ARTICLE IN PRESS

Chemical Engineering Science 61 (2006) 489 – 504

www.elsevier.com/locate/ces

Studies in multiple impeller agitated gas–liquid contactors

Satish D. Shewale, Aniruddha B. Pandit∗
Chemical Engineering Division, Institute of Chemical Technology, University of Mumbai, Matunga, Mumbai 400 019, India

Received 16 October 2004; received in revised form 27 April 2005; accepted 29 April 2005

Abstract
Experiments have been performed to study the effect of the density and the volume of the tracer pulse on the mixing time for two
impeller combinations in the presence of gas in a 0.3 m diameter and 1 m tall cylindrical acrylic vessel. The tall multi-impeller aerobic
fermenters, which require periodic dosing of nutrients that are in the form of aqueous solution, is a classic case under consideration.
Conductivity measuring method was used to measure the mixing time. Two triple impeller combinations; one containing two pitched
blade downflow turbines as upper impellers and disc turbine as the lowermost impeller (2 PBTD–DT) and another containing all pitched
blade downflow turbines (3 PBTD) have been used. Other variables covered during experiments were the density and the amount of the
tracer pulse, the impeller rotational speed and the gas superficial velocity. Fractional gas hold-up, Power consumption and mass transfer
coefficient have also been measured for both the impeller combinations. Influence of aeration and impeller speed on the mixing time has
been explained by the interaction of air induced and impeller generated liquid flows. Three different flow regimes have been distinguished
to explain the hydrodynamics of the overall vessel (i.e., multiple impeller system). A compartment model with the number of compartments
varying with the flow regimes have been used to model liquid phase mixing in these flow regimes. A correlation for the prediction of the
dimensionless mixing time in the loading regime has been proposed in order to account the effect of the density and the amount of the
tracer pulse on the mixing time. Correlations have also been proposed to predict fractional gas hold-up and kL a.
䉷 2005 Elsevier Ltd. All rights reserved.

Keywords: Gas–liquid contactor; Multiple impeller system; Flow patterns; Mixing time; Compartment model; Exchange flow rate

