J. Wind Eng. Ind. Aerodyn.

135 (2014) 5769

Contents lists available at ScienceDirect

Journal of Wind Engineering

and Industrial Aerodynamics
journal homepage: www.elsevier.com/locate/jweia

A consistent vortex model for the aerodynamic analysis of

vertical axis wind turbines
L. Shi a, V.A. Riziotis b,n, S.G. Voutsinas b, J. Wang a
a
b

Huazhong University of Science and Technology, China-EU Institute for Clean and Renewable Energy, LuoYu Road, Hubei Province, Wuhan 1037, China
School of Mechanical Engineering, National Technical University of Athens, 9 Heroon Polytechneiou Street, GR15780 Athens, Greece

art ic l e i nf o

a b s t r a c t

Article history:
Received 5 March 2014
Received in revised form
5 August 2014
Accepted 6 October 2014
Available online 28 October 2014

A consistent two dimensional vortex type aerodynamic model for VAWTs is presented alongside with its
validation against measured data. The ow solver assumes incompressible and inviscid conditions. It
combines a source-vorticity panel formulation for the blades and a vortex blob representation of their
wakes. By construction the model accounts for the effects of curvature of the relative to the blade inow
while blade vortex interactions are modelled by locally correcting the position of the wake vortices when
they impinge on the blade. In order to get realistic loading estimations, lift and drag are corrected using a
modied version of the ONERA model in which only the contribution of the separated generalised
circulation is considered. Comparisons against wind tunnel tests on model rotors as well as full scale,
eld measurements on a 12 kW VAWT indicate that the model predicts well the aerodynamic loads on
the blades and the power output of the rotor.
& 2014 Elsevier Ltd. All rights reserved.

Keywords:
Vertical axis wind turbines
Rotor aerodynamic analysis
Free wake models
Vortex models

1. Introduction
In the early years of wind energy technology, the Vertical Axis
Wind Turbine (VAWT) concept underwent considerable engineering
development. Signicant research effort was directed to the development of new methods for the analysis and optimisation of their
aerodynamic performance. Initially, VAWTs were investigated in
parallel with Horizontal Axis Wind Turbines (HAWTs) and some
medium size turbines in the scale of 100600 kW were installed in
the 80s especially in the US and Canada. Typical examples of such
turbines are the 34 m diameter Sandia prototype DOE 625 kW
(Sutherland et al., 2012) and the commercial FloWind turbines
(Paraschivoiu, 2002). In the early 90s VAWTs were totally abandoned
in favour of HAWTs which ever since prevailed and nally dominated
the market of large scale energy production using wind energy until
today. Recently, the increasing interest in offshore wind farms
targeting to installations in deep water and the continuous research
on oating wind turbine concepts has placed VAWTs again into the
frame. Important drivers are the lower cost of VAWTs, their simple
design architecture and the light weight structure that renders them
ideal for offshore oating applications. Also, the fact that the drive
train system and the generator can be situated on the base of the
n

Corresponding author. Tel.: 30 210 7721101.

E-mail addresses: shilu8893@163.com (L. Shi),
vasilis@uid.mech.ntua.gr (V.A. Riziotis),
spyros@uid.mech.ntua.gr (S.G. Voutsinas), Wangjhust@163.com (J. Wang).
http://dx.doi.org/10.1016/j.jweia.2014.10.002
0167-6105/& 2014 Elsevier Ltd. All rights reserved.

turbine (or even underwater) could considerably reduce maintenance costs and improve the logistics in case of remote installations.
Despite their very simple design philosophy, there are several
challenges with respect to their aerodynamics. The main feature of
VAWTS is that the effective angle of attack seen by the blades
undergoes a very big variation which in moderate to low tip speed
conditions drives the blades into stall both in the negative and the
positive angles of attack regime. The variation of the angles of
attack within the post stall region gives rise to signicant ow
unsteadiness and dynamic stall phenomena. The unsteadiness of
the ow and therefore dynamic stall hysteresis effects strongly
depend on the reduced frequency k that scales the blade motion
which in turn depends on the solidity of the rotor.
Moreover, the blades of VAWTs are subjected to strong Blade
Vortex Interactions (BVIs). The retreating VAWT blade impinges on
its own wake but also on the wakes of preceding blades that are
formed in earlier times. The number over one revolution and the
strength of BVI encountered by each blade depends on the
operating conditions. At high tip speed ratios where wake convection velocity is low, the retreating blades experience multiple
wake crossings. Thereby, BVI gets stronger at high tip speed ratios
while it is considerably weakened at low tip speed ratios where
dynamic stall phenomena dominate.
Another important effect that must be taken into account concerns the curvature of the ow lines in the relative frame. As a result
of the circumferential motion of the blades, the streamlines of the
relative ow are highly curved and therefore the ow characteristics

58

L. Shi et al. / J. Wind Eng. Ind. Aerodyn. 135 (2014) 5769

Nomenclature

G
Green's function Gr 1=2 ln r
1L , 2L , 2D circulation components of the ONERA model

u
w
u
U1
Ub

C D0
C L

ow velocity in the xed coordinate system

ow velocity in the relative coordinate system
vortical part of velocity eld
free stream velocity
body velocity

scalar potential
k stream function
k vorticity eld
p
pressure eld
TE
trailing edge point

source distribution over the airfoil

bound vorticity distribution on the airfoil

circulation around a circuit C

w
vorticity distribution on the wake
xw
arbitrary point on the wake
xcp
collocation point on the airfoil
S
airfoil length
w
intensities of vortex particles on wakes
w
position vectors of vortex particles on wakes
f
cut off function

cut off length of vortex particles

w
surface vorticity of TE wake panel
Sw
lengths of TE and SP near wake panels
w
angles of TE and SP near wake panels
t
time step of numerical simulation

