This article summarizes the results of the “Blind test 5”
workshop, which was held in
Visby, Sweden, in May 2017. This study compares the numerical predictions of
the wake flow behind a model wind turbine operated in yaw to experimental
wind tunnel results. Prior to the workshop, research groups were invited to
predict the turbine performance and wake flow properties using computational
fluid dynamics (CFD) methods. For this purpose, the power, thrust, and yaw
moments for a 30

Four participants submitted predictions using different flow solvers, three of which were based on large eddy simulations (LES) while another one used an improved delayed detached eddy simulation (IDDES) model. The performance of a single yawed turbine was fairly well predicted by all simulations, both in the first and third test cases. The scatter in the downstream turbine performance predictions in the second test case, however, was found to be significantly larger. The complex asymmetric shape of the mean streamwise and vertical velocities was generally well predicted by all the simulations for all test cases. The largest improvement with respect to previous blind tests is the good prediction of the levels of TKE in the wake, even for the complex case of yaw misalignment. These very promising results confirm the mature development stage of LES/DES simulations for wind turbine wake modeling, while competitive advantages might be obtained by faster computational methods.

Wind turbine wake interaction has become a major topic in wind
energy research during the last decades. The power drop between the first and
second turbine can be up to 35 % in an offshore installation, when the
turbines are aligned with the wind direction, while the averaged losses due
to wake interactions are estimated to range between 10 % and 20 %

During the last years CFD models were constantly improved, both by increasing
their accuracy and by reducing computational costs. In order to give the
model developers the possibility to test their CFD models in a complex wake
flow, a fifth blind test was initiated, challenging the modelers with the
dynamic flow situation of a yawed wind turbine. The wakes behind two
different turbines and two inline turbines were investigated. Yaw
misalignment is currently a widely discussed topic in wind energy research.
Intentional yaw misalignment of an upstream turbine in a wind farm is deemed
to have a large potential for increasing the farm's efficiency

The work is organized as follows. Section

In this blind test experiment three different turbine geometries were used.
For the purpose of yaw experiments, a new turbine test rig was constructed at
NTNU, which is called the Laterally Angled Rotating System 1 (LARS1). It
features a shorter nacelle and slimmer tower compared to the turbines used in
previous blind tests in order to minimize the effects on the wake, as shown
in Fig.

NTNU's model wind turbine called T2 was already used in previous blind test
experiments. The sketch in Fig.

The third turbine used in this blind test is the model wind turbine designed
by ForWind at the University of Oldenburg. For the experiments in the NTNU
wind tunnel, the turbine's hub height was increased with four cylindrical
rods, in order to be operated at a height, comparable to the NTNU turbines.
The turbine has a smaller rotor diameter of

Sketches of the model wind turbines with reference coordinate
system,

The NTNU and ForWind rotors are based on two different airfoils. The NREL
S826 airfoil, which is used from root to tip for the NTNU rotor, was
originally designed for application in the tip region of full-scale wind
turbines, a detailed description can be found in

The ForWind rotor is based on the SD7003 airfoil that is defined in detail in

All the experimental data were measured in the closed-loop wind tunnel at the
Department of Energy and Process Engineering at NTNU in Trondheim. The wind
tunnel has a test section length of 11.5 m, a width of 2.7 m, and a height
of 1.8 m. The reference coordinate system is pictured in
Fig.

Reference coordinate system in the wind tunnel and definition of
positive yaw angle

For all test cases a nonuniform shear flow was generated by a grid at the
inlet of the test section. The grid is built from wooden bars with a cross
section of

As the velocities of the shear profile vary in height and are nonuniform over
the rotor area, the reference wind speed

The turbulence intensity (TI) of the inflow is shown in
Fig.

Vertical flow profiles in the empty wind tunnel at different
positions, in which

In this blind test experiment the modelers were asked to simulate three test
cases. In test case 1 the flow

Summary of the parameters that are varied for the three investigated
test cases,

The

The thrust force and yaw moments acting on the upstream and downstream turbine were measured separately using a Schencker six-component force balance, which was installed under the wind tunnel floor. The balance also served as a turning table allowing an exact adjustment of the yaw angle. For the rotor thrust only the load cell parallel to the flow was taken into account. The yaw moment was calculated from a moment equilibrium of three measured forces in the horizontal plane (referenced to the rotor center).

