Professional Documents
Culture Documents
Renewable Energy
journal homepage: www.elsevier.com/locate/renene
a r t i c l e i n f o a b s t r a c t
Article history: Computational Fluid Dynamics is thought to provide in the near future an essential contribution to the
Received 4 December 2013 development of vertical-axis wind turbines, helping this technology to rise towards a more mature in-
Received in revised form dustrial diffusion. The unsteady ow past rotating blades is, however, one of the most challenging ap-
3 April 2015
plications for a numerical simulation and some critical issues have not been settled yet.
Accepted 16 June 2015
Available online xxx
In this work, an extended analysis is presented which has been carried out with the nal aim of
identifying the most effective simulation settings to ensure a reliable fully-unsteady, two-dimensional
simulation of an H-type Darrieus turbine.
Keywords:
CFD
Moving from an extended literature survey, the main analysis parameters have been selected and their
Darrieus inuence has been analyzed together with the mutual inuences between them; the benets and
Wind turbine drawbacks of the proposed approach are also discussed.
Simulation The selected settings were applied to simulate the geometry of a real rotor which was tested in the
Unsteady wind tunnel, obtaining notable agreement between numerical estimations and experimental data.
Sensitivity analysis Moreover, the proposed approach was further validated by means of two other sets of simulations, based
on literature study-cases.
2015 Elsevier Ltd. All rights reserved.
1. Introduction and with very reduced computational cost [9]. In addition, some
more advanced techniques are presently available like wake
Darrieus Vertical-Axis Wind Turbines (VAWTs) are receiving models, vortex models or the Actuator Cylinder ow model [13].
increasing interest in the wind energy scenario, as this turbine As discussed by several authors (e.g. Refs. [14,15]), however, an
typology is thought to represent the most suitable solution in non- accurate modeling of these machines cannot disregard anymore
conventional installation areas, due to the reduced variations of the the recent developments in CFD simulations, as they can signi-
power coefcient even in turbulent and unstructured ows cantly contribute to the technological improvement in designing
(Refs. from Ref. [1e7]), with low noise emissions and high reli- the rotors, needed to rise the technology towards a well-
ability. Moreover, this technology is also gaining popularity for established industrial production. On this basis, one can easily
large-size oating off-shore installations (e.g. Ref. [8]). argue that the goal of assessing a reliable approach to CFD simu-
The design and development of these rotors have been histori- lation of Darrieus turbines is thought to represent one of the most
cally carried out with relatively simple computational tools based challenging prospects for the future wind energy research.
on the BEM (Blade Element Momentum) theory [9e12]. This kind of Some of the most complex and less understood phenomena in
approach can still provide some advantages in many cases, espe- the eld of numerical simulations are involved in the analysis of the
cially concerning the preliminary design of a machine (e.g. overall ow past rotating blades [9]. With particular reference to Darrieus
dimensions and attended power), as it is generally quite reliable wind turbines, the problems to be solved to correctly describe the
ow eld developing around the turbine are increased by the
constant variation of the incidence angle with the azimuthal po-
sition of the blade and the strong interaction between the upwind
* Corresponding author. and the downwind halves of the rotor ([9] and [12]). Moreover, a
E-mail addresses: balduzzi@vega.de.uni.it (F. Balduzzi), bianchini@vega.de.
major aspect of the unsteady aerodynamics of Darrieus rotors is
uni.it (A. Bianchini), maleci@vega.de.uni.it (R. Maleci), giovanni.ferrara@uni.it
(G. Ferrara), lorenzo.ferrari@iccom.cnr.it (L. Ferrari). represented by dynamic stall, which often occurs at low tip-speed
http://dx.doi.org/10.1016/j.renene.2015.06.048
0960-1481/ 2015 Elsevier Ltd. All rights reserved.
420 F. Balduzzi et al. / Renewable Energy 85 (2016) 419e435
ratios (TSRs), where the range of variation of the incidence angle on reliable tools for the development of new machines, with notice-
the airfoils is larger ([9] and [16,17]). able cost and time savings. On the other hand, the great advantages
Within this scenario, a relevant aspect which has not been often of CFD simulations in such complex phenomena, like those con-
discussed in sufcient detail in the technical literature is the nected to rotating blades, can be nullied if unsuitable settings are
philosophical approach to CFD simulations, i.e. the goal of the implemented by the user. In particular, a proper denition of the
simulations themselves and the most suitable tools to achieve it. In computational parameters can be obtained only by means of a
detail, if one would go through the problem with a logical approach, thorough sensitivity analysis on the inuence of these variables on
two main observations can be promptly made: the results accuracy, possibly coupled with a validation study based
on experimental data.
The functioning principle of vertical-axis wind turbines, where Within the present study, a detailed review on the state-of-the-
the ow conditions seen by the blades changes instant by art of numerical approaches for the simulation of Darrieus turbines
instant as a function of the position occupied in the revolution was rst carried out by the authors. Upon examination of the
trajectory, made any typology of simulation denitely ineffec- literature, several works were identied ([15] and [20e39]), all
tive, with the only exception of a fully unsteady approach (i.e. published in the past ve years; as one may then notice, the topic is
the only able to catch the real interactions between the blades). still quite new and there is a lack of extensive studies for the
Based on the above, the circumferential symmetry cannot be denition of practical guidelines to properly model the ow around
exploited like in many other turbomachinery applications: the a Darrieus turbine's blade.
