International Journal of Pharmaceutics 521 (2017) 102109

Contents lists available at ScienceDirect

International Journal of Pharmaceutics

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

Articial neural network modelling of continuous wet granulation

using a twin-screw extruder
Saeed Shirazian* , Manuel Kuhs, Shaza Darwish, Denise Croker, Gavin M. Walker
Department of Chemical Sciences, Bernal Institute, University of Limerick, Limerick, Ireland

A R T I C L E I N F O A B S T R A C T

Article history:
Received 29 November 2016 Computational modelling of twin-screw granulation was conducted by using an articial neural network
Received in revised form 1 February 2017 (ANN) approach. Various ANN congurations were considered with changing hidden layers, nodes and
Accepted 2 February 2017 activation functions to determine the optimum model for the prediction of the process. The neural
Available online 3 February 2017 networks were trained using experimental data obtained for granulation of pure microcrystalline
cellulose using a 12 mm twin-screw extruder. The experimental data were obtained for various liquid
Keywords: binder (water) to solid ratios, screw speeds, material throughputs, and screw congurations. The
Computational modelling granulate particle size distribution, represented by d-values (d10, d50, d90) were considered the
Model predictive control
response in the experiments and the ANN model. Linear and non-linear activation functions were taken
Wet granulation
into account in the simulations and more accurate results were obtained for non-linear function in terms
ANN
Continuous pharmaceutical manufacturing of prediction. Moreover, 2 hidden layers with 2 nodes per layer and 3-Fold cross-validation method gave
the most accurate simulation. The results revealed that the developed ANN model is capable of predicting
granule size distribution in high-shear twin-screw granulation with a high accuracy in different
conditions, and can be used for implementation of model predictive control in continuous
pharmaceutical manufacturing.
2017 Elsevier B.V. All rights reserved.

1. Introduction considered as an advanced control strategy. MPC considers each

The development of continuous pharmaceutical manufacturing
has been a subject of great interest for the pharmaceutical
industry. Currently, manufacturing of solid-dosage pharmaceutical
formulations are carried out in batch-wise operation. In batch-
mode processing, each run that does not meet the requirements is
rejected which results in time and cost decits for the
manufacturing of pharmaceutical compounds. Continuous phar-
maceutical processing can overcome this drawback and therefore
offer more advantages compared to batch processing. In order to
develop continuous pharmaceutical manufacturing, each unit
operation in the manufacturing line should be inter-connected in
an appropriate way (Lee et al., 2015).
A powerful tool for development of continuous pharmaceutical
manufacturing is model predictive control (MPC), which is
Abbreviations: ANN, articial neural network; API, active pharmaceutical
ingredient; DEM, discrete element method; DoE, design of experiment; GSD,
granule size distribution; MCC, microcrystalline cellulose; MPC, model predictive
control; PBM, population balance model; PSD, particle size distribution; RMSE,
root-mean-squared error; SSE, sum of squared error.
* Corresponding author.
E-mail address: saeed.shirazian@ul.ie (S. Shirazian).
unit operation and takes a holistic view of the manufacturing line.
Therefore, each unit operation should be specied and its model
needs to be developed to implement MPC for the manufacturing
line. There are different processes in the manufacturing of solid-
dosage drugs such as milling, mixing, granulation, drying, and
coating. Granulation is the key step in manufacturing pharmaceu-
tical formulations, in which granules are produced from a ne
powder including an active pharmaceutical ingredient (API) and an
excipient. Moreover, wet granulation is the most complex unit
operation in pharmaceutical manufacturing since many
mechanisms are involved in the formation of granules from ne
powder (Rogers et al., 2013). Recently, twin-screw granulation has
gained a lot of attention over other granulation methods due to its
unique characteristics. The main advantage of twin-screw
granulation is that it is an intrinsic continuous process which
can promote development of continuous pharmaceutical
manufacturing (Seem et al., 2015). Other advantages of twin-
screw extruder in the pharmaceutical sector are its ability to mix
and react the feed materials, and its short residence time.
In order to implement MPC approach for continuous
manufacturing, a model of each process step is required which
can be done by mathematical or computational modelling. There

