Chemical Engineering Science 60 (2005) 5151 5163

www.elsevier.com/locate/ces

New methodology for scaling hydrodynamic data from a 2D-uidized bed

Javier Villa Briongosa , Jess Guardiolab,
a Universidad Complutense de Madrid, Facultad de Ciencias Qumicas, Departamento de Ingeniera Qumica, Avda. Complutense S/N,

28040 Madrid, Spain

b Universidad de Alcal, Facultad de Qumica, Departamento de Ingeniera Qumica, Ctra. Madrid-Barcelona, Km 33,6 (Campus),

28871-Alcal de Henares (Madrid), Spain

Received 22 June 2004; received in revised form 14 March 2005; accepted 11 April 2005
Available online 6 June 2005

Abstract
We present a new method of scaling hydrodynamic data obtained from a 2D gassolid uidized bed, establishing links between 2D and
3D geometries. The methodology proposed may also be useful for verifying 3D3D dynamic scaling. According to the chaos scale-up
methodology proposed by Van den Bleek and Schouten [1993. Chemical Engineering Science 48, 23672373], the information balance is
taken into account. The complexity shown by these systems is measured as the Kolmogorov entropy. Fluidized beds of different geometry
(2D, 3D) performing under the bubbling regime were operated at different bed height and bed aspect ratios by uidizing several particle
groups belonging to Geldart groups B and D. Finally, an empirical correlation is proposed to compare uidized bed hydrodynamics. This
correlation relates the global information ow of the system KG = KF r r to the relative Reynolds particle number.
2005 Elsevier Ltd. All rights reserved.
Keywords: Fluidization; Scale-up; Multiphase ow; Non-linear dynamics; 2D dynamic scaling; Free surface uctuation

1. Introduction
Gassolid uidized bed systems are widely used for industrial processes. They can be used as reactors in processes
such as the catalytic production of vinyl acetate, polymerization of olens, etc., and may also serve a physical function
in drying and coating particles. However, the complexity
of their hydrodynamics makes the scale-up of these systems difcult, to the extent that many research efforts have
focused on scaling and modeling gassolid uidized beds
(Glicksman, 1984; Matsen, 1996; Moses, 1996; Glicksman
et al., 1994; Glicksman and Yule, 1995; Levenspiel, 2002).
Moreover, since the non-linear hydrodynamic behavior of
uidized beds may be of a chaotic nature, in this past decade
or so, some authors (Van den Bleek and Schouten, 1993;
Schouten et al., 1996, 1999) have tried to develop a chaos
scale-up methodology. This type of method needs to take

Corresponding author.

E-mail address: [email protected] (J. Guardiola).

0009-2509/$ - see front matter 2005 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ces.2005.04.051

into account the information balance. Namely, the rate of

information lost by the system, expressed as the Kolmogorov
entropy, must be kept constant while scaling-up uidized
bed hydrodynamics if we are to maintain dynamic parity.
Traditionally, 2D uidized bed systems have been used to
obtain hydrodynamic information, such as bubble size and
velocity distribution (Clift and Grace, 1971; Agarwal, 1985,
1987; Lim et al., 1990), to design models aimed at scaling
this information to 3D uidized beds. However, the inuence of the bed walls and geometry on the ow pattern, and
the different coalescence effects in 2D and in 3D beds (Clift,
1986) means there is a need for a suitable criterion for the
appropriate scaling of dynamic data derived from 2D bed to
3D systems. Moreover, to the authors best knowledge, the
literature lacks criteria for scaling uidized bed hydrodynamics from 2D to 3D systems (Geldart, 1970). This study
was thus designed to develop a 2D hydrodynamic data scaling methodology aimed at establishing a criterion to appropriately apply these data to 3D beds.
We used two 2D uidized beds of widths 1 and 2 cm,
and two 3D uidized beds of internal diameters 5.38 and

5152

J. Villa Briongos, J. Guardiola / Chemical Engineering Science 60 (2005) 5151 5163

8.38 cm, to establish hydrodynamic similarities between the

systems. Operating at different bed heights (2D systems) and
bed aspect ratios (3D systems) within the relative velocity
range 1.3 < Ur < 2.7 for a set of Geldart B and D particles,
several time series were collected from free bed surface uctuations to extract hydrodynamic information. Finally, the
information balance was taken into account according to the
chaos scale-up method.

However, when dealing with 2D and 3D uidized beds,

the condition of geometric similarity does not exist such that
the scaling rules derived from the similarity theory cannot
be directly applied. Thus, for information balance, a geometrically independent information ow needs to be dened to
compare the dynamics of the systems, since the ow pattern
is strongly inuenced by column design (shape, size, etc.).

2. Similarity theory

3. Validating uidized bed dynamic scaling

In chemical engineering, small-scale facilities are often

used to predict the behavior of large-scale installations. This
provides maximum knowledge on a system with a minimum
investment. However, the complexity of uidized bed hydrodynamics attributable to the multiphase nature of ow,
makes it difcult to scale-up uidized bed systems, a process frequently ruled by empirical correlations.
During the past few decades, the similarity theory
(Glicksman, 1984) has been widely applied to uidized bed
scale-up (Fitzgerald et al., 1984; Fitzgerald, 1985; Schouten
et al., 1996; Kehlenbeck et al., 2001). According to this
theory, several dimensionless parameters derived from the
governing equation of dynamics should be kept constant
between model and prototype. These dimensionless relations (Table 1) are known as Glicksman scaling rules
(Glicksman et al., 1994) and are usually applied to achieve
dynamic similarity between the scaled beds. As shown in
Table 1, geometric similarityH /D,
, etc.is often preserved to further attain similar dynamicsRe, F r, Gr,
etc.However, because of the complexity of uidized bed
hydrodynamics, the scaling rules sometimes fail. In such
cases, geometric similarity does not ensure the dynamic
similarity needed to preserve, for instance, heat and conversion similarities. Consequently, new balances have to be
taken into account during the scale-up process.
After the chaos scale-up methodology proposed by Van
den Bleek and Schouten (1993), the need to consider the
information generation rate between two scaled uidized
beds using the information group, Kd p /U0 was pointed
out. Thus, to achieve dynamic similarity between the scaled
beds, information ow on a normalized time scale Kd p /U0
should be kept constant during the scale-up process.