are signicantly altered as compared to those of the rectilinear ow. A

symmetric airfoil moving in a curved path will have the aerodynamic
characteristics of a cambered airfoil (non-zero, zero lift angle of
attack). The strength of the virtual cambering effect is related to the
curvature of the path undergone by the blade in relation to the blade
chord. This is directly linked to the parameter c=R.
Finally, the local Reynolds number will also vary signicantly over
the revolution. Maximum Reynolds number occurs at the beginning of
the advancing side where the relative inow velocity takes its
maximum value (sum of wind speed and rotation velocity) and
attains its minimum at the beginning of the retreating side where
the wind velocity is subtracted from the rotation velocity. Clearly the
variation is bigger at higher wind speeds and it is expected to
signicantly affect the aerodynamic loads, especially for smaller size
wind turbines that operate at low average Reynolds numbers, in the
order of 105 .
In the 80s and 90s the aerodynamic modelling of VAWTs was
primarily based on multiple stream tubes, actuator cylinder models coupled with the blade element method (Strickland, 1975;
Paraschivoiu, 1988) and vortex type free or prescribed wake models
(Strickland et al., 1979; Ponta and Jacovkis, 2001;Coton et al., 1994).
Recently, in response to the renewed interest on VAWT, new models
that belong to both categories have been developed with the aim to
provide answers to the deciencies of earlier methods. For example
Madsen et al. (2012) and Larsen and Madsen (2013) developed an
updated non-linear actuator cylinder model in which the assumption
that the ow pressure fully recovers at the centre of the rotor before
the ow crosses the downstream cylinder is no longer made. Also,
new 2D and 3D vortex type free wake models have been developed
with main focus on better representing wake dynamics and modelling
of dynamic stall (Zanon et al., 2013; Simo Ferreira, 2009; Scheurich et
al., 2011). Moreover, as a result of the continuous increase in computer
capabilities, a number of aerodynamic analyses of VAWT employing
CFD methods have been published (Hansen and Srensen, 2001;
Simo Ferreira et al., 2007).

friction drag (minimum of drag coefcient polar)

difference of the real viscous steady state lift coefcient from the corresponding inviscid
C D
difference of friction drag coefcient C D0 from the real
viscous steady state drag coefcient.

rotor rotational speed

c
local blade chord length
R
rotor radius
B
number of rotor blades
R=U 1 tip speed ratio (TSR)
W ef f
local effective ow velocity including wake induced
effects
w0
component of W ef f normal to the blade
chord
q
W
local geometric ow velocity W U 21 R2
ef f
local effective angle of attack of the airfoil section
2 U Wc ef f time constant of the ONERA model
a , r , E empirical parameters of the ONERA model
CL
lift coefcient
CD
drag coefcient
Fn
local blade normal force
Ft
blade tangential/driving force
local 
C n F n = 2W 2 c normal force coefcient
C t F t = 2W 2 c tangential force coefcient
c  c reduced frequency of the airfoil
k 2W
2R
ef f
Bc
s 2R
solidity of the rotor

In the context of wind turbine design verication, the aerodynamic

tool should be of low to moderate computational cost so as to perform
the long list of time domain aeroelastic simulations with turbulent
wind contained in the IEC standard. With regard to HAWTs, actuator
disk models combined with blade element (BEM models) have been
widely used over the years. Through the application of appropriate
engineering corrections that account for dynamic inow, unsteady
aerodynamics and yaw misalignment effects, BEM models constitute a
very powerful design tool of HAWTs (Hansen et al., 2006). The
equivalent in the VAWT case corresponds to the actuator cylinder
model which however cannot properly model phenomena like blade
vortex interaction or virtual/apparent cambering that play here a
decisive role. So the predictions of actuator type models for VAWTs are
compromised and at present no consistent engineering corrections
have been proposed or tuned in order to improve their prediction
capabilities. An efcient modelling alternative, also suitable for the
purposes of aeroelastic analysis, lies in vortex type models. They stand
in between actuator and CFD models. The main advantage of vortex
models is that wake dynamics and all the effects induced by the wake
are build in. Aiming at the lowest possible cost, vortex models are to
be formulated as inviscid ow models in which case viscous effects,
such as ow separation, are added as corrections either a posteriori
directly on the loads (Riziotis and Voutsinas, 1997) or by introducing
additional vorticity emission from the point of separation (Voutsinas
and Riziotis, 1996; Zanon et al., 2013) (usually referred to as the
double wake concept; see Vezza and McD Galbraith, 1985 and
Voutsinas and Riziotis, 1996). In order to become fully predictive, the
double wake model is strongly coupled with a viscous boundary layer
which provides the separation location (Riziotis and Voutsinas, 2008).
In the present paper, a 2D vortex type free wake aerodynamic
model with consistent corrections on loads is presented. A standard
panel method is used in the modelling of the ow around the blades
(Basu and Hancock, 1978) while the wake is represented by freely
moving vortex blobs. In the calculation of the unsteady pressure the
contribution of the wake vorticity is considered separately by solving a

L. Shi et al. / J. Wind Eng. Ind. Aerodyn. 135 (2014) 5769

Poisson equation for the wake induced pressure eld. In this way the
problem of dening the potential function for the wake vorticity eld
is overcome. To the authors knowledge it is the rst time that such a
method is applied in a free wake panel code.
The baseline model is essentially inviscid, therefore for the consistent prediction of VAWT power performance and load characteristics corrections on the potential lift and drag must be applied in
order to account for viscous drag and ow separation. In the present
work, a new correction method is proposed on the basis of the
ONERA model (Petot, 1989). In its extended form, the model splits
the total circulation around the airfoil into two components: the
attached (potential) component and a separated component. The
two circulation components are obtained through the solution of
differential equations, a rst order and a second order. The model
basis assumes rectilinear inow, so in order to be consistent with the
apparent curvature effects that are already contained in the vortex
model, the potential circulation component is taken directly from
there and so only the separated component of the ONERA model is
computed. A similar correction, again on the basis of the ONERA
model, is performed on drag. Calculation of the corrected loads
requires a steady state lift and drag polar and a consistent denition
of the local to the section effective angle of attack which is here
determined kinematically.
The proposed model is validated against measurements from
three tests. The rst one is a water tank test conducted at Sandia
Laboratory (Strickland et al., 1981). The second is a wind tunnel test
conducted at the low speed, low turbulence, wind tunnel of NTUA
(Morathakis, 1991), and the third is a eld test on a small wind
turbine with a rated power of 12 kW (Kjellin et al., 2011). Validation
is carried out in terms of power curve results and azimuth
variations of the aerodynamic loads.

