CFD analysis of 2D and 3D airfoils

using open source solver SU2

Internship Report

Rajesh Vaithiyanathasamy
Sustainable Energy Technology

December 11, 2016

Supervisors:
Prof. Dr. Ir. C. H. Venner (Universiteit Twente)
Dr. H. Ozdemir (ECN)
II
Preface

In my MSc study of Sustainable Energy Technology at University of Twente, Intern-

ship is mandatory. The objectives for the internship is to become acquainted with the
field of interest, apply acquired knowledge and skills in practical situations, work in-
dependently, develop social and communicative skills and carry out a project which
is useful for the employer. During my MSc, I have learnt computational fluid dy-
namics and this internship at ECN provided me the platform to work in a practical
computational environment and implement the theory of Aerodynamics. I would like
to thank Dr. H. Ozdemir for giving me the opportunity to learn and explore at ECN
and guiding me throughout my internship. I would also like to thank Dr. G. Bedon
and Dr. K. Vimalakanthan for helping me with my doubts.

iii
IV P REFACE
Abstract

The open source computational fluid dynamics (CFD) software SU2 is developed for
compressible flows, equipped with approximations like pre-conditioning and artificial
compressibility for incompressible flows. Wind turbines operate at very low velocity
and is a good example of incompressible flows. The performance of SU2 at these
low Mach number flows is analysed with two airfoils FFAW3301 and DU97W300.
The results are compared with Rfoil, Xfoil and experiment. Xfoil and Rfoil are en-
gineering models for aerodynamic design of airfoils. The results from SU2 analysis
shows that the lift coefficient is over predicted in the linear region of non-separated
flow and the over prediction increases to about 30% in the stalled region. Con-
currently, the drag coefficient is under predicted in the linear region and the under
prediction increases in the stalled region of flow. The separation point is predicted
at a later position in the suction side of the airfoil compared to Rfoil but gives better
prediction than Xfoil. Also the wall spacing for y+ = 1 calculated theoretically does
not yield accurate results and need to be reduced to less than half of the value to
get better prediction. Further, the boundary layer thickness estimated from SU2 is
smaller for all angle of attacks than Rfoil and Xfoil for similar cases.

v
VI A BSTRACT
Contents

Preface iii

Abstract v

1 Introduction 1

2 Selection of turbulence scheme and numerical parameters 3

2.1 Requirement to solve Navier-Stokes equations . . . . . . . . . . . . . 3
2.2 Choice of Turbulence model . . . . . . . . . . . . . . . . . . . . . . . . 3
2.2.1 Algebraic model . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.2.2 Spalart Allmaras (SA) model . . . . . . . . . . . . . . . . . . . 4
2.2.3 Two equation models . . . . . . . . . . . . . . . . . . . . . . . 4
2.2.4 7 equation model . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.2.5 Shear Stress Transport (SST) model . . . . . . . . . . . . . . . 5
2.3 Spatial discretisation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.4 Time discretisation scheme . . . . . . . . . . . . . . . . . . . . . . . . 7

3 DU97-W-300 Airfoil 9
3.1 Evaluation of boundary layer to generate mesh . . . . . . . . . . . . . 10
3.2 Lift-coefficient . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
3.3 Drag-coefficient . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
3.4 Pressure coeffiicent (Cp ) . . . . . . . . . . . . . . . . . . . . . . . . . . 13

4 FFA-W3-301 Airfoil 15
4.1 Mesh Generation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
4.2 Finding the right wall spacing: . . . . . . . . . . . . . . . . . . . . . . . 18
4.3 Steady and Unsteady simulation . . . . . . . . . . . . . . . . . . . . . 19
4.3.1 1st order dual time stepping . . . . . . . . . . . . . . . . . . . . 20
4.3.2 2nd order dual time stepping . . . . . . . . . . . . . . . . . . . 20
4.4 Polar plot . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
4.4.1 Choice of the mesh . . . . . . . . . . . . . . . . . . . . . . . . 24
4.4.2 Polar (Lift-coefficient) . . . . . . . . . . . . . . . . . . . . . . . 27

vii
VIII C ONTENTS

4.4.3 Polar (Drag-coefficient) . . . . . . . . . . . . . . . . . . . . . . 28

4.4.4 Vorticity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
4.5 Boundary layer thickness . . . . . . . . . . . . . . . . . . . . . . . . . 31
4.5.1 Velocity method . . . . . . . . . . . . . . . . . . . . . . . . . . 31
4.5.2 Total Pressure method . . . . . . . . . . . . . . . . . . . . . . . 31
4.5.3 Vorticity method . . . . . . . . . . . . . . . . . . . . . . . . . . 32
4.6 3D simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

5 Conclusions 39
 Chapter 1

Introduction

The Stanford University Unstructured (SU2) is a CFD solver for compressible flows
i.e. it is developed for flows of Mach number > 0.3. In general, the cases involv-
ing wind turbine applications have Mach number < 0.3 . In order to understand the
usefulness of SU2 for wind turbine applications, here the SU2 software is validated
for incompressible flows. Further, the SU2 does not have transition model and so
only fully turbulent cases are considered. In this study, SU2 is validated with the
experimental data and the results from other numerical tools. 2D and 3D sections
of airfoils are taken into consideration for analysis.

For incompressible flows SU2 adopts artificial compressibility. Hence the same dis-
cretisation schemes that are applicable for compressible flows are also applicable
for incompressible flows. Generally, in the incompressible form of flow equations,
it is difficult to decouple the pressure term in the Navier-Stokes equation due to
the absence of pressure term in the continuity equation. The usage of artificial
compressibility helps to decouple the pressure and velocity terms in Navier-Stokes
equation by adding an additional pseudo compressible equation. (A. C. Aranake
and Alonso [2014]) Therefore, the difficulty in decoupling of pressure and velocity
terms in incompressible formulation is removed.

The accuracy of results obtained from SU2 for incompressible flows are analysed
for two thick airfoils namely DU97W300 and FFAW3301. Mesh convergence study
is done followed by analysis for optimum y+. Once the suitable mesh is chosen, po-
lar is plotted and the behaviour of SU2 in stalled region is studied with the help of the
lift-drag plot and the position of the separation point. Subsequently, boundary layer
thickness is calculated and compared with Rfoil which generally yields more accu-
rate results for thick airfoils (G. Ramanujam and Hoeijmakers [2016]). Finally 3D
analysis is carrier out for FFAW3301 airfoil to check the accuracy of the solution.