Most techniques used to characterize uidized bed

hydrodynamics are based on numerical analysis of signals collected at several points in the bed using different measurement methods such as capacitance or pressure probes (Werther, 1999). However, the measures
provided by these techniques are highly inuenced by
the local ow pattern around the probe, which is assumed to represent the hydrodynamics of the uidized
bed.
In contrast, the measurement technique used here
(Villa-Briongos and Guardiola, 2003) assumes that the
measure of free bed surface uctuations represents the hydrodynamics of the uidized bed. Bubble motion inside the
bed and bubbles escaping from the top cause uctuations
in the bed surface (Baskakov et al., 1986; Yrquez-Ramirez
and Duursma, 2000). Thus, the dynamics recorded at this
single position in the bed represents the global status
of the uidized system. The method involves measuring
the uctuations produced as a consequence of bed top
particlelaser interactions in a laser sheet that crosses the
bed. The method has the advantage of being non-intrusive as
well as providing a global picture of the systems dynamic
behavior.
To gain insight into uidized bed hydrodynamics, several authors (Skrzycke et al., 1993; Karamavru and Clark,
1997; Marzocchella et al., 1997; Johnsson et al., 2000) advocate the integration of spectral, state space and time domain analysis to characterize bed dynamics. Herein, we use
all these methods in a complementary manner to compare
dynamics among different beds in terms of frequency histograms, normalized power spectral densities (Shrikant and
Klizing, 1993) and the information ow of the system.

Table 1
Scaling parameters
Dimensionless
parameters

Author

f s dp3 f U0 dp
 H
U0
,
,
, s,
,
, etc.
f
2f
(gd p )0.5 f D

Glicksman et al. (1994)

C1 =

Horio et al. (1986)

U0 Umf
(gD)0.5

, C2 =

Umf

(gD)0.5
(U0 Umf )dp f
U0 Umf

KG = K F r r m = K
m, Rer =
0.5
f
(gd p )

Villa & Guardiola similarity hypothesis

J. Villa Briongos, J. Guardiola / Chemical Engineering Science 60 (2005) 5151 5163

100

5153

60
2

40

1
0

25

50
f [Hz]

0.1
-1/

K [bits/s]

Log[E]

10

20

0.01
Exponential region
E~e-f/

potential region
E~f

0.001

Fig. 1. Power spectral density regions for a 3D uidized bed based on

bed surface uctuation signals.

3.1. Kolmogorov entropy

The Kolmogorov entropy is a direct measure of the chaos
level (unpredictability) that determines the rate of information lost by the system. When dealing with fully deterministic periodic data, the Kolmogorov entropy is zero, whereas
for random time series, the entropy is innite. A positive
Kolmogorov entropy value is obtained when deterministic
chaotic processes show intermediate behavior.
The Kolmogorov entropy, usually expressed in units of
bits/s, provides qualitative and quantitative information on
the underlying dynamics and has been widely used to characterize uidized bed hydrodynamics (Vander Stappen et al.,
1993; Marzocchella et al., 1997).
Although there are several approaches to estimating
the Kolmogorov entropy from experimental time series
(Grassberger and Procaccia, 1983; Daw and Hallow, 1993),
due to its reliability and fast computation, the algorithm
used in the present study is based on the maximum likelihood method (Schouten et al., 1994). The average absolute
deviation was used as the cut-off length to estimate the Kolmogorov entropy and since the maximum norm is used to
compute the interpoint distance, an embedding dimension
of 1 is sufcient for a reliable estimate of the entropy.
Schouten et al. (1996, 1999) described the close relationship between the Kolmogorov entropy and ow
macrostructure. These authors used the Kolmogorov entropy to identify the bubble pattern, and related K to the
excess gas velocity, Uex = U0 Umf , assumed to form the
bubbles. This close relationship between Kolmogorov entropy and the macrostructure of ow has also been recently
discussed by Johnsson et al. (2000).
According to the ndings of these last authors, VillaBriongos and Guardiola (2003) found that the power spectral density derived from experimental time series recorded
from bed surface uctuations can also be divided into three
different regions (Fig. 1). Region 1 corresponds to the
macrostructure of ow (visual appearance), which for the

3D1

3D2

3D3

2D1

2D2

2D3

0
0

10

20

30
f[Hz]

40

50

60

Fig. 2. Kolmogorov entropy as a function of the cut-off frequency of the

corresponding low pass lter applied over the signal for several 3D and
2D experiments.

time series considered usually ranges from the minimum

spectral frequency to 7 Hz. Regions 2 and 3 reect the ner
structure of the spectra: region 2, from 7 Hz up to 20 Hz,
comprises the interval of exponential decay (Sigeti and
Horsthemke, 1987) and region 3 corresponds to the potential decay law, i.e., it is the region where the power
spectrum attens out into instrumental noise (stochastic).
Thus, to gain insight into the relative inuence of the different power spectral regions on the Kolmogorov entropy
value, K is estimated by applying subsequent low-pass
lters over the experimental time series (Fig. 2). Then, as
Fig. 2 shows, the Kolmogorov entropy is mostly due to
regions 1 and 2 (macrostructure and ne structure ruled by
exponential decay), in agreement with results described in
the literature. Moreover, from Fig. 2 it can be concluded
that the Kolmogorov entropy used in the calculations is not
inuenced by the low-pass lter applied over the time series
(fc = 60 Hz).
4. Hypothesis of dynamic similarity
As stated before, the literature lacks reports of scaling
2D hydrodynamic data to 3D systems. The similarity theory
fails because of the different geometry shown by these two
systems. Thus, we need to develop a method of dynamically comparing the systems. The essential concept of chaos
scale-up methodology (Van den Bleek and Schouten, 1993)
for preserving information ow in both systems during a
scale-up process will be used as the criterion for dynamic
similarity.
It is assumed that a gassolid uidized bed behaves as
a chaotic system. Its dynamics are therefore characterized
by a strange attractor (Eckman and Ruelle, 1985). To attain

