Applied Energy 185 (2017) 1281–1291

Contents lists available at ScienceDirect

Applied Energy
journal homepage: www.elsevier.com/locate/apenergy

Analysis of cavitation for the optimized design of hydrokinetic turbines

using BEM
Paulo Augusto Strobel Freitas Silva a, Léo Daiki Shinomiya b, Taygoara Felamingo de Oliveira a,
Jerson Rogério Pinheiro Vaz b, André Luiz Amarante Mesquita c,⇑, Antonio Cesar Pinho Brasil Junior a
a
University of Brasília – Faculty of Mechanical Engineering, Campus Darcy Ribeiro, Brasília, DF 70910-900, Brazil
b
Federal University of Pará – Faculty of Mechanical Engineering, Av. Augusto Correa, N 1 – Belém, PA 66075-900, Brazil
c
Federal University of Pará – Amazon Development Center in Engineering – NDAE – Vila Permanente 68.464-000 Tucuruí, PA, Brazil

h i g h l i g h t s

Optimization of horizontal-axis hydrokinetic turbines considering cavitation.

The method corrects the blade chord by a modification on the local thrust coefficient in order to prevent cavitation.

The minimum pressure coefficient is the criterion used for the identification of cavitation on hydrokinetic blades.

The method is evaluated by CFD using the Rayleigh-Plesset model to predict the vapor production rate.

The approach is helpful to design energy generation technologies applied to river, tidal and marine currents.

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

Article history: Hydrokinetic turbines are a promising technology for renewable energy production from river, tidal and
Received 25 June 2015 marine currents. This paper proposes an innovative approach applied to optimization of horizontal axis
Received in revised form 16 February 2016 hydrokinetic turbines (HAHTs) considering the possibility of cavitation. The minimum pressure coeffi-
Accepted 17 February 2016
cient is the criterion used for identifying cavitation on blades. Blade Element Momentum (BEM) theory
Available online 22 March 2016
is employed for the rotor design. During the optimization procedure, chord length at each blade section
is corrected by a modification on the local thrust coefficient in order to prevent cavitation. The hydraulic
Keywords:
parameters as lift, drag and minimum pressure coefficients are calculated by XFoil. Additionally,
Hydrokinetic turbines
Cavitation
Computational Fluid Dynamics (CFD) techniques are used to validate the proposed methodology.
Blade optimization Cavitation volume in the water flow through the rotor, with and without geometrical modifications, is
BEM evaluated using a Reynolds Averaged Navier–Stokes (RANS) approach coupled to the Rayleigh-Plesset
Rayleigh-Plesset model model to estimate the vapor production rate. The methodology is applied to the design of a 10 m diam-
eter Hydrokinetic Turbine (HT) rated to 250 kW output power, for a flow velocity of 2.5 m/s. The flow
around the optimized rotor presents a reduction of the vapor volume without a major variation upon
the turbine output power. A comparison with the Horizontal Axis Rotor Performance Optimization
(HARP_opt) code was carried out, demonstrating good behavior. CFD simulations revealed that the pro-
posed design method minimizes cavitation inception, yielding a useful tool for efficient HT design at
rated conditions.
Ó 2016 Elsevier Ltd. All rights reserved.

1. Introduction The power coefficient maximization is fundamental in HT design

Hydrokinetic turbines (HTs) have recently been used as con-
verters of river, tidal and marine currents to electrical energy
[1,2]. This technology has become significant due to the increasing
use of renewable energy sources with low environmental impact.
⇑ Corresponding author.
E-mail address: andream@ufpa.br (A.L. Amarante Mesquita).
in order to improve energy extraction from water flow. However,
for design, it is important to take into account the possibility of
cavitation. As in the design of classical hydraulic turbines [3], cav-
itation inception is assumed to occur on a blade section when the
minimum local pressure falls below the vapor pressure of the fluid.
Cavitation inception can be predicted by comparing the local min-
imum pressure coefficient with the cavitation number. The
chances of cavitation occurring increases more toward the blade

http://dx.doi.org/10.1016/j.apenergy.2016.02.098
0306-2619/Ó 2016 Elsevier Ltd. All rights reserved.
1282 Paulo Augusto Strobel Freitas Silva et al. / Applied Energy 185 (2017) 1281–1291

Nomenclature
Ns section number
Abbreviations NB number of bubbles per unit of mixture volume
BEM Blade Element Momentum patm , atmospheric pressure (Pa)
CFD Computational Fluid Dynamics p0 ; p1 reference pressure upstream(Pa)
DWST Distance from water surface to tip blade pv vapor pressure (Pa)
HAHT Horizontal Axis Hydrokinetic Turbine r; R radial position and turbine radius (m)
HT Hydrokinetic Turbine rl ; rv volume fraction of liquid and vapor
HARP_opt Horizontal Axis Rotor Performance Optimization r nuc volume fraction of nucleation sites
LES Large Eddy Simulation RB bubble radius (m)
RANS Reynolds Averaged Navier–Stokes Re Reynolds number
SST Shear Stress Model Rec local Reynolds number based on the airfoil chord
ui velocity component (m/s)
u0i velocity fluctuations (m/s)
Arabic symbols
a; a0 axial and tangential induction factor uþ friction velocity (m/s)
aopt optimum axial induction factor u0i u0j Reynolds stress tensor (m2/s2)
V0 velocity in ambient free stream (m/s)
B number of blades
c chord (m) V CAV cavitation velocity
cuc ; cco uncorrected and corrected chord (m) VB bubble volume (m3)
copt optimal chord (m) W relative velocity (m/s)
CL; CD lift and drag coefficients
Dy distance of nearest wall (m)
CP pressure coefficient yþ dimensionless wall distance
C power power coefficient
C Pmin minimum pressure coefficient Greek symbols
Cn normal force coefficients b twist angle (deg)
CT thrust coefficient bopt twist angle (deg)
D turbine diameter (m) k tip-speed ratio
fs safety factor l dynamic viscosity (St)
f force per unit of volume (N/m3) m kinematic viscosity (St)
F Prandtl’s tip loss factor q density (kg/m3)
Fc empirical constant of the cavitation model qm mixture density (kg/m3)
g gravity acceleration (m2/s) r cavitation number
h; H distance between free surface and turbine radial posi- rs surface tension coefficient
tion (m) sij Reynolds stress tensor (m2/s2)
m_ l ; m_ v rate of change mass per unit of volume for liquid and X angular velocity of the rotor (rad/s)
vapor phases

