You are on page 1of 14

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/226360215

Quality by Design Methodology for Development and Scale-up of Batch


Mixing Processes

Article  in  Journal of Pharmaceutical Innovation · December 2008


DOI: 10.1007/s12247-008-9048-9

CITATIONS READS

20 421

7 authors, including:

Marianthi Ierapetritou Maria Silvina Tomassone


Rutgers, The State University of New Jersey Rutgers, The State University of New Jersey
143 PUBLICATIONS   3,655 CITATIONS    107 PUBLICATIONS   1,247 CITATIONS   

SEE PROFILE SEE PROFILE

Donald James Clancy


GlaxoSmithKline
6 PUBLICATIONS   41 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Predictive modeling of pharmaceutical unit operations View project

PAT Real-Time Analysis for the flow synthesis of Oligonucleotides View project

All content following this page was uploaded by Maria Silvina Tomassone on 19 January 2016.

The user has requested enhancement of the downloaded file.


J Pharm Innov (2008) 3:258–270
DOI 10.1007/s12247-008-9048-9

RESEARCH ARTICLE

Quality by Design Methodology for Development


and Scale-up of Batch Mixing Processes
Patricia M. Portillo & Marianthi Ierapetritou &
Silvina Tomassone & Christine Mc Dade &
Donald Clancy & Petrus P. C. Avontuur &
Fernando J. Muzzio

Published online: 2 December 2008


# International Society for Pharmaceutical Engineering 2008

Abstract In this study, a quality by design approach was Introduction


applied to the design and scale-up of a batch mixing
process. Mixtures of acetaminophen and lactose were Powder mixing has been the subject of substantial research.
sampled at different mixing times using a groove sampler. This is motivated by applications in a variety of industrial
Samples were subsequently analyzed using NIR reflection sectors, which include pharmaceuticals, food, ceramics,
spectroscopy. The effects of four processing parameters on catalysts, metals, and polymer manufacturing. Understand-
the empirical mixing rate in a bin blender were examined. ing mixing mechanisms and identifying critical process and
Blender rotation rate (two levels), powder fill level (two material parameters is often a crucial step during process
levels), powder cohesion (two levels), and blender size development. Content uniformity problems have four main
(three levels) represent the four parameters studied. Blender root causes: (a) powder stream flow properties [1], (b) poor
geometry and blender loading method were treated as equipment design or inadequate operation [2], (c) particle
constant parameters. Statistical analysis was used to assess segregation due to differences in particle properties, and (d)
the impact each parameter had on the mixing rate. Blender particle agglomeration, driven by electrostatics, moisture,
size (p=0.02), powder cohesion (p=0.05), and rotation rate softening of low melting point components, as well as other
(p=0.07) all significantly affected the mixing rate. The least factors.
significant parameter was the vessel fill level (p=0.18), Scale-up of mixing operations continues to present a
indicating mixing performance is not strongly affected by concern to the pharmaceutical development process. The
fill level, given the range studied. reliable scaling of a process requires an understanding of
the effects that processing parameters may illicit on
Keywords Quality by design . Batch powder mixing . intermediate- and finished-product properties. Generally,
Bin blender . Statistical analysis . Mixing rate processing conditions are thoroughly examined at small
scales during process development of powder formulations.
The design and scale-up of blending operations is a
P. M. Portillo : M. Ierapetritou : S. Tomassone : F. J. Muzzio (*) multivariate issue; the relative magnitudes of shear,
Department of Chemical and Biochemical Engineering, dispersion, and convective forces may be altered as the
Rutgers University,
Piscataway, NJ 08854, USA process is transferred to larger scales [3].
e-mail: fjmuzzio@yahoo.com A problem with the current scale-up philosophy is a
failure in addressing several critical variables. Shear rate
C. Mc Dade : D. Clancy and total strain have been shown to affect blend micro-
Process Technologies, Glaxo Smith Kline Company,
709 Swedeland Road, structure [4], which may consequently affect the degree of
King of Prussia, PA 19406, USA ingredient agglomeration, blend flow properties, tablet
hardness and final product dissolution, which may ulti-
P. P. C. Avontuur mately result in failures during the scale-up process. An
GlaxoSmithKline R&D, New Frontiers Science Park,
Third Avenue, Harlow, example of this includes blend over-lubrication resulting
Essex CM19 5AW, UK from the increase in shear (per revolution of the blender)
J Pharm Innov (2008) 3:258–270 259

intensity as a function of increasing scale [5]. In a separate 3. Mixing is a function of shear rate for cohesive powders.
study, blender rotation rates were found to affect the relative
standard deviation (RSD) plateau of a given system [6]. In summary, the main variables known to affect
Powder cohesion properties also affect the velocity gradient, mixing performance are: (1) the design of the mixing
where interpaticulate forces dilate the powder bed density. This system (e.g., geometry and blend mechanism), (2)
may have further implications on downstream processing. blender size, (3) the fill level, (4) the blender loading
Optimization of the blending process requires an under- mode, (5) the speed of rotation of the blender, and (6)
standing of blending mechanisms and critical variables. the material properties of the ingredients being mixed
Although modifications to powder cohesion, blender size, (particle size, shape, and density, etc.).
and geometry may not be feasible due to other constraints, Historical practices in pharmaceutical process develop-
operating conditions such as rotation rate and fill level are ment have largely involved univariate (OVAT, “one variable
easier to alter. An understanding of the interactions among at a time”) approaches, where the effects of a single
these variables is essential. variable are examined for a few conditions selected based
V-blenders, tote blenders, and double-cone blenders are on prior experience from a “safe” subset of the permissible
examples of batch blenders that vary in geometric design. design space. A value of the first variable is then selected
For these systems, variables such as blender size and fill and kept constant as a second variable is examined, and so
level may affect mixing behavior [7–9]. Mixing in tumbling forth. However, as suggested in the Process Analytical
blenders is limited in the ability to improve upon Technology (PAT) Guidance [14], the OVAT approach does
component segregation, typically attributable to variations not effectively address the effect of interactions between
in particle characteristics (e.g., size and shape), once it multiple process variables. As a result, unless the effects of
occurs [10]. Further, initial load configuration (top/bottom all variables are nearly independent of one another, the
and left/right) of the active pharmaceutical ingredient (API) optimal conditions for operating the process will not be
and excipients has been shown to affect the mixing rate [11]. determined.
Brone and coworkers [12] examined the effect of The pharmaceutical industry has recently focused sub-
changing the rotation rate from 8 to 24 rpm for glass beads stantial effort on improving its understanding of key unit
in a V-blender. This study illustrated that the mixing operations and developing statistical, instrumental, and
mechanism of free flowing materials was not affected by fundamental methods for characterizing and controlling
rotation rate; however, the total mixing time was ultimately sources of variability in product performance. Beginning
reduced. Sudah and coworkers [13] varied the rotation rate with the introduction of the PAT Guidance in 2003, and
from 5 to 15 rpm for art sand in a rectangular tote blender. continuing with the quality by design (QbD) initiative (for
This study showed rotation rate did not affect mixing rate which draft guidance exists in ICH Q8 [15], Q9 [16], and
(per revolution) in earlier stages of the mixing process (up Q10 [17]), the industry is transitioning from OVAT
to 64 revolutions of the blender) but did affect the approaches to a more multivariate statistical method for
asymptotic variance plateau (total achievable homogeniza- assessing the effects of process variables on product quality.
tion). The studies also showed that for a cohesive blend, The “process understanding” that is considered a keystone
rotating the vessel at 10 rpm resulted in the smallest of the PAT and QbD initiatives enables the process control
asymptotic variance, which suggested the presence of that facilitates the manufacture of quality products meeting
competing mechanisms. Later on, Arratia and coworkers desired performance specifications..
[6] examined the effects of blender fill level (40–85%) on Clearly, application of QbD methods is not a temporally
mixing rate. The study showed a decrease in mixing rate discrete activity to be completed at an early stage of
with increasing fill level for Bohle-Bin blenders. product/process development. Rather, QbD is a longitudinal
Alexander and Muzzio [3] generalized some simple component of the product life cycle, which, early in the
guidelines to help the scale-up process. Some of these may development process serves as a risk assessment and
be applicable for the formulations used in the present case formulation screening methodology, later, as a product/
study. process optimization approach, and finally as a continuous
improvement method during commercial manufacturing.
1. When altering blender size, the dominant mixing While the philosophy of QbD methodologies is straightfor-
mechanism ratio (i.e., gravitational, shear, convective), ward, with the necessary toolbox having been well
order of material addition, and powder arrangement developed and long used in other industries, the actual
should be kept constant. implementation, however, represents a significant task due
2. The number of blender revolutions is a key parameter to the fact that (1) the mechanical and physicochemical
for free flowing powders. Rotation rates, however, are properties of many pharmaceutical APIs and excipients are
largely unimportant. only partially understood, thus limiting identification of
260 J Pharm Innov (2008) 3:258–270

