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

Chemical Engineering Science 59 (2004) 3793 – 3805

www.elsevier.com/locate/ces

Mixing analysis in a tank stirred with Ekato IntermigJ impellers


E.S. Szalaia , P. Arratiaa , K. Johnsonb , F.J. Muzzioa;∗
a Department of Chemical and Biochemical Engineering, Rutgers University, 98 Brett Road, Piscataway NJ 08854-8058, USA
b Independent Consultant, 2884 Sutherland Circle NW, North Canton, OH 44720, USA

Received 4 October 2002; received in revised form 14 November 2003; accepted 1 December 2003

Abstract

The mixing performance of a batch stirred tank with four Ekato IntermigJ impellers is investigated in this paper by experimental
and computational methods. We considered three impeller speeds corresponding to Reynolds numbers 37, 50 and 100, all in the laminar
regime. For the purposes of model development and 9ow validation, Newtonian rheology is assumed, where the 9uid density and viscosity
is set to 1247:6 kg=m3 and 0:4 kg=(m s), respectively. The computed velocity ;eld and mixing patterns are validated using Particle Image
Velocimetry, acid-base visualization experiments and Planar Laser-Induced Fluorescence. All three techniques reveal excellent agreements
between the experiments and computations. Also, detailed Lagrangian analysis of mixing, using particle tracking and stretching simulations,
is presented for two 9owrates in the laminar regime. It is shown that severe compartmentalization exists in the vessel and transport in the
axial direction is very slow. Characterization of local micromixing intensities is presented by computing the distribution of intermaterial
area density and striation thickness distribution (STD) from the stretching ;eld. It is found that the STDs at both 9owrates are identical
despite signi;cant di@erences in the stretching ;eld, suggesting that at low stirring rates micromixing performance is independent of
agitation speed.
? 2004 Published by Elsevier Ltd.

Keywords: Mixing; Fluid mechanics; Microstructure; Homogenisation; Numerical analysis; Visualization

1. Introduction 9ow behavior often causes a decrease in productivity. The


degree of agitation at all levels must be suBcient to elimi-
One of the most challenging tasks in the chemical, phar- nate dead zones, which could become anaerobic, break up
maceutical and biochemical industry is the design and and distribute air bubbles from the sparger and accomplish
scale-up of reactors, when highly viscous 9uids are pro- satisfactory bulk mass and heat transfer. Which one of these
cessed. Commercial applications include antibiotic and processes becomes the most outstanding challenge usually
polysaccharide fermentations, vaccine and glycoprotein depends on the scale of the operation, the exact rheological
production, or distributing additives in creams, pastes and properties of the broth and the sensitivity of the microor-
paint. The formation of persistent slow-mixing regions at ganisms to shear and oxygen levels.
low impeller speed, or the diBculty of handling materials The mixing performance of stirred tanks equipped with
with complex rheological behavior often excludes most tra- various types of agitators, such as Rushton or piched-blade
ditional scale-up methods from the list of engineering tools turbines, has been considered by many researchers (Nienow
designers can use. Handling highly viscous 9uids presents and Elson, 1988; Nienow, 1990; Amanullah et al., 1998,
an interesting problem from a mixing viewpoint, because 2001). Recently high 9ow and low power number impellers,
poor bulk mixing, low oxygen mass transfer rates, and in- such as the Intermig and axial hydrofoil impellers, have
homogeneous distribution of substrate are inherent in these became prevalent (Galindo and Nienow, 1992; Bujalski
applications. While the intrinsic rate of biological reactions et al., 1999; Ibrahim and Nienow, 1999; Houcine et al.,
remains virtually independent of vessel size, the hydro- 2000). Most investigations have examined the mixing time
dynamic environment changes upon scale-up. Non-ideal and power consumption of agitation systems resulting in
highly speci;c empirical correlations (Mavros et al., 1996;
∗ Corresponding author. Tel.: +1-732-445-3357; fax: +1-732-445- Rutherford et al., 1996; Esponosa-Solares et al., 1997;
6758. Armemante and Chang, 1998), but the micromixing prop-
E-mail address: muzzio@soemail.rutgers.edu (F.J. Muzzio). erties of most impeller types have not been considered.

0009-2509/$ - see front matter ? 2004 Published by Elsevier Ltd.


doi:10.1016/j.ces.2003.12.033
3794 E.S. Szalai et al. / Chemical Engineering Science 59 (2004) 3793 – 3805

