Energy 35 (2010) 3349e3363

Contents lists available at ScienceDirect

Energy
journal homepage: www.elsevier.com/locate/energy

Unsteady Aerodynamics of a Savonius wind rotor: a new computational approach

for the simulation of energy performance
V. D’Alessandro*, S. Montelpare, R. Ricci, A. Secchiaroli
Università Politecnica delle Marche, Dipartimento di Energetica, Via Brecce Bianche 1, 60131 Ancona, Italy

a r t i c l e i n f o a b s t r a c t

Article history: When compared with of other wind turbine the Savonius wind rotor offers lower performance in terms
Received 27 October 2009 of power coefficient, on the other hand it offers a number of advantages as it is extremely simple to built,
Received in revised form it is self-starting and it has no need to be oriented in the wind direction. Although it is well suited to be
8 March 2010
integrated in urban environment as mini or micro wind turbine it is inappropriate when high power is
Accepted 10 April 2010
requested. For this reason several studies have been carried-out in recent years in order to improve its
Available online 15 May 2010
aerodynamic performance. The aim of this research is to gain an insight into the complex flow field
developing around a Savonius wind rotor and to evaluate its performance. A mathematical model of the
Keywords:
Savonius
interaction between the flow field and the rotor blades was developed and validated by comparing its
Vertical axis wind turbine results with data obtained at Environmental Wind Tunnel (EWT) laboratory of the “Polytechnic
Computational fluid-dynamics University of Marche”.
Wind tunnel testing
2010 Elsevier Ltd. All rights reserved.

1. Introduction in the prediction of the rotor performance. It is well known that

Wind turbines are usually classified in two main classes: lift
driven wind turbines and drag driven wind turbines. In the former
case the aerodynamic lift is the force responsible for the rotation
whereas in the latter case the aerodynamic drag is the force which
makes the turbine to spin.
The Savonius wind turbine is a vertical axis wind turbine
(VAWT) created for the first time by the Finnish engineer Sigurd
Savonius in 1925. As this rotor is classified as a drag driven device,
aerodynamic theories developed in order to analyze wind turbines
driven by lift force, cannot be applied.
In the field of horizontal axis wind turbine as well as lift driven
vertical axis turbine such as the Darrieus rotor, BEM (Blade Element
Momentum) theory [1] finds its best application.
From what above observed, follows that wind tunnel tests and
computational techniques are the only tools available for studying
Savonius wind rotors.
The aim of this work is to gain an insight into the complex flow
field developing around a Savonius (split-type) wind rotor and to
evaluate its performance.
Several numerical analyses have been performed on Savonius
rotor aerodynamic using both DVM (Discrete Vortex Method) and
CFD methods. Fujisawa [2] and Fernando and Modi [3] used DVM
* Corresponding author. Tel.: þ39 071 220 4359; fax: þ39 071 220 4770.
E-mail address: v.dalessandro@univpm.it (V. D’Alessandro).
this method does not yield a quantitatively accurate prediction of
the rotor performance (in comparison with measured data) but it
reproduces the main features of the performance curves and
flow field as reported in [2,3]. CFD analyses were conducted by
Shinohara and Ishimatsu [4] and Redchyts and Prykhodko [5]. In
particular an unstructured finite volume method was used in [5].
In this analysis Reynolds Averaged NaviereStokes (RANS) equa-
tions were solved for the computation of the turbulent flow using
a third order TVD (Total Variation Diminishing) and a fifth order
Roe flux scheme for the discretization of the convective terms,
while diffusive terms were discretized by a central difference
scheme; SpalarteAllmaras turbulence model [6] was used. Other
numerical studies about Savonius rotor aerodynamic perfor-
mance are available in literature: these simulations were con-
ducted in static conditions, varying the rotor angular position
relative to the wind direction as in [7].
A crucial issue in the analysis of the flow field around the
Savonius rotor is the treatment of the fluidesolid coupling and its
modelling. Modelling the fluid-structure interaction (FSI) is
a problem in many industrial applications. Generally a distinction is
made between three categories of fluidesolid coupling [8]: a one-
way solid to fluid reaction (solid motion influences fluid pattern but
the fluid field does not affects the solid), one-way fluid to solid
reaction (fluid field moves the solid but the latter does not modify
the fluid pattern) and the two-way coupled interaction where
reciprocal influences are modelled. Obviously, in these simulations
the accuracy of the model and its computational cost is heavily

0360-5442/$ e see front matter
2010 Elsevier Ltd. All rights reserved.
doi:10.1016/j.energy.2010.04.021
3350 V. D’Alessandro et al. / Energy 35 (2010) 3349e3363

