Drying Technology

An International Journal

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

Development and validation of a batch fluidized

bed dryer model for pharmaceutical particles

Francis Gagnon, Jocelyn Bouchard, André Desbiens & Éric Poulin

To cite this article: Francis Gagnon, Jocelyn Bouchard, André Desbiens & Éric Poulin (2020):
Development and validation of a batch fluidized bed dryer model for pharmaceutical particles,
Drying Technology

To link to this article: https://doi.org/10.1080/07373937.2019.1701005

Published online: 06 Jan 2020.

Submit your article to this journal

View related articles

View Crossmark data

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

https://www.tandfonline.com/action/journalInformation?journalCode=ldrt20
DRYING TECHNOLOGY
https://doi.org/10.1080/07373937.2019.1701005

Development and validation of a batch fluidized bed dryer model

for pharmaceutical particles
Francis Gagnona , Jocelyn Bouchardb, Andr
e Desbiensa, and
Eric Poulina
a
Department of Electrical and Computer Engineering, LOOP, Universit
e Laval, Quebec City, Quebec, Canada; bDepartment of Chemical
Engineering, LOOP, Universit
e Laval, Quebec City, Quebec, Canada

ABSTRACT ARTICLE HISTORY

Batch fluidized bed drying is a common process for water removal in pharmaceutical par- Received 29 July 2019
ticles. Multiple phenomenological models for this type of dryer are proposed in the litera- Revised 1 December 2019
ture, with many of them relying on the two-phase theory. This article elaborates on a two- Accepted 2 December 2019
phase model and extends it for the pharmaceutical case by improving the particle, batch
KEYWORDS
size and heat loss descriptions. An explicit methodology with low computational cost is also Batch fluidized bed dryer;
developed for solving the fluidized bed equations. Then, based on a grey-box modeling two-phase theory; grey-box
strategy, parameters with known value are fixed, while the unknown parameters are cali- model; system
brated with experimental data collected on a pilot scale fluidized bed dryer. Lastly, the identification;
model is dynamically validated on distinct experimentation datasets. It is shown that the model validation
simulations are able to reproduce the pilot data trends, but there is still room for improve-
ments in the heat loss representation.

Abbreviations: DAE: differential-algebraic equation; FBD: fluidized bed dryer; FE: forward
Euler; K-L: Kunii and Levenspiel; NIR: near-infrared; NRMSE: normalized root mean square
error; ODE: ordinary differential equation; PID: proportionnal integral derivative; PLS: partial
least squares; RK4: forth order Runge-Kutta; SD: standard deviation

1. Introduction range of gas velocity.[2] Dry and Judd[3] also show

that heterogeneous two-phase models with bypasses fit
The pharmaceutical production, and other sectors like
better to gas residence time data. These conclusions
food and mining, frequently relies on batch fluidized
bed dryer (FBD) for their water removal needs in par- are furthermore coherent with the perceptible gas
ticulate materials. While they are efficient dryers, their bubbles, generally occurring in fluidized bed.[4]
operation in industry are typically fixed procedures, Among all the FBD phenomenological models in
which is presumably suboptimal. Process modeling is the literature, one of the most influential work is
one of the most versatile and promising tools to opti- Pal
ancz,[5] which is fundamentally grounded on the
mize production efficiency. In opposition to empirical Kunii and Levenspiel (K–L) two-phase theory.[4] Many
approaches, starting from the first principles can publications are derived from these two references,
potentially increase the applicability range. However, both in batch[6–11] and continuous operation.[12–16]
the growing complexity of modern industrial proc- While they incorporate novel elements to the K-L the-
esses puts a curb on pure first principle developments. ory, common incoherencies appear at the application
Phenomenological modeling is a good compromise in of the K-L interchange coefficients and porosity pat-
such circumstances. terns. Most of them assume an adiabatic vessel (no
The FBD phenomena can be represented with heat losses), an unrealistic scenario when there is min-
either a single homogeneous phase of gas and particles imal insulation like in the pharmaceutical sector. The
or heterogeneous phases with bypasses. Gavi[1] obtains descriptions also sometimes lack explanations, at least
good agreements between a single-phase model and in the author’s point of view, regarding modeling
batch drying experiments at pilot scale. On the other assumptions and limitations. Moreover, the lack of
hand, these approaches normally apply for narrower efficient algorithms for solving the resulting equations

CONTACT Francis Gagnon [email protected] Department of Electrical and Computer Engineering, LOOP, Universit
e Laval, Quebec City,
Quebec, G1V 0A6, Canada.
Color versions of one or more of the figures in the article can be found online at www.tandfonline.com/ldrt.
ß 2019 Taylor & Francis Group, LLC
2 F. GAGNON ET AL.

limits their applicability for optimization. Indeed, coefficients. Indeed, it is well known that overall empir-
many authors use an iterative trial and error search to ical relations fit better fluidized bed experimentations
estimate the nontrivial variables, which is computa- (see Appendix A.3).
tionally expensive. Lastly, the experimental validation To begin with, these assumptions apply throughout
is generally restricted to steady operation of laboratory the model:
scale FBDs, considerably different than the production
ones. Gavi[1] performs a validation at the pilot scale on
The temperature and pressure operating range are
a simpler single-phase model. narrow. Therefore, the change in physical properties
In this work, the pharmaceutical FBD model suc- of components due to temperature and pressure is
cinctly presented in Gagnon et al.[17] is described in neglected. For gases, the changes due to moisture are
details and refined to be coherent with the work of Li neglected too.
and Duncan[10] and Pal
ancz[5], and the underlying
The density and thermal capacity of the solid and
assumptions. The particle shape and the batch size liquid components are much larger than for gases.
calculations are improved for a more realistic repre- As such, steady-state can be assumed for mass and
sentation of pharmaceuticals drying. The heat loss heat transfers on gaseous control volumes.
model is also extended by incorporating losses in the
freeboard region and with the vessel surroundings. The single particle drying model involves mass and
The evaluation sequence of the mathematical expres- heat exchanges between an isolated wet particle and a
sions is described and an explicit method based on circulating convective gas. A stagnant gas film resists
space discretization is proposed to solve the nontrivial exchanges between the two. This model basically sup-
variables. Finally, a grey-box identification algorithm poses that:
with experimental data from a pilot scale FBD enables
a proper calibration of the phenomenological model,
The particle is small and non-hygroscopic, its shape
thereafter validated on similar but distinct datasets. is nonspherical and its volume is invariant during
Validation is performed for the general case of the drying. The internal diffusion of water and heat
unsteady FBD operation (dynamical validation). is neglected. Thereby, the moisture and temperature
This article is organized as follows. Section 2 introdu- within the particle is perfectly uniform.
ces the general methodology, postulates, and limitations
The stagnant gas film is thin and in thermo-
of this work. The specific assumptions and developments dynamic equilibrium with the particle.
of the single particle and the batch FBD model are exten-
The convective gas is uniformly distributed around
sively examined in Sections 3 and 4, respectively. The the particle (no axial or radial dispersion).
resolution of the resulting equations is detailed afterward
in Section 5. Lastly, Section 6 describes the experimental For the FBD model, the exchanges occur between
calibration and validation on the pilot scale process. two heterogeneous phases inside the fluidized bed: an
Further details on the single particle and FBD model are emulsion and a bubble phase. The emulsion is a per-
provided in Appendix A and Appendix B, respectively. fect mix of the wet particles and an interstitial gas.
The gas bubbles pass through the bed but bypass the
emulsion phase thus decreasing gas-particle contact.
2. General modeling assumptions
The freeboard region above the fluidized bed is a per-
As stated by Burgschweiger and Tsotsas,[18] a rigorous fectly mixed gas. The main assumptions of the batch
way of modeling fluidized bed kinetics is by clearly FBD model are:
separating single particle to fluidization and gas-
related phenomena. This methodology facilitates
There is no interaction between the wet particles.
model development for scale-up purposes. It also ena- They all have the same physical properties, mois-
bles independent validations of each subparts. Finally, ture content, and temperature at a given time dur-
it is a modular approach since the particle side can ing the drying.
potentially be any transport processes.
The emulsion phase remains at minimum fluidiza-
Based on this strategy, the convective drying model tion conditions and is homogeneous, i.e., the inter-
of an isolated wet particle is initially developed. stitial gas, particles, and emulsion are perfectly
Afterward, the batch FBD model is derived from the mixed. Direct exchanges do not occur between the
single particle equations. The separation between them bubble phase, particles and vessel wall; only indir-
is systematic with the exception of the particle transfer ect transfer is possible via the interstitial gas. The
 DRYING TECHNOLOGY 3

Figure 1. Cross section of a single wet particle (moisture streams in blue and energy streams in red).

excess inlet airflow above minimum fluidization Two types of moisture can be identified on the par-
conditions goes to the bubble phase. ticle depicted in Figure 1:

The bubble phase contains no particles and the cloud
and wake region is considered negligible (i.e., zero 1. the free moisture, i.e., the water retained on the
thickness). Therefore, the bubbles exchange mass and surface and inside external pores of the solid
energy only with the interstitial gas. They pass material (square marks in Figure 1); and
through the bed in a plug flow manner. Thus, bubble 2. the bound moisture, i.e., the water absorbed
moisture content and temperature are distributed inside internal pores of the solid material (circle
along the height of the fluidized bed. The bubbles marks in Figure 1).
are all identical, spherical at a constant size, and the
In this work, the word particle always refers to the
gas inside them is perfectly mixed.
solid material and associated water seen as a whole.

The freeboard gas region, free of particles,
Its moisture content on a dry basis is:
exchanges heat with the vessel wall by convection
mw
only. vp ¼ (1)

The vessel wall temperature is space invariant, and ms
conduction through the equipment is neglected. with ms, qs, and cs as the mass, density, and specific
Natural convection occurs between the wall and heat capacity of the solid material, and mw, qw, and cw,
ambient gas. the liquid water values (including the two types of mois-

First order dynamics are assumed for the closed- ture). No water vapor is assumed inside the particle.
loop response of the inlet temperature and airflow. When it comes to drying, if the solid material is
supposed non-hygroscopic, a sequential water removal
The next two sections elaborate on the single par- progression can be assumed. Thus, the free moisture
ticle and the batch FBD models respectively. needs to be totally removed first before bound mois-
ture starts to evaporate. The critical moisture content
vpc is defined as the point where all the free moisture
3. Single particle drying model
is removed from the particle. The drying kinetics can
It is necessary to model initially how a single isolated be consequently divided into two distinct parts sepa-
wet particle behaves in the context of convective dry- rated by the critical point vpc:
ing for a given surrounding gas condition. In Section
4, the case of multiple particles with gas interstices
the constant rate period, i.e., the evaporation of the
between them is described as a perfect emulsion for free moisture (square marks in Figure 1), fol-
the FBD modeling. For this purpose, the surrounding lowed by;
convective gas is immediately called the interstitial gas
the falling rate period, i.e., the evaporation of the
in this section, with the subscript e. bound moisture (circle marks in Figure 1).

