2D CFD simulation of dynamic stall on a vertical axis

wind turbine: verification and validation with PIV
measurements
Carlos Sim˜ao Ferreira

Gerard van Bussel

Gijs van Kuik

Delft University of Technology, Delft, 2629 HS, The Netherlands
After a decrease of interest in the 1990’s, the research on Vertical Axis Wind Turbines
(VAWT)
a
has reappeared in the last years as a result of the its increasing application
in the built environment, where VAWTs present several advantages over Horizontal Axis
Wind Turbines (HAWT). The VAWT has an inherent unsteady aerodynamic behavior due
to the variation of angle of attack with the angle of rotation θ, perceived velocity and
consequentially Reynolds number. The phenomenon of dynamic stall is then an inherent
effect of the operation of a Vertical Axis Wind Turbine (VAWT) at low tip speed ratios
(λ), having a significant impact in both loads and power.
The complexity of the unsteady aerodynamics of the VAWT makes it extremely at-
tractive to be analyzed using Computational Fluid Dynamics (CFD) models, where an
approximation of the continuity and momentum equations of the Navier-Stokes equations
set is solved.
The complexity of the problem and the need for new design approaches for VAWT for
the built environment has driven the authors of this work to focus the research of CFD
modeling of VAWT not in the perspective of creating one large academic model to test a
particular situation, but to develop a work that would:
• verify the sensitivity of the model to its grid refinement (space and time),
• evaluate the differences between the different commonly used turbulence models (Laminar,
Spalart −Allmaras and k −), and
• be evaluated using Particle Image Velocimetry (PIV) experimental data, thus deter-
mining the suitability of this data for model validation.
The 2D model created represents the middle section of a single bladed VAWT with
infinite aspect ratio. The model simulates the experimental work
b
of flow field measurement
using Particle Image Velocimetry (PIV) by Sim˜ao Ferreira et al
1
for a single bladed VAWT,
for two different reference Reynolds numbers of Re = 52000 and Re = 70000 for three tip
speed ratios: λ = 2, 3, 4.
The results show the suitability of the PIV data for the validation of the model, the
unsuitability of the application of a single turbulent model and the high sensitivity of the
model to grid refinement.
Nomenclature
c airfoil/blade chord, m
C
N
normal force

PhD Researcher, Faculty of Aerospace Engineering, Kluyverweg 1, 2629 HS, Delft, The Netherlands.

Associate Professor, Faculty of Aerospace Engineering, Kluyverweg 1, 2629 HS, Delft, The Netherlands.

Professor, Faculty of Aerospace Engineering, Kluyverweg 1, 2629 HS, Delft, The Netherlands.
a
In this paper the reference VAWT will be used for lift driven VAWT, unless stated otherwise
b
Note: together with this paper, the authors have submitted to this conference a paper titled ”2D PIV visualization of
Dynamic stall on a Vertical Axis Wind Turbine”, work that precedes the one presented in this paper, describing the experimental
work that originated the data used to validate the results of the simulations presented in this paper.
1 of 11
American Institute of Aeronautics and Astronautics
C
T
tangential force
D rotor diameter, m
k reduced frequency
R rotor radius, m
Re Reynolds number
U