In this paper we focus on the performance of the Ekato the laminar regime. For the purposes of model development
Intermig impellers, considering both the 9ow and mixing and 9ow validation, Newtonian rheology is assumed in this
properties of these low-shear agitators. These impellers are work. The 9uid density and viscosity is set to 1247:6 kg=m3
usually installed in pairs, with each agitator rotated at a and 0:4 kg=(m s), respectively. These values are similar to
90◦ angle with respect to its neighbors. The two pumping the physical properties of a glycerine-water mixture at room
directions of the center and head portions of the Intermig temperature. No-slip boundary conditions were assumed for
is signi;cant for oxygen distribution, but 9ooding occurs the simulations for all 9uid–solid interfaces. The vessel is
more often at a particular aeration rate if the center impeller open to the atmosphere at the top 9uid–9uid interface, so a
pumps downwards (Nienow, 1990). It is suggested that the zero stress boundary condition was applied to that surface.
Intermigs produce a complex 9ow pattern that bears little re-
semblance to the simpli;ed version given by manufacturers
2.1. Solver technology and computational methods
(Ibrahim and Nienow, 1995). In this paper we examine the
performance of a fermenter equipped with four Ekato Inter-
The ORCA CFDJ package is used in the computational
migs by experimental and computational techniques. These
analysis of 9ow and mixing behavior. Details of the compu-
methods and the exact geometry is described in Section 2.
tational methods are given elsewhere (Shakib et al., 1991;
A detailed analysis of the 9ow ;eld and mixing patterns un-
Szalai and Muzzio, 2003; Zalc et al., 2002). Only a brief
der laminar conditions is provided in Section 3. In this sec-
summary is included here for the sake of brevity. The model
tion we also compute mixing intensity based on stretching
of the Intermig impeller geometry is custom-built, accord-
computations (Muzzio and Swanson, 1991) and discuss a
ing to the exact speci;cations by Ekato, and assembled in
quantitative measure for micromixing eBciency.
a CAD-type modular unit to facilitate user-friendly appli-
cations. Numerical solutions for the coupled velocity and
pressure ;eld are computed at the 1.4 million nodal loca-
2. Geometry and methods tions in the three dimensional (3D), unstructured mesh of
the geometry. The solver technology in the ORCA pack-
Flow and mixing is investigated in a stirred tank geometry age is based on a stable iterative method with Galerkin
that is commonly used as a reactor, fermenter or crystallizer. ;nite-element formulations. Similarly to other CFD solvers,
In order to analyze the hydrodynamical environment and it requires a maximum convergence tolerance and forces
mixing performance in the tank by experimental methods, the di@erence between successive iterations below the pre-
a laboratory-scale vessel is built from plexiglass according set limit. In the present analysis, the steady-state 9ow solu-
to the dimensions in Fig. 1(a). An identical computational tions converged below the 10−4 tolerance in seven iterative
model of the geometry is created using the ORCA CFDJ steps in the four-impeller vessel. The divergence-free con-
software package (Dantec Dynamics, Mahwah, NJ), which straint of time-independent 9ows was enforced for all three
consists of 3.7 million tetrahedral volume elements and 1.1 9owrates. The total memory requirement for each solution
million nodes. The tank diameter, T is set to 0:3048 m and was 1:36 Mbytes on a single workstation (Sun Solaris ul-
the height to diameter ratio (H=T ) is 2.75. The vessel is tra60 processor).
agitated by four Ekato IntermigJ impellers, which are con- The nodal values of velocity and pressure ;eld serve as
structed following the manufacturer’s standards. For an en- the starting point for mixing analysis by Lagrangian parti-
larged view, see Fig. 1(b). This type of impeller is usually cle tracking calculations. Particle trajectories are integrated
installed in pairs, where each one is rotated 90◦ relative to via a fourth order, adaptive step-size routine following the
its neighbors. The schematic drawing indicates the position velocity vector ;eld:
of four impellers, spaced at equal distances (7 apart) from
one another and the tank bottom. Subsequently, we number dx
= v(x):
N (2)
each impeller as #1 starting at the tank bottom to #4 at the dt
top. The round bottom of the vessel was simulated as an el-
Mixing processes promote contact between mixture compo-
lipse, where the radius of the major axis is the tank radius
nents, which means increasing the interfacial area available
(equal to 6 ) and the radius of the minor axis is 2 .
for transport of mass and energy and decreasing di@usional
Flow in stirred tanks is generally characterized by the
distances between various species. The increase in interfa-
Reynolds number (Re), de;ned as follows:
cial area that occurs as a result of the mixing process is mea-
ND2 sured by the elongation of small vectors attached to 9uid
Re = ; (1) particles (i.e. stretching) as they move through the 9ow. The

evolution and statistics of stretching in chaotic 9ows has
where N is the impeller speed in revolutions/s, D is the been exploited in detail (Ottino, 1980; Muzzio and Swanson,
impeller diameter,  is the 9uid density in kg=m3 and is 1991; Beigie et al., 1993; Liu et al., 1994), and some rela-
the 9uid viscosity in kg=(m s). Three 9ow conditions are tionships are summarized brie9y here. The elongation of an
examined in this investigation, Re = 37, 50 and 100, all in in;nitesimal vector l with initial length l0 attached to each
E.S. Szalai et al. / Chemical Engineering Science 59 (2004) 3793 – 3805 3795

Fig. 1. (a) The geometry of a tank stirred with four Ekato Intermig impellers. (b) An Ekato Intermig impeller. (c) Experimental velocity ;eld in the
tank, measured by Particle Image Velocimetry, in the region marked by the red box in (a). (d) The simulated 9ow ;eld is shown in the same region
for comparison.

9uid particle is computed by integrating (3): distributing a collection of vectors in the 9ow (as opposed
to a single vector) and measuring a ;eld of stretching in-
dl tensities as a function of position and time. The distribution
= (∇v)T · l (with the initial condition lt=0 = l0 ): (3)
dt of stretching intensities generated by the 9ow then provides
a quantitative assessment of mixing performance. The fre-
The overall length increase, or accumulated stretching (e.g. quency of  values can be computed as
strain), for a given particle trajectory is de;ned as

|l| 1 d N (log )
= : (4) H (log ) = ; (5)
|l0 | N d log 

Since the rate of accumulation of  is exponential, the evolu- where N is the total number of vectors used to compute the
tion of the stretching in chaotic 9ows can be best described stretching ;eld. The base 10 logarithm of  is adapted for
using a logarithmic scale. all results discussed in this paper.
Quantifying the eBciency of a mixing process requires Being able to measure and predict the rate of growth
resolving the emerging spatial structure of the mixture by of interface and its area coverage, or intermaterial area
3796 E.S. Szalai et al. / Chemical Engineering Science 59 (2004) 3793 – 3805

