Continuous Fluidized Bed Drying: Residence Time Distribution Characterization and Effluent Moisture Content Prediction

Download as pdf or txt
Download as pdf or txt
You are on page 1of 13

Received: 16 May 2019 Revised: 10 December 2019 Accepted: 23 December 2019

DOI: 10.1002/aic.16902

PARTICLE TECHNOLOGY AND FLUIDIZATION

Continuous fluidized bed drying: Residence time distribution


characterization and effluent moisture content prediction

Hao Chen1 | Emily Diep1 | Timothy A. G. Langrish2 | Benjamin J. Glasser1

1
Department of Chemical and Biochemical
Engineering, Rutgers University, Piscataway, Abstract
New Jersey The mixing and drying behavior in a continuous fluidized bed dryer were investigated
2
School of Chemical and Biomolecular
experimentally by characterizing the residence time distribution (RTD) and incorpo-
Engineering, The University of Sydney,
Sydney, New South Wales, Australia rating a micromixing model together with the drying kinetics obtained from batch
drying. The RTD of the dryer was modeled using a tank-in-series model. It was found
Correspondence
Benjamin J. Glasser, Department of Chemical that a high initial material loading and a low material flow rate resulted in a reduced
and Biochemical Engineering, Rutgers
peak height and broaded peak width of the RTD curve. To predict the continuous
University, 98 Brett Road, Piscataway, NJ
08854. dryer effluent moisture content, we combined: (a) the drying kinetics as determined
Email: bglasser@rutgers.edu
in a batch fluidized bed dryer, (b) the RTD model, and (c) micromixing models—
Funding information segregation and maximum mixedness models. It was found that the segregation
National Science Foundation, Grant/Award
model overpredicted the effluent moisture content by up to 5% for the cases we
Number: 1537197
have studied while the maximum mixedness model gave a good prediction of the
effluent moisture content.

KEYWORDS

pharmaceutical continuous manufacturing, fluidized bed drying, residence time distribution,


moisture measurement

1 | I N T RO DU CT I O N ingredients (APIs) for the design of experiments (DOE) during


development to support a quality-by-design filing.3 Continuous
Traditional pharmaceutical oral dosage manufacturing is carried out manufacturing also generally reduces the production time because
in a series of batch unit operations. In the past 10 years, the pharma- there is less downtime as compared with the traditional batch
ceutical industry has been transitioning the manufacturing of a num- processes.4
ber of products from traditional batch processes to continuous Tablets are the most commonly used solid dosage forms in the
processes due to the improved agility, flexibility, and robustness in pharmaceutical industry.5 Direct compression or compaction is often
continuous operation.1 There is a large amount of existing “know- considered the easiest and the most cost-effective way of carrying
how” for batch pharmaceutical processes, and batch manufacturing out continuous manufacturing for oral solid dosage.6 Direct compac-
can offer advantages over continuous manufacturing for products tion involves blending the API with excipients and then directly com-
that are produced in small quantities. However, once large quantities pacting the dry powder blend. However, direct compaction remains a
of a product are required, a continuous process allows for a smaller challenge when dealing with tablets which have high drug dose with
footprint and smaller equipment compared with a batch process. poor compressibility. The literature has reported a number of drug
Advanced process control can also be implemented in continuous products with high-dose and/or poor compressibility7 – paracetamol,8
processes. In continuous manufacturing, the lab scale process can be ibuprofen,9 mefenamic acid,10 acetazolamide,11 metformin,12 and
scaled to commercial size by increasing the production time or run- hydroxyapetite.13 At low dosage, the direct compaction process suf-
ning parallel lines (enabled by smaller footprints), which can result in fers from drug uniformity issues due to poorly flowing and cohesive
a significant reductions in cost compared with batch processes.2 drug substances in the powder blend.14,15 Such issues can be resolved
Continuous manufacturing can save expensive active pharmaceutical by the wet granulation process.

AIChE Journal. 2020;66:e16902. wileyonlinelibrary.com/journal/aic © 2019 American Institute of Chemical Engineers 1 of 13


https://doi.org/10.1002/aic.16902
2 of 13 CHEN ET AL.

The wet granulation process involves the addition of a binder that cannot be tested into products; it should be built-in or should be by
adheres APIs and excipients to one another to form granules.16 As a design.”27 Pramod et al.28 listed various CQAs for drug substances,
result, the flowability and the uniformity of the powder blend are excipients, and drug products. As the purpose of the fluidized bed dry-
improved.17,18 Vervaet et al.19 reviewed different types of granulators ing process is to reduce the moisture level while keeping product uni-
used in the pharmaceutical industry. A typical schematic of the contin- formity, the particle size distribution and the moisture level were of
uous wet granulation process is shown in Figure 1. Fonteyne et al.20 most interest.29 The effect of critical CPPs on the CQAs of the fluidized
studied the ConsiGma 25 twin-screw granulator in a continuous pro- bed has been studied. Leersnyer et al.30 studied the breakage and attri-
cess line. Several process analytical technology (PAT) tools—NIR, tion phenomena during fluidized bed drying and after granulation in a
Raman spectroscopy, and on-line particle sized analyses—were used continuous twin screw granulator. It was discovered that the breakage
to identify the critical process parameters (CPPs) that influence the of the granules occurred within the first seconds of drying. The drying
granule properties. The barrel temperature, the amount of liquid addi- time did not affect the breakage and attrition after a critical moisture
tion, and to a lesser extent, the powder feed rate affected the gran- content was reached. In addition, the air temperature and the air flow
ules' properties. Meng et al.21 compared the CPPs and the granulation did not influence the granule size distribution. Chen et al.31 carried out
mechanisms of a continuous high shear wet granulator and a twin a parametric study of initial moisture, loading, airflow, and air tempera-
screw granulator. A DOE was set up based on three process ture on the effect of drying time for different size fluidized bed dryers.
variables—rotation speed of the shaft/screw, liquid to solid ratio, and It was found that the drying process was more efficient when carried
throughput of the material. The high shear granulator showed a high out at a lower temperature and higher airflow. The thermal design of a
dependence on process variables but produced a smaller size variance batch fluidized bed dryer has been described in a number of research
compared with the twin-screw granulator. papers. Kemp et al.32 summarized a drying kinetic model developed by
While the continuous wet granulation process has been widely Reay and Allen33 and extended by Reay,34 and McKenzie and
studied over the past few years, there is limited literature focused on Bahu.35,36 The model predicts the drying rate as a function of the mois-
the subsequent continuous drying process. By reducing the moisture ture content of the solids and the operating conditions of the dryer
level of the wet granules, several problems of the wet granules related (loading, airflow, and air temperature).
to the flowability of the solids, tablet compaction, microbial control, To ensure the product quality is within the desired region at all
and product stability are resolved.22 Fluidized bed drying is one of the times, a continuous manufacturing process needs to operate under a
most widely used unit operations for drying pharmaceutical wet gran- state of control. The process parameters and quality attributes are
ulations due to several advantages. The high heat and mass transfer designed to be kept within specific ranges. However, during the start-
rate between the gas and solid phases provides mild heating which up, shut-down, and transition between states, the process parameters
prevents potential degradation of heat-sensitive APIs. The excellent and quality attributes will be inevitably changed. Disturbances due to
mixing ensures uniformity of the processed materials.23 the deviation of the process parameters could affect the final product
Although batch fluidized bed drying of pharmaceutical materials quality, leading to product failure. For a continuous process, since the
has been studied in numerous previous works,24-26 continuous fluidized regulatory definition of a batch is not restricted regarding the
bed drying of pharmaceutical materials has been studied to a much manufacturing method, it is possible to rescue the process after failure
lesser extent. Understanding the process is a critical prerequisite for has occurred and reject only part of the product by adequately defin-
continuous manufacturing. A deep understanding of the process would ing a “batch” via a thorough understanding of the process dynamics of
enable one to design, analyze, and control manufacturing through mea- how material flows through the process.
surements of several critical quality attributes (CQAs), which is consis- In chemical engineering processing, the residence time distribu-
tent with the current drug quality system requirement that “quality tion (RTD) is a powerful tool used to characterize unit operations by
providing information on how material travels through a reactor and
thus the overall macromixing of the reactor. While RTD has been tra-
ditionally used for liquid or gas based chemical reactors,37-41 applica-
tion of RTDs has been expanded to solid systems in the polymer and
catalyst industries.42-45 Engisch et al.46 studied the traceability of a
continuous direct compaction system. They obtained the RTD and the
cumulative distribution function F(t) of the direct compaction process
from tracer experiments. A pulse-like addition of a component was
introduced to determine the RTD of the system. With the help of
RTD analysis, the effect of disturbances was quantified, allowing
downstream control to be implemented.
In this study, experiments have been carried out to study the
effect of the CPPs on the RTD of a continuous fluidized bed dryer.
F I G U R E 1 A typical schematic of the continuous wet granulation The effluent moisture content was then predicted based on the drying
process [Color figure can be viewed at wileyonlinelibrary.com] kinetics in a batch fluidized bed and the hydrodynamics of the
CHEN ET AL. 3 of 13