5154

J. Villa Briongos, J. Guardiola / Chemical Engineering Science 60 (2005) 5151 5163

dynamic similarity, the attractors of each system must show

the same strangeness, namely, they should have the same
sensitivity to the initial conditions.
Nevertheless, it is well known that the Kolmogorov entropy quanties the strangeness of a strange attractor in
terms of the rate of information generated or lost by the
system. Thus for dynamic similarity, the two systems must
show the same Kolmogorov entropy under the same operational conditions. However, according to the previous discussion, the Kolgomorov entropy is strongly dependent on
the macrostructure of ow (Fig. 2), which, in turn, is mainly
due to the bubble pattern. Hence, a direct comparison of
K between 2D and 3D beds is not suitable, since the distributions of radial and lateral bubbles are affected by the
column design. This inuence is more pronounced in 2D
systems, where wall effects strongly inuence drag prole
forces, modifying bubble distribution and consequently the
solid circulation inside the bed. This can be observed in the
value of Umf estimated experimentally, which is larger in
2D than in 3D systems.
We therefore need to resolve the lack of geometric similarity problem, for instance by eliminating or diminishing
the effect of the geometry of the system on the Kolmogorov
entropy value.
With regard to the macrostructure of ow, Horio et al.
(1986) proposed a rule for dynamic similarity from the
perspective of geometrical similarity. According to these authors, dynamic similarity between a large bed and its model
is achieved when the scaling parameters C1 and C2 in Table 1
are matched. According to Horio et al. (1986), if these relations (Table 1) are kept constant between model and prototype, both systems will show equivalent bubble coalescence,
bubble splitting and interstitial ow pattern. Moreover, they
stated that for powders in which bubble splitting is negligible, the rst condition, C1 , is sufcient for a geometrically
similar distribution of bubble diameter and bubble group.
These authors therefore use the Froude number (C1 ) as a
measure of ow macrostructure. More recently, Sanderson
and Rhodes (2003) used the bubbling scaling rules proposed by Horio et al. (1986) to scale the solid circulation
patterns within a wide range of bed diameters, showing that
solid mixing phenomena are mainly controlled by bubble
behavior in the bed. According to these results, it can be
concluded that for 3D3D conventional scaling, the Froude
number is a reliable measure of the macrostructure of ow
(bubbles plus solid mixing).
Moreover, since the Froude number, C1 , is dened as a
function of the excess gas velocity, Uex = (U0 Umf ) (according to the simple two phase model it is assumed that the
gas passes through the bed as bubbles, thus hereafter Uex will
be used to represent the visible bubble ow). This takes into
account the effect of column design on ow macrostructure,
since as stated before, the Umf estimated for 2D systems
accounts for the different drag prole forces, so the use of
Uex to compute F r partly removes the geometric effects on
macrostructure (bubble ow).

K'
[bit/s]
Rer

KG' = K'Frr' [bits/s]

Rer
1

Fr'r
Yes
KG = KFrr

KG = m0.5 KG' ?
No

K
[bit/s]
Rer
Fr'r

Dynamic
similarity

KG = KFrr [bits/s]
Rer

Fig. 3. Diagram of the 2D3D data-scaling hypothesis.

As indicated in Table 1, the need to compare the two systems on a similar length scale means that although mesoscale effects of bubbles would probably be better accounted
for by db , the need to estimate db for 3D systems led us to
nally use dp as a representative length to dene F r r . Further, the well-known particle size effect on uidization quality and thus on the macrostructure of ow supports the use of
dp . Thus, F r r measures the balance between inertia to gravity forces in the system, responsible for ow macrostructure
irrespective of column design.
To diminish the relative inuence of column design on
the Kolmogorov entropy value, the Froude number relative
to the excess gas ow, F r r dened in Table 1 will be used
as a geometric factor by its product to K.
Now, the overall information ow of the system, KG , can
be dened as
KG = K F r r .

(1)

As shown in Fig. 3, KG represents the information ow

of the bed as well as the Kolmogorov entropy. However,
the relative inuence that the system geometry has on ow
macrostructure, and therefore on the Kolmogorov entropy
value, has been reduced by the product to F r r .
The similarity hypothesis is thus that two uidized beds
with the same global information ow, KG , performing under the same uidization conditions, show the same rate of
generating and destroying information, independent of column design, and are consequently dynamically similar.
Finally, the uidization conditions are given by the
Reynolds particle number relative to the excess gas velocity (Table 1), Rer . This number represents the balance
between inertia to viscous forces for the gassolid uidized
system responsible for the macrostructure of ow, as F r r ,
is independent of column design.

J. Villa Briongos, J. Guardiola / Chemical Engineering Science 60 (2005) 5151 5163

5. Dimensionless scaling of chaotic hydrodynamics

1000

where subscripts 1 and 2 refer to the bed model and its

prototype, respectively.
However, the different geometries of 2D and 3D systems
means a geometric relationship between their scaling factors
cannot be established and, thus, m0.5 cannot be directly inferred as done in conventional 3D3D scaling. Nevertheless,
the dynamic comparison proposed throughout this paper is
a posteriori, meaning that the 2D and 3D systems do not
belong to the same sale-up procedure. Accordingly, the 3D
bed target is xed and a set of 2D experiments are properly
designed and performed to obtain hydrodynamic information on gas/solid dynamics (i.e., bubble size distribution).
Hence, the time factor for comparing the dynamics of 2D
and 3D systems is obtained from the relationship between
the experimental minimum uidization velocity (Eq. (2)),
assuming it gives a suitable relative measure of the different
time scale of the systems.
Thus, for a 3D (1) and a 2D (2) uidized bed, the relationship between their global information ow is
Umf 1
KG2 = m0.5 KG2 .
Umf 2

(3)

Now, when we examine the relationship KG vs the operating conditions (Rer ) for a set of uidized beds of different
geometry operating with the solid group 1 in Table 2, the
loglog plot of KG vs Rer , (Fig. 4) shows how, in this case,
the dynamics of the 2D and 3D systems are characterized
Table 2
Properties of the uidized bed material
Group no.

d p (
m)

p (kg/m3 )

Geldart group

Umf (exp) (m/s)

1
2
3
4

229
458
647
772

2650
2650
2650
2650

B
B
D
D

0.03
0.11
0.18
0.21

3D 2D
2D
Log (KGm0.5)