http://dx.doi.org/10.1016/j.ijpharm.2017.02.009
0378-5173/ 2017 Elsevier B.V. All rights reserved.
 S. Shirazian et al. / International Journal of Pharmaceutics 521 (2017) 102109 103

predict the granule size, liquid content, and porosity as a function

Nomenclature
d D-value of particle size distribution (micron)
f Predicted value
L/S Liquid to solid ratio
K Number of experimental subsets for ANN validation
n Number of experiments
R2 Coefcient of determination
x Linear combination of input factors
y Measured value
Subscript
i Experiment set
are different approaches for mathematical modelling of granula-
tion process including; population balance model (PBM), discrete
element method (DEM), and hybrid models. In hybrid models, both
PBM and DEM models are used to benet the advantages of both
methods in modelling of granulation. For PBM, the changes of
granule properties such as size distribution are calculated.
However, DEM tracks the motion of each individual particle along
the space, this is based on Newtons second law (Rogers et al.,
2013). Given that particle size is an important product character-
istic, these models primarily focus on predicting the particle size
distribution in granulation. However, in wet granulation the liquid
content and porosity are also very important for subsequent
processes, such as tabletting.
Some theoretical and experimental work using twin-screw
extruder has been carried out to simulate continuous wet
granulation. Several researchers have investigated predictive
modelling of wet granulation by using PBM and DEM modelling.
Barrasso et al. (Barrasso et al., 2015) developed a multi-
dimensional population balance model for the prediction of
granule properties in a twin-screw granulation. Lumped-parame-
ter and compartment approach were used for the numerical
solution of population balance equations. This model was able to
of process parameters.
Subsequently, Barrasso et al. (Barrasso et al., 2015 Barrasso and
Ramachandran, 2015) utilized the DEM approach for the estima-
tion of aggregation kernel in solution of population balance model.
The results showed a better prediction of particle size distribution
compared to semi-empirical aggregation kernel.
The results of mechanistic models developed for twin-screw
granulation revealed that these mechanistic models are quite slow
for the use of MPC in the development of continuous pharmaceu-
tical manufacturing. However, these mechanistic models can be
used for the process design and optimization (Barrasso et al., 2015;
Kumar et al., 2013, 2015).
The main disadvantage of aforementioned mechanistic models
is that it is not fast enough to be used as a model for the
development of MPC approach in continuous manufacturing. In
MPC, the model of process should be able to run within a few
seconds in order to predict the future behaviour of the process.
Therefore, these mechanistic models fail to be applied for MPC, and
in fact faster models are required for industrial applications.
Recently, some researchers have tried to reduce the solution time
for mechanistic model by utilizing ANN. Barrasso et al. (Barrasso
et al., 2014) developed a hybrid model by coupling PBM and
articial neural network (ANN) for describing wet granulation. The
main aim of a hybrid model is to make the solution time faster. The
results showed that ANN is capable of simulation for wet
granulation process, although it does not look at the mechanisms
associated with the granulation. A list of different mechanistic and
hybrids models applicable to wet granulation are reported by
Kumar et al. (Kumar et al., 2013; Rogers et al., 2013).
Data-driven models have proved to be robust and efcient for
the application of simulation and prediction of pharmaceutical
processes. An important class of data-driven models is articial
neural network (ANN) which is powerful in process prediction
(Kazemi et al., 2016; Puri et al., 2016). ANN is usually used for
modelling of complex processes in which mechanistic models fail
to predict the process or are computationally expensive.
ANN has been used for predication of some pharmaceutical
processes such as milling. Kazemi et al. (Kazemi et al., 2016)

Table 1
Design of experiments for ANN simulation of wet granulation.