3.1. Assumptions It is also assumed that the surrounding convective

gas around the particle, the so-called interstitial gas, is
A wet particle of a mass mp and density qp is sup- air with uniform moisture content ve and temperature
posed to be small enough so its moisture content vp Te. Additionally, a thin stagnant gas film surrounding
and temperature Tp are both lumped, hence, space the particle (gray region in Figure 1) is assumed in
invariant. The considered shape is nonspherical, as it thermodynamic equilibrium with it and moistened at
is generally the case for pharmaceutical granules.[19] vpf. This film is the boundary layer that resists mass
4 F. GAGNON ET AL.

and heat exchanges between the moving interstitial 3.2. Mass balance
gas and the particle.
With m _ p, pf being the water vapor mass flow rate
During the constant rate period, there is enough
from the particle to the stagnant gas film and t the
free moisture on the particle surface to keep the film
time, the moisture mass balance around the particle
at saturation. Using both the dry air and water vapor _ mass streams in
volume Vp gives (see the blue m
molar mass in an ideal gas mixture, the saturated air Figure 1):
moisture content vsat (or mixing ratio) can be
dmw
extracted from their partial pressures ratio and _ p, pf
¼ m (6)
Dalton’s law:[20] dt
Note that, in this work, the dot notation indicates
Mwv pwv ðTÞ
vsat ðTÞ ¼ (2) rate variables and the comma in subscript refers to
Mg Pa  pwv ðTÞ moisture or heat streams between regions. After sev-
eral substitutions and manipulations, detailed in
where Mwv and Mg are the water vapor and dry air molar
Appendix A.1, Equation (6) is rewritten as:
mass respectively, and Pa, the ambient pressure. The partial
pressure of the water vapor at saturation pwv can be eval- dvp 6
qpd ¼ kpf , e qg ðvpf  ve Þ (7)
uated with an empirical relation like the Tetens equation dt /dp
(high accuracy in the considered range, see Ref.[21]):
with the dry air density qg, particle size dp and spher-


T  273:15 icity /. Calculations of the dry particle skeletal density
pwv ðTÞ ¼ 610:78 exp 17:27 (3) qpd and mass transfer coefficient between the film and
T  35:85
interstitial gas kpf,e are provided in Appendix A.
Equations (2) and (3) are given as generic expres-
sions for a given gas temperature T as they will come
3.3. Energy balance
in handy in multiple contexts.
During the falling rate period, vpf starts to decrease The energy balance around the particle volume Vp gives
since the moisture needs to diffuse from the particle (see the red Q_ and I_ energy streams in Figure 1):
internal pores to the surface to evaporate. This dIp
decrease, which is related to sorption characteristics of ¼ Q_ pf , p  I_ p, pf (8)
dt
the particle material, is positively correlated with the
where Ip is the particle relative enthalpy, Q_ pf , p , the
particle internal moisture content. A simple empirical
heat flow rate from the gas film to the particle, and
function w with fitting coefficients is employed here, I_ p, pf , the enthalpy flow rate from the particle to the
as common in literature:[10,12,15,16] gas film. Since the particle loses energy due both to the
8 water vaporization (i.e., the latent heat) and the water
>
> vp  vpc
<1 mass removal, I_ p, pf needs to be included as an out-
 
wðvp Þ ¼ vp ðvpc þ jÞ (4) going enthalpy flow rate from the particle. Following
>
: vpc ðvp þ jÞ vp < vpc
>
Appendix A.2 manipulations, the energy balance (8)
becomes:
where  and j are the exponent and offset coefficients
dTp 6
respectively. qpd ðcs þ vp cw Þ ¼ fhe, pf ðTe  Tp Þ
dt /dp
As a result, the gas film moisture content on dry (9)
basis vpf for the whole drying process (over and under  kpf , e qg ðvpf  ve Þ½Kwv
the critical point vpc) is the saturated moisture content þ ðcwv  cw ÞðTp  Twv Þg
vsat weighted by the correction function w:
with the water vapor specific heat capacity cwv and
vpf ¼ vsat ðTp Þwðvp Þ (5) vaporization enthalpy Kwv at Twv. The evaluation of
the heat transfer coefficient between the interstitial gas
since the thermodynamic equilibrium implies that the and film he,pf is given in Appendix A.3.
gas film temperature is Tpf ¼ Tp :
Appendix A gives further details on the single par-
ticle drying model assumptions. The following two
4. Batch fluidized bed dryer model
sections apply conservation laws from the moisture The batch FBD model that follows is substantially
and energy streams depicted on Figure 1. based on the K-L two-phase theory. A fundamental
 DRYING TECHNOLOGY 5

Figure 2. Side-view diagram of a two-phase batch fluidized bed dryer (moist gas streams in black, moisture streams in blue and
energy streams in red).

 !
observation about fluidized particles is that they 1 _0
m 1 V_ 0
behave in many points like boiling liquids. The main t0 ¼ ¼ (10)
Sf qg 1 þ v0 Sf 1 þ v0
ideas and assumptions on bubbling fluidized bed kin-
etics are explained in this section. Further details can by assuming that the moist gas density is invariant at
be found in Kunii and Levenspiel.[4] the dry air value qg.1 The inlet gas is divided between
two heterogeneous phases: a dense and dilute phase.
They respectively consist of:
4.1. Assumptions
The fluidized bed consists of wet particles and a fluidizing 1. the emulsion phase, i.e., a perfect emulsion of all
gas that is channeled through a perforated plate at the the particles (subscript pt) and the interstitial gas
bottom of the vessel, called the inlet plate (or distributor). (subscript e); and
The cross-sectional area of the fluidized column is sup- 2. the bubble phase, i.e., spherical gas bubbles (sub-
posed constant at Sf. A side-view diagram of the batch script b) free of particles (hence, perfectly diluted
FBD model is depicted in Figure 2. Blue and red arrows of particles).
denote exchanges between the model compartments:
There seems to be a common confusion in the lit-

the bubble phase Vb erature on proper phases identification of fluidized

the emulsion phase Ve þ Vpt beds. Since the classification is typically based on sep-

the freeboard region Vr arate heterogeneous regions that exchange mass and

the vessel wall Vv energy within the bed, the perfect gas-particle mix of
the emulsion phase is considered as one entity.2
Since moisture contents are provided on a dry basis, Similarly to particle, the expressions inlet gas, emul-
the dry air component needs to be extracted from the sion (interstitial) gas, bubble gas, and outlet gas con-
humid gas flows. The inlet gas, moistened at v0 and sistently refer to the dry air and water vapor mix in
heated at T0, is blown at a total flow rate m _ 0 (or volu- these respective regions. Focusing only on the dry air
metric flow rate V_ 0 ). Knowing that the dry mass flow at component of each phase, the steady-state mass bal-
the inlet is m_ 0 =ð1 þ v0 Þ, the dry air superficial velocity ance around the inlet plate gives (see the black m _
t0 at the inlet can be evaluated with: moist gas streams in Figure 2):
6 F. GAGNON ET AL.




_0
m _e
m _b
m exchanges between the wall and the particles are incor-
0¼   (11)
1 þ v0 1 þ ve 1 þ vb porated into the interstitial gas heat loss model (see
Appendix B.7). There is no heat loss though the vessel
where m _ e and ve are the emulsion gas flow rate and
in the bubble phase. The vessel is externally cooled
moisture content, and m _ b and vb are the bubble gas
down with natural convection along the vessel height zv.
flow rate and moisture content respectively. Figure 2
There are additional heat losses through the vessel
shows that both gas flows arise from the inlet at the
wall in the freeboard region, above the fluidized bed.
bottom, run through the whole fluidized bed, and
The transfer area is inversely proportional to the bed
finally merge in the freeboard. expansion. The freeboard gas is assumed free of particles
The emulsion phase behaves like a perfectly stirred
and perfectly mixed. The particle presence is however
reactor. By assuming minimum fluidization conditions, considered in the heat transfer coefficient calculation (see
the dry air superficial velocity in the interstitial gas is Appendix B.7). From (10), the steady-state dry mass bal-
fixed at tmf. The dry air flow in the emulsion phase is ance around the fluidized bed Vf gives the freeboard
therefore: _ moist gas streams in Figure 2):
flow (see the black m




_e
m m_r m_0
¼ Sf tmf qg (12) ¼ ¼ Sf t0 qg (15)
1 þ ve 1 þ vr 1 þ v0
The minimum fluidization superficial velocity tmf with m_ r and vr as the freeboard flow rate and mois-
represents the minimal inlet gas velocity t0 value to ture content, respectively.
keep the particle bed in the fluidization state. Its value Additional information on the FBD model assump-
is estimated in Appendix B.5. tions is available in Appendix B. Conservation laws
Knowing that the absolute velocity of the dry bubble are applied on all the model compartments in the fol-
gas tb is normalized by the cross-sectional area occupied lowing sections.
by the phase Sb, the dry air flow in the bubble phase is:


m_b
¼ Sb tb qg ¼ dSf tb qg (13) 4.2. Bubble phase
1 þ vb
To simplify the conservation equations, Kunii and
with d, the volume fraction occupied by the bubbles. Levenspiel[4] suggested normalizing the mass and heat
They run through the fluidized bed in a plug flow transfer on the bubbles volume Vb in the fluidized
manner while collecting moisture from, and providing bed. From this standpoint, they defined a normalized
heat to the emulsion gas along the way. This implies mass interchange coefficient K per unit time and vol-
that the moisture content vb and temperature Tb of ume of bubbles. Physically, it can be interpreted as a
the bubble phase is distributed in z along the flow of gas from a phase to another with an equal
expanded bed. Using the batch dry weight mst, skeletal flow in the opposite direction. A normalized heat
density qpd and bed void fraction ef, the expanded bed interchange coefficient H per unit time and volume of
height zf is estimated with: bubbles was analogously defined for the heat transfers.
mst Multiple fluidized bed models in the literature rely
zf ¼ (14)
Sf ð1  ef Þqpd on these normalized interchange coefficients. However,
there is sometimes confusion at applying them correctly
The last expression deserves comments. Most authors in the bubble conservation law. Since they are already
rely on the fluidized bed height zf to parameterize the batch based on a unit volume of bubbles, the volume fraction
size (quantity of particles) in the FBD model. A major occupied by the bubbles d should be absent of the bal-
drawback of this approach comes at model validation, ance equations using absolute velocities tb (as shown in
when an expansion chamber at the base of the vessel (a Appendix B.1 and Ref.[4]).
truncated cone) is present on the real process invalidating
the proportionality assumption between the volume and 4.2.1. Mass balance
height. This is the case for most of the pharmaceutical The steady-state moisture mass balance around the
FBDs. The dry weight parameter mst and (14) represent a infinitesimal control volume dVb gives (see the blue
better choice to model the batch size here. m_ mass streams of the bubble phase in Figure 2):
The fluidized bed loses heat through the cylindrical
_ b, b ðzÞ  m
0¼m _ b, b ðz þ dzÞ þ dm
_ e, b (16)
vessel of inner and outer diameter dv and dvo, respect-
ively. The vessel wall temperature is uniform at Tv, Likewise the bubble moisture content vb, the water
receiving heat from the emulsion gas. The heat _ b, b is a
vapor mass flow rate within the bubble phase m
 DRYING TECHNOLOGY 7

_ e, b is evaluated
function of z. The infinitesimal value dm solution is elaborated at the model resolution part in
from the global water vapor mass flow rate m _ e, b , com- Section 5.
ing from the emulsion gas. As demonstrated in
Appendix B.1.1, Equation (16) can be simplified to: 4.3. Emulsion phase
dv
tb b ¼ Ke, b ½ve  vb ðzÞ (17) The emulsion phase comprises perfectly mixed wet
dz
particles and interstitial gas. As shown in Figure 2, the
The normalized mass interchange coefficient interstitial gas exchanges mass and heat with the inlet
between the emulsion and bubble gas Ke,b and the gas, particles, bubble phase and freeboard gas. Heat
absolute bubble velocity tb is calculated in losses occur through the vessel wall. Transfer phe-
Appendix B. nomena with the particles are based on the single par-
The bubble moisture content next to the inlet plate ticle model developed in Section 3, without
should be the same as the inlet gas moisture content interaction between them. Consequently, conservation
v0, thus: laws for the emulsion particles Vpt lead to the same
expressions (with vp ¼ vpt and Tp ¼ Tpt ). The mass
vb ðz ¼ 0Þ ¼ v0 (18)
and energy balances around Vpt are therefore directly
The ordinary first-order differential equation (17) given by (7) and (9), respectively. The following two
can be solved analytically with the initial condition sections establish mass and energy balances for the
(18), giving: emulsion gas Ve.


