RESEARCH ARTICLE | NOVEMBER 01 2023

Improving Savonius turbine efficiency with splitter and

barrier cylinder deflector design: A Taguchi method study 
Mohamad Hafizul Fikri Mahizam ; Wei Shyang Chang (钟维翔)  ; Esmaeel Fatahian ;
Farzad Ismail ; Mohammad Hafifi Hafiz Ishak

Physics of Fluids 35, 115102 (2023)

https://doi.org/10.1063/5.0162026

CrossMark

 
View Export
Online Citation

02 December 2023 10:07:44

 Physics of Fluids ARTICLE pubs.aip.org/aip/pof

Improving Savonius turbine efficiency with splitter

and barrier cylinder deflector design: A Taguchi
method study
Cite as: Phys. Fluids 35, 115102 (2023); doi: 10.1063/5.0162026
Submitted: 13 June 2023 . Accepted: 10 October 2023 .
Published Online: 1 November 2023

Mohamad Hafizul Fikri Mahizam, Wei Shyang Chang (钟维翔),a) Esmaeel Fatahian, Farzad Ismail,
and Mohammad Hafifi Hafiz Ishak

AFFILIATIONS
School of Aerospace Engineering, Universiti Sains Malaysia, Nibong Tebal, Penang 14300, Malaysia

a)
Author to whom correspondence should be addressed: [email protected]

ABSTRACT
Vertical axis wind turbines are wind turbines with the rotor axis perpendicular to the ground, designed to harness wind energy for electricity
generation. The primary cause of low-efficiency Savonius turbines is the negative torque contribution from the returning blade. A Savonius

02 December 2023 10:07:44

turbine is a type of turbine characterized by its rotational direction orthogonal to the passing fluid flow and the flow interaction with advanc-
ing and returning blades, generating torque primarily through drag forces. Thus, the study proposes a novel design of a cylindrical deflector
with splitters and a barrier to offset the flow field to the returning blades. This study aims to maximize the efficiency of conventional
Savonius turbines using the cylindrical deflector with splitters and a barrier via a combination of computational fluid dynamics simulations
and the Taguchi optimization method. The Taguchi method is used to determine the best combination of specified characteristics such as the
length of the barrier (Ls/D), the barrier attachment angle (a), and the geometric shape of the cylinder deflector with a splitter and a barrier.
The tip speed ratio (k) for this study is fixed at k ¼ 1. The simulations and additive model revealed that the optimal combination in this study
is a cylinder deflector with double wake splitters positioned at the top and middle of the deflector. The deflector is oriented parallel to the
flow, and a barrier placed at the bottom of the deflector is oriented orthogonal to the flow. The optimal configuration has Ls/D ratio of 0.9
and a barrier attachment angle (a) of 10 . Therefore, the ideal combination was found to produce a power coefficient equivalent to 0.459,
indicating that the performance of Savonius turbines increases by around 61% compared to previous studies.
Published under an exclusive license by AIP Publishing. https://doi.org/10.1063/5.0162026

NOMENCLATURE yþ dimensionless wall distance (-)

Dt time-step (s)
A area swept by turbine (m2) h angle associated to time-step (deg)
Cm turbine torque coefficient (-) k tip speed ratio (-)
Cp turbine power coefficient (-) l air viscosity (kg/ms)
d blade diameter (m) q air density (kg/m3)
D rotor diameter (m) x angular velocity of rotor (rad/s)
DR rotating domain diameter (m)
Mt torque (N m)
O location of coordinate system (-) I. INTRODUCTION
P air pressure (Pa) Global energy consumption has significantly increased over the
Pt power (W) past decade, largely due to population growth as well as enhanced
R rotor radius (m) comfort in daily life. However, the widespread use of fossil fuels for
Re Reynolds number (-) power and electricity production has led to major environmental
s overlap (m) issues.1 Creating alternative renewable energy sources is crucial to
U1 flow inlet velocity (m/s) ensuring a sustainable future and advancing eco-friendly technology.

Phys. Fluids 35, 115102 (2023); doi: 10.1063/5.0162026 35, 115102-1

Published under an exclusive license by AIP Publishing
 Physics of Fluids ARTICLE pubs.aip.org/aip/pof

These sources will help mitigate environmental issues such as global Researchers suggest enhancing Savonius turbine power output by
warming, excessive CO2 emissions, and unusual weather patterns. Wind adding splitters, deflectors, guiding vanes, or concentrators. The
energy is a prominent example of a renewable energy source that is returning blades cause negative torque, thereby reducing efficiency.
clean, free, and readily available. Utilizing wind energy can reduce Improving air flow, especially for returning blades, boosts perfor-
dependence on fossil fuel resources.2 Vertical axis wind turbines mance. A direct method uses a flat plate deflector to divert wind,
(VAWTs) are particularly attractive due to their capability to capture reduce negative torque, and enhance efficiency. In the analysis by
and harness wind energy from all directions, compact design, and effi- Zhao et al.,16 the addition of deflectors increased the maximum torque
cient operation in various wind conditions.3 The main rotor shaft of a coefficient by 23%, and its installation also led to varied wake recovery
VAWT is oriented transverse to the wind, and the main parts are situ- effects downstream compared to situations without a deflector.
ated near the base of the turbine for the ease of service and repair. Mohamed et al.17 employed obstacle shielding plate to decrease coun-
Two types of VAWT designs are the drag-type Savonius rotor ter motion in the returning blade of the Savonius rotor. This alteration
and the lift-type Darrieus wind turbines. Both turbines have advan- yielded a substantial 38.9% enhancement in power coefficient com-
tages and drawbacks. For example, a Darrieus rotor turbine has some pared to the conventional design. Layeghmand et al. improved the
issues with dynamic stall and flow wake deformations in 3D aerody- power coefficient of the Savonius VAWT through the application of
namics.4 The performance of the Darrieus turbine is heavily reliant on computational fluid dynamics (CFD) simulations.18 They discovered
the incoming aerodynamic wake but creates a substantially lower wake that an appropriate position and angle value were required to achieve
magnitude than the Savonius turbine.5 By contrast, the Savonius rotor the highest power and torque coefficients, respectively. Overall, the
features a simple build, functions well at low wind speeds, and pos- addition of various types of deflectors has proven to improve the effi-
sesses natural self-starting capabilities.5,6 Thus, the Savonius rotor is ciency of Savonius wind turbines through various methods.
ideal for use in urban environments where installation of VAWTs is The integration of a wind guide as a power augmentation tool for
suitable on building sites and rooftops.7 the conventional Savonius VAWT is crucial to enhancing its power
When the airflow of the wind passing through an object is accel- output efficiency. This phenomenon has recently gained the attention
erating, the gradient of wind speed is usually greater than the natural of scientific researchers. However, the deployment of solid deflectors
wind shear,8 and the velocity of the wind on the rooftop is substantially poses a significant disadvantage due to the intense turbulence and
increased. In urban areas with substantial terrain roughness and wake region generated behind them. These factors markedly influence
obstacles, the wind speed close to the ground is generally slower than flow dynamics, particularly near the returning blade. When flat plate

