Validation of the ORCA CFD Software For

Three Stirred Tank Configurations:

Rushton Turbine Impellers, Pitched Bladed Impeller,
and the Ekato Interming Impeller
E.S. Szalai and F.J. Muzzio
Department of Chemical and Biochemical Engineering, Rutgers University
PO Box 909, Piscataway, NJ 08855

K.J. Bittorf*
Dantec Dynamics Inc., 777 Corporate Drive, Mahwah NJ, 07430
*Corresponding Author: e-mail [email protected]

Stirred tanks are widely used in chemical processing indu This document is comprised of five sections. The first two
tries for blending, liquid-liquid dispersion, gas-liquid disper- sections review the ORCA computational program and
sion, suspension formation, heat transfer, mass transfer and experimental techniques used to validate the results attained
chemical reactions. Efficient mixing is crucial to the out- from the ORCA CFD code. The last three sections provide
comes of all of these processes. For these applications, poor a comprehensive review of the validation studies completed.
mixing can have serious consequences such as excessive by-
product formation, inconsistent product quality in blended The location of dead zones or zones of unmixedness for
products, wide particle size distributions in crystallization three variations of stirred tank reactor systems are numeri-
processes, and slow mass transfer rates in bioreactors. cally modeled and experimentally validated for a stirred tank
operated in the laminar regime. This study reveals corre-
As tighter government regulations in the pharmaceutical sponding quantitative velocity fields, as well as correspon-
industry and lower margins on chemical products increase ding areas of unmixedness through Poincaré sections, in
the importance of optimizing these processes for quality both results. The comparison demonstrates sound qualita-
assurance and product enhancement, use of Computational tive and quantitative agreement between the experimental
Fluid Dynamics (CFD) in design and optimization becomes data and the CFD results.
more prevalent. CFD provides quick results for optimizing
and identifying deficiencies in processes, critically impor-
tant functions for product consistency in pharmaceutical I. ORCA CFD Software
industries which work with FDA testing and extremely tight
regulatory standards. Product incompliance stemming from ORCA uses Galerkin Least-Squares finite element technolo-
inconsistent product can be avoided if the process is either gy, a novel approach to computer aided modeling and opti-
experimentally or computationally characterized. While mizing mixing processes. Computer aided mixing, model-
experimentation can be costly and time consuming com- ing, and analysis involves Lagrangian and Eulerian analysis
pared to characterizing a system computationally, experi- for relative fluid stretching, and energy dissipation concepts
mentation is necessary to validate CFD results before they for laminar and turbulent flows. High quality, conservative,
can be used for design and optimization processes. accurate, fluid velocity, and continuity solutions are required
for determining mixing quality.
This review examines three cases of CFD validation for the
ORCA CFD package These cases, variations of processes The ORCA CFD package, based on a finite element formu-
that are used in the pharmaceutical and chemical process lation, solves the incompressible Reynolds Averaged
industries, are based on a batch reactor with various types of Navier-Stokes (RANS) equations. Though finite element
mixers: three Rushton Turbines in series, Pitched Bladed technology has been well used in areas of heat transfer, solid
Turbine (PBT), and the Ekato Intermig Impeller. The exper- mechanics, and aerodynamics for years, it has been applied
imentation for these systems was completed using Particle only recently to the area of fluid mixing. While most com-
Image Velocimetry (PIV) and Planar Laser Induced mercial technologies solve the resultant Partial Differential
Florescence (PLIF) techniques. Equations (PDEs) as discretized approximate solutions using

January 2002 1
finite volume or finite difference techniques, ORCA uses the tion at many points simultaneously along a planar section of
Galerkin Least-Squares (GLS) finite element technology. the flow. Division of displacement vector information by
The GLS finite element formulation presented provides time yields instantaneous velocity vector maps (2-D or 3-D
another formulation for numerically solving RANS based using stereoscopic imaging).
fluid mechanics equations.

