European Journal of Pharmaceutics and Biopharmaceutics xxx (2014) xxxxxx

Contents lists available at ScienceDirect

European Journal of Pharmaceutics and Biopharmaceutics

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

Research paper

Analysing drying unit performance in a continuous pharmaceutical

manufacturing line by means of mass Energy balances
Sverine Thrse F.C. Mortier a,b, Krist V. Gernaey c, Thomas De Beer b,1, Ingmar Nopens a,
a
Department of Mathematical Modelling, Statistics and Bioinformatics, Ghent University, Ghent, Belgium
b
Department of Pharmaceutical Analysis, Ghent University, Ghent, Belgium
c
Department of Chemical and Biochemical Engineering, Technical University of Denmark, Kgs. Lyngby, Denmark

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

Article history: The current trend in the pharmaceutical industry to move from batch-wise to continuous production
Received 8 July 2013 processes strengthens the need for monitoring and controlling the process in-line. The ConsiGma
Accepted in revised form 20 December 2013 continuous tableting line collects data of the different subunits in real-time, but these are not really used.
Available online xxxx
In this paper the data of the six-segmented uidized bed dryer in the line are used for the development
and evaluation of a mass and energy balance. The objectives are multiple: (1) prediction of the moisture
Keywords: content of the granules leaving the dryer solely based on the currently logged data and (2) prediction of
In-line univariate measurements
the gas outlet temperature to check the mass balances. Once a validated system is established the gas
Mass balance
Energy balance
temperature in different horizontal sections of the drying unit can be predicted. Calculations are also
Process monitoring used to identify errors in the system and to propose alternative sensor locations. A calibration is
Pharmaceutical engineering performed in order to predict the evaporation rate. The balances were able to predict both the moisture
Drying process content of the granules at the end of the drying process and the gas outlet temperature quite accurately.
Combining the gathered information with the height of the bed in the uidized bed can be used to predict
the gas temperature in different horizontal sections of the dryer. An extra sensor measuring the gas tem-
perature and the humidity at the wet transfer line would increase the accuracy of the calculations. An
extra gas velocity sensor at the outlet would be useful to incorporate an extra supervision of the
calculations.
2013 Elsevier B.V. All rights reserved.

1. Introduction adjustment of sensitive process variables would be a major step

1.1. State of the art process monitoring and controlling in
pharmaceutical industry
One recent trend in pharmaceutical manufacturing is the shift
from batch-wise to continuous production processes, not only in
the synthesis of small molecules [1,2], but also in formulation
[3,4]. The traditionally applied batch processes mostly rely on
off-line post-process time-consuming and less efcient laboratory
testing to evaluate the quality of the product [5]. Continuous pro-
duction processes, relying on in-line measurements and real-time
Corresponding author. BIOMATH, Department of Mathematical Modelling,
Statistics and Bioinformatics, Faculty of Bioscience Engineering, Ghent University,
Coupure Links 653, 9000 Ghent, Belgium. Tel.: +32 (0)9 264 61 96; fax: +32 (0)9 264
62 20.
E-mail addresses: severine.mortier@ugent.be (S.T.F.C. Mortier), KVG@kt.dtu.dk
(K.V. Gernaey), thomas.debeer@ugent.be (T. De Beer), ingmar. nopens@ugent.be
(I. Nopens).
URL: http://biomath.ugent.be (S.T.F.C. Mortier).
1
Shared last authorship.
forward towards more efcient production routines. A batch is
well dened, and as such it is easy to perform a quality assurance,
since a batch can simply be accepted or rejected based on a quality
assessment. But since quality is only checked after the process in
traditional manufacturing, it is impossible to control the process
on the basis of such measurements. A bad batch is simply lost.
However, monitoring the process during operation allows interfer-
ing during production (i.e. process control) which is more econom-
ical as the loss in off-spec product is limited [5,1,6]. A general
problem within the pharmaceutical industry is the strict regula-
tions. Once a process is approved by the regulatory authorities, it
is generally accepted that it is almost impossible to change some-
thing in the way of processing without re-documentation and re-
approval [1,5]. The Food and Drug Administration (FDA) intended
to promote innovation by publication of the Process Analytical
Technology (PAT) guidance. An important concept is Quality By De-
sign (QbD), dened as a systematic approach to development that
begins with predened objectives and emphasizes product and
process understanding and process control, based on sound science
and quality risk management [7].

0939-6411/$ - see front matter 2013 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.ejpb.2013.12.014

Please cite this article in press as: S.T.F.C. Mortier et al., Analysing drying unit performance in a continuous pharmaceutical manufacturing line by means of
mass Energy balances, Eur. J. Pharm. Biopharm. (2014), http://dx.doi.org/10.1016/j.ejpb.2013.12.014
2 S.T.F.C. Mortier et al. / European Journal of Pharmaceutics and Biopharmaceutics xxx (2014) xxxxxx

