See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/318603906

Pitch Angle Control for Small-Scale Darrieus Vertical Axis Wind Turbine with
Straight Blades (H-type VAWT)

Article  in  Renewable Energy · July 2017

DOI: 10.1016/j.renene.2017.07.068

CITATIONS READS

65 3,227

3 authors, including:

Gebreel Abdalrahman Fue-Sang Lien

University of Waterloo University of Waterloo
2 PUBLICATIONS   65 CITATIONS    181 PUBLICATIONS   4,538 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

CATHENA 4 Thermalhydraulics Code View project

GPGPU Implementation of a Lattice Boltzmann Methodology for Particle Transport and Deposition in Complex Flows View project

All content following this page was uploaded by Gebreel Abdalrahman on 08 November 2017.

The user has requested enhancement of the downloaded file.

 Renewable Energy 114 (2017) 1353e1362

Contents lists available at ScienceDirect

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

Pitch angle control for a small-scale Darrieus vertical axis wind

turbine with straight blades (H-Type VAWT)
Gebreel Abdalrahman a, b, *, William Melek a, Fue-Sang Lien b
a
The Laboratory of Computational Intelligence and Automation Group, Department of Mechanical and Mechatronics Engineering, University of Waterloo,
Waterloo, Ontario, N2L 3G1, Canada
b
Pollution Management and Atmospheric Modeling Laboratory Group, Department of Mechanical and Mechatronics Engineering, University of Waterloo,
Waterloo, Ontario, N2L 3G1, Canada

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

Article history: Unlike horizontal axis wind turbines (HAWTs), the Darrieus vertical axis wind turbine (H-type VAWT)
Received 25 November 2016 has been the subject of only a few recent studies directed at improving its self-starting capability and/or
Received in revised form aerodynamic performance. The technique currently used for improving the performance of this type of
26 May 2017
turbine is pitch angle control. This paper presents intelligent blade pitch control for enhancing the
Accepted 14 July 2017
Available online 21 July 2017
performance of H-type VAWTs with respect to power output. To determine the optimum pitch angles,
ANSYS Fluent Computational Fluid Dynamics (CFD) software was used for a study of the aerodynamic
performance of a 2D variable pitch angle H-type VAWT at a variety of tip speed ratios (TSRs). For each
Keywords:
Darrieus vertical axis wind turbine (H-type
case examined, the power coefficient (Cp) was calculated and compared to published experimental and
VAWT) CFD findings. The results obtained from the CFD model were then applied for the construction of an
Computational fluid dynamics (CFD) aerodynamic model of an H-type VAWT rotor, which constituted a prerequisite for designing an intel-
Variable pitch angle control ligent pitch angle controller using a multilayer perceptron artificial neural network (MLP-ANN) method.
Multilayer perceptron artificial neural The performance of the MLP-ANN blade pitch controller was compared to that of a conventional
network (MLP-ANN) controller (PID). The findings demonstrate that for an H-type VAWT, compared to a conventional PID
controller, an MLP-ANN results in superior power output.
© 2017 Elsevier Ltd. All rights reserved.

1. Introduction (HAWTs) and vertical axis wind turbines (VAWTs). VAWTs offer a
number of advantages over HAWTs: VAWTs can receive wind
Because of the rising worldwide demand for energy, the blowing from any direction (i.e., omni-directional) so that a yaw
development of alternative and renewable energy sources has mechanism is unnecessary, the simplicity of the blade design
become a vital factor in providing clean energy. All forms of means that the cost is relatively low [3], and VAWT maintenance is
renewable energy, from biofuels and hydro to wind, geothermal comparatively quick and easy since the transmission equipment
energy, and solar power, have been explored with respect to their and generator are located at ground level [2]. VAWTs fall into
practical and economic potential. Stand-alone power supply sys- different categories [2], two of which are the Savonius rotor, which
tems, such as wind turbines and solar cells, are not only suitable is a drag-type device, and the H-Darrieus rotor, which is a lift-type
solutions for producing the electricity needed for rural and remote device in which the blades are straight and parallel to the axis of
areas, but also economical alternatives because they can contribute rotation with a fixed or variable pitch angle.
to a reduction in the cost of grid extensions [1]. For a number of An important parameter associated with VAWT blades is the
years, two main types of wind turbines have been adopted for the operating tip speed ratio (TSR) (l), and a key factor in the selection
extraction of power from wind: horizontal axis wind turbines of the TSR is wind speed, as given in Ref. [4].

ur R
* Corresponding author. The Laboratory of Computational Intelligence and l¼ (1)
Automation Group, Department of Mechanical and Mechatronics Engineering,
u∞
University of Waterloo, Waterloo, Ontario, N2L 3G1, Canada.
E-mail addresses: [email protected] (G. Abdalrahman), william.melek@ where ur is the angular velocity of the rotor, u∞ is the wind speed,
uwaterloo.ca (W. Melek), [email protected] (F.-S. Lien). and R is the rotor radius.

http://dx.doi.org/10.1016/j.renene.2017.07.068
0960-1481/© 2017 Elsevier Ltd. All rights reserved.
1354 G. Abdalrahman et al. / Renewable Energy 114 (2017) 1353e1362

Actual VAWT performance can be predicted based on a deter- However, a large pitch amplitude can cause VAWT performance to
mination of the forces acting on each blade. Fig. 1 illustrates the deteriorate at high TSRs. Conversely, while a small pitch amplitude
velocity and force vectors acting on Darrieus turbine blades. The is sufficient to produce good performance at high TSRs, it will lead

