Drying Technology

An International Journal

ISSN: 0737-3937 (Print) 1532-2300 (Online) Journal homepage: https://www.tandfonline.com/loi/ldrt20

Application of a mechanistic model of batch

fluidized bed drying at laboratory and pilot scale

Emmanuela Gavi

To cite this article: Emmanuela Gavi (2019): Application of a mechanistic model of batch fluidized
bed drying at laboratory and pilot scale, Drying Technology, DOI: 10.1080/07373937.2019.1611594

To link to this article: https://doi.org/10.1080/07373937.2019.1611594

Published online: 13 May 2019.

Submit your article to this journal

Article views: 15

View related articles

View Crossmark data

Full Terms & Conditions of access and use can be found at

https://www.tandfonline.com/action/journalInformation?journalCode=ldrt20
DRYING TECHNOLOGY
https://doi.org/10.1080/07373937.2019.1611594

Application of a mechanistic model of batch fluidized bed drying at

laboratory and pilot scale
Emmanuela Gavi
PTDC Formulation R&D, F. Hoffmann-La Roche, Basel, Switzerland

ABSTRACT ARTICLE HISTORY

The aim of this work was to test the feasibility of reducing experimentation during the Received 19 January 2018
development of the fluid bed drying process of a pharmaceutical product, by conducting Revised 21 April 2019
experimentation at laboratory scale and by using a mechanistic model calibrated on the Accepted 22 April 2019
laboratory scale data to simulate the process at pilot scale. The advantage of this approach
KEYWORDS
with respect to a fully empirical approach, is that a limited number of experiments are run Fluidized bed drying;
at pilot scale with the purpose of scale verification, instead of an extensive experimental mechanistic modeling;
campaign. Batch fluidized bed drying experiments were run with two different formulations scale-up
containing the same active pharmaceutical ingredient (API) at laboratory and pilot scales. A
mechanistic model was calibrated at laboratory scale, and then the same parameters
describing drying behavior were used in simulations of the drying process at the pilot scale.
The setup of the model differed for the two formulations; in particular, for the more hygro-
scopic formulation, the model needs to account for a higher influence of the material on
drying. The simulated product temperature and moisture content in granules were com-
pared with experimental data. Model accuracy is good at steady state for granule moisture
content and for product temperature, in particular for the more hygroscopic formulation.
During dynamic phases the product temperature prediction is less accurate: this effect is
more visible for the less hygroscopic formulation. Finally, the mechanistic model calibrated
at laboratory scale was used to simulate experimental results obtained at pilot scale. Good
agreement between simulations and experiments could be observed, confirming that the
suggested approach could be used in order to considerably decrease experimental effort
during process development with respect to a fully empirical approach.

Introduction Therefore, enhanced process understanding of the

Batch fluidized bed drying is used in the pharmaceut-
ical industry in the manufacture of oral dosage forms
as the final step of a wet granulation process, be it via
high shear wet granulation or via fluidized bed granu-
lation. During the development phase of a wet granu-
lation process it is of interest to understand the effect
of the drying process parameters on the critical qual-
ity attributes (CQAs) of the final product, following a
Quality by Design approach.[1] When the final prod-
uct is a tablet, the residual moisture content of the
granules at the end of the drying process may influ-
ence the tablet performance: Gabbott et al.[2] showed
that granules with higher residual moisture generate
tablets with lower porosity, therefore characterized by
a higher hardness and a slower dissolution. The
residual moisture content in the granules and in the
final tablet may also influence the product stability.[3]
drying process is necessary in order to assess the
impact of the process parameters on the residual
moisture content of the granules and to design a hol-
istic control strategy.
Process development studies in the pharmaceutical
industry are often carried out at pilot scale. Empirical
correlations between process parameters and the
CQAs are then established. As the empirical correla-
tions are then only valid for the pilot scale, further
elaborate experiments are required at commercial
scale, when the process is scaled up.
Use of mechanistic models during process develop-
ment can be used to gain enhanced process under-
standing and reduce the experimental burden during
process development. The fluidized bed drying unit
operation has been studied for a long time (see for
example[4–6]); several mechanistic models of different
levels of complexity have been published, and can be

