Sei sulla pagina 1di 11

G Model

CEP 6674 No. of Pages 11

Chemical Engineering and Processing xxx (2015) xxx–xxx

Contents lists available at ScienceDirect

Chemical Engineering and Processing:

Process Intensification
journal homepage:

Modelling drying pastes in vibrofluidized bed with inert particles

A.B. Silva Costaa , F. Bentes Freireb , J.T. Freireb , M.C. Ferreirab,*
Chemical Engineering Graduating Program, Federal University of São Carlos, São Carlos, SP, Brazil
Chemical Engineering Department, Federal University of São Carlos, São Carlos, SP, Brazil


Article history: In this paper we analysed the use of a hybrid CSTR/neural network model to describe the highly coupled
Received 30 June 2015 heat and mass transfer during paste drying with inert particles in vibrofluidized beds. The model was
Received in revised form 10 September 2015 developed to predict the air temperature and relative humidity at the outlet of dryer throughout drying
Accepted 17 September 2015
time, as well as the moisture content of powder collected in the cyclone. These variables were chosen as
Available online xxx
they are the most important to monitor the process efficiency and to develop strategies for process
control. Drying was modelled based on the global energy and water mass balances with an inter-phase
coupling term that takes into account both the water evaporation and particle coating, and was described
Grey-box model
Neural network
by a fitted neural network (ANN) model. The proposed model was verified by comparing the predictions
Skimmed milk to experimental data obtained in drying calcium carbonate suspensions with different solids contents,
Calcium carbonate skimmed milk and sewage sludge, at a vibration parameter equals to 4.0 and different amplitudes of
vibration. The results showed that the ANN model was effective to estimate the inter-phase coupling
term. The dynamics behavior of the outlet air temperature and relative humidity were well described by
the CSTR lumped model for most conditions evaluated.
ã 2015 Elsevier B.V. All rights reserved.

1. Introduction can handle problematic pasty materials, which are not suitable for
atomization in spray-dryers. Additionally, their installation costs
According to Gupta and Mujumdar [1], the first reported are low and they are very effective fluid-particle contactors,
commercial application of a vibrofluidized bed dryer (VFBD) was in enhancing heat and mass transfer. Another advantage is that the
drying molybdenum and dates back to 1938. During the 1960–70’s, operating settings of a dryer–amplitude (A) and frequency (f) of
in Poland and Russia, fluidized beds began to be applied to dry vibration, and fluid velocity (U)—may be adapted in order to keep
pasty materials as a more effective alternative to the continuous dynamics stability with a variety of pastes of different composi-
operating belt- and cylinder-type dryers used up to then and tions. Such versatility is a rather uncommon feature for dryers.
numerous patents from that period are referenced by Mrowiec and The mechanisms of drying pastes with inert materials were
Ciesielczyk [2]. The use of vibration emerged as a natural thoroughly described by the authors elsewhere [3]. The stages of
development, since it allowed extending the range of operational layer formation, drying, film fracture, and powder elutriation occur
conditions achieved in conventional fluidized bed dryers, where simultaneously and all of them are affected by the interaction
dynamics were easily deteriorated due to dusting and agglomera- mechanisms between the gas phase and particles in movement. In
tion [2]. The most applied technique for drying pastes in moving vibrated beds, the way the phases are mixed depend on the
beds is based on the atomization of a liquid or paste over solid magnitudes of amplitude and frequency of vibration, and on the
particles. The liquid coats the particles with a layer of material that shape and trajectory of the vibration motion. Heat and mass
is heated by conduction from the particles and by convection from transfer during drying are governed primarily by the dynamics
the fluidizing gas. The water is evaporated and the dry layer of behavior of particles and flow, therefore assessing their dynamics
particles is fractured and produces the powder, which is elutriated is an important step to model any process involving vibration.
and separated by a convenient device. Vibrofluidized dryers are a However, in spite of the fact that extensive research has been
good option for powder production at small or moderate scales conducted in this topic, there is still considerable debate in the
since the association of a fluidizing gas and mechanical vibration literature on how the vibration affects the interaction mechanisms
between particles and flow, even if a liquid phase is not present.
Even in dry vibrofluidized beds, defining fundamental parameters
such as the minimum fluidization velocity or stable pressure drop
* Corresponding author.
E-mail address: (M.C. Ferreira). is controversial, because their characteristic curves do not
0255-2701/ ã 2015 Elsevier B.V. All rights reserved.

Please cite this article in press as: A.B. Silva Costa, et al., Modelling drying pastes in vibrofluidized bed with inert particles, Chem. Eng. Process.
G Model
CEP 6674 No. of Pages 11

2 A.B. Silva Costa et al. / Chemical Engineering and Processing xxx (2015) xxx–xxx

data of drying performed using calcium carbonate suspensions of

different solids contents and skimmed milk.

A Amplitude of vibration (m)