Nomenclature r fluid density [kg/m3]

n kinematic viscosity [m2/s]
u velocity vector [m/s] m dynamic viscosity [Pa s]
x position vector [m] l Tip-speed ratio
xG rotation center [m] u angular velocity [rad/s]
P Power [W] 3 rate of dissipation of TKE [m2/s3]
C dimensionless coefficient b blockage factor
f generic time-averaged variable h Kolmogorov length scale [m]
hfi generic angular averaged variable f generic fluid-dynamic variable
R rotor radius [m] 3ijk Ricci-Cubastro tensor
Rmg radius of the rotating domain [m]
N number of cells sliding at each time-step Superscript
S stage number of RungeeKutta method (n) time-step number
H polynomial order of resistant torque (k) generic rotor round
Z Number of time-steps for every round T transpose
ek unit vector normal to the fluid flow panel
M torque [Nm] Subscripts
I moment of inertia [kg m2] R resistant
p pressure [Pa] F fluid
k turbulent kinetic energy [m2/s2] S solid
T stress tensor [Pa] a aerodynamics
¼
I identity tensor in domain inlet
¼
Cp Power coefficient out domain outlet
Cm Torque coefficient w wall
uN wind velocity [m/s] G rotation center
f elliptic relaxation factor [1/s] mg moving grid
v2 Reynolds stress normal to the wall [m2/s2] i RungeeKutta method substep
R2 coefficient of determination g grid
Dt time-step size ^ unit vector
k summation index
Greek Symbols m torque
G domain boundary p power
U generic calculus domain

influenced by the assumptions made on the nature of the fluid wind effects over buildings, structures, orography models which
(viscous-inviscid) and on the structure (rigid-deformable). The are subjected to fully developed environmental boundary layers. A
model treated in this work is a two-way coupling type; the struc- schematic representation of the test section is reported in Fig. 2.
ture (rotor) was treated as a rigid body while the fluid was The wind tunnel is supplied by a fan having a constant rotational
modelled as incompressible and viscous. The main problem in the speed of 975 RPM, consisting of 16 blades with an adjustable pitch
fluid-structure interaction modelling is the procedure used to take that ensure a regulated wind velocity in the test section between
into account the motion of the solid body in the solution of the 6 m/s and 40 m/s. Constant Temperature Hot Wire Anemometer
fluid-dynamics equations. The strategy to solve this problem used (CTA HWA) measurements showed a lack of flow uniformity less
in this work was a SMM (Sliding Mesh Model) approach while
NaviereStokes equations were solved using the finite volume code
FLUENT. The solid body motion was treated solving the second
cardinal equation of dynamics by means of a custom MatLab
numerical algorithm able to import CFD data, calculate the rotor
angular velocity and export this variable as input to the CFD code.
Time marching of the solution of the second cardinal equation has
been executed using an Euler method in the initial steps and a four
stage RungeeKutta (or an AdamseBashfort) scheme in the
following steps.

2. Experimental analysis

Experimental measurements on a full scale Savonius wind

turbine were carried-out in the Environmental Wind Tunnel (EWT)
of the Polytechnic University of Marche, (Fig. 1). The EWT test
chamber consists of three main sections: the first is used for
aerodynamic tests requiring a uniform velocity distribution and
a low turbulence level; the second is used to test reciprocal inter-
ference effects between slender bodies; the latter is used to test Fig. 1. The Environmental Wind Tunnel of the “Polytechnic University of Marche”.
 V. D’Alessandro et al. / Energy 35 (2010) 3349e3363 3351

Fig. 4. Savonius rotor analyzed in this work.

allow to measure the Savonius mean angular velocity by means of

an incremental encoder and the model torque with a load cell;
a ventilated disk brake is first used to slow down the rotor, and then
to keep its angular velocity to a constant mean value. The model
was tested in a range of incoming wind velocities ranging between
Fig. 2. Schematic representation of the Environmental Wind Tunnel of the “Università 6 and 12 m/s with a step of a 1 m/s. The measured quantities were
Politecnica delle Marche”. combined in order to obtain the performance parameters
commonly used in the Aerodynamics of Wind Turbines [1]: i.e. the
torque coefficient (2) and the power coefficient (3). Both were
than 2.5% and a turbulence intensity level less than 0.3% on an area evaluated as function of the dimensionless parameter Tip-Speed
larger than 90% of the test cross section. The wind tunnel is also Ratio (TSR) (4)
equipped with a compact heat exchanger able to control temper-
ature variations inside a range of one Celsius degree. The Savonius M
Cm ¼ (2)
rotor was made in (1:1) scale and placed in the first section of the 1 2
ru A$R
test chamber (Fig. 3); the model overall dimensions were 1 m 2 N
height and 0.4 m diameter, so the occupied frontal area was 0.4 m2
P
(Fig. 3). A cross section of the rotor placed inside the wind tunnel is CP ¼ (3)
reported in Fig. 4. The test section transversal area is 3.16 m2 and
1 3
ru A
the solid blockage factor defined by (1): 2 N
uR
A l ¼ (4)
b ¼ 0:25$ occupied (1) uN
Afree
where A ¼ 2$R$L is the frontal rotor area, R is the rotor radius and L
has a value of 3.2% ensuring the possibility to use experimental data is the rotor height. The equations (2), (3) and (4) are also simply
with a simple blockage factor correction. The experimental facilities linked by the equation (5).