The equipment to produce tablets in a continuous mode is lim- Hence, with no extra cost these data can potentially be sufcient
ited up to now. The ConsiGma continuous tableting line (Col- for monitoring and controlling the process, hereby guaranteeing
lette, GEA Pharma Systems, Wommelgem, Belgium) enables the product quality.
production of tablets from powders in 20 min. It consists of the
ConsiGma high shear granulator and uidized bed dryer, com-
1.2. Error propagation of measured model input variables into model
bined with the GEA Courtoy MODUL P rotary tablet press [8].
output variables
Ldige Process Technology has the CWG line which is a continuous
wet granulation line and is also a complete system from raw mate-
Any measured variable is prone to measurement errors and
rial dosing till tabletting [9,5]. In this work, the focus is on the con-
contains an inherent uncertainty. This will have an impact when
tinuous uidized bed drying system of the ConsiGma continuous
using these measurements in further calculations (e.g. mass bal-
tabletting line. Real-time on-line measurements are continuously
ance checks, computing control actions, etc.). An error propagation
logged. The logged data will be processed using a mass and energy
enables the operator to quantify the cumulative effect of errors in
balance in order to monitor the process in real time.
the input data. Another option would be the use of more accurate
Several in-line measurement tools are available for the moni-
sensors to minimise the error on calculated output. Moreover, the
toring and control of pharmaceutical manufacturing processes.
quest for more accurate sensors results often in more expensive
In-line measurements provide real-time data about a process,
sensors, which is of course important in a production process.
and can thus be used as inputs to control the process operation.
In control applications, one can use data reconciliation tech-
Process analysers can be divided into the univariate (such as tem-
niques in order to reduce the impact of erroneous data [16],
perature, pressure, and gas velocity) and multivariate process mea-
whereas uncertainty propagation can be used for calculations
surements, which provide biological, physical, chemical, etc.
based on the raw data. Error propagation is a methodological tool
characteristics of the material. The determination of the moisture
to quantify the uncertainty on a model output, given the source of
content of material during processing of solids can be achieved
uncertainty and its specications (e.g. variance) are known. Typi-
using several techniques. However, a lot of techniques that are cur-
cally this is done by construction of output uncertainty boundaries.
rently available are not appropriate for use under harsh process
The uncertainty on parameters and variables is hereby propagated
conditions. Time domain reectometry and transmissometry,
to the output. Different methods are available depending on the
capacitance probes, electrical resistance measurements, micro-
model (non-)linearity. Simple linear models can use classical error
wave spectrometry are all sensitive to the bulk density of the mon-
propagation techniques. More complex models can make use of
itored system [10]. Other techniques are not able to measure the
linear error propagation or differential analysis. Uncorrelated in-
moisture content accurately due to the presence of hydrogen
puts for the uncertainty propagation have an advantage as the
atoms in any other substance than water. This is the case for Nucle-
covariance can be discarded from the evaluation [17]. For complex
ar Magnetic Resonance (NMR) or neutron absorption. Besides
non-linear models, a Monte Carlo based method can be used [17].
moisture content, Near Infra-red Spectroscopy (NIRS) is also able
The latter method uses repeated calculation of the output, while
to provide information related to particle size. An advantage of
varying the input [18]. In a rst step, all sources of uncertainty
the technique is its non-destructive nature. Moreover, Near Infra-
should be identied, whereas in the next step these uncertainties
red Spectroscopy (NIRS) is available against a reasonable cost, is
should be quantied [19]. The uncertainty limits can be based on
fast, and is relatively immune to changes in bulk density caused
the measurement error of the used sensors, or by using previously
by granule growth [11,12]. Frake et al. described Near Infra-red
quantied condence intervals. The input is then randomly se-
Spectroscopy (NIRS) as an in-line tool to obtain the granule mois-
lected from its error Probability Density Function (PDF) using a
ture content and particle size of a uidized bed granulation pro-
sampling scheme (e.g. random or Latin Hypercube). The calculated
cess. It allows modication of process conditions during
outputs will also be distributed, which allows dening an output
operation as well as end-point identication [11]. However, atten-
uncertainty distribution. The drawback of the Monte Carlo method
tion should be paid to prevent the formation of solid deposits on
is the number of iterations (i.e. Monte Carlo shots) that should be
the probe [10]. Chablani et al. used Near Infra-red Spectroscopy
carried out (in the order of hundreds to thousands depending on
(NIRS) to measure the granules moisture content produced using
the number of sources of uncertainty considered) before the vari-
the ConsiGma. The application of Near Infra-red Spectroscopy
ance of the output distribution is known accurately [18]. This
(NIRS) as a Process Analytical Technology (PAT) tool during contin-
makes the technique computationally expensive, especially for
uous manufacturing is promising [13]. Besides Near Infra-red Spec-
complex models.
troscopy (NIRS) spectroscopy, several other Process Analytical
Technology (PAT) tools have been examined for moisture content
assessment during processing of pharmaceutical solids. Porthogh- 2. Problem statement and objectives
ese et al. described the use of triboelectric probes to measure solid
moisture content during uidization. Being inexpensive, very sen- As indicated in Section 1.1 process monitoring is mostly done
sitive probes without requiring any maintenance makes them by using off-line time-consuming measurements or by implemen-
attractive [10]. Microwave Resonance Technology (MRT) has been tation of extra sensors. The automatically collected in-line real-
implemented in uidized bed dryers and provides precise and time univariate data is nowadays logged, but is typically not used
accurate results for the granule moisture content independent of for process monitoring and/or controlling.
product density [14]. The technique was validated using reference The main objective of this work is the use of a combined mass
methods (Loss On Drying (LOD)/Infra-red light) and Karl Fisher and energy balance to estimate the moisture content of granules
titration. Wang et al. used Electrical Capacitance Tomography at the end of the drying process in a six-segmented uidized bed
(ECT) for the on-line measurement of the solid moisture content dryer, being part of a full continuous pharmaceutical from pow-
in a batch uidized bed dryer. The measured value can be used der-to-tablet manufacturing line. The energy balance is made to
for implementation of a closed-loop control strategy [15]. Includ- compare the experimental gas outlet temperature with the calcu-
ing in-line monitoring obviously requires extra probes and soft- lated one as a validation of the balances. In the calculations error
ware for data processing, and as such extra costs. In some cases propagation will be accounted for in order to consider uncertainty.
the pharmaceutical production equipment already has sensors in Once the balances are validated a prediction of the gas temperature
place, without using them for specic monitoring or control tasks. in different horizontal sections of each drying segment can be

Please cite this article in press as: S.T.F.C. Mortier et al., Analysing drying unit performance in a continuous pharmaceutical manufacturing line by means of
mass Energy balances, Eur. J. Pharm. Biopharm. (2014), http://dx.doi.org/10.1016/j.ejpb.2013.12.014
 S.T.F.C. Mortier et al. / European Journal of Pharmaceutics and Biopharmaceutics xxx (2014) xxxxxx
made. Furthermore the study can be used to identify errors in the
system and to propose an alternative sensor location.
3. Materials and methods
3.1. Experimental data collection
For the data collection, the ConsiGma continuous from-pow-
der-to-tablet production line from GEA Pharma Systems (Col-
lette, Wommelgem, Belgium) was used. The continuous line
consists of three parts: a continuous twin screw granulator (high
shear), followed by a six-segmented uidized bed dryer system
and a discharge system [20,21].
The automatically in-line logged data are available from two
different production lines, one located at the Laboratory of Phar-
maceutical Process Analytical Technology (Ghent University,
Ghent, Belgium) and the other located at GEA Pharma Systems
(Wommelgem, Belgium). Several datasets were available but only
the results of two datasets (Table 2) are presented in this paper.
The results of other datasets were similar and did not provide extra
information, and were therefore not included in this paper.
For dataset A, only one segment was lled with wet granules,
these data were collected at the system located in Ghent. The for-
mulation of the dry premix consisted of theophylline anhydrate
(Farma-Quimica sur SL, Malaga, Spain) (30%, w/w), lactose mono-
hydrate 200 M (DMV fonterra) (67.5%, w/w) and polyvinylpyrroli-
done (PVP) (Kollidon 30, BASF, Burgbernheim, Germany) (2.5%, w/
w). This premix was granulated with a 0.5% (w/v) sodiumlaurylsul-
fate (SLS)-solution (Fagron, Waregem, Belgium) in distilled water
at a barrel temperature of 25 C. sodiumlaurylsulfate (SLS) was
added to improve the wettability of the dry premix. The screw
speed was held constant at 950 rpm, the powder mass ow at
17.5 kg/h and the liquid mass ow at 32 g/min. The ConsiGma
standard screw conguration was used. The data were continu-
ously logged for several hours (during 8 h), whereby the dryer is
loaded with wet granules at certain moments, and is empty in be-
tween. One single experiment is dened between two periods
where the dryer is empty, and an example is provided in Fig. 4.
These experiments are executed over a period of several days (each
day one run of around 8 h).
Dataset B contains data of the full operating system, where all
six segments were continuously lled with wet granules. As such
a real-time situation is mimicked. The data are collected from
the system situated at GEA Pharma Systems. The formulation con-
sisted of two active ingredients (a slightly soluble Active Pharma-
ceutical Ingredient (API) in a high dose and a soluble Active
Pharmaceutical Ingredient (API) in a lower dose), powdered cellu-
lose, maize starch, pregelatinized starch and sodium starch glyco-
late. The dry ingredients were granulated with distilled water at a
barrel temperature of 25 C and a screw speed of 900 rpm. The
powder mass ow was kept at 20 kg/h, and the liquid mass ow
at 50 g/min. During tableting magnesium stearate was added as lu-
bricant. Johnson & Johnson (Janssen-Cilag, Italy) delivered all
ingredients for the production.
For both datasets the residual moisture content after drying was
measured using Karl Fisher titration using a V30 volumetric KF titra-
tor (Mettler Toledo, USA). Prior to the titration of the granules, they
are stirred and dissolved (methanol (Hydranal, Sigma Aldrich, Ger-
many)) during 5 min. As Karl Fisher titration measures the combi-
nation of the bound water and the free water, the moisture
content of the dry premix should be subtracted from the measured
data to coincide with the calculated moisture content based on the
mass balance (referred to as the corrected value (X KF;corr )). For data-
set A the moisture content at the end of the drying process (X end;KF ) is
measured for each experiment, so just before the dryer is emptied.
Please cite this article in press as: S.T.F.C. Mortier et al., Analysing drying unit performance in a continuous pharmaceutical manufacturing line by means of
mass Energy balances, Eur. J. Pharm. Biopharm. (2014), http://dx.doi.org/10.1016/j.ejpb.2013.12.014
3
3.2. Processing of experimental data
The experimental data were processed off-line, i.e. the data
were extracted from the production line via a SCADA system and
made available in Excel worksheets. The mass and energy balance
was developed in Matlab, and the data processing was not done
on a real-time base. As the data are available in real-time, and con-
sidering that the calculation time is very short, in a later stage the
data processing can almost certainly be done during the operation
of the drying process.
3.3. Error propagation via Monte Carlo simulation
The error propagation is performed for the dataset where all six
segments of the dryer are lled (dataset B). The sampling of the in-
put for the Monte Carlo simulation is done with Sobol sampling
[22]. In total 21 uncertain variables, which are continuously mea-
sured in the six-segmented dryer, are in a rst step identied as
uncertain and thus part of the error propagation. This includes
measurements of temperature, pressure, and gas ow rate. An
uncertainty level of 2% is selected as measurement error, so the
P  1%; P 1% (with P the measured value). Two extra uncertain
inputs are included in the error propagation, which are the param-
eters of the calibration, performed in Section 4.2. For these inputs
the 95% condence interval of the coefcients is calculated, which
is given by:
ci  wi; ci wi 1
where ci is a parameter of the submodel and wi is the half-width of
a 2-sided condence interval based on a specied level of signi-
cance, a. wi is calculated according to:
wi t adf sei 2
with t adf the t-value of the students t distribution with df degrees of
th
freedom for error and sei is the standard error for the i -coefcient.
A sample size of 25,000 is used for the Monte Carlo simulation. The
generated output is analysed by calculating the mean, the standard
deviation and the 95% condence interval. This is done for the mois-
ture content at the end of the drying process, as well as for the out-
let gas temperature.
4. Results
4.1. Deriving the energy and mass balance for the uidized bed dryer
During operation of the ConsiGma 61 univariate variables are
continuously measured and logged. The available information in-
cludes pressure, temperature, ow rate, and humidity for the dif-
ferent compartments of the continuous line. During uidized bed
drying of wet granules, several processes (evaporation, conduc-
tion/convection, heating, ltration) take place, which lead to a de-
crease in the drying air temperature (T g ) in the dryer and an
increase in the humidity of the incoming drying air (RHg ) while
passing through the drying unit. Given the measured variables
and knowing the processes qualitatively, it should be possible to
predict the moisture content of the outgoing drying air. This allows
prediction of the average drying rate as well. A complete mass and
energy balance is thus set up for the six-segmented uidized bed
dryer. This requires in a rst step the denition of the physical
properties of the in- and outgoing gas, liquid and solid streams of
the subprocess (i.e. the six-segmented dryer), where the full pro-
cess can be seen as the whole from powder to tablet manufacturing
line. The inputs, outputs and initial conditions of the subprocess
should be known and one should be able to quantify them. Once
these balances are closed and validated under a set of relevant
4 S.T.F.C. Mortier et al. / European Journal of Pharmaceutics and Biopharmaceutics xxx (2014) xxxxxx