1
2 C HAPTER 1. I NTRODUCTION
 Chapter 2

Selection of turbulence scheme and

numerical parameters

2.1 Requirement to solve Navier-Stokes equations

The flow under analysis is incompressible and with a high Reynolds number. The
medium is air which is of very low viscosity and so inviscid flow assumption might
give good results that are in agreement with experiments for certain cases. Inviscid
flow solution can predict the lift and the pressure distribution around the airfoil but
the drag effects which are mainly caused by the friction can be predicted correctly
by viscous flow analysis. Further the flow over airfoils at high angle of attack tend
to separate at certain point causing stall. This kind of separated flow cannot be well
predicted by the inviscid flow alone. This involve viscous effects too (H. Schlichting
[2000]) which leads to consider boundary layer as well. Additionally, the inviscid
assumption does not satisfy the no-slip boundary condition (ux = 0) which prevents
shear stress from becoming infinite at wall. Hence good prediction of both lift and
drag can be obtained by solving the entire Navier-Stokes equations which involves
both advection and diffusion terms.

2.2 Choice of Turbulence model

Turbulence increases the momentum transport of the flow from free stream to the
flow near the wall. This is due to the Reynolds stresses which delays the separation
making the flow stay more attached as compared to the laminar flows. In laminar
flows the exchange of momentum is insufficient to keep the flow unseparated against
the adverse pressure gradient.

3
4 C HAPTER 2. S ELECTION OF TURBULENCE SCHEME AND NUMERICAL PARAMETERS

In turbulent flows there exist many scales of turbulent eddies. The larger eddies
transfer energy to smaller eddies and in turn the later dissipate them. The difficulty
lies in capturing the entire spectrum of eddies. There are various models of turbu-
lence (G. M. Homsy) that are built into Navier-Stokes equation like Direct Numer-
ical Simulation (DNS), Reynolds Averages Navier-Stokes (RANS) and Large Eddy
Simulation (LES) to capture the turbulence behaviour. In DNS the entire system is
resolved without any turbulence model. Even though it is good to capture all the
spectrum, it is computationally inefficient. RANS is entirely based on modelling.
LES lies in between RANS and DNS. Here it is resolved for large scale eddies and
modelled for small scale which are universal for all the flows. LES is advantageous
since the small eddies are isentropic, simpler algebraic model is sufficient to capture
everything. It uses filtered approach. However, RANS uses modelling for large ed-
dies which are flow dependent and so they are quite difficult. Therefore, LES method
is much more efficient than RANS however, resolving the larger eddies makes the
former computationally expensive. The turbulence model which needs to be used
should be accurate, simple and economical to run. Hence RANS is preferred in SU2
though it is known to be not accurate enough in separated regions, it is computa-
tionally efficient and applicable for industrial applications. The major RANS models
are discussed below.

2.2.1 Algebraic model

This model is very simple but this is not continuous and very difficult to get a con-
vergent solution. It is fine-tuned model and does not work well in regions of adverse
pressure gradient.

2.2.2 Spalart Allmaras (SA) model

This is a one equation model designed for external aerodynamic flows. The con-
stants are tuned for aerodynamic flow field. This is not designed for internal flow
field like stator or rotor in the compressor. The main advantage of this model is that
the equation is very economical and very stable. However, this model does not have
a lot of physics incorporated in it.

2.2.3 Two equation models

k- 

This model involves equation of turbulent kinetic energy (k) and dissipation rate ()
and an empirical relation for  in terms of k and length scale. This model is reason-
2.2. C HOICE OF T URBULENCE MODEL 5

able for many flows and easy to implement. This equation is valid for fully turbulent
flows. However, the prediction is poor for strong separation and rotational flows.
There is also k- RNG model which gives good prediction only for transition flows.

k- ω

This model is based on turbulent kinetic energy (k) and length scale (ω) rather than
dissipation rate and the main advantage is no changes are required to integrate near
the wall in the viscous sublayer.

2.2.4 7 equation model

This model solves Reynolds stress and  equation. This model has no isotropic
assumptions and it is better as turbulence is different in different directions. This
is more complete but it requires six additional equations for each of the turbulent
stresses and an equation for . Therefore, computational effort is large and the
partial differential equations are very stiff. The results are not much better than 1 or
2 equations model to justify extra computational effort.

2.2.5 Shear Stress Transport (SST) model

This combines both the two equation models k- and k-ω. With k-ω it can model
viscous sublayer and switches to k- in the free stream and avoids sensitivity of k-ω
model.
The turbulent boundary layer can be sub-divided into three layers. The laminar sub-
layer (layer close to wall), the log layer and the outer layer. To resolve the boundary
layer there are two approaches as to either use a wall function correlation (30 < y+
< 300) which is empirical or use y+=1. The y+ is a non-dimensional normal spacing
from the airfoil surface whose value determines the part of turbulent boundary layer
that needs to be resolved. The first method of using standard wall function is good
for high Reynolds number flows as the case under consideration. This method re-
solves the log layer where there is equilibrium between production and dissipation of
turbulent energy and therefore avoids the turbulent instabilities in near wall analysis
in viscous sub layer. By this method the order of resolving the boundary layer can
be reduced to 102 compared with the usage of y+ = 1. However, in the separation
region wall function does not give good approximation. Though the second method
of resolving y+=1 is computationally expensive, it gives very good refinement of the
sublayer close to the wall and works fine for complex flows and flows involving sep-
aration. Hence for the analysis here SST turbulent model is utilised with y+ = 1 for
resolving boundary layer.
6 C HAPTER 2. S ELECTION OF TURBULENCE SCHEME AND NUMERICAL PARAMETERS

2.3 Spatial discretisation

In order to resolve Navier-Stokes equations, both the spatial and time domain need
to be discretised. To resolve the spatial domain, it is necessary to define numerical
schemes for the flow variables.

