Powder Technology 386 (2021) 505–518

Contents lists available at ScienceDirect

Powder Technology

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

Optimization of Wurster fluid bed coating: Mathematical model

validated against pharmaceutical production data
Jiří Kolář a,b, Pavel Kovačík b, Tereza Choděrová a, Zdeněk Grof a, František Štěpánek a,⁎
a
UCT Prague, Technická 5, 166 28 Praha 6, Czechia
b
Zentiva, k.s., U Kabelovny 130, 102 37 Praha 10, Czechia

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

Article history: Wurster fluid bed coating is an effective way of manufacturing pharmaceutical pellets requiring specific charac-
Received 29 November 2020 teristics such as high drug loading or enteric protection. The optimization of coating process can be demanding
Received in revised form 20 March 2021 because it necessitates many time-consuming experiments. To reduce the need for experiments and prevent un-
Accepted 24 March 2021
desired agglomeration, a multi-scale model has been developed. On the device-scale, the model predicts the tem-
Available online 29 March 2021
perature, moisture content, and average coating thickness. On the micro-scale, the model estimates the
Keywords:
evaporation time of a single sessile droplet. By coupling models at both length-scales, it is possible to evaluate
Wurster coating droplet evaporation time as a function of process parameters and use it as a tool for process transfer between dif-
Layering ferent production sites and equipment. The model was validated against existing production data. The predicted
Process optimization outlet air temperature was in good agreement, and most importantly, the predicted values of the evaporation
Thermodynamic model time separated undesired agglomeration and successful coating regime.
Heat transfer © 2021 Elsevier B.V. All rights reserved.
Drying

1. Introduction temperature homogeneity. In contrast to conventional fluidized bed

coating, a coating solution is sprayed from a nozzle placed bottom-up
The coating of pharmaceutical dosage forms is an important step in in the center of a cylindrical tube (draft tube). Particles are driven
drug production. It has various purposes, including the application of ac- through the tube and periodically move through the spraying zone
tive pharmaceutical ingredients (APIs), the masking of unpleasant where they collide with small droplets of the coating solution. The circu-
tastes and the formation of a protective barrier against the environment lation of the particles in this way ensures good mixing and provides bet-
[1,2]. The coating also opens up new possibilities in drug formulation ter control of particle movement and the resultant coating properties
design; for instance, layers can enable prolonged drug release or in- [7].
creased drug bioavailability [3,4]. However, pharmaceutical coatings In comparison to other coating methods, the main disadvantage of
must meet strict quality requirements concerning purity, uniformity, Wurster is the tedious setup and process optimization [10]. Currently,
and dosage per tablet/pellet [5,6]. Consequently, setting up a coating manufacturers choose this method mainly for special applications like
process can be both expensive and time-consuming, especially as it is formulation of sustained or enteric release. Coloring, odor, or taste
usually performed on a trial and error basis [7]. For instance, a full facto- masking, where uniform film thickness and coating porosity is less im-
rial design analyzing the influence of 4 variables with only two-level portant, can be done in conventional top-spray fluidized bed [2].
factor consists of 16 trials [8]. Moreover, a higher-order interactions be- Wurster is also less suitable for processing large tablets, where rotating
tween process parameters can complicate the interpretation of results pans are predominately used [10].
[9]. In the case of Wurster, over 10 variables should be considered According to Heinrich et al. [11] and Toschkoff and Khinast [12],
(see Table 1). Thus, mathematical models that can help with the estima- main difficulties of the Wurster coating process are:
tion of process parameters and, thereby, reduce the number of experi-
ments needed are highly valuable. • Disruption of fluidization behavior and formation of stagnant zones
Wurster fluid bed coating is one of the most versatile coating • Low deposition yield of the solute
methods [1,10]. It offers fast heat and mass transfer, as well as • Inter−/intra−particle coating variability
• Particle agglomeration

Airflow characteristics and particulate flow in Wurster device were

⁎ Corresponding author. investigated by many researchers both experimentally and theoreti-
E-mail address: [email protected] (F. Štěpánek). cally. Experimental approaches are usually limited to in-house designed

https://doi.org/10.1016/j.powtec.2021.03.059
0032-5910/© 2021 Elsevier B.V. All rights reserved.
J. Kolář, P. Kovačík, T. Choděrová et al. Powder Technology 386 (2021) 505–518

homogeneous coating distribution [37]. However, the coating distribu-

Table 1 tion and particle wetness can only be controlled indirectly by setting op-
List of process parameters to consider during Wurster coating.
erating parameters, such as temperature, spraying rate, draft tube
Inlet air temperature position, volumetric airflow rate and humidity.
Inlet air humidity In terms of controlling the influence of operating parameters on par-
Fluidization air flow rate ticle wetness, thermodynamic models are required [12]. Dewettinck
Air distributor
Position of draft tube
et al. [38] proposed a steady-state thermodynamic model. Later, a dy-
Batch size namic model additionally extended with population balance was pro-
Spray rate posed by Burgschweiger and Tsotsas [39], Ronsse et al. [40]. Among
Position and type of atomization nozzle other examples steady-state or dynamic models are Larsen et al. [41],
Atomization pressure
Gouin [42], Page et al. [43,44], am Ende et al. [45].
Inlet temperature of coating medium
Composition and concentration of coating medium Quite advanced is a three-dimensional model proposed by Heinrich
et al. [46], which can describe the spatial distribution of solvent in the
device. On the other hand, it is also computationally expensive similarly
to CFD models.
device [13–15], usage of special probes [16] or to indirect measure- Agglomeration can be viewed in terms of probability as a product of
ments, for instance, analysis of pressure drop fluctuations [17]. How- geometrical and the physical collision success factors [28,47]. Geometri-
ever, these results are very important, because they enable validation cal factor is based on surface coverage and physical factor is based on co-
of theoretical models. alescence condition derived from Stokes number analysis [48]. Direct
Regarding the modeling, discrete element method (DEM) coupled usage of this approach in Wurster process can be challenging. For in-
with computational fluid dynamics (CFD) can accurately predict the stance, particle dynamics must be known and investigated either exper-
particulate flow behavior [14,18–22]. The only problem is computa- imentally or by extensive simulations.
tional requirements, which scale rapidly with an increasing number of In case of top spray coating, Rieck et al. [49] presented a model for
particles. Numerical requirements can be partially mitigated by as- agglomeration prediction based on probability of successful wet colli-
sumption of repetitive patterns in particles circulation, so called recur- sions between particles. Model seems to accurately predict majority of
rence CFD-DEM was successfully used by several authors [23–25]. analyzed experiments if layering or agglomeration will be the dominant
On the other hand, multiphase CFD simulations are not limited by enlarging mechanism. However, it should be noted that experiments
the number of particles and several authors reported a good agreement were done with different coating medium, agglomeration experiments
with experimental data [26–28]. However, Eulerian multiphase models mostly with hypromellose, while coating experiments with table salt
cannot describe inter-particle interactions correctly and simulation or sodium benzoate solutions. In contrast to our work, where we com-
setup must be carefully validated [11]. pare different process conditions for the same composition of the coat-
The inter-particle coating variability can be modeled either as a se- ing medium.
ries of mass additions to the particle considering the mass and number Accuracy of a thermodynamic model depends on obtaining correct
of additions as randomly distributed variables [29,30], or as a multi- heat and mass transfer coefficients, which are commonly calculated
compartment model using population balances [31,32] or by using from Ranz–Marchall correlations. Most of the models use this approach
Monte-Carlo techniques [25,33]. Estimation of the inter-particle coating but they were validated only against steady-state operational tempera-
variability requires information about spray characteristics and particle ture. While this approach is sufficient to obtain overall heat losses, it can
movement through the spray zone, which can be obtained by experi- not verify if the model can predict dynamic changes in process
mental measurements or by CFD/CFD-DEM simulations. Models of the parameters.
spray zone are especially useful for the prediction of deposition yield Here, we propose a methodology and model for the lighting fast es-
and intra-particle coating variability [34]. timation of the coating temperature, moisture content and coating
Kieckhefen et al. [24] coupled the CFD-DEM model with droplet de- thickness. The model parameters are obtained by fitting a combination
position and evaluated particle surface coverage. Jiang et al. [25] of both steady-state and dynamic experiments. We show that those ex-
coupled CFD-DEM and Monte Carlo method to completely describe periments can be conducted with just a solvent rather than with a
the coating process including the influence of liquid bridges, droplet de- whole formulation that includes an expensive API. On the micro-scale,
position and drying to simulate coating uniformity and agglomeration we evaluate an evaporation time of a single sessile droplet. And by cou-
behavior. These approaches are again strongly limited by number of pling both models, we express this time as a dynamic function of pro-
particles and they are difficult to apply to a real pharmaceutical process cess parameters.
where several millions of particles are present. The applicability of our model is demonstrated on pharmaceutical
In this work, we mainly focus on the coating of small particles for a production data and can predict the temperature with the average accu-
long period, which is the usual case for the deposition of an API. The racy of ±0.63 °C. Additionally, the value of the evaporation time can be
long process time assures the low inter-particle coating variability used for the prediction of agglomeration.
[30]. However, the agglomeration is here the main concern:
2. Device-scale model
1. Agglomerated pellets must be separated before capsule filling, re-
ducing the yield.
2.1. Model assumptions
2. If agglomerated pellets again disjoin before or during separation,
the resulting inter-particle variability (API dosage) will be higher. On the device-scale, the model can predict temperature and mois-
3. If combined with sustained release or enteric protection, the whole ture content in the Wurster fluid bed process. It consists of three com-
formulation might be ineffective. partments described by differential mass and energy balances (see
Agglomeration has several causes. An insufficient drying rate can Fig. 1): one reaction compartment where coating and drying occurs
lead to excessive wetness and result in the coating solution acting as a and two heat transfer compartments (device shell and surrounding air
binder [35]. Conversely, excessive drying or low air humidity can lead layer). We simplify the coating process into mass and heat transfer be-
dry particles to generate electrostatic charges and, thereby, stick tween two phases, the first denoted as particle phase and the second as
together [36]. To prevent agglomeration, it is necessary to keep the wet- air phase, both ideally mixed similarly to work of Burgschweiger and
ness of the particles within a narrow threshold range while maintaining Tsotsas [39]. Information about temperature and moisture gradient

