Merlin300 Winglet Kai - Thesis

See discussions, stats, and author profiles for this publication at: https://www.researchgate.
net/publication/279884127
Winglet Design for a Fairchild Merlin III using CFD Analysis
Research · July 2015

DOI: 10.13140/RG.2.1.3670.8968
CITATIONS READS
0 963
1 author:
Kai Lehmkuehler
The University of Sydney
10 PUBLICATIONS 70 CITATIONS
SEE PROFILE
Some of the authors of this publication are also working on these related projects:
UAV flight testing and system identification View project
All content following this page was uploaded by Kai Lehmkuehler on 08 July 2015.
The user has requested enhancement of the downloaded file.

Winglet Design for a Fairchild Merlin III
using CFD Analysis
Kai Lehmkuehler
SID 306022249
Semester 1, 2010
Acknowledgements
I would like to thank Dr KC Wong for his support during the past months. Prof.
Steven Armfield kindly allowed me extended access to the computing facilities.
Without that, the scope of the project would not have been possible. Dr Rod Fiford
helped me with moody computer hardware.
The staff at Aeronautical Engineers Australia, especially Doug McPherson and Kim
White, were always happy to answer questions.
Micheal Chadwick from Winrye Aviation kindly provided the wing tip models and
all available information on the aircraft, as well as the initial idea to the project.
Finally, Dr Michael Zedler, Abel-John Buchner and Kim White read draft versions of
the manuscript and provided many useful comments for improvement.
Note:
This work is copyright. Any usage, apart from academic purposes, requires a writ-
ten permission by the author.
i
Abstract
This report investigates the aerodynamic characteristics of a winglet design for a
Fairchild Merlin III 8-seat, twin turboprop aircraft. The winglet was designed by
the aircraft owner to improve the aesthetics of the aeroplane. To ensure that the
winglet will not have any negative effect on the aircraft, a series of comparative
CFD simulations was carried out. These simulations investigated the different wing
tip shapes and were optimised on drag prediction, following guidelines from the
literature. Models of the wing in isolation from the fuselage were used at 1/10th
scale. A hybrid mesh with unstructured surface- and volume grids in addition to a
structured boundary layer mesh was used. This approach is the most efficient one
for limited hardware capabilities.
It was not possible to benchmark the results against a independent viscous solu-
tion, so confidence had to be gained by comparing parts of the results with accepted
computational and empirical methods. Good agreement was achieved for 2-d pres-
sure distributions, as well as for the 3-d lift- and drag curves. The predictions for
the friction drag were within one drag count of the corresponding empirical meth-
ods. Further, no significant problems with solution sensitivity on various parame-
ters, except the mesh size, was found. The mesh size was limited by the available
hardware. The solutions were found to be mesh dependent, but converging with
mesh size. More work on a larger computer is required to carry out a full mesh
dependency study.
The results for the winglet showed a maximum 1.3% increase in L/D of the wing
at α = 4◦ . A strong leading edge vortex created by the high sweep of the winglet
(comparable to a delta wing) severely limited the performance gains by inefficient
lift generation and high friction drag. Further the missing aerofoil section resulted
in high flow gradients with separation near the leading edge. Stalling of the winglet
at high sideslip angles did not cause any interference with the ailerons. The stalling
behaviour of the main wing was not significantly changed at high angles of attack.
The limited efficiency of the winglet resulted in small air loads on it and only mod-
erate wing bending moment gains. The question, whether stalling of the winglet
will cause any structural flutter, could not be answered with the methods available.
Based on the findings for the original winglet, a new, improved design was tested.
It featured less sweep and twice the surface area, with a NACA 64009 aerofoil.
This new design improved the cruise L/D by 5%, which is a threefold improvement
over the original winglet. At the same time, the skin friction drag was found to be
similar, despite the large change in area. Finally, the wing bending moment was
only increased by an additional 2% in cruise. The performance gains of the new
winglet design were comparable to industry standards reported in literature.
ii
Contents
Acknowledgements i
Abstract ii
List of Figures vi
List of Tables viii
Nomenclature ix
1 Introduction 1
2 Background 3
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.2 Aircraft Drag . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.2.1 Fundamentals of Drag . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.2.2 Lift Induced Drag Reduction Concepts . . . . . . . . . . . . . . . . 5
2.2.3 Winglet Principles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.2.4 Design Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.2.5 Winglets on General Aviation Aircraft . . . . . . . . . . . . . . . . . 11
2.3 Computational Fluid Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.3.1 Motivation for CFD . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.3.2 Simulation Accuracy . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.4 Drag Prediction with CFD . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.4.1 Boundary layers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.4.2 Turbulence Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.4.3 Meshing Requirements . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3 Problem Description 25
3.1 The Merlin III Aircraft . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.2 Wing Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.3 Wing Tip Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.4 Winglet Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
iii
Contents Contents
4 Experiment Setup 31
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
4.2 Common settings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
4.2.1 Flight Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
4.2.2 Problem Simplifications . . . . . . . . . . . . . . . . . . . . . . . . . 35
4.3 CFD - Pre-Processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
4.3.1 Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
4.3.2 Geometry Import . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
4.3.3 Mesh Generation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
4.3.4 Surface Mesh . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
4.3.5 Prism Boundary Layer Mesh . . . . . . . . . . . . . . . . . . . . . . . 43
4.3.6 Mesh Quality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
4.4 CFD - Flow Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
4.4.1 Boundary Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
4.4.2 Turbulence Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
4.4.3 Other Solver Settings . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
4.4.4 Solution Convergence . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
4.5 CFD - Post-processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
4.5.1 Force Processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
4.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
5 Verification of the CFD Simulation 53

5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
5.2 Flow field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
5.2.1 2-D Pressure Distribution . . . . . . . . . . . . . . . . . . . . . . . . 54
5.2.2 3-D Pressure Distribution . . . . . . . . . . . . . . . . . . . . . . . . 56
5.2.3 Skin friction Drag . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
5.2.4 Boundary Layer Shape . . . . . . . . . . . . . . . . . . . . . . . . . . 62
5.2.5 Wingtip Vortices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
5.2.6 Flow Similarity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
5.3 CFD Simulation Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
5.3.1 Mesh Size . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
5.3.2 Domain size . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
5.3.3 Choice of Turbulence Model . . . . . . . . . . . . . . . . . . . . . . . 69
5.3.4 SST Model Verification . . . . . . . . . . . . . . . . . . . . . . . . . . 70
5.3.5 Turbulence Intensity . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
5.3.6 Advection Scheme Selection . . . . . . . . . . . . . . . . . . . . . . . 73
5.3.7 Dynamic Viscosity of Air . . . . . . . . . . . . . . . . . . . . . . . . . 73
5.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
6 Results and Discussion of the Original Winglet 75

6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
6.2 Square Wingtip . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
6.3 Winglet . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
6.3.1 Effect on Performance . . . . . . . . . . . . . . . . . . . . . . . . . . 78
6.3.2 Flow Features . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
6.3.3 Effect on Handling Qualities . . . . . . . . . . . . . . . . . . . . . . . 85
6.3.4 Effect on Wing Structure . . . . . . . . . . . . . . . . . . . . . . . . . 90
6.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
iv
Contents Contents
7 Improved Winglet Design 93

7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
7.2 Design Improvements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
7.2.1 Design Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
7.2.2 Geometry of the New Winglet . . . . . . . . . . . . . . . . . . . . . . 95
7.2.3 Performance Assessment . . . . . . . . . . . . . . . . . . . . . . . . 96
7.3 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
8 Future Work 103
9 Conclusion 105
References 106
A List of Mesh Settings 110
B Selected Pressure Distributions

on the Original Winglet 112
C Selected Pressure Distributions on the New Winglet 114
D Limitations of inviscid methods

for winglet design 115
E CFX Solver settings 117
v
List of Figures
2.1 Velocities on wingtip projected in the XY-plane . . . . . . . . . . . . . . . 7

2.2 Structure of a boundary layer . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.3 Boundary layer velocity profiles along the upper wing skin . . . . . . . 18
3.1 Fairchild Merlin III VH-SSM . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

3.2 Fairchild Merlin III 3-view . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.3 Hoerner Tip Geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.4 Original winglet prototype . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.5 Original winglet design modelled onto a Hoerner tip . . . . . . . . . . . 30
3.6 Original Winglet Geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
4.1 Estimate of the effect of flaps on wing CL . . . . . . . . . . . . . . . . . . 34

4.2 Merlin III wing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
4.3 Standard Domain Dimensions in [m] . . . . . . . . . . . . . . . . . . . . . 38
4.4 Mesh overview at the symmetry plane . . . . . . . . . . . . . . . . . . . . . 40
4.5 Surface mesh overview for the standard wing . . . . . . . . . . . . . . . . 41
4.6 Surface mesh details . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
4.7 Matlab tool for boundary layer mesh settings . . . . . . . . . . . . . . . . 44
4.8 Prism mesh details . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
4.9 Wing y + detail for α = 5◦ . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
4.10 Typical CFX mesh quality diagnostics output . . . . . . . . . . . . . . . . 47
4.11 Locations of Maximum Residuals for the winglet at α = 3◦ . . . . . . . . 50
4.12 RMS and MAX residuals for the winglet at α = 3◦ . . . . . . . . . . . . . . 50
5.1 JAVAfoil pressure distribution compared to CFD data at the wing root 55
5.2 Aerofoil shape change due to boundary layer . . . . . . . . . . . . . . . . 55
5.3 Tornado wing geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
5.4 Tornado grid convergence for lift and drag . . . . . . . . . . . . . . . . . 57
5.5 Tornado pressure distribution for α = 5◦ . . . . . . . . . . . . . . . . . . 58
5.6 CFD solution vs. Tornado results for CL and CDp of the square wingtip 58
5.7 CFD solution for CD,viscous versus angle of attack . . . . . . . . . . . . . . 60
5.8 Wall shear stress for α = 13◦ (top) and α = 3◦ (bottom) . . . . . . . . . . 61
5.9 Wing tip vortices visualised in the tow tank . . . . . . . . . . . . . . . . . 62
5.10 Wing tip vortices at α = 3◦ . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
5.11 Effect of Reynolds number on stalling of the winglet at α = 6◦ . . . . . 64
5.12 Lift and drag vs. relative meshsize squared . . . . . . . . . . . . . . . . . 67
5.13 Variation of L/D with domain size at α = 6◦ . . . . . . . . . . . . . . . . . 68
5.14 Residuals for k − turbulence model compared to SST model . . . . . . 69
vi
List of Figures List of Figures
5.15 SST first blending function for α = 5◦ . . . . . . . . . . . . . . . . . . . . . 70

5.16 Velocity contours at α = 5◦ . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
5.17 SST second blending function for α = 5◦ . . . . . . . . . . . . . . . . . . . 71
5.18 L/D sensitivity to turbulence intensity for α = 6◦ . . . . . . . . . . . . . . 72
6.1 L/D for the square tip versus the Hoerner tip . . . . . . . . . . . . . . . . 76
6.2 Vortex comparison on square and Hoerner tip at α = 6◦ . . . . . . . . . 77
6.3 Pressure comparison on square and Hoerner tip at α = 6◦ . . . . . . . . 78
6.4 Performance of the wing with different tip designs . . . . . . . . . . . . 79
6.5 Wall shear and vortices on the winglet . . . . . . . . . . . . . . . . . . . . 81
6.6 Pressure distributions on the winglet . . . . . . . . . . . . . . . . . . . . . 82
6.7 Pressure distributions on the winglet . . . . . . . . . . . . . . . . . . . . . 83
6.8 3-d flow and tip vortices on the winglet at α = 6◦ and zero sideslip . . 84
6.9 Stall comparison of Hoerner wing and winglet at α = 13◦ . . . . . . . . 86
6.10 Vortex development with increasing sideslip, winglet not yet fully stalled 87
6.11 Vortex development with increasing sideslip, winglet fully stalled . . . 88
6.12 Pressure distribution on the wing at 1/4 chord . . . . . . . . . . . . . . . 91
7.1 New Winglet Geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96

7.2 Performance comparison of new winglet . . . . . . . . . . . . . . . . . . . 97
7.3 Vortex comparison on the two winglets . . . . . . . . . . . . . . . . . . . . 98
7.4 Pressure and wall shear comparison on the two winglets . . . . . . . . . 99
7.5 Pressure distribution on the new winglet at quarter chord . . . . . . . . 100
7.6 Pressure distributions on the new winglet for α = 6◦ . . . . . . . . . . . 101
B.1 Locations of pressure distributions . . . . . . . . . . . . . . . . . . . . . . 112

B.2 Winglet pressure distributions for α = 3◦ , β = 0◦ . . . . . . . . . . . . . . 112
B.3 Winglet pressure distributions for α = 3◦ , β = 6◦ . . . . . . . . . . . . . . 113
B.4 Winglet pressure distributions for α = 3◦ , β = 10◦ . . . . . . . . . . . . . 113
C.1 New winglet pressure distributions for α = 3◦ , β = 0◦ . . . . . . . . . . . 114
D.1 Tornado L/D comparison . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116

D.2 Tornado Cp distributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
vii
List of Tables
3.1 Merlin III data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

3.2 Wing data for the wing outboard of the fuselage joint . . . . . . . . . . . 27
3.3 Winglet data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
4.1 Mesh data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
5.1 Parameters for mesh size verification . . . . . . . . . . . . . . . . . . . . . 66

5.2 Domain sizes for verification . . . . . . . . . . . . . . . . . . . . . . . . . . 68
5.3 Sensitivity to free stream ω for α = 5◦ . . . . . . . . . . . . . . . . . . . . 72
5.4 Advection scheme sensitivity for α = 5◦ . . . . . . . . . . . . . . . . . . . 73
5.5 Sensitivity to dynamic viscosity . . . . . . . . . . . . . . . . . . . . . . . . . 74
6.1 Effects of the winglet on the wing 1/4 chord pitching moment . . . . . 89
6.2 Load summary for the main flight conditions . . . . . . . . . . . . . . . . 90
7.1 New Winglet data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95

7.2 New winglet performance comparison . . . . . . . . . . . . . . . . . . . . 97
7.3 Effects of the new winglet on the wing 1/4 chord pitching moment . . 102
A.1 Global mesh settings for all cases . . . . . . . . . . . . . . . . . . . . . . . 110

