Pharmaceutical Development and Technology, 1:17–32, 2005
Copyright D 2005 Taylor & Francis Inc.
ISSN: 1083-7450 print / 1097-9867 online
DOI: 10.1081/PDT-200035869
The Nonsteady State Modeling of Freeze Drying: In-Process Product
Temperature and Moisture Content Mapping and Pharmaceutical
Product Quality Applications
Pharmaceutical Development and Technology Downloaded from informahealthcare.com by McGill University on 11/30/12
M. J. Pikal, S. Cardon, Chandan Bhugra, F. Jameel, and S. Rambhatla
School of Pharmacy, University of Connecticut, Storrs, Connecticut, USA
W. J. Mascarenhas and H. U. Akay
Technalysis Inc., Indianapolis, Indiana, USA
Introduction. Theoretical models of the freeze-drying
process are potentially useful to guide the design of a freeze-
drying process as well as to obtain information not readily
accessible by direct experimentation, such as moisture distribution
and glass transition temperature, Tg, within a vial during
For personal use only.
processing. Previous models were either restricted to the steady
state and/or to one-dimensional problems. While such models are
useful, the restrictions seriously limit applications of the theory.
An earlier work from these laboratories presented a nonsteady
state, two-dimensional model (which becomes a three-dimen-
sional model with an axis of symmetry) of sublimation and
desorption that is quite versatile and allows the user to investigate a
wide variety of heat and mass transfer problems in both primary
and secondary drying. The earlier treatment focused on the
mathematical details of the finite element formulation of the
problem and on validation of the calculations. The objective of
the current study is to provide the physical rational for the choice
of boundary conditions, to validate the model by comparison of
calculated results with experimental data, and to discuss several
representative pharmaceutical applications. To validate the model
and evaluate its utility in studying distribution of moisture and
glass transition temperature in a representative product, calcu-
lations for a sucrose-based formulation were performed, and
selected results were compared with experimental data. Theo-
retical Model. The model is based on a set of coupled differential
equations resulting from constraints imposed by conservation of
energy and mass, where numerical results are obtained using finite
element analysis. Use of the model proceeds via a ‘‘modular
TM
software package’’ supported by Technalysis Inc. (Passage /
Freeze Drying). This package allows the user to define the problem
by inputing shelf temperature, chamber pressure, container
Received 14 January 2004, Accepted 27 March 2004.
Address correspondence to M. J. Pikal, School of Pharmacy,
University of Connecticut, Storrs, CT 06269, USA. Fax: (860)
486-4998; E-mail: pikal@uconnvm.uconn.edu
17
Order reprints of this article at www.copyright.rightslink.com
properties, product properties, and numerical analysis parameters
required for the finite element analysis. Most input data are either
available in the literature or may be easily estimated. Product
resistance to water vapor flow, mass transfer coefficients
describing secondary drying, and container heat transfer coef-
ficients must normally be measured. Each element (i.e., each small
subsystem of the product) may be assigned different values of
product resistance to accurately describe the nonlinear resistance
behavior often shown by real products. During primary drying, the
chamber pressure and shelf temperature may be varied in steps.
During secondary drying, the change in gas composition from pure
water to mostly inert gas is calculated by the model from the
instantaneous water vapor flux and the input pumping capacity of
the freeze dryer. Results. Comparison of the theoretical results
with the experiment data for a 3% sucrose formulation is generally
satisfactory. Primary drying times agree within two hours, and
the product temperature vs. time curves in primary drying agree
within about ± 1°C. The residual moisture vs. time curve is
predicted by the theory within the likely experimental error, and
the lack of large variation in moisture within the vial (i.e., top vs.
side vs. bottom) is also correctly predicted by theory. The
theoretical calculations also provide the time variation of ‘‘Tg – T’’
during both primary and secondary drying, where T is product
temperature and Tg is the glass transition temperature of the
product phase. The calculations demonstrate that with a secondary
drying protocol using a rapid ramp of shelf temperature, the
product temperature does rise above Tg during early secondary
drying, perhaps being a factor in the phenomenon known as ‘‘cake
shrinkage.’’ Conclusion. The theoretical results of in-process
product temperature, primary drying time, and moisture content
mapping and history are consistent with the experimental results,
suggesting the theoretical model should be useful in process
development and ‘‘trouble-shooting’’ applications.
Keywords non-steady state modeling of lyophilization,
temperature and moisture distribution in freeze
drying, lyophilization process, freeze drying, heat
and mass transfer in freeze drying
 18 M. J. Pikal et al.
INTRODUCTION
Freeze drying[1] is a process by which a solution,
normally aqueous, is frozen and the solid solvent (i.e.,
ice) is removed by direct sublimation during ‘‘primary
drying’’ in a vacuum environment. While most of the
solvent (i.e., water) does convert to the crystalline solid,
some solvent does not crystallize and must be removed
by desorption drying during ‘‘secondary drying.’’ While
the operational definition of secondary drying is normally
Pharmaceutical Development and Technology Downloaded from informahealthcare.com by McGill University on 11/30/12
given as the portion of the drying process after all ice has
been removed, it is well known[1] that secondary drying
actually begins in a section of the product once ice is
removed from that local section. At least for pharmaceu-
tical products, primary drying is normally the longest part
of the process, often requiring days for completion.
Because of the long process times and the high cost of
production freeze drying plants, there is considerable
interest in developing processes that minimize process
time and still maintain product quality. Often such
process development is accomplished empirically, and
it is an unfortunate fact that many manufacturing
processes are far from optimal. Furthermore, problems
For personal use only.
often develop, particularly during scale-up, and solutions
need to be found quickly. Again, empiricism is often the
only approach.
Theoretical modeling can provide a better under-
standing of the impact of process and formulation
variations on cycle time and product temperature history
and, thereby, facilitate process development. Likewise,
trouble shooting manufacturing problems can be aided by
theoretical modeling approaches. Of course, experiments
are essential, but experiments whose design and interpre-
tation are guided by theory are often a much more
efficient use of time.
Fundamentally, optimizing and/or trouble shooting
the drying portion of a freeze-drying process is an
exercise in coupled heat and mass transfer. For every
gram of ice sublimed, about 2700 joules of heat must
be supplied and roughly 1500 liters of water vapor must
flow through the previously dried porous product or cake
and ultimately be converted to ice at the condenser.
For most of the primary drying stage, the process is
‘‘pseudo steady state,’’ meaning that as a good first
approximation, the heat input from the shelves to the
product exactly balances the heat removed by sublima-
tion (i.e., the heat needed to change the temperature of
the product is negligible). Indeed, a simple steady-state
model can quantitatively describe primary drying as long
as accurate experimental or estimated heat and mass
transfer coefficients are input into the model.[2] While
the simple steady-state model of sublimation drying[2]
does accurately predict drying times and product
temperature profiles for primary drying, and therefore,
is of great use in the rational design of a freeze-drying
process as well as trouble-shooting-applications, such a
model does not address residual moisture in the dry
layer nor can one obtain information for the nonsteady
state parts of primary drying, such as the time im-
mediately following a change in shelf temperature.
Nonsteady-state models of sublimation and desorption
drying are needed for a complete description of the
drying process. Such models are based on a set of
coupled differential equations and are therefore, much
more complex.[3]
We have previously[4] described the mathematical
details of an extension of the one-dimensional Liapis and
Millman nonsteady-state sublimation:desorption mod-
el.[3] This extension is a finite element analysis of
drying and is a two-dimensional model that can address
spatial distribution of temperature and residual moisture
in a more realistic fashion. In addition, the extension
differs from the Liapis:Millman model in the detail of
several of the boundary conditions, offering a more
physical set of boundary conditions. The earlier treat-
ment focused on the mathematical details of the finite
element formulation of the problem and on validation of
the calculation by comparison of one-dimensional results
on skim milk with the Liapis:Millman study. The
objective of the current study is to provide the physical
rational for the choice of boundary conditions, to validate
the model by comparison of calculated results with
experimental data, and to discuss several representative
pharmaceutical applications.
MATERIALS AND METHODS
Materials
Crystalline Sucrose was obtained from Sigma
Aldrich Co. The vials used were 5 mL and 10 mL
tubing vials with a 20mm neck size designed for freeze
drying (Wheaton Company). For the 10 mL vials, the
inner diameter was 22 mm, and the outer diameter was
24 mm. The 5 mL vials had an inner diameter of
19.2 mm and an outer diameter of 21.6 mm. Stoppers
were 20 mm finish gray butyl stoppers (West Pharma-
ceuticals) designed for lyophilization.
Freeze Drying
Two different freeze-drying operations were per-
formed. In the first experiment (Process 1) 10 mL vials
were filled with 1.55 mL of 5% w/v sucrose solution and
 Nonsteady State Modeling of Freeze Drying
loaded onto the shelf of a laboratory scale Durastop
freeze dryer (Kinetics Thermal Systems). The freeze-
drying procedure involved a normal freezing cycle where
the shelf temperature was first lowered to 5°C at a ramp
rate of 2.5°C/min and held for 15 min, then lowered to