02 December 2023 10:07:44

at rooftop levels. Thus, the Savonius VAWT that exhibits superior self- deflectors are employed, large vortices form behind them, producing
starting capabilities9 even at low wind speeds is a better alternative for flow instability that impacts the returning blade. This instability causes
these circumstances as opposed to the Darrieus VAWT. In addition to cyclic load variations, leading to system fatigue. A strategy to mitigate
the self-starting capability problem, including the dynamic stall phe- these effects involves dissipating the wake zone downstream of the
nomenon and generated drag, the development of Darrieus VAWTs Savonius VAWT, reducing vortex formation behind the deflector and
faces other challenges.10 weakening turbulent flow. A few studies have explored the effects of
Based on previous research, the Savonius VAWT has undergone employing a cylindrical deflector in this context to enhance the turbine
considerable improvements to increase its effectiveness and perfor- efficiency.
mance, particularly in terms of torque and power coefficients. Mat Researchers have recently explored the use of a rotating cylinder
Yazik et al.11 conducted a study that utilized 3D computer simulations as a deflector to improve the performance of Savonius VAWT. In a
to investigate the effects of surface roughness and blade material on a study by Fatahian et al.,19 computational fluid dynamics (CFD)
stationary Savonius turbine under various conditions. The researchers assessed the impact of factors such as distance, cylinder diameter, tor-
suggested that selecting an appropriate turbine material based on its que, angular velocity, power coefficients, and wake zone on Savonius
properties could potentially reduce downtime and expenses. Al Absi rotor performance. Their findings showed that a rotating cylinder
et al.12 proposed enhancing the power coefficient performance of a deflector improved efficiency across all tip speed ratios (k), with a
Savonius elliptical turbine by optimizing the overlap ratio and blade notable 50% increase at k ¼ 0.6 compared to a static deflector.
design. They concluded that the turbine with a zigzag shape inner sur- Subsequently, Fatahian et al.20 introduced an innovative cylinder
face of the concave blade demonstrated the most significant improve- deflector design with downstream wake splitters to boost Savonius
ment in output performance. Furthermore, recent research by Hossein rotor performance. They explored the impact of wake splitter varia-
et al. proposed a novel approach involving a simple modification to tions on torque, power coefficients, and wake zone. Results showed
the turbine blade design that can significantly improve power effi- that such deflectors significantly improved rotor performance, espe-
ciency.13 Results showed that the modified blade design improved cially for k of 0.6 and lower. Using a deflector with two wake splitters
Savonius VAWT power performance by approximately 18%. Al- increased average torque and power coefficients by up to 15% com-
Ghriybah et al. examined the impact of blade spacing inside turbines pared to cases without deflectors. Abdullah et al.21 utilized the concept
on the effectiveness of conventional Savonius rotors featuring both of improvising a Savonius turbine using a rotating cylinder synchro-
inside turbine blades and found 32.9% total improvement over con- nized with a rotating turbine via a timing belt mechanism that is
ventional turbine with one of their configurations.14 Talukdar et al. placed parallel to the turbine position. They found that by using opti-
investigated the performance assessment of Savonius wind (SWT) and mum device parameters, the power coefficient improved by up to
hydrokinetic turbines (SHT) with similar power input.15 Experimental 55.6% compared to the case without the device at a k of 0.9. Setiawan
investigations showed that two-bladed SWTs yield an improved maxi- et al.22 quantitatively analyzes the impact of Savonius vertical axis
mum power coefficient of 47% compared with three-bladed SWTs. hydroturbine performance on the stagger angle change of the circular

Phys. Fluids 35, 115102 (2023); doi: 10.1063/5.0162026 35, 115102-2

Published under an exclusive license by AIP Publishing
 Physics of Fluids ARTICLE pubs.aip.org/aip/pof

cylinder situated in front of the returning blade. The findings reveal that II. GEOMETRICAL MODELS
the highest power coefficient rises by roughly 29.8% for a k of 0.9 for a Numerous researchers have previously investigated a variety of
60 stagger angle. Sewucipto et al.23 experimentally and numerically approaches to determine the optimal parameters for running a
investigated the effects of placing a D-53 type of cylinder facing the Savonius wind turbine. A direct and affordable alternative is to utilize
direction of the returning blades on the Savonius wind turbine perfor- any shape of deflector to improve the overall performance of the drag-
mance. The D-53 type cylinder is a circular cylinder that is sliced only type Savonius rotor, as suggested in earlier studies.2,3,24 Using a specific
on the front side of the cylinder at a cutting angle of 53 . They concluded deflector or barrier to cover the returning blade side could be a way to
that the greatest power coefficient has increased by roughly 24.6%. improve the power coefficient and decrease the negative torque due to
Earlier studies conducted investigations using a flat plate deflector the returning blade. Figure 1 illustrates the evolution of Savonius tur-
as a barrier upstream of wind turbines placed orthogonal to the flow to bine research, showcasing the advancements made to enhance the
improve turbine performance by lessening the negative impact of return- power coefficient compared to (a) the traditional Savonius turbine.
ing blades and guiding the flow toward the advancing blades. Wong Fatahian et al.19 proposed two significant modifications: (b) a static
et al.24 proposed a study of the effect of a flat plate deflector orthogonal cylindrical deflector placed in front of the returning blade of the tur-
to the flow placement at the upwind of the VAWT. They found that the bine, and (c) a rotating cylindrical deflector at the same location to fur-
utilization of a deflector resulted in a significant increase in the averaged ther optimize its performance. Building upon these improvements,
torque coefficient, which was approximately 47.1% higher compared to Fatahian et al.20 continued their research by incorporating (d) two
the VAWT without the deflector. In their empirical research on the wake splitters downstream of the static cylindrical deflector. This
effects of deflector angles on Savonius turbines for hydrokinetic applica- approach aims to enhance turbine efficiency even further. Consistent
tions in small rivers, Salleh et al.25 observed a remarkable improvement with this study, this present investigation aims to advance the power
in the maximum power coefficient. They found a 61.5% increase in the coefficient of the Savonius turbine by introducing (e) a barrier plate
maximum power coefficient compared to the case without the deflector. positioned at the bottom of the static cylindrical deflector, accompa-
They also mentioned that the returning blade deflector angle, which was nied by two wake splitters downstream of the wind direction. The
set at 90 , had a more pronounced impact on turbine power compared wind direction is traveling from left to right.
to changes in the advancing blade deflector angle. Figure 2(a) demonstrates the geometric features of a Savonius
Most of the augmentation strategies are based on a single form of
VAWT that rotates in the positive rotation direction. Figure 2(b)
flow alteration device with significant improvements in turbine perfor-
depicts a schematic of a Savonius VAWT with the addition of a cylin-

02 December 2023 10:07:44

mance. The innovations of this study are highlighted as follows:
drical deflector with splitter and barrier located in front of the return-
• Combination of two or more downstream flow-altering and flow- ing blade of the turbine for improved performance. Meanwhile,
blockage devices. Fig. 2(c) represents the details of the barrier attachment angle on the
• Application of optimization methods to find the optimal design bottom of the cylinder deflector with a wake splitter. In this study, the
or configuration for these devices. inclusion of a splitter wake on the cylinder deflector is crucial, that is,
splitting the flow into distinct paths. By strategically incorporating the
Thus, the aims of this study are formulated as follows:
splitter wake, this research aims to manipulate the fluid dynamics to
• To investigate the effect of combining these devices on the achieve a highly controlled and regulated flow distribution. By con-
Savonius turbine’s performance. trast, the barrier plate positioned on the bottom of the cylinder plays a
• To determine the optimal configuration of the combined devices distinct role within the simulations. The purpose of this plate is specifi-
in terms of the highest power coefficient with respect to three cally defined to impede or block a particular flow component, as pre-
parameters (barrier length, barrier angle attachment, and place- determined by the experimental parameters. This deliberate
ment of splitters) using the Taguchi method. obstruction serves as an investigative tool to study the influence of the

FIG. 1. The evolution of the deflectors

configuration for Savonius VAWT turbine
research.19,20,26

Phys. Fluids 35, 115102 (2023); doi: 10.1063/5.0162026 35, 115102-3

Published under an exclusive license by AIP Publishing
 Physics of Fluids ARTICLE pubs.aip.org/aip/pof

TABLE I. Geometrical values as described in Fig. 2 considered in the study.

Parameters Values

Axis inlet velocity, U1 7 (m/s)

Rotor diameter (D) 0.909 (m)
Blade diameter ratio (d/D) 0.5
Rotating domain diameter (DR/D) 1.1
Deflector distance from rotor center in 1.4
x-direction (Lx/D)
Deflector distance from rotor center in 0.5
y-direction (Ly/D)
Cylinder deflector diameter ratio (dc/D) 0.75
Barrier length ratio (Ls/D) [0.5, 0.7, 0.9]
Barrier attachment angle (a) [0, 10, 20]
Splitter length ratio (L/D) 0.3
Overlap ratio (s/D) 0.1
Blades thickness ratio (t/D) 0.011

III. COMPUTATIONAL MODEL

A. Computational domain and grids
Figure 3 illustrates the computational domain and boundary con-
ditions utilized in this study. The cylinder deflector with wake splitter
and barrier in Fig. 3 is one of the example deflectors used in the simu-

02 December 2023 10:07:44

lations. In CFD, the domain refers to the physical space that is mod-
eled and analyzed numerically, including the fluid volume and solid
boundaries. The boundary conditions specify the prescribed values of
fluid flow parameters or properties at the boundaries of the computa-
tional domain.
For this study, the specified computational domain of the geome-
try is divided into two subdomains: a rotating zone containing turbine
blades with a diameter of D and an extensive stationary zone enclosing
the rotating zone. An interface is defined to separate the two subdo-
mains, allowing the rotating zone to rotate clockwise at the origin axis
at a specific velocity while keeping the mesh of the stationary zone
static. The turbine rotor is positioned along the X axis using the sym-
FIG. 2. Geometrical details of (a) Savonius turbine, (b) static cylinder deflector with metry centerline, and a cylinder barrier with two splitter plates is
splitter and barrier, (c) barrier attachment angle. mounted upstream of the returning blade. The lower and upper

restricted flow on various aspects, such as turbulence characteristics,

pressure distribution, or vortex shedding patterns. This study aims to
gain valuable insight into the overall system behavior and its implica-
tions for practical applications by isolating and studying the effects of
this obstructed flow. The rotor diameter in the present case is 0.909 m,
and the dimensions of the cylinder deflector include a diameter of
0.75/D, a horizontal of 1.4/D, and vertical distances of 0.5/D from the
turbine. These parameters were selected on the basis of the optimal
values reported in previous studies.24,27 Several parameters were incor-
porated to determine the optimal configuration, including the bottom
wake splitter length (Ls/D). Moreover, Table I presents the geometrical
and characteristic values of the Savonius VAWT. The characteristic
and dimensional measurements of a simulated Savonius VAWT were
acquired on the basis of the experimental configuration from Sheldahl
et al.26 and the numerical study by Tian et al.28 FIG. 3. Computational domain in CFD with the locations of boundary conditions.