2. Free wake model formulation

2.1. Flow representation boundary conditions numerical model
For the representation of the inviscid, incompressible ow eld
around the VAWT rotor a standard panel method is employed. The
velocity eld u U; t is represented by means of source i U; t and

59

vorticity i U; t distributions dened on the surfaces Si of the

i 1; B blade sections and vorticity distributions on their wakes
(Basu and Hancock, 1978; Riziotis and Voutsinas, 2008). In the
continuous formulation the wakes of the airfoils are introduced as
vortex sheets Siw U ; t originating from the trailing edges of the
airfoils, carrying surface vorticity iw U ; t (see Fig. 1).
In the numerical model, the vortex sheets Siw are formed by their
i
near parts, consisted of segments of lengths Sw
and carrying

i
constant surface vorticity w and their far parts consisted of a collm
ection of vortex blobs dened by their intensities w and their
m
positions Zw , (index m 1; M counts the number of vortex blobs that
have been released in the wakes in previous time steps) (see again
Fig. 1). Each blade section is divided into N rectilinear elements
(panels) Sie ; e 1; N; i 1; B. On the blade sections, piecewise constant sources ie and constant surface vorticity i are distributed.
Using Green's theorem the velocity eld is written as
"
Z
Z
B
N
N
r
rk
dsx i
dsx
ux0 ; t U1 t
ie
2
i 2 r 2
i
Se
e 1 Se 2 r
i1 e1
#

Swi

2 r

m
m
w x 0  Zw  k

 f
x0  Zm 2
m 1 2
w

rk
wi
dsxw
2
M

where U1 is the velocity at innity, r x0  x is the vector distance

between the point of evaluation x0 and the location x of the inducing
singularity (on the airfoil boundary or the near wake segment), k is
the unit vector normal to the ow plane, f an exponential cut-off
function introduced in order to avoid singularities of the velocity
induced by the vortex blobs of the wake and the core radius of the
vortex blobs (Cottet and Koumoutsakos, 2008).
Numerical solution is obtained through a time marching scheme in
which (in accordance with the formulation of Basu and Hancock
(1978)) at every time step the unknown singularities and the updated
wake geometries (near and far) are dened.
m
Since the intensities of the far wake vortex blobs w are known
from previous time steps, the unknowns of the problem are the
 sources ie , the bound vorticity distributions i and the
i . The total
vorticity distributions of the near wake segments w
2  unknowns are calculated by satisfying (see Riziotis and
Voutsinas (2008) for more details):

Fig. 1. Continuous and discrete representation of the ow around the rotor.

60

L. Shi et al. / J. Wind Eng. Ind. Aerodyn. 135 (2014) 5769

a) The non-entry boundary condition at the elements centres xcp

providing N  independent equations.
b) Kelvin's theorem around every airfoil section. Kelvin's theorem
demands that for any change of the circulation around any
i
i
airfoil section, a vorticity equal to 
w Sw should be
shed in the wake, providing thus independent conditions.
c) Emission condition (Kutta condition) which demands that the
i , at every
intensities of the newly shed vorticities from the TE w
time step of the numerical procedure are given by the local velocity
discontinuities (calculated at the control points adjacent to the TE).
This is in line with the denition of vorticity in the context of
potential ow theory and it is also equivalent to assuming zero
pressure jumps at the TE points (see Riziotis and Voutsinas, 2008
for details). Kutta condition also provides independent equations.
Special care is taken for vortex blobs approaching or crossing
solid boundaries, during wake convection or during the displacement of the blades to their new position, at the end of every time
step. Buffer zones are dened around the solid boundaries that
extend at a distance 0.010.02c (c chord length of airfoil section)
from the blade surface. In case a vortex blob enters the buffer zone
it is displaced normally to the section surface, at the edge of the
zone as shown in Fig. 2.
2.2. Potential loads calculation
The pressure distribution p over the airfoil sections is calculated through the application of Bernoulli's equations which for
unsteady ows takes the form:
u2 p
U2 p
ct 1 1
t
2
2

u u

Accordingly the pressure eld is split into two parts p p p

(potential and vortical) and it is obtained through momentum
equation written in the following form:


u2 p
u
p

 u 0

t
2

p can be specied so that the unsteady Bernoulli equation is

recovered,


u2 p

0
6
t
2

Then, by taking the divergence of (5) and making use of

Uu 0, it takes the form
2 p   u

which can be solved in closed form assuming zero normal

derivative of the pressure on the airfoil surface:
B

p x0 ; t

i1

"Z

Si

p x; t

G
x0  xdS
n

where denotes the stream function. The rest of the singularities

are associated to the potential and then according to Helmholtz's

Z
Siw

 ux; t U Gx0  xdS

8
G being the 2D Green function: Gr 1=2 ln r.
By approximating wake vorticity as a sum of the vortex
blobs, the surface integral over the wakes in (8) takes the
following discrete form:
B

i1

"Z

#
Siw

 ux; t U Gx0 xdS

m 1;M



m k  u m t; t U G x  m t

where p1 t is the static pressure at innity and is the total ow

potential dened by u .
The main difculty in the application of Eq. (2) is the calculation
of the total potential . This is because the potential functions for
the surface vorticity and the point vortices involve the discontinuous arctangent function. A useful corollary of Helmholtz's theorem
is that surface vorticity dened on a surface S is equivalent to a
dipole distribution : =s. If S is smooth and well dened it is
straightforward to substitute vorticity by an equivalent dipole
distribution in the calculation of . Although, this approach can
be followed for the bound vorticity it cannot be always applied in
the wakes. This is especially true in the case of VAWT congurations
where wake vortex sheets undergo a complicated evolution with
ruptures due to BVI and intense deformations. In this case it is
preferred to associate w to a vortical velocity eld u (and thereby
to wake vorticity k), which by denition satises the following conditions:


u  k ; Uu 0; 2 
3

Fig. 2. Treatment of vortex blobs approaching or crossing airfoil boundary.

decomposition theorem:

Integral equation (8) is satised at the elements centres xcp

assuming that p is piecewise constant over the panels and
combined with (9) provides p .
Furthermore, integration of (6) leads to the unsteady Bernoulli
equation for p . When partial derivative in time =t is referenced to the relative blade section frame equation (6) takes the
form (see Katz and Plotkin, 1991)


u2 p w2 U2b p
U2 p

 ct 1 1
u U Ub 

t relative
2
t 2 2
2

10
where Ub U ; t is the velocity of the moving airfoil and w U; t
u U; t  Ub U ; t the relative ow velocity.
To the authors knowledge it is the rst time that such a method
is applied in a free wake panel code.

