Original Article

Journal of Intelligent Material Systems

and Structures
2020, Vol. 31(1) 137–151
Optimal control of a broadband Ó The Author(s) 2019
Article reuse guidelines:
vortex-induced vibration energy sagepub.com/journals-permissions
DOI: 10.1177/1045389X19888711

harvester journals.sagepub.com/home/jim

E Azadi Yazdi

Abstract
A vortex-induced vibration energy harvester consists of a relatively long cylinder mounted on a flexible structure. In a
flow field, the periodically shedding vortices induce transverse vibrations in the cylinder that is converted to electricity
by means of piezoelectric generators. In most vortex-induced vibration harvesters, the output power is considerable
only in a narrow band around the wind speed where the vortex shedding frequency matches the natural frequency of
the structure. To overcome this limitation, a tuned mass mechanism is employed in the proposed vortex-induced vibra-
tion energy harvester that can change the natural frequency of the turbine to match the vortex shedding frequency in a
broad band of wind speeds. The tuned mass mechanism should work in close cooperation with the piezoelectric genera-
tors to maximize the electric power of the turbine. To this end, a nonlinear piezoaeroelastic model of the system is
derived, and a model predictive control technique is formulated to find the optimal control inputs for the tuned mass
actuator and the piezoelectric generators. Results of numeric simulations confirmed that the tuned mass mechanism not
only increases the velocity band over which the turbine is effective but also increases the peak power output of the tur-
bine by 294%.

Keywords
Model predictive control, optimal control, vortex-induced vibrations, tuned mass mechanism, lock-in phenomenon

1. Introduction cylinder that is either flexible or mounted on a flexible

structure (Chizfahm et al., 2018). The fluctuating cross-
Blunt bodies in a steady uniform fluid flow may peri- wise lift force due to the shedding vortices induces
odically shed vortices near their surface. The fluctuat- transverse vibrations in the VIV energy harvesters.
ing pressure field due to the vortices on the surface of Unlike the conventional energy harvesters, the VIV
flexibly mounted bodies yields vortex-induced vibra- energy harvester does not have any moving blades, and
tions (VIVs). Employing the VIV concept, energy har- therefore the rotary bearing and transmission gear box
vesters have been developed that replace batteries to are eliminated in this turbine. Consequently, the manu-
reduce the maintenance cost of low-power devices such facturing and maintenance costs are relatively lower for
as medical implants (Amin Karami and Inman, 2012), VIV energy harvesters. It is expected that a VIV energy
wireless sensors (Roundy and Wright, 2004), and cam- harvester with the height of 150 m can deliver 1 MW
eras (Abdelkefi and Ghommem, 2013). Various struc- of electrical power (Cajas et al., n.d.). However, several
tures have been studied for VIV energy harvesters technical challenges need to be addressed to achieve
including vibrating circular cylinders (Akaydin et al., such a high-power VIV energy harvester.
2012), fluttering airfoils (Abdelkefi et al., 2012b, 2012c; The main challenge in developing a VIV energy har-
Erturk et al., 2010), and galloping sliding structures vester is to achieve a high power generation at a broad-
(Abdelkefi et al., 2012a, 2013; Sirohi and Mahadik, band of wind speeds. It is reported in several VIV
2011; Yang et al., 2013).
Although most of the research studies are restricted
to low power output (less than 1 W) energy harvesters, School of Mechanical Engineering, Shiraz University, Shiraz, Iran
recently VIV wind turbines with a rated mechanical
Corresponding author:
power output of 100 W have been proposed in E Azadi Yazdi, School of Mechanical Engineering, Shiraz University,
Chizfahm et al. (2018). The proposed VIV energy har- Mollasadra St., Shiraz 7193616548, Iran.
vesters consist of a relatively long straight/conical Email: [email protected]
138 Journal of Intelligent Material Systems and Structures 31(1)

research (see, e.g., Abdelkefi et al., 2012a; Akaydin the electric circuit and control algorithm. Most of the
et al., 2012; Hasheminejad et al., 2017) that the VIV studies in VIV energy harvesters used a proportional
power generation is considerable when the vortex shed- controller in the form of a constant resistive electrical
ding frequency is close to the structural natural fre- load (see, e.g., Abdelkefi et al., 2012b; Dai et al.,
quency, that is known as the lock-in phenomenon. The 2014a). The achievable efficiency is very low with con-
vortex shedding frequency for bodies with circular stant resistive loads; in various studies a typical aero-
cross-sections depends linearly on the free-stream wind mechanical efficiency of less than 5% was reported for
speed (Bishop and Hassan, 1964). As the wind speed this approach. It was shown in Vicente-Ludlam et al.
deviates from the one that corresponds to the lock-in (2014) that there exists a theoretical limit on the maxi-
phenomenon, the power generation of the harvester mum achievable efficiency by a constant resistive load.
drops significantly (Abdelkefi et al., 2012a, 2012c; Meliga et al. (2011) proposed a dissipative velocity
Akaydin et al., 2012; Chizfahm et al., 2018). Therefore, feedback control strategy for an flexibly supported
the energy generation of the VIV energy harvesters in cylinder. The main objective of the dissipative velocity
general, and the VIV energy harvesters in particular, is feedback controller is to suppress the vibrations of the
significant only in a narrow band of the wind speeds cylinder, meanwhile it produces a small amount of
around the lock-in wind speed. energy. An adaptive fuzzy sliding-mode controller was
Among several research that reported the degrada- developed in Hasheminejad et al. (2014, 2017) for elas-
tion of the VIV power generation and efficiency outside tically mounted cylinders with circular and elliptical
the lock-in region, only a few have proposed a solution cross-sections. The control objective was VIV suppres-
to overcome this limitation. The efficiency improving sion; however, the energy generation was also studied.
solutions for VIV energy harvesters can be grouped in The maximum achievable aero-mechanical efficiency
four categories: (1) optimizing the geometric shape and was reported to be 5%. Recently, Azadi Yazdi (2018)
orientation of the blunt body, (2) optimizing the electric proposed a nonlinear model predictive control (MPC)
circuit and control algorithm, (3) using nonlinear flex- technique that solely aims to maximize the power out-
ible support structures, and (4) using multiple sources put of a VIV energy harvester. It was shown that the
of kinetic energy. nonlinear MPC technique outperforms the conven-
Several efficiency improving solutions are focused on tional constant resistive load technique by achieving a
finding the optimal shape and orientation of the blunt maximum aero-mechanical efficiency of 32%.
body, because the vortex formation regime depends In the third approach for increasing the power out-
strongly on the geometry of the blunt body. In Kluger put of VIV energy harvesters, the flexible support struc-
et al. (2013), the geometric parameters of the blunt tures were modified. In Pan and Dai (2018), a
body are optimized to achieve the minimum lock-in composite laminate with a stacking sequence design
wind speed, enhance the amplitude of the VIV beyond was proposed for the support structure of a VIV energy
the lock-in wind speed, and minimize amplitude hyster- harvester. It was shown through numerical simulations
esis for varying wind speeds. In Yang et al. (2013), the that the natural frequency of the VIV harvester with
VIVs of blunt bodies of cross-sectional square, rectan- the stacking sequence support can be adjusted to lower
gle, triangle, and D-shape were compared. It was frequency ranges. In Zhang et al. (2017), a support
observed in the experiments that the square-sectioned structure with a nonlinear restoring force was intro-
blunt body outperforms other geometric shapes. Hu duced for a VIV energy harvester. Two repulsive mag-
et al. (2018) investigated the performance of a circular nets were attached on the lower and higher ends of a
cylinder VIV energy harvester with two small-size rod circular cylinder which is subjected to the wind.
attachments. The results of experiments showed that Experimental results confirmed the improvements in
triangular rods significantly extended the lock-in region the performance of the VIV energy harvesting system,
of the VIV energy harvester. Ding et al. (2015) used both in the width of the synchronization region and in
two-dimensional (2D) unsteady computational fluid the level of the harvested power, compared to the clas-
dynamics (CFD) simulations to study the energy gener- sical configuration.
ation of a cylinder with various cross-sectional shapes The last approach to improve the output power of
including square, Q-trapezoid, and triangular sections. the VIV energy harvesters is to use multiple sources of
It was concluded that the Q-trapezoid section exhibits kinetic energy. In Bibo and Daqaq (2013), a VIV har-
larger VIVs than the other sections. In conclusion, most vester was integrated with a vibratory harvester to con-
of the studies suggested non-axisymmetric geometries currently extract energy from wind and base vibrations.
for the blunt body. Although non-axisymmetric blunt The performance of the proposed device was shown to
bodies may improve the efficiency of the VIV energy be high under the combined loading. In Zhao and Yang
harvesters, unlike axisymmetric blunt bodies, their effi- (2018), a mechanical stopper is introduced into a VIV
ciencies are highly sensitive to the air flow direction. harvester in order to concurrently harvest energy from
Another approach to improve the power generation ambient vibrations and wind. In general, significant
and efficiency of VIV energy harvesters is to optimize improvement in the level of the output power can be
Azadi Yazdi 139