For the compressible flows, shock and rarefraction waves appear naturally and they
are well described as Riemann problem. While resolving the hyperbolic part of
Navier-Stokes equation various schemes were established to solve the Riemann
problem for high schock resolution and reduce the oscillations. To solve the appro-
priate Riemann problem, Godunov linear method or non-linear TVD schemes can
be used. But they are of 1st order and 2nd order respectively. Hence in order to
get accurate and smooth results for the Riemann problem it is necessary to make
higher order schemes. However, the exact solution of the Riemann problem be-
comes expensive and practically unusable for complex problems. The higher order
schemes like Roe, JST, etc., which are approximate Riemann solvers which contain
more physical information of the flow can be used as these are computationally ef-
ficient and accurate. The incompressible flow approximation in SU2 adopts artificial
compressibility for pressure and velocity decoupling and so the already established
solutions of the Riemann problem to resolve compressible flow variables can be em-
ployed for incompressible flows (Kong [2011]). The JST schemes is more stable but
requires finer mesh compared to ROE scheme to attain same order of accuracy.
Hence in the analysis of flow variables, ROE scheme has been used.

To solve the spatial gradients Weighted Least Square (WLS) or Green Gauss (GG)
numerical methods can be used in SU2. WLS avoids the error from the outliers
and its performance is good for structured meshes. The GG method has low er-
rors for triangular and quadrilateral cells. Here in all the cases under consideration,
the meshes are structured and so both the type of slope limiter works good. Al-
though the higher order schemes discussed above give good approximations, they
sometimes introduce unphysical wiggles and so flux limiters are used. The Minmod
limiter is a conservative reconstruction with minimum slope available. This is the
most diffusive of all the limiters and least accurate. The Superbee limiter on the
other hand has five segments in it and follows the upper boundary of the viable lim-
iter. This has the steepest slope possible and does not yield good result for smooth
flows. Therefore in SU2 analysis, Venkatakrishnan limiter is used which is smooth
and continuous and lies between Minmod, Superbee limiters.
2.4. T IME DISCRETISATION SCHEME 7

2.4 Time discretisation scheme

The available time discretisation schemes in SU2 are implicit, explicit Euler and
Runge-kutta schemes. The choice of the scheme for time discretisation is implicit
Euler schemes. Though the implicit scheme has one additional step of computation,
higher CFL can be used which facilitates faster convergence without blowing up the
solution convergence and accuracy. This justifies the extra computational effort it
requires.
8 C HAPTER 2. S ELECTION OF TURBULENCE SCHEME AND NUMERICAL PARAMETERS
 Chapter 3

DU97-W-300 Airfoil

One of the airfoils considered for analysis is TU Delft DU97-W-300, non-symmetric

and blunt trailing edge airfoil with 30% thickness and 0.65 m of chord length. The ex-
perimental values are taken from the Avatar project, (Manolesos and Prospathopou-
los [2015]) for this airfoil to compare with the results from SU2. The experiment
is performed in low speed low turbulent wind tunnel with the Reynolds number of
2.106 . The lift and drag coefficients from SU2 is compared with the corresponding
plots from the experiment and with the results from other computational platforms
like MapFlow, Q3UIC, Rfoil and Xfoil.

The MapFlow is a compressible solver adopted with pre-conditioning for low Mach
number (incompressible) flows. The discretisation used is cell centred ROE scheme
for convection term with Venkatakrishnan limiter. The turbulence model used is
SST and time discretisation is implicit second order method. The schemes used in
MapFlow is similar to that being used in SU2 for analysis as described in the previ-
ous section.

The Q3UIC Quasi-three dimensional Unsteady viscous-inviscid Interaction Code is

developed to predict wind turbine aerodynamic performance. The inviscid part is
modelled by a panel method and the viscous part is modelled by solving the integral
form of the equations. Here the tripping of the laminar to boundary layer is done by
either boundary layer tripping or by en transition method.

Xfoil and Rfoil are engineering models for aerodynamics design of airfoils. Rfoil
is an in-house tool developed by ECN based on Xfoil with additional rotational ef-
fects and gives good prediction for thick airfoils. Here the analysis with SU2, Xfoil
and Rfoil are performed for fully turbulent flows whereas the experiment and other
numerical tools involve transition flow.

9
10 C HAPTER 3. DU97-W-300 A IRFOIL

3.1 Evaluation of boundary layer to generate mesh

The main idea is to resolve the boundary layer to get better predictions for sepa-
rated flows. Hence reasonable prediction of boundary layer thickness is estimated
from Xfoil for various angles of attack. The mesh is generated based on the es-
timation from Xfoil tripped cases. The separation point for each angle of attack is
estimated from Xfoil tripped cases and the mesh is refined based on the y+=1 inside
the boundary layer. The separation point is calculated based on the skin friction
coefficient (Cf ) and the shape factor H (H is the ratio of displacement thickness to
momentum thickness). Cf decreases drastically during transition and increases as
the flow become turbulent due to increase in shear stress. Eventually Cf becomes
zero when the flow separates. Further, H increases and then decreases as the flow
changes from transition to turbulent due to the sudden increase in the displacement
thickness. The shape factor reaches the maximum value of about 1.3 and then
dips indicating the point of separation where the displacement thickness increases
rapidly. Therefore, the position along the airfoil where H becomes maximum and Cf
become zero indicates the flow separation.

Figure 3.1: Boundary layer thickness extracted from Xfoil simulations

Figure 3.1 shows the plot of boundary layer thickness along the suction and pres-
sure side of the airfoil for angle of attack 8, 12 and 16 degrees. The boundary layer
thickness is shown only for region of non-separated flow. It can be seen that the
separation starts at 12-degree angle of attack at x = 0.55. Based on the boundary
layer analysis from the above figures, the maximum boundary layer thickness that
can be reached is 0.06 which is a normalised by the chord length. From the chord
length of 0.65m it can be found that the maximum boundary layer thickness is 32.5
mm and so the mesh is resolved finer for this thickness and a growth rate of less
3.2. L IFT- COEFFICIENT 11

than 1.2 is maintained in the region normal to it. The boundary layer is resolved with
y+ = 1 as this gives a good result for separated flows and double precision is used
to avoid round off errors. Four different meshes used for the analysis which are 300
surface points on airfoil with 300 normal points, 500 surface points with 450 normal
points, 500 surface points with 600 normal points and 1000 surface points with 1000
normal points.