Ke, b
vb ðzÞ ¼ ve  ðve  v0 Þ exp  z (19) 4.3.1. Mass balance
tb
Taking the interstitial gas Ve as the control volume, the
steady-state moisture mass balance gives (see the blue m_
4.2.2. Energy balance mass streams of the emulsion phase in Figure 2):
Selecting the same control volume dVb , the steady- _ 0, e  m
0¼m _ e, r þ m
_ pt, e  m
_ e, b (23)
state energy balance is (see the red Q_ and I_ energy
streams of the bubble phase in Figure 2): where m _ pt, e are the incoming water vapor mass
_ 0, e , m
flow rates from the inlet and all the particles respect-
0 ¼ I_ b, b ðzÞ  I_ b, b ðz þ dzÞ  dQ_ b, e þ dI_ e, b (20) ively, and m _ e, r is the outgoing flow rates to the free-
with I_ b, b as the enthalpy flow within the bubble phase, board region. Defining ~v b as the space-average
Q_ b, e , the global heat flow to the emulsion gas, and moisture content in the bubble phase:
ð
I_ e, b , the global enthalpy flow from the emulsion gas. 1 zf
~v b ¼ v ðzÞdz (24)
The incoming enthalpy flow rate from the emulsion is zf 0 b
required because of the vapor mass migration coming
Equation (23) simplifies to (see Appendix B.2.1 for
from the interstitial gas (neglected in Ref.[5]). With
details):
Appendix B.1.2 manipulations and cg, the dry air spe-
cific heat capacity, the bubble phase energy balance tmf 6ð1  ef Þ
ðve  v0 Þ ¼ kpf , e ðvpf  ve Þ  dKe, b ðve  ~v b Þ
becomes: zf /dp
 (25)
  dTb H b, e
tb cg þ vb ðzÞcwv ¼ ½Tb ðzÞ  Te 
dz qg The bed porosity ef and bubble volume fraction d
 are estimated in Appendix B. Knowing that the spatial
þ Ke, b ½ve  vb ðzÞcwv (21) distribution of the bubble phase moisture content is
given by (19), the analytical solution to (24) gives:
Appendix B.6 provides calculations for Hb,e, the 
 
tb Ke, b zf
normalized heat interchange coefficient between the ~v b ¼ ve þ ðve  v0 Þ exp 1
Ke, b zf tb
bubble and emulsion gas. The bubble temperature
(26)
next to the inlet plate should equal to the inlet gas
temperature T0, giving the initial condition:
4.3.2. Energy balance
Tb ðz ¼ 0Þ ¼ T0 (22)
The steady-state energy balance around the interstitial
The ordinary differential equation (21) is nonlinear, gas volume Ve gives (see the red Q_ and I_ energy
making it difficult to solve analytically. A discrete streams of the emulsion phase in Figure 2):
8 F. GAGNON ET AL.

0 ¼ I_ 0, e  I_ e, r  Q_ e, pt þ I_ pt, e þ Q_ b, e  I_ e, b  Q_ e, v 4.4.2. Energy balance

(27) The steady-state energy balance around the freeboard
volume Vr gives (see the red Q_ and I_ energy mass
with Q_ e, v , the heat flow rate to the vessel wall. I_ 0, e streams of the freeboard region in Figure 2):
and I_ e, r are enthalpy flow rates, respectively coming
0 ¼ I_ e, r þ I_ b, r  I_ r, out  Q_ r, v (32)
from the inlet and leaving to the freeboard. Q_ e, pt and
I_ pt, e are the particle heat and enthalpy flows. Lastly, where Q_ r, v represents the heat loss to the vessel wall in
Q_ b, e and I_ e, b represent global energy exchanges with the freeboard. Like the mass balance, the enthalpy flow
the bubble phase. Defining T ~ b as the space-average to the outlet is I_ r, out and the value from the bubble
bubble gas temperature: phase is I_ b, r ¼ I_ b, b ðzf Þ: Following Appendix B.3.2
ð zf manipulations, the freeboard energy balance becomes:
~b ¼ 1
T Tb ðzÞdz (28) ( )
zf 0 4ðzv  zf Þhr, v
t0 ðcg þ vout cwv Þ þ Tout
dv qg
Equation (27) expands to (see Appendix B.2.2 for
details): ¼ dtb ½cg þ vb ðzf Þcwv Tb ðzf Þ (33)
tmf dHb, e ~ 4ðzv  zf Þhr, v
ðcg þ v0 cwv ÞðTe  T0 Þ ¼ ðT b  Te Þ þ tmf ½cg þ ve cwv Te þ Tv
zf qg dv qg
( )
6ð1  ef Þ he, pf with Tout, the outlet gas temperature. An estimate of
þ kpf , e ðvpf  ve Þcwv þ ðTp  Te Þ
/dp qg the heat transfer coefficient between the freeboard gas
and the vessel wall hr,v is provided in Appendix B.7.
4he, v
 ðTe  Tv Þ
dv qg 4.5. Vessel wall
(29)
A constant vessel wall temperature during the batch is
The heat transfer coefficients from the emulsion gas not realistic for fluidized bed drying,[10,23] especially
to the vessel wall he,v is computed in Appendix B.7. when there is no insulation. Hence, likewise
Burgschweiger et al.,[24] heat exchanges between the wall
and the ambient gas is incorporated to the model, yield-
4.4. Freeboard region ing an additional energy balance equation.
Since pharmaceutical production FBDs are tall to
accommodate various batch sizes, they frequently exhibit 4.5.1. Energy balance
a zone between the end of the fluidized bed and the top The energy balance around the vessel wall volume Vv
of the vessel called the freeboard. The outlet gas instru- gives (see the red Q_ energy streams in the emulsion
mentation is typically placed after. Assuming no heat phase and freeboard region of Figure 2):
loss over this wide transfer area is not realistic, especially dIv
¼ Q_ e, v þ Q_ r, v  Q_ v, a (34)
in pharmaceutical applications, operating with minimal dt
insulation for sanitary reasons. where Iv is the vessel relative enthalpy and, Q_ v, a , the heat
flow rate to the ambient gas. With Appendix B.4.1
4.4.1. Mass balance replacements, the vessel wall energy balance simplifies to:
The steady-state moisture mass balance around the
_ mass mv cv dTv
freeboard volume Vr gives (see the blue m ¼ zf he, v ðTe  Tv Þ þ ðzv  zf Þhr, v ðTout  Tv Þ
streams of the freeboard region in Figure 2): pdv dt
dvo
_ e, r þ m
0¼m _ b, r  m
_ r, out (30)  zv hv, a ðTv  Ta Þ
dv
where m _ r, out is the moisture flow to the gas outlet. From (35)
Figure 2, the incoming moisture flow from the bubble
The heat transfer coefficient between the vessel wall
phase m _ b, r equals to m_ b, b (zf). With Appendix B.3.1 sub-
and the ambient gas hv,a is evaluated in Appendix B.7.
stitutions, the freeboard mass balance reduces to:
t0 vout ¼ tmf ve þ dtb vb ðzf Þ (31) 4.6. Inlet humidity, heater and blower
and denoting the outlet gas moisture content Prior to heating, the ambient gas at temperature Ta
with vout. and relative humidity Ua exhibits a moisture content
 DRYING TECHNOLOGY 9

v0 on dry basis given by: equations. All these calculations are implemented in
v0 ¼ Ua vsat ðTa Þ (36) the MATLAB environment. This section describes the
DAEs explicit solving method.
with the function vsat ðTÞ provided in (2).
The inlet heater and blower transient responses
need to be characterized. Typically, the inlet tempera- 5.1. State-space representation
ture and flow rate are under proportionnal integral The manipulated inputs of the model are the inlet
derivative (PID) closed-loop control. Feedback is flow rate V_ sp and temperature Tsp setpoints. The rela-
known to improved system linearity at lower frequen- tive humidity Ua and batch dry weight mst parameters
cies.[25] Thus, empirical first-order linear models are both included in the inputs since their measured
between the controller setpoints and the process varia- values differ significantly between batches. The model
bles are assumed for the sake of simplicity. Knowing input vector u is therefore:
that the PID integral action ensures a closed-loop >
static gain of one, the inlet heater model between the u ¼ ½ V_ sp Tsp Ua mst  (39)
temperature setpoint Tsp and process value T0 is: The other fluidized bed parameters are assumed
dT0 Tsp  T0 both invariant in time and between the batches. The
¼ (37)
dt sT model state vector x are the five differential variables:
and the inlet blower model between the volumetric x ¼ ½ V_ 0 T0 vp Tp Tv > (40)
flow setpoint V_ sp and process value V_ 0 is:
Knowing that all the states are measured except the
dV_ 0 V_ sp  V_ 0 wall temperature Tv, and, with an available outlet tem-
¼ (38)
dt sV_ perature Tout, the state-space representation of the
where sT and sV_ are the closed-loop time constants of batch FBD model is:
the inlet heater and blower controllers respectively. dx h i>
V_ 0 dT0 dvp
¼ fðx, uÞ ¼ ddt dTp dTv
(41a)
dt dt dt dt dt

>
5. Model solving y ¼ hðx, uÞ ¼ ½ V_ 0 T0 wp Tp Tout  (41b)
Solving the equations of a multiple phase reactor where y and h(x, u) are the model output vector and
model is not a trivial task as they frequently form a function respectively, and f(x, u) is the state derivative
coupled system of differential-algebraic equations function. The particle wet basis moisture content wp
(DAEs). The algebraic relations come from the com- is related to the dry basis vp with (A2). The time
mon equilibrium assumptions, and the differential derivative of V_ 0 , T0, vp, Tp, and Tv are provided in
equations arise from the conservation laws. (7,9,37,38), and (35) respectively.
Multiple methods are available to solve DAE sys- It it worth noting that there is no direct trans-
tems. A major challenge is in finding a consistent ini- mission with the manipulated inputs V_ sp and Tsp,
tial solution for handling differential equations.[26] If owing to the heater and blower closed-loop dynam-
computing cost is of concern (e.g., for model opti- ics of (37) and (38). A direct transmission with
mization, process control, state estimation), the simple manipulated variables is an unlikely assumption. As
trial and error search proposed by many a matter of fact, the input vector u appears in the
authors[5,7,9,10,12,15,16] is inefficient. Explicit calcula- output function (41b) strictly because of the param-
tions are preferred in these cases, if available. eters Ua and mst effect on Tout (both are not directly
Pal
ancz[5] and Li and Duncan[10] suggested explicit manipulable).
approaches, but their simplification is incompatible
with the plug flow postulate of the bubble phase.[12]
5.2. Solution overview
In the batch FBD model, there are five differential
equations, giving the five differential variables V_ 0 , T0, The state derivative f(x, u) and the output h(x, u)
vp, Tp, and Tv, and multiple algebraic equations. functions, as defined in (41a) and (41b), respectively,
However, only four algebraic variables are nontrivial: both require a series of pre-calculations before evalu-
ve, ~ ~ b : The outlet values vout and Tout are
v b , Te, and T ating their results. The sequence of these pre-calcula-
calculated from these these five differential and four tions is given in the flowchart of the Figure 3. For the
algebraic variables. The other coefficients in the model output function h(x, u), the algebraic variables ~v b , ve,
can be easily computed with explicit and independent ~ b , and Te current values are necessary for
T
10 F. GAGNON ET AL.