! to poor performance at low TSRs [7,10]. This effect gives rise to an
velocity ð VÞ is the tangential velocity vector of the rotor. The

! important question: How should the pitch amplitude be varied in
resultant velocity vector ð W Þ is the relative velocity that consists
order to optimize VAWT performance as the TSR changes? In fact,
of the induced velocity (U) and blade velocity (V) vectors. The angle
the design of actuators with fast response times in order to vary the
of attack (a) is typically defined as the angle between the direction
pitch angle for a small H-type VAWT is complicated due to the high
of the relative velocity, W, and the chord line of the blade. The blade
frequency of the rotational speeds and the changes in the angle of
pitch angle is as set out in Ref. [5]:
attack. For large turbines, such designs are also expensive. A suit-
able aerodynamic model and a control system that can approxi-
b¼a4 (2)
mate the optimal pitch angle of an H-type VAWT during rotation
!
! are therefore required.
where 4 is the angle between the direction of V and W . Obviously, The goal of the work presented in this paper was a systemati-
both the angle of attack (a) and the relative wind speed (W), which cally designed nonlinear individual active blade pitch control sys-
is a function of the azimuth angle (q), vary during each cycle. tem for H-type VAWTs based on the simulation results obtained
The result is that the magnitude and orientation of both the lift with the commercial ANSYS Fluent CFD software. The CFD model
and drag forces change depending on the azimuthal position of the was used for studying the aerodynamic performance of an H-type
blade. Fig. 1 also shows the main forces acting on VAWT blades: lift, VAWT as well as for determining the optimum pitch angles for each
drag, normal (N), and tangential (T). Rotor performance can be individual blade. These optimal pitch angles were also utilized in
estimated using the tangential force (T) [5]. The power coefficient the design of the proposed control system. If each blade can be
(Cp) is the ratio of the mechanical power produced by the wind controlled individually, it is possible to maximize the moment co-
turbine (Pm ) to the power available in the wind (PW ) [6]: efficient for all blades. Because the complexity of VAWT aero-
dynamics requires a reliable nonlinear control system, this paper
Pm ur tr
Cp ¼ ¼ ¼ CQ l (3) proposes the use of intelligent control techniques based on a neural
PW 1=2rAu3∞ network approach for the design of an individual active blade pitch
control system for an H-type VAWT as a means of improving its
where r is the air density (1.225 kg/m3), and A is the area swept by
power generation performance.
the turbine (e.g., for the H-type Darrieus wind turbine, A ¼ 2RH,
Although VAWTs have been in existence for many years, very
where H is the blade length). However, for VAWTs at low TSRs, a
limited research has been conducted with respect to these systems,
negative torque is often generated because of the large dynamic
in comparison with the extensive examination devoted to HAWTs
cyclic variations in the angle of attack (a). The angle of attack (a) for
[11]. Only recently have VAWTs received increased experimental,
a VAWT with a fixed pitch angle (b ¼ 0 ) can be expressed as fol-
numerical, and analytical attention, a trend that is attributable to
lows [7]:
their ability to achieve useful power production but with less noise
  [12]. This paper provides a review of the previous work, divided
cos q
a ¼ tan1 (4) into two groups: (1) analytical and numerical studies in which the
sin q þ l aerodynamic performance of VAWTs was explored, and (2) pub-
To improve VAWT performance, a variable pitch technique is lished studies of the design of a wind turbine pitch control system.
applied through the modification of the angle of attack (a). A var-
iable pitch angle control system can be either of two main cate- 1.1. Analytical and numerical studies for investigating the
gories: passive or active [8]. In a passive variable pitch system, the aerodynamic behavior of VAWTs
blade is free to pitch about its axis near the leading edge. Theo-
retically, a passive pitch system can greatly improve the starting Numerous analytical, numerical, and experimental methods
torque and performance of a VAWT. However, experimental anal- have been employed for studying the flows around VAWTs and
ysis has shown that the maximum Cp value is less than a third of the their aerodynamic behavior, both with fixed and variable pitch
predicted value [7]. In an active control system, pitch control angles [13]. A double-multiple stream-tube model (DMST) was
mechanisms such as pushrods, cams, or servomotors have been proposed by Paraschivoiu [14]. Paraschivoiu and Saeed [15]
designed to achieve continuous changes in the blade pitch ampli- analyzed an H-type Darrieus VAWT in order to determine the
tude [9]. At low TSRs, a large pitch amplitude is needed in order to optimal variation of the pitch angle of the blades by using two si-
reduce the angle of attack and hence to enhance rotor performance. nusoidal analytical functions. The results of their study demon-
strated that the use of a variable pitch technique can increase
annual energy production by about 30%. Jain and Abhishek [9] also
used DMST for predicting the aerodynamic performance of a VAWT
with dynamic blade pitching (sinusoidal blade pitching). They
concluded that pitch amplitudes should be high (z35 ) for TSRs
below 0.5 and should be reduced to approximately 10 for TSRs
greater than 2.0.
High performance computing systems have recently been
employed as a means of addressing a number of engineering
problems. Computational fluid dynamics (CFD) is a powerful nu-
merical tool for acquiring an understanding of the aerodynamic
response of VAWTs through the solving of the Navier-Stokes
equation, and several studies have reported the use of CFD for
investigating the aerodynamics of VAWTs. Balduzzi et al. [16]
Fig. 1. Forces and velocities acting on the blade of a Darrieus turbine. examined some of the model parameters, such as the dimensions
 G. Abdalrahman et al. / Renewable Energy 114 (2017) 1353e1362 1355