5°C and held for another 15 mins, and then lowered to
– 40°C at a ramp rate of 1°C/min and held for 60 min.
During primary drying, the shelf temperature was raised
to – 20°C and the chamber pressure was set to 100 mTorr
and held until the end of primary drying (nearly 13
Pharmaceutical Development and Technology Downloaded from informahealthcare.com by McGill University on 11/30/12
hours). During secondary drying, the shelf temperature
was raised to 33°C at a ramp rate of 0.28°C/min and held
until the end of the run. In the second operation (Process
2), 5 mL vials were used, and 2 mL of 5% w/v sucrose
solution was filled into each vial. While the freezing
process was similar to the previous run, a lower shelf
temperature of –30°C was maintained during primary
drying. The ramp rate towards secondary drying was
increased to 0.5° C/min.
Dry Layer Resistance Measurement
For personal use only.
Dry layer resistance measurements were made using
the manometric temperature measurement (MTM) proce-
dure.[5,6] The procedure consists of quickly closing the
valve connecting the drying chamber with the condenser
chamber, monitoring the pressure vs. time data generated,
and fitting the pressure rise data to the theoretical
equations describing the pressure rise. This procedure
gives the dry layer resistance, essentially from how fast
the pressure increases initially, and also gives the
temperature of the ice at the sublimation interface.
Vial Heat Transfer Coefficient Measurement
While vial heat transfer coefficients are most
accurately determined with special experiments,[7] data
with acceptable accuracy can also be obtained from
routine freeze-drying experiments; and it was these latter
procedures we used in this research. We also note that,
since in the context of both the experiments reported here
and in the theory, the ‘‘vial heat transfer coefficient’’ is
based upon the shelf temperature ‘‘set point,’’ rather than
on shelf-surface temperature. Thus, the vial heat transfer
coefficients discussed and reported here really reflect
resistance to heat transfer from shelf surface to vial
bottom plus resistance to heat transfer from heat transfer
fluid to shelf surface. That is, in the context of the present
work, the vial heat transfer coefficients are really vial/
shelf heat transfer coefficients. However, for the labora-
19
tory freeze dryer used in these studies, the ‘‘vial/shelf’’
heat transfer coefficients should be only on the order of
10% less than the vial heat transfer coefficient.[2,7] For the
10 mL vials, the heat transfer coefficient was determined
from the process data for the freeze drying runs with
sucrose. Knowing the fill volume and the solution
concentration, one may calculate the total amount of
water removed during freeze drying, which then allows
the calculation of the total heat required to sublime this
amount of water, DQ, from the known heat of sublima-
tion. The duration of primary drying, Dt, is estimated by
the product thermocouple response. Primary drying is
taken as the total time between the decrease of chamber
pressure that marks the start of primary drying and the
time when all product thermocouples in vials in the
interior of the vial array have reached (within 1°C of) the
shelf temperature. Mass loss from the start of pump down
until the pressure approaches the control pressure (100
mTorr) is negligible and is ignored here. Thus, the vial
heat transfer coefficient is calculated from the average
sublimation rate, DQ/Dt, the time average temperature
difference between shelf and product, h Ts –Tpi, and the
area of the vial, Av, as evaluated from the vial outer
diameter, from the expression,
DQ=Dt
Kv ¼
Av hTs
Tp i
This procedure yields data that is of useful accuracy
even though the heterogeneity in drying behavior causes
some ambiguity in defining the end of primary drying.
The heat transfer coefficient for the 5 mL vials was
determined using a procedure similar to the one
described previously for the 10 mL vials, except that
in these experiments, the freeze-drying runs were
stopped well short of the end of primary drying, and
the mass of ice sublimed was calculated as the
difference between the initial and final masses of a
selected group of vials. This procedure was carried out
during freeze drying of several products at several
chamber pressures spanning 100 mTorr, and the value
for 100 mTorr was interpolated from smoothed raw data.
For the 10mL vials used, the value of vial heat transfer
coefficient was found to be: 104Kv = 3.0± 0.1 cal s
1
cm
2 K
1, where the uncertainly represents the standard
error for three measurements. For the 5 mL vials, a
value of 3.7± 0.1 was found for 104Kv. These values are
lower than were previously found for tubing vials.[7]
While some of this difference is due to the fact that the
vial heat transfer coefficient data reported here also
includes thermal resistance in the shelf, the values
obtained do seem low, particularly for the 10 mL vials.
 20 M. J. Pikal et al.
We have no explanation except to note the poor vial
bottom:shelf contact with these vials.
Residual Moisture
To obtain the kinetics of secondary drying, samples
were removed from the freeze-dryer shelf at several times
during secondary drying using a sample extractor. The
sealed samples were then assayed for water content using
Pharmaceutical Development and Technology Downloaded from informahealthcare.com by McGill University on 11/30/12
a coulometric Karl Fischer Titrator (Orion, AF7LC).
Samples were dissolved in 1 mL of ‘‘dry’’ methanol and
about 0.5 mL of solution was injected into the titrator.
Knowing the weight of the sample and the amount of
water in dry methanol, the water content in the sample
could be determined.
Water Sorption Isotherms
Roughly 10 mg of sample was loaded onto a quartz
pan enclosed in a high vacuum electronic microbalance
apparatus (Sartorius). The apparatus consists of a
For personal use only.
microbalance and thermostats for regulating the temper-
ature of the sample (T1) and of water (T2) contained in a
tube connected by a side arm to the sample tube.[8] A
stepwise variation in the temperature of the T2 bath
enabled a stepwise change in solvent partial pressure. For
each step change in partial pressure, the equilibrium water
content of water was determined from the sample mass
recorded under equilibrium conditions (i.e., when the
sample mass remained invariant with time over several
hours). Gravimetric data at equilibrium for each particular
temperature and relative humidity were used to evaluate
the desorption isotherms. The reader is referred to the
literature[8] for additional experimental details. Data were
obtained over a range of relative humidities at 25°C (0 to
21% relative humidity), 0°C (0 to 41% relative humidity),
and
13.4°C (0 to 60% relative humidity). In addition, a
data point corresponding to equilibrium between ice at
– 35°C and the sucrose freeze concentrate was evaluated
based upon the following considerations. We find the
value of Tg’ in frozen aqueous sucrose is
35°C. At