Run L/S Screw speed (rpm) Powder ow rate (g/h) Screw conguration d10 (Micron) d50 (Micron) d90 (Micron)
1 0.54 64 49.75 2 kneading zones 15.07 88.535 328.8
2 0.94 200 98 2 kneading zones 151.35 417.2 1004.15
3 1.22 86 98.2 2 kneading zones 357.35 759.9 1202
4 0.54 200 69.33 2 kneading zones 23.885 130.2 360.3
5 1.22 200 61.47 2 kneading zones 307.15 689.95 1038.7
6 1.22 50 82 2 kneading zones 445.65 767 1105
7 1.21 200 49.4 1 kneading zone 385.75 1008.2 1303
8 0.78 200 49.75 1 kneading zone 65.195 271.6 653.5
9 1.21 115 49.5 1 kneading zone 455.15 1052.65 1313.5
10 0.49 185 67.55 1 kneading zone 14.605 67.46 342
11 0.65 50 97.6 1 kneading zone 10.57 76.865 342.6
12 0.52 50 97.05 1 kneading zone 8.33 35.805 300.5
13 1.22 50 82 1 kneading zone 177.5 440.55 831.8
14 0.55 200 98.6 conveying elements only 25.06 113.25 291.35
15 1.22 200 98.15 conveying elements only 332.55 860.65 1250
16 0.76 200 99.2 conveying elements only 66.91 241.2 641.45
17 0.59 85 51 conveying elements only 35.02 164.7 384.45
18 1.12 72 53.25 conveying elements only 186.775 367.975 820.125
19 0.48 50 49.8 conveying elements only 18.42 102.745 341.2
20 0.50 200 95.05 2 kneading zones with cutting elements 7.585 22.425 113.75
21 1.20 58 82 2 kneading zones with cutting elements 330.7 865.6 1273.5
22 0.66 200 49.65 2 kneading zones with cutting elements 35.605 188 490.7
23 0.54 50 61.45 2 kneading zones with cutting elements 14.875 87.55 342.7
24 1.22 200 97.38 2 kneading zones with cutting elements 325.4 763 1154
104 S. Shirazian et al. / International Journal of Pharmaceutics 521 (2017) 102109

Fig. 1. Screws used for wet granulation experiments.

applied the computational intelligence-based models for simula-
tion of oscillating milling. Genetic programming was coupled with
ANN in order to predict the particle size during the milling process.
The model capability in the prediction of d50 values conrmed that
the ANN is a powerful tool for simulation of pharmaceutical
processes specically for complex unit operations.
Therefore, there is a requirement for the development of a
robust, reliable and fast model to be used for the prediction of high-
shear wet granulation. The main aim this study is to develop a
methodology to predict the wet granulation process in an efcient
manner. The predictive method is based on the simulation of this
process using articial neural network. An effort is made to build a
robust, reliable, and fast model of processing which can be used for
implementation of model predictive control in pharmaceutical
manufacturing. Both linear and non-linear models are investigated
at different network topologies to nd the optimum network for
wet granulation. The model predictions are then compared with
experimental data obtained on a 12 mm twin-screw extruder. To
the best of our knowledge, there is no work on ANN modelling of
high shear twin-screw granulation of pharmaceutical formula-
tions.
2. Experiments
Wet granulation experiments were conducted via a twin-screw
extruder, ZE12 (ThreeTec, Switzerland) with screw diameter of
12 mm and length-to-diameter ratio 40:1. Microcrystalline cellu-
lose (MCC, Avicel pH 101 Pharmatrans) was used as a formulation
model and water as a binder throughout the experiments. A
peristaltic pump (Watson Marlow 520SN) was used for adding
water and a gravimetric feeder (Three-Tec, Switzerland) for feeding
the powder.
Table 2
Comparisons between different ANN topologies in terms of nodes and activation
functions.
The experiments were designed based on a Custom-designed
DoE considering feed ow rate, screw speed, liquid to solid ratio (L/
S), and screw congurations as factors and particle size distribu-
tion in terms of d-values as a response. The designed experiments
and the obtained d-values of particle size are listed in Table 1. Fig. 1
illustrates the photo of two screws used in this work, and the screw
conguration can be seen schematically in the Supplementary le.
The obtained granules were dried overnight in an oven at 40
C
and then characterized using dry powder dispersion laser
diffraction (Microtrac S3500). As the size limit for Microtrac
S3500 is 1.4 mm, a sieve was used prior to laser diffraction to
measure the mass of material greater than 1.4 mm. Consequently
the recorded d-values of particle size are for the mass fraction
smaller than 1.4 mm. As the% mass fraction > 1.4 mm was usually
less than 1%, this was considered adequate.
3. Modelling
3.1. ANN conguration
For the development of the ANN model, in addition to screw
conguration, three process parameters including liquid to solid
ratio (L/S), screw speed, and powder ow rate were considered as
inputs. On the other hand, d-values of particle size distribution
including d10, d50, and d90 were considered as response
parameters. Basically, ANN models consist of three layers; i.e.
input, hidden and output layers (Kazemi et al., 2016). Important
parameters in development of ANN include: validation method,
activation function, hidden layers, and number of nodes in each
layer of network. These parameters were taken into account to nd
out the optimum ANN topology for predicting the process. Finding
the optimum ANN topology is challenging since there is no
straightforward procedure for this, and usually a trial and error

