Вы находитесь на странице: 1из 13

Renewable Energy 117 (2018) 287e299

Contents lists available at ScienceDirect

Renewable Energy
journal homepage: www.elsevier.com/locate/renene

Shape optimization of a Savonius wind rotor with different convex


and concave sides
Wenlong Tian a, b, *, Zhaoyong Mao a, b, Baoshou Zhang a, Yanjun Li c
a
School of Marine Science and Technology, Northwestern Polytechnical University, Xi'an 710072, China
b
Key Laboratory for Unmanned Underwater Vehicle, Northwestern Polytechnical University, Xi'an 710072, China
c
Harbor Branch Oceanographic Institute, Florida Atlantic University, Fort Pierce, FL 34946, USA

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

Article history: This paper introduces an optimization procedure of a modified Savonius rotor with different convex and
Received 1 April 2017 concave sides to maximize the power efficiency. A series of transient computational fluid dynamics (CFD)
Received in revised form simulations are performed to find the peak coefficients of power (Cp) of each blade geometry. Then, a
18 September 2017
global response surface model is created according to the Kriging Method, which defines the relationship
Accepted 23 October 2017
Available online 27 October 2017
between optimization objective Cp and the design parameters. A Particle Swarm Optimization (PSO)
algorithm is applied to finding the optimal design based on the response surface model. The optimal Cp is
0.2580 and is 4.41% higher than the traditional design. Comprehensive comparisons of torque, power and
Keywords:
Savonius
flow structures between the optimal and the traditional designs are performed to illustrate the mech-
Optimization anism of how the blade shapes improve the rotor performance. It is find found that the optimal blade
Blade shape has stronger tip vortices and recovery flows, which contributes to an increase in the performance
CFD of the rotor.
Kriging © 2017 Elsevier Ltd. All rights reserved.
Particle Swarm Optimization

1. Introduction speed and noise emission [4]; (3) high starting torque, good start-
ing performance [5]; and (4) simple structure and low cost.
Due to the increasing oil price and environmental pollution However, the traditional Savonius turbine has relatively lower
considerations, wind energy is attracting increasing attention power efficiency than most lift-type wind turbines, such as HAWTs
comparing with the oil-based power generation [1]. Wind turbines and Darrieus wind turbines. The performance of Savonius turbines
are the devices used for converting wind energy into electricity. is affected by blade shape, gap ratio, overlap ratio, blades number,
Wind turbines are generally divided into two categories, horizontal end plates, stages, guide plates and other accessories [6e8]. In
axis wind turbines (HAWTs) and vertical axis wind turbines previous studies, the efficiency of the Savonius turbine has been
(VAWTs), based on the relative alignment between their axis of increased by using different optimal manners. In the aspect of using
rotation and the wind direction. VAWTs are often favored for small- deflectors, Wahyudi, B. et al. and Kumar, D. et al. [9,10] used
scale power generation because their performance is not depended obstacle plates to optimize the design of Savonius blade turbine.
on the relative wind direction, allowing generation equipment to Similarly, El-Askary, W. A. et al. [11] added a series of guide plates to
be located on the ground shaft [2], resulting in reduced mainte- harvest the incoming wind to generate a wind jet to the concave
nance costs. The Savonius wind turbine is a common type of VAWT side of the advanced blade and prevent the convex side of return
that generates torque through the combined effects of drag and blade from coming upwind stream. As a result, one of them
side forces. These turbines are typically composed of two or three enhanced the performance to reach a power-coefficient peak of
arc-type blades, as shown in Fig. 1. Savonius turbines have the 0.52 with an operation range of tip speed ratio. As the blade is the
following advantages over other types of wind turbines: (1) ability most important part of a Savonius turbine, much more efforts are
to operate under complex turbulent flows [3]; (2) low rotation devoted to the optimization of the blade shape. Mohamed, M. H.
et al. [12] considered 6 free parameters in an optimization process
to increase the efficiency of the Savonius turbine. Driss, Z et al. [13]
* Corresponding author. School of Marine Science and Technology, Northwestern used 12 parameters to define a Savonius turbine and got a segment
Polytechnical University, Xi'an 710072, China. of a circular arc with the blade arc angle j ¼ 120 as the blade.
E-mail address: tianwenlong@mail.nwpu.edu.cn (W. Tian).

https://doi.org/10.1016/j.renene.2017.10.067
0960-1481/© 2017 Elsevier Ltd. All rights reserved.
288 W. Tian et al. / Renewable Energy 117 (2018) 287e299