tip due to low immersion depth, which is an important issue to be
assessed on HT design. Molland et al. [4] showed that the
cavitation-free region changes with respect to the hydrofoil
camber. This observation was confirmed by Batten et al. [5] who
show that the cavitation characteristics for a particular section
can be described by a minimum pressure envelope or cavitation-
free region, as a function of the section cavitation number. Since
the section lift coefficient is a function of the pressure distribution,
this region can be represented as a limiting lift coefficient envelope
for a given cavitation number. In a complementary work, Bahaj
et al. [6] carried out an experimental study of cavitation inception
on HT blades and concluded that the cavitation does not appear
until the cavitation number has been reduced to about 0.9. When
the cavitation number is reduced to below 0.4, there is cavitation
on the suction side over 10–15% of the outer part of the blade.
These authors suggest that cavitation typically appears at
tip-sped ratio (k ¼ XR=V 0 ) values greater than about 7.0, which
indicates a very high thrust loading, as discussed by Goundar
et al. [7].
In general, the optimization models of HT blades are based on
BEM theory. These methods are frequently used for design and
analysis of hydrokinetic rotors. BEM is an integral method, with
semi-empirical information from hydrodynamic forces on hydro-
foil sections, combined with two-dimensional airfoil flow models
or experimental data for section lift and drag [8,9]. Such models
are direct applications of the approaches developed for wind tur-
bines, and are employed for design including optimization models
for chord and twist angle distributions, taking into account the
influence of swirl velocity in the wake [10]. BEM is also applied
successfully in the design of hydrokinetic turbines [5,7,11]. Sale
et al. [12] show a method of optimizing hydrokinetic blades based
on genetic algorithms coupled to the BEM, in which the cavitation
effect was employed. CFD is also used for the analysis of 3-D flow
through HTs as described in Refs. [13,14]. CFD is generally used to
analyze turbine performance and flow field, e.g., Kang et al. [15]
developed a detailed study using large-eddy simulation (LES) in
order to investigate the structure of turbulence in the wake mean-
dering of an axial turbine. Lee et al. [16] used the CFD approach to
obtain a new blade design to delay cavitation inception. The mod-
ified design is a raked tip turbine, which does not degrade the tur-
bine performance.
This paper proposes a new methodology with low computa-
tional cost and easy implementation, using BEM to optimize
hydrokinetic rotors considering the minimum pressure coefficient
as criterion to avoid cavitation. A correction on the local thrust
coefficient is performed, in order to calculate the optimum shape
of the hydrokinetic blade. In the computational algorithm, lift
and drag coefficients are calculated by XFoil [17], which is a linear
vorticity-stream function panel method with an integral boundary
layer and wake model. Both corrected and uncorrected rotor
designs are then analyzed by CFD through the Reynolds Averaged
Navier Stokes (RANS) equations using the Rayleigh-Plesset model
to estimate the vapor production rate. This step is computed by
commercial code ANSYS CFX.
 Paulo Augusto Strobel Freitas Silva et al. / Applied Energy 185 (2017) 1281–1291
The methodology is applied for the design of a three blade HT
with NACA 653  618 hydrofoil, 10 m diameter, D, with a rated
power of 250 kW for a velocity of 2.5 m/s (C power  0:407). As a
result, the blades are corrected from 70% of the length up to the
tip. The design presents good performance without major varia-
tions on the turbine output power. A comparison with HARP_opt
code was also carried out, demonstrating good behavior. The CFD
calculation compared the vapor production rate for both corrected
and uncorrected blades, verifying the modified geometry as a good
design for cavitation delay. Even though the proposed methodol-
ogy needs to be validated by experimental data which will be car-
ried out in the future, it appears to be an efficient method of
hydrokinetic turbine design in a free-cavitation operation regime,
and highly useful in real applications.
2. Cavitation prediction
Cavitation should be considered in the design of HTs [7], as it
causes structural damage to the blades and reduces performance.
During cavitation, liquid vaporizes instantly forming a vapor bub-
ble, which alters the flow. The shape and size of bubbles vary as a
consequence of this action in the pressure and velocity fields.
When the vapor cavity implodes, the pressure on the blade surface
increases, promoting erosion on the blade. The damage decreases
lift and increases drag, leading to a reduction in turbine efficiency.
Cavitation occurrence can be predicted by comparing the local
pressure distribution with the cavitation number [12], which is
classically defined as
patm þ qgh  pv
r¼ ; ð1Þ
1
2
qW 2
where patm is the atmospheric pressure, g is the local gravity, h is the
distance between free surface and the radial position on the
hydrokinetic rotor, pv is the vapor pressure, and
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi by using the chord length before cavitation correction taking
W ¼ ½V 0 ð1  aÞ2 þ ½Xr ð1 þ a0 Þ2 is the relative velocity. Cavitation
will occur if the local minimum pressure coefficient, C Pmin ¼ minðC P Þ
(see Fig. 1), is lower than r. C Pmin is an important parameter for HT
design. It gives information on hydrodynamic loading of blades.
This coefficient can be used as a criterion to avoid cavitation, which
is given by
r þ C Pmin P 0:
The pressure coefficient is defined as
-1
C Pmin = -1.04 Suction side
-0.5
0
CP
NACA 653 - 618
0.5
1
-1 -0.5 0 0.5 1 1.5
x/c
Fig. 1. Typical minimum pressure coefficient around a rotor blade section near the
tip.
1283
p  p0
CP ¼ 1 ; ð3Þ
2
qW 2
where p is the local pressure on the hydrofoil profile at the radial
blade section, and p0 is a reference pressure upstream of the rotor,
given by patm þ qgh.
The hydraulic pressure varies along the rotor blade and cavita-
tion is more likely near the tip of the top blade where static pres-
sure is lower. Fig. 2 illustrates the static pressure condition on an
elementary blade section. Consequently, the cavitation velocity
V CAV can be obtained combining Eqs. (2) and (3).
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
patm þ qgh  pv
V CAV ¼ : ð4Þ
 12 qC Pmin
2.1. Optimum hydrodynamic design considering cavitation
This approach corresponds to a simplified model designed to
avoid cavitation on each blade section, employed together with
the classical Glauert’s optimization [18]. In this case, the criterion
to avoid cavitation requires that the relative velocity, W, at each
blade section along the radial coordinate must be smaller than
V CAV . In other words, if W P V CAV then the local relative velocity
must be corrected. In the context of the BEM method, correction
is done by replacing W by W CAV ¼ ð1  f S ÞV CAV , where f S is an arbi-
trary safety factor defined in the interval 0 6 f S < 1 over the blade
element for the calculation of the chord length. Replacing W by
W CAV in sections where W P V CAV leads to an increase in the chord,
altering the pressure condition on the blade. The classical BEM
 2
method defines the thrust coefficient as C T ¼ VW0 2Bc pr C n , which
can be used to provide an expression for chord length c in terms
of ðV 0 =WÞ2 and V CAV . For blade sections where the criterion is
achieved, it is then easy to find a ratio between the corrected
and uncorrected chord lengths. In order to demonstrate this, start
c ¼ cuc , which is defined by
 2
2pr C T V 0
cuc ¼ : ð5Þ
B Cn W
ð2Þ
2
Fig. 2. Illustration of the static pressure condition on a rotor blade section.
1284 Paulo Augusto Strobel Freitas Silva et al. / Applied Energy 185 (2017) 1281–1291