density, , is an important step towards understanding re- particles moved yields velocity information at all the particle
active processes. Intimate contact between mixture compo- locations simultaneously along the planar cross section. In
nents is necessary for chemical reactions and assures high the validation studies this technique was primarily used to
local rates of mass transfer. The spatial distribution of inter- measure velocity 9ow ;elds which were then compared to
material area is unique in every chaotic 9ow. However, in 9ow ;elds calculated by CFD.
all chaotic 9ows the interface grows exponentially:  ≈ PLIF is an instantaneous, whole-;eld measurement tech-
en . Here  represents the topological entropy of the 9ow, nique, where a 9uorescent dye in solution with one of the
which is the true “mixing rate” in the system, and n is the mixture components is excited with laser light. It reveals
period (n = 2 for one impeller revolution, when using In- the evolution of mixing patterns by dispersing a 9uores-
termigs). With the relationship =∼= (Muzzio et al., cent tracer 9uid in the system. Flow and mixing behavior,
2000), the spatial distribution of intermaterial contact area, concentration ;eld, and reaction rate information can be
scaled by the overall length of the interface , can be pre- measured from the intensity of 9uorescence between the
dicted directly from the stretching ;eld in a mixer: mixture components. A small amount of 9uorescent dye
H (log ) H (log ) 1 dN (log ) (Rhodamine) is injected to the tank and illuminated with a
= = : (6) laser light in the experiments. The spreading of the dye is
  NP  d log 
followed in the 9ow by taking a series of pictures, and the
The stretching ;eld can be formally related to the stria-
concentration is quanti;ed based on the intensity of 9uo-
tion thickness distribution (STD) in chaotic 9ows. Material
rescence. Experimental pictures are taken at various cross
;laments are stretched in one direction and simultaneously
sections in two dimensions, because folds and layers of the
compressed in another direction at the same rate, the striation
illuminated tracer 9uid at di@erent depths in the evolving
thickness is directly related to the stretching of a material
mixing structure would overlap in a 3D view and hide ;ne
;lament: s∼1=∼1=. As a consequence of this relationship,
details.
the time evolution, shape, and scaling of stretching distri-
A third experimental method, a visualization technique
butions in globally chaotic 9ows is also related to the same
based on an acid-base neutralization reaction was also used
properties of striation thickness distributions. Because new
to exploit the size and shape of segregated regions in the
striations are born in a chaotic 9ow after every passing of
mixing tank. This technique is based on adding a soluble,
the impeller, there is a proportionality, the intermaterial area
pH sensitive indicator to the 9uids to be mixed. The indica-
density (), that links the stretching ;eld with the number of
tor appears blue in basic environment and yellow in acidic
striations. Considering that the number of stretching values
media. When the mixture components come in contact dur-
is constant at each 9ow period, while the number of stri-
ing the mixing process, the neutralization reaction instanta-
ations increases (i.e. (Ns )∼), the striation thickness
neously proceeds at the interface and a color change occurs.
distribution, as predicted from the stretching ;eld, follows:
In slow-mixing regions, or islands, the 9uid appears blue for

H (log s) = H (log ): (7) long times, because there is not enough acidic component
 for complete neutralization. The reaction rate is limited there
The real power of this relationship is that while H (log s) by the very slow rate of di@usive transport mechanisms.
is unattainable directly in most 9ows, computing H (log )
is fairly straight-forward. It is shown in this paper that the 3. Results
STD can be predicted in complex 3D 9ows based on the
stretching ;eld and Eq. (7). 3.1. Flow patterns and experimental validation

2.2. Experimental methods Fig. 1(c) and (d) shows the comparison of the com-
puted and experimentally measured 9ow ;eld at Re = 37.
The computational techniques used in this study do not The experimental velocity vectors were measured in a plane
require any experimental input for a 9ow solution, but it is aligned with one of the impeller blades using PIV and col-
always necessary to validate the numerical results. For that ored by increasing magnitude from dark blue to yellow. In
purpose, three non-intrusive optical techniques, Particle Im- order to illustrate the agreement between the computed and
age Velocimetry (PIV), acid-base 9ow visualization exper- experimental 9ow patterns, a section of the mixing vessel
iments and Planar Laser-Induced Fluorescence (PLIF) are framed by a red box in Fig. 1 is enlarged for visual clar-
used to measure the experimental 9ow patterns and compare ity. The velocity measurements indicate that a complex, 3D
with the numerical solutions. 9ow pattern forms in the vessel that creates strong varia-
PIV yields velocity vectors at multiple locations with a tions in the azimuthal direction, so phase locking between
single experiment. It determines the linear displacement of the experimental and computational measurements is criti-
particles suspended in the 9ow. After seeding the 9uid with cal. Fluid motion is dominated by 9ow created at the tip of
tracer particles, a laser is used to illuminate a cross-section of each impeller blade. The tip of the ;rst impeller creates a
the 9ow and two pictures are taken in a short time. Knowing strong axial 9ow, moving 9uid toward the second impeller.
the time interval between the two pictures and the distance An S shape bulk pattern is observed between these two
E.S. Szalai et al. / Chemical Engineering Science 59 (2004) 3793 – 3805 3797

Fig. 2. The probability density function of the velocity is computed from


the experimentally measured and the simulated 9ow ;eld. The statistical
comparison is excellent in both (a) the X direction and (b) the Y direction.

Fig. 3. The velocity ;eld, measured in a cross-sectional plane aligned with


impellers, as the 9uid reaches wake of the second impeller two of the impeller blades, is color-coded by magnitude (a) at Re = 50
and (b) at 100 in the tank.
and is ejected toward the wall by the strong radial 9ow in
that region. A recirculation loop just below the plane of the
second impeller is seen in this section. Overall, the computed
result compares well with the experimentally measured 9ow information in Fig. 2 is that the accuracy of the computed
despite the complexity of the convective motion created by velocity values (red) is reasonable compared to the experi-
the impellers in this section. Surprisingly, the long, pitched mental measurements (black) for both the x (Fig. 2(a)) and
blades closer to the hub has minimal e@ect on axial 9ow and y component (Fig. 2(b)) despite the complex 9ow pattern.
mixing in the vessel. Our observation, that much stronger This comparison establishes a high level of con;dence in the
radial, rather than axial 9ow is created by each impeller, is computational methods and these techniques in turn can be
consistent with results of other investigations ( Ibrahim and used to describe experimentally unattainable characteristics
Nienow, 1995) concluding that a strong radial 9ow develops of the mixing process.
with a small downward discharge angle below Re = 170. The computational velocity ;eld for two 9owrates, Re=50
A quantitative comparison of the experimental and com- and 100, is shown in Fig. 3. Each ;gure shows half of the
putational velocity ;elds can be made by computing the tank cross section, where the shaft is on the left and two
probability density function (pdf) of the velocity compo- of the impellers are aligned with the vertical cross-sectional
nents vX and vY : view. The impeller hubs are shown to mark the location of
  all four impellers. There are large variations from low to
1 dN (va )
pdf (va ) = ; (8) high velocities in the 9ow and a non-intuitive 9ow pattern
Np dva is revealed in the tank cross-section. Several recirculation
where dN (va ) is the number of vectors that have veloc- loops exist in the vessel, one above and one below each im-
ity values between va and (va + dva ). The parameter ‘a’ in peller blade. A large recirculation loop ;lls the bottom por-
Eq. (8) represents x or y, and Np is the total number of ve- tion of the tank. Fluid in this region passes through impeller
locity vectors. The physical meaning of pdf (va ) is the rel- #1, it is pushed towards the vessel wall and returns towards
ative distribution of velocities from smallest to largest val- the shaft.
ues at a selected location. Fig. 2 shows the pdf (vX ) and Above each impeller aligned with the vessel cross-section
the pdf (vY ) at a vertical cross-sectional plane in the tank strong axial 9ow is created towards the tank top, forming an
aligned with two of the impeller blades. The most important S shape bulk 9ow pattern. As the impeller passes through the
3798 E.S. Szalai et al. / Chemical Engineering Science 59 (2004) 3793 – 3805