The aerodynamic power

The experimentally measured values feature several uncertainties. The
statistical uncertainties in every sample of the mean velocity, power,
thrust, and yaw moments are calculated based on a 95 % confidence level
according to the procedure described in

Siemens PLM Software from the United Kingdom (Siemens), the Department of
Mechanical Engineering of the Politecnico di Milano in Italy (POLIMI), the
Facultad de Ingeniería of the Universidad de la República in Uruguay
(UdelaR), and KTH Mechanics Department from the Royal Institute of Technology
in Sweden (KTH) participated in the blind test and submitted computational
results. For clarity, only the abbreviations will be used in the following. A
summary of the simulation methods and mesh properties is presented in
Table

Overview of simulation methods and parameters. Abbreviations: improved delayed detached eddy simulation (IDDES), large eddy simulation (LES), actuator line (ACL), and fully resolved (FR).

Siemens, who previously participated in blind test experiments as CD-adapco,
used the finite volume code STAR-CCM+ v12.04 to mesh and solve all three test
cases. Each simulation resolved the rotor, nacelle, and tower structure
completely, and used the hybrid method improved delayed detached eddy
simulation (IDDES), which resolves the energy-carrying eddies in the free
stream and solves the boundary layer flow with RANS. The Spalart–Allmaras
model was used for closure of the turbulence equations, and the fluid was
considered incompressible. Convective fluxes used a MUSCL third-order scheme
(monotonic upwind scheme for conservation laws), while time was discretized using a second-order implicit
scheme. Each set of blades and hub was contained inside a cylindrical,
rotating volume which was meshed with polyhedral cells, whereas the main
domain used trimmed cells, resulting in a hexahedral dominant mesh in which a
small proportion of cells was trimmed near the boundaries. Due to the
rotation of the cylindrical volumes, the mesh was not conformal at the
interface between the two regions, and flow quantities were interpolated from
one volume to another. All wall surfaces, including the wind turbine bodies
and the wind tunnel walls, were covered in several layers of prismatic cells
to improve the resolution of boundary layers. The resulting

While a rigorous mesh dependency study was not performed, the mesh sizes were
based on previous experience and expected to perform well with an affordable
amount of cells. All simulations were run with a time step of

As inflow the given analytical mean velocity profile

POLIMI submitted a LES that was computed using the ALEVM code. It is an
aerodynamic turbine simulation tool written in C++ and based on pisoFoam,
which is an incompressible transient solver included in the OpenFOAM
framework. The standard PISO (Pressure-Implicit with Splitting of Operators)
solver was modified to include the effect of the turbine blades that are
represented using the lifting line approach. The blade lines are discretized
in segments based on the intersections with the numerical mesh grid, in which
an actuation point acts on each segment. Each point of the actuator line
(ACL) acts as an isolated blade section. More information about the ACL
method can be found in

ALEVM employs the well-known solution of the regularization kernel, smearing
the line forces on the multiple cells following a Gaussian distribution and
thus avoiding abrupt variation in the source term strength between adjacent
cells. The turbulence in the wake region is modeled using a LES, adopting the
Smagorinsky subgrid-scale model. For the time discretization scheme a
first-order implicit approximation is used, while the divergence
discretization scheme and the gradient discretization scheme are approximated
by second order. The simulation is run for a time interval of 20 s, while a
time step of

UdelaR submitted another LES using their in-house developed caffa3d code. It
is an open-source, finite volume code, with second-order accuracy in space
and time, and parallelized with a message passing interface (MPI), in which
the domain is divided into unstructured blocks of structured grids. Complex
geometries are represented by a combination of body-fitted grids and the
immersed boundary method over both Cartesian and body-fitted grid blocks. The
code is F90 and currently runs on CPU, although a CUDA GPU version is
currently being developed. The properties of the geometry and the flow are
expressed as primitive variables in a Cartesian coordinate system, using a
collocated arrangement. An ACL approach is used to discretize the turbine
blades in the simulations. The aerodynamic forces on the blade elements are
computed using the provided XFoil data, and dynamic stall effects are not
considered. The forces, then, are projected onto the computational domain. In
order to compute the additional source term, a Gaussian smearing function is
used, taking into account one smearing factor for each direction: normal,
tangential, and radial to the rotor plane. The domain, representing the wind
tunnel (