Fig. 3. The Savonius rotor model located in the EWT and the measurement system.
3352 V. D’Alessandro et al. / Energy 35 (2010) 3349e3363

CP
Cm ¼ (5)
l
The reported experimental results refer to a Savonius rotor
having no twist (0 ) and having end plates. Twist angle is defined as
the characteristic angle of the helix. The end plates are devices apt
to limit the three-dimensional flow behaviour around the model
tips. This allows the comparison of the experimental results to the
numerical computational data from a 2D domain. The good
agreement between numerical and experimental data reported in
the next sections, confirms as negligible the effects of the residual
3D vortices due to the presence of the end plates in the experi-
mental model.
In Figs. 5 and 6 are represented the power curve and the torque
curve in dimensionless form; data are collected for different
incoming wind velocities; the blue arrow show the direction of
increasing angular velocities. The trend exhibited by the curves is in
a good agreement with previous authors data [2e4]. The Cp versus
TSR curve was interpolated by a cubic polynomial curve obtaining an Fig. 6. The Savonius rotor torque curve.
R2 value greater than 0.96. The Cm versus TSR curve (Fig. 6) shows
instead a linear behaviour down to TSR values of 0.6 and a curve
bending for lower TSR values corresponding, for the experimental K represents a rotating frame around a fixed axis normal to the
conditions of fixed wind velocities, to low rotational speed veloci- represented plane and passing for the point ðxG ; yG Þ:
ties. The dimensionless performance curves reported in Figs. 5 and 6 Inside the K domain an unstructured triangular mesh having
show the absence of a meaningful dependence on the Reynolds a curvilinear orthogonal refinement near the walls (Fig. 8) was
number (6) referred to the investigated incident wind velocities. adopted, in {UF\K} (the white area inside the rectangle) both
curvilinear and orthogonal structured elements were used.
uN $2R
Re ¼ (6)
n
4. Flow field mathematical model
3. Computational model description
The flow field around a Savonius wind rotor at the Reynolds
The Savonius rotor analyzed in this work is a split-type as clearly number tested (ranging from 221,000 to 294,000) exhibits
shown in Fig. 4. The computational 2D domain U (Fig. 7), delimited a turbulent behaviour. In selecting the numerical approach to the
by the boundary G, is occupied by the fluid in a portion named UF simulation of a turbulent flow the main features of turbulence must
and by the solid in a portion named US such that: be considered: highly unsteadiness, three-dimensionality, presence
of different size eddies and coherent structures. Moreover
UF WUS ¼ U (7) a turbulent flow also exhibits random fluctuations of the fluid-
dynamic variables on a broad range of scales (in space and time).
and
This feature makes the Direct Numerical Simulation (DNS) unre-
UF XUS ¼ B (8) alizable for flows of engineering interest. In fact a DNS is based on
the numerical solution of NaviereStokes equations by a computa-
In UF a sub-domain K is defined by the following expression tional grid fine enough to allow for all the significant structures of
n o the turbulence (in space and in time) to be captured. This implies
K ¼ ðx; yÞ˛UF : ðx  xG Þ2 þðy  yG Þ2  R2mg (9) using a grid size determined by the turbulent smallest scale h that
can be estimated using the Kolmogorov theory [11]. A DNS
approach also requires a computational domain dimension as large
as the largest turbulent eddy.
For this reason in a DNS the number of grid points in each
direction must be at least L/h (where L is computational domain
size), but this ratio is proportional to Re3/4 [11].
In the three-dimensional computational domain the total
number of grid points should therefore scaled as Re9/4. The time-

Fig. 5. The Savonius rotor power curve. Fig. 7. Schematic representation of the Computational Domain.
 V. D’Alessandro et al. / Energy 35 (2010) 3349e3363 3353

Fig. 10. Flow chart of fluid-rigid body coupling algorithm.

RANS equations for an incompressible flow are here reported

(11), (12).

V$u ¼ 0 (11)

vt u þ ðu$VÞu ¼ Vðp=rÞ þ nV2 u þ V$ R (12)

Fig. 8. Mesh adopted near the rotor blades. ¼

The system (11), (12) must be closed introducing a suitable

expression for the Reynolds stress tensor R ¼ u0 5u0 by means of
¼
step size, on the other hand, must be at least T/th (where T is the a turbulence model.
temporal scale of the largest turbulent scales and th is the Kolmo- A widely used approach uses the “extra-quantities”: turbulent
gorov time-scale). Kolmogorov theory shows that the ratio T/th is kinetic energy (TKE) k, its dissipation rate 3 and the Boussinesq
proportional to Re1/2 [11] hence the total operations required to approximation (13).
perform a DNS is proportional to Re11/4 which becomes easily   2
prohibitive due to the limitations on computer speed and memory R ¼ nT VuDVuT  k I (13)
¼ 3 ¼
for industrial interest problems.
RANS approach has been developed on the idea of decomposing with k ¼ ð1=2Þu0i u0i :
the velocity components, the pressure and the density (if the flow is The unknown variables k and 3 are obtained solving the two
compressible) into two parts: a time-averaged part and a fluctu- transport equations (14) and (15).
ating part (10).
vt k þ ðu$VÞk ¼ Pk  3 þ V$½ðn þ nT ÞVk (14)
qffiffiffiffiffiffiffiffiffiffiffiffiffi
f ¼ f þ f0 (10)
where Pk ¼ nT S2 and S ¼ 2Sij Sij with Sij ¼ ð1=2Þðvj ui þ vi uj Þ:
RANS equations have been obtained introducing (10) in the   
c31 Pk  c32 3 nT
NaviereStokes equations and assuming that the operator used for vt 3 þ ðu$VÞk ¼ þ V$ nþ V3 (15)
the averaging satisfies the following properties: linearity, deriva- T s3
tives and average commutation, double average (the averaging with T time-scale
have not effect on the averaged variables) [11]. Using this approach
"" rffiffiffi# #
the CPU e time can be heavily reduced and numerical simulations k k 1 ak
of flows with engineering interest can be performed. A RANS T ¼ min max ; 6 ; pffiffiffi pffiffiffiffiffi (16)
3 3 cm 3 S y2
approach implies a reduction of the number details in the
computation of the flow field details as the turbulent kinetic energy
spectra are fully modelled.

