See

discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/274727238

A dimensionless variable for the scale up and

transfer of a roller compaction formulation

Article in Drug Development and Industrial Pharmacy · April 2015

DOI: 10.3109/03639045.2015.1029937 · Source: PubMed

CITATION READS

1 58

4 authors, including:

David Belair
Celgene
37 PUBLICATIONS 165 CITATIONS

SEE PROFILE

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

Engineered three-dimensional multicellular culture model to recapitulate morphogenetic fusion

using human cells View project

Thesis: Regulating Growth Factor Activity via Biomimetic Materials View project

All content following this page was uploaded by David Belair on 26 May 2016.

The user has requested enhancement of the downloaded file.

 http://informahealthcare.com/ddi
ISSN: 0363-9045 (print), 1520-5762 (electronic)

Drug Dev Ind Pharm, Early Online: 1–10

! 2015 Informa Healthcare USA, Inc. DOI: 10.3109/03639045.2015.1029937

RESEARCH ARTICLE

A dimensionless variable for the scale up and transfer of a roller

compaction formulation
Nathan Boersen*, David Belair, Garnet E. Peck, and Rodolfo Pinal

Department of Industrial and Physical Pharmacy, Purdue University, West Lafayette, IN, USA
Drug Dev Ind Pharm Downloaded from informahealthcare.com by Celgene Corp on 04/09/15

Abstract Keywords
Roller compaction is the most commonly employed dry granulation process in the Dimensionless variable, operating
pharmaceutical industry. While this process is increasingly used as an alternative to wet parameters, ribbon porosity, roller
granulation, there are no parameter sets or system of equations to quickly scale up or transfer a compaction, scale up
formulation between two pieces of equipment. In this work, dimensionless variable was
examined as a method to transfer the operating parameters of a formulation between two History
different pieces of equipment. This work was completed to establish the ground work for
the development of a dimensionless relationship relating the operating parameters of the Received 2 September 2014
equipment to the porosity of the ribbon. The working hypothesis was three-fold, namely (i) that Revised 5 February 2015
ribbons of the same porosity made with different equipment will have similar properties, Accepted 9 March 2015
(ii) that it is possible to establish an objective relationship between ribbon porosity and Published online 8 April 2015
For personal use only.

a combination of operating parameters and raw material attributes and (iii) that by express-
ing such parameter combination as a dimensionless variable, it will be possible to use the
same relationship for different pieces of roller compaction equipment. The dimensionless
variable RP/RS*HFS*True Density*D2 was found to correlate well with the ribbon porosity
for the formulations and equipment used in these experiments. Depending on the formulation,
the average difference in ribbon porosity between the two units varied between 0.012 and
0.024.