3. Viscous corrections on loads

In the present model, an a posteriori correction is applied on lift
and drag on the basis of the engineering ONERA model (Petot, 1989).
In the extended ONERA model, the total circulation around the airfoil
is divided into two components 1L and 2L . The rst component
corresponds to the circulation of the purely inviscid ow, while the
second accounts for the effect of stall. The potential circulation
component 1L and thereby the potential part of the lift force on
the blade is taken directly from the vortex model. The 2L component
is obtained by solving a 2nd order differential equation with input the
difference C L C L lin  C Ls of the real viscous steady state lift
coefcient from the corresponding inviscid (C L lin 2 for a symmetric airfoil) at the effective angle of attack ef f (as illustrated in
Fig. 3). In the calculation of the drag, the ONERA model denes an
equivalent circulation component 2D that accounts for the increase
in drag associated with ow separation. The 2D circulation is also

L. Shi et al. / J. Wind Eng. Ind. Aerodyn. 135 (2014) 5769

61

Fig. 3. Denition of C L and C D parameters of the ONERA model.

Fig. 4. Virtual cambering effect on a rotating blade.

obtained by solving a 2nd order differential equation similar to the one

for 2L which takes as input the difference C D C Ds  C D0 of the
friction drag coefcient C D0 (minimum drag coefcient of the steady
polar) from the viscous steady state drag coefcient C Ds (see Fig. 3).
The dening equations for 2L and 2D are
"
#

 E
a
r
r
_  W C w_
L; D 11
2

ef f

ef f

where W ef f is the effective ow velocity, w0 is the component of W ef f

normal to the chord w0 W ef f sin ef f , is a time constant given by
c=2 UW ef f , where c is the airfoil chord, a , r , E with L ; D
are empirical parameters of the model. The range of variation of the
above empirical parameters is dened by Petot (1989). The exact
choice depends on the airfoil geometry and should be tuned separately. Nevertheless, sensitivity of the results on the exact value of the
empirical parameters is rather small and therefore the mean values of
the ranges dened by Petot are considered representative for any
airfoil shape. Calculation of the separated circulation components 2L
and 2D in (11) requires a consistent denition of the local to the
section apparent relative inow velocity W ef f and angle of attack ef f
and the steady state lift and drag polars of the airfoil section at the
corresponding Reynolds number.
The relative inow velocity vector Wef f with respect to the inertial
frame at the quarter chord location of the blade is calculated by
Wef f U1 Uw  Ub

12

where Uw is the wake induced velocity given by

B

Uw xc=4 ; t

i1

"Z

i
i w

Sw

#




M m
xc=4  x  k
xc=4  Zm
w k
w
U 
 f

2 dsxw
xc=4  Zm 2
m 1 2
2 xc=4  x
w

13
In the above expression the contribution of the bound vorticity
and that of the body sources should not be included.
For a consistent calculation of 2L the parameter C L must be
properly dened, taking into account the virtual cambering effect

Fig. 5. Potential and viscous CL loops of a rotating blade.

due to rotation. A symmetric airfoil moving in a curved path (as

illustrated in Fig. 4) will have the aerodynamic characteristics of a
cambered airfoil in a rectilinear ow. The airfoil will have a
positive (non-zero) zero lift angle of attack which implies that it
behaves as a cambered airfoil with a negative curvature on its
suction side (see Fig. 4). As illustrated in Fig. 5, the zero lift angle of
attack of the rotating blade, on the C L  plot is dened as the
angle at which the centre of the unsteady potential C L loop
intersects the abscissa. This is the point wherefrom the steady
curved
state, inviscid, C curved
L lin  polar is also passing (C L lin 2  0 ).
It is noted that, as a result of the wake induction, the unsteady
loop will have a higher than 2 slope.

62

L. Shi et al. / J. Wind Eng. Ind. Aerodyn. 135 (2014) 5769

Fig. 6. Application of the ONERA model in the case of a VAWT blade.

Two alternatives are proposed for the denition of C L . In the

rst (see Fig. 6a)
8
>
< C L1 C L lin  C Ls


C L C L1 1  =s C L2 =s
>
: C C
C
L2

L lin

Ls

ef f o max
max o ef f o max s
ef f 4 max s
14

where ef f  max , max the maximum C L angle of attack either

at positive or negative angles of attack and s is a small range (about
11) within which the blending of the two C L denitions is
performed.
In the second option (Fig. 6b)
curved
C L C curved
L lin  C Ls

15

is the C Ls shifted by the new 0 .

where C curved
Ls
In the rst option, it is implied that neither max nor the post
stall behaviour of the airfoil is affected by virtual cambering. In the
second option max is shifted but the post stall behaviour remains
unchanged. As will be shown later, the rst denition of C L
provides better agreement with measured data. In fact measured
data do not support an earlier stall as a result of the shifting of C Ls .
The nal corrected lift coefcient of the blade section has the
form
C L C L dyn

2L

16

W ef f

where C L dyn is the potential lift coefcient calculated by the purely

inviscid free wake model. It is noted that the circulation component
2L is normalised by the semi-chord. The second term in (16) accounts
for the decrease (in absolute value) in lift as a result of ow separation.
The drag coefcient of the airfoil is written as
C D C D0 C D dyn

2D
W ef f

17

The contribution of the 2D circulation component that accounts

for stall should be added to the friction drag C D0 and the potential drag
C D dyn predicted by the inviscid free wake model (potential wake
induced drag) in order to obtain consistent results for the total drag.
As will be shown in the code validation, consistent prediction of
the rotor power performance requires both corrections to be applied
on loads.