current is passed through the piezoelectric generator, its

voltage drops and vice versa. Therefore, both cases of
increasing and decreasing the generated current may
conflict with the objective of maximizing the electrical
energy output of the turbine. Hence, the determination
of an optimal control strategy for the current of the
piezoelectric patches is non-trivial.
In conclusion, this article has two major contribu-
tions. First, a TMM is proposed to enhance the fre-
quency lock-in phenomenon in VIV energy harvesting
devices. Consequently, the proposed VIV energy har-
vester is able to capture the wind energy for a broad-
band of wind speeds. Second, to maximize the output
electrical power of the turbine, an integrated optimal
control algorithm is implemented for the tuned mass
actuator and the piezoelectric patches. The proposed
optimal control approach respects the limitations of the
actuators and is suitable for real-time implementation.
The article is organized as follows. In section
‘‘Dynamic modeling of the VIV energy harvester,’’ the
nonlinear dynamic model of the VIV energy harvester
Figure 1. The VIV energy harvester: (a) geometrical model and with the TMM is presented. Furthermore, the lineariza-
(b) free-body diagram. tion and discretization of the nonlinear model are dis-
cussed. Section ‘‘MPC’’ describes the MPC design and
implementation. In section ‘‘Results and discussion,’’
numerical simulation results are presented. Effectiveness
attained by using multiple sources of kinetic energy.
of the proposed mechanism in increasing the efficiency
However, Dai et al. (2014a) showed that in some cases
of the turbine is thoroughly investigated. Conclusions
an increase in the level of the base vibrations may result
are presented in section ‘‘Conclusion.’’
in an unfavorable reduction in the VIV effects.
This article proposes a tuned mass mechanism
(TMM) along with an optimal control strategy to 2. Dynamic modeling of the VIV energy
improve the efficiency of a VIV energy harvester with a
cylindrical mast (Figure 1). The proposed TMM con-
harvester
sists of a linear actuator that moves a point mass inside The VIV energy harvester consists of a rigid cylindrical
the hallow mast. It will be shown through analytical mast that is placed on the tip of a flexible beam (Figure
modeling that the TMM can change the effective inertia 1(a)). In a flow field, the periodically shedding vortices
of the cylindrical mast and consequently adjust the nat- exert an oscillatory force on the mast. Due to the oscil-
ural frequency of the VIV energy harvester. Therefore, latory aerodynamic force, the energy harvester exhibits
the resonance can be enhanced by regulating the natu- VIVs. The vibration is converted into electrical energy
ral frequency of the energy harvester close to the vortex by two patches of bimorph piezoelectric generators of
shedding frequency. Consequently, the efficiency of the length Lp that are installed on the right and left sides of
energy harvester can be improved especially in pre- and the flexible stand. A tuned mass is installed inside the
post-lock-in wind speeds. hallow mast that can be moved along the axis of the
The proposed TMM is most effective when it works cylinder by means of a linear actuator. The tuned mass
in close cooperation with the power take-off unit that is responsible for changing the effective mass and iner-
converts the mechanical energy to electricity. In the tia of the mast, and consequently the natural frequency
proposed VIV energy harvester, the power take-off unit of the energy harvester.
consists of two patches of bimorph piezoelectric genera- The dynamic model of the VIV energy harvester is
tors that are installed on the sides of the flexible support derived in this section. The shear deformation and
of the turbine (Figure 1). An optimal control problem is rotary inertia of beam are neglected, because the length
formulated that maximizes the energy generation over of beam is much larger than its thickness. The aerody-
time by adjusting the position of the tuned mass and namic force on the flexible stand is neglected because it
the power extracted from piezoelectric patches. In prin- is narrow compared with the cylindrical mast. Due to
ciple, the electrical energy output of the VIV energy space limitations, some details of the derivation and
harvester is the time integral of the electrical power. verification of the dynamic model have been omitted.
Similar to other electrical energy generators, if a large Interested readers may refer to the study by Chizfahm
140 Journal of Intelligent Material Systems and Structures 31(1)