CONTACT Emmanuela Gavi emmanuela.gavi@roche.com PTDC Formulation R&D, F. Hoffmann-La Roche, Basel, Switzerland.
Color versions of one or more of the figures in the article can be found online at www.tandfonline.com/ldrt.
ß 2019 Taylor & Francis Group, LLC
2 E. GAVI
used to overcome various challenges observed during
the process development stages.
Kemp and Oakley[6] categorized dryer models
according to their complexity and the output informa-
tion that they provide. The simplest models are based
on heat and mass balances, but drying kinetics must
be included in the model when predicting drying time
is of interest. This can be done in single- or multi-
phase models. In single-phase models the bed is
regarded as a continuum where gas and solid particles
are perfectly mixed.[7] Two and three-phase models
attempt to include the effect of fluid dynamics, such
as the by-pass of drying air due to the presence of
bubbles, and back mixing due to turbulence.[8,9] The
bed fluid dynamics can also be described by means of
Computational Fluid Dynamics (CFD), for example
by using the Eulerian-Eulerian Two-Fluid Model.[10,11]
With this approach the bed hydrodynamics is more
accurately described than in two- and three-phase
models: assumptions related to the void fraction in
the bed or empirical relationships describing bed
expansion and bubble size are not necessary, and the
geometry of the equipment can be taken into account.
Drying rate models can be distributed or lumped
models, as described by Kemp and Oakley.[6]
Distributed models make use of fundamental physical
equations and coefficients, however their application
may be limited by the feasibility of measuring the
required parameters. An example of such an approach
is reported by Wang et al.,[10] where the drying in the
falling rate period was described by the diffusion
equation in a spherical particle and solved by applying
a finite volume method. Lumped models are instead
constituted by a smaller number of equations than
distributed models, where materials drying behavior is
described by few lumped parameters that combine dif-
ferent physical phenomena. In this case the drying
rate model is not fully predictive, but experiments are
needed to calibrate the model to obtain the lumped
parameters.[6] An example of such a lumped model is
that developed by Burgschweiger et al.,[8] where the
drying rate in the falling rate period is calculated
through a normalized drying rate, which is a param-
eter fitted on experiments.
Kemp[12] observed that drying rates and times are
affected by macro-scale effects, at the overall bed level,
and therefore there is no need for localized models,
meaning that a single-phase model can be fit for pur-
pose. Also, he observed that, although drying kinetics
is affected by micro-scale effects within particles, it
can be easily characterized using falling kinetics
correlations. Such an approach can be categorized as a
lumped drying rate model.
He and Pinto[13] applied a one phase model to
describe batch fluidized bed drying of pharmaceutical
granules at different scales: laboratory, pilot and full
scale. The drying rate in the falling rate period was
described by a falling-rate kinetics correlation. Two
lumped parameters, the heat loss coefficient and the
moisture diffusivity, were fitted through experiments,
then the adjusted values were used to predict the
other batches, and finally simulations were compared
to experimental results in terms of moisture content
and product temperature. The agreement was found
to be good at laboratory and pilot scales. A greater
level of discrepancy was found between experiments
and simulations at full scale, and it was hypothesized
that the reason for this was that the assumption of a
perfectly mixed system was not fully valid at
that scale.
Kemp and Sohet[14] describe how the deep bed
model can be used to simulate drying rate, gas tem-
perature and humidity for different bed scales and at
different heights of the bed. Simulating the drying
process along the bed height can be useful because the
measured product temperature depends on the tem-
perature probe height, which may be different in dif-
ferent units where the process may be transferred. In
the same paper it is also described how the model was
used to optimize process operating conditions by cal-
culating “what if” scenarios. The model allowed
reducing to less than half the drying time at pilot
scale, and scaling up the process directly to full-scale,
avoiding expensive experiments at both scales.
The literature review above shows that the relation-
ship between the main process parameters of batch
fluidized bed drying and the process responses, such
as product temperature, outlet air humidity and mois-
ture content, are well known. Mechanistic mathemat-
ical models are thus available and have been used in
industrial practice to simulate the drying process
responses at different scales. A logical inference is that
experimental effort during pharmaceutical develop-
ment of a wet granulation process could be reduced,
by only carrying out material-consuming design of
experiments at laboratory scale, in order to study the
influence of granule residual moisture content on tab-
let CQAs. The drying process could then be scaled up
to pilot and commercial scales by making use of a
mechanistic model. At pilot and commercial scales a
reduced number of experiments could be sufficient to
confirm the process understanding gained at labora-
tory scale. An example of potential savings that such

an approach could lead to is the following. A design
of experiments (DoE) for high shear wet granulation
and drying at Roche is typically performed with 11
batches at pilot scale, approximately consuming 15 kg
of drug product for every batch. If the DoE was
instead run at laboratory scale, 2 kg of drug product
would be consumed for every batch. In addition to
the DoE, at least 3 confirmatory batches for the scale-
up approach would be necessary at pilot scale, con-
suming 15 kg of drug product each. The resulting DP
saving would therefore be 98 kg, that for a 50% drug
load product would mean a saving of 49 kg API, or
60% saving with respect to a DoE conducted at pilot
scale. In addition, the knowledge acquired by develop-
ing the scale-up approach could be used for further
scale-up to commercial scale, aiming at a first-time-
right scale-up, without the need for a further DoE to
be performed at that scale. The aim of this work is to
test the feasibility of reducing experimentation in pro-
cess development and scale-up of batch fluidized bed
drying by using a mechanistic model, particularly
regarding the applicability of lumped parameters at
different operating conditions and scales. A mechanis-
tic model of batch fluidized bed drying was used to
study the drying process of two different formulations
containing model API. The model was calibrated at
laboratory scale and it was then used to simulate the
drying process of the same material at pilot scale. In
the next section, the model used in this work will be
described, then the experimental set-up will be
detailed, followed by the model set up and calibration.
The model prediction and the experimental data will
be compared in the results section, followed by a dis-
cussion and conclusions.
Fluidized bed drying model and normalized
drying rate
The fluidized bed drying model applied in this work
is a single-phase model, where the drying rate in the
falling rate period is described by a lumped parame-
ters correlation calibrated through experiments. The
model implementation is available in the commercial
software gSolids 4.1 (PSE Ltd, London, UK) and is
based on a population balance approach. In the fol-
lowing description, simplified equations adapted from
Burgschweiger et al.[8] are reported for illustrative
purposes for a uniform granule population, with gran-
ule size dP. For the actual balance equations for a dis-
tributed granules population applied by the model the
interested reader can refer to.[15] The model imple-
mented in gSolids is a simplification of the model in
DRYING TECHNOLOGY 3
Burgschweiger et al.,[8] because the latter is a two-
phase model, whereas in gSolids the assumption of a
perfectly mixed system is made.
The model is constituted by balance equations for
the transport of vapor mass and energy for the par-
ticles and the suspension gas. The assumption is made
that the unit is perfectly mixed at all times; therefore
the particles temperature, the gas temperature within
the equipment, and the outlet temperature are
the same.
The equation of moisture balance for the solid
phase is
dX
Mp _ pg :
¼ M (1)
dt
The moisture content of the solids in the bed,
Mp X; decreases with time, because of water evapor-
ation due to drying. This rate of drying is equal to the
_ pg ; representing the moisture trans-
mass flow rate M
ferred from the particles to the gas.
The equation for the energy balance for the solid
phase is
dhp
Mp ¼ Q_ gp H_ pg : (2)
dt
The specific enthalpy of the solid particles hp
changes with time in Equation (2) because of the heat
transported from the suspension gas to the particles
Q_ gp ; which allows moisture in the solid to evaporate,
and because of subtraction of the enthalpy associated
with the moisture transported from particles to the
suspension gas, H_ pg :
On the side of the gas the mass and energy balance
equations are:
M_ g ðYg;out  Yg;in Þ ¼ M
_ pg ; (3)


