Professional Documents
Culture Documents
net/publication/337524092
CITATIONS READS
0 12
6 authors, including:
Some of the authors of this publication are also working on these related projects:
Experimental and Numerical Analysis of Wave Energy Conversion Systems View project
All content following this page was uploaded by Tiziano Ghisu on 07 June 2021.
1 Introduction
The growing demand for renewable energy has drawn the atten-
tion to the strong potential of ocean-wave energy [1]. Among the
different technologies that have been studied and tested for the
conversion of ocean-wave energy into electrical energy, oscillat-
ing water column (OWC) systems represent a reliable and simple
solution [2]. Figure 1(a) presents a schematic of an OWC system,
composed of two main units: an open chamber, partially sub-
merged under the sea free surface, where the water movement
induces an alternative movement of a column of air, and a turbine
driven by the air flow. The periodic inversion of the air flow in the
OWC chamber requires a system that is capable of maintaining
the same direction of rotation regardless of the direction of the air
flow. The Wells turbine, patented by Wells in 1970s [3], is charac-
terized by a symmetrical blade profile staggered at 90 deg with
respect to the axis of rotation, and it represents a solution of self-
rectifying air turbine (Fig. 1(b)), which ensures similar perform-
ance during outflow (air flowing out of the chamber) and inflow
(air flowing into the chamber).
Owing to their simplicity of construction and reliability, Wells
turbines have been widely studied as power-take-offs in OWC
systems, both experimentally [4–6] and numerically [7–9]. In
recent years, a number of authors [10–13] have analyzed the tur-
bine performance from a second-law point of view, trying to esti-
mate the turbine efficiency based on an exergy balance,
evaluating the entropy produced by the air flow through the
machine.
1
Corresponding author.
Contributed by the Fluids Engineering Division of ASME for publication in the
JOURNAL OF FLUIDS ENGINEERING. Manuscript received July 14, 2020; final manuscript
received December 27, 2020; published online February 9, 2021. Assoc. Editor: Fig. 1 Working principle of a OWC system and a Wells turbine:
Arindam Banerjee. (a) OWC system and (b) Wells turbine
ð2
TX jW_ pol j dpt
gtt ¼ (3) jwpol j ¼ ¼ (10)
Dpt Q m_ 1 qt
ð2
jW_ j ¼ m_ jwj ¼ m_ dht ¼ mc
_ p ðTt1 Tt2 Þ ¼ T X (20)
1
where r ¼ dsV =dt represents the entropy generation rate per unit
mass, and CV a control volume enclosing the turbine.
Putting together Eqs. (18), (19), and (20) with the definitions of
Fig. 2 Representation of specific works on the Tt–s plane first- and second-law efficiencies, Eqs. (3) and (7), it follows that:
TX TX
ð2 gII ¼ ¼ gtt (22)
W_ lost T X þ Tt;ref S_ G Dpt Q
wlost ¼ ¼ Tt ds (11)
m_ 1
Equation (22) is a direct consequence of the equivalence of
jwj ¼ jwpol j wlost (12) polytropic and isentropic work for a low pressure ratio process. In
light of the growing interest in evaluating the performance of
We can also define an isentropic work, referring to a transfor- Wells turbines with the second-law efficiency, the validity of this
mation at constant entropy between the initial and final total approximation will be verified in this paper with several turbu-
pressures lence models and for different Wells turbine rotors.
ð 2is
jW_ is j 3 Entropy Calculation From Computational Fluid
jwis j ¼ ¼ dht ¼ cp ðTt1 Tt2;is Þ (13)
m_ 1 Dynamics (RANS) Simulations
The quantities required to estimate the efficiency of a Wells tur-
Figure 2 reports graphically the difference between polytropic,
bine can be calculated using numerical simulations, i.e., by solv-
isentropic, actual, and lost specific works for a machine evolving
ing the governing Navier–Stokes (NS) equations numerically.
a compressible flow.
These quantities include the turbine torque T , the static and total
The isentropic work represents the maximum work that a tur-
pressure drops Dp and Dpt , the volumetric flow rate Q, and the
bine can exchange with the fluid (i.e., the available energy per
entropy generation rate S_ G .
unit time), while the polytropic work is the sum of actual work
As a direct numerical simulation of the NS equations for most
and lost work (due to friction). In a turbine, the polytropic work is
flows of industrial interest is still beyond the capability of current
always greater than the isentropic work, with the difference being
computers, the most common and practical approach is the solu-
the heating work gain [27]
tion of the Reynolds–averaged Navier–Stokes (RANS) equations,
jwpol j ¼ jwj þ wlost > jwis j (14) which are derived by replacing the instantaneous flow variables in
the NS with the sum of a mean value and a fluctuating component
Conversely, the isentropic efficiency gis ¼ jwj=jwis j is always with zero mean value (u ¼ u þ u0 ).