Table 1
Characteristics of the Sandia test turbine.
Rotational speed (rpm)
Blade tip speed (m/s)
Number of blades
Solidity

7.2
0.457
2
0.3

Diameter (m)
Blade length (m)
Chord length (m)
Blade airfoil
Re (average)

1.219
0.457
0.0914
NACA0012
40,000

4. Validation against measurements discussion

Steady polars are of particular signicance to the correction of the
loads. Measured data are the best choice. However, quite often
measured data, at the Re number considered, are not available and
therefore one should rely on predicted polars. This is the case of the
validation test cases considered next. So in every test the issue of
predicting polars is addressed. To this end, FOIL2W the viscous
inviscid interaction model presented in Riziotis and Voutsinas (2008)
is used.
The rst validation test was conducted in a towing tank and
concerned an H-type VAWT model at low Reynolds ow conditions (Strickland et al., 1981; Table 1). The turbine was placed
vertically with its upper tip piercing the water surface while its
lower tip had a clearance of 15 cm from the tank bottom. Although
the test was designed to represent the two dimensional ow
around the rotor still some three dimensional effects (three
dimension vortical structures) were identied by Scheurich et al.
(2011). The loads on the blades were measured using strain gauges
and they are provided in the form of dimensionless normal and
tangential to the chord of the blades force coefcients C n and C t
q

2
normalised with W U 21 R
i.e. the relative geometric
velocity (not taking into account wake induction).
Fig. 7 presents comparison of the predicted azimuthal variations of C n and C t against the measured ones for a test case with
5. At this the ow is expected to remain attached both over
the advancing and the retreating side. Simulations are performed
with an azimuthal resolution of 21 while the airfoil sections are
divided into 100 panels. The same numerical parameters have
been used in all simulations presented thereafter. In the plots
three sets of predictions are shown. The rst set is the potential
load, as provided by the vortex model before any correction is
applied, the second set is the ONERA corrected load (correction
based on Eq. (14)) and the third set corresponds to loads calculated
by using the steady state CL and CD data at the effective angles of
attack. The potential and the ONERA corrected (correction based
on FOIL2W polars) C n almost coincide (see Fig. 7a) which indicates

L. Shi et al. / J. Wind Eng. Ind. Aerodyn. 135 (2014) 5769

63

Fig. 7. Comparison of predicted and measured (a) normal and (b) tangential force coefcients for 5 Sandia test turbine potential corresponds to purely inviscid
predictions, steady polars FOIL2W corresponds to corrected results based on steady-state FOIL2W polars, ONERA corrected FOIL2W corresponds to ONERA corrected
results using FOIL2W polars.

that no separation takes place in the examined case. Given that the
maximum angle of attack experienced by the blades is about 81
(absolute value) it is deduced that the resulting corrected loads
will be independent of the polar used in correcting them. The fact
that the correction made on the potential loads is very slight can
be also seen in the CL hysteresis loop plot in Fig. 8 where a very
small deviation between the potential and the corrected loop is
seen at the negative angles of attack regime, during the upstroke
phase (increasing angles of attack). As illustrated in Figs. 5 and 8
anticlockwise CL loop are anticipated for the rotating airfoil. As
discussed in Section 3 the loop is shifted by about 2.51 towards
higher angles with respect to the steady state polar due to virtual
cambering of the airfoil. This explains why the C n variation
obtained by using steady state characteristics is considerably
shifted towards higher values with respect to potential and ONERA
corrected curves. The predictions of the proposed model agree
well with measurements over the biggest portion of the revolution. Some small deviations appear in the vicinity of 2701 azimuth.
Scheurich et al. (2011) attributed this effect to the three dimensionality of the ow and the corresponding cross ow vortical
structures that are being formed in the conditions of the experiment. It is noted that Scheurich et al. (2011) employed a 3D Vortex
Transport Model (VTM) in their analyses. As regards C t , (see
Fig. 7b) the corrected model underestimates the minimum value
at 901 azimuth while a fair agreement is obtained on the retreating
side and the beginning of the advancing side. It is noted that
negative C t corresponds to positive driving force so underestimation of the maximum negative C t value indicates that the average
power will be also underestimated.
The second test is at a moderate Reynolds number (Re100,000).
It was conducted at the low speed, low turbulence, wind tunnel of
NTUA (Morathakis, 1991), on a small H-type VAWT model (see
Table 2). Pressure measurements were taken at the mid-span of one of
the blades, for different values. Integration of the pressure distributions provided azimuthal variations of C n and C t .
Predicted and measured CL and CD polars are shown in Fig. 9
where simulations are performed with FOIL2W, while measured
data, for a slightly higher Reynolds number (Re 160,000) than
that of the test, are taken from Sheldahl and Klimas (1980).
Comparing the two sets of polars it is noted that FOIL2W slightly
over-predicts max . Also, in the measurements, the drop of CL in
the post stall region is deeper and thereafter the increase of CD is
steeper. The agreement in the linear part of CL is perfect while
FOIL2W slightly over-predicts CD0; however this is probably due to

Fig. 8. Predicted CL loop for 5 Sandia test turbine. potential corresponds

to purely inviscid predictions, ONERA corrected FOIL2W corresponds to ONERA
corrected results using FOIL2W polars.

Table 2
Characteristics of the NTUA test turbine.
Rotational speed (rpm)
Blade tip speed (m/s)
Number of blades
Solidity

143
12
2
0.3

Diameter (m)
Blade length (m)
Chord length (m)
Blade airfoil
Re (average)

1.6
1.1
0.12
NACA0018
100,000

the fact that FOIL2W results correspond to a lower Reynolds

number.
In Fig. 10 predictions of the azimuthal variation of C n and C t are
compared to measurements for the 4 case. Since the measured
loads are obtained by integrating pressure recordings, CD0 is not added
in the predictions. As in the rst validation test at this high value the
nal corrected loads (correction made on the basis of Eq. (14))
are independent of the polars (computed or measured) used in
correcting them. Also, the difference between the potential and the
corrected loads is very small and it is mainly limited in the azimuth
range [901, 1801]; over the second quadrant of the advancing side
during the upstroke of the blade. The fact that the corrected loads are
very similar for either of the polars is also depicted by the CL-
hysteresis loops shown in Fig. 11a. The zero lift angle of attack of the
loops is the same as in the rst test i.e. 2.51. A notable change in the

64

L. Shi et al. / J. Wind Eng. Ind. Aerodyn. 135 (2014) 5769

Fig. 9. Computed (using FOIL2W) and measured polars of NACA0018. Left CL, right CD.

Fig. 10. Comparison of predicted and measured (a) normal and (b) tangential force coefcients for 4 NTUA test turbine. potential corresponds to purely inviscid
predictions, steady polars meas. corresponds to corrected results based on steady-state measured polars, ONERA corrected meas. corresponds to ONERA corrected
results using measured polars.