full revolution of the blade must hence be described, leading to As a result, even though the considered studies are all focused
very heavy simulations in terms of mesh size and computational on the evaluation of the average power coefcient as a function of
time. the TSR and/or the instantaneous power coefcient as a function of
the azimuthal position of the blade, poor agreement was found on
Moving forward in the analysis, although the three-dimensional the most suitable settings to be adopted for the simulation. An
approach is the only able to really describe the ow eld around the effective convergence was indeed found only on the bases of the
turbine (i.e. also the real performance), some considerations are simulating approach, i.e.:
here proposed to focus the attention on benets, drawbacks and
requirements of the a 2D or a 3D approach. In particular: The unsteady approach ([ [15] and [20e34]) is largely preferred
to the steady-state ([35e37]) or to the multiple reference frame
A 3D approach is needed in case the simulations are deemed to ([38] and [39]) approaches. In the unsteady approach, the
provide the attended power output of the rotor. In these cases, rotating machine is simulated with two distinct sub-grids: a
the inuence of spanwise velocity components, tip effects and circular zone containing the turbine geometry, and rotating
interactions with the parasitic components (e.g. struts, tower, with its angular velocity, and a xed outer zone (with a rect-
etc.) cannot indeed be neglected ([9] [12], and [18]). angular shape in most cases), which denes the boundaries of
By doing so, enormous computational resources are generally the overall calculation domain. The two regions communicate
needed [19] and in some cases (e.g. Ref. [20]) authors have by means of a sliding interface.
proposed to apply different settings to three-dimensional sim- Boundary Conditions: as widely accepted for similar simula-
ulations with respect to what has been dened for the lighter tions, a velocity inlet and a pressure outlet are used in the
two-dimensional calculations. mainstream direction, whereas lateral boundaries are threated
Some applications, however, do not indeed require an exact either as solid walls or with the symmetry condition ([15]
estimation of the overall performance of the rotor. In particular, [20e22] [24], [27,-28] [30], [33], [35,36]).
if properly assessed, a 2D approach could be successfully applied A fully 3D approach is rarely adopted ([33,34], and [38]). In all
to the analysis of many relevant issues connected to the func- other cases, a 2D approach is basically used, sometimes
tioning of Darrieus rotors, like the dynamic stall, the ow cur- compared with a 3D attempt with very raw settings due to the
vature effects and the wake interaction with the downwind half enormous calculation costs ([20,21,23,30], and [37]).
of the revolution [9]. Moreover, a reliable 2D simulation, The simulations are mainly performed with the commercial
coupled with simplied models to account for the main sec- code ANSYS Fluent ([15] [21e24] [26e29], [31], and [33e36]).
ondary effects [12], could also provide a rst estimation of the The accuracy of the numerical results is usually checked by
overall performance of the rotor, to be compared and integrated means of experimental data derived from wind tunnel mea-
with the results of the BEM codes conventionally exploited by surements ([15] [21e26], [28,-29]).
industrial manufacturers.
Since the present study was conceived in view of a 2D unsteady
Based on the above and observing that no agreement was found approach, a more extensive analysis of the studies [15] and [20e32]
between the most accredited literature sources, in this work an is given in Table 1; the goal of this comparative analysis was in fact
extended analysis on the critical issues to properly perform a 2D to highlight whether some general tendencies could have been
simulation of a Darrieus rotor is presented and discussed. The found among the considered cases.
assessment of a reliable setting for this type of approach can pro- In particular, the benchmark was focused on the followings
vide a very useful tool to more in depth analyze the real functioning parameters:
of the turbines; contemporarily, it could represent the basis for a
future extension of the analysis to full 3D models. Turbulence modeling approach and models
Numerical settings (solution algorithm and methods for the
2. Literature review discretization of the NeS equations)
Time-dependent solution settings (angular discretization and
The ongoing evolution of CFD solvers is providing new oppor- global duration of the calculation)
tunities for wind turbine designers to enhance the comprehension Distance of the domain boundaries (inlet, outlet, lateral and
of the real blades-ow interaction; in addition, the diffusion of sliding interface) from the turbine
commercial codes is thought to guarantee in the next future
F. Balduzzi et al. / Renewable Energy 85 (2016) 419e435 421
Table 1
Comparative analysis of the literature settings for 2D unsteady simulations.
Simulation settings
Domain dimensions
Mesh
Discretization of the boundary layer (y and number of nodes on always smaller than twice the rotor diameter, in order to reduce the
the airfoil contour) computational resources needed in performing the mesh motion.
Overall number of mesh elements and mesh typology The nal section of the comparative literature survey was
dedicated to the analysis of the mesh settings, whose properties
No agreement was found on the choice of the turbulence model heavily affect the accuracy and predictability of the results.
between the various works since all kinds of RANS approaches In particular, in the near-blade region a suitable resolution of the
were proposed at least once, including the one-equation modeling mesh in the direction orthogonal to the solid walls is convention-
and all the most common formulations of the two-equation ally recommended to properly compute the boundary layer prole
models, as well as an application of DES and LES modeling. More- (y), while the number of nodes in which the airfoil is discretized
over, in the strategy for the near-wall treatment both the Wall (NN) is crucial for the determination of both the incidence angle of
Functions and the Low Reynolds approaches were implemented. the incoming ow on the blade and the boundary layer evolution
On the other hand, the numerical settings are almost assessed: the from the leading edge to the trailing edge. Moreover, the dis-
preferred algorithm for the pressureevelocity coupling is the cretization level adopted in the near-blade region also controls the
transient SIMPLE (easier settings, good stability, standard solution total number of mesh elements (NE), since the growth rate of the
for the Fluent code), while the discretization of the NeS equations element's size must be small enough to avoid discontinuities.
