota-
tional speeds; (c) comparison of automatic and manual results for mixing index at 70% fill height; (d) comparison of automatic and manual results for mixing index at 50% fill height.

and mixing due to blade rotation for this mixer type forms and charac- 0.014 (or σ2M = 2 · 10−4) on a volume fraction basis and volumetric
terize this possible equilibrium as a function of size ratio. It should be sample sizes were similar as well. This led to maximum measurable
noted here that the composition of the samples of plastic granules and mixing indices M of 0.9992 or minimum segregation indices 1 − M of
the samples of limestone fractions were analyzed by different methods. 8 · 10− 4 (see Fig. 9) and both experiments produced values above
However, the estimated errors in both analysis methods were ca. σM = this minimum.

Fig. 9. Segregation index (1 − M) results for (a) plastic granule mixtures and (b) limestone mixtures.
T.A.H. Simons et al. / Powder Technology 293 (2016) 15–25 23

The differences in results for different fill height can be observed shows the specific required torque in the mixer per kg of mixture as a
more clearly when plotting 1 − M logarithmically, as shown in Fig. 9. function of volumetric fill height fraction. The required torques for all
The index 1 − M represents a segregation index: complete segregation fill heights were in the order of magnitude of several Nm for the lime-
is at 1 and complete mixing at 1 − M = 0. It can be seen that these stone mixtures and less than 1 Nm for the plastic materials. All materials
results for both mixtures suggested that higher fill heights led to were cohesionless and the plastic materials (both the granules and the
lower final segregation indices. However, the differences between the spheres) possessed relatively low densities as well. The plastic granules
two higher fill heights were not as significant as those between the mixture and the polyethylene spheres mixture produced torques below
lowest and higher fill heights. In addition, these results confirmed that 1 Nm with relatively high standard deviations. These high standard
maximum measurable mixing homogeneity was achieved after approx- deviations can be explained by the very low torques, measured in the
imately 100 rotations for both mixtures. lower measuring range of the torque transducer. No significant influ-
High grades of mixing homogeneity was achieved rather fast in a rib- ence of the rotational speed on the torque was determined from the
bon blade mixer for easy flowing materials, represented by mixtures of data. Although the results for polyethylene spheres at 50% fill height
plastic granules in overlapping size distribution and by mixtures of suggested an increased torque for rotational speeds between 50 and
limestone fractions in different size distribution. A final measurable 80 rpm this was in fact particularly due to low reproducibility of the
mixing state was achieved within 100 blade rotations for both types of empty run measurements.
mixtures, for fill heights from 30 to 90% and for both rotational speeds The 50–50 vol.% mixtures of fine and coarse limestone particles
in this work, at 40 and 80 rpm. Moreover, apart from some observed produced torques on the shaft in the range of 1.5 to 4.5 Nm for different
fluctuations, Figs. 7-9 showed that these high states of mixing homoge- fill heights. For all studied mixtures torque was found no significant
neity were maintained during 400 or 800 rotations respectively. These function of rotational speed in the tested range of rotational speeds.
findings may, after a scale-up study, potentially be useful for optimizing All experiments were conducted in a range of Fr 0.1–1.1. The absence
mixing times in similar mixer types in industry. Large confidence inter- of a significant influence of rotational speed on the required torque for
vals at mixing states below 20 rotations as well as at mixing states for Fr ≤ 1 is in accordance with literature [37,38]. To be able to study the
30% volume fractions indicated difficulties in determining mixing effect of fill height of the mixer on the required torque the mass specific
homogeneity by sampling. Results for 50%, 70% and 90% fill height torques, expressed in Nm per kg material load, were calculated by divid-
were shown to be more accurate in determining mixing homogeneity, ing the required torque on one shaft by the corresponding half the mass
based on Lacey's mixing index. in the mixer. These results are shown in Fig. 10d for better comparison
among several materials and different fill heights. They showed an
3.3. Torque measurements results increase in mass specific torque, i.e. the torque per unit mass, with
increasing mass of the material in the mixer for the limestone mixture
The torque was measured experimentally at one of the shafts of the in a fill height range of 50–90%. This means that torque was not only
twin ribbon blade mixer for several materials at different fill height as a observed to increase with mass in the mixer, but was found to increase
function of rotational speed. All torque measurements shown in this more than proportional with mass for the limestone mixtures, as
work were based on conditions at constant rotational speed. The torque opposed to the linear dependence on mass reported by Cooker and
was observed to show minimal periodical fluctuations with time, which Nedderman for a single vertical ribbon blade mixer [37]. The average
is believed to be an effect of the blade passes near the protruding rods in values of the mass specific torque for both plastic materials, i.e. the plas-
the mixer container. The frequency of the observed torque fluctuations tic granules mixture and the polyethylene spheres mixture, suggested a
corresponded to the frequency of the rotating blades and an example similar increase, but due to error propagation of the standard deviations
measurement of this periodical behavior is shown in Fig. 10 during in the torque measurements it was very difficult to confirm this trend
the first 80 s of the measurement. The measurements usually included significantly.
