Proceedings of ASME Turbo Expo 2010: Power for Land, Sea and Air

GT2010
June 14-18, 2010, Glasgow, UK

GT2010-

OPTIMAL PERFORMANCE OF A MODIFIED THREE-BLADE SAVONIUS TURBINE

USING FRONTAL GUIDING PLATES

Mohamed H. Mohamed
mohamed.mohamed@st.ovgu.de

GaborJaniga
janiga@ovgu.de

Pap
Elemer
elemer.pap@ovgu.de

Dominique Thevenin
thevenin@ovgu.de

Laboratory of Fluid Dynamics and Technical Flows

University of Magdeburg Otto von Guericke
Magdeburg, Germany

during the automatic optimization process, carried out by

coupling an in-house library (OPAL) relying on Evolutionary Algorithms with an industrial flow simulation code
(ANSYS-Fluent). The output power coefficient is the single
target function and must be maximized. The relative performance improvement amounts to more than 50% at the
design point compared with the classical configuration.

ABSTRACT

Wind energy is one of the most promising sources of renewable energy. It is pollution-free, available locally, and
can help in reducing the dependency on fossil fuels. Although a considerable progress has already been achieved,
the available technical design is not yet adequate to develop reliable wind energy converters for conditions corresponding to low wind speeds and urban areas. The Savonius turbine appears to be particularly promising for such
conditions, but suffers from a poor efficiency. The present
study considers an improved design in order to increase the
output power and the static torque of the classical threeblade Savonius turbine, thus obtaining a higher efficiency
and better self-starting capability. To achieve this objective three geometrical properties are optimized simultaneously: 1) the position of an obstacle shielding the returning blade; 2) the position of a deflector guiding the wind
toward the advancing blade; and 3) the blade skeleton line.
As a whole, fifteen free parameters are taken into account

KEYWORDS

Savonius rotor, Wind energy conversion, Optimization,

Evolutionary Algorithms, Turbomachines.

Introduction

Wind power is the conversion of wind energy into a

useful form of energy, such as electricity, using wind turbines. At the end of 2008, worldwide nameplate capacity of wind-powered generators was 121.2 GW. At that
time, wind power accounted for roughly 1.5% of world1

c 2010 by ASME
Copyright

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 01/30/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

Nomenclature

Projected area of rotor (DH)

Cm

Torque coefficient, T /(R2 HU 2 )

Cms

Static torque coefficient, Ts /(R2 HU 2 )

CP

Power coefficient, P/(1/2AU 3 )

Blade chord (2r)

Turbine diameter (2R)

Blade height

Rotational speed of rotor

Output power (2NT /60)

Tip radius of turbine

Rsh

Turbine shaft radius

Radius of semi-cylindrical blade

Gap width

Output torque

Ts

Static torque

Mean velocity in x-direction

Obstacle angle

Deflector angle

Speed ratio

Orientation angle

Density

Angular speed

Figure 1.

Savonius rotor

generation technology that can deliver the necessary cuts in

CO2 during the critical period up to 2020, when greenhouse
gases must begin to decline in order to avoid dangerous climate change. It has been estimated that the installed wind
capacity will produce 260 TWh and save 158 million tons
of CO2 every year [2].
The storage and distribution of electrical power is still
a major problem, in particular when the generated quantity
is varying considerably with time and location, like is the
case for wind energy. A local electricity production, within
urban areas, would help solve this issue. The Savonius turbine appears in principle to be particularly promising for
such conditions, since it is a slow-running machine with a
very compact design.
The Savonius Turbine