1. Introduction offer lower average shear as compared to a single impeller

The multiple impeller-agitated systems are used in several
industrial applications namely; fermentations, gas–liquid re-
actions such as hydrogenation, dissolution and crystalliza-
tion, polymerization reactors, waste water treatment, etc.
Multiple impellers are preferred over a single impeller as
multiple impellers provide better gas utilization in gas–liquid
system due to the higher gas phase residence time, narrower
spread in the residence time distribution in the flow sys-
tems and higher surface area per unit liquid volume for heat
transfer. Also, multiple impellers are preferred over a sin-
gle impeller where shear sensitivity to micro-organisms is
an important criteria for the design, as multiple impellers
∗ Corresponding author. Tel.: +91 22 24145616; fax: +91 22 24145614.
E-mail address: abp@udct.org (A.B. Pandit).
0009-2509/ - see front matter 䉷 2005 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ces.2005.04.078
system due to overall lower operational speeds (Gogate et
al., 2000) at equivalent power dissipation rate and allow the
freedom of controlling the dispersed phase hold-up and the
residence time over a wide range. Multiple impeller biore-
actors are now becoming important due to their efficient
gas-distribution and better oxygen utilization characteris-
tics, higher gas phase residence time, increased gas hold-up
and superior liquid flow (plug flow) characteristics (Gogate
et al., 2000).
The study of the liquid phase homogenization process
in a stirred tank (single or multiple impellers) has of-
ten been approached through the study of ‘mixing time’.
Mixing time is the time necessary for attaining the pre-
defined state of homogeneity of the liquid bulk after the
addition of a tracer pulse or a second liquid, which is to
be homogenized. It is the main parameter used to define
time scale of the macroscopic convective processes in the
 ARTICLE IN PRESS
490 S.D. Shewale, A.B. Pandit / Chemical Engineering Science 61 (2006) 489 – 504
liquid mixing in the stirred reactor when liquid used is
batch-wise.
Smith et al. (1987), Nocentini et al. (1988), Hudcova et
al. (1989), Abrardi et al. (1988, 1990) and Bouaifi et al.
(1997) have investigated flow patterns for gas–liquid con-
tactor agitated by different multiple impeller combinations.
Abardi et al. (1988) have identified different flow regimes
for the lowermost impeller and upper impellers. Abrardi et
al. (1990), Vasconcelos et al. (1995), Vasconcelos and Alves
(1995), Vrabel et al. (1999) and Machon and Jahoda (2000)
have studied the effect of the impeller speed and the gas
velocity on the mixing time. Hristov et al. (2001) have an-
alyzed the gas–liquid mixing accompanied by bioreaction
using a 3-D network-of-zones and have reported that the
simulations can provide detailed predictions of the local gas
hold-up distribution, the local mass transfer area, the par-
tial segregation of both the dissolved gas and the nutrient
and the extent of oxygen depletion of bubbles. Hristov et
al. (2001) have demonstrated these theoretical predictions
using colour-augmented 3D contour maps and solid-body
isosurface images. The analysis reported by Hristov et al.
(2001) is for a real case and they have shown that various
gradients (dissolved gas, nutrient, etc.) were well matched
by the predictions of the 3D model. Alves et al. (1997)
have compared compartment model of mixing, with different
number of compartments per agitation stage with exchange
flow between the adjacent compartments. This compartment
model resembles fluid plug flow model when the number of
compartments are more. Machon and Jahoda (2000) have
used both plug flow model and compartment model success-
fully. In the present study, a compartment model with a sin-
gle compartment per agitation stage has been used. Similar
model has been successfully used by Jahoda and Machon
(1994), Otomo et al. (1995) and Kasat and Pandit (2004)
for unaerated system and by Machon and Jahoda (2000) and
Moucha et al. (2003), for aerated system. Several studies
have been reported (Ahmad et al., 1985; Rielly and Pan-
dit, 1988; Burmester et al., 1991; Bouwmans et al., 1997;
Gogate and Pandit, 1999; Pandit et al., 2000) to account for
the effect of the density difference between the tracer fluid
and the bulk fluid, the type and geometry of the impeller and
the amount of the tracer fluid on the mixing time in a sin-
gle impeller system. Kasat and Pandit (2004) have reported
for the first time the effect of the density and the volume
of the added tracer pulse on the mixing time for unaerated
multiple impeller system.
Maximum cell growth and optimum product yield in the
seed fermenter; and optimum cell growth and maximum
product yield in the production fermenter, are ensured by
controlling the nutrient concentration and other fermenter
parameters precisely. The tall multi-impeller aerobic fer-
menters, which require periodic dosing of nutrients (that are
in the form of aqueous solutions) to maintain desired nutrient
concentration in the fermenter, is a classic case under consid-
eration. In addition to periodic dosing of nutrients, the con-
centration of nutrients in the dose (i.e., density of dose) and
the dose volume are also important parameters. The avail-
ability and the distribution of these nutrients, throughout the
fermenters can significantly affect the metabolic pathways
and the biological product distribution. However, no studies
have been reported to characterize the effect of the tracer
properties (the tracer density and the tracer volume) on the
mixing time in the multiple impeller system in the presence
of gas. The motion of the rising bubbles affects the inter-
action (exchange flows) between the adjacent impellers and
hence acquiring the knowledge of the different flow patterns
under different operating conditions is essential. Hence ob-
jectives of the present work are to identify different flow
regimes, to study the effect of these flow regimes on the
mixing time, if the tracer of different densities is used and
is added in the different quantities, and to get kL a data for
the gas–liquid operation in the multiple impeller system.
2. Experimental
The experiments have been performed in a cylindrical
acrylic vessel of 0.3 m i.d. and 1 m height, with three im-
pellers mounted on the shaft. Four perpendicular baffles hav-
ing width 0.1 T were fitted along the height of the liquid. The
vessel has several apertures for the incorporation of the con-
ductivity measuring probes, having diameters of 26 mm. The
impeller spacing (S) has been maintained at 3D, with a lower-
most impeller located at a distance of 0.5 T from the bottom
of vessel, which was chosen on the basis of the assumption
that each impeller forms its own circulation loop when the
impeller spacing (S)
2D (Hudcova et al., 1989). Air was
sparged through a ring sparger having diameter, same as that
of impeller. The ring sparger with eight symmetrically drilled
holes of 1 mm diameter was located at a distance of 0.25 T
from the bottom of vessel. The ring sparger was designed by
using a criterion reported by Rewatkar and Joshi (1991) for
ensuring that all the holes of the sparger should be effective
over the range of superficial gas velocities covered in this
work. Fig. 1 schematically shows the experimental set up
used in the present study. Two different types of impellers;
six bladed pitched (blade angle,
= 45◦ ) downflow turbine
(D = T /3, blade height (h) = 0.3D), PBDT and six bladed
rushton disc turbine (D = T /3, h = 0.2D, blade length =
0.25D), DT were used to form two triple impeller com-
binations. Power numbers of DT and PBTD were 5 and
1.3, respectively (Kasat and Pandit, 2004). One impeller
combination was comprised of three Pitched Blade down-
flow turbines, while other was containing Disc turbine as the
lowermost impeller and Pitched Blade downflow turbines
as upper two impellers. First and second impeller combina-
tions are denoted as 3 PBTD and 2 PBTD–DT respectively
in the later text. Experiments have been performed with tap
water at 30 ◦ C. The vessel was filled with tap water up to the
height of 2.9 T. Superficial velocities (VG ) of the sparged air
were 5, 10, 15 and 20 mm/s. The impeller rotational speed
(N) was varied in the range of 0–8 rps.
 ARTICLE IN PRESS
S.D. Shewale, A.B. Pandit / Chemical Engineering Science 61 (2006) 489 – 504
P1
P4 P2
P3
0.1T
* ⋅P
1
×
P2
S
H Acrylic tube
⋅
P3
S 0
Aluminum Rod
0.5T ×
P4 0.25T
Rubber Cork
D
T Tracer Addition Tube
× Conductivity probe P1, P2, P3, P4
* Tracer injection point
Fig. 1. Experimental set-up.
The fractional gas hold-up was determined by visually
observing the dispersion heights between the two baffles in
the absence and presence of the aeration. Power consump-
tion was measured using electrical method. Electrical power
consumed by the impeller was determined by using a Tong
tester or Clamp-on meter. Following equation has been uti-
lized for calculating the impeller power consumption,
P = (VR IR + VB IB + VY IY ) cos ,
where the value of power factor, ‘cos ’ is 0.8. Due to the
variation of the power factor, it was independently estimated
before the start of the experimental run on each day. The
power factor varied only within a narrow range of ±3%. The
effective power consumed in the bulk liquid is the difference
between impeller power consumption measured at the exper-
imental conditions and that measured at the same impeller
speed, but in the absence of water. The total power dissipated
in the bulk fluid can be calculated by summing the effective
impeller power (PG ) and gassed power (=QG HD LG g).
Flow regimes for the multiple impeller agitated gas–liquid
contactors were identified by visual observations and by tak-
ing images of the vessel at a time interval of 0.5 s using a
digital camera. Also by giving a pulse of KMnO4 solution
near the free surface of the gas–liquid dispersion, liquid flow
behavior in the vessel was observed and recorded. Based on
these observations, Flow regimes have been distinguished
491
3
Dimensionless concentration
C1
2
C2
C3
1
0 20 40 60 80
time, s
Fig. 2. Experimental conductivity response curve, for 2 PBTD–DT,
VG = 5 mm/s, N = 8 rps, tracer density = 1054 kg/m3 , Vt /Vb = 0.0025
where, C1, C2 and C3 are concentrations in the 1st, 2nd and 3rd com-
partments, using probes P 2, P 3 and P 4, respectively.
for the lowermost impeller, upper impellers and overall ves-
sel as such.
Conductivity measurement method with NaCl solution as
a tracer fluid has been used to measure the mixing time.
The density of the tracer fluid was varied from 1054 to
1178 kg/m3 by dissolving an appropriate quantity of sodium
chloride in water. The amount of the tracer pulse was varied
in the range of 0.25–1.75% of the bulk liquid volume, i.e.,
160–1120 ml. A tracer pulse of known density and volume
was dumped on the surface of the bulk liquid with the help
of an acrylic tube of diameter 0.05 m (Fig. 1), having pull
and push type of an arrangement (an aluminum rod with
rubber cork at one end). With the help of this arrangement
(1) it was also possible to add the high volume tracer almost in-
stantaneously (maximum addition time of 1 s for the largest
volume). The change in the conductivity of the bulk liquid
along the height of the vessel was recorded with the help of
the four conductivity probes. One of the conductivity probes
was located near the surface of the bulk liquid to get the cor-
rect estimate of the time at which the tracer was added. The
remaining three conductivity probes were located just be-
low each impeller plane, which gives an accurate estimation
of the mixing time in each impeller zone and also the over-
all mixing process. Each of the three probes was separated
from its adjacent probe by a vertical distance of 0.3 m and
by an azimuthal angle of 90◦ (Fig. 1). Fig. 2 shows a typical
experimental conductivity response. From the conductivity
vs. time data, the mixing time was calculated for 95% ho-
mogeneity. Details of the calculation of 95% mixing time
are described by Kasat and Pandit (2004). The values of the
experimental mixing time (i.e., the mixing time for slow-
est responding probe which is the lowermost probe) with
 ARTICLE IN PRESS
492 S.D. Shewale, A.B. Pandit / Chemical Engineering Science 61 (2006) 489 – 504
respect to the different parameters, reported later in the text
are typically an average of 3–4 such experiments. Mixing
time measurement was found to be reproducible within the
accuracy limits of ±5%. Bujalski et al. (2001) have strongly
recommended that unshielded probes should not be used
for measuring the mixing time in the presence of gas phase
because the contact of the gas bubbles with the electrode
makes local conductivity signal too noisy to measure the
mixing time accurately. In the present study, each conductiv-
ity probe was covered with the fine mesh screen to prevent
the contact of the gas bubbles with the electrode and to get
a smooth conductivity response curve. No significant differ-
ence in the values of the mixing time was observed when
the experiments were performed for unaerated liquid using
conductivity probes with and without the fine mesh screen
covering. This indicates that the fine mesh screen did not in-
troduce any significant delay in the response to the changes
in the conductivity and also did not affect the liquid flow at
the electrodes significantly.
A dissolved oxygen (DO) probe was located at a distance
of T /4 from the bottom of the vessel in a separate set of
experiments. Experiments have been performed to measure
kL a for both of the impeller combinations, for two values
of VG ; 5 and 10 mm/s and at impeller speed ranging from 0
to 8 rps. DO of the tap water was eliminated by the addition
of the stoichiometrically required quantity of the sodium
sulfite. Once the DO of the bulk liquid reached the range
of 0–0.3 mg/l, i.e., initial DO for the experiment (Ci ), air
flow and impeller action were started simultaneously and the
change in the DO level of bulk liquid (C) was recorded with
respect to the time (t), until C reached the saturation con-
centration (C ∗ ). Data of C varying in the range of 20–60%
of the saturation concentration vs. respective values of the
time was considered for the calculation of kL a. kL a is the
slope of the curve ln((C ∗ − Ci )/(C ∗ − C)) plotted against
time, t. The values of kL a, reported later in the text are typ-
ically an average of 2–3 such experiments. Measurement of
kL a was reproducible within accuracy limits of ±7%.
3. Results and discussion
3.1. Flow patterns
Flow patterns observed in the vessel depend upon VG ,
N and impeller combination. Different flow patterns were
observed for the lower and upper impellers. Fig. 3 shows
some typical flow patterns observed in the vessel; with im-
peller combination of 2 PBTD–DT for VG = 10 mm/s and
with 3 PBTD for VG = 5 mm/s, at different impeller speeds.
Mainly two flow regimes have been identified for the low-
ermost impeller; one Flooding (F) regime and second being
the Loading (L) regime. For upper impellers also mainly two
flow regimes were observed; in the first one, upper impellers
were not able to divert gas bubbles in radial or downward
direction, but due to rotating action of the upper impellers,
diameter of the central gas plume increases and gas bubbles
are poorly dispersed (radially) in the region above the upper
impellers. While in the second one, upper impellers divert
gas bubbles in the radial or in the vertically downwards di-
rection and gas bubbles get well dispersed in the region un-
derneath each of the impeller and/or radially outward. Using
the same terminology as used by Abrardi et al. (1990), the
first regime is defined as Ineffective dispersion (DI) while
the second one is defined as Effective dispersion (DE). The
impeller speed at which the transition occurs from DI to DE
regime for upper impellers is defined as NU , while the im-
peller speed at which the transition occurs from the F to L
regime for the lowermost impeller is defined as NF . Three
hydrodynamic regimes can be distinguished to explain hy-
drodynamics of the overall vessel (i.e., multiple impeller
system); DI–F, DE–F, DE–L for both of the impeller combi-
nations. Shortcut notations used to define hydrodynamics of
the overall vessel are straightforward, e.g. DI–F means up-
per impellers are in the DI regime and lowermost impeller
is in the flooding (F) regime while DE–L means upper im-
pellers are in the DE regime and the lowermost impeller is
in the loading (L) regime. Similarly DE–F means upper im-
pellers are in the DE regime and the lowermost impeller is
in the flooding (F) regime.
It can be seen from Table 1 that the values of NF for 2
PBTD–DT are lower by 6–35% than predicted NF for single
DT using the correlation proposed by Nienow et al. (1985);
whereas the values of NF for 3 PBTD are also lower (devia-
tion of 20–26%) than predicted NF values for single PBTD
using the correlation proposed by Medek and Fort (1997).
Thus it can be concluded from Table 1 that the values of NF
for multiple impeller (actually lowermost impeller in mul-
tiple impeller system) system are lower than that for a sin-
gle impeller. It can be also seen that the difference between
NU and NF is higher, i.e., DE–F regime gets extended over
a wider impeller speed range, for 3 PBTD than that for 2
PBTD–DT. This could be attributed to the difference in the
power dissipation level as discussed later.
3.2. Power consumption studies
Gassed power consumption (PG ) values in the DE–L
regime has been given in the following correlation form,
originally proposed by Michel and Miller (1962), rather than
reporting these values individually for the brevity of the text,