Phys. Fluids 35, 115102 (2023); doi: 10.1063/5.0162026 35, 115102-4

Published under an exclusive license by AIP Publishing
 Physics of Fluids ARTICLE pubs.aip.org/aip/pof

boundaries from the center of the turbine are set at 8D apart. The TABLE II. Quantitative grid details.
upstream boundary is set as 8D from the turbine, while the down-
stream boundary is set to be 16D from the rotor center. Mesh density Coarse Medium Fine
For material properties in the current study, air with constant vis-
Cell numbers 65 000 140 000 200 000
cosity (l ¼ 1.7894  105 kg/ms) and constant density (q ¼ 1.225 kg/
m3 ) is used in the fluid domain. The constant air velocity of 7 m/s is used Cell number in rotating 15 000 30 000 45 000
as an inlet velocity with a Reynolds number of 4.32  105 . Furthermore, zone
the specified turbulence intensity is fixed at 1%,28 and the outlet bound- Cell number in 50 000 110 000 155 000
ary condition is assumed to be atmospheric pressure (pressure outlet). stationary zone
Advancing and returning blades are set as a no-slip boundary condition. Inflation number 10 20 25
The symmetrical boundary condition is applied to the lower and upper Inflation growth factor 1.1 1.1 1.1
boundaries to ensure that the velocity perpendicular to the boundary is Normalized first layer 4.4  104 D 1.1  104 D 5.5  105 D
nullified.29 The sliding mesh method is chosen to observe the rotational thickness
movement of the rotor. The generation of extensive domain grids Average yþ 1.7 0.5 0.3
involves primarily the utilization of structured grids for the outer station- Average orthogonal 0.78 0.83 0.84
ary and the inner rotating subdomains to minimize the degradation of quality
the computational results due to an unstructured mesh when using the Average skewness 0.275 0.264 0.261
finite volume techniques.30 The rotational region of the turbine utilizes Average aspect ratio 7.90 7.63 7.62
finer meshes compared to the stationary region, as shown in Fig. 4(a).
Furthermore, inflation layers are employed in conjunction with the blades
to carefully determine the effect of the boundary layer on their perfor- computational cost. As shown in Fig. 4, a hybrid mesh is used with
mance. A total of 15 inflations on the surfaces of the blades are conducted three distinct grid resolutions for the grid independence test. The ini-
with a growth ratio of 1.1 to resolve the boundary layer close to the rotor tial grid is the fine grid with 200 000 components, followed by the
walls. As required by the k-SST (shear stress transport) model, the thick- medium grid with 140 000 elements and finally the coarse grid with
ness of the first layer cells is precisely computed, revealing a yþ value less 65 000 elements. Figure 5 shows that the Cm for the turbine is plotted
than 1.20,31 Table II provides a summary of the grid characteristics for the

02 December 2023 10:07:44

against the rotational angle across ten iterations, and the Cm curves
Savonius turbine, as recommended by best practices for grids.32 converge after five complete rotations. Furthermore, less than 0.5% of
the average Cm is changed throughout the next several rotation cycles.
B. Grid and time independence test Therefore, the tenth cycle may be used to gauge the performance of a
A grid independence test is used to determine the optimal mesh Savonius turbine rotor.
size that maximizes accuracy and reliability while optimizing the The coefficient of torque is expressed in terms of rotation angle,
as utilized in the grid independence study, as shown in Fig. 6. This
finding shows that while the outputs of the fine grid and the medium
mesh agree rather well, the outcomes of the coarse grid markedly differ
from each other. Thus, the medium mesh is the best option to be uti-
lized in the present study. The rotor performance can be predicted

FIG. 4. Mesh of the computational domain: (a) whole domain, (b) rotating domain,
(c) near-wall cell layer around the turbine blade. FIG. 5. Variations of torque coefficient for ten computation cycles.

Phys. Fluids 35, 115102 (2023); doi: 10.1063/5.0162026 35, 115102-5

Published under an exclusive license by AIP Publishing
 Physics of Fluids ARTICLE pubs.aip.org/aip/pof

1
M ¼ Cm qARU02 ; (1)
2
where the variables M and Cm represent the torque and torque coeffi-
cient, respectively; q denotes the air density; and U0 stands for the
wind speed. Additionally, A and R indicate the swept area and radius
of the wind turbine, respectively. The torque coefficient is interlinked
with the power coefficient, which can be expressed as follows:
Cp ¼ kCm ; (2)
where k is the tip speed ratio, which is defined as
Rx
k¼ ; (3)
U0
where is x the angular velocity of the rotor.
The power coefficient using Eq. (4)33 represents the efficiency of
FIG. 6. Comparison of Cm for three grids with different mesh densities for grid inde- the wind turbine in converting the available power in the airflow into
pendence study. electrical power and is calculated as the ratio of the power generated
by the wind turbine to the power available in the airflow.
with high accuracy using a grid resolution of 140 000, which might be Mathematically, this efficiency can be expressed as follows:
employed in future numerical models to drastically reduce the compu- P
tation time. The result is derived from simulations of unsteady flows. Cp ¼ ; (4)
1
Thus, the time-averaged value is calculated by averaging the perfor- qAU03
2
mance coefficients over many cycles. After completing each CFD anal-
ysis for ten rotation cycles, the coefficients of the final cycle are utilized where P corresponds to the turbine usable power.
in the current study. The convergence analysis utilizes a CFD simula-

02 December 2023 10:07:44

tion of a Savonius rotor operating at k ¼ 1. Numerous time steps, rang- D. Governing equations
ing from 0.5 /time step to 2 /time step, are used to examine the
Modeling the aerodynamics of VAWTs is a challenging task due
reliance on time steps. The instantaneous Cm is then considered, as
to the complex flow patterns around rotating turbines. The large-eddy
shown in Fig. 7. The Cm remains nearly unchanged after four-time
simulation (LES) has been shown to produce accurate results.
steps. The CFD simulations use the time step that is equivalent to 1 /
However, the LES is computationally expensive compared to
time step to increase the simulation accuracy.
Reynolds-averaged Navier–Stokes (RANS) turbulence models, which
are commonly used in simulating the aerodynamics of VAWTs due to
C. Performance parameters their low computational cost. Researchers have proposed various
After the fluid flow around the Savonius VAWT is simulated, the methods to improve the performance of Savonius VAWTs using stan-
torque generated by the wind turbine blades can be calculated using dard CFD simulation procedures. In this study, the incompressible
the following equation:33 URANS equations with the k–x SST turbulence model, which are
commonly employed in many types of CFD simulations, are utilized.
The k–x SST turbulence model is the most suitable for predicting the
entire performance of VAWTs due to its low computational cost, as
supported by several studies.31,34,35 The governing equations of mass
and momentum conservation in this simulation are expressed, respec-
tively, as follows:
@
ðui Þ ¼ 0; (5)
@xi
!
@ @ @P @ @ui @  
q ð ui Þ þ q ðui uj Þ ¼  þl þ qu0 j u0 i ;
@t @xj @xi @xj @xj @xj
(6)
where qu0 u0
j i indicates the Reynolds stress. ui denotes the air velocity
component, P is pressure, and l is the air dynamic viscosity.
The Reynolds stresses can be determined by the gradient trans-
port model by computing the turbulent viscosity via the k–x SST tur-
FIG. 7. Comparison of Cm for three time step sizes for time step independence bulence model, which was originally proposed by Menter.36 This
test. model is mathematically expressed as follows:

Phys. Fluids 35, 115102 (2023); doi: 10.1063/5.0162026 35, 115102-6

Published under an exclusive license by AIP Publishing
 Physics of Fluids ARTICLE pubs.aip.org/aip/pof