3.2 Lift-coefficient

Figure 3.2: Comparison of lift coefficient obtained by various numerical methods

and experimental data

The plot of lift coefficient versus Angle of Attack (AoA) is shown in the figure 3.2.
The results for the four meshes overlap on each other indicating a grid independent
solution. The experimental data and the Mapflow result involves transition and so
they predict higher lift coefficient (Cl ) values than the corresponding AoA from other
numerical tools. The Cl from SU2 are comparatively lower than non-tripped cases
but stalls at around the same position. This is due to the influence of thicker bound-
ary layer for turbulent flows where the velocity deficit is large which eventually leads
to less difference in freestream and local pressure (δP ) and hence lower lift.

Rfoil is known to be more accurate for thick airfoils and so the result from SU2 is
compared with Rfoil tripped cases. It is found that the SU2 over predicts to a max-
imum of 5% in the linear region and over prediction increases in the stall region
to around 30%. However, in deep stalled region (i.e. 20 degrees angle of attack),
the results from SU2 lies closer to Rfoil tripped scenario. At the same time, it is to
12 C HAPTER 3. DU97-W-300 A IRFOIL

be noted that the performance of SU2 in stall is relatively better than Q3UIC panel
method.

3.3 Drag-coefficient

Figure 3.3: Comparison of drag coefficients obtained by various numerical methods

and experimental data.

Figure 3.3 shows the plot of drag for various angle of attack. The drag has two
parts namely skin friction drag and pressure drag. Generally, the skin friction drag
is more predominant over the airfoils. However, as flow separates the pressure drag
(form drag) increases thus causing the overall drag to increase. The reason for the
general trend of increase in drag coefficient as AoA increases. Here, as SU2 and
Rfoil involves tripped flow (fully turbulent) the drag coefficient is comparatively higher
than that of Mapflow and experiment. However, the drag coefficient calculated from
SU2 is under predicting than that of Rfoil tripped case throughout the entire domain
of Aoa -5 to 16 degrees. This underprediction grows when the flow gets separated.
3.4. P RESSURE COEFFIICENT (Cp ) 13

3.4 Pressure coeffiicent (Cp)

(a) AoA -4 (b) AoA 0 (c) AoA 4

(d) AoA 8 (e) AoA 12 (f) AoA 16

Figure 3.4: Comparison of Cp for the various angle of attacks between SU2 and
Rfoil tripped cases.

Figure 3.4 shows the comparison of the plot of Cp from SU2 with that of the Rfoil
tripped cases for all angle of attacks considered in the polar above. The position
of the lowest pressure at the suction side can be seen moving towards the leading
edge as the AoA increases. This indicates the increase in the adverse pressure
gradient. Also the Cp is over predicted compared to the Rfoil tripped case for all
AoAs. The separation for 12 and 16 degrees AoA can be clearly seen as indicated by
the negligible changes in the pressure coefficient in the suction side as x/c increases.
However, the separation is predicted at a later position of x/c in comparison with Rfoil
tripped case.
14 C HAPTER 3. DU97-W-300 A IRFOIL
 Chapter 4

FFA-W3-301 Airfoil

The airfoil under consideration is FFA-W3-301 which is being used in wind turbines.
It has a chord length of 0.6 m with thickness of 30% and it is tested at Reynolds
number of 1.6 million in laminar wind tunnel of Stuttgart University. The numerical
analysis of airfoil is carried out and compared for fully turbulent flows. Turbulent
flows produce more drag compared to laminar flows. However, the flow is stable
and the small disturbances are damped out whereas it is not the case with the lam-
inar flows. Hence even if the drag is higher for turbulent flows, it is preferred for the
reliable performances sacrificing little efficiency. As SU2 is used for turbulent analy-
sis, the airfoil experimental data (P. Fuglsang and Madsen [1998]) with leading edge
roughness (LER) is taken as a benchmark to compare the results. To obtain LER
effect, trip tape is used in experiments. LER is generally observed when dirt is accu-
mulated while operating the wind turbine in dirty environment. The SU2 analysis is
done at the same Reynolds number of 1.6 million and validated with the experiment
and the Navier-Stokes solver Ellipsys 2D for turbulent flow conditions. The Xfoil and
Rfoil tripped cases are also considered for comparison. Tripping of Xfoil and Rfoil
is done at x/c=0.05 from the leading edge both on pressure side and suction side.
The value of x/c for tripping in both Xfoil and Rfoil is chosen in such a way that the
stagnation point remains just before the tripping point even at very high angle of
attack. Since the airfoil under consideration is very thick, Rfoil tripped scenario is
considered a very good benchmark to compare the SU2 results.

4.1 Mesh Generation

2D structured O-mesh is used for the analysis. Structured mesh offers simplicity
and easy data access. The element choice is quadrilateral as this is highly efficient
to get good convergence and high resolution. The boundary layer is refined with
y+ = 1. The wall spacing s is calculated based on, (Schlichting [1979]) :

15
16 C HAPTER 4. FFA-W3-301 A IRFOIL

δs = y + ν/ρu∗ , (4.1)
where ρ is the density of air, µ is the dynamic viscosity and u∗ is the friction ve-
locity. The density and dynamic viscosity are taken at the conditions of standard
temperature and pressure. The friction velocity is given by
p
u∗ = τw /ρ,
where, τw = 0.5Cf ∗ ρ ∗ U 2 , (4.2)
Cf = 0.0576 ∗ Rex−1/5 f or 5 · 105 < Rex < 107 ,

with y+ the wall spacing δs is calculated to be 8.5 · 10−6 . The wall spacing is given
as the constraint at the boundary layer. The far field is fixed at 90 times the chord
length for minimizing the effect of boundary conditions on calculated flow variables.
The O-mesh is preferred due to its larger stability at the trailing edge for high gradi-
ent flows (Lutton [1989]). This is due to the large accumulation of grid points near
the trailing edge which facilitates smooth gradient transformation. The O-mesh also
allows better estimation of angle of attack at Clmax . The skewness of the mesh is
kept under 0.67 and the area ratio between the cells around a maximum of 1.4. The
aspect ratio which is the ratio of length to width is kept under 4000. The larger val-
ues of the above properties than prescribed leads to convergence problem and bad
interpretation of the result.

