PROCESS SYSTEMS ENGINEERING

Design of a High-Efficiency Hydrofoil Through

the Use of Computational Fluid Dynamics and
Multiobjective Optimization
N. Spogis
ESSS, Engineering Simulation and Scientific Software, Rua do Rocio, 423 10 andar conj.1001/1002, Vila Olı́mpia,
CEP: 04552-000, São Paulo, SP, Brazil

J. R. Nunhez
School of Chemical Engineering, State University of Campinas, UNICAMP, P.O. Box 6066, 13083-970,
Campinas, SP, Brazil

DOI 10.1002/aic.11804
Published online May 14, 2009 in Wiley InterScience (www.interscience.wiley.com).

A computational fluid dynamics (CFD) model is proposed, based on ANSYS-CFX tools

coupled to optimization models inside the commercial optimization software mode-
FRONTIER in order to obtain an optimal design of a high-efficiency impeller for solids
suspension. The analysis of impeller shape performance was carried out using the shear-
stress transport (SST) turbulence model with streamline curvature correction. This turbu-
lence model combined the advantages of the j
e and j
x models, ensuring a proper
relation between turbulent stress and turbulent kinetic energy, allowing an accurate and
robust prediction of the impeller blade flow separation. The multiple frames of reference
and the frozen rotor frame change models were used for the rotor/stator interaction
inside the mixing vessel. The optimization procedure used seven design variables, two
nonlinear constraints and two objective functions. The objective functions chosen (among
many other possible options) to evaluate the impeller performance were the maximum
solid distribution throughout the vessel (homogeneous suspension) reflected by a low
variance between local solid concentration and average solid concentration inside the
vessel and the higher pumping effectiveness, which was defined as the quotient of the
flow and power numbers. The first objective function searches for impeller configurations
able to provide good solid suspension, since it aims to achieve homogeneous suspension.
The second objective function aims to reduce power consumption for a high-pumping
capacity of the impeller. These criteria were considered enough to characterize the opti-
mized impeller. Results indicated that the optimized impeller presented an increase of
the pumping impeller capacity and homogeneous solid suspension with low-power con-
sumption, especially when compared with the PBT 45 impeller. V C 2009 American Institute

of Chemical Engineers AIChE J, 55: 1723–1735, 2009

Keywords: mixing, solid suspension, computational fluid dynamics, stirred tank,
optimization, hydrofoil

Correspondence concerning this article should be addressed to N. Spogis at

[email protected] and J. R. Nunhez at [email protected]