35°C, the activity of water is 0.70 (i.e., 70% relative
humidity), and based upon our analysis of literature glass
transition data for sucrose-water systems (to be described
later), the glass transition is – 35°C at a concentration of
water in sucrose of 17.5%, w/w. Thus, to be able to
provide a self-consistent description of the water
desorption characteristics and glass transition character-
istics in the region near Tg’, we use data points of –35°C,
70% relative humidity, and 17.5% water in the construc-
tion of the water desorption isotherm.
RESULTS AND DISCUSSION
The Theoretical Model
The Mass and Heat Transfer Equations
Here, we give only the general forms of the heat
and mass transfer equations with an ‘‘intuitive’’
interpretation. The notation used is defined in the
‘‘nomenclature’’ section. The reader is referred to the
literature[4] for additional mathematical detail. In a finite
element model, the physical system is divided up into a
number of subsystems or ‘‘elements,’’ most being
roughly rectangular in geometry, and the heat and mass
transfer equations are solved for each element. In our
case, each element can, in principle, have different
physical properties. The assembly of elements define a
‘‘grid,’’ which in our case is a two-dimensional grid
comprising on the order of 100 elements, which with an
axis of symmetry can represent a three-dimensional
cylindrical container such as a vial. There is a boundary
representing the sublimation interface that separates
‘‘dry region’’ elements from the ‘‘frozen region’’
elements. As drying proceeds, the grid of elements is
regenerated periodically to form more dry elements at
the expense of frozen elements.
A schematic of the freeze-drying model, but without
defining the grid, is given in Figure 1. Heat flows into the
bottom of the sample from the shelf by radiation heat
transfer, conduction at the points of contact between the
vial and the shelf, and by gas phase heat transfer that is
sometimes denoted as ‘‘convection’’ but which, in reality,
is conduction.[7] We also allow for heat transfer to the top
of the product and the sides of the container, such as
would occur by radiation. In general, the model allows for
both water flux, Nw, and inert gas flux, Nin, through the
dry layer, but we find that in all freeze-drying applications
of pharmaceutical interest, one may ignore the inert gas
flux without introducing significant error. The net flux of
water vapor is, of course, from the sublimation interface,
shown here with slight curvature as found by calculation
and experiment.[7] The mass transfer equations in the dry
layer are given by Eqs. 1 – 4. Eq. 1 describes water flow in
the dry layer where the first term on the left hand side of
the equation represents the change in water in the gas
phase, the second term gives the change in water in the
solid phase, and the term on the right hand side is the flux
of water out of the system. This equation essentially
expresses the conservation of mass for water.
@Cw;g @Cw;s
e þ rI ¼
r  Nw ½1
@t @t
 Nonsteady State Modeling of Freeze Drying 21
Pharmaceutical Development and Technology Downloaded from informahealthcare.com by McGill University on 11/30/12

Figure 1. Schematic of the freeze-drying model showing the sublimation interface and the modes of heat transfer.

The rate of change in water content of the solute phase is where Cw,g is the concentrations of water in the gas phase,
assumed to be proportional to the difference between the P is the total pressure, and the constants, ki (i = 1, 2, 3, 4),
water content and the water content that would represent are mass transfer coefficients defined earlier in the
For personal use only.

equilibrium with the surrounding water activity, aw, at nomenclature section. The terms on the right hand side
temperature T, denoted C*(aw,T), of Eq. 4 represent the contributions of both diffusion and
bulk fluid flow, the latter being mostly Knudsen flow (i.e.,
@Cw; s free molecular flow) in the usual pharmaceutical freeze-
¼
kg ðCw; s
C *ðaw ; TÞÞ ½2
@t drying application. Since the vapor composition in the
cake and in the drying chamber during any practical
where kg is a ‘‘rate constant’’ assumed to exhibit primary drying situation is nearly 100% water [7] one may
Arrhenius temperature dependence. C* must be deter- ignore diffusion in the dry layer during primary drying.
mined by experiment as a function of relative humidity Further, since the rate of secondary drying is normally
and temperature. C* as a function of water activity (or determined by rate processes that occur in the solid state
relative humidity) at selected temperatures is nothing other (i.e., solid-state diffusion) or at the solid:vapor boundary
than the water sorption isotherm as a function of (i.e., evaporation),[10] and since the evolution of the vapor
temperature. C* is normally described by either the composition above the dry layer is set by the flux
GAB Equation,[9] or a generalized combination of the dependent boundary condition (to be discussed later), one
Fruendlich isotherm and the Langmuir isotherm which we could ignore diffusion even in secondary drying without
write in the form, introducing serious error. Using the experimental obser-
  vation that the composition of vapor is essentially all


n A  Pw
Ef 1 Pw L Pw0 water, Eq. 4 reduces to the simplified expression,
C * ¼ Af 0  exp  0
þ  
T Pw 1 þ BL  PPw0
w ½3



 Mw C01 Pw Mw
Ef 2 Ef 3 Nw ¼
 Kw 1 þ rPw 
 Kw rPw
AL ¼ AL0  exp ; BL ¼ BL0  exp RT Kw mmx RT
T T
½5
where Af0, Efi (i = 0, 2, 3), and n, AL0, and BL0 are
constants. Pw is the partial pressure of water, and P0w is the where the term proportional to C01, which is a viscous
vapor pressure of pure water at the sample temperature. flow term, is small compared to unity and may be
The water flux, Nw, is given by neglected. Thus, the only mass transfer coefficient that is
needed to address ‘‘normal’’ freeze-drying problems is
Nw ¼
k1 rCw; g
k2 Cw; g rP ½4 the Knudsen flow mass transfer coefficient, Kw.
 22
Heat transfer in the dry layer is described by,
@T @Cw;s
rI CpI ¼ kI r2 T þ rI DHv
Cp;g rðNt TÞ ½6
@t @T
where the term on the left is the heat used to increase the
local temperature, the first term on the right is the net heat
flow into the local dry layer region, the second term is the
heat needed to desorb water from the solute, and the last
term on the right hand side is the heat carried away from
Pharmaceutical Development and Technology Downloaded from informahealthcare.com by McGill University on 11/30/12
the region by gas flow out. Eq. 6 is essentially the
conservation of energy equation for a local dry layer
region. Heat transfer in the frozen layer is described by
@T
rII CPII ¼ kII r2 T
@t
where the term on the left is the heat used to increase the
local temperature, and the term on the right is the net heat
flow through the frozen layer region.
At the sublimation interface, conservation of energy
requires



@T @T
For personal use only.