S.J. Savonius initially developed the vertical axis Savonius rotor in the late 1920s. The concept of the Savonius
rotor is based on cutting a cylinder into two halves along
the central plane and then moving the two half cylinders
sideways along the cutting plane, so that the cross-section
resembles the letter S (Fig. 1, [3]).
The Savonius rotor, which is a slow-running vertical
axis wind machine (typically used for ' 1.0 or below, see
Eq. 1) has unfortunately a poor efficiency when considering the standard design: theoretically, Cp ' 0.2 at best [4].
Nevertheless, it presents many advantages for specific applications, in particular due to its simplicity, resulting robustness, compactness and low cost. If a higher efficiency
could be obtained, the Savonius rotor would become a very
interesting complementary source of electricity.

wide electricity usage. This amount is growing rapidly,

having doubled between 2005 and 2008. Several countries
have achieved relatively high levels of wind power penetration, such as 19% of stationary electricity production
in Denmark, 11% in Spain and Portugal, and 7% in Germany and the Republic of Ireland. As of May 2009, eighty
countries around the world are using wind power on a commercial basis [1]. Wind energy is perhaps the only power

c 2010 by ASME
Copyright

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 01/30/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

These previous studies have demonstrated that Savonius turbines show considerable drawbacks compared to
conventional turbines, in particular a low efficiency and
poor starting characteristics. Previous investigations of our
research group have shown that guiding and obstacle plates
placed appropriately in front of the turbine might increase
its efficiency [6,7]. Using a contoured shape might be even
more promising but would require considerably more free
parameters and will therefore be considered in a later step.
Building on top of our previous results, the blade shape will
be simultaneously modified in the present work. The optimization process thus relies on free design variables that
describe the position and angles of the plates, the blade
shape (skeleton line) as well as the gap width s (Fig. 3).
At the end, fifteen free parameters are thus considered (X1 ,
Y1 , X2 , Y2 , Xd1 , Yd1 , Xd2 , Yd2 , XP1 , YP1 , XP2 , YP2 , XP3 , YP3
and s). The objective function considers only one output of
the simulation, that should be maximized: the output power
coefficient Cp .

Advancing
Blade

Wind
(0,0)

Returning
Blade

Figure 2.

Schematic description and main parameters characterizing a

Savonius rotor

Performance of a Savonius Turbine

Using the notations of Fig. 2, the velocity coefficient is

defined as:
= R/U

(1)

For a Savonius rotor of height H, a wind of incoming

velocity U, the mechanical power P and the mechanical
torque on the axis of a Savonius turbine can respectively be
written as follows:
Cp =

P
RHU 3

Optimization Methodology

Optimization is a body of mathematical results and numerical methods for finding and identifying the best candidate from a collection of alternatives, without having to
explicitly enumerate and evaluate all possible alternatives.
Optimization is a key engineering task, since the function
of any engineer is to design new, better, more efficient, and
less expensive systems as well as to devise plans and procedures for an improved operation of existing systems. Nevertheless, such a real optimization relying on suitable algorithmic procedures is still a relatively new approach, in
particular when considering turbomachines [8, 9].
The central goal when designing an improved Savonius turbine is to achieve high efficiency, i.e., high power
output. Furthermore, it must be kept in mind that turbomachines often operate outside the nominal (or design) conditions. Therefore, after optimizing the configuration for the
maximum output power coefficient, known to occur for a
speed ratio ' 0.7, the full range of speed ratios will be
considered.
In our group, a considerable experience is available
concerning the mathematical optimization relying on evaluations based on Computational Fluid Dynamics (CFD)
[10]. We therefore employ our own optimization library,
OPAL (for OPtimization ALgorithms), containing many

(2)

and
Cm =

T
R2 HU 2

(3)

where Cp and Cm are respectively the power coefficient

and the torque coefficient. In the following sections, a rotor is called a conventional Savonius rotor if semi-cylinder
blades are used without any flow guiding plates.
Purpose of the Present Work

The conventional, three-blade configuration of the rotor has been extensively studied in the past [5]. The corresponding values of Cp and Cm have been determined numerically and sometimes experimentally as a function of
the speed ratio . This has been used to validate extensively our numerical procedure by comparison with published, reference data [6].
3

c 2010 by ASME
Copyright

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 01/30/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