V
C 2009 American Institute of Chemical Engineers

AIChE Journal July 2009 Vol. 55, No. 7 1723

Introduction following the mesh constraints specification), its highly auto-
mated batch processing and the scripting framework for the
Mixing vessels are widely used in the chemical, petro-
batch preprocessing, processing and post-processing opera-
chemical, pharmaceutical, biotechnological and food process-
tions were incorporated inside modeFRONTIER.23 In other
ing industries to optimize mixing and/or heat transfer. Mix-
words, after defining all the impeller specifications, all these
ing must be efficient to ensure optimum product quality.1,2,3
tools work together generating the geometry of an impeller
Variables of interest may include mixing times, gas holdup,
obtained by the optimization procedure followed by the
power draw, local shear, strain rates and solid distribution.
mesh generation, model simulation and post processing of
In the chemical industry, for example, the mixing of react-
the most important results. All these steps were done for ev-
ing substances should occur as fast as possible in order to
ery geometry until the optimum geometry was found follow-
achieve an efficient reaction and this aspect is often a pro-
ing the model constraints. The complete procedure is carried
cess problem. In this case, an impeller which produces a
out by the model without any further human intervention.
highly turbulent flow is needed to reduce segregation and
The optimization software ModeFRONTIER is composed
minimize the mass and energy transport limitations for
of models that range from gradient-based methods to genetic
chemical reactions.
algorithms.24–26 It allows the specification of the two objec-
In biochemistry, on the other hand, there is a need to sus-
tive functions used to design the impeller namely: (1) the
pend microorganisms to improve the efficiency of bioreac-
minimization of the concentration variance inside the vessel,
tors.
and (2) the maximization of the impeller efficiency, defined
Another macro-level mixing aspect is related to how the
by the quotient (flow number/power number). Factors that
flow pattern generated by the impeller affects both the solid
influence the impeller geometry were successfully investi-
suspension and particles incorporation and distribution within
gated using this tool. This optimization software in effect
the vessel. Parameters that affect Liquid-solid mixing are the
became a link between the CAE tools (ICEM and CFX)
shape of the solids, solid-size distribution area, solid concen-
optimizing the procedure by modifying the value assigned to
tration, solid density, liquid density and viscosity.1,3,4–12,14
the input variables (design parameters and flow requirement),
The quality of solid distribution includes the description
and monitoring the outputs. The section entitled ‘‘Prelimi-
of fillets in on and off bottom motions and the uniform solid
nary investigations’’ explains the optimization procedure in
suspension. These effects are very important in some mixing
more detail.
applications such as in the mining industry, rubber crumb,
crystallization and precipitations. Abrasion and impeller
wear are important factors to be considered in solid-liquid
mixing.3 Basic Numerical Concepts
There are also some important economic issues involved. In the following sections, a brief introduction to the CFD
The minimization of the amount of power consumption to model is presented and a description of the employed numer-
ensure certain mixing conditions is one of these issues. The ical tools is also given. Then the coupling of the models
cost of the impeller and vessel materials, and the lifetime of within the commercial program (modeFRONTIER) is also
the equipment are also important.2,3,15 considered.
All the aforementioned aspects depend strongly on the ge-
ometry of the impeller, vessel parameters as well as the rota-
tional speed of the stirrer and the fluid properties.1,3 Governing equations for fluid flow
The variety of mixing applications has led to a tremen- The set of equations solved numerically by ANSYS CFX
dous number of different types of impeller and vessels which were the unsteady Navier-Stokes equations in their conserva-
are in use at present. Due to the great amount of factors that tion form and the mass conservation equation. It also used
affect the impeller performance, it is very difficult to select the alternate rotation model that solves numerically the abso-
a proper impeller design for a specific process. Experimental lute frame velocity instead of the relative velocity.
investigations are usually very costly and time-consuming. The frozen rotor model was used in a 120 sector of the
The application of computational fluid dynamics (CFD) reactor. The shear-stress transport (SST) turbulence model
results in a faster and lower design cost, reducing experi- was used since it is more accurate and robust for the predic-
ments and providing a more reliable scale-up.22 This also tion of problems involving flow separation. This aspect is
gives a better understanding of processes, leading to higher important in this project since it is paramount that no bound-
yields and reduced waste.In this study, the numerical investi- ary layer separation exists in the optimized impeller.27–29
gation of the flow of a stirred vessel using a CFD approach Vessel Parameterization A schematic representation of
coupled with a multiobjective design optimization method is the vessel configuration is shown in Figure 1. The system
presented, and results indicate that it can be a very useful consists of a torispherical-bottomed cylindrical vessel with
tool offering new possibilities for higher product quality, diameter T and height H, which equals the height of the liq-
cost reduction and power consumption minimization. uid. The off-bottom clearance is constant and equals H/4.
The ANSYS CFX software was chosen for the numerical The torispherical 100-6 head has a crown radius of 100%
CFD results. ANSYS ICEM CFD was chosen for the geome- (or equal to) the diameter of the head with a knuckle radius
try and mesh generation process in order to provide sophisti- of 6% of the diameter of the head. The shaft of the impeller
cated geometry acquisition, mesh generation, mesh editing is concentric with the axis of the vessel. The actual geomet-
and a wide variety of solver outputs. The built-in geometry rical parameters, which are considered as a standard configu-
creation of the mesh generating software (automatically done ration, are summarized in Figure 1. The fluid is water at 25

1724 DOI 10.1002/aic Published on behalf of the AIChE July 2009 Vol. 55, No. 7 AIChE Journal
 Figure 1. Schematic representation of the stirred tank studied in this work.

C, with density q ¼ 997.0 kg/m3, and viscosity l ¼ 0.8899 consideration is the number and the shape of the blades
cP. used. Increasing the aspect ratio of the blades reduces drag.
Impeller Design Variables and Meshing The optimization However, the amount of pumping produced depends on the
of the impeller blades requires a representation of the blade blade area, so using high-aspect blades can lead to the need
that should be flexible enough to produce a very wide range of an impeller diameter which is unviable.
of impeller shapes. Conversely, it should use the minimum A further balance is that a small number of blades reduces
number of variables possible to save computational time and the interference between blades, but there is a need of a suf-
to allow an efficient storage and manipulation of the data for ficient total blade area to suspend the solids, so a compro-
a wide variety of viable breeding population of candidate mise between these opposing effects is needed. Increasing
impellers. the number of blades also decreases the amount of work
Changes on the efficiency of the impeller are produced by each blade is required to perform.
a wide number of factors, namely: adjustments to the helix A schematic representation of the impeller configuration is
angle, the angle between the resultant relative velocity and shown in Figure 2. For the model of this work, the impeller
the blade rotation direction, and also to the blade pitch. Very blade can be represented by seven construction parameters,
small pitch and helix angles give a good performance against which are even able to allow a twisting of the blade. The pa-
resistance, but provide little thrust and also little pumping. rameters are:
Larger angles have the opposite effect. 1 Impeller diameter ratio;
The best helix angle is when the impeller blade is acting 2 Root chord;
as a wing producing much more lift than drag, roughly 45 3 Tip chord;
in practice. However, due to the shape of the impeller, only 4 Root chord angle;
part of the blade can actually be operating at peak efficiency. 5 Tip chord angle;
The outer part of the impeller blade produces the most 6 Root profile;
pumping, and so the blades are positioned at a pitch that 7 Tip profile.
gives optimum angle to that portion. Since a large portion of Two of these parameters are discrete variables (root and
the blade is, therefore, at an inefficient angle, the inboard tip profile) and five are continuum variables. Table 1 shows
ends of the impeller blade are hidden by a streamlined spin- the range of variation of these parameters. All the variables
ner to reduce the resistance torque that would otherwise be defined above were changed in a range of values to allow a
created. Very high-efficiency pumping impellers are similar. great number of candidate numerical prototypes to be tested
They need an optimum angle of attack in its airfoil section in the optimization procedure (the ranges were determined
to work properly. An impeller working at a pitch angle of based on the values used for commercial impellers).
45 at high-rotational speeds, presents a very high-angle of This new hydrofoil impeller was designed to maximize
attack. There is a need to adjust the impeller pitch angle to solid dispersion in stirred vessels at the lowest possible
alter resistance to torque and improve efficiency. A further power consumption.