No. First layer Second layer Average RMSE

Linear Non-linear Linear Non-linear

1 1 1 1 1 102.98
2 1 1 1 2 84.68
3 1 1 2 1 105.95
4 1 1 2 2 80.06
5 1 2 1 1 82.26
6 1 2 1 2 71.37
7 1 2 2 1 61.04
8 1 2 2 2 93.16
9 2 1 1 1 78.12
10 2 1 1 2 89.85
11 2 1 2 1 65.44
12 2 1 2 2 93.83
13 2 2 1 1 60.57
14 2 2 1 2 65.31
15 2 2 2 1 85.69
16 2 2 2 2 48.95
17 0 1 0 1 89.40
18 0 2 0 2 71.98
19 1 0 1 0 104.60
20 2 0 2 0 107.10
Fig. 2. Topology of boosted ANN.
 S. Shirazian et al. / International Journal of Pharmaceutics 521 (2017) 102109 105

approach is utilized for building an appropriate model. Kfold show the model accuracy. Coefcient of determination which is
method was used as validation in which K = 3 showed the best dened as (Barrasso et al., 2015):
results. The latter means that the whole experimental measure- X 2
ments are divided into three sections, and each set of data is 2
f i  yi
R 1  Xi 1
evaluated for validation and the best t is selected. Therefore, y  yi 2
i i
among 24 experimental runs listed in Table 1, 16 runs were used for
training of the network and 8 runs were used to validate the built where R2 refers to coefcient of determination. f and y refer to
ANN model. JMP Pro 12 software was used for ANN modelling and predicted and observed values, respectively. i also refers to the set
analysis of the results. The software is capable of considering of experimental runs.
categorical factors as input for ANN tting. Moreover, the goodness of ANN t is reported as root-mean-
squared error which is dened as follows:
3.2. Model assessment sX
2
f  yi
i i
RMSE 2
The developed ANN model is used for prediction of d-values in n
twin-screw granulation in which the predictability of model is where n refers to the number of measurements
assessed through comparing the predicted values with observed
(experimental) values. Two statistical parameters were used to

Fig. 3. Experimental versus predicted d-values.

106 S. Shirazian et al. / International Journal of Pharmaceutics 521 (2017) 102109
4. Results and discussion
4.1. Experimental results
The experimental data obtained from measuring d-values of
particle size distribution are reported in Table 1. As seen, 24
experiments were run and d-values for each run were measured
and reported as d10, d50, and d90. Rather than tting the ANN
model to all d-values and particle sizes, these three d-values were
chosen to t the model to avoid the complexity of objective
function in parameter estimation. Moreover, for industrial
applications the granule size range are of great importance rather
than the whole particle size distribution (Barrasso et al., 2015).
4.2. Determination of hidden layers and nodes
Basically, ANN consists of layers and nodes in which increasing
hidden layers and notes would make the ANN model more exible.
In this work, two hidden layers and up to four nodes per layer were
considered for the development of ANN. However, more nodes and
layers could be considered but the solution time would increase
considerably. Since the main aim of ANN model in this work is to be
used for MPC application, the solution time is the most important
criterion for ANN development. For each hidden layer, an
activation function is required to estimate the value of response.
The activation function applies at the nodes of hidden layers and is
a transformation of a linear combination of input variables. For this
work, two different activation functions are considered including
linear and non-linear. For linear activation function, the linear
combination of input variables are not transformed whereas for
non-linear, the hyperbolic tangent function is used as follows
Fig. 4. Residuals of ANN ttings for different d-values.
(SAS_Institute, 2016):
e2x  1
3
e2x 1
where x is a linear combination of input variables
Up to four nodes were considered for each layer (two linear and
two non-linear) and the results were compared in terms of average
RMSE for overall training and predictions. Increasing the number of
nodes greater than two increased the solution time and no
signicant improvement was observed. Since the objective
function may have different local minima, multi-start technique
was used for parameter estimation and tting. This means that the
ANN tries to estimate the tting parameters by different initial
points in order to nd the global minimum of objective function.
The iteration parameter was set to be 10 for multi-start technique
(Kazemi et al., 2016).
The results are listed in Table 2. As shown, the ANN topology
with two nodes for both hidden layers considering a combination
of non-linear and linear activation functions shows the best
performance in tting the experimental data. However, the results
are not still satisfactory given that the model could be used as
reliable tool for prediction of process.
4.3. Boosting the network
The ANN developed so far with two hidden layers and two
nodes per layer has shown to be weak for accurate prediction of
granule size in twin-screw granulation. Considering this, an
attempt was made to build a more robust and reliable model to
predict the process. To do this, the trained networks were made
more robust by boosting approach. Boosting is used of construct
 S. Shirazian et al. / International Journal of Pharmaceutics 521 (2017) 102109 107

