Navier-Stokes Simulation of Wind Turbine Consider Laminar-Turbulent Transition

Effects
Bum-suk KIM, Korean Resisters of Shipping, bskim@krs.co.kr
Hark-jin EUM, Korean Resisters of Shipping, hjeum@krs.co.kr
Jong-bum WON, Korean Resister of Shipping, jbwon@krs.co.kr
Mann-eung KIM, Korean Resister of Shipping, mekim@krs.co.kr
Key words: CFD, transitional turbulence model, performance, flap-wise bending
moment, Phase IV wind turbine
Abstract
Despite of the laminar-turbulent transition region co-exist with fully turbulence region
around the leading edge of an airfoil, still lots of researchers apply to fully turbulence
models to predict aerodynamic characteristics. It is well known that fully turbulent
model such as standard k- model, couldn't predict the complex stall phenomenon
and the separation behavior on an airfoil accurately, it usually leads to over prediction
of the aerodynamic characteristics such as lift and drag forces. So, we apply
correlation based transition model to predict aerodynamic performance of the NREL
(National Renewable Energy Laboratory) Phase IV wind turbine. Simulated results by
applying the transition model compared with the experimental results. It compared
with the fully turbulence results also. Several post processing results such as low
speed shaft torque, flap-wise root bending moment and aerodynamic coefficients of
2D airfoil included in this study. As a result, the low speed shaft torque predicted by
use of the transitional turbulence model is very good agree with the experimental
result in whole operating conditions but fully turbulent model, standard k-, over
predict the shaft torque after 7m/s. Flap-wise root bending moment is also good
agreement between the numerical simulation and the experimentation for most of the
operating conditions, especially with the transition model.
Introduction
In order to enhance the efficiency and performance of blades for wind turbine,
which require aerodynamically optimal design, first we need to understand
clearly the characteristics of flow field. For extracting the factors of optimal
design, we should obtain extensive data on flow characteristic and performance
characteristic from reliable experiments, but such experiments cost a prohibitive
amount of time and money. Accordingly, for more efficient examination of the

flow and performance characteristics of blades, various numerical analyses are

being conducted using wake code, aero-elastic code, performance code, CFD
code, etc.
In terms of the reliability of the results of analyzing the aerodynamic
characteristic of airfoil through CFD, the standard k- turbulence model, which
is adopted by most of general CFD codes, is well known to overestimate lift
force because there happen problems such as the inaccurate estimation of
separation points and stall area in case the angle of attack of airfoil is
maintained larger than the stall angle. This is caused by failure to estimate
boundary layer separation in the viscosity sub-layer. Thus, in order to improve
the estimation of wall shear stress in the region of viscosity sub-layer, new
models have been developed and applied including the Wilcox model,
BSL(baseline model) model that takes the advantages of Wilcox model, and
SST(Shear Stress Transport) model that calculates the transport term of
turbulent shear stress. In particular, the SST model is considered to estimate
the size of vortex caused by adverse pressure gradient and separation points
relatively accurately [1].
The flow field on an airfoil has laminar-transition-turbulent areas simultaneously,
and shows the characteristic that the transition point moves toward the leading
edge with the increase of the angle of attack. However, the results of analysis
by applying the fully turbulence model have a problem in that flow field is
assumed to be fully turbulent.
In general, flow separation in the laminar boundary layer is faster than that in
the turbulent boundary layer, and this phenomenon is an important reason that
causes difference between analysis results using a fully turbulent model without
considering laminar-transition effects and actual experiment results. In particular,
it can have a significant effect on the estimation of the performance of wind
turbine blades made of airfoil bundles. Accordingly, in order to get more
accurate aerodynamic characteristic analysis results, we need to consider the
application of a sophisticated turbulence model that can accurately estimate
laminar separation bubble and turbulent reattachment phenomenon.
Numerical simulation
The present study chose the Phase IV wind turbine blade, with which we can
implement accurate configuration data and obtain reliable experiment results
relatively easily, as an object of numerical simulation.
Phase IV wind turbine generates 20kW rated power, and is of stall controlled
type. The diameter of the blade is 10.058m, rotation speed is 71.63rpm, and