Fig. 9. Domain boundaries. Fig. 11. Aerodynamic torque treatment in the RungeeKutta sub-steps.
3354 V. D’Alessandro et al. / Energy 35 (2010) 3349e3363

Fig. 12. Tip-speed ratio versus time (Experimental value: l ¼ 0.735).

The model used in this work also includes a transport equation cm ¼ 0:22 c31 ¼ 1:4 c32 ¼ 1:9
for y2 (Reynolds stress normal to the wall) and for f (elliptic relax- cf 1 ¼ 1:4 cf 2 ¼ 0:3 s3 ¼ 1:3
ation factor) representing non-local effects [13]. The turbulent CL ¼ 0:23 Ch ¼ 0:23 a ¼ 0:6
viscosity nT is defined as:
The y2  f model was used for the CFD computations of the flow
field around a Savonius wind rotor because it was designed in order
vT ¼ cm y2 T (17)
to represent the tendency of the wall to suppress transport in
A transport equations was also solved for y2 and f. normal direction without requiring a full second moment closure
hence limiting the CPU e time [13]. Moreover this model showed
y2 h i good performance in the prediction of the flow fields with strong
vt y2 þ ðu$VÞy2 ¼ kf  3 þ V$ ðn þ nT ÞVy2 (18) separations [14] that affects Savonius rotors too.
k
In order to take in account the rotor rotation in the computa-
 2=3  y2 =k Pk tional domain the fluid flow equations can be solved following two
f  L 2 V2 f ¼ cf1  1 þ cf2 (19) different approaches: locating the rotor in a non-inertial frame or in
T k
an inertial one.
L is the length scale In the first case the solution obtained is strictly connected with
" 3
" 3# 3
# the angle of attack of the rotor (angle between the cord of the
1 k2 k2 v4 rotor and the wind direction). Moreover if a constant angular
L ¼ CL max min ; pffiffiffi pffiffiffiffiffi ; ch 1 (20)
3 cm 3 S y2 34 velocity is imposed the Coriolis force effect on the turbulence
scalar quantities is completely avoided. Equation (10), in tensorial
The following values for the calibration constants included in notation, for a two-dimensional flow field in a non-inertial frame
the turbulence model [13] were used is expressed by (21)

Fig. 13. Plot 3d of Moment coefficient, tip-speed ratio and time (Experimental value: l ¼ 0.735).
 V. D’Alessandro et al. / Energy 35 (2010) 3349e3363 3355

Fig. 14. Irregularity degree as function of the mean tip-speed ratio.

vui v  1 vP v  ’ ’
þ ui uj ¼  þ n V2 u i  u u  23i3k uuk (21) Closure (SMC) can solve this problem producing an increasing in
vt vxj r vxi vxj i j CPU e time respect every EVM.
Locating the rotor in a inertial reference frame with constant
in which P is the pressure term modified in order to include the
angular velocity a not high-fidelity physical modelling is performed
centrifugal potential as in (22)
because the Savonius during its rotation exhibits a strong variation
1 in angular velocity; for this reason in this work an approach based
P ¼ p  ru2 r 2 (22)
2 on an inertial frame with variable rotor angular velocity was used.
It’s well know from turbulent flows theory [11,12] that multi- In particular a SMM approach was adopted in order to take into
plying (21) by u’i and averaging, a transport equation for TKE can be account the rotation of the rigid body around a fixed axis. The
obtained. Starting from (21) an equation for TKE in non-inertial computational grid is decomposed into two sub-grids, where the
frames can be derived; in this case it’s easy to note as the term inner K domain is able to rigidly rotate with respect to the outer
related to the Coriolis force in (21) disappeared (23). one. The rotational sliding grid set-up, used in this work is dis-
played in Fig. 7. The commonly used Fluid-Dynamics conservation
23i3k u’i uuk ¼ 23i3k u’i uuk ¼ 0 (23) laws (i.e. momentum, mass and turbulence quantities) must be
suitably modified in order to take into account the grid motion. At
hence Coriolis force effect is completely neglected in Eddy- this purpose a mesh motion flux term, related to mesh motion
Viscosity Models (EVM) using the transport equation for TKE. This velocity ug, was introduced. Hence the transport equation for
is directly related to the scalar nature of the TKE. Second Moment a generic fluid flow variable f becomes (24) [9]:

Fig. 15. Cm  q trend polar plot (the wind direction is parallel to q ¼ 90 ).
3356 V. D’Alessandro et al. / Energy 35 (2010) 3349e3363
Z I I

vt ðrfÞdU þ r u  ug f$ndG  ðGVfÞ$ndG ¼ 0 (24)