The Galerkin finite element method serves as the basis of the

formulation but is insufficient to yield a stable solution for
the incompressible Navier-Stokes equations. Instabilities in
the finite element technique arise from the continuity equa-
tion and the convective term in the momentum equations.
However, the added least-squares operators provide rigorous
mathematical stability and convergence without sacrificing
accuracy. The GLS finite element technology, a central dif-
ference operator that uses weighted residual methodology
for momentum, energy and turbulence equations, minimizes
error in the approximating functions while satisfying all con-
servation equations (heat, energy, momentum), both locally Figure 1: Shows the set up of a PIV/PLIF system. A
on the elements and globally on the entire system, while laser sheet projects into the tank illuminating particles
maintaining system stability. Conservation principles are or fluorescent dye tracer. The camera is set perpendi-
specifically formulated and solved. The GLS technique pro- cular to the laser sheet taking images of takes images
vides the basis of the confidence, robustness, accuracy, con- of particles or the fluorescent dye tracer.
vergence, and stability of ORCA. It provides 4th order accu-
racy with respect to the spatial discretization and 2nd order
accuracy with respect to time. This method ensures Typical quantities obtained are two and three velocity com-
extremely accurate solutions for a wide variety of flows, as ponents, as well as Reynolds and shear stress information.
shown in the following validation studies. In the validation studies, this technique was used primarily
to measure velocity flow fields, which were then compared
The ORCA CFD code runs on a Linux platform to optimize to flow fields calculated by the CFD.
memory usage for the computations. Presently (2002), a
typical system consists of one to four desktop computers in Planar Laser Induced Fluorescence (PLIF) is an instanta-
parallel, each with 515 GB Ram and Intel® or Athlon® neous, whole-field measurement technique, during which a
processors. Solution times on a 3 million tetrahedral mesh fluorescent dye in solution with one of the mixture compo-
typically vary from ten minutes to five hours depending on nents is excited with laser light. Flow and mixing behavior,
problem complexity. In this study, the typical solution time concentration field, and reaction rate information can be
was one hour for solutions with an accuracy or residual ratio measured from the relative intensity of fluorescence between
of 10-4 for both pressure and velocity. the mixture components. The properties of the fluorescent
tracer depend on the excitation wavelength of the laser used.
Typically, a linear response between intensity and molecular
II. Experimental Equipment Used for Validation concentration is assumed when computing the instantaneous
concentration of the mixture.
Three non-intrusive experimental techniques were used to
validate the ORCA CFD program. The first technique, When coupled with PIV, this process can obtain information
Particle Image Velocimetry (PIV), measures velocities found on mass flux and Reynolds flux terms. In the experiments
within the system. The second technique, Planar Laser under consideration, a small amount of fluorescent dye
Induced Fluorescence (PLIF), measures dispersion of tracers (Rhodamine) is injected into the process volume and illumi-
or reactants within system. The third technique, a flow visu- nated with a laser light. The spreading of the dye can then
alization method, is based on an acid-base neutralization be followed in the flow, and the concentration can be quan-
reaction. Both PIV and PLIF are optical techniques that tified based on the intensity of fluorescence. Experimental
require high-speed cameras and a laser sheet to illuminate pictures are taken at various cross sections in two dimen-
either particles or fluorescent dye. Figure 1 shows an illus- sions, because folds and layers of the illuminated tracer fluid
tration of the system used. at different depths in the evolving mixing structure overlap
in a three-dimensional view and conceal fine details.
Particle Image Velocimetry (PIV) determines the linear dis-
placement information of groups of particles suspended in A third experimental method, a visualization technique
the flow at two instants in time and yields velocity informa- based on an acid-base neutralization reaction, also exploits
the size and shape of segregated regions in the mixing tank.

January 2002 2
A soluble, pH-sensitive indicator that appears blue in basic Bulk flow patterns from PIV and CFD
media and yellow in acidic media is added to each of the flu-
ids. When the mixture components come into contact during The figures in this section present comparisons between the
the mixing process, the neutralization reaction instanta- velocity field from CFD and PIV results for four flow con-
neously proceeds at the interface and a color change occurs. ditions: Re=20, 40, 80 and 160. Velocity vectors in the left
In poorly-mixing regions, or islands, the fluid appears blue half of each figure are based on CFD solutions for the flow
for long periods, as the acidic component is not sufficient for fields. All of the vectors are shown with the same length,
complete neutralization. Furthermore, the reaction rate is and the color-coding is based on the planar velocity magni-
limited in islands by the slow rate of diffusive transport tude normalized with respect to the tip speed of the impeller.
mechanisms. The vectors in the right half of each figure correspond to PIV
results for the same cross-sectional plane.
III. Flow Field Validation of a Stirred Tank Mixed
with 3 Rushton Impellers In Figure 3a, both the computational and experimental
results indicate that six recirculation loops exist, with one
Using ORCA CFD technology, the flow field was modeled located above and one located below each impeller. For this
in a stirred tank equipped with three Rushton impellers case, the maximum velocity magnitude in this vertical plane
(Figure 2) and the computational results compared with is only 20% of the tip speed of the impeller, indicating that
experimental measurements of the velocity field. the flows associated with recirculation regions are relatively
weak. The computational and experimental locations of the
centers of these recirculation regions coincide almost per-
fectly.