_ g hg;out  hg;in ¼ Q_ gp þ H_ pg Q_ gw :
M (4)
where Yg,in and Yg,out are the absolute humidities in
the inlet and exhaust gas respectively, hg,in and hg,out
are the enthalpies of the inlet and exhaust gas respect-
ively. Equation (3) expresses that the difference
between the mass flow rate of humidity entering and
exiting with the gas is equal to the rate of drying. In
Equation (4), the difference in enthalpy between the
inlet and outlet associated with the gas flow is equal
to the difference of the latent heat associated with the
evaporating water H_ pg ; minus the sensible heat trans-
ported from the gas to the solid Q_ gp and the heat loss
with the environment Q_ gw :
The rates of the various transport phenomena and
the transport coefficients are reported in the
Appendix. Here the model of the drying rate and of
the heat loss with the environment are described.
4 E. GAVI

The drying process of particles is typically repre- Table 1. Composition of formulation A.

sented as a two stage process.[7] During the first dry- Ingredient Fraction on total solids
ing stage, the drying rate is controlled by the rate of Proprietary API Less than 0.1
Lactose Less than 0.6
transport of the vapor from the solid surface to the Fillers Less than 0.3
bulk gas and it is approximately constant, therefore it Binder Less than 0.05
Water (0.29)
is often termed the constant rate period (CRP).
Burgschweiger et al.,[8] however, observe that for
hygroscopic materials, sorptivity may influence the Table 2. Composition of formulation B.
drying rate already in this phase. A constant drying Ingredient Fraction on total solids
Naproxen 0.52
rate may therefore not exist, and[8] preferred referring Lactose 0.17
to this phase as the first drying period. As the drying Avicel PH102 0.21
process continues, a critical moisture content is Povidone K30 0.10
Water (0.22)
reached for which the transport of moisture to the
solid surface is not sufficient anymore to maintain a
The heat loss with the environment Q_ gw can be
saturated surface and the drying rate becomes con-
described as
trolled by resistance to diffusion of moisture within


the solid. After the critical moisture content is reached Q_ gw ¼ UAl Tg  Tamb ; (8)
the rate of drying therefore decreases; this drying
where U is the overall heat loss coefficient, Al is the
period is called the falling rate period (FRP)[7] or
lateral area of the equipment, Tg is the gas tempera-
second drying period.[8]
ture in the equipment, and Tamb is the ambient tem-
The model used in this work described the rate of
drying M _ pg as[8] perature outside the equipment.


M_ pg ¼ _ ðgÞqg bpg Apg Yeq X; Tp  Yg (5)
Experimental details
where Yeq is the humidity at the solid surface, calcu-
Two formulations were used, one containing a propri-
lated from the desorption isotherm curve of the
etary API, the other containing Naproxen. The com-
material as a function of the moisture content of the
position of the two formulations, from now on
solid X and the solid temperature Tp, Yg is the humid-
referred to as Formulation A and B, are reported in
ity in the bulk gas, bpg is the particle-gas mass trans-
Tables 1 and 2. The composition of Formulation A is
port coefficient, qg is the gas density, Apg represents
the total particle-gas contact surface area and mðgÞ_ reported in approximate percentages because it is
represents the normalized single particle drying kinet- proprietary.
The list of the granulation and drying equipment
ics as a function of the particle normalized moisture
content g: used for the experiments is reported in Table 3.
The normalized drying rate allows modeling of the The experiments listed in Table 4 were performed
second drying rate period. It is defined as the quotient at laboratory scale. The wet granules were manufac-
of the drying rate m _ pg and the drying rate of the first tured via high shear wet granulation and then trans-
drying period m_ pg;I for a single particle: ferred in a closed container, sieved through a mill
equipped with a 10 mm screen, transferred to a
m_ pg
_ ðgÞ ¼ : (6) second container, and finally loaded in a pre-heated
_ pg;I
m fluid bed dryer. The mass of wet material in the dryer
The normalized drying rate is a function of the was calculated by weighing the second container
normalized particle moisture content which is defined before and after loading the dryer. As will be
as: explained in the results section, formulation B pre-
 sented different drying behavior when granulation was
XX eq
g¼ ; (7) carried out at laboratory or pilot scale. Therefore, part
Xcr  Xeq
of the wet granules manufactured at pilot scale for the
with X the average moisture content in the solid, Xeq batches B_Pilot1L_T45 and B_Pilot1L_T60 were then
the solid moisture content in equilibrium with the dry dried at laboratory scale in the batches indicated as
gas, Xcr the critical moisture content. It is assumed B_Lab2_T60 and B_Lab2_T45 in the table. The drying

that while X>X _
cr and g>1; mðgÞ ¼ 1; whereas when experiments at laboratory scale were carried out by
 _
drying reaches X<Xcr and g<1; 0<mðgÞ<1: setting the inlet air temperature at a constant value
 DRYING TECHNOLOGY 5

Table 3. Equipment used in experiments.

Granulation lines High shear wet granulator make and model Nominal volume, l Fluid bed dryer make and model Nominal volume, l
Laboratory 1 Diosna P10 10 Glatt GPCG2 6
Laboratory 2 — — Glatt GPCG2 12
Pilot 1 Diosna P150 150 Diosna CAT 150 50 (small)
150 (large)
Pilot 2 Diosna P70 70 Diosna CAP70 70

Table 4. Experiments performed at laboratory scale.