is mainly based on a 2nd order scheme, both for the time and the Upon examination of Table 1, it is readily noticeable that most of
spatial derivatives (the upwind scheme is preferred, due to a good the analyzed studies chose a direct resolution of the boundary layer
compromise between stability and accuracy). prole, with y values essentially lower than 5. Notwithstanding
The choices for time-dependent solution parameters show again this, the resolutions of both the blade prole and the global
uncertainty on the values needed to achieve a proper accuracy: the computational domain can vary by more than one order of
angular time-step (Dw) is ranging from 1/15 e2 depending on the magnitude and the strategies implemented cannot be standardized
specic application, while the revolutions completed by the since a common rule cannot be found. The elements type used for
rotating region in order to reach a stable and a repetitive torque the grid generation is also not shared: both structured (quadrilat-
prole vary from 4 to about 15. As a general indication, the most eral elements) and unstructured (triangular elements) meshes
exploited convergence criterion in the literature is to compare the were applied and an extrusion of layers of quadrilateral elements in
average value of the torque over a complete rotation between two the near-wall zone is alternatively used.
subsequent revolutions; in most of the works, simulations are
stopped when this difference becomes lower than 1%. 3. Study case
The overall domain dimensions used in the majority of the
studies are comparable to the usual values for a free ow around an A reference rotor for the study was rst selected. Thanks to the
obstacle, being proportional to the size of the obstacle itself (in this possibility of exploiting a real full-scale model of an industrial rotor
case, the rotor diameter D). Both the inlet and the lateral boundaries
are placed at a distance (L1 and W) from the rotational axis of the
turbine of about 5e10 diameters, while the outlet boundary at a Table 2
Main features of the analyzed rotor.
distance (L2) of 10e20 diameters. Only one exception [15] conicts
with the global trend, characterized by an overall domain length of Blades number (N) 3
about 100 diameters and a width of 80 diameters. It has to be Blades Shape Straight
Blades Airfoil (attended) NACA0018
noticed that the simulations can be performed for a turbine placed
Radius (R) [m] 0.850
either in an open eld or inside a wind tunnel for the validation with Chord (c) [m] 0.246
experimental data, determining in the latter case a constraint for the AR blades 12
domain width. Finally, the diameter of the rotating region (DRR) is Solidity (s) 0.44
422 F. Balduzzi et al. / Renewable Energy 85 (2016) 419e435
[40,41] for the experimental validation, the geometry considered in The introduction of a conservative circular sliding interface,
the study is summarized in Table 2. dened by the diameter of the rotating region (DRR), guaranteed the
The turbine had three very long straight blades (Aspect Ratio connectivity between the two separated regions as well as the
higher than 12), realized with an extruded aluminum technology relative motion of the components. For the denition of the rotor
which allowed a very accurate reproduction of the airfoils' geom- geometry, only the turbine's blades were taken into account,
etry. Moreover, rectangular end-plates with rounded off angles neglecting the presence of both the supporting struts and the shaft.
were added at the end on each blade in order to further mitigate The main parameters of the xed area are shown in Fig. 1: the
the tip-losses. Each blade was supported by two airfoil-shaped velocity inlet and pressure outlet boundary conditions were placed
struts and a central thin tie-rod, which connected it to a steel respectively at a L1 distance upwind and L2 distance downwind
central shaft with a very small diameter (less than 0.05 m), in order with respect to the rotational axis of the turbine, while a symmetry
to minimize the shadowing effect on the downwind blades. condition was assigned to the lateral boundaries, identied by the
It is worth pointing out that, based on the indications by width W. The symmetry condition for lateral boundaries is indeed
Migliore [44], the geometric airfoils tested onboard the model were the most common solution for this type of simulations (see litera-
cambered proles, obtained by a conformal transformation of the ture survey). An alternative option, however, could be represented
NACA0018 section by the turbine's radius to compensate the ow by opening-type conditions (i.e. able to support simultaneous
curvature effects [41]. inow and outow over a single region), which could enable a
As the test model was a pre-industrial prototype of a real ma- reduction of domain width. Due to possible instabilities generated
chine, no images of the rotor can be shown here due to a non- by this type of setting, however, in this work the conservative
disclosure agreement with the industrial partner. choice of symmetry conditions was maintained.
The turbine was tested in a large fully open-jet wind tunnel in Italy, Due to the remarkable inuence of the domain sizes on the
able to provide an oncoming ow velocity in the testing section up to correct description of the ow eld past the turbine, a specic
70 m/s, with a ow distortion in terms of velocity variation lower than analysis on these parameters was carried out, whose results will be
0.5% [40,41]. The testing section was more than 8 times larger than the discussed later in the study.
front area of the rotor, whereas the jet length was more than 5 times The 2nd order upwind scheme was used for the spatial dis-
the turbine's diameter. As the ow can pass around the object freely, cretization of all the equations including a transport term (i.e.
this tunnel type is thought to avoid any blockage effect on models up momentum, energy and turbulence), as well as the bounded 2nd
to two times larger than the present rotor, even if supposed to be order implicit for the time differencing, to achieve a good resolu-
totally solid. Although proper blockage corrections for VAWTs in open tion. In order to ensure a stable simulation, the initialization of the
wind tunnels are presently missing, the authors have, however, solution was performed by rst calculating the steady-state ow
estimated that blockage can be here neglected based on analogies around the rotor, obtaining an initial guess for the unknown vari-
with some literature works on open-jet wind tunnels [42,43]. ables. Then, the time-dependent simulation started with a 1st order
Both numerical simulations and experimental tests were per- differencing scheme in both time and space, switching only after
formed with an undisturbed wind speed of 8 m/s, corresponding to few revolutions to the 2nd order scheme.
local Reynolds numbers on the blades in the order of 2 105. As indicated by the literature, a converged solution was ach-
The rotating axis was connected to an electric motor/brake, ieved by running the simulations until a periodic behavior was
which was used to explore the entire power curve of the machine. reached; the global convergence was monitored comparing the
The torque output and the turbine revolution speed were measured average value of the torque over a complete revolution between
with a high precision torque meter inserted between the shaft and two subsequent revolutions. Contrary to the literature, however,
the pulley of the driving belt: the torque meter had a full-scale of the sensitivity of the results on the selected threshold for torque
100 Nm with an accuracy of 0.1% of the FSO. assessment was specically investigated in this work (see Section
In particular, in order to obtain experimental data that could be 4.2).