of domains, mesh sizes, and turbulence models, in order to examine simulation of both fixed and variable pitch angle 2-D three-bladed
their influence on the performance of a 2D H-type VAWT with a H-Darrieus VAWTs in order to predict their aerodynamic perfor-
fixed pitch angle. Their CFD results agreed well with the experi- mance. The flow was also analyzed using both multiple reference
mental data. Chen and Zhou [17] used the ANSYS Fluent CFD frame (MRF) and sliding mesh (SM) techniques. The results of the
package to assess the aerodynamic performance of a VAWT by final revolution were chosen for the following analysis. The data-
setting the pitch angle of the blades in the range of b ¼ 5 e5 , base generated by ANSYS Fluent was then used for the design of the
which in turn, changed the angle of attack. A 2D study was con- proposed blade pitch control algorithm. Table 1 shows the main
ducted on a three-bladed VAWT with an NACA 0018 airfoil. It was dimensions and parameters of the CFD model, which were based
shown that the power coefficient increased slightly for pitch angles on previous studies [16,19].
from 5 to 2 . Performance was also improved by 7% at b ¼ 2
compared to the fixed pitch angle case (i.e., b ¼ 0 ). Miau et al. [18] 2.1. Computational domains
argued that the aerodynamic characteristics of a small three-bladed
VAWT with an NACA 0015 airfoil can be improved with the use of a The H-type VAWT CFD model used in the study presented in this
pitch control strategy. They examined different pitch angles paper consists of three main subdomains: a stationary subdomain,
numerically and validated the findings experimentally. The results a rotating subdomain, and a blade subdomain [16]. Fig. 2 shows
showed that, compared to a fixed pitch angle, their proposed var- these computational subdomains and their dimensions.
iable pitch control strategy enhanced VAWT performance for a
variety of TSRs. Using the ANSYS CFX CFD package, Sumantraa et al.
2.2. Grid generation
[19] employed numerical techniques to focus on the effect of a
preset pitch angle on the performance of an H-type VAWT. The
The accuracy of the model results is sensitive to the size and
model simulated and analyzed three pitch angles (6 , 0 ,
distribution of the mesh. In the 2-D simulations for this study, to
and þ6 ), with varied TSRs and wind speeds. For each pitch angle
resolve the boundary layer of the airfoils, the mesh is very fine
configuration, flow field characteristics were investigated, and the
around the blades, as can be seen in Fig. 3 (a) and (b). To reduce the
power curves were compared. The authors concluded that the best
computational cost of the overall CFD model, a close-to-equilateral
performance occurred at a pitch angle of 6 for all TSRs and wind
coarse mesh is generated in the stationary domain, as shown in
speeds. Erickson et al. [20] explored the impact of sinusoidal pitch
Fig. 3 (c). The inlet and outlet boundary conditions are also defined
actuation on the performance of a high-solidity VAWT in a wind
as a velocity inlet (u∞ ¼ 10 m/s) and pressure outlet (atmospheric
tunnel over a wide range of design and operational conditions.
pressure), respectively.
Their experimental work was conducted using a cam and control
The CFD model was validated through comparison with pub-
rod mechanism. The variable pitch VAWT achieved a 35% greater
lished wind tunnel experimental and CFD results for an H-type
power coefficient than the fixed blade configuration.
Darrieus VAWT [16]. Because the experimental data are available
only for a fixed pitch angle, the CFD simulation results were vali-
1.2. Studies of wind turbine pitch control system design
dated only for the fixed pitch angle case. The 2-D H-type VAWT
model with a fixed pitch angle that was simulated using a CFD
Numerous studies have suggested a variety of pitch angle con-
software package and reported by Balduzzi et al. [16] was repro-
trol methods. Pitch angle controllers based on proportio-
duced with some differences, such as the shapes of the computa-
naleintegralederivative (PID) and linear quadratic Gaussian (LQG)
tional subdomains and the number of elements. To reduce the
control have been used for power regulation. These linear control
computational cost, some parameters, such as the number of iter-
methods offer only limited performance, especially in the face of
ations and the time step size, were also reduced in the ANSYS
the uncertainties associated with the nonlinear dynamic properties
Fluent setup used in this work.
of wind turbines [21]. Sliding-mode control techniques have also
been applied for pitch angle control. These techniques provide
2.3. Identification of the desired blade pitch angles based on the
robust performance but require knowledge of the dynamic model
CFD model
of system [22]. Intelligent control methods that employ approaches
such as neural networks and fuzzy logic have also been proposed
In the study presented in this paper, the analysis was carried out
for the modeling and control of nonlinear dynamical systems [23].
for pitch angles of b ¼ 6 , 4 , 0 , 4 , and 6 ; TSRs of l ¼ 1,1.7,2,2.5,
An artificial neural network (ANN) can not only estimate a number
and 3.3; and a mean wind speed of u∞ ¼ 10 m/s. Fig. 4 shows the
of nonlinear functions based on the information available for
power coefficient (Cp) curves at the different pitch angles, including
training but also provide a high degree of accuracy under specific
the fixed pitch angle case (b ¼ 0 ). The predicted curves were then
system conditions [24]. All of the above research related to pitch
compared with other published experimental and CFD results with
control systems has been conducted for HAWTs. Unfortunately, to
respect to the power coefficient.
the authors' knowledge, the published literature contains no
Although the wind tunnel prototype (i.e., the experimental
studies related to control of the pitch angle in Darrieus VAWTs.
study) is a 3-D model, the predicted power curves obtained for this
Sargolzaei [25] applied ANNs for predicting the power coefficient
and torque for Savonius VAWTs that have six different shapes of
rotor blades at different Reynolds numbers. The simulation results, Table 1
which were compared with experimental data, revealed that the Main features of the Darrieus wind turbine analyzed in the present study.
use of the proposed ANN technique led to an increase in power.
Feature Value