n
Po2 N D 3
PG =
. (2)
Q0.56
G
Values of
and n, required to fit the gassed power consump-
tion data for both of the impeller combinations are given in
the Table 2 along with the values reported by Michel and
Miller (1962) which are for a single DT. It can be seen from
the Table 2 that the values of ‘n’ obtained for both of the
impeller combinations are much closer to the value of ‘n’
reported by Michel and Miller (1962), i.e., 0.45; while the
 ARTICLE IN PRESS
S.D. Shewale, A.B. Pandit / Chemical Engineering Science 61 (2006) 489 – 504 493

Fig. 3. Flow patterns for 2 PBTD–DT (A,B,C) and for 3 PBTD (D,E,F) at VG = 10 mm/s and 5 mm/s respectively. (A, D) DI–F regime; (B, E) DE–F
regime; (C, F) DE–L regime. (A) N = 1.67 rps; (D) N = 0 rps; (B,E) N = 3.75 rps; (C,F) N = 5.08 rps.

Table 1
Values of NU and NF for 2 PBTD–DT and 3 PBTD, with predicted values of NF for single DT and single PBTD

VG (mm/s) 2 PBTD–DT 3 PBTD DT (Nienow et al., 1985) PBTD (Medek and Fort, 1997)

NU (rps) NF (rps) NU (rps) NF (rps) NF (rps) using NF (rps)

QG
nb−0.542 (sin
)−0.58
2
= 30(D/T )3.5 N gD =854.5Q0.558
G
ND 3

5 <1.67 2.42 <1.67 3.75 3.76 4.67

10 1.67 3.75 <1.67 5.08 4.75 6.87
15 2.42 5.08 1.67 6.5 5.4 8.61
20 2.42 6.5 2.42 8 6 10.1

Table 2 3.3. Fractional gas hold-up

Values of
and n for PG correlation, Eq. (2)

Impeller combination PG correlation, Eq. (2) It can be seen from Fig. 4 that for any value of VG ,
at equivalent PG /V , both impeller combinations gives ap-

n Std. dev. (%)
proximately the same (within ±10%) fractional gas hold-
DT (Michel and Miller, 1962) 0.812 0.45 — up. Our gas hold-up data for the DE–L regime is compared
2 PBTD–DT 1.501 0.425 9 with the correlation reported by Moucha et al. (2003) as
3 PBTD 1.52 0.427 4 shown in Fig. 5. It can be observed from Fig. 5 that for
2 PBTD–DT, deviation in the gas hold-up values observed
by Moucha et al. (2003) and our experimental gas hold-
values of ‘
’ obtained for both of the impeller combinations up values is large (Moucha et al. (2003) reported 40–200%
are approximately same (
1.52 ± 0.05) but significantly higher G values), while for 3 PBTD, it is low (anywhere
higher (nearly two times) than the value of ‘
’ reported by between 10% and 90%). This can be attributed to the fact
Michel and Miller (1962), i.e., 0.812. This also shows that that tap water has been used as the experimental liquid in
the lowermost impeller follows the classical gassed impeller the present work while 0.5 M Na2 SO4 aqueous solution was
flow regimes and gets affected by the gas even in multi- used by Moucha et al. (2003). The effect of the solutes on
impeller system as if it was the only impeller present in the the fractional gas hold-up is mainly through the property of
system. The variation in the value of ‘
’ shows that ‘
’ in- the coalescence and the presence of coalescence inhibiting
creases essentially due to an additive effect (addition of the solutes results into an increase in the fractional gas hold-
individual impeller power number) of the upper impellers, up. This is in accordance with the results of Pinelli et al.
but remains approximately constant irrespective of the im- (1994) who have reported that for triple impeller configura-
peller combination indicating the relative insensitivity of the tions containing all DTs, the coalescence inhibiting solutions
upper impellers to the gassed conditions. of sodium sulphate and polyvinyl-pyrrolidone (PVP) exhibit
 ARTICLE IN PRESS
494 S.D. Shewale, A.B. Pandit / Chemical Engineering Science 61 (2006) 489 – 504

0.08 correlating gas hold-up with impeller power consumption

VG =
per unit volume (PG /V ) instead of Ptot /V gave better data
0.07 2 PBTD-DT fitting in this work, fractional gas hold-up was correlated as
3 PBTD
20 mm/s a function of PG /V and VG using the following type of an
0.06 empirical equation:
VG =
15 mm/s  B
0.05 PG
G = A VGC . (3)
V
εG

0.04
VG = Values of constants A, B, C obtained by the regression of
0.03 10 mm/s the experimental data in the DE–L regime for both of the
impeller combinations are given in the Table 3 along with
0.02 the values of A, B, C given by Moucha et al. (2003) who
reported correlation for G in similar form as that of Eq.
0.01 (3). It can be seen from Table 3 that values of the constants
VG = 5 mm/s A, B, C for 2 PBTD–DT are higher than that for 3 PBTD.
0 It can be also seen from Table 3 that for both of the impeller
0 1 2 3 combinations, the value of B (exponent over PG /V ) reported
log (PG/V, W/m3) by Moucha et al. (2003) is higher than that reported here;
while the value of C (exponent over VG ) reported by Moucha
Fig. 4. Effect of PG /V on G with VG as a parameter (empty et al. (2003) is lower than that obtained in this work. This
points = DI– F regime; shadowed empty points = DE– F regime; filled
could be also attributed to the difference in the type of the
points = DE– L regime).
gas sparger used in this work.

0.16 3.4. Mixing time studies

3.4.1. Effect of impeller rotational speed

Fig. 6 shows a typical variation in the mixing time for 2
0.12 PBTD–DT with impeller speed at a VG of 10 mm/s. Similar
trend was observed for all values of VG for both impeller
combinations (2PBTD–DT and 3 PBTD). It can be seen
Predicted εG

from Fig. 6 that lowest values of the mixing time (which

0.08 are changing marginally with an increase in the impeller
speed) are observed in the DI–F regime. This is due to the
fact that the liquid mixing is controlled by the liquid flow
induced by the gas bubbles (Fig. 3A). When impeller speed
0.04 2PBTD-DT and reaches just beyond NU , the liquid flow generated by the
3PBTD, Eq. 3 middle impeller, PBTD in the vertically downward direction
2PBTD-DT and increases and becomes comparable to the air-induced liquid
3PBTD, Moucha et al. (2003) flow in the vertically upward direction. Due to these equal
0 and opposite flows, local turbulence is generated. Since en-
0 0.04 0.08 0.12 0.16 ergy gets dissipated in this local turbulence, less energy is
Experimental εG
available towards the overall liquid circulation and convec-
tive mixing and the mixing time increases due to the transi-
Fig. 5. The comparison of experimental G in the DE–L regime with the
literature data. tion from DI to DE regime for the upper impellers as seen in
Fig. 6. The role of the convective and turbulent mixing has
been discussed by Patwardhan and Gaikwad (2003) and this
80–150% and 130–300% higher hold-up, respectively, than observation is consistent with their conclusion. If the im-
water. Nevertheless, near parallel lines to the 45◦ line indi- peller speed is further increased to reach the DE–F regime,
cate that the type of the parameter dependence proposed by the middle impeller starts diverting gas bubbles in the radi-
Moucha et al. (2003) is still valid for the impeller combina- ally outwards and/or the downward direction (refer Fig. 3B).
tions explored in this work. Rushton and Bimbinet (1968) These gas bubbles rise near the wall region and continue to
had correlated fractional gas hold-up as a function of the travel along the wall region due to the uppermost impeller.
total power consumption per unit volume (Ptot /V ) and the In the DE–F regime, in the upper compartment of the ves-
superficial gas velocity (VG ) with exponent over Ptot /V as sel, liquid flow induced by the motion of the gas bubbles
0.37 and exponent over VG as 0.61 for disc turbine. Since is in the vertically upward direction in the wall region and
 ARTICLE IN PRESS
S.D. Shewale, A.B. Pandit / Chemical Engineering Science 61 (2006) 489 – 504 495