Figure 4.1 shows the mesh generated as per the above mentioned parameter with
300 points on the airfoil surface and 400 points normal to it. On performing simula-
tion, it is found that in some points the y+ exceeds 1 even though the wall spacing
of 8.5 · 10−6 used in the mesh is calculated theoretically using y+ = 1.
4.1. M ESH G ENERATION 17

Figure 4.1: Mesh with approximately 120000 cells

18 C HAPTER 4. FFA-W3-301 A IRFOIL

4.2 Finding the right wall spacing:

Figure 4.2: y+ for theoretical and altered wall spacing

Figure 4.2 shows the plot of y+ versus dimensionless x-coordinate of the airfoil
for the previously shown mesh at 10 degrees angle of attack. The theoretically
calculated wall spacing of 8.5 · 10−6 for y+ = 1 yields y+ value of 1.8 at the suction
side of the airfoil near the leading edge. Even though the calculated wall spacing
utilises the local Reynolds number, this higher value of y+ at the leading edge is due
to the high angle of attack which is not taken into account while calculating the wall
spacing. In order to get the y+ less than 1 even at higher angle of attack, the wall
spacing is linearly reduced to 3.5 · 10−6 for which the y+ remains less than 1 around
the entire airfoil as shown in the figure as altered wall spacing.

Figure 4.3: Comparison of Cp for theoretical and altered wall spacing with experi-
ment
4.3. S TEADY AND U NSTEADY SIMULATION 19

Figure 4.3 shows the plot of pressure coefficient (Cp ) versus dimensionless x-
coordinate from the SU2 analysis for the same mesh. This includes the results for
calculated and altered wall spacing along with the experimental values for angle of
attack of 10 degrees. At this angle of attack the flow begins to separate as evident
from the experimental data where the change in pressure coefficient is very little
towards the trailing edge beyond x/c = 0.6. Also SU2 predicts the separation but
at a farther position in the chord at x/c = 0.83. At this point the change in pressure
coefficient starts to become insignificant. The separation is almost at the same point
for both theoretically calculated δs and corrected δs. However, for the corrected wall
spacing the Cp distribution lies closer to the experimental value. Hence the corrected
wall spacing of 3.5 · 10−6 is utilised in the remaining analysis.

4.3 Steady and Unsteady simulation

Both the steady and unsteady simulations are performed to compare the result with
the experimental data. The steady state simulation is based on the method previ-
ously described with the choice of turbulence model as SST. Further the unsteady
simulation is carried out to check for the improvement of the solution. For the un-
steady simulation, dual time stepping is used to achieve faster convergence. Implicit
method can also be used which retains higher order of accuracy. This facilitates
larger time step without affecting the solution convergence. But the usage of implicit
method is expensive for both 2D and 3D cases as it involves large number of points
and require huge system of equations to solve. Therefore dual time stepping is pre-
ferred.

While solving the Navier-Stokes equations, the momentum equation is uncoupled

as velocity and pressure terms and solved separately. The velocity is calculated as
an intermediate step followed by the pressure in the next time step. Finally the ve-
locity is corrected in 3rd step. This method of solving is 1st order pressure accurate
as pressure is completely disintegrated in the equation. This is overcome by includ-
ing pressure partially in step 1 while evaluating intermediate velocity. However, 2nd
order accuracy for pressure is difficult to obtain due to weakly coupled steps. This is
eliminated in dual time stepping. The basic idea is to introduce a fictitious time step
to make unsteady problem into a pseudo-steady problem. This brings in inherent
advantages of a steady-state problem with local time stepping and usage of implicit
methods for convergence acceleration. Dual time stepping can be of 1st or 2nd order.
20 C HAPTER 4. FFA-W3-301 A IRFOIL

4.3.1 1st order dual time stepping

Pseudo time stepping has been used in steady state simulations for rapid conver-
gence of the solution in a robust way. However, it uses 1st order backward temporal
discretisation which is not very good for simulation of unsteady flows.

4.3.2 2nd order dual time stepping

The 2nd order method is implemented to get 2nd order solution for the transient flow.
It consists of two temporal terms: a physical and a global time integration term. The
former term is used to fit the physical phenomena under consideration while the later
is used to facilitate quicker convergence. The solution of each unsteady time step
is like the converged result obtained with the pseudo time step. While implementing
dual time stepping, only the diagonal terms of the Jacobian in the steady state need
to be changed with the addition of source term from previous time step. However,
it has a drawback that it requires to store three different time steps. i.e. the current
time step and two past time steps.

For the unsteady simulation in SU2, the inner iteration is set to 3000 for every exter-
nal iteration and the CFL number for inner loop is kept high to reach rapid conver-
gence. However, the external time step is chosen with care to get solution without
blowing up. For the total far field distance of 90 chords length and input velocity of 20
to 30 m/s for the chosen Reynolds number, the unsteady time to capture the entire
flow physics is 4.5 seconds. Nevertheless, the choice of the total unsteady time is
taken as 50 seconds.
4.3. S TEADY AND U NSTEADY SIMULATION 21

Figure 4.4: Cp for angle of attack of 6 degrees

22 C HAPTER 4. FFA-W3-301 A IRFOIL

Figure 4.5: Cp for angle of attack of 10 degrees

4.3. S TEADY AND U NSTEADY SIMULATION 23

Figures 4.4 and 4.5 show the steady state and unsteady simulations for differ-
ent meshes for angle of attack of 6 and 10 degrees respectively. The mesh utilised
include 500 points on the airfoil surface and 750 points in the normal direction (i.e.
approximately 375000 cells), 1000 points on the airfoil surface and 750 points in the
normal direction (i.e. approximately 750000 cells), 1000 points on the airfoil surface
and 1000 points in the normal direction (i.e. approximately 1000000 cells). For both
cases of 6 and 10 degrees angles of attack the results obtained from all meshes
converge to experimental data better than Xfoil tripped case but not as good as that
of Rfoil. The results from Rfoil lies closer to the experimental data but still there is
an over prediction.