greater than the polytropic efficiency gpol ¼ jwj=jwpol j. The latter For a compressible flow, assuming negligible heat transfer as
is more representative of the inefficiencies in turbomachine, as it common in turbomachinery applications, the RANS equations of
does not take into account the heating work gain and only consid- conservation of mass, momentum, and energy read as follows
ers the lost work. [33]:
For low pressure ratio machines (as in the case of Wells tur- 8
bine) ðTt1 Tt2;is Þ Tt1 , and the following approximations can > @q
>
> þ r qV ¼ 0
be made >
>
>
> @t
>
>
wpol wis ) gpol gis (15) <
@qV
þ r qV V þ qV0 V0 ¼ rp þ r P (23)
ð2 > @t
>
>
>
Tt ds Tt;ref ðs2 s1 Þ (16) >
> @qh
>
> Dp
1 >
: þ r qV h ¼ þ P : rV þ P0 : rV0
@t Dt
provided that Tt;ref is chosen appropriately (i.e., in the range of
temperatures involved in the transformation). Under these where V is the velocity vector, h is the static enthalpy, and P is
assumptions the deviatoric stress tensor, that for a Newtonian flow can be
expressed as the sum of the contributions due to the strain rate ten-
jwj jwpol j Tt;ref ðs2 s1 Þ (17) sor rS V and the volumetric tensor ðr VÞI, each one multiplied
has been proposed in Ref. [39], where the authors note that
the second approach (Eq. (34)) is more reliable, because in
RANS approaches, e is only used as an intermediate quan-
tity to calculate lt and PR , which interact with the mean
flow through the momentum and energy equations.
(2) In Spalart–Allmaras (S–A) models only the second Fig. 3 Computational domain and mesh for simulations: (a)
approach can be used, as the method solves directly a trans- computational domain and (b) blade and hub mesh
port equation for kinematic eddy viscosity ~ , which is
related to lT through the following expressions:
[9,13,21,48–51]. The OWC chamber is not included in the domain
ð~ = Þ3 (as done, for example, in Refs. [52–54]), because its main effect is
lT ¼ q~
fv1 fv1 ¼ (36) to cause a delay between the movement of the water level in the
ð~ = Þ3 þ C3v1
chamber and the mass-flow in the turbine duct, without significant
where fv1 is the viscous damping function. In addition to the stand- modifications to the turbine performance.
ard formulation of the k e model, the realizable (REAL) k e Uniform inlet boundary conditions have been used for velocity,
formulation has also been considered. It is a newer implementa- total temperature, and turbulent quantities. The velocity has been
tion [41] that differs from the original for a new formulation of set to a value, fixed in steady (fixed-/) simulations and sinusoi-
the turbulent viscosity and a new transport equation for the dissi- dally variable in time in dynamic simulations, chosen to obtain
R
pation rate e. The FLUENTV User’s Guide [42] states that the k e the required value or range of flow coefficients (see Eq. (1)).
REAL model provides superior performance for flows involving Figure 4 shows a typical variation of flow coefficient /, where Tw
rotation, boundary layers under strong adverse pressure gradients, represents the wave period. The inlet total temperature has been
separation, and recirculation. set to 288 K and the turbulent quantities are calculated by the
solver based on the values set for turbulent intensity and length
scale (2% and 7% of the blade height, as suggested in Ref. [42]).
4 Methodology A uniform value of static pressure has been specified at the outlet,
The turbine geometry and operating conditions simulated in while periodic boundary conditions have been used at the two
this work are the ones presented in the experimental work from sides of the passage of the computational domain (Fig. 3). Inlet
Setoguchi et al. [43]. The main details are summarized in Table 1. and outlet are inverted for negative values of the flow coefficient.
The domain for the numerical simulations is reported in A multiblock structured grid has been used to discretize the vol-
Fig. 3(a): it is a straight duct representing a single blade passage ume, with a C-grid around the blade able to capture the boundary
of the turbine, with periodic boundary conditions at the two sides. layer flow and an H-grid in the rest of domain, see Fig. 3(b).
