RESEARCH ARTICLE

Drying Process Optimization for an API Solvate Using Heat

Transfer Model of an Agitated Filter Dryer
NANDKISHOR K. NERE, KIMBERLEY C. ALLEN JAMES C. MAREK, SHAILENDRA V. BORDAWEKAR
Process Research and Development, Global Pharmaceutical Research and Development, Abbott Laboratories, North Chicago,
Illinois 60064

Received 8 November 2011; revised 24 May 2012; accepted 30 May 2012

Published online in Wiley Online Library (wileyonlinelibrary.com). DOI 10.1002/jps.23237

ABSTRACT: Drying an early stage active pharmaceutical ingredient candidate required ex-
cessively long cycle times in a pilot plant agitated filter dryer. The key to faster drying is to
ensure sufficient heat transfer and minimize mass transfer limitations. Designing the right
mixing protocol is of utmost importance to achieve efficient heat transfer. To this order, a com-
posite model was developed for the removal of bound solvent that incorporates models for heat
transfer and desolvation kinetics. The proposed heat transfer model differs from previously
reported models in two respects: it accounts for the effects of a gas gap between the vessel wall
and solids on the overall heat transfer coefficient, and headspace pressure on the mean free path
length of the inert gas and thereby on the heat transfer between the vessel wall and the first
layer of solids. A computational methodology was developed incorporating the effects of mixing
and headspace pressure to simulate the drying profile using a modified model framework within
the Dynochem software. A dryer operational protocol was designed based on the desolvation
kinetics, thermal stability studies of wet and dry cake, and the understanding gained through
model simulations, resulting in a multifold reduction in drying time. © 2012 Wiley Periodicals,
Inc. and the American Pharmacists Association J Pharm Sci
Keywords: desolvation; drying; mixing; kinetics; solvate

INTRODUCTION ist several parameters that can be varied to achieve

optimal drying performance, including the extent
Drying of an active pharmaceutical ingredient (API)
and intensity of mixing, headspace pressure, jacket
is an important unit operation in the pharmaceuti-
temperature, inert gas temperature, and flow rate.
cal industry as it is used to control residual solvent
Nevertheless, there is no universal drying protocol
content, in addition to preserving or producing the
applicable to all pharmaceutical compounds, due to
right crystal form. Drying operation is often time con-
variations in physicochemical properties of the wet
suming, and may turn out to be a bottleneck in the
cake and the desired dried product attributes. Be-
overall process cycle time. Many APIs are potent com-
cause of these challenges, modeling can be of great
pounds that require filtration and drying to be done
utility in designing dryer operating conditions. In-
in a single contained vessel, known as a filter dryer,
sights from the modeling can reduce experimental ef-
to minimize operator exposure. The present work fo-
forts, thereby saving expensive material required to
cuses on the agitated filter dryer wherein the dryer is
understand the drying process at bench scale and also
equipped with an agitator for mixing the solids and
ensure predictable robust performance on scale-up.
promote better heat transfer. Given a fixed geomet-
In the present paper, we demonstrate an approach
ric configuration of an agitated filter dryer, there ex-
that was used for a significant reduction in drying
time of an API ethanol bis-solvate. We will refer to
Correspondence to: Nandkishor K. Nere (Telephone: +91- ethanol as “solvent” and the bis-solvate as “solvate”
22-2740-3140; Fax: +91-22-2740-3140; E-mail: nandkishor.nere@
adityabirla.com) in the discussion. In an early pilot plant campaign
Nandkishor K. Nere’s present address is Aditya Birla Science (batches I and II), the drying process took nearly a
and Technology Company Ltd., Panvel, Maharashtra, India. week to achieve an acceptable solvent content of less
Kimberley C. Allen is currently not affiliated with Abbott Lab-
oratories. than 0.1% (w/w). The dryer was operated at a nominal
Journal of Pharmaceutical Sciences jacket temperature of 50◦ C, which resulted in a maxi-
© 2012 Wiley Periodicals, Inc. and the American Pharmacists Association mum internal temperature of about 40◦ C. The outline