and a2, which stands for the height of each blade side. The per-
Nomenclature formance of each geometric design is then evaluated by CFD
method. The accuracy of CFD method is proved by similar research
D Diameter of the rotor (m) conducted in previous studies [17e19]. An optimization study
d Blade chord length (m) based on the Kriging surrogate model [20] and the Particle Swarm
S0 Gap between the two blades (m) Optimization (PSO) [21] is performed to find the optimal design.
h1 Height of the convex side (m) Kriging Method is an unbiased estimation model which establishes
h2 Height of the concave side (m) the relationship (response surface model) between optimization
a1 Dimensionless height of the convex side objective and design parameters. PSO algorithm is then used to find
a2 Dimensionless height of the concave side the optimal design parameters for the blade based on the surrogate
H Height of the rotor (m) model. The PSO is inspired by the natural swarming behavior of
x Radial location on the blade (m) bird flocking and simple in concept, easy to implement and
P Power (W) computationally inexpensive [22].
Cp Coefficient of power
M Torque (Nm) 2. Geometry configuration
Cm Coefficient of torque
v∞ Upstream velocity (m/s) 2.1. Parameter definition
u Angular velocity (rad/s)
r Water density (kg/m3) In this paper, the modified Savonius blade has its sides gener-
TSR Tip speed ratio ated from a semi-ellipse. Typically, when the two axes of the ellipse
have the same length, the modified blade represents a classical
semi-circle blade which is commonly used in Savonius rotors. Fig. 2
(a) gives a solid view of a typical two-bladed counterclockwise-
Ahmed, M. R. et al. [14] suggested the optimal blade curvature for rotating Savonius wind rotor with endplates. It should be noted
maximum energy extraction was found to be 70 (blade arc angle). that the rotation direction of the rotor only influences the sym-
Al-Faruk, A. et al. [15] studied the blade overlap ratio, the blade arc metry of the flow structures and has no effect on the forces and
angle and the higher aspect ratio at the same time. The perfor- power of the rotor. There are several parameters for the rotor
mance comparison of optimum configuration with conventional description shown Fig. 2 (b) and (c), where D is the diameter of the
Savonius rotor showed an increase of 24.12% in the coefficient of rotor, H is the height of the rotor, S0 is the gap between the two
power. Abdullah Al-Faruk suggested power coefficients increase blades, d is the blade chord length. The shape of the blade is
with the increase of blade arc angle up to a certain optimal value of generated from a half ellipse with h1 and h2 being the height of the
195 which is 29% higher than the conventional blade arc angle of outer (convex) and inner (concave) sides, respectively. Therefore, a
180 , which is quite different from the conclusions of other classical semi-circle Savonius blade can be described as h1 ¼ h2 ¼ d/
scholars. Kacprzak K. et al. [16] suggested the optimum blade was 2.
composed of a segment of a circular arc and a short straight line We can consider the rotor with h1 ¼ h2 as the Classical Savonius
segment. rotor, and the others are Modified Savonius rotors. Unlike Classical
From the brief literature review, it is evident that researchers Savonius rotor, the Modified Savonius rotors have different sides
have done extensive work to optimize the shape of the Savonius with h1  h2. As this paper deals with the blade shape optimization,
blade. However, the optimizations are all performed on ‘thin’ two parameters, h1 and h2, will be considered and optimized. The
blades which have the same convex and concave sides. The concave other parameters, such as D, S0 and d, will be treated as constant.
side of the blade will affect the flow inside the rotor. Therefore, it is Hence, the overall size of the Modified Savonius is set equal to a
necessary to investigate the blades with different convex and Classical Savonius experimentally studied before [23]. To eliminate
concave sides. This paper deals with the optimization of a modified the dimensional effects, h1 and h2 are normalized by dividing the
Savonius blade with different convex and concave surface shapes, blade chord length (Eq. (1)). The main geometrical parameters of
which are generated from a series of semi-ellipses. The shape of the the Classical Savonius and the Modified Savonius are listed in
modified blade is defined by two non-dimensional parameters, a1 Table 1.

hi
ai ¼ ; i ¼ 1; 2 (1)
d
End plate
A series of blades can be described by the two design parame-
ters, a1 and a2. Before the optimization procedure, 68 groups of CFD
simulations are performed as sample points with a1 and a2 varying
Overlap ratio
from 0.2000 to 0.5500. The detailed values of the two parameters
are listed in Table 2. Among all the cases, the C5 rotor is the Classical
Stages Savonius turbine with the same dimensions with that tested in the
Single blade wind tunnel [23] and will be used for the model validation and the
evaluation criteria of the performance of the modified rotors.
Gap ratio
2.2. Coefficients of performance
Blade arc
angle Tip speed ratio (TSR) means the ratio between rotor blade tip
speed and inflow airspeed and is defined as follows:

TSR ¼ uD=ð2v∞ Þ (2)


Fig. 1. Some key terms of a Savonius rotor.
W. Tian et al. / Renewable Energy 117 (2018) 287e299 289

Fig. 2. Solid view of A Savonius turbine (a) and Simplified 2D model for Classical Savonius(b) and Modified Savonius(c).

Table 1 parameters (a1 and a2). Finally, PSO method is used to find the
Main geometrical parameters of the Savonius rotors. optimal Cp together with the values for the corresponding config-
D (m) H (m) d (m) S0 (m) a1 a2 uration parameters based on the surrogate model.
Classical Savonius 0.909 1.0 0.5 0.091 0.5 0.5
A series of transient, 2D representation of the flow field is built
Modified Savonius 0.909 1.0 0.5 0.091 variation variation to ensure the accuracy of calculation and improve the efficiency of
the optimization process. Hence, the flow in the vertical direction is
neglected, which is equivalent to 3D Savonius rotors with end-
Performance of Savonius wind turbines can be characterized by plates. The 2D simulations have been successfully proved with high
the manner in which the two leading indicators: torque coefficient accuracy in similar studies using Ansys Fluent [12,16,18,24]. Further,
(Cm) and power coefficient (Cp) varying with TSR. Cp represents the the results of this numerical model will be compared with existing
fraction of extracted power from the total power of air flow which experimental data for the model validation, which will be discussed
runs through the projected area of turbines in the flow direction. Cm in Section 3.2. Due to the low velocity (7 m/s) flow condition [23],
and Cp can be obtained as follows: basic assumptions are presented:
. 
1) Temperature and density variation of the flow is negligible.
Cm ¼ 4M rHv2∞ D2 (3)
2) The relative velocity is assumed to be a vector with constant
.  .  direction.
Cp ¼ P 0:5rHDv3∞ ¼ 2M u rHDv3∞ ¼ Cm $TSR (4) 3) The resistance of the rotor shaft is not included.

where M represents torque and P is power. Due to the two-


dimensional assumption, the unit height H ¼ 1 m is used. The 3.1. Computational domains and grid generation
power coefficient Cp denotes the energy capture capability of a
rotor and is then set as the optimal objective in the optimization The overall domain is split into two subdomains to facilitate
code. simulating the rotating of the rotor (Fig. 3). The highest resolution
subdomain contains the grid elements surrounding the rotor
(referred to as the rotating domain). The second low resolution
3. Numerical method subdomain contains the cells in the outer region (referred to as the
stationary domain). The boundary conditions include:
In the optimization process, CFD method using Ansys Fluent
provides foundation elements for the modeling of the response 1) Inlet: Uniform and constant velocity of 7 m/s.
surface surrogate model obtained by the Kriging method. The 2) Outlet: Pressure outlet with a relative atmospheric pressure of
aerodynamics (Cm and Cp) and the flow structures for each simu- 0 Pa.
lation case are obtained by the CFD approaches separately. For each 3) Interface: The overlapped edges between the rotating domain
rotor listed in Table 2, simulations are performed at several TSRs to and the stationary domain are set as interfaces.
find its peak Cp. The results are then used to build the relation 4) No slip wall: No slip wall conditions are imposed at the surfaces
model (Kriging surrogate model) between the Cp and the design of the blades.