UF v UF v UF

the main problem due to the relative motion between the mesh
elements of the K and {UF\K} domains is related to the fluxes at the
boundaries. The adopted solution procedure provides an appropriate
choice of time-step size (25) obtained by fixing the number of nodes
that slide in each time-step and knowing the distance between two
consecutive nodes on the grids interface Ds, the angular velocity at the
previous time-step solved u(n  1) and the radius of the rotating frame
R. In this way there are not overlapped cells at the sliding boundary
and a conformal grid is assured in every time-step solved.
Ds$N
Dt ðnÞ ¼ (25)
uðn1Þ $Rmg
RANS equations closed with y2  f model must be, obviously,
completed with a set of boundary conditions. The imposed
boundary conditions are:

 a fixed inflow velocity at Gin (26), (see Fig. 9);

uðxÞ ¼ uin cx˛Gin (26)

 a zero normal pressure gradient at Gin;

 no-slip at the rotor blades (27); Fig. 17. Torque release comparison between the approach developed in this work and
the approach with a constant tip-speed ratio e Polar frame (Experimental value:
l ¼ 0.735).
uðxÞ ¼ uek ^ðx  xG Þ cx˛vUs (27)

 Gw is treated as wall in order to reproduce the test chamber
of the Environmental Wind Tunnel Laboratory of the Poly-
technic University of Marche and to create a virtual wind
tunnel.
 an outflow physical boundary condition was used at Gout [10].
 the turbulent intensity and the hydraulic diameter were fixed
on Gin (the same values used in the experimental tests were
reproduced).
 At Gin, Gw and on the rotor blades a zero normal gradient
condition for the turbulent quantities was fixed
 For the fluid domain the following initial condition was set:
uðx; t ¼ 0Þ ¼ 0 cx˛UF yK
5. Solid body mathematical model
The fluid-structure coupling algorithm presented in this paper
provides, for each time-step, the solution of the fluid-dynamic field
(by means of the finite volume code FLUENT) and then evaluates
the angular velocity of the rotating system. These analyses were
performed by a numerical algorithm written in MatLab language. A
flow chart of the algorithm is shown in Fig. 10.
The fluid flow induces the motion of the solid body by means of
aerodynamic interactions. The structural deformations were
neglected and only the rigid body kinematics was considered;
hence the motion of the solid is evaluated integrating the second
(28) cardinal equation of dynamics for 1-DOF (Degree of Freedom)
rotating system (29)

Fig. 16. Torque release comparison between the approach developed in this work and the approach with a constant tip-speed ratio e Cartesian Frame (Experimental value:
l ¼ 0.735).
 V. D’Alessandro et al. / Energy 35 (2010) 3349e3363 3357

Fig. 18. Mean tip-speed ratio versus rotation number (Experimental value: l ¼ 0.735).

X
I u_ ¼ ext
MG;k (29) In this work a linear function of the angular velocity (posing
k H ¼ 1 in (32)) was used as the resistant torque in order to reproduce
the experimental conditions previously described.
which is referred to the fixed rotation axis of the system. Torques
As a consequence the equation (29) become:
acting on the rigid body are the aerodynamic moment Ma and
resistant torque Mr. The first is expressed per unit of rotor length by I u_ ¼ LjMa j  jMr j (33)
(30), in which n is the outward normal unit vector while the second
one can be expressed in a general way by means of a polynomial where L is the rotor height. The initial condition adopted in solving
function of the angular velocity with coefficients ak (32). (33) is given by:
I uð0Þ ¼ u0 (34)
Ma ¼ ðx  xG Þ^½ T ðu; pÞ  ndG (30)
¼
u0 was assumed to be 16 rad/s in order to achieve a suitable time-
v US
step size in the first time-step solved for flow governing equations
with numerical solution. The rotor moment of inertia resulted from
  a CAD evaluation of the model geometry.
T ðu; pÞ ¼ m Vu þ VuT  p I (31)
¼ ¼
6. Numerical methods
!
X
H
Mr ¼ br
ak uk M (32) RANS equations coupled with the turbulence model, were
k¼0 solved by means of a second order Finite Volume Method (more

Fig. 19. Mean power Coefficient versus rotation number (Experimental value: l ¼ 0.735).
3358 V. D’Alessandro et al. / Energy 35 (2010) 3349e3363