X
et al. (2018) for more details on the aero-elastic model Y ðx, tÞ = fi ð xÞyi ðtÞ ð3Þ
of the turbine.
where fi (x) is the ith mode shape, and yi (t) is the ith
modal response factor. The kinetic energy of the stand
2.1. Aerodynamic force model and the piezoelectric patches is given by
The amplitude of the VIVs, and consequently the power ð ð
2 2
transmitted to the structure, is considerable when the 2T1 = rs As Y ðx, tÞdx + rp Ap Y_ ðx, tÞdx
_
vortices are shed at a frequency close to the natural fre- Ls Lp
quency of the flexible structure, which is known as the XX ð
lock-in phenomenon. One of the most important chal- = r s As y_ i ðtÞ_yj ðtÞ fi ð xÞfj ð xÞdx
lenges in modeling of the VIV is finding an appropriate i j
Ls
expression for vortex-induced forces that captures the XX ð X
main dynamic features of the in VIV such as the lock-in + r p Ap y_ i ðtÞ_yj ðtÞ fi ð xÞfj ð xÞdx = Mi y_ 2i ðtÞ
phenomenon. Numerous aerodynamic force expres- i j i
Lp
sions have been proposed for an oscillatory cylinder in ð4Þ
the past few decades (Gabbai and Benaroya, 2005). A
widely used model for the aerodynamic lift on an oscil- where rs and rp are the mass densities, As and Ap are
latory cylinder in a 2D flow is the semiempirical wake the cross-sectional areas, Ls and Lp are the lengths, ssx
oscillator model (Bishop and Hassan, 1964). In this and sPx are the stresses in the x-direction, and esx and ePx
article, since the mast is relatively long, we may assume are the strains in the x-direction of the stand and the
a 2D flow for each slice of the mast (Olinger, 1998). Ðpiezoelectric substrates,
Ð Lp 2 respectively, and Mi =
Ls 2
According to the wake oscillator model, at a time t, the 0 f i (x)A s r s dx + 0 fi (x)Ap rp dx is the generalized
lift coefficient CL (x, t) of a slice of the mast at a height x mass of the ith mode.
is defined as (Bishop and Hassan, 1964) The potential energy of the stand and the piezoelec-
tric patches is given by
2a _
CL ðx, tÞ = Qðx, tÞ  Y ðx, tÞ ð1Þ ð ð ð
Dvs
2U1 = ssx esx d
V+ spx epx d
V E3 D3 d
V ð5Þ
where Q(x, t) is the excitation component, Y_ (x, t) is the 
Vs 
Vp 
Vp
transverse velocity, a is an empirical constant deter-
mined through experiments, vs :¼ 2pSU0 =D is the vor- where  V s is the volume of the stand, 
V p is the volume
tex shedding frequency, S is the Strouhal number of the piezoelectric, E3 and D3 are, respectively, the elec-
defined in Skop and Balasubramanian (1997), U0 is the tric field and the electric displacement in the piezoelec-
free-stream air speed, and D is the diameter of the mast. tric patches. The strains and stresses in the stand and
In equation (1), the excitation component, Q(x, t), satis- piezoelectric substrates are given by
fies the nonlinear van der Pol oscillator
∂2 Y ðx, tÞ
  esx = epx =  y ð6Þ
€ ðx, tÞ  vs G C 2  4Q2 ðx, tÞ Q_ ðx, tÞ
Q ∂x2
L0
2 vs F _ ssx = Es esx ð7Þ
+ vs Qðx, tÞ = Y ðx, tÞ ð2Þ  
D spx = Ep epx  d31 E3 ð8Þ
where G and F are empirical constants (Skop and
Balasubramanian, 1997) evaluated from experiments. where Es and Ep are Young’s moduli of the stand and
For stationary cylinders, the right-hand side of equa- piezoelectric layers at constant electric field, and d31 is
tion (2) vanishes; therefore, the solution to the afore- the strain coefficient of the piezoelectric patch. The elec-
mentioned equation is Q = CL0 sinvs t. This means that tric displacement is defined as
CL0 is the amplitude of the fluctuating lift coefficient on
D3 = Ep d31 epx + e33 E3 = e31 epx + e33 E3 ð9Þ
a stationary cylinder.
where e33 is the permittivity component at constant
strain, and e31 = Ep d31 is the stress coefficient of the
2.2. Piezoelastic model piezoelectric patch.
To obtain the piezoelastic model, the stand is modeled as If the right and left layers of the piezoelectric sheet
a multilayered clamped-free beam of length Ls which is are connected in series, the electric fields have the same
partially covered by piezoelectric patches on both sides. value and direction in both layers. The value of this
To obtain a finite dimensional model, the normal modes electric field for a piezoelectric layer of thickness tp is
expression is used for the deflection of the stand as given by (Dai et al., 2014b)
Azadi Yazdi 141

"  #
V ðtÞ 1 ∂Y ðLs , tÞ 2
E3 =  ð10Þ T2 = mTM x_ 2TM ðtÞ + + xTM ðtÞu_ m ðtÞ
2tp 2 ∂t
where V (t) is the generated voltage across the series ð15Þ
connection of the piezoelectric patches. Therefore, the U2 = mTM gxTM ðtÞ ð16Þ
potential energy is given by
where mTM is the mass of the tuned mass, and g is the
Ap e33 2 X gravitational constant.
2U1 =  V ðt Þ + Ki y2i ðtÞ + Gi yi ðtÞV ðtÞ ð11Þ
2tp i

where 2.4. Piezoaeroelastic model

ð ð This section presents the equations of motion of the
 2  2 vibrations of the proposed VIV energy harvester. The
Ki = Es Is f00i ð xÞ dx + Ep Ip f00i ð xÞ dx ð12Þ
wind flow is assumed to be uniform along the energy
Ls Lp
harvester and the three-dimensional boundary layer
ð effects are neglected on both ends. The diameter of the
1   00
Gi = d31 Ep + e31 Ip fi ð xÞdx ð13Þ mast is assumed to be constant, that is, D(x) = D.
2tp
Lp The non-conservative work consists of the work of
the electrical field, aerodynamic force, and tuned mass
where Is and Ip are the second moment of area of the actuator, given by
cross-section of the stand and piezoelectric patches with
respect to the Z axis, respectively. dW =  V dQR + FM dY ðLs , tÞ + MM duðLs , tÞ + FTM dxTM
ð17Þ
2.3. Tuned mass actuator The force FTM is the force exerted by the tuned mass
actuator, and FM and MM are, respectively, the force
The energy production of most VIV energy harvesters
and moment exerted from the mast on the tip of the
is substantial near the lock-in region; however, it is
stand (as shown in Figure 1(b)), given by
observed that the energy output drops sharply as the
wind speed deviates from the lock-in region (Abdelkefi
ð Lm
Ls +
et al., 2012a, 2012c; Akaydin et al., 2012; Dai et al., 1 r U 2 DC ðx, tÞdx  m €y
FM = L m m ð18Þ
2014b; Hasheminejad et al., 2017). Therefore, the har- 2 a 0
vester becomes practically ineffective at high winds due Ls

to the mismatch between the vortex shedding frequency ð Lm

Ls +
and the structural natural frequency. To overcome this 1
MM = r U 2 DCL ðx, tÞðx  Ls Þdx  mm L ym  Im €
m€ um
shortcoming, we propose to use a TMM inside the mast 2 a 0
Ls
that can change the generalized mass, and consequently
the natural frequency, of the VIV energy harvester. The ð19Þ
tuned mass can move along the axis of the mast by where ra is the air density, and for the center of mass
means of a linear actuator, for example, hydraulic of the mast, L m is the distance to the tip of the stand,
cylinder, pneumatic cylinder, ball-screw, or linear and Im is the moment of inertia about the z-axis. Also,
motor. For simplicity, in this article the linear actuator the displacement of the center of mass of the mast ym is
is represented by a control force, and its dynamic model given by
is not included in the governing equations of the VIV
energy harvester. Moreover, the TMM is assumed to Lm
be frictionless. ym ’Y ðx, tÞjx = Ls + um =
X X2
Suppose that the angular rotation of the mast, m
fi ðLs Þyi ðtÞ + L f0 i ðLs Þyi ðtÞ ð20Þ
um (t) :¼ (∂Y (x, t)=∂x)jx = Ls , is small ( tan (um )’um ), the
position vector of the tuned mass is given by The aerodynamic force and moment in FM and MM
are computed by assuming the mode shapes of Q(x, t) to
RTM=G = xTM ðtÞ^i + ðY ðLs , tÞ + xTM ðtÞum ðtÞÞ^j ð14Þ be the same as the mode shapes of the mast (Chizfahm
et al., 2018)
where xTM (t) is the displacement of the tuned mass rela-
X
tive to the tip of the stand, ^i and ^j are the unit vectors Qðx, tÞ = qi ðtÞ½fi ðLs Þ + ðx  Ls Þf0 i ðLs Þ ð21Þ
in the x and y directions, respectively. Therefore, for a
small um , the kinetic and potential energy of the tuned where qi (t) is the modal response factor of the fluctuat-
mass are given by ing lift excitation component.
142 Journal of Intelligent Material Systems and Structures 31(1)