Table 2
Design parameters.

Simulation cases a1 a2

C1~ C4 0.5500 0.5500, 0.5000, 0.4500, 0.4000


C5~C15 0.5000 0.5000, 0.4750, 0.4500, 0.4250, 0.4000, 0.3750, 0.3500, 0.3250, 0.3000, 0.2750, 0.2500
C16~C24 0.4500 0.4500, 0.4250, 0.4000, 0.3750, 0.3500, 0.3250, 0.3000, 0.2750, 0.2500
C25~C33 0.4250 0.4250, 0.4000, 0.3750, 0.3500, 0.3250, 0.3000, 0.2750, 0. 2500, 0.2250
C34~C42 0.4000 0.4000, 0.3750, 0.3500, 0.3250, 0.3000, 0.2750, 0. 2500, 0.2250, 0.2000
C43~C50 0.3750 0.3750, 0.3500, 0.3250, 0.3000, 0.2750, 0.2500, 0.2250, 0.2000
C51~C56 0.3500 0.3500, 0.3250, 0.3000, 0.2750, 0.2500, 0.2250
C57~C61 0.3250 0.3250, 0.3000, 0.2750, 0.2500, 0.2250
C62~C65 0.3000 0.3000, 0.2750, 0. 2500, 0.2250
C66~C68 0.2750 0.2750, 0.2500, 0.2250
290 W. Tian et al. / Renewable Energy 117 (2018) 287e299

Wall

Central axis
Blade 2

Velocity-inlet
Blade 1

Pressure-outlet
o
Rotating
Interface
Rotating Domain

Stationary Domain
Wall

Fig. 3. Simulation domains.

different contours at different rotor angle. A sliding mesh method is


used to simulate the rotation of rotor. The rotating domain moved
exactly 1.0 in each simulation step. Four type relative rotation
positions and details of the local mesh are shown in Fig. 4.

3.2. Numerical method validation

Numerous simulations are carried out to optimize the perfor-


mance of the modified Savonius rotor. Each simulation lasts for 10
rotation periods, and the coefficients in the last cycle are used for
the analysis. Typically, simulation for the Classical Savonius at
TSR ¼ 1.0 is selected for the convergence analysis. The Cm of the
blades and the rotor in 10 rotation periods is plotted against rota-
tion angle in Fig. 5. It can be seen that the Cm curves get converged
after 5 rotation periods. The relative difference of the averaged Cm
Fig. 4. Relative rotation positions around the blades at different rotor angle.
in the following rotation cycles is less than 0.5%. Therefore, it is
reasonable to take the coefficients in the 10th cycle to predict the
5) Sliding wall: Sliding wall conditions are imposed at both the top performance of the rotor.
and the bottom edges. A mesh independence validation study and a time step inde-
pendence study are conducted. The verification simulations are
The Savonius rotor rotates around the central axis and exhibits performed using the Classical Savonius running at TSR ¼ 1.0. The
Shear Stress Transport (SST) k-u turbulence model is used in the

Fig. 5. The torque coefficient during 10 rotation periods at TSR ¼ 0.9 in the process of CFD simulation.
W. Tian et al. / Renewable Energy 117 (2018) 287e299 291

0.8 0.8
61000 0.5°
86900 1.0°
0.6 117000 0.6 2.0°

Torque coefficient of a single blade


Torque coefficient of a single blade

4.0°

0.4 0.4

0.2 0.2

0 0

-0.2 -0.2

-0.4 -0.4
0 50 100 150 200 250 300 350 0 50 100 150 200 250 300 350
Rotor angle (degree) Rotor angle (degree)
(a) mesh density verification (b) time step verification
Fig. 6. Mesh density verification and time step verification.

Table 3 steps ranging from 0.5 /step to 4 /step. The instantaneous Cm and
Average Cp for different densities and time steps. the average Cp are then evaluated, as shown in Fig. 6(b) and Table 3.
Grid Number Average Cp Time step Average Cp Both the Cm and Cp have the good consistency for the four-time
61000 0.2471 0.5 0.2472
steps. In consideration of the efficiency of the CFD simulation, a
86900 0.2482 1.0 0.2471 time step of 1.0 /step is selected for subsequent studies.
117000 0.2468 2.0 0.2478 To further verify the accuracy of the CFD method in this paper,
e e 4.0 0.2483 the averaged Cm of the rotor is compared with the wind tunnel data
[23]. Fig. 7 suggests that the simulation results are in good agree-
ment with the experimental results [23]. The CFD method slightly
simulation. The mesh independence is tested by assessing the overestimates the results when TSR > 1. The maximum relative
instantaneous torque of a single blade for different mesh densities. error between the numerical and the experimental data is 3.4%
The instantaneous Cm is presented in Fig. 6(a) and the averaged Cp is occurred at TSR ¼ 1. Therefore, it is acceptable to use the proposed
listed in Table 3. Both the Cm and Cp have the good consistency for CFD method to predict the performance of Savonius rotors.
different mesh densities (61,000/86900/117000 elements).
Accordingly, the mesh with approximately 60,000 elements can 4. Optimal design method
predict the performance of the rotor with sufficient accuracy and
will be used in the later simulations. Particle swarm optimization (PSO) is a method to search for the
The time step independence is verified using different time optimal value of a surrogate model. In this paper, the surrogate
model is proposed based on the Kriging Method, which establishes

([SHLUPHQWV>@
Experimental Data
0.4 Simulation Data

0.35
Average torque coefficient

0.3

0.25

0.2

0.15
0.4 0.5 0.6 0.7 0.8 0.9 1 1.1 1.2 1.3 1.4
Tip speed ratio

Fig. 7. Comparison of the averaged Cm at different TSRs.