506
J. Kolář, P. Kovačík, T. Choděrová et al. Powder Technology 386 (2021) 505–518

dZ :
mpar ¼ ws mspray , ð2cÞ
dt : :
where mvap is the mass flux due to evaporation, mspray is the inlet mass
:
flow of the coating medium, ws weight is the fraction of solute, mair is
:
the inlet mass flow of dry air and mat is the atomization air flow.
:
The heat fluxes between air and particles Q par and between air and
:
device Q dev are described as
:

Q par ¼ α par Apar T air −T par , ð3Þ

:
Q dev ¼ α dev Adev ðT air −T dev Þ, ð4Þ

where Tair, Tpar, Tdev are the temperatures of air, particles and device, re-
spectively, αi are the corresponding heat transfer coefficients, Adev is the
surface area of device and Apar is the surface area of particles. Apar is re-
lated to the initial particle diameter and solid relative mass fraction Z as
follows
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
3 ρpar
dpar ¼ dpar,0 1þZ , ð5Þ
ρs

Fig. 1. Illustration of the model.

2
mpar dpar
Apar ¼ 6 3
, ð6Þ
inside the device is lost with this approach. However, as we will show, ρpar dpar,0
the outlet air temperature is still predicted with high accuracy.
The main assumptions of the model are: where ρpar is the density of particles and ρs is the density of
solute. The average coating thickness hc is readily available as hc = 0.5
• Air and particles are considered as two ideally mixed phases.
⋅ (dpar − dpar, 0).
• All gases behave ideally.
The heat transfer to surroundings is given by
• Coating medium contains solute (micronized suspension of an API,
polymer) and solvent. :

Q ⁎ ¼ α ⁎am Adev T dev −T ⁎am , ð7Þ
• Only solvent can evaporate. am
• Solute is a multi-component mixture but it is represented by mean
thermodynamic quantities and single weight fraction for model pur-
:

Q am ¼ α am Aam T ⁎am −T am , ð8Þ
poses.
• Particles are spherical pellets with initial constant diameter. where Tam ∗
is the temperature of air layer around the device, which
• Coating medium is sprayed evenly on the particles. changes dynamically with the process and Tam is the ambient tempera-
• Evaporation is described by Eq. (12) based on micro-scale model ture far from the device.
analysis. All heat transfer coefficients are determined numerically so for sim-
plicity we assume Aam = Adev and we can define overall heat transfer co-
2.2. Device-scale model description efficient:
 −1
Quantity of the coating medium is described by its solvent relative 1 1 1
α overall ¼ þ þ , ð9Þ
mass fraction in the particle phase X and in the air phase Y, which we α dev α ⁎am α am
refer to as moisture content, and by its solid relative mass fraction in
the particle phase Z: Enthalpy balances are in the form:
ðl Þ
msol   dT :

ðlÞ ðsÞ
mpar Xcp,sol þ Zcðp,s
sÞ par
X¼ , ð1aÞ þ cp,par ¼ −mvap ΔH vap T par
mpar dt
Z T par
: ðl Þ
ðgÞ
msol −ð1−ws Þmspray cp,sol dT ð10aÞ
Y¼ , ð1bÞ T spray
mair Z T par :
: ðsÞ
−ws mspray cp,s dT þ Q par ,
ms T spray
Z¼ , ð1cÞ
mpar
(g) (l)   dT Z T in Z
where msol , msol are masses of solvent, ms is mass of the solute and mpar, ðgÞ ðgÞ air : air
ðgÞ : T am
ðgÞ
mair are masses of dry particles and dry air, respectively. Average coating mair Ycp,sol þ cp,air ¼ mair cp,air dT þ mat cp,air dT
dt T air T air
thickness, increased particle diameter and surface area is calculated Z ð10bÞ
: T air : :
ðgÞ
from Z and known density of solute. Mass balances of the coating me- −mvap cp,sol dT−Q par −Q dev ,
T par
dium can be written as follows,
dT dev : :
dX ðsÞ
¼ Q dev −Q ⁎, ð10cÞ
mpar _ vap þ ð1−ws Þm
¼ −m _ spray ; ð2aÞ mdev cp,dev
dt dt am

dY ⁎ : :
mair _ air þ m
¼ ðm _ at ÞðY in −Y Þ þ m
_ vap ; ð2bÞ dT am
dt m⁎am c⁎p,am ¼ Q ⁎−Q am , ð10dÞ
dt am

507
J. Kolář, P. Kovačík, T. Choděrová et al. Powder Technology 386 (2021) 505–518

dT RTD
¼ α 0RTD ðT air −T RTD Þ, ð10eÞ 3.2. Micro-scale model description
dt
where cp,i(g,l,s) are the appropriate heat capacities, ΔHvap(Tpar) is the spe-
∗ Sauter diameter of the droplet is estimated from equation proposed
cific evaporation enthalpy of the solvent and mdev, mam are the weights
by Masters [50]:
of the device and the air layer, respectively.
Because the experimental temperature is measured by an RTD ther- pffiffiffiffiffiffiffiffiffi  0:45 : !1:5
585⋅103 3 σ cm μ cm 1000V cm
mometer (see Section 4), which has a longer response time in compar- dd ¼ pffiffiffiffiffiffiffiffi þ 597 pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi : , ð13Þ
urel ρcm σ cm ρcm V at
ison to a thermocouple, we could not neglect it. So, in Eq. (10e) we
introduce a simple one parametric model of thermometer response
time, where αRTD′ is the reduced heat transfer coefficient. where urel is the relative velocity of atomization air flow to sprayed liq-
The mass flux of evaporation is a function of the difference between uid, σcm, μcm, ρcm surface
:
tension,
:
viscosity and density of the coating
the equilibrium moisture content Y ∗ and the actual moisture content Y medium, respectively, V at , V cm volumetric flow rate of atomization air
in the air phase. Y ∗ is computed from and coating medium.
A single droplet creates a spherical cap upon collision with the sur-


