## Are you sure?

This action might not be possible to undo. Are you sure you want to continue?

**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

,µ

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.

- Ansys Mesh Introduction
- Fluent Tutorials 1
- CFD Study of a Wind Turbine Rotor
- CFD Analysis of a Horizontal Axis Wind Turbine (HAWT) Blade
- v6-57
- Wind Turbine Design
- Ansys Modeling and Meshing Guide
- Vertical Axis Wind Turbine
- Vertical Axis Wind Turbine
- VAWT Thermal Stresses Analysis
- Career Profile Handbook
- CFD Analysis of a Twisted Savonius Turbine
- Project Final
- Airfoil Lift 1
- AIEEE_Physics by DB Singh
- 41951343 Mathematical Olympiad Challenges Titu Andreescu Razvan Gelca
- 100 Winning Resumes for $100,000+ Jobs
- 77547452 Wind Turbine
- Java Programming Fundamentals
- Physics Kinematics
- An Introduction to Computational Fluid Dynamics - Versteeg
- Modelling and Numerical Simulation for Yacht Engineering
- Ency Turb
- Oral Training in Fluency Vocabulary
- CFD (Introduction)
- Fluent Tutorials - Cornell University
- Maths+quest
- Own Your Future Ideas Skills From Microsoft
- JP 2007 Blocken 1
- TF 93 Report

- tmpA063.tmp
- CFD Analysis of Convergent-Divergent Nozzle using Standard K-ÃŽâ€¢ Model
- tmp6BEE
- Tmp 2924
- An Overview of Vortex Tube on CFD with Straight and Helical Nozzle
- tmpCFAB
- tmpA422.tmp
- tmp7E6C.tmp
- Computational Study of Turbulent Flow around a Generic Car Body (Ahmed BODY)
- tmp734D
- tmpFAEF.tmp
- NASA Facts ACLAIM Airborn Coherant LIDAR for Advanced in-Flight Measurement
- Investigation On Computational Fluid Dynamics (CFD) Simulation for Wind Effects On Tall Tapered Buildings
- John J. Magan v. Lufthansa German Airlines, 339 F.3d 158, 2d Cir. (2003)
- Comparative Assessment of Aerodynamic Forces Generated Due to Wind Deflector of Passenger Car by CFD and Experimental Method
- 090610_mandelbrot

- Design Standards for Offshore Wind Farms-ABS
- On the Design and testing of Ocean Wave Energy Convertor.pdf
- Design Analysis and Optimisation of Mooring System for Floating Wind Turbines
- Example Technical Background Report_ Global Warming
- Guideline for the Certification of Wind Turbines Edition 2010 R0 2
- In Halt

Sign up to vote on this title

UsefulNot usefulClose Dialog## Are you sure?

This action might not be possible to undo. Are you sure you want to continue?

Close Dialog## This title now requires a credit

Use one of your book credits to continue reading from where you left off, or restart the preview.

Loading