Xd2

(a)

CFD coupling with OPAL (optimizer)

Xd1

New configuration

X1>R

Input
Input file
file
(parameter
(parameter values)
values)

Advancing
Blade

Yd1

Yd2

Deflector

Wind

Gambit
Gambit
Journal
Journal file
file

Fluent
Fluent
Journal
Journal file
file

Gambit 2.4

Fluent 6.3

Geometry
++
Mesh
Mesh

CFD
Simulation
+

OPAL
Evolutionary
Evolutionary
Algorithms

(0,0)

Y2

Obstacle

Y1

x
Output
Output file
file
(objective
(objective value)
value)

Post-processing
Post-processing

X2>R
Returning
Blade

C
C program
program on
on Linux
Linux for
for automatization
automatization

P4

(b)

P3(XP3,YP3)
Y
Fixed points

Blade
center
(P0)

Figure 4.
pling.

r
X

a mutation probability of 100% with decreasing mutation

amplitude is considered. All further details can be found
elsewhere [6], where the same procedure has been applied
to a simpler configuration.
A fully automatic optimization finally takes place, using OPAL (decision-maker for the configurations to investigate), the commercial tool Gambit for geometry and grid
generation (including quality check) and the industrial CFD
code ANSYS-Fluent to compute the flow field around the
Savonius turbine. As a result of the CFD computation the
output power coefficient is determined, and is stored in a result file. The procedure is automated using journal scripts
(to restart Gambit, Fluent) and a master program written
in C, calling all codes in the right sequence as shown in
Fig. 4. By checking the values stored in the result file,
OPAL is able to decide how to modify the input parameters
before starting a new iteration. The fully coupled optimization procedure is a complex task, which has been described
in detail in previous publications [1012].

P1(XP1,YP1)
Variable points

S
P2(XP2,YP2)
P5
Turbine shaft

Figure 3.

Schematic description of optimizer (OPAL) and CFD code cou-

Schematic description of the free optimization parameters

characterizing a three-blade Savonius rotor :

a) plate parameters

(X1 ,Y1 , X2 ,Y2 , Xd1 ,Yd1 , Xd2 and Yd2 ); b) XP1 ,YP1 , XP2 ,YP2 , XP3
and YP3 used to modify the blade shape; additionally, the gap width s.

different optimization techniques. Different CFD solvers

(in-house codes, ANSYS-Fluent, ANSYS-CFX) have been
coupled in the past with this optimizer. It has already
been employed successfully to improve a variety of applications like heat exchangers [11], burners [12] or turbomachines [6, 7, 13, 14]. These studies have in particular demonstrated the efficiency of evolutionary algorithms
(EA) for CFD-based optimization. Using EA, a very robust procedure can be obtained and local extremal values
do not falsify the results [15, 16]. Therefore, the present
study relies again on EA with a population size of 20 in
the first generation and a total number of generations of
23. The optimization process stops automatically after this
twenty-third generation, since the observed progress in the
objective function falls below the user-prescribed threshold. When computing a new generation from the previous
one, a survival probability of 50%, an averaging probability of 33.3% and a crossover probability of 16.7% (total
100%) are implemented. In the additional mutation step,

Numerical Flow Simulations

From the literature it is known that an accurate CFD

simulation of the flow around a Savonius turbine is a challenging task, mainly due to its highly time-dependent nature and to the fact that flow separation plays an important
role for the efficiency of the system. It is therefore necessary to check the full numerical procedure with great care.
Afterwards, the resulting methodology must be validated.
All flow simulations presented in this work rely on
the industrial software ANSYS-Fluent 6.3. The unsteady
Reynolds-Averaged Navier-Stokes equations are solved
4

c 2010 by ASME
Copyright

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 01/30/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