AIChE Journal July 2009 Vol. 55, No. 7 Published on behalf of the AIChE DOI 10.1002/aic 1725
 One of the most important aspects of the model of this
work is that it needs to predict with accuracy the boundary
layer on the impeller blade, which makes paramount the
choice of the turbulence model. The SST turbulence model
with curvature correction was chosen to properly predict the
boundary layer behavior. Two criteria need to be respected
in order to generate meshes which satisfy the minimal
requirements for accurate boundary layer predictions:
• Minimum spacing between walls and first node in the
boundary layer.
• Minimum number of nodes in the boundary layer.
The mesh near the wall followed size restrictions. The
size of the layer of control volumes at the wall are predicted
by simplifications of laminar boundary layer equations for
flat plates. Respecting these requirements, fundamental for a
correct prediction, a tetrahedral mesh was generated. A
smoothing algorithm was chosen in order to provide high
element quality (avoiding distorted control volumes). The ac-
curacy of calculation was improved by placing a thin layer
of prism elements near the walls. The mesh generated is
shown in Figure 4.
Figure 2. Schematic representation of the impeller with
the variables used in the optimization proce-
dure. Numerical principles of the optimization method
It is well known that classical gradient-based algorithm
methods use the ‘‘direction of improvement’’ information in
A note should be made that the power number of the order to achieve a fast and accurate convergence toward the
impeller determined in this work could lead to a working optimal solution. It requires an accurate gradient evaluation.
condition that might not be ideal for some industrial use. Holland34 is generally credited for the creation of the
The commercial hydrofoils have a power number which are genetic algorithm model, and he gives an introduction to this
normally higher than 0.1 to guarantee a wide range of use. method in Holland.34 He points out that the success of this
Therefore, commercial hydrofoils have a power number computational method, which uses optimization techniques
higher than the power number of the impeller obtained in such as simple hill climbing, is due to its ability to properly
this work through the optimization procedure. The model use partial solutions in the optimization problem. Mode-
developed in this work does not use this value of power FRONTIER has implemented an improved version of multi-
number as a lower limit. Further work may take this aspect objective genetic algorithm (MOGA-II). It uses a smart mul-
into consideration for selection of geometries for general tisearch algorithm for robustness and directional crossover
industrial use. for fast convergence. The efficiency of the method is ruled
Since avoiding flow separation is an important aspect of by its reproduction operators: classical crossover, directional
this work, several airfoil shapes were studied. An airfoil crossover, mutation and selection.35
with the shape of a wing is seen at its cross-section in Figure This project employed the multiobjective genetic algo-
3. It is passed through a fluid in order to provide either lift rithm to determine a Pareto optimal set of impeller blade
or down force, depending on its application. This force is designs, with each member offering a unique trade-off
generated by a pressure gradient. The pressure gradient and between conflicting objectives of low-energy consumption,
the impeller blade size are responsible for the impeller increasing the impeller pumping capacity and improving
pumping. tank homogeneity. The Pareto optimal design consists of all
Subsonic and low-Reynolds airfoils have a characteristic
shape. It starts with a rounded leading edge, followed by a Table 1. Input Variables
sharp trailing edge, often with camber. A comprehensive
Minimum Maximum
source of low-speed airfoil geometries and performance data Variable Value Value Discrete/Continuum
is provided by Goparathnam and Selig30 at the University of
Impeller 0.4 0.5 Continuum
Illinois at Urbana-Champaign (UIUC). Other works are Xu diameter
and Sankar,31 Lyon et al.32 and Caradonna and Tung.33 Root chord 0.2 0.2 Continuum
The criteria adopted for the airfoil selection in this study Tip chord 0.1 0.2 Continuum
are: high lift at a low angle of attack, high Clmax, gentle stall Root chord 20 degrees 70 degrees Continuum
characteristics, relative low coefficient of moment, suffi- angle (related to (related to
rotation axis) rotation axis)
ciently low drag, easy manufacturability and good operations Tip chord 30 degrees 95 degrees Continuum
at low Reynolds number. angle (related to (related to
Based on these criteria, four possible airfoils were selected rotation axis) rotation axis)
and used as root and tip blade airfoil parameters, as shown Root profile DAE11, S1223, E387, FX 63-137 Discrete
Tip profile DAE11, S1223, E387, FX 63-137 Discrete
in Figure 3.