JOURNAL OF PHARMACEUTICAL SCIENCES 1

2 NERE, ALLEN JAMES C. MAREK, AND BORDAWEKAR
Figure 1. PXRD patterns indicate the form conversion from solvate to anhydrous form during
the course of drying. Red line, initial wet cake; blue line, partially converted solvate; green line,
anhydrous form.
of this paper is as follows: We present a brief analysis
of the drying using powder X-ray diffraction (PXRD)
patterns of solids and monitoring of the exit gas using
mass spectrometry, followed by the estimation of in-
trinsic desolvation kinetics. We describe the various
transport processes, and their interactions during the
course of drying, that were used to guide development
of a composite model. A heat transfer model similar
to that of Tsotsas and Schlünder1 is described, which
is followed by the algorithm for simulating the dry-
ing profile in a filter dryer. Prediction of past drying
performance is presented, followed by optimization of
the drying process and the subsequent verification.
Finally, we propose general guidelines to operate fil-
ter dryers for efficient drying.
Process Description
A typical agitated filter dryer is equipped with a
heated agitator in addition to the heated jacket. Inter-
nal diameter of the pilot plant filter dryer is 0.6 m and
the agitator diameter is 0.58 m. The maximum possi-
ble agitator speed in the filter dryer is approximately
20 rpm. Because of hydraulic controls, start-up speed
generally goes high and settles on the control speed
after few seconds. Besides raising or lowering the agi-
tator to the desired axial location, the direction of ag-
itator rotation can also be changed to achieve either
cake mixing or smoothing of the cake surface. Nitro-
gen gas can be purged through the cake at a desired
flow rate and vacuum can be applied below the bottom
filter plate. Headspace pressure and solids tempera-
ture can be monitored online. Total heat transfer area
provided by the filter dryer includes the surface area
JOURNAL OF PHARMACEUTICAL SCIENCES
provided by the heated agitator, in addition to the
jacketed surface in contact with the cake. In the case
under consideration, the dryer was operated at max-
imum fill capacity, thereby utilizing the entire heat
transfer area afforded by the jacket.
Desolvation Kinetics
Before describing the modeling of the drying process,
it is of interest to briefly discuss relevant studies car-
ried out using a process analytical technology tool
to understand the drying process at laboratory scale.
Figure 1 shows the PXRD patterns that demonstrate
crystal form conversion (from solvate) to anhydrous
API during the course of drying. The analysis of crys-
tal structure obtained using a single solvate crystal
identified it as a bis-solvate of API. Online mass spec-
trometry analysis of the exit gas from the dryer op-
erated at two different temperatures showed signifi-
cant enhancement in the solvent removal rate at 60◦ C
compared with 40◦ C. This also indicates that the des-
olvation is accelerated beyond a certain temperature,
which is certainly above the nominal internal dry-
ing temperature of 40◦ C observed in the early pilot
plant campaign. Intrinsic kinetics of desolvation was
obtained by tracking the solvent content as function
of time in dynamic vapor sorption (DVS) experiments
carried out at three different temperatures under ni-
trogen purge. In the DVS experiment, 50 mg of mate-
rial was placed in a pan as a thin layer and kept at a
constant temperature of 40◦ C. The loss of weight was
tracked as a function of time to calculate the rate of
solvent loss, which is readily related to the rate of
desolvation. The dynamic desolvation data were
then fitted to first-order kinetics to extract the rate
DOI 10.1002/jps
 DRYING PROCESS OPTIMIZATION FOR AN API SOLVATE IN AGITATED FILTER DRYER
Figure 2. Arrhenius plot for the determination of desol-
vation kinetics. Filled diamonds, experimentally measured
rate constants; , fit; equation of line, ln(kD ) = −0.0002T
+ 0.0026.
constant. The experiment was repeated at two more
temperatures of 60◦ C and 70◦ C. The desolvation rate
constants were obtained at three temperatures, which
were used to obtain the Arrhenius plot as shown in
Figure 2. The temperature dependence of the desolva-
tion rate constant kD (T) was obtained from the equa-
tion of fitted line. We defer the detailed explanation
of the rate constant in the context of broader process
to the later section on Composite Drying Model. In
order to determine the maximum acceptable temper-
ature for drying, thermal stability studies were car-
ried out on the wet cake. It was found that the wet
cake begins to change color at temperatures above
80◦ C, indicating thermal degradation. The transfor-
mation of the API solvate to an amorphous state was
only observed in cases where the ethanol content was
greatly in excess of the stoichiometric amount for sol-
vation, essentially from improperly deliquored wet
cakes. Transformation of the desolvated ethanolate
form to the amorphous state was not observed in in-
dependent laboratory studies. The API desolvate was
in fact the stable form at elevated temperatures, at
least until the melting point (ca. 150◦ C) was reached.
The desolvation kinetics suggests that the drying rate
can be significantly enhanced by increasing the tem-
perature above the nominal operating temperature of
40◦ C. However, the maximum temperature should be
limited to 80◦ C to avoid thermal degradation and/or
amorphosization. The task at hand was to identify a
temperature profile that would facilitate an increase
in the desolvation rate without compromising product
quality. With these considerations, it was thought de-
sirable to increase the jacket temperature gradually
during the course of drying, with an upper limit of
70◦ C.
Composite Drying Model
In order to formulate a model, it is useful to define the
various steps in the drying process. Figure 3 shows
the schematic of various rate processes, along with
DOI 10.1002/jps
3
Figure 3. (a) Schematic of desolvation and evaporation
rate processes. (b) Schematic of drying phases in the com-
posite model.
their interactions that form the basis of the drying
model used for simulation and optimization. Phase I
consists of evaporation of the unbound solvent, which
is relatively fast and can be easily ensured by proper
cake deliquoring and nitrogen blow down under vac-
uum. Hence, this would not be considered further for
the modeling purposes. We assume that the rate of
desolvation is insignificant during phase I due to a
higher energy barrier for desolvation. We focus here
on the slowest drying step, designated as a combi-
nation of phases II and III, which represent the re-
moval of bound solvent primarily present in solvated
form for the compound under consideration. There are
three key steps in the removal of bound solvent. The
first step is the desolvation of the solvate, the rate of
which is governed by temperature. We assume that
the desolvation results in the separation of solvent in
liquid form. The second step is the evaporation of the
free liquid solvent after desolvation. Subsequently, in
the third step, the evaporated solvent is convected
away by the inert gas. We assume that the inert gas
is flowing around the particles at a sufficiently high
velocity to alleviate diffusion limitations in the third
step. The desolvation and evaporation rates depend
on the heat transfer rate, which is governed by the
physical properties of the cake, jacket fluid temper-
ature, heat transfer area, and the mixing provided
by the agitator. Desolvation is reversible in nature;
however, we resort to its representation in terms of
the net desolvation rate constant denoted by kD for
the sake of model simplification and due to intrinsic
difficulties associated with the characterization of its
reversibility. The corresponding net desolvation rate
JOURNAL OF PHARMACEUTICAL SCIENCES
4 NERE, ALLEN JAMES C. MAREK, AND BORDAWEKAR