Experimental Method

Blender Specifications

The blenders used in this study are shown in Fig. 1. They


have a fixed geometry comprised of a rectangular bin
blender and a pyramidal hopper that forms the tote. This
design was selected for several reasons, including: (1) it is
common, (2) it is available in our lab in three sizes, (3) it
has been thoroughly examined in previous studies, and (4)
a good working knowledge of the critical-to-performance
variables is available. The two larger blenders were
Fig. 1 The three tote blenders used in this study. From left to right:
10, 0.5, and 2 ft3 blender manufactured by the GEA group (Birmingham, UK) and
the smallest blender is a custom-made vessel run under a
step motor control. The rectangular bin blenders used in
this study have been previously employed in the assessment
critical material variables, and (2) there is an incomplete of blender performance [9, 13, 19]. A schematic of the
knowledge of critical process variables for many pharma- blender geometry is illustrated in Fig. 2. The corresponding
ceutical operations. dimensions for the three different blenders are shown in
As a result, prioritization (early screening) of factors Table 2. Baffles were used since they have been shown to
is difficult, and studies need to address the large improve mixing performance [20] by increasing axial flow.
comprehensive parametric space of all conceivably
relevant variables. Because of this incomplete knowledge Powder Materials
of what is truly critical, naive attempts at application of
QbD methodologies are likely to be suboptimal due to a A binary blend system was investigated. The two blends
poorly constructed design space. In this paper, we examined contained acetaminophen (Mallinkcrodt, milled
attempt to introduce some prioritization guidelines for to 30 μm) and one of two grades of lactose (De
batch blending in tumbling blenders, where a knowledge Melkindustrie Veghel): lactose 125 (55 μm) or lactose
base exists. The scale-up of blending and assurance of 100 (130 μm). The bulk densities of the three powders used
homogeneity may benefit substantially from the devel- were measured by taking the weight of a finite portion of
opment and incorporation of a realistic QbD strategy. powder and dividing it by the volume the powder sample
QbD has already begun to be capitalized where new occupied. The procedure was repeated ten times for each
methods for quantitatively assessing product quality are powder. Experimental densities for lactose 100M, Lactose
being developed. A risk-based strategy for quality 125M and acetaminophen are reported as 0.785±9.02E−03,
assessment which uses model-based simulation to link 0.765±1.33E−02, and 0.297±1.57E−02 g/ml.
variation in drug product parameters and clinical perfor-
mance has been recently presented [18]. Powder Loading
As mentioned, this paper outlines a QbD strategy for the
design of batch blending operations. In the next section, we The method by which materials are initially loaded into the
discuss the experimental method. This provides the scope of blender vessel is a parameter that has been shown to affect the
experimental conditions investigated and the details behind mixing performance of tumbling blenders. Two particular
the fixed parameters, such as blender specifications, loading loading patterns previously examined are the top-to-bottom
pattern, and materials. This section covers all necessary and left-to-right starting configurations. In the top-to-bottom
details concerning the experimental and analytical methods
used in obtaining the data. This includes the sampling
Fig. 2 Bin blender schematic D
method, analysis technique, and the quantification methods
L
used to measure mixing performance. The “Effects and
Axis of
Characterization of Cohesion” section describes cohesion, its rotation H1
effect on mixing, and the methods used to quantify different
levels of cohesion. The “Results” section is devoted to a
statistical analysis of the factorial batch-blending experi- θ
H2
ments. Finally, the “Outlook and Conclusions” section
provides closing comments and conclusions. V
J Pharm Innov (2008) 3:258–270 261

Table 1 Experimental conditions examined nondimensional parameter comparing inertial and gravita-
Tote size Fill Speed (rpm) Cohesion Fr no. tional forces. In theory, if you keep this number the same
(ft3) level for the varying blender sizes, the powder blends within the
mixers are maintaining a constant ratio of inertial and
0.5 50 Low (10) Lactose 100 4.11E−04 gravitational forces [3]. No first principle approaches
0.5 70 Low (10) Lactose 100 4.11E−04
currently exist for the design of a dry blend process scale-
0.5 50 Low (10) Lactose 125 4.11E−04
up, except for the application of the Froude number. The
0.5 70 Low (10) Lactose 125 4.11E−04
2 50 Low (10) Lactose 100 4.29E−04 Froude number can be justified from a generalized
2 70 Low (10) Lactose 100 4.29E−04 momentum balance under fairly drastic assumptions. The
2 50 Low (10) Lactose 125 4.29E−04 Froude number for tumbler mixers is given by:
2 70 Low (10) Lactose 125 4.29E−04
10 50 Low (8) Lactose 100 4.22E−04
Ω2 R
10 70 Low (8) Lactose 100 4.22E−04 Fr ¼ ð1Þ
10 50 Low (8) Lactose 125 4.22E−04 g
10 70 Low (8) Lactose 125 4.22E−04
0.5 50 High (14) Lactose 100 8.06E−04
0.5 70 High (14) Lactose 100 8.06E−04 where Ω is the rotation rate (revolutions per minute), R is
0.5 50 High (14) Lactose 125 8.06E−04 the radius (mm), and g is the acceleration from gravity
0.5 70 High (14) Lactose 125 8.06E−04 (millimeter per square minute). Conversely, semi-empiri-
2 50 High (14) Lactose 100 8.42E−04 cal approaches are based on the Rayleigh method to
2 70 High (14) Lactose 100 8.42E−04 develop similarity criteria. This yields a resulting equation
2 50 High (14) Lactose 125 8.42E−04 that is a function of particle velocity, vessel rotation rate,
2 70 High (14) Lactose 125 8.42E−04
radius, particle diameter, and gravitational acceleration
10 50 High (11) Lactose 100 7.97E−04
10 70 High (11) Lactose 100 7.97E−04
[3].
10 50 High (11) Lactose 125 7.97E−04
10 70 High (11) Lactose 125 7.97E−04 Sampling Method