experimental conditions, they can be used to determine the nal in the incoming and outgoing gas and the gas of the wet transfer
moisture content of the granules and the temperature prole of line is calculated based on RHg ; T g and Pg . The relative humidity
the drying air in the dryer. For the latter each drying cell is divided (RHg ) (a measured variable) in % is expressed as:
in several horizontal sections. The combination of the temperature
ew q
prole in each drying cell with knowledge about the trajectory fol- RHg 100 w 100 4
ew qw;s
lowed by granules throughout the dryer, which can be investigated
by a Computational Fluid Dynamics (CFD)-model, is useful to cal- with ew the partial pressure of water vapour (Pa), ew the saturated
culate the drying behaviour of particles. A combination of the vapour pressure of water (Pa), qw the vapour density (kg/m3) and
description of the spatial behaviour of the granules with a Popula- qw;s the vapour density at saturation (kg/m3). The saturated vapour
tion Balance Model(ling) (PBM)-model would be very useful to pressure of water (Pa) can be calculated using [24]:
understand the process in the uidized bed dryer in more detail.
Population Balance Model(ling) (PBM) is a tool that can for exam- e77:34500:0057T g 7235=T g
ew 5
ple be applied to describe the distributed behaviour of a population T 8:2
g
of particles interacting with each other and the surroundings [23].
Fig. 1 presents the inlets and outlets of the dryer unit. with T g the dry bulb temperature of the moist air (K) (tted to the
experimental data of Landolt-Brnstein [24]). Another alternative
4.1.1. Mass balance relation, which also takes the pressure into account, was presented
The moisture content of the wet granules, introduced in the by Buck [25]:
dryer, can be calculated based on the amount of liquid and solids
introduced in the granulator as the entire mass is transported from ew 6:11211:0007 3:46  106 Pg e17:502T g =240:97T g 6
the granulator to the drying unit. The humidity, temperature and
where P g is the absolute pressure in hPa. This last relation is used
pressure of the incoming and outgoing gas are measured (Fig. 1),
for the calculations. Using the ideal gas law qw can be calculated:
which enables to calculate the amount of water in the gas entering
and leaving the dryer. The resulting steady-state mass balance is ew m
qw 7
given by: RT g
mwat;g;inlet mwat;gran;0 mwat;g;WTL mwat;g;outlet mwat;gran;end 3 with R the universal gas constant (J/mol/K) and m the molar mass of
where mwat;g;inlet and mwat;g;outlet are the mass of water in respectively water (18.01528 g/mol). The water content of gas mwat;g (kg/h) can
the incoming gas stream and the outgoing gas stream, mwat;gran;0 and be calculated by:
mwat;gran;end are the mass of water in respectively the incoming (ini- mwat;g qw V g 8
tial moisture content of the granules) and outgoing granules (result-
3
ing moisture content of the granules at the end of the drying with V g (m /h) the gas ow rate.
process) and mwat;g;WTL represents the mass of water in the gas enter- The humidity and the gas temperature of the gas entering the
ing the dryer via the wet transfer line together with the wet gran- dryer via the wet transfer line are not measured. However, it is as-
ules (Fig. 1). The difference between the water content of the gas sumed that the humidity is high as this gas ow is in contact with
entering and leaving the dryer is the amount of water evaporated the granulator. The temperature is assumed to be a normal ambi-
during drying (mv ap ). As it is a continuous drying process, the unit ent temperature. The assumed corresponding values were taken
of these variables is per unit of time (kg/h). The amount of water as 30% for the RHg;WTL and 20 C for T g;WTL .

Fig. 1. Boundaries of the mass and energy balance for the dryer unit. On the left, a representation of the way the uidized bed dryer system operates is shown, supplemented
with the measured variables. The question mark indicates that these 2 variables are not measured. On the right the mass balance is schematically represented. The calculated
variables are indicated in italics.