Figure 3. Workflow of the batch fluidized bed dryer model.

determining the outlet temperature Tout. All the 5.3. Nontrivial algebraic variable calculations
returned results of Figure 3 are also required for the
The emulsion gas moisture content ve and bubble gas
state derivative function f(x, u). The calculations of
~ b , and Te in average value ~v b are to be computed before the evalu-
the nontrivial algebraic variables ~v b , ve, T ~ b : Since
ation of their respective temperature Te and T
Figure 3 are explained in the next section.
a solution is available for the bubble mass balance
Having all the necessary algebraic variable values at
ODE, provided in (19), the relation between the aver-
a given time (or a consistent initial condition), the
age bubble moisture ~v b and emulsion moisture ve is
DAE system is transformed into ordinary differential explicitly given by (26). Combining (26) with the
equations (ODEs). It can be solved thereafter with any emulsion mass balance of (25) produces a linear equa-
numerical ODE solvers, with fixed or variable time tion system:
steps. The fixed step forth order Runge-Kutta (RK4)
method yields stable and coherent results with time h i ~v
h i
av11 av12 b
¼ bbv1 (42)
step Dt up to 30 s for the parameter set in this work. av21 av22
|fflfflfflfflfflffl{zfflfflfflfflfflffl}
ve v2
|ffl{zffl}
Note that the sampling time is limited by the fastest Av bv
dynamics for fixed step algorithms, which is typically
the blower closed-loop model of (38). with the Av matrix coefficients given by:
 DRYING TECHNOLOGY 11

av11 ¼ 1 (43a) Tb ½n ¼ Te þ ðT0  Te Þh½n (48)


 
tb Ke, b zf and the coefficients h½n are:
av12 ¼ 1 þ exp 1 (43b) 8
Ke, b zf tb >
>
<1 n¼0
av21 ¼ dKe, b (43c) h½n ¼ Y
n1
(49)
>
> 1  Dzf ½j n > 0
tmf 6ð1  emf Þ : j¼0
av22 ¼   kpf , e  dKe, b (43d)
zf /dp
The FE method corresponds to the left-endpoint
and the bv vector components as: rectangle method for integration. Hence, the average

  ~ b is computed from (48) and:
tb Ke, b zf temperature T
bv1 ¼ v0 exp 1 (44a) ð
Ke, b zf tb
~ 1 zf 1 nX z 2

Tb ¼ Tb ðzÞdz  Tb ½n
tmf 6ð1  emf Þ zf 0 nz  1 n¼0
bv2 ¼ v0  vpf kpf , e (44b) (50)
zf /dp
1 nX z 2

¼ Te þ ðT0  Te Þ h½n
The ~ v b and ve values are calculated by solving the nz  1 n¼0
linear system (42). The bubble gas moisture content at
the top of the fluidized bed vb ðzf Þ is computed after- Combining (50) with the emulsion energy balance
ward by inserting the evaluated ve and zf in (19). of (29) leads to:
The spatial ODE of the bubble phase energy balance h i ~
h i
Tb
aT11 aT12
¼ bbT1 (51)
in (21) is nonlinear. Strictly speaking, the model is a sys- aT21 aT22 T T2
|fflfflfflfflfflffl{zfflfflfflfflfflffl} e |ffl{zffl}
tem of partial differential and algebraic equations. AT bT
However, partial derivatives can be avoided by noting
that time and space derivatives are decoupled in the sys- with the AT matrix coefficients given by:
tem, meaning that they can be solved as separate ODEs. aT11 ¼ 1 (52a)
The z coordinate is solved with the explicit forward
1 nX z 2
Euler (FE) method to keep these internal calculations as aT12 ¼ 1  h½n (52b)
simple as possible. Similarly to the dimensionless height nz  1 n¼0
strategy of Burgschweiger et al.,[24] instead of fixing the dHb, e
sampling distance at Dz, the sample quantity is main- aT21 ¼ (52c)
qg
tained constant at nz. This has the advantages of nor- ( )
malizing the spatial coordinate and dissociating the 6ð1  ef Þ he, pf
problem from the varying bed expansion. aT22 ¼ kpf , e ðvpf  ve Þcwv þ
/dp qg
For a specified bed height zf, the sampling distance
Dz is adapted with: dHb, e 4he, v tmf ðcg þ v0 cwv Þ
  
zf qg dv qg zf
Dz ¼ (45)
nz  1 (52d)
to produce nz samples from z ¼ 0 to zf inclusively. and the bT vector components as:
The ODE of the bubble phase energy balance is:
1 nX z 2
Hb, e
þ Ke, b ½ve  vb ðzÞcwv bT1 ¼ T0 h½n (53a)
dTb qg nz  1 n¼0
¼   ½Te  Tb ðzÞ (46)
dz tb cg þ vb ðzÞcwv ( )
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl} 6ð1  ef Þ he, pf
f ðzÞ bT2 ¼ Tp kpf , e ðvpf  ve Þcwv þ
/dp qg
with vb ðzÞ provided in (19). Using the bracket nota-
4he, v tmf ðcg þ v0 cwv Þ
tion for the discretized function f ½n ¼ f ðnDzÞ, with  Tv  T0
n, a nonnegative integer for the nth sample, the FE dv qg zf
method on (46) gives: (53b)
Tb ½n þ 1 ¼ Tb ½n þ Dzf ½nðTe  Tb ½nÞ (47) ~ b and Te is done by solving the
The calculation of T
linear system (51). The bubble temperature at the top
With the initial condition T0 , inserting some n val-
of the fluidized bed Tb ðzf Þ is thereafter evaluated from
ues in (47) exposes the following pattern:
(48), with Te and the last h coefficient:
12 F. GAGNON ET AL.

high-shear step with a chopper and impeller is per-

formed for four minutes. The produced wet granules
are first sieved through a 1.40 mm mesh before split-
ting them into different batch sizes for drying.
Since some parameters of the FBD model are
unknown or hard to measure, they need to be cali-
brated with experimental data. As detailed in Section
6.2, the sorption parameter , critical moisture vpc,
particle size dp, sphericity /, and bubble size db
require calibration. Thus, a nonlinear grey-box model3
estimation algorithm, available in the system identifi-
cation toolbox of MATLAB, is used on data coming
from multiple batches (multiple experiment identifica-
tion). In short, the grey-box identification algorithm is
a nonlinear least squares optimization problem mini-
mizing the distance between simulated model outputs
and experimental data, using parameters as decision
variables. The resulting calibrated model is cross-vali-
dated on distinct datasets. All the model fit bench-
marks provided in this section are based on the
normalized root mean square error (NRMSE), relying
on the averaged measurement for normalization.
Figure 4. Pilot scale batch fluidized bed dryer.

Tb ðzf Þ ¼ Tb ½nz  1 ¼ Te þ ðT0  Te Þh½nz  1 (54) 6.1. Inlet heater and blower closed-loop models

It is stressed here that the sample quantity nz Before proceeding with the grey-box calibration, the
should be sufficiently high to minimize the error closed-loop time constants sT and sV_ are separately
induced by the first-order FE method. Several simula- estimated on identification data coming from the pilot
tions suggest that nz  40 is enough to keep the simu- FBD. Running the dryer empty, identification and val-
lated Tb ðzÞ curves unchanged (with the parameter set idation tests are achieved both for the heater and
of this work). blower, presuming a stable and sufficiently damp-
ened response.
The identified time constants of the linear systems
6. Calibration and validation are shown at the end of Table 1 (highlighted in gray).
The model is validated on a pilot scale FBD. The Comparisons of the calibrated heater and blower
experimental setup described by Gagnon et al.[17] is models with the validation data are presented at
used: an Aeromatic MP-1 fluidized bed instrumented Figures 5 and 6, respectively. High values of NRMSE
with multiple sensors and actuators (max capacity of (72% and 73% in Figures 5 and 6), and narrow par-
25 L). An inline near-infrared (NIR) probe made by ameter standard deviations (SDs), provided in the last
Viavi Solution (formerly JDSU) gives a particle mois- column of Table 1, indicate that the inlet closed-loop
ture measurement with a pre-calibrated partial least models are sufficiently accurate.
squares (PLS) model. All the states x and outputs y
are measured except the vessel wall temperature. The
6.2. Batch fluidized bed dryer model
data sampling period is 1 s. A picture of the experi-
mental setup is shown at Figure 4. When possible, model parameters that are known are
The wet granules are prepared with 3.14 kg of cal- set to their measured or expected values. They are all
cium carbonate powder granulated with two exci- listed in Table 1 (not highlighted), with details on the
pients in small quantity. The dry ingredients are selected values provided in the last column. The other
mixed on a vertical high-shear granulator for five parameters are estimated from the calibration dataset,
minutes. Afterward, 0.712 kg of purified water is made of three batches, as described in Table 2 and
slowly incorporated through an atomizer. The final illustrated in Figure 7.
 DRYING TECHNOLOGY 13

Table 1. Batch fluidized bed dryer model parameters.

Par. Value Comment
nz 40 Sufficient for small FE error
g 9.807 m/s2 Standard value
qg 1.127 kg/m3 Value at 40  C
cg 1.007 kJ/(kg K) Value at 40  C
Mg 28.97 g/mol Standard atomic weights
lg 19.07 mPa s Value at 40  C
kg 27.35 mW/(m K) Value at 40  C
qw 992.3 kg/m3 Value at 40  C
cw 4.180 kJ/(kg K) Value at 40  C
cwv 1.867 kJ/(kg K) Value at 40  C
Mwv 18.02 g/mol Standard atomic weights
Dwv 27.33 mm2/s Value at 40  C
Kwv 2.406 MJ/kg Value at Twv
Twv 313.15 K Plausible constant value
qs 2765 kg/m3 Calcium carbonate value
cs 834.3 J/(kg K) Calcium carbonate value
j 0.010 Small positive value
 2.604 Estimated SD: 0.048
vpc 0.04011 Estimated SD: 0.00037
dp 498.6 mm Estimated SD: 12.1 mm
/ 0.5206 Estimated SD: 0.1808
db 9.369 mm Estimated SD: 3.438 mm
mv 110.0 kg Approximate weight from dimensions and density
cv 500.0 J/(kg K) Stainless steel 316 value
dv 0.250 m Expansion chamber lower and upper inner diameter mean
dvo 0.260 m Wall thicknesses added to dv
zv 2.200 m Cylinder height at dv giving a 1.7 m2 inner surface
Ta 298.15 K Conventional normal value
Pa 101.3 KPa Conventional normal value
sT 86.74 s Estimated SD: 1.77 s
sV_ 24.67 s Estimated SD: 0.36 s

Figure 5. Heater closed-loop model and validation data comparison.

Figure 6. Blower closed-loop model and validation data comparison.