For each treatment condition, powder samples were taken


after 5, 25, 50, and 100 revolutions using a groove sampler.
configuration, one powder is initially loaded into the vessel, The groove sampler consists of a hollow sleeve (1 in. in
above this powder bed another powder is layered. In the left- diameter) surrounding a solid inner steel rod possessing a
to-right configuration, a separator is initially inserted into the groove along most of the length of the rod. The inner pipe
vessel where one side of the mixer is loaded with one powder has a sampling cavity that is 1/2 in. deep and 1/2 in. wide
and the other side with another. In previous experimental along the middle 80% of the rod. Rotating the inner pipe
studies [11], the axial variance, defined as the variability relative to the outer pipe opens and closes the groove
within the axial length of the vessel (axial length is sampler. The sampler (in its open position) was inserted
perpendicular to the axis of rotation) of a powder blend, into the powder bed and subsequently rotated to trap
was examined in a 56-L tote blender for both loading material within the sampling cavity. After being removed
patterns. The top-to-bottom loading resulted in significantly from the powder bin, the sampler was then placed
faster mixing rates than left-to-right loading. Thus, the top- horizontally on a stand while open, and the entire device
to-bottom loading was used for this study. was rotated to discharge the collected material into a series
of small trays. Sample size can vary with size or width of
Experimental Design the sampler [10, 18].
Sample size and sampling location were constant for all
A randomized full factorial design was used to determine experiments. The axial sampling locations lie on the axis of
the effect of each parameter and their interactions on the rotation, and the vertical sampling locations lie perpendic-
rate of the mixing process. The full list of experimental ular to the axis of rotation. Samples were retrieved from
conditions from the combined variable levels is shown in five radial positions for all three mixing vessels. Ten axial
Table 1. Experiments are conducted for three tote sizes (0.5, positions were examined for the two larger mixers (10 and
2, 10 ft3) and two volumetric fill levels (50% and 70%). 2 ft3) and five axial positions were examined for the smaller
Each tumbler gyrates at two speeds (low and high) as (0.5 ft3) mixer due to the smaller axial bed. The analysis
shown in Table 1. The speeds are dependent on tumbler and limitations of accuracy and precision for the sampling
size and are adjusted to keep the Froude number constant, device used here have been discussed in detail by Muzzio
as illustrated in Table 1. The Froude number is a and coworkers [19, 21].
262 J Pharm Innov (2008) 3:258–270

NIR Analysis on Homogeneity Effects samples. The results indicated no real advantage from
agitating and scanning the sample multiple times, for the
Near-infrared spectroscopy (NIR) was used in this study system under study. Further, some potential segregation was
due to its rapid real-time analysis, high-throughput capa- detected. Therefore, samples were assayed a single time.
bility, and nondestructive nature. NIR technique details may
be found elsewhere [6, 20, 22]. Following extraction, Quantification of Mixing Performance
samples were collected into glass vials (Fisher, Pittsburgh,
PA, USA) and analyzed using single-point near-infrared Homogeneity Measurements
spectroscopy (FOSS NIR Systems, Silver Spring, MD,
USA). The composition of each sample was examined In order to determine powder homogeneity, the sample-to-
using a Rapid Content Analyzer (NIR 5082) manufactured sample variability was quantified in terms of the RSD. The
by FOSS NIR Systems, Inc. Empirical modeling of the full standard definition of RSD (also known as coefficient of
NIR (1,100–2,500 nm) portion of the spectra (second variance (CoV)) is given by:
derivative pretreatment) was performed using partial least
squares regression. The amounts of acetaminophen and s
RSD ¼ CoV ¼ ð2Þ
lactose varied in the calibration and validation samples. x
Samples ranged from 0% to 5% w/w acetaminophen (0.1%
increments) and 5–10% acetaminophen (1% increments). where s represents the sampling estimate of the standard
Each sample was scanned 32 times without repositioning deviation and x the average of all the samples. As
the vial to obtain an NIR spectrum. The performance of the mentioned, samples were retrieved from the vessel using a
calibration model was validated using 2-g samples with sampler and subsequently measured via NIR. For each
known acetaminophen concentrations. The NIR perfor- radial core position (denoted as j), xij is a sample
mance method was assessed by the root mean square error concentration, xj is the mean concentration, and Nj is the
of calibration (RMSEC) and the root mean square error of number of samples in that core [23, 24]. The standard
prediction (RMSEP). RMSEC and RMSEP values obtained definition of variance (s2) is given by Eq. 3:
in this study were 0.997 and 0.976, respectively. The
resulting model was use to predict the concentration of  
X X xij  x 2
acetaminophen in each sample obtained from the groove s ¼
2
ð3Þ
sampler. Powder samples were collected in the same glass j i
Ni
vials (28 mm diameter, 61 mm long, 20 mL volume capacity)
used to analyze the calibration and validation samples. P
Sampling volume and size may substantially affect the i xij
xj ¼ ð4Þ
estimate of variability between sample concentrations. Due Ni
to the increase in scale of scrutiny, larger samples exhibit
smaller variability between acetaminophen concentrations. where N is the number of samples and x is the mean
When the analytical method interrogates the same surface composition found using Eq. 4. The total variance is
area of the powder, however, the measurement sample size decomposed into two components for axial and radial
is constant. The depth of powder in the vial should exceed variability (Eq. 5):
4 mm for reflectance measurements to ensure the effective
beam penetration depth does not exceed the powder depth 1 X  2 1 X X  2
bed. Reflectance spectroscopy may assay a small fraction of s2 ¼ Nj xj  x þ xij  xj ð5Þ
N j N j i
the entire sample, making it relatively more vulnerable to
inhomogeneities within the sample. To mitigate such
problems, multiple cycles of sample agitation followed by The first term is an estimate of axial variance (s2A ) and
sample analysis may be employed; however, this technique the second term, radial variance (s2R ) [11]. Axial variance
may result in sample segregation. When the particle size measures the differences in concentrations between the top
distribution is broad, larger particles will segregate by and bottom of the powder bed.
levitating to the top of the powder bed. To check for such Statistically, if random samples are taken from a
effects, we evaluated the effect of agitation and repeated mixture of average composition q, given the fraction of
assaying. Figure 2 illustrates the RSD profile of all 50 the first component is P and second component is (1−P),
samples (ten samples along the axial bed at 5 radial and the mixture has a random structure, the composition
positions) scanned a single time. The second RSD profile of the samples will be normally distributed. The theoret-
resulted from agitating and subsequently rescanning all the ical variance can be calculated for completely random
J Pharm Innov (2008) 3:258–270 263

mixtures using Eq. 6 and for nonrandom mixtures using Effects and Characterization of Cohesion
Eq. 7:
Cohesion
 
