Accepted Manuscript

A novel wake energy reuse method to optimize the layout for Savonius-type
vertical axis wind turbines

Baoshou Zhang, Baowei Song, Zhaoyong Mao, Wenlong Tian

PII: S0360-5442(17)30004-X

DOI: 10.1016/j.energy.2017.01.004

Reference: EGY 10139

To appear in: Energy

Received Date: 02 August 2016

Revised Date: 25 December 2016

Accepted Date: 02 January 2017

Please cite this article as: Baoshou Zhang, Baowei Song, Zhaoyong Mao, Wenlong Tian, A novel
wake energy reuse method to optimize the layout for Savonius-type vertical axis wind turbines,
Energy (2017), doi: 10.1016/j.energy.2017.01.004

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to
our customers we are providing this early version of the manuscript. The manuscript will undergo
copyediting, typesetting, and review of the resulting proof before it is published in its final form.
Please note that during the production process errors may be discovered which could affect the
content, and all legal disclaimers that apply to the journal pertain.
 ACCEPTED MANUSCRIPT

Highlights

A novel wake energy reuse method is proposed to optimize the layout for S-VAWT
Response surface models for Cp and layout positions is created by Kriging Method
Energy distributions of the turbine wake are investigated
The performance of the turbine at optimal position is increased by 22.89%
The optimal position is suitable for multi-turbines layout in a large wind farm
 ACCEPTED MANUSCRIPT

1 A novel wake energy reuse method to optimize the layout

2 for Savonius-type vertical axis wind turbines
3 Baoshou Zhang,1Baowei Song, Zhaoyong Mao, Wenlong Tian
4 School of Marine Science and Technology, Northwestern Polytechnical University
5 No. 127, Youyi Road(West), Beilin, Xi'an City, Shaanxi Province, PRC
6

7 ABSTRACT
8 The long wake of a wind turbine has a significant impact on the performance of downstream turbines.
9 Under the inspiration of migrating geese flying in a V or I formation to save energy, a novel wake energy
10 reuse method is proposed to optimize the layout for Savonius-type vertical axis wind turbines (S-VAWT).
11 VAWT wakes include a series of high speed and energy zones. On both sides of the upstream turbine,
12 7  16 transient two-dimensional numerical simulations are performed with Fluent to investigate wake
13 structure, interaction effect and power coefficients (Cp) of downstream turbines. Based on Kriging Method,
14 a response surface model (Surrogate model) is created to describe the relationship between the optimization
15 objective Cp and layout positions. Finally, particle swarm optimization algorithm is applied to find the
16 optimal relative layout position (5.25m, -2.18m) of the downstream turbine. The optimal position is located
17 in the periodic high speed zone of the wake on the advancing blade side. And the optimal position is
18 suitable for multi-turbines in a large wind farm. The optimization results show that Cp of downstream
19 turbines at optimal layout position is significantly increased from 0.2477 to 0.3044 (22.89% higher).

20 KEYWORDS
21 Savonius-type vertical axis wind turbines; Wake; Layout; Numerical simulation; Interaction effect
22
23

24 1. INTRODUCTION
25 The gradually exhausting non-renewable energy resources have stimulated the development of the
26 renewable energy technology. Wind is an important renewable energy source. Recently, a new challenge
27 arises in the design of the wind turbine layout. The wake of the upstream turbine has a detrimental effect in
28 a wind farm. In the turbine wake, wind speed decreases, while the turbulence intensity increases. These
29 effects result in a large disturbance for downstream wind turbines. Previous studies have found that uneven
30 load stress appears on downstream wind turbines, and such fatigue load will reduce the life of turbines.
31 [1,2,3] Meanwhile, the unsteady wake will deteriorate the performance of downstream wind turbines,
32 which further reduces the economic efficiency of a wind farm. According to previous work, 10–20% of the
33 total power output of large wind farms will be wasted [2]. Thus, it is quite meaningful to study the wake to
34 wake interaction and the characteristics of the wake. The traditional method for wind turbines layout aims

Corresponding author: Baoshou Zhang, Telephone/fax number: +86 18792547553; E-mail address: 1262034751@qq.com

1
 ACCEPTED MANUSCRIPT

1 to reduce interaction effects between different turbines and increase power. In other words, the wake of the
2 upstream turbine should not overlap with the downstream turbine [1]. Other researches have shown that
3 increasing the number of turbines could appropriately improve the economic efficiency in equal surfaces
4 [2,4,5]. However, there are no public literatures relating to the reuse of the wake energy to enhance the
5 performance of downstream turbines. And the relationship between the layout position and the performance
6 of vertical axis wind turbine has not been investigated in detail. Therefore, it is essential to conduct
7 research on wind turbines layout optimization for energy efficiency improvement.
8 This study takes Savonius-type vertical axis wind turbines (S-VAWT) as an example. S-VAWT is a
9 type of vertical axis wind turbines. Main advantages of S-VAWTs can be described as follows:
10 • High starting up and full operation moment [6]
11 • Operation in a wide range of wind conditions
12 • Simple construction and low noise emission [7,8]
13 These advantages make S-VAWTs suitable for small and medium-sized distributed power generation.
14 In general, this power generation is between tens of kW and hundreds of kW. Wind energy utilization
15 efficiency of S-VAWTS is 15-25% in past simulations and experiments [6,8]. There is still room to
16 improve the relatively low efficiency. For S-VAWTs, after the wind bypasses blades, the incoming flow
17 becomes more complicated. According to previous studies [9,10,11,12], the performance of the
18 downstream turbine is deteriorated by the wake, and therefore the distance between two S-VAWTs should
19 be large enough to reduce the wake effect. The distance for VAWT is usually 15 times larger than the rotor
20 diameter [1]. The distance limits the maximum number of turbines in equal available surfaces. However, a
21 series of high speed and energy zones are also observed in the wake, and so far few investigations were
22 performed to reuse the wake energy of upstream turbines. Thus, a novel wake energy reuse method is
23 proposed for the turbine layout optimization in the current work. Downstream turbines are placed in the
24 high speed and energy zones of the wake to reuse the wake energy. The results show that, compared with
25 the traditional layout, the wind turbine distances are far less, i.e., the number of turbines increases in equal
26 surfaces, and the downstream turbine acquires a higher efficiency.
27 Optimization method is essential in the current work to find the optimal turbine layout. Considering
28 the large number of design schemes and the complexity of the wake, wind tunnel experiments are difficult
29 to implement, while numerical method is efficient and intuitive [4,6]. To obtain the optimal solution for the
30 turbine layout, a series of transient CFD simulations is performed to investigate different layout
31 characteristics with Fluent. Surrogate model is always used in the optimization process when the evaluation
32 of the objective function is time-consuming, as in the current problem. Kriging Method proposed by Danie
33 G. Krige in 1951 is a widely used surrogate model to establish the response surface [13]. The relationship
34 (response surface model) between optimization objective (energy efficiency) and layout positions could be
35 precisely defined based on Kriging model.
36 Then, a highly effective optimization method is needed to find the optimal solution based on Kriging
37 model. Particle swarm optimization (PSO) algorithm is an optimization method that is simple in concept,
38 easy to be implemented and computationally inexpensive [14]. Accordingly, the PSO algorithm is adopted
39 to find the optimal relative position of the downstream turbine. The optimal layout position of the