A.2 Common mesh settings for all cases . . . . . . . . . . . . . . . . . . . . . . 110
A.3 Mesh sizing for Hoerner wing . . . . . . . . . . . . . . . . . . . . . . . . . . 111
A.4 Mesh sizing for wing with original winglet . . . . . . . . . . . . . . . . . . 111
A.5 Mesh sizing for wing with new winglet design . . . . . . . . . . . . . . . . 111
viii
Nomenclature
a = Lift curve slope CD = Coefficienct of drag: D/(qS)

b = Wing span CD,0 = Coefficienct of form drag
c = Wing chord CDp = Coefficient of pressure drag
D = Drag CD,visc = Coefficient of viscous drag
e = Oswald wing efficiency Cf = Coefficient of friction
F1 , F2 = SST model blending functions CL = Coefficient of lift: L/(qS)
I = Turbulence intensity Cp = Pressure coefficient: (p − p∞ )/q∞
k = Turbulence kinetic energy Cy,β = Coefficient of sideforce due
l = Turbulence length scale to sideslip
L = Lift Cl,β = Coefficient of roll moment due
L = Characteristic length of a flow to sideslip
L/D = Lift on drag ratio Cn,β = Coefficient of yawing moment
M = Mach number due to sideslip
p = Pressure X∞ = Free stream conditions of X
2
q = Dynamic pressure: 0.5ρV
Re = Reynolds number: ρV L/µ α = Angle of attack
S = Wing area β = Angle of sideslip
V = Velocity δ = Boundary layer thickness
+
y = Dimension-less wall distance = Turbulence dissipation rate
λ = Wing taper ratio
2
AR = Aspect ratio: b /S Λ = Wing leading edge sweep
BM = Wing bending moment µ = Dynamic viscosity
CFD = Computational fluid dynamics ν = Kinematic viscosity: µ/ρ
GA = General aviation ρ = Air density
MAC = Mean aerodynamic chord ω = Turbulence eddy frequency
MTOW = Maximum take off weight
NACA = National Advisory Committee
for Aeronautics
SST = Shear stress transport
turbulence model
T AS = True air speed
ix
1
Introduction
The purpose of this thesis is the aerodynamic characterisation of a winglet design

for a Fairchild Merlin III aircraft. The task was presented to me by Doug McPherson
from Aeronautical Engineers Australia1 . He introduced me to Micheal Chadwick
from Winrye Aviation in Bankstown, who are currently restoring and upgrading a
Merlin III aircraft. Once completed the aeroplane is designated to be chartered out
to business customers. With that in mind, Mr Chadwick designed the winglet to
improve the visual appearance of the aeroplane.
The winglet was designed using resources for homebuilders, as the main purpose
was not to maximise performance gains. Nevertheless the addition of a winglet
onto a existing wing will have an influence on the aerodynamic and structural char-
acteristics of that wing. It is therefore necessary to investigate these effects to
ensure the winglet will not cause any harm or unwanted performance degradation.
This was the main task of this project. After the original winglet design was investi-
gated, several flaws of the design emerged. To improve on the design, a new winglet
concept was then conceived and compared to the results of the original shape.
To perform this investigation within the short remaining time frame, it was de-
cided to attempt a high quality CFD solution, benchmarked against other numerical
methods. There was no time to do experimental work to confirm the findings.
In Chapter 2 the theoretical concepts of winglets and CFD are introduced, based on
the available literature. The chapter focuses on winglet design methods which are
applicable to low speed aeroplanes. In the CFD part, the main emphasis is placed
on drag prediction methods, as these are most important for winglet design.
The aircraft and the different geometries of wing tips used for this project are
introduced in Chapter 3.
Chapter 4 details the setup of the CFD simulation. The process of making the CAD
models and the creating the meshes described, as well as the assumptions used
1
after my initial project, a co-operation with a company called AirAffairs, stalled after 6 months
due to zero support by that company.
1
1. Introduction
to configure the CFD solver. The chapter also contains some discussion of the
difficulties encountered during this project to assist future research in this field.
To gain confidence in the simulation results, they are verified against several ac-
cepted empirical and numerical methods for aerodynamic predictions in Chapter 5.
These methods include inviscid panel codes (two- and three dimensional), together
with accepted equations for skin friction drag. The second part of the chapter con-
tains studies of the CFD solution sensitivity to the main simulation parameters, for
example the mesh size and the turbulence model selection.
The results for the original winglet are discussed and interpreted in Chapter 6.
Detailed descriptions of the influences of the winglet on performance, handling
qualities and wing structure are given. The effects are explained with the aid of
flow field visualisations obtained from the CFD solutions.
In Chapter 7, a new and improved design is introduced to resolve some shortcom-

ings of the original winglet. Results for this design are presented and compared to
the original winglet.
Finally, topics for future work are presented in Chapter 8, which are mainly con-
cerned with some detailed design issues of the new winglet that could not be com-
pleted on time.
2
2
Background
2.1 Introduction
This section contains a review of the concepts used in this project. It is based on the
available literature on the subjects. First, the properties of aircraft drag are intro-
duced, focusing on the lift induced component, and including various techniques
for the reduction of this drag. Winglets will be covered in detail as they are the
main topic of this report. Then concepts of CFD analysis methods are discussed,
with emphasis on drag prediction.
This chapter introduces the concepts only, more information on the application of
these methods is given in the relevant sections later on.
2.2 Aircraft Drag

In this section a brief overview on aircraft drag in general is given. This is based
on the very informative and complete paper by Kroo [1]. The rest of this section
will then discuss the winglets in more detail, focusing on the characterisation and
design approaches. Most of the existing literature treats the question of winglet
efficiency gains over other concepts for drag reduction. While being instructive in
understanding the physical principles of winglets, this topic will only be visited
briefly as this aircraft is going to be fitted with winglets for aesthetic reasons in any
case. Therefore the main question is not whether the winglet is the most efficient
way of reducing drag but to design a winglet that can be fitted without causing any
problems. Any performance gains are a welcome side effect.
2.2.1
Fundamentals of Drag
Aircraft drag is influenced by many properties. The three main categories are vis-
cous or form drag, lift induced drag and wave drag. The first is relatively indepen-
3
2. Background 2.2. Aircraft Drag
dent of the flight condition, the second varies roughly with the square of the lift
coefficient and the last is only applicable at speeds near or above M = 1. The wave
drag will be ignored here because the maximum speed of the aircraft of concern is
less than M < 0.4.
This leads to the standard drag equation:
CD = CD,0 + KCL2 (2.1)
where K is a correction factor for the additional losses due to the finite wing span,
usually written as 1/π eAR. Here e is the ‘Oswald efficiency’, a correction for a non-
elliptical lift distribution. Theoretically e is 1 for an elliptical lift distribution and
about 0.8 for a not excessively tapered, rectangular wing. In reality other factors
caused by the viscosity of the air also influence the value of e [1]. The value for e
can be found by plotting CD versus CL2 and finding the slope. It is a good measure-
ment to compare the drag of wings with identical aspect ratio. The base drag CD,0
contains all effects that arise due to viscosity and boundary layer interference. A
breakdown of the influence of different aircraft components on the base drag can
be found in most books on aircraft performance [2].
Equation (2.1) can be written as
L2
D = CD,0 qS + K (2.2)
qS
This implies that for constant lift the drag is minimum if the two terms are equal.
Consequently is the lift induced drag about half of the total drag when the aircraft
is operated at the maximum lift to drag ratio (L/D), which will be the case for
cruise [1]. At low speeds, the required high lift coefficients cause the total drag to
be dominated by the lift dependent term (up to 90%). Therefore reducing the lift
induced drag is important for the performance and efficiency of the aeroplane. The
cruise efficiency and range will be improved as well as the climb performance.
If the standard drag polar is written in dimensional form as before with K expanded,
L2
D = CD,0 qS + (2.3)
qπ b2 e
we can see that the drag is actually dependent on the wing span and not on the
aspect ratio. This will become an important observation when methods for drag
reduction are discussed in the next section.
This drag polar is the result of Prandtl’s lifting line theory and has been found to
be quite accurate for most untwisted wings of AR > 3. For a twisted wing the value
of e is no longer a constant as the local angle of attack varies along the wing span.
This makes comparisons of the wing efficiency more complicated. The drag polar
4
for a twisted wing contains an additional linear term in CL ,
CD = CD,0 + K1 CL + K2 CL2 (2.4)
with K1 depending on the amount of twist and K2 being the same as K before.
The lift induced drag is caused by the downwash created by the vortices originat-
ing at the wing tips. These vortices are produced by the cross-flow around the wing
tip from the high pressure below the wing to the low pressure above the wing. In
addition, energy is lost in these vortices due to frictional effects [3]. More theoret-
ical derivations can be found in [1], [4] or [5]. These more advanced concepts are
beyond the scope of this project as instead of a purely theoretical drag prediction
an experimental approach is used.
Aircraft designers have always tried to minimise the generation of lift induced drag
for obvious reasons: The lower the drag, the more fuel efficient the aeroplane be-
comes. Additionally, the performance in critical flight conditions like take off and
landing will be improved. Several methods to achieve this task exist and others are
still in the research stage. This will be the topic of the next section.
2.2.2
Lift Induced Drag Reduction Concepts
By inspecting equation (2.3), the most straight forward method to reduce the in-
duced drag is to extend the wing span b. Unfortunately there are several factors
that prevent this method to be universally applicable. For example:
Wing span restrictions: Some aeroplanes are span restricted due to airport clear-
ance (Airbus A380) or competition rules (Sailplanes).
Effect on wing root bending moment: Adding more wingspan will create extra air
loads with a large moment arm about the wing root. This will negatively affect
the root bending moment and therefore the structural weight of the wing.
This is especially important for a retro-fit of a drag reduction device, where
the wing structure is given. Any increase in loads will require reinforcements,
that will add extra weight. This offsets some of the efficiency gains.
Additional surface friction drag: The addition of a span extension will increase
the wetted area of the wing. This results in a larger skin friction drag that will
reduce the overall drag reduction of the extension. This is a major factor in
determining the advantages of various drag reduction methods.
5
For these reasons, several other methods are considered by aircraft designers. The
most popular is the addition of near vertical span extensions, called Winglets, as
introduced by Whitcomb in 1976. These are discussed in great detail in the next
section.
Other methods include the use of ragged wing tips, which are highly swept, high
aspect ratio extensions in the wing plane as introduced by [6] and used on the latest
Boeing airliners. These tip designs appear to be more efficient than winglets at low
lift coefficients (CL < 0.3), as required in cruise by transonic airliners.
Further, several concepts are studied theoretically or experimentally without being

used today. These include spiroid tips, currently evaluated by [7], which make a
full circle and join back onto the wing. Here the question is, whether the additional
interference of the spiroid ending on the wing surface will offset the efficiency gains
by the device.
Kroo proposes a C-wing in several of his papers, for example [1] and [8]. While these
configurations improve the tip efficiency of a wing even further than a winglet, they
introduce structural problems that lead to weight penalties. To the knowledge of
this author, no C-wing has been used on a production aircraft.
Finally tip sails, that are multiple high aspect ratio surfaces extending from the wing
tip like feathers of birds were studied. According to Kroo [1] these are currently less
effective than a single surface winglet.
After this general overview, the next sections introduce winglets and their design
requirements in detail as they are the main topic of this text.
2.2.3
Winglet Principles
Winglets are small, near vertical lifting surfaces mounted on the wing tips. If care-
fully designed they reduce the induced drag of the wing and therefore increase the
performance either in terms of less fuel consumption or longer range. Other per-
formance parameters like rate of climb are also positively affected. The addition of
the winglet always adds structural weight. This increase has to be overcompensated
by the drag reduction to be viable.
The design of a successful winglet is a difficult process, where many variables need
to be optimised before a real performance gain will be achieved. Depending on the
initial wing design it might even not be possible to design a winglet that improves
the overall characteristics of the aircraft.
A lot of literature exists on this topic, but mostly this concerns a transonic flight
regime as required by transport airliners and business jets. NASA [9], Boeing [10]
6
Figure 2.1: Velocities on wingtip projected in the XY-plane
and the Douglas Aircraft Corporation [11], as well as many others have studied
winglets for high speed flight. As these findings only have a limited applicability
for the current project, they will not be discussed in more detail. The remaining
text will concentrate on the design principles of winglets and their application for
low speed flight.
Apart from the possible performance gains winglets are often used for aesthetic
reasons. These swept surfaces at the wing tip simply look attractive on an aero-
plane. Most business jets use winglets for that reason even on new designs, al-
though it can be shown that on new wing designs a planar extension of identical
surface area and perimeter length is more effective [12]. This leads to the ragged
wing tips mentioned above.
Winglets alter the pressure field at the wing tip. To do that, they have to produce
a significant inboard force normal to the winglet [13]. This is achieved as shown
in Figure 2.1. The figure shows a projection of the velocity streamlines into the
XY-plane at the wing tip. In the vector diagram it can be seen that the developing
wing tip vortex creates an effective angle of attack for a vertical surface parallel to
the wing tip. A winglet placed in this flow as shown in the top part of the figure
will create a normal force in the Y direction in the same manner as the wing itself
creates lift [14].
This side force reduces the lift induced inflow above the wing tip, which can be seen
in the figure. Comparing the streamlines, the flow inboard of the tip is straightened
7
in Y-direction by the presence of the winglet. The spanwise velocity components

(Vy ) in the core of the vortex flow around the tips, which contain the highest energy,
are substantially reduced. At the same time the tip of the winglet itself creates a
new, smaller vortex (which can also be seen in the figure at the trailing edge of the
winglet). Therefore Whitcomb [13] calls winglets ‘vortex diffusors’, as they spread
the tip vortices over a larger area and reduce their energy losses by that process.
2.2.4
Design Parameters
The ESDU [15] lists the following design objectives for winglets (similar objectives
can also be found in most other papers covering this topic):
• To generate the required lift distribution required for the desired drag reduc-
tion. Whitcomb [13] states that the optimum span loading for a winglet should
have a relatively high load near the root compared to an optimum elliptical
loading for a normal wing.
• To generate this lift distribution as efficiently as possible.
• To minimize the effects if interference at the junction. For transsonic con-

ditions this also includes the formation of shock waves, but for the flight
conditions of concern here this not important.
• To avoid separated flow at all flight conditions. This objective is questionable

in this general formulation as it is not possible to achieve. A winglet that is
effective at zero sideslip conditions will always stall at high sideslip angles
simply because the angle of attack range of a low aspect ratio wing is only
10-12 degrees. If the winglet is not to stall at 10 degrees of sideslip it has to
operate near α = 2◦ in cruise. In this condition it will not create the required
lift distribution to reduce the drag effectively. It would be better to formulate
this objective as not to stall at any zero sideslip conditions and not to create
any hazards in the stalled condition. Interestingly there is very little published
data on the effects of winglet stall. The only one is in [16] for general aviation
aircraft, where it states that the effects vary significantly with aircraft type.
• to minimise the structural impact of the winglet.
For winglets to work effectively, the tip loads of the wing should be high. The orig-
inal design by Whitcomb was done for a first generation transport wing with nearly
elliptical loading. Much less performance improvement was found for wings with
smaller outboard loadings on newer transport designs [15]. These newer designs
offload the wing tips to reduce the structural weight of the wing, while accepting a
small aerodynamic efficiency loss.
8
Adding a winglet increases the viscous drag of the wing due to the increase in
wetted area. Also interference drag may be produced at the junction to the wing.
Additionally, a winglet may produce a change to the wing pitching moment due to
the location aft of the wing’s aerodynamic centre. This can lead to a higher trim
drag on the tail. These drag increases will reduce the overall efficiency gains of the
winglets.
To design a winglet that achieves all the stated objectives, every source [13][15] etc.
proposes a procedure that starts with a basic design resulting from previous data
and then refine the design by experiments or simulations. As Kroo [1] states, a
winglet design is a highly multidisciplinary approach as several fields are affected
(Aerodynamics, Stability and Control, Structures etc). Due to this complexity there
is no general design that works for all aircraft. Each design has to be optimised to
the specific requirements of the wing design that it will be fitted to.
There was no literature available on winglet design with CFD. All authors use po-
tential flow methods with at best corrections for viscous effects. Therefore many
papers miss out on the important interference effects near the junction [17]. This
can have a significant effect on the final results, with the predictions of efficiency
gains usually too optimistic. Consequently it is necessary to introduce a viscous
solution into the design process and not to rely on inviscid methods alone. This
can be achieved by CFD or testing.
The ‘toolbox’ of the winglet designer contains several factors that can be adjusted
to achieve the above mentioned objectives. These are [13][14][15]:
Planform: This includes the root chord, the taper ratio and the winglet height. As
for the root chord length, these are usually either about 60% of the wing tip
chord or the full tip chord length. The reason for these two distinct cases it
given in [13]. To minimise the interference drag the winglet should begin be-
hind the minimum pressure region of the wing. This is even more important
for high speed cases. On the other hand a discontinuity in the wing chord will
always produce unwanted interference. Thus, more designers today favour
the full wing tip chord length for the root of the winglet. The taper ratio to-
gether with the optional twist should produce a nearly constant normal force
coefficient distribution along the span of the winglet.
The induced drag of a wing-winglet combination decreases nearly linearly with