material during the continuous fluidized bed drying process. This work the wall of Segment I, and the exit port is on the wall of segment
provides insights on the continuous fluidized bed process and aims to IV. Small gaps are left between the bottom of the Segments II and III
present a method of characterizing and predicting a continuous fluid- to allow particle movement from one segment to another.
ized bed drying process that can be migrated to other continuous dry- Before the process started, the product chamber was prefilled
ing processeses. with a certain amount of materials (initial loading). The system was
thus first run as a batch process. The materials were then fed using
the KT20 feeder. Upon reaching steady state, the powders were
2 | MATERIALS pushed forward in a counter clock-wise motion at a constant mass
flow rate (Figure 3a, top view) from Segment I–II to III–IV. The rotary
2.1 | Materials valves helped to seal the fluidized bed system and ensure the material
flow was steady. Samples were taken from the effluent stream to
The material used for the study is Fujicalin—a dibasic calcium phosphate measure the moisture content and the tracer concentration. Taking
anhydrous (DCPA) purchased from Fuji Chemical Industries (USA). the samples disturbed the mass flow measurement, which in turn dis-
DCPA is a common excipient in pharmaceutical manufacturing. The par- turbed the control of the mass flow rate. Therefore, with the time
ticle size distribution was characterized using laser diffraction (model LS scale of most our experiments around 60 min (three mean residence
13 320, Beckham Coulter). The DCPA used in this study has a unimodal times), samples were taken every 3 min to ensure enough data points
size distribution. The average diameter of DCPA is 120 μm. The flow- without large fluctuations in the mass flow rate.
ability of the DCPA powder was measured by running shear tests at
3 kPa using the FT4 powder rheometer (Freeman Technology, Tewkes-
bury, UK). Nigrosin water soluble dye was purchased from Sigma- 3 | METHODS
Aldrich (Darmstadt, Germany) to prepare the colored tracer powder.
The ultimate goal of this study is to develop a method to predict the
moisture content of the continuous fluidized bed—one of the CQAs.
2.2 | Description of the Glatt GPCG-2 system

Drying experiments were performed on a Glatt GPCG-2 system (Glatt


Air Techniques Inc., Ramsey, NJ). The continuous line includes a loss-
in-weight feeder (model KT20 from Coperion GmbH), two rotary val-
ves, and a Glatt GPCG-2 fluidized bed dryer. The units are connected
vertically as shown in Figure 2. The GPCG-2 fluidized bed dryer has
an inlet port and an outlet port on the product chamber to allow con-
tinuous operation. Figure 3. shows a top view (Figure 3a) and side
view (Figure 3b) of the product chamber. The product chamber is
divided into four segments as shown in Figure 3a. The inlet port is on

F I G U R E 2 Experimental setup of the continuous fluidized bed F I G U R E 3 Top view (a) and side view (b) of Glatt GPCG-2
drying line [Color figure can be viewed at wileyonlinelibrary.com] continuous fluidized bed system
4 of 13 CHEN ET AL.

another and the parameters values are such that the exponential
decay dominates over the linear increase. However, the linear term
serves to reduce the rate that the moisture content would decrease
with time if there was solely an exponential decay. The parameters a,
k, n, and b are treated as fitting parameters that are used to fit the
experimental data.
F I G U R E 4 Method of obtaining exit conversion for a continuous
fluidized bed dryer