kI þ kII ¼
Ntn Cp;g T
Nwn DHs ½8 tact’’ contribution. Experimental values of Kv are
@n I @n II
where the first term on the left is the heat flow to the
interface from the dry region, and the second term is the
heat flow to the interface from the frozen region. The first
term on the right hand side is the heat removed by gas
flow from the interface, and the second term is the heat
removed by sublimation of ice, which is normally the
largest term on the right.
Heat Transfer Boundary Conditions
The heat transfer boundary conditions are given by,
q1 ¼ h1 ðT
Tup Þ top surface
q2 ¼ h2 ðT
Ts Þ bottom surface
q3 ¼ h3 ðT
Tw Þ side surface
where the units of the heat transfer coefficients, hi, are
power/area K (i.e., Kw/m2K) and qi are heat fluxes in
power/area (i.e., KW/m2). In the low pressure environ-
ment characteristic of freeze drying, heat transfer
coefficients contain contributions from three ‘‘mecha-
nisms’’: 1) conduction through the solid state at the points
of contact between the vial and the shelf, 2) radiation heat
transfer from dryer surfaces (i.e., shelves above the vials
and under the vials and the dryer walls), and 3) heat
transfer via collisions of gas molecules with the hot
surface (i.e., shelf) and the cold surface (i.e., bottom of
vial).[7] The latter mechanism which is essentially heat
M. J. Pikal et al.
transfer by gas phase conduction, increases as the
chamber pressure increases, and is normally the largest
single mode of heat transfer.[7] Thus, h1 and h3 would
normally be radiation heat transfer while h2 represents
heat transfer from the shelf to the bottom of the vial and
would include all three heat transfer mechanisms.
For maximum accuracy in the modeling, hi are
evaluated from the experimental heat transfer coefficient,
Kv, which is defined by,
dQ
¼ Av Kv ðTs
Tp Þ ½10
dt
where Av is the cross sectional area of the vial based upon
the outer diameter, Ts is the shelf surface temperature, and
½7 Tp is the product temperature in the bottom center of the
vial. The vial heat transfer coefficient may be written in
the form,
Kp P
Kv ¼ KK þ ½11
1 þ KD P
where KK is a constant representing the sum of the
radiation contributions (top and bottom) and the ‘‘con-
normally reported for vials in the interior of an array of
other vials where there is essentially no side radiation (i.e.,
h3 = 0). KD is a constant proportional to the average
separation distance between the bottom of the vial and the
shelf, and KP is a constant that appears to be essentially the
same for all vials.[7] Experimental values of Kv have been
reported in units of ‘‘cal s
1cm
2K
1,’’ where pressure in
Eq. 11 is in Torr.[7] Here, the value of Kp = 0.00332 cal s
1
cm
2 Torr
1, and the top radiation term (i.e., analogous to
h1) is 0.84 10
4 cal s
1 cm
2 for all vials. To obtain the
value of h2 from experimental data we must first calculate
the portion of the experimental Kv that represents only
bottom heat transfer, denoted Kvb, by subtracting the top
radiation term (0.84  10
4) from the experimental Kv.a
Secondly, heat transfer in the freeze-drying model
½9
a
This procedure assumes the ‘‘conventional’’ description[7] of
radiation heat transfer in terms of a radiation heat transfer
coefficient which is given in units of cal/sec, as 1  10
4ei (Ts –
Tp), where ei is the relevant emissivity, and the numerical factor,
10
4, is an approximation for ‘‘4 < T3>,’’ consistent with a mean
temperature of 264 K. In our studies, the mean temperature (i.e.,
of shelf and product) in primary drying is about
30°C and
about + 10°C in early secondary drying, giving 263 K as a mean.
However, this simplification does partition too much of the heat
transfer to top radiation during primary drying and too little in
secondary drying. Overall heat transfer is not impacted. The
only impact on the present study is in the distribution of heat
transfer between top and bottom, with the vial ‘‘bottom heat
transfer coefficient, Kvb’’ being in error by about
7% in
primary drying and + 7% in early secondary drying.
 Nonsteady State Modeling of Freeze Drying 23

Figure 2. Schematic of elements along vial bottom and assignment of heat transfer coefficients, h21.
Pharmaceutical Development and Technology Downloaded from informahealthcare.com by McGill University on 11/30/12

(Figure 1) is based upon heat flow through the cross- and then rearranged to yield
sectional area of the product. Thus, the relationship be-
 
tween Kvb (cals
1 cm
2 K
1 units) and h2 (MKS units) is, h2c
h21 ¼ 1:385  hh2 i  1
0:278  ½15

 hh2 i
Av
hh2 i ¼ 41:876   Kvb ½12
Ap
with < h2> being given in terms of the experimental
where <h2> is the ‘‘average’’ heat transfer coefficient for quantity, Kvb by Eq. 12. The heat transfer coefficient for
heat flow across the vial bottom. The number, 41.876, elements 10 and 11 is calculated from the expression for
comes from the units’ conversion. The ratio of the vial the vial-bottom heat transfer coefficient, Kvb, where the
‘‘outer’’ area to vial inner or ‘‘product’’ area, Av/Ap, is separation distance is set equal to zero (i.e., KD = 0),
about 1.2– 1.3 for most tubing vials. The heat transfer
coefficient for the top boundary, h1, is given in terms of the Kvb ðKD ¼ 0Þ
emissivity for top radiation, ev (= 0.84), h2c ¼ hh2 i ½16
Kvb
For personal use only.


Av
h1 ¼ 41:876   ev  10
4 ½13 If the pressure dependence of the vial heat transfer
Ap coefficient has not been experimentally determined, the
While Eq. 11 could be used to specify the heat transfer value of KD will not be known. Here, we simply use the
coefficients all across the bottom boundary, we generally
choose to specify heat transfer coefficients that recognize
the variation in heat transfer coefficient with variation in
the separation distance between shelf and vial bottom
according to the schematic shown in Figure 2 for a grid that
has 12 elements in the radial direction (i.e., 12 elements
from the center of the vial to the edge). Generalization to
‘‘n’’ elements in the radial direction is obvious. Here, each
element has ‘‘radial length’’ (1/12) r, where r is the radius
of the product cake. As an approximation to the actual
contour for a vial bottom, we take elements 1 –9 and
element 12 to have a roughly equal separation between the
vial bottom and the shelf, and have the same heat transfer
coefficient, h21. Elements 10 and 11 are taken to be in
contact with the shelf (i.e., heat transfer coefficient
corresponding to KD [Eq. 10] = 0, and have the heat
transfer coefficient, h2c. The total heat flow into the vial,
q2 = pr2 < h2>, is given by the sum of the heat transfer into
the various elements, which may be expressed in the form,
"


2 2

2 9 11 9 constant chamber pressure, the partial pressure of water
pr  hh2 i ¼ p r h21 þ p r

12 12 12
"
2 #
11
þ p r2
r  h21
12
approximation that for tubing vials, h2c/ < h2>  1.5. With
this procedure, we force the overall heat transfer
coefficient to correspond to the experimental value but
yet allow spatial variation in heat transfer in the radial
direction to match physical reality.
The Flux Dependent Pressure Boundary Condition:
Evaluation of P w
The theoretical model allows the drying process to be
divided into a number of steps, with the chamber pressure
and shelf temperature held constant throughout each
individual step. As a boundary condition, it is necessary to
fix the value of the partial pressure of water above the dry
layer, Pw. In conventional freeze drying practice, the
chamber pressure is held constant throughout the process.
However, the composition of the gas in the drying
chamber, and therefore, above the dry cake, changes
during secondary drying from nearly all water vapor to
nearly all inert gas as the product dries.[1] Thus, even at
2 #
r  h2c changes with time during drying. In the theoretical model,
the partial pressure of inerts, Pin, is calculated as a
function of time during secondary drying from a
½14 knowledge of the total water flux coming from the
sample normalized to the total area of the dry cake (i.e.,
 24 M. J. Pikal et al.

the area of the interface between the dry layer and the The molar flux of inerts may be written in the form,
‘‘outside’’) and the pumping characteristics of the freeze
dryer. The following analysis develops the appropriate ðPc
Jw  R0c Þ
Jin ¼ ½24
equations by an analysis of mass transfer through three ½R0c þ R0P ð1 þ k  Jw Þ
external barriers, or resistances.
which may, as a good approximation, be simplifiedb as
The barrier to total mass flow from the drying
chamber (i.e., outside the dry layer) to the condenser Pc
chamber, expressed in molar units, R’c , is given by Jin ffi ½25
R0P
ðPc
PCD Þ Note that the partial pressure of inerts is given by,
R0c ¼
Pharmaceutical Development and Technology Downloaded from informahealthcare.com by McGill University on 11/30/12

½17
ðJw þ Jin Þ
Jin
Pin ¼ Pc  ½26
where PCD is the total pressure in the condenser chamber ðJin þ Jw Þ
and Jw and Jin are the molar fluxes of water and inerts,
respectively, in mol m
2 s
1. The resistance to flow from and the mole fraction of inerts in the drying chamber,
condenser chamber to surface of condenser plates, R’CD, is Xin, is given by,

ðPCD ice Jin

w
Pw Þ Xin ¼ ½27
R0CD ¼ ½18 ðJin þ Jw Þ
Jw
where Pice Calculations proceed by evaluation of Jw from the model
w is the vapor pressure of ice at the surface
temperature of the condenser plates. This pressure is and Jin from Eq. 24 [or Eq. 25]. Thus, Pin may be
normally small compared to all other pressures in the evaluated from Eq. 26, with Pw then being evaluated as
system, and we may set it equal to zero with little error. the difference, Pw = Pc
Pin. The calculations start (in
early primary drying) with Pw = Pc, and thereafter the nth
For personal use only.