292 W. Tian et al. / Renewable Energy 117 (2018) 287e299

Start

Initialize particles for each particle

Numerical method using


Ansys Fluent provides
foundation elements for Calculate the fitness value according to the
modeling of the response response surface surrogate model ( established
surface surrogate model by using Kriging Method)
( established by using
Kriging Method)
Set the particle with the best fitness value as
personal optimal solution.

Choose the particle with the best fitness value


Update particle fitness value for each
of all the particles as the global optimal
particle
solution.

No
The maximum number of iterations is attained Calculate particle speed for each particle

Yes

End

Fig. 8. The general procedure of PSO.

relation model between optimization objective Cp and design pa- easy understanding, the power is presented in the form of a ratio
rameters, a1 and a2. The CFD method proposed in Section 3 pro- between the modified Savonius and the Classical Savonius (0.2471).
vides data for the sample elements of the response surface The optimal point was marked in red in Fig. 9. Fig. 9 suggests that
surrogate model. Finally, PSO is used to find the maximum value of there exists two main regions where the Cp of the rotor is improved,
Cp and the corresponding design parameters of the blade. The which are referred to as the Weak Improved Region (WIR) and the
detailed Kriging surrogate model and the PSO algorithm can be Highly Improved Region (HIR), respectively. The WIR locates near
found in the Appendix. (0.4500, 0.4500) where the blade has nearly the same sides. The
In this paper, the PSO is implemented in a Matlab code in peak value in the WIR is about 1.03. The HIR locates near (0.4000,
conjunction with Kriging model according to the flow diagram 0.2750), where the blade is flatter than a classical semi-circle blade.
Fig. 8. The procedure outputs are optimal value Cp. Then, two var- The optimal global value of the power ratio is 1.0449, and the co-
iables representing design parameters (a1 and a2) of the turbine are ordinate values of the optimal point represent the design param-
translated from Kriging model. Hence, with the help of PSO, the eters (a1 ¼ 0.3936 and a2 ¼ 0.2743). The optimal design will be
optimal design parameters can be found. referred to as Optimal Savonius in the following contents. There-
By adding the results of the 68 simulation groups as the sample fore, the PSO suggests an Optimal Savonius with a Cp increment of
points in the optimization region, a Kriging response surface model 4.49%. The characteristic of the Optimal Savonius rotor will be
is established, as shown in Fig. 9(a). After an iterative search in the investigated in detail in the following section.
optimization region, the optimal point was obtained using PSO. For

Fig. 9. Response surface and optimal point.


W. Tian et al. / Renewable Energy 117 (2018) 287e299 293

5.3 and 5.4.


After investigating the variation of the average Cm and Cp with
TSRs, the Optimal Savonius turbine exhibits superior characteristics
at most of the TSRs. Comparisons of the averaged Cm and Cp are
shown in Fig. 13(a) and (b), respectively. The Cm of both rotors
decreases approximately linearly with the increase of TSR. Although
the averaged Cm of the Optimal Savonius is lower before TSR ¼ 0.7, it
is increased by a larger increment at the following TSRs. This phe-
Fig. 10. Comparison of the shape for the Classical Savonius and the Optimal Savonius. nomenon implies that the Optimal Savonius is more suitable for
high-speed rotation cases. There is a peak for each Cp curve, and the
power coefficients increase with TSR to TSR ¼ 1.0 after which it
5. Results and discussion
drops as TSR further increases. The peak Cp for the two rotors are
listed in Table 4. The Optimal Savonius predicts a peak value of
5.1. Comparison of the geometries
0.2580, which is 4.41% higher than the Classical Savonius. It should
be noted that this improvement in Cp is better than the accuracy of
In order to give a more intuitive expression of the Optimal
the CFD method because the results in Fig. 5 and Table 3 suggest a
Savonius blade, the blade shapes of the Optimal Savonius and the
maximum relative error of less than 0.5% for the CFD method.
Classical Savonius are compared in Fig. 10. Generally, the optimal
Therefore, the optimal result can be trusted. In addition, the per-
shape is flatter and thicker than the classical semi-circle blade. Due
centages of power increase suggested by the CFD approach (4.41%)
to the changes in the geometry, the aerodynamic forces and flow
and the PSO method (4.49%) are very close, which in turn proves
structures of the Optimal Savonius must be different from those of
the accuracy of the PSO method.
the Classical Savonius and will be discussed in the following
contents.
5.3. Pressure distributions

5.2. Torque and power analysis Pressure distributions of the flow fields around the blades are
presented to explore the flow characteristics, as shown in Fig. 14. To
Fig. 11 shows the variations of the instantaneous Cm of the give a clear illustration, the pressure contours at different rotation
Classical and the Optimal Savonius rotors in a rotation period and at angles, from 0 to 315 with a step 45 , are marked along the
different tip speed ratios. The instantaneous torque experienced instantaneous Cm curve of the Optimal Savonius at TSR ¼ 1.0. At the
two cycles in one revolution. The Cm of both rotors shows the same rotor angle of 0 and 45 , a high-pressure area can be observed on
trend. The amplitude of instantaneous Cm increases with the TSR. the concave side, and a low-pressure area can be observed on the
The instantaneous Cm of the two rotors at TSR ¼ 1.0 (the TSR convex side. The differential pressure on both sides leads to a high
corresponding to the peak Cp) are compared to show the difference positive torque to drive the rotor. At 90 , a great low-pressure area
between both rotors, as illustrated in Fig. 12 (a). It is clear that the appears outside the blade and near the tip, which contributes to
instantaneous Cm is obviously improved, especially in the region increasing the driving torque for the rotor. This low-pressure area is
 
between 25 and 100 . Since the rotor torque is the combined re- the result of tip vortex shedding and is too far to influence the blade
sults of the two blades, the Cm of a single blade can give a better after 180 . The blade generates negative torque in the ‘returning’
comparison, as shown in Fig. 12 (b). The instantaneous Cm of the process (rotation angle between 150 and 320 ), in this region

Optimal Savonius is apparently increased in the region between 25 positive pressure acts on the convex side, and negative pressure
  
and 150 , but is decreased in the region between 225 and 330 . acts on the other side. The net pressure leads to a negative torque
This variation in the instantaneous Cm will be explained in Section inhibiting the rotation of the blade.