Experiment Dry solids Wet material Inlet temperature Air flow rate
name Formulation granulated (kg) dried (kg) Line ( C) (m3/h)
A_Lab1_T60 A 2 2.2 Laboratory 1 60 70–50
A_Lab1_T70 A 2 2.4 Laboratory 1 70 70–50
A_Lab1_T50 A 2 2.5 Laboratory 1 50 70–50
A_Lab1_T50-55-60 A 2 2.4 Laboratory 1 50, 55, 60 70–50
B_Lab1_T60 B 2 2.1 Laboratory 1 60 70–50
B_Lab1_T50 B 2 2.3 Laboratory 1 50 70–50
B_Lab1_T70 B 2 2.4 Laboratory 1 70 70–50
B_Lab1_T65 B 2 2.4 Laboratory 1 65 70–50
B_Lab2_T60 B — 2 (sample B_Pilot1L_T45) Laboratory 2 (drying only) 60 80–70
B_Lab2_T45 B — 2 (sample B_Pilot1L_T60) Laboratory 2 (drying only) 45 80–70

and by decreasing the inlet air flow rate in a stepwise
manner. In each batch, the inlet air flow rate was
decreased twice as the material became progressively
drier, to avoid over-fluidization of the bed and poten-
tial blockage of the filters. The decision on when to
decrease the inlet air temperature was taken based on
visual observation of the fluidized bed. Additionally to
this, during one batch, A_Lab1_T50-55-60, the inlet
air temperature was increased in a stepwise manner:
in the first 30 minutes of drying, the inlet air tempera-
ture was increased twice, first after 10 minutes from
50 to 55  C, then after 10 minutes from 55 to 60  C.
Granulation and drying experiments were carried
out at pilot scale on two different granulation lines, as
described in Table 5. The wet granules were vacuum
transferred through a mill equipped with a 10 mm
screen to the pre-heated fluid bed dryer. The mass of
wet granules in the dryer was calculated based on the
initial dry solids mass and the added mass of water.
The inlet air temperature was set at a constant value
for the duration of the experiment and the flow rate
was decreased in a stepwise manner. At pilot scale the
inlet air flow rate was determined based on a scale-up
calculation from the inlet air flow rate used at labora-
tory scale. The scale-up calculation aimed at keeping
constant surface velocity across scales, which was cal-
culated as the ratio between volume flow rate and air
distributor section area. The inlet air flow rate was
decreased at pilot scale when the granules moisture
content was reached, at which the inlet air flow rate
was decreased at laboratory scale. The isotherm sorp-
tion curve of each formulation was measured experi-
mentally at 25  C with the Dynamic Vapor Sorption
method. The isotherm curves are shown in Figures 1
and 2 for formulations A and B respectively.
Formulation B is seen to be more hygroscopic than
formulation A when the relative humidity is above
50%. Additionally, formulation B shows a large hys-
teresis during desorption, which may be due to the
formation of a hydrate form. The sorption curve of
the second cycle is however very similar to that of the
first cycle, so the hydrate formation is assumed to be
completely reversible.
The particle size distribution of the dry granules
was measured by passing two samples of each batch
through two different sieve towers, one with eight
sieves between 90 and 1000 mm, the other with eight
sieves between 1000 and 5000 lm. It was observed
that granules of formulation A tend to be mono-dis-
tributed with a d50 around 200 lm, whereas granules
of formulation B tend to be characterized by a bi-
modal distribution, with d50 around 450 lm.
Model set up and calibration
The fluid bed drying model was set up in the com-
mercial software gSolids 4.1 (PSE, London, UK) by
using the software’s standard libraries. The resulting
flowsheet is shown in Figure 3.
The fluid bed dryer was set up as a
batch operation.
The input information to feed to the model is
the following:
1. Material properties (with regards to drying):
desorption isotherm, normalized drying rate and
critical moisture content.
2. Granule properties: granule size distribution.
6 E. GAVI

Table 5. Experiments performed at pilot scale.

Experiment name Formulation Dry solids granulated (kg) Wet material dried (kg) Line Inlet temperature ( C) Air flow rate (m3/h)
A_Pilot2_T55 A 26 33.5 Pilot 2 55 520–350
A_Pilot2_T65 A 26 33.5 Pilot 2 65 520–350
B_Pilot1S_T55 B 30 21 Pilot 1 small 55 540–420
B_Pilot1L_T45 B 30 34.6 Pilot1 large 45 1050–750
B_Pilot1L_T60 B 30 34.6 Pilot1 large 60 1050–750

Figure 1. Isotherm curve of formulation A.

Figure 2. Isotherm curve of formulation B.

3. Equipment geometrical parameters: area of distri-
bution plate of inlet air, lateral area of mater-
ial container.
4. Equipment heat loss parameter: heat loss
coefficient.
5. Process operating and initial conditions: tempera-
ture, flow rate, and humidity of inlet air (which
may be time dependent), initial mass of wet gran-
ules, initial moisture content, initial product
temperature.
The particle size distribution of the granules was
introduced in the model by reporting the sieve ana-
lysis data of the dry granules. The distribution was
represented in the model by using a number of grid
points equal to 20, spaced with a log normal distribu-
tion between 49 and 7001 lm.
The drying rate and the heat loss in the fluid bed
drying model contain parameters that need to be cali-
brated, as follows.
The first step in this work consisted in using
laboratory-scale data to calibrate the normalized dry-
ing rate and the critical moisture content for each for-
mulation, and the heat loss coefficient for the
laboratory dryer.
In order to determine the normalized drying rate,
first the drying rate was calculated from the experi-
mental data; at this time it was determined if a con-
stant rate period and a falling rate period could be
observed. The critical moisture content was set at the