Now, defining the Lagrangian as L = T1 + The dynamic model equations (23) to (25) can be
T2  U1  U2 , Euler–Lagrange principle can be used in written in state-space form as
the ~q :¼ (yi , X , xTM ) generalized coordinates, with
X_ = V , to obtain the piezoaeroelastic model of the VIV x_ ðtÞ = f ðxðtÞ, vðtÞÞ + BuðtÞ ð32Þ
energy harvester
where x(t) :¼ ½ y y_ V xTM x_ TM q q_ T is the
∂ ∂L  ∂L = dW state vector, v(t) :¼ ½ U0 g T is the exogenous input
ð22Þ
q_i ∂~
∂t ∂~ qi d~
qi vector that consists of the wind speed and the gravita-
tional constant, u(t) :¼ ½ i FTM T is the control input
For simplicity and without loss of generality, we only
vector that consists of the piezoelectric control
consider the first mode of the stand. From equations
current and the tuned mass actuator force, and the
(2) and (22), the piezoaeroelastic equations are given by
vector f (x(t), v(t)) and matrix B are determined
~ y + 2mTM ½uðLs Þ + xTM u0 ðLs Þu0 ðLs Þ_xTM y_ + from equations (23) to (25). The dynamic model can be
M€
linearized around an operating point x0 =
2a ~  gV = Gq
G_y + Ky ð23Þ ½ 0 0 0 x0TM 0 q0 0 T as
Dvs
 p V_ + g y_ =  iðtÞ
C ð24Þ x_ ðtÞ = Ac xðtÞ + Bc1 uðtÞ + Bc2 vðtÞ ð33Þ

FTM ðtÞ where

€xTM  u0 ðLs Þ½uðLs Þ + xTM u0 ðLs Þ_y2 =  g +
mTM 2 3
0 1 0 0 0 0 0
ð25Þ 6 K~ 7
2a g G
6  M~ ~0 G 0 0 0 7
The derivations of the aforementioned equations are 6 0 Dvs M ~0
M ~0
M 7
6 7
omitted due to space limitations. Here, u(x) :¼ f1 (x), 6 0  Cgp 0 0 0 0 0 7
6 7
y :¼ y1 (t), and i(t) is the current through the piezoelec- 6 7
Ac = 6 0 0 0 0 1 0 0 7
tric, and 6 7
6 0 0 0 0 0 0 0 7
6 7
6 7
ðs
L Lðp 6 0 0 0 0 0 0 1 7
4 5
~=
M u2 ð xÞAs rs dx + u2 ð xÞAp rp dx 0 vs F
0 0 0 v2s vs GCL20
D
0 0
 ð34Þ

2 m uðLs Þu0 ðLs Þ + L
+ mm u ðLs Þ + 2L 2 + I m u0 2 ðLs Þ " #T
m
mm 0 0  C1p 0 0 0 0
0 2 B c1 = ð35Þ
+ mTM ½uðLs Þ + xTM u ðLs Þ 0 0 0 0 1
0 0
mTM
ð26Þ " #
ð ð 0 0 0 0 0 0  4pS
D vs q0
2 2 Bc 2 = ð36Þ
~=
K Es Is ½u00 ð xÞ dx + Ep Ip ½u00 ð xÞ dx ð27Þ 0 0 0 0 1 0 0
Ls Lp
and M~ 0 is the value of M
~ in equation (26) obtained at
ð Lm
Ls + x = x0 .
1
G= r U 2D hð xÞ½uðLs Þ + ðx  Ls Þu0 ðLs Þdx
2 a 0
Ls 2.5. Discretization and augmentation
ð28Þ In order to formulate the finite horizon MPC design
Ð Lm
Ls + problem, the system must be presented in discrete-time.
h4 ð xÞdx A first-order hold (Franklin et al., 1998) for continuous
Ls
L = 4vs G L ð29Þ inputs u(t) and v(t), and a zero-order hold for states are
s Ð Lm
+
preferred here because they simplify the formulation of
h2 ð xÞdx
Ls the MPC cost function. Therefore, the discrete-time
    representation of the model (equation (33)) is given by
1
g= e31 wp tp + 2ts u0 Lp ð30Þ
2
xððk + 1ÞhÞ = ehAc xðkhÞ + aðBc1 uðkhÞ + Bc2 vðkhÞÞ
 p = e33 Lp wp
C ð31Þ + b½Bc1 Duððk + 1ÞhÞ + Bc2 Dvððk + 1ÞhÞ ð37Þ
4tp
where h is the sampling interval, and
where h(x) = f(Ls ) + (x  Ls )f0 (Ls ), and ws and wp are
the widths of the stand and piezoelectric layers,  hA 
a = A1
c e c I ð38Þ
respectively.
Azadi Yazdi 143

1 1 and the piezoelectric patches. In the following sections,

b= A ða  hI Þ ð39Þ
h c first, the optimal energy generation cost function asso-
Duððk + 1ÞhÞ = uððk + 1ÞhÞ  uðkhÞ ð40Þ ciated with each actuator is derived, then, the cost func-
tions will be integrated into the MPC synthesis
To convert equation (1) to the discrete-time linear problem.
state-space model, let us introduce the augmented dis-
crete state vector xk as
3.1. Tuned mass actuator
T
xk = ½ xðkhÞ uðkhÞ vðkhÞ  2 R n + nu + nv
ð41Þ In this article, the tuned mass actuator is employed to
adjust the natural frequency of the energy harvester by
The associated state-space equation is ~ in equation (23). To
changing the effective mass M
xk + 1 = Axk + B1 Duk + 1 + B2 Dvk + 1 ð42Þ maximize the energy generation, the tuned mass posi-
tion should be regulated such that the resonance is
where enhanced. It is reported in Azadi Yazdi (2018) that the
2 3 energy generation of the energy harvester is maximized
ehAc aBc1 aBc2 when the non-dimensional frequency O = vs =vn is
A=4 0 I 0 5 2 Rðn + nu + nv Þ 3 ðn + nu + nv Þ slightly greater than 1. To obtain a non-dimensional
0 0 I frequency of Odes , the equivalent mass M ~ should be
ð43Þ tuned to the following value
 
B1 = ½ bBc1 I 0 T 2 Rðn + nu + nv Þ 3 nu ð44Þ 2 2
M ~ Odes
~ des = K ~
=K
DOdes
ð47Þ
vs 2pSU0
B2 = ½ bBc2 0 I T 2 Rðn + nu + nv Þ 3 nv ð45Þ
To achieve the aforementioned value of M ~ des , the
The discrete output for MPC synthesis problem is des
desired position of the tuned mass xTM can be calculated
defined as yk = ½ x3 u1 x4 u2 T that is given by
from equation (26). Without loss of generality, suppose
yk = Cxk = the value of xdes
TM is used as the operating point of the lin-
2 3 earized plant (equation (33)). Therefore, the resonance
0 0 1 0 0 0 0 0 0 0 0
60 0 in the energy harvester is enhanced if the position of the
6 0 0 0 0 0 1 0 0 077x
40 0 ð46Þ tuned mass, that is, the fourth element in the state vec-
0 1 0 0 0 0 0 0 05 k
tor of the linearized plant, is regulated to zero. To regu-
0 0 0 0 0 0 0 0 1 0 0
late the position of the tuned mass, the following cost
function should be minimized for a control horizon of
3. MPC N samples