Unperturbed velocity m/s
α angle of attack
θ azimuth angle
ω perturbation frequency
Γ circulation m
2
/s
ρ density of the fluid kg/m
3
µ
Γ
average value of circulation m
2
/s
I. Introduction
The increasing awareness of the need for environmentally sustainable housing and cities has driven the
promotion of wind energy conversion systems for the built environment. One of the results of the development
of solutions for the built environment is the reappearance of Vertical Axis Wind Turbines (VAWTs). In the
built environment, the VAWT presents several advantages over the more common Horizontal Axis Wind
Turbines (HAWTs), namely: its low sound emission (consequence of its operation at lower tip speed ratios),
better esthetics due to its three-dimensionality (more suitable for integration in some architectural projects,
since it follows the sense of volume of the building), its insensitivity to yaw and its increased performance
in skewed flow (see Mertens et al
2
and Sim˜ao Ferreira et al
3,4
).
The phenomenon of dynamic stall is an inherent effect of the operation of a VAWT at low tip speed
ratios (λ). The presence of dynamic stall has a significant impact in both load and power.
Modeling the VAWT in dynamic stall presents five immediate challenges:
• the unsteady component of the flow requires a time accurate model, adding an extra dimension (time)
to the numerical grid,
• the geometry of the rotor does not allow for important spatial/time grid simplifications to be applied
(example: moving reference frames or radial symmetry),
• the large amount of shed vorticity implies that the model should be sensitive to numerical dissipation
and viscous modeling of the vorticity,
• the geometry of a Vertical Axis Wind Turbine results in blade-vortex interaction at the downwind
passage of the blade between the blade and the shed vorticity that was generated at the upwind
passage. This means that the development of the shed vorticity must be correctly modeled inside the
entire rotor diameter; in order to avoid numerical dissipation, the spatial resolution of the grid must
be very fine not only in the immediacy of the blades but over the entire rotor, and
• the variation of angle of attack of the blade with azimuth angle implies a varying relation/dominance
between lift and drag force on the blade (resulting in moments during the rotation where the VAWT is
actually being decelerated because only drag force is present). The correct use of a turbulence model
and near-wall models are dominant in these situations. This is particulary important at low tip speed
ratios, where the power output of the VAWT is negative (for a certain range of λ); the performance at
low tip speed ratios is highly important for start-up behavior, one of the disadvantages usually pointed
out to VAWT.
This numerical model aims at simulating the experimental work by Sim˜ao Ferreira et al;
1
this experi-
mental work comprises several flow conditions, covering two different Reynolds numbers and three tip sped
ratios. Yet, only one Reynolds number (Re = 50000) and a single tip speed ratio (λ = 2) will be considered.
Although simulations at other Reynolds numbers and tip speed ratios were performed their presentation in
the scope of this paper does not add more insight to the main conclusions.
Contrary to the previous attempts to model the VAWT using CFD (see Hansen et al
5
, Allet et al
6
,
Paraschivoiu et al
7
, Paraschivoiu et al
8
and Paraschivoiu
9
), the validation of the results is achieved not by
2 of 11
American Institute of Aeronautics and Astronautics
comparing the load on the blades but by comparing the vorticity in the rotor area. This work aims also at
demonstrating the suitability of PIV data for validation of numerical models for this specific problem.
The results of the simulations show the impact of grid refinement and choice of turbulence model. The
initial focus of this research (quantification of accuracy) is thus changed into the pursuit of a correct modeling
strategy.
II. Model geometry and computational grid
The geometry of the model is a 2D representation of the experimental setup of Sim˜ao Ferreira et al.
1
The
model’s wall boundary conditions consists of two walls spaced 1.25m, where a 0.4m diameter single-bladed
Darrieus VAWT is placed. The rotor is represented by an 0.05m chord NACA0015 airfoil and the 0.05m
rotor axis. The rotor axis is placed over the symmetry position of the wind tunnel. Comparative simulations
with and without wind tunnel wall boundary conditions (with and without blockage) showed that there is
no significant difference in results and any effect of the wind tunnel walls is negligible. The inlet and outlet
boundary conditions are placed respectively 10D and 14D upwind and downwind of the rotor, allowing a
full development of the wake.
The model comprises a 2D spatial grid, simulating the conditions at the middle cross-section of the exper-
imental setup. The grid is composed of four non-conformal sub-grids, each a structured grid of quadrilateral
elements. Figure 1 presents a schematic of the shape and location of each sub-grid and the wall boundary
conditions representing the airfoil/blade, wind tunnel walls and rotor axis (the flow inlet and outlet boundary
conditions are not represented).
The use of sub-grids is necessary due to the movement of the rotor elements. Thus, the sub-grids Rotor
Diameter, Blade sub-grid I, and Blade sub-grid II rotate with an angular velocity Ω, while the sub-grid
Wind Tunnel remains fixed. The option of dividing the rotor space in three sub-grids allows the use of a
structured grid without compromising its quality or requiring a large mesh with over refinement in areas
of lower importance. A quadrilateral structured grid also allows an easier control of the refinement of the
grid; this aspect is crucial in the present work, since the sensitivity to the refinement of the space grid is
evaluated.
The most refined sub-grid is Blade sub-grid I which defines the geometry of the airfoil and its immediate
wake development flow region. The reference fine grid comprises 3305 nodes over the airfoil surface, where
the height of the first row of cells is set to y
+
≈ 1 (when θ = 90

, k − model). The total model size
comprises approx. 1.6 ∗ 10
6
cells.
Blade
Blade sub-grid I
Blade sub-grid II
Rotor Diameter grid
Rotor Axis
Wind Tunnel grid
Wind Tunnel wall II
Wind Tunnel wall I
U
’
Figure 1. Partial schematic of the model geometry, sub-grid distribution and boundary conditions.
3 of 11
American Institute of Aeronautics and Astronautics
III. Simulated flow conditions
The simulation aimed at representing the flow conditions of the experimental work. Simulations have
been performed for λ = 2, 3, and 4, and incoming flows of U

= 7.5m/s and 10.5m/s. Yet, for the scope of
this paper, only the simulations regarding the case of λ = 2 and U

= 7.5m/s (resulting in Ω = 75rad/s) are
discussed. The level of unsteadiness is determined by the reduced frequency
c
k, which for this experimental
work was k = 0.125, placing the work in the unsteady aerodynamics region.
Due to the importance of the induction of the rotor, it is necessary to perform a simulation for several
rotations until a long fully developed wake is present. All values presented in this paper relate to the
revolutions of the rotor after a periodic post-transient solution is found. This transient solution took up to
7 revolutions from the initial stand-still situation to the fully developed wake. This solution was used as a
base for the sensitivity analysis; a second transient solution, due to change of grid refinement or turbulence
model, took 2 −3 revolutions to overpass.
IV. Verification and validation of Laminar, k − and S −A models
The limits of the computational hardware and time frame of the computations do not allow the use of a
space-time grid fine enough to calculate all scales of turbulence. The use of the different turbulence models,
although allowing a coarser grid, impacts the result of the simulation. The increase of computational power
in the latest years has lead to the use of increasingly more computationally expensive turbulence models,
in the hope of achieving more accurate simulations. Verification of simulation results of an airfoil in a
similar Darrieus motion was previously performed by Paraschivoiu
8,9
), with discussion of the differences of
results for several one-equation/two-equations Reynolds Averaged turbulence models. Yet, the phenomenon
of dynamic stall of the VAWT at low tip speed ratio is highly dependent of the laminar separation occurring
at the leading edge. The use of fully turbulent models can limit the occurrence of this laminar separation,
resulting in an incorrect turbulent flow development.
This work presents a verification of the effect of applying either a laminar or a turbulent model in the
entire flow. The Laminar consists of the momentum and continuity equations without closure relations.
The Spalart −Allmaras (S −A) and k − models assume, respectively, a one-equation and a two-equation
model for the eddy viscosity. The description of the models can be found in the Fluent c User’s Manual.
10
Figure 2 shows the values for tangential and normal force over the first 180