Please cite this article in press as: S.T.F.C. Mortier et al., Analysing drying unit performance in a continuous pharmaceutical manufacturing line by means of
mass Energy balances, Eur. J. Pharm. Biopharm. (2014), http://dx.doi.org/10.1016/j.ejpb.2013.12.014
 S.T.F.C. Mortier et al. / European Journal of Pharmaceutics and Biopharmaceutics xxx (2014) xxxxxx 5

The total gas ow rate leaving the dryer is calculated based on Table 2
conservation of mass. As the amount of air entering and leaving the Overview of the data used in the mass/energy analysis.

dryer are equal (in kg/s), this amount can be used to calculate the System Remark
gas ow rate leaving the dryer as the relative humidity RHg;outlet , the A Ghent Only one segment is lled calibration dataset
gas temperature T g;outlet and the pressure P g;outlet of the gas ow B Wommelgem Dataset with the full operating system
leaving the dryer are measured.
The partial vapour pressure of dry air (Pa) is calculated by:
PDA P g  ew 9 T g;inlet , one part in the middle, at the level of the six segments,
and one part where all six segments are joining again. As the
The density of humid air (kg/m3) is given by:
temperature of the gas is measured in each cell separately at
PDA =Rd ew =Rv only one location, it is assumed that the temperature equals
qHA 10
these values in the entire segment. For the upper part of the
Tg
dryer the mean of the temperature in the six cells is used.
with Rd the specic gas constant for dry air (287.058 J/kg/K) and Rv For each part the heat transfer is calculated using an overall
the specic gas constant for water vapour (461.495 J/kg/K). The heat transfer coefcient, which takes the conduction through
knowledge of the density of humid air enables one to calculate the wall and the convection outside the dryer into account. It
the mass of humid air (mHA ): is assumed that there is no barrier inside the dryer for heat
mHA qHA V g 11 transfer, meaning that the temperature at the wall equals the
temperature in the middle of the dryer. As the gas ow inside
The mass of dry air is computed by: the dryer is high, and since there is a lot of turbulence, this
mDA mHA  mwat;g 12 assumption is reasonable. At the outerside of the wall free con-
vection occurs with air as uidum (Fig. 2).
An overview of the different equations used to calculate mwat;g
The overall heat transfer coefcient U (W/m2/K) is calculated
and mDA is given in Table 1.
by:

4.1.2. Energy balance 1

U 14
While passing the drying unit the temperature of the gas de- 1 1=h Dx=k
creases. When the complete energy balance is made, it should be
with h the convective heat transfer coefcient (W/m2/K), Dx the
possible on the one hand to calculate the gas temperature at the
thickness of the wall (m) and k the thermal conductivity (W/m/K).
outlet taking all sinks of energy into account. This calculated value
It is assumed that h and k equal respectively 8 W/m2/K and
can be compared with the measured value in order to check the
16.3 W/m/K.
accuracy of the balance. On the other hand, this obtained knowl-
The heat transfer due to conduction and convection (Q con ) can be
edge can be used to determine the spatial prole of the tempera-
calculated by:
ture in a segment of the dryer. This information is useful when
the trajectory of granules into a segment of the dryer is known. Q con UADT con 15
The combination of both yields the temperature prole that gran-
ules undergo during drying. Several heat sinks can be distin-
guished: energy required to evaporate the water in the granules,
energy loss through the wall of the dryer, energy required to heat
the granules entering the dryer and energy loss due to passing the
lter.

1. Energy required for the evaporation of water from the wet

granules: The energy required to evaporate water (Q v ap (kJ/h))
is calculated by:
Q v ap DHv ap mv ap 13
with DHv ap the heat of evaporation (2257 kJ/kg).
2. Energy loss through the wall of the dryer: The wall of the dryer
consists of stainless steel, which is a good conductor of heat. As
the dryer is not insulated, some heat will be lost through the
wall. To determine the amount of energy lost through the wall,
Fig. 2. Schematic representation of the heat transfer trough the wall of the dryer,
the dryer was split into three parts. One part under the distrib- T ambient and T inside represent respectively the temperature outside the dryer and
utor plate, where it is assumed that the temperature is equal to inside the dryer.

Table 1
Schematic overview of equations to calculate mwat;g and mDA .

RHg
Buck 25 ew Ideal gas law qw
) )
ew (Eq. (6)) Eq:4 Eq:7
) mwat;g
Eq:8
Vg
Pg
) P DA ) qHA ) mHA ) mDA
Eq:9 Eq:10 Eq:11 Eq:12
ew

Please cite this article in press as: S.T.F.C. Mortier et al., Analysing drying unit performance in a continuous pharmaceutical manufacturing line by means of
mass Energy balances, Eur. J. Pharm. Biopharm. (2014), http://dx.doi.org/10.1016/j.ejpb.2013.12.014
6 S.T.F.C. Mortier et al. / European Journal of Pharmaceutics and Biopharmaceutics xxx (2014) xxxxxx
with A the surface area of the wall (m2) and DT con the tempera-
ture difference between the inside and the outside of the dryer.
3. Energy required for heating up the granules: Since the granules
originating from the granulator are at a lower temperature than
the drying air temperature, as such some energy is required to
heat them up. Wet granules consist of water and solid material,
and both need to be heated. The energy required to heat up
material (Q heat (kJ/h)) is:
Q heat mm cp;m DT heat
with mm the mass of the material being warmed, cp;m the specic
heat of the material (kJ/kg/K) and DT heat the difference between
the temperature of the particles originating from the granulator
and the temperature in the cell. Here, it is assumed that the tem-
perature of the particles at the end of the drying process equals
the temperature in the cell. The specic heat of water cp;wat is cal-
culated by [26]:
cp;wat 2:7637  102  2:0901T 8:1250  103 T 2
 1:4116  105 T 3 9:3701
 109 T 4 =M w =1000
with Mw molecular weight of water (18.015  10 kg/mol) and 3
T the temperature in K. The temperature to calculate the specic
heat of water is assumed to equal the temperature of the gran-
ulator barrel. The specic heat of the solid is assumed to equal
1252  103 kJ/kg/K.
4. Energy loss due to passing the lter: Depending on the experi-
mental conditions a lower or higher pressure difference sets in
over the lter. Due to this increasing pressure difference the
energy (Q fil (kJ/h)) needed to pass the gas through the lter
increases.
Q fil DPfil V g
with DPfil the pressure difference over the lter. Two lters are
being passed by the gas and V g is the volume (gas ow rate) of
the gas passing the lter (gas ow rate) (m3/h).
5. Complete energy balance: The total energy required during dry-
ing is the sum of Q v ap ; Q con ; Q heat and Q fil (Table 3). The sum of
the four energy sinks should balance with the energy that is
released due to the cooling of the gas from the inlet tempera-
ture to the outlet temperature (source of energy):
Q air mair cp;air DT air
with mair the mass of air being cooled during drying, cp;air the
specic heat of air (kJ/kg/K) and DT air the temperature difference
between the incoming and outgoing gas.
The energy balance can be used to calculate the temperature at
the outlet of the dryer based on the amount of evaporated water,
the temperature at the inlet and the temperature in each cell of
the dryer.
4.2. Dataset A
Dataset A consists of several runs of experimental data, each
collected on a distinct day. During these runs the gas inlet
Table 3
Schematic overview of the energy balance.
Variables Constants
Sinks Q v ap mv ap DHv ap
Q con DT con U; A
Q heat mm ; DT heat cp;m
Q fil DP fil ; V g
Sources Q air mair ; DT air cp;air
Please cite this article in press as: S.T.F.C. Mortier et al., Analysing drying unit performance in a continuous pharmaceutical manufacturing line by means of
mass Energy balances, Eur. J. Pharm. Biopharm. (2014), http://dx.doi.org/10.1016/j.ejpb.2013.12.014
temperature is not constant. Fig. 3 presents an example of the inlet
gas temperature T g;inlet dynamics used for one day of experiments.
The humidity data of the drying air at the inlet RHg;inlet are pre-
sented in Fig. 4. The dependency of the humidity of the drying air
on temperature is clearly visible. The higher the drying air temper-
ature, the lower the corresponding value for the relative humidity.
In Fig. 4 the measured relative humidity at the outlet RHg;outlet is pre-
sented. The peaks coincide with the time when the dryer unit is
lled with wet granules. Each peak corresponds to one experiment.
16
The calculated mass ux of water in the gas phase at the inlet
mwat;g;IN (sum of mwat;g;inlet and mwat;g;WTL ) and outlet are presented
in Fig. 5. The water content of the gas at the outlet is remarkably
higher compared to the inlet, which is to be expected as the differ-
ence between both should represent the evaporation rate mv ap .
However, when the dryer unit is empty between the experiments
(e.g., time points 1.82 in Fig. 4), the calculated mass of water of
the gas at the outlet and inlet should normally be equal. This
appears not to be the case and an offset (difference between
mwat;g;IN and mw;gas;outlet for an empty dryer) is observed. Moreover,
this offset seems to be correlated with the gas inlet temperature
T g;inlet . For the periods where the dryer is empty, the offsets are cal-
17
culated and plotted against T g;inlet (Fig. 5). Based on this a relation
between the offset and T g;inlet is established. This relation could
be established both for the different runs together, i.e. taking all
data points, (further referred to as case 1) as well as separately,
i.e. for each run another relation (further referred to as case 2)
(i.e. for each day (run) another relation is made between the offset
and the gas inlet temperature). The result of the linear regression
for case 1 is presented in Fig. 5 and results in:
offset 0:0187T g;inlet  0:3763 20
18 with 0:0175 0:0199 and [0.4416 0.3111] the condence inter-
vals (signicance level of 5%) for the coefcients. The R2 of the linear
regression equals 0.95 (Case 1), the R2 for the linear regression for
the independent runs varied between 0.87 and 1. Performing a lin-
ear regression for the different runs separately allows to calculate
the moisture content of the granules at the end of the drying pro-
cess more accurately for each experiment, however, the drawback
is that the calibration is based on less data points. The offset seemed
to be a general problem for all collected datasets (of which only two
are discussed in this contribution), no matter in which system the
19 experiments were performed (cfr. Ghent or Wommelgem). As such,
it was decided to perform one calibration (Case 1) and use the linear
regression for all datasets. But both cases (Case 1 and 2) are used for
Equation
Eq. (13)
Eq. (15)
Eq. (16)
Eq. (18)
Eq. (19)
Fig. 3. T g;inlet for dataset A. The temperature is varying between 35 and 75 C.
 S.T.F.C. Mortier et al. / European Journal of Pharmaceutics and Biopharmaceutics xxx (2014) xxxxxx 7