the winglet height. For a higher winglet, however, larger normal force coeffi-
cients are required. This has a negative effect on the structural weight of the
winglet and the wing bending moment. Thus the height of a winglet is always
a compromise between aerodynamic efficiency and structural penalties. The
original winglet by Whitcomb has the height of the matching tip chord, other
designs are much larger (Boeing 737).
9
A winglet should have the largest possible aspect ratio to be as effective in

generating the loading required as possible. This leads to a shorter chord
than the main wing. According to [12], this possibility of a winglet to have
a narrower chord than the main wing and thus less wetted area, is the main
reason, why a winglet is more effective than a simple span extension with a
large chord.
Aerofoil Section: The wing section should operate at its maximum L/D at de-
sign conditions. Also it must have favourable stalling conditions as the local
Reynolds numbers at the winglet are usually small due to the limited chord
length. On some applications, like Sailplanes, there is also a large variation in
Reynolds number at different flight conditions. For the Merlin III this is not
critical, as shown in Section 4.2.1, because of the exchange between air density
and speed.
Incidence and Twist: The incidence of the winglet has to be set such that it oper-
ates at maximum L/D and does not stall at high lift conditions. This is done
by setting the winglet at an angle with the aircraft centre line (toe in or out)
The amount of lift produced by the wing will determine the strength of the
tip vortex and thus the size of Vy in Figure 2.1. This will then produce the
angle of attack at the winglet. It may be tempting to try and find the angle
of attack from a graph like Figure 2.1 but the subsonic flow conditions will
result in a change in the flow field once the winglet is present as shown in
the upper part of Figure 2.1. This is due to the forward communication of the
pressure changes created by the winglet. Therefore these graphs cannot be
used for this purpose. The actual flow conditions at the winglet can only be
determined by simulations or experiments.
Twist can be used to fine-tune the lift distribution but it is not used in any
of the papers studied. The variation in strength of the wing tip vortex with
increasing distance from the wing tip produces a natural twist [13]. Further
gains by actual winglet twist are usually too small to justify the effort.
Cant Angle: The cant angle is the angle of the winglet with the vertical. It deter-
mines what amount of the winglet’s normal force is generating additional lift
and consequently increases in wing bending moment. It is also important for
stability criteria [16] as the winglet adds to the effective dihedral of the wing.
A standard value for the cant angle is 15 degrees as introduced by Whitcomb.
Sweep Angle: The sweep of the winglet is mainly important for high speed flight
where the critical Mach number of the wing section is reached. A guide is to
sweep the winglet in the same way as the wing. For slow flight a winglet sweep
does not have any positive impact [14]. It is often used for aesthetic reasons.
10
2. Background 2.3. Computational Fluid Dynamics
Junction Design: The junction design determines the amount of interference drag
created because of the interactions of the wing and winglet boundary layers.
This can lead to flow separation. The best way is to blend the wing and the
winglet in a smooth curvature.
The original Whitcomb winglet also featured a lower winglet that creates an out-
board normal force. It decreases the inflow velocities onto the upper winglet and
therefore prevents flow separation [13]. The performance gains are minimal and
such a lower winglet was last used on a Boeing MD-11 [15].
2.2.5
Winglets on General Aviation Aircraft
A limited amount of research literature is available for winglets on general aviation

(GA) aircraft [14][16][18]. They found similar advantages as with the high speed
applications and several differences for the design parameters due to the lower air-
speeds. This mainly concerns the incidence angle and the aerofoil sections. More
details from these papers will be used extensively in the later chapters. Neverthe-
less, winglets are rarely used on GA aircraft. Most likely it is easier and cheaper
to use simpler methods like span extensions as GA aircraft are usually not span
limited. On the other hand winglets are very common on sailplanes where come
class rules for competition limit the span of the wings [19][20].
2.3 Computational Fluid Dynamics

Any fluid flow is governed by three basic principles: Conservation of mass, New-
ton’s second law: F = ma and the conservation of energy. These principles can
be expressed in the Navier-Stokes equations, either in differential or integral form.
Due to the non-linearity of the eqautions no analytical solution exists for these
equations as of today, except for the most basic flows. To solve the equations for
a complicated flow problem, digital computers can be used to perform a technique
called Computational Fluid Dynamics (CFD). It is defined as [21]:
“CFD is the art of replacing the partial differentials or integrals in the

Navier Stokes equations with discretized algebraic forms.1 These equa-
tions are then solved to obtain a set of discrete numbers for the flow field
instead of a analytical solution”
CFD is a complicated discipline with challenges caused by the nature of the physics
involved, the mathematical behaviour of the governing equations and their discrete
counterparts and issues with numerical stability of the algorithms used. Also, the
1
And then trying not to get into too much trouble, one might want to add.
11
iterative nature of the method can make it very time consuming if the user is not
sufficiently experienced and the problem size is large.
2.3.1
Motivation for CFD
The motivation to use CFD over inviscid methods is the ability to predict viscous
effects and flow separation. A CFD solution gives a more detailed picture about
the flow field and shows additional information like stall which may be important
for performance assessment. Also, interference effects at a wing-winglet junction,
like vortex generation or boundary layer separation, can only be predicted with CFD.
An example, how neglecting these effects can change the solution for a complicated
winglet flow will be presented in Chapter 7. On the other hand, CFD contains several
sources of error itself. These errors and their implications are discussed in the next
section.
An overview over the applications of CFD for aerodynamic predictions is given in

[22]. It states that CFD is currently mature enough to perform cost effective sim-
ulations during the conceptual and preliminary design stages, where mainly com-
parative studies are required. For the absolute data in the final design stage a wind
tunnel test series may still be preferable, if only to verify the CFD results with great
precision. Aerodynamic load data can still be predicted more reliably through prac-
tical tests, according to the reference.
2.3.2
Simulation Accuracy
CFD is an approximate technique and thus the results will contain errors. The aim
of each CFD study must be to keep these errors as small as possible and to estimate
their magnitude to judge the quality of the simulation results. To minimise these
errors one has to understand their sources. The CFX manual [23] lists the following
categories:
Numerical Errors: These arise from the differences between the continuous differ-
ential equations and their discretized forms. There are spatial and time errors
for the discretization in space and time, respectively.
Modelling Errors: Caused by the necessary use of models for turbulence (see Sec-
tion 2.4.2) and other processes like combustion. These models simplify pro-
cesses that cannot be solved directly but also introduce uncertainties into the
solution, as they are not an accurate description of the process they model.
Further any discrepancies between the real world geometry and the simulated
geometry will cause errors. Usually the real shape will have to be simplified to
12
be accessible for CFD (see Section 4.2.2). Also, the quality of the CAD models
depends on the available input data (see Section 4.3.1).
User Errors: These errors are caused by an inexperienced or sloppy user, selecting
wrong simulation parameters. This includes the use of inadequate boundary-
and initial conditions.
The last two error sources can be controlled by carefully selecting the simulation
parameters and performing verifications as discussed in Chapter 5. Also, careful
documentation is required to show the solution process. The errors caused by the
geometry differences have to be judged by comparison with another source of data.
The numerical errors are inherent to the CFD method and cannot be fully elimi-
nated. There exist several techniques to estimate and minimise these errors. Some
of them are presented next.
The solution error is defined as [23]
fexact − fnumeric
Es = (2.5)
fexact
where f can be a direct value at a grid node, like pressure, or an integrated solu-
tion parameter like lift or drag. As the exact solution for the problem is unknown
(otherwise there would not be a reason to do the simulation in the first place), it is
difficult to quantify the solution error. Therefore other methods of predicting the
error are required.
The most important numerical errors for drag prediction are the spatial discreti-
sation errors [24]. They arise from the use of numerical approximations for the
analytical derivatives or integrals in the flow equations. These numerical schemes
have a truncation error as it is impractical to use schemes of very high order. Typ-
ically second order schemes are used for high quality CFD solutions. These are
second order accurate as with a truncation error of order two. As an example a
second order central difference scheme can be written as (see [21] for more details)
!
δu δ3 x (∆x)3

ui+1,j − ui−1,j = 2 ∆x + 2 + HOT (2.6)
δx i,j δx 3 i,j 6
Thus the derivative δu/δx can be represented as
δu ui+1,j − ui−1,j

= + O(∆x)2 (2.7)
δx i,j 2∆x
In other words, while the continuous derivatives (or integrals) in the Navier-Stokes
equations depend on the entire flow field, does the discrete representation only
take into account the values at the grid nodes right next to it. Any information
further away is only considered indirectly by the iteration along the grid.
13
These spatial discretisation errors contribute to a phenomenon called numerical

diffusion and are most prominent in 3-d solutions where the grid is not aligned with
the flow direction [23]. This is the reason why structured grids are preferred. This
numerical diffusion changes the behaviour of the diffusion terms in the governing
equations and thereby causes errors. A similar analysis can be performed for the
time discretisation errors.
The most practical method to estimate these errors is to use a series of grid refine-
ments. For unstructured meshes, a global mesh spacing factor is required, which
is used to rescale the spacings in all areas of the domain in a uniform manner. At
least three data points are required with the mesh spacing reduced by a factor of
two each time. This leads to a growth in mesh size by a factor of eight for each
step for a 3-d problem. This, and the fact that the coarsest mesh has to be in the
asymptotic convergence region, presents challenges for the hardware requirements
to solve complex 3-d problems. The typical grid sizes for the AIAA drag prediction
workshops were [2.7, 8, 24] million grid points [24]. By comparison, the largest grid
possible in this project had 2.37 million nodes.
In the asymptotic convergence region ‘the principle concept of grid convergence

states that as the mesh resolution is increased in all areas of the domain, the dis-
cretization error in the solution should vanish and the discretized solution should
gradually tend toward the continuous solution. For a second-order accurate scheme,
where the discretization error varies as O(h2 ) and where h represents some mea-
sure of the mesh spacing, solution values such as lift and drag plotted versus h2
should produce a straight-line plot that extrapolates to the continuous solution
values in the limit as h → 0, provided that the grids are of the same family’ [24].
This principle leads to a definition of a target variable error via the Richardson
extrapolation [23],
|X1 − X2 | 1
Erms = p
(2.8)
|X1 | (r − 1)
where X is the target variable, p is the order of accuracy of the scheme and r can
be written for an unstructured mesh as
1/D
N1

r = (2.9)
N2
with Ni the number of grid points and D the dimension of the problem.
An indicator if the solutions are in the asymptotic range is given as
Erms
Eh = (2.10)
hp
and should be close to constant.
14
2. Background 2.4. Drag Prediction with CFD
These definitions will be used in Section 5.3.1 to estimate the errors in the solutions
for this project.
The other numerical errors are the iteration error determined by the solution resid-
uals and round off errors. These can be controlled by ensuring appropriate residual
reduction during the solution and invoking the double precision solver that uses
two data words to represent the flow variables.
A method for managing CFD errors based on quality management principles is

presented in [25]. It is mainly targeted at the industry, where proper results have
to be delivered within a restricted time frame at predictable cost. The paper is very
instructive but the method is beyond the scope of this project.
Generally the best way of estimating errors is the comparison of the results with an
independent data source as will be done in Chapter 5.
2.4 Drag Prediction with CFD

The prediction of the drag of an aerodynamic body is probably one of the most
difficult tasks in CFD. Drag is caused by several different mechanisms (see Section
2.2.1), some of which require the resolution of very small scales (viscous drag) in
the flow field.
The AIAA has held four drag prediction workshops so far to assess the state of the
art in this field [26]. Numerous papers have been published on the results. The
bottom line is that even with grids of 100 million nodes there is still a substantial
variation in the results. These are not only caused by the grid resolution but also
by the choice of the solver, the turbulence model and the grid type [24].
There is still a large amount of research into this field, with the results for the
prediction of the absolute drag of a body not being satisfactory yet [26]. Better
results are achieved for comparative studies, like this one, where only changes in
drag due to modifications of the geometry are required.
Next, the main factors that create challenges for drag prediction with CFD are dis-
cussed. These are the boundary layers and the turbulent flow.
15
2.4.1
Boundary layers
The boundary layer is the small region near the surface of a wing with a large
velocity gradient, from V = 0 right at the wall due to the no-slip condition, to the
free stream velocity V∞ . The concept of the boundary layer was first introduced by
Ludwig Prandtl in his famous paper as [3]:
“A very satisfactory explanation of the physical processes in the bound-

ary layer between a fluid and a solid could be obtained by the hypothesis
of an adhesion of the fluid to the walls, that is, by the hypothesis of a zero
relative velocity between fluid and wall. If the viscosity was very small
and the fluid path along the wall not too long, the fluid velocity ought to
resume its normal value at a very short distance from the wall. In the thin
transition layer however, the sharp changes of velocity, even with small
coefficient of friction, produce marked results.” Ludwig Prandtl, 1904
This concept means that friction effects are only important in this thin layer near
the surface of the body and not in the entire flow field around it. Also, it can be
shown that the pressure is constant through the boundary layer. This explains the
success of potential flow methods that are only concerned with frictionless flow in
the far field around the body. Using the constant pressure property of the boundary
layer, any pressure distribution computed by these inviscid methods is also valid
on the surface itself.
The boundary layer is a thin layer with high velocity gradients and thus large shear
forces. For this project, the boundary layer is on average only δ = 0.8 mm thick
(on the 1/10th scale model, see Section 4.3.5). In this distance the flow accelerates
from zero to 180 m/s near the leading edge of the wing. Therefore, one of the main
difficulties is to resolve this thin layer with sufficient accuracy to capture its velocity
profile accurately.
As this project is a fully turbulent solution, only this case will be discussed. In
reality, there will be a section of laminar boundary layer near the leading edge,
transitioning into a turbulent one at the transition point. This is very difficult to
model in CFD (see Section 5.2.3). To avoid these complications, the problem has
been simplified to use a fully turbulent flow.
A turbulent boundary layer has three distinct layers as shown in Figure 2.2. Near
the surface is the laminar sublayer up to y + = 5. The parameter y + is a measure
for the distance from the wall and is defined in Section 2.4.2. Next comes the mixing
layer between y + = 5 and y + = 30, where laminar and turbulent flow regions mix.
From y + = 30 to 300 there is the log-layer, named after the shape of the velocity
profile. The egde of the boundary layer is usually defined where the flow velocity
16
Figure 2.2: Structure of a boundary layer. Source: [27], modified
reaches 0.98V∞ . This defines the boundary layer thickness δ. The different sub-
layers are important for the turbulence modelling and meshing discussed later.
The shape of the boundary layer develops along the wing chord due to the exist-
ing adverse pressure gradients and shear stresses. Figure 2.3 shows a preview of
some results calculated in this project along the upper wing surface to illustrate
this. Near the leading edge the speed of the flow is high and the boundary layer
is very thin, as shown in Figure 2.3(a). The logarithmic region can be clearly seen,
starting at the cyan velocity vectors. Further along the upper surface in Figure
2.3(b) the flow slows down due to internal shear and has to work against the ad-
verse pressure gradient. The boundary layer becomes markedly thicker. Near the
trailing edge, Figure 2.3(c) shows that the internal friction and the pressure gra-
dient have almost annihilated the momentum of the flow. The region of very low
velocities becomes very thick, with the total boundary layer extending outside the
structured mesh region. In Figure 2.3(d), the flow close to the surface ran out of
momentum and there is a point with zero velocity (see arrow). Behind this separa-
tion point the adverse pressure gradient causes the flow velocity to invert, creating
a re-circulation zone that can be clearly seen in the Figure. From this discussion we
can see that the boundary layer thickness δ mentioned above is not constant. Any
approximations used to design the near surface mesh are only valid as an average,
with the calculated values found to be close to reality near mid wing for this project
(see Section 4.3.5).
The effects described in this section highlight the difficulty in resolving the bound-
ary layer with sufficient accuracy to capture all the important influences on the drag
of the wing. Even the largest computational meshes used by researchers still strug-
gle in this field, as described in the various papers on the AIAA drag prediction
workshops [24][28][26]. Further, to compute the turbulent properties of the flow
and the boundary layer in particular, models for turbulence are required. These are
discussed next.
17
(a) Near leading edge (b) Mid-wing
(c) Near trailing edge (d) Separation at α = 10◦
Figure 2.3: Boundary layer velocity profiles along the upper wing skin
2.4.2
Turbulence Modeling
Turbulence is described by many authors as one of the last major unsolved prob-
lems in classical physics [3]. Up to today there is no satisfactory theory about the
mechanisms involved. Its inherently random properties pose a difficult problem in
establishing universal laws that could be used to calculate turbulent motion.
Any flow above a Reynolds number of approximately 500,000 is turbulent. Thus,