To obtain the corrected formulation for the chord, it is necessary optimization procedure described in Section 2.1. The approach
to replace W by W CAV , yielding assumes the flow as being statistically stationary. In this sense,
 2 the simulations need to provide averaged fields around the rotor
2pr C T V0 blades. In order to consider the turbulent phenomena without
cco ¼ ; ð6Þ
B C n ð1  f S ÞV CAV numerically solving all the eddy scales, the Reynolds Average
where C T is given by simple axial momentum theory as Navier–Stokes (RANS) approach is adopted. This technique has
been historically used to simulate HTs [23] and currently RANS
C T ¼ 4að1  aÞF: ð7Þ evolved to a standard in the accurate prediction free flow turbines
performance [24]. In this methodology, the contribution of the tur-
where F is the Prandtl’s approximation to the tip-loss factor [19].
bulent velocity fluctuations u0i to the temporal average velocity and
The indices uc and co refer to the uncorrected and corrected chord,
respectively. Dividing Eq. (6) by (5) yields the ratio between cco and pressure fields are given by the Reynolds Stress Tensor sij ¼ u0i u0j ,
cuc at each blade section, given by which must be modeled [25]. In the present work, the so-called
 2 j  x Shear-Stress Transport (SST) model is employed [26].
W In this study, the simulations of cavitating flows are performed
cco ¼ cuc : ð8Þ
ð1  f S ÞV CAV considering a two-phase continuum mixture of liquid water and
vapor. In this case, the phase continuity equations demand that
Note in Eq. (8), under cavitating condition (W P V CAV ) the term
h i2
@ra qa @r a qa ui
¼ m_ a ;
W
ð1f S ÞV CAV
will be always larger than 1 increasing the chord cco . þ ð10Þ
@t @xi
This equation corrects the chord distribution on the rotor sections
subject to cavitation. Eq. (8) can be used in any optimization where ui are the components of the mean velocity field, q is the
model, and here is applied to the classical Glauert’s optimization density and m _ a represents the rate of change of mass for each phase,
[18]. The Glauert’s optimization is designed to ensure that the per unit of mixture volume. These source terms account for mass
effective angle of attack has an optimum value along the span. exchange between the vapor and liquid phases. The subscript
The optimum angle of attack is assumed to be the one at maximum a ¼ ðl; v Þ discriminates the liquid and vapor phases and ra is the vol-
C L =C D . The optimum relationship between the local-speed ratio, ume fraction of each one. Taking into account that r l þ r v ¼ 1 and
x ¼ Xr=V 0 , and the axial induction factor, a ¼ aopt , is obtained by assuming that there are no thermal barriers, the species conserva-
maximizing the power coefficient. A detailed description is tion principle requires that
reported in Reference [19]. The optimum axial induction factor
m_ v ¼ m_ l : ð11Þ
aopt , is

Assuming that both phases have the same velocity, the mean
16a3opt  24a2opt þ aopt 9  3x2  1 þ x2 ¼ 0: ð9Þ momentum conservation equation is modified only by the substi-
It is noteworthy that Eq. (9) is assumed independent of Prandtl’s tution of the fluid density q by the mixture density
approximation F. This limitation is employed because F is variable qm ¼ ql rl þ qv rv , in such a way that
with the axial induction factor a, making the optimization analysis @uj @uj @p @  
rather complex. Studies on this were previously analyzed by Bur- qm þ qm ui ¼ þ 2lSij  qm u0i u0j þ qm f ; ð12Þ
@t @xi @xj @xi
ton et al. [20] and further detailed by Clifton-Smith [21]. Thus, in
the proposed optimization procedure, F is taken into account only where p is the mean mechanical pressure, l is the dynamic viscos-
in Eq. (7). ity, f is a force per unit of volume which may represent the gravita-
Having derived all the necessary equations to compute a given tional, Coriolis and centrifugal contributions and Sij are the
hydrokinetic blade geometry, it is possible to calculate the chord components of the symmetric part of the velocity gradient tensor.
and twist angle distributions considering the possible occurrence As previously mentioned, the Reynolds Stress Tensor u0i u0j is mod-
of cavitation. The rotor optimization can be expressed as a function eled by the j  x SST.
of the induction coefficients. The procedure to calculate the opti- Eq. (11) allows the two-phase calculations to be done by solving
mal chord copt and twist angle bopt for each radial section along only one differential continuity equation in addition to the
the blade, starting at its most inner section, is described in Appen- momentum equation, written in the form of Eq. (12). To proceed
dix A. in this manner, an equation for the mass source term m_ v is
required, which may be derived from the so called Rayleigh-
2.2. Numerical methodology Plesset model [22] for a spherical vapor bubble
 2
2rs
2
d RB 3 dRB p p
A numerical modeling using CFD (ANSYS CFX) is used in this RB þ þ ¼ v ; ð13Þ
study to simulate the flow throughout the hydrokinetic rotor. dt
2 2 dt q l RB ql
The main objective is to validate the optimization procedure in
where RB is the bubble radius, rs is the surface tension coefficient
free-cavitation condition as described previously. The CFD proce-
and pv is the pressure in the bubble, assumed to be the vapor pres-
dure is a good tool for this purpose, since it is able to predict accu-
sure at the liquid temperature. Assuming that only large bubbles
rately the flow around the blade, allowing to assess the cavitation
are important in cavitation, the second order terms and the surface
constraint introduced into the turbine thrust coefficient. Thus, the
tension are very small and can be neglected when compared with
cavitating flow using the Rayleigh-Plesset model [22], as well as all
the pressure difference [27]. After these simplifications, Eq. (13)
computational conditions are described in the next subsections.
reduces to
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2.2.1. Cavitating flow simulation dRB 2 pv  p
¼ : ð14Þ
The flow through the turbine rotor is assumed to be incom- dt 3 ql
pressible and fully turbulent. Consequently, velocity and pressure
fields are governed by the Navier–Stokes equations. CFD simula- Defining the bubble density number, N B , as being the number of
tions carried out in this work were conceived to provide quantita- bubbles per unit of volume of the mixture, the rate of change of
tive support to the cavitation prevention in the proposed vapor mass per unit of volume can be calculated by
 Paulo Augusto Strobel Freitas Silva et al. / Applied Energy 185 (2017) 1281–1291 1285