Table 1 after solution initialization the RungeeKutta method was

Comparison between numerical and experimental results. employed, hence
Numerical data Experimental data D[%]
X
s
l 0.513 l 0.53 4.04
uðnþ1Þ ¼ uðnÞ þ Dt bi Ki (36)
Cp 0.200 Cp 0.20 0.5
i¼1
l 0.617 l 0.64 3.66
Cp 0.221 Cp 0.22 0.48 with
l 0.735 l 0.76 3.27 1
Cp 0.240 Cp 0.23 2.5 K ¼ ðLjM’a j  jMr jÞ (37)
I
l 0.847 l 0.90 5.77
Cp 0.234 Cp 0.24 0.67 and
l 1.009 l 1.02 1.09 0 1
Cp 0.237 Cp 0.24 2.95 X
s
Ki ¼ f @t ðnÞ þ ci Dt; uðnÞ þ Dt aij Kj A (38)
l 1.090 l 1.10 0.92 j¼1
Cp 0.216 Cp 0.22 3.7
the coefficients {aij}, {ci} and {bi} completely characterize the
generic RungeeKutta method and are collected in the well known
details can be found in [16]). The discretization of the convective Butcher matrix (39).
terms was obtained by means of MUSCL [10] approach. All diffusive
terms were discretized with a central difference scheme and non-
orthogonal correction was treated using a Green-Gauss cell based
approach [10]. Time-integration was performed with a second
order accurate implicit scheme in order to cancel numerical diffu-
sion due to temporal discretization [16]. Pressureevelocity
coupling was solved using a SIMPLEC [17] approach. The fluid-
structure coupling algorithm, presented in this paper, implies that used in the form given by (40)
the aerodynamic torque is known only from previous time-step.
This requires an explicit scheme for the time-integration of the
second cardinal equation of dynamics with a time-step size fixed
for mesh conformity reasons, see (25). An a priori stability analysis
could not be performed hence strong stable time-integration
method was needed. For this reason a four stage RungeeKutta
time-integration method was used for the second cardinal equation
of dynamics. The fourth order accuracy guaranteed by the Run- The condition (41) ensures the scheme consistency [15] and
geeKutta method is a surplus, because of the time-integration being the RungeeKutta method a one step scheme then the
scheme used in fluid-dynamic code is second order accurate; consistency implies stability and hence convergence [15].
anyway this does not heavily increase the computational load. A
two point and a tree point explicit AdamseBashfort scheme were
X
s
bi ¼ 1 (41)
tested too. i¼1
In the first time-steps an Euler method [15] was used, obtaining
from (33) u(n þ 1). The aerodynamic moment is known only in the time-step (n)
and it is unknown in the sub-steps between (n) and (n þ 1): so, in
Dt   order to limit the computational load, these values were calculated
uðnþ1Þ ¼ uðnÞ þ LjM’a jðnÞ  jMr jðnÞ (35)
I by means of an interpolating fourth order polynomial function

Fig. 20. Comparison between numerical and experimental power curve. Fig. 21. Comparison between numerical and experimental torque curve.
 V. D’Alessandro et al. / Energy 35 (2010) 3349e3363 3359

Fig. 22. Comparison between results obtained with time-integration schemes (Experimental value: l ¼ 0.735).

using previous time-steps data and the coefficients reported in Dt ðnÞ  

uðnþ1Þ ¼ uðnÞ þ 23K ðnÞ  16K ðn1Þ þ 5K ðn2Þ (45)
Butcher matrix (40) (see Fig. 11). 12
The fourth order polynomial coefficients were calculated on As shown in the following paragraphs, results are quite inde-
data reported in (42). pendent from the time marching scheme. A comparison between
n      the approach just described, based on the fluid-rigid body model-
ði4Þ ði3Þ ði3Þ ði2Þ ði2Þ
X ¼ Ma ; t ði4Þ ; Ma ;t ; Ma ;t ; ling, and the widely used approach in the aerodynamics of wind
   o turbines based on a constant tip-speed ratio is also presented in the
ði1Þ ði1Þ ðiÞ
Ma ;t ; Ma ; t ðiÞ ð42Þ
following.
Hence they were used for calculating the requested aero-
dynamics torque at the time t ðiþ1=2Þ (43) (according to the Butcher
matrix). 7. Numerical results

 
ðiþ1=2Þ Savonius wind rotor performance, obtained from numerical
Ma ; t ðiþ1=2Þ (43)
computations, was evaluated for every angular position occupied
The second cardinal equation of dynamics was also solved using by the rotor using the parameters (2), (3) as function of (4).
both explicit 2-points and 3-points AdamseBashforth schemes [15] Performance parameters were also averaged on the single rotor
expressed in (44) and (45) respectively. rotation obtaining (46) and (47)

Z2p
Dt ðnÞ   1
uðnþ1Þ ¼ uðnÞ þ 3K ðnÞ  K ðn1Þ (44) hli ¼ lðqÞdq (46)
2 2p
0

Fig. 23. Comparison between results obtained with time-integration schemes (Experimental value: l ¼ 0.735).
3360 V. D’Alessandro et al. / Energy 35 (2010) 3349e3363

Fig. 24. Comparison between tip-speed ratio results obtained with two different grid refinements (Experimental value: l ¼ 0.833).

Z2p due to a variation in N caused by the necessity of diminishing time-

1
hCP i ¼ CP ðqÞdq (47) step size values at TSRs greater or equal than one.
2p A stable periodic operating condition was reached by the
0
Savonius rotor model as clearly visible in Fig. 12, where the tran-
the integral averages were numerically calculated by means of sient behaviour ended in about 3 s. The periodic value of the TSR is
a second order accurate method (48) and (49). directly related to the variable torque that a drag based wind
( ) turbine shows in relation with the angular rotor position.
R 1 X
Z1
1 The unstable aerodynamic torque experienced by the rotor was
hli ¼ ½uðqk Þ þ uðqkþ1 Þ (48)
uN ðZ  1Þ 2 calculated and reported in Fig. 13. It is evident that there is a strong
k¼1
variability during rotor operation that relates to time and TSR. This
( is in agreement with the strong irregularity degree of this kind of
1 1 X
Z1
1 wind turbine that was also calculated.
hCP i ¼ ½Mðqk Þuðqk Þ
ð1=2ÞruN A ðZ  1Þ k ¼ 1 2
3 It is worthy introducing EðkÞ as the vector that contains the
) angular velocities for the generic (k)-th rotation.
þ Mðqkþ1 Þuðqkþ1 Þ ð49Þ n o
EðkÞ ¼ uðkÞ uðkÞ uðkÞ
1 ; 2 ; .; Z (50)
In these equations qk is the angular position of the rotor at
a generic time-step (a zero value corresponds to the rotor aligned k ¼ 1; 2; .; Nsp with Nsp defined as the number of the rotor rota-
with flow direction as showed in Fig. 15) and Z is the number of tions in the periodic steady state.
time-steps in each rotor rotation. Z is fixed at 68 for mean TSRs Maximum and minimum angular velocity for the (k)-th round
ðkÞ ðkÞ
lower than one and at 157 for TSRs greater or equal to one. This is are respectively umax and umin :