more than 1000 data points. The increase in specific torque in this work for the twin ribbon blade
Fig. 11 shows the results from torque measurements for mixer is assumed to be a consequence of a slight increase in bulk density
(a) 50–50 vol.% mixtures of differently colored plastic granules, of the material with increasing fill height as well as the presence of dou-
(b) 3 mm spherical polyethylene particles and (c) 50–50 vol.% mixtures ble blades. In the center of the mixer a shear zone is created in the region
of fine and coarse fractioned limestone particles respectively. Fig. 11d where the two blades overlap and material is exchanged. With increas-
ing fill height this overlap exists over a larger region and may therefore
contribute to an increase in torque on the blade. Although abrasion
might increase the required torque it was found that abrasion of the
limestone mixtures during the mixing procedure was negligible until
800 rotations. Research on local stress on the blades e.g. using pressure
sensors on the blade may provide more detailed information on the
local stress phenomena in the center of the mixer and the origins of
the observed torque increase. DEM simulation may assist in studying
local shear zones in the center of the mixer. In addition, torque measure-
ments on more model mixtures are required to confirm the increase in
mass specific torque with increasing fill height of the twin ribbon blade
mixer as a general mixer property or for specific materials only.
Furthermore, the results for the mixing homogeneity over mixing
time and the required torque on shaft may be used as key validation
parameters for validation of simulations using the discrete element
method (DEM), since both parameters can be easily determined from
DEM simulation results. Numerical studies using DEM are planned to
be presented in a later paper.
As both mass specific torque and final mixing homogeneity
increased with increasing fill height for the tested limestone mixtures
Fig. 10. Exemplary torque measurement on one of the mixing shafts of the twin ribbon in the twin ribbon blade mixer, future work may also be dedicated to
blade mixer for a limestone mixture. couple torque results to the results for mixing homogeneity in order
24 T.A.H. Simons et al. / Powder Technology 293 (2016) 15–25

Fig. 11. Net measured torque on shaft in ribbon blade mixer for experiments with (a) 50/50 mixtures of plastic granules, (b) 3 mm spherical polyethylene particles and (c) 50/50 fine and
coarse fractioned limestone particles respectively; (d) mass specific torque for ribbon blade mixer as a function of fill height fraction.

to confirm whether a correlation between torque results and mixing ho- 2. High mixing homogeneities have been achieved within 100 blade
mogeneity holds and establish a means for determining mixing efficien- rotations for both model mixtures and, apart from some fluctuations
cy. This may offer potential for determining required energy inputs to at lower fill heights, are maintained during 400 or 800 rotations at 40
achieve a predefined mixing state while coupled to material properties, or 80 rpm respectively. These results showing fast and stable mixing
such as particle size, shape, cohesion and adhesion, using a larger may be of potential use for optimizing mixing times in similar
dataset of mixtures in this mixer. mixers, using cohesionless materials.
3. Higher filling degree in the studied twin ribbon blade mixer have led
4. Conclusion and outlook to better mixing performance in terms of higher mixing index
achieved faster for the plastic granules mixture and has been
Experimental results of bulk solids mixing in a twin ribbon blade suggested from the data with limestone mixtures as well.
blender have been shown in this work in order to characterize mixing 4. No significant, clear dependence of the required torque on blade rota-
behavior in such a mixer. Characterization was done by identifying tional speed has been found for rotational speeds with Fr ≤ 1 at several
mixing homogeneity by sampling and by measuring the torque on the fill heights for plastic granules mixtures and for polyethylene spheres
mixing shaft. Mixing homogeneity was studied over the number of mixtures and at 70% fill height for differently sized limestone fractions.
rotations for two different mixtures of materials, as a function of blade This has been reflected in experimental results for torque on shaft at
rotation speed and fill height for constant volume fractions of each com- rotational speeds corresponding to a range of Fr 0.1–1.1.
ponent. The following was observed from laboratory experiments with
The results for both mixing homogeneity and torque on one of the
a mixture of plastic granules in overlapping particle size distributions
shafts may be used as validation experiments for simulation with the
and with another mixture of limestone in fractions with different size
Discrete Element Method (DEM) and a numerical study of this mixer
using DEM is planned to be presented in a later paper.