Figure 3b, which shows results for the second case where the
impeller speed is set to Re=40, reveals flow patterns sub-
stantially different from those presented in Figure 3a. Here,
the planar velocity magnitudes reach 32% of the impeller tip
speed. Again, velocity directions and magnitudes agree
between the CFD and PIV results.

The third flow condition for which validation of the flow

patterns is performed occurs at an impeller speed of Re=80,
as shown in Figure 3c. Here, four recirculation regions are
apparent, with the centers of two located below and above
the upper and lower impeller blades, respectively, and the
centers of the other two located near the mid-height of the
vessel and close to the wall. Under these flow conditions,

Figure 2a. Figure 2b.

Three dimensional view Top view of a six-
Velocity Magnitude
of a tank with 3 Rushton blade Rushton tur- 0.2 RPM

turbines. bine

Particle Image Velocimetry (PIV), an optical experimental

0.1 RPM
technique, measures bulk flow patterns in the stirred tank at
any cross-sectional plane. The tank is filled with glycerine,
creating a laminar and Newtonian flow field . The fluid is
seeded with fluorescent particles in the experiments and a
laser illuminates these tracers. When a high-speed camera
0.0
captures two consecutive snapshots of the flow, particle
velocities are calculated based on the distance traveled
between the two snapshots. The figures within the document
Figure 3a.
contain planar velocity vectors aligned with one of the
Bulk flow pattern at Re=20. Left half is from CFD, right
impeller blades.
half is from PIV. (from Alvarez, 2000 and Zalc, 2000)

January 2002 3
 Observations concerning the maximum percentage of the
impeller tip speed that is attained by the planar velocity mag-
Velocity Magnitude nitudes reveal that, at lower tip speeds, the flow within the
0.32 RPM
recirculation regions is relatively weak and the dominant
flow is azimuthal.

The flow pattern at an impeller speed of 8 RMP is shown in

0.16 RPM Figure 3d. As impeller speed is increased, inertial effects
become more significant and the relative strength of the
recirculation regions becomes greater. Both experiments
and computations reveal significant differences in flow
structures as a function of agitation rate.
0.0

Figure 3b.
Bulk flow pattern at Re=40. Left half is from CFD, right
half is from PIV. (from Alvarez, 2000 and Zalc, 2000) Quantitative Comparison of Velocity profiles from PIV and
CFD

The second set of figures represents a quantitative compari-

Velocity Magnitude son between the PIV and CFD results for the flow fields in
0.52 RPM
the stirred tank. The velocity components in the flow are
compared along the tank height at a vertical line. Consider
a vertical line situated midway between the impeller shaft
and the tank wall, aligned with one of the impeller blades
0.26 RPM and extending the entire height of the vessel. The velocity
components along this line can be interpolated using the
CFD results and compared with PIV data, as shown in
Figures 4 and 5.
0.0
In each case, the horizontal axis represents the vertical dis-
Figure 3c.
tance along the line, while the vertical axis represents the
Bulk flow pattern at Re=80. Left half is from CFD, right vertical and radial velocity components normalized with
half is from PIV. (from Alvarez, 2000 and Zalc, 2000) respect to the impeller tip speed. CFD results are shown as
thin lines (normalized vertical velocity) and thick lines (nor-
the planar velocity magnitudes reach 52% of the tip speed.
malized radial velocities). PIV data are shown as circles
(normalized vertical velocity) and squares (normalized radi-
Normalized
al velocities). The results provide verification between the
planar velocity experimental and computational methods. Rotating the pic-
magnitude
Velocity Magnitude
0.520.52 RPM
ture in Figure 4 90 degrees to the right, makes evident the
relation of these maxima to vertical heights corresponding
with the positions of the three impellers. At Re=20, three
0.39
local maxima for the normalized radial velocity exist at very
similar values, with the middle peak slightly lower than the
0.260.26 RPM other two. Although the boundary conditions at the top and
bottom of the tank are not the same, the profiles are quite
0.13 symmetric about the mid-height of the vessel (y = 0.18m).
The normalized vertical velocities are of smaller magnitude
0.000.0
than the radial velocities.