Fig. 4. RHg;inlet (left) and RHg;outlet (right) for dataset A. The humidity at the inlet is changing due to the change in T g;inlet . An example of one experiment is indicated in the gure
of RHg;outlet . The humidity of the gas is increasing when a drying cell is lled with wet granules.

Fig. 5. mwat;g;IN and mw;gas;outlet (left) for dataset A, where the difference between both should be approximately zero when the dryer is empty. On the right gure the offset
(difference between mwat;g;IN and mw;gas;outlet for an empty dryer) in function of the gas inlet temperature T g;inlet is presented. The offset is calculated based on the calibration
dataset. The linear regression for case 1, where the data points for all runs are used, is shown.

dataset A in order to compare the results. The cause for this offset
remains unclear.
Based on the liquid and solid addition to the granulator the
moisture content of the granules entering the dryer can be calcu-
lated. Using the calculated evaporation rate, based on the proper-
ties of the gas entering and leaving the granulator, the moisture
content of the granules leaving the dryer is computed. Here, the
evaporation rate is corrected for the above discussed offset. As
one of the objectives is to monitor the moisture content of the
granules leaving the dryer by using the real-time in-line measure-
ments, it is important that the calculated moisture content is close
to the moisture content measured by Karl Fisher titration
(validation). The results are presented in Table 4 for case 1 and 2
(X end;cal;1 ; X end;cal;2 respectively). X end;KF and X end;KF;corr are respectively
the mean of three Karl Fisher measurements and the corrected va-
lue. The standard deviation (rX end;KF ) is also calculated and listed in
Table 4. The experiments with an overprediction are indicated in
bold, whereas the predictions with an underprediction are given
in italics. The best result (Case 1 or 2) for each experiment is under-
lined. The scenario with the different linear regressions yields the
best result. However, as the linear regression will be used for all
datasets, collected on both the system in Ghent as in Wommelgem
and operating in such a way that only one segment is lled with
wet granules or for the full operating system, the linear regression
of case 1 will be implemented in the mass and energy balance. This
regression is more generally applicable as it is based on more data
points, collected over several days.
4.3. Dataset B
When collecting dataset B, the full dryer equipment is used in a
continuous operation (6 h), and the set of process parameters are
not changed during operation. Fig. 6 presents the measured tem-
perature at the inlet T g;inlet and shows that after the initial dynamic
behaviour the temperature is close to the set point of 45 C.
The measured outlet temperature T g;outlet is obviously lower
than T g;inlet , as well as the calculated mean temperature over the
six segments (T g;mean ) (Fig. 7). The gas temperature measured in
the six cells is also added in Fig. 7. The gas temperature decreases
in the cell when wet granules are lled in the segment. During dry-
ing the gas temperature increases again as less and less energy is
required for evaporation of water. Based on these measurements

Please cite this article in press as: S.T.F.C. Mortier et al., Analysing drying unit performance in a continuous pharmaceutical manufacturing line by means of
mass Energy balances, Eur. J. Pharm. Biopharm. (2014), http://dx.doi.org/10.1016/j.ejpb.2013.12.014
8 S.T.F.C. Mortier et al. / European Journal of Pharmaceutics and Biopharmaceutics xxx (2014) xxxxxx