⁎ p0sol T par Msol face area given by the contact angle θ:
Y ¼
 ð11Þ
p−p0sol T par Mair 0 1ð1=3Þ
3V d sin ðθÞ3
where p0sol is the equilibrium vapor pressure estimated from the Antoine r base ¼ @  A ð14Þ
π 2−3 cos ðθÞ þ cos ðθÞ3
equation at particle temperature. Effect of solute content will be
discussed in Section 3.  
During a coating in the fluidized bed, only a fraction of particle sur- r base 2
Acap ¼ 2πð1− cos ðθÞÞ , ð15Þ
face is covered by the coating medium and the covered surface can be sin ðθÞ
expressed as a function of the moisture content:
where Vd = md/ρcm is the droplet volume. During evaporation of the
:

mvap ¼ kY Avap ðX Þ Y ⁎ −Y ð12Þ droplet, we will assume that contact angle is still constant and radius
is shrinking. Mass balance for the droplet then can be expressed as:
where kY is the evaporation rate coefficient and Avap(X) is assumed to
dmd

depend linearly on the moisture content, thus we search for relation ¼ −Acap ðmd ÞkY Y ⁎ −Y : ð16Þ
Avap(X) = avapX, where avap is a specific evaporation area, which can dt
be estimated with micro-scale approach, see Section 3 for definition.
Mixture of the solvent, API, polymer and other compounds does not
behave ideally. However, for purpose of our model we will assume that
3. Micro-scale model
saturated vapor pressure can be described by a simplified Florry-
Huggins equation [51], where we neglect any mutual affinity between
3.1. Model assumptions
solvent and other compounds:
On the scale of a single droplet, there can be two evaporation re- p0fh,sol ¼ p0sol ϕsol exp ð1−ϕsol Þ ð17Þ
gimes. The first assumes a sessile droplet on the inert surface, where liq-
uid evaporates directly from the liquid-air interface (see Fig. 2a). p0fh,sol Msol
Whereas, in the second, liquid can penetrate the pellet and the evapora- Y⁎ ¼ , ð18Þ
p−p0fh,sol M air
tion is slowed down by diffusion (see Fig. 2b). In our case, the low poros-
ity of the experimentally used material (≈5%) and higher viscosity due
where ϕsol is the volume fraction of the solvent and is defined as
to usage of polymer favor evaporation directly from the surface.
Summarized main assumptions of the micro-scale model are:  −1
ρsol ms
ϕsol ¼ 1þ , ð19Þ
• Inert pellet surface ðmd ðt Þ−ms Þρs
• No further droplet breakup
• Immediate spreading of the droplet to a shape given by contact angle where ms = wsmd, 0 is the weight of solute in the droplet, ρsol is the den-
• No droplet stacking (only a single droplet is deposited) sity of solvent and ρs is the density of solute. Density of the mixture
• Constant temperature and air moisture content (coating medium) follows ideal mixing rule:
• Saturated vapor pressure of solvent can be described by Flory-  
1−ws ws −1
Huggins equation ρcm ¼ þ : ð20Þ
ρsol ρs

The differential equation for droplet mass balance Eq. (16) can
be now numerically solved in order to obtain total evaporation
time of a droplet at a constant air temperature and moisture
content.
In the case of pure solvent, we can further derive a relation for esti-
mation of evaporation area as a function of the moisture content X. In
the following we assume ws = 0, which enables to analytically solve
Eq. (16). After substitution of Eq. (15) to Eq. (16) and rearranging we
get:

dmd ð2=3Þ

Fig. 2. Illustration of two evaporation regimes during coating.
¼ −Bmd kY Y ⁎ −Y , ð21Þ
dt

508
J. Kolář, P. Kovačík, T. Choděrová et al. Powder Technology 386 (2021) 505–518

B ¼ 2πð1  cosðθÞÞ 4. Experimental

0 1ð2=3Þ
@ 3 ð22Þ
  A ; 4.1. Experimental setup
ρsol π 2  3 cosðθÞ þ cosðθÞ3

All experiments were carried out in Glatt GPCG 2 fluid bed equipped
where we use ρcm = ρsol. Eq. (21) can be then solved for initial condition with a 7″ conical Wurster chamber and a cylindrical expansion cham-
md = md, 0, obtained from the initial droplet volume Vd, 0 (given by ber. The experimental setup can be seen in Fig. 3. A mixture of solvent,
Eq. (13)): API, and polymer is supplied via the peristaltic pump to the spray nozzle

 !3 and the exact spraying rate is determined from a weight difference re-
ð1=3Þ BkY Y ⁎ −Y corded by the digital balance (±0.1 g min−1). The inlet fluidization
md ðt Þ ¼ md,0 − t , ð23Þ
3 air volumetric flow rate is measured with accuracy of ±2 m3 h−1.
Three PT 100 RTD thermometers with the accuracy of ±0.1 °C are re-
total evaporation time of a single droplet is then: sponsible for measuring the temperature of the air at the inlet, outlet
and product temperature. The response time of the RTD thermometer
3 ð1=3Þ was only known for water (t50 = 3.5 s, t90 = 8 s) and we expect it to
t evap ¼
 md,0 : ð24Þ
BkY Y ⁎ −Y be within 1–2 min for a rapidly flowing air. The ambient temperature
is measured by a separate digital thermometer with the same accuracy,
placed aside at a room corner. The air humidity is measured by the ab-
With usage of Eq. (23) and similarly derived equation for sessile
solute humidity sensor placed at the inlet (±0.5 g kg−1).
droplet surface Ad(t), we calculate the time averaged droplet mass and :
area: The atomization air volumetric flow rate V at m3 s−1 for the nozzle
is given as a function of the atomization pressure pat [bar] obtained from
Z t evap
1 md,0 Glatt [52]:
md ¼ md ðt Þdt ¼ ð25Þ
t evap 0 4
: 1
V at ¼ ð1:155⋅pat þ 1:411Þ for 0:8<pat <4: ð31Þ
Z t evap 2=3 3600
1 Bmd,0
Ad ¼ Ad ðt Þdt ¼ , ð26Þ
t evap 0 3

4.2. Validation experiments

Evaporation time of a single droplet is generally low, at maximum a
few seconds. Whereas, Wurster process parameters are usually constant
Pellets made of tartaric acid with a mean diameter 500 μm or 700 μm
for a longer period of time (minutes). Therefore we assume a quasi-
were used as the core material. Their porosity is ≈5%, measured by mer-
steady state and the total evaporation area on the device-scale can be
cury porosimetry. Totally, 800 g of the pellets was fluidized in a single
estimated as:
batch. The atomization pressure was set to 1.5 atm if not stated otherwise.
Ad 4 −1=3
Avap ¼ mpar X ¼ mpar XBmd,0 : ð27Þ 4.2.1. Dry experiments
md 3
During the dry experiments, there was no solvent inlet, but the at-
omization air flow was kept on to support the fluidization. The steady-
Finally, after rearranging we obtain a suitable relation for evapora-
tion area estimation:

1
Avap ¼ avap X ¼ f cap ðθÞ mpar X ð28Þ
dd ρsol

!2=3
1
f cap ðθÞ ¼ 21=3 ⋅8⋅ð1− cos ðθÞÞ⋅ , ð29Þ
2−3 cos ðθÞ þ cos ðθÞ3

where fcap is a spherical cap shape factor.

3.3. Coupling to device-scale model

In the case of pure solvent, Eq. (28) can be readily used. In the case of
nonzero solute content, micro-scale model must be integrated for every
time step of device scale model. Integrals from Eqs. (25) and (26) must
be solved numerically together with averaged volume fraction
Z t evap
1
ϕsol ¼ ϕsol ðt Þdt: ð30Þ
t evap 0

Additionally, Florry-Huggins theory should be included, Eq. (11) is

substituted with Eqs. (17) and (18) using ϕsol .
However, it should be noted that it influences only the liquid mois-
ture content and has a negligible effect on the temperature and the air Fig. 3. Experimental setup, 1 - air compressor with heating control, 2 - Wurster chamber
with atomization nozzle, 3 - Expansion chamber, 4 - Sprayed solution (mass flow rate
moisture content. Thus, usage of pure solvent approximation (only measured by balance), 5 - peristaltic pump, 6 - filter; thermometers: T1 - inlet air
Eq. (28)) to increase the model performance is a viable choice if liquid temperature, T2 - outlet air temperature, T3 - product temperature; V - volumetric
moisture content is not of interest. flowmeter.

509
J. Kolář, P. Kovačík, T. Choděrová et al. Powder Technology 386 (2021) 505–518

Table 2 4.3. Production data

Summary of dynamic experiments.

id # in Tin
air [°C] _ spray ½g min−1  pat [bar] dpar [μm] Analyzed productions were performed in the same device. The coat-
V_ air ½m3 h 
−1 m
ing solution was prepared by mixing ethanol or isopropanol with API
e, 1i 100 35.0 0–18 1.5 700
and other compounds. The exact formulation is a know-how of Zentiva
e, 2i 120 35.0 0–18 1.5 700
e, 3i 80 40.0 0–26 1.5 500
and cannot be published, but the overall solid content will be given. All
e, 4i 100 40.0 0–26 1.5 500 process parameters were taken from the production records including
e, 5i 120 40.0 0–26 1.5 500 batch size, air inlet, outlet temperature, solution composition, and
e 120 25.0 7–27 1.2 500 spray rate. In total, we have analyzed 12 production records.
n 120 25–65 0.0 1.2 500
n 140 25–65 0.0 1.2 500
n 160 25–65 0.0 1.2 500
5. Determination of model parameters
e – ethanol, i – isopropanol, n – no solvent.

