turbines start-up
A. Rossetti
*
, G. Pavesi
University of Padua, Department of Mechanical Engineering, Via Venezia 1, 35131 Padova (PD), Italy
a r t i c l e i n f o
Article history:
Received 2 December 2011
Accepted 7 June 2012
Available online 16 July 2012
Keywords:
VAWT
Darrieus
Self-start
3D effects
CFD
BEM
a b s t r a c t
Self-start capability is an important feature of wind turbines. It allows to obtain simpler and cheaper
turbines not actively controlled. Different approaches to describe the self-start of an H-blade Darrieus
rotor are presented and compared in the present work. The Blade Element Momentum (BEM) approach
was compared with two and three-dimensional CFD simulations. The tipespeed ratio versus power
coefcient curves and the evolution of the trust forces over a blade revolution highlighted the limits and
the strengths of each approach.
The BEM model showed remarkable limits to describe to describe the self-start behaviour of the tested
geometry. The principal limits of the BEM approach can be ascribed to the absence of well documented
aerofoil databases for low Reynolds number and the inadequate modelling of dynamics effects. The 2D
simulation allowed to highlight the unsteady features of the ow elds, and the presence of a complex
vortices pattern which interact with the blade. Furthermore the comparison between 2D and 3D data
demonstrated the importance of 3D effects such as secondary ows and tip effects. These effects were
proved to have a positive effect on start-up, increasing the torque characteristic for tipespeed ratio of 1.
The start-up capability of H-Darrieus appears to be inuenced by many different factors, which include
secondary ows, three-dimensional aerodynamic effects and the nite aspect-ratio of the blades.
2012 Elsevier Ltd. All rights reserved.
1. Introduction
The increasing energy cost and the presence of support policy
towards renewable energy are feeding the interest to develop small
power wind turbines suitable for urban and suburban applications.
The soft sound emissions, due to the lower optimum tipespeed
ratios compared with the three blades HAWT, the insensitivity to
the wind direction, and the lack of need of passive or active yawl
control are attractive aspects. Moreover, the easy accessibility of
the electrical and mechanical part of the VAWT decreases the
maintenance costs of the turbine. Notwithstanding, the higher
performances of the HAWT and the more complex structural
design of the VAWT, the Darrieus turbines presents many
advantages when small applications in urban agglomerations are
considered.
The ability of self-start is an important quality/property for
small turbines, to obtain simpler and cheaper actively controlled
turbines. The need of the turbine assisted launch causes energy
expense, could impact the turbine overall production and increases
the control system complexity [1]. The self-start capability of xed
pitch Darrieus turbine is not completely documented yet and
conicting data and conclusions can be found. Baker [2] identied
the presence of a negative torque band, on a l range from 0.5 to 2,
depending on the blade geometry consistently with the experi-
mental data reported in literature [3]. Baker suggested the use of
inclined or wrapped blades and a series of desirable aerofoil section
feature to aid self-starting. Other authors [4,5] have pointed out the
effect of the blade Reynolds number on the self-start behaviour,
demonstrating the positive effects of high Re numbers. This has
encouraged the construction of turbines with long chord blades
that is with high solidity. The capability of high solidity Darrieus to
self-start is consistent with some producers claims [6], but the
excessive increase in solidity leads to the decrease of the maximum
power coefcient. For the Troposkien blade turbine, the optimum
solidity was calculated by Strickland to be about 0.3, on the base of
Blade Element Momentum analyses [7] and conrmed by wind
tunnel experimental tests conducted by Sandia National Labora-
tories [8]. Recent papers demonstrated the capability of self-
starting of an H-blade, low solidity Darrieus in open section wind
tunnel at low Re [9,10]. The papers highlighted the difculty of
predicting self-start using analytical models, due to the uncertainly
of aerodynamic database at low Re and to the presence of complex
* Corresponding author. Tel.: 39 049 827 7474; fax: 39 049 827 6785.
E-mail address: antonio.rossetti@unipd.it (A. Rossetti).
Contents lists available at SciVerse ScienceDirect
Renewable Energy
j ournal homepage: www. el sevi er. com/ l ocat e/ renene
0960-1481/$ e see front matter 2012 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.renene.2012.06.025
Renewable Energy 50 (2013) 7e19
and unsteady phenomena at low tipespeed ratios [1,11]. In the
same way, the studies on the reduced aspect-ratio effects carried
out contradictory conclusions. Musgrove and Mays [5] suggested
a positive effect whereas Kirke [12] presented opposite argumen-
tation based on theoretical considerations. Despite the apparent
discrepancy between the cited papers, all concur to suggest that the
xed blade Darrieus could self-start but only under some circum-
stances and with suitable geometry.