Table 4 At the early start it is obvious that the drying air temperature at the
Moisture content of the granules at the end of the drying process for dataset A (Bold outlet is higher, as there is only one segment in which particles are
indicates an overprediction, italics an underprediction, underlined the best result).
X end;KF;corr represents the corrected value (dry premix correction), which is the target
being dried. Only after all segments have been lled with particles,
for our calculations. X end;cal;1 and X end;cal;2 are the calculated values for respectively case the drying air temperature could reach a dynamic steady state.
1 and case 2. Also, at the end of the 6 h process the drying air temperature in-
T g;inlet (C) X end;KF (%) rXend;KF X end;KF;corr (%) X end;cal;1 (%) X end;cal;2 (%)
creased again, as the experiment was stopped and no particles
were lled in the dryer anymore. The drying air temperature mea-
35 6.57 0.33 2.97 3.37 3.65
sured in each segment follows a cyclic behaviour. The six-seg-
35 6.66 0.14 3.06 3.26 3.54
mented dryer is operating in a continuous way, i.e. one segment
35 7.06 0.13 3.46 3.57 3.74
is lled, while in the other segments granules are being dried.
35 5.57 0.30 1.97 2.81 2.42 T g;mean is remarkably higher compared to T g;outlet , which could be re-
35 5.68 0.10 2.08 2.99 2.61 lated to the location of the sensors, and/or the presence of lters
35 5.81 0.08 2.21 2.85 2.42
between the drying unit itself and the outlet.
35 7.05 0.15 3.45 1.52 2.98
The measured relative humidity at the inlet (RHg;inlet ) and at the
35 6.80 0.03 3.20 2.09 3.22
outlet (RHg;outlet ) is presented in Fig. 8. The relative humidity at the
35 6.44 0.15 2.84 2.28 3.32
inlet was around 10%, and only at the start the humidity was
35 6.28 0.15 2.68 2.81 3.00
clearly higher. At the outlet the humidity increases: as in the
35 6.09 0.12 2.49 2.43 2.62
beginning of the experiment it took some time until all segments
50 4.92 0.02 1.32 1.68 1.41
50 4.91 0.19 1.31 2.11
are lled, a quite steep increase was observed. However, it took
1.80
50 4.83 0.45 1.23 1.82
again some time until a steady state was reached. The cyclic behav-
1.47
50 5.36 0.14 1.76 2.22
iour, also present in the gas outlet temperature, is clearly visible
2.20
50 4.89 0.14 1.29 1.05 here as well.
1.08
55 4.24 0.09 0.64 1.54 In Fig. 9 the pressure difference over the lters is shown. A stea-
1.49
60 3.70 0.14 0.10 1.46 dy increase in pressure difference is obvious. This is understand-
0.17
60 3.82 0.31 0.22 1.51 able as more and more dust is stacked into the lters. This
0.20
60 4.22 0.05 0.62 0.08 1.87 increase will contribute to the difference between T g;mean and
60 4.06 0.10 0.46 0.68 0.81 T g;outlet .
75 3.47 0.04 0 0.25 0.85 Fig. 10 presents the gas ow rate at the inlet and the outlet, and
75 3.63 0.04 0.03 2.11 1.66 obviously the ow is higher at the outlet. The ow at the outlet is
75 3.73 0.07 0.13 1.57 1.03 rstly the combination of the ow at the bottom of the dryer and
75 3.66 0.04 0.06 1.10 0.69 the wet transfer line. Moreover, due to a difference in temperature,
75 3.60 0.13 0 1.00 0.47 pressure and water content the gas ow rate at the top can differ
75 3.57 0.08 0 1.64 0.86 from the bottom of the dryer.
75 3.53 0.06 0 1.57 0.87 In Fig. 11 the evaporation rate is presented, calculated using the
linear regression (see Eq. (20)). The increase and decrease of mv ap at
the start and the end of the run, respectively, are clearly visible. At
the start one segment is lled after the other which causes the
gradual increase in mv ap , and the same reasoning is valid at the end.
Based on the liquid and solid addition to the granulator the
moisture content of the granules entering the dryer was calculated.
Using the calculated (and corrected-validated) evaporation rate the
moisture content of the granules leaving the dryer is computed
(Fig. 12). It is impossible to calculate the moisture content of the
granules for each cell separately as only the cumulative humidity
of the gas leaving the dryer is measured. Karl Fisher titration mea-
sures a value of 13% for the initial moisture content and 1.351.45%
for the residual moisture content. In Fig. 12 the moisture content of
the granules entering the dryer (X 0 ) and leaving the dryer (X end ) is
presented. The calculated mean over a time span of 2 h equals
13.05% for X 0 and 1.37% for X end , values which are in very good
agreement with the Karl Fisher results.
The experimental outlet drying air temperature T g;outlet;exp is
compared with the calculated temperature T g;outlet;cal in Fig. 12. In
general the experimental and calculated results coincide well,
and the overall cyclic behaviour is captured in the calculations.
Fig. 6. T g;inlet for dataset B. After a short transient the set point is reached. However, at a smaller scale the peaks of the calculation do not
coincide with the experimental results. The combination of the
the process in each segment can be followed, however, the mois- outlet gas temperature and the humidity causes the uctuation
ture content of the drying air is not measured in each cell. The in the qHA of the gas at the outlet. This strong uctuation causes
latter would provide extra information about the process and could the uctuation in T g;outlet;cal , and causes the shift in the time instants
be used to detect problems during operation. At the outlet a convo- of the peaks. The calculated mean temperature is 28.20 C.
lution of two sinusoidal functions with different frequencies and Finally, an uncertainty propagation was performed on this data-
amplitudes can be observed for T g , an explanation for this set. The standard deviation for the moisture content X end is 0.2371.
behaviour has been mentioned in the discussion (Section 5.1). The resulting 95% condence interval is 0:9383 1:8382. The exper-
However, it is noteworthy that it took almost 1.5 h until T g;outlet imentally measured moisture content is clearly situated in the con-
reached a steady-state value of around 28 C (dynamic behaviour). dence interval. For the gas temperature the standard deviation is

Please cite this article in press as: S.T.F.C. Mortier et al., Analysing drying unit performance in a continuous pharmaceutical manufacturing line by means of
mass Energy balances, Eur. J. Pharm. Biopharm. (2014), http://dx.doi.org/10.1016/j.ejpb.2013.12.014
 S.T.F.C. Mortier et al. / European Journal of Pharmaceutics and Biopharmaceutics xxx (2014) xxxxxx 9

Fig. 7. T g;outlet and in the different segments, as well as the mean of these temperatures for dataset B: entire experiment (left), detail of approximately 2.5 h (right).

Fig. 8. RHg;inlet (left) and RHg;outlet (right) for dataset B. After some dynamic behaviour at the start the humidity stays around 10% for RHg;inlet . The cyclic behaviour of RHg;outlet is
clearly visible.

Fig. 10. V g;inlet and V g;outlet for dataset B. V g;outlet is calculated since this variable is not
Fig. 9. DP fil over the lters for dataset B. An increase in DP fil is obvious. measured.

Please cite this article in press as: S.T.F.C. Mortier et al., Analysing drying unit performance in a continuous pharmaceutical manufacturing line by means of
mass Energy balances, Eur. J. Pharm. Biopharm. (2014), http://dx.doi.org/10.1016/j.ejpb.2013.12.014
10 S.T.F.C. Mortier et al. / European Journal of Pharmaceutics and Biopharmaceutics xxx (2014) xxxxxx

Fig. 11. mv ap for dataset B. The calculation of mv ap is based on the measured

variables of the drying air.
Fig. 13. T g;outlet;exp and 95% condence interval for dataset B.

0.2115, and the 95% condence interval 27:7994 28:6021. The

result for the condence interval of the gas temperature at the out-
let is presented in Fig. 13. The condence interval is small, and the
experimental measured temperature does not fall into the interval.
5. Discussion
5.1. Use of in-line data for process monitoring of the ConsiGma
The use of the standard logged data and a mass and energy bal-
ance can be compared with the use of expensive sensors, which are
typically not standard available in most industrial pharmaceutical
equipment. The use of the univariate data enables us to determine
directly the end of the drying process by monitoring of the temper-
ature prole of the different segments of the dryer, however, as it is
not always desired to dry the granules till a moisture content of 0%,
these proles are not always sufcient. The use of a mass balance
would be a powerful tool to quantify the moisture content of the
granules at the end of the process. Moreover, the implementation
of the balances in a real-time system enables the operator to follow
up the process in a better way. A deviation in the gas outlet tem-
perature gives the operator an indication that something is chang-
ing, but the calculated evaporation rate gives more information
about the drying process itself, as it takes the inlet gas temperature
into account. Combining the information about the moisture con-
tent with an appropriate controller in order to increase or decrease
the evaporation rate would be a major step forward towards im-
proved process control of the drying process. However, the imple-
mentation of a controller requires more information about the
drying process, for this reason mechanistic models are a powerful
tool.
In our case the calculated moisture content of the granules at
the end of the drying process was lower than the result obtained
by Karl Fisher titration. The granules are brought into the dryer
using a transfer line, which is in contact with the granulator, and
where the humidity of the gas is high. Together with these gran-
ules, water-rich air is introduced into the dryer, but no sensors
are located at this inlet to measure the humidity and the gas tem-
perature. As a consequence, an assumption has to be made about
the humidity and the gas temperature. To calculate the amount
of water evaporated during drying, it is assumed that the differ-
ence between the water content of the gas at the inlet and the out-
let is only related through the drying process. However, in our case
an unknown process takes place which resulted in a higher mois-
ture content of the drying gas at the outlet compared to the inlet

Fig. 12. X 0 and X end (left) and T g;outlet;exp and T g;outlet;cal (right) for dataset B. X 0 and X end are calculated based on properties of the input of the granulator and the mv ap . Both
T g;outlet;exp and T g;outlet;cal display a cyclic behaviour, however at a smaller scale the peaks do not match entirely.