P
s ¼P 1
2
ð6Þ Powder cohesion is one of the most important properties
N
affecting flow, and by extension, powder mixing. Cohesion
is caused by Van der Waals, capillary, and electrostatic
 
ð P ð 1  P Þ  LÞ 2 attraction forces. Cohesion of a bulk solid is defined as its
s2 ¼ L þ L ð7Þ resistance to shear (shear strength) under null normal stress
N
on the plane of failure [24, 27]. Cohesion of powders,
although poorly understood is known to depend on powder
where L represents a constant for a given mixture or state of
moisture content, presence/absence of glidants, and particle
mixedness and may be determined experimentally when the
size. As the size of a particle decreases, the ratio of its
value of σ is known for a given value of N. For a system
surface area to mass increases. A cohesive powder can be
where the two components are completely unmixed, the
defined as a material where the interparticle adhesive forces
initial variance (s 20 ) of the sample composition may be
exceed the particle weight by at least an order of
calculated using Eq. 8 [21, 25]:
magnitude. In such systems, particles no longer flow
independently; rather, they move in “chunks” where the
s 20 ¼ Pð1  PÞ ð8Þ size is dependent upon the intensity of the cohesive stresses
[25, 28]. The effective magnitude of cohesive effects
However if Eq. 8 was followed for a mass ratio of 97% depends primarily on two factors: (1) the intensity and
lactose and 3% acetaminophen, the initial RSD would nature of the cohesive forces and (2) the number of
be ∼3%. This value should represent the largest obtainable interparticle contacts per unit area (packing density). The
variance for this system. The initial RSD (RSD0) for these effect of cohesion on mixing behavior is not always
experiments was calculated from a statistical approach obvious. Slightly cohesive powders have been observed to
published elsewhere [22, 26]. The RSD0 is 2.45% for the mix faster than free flowing materials. Strongly cohesive
larger and medium tote (50 samples per time point) and powders, however, blend much more slowly and often
2.10% for the smaller tote (25 samples per time point). The require externally applied shear, which may be supplied by
discrepancies in RSD values may be due to the differences an impeller [3]. Chaudhuri and coworkers [26, 29] showed
in the number of samples retrieved relative to the total that material possessing a low intensity of cohesion
powder mass in each vessel. (defined as a bond number of 0.1) enhanced mixing relative
to no cohesion (bond number equal to 0) in a rotating drum.
Conversely, high values of cohesion (bond numbers greater
Mixing Rate
than 60) reduced the mixing rate. Slight cohesion was
shown to improve mixing rate in a rotating drum blending
Process performance monitoring for each parameter com-
process [27, 30]. In a separate report, slight cohesion may
bination was evaluated by the mixing rate of acetamino-
also result in agglomerates [28, 29, 31, 32] that ultimately
phen in lactose. The mixing rate was computed by
result in segregation due to the variability in particle
retrieving powder samples from the blender as a function
diffusion in the vessel [26, 29]. Depending on the blending
of revolutions. Sample variance and relative standard
process, size, and design conditions of the equipment, the
deviation as a function of vessel revolutions were deter-
effect of cohesion varies as shown by several references in
mined from NIR-predicted concentrations. In the absence
the “Introduction” section. The following sections discuss
of segregation, the variance in a blender typically decays to
two methods for quantifying cohesion.
its asymptotic value as an exponential of time. Therefore,
mixing rate is measured as the slope of the logarithm of the
Hausner Ratio
variance. The slope, m, is determined from Eq. 9:
P     Interparticle surface forces such as friction and cohesion are
revolutionsi  revolutions log s2i  logðs2 Þ dependent on the total surface area. Since mass is
m¼ P 2
revolutionsi  revolutions proportional to the volume, the surface area to volume
ratio is a good general indication of the “flowability” of a
ð9Þ
powder system [30, 33]. The Hall flowmeter and Hausner
where revolutions is the revolution average and logðs2 Þ is ratio are two common techniques for analyzing the effect of
the logarithmic variance average. interparticle forces on the flow behavior of powder systems
264 J Pharm Innov (2008) 3:258–270

Table 2 Bin blender and baffle dimensions

Mixer (ft3) D (mm) H1 (mm) H2 (mm) L (mm) V (mm) θ Baffle width (mm) Baffle height (mm)

0.5 279.4 153 150 228.6 88.9 35 31.75 50.8


2 457.2 318.8 240 381 152.4 35 50.8 82.55
10 762 516 500 609.6 203.2 35 152.4 266.7