Once the hypothesis of dynamic similarity has been stated,

the procedure used to compare KG between different uidized beds needs to be established, taking into account the
different sizes of the 2D and 3D systems used.
The dimensionless scaling of chaotic hydrodynamics was
rst addressed by Van den Bleek and Schouten (1993), who
without experimental evidence proposed the use of dp /U0
as the time factor. Later, Schouten et al. (1996) successfully
used m0.5 as the time factor to scale the Kolmogorov entropy
between two scaled beds of different size. In conventional
3D3D scaling, m0.5 can be deduced from the relationship
among the four scale factors suggested by Fitzgerald et al.
(1984). The time factor is nally given by (Kunii and
Levenspiel, 1991):

0.5
Umf 1
t1
L1
= =
= m0.5 ,
(2)
Umf 2
t2
L2

KG1 =

5155

100

3D 2D

T1 L/D = 1.4
B2 L = 6
B2 L = 7
B2 L = 8
B2 L = 9

10

B2 L = 10

1
0.1

1
Log Rer

10

Fig. 4. Information ow diagram for a 3D bed (T1) and a set of 2D

uidized beds (B2) performing with the rst solid group (d p = 229
m)
indicated in Table 2.

by a similar KG value within a certain range of operating

conditions. The two systems can therefore be dynamically
compared within this region according to the previous hypothesis.

6. Experimental
6.1. Fluidized beds and experimental conditions
The experimental time series were collected from a set
of 2D and 3D uidized beds. The 2D systems (B1, B2)
were 1 and 2 cm wide, respectively, and the 3D uidized
beds (T1, T2) had internal diameters of 5.38 and 8.38 cm,
respectively. The beds were operated with a porous plate
distributor. They were lled to varying bed heights (from
4 to 9 cm for the 2D systems) and bed aspect ratios (from
0.7 to 2 for the 3D systems). The solid group was uidized
with air (atmospheric conditions) at relative gas velocities
ranging from 1.3 < Ur < 2.7.
Once the steady state is achieved, time series are obtained
by measuring the uctuations of the free bed surface. These
are measured when a laser sheet previously focused using
an ORIEL 12743 convergent lens, perpendicularly crosses
the bed, interacting with the bed top particles. The output
signal obtained from laserparticle interactions is collected
by a photodiode PSD, 4 cm in height, and is later amplied
before A/D conversion in the PC controlled by a software
package supplied by Monocrom S.A. (Barcelona, Spain).
The sample frequency used was 175 Hz and a low-pass
lter with a cut-off frequency of 60 Hz was simultaneously
applied over the signal, satisfying Nyquist criteria to avoid
possible aliasing problems. The length of the experimental
time series was 42 000 points (4 min). The time series was
divided into four blocks of 10 500 points before state space

J. Villa Briongos, J. Guardiola / Chemical Engineering Science 60 (2005) 5151 5163

analysis, and the average of the variables computed from

each block was presented as a result. The reliability of the
results was conrmed by the discrete variation around the
mean (less than 5%) of the Kolmogorov entropy computed
for each block. During the performance of the 2D gassolid
uidized bed of 1 cm width, video recordings were made to
evaluate the quality of uidization.
6.2. Bed materials and uidization quality
To design and scale-up a specic uidized bed operation,
it is worth knowing the contacting regime that will be ruling
the dynamics, as well as the uidization quality promoted
by the solid group used. Accordingly, the properties of the
powder and the way it uidizes need to be characterized.
We used solid particles belonging to B or D Geldart groups
(Table 2). However, after using complementary information
to the experimental observations in the video recordings and
ow regime diagrams (Kunii and Levenspiel, 1991), uidization behavior was classed as: A powder-type (group 1),
B-like (group 2) and D-like (groups 3, 4) (Villa-Briongos
and Guardiola, 2004). Thus, throughout this paper, particle group refers to Geldart classication criteria (Table 2)
whereas powder type is related to the observed uidization quality promoted by the solid within the uidized system. Moreover, according to the ow diagrams (Kunii and
Levenspiel, 1991), the supercial velocity range covered indicated that the beds performed in the bubbling regime.
7. Data-scaling methodology
The dynamic similarity hypothesis was implemented in
two steps (Fig. 3):
1. The effect of bed geometry on the Kolmogorov entropy
value was removed or diminished by its product to F r r .
2. The time factor, m0.5 , was used for suitable comparison
between beds of different sizes.
The ordinates of the information diagrams of the gures
presented here were estimated according to Eq. (3), where
the time factor m0.5 was drawn from the relation between
Umf 1 of the T1 system and Umf 2 of the other system compared. So the KG is recalculated for all systems but T1,
and when two beds that differ from T1 are compared, both
KG are computed taking as reference their corresponding
T1 system.
It may be seen in Fig. 4 that the dynamic behavior of the
2D and 3D systems is very similar, and it is possible that
in some situations both systems almost meet the conditions
required for the similarity hypothesis, allowing comparison
of their dynamics. However, this rarely occurs in practice.
Situations such as that shown in Fig. 5, in which there are
large hydrodynamic differences, are frequently found when
comparing 2D and 3D systems. Thus, one cannot simply
assume that, for instance, bubble size distribution measured
in a 2D bed can be translated to a 3D bed performing with

1000
3D - 2D
2D

Log (KGm0.5)

5156

100
T1 L/D = 0.7
3D

B1 L = 4

10

B1 L = 5
2D

B1 L = 6
B1 L = 7
B1 L = 8

1
0.1

1
Log Rer

10

Fig. 5. Information ow diagram for a 3D bed (T1) and set of 2D uidized

beds (B1) performing with the rst solid group (d p = 229
m) indicated
in Table 2.