degrees of rotation, made
non-dimensional by the scale of the airfoil and its rotational velocity (
1
2
ρλ
2
U
2

c). Notice that the models
assuming turbulent flow (k − and S − A) present similar results, diverging from the laminar model. The
normal force C
N
d
-force in the radial direction of the rotor- is, in the laminar case, larger in the first quarter
of rotation and lower in the second quarter of rotation, in comparison with the turbulent flow simulations.
The laminar flow also presents non-zero normal force at azimuthal angle θ = 0, thus showing an effect on
airfoils circulation of the vorticity generated in the downwind portion of the rotation, dragged with the
airfoil. A more significant difference is in the tangential component of the force on the airfoil (C
T
), which,
in the upwind half of the rotation, presents an almost opposite sign of force; the tangential force is very
dependent on the friction force, and thus, on the wall treatments associated with each turbulence model. The
differences in forces are, despite large, in the range of the difference found in several works where numerical
simulations are compared with experimental results of forces or pressure measurements; this result opens
some questions regarding the suitability of integral lift force data for the validation of simulations.
The results from the current simulation in terms of flow field measurements are compared with the
experimental flow field measurements using PIV of Sim˜ao Ferreira et al.
1
Figure 3 shows the phase locked
average at θ = 96.5

of the vorticity field (non-dimensioned by the scale of the airfoil and its rotational
velocity) for the three simulations with different turbulence models. Comparing these with the experimental
results of the phase locked average vorticity obtained with PIV in the work of Sim˜ao Ferreira et al
1
(see
figure 4), the Laminar model simulation better predicts the development of the large leading edge separated
vorticity and the rolling up of the counter-clockwise vorticity at the trailing edge; in fact, the models
assuming fully turbulent flow suppress the development of the leading edge separation and present a vorticity
c
The reduced frequency k is defined as k = ω.c/2V , where ω is the angular frequency of the unsteadiness, c is the blade’s
chord and V is the velocity of the blade. In this experience, due to the variation of V with rotation angle, k was defined as
k = ω.c/(2.λ.V

) = ω.c/(2.ω.R) = c/(2.R), where λ is the tip speed ratio and R is the radius of rotation
d
The normal and tangential forces are presented in their non-dimensioned form C
N
and C
T
. The scaling is done through
the dimensions associated with the airfoil movement
1
2
ρλU
2

c.
4 of 11
American Institute of Aeronautics and Astronautics
distribution that does not fit the experimental data. This results shows the unsuitability of using a turbulent
model over the entire flow for the simulation of dynamic stall in this Reynolds range.
0 15 30 45 60 75 90 105 120 135 150 165 180
−1
0
1
2
3
4
θ (degrees)


C
N
C
T
C
N
C
T
laminar
k−e
SA
Figure 2. Comparison of normal and tangential force on airfoil calculated with Laminar, k − and S −A models
(first 180

). The results are the phase locked average over 10 rotations. Error bar represents uncertainty at
95% confidence level.
Figure 3. Vorticity field around airfoil for three turbulence model, θ = 96.5

. Each result is the phase locked
average of 10 samples.
V. Verification and validation of time grid refinement
On of the most important parameters in a time accurate simulation is the definition of the time step.
Figure 5 presents the value of normal and tangential force in the airfoil (first 180

) for four different time
steps. The results of normal and tangential force show a significant sensitivity for the refinement of the time
grid; the simulations show a divergence between the azimuthal positions of 60

and 160

. This difference can
once again be related to the accuracy of the simulation of the development of the leading edge separation
vorticity and the consequent development of the trailing edge vorticity roll-up.
5 of 11
American Institute of Aeronautics and Astronautics
Figure 4. Experimental results of the vorticity (Sim˜ao Ferreira et al
1
).
Figure 6 presents the phase locked average of the vorticity around the airfoil for four different time grid
refinements at four different positions (in the 90

region). Comparing the simulations with the experimental
observations of figure 4 and remaining results of Sim˜ao Ferreira et al,
1
it is possible to see that the simulation
with ∆ t = 1

∗ Ω
−1
presents an underestimation of the generation and evolution of the vorticity, both the
one detached at the leading edge and the one shed at the trailing edge. This underestimation is smaller for
∆ t = 0.5

∗Ω
−1
and the simulations with ∆ t = 0.25

∗Ω
−1
and ∆ t = 0.125

∗Ω
−1
present an overestimation
of the shed vorticity.
The results show that with increasing refinement of the time grid (∆t), an increasing generation of
vorticity at the leading edge occurs, paired by an earlier (in azimuthal position) concentration and rolling up
of counter-clockwise vorticity at the trailing edge. This trend can be observed in the value of circulation of
the clockwise vorticity shed from the leading edge when compared with the experimental values (figure 7).
The value of circulation is calculated in the same procedure as used by Sim˜ao Ferreira et al.
1
0 15 30 45 60 75 90 105 120 135 150 165 180
−1
0
1
2
3
4
5
θ (degrees)


C
N
C
T
C
N
C
T
∆t=1
°
*Ω
−1
∆t=0.5
°
*Ω
−1
∆t=0.25
°
*Ω
−1
∆t=0.125
°
*Ω
−1
Figure 5. Comparison of normal and tangential force on airfoil with Laminar model, for the cases ∆ t =
1