under the influence of gravity. The Hausner ratio is the ratio of Results from this method were used to select grades of
the tapped density to the apparent (poured) density of the lactose with different cohesiveness. The measured flow
powder. The apparent density tends to decrease as the indices of lactose 100M and lactose 125M were 17.3 and
interparticle friction in a powder system increases. The tapped 20.26, respectively.
density tends to decrease as well, albeit a lesser extent due to
the additional energy imparted from tapping. The cohesive
behavior of a powder is a qualitative description of how Results
powder moves. Yield strength increases with powder cohesion
because a larger stress is required to deform the powder In this section, we discuss the effects of four parameters
[25, 28]. A cohesive powder will have a higher Hausner ratio (tote size, cohesion, rotation rate, and fill level) on the
relative to one that is free flowing. The US Pharmacopeia mixing rate of typical pharmaceutical materials in a tote
[34] defines ranges of the Hausner ratio which describe blender. The experiments were performed using acetamin-
powder flowability. Hausner ratio values between 1 and 1.11 ophen, which is typically regarded as a difficult-to-mix
are considered to reflect excellent flow properties. Values API. Two lactose grades were used to simulate formulations
greater than 1.6 typically suggest very poor flow, a character- of varying cohesion (based on the Hausner ratio). The main
istic of cohesive powders. The larger the surface area to effects and interactions were examined. Statistical analysis
volume ratio, the greater the probability a particle will cling to was used to: (1) identify the critical parameters and (2)
another. DMV International reports Hausner ratios of 1.21 and develop proper analytical methodology. The variance at
1.28 for lactose 100M and lactose 125M, respectively. each experimental condition is listed in Table 2 for the 0.5-,
2-, and 10-ft3 blenders. The remainder of this section
Gravitational Displacement Rheometer presents the effects of the four experimental parameters on
mixing rate.
Many studies measure the magnitude of cohesive forces
indirectly using “flow testers.” A common approach is to Statistical Analysis Methodology
use a “shear cell” to measure cohesion in the incipient
failure regime [24, 27]. A different approach to character- Analysis of variance (ANOVA) is a mathematical proce-
izing cohesion of powder in the dilated state was adopted dure for partitioning the variability of a data set into
here. A method using a gravitational displacement rheom- components associated with different main and interaction
eter (GDR) to characterize powder flowability, originally effects. The information provided by ANOVA is used to
developed by Faqih and coworkers [32, 35], is employed in construct statistical tests to determine the statistical signif-
this study. This device produces a “flow index” that is icance of main effects and their interactions. An F-statistic
proportional to the yield strength of the dilated powder. is computed for each effect (including main and inter-
Higher flow index values indicate a more cohesive powder actions), which is used to test hypotheses about the
and consequently decreased flowability. existence of the effects of variables [33, 36].
In the GDR, a cylinder containing the powder sample is In this work, a fully randomized factorial design of the four
mounted on a pivoted table supported by a load cell. As the processing parameters (n=1) was examined. All four
cylinder rotates, the powder avalanches down the cascading variables were treated as fixed-level variables (tote size 0.5,
surface, thereby resulting in a change to the center of mass. 2, and 10 ft3; speed: H, L; cohesion: lactose 100, lactose 125;
This shift is measured by the force registered by the load fill level 50% and 70%) resulting in a total of 24 treatment
cell. The size and frequency of avalanches are analyzed. As conditions (Table 3). Data were analyzed under the usual
the cohesive nature of the material increases, the resulting assumptions of normality and independence. Given the large
avalanches become larger and the frequency of avalanche number of variables, which prevented replications, it was
tumbling decreases. The RSD of the force signal is necessary to assume that higher order interactions did not
calculated at several rotation rates and averaged to give an exist in order to release degrees of freedom to estimate an
accurate characterization of avalanche size. error term. Two models were examined: (1) a model where
J Pharm Innov (2008) 3:258–270 265

Table 3 Variance results for the three tote blenders 0.5, 2, and 10 ft3

Tote 0.5 ft3 Tote 2 ft3 Tote 10 ft3

Rev Var Log(var) Rev Var Log(var) Rev Var Log(var)

Details: low rpm, 50% fill, lactose 100


5 1.353 0.131 5 0.468 −0.330 5 0.187 −0.727
25 0.015 −1.825 25 0.159 −0.800 25 0.048 −1.322
50 0.014 −1.866 50 0.032 −1.490 50 0.020 −1.692
100 0.011 −1.948 100 0.011 −1.976 100 0.012 −1.932
Slope= Mixing rate= −0.017 Slope= Mixing rate= −0.017 Slope= Mixing rate= −0.012
Details: low rpm, 70% fill, lactose 100
5 1.210 0.083 5 0.652 −0.186 5 0.124 −0.906
25 0.343 −0.464 25 0.286 −0.544 25 0.064 −1.192
50 0.172 −0.765 50 0.038 −1.425 50 0.024 −1.621
100 0.076 −1.120 100 0.015 −1.816 100 0.016 −1.789
Slope= Mixing rate= −0.012 Slope= Mixing rate= −0.018 Slope= Mixing rate= −0.009
Details: low rpm, 50% fill, lactose 125
5 1.011 0.005 5 0.065 −1.190 5 0.067 −1.175
25 0.112 −0.952 25 0.016 −1.792 25 0.050 −1.303
50 0.042 −1.377 50 0.009 −2.067 50 0.023 −1.646
100 0.013 −1.881 100 0.005 −2.302 100 0.021 −1.670
Slope= Mixing rate= −0.018 Slope= Mixing rate= −0.011 Slope= Mixing rate= −0.005
Details: low rpm, 70% fill, lactose 125
5 1.472 0.168 5 0.520 −0.284 5 0.138 −0.859
25 0.123 −0.910 25 0.053 −1.275 25 0.049 −1.310
50 0.072 −1.144 50 0.059 −1.232 50 0.025 −1.610
100 0.014 −1.867 100 0.025 −1.609 100 0.023 −1.644
Slope= Mixing rate= −0.019 Slope= Mixing rate= −0.011 Slope= Mixing rate= −0.008
Details: high rpm, 50% fill, lactose 100
5 0.912 −0.040 5 0.173 −0.762 5 0.114 −0.942
25 0.194 −0.712 25 0.035 −1.454 25 0.029 −1.545
50 0.037 −1.434 50 0.002 −2.786 50 0.024 −1.626
100 0.005 −2.314 100 0.002 −2.754 100 0.004 −2.433
Slope= Mixing rate= −0.023 Slope= Mixing rate= −0.021 Slope= Mixing rate= −0.015
Details: high rpm, 70% fill, lactose 100
5 1.668 0.222 5 0.439 −0.357 5 0.177 −0.752
25 0.103 −0.988 25 0.091 −1.039 25 0.018 −1.748
50 0.020 −1.705 50 0.019 −1.728 50 0.015 −1.825
100 0.009 −2.024 100 0.009 −2.047 100 0.015 −1.832
Slope= Mixing rate= −0.022 Slope= Mixing rate= −0.017 Slope= Mixing rate= −0.009
Details: high rpm, 50% fill, lactose 125
5 0.275 −0.561 5 0.226 −0.645 5 0.156 −0.807
25 0.045 −1.349 25 0.016 −1.783 25 0.048 −1.320
50 0.035 −1.456 50 0.007 −2.156 50 0.035 −1.453
100 0.002 −2.737 100 0.006 −2.245 100 0.023 −1.629
Slope= Mixing rate= −0.021 Slope= Mixing rate= −0.014 Slope= Mixing rate= −0.008
Details: high rpm, 70% fill, lactose 125
5 1.117 0.048 5 0.260 −0.584 5 0.184 −0.734
25 0.105 −0.977 25 0.083 −1.081 25 0.029 −1.534
50 0.023 −1.634 50 0.039 −1.406 50 0.015 −1.832
100 0.013 −1.890 100 0.032 −1.498 100 0.012 −1.914
Slope= Mixing rate= −0.019 Slope= Mixing rate= −0.009 Slope= Mixing rate= −0.011

all two-way and three-way interactions are considered where where 9 degrees of freedom remain for the error term. The
2 degrees of freedom remain for the error term (from results are summarized in Tables 4 and 5.
assuming a four-way interaction did not exist), and (2) a The sums of squares, SS, were calculated using SAS
model where only the two-way interactions are considered version 9.1 (SAS Institute Inc., Cary NC, USA). The p
266 J Pharm Innov (2008) 3:258–270