Introduction the post-compacted ribbon, granule and tablet properties can be

difficult. This is due to the complexity of the material’s
Roller compaction is the most commonly employed dry granu-
interaction with the operating parameters of the equipment.
lation process in the pharmaceutical industry. This unit operation
Numerous attempts have been made to model the roller compac-
exploits a powder’s ability to plastically deform to form stable
tion process. These include theoretical models6, finite element
compacts. Roller compaction has seen increased use in the
methods7, neural networks8–10, quadratic regression11, multilinear
pharmaceutical industry over the past few decades because many
stepwise regression12, thin layer13 and polynomials14.
newly developed active pharmaceutical ingredients (APIs) can be
Recently, various authors have begun to elucidate the
sensitive to the liquid addition and/or drying stage applied during
interactions between the material attributes of the formulation
wet granulation. Roller compaction is also growing in popularity
and the operating parameters of the equipment15–18. While
because it is a one-step process that lends itself to continuous
previous research had demonstrated the importance of the
manufacturing and has a higher production throughput than other
operating parameters, this research revealed, for certain responses,
granulation techniques.
the raw material attributes of a formulation were as important, if
Roller compaction, like many granulation techniques, has
not more so, than the operating parameters of the equipment.
traditionally used a trial and error approach towards developing a
While Soh et al. used a placebo formulation, Boersen et al., used
formulation and the operating parameters used for manufactur-
a similar approach to predict the ribbon porosity using the same
ing1. Design of experiments has been used to determine which
formulation, but different active ingredients17.
operating parameters give the best product performance2–4. Yet,
Work by Boersen et al. concentrated on trying to predict
this can be time consuming and could result in significant
ribbon porosity because this property is often employed as a
losses of API when scaling to larger pieces of equipment5.
measurement to describe the density of a ribbon19. Ribbon
While scientists are increasingly using this technology, predicting
porosity can often be difficult to measure if rollers with a surface
texture are used to effectively grip the powder to bring it between
the rollers20. However, it is a useful parameter that allows one to
directly compare different concentrations of the same API or two
Address for correspondence: Nathan Boersen, (*Present Address)
Department of Formulations Research and Development, Celgene
different formulations. Since porosity is a dimensionless variable,
Corporation, 86 Morris Ave, Summit, NJ 07901, USA. Tel: +1 908 673 it may be a good parameter to use in the transfer and scale up of
9540. Fax: +1 908 673 2788. E-mail: nboersen@celgene.com a formulation between different pieces of equipment.
 2 N. Boersen et al.
To date, there are no parameter sets or system of equations that
can quickly scale up or transfer a formulation between two pieces
of equipment. A few scale-up approaches have been detailed in
the literature21–23. Most recently, Liu et al. have explored Joint Y
Partial Least squares (PLS) for product scale up of formulations
from a lab scale Fitzpatrick IR 220 to a full scale Fitzpatrick IR
52021. In their work, the authors used historical data to generate a
model from formulations compacted on both roller compactors.
Ribbon density and solid fraction were modeled by API and
excipient mass fraction, along with roll speed (RS) and pressure
and feed screw speed. The authors were able to develop models
that could predict the ribbon properties. However, it was noted
that it was not possible to scale up additional products outside of
the model constraints. This suggests that the formulation proper-
ties may play a critical role when trying to model products that
differ from the original model.
Trial and error, or a design of experiments, is commonly
employed to determine the initial operating conditions for a
particular equipment train during preliminary roller compaction
Drug Dev Ind Pharm Downloaded from informahealthcare.com by Celgene Corp on 04/09/15
experiments. While time consuming, this approach may not result
in large losses of a potentially expensive API. Should the project
proceed beyond the laboratory, larger pieces of equipment would
be required to produce pilot and production size lots. Trial and
error or a design of experiments can be applied to determine the
relationship between variables, however, this can be time
consuming and expensive.
Due to the complexity of the interactions between the
operating parameters and raw material attributes, developing a
first principles model to predict ribbon porosity may be difficult.
To circumvent these difficulties, relevant variables can be
For personal use only.
grouped into a dimensionless parameter. When developing a
dimensionless parameter, it is important to first establish which
variables influence the outcome of the results. After determining
the number of reference dimensions, an algebraic algorithm is
applied and the important variable(s) relating to the problem will
be established in a simplified dimensionless form24. Dimensional
analysis is valuable because it reduces the number of experiments
needed to determine the relationship between variables. These
new variable(s) now represent a ratio between the important terms
in the equation25.
The objective of this study was to explore a science-based
approach for the scale-up and transfer of roller compacted
formulations. Specifically, this project looked to establish if
dimensionless variables could be used as criteria for scale up and
transferability. To this effect, the investigation focused on ribbon
porosity as the ribbon parameter of central interest. The working
hypothesis was three-fold, namely (i) that ribbons of the same
porosity made with different equipment will have similar
properties, (ii) that it is possible to establish an objective
relationship between ribbon porosity and a combination of
operating parameters and raw material attributes and (iii) that
by expressing such a parameter, combination as a dimensionless
variable, it will be possible to use the same relationship for
different pieces of roller compaction equipment. Multivariate
relationships were also established to validate the relationships
between the variables used to create the dimensionless relation-
ship and the porosity of the ribbon.
Experimental
Materials
For all roller compaction experiments, microcrystalline cellulose
(MCC) served as the model plastically deforming excipient.
Avicel PH-102 (FMC Corporation, Newark, DE) was separated
into three separate sieve cuts, #100 (4150 lm), #200 (75–106 lm)
and the pan (553 lm) using US Standard Test Sieves (W.S. Tyler,
Drug Dev Ind Pharm, Early Online: 1–10
Mentor, OH), blended with magnesium stearate and roller
compacted. Blends of uncut Avicel PH-102 and 25% of either
tolmetin sodium dihydrate (Noramco Inc., Athens, GA) or
acetaminophen (Tyco Healthcare, Raleigh, NC) were blended
along with 1% magnesium stearate and 1% silicon dioxide
(Aerosil, Degussa Corporation, NJ) and compacted. Emcocel
SCG (J.R.S. Pharma LP, NY) was also blended with 25% tolmetin
sodium dihydrate, 1% magnesium stearate and 1% silicon dioxide
and roller compacted.
Equipment
Alexanderwerk WP 120  40
Roller compaction experiments were first completed using the
Alexanderwerk WP 120  40 unit (Alexanderwerk Inc., Horsham,
PA). The WP 120  40 has three controllable operating param-
eters: RS, roll pressure (RP) and a horizontal feed screw (HFS).
High and low operating parameters varied between 4 and 12 rpm
for roll speed (RS) and 20 and 40 bar for RP. The HFS speed was
held constant at approximately 20 rpm. Roll gap was monitored
but not held constant throughout any of the experiments. Ribbon
samples of roughly 25 cm in length were taken every 20 s for 60 s.
All compaction experiments on the Alexanderwerk used upper
and lower knurled rollers.
Fitzpatrick IR-220
Following the experiments on the Alexanderwerk roller com-
pactor, the same blends were compacted using the Fitzpatrick
Chilsonator IR-220 unit (The Fitzpatrick Co., Elmhurst, IL). The
Chilsonator IR-220 has four controllable operating parameters:
RS, RP and a horizontal (HFS) and vertical (VFS) feed screw.
High- and low-operating parameters were varied between 3 and
9 rpm for RS and 700 and 1000 psi for RP. The VFS speed was
varied between 150 and 170 rpm and the HFS speed was varied
between 19 and 21 rpm. Real time operating parameters were
downloaded from the roller compactor interface and included in
all analyses. Samples approximately 25 cm in length were taken
every 30 s for 90 s. All compaction experiments on the Fitzpatrick
unit were completed using two smooth rollers whose surfaces
had been sand blasted, allowing the rollers to effectively grip
the powder.
Differences in equipment design
While both the Alexanderwerk and Fitzpatrick units compact
powder to form a ribbon, the two pieces of equipment have very
different design features. These different design features may limit
the practicality of applying a dimensionless variable across
different pieces of equipment. Therefore, one of the goals of this
research was to find a dimensionless relationship that eliminated
differences in equipment design, or at least minimizes them.
Table 1 lists many of the design differences for the
Alexanderwerk and Fitzpatrick units. The arrangement of the
rollers, either vertically, horizontally or at a 45 angle, will
influence the powder’s rearrangement in the slip region. These
roller arrangements will also apply slightly different compressive
and shear forces across the diameter of the rollers. Another
important difference is the types of rollers used on each
compactor. For this study, smooth rollers were employed for the
Fitzpatrick unit while knurled rollers were employed for the
Alexanderwerk unit. The knurled rollers allow the material to be
better pulled into the slip and nip regions of the compactor and
may affect the compaction of the powder.
The different feed screw designs may also have an effect of the
relationship between the dimensionless variable and ribbon
porosity. The Fitzpatrick unit is comprised of both a VFS and
 DOI: 10.3109/03639045.2015.1029937 Roller compaction with dimensionless variables 3
Table 1. Design features of the Fitzpatrick and Alexanderwerk roller 0.22 mm3/mm2. The porosity of the ribbon was determined
compactors. using the following equation:
Design feature Unit 1 (Fitzpatrick) Unit 2 (Alexanderwerk) app
Ribbon Porosity ¼ 1 
Powder feed Vertical and horizontal Horizontal deed screw
true
feed screws
Rollers Smooth Knurled where app is the apparent density of the ribbon and true is the
Outflow Vertical Horizontal true density of the formulation. Three samples were taken from
Roll width 0.02 m 0.04 m each time point for a total of nine samples per compaction
Roll diameter 0.20 m 0.12 m experiment. The nine samples were then averaged to obtain the
Vacuum deaeration Available with the Available in unit ribbon porosity for each set of experiments.
addition of a pump
Roll gap control Not available Available
function
Calculation of the dimensionless variable
The transfer and scalability of a roller compacted formulation is
difficult since no set parameters or system of equations exist to
HFS while the Alexanderwerk unit only has one HFS. In addition quickly scale up or transfer a formulation between two pieces of
to the feed screw design, the screw flights are different for the two equipment. Dimensionless variables are commonly employed in
units. The differences in design and in screw flight will deliver many areas of engineering as a science-based approach to transfer
different amounts of powder between the rollers and cause the operating settings between pieces of equipment. Therefore,
Drug Dev Ind Pharm Downloaded from informahealthcare.com by Celgene Corp on 04/09/15