" #
@ @ @ @k E. Taguchi method
ðqkÞ þ ðqkui Þ ¼ Ck þ Gk  Yk ; (7)
@t @xi @xj @xj The power coefficient production of a VAWT system can be ana-
" # lyzed and optimized by using the Taguchi method.39 This methodol-
@ @ @ @x ogy includes the following three steps.21,40 The best design parameters
ðqxÞ þ ðqxui Þ ¼ Cx þ Gx  Yx þ Dx : (8) are found by analyzing the signal-to-noise ratio (S/N ratio) of test data,
@t @xi @xj @xj
which is conducted by first defining the objective function, then find-
Within the equations of motion, the terms Uk and Ux signify the ing the right set of orthogonal arrays based on factors and levels, per-
diffusivity of k and x, respectively. Gk and Gx indicate the production forming the fewest number of experiments, and finally using the test
of k and x due to mean velocity gradients, respectively. Furthermore, data to make those determinations. The S/N ratio can be used to exam-
Yk and Yx represent the dissipation of k and x, respectively, and Dx ine the quality features of product or process parameters, and the
denotes the cross-diffusion term. approach is characterized by an orthogonal array comprising factors
The numerical simulations are performed using the Ansys Fluent and levels.41,42
2019 R3 CFD code. The pressure-based solver is utilized to solve the
momentum and pressure correction equations. The flow solver 1. Objective function
employs the semi-implicit method for pressure-linked equations
(SIMPLE) method for pressure–velocity coupling to ensure computa- The effectiveness of the system is assessed using a loss function
tional stability and produce reasonable results. Additionally, a second- constructed from statistics43 and the divergence of a characteristic
order upwind approach is utilized for the convective terms, as from its intended value. The loss function is written as follows:44
suggested by previous studies.36–38 The convergence criterion for each
variable is set to 106. The solver settings for the CFD simulation are Lð y Þ ¼ K ð y  T Þ 2 ; (10)
presented in Table III. The turbulence intensity in Table III for this where L is the loss, K is a constant whose value depends on the size of
study was chosen to be 1% following the procedure of the previous the characteristic, T is the desired response, and y is the actual
research.19,20 This choice aims to minimize simulation result errors. response. The approach defines three distinct types of S/N ratios: nom-
However, this choice also raises reliability issues with the simulation inal-the-better (NB), larger-the-better (LB), and smaller-the-better
procedure. Therefore, additional simulations were conducted for a tur- (SB). A considerable characteristic performance is correlated with a

02 December 2023 10:07:44

bulence intensity sensitivity study, ideally encompassing a low-TI and high S/N ratio. The following loss function is used to compute the S/N
a high-TI (approximately 10%) scenario, to observe the reaction of the
ratio:
results to changes in turbulence intensity. Equation (9) is employed to
calculate the time steps S  
¼ 10 log ð y  T Þ2 : (11)
h N
Dtjh ¼; (9)
180 According to Betz’s law or Betz limit, the theoretical maximum power
x
p coefficient of a single turbine is 0.593.45 Hence, T in Eq. (11) is defined
where h is the azimuthal angle of the turbine rotor related to time as having a value of 0.593. The performance of the investigated turbine
steps. system will improve with an increasing average power coefficient.

TABLE III. Summary of solver setting.

Grid number 140 000

Time-step 0.001 132 76 s (1 per time step), k ¼ 1
Sub-iterations per time step 20
Solution methods Solver Transient
Pressure-based
SIMPLE
Discretization methods Least squares cell-based method for gradients
Second-order upwind discretization for equations
Solution controls Under-relaxation factors Pressure ¼ 0.3
Momentum ¼ 0.7
Rotational model Sliding mesh model (SMM)
Turbulence model SST k–x
Boundary conditions Incoming flow velocity V ¼ 7 m/s
Angular velocity X ¼ 15.402 rad/s
Turbulence boundary conditions Inlet TI ¼ 1% - TVR ¼ 10
Outlet TI ¼ 5% - TVR ¼ 10
Residual criteria 10–6

Phys. Fluids 35, 115102 (2023); doi: 10.1063/5.0162026 35, 115102-7

Published under an exclusive license by AIP Publishing
 Physics of Fluids ARTICLE pubs.aip.org/aip/pof

TABLE IV. Design control parameters applied in the Taguchi method.

Level

Factor Control parameter Notation 1 2 3

A Barrier length Ls/D 0.5 0.7 0.9

B Barrier attachment angle a 0 10 20
C Geometry shape of cylinder deflector Double wake splitters Wake splitter in the middle Wake splitter on top
with splitter(s) and a barrier with a barrier with a barrier with a barrier

FIG. 8. Three different geometry shapes

of cylinder deflector with splitter and
barrier.

2. Factors and levels their chosen levels. This approach provides an array of minimum tests

02 December 2023 10:07:44

that result in the ideal design of variable values (factors). Every aspect
Numerous aspects might influence product quality. Several geo- and its related levels are evaluated equally by the OA because it is a
metric parameters in the blade designs of turbines for a wind turbine well-balanced method. Building an appropriate orthogonal array
system can be described as factors, and different parametric values on depends on the quantity of design factors (variables) and the levels
every factor are referred to as levels. As identified in the current study, (values) of each design component. The degrees of freedom technique
the Taguchi design will consider the following variables: the length of may be used to compute the necessary minimum tests. While only
the barrier (Ls/D), the barrier attachment angle (a), and the position of three variables and three levels are included in the current investiga-
the barrier and splitters on the cylindrical deflector. Table IV includes tion, using the standard OA L9(34) is also possible by eliminating the
a list of the chosen design parameters. A, B, and C are used in Table IV fourth factor. Therefore, the analysis could be completed with only
to represent the factors, including the length of the barrier, the attach- nine tests by using a common OA L9(33). Only three factors and three
ment angle of the barrier, and the various geometric shapes of the cyl- levels for each component are utilized in Table V L9 orthogonal array.
inder deflector with splitters and barriers. This representation is For emphasis, the orthogonality of the array is unaffected by the elimi-
conducted to facilitate a clear factor denotation in Secs. III E 3–III E 5. nation of the fourth element; hence, the levels would still appear three
Figure 8 illustrates the three different geometry shapes of cylinder times for each factor to maintain the array’s balance.
deflectors with splitters and barriers, which are labeled as C1, C2, and
C3. These variations were studied to evaluate their influence on the
aerodynamic performance of the Savonius turbines. The simulation TABLE V. L9 orthogonal array.
setup involved placing each of these configurations where detailed
measurements of torque, power coefficients, and wake patterns were Barrier Geometry shape of cylinder
taken under varying flow conditions. The results provided valuable Barrier attachment deflector with splitter
Experiment length (A) angle (B) and barrier (C)
insight into the impact of these design modifications on turbine effi-
ciency and wake behavior. This study contributes to our understanding 1 1 1 1
of optimizing Savonius turbine designs for enhanced hydrokinetic 2 1 2 2
applications in small rivers. All three geometry shapes have barriers on 3 1 3 3
the cylinder deflector, with C1 featuring two wake splitters, while C2
4 2 1 2
and C3 have a wake splitter positioned at the middle and top of the
5 2 2 3
downstream of the cylinder deflector, respectively.
6 2 3 1
7 3 1 3
3. Orthogonal array
8 3 2 1
While using the Taguchi approach, the orthogonal array (OA) 9 3 3 2
would be essential in determining the importance of components and

Phys. Fluids 35, 115102 (2023); doi: 10.1063/5.0162026 35, 115102-8

Published under an exclusive license by AIP Publishing
 Physics of Fluids ARTICLE pubs.aip.org/aip/pof

represent the mean S/N ratios for experiments with factor B at level j
and factor C at level k, respectively. In most cases, the last term e,
which represents the error, can be ignored. The S/N ratio for all experi-
ments can be determined by performing the minimum number of
experiments specified in the OA. This method is generally effective for
practical problems, assuming the absence of interactions between any
factors in the OA. Readers are referred to Phadke44 for additional
mathematical details on this model.