Table 1 Table 2
Material constants and general conditions for the Rayleigh-Plesset model at 25 °C. Design parameters used in the simulation of the HAHT.

Quantity Value Parameters Values

Fc 0:01 (condensation) and 50 (vaporization) Turbine diameter (D) 10:0 m
ql 997 kg/m3 Hub diameter 1:5 m
qv 0:02308 kg/m3 Number of blades 3
r nuc [32] 5  104 Current velocity ðV 0 Þ 2:5 m/s
Water density ðqÞ at 25 °C 997 kg/m3
Mean bubble diameter (RB ) [32] 1  106 m
Submergence of the turbine (H) 6m
Pressure of reference (p1 ) 162 kPa
Pressure of vapor (pv ) 3.170 kPa
P atm 1  105 Pa
Pv 3:17  103 Pa
Gravity (g) 9:81 m/s2
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Safety factor 5%
2 pv  p Angular velocity 35 rpm
_ v ¼ NB qv 4pR2B
m : ð15Þ
3 ql

Additionally r v ¼ N B 4pR3B =3, in such a way we can rewrite Eq. (15)
in terms of the fraction of vapor in the form
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
3V v qv 2 pv  p
_v ¼
m :
RB 3 ql
Eq. (16) relates the mass exchanging rate between phases to the
square root of the difference between the vapor pressure inside the
bubble and local mechanical pressure. In this form, the cavitation
model is restricted only to vaporization, i.e. m_ v > 0, because
pv  p must be positive. In other words, Eq. (16) cannot handle a
radius decrease (dRB =dt < 0) due to condensation, which takes
place inside the bubble when p > pv . In addition, the original
Rayleigh-Plesset model does not consider nucleation phenomena.
Vaporization begins at nucleation sites and, as the vapor volume
fraction increases, the nucleation site density decreases. Thus, only
for vaporization, rv is replaced by r nuc ð1  r v Þ, where rnuc is the vol-
ume fraction of the nucleation sites [28]. Barkir et al. [29] propose a
heuristic modification that, together with nucleation accounting,
leads to Eq. (17), where F c is an empirical constant used to discrim-
inate condensation and vaporization. In that work, Barkir et al. per-
formed a numerical calibration using experimental data [30,31]
and found that F c ¼ 50 for vaporization and F C ¼ 0:01 for conden-
sation accurately reproduces leading edge and mid-chord hydrofoil
cavitation. The values of the parameters of Eq. (17) found by Bakir
et al. [29] are given in Table 1.
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
3r nuc ð1  rv Þqv 2 jpv  pj
m_ v ¼ F c signðpv  pÞ:
RB 3 ql
2.2.2. Numerical setup
The HT geometric model was constructed based on the shape
data information given by the algorithm described in Section A
(Appendix B, Tables 6 and 7) for the corrected and uncorrected
ð16Þ
blades. In order to evaluate the effectiveness of this algorithm to
generate a non-cavitating blade, two rotor geometries were con-
structed: one with the shape corrected to avoid cavitation, and
another without this correction. The computational domain is
31 m  50 m 150 m as illustrated in Fig. 3. The rotor is positioned
at 2.5 D from the inlet boundary and 12.5 D from the outlet, in order
to avoid any spurious influence of the inlet and outlet sections. The
HT axis is placed 6 m from the river surface (H ¼ 6 m). In this man-
ner, a proximity of 1 m between the blade tip and river surface is
set, which provides a favorable environment to cavitation occur-
rence because of the low hydrostatic pressure in this region.
Most published literature assumes an axisymmetric flow
around the turbine and only an angular periodic section needs to
be simulated in steady flow approaches [33,34]. However, in this
case the entire domain must be considered because the flow is
non axisymmetric due to the proximity between the free surface
and the rotor, which is relevant for cavitation. Therefore, the com-
putational domain is divided in two different subdomains: a rotat-
ing cylindrical subdomain around the rotor, and a stationary flow
region. The cylindrical subdomain is constructed with a 11 m
diameter and 2 m length. Targeting to avoid a high computational
demand with sliding meshes, the interaction between the rota-
tional and stationary subdomains is made by imposing a rotating
ð17Þ reference frame only in the flow inside the cylindrical subdomain.
With this technique, the rotor is kept in a fixed position and the

Fig. 3. General setup of the computational domain.