powder to have dissimilar pre-densification and deaeration rates.
While there are many design differences between the two units,
both are laboratory scale pieces of equipment. As a result,
differences in design may only be expressed when scaling to pilot
size or manufacturing pieces of equipment.
Methods
Blending
For all experiments, powder was weighed and sifted through a 30
For personal use only.
dimensional analysis was investigated as an approach to transfer
the operating parameters between roller compactors made by two
different manufacturers. If a common relationship brought out by
the dimensionless parameters exists between the two compactors,
ribbons of the same porosity made with different equipment may
exhibit similar properties.
Analysis from Boersen et al. revealed the RP, RS, feed screw
speed, true density, flowability as measured by angle of repose
and the mean particle size of the formulation were the critical
factors needed to model and predict the ribbon porosity of a
formulation, containing the same excipients, but different APIs17.

mesh screen. The powder was pre-mixed using a large scoop

before it was transferred to a 1.5 cubic foot tote blender (Model
760092, Tote Systems, TX) and blended for 8 min at 16 rpm.
Batch sizes ranged between 3 and 5 kg.
True density
The true density of all excipients and APIs were measured using a
helium pycnometer (Micromeritics AccuPyc 1330, Micromeritics,
GA). MCC was dried for 4 h in a 105 C oven. Tolmetin,
acetaminophen, silicon dioxide and magnesium stearate were
dried overnight in a 50 C oven. After drying, approximately 3–5 g
of powder was placed in a stainless steel cylinder where one
sample had its true density measured six times. Experiments were
completed in triplicate. The true density of the blends was
determined by taking the summation of the weight fraction of the
excipient and multiplying by the true density of each material
X
i
true ¼ wi i
n¼1
where true is the true density of the formulation, and wi and i are
the weight fraction and density of component i in the blend,
respectively.
Ribbon property measurements
After compaction was completed, the ribbons were stored in a
humidity controlled room at approximately 25 C/25% RH for
approximately 12 h allowing for relaxation. The ribbons were
weighed using a Mettler AE100 balance (Mettler Toledo,
Columbus, OH) and their length and width were measured once
and thickness was measured in triplicate using digital calipers
(VWR, West Chester, PA). Since knurled rollers were used on
the Alexanderwerk roller compactor, a correction factor was
introduced to account for the void volume left by the rollers. The
void volume correction factor for two knurled rollers was
This research agrees well with other authors who also determined
that RS, RP and HFS speed were the critical operating parameters
needed to model various post-compacted ribbon, granule and
tablet properties8,11,26,27.
Using this previous research as a guide, a dimensionless
parameter was created to transfer operating parameters between
different pieces of equipment. The dimensionless parameter used
in this study was a combination of five variables, RS, RP, HFS
speed, true density (True) and roll diameter (D). Since the raw
material attributes of the formulation are constant between the
pieces of equipment, parameters like true density, flowability, and
mean particle size do not need to be included in the dimensionless
parameter. However, to make a truly dimensionless parameter,
true density was added to cancel the mass units in the pressure
term. Therefore, the relationship between the compactor and
ribbon porosity may only be a function of the operating
parameters of the equipment. Material attributes such as particle
size and moisture content would be expected to only change the
ð1Þ
slope and intercept of the relationship. As a result, different
formulations may not be directly comparable for the same
dimensionless variable. Roll diameter was also included in the
relationship as an additional scaling factor for the two pieces
of equipment.
RS will influence ribbon porosity due to its effect on the time
the powder spends in the compaction zone. A lower RS increases
time in the compaction zone, increasing density and decreasing
porosity. An increased RP will also increase density and decrease
porosity. Feed screws help to maintain a constant flow of material
between the rollers26. Feeding greater amounts of material into
the slip and nip regions of the roller compactor will also densify
the ribbon and decrease ribbon porosity. Roll diameter is an
important design factor influencing size of the equipment.
RS, RP, the HFS speed, true density and roll diameter have
units of (1/time), (mass/time2 * length), (1/time), (mass/length3)
and (length), respectively. A total of three independent physical
units, mass, length and time are needed to compose the units of
 4 N. Boersen et al. Drug Dev Ind Pharm, Early Online: 1–10