Rotor radius (R) [mm] 850

2. Computational fluid dynamics (CFD) model
Blades number (Nb) [] 3
Blade profile NACA 0018
This section introduces the modeling of a 2-D H-type VAWT Chord (c) [mm] 246
using a CFD method. CFD techniques have recently been applied to Pitch angle (b) [degree] 6, 4, 0, 4, 6
the study of the aerodynamic behavior of VAWTs [11,26]. In the Azimuth angle (q) [degree] 0 to 360
Tip speed ratio (l or TSR) [] 1,1.7,2,2.5,3.3
study presented in this paper, ANSYS Fluent was utilized for the
1356 G. Abdalrahman et al. / Renewable Energy 114 (2017) 1353e1362

Fig. 2. Main computational subdomains used in ANSYS DesignModeler.

conditions.
In the CFD models developed for this study, these angles might
not be large enough to cause a dynamic stall. It could be observed
that the Cp peaks for all curves occur at TSRs between 2 and 2.5 (i.e.,
the optimal TSR, or lopt). As well, Table 2 shows that, with positive
pitch angles, positive torque is generated at a low TSR of l ¼ 1.
Fig. 5 shows the desired pitch angles (bopt) that provide the
highest Cp value at the different TSRs, as a function of the position
angle (q).

3. Modeling of the H-type VAWT

Examining the effect of pitch angle on VAWT power output

required the building of a dynamic model of the VAWT. Several
Fig. 3. Mesh structure of the Darrieus wind turbine modeled for this study. studies [27e29] describe dynamic wind turbine models, but their
primary focus was the modeling of HAWTs.
The main wind turbine model consists of the following
0.35 components:
0.3
a. Plant:
0.25  Aerodynamic block: Turbine rotor and blades
 Mechanical block: Shaft and gearbox unit (drive-train)
Power coefficient Cp

0.2
 Electrical block: Induction generator
0.15 b. Control block (pitch angle control system)
0.1
Fig. 6 is a block diagram of the H-type VAWT modeled for this
0.05 study, which was based on the HAWT model described in Ref. [27].
The H-type VAWT model was implemented using Matlab/Simulink.
0
In the wind speed component of the model, two cases are
0.5 1 1.5 2 2.5 3 3.5
-0.05 considered individually in order to represent the wind speed ðu∞ Þ:
(1) a constant wind speed and (2) a constant wind speed on which
-0.1
TSR random Gaussian noise is superimposed. The constant wind speed
Pitch=-6 deg Pitch=-4 deg Pitch=0 deg investigated for both cases is the same as the inlet wind speed in
Pitch=+4 deg Pitch=+6 deg Exp. Data [16] the CFD model (i.e., u∞ ¼ 10 m=s). The power extracted from the
CFD Results [16] wind is proportional to the cube of the wind speed and can be
expressed as follows [27]:
Fig. 4. Comparison of the CFD results for this study with other published experimental
and CFD results, with respect to the power coefficient versus the TSR at different pitch
angles.
Table 2
Power coefficients (Cp) at different TSRs and pitch angles.

work follow a trend that is similar to the published experimental TSR Pitch angles (degree)
and CFD data. For the fixed pitch angle case, good agreement 6 4 0 (fixed) 4 6
among all results was observed at TSR values greater than 2.5.
1 0.01519 0.01201 0.00125 0.014456 0.013229
However, a discrepancy can be noted in the peak values of the 1.7 0.040504 0.0643 0.205944 0.144615 0.089795
power curve: the CFD results overestimate the value of Cp. This 2 0.140547 0.145001 0.292979 0.265298 0.111846
behavior is probably due to the differences evident between the 2.5 0.145308 0.181745 0.244328 0.242333 0.199989
angles of attack for the computational and experimental 3.3 0.05182 0.046705 0.056599 0.044126 0.012437
 G. Abdalrahman et al. / Renewable Energy 114 (2017) 1353e1362 1357

Pitch angle varia ons at TSR=1 Pitch angle varia ons at TSR=1.7
8 (a) 8
(b)
6 6

Pitch angle (deg.)

Pitch angle (deg.)
4 4
2 B1 B1
2
0
B2 0 B2
2 0
-2 100 200 300 400
0 100 200 300 400
B3 -2 B3
-4
4
-4
-6
6
-6
-8
8 Azimuth angle (deg.)
Azimuth angle (deg.)

Pitch angle varia ons at TSR=2 Pitch angle varia ons at TSR=2.5
8
8
(c) 6
(d)
6

Pitch angle (deg.)

Pitch angle (deg.)

4 4
2 2 B1
B1
0 0
B2 B2
2 0
-2 100 200 300 400 2 0
-2 100 200 300 400

-4
4 B3 -4
4 B3
-6
6 -6
6
-8
8 -8
8
Azimuth angle (deg.) Azimuth angle (deg.)

Pitch angle varia ons at TSR=3.3

8
6
(e)
Pitch angle (deg.)

4
2 B1
0
B2
2 0
-2 100 200 300 400
-4
4 B3
-6
6
-8
8
Azimuth angle (deg.)