1 1
TSR=0.6 TSR=0.6
TSR=0.8 TSR=0.8
0.8 0.8
TSR=1 TSR=1
TSR=1.2 TSR=1.2
0.6 0.6

0.4 0.4
Cm

Cm

0.2 0.2

0 0

-0.2 -0.2

-0.4 -0.4
0 50 100 150 200 250 300 350 400 0 50 100 150 200 250 300 350 400
Rotation angle / ° Rotation angle / °
(a) Classical Savonius (b) Optimal Savonius
Fig. 11. Classical Savonius (a) and Optimal Savonius (b): Comparison of the torque coefficient for the four TSRs.
294 W. Tian et al. / Renewable Energy 117 (2018) 287e299

0.8 1
Classical Classical
0.7 Optimal Optimal

0.6 Average
Cp=0.2580
0.5 0.5

0.4
Cm

Cm
0.3
Average
0.2 Cp=0.2471 0

0.1

-0.1 -0.5
0 50 100 150 200 250 300 350 400 0 50 100 150 200 250 300 350 400
Rotation angle / ° Rotation angle / °
(a) Torque coefficient of the rotor (b) Torque coefficient of a single blade
Fig. 12. Comparison of the instantaneous Cm between the Classical and the Optimal Savonius rotors: torque of the rotor (a) and a single blade (b).

0.38 0.28
Classical Classical
0.36 Optimal Optimal
0.27

0.34
0.26
0.32
4.41%
0.25
0.3
Cm

Cp

0.28 0.24

0.26
0.23

0.24
0.22
0.22
0.21
0.2

0.18 0.2
0.6 0.7 0.8 0.9 1 1.1 1.2 0.6 0.7 0.8 0.9 1 1.1 1.2
TSR / ° TSR / °
(a) Torque coefficient (b) Power coefficient
Fig. 13. Comparison of (a) average Cm and (b) average Cp at different TSRs for the Classical Savonius and the Optimal Savonius.

Table 4
Comparison of main performance parameters.

Initial design(Classical) Optimal design (PSO) Percentage of Improvement(PSO) Optimal design (CFD) Percentage of Improvement(CFD)

Cp 0.2471 0.2582 4.49% 0.2580 4.41%

As can be seen from Fig. 12(b) that the main difference between To give a further insight into the quantitative comparison of
the Cm curves of the Classical and Optimal rotors occur in the re- pressure at the selected rotation angles, pressure distributions
   
gions of (25 , 150 ) and (225 , 330 ). Comparison of pressure dis- along the blade are plotted against the normalized radial location,
tributions on the two blades will give a precise explanation for the x/d, where x represents the radial location on the blade and d is the
 
variation of the Cm. Typically, two rotation angles, 90 and 270 , blade chord length (0.5 m). The comparisons at rotation angles of
  
locate in the two regions and are selected for the following com- 90 and 270 are shown in Fig. 16 (a) and (b), respectively. At 90 ,
parisons. There are two characteristic features in the pressure the pressure on the concave side of the Optimal Savonius is
distributions in Fig. 15, with the first one being a low-pressure zone slightly lower than that of the Classical Savonius. But the pressure

near the tip at 90 and the second one being a high-pressure zone on the convex is significantly reduced, especially near the tip

on the mid of the convex surface of the blade at 270 . where the low-pressure zone exists (Fig. 15). In general, the net
W. Tian et al. / Renewable Energy 117 (2018) 287e299 295

Fig. 14. Pressure distributions of the local flow at different rotors angular positions.

 
Fig. 15. Comparison of the pressure distributions around the blades for (a) the Classical Savonius and (b) the Optimal Savonius for TSR ¼ 1.0 and at rotation angles of 90 and 270 .

pressure on the blade is reduced, and torque of the blade is (1) Reverse Flow: flow deflected by the convex surface of the
improved. This explains the variation of blade Cm in the region of upper blade and the concave side of the lower blade and
  
(25 , 150 ). While at 270 , the blade is exactly opposite the up- moving towards the tip of the lower blade.
stream flow and generates high pressure on the convex side. The (2) Tip Vortices: Reverse Flow meets the upstream flow, sepa-
Optimal Savonius has higher positive pressure on the convex side rates at the tip of the lower blade and forms high-speed
and lower negative pressure on the concave side. There is no vortices.
doubt that the optimal blade has a worse performance in the re- (3) Recovery Flow: flow moving from the tip of the lower blade
 
gion of (225 , 330 ). along its convex side to the concave side of the upper blade.
(4) Jet Flow: flow passing from the lower blade to the upper
blade through the gap between the two blades.
(5) Stagnation: stagnation point on the convex side of the blade
5.4. Flow structures with zero velocity.
(6) Upwind Separation: separation flow at the tip of the upper
Although the difference in the performance of Optimal and blade, resulting with low-velocity vortices.
Classical Savonius blades has been explained from the pressure (7) Suction Vortices: a combined result of the Recovery flow, the
perspective, the reason why the pressure changes occur needs to be Jet flow, and the Upwind separation, covering most of the
explained. In this section, an explanation for the variation in concave side of the upper blade.
pressure and finally in the Cm and Cp of the blade will be given.
Fig. 17 shows the comparison of flow structures around the Optimal Now that the flow structures are clear, it is easy to find the
 
and Classical Savonius blades at rotation angles of 90 and 270 , for difference between the two rotors. Before the analysis, one basic
a simple expression, blades at these two locations are denoted as rule from the Bernoulli equation should be bear in mind that where
the lower blade and the upper blade, respectively. It can be seen higher velocity occurs, lower pressure follows. The main difference
from the velocity and vorticity contours and the streamlines in between the flow structures of the Optimal and the Classical
Fig. 17 that the rotor is under rather complicated flows. There are Savonius rotors are as follows:
several main features of the rotor flows which are marked in the
figure:
296 W. Tian et al. / Renewable Energy 117 (2018) 287e299

20 80
Classical Savonius
0
60 Optimal Savonius
-20 Convex surface
Concave surface 40
-40

-60 20
Pressure/ Pa

Pressure/ Pa
-80 0
Concave surface
-100
-20
-120
-40
-140

-160 Classical Savonius -60


Optimal Savonius
Concave surface
-180
-80
-200
-100
-0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 -0.2 0 0.2 0.4 0.6 0.8 1 1.2
x/d x/d
(a) 90° (b) 270°
 
Fig. 16. Comparison of the pressure distributions on the blades for the Classical Savonius and the Optimal Savonius for TSR ¼ 1.0 and at rotation angles of (a) 90 and (b) 270 .