1286 Paulo Augusto Strobel Freitas Silva et al. / Applied Energy 185 (2017) 1281–1291

Table 3 Table 2, is considered. For simulations carried out with classical

Hydrokinetic rotor geometry and hydrodynamic characteristics on each blade section. and modified Glauert’s optimization, the hydrodynamic parame-
Nsa r (m) c (m) b ( ) Rec  106 C Pmin r C Pmin þ r ters such as lift, drag, and minimum pressure coefficients are
obtained using the free software XFoil, a coupled panel/viscous
1 0.9737 0.5637 20.34 2.9230 1.140 14.5033 13.3633
2 1.4211 0.5273 14.09 3.3957 1.142 8.0703 6.9283 code developed at MIT [17]. In Table 2 (see also Fig. 2), H is the sub-
3 1.8684 0.4273 10.37 3.4719 1.142 4.9702 3.8282 mergence of the turbine. For engineering purposes, it is important
4 2.3158 0.3552 7.94 3.5081 1.143 3.3000 2.1570 to consider a safety margin in the calculation procedures. There-
5 2.7632 0.3028 6.24 3.5287 1.143 2.3154 1.1724
fore, the safety factor f S ¼ 5% is considered in this methodology
6 3.2105 0.2632 4.99 3.5381 1.143 1.6932 0.5502
7 3.6579 0.2583 4.04 3.5285 1.144 1.0747 0.0693
to assure that cavitation is prevented. In a real application, the
8 4.1053 0.2535 3.29 3.4527 1.144 0.9891 0.1549 parameters shown in Table 2 are input data for the algorithm
9 4.5526 0.2620 2.68 3.0972 1.145 0.7809 0.3641 depicted in Appendix A, through which the rotor blade can be
10 5.0000 0.1067 2.18 1.1105 1.146 0.6264 0.5196 designed for a free-cavitation condition.
a
Ns is the section number, usually called blade elements.

governing equations are solved considering a modified gravita-
tional acceleration, taking into account the Coriolis and centrifugal
components.
A uniform velocity field of 2.5 m/s and 5% turbulence intensity
is applied as a Dirichlet boundary condition at the inlet section. For
the outlet region, a zero pressure gradient condition is applied,
leaving the velocity field to be determined to satisfy the momen-
tum equation. The non-slip condition was applied at the blade sur-
faces, while the free slip condition is applied at the top, bottom and
lateral faces of the stationary domain. The rotational subdomain
has an angular velocity of 35 rpm.
3. Results and discussion
3.1. Design information and operational conditions
3.2. Correction method results
The results are obtained using the information from Table 2 and
the hydrokinetic rotor geometry presented in Table 3, which shows
the local Reynolds number based on the chord length and relative
velocity (Rec ¼ Wc=m, where m is the kinematic viscosity), the min-
imum pressure coefficient, and cavitation number or Thoma num-
ber as function of the radial position. Note that
Cpmin þ r ¼ 0:0693 for Ns ¼ 7 and greater, indicating that the cri-
terion Cpmin þ r P 0 is not satisfied from r P 3:66 m. Fig. 4a shows
the corrected and uncorrected chord distributions along the blade.
Cavitation occurs at approximately 70% of the span for the rotor
when operating at 35 rpm. The correction on the blade chord is a
consequence of the modification on the relative velocity, as shown
in Fig. 4b. This occurs due to W becoming higher than V CAV for
h i2
radial positions r > 3:5 m, where the term ð1f WÞV CAV is greater
S

To evaluate the proposed methodology, an HAHT design using than 1. In order to avoid cavitation, the proposed algorithm cor-
the NACA 653  618 hydrofoil, with main parameters given by rects W and assumes values always lower than V CAV .

Fig. 4. (a) Chord distribution along the blade. (b) Relative and cavitation velocities as functions of the radial position.

Fig. 5. (a) Power coefficient and (b) the output power as a function of undisturbed free-stream velocity.
 Paulo Augusto Strobel Freitas Silva et al. / Applied Energy 185 (2017) 1281–1291
Table 4
Processing time results.
DWST (m) Methodology Processing time (in min)
12 HARP_opt 35
11 HARP_opt 86
10 HARP_opt 139
1, 5, 10, 11 and 12 Present work <2
For the uncorrected rotor, the power coefficient C power , is slightly
higher than the corrected case for V 0 > 2:5 m/s (Fig. 5a). However,
C power is higher for the corrected rotor for V 0 < 2:5 m/s. The curve of
efficiency is changed by the increase of the chord required to avoid
cavitation. This effect is expected since the rotor geometry is mod-
ified. Fig. 5b shows the output power as a function of the free-
stream velocity. In this case, for the uncorrected rotor, P modifies
slightly higher when compared with the corrected rotor. This is
due to the change with the term V 30 , once P ¼ 12 qAV 30 C power , and
for low water velocities the modification in P is equally low. These
results show that the proposed methodology potentially prevents
cavitation without leading to significant reductions in turbine out-
put power.
To perform a comparison with another method available in the
literature, HARP_opt code is used. As described by Sale [35], HAR-
P_opt utilizes a multiple objective genetic algorithm associated to
BEM theory applied to design horizontal axis wind and hydroki-
netic turbine rotors. A cavitation constraint is introduced into
BEM to ensure that the calculated outputs are valid. An equivalent
procedure using the minimum local pressure coefficient criterion is
made within HARP_opt. However, the rotor blade optimization
occurs using a genetic algorithm, which obviously presents very
slow convergence when compared with traditional gradient-
based optimization approaches, such as Newton’s method [12]
which is limited and can fail for global optimal solutions. Hence,
to get comparative results, an HT with the same design parameters
as shown in Table 2 is simulated. The adopted genetic algorithm
configurations are: 30 generations and 15 individuals for each sim-
ulation. Table 4 shows the results considering the processing time
using a modern PC (core-i7 and 16 GB – RAM). Note that when the
distance from water surface to tip blade (DWST) decreases, the
processing time heavily increases up to 2 h and 19 min, demon-
strating the higher computational cost spent by the genetic algo-
rithm. A similar processing time was also reported by Sale et al.
[12] for a turbine with 3 blades, 5 m diameter and 35 kW rated
power (about 2 h). In the simulation it is not possible to obtain
results for DWST less than 10 m, due to it being very difficult for
HARP_opt to find feasible individuals per generation. On the other
hand, the present methodology is very fast because the cavitation
Fig. 6. (a) Chord distribution along the blade. (b) Output power as a function of the undisturbed free-stream velocity.
1287
Fig. 7. View of the turbine with streamlines.
constraint is imposed on the rotor load coefficient through
Glauert’s optimization [18], allowing chord correction in a simple
iteractive form. The present model can quickly solve problems
for any DWST in a processing time of less than 2 min.
An important concern is that HARP_opt code seems to have dif-
ficulty in simulating rotors more susceptible to cavitation. In other
words, when the rotor is closer to the water surface, HARP_opt
demands a huge processing time, mainly to find feasible individu-
als. This occurs only when the cavitation analysis is activated into
the code. For simulation without cavitation, HARP_opt is very fast
(about 2 min). Another advantage is that HARP_opt is able to search
effectively for solution space and convergence on a globally optimal
solution. This is a limitation for the present methodology. Fig. 6
shows the cavitation effect on chord distribution and output power
only for HARP_opt code. Using the present model, the cavitation
occurs for DWST up to 9 m, where HARP_opt was not able to solve.
Thus, for values higher than 9 m, the chord is not corrected, and for
this reason, only results for the uncorrected form are shown in Fig. 6
for the present work. Note that at 10 m, the cavitation occurs for
HARP_opt, and the chord is strongly increased along the entire
blade under the same operating condition (Fig. 6a). Comparing only
uncorrected forms, the present optimization has better output
power for V 0 > 2:5 m/s, and practically the same output power
when compared with the corrected rotor by HARP_opt (Fig. 6b).
3.3. Validation using CFD
Cavitation inception takes places when the nuclei of bubbles are
exposed to a pressure lower than vapor pressure. The lower peak at
1288 Paulo Augusto Strobel Freitas Silva et al. / Applied Energy 185 (2017) 1281–1291