using the SIMPLE (Semi-Implicit Method for Pressurelinked Equations) algorithm for pressure-velocity coupling.
The flow variables and all turbulent quantities are discretized in a finite volume formulation using a secondorder upwind scheme. For the present configuration, twodimensional simulations are sufficient (no geometry change
in the third direction when excluding boundary effects), so
that very fine grids can be employed.
The unsteady flow is solved by using the Sliding Mesh
Model (SMM). Three complete revolutions are always
computed, using a constant time-step; the first one is used
to initiate the correct flow solution, while the flow properties (in particular the power coefficient Cp and the torque
coefficient Cm ) are obtained by averaging the results during
the last two revolutions. On a standard PC, one evaluation
(i.e., three revolutions for one specific configuration) takes
about 280 minutes of computing time.
A grid-independence study has been first carried out
for one geometrical configuration. Several different twodimensional, unstructured grids of increasing density and
quality, composed of 3 400 up to 116 000 cells, have been
tested for the standard Savonius turbine with a specified obstacle plate. All grids employing more than 71 000 cells
lead to a relative variation of the output quantity below
1.8% [6]. Since the cost of a CFD evaluation obviously
increases rapidly with the number of grid cells, the intermediate grid range between 75 000 and 95 000 cells has been
retained for all further results shown in the present paper.
The grid is refined in the vicinity of the turbine and of the
solid surfaces, capturing all relevant flow features (Fig. 5).
The minimum size of the computational domain has been
checked in a separate project [17]. It has been found that a
domain size equal to 27 times the rotor radius is needed to
get a result independent from any influence of the boundary conditions. This is in agreement with previous studies,
mostly recommending 10 times the rotor radius on each
side of the turbine.

Figure 5.

Zoom on the two-dimensional, unstructured grid in the vicinity

of the turbine

strate the excellent agreement obtained between CFD and

experiments for the target function, Cp , when using the realizable k turbulence model. A similar tendency has
been observed for other studies involving cambered blades
[13, 14], proving the interest of the realizable k model
for fast CFD simulations. The fact that the Reynolds Stress
Model does not lead to an improvement compared to standard two-equation models is probably a result of its higher
sensitivity toward inflow turbulent boundary conditions,
which are usually not measured in the experiments.
The near-wall treatment relies on standard wall functions. The y+ -values found near all walls in the employed
grid are around 60 and fall therefore within the recommended range for best-practice CFD (30 < y+ < 300). A
boundary layer grid is employed near the surfaces.
Since many different configurations must be evaluated
during the optimization, the computing time associated
with one single CFD computation must be kept acceptable.
As a consequence, only three rotor revolutions are considered, using an appropriate, constant value of the time-step.
The first revolution is only used to initiate a correct flow
solution, while the interesting flow properties (in particular
the power coefficient Cp and the torque coefficient Cm ) are
obtained by averaging the CFD results during the two further revolutions. We have checked separately the influence
of the number of revolutions on the computed turbine performance for the optimal design (Fig. 7), by continuing the
CFD simulation. After 10 revolutions, the average power
coefficient reaches a constant value. The absolute differ-

CFD Validation

After an acceptable grid and domain size have been

identified, the full numerical procedure and in particular the
employed turbulence model have been validated by comparison with published experimental results for a classical
three-blade Savonius turbine [5]. The influence of the turbulence model is shown in Fig. 6. These results demon-

c 2010 by ASME
Copyright

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 01/30/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

No. of revolutions

0.2

20

0.15

0.1

60

80

100

Exp. (K. Irabu & J. Roy, 2007)

Realizable k-e model
SST model
Standard k-e model
RSM model

0.05

l =0.7
Instantaneous power coeff.
Average Power Coeff.

0.8

Employed number of revolutions

0.4

0
0.2

0.4

0.6

0.8

1.2

15

30

45

60

75

Time

Speed ratio(l)
Figure 6.

40

1.2

Power coefficient (Cp)

Power Coefficient (Cp)

Figure 7.