coherently compared with those coming from 2D simulations, the Beyond these basic assumptions, however, the lack of agree-
net torque of the blades was calculated by purging the global torque ment in the technical literature led this study to investigate the
output from the parasitic torque contribution of the rotating struts following parameters within the indicated ranges:
and the tower, whose torque was measured in a second run after
the blades were removed. Turbulence model: Standard k-, RNG k- and k-u SST
Solver type: pressure-based, density-based
4. Main simulation settings
Fluid properties: compressible, incompressible the Shear-Stress Transport (SST) k-u model was chosen to model the
Solution Algorithm: SIMPLE, PISO, Coupled turbulence. The Enhanced Wall Treatment was implemented for the
Angular time-step: 0.27, 0.9 computation of the boundary layer in the near-wall regions, which
Iterations per timestep: 20, 30, 40 introduces a modication in the turbulence model to enable the
Domain dimensions: L1, L2, W viscosity-affected region to be resolved up to the wall.
Number of revolutions to convergence: variable with the tip- The SST model is based on a zonal formulation, which makes use
speed ratio of blending functions in order to switch from a u-based formulation
inside the boundary layer to a -based formulation in the core re-
It is worth pointing out that all the considered variables not only gion of the ow [47]. This model was chosen mainly because it
directly affect the accuracy of the nal result but also have a strong offers the typical advantages of the most used conventional two-
mutual inuence between themselves; as a result, the analysis tried equation turbulence models, k- and k-u, avoiding their respec-
on one hand to decouple the effects of the variables and, on the tive basic shortcomings. In detail, the -based models fail to predict
other hand, to highlight their inuence on other components as the proper behavior of turbulent boundary layers up to separation,
well as on the simulation outcomes. while their application is recommended in free shear ows, i.e. jet
In particular, in the rst phase of the study, focused on the and wake, mixing layers, etc. On the other hand, the use of u-based
assessment of the most effective numerical setting, the boundaries models is suggested for compressible ows and separating ows
were initially placed very far away from the rotor (similar to [15]), under adverse pressure gradients, although they reveal strong
in order to avoid any distortion on the ow eld and allow a specic sensitivity of the solution on arbitrary freestream values of both k
investigation on the parameters affecting the accuracy of the model and u outside the shear layer ([46], [48,49]). Moreover, the cali-
in describing the ow-blades interaction. Moreover, a reference bration of the SST model was originally focused especially on
mesh was created on the basis of the highest renement level smooth surfaces [45].
found in the literature [15]. Both these parameters then became In the case of a real Darrieus functioning, both phenomena
part of the analyzed variables. As a general indication, the multi- actually occur, making the k-u SST approach by far preferable than
variate matrix of tests obtained with the aforementioned parame- other solutions.
ters was in fact analyzed, with the exception of those combinations
involving elements that were already discarded based on specic 4.2. Convergence criteria
evidence. Throughout the study, the general criteria that were used
for the evaluation of the acceptability of the results were: The usual convergence criterion based on the deviation of the
averaged torque value of a blade (or power coefcient) over a
A satisfactory matching between simulation results and exper- complete revolution between two subsequent cycles was here
imental data; thought to represent the most effective solution to ensure a stable
The achievement of insensitivity to the variation of a parameter; behavior of the simulation.
The convergence of residuals (all quantities to 105, except for Contrary to the literature, however, in which the calculations are
momentum to 106) and computational limits. generally stopped as soon as the difference becomes lower than 1%,
a sensitivity study was here carried out.
Moreover, it is here proposed that the most correct way of In particular, it was noticed that the required number of revo-
comparing two simulations should be not only based on the com- lutions cannot be estimated a priori, being notably dependent on
parison between the calculated torque values or torque coefcients, the tip-speed ratio of the turbine. In the study, the minimum
as very often made in the literature. As aggregate parameters, they number of cycles felt in a range between 20 and 90. For example, in
are in fact deemed to potentially hide differences between the Fig. 2, the convergence histories of two simulations at TSR 1.1 and
simulations, due to undesired compensations between different TSR 2.2 with the same settings are reported: the power coef-
zones of the torque prole. On this basis, in the present study the cient was divided by its nal value (CP,F).
attention was focused both on the averaged power coefcient From a perusal of Fig. 2, it is readily noticeable that the number
values over a revolution and on the evaluation of the mean error of revolutions can substantially vary from an operating point to
between the instantaneous torque coefcient proles. another. In addition, the 1% threshold appears not suitable for
similar simulations as the very at convergence trend can intro- computing effort.
duce signicant variations in the nal torque value. For example, in Finally, since the pressure-based solver was specied, the last
case of the TSR 2.2 simulation, the deviation of CP between two requirement was the choice of the algorithm to solve the linkage
subsequent revolutions goes under 1% just after 5 rounds: the value between pressure and velocity. An iterative solution is in fact
at that cycle (i.e. the one accepted on the basis of literature criteria) needed due to both the non-linearity in the NeS equations set and
would however overestimate the nal value (CP,F) of about 9%, the lack of an independent equation for the pressure, as mentioned
introducing a non-negligible detriment of the simulations accuracy. before. The two standard alternative formulations for the pressur-
To overcome this limit, the convergence threshold applied in the eevelocity coupling are the Semi-Implicit Method for Pressure-
study was reduced by one order of magnitude and xed to 0.1% of Linked Equations (SIMPLE) and the Pressure Implicit with Splitting
the CP value between two subsequent revolutions. of Operators (PISO), both based on the solution of an additional
equation for the pressure eld (pressure-correction equation).