is indicated by RD . The solvent evaporation rate is

denoted by Ẋ C , which is determined by heat transfer
rate into the solids, Qin . The heat transfer is a function
of the overall heat transfer coefficient and the temper-
ature driving force. On the basis of this mechanism,
rapid withdrawal of solvent vapor by providing suffi-
cient inert gas flow rate will result in desolvation and
evaporation being the rate-determining steps. Both
desolvation and evaporation are governed by local
temperature and hence the underlying heat trans-
fer rate. Thus, the effective rate of solvent loss will be
the smaller of the two rates, that is, the desolvation
rate (RD ) and the solvent evaporation rate (Ẋ C ). This
was incorporated in the simulation methodology. At a
given temperature T, the rate of desolvation is readily
calculated using the equation, RD = kD (T)Cs, b , where
Cs,b is the ratio of the mass of the bound solvent in
the form of solvate to the mass of anhydrous solids.
Now we shed light on determination of the over- Figure 4. Schematic representation of temperature gra-
all heat transfer coefficient, which is a key parameter dients in filter dryer. , vessel wall; , first layer of solids
that acts in conjunction with the temperature differ- and gas gap; , dry solids; , wet solids.
ence (i.e., driving force) to determine the rate of heat
transfer. The heat transfer model based on the pen- heat transfer coefficient for the wall and hence the
etration theory proposed by Tsotsas and Schlünder1 1
corresponding resistance hwall is negligible in compar-
along with a few modifications forms the backbone of ison to the other resistances, and can be neglected.
the heat transfer model used for simulations. For the This also means that the wall temperature is nearly
sake of completeness, we describe the modified ver- equal to the jacket temperature (i.e., Tj = TW ). With
sion of the heat transfer model applied to the system these assumptions, the equations for heat transfer
under consideration. The mechanism for heat trans- rate and the overall heat transfer coefficient simplify
fer is assumed to be dominated by thermal conduc- to Eqs. 2 and 3:
tion. The temperature gradients in a filter dryer are
depicted schematically in Figure 4. The overall gra- Q = UA(TW − Tbed ) (2)
dient is assumed to have contributions from (i) the
vessel wall, (ii) the gap between the outer layer of dry and
solid bed and the vessel wall, which is assumed to be
composed of the first layer of solids and the net gas 1 1 1
=  + (3)
gap, and (iii) the dry solids. The third layer of solids, U hws hsb,wet
composed of wet solids, is assumed to be at a uniform
temperature. A key difference between the present study and
Thus, the rate of heat transfer is given by Q = UA(Tj earlier published literature is the modified approach
− Tbed ) where the overall heat transfer coefficient U, to calculating the effective contact heat transfer co-
follows from Eq. 1: efficient (h ws ). We assume the equation given by
Hoekstra et al.2 to hold for the drying in the presence
1 1 1 1 of agitation, wherein the solids are pushed toward the
= +  + (1)
U hwall hws hsb,wet vessel wall, leaving behind no inert gas gap along the
vessel periphery. However, in the absence of agitation,
where hsb, wet is the heat transfer coefficient for the there is a higher likelihood that a gas gap will form,
solid bed and represents the inverse of penetration created by preferential flow of inert gas along the ves-
resistance, h ws is the contact heat transfer coefficient sel wall. This would offer additional resistance to heat
for the first layer of solids near the vessel wall and transfer and needs to be accounted for in the model.
accounts for a gas gap between the outer layer of Thus, the resistance to heat transfer from the vessel
dry solid bed and the vessel wall, and hwall is the wall to the outer layer of dry solid bed is assumed to
heat transfer coefficient for the vessel wall. The heat be composed of the first layer of solids on the vessel
transfer coefficient for the vessel wall is given by wall (also termed contact resistance) and the gas gap,
hwall = kmetal
dw
, where kmetal is the thermal conductivity and can be represented as two resistances in series.
of the vessel wall and dw is the thickness of the wall. The effective resistance is then the summation of two
High metal conductivity results in a relatively high resistances, and its inverse represents the effective