2
 ACCEPTED MANUSCRIPT

1 downstream turbine is obtained through a series of iterations. Energy distributions, typical velocity
2 distributions, pressure distributions around different blades and interaction effect of multi-turbines are also
3 illustrated and analyzed. Compared with traditional layout, the proposed optimal layout has significant
4 improvements. The power coefficient Cp of downstream turbines is increased by 22.89%. And the distance
5 between the upstream turbine and the downstream turbine is obviously reduced. So in equal surfaces, the
6 number of turbines using the new layout method is about 4 times the number of turbines using past no
7 overlap layout method.
8 The main contribution of the paper is that a new way to regard the turbine’s wake is proposed. The
9 high speed zone in the wake created by the upstream turbine is exploited to improve the performance of
10 downstream turbines. It is a completely new thought to optimize the layout of S-VAWTs. This novel layout
11 method has both theoretical and practical values, and gives a new idea for further development of wind
12 energy utilization.
13

14 2. GEOMETRY CONFIGURATION
15 As known to all, migrating geese fly in a V or I formation, as shown in Fig. 1(a). The goose in front
16 produces high speed updrafts in wake, so downstream geese fly in updrafts or use the pressure difference
17 to save energy. Similarly, upstream turbine produces a long wake and a periodic high speed zone on both
18 sides of the wake. As the upstream turbine rotates, high speed rotating vortices appear behind the blades
19 [11]. The periodic high speed zone appears accordingly. The velocity difference and the high speed zone
20 could be used to enhance the performance of downstream turbines.
21 In particular, there are two types of layout positions for each downstream turbine. Type I: locating
22 returning blade in low speed zone but advancing blade in periodic high speed zone, as shown in Fig. 1(b)
23 Type I position. Type II: locating two blades in the periodic high speed zone, as shown in Fig.1(b) Type II
24 position. This difference will be discussed in detail in following sections.
25 VAWT is a type of drag-driven devices. [6] Advancing blade which moves in the same direction with
26 flow provides thrust, while returning blade which moves in the opposite direction with flow provides drag.
27 Generally, downstream turbines could be arranged into one side type layout (on advancing blade side or
28 returning blade side) or V type layout (on both sides), as shown in Fig. 2.
29 Geometric model of S-VAWT is shown in Fig. 3(a). According to previous studies [6,7,15], a
30 simplified two-dimensional model can be used to describe the wind turbine, as shown in Fig. 3(b). Some
31 parameters are necessary to define the turbine profile, as listed in Table 1. D is the turbine diameter; H is
32 the turbine height; d is the length of the blade; r is the radius of the blade; S0 is the length of blades overlap.
33 These turbines are exposed to the wind with the rotation rate  . The undisturbed wind speed is v  . [15,16]
34 Tip speed ratio (TSR) means the ratio between blade tip speed and inflow air speed, i.e., TSR is defined
35 as Eq. (1):

36 TSR  D (2v  ) (1)

37 The performance of a turbine can be characterized by using the two main indicators: torque coefficient
38 (CT ) and power coefficient (Cp). Cp represents the fraction of extracted power from the total power of air

3
 ACCEPTED MANUSCRIPT

1 flow which runs through the projected area of the turbine at the flow direction. CT and Cp can be obtained as
2 follows:

3 CT  4 M ( ρHv2 D 2 ) (2)

4 C p  P (0.5 HDv3 )  C m  TSR (3)

5 where M is the torque acting on turbines; P is the power of the turbine;  is the rotational speed;  is the
6 density of water. Since only two-dimensional simulations are performed, the height H of turbines is 1m.
7 [17]
8 Generally, power coefficient Cp is used to evaluate the comprehensive performance of Savonius
9 turbines. In this paper, the optimization objective is to attain the best Cp by locating the downstream turbine
10 at the optimal position in the wake.
11

12 3. NUMERICAL METHOD
13 Numerical simulation is a precise and high-efficient investigation approach for the wake flow. A series
14 of two-dimensional representations of the flow field is built. Hence, the flow in the vertical direction is
15 neglected, which is equivalent to 3D Savonius turbines with endplates. It has been already successfully
16 proved in similar studies by using Fluent [6,17-19]. Then, the results of the numerical simulation are
17 validated and compared with some previous typical experimental data. The results of these two methods are
18 in good agreement, as will be discussed in the upcoming sections.
19 In the optimization process, numerical method using Fluent provides fundamental elements for the
20 modeling of the response surface surrogate model (established by using Kriging Method). Numerical
21 method could simulate wake flow around the turbines. Correspondingly, fluid forces/ moments/ angular
22 velocities/ etc. acting on these turbines are obtained separately. Meanwhile, the performances (Cp power
23 coefficient) of the turbines are obtained. For a certain relative position between the upstream turbine and
24 the downstream turbine, a certain performance of the turbine is obtained. In other words, Fluent translates
25 the information of layout position (x, y) into the performance (Cp power coefficient) of the turbine. Then,
26 Kriging Method establishes the relation model (Surrogate model) between a series of Cp and layout
27 positions (x, y). Finally, the optimal value of Cp is found based on PSO and the surrogate model.
28
29 3.1. Computational domains and grid generation
30 The computational domain is shown in Fig. 4. A 3-unit layout is a fundamental element of a larger
31 amount of units in a large wind farm. Interaction effect between three turbines is the most representative
32 example. Wind turbines at A and B positions have the same direction of rotation. Relative position between
33 the rotation centers of two turbines at A or B positions is ( xB , y B ) . 7  16 schemes are simulated to cover the
34 wake on the advancing blade side. x B changes from 2m to 8m with a 1m interval; y B changes from -0.4m to
35 -3.4m with a 0.2m interval. Similarly, relative position between A and C positions is ( xC , y C ) . The wind
36 turbine at C position rotates in the opposite direction. Wind speed is 7.0 m/s. The initial conditions are the
37 same as those used in previous studies. [6,17] Simulation results provide basic component factors for