Validation of computational model: power coefficient com-

Influence of the number of revolutions on the instantaneous

and on the average power coefficient C p computed by CFD for the opti-

pared to experimental results for a three-blade Savonius turbine [5]

mum design shown later.

ence in Cp between this value and the one obtained after

only three revolutions equals 0.024. This is an estimation
of the uncertainty associated with the considered optimization process. As shown later in Fig. 8, this inherent uncertainty is very small compared to the range of Cp explored
during the optimization and is thus deemed acceptable. It
amounts to only 6% of the pressure coefficient associated
with the optimal design. The influence of the number of
revolutions on the estimation of Cp by CFD has been investigated systematically in a separate project [17], confirming
the present findings. Only three revolutions have been thus
computed for each design in order to reduce the needed
computational time.

all 5 points, the full profile is reconstructed using standard

splines (Nonuniform rational B-splines, NURBS). The order of a NURBS curve defines the number of nearby control points that influence any given point on the curve. The
curve is represented mathematically by a polynomial of degree one less than the order of the curve. This means that
the spline order is 5 in our case and the degree of the polynomial is 4. The objective function contains one single output of the simulation, that should be maximized as far as
possible: the power coefficient Cp . The parameter space
considered in the optimization has been defined as documented in Table 1. These domains are selected to prevent
any domain overlap along the Y -direction, to keep realistic
blade shapes and to cover a wide region for positioning the
guiding plates. The reference point of the parameter space
for the blade skeleton line is point P0 , which is the center of
the original, semi-cylindrical shape with radius r as shown
in Fig. 3. The reference point for the remaining space parameters (guiding plates and gap width) is the global center of turbine rotation. During the calculations, a circular
turbine shaft is included with a radius Rsh computed from
Rsh /R = 0.03.
The results presented in Fig. 8 indicate that the considered objective is indeed considerably influenced by the
fifteen free parameters. As a whole, 240 different geometrical settings have been evaluated by CFD, requesting 47
days of total computing time on a standard PC. Note that
the user-waiting time could be considerably reduced by carrying out the requested CFD on a more powerful computer

Results and Discussions

Optimization of all turbine parameters

The mathematical optimization procedure described

previously (Evolutionary Algorithms relying on automated
evaluations through CFD) can now be employed to find the
optimal position of both guiding plates, the gap width and
the blade shape. This is done first for a constant speed
ratio = 0.7, considering a fixed incident wind velocity
U = 10 m/s. This value of is retained since it is known
from the literature that it corresponds to the zone of peak
power coefficient of the conventional Savonius turbine. As
explained previously, fifteen degrees of freedom are left
simultaneously for the optimization (Fig. 3). Regarding
the blade shape, the points P4 and P5 are considered to
be fixed; only P1 , P2 , P3 are changing position. Knowing

c 2010 by ASME
Copyright

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 01/30/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

Table 1.

Optimum
configuration

Acceptable range for the input parameters (parameter space)

Parameter

Minimum allowed

Maximum allowed

Blade shape

Guiding plates positions

Gap width

Blade shape
Xp1/r

XP1 /r

0.53

1.2

YP1 /r

0.24

0.24

XP2 /r

0.24

1.1

YP2 /r

0.94

0.24

XP3 /r

0.24

1.1

YP3 /r

0.24

0.94

s/R
X2/R
Xd1/R
Xd2/R
X1/R
CP
X2/R
Y1/R
Yd2/R
Yp2/r
Yd1/R
Yp3/r
1.1 -0.24 1.1 0.94 0 1.65 0 2.6 -1.1 0 -1.1 -0.7 0.18 0.38

Xp2/r

Yp1/r
1.2 0.24

Xp3/r

0.363
Optimum
configuration

0.024 Absolute
uncertainty

0.156

Three-blade
Savonius turbine
without guiding
plates

Guiding plates
X1d /R

1.2

0.0

Yd1 /R

1.1

1.65

X2d /R

1.88

0.0

Figure 8.