for all full scale aircraft applications, the flow conditions always involve dealing
with turbulence. As will be shown later, turbulence has several major effects on
the drag of an aircraft. It is therefore necessary to deal with these effects during a
detailed analysis.
18
Wilcox [29] gives the followng definition of turbulence:
“Turbulent fluid motion is an irregular condition of flow in which the

various quantities show a random variation with time and space co-ordinates,
so that statistically average values can be discerned. Turbulence has a
wide range of scales.”
In contrast to laminar motion, turbulence can be described as the motion of locally

swirling eddies with possibly large vorticity. These eddies introduce mixing and
turbulent stresses to the fluid that can be large compared to laminar flow.
Some general properties of turbulence are [29]:
Instability: Turbulence develops from laminar flow through instabilities and oscil-
lations caused by small disturbances like surface imperfections. This is the
mechanism on an aircraft wing, where the flow around the leading edge is
normally laminar and then transitions to turbulent flow at some point where
the instabilities become too large to sustain laminar conditions.
Nonlinearity: Any turbulence is strongly non-linear and 3-dimensional. It is a con-

sequence of interaction of the inertial and viscous terms of the Navier-Stokes
equations, which makes it very difficult to analyse.
Statistical aspects: As turbulence is also time dependent, it can best be described

by statistical methods. For engineering work usually only a time average is
required. This reduces the dependence on exact initial conditions, which are
usually unknown.
Turbulence is continuous: As will be shown later the smallest scales of turbulence

are very small but always much larger than the molecular scale. Discrete in-
teractions between molecules of the fluid do not have a direct effect on the
properties of turbulence.
Turbulence scales: A turbulent flow usually contains eddies of a large range of

sizes. The largest are of the size of the characteristic length of the flow and
the smallest eddies can be described by Kolmogorov’s scales of length, time
and velocity given as
η = (ν 3 /)1/4 τ = (ν/)1/2 v = (ν)1/4 (2.11)
respectively. Wilcox gives an example for the average smallest length scale at
a car moving with 28.8 m/s, where η ≈700 times the molecular diameter.
Turbulence can only be fully described by resolving the full spectrum of scales
involved. This is the reason why today’s computers are orders of magnitude
too slow to perform these calculations, except for the simplest problems. This
discipline is called Direct Numerical Simulation (DNS). Results from DNS cal-
19
culations can be used to benchmark turbulence models but some authors ex-
press concern over the applicability of these comparisons [30].
The large eddies in the turbulent flow carry most of the energy. This turbulent
energy is transferred to the smaller scale eddies where it dissipates into heat
by molecular interactions. These small eddies carry most of the vorticity of
the flow. This leads to two main parameters of turbulence modelling, namely
the turbulent energy k and the dissipation rate , which are defined later.
Enhanced Diffusivity: The enhanced diffusion in turbulent flows increases the trans-
fer of mass, momentum and energy. This is the reason for the increase in drag
in turbulent flows, as compared to laminar flows.
For an in-depth coverage of this topic, the reader is referred to reference [29].
From the above discussion it becomes clear that, to predict turbulent flows with nu-
merical methods, turbulence models are required. These models reduce the com-
plexity of the problem to a scale that can be handled by today’s computers. To
introduce some of the models and techniques used for aerodynamics, some com-
mon symbols need to be defined:
Turbulence kinetic energy k [m2 s2 ]: A measurement for the energy that is car-
ried by the large scale eddies. This energy is transferred to smaller eddies at
a rate where is dissipates into heat by molecular viscosity. An estimate for
setting the solution boundary conditions is
3
k≈ 2 (Vav I)2 (2.12)
where Vav is the average flow velocity and I the turbulence intensity.
Turbulence dissipation rate [m2 s3 ]: The energy transfer rate from large eddies
to smallest eddies. Formally defined as
dk
=− (2.13)
dt
it can be estimated to establish the boundary conditions as
k3/2
≈ (2.14)
l
Turbulence eddy frequency ω [s−1 ]: No precise physical definition is available

for the meaning of this parameter, which was derived from dimensional anal-
ysis.
20
Wilcox [29] cites the original developer of the k − ω model, Kolmogorov, as:
“omega is some mean frequency determined by ω = ck1/2 /l, where c is a
constant.” An estimate for ω to set the boundary conditions is
k1/2
ω≈ 1/4 (2.15)
Cµ l
where Cµ or β∗ is an empirical constant, normally given as 0.09 [30].
Turbulence length scale l [m]: A measurement for the largest scale of eddies. For
external flow problems this is usually the characteristic length of the flow,
namely, the mean aerodynamic chord for wings.
Turbulence intensity I: A measurement for the free stream turbulence intensity

of the flow. Directly related to k as seen above. For aircraft aerodynamics this
is usually small as it is assumed the aircraft will fly into still air. For wind
tunnel testing this is a criterion of the quality of the tunnel, as one tries to
approximate still air conditions as well as possible.
Wall distance y + : The dimension-less distance of mesh cells normal to the surface
of a obstacle in the flow. It is relevant for a turbulence model to be able to
predict the viscous drag accurately. It is defined as
ut y
y+ = (2.16)
ν
p
with ut = τw /ρ being the friction velocity (τw is the wall shear stress), y the
wall distance and ν the kinematic viscosity. The value of y + depends on the
local flow conditions and can only be estimated before the solution.
These definitions will be used later to set up the CFD simulation in Section 4.3.5
and Section 4.4.1.
The majority of the turbulence models used today in CFD are based on empirical
formulations. These are often derived by dimensional analysis. The turbulence
models used in this project are from the ‘eddy viscosity’ class. They use a mod-
ification to the governing flow equations called Reyonlds Average Navier Stokes
equations (RANS). These equations contain a statistical process of averaging the
flow characteristics, to account for the fluctuating nature of the turbulent flow.
The RANS equations make the turbulence more accessible as they contain only the
mean values of all variables plus a new set of unknowns, called the ‘Reynolds stress
tensor’. This tensor adds 6 unknowns to the four basic variables (pressure and the
three velocities u, v, w). The RANS equations provide four equations (mass con-
servation and three velocity components). This makes it impossible to solve the
problem, as the number of unknowns is larger than the number of equations. Fur-
ther equations are required to proceed. These additional equations come from the
turbulence models to close the system.
21
The most commonly used turbulence models for aerodynamics are the two-equation
models k − and k − ω. Both contain an additional transport equation for k, and
a further equation for or ω depending on the model. The remaining 4 unknowns
are solved by supporting equations and constants that were derived from empirical
work or dimensional analysis. Next some details for the three turbulence models
considered for this project are given:
k − model: This model was quasi-standard during the 90s. It is quite robust but
does not perform well inside the boundary layer. Complicated, non-linear
damping functions are required to achieve proper results [23]. These func-
tions add errors to the results of the model and render it less precise in pre-
dicting viscous drag effects.
k − ω model: The k − ω model overcomes this problem by using different assump-

tions about the eddy viscosity. It works well inside the boundary layer and is
very robust but has a major disadvantage in being sensitive to the free stream
value of ω. This is usually specified by the user as the inlet boundary condi-
tion and may not be known for the flow to analyse, given the sketchy definition
from above. Therefore, if ω is not specified correctly, the final solution may
contain significant errors [23],[30]. Recently, Wilcox [29] improved the k − ω
model with respect to this boundary condition dependence, but the current
CFD codes still contain the older, flawed version [23].
k − ω SST model: The shear stress transport (SST) model was developed by F. R.
Menter. A complete description can be found in his paper [30]. The important
aspects for this project are as follows:
The SST model was developed in two stages. Firstly, a so called base line model
(BSL) was introduced to combine the advantages of the a k − ω and a k − for-
mulation. A blending function F1 is introduced to use the k − ω model in the
boundary layer and the k − model in the free stream. This blending function
takes a value of one close to the walls and reduces to zero outside the loga-
rithmic region of the boundary layer. This BSL model shows robust behaviour
without the dependency on the free-stream value of ω. Despite this, the BSL
model still does not correctly predict the flow separation under adverse pres-
sure gradients, because it does not model the effects of the turbulent shear
stress transport. Therefore, further additions were introduced by Menter to
create the SST turbulence model.
The conventional two equation models from above over-predict the turbulent
shear stress τ in regions of adverse pressure gradients because of their initial
assumptions. To account for this, the SST model introduces a new formula-
tion for the eddy viscosity. It features a second blending function F2 . This
additional blending function takes the value of one inside the boundary layer
and zero in the free stream. It blends the new eddy viscosity formulation with
22
the standard formulation, vt = k/ω, in a similar way the BSL model combined
the two turbulence models. Plots of the two blending functions are shown and
discussed in Section 5.3.3.
The SST model has shown superior performance in various experiments, see [23]
and [30], and is known as a very robust turbulence model for external aerodynam-
ics. It will be used exclusively during this project.
For all turbulence models described above to work properly, it is necessary to re-
solve the boundary layer structure (see Figure 2.2) with a certain accuracy. This re-
sults in demanding meshing requirements (see Section 2.4.3). If these requirements
cannot be met due to hardware restrictions or other limitations, mathematical mod-
els of the boundary layer can be used. These are called ‘Wall Functions’ and are an
integral part of most turbulence model formulations [23]. CFX has a automatic wall
function for the SST model, that automatically determines whether the quality of
the boundary layer resolution is adequate. It then automatically switches between
the use of the wall function or the direct solution. The governing parameter for this
is the value of y + . More information on these wall functions can be found in [31].
Reference [32] presents an outlook into the future of predicting turbulent boundary
layers. It shows, how a more advanced turbulence modelling technique called ‘Large
Eddy Simulation (LES)’ can be used inside the boundary layer and a two equation
model in the free stream. LES is too hardware intensive to be used solely for large
problems yet, but this combination approach might prove a step forward towards
better prediction accuracy.
2.4.3
Meshing Requirements
The two previous sections on boundary layers and turbulence modelling define the
requirements for the computational meshes used for the CFD simulations. These
requirements are quite demanding in terms of the hardware needed to solve the re-
sulting grids. There are several best practise guidelines on how to create a suitable
mesh for aerodynamics computations, that are based on previous experience and
requirements of the methods used.
The guidelines for the grid generation at the AIAA drag prediction workshops are
given in [24]. These were used as a guide in this project although some of the
requirements could not be achieved due to hardware restrictions. Also, some of
the guidelines are only necessary for the transonic flight conditions used in the
workshop. The guidelines are:
• y + < 1 for the boundary layer, also, two layers of constant spacing next to the
wall and then a growth rate of 1.25 or less.
23
2. Background 2.5. Summary
• Domain walls 100 chord lengths away from the model
• Chord-wise surface mesh spacings 0.1% of chord at leading- and trailing edge
of the wing
• At least 8 cells on the blunt trailing edge
• Create a family of grids growing 3 times with each refinement for convergence
studies
An additional requirement is the boundary layer resolution as given in [23]. For the
use of the wall functions, at least 10 cells must be in the boundary layer and the
value for y + must be less than 50. To avoid the wall functions and use the full
boundary layer resolution models, at least 15 cells are required in the boundary
layer in addition to the requirement y + ≤ 1 from above.
2.5 Summary
This chapter introduced the theoretical concepts required for this project, based
on the available literature. Firstly, aircraft drag was discussed, with emphasis on
the lift induced drag component. Some basic ideas on methods to reduce this
component were given. Next, winglets were introduced, illustrating their working
principles and methods for designing them. The second part of the chapter con-
tained theory on CFD methods, concentrating on the prediction of solution errors
and the factors important for drag prediction. Of those, boundary layers and tur-
bulence models were discussed in detail. Finally, the requirements for a suitable
computational mesh, as proposed by literature, were given.
24
3
Problem Description
3.1 The Merlin III Aircraft

The aircraft is a twin engine executive transport, originally designed by Svearingen,
then Fairchild and now M7. It is driven by two turboprop engines driving four
bladed propellers. The design is a shortened version of the popular Metro twin
turboprop transport. It shares the same wing structure as the larger Metro which
has a outboard extension of 1.82m to cater for the higher weight.
The fuselage is a semimonocoque structure. An air conditioning and pressurization

system conditions and pressurizes the cockpit, passenger cabin, and aft baggage
compartment. The aircraft is equipped with an all-movable horizontal tail. The
tricycle landing gear has steerable nose wheels. The internal configuration in the
standard arrangement accommodates six to nine passengers and a crew of two.
Some basic data from the flight manual are shown in Table 3.1 (more details below
for the wing). A 3-view is shown in Figure 3.2 together with a photo of the aircraft.
The aircraft with the serial number 213 is currently being restored and upgraded
at Winrye Aviation. It receives more modern engines and other modifications from
later models of the Merlin family.
Dimensions/Weights: Performance:
Length: 12.85m Engine shaft HP: 900
Wingspan: 14.1m Fuel capacity: 2460 l Jet A-1
Height: 5.13m Stall speed clean: 103kn CAS
MTOW: 5670kg Approach speed (flaps down): 116kn CAS
Zero Fuel: 4762kg Cruise speed: 250kn
Max. Landing weight: 5670kg Max. altitude: 9448m
Empty: 3447kg Max. crosswind component: 22kn
Table 3.1: Merlin III data
25
3. Problem Description 3.1. The Merlin III Aircraft
Figure 3.1: Fairchild Merlin III VH-SSM
Figure 3.2: Fairchild Merlin III 3-view, Source:

http://members.lycos.fr/wings2/3views/merlin3v.jpg
26
3. Problem Description 3.2. Wing Properties
3.2 Wing Properties

The wing is a standard metal design of trapezoid planform. It features a taper
ratio of 0.45 and a small leading edge sweep. The wing is twisted slightly for some
tip washout and uses two different aerofoils at root and tip. The high lift devices
are double slotted flaps along 60% of the span. The flight manual only lists the
total wing planform area. As this project will only simulate the portion of the wing
outside the fuselage, all dimension-less coefficients will be based on the area of this
smaller wing. The wing area S of a simple tapered wing is given by
croot + ctip