these variables. Dimensionless Variable 1 shows the dimension- calculated using cross-validation29. Values greater than 0.5
less parameter created to produce ribbons of the same porosity on indicate a good predictability, while values greater than 0.9
units by two different manufacturers. indicate an excellent predictability. External validation was
applied to validate a new model set combining the formulations
RP
Dimensionless Variable 1 ¼ DV ¼ using two different APIs. The residuals, root mean square error of
RS  HFS  True  D2 prediction (RMSEP) and the percent error of the prediction were
where RP has units of (Pa), RS has units of (1/min), HFS speed determined.
units of (1/min), true density units of (kg/m3) and roll diameter
units of (m). The dimensionless parameter was then divided by
Results and discussion
10 000 to reduce the magnitude of the variable.
To test the feasibility of using the dimensionless variable as a Relationships between ribbon porosity and
way to transfer the operating parameters between different pieces RP/RS*HFS*True*D2 for sieved Avicel PH-102
of equipment, pure MCC and blends of an API and MCC were
It was hypothesized that differences in the raw material attributes
roller compacted. Pure MCC was compacted to investigate if a
of the formulation would be directly reflected in the slope and
relationship exists between the dimensionless variable and ribbon
intercept of the line of best fit through the data. Three sieve cuts
porosity. Blends of MCC and an API were compacted to further
of Avicel PH-102 were compacted to investigate the influence of
confirm any relationship and to more realistically portray
particle size on the dimensionless parameter. Figures 1–3 plot the
formulations used in industry.
ribbon porosity versus the dimensionless parameter for the three
Compaction was first completed on the Alexanderwerk unit.
Drug Dev Ind Pharm Downloaded from informahealthcare.com by Celgene Corp on 04/09/15

cuts of Avicel PH-102. These sieve cuts ranged in size from

During compaction runs, real-time operating parameters were
greater than 150 mm, 75–106 mm and less than 53 mm. The plots
manually recorded from the display panel. After compaction
show ribbon porosity was highly correlated to the dimensionless
was complete, the dimensionless variable for each set of
parameter for all sieve cuts, R240.850.
Alexanderwerk operating parameters was calculated. The operat-
At the same set of operating parameters, ribbon porosity
ing parameters for the corresponding Fitzpatrick trial were back
should increase with increasing particle size. The influence of
calculated using the same dimensionless number from the
particle size on ribbon porosity is somewhat reflected by the y
Alexanderwerk trial. While an infinite number of Fitzpatrick
intercept of the line of best fit through the data. As the particle
operating parameter combinations exists, the values input into the
size decreases from greater than 150 lm to 75–106 lm, the ribbon
variable are limited by the minimum and maximum operating
porosity at a dimensionless parameter of zero, or the maximum
parameter values. In initial experiments completed with the
ribbon porosity, decreases from 0.7022 to 0.6873. However, an
For personal use only.

Fitzpatrick unit, the RP and feed screw speed were held constant
while the RS was altered to change the value of the dimensionless
parameter. In later experiments, either the feed screw speed or RP
was simultaneous varied with the RS while remaining parameter
was held constant.
Multivariate data analysis
PLS analysis using the SIMCA-P + Version 11.0.0.0 (Umetrics,
Umeå, Sweden) software was employed to model the roller
compaction operating parameters, raw material attributes and the
ribbon porosity. Unit variance scaling was applied to all data. UV
scaling is commonly employed when there is no prior knowledge
of a variables’ influence28. The predictability of the models was
increase in the maximum ribbon porosity was observed when the
particle size was further decreased to below 53 lm. As compac-
tion progressed with the powder less than 53 mm, a significant
amount of powder leaked before the powder entered the rollers
for the Alexanderwerk unit and from between the rollers of the
Fitzpatrick unit. Powder leaking between the rollers would
prevent the maximum amount of powder entering the slip and
nip regions of the rollers resulting in a ribbon with a lower than
expected density. The uncertainties in the data for Avicel PH-102,
sieve cut pan (553 mm) may be caused by this phenomenon. The
data also indicate there may be a minimum particle size to
effectively convey all powder through the slip and nip regions of
the roller compactor.