3.2 | Residence time distribution


The GPCG-2 continuous fluidized bed has been characterized by RTD
models and drying kinetic models which have been combined to pre- 3.2.1 | Basics of RTD
dict the effluent moisture content as shown in Figure 4.
The drying process in the continuous fluidized bed has been The RTD reflects the time each portion of material spends in the reac-
treated as a chemical reaction process. The flow and mixing of the tor/process. It can be obtained for a unit operation with a tracer
solid particles has been characterized using the RTD data and the response experiment.50 Traditionally, either a pulse or a step change
RTD model. of tracer is added to the inlet of the unit operation, and the response
of the tracer concentration profile is measured at the exit. Pulse tests
were conducted in this study. During a pulse test, an amount of the
3.1 | Drying kinetics tracer (N0) was suddenly introduced into the inlet. At the exit, the
amount of tracer coming out (ΔN) over a small increment of time Δt
Drying of porous solid particles usually contains three periods: the was measured. The RTD function E(t) has been defined as the rate of
preheating period, the unhindered or constant rate period, and hindered the total percentage of tracer leaving the reactor as a function
or falling rate period. During the preheating period, the heat absorbed of time:
by the particles is mainly used to heat up the particles and the surround-
ing environment. The fluidized bed was warmed up prior to all experi- ΔN=Δt vCðtÞ
EðtÞ = = ð2Þ
N0 N0
ments by flushing heating gas into the empty product chamber to
reduce the preheating period and thus decrease the time for the bed to
reach steady state. During the unhindered period, the amount of heat where v is the mass flow of the material and C is the concentration of
transferred to the solid is equal to the amount of liquid vaporized multi- tracer in the effluent stream. E(t) has units of time−1.
plied by the latent heat of vaporization. During the unhindered period, It is desirable to obtain the the full concentration profile, including
the wet particles keep supplying liquid to the surface, and the drying the entire tail of the curve in order to achieve an accurate RTD. In
front recedes toward the center. Finally, in the hindered period, the sup- practice, where a complete tail may not be achievable, extrapolating
ply of liquid to the surface is insufficient to maintain a sufficiently wet- the tail by fitting an exponential decay may improve the accuracy.
ted surface, and thus the drying rate starts to fall.26,47,48 Integrating the RTD as a function of time gives the cumulative
The following assumption is made: the continuous fluidized bed distribution function F(t):
drying process has the same drying kinetics for a single particle as in a
ðt
batch fluidized bed when the loading, air temperature, and airflow are
F ðtÞ = EðtÞdt ð3Þ
the same. For simplicity, the drying curves obtained from the batch 0

drying experiments were fitted with an empirical model proposed by


Midilli et al.49: The cumulative distribution function represents the fraction of
effluent that has been in the reactor for less than time, t.
CA = a expð − ktn Þ + bt ð1Þ From the RTD, the average time that all tracers spend in the reac-
tor is:
where CA is the moisture concentration of the particles, t is the time,
Ð∞
a is the intial moisture content, k is a pre-exponential factor, and n is tEðtÞdt
tm = Ð0∞ ð4Þ
an exponent that together determine the exponential decay of the 0 E ðtÞdt

moisture content, and b is a constant that determines the linear por-


tion of the model with time. Equation (1) is an empirical model which Instead of comparing the whole curve of several RTDs, variance
has been found to be able to capture the drying behavior of a variety and skewness are usually compared and are defined in the following
of materials. The first term in Equation (1) represents an exponetial equations, respectively:
decay of the moisture content with time from the initial moisture con-
ð∞
tent while the second term represents a linear increase of the mois- σ2 = ðt− tm Þ2 EðtÞdt ð5Þ
ture content with time. These two terms “compete” against one 0
CHEN ET AL. 5 of 13

3.2.3 | Calibration curve development


ð∞
1 3
s3 = ðt− tm Þ EðtÞdt ð6Þ
σ 3=2 0 The amount of tracer was measured using a color spectrometer
(or colorimeter). The colorimeter shines a light on a sample and mea-
A dimensionless number, the Peclet number (Pe), is used to char- sures the reflectance as a function of wavelength. A calibration curve
acterize the continuous fluidized bed behavior. Pe is defined as: was required to correlate the color reflectance and the tracer concen-
tration of the sample. The calibration method we have used in this
Rate of transport by convection UL study was employed by Emady et al.51 Tracer concentrations ranging
Pe = = ð7Þ
Rate of transport by dispersion D
from 0 to 20% (wt%) with 1% intervals were prepared. Data taken
from a single wavelength (410 nm) were plotted as functions of the
where U is the superficial velocity of the particles, L is the charac- concentration of tracer (see Figure 6). The wavelength of 410 nm was
teristic length of the fluidized bed, and D is the dispersion coeffi- chosen because it gave the lowest standard deviation across each set
cient. Pe can be measured experimentally from the following of samples. A fourth order polynomial equation was used to fit the
equation50: calibration curve (see curve labeled “Poly.” in Figure 6).

σ2 2 2  
2
= − 2 1 −e −Pe ð8Þ
tm Pe Pe 4 | RESULTS AND DISCUSSION

To investigate the effect of the CPP on the RTD, a two-level fac- 4.1 | Validation of the tracer experiment
torial design was adopted in this study, as shown in Table 1.
In tracer response tests, it was necessary to ensure that the tracer had
similar properties to the original material. The surface, particle size dis-
3.2.2 | Tracer preparation tribution, and flowability were compared between the tracer material
and the original DCPA material. Figure 7 shows an SEM image of the
In order to carry out a successful pulse response experiment, the DCPA particles before and after nigrosin coating. Both particles
tracer material must be similar to the particles of interest to provide appear to have a spherical shape, and the surface appears similar for
a representative RTD. In this study, the tracer material was prepared
by coating nigrosin onto the DCPA via the top spraying technique in
a Glatt GPCG-1 batch fluidized bed. The nigrosin water solution
was prepared at its maximum water solubility (10 mg/ml). The solu-
tion was sprayed into a fluidized DCPA powder bed at room tem-
perature. After the top spraying, the DCPA was dried in the
fluidized bed to provide dry tracer powders. Figure 5a shows the
DCPA particles before and after coating treatment. The tracer parti-
cles (DCPA + nigrosin) had a black color due to the nigrosin solution.
The tracer powder was then mixed with different amounts of the
original white DCPA powder (see Figure 5b) for calibration curve
development.

TABLE 1 Design of experiment

Initial loading Air flow rate Material flow


Experiment # (kg) (m3/hr) rate (kg/hr)
1 4 90 10
2 4 60 5
3 4 90 5
4 2 60 10
5 2 90 10
6 2 90 5 F I G U R E 5 Tracer powder. (a) dibasic calcium phosphate
anhydrous (DCPA) tracer (left) versus DCPA powder (right) and
7 1 90 5
(b) DCPA powder mixed with different concentrations of tracer [Color
8 2 90 15 figure can be viewed at wileyonlinelibrary.com]
6 of 13 CHEN ET AL.

F I G U R E 9 Shear cell test at 3 kPa before and after coating [Color


figure can be viewed at wileyonlinelibrary.com]

F I G U R E 6 Calibration for tracer [Color figure can be viewed at


to the powder bed under different normal stresses, and the yield
wileyonlinelibrary.com]
stress was measured (see Figure 9). Extrapolating the curve to the y-
axis gives the cohesion strength, which is the shear strength at zero
normal strength. The cohesion strengths of the original DCPA and the
tracers are 0.36 and 0.31 kPa, respectively.
Since most of the tracer experiments were carried out for dry
materials in this study, it was necessary to investigate whether the dry
powders can represent the flowability of our wet powders. Figure 10
shows the RTD curves from otherwise identical experiments with the
exception of the initial moisture content of the powders. The wet
material had 20% moisture content (dry basis weight percent), and the
dry material had about 1% moisture content (dry basis weight per-
cent). The RTD curves showed similar peaks and tails, indicating that
F I G U R E 7 SEM image of dibasic calcium phosphate anhydrous
particle before coating (a) and after coating (b) the tracer tests conducted with dry materials can represent the hydro-
dynamics of the wet materials.