1726 DOI 10.1002/aic Published on behalf of the AIChE July 2009 Vol. 55, No. 7 AIChE Journal
 Figure 3. Four different airfoils used for low Reynolds number applications.

the nondominated solutions possible in the N-dimensional tion tool (ANSYS ICEM CFD) becomes active and creates
parameter domain being considered, and forms an N-dimen- the new geometry and the corresponding numerical grid.
sional surface. 3 Flow simulation: Np estimation: The solver computes
the flow field and estimates the power number Np when it
Control program gets the signal that the new grid is available. This simulation
is made using a steady state approach.
The components described in the preceding sections are 4 New rotation computation: The value of the power
integrated and coupled to modeFRONTIER optimization number Np estimated for the geometry normally provides
software, which is illustrated schematically in Figure 5. power consumption different from the 2 kW/m3 criteria cho-
The optimization procedure involves the following major sen in this work. With the knowledge of the impeller diame-
steps: ter and the fluid density, a new rotational velocity is calcu-
1 Optimizer: The optimizer is started and computes a lated in order to conserve the power consumption. This pro-
new set of design variables. Afterward it is turned on stand cedure is described in the next section entitled Power
by. Consumption Conservation.
2 Geometry and Grid generation. On getting the signal 5 Steady-state flow simulation: With the new rotation
that the new design variables are available, the grid genera- velocity, a new steady-state simulation is performed in order

AIChE Journal July 2009 Vol. 55, No. 7 Published on behalf of the AIChE DOI 10.1002/aic 1727
 understood and practical. It is certainly more conservative
than the impeller tip speed criteria, which is also widely
used.
As the vessel volume in this optimization procedure is
constant, it is only necessary to maintain the power con-
sumption constant. As the impeller geometry is not constant,
the power number (Np) is not constant either, so rotational
speed corrections are needed in order to maintain the power
consumption constant.
A power consumption of 2 kW/m3 is used in this work to
heavy solid suspension processes. This value is a rule of
thumb. However, it is true in many industrial applications.
This fixed value was chosen for the optimization process, so
the different impellers generated could be compared only in
terms of characteristic flow and its capacity to homogenize
the stirred vessel.
The following algorithm was used for steps 3 to 5 of the
optimization procedure to maintain the same power con-
sumption.
1 Evaluate the torque
2 With the blade torque and initial rotational velocity,
calculate the Power consumption: P ¼ x:T
3 With the power consumption, fluid density, initial rota-
tion velocity and impeller diameter, calculate the power
number: Np ¼ q:NP3 :D5
4 With the power number, impeller diameter, fluid den-
sity and the desired power consumption, a new rotational
qffiffiffiffiffiffiffiffiffiffiffiffi
speed is calculated and used in the step 5 of the optimization
procedure: N ¼ 3 q:DP5 :Np

Constraints
Constraint handling is an integral part of any general pa-
rameter of an optimization method. In order to restrict the
solution to a limited area, only two of the defined constraints

Figure 4. An example of a mesh generated in this work

using adaptive meshing.

to determine a starting value for the next step that will deter-
mine the solid distribution. The flow number Nq is also esti-
mated with the new rotational speed.
6 Transient solid dispersion: Estimates the solid distri-
bution within the stirred vessel.
7 Post process results: In this step ANSYS CFX-Post
computes the output variables and objective functions for the
new geometry, and writes them in an output ASCII file of
modeFRONTIER.
8 Test of optimizer convergence: The optimizer
decides, by the given criteria, if the current value of the
objective functions should be accepted as optimum. If it is
confirmed, the procedure is finished, if not, the procedure is
automatically repeated from step 1.

Power Consumption Conservation

Scale-up based on equal power per volume P/V, is prob- Figure 5. Schematic Diagram of the optimization pro-
ably the most accepted criteria in mixing because it is easily cedure.

1728 DOI 10.1002/aic Published on behalf of the AIChE July 2009 Vol. 55, No. 7 AIChE Journal
 tions, which are the pumping effectiveness (quotient of the
flow and power numbers) and the vessel solid concentration
variance. The pumping effectiveness was maximized and the
vessel solid concentration variance was minimized.
The solid concentration variance of the vessel was esti-
mated by the well known statistical formula

1 X n
s2 ¼ ðCi
CÞ2 (1)
n
1 i¼1

Where n is the number of nodes in the mixing vessel

mesh. As Eq. 1 shows, a truly complete homogeneous solid
suspension would provide s2 ¼ 0. Therefore, the optimized
impeller has a low variance.
The pumping effectiveness is defined as
Nq
n¼ (2)
NP
The higher the pumping effectiveness is, the higher the
pumping capacity will be for small power consumption.