In this paper, different approaches to describe the self-start of
an H-blade Darrieus rotor are presented and compared. The
start-up characteristics of the turbine are computed using both
a Blade Element Momentum (BEM) model and 2D and 3D CFD
analyses. The 3D analyses are compared to 2D approach to
identify the effects of nite aspect-ratio on start-up process, to
highlight the strong 3D effects on start-up mechanics and the
inadequacy of analytical models to predict self-start worsened
by the absence of accurate aerodynamic database for low Rey-
nolds numbers.
2. Self-starting
Two different approaches can be distinguished in the self-
starting denition. The rst one denes the self-starting deni-
tion on the basis of the output power delivered by the turbine.
Ebert and Wood considered the starting process completed when
usable power extraction starts up [1]. In a similar way, Kirke
dened the self-starting feature as the capability to accelerate from
not running to the point at which a useful output is produced [12].
The second approach denes the starting as a function of the
tipespeed ratio (TSR) l. Lunt identied the self-starting feature as
the turbine capability to reach a TSR greater than 1 [13]. This
denition is based on the assumption that to reach TSR values
greater than 1 the turbine has to overcome the drag-driven motion
and enter into the lift-driven motion. These denitions are limited.
The rst approach suffers from the absence of a precise threshold
for the output power. Moreover, the output power depends on the
wind velocity. On the basis of the given denition, every machine
that allows a drag-driven motion can be dened as self-starting
under high wind. As opposed to the second denition, which
assures the presence of a lift-driven behaviour, but does not assure
any energy production.
For the present work a restrictive formulation was considered.
The self-start capability of the turbine was dened as the ability to
reach the escape velocity, starting from rest, under constant wind
speed when no power extraction occurs in the start-up process.
This denition is compatible with the previous denitions. The
request to accelerate the turbine up to the escape velocity assures
the presence of a net power production because the turbine over-
steps the TSR with the maximum Cp and consequently the turbine
gets to l greater than 1. Moreover the dened start-up capability is
explicitly related to the wind velocity and to the inuence of the
Reynolds number on the start-up [4,5,14] and consistently allows to
dene the minimum wind velocity at which the start-up takes
place.
3. Reference turbine
The reference turbine is an H-Darrieus 3-bladed rotor [10] with
the main geometrical characteristics listed in Table 1. According to
Hill, this Darrieus geometry was proved to be able to self-start with
a constant wind of 6 [m/s] when no electrical load is applied to the
rotor. The turbine tipespeed ratio verses time from rest to the
equilibrium velocity is reported in Fig. 1 [10].
The velocity trend shows a sequential behaviour. After quite
a linear acceleration, due to the drag-driven motion of the turbine,
the rotor speed shows a relatively long plateau period before it is
accelerate by the lift-driven motion and reaches the equilibrium
condition. The small inertia of the rotor and the persistence in the
plateau zone allows to exclude the inertial effects to help overcome
the dead zone. The experimental data are compatible both with
a slightly positive torque for l in the range between 1.2 and 1.3 and
with the presence of a negative torque band in the same range.
In eld test, the rise to overcome the dead zone could be
reasonably explained by wind instability according to Baker [2], and
a temporary drop in wind intensity results in an increase of l to
a value outside the negative torque region. Despite signicant
uctuation observed in eld test, the wind tunnel test conditions
are carried out with marginal wind variation. According to this
observation, only a small dead band in the range of l 1.2e1.3 could
match with the experimental data.
4. BEM-software
The Blade Element Momentum (BEM) model, originally pre-
sented by Templin [15] and developed by Strickland [7], adapted
the actuator disk approach to predict the performance of a Darrieus
turbine. The model was improved over the years by many authors
[16e19], to take into account dynamic stall and secondary effects.
The BEM-software used for this paper was implemented in Matlab,
according to the Double-Disk Multiple stream tubes presented by
Paraschivoiu. Moreover, the basic model was improved using
appropriate sub models to copy dynamic behaviour and secondary
effects:
- Momentum equation: The original momentum equation
resulting from the axial equilibrium of the turbine is inappro-
priate for turbulent wake. The empirical momentum equation
proposed by Glauert [17,20] was implemented to overcome the
original limits.