4
 ACCEPTED MANUSCRIPT

1 layout optimization.
2 Moving mesh technology also known as dynamic mesh technology is used to simulate the rotation of
3 the wind turbine. Rotating domain and local details of the mesh are shown in Fig. 5.
4
5 3.2. Numerical method validation
6 In this section, mesh independence validation and time step independence validation are conducted.
7 Mesh independence is tested by assessing the instantaneous CT of the wind turbine for different mesh
8 densities [20,21]. CT is presented in Fig. 6(a). Since two blades are origin-symmetric, the instantaneous CT
9 is cycled twice in a rotation period. The different CT has the great consistency, though the mesh densities
10 are different (61000/86900/117000 elements). Accordingly, the mesh with approximately 60,000 elements
11 is sufficient for the mesh independence study. And this mesh density is propitious to ensure the higher
12 computation efficiency.
13 The time step independence validation is conducted using different time steps ranging from 0.5
14 degrees/step to 4 degrees/step, as shown in Fig. 6(b). The different instantaneous Cp ranging from 0.5
15 degrees/step to 2 degrees/step have the great consistency. Taking the efficiency of simulated into
16 consideration, 1.0 degree/step is selected in the optimization process.
17 To verify the feasibility of the numerical method, simulation results are compared with wind tunnel
18 experimental data presented by US Sandia Laboratories [22]. Fig. 7 shows the comparison of the average
19 Cp at different TSR for experimental data and simulation data obtained by using this numerical method.
20 The values of averaged power coefficient increase with the TSR. The maximum of averaged power
21 coefficient (Cp=0.2477) appears at about TSR=0.8-1.0. After a certain value, Cp decreases. It is clear that
22 simulation results fit well with the experimental data. Therefore, the numerical method using Fluent is
23 acceptable. The simulation results of the single turbine are used as reference values in the following
24 optimization process.
25
26 3.3. Simulation example
27 Velocity distributions and turbulence kinetic energy distributions of the wake are shown in Fig. 8 and
28 Fig. 9. The upstream turbine creates periodic high speed vortex zones and an unsteady flow with increased
29 turbulence. The shedding vortex is stretched with the increase of distance.
30 In the simulation process, the cyclical CT during 15 rotation periods is monitored, as shown in Fig.
31 10(a). The instantaneous CT cyclically varies with the rotor angle. The variation of CT of the upstream
32 turbine keeps the same in the sixth and subsequent rotation periods. Due to the wake effect, the variation of
33 CT of downstream turbines is more complex.
34 With these coefficient curves, the performance of S-VAWT can be defined and compared, which
35 provides the parameterization basis for the layout optimization. The instantaneous CT of a single blade is
36 shown in Fig. 10(b). Average power coefficients (Cp) of different turbines are compared in Fig. 10(c). Ref.
37 value represents Cp of a single wind turbine obtained without considering the effect from other turbines. Cp
38 of the upstream turbine is close to Ref. value (Cp=0.2477); At B position, the wind turbine is located in the
39 periodic high speed zone. Correspondingly, Cp is improved obviously. At C position, the returning blade of

5
 ACCEPTED MANUSCRIPT

1 the wind turbine is located in low speed zone, and the advancing blade is located in the periodic high speed
2 zone. However, Cp fells significantly. It means that performance degradation occurs. These layout
3 differences will be discussed in section 5.2.
4

5 4. OPTIMAL DESIGN METHOD

6 Generally, particle swarm optimization (PSO) is a method of searching for the optimal value according
7 to a certain known model (Surrogate model). In this paper, the certain known model is established by using
8 Kriging Method. Kriging model describes the relationship between optimization objective Cp and layout
9 positions (x, y). Cp represents the performance of the downstream turbine. PSO could find the maximum
10 value of Cp. Finally, the maximum value of Cp corresponds to the optimal layout position (x, y) of the
11 downstream turbine.
12
13 4.1. Kriging Method
14 Numerous adaptations of layout positions (x and y) are performed to establish the surrogate model
15 using Kriging Method. Kriging Method defines the relationship between optimization objective Cp and
16 layout position parameters accurately. Kriging Method expresses the unknown function ŷ (X) as:
17 yˆ (X)    z (X) (4)
18 In this paper, X is a 2-dimensional vector including position parameters (x and y);  is the linear
19 regression part; z (X) is the local deviation. z (X) represents a model of a Gaussian and stationary random
20 process with zero mean and covariance:
21 E[ z (X)]  0 (5)

22 Var[ z (X)]   2 (6)

23 E[ z (X) z (X i )]   2 R (X, X i ) (7)

24 where  2 is the variance of the stationary random process; R(X, X i ) is the spatial correlation function which
25 represents the correlation between z (X) and z (X i ) . The Spatial Correlation Function (SCF) is defined as:
n
26 R(X, X i )  exp[ 
i 1
k (X ki  X k ) 2 ] (8)

27 where  k is the kth element of the correlation vector parameter  . The correlation matrix can be obtained as
28 follows:
 R( X1 , X1 )  R( X1 , X N )
 
29 R     (9)
 R( X , X )  R( X , X )
 N 1 N N

30 where R(X N , X N ) is a SCF of two known points, and each possible combination of these simple points is

31 described by matrix R. The correlation between an unknown prediction point X and the N sample points is
32 defined as follows:

6
 ACCEPTED MANUSCRIPT

 R(X, X 1 ) 
 
1 r (X)     (10)
 R(X, X )
 N 

2 The main goal of the paper is to optimize the layout of turbine, so the details of the derivations details are
3 not described here. The final Kriging Method predictor ŷ (X) is obtained [13,23]:

4 yˆ (X)  ˆ  r T (X) R 1 (Y  E ( gˆ (X))) (11)

5 where ̂ is the least-squares estimated of  ; Y is N-dimensional vector:

6 ˆ  ( E T R 1 E ) 1 E T R 1Y (12)

7 Y  y (X 1 ) y (X 2 )  y (X N ) (13)
T

8 This uncertainty of Kriging predictor ŷ (X) is expressed as follows.

MSE  E[( yˆ (X)  y (X)) 2 ]

9 (14)

  2 1  r T (X) R 1 r (X)  [1  E T R 1 r (X)]2 ( E T R 1 ) 1 
10 where MSE represents the mean squared error of y (X) . MATLAB is used for the Kriging Method program
11 writing. Finally, the response surface is built as shown in Fig. 12 and Fig. 13 in the next section.
12
13 4.2. Particle swarm optimization (PSO)
14 PSO has been recognized as much effective algorithm both in speed and quality of solution. PSO
15 algorithm possesses the advantages of a simple concept, global optimization performance, and less
16 controlling parameters, et al [24]. Although, when it comes to multivariate problems or sudden changes, it
17 is not validated by a mathematical approach that PSO can ensure optimality. Because it is impossible to set
18 countless sample points to cover global variable space. But for the two-dimensional layout (x, y) in
19 empirical evidence, PSO is suitable for the less optimization variables problem in practice. Great ability in
20 global optimal search and avoiding trapping in local optimum are also attractive. In this paper, PSO
21 satisfies the requirements of optimization. Thus, PSO is adopted to solve the relatively simple two-
22 dimensional optimal layout. The steps of PSO are as follows:
23 (1) Initialize particles (initial layout position)
24 In this paper, the number of random particles is 500:
25 P1  ( x1, y1), P 2  ( x 2, y 2), P3  ( x3, y3)...P500  ( x500, y500) (15)
26 (2) Calculate the fitness value (Cp) for each particle according to the certain known model f (Surrogate
27 model):
28 Cp  f (x,y ) (16)
29 where f is derived from Kriging Method ŷ (X) according to Eq. (11).
30 (3) Set the particle Pi with the best fitness value (Cp) in history as personal optimal solution (pBest).
31 (4) Choose the particle Pg with the best fitness value (Cp) of all the particles as the global optimal
32 solution (gBest). Maximum optimization objective Cp is the best choice of wind turbine layout patterns.
33 Finally, optimal Cp could be converted into the optimal layout position (x, y) according to Kriging model.