Numerical Results
In the next sections, a description is given of the grid
requirements followed by a discussion of the preliminary
investigations of the design space. These sections propose a
strategy to reduce computational errors and minimize the
computational time required to optimize and design the
high-efficiency hydrofoil prototype.

Grid requirements
In order to predict turbulence properly, it is important to
have at least a layer of control volumes capturing the effect of
the boundary layer, which is assessed by the dimensionless
variable yþ. Studies of the sensitiveness of yþ value in the
impeller for values between 0.001 and 1 show that there is
almost no effect of these changes on the final solution. Once
the maximum yþ increases to values above 10, the transition
location of the boundary layer begins to move upstream. At a
Figure 6. Population distribution in the design space maximum yþ of 25, the boundary layer is almost completely
for the initial impellers used for the optimiza- turbulent. Preliminary investigations also indicate that for yþ
tion method. values below 0.001, the transition location appears to move
downstream. This is presumably caused by the large surface
relate specifically to the creation of ‘‘realistic impeller values of the specific turbulence frequency, which increases
blades’’ in the optimization problem: proportionally with the height of the first grid point.
• Tip chord angle <5 root chord angle. The effect of expansion ratios of the wall for a yþ value
• Tip chord \¼ root chord. of 1 was studied in order to determine the number of cells in
The angles of the root and tip chord vary, so the CAD the boundary layer. Expansion factors between 1.05 and 1.1,
generating model create the surface using a linear interpola- present no effect on the solution. For larger expansion fac-
tion of the values of the angles starting at the hub (root tors between 1.2 and 1.4, there is a small, yet noticeable
chord), and finishing at the impeller tip (tip chord). The upstream shift in the transition location.
width of the blade is also an interpolation of the values of The effect of the streamwise grid refinement shows that
the root chord and tip chord. The final surface is obtained by the model was not very sensitive to the number of stream-
the CAD model using a loft surface. wise nodes. The grid independent solution appears to occur
when there are approximately 80 streamwise grid points.

Output variables and objective functions Preliminary investigations

Ten output variables were used to monitor the mixing effi- A preliminary investigation was carried out to determine
ciency but only two of them were used as objective func- the behavior and the main characteristics of the problem that

AIChE Journal July 2009 Vol. 55, No. 7 Published on behalf of the AIChE DOI 10.1002/aic 1729
Figure 7. Response surfaces of the optimization procedure for the variance of the solid concentration inside the
vessel.

is being examined. The aim is to analyze which variables the interaction between the root chord size and the tip chord
affect most the problem. The distribution of the initial impel- size as a function of the variance. The surface presents a
lers and their variation on the design space is shown in Fig- minimum and it indicates a root chord and tip chord around
ure 6.36,37 15 cm. The interaction between the D/T ratio and tip chord
Initial 2-D (two-dimensional) CFD studies show that the angle in Figure 7c also shows a minimum for a small var-
DAE 11 airfoil provide a better lift/drag relationship, and iance. The tip chord angle is again around 90 and the D/T
have a gentle stall behavior when compared to the other air- ratio is around 0.5. Figure 7d shows the D/T ratio interaction
foils studied. Therefore, in order to minimize the number of with the root chord angle. In this case there is more than
input parameters and reduce the computational time require- one minimum for a small variance. The root chord angle
ments, the root profile and the tip profile was fixed as the should be between 40 and 55 and the D/T ratio between
DAE 11. 0.46 and 0.5.
Figure 7 shows the relationship between the solid concen- Figure 8 shows the relationship between the impeller
tration variance and the input parameters (root chord angle, pumping effectiveness and the input parameters (root chord
tip chord angle, impeller diameter, root chord and tip chord). angle, tip chord angle, impeller diameter, root chord and tip
Figure 7a indicates that for a minimum variance (one of the chord). In this case the aim is to achieve a high pumping
objectives of this work) the interaction between the tip chord effectiveness. Figure 8a presents the maximum impeller
angle and the root chord angle indicates that the tip chord effectiveness for a tip chord angle around 90 , and a root
angle should be around 90 or a little over, and the root chord angle around 70 . Figure 8b shows that the maximum
chord angle should be around 50 and 55 . Figure 7b shows pumping effectiveness is for a 19 cm root chord size and a

1730 DOI 10.1002/aic Published on behalf of the AIChE July 2009 Vol. 55, No. 7 AIChE Journal
 Figure 8. Optimization response surfaces for pumping effectiveness.