Table 1
Reference geometry.
Aerofoil NACA 0018
Chord 0.083 [m]
Span 0.600 [m]
Diameter 0.750 [m]
Moment of inertia 0.018 [kg m
2
]
Turbine Ar. 0.80 []
Blade Ar. 7.22 []
Fig. 1. Start-up characteristic of the reference rotor under constant wind of 6 [m/s].
A. Rossetti, G. Pavesi / Renewable Energy 50 (2013) 7e19 8
- Dynamic stall model: The dynamic stall was modelled according
to Mass and Berg [21].
- Stream tube expansion model was implemented to improve the
software accuracy [18,19].
The aerodynamic proles were characterized for different Rey-
nolds numbers in the range from 40,000 to 1e7 and for an angle of
attach from 0
to 180
/u
were employed. Results of the sensitivity analyses encouraged the
use of the nest mesh for 2D simulation, while 100.000 nodes per
layer were assumed to be a good compromise between accuracy
and computational time for the 3D case.
The mesh structure and renements are presented in Fig. 5. The
grid size and elements dimensions are reported in Tables 3 and 4
for the different domains.
The inlet condition was set on the upwind face of the stator
domain, imposing a constant wind prole equal to 6 [m/s] with
stochastic turbulence of 5% of the free stream velocity. The static
atmospheric pressure was imposed on the remaining surfaces of
the model. Symmetry was employed on both the lower and the
upper faces for the 2D simulation, while for the 3D analyses it was
substituted on the upper part with a static pressure condition.
Table 2
Benchmark geometry for the BEM-software.
Aerofoil NACA 0015
Chord 0.152 [m]
Span 5.10 [m]
Diameter 5.00 [m]
Fig. 2. Comparison between the eld data and BEM model for the Sandia 5 m turbine
for constant speed of 150 [rpm].
Fig. 3. 3D domain dimension.
Fig. 4. Sensitivity study results.
A. Rossetti, G. Pavesi / Renewable Energy 50 (2013) 7e19 9
Time discretization was dened according to a sensitivity study
presented in Fig. 4. A time step dt corresponding to a 2
[deg] of the
rotor revolution was employed for all the analyses. The analyses
were carried out for a total time corresponding to at least 8
complete revolutions of the rotor, to gain the independence of the
solution from the initial condition. Convergence to the periodic
behaviour of the system was monitored comparing the mean
power coefcient between two subsequent revolutions. For all the
considered cases, this parameter was kept below the 3% of the
mean Cp.
The air was modelled as constant propriety gas at 25 [
C]. A
specic attention was given to the turbulence model. The SAS
model was preferred to more classical turbulence models. This
model results in a compromise between LES and RANS formulation,
providing LES-like behaviour in unsteady regions of the ow eld
and standard RANS capabilities in stable ow regions [25]. In order
to further improve the near wall ow model accuracy, the
laminareturbulent transition were modelled employing the two
equation model developed by Langtry and Menter [26,27].
6. Results
6.1. BEM model
The torque characteristic of the turbine calculated using the
BEM model is plotted in Fig. 6.
Fig. 5. Mesh structure; a) 2D analyses computational domain; b) 2D rotor detail; c) 2D O-grid structure detail; d) 3D mesh on blade tip.
Table 3
elements dimension.
2D case 3D case
Stator
Maximum element 80 [mm] 80 [mm]
Elements on interface and wake 8 [mm] 12 [mm]
Rotor
Maximum element 8 [mm] 12 [mm]
Maximum element on blade 100 [mm] 150 [mm]
Maximum height on blade 40 [mm] 40 [mm]
Expansion ratio 1.1 1.15
Table 4
Mesh size per layer.
Stator
2D model 29,788 nodes
3D model 1,883,780 nodes
Rotor
2D model 434,508 nodes
3D model 6,986,796 nodes
Fig. 6. Tipespeed ratio curve predicted by the BEM model.
A. Rossetti, G. Pavesi / Renewable Energy 50 (2013) 7e19 10
The results suggest that the model drastically overestimated the
dead band of the turbine. The negative torque band extends from
l 0.2 to l 3. Similar results were obtained by Hill [10], which
reports two distinct negative torque bands between 1.2 and 3 for
a 0012 prole.