4.3. Solver settings Originally, the PISO method was preferable for transient problems
since it was derived from the SIMPLE one specically for unsteady
The assessment of the numerical approach is here discussed. calculations: with sufciently small timesteps, accurate results
First, some considerations on the effects of compressibility are could be obtained with lower computational costs than the SIMPLE
provided. Then, the attention is focused on the choice of the type of algorithm [50]. Both these semi-implicit solution methods are,
solver and solution algorithm to be used for the simulations, ac- however, known to converge slowly since the momentum equation
counting also for the effects of the angular time-step and the and pressure-correction equation are solved separately [45], i.e.
number of inner iterations per timestep. with a segregated approach.
Two different revolution regimes were considered for the Starting from these considerations, a coupled approach (non-
analysis (TSR 1.1 and TSR 2.2), in order to evaluate the effec- segregated) was tested in addition to SIMPLE and PISO algorithms. In
tiveness of the approach for different working conditions of the the Coupled algorithm, the NeS equations set is directly solved
machine (i.e. relative wind speeds and ranges of incidence angles); through an implicit discretization of pressure in the momentum
the two functioning points are in fact located in the unstable and equations, with benets in terms of robustness and convergence,
stable part of the torque curve, respectively. especially with large timesteps or with a poor-quality mesh ([45,52]).
Concerning the method for the treatment of the gas density, it is The worst performance in terms of accuracy of the results was
worth pointing out that the ow through a Darrieus turbine is found with the PISO algorithm: an excessive overestimation of the
characterized by low Mach number values. Therefore, the effects of experimental torque was observed for both the simulated regimes.
compressibility can be considered mild or even null. In these con- As an example, Fig. 3 shows the single blade instantaneous torque
ditions, since the pressure is almost not linked to the density, the coefcient at TSR 1.1 as a function of the azimuthal position
most straightforward choice for the solver type is represented by during a complete revolution.
the pressure-based approach, in which the continuity equation is The simulated case with the PISO algorithm is compared to the
used in combination with the momentum equation to derive an results obtained with the nal settings established at the end of the
equation for the direct solution of the pressure eld. sensitivity analysis, which are more closely matching experimental
On the other hand, the application of a density-based approach data: a complete mismatch in the upwind zone of the rotation is
might not be recommended, as the continuity equation is used as a clearly visible for the PISO curve, with a peak overestimation of
transport equation for density, while the pressure eld is deter- approximately 30%. This behavior is due to a wrong detection of the
mined from the equation of state ([50,51]). Notwithstanding this, stall onset, since the ow remains attached to the blade for further
recent developments and modications in the numerical method- 10 of rotation.
ologies led to the extension of the applicability of both solvers in The attention was then focused on the comparison between
order to properly perform for a wider range of ow conditions [45]. SIMPLE and Coupled algorithm. The torque coefcient prole at
On these bases, a comparative analysis was carried out between TSR 2.2 is reported in Fig. 4 with azimuthal increments of 0.9
the two approaches, whose results are not reported here for and 0.27 for both algorithms: considering the results with the
conciseness reasons. The results demonstrated, however, that the smaller timestep, the two curves are nearly coincident, indicating
pressure-based approach is more stable and has a faster conver- an almost equivalent response of the two algorithms. Nevertheless,
gence rate than the density-based one; moreover, due to its
intrinsic formulation, this latter approach would require very low
residuals and very short timesteps to ensure an accurate solution,
which are, however, very hardly achievable in case of large and
unsteady simulations like those considered in this study. For
example, if one imposes an angular timestep of 0.27 together with
40 iterations per timestep, the residuals were not able to become
lower than 103. As a result, the pressure-based formulation was
selected for the simulations.
The ideal gas law was enabled for the required material prop-
erties; a sensitivity analysis was indeed carried out also on the
proper choice between the formulations for incompressible or
compressible ow: the resulting elds of density, temperature and
Mach were analyzed. The results demonstrated the absence of
appreciable compressibility effects, although a slight over-
estimation of the power coefcient was observed. The convergence
rate was not, however, speeded up by the simpler formulation, i.e.
the incompressible law. Therefore, the more complex and accurate
model, i.e. the compressible law, was preferred since it was Fig. 3. Sensitivity analysis @ TSR 1.1: instantaneous torque coefcient for PISO al-
assumed to guarantee a higher degree of detail without additional gorithm vs. nal settings.
F. Balduzzi et al. / Renewable Energy 85 (2016) 419e435 425
Fig. 4. Sensitivity analysis @ TSR 2.2: instantaneous torque coefcient for SIMPLE Fig. 5. Sensitivity analysis @ TSR 1.1: instantaneous torque coefcient for Coupled
and Coupled algorithms with azimuthal increments of 0.9 and 0.27. algorithm with 20, 30 and 40 iterations per timestep and azimuthal increment of 0.27.
Case A B C D E
4.4. Boundaries
L/D (L1 L2)/D 15 30 60 90 140
The effects of the domain extent on the torque output were W/D 10 20 40 60 60
Aspect Ratio 1.5 2.33
nally analyzed. The goal was to place the boundaries at a distance
426 F. Balduzzi et al. / Renewable Energy 85 (2016) 419e435
Once the main simulation settings have been assessed, the right
choice of both the spatial and the temporal discretizations becomes
the key point for a successful simulation.
In particular, while in common RANS simulations the attention
is primarily focused on ensuring a sufciently rened mesh (spatial
discretization), in the present case the sudden variation of the ow
conditions on the airfoil during the revolution imposes a specic
analysis of the temporal discretization. More specically, the
azimuthal increment between two subsequent steps of analysis
must be small enough to correctly describe every ow structure
(vortices, wakes, etc.) that is originated in the ow eld; otherwise,
signicant errors could be introduced in the predicted torque
prole over a revolution.