plane, the 9ow direction reverses in the same region towards


the bottom of the vessel. This behavior is also observed as
the Re number is increased to 100 (Fig. 3(b)). The overall
9ow pattern is almost identical to the one at Re = 50, but
more intense gradients are apparent at the higher Re number
case, due to the increased amount of inertia at the larger
agitation rate.
After establishing a high level of con;dence in the com-
putational techniques, the following sections focus on the
evolution of mixing structures. Because the 9ow ;eld is
fully 3D, it is inappropriate to conclude that the recircula-
tion loops represent segregated regions in the tank without
performing a Lagrangian mixing analysis.

3.2. Short-term mixing structures

Recently, it has been shown that the chaotic mixing pro-


cess in stirred tanks is driven by small 9ow perturbations
caused by the passing of the impeller blades (Zalc et al.,
2002). In fact, no laminar mixing is possible if the impeller
blades are removed and replaced by concentric disks, be-
cause at low Reynolds numbers the 9ow is steady and 2D.
In such 9ows, particle trajectories follow closed streamlines
and simply trace out nested tori. There is no convective
exchange of material among these tori, hence mixing can
occur only by slow di@usive transport. When 9uid is agi-
tated by impellers, the passage of 9uid elements through the
near-blade regions brings about stretching and reorientation. Fig. 4. The mixing structure after (a) 10 and (b) 20 impeller revolutions
The 9uid elements must move towards the wall to allow at Re = 50.
for the passing of the impeller blades. This mechanism of
stretching and reorientation is illustrated by the computed segregation of the red and blue colored tracer 9uid in the
short-term mixing structures in Fig. 4. top and the bottom of the tank, respectively. Red tracer 9uid
Since particles located near the shaft tend to be ejected was injected only in the top region and it remains there
outward radially by the impellers, 200,000 tracer particles after 10 impeller revolutions (Fig. 4(a)). As the impellers
are injected initially along the shaft in a vertical line extend- rotate for another 10 revolutions, more 9uid striations are
ing along the height of the tank. Radial mixing e@ects are created around the blades, but the axial mixing barrier at the
most noticeable because tracer particles are initially placed mid-height of the tank is not penetrated.
in regions where they quickly pass through the blade section The e@ect of poor axial mixing persistently remains as the
of the impellers. When distributing additives, this region is agitation speed is increased to Re=100 in Figs. 5(a) and (b).
best for injections. The position of particles is recorded after The computed mixing patterns after 20 and 40 impeller rev-
every 10 impeller revolutions (20 9ow periods). The com- olutions are shown at the higher agitation rate corresponding
puted mixing structures at Re = 50 are shown in Figs. 4(a) to 42 RPM. Fluid tracer particles are color-coded similarly
and (b) after 10 and 20 revolutions, respectively. Fluid tracer to the Re = 50 case (tracer particles initially injected in the
particles initially injected in the top half of the vessel are top half of the vessel are colored red and tracers initially
colored red and tracers initially placed in the bottom half are placed in the bottom half are blue in the ;gure). The two
colored blue in the ;gure. In Fig. 4(a) particles are ejected 9uids remain almost completely segregated as they spread
along the center of each impeller towards the tank wall and throughout the chambers around each impeller. In Fig. 5(a),
then invade the top and bottom section of the compartments no red dye visits the bottom half of the vessel and conversely
surrounding each impeller. As particles reach the tank wall, no blue dye enters the top region. After 20 revolutions some
the stream splits and feeds the top and bottom part of each particles are entrained along the shaft, but there is essen-
chamber returning toward the shaft to be re-injected along tially no convective mixing across the separation plane at the
the centerline again. mid-height of the tank. Such severe compartmentalization in
Severe axial mixing limitations are immediately apparent the 9ow can cause signi;cant mixing limitations. Transport
from the mixing patterns at Re = 50 in Figs. 4(a) and (b). across this permanent boundary only occurs via di@usion,
The e@ect of poor axial mixing is revealed by the complete which is generally too slow to assure complete homogeneity
E.S. Szalai et al. / Chemical Engineering Science 59 (2004) 3793 – 3805 3799