Fig. 7. Lift and trust forces calculated on the overall blade revolution; (a) l 1; (b) l 2.
Fig. 8. Lift characteristic of the NACA 0018 prole as described by different authors; a) lift coefcient versus incidence curves; b) characteristic points of the curves as function of the
Reynolds number.
Fig. 9. Modied aerodynamic database according to the ViternaeCorrigan correlation.
Fig. 10. Tipespeed ratio versus torque curve as predicted by the BEM model using
unmodied and modied aerodynamic data.
A. Rossetti, G. Pavesi / Renewable Energy 50 (2013) 7e19 11
The torque coefcient and the lift coefcient versus the blade
position over the overall revolution are shown in Fig. 7. The nega-
tive torque is caused by the sudden decrease of the lift in the after-
stall region [2], till the increase of the tipespeed ratio decreases the
angle of attack below the stall incidence (l > 3.5).
The drop of the lift in the after-stall region as reported by
Sheldahl and Klimas is, however, in contradiction with other
databases available in literature. Fig. 8 compares the lift coefcient
characteristic of the NACA0018 prole for Reynolds number in the
interval 80.000e150.000.
Sheldahl and Klimas data report both the lower maximum lift
and the higher lift loss in the after-stall region. The after-stall lift
drop is in accordance with recent data published by Timmer [28],
but the data appear to overestimate the maximum lift. The data
proposed by Jacobs and Sherman [29] present values of maximum
lift intermediate between the other two series of data. On the other
hand, the after-stall characteristic is strongly different and it shows
a smoother transition to the full stalled zone.
Until Timmer, and Jacobs and Shermans data could not be used
effectively for the present work due to the absence of low Reynolds
series [Re < 1e5]. To evaluate the inuence of the post-stall region
on the BEM results, the analyses were repeated using a database
derived from the Sheldahl and Klimas data, assuming the post-stall
Fig. 12. Start-up characteristics expected using tipespeed ratio versus torque curve
of Fig. 10.
Fig. 13. Tipespeed ratio versus torque curve. 2D CFD and BEM results. Fig. 14. Trust force comparison between CFD and BEM for l 1.
Fig. 11. Lift and Trust forces calculated over a whole blade revolution using modied aerodynamic data; (a) l 1; (b) l 2.
A. Rossetti, G. Pavesi / Renewable Energy 50 (2013) 7e19 12
Fig. 15. Blade loading charts for different blade positions.
A. Rossetti, G. Pavesi / Renewable Energy 50 (2013) 7e19 13
characteristic is in accordance with the model proposed by Viterna
and Corrigan [30]. The blend between the original data and the
post-stall model were arranged to obtain a post-stall characteristic
similar to the one observed in Jacobs and Sherman data. The
aerodynamic data obtained are compared to the original data in
Fig. 9.The torque curve obtained with the modied database is
reported in Fig. 10.
The local modication of the aerodynamic data had strong
impact on the torque curve. The torque characteristic obtained with
the modied data presents always positive torque except for l 1
Fig. 16. Adimensional vorticity development for step of 20 [
higher.
This is explainable considering the velocity component orthogonal
to the undisturbed wind due to the vorticity induced by the wind
turbine and the expansion of the ow streamlines (Fig. 17) which
are computed in CFD model while they are not accounted accu-
rately in the BEM approach. The deection of the undisturbed
wind reduces the effective incidence on the blades for small q
values and causes the delay of the trust ramp in CFD model.
Consistently, the blade loading chart highlights the presence of
a small negative incidence when q 0
later than
the BEM model (Fig. 16). This increases the forces in a
0
and
decreases the trust in b
0
where stall occurs. The different blade stall
characteristic is directly related to the dynamic effects, which are
dominant at low tipespeed ratios [11,30] and are not adequately
considered in the BEM.
The after-stall region up to q 180
e
q 100
sin q
l* cos q
(3)
Fig. 22. Trust force on the six elements of the blade; a) sections S1 (tip) and S2; b) sections S3 and S4; c) section S5 and S6 (symmetry).
Fig. 23. Mean trust on a revolution for the six elements of the blade.
A. Rossetti, G. Pavesi / Renewable Energy 50 (2013) 7e19 17
The incidence reduction could be appreciated in Fig. 24, noticing
that for q 80