Fig. 11. Predicted (a) CL and (b) CD loop for 4 NTUA test turbine potential corresponds to purely inviscid predictions, ONERA corrected meas. corresponds
to ONERA corrected results using measured polars, ONERA corrected FOIL2W corresponds to ONERA corrected results using FOIL2W polars.

slope of the CD loop with respect to the steady-state curve is obtained

as a result of the wake induced effects (see Fig. 11b). The agreement
between predicted and measured C n azimuthal variation is again very

good in this high case (see Fig. 10a). As in the rst test, the minimum
C t value at 901 azimuth is slightly under-predicted (see Fig. 10b). A
slight shift of the predicted C t variation is also observed in the rst

L. Shi et al. / J. Wind Eng. Ind. Aerodyn. 135 (2014) 5769

65

Fig. 12. Comparison of predicted and measured (a) normal and (b) tangential force coefcients for 3 NTUA test turbine ONERA corrected meas. corresponds to
ONERA corrected results using measured polars, ONERA corrected FOIL2W corresponds to ONERA corrected results using FOIL2W polars.

Fig. 13. Comparison of the two correction methods provided by Eq. (14) and Eq. (15) on the prediction of (a) normal force coefcient and (b) the CL loops for 3 NTUA test
turbine. Corrections made using measured polars.

quadrant of the advancing side; however, the agreement on the

retreating side is almost perfect.
Next two lower cases are considered ( 3 and 2) in which
pronounced ow separation takes place at least on the advancing side.
In Fig. 12 predicted azimuthal variations of C n and C t are compared to
measurements for the 3 case. In this case, dynamic stall takes place
on the advancing side while on the retreating side the ow remains
attached. Predictions employing both sets of steady-state polars
(measured and computed by FOIL2W) for the correction of the
potential loads are presented in the plots. Small differences are noted
in the prediction of C n between the two sets (see Fig. 12a). Deviations
are localised in the azimuth range [1201, 1351]. The deeper knee
obtained when correction is based on the measured polar is due to the
deeper CL drop in the post stall region exhibiting in the measured
steady state CL curve. The agreement with measurements is reasonable
while bigger differences are seen in the second quadrant of the
advancing side. The second quadrant corresponds to the end of the
blade downstroke, where the blade reaches the maximum negative
angle of attack, and the beginning of the upstroke. In this azimuth
range, dynamic stall develops; so C n rapidly drops, mainly because of
the drop of CL. C t increases both because of the drop of CL and the
consequent increase of CD. The increase in C t in the azimuth range of
[901, 180;1] is clearly seen in Fig. 12b. The agreement of the corrected
C t with the measured one, in this azimuth range, is very good

especially for the measured polars. A good agreement is also noted

on the retreating side while, as in the 4 case, a slight phase shift is
observed in the beginning of the advancing side.
In Fig. 13, the two correction options proposed in Section 3 are
compared. Clearly the usage of Eq. (14) (see Fig. 13a) gives better
prediction of the minimum C n (end of downstroke phase). The
earlier loss of lift imposed by Eq. (15) results in an underprediction of the minimum C n . Due to earlier stall, deeper dynamic
stall conditions are encountered by the airfoil, followed by a
retarded reattachment of ow during the upstroke as seen in
Fig. 13b. The deeper dynamic stall conditions in Fig. 13b are
indicated by the wider CL hysteresis loop. The retarded reattachment of the ow appears in the azimuth plot of C n as a phase shift
in the azimuth range [1351, 2251].
In Fig. 14 predicted azimuthal variations of C n and C t are compared
to measurements for the 2 case. In this low case, dynamic stall
takes place on both the advancing and the retreating side. Again
predictions employing both sets of steady-state polars for the correction of the potential loads are shown in the plots. Deviations between
predictions and measurements are bigger in this case. In measurements, C n exhibits a smooth sinusoidal behaviour on the advancing
side while in the predictions a knee associated with stall is seen at 901
azimuth. In the 3 case a similar knee appeared at 1201 azimuth.
Due to dynamic stall, an increase in C t is predicted at the same

66

L. Shi et al. / J. Wind Eng. Ind. Aerodyn. 135 (2014) 5769

azimuth angle. Checking the measured pressure distributions (not

shown herein), it was found that a at Cp is obtained over the whole
suction side of the blade throughout the advancing side. This implies
that the ow is fully separated even at low angles of attack at the
beginning of the advancing side. The explanation for this behaviour
could be that due to dynamic stall, which now occurs also on the
retreating side, separated ow over the blade has no time to recover
and reattach. So, despite the low angles of attack at the beginning of
the advancing side the ow remains fully separated. Such behaviour is
expected when the reduced frequency of the unsteady motion is very
high. Taking into account that the relative inow velocity W is
dominated by R, a rather moderate reduced frequency k  c=2R
0:075 is obtained. On the retreating side the agreement is better
except that the model over-predicts the maximum C n value at the
onset of dynamic stall, in the vicinity of 2251 azimuth. At this azimuth

angle C t appears to have a dip. This is a result of the rapid increase in

drag which also takes place in the onset of dynamic stall. Beyond 2251
azimuth, and within a very short azimuth range C n drops as a result of
the loss of lift following the occurrence of dynamic stall.
In Fig. 15a and b, the corrected (correction based on measured
polars) CL and CD hysteresis loops for the three values ( 2,3,4)
are compared. It is clear that in the 2 case, dynamic stall occurs
both at the positive and negative angles of attack regime. It is also
seen that as a result of dynamic stall an overshoot in the
maximum CL (positive and negative) is obtained beyond the steady
state CL max value. In dynamic stall conditions, CD loop takes an
eight like shape typical also to airfoils undergoing pitching
motion or periodic effective angle of attack variation.
The third case concerns a 12 kW H-type VAWT small wind
turbine tested at the Uppsala University (see Table 3). The blades

Fig. 14. Comparison of predicted and measured (a) normal and (b) tangential force coefcients for 2 NTUA test turbine ONERA corrected meas. corresponds
to ONERA corrected results using measured polars, ONERA corrected FOIL2W corresponds to ONERA corrected results using FOIL2W polars.