tip chord size around 13.5 cm. The interaction between the 15 generations of 30 individuals (450 optimization cycles),
D/T ratio and tip chord angle in Figure 8c also shows that which resulted in about 98 days of computing time.
the maximum for the tip Chord angle is again around 90 , The performed evaluations allowed for the establishment
and the D/T ratio is around 0.44 and 0.5. Figure 8d shows of a general tendency and for the definition of the influence
the D/T ratio interaction with the root chord angle. In this of each parameter on the estimated values. Furthermore, an
case there is more than one maximum at the response sur- experimental validation was performed (the results will be
face. The root chord angle should be around 70 , and the D/ discussed in the following section), and the solid concentra-
T ratio range between 0.44 or 0.5. tion variance was reduced by 48.5%, the power consumption
reduced by 84.4%, and the pumping effectiveness increased
by 410.2%, when compared to the performance of a pitched
blade impeller (also with a DAE 11 blade profile).
Optimization Results The initial pitched blade impeller (constant tip chord angle
It is important to mention that the computational require-
45 ), has a low-discharge angle and low-solid suspension.
ments for one optimization step (one impeller design) took The pumping effectiveness is low due to the radial velocity
approximately 5.2 h of computing time on a two processor component of the PTB45. The generated flow shows a
AMD AthlonTM MP 2800þ machine, 2 GB RAM. The opti- boundary layer separation and there is a blade stall due to
mization process was configured so as to guarantee the the high tip chord angle.
robustness of the calculation. The optimization method Figure 9a shows the velocity contour plot and the velocity
arrived at some very interesting results, and it required only vector plot for the PBT45. It can be easily noticed that the

AIChE Journal July 2009 Vol. 55, No. 7 Published on behalf of the AIChE DOI 10.1002/aic 1731
 objective of the work specified in the optimization procedure
by a higher pumping effectiveness.
After the optimization steps using modeFRONTIER, a fur-
ther refinement on the impeller tip was done. It was tapered
as most high-efficiency impellers are. The objective func-
tions input into the optimization model do not account for
this kind of refinement. A smooth joining between the impel-
ler blade and the hub was also designed to reduce the bound-
ary layer separation on this region and improve the blade
suction. These changes improved the impeller performance
as expected. The final impeller design is shown in Figure 11.
It is important to notice that the results in Figure 10 do not
show the tapering of the impeller tip.

Figure 9. Velocity for the 45 pitched blade and opti-

mized Impellers.

PBT does not have a pure axial flow. This fact has already
been mentioned before in the literature and can even change
the pattern of the PBT 45 (see, for example, Kresta and
Wood38). Due to that, there is a poor solid suspension in the
region below the impeller, as can be noticed in Figure 10a.
Figures 9b shows that the optimized impeller has a very
strong axial flow, resulting in a higher pumping effectiveness
and, consequently, a higher solid suspension, as shown in
Figure 10b. It can be noticed that the solid concentration at
the bottom of the vessel is very low, resulting in a low var-
iance, which is the characteristic of a very well homogenized
suspension.
It can be said that the optimization procedure resulted in
an impeller with a very pure axial flow below the impeller.
It makes sense since a pure axial flow results in a flow direc-
tion which minimizes solid settling at the bottom of the ves-
sel (low variance as specified by the optimization constraint).
The optimized impeller has only a very small region with
boundary layer separation, just a small point near the hub
and it shows a very low-tip vortex, when compared to the
PBT 45 . It also makes sense, since this shows that no
energy consumption is lost in vortex generation or radial Figure 10. Solids distribution for the 45 pitched blade
flow. The energy is only for axial pumping, which is the and optimized impellers.

1732 DOI 10.1002/aic Published on behalf of the AIChE July 2009 Vol. 55, No. 7 AIChE Journal
 Figure 13. Comparison between computational and ex-
perimental results.
Figure 11. Optimized blade design.
noticed that both impellers are suspending the solids, but the
optimized impeller is suspending much more solids. Figure
Experimental Tests
13 shows the power consumption obtained by the CFD
In order to validate the CFD/Optimization model, some model and the experimental tank. The results are very simi-
experimental tests were performed. Figure 12 compares two lar in the region where experimental results were obtained.
situations in which the optimized impeller performed much
better than the PBT impeller. The first two figures in Figure
12 show that for the same power input of approximately 6 Summary and Conclusions
W, the optimized impeller suspended much more solids than This multiobjective optimization model for an optimal
the PBT impeller. For a power input of 11 W, it can be impeller design contains many innovative elements. It is

Figure 12. Experimental comparison between proposed impeller and the PBT impeller.