mixing patterns after 20 revolutions can be compared for


two agitation rates, Re = 50 and 100, respectively. Examine
the region below impeller #1, where a large recirculation
loop forms a slow-mixing region. The short-time evolution
of the mixing patterns indicate that even though segregated
regions are small after long times, the 9uid surrounding
these regions mixes at a very slow rate. Colored 9uid tracer
particles invade the star-shaped islands very slowly and this
e@ect increases at larger rates of agitation. The size of the
slow-mixing region devoid of dye in Fig. 4(b) after 20 rev-
olutions is signi;cantly smaller than in 5(a) after the same
number of impeller rotations. However, 20 revolutions at
Re = 100 corresponds to half of the processing time as for
Re = 50. Figs. 4(b) and 5(b) indicate mixing after the same
amount of time (1 min) for the two agitation rates. The same
observation can be made about the e@ect of the slow-mixing
regions as before: mixing in the Re = 100 case is a@ected
more signi;cantly. The size of the unmixed region in 5(b)
is approximately double compared to 4(b).
An acid-base visualization experiment, as described in
Section 2, was carried out to con;rm the computational mix-
ing analysis and examine axial mixing in the tank. Figs. 6(a)
–(f) shows a series of experimental photographs, where the
tank is originally ;lled with a viscous glycerine-water mix-
ture that contains NaOH base and Bromothymol blue pH
indicator. The impeller speed in these experiments corre-
sponds to 21 RPM, the same as in Figs. 4(a) and (b). The
entire vessel initially appears blue, because the pH is slightly
above 7. A small amount of acidic 9uid is then injected
Fig. 5. The mixing structure after (a) 20 and (b) 40 impeller revolutions near the top of the vessel at the start of the experiment. As
at Re = 100. the mixing process evolves and the two 9uids come to in-
timate contact, the neutralization reaction changes the pH.
in most practical applications. As a consequence, large vari- Thus, the 9uid appears yellow in all well-mixed areas, and
ations in yield or 9uctuations in eSuent product and/or waste remains blue in regions where acidic 9uid is yet not trans-
concentrations may be observed. Moving towards the top of ported. Fig. 6(a) reveals that acidic 9uid invades the cham-
the vessel, some very small segregated regions are also no- ber around the top impeller and indicates the location of the
ticeable along the shaft. Due to their small size, these islands separation plane and the doughnut-shaped, slow-mixing re-
do not signi;cantly a@ect mixing performance. gions above and below the impeller. After 15 min Fig. 6(c),
The mixing patterns in Figs. 4 and 5 reveal that a pro- the top compartment is almost completely mixed, but only a
nounced structure of lobes is created as a result of the peri- very small amount of acidic 9uid gets to the region around
odic stretching and folding caused by the impellers. These the second impeller from the top. In an additional 15 min
lobes gradually wrap around a slow-mixing region at the most of the top half of the vessel is well-mixed, except for
center of each chamber. The overall appearance of the mix- two slow-mixing regions around the second impeller. After
ing pattern from 10 to 20 revolutions (Figs. 4(a) and (b)) 90 min the slow-mixing regions in the top are almost com-
remains similar, only more folds appear in fast-mixing areas. pletely destroyed, but there is a clear boundary to convective
Self-similarity, the evolution of mixing patterns towards an transport between the top two and the bottom two impellers.
asymptotic structure, is characteristic of all periodic chaotic A small discoloration of the third chamber from the top in-
9ows. An underlying ;eld of orientations controls the struc- dicates that some 9uid is slowly transported to the bottom
ture of 9uid striations at each location in the 9ow. The orien- along the shaft, which is consistent with the results of the
tation process, controlled by a property named Asymptotic computations in Fig. 4. Clearly, there is a severe limitation
Directionality, is very fast and the dependence on initial in- in axial mixing, despite indications of axial 9ow created by
jection location is lost exponentially fast (Giona et al., 1998; the Intermig impellers.
Muzzio et al., 2000). Almost all mixing tanks are designed with the shaft
Recirculating regions create multiple slow-mixing regions placed at the center of the vessel and agitators usually have a
in the 9ow above and below each impeller. Figs. 4(b) and plane of symmetry. The symmetric con;guration is the least
5(a) reveal the e@ect of these slow-mixing regions, where e@ective from a theoretical standpoint, because it promotes
3800 E.S. Szalai et al. / Chemical Engineering Science 59 (2004) 3793 – 3805

Fig. 6. Mixing patterns captured by an acid-base experiment using constant agitation speed at Re = 37.

Fig. 7. Mixing patterns captured by an acid-base experiment using variable agitation speed.

segregation and creates time-periodic conditions. One ap- the agitation rate is alternated between 21 and 42 RPM
proach to overcome the limitations of such geometries every 1 min. Similarly to the previous experiment shown
is to vary the speed of the impellers. It has been shown in Fig. 6, the tank is initially ;lled with a mixture of glyc-
that systematically alternating the agitation rate destroys erine and water containing base and bromothymol blue
slow-mixing regions below and above a single Rushton pH indicator. After the acidic 9uid is injected near the
impeller (Lamberto et al., 1996). In this paper this tech- top, the impeller speed is systematically increased and de-
nique is applied to perturb the separation plane at the creased every 5 min. Using the variable speed technique,
mid-height of the vessel and promote rapid and eBcient the top-bottom segregation at the center of the tank is
mixing. The results of variable speed agitation carried out completely eliminated in 30 min (Fig. 7(d)). In contrast to
in the tank with Intermig impellers is indicated in Fig. 7, Fig. 6(d), where top to bottom segregation persists after
where Figs. 7(a)–(d) shows a series of photographs as 90 min when the agitator speed is constant, Fig. 7 indicates
E.S. Szalai et al. / Chemical Engineering Science 59 (2004) 3793 – 3805 3801

Fig. 8. Comparison of (a) PLIF and (b) computed 9ow structures below an Intermig impeller.

that the variable speed technique e@ectively destroys axial suBcient area for di@usional transport. As 9uid ;laments
segregation. are stretched and folded by the 9ow, they gradually deform,
Next, the mixing behavior in a small section of the tank increase in length and simultaneously decrease in width.
is examined in detail. A very complex, ;lamented mix- The length increase which the 9uid ;laments experience
ing structure is created by the impellers. The details of the can be measured by stretching computations. The stretching
mixing pattern around one of the impellers is enlarged in ;eld captures the increase in the contact area between 9uid
Figs. 8(a) and (b). The experimental mixing pattern in Fig. ;laments during the mixing process (Muzzio and Swanson,
8(a) is captured by LIF measurements, described earlier in 1991; Szalai and Muzzio, 2003).
Section 2. In these experiments 9uorescent rhodamine dye The stretching of material ;laments was computed in the
is injected initially at the tip of the impeller blades and a tank by placing 200,000 small vectors in the 9ow and inte-
laser is used to illuminate a cross-section of the vessel near grating Eqs. (2) and (3) simultaneously. Fig. 9 shows the
the center. As mixing proceeds, a series of pictures are taken stretching ;eld in the tank at Re = 50. The colors from blue
by a CCD camera. Well-mixed areas appear yellow and the to red represent the logarithm of the stretching as a function
intensity of 9uorescence is an indication of mixing inten- of initial position, from low to high intensities. After 10 rev-
sity in each region of the 9ow. The computational mixing olutions in (a), a complex, highly inhomogeneous stretching
pattern in Fig. 8(b) appears almost identical to the experi- ;eld emerges. The slow-mixing regions observed in the mix-
mental mixing structure. Hence, the physics of 9uid motion ing patterns (Fig. 4) coincide with the blue, low-stretching
is captured correctly in the computational model. Addition- regions in Fig. 9(a). It is important to note that the color as-
ally, both pictures reveal the formation of lobes near the tip signment is based on a logarithmic scale, so there is multiple
of the impeller that travel towards the tank wall and then orders of magnitude di@erence between the high stretching
wrap around the slow-mixing region below the impeller. The (red) and low stretching (blue) regions. A very inhomo-
periodic passing of the impeller blades creates new folds geneous intensity ;eld is not bene;cial for shear-sensitive
each time and the 9ow stretches and folds the already exist- materials, such as mammalian cell cultures. As cells visit
ing ones. A similar mixing mechanism has been observed di@erent regions of the 9ow, they experience orders of
when using Rushton impellers (Zalc et al., 2002), where the magnitude changes in shear and can su@er signi;cant
perturbations created by the Rushton impeller blades were damage.
signi;cantly smaller in size. These folds are created as the The stretching ;eld for Re = 100 is shown in Fig. 9(b).
impeller blades periodically increase the intermaterial con- Although the distribution of high- and low-intensity regions
tact area between two 9uids. The more ;lamented the struc- is somewhat di@erent for this 9owrate than for Re=50, some
ture, the more area is available for di@usion to occur and important characteristics remain unchanged. The stretching
ultimately complete the homogenization on the micromix- ;eld in Fig. 9(b) is also very inhomogeneous, and high-shear
ing scale. The folds created by the Intermigs invade a larger regions appear at the plane of every impeller and at the
region of the 9ow compared to other impeller types such mid-plane between each blade. The ;eld of mixing inten-
as the Rushton or the pitched-blade (Lamberto et al., 1999; sities measured by stretching represents the micromixing
Alvarez et al., 2002). properties of the 9ow for the duration of the mixing process.
The spatial inhomogeneities remain unchanged and do not
disappear over time. As shown in Figs. 4 and 5, a character-
3.3. The stretching :eld istic property of chaotic 9ows is that contact area between
mixture components always accumulates at the same spa-
The ultimate goal of a mixing application is to increase the tial locations. Because there is an invariant ;eld of orien-
intimacy of contact between mixture components and create tations dictating how 9uid striations form and orient in the
3802 E.S. Szalai et al. / Chemical Engineering Science 59 (2004) 3793 – 3805