Moreover, it is worth remarking that a strong mutual inuence
Fig. 6. Boundary inuence @ TSR 2.2: power coefcient as a function of the domain is established between the temporal and spatial characteristic
length. scales; in order to accurately describe a structure, e.g. a dynamic-
stall vortex, it is in fact important both a ne mesh, to capture
the gradients, and a very small advance of the rotating frame, to
avoid any undesired discontinuity of the variables between two
instants.
On these bases, a multivariate sensitivity analysis was carried
out accounting for the mesh features and the timestep. In addition,
a check on the number of inner iterations needed for each timestep
was constantly ensured, conrming, however, that 30 iterations
were generally sufcient to a stable solution also in the most critical
functioning points.
Table 4
Mesh sensitivity analysis.
Reference Number of elements (NE) Boundary Number of nodes on the airfoil (NN) Number of quads rows (NBL) Element sizing [mm]
name Layer
Rotating domain Stationary domain yp Sliding
(NER) (NES) interface
Fig. 10. Sensitivity analysis @ TSR 1.1: vorticity contours with G4 and G6 meshes and azimuthal increments of 0.27 and 0.135 .
Fig. 13. Sensitivity analysis @ TSR 2.2: gauge pressure contours with G3 and G4 meshes and an azimuthal increment of 0.27.
Dt
Fig. 14. Sensitivity analysis @ TSR 2.2: power coefcient as a function of the global
Co V (1)
Dx
cells number.
430 F. Balduzzi et al. / Renewable Energy 85 (2016) 419e435
Based on its formulation, the number expresses the ratio be- discretization in the orthogonal direction from the blade wall.
tween the temporal timestep (Dt) and the time required by a uid Fig. 17 conversely shows the average Courant Number within
particle moving with V velocity to be convected throughout a cell of the rst region (<1 mm from the blade wall, i.e. the most critical
dimension Dx. While in case of explicit schemes for temporal dis- one) over a revolution with G4, G5 and G6 meshes and angular
cretization the Courant-Friedrichs-Lewy (CFL) criterion imposes a timesteps of 0.135 and 0.27.
limit on the maximum allowed value of Co (i.e. Co < 1 [51] [53], and All the analyzed congurations respected the general prescrip-
[54]) to ensure the stability of the calculation, implicit methods are tion of Co < 10. If very ne spatial discretization are used (e.g. G6),
thought to be unconditionally stable with respect to the timestep the timestep must be likewise reduced to contain the Courant
size ([45] and [51]). Number, whereas, as expected, the coarsest mesh (G4) is that
Although theoretically valid if the problem is studied with a ensuring the highest degree of stability (minimum Co).
linear stability analysis, when the timestep is increased non- Based on the above the following observations were made:
linearity effects would become prominent and oscillatory solu-
tions may occur. On these bases, the literature indicates that an The G4 mesh is a very suitable choice for the present study case
operational Co between 5 and 10 for viscous turbomachinery ows, as it offers sufcient accuracy (see Figs. 9, 12 and 15), an
solved with an implicit scheme, provides the best error damping affordable computational effort and good stability (Fig. 17).
properties ([53] and [54]). This solution must be, however, coupled with a proper man-
Recently, an interesting analysis on the application of the CFL agement of the timestep as a function of the revolution speed of
criterion to unsteady simulations of Darrieus turbines was pre- the rotor: in detail, when the revolution regime decreases, the
sented by Ref. [24], which focused the attention on the rotating grid azimuthal increment between two subsequent steps must be
interface; in particular, the satisfaction of an upper bound of 0.15 for proportionally reduced to ensure both a good accuracy (see
the Courant Number in the interface cells is assumed to dene the Figs. 9, 12 and 17) and a proper stability of the calculation.
maximum required angular timestep in order to obtain accurate Only in case of very low revolution speeds, were the airfoils
results. experience very unstable functioning conditions, the use of ner
In the present study, a specic analysis was instead carried out meshes can be considered in order to enhance the accuracy of
on the Courant Number conditions in proximity of the blades, as a the simulation (Fig. 11).
correct description of the ow in these zones was in fact deemed to
be the most restrictive requisite to accurately predict the torque Within these assumptions, the analysis was further focused on a
output; a verication on the conditions at the interface was how- more in-depth evaluation of the results obtained with the G4 mesh.
ever made, obtaining Co values very close to the limit suggested by In particular, a global check on the suitability of the dimensionless
Ref. [24]. parameters was carried out for the three reference revolution re-
First, a check on both the y and the Courant Number was made gimes of TSR 1.1, TSR 2.2, TSR 3.3 and their corresponding
on all the analyzed cases (in terms of mesh, timestep and revolution angular timesteps. In particular, Fig. 18 and Fig. 19 report the
speed). The y (average and maximum value) was calculated on the average and maximum values of the y as a function of the
airfoil surface, whereas the Co (average and maximum value) was azimuthal position, respectively.
evaluated in three different zones around the blades, compre- Upon examination of the results, it is readily noticeable that the
hended within a distance from the wall of 1, 5 or 10 mm, mesh properties in the near-wall region were correctly dened
respectively. since the average values of y were all around the limit of 1 that
For conciseness reasons, only selected results are reported here. represented the target of the simulations, although a small
Analogous to the mesh sensitivity analysis, specic attention was (acceptable) increase is observed, due to the increase in the relative
paid to the TSR 1.1 regime, which showed the most critical ow velocity, hence the wall shear stress. Moreover the same in-
behavior in terms of ow conditions of the airfoils. In particular, crease was not observed for the maximum values of y, ensuring
Fig. 16 reports the average values of the y for the analyzed meshes: similar values (in the order of 3.5) for all regimes, which were
in this case, the timestep is arbitrary as it denitely does not affect registered only in very few cells of the domain.
the calculated ow velocity in the boundary layer. Upon examina- At the same time, Fig. 20 and Fig. 21 report the average and
tion of Fig. 16, it is readily noticeable that the prescribed limit was maximum values of the Courant Number over a revolution,
always satised, indicating that all meshes had a suitable spatial
Fig. 17. Average Courant Number (1 mm region around blade prole) over a revolution
Fig. 16. Average y values over a revolution @ TSR 1.1 with all the analyzed meshes. @ TSR 1.1 with G4, G5 and G6 meshes and angular timesteps of 0.135 and 0.27.