Fig. 15. Comparison of the (a) CL and (b) CD loops for 2,3,4 NTUA test turbine ONERA corrected results using measured polars are presented.

Table 3
Characteristics of the 12 kW H-type VAWT.
Power (kW)
Rotational speed (rpm)
Blade tip speed (m/s)
Rated wind speed (m/s)
Number of blades
Solidity

12
127
40
12
3
0.25

Hub height (m)

Diameter (m)
Blade length (m)
Chord length at mid-section (m)
Blade airfoil
Re (average)

6
6
5
0.25
NACA0021
650,000

L. Shi et al. / J. Wind Eng. Ind. Aerodyn. 135 (2014) 5769

67

Fig. 16. Computed (using FOIL2W) and measured polars of NACA0021. Left CL, right CD.

of the turbine are tapered at the two ends (upper and lower). The
tapering begins at 1 m from the tips and gives a chord at the tip
which is 60% of the chord at the mid-section of the blade. Power
measurements have been conducted for this wind turbine and
about 350 h of valid data in normal operation have been collected.
The average Reynolds number of this test was 650,000. Predicted
and measured CL and CD polars are shown in Fig. 16. Predictions are
obtained with FOIL2W, while measured data, for a slightly higher
Reynolds number (Re700,000) than that the turbine operates at, are
taken from Sheldahl and Klimas (1980). Comparing the two sets it is
noted that FOIL2W slightly over-predicts C L max . The at top behaviour
of CL in the vicinity of max is well predicted for the thick NACA 0021.
The agreement in the linear part of CL is perfect. Concerning drag,
FOIL2W captures the CD0, however the increase in drag in the post stall
region is steeper in the measurements.
Since the present model is two dimensional, the tapering of the
blades has only been taken into account indirectly in the analysis.
Two dimensional simulations are performed for the mid-section of
the rotor. Then for evaluating the power of the wind turbine the
2D driving force is integrated along the blade span after an a
posteriori correction for 3D effects is applied. The correction is
based on the local C L3D =C L2D ratio and on the wake induced angle
of attack W (induced by the chordwise 3D vortices) of the blade
calculated by a standard lifting line model in uniform axial inow
corresponding to the problem of xed wing. The C L3D =C L2D ratio is
directly applied as a correction to the lift calculated by the 2D
model. By means of the 3D induced angle of attack W an
additional wake induced drag component is calculated. It is noted
that due to the trapezoidal shape of the blade the 3D effects will
be rather limited. Both C L3D =C L2D and W exhibit a similar at
shape along the biggest portion of the span and they both
suddenly drop very close to the tips.
The mean power coefcient Cp and the maximum possible
uncertainty of the measurement as provided in Kjellin et al. (2011)
are shown in Fig. 17 against present predictions. The large error
according to the experimentalists is caused by uncertainties in the
wind speed measurement. Three predicted Cp curves are shown
in the plot. The rst, power curve is calculated by the purely
inviscid code (potential), without making any correction for
viscous effects. It is seen that the maximum efciency of the rotor
for ideal ow conditions is slightly above 0.40, far from Betz's
limit. The second power curve (steady) is calculated using the
steady state CL, CD polars of the NACA0021 (measured polars are
employed) and the local ow angle of attack provided by the free
wake code. Finally the third power curve (onera corrected) is
calculated using the ONERA corrected loads (correction based on

Fig. 17. Comparison of predicted and measured power coefcient of the 12 kW

turbine potential corresponds to purely inviscid predictions, steady corresponds to corrected results based on steady-state measured polars, ONERA
corrected corresponds to ONERA corrected results using measured polars.

Eq. (14)). For values less than 4 the ONERA corrected model over
estimates the measured mean power by  10%. This is a reasonable
difference taking into account that the measured curve corresponds to the electric power while the predicted one is the
aerodynamic power of the rotor. Deviations are much higher for
values higher than 4. But this is also expected given that the
efciency of the generator is considerably lower at low torques. If
the steady state polars are directly employed, without taking into
account the effect that virtual cambering has on loads, signicantly lower power coefcients are obtained.
In Fig. 18, the CL hysteresis loops for four different values lying in
different operational regimes are shown. At 4 (low wind speed) the
ow remains attached. The zero lift angle of attack of the CL loop is
1.61, lower than that of the two previous tests in which 0 2:51. The
virtual cambering effect is directly associated with the c=R parameter.
In the present rotor c=R 0:083 while the rst two rotors have
c=R 0:15. As in the previous tests, the slope of the loop is
signicantly higher than the slope of the steady state CL polar in the
linear part. At 3 and 3.5 (moderate wind speeds) where maximum
Cp is obtained, light separation of the ow takes place on the
advancing side of the rotation. At 2 (high wind speed) dynamic
stall takes place both on advancing and the retreating side. Especially
on the advancing side the blade encounters deep stall conditions as
indicated by the wide CL loop.

68

L. Shi et al. / J. Wind Eng. Ind. Aerodyn. 135 (2014) 5769

Fig. 18. Comparison of the CL loops for (a) 2, (b), 3, (c) 3.5, (d) 412 kW turbine.

5. Conclusion
A consistent vortex aerodynamic model for the analysis of VAWT
was presented. The model combines a vortex type free wake potential
ow solver with viscous corrections of loads. In particular the model
addresses the onset of BVI encounters by correcting the trajectories of
the vortices when approaching the blades and the virtual cambering
effect on the loads. The latter is insured by keeping the circulation
directly obtained by the potential solver as the attached generalised lift
circulation in the ONERA model.
Based on the three validation tests considered, the following
conclusions can be drawn:
1) At high to medium tip speed ratios the model correctly predicts
the aerodynamic performance of VAWT as depicted by the
azimuth variations of the blade loads. These conditions correspond to up to light stall conditions.
2) At low tip speed ratios, the model predictions are acceptable.
Under such conditions the blade undergoes deep stall and in
this sense the approximation of dynamic stall behaviour contained in the model are expected to work less well. The main
source of errors is connected to the 2D polars on which the
calculation of loads is based. However on the other hand this
renders the computation cost effective which is a feature
needed in design procedures.
Increasing the accuracy of the predictions will necessarily come
along with increased computational cost. One option in this respect is

to apply strong viscousinviscid interaction modelling in combination

with the double wake concept. In this connection a critical point
concerns the signicant azimuth variation of the Re number at low a
missing feature in the existing boundary layer models. The other
option is to better tune the correction model in which case data from
CFD simulations could signicantly contribute.
Finally as regards the 3D case, the modelling principles followed in the present work are readily extendable to 3D modelling
as reported in Shi (2012).