∗ Ω
−1
,∆ t = 0.5

∗ Ω
−1
,∆ t = 0.25

∗ Ω
−1
,and ∆ t = 0.125

∗ Ω
−1
(first 180

). The results correspond to the phase
locked average of 20, 10, 5 and 5 samples.
The anticipation/delay of flow phenomena is important to notice. Comparing with experiments, similar
vorticity patterns to those observed in the four simulations occur in reality. However, it is the divergence of
azimuthal angle at which they occur that is the basis of error of the simulation.
One of the properties of the phenomena of dynamic stall in a Darrieus is the occurrence of a strong
random component. Figures 8 and 9 present the phase locked average distributions of vorticity and three
instantaneous distributions, for two simulations at different refinements of the time grid. There are two
6 of 11
American Institute of Aeronautics and Astronautics
Figure 6. Vorticity field around airfoil for four simulation time steps, after the 90

position (Laminar model).
The results correspond to the phase locked average of 20, 10, 5 and 5 samples.
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
4.5
5.0
60 75 90 105 120 135 150
u (degrees)
Experimental
CFD simulation
ì=2.0

I
,
c. ì.U
’
At=1°.ȍ
-1
At=0.5°.ȍ
-1
At=0.25°.ȍ
-1
At=0.125°.ȍ
-1
Figure 7. Circulation of the separated leading edge vorticity. Comparison of simulation values with experi-
mental results.
7 of 11
American Institute of Aeronautics and Astronautics
aspects to compare between the two cases: the presence of more and smaller vortices relating to the leading
edge separated vorticity in the case of lower refinement, as compared with the presence of only two large
clockwise vortices in the case of a more refined time grid; and the larger size of the counter-clockwise vorticity
generated over the surface of the airfoil (past the first leading edge vortex) in the finer grid case.
Figure 8. Comparison of phase locked average with instantaneous sample, ∆ t = 0.5

∗ Ω
−1
, θ = 96.5

.
Comparing the simulations with the phase locked averaged and instantaneous experimental measurements
of figures 10 and 11 (neglecting the difference of azimuthal position), it is possible to see that in the coarser
time grid there is an over fragmentation/spreading of the vorticity, in the case of the finer time grid, there
is a over concentration of vorticity in two single large vortices.
Although the results do not represent accurately all the properties of the real flow, they point to the
possibility of the random component having a mainly 2D origin and thus, being passible of simulation
through a 2D model (avoiding the expenses of a 3D representation).
Figure 9. Comparison of phase locked average with instantaneous sample, ∆ t = 0.25

∗ Ω
−1
, θ = 96.5

.
8 of 11
American Institute of Aeronautics and Astronautics
Figure 10. Vorticity field around airfoil (experimental PIV data), phase locked average, θ = 113

.
Figure 11. Vorticity field around airfoil (experimental PIV data), instantaneous sample, θ = 113

.
9 of 11
American Institute of Aeronautics and Astronautics
VI. Verification of the influence of spatial grid size
Having verified the influence of the refinement of the grid in time, it is now necessary to view the
implication of a coarsening of the spatial grid.
In this section we compare the results of the reference fine grid with a model in which the sub-grid Blade
sub-grid I is coarsened by a factor of 4 in area (the distance between nodes in x an y are doubled). The
comparison for the simulated resulting forces for the reference and coarsened grid is shown in figure 12. The
forces for the coarse space grid follow the trend seen in the case of refinement of the time grid. These force
development are a result of the characteristics of vorticity development previously observed.
0 15 30 45 60 75 90 105 120 135 150 165 180
-1
0
1
2
3
4
T (degrees)
C
N
C
T
C
N
C
T
Coarse grid
Fine grid
Figure 12. Comparison of normal and tangential force on airfoil , reference fine and coarse grids, ∆ t = 0.5

∗Ω
−1
.
The results correspond to the phase locked average of 10 samples.
Figure 13 compares the phase locked vorticity distribution of fine and coarse mesh at θ = 96.5

. In
the coarsened mesh the over estimation of the leading edge vorticity and early roll-up of the trailing edge
vorticity occurs, as previously seen in the case of refinement of the time grid, but in this case in opposite
direction.
VII. Conclusions
The results of the CFD simulations indicate the unsuitability of using a single turbulence model scheme
for the numerical simulation of a 2D Darrieus in dynamic stall at low tip speed ratios. The use of 1-
equation/2-equation models that assume turbulent flow over the entire field results in a partial suppression
of the magnitude of the leading edge laminar separation and in a reduction of maximum normal force.
Although the resulting flow field of the three turbulent models differ totally in the spatial distribution of
vorticity, the average tangential and normal forces over the upwind semi-rotation of the airfoil is similar,
thus showing the difficulty of using experimental force/performance data for the validation at low λ. The
Laminar model, although more suitable for the simulation of the leading edge separation, leads,at increasing
refinement of the time grid, to over prediction of the generation and evolution of the vorticity.
The experimental flow field data obtained using Particle Image Velocimetry proved to be a very useful
method of validation of the rotor aerodynamics, since it gives direct insight into the the flow behavior
(contrary to the total force, which is an integral result of the aerodynamics).
Further research will aim at developing numerical simulations with multiple turbulence models over the
grid, thus aiming to improve the simulation of the development of the counter-clockwise vorticity shed at
the trailing edge (resulting in a more correct development of the leading edge separation vorticity).
References
1
Sim˜ao Ferreira, C., van Bussel, G., and van Kuik, G., “2D PIV visualization of dynamic stall on a vertical axis wind
turbine,” Abstract submited for the 2007 AIAA/ASME Wind Energy Symposium, 2007.
10 of 11
American Institute of Aeronautics and Astronautics
Figure 13. Comparison of the vorticity field around airfoil for coarse and fine grid, θ = 96.5