7
 ACCEPTED MANUSCRIPT

1 (5) Then, for each particle, calculate particle speed according to Eq. (17) and Eq. (18) in the x and y
2 direction.
3 v x = w  v x + c1  rand ()  ( pBest x - present x )+ c 2  rand ()  ( gBest x - present x ) (17)

4 v y = w  v y + c1  rand ()  ( pBest y - present y )+ c 2  rand ()  ( gBest y - present y ) (18)

5 where v x and v y are particle speed; pBest x and pBest y are personal optimal solution; present x and
6 present y are present position. Acceleration coefficient c1 is 2; acceleration coefficient c2 is 2; the inertia
7 weight w is 0.8.
8 (6) Update particle position according Eq. (19) and Eq. (20) per unit time t.
9 present x = present x + v x  t (19)

10 present y = present y + v y  t (20)

11 (7) If the maximum number of iterations or minimum error criteria is attained, the optimization
12 process stops. In this paper, the maximum number of iterations is 500, which is set as the stop criterion.
13 For this two-dimensional layout optimization problem, the constraint is the position and size of these
14 turbines. First, impact is not allowed. The distance between the upstream turbine and the downstream
15 turbine should be greater than the turbine diameter (1D). Then, relative position shall not exceed the range
16 of a turbine wake (15D). All the constraints have been written in the optimization program. The layout
17 optimization program based on the PSO is designed in Fig. 11.
18 PSO is implemented in a computer program (using Matlab language) in conjunction with Kriging
19 model according to the flow diagram Fig.11. And we have compared PSO with Genetic Algorithm (GA)
20 when choosing an optimization method. In this optimization using PSO, 500 particles are randomly
21 selected. The optimal solution is obtained after about 200-250 iterations. But in the similar cases, GA-based
22 iteration takes up to 2-3 times as long as PSO-based iteration.
23 In this paper, the optimization program has been repeated for 20 times with different initial particles.
24 Compared with average value, all deviations are mainly in about 0.15%-0.18%. Meanwhile, experimental
25 results from practical project fit well with these simulation results. It is clearly that random initial particles
26 and outside initial disturbance is not obvious. All these results prove that the optimization process has high
27 efficiency and strong robustness in practice. The output in this optimization process is the optimal value Cp.
28 Then, 2 variables representing the X and Y coordinates of the downstream turbine’s position are translated
29 from Kriging model. Hence, with the help of PSO, the optimal layout solution can be found.
30
31 4.3. Optimization results
32 The optimal position of the downstream turbine obtained by the PSO algorithm can effectively
33 enhance the turbine performance. On the returning blade side, the response surface model respecting to the
34 layout positions (x, y) is established, as shown in Fig. 12(a). The response surface represents the energy
35 distributions diagram for the downstream turbine. Two green dashed lines represent reference value
36 (Cp=0.2477). Through the iterative search, the optimal point is directly obtained by using PSO. The optimal
37 point is marked in Fig. 12(b). The coordinate values of the optimal point represent the position of the
38 downstream turbine (x=4.91 and y=2.24). The predictive value of Cp is 0.29116.

8
 ACCEPTED MANUSCRIPT

1 Similarly, on the advancing blade side, the response surface model respecting to the layout position (x,
2 y) is shown in Fig. 13(a). The optimal point found by PSO is marked in Fig. 12(b). The coordinate values
3 of the optimal point represents the layout position of the downstream turbine (x=5.25 and y=-2.18). The
4 predictive value of Cp is 0.30722. It is clear that the optimal position on the advancing blade side is better
5 than that of on the returning blade side.
6

7 5. RESULTS AND ANALYSIS

8 5.1 Wake structure

9 Based on the optimization result in Section 4, the optimal position of the downstream turbine is
10 completely located in the periodic high speed zone, and the optimal layout position on the advancing blade
11 side has a definite advantage. In order to test the accuracy of the layout optimization process, experimental
12 value of Cp is used to verify the correctness of the optimal results. Optimal layout positions, predictive
13 value of Cp, experimental value of Cp, and optimization effect of are compared in Table 2 below.
14 Experimental value of Cp is consistent well with the predictive value of Cp. It also proves that the optimal
15 layout on the advancing blade side has better performance. Therefore, the optimal layout on the returning
16 blade or V type layout would not be taken into account.
17 In order to analyze the advantages of the optimal layout, simplified wake structure is illustrated in Fig.
18 14. Analogous to flow past circular cylinder, the flow field velocity on both sides of circular cylinder
19 increases due to blockage effect. The wake of the upstream turbine will be separated into 3 parts. The
20 middle part is the main wake flow (Low speed zone). The down part is a periodic high speed zone (on the
21 advancing blade side). Because the blade moves in the same direction with the flow, a lot of periodic high
22 speed zones appear. Therefore, the averaged incoming flow speed for downstream turbines increases. The
23 upper part is a secondary-periodic high speed zone (on the returning blade side). The blade moves in the
24 opposite direction with flow, so it will impede wind. Correspondingly, the secondary-periodic high speed
25 zone appears behind the returning blade. That's the explanation for why the optimal layout position is on
26 the advancing blade side.
27

28 5.2 Typical layout analysis

29 The optimal layout position of the downstream turbine is located in periodic high speed zone.
30 However, a special layout position is also noteworthy. Returning blade is located in low speed zone and
31 advancing blade is located in periodic high speed zone. A typical layout is analyzed in detail here. The
32 relative position of the downstream turbine is x=4m and y=-0.8m. To explain the interaction between two
33 turbines, the velocity distributions are shown in Fig. 15. Due to the wake effect, the flow structure around
34 the downstream turbine is very complex. A series of vortexes are generated by the upstream turbine. The
35 wake of the upstream turbine is truncated by the downstream turbine, as a result, the wake is divided into
36 two parts. It is clearly seen that different blades of the downstream turbine are affected separately by
37 different flow. Returning blade is affected by low speed flow. Advancing blade is influenced by periodic

9
 ACCEPTED MANUSCRIPT

1 high speed flow.