AIChE Journal July 2009 Vol. 55, No. 7 Published on behalf of the AIChE DOI 10.1002/aic 1733
believed that the incorporation of parameterization refine- T ¼ tank diameter, dimensions L
ments such as tip construction and smoothness of the impel- C ¼ bottom clearance, dimensions L
H ¼ height of the liquid, dimensions L
ler surface helped in achieving better results.
This research shows that an optimization process is viable
to determine promising impeller designs and a brief outline
of the methodology to achieve that purpose has been pre-
Literature Cited
sented.
This work also indicates the advantages of coupling com- 1. Joaquim Jr CF, Cekinski E, Nunhez JR, Urenha LC. Agitação e
Mistura na Indústria. 1st ed. Rio de Janeiro, Brazil: LTC Editora;
putational fluid dynamics and multiobjective design optimi- 2007.
zation methods. 2. Oldshue JY. Fluid Mixing Technology. 1 ed. New York: McGraw
Hill; 1983.
3. Paul EL, Atiemo O, Kresta SM. Handbook of Industrial Mixing,
Acknowledgments Science & Practice. 1st ed. Hoboken, NJ: John Wiley & Sons, Inc;
The authors would like to thank all colleagues of ESSS and Esteco, 2004.
who helped in the development of this work and also provided the soft- 4. Fasano JB, Bakker A, Penney WR. Advanced impeller geometry
ware licenses used in this project, and CenPRA, that kindly manufac- boosts liquid agitation. Advanced liquid agitation. Chemineer Chem
tured the impeller prototypes. Eng. May 1999.
5. Myers KJ, Bakker A. Solids suspension with up-pumping pitched-
blade and high-efficiency impellers. Can J Chem Eng.
Notation 1998;76:4233–440.
6. Pantula PRK, Ahmed N. Solids Suspension and Gas Holdup in
Throughout this article, dimensions are given in terms of the Three Phase Mechanically Agitated Reactors. In: Chemeca ’98, Port
fundamental magnitudes of length (L), mass (M), time (T). Douglas, Queensland, Australia, 1998. Paper No. 132.
7. Pantula PRK, Ahmed N. The Impeller Speed Required for Com-
This section lists symbols used in this article, their meaning, plete Solids Suspension in Aerated Vessels: A Simple Correlation?
dimensions and, where applicable, their values. Dimensionless In: Re´cents progres en genie des procéde´s (Mixing IX), Paris,
quantities are denoted by 1. The values of physical constants France; 1997:11–18.
(or their default values) are also given. 8. Ibrahim S, Nienow AW. Particle suspension in turbulent regime:
the effect of impeller type and impeller/vessel configuration. Trans
IChemE. 1996;74(A):679–688.
Symbols 9. Hicks MT, Myers KJ, Bakker A. Cloud height in solids suspen-
sion agitation. Chem Eng Commun. 1997;160:137–155.
Ce1 ¼ k-e turbulence model constant, dimensions 1, value 1.44 10. Drewer GR, Ahmed N, Jameson GJ. Suspension of High Concen-
Ce2 ¼ k-e turbulence model constant, dimensions 1, value 1.92 tration Solids in Mechanically Stirred Vessels. IChemE Symp. Se-
Cl ¼ k-e turbulence model constant, dimensions 1, value 0.09 ries. No. 136. Proceedings of the 8th European Conference on Mix-
E ¼ constant used for near-wall modeling, dimensions 1, value ing. 1994:41–48.
9.793 11. Nienow AW. The Suspension of Solid Particles. In: Harnby N,
g ¼ gravity vector, dimensions L T
2, value 9.81 Edwards MF, Nienow AW, eds. Mixing in the Process Industries.
k ¼ turbulence kinetic energy per unit mass, dimensions L2 T
2 2nd ed. London, U.K: Butterworths; 1992:364–393.
Pk ¼ shear production of turbulence, dimensions M L
1 T
3 12. Nienow AW, Konno M, Bujalski W. Studies on Three-phase Mix-
p, pstat ¼ static (thermodynamic) pressure, dimensions M L
1 T
2 ing: A Review and Recent Results. In: Proceedings of the 5th Euro-
pref ¼ reference pressure, dimensions M L
1 T
2 pean Conference on Mixing. Wurzburg, West Germany; 1985:1–13.
ptot0 ¼ total pressure, dimensions M L
1 T
2 13. Weetman RJ, Coyle CK. The Use of Fluidfoil Impellers in Viscous
p ¼ modified pressure, dimensions M L
1 T
2 Mixing Applications. AIChE 1989 Annual Meeting, San Francisco,
Re ¼ Reynolds number, dimensions 1 CA; Nov. 5–10, 1989.
r ¼ location vector, dimensions L 14. Buurman C, Resoort G, Plaschkes A. Scaling-up rules for solids
SM ¼ momentum source, dimensions M L
2T
2 suspension in stirred vessels. Chem Eng Sci. 1986;41:2865–2871.
Sct ¼ turbulent Schmidt number, dimensions lt/Ct, value 1 15. Rushton JH, Costich EW, Everett HJ. Power characteristics of
U ¼ vector of velocity Ux,y,z, dimensions L T
1 mixing impellers Part II. Chem Eng Prog. 1950;46(9):467–476.
U ¼ velocity magnitude, dimensions L T
1 16. Wu JY, Zhu PC, Pullum L. The effect of impeller pumping and
u ¼ vector of fluctuating velocity component ux,y,z fluid rheology on solids suspension in a stirred vessel. Can J Chem
¼ turbulent flow, dimensions L T
1 Eng. 2001;79:177–186.
e ¼ turbulence dissipation rate, dimensions L2 T
3 17. Wu JY, Zhu PC, Bandopadhayay L, Pullum PC, Shepherd IC.
n ¼ impeller pumping effectiveness Solids suspension with axial flow impellers. AIChE J. 2000;46:647–
f ¼ bulk viscosity, dimensions M L
1 T
1 650.
j ¼ Von Karman constant, dimensions 1, value 0.41 18. Rewatkar VB, Raghava Rao KSMS, Joshi JB. Critical speed for
l ¼ molecular (dynamic) viscosity, dimensions M L
1 T
1 solid suspension in mechanically agitated three-phase reactors: 1.
lt ¼ turbulent viscosity, dimensions M L
1 T
1 Experimental part. Ind Eng Chem Res. 1991;30:1770–1784.
leff ¼ effective viscosity, lþlt, dimensions M L
1 T
1 19. Raghava Rao KSMS, Rewatkar VB, Joshi JB. Critical impeller
q ¼ density, dimensions M L
3 speed for solid suspension in mechanically agitated contactors.
rk ¼ turbulence model constant for the k equation, dimensions 1, AIChE J. 1988;34:1332–1340.
value 1.0 20. Wong CW, Wang JP, Huang ST. Investigations of fluid dynamics
re ¼ k-e turbulence model constant, dimensions 1, 1.3 in mechanically stirred aerated slurry reactors. Can J Chem Eng.
rw ¼ k-x Turbulence model constant, dimensions 1, 2 1987;65:412–419.
s ¼ shear stress or sub-grid scale stress, dimensions M L
1 T
2 21. Zwietering TN. Suspending of solid particles in liquid by agitators
x ¼ angular velocity, dimensions T
1 Chem Eng Sci. 1950;8:244–253.
yþ ¼ Y plus 1 22. Aidari AH, Bakker A, Oshinowo LM. Realize greater benefits
P ¼ power consumption from CFD. Chem Eng Prog. 2001;97(3):45–53.
N ¼ rotational speed, dimensions T
1 23. modeFRONTIER:The Multi-Objective Optimization and Design
Np ¼ power number, dimensions 1 Environment. http://www.esteco.it; 2008.
Nq ¼ pumping number, dimensions 1 24. Boyd S, VandenBerghe L. Convex Optimization. Cambridge Uni-
D ¼ impeller diameter, dimensions L versity Press 2004.