4.2 | Effect of the operating conditions on


residence time distribution

In this section, the effect of several process parameters (loading,


material flow rate, and the air flow rate) will be discussed. The RTD
curves are shown in Figure 11 for all operating conditions listed in
Table 1. At first glance, a single peak followed by an exponential decay
was observed for all curves. A short delay (around 1 min) was
observed before a peak appeared. Table 2 summarizes some impor-
tant criteria of the RTD curves. For an ideal reactor with fixed volume
and material flow rate, the theoretical space time is the ratio of the
volume to the material flow rate. Similarly, for the GPCG-2 fluidized
F I G U R E 8 Particle size distribution of dibasic calcium phosphate
anhydrous before and after coating [Color figure can be viewed at bed, the theoretical space time was defined as the ratio of the loading
wileyonlinelibrary.com] to the material flow rate. It was observed that the experimentally
observed mean residence time was smaller than the theoretical space
both particles. The particle size distribution for the material before time. The effective material flow rate was less than the setpoint, indi-
and after coating was measured and is presented in Figure 8. It can be cating dead-zones in the fluidized bed. For the 4 kg cases, the theoret-
seen that no significant agglomeration or attrition happened during ical space time was almost twice the value of the mean residence
coating. The cohesion strength was examined by carrying out a shear time. For the 2 kg cases, the theoretical space time was closer to the
cell test at 3 kPa. During the shear cell test, vertical stress was applied mean residence time. The higher initial loading in general resulted in
CHEN ET AL. 7 of 13

more stagnant flow or dead space. The large effect of the initial load- 4.2.1 | Effect of initial loading on residence time
ing on the RTD makes it difficult to predict the dryer behavior using distribution
traditional RTD models since one needs to account for the effect of
the initial loading. During the process, the materials experienced two motions—the circu-
lation inside each segment due to the fluidization, and the forward
motion due to the materials being pushed by the inlet stream toward
the outlet. Therefore, instead of filling materials into an empty fluid-
ized bed dryer, the GPCG-2 was prefilled with a certain amount of
materials and operated as a batch unit at the beginning of the process,
and then the inlet stream was introduced to achieve continuous oper-
ation. Such an operating procedure allowed the unit to reach steady
state more rapidly during the startup process. As the inlet and outlet
streams were kept constant throughout the process, it was expected
that once the system reached steady state, the resulting operation
would be independent of the initial loading and there would thus be
one steady state. However, this is not what was observed. Figure 12
shows the steady state RTD curves from experiments, otherwise iden-
tical, with different loadings. It can be seen that the system exhibited
multiple steady states, and the resulting steady state was a function
F I G U R E 1 0 Residence time distribution of dry and wet materials
in GPCG-2 with 2 kg initial loading, air flow rate of 90 m3/hr, and of the initial loading. This result is not surprising considering the com-
material flow rate of 5 kg/hr [Color figure can be viewed at plexity of our fluidized bed system where the initial loading affects
wileyonlinelibrary.com] the extent of fluidization, which in turn affects the flow of solids
through the segments of the fluidized bed. As can be seen in

F I G U R E 1 1 Residence time distribution of all experiments. The


results correspond to 1, 2, and 4 kg initial loadings (first column of
legend); 5, 10, and 15 kg/hr material flowrates (second column of F I G U R E 1 2 Comparison of residence time distribution curve at
legend); and 60 and 90 m3/hr air flowrates (third column of the legend) different initial loadings [Color figure can be viewed at
[Color figure can be viewed at wileyonlinelibrary.com] wileyonlinelibrary.com]

TABLE 2 Residence time distribution results

Experiment # tspace (min) tm (min) σ 2 (min2) s3 (min3) Pe


1 24 13.62 101.80 1.36 2.14
2 48 28.34 426.70 1.19 2.29
3 48 23.52 268.50 0.742 2.70
4 12 10.68 66.89 1.05 1.87
5 12 10.53 66.57 1.39 1.77
6 24 16.53 116.90 1.03 3.31
7 12 13.13 41.33 0.804 7.18
8 8 7.04 20.30 0.835 3.55
8 of 13 CHEN ET AL.

Figure 12, higher initial loading usually results in longer residence with 66.57 min2 (#5). The increase in air flow rate did not substantially
time. As the initial loading increased from 1 (#7) to 2 (#6) and to 4 kg enhance the dispersion transport of the particles, since the Pe
(#3), the resulting RTD curve became broader (the variance increased remained similar (1.87 and 1.77 for Experiments #4 and #5). Similar
from 41.33 to 116.90 and 268.50 min2, respectively), and the peak of results were also found for Experiments #2 and #3 at low solids flow
the curve became higher than the peak of the lower loading condi- rates. The above results indicated that, for the air flow rate we exam-
tions (#7 and #6). As the initial loading increased from 1 (#7) to 2 (#6) ined, the air flow rate did not affect the behavior of the fluidized bed
and to 4 kg (#3), the mean residence time increased from 13.13 to dryer significantly.
16.53 and 23.52 min, respectively. Similar results were found for
Experiments #5 and #1 at a higher level of inlet material flow rate (see
Figure 11). A higher loading generally results in a more CSTR-like 4.2.3 | Effect of material flow rate on residence
behavior, which has a longer tail in the RTD curves (see Figure 11 time distribution
Experiments #1, #2, and #3).
The effect of the material flow rate on the RTD curves is shown in
Figure 14 for a low initial loading (2 kg). As the material flow rate
4.2.2 | Effect of air flow rate on residence time increases from 5 (#6) to 10 (#5) and to 15 kg/hr (#8), the peak height
distribution increased almost proportionally, and the peak width was reduced. The
mean residence time reduced from 16.53 (#6) to 10.53 (#5), and then
The air flow rate is a critical parameter for fluidized bed drying pro- to 7.04 min (#8). The variance reduced from 116.90 (#4) to 66.57 (#5)
cesses. An enhanced drying rate may be achieved by increasing the air and then to 20.30 min2 (#8). The increase in material flow rate slightly
flow rate, thus reducing the drying time. However, since the inlet air enhanced the dispersion transport of the particles overall as the Pe
blows in from the bottom of the fluidized bed and promotes the decreased from 3.31 (#6) to 1.77 (#5) at a lower initial loading and
motion of the particles, a high air flow rate may cause extra back- from 2.7 (#3) to 2.14 (#1) at a higher initial loading. Despite the
mixing, which is generally undesirable in continuous drying, as this change in Pe, the fluidized bed dryer still operates as a fairly well-
causes wet material to mix with more dry material. The Peclet num- mixed system with a large amount of dispersion.
ber, Pe, was used to quantify the mixing of the particles in the fluid-
ized bed. Pe characterizes the ratio of convective transport to the
dispersion transport in the reactor. The reactor behaves like a CSTR 4.3 | Tank in series model fitting
when Pe = 0, where dispersion transport dominates. As the Pe
increases, clearer plug flow behavior will be observed for the reactor. In many chemical processes, the continuous stirred tank reactor
The effect of the air flow rate on the RTD curve is shown in (CSTR) model is used to describe an ideal reactor with perfect mixing.
Figure 13 at a high solids/material flow rate. For an air flow rate However, in reality, not all reactors are CSTRs, and consequently,
increase of 50% from 60 (#4) to 90 m3/hr (#5), only a slight change in many RTD models are developed to allow some deviation from the
the peak height can be observed. The mean residence time was similar ideal reactor. Since the GPCG-2 fluidized bed system has a four-
for both cases—10.68 min (#4) as compared with 10.53 min (#5). Both segment product chamber, a tank-in-series (TIS) model seemed to be
2
RTD curves also had similar variance—66.89 min (#4) as compared a reasonable start for a fitting model to the RTD curves. The TIS