Figure 5 illustrates results for Re=40. Once again, the results

Figure 3d. provide verification between the numerical and experimental
Bulk flow pattern at Re=160. Only CFD results are
methods. However, the magnitudes of the normalized verti-
shown for this case (from Alvarez, 2000 and Zalc, 2000).
cal velocities are larger than for the previous case, at lower
impeller speed. A large positive peak at a height of approx-
imately 0.12 m contrasts with a large negative peak at rough-
ly 0.24 m. Overall, these results indicate that as the agitation
rate increases within this range, the outward radial flow

January 2002 4
 associated with the middle impeller becomes more domi-
Normalized radial and vertical components of the velocity
nant. In addition, the normalized vertical velocities become
increasingly dominant because the fluid that is ejected out-
ward radially by the middle impeller must recirculate
through the impeller regions.

Mechanism of mixing illustrated

Recently, the chaotic mixing process in stirred tanks

equipped with Rushton turbines has been shown to be driv-
Vertical Height (m)
en by flow perturbations caused by the presence of the
impeller blades (Alvarez et al., 2000a). As fluid filaments
Normalized radial and vertical components of the velocity

are advected, they experience continuous length stretch and

reorientation in a chaotic flow environment, such as in the
tank with Rushton turbines. Fluid filaments in the impeller
region must allow the periodic passing of the impeller
blades. Thus, the flow is perturbed periodically, creating a
characteristic wiggle-like structure in the mixing patterns.
This mechanism of stretching and reorientation is illustrated
next by computing what is essentially a short-term Poincaré
section.
Vertical Height (m)

Figure 4. Since particles located near the shaft tend to be ejected radi-
A comparison of experimental and computational val- ally outward by the impellers, tracer particles are placed in
ues for the radial and vertical velocity components. that region initially to capture the mixing behavior in the
The results are for Re=20 impeller speed in (a) and
impeller region. The specific initial condition for calculating
Re=40 in (b). The thick line indicates the CFD data for
the trajectories of tracer particles is a vertical line just out-
the radial component of the velocity and the thin line is
side the shaft that extends the entire height of the tank. The
the vertical component of the velocity in each figure. line consists of 30,000 equally spaced passive tracer parti-
The corresponding experimental data is plotted as cir- cles. Their trajectories are computed for 60 revolutions and
cles and squares, respectively. (from Alvarez, 2000 and position and residence time is recorded every time a particle
Zalc, 2000)
crosses a plane fixed to one of the impeller blades. Since the
flow is periodic, plotting all of the intersections that occur up
to a particular time on a single figure reveals the temporal
evolution of mixing structures.

PLIF reveals whether segregated regions exist in the flow.

Normalized radial and vertical components of the velocity

Pictures are taken at various cross sections in two dimen-

sions, because folds and layers of the illuminated tracer fluid
at different depths in the evolving mixing structure overlap
in a three-dimensional view and hide fine details. The view-
ing plane was aligned with one of the impeller blades and
pictures of the flow were taken after 60 impeller revolutions.

The lower portion of Figure 6a represents all particle loca-

tions up to 60 impeller revolutions for the flow at Re=20,
whereas Figure 6b denotes similar results for Re=40. The
Vertical height (m)
upper sections of the figures show the corresponding exper-
imental structures at the same conditions. Both pictures
Figure 5.
illustrate the numerous particles that have passed through the
Comparison of experimental and computational values
of the radial and vertical velocity components. The
blade and the pronounced "wiggle" structures that emanate
impeller speed is set to Re=80. The thick line indicates from the top and bottom impellers. It should be noted that
the CFD data for the radial component of the velocity these structures are apparent at this time because the parti-
and the thin line is the vertical component of the veloci- cles were initially located along the shaft; under other initial
ty. The corresponding experimental data is plotted as conditions, more impeller revolutions would be required for
circles and squares, respectively. (from Alvarez, 2000 tracer particles to pass through the impeller regions and
and Zalc, 2000 exhibit such structure. Therefore, it is most beneficial to

January 2002 5
 (a) (b) density of points. Particle trajectories are computed for 600
rotations and the position is recorded every time a particle
crosses the intersection surface. Once all the positions for
every impeller revolution are superimposed, segregated
regions are indicated as concentric rings on the plot. In these
regions (called "islands"), material exchange with the outer
region of the tank occurs by slow diffusive transport. Well-
mixing areas, where the flow is chaotic, appear as a random
cloud of points, devoid of large-scale structure. The general
notion that chaos means randomness is partially due to the
traditional use of this method to examine the flow structure.