From the figures, it can be seen that the higher the number of cells on the sur-
face or the normal direction the better the accuracy of the result as expected. The
unsteady simulations provide better results for the corresponding number of cells
than that of the steady case. As can be seen from the figures the increase in num-
ber of points makes the Cp value closer to that of the experiment with the unsteady
case 1000*1000 points reaching the highest accuracy. This seems that as number
of cells increases, results that are closer to the experiment could be obtained.

For 6 degrees angle of attack where there is no separation, the trend of Cp is similar
to that of the experiment and the result lies closer to Rfoil. However, for 10 degrees
angle of attack when there is separation, the result from SU2 is not closer to Rfoil
or the experiment. Also the trend is not the same .i.e. SU2 predicts separation at
a later point along the chord as evident from the position where the change in Cp
starts to be negligible. Rfoil predicts the separation at x/c=0.6, according to the ex-
perimental data the separation is at x/c=0.5 whereas from SU2 it is estimated to be
x/c=0.833. The plot of Cf from the result of one of the meshes is shown below which
indicates that the point of separation occurs at x/c = 0.833
24 C HAPTER 4. FFA-W3-301 A IRFOIL

Figure 4.6: Cf for Angle of attack of 6 and 10 degrees

Figure 4.6 shows the skin friction coefficient for 6 and 10 degrees angle of attack,
with the later indicating the separation at x/c = 0.833. This can be visualised from
the point where Cf reaches zero indicating no shear stress as the flow is separated.
The same can also be visualised from the vorticity plot shown below in figure 4.11
because the flow takes the form of eddies or vortices when the flow separates.

4.4 Polar plot

4.4.1 Choice of the mesh

The polar plot is then analysed to check for the accuracy of SU2 for angle of attack
of -5 to 16 degrees with the step size of 4 degrees. Various meshes were analysed
for 6 and 10 degrees angle of attack to choose the optimum mesh for the polar plot.
4.4. P OLAR PLOT 25

Figure 4.7: Cp for angle of attack of 6 and 10 degrees for various meshes
26 C HAPTER 4. FFA-W3-301 A IRFOIL

Figures 4.7 show the plot for 6 degrees and 10-degree angle of attack for differ-
ent meshes with 300, 400, 500 points on airfoil surface and varying the points in the
normal direction. For both 6 and 10 degrees Cp follows the same trend as discussed
in the previous section. The mesh with 300 points on the airfoil surface and 400
points in the normal direction seem to give better prediction at 10 degrees which is
the beginning of the stall region. Also the choice of 120000 cells mesh is made by
keeping in mind that the same mesh will be used for 3D simulation where additional
points needs to be added in 3rd dimension which increases computational cost.

As described before in section 4.2, the solution converges better if the y+ remains
less than 1. For the chosen mesh, the wall spacing is kept as 3.5 · 10− 6 (which is the
altered value as in section 4.2 ) and the y+ obtained for all angle of attacks is shown
below.

Figure 4.8: y+ for various angle of attacks

Figure 4.8 shows the y+ plot versus dimensionless distance of airfoil. It is evident
that with the corrected value of wall spacing the y+ remains less than 1 even at high
angle of attack around the entire airfoil. This would ensure better resolving and good
and closer result to that of the experiment.
4.4. P OLAR PLOT 27

4.4.2 Polar (Lift-coefficient)

Figure 4.9: Plot of lift coefficient (Cl ) versus angle of attack

Figure 4.9 shows the plot of lift coefficient of the FFAW3301 from SU2 simulation,
for the mesh size of 120000 including both steady and unsteady flows, with the
experimental values from Riso for leading edge roughness (denoted as RISO-LER
tape) and the Ellipsys 2D turbulent simulations. The Rfoil and Xfoil tripped cases are
also compared with them. All the simulation results are over predicted compared
to the experiment. Xfoil tripped case over predicts the lift completely in the entire
domain. The results from Rfoil tripped case lies closer to the experiment. The
results from SU2 over predicts the lift coefficient in comparison to Rfoil tripped case
and Ellipsys 2D turbulent simulations. For the steady state simulation, in the linear
region of the lift coefficient (non-separated flow), the over prediction of SU2 is to
the maximum of 10% compared to the experiment. However, in the stalled region
(i.e. between angle of attack 10 and 16) the over prediction is around 30 to 35%.
For completely separated flow as in angle of attack 16 degrees the SU2 result is
over predicted by 15%. For the unsteady simulation, the trend is the same in the
linear and stalled region. However, in the fully separated flow at 16 degrees and
20 degrees, the prediction is closer to the experiment. In the negative angle of
attack, Rfoil tripped case completely over predicts whereas the Xfoil tripped case lies
closer to the experiment. Both the steady and unsteady simulation under predicts
for negative angle of attack to around 20% and 16% respectively. It can be seen
that unsteady simulation gives better result in the fully separated region but with the
increase in computational cost.
28 C HAPTER 4. FFA-W3-301 A IRFOIL

4.4.3 Polar (Drag-coefficient)

Figure 4.10: Plot of drag coefficient (Cd ) versus angle of attack

Figure 4.10 shows the plot of drag coefficient for the same airfoil. The results from
numerical tools under predict the drag value compared to experiment. SU2 results
are comparatively under predicted in the entire domain than Ellipsys or Rfoil tripped
simulations. In the linear region of unseparated flow, the steady and unsteady flows
under predicts to around 30 % in comparison with experimental data. However, in
the separated region the under prediction drastically increases to 200 to 300%. SU2
under predicts the drag coefficient to a maximum of 40% and 3% in the linear region
in comparison with Ellipsys simulation and Rfoil tripped cases respectively. However,
this under prediction decreases to around 25% and increases to 80% respectively
in the separated region.