MPC refers to a class of control schemes, which finds kX

+N
the control input that optimizes a pre-defined cost func- J1 ðk Þ :¼ y2j, 3 ð48Þ
tion over a finite future time horizon. Often, MPC uses j=k +1

a receding strategy that means, at each time instance,

According to the definition of the augmented state
only the first element of the calculated optimal inputs is
vector (equation (41)), yj, 3 is the third element of the
used, and for the next step, the optimization horizon is
output vector that is the position of the tuned mass.
shifted toward future. The MPC scheme originated in
late 1970s (Richalet et al., 1978). It has been further
developed for the controlled autoregressive integrating 3.2. Piezoelectric patches
moving-average (CARIMA) input–output models and
The electric energy output of the piezoelectric patches is
state-space models in Clarke et al. (1987) and Li et al.
given by the time integral of the instantaneous electrical
(1989), respectively. For linear systems, the MPC opti-
power as
mization problem is convex. In this case, efficient
numerical algorithms exist to solve the MPC problem ð
(Camacho and Alba, 2013). This section formulates E =  V ðtÞiðtÞdt ð49Þ
and solves an MPC synthesis problem that aims to
maximize the electrical energy generation in the VIV In the discrete output vector, equation (46), the first
energy harvester. element (yk, 1 ) and the second element (yk, 2 ), respec-
In the proposed energy harvester, the control objec- tively, represent V (tk ) and i(tk ), for instance tk = kh.
tive of maximizing the electrical energy generation can The maximum power generation objective can be
be attained by two actuators: the tuned mass actuator achieved by minimizing the following cost function
144 Journal of Intelligent Material Systems and Structures 31(1)

kX
+N   In conclusion, the MPC design problem can be
J2 ðk Þ :¼ yj, 1 yj, 2 ð50Þ stated as
j=k +1

Essentially, to guarantee that the amount of energy min J ðDUÞ ð59Þ

DU
generated by the piezoelectric patches is positive, the
achieved minimum value of the cost function J2 (k) Subject to the constraint (equation (39)), since the
should be negative. cost function, J (DU), is quadratic and the constraints
are linear with respect to the optimization variable DU,
the optimization problem (equation (19)) is an LQP.
3.3. MPC synthesis Several efficient numerical methods are available for
Based on the cost functions J1 (k), and J2 (k), let us for- LQP, see, for example, interior-point method (Rao
mulate the MPC synthesis problem as et al., 1998) and active-set method (Ferreau et al.,
2008).
arg min J3 ðk Þ ð51Þ
Duk , ..., Duk + N 1

For a control horizon of N samples 4. Results and discussion

kX
+N   This section investigates the performance of the pro-
J3 ðk Þ :¼ l1 J1 ðk Þ + l2 J2 ðk Þ + l3 y2j, 2 + l4 y2j, 4 posed VIV energy harvester with the TMM through
j=k +1 numerical simulations. We aim to determine the power
ð52Þ
generation of the VIV energy harvester and to investi-
gate effectiveness of the proposed TMM in improving
The terms in the summation in equation (12) repre- the power generation in various wind speeds.
sent the power losses associated with the control action The physical properties of the VIV energy harvester
required by the piezoelectric and the tuned mass actua- are given in Table 1. The material properties and geo-
tor. Here, constants l1 , l2 , l3 , and l4 are positive metric parameters of the VIV energy harvester are cho-
weights that are used to adjust the relative importance sen to match with those of Azadi Yazdi (2018). The
of the four objectives. The minimization (equation parameters of the aerodynamic force expression (i.e.
(11)) is constrained to system dynamics (equation (6)) CL0 , F, G, and S) are borrowed from the study by Skop
and path constraints and Balasubramanian (1997). The values of the MPC
parameters are chosen by trial and error. The physical
yj, i ł ci , 8j = k + 1, . . . , k + N  1 and 8i = 1, . . . , 4 properties of the piezoelectric patches are taken from
ð53Þ the study by Dai et al. (2014b).

where ci is the bound on the ith discrete state.

To formulate the MPC optimization problem as a 4.1. Preliminary results with a proportional controller
linear quadratic problem (LQP), let us introduce pre-
diction vector notation for states, inputs, and outputs The proportional controller has been widely used in the
as literature of VIV energy harvesters in the form of con-
stant load resistor (Abdelkefi et al., 2012a; Akaydin
 T et al., 2012; Dai et al., 2014b). In our device, a constant
Xk = xTk + 1 , xTk + 2 , . . . , xTk + N ð54Þ
controller Kp is equivalent to a load resistor of
 T RL = 1=Kp .
DUk = DuTk , DuTk + 1 , . . . , DuTk + N1 ð55Þ
To investigate the efficiency of the VIV energy har-
 T
DVk = DvTk , DvTk + 1 , . . . , DvTk + N1 ð56Þ vester, let us define the efficiency as
 T ð ð
Yk = yTk , yTk + 1 , . . . , yTk + N 1 ð57Þ Pð t Þ V ðtÞiðtÞ
h= dt = dt ð60Þ
Pf Pf
where Xk 2 R(n + nu + nv )N 3 1 , DUk 2 Rnu N 3 1 , DVk 2
Rnv N 3 1 , and Yk 2 Rny N 3 1 . Using the prediction vector where Pf is the available aerodynamic power in the flow
notation, the MPC cost function (equation (12)) is passed through the harvester’s effective frontal area, Af ,
given by that is given by Dai et al. (2014b)
 
 T u DU + xT T T + DVT T T Q
J ðDUÞ = DUT T Tu Q  T u DU
k x v Pf = 0:5Af ra U03
 
ð58Þ D 1
= 0:5 2 Lm + Ym, max + L2m tan um, max ra U03
 T , T , and T 2 2
where Q, u x v are constant matrices
defined in Appendix 1. ð61Þ
Azadi Yazdi 145

Table 1. Properties of the VIV energy harvester and the MPC scheme.

Variable Description Value

Mast Lm Length 5m
D Outer diameter 0:6 m
t Thickness 0:002 m
rm Mass density 1040 kg=m3
mTM Tuned mass 10 kg
Stand Ls Length 1m
ws Width 0:017 m
ts Thickness 0:017 m
rs Mass density 2730 kg=m3
Es Young’s modulus 90 GPa
Piezoelectric Lp Length 0:015 m
wp Width 0:017 m
tp Thickness 0:0034 m
e33 Permittivity at constant strain 13:28 nCm1
d31 Strain coefficient 190 pm V1
P
C Capacitance 20:3 nF
rp Mass density 7800 kg=m3
Ep Young’s modulus 66 GPa
Aerodynamic CL0 Lift coefficient of stationary cylinder 0.28
ra Air density 1:22 kg=m3
F van der Pol’s empirical constants 0.42
G van der Pol’s empirical constants 0.028
S Strouhal number 0.21
MPC h Sampling time 0:02 s
T Optimization horizon 0:1 s
N Number of samples 5
Odes Desired non-dimensional frequency 1.2

VIV: vortex-induced vibration; MPC: model predictive control.

where Ym, max is the maximum displacement of the tip of

the stand, and um, max is the maximum slope of the tip of
the stand.
The average closed-loop power output and efficiency
of the VIV energy harvester equipped with propor-
tional controllers of Kp = 5 3 106 , 1 3 107 , 5 3
107 , 1 3 108 is shown in Figure 2. As shown in
Figure 2, and affirmed in Abdelkefi et al. (2012a),
Akaydin et al. (2012), and Dai et al. (2014b), VIV
energy harvesters equipped with proportional control-
lers experience the resonance only in a narrow band
around the lock-in wind speed of Ulockin = 0:97 m=s.
Outside this narrow band, the power output and the
efficiency of the VIV degrade sharply due to a mis-
match between the natural frequency and the vortex
shedding frequency. Previous research on VIV energy
harvesting devices (Abdelkefi et al., 2012b; Dai et al.,
2014b) reported similar degradation trends in the effi-
ciency and power outside the lock-in region.