The ‘‘condenser resistance,’’ R’CD , is given in terms of the

partial pressure of inerts in the condenser chamber, PCD time step uses the previous calculated value of Pw.
in , by
The validity of Eqs. 25 –27 and this approach for
evaluation of vapor composition in the freeze dryer is
R0CD ¼ k  PCD
in ½19
demonstrated by Figure 3. Here, a double reciprocal plot,
based on Eq. 27, is given for a freeze drying run with 5%
where k is a constant characteristic of the freeze dryer.[2]
sucrose (Process 1). Such a plot should be a straight line
The resistance to pumping the inerts out of the condenser
with an intercept of 1.0, as observed (Figure 3), with the
chamber with the vacuum pump, denoted, R’p . is given by
slope giving Jin, from which the value R0P may be
PCD calculated from Eq. 25. The value found is 1.56 105 Pa s
R0P ¼ in
½20 m2 mol
1. The mass transfer constants are somewhat
Jin
specific for the dryer and product area loaded.b
After some algebra with the defining equations Eqs. 17 –19,
one may develop the working equations, which allow
Evaluation of Mass Transfer Coefficient for Water
calculation of Pin. First, the molar flux of water is related to
Vapor Transport
the mass flux by,
1000 Mass transfer in the dry layer is dominated by
Jw ¼  Nw ½21 Knudsen flow[1,2,12] , so the most critical mass transfer
Mw
coefficient is the Knudsen flow constant, Kw. The
where Mw is the molecular weight of water. The resistance Knudsen flow coefficient may be written in the form [3,4]
of the chamber may be written as a function of pressure in
the form[2,7] ,
Kw ¼ f C1 ðTtop þ Tinterface Þ1=2 ½28
1
R0c ¼ ½22
b1  P
where b1 is a constant and P is the mean pressure which b
Values of the constants for other dryers are as follows:
may be calculated from, R’p = 3.84  10 5 Pa s m 2 mol
1 ; k = 1.368 s m 2 mol
1 ,
( sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi) b1 = 6.59  10
3 mol (Pa)
2 s
1 m
2. The values of R’p and b1
Pc 2Jw are for a normal production freeze dryer.[2] The value for k is
P ¼  1þ 1
½23 taken from experiments done with a small laboratory freeze
2 b1 P2c
dryer.[7]
 Nonsteady State Modeling of Freeze Drying 25

where dm/dt is the sublimation rate, Ap is the cross-

sectional area of the dry layer, P0 is the vapor pressure of
ice, Pv is the pressure above the dry layer, and R ^ is the
p
area normalized product resistance. The usual units for R ^
p
2
1 ^
are cm hr Torr g . We find that Rp increases with an
increase in dry layer thickness, but not always in linear
fashion and the extrapolated value of R ^ at zero product
p
thickness is normally significantly higher than zero.[1] In
fact, the variation of resistance with thickness normally
fits an empirical expression of the form,
Pharmaceutical Development and Technology Downloaded from informahealthcare.com by McGill University on 11/30/12

^ ¼ R þ A1
l
R p 0 ½30
1 þ A2
l
Figure 3. Mole fraction of inerts in secondary drying:
Comparison of values calculated from Eqs. 25 – 27 with where R0, A1, and A2 are constants and l is the thickness
experimental measurements from an electronic moisture sensor of the dry layer. Except for possible units differences,
(Ondyne). Values of Jw are evaluated from residual moisture vs. Kw is essentially the reciprocal of the derivative of R ^
p
time data for bovine somatotropin (BST) for a 1 cm fill depth
with respect to dry layer thickness. Note that Kw is a
(data from Ref. [11]). Freeze drying was with a Virtis 25 SRC-X ^ is linear in
constant throughout the dry layer only if R p
lab scale freeze dryer. Freeze drying conditions are: chamber
pressure = 0.1 Torr, shelf temperature = + 12°C, product temper- dry layer thickness.
ature 
28°C in steady state primary drying. Product descrip-
tion is: 5% solids (BST) in solution with 0.19 g solids per vial
after drying, maximum cake thickness = 1 cm., area of cake
For personal use only.
Calculation of the heterogeneity factor in such a way
as to maintain correspondence with the experimental R ^
p
vs. l data may be done with the following procedure. For
each row, calculate the value of Kw for that row by,c

surface = 3.8 cm2. The symbols are experimental data while the
straight line is a linear fit to the data.
0:0292
Kwi ¼ ½31
ðdRp =dlÞi
where Ttop and Tinterface are the temperatures (K) of the
top of the dry layer and the sublimation interface,
where the numerical value of 0.0292 represents units
respectively, C1 is a constant dependent on the structure ^ and cm for
conversion from the ‘‘usual’’ units used for R p
of the porous system, and f is the ‘‘heterogeneity
dry layer thickness to the mks units used in the theoretical
factor.’’ The heterogeneity factor allows for variation in
model for Kw. We next calculate an ‘‘average’’ value of
pore size and Knudsen flow constant as a function of
the Knudsen permeability, denoted Kwo, from the average
position in the cake. Thus, at an extreme, the model can
value of the derivative of Rp with respect to thickness, l,
accommodate a unique value of Kw for each element ^ /dli
denoted hdR p lmax, by the equation,
through assignment of a unique value for the heteroge-
neity factor, f, for each element. On the other extreme, if 0:0292
the permeability is constant throughout the system, all Kw0 ¼ ½32
^
hdRp =dlil max
elements would have the same value of Kw, and we
would set the heterogeneity factor equal to one. In The heterogeneity factor for row i is then calculated as,
practice, permeability usually does not vary greatly in
the radial direction, but significant variations in per- Kwi
fi ¼ ½33
meability occur as one moves from the top to the bottom Kw0
of the dry layer, so reality dictates we assign different
heterogeneity factors for each row of elements. We now
Evaluation of Desorption Parameters
consider how to set the heterogeneity factor for each row
to reproduce the experimental mass transfer behavior in
Experimental values of C* at various water
the Knudsen permeability format.
activities and temperatures were determined as described
We normally[1] experimentally describe mass transfer
earlier, the obtained experimental data were then fitted
in the dry layer in terms of product resistance, where
resistance is essentially defined by,
c
Dividing the dry layer into sections or ‘‘elements,’’ it is
dm ðP0
Pv Þ straightforward to show that the steady-state mass transfer rate is
¼ Ap ½29 ½PðlþDlÞ
PðlÞ
dt R^ given by, dm
dt ¼ ^ ^ ¼ C  Ap  ½PðlþDlÞ
PðlÞ
Dl  Kwi .
p ½ Rp ðlþDlÞ
Rp ðlÞ
 26 M. J. Pikal et al.