4.4.4 Vorticity
In an airfoil the lowest pressure appears in the suction side at a point near the
leading edge. From the leading edge till this point there is a favourable pressure
gradient (pressure decreases). Beyond this point till the trailing edge there is an
adverse pressure gradient (increase in pressure). The flow is carried out along the
adverse gradient by the momentum of the surrounding air and it loses velocity in this
region. At some point the adverse pressure gradient becomes so large that the flow
reverses its direction leading to separation and the velocity at this point becomes
negative. The separated flow takes the form of eddy or vortices and therefore can
be better visualised from the vorticity plot.
4.4. P OLAR PLOT 29

(a) AoA -4 (b) AoA 0

(c) AoA 4 (d) AoA 8

(e) AoA 10 (f) AoA 12

(g) AoA 16

Figure 4.11: Vorticity contours obtained by SU2 for various Angle of Attacks with
separation occuring for high angle of attacks (> 100 )
30 C HAPTER 4. FFA-W3-301 A IRFOIL

Figure 4.11 shows the plot of vorticity contours for angle of attacks -4,0,4,8,10,12
and 16 from top left. From the plot it is clear that the separation begins to happen
at angle of attack of 10 degrees. The flow becomes fully separated at 16 degrees
as can be seen from the figure. For negative angle of attack at -4 degrees (top left)
the flow separates due to the camber of the airfoil. Even though the vorticity gives
the clear picture of the flow separation, the exact point of separation can be studied
from the skin friction. At this point skin friction eventually become zero as there is
no wall shear stress when the flow separates.

Figure 4.12: Skin friction coefficient for various angle of attacks

Figure 4.12 shows the plot of skin friction coefficient (Cf ) in the suction side of
the airfoil to indicate the point of separation. The plot uses the result from SU2-fully
turbulent, Rfoil tripped and Xfoil tripped cases for angle of attack of -4 till 12. In Rfoil
the value of H becomes maximum at the point where Cf reaches minimum. The re-
sults from Rfoil are given for fully tripped flow and the maximum value of H indicates
flow separation, where the displacement thickness rapidly increases. In the results
from Xfoil and SU2, the separation is marked by the point where Cf becomes zero
due to zero wall shear stress on flow separation. For all cases involving separation
i.e. angle of attack -4, 10 and 12 degrees, SU2 predicts separation point at a later
position along the airfoil compared to Rfoil but earlier than Xfoil.
4.5. B OUNDARY LAYER THICKNESS 31

4.5 Boundary layer thickness

The boundary layer is formed at the immediate vicinity of the airfoil where significant
viscous effects prevail. For the turbulent flow as in here, the boundary layer contains
eddies and swirls and is more stable with larger thickness than laminar boundary
layer as larger momentum is imparted into. The boundary layer tends to grow as
the result of adverse pressure gradient and velocity reduction in the direction from
leading edge to trailing edge. The geometry used for SU2 simulation is resolved in
the near wall region with y+=1 as this would give a very good estimation for sep-
arated flow. Also the usage of y+=1 would give a very good approximation of the
boundary layer thickness. The results from SU2 can be used to calculate boundary
layer thickness in three ways namely velocity, vorticity, and total pressure methods
(Zafar [2011]).

4.5.1 Velocity method

The boundary layer thickness can be obtained by locating the edge of the boundary
layer. It is at the position where the local velocity becomes 99% of the free stream
velocity in the direction normal to the airfoil. Generally, the velocity increases in
the region of favourable pressure gradient and decreases from the point of lowest
pressure till the trailing edge in the suction side of the airfoil. Due to this decrease in
the velocity, the position where the velocity becomes 99% of Uinf tends to increase
in the direction towards the trailing edge and so the boundary layer thickness. The
value of Cp at the surface of the airfoil from the result of SU2 can be used to calculate
the velocity at the edge of the boundary layer using,

(4.3)
p
u = Uinf 1 − Cp .

The position where this value of u becomes equal to 99% of freestream velocity
gives the boundary layer edge. However, the curvature shape of the airfoil tends
to accelerate the flow increasing the velocity. Therefore, this method cannot predict
the boundary layer thickness properly.

4.5.2 Total Pressure method

From the previous section, it is understood that the position where the 99% of
freestream velocity exists is the edge of boundary layer. Inside the boundary layer
the total pressure remains very small and increases in the normal direction to the air-
foil eventually becomes close to zero approximately near the edge of the boundary
layer. Hence the position where the total pressure reaches the lowest positive value
32 C HAPTER 4. FFA-W3-301 A IRFOIL

gives the edge of boundary layer. By this way the entire boundary layer thickness
can be calculated in the suction side of the airfoil.

4.5.3 Vorticity method

Vorticity is the measure of rotational effect of the fluid and for two dimensional flow
is given by,

∂v ∂u
w= − . (4.4)
∂x ∂y

Inside the boundary layer the velocity gradient is very large and so shear stress de-
veloped between different streamlines leads to rotational effect and thus non-zero
vorticity. However, outside the boundary layer the velocity remains almost close to
the freestream velocity and so the vorticity is close to zero. Therefore, the position
closer to airfoil in the normal direction where the vorticity becomes zero gives a point
of the boundary layer edge. When this is calculated along the entire suction side of
the airfoil the boundary layer thickness can be obtained.

Figure 4.13: Boundary layer thickness for AoA of 0 degrees for FFA-W3-301 airfoil
4.5. B OUNDARY LAYER THICKNESS 33

Figure 4.14: Boundary layer thickness for AoA of 4 degrees for FFA-W3-301 airfoil

Figure 4.15: Boundary layer thickness for AoA of 8 degrees for FFA-W3-301 airfoil
34 C HAPTER 4. FFA-W3-301 A IRFOIL

Figure 4.16: Boundary layer thickness for AoA of 12 degrees for FFA-W3-301 airfoil

Figure 4.13 to 4.16 give the plot of boundary layer thickness calculated from
vorticity method, Total pressure method, Rfoil tripped and Xfoil tripped cases for
angle of attack of 0,4,8 and 12 degrees in the suction side of the airfoil. In Xfoil and
Rfoil the boundary layer is calculated from the correlation,

H1 = 3.15 + 1.72/(H − 1),

H = δ ∗ /θ, (4.5)
δ = θ.H1 + δ ∗ ,

where δ is the boundary layer thickness, δ ∗ is the displacement thickness and θ is

the momentum thickness

As can be seen the boundary layer thickness increases as AoA increases. Xfoil
predicts larger boundary layer thickness for all AoA. Rfoil gives boundary layer thick-
ness smaller than Xfoil but larger than the two methods from SU2. As the Rfoil is
successfully tested for many thick airfoils previously this can be used as the stan-
dard for comparison. The boundary layer thickness from vorticity method is obtained
by tuning the vorticity (ω) between 0.1 and 0.01 and this method gives the largest
thickness near the leading edge and less than the thickness predicted by Rfoil to-
wards the trailing edge. The pressure method gives the lowest approximation of
the boundary layer thickness for all angle of attacks. Hence from SU2 the vorticity
method gives good approximation of the boundary layer thickness closer to Rfoil.
4.5. B OUNDARY LAYER THICKNESS 35