2 It was once believed that this type layout is the optimal layout, but in fact it isn't. In order to make a
3 qualitative analysis for the performance of the downstream turbine, instantaneous CT of turbine is
4 compared with the reference value. Variations of CT of a single blade are shown in Fig. 16(a) and Fig.
5 16(b). Two CT curves of the blades have an obvious difference. Because the returning blade is located in
6 low speed zone, the minimum value of CT becomes greater compared with the reference value. As for
7 another blade shown in Fig. 16(b), maximum value of instantaneous CT is reduced. Although, the minimum
8 value of CT becomes greater, the performance of the blades may be not improved in a whole rotation
9 period. Fig. 16(c) shows the variations of CT of the downstream turbine (including two blades). Since the
10 two blades are origin-symmetric, CT of the turbine shows a double peak profile in a rotation period.
11 Average power coefficient (Cp) is 0.2235, which is less than the reference value 0.2477. For this layout
12 method, the wake effect is significantly negative on the downstream turbine. The performance of the whole
13 turbine is not improved in this layout position.
14

15 5.3 Optimal layout analysis

16 According to the current results, one side type layout (on the advancing blade side) is the best. The
17 optimal layout position of the downstream turbine is located in the periodic high speed zone where the
18 averaged incoming flow speed for the downstream turbine increases. The flow structure of the downstream
19 turbine is almost the same to that of the upstream turbine, as shown in Fig. 17.
20 The performance of this optimal layout is validated and compared with the reference value. They all
21 are similar in profile. CT of the turbine is compared in Fig. 18. Obviously, not only a single blade, but also
22 the whole downstream turbine is significantly improved. Average Cp of the downstream turbine at optimal
23 layout position is increased from 0.2477 to 0.3044 (22.89% higher).
24 The conventional multi-turbines layout method focuses much more on reducing the interaction effect
25 between different turbines and increase power. Briefly, the wake of the upstream turbine should not overlap
26 with the downstream turbine. The wake is detrimental to downstream turbines. No overlap distance should
27 be more than 15D in flow direction. And the lateral distance should be more than about 3D. In the current
28 work, downstream turbines are placed in the periodic high speed zone of the wake and reuse the wake
29 energy, as shown in Fig. 17. In flow direction, the distance is about 5.2D. The lateral distance is about
30 2.1D. The new layout distances are far less than the conventional layout. In equal surfaces, the number of
31 turbines using the new layout method is about 4 times larger than the number of turbines using no overlap
32 layout method.
33 It is essential to investigate pressure distributions around the wind turbines [25]. At the rotor angle of
34 0 degrees (Fig. 19(a)), a high pressure zone appears on the concave side of the advancing blade, and a low
35 pressure zone appears on the convex side of the advancing blade. At 45 degrees (Fig. 19(b)), pressure
36 distributions around the advancing blade are similar to that of 0 degrees rotor angle. For the returning
37 blade, a high pressure area appears on the convex side, and a low pressure area appears on the concave
38 side. The pressure differential provides a great torque inhibiting the rotation of the turbine. At 90 degrees
39 (Fig. 19(c)), a great low pressure zone appears on the top of the advancing blade. The pressure difference

10
 ACCEPTED MANUSCRIPT

1 will increase the driving torque. However, the pressure differential on the returning blade creates negative
2 torque inhibiting the rotation of the turbine. At 135 degrees (Fig. 19(d)), a great low pressure vortex leaves
3 the top of the advancing blade. Then, the turbine rotates in these previous steps.
4 The changing curves of the pressure distributions on a single blade at 0, 45, 90, 135, 180 and 270
5 degrees are also compared, as shown in Fig. 21(a)-(f). Comparison of pressure distributions on the blade
6 will give a precise explanation where the improvement occurs. The horizontal axis represents the length of
7 the blade (m). The vertical axis represents pressure (Pascal).
8 Pressure distributions on the concave side of the reference turbine (C-Ref), the convex side of the
9 reference turbine (V-Ref), the concave side of the optimal layout turbine (C-Op) and the convex side of the
10 optimal layout turbine (V-Op) are drawn. Meanwhile, pressure differential between concave side and
11 convex side of the reference turbine (All-Ref) and the optimal layout turbine (All-Op) are also drawn in
12 Fig. 20. The curve of All-Op is evidently located upper side of the curve of All-Ref in most cases.
13 Therefore, the downstream turbine placed in the optimal layout position is obviously improved.
14

15 5.4 Multi-turbines interaction effect

16 Multi-turbines interaction effect should not be ignored. It is inevitable that the downstream turbine
17 impedes the flow and increases blocking ratio [11, 26]. The performance of the upstream turbine may be
18 affected. Response surface of Cp for the upstream turbine is illustrated in Fig. 21.
19 The response surface represents the energy distributions diagram of the upstream turbine respecting to
20 the layout position of the downstream turbine. In the region of a green dashed line, Cp is less than the
21 reference value (Cp=0.2477), which means the interaction effect from downstream turbine can’t be ignored.
22 But clearly, the optimal layout position (x=5.25 and y=-2.18) is out of the region. The interaction effect on
23 the upstream turbine could be ignored. Under the condition, the optimal layout position can be extended.
24 Compared with past relative essays, multi-turbines layout is more closely spaced. It means that the number
25 of turbines increases. A 5-unit layout or a larger amount of units could be created, as shown in Fig. 22. Five
26 wind turbines are investigated at the same time to validate the optimal position and multi-turbines
27 interaction effect. The wake structure of downstream wind turbines is similar to that of the upstream
28 turbine.
29 Instantaneous CT of a single blade are compared in Fig. 23(a). CT of No.2,3,4,5 turbines are
30 significantly higher than the reference value. Average power coefficients are compared in Fig. 23(b).
31 Correspondingly, Cp of No. 2,3,4,5 turbines are higher than Cp of the upstream turbine. Cp of the upstream
32 turbine is close to reference value. Cp of No. 2,3,4,5 turbines are close to Cp of the downstream turbine at
33 the optimal relative position (predictive value of Cp is 0.30722).
34 Meanwhile, experimental data from practical project are added to validate the simulation results. The
35 experimental results fit well with the simulation results, as shown in Table 3. The optimal position is
36 suitable for multiple turbines in a large wind farm. This conclusion can be used to study a larger amount of
37 turbines in a large farm.
38