.
2
Mertens, S., van Kuik, G., and van Bussel, G., “Performance of a H-Darrieus in the Skewed Flow on a Roof,” Journal of
Solar Energy Engineering, Vol. 125, 2003, pp. 433–440.
3
Sim˜ao Ferreira, C., van Bussel, G., and van Kuik, G., “An analytical method to predict the variation in performance of
an H-Darrieus in skewed flow and its experimental validation,” Proceedings of the European Wind Energy Conference 2006,
2006.
4
Sim˜ao Ferreira, C., van Bussel, G., and van Kuik, G., “Wind tunnel hotwire measurements, flow visualization and thrust
measurement of a VAWT in skew,” AIAA/ASME Wind Energy Symposium, 2006.
5
Hansen, M. and Søresen, D., “CFD Model for Vertical Axis Wind Turbine,” Wind Energy for the New Millennium-
Proceedings of the European Wind Energy Conference, Copenhagen, Denmark, 2001.
6
Allet, A., Hall´e, S., and Paraschivoiu, I., “Numerical Simulation of Dynamic Stall Around an Airfoil in Darrieus Motion,”
Journal of Solar Energy Engineering, Vol. 121, February 1999, pp. 69–76.
7
Paraschivoiu, I. and Allet, A., “Aerodynamic Analysis of the Darrieus Wind Turbines Including Dynamic-Stall Effects,”
Journal of Propulsion and Power, Vol. 4, No. 5, 1988, pp. 472–477.
8
Paraschivoiu, I. and B´eguier, C., “Visualization, Measurements and Calculations of Dynamic Stall for a Similar Motion
of VAWT,” Proceedings of the European Wind Energy Conference, Herning, Denmark, 1998.
9
Paraschivoiu, I., Wind Turbine Design - With Emphasis on Darrieus Concept, Polytechnic international Press, 2002.
10
Fluent User’s Manual , Fluent Inc.
11 of 11
American Institute of Aeronautics and Astronautics

This means that the development of the shed vorticity must be correctly modeled inside the entire rotor diameter. Introduction The increasing awareness of the need for environmentally sustainable housing and cities has driven the promotion of wind energy conversion systems for the built environment. In the built environment. its insensitivity to yaw and its increased performance in skewed flow (see Mertens et al2 and Sim˜o Ferreira et al3. • the geometry of a Vertical Axis Wind Turbine results in blade-vortex interaction at the downwind passage of the blade between the blade and the shed vorticity that was generated at the upwind passage. The correct use of a turbulence model and near-wall models are dominant in these situations. Allet et al6 . • the geometry of the rotor does not allow for important spatial/time grid simplifications to be applied (example: moving reference frames or radial symmetry). where the power output of the VAWT is negative (for a certain range of λ). since it follows the sense of volume of the building). m reduced frequency rotor radius. Paraschivoiu et al8 and Paraschivoiu9 ). Although simulations at other Reynolds numbers and tip speed ratios were performed their presentation in the scope of this paper does not add more insight to the main conclusions. namely: its low sound emission (consequence of its operation at lower tip speed ratios).CT D k R Re U∞ α θ ω Γ ρ µΓ tangential force rotor diameter. One of the results of the development of solutions for the built environment is the reappearance of Vertical Axis Wind Turbines (VAWTs). This is particulary important at low tip speed ratios. one of the disadvantages usually pointed out to VAWT. m Reynolds number Unperturbed velocity m/s angle of attack azimuth angle perturbation frequency circulation m2 /s density of the fluid kg/m3 average value of circulation m2 /s I. in order to avoid numerical dissipation. • the large amount of shed vorticity implies that the model should be sensitive to numerical dissipation and viscous modeling of the vorticity. the spatial resolution of the grid must be very fine not only in the immediacy of the blades but over the entire rotor. Contrary to the previous attempts to model the VAWT using CFD (see Hansen et al5 .4 ). Yet. better esthetics due to its three-dimensionality (more suitable for integration in some architectural projects. covering two different Reynolds numbers and three tip sped ratios.1 this experia mental work comprises several flow conditions. only one Reynolds number (Re = 50000) and a single tip speed ratio (λ = 2) will be considered. This numerical model aims at simulating the experimental work by Sim˜o Ferreira et al. a The phenomenon of dynamic stall is an inherent effect of the operation of a VAWT at low tip speed ratios (λ). The presence of dynamic stall has a significant impact in both load and power. Paraschivoiu et al7 . and • the variation of angle of attack of the blade with azimuth angle implies a varying relation/dominance between lift and drag force on the blade (resulting in moments during the rotation where the VAWT is actually being decelerated because only drag force is present). the performance at low tip speed ratios is highly important for start-up behavior. Modeling the VAWT in dynamic stall presents five immediate challenges: • the unsteady component of the flow requires a time accurate model. the VAWT presents several advantages over the more common Horizontal Axis Wind Turbines (HAWTs). the validation of the results is achieved not by 2 of 11 American Institute of Aeronautics and Astronautics . adding an extra dimension (time) to the numerical grid.