Table 4 Four-way ANOVA

Source DF SS MS F p

Rotation rate 1 4.27E−05 4.27E−05 12.97 0.07


Fill level 1 1.35E−05 1.35E−05 4.10 0.18
Cohesion 1 6.02E−05 6.02E−05 18.29 0.05
Tote 2 3.44E−04 1.72E−04 52.22 0.02
Fill × tote size 2 1.75E−06 8.75E−07 0.27 0.79
Fill × cohesion 1 1.35E−05 1.35E−05 4.10 0.18
Fill × rotation rate 1 6.00E−06 6.00E−06 1.82 0.31
Rotation rate × tote 2 1.46E−05 7.29E−06 2.22 0.31
Rotation rate × cohesion 1 6.00E−06 6.00E−06 1.82 0.31
Tote size × cohesion 2 6.01E−05 3.00E−05 9.13 0.10
Fill × tote × rotation 2 7.75E−06 3.88E−06 1.18 0.46
Fill × cohesion × rotation 1 6.70E−07 6.70E−07 0.20 0.70
Fill × cohesion × tote 2 1.83E−05 9.13E−06 2.77 0.27
Tote × cohesion × rotation 2 1.68E−05 8.38E−06 2.55 0.28
Error 2 6.58E−06 3.29E−06

value was calculated using Microsoft® Excel’s pdist term, again, was estimated through degrees of freedom
function. A value of p≤0.05 was selected as the threshold obtained by neglecting three-way interactions.
for considering an effect to be significant. Further consolidation of the model does not change the
The first model (Table 4) indicates only tote size to be significance of factors mentioned above. A model (not shown)
significant. The calculated p values for the first model are as where only the tote size × cohesion and the fill level × cohesion
follows: tote size (p=0.02), cohesion (p=0.05), rotation rate interactions were retained gave nearly identical results. The
(p=0.07), and fill level (p=0.18). The error estimate, once same three main effects, again, were significant (tote size (p<
again, was obtained from neglecting a four-way interaction. 0.0001), speed (p=0.016), cohesion (p=0.006)), while fill
The results obtained from the second model are given in level was not significant (p=0.15). The tote size × cohesion
Table 5. This model assumed all three-way interactions to interaction was significant (p=0.02) and fill level × cohesion
be negligible, a reasonable assumption considering the interaction was not significant (p=0.15). It must be stressed
range of p values (0.27–0.70). The second model, therefore, that failure to prove significance for a given set of experi-
focuses on the contribution of the main and two-way ments is not proof of statistical nonsignificance, especially
effects. This model resulted in p values lower than 0.05 for when a borderline p value is obtained.
three main effects (tote size, speed, and cohesion), while fill
level was not significant. Only one of the two-way Effect of Tote Size
interactions (tote size × cohesion) resulted in a low p
value. This is consistent with the fact that both tote size and The effects of tote size have been examined previously
cohesion were the most significant main effects. The error [13], where it has been shown that the magnitude of the

Table 5 Four-way ANOVA with no three-way interactions

Source DF SS MS F p

Rotation rate 1 4.27E−05 4.27E−05 7.68 0.02


Fill level 1 1.35E−05 1.35E−05 2.43 0.15
Cohesion 1 6.02E−05 6.02E−05 10.83 0.01
Tote 2 3.44E−04 1.72E−04 30.92 0.00
Fill × tote size 2 1.75E−06 8.75E−07 0.16 0.86
Fill × cohesion 1 1.35E−05 1.35E−05 2.43 0.15
Fill × rotation rate 1 6.00E−06 6.00E−06 1.08 0.33
Rotation rate × tote 2 1.46E−05 7.29E−06 1.31 0.32
Rotation rate × cohesion 1 6.00E−06 6.00E−06 1.08 0.33
Tote size × cohesion 2 6.01E−05 3.00E−05 5.41 0.03
Error 9 5.00E−05 5.56E−06
J Pharm Innov (2008) 3:258–270 267

1.4
compressibility, dissolution, chemical and physical compat-
1st RSD 2nd RSD
1.2 ibility). Further, the “materials properties” subdomain is
1.0 limited to selecting different grades (e.g., particle size),
0.8 brands, and concentrations of a given excipient. Consistent
RSD

with this “real world” situation, the effect of powder


0.6
properties was examined by selecting two lactose grades,
0.4
differing both in cohesion and particle size. Only one type
0.2 of acetaminophen was used in the study.
0.0 The increase in cohesion reduced the mixing rate for the
0 20 40 60 80 100 120 medium tote size blender (Fig. 4) and large tote size blender
Revolutions (not shown). The less cohesive lactose gave a lower final
Fig. 3 The results of a blending experiment with acetaminophen and blend variance and higher mixing rates. The low p value
lactose (highly cohesive) in the 2-ft3 mixer filled to 60% capacity. The (p=0.01) indicates a significant effect. Additionally, the
blending speed was set to 12 rpm most significant interaction was tote size × cohesion (p=
0.03) indicating tote size dependence of the cohesion effect
scale-up effect on mixing performance is dependent upon in mixing performance evaluation.
blend flow properties. A number of process problems are
caused by powder cohesion. This is illustrated by interpar- Effect of Rotation Rate
ticle forces resulting in API agglomeration.
Figure 3 illustrates the effect of tote size on mixing rates. Previous work by Sudah et al. [13] has shown that for free
As expected, the smaller tote had the fastest mixing rate, flowing materials, the mixing rate in bin blenders was
followed by the medium tote and large tote, respectively. largely independent of the blender rotational speed.
On average, the smallest tote had a mixing rate that was ∼2 However, different behaviors were observed for cohesive
times greater than the largest tote and 1.32 times greater materials. The effect of tumbler rotational speed on the
than the medium-sized tote. The ANOVA indicated that this homogeneity of cohesive mixture was investigated compu-
parameter had the lowest p value (p<0.0001). In the current tationally by Chaudhuri and coworkers [26, 29]. Simula-
study, therefore, tote size had the most statistically tions were employed where a drum was rotated at three
significant effect on mixing performance. different speeds. These simulations showed that the
homogeneity of cohesive mixtures could be affected by
Effect of Cohesion blender rotational speed. Mixing performance was im-
proved at higher rotational speeds and optimal performance
As mentioned above, material properties may also affect was observed for an intermediate cohesion level.
mixing performance. This is a very wide subdomain of the In this study, the majority of the experiments showed an
parametric space, potentially encompassing multivariate improvement in mixing rate when the rotation rate was
ranges of particle size, morphology, and cohesion. In real increased. The effect of rotation rate on mixing rate in the
applications, however, ingredients are typically selected in 0.5-ft3 tote size is shown in Fig. 5. By increasing the
response to multiple performance requirements (e.g., rotation rate from 10 to 14 rpm, a well-mixed state was
reached in fewer total revolutions. Changing the rotation
0.025 rate from 8 to 11 rpm for the 10-ft3 mixer improved the
.5 cu. ft 2 cu. ft 10 cu. ft
0.020
0.025
-Mixing Rate