11
 ACCEPTED MANUSCRIPT

1 6. CONCLUSIONS
2 The wake effect of Savonius-type vertical axis wind turbines is investigated using CFD method. Based
3 on Kriging Method, the response surface model is established to define the relationship between
4 optimization objective Cp (power coefficient) and layout parameters. PSO algorithm is presented to obtain
5 the optimal position of the downstream turbine. Previous relative essays and results are committed to
6 reducing the interaction effect between different turbines. In other words, the wake of the upstream turbine
7 should not overlap with the downstream turbine. However, a new thought is put forward to reuse the wake
8 energy in the current study. The high speed zone in the wake created by the upstream turbine could be
9 reused to enhance the performance of downstream turbines. Therefore, wind energy is fully utilized.
10 Detailed conclusions can be summarized as follows:
11 1) The power coefficient Cp of the downstream turbine changes with the layout position. On the
12 returning blade side, the optimal layout position of the downstream turbine is x=4.91 and y=2.24.
13 Correspondingly, the optimal Cp is 0.2913. While on the advancing blade side, the optimal position is
14 x=5.25 and y=-2.18. The optimal Cp is 0.3044. The optimal layout on the advancing blade side is visibly
15 superior to that of on the returning blade side.
16 2) The optimal layout position is located in the periodic high speed zone. In this zone, the averaged
17 incoming flow speed for the downstream turbine increases obviously. The performance of the downstream
18 turbine is improved in a whole rotation period. It is noteworthy that, for a typical layout, in which the
19 returning blade located in low speed zone and the advancing blade located in the periodic high speed zone,
20 the performance of the downstream turbine is just improved in returning process.
21 3) When the distance between the upstream turbine and the downstream turbine is relatively small,
22 multi-turbines interaction effect cannot be ignored. The performance of the upstream turbine could be
23 largely affected by the downstream turbine. With the distance increasing, the Cp of the upstream wind
24 turbine increases. When the distance is longer than a certain distance, the interaction effect on the upstream
25 turbine could be neglected. In the current study, the optimal layout position exceeds the certain distance.
26 4) The optimal layout position of the downstream turbine could be applicable for multi-turbines
27 layout. As the number of downstream turbines increases, all power properties are improved. Cp of each
28 downstream turbine is close to predictive value (Cp=0.30722).
29 5) In practice, more turbines can be extended as shown in Fig. 22. Compared with past research,
30 multi-turbines layout can be more closely spaced. In equal surfaces, the number of turbines using the new
31 layout method is about 4 times larger than the number of turbines using conventional no overlap layout
32 method.
33 In this study, it is intended to provide a novel wake energy reuse ideology for the layout optimization
34 of Savonius-type vertical axis wind turbines in a large wind farm. Using this method, the wake energy
35 caused by the upstream turbine is not detrimental to downstream turbines. On the contrary, the wake energy
36 is used to enhance the performance of downstream turbine. Therefore, applying the wake energy reuse
37 method in VAWT layout optimization is worthy of research and promotion in the future.
38

12
 ACCEPTED MANUSCRIPT

1 ACKNOWLEDGEMENT
2 This research is supported by the National Science Foundation of China (Grant No. 61572404) and the
3 Science and Technology Nova Plan of Shaanxi Province (Grant No. 2016KJXX-57).
4
5

6 REFERENCES
7 [1] Zuo W, Wang X, Kang S. Numerical simulations on the wake effect of h-type vertical axis wind
8 turbines. Energy 2016;106: 691-700.
9 [2] Gao X, Yang H, Lu L. Optimization of wind turbine layout position in a wind farm using a newly-
10 developed two-dimensional wake model. Applied Energy 2016;174:192-200.
11 [3] Mo JO, Choudhry A, Arjomandi M, Lee YH. Large eddy simulation of the wind turbine wake
12 characteristics in the numerical wind tunnel model. Journal of Wind Engineering and Industrial
13 Aerodynamics 2013;112:11-24.
14 [4] Guerri O, Sakout A, Bouhadef K. Simulations of the fluid flow around a rotating vertical axis wind
15 turbine. Wind Engineering 2007;31(3): 149-163.
16 [5] Park J, Law K H. Layout optimization for maximizing wind farm power production using sequential
17 convex programming. Applied Energy 2015;151:320-334.
18 [6] Kacprzak K, Liskiewicz G, Sobczak K. Numerical investigation of conventional and modified
19 Savonius wind turbines. Renewable Energy 2013; 60(4):578-585.
20 [7] Whitehouse GR, Sibilia MJ, Bilanin AJ, et al. Variable geometry wind turbine for performance
21 enhancement, improved survivability and reduced cost of energy. Wind Energy 2015; 18(7): 1303-
22 1311.
23 [8] Altan, Deda B, Atılgan. An experimental study on improvement of a Savonius rotor performance with
24 curtaining. Experimental Thermal and Fluid Science 2008; 32(8):1673-1678.
25 [9] Tian W, Ozbay A, Hu H. Effects of incoming surface wind conditions on the wake characteristics and
26 dynamic wind loads acting on a wind turbine model. Physics of Fluids 2014; 26(12), 1073–1079.
27 [10] Leu TS, Yu JM, Hu CC, Miau JJ, Liang SY, Li JY. Experimental study of free stream turbulent effects
28 on dynamic stall of pitching airfoil by using particle image velocimetry. Applied Mechanics &
29 Materials 2012; 225:103-108.
30 [11] Rockel S, Peinke J, Hölling M, Cal RB. Wake to wake interaction of floating wind turbine models in
31 free pitch motion: an eddy viscosity and mixing length approach. Renewable Energy 2016; 85:666-
32 676.
33 [12] Chamorro LP, Portéagel F. Turbulent flow inside and above a wind farm: a wind-tunnel study.
34 Energies 2011; 4(12):1916-1936.
35 [13] Martin JD, Simpson TW. Use of kriging models to approximate deterministic computer models. AIAA
36 journal 2005; 43(4): 853-863.
37 [14] Eberhart RC, Kennedy J. A new optimizer using particle swarm theory. In Proceedings of the sixth
38 international symposium on micro machine and human science, 1995, October; Vol.1, pp. 39-43.

13
 ACCEPTED MANUSCRIPT

1 [15] Kacprzak K., Liskiewicz G, Sobczak K. Numerical investigation of conventional and modified
2 savonius wind turbines. Renewable Energy 2013; 60(4), 578-585.
3 [16] Kamoji MA., Kedare SB, Prabhu SV. Experimental investigations on single stage modified savonius
4 rotor. Applied Energy 2009; 86(7–8): 1064-1073.
5 [17] Wenlong T, Baowei S, Zhaoyong M. Conceptual design and numerical simulations of a vertical axis
6 water turbine used for underwater mooring platforms. International Journal of Naval Architecture and
7 Ocean Engineering 2013; 5(4): 625-634.
8 [18] Akwa JV, Vielmo HA, Petry AP. A review on the performance of Savonius wind turbines. Renewable
9 and Sustainable Energy Reviews 2012; 16(5): 3054-3064.
10 [19] Kacprzak K, Liskiewicz G, Sobczak K. Numerical investigation of conventional and modified
11 Savonius wind turbines. Renewable energy 2013; 60: 578-585.
12 [20] McTavish S, Feszty D, Sankar T. Steady and rotating computational fluid dynamics simulations of a
13 novel vertical axis wind turbine for small-scale power generation. Renewable energy 2012; 41: 171-
14 179.
15 [21] Altan BD, Atilgan M. An experimental and numerical study on the improvement of the performance
16 of savonius wind rotor. Energy Conversion and Management 2008; 49(12):3425-3432.
17 [22] Sheldahl RE, Feltz LV, Blackwell BF. Wind tunnel performance data for two-and three-bucket
18 Savonius rotors. Journal of Energy 1978; 2(3): 160-164.
19 [23] Sun C, Song B, Wang P. Parametric geometric model and shape optimization of an underwater glider
20 with blended-wing-body. International Journal of Naval Architecture and Ocean Engineering, 2015,
21 7(6): 995-1006.
22 [24] Chen HN, Zhu YL, Hu KY, et al. Global optimization based on hierarchical coevolution model.
23 Evolutionary Computation 2008. CEC 2008. (IEEE World Congress on Computational Intelligence).
24 IEEE Congress on. IEEE, 2008: 1497-1504.
25 [25] Jaohindy P, Ennamiri H, Garde F, et al. Numerical investigation of airflow through a Savonius rotor.
26 Wind Energy 2013; 17(6):2-18.
27 [26] Shigetomi A, Murai Y, Tasaka Y, Takeda Y. Interactive flow field around two savonius turbines.
28 Renewable Energy 2011; 36(2):536e45.