Please cite this article in press as: S.T.F.C. Mortier et al., Analysing drying unit performance in a continuous pharmaceutical manufacturing line by means of
mass Energy balances, Eur. J. Pharm. Biopharm. (2014), http://dx.doi.org/10.1016/j.ejpb.2013.12.014
 S.T.F.C. Mortier et al. / European Journal of Pharmaceutics and Biopharmaceutics xxx (2014) xxxxxx
for an empty dryer. This means an overprediction of the evapora-
tion rate, which is solved by implementing a correction term, but
needs further investigation. With this correction reasonable results
could be found for several data sets, of which part of them are pre-
sented in this paper. This means that mass and energy balances are
a powerful tool to monitor the process without the need for extra
sensors or time-consuming off-line experiments. An important
conclusion is that it is very important not to trust the available data
blindly, but perform basic reconciliation checks. In this case the
problem with the data was revealed by making a mass and energy
balance.
An important issue is the unequal behaviour of the different
cells. This is clearly visualised by the sinusoidal function with low-
er frequency (Fig. 7). The higher frequency sinusoidal behaviour
corresponds with one lling period (i.e. 180 s for dataset B),
whereas the lower frequency sinusoidal behaviour corresponds
with the total cell cycle (i.e. six times the lling period or 1080 s
for dataset B). As the cells do not operate similarly, the relative
humidity at the outlet is uctuating according to which cell is in
a certain state of the cycle (Fig. 8). For instance, each time that cell
1 is lled with wet granules more water is evaporated and RHg;outlet
is higher.
A good prediction of the gas outlet temperature gives the
opportunity to use this information and to combine with the tra-
jectory that particles are following during uidization. The study
about the ow of particles can be done using Computational Fluid
Dynamics (CFD), which is also able to predict the bed height of the
granules. The gas temperature is not uniform in the dryer, but de-
creases due to several sinks (Section 4.1.2 (Fig. 14)). The bed height
will determine in which section of the segment the gas tempera-
ture will decrease fast, as the heat for evaporation is the largest
sink. The bed height can, together with assumptions about the uni-
formity of particles, be able to predict the amount of heat loss in
different horizontal sections of the dryer. Another solution to cal-
culate the gas temperature in different horizontal sections or in
more detail at all locations is the use of a combined Computational
Fluid Dynamics (CFD)Population Balance Model(ling) (PBM)-
model. Before the combination can be realised, a thorough valida-
tion of both models is necessary. Nevertheless, the combined
model will be computationally very expensive. The use of the
developed energy balance can give a rst insight in the gas temper-
ature at different locations without the need for a validated Popu-
lation Balance Model(ling) (PBM)-model and the combined model.
As indicated extra sensors at the wet transfer line would give
the opportunity to eliminate the assumptions about the humidity
and the gas temperature. Furthermore, the possibility to measure
the humidity and the temperature of the outlet air from each
Fig. 14. Schematic representation of the combination of the bed height with the
sinks of heat in order to predict the gas temperature in different horizontal sections
of one segment of the dryer.
Please cite this article in press as: S.T.F.C. Mortier et al., Analysing drying unit performance in a continuous pharmaceutical manufacturing line by means of
mass Energy balances, Eur. J. Pharm. Biopharm. (2014), http://dx.doi.org/10.1016/j.ejpb.2013.12.014
11
segment of the dryer would even provide additional information
about the process. The calculation of the evolution of the moisture
content during drying would give the operator the freedom to
interfere within the process in a more accurate way.
5.2. Prerequisites, assumptions and shortcomings when using mass
and energy balances
However, using measurements to follow the process dynamics
and perform process control in continuous production processes
requires detailed information and has also some shortcomings. A
mass and energy balance requires rst of all the denition of the
subprocesses, and all inlets and outlets should be known if it is
an open process. However, there should be data available to quan-
tify the important variables. Furthermore, assumptions will typi-
cally need to be made in order to simplify the process, and these
assumptions should be carefully checked in order to justify them.
 The exact location of sensors should be known in order to (1)
know which processes have to be taken into account and (2)
to be sure that the measurements deal with the mentioned
ow. For instance, the water content of the gas is dependent
on the gas temperature (T g ), the pressure (Pg ) and the volume
of the gas (V g ). If not all sensors are located at the same location,
the water content of the gas cannot be determined correctly.
Moreover, a sensor which is in contact with the material will
produce other results compared to a sensor in contact with
the gas phase.
 The presence of lters is important to include pressure drops or
pressure rises, which results in an energy release or uptake.
 A mass and energy balance requires the knowledge of all inputs
and outputs of the subprocess. If there is a lack in measure-
ments at the boundaries of the subprocess the mass balance will
be incorrect, leading to under- or overestimation of the calcu-
lated output variable. As such it is important to check the bal-
ances with validation experiments a priory (i.e. before the use
in industrial applications) and/or to include a real-time valida-
tion of the measurements by performing basic checks (e.g. com-
paring measured with calculated variables).
 An important shortcoming of mass and energy balances is their
inability to assist in gaining detailed insight in the process since
they are very crude. More detailed mechanistic models are very
powerful tools to understand the process in more detail, and
understand inputoutput relations. With the balances the inu-
ence of changing input variables during operation cannot be
investigated as the effect of the input variables on the output
variable of interest, i.e. the moisture content, is not known
[5]. Once the effect of the input variables on the output variable
of interest is known, the operator has freedom to change the
inputs during operation in order to end up with the desired
end quality.
6. General conclusion
The automatically collected in-line real-time data of the Cons-
iGma continuous tableting line, and more especially the uidized
bed dryer unit, is processed. The use of a mass and energy balance
enables to monitor the process without the need for extra sensors
or time-consuming off-line experiments. However, a calibration in