Mid Cohesion
0.015
0.02 High Cohesion
-Mixing Rate

0.010
0.015

0.005
0.01

0.000 0.005
Lactose Lactose Lactose Lactose Lactose Lactose Lactose Lactose
100 100 125 125 100 100 125 125
0
Fill 50 % Fill 70 % Fill 50 % Fill 70 % Fill 50 % Fill 70 % Fill 50 % Fill 70 %
Fill 50 % Fill 50 % Fill 70 % Fill 70 %
Low Low Low Low High High High High
Speed Speed Speed Speed Speed Speed Speed Speed Low Speed High Speed Low Speed High Speed

Fig. 4 Mixing rate [(APAP%)2/rev] as a function of three varying tote Fig. 5 Mixing rate [(APAP%)2/rev] as a function of cohesion for the
sizes for all eight experiments 2-ft3 mixer
268 J Pharm Innov (2008) 3:258–270

0.025
Low Speed High Speed fill level for the 2-ft3 tote size exhibited faster mixing rates
0.02 when conducted at a high rotation rate. The largest tote size
mixer showed a faster mixing rate at 50% fill levels when
-Mixing Rate

0.015
using lactose 100 (Fig. 7). These observations indicate the
0.01 lactose/acetaminophen formulations used here are not
necessarily affected adversely at high fill levels. Given the
0.005
various findings for different size tote blenders, the effect of
0 fill level on mixing performance generally remained small
Lactose 100 Lactose 100 Lactose 125 Lactose 125 for the range examined. Supporting this conclusion,
Fill 50 % Fill 70 % Fill 50 % Fill 70 % statistical analysis indicated that fill level was the least
significant parameter affecting mixing performance (p=
Fig. 6 Mixing rate [(APAP%)2/rev] as a function of rotation rate on
0.15).
mixing rate for the 0.5-ft3 tote mixer. For the 0.5-ft3 mixer, 10 rpm is
considered the low speed and high speed is 14 rpm

Outlook and Conclusions


mixing performance for all conditions, with the exception of
the experiment conducted at a 70% fill level using lactose This study presented an example of the application of
100, where the mixing performance remained similar. statistical methods to examine the effects of multiple
Changing the revolutions per minute from 10 to 14 rpm for variables on mixing performance. Further, this study
the 2-ft3 mixer improved the mixing performance at the enabled the prioritization of the impact of variables using
50% fill level for both lactose grades. Moreover, the the following order of significance: tote size (p<0.0001),
statistical analysis showed that the rotation rate affected cohesion (p=0.01), rotation rate (p=0.02), and fill level
the mixing performance with a significant p value of 0.02. (p=0.15). Additionally, the results indicated main variables,
such as tote size, inherently interact. While this may explain
Fill Level the obscure nature of blending process scale-up, it also
suggests that ANOVA methods may be used to unravel
From a production standpoint, finding an optimum fill level some of its complexities. Unfortunately, some of the
is important given the desire to both maximize batch size interactions are difficult to examine in practice. Given the
and avoid problematic processing conditions. Early studies typical small number of experiments that can be conducted
using free flowing materials showed extreme variations in in a blending study, the risk of making erroneous
fill level (20% to 85%) and substantially affect mixing rates assumptions concerning large-scale batch behavior based
in tumbling blenders, where higher fill levels correlate to on small-scale observations is high. For example, small-
slower mixing [6]. Sudah and coworkers [9] found that scale experiments may indicate fill level to have a
optimum mixing rates were obtained at 60% fill, where the significant effect on the mixing performance; however, this
slowest mixing rate was observed at 80% fill over a 20– may not be the case when the same experiments are
80% range. In addition to a diminishing mixing rate, Arratia analyzed at a larger scale. This, in fact, is caused by the
and coworkers [6] measured concentration profiles for high interaction between scale and cohesion.
fill levels (greater than 85%) at low rotational speeds and If the study described above were conducted as part of a
showed segregated regions present in the center of the pharmaceutical product development campaign, the next
blender. Alexander and coworkers [10, 18], however, found
no significant difference in mixing rates for a sand mixture 0.025
employing a more practical range (40–60% fill), thereby 50% Fill
suggesting process flexibility within robust operation. 0.02 70% Fill
-Mixing Rate

Moreover, using fill levels in this range was shown to 0.015


reduce the probability of dead zones formation [3].
However, the bulk of this work was performed on free 0.01

flowing powders. For the slightly cohesive systems studied 0.005


in this paper, the effect of fill level is largely unknown.
Building on the previous referenced tote blender work, the 0
Lactose 100 Lactose 125 Lactose 100 Lactose 125
effects of fill level on mixing performance at 50% and 70%
Low Speed Low Speed High Speed High Speed
of total blender volume were examined.
Fill levels of 50% in the small tote size lead to higher Fig. 7 Mixing rate [(APAP%)2/rev] as a function of fill level for
mixing rates relative to the 70% fill level (Fig. 6). The 50% experiments conducted in the 0.5-ft3 tote mixer
J Pharm Innov (2008) 3:258–270 269