Fig. 25. Comparison between torque coefficient results obtained with two different grid refinements (Experimental value: l ¼ 0.833).
 V. D’Alessandro et al. / Energy 35 (2010) 3349e3363 3361

Fig. 26. Contour of turbulence intensity (Experimental value: l ¼ 0.735).



uðkÞ ðkÞ
max ˛E cu ˛E; u
ðkÞ ðkÞ
 umax (51) (see Fig. 9). The integral (56) was calculated with the same
approach adopted for (46) and (47).


uðkÞ ðkÞ
˛E cuðkÞ ˛E; uðkÞ  umin (52) Zt2
min 1
Cm ðqÞ ¼ Cm ðq; tÞdt (56)
The mean angular velocity for the (k)-th round is defined as: t2  t1
t1

1 ðkÞ 
uðkÞ
m ¼ umax þ uðkÞ
min
(53) The polar plot reported in Fig. 15 shows a clockwise rotation
2
of the Cm  q trend by increasing the mean tip-speed ratio. This
hence the irregularity degree for the (k)-th rotation is defined as: effect is evident for tip-speed ratio values less than one and
seems to disappear, or even revert, at tip-speed ratios greater
uðkÞ ðkÞ
max  umin than one. This also allows to compare the modelling approach
iðkÞ ¼ (54) used in this work with the more classical approach based on
uðkÞ
m
a constant tip-speed ratio during all the numerical simulation.
The irregularity degree for a given functioning point is Figs. 16 and 17 clearly show a strong difference in the torque
defined as prediction during the rotor rotation and moreover the overlap jet
effect is less important in the approach based on a constant tip-
1 XNs
speed ratio.
I ¼ iðkÞ (55) The Savonius rotor mean performance, at the operating point,
Ns
k¼1
was evaluated averaging the parameters reported in Figs. 18 and 19
where Ns is the number of complete rotor rotations in which the once the asymptotic behaviour was reached. The average was
mean performance parameters, previously, defined, exhibit performed on the samples number corresponding to the steady
a steady behaviour. condition. In Fig. 18 the effect of the average on the instantaneous
The parameter reported in (55) was calculated for all mean tip- tip-speed ratio is also shown. These data were used for numerical/
speed ratios computed in this work. The trend is reported in Fig. 14. experimental data comparison.
The irregularity degree (55) shows a decrease with the increasing of Several operating points were simulated in order to obtain
the mean tip-speed ratio. complete numerical performance curves. Numerical and exper-
Aerodynamic torque coefficient data averaged in time (56) imental data are reported in Table 1 and graphed in Figs. 20 and
(averaging procedure was performed only in the periodic steady 21. In all the simulations reported in this paper the inflow wind
state) makes it possible to evaluate Savonius rotor torque release velocity has been fixed to 9 m/s and the turbulent intensity

Fig. 27. Contour of velocity magnitude (Experimental value: l ¼ 0.735).