Savonius flatter convex surface, resulting with a faster Re-


covery low.
(3) The Optimal Savonius has a larger Stagnation zone, which
shows an increase in pressure on the convex side of the
upper blade (Fig. 16 (b)). The reason is that the blade is
blunter and has a larger block effect on the upstream flow.
Besides, the Stagnation point is closer to the tip of the blade
(a larger radial position), this increases the arm length of the
drag and leads to a lower negative torque.
(4) The Optimal Savonius has stronger Suction Vortices, result-
ing in a decrease in pressure on the concave side of the upper
blade (Fig. 16 (b)). The stronger Suction vortices are the result
of the stronger Jet Flow, Recovery Flow and Upwind
Separation.

Through the above investigation of the flow structures, the


reasons for the variation in pressure (Fig. 16) and finally in the
instantaneous Cm (Fig. 12 (b)) are clear. In summary, the improved
Tip Vortices and Recovery Flow contribute to the increase of the
performance of the Optimal Savonius rotor. Notably, the influences
of the two sides of the blade on the performance of the rotor are
summarized and listed in Table 5. The results will help the future
design and optimization of Savonius rotors.
Fig. 17. Comparison of flow structures around the blades for the Classical and Optimal A further comparison of the wake flow structures of the Optimal
 
Savonius rotors for TSR ¼ 1.0 and at rotation angles of 90 and 270 : (a) contours of
and Classical Savonius is also performed. Fig. 18 shows the contours
velocity and streamlines, (b) contours of vorticity.
of velocity and vorticity of the two rotors. Both rotors have similar
wake structures. There are three main features in the velocity
(1) The Optimal Savonius has a stronger Reverse Flow and Jet contours. The first one is the low-speed wake region directly
Flow, which leads to a loss of pressure on the concave side of downstream of the rotor. This region is about 6D in length
the lower blade (Fig. 16 (a)). It is because the concave side of
the optimal blade is flatter, which makes it easier for the
Table 5
upstream flow passing from both the root and the tip of the
Influences of the blade sides on the performance of the rotor.
blade.
(2) The Optimal Savonius has stronger Tip Vortices and Recovery Blade side Flow structure Cm and Cp
Flow, which leads to a reduction in pressure on the convex Convex Increase Stagnation Decrease
side of the lower blade (Fig. 16 (a)). It is due to the Reverse Increase Recovery Flow Increase
Increase Suction Vortices Decrease
Flow of the Optimal Savonius is stronger, and the velocity
Concave Increase Reverse Flow Decrease
near the tip of the blade is higher and stronger Tip Vortices Increase Tip Vortices Increase
occur. Also, it is easier for the air flowing along the Optimal Increase Jet Flow Decrease

The items in bold indicate positive effect on the Cm and Cp.


W. Tian et al. / Renewable Energy 117 (2018) 287e299 297

Fig. 18. Comparison of flow structures in the wake of the Optimal Savonius and the Classical Savonius: (a) contours of velocity, (b) contours of vorticity.

Table 6
Comparison of the power improvement with previous studies.

Source El-Askary et al. [11] Mohamed et al. [12] Al-Faruk et al. [15] Kacprzak et al. [16] Kamoji et al. [25] Present study

method flow augmentation deflector new blade shape new blade shape new blade shape double side optimization
increase percent 165% 40% 29% 26% 11% 4.41%

(contours beyond 4D are not shown in the figure), and the velocity (2) Regarding the average Cp, the Optimal Savonius rotor is su-
in this region is significantly decreased. The second one is the perior to the classical design when TSR  0.7. The maximum
strong speed-up region, which is the result of the Tip Vortices and average Cp is 0.2580, which is 4.41% higher than the classical
located along the lower boundary of the low-speed wake region. design.
The size of the strong speed-up region increases with downstream (3) The optimal blade shape helps to increase the instantaneous
 
distance, but its local velocity decreases due to the dissipation of Cm at rotation angles between 25 and 150 , but has adverse
 