4.2. MPC scheme without the TMM

Figure 2. The closed-loop power characteristics of the VIV
Before using the tuned mass to improve the perfor- energy harvester without tuned mass mechanism equipped with
mance of the VIV energy harvester, let us investigate proportional controllers of Kp = 53106 , 13107 , 53107 ,
the response of the energy harvester equipped with the 13108 .
146 Journal of Intelligent Material Systems and Structures 31(1)

The electrical power output of the turbine is shown in

Figure 4. It follows from Figure 4 that the VIV energy
harvester with the MPC scheme has a mean power out-
put of Pmean = 6:1 W in the steady state. Using equa-
tion (60), the mean efficiency is h = 7:11% in the wind
speed of 1.6 m/s. In comparison with the proportional
controller, at the same wind speed, the MPC scheme
improved the mean power output and the efficiency sig-
nificantly. The optimal control current through piezo-
electric patches, i(t), is shown in Figure 5.
The average power output and efficiency of the VIV
energy harvester equipped with the MPC scheme for
wind speeds 0:5 m=s\U0 \4 m=s are demonstrated in
Figure 3. Closed-loop responses of the transverse displacement Figure 6. In comparison with the best performance of
of the tip of the VIV energy harvester without the TMM equipped the proportional controllers in Figure 2, the peak value
with the MPC scheme at U0 = 1:6 m=s. of the power is improved by 304% by using the pro-
posed MPC method. The MPC scheme not only
improves the power output and efficiency in the lock-in
region (Ulockin = 0:97 m=s) but also, and more impor-
tantly, increases the power output in the post-lock-in
region drastically. The power output of the turbine
with the MPC scheme is greater than 3:19 W in the
post-lock-in region (1:2\U0 \4), where the conven-
tional proportional controllers deliver virtually no
power in this region, as shown in Figure 2.
Note that although the efficiency starts to decay at
U0 = 1:2 m=s in Figure 6, the electrical power output is
still considerable beyond this point, owing to the exces-
sive amount of energy stored in the higher winds.

4.3. MPC scheme with the TMM

Figure 4. The power output of the VIV energy harvester To show the effectiveness of the proposed TMM in
without the TMM equipped with the MPC scheme at U0 =
improving the performance of the VIV energy har-
1:6 m=s.
vester, the closed-loop performance of the turbine with
the tuned mass is investigated in this section. As
explained earlier, the key feature of the VIV energy
proposed MPC scheme. Since the tuned mass is not
harvester that enhances the power generation is the
considered in this section, the terms that are related to
lock-in phenomenon (i.e. vs = vn ). Using the definition
the regulation of the tuned mass (i.e. Q1 and Q4 ) in the
of the vortex shedding frequency, the lock-in wind
MPC cost function (equation (79)) are eliminated.
speed is given by
The parameters of the MPC scheme are given in
Table 1. The prediction horizon of T = 0:1 s is used to vn D
ensure the convergence of the MPC optimization prob- Ulockin = ð62Þ
2pS
lem within each sampling time of h = 0:02 s in real-time
implementation. The wind speed V is assumed to be In the absence of the tuned mass, the VIV energy
constant and known in the prediction time horizon. The harvester has a single lock-in wind speed of
‘‘quadprog’’ function in the MATLAB’s Optimization Ulockin = 0:97 m=s. However, if the VIV energy har-
Toolbox is used for solving the optimization problem vester is equipped with a tuned mass, the natural fre-
(equation (59)) that uses the interior-point algorithm. quency of the turbine can be tuned by changing the
The solver ‘‘quadprog’’ handles box constraints and lin- position of the tuned mass. Figure 7 shows the natural
ear equality and inequality constraints. frequency of the VIV energy harvester for various posi-
The transverse displacement of the tip of the VIV tions of the tuned mass. Consequently, according to
energy harvester, Y (Ls + Lm , t), equipped with the equation (62) the lock-in wind speed depends strongly
MPC scheme at the wind speed of U0 = 1:6 m=s is on the position of the tuned mass as shown in Figure 8.
shown in Figure 3. The maximum displacement of the To allow for a fair comparison between two types of
tip of the energy harvester is 0:23 m in the steady state. the VIV harvesters without and with the TMM, it is
Azadi Yazdi 147

Figure 5. The optimal control current through piezoelectric Figure 8. The lock-in wind speed of the VIV energy harvester
patches of the VIV energy harvester without the TMM equipped versus the position of the tuned mass for mTM = 1, 10, and
with the MPC scheme at U0 = 1:6 m=s. 50 kg.

case, the two types of the VIV harvesters would have

similar dynamic characteristics if their centers of mass
coincide. Consequently, the lock-in wind speed of the
harvester with the TMM matches the one of the har-
vester without the TMM (Ulockin = 0:97 m=s) when
the tuned mass is placed at the center of mass of the
cylinder (i.e. xTM = 2:5 m). This can be verified in
Figure 8 for the tuned mass of mTM = 10 kg.
Let us define the tuned mass ratio as the mass of the
tuned mass to the total mass of the cylindrical mast,
given by
mTM
MR = 100 3 ð63Þ
mm + mTM
Figure 6. The optimal power characteristics of the VIV energy
harvester without the TMM equipped with the MPC scheme. In the proposed VIV energy harvester, where the
cylindrical mast has a mass of 10 kg, tuned masses of
1, 10, and 50 kg, in Figures 7 and 8, correspond to
tuned mass ratios of 9%, 50%, and 83%, respectively.
According to Figures 7 and 8, as the mass ratio
increases, the position of the tuned mass has a stronger
effect on the natural frequency and the lock-in wind
speed of the turbine. However, large values of tuned
mass ratio may not be practically feasible, due to the
excessive load exerted by the tuned mass on the sup-
porting structures of the VIV harvester. Hereafter, we
use a tuned mass of 10 kg.
The transverse displacement of the tip of the VIV
energy harvester equipped with the TMM at the wind
speed of U0 = 1:6 m=s is shown in Figure 9. The
maximum displacement of the tip of the energy harvester
is 0:23 m in the steady state. The tuned mass is initially
placed at the mid-span of the mast that is
Figure 7. The natural frequency of the VIV energy harvester xTM (0) = 2:5 m. According to equation (47), to acquire a
versus the position of the tuned mass for mTM = 1, 10, and non-dimensional frequency of Odes = 1:2, the desired
50 kg. position of the tuned mass is at xdes
TM = 0 m. The closed-
loop response of the tuned mass is shown in Figure 10.
assumed that the total mass of the cylinder and the The tuned mass has been moved toward the desired
tuned mass (mTM = 10 kg) in the latter type are equal position owing to the term l1 Q1 in the MPC cost func-
to the mass of the cylinder in the former type. In this tion (79).
148 Journal of Intelligent Material Systems and Structures 31(1)

Figure 12. The optimal control current through piezoelectric

Figure 9. Closed-loop responses of the VIV energy harvester
patches of the VIV energy harvester with the TMM at U0 =
with the TMM equipped with the MPC scheme at U0 = 1:6 m=s.
1:6 m=s.