II. The results of the simulations show the impact of grid refinement and choice of turbulence model. the sub-grids Rotor Diameter. k − model). Thus. The grid is composed of four non-conformal sub-grids. wind tunnel walls and rotor axis (the flow inlet and outlet boundary conditions are not represented). each a structured grid of quadrilateral elements. Wind Tunnel wall I U Blade Rotor Diameter grid Blade sub-grid I Rotor Axis Blade sub-grid II Wind Tunnel grid Wind Tunnel wall II Figure 1.6 ∗ 106 cells. since the sensitivity to the refinement of the space grid is evaluated. Model geometry and computational grid The geometry of the model is a 2D representation of the experimental setup of Sim˜o Ferreira et al. The model comprises a 2D spatial grid. The rotor axis is placed over the symmetry position of the wind tunnel. 3 of 11 American Institute of Aeronautics and Astronautics .1 The a model’s wall boundary conditions consists of two walls spaced 1. 1. The initial focus of this research (quantification of accuracy) is thus changed into the pursuit of a correct modeling strategy.05m rotor axis. where a 0. Blade sub-grid I.comparing the load on the blades but by comparing the vorticity in the rotor area. and Blade sub-grid II rotate with an angular velocity Ω.25m. where the height of the first row of cells is set to y + ≈ 1 (when θ = 90◦ .05m chord NACA0015 airfoil and the 0. The option of dividing the rotor space in three sub-grids allows the use of a structured grid without compromising its quality or requiring a large mesh with over refinement in areas of lower importance.4m diameter single-bladed Darrieus VAWT is placed. while the sub-grid Wind Tunnel remains fixed. Comparative simulations with and without wind tunnel wall boundary conditions (with and without blockage) showed that there is no significant difference in results and any effect of the wind tunnel walls is negligible. The most refined sub-grid is Blade sub-grid I which defines the geometry of the airfoil and its immediate wake development flow region. simulating the conditions at the middle cross-section of the experimental setup. This work aims also at demonstrating the suitability of PIV data for validation of numerical models for this specific problem. sub-grid distribution and boundary conditions. A quadrilateral structured grid also allows an easier control of the refinement of the grid. allowing a full development of the wake. The total model size comprises approx. Partial schematic of the model geometry. Figure 1 presents a schematic of the shape and location of each sub-grid and the wall boundary conditions representing the airfoil/blade. this aspect is crucial in the present work. The rotor is represented by an 0. The inlet and outlet boundary conditions are placed respectively 10D and 14D upwind and downwind of the rotor. The use of sub-grids is necessary due to the movement of the rotor elements. The reference fine grid comprises 3305 nodes over the airfoil surface.

The increase of computational power in the latest years has lead to the use of increasingly more computationally expensive turbulence models.5m/s. 2 4 of 11 American Institute of Aeronautics and Astronautics . This solution was used as a base for the sensitivity analysis. The Laminar consists of the momentum and continuity equations without closure relations. the phenomenon of dynamic stall of the VAWT at low tip speed ratio is highly dependent of the laminar separation occurring at the leading edge. respectively. a second transient solution. which.5◦ of the vorticity field (non-dimensioned by the scale of the airfoil and its rotational velocity) for the three simulations with different turbulence models. The use of the different turbulence models.R) = c/(2.5m/s (resulting in Ω = 75rad/s) are discussed. a one-equation and a two-equation model for the eddy viscosity. in the hope of achieving more accurate simulations. k was defined as k = ω. where ω is the angular frequency of the unsteadiness.c/(2. IV.1 Figure 3 shows the phase locked a average at θ = 96. Yet. where λ is the tip speed ratio and R is the radius of rotation d The normal and tangential forces are presented in their non-dimensioned form C N and CT .V∞ ) = ω. A more significant difference is in the tangential component of the force on the airfoil (CT ). dragged with the airfoil. In this experience.c/2V .9 ). on the wall treatments associated with each turbulence model. in the laminar case. c is the blade’s chord and V is the velocity of the blade. Notice that the models 2 assuming turbulent flow (k − and S − A) present similar results. despite large. All values presented in this paper relate to the revolutions of the rotor after a periodic post-transient solution is found. in fact. Simulations have been performed for λ = 2. with discussion of the differences of results for several one-equation/two-equations Reynolds Averaged turbulence models. This work presents a verification of the effect of applying either a laminar or a turbulent model in the entire flow. The use of fully turbulent models can limit the occurrence of this laminar separation. The differences in forces are.5m/s and 10. The Spalart − Allmaras (S − A) and k − models assume. Comparing these with the experimental results of the phase locked average vorticity obtained with PIV in the work of Sim˜o Ferreira et al1 (see a figure 4).III.λ. The level of unsteadiness is determined by the reduced frequency c k.R). and 4. only the simulations regarding the case of λ = 2 and U∞ = 7. although allowing a coarser grid. for the scope of this paper. The scaling is done through 2 the dimensions associated with the airfoil movement 1 ρλU∞ c. Simulated flow conditions The simulation aimed at representing the flow conditions of the experimental work. Due to the importance of the induction of the rotor. the Laminar model simulation better predicts the development of the large leading edge separated vorticity and the rolling up of the counter-clockwise vorticity at the trailing edge. the tangential force is very dependent on the friction force. thus showing an effect on airfoils circulation of the vorticity generated in the downwind portion of the rotation. k − and S − A models The limits of the computational hardware and time frame of the computations do not allow the use of a space-time grid fine enough to calculate all scales of turbulence. placing the work in the unsteady aerodynamics region. the models assuming fully turbulent flow suppress the development of the leading edge separation and present a vorticity c The reduced frequency k is defined as k = ω.125. this result opens some questions regarding the suitability of integral lift force data for the validation of simulations. and incoming flows of U∞ = 7. 3. in the range of the difference found in several works where numerical simulations are compared with experimental results of forces or pressure measurements. The laminar flow also presents non-zero normal force at azimuthal angle θ = 0. due to change of grid refinement or turbulence model. in comparison with the turbulent flow simulations. diverging from the laminar model. due to the variation of V with rotation angle. The results from the current simulation in terms of flow field measurements are compared with the experimental flow field measurements using PIV of Sim˜o Ferreira et al. larger in the first quarter of rotation and lower in the second quarter of rotation. impacts the result of the simulation. took 2 − 3 revolutions to overpass. in the upwind half of the rotation. it is necessary to perform a simulation for several rotations until a long fully developed wake is present. presents an almost opposite sign of force. This transient solution took up to 7 revolutions from the initial stand-still situation to the fully developed wake. Verification and validation of Laminar. which for this experimental work was k = 0.ω. made 2 non-dimensional by the scale of the airfoil and its rotational velocity ( 1 ρλ2 U∞ c). Yet.10 Figure 2 shows the values for tangential and normal force over the first 180◦ degrees of rotation. The description of the models can be found in the Fluent c User’s Manual. The normal force CN d -force in the radial direction of the rotor. and thus. resulting in an incorrect turbulent flow development.is.c/(2. Verification of simulation results of an airfoil in a similar Darrieus motion was previously performed by Paraschivoiu8.