JOURNAL OF PHARMACEUTICAL SCIENCES DOI 10.1002/jps

 DRYING PROCESS OPTIMIZATION FOR AN API SOLVATE IN AGITATED FILTER DRYER
heat transfer coefficient, h ws . We adopt the equation
given by Hoekstra et al.2 to calculate the contact heat
transfer coefficient, hws as:
  2kg /dp
hws = ψ a hwp + 1 − ψ a √ (4)
2 + 2 (lmf + lR ) /dp
where ψ a is the fractional vessel wall coverage by
solids; hwp is the heat transfer coefficient for a single
particle; kg is the nitrogen gas conductivity, and lmf ,
lR , and dP are the mean free path length for nitrogen
gas molecules, surface roughness of the vessel wall,
and particle diameter, respectively.
We denote the heat transfer coefficient correspond-
ing to the gas gap by hgg . The corresponding resis-
tance is given by h1gg , which can be easily calculated
k tion of the agitation profile throughout the course of
using, hgg = dngg , where dng is the net gas gap. The net
gas gap is defined as the clearance between the first
layer of solids on the vessel wall and the outer layer
of solid bed.
Thus, the total resistance is given by h1 = h1ws + h1gg ,
ws
whereby it follows that
 −1 

1   2kg /dp kg −1
 = ψ a hwp + 1 − ψ a √ +
hws 2 + 2 (lmf + lR ) /dp dng
(5)
The net gas gap is equal to zero in the presence of
agitation and under that condition Eq. 5 reduces to
the one given by Hoekstra et al.2 The calculation of
the heat transfer coefficient for a single particle, hwp
follows from Schlüder and Mollekopf.3
Note that we neglect the radiative heat trans-
fer mode due to the inherently low jacket tempera-
tures typically used in API drying. The mean free
path (lmf ) was calculated using the following equation
lmf = √k2Bd
B TW
2P
, where kB is the Boltzmann constant and
g
dg is the diameter of nitrogen gas molecule. Use of this
equation, instead of a fixed value of lmf corresponding
to an average pressure, allows us to simulate the ef-
fect of dynamic pressure, P, in the filter dryer. The
variation in headspace pressure may be significant
given the dynamics of pressure drop across the cake,
inert gas flow rate, and the vacuum drawn from the
bottom of the filter dryer. It can be readily calculated
from the Eq. 5 that the increase of headspace pressure
from 50 to 1000 mbar results in an order of magni-
tude decrease in mean free path length and the in-
crease in heat transfer coefficient for the wall to first
layer of solids by a factor of 2.5. Thus, it is clear that
the headspace pressure can have a significant effect
on the heat transfer coefficient and hence deserves
due consideration throughout the drying simulation.
It should also be noted that the increase in headspace
DOI 10.1002/jps
5
pressure will result in an increase in boiling point of
the solvent. This will result in decreased driving force
for the heat flux from the jacket. Thus, an increase in
headspace pressure increases heat transfer coefficient
and decreases temperature-driving force, both being
counteracting in deciding the effective heat flux that
governs the solvent evaporation rate. It is also worth
noting that the contribution of a gas gap to overall
heat transfer coefficient ranges from 10% to 50%, de-
pending on the fraction of wet solids, which justifies
its incorporation in the heat transfer model.
During the course of drying, there are periods of
intermittent agitation. Hence, the heat transfer coef-
ficients must be determined for two separate scenar-
ios, one during the mixing and one in the absence of
mixing. These must be used appropriately as a func-
simulation.
Now, we consider determination of the solid bed
heat transfer coefficient hsb, wet , which bears depen-
dence on the mixing. Although its determination in
the absence of mixing is straightforward, we need
to apply the penetration model proposed by Tsotsas
and Schlünder1 to calculate hsb, wet during a period of
mixing. For sake of completeness, we revisit the heat
transfer model to calculate various heat transfer co-
efficients involved in the computational methodology.
Heat Transfer Model of Tsotsas and Schlünder
In a penetration model proposed by Tsotsas and
Schlünder1 , the mixing period is assumed to be com-
posed of a sequence of unsteady steps separated by
fictitious periods of contact time, tc during which the
bed is assumed to be static. A distinct drying front
penetrates from the hot surface into the bulk and rep-
resents a moving front. It is assumed that all the par-
ticles between the hot surface and the moving front
are dry and the ones beyond the moving drying front
are wet. At the end of the contact time, the bulk is
assumed to be perfectly mixed. The contact time is
a function of mixing and bears a relationship, tc =
tmix Nmix where Nmix is the mixing number, which is
(2BN)2 Dimp
empirically related to Froude number, Fr = 2g
by Nmix = C(Fr)d , where the constants, C and d are
characteristics of the agitator type.
Consider a scenario at any time t within the contact
time tc (0 < t < tc ) where the drying front is located at a
distance x = xF from the vessel wall. The layer beyond
the drying front is assumed to have uniform tempera-
ture and solvent content. The transient temperature
for a dry layer can be described by the Fourier equa-
tion along with the appropriate boundary conditions
as follows:
∂2 T ∂T
ksb,dry 2
= ρ s Cp,s (6)
∂x ∂t
JOURNAL OF PHARMACEUTICAL SCIENCES
6 NERE, ALLEN JAMES C. MAREK, AND BORDAWEKAR