Figure 3. Model flowsheet of the fluid bed dryer set up in gSolids 4.1 (PSE, London, UK).
Figure 4. Normalized drying rate for formulation A.
transition between CRP and FRP. With this informa-
tion the normalized drying rate was calculated using
Equations (6) and (7). This approach was followed for
both formulations, however for formulation B it was
observed that the normalized drying rate so deter-
mined did not lead to good agreement between simu-
lation results and experimental data of the drying
process. The normalized drying rate was therefore set
up as a linear normalized drying rate, effective for the
whole duration of drying
DRYING TECHNOLOGY 7
_ ðgÞ ¼ Ag (9)
where A was a fitting parameter. The coefficient A was
set by manual trial and error, because the software does
not allow to use it as a calibration parameter. In the
results it will be further explained why this approach
seemed to be the most appropriate for formulation B.
The estimation of the heat loss coefficient was per-
formed with the parameter estimation tool available in
gSolids. The measured product temperature and
8 E. GAVI
Figure 5. Comparison of simulated and experimental drying rate for: (a) A_Lab1_T50; (b) A_Lab1_T70.
moisture content (wet basis) were used for the estima-
tion with constant variances of measurement equal to
1  C and 0.005 respectively.
The normalized drying rate determined at labora-
tory scale was used in the pilot scale model, where
only the heat loss coefficient was adjusted with par-
ameter estimation.
Results and discussion
Model calibration at laboratory scale
In this section it is described how model calibration
was carried out for application at the laboratory scale
to characterize the drying behavior of the material
and to obtain the heat loss of the equipment.
The drying rate was calculated for experiments
A_Lab1_T50 and A_Lab1_T70. These experiments
were chosen because they represent the extremes of
the inlet air temperature range covered. It was
observed that at X ¼ 0.0775 a transition regime takes
place from approximately constant to falling drying
rate, therefore the critical moisture content was fixed
to that value in the model. A normalized drying rate
was calculated from the experimental drying rate for
the two experiments and fitted by a second order
polynomial (Figure 4), obtaining the expression
 DRYING TECHNOLOGY 9

Figure 6. Comparison between experimental data and simulation results in terms of product temperature and moisture content:
(a) A_Lab1_T50; (b) A_Lab1_T70.

_ followed by values close to the experiments.

 ¼ 0:51g2 þ 1:55g: (10)
This normalized drying rate was used in the simu-
lation of drying of formulation A granules. A param-
eter estimation was run, including experiments
A_Lab1_T50, A_Lab1_T60, A_Lab1_T70; the heat loss
coefficient U ¼ 53 J/s m2 K was then determined.
In Figure 5, the simulated drying rate is shown and
compared with experimental data for A_Lab1_T50
and A_Lab1_T70. The model describes well the
experimental drying rate, where a constant and a fall-
ing rate period are observed. At the beginning of the
drying process, high drying rates are calculated
Quantitatively, it can be noticed that for the initial
30 minutes of A_Lab1_T50 and the initial 20 minutes
of A_Lab1_T70, the modeled drying rate is higher
than the experimental one. This overestimate may be
due to insufficient mixing at the start of the process.
Towards the end of the process the agreement
improves, although the model slightly underestimates
the drying rate.
In Figure 6 the results in terms of moisture content
and temperature are shown for the same experiments.
The simulated moisture content is in good agreement
with experimental data particularly at the start and at
10 E. GAVI
Figure 7. Comparison between experimental data and simulation results in terms of drying rate for B_Lab2_T60. The effect of
choosing different settings for the normalized drying rate in the model is shown.
the end of the drying process. It is suggested that the
agreement of the model with experiments could be
improved by decreasing the drying rate in the range
X > Xcr. The simulated product temperature trend
agrees qualitatively with experiments, in that a steady-
state is predicted at the start of the drying process,
followed by a temperature increase. However in
experiments it can be observed that the temperature
change takes place more slowly than is predicted by
the model. This behavior might be due to the effect of
the equipment heat capacity that the model does not
take into account. As a consequence, the simulated
temperature at the start of the process decreases rap-
idly, following the introduction of wet granules in the
pre-heated equipment, and it conversely increases rap-
idly at the end of drying, leading to a steady-state
temperature, not observed in experiments. These
observations are also valid for the remaining two
experiments performed at laboratory scale with for-
mulation A, the results of which are not shown
for brevity.
For formulation B, experiment B_lab2_T60 was
used to determine the normalized drying rate. Firstly
it was assumed that drying is limited by external con-
ditions only, and experiments B_lab2_T60 and
B_lab2_T45 were used to determine an approximated
heat loss coefficient. Then the experimental drying
rate was compared with the simulated one (see Figure
7). It can be observed that when only external condi-
tions are considered, the drying rate is higher than
experiments at the start of the process, and then it is
much lower than experiments towards the end of the
process. As a consequence, the moisture content in
the simulations decreases much faster than in the
experiments, and it reaches a lower final value, as
shown in Figure 8. The introduction of a normalized
drying rate at the transition between the observed
constant and falling rate periods did not suffice in this
case to obtain an agreement with experimental data,
due to the higher drying rate present already at the
start of the drying process. Therefore it was decided
to set a linear normalized drying rate, as described in
Equation (9), starting from Xcr ¼ 0.186, i.e., covering
the entire duration of the drying process. The coeffi-
cient A was set to be equal to 0.01 and 0.015 and for
this last value it was possible to find a good agree-
ment both in terms of drying rate and moisture con-
tent, as shown in Figures 7 and 8, respectively.
The chosen approach, which imposes a falling rate
period for the whole drying process, is in disagree-
ment with the experimental observation that can be
made in Figure 7. As mentioned before, however,
Burgschweiger et al.[8] observed that for hygroscopic
materials a constant drying rate may not exist, and
although a constant rate period can be observed for
this specific experiment B_lab2_T60, it is absent in
other experiments with this formulation (as for
example shown in Figure 10(a), which will be com-
mented later). This choice of drying rate for
Formulation B is therefore justified.