Table 3
Values of A, B and C for gas hold-up correlation, Eq. (3) and kL a correlation, Eq. (8)

Impeller combination Gas hold-up correlation, Eq. (3) kL a correlation, Eq. (8)

A B C Std. dev. (%) A B C Std. dev. (%)

2 PBTD–DT 1.544 0.263 1.156 7 0.0191 0.57 0.7 6.5

3 PBTD 1.504 0.2 1.079 6 0.0057 0.701 0.59 5

Moucha et al. (2003)

2 PBTDT–DT 0.101 0.518 0.707 6.2 9.577e−4 1.138 0.462 11.7
3 PBTD 0.347 0.288 0.795 3.7 1.133e−3 1.154 0.508 8.6

in the vertically downward direction in the central region 60

(Fig. 3B). This vertically downward liquid flow in the central
region gets supplementary action of the liquid flow generated 50
DE-L
by the PBTDs in the downwards direction and the overall

Mixing time, sec

liquid circulation flow rate in the upper compartment of the
vessel increases improving the interaction between the upper
and lower compartments of the vessel and hence the mixing
time decreases with an increase in the impeller speed in the
DE–F regime as can be seen in the Fig. 6. If the impeller
speed is further increased, above NF , transition from flood-
ing to loading regime occurs for the lowermost impeller.
In this DE–L regime, each impeller is having its own flow
pattern, i.e., it forms a circulation cell around itself (refer
Fig. 3C). In the DE–L regime, mixing performance of the
entire vessel is dominated by the exchange flow rate between
the adjacent circulation cells (compartments) instead of the
overall circulation flow rate generated by each impeller and
hence the mixing time increases sharply (refer Fig. 6). If
impeller speed is further increased in the DE–L regime, ex-
change flow rate between circulation cells further increases
and the mixing time decreases again (refer Fig. 6). Abrardi
et al. (1990) have reported similar observation (decreasing
mixing time with an increasing N) in the DE–L regime for
a dual impeller system.
3.4.2. Effect of impeller rotational speed and gas
superficial velocity
Fig. 7 shows that for both of the impeller combinations,
in the DI–F and DE–F regimes, mixing time decreases with
an increase in VG at any impeller speed, because the liquid
flow induced by the gas bubbles obviously increases with
an increase in VG and hence improves the mixing; while in
the DE–L regime, if VG is increased at a constant N, mix-
ing time increases because increase in VG at constant N de-
creases the impeller power consumption reducing the liquid
circulation velocity generated by impeller and hence the ex-
change flow rate between adjacent circulation cells. Pandit
and Joshi (1983) have reported similar behavior for a sin-
gle impeller system. Pandit and Joshi (1983) have assumed
two different regimes separated by a critical impeller speed
for gas-phase dispersion (NCD ) for a specific gas flow rate
(QG ) and have reported that the mixing time decreases with
an increase in the QG at any impeller speed which is less
40
DE-F
DI -F
30
Transition from DE-F
NF
20 NU to DE-L regime
Transition from DI-F
10
to DE-F regime
0
0 2 4 6 8
Impeller speed, N, rps
Fig. 6. Effect of Impeller speed on mixing time, for 2 PBTD–DT,
VG = 10 mm/s, tracer density = 1054 kg/m3 , Vt /Vb = 0.0025 (empty
points = DI– F regime; shadowed empty points = DE– F regime; filled
points = DE– L regime).
than NCD and the mixing time increases with an increase in
the QG at any impeller speed which is greater than NCD .
Fig. 8 shows that at same values of Ptot /V , mixing time
for 3 PBTD is always less than that for 2 PBTD–DT. This
could be because of better axial dispersion efficiency of the
axial impellers as compared to that of radial one. This also
confirms the conclusion from the past literature (Prajitno et
al., 1998; Vrabel et al., 2000; Moucha et al., 2003) that the
impeller combinations having axial liquid flow give better
mixing efficiency as compared to others at the same power
consumption per unit volume in the case of multiple impeller
system.
3.4.3. Effect of tracer properties
It has been observed that in the DE–L flow regime, for
low tracer volume (i.e., Vt /Vb = 0.0025), NTmix marginally
changes (Fig. 9A) and for higher tracer volume (i.e., Vt /Vb =
0.0175), NTmix increases (Fig. 9B) significantly, with an
increase in
/LG or tracer density at a constant N and VG .
This could be attributed to the reason that though increase
in the
/LG is significant due to an increase in the tracer
density and decrease in the bulk density from L to LG ,
increase in the buoyancy force exerted on the tracer packet
 ARTICLE IN PRESS
496 S.D. Shewale, A.B. Pandit / Chemical Engineering Science 61 (2006) 489 – 504

120 120
(a) VG, mm/s (b) VG, mm/s
100 0 100 0
5 5
80 10 80 10
Mixing time, s
15 15
60 20 60 20
VG VG
40 40

20 20
VG VG
0 0
0 2 4 6 8 0 2 4 6 8
Impeller speed, rps

Fig. 7. Mixing time vs. N, with VG as a parameter, tracer density = 1054 kg/m3 , Vt /Vb = 0.0025 (empty points = DI– F regime; shadowed empty
points = DE– F regime; filled points = DE– L regime) (a) 2 PBTD–DT and (b) 3 PBTD.

60 Similar effect of tracer properties was also observed for DI–F

and DE–F regimes (Fig. 10). It can be also observed that
55 the extent of increase in NTmix , with an increase in
/LG
at higher Vt /Vb (Fig. 9B) and with an increase in Vt /Vb at
50 any value of the tracer density (Fig. 9C and D), is lower at
low impeller speed and higher at high impeller speeds in the
45 DE–L regime at constant VG . Fig. 10 depicts that the effect
Mixing time, s

of tracer properties on the mixing time is more pronounced

40
35
30
3 PBTD
25
2 PBTD-DT
20
0 200 400
Ptot/V, W/m3
Fig. 8. Mixing time vs. Ptot /V with impeller combination as parameter,
VG = 10 mm/s, tracer density = 1054 kg/m3 , Vt /Vb = 0.0025 (empty
points = DI– F and DE–F regime; filled points = DE– L regime).
(FB =
Vt g) is significant, at higher Vt than that at lower
Vt . It can be seen from the Fig. 9C and D that NTmix in-
creases with an increase in the tracer volume (Vt ) for all the
values of tracer densities covered in this work. This could
be attributed to the fact that the lower volume of the tracer
pulse added loses its identity very soon against the liquid
turbulence generated by the impeller, while higher volume
of the tracer pulse retains its identity for some more time
due to relatively large size and lower rate of turbulent ero-
sion of its identity. Also for higher tracer volumes, gravity
controlled liquid flow exists for unaerated case (Kasat and
Pandit, 2004) that results into lower mixing time, but in the
aerated case, gassing effect opposes gravity controlled liq-
uid flow significantly and results into higher mixing time.
at higher VG in the DI–F and DE–F regimes. This is be-
cause at higher speeds (for a constant value of VG ) in the
DE–L regime and at higher VG (for a constant value of N)
in the DI–F and DE–F regime, high values of gas hold-up
exists. This decreases average bulk density (LG ), increas-
ing the buoyancy force exerted on the tracer packet (FB )
significantly at higher Vt and resulting in an increase in the
mixing time.
600
3.4.4. Richardson number
The Richardson number is the ratio of the static head of
liquid to the dynamic head of the flowing liquid generated
by the impeller action. Since the buoyancy forces exerted
on the tracer packet also depends on the density and the
volume of the tracer packet. Pandit et al. (2000) have mul-
tiplied Richardson number (Ri) by Vt /Vb and modified the
conventional definition of it (which is valid in the range of
Vt /Vb from 0.0025 to 0.08), as follows:

gHD Vt
Modified Richardson number (Rio ) = . (4)
LG N 2 D 2 Vb
It can be observed from the Fig. 11 for DE–L regime
that the y-intercepts of the lines, NTmix vs. Rio , for different
impeller speeds at VG = 5 mm/s are approximately same
(average = 241.1); while the y-intercept increases with an
increase in VG (for VG = 20 mm/s, y-intercept is 314.2) and
the slope of the lines of NTmix against Rio at low N like
3.75 rps is low (224) while at high N like 8 rps it is much
higher (4212). The increase in the slope with an increase in
the impeller speed at constant VG and the increase in the
 ARTICLE IN PRESS
S.D. Shewale, A.B. Pandit / Chemical Engineering Science 61 (2006) 489 – 504 497