the Tip Vortices. The third one is the weak speed-up region which is effect at rotation angles between 225 and 330 .
the result of the Upwind Separation and located along the upper (4) The blade shape has distinct effects on the flow structures
boundary of the low-speed wake region. The weak speed-up region around the rotor. The reason for the improvement of the
has the similar distribution with the strong speed-up region but is turbine performance is the increased Tip vortices and the
smaller in both size and velocity. Recovery flow.
The contours of vorticity in Fig. 18 (b) clearly shows the gener- (5) The influences of each side of the blade are analyzed and
ation, development, and dissipation of the tip vortices. Vortices summarized.
shed from both blades with greater intensity for the lower blade.
The contra-rotating scheme of vortices is similar to the Karman The performance of the rotor is increased by 4.41%, which is not
vortex street. These vortices dissipate quickly with the increased as obvious as other optimization studies on a traditional Savonius
downstream distance. The value of vorticity near the blade tip is on rotor [11,12,15,16,25] (Table 6). But it should be noted that these
the order of 1000/s (a maximum value of 100 is used in the contours approaches are complementary because previous studies mainly
for a better visualization). However, it decreases to about 20/s at 4D work on the optimization of blades with the same sides and this
downstream. paper is on the blade with different sides. If doing a combination of
previous ‘thin’ blade optimization and the double side optimization
6. Conclusions in this study, the performance of the rotor could be further
improved. Fig. 16 (a) suggests that the pressure on both sides of the
A novel Savonius blade with different sides is proposed to in- blade is negative, which means that the rotor is driven by lift,
crease the performance of the rotor. The shape of the blade surface although it is called as a ‘drag-type’ rotor. Lift-type rotors predict
is generated from a semi-ellipse, and two design parameters are set higher Cp values. If a full parameter optimization, which includes all
to describe a blade. 68 Savonius rotors with different blades are the design parameters, such as the shape of both sides, gap ratio,
numerically analyzed using CFD method. Simulation results are overlap ratio, tip deflector, etc., could be performed, the rotor
compared to the experimental data coming from Ref. [22], which performance may be comparative to the Darrieus rotor.
verifies the feasibility of the numerical method. Based on 68 sample It should be added that although the new design proposed in
points, a response surface surrogate model is established according this paper can improve the power coefficient of the Savonius rotor,
to the Kriging Method. Accordingly, PSO algorithm is presented to it increases the difficulty of manufacture because of the different
search for the optimal shape with the highest Cp. Comprehensive shapes on both sides. A possible solution to this problem is to
comparisons of torque, power and flow structures between the fabricate the two sides separately using thin plates and then fix
Optimal Savonius and the Classical Savonius are performed to find them together as a whole blade. Another consideration on the new
how the blade shapes improve the rotor performance. The main design is the weight of the rotor, which would be heavier due to the
conclusions of this paper are summarized as follows: more material used and have a worse starting performance. This
problem could be possibly solved by using light and high-strength
(1) The Optimal Savonius blade is flatter and thicker than the materials to build the blade.
classical one, with design parameters of a1 ¼ 0.3936 and
a2 ¼ 0.2743.
298 W. Tian et al. / Renewable Energy 117 (2018) 287e299

Acknowledgement
Y ¼ fyðX1 ÞyðX2 Þ/yðXN ÞgT (14)
This research was supported by the National Science Foundation
According to Kriging Method, the response surface model is
of China (Grant No. 61572404).
built as optimization function using Matlab code.

APPENDIX
2. Particle swarm optimization (PSO)
1. Kriging Method
PSO algorithm simulates the natural swarming behavior of bird
Kriging Method defined the relationship between a series of flocking [21]. Using the information from individuals in a group, the
optimization objective Cp and design parameters. Kriging Method movement of the whole group is generated from disorder to order
expresses the unknown function by ðXÞ as: in the solving space. PSO algorithm possesses the advantages of
simple concept, global optimization performance, and less con-
y ðXÞ ¼ b þ zðXÞ
b (5) trolling parameters et al. The great ability in global optimal search
and avoiding trapping in local optimum are very attractive [27]. The
where X is an n-dimensional vector (n design variables, a1 and a2), b steps of PSO are as follows:
is the linear regression part and zðXÞ is the local deviation. zðXÞ
represents a model of a Gaussian and stationary random process (1) Initialize N particles (initial group N design parameters (a1,
with zero mean and covariance: a2)), In this paper, the number of random particles is
N ¼ 500:
E½zðXÞ ¼ 0 (6)
P1 ¼ ða1 1; a2 1Þ; P2 ¼ ða1 2; a2 2Þ; P3 ¼ ða1 3; a2 3Þ::P500
Var½zðXÞ ¼ s2 (7) ¼ ða1 500; a2 500Þ (15)

E½zðXÞzðXi Þ ¼ s2 RðX; Xi Þ (8)


(2) Calculate the fitness value (Cp) for each particle according to
where s2 is the variance of stationary random process; RðX; Xi Þ is the certain known model f (Surrogate model):
the spatial correlation function which represents the correlation
between zðXÞ and zðXi Þ [26]. The spatial correlation function is Cp ¼ f ða1 ; a2 Þ (16)
defined as:
where f is derived from Kriging Method b
y ðXÞ according to Eq. (12).
" #
X
n
RðX; Xi Þ ¼ exp  qk ðXki  Xk Þ2 (9) (3) Set the particle Pi with the best fitness value (Cp) in history as
i¼1 personal optimal solution (pBest).
(4) Choose the particle Pg with the best fitness value (Cp) of all
where qk is the kth element of correlation vector parameter q. Ac- the particles as the global optimal solution (gBest). The
cording to the Spatial Correlation Function, the correlation matrix maximum optimization objective Cp is the best choice of
can be obtained as follows: design parameters. Finally, the optimal Cp could be converted
2 3 into the optimal design parameters (a2 and a1) according to
RðX1 ; X1 Þ / RðX1 ; XN Þ
Kriging model.
R¼4 « 1 « 5 (10)
(5) Then, for each particle, calculate particle speed according Eq.
RðXN ; X1 Þ / RðXN ; XN Þ
(17) and Eq. (18) in the a2 and a1direction.
where RðXN ; XN Þ is a spatial correlation function of two known
va2 ¼ w  va2 þ c1  randðÞ  ðpBesta2  presenta2 Þ þ c2
points, and each possible combination of these simple points is
described by matrix R. The correlation between an unknown pre-  randðÞ  ðgBesta2  presenta2 Þ (17)
diction point x and the N known sample points is defined as
follows: va1 ¼ w  va1 þ c1  randðÞ  ðpBesta1  presenta1 Þ þ c2
2 3  randðÞ  ðgBesta1  presenta1 Þ (18)
ðRðX; X1 Þ
rðxÞ ¼ 4 « 5 (11)
RðX; XN Þ where va2 and va1 are particle speed; pBesta2 and pBesta1 are per-
sonal optimal solution; presenta2 and presenta1 are present position.
The main goal of the paper is to optimize the design of the blade, Acceleration coefficient c1 is 2; acceleration coefficient c2 is 2; the
so the details of the derivations details are not described here. The inertia weight w is 0.8.
final Kriging predictor by ðXÞ is obtained:
   (6) Update particle position according Eq. (19) and Eq. (20) per
y ðXÞ ¼ b
b b þ rT ðXÞR1 Y  E bg ðXÞ (12) unit time t.