Figure 8. Comparison between experimental data and simulation results in terms of moisture content for A_Lab1_T50. The effect
of choosing different settings for the normalized drying rate in the model is shown.
The parameter estimation was then run to deter-
mine the heat loss coefficient, obtaining a result of
U ¼ 15 J/s m2 K. The simulation results for the two
experiments are shown in Figure 9. Since there was
an initial pre-heating phase of the equipment, the
experimental product temperature at the start of the
process was around 30  C. As observed for the previ-
ous formulation, the model predicts quick decreases
or increases of temperature, whereas experimentally
this is not observed. Both in experiments and simula-
tions, there is no extended steady-state period at wet
bulb temperature, possibly due to the slow drying rate
presented by this material. After reaching a minimum
value, the temperature starts increasing, until a steady
state at high temperature is reached. The model can
describe this final steady-state qualitatively for both
experiments. For B_lab2_T60 the model also agrees
quantitatively with experimental data, whereas for
B_lab2_T45 the final temperature is underestimated
by 1 to 2  C. The moisture content is well predicted
by the model in both cases.
It was also attempted to develop the drying model
with experiments performed with granules produced
at laboratory scale and dried in the laboratory equip-
ment Laboratory 1. The experimental drying rate was
calculated and compared with the simulated data
obtained by setting a linear normalized drying rate
using A ¼ 0.015, as had been done for the experiments
performed in the Laboratory 2 equipment. The results
in terms of drying rate and moisture content are
reported in Figure 10 for experiment B_lab1_T60. It
DRYING TECHNOLOGY 11
can be observed that when using A ¼ 0.015, which
was considered appropriate for the granules produced
at pilot scale, the drying rate is under-estimated,
resulting in an over-estimated moisture content by the
model. When A was set to 1, it was instead possible
to better match the experimental drying rate and the
moisture content. This result showed that the material
granulated at pilot scale dries more slowly than that
granulated at laboratory scale. Gupta[16] also observed
that the available water for evaporation is dependent
on the granulation time. In the present case, it is pos-
sible that the same effect resulted from the scale-up of
the granulation process. In the following part con-
cerning the model application at pilot scale, the nor-
malized drying rate determined in the equipment
Laboratory 2 was used.
Simulation of the drying process at pilot scale
In the pilot-scale model the normalized drying rate
and critical moisture content determined at the
laboratory scale for the two formulations were used.
The experimental data collected for the two batches
of formulation A at Pilot2, A_Pilot2_T55 and
A_Pilot2_T65, were used to calibrate the model with
respect to the heat loss coefficient. The initial mois-
ture content was not known at the start of drying,
because sampling was not possible. The drying process
was then simulated starting from the moisture content
value measured on the first sample collected during
the drying process, and the heat loss coefficient was
12 E. GAVI
Figure 9. Comparison between experimental data and simulation results in terms of product temperature and moisture content:
(a) B_Lab2_T45; (b) B_Lab2_T60.
determined by running a parameter estimation with
the two experiments: U ¼ 18 J/s m2 K. In order to
develop a model that could be used predictively, it
was attempted to run the model from the initial dry-
ing rate assumed to be equal to that obtained at the
laboratory scale. By using this initial moisture content
value, the moisture content was overestimated by the
model. Vacuum loading was used to transfer the wet
granules from the granulator to the dryer, therefore it
was hypothesized that the granules undergo drying
during the transfer process that takes place at a higher
rate than that occurring in the fluid bed dryer. To
account for this, a fictitious additional drying time
was added to the real drying process. By trial and
error it was found that if the loading phase is set in
the simulation to last a total of 15 minutes, then the
process can be simulated starting from the initial
moisture content determined at pilot scale.
The simulation of one of the two batches A_Pilot2
_T65 is reported in Figure 11.
It can be observed that the agreement with experi-
mental data is good in terms of moisture content and
very good in terms of temperature. It is interesting to
notice that unlike at the laboratory scale, the model
 DRYING TECHNOLOGY 13

Figure 10. Graphs showing the effect of choosing different coefficient values for the normalized drying rate expression for
B_Lab1_T60: (a) Comparison in terms of drying rate between experimental data and simulation results with two A coefficient val-
ues; (b) Comparison in terms of moisture content between experimental data and simulation results with two A coefficient values.

predicts the temperature increase at the same rate that
is observed experimentally. The closer agreement was
ascribed to the fact that the ratio product volume to
equipment surface area is higher at pilot scale and
therefore the equipment heat capacity has a lower
influence on the temperature profile.
For formulation B, in equipment Pilot1S and
Pilot1L, the heat loss coefficient was estimated to be
U ¼ 8 J/s m2 K, by running a parameter estimation
with the data from the three experiments performed
in that equipment. The results are reported in Figures
12 and 13. The moisture content is well predicted in
Pilot1S and in Pilot1L for B_Pilot1L_T60. In the
experiment at lower temperature, B_Pilot1L_T45, the
agreement for moisture content in the final part of
the process is poor. The experimental data also show
that the moisture reaches briefly a plateau at LOD ¼
4%, before start decreasing again. The product tem-
perature is instead satisfactorily predicted for all three
experiments. As at the laboratory scale, it is observed
that the product temperature does not settle in a
steady-state at the start of drying, and increases
quickly. The model predicts this but overestimates the
product temperature until the temperature plateau is
14 E. GAVI

Figure 11. Comparison between experimental data and simulation results in terms of product temperature and moisture content
for A_Pilot2_T65.