the same particle group. Hence, the need arises for a criterion
that will indicate when hydrodynamic data collected from a
2D uidized bed can be appropriately scaled to a 3D system.
At this point, let us establish a practical case to apply
our methodology for comparing 2D and 3D dynamics. To
gain information on its hydrodynamics, several experiments
were performed using a 3D uidized bed of bed aspect ratio L/H = 1.4 and internal diameter 5.38 (T1 system). To
model its dynamics, a set of 2D experiments (i.e., B2 system) at varying bed heights were performed to estimate the
bubble size distribution. As far as we are aware, there are
no indications in the literature of the operational conditions
that will give rise to closest 2D3D dynamics.
We therefore monitored uidized bed hydrodynamics in
both systems by measuring free bed surface uctuations over
a broad range of uidizing conditions. Subsequently, KG
was used as the criterion for dynamic similarity according to
the dynamic similarity hypothesis (Fig. 4). Then, following
data-scaling methodology (points 1, 2), the KG values for the
3D and the 2D systems were compared within the common
operational range in the information ow diagram (Fig. 4).
In the end, the 2D-system showing a KG value closest to
the 3D bed at the same Rer will be dynamically the most
similar.
7.1. Evaluating the data-scaling method by comparing
2D3D dynamics
According to the data-scaling methodology described, the
B1, B2, T1 and T2 systems were used to evaluate the similarity hypothesis. Several comparisons of dynamics in the
experimental conditions described in Section 6 were made.
Fig. 6 shows the corresponding power spectral analysis
(Fig. 6a) (the abscissa has been properly scaled according to
the time factorEq. (1)therefore f2 =m0.5 f1 ) and the frequency histogram (Fig. 6b) for two uidized beds (2D3D)
of similar KG values. The beds were operated using the

J. Villa Briongos, J. Guardiola / Chemical Engineering Science 60 (2005) 5151 5163

5157

Fig. 6. Comparison between (a) the corresponding dimensionless power

spectrum and (b) the frequency histogram for two uidized beds performing with the second solid group (d p = 458
m) under the following
uidizing conditions: B2) H = 8 cm, Rer = 3.14, KG m0.5 = 24.4 bits/s;
T1) H /D = 1.4, Rer = 3.3, KG = 24.7 bits/s.

Fig. 7. Comparison between (a) the corresponding dimensionless power

spectrum and (b) the frequency histogram for two uidized beds performing with the second solid group (d p = 458
m) under the following
uidizing conditions: B1) H = 6 cm, Rer = 5.6, KG m0.5 = 57.7 bits/s;
T1) H /D = 1.4, Rer = 5.7, KG = 203 bits/s.

second solid group (Table 2) in the same uidizing conditions (see caption). The frequency histogram (Fig. 6b) indicates that the statistical distribution of the time series is
qualitatively similar for the two systems. Complementary
power spectral densities were well matched. As shown in
Fig. 6a, the energy value and energy distribution across the
frequency range were very similar in both systems.
In contrast, Fig. 7 shows a dynamic comparison in which
the similarity hypothesis was not satised: the two uidized
beds (2D3D) had different KG values when performed with
the second solid group under the same operational conditions. In this case, despite the frequency histogram (Fig. 7b)
indicating a qualitatively similar trend, the power spectral
density conrmed that the systems were ruled by a markedly
different ow pattern. It can thus be concluded that when the
similarity hypothesis is satised, time and frequency domain
analyses of the compared systems show reasonable agreement, indicating dynamic similarity. However when the hypothesis criterion is not fullled, the dynamics under comparison substantially differ.

Figs. 810 show the power spectral densities and frequency histograms for several 2D3D comparisons, according to the similarity hypothesis. The beds were operated
with the 1, 3 and 4 solid groups (Table 2). As for the second solid group, time and frequency analyses indicate similar dynamic behavior for the 2D and 3D systems compared,
thus validating the similarity hypothesis. Moreover, it seems
that for the 2, 3 and 4 solid groups (B and D uidization
like behavior), both the time and frequency domain analyses were more closely matched than for the rst solid group
(A-like behavior). This is explained by the different uidization quality promoted by each solid group and the F r r used
to reduce the geometric dependence of K. Thus, according
to the two-phase model, it was assumed that excess gas velocity through the bed occurs as bubbles. However, it is well
known that visible bubble ow depends on the solid group
used (Kunii and Levenspiel, 1991) and that the two phase
theory tends to overestimate ow in the bubble phase (Grace
and Harrison, 1969). According to the analysis described in
Section 6.2, particles in the 2, 3 and 4 solid groups uidized

5158

J. Villa Briongos, J. Guardiola / Chemical Engineering Science 60 (2005) 5151 5163

Fig. 8. Comparison between (a) the corresponding dimensionless power

spectrum and (b) the frequency histogram for two uidized beds performing with the rst solid group (d p = 229
m) under the following
uidizing conditions: B2) H = 7 cm, Rer = 0.93, KG m0.5 = 29.5 bits/s;
T1) H /D = 1.4, Rer = 0.94, KG = 29.7 bits/s.

like B and D powders, stabilizing bubbles and making the

splitting process negligible. This means that excess gas velocity could be a reliable measure of the visible bubble ow;
F r r (Table 1) being a proper measure of ow macrostructure indicating the complete abolishing of the geometric dependence of K by the product of K to F r r . In contrast, the
rst solid group promoted A solid-like uidization, meaning
the splitting process often occurs and that the macrostructure of ow is not well represented by the F r r number.
Thus, KG maintains some geometric dependence. These results are in agreement to those of Horio et al. (1986) who
found that the Froude number provided a reliable measure
of ow macrostructure only when bubble splitting was negligible (powders of the B and D type).
It should be noted that despite the large difference between
the uidized beds compared in Fig. 10a (1 cm wide 2D bed
vs 3D bed of D = 5.38 cm), the time and frequency domain
analyses matched well, validating the similarity hypothesis.
Thus, despite the scarce variation in frequencies in the power
spectral analysis (Fig. 10a), in which 2D frequencies shifted

Fig. 9. Comparison between (a) the corresponding dimensionless power

spectrum and (b) the frequency histogram for two uidized beds performing with the third solid group (d p = 647
m) under the following
uidizing conditions: B2) H = 7 cm, Rer = 22, KG m0.5 = 639 bits/s; T1)
H /D = 1, Rer = 21.4, KG = 648 bits/s.

to higher values once the frequency was properly scaled,

both uidized systems qualitatively showed the same bubble
ow pattern. Such shifting could be due to the fact that in this
case, the time factor used to scale the frequencies does not
properly account for the different diameter of the bursting
bubbles at the surface promoted by solid group No. 4.
7.2. Comparing different particle systems
When the whole set of uidized beds (B1, B2, T1, T2)
performing with the four particle groups (Table 2) is included in the same information ow diagram (Fig. 11 ), it
may be observed that particle groups showing similar experimental behavior are represented by points close together
in the diagram. Thus, the information ow diagram reects
the Geldart classication of uidized powders. In Fig. 11,
global information ow has been properly time scaled to a
T1 system. Hence, according to this gure, particle groups
3 and 4 show dynamically closer behavior under some

J. Villa Briongos, J. Guardiola / Chemical Engineering Science 60 (2005) 5151 5163

5159

1.E+04
D

Log (KGm0.5)

B
A

1.E+03

dp

1.E+02

1.E+01
1.E+02

229
458
647
772
1.E+03

1.E+04
Log Rer

1.E+05

1.E+06

Fig. 12. Information ow diagram provided by the particle array model.