3362 V. D’Alessandro et al. / Energy 35 (2010) 3349e3363
Fig. 28. Flow pattern near rotor blades (Experimental value: l ¼ 0.735).
at the inflow section was fixed to 0.2% as in the experimental
set-up.
Aerodynamic performance, obtained by the numerical simula-
tions, shows a very good agreement with experimental data. This
allows to use the computational model for a fluid-dynamic analysis
of a Savonius wind rotor.
Time marching of the second cardinal equation was performed
using an Euler method in the initial step and a four stage Run-
geeKutta or the AdamseBashfort schemes (two points and three
points) in the following steps. Figs. 22 and 23 show small differ-
ences in rotor performance calculation using these three different
time-integration schemes; the main differences are evident in the
run up phase. RungeeKutta scheme offers a better estimation of the
power coefficient of the rotor while a not meaningful difference can
be noted in the tip-speed ratio calculations between the schemes
adopted.
A sensitivity analysis on the results was performed about the
influence of the yþ (57) value.
Fig. 29. Velocity magnitude profiles in the wake region e rotor aligned with the wind direction.
us y
yþ ¼ (57)
n
pffiffiffiffiffiffiffiffiffiffiffi
where us is the friction velocity defined as us ¼ sw =r and y is the
distance of fist computational cell from the closest wall.
A coarser grid near the rotor blades was built in order to
guarantee yþ values ranging from 0.1 to 90 on the blade suction
side and from 0.1 to 40 on the pressure side. The results obtained
on the coarse grid configuration was compared with experi-
mental results and with computational results calculated on
a finer grid with low yþ values (ranging from 0.1 to 4 on the
suction side and from 0.1 to 2 on the blade pressure side).
Comparisons show a little influence of the grid refinement on the
prediction of the Savonius rotor performance as enlightened in
the Figs. 24 and 25.
Moreover the computational model developed in this work
allow to highlight the following flow features:
1. The wake induced by the Savonius rotor is showed in Fig. 26
where a turbulent intensity map is reported. The flow pattern
enlightens a cyclic behaviour. A velocity magnitude map of the
wake is presented in Fig. 27.
2. A quantitatively description of the wake is presented in Figs. 29
and 30 where the velocity magnitude profiles in the wake
region are reported. It’s quite easy to understand as the recir-
culation zones produce a velocity decrease.
3. An analysis of sequential frames allows to distinguish that the
central vortex is continuously supplied by the vortices devel-
oping on the advancing blade tip and behind the returning
blade.
Fig. 28, in which the maximum value of the velocity magnitude
field was fixed at 12 m/s, shows a zoom of the flow pattern previ-
ously highlighted; it shows the presence of some recirculation
zones in the suction side of the advancing blade and of the overlap
jet. This jet starts from the concave side of the advancing blade and
develops toward the returning blade. The increment of the pressure
on the impinging side reduces the negative contribution of the
 V. D’Alessandro et al. / Energy 35 (2010) 3349e3363
Fig. 30. Velocity magnitude profiles in the wake region e rotor orthogonal to the wind direction.
returning blade to the overall torque. This is confirmed by the
experimental tests. Moreover counter-rotating vortices are present
behind the returning blade. Both these structures evolve in the
bigger vortex located downstream.
8. Conclusions
An extended wind tunnel testing program was conducted in the
Environmental Wind Tunnel of the “Polytechnic University of
Marche” on a split-type Savonius wind rotor. The experimental
facilities allowed to evaluate rotor performance expressed by
means of the dimensionless parameters usually used in the aero-
dynamics of wind turbines. The use of dimensionless quantities
allows to verify the Reynolds number effect on the Savonius
performance. Graph reported in Figs. 5 and 6 underlines the
absence of a meaningful Reynolds number dependence on the
investigated wind velocities. A computational methodology able to
calculate the flow field around the rotor was also developed. RANS
equations (closed using y2  f turbulence model) were solved in
order to obtain accurate information about the flow field. The rotor
blades were treated as rigid bodies and their behaviour were
modelled by means of the Second Cardinal Equation of Dynamic.
The experimental data were used to validate the developed
computational methodology. The comparison of performance data
obtained by numerical simulations and experimental measure-
ments shows very good agreement. This suggests the use of this
numerical method for studying new blades shapes in order to
produce better rotor performance. The results exhibited by
numerical simulations allowed to gain an insight into the flow field
mean features.
3363
References
[1] Hansen MOL. Aerodynamics of wind turbines. 2nd ed. London, Sterling VA:
Earthscan; 2008.
[2] Fujisawa N. Velocity measurements and numerical calculations of flow fields
in and around Savonius rotors. Journal of Wind Engineering and Industrial
Aerodynamics 1996;59:39e50.
[3] Fernando MSUK, Modi VJ. A numerical analysis of the unsteady flow past
a Savonius wind turbine. Journal of Wind Engineering and Industrial Aero-
dynamics, 32, 3, p. 303e27.
[4] Shinohara T, Ishimatsu K. Simulation of flow around rotating Savonius rotors.
In: Proceedings of the 6th National Symposium on Computational Fluid
Dynamics, p. 691e694, (SEE N94-34731 10-34).
[5] Redchyts D, Prykhodko O. Numerical modelling of dynamics and aero-
dynamics processes of Darrieus and Savonius rotors. Geophysical Research
Abstract 2006;8(03083).
[6] Spalart PR, Allmaras SR. A one-equation turbulence model for aerodynamic
flows. AIAA-Paper 92-0439, 1, 5e21; 1992.
[7] Altan BD, Atilgan M. An experimental and numerical study on the improve-
ment of the performance of Savonius wind rotor. Energy Conversion and
Management 2008;49:3425e32.
[8] O’Brien JF, Zordan V, Hodings JK. Combining active and passive simulations for
second motion. IECC Computer Graphics and Applications 2000;20(4):86e96.
[9] Hirsch C. Numerical computation of internal and external flows. 2nd ed.
Elsevier; 2007.
[10] FLUENT users’ guide release 6.3.26. USA: Ansys Inc.; 2006.
[11] Pope SB. Turbulent flows. UK: Cambrigde University Press; 2000. 0521598869l.
[12] Wilcox DC. Turbulence modelling for CFD. California, USA: DCW Industries,
ISBN 0963605100; 1994.
[13] Durbin P. Near-wall turbulence closure modelling without damping functions.
Theoretical and Computational Fluid Dynamics 1991;3:1e13.
[14] Durbin P. Separated flow computation with k-3-v2 model. AIAA Journal, 33, p.
659e64.
[15] Quarteroni A, Sacco R, Saleri F. Matematica numerica. 2nd ed. Springer; 2005.
[16] Jasak H. Error analysis and estimation for the finite volume method with
applications to fluid flows. PhD Thesis, Imperial College of Science, Tech-
nology and Medicine e London, UK; 1996.
[17] Versteeg HK, Malalasekera W. An introduction to computational fluid
dynamics, the finite volume method. Longman Scientific & Technical; 1995.