Figure 12. Comparison between experimental data and simulation results in terms of product temperature and moisture content
for B_Pilot1S_T55.

reached. Once the final steady state is reached, the
agreement of the simulated product temperature with
experiments is satisfactory for Pilot1S, with the model
underestimating by 1  C, and good for Pilot1L.
Conclusions
The purpose of this work was to explore the possibil-
ity of reducing experimentation at pilot scale during
development of a fluid bed drying process, by
investigating experimentally the drying behavior of a
pharmaceutical formulation at the laboratory scale
and then by using a mechanistic model of the fluid
bed drying process calibrated at the laboratory scale
to simulate experiments at the pilot scale.
It was observed that the more hygroscopic
Formulation B required imposing a normalized drying
rate for the entire duration of the drying process. On
the contrary, for Formulation A a normalized drying
rate was imposed only at the end of the drying

Figure 13. Comparison between experimental data and simulation results in terms of product temperature and moisture content:
(a) B_Pilot1L_T45; (b) B_Pilot1L_T60.
process, where a transition from a constant to falling
drying rate could be observed.
It was verified that the material drying behavior
could be described in the model with the same
parameters and values of granule properties, when
simulating two different scales of fluid bed dryer. The
drying behavior of the granules of Formulation A was
not affected by granulation conditions. For
Formulation B, by contrast, granules manufactured in
the pilot scale high shear wet granulator dried more
slowly than those manufactured in the laboratory scale
granulator. This indicates that when undertaking
DRYING TECHNOLOGY 15
drying scale-up it is important to verify if the granules
manufactured with a given formulation are sensitive
to the granulation process, and if so, an accurate high
shear wet granulation scale-up approach should also
be developed, to obtain granules with same drying
behavior at different scales.
At the laboratory scale the calibrated model
described well the moisture content and the product
temperature at steady state for formulation B. The
agreement is not as good for Formulation A. For this
formulation, the agreement in terms of moisture con-
tent may be improved by modifying the normalized
16 E. GAVI

drying rate also in the constant rate period, possibly Shpg Sherwood number
to describe a non-ideal bed movement at the start of t time, s
Tamb ambient temperature, K
drying. The measured product temperature for this
Tg temperature of gas, K
formulation shows that the model is not able to Tp temperature of bed of solid particles, K
describe the dynamic behavior of the product tem- u0 particle surface velocity, m/s
perature. This could be improved by resorting to a U overall heat loss coefficient, J/s m2 K
dynamic heat loss model.[17] This phenomenon is less X mass fraction of water in solids (wet based),
kg/kg
important for Formulation B, where the rate of drying 
X average moisture content in solids, kg/kg
is slower and the product temperature reaches steady Xcr critical moisture content in solids, kg/kg
state at the end of drying, and also at pilot scale, Xeq solid moisture content in equilibrium with dry
where the effect of the equipment heat capacity gas, kg/kg
is lower. Yeq equilibrium absolute humidity, g water/kg dry gas
At the pilot scale the model, which was calibrated Yg;out absolute humidity in exhaust gas, g water/kg
dry gas
at laboratory scale with regards to the formulation Yg;in absolute humidity in inlet gas, g water/kg dry gas
drying behavior, simulates well the drying process of apg heat transfer coefficient, W/m2 K
both formulations. bpg particle-gas mass transport coefficient, m/s
These results show that it is possible to study dg conductivity of the gas, W/m K
experimentally the fluid bed drying process at the lg gas viscosity, Pa
g normalized moisture content, -
laboratory scale and to use a mechanistic model cali- _ normalized single particle drying rate, -
brated on laboratory scale data, to simulate the pro- qg gas density, kg/m3
cess at the pilot scale, potentially leading to
considerable savings in terms of API used in the
Acknowledgments
experimental campaign for developing the drying pro-
cess of a new pharmaceutical product granulated via The author would like to acknowledge Pranay K. Ghosh,
high shear wet granulation. Susanne Page, Oskar Kalb and Dharam Singhal for the sup-
port given through discussions and feedback, Davide De
Franco and J€urgen H€orth for support in performing experi-
Nomenclature ments, and David Slade (PSE, London, UK) for technical
A fitting parameter support in using the software and technical discussions.
Abed bed section area, m2
Al lateral surface area of equipment, m2
Apg total particle-gas surface contact area, m2
References
cg specific heat capacity of the gas, J/(kg K) [1] Yu, L. X. Pharmaceutical Quality by Design: Product
cw;g specific heat capacity of water vapor, J/(kg K) and Process Development, Understanding and
dp particle diameter, m Control. Pharm. Res. 2008, 25, 781–791. DOI:
D diffusivity of moisture in suspension gas phase, 10.1007/s11095-007-9511-1.
m2/s [2] Gabbott, I. P.; Husban, F. A.; Reynolds, G. K. The
hg;in specific enthalpy of inlet gas, J/kg Combined Effect of Wet Granulation Process
hg;out specific enthalpy of exhaust gas, J/kg Parameters and Dried Granule Moisture Content on
hp specific enthalpy of solid particles, J/kg
Tablet Quality Attributes. Eur. J. Pharm. Biopharm.
Dhv specific heat of evaporation, J/kg
2016, 106, 70–78. DOI: 10.1016/j.ejpb.2016.03.022.
H_ pg enthalpy associated with the moisture transported
[3] Snider, B.; Liang, P.; Pearson, N. Implementation of
from particles to the suspension gas, J/s
Water Activity Testing to Replace Karl Fischer Water
_ pg
m single particle drying rate, kg/s
_ pg;I
m single particle drying rate in the first drying Testing; Pharmaceutical Technology, 2007, pp 56–71.
period, kg/s [4] Bahu, R. E. Fluidised Bed Dryer Scale-up. Dry.
_g
M gas mass flow rate, kg/s Technol. 1994, 12, 329–339. DOI: 10.1080/
Mp mass of solid particles in the fluidized bed, kg 07373939408959959.
_ pg
M moisture transferred from solid particles to gas or [5] Groenewold, H.; Tsotsas, E. A New Model for Fluid
total drying rate, kg/s Bed Drying. Dry. Technol. 1997, 15, 1687–1698.
Nu Nusselt number DOI: 10.1080/07373939708917318.
Pr Prandtl number [6] Kemp, I. C.; Oakley, D. E. Modelling of Particulate
Q_ gp heat flow from suspension gas to solid particles, Drying in Theory and Practice. Dry. Technol. 2002,
J/s 20, 1699–1750. DOI: 10.1081/DRT-120015410.
Q_ gw heat loss with the environment, J/s [7] Mujumdar, A. S. Handbook of Industrial Drying;CRC
Re0 particle Reynolds number at surface velocity Press, Taylor and Francis Group: Boca Raton, FL;
Sc Schmidt number 2007.
 DRYING TECHNOLOGY 17