350 500
(A) (B)
450
300
400

250 350
300
200 250
VG, mm/s-N, rps VG, mm/s-N, rps
5- 3.75 5- 5.08 200 5- 3.75 5- 5.08
150
5- 6.5 5- 8.0 150 5- 6.5 5- 8.0
15- 6.5 15- 8.0 15- 6.5 15- 8.0
100 100
0 0.1 0.2 0.3 0 0.1 0.2 0.3
N×Tmix

∆ρ/ρLG

400 500
(C) (D)
350 450
400
300
350
250 300
200 250
150 200
VG, mm/s-N, rps 150 VG, mm/s-N, rps
100 10- 5.08 10- 5.08
100
50 10- 6.5 10- 6.5
50 10- 8.0
10- 8.0
0 0
0 0.005 0.01 0.015 0.02 0 0.005 0.01 0.015 0.02
Vt/Vb

Fig. 9. Effect of the tracer properties on NTmix in the DE–L regime for 2 PBTD–DT: (A) Vt /Vb = 0.0025, (B) Vt /Vb = 0.0175, (C) tracer
density = 1054 kg/m3 and (D) tracer density = 1178 kg/m3 .

y-intercept with an increase in the VG are probably due to the
fact that due to the better distribution of the gas, the gassing
effect needs to be overcome to bring the tracer down first
after its addition at the top (relative increase in the
/LG
value) and then mix with the liquid bulk. This gassing effect
is high when high gas hold-up exists in the vessel, i.e., low
QG and high N or high QG and low N in the DE–L regime.
Another reason that can be put forward to explain this is that,
at low impeller speeds in the DE–L regime, existence of the
gravity controlled regime improves the mixing performance
by bringing the tracer down in the impeller zone quickly and
low dependency on the Rio was thus found; while at high
impeller speeds in the DE–L regime, existence of the stirrer
controlled regime results in the delayed entrainment of the
added tracer in the impeller stream giving high dependency
of the mixing time on the Rio which is an effect undesirable
in the normal operation, but has been observed in this study,
mainly attributed to the fractional gas hold-up values.
Fig. 12A, which is constructed for 2 PBTD–DT in the
DE–L regime, indicates that the slopes of the lines (NTmix
vs. Rio ) for a typical N at different values of VG are approx-
imately the same (at N = 8 rps, slopes are 4212, 3524, 2239,
3309 for VG = 5, 10, 15, 20 mm/s respectively.). Fig. 12B,
which is constructed for 3 PBTD in the DE–L regime, also
shows similar behavior. This indicates that the mixing time
mainly depends upon the inertial forces generated by the
impeller in the DE–L regime, as the role of the gas sparging
is much lower but the value of the fractional gas hold-up
contributes significantly. This observation is consistent with
the observed variation in the mixing time.
For unaerated single impeller system, Ahmad et al. (1985)
have reported that NTmix is proportional to Ri0.8 . Pandit et al.
(2000) have reported existence of critical Rio for unaerated
single impeller system. Pandit et al. (2000) also reported that
at low Rio (i.e., less than critical Rio ), NTmix is independent
of Rio and at higher Rio (i.e., higher than critical Rio ), it
is proportional to Rino where the value of n is 0.58 for DT
and 0.76 for PBTD. It can be observed from Fig. 12 that
at low values of Rio (which exists for Vt /Vb = 0.0025),
the values of NTmix are approximately the same (nearly the
same intercept), regardless of N (in the DE–L regime) at
constant VG . This is because though fractional gas hold-up
increases with an increase in N, due to low Vt /Vb increase
in the FB is not significant. At high values of Rio (which
exists for Vt /Vb = 0.01 and 0.0175), higher values of NTmix
has been observed at higher speeds (Fig. 12). This could be
attributed to the higher values of fractional gas hold-up and
higher tracer volume as discussed earlier in this section.
 ARTICLE IN PRESS
498 S.D. Shewale, A.B. Pandit / Chemical Engineering Science 61 (2006) 489 – 504

80
(A) (B)
50 70

40 60
50
30
40

20 VG, mm/s-N, rps 30

VG, mm/s-N, rps
5- 0.0 5- 1.67 20 5- 0.0 5- 1.67
10 5- 2.42 15- 0 5- 2.42 15- 0
15- 1.67 15- 3.75 10 15- 1.67 15- 3.75
15- 5.08 15- 5.08
Mixing time, s

0 0
0 0.1 0.2 0.3 0 0.1 0.2 0.3
∆ρ/ρLG

50 60
(C) (D)
45
50
40
40
35
30 30
VG, mm/s-N, rps VG, mm/s-N, rps
25
10- 0.0 20 10- 0.0
20 10- 1.67 10- 1.67
10- 2.42 10 10- 2.42
15
10- 3.75 10- 3.75
10 0
0 0.005 0.01 0.015 0.02 0 0.005 0.01 0.015 0.02
Vt/Vb

Fig. 10. Effect of tracer properties on Tmix in flooding regime for 2 PBTD–DT: (A) Vt /Vb = 0.0025, (B) Vt /Vb = 0.0175, (C) tracer density = 1054 kg/m3
and (D) tracer density = 1178 kg/m3 .

3.4.5. Development of correlation 600

Since as discussed earlier for the plots of NTmix vs. Rio , y = 3308.7x + 314.2
y-intercept is a function of VG while the slope is mainly 500
a function of impeller speed, it was decided to correlate y = 4211.9x + 242.4
NTmix in the similar format. The correlation of NTmix con- 400 y = 1973.6x + 244
tains two terms; first term being dimensional and a function y = 900.4x + 231.6
N×Tmix

of VG , while second term being dimensionless, with inclu- 300

sion of dimensionless numbers. The following correlation y = 224.2x + 246.5
200
has been developed to predict NTmix in the DE–L regime
for 2 PBTD–DT, VG, mm/s - N, rps
100
  5-3.75 5-5.08 5-6.5
HD G −0.496 5-8.0 20-8
NTmix = 453.7VG 0.127
+ 583.95 0
T
  0 0.05 0.1 0.15 0.2 0.25
PG −1.297 2 Modified Richardson number, Rio
× (N D/g)1.756 (Rio )0.711 . (5)
PO
Fig. 11. Variation in the slope of curve NTmix vs. Rio , with VG –N as a
Eq. (5) has correlation coefficient = 0.89 and std. dev. = parameter in the DE–L regime, for 2 PBTD–DT.
6.85%, where all the variables are in MKS units. Correlation
for the NTmix in the DE–L regime for 3 PBTD having similar
form as that of Eq. (5) is as follows: Eq. (6) has correlation coefficient=0.9 and std. dev.=6.7%,
  where all the variables are in MKS units. Parity plots for
HD G −0.286
NTmix = 333.82VG 0.184
+ 553.63 correlations, i.e., Eqs. (5) and (6) are shown in Fig. 13A and
T B, respectively, and it can be seen that the agreement is rea-
 −0.39
PG sonable. From the Eq. (5) for 2 PBTD–DT and Eq. (6) for
× (N 2 D/g)0.858 (Rio )0.294 . (6)
PO 3 PBTD, it can be seen that proportionality constant in the
 ARTICLE IN PRESS
S.D. Shewale, A.B. Pandit / Chemical Engineering Science 61 (2006) 489 – 504 499

600 700
(A) (B)
500 600

500
400

N×Tmix
400
300
300
VG, mm/s - N, rps
200 VG, mm/s - N, rps
5- 3.75 5- 5.08 5- 6.5 200
5- 8.0 10- 5.08 10- 6.5 5- 5.08 5- 6.5
100 100 5- 8.0 10- 6.5
10- 8.0 15- 6.5 15- 8.0
20- 8.0 10- 8.0 15- 8.0
0 0
0 0.1 0.2 0 0.05 0.1 0.15
Modified Richardson number, Rio

Fig. 12. NTmix vs. Rio , in the DE–L regime: (A) 2 PBTD–DT and (B) 3 PBTD.