F. Balduzzi et al. / Renewable Energy 85 (2016) 419e435 431
Fig. 18. Average y values over a revolution with the G4 mesh at different revolution Fig. 21. Maximum Courant Number (1 mm region around blade prole) over a revo-
regimes. lution with the G4 mesh at different revolution regimes.
Fig. 22. Comparison between simulations and experiments: power coefcient vs. TSR. Fig. 23. Sensitivity analysis on non-optimized numerical setting: power coefcient vs.
TSR.
0.65 m from the tunnel outlet [57]. From experience of the present study, the following indications
Based on the above, the numerical set-up of simulations was are proposed:
slightly modied to account for the jet effect connected to the
specic layout of the tunnel. In detail, the inlet of the domain was The pressure-based solver using a coupled algorithm for a
limited to a section of 1.82 m and get near to the turbine at the compressible ow is thought to represent the best numerical
prescribed distance of 0.65 m. Both the lateral and the downwind approach, if combined with the 2nd order upwind scheme for
boundaries were instead maintained at distances of 60D and 100D, the spatial discretization of all the transport equations and the
respectively. bounded 2nd order implicit for the time differencing.
In order to assess the proper spatial and temporal discretiza- The k-u SST turbulence model is the most suitable choice for the
tions, a preliminary analysis of the rotor was carried out with a simulation of Darrieus turbines functioning, which contempo-
proprietary BEM code [58,59], as no information on the power rarily involves the presence of both boundary layer separation
coefcient curve was provided in the reference. The code revealed and free-shear ows.
that the proposed rotor is credited of a very poor performance at A convergence criterion based on the variation of the torque
the considered wind speed (i.e. U 3.2 m/s) with an almost null output can be a valuable solution but only on condition that a
power coefcient. In these conditions, despite the TSR equal to 3.0, very strict threshold is imposed. In detail, the convergence
the rotor was simulated as if working in an unstable regime. In history of the simulations revealed that, although a torque
further detail, the two NACA0018 airfoils were discretized with a variation lower than 1% (i.e. the value generally adopted in the
mesh comparable to the original G6, i.e. approximately 1800 nodes literature) between two subsequent revolutions is generally
on the airfoil, 9.0 105 elements in the rotating region and y~1. achieved after only few cycles, the torque value measured after
Moreover, in order to ensure a correct evaluation of ow separation this threshold can be sensibly different (up to approximately
and vortices propagation, a small timestep was preferred, equal to 10%) from the nal value at convergence. A convergence crite-
0.15 . rion is then proposed using a limit on the torque variation lower
Fig. 24 reports the comparison between simulations and ex- than 0.1%.
periments in terms of instantaneous tangential force coefcient The domain dimensions to ensure an open-eld-like behavior of
(Ctang) along revolution for a declared revolution speed of 300 rpm the rotor must be extended signicantly with respect to what is
and a TSR 3.0. generally suggested in the literature. In the present case, the
Despite the uncertainties connected to the experiments and the results insensitivity to boundaries' distance was obtained by
highly complex functioning point (average Ctang almost zero), the placing the inlet section 40 rotor's diameters upwind the center
agreement is impressively good, conrming that the proposed of the turbine, the outlet section 100 diameters downwind and
computational approach can effectively reproduce the physics of each lateral boundary 30 diameters away. Moreover, to avoid
Darrieus wind turbines, opening interesting prospects for future undesired disturbances generated at the sliding interface, the
analyses based on CFD. rotating region had a double diameter with respect to the
turbine.
7. Conclusions A specic mesh region for boundary layer resolution can provide
notable advantages in terms of accuracy, on condition that y
In the present study, an extended analysis was carried out to values in the order of 1 are guaranteed. This simulation cases
highlight some critical issues for an accurate two-dimensional CFD generally require very heavy meshes, particularly due to the
simulation of Darrieus wind turbines. high number of nodes needed on the airfoils' surface. The
All the choices made in the study concerning each of the main renement level must be further increased in case of low tip-
numerical parameters of the model were based on specic speed ratios of the rotor, where the large incidence angle vari-
comparative analyses, which assessed the inuence of the param- ations make a correct localization of the stall point on the pro-
eter itself both on the solution stability and on the accuracy with les even more important.
respect to purposefully collected experimental data on the study The study showed that the required temporal discretization is
turbine. Moreover, a literature case study also conrmed the suit- almost independent on the revolution regime considered for the
ability of the proposed numerical approach also for the prediction simulation. In detail, a constant timestep was identied which
of the instantaneous torque prole of a Darrieus turbine. allowed to contain both the average and the maximum Courant
Number in proximity of the blades. As a result, the angular
timestep becomes directly proportional to the revolution speed
of the rotor. In the present case study, angular timestep in the
range between 0.135 and 0.405 were used. Small angular
timesteps are also suggested in case of low wind speeds were
the torque output of the rotor is reduced and large separate
regions occur.
Acknowledgments
Power Generation Systems with CFD Computations, in: Proceedings of Tunnel, J. Energy Resour. Technol. 133 (3) (2011), 031201e031201.