Yd2 /R

1.76

2.6

uration are connected with a thick gray line. The power coefficient of the

X1 /R

1.88

1.1

conventional three-blade turbine (semi-cylindrical shape) is also shown

Y1 /R

0.7

0.0

X2 /R

1.88

1.1

Y2 /R

1.88

0.7

0.53 -0.24 0.24 -0.94 0.24 0.24 -1.2 1.1 -1.88 1.76 -1.88 -0.7 -1.88 -1.88 0.03 0.024

Input parameters of the optimization and power coefficient rep-

resented using parallel coordinates. The parameters of the optimal config-

with a black circle.

Deflector

Gap width
s/R

0.03

Advancing
Blade

Wind

0.18

or in parallel on a PC cluster [10]. Such a parallel procedure, already implemented in OPAL, has not been used in
the present case but could reduce the needed time by orders
of magnitude, as demonstrated in other studies. Relying
on parallel computers and possibly carrying out each CFD
evaluation again in parallel [10] is clearly necessary when
considering three-dimensional problems. Fortunately, it
is quite straightforward to implement, so that researchers
having access to parallel clusters can solve corresponding
problems within an acceptable lapse of time.
The optimal configuration (highest point in the right
column in Fig. 8, all corresponding parameters being connected by a thick gray line) can now readily be identified for
= 0.7. The corresponding geometry is shown in Fig. 9.
The optimum parameter values are listed in Table 2.

Obstacle
Returning blade
Figure 9.

Optimum configuration obtained with the optimization proce-

dure.

One instantaneous picture of the velocity field is shown

as an example in Fig. 10, demonstrating that the employed
grid captures all important flow features in the vicinity of
the rotor and guiding box. This is of course a dynamic
process, difficult to illustrate in a static figure.
At = 0.7 the optimal point found by the optimization
procedure corresponds to an absolute increase of the power
7

c 2010 by ASME
Copyright

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 01/30/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

Table 2.

Optimal configuration

Part

Parameter

Value

Angle

Blade shape

XP1 /r

0.6077

YP1 /r

-0.1338

XP2 /r

0.2735

YP2 /r

-0.7136

XP3 /r

0.7065

YP3 /r

0.5901

Xd1 /R

-0.3089

Yd1 /R

1.436

Xd2 /R

-0.4591

Yd2 /R

2.388

X1 /R

-1.3638

Y1 /R

-0.1075

X2 /R

-1.691

Y2 /R

-1.5935

s/R

0.0988

Guiding plates

Gap width

= 81.13
Figure 10.

Instantaneous velocity vectors magnitude (m/s) around the

optimum configuration (zoom) at the design point (

= 0.7).

optimized configuration.
= 77.58

Self-starting capability

For decentralized, low-cost wind-energy applications

as considered here, it is essential to obtain a self-starting
system. To investigate this issue, the static torque exerted
on a turbine at a fixed angle has been computed by CFD
as a function of this angle . Figure 12 shows the obtained
static torque coefficient Cms obtained for the optimal design
compared to the classical three-blade turbine. The experimental results of [5] for a conventional three-blade turbine
are also shown for comparison in Fig. 12. Due to periodicity, the results are only plotted for between 0 and
120 . Compared to the classical turbine, these computations demonstrate that the modifications have a considerable and positive effect on the static torque coefficient, except in a small range (90 100 ). There, the static
torque coefficient is less than the classical one, but remains
strictly positive. Averaging over all angle positions, Cms is
increased by 0.091 for the optimum design.

coefficient by 0.207 compared with the conventional threeblade Savonius turbine (semi-cylindrical blade shape). As
a whole, this means a relative increase of the performance
(measured by the power output coefficient) by 57% for the
optimum design.
Off-design performance