600 600
(A) (B)
500 500
Predicted N×Tmix

400 400

300 300

200 200

100 100

0 0
0 100 200 300 400 500 600 0 100 200 300 400 500 600
Experimental N×Tmix

Fig. 13. Comparison of predicted NTmix with experimental NTmix , in the DE–L regime: (A) using Eq. (5) for 2 PBTD–DT and (B) using Eq. (6) for 3
PBTD.

first term for 2 PBTD–DT (453.7) is higher than that for 3
PBTD (333.82) and proportionality constant in the second
term for 2 PBTD–DT (583.95) is also higher than that for 3
PBTD (553.63), which indicates that the mixing efficiency
of 3 PBTD is better than that of 2PBTD–DT. Exponent over
VG in the first term for 2 PBTD–DT (0.127) is lower than that
for 3 PBTD (0.184) indicating that effect of VG over the mix-
ing time is more pronounced for 3 PBTD as compared to 2
PBTD–DT. Exponent over (HD G /T ) for 3 PBTD (−0.286)
is higher than that for 2 PBTD–DT (−0.496); which indi-
cates that the gassing effect is more pronounced for 3 PBTD
than 2 PBTD–DT which also results into a relatively differ-
ent bubble size distribution (and hence different variation in
it’s rise velocity) and is consistent with the observed varia-
tion in the kL a as discussed later. Exponent over (N 2 D/g)
for 2 PBTD–DT (1.756) is higher than that for 3 PBTD
(0.858). This again indicates that for 2 PBTD–DT, mixing
time is a strong function of N as compared to 3 PBTD. The
exponent over Rio for 2 PBTD–DT (0.711) is higher than
that for 3 PBTD (0.294), indicating that the effect of tracer
properties on NTmix is more pronounced for 2 PBTD–DT
than that for 3 PBTD, i.e., 2 PBTD–DT is more suscepti-
ble to the variation in the tracer density and Vt /Vb than 3
PBTD. This could be attributed to the fact that at equiva-
lent values of Ptot /V , due to lower value of impeller speed
for 2 PBTD–DT, it produces lower hydrostatic head (which
is proportional to N 2 D 2 ) than that produced by 3 PBTD
which at equivalent power values operate at higher rotational
speeds and also at same values of N and VG , due to higher
values ofG and hence lower LG , though 2 PBTD–DT is
required to produce higher hydrostatic head as compared to
3 PBTD. This also indicates that though 2 PBTD–DT draws
higher power, it produces lower hydrostatic head as com-
pared to that produced by 3 PBTD and hence the depen-
dency as observed on the Rio (in terms of it’s exponent) is
contrary to the expection (based on the power dissipation).
3.5. Compartment model
A single circulation cell (or compartment) model is shown
in the Fig. 14A for DI–F regime based upon the visual obser-
vation (Fig. 3A and D). Joshi et al. (1982) have discussed a
theoretical procedure for the calculation of the mixing time,
which is based on the knowledge of the circulation velocity
 ARTICLE IN PRESS
500 S.D. Shewale, A.B. Pandit / Chemical Engineering Science 61 (2006) 489 – 504
QE
QE
QE
D 0
(A) (B) (C)
Fig. 14. Physical representation of the compartmental model for flow
regimes: (A) DI–F, (B) DE–F and (C) DE–L.
(VC ) and the length the flow path (L). They assumed that the
mixing time is five times the circulation time (=L/VC ). L
is the length of longest loop. Average liquid circulation ve-
locity (VC ), which is responsible for the overall mixing in
the DI–F regime, is calculated using the above definition of
the mixing time and experimentally measured mixing times.
Visual observation (Fig. 3B and E) depicts the presence of
two compartments in the vessel for DE–F regime. Location
of an imaginary boundary (indicated by dotted line in the
Fig. 14B) between the two compartments vary from distance
of H /2 (when N is just above NU ) to H /3 (when N is just
below NF ) from the bottom of the vessel. Liquid exchange
flow rate between the adjacent compartments was estimated
by least-square fitting the conductivity responses utilizing
the model of cascade of two perfectly mixed compartments
for DE–F regime. Model of three perfectly mixed circulation
cells with exchange flow rate QE between adjacent cells for
DE–L regime is shown in the Fig. 14C. The numbers of com-
partments are taken to be equal to the number of impellers;
since in the present work the impeller spacing is equal to
3D(=T ) and for such a spacing it has been proved previ-
ously (Hudcova et al., 1989; Mishra and Joshi, 1994) that
the impellers are hydro-dynamically independent of each
other. All the compartments have the same volume (Vcell ).
Exchange flow rate (QE ) was estimated by least-square fit-
ting the conductivity responses using the triple compartment
model for DE–L regime.
It can be observed from Fig. 15 that the exchange flow
rate (QE ) increases with an increase in the impeller speed
(QE ∝ N
) at a constant VG in the DE–L regime. Exponent
over N is approximately equal to unity (i.e., 1.02 and 0.95,
for VG = 5 and 10 mm/s, respectively), indicating that QE
is some constant fraction of the impeller discharge flow rate
(which is directly proportional to N). Similar behavior was
2.5
2
QE
QE, 10-3 m3/s
1.5
1 VG, mm/s
QE
0 5
0.5
10 15
20
0 3 6 9
N, rps
Fig. 15. Exchange flow rate vs. N in the DE–L regime, for 2 PBTD–DT,
tracer density = 1054 kg/m3 , Vt /Vb = 0.0025.
observed for other sets of the tracer density and Vt /Vb in
the DE–L regime. Another observation which can be made
from Fig. 15 that contrary to the expectation of increased
exchange flow rate due to the vertical motion of the bubbles
through all the cells sequentially, the exchange flow rates
were found to decrease with an increase in the gas flow rate
at any specific impeller rotational speed in the DE–L regime.
This also suggests that the decrease in the impeller power
dissipation rate due to the gas sparging and its net effect on
the reduction in the overall circulation velocities is far more
significant as compared to the additional liquid circulation
generated by the gas sparging. The observed increase in the
overall mixing time is also consistent with the above argu-
ment and the observation. Kasat and Pandit (2004) have re-
ported that for unaerated multiple impeller system, exchange
flow rate increases with an increase in the impeller speed
for both of the impeller combinations (2 PBTD–DT and 3
PBTD). Similar observation has been made for aerated mul-
tiple impeller system in the DE–L regime at a constant gas
flow rate. However the exchange flow rate (QE ) was found
to decrease with an increase in the gas flow rate (Fig. 15)
due to a reduced impeller power as discussed earlier.
Effective liquid velocity (Vl ) responsible for the liquid
mixing in the vessel is equal to the liquid circulation velocity
which was calculated using a single circulation cell model
for DI–F regime, while for DE–F and DE–L regimes Vl is
equal to QE /(T 2 /4). Vl has been plotted against Ptot /V
in Fig. 16 for both of the impeller combinations. It can be
seen from Fig. 16 that Vl in the DI–F regime is high as
compared to Vl in the DE–F and DE–L regimes. It has been
also observed that at a constant VG in the DI–F regime, the
values of Vl are approximately same (Fig. 16) and are nearly
equal (within ±5%) to the liquid circulation velocity in the
case of only sparging (i.e., VCS =1.31{gT (VG −G Vb∞ )}1/3 ,
 ARTICLE IN PRESS
S.D. Shewale, A.B. Pandit / Chemical Engineering Science 61 (2006) 489 – 504 501

0 0
(A) (B)

-0.5 -0.5
VG, mm/s

log (Vl, m/s)

-1 -1 0 5
10 15
20
-1.5 -1.5
VG, mm/s
-2 0 5 -2
10 15
20
-2.5 -2.5
0 200 400 600 800 0 100 200 300 400
Ptot/V, W/m3

Fig. 16. Effect of Ptot /V on effective liquid velocity responsible for the liquid mixing (Vl ): (A) 2 PBTD–DT and (B) 3 PBTD.