where ksb, dry is the dry bed conductivity, ρs is the For completely dry bed, .goes to infinity and hence
solids density, and Cp, s is the specific heat capacity. erf.becomes unity. Hence,
The equation is subjected to the boundary conditions: 
(i) at x = 0, T = TD , which assumes the edge of first 
2 kcP,s ρ bed
layer of solids near the vessel wall at a constant tem- hsb,dry =√ (13)
B tc
perature, TD and (ii) at x = xF , T = Tbed representing
the temperature at the drying front or the interface
It readily follows from Eqs. 12 and 13 that hsb,wet =
between dry and wet solids. The dynamic evolution of hsb,dry
.
the distance of the drying front is described by erf .
Thus, the bed overall heat transfer coefficient is
given by Eq. 14:
∂xF ksb,dry ∂T
=− (7)     
∂t ρ s XHEvp ∂x 1 1 hws
=  1+ −1 . (14)
U hws hdry
where X is the solvent content and ΔHEvp is the heat
of solvent evaporation. Hence, the heat transfer rate into the “wet part” of
The analytical solution to Eq. 6 follows from the cake, determined by the temperature gradient at
Neumann4 and is given by the boundary between the dry and wet cake, is given
by



T − Tbed xF ρ s Cp,s x ρ s Cp,s 

erf = erf (8) Q in = UA (TW − Tbed ) exp −ζ 2 (15)
TD − Tbed 2 ksb,dry t 2 ksb,dry t

and in turn is used to estimate the possible sol-

The reduced distance
  of the drying front can be de- vent evaporation rate as Ẋ C = H Q in
. The change
ρ s Cp,s Evp
fined as ζ = erf xF
2 ksb,dry t
, which can be calculated in average temperature T in a contact time tc
by using Eq. 9: corresponding to a change in the solvent content,
min(Ẋ C ,RD )tc
X = , after perfect mixing can be calcu-
√    
hws
  W
lated by subtracting the heat used for solvent evap-
Bζ exp ζ 2 1 + − 1 erf ζ
hdry oration, Qin given by Eq. 15 from the total heat sup-
   plied, Q given by Eq. 2. The equation for T is as
Cp,s (TW − Tbed ) hws
= −1 (9) follows:
XHEvp hdry
HEvp   
T = X exp ζ 2 − 1 (16)
where, hdry is the coefficient for heat transfer from the cp,s + cp,l X
hot surface to the drying front, and it is composed of
contributions from h ws and the heat transfer coeffi- where cp, s , cp, l are the specific heat capacities of solids
cient for dry solid part of solid bed, hsb, dry according and the solvent (liquid), respectively.
to: Model Simulations and Optimization