1.1. Background
Cp Specific heat capacity (J kg1 K1)
f frequency of vibration (Hz)
Long before their application in drying, the mechanics of
F Paste flow rate (g s1)
particles in vibrofluidized beds were investigated based on
g Gravitational acceleration (m s2)
theoretical approaches. Bachmann [7] derived a mathematical
H0 Inert particles static height (m)
model to describe the dynamics behavior of particles in a
k Phase-coupling term (g s1)
mechanically vibrated chamber [8]. Kröll [9,10] proposed a
m Mass (g)
simplified model to predict particle movement in a vibrated
P Pressure (Pa)
column [11]. Gallas et al. [12,13] worked on two- and three-
q Rate of heat loss from the dryer (J)
dimensional molecular dynamics simulation of particle segrega-
RH Relative humidity (–)
tion by shaking, for different amplitudes and frequencies of
T Temperature ( C)
vibration. Those researchers were mainly interested in investigat-
t Time (s)
ing segregation mechanisms in granular particles beds and their
U Inlet air velocity (m s1)
models did not consider air flow through the beds. In recent years,
W Air mass flow rate (kg s1)
models for fluidized systems were developed focused mainly on
x Mass fraction of water in the paste (–)
investigating the effect of applying vibrations on the mechanics of
y Mass fraction of water in the air (–)
powder beds of group A and C particles [14–16]. When drying
pastes using inert particles, the particles are generally coarse and
Greek symbols
models developed for powders are not adequate to describe their
D Variation (–) motion. A limited number of studies has been found aimed at
l Latent heat of vaporization (J kg1) describing the dynamics of granular or coarse particles under
’ Particle shape factor (-) vibration, such as in Weir [17] and Xiang et al. [18]. The focus of
G Dimensionless vibration number (–) these studies is mainly in predicting how the vibration waves are
imparted to gas and solid phases. The flow behavior of particles is
simulated by the discrete element method (DEM), using models
0 Initial value (at t = 0)
that require the definition of several interaction parameters. The
exp Experimental
presence of a liquid phase has not been considered by any of these
g Gas
i Inlet
In contrast to the restricted number of studies on theoretical
j Spouted bed
models, there is a considerable amount of experimental research in
mf Minimum fluidization
the literature focusing on the analysis of dynamics of vibroflui-
o Outlet
dized beds of particles, and many of them are aimed at drying
p Paste
applications [1,5,19,20]. A common approach adopted by many
s Inert particles
authors is to compare the dynamics of vibrofluidized beds to the
conventional fluidized ones to identify how the application of
vibration affects the system. As an example, experiments are
necessarily present similar patterns to those observed in carried out to obtain the characteristic curves, in which the bed
conventional fluidized beds [1,4–6]. Therefore, the design of a pressure drop is plotted against increasing values of air velocity. In
VFBD and selection of operating conditions involves a considerable conventional fluidized beds, from the characteristic curve,
amount of empirical expertise. The development of adequate parameters such as Umf and DP at stable operation can be
models to describe the process is an important step to allow determined. However, questions have been raised about applying
further work on evaluating economic feasibility, process control, the same procedure to identify those parameters in VFBDs, because
scale-up, and also to obtain reliable simulation and predictions of there is evidence that vibration may prevent a clear identification
dryer performance without investing significantly in pilot plants of the transition between the fixed and expanded regimes. Based
and equipment. on a review of experimental results, Gupta and Mujumdar [1]
The purposes of this paper are to propose and validate a stated that the mechanism of fluidization in VFBDs is very different
versatile and easily to implement model to be applied to describe from that of conventional ones, with no pressure peak observed on
paste drying in VFBDs. To accomplish these objectives, an overview the onset of fluidization. Strumillo and Pakowski [19] reported on
of published research on the analysis of dynamics behaviour and characteristic curves of very different patterns depending on the
drying in VFBDs will be presented, with the purpose of values of A and f, and properties of inert particles. These results
summarizing the current state of knowledge on this topic. Based have been further corroborated by other researchers who obtained
on this review, information will be collected to fundament the experimental data on VFBD’s, such as Daleffe et al. [5], Zhou et al.
mathematical model. The developed model is based on the [21], Silva-Moris and Rocha [22], Yang et al. [23], Meili et al. [24],
approach known as the grey box model, which combines Meili et al. [25], Nunes et al. [26]. Therefore, is not yet clear how to
fundamental based mechanistic models and empirical models. identify the onset of fluidization in systems with vibration, which
The balance equations of heat and mass transfer are applied to is a clear drawback as one intend to compare results obtained by
describe the drying process, and an empirical equation is fitted to different research groups.
predict the drying kinetic rate for different pastes. The model is The vibration dimensionless number, defined in Eq. (1), is the
able to predict the following variables: outlet air temperature, air most commonly parameter used to characterize the vibrational
relative humidity and powder moisture content, as a function of settings in VFBDs.
time for a broad range of experimental conditions. The verification
Að2pf Þ
of this model is done by comparing the predictions to experimental G¼ ð1Þ

Please cite this article in press as: A.B. Silva Costa, et al., Modelling drying pastes in vibrofluidized bed with inert particles, Chem. Eng. Process.
G Model
CEP 6674 No. of Pages 11

A.B. Silva Costa et al. / Chemical Engineering and Processing xxx (2015) xxx–xxx 3