5. Factor effect value

In the Taguchi method, understanding and quantifying factor
effect values is essential for achieving optimal results. These values play
a pivotal role in pinpointing the factors that wield the greatest influence
over the studied performance or quality characteristic. To calculate fac-
tor effect values, the signal-to-noise (S/N) ratio for various factor levels
or settings is determined, with specific formulas chosen based on the
optimization goal. Subsequently, the average S/N ratio is computed for
each factor setting, and the range of S/N ratios is assessed by comparing
the highest and lowest averages. This leads to the calculation of the fac-
tor effect (FE) value, which indicates the relative impact of each factor.
Factors with high factor effect values are prioritized for further optimi-
zation, streamlining the path to achieving the desired outcome. In
essence, Factor effect values serve as quantitative guides, revealing the
most influential factors and paving the way for effective optimization,

Factor effect ðFEÞ ¼ highest S=N ratios for a factor

02 December 2023 10:07:44

 lowest S=N ratios for a factor: (13)

F. Validations
The present study employed the experimental examination con-
ducted by Sheldahl et al. to verify the accuracy and dependability of
the numerical outcomes.26 The experimental data were used to evalu-
FIG. 9. Validation of average computed moment coefficient (Cm) and power coeffi- ate the aerodynamic performance of a drag-type Savonius turbine
cient (Cp) with Sheldahl et al.26 experimental data.
without a deflector case at Re ¼ 4.32  105, as depicted in Fig. 9. The
results suggest that the current two-dimensional (2D) CFD modeling
4. Additive model agrees with the three-dimensional (3D) experimental findings, with
Equation (12) can be used to achieve the three elements that the minor discrepancies of no more than 5%, and closely matches the
additive model uses to represent the S/N ratio from outside OA trend of average torque and power coefficients (Cm and Cp) in terms
of k (Table VI). Table VII compares the average CP obtained from the
ge ðAi ; Bj ; Ck Þ ¼ l þ ai þ bj þ ck þ e; (12) present numerical work for Savonius wind turbines without deflector
cases13,28,46 at two different k equal to 0.6 and 1 to evaluate the preci-
where sion of the numerical outcomes effectively. The comparison indicates
ai ¼ lAi  l; bj ¼ lBj  l and ck ¼ lCk  l: that the present CFD modeling and other numerical studies are in
good agreement, with a maximum difference of 3.4%.
The above equation represents the mean S/N ratio for all experiments Fatahian et al. conducted an additional investigation, specifically
in an OA as follows: l, where lAi is the average S/N ratio for experi- focusing on the performance of the cylinder deflector with a wake
ments (i.e., Ai) conducted at level i for factor A. Similarly, lBj and lCk splitter, to validate the simulation results. Their findings were

TABLE VI. Comparison between the average moment coefficients (Cm) of the present study and experimental data.

TSR (k) 0.6 0.7 0.8 0.9 1.0 1.1 1.2 1.3

Cm of Experimental data (3D) 0.365 0.313 0.294 0.270 0.235 0.213 0.815 0.156
Cm of simulation study (2D) 0.362 0.312 0.284 0.261 0.232 0.210 0.183 0.161
Percentage difference (%) 0.82 0.32 3.40 3.33 1.28 1.41 0.92 3.21

Phys. Fluids 35, 115102 (2023); doi: 10.1063/5.0162026 35, 115102-9

Published under an exclusive license by AIP Publishing
 Physics of Fluids ARTICLE pubs.aip.org/aip/pof

TABLE VII. Comparison between the average power coefficient (CP) of the present study and previous numerical studies.

Average CP

Numerical studies Turbulence model k ¼ 0.6 Difference (%) k¼1 Difference (%)

Present study SST k–x 0.2211  0.2492 

Tian et al.28 SST k–x 0.219 0.96 0.246 1.30
Alizadeh et al.13 SST k–x 0.2183 1.28 0.242 2.98
Roshan et al.46 SST k–x 0.215 2.84 0.258 3.41

compared to the simulation results obtained in the present study. IV. RESULTS AND DISCUSSION
Figure 10 demonstrates the comparison, showing that the maximum A. Torque and power analysis in orthogonal array
percentage error in the simulation was 3.8% at k ¼ 0.6, while the low-
As stated earlier, the Taguchi method was employed in this study,
est percentage error was recorded at 0.54% at k ¼ 1, based on the work
with the power coefficient serving as the objective function. The
of Fatahian et al.20 These results indicate a close agreement between
instantaneous turbine blade torque was obtained from the simulations
the simulation data and the experimental findings, providing further
to compute the power coefficient for each VAWT model. The moment
confidence in the accuracy and reliability of the simulation model.
coefficient (Cm) of advancing and returning the blade of Savonius
VAWTs was calculated using the following equation:
T
Cm ¼ ; (14)
1
qU1
2
SR
2
where T represents the torque of the advancing and returning blade of
the Savonius VAWT. The simulations were performed for three differ-

02 December 2023 10:07:44

ent geometry shapes of cylinder deflector with splitter and barrier, as
shown in Fig. 8, at three different barrier lengths (Ls/D) and three dif-
ferent barrier attachment angles (a) as in Table IV, to obtain the flow
characteristics and by using the level combinations. Table VIII lists the
predicted values of the average power coefficient and S/N ratio. In all
nine simulations conducted in the study, a fixed k value of 1 was main-
tained to ensure consistency and accuracy in the results obtained. This
study only considers one type of tip speed ratio, which is k ¼ 1. This
choice is based on the fact that k ¼ 1 generally yields a high coefficient
of performance for Savonius turbines. Therefore, this study aims to
assess the extent of improvement achievable under optimal rotational
speed conditions, which correspond to the stage of best performance.
This decision was made on the basis of previous research19,20 that
showed a significant impact of k on the performance of the Savonius
turbine. However, the current study disregards the range of k values
yet acknowledges the broad range of k values that could influence the
outcomes of the research, which might be of interest to other research-
ers or practitioners in the field. However, considering all these values
would substantially increase the number of runs required, which is
beyond the scope of this study. Exploring additional k values will be
considered in future research. Based on previous studies,19,20,26 a trend
can be observed where the performance of Savonius turbines increases
from k 0.6 to 0.9 but then starts to decline from k 0.9 to 1.3. Therefore,
compared with the performance achieved at k ¼ 1, the coefficient of
performance would expectedly decrease after k ¼ 1. The study isolated
the effects of various design parameters under investigation, including
the position and angle of the wake splitter and the length of the barrier,
by keeping k constant across all simulations. This approach allowed
for a highly accurate comparison of the performance of different
FIG. 10. Validation of average computed moment coefficient (Cm) and power coeffi- designs and a clear identification of the ideal combination of design
cient (Cp) with Fatahian et al.20 parameters that led to the highest power coefficient. A high Cp average

Phys. Fluids 35, 115102 (2023); doi: 10.1063/5.0162026 35, 115102-10

Published under an exclusive license by AIP Publishing
 Physics of Fluids ARTICLE pubs.aip.org/aip/pof

TABLE VIII. Comparison of power coefficients and S/N ratios for all nine simulations.

Barrier Barrier attachment Geometry shape of cylinder deflector

Simulation length (A) angle (B) with splitter and barrier (C) Cp S/N ratio

E1 0.5 0 Double wake splitters with barrier 0.410 14.61
E2 0.5 10 Wake splitter on middle with barrier 0.395 13.94
E3 0.5 20 Wake splitter on top with barrier 0.413 14.75
E4 0.7 0 Wake splitter on middle with barrier 0.200 8.05
E5 0.7 10 Wake splitter on top with barrier 0.409 14.56
E6 0.7 20 Double wake splitters with barrier 0.438 16.03
E7 0.9 0 Wake splitter on top with barrier 0.449 16.65
E8 0.9 10 Double wake splitters with barrier 0.459 17.27
E9 0.9 20 Wake splitter on middle with barrier 0.310 10.87

generally corresponds to a large S/N ratio, indicating superior system values obtained for the three level 1 trials of factor A, namely, 14.610,
performance. Run 8 demonstrated the highest system performance, 13.936, and 14.751, as presented in Table VIII. The S/N ratios are also
with average power coefficient and S/N ratio values of 0.459 and computed similarly for other factor levels. Chen et al. proposed that an
17.266, respectively, while Run 4 had the lowest performance, with increase in the mean S/N ratio is indicative of an improvement in the
average power coefficient and S/N ratio values of 0.200 and 8.046, performance of the wind turbine system regarding power output.39
respectively. The corresponding S/N ratios were also calculated using Thus, the order of influence of the chosen factors can be deduced from
Eq. (11) and listed in Table VIII, showing a proportional relationship the variations in values presented in this figure. Specifically, the geo-
with the power coefficient value. Thus, the highest and lowest S/N metric shape of a cylinder deflector with splitter and barrier is regarded
ratios occurred in E8 and E4, respectively. However, determining as the most critical factor that affects power performance among the
which factors had the most substantial influence on the power output

02 December 2023 10:07:44

three factors. The results depicted in Fig. 11 demonstrate that the com-
based on the results presented in this table is challenging. Therefore, bination of A3, B2, and C1 is the optimal choice for predicting the
the Taguchi method will be utilized in Sec. IV B to avoid the need for a highest power output performance based on the mean S/N ratio values
“real” simulation and determine the optimal combination and influ- obtained through the Taguchi method. The averaged value of the S/N
ence order of the selected design parameters. ratio from different cases, each with a particular level of one factor,
The data presented in Table VIII allow for the computation of could analytically reflect the individual contribution of the levels.
the mean S/N ratio for each level of every individual factor. Figure 11 The factor effect value in terms of mean power coefficients pre-
provides an overview of the profiles of mean S/N ratios for the three sented in Fig. 12 is defined as the difference between the highest and
factors at various levels. For instance, the mean S/N ratio of factor A lowest mean S/N ratios for each factor. The differences in these factors
with level 1, which is plotted as A1 on the x axis, is determined by aver- can be observed in Fig. 11. The significance of the factor effect is to
aging the S/N ratio values from Table VIII that correspond to the three observe the sensitivity among different factors. A high factor effect
level 1 values for factor A. The mean S/N ratio of factor A with level 1 indicates high sensitivity of a certain factor. Figure 12 demonstrates
is determined by calculating the arithmetic mean of the S/N ratio that the effect values are ordered as follows: geometry shape of cylinder

FIG. 11. Profiles of mean S/N ratio. FIG. 12. Factor effect value in terms of mean power coefficient.