Fig. 10. Comparison between (a) the corresponding dimensionless power

spectrum and (b) the frequency histogram for two uidized beds performing with the fourth solid group (d p = 772
m) under the following
uidizing conditions: B1) H = 7 cm, Rer = 9.5, KG m0.5 = 26 bits/s; T1)
H /D = 1.4, Rer = 9.7, KG = 31.5 bits/s.

1.E+04
D
1.E+03

Log (KGm0.5)

A
1.E+02

1.E+01

dp
229
458
647
772

1.E+00

1.E-01
1.E-01

1.E+00

1.E+01

Fig. 13. Comparison between (a) the corresponding dimensionless power

spectrum and (b) the frequency histogram for two uidized beds performing under the following uidizing conditions: B2) d p =647
m, H =7 cm,
Rer = 13, KG m0.5 = 198 bits/s; T1) d p = 772
m, H /D = 1.4, Rer = 13,
KG = 204 bits/s.

1.E+02

Log Rer
Fig. 11. Information ow diagram for the whole set of beds and particle
groups used.

uidizing conditions, and some of their operational points

overlap in the chart. This similar behavior is in agreement with the D-like uidization predicted from: (1) the

5160

J. Villa Briongos, J. Guardiola / Chemical Engineering Science 60 (2005) 5151 5163

1000
Log (KGm0.5)

Log (KGm0.5)

100

10

1
T1
T2

0.1
0.1
(a)

1
Log Rer

10
T1
T2

1
1

10
(c)

100

10
Log Rer

Log (KGm0.5)

10

100

10

T1
T2

1
0.1
(b)

1
Log Rer

100

1000

T1 T2

Log (KGm0.5)

100

T1
T2

10

1
(d)

10
Log Rer

100

Fig. 14. Information ow diagram for a set of 3D uidized beds scaled according to conventional similarity rules performing under the following
conditions: (a) d p = 229
m, T1 (H /D = 1) vs T2 (H /D = 1); (b) d p = 229
m, T1 (H /D = 1.7) vs T2 (H /D = 1.7); (c) d p = 458
m, T1 (H /D = 0.7)
vs T2 (H /D = 0.7); (d) d p = 458
m, T1 (H /D = 1) vs T2 (H /D = 1).

experimental behavior observed in the video recording; (2)

the uidization contacting modes deduced from ow regime
diagrams (Kunii and Levenspiel, 1991); and (3) the properties of uidized powders (Geldart, 1986). Moreover, the diagram in Fig. 11 shows how the rst group (A-type) behaves
dynamically different to groups 3 and 4. Finally, group 2
(B-type) shows a uidization behavior that is intermediate
between A- and D-like.
If we consider the information ow diagram (Fig. 11), the
results provided by a simple uidized bed dynamic model
like the particle array model are able to qualitatively predict
experimental trends fairly well (Fig. 12 ). This conrms the
fact that different uidizing powders can be sorted in terms
of global information ow. Further, Fig. 12 shows how powders that experimentally promote similar uidization quality
(groups 3 and 4) appear close in the diagram, according to
the experimental results (Fig. 11).
According to the above, Fig. 13 compares two geometrically different uidized beds (2D3D) performing with
different particle groups. As described in the caption to
Fig. 13, both systems show similar global information ow
in the same uidizing conditions. Thus, their corresponding
operational points overlap on the diagram in Fig. 11. According to the similarity hypothesis, both uidized systems
are dynamically similar. Consequently, the two power spectral densities and frequency histograms show good t, conrming the dynamic similarity predicted by the hypothesis
of dynamic similarity.

The similarity hypothesis is, therefore, also useful for

comparing different uidized systems operating with different particle groups, which in principle do not belong to the
same scale-up procedure. This opens up the possibility of
reusing previous knowledge by comparing the global information ow of the systems in the information diagram.
7.3. Extending the hypothesis for scaling 3D3D uidized
bed hydrodynamics
The nal purpose of a scale-up process is to preserve
the dynamics between the prototype and model to further
attain thermal and conversion similarities. Thus, KG /Rer
diagrams can be very useful for verifying the reliability of
the scale-up process during 3D3D scaling.
Glicksman scaling rules (Table 1) based on geometric
similarity were therefore followed to scale-up the dynamics
between two 3D uidized beds of internal diameter 5.38
(T1) and 8.38 cm (T2). The beds were operated using groups
1 and 2 (Table 1) as bed material and for several bed aspect
ratios. Fig. 14shows the KG /Rer diagram for both T1 and
T2; m0.5 was used as the time factor to scale KG . Then,
according to the results derived from the diagram (Fig. 14), if
the model and prototype have dynamic similarity, the global
information ow of the system, KG , has the same behavior
on the Rer for both uidized beds (their curves have the
same slope). Namely, their operational points lie on the same
line of the loglog plot.

J. Villa Briongos, J. Guardiola / Chemical Engineering Science 60 (2005) 5151 5163

5161

bubbles (Horio et al., 1986), fail to show dynamic similarity

after the scaling process. In this case, the scale-up procedure
should be revised.

8. Conclusions and limitations

Fig. 15. Comparison between (a) the corresponding dimensionless power

spectrum and (b) the frequency histogram for two uidized beds performing under the following uidizing conditions: T1) d p = 229
m,
H /D = 1.7, Rer = 0.52, KG = 11.6 bits/s; T2) d p = 229
m, H /D = 1.7,
Rer = 0.52, KG m0.5 = 11.4 bits/s.