14
 ACCEPTED MANUSCRIPT

1
2 Fig. 1 (a) Migrating geese and (b) Wake of a single turbine
3 Fig. 2 (a) One side type layout and (b) V type layout
4 Fig. 3 (a) Savonius wind turbines and (b) Simplified 2D model
5 Fig. 4 Computational domains
6 Fig. 5 Grids of rotating domain and details
7 Fig. 6 (a) 3 mesh densities and (b) 4 time steps: Comparison of the torque coefficient in a rotation period
8 Fig. 7 Comparison of the average Cp at different TSR
9 Fig. 8 Velocity distributions of the wake
10 Fig. 9 Turbulence Kinetic Energy distributions of the wake
11 Fig.10 Comparison of (a) CT of a turbine, (b) CT of a single blade and (c) Average Cp
12 Fig. 11 The general procedure of PSO
13 Fig. 12 (a) Response surface and (b) Optimal point on the returning blade side
14 Fig. 13 (a) Response surface and (b) Optimal point on the advancing blade side
15 Fig. 14 Simplified wake structure
16 Fig. 15 Velocity distribution at a typical layout position
17 Fig. 16 The variations of instantaneous CT of (a)Blade 1, (b)Blade 2 and (c)Turbine
18 Fig. 17 Velocity distribution at the optimal layout position
19 Fig. 18 The variations of instantaneous CT of (a)Blade 1, (b)Blade 2 and (c)Turbine
20 Fig. 19 Pressure distributions around the downstream turbine at optimal layout position
21 Fig. 20 Comparison of pressure distributions on a single blade at different rotor angle position
22 Fig. 21 (a) Response surface of Cp for the upstream turbine and (b) Region of interaction effect
23 Fig. 22 Velocity distributions at optimal layout position for multi-turbines
24 Fig. 23 Comparison of (a) CT of a single blade and (b) average Cp for multi-turbines

25
26
27
Periodic high speed zone Returning blade side

High speed rotating vortices Type I position

Low speed zone

A

High speed rotating vortices

Type II position
Advancing blade side
Periodic high speed zone

28
29 Fig. 1 (a) Migrating geese and (b) Wake of a single turbine

15
 ACCEPTED MANUSCRIPT

Wind vane Wind vane

Turbine
Turbine

Generator
Generator

Locking bridge

Locking bridge
Yaw motor Yaw motor

o
o

o o

o o

o
o

1 (a) (b)

2 Fig. 2 (a) One side type layout and (b) V type layout
3
4
5

Returning blade

d
v r

H s0 o D

Wind

Advancing blade

6 (a) Savonius turbines with endplates (b) Classical Savonius

7 Fig. 3 (a) Savonius wind turbines and (b) Simplified 2D model

8
9
10

16
 ACCEPTED MANUSCRIPT

y Wall
Stationary Domain
xC o
Central axis
Rotating C
Balde1
Velocity-inlet yC

Pressure-outlet
Balde2

o x
Interface yB
A
Rotating Domain
xB
o

B
Wall
1
2 Fig. 4 Computational domains
3
4
5

Local mesh

Rrotating

6
7 Fig. 5 Grids of rotating domain and details
8
9
10

0.6 61000grids 0.6 0.5°

86900grids 1.0°
0.5 117000grids 0.5 2.0°
4.0°
0.4 0.4
Torque coefficient

Torque coefficient

0.3
0.3

0.2
0.2

0.1
0.1
0
0
-0.1
-0.1
0 50 100 150 200 250 300 350 0 50 100 150 200 250 300 350
Rotor angle (degree) Rotor angle (degree)

11 (a) 3 mesh densities (b) 4 simulated time steps

12 Fig. 6 (a) 3 mesh densities and (b) 4 time steps: Comparison of the torque coefficient in a rotation period

17
 ACCEPTED MANUSCRIPT

0.26
Experimental Data
0.25 Simulation Data

0.24

Average power coefficient

0.23

0.22

0.21

0.2

0.19

0.18
0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1.1 1.2 1.3
Tip speed ratio
1
2 Fig. 7 Comparison of the average Cp at different TSR
3
4
5

6
7 Fig. 8 Velocity distributions of the wake
8
9
10

18
 ACCEPTED MANUSCRIPT

1
2 Fig. 9 Turbulence Kinetic Energy distributions of the wake
3
4
5
1.5 1

A position Blade1
Blade2
A position
1 B position
All
C position
Torque coefficient

0.5
0.5

Torque coefficient
0

-0.5

-1 0
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
Rotor angle (degree)

1.5

B position Blade1
Blade2
1 All -0.5
0 50 100 150 200 250 300 350
Torque coefficient

Rotor angle (degree)

0.5
(b)
0
0.35
-0.5
0.3 0.2823
-1
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
0.2354 Ref. value
Average power coefficient

Rotor angle (degree) 0.25

1.5
Blade1

1 C position Blade2
All
0.2
0.1627
Torque coefficient

0.5 0.15

0 0.1

-0.5 0.05

-1 0
0 500 1000 1500 2000 2500 3000
Rotor angle (degree)
3500 4000 4500 5000
A
1 B
2 C3
Position

6 (a) (c)

7 Fig.10 Comparison of (a) CT of a turbine, (b) CT of a single blade and (c) Average Cp

19
 ACCEPTED MANUSCRIPT

Start

Initialize particles for each particle

Numerical method using

Ansys Fluent provides
foundation elements for the 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

1
2 Fig. 11 The general procedure of PSO
3
4
5
6

0. 25 0.25
0.26 0.26
3 0.25
0.26
Optimal layout point
0.27
7
0.2 8
2.5 0.2 0.
0.26 28
0.2

5 0.2
0.2 28
29

0.2 6
85

6 0. 7
0.