5.1. Initial estimation

state temperature was taken after at least 45 min of a steady run. This
temperature was taken before and after each wet experiment discussed At first, we would like to obtain a reasonable guess of the model pa-
below. Regarding dynamic dry experiments, the pellets were fluidized rameters. All thermodynamic properties and device geometry-based
by air at room temperature and subsequently, the inlet air temperature parameters are summarized in Table 3.
was gradually increased in two stages and then decreased. The device weight was estimated as the total mass of the steel com-
ponents in direct contact with the airflow. Most of the components
could not be directly weighted so roughly mdev ≈ 50 kg. RTD thermom-
4.2.2. Wet experiments
eter response time is expected to be 1–2 min, and corresponding value
During the wet experiments, only a pure solvent (ethanol or
of αRTD′ ≈ 0.02 s−1. On the other hand, remaining heat and mass trans-
isopropanol) was sprayed onto the particles. The inlet air velocity was
fer rates are more complex and depend on the character of the airflow in
set to 80, 100 or 120 m3 h−1 and the temperature to 25, 35 or 40 °C.
the Wurster device. It is influenced by several factors including the dis-
The device was first heated until a steady initial temperature was ob-
tributor plate, the device conical geometry, and the particular flow. In
tained. Subsequently, the spraying mass flow rate was gradually in-
general, it is difficult to find a representative length and air velocity
creased in two or three stages and kept constant for each stage. Then,
for correct evaluation of Reynolds number, so for our purpose, we will
it was gradually decreased in the same manner.
simply take an average superficial velocity based on chamber diameter
(0.180–0.295 m) and volumetric flow rate. Theoretical background for
4.2.3. Usage of experimental data the determination of the heat and mass transfer coefficients is described
:
A summary of all dynamic experiments can be seen in Table 2. The below and the results for V air ¼ 120 m3 h−1 and an average tempera-
recorded dynamic response of outlet air temperature together with ture of T = 30∘C are summarized in Table 4. The final comparison to
the steady-state temperature from the first 4 experiments (1e, 1i, 2e, the experimental data can be seen in Fig. 4.
2i) were used for the determination of the mass and heat transfer coef-
ficients. This procedure is discussed in details in Section 5. The remain-
5.1.1. Particle-air mass transfer
ing experimental data were used for model validation.
We estimated mass transfer coefficient, kY, from correlation for Sher-
wood, Reynolds and Schmidt number for a single sphere particle with
diameter, dpar, moving through air at relative velocity u0 [53]:
Table 3
Model parameters (thermodynamic properties are considered as functions of tempera-
ture, here values at T = 30 ° C).
Table 4
Device parameters
Values of used parameters at T = 30∘C and estimated model parameters from correlations.
Total device area Adev 1.6 m2
Total device air volume Vair 0.1 m3 Parameters Value Units
Device specific heat capacity (steel) c(s)
p, dev 502 J kg−1 K−1
dpar 700 μm
:
Pellet material V air 120 m3 h−1
Particle diameter dpar 500, 700 μm dchamber 0.180–0.295 m
Batch size mpar 0.8 kg Lnc 1.678 m
Particle density ρpar 1640 kg m−3 u0 (average) 0.752 m s−1
Particle specific heat capacity c(s)
p,par 1229 J kg−1 K−1 DEtOH 1.35 ⋅ 10−5 m2 s−1
DIPA 1.06 ⋅ 10−5 m2 s−1
Air properties
μair 1.864 ⋅ 10−5 Pa s
Ambient pressure p 101,325 Pa
λair 0.027 W m−1 K−1
Air density ρair 1.159 kg m−3
λpar 1.123 W m−1 K−1
Air specific heat capacity cp, air 1017 J kg−1 K−1
Repar 33 −
Air specific humidity hair 0.008 kg kg−1
Redev 11,000 −
Solvent properties Ra 3 ⋅ 109 −
Ethanol specific heat capacity (g) c(g)
p, EtOH 1438 J kg−1 K−1 ψwp 0.038 −
Ethanol specific heat capacity (l) c(l)
p, EtOH 2480 J kg−1 K−1 ϕw 0.175 −
Ethanol evaporation enthalpy ΔHvap, EtOH 914 kJ kg−1 ε 0.476 −
Isopropanol specific heat capacity (g) c(g)
p, IPA 1545 J kg−1 K−1 mdev 50 kg
Isopropanol specific heat capacity (l) c(l)
p, IPA 2980 J kg−1 K−1 αpar 200.0 Wm−2 K−1
Isopropanol evaporation enthalpy ΔHvap, IPA 749 kJ kg−1 αwa 4.0 Wm−2 K−1
αwp 254.0 Wm−2 K−1
Process parameters αdev 20.0 Wm−2 K−1
Inlet air temperature Tin
air 25–65 °C αam 3.5 Wm−2 K−1
Ambient temperature Tam 22–26 °C α′RTD 0.02 s−1
Temperature of sprayed liquid Tspray 22–26 °C kY, EtOH 0.13 kg m−2 s−1
Solute specific heat capacity c(s)
p, s 1300 J kg−1 K−1 kY, IPA 0.11 kg m−2 s−1

510
J. Kolář, P. Kovačík, T. Choděrová et al. Powder Technology 386 (2021) 505–518

:
Fig. 4. Results of the model with initially estimated parameters V air ¼ 120 m3 h−1, (a) - ethanol, (b) - isopropanol.

and Adev is the total area of the device wall. In Wurster, particles are
kY dpar 1=3
Sh ¼ ¼ 2 þ 0:6 Re 1=2
par Sc , ð32Þ lifted through the Wurster tube, hit randomly the wall (splash zone)
Dρair and then rest in the annular zone (bed zone) with overall downward
movement and slight bubbling. Determination of the total area in con-
dpar u0 ρair μ air
Re par ¼ , Sc ¼ , ð33Þ tact with particles as well as the exact contribution of the splash zone
μ air ρair D
is challenging. So for an initial estimation, we assume that Awp is the
area covered by the loosely packed particles in the annular zone and
where D is diffusivity of the solvent, ρair, μair are density and viscosity of
we neglect the contribution of the splash zone.
the air, respectively.
Heat transfer coefficient for airflow through a tube in a thermally de-
It should be noted that we neglect the effect of possible back-mixing
veloped region can be estimated from the Dittus-Boelter equation for
of air resulting from turbulence and do not correct the Sherwood num-
cooling [55]:
ber as proposed by [54] for regular fluid bed of particles. We assume
that majority of the solvent evaporates during flight of the particle α ⁎wa ddev
and without space-resolution of the moisture content, fluid velocity Nu ¼ ¼ 0:023 Re 0:8
dev Pr
0:3
, Re dev >10000 ð37Þ
λair
and particle movement, the single sphere mass transfer coefficient
seems more appropriate to us. ddev u0 ρair cp,air μ air
Re dev ¼ , Pr ¼ , ð38Þ
μ air λair
5.1.2. Particle-air heat transfer
∗
We estimated heat transfer coefficient for particles, αpar, from corre- where ddev is the characteristic length, and αwa heat transfer coefficient
lation for Nusselt, Reynolds and Prandtl number [53]: for device wall in developed region. In the matter of the Wurster device,
the ratio of its length and diameter is low: Ldev/ddev ≈ 6 − 9, so the flow
α par dpar 1=2 1=3 is not fully thermally developed and the heat transfer coefficient will be
Nu ¼ ¼ 2 þ 0:6 Re par Pr , ð34Þ
λair higher [56]:

dpar u0 ρair cp,air μ air  ð0:7Þ

Re par ¼ , Pr ¼ , ð35Þ α wa ddev Ldev
μ air λair ¼ 1 þ , 2< <20: ð39Þ
α ⁎wa Ldev ddev

where λair is air conductivity and cp, air specific heat capacity of air. Again
Presence of solids significantly increases the heat transfer in wall-
no correction to back-mixing was done.
particle region. Kunii and Levenspiel [57] recommend following expres-
sion for the wall-particle heat transfer coefficient:
5.1.3. Wall-air and Wall-particle heat transfer
The heat transfer to the device wall can be divided into two regions: 2λ0ewp
a particle-free region where heat is transferred mostly by convection α wp ¼ þ 0:05Prair Re par , ð40Þ
dpar
and to a particle region where the conduction of solids also contributes.
The radiant heat transfer can be neglected because of the low- where λ0ewp is the effective conductivity of the wall-particle region
temperature difference between surfaces [53]. With these assumptions, with stagnant gas. If we assume that heat flows in parallel paths
the device heat transfer coefficient is through the stagnant gas and solids, the effective conductivity can be
  expressed as
α dev ¼ 1−ψwp α wa þ ψwp α wp , ð36aÞ
λ0ewp ¼ ελair þ ð1−εÞλpar , ð41Þ
Awp
ψwp ¼ , ð36bÞ
Adev where ε is the void fraction of gas and λg, λpar are the thermal conduc-
where αwa is the wall-air heat transfer coefficient, αwp is the wall- tivities of gas and particles, respectively. However, in fact, there is a
particle heat transfer coefficient, Awp is the area in contact with particles much smaller contact area between adjacent spherical particles and