order to predict the moisture content of the granules leaving the
dryer was necessary. This was done by implementation of a linear
regression to calculate the evaporation rate. The observed differ-
ence in water content of the gas at the inlet and the outlet for an
empty dryer remains unsolved. For dataset B the experimental re-
sults, measured by Karl Fisher titration, and the calculated results
were in very good agreement. The error propagation performed on
12 S.T.F.C. Mortier et al. / European Journal of Pharmaceutics and Biopharmaceutics xxx (2014) xxxxxx
the data was of interest to calculate the condence interval on the
calculated values and investigate whether measurement errors
could explain the observed deviations between calculations using
balances and actual measurements.
Mass and energy balances are powerful to monitor the process
on-line, however gaining more detailed insight in the process is
impossible. For the latter, mechanistic models remain necessary.
Acknowledgment
Financial support for this research from the Fund for Scientic
Research Flanders (FWO Flanders Ph.D. fellowship Sverine
Thrse F.C. Mortier) is gratefully acknowledged.
References
[1] K. Plumb, Continuous processing in the pharmaceutical industry: changing the mind
set, Chem. Eng. Res. Des. 83 (6) (2005) 730738, http://dx.doi.org/10.1205/
cherd.04359. <http://linkinghub.elsevier.com/retrieve/pii/S0263876205727556>.
[2] A. Cervera-Padrell, T. Skovby, S. Kiil, R. Gani, K. Gernaey, Active pharmaceutical
ingredient (API) production involving continuous processes a Process Systems
Engineering (PSE)-assisted design framework, Eur. J. Pharm. Biopharm. 80
(2012) 437456.
[3] C. Vervaet, J. Remon, Continuous granulation in the pharmaceutical industry,
Chem. Eng. Sci. 60 (14) (2005) 39493957, http://dx.doi.org/10.1016/j.ces.2005.
02.028. <http://linkinghub.elsevier.com/retrieve/pii/S0009250905001284>.
[4] H. Leuenberger, New trends in the production of pharmaceutical granules: the
classical batch concept and the problem of scale-up, Eur. J. Pharm. Biopharm.
52 (3) (2001) 279288. <http://www.ncbi.nlm.nih.gov/pubmed/11677070>.
[5] S. Mortier, T. De Beer, K. Gernaey, J. Remon, C. Vervaet, I. Nopens, Mechanistic
modelling of uidized bed drying processes of wet porous granules: a review,
Eur. J. Pharm. Biopharm. 79 (2) (2011) 205225, http://dx.doi.org/10.1016/
j.ejpb.2011.05.013.
[6] H. Leuenberger, New trends in the production of pharmaceutical granules:
batch versus continuous processing, Eur. J. Pharm. Biopharm. 52 (3) (2001)
289296. <http://www.ncbi.nlm.nih.gov/pubmed/11677071>.
[7] International Conference on Harmonisation of Technical Requirements for
Registration of Pharmaceuticals for Human Use. <http://www.ich.org/cache/
compo/276-254-1.html>.
[8] GEA Pharma Systems. <http://www.gea-ps.com/npsportal/cmsdoc.nsf/
WebDoc/misy6lmeqe>.
[9] Ldige Process Technology. <http://www.loedige.de/Startseite.html>.
[10] F. Portoghese, F. Berruti, C. Briens, Continuous on-line measurement of solid
moisture content during uidized bed drying using triboelectric probes,
Powder Technol. 181 (2) (2008) 169177, http://dx.doi.org/10.1016/j.powtec.
2007.01.003. <http://linkinghub.elsevier.com/retrieve/pii/S0032591007000095>.
[11] P. Frake, D. Greenhalgh, S. Grierson, J. Hempenstall, D. Rudd, Process control
and end-point determination of a uid bed granulation by application of near
Please cite this article in press as: S.T.F.C. Mortier et al., Analysing drying unit performance in a continuous pharmaceutical manufacturing line by means of
mass Energy balances, Eur. J. Pharm. Biopharm. (2014), http://dx.doi.org/10.1016/j.ejpb.2013.12.014
infra-red spectroscopy, Int. J. Pharm. 151 (1) (1997) 7580, http://dx.doi.org/
10.1016/S0378-5173(97)04894-1. <http://linkinghub.elsevier.com/retrieve/
pii/S0378517397048941>.
[12] M. Sarragua, A. Cruz, S. Soares, H. Amaral, P. Costa, J. Lopes, Determination of
ow properties of pharmaceutical powders by near infrared spectroscopy, J.
Pharm. Biomed. Anal. 52 (4) (2010) 484492, http://dx.doi.org/10.1016/
j.jpba.2010.01.038. <http://www.ncbi.nlm.nih.gov/pubmed/20167448>.
[13] L. Chablani, M. Taylor, A. Mehrotra, P. Rameas, W. Stagner, Inline real-time
near-infrared granule moisture measurements of a continuous granulation-
drying-milling process, AAPS PharmSciTech 12 (4) (2011) 10501055, http://
dx.doi.org/10.1208/s12249-011-9669-z. <http://www.pubmedcentral.nih.gov/
articlerender.fcgi?artid=3225541&tool=pmcentrez&rendertype=abstract>.
[14] C. Buschmller, W. Wolfgang, C. Dscher, J. Dressler, J. Breitkreutz, In-line
monitoring of granule moisture in uidized-bed dryers using microwave
resonance technology, Eur. J. Pharm. Biopharm. 69 (2008) 380387.
[15] H. Wang, P. Senior, R. Mann, W. Yang, Online measurement and control of
solids moisture in uidised bed dryers, Chem. Eng. Sci. 64 (2009) 28932902.
[16] P. Vachhani, R. Rengaswamy, V. Gangwal, S. Narasimhan, Recursive estimation
in constrained nonlinear dynamical systems, AIChE J. 51 (3) (2005) 946959.
[17] J. Helton, F. Davis, Latin hypercube sampling and the propagation of
uncertainty in analyses of complex systems, Reliab. Eng. Syst. Saf. 81 (1)
(2003) 2369, http://dx.doi.org/10.1016/S0951-8320(03)00058-9. <http://
linkinghub.elsevier.com/retrieve/pii/S0951832003000589>.
[18] G. Anderson, Error propagation by the Monte Carlo method in geochemical
calculations, Geochim. Cosmochim. Acta 40 (12) (1976) 15331538, http://
dx.doi.org/10.1016/0016-7037(76)90092-2. <http://linkinghub.elsevier.com/
retrieve/pii/0016703776900922>.
[19] M. Herrador, A. Gonzlez, Evaluation of measurement uncertainty in analytical
assays by means of Monte-Carlo simulation, Talanta 64 (2) (2004) 415422,
http://dx.doi.org/10.1016/j.talanta.2004.03.011. <http://www.ncbi.nlm.nih.
gov/pubmed/18969620>.
[20] S. Mortier, T. De Beer, K. Gernaey, J. Vercruysse, M. Fonteyne, J. Remon, C.
Vervaet, I. Nopens, Mechanistic modelling of the drying behaviour of single
pharmaceutical granules, Eur. J. Pharm. Biopharm. 80 (3) (2012) 682689,
http://dx.doi.org/10.1016/j.ejpb.2011.12.010.
[21] M. Fonteyne, J. Vercruysse, D. Daz, D. Gildemyn, C. Vervaet, J. Remon, T. De Beer,
Real-time assessment of critical quality attributes of a continuous granulation
process, Pharm. Dev. Technol. (2011) 113, http://dx.doi.org/10.3109/
10837450.2011.62786. <http://www.ncbi.nlm.nih.gov/pubmed/22023327>.
[22] I. Sobol, On the systematic search in a hypercube, SIAM J. Numer. Anal. 16 (5)
(1979) 790793.
[23] D. Ramkrishna, Population Balances: Theory and Applications to Particulate
Systems in Engineering, Academic Press, New York, 2000. <http://
www.amazon.com/Population-Balances-Applications-Particulate-
Engineering/dp/0125769709>.
[24] Landolt-Brnstein, Numerical data and functional relationships in science and
technology, Phys. Chem. Prop. Air (1990).
[25] A. Buck, New equations for computing vapor pressure and enhancement
factor, J. Appl. Meteorol. 20 (1981) 15271532.
[26] R. Perry, D. Green, Perrys Chemical Engineers Handbook, seventh ed.,
McGraw-Hil, New York, 2007.