Table 4
Average values of QE /QCG w.r.t. tracer properties

Vt /Vb QE /QCG

2 PBTD–DT 3 PBTD
Tracer density (kg/m3 ) Tracer density (kg/m3 )

1054 1114 1178 1054 1114 1178

0.0025 0.293 0.308 0.298 0.333 0.336 0.341

0.01 0.277 0.257 0.244 0.281 0.242 0.233
0.0175 0.26 0.235 0.21 0.241 0.233 0.218

Joshi, 1980), which is calculated using Vb∞ = 0.3 m/s, i.e.,
system essentially behaves like a bubble column in DI–F
regime.
The flow pattern generated by the impeller is characterized
by a circulation loop, involving a liquid circulation flow rate
(QCO = NCO N D 3 ) in un-gassed condition. For DT with
D = T /3, NCO
2NQ
1.5 (Mahouast et al., 1991); while
for PBTD, NCO
NQ
0.8. NCO and NQ are circulation
number and flow number, respectively. The liquid circulation
flow rate under gassed conditions can be easily obtained by
QCG = QCO (PG /PO )1/3 .
Average values of the ratio, QE /QCG , for all the sets of the
tracer density and Vt /Vb are given in the Table 4. It was ob-
served that for 2 PBTD–DT, average values of the QE /QCG
are 0.30, 0.259 and 0.235 and for 3 PBTD, they are 0.337,
0.252 and 0.23, for the values of Vt /Vb ; 0.0025, 0.01 and
0.0175, respectively. It can be seen from the Table 4 that
for both of the impeller combinations, for Vt /Vb = 0.0025,
the ratio, QE /QCG , is approximately constant irrespective
of the change in the tracer density; while for Vt /Vb = 0.01
and 0.0175, the ratio decreases with an increase in the tracer
density. These results are consistent with the results ob-
tained from the circulation cell model (Joshi and Sharma,
1979) used to explain the behavior of the gas–liquid bub-
ble column (similar to the compartment model) that the
inter-cell exchange velocity is 0.31–0.33 times the average
circulation velocity. Kasat and Pandit (2004) have reported
that QE /QCO varies from 0.34 to 0.44 for unaerated system
for the same impeller combinations.
Vasconcelos et al. (1995) reported that vax is a scale inde-
pendent, characteristic velocity of liquid flow induced by air
having average value of 0.066 m/s. Vasconcelos et al. (1998)
have confirmed the scale independency of vax with an in-
variant value of 0.07 m/s for gassed tank stirred by multi-
ple Rushton turbines. Vrabel et al. (1999) also reported that
the value of vax is 0.054 m/s. By using the methodology of
Vasconcelos et al. (1995), value of vax has been calculated
(7) in the present work. The values of vax for 2 PBTD–DT and
3 PBTD were found to be 0.032 and 0.015 m/s, respectively.
These values are lesser (but of the same order of magni-
tude) as compared to the value of vax (0.066 m/s) reported
by Vasconcelos et al. (1995) for impeller combination com-
prising Rushton turbines.
3.6. Mass transfer studies
Fig. 17 depicts that at same value of Ptot /V , both the
impeller combinations gives approximately same (within
±7%) values of the kL a (MTC) in all regimes. This is in ac-
cordance with the conclusion of Moucha et al. (2003) who
studied the effect of the impeller configuration on kL a at con-
stant Ptot /V and reported that at lower Ptot /V (300 W/m3 )
the kL a values are approximately independent of impeller
 ARTICLE IN PRESS
502 S.D. Shewale, A.B. Pandit / Chemical Engineering Science 61 (2006) 489 – 504
0.03
0.025
VG = 10 mm/s
0.02
kLa, s-1
0.015
VG = 5 mm/s
0.01
0.005
2 PBTD-DT
3 PBTD
0
0 200 400 600
Ptot /V, W/m3
Fig. 17. kL a vs. Ptot /V , for both of the impeller combinations (empty
points = DI– F and DE–F regime; filled points = DE– L regime).
configuration, while at higher Ptot /V (800 W/m3 ), the
impeller configurations with high power number provide
significantly higher mass transfer coefficients. Linek et al.
(1987) have shown kL a to be independent of reactor size and
only dependent upon power consumption per unit volume
and VG .
Mass transfer coefficient (kL a) is correlated as a function
of total power consumption per unit volume (Ptot /V ) and
gas superficial velocity (VG ) by the following equation:
 
Ptot B C
kL a = A VG . (8)
V
The values of A, B, C were obtained by regression of the
experimental data in the DE–L regime for both of the im-
peller combinations and are given in Table 3 along with the
values of A, B, C reported by Moucha et al. (2003), who
proposed correlation for kL a in the similar form. Our ex-
perimental kL a data are compared with the correlation of
Moucha et al. (2003) in the Fig. 18. Values predicted by the
correlation of Moucha et al. (2003) are much higher as com-
pared to our experimental data. This can be explained on
the basis of the reason that Moucha et al. (2003) have used
0.5 M Na2 SO4 aqueous solution as an experimental fluid,
i.e., the presence of coalescence inhibiting solutes increases
the fractional gas hold-up (as explained earlier) and subse-
quently the MTC. It can be also seen from Table 3 that expo-
nent over Ptot /V reported by Moucha et al. (2003) is higher
than that observed in the present work; while exponent over
VG reported by Moucha et al. (2003) is lower than that in
the present work. These results are in accordance with the
findings of Linek et al. (1987) who reported that exponent
over power dissipation per unit liquid volume increases with
0.16
0.12
Predicted kLa, s-1
0.08
□ 2PBTD-DT and
0.04 ◊ 3PBTD, Eq. 8
䡵 2PBTD-DT and
♦ 3PBTD, M oucha et al. (2003)
0
0 0.04 0.08 0.12 0.16
Experimental kLa, s-1
Fig. 18. The comparison of experimental kL a in the DE–L regime with
literature data.
an increase in the concentration of coalescence inhibiting
solutes in the liquid.
4. Conclusions
Three hydrodynamic flow regimes have been identified
for the vessel having multiple impeller gas–liquid system,
namely; DI–F, DE–F and DE–L. DI–F means upper im-
pellers are in the ineffective dispersion (DI) regime and low-
ermost impeller is in the flooding (F) regime; while DE–L
means upper impellers are in the effective dispersion (DE)
regime and the lowermost impeller is in the loading (L)
regime. Similarly DE–F means upper impellers are in the
effective dispersion (DE) regime and the lowermost im-
peller is in the flooding (F) regime. Lowest values of mix-
ing times were observed at zero and low impeller speeds
(DI–F regime) with only gas sparging. If the impeller speed
is increased at a constant VG , above NU mixing time in-
creases due to transition from DI–F to DE–F regime and
then decreases in the DE–F regime. With further increase
in the impeller speed, above NF , mixing time increases due
to transition from DE–F to DE–L regime and decreases in
the DE–L regime with an increase in impeller speed. In the
DI–F and DE–F regimes, mixing time decreases while it in-
creases with an increase in the gas flow rate in the DE–L
regime. At equivalent values of Ptot /V , mixing efficiency
of 3 PBTD was observed to be better than 2 PBTD–DT,
whereas comparable values of kL a were observed for both
of the impeller combinations. At low impeller speed in the
DE–L regime, low dependency of Rio is observed on the di-
mensionless mixing time because mixing phenomenon gets
a supplement from the buoyancy generated liquid flow by
 ARTICLE IN PRESS
S.D. Shewale, A.B. Pandit / Chemical Engineering Science 61 (2006) 489 – 504 503

naturally distributing the tracer packet. Whereas at higher References

impeller speed in the DE–L regime, high dependency on Rio
has been observed probably due to the fact that gassing ef- Abardi, V., Rovero, G., Sicardi, S., Baldi, G., Conti, R., 1988. Sparged
fect needs to be overcome to bring tracer down (when added vessels agitated by multiple turbines. Proceedings of the Sixth European
Conference on Mixing, pp. 329–336.
at the top). Mixing under these conditions is totally gov- Abrardi, V., Rovero, G., Sicardi, S., Baldi, G., Conti, R., 1990.
erned by the stirrer generated flow and the power drawn by Hydrodynamics of gas–liquid dispersion with a multi impeller system.
these impellers in the presence of gas. Transactions of the Institution of Chemical Engineers 68 (A), 516–522.
Ahmad, S.W., Latto, B., Baird, M.H.I., 1985. Mixing of stratified liquids.
Chemical Engineering Research Design 63, 157–167.
Alves, S.S., Vasconcelos, J.M.T., Barata, J., 1997. Alternative compartment
Notation models of mixing in tall tanks agitated by multi-Rushton turbines.
Transactions of the Institution of Chemical Engineers 75 (A), 334–338.
Bouaifi, M., Roustan, M., Djebbar, R., 1997. Hydrodynamics of multi-
C concentration, mg/l stage agitated gas–liquid reactors. Proceedings of the Ninth European
D impeller diameter, m Conference on Mixing, pp. 137–144.
FB buoyancy forces exerted on the tracer packet Bouwmans, I., Bakker, A., Van den Akker, H.E.A., 1997. Blending liquids
(
Vt g), N of differing viscosities and densities in stirred tank vessels. Transactions
of the Institution of Chemical Engineers 75 (A), 777–783.
g gravitational constant, m/s2
Bujalski, W., Pinelli, D., Nienow, A.W., Magelli, F., 2001. Comparison
HD dispersion height in the vessel, m of experimental techniques for the measurement of mixing time in
IR , IY , IB electric current flowing through R, Y, B gas–liquid system. Chemical Engineering Technology 24 (9), 919–923.
phases, respectively, A Burmester, S.S.H., Rielly, C.D., Edwards, M.F., 1991. The mixing
kL a volumetric mass transfer coefficient, s −1 of miscible liquids with large difference in density and viscosity.
Proceedings of the Seventh European Conference on Mixing, pp. 9–16.
nb number of blades Gogate, P.R., Pandit, A.B., 1999. Mixing of miscible liquids with density
N impeller rotational speed, rps difference: effect of volume and density of the tracer fluid. Canadian
NF impeller speed required for transition from F Journal of Chemical Engineering 77, 988–996.
to L regime, rps Gogate, P.R., Beenackers, A., Pandit, A.B., 2000. Multiple impeller
NU impeller speed required for transition from systems with a special emphasis on bioreactors: a critical review.
Biochemical Engineering Journal 6, 109–144.
DI to DE regime, rps Hristov, H., Mann, R., Lossev, V., Vlaev, S.D., Seichter, P., 2001. A 3-D
PG gassed impeller power consumption, W analysis of gas–liquid mixing, mass transfer and bioreaction in a stirred
PO unaerated impeller power consumption, W bio-reactor. Transactions of the Institution of Chemical Engineers 79
Ptot total power consumption (C), 232–241.
(PG + gassed power(=QG HD LG g)), W Hudcova, V., Machon, V., Nienow, K.C.A.M., 1989. Gas–liquid dispersion
with dual Rushton turbine impellers. Biotechnology and Bioengineering
QG gas flow rate, m3 /s   34, 617.