Figure 1. Relationship between ribbon por-

osity and RP/RS*HFS*True*D2 for Avicel
PH-102, sieve cut #100 (4150 lm).
 DOI: 10.3109/03639045.2015.1029937
Figure 2. Relationship between ribbon por-
osity and RP/RS*HFS*True*D2 for Avicel
PH-102, sieve cut #200 (75–106 lm).
Drug Dev Ind Pharm Downloaded from informahealthcare.com by Celgene Corp on 04/09/15
Figure 3. Relationship between ribbon por-
osity and RP/RS*HFS*True*D2 for Avicel
PH-102, sieve cut pan (553 lm).
For personal use only.
Tolmetin and acetaminophen blends
Figures 4–6 plot the ribbon porosity versus the dimensionless
variable for three blends containing two different APIs. Figures 5
and 6 are blends of tolmetin or acetaminophen and Avicel PH-
102, respectively. Figure 7 shows a plot of tolmetin and Emcocel
SCG. Blends containing tolmetin have R2 correlations greater
than 0.870, while the blend containing acetaminophen have a R2
correlation approximately 0.820.
Tables 2–4 compare the difference between the Alexanderwerk
and Fitzpatrick units using the Alexanderwerk values as the
reference value. The percent error in all cases is less than 16%,
translating to a difference in porosity no greater than 0.07.
However, approximately three quarters of the trials have a percent
error less than 8.3%, or an average porosity difference of 0.024.
Due to the inherent variability of the applied RP, feed
screw or RS, a variation in the ribbon porosity will exist during
the production of the ribbon and may account for the observed
Roller compaction with dimensionless variables 5
disparity. This small difference could also be a function of the
different equipment designs. The effect of such design differences
could be systematically investigated in order to discriminate
equipment design attributes that influence the process from those
that do not.
The differences in design may be more pronounced when
scaling from a laboratory to pilot or production scales. The
variable with the greatest influence on ribbon porosity was RS.
RS controls the time the powder spends between the rollers, with
a lower RS leading towards longer dwell times. When scaling to
pieces of equipment with larger rollers, the RS would have to
decrease if the dwell time is to be kept constant. Dwell times can
be kept constant by calculating the linear roll velocity, RS
multiplied by the circumference of the roller. Constant linear roll
velocities help to keep the dwell time consistent between roller
compactors. For the MCC and acetaminophen/tolmetin blends,
the linear roll velocity was on average 24% higher for the
 6 N. Boersen et al.
Figure 4. Relationship between ribbon por-
osity and RP/RS*HFS*True*D2 for a blend
of 75% Avicel PH-102 and 25% tolmetin.
Drug Dev Ind Pharm Downloaded from informahealthcare.com by Celgene Corp on 04/09/15
Figure 5. Relationship between ribbon por-
osity and RP/RS*HFS*True*D2 for a blend
of 75% Avicel PH-102 and 25%
acetaminophen.
For personal use only.
Fitzpatrick roller compactor. This may help to explain why the
Fitzpatrick roller compactor tended to produce ribbons with
higher ribbon porosities.
Multivariate modeling of ribbon porosity
PLS analysis was employed to model the dimensionless variable,
RP, RS, true density, roll diameter and HFS speed with the
porosity of the compacted ribbon. Initial models used 24 tolmetin
experiments and 23 acetaminophen experiments to predict ribbon
porosity. The scores plot of the initial model revealed three
potential outliers. These observations were subsequently removed
from the remainder of the analyses.
The remaining 44 observations were used to model ribbon
porosity. The fit and predictability of two models, one with and
one without the dimensionless variable, are listed in Table 5. It
took two components to model ribbon porosity with the dimen-
sionless variable, producing a predictability of 0.874. Without the
Drug Dev Ind Pharm, Early Online: 1–10
dimensionless variable, it took three components to model ribbon
porosity, producing a predictability of 0.849. This value indicated
these variables could sufficiently predict the ribbon porosity of a
formulation of the same excipients, but different APIs.
Figure 7 shows the coefficient plot for the six variables included
in model 1. The coefficient plot reveals both the dimensionless
variable and RS are influential, with the dimensionless variable
contributing more information to the prediction of ribbon porosity.
RP, true density, the HFS speed and roll diameter contribute little
meaningful information to predict ribbon porosity. While RS does
contribute a significant amount of information to the model, the
addition of these other variable rolled into one dimensionless
variable is able to account for much more variability. Figure 8(a)
and (b) shows a simple linear regression of ribbon porosity with
either the RS or the dimensionless variable. Linear regression
shows a R2 of 0.639 and 0.812 for ribbon porosity versus RS and
ribbon porosity versus the dimensionless variable, respectively.
The simple linear regression shows the dimensionless variable is
 DOI: 10.3109/03639045.2015.1029937 Roller compaction with dimensionless variables 7
Figure 6. Relationship between ribbon por-
osity and RP/RS*HFS*True*D2 for a blend
of 75% Emcocel SCG and 25% tolmetin.
Drug Dev Ind Pharm Downloaded from informahealthcare.com by Celgene Corp on 04/09/15
For personal use only.