 >
The grey-box identification algorithm is a least w ¼ 0 0 1:0 0:5 0:8 (55)
squares optimization problem between simulations
and data. The problem is solved with a trust region excluding V_ 0 and T0 since they are already calibrated
reflective method. The objective function allows out- in Section 6.1.
put weighting with w, which is selected from the max- All parameters that are a function of temperature
imal range of each outputs y in data: are evaluated at 40  C, approximately corresponding
14 F. GAGNON ET AL.

to the middle point of the operation range.4 As shown corresponding to the beginning of Tp increase, vp 
in Figure 4, the pilot vessel is not completely cylin- 0:03 here), the optimization problem uses a ± 0.015 con-
drical. The mean value of the expansion chamber strained range around this reading. The bubble size db,
lower and upper diameter measurements gives the particle size dp, and sphericity / are all estimated since
inner diameter dv. The wall thickness dimensions are they are distributed in practice and their effective value
added to this value for the dvo parameter. The height is uncertain. This is especially true given that a particle
of a cylinder having a diameter dv and the same inner sieve analysis resulted in a wide distribution
surface as the pilot represents the vessel height zv. The (span ¼ 3:0) with a bimodal shape. The particle size dp
inner surface on the pilot vessel is 1.7 m2. is also constrained between the d10 and d90 measure-
Even if both parameters j and  are empirical ments, which are 79.51 mm and 1084 mm respectively.
(thus unknown), j is not estimated since its influence The initial state values V_ 0 , T0, vp, Tp, and Tv are also
on the model is negligible, resulting in a high uncer- estimated in each batches both at calibration and valid-
tainty (also corroborated by Ref.[10]). The critical ation. They are respectively constrained at ± 0.005 m3/s,
moisture vpc can slightly vary from batch to batch. ±10.0 K, ±0.05, ±10.0 K, and ±20.0 K of their initial
The calibrated value thus represents an optimal aver-
measurement, using Tout as an approximation for the
age value. Knowing that vp can be roughly estimated
initial Tv value. The calibrated parameters and their SDs
graphically on data from one batch (vp value
are available in Table 1 (highlighted in blue). It’s worth
Table 2. Available data collected on the pilot scale process.
mentioning that the SD of / is higher because of its
NRMSE (%)
lower sensitivity in the model (at the current dp value).
Table 2 provides the fit between the calibrated
Batch Usage mst (kg) wp Tp Tout
B1 Calibration 4.38 78.49 50.46 33.73
model and the validation dataset, made of three
B2 Calibration 3.91 76.36 84.05 37.02 batches. It must be emphasized that step changes in
B3 Calibration 3.78 76.03 78.23 53.62 operating conditions were imposed to obtain input
B4 Validation 4.71 70.75 73.60 87.84
B5 Validation 3.98 70.19 60.42 39.08 signals displaying a rich frequency content in order to
B6 Validation 3.67 83.92 86.64 51.69 capture the process dynamics. To facilitate

Figure 7. Model calibration dataset: batches B1, B2, and B3.

 DRYING TECHNOLOGY 15

Figure 8. Calibrated model comparison with B4 data: steady operation at V_ sp ¼ 150 m3/h and Tsp ¼ 65 C.

Figure 9. Calibrated model comparison with B5 and B6 data: unsteady operation.

16 F. GAGNON ET AL.

interpretation, Figure 8 compares the calibrated model representation of the particle size could also improve
outputs with B4 data, operated with constant inlet set- simulations, since the measured distribution is
points, a common practice in the industry. A simulation wide here.
with the adiabatic vessel assumption, as typical in the lit-
erature, is also depicted in dashed line (he, v ¼ hr, v ¼ 0).
7. Conclusion
These results highlight that the proposed heat loss
model brings distinct improvement in Tout simulation This article revisits the fundamentals of batch fluidized
and Tp, for the falling rate period. bed dryer modeling and proposes a validation for
Figure 9 shows a comparison with B5 and B6 data, pharmaceutical applications. The main contributions are
including steps on setpoints to validate effects of to clarify assumptions made in the literature, develop
dynamical changes. As expected, the model perform- heat loss descriptions, propose a computationally low-
ances are generally lower compared to the constant cost workflow for the evaluation of non-trivial variables,
operation case in batch B4. The model is nevertheless and clearly describe the equation solving sequence. The
able to reproduce the general trends, especially for the model is calibrated with a grey-box identification algo-
particle moisture content wp and temperature Tp. rithm and validated on a pilot scale process. Results
However, the outlet temperature Tout predictions are showed that the simulations can reproduce the pilot
less accurate, particularly for the batch B5. data, but the heat loss representation would require to
For 5 < t < 8 min in B5, Figure 9 shows a dip in the be upgraded for additional accuracy.
simulated Tp caused by inlet temperature drop, which is For future work, the low computational cost of the
absent from the data. This dissimilarity is presumably solving strategy could be exploited for real-time appli-
caused by the mismatch on the moisture content wp. cations. Moisture sensors for solids are expensive and
While the model is still in the constant rate (vp > vpc ), not fully fault-proof. State estimation and virtual
the falling rate is beginning on the real system. The inlet measurement are concrete examples that would bene-
temperature decrease cancels out roughly with the nat- fit both from the model low computational cost, and
ural temperature increase of the falling rate, giving an the improved accuracy at the product and outlet tem-
apparent constant response on the pilot. peratures. Also, the batch processing is based on the
Note that a 19% solid material loss was detected single particle kinetic and most of the parameters
during the batch B6, caused by an imperfect sealing in have a physical meaning. These aspects could be bene-
the outlet filter assembly. Since particles are hotter ficial for process development and scale-up purposes.
and more prone to elutriation at the end of a batch Finally, the generalization to continuous dryers could
(lower density, fines generation, etc.), the high Tout be interesting to explore, knowing that this type of
value at the end of B6 data could be explained by this production is a promising path for the pharmaceut-
loss. Notwithstanding, the model is still able to repli- ical industry.
cate the overall trends. Furthermore, since a 19%
reduction of mst does not significantly impact model Nomenclature
simulations, the mass loss effect in B6 seems less Latin symbols
important compared to the heat loss.
Because the drying in B5 was significantly faster, the c isobaric specific heat capacity, J kg–1 K–1
majority of the batch time was spent in the falling rate D molecular diffusion coefficient, m2 s–1
d diameter, m
period. Hence, the proposed heat loss model seems to f bubble temperature derivative function, m–1
be less representative in the falling rate than the con- g gravitational acceleration, m s–2
stant rate period. This is not surprising knowing that H heat interchange coeffcient normalized on bubble
the temperature gradients in the system are large during volume, W m–3 K–1
the falling rate period. Tout simulations nonetheless sig- h heat transfer coeffcient, W m–2 K–1
I relative enthalpy, J
nificantly improved compared to results using the adia- i specific relative enthalpy on dry basis, J kg–1
batic vessel assumption as typical in the literature. K mass interchange coeffcient normalized on bubble
Further developments include asserting the mixing volume, s–1
model of the freeboard and modeling the wall tempera- k mass transfer coeffcient, m s–1
ture as a distributed variable by considering conduction M molar mass, kg mol–1
m mass, kg
through the vessel. The freeboard model could possibly n sample
include the particle presence with the transport disen- P absolute pressure, Pa
gagement height concept.[4] A nonuniform p partial pressure, Pa
 DRYING TECHNOLOGY 17

Q heat, J wv water vapor/vaporization

S surface area, m2 w liquid water of the particle
T thermodynamic temperature, K
t time, s Operators
V volume, m3
w moisture content on wet basis
_ flow rate of
, s–1
z height above inlet plate, m ~
space-average of
along the expanded bed height

, in subscript: moisture or heat from region
to 
Greek symbols
d bubble fraction in the bed volume Notes
e void fraction
1. A model sensitivity analysis of qg limiting cases
h discrete bubble temperature coeffcients
j offset constant in w function validates this assumption.
K specific enthalpy of vaporization, J kg–1 2. A three-phase model would include a separate region (thus,
k thermal conductivity, W m–1 K–1 separate balance equations) between the bubbles and the
l dynamic viscosity, Pa s emulsion often called the cloud and wake region.[4,22]
 exponent constant in w function 3. A grey-box model combines features of both first-
q density, kg m–3 principles (e.g. physical laws, conservation principles) and
s closed-loop response time constant, s empirical models (e.g. regressions or transfer functions).
t dry air velocity, m s–1 4. A sensitivity analysis on the dependency of parameters
U relative humidity on temperature, for limit cases (20 and 100  C), attests
/ particle sphericity this simplification.
v moisture content on dry basis
w gas film correction function based on the particle
sorption characteristics Acknowledgments
The authors would like to thank Pfizer Montreal and the
Vectors and matrices National Sciences and Engineering Council of Canada
(NSERC) for funding this research, and Pfizer Montreal for
A equation system coeffcients, (2  2) providing the necessary raw materials and laboratory equip-
b equation system constants, (2  1) ment. They specially acknowledge the valuable time, sup-
f model state derivative function, (5  1) port, and efforts invested by Pierre-Philippe Lapointe-
h model output function, (5  1) Garant, B
arbara Santos Silva, Sophie Hudon and Yuwei
u model inputs, (4  1) Zhang (Pfizer Chair, Universit
e de Sherbrooke).
w output weigths for model calibration, (5  1)
x model states, (5  1)
y model outputs, (5  1) ORCID
Francis Gagnon http://orcid.org/0000-0002-5037-0232
Subscript
0 inlet gas
a ambient gas References
b bubble phase gas [1] Gavi, E. Application of a Mechanistic Model of
c cloud phase Batch Fluidized Bed Drying at Laboratory and Pilot
e emulsion phase: interstitial gas Scale. Drying Technol. 2019, 1–17. DOI: 10.1080/
f expanded fluidized bed
07373937.2019.1611594.
g dry air
[2] Mostoufi, N.; Cui, H.; Chaouki, J. A Comparison of
mf minimum fluidization condition
Two- and Single-Phase Models for Fluidized-Bed
out outlet gas
pc particle critical moisture content Reactors. Ind. Eng. Chem. Res. 2001, 40, 5526–5532.
pd dry particle skeleton DOI: 10.1021/ie010121n.
pf emulsion phase: particle stagnant gas film [3] Dry, R. J.; Judd, M. R. Fluidised Beds of Fine, Dense
pt emulsion phase: total of all the wet particles Powders: scale-up and Reactor Modelling. Powder
p emulsion phase: wet particle Technol. 1985, 43, 41–53. DOI: 10.1016/S0032-
ref reference state for enthalpy 5910(85)80078-4.
r freeboard region above the fluidized bed [4] Kunii, D.; Levenspiel, O. Fuidization Engineering,
sat saturation 2nd ed.; Butterworth Heinemann: Oxford, UK, 1991.
sp controller setpoint [5] Pal
ancz, B. A Mathematical Model for Continuous
st total of all the particle solids Fluidized Bed Drying. Chem. Eng. Sci. 1983, 38,
s particle solid material 1045–1059. DOI: 10.1016/0009-2509(83)80026-8.
vo vessel outer surface [6] Kannan, C. S.; Rao, S. S.; Varma, Y. B. G. A Kinetic
v vessel Model for Drying of Solids in Batch Fluidized Beds.
18 F. GAGNON ET AL.