Physically, G is interpreted as a ratio of vibrational to Daleffe et al. [5] used glycerol to evaluate the dynamics
gravitational forces. In gas fluidized beds, vibrations may be behavior of VFBDs and compared fluidized and vibrofluidized beds
imparted differently to particles depending on the air velocity (U). operating at different values of dp and ’. The vibration number was
At U < Umf, vibration would be the dominant mechanism to set to values of G = 1, 2 and 4, and a particularity of their work was
interact with particles, thus enhancing mixing and breaking-up that these values could be obtained using different combinations of
aggregates. At U > Umf, these effects might be partially accom- A and f. Therefore, they could set a given G by associating either a
plished by the flow itself and the influence of vibration is weaker. high frequency and a low amplitude of vibration or a low frequency
According to Chelenov and Mikhailov [27], in a vibrated bed (with and a high amplitude. The experiments covered a wide range of
no air flow), the movement of particles is governed predominantly conditions and provided useful information to compare the
by vibration at G > 1, which results in expanded particles beds. At dynamics of fluidized and vibrofluidized beds. Bed dynamics
this condition, the bed voidage is high and the pressure drop tends was affected as f or A were varied, even keeping a constant value of
to decrease in comparison to a conventional fluidized bed. At G < 1, G . If a given G was associated with a low frequency and a high
however, the gravitational force prevails and the particles beds amplitude, the bed structure expanded and DP at stable operation
turns to a compacted structure, leading to increased pressure was lower than the observed for a fluidized bed in similar
drops. Some authors reported on similar behavior in fluidized conditions. However, if obtained by combining a high frequency
systems with vibration, such as Gupta and Mujumdar [1], who and a low amplitude of vibration, the opposite effect was observed,
observed that Umf and DP may be reduced by using G > 1. Erdész that is, the bed structure compacted and DP increased. These
and Ormós [28] investigated the behavior of DP on the onset of findings corroborated that the interactions between the amplitude
fluidization for a wide range of G , from 0.2 to 13. They observed an and frequency of the vibration should be investigated in order to
increase in DP at values of G up to 1.5. At G > 1.5 however, DP clarify the effects of the vibration on the bed dynamics. They also
decreased due to the expansion of bed of particles. Malhota et al. suggest that fitted correlations to predict DP and Umf in
[29] reported that beds compacted using G up to 2.0 and expanded vibrofluidized beds should consider both vibration parameters
for higher values of G. instead of only G . Meili et al. [24,25] extended this research
In their simulation of the dynamics of a vibrated bed of steel evaluating drying of different solutions and suspensions. The
spheres, Gallas et al. [12] predicted qualitative differences on the investigation was aimed at identifying the effects of the character-
dynamics behavior of a particles bed using different combinations istics of the paste on the bed dynamics and with that purpose, the
of A and f and a fixed G . This was one of the first studies to report authors measured the variation of air velocity, air temperature and
that G may not be a good parameter to quantify the vibration powder moisture content throughout time, for drying of different
effects in vibrated beds, since different values of A and f might be pastes. The dynamic responses of measured variables, as well as
combined to yield identical values of G . Daleffe et al. [5] carried out the powder production rate, were strongly dependent on the
a number of experimental investigations on the dynamics of characteristics of the paste. Generally, the maximum admissible
vibrofluidized beds and reported on very different dynamic paste feed rate and powder production rate increased as vibration
patterns with identical values of G obtained from different was applied. The authors also noted that the combination of a high
combinations of A and f. The authors observed that at a constant frequency of vibration and a low amplitude was the best way to
value of G , reducing f and increasing A expanded the beds. enhance particles mixing and resulted in higher powder produc-
Therefore, DP decreased and Umf increased tion rates. This study showed clear evidence that, provided that
Due to the difficulty involved in deriving theoretical models to vibration settings are adequately chosen, the VFBDs behave as a
predict Umf and DP, many fitted equations were proposed to well-mixed vessel. Therefore, the process might be well described
overcome this difficulty [28,30–32]. These equations are usually by using a concentrated parameter for variables associated to heat
expressed as a function of Γ , particle properties (rp, dp, ’) and bed and mass transfer, as proposed by Freire et al. [33] for spouted bed
static height (H0). However, they cover a restrict range of possible drying.
operating settings and may produce inconsistent values if Although considerable progress has been reached in describing
extrapolated to different conditions. Another drawback that limits the dynamics of gas and particles in VFBDs using mechanistic
the use of these equations is that in most of them, the parameter G models, they hardly will be adequate for use in drying applications.
is the only one to account for the effects of vibration. This may lead Particularly in process control, it is convenient to use robust and
to misleading results, since there is enough evidence that a given simple models in order to have fast responses for the variables of
value of G may result in very different dynamics conditions if interest. Solutions based on numerical approaches, such as those
different combinations of A and f are used [5,24]. obtained from DEM models do not meet these requirements. On
Different authors have carried out experimental investiga- the other hand, the use of purely empirical models limits the range
tions to identify the effects of the presence of a liquid or slurry on of applicability, do not consider the physical phenomena involved,
the dynamics of VFBDs [5,29,31]. Changes in rheological, and are often oversimplified. The proposal of developing a model
physical, and chemical properties commonly observed during that combines phenomenological and empirical approaches
drying often lead to unstable conditions, so experimental data intends to contribute to fill in this niche and to contribute to
can be obtained for only a narrow range of experimental advances in this area.
conditions. This problem has been overcome by using pure
liquids, such as water or glycerol to keep the beds wet, thus 2. Methodology
simulating to some degree the presence of a paste. Malhota et al.
[29] were the first to investigate dynamics of VFBDs using 2.1. Drying model
glycerol, a substance with high vapor pressure, which does not
evaporate at the usual drying temperatures. The authors tested To describe heat and mass transfer during paste drying in a
different particle diameters (0.353 and 0.667 mm) and vibration vibrofluized bed under transient conditions, a lumped parameter
frequencies, with G from 0 to 4 and varied the mass of glycerol, model was derived from the overall balance equations. The main
corresponding to pore volume saturations (’) of 0.0108 and assumptions adopted in this study are the following: (1) the
0.0216. They showed that the effect of vibration on dynamics vibrofluidized bed behaves as a continuously stirred tank, (2) the
behavior was positive, particularly at the higher saturation, a gas phase behaves as an ideal gas, (3) the solid temperature at the
condition that favored particle agglomeration. bed outlet is equal to the air temperature at the cyclone exhaust,

Please cite this article in press as: A.B. Silva Costa, et al., Modelling drying pastes in vibrofluidized bed with inert particles, Chem. Eng. Process.
G Model
CEP 6674 No. of Pages 11

4 A.B. Silva Costa et al. / Chemical Engineering and Processing xxx (2015) xxx–xxx

(4) the paste retention in the bed is neglected, and (5) the water
diffusion in the paste film coating the inert particles is neglected.
The resulting overall mass balance for the gas phase is given by
Eq. (2), where W is the mass gas flowrate, y is the water mass
fraction in the gas and k is a phase-coupling term. The indexes i and
o refer to the inlet and outlet fluxes.
dyo 1
¼ fW i yi  W o yo þ kg ð2Þ
dt mg