OCEANS, Kobe, Japan, April 8-11, 2008. [44] P.G. Migliore, W.P. Wolfe, J.B. Fanucci, Flow Curvature Effects on Darrieus
[33] M. Raciti Castelli, A. Dal Monte, M. Quaresimin, E. Benini, Numerical evalua- Turbine Blade Aerodynamics, J. Energy 4 (2) (1980) 49e55.
tion of aerodynamic and inertial contributions to Darrieus wind turbine blade [45] Ansys, Inc, Fluent Theory Guide, 2013 release 14.5.
deformation, Renewable Energy 51 (2012) 101e112. [46] D.C. Wilcox, Turbulence Modeling for CFD, DCW, Industries Inc, La Canada
[34] C. Jian, J. Kumbernuss, H.X. Yang, X.Y. Wang, Inuences of helix blade and (CA), 1998.
unsymmetrical airfoil on theperformance of the Darrieus wind rotor, in: [47] F.R. Menter, M. Kuntz, R. Langtry, Ten Years of Industrial Experience with the
Proceedings of the International Conference on Applied Energy, Perugia, Italy, SST Turbulence Model, Turbulence, Heat Mass Transfer 4 (2003) 625e632.
May 16-18, 2011. [48] F.R. Menter, Two-Equation Eddy-Viscosity Turbulence Models for Engineering
[35] R. Gupta, A. Biswas, Computational uid dynamics analysis of a twisted three- Applications, AIAA J. 32 8 (1994) 1598e1605.
bladed H-Darrieus rotor, J. Renewable Sustainable Energy 2 4 (2010) [49] J. Blazek, Computational Fluid Dynamics: Principles and Applications, Elsevier
043111e043126. Science, Oxford (UK), 2001.
[36] R. Gupta, S. Roy, A. Biswas, Computational uid dynamics analysis of a twisted [50] H.K. Versteeg, W. Malalasekera, An Introduction to Computational Fluid Dy-
airfoil shaped two-bladed H-Darrieus rotor made from breglass reinforced namics. The Finite Volume Method, Longman, UK, 1995.
plastic (FRP), Int. J. Energy Environ. 1 6 (2010) 953e968. [51] J. Ferziger, M. Peric, in: Computational Methods for Fluid Dynamics, 3rd rev,
[37] C. Kaminsky, A. Filush, P. Kasprzak, W. Mokhtar, A CFD Study of Wind Turbine Springer, Berlin, 2002.
Aerodynamics, in: Proceedings of the 2012 ASEE North Central Section Con- [52] R. Peyret, Handbook of Computational Fluid Mechanics, Academic Press Inc.,
ference, Ohio Northern University, March 24, 2012. Oxford, 1996.
[38] G. Colley, R. Mishra, H.V. Rao, R. Woolhead, Computational ow eld analysis [53] R.S. Amano, B. Sunden, Computational Fluid Dynamics and Heat Transfer:
of a Vertical Axis Wind Turbine, in: Proceedings of the International Confer- Emerging Topics, WIT Press, Southampton, UK, 2011.
ence on Renewable Energies and Power Quality, Las Palmas de Gran Canaria, [54] C.J. Hearn, The Dynamics of Coastal Models, Cambridge University Press,
Spain, April 13-15, 2011. Cambridge, UK, 2008.
[39] P. Sabaeifard, H. Razzaghi, A. Forouzandeh, Determination of Vertical Axis [55] F. Balduzzi, A. Bianchini, F.A. Gigante, G. Ferrara, M.S. Campobasso, L. Ferrari,
Wind Turbines Optimal Conguration through CFD Simulations, in: Pro- Parametric and Comparative Assessment of Navier-Stokes CFD Methodologies
ceedings of the International Conference on Future Environment and Energy, for Darrieus Wind Turbine Performance Analysis, in: Proceedings of the ASME
Singapore, 2012. Turbo Expo 2015, Montreal, Canada, June 15-19, 2015.
[40] A. Bianchini, L. Ferrari, S. Magnani, On the effects of a skewed ow on the [56] P. Vittecoq, A. Laneville, The aerodynamic forces for a Darrieus rotor with
performance of a three-bladed H-Darrieus turbine: experimental and theo- straight blades: Wind tunnel measurements, J. Wind Eng. Ind. Aerodynamics
retical analyses, in: Proceedings of the International Conference on Applied 15 (1e3) (1983) 381e388.
Energy (ICAE) 2012, Suzhou (China), July 5-8, 2012. [57] I. Paraschivoiu, Predicted and Experimental Aerodynamic Forces on the Dar-
[41] F. Balduzzi, A. Bianchini, G. Ferrara, L. Ferrari, R. Maleci, Blade design criteria to rieus Rotor, J. Energy 7 (6) (1983) 610e615.
compensate the ow curvature effects in H-Darrieus wind turbines, [58] A. Bianchini, L. Ferrari, E.A. Carnevale, A model to account for the Virtual
J. Turbomach. 137 (1) (2015) 1e10. Camber Effect in the Performance Prediction of an H-Darrieus VAWT Using
[42] K. Cooper, Bluff-body blockage corrections in closed- an open-test-section the Momentum Models, Wind Eng. 35 (4) (2011) 465e482.
wind tunnels, Wind Tunnel Wall Correction, in: B.F.R. Ewald (Ed.), AGARD- [59] A. Bianchini, L. Ferrari, S. Magnani, Start-up behavior of a three-bladed h-
AG-336, RTO/NATO, Brussels (Belgium), 1998. Darrieus VAWT: experimental and numerical analysis, in: Proceedings of the
[43] L. Battisti, V. Dossena, G. Persico, L. Zanne, S. DellAnna, B. Paradiso, Aero- ASME Turbo Expo 2011, Vancouver, Canada, June 6-10, 2011.
dynamic Measurements on a Vertical Axis Wind Turbine in a Large Scale Wind