presenta2 ¼ presenta2 þ va2  t (19)


where b b is the least square estimation of b and Y is N-dimensional
vector:
presenta1 ¼ presenta1 þ va1  t (20)
 1
b
b ¼ ET R1 E ET R1 Y (13)
(7) If the maximum number of iterations or minimum error
criteria is attained, the optimization process stops. In this
W. Tian et al. / Renewable Energy 117 (2018) 287e299 299

paper, the maximum number of iterations is 200, which is set experimental validation of the turbulent flow around a small incurved
Savonius wind rotor, Energy 74 (2014) 506e517.
as the stop criterion.
[14] M.R. Ahmed, M. Faizal, Y.H. Lee, Optimization of blade curvature and inter-
rotor spacing of Savonius rotors for maximum wave energy extraction,
References Ocean. Eng. 65 (65) (2013) 32e38.
[15] A. Al-Faruk, A. Sharifian, Geometrical optimization of a swirling Savonius
[1] A.A. Khan, Z. Mehmood, A. Shahzad, K.A. Chughtai, A. Javed, Evaluation of wind turbine using an open jet wind tunnel, Alexandria Eng. J. 55 (3) (2016)
wind energy potential alongside motoways of Pakistan, Asian J. Appl. Sci. Eng. 2055e2064.
3 (4) (2014) 375e381. [16] K. Kacprzak, G. Liskiewicz, K. Sobczak, Numerical investigation of conven-
[2] B. Yang, C. Lawn, Fluid dynamic performance of a vertical axis turbine for tidal tional and modified Savonius wind turbines, Renew. Energy 60 (4) (2013)
currents, Renew. Energy 36 (12) (2011) 3355e3366. 578e585.
[3] K. Pope, I. Dincer, G.F. Naterer, Energy and exergy efficiency comparison of [17] V. D'Alessandro, S. Montelpare, R. Ricci, A. Secchiaroli, Unsteady aerodynamics
horizontal and vertical axis wind turbines, Renew. Energy 35 (9) (2010) of a Savonius wind rotor: a new computational approach for the simulation of
2102e2113. energy performance, Energy 35 (8) (2010) 3349e3363.
[4] A. Shigetomi, Y. Murai, Y. Tasaka, Y. Takeda, Interactive flow field around two [18] A. Dragomirescu, Performance assessment of a small wind turbine with
Savonius turbines, Renew. Energy 36 (2) (2011) 536e545. crossflow runner by numerical simulations, Renew. Energy 36 (3) (2011)
[5] S. Sivasegaram, Design parameters affecting the performance of resistance- 957e965.
type, vertical-axis windrotors - an experimental investigation, Wind Eng. 1 [19] Z. Zhao, Y. Zheng, X. Xu, W. Liu, D. Zhou, Optimum design configuration of
(5632) (1977) 8. helical Savonius rotor via numerical study, in: ASME Fluids Engineering Di-
[6] J.V. Akwa, H.A. Vielmo, A.P. Petry, A review on the performance of Savonius vision Summer Meeting, 2009, pp. 1273e1278.
wind turbines, Renew. Sustain. Energy Rev. 16 (5) (2012) 3054e3064. [20] J.D. Martin, T.W. Simpson, Use of Kriging models to approximate deterministic

[7] B.D. Altan, M. Atılgana, A. Ozdamarb, An experimental study on improvement computer models, AIAA J. 43 (4) (2004) 853e863.
of a Savonius rotor performance with curtaining, Exp. Therm. Fluid Sci. 32 (8) [21] R. Eberhart, J. Kennedy, A new optimizer using particle swarm theory, in:
(2008) 1673e1678. International Symposium on MICRO Machine and Human Science, 1995,
[8] Z. Zhao, Y. Zheng, X. Xu, W. Liu, Research on the improvement of the per- pp. 39e43.
formance of savonius rotor based on numerical study, in: International Con- [22] C.F. Juang, A hybrid of genetic algorithm and particle swarm optimization for
ference on Sustainable Power Generation and Supply, 2009, pp. 1e6. recurrent network design, IEEE Trans. on Syst. Man Cybern. 34 (2) (2004)
[9] D. Kumar, S. Sarkar, Numerical investigation of hydraulic load and stress 997e1006.
induced in Savonius hydrokinetic turbine with the effects of augmentation [23] R.E. Sheldahl, L.V. Feltz, B.F. Blackwell, Wind tunnel performance data for two-
techniques through fluid-structure interaction analysis, Energy 116 (2016) and three-bucket Savonius rotors, J. Energy 2 (3) (1978) 160e164.
609e618. [24] M.M.A. Bhutta, N. Hayat, A.U. Farooq, Z. Ali, S.R. Jamil, Z. Hussain, Vertical axis
[10] B. Wahyudi, S. Soeparman, H.W.M. Hoeijmakers, Optimization design of wind turbine e a review of various configurations and design techniques,
savonius diffuser blade with moving deflector for hydrokinetic cross flow Renew. Sustain. Energy Rev. 16 (4) (2012) 1926e1939.
turbine rotor, Energy Procedia 68 (2015) 244e253. [25] M.A. Kamoji, S.B. Kedare, S.V. Prabhu, Experimental investigations on single
[11] W.A. El-Askary, M.H. Nasef, A.A. Abdel-Hamid, H.E. Gad, Harvesting wind stage modified Savonius rotor, Appl. Energy 86 (7) (2009) 1064e1073.
energy for improving performance of Savonius rotor, J. Wind Eng. Ind. Aer- [26] C. Sun, B. Song, P. Wang, Parametric geometric model and shape optimization
odyn. 139 (2015) 8e15. of an underwater glider with blended-wing-body, Int. J. Nav. Archit. Ocean
[12] M.H. Mohamed, G. Janiga, E. Pap, D. The venin, Optimal blade shape of a Eng. 7 (6) (2015) 995e1006.
modified Savonius turbine using an obstacle shielding the returning blade, [27] H.N. Chen, Y.L. Zhu, K.Y. Hu, T. Ku, Global optimization based on hierarchical
Energy Convers. Manag. 52 (1) (2011) 236e242. coevolution model, Evol. Comput. (2008) 1497e1504.
[13] Z. Driss, O. Mlayeh, D. Driss, M. Maaloul, M.S. Abid, Numerical simulation and

Вам также может понравиться