Table 5 entire boundary layer region on the turbine blades, the near wall
Mesh refinement study. zone is represented with a spacing ratio of 1.2 in the normal direc-
Mesh Nodes ½106  yþ
max Power (kW) maxfrv g tion using 35 grid element layers. To get a detailed solution of the
boundary layer using SST turbulence model, the prism layers must
Mesh A 4.3 1.54 291 0.32
Mesh B 7.5 1.07 255 0.84 contain the entire boundary layer. It is well known that a common
Mesh C 8.5 1.20 251 0.81 parameter to identify the subparts of the boundary layer is the
value of yþ [36],

the trailing edge of the blade and viscous effects of the boundary Dyuþ
layer make a favorable environment for cavitation [22]. This region yþ ¼ ð18Þ
m
has high normal velocity gradients, which require a greater refine-
ment of the computational mesh. On the other hand, the stream- where uþ is the wall shear velocity. In this manner, the near-wall
wise gradients of velocity are low. To accurately capture the region can be roughly subdivided into three layers: viscous layer

(a) (b) (c)

Fig. 8. Cavitation in the rotors at 35 rpm. (a) corrected (b) uncorrected (c) geometrical comparison.

Fig. 9. Local comparison of cavitation occurrence obtained by CFD, XFoil and BEM for the uncorrected geometry.
 Paulo Augusto Strobel Freitas Silva et al. / Applied Energy 185 (2017) 1281–1291 1289

(yþ < 5), buffer layer (5 < yþ < 30) and the fully turbulent layer Table 6
(yþ > 30) [36]. Consequently, to solve the viscous sublayer, the val- Blade chord and twist distributions for foil NACA 653  618 for rotor with blade
corrected.
ues of yþ generally must be less than 5. Fig. 7 shows a view of the
turbine used in the simulations performed in this work. Radial distance, Span station Chord length Twist Twist axis
Numerical solutions obtained by CFD techniques can be differ- r (m) (r/R) (m) (°) (% chord)

ent depending on the discretization resolution of the computa- 0.743 0.148 0.27 24 30
tional domain. Therefore, three different meshes were developed 0.949 0.189 0.56 20 30
1.185 0.237 0.55 16 30
for the corrected blade with different configuration of local refine-
1.400 0.280 0.51 13 30
ment as shown in Table 5. Beyond the number of nodes in the 1.635 0.327 0.47 11 30
mesh, it is important to identify the regions where the refinement 1.860 0.372 0.43 10 30
will be applied. Two regions where mesh refinement is more 2.086 0.417 0.39 8 30
2.311 0.462 0.35 7 30
important in rotor simulations is the boundary layer and the
2.536 0.507 0.33 6 30
near-wake zone. Mesh A aimed to obtain a good resolution in the 2.761 0.552 0.3 5 30
boundary layer. The thickness of the first layer was set to 106 m, 2.985 0.597 0.28 5 30
which leads to yþ max ¼ 1:54. As this value is lower than 5, the sub-
3.210 0.642 0.26 4 30
3.432 0.686 0.26 4 30
layer is well resolved. Next, a grid refinement was applied in the
3.657 0.731 0.26 3 30
turbulent wake for 17 m downstream of the rotor to obtain a high 3.880 0.776 0.26 3 30
discretization of the near-wake. As shown in Table 5, even with a 4.101 0.820 0.25 2 30
difference of one million nodes between Meshes B and C, there is 4.328 0.865 0.26 2 30
no significant change in values of mechanical power and maximum 4.550 0.910 0.26 2 30
4.776 0.955 0.26 2 30
vapor volume fraction rv , evidencing the convergence of numerical 5.000 1.000 0.11 1 30
results negating any further mesh refinement. Due to geometrical
similarity between the corrected and uncorrected rotor, another
mesh convergence was not done for the uncorrected rotor and
Table 7
the features of Mesh C were replicated to generate the mesh of this
Blade chord and twist distributions for foil NACA 653  618 for rotor with blade
rotor. uncorrected.
The total vapor volume was used to quantify cavitation. In this
Radial distance, Span station Chord length Twist Twist axis
way, it was considered that cavitation occurs only when the vol-
r (m) (r/R) (m) (°) (% chord)
ume fraction of vapor is greater than 0.01. For the uncorrected
0.7474 0.14948 0.27 24 30
rotor, the simulation indicates a vapor volume equal to 228:7 ml,
0.9611 0.19222 0.56 20 30
while for the corrected rotor it indicates 18:9 ml. In other words, 1.175 0.235 0.58 16 30
the numerical simulations show that the uncorrected design, gen- 1.4176 0.28352 0.53 13 30
erated by the classical Glauert’s method [18] produces 12 times 1.6424 0.32848 0.47 11 30
more vapor than the corrected one designed using the present 1.8632 0.37264 0.43 10 30
2.0918 0.41836 0.39 8 30
work. Fig. 8 shows the region where cavitation occurs for the cor- 2.3143 0.46286 0.35 7 30
rected rotor (Fig. 8a) and uncorrected rotor (Fig. 8b). Fig. 8c clarifies 2.5381 0.50762 0.33 6 30
the geometrical differences between the designs. The region where 2.7631 0.55262 0.3 5 30
the blades were modified matches where cavitation occurs. 2.9854 0.59708 0.28 5 30
3.2088 0.64176 0.26 4 30
As specified by the Eq. (2), cavitation inception starts when
3.4334 0.68668 0.25 4 30
r ¼ C Pmin . Fig. 9 illustrates a local comparison between C Pmin as 3.6269 0.72538 0.23 3 30
3.8815 0.7763 0.22 3 30
4.1036 0.82072 0.2 2 30
4.329 0.8658 0.19 2 30
4.55 0.91 0.16 2 30
4.77 0.954 0.13 2 30
5 1 0.05 1 30