At 12 degree AoA, the boundary layer thickens drastically at the point of separa-
tion and then goes off from the surface due to the complete loss of skin friction
coefficient (wall shear stress). This behaviour is predicted by Rfoil beyond x/c = 0.4
which is the point of separation. However, for the other methods the boundary layer
thickness increases after a small bump at separation.
36 C HAPTER 4. FFA-W3-301 A IRFOIL

4.6 3D simulation
3D simulation is performed on the same airfoil for the mesh of 120000 (300*400)
points in 2D. This mesh is extruded with 17.5 mm thickness including 60 points in
the lateral direction which leads to 7.2 million cells in total for 3D mesh. The 3D sim-
ulations for AoA of 6 and 10 degrees are performed and Cp along the dimensionless
distance over the airfoil are plotted below.

Figure 4.17: Cp for angle of attack of 6 and 10 degrees for 3D and 2D cases/

The figure in the top shows the Cp for AOA of 6 degrees and the one on the
bottom is for AoA 10 degrees along with experiment, Rfoil and Xfoil tripped results.
The 2D plot shows the Cp for 3D along the length of the chosen span and so multiple
values of Cp can be visualised on the same position. Rfoil results lies closer to the
experiment but it is overpredicted. For 6 degree AoA the 3D simulation is closer
4.6. 3D SIMULATION 37

to the Rfoil tripped case which is a good approximation for thicker airfoils. For 10
degrees of AoA which is a separated flow, 2D results predicts separation later than
Rfoil. However, the 3D approximation gives results closer to the experiment and the
separation is predicted earlier than Rfoil. Eventhough the results look better for 3D
simulations, it should be noted that the amount of computational effort for 3D is quite
higher.
38 C HAPTER 4. FFA-W3-301 A IRFOIL
 Chapter 5

Conclusions

The compressible flow solver, SU2, performance for incompressible flows is studied
with two different airfoils namely FFAW3301 and DU97W301. It is found that the
lift coefficient is over predicted and the drag coefficient is under predicted than Rfoil
and experimental data. The percentage of variation in prediction differs in the linear
and the stalled region for both the lift and drag coefficients. The results from the
two airfoils depict that the lift coefficient are over predicted to a maximum of 9 %
and 5% than the experiment and Rfoil respectively in the linear region. However, in
the stalled region the over prediction increases to around 30 to 35% and 15 to 20%
respectively. At very high angle of attack say 20 degrees, the Cl over prediction is
reduced to 4 to 7% in comparison with Rfoil. On the other hand, the drag coefficient
from SU2 is under predicted to around 15 to 20 % as compared with Rfoil in the
linear region and to more than 100% in stalled region. The preliminary 3D results
lie in close proximity to Rfoil and so more accurate than 2D results. However, this
is acheived at the expense of the computional cost. Furthermore, the usage of
theoretically calculated wall spacing makes y+ > 1 in region closer to leading edge
at high angle of attack. This also affects the accuracy of the results. So the wall
spacing need to be reduced to around half of the calculated value to maintain y+
less than one at all positions around airfoil and to get accurate results. Finally, it is
found that the SU2 predicts separation exactly at the same angle of attack as that of
experiment and Rfoil. However, the point of separation is predicted at a later position
in the suction side of the airfoil surface than that of experiment and Rfoil. The results
from SU2 looks satisfactory with small deviation in the region of non-separated flow.
Its performance still need to be improved in the stalled region.

39
40 C HAPTER 5. C ONCLUSIONS
Bibliography

A. K. Lonkar T. W. Lukaczyk D. E. Manosalvas K. R. Naik A. S. Padron B. Tracey

A. Variyar A. C. Aranake, S. R. Copeland and J. Alonso. Stanford University
Unstructured (SU2): Open-source Analysis and Design Technology for Turbu-
lent Flows. AIAA Journal, 51(January):1–19, 2014. ISSN 0001-1452. doi:
10.2514/1.C032491.

K. S. Breuer J. W. M. Bush C. Clanet M. Fermigier S. Hochgreb J. R. Koseff B.

R. Munson K. G. Powell D. Quere J. J. Riley C. R. Robertson A. J. Smits S. T.
Thoroddsen J. M. Wallace G. M. Homsy, H. Aref. Multimedia Fluid Mechanics.
Cambridge University Press.

H. Ozdemir G. Ramanujam and H. W. M. Hoeijmakers. Improving Airfoil Drag Predic-

tion. 34th Wind Energy Symposium, (January):1–12, 2016. ISSN 2016-0748. doi:
10.2514/6.2016-0748. URL http://arc.aiaa.org/doi/10.2514/6.2016-0748.

K. Gersten H. Schlichting. Boundary layer theory. Mc Graw Hill, eighth edition,

2000.

C. Kong. Comparison of Approximate Riemann Solvers. Thesis, (May):1–69, 2011.

URL http://wap.rdg.ac.uk/web/FILES/maths/CKong-riemann.pdf.

M. J. Lutton. Comparison of C-and O-Grid Generation Methods Using a NACA

0012 Airfoil. pages 1–129, 1989. URL http://oai.dtic.mil/oai/oai?verb=
getRecord{&}metadataPrefix=html{&}identifier=ADA216375.

M. Manolesos and J. Prospathopoulos. CFD and experimental database of flow

devices, comparison. Avatar Task 3.1 Report, (February):1–106, 2015.

K. S. Dahl P. Fuglsang, I. Antoniou and H. Madsen. Wind tunnel tests of the FFA-
W3-241, FFA-W3-301 and NACA 63-430 airfoils, volume 1041. 1998. ISBN
8755023770.

H. Schlichting. Boundary Layer Theory. Mc Graw Hill, 1979.

I. Zafar. Numerical Investigation of Boundary Layer Characteristics for Transitional

Flow over 2D Airfoils. Thesis, (October):1–102, 2011.

41