Figure 10. The tuned mass position at U0 = 1:6 m=s. Figure 13. The optimal control force of the tuned mass
actuator at U0 = 1:6 m=s.

than the mean output power of the turbine in the absence

of the TMM. Using equation (60), the mean efficiency is
h = 27:7%, that is, much larger than the efficiency of the
turbine without the TMM. In conclusion, the TMM
increases the power output and efficiency of the VIV
energy harvester significantly. The optimal control current
through piezoelectric patches, i(t), is shown in Figure 12.
Upon comparing with Figure 4, the control current has
larger peaks in the VIV energy harvester with the TMM.
Figure 13 shows the control force of the tuned mass actua-
tor. The peak value of the control force reaches the rea-
sonable value of 157 N that can be provided by many
types of linear actuators. Also note that the fluctuations in
Figure 11. The optimal power output of the VIV energy the steady state control force are due to inertial forces
harvester with the TMM at U0 = 1:6 m=s. induced by the vibrations of the harvester, which should
be overcome by the tuned mass actuator.
Let us investigate the performance of the proposed
The electrical power output of the turbine equipped TMM in other wind speeds. The average power output
with the TMM is shown in Figure 11. The VIV energy and efficiency of the VIV energy harvester equipped
harvester with the TMM captures a mean power output with the MPC scheme for wind speeds 0:5 m=s\
of Pmean = 16:3 W in the steady state, that is, 167% larger U0 \7 m=s are demonstrated in Figure 14. In
Azadi Yazdi 149

Table 2. The 70% effective wind speed band.

hmax Uhmax U70% B70%

Proportional 24.4 1.05 [1.01, 1.104] 0.094

MPC without TMM 25.1 1.2 [1.06, 1.36] 0.3
MPC with TMM 27.7 1.6 [1.05, 1.92] 0.87

MPC: model predictive control; TMM: tuned mass mechanism.

electric power output of the turbine, the TMM should

work in close cooperation with the piezoelectric genera-
tors. An MPC technique is formulated to find the opti-
mal control inputs for the tuned mass actuator and the
Figure 14. The optimal power characteristics of the turbine
piezoelectric generators. Results of the numerical simu-
with the TMM equipped with the MPC scheme. lations showed that the TMM increased the width of
the velocity band over which the turbine has a rela-
tively high efficiency by 190%. The proposed VIV
comparison with the VIV energy harvester without energy harvester with TMM was able to achieve a
the TMM (Figure 6), the decaying trend in the effi- power output of 26.7 W at the wind speed of 2 m=s
ciency is delayed to significantly higher wind speeds. that is 294% larger than the maximum power gener-
Consequently, the peak value of the power is increased ated without TMM.
by 294% using the proposed TMM compared with the
best performance of the VIV energy harvester without Declaration of conflicting interests
TMM. The power output of the VIV turbine with the The author(s) declared no potential conflicts of interest with
TMM in the post-lock-in is greater than 18.3 W that respect to the research, authorship, and/or publication of this
shows 474% improvement over the turbine without the article.
TMM.
Now, let us define a measure to investigate the width Funding
of the wind speed region over which the turbine has a The author(s) received no financial support for the research,
relatively high efficiency. Let hmax be the maximum effi- authorship, and/or publication of this article.
ciency achievable by the energy harvester, and Uhmax be
the corresponding wind speed. Define the 70% effective ORCID iD
wind speed set, U 70% , as the compact set of wind speeds
E Azadi Yazdi https://orcid.org/0000-0001-6615-8327
around Uhmax for which the turbine has an efficiency
greater than or equal to 0:7hmax . The effective wind
speed band is the width of the wind speed region U 70% , References
that is, defined as Abdelkefi A and Ghommem M (2013) Piezoelectric energy
  harvesting from morphing wing motions for micro air
B70% = max U 70%
 min U 70%
ð64Þ vehicles. Theoretical and Applied Mechanics Letters 3(5):
052004.
The effective wind speed band of the proposed VIV Abdelkefi A, Hajj MR and Nayfeh AH (2012a) Power har-
energy harvester is given in Table 2. The VIV energy vesting from transverse galloping of square cylinder. Non-
harvester with TMM has a higher effective wind speed linear Dynamics 70(2): 1355–1363.
band, B70% , compared with other two designs. Abdelkefi A, Hajj MR and Nayfeh AH (2013) Piezoelectric
energy harvesting from transverse galloping of bluff bod-
ies. Smart Materials and Structures 22(1): 015014.
Abdelkefi A, Nayfeh AH and Hajj MR (2012b) Design of
5. Conclusion piezoaeroelastic energy harvesters. Nonlinear Dynamics
In this article, the optimal control problem of a VIV 68(4): 519–530.
energy harvester with a TMM was investigated. The Abdelkefi A, Nayfeh AH and Hajj MR (2012c) Modeling and
analysis of piezoaeroelastic energy harvesters. Nonlinear
turbine consists of a relatively long rigid hollow cylin-
Dynamics 67(2): 925–939.
der mounted on a flexible beam that is partially cov-
Akaydin HD, Elvin N and Andreopoulos Y (2012) The per-
ered by piezoelectric generators. It was demonstrated formance of a self-excited fluidic energy harvester. Smart
through dynamic modeling of the turbine that the Materials and Structures 21(2): 025007.
TMM, which is placed inside the cylinder, can change Amin Karami M and Inman DJ (2012) Powering pacemakers
the natural frequency of the structure and consequently from heartbeat vibrations using linear and nonlinear
enhance the VIV of the cylinder. To maximize the energy harvesters. Applied Physics Letters 100(4): 042901.
150 Journal of Intelligent Material Systems and Structures 31(1)