The lower portion of Figure 7a represents the Poincaré sec-

tion computed for Re=20 impeller speed, while the upper
portion of this figure shows an experimental image of the
flow at a cross sectional plane. This evolving mixing struc-
ture was captured using PLIF, or Planar Laser Induced
Fluorescence experimental techniques.

(a) (b)

Figure 6.
The structures of segregated regions in the stirred tank
are revealed by the calculation of Poincaré sections.
Results are shown for Re=20 in (a) and Re=40 in (b).
Also, the top half of (a) and the bottom half of (b) repre-
sent the corresponding PLIF experimental results.
(from Alvarez, 2000 and Zalc, 2000)

inject in a region that will cause material to pass through the

impeller blades as early as possible (i.e., upstream of the
impeller). This seemingly obvious modification is often
overlooked in actual chemical processes where material is
often injected only at the impeller discharge.

Also, the frequency and amplitude of the wiggles rise with

increasing distance from the impeller blades. The area
densely covered with particles near the tank wall and bottom Figure 7.
evolves as the wiggle structures created by the top and bot- A structure created by the periodic passing of the
tom impellers "wrap" around the toroidal-shaped segregated impeller blades is revealed in experimental (top) and
regions and recirculate through the blade regions more than computational (bottom) mixing analysis. The trajecto-
once. ries of tracer particles are perturbed by the impeller
blade resulting in a wiggle-like structure that creates
Asymptotic mixing performance mixing between the layers. Two flow conditions are
shown: (a) at Re=20 and (b) at Re=40. (from Alvarez,
2000 and Zalc, 2000)
A common computational approach to understanding com-
plex fluid motion in periodic chaotic flows reduces the num-
ber of dimensions by exploiting time-discrete dynamics.
Computing Poincaré sections is a traditional tool to charac- A PLIF image of the mixing structure taken after several
terize asymptotic mixing behavior. In the case of the stirred hours in this set of experiments for an injection initially
tank, consider one vertical plane cutting through the vessel, located in the chaotic region yielded intriguing information.
aligned with one of the impeller blades. Tracer particles in Both the experimental and computational views of the flow
the fluid pass through that reference plane multiple times, as field indicate the existence in the tank of six macroscopic
the impeller rotates and all of the intersections of particle tra- segregated regions, which were forecasted by six recircula-
jectories are plotted on a single graph. tion loops visible in the velocity field (Figure 3a). The sizes
and locations of the segregated poorly-mixing regions are
In Figure 7, tracer particles initially placed in the flow ensure similar in the Poincaré section and the experimental PLIF
that the entire cross-sectional plane is covered with equal results. It should be noted that, in the experimental image,

January 2002 6
no details within the islands of regular flow can be seen as
the Rhodamine was initially injected into the chaotic region
and can only enter the islands by diffusion after long periods.
8mm

Figure 7b presents similar results for a Re=40. The upper

portion of the figure shows the computed Poincaré section,
while the lower portion indicates PLIF results. Again, the 0.36m
simulations and experiments agree. The flow field at this
stirring rate differs considerably from the previous case, as
seen in the velocity computations; however, higher impeller
speed does not necessarily ensure better or faster mixing.
Increasing the impeller speed shifts the location of, but does 0.24m
not eliminate, islands in the vessel. In this case, two small-
er islands, one near the top impeller and one near the bottom Figure 8.
impeller, and two large kidney shaped islands, outward from Schematic drawing of the laboratory-scale mixing tank
the impeller assembly, are evident. equipped with a 45o pitched blade impeller.