1. The sampling method is appropriate for determining mixing In addition, to study the possible equilibrium between mixing by the
homogeneity after 20 rotations, but shows very broad confidence blades and segregation due to size differences in the material may be
intervals at low mixing degree and when operated at 30% fill studied experimentally testing a range of binary mixtures of compo-
height. Higher accuracy in the beginning of the mixing event nents with different size ratios in the twin ribbon blade mixer.
(b 20 rotations) may be achieved by taking more samples for the Local force studies on the blades may in future help in explaining the
first rotations. increase in mass specific torque, possibly aided by DEM simulations.
T.A.H. Simons et al. / Powder Technology 293 (2016) 15–25 25

Finally, the combined results of mixing homogeneity over time with [17] J.R. Jones, D.J. Parker, J. Bridgwater, Axial mixing in a ploughshare mixer, Powder
Technol. 178 (2007) 73–86.
the torque input might be used in future to determine required energy [18] A. Hassanpour, H. Tan, A. Bayly, P. Gopalkrishnan, B. Ng, M. Ghadiri, Analysis of
inputs to achieve a predefined mixing state and in such a way evaluate particle motion in a paddle mixer using Discrete Element Method (DEM), Powder
the efficiency of the combination of a mixer and agitator. However, a Technol. 206 (2011) 189–194.
[19] M. Marigo, M. Davies, T. Leadbeater, D.L. Cairns, A. Ingram, E.H. Stitt, Application of
larger database of combinations of mixers and mixtures would be Positron Emission Particle Tracking (PEPT) to validate a Discrete Element Method
required to develop such a model. (DEM) model of granular flow and mixing in the Turbula mixer, Int. J. Pharm. 446
(2013) 46–58.
[20] C.R.K. Windows-Yule, T. Weinhart, D.J. Parker, A.R. Thornton, Effects of packing
Acknowledgment density on the segregative behaviors of granular systems, Phys. Rev. Lett. 112
(2014) 098001–098005.
The presented work was part of the PARDEM project and the [21] C.R.K. Windows-Yule, A.D. Rosato, N. Rivas, D.J. Parker, Influence of initial conditions
on granular dynamics near the jamming transition, New J. Phys. 16 (2014) 1–18.
funding by the EU Marie Curie ITN (Project No. ITN-238577) is gratefully
[22] S.S. Manickam, R. Shah, J. Tomei, T.L. Bergman, B. Chaudhuri, Investigating mixing in
acknowledged. In addition, the authors would like to thank Paul a multi-dimensional rotary mixer: experiments and simulations, Powder Technol.
Klassen, Julia Wittmann, Juri Wittmann and Christian Sandy for their 201 (2010) 83–92.
help with the mixing experiments as well as Markus and Robert Heinz [23] A.W. Jenike, Storage and flow of solids, Bull. Utah Eng. Exp. Station 123 (1964)
for their help with the flood light setup. TAHS is grateful for the [24] D. Schulze, Powders and Bulk Solids, Springer, Berlin, 2008 (Chapter 3).
additional image material provided by Ramon Cabiscol. [25] N. Estrada, A. Taboada, F. Radjaï, Shear strength and force transmission in granular
media with rolling resistance, Phys. Rev. E 78 (2008) 1–11.
[26] J. Härtl, J.Y. Ooi, Experiments and simulations of direct shear tests: porosity, contact
References friction and bulk friction, Granul. Matter 10 (2008) 263–271.
[27] J. Härtl, J.Y. Ooi, Numerical investigation of particle shape and particle friction on
[1] P.M.C. Lacey, Developments in the theory of particle mixing, J. Appl. Chem. 4 (1954) limiting bulk friction in direct shear tests and comparison with experiments,
257–268. Powder Technol. 212 (2011) 231–239.
[2] K.W. Carley-Macauly, M.B. Donald, The mixing of solids in tumbling mixers — I, [28] S. Strege, A. Weuster, H. Zetzener, L. Brendel, A. Kwade, D.E. Wolf, Approach to
Chem. Eng. Sci. 17 (1962) 493–506. structural anisotropy in compacted cohesive powder, Granul. Matter 16 (2014)
[3] W. Müller, H. Rumpf, Das Mischen von Pulvern in Mischern mit axialer 401–409.
Mischbewegung, Chem. Ing. Tech. 39 (1967) 365–373. [29] T.A.H. Simons, R. Weiler, S. Strege, S. Bensmann, M. Schilling, A. Kwade, A ring shear
[4] N. Harnby, The statistical analysis of particulate mixtures, part I. The sampling of tester as calibration experiment for DEM simulations in agitated mixers — a sensi-
mixtures and the resultant precision of estimates based on the sample, Powder tivity study, Proc. Eng. 102 (2015) 741–748.