gH
Ri Richardson number  N 2 DD 2 Jahoda, M., Machon, V., 1994. Homogenization of liquids in tanks stirred
LG by multiple impellers. Chemical Engineering Technology 17, 95–101.
Rio modified Richardson number (RiV t /Vb ) Joshi, J.B., 1980. Axial mixing in multiphase contactors — A unified
T tank diameter, m correlation. Transactions of the Institution of Chemical Engineers 58,
Tmix mixing time, s 155–165.
V , Vb bulk volume, m3 Joshi, J.B., Sharma, M.M., 1979. Some design features of radial baffles
in sectionalized bubble columns. Canadian Journal of Chemical
VG superficial gas velocity, m/s
Engineering 57, 375–377.
Vl effective liquid velocity responsible for liquid Joshi, J.B., Pandit, A.B., Sharma, M.M., 1982. Mechanically agitated
mixing, m/s gas–liquid reactors. Chemical Engineering Science 37, 813–844.
VR , VY , VB voltage across phases R, Y, B, respectively, Kasat, G.R., Pandit, A.B., 2004. Mixing time studies in multiple impeller
V agitated reactors. Canadian Journal of Chemical Engineering 82,
892–904.
Vt volume of the tracer added, m3 Linek, V., Vacek, V., Benes, P., 1987. A critical review and experimental
verification of correct use of dynamic method for the determination
Greek letters of oxygen transfer in aerated agitated vessels to water, electrolytic
solutions and viscous liquids. Chemical Engineering Journal 34, 11.

 difference between the densities of tracer Machon, V., Jahoda, M., 2000. Liquid homogenization in aerated multi-
fluid and gas–liquid dispersion impeller stirred vessel. Chemical Engineering Technology 23 (10), 869
L density of the liquid, kg/m3 –876.
Mahouast, M., Fontaine, P., Mallet, J., 1991. A two-component LDV bench
LG density of the gas–liquid dispersion, kg/m3 for automated determination of flowrates and energy transfers generated
by agitators. Proceedings of the Seventh European Conference on
Mixing, pp. 181–185.
Acknowledgements Medek, J., Fort, I., 1997. Dispersion effect of pitched blade impellers.
Proceedings of the Ninth European Conference on Mixing, pp.
107–114.
Authors would like to thank to University Grant commis- Michel, B.J., Miller, S.A., 1962. Power requirements of gas–liquid agitated
sion for providing research fellowship to SDS. systems. A.I.Ch.E. Journal 8 (2), 262–266.
 ARTICLE IN PRESS
504 S.D. Shewale, A.B. Pandit / Chemical Engineering Science 61 (2006) 489 – 504
Mishra, V.P., Joshi, J.B., 1994. Flow generated by disc turbine. Part
IV: multiple impellers. Chemical Engineering Research Design 73,
657–668.
Moucha, T., Linek, V., Prokopova, E., 2003. Gas hold-up, mixing time
and gas–liquid mass transfer coefficient of various multiple-impeller
configurations: rushton turbine, pitched blade and techmix impeller and
their combinations. Chemical Engineering Science 58, 1839–1846.
Nienow, A., Warmoeskerken, M., Smith, J., Konno, M., 1985. On the
flooding loading transitions and the complete dispersion dispersed
condition in aerated vessels agitated by a Rushton turbine. Proceedings
of the Fifth European Conference on Mixing, pp. 143–154.
Nocentini, M., Fajner, D., Pasquali, G., Majeli, F., 1988. A fluid dynamic
study of gas–liquid, non standard vessel stirred by multiple-impellers.
Chemical Engineering Journal 37, 53.
Otomo, N., Bujalski, W., Nienow, A.W., 1995. The application of a
compartment model to a vessel stirred with either dual radial or dual
axial flow impellers. The 1995 IChemE Research Event, The University
of Birmingham, vol. 2, pp. 829–831.
Pandit, A.B., Joshi, J.B., 1983. Mixing in mechanically agitated gas–liquid
contactors, bubble columns and modified bubble columns. Chemical
Engineering Science 38, 1189–1215.
Pandit, A.B., Gogate, P.R., Dindore V.Y., 2000. Effect of tracer properties
(volume, density and viscosity) on mixing time in mechanically agitated
contactors. Proceedings of the 10th European Conference on Mixing,
pp. 385–393.
Patwardhan, A.W., Gaikwad, S.G., 2003. Mixing in tanks agitated by jets.
Transactions of the Institution of Chemical Engineers 81 (A), 211–220.
Pinelli, D., Nocentini, M., Magelli, F., 1994. Hold-up in low viscosity
gas–liquid systems stirred with multiple impellers: comparison of
different agitator types and sets. Proceedings of the Eighth European
Conference in Mixing, pp. 81–84.
Prajitno, D.H., Mishra, V.P., Takenaka, K., Bujalski, W., Nienow, A.W.,
McKeemie, J., 1998. Gas–liquid mixing studies with multiple up- and
down-pumping hydrofoil impellers: power characteristics and mixing
time. Canadian Journal of Chemical Engineering 76, 1056–1068.
Rewatkar, V.B., Joshi, J.B., 1991. Role of sparger design in mechanically
agitated gas–liquid reactors. Part I: power consumption. Chemical
Engineering and Technology 14, 333–347.
Rielly, C.D., Pandit, A.B., 1988. The mixing of Newtonian liquids
with large density and viscosity differences in mechanically agitated
contactors. Proceedings of the Sixth European Conference on Mixing,
pp. 69–77.
Rushton, J.H., Bimbinet, J., 1968. Holdup and flooding in air liquid
mixing. Canadian Journal of Chemical Engineering 46, 16–21.
Smith, J.M., Warmoeskerken, M., Zeef, E., 1987. Flow conditions
in vessels dispersing gases in liquids with multiple impellers.
Biotechnology Processes, A.I.Ch.E. Journal 107–115.
Vasconcelos, J.M.T., Alves, S.S., 1995. Mixing in gas–liquid contactors
agitated by multiple turbines in the flooding regime. Chemical
Engineering Science 50, 2355–2357.
Vasconcelos, J.M.T., Alves, S.S., Barata, J.M., 1995. Mixing in gas–liquid
contactors agitated by multiple turbines. Chemical Engineering Science
50, 2343.
Vasconcelos, J.M.T., Alves, S.S., Nienow, A.W., Bujalski, W., 1998. Scale-
up of mixing in gassed multi-turbine agitated vessels. Canadian Journal
of Chemical Engineering 76, 398–404.
Vrabel, P., Van der lans, R., Cui, Y.Q., Luyben, K., 1999. Compartmental
model approach: mixing in large scale aerated reactors with multiple
impellers. Transactions of the Institution of Chemical Engineers 77
(A), 291–302.
Vrabel, P., Van der lans, R., Luyben, K., Boon, L., Nienow, A.W., 2000.
Mixing in large-scale vessels stirred with multiple radial or radial and
axial up-pumping impellers: modeling and measurements. Chemical
Engineering Science 55, 5881–5896.