media and yellow in acidic media is added to both mixture

Laminar flow and mixing in a stirred tank equipped with components. Starting with an initially blue, basic solution in
three Rushton turbines were investigated both computation- the tank, well and poorly-mixing regions can be identified by
ally and experimentally. Simulations under four flow condi- injecting an acidic solution, also marked with the same pH
tions were performed with the ORCA CFD package and the indicator, to the tank during the experiments. As the reaction
results were validated by two non-intrusive experimental proceeds, the basic environment is neutralized and colored
methods. An unstructured computational grid of approxi- yellow in regions of intense mixing.
mately 2 million tetrahedra was sufficient to resolve the lam-
inar flow field. Using the ORCA software suite, geometry Fine-scale resolution of flow structure
modeling, mesh generation, flow field solution and analysis
can be performed within 24 hours. This time frame indicates The laboratory-scale experimental tank is initially filled with
that CFD is now a mature tool for the analysis of industrial- a viscous solution of indicator and base (blue) in Figure 9a.
ly relevant mixing devices with complex geometries. A Then yellow solution, which contains enough acid to change
remarkable comparison between the CFD results and PIV the environment in the entire process volume to acidic (yel-
data was observed for bulk flow patterns at several different low), if completely mixed, is injected near the impeller.
agitation rates. After several hours of stirring, the bulk of the tank is yellow,
indicating that the reaction has completed in those areas.
Further, the sizes and locations of poorly-mixing regions in However, two toroidal-shaped regions persist, and only mix
the computed Poincaré sections agreed very well with the with the rest of the flow by molecular diffusion.
results of PLIF experiments. The size and location of
(a)
unmixed regions appears to depend on Reynolds number in (b)
a non-intuitive manner. In practice, this finding implies that
mixing rates, concentration profiles, and optimal injection
locations also have a notable dependence on Reynolds num-
ber. Clearly, since the parameter space is simply too large to
explore experimentally, the practitioner needs access to CFD
toolboxes to undertake optimal design of laminar mixing
processes.

IV. Flow Field Validation of Batch Stirred Tank

Mixed with a 45o Pitched-blade Turbine
Figure 9.
ORCA was used to model an experimental system involving Computational (a) and experimental (b) pictures of the
an unbaffled tank with a single, 45o pitched-blade turbine. flow in a tank with one pitched blade turbine. The
Figure 8 provides a schematic drawing of the experimental regions above and below the impeller only exchange
mixing device and the relevant dimensions. A visualization material with the rest of the flow by slow diffusion. The
technique based on an acid-base neutralization reaction is structures are persistent even after many hours of
used to examine mixing behavior in the tank. A pH sensitive operation. (Fig 9a from Alvarez, 2000 and Fig. 9b from
indicator, bromothymol blue, which appears blue in basic Zalc, 2000)

January 2002 7
In the computations in Figure 9b, one tracer particle is is fitted with a concentric shaft and four Ekato Intermig®
placed inside each segregated area and their trajectories are impellers. These complex impellers, commonly used in
computed for several thousand impeller revolutions. Note pairs, are each rotated 90o on the shaft. Figure 11 shows the
that the simulated massless tracers do not leave the toroidal surface mesh of the tank and the four impellers considered in
regions, even after thousands of impeller revolutions, but this analysis, which are placed at equal distances along the
remain trapped in those zones as in the experiments. shaft. The unstructured volumetric mesh of the mixing tank
necessary to obtain the required solution quality was approx-
More detailed illustrations of the upper segregated region in imately 6 million elements. The velocity and pressure field
the tank are shown in Figure 10. A cross-section of the was obtained by iteration until convergence beyond 0.0001
doughnut-shaped torus is enlarged in the plane of one of the was achieved.
impeller blades to refine the details of the structure. The
computational result appears on the left side of Figure 10a,
whereas the experimentally resolved structure appears on the
right Figure 10b. The experimental picture was obtained by
using small, neutrally buoyant seeding particles that follow
the fluid motion. A laser sheet is passed through the tank
that is made of Plexiglas and a picture is taken with a digital
camera. A chain of small islets, each less than 1 mm across,
surrounds the large segregated region..

The computational results for the flow field were validated

by comparisons to experimental measurements by acid-base
visualization technique. Very good agreement was obtained
between the experimental and simulated flow fields.

(a) (b)

Figure 11
(a) Surface mesh of a tank equipped with 2 pairs of
Ekato Intermig? impellers.(b) Geometry of a single
Intermig impeller.