Fig. 9. The stretching ;eld (), color-coded by the magnitude of the ac-
cumulated stretching on a logarithmic scale, is shown after 20 revolutions
(a) at Re = 50 and (b) at Re = 100.

9ow, the intensity ;eld as well as the mixing patterns re-


main self-similar. Thus, it is suBcient to describe the local
mixing intensities at one instant in time to fully characterize
the entire mixing process.
This information can be used to identify areas in the tank
with high local intensity of mixing. In other words, e@ective
injection locations can be identi;ed by the high-stretching,
red areas in the tank. For instance, if a small amount of ad- Fig. 10. The logarithm of the probability density function of log(),
ditive needs to be distributed relatively fast in the mixing (log(H (log())), at (a) Re = 50 and (b) at Re = 100.
vessel, areas near the tip of each blade are recommended
for eBcient mixing. Blue, low-stretching areas are very
slow-mixing regions, as seen earlier in the mixing patterns is the total number of stretch vectors. In physical terms,
in Figs. 4 and 5. If nutrients are added here for distribu- Hn (log()) represents the spectrum of intensities in the mi-
tion, they remain trapped inside these regions for long times. cromixing process. Fig. 10(a) shows plots of log(H (log()))
Only di@usion transports material from these low stretching versus log() for the Re=50 9ow after 10, 20, 30, 40, 50 and
regions to the bulk 9uid. 60 impeller revolutions. The distribution curves shift toward
It is instructive to look at the distribution of mixing in- higher stretching values as the total number of revolutions
tensities in a quantitative manner by computing the pdf of increases. As stretching accumulates, the curves broaden,
the natural logarithm of the stretching values, H (log()): but retain a constant shape.
  Fig. 10(b) shows similar results for Re =100. In all cases,
1 dN (log )
H (log ) = ; (9) the distributions exhibit multiple peaks (with mean values in
Np d log  the low and high stretching regions). The distinct peaks near
where dN (log()) is the number of vectors that have stretch- log()–0 for each Re number represent the stretching expe-
ing values between log() and log() + d log() and Np rienced by ;laments initially placed within the segregated
E.S. Szalai et al. / Chemical Engineering Science 59 (2004) 3793 – 3805 3803