511
J. Kolář, P. Kovačík, T. Choděrová et al. Powder Technology 386 (2021) 505–518

instead, heat is partly transferred through a gas layer between these reduced version of the model which exclude the air layer. Reduced
∗ ∗
particles. A more realistic model proposed by Kunii and Levenspiel model can be obtained by setting the mam = 0 and αam → ∞, which im-
∗
[57] is then plies Tam = Tdev and
!  −1
1 1 1
λ0ewp ¼ ελair þ ð1−ε Þλpar
 , ð42Þ α overall ¼ þ , ð46Þ
ϕw ⋅ λpar =λair þ 1=3 α dev α am

where ϕw is a dimensionless thickness of the equivalent gas layer, for its In this case, αoverall was computed as an average from all steady-state
exact evaluation see work of Kunii and Levenspiel [57]. In our case λpar/ experiments (1–4).
λair ≈ 42 and ϕw ≈ 0.175 for loosely packed particles with ε ≈ 0.476 in To determine the remaining parameters (mdev, αdev, αpar, αRTD′ and
∗ ∗
the Wurster annular region. additionally αam , mam for the full model) a set of dynamic experiments
with time dependent solvent spray rate was carried out (see
5.1.4. Device-ambient heat transfer Section 4.2, only solvent, ws = 0). We used set of 2 experiments for
Heat to the ambient is transferred mostly by natural convection and each solvent with the volumetric air flow rate of 100 and 120 m3 h−1.
its rate depends on Grashof and Rayleigh number The resulting dynamic response of the temperature was compared
with the predicted temperature from the model. As αpar had an insignif-
gβðT dev −T am ÞL3nc ρair icant influence on the resulting outlet air temperature, it was set to an
Gr ¼ , ð43Þ
μ air initially estimated value of 200 W m−2 K−1 (see Table 4). The other
model parameters were found in order to minimize the sum of squared
Ra ¼ GrPr, ð44Þ differences between the outlet air temperature from the model (given
by temperature of the RTD sensor) and from the experiments at mea-
where β is the coefficient of volume expansion (β = 1/T for ideal gases) sured time intervals τi:
and Lnc is the characteristic length of natural convection, which was
chosen to be the total height of the device due to its overall cylindrical
 1 N  2
E mdev , α dev , kYX , α ⁎am , m⁎am ¼ ∑ T mod exp
air ðτ i Þ−T air ðτ i Þ : ð47Þ
shape. In case of high Rayleigh number Ra ≈ 3 ⋅ 109, we can expect an N i¼1
onset of turbulence and higher heat transfer rate in comparison to lam-
inar flow. In the air, the following correlation for Nusselt number was Eq. (47) has many local minima, so a particle swarm method [59]
proposed [58]: was utilized to explore the domain and a resulting minimum was
found by a pattern search [59] from the most promising point. The
α am Lnc final values of the parameters are summarized in Table 6 and they are
Nunc ¼ ¼ 0:94Ra0:25 : ð45Þ
λair compared to initially estimated parameters from Section 5.3.
In Figs. 5 and 6, it can be seen that both models can predict the tem-
perature response with a high accuracy. The average temperature dif-
5.2. Obtaining parameters by fitting the experimental outlet air ference per measured point ΔT was less than 0.11 °C for the reduced
temperature model. However, as heating of the air around the device is not
accounted for, it slightly overestimates the temperature in Fig. 5
In the previous section, we had to make several assumptions, which where overall heat loss was higher. On the other hand, it underesti-
can be inappropriate because of our limited knowledge of the process. mates the temperature for the process carried out in the heated device
Therefore, we propose a different strategy for estimation of the heat (see Fig. 6). The modeling of the air layer increases the model accuracy
transfer coefficients based on the fitting of the experimental outlet air to the average difference ΔT of 0.074 °C per measured point.
temperature. Although, the mass transfer coefficient can not be ob-
tained in this way. In all experiments with pure solvent, we estimate 5.3. Comparison of obtained model parameters
the evaporation area by Eq. (28), where we use average droplet diame-
ter for the process (dd = 17 μm). Implementation of the whole proce- A comparison of the initially estimated parameters and the optimal
dure is automated in Matlab and can be used for the analysis of any values can be seen in Table 6. The values of αRTD′ are in agreement,
experimental data. and the corresponding RTD thermometer response time is indeed
The overall heat transfer coefficient of the device αoverall was around 2 min. The obtained mass of the device mdev matches the weight
determined from Eq. (9) with the usage of the set of steady-state of the steel components of the Wurster apparatus in order of magni-
experiments without solvent (Table 5). During long processes, the air tude. However, the high mass of the air layer corresponds to the ability
layer around the device is slowly heated up, which results in a long of the air layer to absorb heat, and its numerical value is influenced by
term decrease of overall heat loss. So, we used only the temperatures the ambient air mixing and ventilation.
taken after several hours of the process run (3 and 4 from Table 5) to
obtain αoverall.
In order to verify if the modeling of the air layer significantly in- Table 6
creases accuracy of predicted temperature, we additionally define a Optimal values of model parameters.

Parameter Reduced Full Estimated Units

αpar 200.0 200.0 200.0 W m−2 K−1

Table 5
αdev 28.13 34.96 20.0 W m−2 K−1
Steady-state experimental data and determined coefficients.
αama 8.05 25.78 3.5 W m−2 K−1
# Tin
air [°C] Tpar [°C] Tair [°C] Tam [°C] pat [bar] mdev 46.94 37.71 50 kg
V_ air [m3 h−1]
in
kYX, EtOH 0.13 0.13 0.13 kg m−2 s−1
1 100 35.0 32.20 32.20 24.7 1.5 kYX, IPA 0.11 0.11 0.11 kg m−2 s−1
2 100 35.0 32.3 32.3 24.5 1.5 α′RTD 0.02 0.024 0.02 s−1
∗
3 120 35.0 32.80 32.80 24.7 1.5 αam ∞ 9.71 ∞ W m−2 K−1
∗
4 120 35.0 32.80 32.80 24.6 1.5 mam 0 179.12 0 kg
αoverall 6.26 [W m−2 K−1] (Reduced) Error E(…) 0.016 0.0079 0.23
αoverall 5.87 [W m−2 K−1] (Full) a
Calculated from αoverall.

512
J. Kolář, P. Kovačík, T. Choděrová et al. Powder Technology 386 (2021) 505–518

:
Fig. 5. Results of the model fitting V air ¼ 100 m3 h−1, ethanol, (a) - reduced, (b) - with air layer.

:
Fig. 6. Results of the model fitting V air ¼ 120 m3 h−1, isopropanol, (a) - reduced, (b) - with air layer.

:
Fig. 7. Model temperature prediction, V air ¼ 80 m3 h−1, (a) - ethanol (ΔT = ± 0.28 °C), (b) - isopropanol (ΔT = ± 0.08 °C).

513
J. Kolář, P. Kovačík, T. Choděrová et al. Powder Technology 386 (2021) 505–518

:
Fig. 8. Model temperature prediction, V air ¼ 120 m3 h−1, (a) - ethanol (ΔT = ± 0.16 °C), (b) - no solvent (ΔT = ± 1.52 °C).