Eq. (3) represents the overall mass balance for the liquid phase,
where F is the paste flow rate and x is the water mass fraction in the
F i xi  k
xo ¼ ð3Þ Fig. 2. Schematic diagram of the multilayer artificial neural network.
The temperature at the exhaust air (Tgo) is estimated from an the available data, which were the moisture at time zero and those
overall heat balance (Eq. (4)), which considers the temperature measured after 600 s. The fitted equation was then used to predict
variation of gas and liquid phases between the dryer inlet and the moisture content in the whole time interval and these values
outlet, the heat consumed to raise the temperature of inert were supplied to the ANN as input data to estimate k. It must be
particles, the latent heat used in water evaporation (l), and the noted that although the choice of an exponential function was
convective heat loss between the chamber wall and the ambient arbitrary, its adequacy has been further checked by comparing the
air, given by the term q. This heat loss was calculated by measuring predicted and experimental values of the air temperature and
the temperatures of the ambient air and dryer wall, and using an moisture content. Both variables have been measured throughout
adequate correlation to estimate the heat transfer coefficient. the whole drying time in intervals of 30 s. The good agreement
dT go W i cpgi T gi þ F i cppi T pi  W o cpgo T go  F o cppo T go  lk  q observed between the predicted and experimental values for those
ð4Þ variables allows to infer that this procedure was effective.
dt mg C pgo þ ms cps þ mj cpj
The design of the neural network was done using the Neural
In Eqs. (2)–(4), k is the phase-coupling term which, along with Network Toolbox available in the MatLab software, which allows a
the energy net rate and the mass efflux from the dryer, describes fast model creation and validation in a single framework. Standard
the simultaneous phenomena of heat exchange, water evaporation Bayesian regularization back propagation training algorithm was
and particle coating with the paste. The model was simulated used for training the network. This training algorithm updated the
according to the flow diagram shown in Fig. 1. weights between adjacent neurons according to Levenberg–
The phase-coupling term for each paste was estimated by an Marquardt algorithm. It minimizes a linear combination of squared
artificial neural network having 5 neurons in the input layer, 1 errors multiplied by weights to produce a network that best fits the
neuron in the output layer, 6 neurons in the hidden layer. The experimental data. The number of neurons in the hidden layer was
inputs were the drying time, the inlet air temperature, the paste chosen by trial and error, as suggested by Himmelblau [34],
feed rate, and the amplitude and frequency of vibration (Fig. 2). starting with 2 neurons and adding up some more until the
Part of experimental data obtained using water, skimmed mill, network performance in estimating the correct output is
calcium carbonate suspensions and sewage sludge was used for satisfactory. A reasonable number of neurons for this application
training the ANN model. From this set of data, 80% was used for was found to be 6. The lumped model f ðÞ was solved numerically
training, 10% for testing and 10% for validation of the neural using 4–5th order Runge–Kutta method implemented in the
network. MatLab code. The initial conditions for water mass fraction in the
The inlet air temperatures were 80 and 100  C and the paste gas phase and air temperature were determined experimentally.
feed rates varied depending on the paste. Experimental values of
powder moisture content measured from samplings collected at 2.2. Equipment and materials
the cyclone exit since drying started up to 800 s were supplied as
input into the neural network to estimate the values of k. A A detailed description of experimental unit and experimental
shortcoming in this procedure is that the powder production is procedures is given elsewhere [24]. Only essential information
incipient up to approximately 500 s and samples could not be about the dryer, pastes and inert particles and a summary of
collected, causing a gap of experimental data in this period. This experimental procedure will be given here. Experimental unit is
difficulty has been overcame by fitting an exponential function to depicted in Fig. 3.
The vibrofluidized bed is composed of a cylindrical glass
chamber 0.5 m high and with a 0.114 m inside diameter, with an
eccentric mechanism used to adjust the amplitude of vibration and
a mechanical controller located at the axle of the electric motor
used to adjust the frequency of vibration. The paste was fed by a
peristaltic pump and uniformly dripped over the particles. Dried
powder was collected at the exit of a Lapple cyclone connected to
the dryer. As inert particles were used glass spheres with a particle
density of 2500(5) kg/m3 and a mean diameter of 2.19  103
(0.05  103) m. The total mass of inert used in every essay was
1.5 (0.1 103) kg, corresponding to a static bed height of 0.09
(0.05  102) m. The evaporation experiments were performed
using distilled water, with air velocities equal to 1.2  Umf and
1.4  Umf, at air temperatures of 80 and 100  C. Drying tests were
Fig. 1. Data flow diagram of the process.

Please cite this article in press as: A.B. Silva Costa, et al., Modelling drying pastes in vibrofluidized bed with inert particles, Chem. Eng. Process.
G Model
CEP 6674 No. of Pages 11

A.B. Silva Costa et al. / Chemical Engineering and Processing xxx (2015) xxx–xxx 5

Fig. 3. Schematic diagram of experimental set-up [25].

initiated by adjusting the parameters for a given experiment, and 3. Results

then keeping the bed fluidized until stabilization under a steady-
state condition. The feeding of a paste (at a temperature of about 3.1. Fluid dynamics behaviour
30  C) was always initiated at a rate of 14 mL/min. The moisture
content of air at the bed entrance was monitored using a According to the review presented in Section 1.1, correlations to
psycrometer linked to the data acquisition system. Along the predict the stable pressure drops in dry beds are very restrict with
process, the dry and wet bulb temperatures were measured at the regards to their range of applicability. Additionally, no model to
bed exit. The absolute humidity and relative humidity were predict the dynamics behaviour in the presence of a paste was
obtained indirectly from these measured temperatures, using a found. The values of minimum fluidization velocity and pressure
computer routine developed in MatLab for the psychometric drops used in modelling were obtained by Meili et al. [24]. The
calculations. The dynamic behavior of particles was visually characteristic curves of pressure drop versus air velocity under
inspected during the experiments to evaluate bed expansion, different amplitudes and frequencies for the dry beds (no paste
and possible agglomeration or channeling. At intervals of 30 s, the added) may be visualized in Fig. 4.
data were collected at a frequency of 512 Hz during 2 s, giving a The curves are typical of fluidization, with a well-defined
total of 1024 measurement points for each sampling. An interface transition between the fixed and fluidized regions at Umf = 1.1 m/s.
was developed for the data acquisition device to provide the In comparison to the conventional fluidized bed (G = 0), the use of
average values and the standard mean deviations of the dry and vibration causes a reduction in the pressure drop, indicating that
wet bulb temperatures. In each experiment, a data file with the bed permeability was improved. It is worth noting that the
extension .xls was generated by this interface. As constant values of vibration amplitude is the parameter with major influence on the
moisture content at the bed exit were observed, indicating a magnitude of the pressure reduction. In spite of the fact that G is
steady-stare operation condition, the paste feed rate was increased the same, equal to 4.0, at A = 0.015 m (and f = 8.12 Hz) the pressure
and the whole procedure repeated until a new steady-state was drop is lower than that observed at A = 0.003 m (f = 8.14 Hz).
reached. The gradual increasing of paste feeding rate continued A condition to successful drying a paste in a VFBD is to assure
until either a saturated moist air was observed at the bed exit or a that the dynamic stability will be preserved in the presence of the
collapse in the bed dynamics caused by liquid surplus into the bed paste, that is, that the pressure drop will either remain stable or
was detected. The dry powder was collected at the cyclone oscillate in a restricted range to keep a steady-state operation. The
underflow and the powder production rate was estimated by ranges of stability will depend on the paste characteristics and feed
weighting the mass collected in 10 min samplings performed along rate. They were verified by Meili et al. [24] in experiments using a
the drying essays. variety of pastes, under air velocities equal to 1.33 m/s and 1.55 m/s