Fig. 5. Optimum pitch angles for the three blades of the H-type VAWT at different TSRs.

Fig. 6. Main components of the H-type VAWT modeled for this study.

In fact, only a fraction of this power can be captured by the

1
PW ¼ rAu3∞ (5) turbine, representing the mechanical power from the wind turbine
2 rotor (Pm ), defined as in Ref. [27]:
1358 G. Abdalrahman et al. / Renewable Energy 114 (2017) 1353e1362

1
Pm ¼ rAu3∞ Cp ðl; bÞ (6)
2

where Cp is the power coefficient. Based on the CFD results, It can be

seen that the power coefficient of the H-type VAWT is a function of
the tip speed ratio (l) and the blade pitch angle (b), including the
fixed pitch angle case (b ¼ 0 ). The CFD results are also used for
determining the optimum blade pitch angles (bopt) of the H-type
VAWT and their maximum power coefficients (Cp,max) at different
TSRs. To keep the power coefficient (Cp) close to its maximum value
(Cp,max) at a given wind speed, the corresponding TSR should also be
close to its optimal value (lopt), in the range of 2e2.5, as shown in
Fig. 4. The CFD results are then mapped to the aerodynamic model
of the H-type VAWT rotor (i.e., “aerodynamic block” in Fig. 6),
which is used for calculating the aerodynamic torque of that rotor Fig. 8. MLP-ANN mapping of the power coefficient (Cp) for a variable pitch angle
(tr ). VAWT.

3.1. Mapping of the H-type VAWT rotor (aerodynamic block) relationship between the input and the output.
A LUT is also employed for estimating the optimum pitch angle,
The aerodynamic models of the H-type VAWT rotor for the cases which is then used as MLP-ANN input, as shown in Fig. 8. These
of both fixed and variable pitch angles are formulated based on the angles also function as reference signals for the pitch angle control
CFD results. The dynamic models can be identified by mapping the system (Fig. 9). More details about the modeling of the other blocks
system input to the system output [30]. In this research, the rotor can be found in Ref. [27].
part of the H-type VAWT is identified using both a look-up table
(LUT) and a multilayer perceptron artificial neural network (MLP-
ANN) for cases involving fixed and variable pitch angles, respec- 3.3. Proposed H-type VAWT blade pitch control system
tively. The rotor parameters available from the CFD results, such as
power coefficients (Cp), TSRs, azimuth angle (q), and pitch angles This paper proposes blade pitching as a means of improving the
(b), are used as additional input and output data. performance of an H-type VAWT. The H-type VAWT model shown
To estimate the average power coefficients (Cp,avg) over three in Fig. 6 is nonlinear because the rotor efficiency, which is depen-
blades, a LUT is implemented for each blade using two types of CFD dent on the rotor speed, the wind speed, and the blade pitch angle,
input (TSR, q) and one type of CFD output (Cp), as illustrated in Fig. 7. is highly nonlinear. Therefore, linear control methods alone might
The average power coefficient (Cp,avg) is used for determining the not be sufficient for ensuring the desired performance. However,
aerodynamic torque of the H-type VAWT rotor (tr). the three blades of the H-type VAWT are controlled individually
using both a global ANN and PID controllers, as shown in Fig. 9.
3.2. Mapping of a variable pitch angle H-type VAWT rotor Multiple-input multiple-output (MIMO) MLP-ANNs are pro-
posed for designing an intelligent nonlinear blade pitch control
Although an LUT can provide reasonably accurate results system for an H-type VAWT. Introduced as input to the proposed
_ the refer-
MLP-ANN controller are the error (e), its derivative (e),
through linear interpolation, it is difficult to extend its use for more
than two input items. In the variable pitch angle H-type VAWT ence power (Pref), the tip speed ratio (TSR), and the azimuth angle
model, the input consists of the TSR, q, and the optimum pitch (q), while the output consists of the pitch angle command to sup-
angles (bopt), while the output is only the maximum power coeffi- plement the linear control command for each blade.
cient (Cp,max) for each blade. In this case, the MLP-ANN is utilized in
order to approximate this H-type VAWT nonlinear system with a
variable pitch angle. The MLP-ANN output is an estimate of the
power coefficient for each blade at different operating conditions.
Specifically, the MLP-ANN is trained using I/O data obtained from
the CFD simulations in order to describe the aerodynamic proper-
ties of the rotor of the H-type VAWT, which represent a nonlinear

Fig. 9. Proposed pitch control system: (a) MLP-ANN pitch control system, (b) PID pitch
Fig. 7. LUT-mapping of the power coefficient (Cp) for a fixed pitch angle VAWT. control system.
 G. Abdalrahman et al. / Renewable Energy 114 (2017) 1353e1362 1359

As shown in Fig. 9 (a), the error (e) is defined in terms of the

reference and actual electrical power generation (Pg ), as follows:

e ¼ Pref  Pg (7)

(i) Using an MLP-ANN for mapping a variable pitch H-type

VAWT rotor

For each blade, a fully connected two-layer feedforward MLP-

ANN with three types of input and one type of output is used for
mapping the blade dynamics when the pitch angle is variable. The
numbers of nodes in the hidden and output layers are 20 and 1,
respectively. The features of the MLP-ANN are the given TSRs, azi-
muth angle (0 e360 ), and the optimum pitch angles (bopt). How-
ever, the maximum power coefficient (Cp,max) is chosen as the
output data (target). These data sets are also divided into two parts:
75% for training and the remaining 25% for testing. The Levenberg-
Marquardt back propagation algorithm is preferred as the training
method because it can provide accurate predictions [31], and the
sigmoid function is used as the activation function for the neurons
in the hidden layer.