Regarding the device heat transfer coefficient αdev, both optimal 7.1. Results for device-scale model
values are higher, than we estimated, so the contribution of the particles
to the total heat transfer is larger. This might be caused especially by the Regarding the device-scale model, we used the previously obtained
heat transfer in the splash zone, which we inappropriately neglected. parameters for the full version of our model (see Table 6). We also ne-
Also, ambient heat transfer coefficients are higher, here we compare glect the presence of solute in the calculation of saturated vapor pres-
∗
αam for the full model and αam for reduced model. This could be related sure and the evaporation area, because we have found out that it has a
to the conductance of heat through the steel components to other parts little influence on the resulting temperature.
of the device and to an overall increase in the heat transfer area. It should be noted that the quality of the production experimental
data varied. First, process conditions were sampled in long intervals,
in some cases, there were only 3 recorded points. Second, the ambient
6. Validation of the model temperature was measured rarely, so we used a default value of 23 °C
where it was unknown.
6.1. Model validation at different flow rates and temperatures

The model with previously obtained optimal parameters (Table 6)

was validated against similar experiments under different process con- Table 7
ditions, which were not used during the temperature fitting. Totally, we Pharmaceutical production data: process conditions and prediction results as average
temperature differences ΔT.
analyzed 7 wet experiments (3-5e, 3-5i, 6e), where the average temper-
ature difference ΔT was ±0.15°C and 3 dry experiments (7-9n), where Ethanol
it was ±1.43°C. # Tin
air [°C]
:
mspray [g min−1] ws dd [μm] ΔT [°C]
V_ air [m3 h−1]
in

In Fig. 7, we can compare model prediction at the higher tempera-

ture and lower volumetric flow rate. In the case of the experiments 1 70–90 25.5–27.0 0.0–37.0 0.16 45.8 ±0.88
2 70–95 23.8–28.1 0.0–37.0 0.16 67.3 ±0.36
with ethanol, the device was slightly overheated probably by the previ- 3 70–100 24.4–35.0 0.0–39.0 0.16 62.8 ±0.99
ous experiment at higher temperature or velocity, which leads to a 4 105–105 24.2–30.4 0.0–38.0 0.16 53.9 ±0.57
higher temperature difference (±0.28°C). Whereas, the error was as 5 105–105 30.0–30.5 0.0–41.0 0.16 68.2 ±0.45
low as ±0.08°C for isopropanol. Model performs well also at lower tem- 6 105–105 23.0–28.0 30.0–30.0 0.16 69.3 ±0.99
7 140–140 21.0–40.0 0.0–35.0 0.16 35.1 ±0.27
perature (see Fig. 8(a)).
8 125–125 24.7–39.9 0.0–35.0 0.16 57.2 ±1.83
Dry experiments prediction has a higher deviation during heating 9 120–120 24.6–28.7 34.0–40.0 0.16 89.1 ±1.34
(see Fig. 8(b)). The device is quite big and heavy and there are addi-
tional initial heat losses that are not considered in the model. However, Isopropanol
once these components reach a steady temperature, the prediction is # Tin
air [°C]
:
mspray [g min−1] ws dd [μm] ΔT [°C]
V_ air [m3 h−1]
in

again precise.
10 75–100 24.7–26.0 8.3–15.0 0.22 34.1 ±0.38
11 100–120 23.3–34.0 0.0–30.0 0.22 40.1 ±0.94
12 100–115 20.9–29.8 10.0–20.0 0.18 24.5 ±0.32
7. Analysis of pharmaceutical production data 13 120–130 23.2–26.2 9.0–18.0 0.20 26.8 ±0.12
14 115–130 22.6–26.3 0.0–23.0 0.20 30.9 ±0.38
We have analyzed data from 22 production records in total. Spraying 15 150–150 24.7–25.8 0.0–22.0 0.20 33.1 ±0.59
solution consisted of ethanol or isopropanol together with a full phar- 16 126–131 24.9–26.4 9.0–18.0 0.20 28.1 ±0.14
17 115–130 24.1–26.0 9.0–18.0 0.20 28.1 ±0.49
maceutical formulation including the API and polymer. We are not 18 125–130 23.0–25.9 10.0–20.0 0.22 29.0 ±0.48
allowed to publish the exact composition of the solution, but we provide 19 130–135 22.0–25.4 10.5–23.0 0.20 33.0 ±0.31
the mass fraction of solid compounds ws and estimated average droplet 20 80–95 30.0–34.1 0.0–17.5 0.20 19.5 ±0.53
size, which was higher especially due to an increase in solution viscos- 21 130 24.2–26.1 10–19 0.20 29.8 ±0.24
22 110–140 26.0–40.0 22–30 0.26 38.2 ±1.45
ity. Process conditions of all experiments are summarized in Table 7.

514
J. Kolář, P. Kovačík, T. Choděrová et al. Powder Technology 386 (2021) 505–518

: :
Fig. 9. Comparison of the best and the worst model prediction (ethanol), (a) - process 7 (ΔT = ± 0.27 °C), (b) - process 8 (ΔT = ± 1.83 °C), m ðlÞ ¼ ð1−ws Þmspray .
spray

: :
Fig. 10. Comparison of the best and the worst model prediction (isopropanol), (a) - process 13 (ΔT = ± 0.12 °C), (b) - process 22 (ΔT = ± 1.45°C), m ðlÞ ¼ ð1−ws Þmspray .
spray

Despite the limited data quality, our model predicted the tempera- solution composition. The mass transfer coefficient covers the effect of
ture response with high accuracy, the average temperature difference the current airflow and particle size. Thus, the resulting evaporation
ΔT was ±0.85 °C (ethanol) and ±0.51 °C (isopropanol). All results are time includes the effect of all process parameters.
also noted in Table 7.
In Figs. 9 and 10, we compare the temperature profile for the best
and the worst performing experiment for each solvent. Although the
process conditions are dynamically changing, our model can still predict
the temperature precisely.

7.2. Results for micro-scale model

On the device-scale, we greatly simplified the evaporation kinetics,

however, it had a negligible effect on the resulting temperature and pre-
diction accuracy. Also, the air moisture content Y reaches its equilibrium
rapidly after any change in process conditions. This enables us, to study
the evaporation on the micro-scale under a quasi-steady state. We can
solve Eq. (16) to obtain droplet evaporation time for the constant values
of Tair and Y at every time step of the device-scale model. Droplet size Fig. 11. Estimated maximal evaporation time of a single sprayed droplet against
was calculated for the current spray rate, atomization pressure, and corresponding air flow rate.

515
J. Kolář, P. Kovačík, T. Choděrová et al. Powder Technology 386 (2021) 505–518

We will now analyze production processes 10–22, where together 7

agglomerated and 6 did not. Two of the unsuccessful experiments were
reported to be additionally highly static due to the very low humidity of
the inlet air.
We evaluated the maximal evaporation time of a single droplet and
the corresponding average airflow rate for each process. Higher airflow
rate is related to the higher kinetic energy of particles and therefore it
should decrease the probability of agglomeration. All successful experi-
ments can be separated by a straight line from the agglomerated ones
(see Fig. 11). The area under this line can be called a safe zone, where
no agglomeration occurs. We may also deduce that one of the static ex-
periments probably agglomerated solely due to the low air humidity,
whereas the other would probably agglomerate also in the more
humid air.
The Wurster annulus, where particles move slowly, is the most crit-
Fig. 12. Iso-surfaces of evaporation time of a single droplet for used coating medium ical zone. If the particles are dried before landing in this zone, they will
(isopropanol) at pat = 1.2 atm. not agglomerate. So here, the evaporation time is a better descriptor for
estimation of agglomeration than particle moisture content, because, on
the device scale, we can not calculate the wetness space distribution.

Fig. 13. Evaporation time of a single droplet for coating medium used in processes 10–22, green area denotes predicted safe zone, (a)–(c) depicted by red rectangles in Fig. 12.