Please cite this article in press as: A.B. Silva Costa, et al., Modelling drying pastes in vibrofluidized bed with inert particles, Chem. Eng. Process.
G Model
CEP 6674 No. of Pages 11

6 A.B. Silva Costa et al. / Chemical Engineering and Processing xxx (2015) xxx–xxx

maximum feed rate is much lower than the value observed for
water, a clear indication that the bed dynamics is affected by the
paste characteristics. This is because the paste composition affects
the inter-particle adhesion forces and agglomeration features.
To evaluate the effect of solids concentration of a paste in the
bed dynamics behaviour, Fig. 7 shows the pressure variation for
calcium carbonate solutions under two different solids concen-
trations, namely 3% (Fig. 7a and b) and 9% (Fig. 7c and d). The
increase in the solids concentration provoked a decrease in the
maximum admissible paste feed rate, from 38 mL/min at 3%
concentrated solution to 30 mL/min, at 9% solution. It is worth
noting that, for a 3% concentrated carbonate solution, stable
operation was observed for the same range of paste feed rate
observed using pure water, which is explained by the low solids
concentration of that suspension. The results were not significantly
affected when the air velocity and temperature were reduced to
1.33 m/s and 80  C. In drying sewage sludge, which has a low solids
content, the results were also similar to water and will not be
showed here.
Fig. 4. Pressure drop versus air velocity in dry beds [24]. It is evident that the bed dynamics and the range of paste feed
rates under stable operation is affected by the paste and its solids
(1.2Umf and 1.4Umf, respectively) and temperatures equal to 80  C concentration. It is not practical to consider these effects in a
and 100  C. phenomenological and comprehensive model. The strategy of
To illustrate the effect of a paste on the bed dynamics, Fig. 5 using a neural network approach is aimed at overcoming this
shows the measured values of ratio DP/DP0 versus time, in water difficulty.
evaporation using air at 100  C and 1.55 m/s, under different
vibration parameters and increasing values of paste feed rates. The 3.2. Verification of the ANN model
data show how the instantaneous pressure (DP) deviates from the
initial value (DP0), measured as the chamber was still dry. As water As described in the Section 2.1, a significant number of data was
is a pure fluid, the bed may be operated at high saturations. Under previously obtained with the purpose of training the ANN model
the range from 10 mL/min up to 42 mL/min, stable dynamics was for estimation of the phase coupling term, k. The performance of
always observed. It is noted that the pressure oscillations are wider this model was verified by comparing the ANN model predictions
at A = 0.015 m, as the vibration imparted to the bed of particles is to the values of k estimated from the drying experiments using
more intense at this condition. In the conditions evaluated, the different pastes. Some results, showing the measured data and the
variation of vibration amplitude or frequency did not affect the predicted values obtained using skimmed milk and a 3% solids
maximum admissible paste feed rate. Generally, the pressure concentration carbonate solution can be seen in Fig. 8.
fluctuations increased as the paste feed rate was raised. An excellent agreement is observed between the experimental
For comparison, Fig. 6 shows the data of dimensionless pressure and the predicted values, suggesting that the ANN model is
obtained for drying skimmed milk, in the same conditions working effectively. It must be noted that the experimental
described in Fig. 4. Unlike water, skimmed milk contains sugar conditions used for training were randomly chosen among the
and dispersed solids, but essentially the variation of pressure drops available data. The predictions obtained from the designed ANN
is very similar to that observed for water. The main difference were always good, regardless of the type of paste, inlet air
concerns to the range of paste feed rates that allowed stable temperature or paste feed rate used in the experiments. On
operation, which varied from 14 mL/min up to 22 mL/min. The average, the deviations between the experimental and predicted

(a) (b)

Fig. 5. Dimensionless pressure drop versus time; water evaporation at T = 100 C and U = 1.55 m/s, (a) A = 0.015 m and f = 8.12 Hz; (b) A = 0.003 m and f = 18.14 Hz.

Please cite this article in press as: A.B. Silva Costa, et al., Modelling drying pastes in vibrofluidized bed with inert particles, Chem. Eng. Process.
G Model
CEP 6674 No. of Pages 11

A.B. Silva Costa et al. / Chemical Engineering and Processing xxx (2015) xxx–xxx 7

(a) (b)
Fig. 6. Dimensionless pressure drops versus time, drying of skimmed milk, T = 100  C and U = 1.55 m/s; (a) A = 0.015m and f = 8.12 Hz and (b) A = 0.003 m and f = 18.14 Hz.

(a) (b)

(c) (d)

Fig. 7. Dimensionless pressure drop versus time, drying of carbonate suspensions, T = 100 C and U = 1.55 m/s; (a) 3% concentration, A = 0.015 m and f = 8.12 Hz; (b) 3%
concentration, A = 0.003 m and f = 18.14 Hz; (c) 9% concentration, A = 0.015 m and f = 8.12 Hz; (d) 9% concentration, A = 0.003 m and f = 18.14 Hz.