A third LES was submitted by KTH. The spectral element code Nek5000

The modelers were asked to predict the power coefficients

The modelers were asked to provide predictions of the velocities and TKE in
full wake planes in the ranges

Measurement grid in the wake consisting of 357 points, the blue
tower and nacelle represents the NTNU turbine LARS1, the green tower and
nacelle represents the ForWind turbine, the dashed line corresponds to the
projection of the rotor diameter

Two-dimensional wake contours are difficult to compare quantitatively as they cannot be plotted in the same diagram. However, they provide valuable insight into the shape and position of the wake. Therefore, the wake shapes are in a first iteration compared qualitatively. To obtain quantitative measures of comparison, different methods to compute the wake position, the energy content in the wake, and the magnitudes of the wake parameters are applied. These are described below.

In order to quantify the wake deflection, a method approximating the
available power is used, which was previously described by

From the statistical error measures proposed by

The results of

The blade element momentum (BEM) tool Ashes

The thrust coefficients

Power coefficient

Numerical values of power coefficient

Figure

Next, Fig.

The normalized TKE

The comparisons of

Comparison parameters: skew angle (

In test case 2 an aligned turbine array with both NTNU turbines LARS1 and T2
is investigated. The upstream turbine LARS1 is operated at

Power coefficient

This section discusses the wake characteristics

The contours of the vertical velocity component

The TKE

Comparison parameters: skew angle (

In the third test case the wake behind the yawed ForWind turbine is
investigated. It was simulated by three of the modelers, while POLIMI did not
submit predictions for this test case. The contours of the streamwise
velocity

The contours of the normalized vertical velocity

The TKE contours presented in Fig.

Comparison parameters: skew angle (

The comparison of the wake characteristics

The results of four different computational contributions were compared to experimental wind tunnel results in this blind test experiment. The modelers submitted predictions for the performance of two single yawed turbine models and an aligned turbine array where only the upstream turbine is yawed. Furthermore, they predicted the mean and turbulent wake flow behind two different model turbines and the turbine array.

The power of a single yawed turbine

The predictions of the thrust coefficients

When comparing CFD predictions to experimental measurements it is important
to quantify the differences. Therefore, different techniques have been
applied to analyze the wake properties. The statistical methods NMSE and

The comparison of the mean streamwise velocity

Furthermore, the good results of the simulations based on a lower cell count indicate a new trend towards CFD codes that are able to perform accurate wake flow predictions at significantly lower computational cost. This becomes especially important for wake predictions of full-scale turbines in which the dimensions and Reynolds numbers exceed those of the experiments. Consequently, simulations with a fine grid may be very hard to realize in such a case. Nevertheless, the good performance of the coarse-grid simulations in the blind test shows that they are a promising tool for full-scale wake predictions.

Overall, the results of this blind test comparison confirm a continuous improvement in performance and wake flow predictions from Blind test 1 to Blind test 5. LES-ACL approaches as well as the hybrid IDDES technique were confirmed to be able to perform accurate predictions, also for complex setups featuring highly unsteady flow in yawed and partial wake operation.

All presented wake data in this paper is available at

FM, JS and JB planned the carried out the experiments. RF and SE carried out the simulations indicated as “Siemens”. LB and PS carried out the simulations indicated as “Polimi”. MD and AG carried out the simulations indicated as “UdelaR”. EK and DH carried out the simulations indicated as “KTH”. MH, JP, MA and LS initiated and supervised the project. The workshop was initiated and prepared by JB, FM, JS and LS. The data was post-processed and compared by FM. The manuscript was written by FM, and revised by JS and JB. The description of the simulation methods was provided by Siemens, Polimi, UdelaR and KTH. All authors provided critical feedback and contributed to the shape of the manuscript.

The authors declare that they have no conflict of interest.

The authors would like to thank Stefan Ivanell and the staff of the Wind Energy group from Uppsala University and Campus Gotland for providing the venue for the workshop. Edited by: Alessandro Bianchini Reviewed by: Gerard Schepers and one anonymous referee