Figure 7. Coefficient plot for the six variables included in model 1.

not only accounting for the large information input by ribbon
porosity, but also the little information input by RP, true density,
the HFS speed and roll diameter.
To externally validate the data, a new modeling set was created
using 34 of the 44 observations. Ten observations were used to
validate the new modeling sets. The fit and predictability of these
models are listed in Table 5. Models 3 and 4, with and without the
dimensionless variable, produced predictabilities of 0.882 and
0.827, respectively. For these models, the residuals for the
predicted values were determined along with the RMSEP and the
percent error of the prediction. Residuals are the difference
between the actual values and the predicted values, RMSEP
shows the average uncertainty expected and the percent error
shows the deviation of the prediction. A valid model will have
 8 N. Boersen et al.
Figure 8. Linear regression for ribbon poros-
ity versus roll speed (a) and ribbon porosity
versus the dimensionless variable (b) for all
observations.
Drug Dev Ind Pharm Downloaded from informahealthcare.com by Celgene Corp on 04/09/15
For personal use only.
residuals, RMSEP and a percent error close to zero. These values
are listed in Table 6.
The multivariate model showed a correlation could be drawn
using the five variables to establish a relationship with the
porosity of the ribbon. In addition, the error of prediction
calculated by the model (0.023) was similar to the average
difference in porosity calculated between the Fitzpatrick and
Alexanderwerk units (0.024).
The multivariate relationship also confirmed true density did
not significantly influence the ribbon porosity. While not as
influential, true density would be a useful parameter when the
excipients of a formulation are the same, but the API varies. If the
concentration of the API was held constant between the two
formulations, the powders may have similar bulk properties. The
dimensionless relationship could then be used to determine the
initial operating parameters on a similar roller compactor, but a
different formulation.
Drug Dev Ind Pharm, Early Online: 1–10
For the formulations and equipment used in this research, the
dimensionless variable RP/RS*HFS*True Density*D2 was shown
to correlate the ribbon porosity between roller compactors from
two different manufacturers. Using these formulations, the
compaction of the same formulation on two different compactors
appears to be a function of the operating parameters of the
equipment and not the attributes of formulation. Therefore, a
relationship between the two compactors could be established.
While the raw material attributes’ influence may not be as evident
on smaller pieces of equipment, these interactions may be more
pronounced as a formulation is scaled from laboratory to pilot to
production pieces of equipment.
While methods like PLS, employed by Liu et al., may provide
better prediction of the ribbon porosity on different scales,
significant time and quantities of material are needed to establish
such relationships21. These newly developed models may then
be product or API specific, limiting predictive capabilities.
 DOI: 10.3109/03639045.2015.1029937 Roller compaction with dimensionless variables 9
Table 2. Ribbon porosity values for blends of tolmetin and Avicel PH-
Table 4. Ribbon porosity values for blends of tolmetin and Emcocel SCG
102 for both the Alexanderwerk and Fitzpatrick units.
for both the Alexanderwerk and Fitzpatrick units.
RP/RS*HFS* Ribbon Percent
RP/RS*HFS* Ribbon Percent
Trial Compactor True*D2 porosity error
Trial Compactor True*D2 porosity error
F1-1-1 Fitzpatrick 248 0.487 6.4
F1-1 Fitzpatrick 255 0.365 6.7
A1-1-1 Alexanderwerk 240 0.458
A1-1 Alexanderwerk 261 0.391
F2-1-1 Fitzpatrick 228 0.512 10.9
F2-1 Fitzpatrick 287 0.347 1.6
A2-1-1 Alexanderwerk 228 0.462
F2-2 Fitzpatrick 284 0.340 3.7
F1-2-1 Fitzpatrick 272 0.445 3.7 A2-1 Alexanderwerk 290 0.353
A1-2-1 Alexanderwerk 269 0.429
F3-1 Fitzpatrick 337 0.292 3.9
F2-2-1 Fitzpatrick 275 0.446 11.7 F3-2 Fitzpatrick 332 0.307 1.2
A2-2-1 Alexanderwerk 278 0.417 A3-1 Alexanderwerk 342 0.304
F1-3-1 Fitzpatrick 337 0.358 4.6
F1-3-2 Fitzpatrick 345 0.383 11.8 A negative percent error indicates the Fitzpatrick ribbon porosity was
F1-3-3 Fitzpatrick 331 0.386 12.5 below the reference value.
F1-3-4 Fitzpatrick 335 0.356 4
F1-3-5 Fitzpatrick 330 0.355 3.6
F1-3-6 Fitzpatrick 329 0.341 0.7
F1-3-7 Fitzpatrick 323 0.342 0.2 Table 5. Fit and predictability predicting ribbon porosity.
Drug Dev Ind Pharm Downloaded from informahealthcare.com by Celgene Corp on 04/09/15

F1-3-8 Fitzpatrick 336 0.347 1.2

A1-3-1 Alexanderwerk 338 0.343 R2Y Q2(cum)
F2-3-1 Fitzpatrick 348 0.287 0.4 Model Observations Variables Components (Fit) (Predictability)
A2-3-1 Alexanderwerk 352 0.301 1 44 RS, RP, HFS, 2 0.905 0.874
F1-4-1 Fitzpatrick 441 0.287 7.3 True, D, DV
F1-4-2 Fitzpatrick 443 0.301 3.1 2 44 RS, RP, HFS, 3 0.896 0.849
F1-4-3 Fitzpatrick 448 0.274 11.7 True, D,
F1-4-4 Fitzpatrick 448 0.274 11.6 3 34 RS, RP, HFS, 2 0.915 0.882
A1-4-1 Alexanderwerk 450 0.310 True, D, DV
4 34 RS, RP, HFS, 3 0.898 0.827
A negative percent error indicates the Fitzpatrick ribbon porosity was True, D,
below the reference value.
For personal use only.