Please cite this article in press as: A.B. Silva Costa, et al., Modelling drying pastes in vibrofluidized bed with inert particles, Chem. Eng. Process.
G Model
CEP 6674 No. of Pages 11

8 A.B. Silva Costa et al. / Chemical Engineering and Processing xxx (2015) xxx–xxx

0.65 0.65
14 mL/min exp.
0.60 0.60
18 mL/min exp.
0.55 0.55
22 mL/min exp.
0.50 0.50
Coupling term, k (g/s)


Coupling term, k (g/s)

0.20 14 mL/min exp.
0.15 22 mL/min exp.
0.10 30 mL/min exp.
0.05 38 mL/min exp.
0.00 0.05
0 500 1000 1500 2000 2500 3000 0.00
0 500 1000 1500 2000 2500 3000
Time (s)
Time (s)
Fig. 8. Values of k estimated using the neural network and experimental values in drying of (a) skimmed milk; (b) calcium carbonate suspension with 3% solids concentration,
for different paste feed rates; T = 100  C, U = 1.55 m/s

values were 5%. The results shown in Fig. 8 are representative of air temperature, the outlet air relative humidity, and the powder
other experimental conditions. moisture content.
The adequacy of the designed ANN to estimate the phase A comparison between predicted and estimated values of the
coupling term for operational conditions different from those used outlet air temperature, the air relative humidity and the powder
in training the network was verified by selecting some conditions moisture content over time in the drying of skimmed milk can be
arbitrarily and comparing the model predictions to estimates seen in Fig. 10(a) and (b) respectively.
obtained from experimental data. As an example, values of k In Fig. 10(a), the temperature decreases and air relative
estimated for drying sewage sludge are shown in Fig. 9. humidity rises with increasing time. This is the expected
The good fitting provided by the model observed in Fig. 9 was behaviour, as heat is transferred from the air to the particles
reproduced for other pastes and conditions, confirming that the bed and moisture is evaporated and transferred to the gas phase.
ANN model estimates the phase coupling term adequately. It can Eventually, as the drying proceeds, a steady-state condition is
be seen that the model is able to predict also the values of k for reached and the temperature becomes constant. The temperature
paste feed rates which were not tested in the experiments (dot values were slightly underestimated while the relative humidity
lines). was a little overestimated by the lumped CSTR model, but their
global dynamics behaviour was consistently predicted, and even
3.3. Performance of the CST lumped model the time required for reaching the steady-state was accurately
predicted by the model as well.
The designed ANN was used to predict the phase coupling terms The CSTR model predictions and experimental values of powder
for the range of conditions investigated, and was then provided as moisture may be compared in Fig. 10(b). It can be seen that
input data for the lumped model, according to the diagram shown moisture reduction in the powder is practically instantaneous. This
in Fig. 1. Eqs. (2)–(4) were then simultaneously solved for the outlet is a dynamics behaviour difficult to be consistently predicted by a
model, since it shows a severe reduction in a very short time. It can
be seen in Fig. 10(b) that the model estimations are in excellent
agreement with experimental data in the ranges in which the
0.5 10 mL/min
12 mL/min powder moisture can be measured. The deviations between
14 mL/min predicted and experimental data in Fig. 10(b) were 4% on average.
16 mL/min As the paste feed rate rises, an increase in the deviations
0.4 18 mL/min
between predicted and experimental data of air temperature and
Phase coupling term, k (g/s)

20 mL/min
22 mL/min relative humidity was observed, as can be noted in Fig. 11(a), for a
0.3 24 mL/min
paste feed rate of 18 mL/min. At this feed rate, the model
26 mL/min
predicted performance is still good to predict the global dynamics behaviour
of these variables, but the deviations between predicted and
experimental values in a steady-state condition reached values of
10% for air temperature and 30% for relative moisture. For the
0.1 powder moisture content, however, the predictions continue to be
very good, with deviations of 5% on average. As the paste feed rate
rises to 22 mL/min, which is the maximum admissible value for a
0.0 dryer stable operation, the model did not predict well the global
0 500 1000 1500 2000 2500 3000
behaviour of air temperature and relative humidity (Fig. 12a), but
Time (s) still provided excellent predictions of powder moisture content
(Fig. 12b).
Fig. 9. Values of phase coupling term, k versus time: comparison among
estimations from ANN model and values calculated from experimental data; The increase in the deviations observed as the paste feed rate
A = 15 mm and f = 8.12 Hz, T = 100  C and U = 1.55 m/s. rises may be explained because the increase in bed saturation

Please cite this article in press as: A.B. Silva Costa, et al., Modelling drying pastes in vibrofluidized bed with inert particles, Chem. Eng. Process.
G Model
CEP 6674 No. of Pages 11

A.B. Silva Costa et al. / Chemical Engineering and Processing xxx (2015) xxx–xxx 9

(a) (b)
Fig. 10. Experimental and simulated results in drying of skimmed milk at 14 mL/min (a) outlet air temperature and relative humidity versus time; (b) powder moisture
content versus time. T = 100  C, U = 1.55 m/s; A = 0.003 m; f = 18.2 Hz.

(a) (b)
Fig. 11. Experimental and simulated results in drying of skimmed milk at 18 mL/min (a) outlet air temperature and relative humidity versus time; (b) powder moisture
content versus time. T = 100  C, U = 1.55 m/s; A = 0.003 m; f = 18.2 Hz.

(a) (b)
Fig. 12. Experimental and simulated results in drying of skimmed milk at 22 mL/min (a) outlet air temperature and relative humidity versus time; (b) powder moisture
content versus time. T = 100  C, U = 1.55 m/s; A = 0.003 m; f = 18.2 Hz.

Please cite this article in press as: A.B. Silva Costa, et al., Modelling drying pastes in vibrofluidized bed with inert particles, Chem. Eng. Process.
G Model
CEP 6674 No. of Pages 11

10 A.B. Silva Costa et al. / Chemical Engineering and Processing xxx (2015) xxx–xxx

(a) (b)

Fig. 13. Experimental and simulated results in drying calcium carbonate suspensions at 14 mL/min; A = 3 mm, 80 C and 1.33 m/s. (a) outlet air temperature and relative
humidity versus time at 3% concentration; (b) outlet air temperature and relative humidity versus time at 9% concentration.