[8] Burgschweiger, J.; Groenewold, H.; Hirschmann, C.; 1999, 17, 1557–1570. DOI: 10.1080/
Tsotsas, E. From Hygroscopic Single Particle to 07373939908917635.
Batch Fluidized Bed Drying Kinetics. Can. J. Chem.
Eng. 1999, 77, 333–341. DOI: 10.1002/
cjce.5450770220. Appendix
[9] Wang, H. G.; Dyakowski, T.; Senior, P.; Raghavan,
R. S.; Yang, W. Q. Modelling of Batch Fluidised Bed Rates of transports and transport coefficients
Drying of Pharmaceutical Granules. Chem. Eng. Sci. In the drying rate M_ pg reported in Equation (5), the coeffi-
2007, 62, 1524–1535. DOI: 10.1016/j.ces.2006.11.047. cients of transport are calculated via the following
[10] Wang, H. G.; Yang, W. Q.; Senior, P.; Raghavan, equations[18,19]
R. S.; Duncan, S. R. Investigation of Batch Fluidized-  
Bed Drying by Mathematical Modeling, CFD bpg dp Re0 Sc ShApg =Abed
Shpg ¼ ¼ ln 1 þ (A1)
Simulation and ECT Measurement. AIChE J. 2008, D Apg =Abed Re0 Sc
54, 427–444. DOI: 10.1002/aic.11406.
where
[11] Jang, J.; Arastoopour, H. CFD Simulation of a
Pharmaceutical Bubbling Bed Drying Process at dp u0 qg
Re0 ¼ ; (A2)
Three Different Scales. Powder Technol. 2014, 263, lg
14. DOI: 10.1016/j.powtec.2014.04.054. 1=2
[12] Kemp, I. C. Drying Models, Myths and Sh ¼ 2 þ 0:6Re0 Sc1=3 ; (A3)
Misconceptions. Chem. Eng. Technol. 2011, 34, lg
Sc ¼ ; (A4)
1057–1066. DOI: 10.1002/ceat.201100061. qg D
[13] He, X.; Pinto, M. Modelling of Fluid Bed Drying at
Different Production Scales in Pharmaceutical and the symbols represent: dp particles diameter, D diffusiv-
Manufacturing. In Advanced Process Modelling ity of moisture in suspension gas phase, qg and lg density
Forum; 2013, London, UK. and viscosity of the gas respectively, Apg total surface area
[14] Kemp, I. C.; Sohet, Q. Scale-up, Optimization, and of particles, Abed bed section area, u0 gas surface velocity.
Control of Industrial Batch Fluidized Bed Dryers The rate of transport of enthalpy from the particles to
Using Multilevel Theoretical Models. Dry. Technol. the suspension gas, in Equations (2) and (4), is expressed
2010, 28, 710–722. DOI: 10.1080/ as[8]


07373931003799285. H_ pg ¼ M
_ pg cw;g Tp þ Dhv (A5)
[15] Pinto, M.; Kemp, I.; Bermingham, S.; Hartwig, T.;
Bisten, A. Development of an Axisymmetric where cw,g is the specific heat capacity of water vapor, Tp is
Population Balance Model for Spray Drying and the particle temperature and Dhv is the specific heat of
Validation against Experimental Data and CFD evaporation.
Simulations. Chem. Eng. Res. Des. 2014, 92, 619–634. Finally the rate of transport of heat from particles to the
DOI: 10.1016/j.cherd.2013.12.019. gas, in Equations (2) and (4), is expressed as[6]
[16] Gupta, R. Fluid Bed Granulation and Drying, In

Q_ gp ¼ apg Apg Tp  Tg ; (A6)
Predictive Modeling of Pharmaceutical Unit
Operations; Pandey, P., Bharadwaj, R., Eds.; where Tg is the gas temperature and the heat transfer coeffi-
Woodhead Publishing, 2017 pp 137–158. cient apg is obtained from the following equations[5]:
[17] Garcıa-Mu~noz, S.; Moreno, M.; Cano, A.; Leyland, S.  
Modeling the Dynamics of the Film Coating apg dp Re0 Pr NuApg =Abed
Nups ¼ ¼ ln 1 þ ; (A7)
Operation for Pharmaceutical Tablets. AIChE dg Apg =Abed Re0 Pr
Annual Meeting, October 2012, Pittsburgh, PA.
with
[18] Burgschweiger, J.; Tsotsas, E. Experimental
1=2
Investigation and Modelling of Continuous Fluidized Nu ¼ 2 þ 0:6Re0 Pr1=3 ; (A8)
Bed Drying under Steady-State and Dynamic cg lg
Conditions. Chem. Eng. Sci. 2002, 57, 5021–5038. Pr ¼ ; (A9)
dg
DOI: 10.1016/S0009-2509(02)00424-4.
[19] Groenewald, H.; Tsotsas, E. Predicting Apparent and where cg and dg are the specific heat and the conductiv-
Sherwood Numbers for Fluidized Beds. Dry. Technol. ity of the gas respectively.