S=b (3.1)
2
This area is the projected area onto the XY plane. There is a small difference due to
the dihedral. This is ignored for this project.
To calculate the Reynolds number and other required flow parameters, the charac-
teristic length L of the flow is required. For a non-rectangular wing this is usually
the mean aerodynamic chord (MAC). It can be computed from
" #
2 1 + λ + λ2
MAC = cr oot (3.2)
3 1+λ
for a given taper ratio λ. This equation is independent of the wing sweep. The wing
data is summarized in Table 3.2 using data from the flight manual and measure-
ments of the aircraft. A drawing of the wing model is shown in Figure 4.2, which
also includes the the simplifications made for this project as described in Section
4.2.2.
Root section: NACA 652 A215 Dihedral: 5◦

Root chord: 2440 mm Root incidence: +1◦
Tip section: NACA 642 A415 mod. Tip incidence: −1◦
Tip chord: 1090 mm Quarter chord sweep Λ: 0◦ 5400
Half span: 6368 mm Taper ratio λ: 0.447
Half wing area: 11.24 m2 MAC: 1851 mm
Aspect ratio: 7.216
Table 3.2: Wing data for the wing outboard of the fuselage joint
3.3 Wing Tip Properties

The wing tips are a modified Hoerner design [33]. It is characterized by tapering
the wing thickness into a sharp edge, while keeping the upper surface of the wing
27
3. Problem Description 3.3. Wing Tip Properties
(a) Photograph of the Hoerner wing tip installed on VH-SSM
(b) Rendering of the tip model
Figure 3.3: Hoerner Tip Geometry (Model 1/10th scale)
section unchanged. Figure 3.3 shows the geometry. The original Hoerner design
was also tapered spanwinse with the maximum span near the quarter chord station.
This has been omitted by the designers of the Merlin III wing. The tip has a span of
300mm from the most outboard rib in the wing, where it is attached.
In Hoerner’s paper [33] and the book [34], there is no mention of any advantage of
this wing tip design over a simple square tip, where the aerofoil is continued to the
tip and then cut of square. Therefore the question is, why such a complicated tip
design was used for this aeroplane (and for many other designs). To investigate this,
a square tip has been tested and will be compared to the results for the Hoerner tip
in Section 6.2.
28
3. Problem Description 3.4. Winglet Properties
3.4 Winglet Properties

The winglet is a relatively small, highly swept design. It does not have a defined
aerofoil section, the section is best described as a flat plate that is thicker in the
front with a rounded leading edge and a blunt trailing edge 10mm wide. The winglet
is modelled onto the existing Hoerner tip and made from fiberglass. The proto-
type is made from two surface shells that are joined along the leading edge of
the winglet. It does not have any internal structure yet, but a large wall thickness
renders the structure very sturdy.
Table 3.3 lists the dimensions that were measured from the prototype shown in
Figs. 3.4 and 3.5 as described in Section 4.3.1. Figure 3.6 shows the geometry of
the winglet model, including the sections along the span.
Section: ≈ flat plate Cant angle: 10◦

Root chord: 530 mm Root toe angle: 0◦
Tip chord: 180 mm Tip toe angle: −3◦
Span: 450 mm Leading edge sweep Λ: 59◦
Area: 0.14 m2 Taper ratio λ: 0.34
Aspect ratio: 1.4 MAC: 383.8 mm
Table 3.3: Winglet data
Figure 3.4: Original winglet prototype (Position light cap missing)
29
3. Problem Description 3.4. Winglet Properties
Figure 3.5: Original winglet design modelled onto a Hoerner tip (Position
light cap missing)
(a) (b)
(c)
Figure 3.6: Original Winglet Geometry (Model 1/10th scale)
30
4
Experiment Setup
4.1 Introduction
This chapter describes the setup of the experiments performed in this study. They
will use CFD to simulate the flow around the Merlin III wing in isolation from the
fuselage to keep the hardware requirements within reasonable bounds. The dif-
ferent wing tips will be mounted to the wing. Then their influence on the lift and
drag characteristics of the wing with changing angle of attack and sideslip will be
measured.
To set up the CFD solution, firstly suitable CAD models for the structures are re-
quired. Then a computational grid has to be generated from these geometries and
the CFD solver needs to be configured appropriately. For this task, the flight con-
ditions for the simulations need to be known. Finally, the quality of these meshes
has to be benchmarked to ensure proper solution convergence.
This author encountered many difficulties in the process, that could have been
avoided if suitable literature and guidelines would have been available. Therefore,
to assist future students working on a similar problem, all steps necessary to set
up a successful simulation will be described in detail (sometimes maybe with more
than necessary detail on specific problems, where assumed instructive). It is hoped
that this will be of use for a prospective reader.
4.2 Common settings

The settings and assumptions discussed in this first section apply to both the CFD
solution as well as the inviscid vortex lattice methods used for this analysis. These
include the estimates for the flight conditions of concern and the simplifications to
the geometry, that were necessary for a successful simulation.
31
4. Experiment Setup 4.2. Common settings
4.2.1
Flight Conditions
There are two important flight conditions for the evaluation of winglets. The first
is the cruise condition where minimum drag is desired. The second is the landing
(or take off) condition where high lift coefficients and possible sideslip may cause
unwanted flow separation on the winglet.
Thanks to Winrye Aviation the flight manual for the Merlin III was available to this
author. It contains a sample flight to demonstrate the performance calculations for
the type. This data was used to define the two conditions.
Cruise The flight manual sample flight defines a long range cruise condition
(several are possible: maximum range, maximum endurance and so on) as:
• Altitude: 25000ft = 7620m
• Airspeed: 254kn TAS = 130m/s = M0.42
• Aircraft weight at cruise: on average 11000lb = 5000kg
• Air density at altitude = 0.5489 kg/m3
This results in a required lift coefficient
L 5000 × 9.81
CL,cruise = = = 0.41 (4.1)
qS 0.5 × 0.5489 × 1302 × 25.78
From lift curve graph in Figure 6.4(a), obtained during the analysis, the required
angle of attack is about 3 degrees. The resulting flight Reynolds number is 8 million,
based on the mean aerodynamic chord.
Landing The flight manual sample flight defines a heavy landing (full MTOW)
condition as:
• Altitude: 0ft = 0m
• Airspeed: 116kn TAS = 59.67m/s
• Aircraft weight at landing: 12500lb = 5669.9kg
• Air density at sea level = 1.225 kg/m3
The required lift coefficient computes as
L 5669.9 × 9.81
CL,land = = = 0.99 (4.2)
qS 0.5 × 1.225 × 59.672 × 25.78
32
The landing Reynolds number is 7.6 million, based on the MAC, which is similar
to the cruise Reynolds number due to the exchange between airspeed and density.
This high lift coefficient can only be achieved using flaps. A simulation of the flaps
is beyond the scope of this project, so a simpler approximation had to be used to
estimate the required angle of attack with flaps extended. A method outlined in
reference [2], based on experimental data obtained by NACA, and some wing sec-
tion data from [35] will be used. The section data shows the change in lift curve
for a NACA 653 − 118 aerofoil with a double slotted flap, similar to the arrange-
ment found on the Merlin III. The aerofoil section should be similar enough for this
estimate, as it is only 3% thicker than the aerofoil used on the Merlin III.
A trailing edge flap simply shifts the lift curve of the aerofoil upwards without
changing the slope a2d . The resulting change in lift curve slope for the 3-d wing
can be calculated from lifting line theory as
a2d
a3d = (4.3)
1 + (a2d /π AR)(1 + τ)
where τ is the correction factor for a non elliptical lift distribution. From this
equation we can see that the parallel shift of the 2-d lift curve by the flaps will also
shift the 3-d curve in a similar way as a2d does not change. τ, however, will be
affected by the change in lift distribution due to the flaps and introduce a small
change to the slope of a3d . This, and the possible increase of the wing area by the
flaps, will be ignored for this estimate, because the resulting changes to the landing
angle of attack will be in the order of ±1◦ . This is not significant for this project.
To estimate the amount of parallel shift of the 3-d lift curve, the angle of zero lift
is required. For an un-flapped 3-d wing this angle is approximately the same as for
the 2-d section. Small corrections may be necessary due to the twist of the wing
but will be neglected here. The change in α0 for the flapped wing can be written as
∆α0 = −K∆CL (4.4)
where K is a constant given in graphs in reference [2] and ∆CL is the change in
maximum section CL due to the flaps.
Thus, from [35] for a landing flap deflection of 36◦ the increase in maximum CL is
approximately 1.25. The value for K is determined as 7. This results in a change of
α0,f of −8.75◦
Now, the wing is only flapped for 60% of its span. An average of the two parts of
the wing can be found from
α¯0 = 0.6α0,f + 0.4α0,clean = −(0.6 × 10.75 + 0.4 × 2) = −7.25◦ (4.5)
33
1.4
1.2
0.8
CL
0.6
0.4 Plain wing

Flaps extended
0.2
0
−2 0 2 4 6 8 10
α
Figure 4.1: Estimate of the effect of flaps on wing CL
This corresponds to a change in CL,0 of 0.41. The result is plotted in Figure 4.1,
using the lift curve from an inviscid solution for the wing (as calculated in Section
5.2.2). We find, that for the landing CL of 0.99 an angle of attack of about 5.5
degrees is required. This appears to be a sensible but slightly optimistic estimate.
It will be shown later that the wing experiences the beginning of flow separation at
α = 8◦ , with noticeable loss of lift above α = 10◦ . For the course of this project an
angle of attack of 6 degrees will be used for the landing case.
Simulation The flight Reynolds numbers of about 8 million are too large to be
simulated with the mesh sizes possible. The boundary layer becomes thinner with
increasing Reynolds number (see Section 4.3.5) and with that the requirements for
the near wall mesh increase, too. Some calculations at the correct Reynolds number
will be performed for verification but these runs did not converge fully.
The largest model that could be used was 1/10th scale. This would reduce Re by
a factor of ten. To maximise the available Reynolds number the simulation flight
condition was chosen to be a combination of cruise and landing. The speed used
is 125m/s at standard sea level conditions. This equals M=0.36 and the resulting
Reynold’s number based on the MAC is 1.58 million. The Mach number is slightly
inside the compressible region (M > 0.3) but any errors will be cancelled during
the comparison of the cases. No changes to the flow field are expected as the
Mach number is not high enough to generate shocks (see also Section 4.4). The
local Reynolds number on the winglet at these conditions is ≈ 330000, based on its
mean aerodynamic chord.
34
Figure 4.2: Merlin III wing
4.2.2
Problem Simplifications
To keep the mesh size within reasonable bounds, the wing of the Merlin III had to
be simplified. Firstly, only the wing outboard of the fuselage could be simulated.
The fuselage or the tail would have added too many mesh cells and most likely also
complications due to the wing-fuselage interference. This study only compares the
performances of different wing tip designs. Therefore the removal of the fuselage
appears to be an adequate simplification, as it has only a minor effect onto the wing
tips
For the same reasons the engine nacelle had to be removed, too. The main effects of
the nacelles would have been the propeller slipstream over the wing and an increase
in wetted surface area. Also the edges where the nacelle intersects with the wing
could have added more interference effects. These interference effects can lead to
areas of flow separation, which is problematic to solve with currently available CFD
techniques [24].
Also, a small fence on the leading edge near the tip was removed (see Figure 3.3).
This fence was installed to prevent the position lights from blinding the pilots and
not for aerodynamic reasons. It will have a small effect on the flow field at the
tip but it introduced major complications with the boundary layer mesh in its sur-
roundings. To improve the overall mesh quality, this fence was removed. The final
wing model used for this project is shown in Figure 4.2.
35
4. Experiment Setup 4.3. CFD - Pre-Processing
4.3 CFD - Pre-Processing

The CFD pre-processing is often the most difficult and time consuming part of any
CFD work. The process involves the creation of the geometry using a CAD system,
importing and meshing the geometry in the meshing application and finally creat-
ing the solver input file with a suitable pre-processing software. We immediately
see, that these steps involve multiple data conversions between the at least three
different applications. In this study, almost all complications arose from incompat-
ibilities between the programs.
The original plan was to use SOLIDWORKS to generate the geometry, simply be-
cause of this author’s familiarity with this program. Then Ansys ICEM would be
used to create the meshes. ICEM is described in the literature [26] as the most pow-
erful and versatile meshing application and contains a wealth of tools to process
geometry and meshes. Ansys FLUENT was initially chosen as the solver, again due
to familiarity with the program from previous work.
As it turns out, the geometries required for this study are very complex and it
presented quite a challenge to just model them precisely in SOLIDWORKS. On top
of that does the CFD meshing impose stringent restrictions on the quality of the
geometry created. The meshing applications require clean surfaces without any
gaps or holes. So, while a model is completely valid for use in manufacturing or for
illustrations it may not have the quality for meshing. A lot of time was spend to
create an accurate representation of the wing tips that also fulfilled the geometry
quality criteria.
Unfortunately did ICEM not accept any of the files created by SOLIDWORKS to be
used with the meshing mode necessary for CFD (other modes, like Octree, more ap-
propriate for finite element modeling, worked but did not produce valid meshes for
CFD). Luckily for this author, a new version of Ansys Workbench (Version 12) be-
came available during the project and its meshing application now contains several
of the ICEM methods. Workbench accepts geometries of much lower quality than
ICEM, which solved a lot of the problems with the file transfer. Sadly it does not
yet contain any of the mesh quality improvement tools from ICEM, so that a mesh
created in Workbench cannot be improved any further. This functionality was often
missed in the later stages of the project, but the working geometry import and an
acceptable mesh quality outweighed these concerns.
Finally FLUENT was abandoned for Ansys CFX as CFX appears to be able to handle
meshes of lower quality, with the mesh expansion ratio (how fast the cells grow
away from the wing surface) being the major issue. As will be discussed later, the
transition from the structured boundary layer mesh to the unstructured tetrahedral
volume mesh posed the main difficulty for FLUENT at the grid sizes used.
36
4.3.1
Models
The wing models were created as surface models in Solidworks. The Hoerner tip
and the winglet tip were physically available for measurement, thanks to Winrye
Aviation. Bot shapes are quite complex and the optimal way would have been a
3-d digital scan to represent them accurately in the CAD environment. A scan
was quoted at about $2000 which was beyond the budget available. Therefore, a
different method of creating the models had to be invented.
For the Hoerner tip it proved to be acceptable to determine 10 chordwise cross-