regions in the 9ow. These peaks of low stretching values The validity of Eq. (13) was demonstrated recently (Alvarez
remain virtually ;xed as the number of impeller revolutions et al., 1998), where the micromixing structure H (log s) in
increases, because the stretching inside slow-mixing islands a simple theoretical 9ow was computed directly and also
increases at a slow linear rate, if at all. obtained from stretching computations. More than 109 tracer
particles and over 106 stretching values were required to
3.4. Micromixing intensities compute the STD directly and compare it with the prediction
of Eq. (13). Such a level of numerical resolution is usually
The thickness of 9uid ;laments is the most fundamental impossible and almost always impractical when dealing with
measure of the di@usive length scale in the mixture. This realistic 9ow systems. Eq. (13) cannot be applied directly
is the distance 9uid molecules must ultimately travel to to 3D systems such as the example studied here, for several
complete the homogenization process. However, this mea- reasons. First, the value of  is very noisy in 3D systems,
sure of local micromixing intensity cannot be assessed by because of limited statistical resolution. Furthermore,  is the
experimental techniques, because the average thickness of statistical weighting function for H (log s), and it is mainly
9uid striations rapidly decreases below micron size in the determined by the high end tail of H (log ), which is also
course of a mixing process. The average length scale rapidly very noisy for small data sets.
falls below experimental resolution in complex 9ows. The Fortunately additional theoretical developments allow one
scaling properties of stretching discussed above can be ex- to circumvent the need for direct computation, simply by
tended to generate useful predictions of micromixing in- recognizing that the high  region of H (log ) has the func-
timacy, and compute the striation thickness. Clearly, the tional form:
same iterative stretching-and-folding process that generates 
self-similar stretching distributions drives the evolution of H (log ) ≈ (log  )−1 H (ŵ)e(w−ŵ) =c ; (14)
striations, and therefore controls the dynamics of Striation
Thickness Distributions (STDs). where ŵ is the mode of w, and H (ŵ) is the value of H (w)
STDs could be computed directly from distributions of at the mode. The constants C and  can be easily deter-
stretching values by realizing that as a portion of 9uid is mined from a linear ;t of a double-logarithmic plot of
stretched, it generates striations with a thickness (s) in- the H (log ) data shown earlier in Fig. 10. As shown in
versely proportional to the amount of stretching () applied Fig. 11(a) for Re = 50 and in Fig. 11(b) for Re = 100
to the 9uid (Muzzio and Swanson, 1991). Moreover, the Eq. (14) accurately captures the high-stretching tail of
number of such striations in a portion of 9uid (dN (s)), which H (log ) for several decades of probability. Interestingly,
is proportional to the amount of contact area , is also pro- the ;t improves for larger values of n as the number of
portional to the stretching: impeller revolutions increases; more and more of the dis-
tribution tail converges toward the expected theoretical
s∼1=; (10a) behavior. Somewhat surprisingly, the ;t is better for the
higher value of Re, suggesting that, as intuitively expected,
ds∼ − 2 dN (); (10b)
the strength of correlations for high stretching values is
dN (s)∼ − dN (); (10c) weaker at higher Reynolds numbers.
Subsequently, H (log s) can be predicted by inserting
where the minus sign in Eq. (10c) re9ects that a list or- Eq. (14) into Eq. (13), i.e., using the ;t, rather than the
dered in increasing magnitude of s results in a list ordered original data. H (log s) vs. (−log s) calculated by assuming
in decreasing order of . Eq. (10) can be simply re-stated so ≈ 1 (the size of the unmixed region initially is assumed
in terms of the logarithms of s and  as to be of the same order of magnitude as the entire system)
d(log s)∼ − d(log ); (11a) is shown by the dashed curves in Fig. 11(a) for Re = 50 and
in Fig. 11(b) for Re = 100, n = 20 and 60. Since  acts as
dN (log s)∼ − dN (log ): (11b) a statistical weight on H (log ), H (log s) shifts to the right
of H (log ). In other words, the right branch of H (log )
Finally, one needs to take into account the fact that while
generates essentially the entire distribution H (log s), and
the number of stretching values remains constant with n, the
the mode of H (log s) corresponds to stretching values far
number of striations, Ns , increases as
along the high stretching tail of H (log ). This essentially
Ns ∼∼: (12) means that from the narrow viewpoint of generating inter-
In principle, Eqs. (11) and (12) are everything one needs to material surface for transport, the entire left branch of the
predict the striation thickness distribution: stretching distribution is completely irrelevant.
1 dN (log s) 1  d N log ) This observation has deep physical meaning. Fig. 11
H (log s) = ∼ demonstrates that the overwhelming majority of striations,
Ns d(log s)  d(log )
corresponding to most of the interfacial contact area in the
 tank, is generated by a set of very rare, but very intense
= H (log ): (13) stretching events.

3804 E.S. Szalai et al. / Chemical Engineering Science 59 (2004) 3793 – 3805

0 1.0 0
n
-2

log(H(log(s))
-4
-6
-2
-8
10-2
log(H(log(λ))), log(H(log(s)))

-10
-12
-4 -10 10 30 50 70

log(H(w))
-log(s)

-6 10-4

-8
10-6
-10

-12 10-8
-10 10 30 50 70 -8 -6 -4 -2 0 2 4 6 8 10 12
(a) log(λ), -log(s) (a) log(w)

0 1.0 0
-2 n

log(H(log(s))
-1 -4
-6
-2 10-2 -8
-10
log(H(log(λ))), log(H(log(s)))

-3 log(H(w)) -12
-10 10 30 50
-log(s)
-4
10-4
-5

-6
10-6
-7

-8

-9 10-8
-10 0 10 20 30 40 50 60 -8 -6 -4 -2 0 2 4 6 8 10 12
(b) log(λ), -log(s) (b) log(w)

n=20, H(log λ), computed Fig. 12. Main panel: The scaled striation thickness distributions are in-
n=20, H(log λ), predicted variant after 20 periods for both (a) Re = 50 and (b) Re = 100, indicating
n=20, H(log s), predicted that micromixing intensity is independent of agitation rate at low stirring
n=60, H(log λ), computed rates. Inset: log(H (log s)) for n = 20, 40, 60, 80, 100 and 120.
n=60, H(log λ), predicted
n=60, H(log s), predicted

Fig. 11. log(H (log )) for n = 20 ( ) and 60 ( ). The right branch 4. Conclusions
for n = 20 (—) and 60 ( ) is captured by Eq. (14). log(H (log(s)))
predicted from Eq. (13) for n = 20 (....) and 60 ( ). The objective of this paper was to analyze 9ow and mixing
in a vessel equipped with Ekato Intermig impellers. We
examined the 9ow ;eld and mixing structures for two cases
in the laminar regime. We showed that despite the complex
9ow ;eld created by the impellers, homogeneous mixing is
The evolution of H (log s), and its scaling properties are not achieved even after several hours of processing time if
illustrated in Fig. 12(a) for Re = 50 and in Fig. 12(b) for highly viscous materials are used.
Re = 100, n = 20–120. The inset shows H (log s) in each The computational results for the 9ow ;eld were vali-
;gure and the main panel shows the same data after the dated by comparisons to experimentally measured values.
distributions have been re-scaled by standardization. As re- A laboratory-scale prototype of the industrial mixing vessel
vealed in Fig. 12(a) for Re = 50, the shape of the distribu- was constructed from clear acrylic and the velocity ;eld at
tion rapidly approaches a self-similar distribution, where the Re = 37 was measured by Particle Imaging Velocitmetry.
curves increasingly overlap one another as n increases. Ac- Very good agreement was obtained between the experimen-
cording to Fig. 12(b) the convergence to a self-similar STD tal and simulated 9ow ;elds.
is also fast for Re=100, where the distributions overlap over Once the computational tools are validated by experi-
three decades of probability for n ¿ 20, and become indis- ments, they can be used to evaluate mixing performance and
tinguishable for over more than ;ve decades of magnitude. predict macro- and micromixing eBciency. In this paper we
E.S. Szalai et al. / Chemical Engineering Science 59 (2004) 3793 – 3805 3805