It can be inferred from Figs. 14a, c, and d, that for these

bed aspect ratios, both beds behave dynamic similarly within
a broad range of operational conditions (their points appear
on the same line). Thus, it would be correct to assume that
previous knowledge of the smaller system will be useful for
the larger one (i.e., bubble properties extracted from a 2D
uidized bed and scaled to the 3D model).
In contrast, Fig. 14b shows that model and prototype are
only dynamically similar over a particular range of uidizing
conditions. Thus the scale-up procedure was only successful
over the operational range in which their dynamics matched.
This small-scale experiment would thus not be useful for
modeling bed behavior on a larger scale outside the corresponding operational range. Fig. 15 shows that within the
operational range commonly applied, model and prototype
show dynamic similarity.
Fig. 14b is an example of the difculty involved in uidized bed scale-up. Two beds showing both geometric similarity (same column design, bed aspect ratio, s /f ,
)
and geometric similarity in the hydrodynamics of the scaled

A method is proposed to compare and properly scale

2D3D hydrodynamic data. This methodology, which may
also be used to validate 3D3D dynamic scaling, is based
on the deterministic chaos theory and takes into account
the main force balances of the system (represented by Fr
and Re numbers). In addition, the method may be used to
qualitatively and quantitatively compare dynamic similarity
between two beds within a broad range of uidizing conditions using only the information ow diagram (Fig. 11).
The information ow diagram also enables us to compare
uidized beds operating with different particle groups, that
in principle do not belong to the same scale-up procedure.
This makes it possible to reuse the knowledge acquired in
previous research, comparing global information ow in the
information diagram.
Moreover, the fact that this methodology is based on
the deterministic chaos theory makes it suitable for studying non-linear dynamics, rather than using the more common frequency and time domain analyses. It is well known
that the linear characteristics of spectral analysis may not
describe some nonlinear aspects of the measurements obtained from complex systems (Morll and Bunk, 2001) and
that two very different time series can yield the same time
domain analysis.
The global information ow of the system was used as the
criterion for dynamic similarity. The similarity hypothesis
proposed is that two uidized beds that have the same global
information ow, KG , under the same uidization conditions will be dynamically similar. However, when the bed
performs with type A powders, the KG may still preserve
some geometric dependence due to the bubble splitting process, making the dynamic comparison less reliable. Future
studies will therefore need to focus on improving the geometric factor, for instance, by taking into account bubble
splitting, to completely abolish the geometric dependence
of K in these systems.
The results presented are obviously not devoid of certain
limitations. First, it should be mentioned that the small supercial gas velocity range covered complementary to the
small bed sizes limit the generality of the method. This implies the need to work at higher gas velocities and use larger
facilities to conrm the trends in global information ow
observed. In addition, the fact the beds were mostly performing at a transition like-state (high ow instability) will
make working at higher gas velocity particularly interesting
in terms of observing the behavior of KG in such conditions.
The different solid groups used were all Ballotini glass
beads of different size cuts. Thus, different types of solids
would also need to be tested in future work.

5162

J. Villa Briongos, J. Guardiola / Chemical Engineering Science 60 (2005) 5151 5163

Finally, we would like to point out that the measurement technique used has not yet been adapted for
measuring large-diameter 3D beds operating at large supercial gas velocities. It would be very interesting to
check our scale-up methodology on data collected using measurement techniques that are broadly applied
in uidization engineering such as pressure uctuation
measurements.

Acknowledgements
J. Villa-Briongos would like to thank Ruud van Ommen
and Professor Cor M. van den Bleek from TU DELFT
for their useful advices and inspiring work. Moreover, the
nancial support from project PPQ2003-08742 is kindly
acknowledged.
References

Notation
dp
dp
D
E
f
F
g
H
K
KG
m0.5
U0
Umf
Ur

particle diameter,
m
particle mean diameter,
m
bed diameter, m
power spectrum energy, dimensionless
spectral frequency, Hz
histogram frequency, dimensionless
gravity acceleration (9.81 m/s2 )
bed height, m
Kolmogorov entropy, bits/s
global information ow, bits/s
time scale factor, dimensionless
supercial gas velocity, m/s
minimum uidization velocity
relative gas velocity dened as U0 /Umf

Greek letters



f


f
S

exponent governing potential decay in region 3

of the power spectrum
inverse time scale governing the exponential decay region of the power spectrum
uid viscosity, Pa s
particle-shape factor
uid density, kg/m3
solid density, kg/m3

Abbreviations
B1
B2
Bins
Cn
F rr
Rer
T1
T2

2D uidized bedwidth 1 cm
2D uidized bedwidth 2 cm
refers to normalized measure of surface uctuation, dimensionless
n = 1, 2, dimensionless parameters dened in
Table 1
Froude number relative to the excess gas velocity (U0 Umf )
Reynolds particle number relative to the excess
gas velocity (U0 Umf )
3D uidized bedinternal diameter 5.38 cm
3D uidized bedinternal diameter 8.38 cm