Phys. Fluids 35, 115102 (2023); doi: 10.1063/5.0162026 35, 115102-11

Published under an exclusive license by AIP Publishing
 Physics of Fluids ARTICLE pubs.aip.org/aip/pof

deflector with splitter and barrier > barrier attachment angle > barrier A3, B2, and C1 has the greatest S/N ratio, which is equivalent to 17.99.
length. The effect profile indicates that the performance coefficient of This finding indicates that the turbine with this setup and geometry
VAWTs is particularly sensitive to changes in the levels of Factors C shape of cylinder deflector with splitter and barrier would produce the
(geometry shape of cylinder deflector with splitter and barrier) while most power coefficient in between parameters; thus, this turbine has
remaining relatively less sensitive to changes in Factors A (barrier been chosen for this study. This result corresponds to the one depicted
length) and B (barrier attachment angle). in Fig. 11. A3, B2, and C1 are equivalent to the simulation of E8.
Therefore, E8 has the highest coefficient power for the Savonius tur-
B. Additive model bine compared to other cases. The additive model points in the proper
direction for improving power performance, but the ideal state was not
The estimated S/N ratios for all 27 combinations of factors and
accurately anticipated. This approach performs exceptionally well in
levels were calculated using the additive model by applying Eq. (12),
problems with no-interaction components, as was already previously
utilizing the data from Table VIII. Consider the combinations of A1,
mentioned. Therefore, determining whether any connection between
B2, and C3 as an example, where is the mean of all nine S/N ratios and
factors exists is the next stage in this study.
a1 is the mean of the S/N ratios for E1, E2, and E3 as they incorporate
factor A with level 1. Similarly, b2 represents the mean S/N ratio from
E2, E5, and E8, whereas c3 represents the mean S/N ratio from E3, E5, C. Interaction analysis
and E7. The estimated mean S/N ratios for the complete combinations A linear graph was created to assess the interaction data easily
are listed in Table IX. The numbers 1, 2, and 3 in each barrier length, and determine the presence of interactions that occur between the two
barrier attachment angle, and geometry shape of cylinder deflector chosen factors. Figure 13 illustrates a scenario with two factors and
with splitter and barrier columns of Table V indicate the levels illus- three levels. Figure 13 shows a parallelism of the impact of the compo-
trated. Based on the result, case number 22 with the combination of nents, assuming the absence of interaction between the two in
Fig. 13(a). The two components can be viewed as interacting despite
TABLE IX. Estimated S/N ratio obtained from the additive model Eq. (12). the non-parallelism between them. The scenario with a synergistic
interaction, also known as a weak interaction between the two ele-
Barrier Geometry shape of ments, is shown in Fig. 13(b). By contrast, the example of an anti-
Barrier attachment cylinder deflector Estimated synergistic interaction is shown in Fig. 13(c) and can be referred to as a
strong interaction. If an anti-synergistic interaction occurs, then the

02 December 2023 10:07:44

Number length angle with splitter and barrier S/N ratio
mutual impact of the components should be regarded as an important
1 1 1 1 15.34 parameter impacting the product quality. In the present case, the addi-
2 1 1 2 10.33 tive model is valid, but the anticipated optimal levels of the Taguchi
3 1 1 3 14.70 method may be deceptive for product designs. The interaction analysis
4 1 2 1 17.49 between any two factors, including A B, A C, and B C, is carefully stud-
5 1 2 2 12.48 ied by applying the technique outlined above. Notably, the denotations
6 1 2 3 16.85 A B, A C, and B C signify solely the effect of the two chosen factors,
ignoring the third. For instance, if A B, then only factors A and B are
7 1 3 1 16.12
used in the calculation, regardless of the number provided for factor C.
8 1 3 2 11.11
For this investigation, a relationship between any two parameters is
9 1 3 3 15.48 constructed in Fig. 14 based on the mean S/N ratio values displayed in
10 2 1 1 13.79 Table VIII. All three factor combinations exhibit an interaction effect,
11 2 1 2 8.77 which indicates that all two-factor interactions are either synergistic or
12 2 1 3 13.14 anti-synergistic. Among these plots, the connection between A and B
13 2 2 1 15.94 has a common pattern that can be viewed as an interaction with an
14 2 2 2 10.92 anti-synergistic effect. The connection between B and C is another
15 2 2 3 15.30
16 2 3 1 14.57
17 2 3 2 9.55
18 2 3 3 13.92
19 3 1 1 15.84
20 3 1 2 10.82
21 3 1 3 15.20
22 3 2 1 17.99
23 3 2 2 12.98
24 3 2 3 17.35
25 3 3 1 16.62
26 3 3 2 11.60
27 3 3 3 15.98
FIG. 13. Example of interactions between factors.44

Phys. Fluids 35, 115102 (2023); doi: 10.1063/5.0162026 35, 115102-12

Published under an exclusive license by AIP Publishing
 Physics of Fluids ARTICLE pubs.aip.org/aip/pof

02 December 2023 10:07:44

FIG. 14. Interaction analysis between two factors.

example of this phenomenon. This finding indicates that factor B has a

significant interaction effect with the two other factors.

D. Flow structures
This study presents visual representations of static pressure
contours and velocity profiles to comprehensively analyze the impact
of the cylinder deflector with wake splitters and that with barrier
plate on the performance of the Savonius turbine rotor. Specifically,
at h ¼ 300 , which corresponds to the azimuthal location where the
maximum Cm (moment coefficient) is achieved, the findings are
presented in Fig. 15. This figure depicts the correlation between the
instantaneous torque coefficient (Cm) and the rotation angle, com-
paring the cases of cylinder deflectors with two wake splitters and an
additional barrier plate, to the scenario without a deflector, all at
k ¼ 1. The purpose is to demonstrate the contribution of the FIG. 15. Changes of Cm for the deflector with splitters and barrier plate at k ¼ 1.

Phys. Fluids 35, 115102 (2023); doi: 10.1063/5.0162026 35, 115102-13

Published under an exclusive license by AIP Publishing
 Physics of Fluids ARTICLE pubs.aip.org/aip/pof

presence of a barrier plate attached onto cylindrical deflector to tor- Understanding the flow dynamics of the rotor is crucial, and the
que augmentation. velocity contour provides valuable insight. Figure 17 demonstrates
Figure 16 provides a comparative investigation of static pressure the impact of the cylinder deflector attached by two wake splitters on
contours surrounding the Savonius turbine at k ¼ 1. Considering three the downstream (with and without barrier plate) on the velocity field
scenarios, the study examines surrounding the Savonius rotor at k ¼ 1. As the rotor rotates, two sig-
nificant zones emerge: a high-velocity (HV) zone near the advancing
(1) a case without a deflector, blade and a wake zone downstream of the turbine. A large wake zone
(2) a case with a deflector followed by two wake splitters, and forms behind the rotor, where the flow decelerates after passing
(3) a case with a deflector followed by two splitters and a barrier around the two blades. Moreover, the cylindrical deflector attached
at h ¼ 300 on the convex side of the advancing blade. The observed with two wake splitters and barrier plate produced the largest wake
static pressure fields exhibit similarities to those reported in previous zone compared to others. Introducing a circular deflector in front of
studies. Notably, two significant zones can be identified in Fig. 16: (1) the turbine redirects the airflow toward the advancing blade instead of
the high-pressure (HP) zone located on the returning blade, and (2) directly hitting the convex surface of the returning blade. The intro-
the low-pressure (LP) zone situated on the advancing blade. duction of splitter wakes on the cylinder deflector serves to divide the
flow into separate paths, while the barrier plate plays a role in obstruct-
ing a specific flow component. This obstruction has the potential to

02 December 2023 10:07:44

FIG. 16. Comparison of static pressure distributions around the Savonius turbine FIG. 17. Comparison of velocity distributions around the Savonius turbine for three
for three different configurations at k ¼ 1 at 4.22 s. different configurations at k ¼ 1 at 4.22 s.