(ii) Using the MLP-ANN for the pitch angle control system

The MLP-ANN method was also adopted for designing the

intelligent control block depicted in Fig. 9 (a) as the blade pitch
angle controller for the H-type VAWT. The structure of the layered
feedforward MLP-ANN for the control system is similar to that of
the algorithm discussed in Section (i), but the training data differ
with respect to input and output. Because a well-trained ANN with
the correct structure is important for capturing the I/O relationship,
the MLP-ANN model is trained based on more than 250,000 data
points generated from the CFD calculations. Fig. 10 presents the
mean square error of the training process for the MLP-ANN model.

4. Numerical results

Fig. 11 (a) shows that the LUT can estimate the power co-
efficients (Cp) accurately at the mean wind speed (u∞ ¼ 10 m/s). The

Fig. 11. Mapping of the power coefficient (Cp) at a wind speed of 10 m/s by (a) the LUT
for a fixed pitch angle VAWT and (b) the MLP-ANN for a variable pitch angle VAWT.

LUT is implemented using the exact CFD results, using the azimuth
angle (0 e360 ) and optimum TSR (lopt ¼ 2.5) as input. As a result,
there is no error between the value of Cp obtained from the CFD
model and the value of Cp based on the LUT.
The power coefficient for a variable pitch angle H-type VAWT
rotor was mapped using the MLP-ANN technique in order to esti-
mate the value of Cp, as shown in Fig. 11 (b). In this case, the mean
square error (MSE) is equal to 2.62  104.

4.1. Results for a pitch angle control system

A blade pitch angle control strategy based on both PID and MLP-
ANN controllers is proposed for generating the power output under
different operating conditions, as shown in Fig. 9. The gains of the
Fig. 10. Performance of the MLP-ANN model during the training process. PID feedback control system are chosen as Kp ¼ 0.00953,
1360 G. Abdalrahman et al. / Renewable Energy 114 (2017) 1353e1362

Ki ¼ 8.583  105, and Kd ¼ 0.00192 for proportional, integral, and

derivative gains, respectively. The performance of the proposed
control system is evaluated at a uniform wind speed at a discrete
time of Ts ¼ 2  105 s.
For the case of a uniform wind speed of u∞ ¼ 10 m/s, Fig. 12
shows the reference pitch angle and command pitch angle pre-
dicted by both the MLP-ANN and PID controllers for each blade. It
can be observed that, in general, the MLP-ANN controller produced
some overshoots, possibly attributable to outliers in the training
data. However, it also resulted in a shorter control rise time (Tr ),
which is the time required for the signal to change from a given low
value to a given high value, and is also used for measuring the
performance of the control system. The rise time values for both the
PID and MLP-ANN pitch control systems are listed in Table 3.
Table 4 shows the MSE between the reference pitch angle and
the control pitch signals for a uniform wind speed over the three
blades. The MSE for each blade can be calculated as follows:

n  2
1X
MSE ¼ b  bi (8)
n i¼1 ref ;i

where n is the number of samples generated by the model simu-

lation. Although the PID control is a feedback control system, the
MSE values for the MLP-ANN controller are less than those for the
PID controller. Consequently, with respect to rise time and MSE, the
performance of the MLP-ANN controller is superior to that of the
PID-based controller.

4.2. Power output results

A comparison of the power output achieved by the PID and MLP-

ANN controllers at uniform and variable wind speeds is presented
in Fig. 13, which reveals that the Pref curve and the power genera-
tion (Pg) curves for both control scenarios are close to each other.
Although the ability to track the desired pitch angle is comparable
for both controllers, the improvement achieved when the MLP-
ANN is used is clearly observable in the power generation (Pg)
curves, where the effect of variations in the pitch angle associated
with both controllers, as shown in Fig. 12, can be clearly seen. Small
discrepancies between the actual and reference power waveforms
first occur during the second half of the simulation when the time is
greater than approximately 36 s. These discrepancies might be due
to the large changes in the pitch angles for both the second
(from 4 to 6 ) and third (from 0 to 6 ) blades that occur at the
same time (z37.6 s). However, the discrepancies are overcome
when the MLP-ANN control system is used for the uniform wind
speed case. An additional observation is that, for all curves, the
peaks of the output power occur when the pitch angles of any two
blades are positive. This finding means that the positive pitch an-
gles decrease the angle of attack and that the lift forces then in-
crease accordingly.
For both wind speed cases, the simulation results for the fixed
pitch and variable pitch H-type VAWTs were also compared and are
shown in Fig. 13. Based on the individual blade pitching, the per-
formance of the H-type VAWT is improved in terms of power
output. The H-type VAWT with the pitch control systems depicted
in Fig. 9 can harness more wind energy under different operating
conditions. In fact, compared to the fixed pitch VAWT cases, the
power output is increased by around 29% for both wind speed
scenarios.
The areas below all of the power curves displayed in Fig. 13
represent the total gross wind power harnessed by the H-type Fig. 12. Response of the pitch angle control system at a uniform wind speed of 10 m/s.
VAWT for each of the four operating scenarios. The results are listed
in Table 5.
 G. Abdalrahman et al. / Renewable Energy 114 (2017) 1353e1362 1361

Table 3
Rise time for both the MLP-ANN and PID pitch control systems at a uniform wind
speed.