a function of radial position obtained by CFD simulations and by

the analytical methodology. Therefore, when the curve of cavita-
tion number r cross the curves of C Pmin , cavitation takes place
in the flow. As can be seen in Fig. 9, both methods present similar
results for C Pmin and accurately identify the location where cavita-
tion may occur (r > 3:5 m).

4. Conclusions

An efficient methodology for non-cavitating HT blade design at

Fig. 10. Velocity diagram for the section of the rotor blade.
rated conditions was performed, extending the Glauert’s optimiza-
tion. A cavitation constraint was implemented in order to avoid
such phenomenon on hydrokinetic blades which commonly
demages rotors. A modification on the load factor (thrust coeffi-
cient) was then employed, and the minimum pressure coefficient
criterion calculated by the XFoil boundary layer code was utilized.
This technique alters the blade local chord without major changes
on the turbine power coefficient, demonstrating its huge advan-
1290 Paulo Augusto Strobel Freitas Silva et al. / Applied Energy 185 (2017) 1281–1291
tage when compared with other models available in the literature.
The approach was evaluated by comparisons with HARP_opt and
computational fluid dynamics using ANSYS CFX with the
Rayleigh-Plesset model to predict vapor production rates. For tip
speed ratios lower than the rated value, the proposed methodology
/iterþ1 opt  /iter
.
ensures that cavitation does not occur due to low relative velocities
at each blade section. On the other hand, for higher tip speed ratios,
cavitation occurs obviously due to increased relative velocity. Even
though acceptable amounts of cavitation are currently not yet
known, this approach might be adopted for HT blade design,
although cavitation erosion performance for suitable blade materi-
als must be analyzed. The main contribution of this methodology is
the development of a new optimization approach applied to
hydrokinetic blades coupled with the minimum pressure coeffi-
cient criterion, which is responsible for the correction on chord dis-
tribution. It is important to note that the twist angle is not
significantly altered by the chord correction. The present work
can be helpful in the development of technologies for energy con-
version of river, tidal and marine currents. Furthermore, as the
approach is an extension of the Glauert’s optimization, which is
one of the most widely used models, its applicability might be
huge, primarily for real situations. However, some limitations
should be analyzed carefully, and there is a need for model valida-
tion using experimental data, as well as analysis of the model in
off-design conditions.
Acknowledgments
The authors would like to thank the Brazilian National Council
for Scientific and Technological Development (CNPq), the Coordi-
nation for the Improvement of Higher Education Personnel of the
Brazilian government (CAPES), the Centrais Elétricas do Brasil (Ele-
tronorte), and the Dean of Research and Graduate Studies of the
University of Pará – Brazil (PROPESP/UFPA) for financial support.
A special thanks to Professor David H. Wood of the Department
of Mechanical and Manufacturing Engineering, Schulich School of
Engineering, University of Calgary, Alberta, Canada, for valuable
technical suggestions.
Appendix A. Algorithm
Iterative procedure for the calculation of optimum chord and
twist angle at each blade section. Mostly equations can be derived
from Fig. 10.
Require: r; X; C L ðaopt Þ; C D ðaopt Þ and V 0
Set initial values for aopt and a0opt . In this work aopt ¼ 1=3 and
a0opt ¼ 0;
for i ¼ 1 to Ns (Number of sections) do
while error > TOL do
iter ¼ iter þ 1;
V 0 ð1aopt Þ
Compute /opt ¼ Xr 1þa0 ;
ð opt Þ
Compute C n ¼ C L ðaopt Þ cos /opt þ C D ðaopt Þ sin /opt and
C t ¼ C L ðaopt Þ sin /opt  C D ðaopt Þ cos /opt , respectively, where
aopt is obtained from maximum C L =C D ;
Compute C T , using Eq. (7);
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
end if
Compute new aopt , using Eq. (9), and new
a0opt ¼ 8prF sin /opt1cos /opt ;
1
BcC t
Compute error ¼
opt
end while
end for
Compute blade geometry.
Appendix B. Geometrical information
Tables 6 and 7 expose the blade chord and twist distributions of
the rotors corrected and uncorrected, respectively.
References
[1] Khan MJ, Bhuyan G, Iqbal MT, Quaicoe JE. Hydrokinetic energy conversion
system and assessment of horizontal and vertical turbines for river and tidal
application: a technological status review. Appl Energy 2009;86:1823–35.
[2] Gney MS, Kaygusuz K. Hydrokinetic energy conversion systems: a technology
status review. Renew Sustain Energy Rev 2010;14:2996–3004.
[3] Raab J. Hydro power: the design, use and function of hydro mechanical,
hydraulic and electrical equipment. Düsseldorf, Germany: VDI-Verlag GmbH;
1985.
[4] Molland AF, Bahaj AS, Chaplin JR, Batten WMJ. Measurement and prediction of
forces pressure and cavitation on 2-D sections suitable for marine current
turbines. Proc Inst Mech E, Part M 2004;218:127–38.