1 1 1 As shown earlier, the rate of heat transfer is deter-

=  + (10) mined by the overall heat transfer coefficient, which
hdry hws hsb,dry
is governed by various contributing coefficients. Al-
though the heat transfer coefficient corresponding to
The temperature profile given by Eq. 8 can be used the first layer of solids is fixed for a specified state of
to calculate the penetration heat transfer coefficient mixing (i.e., on or off) due to its dependence on phys-
for the wet part of the bed by using following integral ical properties, the heat transfer coefficients hsb, dry
equation: and hsb, wet are variables. To calculate these, a corre-
lation for mixing number has to be established. To
tc   this end, the constants C and d characterizing the
1 ksb,dry ∂T
hsb,wet = dt (11) mixing number that depend on the agitator geometry
tc (TD − Tbed ) ∂x x=0
0 were estimated by forcing a reasonable fit of the model
predictions with the experimental data collected from
Thus, from Eqs. 8 and 11, we get the early pilot plant batch II. All simulations were car-
ried out using modified model framework within the
  Dynochem R
software.5 Impeller diameter of 0.58 m,
2 kcP,s ρ bed 1 solvent and solids specific heats of 2.85 and 2.00 kJ
hsb,wet =√ (12)
B tc erf . kg−1 K−1 , respectively, solid density of 1320 kg m−3 ,