sections by using cardboard templates at every 10% of the chord line. Some engi-
neers still refer to this method as CAD (Cardboard Aided Design), as it is a quick
and reasonably accurate procedure. The cross-sections were drawn in Solidworks
as splines and a chordwise loft created the surface as shown in Figure 3.3.
The tip wing section of the Merlin III is listed in [36] as a NACA 64A415. This
information is either incorrect or during manufacturing of the wing tip it became
something slightly different. In any case, it was impossible to match the dimensions
obtained from the wing tip with this or any other similar aerofoil section, even by
deforming the flexible skin of the wing tip. As the exact section at the tip of the wing
will only have a small effect on the overall performance, and for this comparative
study it will be the same in each case, it was decided to simply use the section that
was measured from the wing tip. A full scale drawing was created and measured
for modelling in Solidworks. The wing was then lofted spanwise using the NACA
65A215 root section as specified for the aircraft.
The shape to the trailing edge varies along the wing span due to the different control
surfaces. The flaps have a sharp trailing edge, whereas the ailerons and the wing tip
have a 7 mm high blunt trailing edge. A blunt trailing edge is easier to mesh because
the wing section does not end in a single point, but the drag characteristics can be
difficult to predict at the possible mesh size. Nevertheless, some researchers have
successfully used a coarse resolution of the trailing edge [28] and therefore a 7 mm
blunt trailing edge was used on the entire wing to avoid issues with transitions from
one shape to another.
The winglet was measured with the same procedure in spanwise, horizontal sec-
tions. Creating this model clearly showed the limitations of what SOLIDWORKS can
do, especially at the wing-winglet joint. Several different methods were tried to
create this fillet in a tedious process 1 . Finally, a method was found that created a
geomerty which was similar to the real design and that would import into the mesh-
ing software. The emphasis, however, must be on similar as the program features
did not allow to fully control the shape of this joint. In retrospective, it would have
1
Murphy’s law’s [37] universal applicability was proven here beyond reasonable doubt.
37
Figure 4.3: Standard Domain Dimensions in [m]
been helpful to use a different program for geometry creation like Rhinoceros, that
is better suited for complex surface modelling.
Any additional model was a variation of the two discussed above and will not be
described in detail. All models were created full scale to avoid measurement con-
version errors.
Finally, the meshing software requires the domain box to be in the model, so a
surface cube was created as shown in Figure 4.3 and merged with the wing model2 .
The geometry was then ready to be transferred to the meshing software.
4.3.2
Geometry Import
In Ansys Workbench the geometries are handled inside the ‘Design Modeller’. It
allows to import different file formats and to perform some manipulations, like
scaling. Simple geometries can also be created directly with this application. The
Parasolid format was used for file transfer. It has proven to be robust for the shapes
involved in this project.
The meshing software required the domain to be a solid with the wing model sub-
tracted from this block. Solidworks was not able to perform this operation although
2
The intersection of the wing root with the symmetry plane was created with surface trim opera-
tions. This turned out to be the main problem while using ICEM, where mysterious gaps were present
at this joint that caused the mesher to fail despite a successful geometry check. The reason for this
could not be determined during this project. Later in the project, it was found that the SOLIDWORKS
feature ‘knit’, used to merge the surfaces, is actually responsible for the bad surface quality
38
it has a function called ‘Surface Knit’ with that option. Luckily the Design Modeller
can do this using a ‘Sew’ operation. This was one of the most crucial points in
the entire project. If this sew function would have failed in the same way as the
Solidworks function, it would have been impossible to continue with this software
combination. This also shows the problem with today’s CAD applications. None of
the programs available are truly universal and it may be necessary to have several
different ones available to solve a problem.
The two final steps were scaling the geometry by 1/10th to create the right size
for the mesh generation and define the domain as a fluid region. The setup was
then ready for the mesh application. Workbench automatically transfers the data
between its modules so that no further problems with file conversions occur. This
is a very convenient feature.
4.3.3
Mesh Generation
The only method to fulfil the meshing guidelines from Section 2.4.3 with limited
computational resources is the use of a hybrid mesh, consisting of an unstructured
surface- and volume mesh and a structured boundary layer mesh near the walls.
This technique is widely used in the CFD community, although the change in mesh
topology introduces uncertainties that are hard to quantify [38]. These errors are
smaller, however, than the errors caused by a insufficient boundary layer resolution
and must be accepted. A different approach would be to use a fully structured mesh
with only hexahedral elements. These meshes offer the best accuracy, but are very
complicated to generate [26]. Without any experience in this field, this option was
not possible in the timeframe available.
The advantages of the hybrid mesh are:
• The surface mesh uses triangular elements that can easily represent compli-
cated surface geometries. This mesh is generated first.
• The boundary layer mesh uses hexahedral (prism) elements to create very thin
copies of the surface triangles into space away from the surface. This is often
referred to as inflation. The result are elements with a very high aspect ratio,
they are very long and thin. This allows cells near the surface without exces-
sive surface mesh sizes. Depending on the software used, this mesh is created
before or after the unstructured volume mesh. For this project, is was set up
before the volume mesh for greater quality.
• The remaining volume is filled with an unstructured mesh using tetrahedral

elements. These start from the surface triangles of the inflation layers up to
the domain walls.
39
Figure 4.4: Mesh overview at the symmetry plane
The proportion of prism elements to hexahedral elements shows the demanding

requirements for the boundary layer mesh. From Table 4.1 we find that more than
60% of all cells are within the first 5 mm of the wing surface. Figure 4.4 shows an
overview of the mesh near the wing root.
A major concern is the interface between the prism mesh and the tetrahedral vol-
ume mesh. The increase in cell volume cannot be too large for the solution algo-
rithms to work properly. Thus a compromise between the number of prism layers
and their expansion ratio against the resulting mesh size has to be found. Also,
some solvers tolerate a larger jump at this interface than others. This is the most
probable reason why FLUENT did not work but CFX does, as they use different al-
gorithms (node centered vs. cell centered) for the solution.
The generation of a high quality mesh is most important, as the accuracy and sta-
bility of the solution depend on this. Therefore, a lot of time was spend on this
topic. The process is mainly by iteration, starting with a ‘good practise’ mesh and
then refining it, based on the results of preliminary solutions. It took two months
for the first working mesh with sensible results and then two weeks of refinement.
4.3.4
Surface Mesh
The triangular surface mesh is the basis for the volume mesh and was created first.
It determines the resolution of the surface features like leading edge curvature.
40
(a) Surface mesh sections
(b) Final mesh
Figure 4.5: Surface mesh overview for the standard wing
The size of the surface mesh is defining the final mesh size, so it has to be carefully
designed to fit the hardware constraints.
The general method is to create a fine mesh at areas where large flow gradients
occur and use larger mesh sizes in the other regions. For an aircraft wing these
are mainly the leading and trailing edges, as well as the tip. As the surface mesh
size is set in Ansys Meshing by defining the size on a surface (compared to sizes on
the surrounding curves in other programs), the wing was split into several sections
in SOLIDWORKS as shown in Figure 4.5(a). Each of these sections then received
its own sizing, growing from the leading- and trailing edges to the middle of the
wing. Virtual geometry was used on the winglet to fix some bad surface edges at
the wing-winglet juncture.
The meshing guidelines from Section 2.4.3 call for a mesh size of 0.1% local chord
at the leading- and trailing edge. In this project this was simplified to 0.1% of the
MAC to be able to handle this requirement with the options given by the meshing
application. None of the available papers specified the maximum size on the tip
and in the middle of the wing, but several papers contained images of the meshes
41
Figure 4.6: Surface mesh details
[28][24] etc. These images were used to establish a size relationship between the
smallest and largest mesh sizes.
Originally the sections ‘LE’ and ‘p0’ in Figure 4.5(a) were a single unit. It was, how-
ever, quickly established that the size requirements were too demanding for the
available memory size. On the other hand, researchers at NASA have created a
mesh similar to the size possible for this project and published details in [28]. They
describe their mesh as a more realistic example for a current industrial application
and specify a leading edge spacing of 0.5%, a trailing edge spacing of 0.33% and 2-4
cells across the blunt trailing edge. By splitting the leading edge section as shown
in the Figure, it was possible to achieve all these guidelines. The final surface mesh
is shown in Figure 4.5(b).
The Hoerner tip created a few challenges due to its small curvatures, as shown
in Figure 4.6(b), and the transition from a round side edge into a blunt trailing
edge. On the real aircraft the tip has a round trailing edge and a discontinuity
between the aileron and the tip, but this would have been even harder to model.
To improve the mesh quality, the blunt trailing edge was extended to the wing tip
as shown in Figure 4.6(a). A very fine mesh had to be used in these areas after it
was established that especially the trailing edge section of the tip was critical for
solution convergence. The remaining wing tip was meshed as small as possible to
capture the complicated flow in this area. A detailed list of all mesh settings is given
in Appendix A. The mesh size for the Hoerner tip was kept below the maximum
possible cell count to have some reserves when adding the extra surfaces of the
winglets.
42
One main goal during the project was to keep the meshes as similar as possible to
be able to compare their results. The possible mesh sizes were not fine enough to
expect the solutions to be mesh independent. Therefore, the results depend on the
mesh topology and big changes here will introduce errors into the comparison. To
achieve this mesh similarity, all settings on the main wing were kept constant for
all cases. The only changes occured near the winglet junction and on the winglet
itself.
Figure 4.6(c) shows the surface mesh on the original winglet. The leading- and
trailing edges were meshed as fine as the remaining memory allowed. Additionally,
the tip of the winglet was found to be critical for convergence and had to be re-
meshed very finely. Finally, the main surfaces of the winglet received the same
mesh size as the wing tip. These settings are also listed in Appendix A.
4.3.5
Prism Boundary Layer Mesh
After completing the surface mesh, the boundary layer mesh was designed. As
mentioned in Section 4.3.3, about 60% of all cells in the mesh will be within the first
5mm of the wing surface. Therefore, it was critical to find a prism mesh setting
that would fulfil the mesh requirements of Section 2.4.3, as well as keep the mesh
size within reasonable bounds.
The requirements state a value of y + ≤ 1 and call for 10-15 cells in the boundary
layer. To achieve these objectives, the meshing software has three parameters
• Initial layer height
• Expansion ratio
• Number of prism layers
available to configure the boundary layer mesh. To use these settings, the mesh
requirements have to be converted into the dimensions of the actual boundary
layer expected for this flow. The CFX manual [23] contains some approximations to
convert a y + value into the required spacing of the first layer off the walls and to
determine the boundary layer thickness δ. These yield with the data for the 1/10th
scale model:
43
−3
x 10
5
4.5
3.5
2.5
1.5
0.5
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Figure 4.7: Matlab tool for boundary layer mesh settings
√ − 13
∆y = L ∆y + 74 ReL 14 (4.6)
= 2.68e−6 m
−1/7
δ = 0.035 L ReL (4.7)
= 8.2e−4 m
where L is the characteristic length of the flow (here the MAC) and ReL the simu-
lation Reynolds number based on L. The value for y + was restricted to 3 as the
requirement of y + ≤ 1 was not possible. This proved to be sufficiently close to
the walls as discussed in Section 5.2.3. Most researchers use advanced meshing
methods that include spanwise stretching of the mesh to reduce the resolution per-
pendicular to the flow [28] and use the gained memory capacity for achieving the
y + ≤ 1 requirement.
To assist with the choice of the prism mesh settings, a tool was developed in Mat-
lab. This was done to reduce the time required for this process, as the automated
generation of the meshes took about 45 min each time.
The tool contains the equations used by the software to generate the prism lay-
ers and takes as input the prism mesh parameters from above. It then plots the
44
(a) trailing edge on symmetry plane (b) Spanwise cut with winglet
Figure 4.8: Prism mesh details
prism layers as shown in Figure 4.7 and calculates the number of layers inside the
boundary layer based on equation (4.7). This allowed to experiment with a number
of settings to optimise the parameters to achieve the maximum number of layers
in the boundary layer and the best transition to the tetrahedral mesh versus mesh
size. The final settings are listed in Appendix A.
Views of the prism layers are presented in Figures 4.8 and 4.4. Ideally, the aspect
ratio of the prism cells would be close to unity before transitioning into the tetrahe-
dral volume mesh. This would have required around 30 prism layers, compared to
the 22 layers available. Nevertheless, the resulting grid expansion ratio was within
the acceptable limits as discussed in the next section and therefore an usable mesh
quality was achieved.
4.3.6
Mesh Quality
Table 4.1 lists the final mesh sizes for the different cases. The maximum solvable
size was about 6.6 million cells with 8 Gb of computer memory. The pyramid
elements for the original winglet occur near the winglet leading edge joint, where
the mesher could not produce a continuous prism mesh. As the number of pyramid
elements is small, the defect in the prism mesh is small and does not cause any
concern. Still, this was later solved by using a different set of virtual topologies for
the new winglet design.
Figure 4.9 shows a plot of the y + values computed during the solution. In most
regions on the wing the value is 3 or below as specified. The approximation worked
45
Model Elements Tetra Pyramids Prisms

Hoerner 5,595,730 2,223,658 - 3,372,072
Square 5,712,783 2,262,149 - 3,450,634
Winglet 6,086,969 2,428,505 1,928 3,656,536
Winglet 15◦ sideslip 6,317,652 2,520,130 1,897 3,795,625
New Winglet 6,291,584 2,481,580 - 3,810,004
Table 4.1: Mesh data
Figure 4.9: Wing y + detail for α = 5◦
well. Only near the leading edges, where the local Reynolds number is very high
(especially on the winglet), the y + values are between 3 and 5.7. This appears to
be a good compromise, as the parameters for the inflation layers were only set
globally.
The Workbench Meshing application is able to calculate several mesh quality met-
rics. These are important for a well conditioned system of equations to be solved
by the solution algorithm. One metric is the element quality, which is defined as
‘The element quality is based on the ratio of the volume to the edge length for a
given element. A value of 1 indicates a perfect cube or square while a value of 0
indicates that the element has a zero or negative volume’ [39]. For this project,
the results for element quality had typically a mean of 0.45 to 0.5 with a standard
deviation of about 0.35. Sadly this software version does not give any further in-
formation on how many good and bad elements there are. Also the function for
showing the worst element is totally useless, as it highlights the element only on
cut planes through the mesh. The user has to define these planes himself. With
46
4. Experiment Setup 4.4. CFD - Flow Solution
+---------------------------------------------------------------+
| Mesh Statistics |
+---------------------------------------------------------------+
| Domain Name | Orthog. Angle | Exp. Factor | Aspect Ratio |
+-----------------+---------------+--------------+--------------+
| | Minimum [deg] | Maximum | Maximum |
+-----------------+---------------+--------------+--------------+
| Domain | 13.2 ! | 42 ! | 1475 OK |
+-----------------+---------------+--------------+--------------+
| | %! %ok %OK | %! %ok %OK | %! %ok %OK |
+-----------------+---------------+--------------+--------------+
| Domain | <1 17 83 | <1 2 98 | 0 0 100 |
+-----------------+---------------+--------------+--------------+
Figure 4.10: Typical CFX mesh quality diagnostics output
the smallest elements just 3e−4 mm wide, there are about 2100 possible locations
along the wing span. In any case, if by luck the worst cell could be found, there are
no tools available to fix the problem without a complete re-mesh like there are in
ICEM.
Another, more useful tool to judge the mesh quality was available on start-up of
the CFX solver. It displays a more informative statistic of mesh parameters that
are most important for the solver to work properly. Figure 4.10 shows a typical
example. Here we can see the percentage of bad cells in the mesh. On closer
investigation in post processing, the number of cells with a critical expansion factor,
for example, was found to be about only 30 out of the total 6 million cells. This is
clearly a negligible number.
These results together with the proper convergence of the solutions discussed in
Section 4.4.4 and the verification results from Chapter 5 proved that an acceptable
mesh quality had been achieved.
4.4 CFD - Flow Solution