The results are the phase locked average over 10 rotations. This results shows the unsuitability of using a turbulent model over the entire flow for the simulation of dynamic stall in this Reynolds range. V. Error bar represents uncertainty at 95% confidence level. 4 C 3 CN C 2 T N laminar k−e SA 1 C T 0 −1 0 15 30 45 60 75 90 105 120 135 150 165 180 θ (degrees) Figure 2. Figure 3. The results of normal and tangential force show a significant sensitivity for the refinement of the time grid. the simulations show a divergence between the azimuthal positions of 60◦ and 160◦ . Verification and validation of time grid refinement On of the most important parameters in a time accurate simulation is the definition of the time step.distribution that does not fit the experimental data. This difference can once again be related to the accuracy of the simulation of the development of the leading edge separation vorticity and the consequent development of the trailing edge vorticity roll-up. k − and S − A models (first 180◦ ).5◦ . Figure 5 presents the value of normal and tangential force in the airfoil (first 180◦ ) for four different time steps. 5 of 11 American Institute of Aeronautics and Astronautics . Vorticity field around airfoil for three turbulence model. Comparison of normal and tangential force on airfoil calculated with Laminar. θ = 96. Each result is the phase locked average of 10 samples.

125°*Ω−1 3 CN CT 2 CT 1 0 −1 0 15 30 45 60 75 90 105 120 135 150 165 180 θ (degrees) Figure 5. There are two 6 of 11 American Institute of Aeronautics and Astronautics . Experimental results of the vorticity (Sim˜o Ferreira et al1 ).Figure 4. The anticipation/delay of flow phenomena is important to notice.∆ t = 0. for two simulations at different refinements of the time grid.1 a 5 CN ∆t=1 *Ω 4 ° −1 ∆t=0. paired by an earlier (in azimuthal position) concentration and rolling up of counter-clockwise vorticity at the trailing edge.125◦ ∗ Ω−1 (first 180◦ ). 5 and 5 samples. it is the divergence of azimuthal angle at which they occur that is the basis of error of the simulation. One of the properties of the phenomena of dynamic stall in a Darrieus is the occurrence of a strong random component. Comparing the simulations with the experimental observations of figure 4 and remaining results of Sim˜o Ferreira et al.1 it is possible to see that the simulation a with ∆ t = 1◦ ∗ Ω−1 presents an underestimation of the generation and evolution of the vorticity.25◦ ∗ Ω−1 .5°*Ω−1 ∆t=0.and ∆ t = 0. for the cases ∆ t = 1◦ ∗ Ω−1 . However. both the one detached at the leading edge and the one shed at the trailing edge.∆ t = 0. The results correspond to the phase locked average of 20. This underestimation is smaller for ∆ t = 0. Figures 8 and 9 present the phase locked average distributions of vorticity and three instantaneous distributions.25°*Ω−1 ∆t=0. a Figure 6 presents the phase locked average of the vorticity around the airfoil for four different time grid refinements at four different positions (in the 90◦ region). The value of circulation is calculated in the same procedure as used by Sim˜o Ferreira et al.25◦ ∗Ω−1 and ∆ t = 0. The results show that with increasing refinement of the time grid (∆t). an increasing generation of vorticity at the leading edge occurs.125◦ ∗Ω−1 present an overestimation of the shed vorticity.5◦ ∗ Ω−1 . Comparing with experiments. This trend can be observed in the value of circulation of the clockwise vorticity shed from the leading edge when compared with the experimental values (figure 7).5◦ ∗Ω−1 and the simulations with ∆ t = 0. Comparison of normal and tangential force on airfoil with Laminar model. 10. similar vorticity patterns to those observed in the four simulations occur in reality.

5 4.5 1.25°. 7 of 11 American Institute of Aeronautics and Astronautics .0 1.0 2.Figure 6. Vorticity field around airfoil for four simulation time steps.5 0.5 t=0. -1 -1 3. Circulation of the separated leading edge vorticity.5 t=1°. 5 and 5 samples.0 3. -1 c. U 2.0 60 75 90 105 120 135 (degrees) Experimental CFD simulation 150 Figure 7. Comparison of simulation values with experimental results.5°. 10. 5. after the 90◦ position (Laminar model).125°. t=0.0 0. The results correspond to the phase locked average of 20.0 4. -1 t=0.