JOURNAL OF PHARMACEUTICAL SCIENCES DOI 10.1002/jps

 DRYING PROCESS OPTIMIZATION FOR AN API SOLVATE IN AGITATED FILTER DRYER
Figure 5. Model fit of the observed performance in the
early pilot plant campaign (batch-II). Filled diamonds, pilot
plant data; , model fit.
and the bed thermal conductivity of 0.08 W m−1 K−1
were used for the simulations. Predictions of the mea-
sured drying profile (i.e., solvent content) and solids
temperature as a function of time were used to judge
the quality of fit. The fitting exercise was conducted
using the actual agitation and headspace pressure
profiles. The drying profiles for solvent content and
solids temperature depicted in Figures 5 and 6, re-
spectively, correspond to batch II from the early pilot
plant campaign. Also the starting solvent composi-
tion (5.4%, w/w) used for the fitting was chosen based
on the first available measurement and it falls well
below the bis-solvate composition of 8.8% (w/w). The
predicted solvent content is in good agreement with
the measured values as shown in Figure 5. Further-
more, it can be seen from Figure 6 that the predicted
temperature profile is also in reasonable agreement
with the measurements. Note the initial phase where
solids temperature increases before attaining a some-
what steady value. From Figures 5 and 6, it can be
noted that the time at which temperature assumes a
somewhat steady value corresponds to mono-solvate
Figure 6. Comparison of measured and fitted internal (solids) temperature profile (early pilot
plant campaign, batch-II). Diamonds, measured internal temperature;
temperature.
DOI 10.1002/jps
7
composition. We believe that the first solvent molecule
of the bis-solvate has a lower energy barrier for des-
olvation compared with the second solvent molecule.
Thus, during removal of the first solvent molecule,
evaporation of the desolvated solvent may be the rate-
determining step, thereby increasing the solids tem-
perature. It should also be noted that the temperature
probe resides on the dryer wall and is in contact with
the gas flowing through the gap near the wall and
the solids. Thus, the measured temperature does not
exactly match the dry solids temperature predicted
by the model. This is also supported by the observed
jumps in the measured temperature during the mix-
ing period as shown in Figure 6. The values of con-
stants C and d that resulted in good predictions of
time profiles for both the solvent content and tem-
perature were found to be 20 and 0.25, respectively.
These were kept constant in all of the subsequent
simulations.
It follows that there are two parameters that can be
varied to enhance the heat transfer rate, one being the
jacket temperature (which decides the temperature-
driving force) and the other being the agitation speed
in conjunction with the agitation time. In view of this,
various drying scenarios involving agitation speed,
frequency and time span of agitation, and jacket tem-
perature were simulated to understand the effect on
the drying time. It was seen that agitation plays a sig-
nificant role in enhancing the drying rate. It is clear
that continuous mixing results in the fastest drying
at a given jacket temperature. However, it can also
result in significant attrition of product crystals. To
limit generation of fines, intermittent agitation is of-
ten used in practice.
Factors that must be considered to identify an ac-
ceptable jacket temperature include an upper limit to
assure thermal stability of the product and acceptable
crystal form, which in this case demanded a gradual
, predicted internal
JOURNAL OF PHARMACEUTICAL SCIENCES
8 NERE, ALLEN JAMES C. MAREK, AND BORDAWEKAR
Figure 7. Schematic of desirable temperature, cake
smoothing, and mixing profile based on the understanding
sought through model simulations and practical considera-
tions.
increase in the drying temperature. The mixing pe-
riod and intensity that produces acceptable particle
attrition forms a practical basis for the design of the
agitator operating protocol. The schematic of the dry-
ing protocol is depicted in Figure 7. Note that the
jacket temperature was gradually increased to avoid
excessive heating of “wet cake” (i.e., cake with un-
bound solvent) as it may result in the formation of
a dry coating on the jacketed wall, and hence ham-
per the heat transfer rate. Furthermore, no agitation
was sought in the presence of wet cake to avoid gran-
ulation by mixing, which may result in the forma-
tion of hard lumps upon heating and slow down the
drying process. Table 1 presents the summary of the
direct comparison between the older drying protocol
with the new one for the sake of clarity. It should be
noted that the rpm was reduced to reduce the extent
of particle breakage/attrition as the total mixing was
increased in the new drying protocol.
Next, we discuss the verification of model predic-
tions for the proposed drying protocol. Model simu-
lations were run using various headspace pressures
and agitation profiles to determine an efficient dryer
operational protocol. The chosen protocol was imple-
mented in pilot plant batch III. The actual measured
profiles were used to simulate the drying operation
for model verification. Figure 8a shows the compari-
son between the model predictions and observed dry-
ing performance, demonstrating that the model sim-
Table 1. Direct Comparison of the Older Drying Protocol with the New Protocol
Operational Parameter Older Drying Protocol
Jacket temperature (◦ C) 50
Average headspace pressure (mm Hg) 300
Total mixing time (min) 175
Mixing rpm and frequency 10 rpm for 5 min every hour for first 15 h,
followed by mixing at 10 rpm for 5 min
every 6 h
JOURNAL OF PHARMACEUTICAL SCIENCES
ulations are in close agreement with the measured
solvent content throughout the drying process. The
drying time (time to reach solvent content of 0.1%, w/
w starting from the solvent content of 8.8%, w/w) was
approximately 35 h as compared with the estimated
140 h required in the early pilot plant campaign batch
I. Four additional pilot plant batches of the API were
dried using these new operating conditions. Figure 8b
depicts the time profiles of measured solvent content
for all five batches (batches II through VII from later
campaign) dried using the new protocol. All of these
batches showed similar drying performance, thereby
verifying the model description of the drying process.
Furthermore, it shows consistent dryer performance
in all of the batches indicating process robustness.
The model robustness beyond one set of operating pa-
rameters could not be verified at scale due to obvious
constraints on the use of pilot plant equipment and
expensive API. This would certainly be a useful exer-
cise for the future.
GENERAL GUIDELINES FOR OPERATING
AGITATED FILTER DRYER
On the basis of the model studies, demonstration,
and practical considerations, we propose the follow-
ing guidelines for efficient filter dryer operation. Note
that in practice any particular protocol will need to be
changed, depending on specific issues that may be re-
lated to the desirable physicochemical characteristics
of the product.
(1) Efficient inert gas flow through the entire cake
is important. Intermittent cake smoothing must
be carried out to prevent any preferential inert
gas channeling or bypassing. This will alleviate
any mass transfer limitations to the movement
of solvent vapors. It should be noted that there
is a limit on the maximum nitrogen flow rate
given the desired pressure in the headspace and
the cake properties. Under no circumstances,
the headspace pressure should be allowed to in-
crease beyond a limit that will result in cake
compression, decreased inert gas flow rate, and
hence increased resistance to mass transfer. The
combination of desired inert gas flow rate and
headspace pressure can be achieved by ensuring
New Drying Protocol
70
250
360
No mixing for first 6 h, followed by 30 min of
mixing at 2 rpm every 3 h for first 9 h, followed
by 30 min of mixing at 3 rpm every 2 h
DOI 10.1002/jps
 DRYING PROCESS OPTIMIZATION FOR AN API SOLVATE IN AGITATED FILTER DRYER 9

Figure 8. (a) Predicted versus measured drying profile for pilot plant batch-III. Diamonds,
pilot plant data; , model prediction. (b) Measured time profiles of solvent content (%, w/w)
during the drying of five pilot plant batches (batches-III through VII). Open diamonds, batch-III;
filled squares, batch-IV; filled diamonds, batch-V; crosses, batch-VI; pluses, batch-VII.