Figure 10. Laminar flow of a viscous Newtonian fluid is examined by

Enlargerd view of the segregated regions (a.k.a.
comparing the computational velocity field to experimental
islands) near the top part of the pitched turbine: (a)
computational data (b) experimental picture. The small measurements obtained by PIV measurements. The compu-
chain of islands surrounding the main region are only a tational and experimental velocity field was compared at a
few millimeters in size in a 12 in. radius laboratory- vertical plane 2 mm from the center of the tank, which was
scale tank, yet they are clearly resolved in both the broken into several smaller interrogation zones. Two areas,
computations and experiments. Fig 10a from Zalc, 2000 shown in Figure 12 and Figure 13, demonstrate good agree-
and Fig. 10b from Alvarez, 2000) ment between the computational and experimental flow
solutions without additional scaling. In Figure 12a recircula-
tion loop can be observed below the bottom impeller..

VII. FlowField Validation for a Stirred Tank Mixed The impellers create strong vertical flow, and the areas of
with 4 Ekato Intermig® Impellers highest magnitude form a complex S-shapd region between
impellers as demonstrated by experimental and simulated
Computations of the Velocity Field flow patterns between impeller 1 (closest to the bottom of
the vessel) and impeller 2. Again, a recirculation loop forms
The Ekato Intermig® vessel geometry commonly used fer- below the impeller, and the computational and experimental
mentation applications is modeled here by the ORCA CFD locations coincide exactly.
technology. A tank with a height to diameter ratio close to 3

January 2002 8
 in a vessel equipped with Ekato Intermig impellers. This
PIV
complex impeller design is often used in fermentation
processes for oxygen distribution. Because the viscosity of
many fermentation broths changes during processing, a mix-
ing operation that started under highly turbulent conditions
frequently ends up in the laminar regime at completion.

The flow field and mixing were examined at Re = 37 in the

laminar regime. The computational results for the flow field
were validated by comparisons to experimental measure-
ments by Particle Image Velocimetry. Very good agreement
CFD
was obtained between the experimental and simulated flow
fields.

VIII. Conclusions

Three case studies were examined with the objective of val-

idating the GLS Finite Element formulation used within
ORCA CFD package. The investigation showed good
agreement between the experimental results from the
Figure 12.
Rushton Turbine, PBT, and the Ekato Intermig and the CFD
Experimental and computational flow patterns in a mix- results. The results not only qualitatively predicted the var-
ing tank with 4 Ekato Intermig impellers. The region in ious flow structures but also quantitatively predicted the
the red box is enlarged for better view on the right- velocity flow fields. Based upon these results ORCA can be
hand side. The experimental image of the velocity field, considered an excellent tool for design and optimization.
captured using Particle Image Velocimetry or (PIV),
demonstrates that the CFD technology is able to Optimization of stirred reactors through the assistance of
resolve flow in a complex mixing apparatus with large CFD simulations will minimize or eliminate inconsistent
process volume.(Szalai, 2002)
products associated with suboptimal mixing. While reduc-
The objective of this section was to analyze flow and mixing ing costs in the chemical, food, and personal products indus-
tries, CFD simulations will help maintain tight control of
PIV
highly sensitive goods like pharmaceuticals.

References

Alvarez, M.M. "Using Spatio-Temporal Asymmetry to

Enhance Mixing in Chaotic Flows: From Maps to Stirred
Tanks", 2000, Ph.D. Dissertation, Rutgers, The State
University of New Jersey, New Brunswick, New Jersey.

CFD
Zalc, J.M. "Simulation of Flow and Mixing in a Three-
Impeller Rushton Stirred Tank and an SMX Static Mixer",
2000, Ph.D. Dissertation, Rutgers, The State University of
New Jersey, New Brunswick, New Jersey.

Zalc, J.M., Alvarez, M.M., Muzzio, F.J. and Arik, E.;

"Extensive Validation of Computed Laminar Flow In a
Stirred Tank With Three Rushton Turbines", AIChE J.,
2002, v. 47 (10), p. 2144-2154.
Figure 13.
The mid-section of the tank is enlarged, where the flow
between impellers 1 and 2 show strong axial flow. The
location of the reciculation loop just below impeller 2
shows that computations resolve the flow accurately.
(Szalai, 2002)

January 2002 9
January 2002 10