The general solution method was chosen to be a steady state, incompressible flow
and fully turbulent solution. This method proved to be the most stable and acces-
sible procedure for this problem with the available hardware.
The flow Mach number is about 0.36, which is at the upper boundary for incom-
pressible flow. Most authors treat M = 0.3 as the speed where compressibility ef-
fects become significant. The flow Mach number was chosen to keep the Reynolds
number as high as possible as discussed in Section 4.2.1. If the aim of the project
had been to determine the actual drag figures, this flow should have been solved
using compressible flow. This involves the solution of another equation, the energy
equation, during the CFD runs. The energy equation introduces further instability
to the solution [23] and requires additional memory as well as extra solution time.
47
This comparative study, however, will cancel any errors due to compressibility as
discussed in 4.2.1 and the limited mesh quality required special attention to a very
stable solution method. Thus, it was decided to use the incompressible solver after
a test run for compressible flow failed in the early stages of the project.
4.4.1
Boundary Conditions
The boundary conditions on the domain surfaces were standard for an external
flow problem, except for rotating the flow around the model to generate the angle
of attack. In detail:
Inlet, Bottom: Both faces serve as inlets. The inlet velocity magnitude was 125 m/s
for all cases. The velocity components were given in Cartesian components to
result in angles of attack on the horizontal model between 1 and 13 degrees.
The turbulence settings will be listed in the next section.
Out, Top: These faces are the flow outlets with a relative reference pressure of 0
Pa.
Side, Symmetry Plane: Symmetry boundary conditions avoid friction effects on

the walls.
Wing, Winglet: No slip wall with zero roughness was used for full wall treatment
on the wing.
4.4.2
Turbulence Model
The SST turbulence model was used exclusively. There are no parameters to set for
this model. Only the incoming flow conditions were specified in the Inlet boundary
condition to define the values for k and ω via the equations given in Section 2.4.2.
A turbulence intensity of I = 0.1% was used as the aircraft was assumed to fly into
a still atmosphere. For the length scale the option ‘Auto calculate’ was used on
advise. Late in the project it was found out that this option is wrong for external
flows as it results in a large value for the free stream ω∞ , and should only be used
for internal flows. The length scale should have been set as the length of the MAC
of the wing, which is the characteristic length of the flow. Fortunately, this error
had no implications on the results as discussed in Section 5.3.4 as the SST model
removes the dependence of the results on ω∞ . For the improved design runs the
length scale was set properly.
48
4.4.3
Other Solver Settings
The working fluid was air at standard atmospheric conditions (Pressure: 101325 Pa,
Density: 1.225 kg/m3 and Temperature: 15 degrees Celsius) with constant density
and isothermal properties to maximise the flow Reynolds number as discussed in
Section 4.2.1. The advection schemes were of second order. To save space the full
list of settings is given in Appendix E.
As the computer used was equipped with a multicore CPU, the solution was run in
local parallel mode. This partitions the mesh into two parts, which are solved in-
dependently. Information about the interface is exchanged between the CPU cores
as it would be in a distributed computing setup over a network. This method re-
quires more memory but cuts the solution time by at least a third. Overhead for
the exchange of data between the processes accounts for the deviation from the
theoretically possible doubling in solution speed.
4.4.4
Solution Convergence
The solution convergence was determined by the RMS residuals. The threshold
was set to be a value of 2 × 10−5 for all of the transport equations. At this level
the forces were usually converged to the second or third decimal place, which is
accurate enough for this analysis. Intuitively, one would select a value of 1 × 10−5
for the residuals, but in the early test runs it was found that this final reduction
in residuals increased the solution time considerably (sometimes no convergence
would be achieved) without improving the accuracy significantly. Plots of the force
convergence have been omitted due to space restrictions and limited information
content.
A helpful feature would be a convergence monitor on the forces themselves. This

could be realised by specifying a number of iterations where the force fluctuation
has to stay within specified bounds. This could reduce unnecessary solution time,
as often the forces converged earlier than the residuals to an acceptable accuracy.
Sadly neither FLUENT nor CFX feature this. As a workaround, the number of itera-
tions was limited to 120 for each run. If the solution residuals were not converged
by then, the force convergence would be examined and further steps run, if re-
quired.
Every new simulation was initialised with the closest converged solution available.
This method reduced the simulation time by a third (from 12 to 8 hours). At the
end of each run a table of the locations of the maximum residuals is printed as
shown in Figure 4.11. This example is for the winglet at α = 3◦ . Usually, as in this
49
+----------------------------------------------------------------+
| Locations of Maximum Residuals |
+----------------------------------------------------------------+
| Equation | Node # | X | Y | Z |
+----------------------------------------------------------------+
| U-Mom | 2056493 | 5.692E-02 |-2.643E-02 | 7.490E-02 |
| V-Mom | 2056493 | 5.692E-02 |-2.643E-02 | 7.490E-02 |
| W-Mom | 1359534 | 5.737E-02 |-2.626E-02 | 7.550E-02 |
| P-Mass | 2051605 | 5.818E-02 |-2.612E-02 | 7.476E-02 |
| K-TurbKE | 1969984 | 5.691E-02 |-2.642E-02 | 7.491E-02 |
| O-TurbFreq | 1964346 | 6.057E-02 |-2.887E-02 | 8.345E-02 |
+----------------------------------------------------------------+
Figure 4.11: Locations of Maximum Residuals for the winglet at α = 3◦
−1 0
10 10
Mass
U
10
−2 V
−1
W 10
K
ω
−3
10
−2
10
MAX Residuals
RMS Residuals
−4
10
−3
10
−5
10
−4
−6
10
10
−7
10
0 20 40 60 80 100 0 20 40 60 80 100
Iterations Iterations
Figure 4.12: RMS and MAX residuals for the winglet at α = 3◦
example, all components peak at the same location, here near the leading edge of
the winglet on the inside, where the junction vortex is created. It is a convenient
feature to find the location where the problems are caused. The grid in this region
can then be refined. This feature was used extensively during grid development.
The residuals are plotted against iteration number in Figure 4.12. The RMS residuals
converge smoothly with comparable gradients. The insufficient mesh quality near
the winglet leading egde causes the MAX residuals to converge much more untidy
with problems especially in the K equation. The smooth RMS convergence shows
that the problems are confined locally and are not general. Therefore, the solution
error created will not be significant.
To save time, all solutions were run in batch mode, with a list of required com-
putations always waiting in the background. This ensured optimal usage of the
hardware, as immediately after one simulation finished the next would start.
50
4. Experiment Setup 4.5. CFD - Post-processing
This completes the experiment setup. The next section will introduce a few con-
cepts used during the post processing of the results, before moving on to the veri-
fication of the results in Chapter 5.
4.5 CFD - Post-processing

The results of a CFD solution are the numerical values of the flow variables at each
node of the mesh. This data can then be used to display this data or calculate
secondary variables like forces. The package used for this project was ANSYS CFX-
Post. It is a very powerful application, that allows to display the results in many
different ways. The major difficulty in CFD post processing is the visualisation of
3-d data on a 2-d screen. As long as 3-d monitors are not yet standard, one has to
use features like stream lines or vector plots for that purpose. One big advantage
of CFX-Post is the capability to load several result files for comparison.
The next sections introduce the concepts used to process the forces calculated
by CFX-Post and convert them into the standard formats used in the aeronautical
community.
4.5.1
Force Processing
The standard way of expressing forces on aircraft is to use lift and drag coefficients.
These are defined as dimension-less forces parallel to the free stream (drag) and
perpendicular to the free stream (lift). The forces are normalised by the dynamic
pressure q and the wing area S. This conversion was done in Matlab, using the
raw data from the result files and the required co-ordinate transforms discussed
next. CFX reports the pressure- and viscous forces separately. Therefore, it was
possible to process and interpret them independently, as well as to add the viscous
components to the inviscid solutions used for the verifications in Chapter 5.
Co-ordinate transforms The models were at a fixed position inside the domain
and the flow was rotated around them to generate the angle of attack. CFX calcu-
lates the forces along the main co-ordinate axes, which were fixed at the wing. The
axes were defined in a standard aircraft co-ordiante system as
• X: Along the chord line to the back of the wing
• Y: Along the span, positive out the right wing
• Z: positive up
Thus, the results are in body axes. To transform from body to wind axes to obtain
lift and drag in their usual definition, the following transformation was used. With-
51
4. Experiment Setup 4.6. Summary
out sideslip, it is a simple rotation about the y-axis through the negative angle of
attack. The matrix is  
 cos α 0 − sin α 
 
Lwb =
 0 1 0 
 (4.8)
 
sin α 0 cos α
With body forces X, Y and Z the result is

   
 D   X 
   
 Y  = Lwb (−α) ×  Y  (4.9)
   
   
L Z
For the sideslip cases the wing itself was rotated inside the domain, as this is the
only possible way of using a half model for sideslip. This means, that the results
are no longer in body axes but in stability axes. The transformation to wind axes for
lift and drag remains the same, but the side forces require another transformation
about the Z-axis through the sideslip angle β. To save space, these have not been
listed here.
Moments The bending- and pitching moments of the wing, with or without the
winglets, were determined by defining a local co-ordinate system at the wing root
along the chord line with the same orientation as the system co-ordiantes centred
near the wing tip joint (This was a relict of the model set up in Solidworks). The
origin of this new co-ordinate system was at the quater chord location. Then the
function calculator in CFX-Post was used to find the total moments about the X-
and Y-axis of this co-ordinate system for the bending- and pitching moments, re-
spectively.
4.6 Summary
This chapter discussed the setup for the CFD solution. The creating of the CAD
models was described, pointing out some difficulties with data transfer. Next, the
meshing strategy has been discussed, with images of the surface- and boundary
layer mesh reinforcing the ideas. Benchmarking the mesh quality and the solution
convergence was another important step in obtaining a working solution. Here, the
mesh quality was found to be acceptable for the scope of this project. Finally, the
solver settings were described to illustrate the assumptions made.
52
5
Verification of the CFD Simulation
5.1 Introduction
A major part of every CFD simulation is the verification of the results using an
independent source. Every CFD simulation contains errors as discussed in Section
2.3.2. Some of these can become large and render the solution useless. To gain
confidence in the results, the solutions themselves need to be verified, as well as
their sensitivity to several key parameters in the set up. This will be the aim of this
chapter.
Ideally one would compare a CFD solution to other viscous data obtained from
wind tunnel testing or even flight tests. This was not possible for this project due
to time and budget constraints. There are, however, other computational methods
like inviscid panel methods or empirical formulas for viscous drag, that can be used
to verify different parts of the CFD solution. If all these parts themselves compare
well to these established methods, there will be good degree of confidence in the
overall results.
Also, for this comparative study some errors between the solution methods (a con-
stant drag offset for example) may be neglected because they cancel out by com-
paring two cases.
Now, firstly some features of the flow field will be compared to inviscid methods.
Then the viscous drag predictions are benchmarked against some well known em-
pirical equations. The final part of this chapter investigates the sensitivity of the
solution to the experiment set up.
5.2 Flow field

The flow field of a 3-d wing with a winglet contains several characteristic features
to be compared with existing predictions. There are, for example, the chord-wise
and span-wise pressure distributions that are unique for the aerofoil section and
53
5. Verification of the CFD Simulation 5.2. Flow field
the wing planform used. Then, the viscous drag depends mainly on the wetted area
of the wing. Finally, the vortices generated at the wing tip can be studied, as they
have characteristic features depending on the tip shape.
5.2.1
2-D Pressure Distribution
On the symmetry plane the flow is nearly two dimensional, as the wing is simu-
lated in isolation. Thus, it is possible to compare the pressure distribution on the
wing-symmetry plane intersection to a solution of a 2-d inviscid panel method for
aerofoils. The software JAVAfoil [40] was used for this purpose. The case investi-
gated was the wing with the Hoerner tips at α = 5◦ .
To find the correct local angle of attack for the 2-D solution, the results of the
standard lifting line theory were used. It states, that the local angle of attack is
reduced by the downwash as
CL
α2d = α3d − (5.1)
π eAR
0.5693

=5− × 180/π
π × 0.7723 × 7.216
= 3.14◦
with the data for CL from the CFD solution and the Oswald efficiency e approxi-
mated from the same results by fitting a line through the plot CD vs. CL2 .
In Figure 5.1, it can be seen that the CFD solution matches the inviscid solution
accurately in regions, where the boundary layer is thin. This is in the front of the
aerofoil and along the bottom surface. In the back of the top half, from about
half chord, the thickening boundary layer forms a virtual barrier for the flow. This
is shown in Figure 5.2, where the black line is an iso-contour of V = 100 m/s.
The bubble near the stagnation point near the leading edge is of no interest here.
Near the trailing edge the plot illustrates, how the effective shape of the aerofoil
outside the boundary layer changes. The potential flow ‘sees’ a different shape of
the aerofoil, one that is defined by the outer edge of the boundary layer and that is
much thicker in the aft parts. Therefore, the pressure distribution in these regions
does not match the panel code solution as closely as on the leading edge. To achieve
a better match, it is necessary to analyse the new section shape from Figure 5.2 in
the panel code to account for the changes to the aerofoil. The spikes at the trailing
edge in the inviscid solution are typical for this method and can be ignored.
54
CFD
JAVAfoil
−1
−0.5
Cp
0.5
1
0 0.2 0.4 0.6 0.8 1
x−station
Figure 5.1: JAVAfoil pressure distribution compared to CFD data at the

wing root
Figure 5.2: Aerofoil shape change due to boundary layer for potential
flow solution
With this first step showing good agreement of the solutions within the limitations
of the methods, the next step will be to compare the 3-d results for lift and drag
with a vortex lattice method.
55
Figure 5.3: Tornado wing geometry
5.2.2
3-D Pressure Distribution
A 3-d vortex lattice method can be used to compare the CFD results with an inviscid
potential flow solution for the entire wing. This will give a good indication if the
3-d pressure field is calculated accurately.
The Matlab code TORNADO [41] was used for this purpose. This code is very easy
to use and has been shown by its author [42] and by this author [43] to compare
well with published flight test data of simple (Cessna 172) and very complicated
shapes 1 . Nevertheless, this project demonstrated some limitations of the code,
as presented in Appendix D. These limitations, however, were not critical for the
work planned with TORNADO. One main advantage of TORNADO over other panel
methods is the availability of the source code. It is very easy to extend the features
of the code and to add functionality as required. For this project this was used
extensively for the post-processing of the results.
TORNADO neglects the wing thickness and uses only the camber of the mean line
for its calculations. This is a standard procedure with 3-d panel methods, since the
effects due to thickness are usually small compared to the overall results. The CFD
results for the square wingtip were used for this comparison, because this shape
can be modelled identically in TORNADO.
The wing geometry, as modelled in TORNADO, is shown in Figure 5.3. This pan-
elling scheme is the result of a grid convergence study performed to find the mini-
mum grid resolution required for grid independence. The results of this study are
1
This author’s investigation into the low speed characteristics of a double arrow supersonic trans-
port as part of the Aerospace Design 2 class
56
0.9 0.03
4x6 4x6
0.8 8x12 8x12
12x18 0.025 12x18
0.7 16x24 16x24
20x30 20x30
0.6 0.02
0.5
CD
CL
0.015
0.4
0.3 0.01
0.2
0.005
0.1
0 0
−1 0 1 2 3 4 5 6 7 8 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
α CL
(a) (b)
Figure 5.4: Tornado grid convergence for lift and drag (Legend: chord-
wise panels x spanwise panels)
shown in Figures 5.4(a) and 5.4(b). The panel method is robust with respect to grid
sizes, compared to a CFD simulation. Even a very coarse grid with 4 by 6 panels
does not produce large errors. Nevertheless, only the results with a grid size above
16 by 24 cells are nearly identical which indicates grid independence. Therefore,
a grid of 16 by 24 panels was used for the main wing for all computations in this
project.
TORNADO by default calculates only one side of the wing, if it is symmetrical, as