Figure 9. Comparing the simulations with the phase locked averaged and instantaneous experimental measurements of figures 10 and 11 (neglecting the difference of azimuthal position). and the larger size of the counter-clockwise vorticity generated over the surface of the airfoil (past the first leading edge vortex) in the finer grid case. Although the results do not represent accurately all the properties of the real flow.5◦ . there is a over concentration of vorticity in two single large vortices. being passible of simulation through a 2D model (avoiding the expenses of a 3D representation). ∆ t = 0. θ = 96. as compared with the presence of only two large clockwise vortices in the case of a more refined time grid. Comparison of phase locked average with instantaneous sample. it is possible to see that in the coarser time grid there is an over fragmentation/spreading of the vorticity.5◦ ∗ Ω−1 .5◦ .aspects to compare between the two cases: the presence of more and smaller vortices relating to the leading edge separated vorticity in the case of lower refinement. Comparison of phase locked average with instantaneous sample. in the case of the finer time grid. they point to the possibility of the random component having a mainly 2D origin and thus. Figure 8.25◦ ∗ Ω−1 . ∆ t = 0. θ = 96. 8 of 11 American Institute of Aeronautics and Astronautics .

phase locked average. θ = 113◦ . θ = 113◦ . Vorticity field around airfoil (experimental PIV data). 9 of 11 American Institute of Aeronautics and Astronautics . instantaneous sample. Figure 11.Figure 10. Vorticity field around airfoil (experimental PIV data).

” Abstract submited for the 2007 AIAA/ASME Wind Energy Symposium. The forces for the coarse space grid follow the trend seen in the case of refinement of the time grid. which is an integral result of the aerodynamics).VI. thus aiming to improve the simulation of the development of the counter-clockwise vorticity shed at the trailing edge (resulting in a more correct development of the leading edge separation vorticity). Conclusions The results of the CFD simulations indicate the unsuitability of using a single turbulence model scheme for the numerical simulation of a 2D Darrieus in dynamic stall at low tip speed ratios.. Comparison of normal and tangential force on airfoil . The comparison for the simulated resulting forces for the reference and coarsened grid is shown in figure 12. Although the resulting flow field of the three turbulent models differ totally in the spatial distribution of vorticity. In this section we compare the results of the reference fine grid with a model in which the sub-grid Blade sub-grid I is coarsened by a factor of 4 in area (the distance between nodes in x an y are doubled). G. References 1 Sim˜o Ferreira. van Bussel. The use of 1equation/2-equation models that assume turbulent flow over the entire field results in a partial suppression of the magnitude of the leading edge laminar separation and in a reduction of maximum normal force.5◦ . since it gives direct insight into the the flow behavior (contrary to the total force. leads. G. reference fine and coarse grids. The results correspond to the phase locked average of 10 samples. 2007. although more suitable for the simulation of the leading edge separation. C. These force development are a result of the characteristics of vorticity development previously observed. VII. Further research will aim at developing numerical simulations with multiple turbulence models over the grid. the average tangential and normal forces over the upwind semi-rotation of the airfoil is similar. and van Kuik. to over prediction of the generation and evolution of the vorticity.at increasing refinement of the time grid. The experimental flow field data obtained using Particle Image Velocimetry proved to be a very useful method of validation of the rotor aerodynamics. but in this case in opposite direction. as previously seen in the case of refinement of the time grid. “2D PIV visualization of dynamic stall on a vertical axis wind a turbine.5◦ ∗Ω−1 . it is now necessary to view the implication of a coarsening of the spatial grid. 10 of 11 American Institute of Aeronautics and Astronautics . The Laminar model. Figure 13 compares the phase locked vorticity distribution of fine and coarse mesh at θ = 96. thus showing the difficulty of using experimental force/performance data for the validation at low λ. ∆ t = 0. Verification of the influence of spatial grid size Having verified the influence of the refinement of the grid in time. In the coarsened mesh the over estimation of the leading edge vorticity and early roll-up of the trailing edge vorticity occurs... 4 3 CN CT 2 CN 1 CT 0 Coarse grid Fine grid -1 0 15 30 45 60 75 90 105 120 135 150 165 180 (degrees) Figure 12.

van Bussel. 1988.. and Søresen.” AIAA/ASME Wind Energy Symposium.. Denmark. and van Kuik. S. S. C. 3 Sim˜o Ferreira.. 4. Measurements and Calculations of Dynamic Stall for a Similar Motion e of VAWT. 2 Mertens.. flow visualization and thrust a measurement of a VAWT in skew. 2001. 2003.. 7 Paraschivoiu. Polytechnic international Press.” Wind Energy for the New MillenniumProceedings of the European Wind Energy Conference.” Journal of Solar Energy Engineering. A.” e Journal of Solar Energy Engineering. C. Fluent Inc. 121.. 6 Allet. G. 2002. Vol. van Kuik..5◦ . 433–440.” Proceedings of the European Wind Energy Conference 2006 . I. pp. “Wind tunnel hotwire measurements. pp. and van Bussel. 125.. 1998.. “Visualization. G. G. 69–76. 10 Fluent User’s Manual. 9 Paraschivoiu. “Aerodynamic Analysis of the Darrieus Wind Turbines Including Dynamic-Stall Effects. Herning. pp. Vol.” Proceedings of the European Wind Energy Conference. Vol.. A. Wind Turbine Design .Figure 13. Denmark. 5 Hansen. van Bussel. and Allet. and B´guier. and Paraschivoiu. “CFD Model for Vertical Axis Wind Turbine. 8 Paraschivoiu. I. February 1999. No. “Numerical Simulation of Dynamic Stall Around an Airfoil in Darrieus Motion. and van Kuik.. Hall´. θ = 96. Comparison of the vorticity field around airfoil for coarse and fine grid.. D. 5. “An analytical method to predict the variation in performance of a an H-Darrieus in skewed flow and its experimental validation. C.. Copenhagen. I. 4 Sim˜o Ferreira. I. 2006.. G. 2006.” Journal of Propulsion and Power .. M.With Emphasis on Darrieus Concept. 11 of 11 American Institute of Aeronautics and Astronautics . “Performance of a H-Darrieus in the Skewed Flow on a Roof. 472–477.. G. G.

Sign up to vote on this title
UsefulNot useful