1734 DOI 10.1002/aic Published on behalf of the AIChE July 2009 Vol. 55, No. 7 AIChE Journal
25. Papalambros PI, Wilde DJ. Principles of Optimal Design: Modeling 33. Caradonna FX, Tung C. Experimental and analytical studies of a
and Computation. Cambridge University Press; 2000. model helicopter rotor in hove. Vertica. 1981;5:149–161.
26. Dixon JW, Mood AM. A Method for Obtaining and Analyzing 34. Holland JH. Adaptation in Natural and Artificial Systems: An Intro-
Sensitivity Data. Journal of the American Statistical Association. ductory Analysis with Applications to Biology, Control and Artificial
1948;43:109–126. Intelligence. 2nd ed. MIT Press; 1992.
27. Menter FR. Two-equation eddy-viscosity turbulence models for en- 35. Avriel M. Nonlinear Programming: Analysis and Methods. Dover
gineering applications. AIAA J. 1994;38:1598–1605. Publishing 2003.
28. Wilcox DC. Turbulence Modeling for CFD. 2nd ed. DCW Industries; 36. Neyer BT. A D-optimality-based sensitivity test. Technometrics.
1998a. 1994;36:61–70.
29. Wilcox DC. Reassessment of the scale-determining equation for 37. Neyer BT. Analysis of Sensitivity Tests, MLM-3736. EG&G Mound
advanced turbulence models. AIAA J. 1988b;26:1299–1310. Applied Technologies, Miamisburg, Ohio; 1992.
30. Gopalarathnan A, Selig MS. Low speed natural-laminar-flow air- 38. Kresta SM, Wood PE. The mean flow field produced by a 45
foils: case study in inverse airfoil design. J Aircraft. 2001;5:38. pitched blade turbine: changes in the circulation pattern due to off
31. Xu G, Sankar LN. Computational study of horizontal axis wind bottom clearance. Can J Chem Eng. 1983;71:42–53.
turbines. J Sol Energy Eng. 2000;5:122.
32. Lyon CA, Broere AP, Giguere P, Gopalarathnam A, Selig MS. Sum- Manuscript received Mar. 4, 2008, revision received Oct. 31, 2008, and final
mary of Low-Speed Airfoil Data. SoarTech Publications; 1997:3. revision received Dec. 12, 2008.

AIChE Journal July 2009 Vol. 55, No. 7 Published on behalf of the AIChE DOI 10.1002/aic 1735