larger additive ANN models by tting a sequence of smaller

models. Basically, in boosting a model, each of the smaller models
which is already built is t on the scaled residuals of the previous
model. Afterward, the models are combined to form the larger nal
models. It should be pointed out that boosting uses validation to
assess how many component models are required to t. Boosting is
often faster than tting a single large ANN model. However, the
base model should consist of a 1 to 2 node model, or else the
advantage of faster tting may be lost if a large number of models is
specied (SAS_Institute, 2016). The boosting was performed for all
networks consisting of a maximum of 2 nodes per layer. (see
Table 2). For boosting the network, the maximum number of
models were set to be 15 and the best results were obtained with
the network consisting of 2 nodes per layer with non-linear
activation function (No. 18 in Table 2). Finally, the model consists of
15  2 = 30 nodes in each hidden layer with non-linear activation
function. The topology of boosted ANN is shown in Fig. 2.

4.4. Model validation

The boosted ANN is rst trained/calibrated using experimental

data (16 runs) in order to estimate the unknown tting parameters.
Once the model is calibrated, it is then used for the prediction of
experimental data (8 runs). The trained and predicted results
developed by ANN were compared with experimental data to
assess the accuracy of the ANN. The results are shown in Figs. 3 and
4. The statistical parameters of ttings are also shown in Fig. 5. In
Fig. 3, predicted values for d10, d50, and d90 are compared with
experimental values for both training and validation. In Fig. 4
residuals of ttings are shown for training and validation also. As
shown, the predicted results match rather well with experimental
results for different d-values, however a slight deviation is
observed for validation d90 values. The coefcient of determina-
tion in all cases equal 0.99 except for training of d90. Therefore, the
results reveal that boosting is a powerful tool for making more
robust and reliable ANN models. Furthermore, the developed ANN Fig. 5. Fitting parameters of training and validation.
is very fast and can predict the effect of various parameters on
granule size with high accuracy. This is a promising approach for congurations on the granule properties. The effect of three
development of MPC for continuous pharmaceutical manufactur- process parameters and screw conguration on granule size are
ing. illustrated in Fig. 6. As shown, the most paramount parameter is
The model ndings reveal that the developed ANN is a liquid to solid ratio in which increasing L/S increases the granule
promising tool for modelling of twin-screw granulation of size signicantly. This behaviour is attributed to the increasing
pharmaceutical formulations. However, the results of ANN aggregation rate with L/S so that larger particles/granules are
modelling are compared with other modelling approaches used formed during the process. However, this is not always favourable
for twin-screw granulation. The main modelling approach used for in wet granulation as it may result in formation of over-sized
twin-screw granulation is population balance model in which the granules that must be milled after granulation. Therefore, the
change of particle properties are tracked during the process. liquid addition ow rate should be closely/accurately controlled. It
Barrasso et al. (Barrasso et al., 2015, 2014) developed multi- is also observed that increasing powder ow rate results in lower
dimensional population balance models for twin-screw granula- particle size. The latter could be due to reduction of residence time
tion. The model could predict the granule size distribution, liquid of particles inside the granulator with increasing powder ow rate.
content, and porosity. The modelling results were compared with Therefore, the particles do not have enough time to get wetted and
experimental data obtained from a 16 mm extruder. The model was agglomerate inside the extruder.
further developed by coupling PBM with DEM and ANN. The Screw speed does not have considerable effect on granule size,
comparisons are shown in Table 3 in terms of overall R2 in however increasing screw speed results in a small increase in
prediction of granule size. It is evident that the developed ANN is particle size which could be due to a higher mixing rate between
more accurate than other developed mechanistic models devel- the powder and liquid binder. Finally, the largest particles are
oped to date for prediction of twin-screw granulation. Moreover, observed for screw conguration with two kneading zones. With
ANN method is much faster than mechanistic models for the
prediction of process. In the following sections, the effect of
Table 3
different parameters on granule size are investigated by using the Comparisons between ANN results developed in this work and other modelling
developed ANN. results in literature.