to the generalized sorption equation [Eq. 3], and the

parameters obtained are shown in Table 1. The
isotherms reproduced by these parameters fit the data
available reasonably well, but there were systematic
deviations (Figure 4). The deviations seem to be beyond
anticipated experimental error and probably reflect the
inability of the sorption equation to fit the data, even
though a large number of parameters were used. Note
that this treatment is only intended to reproduce and
extrapolate the experimental observations and yield a
Pharmaceutical Development and Technology Downloaded from informahealthcare.com by McGill University on 11/30/12

self-consistent parameterization where the value of C* at

35°C (i.e., Tg’) is forced to be 0.175 g/g, so that the
composition of the maximally freeze-concentrated solute
will be the equilibrium water content at Tg’. No physical Figure 4. Water desorption isotherms for amorphous sucrose
at selected temperatures: C* as a function of water activity and
meaning of the parameters evaluated is intended. The
temperature. The symbols represent experimental data (open
desorption parameter kg, was calculated using data
circle =
35°C, triangles =
13.4°C, squares = 0°C, and closed
obtained from one of the experiments denoted, Process circles = 25°C. The solid lines represent the corresponding data
1. To estimate the desorption coefficient kg, vials were calculated from the ‘‘best fit’’ parameters of Eq. 3.
removed from the freeze dryer at several time intervals
during the secondary drying stage of the freeze-drying
process. The concentration of unfrozen water C(t) was expected, kg did exhibit an Arrhenius temperature de-
assayed using a Karl-Fischer titrator. For each sampling, pendence (Figure 5).
values of the partial pressure of water vapor, Pw were
For personal use only.

calculated from the dew point measurement and the Finite Element Analysis Solution
vapor pressure of supercooled water, P0w was calculated
from the product temperature and the known relationship The preprocessor module of the finite element
between temperature and vapor pressure. These values analysis software involves the use of input parameters
were then used to calculate C*, i.e., from Eq. 3. At each to create the model for the freeze-drying analysis. A list of
time, the value of C(t), C*, and dC/dt (obtained by input parameters required to run the model is shown in
numerical differentiation) were calculated. Using Eq. 2, Table 2. While some parameters are standard handbook
the desorption parameter, kg, was then calculated. As values and do not change as the application changes,
some others are available by estimation with negligible
compromise on accuracy. For the effective heat capacity,
heat conductivity, and density of the dried material (i.e.,
Table 1
solute), one may use the corresponding property of any
Parameters for the generalized sorption model (Eq. 3) and
similar material, or the Passage default value,d with
parameters for a modified Gordon-Taylor equation describing
glass transition temperature as a function of water content, acceptable accuracy. Since the gas is always mostly water
shown in the table. The modified Gordon-Taylor equation vapor during primary drying, when heat transfer effects
was fit to literature data[13 – 15] over a wide range of are most critical, the standard values for water vapor may
water contents from zero to 25% be used for the heat capacity and heat conductivity of the
gas in the dry layer (i.e., the Passage default valuesb). The
½KTg  Tg2
Cw  ðKTg  Tg2
Tg1 Þ gases are the same in nearly all freeze-drying applications
Tg ¼ þ a1  Cw  ð1
Cw Þ
½Cw  ð1
KTg Þ þ KTg (i.e., water and nitrogen), so for the free gas mutual
þ a2  Cw2  ð1
Cw Þ
diffusivity constant, one may use the default value.b With
essentially no loss in accuracy, the values of heat
Parameter Value Parameter Value capacity, heat conductivity, and density for the frozen
layer may be taken as the weighted average of dry solid
Tg2 348.2 AL0 6.42  10
5 and ice values. Input parameters such as internal mass
Tg1 135 BL0 7.88  106
transfer coefficient for desorption, kg, and Knudsen flow
KTg 0.0920 Af0 4.26  105
a1 480.8 Ef1 1929
a2
1224 n 0.3 d
Ef2, Ef3
5028 The default values are the values from the skim milk problem
studied by Millman, Liapis, and Marchello.[3]
 Nonsteady State Modeling of Freeze Drying 27
Pharmaceutical Development and Technology Downloaded from informahealthcare.com by McGill University on 11/30/12

Figure 5. Arrhenius plot of the desorption constant evaluated from secondary drying data for Process 1. A fit to the Arrhenius
equation results in an activation energy Ea of 8,136 J/mol, and a preexponential factor of 3.34.10
3 s
1.

constant, Kw, may be estimated but are best determined
from experiments on the system of interest, sucrose in this
case, using procedures described earlier. Heat transfer
parameters, hi (i = 1, 2, 3), may be calculated from
For personal use only.
estimated or measured vial heat transfer coefficients, Kv,
as described earlier.
The solver or the processor module performs the
finite element analysis of the model using input files

Table 2
Input parameters and property data for the model

Property Description Value

C0 1 Constant for viscous flow, depends on cake structure 0 (term not used)
C1 Constant for Knudsen flow, depends on cake structure 4.64  10
4
C2 Constant for diffusivity in cake, depends on structure 0.311
CpI Effective heat capacity of dry layer (kJ/Kg.K) 0.2595
CpII Effective heat capacity of frozen layer (kJ/Kg.K) 1.94
Cpg Effective heat capacity of gas in dry layer (kJ/Kg.K) 2.01
C0s, w Initial concentration of sorbed water(kg water/kg solid) 0.2059
k1 Constant for mass transfer C2D0w, inKw/(C2D0w, in+ KmxP)
k2 Constant for mass transfer KwKin/(C2D0w,in+ KmxPT) + C01/mmx
kI Effective thermal conductivity of dry layer 8.826  10
8*P + 2.706  10
5
kII Effective thermal conductivity of frozen layer 0.002763
kg Mass transfer coefficient for desorption (s
1) see Figure 5
Kw Knudsen diffusivity fC1(Ttop+ Tintf)^0.5
Min Molecular weight of inert gas (kg/kg – mole) 28.02
Mw Molecular weight of water (kg/kg – mole) 18.016
R Gas law constant g (J/kg – mole.K) 8314.0
DHs Heat of sublimation of ice (kJ/kg) 2834.6
DHv Heat of vaporization of bound water (kJ/kg) 2499.6
e Void fraction in the dried region 0.95
mmx Viscosity of the binary mixture of water vapor and inert gas term not used
rI Density of dry region (kg/m3) 55.0
rII Density of frozen region (kg/m3) 927.1
R0 Constant term in dry layer resistance equation (Eq. 30) 0.2
A1 Constant in dry layer resistance equation (Eq. 30) 67.9 (Process 1); 14.6 (Process 2)
A2 Constant in dry layer resistance equation (Eq. 30) 26.2 (Process 1); 1.51 (Process 2)
h1 Heat transfer coefficient for top radiation (MKS units) 0.0042
h21 Heat transfer coefficient for shelf-vial bottom (MKS units) 0.00873(Process 1), 0.01223 (Process 2)
h2c Heat transfer coefficient for shelf-vial bottom at ‘‘contact’’ (MKS units) 0.01622(Process 1), 0.02272 (Process 2)
 28 M. J. Pikal et al.