516
J. Kolář, P. Kovačík, T. Choděrová et al. Powder Technology 386 (2021) 505–518

8. Equilibrium values of the evaporation time Declaration of Competing Interest

We conducted a parameter screening for inlet air temperature, spray The authors declare that they have no known competing financial
rate, airflow, and atomization pressure. The resulting diagram for the interests or personal relationships that could have appeared to influ-
same coating medium as used in processes 10–22 can be seen in ence the work reported in this paper.
Fig. 12. The drawn isosurfaces corresponds to the constant evaporation
time value. Extracted slices (see Fig. 13) can be used for a quick process Acknowledgment
comparison, green area denotes predicted safe zone. As long as, the pro-
cess evaporation time is in the safe zone, the particles should not We would like to thank Zentiva, k.s. for providing the experimental
agglomerate. data and supporting this work through the Parc (Applied pharmaceuti-
In terms of controlling the process, the highest influence has cal research center).
the inlet spray rate and airflow. However, lowering the spray rate
means extending greatly the process time. Whereas, increasing References
the airflow might cause more intense abrasion of material and re-
[1] R. Turton, Challenges in the modeling and prediction of coating of pharmaceutical
duction of yield. Similarly, very high atomization pressure reduce dosage forms, Powder Technol. 181 (2008) 186–194, https://doi.org/10.1016/j.
the yield because very small droplets can dry out before deposi- powtec.2006.12.006.
tion. The drying temperature is also limited, usually by polymer [2] L.A. Felton, S.C. Porter, An update on pharmaceutical film coating for drug delivery,
Expert Opin. Drug Deliv. 10 (2013) 421–435, https://doi.org/10.1517/17425247.
glass-transition temperature or thermal-stability of the API. So 2013.763792.
the optimal approach is probably take the highest possible spray [3] F. Lecomte, J. Siepmann, M. Walther, R.J. MacRae, R. Bodmeier, pH-sensitive polymer
rate for the constrained values of inlet airflow, temperature and at- blends used as coating materials to control drug release from spherical beads: elu-
cidation of the underlying mass transport mechanisms, Pharm. Res. 22 (2005)
omization pressure.
1129–1141, https://doi.org/10.1007/s11095-005-5421-2.
[4] Y. Karrout, C. Neut, D. Wils, F. Siepmann, L. Deremaux, P. Desreumaux, J. Siepmann,
Characterization of ethylcellulose: starch-based film coatings for colon targeting,
9. Conclusion Drug Dev. Ind. Pharm. 35 (2009) 1190–1200, https://doi.org/10.1080/
03639040902858868.
We presented a multi-scale thermodynamic model suitable for [5] J. Rantanen, J. Khinast, The future of pharmaceutical manufacturing sciences, J.
Pharm. Sci. 104 (2015) 3612–3638, https://doi.org/10.1002/jps.24594.
the optimization of Wurster fluid bed coating. On the device-scale, the [6] K. Naelapää, P. Veski, H.G. Kristensen, J. Rantanen, P. Bertelsen, Building quality into
model is able to accurately predict temperature changes during the a coating process, Pharm. Dev. Technol. 15 (2010) 35–45, https://doi.org/10.3109/
coating process. Because of the model's low computational require- 10837450902882377.
[7] E. Teunou, D. Poncelet, Batch and continuous fluid bed coating – review and state of
ments, it can be used for fast screening of many different process
the art, J. Food Eng. 53 (2002) 325–340, https://doi.org/10.1016/S0260-8774(01)
conditions. 00173-X.
To obtain model parameters, we suggest to carry out a set of two ex- [8] A. van Kampen, R. Kohlus, Systematic process optimisation of fluid bed coating,
periments: steady-state experiment without spraying any liquid to ob- Powder Technol. 305 (2017) 426–432, https://doi.org/10.1016/j.powtec.2016.10.
007.
tain overall heat loss of the device; dynamic experiment with the [9] K. Naelapää, M. Allesø, H.G. Kristensen, R. Bro, J. Rantanen, P. Bertelsen, Increasing
spraying of coating medium (CM) to obtain the remaining parameters. process understanding by analyzing complex interactions in experimental data, J.
In the case of dynamic experiments, it is possible to use only an organic Pharm. Sci. 98 (2009) 1852–1861, https://doi.org/10.1002/jps.21565.
[10] P. Hadfield, Processing and Equipment Considerations for Aqueous Coatings, 4th ed.
solvent as a model CM instead of the complete formulation including CRC Press, New York, 2016 49–82, https://doi.org/10.1201/9781315369938-4.
API, and other compounds. Despite, usage of simplified CM, the model [11] S. Heinrich, M. Dosta, S. Antonyuk, Multiscale Analysis of a Coating Process in a
was able to successfully predict temperature profile even for processes Wurster Fluidized Bed Apparatus, 2015 83–135, https://doi.org/10.1016/bs.ache.
2015.10.012.
with CM involving an API.
[12] G. Toschkoff, J.G. Khinast, Mathematical modeling of the coating process, Int. J.
The usage of simplified evaporation does not influence the resulting Pharm. 457 (2013) 407–422, https://doi.org/10.1016/j.ijpharm.2013.08.022.
temperature and air moisture content. Wurster process reacts very fast [13] S. Karlsson, I. Niklasson Björn, S. Folestad, A. Rasmuson, Measurement of the particle
to changes in process parameters, and therefore, it is possible to analyze movement in the fountain region of a Wurster type bed, Powder Technol. 165
(2006) 22–29, https://doi.org/10.1016/J.POWTEC.2006.03.014.
the drying rate in detail under quasi-steady state where the process pa- [14] H. Wang, G. Qiu, J. Ye, W. Yang, Experimental study and modelling on gas–solid flow
rameters are taken as constants. in a lab-scale fluidised bed with Wurster tube, Powder Technol. 300 (2016) 14–27,
On the micro-scale, we calculated the evaporation rate of a single https://doi.org/10.1016/j.powtec.2016.01.025 arXiv:arXiv:1402.6991v1.
[15] D. Kitak, R. Šibanc, R. Dreu, Evaluation of pellet cycle times in a Wurster chamber
sprayed droplet. We used this time to compare real pharmaceutical using a photoluminescence method, Chem. Eng. Res. Des. 2 (2018) 1170–1179,
data and it seems that the tendency of agglomeration can be directly re- https://doi.org/10.1016/j.cherd.2018.02.022.
lated to the droplet evaporation time and inlet air velocity. [16] L. Li, A. Rasmuson, A. Ingram, M. Johansson, J. Remmelgas, C. von Corswant, S.
Folestad, PEPT study of particle cycle and residence time distributions in a Wurster
In spite of the model can not a priori predict when particles
fluid bed, AICHE J. 61 (2015) 756–768, https://doi.org/10.1002/aic.14692 arXiv:
would agglomerate, it transforms the design space of many process arXiv:1402.6991v1.
variables into a few which offers a clearer description of the pro- [17] M. Perpar, M. Luštrik, R. Dreu, S. Srčič, I. Žun, Estimating coating quality parameters
cess. In this way, it facilitates the comparison of the conducted ex- on the basis of pressure drop measurements in a Wurster draft tube, Powder
Technol. 246 (2013) 41–50, https://doi.org/10.1016/J.POWTEC.2013.05.005.
periments and helps to design a new one. And thereby, it makes [18] L. Fries, S. Antonyuk, S. Heinrich, S. Palzer, DEM–CFD modeling of a fluidized bed
possible to reduce the number of experiments needed for the spray granulator, Chem. Eng. Sci. 66 (2011) 2340–2355, https://doi.org/10.1016/j.
optimization. ces.2011.02.038.
[19] L. Li, J. Remmelgas, B.G.M. van Wachem, C. von Corswant, M. Johansson, S. Folestad,
Future work should focus on the particle dynamics and micro-scale A. Rasmuson, Residence time distributions of different size particles in the spray
description of droplets deposition and change of physical properties zone of a Wurster fluid bed studied using DEM-CFD, Powder Technol. 280 (2015)
during drying which should make the prediction of agglomeration 124–134, https://doi.org/10.1016/J.POWTEC.2015.04.031.
[20] L. Li, J. Remmelgas, B.G.M. van Wachem, C. von Corswant, S. Folestad, M. Johansson,
more precise.
A. Rasmuson, Effect of drag models on residence time distributions of particles in a
wurster fluidized bed: a DEM-CFD study, KONA Powder Part. J. 2016 (2016)
264–277, https://doi.org/10.14356/kona.2016008.
Funding [21] Z. Jiang, A. Bück, E. Tsotsas, CFD–DEM study of residence time, droplet deposition,
and collision velocity for a binary particle mixture in a Wurster fluidized bed coater,
This work was supported by Specific university research MŠMT Dry. Technol. 36 (2018) 638–650, https://doi.org/10.1080/07373937.2017.1319852.
[22] P. Breuninger, D. Weis, I. Behrendt, P. Grohn, F. Krull, S. Antonyuk, CFD–DEM simu-
No.21-SVV/2019 and the Czech Science Foundation (GAČR project no. lation of fine particles in a spouted bed apparatus with a Wurster tube, Particuology
19-26127X). 42 (2019) 114–125, https://doi.org/10.1016/J.PARTIC.2018.03.015.