2
y (m)

0.2
85 8
0.2
0.26

0. 0.
0.26 0.27

28 0.28
0.2

27
1.5
7

0. 27
0.26 0. 25
6
0.2 0.2
6
0.27 0.2 0.25 5
1 5 0. 25 0.1
0.2 0.2
0.15 0.1
0.2
0.15 0.05
0. 1 0.1
0.5 0.05 0.05 0
0
2 3 4 5 6 7 8
x (m)
7 (a) (b)

8 Fig. 12 (a) Response surface and (b) Optimal point on the returning blade side
9
10
11

20
 ACCEPTED MANUSCRIPT

0. 05 0.05
-0.5 5
0.1 0.1 0.0
0. 15
0.2 0. 15 0.1
0.25 0.2
-1 0.27 0.25 0.15
0.28
0.2
0. 27 0. 29 0.2
7 0. 25
0.28 0.
0.2 27
95 0.29

0.28
-1.5
0. 3 29

28
0.

0.28
0.
0.3

y (m)

95
0.3
0.303

0. 29
0.2
-2
3
03 0.

0.
0.3

30

5
0.306

29
3

0.
0.28
-2.5
0. 3

29
0.2
8 0.295

0.
0.29
Optimal layout point
-3 0.27 0.28
0. 28
0. 25 0.27
0.2 0.25
7
2 3 4 5 6 7 8
x (m)

1 (a) (b)

2 Fig. 13 (a) Response surface and (b) Optimal point on the advancing blade side
3
4
5

o
Secondary-periodic high speed zone
C

o Low speed zone

Periodic high speed zone

6
7 Fig. 14 Simplified wake structure
8
9
10

21
 ACCEPTED MANUSCRIPT

4m

Low speed wake

-0.8m
High speed vortex

1
2 Fig. 15 Velocity distribution at a typical layout position
3
4
5
1 1 0.7
B position B position B position
A rotation period A rotation period A rotation period
Reference Reference 0.6 Reference
down
0.5

0.5 0.5 0.4

Torque coefficient

Torque coefficient

Torque coefficient

0.3

0.2

0 0 0.1

-0.1
up up
-0.5 -0.5 -0.2
0 200 400 600 800 1000 0 200 400 600 800 1000 0 200 400 600 800 1000
Rotor angle (degree) Rotor angle (degree) Rotor angle (degree)

6 (a) (b) (c)

7 Fig. 16 The variations of instantaneous CT of (a)Blade 1, (b)Blade 2 and (c)Turbine

8
9
10
11
-2.18m

5.25m

12
13 Fig. 17 Velocity distribution at the optimal layout position

22
 ACCEPTED MANUSCRIPT
1 1 0.8
A rotation period B position A rotation period B position A rotation period B position
Reference Reference 0.7 Reference

up up 0.6

0.5 0.5 0.5 up

Torque coefficient
Torque coefficient

Torque coefficient
0.4

0.3

0 0 0.2

0.1

-0.5 -0.5 -0.1

0 200 400 600 800 1000 0 200 400 600 800 1000 0 200 400 600 800 1000
Rotor angle (degree) Rotor angle (degree) Rotor angle (degree)

1 (a) (b) (c)

2 Fig. 18 The variations of instantaneous CT of (a)Blade 1, (b)Blade 2 and (c)Turbine

3
4
5
6

High pressure zone

High pressure zone

Low pressure zone

Low pressure zone

(a) 0° (b) 45°

High pressure zone

Great low pressure zone

Great low pressure zone

(c) 90° (d) 135°

7
8 Fig. 19 Pressure distributions around the downstream turbine at optimal layout position

23
 ACCEPTED MANUSCRIPT
0° 45° 90°
250
V-Ref V-Ref
200 C-Ref
100 C-Ref 100
All-Ref All-Ref
150
V-Op V-Op
C-Op 100 C-Op
50 50
All-Op All-Op

Pressure (pascal)

Pressure (pascal)
Pressure (pascal)

50

0 0 0

-50
V-Ref
-50 -50
C-Ref -100
All-Ref
-150
V-Op
-100 -100
C-Op -200
All-Op
-250
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
Curve length （ m（ Curve length （ m（ Curve length （ m（
(a)135°
0° (b) 45°
180°
(c)270°90°
40
V-Ref V-Ref 60 V-Ref
30 10
C-Ref C-Ref C-Ref
All-Ref All-Ref 40 All-Ref
20
V-Op 0 V-Op V-Op
10 C-Op C-Op 20 C-Op
All-Op All-Op All-Op

Pressure (pascal)
Pressure (pascal)

Pressure (pascal)

0 -10
0

-10
-20 -20
-20
-40
-30 -30
-60
-40
-40
-80
-50

-60 -50 -100

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
Curve length （ m（ Curve length （ m（ Curve length （ m（

(d) 135°
1 (e) 180° (f) 270°

2 Fig. 20 Comparison of pressure distributions on a single blade at different rotor angle position
3
4
5
6
-0.4
0.235
0.0.
22225
-0.6 0.24

0.245
0.2

-0.8
0.247
3

0.22
4
0.2

-1 0.225
0.23 0.235
-1.2 0.235
0.24 0.24
-1.4
y (m)

0.245
0.247 45
0.2
-1.6 0.24 4 7
5 0. 2
-1.8 0.2
47
-2

-2.2

-2.4 Optimal layout point

2 3 4 5 6 7 8
x (m)

7 （a） （b）
8 Fig. 21 (a) Response surface of Cp for the upstream turbine and (b) Region of interaction effect

24
 ACCEPTED MANUSCRIPT

Ref
-2.18m

2 E xt e n
d mo
re t ur
5.25m bi ne s

-2.18m
3
5.25m

-2.18m
4
Extend more turbines
5.25m

-2.18m
5
5.25m

1
2 Fig. 22 Velocity distributions at optimal layout position for multi-turbines
3
4
5
6
0.35
Ref
1
2 0.3
Predicted value
3
0.8
4
Ref. value
Average power coefficient

5 0.25
0.6
Torque coefficient

0.4 0.2

0.2
0.15

0
0.1
-0.2

0.05
-0.4

0
0 50 100 150 200 250 300 350 Ref
1 2 3 4 5

7 Rotor angle (degree) Position

8 Fig. 23 Comparison of (a) CT of a single blade and (b) average Cp for multi-turbines

9
10
11

25
 ACCEPTED MANUSCRIPT

1
2 Table 1 Main geometrical parameters of turbine

D (m) H (m) d (m) S0 (m) r (m)

Savonius 0.909 1.0 0.5 0.91 0.25

3
4
5
6
7 Table 2 Comparison of main performance parameters

Optimal layout Predictive value of Cp Experiment value of Cp Optimization effect

Ref. value -- -- 0.2477 --

Returning side x=4.91; y=2.24 0.29116 0.2913 17.60%
Advancing side x=5.25; y=-2.18 0.30722 0.3044 22.89%

8
9
10
11
12
13 Table 3 Comparison between experimental results and simulation results

Ref #2 #3 #4 #5

Experiment 0.239 0.281 0.303 0.293 0.290

Simulation 0.24324 0.29033 0.30597 0. 29947 0. 30164

14

26