sufficient vacuum draw from the bottom of a fil-
ter dryer.
(2) No agitation or mixing should be sought in the
presence of substantially wet cake to avoid gran-
ulation by mixing. The wet granules can form
hard lumps upon heating that can be difficult to
dry, eventually increasing the drying time.
(3) Heat the wet cake gradually after ensuring del-
iquoring to the fullest possible extent. Heating
of “wet cake” (i.e., cake with unbound solvent)
in the proximity of vessel wall may result in for-
mation of dry coat on the jacketed wall, fouling
it and hence reduce the heat transfer rate.
CONCLUSIONS
The use of PXRD and online mass spectrometry fa-
cilitated the understanding of desolvation via form
conversion. A composite drying model incorporating
constituent models for heat transfer and desolvation
kinetics was developed to simulate the drying of the
solvate to anhydrous product. Model simulations were
used to understand the effect of key operating vari-
ables, namely mixing and jacket temperature. The op-
erational protocol was designed to (1) avoid thermal
decomposition of product, (2) alleviate mass trans-
fer limitations by ensuring sufficient nitrogen flow
through the cake, and (3) have balanced intermittent
mixing to promote heat transfer. This resulted in a
consistent demonstration of the reduction in drying
time by a factor of 4.
NOMENCLATURE
A heat transfer surface area (m2 )
cp,s specific heat capacity of solid (kJ kg−1 K−1 )
cp, l specific heat capacity of solvent (kJ kg−1 K−1 )
Cs, b ratio of the mass of the bound solvent per
unit mass of dry solids
Dimp impeller diameter (m)
dnc net gas gap between the first layer of solids
and the outer layer of dry solids (m)

DOI 10.1002/jps JOURNAL OF PHARMACEUTICAL SCIENCES

10 NERE, ALLEN JAMES C. MAREK, AND BORDAWEKAR

dp particle diameter (m) tc contact time (s)

dw thickness of the wall (m) Tw vessel wall temperature (K)
dwc clearance between the agitator tip and the Tbed bed interface temperature (K)
vessel wall (m) TD temperature at the end of the first layer of
Fr Froude number solids (K)
hgg heat transfer coefficient for the gas gap Tj jacket temperature (K)
(W m−2 K−1 ) U overall heat transfer coefficient (W m−2 K−1 )
hsb, dry heat transfer coefficient for the dry part of W mass of dry solids (kg)
solid bed (W m−2 K−1 ) X solvent content per unit mass of the dry solids
hdry heat transfer coefficient for the dry solids (kg kg−1 )
(W m−2 K−1 ) x distance from the wall (m)
hsb, wet heat transfer coefficient for the wet part of xF distance of a drying front from the wall (m)
solid bed (W m−2 K−1 ) Ẋ C possible drying rate dictated by the heat trans-
hws heat transfer coefficient for the first layer of fer rate (kg s−1 )
the solids near the vessel wall (W m−2 K−1 )
Greek letters
h ws heat transfer coefficient corresponding to the
distance between the vessel wall and the outer layer ψa fractional vessel wall coverage by solids
of dry solid bed (includes contribution from gas gap) ς dimensionless distance of drying front from
(W m−2 K−1 ) the vessel wall
hwall heat transfer coefficient for the vessel wall ρs density of solids (kg m−3 )
(W m−2 K−1 )
ΔHEvp latent heat of vaporization of solvent (J kg−1 )
kB Boltzmann constant ACKNOWLEDGMENTS
kmetal thermal conductivity of the vessel wall The authors kindly acknowledge the help from
(W m−1 K−1 ) Michael Rog, Yanqun Zhao, and Venkateswarlu
ksb, dry solid bed conductivity (W m−1 K−1 ) Bhamidi for the sampling and analysis and the API
kD rate constant of desolvation (s−1 ) project team for their support.
kg nitrogen gas conductivity (W m−1 K−1 )
kp single solid particle conductivity (W m−1 K−1 )
lmf mean free path length of nitrogen gas REFERENCES
molecule (m)
1. Tsotsas E, Schlünder EU. 1987. Vacuum contact drying of me-
lR surface roughness of the vessel wall (m) chanically agitated beds: The influence of hygroscopic behaviour
N speed of the impeller (s−1 ) on drying rate curve. Chem Eng Process 21:199–208.
Nmix mixing number 2. Hoekstra L, Vonk P, Hulshof LA. 2006. Modeling the scale-up
P headspace pressure (N m−2 ) of contact drying processes. Org Process Res Dev 10:409–416.
Q total heat transfer rate from the jacket to the 3. Schlünder EU, Mollekopf M. 1984. Vacuum contact drying of
free flowing mechanically agitated particulate material. Chem
solids (Eq. 2) (W m−2 ) Eng Process 18:93–111.
Qin heat transfer rate to the wet part of the bed 4. Carslaw HS, Jaeger JC. 1993. Conduction of heat in solids. 2nd
(Eq. 13) (W m−2 ) ed. New York: Clarendon Press.
RD rate of desolvation (kg s−1 ) 5. Scale-up Systems Limited, DynoChem 2008 Model Library. Ac-
tmix mixing time (s) cessed in October 2008, at: http://dcresources.scale-up.com.

JOURNAL OF PHARMACEUTICAL SCIENCES DOI 10.1002/jps