Technol. 5 (1971/1972) 81–86. [30] A. Aigner, S. Schneiderbauer, C. Kloss, S. Pirker, Determining the coefficient of
[5] J.A. Hersey, Ordered mixing: a new concept in powder mixing practice, Powder friction by shear tester simulation, in: M. Bischoff, E. Oñate, D.R.J. Owen, E. Ramm,
Technol. 11 (1975) 41–44. P. Wriggers (Eds.),III International Conference on Particle-Based Methods — Funda-
[6] J. Bridgwater, Fundamental powder mixing mechanisms, Powder Technol. 15 mentals and Applications (PARTICLES 2013) Stuttgart, Germany, 2013.
(1976) 215–236. [31] A. Kwade, D. Schulze, J. Schwedes, Determination of the stress ratio in uniaxial
[7] M. Poux, P. Fayolle, F. Bertrand, D. Bridoux, J. Bousquet, Powder mixing: some compression tests — part 2, Powder Handl. Proc. 6 (1994) 199–203.
practical rules applied to agitated systems, Powder Technol. 68 (1991) 213–234. [32] T.A.H. Simons, M. Combarros-Garcia, P. Gupta, U. Tüzün, S. Zigan, J. Sun, M. Schilling,
[8] K. Sommer, Mixing of Solids, Ullmann's Encyclopedia of Industrial Chemistry, S. Bensmann, H. Zetzener, H.J. Feise, A. Kwade, F. Kleine-Jäger, J.Y. Ooi, Segregation
Wiley-VCH, Weinheim, Germany, 2005. and mixing of granular material in industrial processes, in: M. Bischoff, E. Oñate,
[9] C. Schofield, The definition and assessment of mixture quality in mixtures of partic- D.R.J. Owen, E. Ramm, P. Wriggers (Eds.),III International Conference on Particle-
ulate solids, Powder Technol. 15 (1976) 169–180. Based Methods — Fundamentals and Applications (PARTICLES 2013) Stuttgart,
[10] J. Raasch, K. Sommer, Anwendung von statistischen Prüfverfahren im Bereich der Germany, 2013.
Mischtechnik, Chem. Ing. Tech. 62 (1990) 17–22. [33] M. Stiess, Mechanische Verfahrenstechnik 1, 2nd ed. Springer, Berlin, Germany,
[11] F.J. Muzzio, P. Robinson, C. Wightman, D. Brone, Sampling practices in powder 1995 (Chapter 7).
blending, Int. J. Pharm. 155 (1997) 153–178. [34] K. Sommer, H. Rumpf, Varianz der stochastischen Homogenitaet bei
[12] P.A. Cundall, O.D.L. Strack, A discrete numerical model for granular assemblies, Koernermischungen und Suspensionen und praktische Ermittlung der Mischguete,
Géotechnique 29 (1979) 47–65. Chem. Ing. Tech. 46 (1974) 257.
[13] B. Remy, T.M. Canty, J.G. Khinast, B.J. Glasser, Experiments and simulations of [35] HBM Data sheet 'Datenblatt T5 Drehmomentmesswellen', Hottinger Baldwin
cohesionless particles with varying roughness in a bladed mixer, Chem. Eng. Sci. Messtechnik GmbH, Darmstadt, not dated.
65 (2010) 4557–4571. [36] B. Cooker, R.M. Nedderman, A theory of the mechanics of the helical ribbon powder
[14] S.L. Conway, A. Lekhal, J.G. Khinast, B.J. Glasser, Granular flow and segregation in a agitator, Powder Technol. 50 (1987) 1–13.
four-bladed mixer, Chem. Eng. Sci. 60 (2005) 7091–7107. [37] B. Cooker, R.M. Nedderman, Circulation and power consumption in helical ribbon
[15] R.L. Stewart, J. Bridgwater, D.J. Parker, Granular flow over a flat-bladed stirrer, Chem. powder agitators, Powder Technol. 52 (1987) 117–129.
Eng. Sci. 56 (2001) 4257–4271. [38] F. Vock, Zur Rührmechanik von FeststoffschüttungenPhD thesis Faculty of Chemical
[16] R.L. Stewart, J. Bridgwater, Y.C. Zhou, A.B. Yu, Simulated and measured flow of Engineering, University of Karlsruhe, Karlsruhe, 1975.
granules in a bladed mixer — a detailed comparison, Chem. Eng. Sci. 56 (2001)