Model R2 Ref.
4.5. Effect of process parameters on granule size
Multi-dimensional PBM 0.805 Barrasso et al. (2015)
Coupled ANN-PBM-DEM 0.97 Barrasso et al. (2014)
Once the ANN model is validated, it can be used as a predictive ANN 0.99 (This work)
tool to evaluate the effect of various process parameters and screw
108 S. Shirazian et al. / International Journal of Pharmaceutics 521 (2017) 102109
Fig. 6. Effect of process parameters and screw congurations on granule d-values
calculated by developed ANN.
two kneading zones, there is more mixing between powder and
liquid binder and the number of collision between particles
increases. Therefore this results in the formation of larger particles.
Fig. 7. Effect of process parameters and screw congurations on span of particle size distribution.
4.6. Effect of process parameters on span of GSD
The span of GSD is important for the analysis and investigation
of granule size distribution (GSD) obtained from continuous wet
granulation. Span of size distribution is a parameter which
indicates the width of size distribution. It shows if the granules
are distributed in a narrow or broad way and is important for
tabletting since the granule size affects the tablet properties, for
example dissolution and disintegration. The span of granule size
distribution can be formulated as:
d90  d10
Span 4
d50
Therefore, span of particle size distribution indicates how far
apart the 10 percent (d10) and 90 percent (d90) points are,
normalized with the midpoint of PSD (d50). In granulation of
pharmaceutical formulations, a narrow particle size is desired.
Given that the values of d90, d10, and d50 from the granule size
distribution are calculated, the span of granule size distribution for
each experimental run can be easily calculated by using Eq. (4). The
effect of process parameters and screw congurations on span of
GSD for all 24 experimental runs are illustrated in Fig. 7. The results
are shown as contour of span (response variable) versus screw
speed and liquid to solid ratio. Four screw congurations are
 S. Shirazian et al. / International Journal of Pharmaceutics 521 (2017) 102109
considered to investigate the effect of this parameter on span. As
shown, the most important parameter affecting the span values is
L/S ratio in which the span values change from 1 up to 7. The reason
for this behaviour could be due to the liquid distribution during the
process. At low L/S ratio, the liquid binder is not distributed well
and some particles are not wet enough to form granules. The latter
would cause formation of a non-uniform or broad granule
distribution containing ne powder, medium-sized and large
granules. On the other hand, it is evident that increasing screw
speed slightly decreases span values. This behaviour could be
attributed to mixing the powder and liquid binder thoroughly by
increasing the screw speed in which results in narrower granule
size distribution.
The inuence of screw conguration reveals that the lowest
span values (narrower GSD) are obtained for screw with 2
kneading zones. The latter could be due to well mixing of liquid
binder and powder and higher formation of granules during the
process.
5. Conclusions
Computational intelligent modelling of wet granulation was
studied in this work. An articial neural network (ANN) was
developed and trained to predict the granule size distribution.
The experimental work consists of wet granulation of microcrys-
talline cellulose using a twin-screw extruder. Process parameters
including liquid to solid ratio, screw speed, and powder ow rate
were considered as inputs. Also, screw conguration was
considered as input by changing the screw conguration at four