Rise time (Tr, sec) Uniform wind speed at 10 m/s

MLP-ANN controller PID controller

Blade 1 0.0518 0.8064

Blade 2 0.0852 0.7899
Blade 3 0.0418 0.0788

Table 4
MSEs for both the MLP-ANN and PID control systems at a uniform wind speed.

(MSE) Uniform wind speed

MLP-ANN controller PID controller

Blade 1 0.0040 0.0039

Blade 2 0.0080 0.0314
Blade 3 0.0140 0.0266

As well, for both the variable and fixed pitch angle models, the
area between the power output curves represents the additional
energy (Ea ) harnessed through pitching. It can be calculated for
both wind speed cases (uniform and variable) as follows:
For uniform wind speed using MLP-ANN results:

Eauni ¼ Evuni  Efuni ¼ 62640 Watt$sec (9)

For variable wind speed using MLP-ANN results:

Eavar ¼ Evvar  Efvar ¼ 66250 Watt$sec (10)

Based on Equations (9) and (10), the results are roughly equal,
which suggests that the simulation model incorporating different
pitch control systems is able to predict the desired power output for
different operating conditions. In addition, power analysis reveals
that about 1% of the total gross power (Evuni ) is required for the
control devices.

5. Conclusion

Small- and medium-sized VAWTs can be utilized effectively as

stand-alone wind energy generation sources if their efficiency can
be further enhanced. This paper proposes intelligent control tech-
niques based on a neural network approach for designing an in-
dividual active blade pitch control system for an H-type VAWT as a
means of improving its power generation performance. In the work
Fig. 13. Power output from fixed and variable pitch angle H-type VAWTs at uniform
presented in this paper, a computational fluid dynamics (CFD) and variable wind speeds whose mean value is 10 m/s.
package, ANSYS Fluent, is employed for analyzing the performance
of a 2-D variable pitch angle H-type Darrieus VAWT at different
TSRs. For validation purposes, the resulting power coefficient (Cp) Table 5
Potential wind energy predicted for fixed and variable pitch angle H-type VAWTs at
value has been calculated for each case and then compared with
uniform and variable wind speeds whose mean value is 10 m/s. Note that Ref. data,
published experimental and CFD results. The pitch angles that MLP-ANN and PID are only for Evuni and Evvar .
maximize the power coefficients have been identified for a variety
Energy Uniform wind Variable wind Speed
of TSRs. The CFD results have then been adopted for designing
Watt.sec Speed (10 m/s)
control parameters for an H-type VAWT using Matlab\Simulink.
Because of the mathematical complexity associated with modeling Fixed pitch Variable pitch Fixed pitch Variable pitch
angle Efuni angle Evuni angle Efvar angle Evvar
the dynamic response of a VAWT rotor, the CFD results have been
applied for mapping fixed and variable pitch angle system models Ref. data 1.48eþ05 2.1316e þ 05 1.51e þ 05 2.1713e þ 05
MLP-ANN 2.1116e þ 05 2.1650e þ 05
of the H-type VAWT rotor using a LUT as well as an MLP-ANN. A
PID 2.0692e þ 05 2.1354e þ 05
nonlinear controller based on a neural network (MLP-ANN) has also
been developed for controlling the blade pitch angle of a Darrieus
H-type VAWT. To investigate the robustness of the proposed
intelligent blade pitch control system, a conventional controller significant improvement with respect to the power output from the
(PID) was also implemented to control each blade individually. The H-type VAWT compared to that provided by the linear PID control
blade pitch control system based on the MLP-ANN produced a method. The results also reveal that, compared to fixed pitch angle
1362 G. Abdalrahman et al. / Renewable Energy 114 (2017) 1353e1362