Agarwal, P.K., 1985. Bubble characteristics in gas uidized beds. Chemical

Engineering Research and Design 63, 323336.
Agarwal, P.K., 1987. Effect of bed diameter on bubble growth an incipient
slugging in gas uidised beds. Chemical Engineering Research and
Design 65, 345354.
Baskakov, A.P., Tupogonov, V.G., Filippovsky, N.F., 1986. A study of
pressure uctuations in a bubbling uidized bed. Powder Technology
45, 113117.
Clift, R., 1986. Hydrodynamics of bubbling uidized beds. In: Geldart,
D. (Ed.), Gas Fluidization Technology. Wiley, Chichester. (Chapter 4).
Clift, R., Grace, J.R., 1971. Coalescence of bubbles in uidized beds.
A.I.Ch.E. Symposium Series 67 (116), 2333.
Daw, C.S., Hallow, J.S., 1993. Evaluation and control of uidization
quality through chaotic time series analysis of pressure drop
measurements. A.I.Ch.E. Symposium Series 89 (296), 103122.
Eckman, J.P., Ruelle, D., 1985. Ergodic theory of chaos and strange
attractors. Reviews of Modern Physics 57, 617656.
Fitzgerald, T., Bushnell, D., Crane, S., Yeong-Cheng Shieh, 1984. Testing
of cold scaled bed modeling for uidized-bed combustors. Powders
Technology 38, 107120.
Fitzgerald, T.J., 1985. Coarse particle systems. In: Davison, J.F., Clift,
R., Harrison, D. (Eds.), Fluidization. Academic Press, London, UK.
(Chapter 12).
Geldart, D., 1970. The size and frequency of bubbles in two and threedimensional gas-uidised beds. Powder Technology 4, 4155.
Geldart, D., 1986. Gas Fluidization Technology. Wiley, Chichester.
Glicksman, L.R., 1984. Scaling relationships for uidized beds. Chemical
Engineering Science 39, 13731379.
Glicksman, L.R., Yule, T., 1995. Prediction of the particle ow conditions
in the freeboard of a freely bubbling uidized bed. Chemical
Engineering Science 50, 6979.
Glicksman, L.R., Hyre, M.R., Farrel, P.A., 1994. Dynamic similarity in
uidization. International Journal of Multiphase Flow 20, 331386.
Grace, J.R., Harrison, D., 1969. The behaviour of freely bubbling uidised
beds. Chemical Engineering Science 24, 497508.
Grassberger, P., Procaccia, I., 1983. Estimation of Kolmogorov entropy.
Physical Review A 28, 25912593.
Horio, M., Akira, N., Yoshitaka, S., Iwao, M., 1986. A new similarity
rule for uidized bed scale-up. A.I.Ch.E. Journal 32, 14661482.
Johnsson, F., Zjierveld, R.C., Schouten, J.C., Van den Bleek, C.M.,
Leckner, B., 2000. Characterization of uidization regimes by timeseries analysis of pressure uctuations. International Journal of
Multiphase Flow 26, 663715.
Karamavru, A.I., Clark, N.N., 1997. Local differential pressure analysis in
a slugging bed using deterministic chaos theory. Chemical Engineering
Science 52, 357370.
Kehlenbeck, R., Yates, J., Di Felice, R., Hofbauer, H., Rauch, R.,
2001. Novel scaling parameter for circulating uidized beds. A.I.Ch.E.
Journal 47, 582589.
Kunii, D., Levenspiel, O., 1991. Fluidization Engineering. ButterworthHeinemann, Boston, MA.
Levenspiel, O., 2002. G/S reactor models-packed beds, bubbling uidized
beds, turbulent uidized beds and circulating (fast) uidized beds.
Powder Technology 122, 19.

J. Villa Briongos, J. Guardiola / Chemical Engineering Science 60 (2005) 5151 5163

Lim, K.S., Agarwal, P.K., ONeill, B.K., 1990. Measurement and modeling
of bubble parameters in a two dimensional gas-uidized bed using
image analysis. Powder Technology 60, 159171.
Marzocchella, A., Zijerveld, R.C., Schouten, J.C., Van den Bleek, C.M.,
1997. Chaotic behavior of gassolid ow in the riser of a laboratoryscale circulating uidized bed. A.I.Ch.E. Journal 43, 14581468.
Matsen, J., 1996. Scale-up of uidized bed processes: principle and
practice. Powder Technology 88, 237244.
Morll, G., Bunk, W., 2001. New design on complex patterns. Europhysics
news, May/June, 7781.
Moses, N.B., 1996. A rigorous description of gassolid uidized beds.
Chemical Engineering Science 51, 603622.
Sanderson, J., Rhodes, M., 2003. Hydrodynamics of solid motion and
mixing in bubbling uidized beds. A.I.Ch.E. Journal 49, 23172327.
Schouten, J.C., Takens, F., Van den Bleek, C.M., 1994. Maximum
likelihood estimation of the entropy of an attractor. Physical Review
E 49, 126129.
Schouten, J.C., Vander Stappen, M.L.M., Van den Bleek, C.M.,
1996. Scale-up of chaotic uidized bed hydrodynamics. Chemical
Engineering Science 51, 19912000.
Schouten, J.C., Zijerveld, R.C., Van den Bleek, C.M., 1999. Scale-up
of bottom bed dynamics and axial solids-distribution in circulating
uidized beds of Geldart-B particles. Chemical Engineering Science
54, 21032112.
Shrikant, V.D., Klizing, G.E., 1993. Pressure uctuation analysis for
uidized bed. A.I.Ch.E. Symposium Series 89 (296), 170183.

5163

Sigeti, D., Horsthemke, W., 1987. High-frequency power spectra for

systems subject to noise. Physical Review A 35, 22762282.
Skrzycke, D.P., Nguyen, K., Daw, C.S., 1993. Characterization of
uidization behavior of different solid types based on chaotic time
series analysis of pressure signals. In: Rubow, L.N. (Ed.), Proceedings
of the 12th International Conference on Fluidized Bed Combustion,
vol. 1. ASME, New York, p. 155.
Van den Bleek, C.M., Schouten, J.C., 1993. Can deterministic chaos create
order in a uidized-bed scale-up. Chemical Engineering Science 48,
23672373.
Vander Stappen, M.L.M., Schouten, J.C., Van den Bleek, C.M., 1993.
Deterministic chaos analysis of the dynamical behaviour of slugging
and bubbling uidized beds. In: Rubow, L.N. (Ed.), Proceedings
of the 12th International Conference on Fluidized Bed Combustion,
vol. 1. ASME, New York, p. 129.
Villa-Briongos, J., Guardiola, J., 2003. Free top uidized bed surface
uctuations as a source of hydrodynamic data. Powder Technology
134, 133144.
Villa-Briongos, J., Guardiola, J., 2004. Using free surface uctuations in
a 3D uidized bed for dynamic characterization. A.I.Ch.E. Journal 50,
30603067.
Werther, J., 1999. Measurement techniques in uidized beds. Powder
Technology 102, 1536.
Yrquez-Ramirez, M.I., Duursma, G.R., 2000. Study of the ow pattern
above an erupting bubble in an incipiently uidised bed using image
shifting. Chemical Engineering Science 55, 20552064.