Table 3
Comparison of experimental and theoretical primary drying times and temperatures. Experimental drying times are from dew
point sensor data, and experimental temperatures are mean temperatures from the start of primary drying to the
sharp increase in temperature indicating that thermocouple has lost contact with ice

Average product temperature in

Primary drying time (hours) primary drying, 8C

Freeze drying runs Experiment Passage Experiment Passage

Process 1 13.0 12.7
36.0
36.5

Pharmaceutical Development and Technology Downloaded from informahealthcare.com by McGill University on 11/30/12

Process 2 37.2 39.4
36.2
35.5

created in the preprocessor. The primary freeze-drying
part of the solver solves the heat and mass transfer
equations associated with primary drying while the
secondary drying module solves the corresponding
coupled equations associated with secondary drying.
The output of the finite element analysis calculations
includes time and spatial distribution of residual water
contents, product temperature, vapor pressure and the
difference between the glass transition temperature and
product temperature (Tg
T).
For personal use only.
represents a comparison of experimental and calculated
product temperatures during drying for Process 2. Good
agreement is observed with the difference between
predicted and experimental product temperatures being
less than 1°C in the initial part of primary drying and
throughout secondary drying. Likewise, the mean temper-
atures in primary drying (Table 3) are in good agreement.
For reasons we do not fully understand, the approach
towards the shelf temperature at the end of primary drying
is more gradual for thermocouple-measured product
temperatures as compared to the sharp increase in

Validation of the Model simulated product temperatures at the end of primary

drying. Perhaps some of this difference may be attributed
Product Temperatures and Drying Times . Vali- to the heat capacity of the glass in the vial, which the
dation of the model was carried out by comparison of data calculations ignore, and our approximations for thermal
obtained from an experiment with data obtained by conductivity and heat capacity in the dry layer. There also
simulation using Passage software. A comparison of may be some residual ice in the vial well after the sharp
primary drying times predicted by Passage and those break to higher temperature. In support of this specula-
obtained from experiment for Processes 1 and 2 is shown tion, we note that the end of primary drying as measured
in Table 3. Experimental primary drying times were by dew point response only occurs well after the
determined from the electronic moisture sensor response temperature break.
(i.e., a sharp decrease in dew point indicates the end of
primary drying). Good agreement between experiment
and theory is observed. It should be noted that primary
drying times differ according to position on the shelf.
Experiments conducted in our laboratory indicate that
vials located closer to the freeze dryer door have higher
sublimation rates and lower primary drying times. A
plausible explanation for this effect is atypical radiation
heat transfer that edge vials receive due to their direct
view of a warmer surface. Both experimental and
simulation results reported here refer to typical vials
located in the interior of the vial array (i.e., vials that are
enclosed in a hexagonal arrangement).
Product temperatures recorded by thermocouple
measurements during freeze drying were compared with
product temperatures predicted by Passage (Table 3 and
Figure 6). Theoretical product temperatures were taken as
the average of 1  2 elements, one radial and two Figure 6. Comparison of product temperatures determined
lengthwise, so as to compare with the dimensions of the from experiment (solid triangles) and passage (open circles) for
product thermocouple located inside the vial. Figure 6 Process 2. Solid line represents the shelf temperature.
 Nonsteady State Modeling of Freeze Drying 29

experimentally. Predicted product temperature profiles

during primary drying are shown for three different
locations in Figure 8a. Since the sublimation interface
moves from the top downwards with the top of the
material drying the earliest, there is about a 3° difference
in temperature between the core and the top of the sample.
There is about a 1° variation in temperature in the radial
direction. The difference between the glass transition
temperature and the product temperature (T
Tg) is an
important factor to be considered throughout the drying
Pharmaceutical Development and Technology Downloaded from informahealthcare.com by McGill University on 11/30/12

process, but the glass transition temperature during

secondary drying is difficult to directly measure. Figure
8b show predicted profiles of (T
Tg) during the
secondary drying stage. Three different locations, the
top, the bottom, and the core of the sample were chosen.
Notice from Figure 8b that the product temperature is well
above the glass transition temperature for several hours
Figure 7. Comparison of residual water contents obtained
from experiment (open and closed circles) and theory (repre-
sented by the solid line) for Process 2. Experiment 1 (open
circles) represents data obtained from a freeze-drying experi-
ment without the use of radiation shields and experiment 2
(filled circles) are data obtained using aluminum foil on the
interior surface of the door as a radiation shield. Error bars
For personal use only.

represent standard deviations.

Residual Moisture vs. Time . Experimental resid-

ual water contents were determined from Karl Fischer
analysis of samples removed at different times during the
freeze-drying process. Figure 7 represents a comparison
of experimental and theoretical water contents with time
obtained for Process 2, (i.e., a validation run that was
operated using a different freeze drying protocol than did
Process 1). Good agreement is observed, although it does
appear that agreement would be better if the initial water
content were lower. That is, our choice of 17.5% water
in the freeze concentrate causes the theoretical results to
be high relative to the experiment at early times. How-
ever, the correspondence is close enough so that the
calculated glass transition temperature, whose accuracy
depends upon accuracy of moisture content, should have
useful accuracy.
It is important, also, to note that heterogeneity in
sublimation rates is more pronounced at lower shelf
temperatures during primary drying, and one can observe
that the variation in data was reduced significantly by the
use of aluminum foil as a radiation shield on the door of
the freeze dryer.
Figure 8. a) Spatial distribution of product temperature (T)
Spatial Distribution of Temperature and Mois- during primary drying, and b) (Tg
T) during secondary drying.
ture . The finite element analysis model can be used to Data are shown for three different positions; bottom center (bold
monitor the time and spatial distribution of several solid lines), top (bold dashed lines), and edge-center (thin dashed
important parameters that may be difficult to determine lines). Process conditions were as in Process 2.
 30 M. J. Pikal et al.

during the secondary drying stage. In this particular
process, a ramp rate of 0.5 C/min was used to advance
from primary drying to secondary drying. Thus, it may be
important to use a low ramp rate towards secondary
drying to avoid partial product collapse, or ‘‘cake
shrinkage’’ that can occur if the product temperature
exceeds the Tg.
Another important observation from the spatial
distribution of residual moisture during freeze drying of
sucrose samples was the absence of large differences in
Pharmaceutical Development and Technology Downloaded from informahealthcare.com by McGill University on 11/30/12
ples, which were cut into three sections consisting of the
bottom, top, and side of the sample, show no significant
difference between the sections (Figure 9b). This
observation is in sharp contrast to both the experimental
data obtained earlier for drying bovine somatotropin
(BST)[11] and the Passage simulation results for BST.[4]
Experimental data for BST[11] showed very large moisture
variation with position in the cake. There was evidence
for significantly higher moisture content in the bottom
section, and the sides section typically had the lowest

water content between the bottom, top, and sides (i.e.,
outer section, near the vial wall) of the sample (Figure
9a). The theoretical results show only small differences
between top and bottom early in secondary drying. Karl
Fischer analysis for water content of these sucrose sam-
For personal use only.
moisture content. The lower moisture in the side (or outer)
section was attributed to preferential drying near the side,
likely caused by the cake shrinking away from the vial
wall during primary drying. For BST, Passage simulation
results[4] showed the following order in residual moisture

Figure 9. Spatial distribution of residual water contents during secondary drying. (a.) Calculated by Passage; (b.) Experimental
measurements. Symbols key: sides (vertical bars), bottom (solid bars) and top (horizontal bars). The experimental values were by
Karl Fischer analysis of sections of the cake using previously described procedures.[11] The calculated results refer to elements
located as follows: side (two elements from the edge, middle of cake), bottom (three elements up from bottom near center of vial),
and top (two elements down from top near center of vial). Error bars represent standard errors. The asterisk above the 9.8 hour time
point in (a.) indicates only one measurement available for that time point.
 Nonsteady State Modeling of Freeze Drying 31

content, top <sides <bottom, with the difference between ev vial top emissivity
top and bottom being about a factor of 20. Cake shrinkage f heterogeneity factor
was not accounted for in the Passage model, so it is h heat transfer coefficient (kW/m2 K)
understandable that the theoretical results did not exactly Jw molar flux of water (mol/m2 s)
reproduce the experimental observations with regard to Jin molar flux of inerts (mol/m2 s)
lowest moisture in the sides section. However, it is kI thermal conductivity of dry layer(kW/mK)
interesting and significant that the qualitative experimen- kII thermal conductivity of frozen layer(kW/mK)
tal results of no large spatial variation in moisture for k1 bulk diffusivity constant C2D0w,inKw/(C2D0w,in
sucrose and large spatial variation in moisture for BST are + KmxP)
predicted by the calculations. k2, k 4 self diffusivity constant KwKin/(C2D0w,in+KmxP)+
Pharmaceutical Development and Technology Downloaded from informahealthcare.com by McGill University on 11/30/12