Table 3. Ribbon porosity values for blends of acetaminophen and Avicel Table 6. Residuals, root mean square error of prediction (RMSEP) and
PH-102 for both the Alexanderwerk and Fitzpatrick units. the percent error of the prediction for externally validated modeling set.

RP/RS*HFS* Ribbon Percent Model Variables Residuals RMSEP Percent error

Trial Compactor True*D2 porosity error
3 RS, RP, HFS, 0.0003 0.0232 11.0%
F1-1-1 Fitzpatrick 229 0.464 4.7 True, D, DV
A1-1-1 Alexanderwerk 232 0.487 4 RS, RP, HFS, 0.0632 0.0236 11.2%
F1-2-1 Fitzpatrick 285 0.387 15.4 True, D,
F1-2-2 Fitzpatrick 251 0.456 0.2
F1-2-3 Fitzpatrick 263 0.458 0.1
F1-2-4 Fitzpatrick 278 0.457 0.1
A1-2-1 Alexanderwerk 290 0.457
parameters2–4,11,33 influence the density of the ribbon. Along with
F2-2-1 Fitzpatrick 268 0.445 1.7
A2-2-1 Alexanderwerk 274 0.453
differences in equipment design, it will be difficult to develop any
one size fits all relationship that will apply across different
F1-3-1 Fitzpatrick 385 0.321 13.9
F1-3-2 Fitzpatrick 329 0.342 8.1
equipment trains and formulations. The effectiveness of this
F1-3-3 Fitzpatrick 346 0.354 4.8 dimensionless variable applied across two equipment trains while
F1-3-4 Fitzpatrick 385 0.353 5.2 using a similar formulation with different active ingredients
A1-3-1 Alexanderwerk 395 0.372 makes it a good first approach to transfer a formulation between
F2-3-1 Fitzpatrick 340 0.361 7.3 different pieces of equipment and gain an estimation of the ribbon
A2-3-1 Alexanderwerk 347 0.390 porosity. Along with the knowledge of the formulator and
F1-4-1 Fitzpatrick 415 0.314 8.3 operator, using this dimensionless variable could help save time,
F1-4-2 Fitzpatrick 366 0.332 3.0 money and materials during the initial phases of scaling to
F1-4-3 Fitzpatrick 380 0.315 7.9 contract manufacturer with a different set of equipment, or to
F1-4-4 Fitzpatrick 409 0.331 3.1 larger equipment trains.
A1-4-1 Alexanderwerk 418 0.342
F2-4-1 Fitzpatrick 430 0.297 16.0 Conclusions
A2-4-1 Alexanderwerk 433 0.354
This study was completed to bring a more scientific approach
A negative percent error indicates the Fitzpatrick ribbon porosity was towards the scaling up and transferring of roller compacted
below the reference value. formulations. In addition, this methodology helps to lay the
ground work for the development of a dimensionless relationship,
In addition, knowledge of multivariate analysis and a multivariate relating the selected operating parameters of the equipment to the
software are required to model the post-roller compacted ribbon porosity of the ribbon. While there were many design differences
properties. The difficulty in trying to predict ribbon properties is between the two units, fairly strong relationships were observed
the fact that both raw material properties14,30–32 and operating relating ribbon porosity to the dimensionless variable for
 10 N. Boersen et al.
acetaminophen and tolmetin. However, further studies are needed
to investigate if this relationship can be applied to equipment of
different manufacturers, larger scale processes or different
formulations. While this work is preliminary, this dimensionless
variable may be a good first approach equation to transfer the
operating parameters between different pieces of equipment.
Acknowledgements
The authors thank The Fitzpatrick Corporation for donating the IR-220
roller compactor, Alexanderwerk, Inc. for lending the WP 120  40 roller
compactor and FMC Biopolymer for donating the MCC used in these
experiments.
Declaration of interest
The authors report no declarations of interest.
References
1. Kleinebudde P. Roll compaction/dry granulation: pharmaceutical
Drug Dev Ind Pharm Downloaded from informahealthcare.com by Celgene Corp on 04/09/15
applications. Eur J Pharm Biopharm 2004;58:317–26.
2. Hariharan M, Wowchuk C, Nkansah P, Gupta V. Effect of
formulation composition on the properties of controlled release
tablets prepared by roller compaction. Drug Dev Ind Pharm 2004;
30:565–72.
3. von Eggelkraut-Gottanka S, Abed S, Müller W, Schmidt P. Roller
compaction and tabletting of St. John’s wort plant dry extract using a
gap width and force controlled roller compactor. II. Study of roller
compaction variables on granule and tablet properties by a 33
factorial design. Pharm Dev Technol 2002;7:447–55.
4. Rambali B, Baert L, Jans E, Massart D. Influence of the roll
compactor parameter settings and the compression pressure on the
buccal bio-adhesive tablet properties. Int J Pharm 2001;220:129–40.
For personal use only.
5. Sheskey P, Pacholke K, Sackett G, et al. Roll compaction
granulation of a controlled-release matrix tablet formulation
containing HPMC: effect of process scale-up on robustness of
tablets and predicted in vivo performance. Pharm Technol 1999;23:
6–21.
6. Johanson J. A rolling theory for granular solids. J Appl Mech 1965;
32:842–8.
7. Cunningham J, Winstead D, Zavaliangos A. Understanding variation