Ind. Eng. Chem. Res. 1994, 33, 363–370. DOI: 10. Processes: review and State of the Art. J. Mater. Res.
1021/ie00026a029. Technol. 2015, 4, 208–216. DOI: 10.1016/j.jmrt.2014.
[7] Aly, S. E.; Fathalah, K. A. A Three Phase Model of a 10.018.
Batch Fluidized Bed. Heat Mass Transf. 1999, 34, [23] Villegas, J.; Duncan, S.; Wang, H.; Yang, W.;
405–412. DOI: 10.1007/s002310050. Raghavan, R. Optimal Operating Conditions for a
[8] Ciesielczyk, W. Batch Drying Kinetics in a Two- Batch Fluidised Bed Dryer. IET Control Theory Appl.
Zone Bubbling Fluidized Bed. Drying Technol. 2005, 2010, 4, 294–302. DOI: 10.1049/iet-cta.2009.0039.
23, 1613–1640. DOI: 10.1081/DRT-200065013. [24] Burgschweiger, J.; Groenewold, H.; Hirschmann, C.;
[9] Wang, H.; Dyakowski, T.; Senior, P.; Raghavan, R.; Tsotsas, E. From Hygroscopic Single Particle to Batch
Wang, W. Modelling of a Batch Fluidised Bed Fluidized Bed Drying Kinetics. Can. J. Chem. Eng.
Drying of Pharmaceutical Granules. Chem. Eng. Sci. 1999, 77, 333–341. DOI: 10.1002/cjce.5450770220.
2007, 62, 1524–1535. DOI: 10.1016/j.ces.2006.11.047. [25] Kuo, B. C.; Golnaraghi, M. F. Automatic Control
[10] Li, M.; Duncan, S. Dynamic Model Analysis of Batch Systems, 8th ed.; John Wiley & Sons: Hoboken, NJ, 2003.
Fluidized Bed Dryers. Part. Part. Syst. Charact. 2008, [26] Kumar, A.; Daoutidis, P. Feedback Control of Nonlinear
25, 328–344. DOI: 10.1002/ppsc.200800033. Differential Algebraic Equation Systems. AIChE J. 1995,
[11] Zammouri, A.; Mihoubi, N. B.; Kechaou, N. 41, 619–636. DOI: 10.1002/aic.690410319.
Modeling Batch Fluidized Bed Drying of [27] Groenewold, H.; Tsotsas, E. A New Model for Fluid
Pharmaceutical Powder: Influence of Hydrodynamic Bed Drying. Drying Technol. 1997, 15, 1687–1698.
Regime and Sorption Isotherms. Drying Technol. DOI: 10.1080/07373939708917318.
2019, 1–16. DOI: 10.1080/07373937.2019.1663867. [28] Kloppers, J. C.; Kr€ oger, D. G. The Lewis Factor and
[12] Lai, F. S.; Chen, Y.; Fan, L.-T. Modelling and Its Influence on the Performance Prediction of Wet-
Simulation of a Continuous Fluidized-Bed Dryer. Cooling Towers. Int. J. Therm. Sci. 2005, 44,
Chem. Eng. Sci. 1986, 41, 2419–2430. DOI: 10.1016/ 879–884. DOI: 10.1016/j.ijthermalsci.2005.03.006.
0009-2509(86)85092-8. [29] Broadhurst, T. E.; Becker, H. A. Onset of
[13] Panda, R. C.; Ramachandra IIao, V. S. Dynamic Model Fluidization and Slugging in Beds of Uniform
of a Fluidized Bed Dryer. Drying Technol. 1993, 11, Particles. AIChE J. 1975, 21, 238–247. DOI: 10.1002/
589–602. DOI: 10.1080/07373939308916846. aic.690210204.
[14] Zahed, A. H.; Zhu, J.-X.; Grace, J. R. Modelling and [30] George, S. E. Heat Transfer to Tubes in the
Simulation of Batch and Continuous Fluidized Bed
Freeboard Region of a Fluidized Bed. Ph.D. thesis,
Dryers. Drying Technol. 1995, 13, 1–28. DOI: 10.
McGill University, Montreal, Canada, 1980.
1080/07373939508916940.
[31] Grewal, N.; Saxena, S. Heat Transfer between a
[15] Garnavi, L.; Kasiri, N.; Hashemabadi, S.
Horizontal Tube and a Gas-Solid Fluidized Bed. Int.
Mathematical Modeling of a Continuous Fluidized
J. Heat Mass Transf. 1980, 23, 1505–1519. DOI: 10.
Bed Dryer. Int. Commun. Heat Mass Transf. 2006,
1016/0017-9310(80)90154-4.
33, 666–675. DOI: 10.1016/j.icheatmasstransfer.2006.
[32] Osman, M.; Upadhyay, S.; Saxena, S. Heat Transfer
02.014.
Investigations for in-Bed and Freeboard Vertical U-
[16] Wanjari, A. N.; Thorat, B. N.; Baker, C. G. J.;
Tubes in a Fluidized Bed at Moderate Temperatures.
Mujumdar, A. S. Design and Modeling of Plug Flow
Fluid Bed Dryers. Drying Technol. 2006, 24, Energy 1982, 7, 465–472. DOI: 10.1016/0360-
147–157. DOI: 10.1080/07373930600558946. 5442(82)90056-1.
[17]
Lapointe-Garant,
Gagnon, F.; Desbiens, A.; Poulin, E.; [33] Churchill, S. W.; Chu, H. H. Correlating Equations
P.-P.; Simard, J.-S. Nonlinear Model Predictive Control for Laminar and Turbulent Free Convection from a
of a Batch Fluidized Bed Dryer for Pharmaceutical Vertical Plate. Int. J. Heat Mass Transf. 1975, 18,
Particles. Control Eng. Pract. 2017, 64, 88–101. DOI: 1323–1329. DOI: 10.1016/0017-9310(75)90243-4.
10.1016/j.conengprac.2017.04.009.
[18] Burgschweiger, J.; Tsotsas, E. Experimental
Investigation and Modelling of Continuous Fluidized Appendix A. Single particle model detailed
Bed Drying under Steady-State and Dynamic development
Conditions. Chem. Eng. Sci. 2002, 57, 5021–5038.
DOI: 10.1016/S0009-2509(02)00424-4. The particle total mass mp is:
[19] Cavinato, M.; Franceschinis, E.; Cavallari, S.;
mp ¼ ms þ mw (A1)
Realdon, N.; Santomaso, A. Relationship between
Particle Shape and Some Process Variables in High Dry basis values simplify calculations for modeling. It is
Shear Wet Granulation Using Binders of Different however more common to express moisture contents on a
Viscosity. Chem. Eng. J. 2010, 164, 292–298. DOI: wet basis. The wet basis moisture content of the particle wp
10.1016/j.cej.2010.08.029. can be evaluated with:
[20] Mujumdar, A. S. Handbook of Industrial Drying, 4th mw mw vp
ed.; CRC Press: Boca Raton, FL, 2014. wp ¼ ¼ ¼ (A2)
mp ms þ mw 1 þ v p
[21] Monteith, J. L.; Unsworth, M. H. Principles of
Environmental Physics, 3rd ed.; Academic Press, 2008. Since there is a phase change in the model (water vapor-
[22] Philippsen, C. G.; Vilela, A. C. F.; Zen, L. D. ization), it is important to fix a reference state in the
Fluidized Bed Modeling Applied to the Analysis of enthalpy calculation. Here, the reference state for the water
 DRYING TECHNOLOGY 19

is chosen to be in its liquid state at temperature Tref. The enthalpy on dry basis, i.e., is (per unit of dry air mass mg):
reference point for the other substances of the model (solids
ie ¼ cg ðTe  Tref Þ
and dry air) is their current states of matter at Tref. Thus,
all enthalpies I are related to this reference state. þ ve ½cw ðTwv  Tref Þ þ Kwv þ cwv ðTe  Twv Þ (A9)
Analogous to the mass in (A1), the particle enthalpy Ip
can be calculated by summing its constituting components where cg and cwv are the specific heat capacity of the dry air
enthalpies (extensive property). It is composed of the solid and the water vapor respectively, and Twv, the reference
material enthalpy Is, the liquid water enthalpy Iw and the vaporization temperature. The specific enthalpy of water
bonds between them, or, the sorption enthalpy. As men- vaporization Kwv (or latent heat) is to be evaluated at Twv.
tioned by Burgschweiger and Tsotsas[18] for a highly sorp- The enthalpy path can be simplified by choosing Tref ¼
tive material (i.e., hygroscopic), the sorption enthalpy is still Twv , giving:
significantly smaller than the other enthalpies in the system. ie ¼ cg ðTe  Twv Þ þ ve ½Kwv þ cwv ðTe  Twv Þ
(A10)
Neglecting the latter, with a uniform temperature at Tp, the ¼ ðcg þ ve cwv ÞðTe  Tref Þ þ ve Kwv
particle relative enthalpy is:
Note that Tref is used in the remaining enthalpy calcula-
Ip ¼ Is þ Iw ¼ ðms cs þ mw cw ÞðTp  Tref Þ (A3) tions to ease reading, having in mind that its value at reso-
lution is Twv. Likewise, the specific enthalpy of the particle
where cs and cw are the specific heat capacity of the solid
gas film ipf on dry basis is:
and liquid water respectively. It can also be expressed as a  
particle specific enthalpy ip on dry basis according to: ipf ¼ cg ðTp  Tref Þ þ vpf Kwv þ cwv ðTp  Tref Þ (A11)
Ip
ip ¼ ¼ ðcs þ vp cw ÞðTp  Tref Þ (A4)
ms
To simplify the problem, no particle attrition or deform-
Appendix A.1. Mass balance details
ation is assumed during the drying. Further simplification is Since ms is a constant by postulating no particle attrition,
achieved by neglecting the volume of free moisture. In other the dry basis moisture content allows the substitution:
words, the volume occupied by a wet particle Vp is fixed at
the skeletal volume (including internal pores) of the dried dmw dðmw =ms Þ dvp
¼ ms ¼ ms (A12)
one Vpd. Assuming that all the internal pores are filled with dt dt dt
water at the critical point vpc, they occupy the same volume And, as shown in Figure 1, the steady-state mass balance
as the water at this specific point. That is why the particle around the stagnant gas film volume gives:
volume is approximated with: _ p, pf  m
_ pf , e

 0¼m (A13)
1 vpc
Vp  Vpd ¼ Vs þ Vw jvpc ¼ ms þ (A5) with m _ pf , e as the water flow rate from the gas film to the
qs qw
interstitial gas. Inserting (A12) and (A13) in (6) gives the
where Vs, and qs are the solid material volume and density, particle mass balance expressed with the interstitial gas:
and Vw, and qw are the water volume and density, respect- dvp
ively. The dry particle skeletal density is then: ¼ m
ms _ pf , e (A14)
dt
ms ms qs qw
qpd ¼ ¼  ¼ (A6) By introducing kpf,e as the mass transfer coefficient
Vpd ms 1 þ vpc qs vpc þ qw between the particle film and the interstitial gas on the
qs qw
transfer area, the mass flow rate is:
and, from (A1), (A5), and (A6), the wet particle density is:
_ pf , e ¼ Sp kpf , e qg ðvpf  ve Þ
m (A15)
mp ms þ mw ms
qp ¼  ¼ ð1 þ vp Þ ¼ qpd ð1 þ vp Þ (A7)
Vp Vpd Vpd since the film thin thickness implies Spf  Sp : The driving
force concentrations are typically expressed in kg m3.
These assumptions enable a concise characterization of They are converted to moisture content v here knowing
the particle shape with: that they approximately equal qg v for low v value. Noting
pdp3 pdp2 from (A5) and (A6) that ms ¼ qdp Vpd  qpd Vp , and using
Vp ¼ and Sp ¼ (A8) (A8), the mass balance equation of the particle reduces to:
6 /
pdp3 dvp pdp2
where Vp and Sp are the particle volume and surface area qpd ¼ kpf , e qg ðvpf  ve Þ (A16)
respectively. The sphericity / (also called shape factor) is 6 dt /
included to consider the particle nonspherical shape. For a or
given / value, the particle diameter dp corresponds theoret-
ically to the diameter of a sphere having the same volume dvp 6
qpd ¼ kpf , e qg ðvpf  ve Þ (A17)
as the particle. However, dp can be approximated with the dt /dp
particle size given by a screen analysis.[4]
Similarly to the particle enthalpy in (A3), the interstitial
Appendix A.2. Energy balance details
gas enthalpy can be evaluated by summing the enthalpy of
the dry air and water vapor in it. For calculating the vapor Like in the mass balance, the particle enthalpy Ip can be
enthalpy, the path must start at the selected reference, i.e., conveniently substituted by the dry basis specific enthalpy
the liquid state at Tref. Hence, the interstitial gas specific ip. With ms constant, the accumulation term becomes:
20 F. GAGNON ET AL.