517
J. Kolář, P. Kovačík, T. Choděrová et al. Powder Technology 386 (2021) 505–518

[23] T. Lichtenegger, S. Pirker, Recurrence CFD – a novel approach to simulate multiphase [40] F. Ronsse, J.G. Pieters, K. Dewettinck, Combined population balance and thermody-
flows with strongly separated time scales, Chem. Eng. Sci. 153 (2016) 394–410, namic modelling of the batch top-spray fluidised bed coating process. Part I—
https://doi.org/10.1016/j.ces.2016.07.036arXiv:1606.00586. model development and validation, J. Food Eng. 78 (2007) 296–307, https://doi.
[24] P. Kieckhefen, T. Lichtenegger, S. Pietsch, S. Pirker, S. Heinrich, Simulation of spray org/10.1016/j.jfoodeng.2005.09.029.
coating in a spouted bed using recurrence CFD, Particuology 42 (2019) 92–103, [41] C.C. Larsen, J.M. Sonnergaard, P. Bertelsen, P. Holm, A new process control strategy
https://doi.org/10.1016/J.PARTIC.2018.01.008. for aqueous film coating of pellets in fluidised bed, Eur. J. Pharm. Sci. 20 (2003)
[25] Z. Jiang, C. Rieck, A. Bück, E. Tsotsas, Modeling of inter- and intra-particle coating 273–283, https://doi.org/10.1016/S0928-0987(03)00206-9.
uniformity in a Wurster fluidized bed by a coupled CFD-DEM-Monte Carlo ap- [42] S. Gouin, Fluidized bed microencapsulation: thermodynamics of aqueous and
proach, Chem. Eng. Sci. 211 (2020) 115289, https://doi.org/10.1016/j.ces.2019. ethanolic coating processes, J. Microencapsul. 22 (2005) 829–839, https://doi.org/
115289. 10.1080/02652040500317527.
[26] R. Šibanc, S. Srčič, R. Dreu, Numerical simulation of two-phase flow in a Wurster [43] S. Page, K.H. Baumann, P. Kleinebudde, Mathematical modeling of an aqueous film
coating chamber and comparison with experimental results, Chem. Eng. Sci. 99 coating process in a Bohle Lab-Coater, part 1: Development of the model, AAPS
(2013) 225–237, https://doi.org/10.1016/j.ces.2013.05.057. PharmSciTech 7 (2006)https://doi.org/10.1208/pt070242.
[27] S. Karlsson, A. Rasmuson, B. Van WaChem, I.N. Björn, CFD modeling of the wurster
[44] S. Page, K.-H. Baumann, P. Kleinebudde, Mathematical modeling of an aqueous film
bed coater, AICHE J. 55 (2009) 2578–2590, https://doi.org/10.1002/aic.11847
coating process in a Bohle lab-coater: part 2: application of the model, AAPS
arXiv:0201037v1.
PharmSciTech 7 (2006) E1–E8, https://doi.org/10.1208/pt070242.
[28] P. Rajniak, F. Stepanek, K. Dhanasekharan, R. Fan, C. Mancinelli, R. Chern, A com-
bined experimental and computational study of wet granulation in a Wurster [45] M.T. am Ende, A. Berchielli, A thermodynamic model for organic and aqueous tablet
fluid bed granulator, Powder Technol. 189 (2009) 190–201, https://doi.org/10. film coating, Pharm. Dev. Technol. 10 (2005) 47–58, https://doi.org/10.1081/PDT-
1016/J.POWTEC.2008.04.027. 35915.
[29] X.X. Cheng, R. Turton, The prediction of variability occurring in fluidized bed coating [46] S. Heinrich, J. Blumschein, M. Henneberg, M. Ihlow, M. Peglow, L. Mörl, Study of dy-
equipment. I. The measurement of particle circulation rates in a bottom-spray fluid- namic multi-dimensional temperature and concentration distributions in liquid-
ized bed coater, Pharm. Dev. Technol. 5 (2000) 311–322, https://doi.org/10.1081/ sprayed fluidized beds, Chem. Eng. Sci. 58 (2003) 5135–5160, https://doi.org/10.
PDT-100100546. 1016/J.CES.2003.08.010.
[30] X.X. Cheng, R. Turton, The prediction of variability occurring in fluidized bed coating [47] P. Rajniak, C. Mancinelli, R.T. Chern, F. Stepanek, L. Farber, B.T. Hill, Experimental
equipment. II. The role of nonuniform particle coverage as particles pass through the study of wet granulation in fluidized bed: impact of the binder properties on the
spray zone, Pharm. Dev. Technol. 5 (2000) 323–332, https://doi.org/10.1081/PDT- granule morphology, Int. J. Pharm. 334 (2007) 92–102, https://doi.org/10.1016/j.
100100547. ijpharm.2006.10.040.
[31] D.F. Sherony, A model of surface renewal with application to fluid bed coating of [48] L.X. Liu, J.D. Litster, S.M. Iveson, B.J. Ennis, Coalescence of deformable granules in wet
particles, Chem. Eng. Sci. 36 (1981) 845–848, https://doi.org/10.1016/0009-2509 granulation processes, AICHE J. 46 (2000) 529–539, https://doi.org/10.1002/aic.
(81)85037-3. 690460312.
[32] S.J. Maronga, P. Wnukowski, Modelling of the three-domain fluidized-bed particu- [49] C. Rieck, A. Bück, E. Tsotsas, Estimation of the dominant size enlargement mecha-
late coating process, Chem. Eng. Sci. 52 (1997) 2915–2925, https://doi.org/10. nism in spray fluidized bed processes, AICHE J. 66 (2020), e16920, https://doi.
1016/S0009-2509(97)00112-7. org/10.1002/aic.16920.
[33] K.Z. KuShaari, P. Pandey, Y. Song, R. Turton, Monte Carlo simulations to determine [50] K.K. Masters, Spray Drying Handbook, Longman Scientific & Technical 1991.
coating uniformity in a Wurster fluidized bed coating process, Powder Technol. [51] B. Guerrier, C. Bouchard, C. Allain, C. Bénard, Drying kinetics of polymer films, AICHE
166 (2006) 81–90, https://doi.org/10.1016/j.powtec.2006.05.001. J. 44 (1998) 791–798, https://doi.org/10.1002/aic.690440404.
[34] G. Toschkoff, S. Just, A. Funke, D. Djuric, K. Knop, P. Kleinebudde, G. Scharrer, J.G.
[52] GLATT, Manual of System GLATT LabSystem, 2006.
Khinast, Spray models for discrete element simulations of particle coating processes,
Chem. Eng. Sci. 101 (2013) 603–614, https://doi.org/10.1016/J.CES.2013.06.051. [53] D. Kunii, O. Levenspiel, Particle-to-gas mass and heat transfer, Fluidiz. Eng. 80
[35] K. Christoph Link, E.-U. Schlünder, Fluidized bed spray granulation: investigation of (1991) 257–276, https://doi.org/10.1016/B978-0-08-050664-7.50017-2.
the coating process on a single sphere, Chem. Eng. Process. Process Intensif. 36 [54] H. Groenewold, E. Tsotsas, Predicting apparent Sherwood numbers for fluidized
(1997) 443–457, https://doi.org/10.1016/S0255-2701(97)00022-6. beds, Drying Technology, Vol. 17, Marcel Dekker Inc 1999, pp. 1557–1570, https://
[36] F. Fotovat, X.T. Bi, J.R. Grace, Electrostatics in gas-solid fluidized beds: a review, doi.org/10.1080/07373939908917635.
Chem. Eng. Sci. 173 (2017) 303–334, https://doi.org/10.1016/J.CES.2017.08.001. [55] R. Winterton, Where did the Dittus and Boelter equation come from? Int. J. Heat
[37] A.S. Prata, A. Maudhuit, L. Boillereaux, D. Poncelet, Development of a control system Mass Transf. 41 (1998) 809–810, https://doi.org/10.1016/S0017-9310(97)00177-4.
to anticipate agglomeration in fluidised bed coating, Powder Technol. 224 (2012) [56] C.J. Geankoplis, Transport Processes and Unit Operations, PTR Prentice Hall, 1993.
168–174, https://doi.org/10.1016/J.POWTEC.2012.02.049. [57] D. Kunii, O. Levenspiel, Heat transfer between fluidized beds and surfaces, Fluidiz.
[38] K. Dewettinck, A.D. Visscher, L. Deroo, A. Huyghebaert, Modeling the steady-state Eng. (1991) 313–336, https://doi.org/10.1016/B978-0-08-050664-7.50019-6.
thermodynamic operation point of top-spray fluidized bed processing, J. Food Eng. [58] A. Ede, Advances in Free Convection, National Engineering Laboratory, East Kilbride,
39 (1999) 131–143, https://doi.org/10.1016/S0260-8774(98)00144-7. Scotland, 1967 1–64, https://doi.org/10.1016/S0065-2717(08)70272-7.
[39] J. Burgschweiger, E. Tsotsas, Experimental investigation and modelling of continu- [59] Matlab, Version 9.1 (R2016b). , The MathWorks Inc., Natick, Massachusetts, 2016.
ous fluidized bed drying under steady-state and dynamic conditions, Chem. Eng.
Sci. 57 (2002) 5021–5038, https://doi.org/10.1016/S0009-2509(02)00424-4.

518