(C01/mmax)
k3 bulk diffusivity constant C2D0w,in Kin/(C2D0w,in+
CONCLUSIONS KmxP)
kg mass transfer coefficient for desorption (s
1)
The theoretical results of in-process product temper- Kv vial heat transfer coefficient (cal/sec cm2 K)
ature, primary drying time, and residual moisture content Kvb vial bottom heat transfer coefficient (cal/s cm2 K)
variation, both spatial and temporal, are fully consistent Kw Knudsen diffusivity for water vapor C1 (RT/
with the experimental results, suggesting the theoretical Mw)0.5
model should be useful in process development and Kin Knudsen diffusivity for interts C1 (RT/Min)0.5
trouble-shooting applications. Of particular use, is the Kmx mean Knudsen diffusivity for binary gas mixture,
feature of spatial and temporal variation in residual (Kmx = ywKin+ yinKw)
moisture during secondary drying. Such data are ex- l thickness of dry layer, cm
tremely difficult to obtain by direct experiment and are M molecular weight (kg/kg mole)
For personal use only.

quite relevant to questions of collapse during secondary Nin molar flux of inerts (kg/m2sec)
drying as well as perhaps relevant to predictions of Nw molar flux of water (kg/m2sec)
protein stability during drying. That is, since the P total pressure
calculations can give the Tg history throughout the vial Pc chamber pressure (Pa)
during the secondary drying process, stability predictions Pcd condenser pressure (Pa)
can, perhaps, be made for those cases where T
Tg is an Pin partial pressure of inert gas (Pa)
important stability variable. P0 vapor pressure of ice (Torr)
Pv pressure above the dry layer (Torr)
Pw partial pressure of water vapor (Pa)
NOMENCLATURE q heat flux (kW/m2)
R gas law constant
A cross-sectional area of the dry layer, cm2 Rp area normalized product resistance (cm2 hr
C* equilibrium concentration of sorbed water (kg Torr/g)
water/kg solid) R’c chamber resistance
Cp heat capacity (kJ/kg K) RCD’ resistance to flow from condenser chamber to
Cw, s concentration of sorbed water (kg water/kg condenser plates
solid) t time (s)
C01 constant dependent only upon structure of po- T temperature (K)
rous medium and giving relative D’Arcy flow Tint temperature of the sublimation interface (K)
permeability Tp temperature of the product (K)
C1 constant dependent only upon structure of po- Ts shelf Temperature (K)
rous medium and giving relative Knudsen flow Ttop temperature of the top of the dry layer (K)
permeability Tup upper plate temperature (K)
C2 constant dependent only upon structure of po- Xin mole fraction of inerts
rous medium and giving the ratio of bulk dif- yw mole fraction of water vapor
fusivity within the porous medium to the free
gas bulk diffusivity (dimensionless) Greek symbols
Dw, in free gas mutual diffusivity in a binary mixture
of water vapor and inert gas DHs heat of sublimation of ice (kJ/kg)
D0w, in Dw, in P DHv heat of vaporization of bound water (kJ/kg)
 32 M. J. Pikal et al.
e void fraction in the dried region
mmx viscosity of the binary mixture of water vapor and
inert gas in the porous dried layer (kg/m.s)
r density (kg/m3)
Subscripts
I dry region
II frozen region
n normal component
Pharmaceutical Development and Technology Downloaded from informahealthcare.com by McGill University on 11/30/12
x x-component
y y-component
ACKNOWLEDGMENTS
This project was funded by a grant from the National
Science Foundation – Center for Pharmaceutical Process-
ing Research (NSF –CPPR).
REFERENCES
For personal use only.
1. Pikal, M.J. Freeze drying. In Encyclopedia of
Pharmaceutical Technology, 6; Swarbick, J., Boylan,
J., Eds.; Marcel Dekker, 2001. See also, the Dekker
web site, www.Dekker.com. The DOI number is,
10.1081/E-EPT-100001712, pages 1299 –1326.
2. Pikal, M.J. Use of laboratory data in freeze drying
process design: heat and mass transfer coefficients
and the computer simulation of freeze-drying. J.
Parenter. Drug Assoc. 1985, 39 (No. 3), 115 – 138.
3. Millman, M.J.; Liapis, A.I.; Marchello, J.M. An
analysis of the lyophilization process using a
sorption-sublimation model and various operational
policies. AICHE J. 1985, 3, 1594– 1604.
4. Mascarenhas, W.J.; Akay, H.U.; Pikal, M.J. A
computational model for finite element analysis of
the freeze-drying process. Comput. Methods Appl.
Mech. Eng. 1997, 148, 105– 124.
5. Milton, N.; Pikal, M.J.; Roy, M.L.; Nail, S.L.
Evaluation of manometric temperature measurement
as a method of monitoring product temperature
during lyophilization. PDA J. Pharm. Sci. Technol.
1997, 5, 7 –16.
6. Tang, X.; Pikal, M.J. Evaluation of manometric
temperature measurement (MTM) in freeze drying.
Part II: measurement of dry layer resistance, (under
review).
7. Pikal, M.J.; Roy, M.L.; Shah, S. Mass and heat
transfer in vial freeze-drying of pharmaceuticals:
role of the vial. J. Pharm. Sci. 1984, 73, 1224 –
1237.
8. Pikal, M.J.; Lang, J.E.; Shah, S. Desolvation kinetics
of cefamandole sodium methanolate: the effect of
water vapor. Int. J. Pharm. 1983, 17, 237– 262.
9. Zografi, G.; Kontny, M.J. The interaction of water
with cellulose-and starch derived pharmaceutical
excipients. Pharm. Res. 1986, 187 –194.
10. Pikal, M.J.; Shah, S.; Roy, M.L.; Putman, R. The
secondary drying stage of freeze drying: drying
kinetics as a function of temperature and chamber
pressure. Int. J. Pharm. 1990, 60, 203– 217.
11. Pikal, M.J.; Shah, S. Intravial distribution of
moisture during the secondary drying stage of
freeze drying. PDA J. Pharm. Sci. Technol. 1997,
51, 17 –24.
12. Pikal, M.J.; Shah, S.; Senior, D.; Lang, J.E. Physical
chemistry of freeze-drying: measurement of subli-
mation rates for frozen aqueous solutions by a micro
balance technique. J. Pharm. Sci. 1983, 72, 635 –
650.
13. Saleki-Gerhardt, A.; Zografi, G. Non-isothermal and
isothermal crystallization of sucrose from the
amorphous state. Pharm. Res. 1994, 11, 1166 –1173.
14. Hatley, R.H.M.; van den Berg, C.; Franks, F. The
unfrozen water content of maximally freeze con-
centrated carbohydrate solutions: validity of the
methods used for its determinaton. Cryo-Lett. 1991,
12, 113 –124.
15. Hatley, R.H.M.; Mant, A. Determination of the
unfrozen water content of maximally freeze-con-
centrated carbohydrate solutions. Int. J. Biol.
Macromol. 1993, 15, 227 –232.