two blades are used. The blades of Phase IV wind turbine are made of S809
airfoil. It has a thickness of 21%, and was designed to be relatively less
sensitive to the surface roughness of the leading edge for the stall regulated
type wind turbines [2].
Figure 1 shows the configuration of S809 airfoil, and Figure 2 shows the
configuration of the Phase IV wind turbine blades.
This study is divided into the analysis of 2D aerodynamic characteristics of
S809 airfoil and the analysis of the 3D performance characteristics of Phase IV
wind turbine. We applied the correlation-based transition [3], SST, RNG k- and
standard k- models in order to test the accuracy of S809 airfoil aerodynamic
characteristic analysis according to turbulence model. For comparison with the
results of CFD analysis, we used the results of experiment, performed in the
low speed wind tunnel at the Delft University of Technology [4]. The angle of
attack of airfoil was changed from 0 to 1.02, 5.22, 9.22, 14.24 and 20.15
and Re=2106.
In Phase IV wind turbine analysis, we selected the correlation-based transition
model and the standard k- model, which is widely applied for getting results at
an engineering level and is adopted as the basic turbulence model in a general
CFD codes, and compared their results. Inlet wind velocity was set to 7m/s,
10m/s, 13m/s, 15m/s, 20m/s and 25m/s, and the results of CFD analysis were
compared with the results of an experiment conducted in the NASA Ames wind
tunnel by NREL.
2D airfoil analysis and 3D phase IV wind turbine analysis were performed with
ANSYS-CFX V11.0 using a 2nd order high resolution scheme. For the
determination of convergence, RMS residual was set to 10-5, and in S809 airfoil
analysis calculation was performed sufficiently up to the point where the value
did not change any more while monitoring lift and drag forces.
For precise analysis of the laminar-transition boundary layers through the
application of the transition turbulence model, we can get reliable analysis
results when the spatial resolution of grid around the near-wall satisfies the
condition of y+ < 1. In case of the k- model, however, we need to satisfy the
condition of y+ > 30. Accordingly, this study considered the near-wall grid
density reflecting best the characteristic of the applied turbulence model in the
stage of computational grid generation, and satisfied the condition of y+ < 1 for
the transition model and y+ > 30 for the k- model [5]. Computational grid was
generated using ICEM-CFD HEXA, a grid generation tool, and it was composed
of a complete O-type HEXA grid system with 153,600 nodes for 2D airfoil and
with about 7,500,000 nodes for Phase IV simulations.
The wall was treated to be no-slip condition, and a constant wind velocity was

applied as an inlet condition. The wind velocity at the inlet was set to satisfy
the condition of Re=2106 in S809 analysis, and 6 wind velocity conditions
from 7m/s to 25.06m/s were defined in 3D wind turbine analysis. Inlet
turbulence intensity was set to 0.2%.
Figure 3 and Figure 4 show the computational grid of 2D airfoil and 3D wind
turbine respectively.

Results and Discussions

Generally in case of airfoil, 3D vortex breakdown takes place with the increase
of AOA on the suction side. The 3D stall phenomenon affects the aerodynamic
characteristic of airfoil, and therefore, the estimation of aerodynamic
characteristic in the post-stall area through 2D numerical analysis estimates
results somewhat different from the experiment result.
Figure 5 compared lift coefficient between experiment results and CFD results.
Until the point of 9.22, the area before the occurrence of stall, the result of
transition model analysis fairly agrees with the experiment result, but from
14.24 they show some difference due to the occurrence of complex 3d flows
on the suction side.
Figure 6 compared drag coefficient. As in Figure 5, in the area before the
occurrence of stall, the result of transition model analysis exactly agrees with
the experiment result, but in the area after the stall, it shows difference from
the experiment result. As a whole, compared to that by fully turbulent models,
the result of estimation by the transition model agrees better with the
experiment result, and particularly in the area before the occurrence of stall, it
agrees very accurately with the experiment result.
Figure 7 compared pitching moment coefficient. The result of transition model
analysis fairly agrees with the experiment result except at 20.15 where
complete stall occurs.
CFD analysis results were close to the experiment results in order of transition,
SST, RNG k- and standard k- model, and in particular, k- type turbulence
models could not estimate the accurate aerodynamic characteristics. Thus, it is
essentially required to apply the transition model to analysis in order to
estimate aerodynamic characteristic accurately.
Figure 8 is the results of applying the transitional turbulence model, showing
turbulent intermittency on the suction side of blades according to inlet wind
velocity in order of 7m/s~20.13m/s. In the results, red color means the turbulent
area and blue color the laminar area, and the result of 25.06m/s was almost

identical with that of 20.13m/s corresponding to full turbulence, so it was not

shown. In case of the result of 7m/s, a relatively wide laminar area exists
around the tip and the hub. Around the tip, it extends from the leading edge to
the position of x/chord=0.5, but with the increase of inlet wind velocity the
laminar shrinks gradually to the surroundings of the leading edge and the tip
by the effect of change in the angle of attack at the local span of the blade.
When inlet wind velocity reaches 15.1m/s, the laminar area exists limitedly in a
narrow part of the leading edge near the tip, and at 20.13m/s the suction side
of the blade becomes the fully turbulent state.
With the increase of wind velocity into the rotor, the angle of attack at the
local span of the blade increases because of interaction between axial velocity
and tangential velocity, and there happens a complex stall phenomenon on the
suction side due to the effect of laminar separation and adverse pressure
gradient. Accordingly, for the accurate prediction of the change of aerodynamic
characteristic caused by the stall phenomenon, we need to estimate the
accurate position and size of laminar separation inside the boundary layer and
the separation caused by adverse pressure gradient.
In case of a flow attached on the blade surface, commonly used the standard
k- model can produce relatively accurate results, but has the problem that it
does not consider the transportation of shear stress in the condition where
separation is caused by adverse pressure gradient and as a result it
overestimates eddy viscosity and consequently estimates separation point and
region inaccurately and has stall delay phenomenon.
To solve this problem, the SST model was developed by Menter et al. in 1994.
Compared to the k- model, the turbulence model is known to estimate
relatively accurately the start point and region of separation caused by adverse
pressure gradient, but still cannot estimate laminar separation on wind turbine
blades.
In general, it is known that separation happens more easily in the laminar
boundary layer than in the turbulent boundary layer, and failure in accurate
estimation of separation point in the boundary layer results in the shrinkage or
expansion of the separation area and inaccurate estimation of separation point
and consequently the overestimation or underestimation of torque on the blade.
It is a well-known fact that the k- model, a fully turbulent model adopted by a
general CFD codes, fails to estimate aerodynamic characteristic accurately for
the post-stall region, and this is because it processes the boundary layer with
a wall function and therefore cannot reflect the flow separation phenomenon
exactly. Accordingly, the result of analysis without using the transition model
may show the tendency of agreeing relatively well with the experiment result in