dIp dðIp =ms Þ dip Since they are defined equal, Tref is substituted by Twv in
¼ ms ¼ ms (A18) (A26) to indicate that the vaporization enthalpy Kwv is eval-
dt dt dt
uated at this temperature.
With this substitution, the energy balance (8) becomes:
dip
¼ Q_ pf , p  I_ p, pf
ms (A19) Appendix A.3. Particle transfer coefficients
dt
As shown by the Figure 1, the steady-state energy bal- The mass and heat transfer coefficients between the particle
ance around the stagnant gas film volume gives: and interstitial gas can be predicted with multiple expres-
sions available in the literature. The main difficulties arise
0 ¼ Q_ e, pf  Q_ pf , p þ I_ p, pf  I_ pf , e (A20)
from the fact that both coefficients differ experimentally for
with Q_ e, pf as the heat flow rate from the interstitial gas to the a single particle (true value) or a bed of particles (apparent
gas film and I_ pf , e as the enthalpy flow rate from the gas film value) at low Reynold numbers.[4] The theoretical causes of
to the interstitial gas. Inserting (A20) in (A19), the particle this difference are still debated and not fully understood.[27]
energy balance can be expressed with the interstitial gas: The calculations given here are empirical relations valid for
fluidized particles. It is however important to mention that
dip
ms ¼ Q_ e, pf  I_ pf , e (A21) the FBD model sensitivity to them is exceptionally low,
dt attributable to their high values relative to the phase inter-
Knowing that Tpf ¼ Tp , the heat flow rate provided by change coefficients.
the interstitial gas is: The mass transfer coefficient kpf,e is calculated from the
heat transfer coefficient he,pf by directly applying the Lewis
Q_ e, pf ¼ Sp he, pf ðTe  Tp Þ (A22)
relation for turbulent air-water systems (unit Lewis factor,
with he,pf as the heat transfer coefficient between the inter- see Refs.[20,28]):
stitial gas and the particle gas film on the transfer area he, pf
Spf  Sp : The water mass flows m _ p, pf and m
_ pf , e are assumed kpf , e ¼ (A27)
cg
to be in the vapor state. Using only the water vapor
enthalpy part of the gas film in (A11) and the mass flow with the specific heat capacity of the gas cg assumed con-
rate at (A15), the enthalpy flow rate from the gas film to stant at the dry air value. It is worth mentioning that (A27)
the interstitial gas is: is a simpler but equivalent expression for the evaporation
  coefficient given in Refs.[5,7,10,12,15,16]
I_ pf , e ¼ m_ pf , e Kwv þ cwv ðTp  Tref Þ
  (A23) The heat transfer coefficient he,pf is approximated with
¼ Sp kpf , e qg ðvpf  ve Þ Kwv þ cwv ðTp  Tref Þ the simple Ranz-Marshall correlation for spherical
particles:[4]
where the temperature at the interface between the gas film 8 9
!0:5

and the interstitial gas is supposed to be at Tpf ¼ Tp : kg < dp t0 qg cg lg 1=3 =
Inserting (A22) and (A23) into (A21) gives: he, pf ¼ 2 þ 0:6 (A28)
dp : lg kg ;
dip n
ms ¼ Sp he, pf ðTe  Tp Þ
dt where lg and kg are the dynamic viscosity and thermal con-
 o ductivity of air respectively. The inlet dry air superficial vel-
 kpf , e qg ðvpf  ve Þ Kwv þ cwv ðTp  Tref Þ
ocity t0, defined in (10), is not related to the single particle
(A24) but the batch FBD model.
And replacing ip by (A4) in the accumulation term leads to:

dip d ðcs þ vp cw ÞðTp  Tref Þ Appendix B. Batch model detailed
ms ¼ ms
dt dt development
dTp dvp
¼ ms ðcs þ vp cw Þ þ cw ðTp  Tref Þms Assuming a cylindrical FBD vessel of inner diameter dv, the
dt dt
volume Vf and the cross-sectional area Sf of the whole fluid-
dTp
¼ ms ðcs þ vp cw Þ  cw ðTp  Tref ÞSp kpf , e qg ðvpf  ve Þ ized bed column are:
dt
(A25) pdv2
Vf ¼ Sf zf and Sf ¼ (B1)
4
where the dvp =dt term was replaced by the particle mass
balance described by (A14) and (A15). Inserting (A25) in From Figure 2, the wall contact areas with the emulsion
(A24) and relating ms to the particle shape with (A5)–(A8) gas Sv and freeboard Sr are:
(like in the mass balance), the energy balance of a single Sv ¼ pdv zf and Sr ¼ pdv ðzv  zf Þ (B2)
particle becomes:
and the external vessel surface Svo is:
dTp 6 n
qpd ðcs þ vp cw Þ ¼ he, pf ðTe  Tp Þ Svo ¼ pdvo zv (B3)
dt /dp
 kpf , e qg ðvpf  ve Þ½Kwv (A26) By introducing d as the fluidized bed volume fraction
o that is occupied by the bubble phase, the bubble phase vol-
þ ðcwv  cw ÞðTp  Twv Þ ume Vb and cross-sectional area Sb is:
 DRYING TECHNOLOGY 21

If the inlet gas temperature is uniform at T0, its specific

enthalpy on dry basis i0 can be calculated likewise the emul-
sion gas in (A10):
i0 ¼ ðcg þ v0 cwv ÞðT0  Tref Þ þ v0 Kwv (B10)
Similarly, the outlet gas specific enthalpy iout is:
iout ¼ ðcg þ vout cwv ÞðTout  Tref Þ þ vout Kwv (B11)
with the outlet gas moisture content vout and temperature
Tout. Remembering that vb and Tb is spatially distributed
Figure B1. Stacked phases representation of the fluidized from the plug flow postulate, the bubble specific enthalpy ib
bed column. is also distributed:
ib ¼ ðcg þ vb cwv ÞðTb  Tref Þ þ vb Kwv (B12)
Vb The vessel relative enthalpy Iv can be calculated from the
Vb ¼ dVf and Sb ¼ ¼ dSf (B4)
zf uniform vessel temperature Tv:
Its value is calculated by inserting (10,12), and (13) in Iv ¼ mv cv ðTv  Tref Þ (B13)
the inlet mass balance (11) and simplifying:
where mv and cv are respectively the mass and the specific
t0  tmf heat capacity of the vessel material.
d¼ (B5)
tb
In the K-L model, the absolute bubble velocity tb (or ris- Appendix B.1. Bubble phase
ing velocity) is supposed invariant in height and directly
Appendix B.1.1. Mass balance details
proportional to the excess inlet superficial velocity t0 above
Based on the bubble phase dry air flow definition of (13),
tmf up to a constant:
the water vapor mass flow rate within the bubble phase
tb ¼ ðt0  tmf Þ þ 0:711ðgdb Þ1=2 (B6) m_ b, b is a function of z according to:

where g is the gravitational acceleration, and db, the effective _ b ðzÞ

m
_ b, b ðzÞ ¼
m v ðzÞ ¼ dSf tb qg vb ðzÞ (B14)
single bubble size. The following assumptions only concern 1 þ vb ðzÞ b
the volume fraction of each phases in the bed column. It is _ e, b of the global water
And the infinitesimal portion dm
thus possible to redistribute and stack each phase vertically vapor mass flow rate from the emulsion gas m _ e, b is defined
for the sole purpose of those explanations, as depicted by as:
Figure B1.
With the emulsion at minimum fluidization postulate, _ e, b ¼ Ke, b qg ½ve  vb ðzÞdVb
dm (B15)
the fraction of emulsion phase volume occupied by the where Ke,b is the normalized mass interchange coefficient
interstitial gas, or void fraction, is emf. Its value is estimated between the emulsion and bubble gas. From (B4), the con-
in Appendix B.5. Based on the total volume of all the par- trol volume dVb is in fact:
ticles Vpt and the interstitial gas Ve, Figure B1 reveals that:
dVb ¼ Sb dz ¼ dSf dz (B16)
Ve ¼ emf ðVe þ Vpt Þ ¼ emf Sf zmf (B7)
Inserting (B14)–(B16) in the bubble mass balance (16)
with zmf as the fluidized bed height at the minimum fluid- and simplifying leads to:
ization conditions. Another implication of the emulsion at vb ðz þ dzÞ  vb ðzÞ
minimum fluidization is that the expansion of fluidized bed tb ¼ Ke, b ½ve  vb ðzÞ (B17)
dz
is entirely explained by the varying bubble volume Vb.
Relying on (B7) and (B4), the total void fraction of the or
expanded bed ef is: dvb
Ve þ Vb ¼ Ke, b ½ve  vb ðzÞ
tb (B18)
ef ¼ ¼ ð1  dÞemf þ d (B8) dz
Vf As expected, the normalized coefficient Ke,b makes the
Moreover, it is assumed in Appendix A that the volume bubble volume fraction d disappear from (B18).
of a single wet particle Vp is invariant and equal to its dry
skeletal volume Vpd. Here, this postulate means that the Appendix B.1.2. Energy balance details
particles total volume Vpt is fixed at the one occupied by all Relying on the dry air flow of (13) and the bubble dry basis
the dry skeletons in the batch (neglecting mass losses). enthalpy of (B12), the enthalpy flow within the bubble
Hence, the total mass of all the particle solids mst, or the phase I_ b, b is:
dry weight of the batch, is related to the dry skeletal density
_ b ðzÞ
m
qpd in (A6) according to: I_ b, b ðzÞ ¼ ib ðzÞ ¼ dSf tb qg ib ðzÞ (B19)
1 þ vb ðzÞ
mst  qpd Vpt ¼ qpd ð1  ef ÞVf (B9)
and, by inserting (B19) in the bubble energy balance (20),
The expended bed zf calculation of (14) derives from the two I_ b, b terms of (20) can be substituted by:
(B1) and (B9) assumptions.
22 F. GAGNON ET AL.

I_ b, b ðzÞ  I_ b, b ðz þ dzÞ ¼ dSf tb qg ½ib ðzÞ  ib ðz þ dzÞ _ 0, e ¼

m
_e
m
v ¼ Sf tmf qg v0 (B28)
1 þ ve 0
¼ dSf tb qg dib
(B20) and the vapor flow rate leaving to the freeboard region is:
The infinitesimal outgoing heat flow to the emulsion _e
m
_ e, r ¼
m v ¼ Sf tmf qg ve (B29)
dQ_ b, e is defined as: 1 þ ve e
dQ_ b, e ¼ Hb, e ½Tb ðzÞ  Te dVb (B21) remembering that the emulsion is perfectly mixed, which
implies that the exit gas moisture content is equal to the
with Hb,e being the normalized heat interchange coefficient internal value ve. The water vapor flow rate from all the
between the bubble and emulsion gas, analogously to the particles is supposed directly proportional to the single par-
mass coefficient Kb,e. Similarly to single particle outgoing ticle rate in (A15). Knowing that the single particle rate
enthalpy flow at (A23), the infinitesimal incoming enthalpy m_ pf , e is for a given solid mass ms, and the global rate m _ pt, e ,
flow from the emulsion dI_ e, b is calculated with the vapor for the total mass mst, the proportionality relation is:
mass migration dm _ e, b in (B15):
mst
m_ pt, e ¼ m_ pf , e
dI_ e, b ¼ dm
_ e, b ½Kwv þ cwv ðTe  Tref Þ (B22) ms
by supposing the temperature at the interface between the qpd ð1  ef ÞVf
¼ Sp kpf , e qg ðvpf  ve Þ (B30)
emulsion and bubble phase to be Te. qpd Vp
Inserting (B15), (B16), and (B20)–(B22) in the bubble 6ð1  ef ÞVf
energy balance (20), and rearranging leads to: ¼ kpf , e qg ðvpf  ve Þ
/dp
dib Ke, b
¼ ½v  vb ðzÞ½Kwv þ cwv ðTe  Tref Þ using (A5)–(A8), (B1), and zf of (14) for substitutions. By
dz tb e
inserting (B1), (B4), and (B27)–(B30) in the emulsion mass
Hb, e balance (23), the equation reduces to:
 ½Tb ðzÞ  Te  (B23)
qg tb
tmf 6ð1  ef Þ
ðv  v0 Þ ¼ kpf , e ðvpf  ve Þ  dKe, b ðve  ~v b Þ
Expanding the bubble enthalpy derivative yields: zf e /dp
  