logical step would be to design a follow-up study for 3. Alexander AW, Muzzio FJ. Batch size increase in dry blending
and mixing. In: Levin M, editor. Pharmaceutical process scale-up.
concentrating on significant variables to explore, in more
New York: Marcel Dekker; 2001.
detail, their effects and interactions. Further, this would 4. Hassanpour A, Antony SJ, Ghadiria M. Effect of size ratio on the
enable determination of the range of variable values over behaviour of agglomerates embedded in a bed of particles
which the process could be operated with satisfactory subjected to shearing: DEM analysis. Chem Eng Sci 2007;62
(4):935–42.
results. An iterative approach, where the findings from each
5. Llusa M, Muzzio F. The effect of shear mixing on the blending
study are carried into the next study design, is clearly of cohesive lubricants and drugs, Pharmaceutical Technology,
needed. http://pharmtech.findpharma.com/pharmtech/Mixing+and+
Several additional comments deserve attention: blending/The-Effect-of-Shear-Mixing-on-the-Blending-of-Cohe/
ArticleStandard/Article/detail/283906. Accessed 1 Oct 2008;
– Given that blender size likely interacts with other 2005.
variables, the results obtained at smaller scales (a 6. Arratia PE, Duong N-H, Muzzio FJ, Godbole P, Lange A,
Reynolds S. Characterizing mixing and lubrication in the Bohle
common practice in many industries) should be Bin blender. Powder Technol 2006;161:202–8.
evaluated very carefully. Scale-up studies going beyond 7. Alexander A, Shinbrot T, Muzzio FJ. Granular segregation in the
mere “performance equivalence” and aimed at devel- double-cone blender: transitions and mechanisms. Phys Fluids
opment of effective scale-up correlations are critical. 2001;13(3):579–87.
8. Alexander A, Shinbrot T, Muzzio FJ. Segregation patterns in V-
Material properties of powders clearly play an impor- blenders. Chem Eng Sci 2003;58:487–96.
tant, scale-dependent role in blending performance. 9. Sudah OS, Coffin-Beach D, Muzzio FJ. Quantitative character-
This study merely touched the surface of assessing ization of mixing of free-flowing granular material in tote (bin)-
these phenomena. Going forward, the pharmaceutical blenders. Powder Technol 2002a;126:191–200.
10. Alexander A, Arratia P, Goodridge C, Sudah O, Brone D, Muzzio
industry should continue to enhance characterization of F. Characterization of the performance of bin blenders. Part 3:
relevant material properties. cohesive powders. Pharm Tech 2004;28:54–74.
– As previously addressed, materials and processes both 11. Muzzio FJ, Alexander A, Goodridge C, Shen E, Shinbrot T. Solid
exhibit constant variability, making the optimum mixing. In: Paul EL, Atiemo-Obeng VA, Suzanne K, editors.
Chapter 15A, handbook of industrial mixing: science and practice.
process a moving target. Additionally, the amount of New York: Wiley; 2004. p. 887–923.
work needed to identify, characterize, and control all 12. Brone D, Alexander A, Muzzio FJ. Quantitative characterization
variables affecting product performance is significant. of mixing of dry powders in V-blenders. AlChE J 1998;44
With this in mind, only a first-pass design may be (2):271–8.
13. Sudah OS, Coffin-Beach D, Muzzio FJ. Effects of blender rotation
achieved within the short time frames associated with speed and discharge on the homogeneity of cohesive and free-
current pharmaceutical product development cycles. flowing mixtures. Int J Pharm 2002b;247:57–68.
However, valuable information is generated by the 14. FDA. Powder blends and finished dosage units—stratified in
manufacturing operation, for further refinement of process dosage unit sampling and assessment, Pharmaceutical
CGMP’s, Guidance for Industry; 2003.
models for the improvement of product performance. 15. International Conference on Harmonisation of Technical Require-
In an enlightened QbD-enabled regulatory framework, ments for Registration of Pharmaceuticals for Human Use, ICH
Harmonised Tripartite Guideline, Pharmaceutical Development
continuous improvement is enabled through the process Q8(R1), 1-November-2007.
understanding that is gained during process development 16. International Conference on Harmonisation of Technical Require-
and manufacturing. An idea advanced by this work is that ments for Registration of Pharmaceuticals for Human Use, ICH
of dynamic quality specifications. These ultimately allow Harmonised Tripartite Guideline, Quality Risk Management Q9,
Step 4, 9-November-2005.
for more flexibility at the beginning of the manufacturing 17. International Conference on Harmonisation of Technical Require-
life cycle, when knowledge is sparse, and lead to an ments for Registration of Pharmaceuticals for Human Use, ICH
improved, but more narrowly defined process, as the Harmonised Tripartite Guideline, Pharmaceutical Quality System
process approaches maturity. Q10, Step 4, 4-June-2008.
18. Cogdill RP, Drennen JK. Risk-based quality by design (QbD): a
Taguchi perspective on the assessment of product quality, and the
quantitative linkage of drug product parameters and clinical
performance. J Pharm Innov 2008;3(1):23–29.
References 19. Lemieux M, Bertrand F, Chaouki J, Gosselin P. Comparative
study of the mixing of free-flowing particles in a V-blender and a
1. Mehrotra A, Chaudhuri B, Faqih A, Tomassone MS, Muzzio FJ. bin-blender. Chem Eng Sci 2007;62(6):1783–802.
A modeling approach for understanding effects of powder flow 20. Muguruma Y, Tanaka T, Kawatake S, Tsuji Y. Discrete particle
properties on tablet weight variability. Powder Technol. 2008. simulation of a rotary vessel mixer with baffles. Powder Technol
doi:10.1016/j.powtec.2008.05.016 (in press). 1997;93(3):261–66.
2. James MN. Design, manufacture and materials; their interaction 21. Muzzio FJ, Robinson P, Wightman C, Brone D. Sampling
and role in engineering failures. Eng Fail Anal 2005;12:662–78. practices in powder blending. Int J Pharm 1997;155(2):153–78.
270 J Pharm Innov (2008) 3:258–270

22. Duong N-H, Arratia PE, Muzzio FJ, Lange A, Timmermans J, 29. Chaudhuri B, Mehrotra A, Muzzio FJ, Tomassone MS. Cohesive
Reynolds S. Homogeneity study using NIR spectroscopy: tracking effects in powder mixing in a tumbling blender. Powder Technol
magnesium stearate in Bohle Bin-blenders. Drug Dev Ind Pharm 2006;165:105–14.
2003;29:679–87. 30. Chang R-K, Badawy S, Hussain MA, Buehler JD. Drug development
23. Alexander A, Goodridge C, Muzzio F, Arratia P, Brone D, Sudah and industrial pharmacy. Drug Dev Ind Pharm 1995;21(3):361–8.
O. Part 1. Pharm Tech. 2004. 31. Li H, McCarthy JJ. Controlling cohesive particle mixing and
24. Brone D, Muzzio FJ. Enhanced mixing in double-cone blenders. segregation. Phys Rev Lett 2003;90:184301–4.
Powder Technol 2000;110:179–89. 32. Weber MW, Hrenya CM. Square-well model for cohesion in
25. Manjunath K, Dhodapkar S, Jacob K. Solid mixing: mixing of fluidized beds. Chem Eng Sci 2006;61(14):4511–27.
pariculate solids in the process industries. In: Paul EL, Atiemo- 33. Stanford MK, DellaCorte C, Eylon D. Particle morphology effects
Obeng VA, Suzanne K, editors. Chapter 15B, handbook of on flow characteristics of PS304 plasma spray coating feedstock
industrial mixing: science and practice. New York: Wiley; 2004. powder blend. NASANASA/TM-2002-211206. Springfield, VA:
p. 887–923. National Technical Information Service; 2002.
26. Portillo P, Muzzio F, Ierapetritou MG. Characterizing powder 34. Sheehan C, United States Pharmacopeia, 1174, powder flow,
mixing processes utilizing compartment models. Int J Pharm http://www.pharmacopeia.cn/v29240/usp29nf24s0_c1174.html.
2006;320:14–22. Accessed 1 Oct 2008.
27. Schwedes J. Review of testers for measuring flow properties of 35. Faqih A, Chaudhuri B, Alexander A, Davies C, Muzzio FJ, Tomassone
bulk solids. Granul Matter 2003;5:1–43. MS. An experimental/computational approach for examining uncon-
28. Walton OR, De Moor CP, Gill KS. Effects of gravity on cohesive fined cohesive powder flow. Int J Pharm 2006;324:116–27.
behavior of fine powders: implications for processing Lunar 36. Dehlert GW. A first course in design and analysis of experiments.
regolith. Granul Matter 2007;9:353–63. New York: Freeman; 2000. p. 48.

View publication stats

You might also like