applied computational techniques to examine the evolution Galindo, E., Nienow, A.W., 1992. Mixing of highly viscous simulated
of mixing patterns and predict severe vertical segregation Xanthan fermentation broths with the Lightnin a-315 impeller.
Biotechnology Progress 8 (3), 233–239.
at low agitation rates. It was shown that a separation plane
Giona, M., Cerbelli, S., Muzzio, F.J., Adrover, A., 1998. Non-uniform
exists in the tank at mid-height from top to bottom, and this stationary measure properties of chaotic area-preserving dynamical
plane forms a signi;cant barrier to mixing in the axial direc- systems. Physica A 254, 451–465.
tion. Convective mixing is very slow at constant agitation Houcine, I., Plasari, E., David, R., 2000. E@ects of the stirred Tank’s
rate across the mid-plane. Design on power consumption and mixing time in liquid phase.
It was also shown that slow-mixing regions form around Chemical Engineering Technology 23 (7), 605.
Ibrahim, S., Nienow, A.W., 1995. Power curves and 9ow patterns for a
recirculating loops in the 9uid at low agitation rates. The range of impellers in Newtonian 9uids. Transactions of the Institution
shape, size and location of these islands is non-intuitive, of Chemical Engineers—London 73 (Part A), 485–491.
but can be accurately predicted by the computational mix- Ibrahim, S., Nienow, A.W., 1999. Comparing impeller performance for
ing patterns. The mixing structure at each impeller speed solid-suspension in the transitional 9ow regime with Newtonian 9uids.
evolves along a self-similar template that is controlled by Transactions of the Institution of Chemical Engineers 77 (Part A), 721.
Lamberto, D., Alvarez, M.M., Muzzio, F.J., 1999. Experimental and
an invariant ;eld of orientations. computational investigation of the laminar 9ow structure in a stirred
Understanding this property of laminar chaotic 9ows of- tank. Chemical Engineering Science 54 (7), 919–942.
fers a solution to the axial segregation problem. Varying the Lamberto, D.J., Muzzio, F.J., Swanson, P.D., 1996. Using time-dependent
agitation rate at periodic intervals of time breaks the vertical Rpm to enhance mixing in stirred vessels. Chemical Engineering
separation plane, because the mixing pattern changes at each Science 51 (5), 733–741.
Liu, M., Peskin, R.I., Muzzio, F.J., Leong, C.W., 1994. Structure of
di@erent agitation rate. Using the variable rpm technique, the stretching ;eld in chaotic cavity 9ows. A.I.Ch.E. Journal 40 (8),
convective mixing occurs in the axial direction signi;cantly 1273–1285.
faster than at a constant rate of agitation. Mavros, P., Xuereb, C., Bertrand, J., 1996. Determination of 3d 9ow ;elds
in agitated vessels by laser-doppler velocimetry: e@ect of impeller
type and liquid viscosity on liquid 9ow patterns. Transactions of the
References Institution of Chemical Engineers—London 74 (Part A), 658–668.
Muzzio, F.J., Swanson, P.D., 1991. The statistics of stretching and stirring
Alvarez, M.M., Muzzio, F.J., Cerbelli, S., Adrover, A., Giona, M., 1998. in chaotic 9ows. Physics of Fluids 3 (5), 822–834.
Self-similar spatiotemporal structure of intermaterial boundaries in Muzzio, F.J., Alvarez, M.M., Cerbelli, S., Giona, M., Adrover, A., 2000.
chaotic 9ows. Physical Review Letters 81 (16), 3395–3398. The intermaterial area density generated by time- and spatially periodic
Alvarez, M.M., Shinbrot, T., Zalc, J., Muzzio, F.J., 2002. Practical chaotic 2d chaotic 9ows. Chemical Engineering Science 55, 1497–1508.
mixing. Chemical Engineering Science 57 (17), 3749–3753. Nienow, A.W., 1990. Gas dispersion performance in fermenter operation.
Amanullah, A., Serrano-Carreon, L., Castro, B., Galindo, E., Nienow, Chemical Engineering Progress. February 61.
A.W., 1998. The in9uence of impeller type in pilot scale Xanthan Nienow, A.W., Elson, T.P., 1988. Aspects of mixing in rheologically
fermentations. Biotechnology and Bioengineering 57 (1), 95–108. complex 9uids. Chemical Engineering Research and Design 66, 5–15.
Amanullah, A., McFarlane, M., Emery, A.N., Nienow, A.W., 2001. Ottino, J.M., 1980. Lamellar mixing models for structured chemical
Scale-down model to simulate spatial pH variations in large-scale reactions and their relationship to statistical models; Macro- and
bioreactors. Biotechnology and Bioengineering 73 (5), 390. micromixing and the problem of averages. Chemical Engineering
Armemante, P.M., Chang, G., 1998. Power consumption in agitated Science 35 (6), 1377–1381.
vessels provided with multiple-disk turbines. Industrial Engineering Rutherford, K., Mahmoudi, M.S., Lee, K.C., Yianneskis, M., 1996. The
and Chemistry Research 37, 284–291. in9uence of rushton impeller blade and disk thickness on the mixing
Beigie, D., Leonard, A., Wiggins, S., 1993. Statistical relaxation characteristics of stirred vessels. Transactions of the Institution of
under nonturbulent chaotic 9ows: Non-gaussian high stretch tails of Chemical Engineers—London 74 (Part A), 369–378.
;nite-time Lyapunov exponent distributions. Physical Review Letters Shakib, F., Hughes, T.J.R., Johan, Z., 1991. A new ;nite element
70 (3), 275–278. formulation for computational 9uid dynamics: X. The compressible
Bujalski, W., Takenaka, K., Paolini, S., Jahoda, M., Paglianti, A., Euler and Navier–Stokes Equations. Computer Methods in Applied
Takahashi, K., Nienow, A.W., Etchells, A.W., 1999. Suspension and Mechanics and Engineering 89, 141–219.
liquid homogenization in high solids concentration stirred chemical Szalai, E.S., Muzzio, F.J., 2003. A fundamental approach to the design and
reactors. Transactions of the Institution of Chemical Engineers 77 (Part optimization of static mixers. A.I.Ch.E. Journal 49 (11), 2687–2699.
A), 241. Zalc, J.M., Szalai, E.S., Ja@er, S., Muzzio, F.J., 2002. Characterization
Esponosa-Solares, T., Brito-De La Fuente, E., Tecante, A., Tanguy, P.A., of 9ow and mixing in an SMX static mixer. A.I.Ch.E. Journal 48 (3),
1997. Power consumption of a dual turbine-helical ribbon impeller 427.
mixer in ungassed conditions. Chemical Engineering Science 67,
215–219.

You might also like