[5] Batten WMJ, Bahaj AS, Molland AF, Chaplin JR. Hydrodynamics of marine
current turbines. Renew Energy 2006;31:249–56.
[6] Bahaj AS, Molland AF, Chaplin JR, Batten WMJ. Power and thrust
measurements of marine current turbines under various hydrodynamic flow
conditions in a cavitation tunnel and a towing tank. Renew Energy: Int J
2007;32(3):407–26.
[7] Goundar JN, Ahmed MR, Lee YH. Numerical and experimental studies on
hydrofoils for marine current turbines. Renew Energy 2012;42:173–9.
[8] Mesquita ALA, Alves ASG. An improved approach for performance prediction of
HAWT using strip theory. Wind Eng 2000;24(6):417–30.
[9] Vaz JRP, Pinho JT, Mesquita ALA. An extension of BEM method applied to
horizontal-axis wind turbine design. Renew Energy 2011;36:1734–40.
[10] Rio Vaz DATD, Vaz JRP, Mesquita ALA, Pinho JT, Brasil Junior AP. Optimum
aerodynamic design for wind turbine blade with a rankine vortex wake.
Renew Energy 2013;55:296–304.
[11] Goundar JN, Ahmed MR. Marine current energy resource assessment and
design of a marine current turbine for Fiji. Renew Energy 2014;65:14–22.
[12] Sale D, Jonkman J, Musial W. Hydrodynamic optimization method and design
code for stall-regulated hydrokinetic turbine rotors. In: ASME 28th
international conference on ocean, offshore, and arctic engineering, Hawaii,
Honolulu; 2009.
[13] Jo CH, Yim JY, Lee KH, Rho YH. Performance of horizontal axis tidal current
turbine by blade configuration. Renew Energy 2012;42:195–206.
[14] Singh PM, Choi YD. Shape design and numerical analysis on a 1MW tidal
current turbine for the south-western coast of Korea. Renew Energy
2014;68:485–93.
[15] Kang S, Yang X, Sotiropoulos F. On the onset of wake meandering for an axial
flow turbine in a turbulent open channel flow. J Fluid Mech
2014;744:376–403.
[16] Lee JH, Park S, Kim DH, Rhee SH, Kim MC. Computational methods for
performance analysis of horizontal axis tidal stream turbines. Appl Energy
2012;98:512–23.
[17] Drela M. XFoil: analysis and design system for low Reynolds number airfoils.
In: Conference on low Reynolds number airfoil aerodynamics, University of
Notre Dame; 1989.
[18] Glauert H. Aerodynamic theory. In: Durand WF, editor. Chapter XI. Division L.
Airplanes Propellers, vol. 4; 1935. p. 191–5 [reprinted, Dover, NewYork; 1963].
[19] Hansen M. Aerodynamics of wind turbines. 2nd ed. Earthscan; 2008.
[20] Burton T, Sharpe D, Jenkins N, Bossanyi E. Wind energy handbook. John Wiley
and Sons; 2001.
[21] Clifton-Smith MJ. Wind turbine blade optimisation with tip loss corrections.
2 h  i2 Wind Eng 2009;33(5):477496.
Compute W ¼ V 0 1  aopt þ Xr 1 þ a0opt ; [22] Plesset MS, Prosperetti A. Bubble dynamics and cavitation. Annu Rev Fluid
Mech 1977;9:145–85.
Compute cucopt , using Eq. (5) and bopt ¼ /opt  aopt ; [23] Benjanirat S, Sankar LN. Evaluation of turbulence models for the prediction of
Compute V CAV , using Eq. (4); wind turbine aerodynamics. AIAA; 2003, 0517.
if W > V CAV then [24] Lanzafame R, Mauro S, Messina M. Wind turbine CFD modeling using a
correlation-based transitional model. Renew Energy 2013;52:31–9.
Compute ccoopt , using Eq. (8) [25] Pope SB. Turbulent flows. Cambridge; 2000.
[26] Menter FR. Two-equation eddy-viscosity turbulence models for engineering
applications. AIAA J 1994;32(8):1598–605.
 Paulo Augusto Strobel Freitas Silva et al. / Applied Energy 185 (2017) 1281–1291
[27] Znidarcic A, Mettin R, Dular M. Modeling cavitation in a rapidly changing
pressure field: application to a small ultrasonic horn. Ultrasonics Sonochem
2015;22(January):482–92.
[28] Morgut M, Nobile E. Numerical predictions of cavitating flow around model
scale propellers by CFD and advanced model calibration. Int J Rotating
Machinery 2012:11. http://dx.doi.org/10.1155/2012/618180 618180.
[29] Bakir F, Rey R, Gerber AG, Belamri T, Hutchinson B. Numerical and
experimental investigations of the cavitating behaviour of an inducer. Int J
Rotating Machinery 2004;10:15–25.
[30] Frank T, Lifante C, Jebauer S, Kuntz M, Rieck K. CFD simulation of cloud and tip
vortex cavitation on hydrofoils. In: 6th International conference on multiphase
flow; 2007.
[31] Gerber AGA. CFD model for devices operating under extensive cavitation
conditions. IMECE 2002, New Orleans, November 17–22; 2002, 39315.
1291
[32] Benini E. Significance of blade element theory in performance prediction of
marine propellers. Ocean Eng 2004;31:957–74.
[33] Mo Jang-Oh, Lee Young-ho. CFD investigation on the aerodynamic
characteristics of small-sized wind turbine of NREL PHASE VI operating with
a stall-regulated method. J Mech Sci Technol 2012;26:81–92.
[34] Yelmule MM, EswaraRao AVSJ. CFD predictions of NREL phase VI rotor
experiments in NASA/AMES wind tunnel. Int J Renew Energy Res (IJRER)
2013;3(2):261–9.
[35] Sale D. HARP_Opt user’s guide, National Renewable Energy Laboratory (NREL)
United States of America. Last revised on June 26, 2010 for version 2.00.00.
[36] Schlichting H, Gersten K, Gersten K. Boundary-layer theory. 8th ed. Springer;
2000.