Acknowledgements
This work was partly funded by the European Commission
(Contract Grant no: DCI-ASIE/2010/240213) under ICARE project.
References
Basu, B.C., Hancock, G.J., 1978. The unsteady motion of a two-dimensional airfoil in
incompressible, inviscid ow. J. Fluid Mech. 87 (part 1), 159178.
Coton, F.N., Jiang, D., Galbraith, R.A. McD, 1994. An unsteady prescribed wake model
for vertical axis wind turbines. J. Power Energy 208, 1320.
Cottet, G.H., Koumoutsakos, P., 2008. Vortex Methods: Theory and Practice.
Cambridge University Press, Cambridge, UK, ISBN: 9780521061704.
Hansen M.O.L., Srensen D.N. CFD model for vertical axis wind turbine. Wind
energy for the new millennium. In: Proceedings of the European Wind Energy
Conference. Copenhagen, 2001. pp. 485488.
Hansen, M.O.L., Srensen, J.N., Voutsinas, S., Srensen, N., Madsen, H., 2006. Aa.
State of the art in wind turbine aerodynamics and aeroelasticity. Prog. Aerosp.
Sci. 42, 285330.

L. Shi et al. / J. Wind Eng. Ind. Aerodyn. 135 (2014) 5769

Katz, J., Plotkin, A., 1991. Low Speed Aerodynamics: From Wing Theory to Panel
Methods. McGraw-Hill, International Editions, Singapore, ISBN: 0-07-100876-4.
Kjellin, J., Bolow, F., Eriksson, S., Deglaire, P., Leijon, M., Bernhoff, H., 2011. Power
coefcient measurements on a 12 kW straight bladed vertical axis wind
turbine. Renew. Energy 36, 30503053.
Larsen, T.J., Madsen, H.Aa. On the way to reliable aeroelastic load simulation on
VAWT's. In: Proceedings of EWEA Annual Event 2013. EWEA The European
Wind Energy Association. Vienna, 2013.
Madsen, H.Aa., Paulsen, U.S., Vitae, L. Analysis of VAWT aerodynamics and design
using the Actuator Cylinder ow model. In: Proceedings of Making Torque from
the Wind Conference. Oldenburg, 912 October, 2012.
Morathakis, E., 1991. Theoretical and Experimental Study of Unsteady Phenomena
in Turbines (Ph.D. thesis). National Technical University of Athens.
Paraschivoiu, I., 1988. Double-multiple stream tube model for studying vertical-axis
wind turbines. J. Propuls. Power 4, 370377.
Paraschivoiu, I., 2002. Wind Turbine Design with Emphasis on Darrieus Concept.
Presses Internationales Polytechnique, Quebec, Canada, ISBN: 978-2-55300931-0.
Petot, D., 1989. Differential equation modelling of dynamic stall. Recherch
Aerospatiale 5, 5972.
Ponta, F.L., Jacovkis, P.M., 2001. A vortex model for Darrieus turbine using nite
element techniques. Renew. Energy 24, 118.
Riziotis, V.A., Voutsinas, S.G. Dynamic stall on wind turbine rotors: comparative
evaluation study of different models. In: Proceedings of the EWEC'97. Dublin,
Ireland, 1997.
Riziotis, V.A., Voutsinas, S.G., 2008. Dynamic stall modelling on airfoils based on
strong viscousinviscid interaction coupling. J. Num. Methods Fluids 56,
185208.
Scheurich, F., Fletcher, T.M., Brown, R.E., 2011. Simulating the aerodynamic
performance and wake dynamics of a vertical-axis wind turbine. Wind Energy
J. 14 (2), 159177.

69

Simo Ferreira C.J., van Bussel G., van Kuik G. 2D CFD simulation of dynamic stall on a
vertical axis wind turbine: verication and validation with PIV measurements. In:
45th AIAA Aerospace Sciences Meeting and Exhibit/ASME Wind Energy Symposium. Reno, 2007, pp. 1619116201.
Sheldahl, R.E., Klimas, P.C., 1980. Aerodynamic characteristics of seven symmetrical
airfoil sections through 180-degree angle of attack for use in aerodynamic
analysis of vertical axis wind turbines, Sandia Technical Report SAND80-2114.
Shi, L., 2012. Assessment of Free Wake Aerodynamic Models on the Analysis of
Vertical Axis Wind Turbines (VAWT) (Master thesis). China-EU Institute for
Clean And Renewable Energy ICARE.
Simo Ferreira, C.J.S., 2009. The Near Wake of the VAWT 2D and 3D Views of the
VAWT Aerodynamics (Ph.D. thesis). Technical University of Delft.
Strickland, J.H., 1975. The Darrieus Turbine: a Performance Prediction Model Using
Multiple Streamtubes, Sandia Technical Report SAND75-0431.
Strickland, J.H., Webster, B.T., Nguyen, T., 1979. A vortex model of the Darrieus
turbine: an analytical and experimental study. Trans. ASME. J. Fluids Eng. 101,
500505.
Strickland, J.H., Smith, T., Sun, K., 1981. A Vortex Model of the Darrieus Turbine: an
Analytical and Experimental Study. Sandia Technical Report SAND81-7017.
Sutherland, H.J., Berg, D.E., Ashwill, T.D., 2012. A Retrospective of VAWT Technology, Sandia Technical Report SAND 2012-0304.
Vezza, M., McD Galbraith, R.A., 1985. An inviscid model of unsteady aerofoil ow
with xed upper surface separation. J. Num. Methods Fluids 5, 577592.
Voutsinas, S.G., Riziotis, V.A. Vortex particle modelling of stall on rotors. Application
to wind turbines. In: Proceedings of the Fluids Engineering Division Summer
Meeting, ASME. San Diego, California, USA, 1996.
Zanon, A., Giannattasio, P., Simo Ferreira, C.J.S., 2013. A vortex panel model for the
simulation of the wake ow past a vertical axis wind turbine in dynamic stall.
Wind Energy J. 16 (5), 661680.