results in stronger fluid dynamics instabilities. In these situations, For the same feed rate and suspension with 9% solids concentra-
the assumption that the bed is a well-mixed vessel does not hold, tion, the model did not converge to a consistent solution.
and the model assumptions are no longer consistent with In spite of the fact that the dryer could be operated at paste feed
experimental conditions. rates up to 38 mL/min and 30 mL/min for 3% and 9% concentrated
The results obtained in drying skimmed milk at A = 0.015 m and calcium carbonate suspensions, respectively, the model was not
f = 8.12 Hz were very similar, as well as those obtained at 80  C and able to predict the dynamics behaviour of the air relative moisture
U = 1.33 m/s. From the variables tested, the variation in the paste for feed rates greater than 22 mL/min. In those conditions,
feed rate has a major influence on the performance of the CSTR instabilities were already observed during operation and the
model. model is no longer adequate. At a paste feed rate of 30 mL/min, the
The experimental and estimated values of air temperature and variation of relative moisture throughout time predicted by the
relative humidity as a function of time, for drying calcium model was not suitable, as it did not converge to a constant value as
carbonate suspensions with different solids concentrations are experimentally observed. The results for drying sewage sludge
shown in Fig. 13(a) (3%) and 13(b) (9%), for a paste feed rate of were not presented here because they were very similar to those of
14 mL/min. calcium carbonate suspensions at a solids concentration of 3%.
Comparing the predictions for 14 mL/min (Fig. 13a and b) at It is seen that the CSTR/ANN hybrid model predicts the variation
different concentrations, it is seen that the air temperature of air temperature and relative moisture well at low paste feed
estimates are good in both cases, but the model performance is rates, with pastes of low or moderate solids content, although the
not so good to predict the air relative moisture. In this case, the quality of fitting changes slightly depending on the paste and
deviations between predicted and experimental values increased operation conditions. The predictions of powder moisture were
significantly, as the solids concentration in the paste increased. always good, regardless of the type of paste or feed rate. This is a
Fig. 14 shows the model predictions and experimental results very encouraging result, once the powder moisture content is an
for a paste feed rate of 22 mL/min, solids concentration of 3% and important variable to be monitored in commercial production and
operating conditions similar to those of Fig. 13. The predictions of in control operations.
relative moisture clearly worsen as the paste feed rate increases. The results demonstrate that the modelling strategy adopted in
the present work, based on the introduction of an empirically
estimated phase coupling term, applied as a closure for the mass
and energy conservation balances (Eqs. (2)–(4)), was effective for
modelling paste drying on inert particles in vibrated beds.
Generally, the performance of the model regarding drying in
VFBDs was similar to that obtained in simulations for paste drying
in spouted beds [34]. An advance of the present study is that a
single ANN was used successfully for estimating k for a different
variety of pastes, and then the dynamic behaviour of global
variables might be accurately estimated from the phenomenologi-
cal conservation balances. Using ANN overcomes important
difficulties concerned paste drying modelling, which is the
necessity of obtaining reliable estimates of global drying kinetics.
Because this drying kinetics depends not only on the moisture
evaporation, but also on the film coating and removing mecha-
nisms, it is very difficult to be obtained experimentally. The poor
performances of the model were observed for conditions
associated to hydrodynamic instabilities caused by the increase
in bed saturation. In these situations, the assumption that the bed
Fig. 14. Outlet air temperature and relative humidity versus time; experimental
and simulated results in drying of calcium carbonate suspension at a solids
behaves as a completely agitated vessel does not hold, and the
concentration of 3% and 22 mL/min; A = 3 mm, 80  C and 1.33 m/s. model will no longer be adequate to describe the drying.

Please cite this article in press as: A.B. Silva Costa, et al., Modelling drying pastes in vibrofluidized bed with inert particles, Chem. Eng. Process.
G Model
CEP 6674 No. of Pages 11

A.B. Silva Costa et al. / Chemical Engineering and Processing xxx (2015) xxx–xxx 11