Simulating a single passage of a tubomachinery’s blade row (or Four different turbulence closure models have been compared:
even multiple single passages from different blade rows, with an the k x shear stress transport (SST), the standard (STD) and
appropriate treatment of the interrow interface) is a common prac- REAL k e and the Spalart–Allmaras (S–A) models. Numerical
tice in turbomachinery RANS simulations [44–46] when the simulations have been conducted using the commercial CFD soft-
hypothesis of periodic flow with respect to the blade pitch is valid, ware ANSYS FLUENT 17.0. The SIMPLEC algorithm has been used
i.e., in the absence of flow structures larger than the blade pitch. for the pressure–velocity coupling, a second-order centered
This approximation has been often used in Wells turbine simula-
tions [8,9,21,47]. A comparison between the simulation of the
Wells turbine’s full rotor and of a single passage was conducted
by the authors, showing a maximum difference in the performance
coefficients of less than 0.1%. The interaction between stationary
and moving parts has been modeled using a “frozen rotor
approach,” also referred to as the “multiple reference frame”
model in the FLUENT User’s Guide [42]. The model can be used for
turbomachinery applications in which rotor–stator interaction is
relatively weak, and the flow is relatively uncomplicated at the
interface between the moving and stationary zones. This seems to
be a good approximation for the present case study, where the
interaction between moving parts is only at the hub and casing of
the duct where the turbine is housed. The same model has been Fig. 4 Inlet boundary condition for dynamic simulations, in
used in the majority of previous CFD analyses of Wells turbines terms of flow coefficient /
Sext Sj
N3 ¼ NE 17,376,540
GCIj ¼ 1:25
(39) r12 1.43
Sext
r23 1.43
S1 =Sext 0.9964 1.0006
while [57], in the presence of oscillatory convergence, suggest S2 =Sext 0.9995 0.9994
using half the range of variability in the data S3 =Sext 0.9999 1.0003
p 5.37 0.69
1 GCI1 0.062% 0.244%
UStern ¼ jmaxðSj Þ minðSj Þj (40)
2 GCI2 0.009% 0.244%
GCI3 0.001% 0.133%
Figure 5 presents the convergence of p and T for a flow coef- UStern — 0.061
ficient / ¼ 0:16, using five grids with different number of cells
Fig. 7 Verification of the temporal discretization using five differ- 5 Results—Comparison Among Different Turbulence
ent time-step sizes with a maximum flow coefficient /max 5 0.23: (a) Models
non-dimensional torque history (final period), (b) Non-dimensional
torque as a function of the flow coefficient, (c) non-dimensional In this paragraph, simulations with different turbulence models
entropy generation (final period), and (d) non-dimensional entropy are reported for the same rotor geometry, i.e., the rotor with six
generation as a function of the flow coefficient blades (rotor tip solidity equal to 0.57) and NACA0015 blade
The last two methods are the indirect and direct approaches
described by Herwig [36]. The exergy calculation ES has been Fig. 13 Rotor efficiencies averaged on cycle and calculated with
evaluated both considering the generation, ES;g , and dissipation, different turbulence closure models: (a) Non-dimensional torque,
ES;d , of turbulent kinetic energy (Eqs. (41) and (42)). (b) non-dimensional total pressure drop, (c) non-dimensional
It is interesting to note how the traditional approach (EP) and static pressure drop, and (d) non-dimensional entropy production
the indirect method lead to same results, and this is a confirmation
of the validity of the assumptions made in Sec. 2.3
more evident near the suction side of the blade and near the trail-
(Dpt Q mc _ p DTt mT
_ t;ref Ds). On the contrary, the direct method
ing edge, at all spanwise positions from hub to tip.
(which requires the integration of the entropy production in the
The above results have been used to evaluate the turbine effi-
computational domain) based on the turbulent kinetic energy pro-
ciencies defined in Sec. 2. The values reported in Fig. 13 are a
duction (ES;g ) leads to an overestimation of the available energy.
direct consequence of the results in Figs. 9–11. The aerodynamic
This difference is not too significant for the k x model (about
efficiency gad is lower than the total-to-total efficiency gtt, as the
2%), larger for the other turbulence models, and especially for the
static pressure drop that appears in its denominator is larger than
k e STD (about 9%). This result is in line with the differences
the total pressure drop that is used to calculate the first-law effi-
encountered when estimating the entropy generation in other
ciency (the former includes the exit dynamic head, which in a
applications: differences as large as 15% are not uncommon [72],
Wells turbine is lost). The second-law efficiency (which theoreti-
and are due to the fact that CFD software do not solve the entropy
cally should be approximately equal to the latter) has a very simi-
equation, which therefore can be not strictly satisfied due to
lar value only when the denominator is calculated using the
numerical errors, as explained in Ref. [73]. Lower discrepancies
indirect method, while it is lower when the direct method is
between the direct method and the other ones can be obtained
selected. A further difference exists depending on whether the
when using the turbulent kinetic energy dissipation (ES;d ). The
entropy generation S_ G is calculated using the dissipation or the
overestimation still remains high for k e models, but it almost
production of turbulent kinetic energy, being the latter slightly
disappears for the k x SST model.
larger than the former (see Fig. 10). These differences are smaller
The larger estimation of the available energy from
for the k x model than for Spalart–Allmaras and k e models.
Spalart–Allmaras and k e models is linked to the overestimation
The second-law efficiency calculated using the turbulent kinetic
of local entropy generation (see Fig. 10), which is due to the dif-
energy dissipation with the k x SST model is remarkably close
ferent treatment of the boundary layer region, assumed fully tur-
to the first-law efficiency.
bulent [70]. On the contrary, the k x SST model adopts the
standard k e model only away from the walls and an improved
formulation within the boundary layer [71], where viscous effects 5.2 Simulations With /max 5 0:345. Figure 14 reports the
predominate over turbulent ones. performance parameters of the Wells turbine with an operating
This difference in boundary layer treatment among k x SST (sinusoidal) cycle with a maximum flow coefficient sufficient to
and k e STD model is highlighted in Fig. 12, that clearly shows produce deep stall conditions.
higher intensity of nondimensional entropy production in the When comparing the curves in Fig. 14, the k e model in its
boundary layer region for the k e model. This overestimation is standard formulation is unable to predict the turbine stall and the
7 Conclusions