It is now important to check how this gain would

change as a function of , since such a turbine must be
able to work also for off-design conditions. Therefore, the
performance of the optimal configuration has been finally
computed for the full range of useful -values, as shown in
Fig. 11. This figure demonstrates that the improvement of
both torque coefficient and power output coefficient is observed throughout for all values of , compared to the conventional three-blade Savonius turbine. The relative performance increase compared to the standard Savonius configuration is always higher than 50% in the usual operating
range (0.6 1), demonstrating again the interest of the

Practical realization

From the technical point of view many existing systems already rely on a tail vane for optimal alignment into
the wind direction. A similar technical solution would be
used for the Savonius turbine using guiding plates. In this
manner the orientation of the system can be simply, efficiently and automatically controlled. As a whole, the optimized configuration is only slightly more complex, more
expensive and heavier than the original system. Therefore,
8

c 2010 by ASME
Copyright

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 01/30/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

Static torque coefficient (Cms)

Conventional three blade

Savonius

(a)

Torque Coefficient (Cm)

0.7
Optimum design
% Relative increase

0.6
0.5
0.4
0.3
0.2
0.1

Exp. K. Irabu & J. Roy (2007)

(Conventional design)

0.6

Our CFD results:

Conventional design
Optimum design

0.4

0.2

0
Negative Torque (no self-starting capability)

0
0.2

0.4

0.6

0.8

1.2

1.4

1.6

(b)

angle

0.45
0.4

80

0.35
0.3

60

0.25
0.2

40

0.15
0.1

20

0.4

0.6

0.8

1.2

1.4

Static torque coefficient Cms as a function of the fixed rotor

for the optimal design (filled squares) compared to the classical

Conclusions

The Savonius turbine is a promising concept for smallscale wind-energy systems, but suffers from a poor efficiency. Therefore, the major objective of the present study
is to identify an improved design, leading to higher values
of the power coefficient and of the static torque of the threeblade Savonius turbine, thus obtaining a higher efficiency
and better self-starting capability. For this purpose, frontal
plates guiding the wind toward the advancing blade are introduced. Simultaneously, the installation of these guiding
plates improve the self-starting capability of the system.
After validating the numerical procedure against experimental measurements, accurate CFD simulations of the
unsteady flow around a conventional three-blade Savonius
turbine have been carried out. The realizable k turbulence model can be employed for a quantitative analysis of
the performance, provided a sufficiently fine grid is used.
The blade shape, position and angles of the guiding
plates and gap width have then been optimized in a fully
automatic manner, in order to obtain the best possible performance, as measured by the power coefficient Cp . The
optimization relies on evolutionary algorithms, while all
geometrical configurations are evaluated by CFD. This optimization procedure is able to identify considerably better configurations than the conventional three-blade Savonius turbine, leading in particular to a relative increase of
the power output coefficient by 57% at = 0.7. A perfor-

1.6

Speed ratio(l)
Figure 11.

120

are also shown for comparison (empty squares).

0
0.2

90

three-blade Savonius turbine (black plus). The experimental results of [5]

0.05
0

60

Figure 12.

100

Relative incease (%)

Power Coefficient (Cp)

0.5

30

Rotation angle(q)

Speed ratio (l)

Performance of the optimized configuration (black squares)

compared to the conventional three-blade Savonius turbine (empty

squares): a) torque coefficient; b) power coefficient. The corresponding
relative increase compared to the standard configuration is shown with
black plus.

the improved power and torque coefficients should easily

compensate these drawbacks within a short time after installation.
It has been shown in Fig. 11 that the optimum design
leads to a broader operating range. It can be in particular
employed for higher speed ratios , making it attractive to
exploit wind energy at higher velocity. The present study
does not consider transient effects (wind gusts, storms), a
problem common to all wind turbines and mostly leading
to material limitations.

c 2010 by ASME
Copyright

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 01/30/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

mance gain of at least 25% is found for the full operating

range of the conventional design. At the same time, the
operating range is extended up to = 1.5. This positive effect is also observed for the torque coefficient. The optimal
design still ensures self-starting capability for all rotating
angles. Therefore, this optimal configuration appears to be
very promising for wind energy generation, in particular in
urban areas.
A further optimization should consider a contoured
nozzle as guiding box in front of the turbine. It would also
be interesting to consider simultaneously the optimization
over the full operating range. In both cases, the computational costs will increase considerably.