Phys. Fluids 35, 115102 (2023); doi: 10.1063/5.0162026 35, 115102-14

Published under an exclusive license by AIP Publishing
 Physics of Fluids ARTICLE pubs.aip.org/aip/pof

create a substantial wake zone behind the rotor. Therefore, the negative geometric shape of the cylinder with splitters had a significant effect
torque from the returning blade will diminish, while the positive tor- on the power performance of the Savonius turbine, while the barrier
que from the advancing blade will rise. length and barrier attachment angle had minimal impact.
The analysis of the S/N ratio indicated that the use of double
E. Turbulence intensity wake splitters with a barrier on the cylinder under specific operating
conditions produced significant findings. A barrier length (Ls/D) of
Turbulence intensity is defined as the ratio of the root mean
square (RMS) of the turbulent velocity fluctuations to the mean veloc- 0.9 and a barrier attachment angle (a) of 10 for k ¼ 1 resulted in max-
ity of the flow. This intensity is typically expressed as a percentage and imum power performance of the system by efficiently harnessing wind
is used to describe the level of turbulence in the initial conditions of energy from the interaction of the Savonius turbine. Consequently, the
the simulation. High turbulence intensity values indicate considerable optimal combination of design parameters was identified, resulting in
turbulent and fluctuating flow behavior, while low values represent a power coefficient equivalent to 0.459. This finding signifies a notable
smooth and less turbulent flow conditions. The past research method- improvement in the overall performance of the Savonius system, with
ology employed a 1% turbulence intensity. Thus, the intention is to an increase in approximately 61% compared to the highest power coef-
conduct a comparison between the outcomes of real-world 10% turbu- ficient obtained for k ¼ 1 recorded in the previous study.20
lence and the simulation involving a 1% turbulence intensity. Furthermore, the utilization of a 10 barrier attachment angle holds a
Therefore, a comprehensive evaluation of all cases has been conducted remarkable potential for achieving a high power coefficient, as demon-
to observe any disparities. The representation in Fig. 18 illustrates the strated on the data presented in Fig. 11.
Notably, the position of the wake splitter on top of the cylinder
discernible difference in turbulence intensity between 1% and 10% at
deflector plays a crucial role in achieving a high power coefficient. The
the advancing blade throughout one complete rotation of the Savonius
turbine, featuring a cylinder deflector, two wake splitters, and a barrier. inclusion of a wake splitter on top of the cylinder deflector in all simu-
The insights derived from Fig. 18 reveal that the percentage disparity lations resulted in a high power coefficient, which can be attributed to
between turbulence intensities of 1% and 10% is notably minimal. the obstruction of the flow that should move toward the returning
Notably, for the specific purpose of comparing turbulence intensity, blade and redirect the flow toward the advancing blade. More impor-
only one case has been chosen for presentation. Consequently, the fol- tantly, this study has not asserted that k ¼ 1 results in the best perfor-
lowing conclusion can be drawn: the utilization of a 1% turbulence mance of the Savonius turbine when compared to other k values.
Therefore, additional investigations involving different k values are

02 December 2023 10:07:44

intensity for these simulations could lead to outcomes that are compa-
rable to those with 10% turbulent intensities. necessary to determine the existence of an optimal configuration for
the Savonius turbine. Furthermore, investigations using physical
V. CONCLUSIONS experiments should be conducted for the proposed Savonius turbine
CFD simulations have been conducted herein to determine the with a cylinder deflector–double wake splitter plus a barrier to verify
power performance of Savonius VAWTs with three different geomet- the overall performance improvement.
ric shapes of a cylinder-type deflector with a splitter and a barrier
under various operating circumstances. Using the Taguchi approach, ACKNOWLEDGMENTS
the performance of the Savonius wind turbine has been examined by
This work was supported by Universiti Sains Malaysia under
integrating three operational elements under three levels: barrier length
the Short-Term Scheme with Grant No. 304/PAERO/6315475.
(Ls/D), barrier attachment angle (a), and geometric shape of the
cylinder-type deflector. The Taguchi method showed that the
AUTHOR DECLARATIONS
Conflict of Interest
The authors have no conflicts to disclose.

Author Contributions
Mohamad Hafizul Fikri Mahizam: Conceptualization (equal);
Data curation (equal); Formal analysis (equal); Investigation
(equal); Methodology (equal); Validation (equal); Visualization
(equal); Writing – original draft (equal); Writing – review & edit-
ing (equal). Wei Shyang Chang: Conceptualization (equal);
Funding acquisition (equal); Methodology (equal); Project
administration (equal); Software (equal); Supervision (equal);
Visualization (equal); Writing – review & editing (equal).
Esmaeel Fatahian: Conceptualization (equal); Methodology
(equal); Project administration (equal); Visualization (equal);
Writing – review & editing (equal). Farzad Ismail: Project admin-
istration (equal); Visualization (equal); Writing – review & editing
FIG. 18. Comparison of turbulence intensity between 1% TI and 10% TI for E8 (equal). Mohammad Hafifi Hafiz Ishak: Project administration
case. (equal); Visualization (equal); Writing – review & editing (equal).

Phys. Fluids 35, 115102 (2023); doi: 10.1063/5.0162026 35, 115102-15

Published under an exclusive license by AIP Publishing
 Physics of Fluids ARTICLE pubs.aip.org/aip/pof

DATA AVAILABILITY 20
E. Fatahian, F. Ismail, M. H. H. Ishak, and W. S. Chang, “The role of wake
splitter deflector on performance enhancement of Savonius wind turbine,”
The data that support the findings of this study are available from Phys. Fluids 34, 095111 (2022).
the corresponding author upon reasonable request. 21
M. S. Abdullah, M. H. H. Ishak, and F. Ismail, “Performance improvement of
the Savonius turbine using a novel augmentation device with the Taguchi opti-
REFERENCES mization method,” Phys. Fluids 35(1), 15108 (2023).
1
22
P. A. Setiawan, T. Yuwono, and W. A. Widodo, “Numerical study of the stagger
OECD, OECD Green Growth Studies (OECD, 2012), p. 104.
2 angle effect of a circular cylinder installed in front of returning blade toward
W. T. Chong, W. K. Muzammil, H. C. Ong, K. Sopian, M. Gwani, A. Fazlizan,
the vertical axis Savonius water turbine performance,” J. Phys.: Conf. Ser.
and S. C. Poh, “Performance analysis of the deflector integrated cross axis wind
1179(1), 012107 (2019).
turbine,” Renewable Energy 138, 675–690 (2019).
3
23
S. Sewucipto and T. Yuwono, “The influence of upstream installation of D-53
K. H. Wong, W. T. Chong, N. L. Sukiman, Y. C. Shiah, S. C. Poh, K. Sopian,
type cylinder on the performance of Savonius turbine,” J. Adv. Res. Exp. Fluid
and W. C. Wang, “Experimental and simulation investigation into the effects
Mech. Heat Transfer 3, 36–47 (2021).
of a flat plate deflector on vertical axis wind turbine,” Energy Convers. Manage. 24
K. H. Wong, W. T. Chong, S. C. Poh, Y. C. Shiah, N. L. Sukiman, and C. T.
160, 109–125 (2018).
Wang, “3D CFD simulation and parametric study of a flat plate deflector for
4
C. Sim~ao Ferreira, G. Van Kuik, G. Van Bussel, and F. Scarano, “Visualization
vertical axis wind turbine,” Renewable Energy 129, 32–55 (2018).
by PIV of dynamic stall on a vertical axis wind turbine,” Exp. Fluids 46(1), 25
M. B. Salleh, N. M. Kamaruddin, and Z. Mohamed-Kassim, “Experimental
97–108 (2009).
investigation on the effects of deflector angles on the power performance of a
5
A. Pallotta, D. Pietrogiacomi, and G. P. Romano, “HYBRI—A combined
Savonius turbine for hydrokinetic applications in small rivers,” Energy 247,
Savonius-Darrieus wind turbine: Performances and flow fields,” Energy 191,
123432 (2022).
116433 (2020). 26
R. E. Sheldahl, B. F. Blackwell, and L. V. Feltz, “Wind tunnel performance data
6
H. Fatahian, H. Salarian, J. Khaleghinia, and E. Fatahian, “Improving the effi-
for two- and three-bucket Savonius rotors,” J. Energy 2(3), 160–164 (1978).
ciency of a Savonius vertical axis wind turbine using an optimum parameters,” 27
K. Golecha, T. I. Eldho, and S. V. Prabhu, “Influence of the deflector plate on
Comput. Res. Prog. Appl. Sci. Eng. 4, 27–32 (2018).
the performance of modified Savonius water turbine,” Appl. Energy 88(9),
7
W. Xu, G. Li, X. Zheng, Y. Li, S. Li, C. Zhang, and F. Wang, “High-resolution
numerical simulation of the performance of vertical axis wind turbines in 3207–3217 (2011).
28
W. Tian, Z. Mao, B. Zhang, and Y. Li, “Shape optimization of a Savonius wind
urban area. I. Wind turbines on the side of single building,” Renewable Energy
177, 461–474 (2021). rotor with different convex and concave sides,” Renewable Energy 117,
8
J. Halliday, “The feasibility of building-mounted/integrated wind turbines 287–299 (2018).
29
(BUWTs): Achieving their potential for carbon emission reductions,” Report A. S. Saad, A. Elwardany, I. I. El-Sharkawy, S. Ookawara, and M. Ahmed,
“Performance evaluation of a novel vertical axis wind turbine using twisted