dib d cg þ vb ðzÞcwv ½Tb ðzÞ  Tref  þ vb ðzÞKwv (B31)
¼
dz dz
 dTb  dvb
¼ cg þ cwv vb ðzÞ þ Kwv þ cwv ½Tb ðzÞ  Tref  Appendix B.2.2. Energy balance details
dz dz
Based on the infinitesimal values of (B16), (B21), and (B22),
(B24)
the total heat Q_ b, e and enthalpy I_ e, b exchanges with the
By inserting (B24) in (B23) and substituting the dvb =dz bubble phase are cumulated using:
with the mass balance (B18), the Tref and Kwv terms cancel ð zf ð zf
out. The bubble phase energy balance becomes: Q_ b, e ¼ dQ_ b, e ¼ Hb, e ½Tb ðzÞ  Te dSf dz
0 0 (B32)
  dTb
tb cg þ vb ðzÞcwv ¼ Vb Hb, e ðT ~ b  Te Þ
(dz ) (B25)
Hb, e and
¼ ½Tb ðzÞ  Te  þ Ke, b ½ve  vb ðzÞcwv ð zf
qg
I_ e, b ¼ dI_ e, b
Note that there is still no d coefficient in (B25) because 0 (B33)
of the K-L normalization method. ¼ Vb Ke, b qg ðve  ~v b Þ½Kwv þ cwv ðTe  Tref Þ
~ b in
with (B1), (B4), and the space-average values v~ b and T
(24) and (28) and. From the dry basis enthalpies and dry
Appendix B.2. Emulsion phase
air flow of (A10), (B10), and (12), the enthalpy flow rate
Appendix B.2.1. Mass balance details I_ 0, e coming from the inlet is:
The total moisture exchange with the bubble phase m _ e, b is _e
m
obtained with: I_ 0, e ¼ i0 ¼ Sf tmf qg i0 (B34)
1 þ ve
ð zf ð zf
m_ e, b ¼ _ e, b ¼
dm Ke, b qg ½ve  vb ðzÞdSf dz (B26) and the enthalpy rate I_ e, r leaving to the freeboard region is:
0 0
_e
m
relying on the infinitesimal values of (B15) and (B16). I_ e, r ¼ ie ¼ Sf tmf qg ie (B35)
1 þ ve
Equation (B26) can be rewritten with the space-average
value ~v b of (24): since the emulsion is perfectly mixed. Like for the vapor
mass flow m _ pt, e , the heat Q_ e, pt and enthalpy I_ pt, e flow from
_ e, b ¼ dSf zf Ke, b ðve  ~v b Þ ¼ Vb Ke, b qg ðve  ~v b Þ
m (B27) the particles are directly proportional to the single particle
based on (B1) and (B4). From the interstitial gas dry air flow rates Q_ e, pf and I_ pf , e : Thus, based on the single particle def-
definition of (12), the vapor flow rate from the inlet gas is: initions of (A22) and (A23), the global rates are:
 DRYING TECHNOLOGY 23

mst _ 6ð1  ef ÞVf heat loss to the vessel wall in the freeboard Q_ r, v is calcu-
Q_ e, pt ¼ Q ¼ he, pf ðTe  Tp Þ (B36) lated with (B2) and:
ms e, pf /dp
Q_ r, v ¼ Sr hr, v ðTr  Tv Þ ¼ pdv ðzv  zf Þhr, v ðTout  Tv Þ (B45)
and
mst _ giving hr,v as the heat transfer coefficient between the free-
I_ pt, e ¼ I pf , e board gas and the vessel wall.
ms
By inserting (A10), (B12), (B11), (B35), and (B43)–(B45)
6ð1  ef ÞVf  
¼ kpf , e qg ðvpf  ve Þ Kwv þ cwv ðTp  Tref Þ in the freeboard energy balance (32), and simplifying with
/dp the t0 vout term of (B42), the expression becomes (the Tref
(B37) and Kwv terms vanish):
( )
With the transfer surface of (B2), the heat flow rate to 4ðzv  zf Þhr, v
the vessel wall Q_ e, v is defined as: t0 ðcg þ vout cwv Þ þ Tout
dv qg
Q_ e, v ¼ Sv he, v ðTe  Tv Þ ¼ pdv zf he, v ðTe  Tv Þ (B38)  
¼ dtb cg þ vb ðzf Þcwv Tb ðzf Þ (B46)
where he,v is the heat transfer coefficients from the emulsion 4ðzv  zf Þhr, v
gas to the vessel wall. þ tmf ½cg þ ve cwv Te þ Tv
dv qg
By inserting (A10), (B1), (B10), (B4), and (B32)–(B38) in
the emulsion energy balance (27), and simplifying with the
dKe, b ðve  ~v b Þ term of (B31), the emulsion phase energy
balance becomes (the Tref and Kwv terms vanish): Appendix B.4. Vessel wall
tmf dHb, e ~ Appendix B.4.1. Energy balance details
ðcg þ v0 cwv ÞðTe  T0 Þ ¼ ðT b  Te Þ
zf qg The heat flow rate from the vessel wall to the ambient gas
( ) Q_ v, a on the outer transfer area of (B3) is:
6ð1  ef Þ he, pf
þ kpf , e ðvpf  ve Þcwv þ ðTp  Te Þ
/dp qg Q_ v, a ¼ Svo hv, a ðTv  Ta Þ ¼ pdvo zv hv, a ðTv  Ta Þ (B47)
4he, v with hv,a as the heat transfer coefficient between the vessel
 ðTe  Tv Þ wall and the ambient gas. By inserting the wall enthalpy
dv qg
and the heat flow rates of (B13), (B38), (B45), and (B47) in
(B39) the wall energy balance (34), the equation simplifies to:
Appendix B.3. Freeboard region mv cv dTv
¼ zf he, v ðTe  Tv Þ þ ðzv  zf Þhr, v ðTout  Tv Þ
pdv dt
Appendix B.3.1. Mass balance details
The incoming moisture flow from the bubble phase m _ b, r is dvo
 zv hv, a ðTv  Ta Þ
_ b, b in
equal to the moisture flow within the bubble phase, m dv
(B14), evaluated at the top of the fluidized bed: (B48)
_ b, r ¼ m
m _ b, b ðzf Þ ¼ dSf tb qg vb ðzf Þ (B40)
Appendix B.5. Fluidization regime coefficients
The moisture flow to the gas outlet m _ r, out is calculated
with the freeboard region dry air flow of (15) and: The void fraction at minimum fluidization conditions emf is
estimated with:[29]
_r
m " #0:029

_ r, out ¼
m v ¼ Sf t0 qg vout (B41) l2g
1 þ vr out 0:72
qg 0:021
emf ¼ 0:586/ (B49)
By inserting (B29), (B40), and (B41) in the freeboard qg ðqp  qg Þgdp
3 qp
mass balance (30), the equation reduces to:
relying on the wet particle density qp calculated by (A7).
t0 vout ¼ tmf ve þ dtb vb ðzf Þ (B42) The minimum fluidization superficial velocity tmf is gener-
ally predicted with the Ergun equation, which requires find-
ing the positive root of a quadratic function. Since typical
Appendix B.3.2. Energy balance details pharmaceutical granules are sufficiently small, the root find-
The enthalpy flow rate coming from the bubble phase I_ b, r ing can be avoided by simplifying the Ergun equation, giv-
is the bubble phase enthalpy flow of (B19) evaluated at the ing an explicit method:[4]
top of the fluidized bed:
dp2 ðqp  qg Þg e3mf /2
I_ b, r ¼ I_ b, b ðzf Þ ¼ dSf tb qg ib ðzf Þ (B43) tmf ¼ (B50)
150lg 1  emf
The enthalpy flow rate going to the outlet I_ r, out is given
by the freeboard region dry flow of (15) and:
_r
m Appendix B.6. Phase interchange coefficients
I_ r, out ¼ iout ¼ Sf t0 qg iout (B44)
1 þ vr Even if the particle clouds around the bubbles are neglected,
Knowing that the freeboard temperature Tr equals to the they cannot be ignored at the calculation of the transfer
outlet value Tout from the perfect mixing assumption, the coefficients since they resist exchanges between the bubble
24 F. GAGNON ET AL.

and emulsion phase. From the bubble volume normalization be neglected for small particles.[4] Because of the FBD rapid
concept of Kunii and Levenspiel,[4] the normalized mass mixing, both convection and conduction are lumped into
interchange coefficient between the bubble and emulsion the interstitial gas heat loss model by assuming that Te 
phase Ke,b is calculated with: Tp : The heat transfer coefficient between the emulsion gas

 and the vessel wall he,v is calculated with a correlation from
1 1 1 Grewal and Saxena[31] that incorporates both transfer mech-
K e, b ¼ þ (B51)
Kb, c Kc, e anisms in a single coefficient:
by noting that Ke, b ¼ Kb, e : The normalized mass inter-
 !0:325
kg t0 dv lg
change coefficients from bubble to cloud phase Kb,c and he, v ¼ 47 ð1  ef Þ
from cloud to emulsion phase Kc,e are:[4] dv qs dp3 g
!0:23
(B57)
tmf D0:5 g 0:25 qs cs dv1:5 g 0:5 cg lg 0:3
Kb, c ¼ 4:5 þ 5:85 wv1:25 (B52) 
db db kg kg
and with the fluidized bed void fraction ef provided by (B8).
!0:5 Note that the characteristic length of the Nusselt number is
0:5
emf Dwv g the vessel inner diameter dv and the moisture influence on
Kc, e ¼ 5:71 (B53)
db2:5 the particle density and specific heat is negligible here.
The literature is scarcer for heat transfer with walls in
with Dwv being the molecular diffusion coefficient of water the freeboard. It is however observed that the presence of
vapor in the air. the highly turbulent and elutriated particles cannot be
The normalized heat interchange coefficients can be neglected. This leads to a poor applicability of the classical
analogously predicted from the mass coefficient convection formula for turbulent gas in a cylinder.[30,32]
calculations: The Osman et al.[32] correlation is used to estimate the heat


1 1 1 transfer coefficient from the freeboard region to the vessel
Hb, e ¼ þ (B54) wall hr,v. The particle influence is implicitly included since
Hb, c Hc, e
it was extrapolated from pilot fluidized bed data. By insert-
where Hb,e is the normalized heat interchange coefficients ing a vertical tubular surface in the freeboard with small sil-
between the bubble and emulsion phase. The interchange ica sand, they proposed:
coefficients from bubble to cloud Hb,c and from cloud to
 !0:55

emulsion Hc,e are:[4] kg dv t0 qg cg lg 1=3
hr, v ¼ 3:95 (B58)
dv lg kg
tmf qg cg ðkg qg cg Þ0:5 g 0:25
Hb, c ¼ 4:5 þ 5:85 (B55) The evaluation of the heat transfer coefficient between
db db1:25
the vessel wall and the ambient gas hv,a is achieved with
and expressions from Churchill and Chu.[33] Free convection on
!0:5 a vertical plate is assumed, meaning that the curvature of
emf kg qg cg g 0:5 the cylindrical vessel is neglected here. By assigning the
Hc, e ¼ 5:71 (B56)
db2:5 Nusselt number characteristic length to the vessel total
height zv, the heat transfer coefficient to the ambient gas is:
8  c l 1=6 9 2
>
> >
>
Appendix B.7. Vessel wall transfer coefficients >
kg < 0:387 Gr g g
kg
>
=
hv , a ¼ 0:825 þ
9=16
8=27 >
(B59)
There is a vast amount of literature on heat transfer zv > 
>
> þ
0:492kg >
>
between a fluidized bed and a cooling or heating surface : 1 cg l g ;
(walls), with sometime contradictions between results.[30]
The wall heat transfer phenomena are recognized to be the with the Grashof number Gr evaluated by:
sum of a gaseous convection and a particle conduction gzv3 jTv  Ta jq2g
components (neglecting radiation for low temperature). It is Gr ¼ (B60)
0:5ðTv þ Ta Þl2g
well known that the particle conduction component cannot