levels. On the other hand, d10, d50, and d90 of particle size
distribution were considered as outputs. Various ANN topologies
were investigated to nd the optimum hidden layers and nodes.
The best results were obtained for 2 hidden layers containing 2
nodes per layer with non-linear activation function for both
layers. Moreover, the network was boosted by building larger
models up to 15. The comparison between model predictions and
experimental data showed great corrilation with R2 = 0.99. The
investigation on the effect of various parameters on the granule
size indicated that L/S is the most important parameter which can
cause formation of over-sized granules. The developed modelling
methodology is fast and reliable and can be used for implementa-
tion of model predictive control in developing continuous
pharmaceutical manufacturing.
109
Acknowledgments
This work was funded by Science Foundation Ireland (SFI)
under the following grants:
Model Predictive Control of Continuous Pharmaceutical Pro-
cesses(13/IA/1980).
The equipment used in this work was funded by Science
Foundation Ireland (SFI) under the following grants:
Model Predictive Control of Continuous Pharmaceutical Pro-
cesses(13/IA/1980).
Appendix A. Supplementary data
Supplementary data associated with this article can be found, in
the online version, at http://dx.doi.org/10.1016/j.
ijpharm.2017.02.009.
References
Barrasso, D., Ramachandran, R., 2015. Multi-scale modeling of granulation
processes: bi-directional coupling of PBM with DEM via collision frequencies.
Chem. Eng. Res. Des. 93, 304317.
Barrasso, D., Tamrakar, A., Ramachandran, R., 2014. A reduced order PBMANN
model of a multi-scale PBMDEM description of a wet granulation process.
Chem. Eng. Sci. 119, 319329.
Barrasso, D., El Hagrasy, A., Litster, J.D., Ramachandran, R., 2015. Multi-dimensional
population balance model development and validation for a twin screw
granulation process. Powder Technol. 270, 612621 Part B.
Kazemi, P., Khalid, M.H., Szlek, J., Mirti9
c, A., Reynolds, G.K., Jachowicz, R., Mendyk, A.,
2016. Computational intelligence modeling of granule size distribution for
oscillating milling. Powder Technol. 301, 12521258.
Kumar, A., Gernaey, K.V., Beer, T.D., Nopens, I., 2013. Model-based analysis of high
shear wet granulation from batch to continuous processes in pharmaceutical
productiona critical review. Eur. J. Pharm. Biopharm. 85, 814832.
Kumar, A., Gernaey, K.V., De Beer, T., Nopens, I., 2015. Model-based characterisation
of twin-screw granulation system for continuous solid dosage manufacturing.
In: Krist, V., Gernaey, J.K.H., Raqul, G. (Eds.), Computer Aided Chemical
Engineering. Elsevier, pp. 21652170.
Lee, S.L., OConnor, T.F., Yang, X., Cruz, C.N., Chatterjee, S., Madurawe, R.D., Moore, C.
M.V., Yu, L.X., Woodcock, J., 2015. Modernizing pharmaceutical manufacturing:
from batch to continuous production. J. Pharm. Innov. 10, 191199.
Puri, M., Solanki, A., Padawer, T., Tipparaju, S.M., Moreno, W.A., Pathak, Y., 2016.
Chapter 1Introduction to Articial Neural Network (ANN) as a Predictive Tool
for Drug Design, Discovery, Delivery, and Disposition: Basic Concepts and
Modeling, Articial Neural Network for Drug Design, Delivery and Disposition.
Academic Press, Boston, pp. 313.
Rogers, A., Hashemi, A., Ierapetritou, M., 2013. Modeling of particulate processes for
the continuous manufacture of solid-based pharmaceutical dosage forms.
Processes 1, 67.
SAS_Institute, 2016. Using JMP1 12.2.0, Cary, NC.
Seem, T.C., Rowson, N.A., Ingram, A., Huang, Z., Yu, S., de Matas, M., Gabbott, I.,
Reynolds, G.K., 2015. Twin screw granulationa literature review. Powder
Technol. 276, 89102.