02 December 2023 10:07:44

No. 2002-07-028-1-6 (Final Report of Carbon Trust Contract Science and
Technology Facilities Council, 2005). blades in multi-stage Savonius rotors,” Energy Convers. Manage. 235, 114013
9
L. Zhang and J. Qu, “Study on aerodynamic performance of a combined verti- (2021).
cal axis wind turbine based on blade element momentum theorem,”
30
H. Chizari and F. Ismail, “A grid-insensitive LDA method on triangular
J. Renewable Sustainable Energy 13, 33304 (2021). grids solving the system of Euler equations,” J. Sci. Comput. 71(2), 839–874
10
S. Li, Y. Li, C. Yang, Q. Wang, B. Zhao, D. Li, R. Zhao, T. Ren, X. Zheng, Z. (2017).
Gao, and W. Xu, “Experimental investigation of solidity and other characteris-
31
N. Alom and U. K. Saha, “Influence of blade profiles on Savonius rotor perfor-
tics on dual vertical axis wind turbines in an urban environment,” Energy mance: Numerical simulation and experimental validation,” Energy Convers.
Convers. Manage. 229, 113689 (2021). Manage. 186, 267–277 (2019).
11
M. H. Mat Yazik, W. S. Chang, M. H. H. Ishak, E. Fatahian, and F. Ismail,
32
F. Ismail, P. M. Carrica, T. Xing, and F. Stern, “Evaluation of linear and nonlin-
“Effect of surface roughness and blade material on the performance of a station- ear convection schemes on multidimensional non-orthogonal grids with appli-
ary Savonius wind turbine under different operating conditions,” Phys. Fluids cations to KVLCC2 tanker,” Int. J. Numer. Methods Fluids 64(8), 850–886
35(3), 35133 (2023). (2010).
12
S. M. Al Absi, A. H. Jabbar, S. O. Mezan, B. A. Al-Rawi, and S. T. Al_Attabi, “An
33
M. Hassanpour and L. N. Azadani, “Aerodynamic optimization of the configu-
experimental test of the performance enhancement of a Savonius turbine by modify- ration of a pair of vertical axis wind turbines,” Energy Convers. Manage. 238,
ing the inner surface of a blade,” Mater. Today Proc. 42, 2233–2240 (2021). 114069 (2021).
13
H. Alizadeh, M. H. Jahangir, and R. Ghasempour, “CFD-based improvement of
34
A. Sagharichi, M. Zamani, and A. Ghasemi, “Effect of solidity on the perfor-
Savonius type hydrokinetic turbine using optimized barrier at the low-speed mance of variable-pitch vertical axis wind turbine,” Energy 161, 753–775
flows,” Ocean Eng. 202, 107178 (2020). (2018).
14
M. Al-Ghriybah, M. Fadhli Zulkafli, D. Hissein Didane, and S. Mohd, “The
35
M. Eshagh Nimvari, H. Fatahian, and E. Fatahian, “Performance improvement
effect of spacing between inner blades on the performance of the Savonius wind of a Savonius vertical axis wind turbine using a porous deflector,” Energy
turbine,” Sustainable Energy Technol. Assess. 43, 100988 (2021). Convers. Manage. 220, 113062 (2020).
15
P. K. Talukdar, V. Kulkarni, and U. K. Saha, “Performance estimation of
36
F. R. Menter, “Two-equation eddy-viscosity turbulence models for engineering
Savonius wind and Savonius hydrokinetic turbines under identical power applications,” AIAA J. 32, 1598–1605 (1994).
input,” J. Renewable Sustainable Energy 10(6), 64704 (2018).
37
A. S. Saad, I. I. El-Sharkawy, S. Ookawara, and M. Ahmed, “Performance
16
P. Zhao, Y. Jiang, S. Liu, T. Stoesser, L. Zou, and K. Wang, “Investigation of enhancement of twisted-bladed Savonius vertical axis wind turbines,” Energy
fundamental mechanism leading to the performance improvement of vertical Convers. Manage. 209, 112673 (2020).
axis wind turbines by deflector,” Energy Convers. Manage. 247, 114680 (2021).
38
M. Mosbahi, A. Ayadi, Y. Chouaibi, Z. Driss, and T. Tucciarelli, “Performance
17
M. H. Mohamed, G. Janiga, E. Pap, and D. Thevenin, “Optimal blade shape of a study of a helical Savonius hydrokinetic turbine with a new deflector system
modified Savonius turbine using an obstacle shielding the returning blade,” design,” Energy Convers. Manage. 194, 55–74 (2019).
Energy Convers. Manage. 52(1), 236–242 (2011).
39
W. H. Chen, C. Y. Chen, C. Y. Huang, and C. J. Hwang, “Power output analysis
18
K. Layeghmand, N. G. Tabari, and M. Zarkesh, “Improving efficiency of and optimization of two straight-bladed vertical-axis wind turbines,” Appl.
Savonius wind turbine by means of an airfoil-shaped deflector,” J. Braz. Soc. Energy 185, 223–232 (2017).
Mech. Sci. Eng. 42(10), 528 (2020).
40
Z. Wang, Y. Wang, and M. Zhuang, “Improvement of the aerodynamic perfor-
19
E. Fatahian, F. Ismail, M. Hafifi Hafiz Ishak, and W. Shyang Chang, “An inno- mance of vertical axis wind turbines with leading-edge serrations and helical
vative deflector system for drag-type Savonius turbine using a rotating cylinder blades using CFD and Taguchi method,” Energy Convers. Manage. 177,
for performance improvement,” Energy Convers. Manage. 257, 115453 (2022). 107–121 (2018).

Phys. Fluids 35, 115102 (2023); doi: 10.1063/5.0162026 35, 115102-16

Published under an exclusive license by AIP Publishing
 Physics of Fluids ARTICLE pubs.aip.org/aip/pof

41
W.-H. Chen, C.-J. Chen, and C.-I. Hung, “Taguchi approach for co-gasification opti- 44
M. S. Phadke, Quality Engineering Using Robust Design (Prentice Hall,
mization of torrefied biomass and coal,” Bioresour. Technol. 144, 615–622 (2013). Englewood Cliffs, NJ, 1989).
42
W.-H. Chen, S.-R. Huang, and Y.-L. Lin, “Performance analysis and optimum 45
H. Hu, Z. Wang, A. Ozbay, W. Tian, and A. Sharma, An experimental investi-
operation of a thermoelectric generator by Taguchi method,” Appl. Energy gation on the wake characteristics behind a novel twin-rotor wind turbine,”
158, 44–54 (2015). AIAA Paper No. 2015-1663, 2015.
43
C.-N. Liao and H.-P. Kao, “Supplier selection model using Taguchi loss func- 46
A. Roshan, A. Sagharichi, and M. Maghrebi, “Nondimensional parameters’
tion, analytical hierarchy process and multi-choice goal programming,” effects on hybrid Darrieus-Savonius wind turbine performance,” J. Energy
Comput. Ind. Eng. 58(4), 571–577 (2010). Resour. Technol. 142, 011202 (2020).

02 December 2023 10:07:44

Phys. Fluids 35, 115102 (2023); doi: 10.1063/5.0162026 35, 115102-17

Published under an exclusive license by AIP Publishing