4. Conclusions [14] D. Barletta, P. Russo, M. Poletto, Dynamic response of a vibrated fluidized bed
of fine and cohesive powders, Powder Tecnhol 237 (2013) 276–285.
[15] A. Ajbar, Y. Bahkbakhi, S. Ali, M. Asif, Fluidization of nano-powder: Effect of
The use of a grey box modelling approach was proposed for sound vibration and pre-mixing with group A particles, Powder Tecnhol. 206
modelling paste drying with inert particles in vibrofluidized bed (2011) 327–337.
dryers. The model was verified for a variety of pastes of very [16] S. Limtrakul, W. Rotjanavijit, T. Vatanatham, Lagrangian modeling and
simulation of effect of vibration on cohesive particle movement in a fluidized
different compositions, namely skimmed milk, calcium carbonate bed, Chem. Eng. Sci. 62 (2007) 232–245.
suspensions with different solids concentrations and sewage [17] G.J. Weir, Sound speed and attenuation in dense, non-cohesive air-granular
sludge. The model yielded good estimates of the outlet air systems, Chem. Eng. Sci. 56 (2001) 3699–3717.
[18] L. Xiang, W. Shuyan, L. Huilin, L. Goudong, C. Juhui, L. Yikun, Numerical
temperature, the relative humidity, and the powder moisture simulation of particle motion in vibrated fluidized beds, Powder Tecnhol. 197
content as a function of time, for air inlet temperatures of 80 and (2010) 25–35.
100  C and paste feed rates from 14 to 48 mL/min. Besides being [19] C. Strumillo, Z. Pakowski, Drying of granular products in vibrofluidized beds,
Drying ‘80, Proceedings of 2th International Drying Symposium, vol. 1,
simple in terms of implementation, the model is also flexible as it
Hemisphere Publishing Corporation, New York, Montreal, Canada, 1980, pp.
can be extended for other types of paste simply by including 211–226.
additional data into the ANN training procedure. [20] Z. Pakowski, A.S. Mujumdar, C. Strumillo, Theory and application of vibrated
beds and vibrated fluid beds for drying process, in: A.S. Mujumdar (Ed.),
Advances in Drying, 3, Hemisphere Publishing Corporation, Washington, 1984,
Acknowledgements pp. 245–306.
[21] T. Zhou, H. Kage, S. Funaoka, H. Ogura, Y. Matsuno, Fluidization behavior of
The authors would like to thank the Brazilian funding agencies glass beads under different vibration modules, Adv. Powder Technol. 12 (4)
(2001) 559–575.
CAPES and CNPq for their financial support. [22] V.A. Silva-Moris, S.C.S. Rocha, Vibrofluidized bed drying of adipic acid, Dry.
Technol. 24 (2006) 303–313.
References [23] J. Yang, T. Zhou, L. Song, Agglomerating vibro-fluidization behavior of nano-
particles, Adv. Powder Technol. 20 (2009) 158–163.
[24] L. Meili, R.V. Daleffe, M.C. Ferreira, J.T. Freire, Analysis of the influence of
[1] R. Gupta, A.S. Mujumdar, Aerodynamic and thermal characteristics of vibrated
dimensionless vibration number on the drying of pastes in vibrofluidized beds,
fluid: a review, Montreal, Canada, Hemisphere Pub., New York, Drying ’80, Vol.
Dry. Technol. 28 (2010) 402–411.
1, Proceedings of 2th International Drying Symposium, 11980, pp. 141–150.
[25] L. Meili, L.F.B. Freire, M.C. Ferreira, J.T. Freire, Fluid dynamics of vibrofluidized
[2] M. Mrowiec, W. Ciesielczyk, Fluidized-bed dryers for paste materials, Int.
beds during the transient period of water evaporation and drying of solutions,
Chem. Eng. 17 (2) (1977) 373–379.
Chem. Eng. Technol. 35 (2012) 1803–1809.
[3] J.T. Freire, M.C. Ferreira, F.B. Freire, Drying of solutions, slurries, and pastes, in:
[26] J.F. Nunes, F.C.A. Alcântara, V.A. Silva-Moris, S.C.S. Rocha, Fluid dynamics and
N. Epstein, J.R. Grace (Eds.), Spouted and Spout-fluid Beds—Fundamentals and
coating of sodium bicarbonate in a vibrofluidized bed, Chem. Eng. Process. 52
Applications, Cambridge University Press, New York, 2011, pp. 206–221.
(2012) 34–40.
[4] R.V. Daleffe, J.T. Freire, Analysis of the fluid dynamic behavior of fluidized and
[27] V.A. Chlenov, N.V. Mikhailov, Some properties of a fluidized bed under
vibro-fluidized bed containing glycerol, Braz. J. Chem. Eng. 21 (1) (2004) 35–
vibration, Inzhenerno Fizicheskii Zhurnal Minski 9 (2) (1965) 196–200 (in
[5] R.V. Daleffe, M.C. Ferreira, J.T. Freire, Drying of paste in vibro-fluidized beds:
[28] K. Erdész, Z. Ormós, Drying of past-like materials in vibrofluidized bed of inert
effects of the amplitude and frequency of vibration, Dry. Technol. 23 (9-11)
packing, Drying ‘86, Proceeding of 5th International Drying Symposium (IDS’
(2005) 1765–1781.
86), 1, Hemisphere Publishing Corporation, Boston, USA, New York, 1986, pp.
[6] R.V. Daleffe, M.C. Ferreira, J.T. Freire, Effects of binary particle size distribution
on the fluid dynamic behavior of fluidized, vibrated and vibrofluidized beds,
[29] K. Malhotra, L.L.K. Cheong, A.S. Mujumdar, Pressure drop characteristics for
Braz. J. Chem. Eng. 25 (2008) 83–94.
vibrated beds of dry and sticky particles, Powder Technol. 39 (1) (1984) 101–
[7] D. Bachmann, Verfahrenstechnik. Zeitschrift, 2, Verein Deustscher Ingenieure
Beiheft, Dusseldorf, Germany, 1940 43.
[30] E. Bratu, G.I. Jinescu, Effect of vertical vibrations on pressure drop in a fluidized
[8] R.G. Gutman, Vibrated beds of powders, Part I: a theoretical model for the
layer, Br. Chem. Eng. 17 (1–2) (1971) 49–56.
vibrated bed, Trans. Chem. Eng. 54 (3) (1976) 174–183.
[31] V.A. Chevilenko, V.I. Mushtaev, B.M. Korotkov, G.L. Izhorin, V.A. Lyalin, A.N.
[9] W. Kroll, Uber das verhalten von schuttgut in lotrecht schwingenden gefassen.
Planovsky, Study of hydrodynamics and structure of vibro pseudo-fluidized
Forschung Auf Dem Gebiete Des Ingenieurwesens, Forschung Im
bed in drying of dispersed materials, Khim. Promst. (in Russian) (1979) 356–
Ingenieurwesen 20 (2) (1954) 2–15 Düsseldorf, Germany,.
360 Moscou, n. 6.
[10] W. Kroll, Fliesserscheinungen na haufwerken in schwingenden gafassen,
[32] R. Gupta, A.S. Mujumdar, Aerodynamic of a vibrated fluid bed, Can. J. Chem.
Chemie Ingeneieur Technik 27 (1) (1955) 33–38 Weinheim, Germany,.
Eng. Ottawa 58 (3) (1980) 332–338.
[11] B. Thomas, Y.A. Lui, M.O. Mason, A.M. Squires, Vibrated beds: new tools for
[33] J.T. Freire, F.B. Freire, M.C. Ferreira, B.S. Nascimento, A hybrid lumped
heat transfer, Chem. Eng. Prog. 84 (6) (1988) 65–75.
parameter/neural network model for spouted bed drying of pastes with inert
[12] J.A.C. Gallas, H.J. Herrmann, S. Sokolowski, Molecular dynamics simulation of
particles, Dry. Technol. 30 (2012) 1342–1353.
powder fluidization in two dimensions, Phys. A. 189 (1992) 437–446.
[34] D.M. Himmelblau, Accounts of experiences in the application of artificial
[13] J.A.C. Gallas, H.J. Herrmann, T. Poschel, S. Sokolowski, Molecular dynamics
neural networks in chemical engineering, Ind. Eng. Chem. Res. 47 (2008)
simulation of size segregation in three dimensions, J. Stat. Phys. 82 (1996) 443–

Please cite this article in press as: A.B. Silva Costa, et al., Modelling drying pastes in vibrofluidized bed with inert particles, Chem. Eng. Process.