F I G U R E 1 3 Comparison of residence time distribution curves for F I G U R E 1 4 Comparison of residence time distribution curves for
different air flow rates [Color figure can be viewed at different material flow rates [Color figure can be viewed at
wileyonlinelibrary.com] wileyonlinelibrary.com]
CHEN ET AL. 9 of 13

model assumes the reactor to be equivalent to several identical CSTRs A noninteger n value was therefore required to obtained a better
that run in series. When the number of CSTRs approaches infinity, the degree of fitting. While a noninteger value of n is a theoretical repre-
reactor behaves identically to a plug flow reactor (PFR). A generalized sentation, it is useful to consider noninteger values of n to character-
TIS model with a lag time term included is shown in Equation (9). ize the extent of mixing in the fluidized bed as has been done for
chemical reactors and fluidized beds in the past.52,53 A gamma func-
n −1
ðt−tp Þ −
ðt− tp Þ tion was used to approximate the factorial term in Equation (9) to
EðtÞ = e τi ð9Þ
ðn− 1Þ!τi n allow for noninteger values of n and achieve better fitting. R2 ranged
from .935 to .994 for the TIS model fittings. We have also examined
where tp is the lag time, n is the number of CSTRs, and τi is the mean scaling the RTD curves by the mean residence time, and these results
residence time of each CSTR where: are presented in Supporting Information. As can be seen in Figure S1
the scaled curves have a similar peak height and corresponding time
tm for the peak in the RTD. This shows the similarity in behavior of the
τi = ð10Þ
n
fluidized bed for different operating conditions and shows that most
of the differences in behavior can be captured by the mean residence
The RTD experimental data was fitted to the TIS model using the time. In addition, we have examined the RTD results on semi-log axes
OriginPro 2018 (OriginLab) nonlinear fitting tool. The orthogonal dis- (see Supporting Information). On semi-log axes, the tail of a single
tance regression algorithm was used for curve fitting, where the algo- CSTR appears as a straight line so one can clearly see the deviation of
rithm minimizes the orthogonal distance from the data to the fitted the fluidized bed from the behavior of one CSTR (see Figure S2).
curve. Table 3 summarizes the TIS model parameters for the fits of Despite the fact that the GPCG-2 chamber had four segments
the experimental data. While a typical TIS model uses an integer value (compartments), the number of CSTRs was not related to the number
of n, our results showed that the model fittings with integer n values of compartments. Moreover the solids do not move from one com-
did not agree with the experimental data. Figure 15 compares the TIS partment to the next compartment (I–IV) like a series of CSTRs in
model fitting of Experiment #8 with different n. The cases where series. Instead we see that the solids at the entrance clearly have a
n = 1 and 2 failed to indicate when the peak of the tracer appeared. chance to mix and move to Compartment IV (exit) at a very early time.
The number of CSTRs was less than two for all experiments, which
also confirms the conclusion that changing the material flow rate did
TABLE 3 Results of tank-in-series model fitting
not affect the mixing significantly. A lag time ranging from 0.59 to
Experiment # n tm(min) tp(min) R2 1.94 min was found for all tests except for Experiment #7, which had
1 1.23 8.39 1.93 .994 a much higher lag time—we believe this was due to experimental
2 1.14 33.15 1.94 .935 error. The mean residence time from the TIS model fit agreed with the
3 1.25 23.08 1.53 .952 experimentally obtained value. Overall, the TIS model fitting showed a
4 1.15 10.27 0.99 .987 good representation of the RTD curves, and it was therefore used for

5 1.16 9.32 1.60 .960 effluent moisture content prediction.

6 1.89 15.21 0.59 .991


7 1.40 8.91 4.89 .987
4.4 | Micromixing in a continuous fluidized bed
8 1.48 6.48 0.95 .997

During a continuous drying process, the feed stream is broken up due


to fluidization right after entering the fluidized bed. The RTD provides
information on how long the particles stay in the fluidized bed; How-
ever, the mixing state of the feed stream with the materials within the
fluidized bed is unknown. A model describing the micromixing is cru-
cial to understand and to control the continuous fluidized bed drying
process. Zwietering et al.54,55 distinguished two extreme cases
through two models–the segregation model (SM) and the maximum
mixedness model (MMM).
In the SM, the powders are broken up into discrete fragments
characterized by particles with identical moisture contents. Figure 16
depicts a schematic of the SM.54,55 The particle stream flows through
a PFR with outlets following the arrows. The particles will leave
the reactor in a manner such that the RTD of the PFR is identical to
F I G U R E 1 5 TIS model fitting of Experiment #8 with n = 1.00, the real reactor which, in our case, is the GPCG-2 fluidized bed dryer.
1.48, and 2.00 [Color figure can be viewed at wileyonlinelibrary.com] The circles in Figure 16 represent the particle fragments. Each
10 of 13 CHEN ET AL.

coming from the prefilled particles. In a SM, the moisture content of


these two groups of particles can be calculated separately, and the
effluent moisture content would be an average moisture content of
the two groups:

Ðt
F ðtÞ 0 EðtÞX ðtÞdt + ½1 −F ðtÞ X ðtÞ
X effluent ðtÞ = ð13Þ
FðtÞ + ½1 − FðtÞ

While SM gives the least amount of mixing, the MMM assumes


the maximum possible mixing in the reactor. The MMM assumes that
the inflowing material is dispersed on the smallest scale possible,
immediately after being fed into the fluidized bed. MMM can be con-
sidered as a PFR with inlet flow entering from side entrance (see
Figure 17).54 For an arbitary RTD, the inlet stream was fed in such a
F I G U R E 1 6 Conceptual basis for segregation model. Schematic
manner that the RTD of the PFR is identical to the fluidized bed RTD.
of the segregation model (top) and resulting residence time
distribution (bottom) From the side entrance, let λ (life expectancy) be the time the particles
take to leave the fluidized bed. It should be noted that particles with a
life expectancy of 0 are at the right end of the reactor, and particles
fragment dries independently in the fluidized bed and does not mix with a life expectancy of infinity are at the left end of the reactor. In
until the exit of the fluidized bed. Therefore, the moisture content of MMM, the moisture content of the particles in the fluidized bed is an
each fragment is directly correlated to the time spent in the fluidized average of all particles within the fluidized bed. This situation is in
bed dryer. Each fragment has its own moisture content depending on contrast to the SM where all the particles have their own moisture
how long it has stayed in the fluidized bed. Because of the distribution content and do not mix until the exit.
of the residence time, a fragment (E(t)) of the incoming stream is dried Using this approach, the model is developed based on a mass bal-
with a conversion of moisture content X(t). Therefore, the mean con- ance of the moisture contents over the increment between λ and
version of the drying process in the effluent stream will be an average λ + Δλ. At λ, the amount of moisture entering the reactor equals v0C0E
of the conversions of all particles in the exit stream: (λ)Δλ, where v0 is the inlet flow rate. The amount of moisture entering
at λ + Δλ is v0[1 − F(λ)]C|λ + Δλ. The amount of moisture exiting from λ
ð∞
=
X EðtÞX ðtÞdt ð11Þ is v0[1 − F(λ)]C|λ. The moisture removed due to drying is rAv0[1 − F
0
(λ)]Δλ. Carrying out a mass balance on the moisture gives:

where, v 0 ½1 −F ðλÞCj λ + Δλ + v0 C0 EðλÞΔλ− v0 ½1− F ðλÞCjλ + rA v0 ½1 − FðλÞΔλ = 0


ð14Þ
CA ðtÞ
XðtÞ = 1 − ð12Þ
CA0
This equation can then be rearranged to:

and CA(t) can be obtained from drying kinetics as described in dCA EðλÞ
= − rA + ðCA −CA0 Þ ð15Þ
dλ 1 −F ðλÞ
Equation (1).
Equation (11) gives a prediction of the moisture content after
reaching steady state operation. However, the GPCG-2 fluidized bed Equation (15) has been solved numerically using the Runge–Kutta
dryer was first operated as a batch unit, and then the inlet and outlet method.
stream were “turned on” to allow continuous operation. Therefore,
we were also interested in how the system reaches steady state dur-
ing the “start-up” phase after switching from batch to continuous 4.5 | Predicting the effluent moisture content
operation. To obtain the drying curve before steady state operation, a
mass balance was carried out for the effluent stream. As described in the previous section, the continuous drying process
Upon switching from batch to continuous operation, the effluent can, in principle, be predicted by combining the drying kinetics, the
stream consisted of two groups of particles. The first group of parti- RTD, and the micromixing model. In this section, the model has been
cles came from the inlet stream. According to Equation (3), F(t) of the used to try to predict results for operation of the continuous fluidized
inlet stream left the fluidized bed dryer at time t. The second group of bed for 20% initial moisture content, 2 kg initial loading, 90 m3/hr air
particles came from the prefilled particles based on the initial loading flow rate, and 80 C air temperature. The model has been used to pre-
of the fluidized bed. The effluent stream flow rate is equivalent to the dict both a 5 kg/hr material flow rate and a 10 kg/hr material
inlet stream flow rate. Therefore, [1 − F(t)] of the effluent stream was flow rate.
CHEN ET AL. 11 of 13

The drying kinetics were obtained by carrying out a batch experi- fitted using the Midilli et al.49 empirical model. By fitting the experi-
ment using the same operating conditions as in continuous operation. mental data to Equation (1) we obtained parameter values of a = 18.7,
Figure 18 shows the drying curve from a batch experiment conducted b = 0.07, k = 0.06, and n = 1.5. Since a is the intial moisture content it
for 2 kg initial loading with 20% initial moisture, 80 C, and 90 m3/hr is constrained to be a positive number; k is a pre-exponential factor
air flow rate. As discussed previously, the batch drying curve was and it is also constrained to be a positive number (as otherwise the
moisture content will increase with time); n is an exponent that is also
constrained to be positive, and b is a constant that determines the lin-
ear portion of the model with time, which is also constrained to be
positive. The resulting moisture content as a function of time from
Equation (1) with these parameter values is plotted in Figure 18
(orange line). If one examines the rate of drying (not shown), the dry-
ing process remains at a constant drying rate of 1.64%/min until the
moisture content drops to around 4% moisture content.
Using the method discussed in the method section, the drying
curve was predicted and compared with the experimental data from
the outlet of the continuous drying process. Figure 19 shows a com-
parison between the SM prediction, the MMM prediction and the
experimental data for two different material flow rates. The experi-
mental data in Figure 19 showed that effluent moisture content from
the dryer started off at an initial value close to the feed material (20%
moisture content) and decreased nonlinearly until a steady state mois-
F I G U R E 1 7 Conceptual basis for maximum mixedness model.
ture was achieved. The steady state effluent from the dryer was
Schematic of the maximum mixedness model (top) and resulting
residence time distribution (bottom) around 5% for 10 kg/hr (see Figure 19a) and around 1% for 5 kg/hr
(see Figure 19b). The steady state moisture content was higher for
the higher material flow rate due to the smaller residence time of the
material at higher material flow rates. With the higher material flow
rate of 10 kg/hr (see Figure 19a), the SM was able to predict the mois-
ture content until around 10 min but overestimated the steady state
moisture content by about 2.4%. Similar results were found for the
case with the material flow of 5 kg/h (see Figure 19(b)); the SM was
able to predict the moisture content until around 10 min but over-
estimated the equilibrium moisture content by about 3.2%. While the
SM overestimated the equilibrium moisture content, it is observed
that the MMM was able to better capture both the evolution of the
moisture content and the steady state moisture content in both cases.
The SM started to overestimate the moisture content after
10 min because the SM assumes that the mixing occurs at the exit of
the reactor. Therefore, the drying rate solely depended on the age
F I G U R E 1 8 Drying curve obtained from a batch experiment
conducted at 2 kg initial loading with 20% initial moisture, 80 C, and (time spent in the fluidized bed dryer) of each particle. If one examines
90 m3/hr air flow rate [Color figure can be viewed at the batch drying curve (see Figure 18) then one can see that the dry-
wileyonlinelibrary.com] ing rate reaches the hindered drying rate (falling rate period) after

F I G U R E 1 9 Comparison of
SM and MMM prediction of
drying at 2 kg initial loading
with 20% initial moisture, 80 C,
and 90 m3/hr air flow rate with
a material flow of (a) 10 kg/hr
and (b) 5 kg/hr [Color figure can
be viewed at
wileyonlinelibrary.com]
12 of 13 CHEN ET AL.