an operation condition where the wind turbine blades form an attached flow,
but it cannot estimate accurate power performance in the operation range
where stall occurs.
Figure 9 and 10 show representative surface streamline distribution according to
inlet wind velocity (7m/s, 10.02m/s, 20. 13m/s) on the suction side of the blade.
Figure 9 is the result of laminar-transition analysis, and Figure 10 is the result
of applying the standard k- model.
According to Figure 9, at 7m/s an attached flow is formed in most of the area
of the blade except stall appearing limitedly in a narrow part around the hub.
In Figure 10 as well, a stable attached flow is formed throughout the entire
blade except a local area around the hub.
Different from the result at 7m/s, at 10.02m/s an attached flow is formed until
around 35% point from the blade tip but from that point to the hub a stall area
is observed. The flow separated from the hub moves toward the tip along the
suction side surface of the blade due to centrifugal acceleration force and
pressure difference. In Figure 10, however, at 10.02 m/s the attached flow is
maintained almost constant around the leading edge of the blade and the
attached flow is maintained throughout an extensive range except some area
on the posterior margin near the hub. Under the same wind velocity condition,
the results of estimation by the two models show different characteristics, and it
is expected that the result in Figure 10 maintaining the attached flow in a more
extensive area will produce larger torque than that in Figure 9.
Figure 11 is a graph that compared the result of torque experiment conducted
in the NASA Ames wind tunnel by NREL with the result of CFD analysis.
The torque estimated by the transition model is almost identical with the
experiment result, and particularly also in the operation condition of over 10.02
m/s where severe stall happens throughout the entire area of the blade suction
side, the experiment result agrees well with the result of numerical analysis.
Accordingly, it is concluded that laminar separation can be analyzed accurately
and the start point and region of flow separation caused by back pressure
gradient can be estimated relatively accurately by the transition model and, on
the contrary, the standard k- model, a typical fully turbulent model,
overestimates blade torque compared to the experiment result in all cases
except 7m/s, the area before the occurrence of stall. This is mainly because of
failure in the estimation of flow separation as shown in Figure 9 and 10 in
which the attached flow on the blade surface is formed throughout a wider
range than the result of the transition model under the same inlet wind velocity
condition.
Figure 12 is the result of comparing root flap bending moment. The result of

the transition model agrees best with the experiment value, and the k- model
estimates somewhat higher than the experiment value in the segment before
15.06m/s.
Conclusions
The present study performed CFD analysis on S809 airfoil and NREL Phase IV
wind turbine blades. According to the results of analyzing airfoil and 3D wind
turbine blades, the transition model estimated aerodynamic forces, torque and
bending moment most closely to the experiment results. On the contrary, the
standard k- model, which is adopted as a default turbulence model in a
general CFD codes, failed to estimate aerodynamic forces for the area after
stall angle due to failure in estimating the region and position of separation,
and overestimated torque.
In this study estimated aerodynamic forces and performance relatively
accurately using the transition turbulence model. Accordingly, when the
separation in the laminar boundary layer has a considerable effect on the
characteristic of the entire system or when estimating wind turbine blade
performance that needs to estimate the area of stall and the separation point
accurately, the transition model is more recommendable than a fully turbulent
model.
References
[1] Menter, F.R. Two-equation eddy-viscosity turbulence models for engineering
Applications. AIAA Journal, Vol. 32, No. 8, 1994; 1598-1605.
[2] Sommers, D.M. Design and Experimental Results for the S809 airfoil. Airfoils,
Inc., State College, 1989.
[3] Langtry, PR.B, Menter, F.R. Transitional Modeling for general CFD
Applications in Aeronautics. AIAA Paper 2005-522, 2005.
[4] Sommers, D.M. Design and Experimental Results for the S809 Airfoil.
NREL/SR-440-6918, 1997.
[5] R.B. Langtry, J.Gola, F.R. Menter. Predicting 2D Airfoil and 3D Wind Turbine
Rotor Performance using a Transition Model for General CFD Codes. AIAA
2006-0395, 2006.