Azadi Yazdi E (2018) Nonlinear model predictive control of energy harvester with different rod-shaped attachments.
a vortex-induced vibrations bladeless wind turbine. Smart Applied Energy 226: 682–689.
Materials and Structures 27: 075005. DOI: 10.1088/1361- Kluger JM, Moon FC and Rand RH (2013) Shape optimiza-
665X/aac0b6. tion of a blunt body Vibro-wind galloping oscillator. Jour-
Bibo A and Daqaq MF (2013) Investigation of concurrent nal of Fluids and Structures 40: 185–200.
energy harvesting from ambient vibrations and wind using Li S, Lim KY and Fisher DG (1989) A state space formula-
a single piezoelectric generator. Applied Physics Letters tion for model predictive control. AIChE Journal 35(2):
102(24): 243904. 241–249.
Bishop RED and Hassan AY (1964) The lift and drag forces Meliga P, Chomaz JM and Gallaire F (2011) Extracting
on a circular cylinder in a flowing fluid. Proceedings of the energy from a flow: an asymptotic approach using vortex-
Royal Society A: Mathematical, Physical and Engineering induced vibrations and feedback control. Journal of Fluids
Sciences 277(1368): 32–50. and Structures 27(5–6): 861–874.
Cajas J, Houzeaux G, Yáñez D, et al. (n.d.). SHAPE project Olinger D (1998) A low-order model for vortex shedding pat-
vortex bladeless: parallel multi-code coupling for fluid- terns behind vibrating flexible cables. Physics of Fluids
structure interaction in wind energy generation. Available 10(8): 1953–1961.
at: http://www.prace-ri.eu/IMG/pdf/WP216.pdf Pan D and Dai F (2018) Design and analysis of a broadband
Camacho EF and Alba CB (2013) Model Predictive Control. vibratory energy harvester using bi-stable piezoelectric
London: Springer Science & Business Media. composite laminate. Energy Conversion and Management
Chizfahm A, Yazdi EA and Eghtesad M (2018) Dynamic 169: 149–160.
modeling of vortex induced vibration wind turbines. Rao CV, Wright SJ and Rawlings JB (1998) Application of
Renewable Energy 121: 632–643. interior-point methods to model predictive control. Journal
Clarke DW, Mohtadi C and Tuffs PS (1987) Generalized pre- of Optimization Theory and Applications 99(3): 723–757.
dictive control—part I. The basic algorithm. Automatica Richalet J, Rault A, Testud JL, et al. (1978) Model predictive
23(2): 137–148. heuristic control: applications to industrial processes.
Dai HL, Abdelkefi A and Wang L (2014a) Piezoelectric Automatica 14(5): 413–428.
energy harvesting from concurrent vortex-induced vibra- Roundy S and Wright PK (2004) A piezoelectric vibration
tions and base excitations. Nonlinear Dynamics 77(3): based generator for wireless electronics. Smart Materials
967–981. and Structures 13(5): 1131–1142.
Dai HL, Abdelkefi A and Wang L (2014b) Theoretical model- Sirohi J and Mahadik R (2011) Piezoelectric wind energy har-
ing and nonlinear analysis of piezoelectric energy harvest- vester for low-power sensors. Journal of Intelligent Mate-
ing from vortex-induced vibrations. Journal of Intelligent rial Systems and Structures 22(18): 2215–2228.
Material Systems and Structures 25(14): 1861–1874. Skop RA and Balasubramanian S (1997) A new twist on an
Ding L, Zhang L, Wu C, et al. (2015) Flow induced motion old model for vortex-excited vibrations. Journal of Fluids
and energy harvesting of bluff bodies with different cross and Structures 11(4): 395–412.
sections. Energy Conversion and Management 91: 416–426. Vicente-Ludlam D, Barrero-Gil A and Velazquez A (2014)
Erturk A, Vieira WGR, De Marqui C, et al. (2010) On the Optimal electromagnetic energy extraction from transverse
energy harvesting potential of piezoaeroelastic systems. galloping. Journal of Fluids and Structures 51: 281–291.
Applied Physics Letters 96(18): 184103. Yang Y, Zhao L and Tang L (2013) Comparative study of tip
Ferreau HJ, Bock HG and Diehl M (2008) An online active cross-sections for efficient galloping energy harvesting.
set strategy to overcome the limitations of explicit MPC. Applied Physics Letters 102(6): 064105.
International Journal of Robust and Nonlinear Control Zhang LB, Dai HL, Abdelkefi A, et al. (2017) Improving the
18(8): 816–830. performance of aeroelastic energy harvesters by an inter-
Franklin GF, Powell JD and Workman ML (1998) Digital ference cylinder. Applied Physics Letters 111(7): 073904.
Control of Dynamic Systems. Reading, MA: Addison- Zhao L and Yang Y (2018) An impact-based broadband
Wesley. aeroelastic energy harvester for concurrent wind and base
Gabbai RD and Benaroya H (2005) An overview of modeling vibration energy harvesting. Applied Energy 212: 233–243.
and experiments of vortex-induced vibration of circular
cylinders. Journal of Sound and Vibration 282(3–5):
575–616. Appendix 1
Hasheminejad SM, Rabiee AH and Jarrahi M (2017) Semi-
active vortex induced vibration control of an elastic ellipti- Consider the following prediction vector notation for
cal cylinder with energy regeneration capability. Interna- states, inputs, and outputs as
tional Journal of Structural Stability and Dynamics 17(9):
 T
1750107. Xk = xTk + 1 , xTk + 2 , . . . , xTk + N ð65Þ
Hasheminejad SM, Rabiee AH, Jarrahi M, et al. (2014)
Active vortex-induced vibration control of a circular cylin-  T
DUk = DuTk , DuTk + 1 , . . . , DuTk + N1 ð66Þ
der at low Reynolds numbers using an adaptive fuzzy slid-
ing mode controller. Journal of Fluids and Structures 50:  T
49–65. DVk = DvTk , DvTk + 1 , . . . , DvTk + N1 ð67Þ
Hu G, Tse KT, Wei M, et al. (2018) Experimental investiga-  T
tion on the efficiency of circular cylinder-based wind Yk = yTk , yTk + 1 , . . . , yTk + N 1 ð68Þ
Azadi Yazdi 151

 
Based on the state-space model (equation (6)), the 1
Qi = diag Pi , Pi , . . . , Pi ð76Þ
values of the future state variables are calculated 2
sequentially using the set of previous and future control
and diagfa1 , . . . , an g is a diagonal matrix whose diago-
parameters as
nal entries are a1 , . . . , an , and P1 , P2 , P3 , P4 2 Rny 3 ny
xk + 1jk = Axk + B1 Duk + B2 Dvk are given by
xk + 2jk = Axk + 1jk + B1 Duk + 1 + B2 Dvk + 1 P1 = diagf0, 0, 1, 0g ð77Þ
2
= A xk + AB1 Duk + B1 Duk + 1 + AB2 Dvk + B2 Dvk + 1 2 3
0 1 0 0
.. 61 0 0 07
. P2 = 6
40 0 0 05
7 ð78Þ
ð69Þ 0 0 0 0
Using yk = Cxk and equation (24), the predicted vec- Consequently, the MPC cost function (12) is given
tors of output can be written as by

Yk = T x xk + T u DUk + T v DVk ð70Þ 1 T

J3 ðk Þ = Y ðl1 Q1 + l2 Q2 + l3 Q3 + l4 Q4 ÞYk ð79Þ
where 2 k
where Q3 , Q4 2 Rny N 3 ny N are block diagonal matrices
h iT
T 2 T N T defined in equation (31) with
T x = (CA) , (CA ) , . . . , (CA ) ð71Þ
2 3 P3 = diagf0, 1, 0, 0g ð80Þ
CB1 0  0
6 CAB1 CB1  0 7 P4 = diagf0, 0, 0, 1g ð81Þ
6 7
Tu = 6 .. .. .. .. 7 ð72Þ
4 . . . . 5 Substituting equation (25) in equation (34) and dis-
CAN 1 B1 CAN 2 B1    CB1 carding the terms that do not depend on the optimiza-
2 3 tion variable DU leads to the definition of a new
CB2 0  0
6 CAB2 CB2  0 7 objective function
6 7
Tv = 6 .. .. .. .. 7 ð73Þ  
4 . . . . 5  u DU + xT TT + DVT TT QT
J ðDUÞ = DUT TuT QT  u DU
k x v
N 1 N 2
CA B2 CA B2    CB2
ð82Þ
Using the aforementioned notation, the predicted where
values of all future outputs in the control horizon of
length N samples are stored in the variable Yk .  = l1 Q1 + l2 Q2 + l3 Q3 + l4 Q4
Q ð83Þ
Therefore, the cost functions J1 (k) and J2 (k) can be
written as functions of Yk The path constraint (equation (13)) can be written as

kX
+N
1 T  Cs  T x xk  T v DV ł T u DU ł Cs  T x xk  T v DV
J1 ðk Þ = y2j, 3 = Y Q1 Yk ð74Þ ð84Þ
j=k +1
2 k
where Cs 2 R7N is defined as

kX
+N  1
J2 ðk Þ = y, 3 yj, 2 = YTk Q2 Yk ð75Þ
j=k +1
2 Cs = ½cs, 1 , cs, 2 , cs, 3 , cs, 4 , cs, 5 , cs, 6 , cs, 7 , cs, 1 , . . . , cs, 7 T
ð85Þ
where Qi 2 Rny N 3 ny N is a block diagonal matrix defined
as