ACKNOWLEDGMENT

The Ph.D. work of Mr. Mohamed is supported financially by a bursary of the Egyptian government.

REFERENCES

[1] http://en.wikipedia.org/wiki/Wind power. Last accessed November 2009.

[2] http://www.gwec.net/index.php?id=13. Global Wind
Energy council (GWEC)-Global trends, last accessed
november 2009.
[3] Gupta, R. Biswas, A. and Sharma, K.K., 2008, Comparative study of a three-bucket Savonius rotor with a
combined three-bucket Savonius-three-bladed Darrieus
rotor, Renewable Energy, 33(9), pp. 1974-1981.
[4] Menet, J., 2004, A double-step Savonius rotor for local production of electricity: A design study, Renewable Energy, 29(11), pp. 1843-1862.
[5] Irabu, K. and Roy, J.N., 2007, Characteristics of wind
power on Savonius rotor using a guide-box tunnel, Exper. Thermal Fluid Sci., 32(2), pp. 580-586.
[6] Mohamed, M.H., Janiga, G., Pap, E. and Thevenin,
D., 2010, Optimization of Savonius turbines using an
obstacle shielding the returning blade, Renewable Energy, accepted for publication.
[7] Mohamed, M.H., Janiga, G., Pap, E. and Thevenin,D.,
2008, Optimal performance of a Savonius turbine using an obstacle shielding the returning blade, In Ninth
International Congress of Fluid Dynamics and Propulsion (ASME ICFDP9-EG-249), Alexandria, Egypt.
[8] Bonaiuti, D. and Zangeneh, M., 2009, On the cou10

pling of inverse design and optimization techniques for

the multiobjective, multipoint design of turbomachinery blades, J. Turbomachinery, 131(2) pp. 021014.
[9] Van den Braembussche, R.A., 2008, Numerical optimization for advanced turbomachinery design, In: Optimization and Computational Fluid Dynamics, Berlin,
Heidelberg, Springer-Verlag, pp. 147-188.
[10] Thevenin, D. and Janiga,G., 2008, Optimization
and Computational Fluid Dynamics, Springer, Berlin,
Heidelberg.
[11] Hilbert, R. Janiga, G. Baron, R. and Thevenin, D.
2006, Multiobjective shape optimization of a heat exchanger using parallel genetic algorithms, Int. J. Heat
Mass Transf., 49, pp. 2567-2577.
[12] Janiga, G. and Thevenin, D. ,2007, Reducing the CO
emissions in a laminar burner using different numerical
optimization methods, J. Power Energy, 221(5), pp.
647-655.
[13] Mohamed, M.H., Janiga, G. and Thevenin, D., 2008,
Performance optimization of a modified Wells turbine
using non-symmetric airfoil blades, In ASME Turbo
Expo Conference, (GT2008-50815), Berlin, Germany.
[14] Mohamed, M.H., Janiga, G., Pap, E. and Thevenin,
D., 2008, Optimal shape of a modified Wells turbine
considering mutual interaction between the blades, In
First International Conference of Energy Engineering
ICEE-1, Aswan, Egypt.
[15] Deb, K., 2001, Multi-Objective Optimization using
Evolutionary Algorithms, John Wiley & Sons.
[16] Coello Coello, C.A., Lamont, G.B. and Van Veldhuizen, D.A., 2007, Evolutionary Algorithms for
Solving Multi-Objective Problems, Springer.
[17] Yu, H., 2009, Numerical investigation of a rotating system using OpenFOAM, Masters Thesis LSSM02/09, Univ. of Magdeburg.

c 2010 by ASME
Copyright

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 01/30/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use