~10 min. Since the SM assumes that mixing occurs at the exit of the results for additional operating parameters and additional powders. It
reactor, after around 10 min a large number of particles would have is also of interest to see if the approach that we adopted can be used
reached the hindered drying period, and the drying rate would drop. to predict other continuous dryers.
This situation can be observed in Figures 19a and 19b where the
predicted curves changed slope distinctly at 10 min once the drying ACKNOWLEDG MENT
reached the hindered period (see Figure 18 at t = 10 min). On the The authors would like to acknowledge Glatt Air Inc. for offering the
other hand, the MMM assumes that the mixing occurs once the parti- opportunity to use the GPCG-2 continuous fluidized bed system. We
cles enter the reactor. Even though the particles have different ages, would like to thank Subham Rustagi for help with the experiments.
the moisture content at a certain point is based on a mixture of the Some of the material in this paper is based upon work supported by
particles and thus is an average of those particles. The overall drying the National Science Foundation under Grant Number 1537197. Any
rate would still be unhindered even if, in practice, some of the individ- opinions, findings, and conclusions or recommendations expressed in
ual particles had a lower moisture content and fell into the hindered this material are those of the authors and do not necessarily reflect
period. The successful prediction of the effluent moisture content the views of the National Science Foundation.
from the MMM is consistent with the fact that we observed that a
TIS model, with a n value of around 2, succesfully modeled the the OR CID
GPCG-2 continuous fluidized bed dryer. This result indicates that the Hao Chen https://orcid.org/0000-0002-7751-1859
GPCG-2 continuous fluidized bed is close to a well-mixed system, and
it can be approximated by the MMM. RE FE RE NCE S
Continuous chemical reactors have been studied extensively for 1. Oka SS, Sebastian Escotet-Espinoza M, Singh R, et al. Design of an
many decades. Here, we have utilized work that has been developed integrated continuous manufacturing system. In: Continuous
Manufacturing of Pharmaceuticals.2017:405–446.
to predict the exit concentration of continuous chemical reactors to
2. Srai JS, Badman C, Krumme M, Futran M, Johnston C. Future supply
predict the effluent moisture content of a continuous dryer. While chains enabled by continuous processing-opportunities and chal-
continuous dryers have many advantages over batch dryers, one of lenges May 20-21, 2014 continuous manufacturing symposium.
the significant disadvantages is the large amount of powder that is J Pharm Sci. 2015;104(3):840-849.
3. FDA US. Paving the way for personalized medicine: FDA's role in a
needed to do process development in a continuous system. Having a
new era of medical product development. 2013:1-74.
model to predict the moisture content in the continuous dryer is of 4. Schaber SD, Gerogiorgis DI, Ramachandran R, Evans JMB, Barton PI,
critical importance as this model can substantially reduce the amount Trout BL. Economic analysis of integrated continuous and batch phar-
of powder needed in process development. In addition, such a model maceutical manufacturing: a case study. Ind Eng Chem Res. 2011;50
(17):10083-10092.
can be used for process control and process optimization.
5. Gohel MC, Jogani PD. A review of co-processed directly compressible
excipients. J Pharm Pharm Sci. 2005;8(1):76-93.
6. Van Snick B, Holman J, Vanhoorne V, et al. Development of a contin-
5 | C O N CL U S I O N uous direct compression platform for low-dose drug products. Int J
Pharm. 2017;529(1):329-346.
7. Patel S, Kaushal AM, Bansal AK. Compression physics in the formula-
In this work, the GPCG-2 continuous fluidized bed dryer was charac-
tion development of tablets. Crit Rev Ther Drug Carrier Syst. 2006;23
terized by studying the effect of several CPPs on the RTD. The RTD (1):1-65.
was measured by a tracer response experiment. An increase of initial 8. Garekani HA, Ford JL, Rubinstein MH, Rajabi-Siahboomi AR. Forma-
loading reduced the peak height and increased the peak width of the tion and compression characteristics of prismatic polyhedral and thin
plate-like crystals of paracetamol. Int J Pharm. 1999;187(1):77-89.
RTD curve. The air flow rate had an insignificant effect on the RTD
9. Pawar AP, Paradkar AR, Kadam SS, Mahadik KR. Crystallo-co-agglom-
curve. The material flow rate increased the peak height and decreased eration: a novel technique to obtain ibuprofen-paracetamol agglomer-
the peak width of the RTD curve. Despite the effect of the aforemen- ates. AAPS PharmSciTech. 2004;5(3):57-64.
tioned CPPs, the Pe number was low for all the cases studied, indicat- 10. Adam A, Schrimpl L, Schmidt PC. Factors influencing capping and
cracking of Mefenamic acid tablets. Drug Dev Ind Pharm. 2000;26(5):
ing a high degree of dispersion in the continuous fluidized bed.
489-497.
The effluent moisture content was predicted by combining the 11. Di Martino P, Scoppa M, Joiris E, et al. The spray drying of acetazol-
drying kinetics from a batch experiment, a tank-in-series model by amide as method to modify crystal properties and to improve com-
fitting the RTD data, and micromixing models. Two micromixing pression behaviour. Int J Pharm. 2001;213(1):209-221.
12. Gohel MC, Jogani PD. Exploration of melt granulation technique for
models were examined—the SM and MMM. The SM overestimated
the development of coprocessed directly compressible adjuvant con-
the steady state moisture content in the effluent stream because the taining lactose and microcrystalline cellulose. Pharm Dev Technol.
drying rate dropped significantly after a certain amount of particles 2003;8(2):175-185.
fell into the hindered drying period. The MMM gave a good prediction 13. Pontier C, Viana M, Champion E, Bernache-Assollant D, Chulia D.
About the use of stoichiometric hydroxyapatite in compression—
of the effluent moisture content for the dryer that we examined, and
incidence of manufacturing process on compressibility. Eur J Pharm
the conditions that we studied. These results are for a limited set of
Biopharm. 2001;51(3):249-257.
parameters and for one particular dryer and one particular powder. 14. Bi M, Sun CC, Alvarez F, Alvarez-Nunez F. The manufacture of low-
Further work should be carried out to examine the generality of these dose oral solid dosage form to support early clinical studies using an
CHEN ET AL. 13 of 13