operation, the blade pitch angle technique clearly increases the symmetric NACA 4-digit airfoils on self starting capability of a 1kW H-type
vertical Axis wind turbine, Int. J. Mech. Eng. Appl. 3 (2015) 7e16.
power output of the H-type VAWT, with a percentage improvement
[14] I. Paraschivoiu, Wind turbine design: with emphasis on Darrieus concept,
that exceeded 25% in some of the simulated cases. Presses inter Polytechnique, 2002.
[15] I. Paraschivoiu, O. Trifu, F. Saeed, H-Darrieus wind turbine with blade pitch
Acknowledgment control, Int. J. Rotating Mach 2009 (2009) 1e7.
[16] F. Balduzzi, A. Bianchini, R. Maleci, G. Ferrara, L. Ferrari, “Critical issues in the
CFD simulation of Darrieus wind turbines, Renew. Energy 85 (2016) 419e435.
The authors would like to take this opportunity to express our [17] W.Y. Chen, C.Y. Zhou, Application of numercal simulation to obtain the opti-
genuine thanks for the financial support of the Libyan Ministry of mization pitch angle for VAWT, in: 2009 World Non-grid-connected Wind
Power Energy Conf., 2009, pp. 110e114.
Higher Education This work was made possible by the facilities of [18] J.J. Miau, S.Y. Liang, R.M. Yu, C.C. Hu, T.S. Leu, J.C. Cheng, S.J. Chen, Design and
the Shared Hierarchical Academic Research Computing Network test of a Vertical-Axis wind turbine with pitch control, Appl. Mech. Mater. 225
(SHARCNET: www.sharcnet.ca) and Compute Canada (www. (2012) 338e343.
[19] R.B. Sumantraa, S. Chandramouli, T.P. Premsai, P. Prithviraj, M. Vivek,
computecanada.ca). V.R. Kishore, Numerical analysis of effect of pitch angle on a small scale
vertical Axis wind turbine, J. Renew. Energy Res. 4 (no. 4) (2014) 929e935.
References [20] D.W. Erickson, J.J. Wallace, J. Peraire, Performance characterization of cyclic
blade pitch variation on a vertical Axis wind turbine, in: 49th AIAA Aerospace
Sciences Meeting, Including the New Horizons Forum and Aerospace Expo-
[1] W. Peter, W. Uli, Physics of Solar Cells: from Basic Principles to Advanced
sition, 2011, pp. 1e23.
Concepts, third ed., John Wiley & Sons, 2016.
[21] T.L. Van, T.H. Nguyen, D. Lee, S. Member, Advanced pitch angle control based
[2] Y. Chen, Numerical Simulation of the Aerodynamic Performance of an H-rotor,
on fuzzy logic for variable-speed wind turbine systems, IEEE Trans. Energy
Master's Thesis, University of Louisville, 2011.
Cone 30 (no. 2) (2015) 578e587.
[3] R. Howell, N. Qin, J. Edwards, N. Durrani, “Wind tunnel and numerical study of
[22] S. Lee, Y. Joo, J. Back, J.H. Seo, Sliding mode controller for torque and pitch
a small vertical Axis wind turbine, Renew. Energy 35 (no. 2) (2010) 412e422.
control of wind power system based on PMSG, in: International Conference on
[4] T.J. Carrigan, B.H. Dennis, Z.X. Han, B.P. Wang, “Aerodynamic shape optimi-
Control, Automation and Systems, 2010, pp. 1079e1084.
zation of a Vertical-Axis wind turbine using differential evolution, ISRN
[23] N.K. Kasabov, Foundations of Neural Networks, Fuzzy Systems, and Knowl-
Renew. Energy 2012 (2012) 1e16.
edge Engineering, The MIT Press, Cambridge, Massachusetts, 1998.
[5] E. Dyachuk, M. Rossander, A. Goude, H. Bernhoff, “Measurements of the
[24] O. Barambones, A robust wind turbine control using a neural network based
aerodynamic normal forces on a 12-kW straight-bladed vertical Axis wind
wind speed estimator, in: Inte. Joint Conf. On Neural Networks Ijcnn, 2010,
turbine, Energies 8 (no. 8) (2015) 8482e8496.
pp. 1e8.
[6] R. Gosselin, G. Dumas, M. Boudreau, “Parametric study of H-Darrieus Vertical-
[25] J. Sargolzaei, Prediction of the power ratio and torque in wind turbine Savo-
Axis turbines using uRANS simulations, Lmfn.Ulaval.Ca 178 (2013).
nius rotors using artificial neural networks, WSEAS Int. Conf. Renew. Energy
[7] R. Bos, Self-starting of a Small Urban Darrieus Rotor. Strategies to Boost Per-
Sources 1 (no. 2) (2007) 7e12.
formance in Low-Reynolds-number Flows, Master’s Thesis, Delft University of
[26] S. Lin, Y. Lin, C. Bai, W. Wang, Performance analysis of vertical-Axis-wind-
Technology, 2012.
turbine blade with modified trailing edge through computational fluid dy-
[8] S.S. Khalid, Z. Liang, S. Qi-hu, Z. Xue-Wei, “Difference between fixed and
namics, Renew. Energy 99 (2016) 654e662.
variable pitch vertical Axis tidal turbine-using CFD analysis in CFX, Res. J. Appl.
[27] M. Singh, Dynamic models for wind turbines and wind power plants, Natl.
Sci. Eng. Technol. 5 (no. 1) (2013) 319e325.
Renew. Energy Lab. 1 (no. May) (2011) 115.
[9] P. Jain, A. Abhishek, Performance prediction and fundamental understanding  Luna, J. Rocabert, D. Aguilar, G. Va
[28] A. Rola n, A. zquez, An approach to the
of small scale vertical Axis wind turbine with variable amplitude blade
performance-oriented model of variable-speed wind turbines, in: IEEE In-
pitching, Renew. Energy 97 (2016) 97e113.
ternational Symposium on Industrial Electronics, 2010, pp. 3853e3858.
[10] L. Lazauskas, Three pitch control systems for vertical Axis wind turbine
[29] X. Zhang, D. Xu, Y. Liu, Intelligent control for large-scale variable speed,
compared, Wind Eng. 16 (No. 5) (1992) 269e282.
J. Control Theory Appl. 3 (no. December 2003) (2004) 305e311.
[11] C. Li, S. Zhu, Y. Xu, Y. Xiao, 2.5D large eddy simulation of vertical Axis wind
[30] S. Mohanty, Artificial neural network based system identification and model
turbine in consideration of high angle of attack flow, Renew. Energy 51 (2013)
predictive control of a flotation column, J. Process Control 19 (no. 6) (2009)
317e330.
991e999.
[12] A. Bianchini, L. Ferrari, S. Magnani, Start-up behavior of a three-bladed H-
[31] S. Sapna, A. Tamilarasi, M. Pravin Kumar, Backpropagation learning algorithm
Darrieus VAWT: experimental and numerical analysis, in: ASME Turbo Expo:
based on levenberg marquardt algorithm, Comp. Sci. Inf. Technol. (CS IT) 2
Turbine Technical Conference and Exposition, 2011, pp. 811e820.
(2012) 393e398.
[13] C. Nguyen, T. Le, P. Tran, A numerical study of thickness effect of the

View publication stats