in roller compaction through finite element-based process modeling.
Comput Chem Eng 2010;34:1058–71.
8. Inghelbrecht S, Remon J, Fernandes de Aguiar P, et al.
Instrumentation of a roll compactor and the evaluation of the
parameter settings by neural networks. Int J Pharm 1997;148:
103–15.
9. Turkoglu M, Aydin I, Murray M, Sakr A. Modeling of a roller-
compaction process using neural networks and genetic algorithms.
Eur J Pharm Biopharm 1999;48:239–45.
10. Mansa R, Bridson R, Greenwood R, et al. Using intelligent software
to predict the effects of formulation and processing parameters on
roller compaction. Powder Technol 2008;181:217–25.
11. Falzone A, Peck G, McCabe G. Effects of changes in roller
compactor parameters on granulations produced by compaction.
Drug Dev Ind Pharm 1992;18:469–89.
12. Chang C, Alvarez-Nunez F, Rinella J, et al. Roller compaction,
granulation and capsule product dissolution of drug formulations
containing a lactose or mannitol filler, starch, and talc. AAPS
PharmSciTech 2008;9:597–604.
13. Peter S, Lammens R, Steffens K. Roller compaction/dry granu-
lation: use of the thin layer model for predicting densities and
forces during roller compaction. Powder Technol 2010;199:
165–75.
View publication stats
Drug Dev Ind Pharm, Early Online: 1–10
14. Inghelbrecht S, Remon J. The roller compaction of different types of
lactose. Int J Pharm 1998;166:135–44.
15. Soh J, Boersen N, Carvajal M, et al. Importance of raw material
attributes for modeling ribbon and granule properties in roller
compaction: multivariate analysis on roll gap and NIR spectral slope
as process critical control parameters. J Pharm Innov 2007;2:
106–24.
16. Soh J, Wang F, Boersen N, et al. Utility of multivariate analysis in
modeling the effects of raw material properties and operating
parameters on granule and ribbon properties prepared in roller
compaction. Drug Dev Ind Pharm 2008;34:1022–35.
17. Boersen N, Carvajal M, Morris K, et al. The influence of API
concentration on the roller compaction process: modeling and
prediction of the post compacted ribbon, granule and tablet
properties using multivariate data analysis. Drug Dev Ind Pharm
2014. [Epub ahead of print].
18. Boersen N. The development of roller compacted formulations using
multivariate and dimensional analysis. West Lafayette: Purdue
University; 2010.
19. Zinchuk A, Mullarney M, Hancock B. Simulation of roller
compaction using a laboratory scale compaction simulator. Int J
Pharm 2004;269:403–15.
20. Nkansah P, Wu S, Sobotka S, et al. A novel method for estimating
solid fraction of roller-compacted ribbons. Drug Dev Ind Pharm
2008;34:142–8.
21. Liu Z, Bruwer M, MacGregor J, et al. Scale-up of a pharmaceutical
roller compaction process using a joint-Y partial least squares model.
Ind Eng Chem Res 2011;50:10696–706.
22. Sheskey P, Pacholke K, Sackett G, et al. Effect of process scale-up
on robustness of tablets, tablet stability, and predicted in vivo
performance. Pharm Technol 2000;24:30–52.
23. Dehont F, Hervieu P, Jerome E, et al. Briquetting and granulation by
compaction new granulator-compactor for the pharmaceutical
industry. Drug Dev Ind Pharm 1989;15:2245–63.
24. McCabe WL, Smith JC, Harriott P. Unit operations of chemical
engineering. New York: McGraw-Hill, Inc.; 1993.
25. Rippe EG. Mixing. In: Lachman L, Liberman HA, Kanig JL, eds.
The theory & practice of industrial pharmacy. 3rd edn. Champaign:
Stipes Publishing, L.L.C.; 2004.
26. Guigon P, Simon O. Roll press design – influence of force feed
systems on compaction. Powder Technol 2003;130:41–8.
27. von Eggelkraut-Gottanka SG, Abu Abed S, Muller W, Schmidt PC.
Roller compaction and tabletting of St. John’s wort plant dry extract
using a gap width and force controlled roller compactor. II. Study
of roller compaction variables on granule and tablet properties by a
3(3) factorial design. Pharm Dev Technol 2002;7:447–55.
28. Keun HC, Ebbels TMD, Antti H, et al. Improved analysis of
multivariate data by variable stability scaling: application to NMR-
based metabolic profiling. Anal Chim Acta 2003;490:265–76.
29. Eriksson L, Johansson E, Kettaneh-Wold N, et al. Mulit- and
megavariate data analysis, Part I: basic principles and applications.
Umea, Sweeden: Umetrics; 2006.
30. Herting M, Kleinebudde P. Roll compaction/dry granulation: effect
of raw material particle size on granule and tablet properties. Int J
Pharm 2007;338:110–18.
31. Sun C, Himmelspach M. Reduced tabletability of roller compacted
granules as a result of granule size enlargement. J Pharm Sci 2006;
95:200–6.
32. Bacher C, Olsen P, Bertelsen P, et al. Improving the compaction
properties of roller compacted calcium carbonate. Int J Pharm 2007;
342:115–23.
33. Am Ende M, Moses S, Carella A, et al. Improving the content
uniformity of a low-dose tablet formulation through roller compac-
tion optimization. Pharm Dev Technol 2007;12:391–404.