automated micro-filing system. AAPS PharmSciTech. 2011;12(1): 37. Adeosun JT, Lawal A. Numerical and experimental studies of mixing
88-95. characteristics in a T-junction microchannel using residence-time dis-
15. Byrn S, Futran M, Thomas H, et al. Achieving continuous manufactur- tribution. Chem Eng Sci. 2009;64(10):2422-2432.
ing for final dosage formation: challenges and how to meet them May 38. Rakoczy R, Kordas M, Story G, Konopacki M. The characterization of
20–21 2014 continuous manufacturing symposium. J Pharm Sci. the residence time distribution in a magnetic mixer by means of the
2015;104(3):792-802. information entropy. Chem Eng Sci. 2014;105:191-197.
16. Dhenge RM, Fyles RS, Cartwright JJ, Doughty DG, Hounslow MJ, 39. Jafarikojour M, Sohrabi M, Royaee SJ, Rezaei M. A new model for res-
Salman AD. Twin screw wet granulation: granule properties. Chem idence time distribution of impinging streams reactors using
Eng J. 2010;164(2):322-329. descending-sized stirred tanks in series. Chem Eng Res Des. 2016;109:
17. Saleh MF, Dhenge RM, Cartwright JJ, Hounslow MJ, Salman AD. 86-96.
Twin screw wet granulation: binder delivery. Int J Pharm. 2015;487 40. Chen L, Hu G-H. Applications of a statistical theory in residence time
(1):124-134. distributions. AIChE J. 1993;39(9):1558-1562.
18. Ebrahimi A, Saffari M, Dehghani F, Langrish T. Incorporation of acet- 41. Dong Z, Zhao S, Zhang Y, Yao C, Yuan Q, Chen G. Mixing and resi-
aminophen as an active pharmaceutical ingredient into porous lac- dence time distribution in ultrasonic microreactors. AIChE J. 2017;63
tose. Int J Pharm. 2016;499(1):217-227. (4):1404-1418.
19. Vervaet C, Remon JP. Continuous granulation in the pharmaceutical 42. Wolf D, White DH. Experimental study of the residence time distribu-
industry. Chem Eng Sci. 2005;60(14):3949-3957. tion in plasticating screw extruders. AIChE J. 1976;22(1):122-131.
20. Fonteyne M, Vercruysse J, Díaz DC, et al. Real-time assessment of 43. Chen L, Pan Z, Hu G-H. Residence time distribution in screw
critical quality attributes of a continuous granulation process. Pharm extruders. AIChE J. 1993;39(9):1455-1464.
Dev Technol. 2013;18(1):85-97. 44. Gao Y, Glasser BJ, Ierapetritou MG, et al. Measurement of residence
21. Meng W, Kotamarthy L, Panikar S, et al. Statistical analysis and com- time distribution in a rotary calciner. AIChE J. 2013;59(11):4068-
parison of a continuous high shear granulator with a twin screw gran- 4076.
ulator: effect of process parameters on critical granule attributes and 45. Zhang C-L, Feng L-F, Hoppe S, Hu G-H. Residence time distribution:
granulation mechanisms. Int J Pharm. 2016;513(1):357-375. an old concept in chemical engineering and a new application in poly-
22. Gerhardt AH. Moisture effects on solid dosage forms formulation, mer processing. AIChE J. 2009;55(1):279-283.
processing and stability. J GXP Comp. 2009;13:58-66. 46. Engisch W, Muzzio F. Using residence time distributions (RTDs) to
23. Chew JW, Hrenya CM. Link between bubbling and segregation pat- address the traceability of raw materials in continuous pharmaceutical
terns in gas-fluidized beds with continuous size distributions. AIChE J. manufacturing. J Pharm Innov. 2016;11:64-81.
2011;57(11):3003-3011. 47. Schlünder EU. Drying of porous material during the constant and the
24. Langrish TAG, Bahu RE, Reay D. Drying kinetics of particles from thin falling rate period: a critical review of existing hypotheses. Drying
layer drying experiments. Chem Eng Res Des. 1991;69(5):417-424. Technol. 2004;22(6):1517-1532.
25. Langrish TAG, Kockel TK. The assessment of a characteristic drying 48. Zhao H, Chen G. Heat and mass transfer in batch fluidized-bed drying
curve for milk powder for use in computational fluid dynamics model- of porous particles. Chem Eng Sci. 2000;55(10):1857-1869.
ling. Chem Eng J. 2001;84(1):69-74. 49. Midilli A, Kucuk H, Yapar Z. A new model for single-layer drying. Dry-
26. Wang HG, Yang WQ, Senior P, Raghavan RS, Duncan SR. Investiga- ing Technol. 2002;20(7):1503-1513.
tion of batch fluidized-bed drying by mathematical modeling, CFD 50. Fogler HS. Elements of chemical reaction engineering. Upper Saddle
simulation and ECT measurement. AIChE J. 2008;54(2):427-444. River, NJ: Prentice-Hall; 1986.
27. FDA US. Guidance for Industry PAT—a framework for innovative 51. Emady HN, Wittman M, Koynov S, et al. A simple color concentration
pharmaceutical development, manufacturing, and quality assurance. measurement technique for powders. Powder Technol. 2015;286:
In: Guidance for Industry. http://www.gmp-compliance.org/guidemgr/ 392-400.
files/PAT-FDA-6419FNL.PDF2004. 52. Stokes RL, Nauman EB. Residence time distribution functions for
28. Pramod K, Tahir M, Charoo N, Ansari S, Ali J. Pharmaceutical product stirred tanks in series. Canad J Chem Eng. 1970;48(6):723-725.
development: a quality by design approach. Int J Pharm Invest. 2016;6 53. Babu MP, Setty YP. Residence time distribution of solids in a fluidized
(3):129-138. bed. Canad J Chem Eng. 2003;81:118-123.
29. Kemp IC. Drying of pharmaceuticals in theory and practice. Drying 54. Zwietering TN. The degree of mixing in continuous flow systems.
Technol. 2017;35(8):918-924. Chem Eng Sci. 1959;11(1):1-15.
30. De Leersnyder F, Vanhoorne V, Bekaert H, et al. Breakage and drying 55. Zwietering TN. A backmixing model describing micromixing in single-
behaviour of granules in a continuous fluid bed dryer: influence of phase continuous-flow systems. Chem Eng Sci. 1984;39(12):1765-
process parameters and wet granule transfer. Eur J Pharm Sci. 2018; 1778.
115:223-232.
31. Chen H, Liu X, Bishop C, Glasser BJ. Fluidized bed drying of a phar-
maceutical powder: a parametric investigation of drying of dibasic cal- SUPPORTING INF ORMATION
cium phosphate. Drying Technol. 2016;35(13):1602-1618. Additional supporting information may be found online in the
32. Kemp IC, Oakley DE. Modelling of particulate drying in theory and
Supporting Information section at the end of this article.
practice. Drying Technol. 2002;20(9):1699-1750.
33. Reay D. Allen RWK the effect of bed temperature on fluid bed batch
drying curves. J Separ Proc Technol. 1982;3(4):11-13.
34. Reay DA, Allen RWK. Predicting the Performance of a Continuous Well- How to cite this article: Chen H, Diep E, Langrish TAG,
mixed Fluid Bed Dryer from Batch Tests. Vol 2. Birmingham: Drying Glasser BJ. Continuous fluidized bed drying: Residence time
Research Ltd.; 1982. distribution characterization and effluent moisture content
35. Bahu RE. Fluidised bed dryer scale-up. Drying Technol. 1994;12(1–2):
prediction. AIChE J. 2020;66:e16902. https://doi.org/10.
329-339.
36. McKenzie K, Bahu RE. Material model for fluidised bed drying. Prague, 1002/aic.16902
Czechoslovakia: Elsevier; 1991.

You might also like