it was done in the CFD solution to save resources. Then, it mirrors the results
and plots the full configuration. This approach has been kept when presenting
TORNADO results in this text. It would have been difficult to change all the code to
account only for one half wing. All numerical results are based on the correct wing
areas in each case and thus the coefficients are comparable with the single wing
from the CFD solution.
To illustrate the the pressure distribution on the wing, Figure 5.5 shows the result
calculated by TORNADO. Clearly visible is the general reduction of lift production
near the tips, as expected, and the effect of the wing twist, that reduces the local
angle of attack near the tips to prevent stalling of the ailerons. This unloads the
wing tips further. The chordwise pressure distribution from TORNADO is based
on only 16 panels, as compared to 61 for the JAVAfoil 2-D solution. This limited
resolution results in a pressure distribution that is shaped differently than the one
shown in Figure 5.1. It integrates to the same lift coefficient, but cannot be used
for a comparison with the chordwise CFD solution. A spanwise comparison at
the quarter chord line would be possible, but could not be completed due to time
constraints and the amount of programming required to process the TORNADO
results into a form that is comparable with the CFD data.
57
−0.2
0.04
0.02
0
−0.4
0.6
−0.6
0.4
−0.8
0.2 −1
−1.2
0
−1.4
−0.2
−1.6
−0.4
−1.8
0.2
0.15
0.1
−0.6 0.05 −2
Figure 5.5: Tornado pressure distribution for α = 5◦ . The minimum Cp

at the leading edge has been limited to -2.0 to show the other
regions more clearly.
1 0.06
CFD
CFD
0.9 TORNADO
TORNADO
0.05
0.8
0.04
0.7
CD
CL
0.6 0.03
0.5
0.02
0.4
0.01
0.3
0.2 0
1 2 3 4 5 6 7 8 9 10 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
α CL
(a) (b)
Figure 5.6: CFD solution vs. Tornado results for CL and CDp of the square
wingtip
The lift curve and the drag polar for the TORNADO solution versus the CFD results
are shown in Figure 5.6(a) and 5.6(b), respectively. The lift curve agrees well with
the CFD results below α = 6◦ . Only a small difference in slop is visible. Above
α = 6◦ the effects of viscosity and beginning flow separation start to influence the
CFD solution. The boundary layer near the trailing edge increases in thickness with
larger angles of attack, as the adverse pressure gradient becomes larger. This leads
to separation above α = 8◦ and consequently, a loss in lift. This is clearly visible
in the reduction in lift curve slope of the CFD solution. The inviscid results do not
include these effects, which explains the constant slope for all angles of attack.
58
The main flight conditions considered in this project were at α = 3◦ for cruise and
α = 6◦ for landing. These are within the linear range of the lift curve and compare
well with the inviscid solution.
Any potential flow solution can only predict the lift dependent drag, which is the
second part in the standard drag polar CD = CD,0 + KCL2 . The zero lift drag CD,0 is
mainly caused by skin friction, a component that cannot be predicted using inviscid
flow. Therefore, only the lift dependent drag is compared in this section, which is
reported by CFX as pressure drag CDp .
The drag polar in Figure 5.6(b) shows less agreement on first inspection. A closer
look reveals the difference in the graphs as an almost constant offset, especially
between 0.3 < CL < 0.6, which corresponds to the main operating range of the
aircraft. This constant offset is most likely caused by the grid dependence of the
CFD solution, which will be shown in Section 5.3.1. At high lift coefficients the
induced drag of the viscous solution grows faster than the inviscid results. This
proves, that the induced drag contains a small viscous component [1].
Again, good agreement has been found within the limitations of the two solution
methods. Next, the viscous drag component will be compared to empirical methods
for drag prediction on flat plates.
5.2.3
Skin friction Drag
Skin friction drag is caused by shear stresses inside the boundary layer near the
walls of the wing. It is very difficult to predict accurately, because it depends on
the resolution of the boundary layer and the quality of the turbulence model. This
was discussed in Section 2.4.1. The skin friction drag is output separately by CFX
and thus can be compared individually. The CFD solution used for this project is a
fully turbulent solution as opposed to reality where there will be a laminar section
of the flow near the leading edge of the wing. This laminar flow will transition into
turbulent flow at some point along the chord, depending on surface roughness,
pressure gradients and other factors. These transition effects can be simulated in
CFD, but add further complications and hardware requirements (more equations
need to be solved during each step). For this reason, the fully turbulent solution
was chosen, where the flow entering the computational domain is already turbulent.
For the purpose of this chapter the wing is approximated with a flat plate of the
same area.
The CFD solution has an almost constant coefficient of friction, Cf , for low angles
of attack as shown in Figure 5.7. The drag due to friction is created in the boundary
layer which does not change significantly at these small angles of attack. This
changes as α increases beyond 6 degrees. From here on the skin friction drag
59
−3
x 10
10
9.5
8.5
8
CD
7.5
6.5
5.5
5
0 2 4 6 8 10 12 14
α
Figure 5.7: CFD solution for CD,viscous versus angle of attack
decreases steadily. This is due to the beginning separation near the trailing edge as
discussed in Section 5.2.2. To illustrate this, Figure 5.8 presents a plot of the wall
shear stress on the upper side of the wing at 3 and 13 degrees angle of attack. It
can be seen that the area of very low shear stress is much larger for the 13 degree
case, as the wing is near stall. This reduces the velocity gradients near the wing
surface and the shear stresses become smaller. Near the leading edge, the shear
stress increases but this cannot compensate the reduction due to the separation.
As a result, the total amount of viscous drag reduces, as shown in Figure 5.7.
To verify the magnitude of the skin friction drag, Hoerner [34] presents two semi-
empirical equations to estimate the skin friction drag of a fully turbulent flow over
a flat plate. The first was derived by Prandtl and v. Karman as
0.074
Cf = 1/5 (5.2)
Rec
for Reynolds numbers up to 1 × 106 . This is lower than the simulation Reynolds
number of 1.58 × 106 , but the equation is used in other references [3] without this
restriction. It results in a skin friction coefficient of Cf = 0.00426 at the simulation
Reynolds number. This result is for one side of the wing. The total skin friction
coefficient is then twice the result from above, namely Cf ,wing = 0.00852.
The second equation, derived by Schultz-Grunow, is a result of testing plane walls

compared to pipes for the formula before. This equation is valid for Reynolds
numbers larger than 1 × 105 :
60
Figure 5.8: Wall shear stress for α = 13◦ (top) and α = 3◦ (bottom)
0.427
Cf = (5.3)
(log Rec − 0.407)2.64
= 0.00414
This equation is Hoerner’s preference for calculation skin friction drag. Again a
factor of two is required to find the total friction drag coefficient. The result is
Cf ,wing = 0.00828.
For the wing with Hoerner tip, the Cf varies between 0.00835 at α = 1◦ and 0.00798
at α = 8◦ . Taking the mean Cf = 0.00817, and comparing it to the second equation,
shows excellent agreement with a difference of about one drag count. Even the
more basic first formula compares well with the CFD result. During the AIAA drag
prediction workshop the variations in drag prediction due to mesh dependency
and solver settings were actually larger than the error between the two empirical
methods [24].
As shown above the semi-empirical methods derived from experiments in the 1940s
compare very well with the CFD results for this incompressible flow regime. The
combination of boundary layer meshing and SST turbulence model appears to work
satisfactory. Further benchmarking of the skin friction drag would require highly
accurate wind tunnel tests, that are able to measure to a precision of a single drag
count. This, however, calls for the state of the art measurement equipment and a
61
very precise model. It is easy to imagine the price tag of such a test, which can
only be performed during a well funded industry- or research project. For the aims
of this study, the achieved accuracy in skin friction drag appears to be more than
acceptable.
5.2.4
Boundary Layer Shape
The development of the boundary layer along the wing chord has already been
discussed in Section 2.4.1 using graphs produced by the CFD solution. It was deter-
mined, that all typical details of a boundary layer were adequately resolved. These
were shown in Figure 2.3. To save space, the results are not repeated here and the
reader is referred to Section 2.4.1 for further information.
5.2.5
Wingtip Vortices
A qualitative comparison can be made for the wing tip vortices generated by the
wing with a winglet. Marchmann et. al. [18] have visualised these vortices on a
general aviation wing with winglets in a tow tank. This was before the age of CFD
and computer graphics. The authors report two distinctive vortices as shown in
Figure 5.9, one originating from the wing tip and one from the tip of the winglet.
The vortices merged at some distance behind the wing. This is a different behaviour
than the wing tip vortices of a high speed aeroplane, which usually only has a
single vortex leaving the tip of the winglet [7]. The vortex shape probably strongly
depends on the Reynolds number of the flow.
Figure 5.9: Wing tip vortices visualised in the tow tank by Marchmann et.
al. Source: [18]
62
Figure 5.10: Wing tip vortices at α = 3◦
Figure 5.10 shows the vortices simulated by the CFD solution at α = 3◦ . The visu-
alisation was achieved by using a function in CFX-Post called ‘vortex core region’.
This creates an iso-surface for the variable ‘swirling strength’, which is a measure
for the vorticity of the flow. This function will be used extensively later on. Here it
shows a similar vortex distribution as in Figure 5.9, except for the rotation of the
winglet tip vortex around the main wing vortex, which cannot be visualised by this
method.
5.2.6
Flow Similarity
The laws of flow similarity require scaled tests in wind tunnels or CFD to be run
at a similar Reynolds- and Mach numbers to the real flow conditions. Only then
will the flow field be similar and forces and moments be directly transferable from
the model to reality. However, it is seldom possible to achieve this match due to
the technical complications involved. For the current project, for example, with a
1/10th scale model one would either have to increase the velocity by a factor of
10, which would render it supersonic, or increase the air density by a significant
amount. This is possible in some wind tunnels by cooling the air down or using
a different gas, but it is very expensive. In CFD it is much easier, one has just to
change the value for the density in the setup. Still, it was not possible for this
project to use the correct Reynolds number due to mesh quality issues for the
majority of the simulations.
A mismatch in Reynolds numbers can cause several errors to lift and drag predic-
tions. These are listed in [45] with methods for correction. Most of these errors are
63
5. Verification of the CFD Simulation 5.3. CFD Simulation Parameters
Figure 5.11: Effect of Reynolds number on stalling of the winglet at α =

6◦
of no consequence for this comparative study. Only the effect of lower Reynolds
numbers on the stalling characteristics of a wing gave rise to a more detailed in-
quiry.
In Figure 5.11 the newly designed winglet from Chapter 7 is shown at the simulation
Reynolds number on the left and at the flight Reynolds number on the right. The
new design is used for this comparison because the effects are more pronounced
for this case. The landing flight condition (see Section 4.2.1) was achieved by setting
the air density ρ = 5.88 kg/m3 at a speed of V = 125 m/s.
The left image shows a region of beginning flow separation caused by the interac-
tion of the junction vortex (arrow) with the boundary layer near the trailing edge of
the winglet. At the correct flight condition it is gone. In [45], the change in stalling
behaviour with increasing Reynolds number is described as a more abrupt effect
at higher angles of attack. The stalling patterns itself are not affected in terms of
location and spatial development of the separation, but simply shifted to higher
angles of attack. Therefore, whenever stalling was observed on the winglets, a sim-
ulation at the flight Reynolds number was run to obtain at least some qualitative
results for the extend of the flow separation at the correct flight conditions. These
results can then be used to judge the impact of the Reynolds number mismatch on
the simulation data.
64
5.3 CFD Simulation Parameters

Every CFD simulation requires many parameters to be set correctly, and other fac-
tors like the size of the computational domain chosen suitably, to produce correct
results. Often the values for these settings are not known, because they may de-
pend on the solution itself, or they are simply unknown for the flow of concern.
Approximations and assumptions based on experience are used in these cases. A
study of the solution sensitivity to these parameters is required to justify the selec-
tions. This is the aim of the next section.
5.3.1
Mesh Size
One of the most important parameters for any CFD solution is the size of the com-
putational mesh. The finer the mesh, the better the resolution of the flow physics.
On the other hand, the available hardware dictates the maximum mesh size. A
compromise between accuracy and usability has to be found in most simulations.
It is then necessary to verify, that the mesh size is small enough to capture the
important physics of the flow to obtain correct results.
Usually, this is done by doubling the mesh size at least twice and comparing the re-
sults using the Richardson extrapolation [29] or more advanced methods [44]. The
evaluation of unstructured grids is especially difficult, as the cell volume changes
not necessarily proportional to the mesh size factor. Also, compared to a struc-
tured mesh, the node locations are not constant. This results in errors, if direct
nodal values are compared between different mesh sizes.
For this project, it was not possible to increase the mesh size by a factor of two
since the working mesh was already near the capacity of the available computer. It
was also not possible to half the mesh size, as the solution would not converge on
such a coarse mesh. To get an idea of the solution sensitivity on the mesh size, a
series of small changes was prepared for comparison with the working mesh. The
parameter used for these changes was the mesh expansion ratio, which determines
the size increase between each layer. The expansion ratio was varied only for the
tetrahedral surface- and volume mesh. Although it violates the guidelines specified
in Section 2.3.2, the boundary layer mesh was kept constant. A change for this
prism mesh would have been very time consuming, and would have changed the
overall characteristics of the mesh in a significant way. As the solutions on the
current grid size are strongly mesh dependent, this would have made it difficult to
compare the different meshes.
Table 5.1 shows the sizes of the meshes tested. All were run with the Hoerner
wing at α = 6◦ . The first observation during the runs was a slower convergence the
65
Case Expansion ratio Mesh size (Cells) No. of nodes

1 1.21 6.498e6 2,311,835
2 1.23 5.986e6 2,192,322
standard 1.25 5.596e6 2,098,998
3 1.27 5.280e6 2,020,981
4 1.29 5.026e6 1,959,385
Table 5.1: Parameters for mesh size verification
further the grid size diverged from the standard mesh. Case 4 and especially case 1
took about 50 iterations longer than usual. This was most likely caused by the non
optimal interface between the boundary layer mesh and the tetrahedral volume
mesh, which was optimized for the 1.25 expansion ratio. Although expected for
the mesh enlargement, this behaviour was surprising for the mesh refinement. It
further demonstrates, that the current mesh size is just large enough to be solved
properly and that the solution is very sensitive to mesh disturbances.
The results for lift- and drag coefficients are shown in Figures 5.12. The left graphs
show the results of the computations plotted against the square of the relative
mesh size. As second order discretisation schemes were used for the CFD solution,
the error converges quadratic with decreasing mesh size within the asymptotic re-
gion as discussed in Section 2.3.2. Consequently, if in the asymptotic region, the re-
sults should change monotonically with grid size and the plot of mesh size squared
versus results should be a straight line. This line can then be extrapolated to yield
the result for an infinitesimally small mesh.
Inspecting the graphs, the plots show a linear trend with some disturbances. Es-
pecially the lift curve contains a noticeable break in slope below for the smaller
meshes, whereas this break is less pronounced in the drag graph. One has to
be careful to conclude an overall linear trend from this data as the variation in
mesh size was below 25%. This indicates convergence with mesh size, but not if
the asymptotic region has been reached. Therefore, grid convergence cannot be
asumed from this limited data without further investigations.
If we assume, for the sake of the argument, that these meshes are in the asymptotic
convergence region, we can extrapolate the line as shown in the right plots of Figure
5.12 to an infinitely small mesh size. This results in predictions of CL = 0.616 and
CD = 0.023 as shown in the figure. Especially the drag result is then much closer
to the TORNADO prediction of CD = 0.0183 + 0.0082 = 0.0265, if the viscous
component of drag is added to the potential flow solution. From this, it appears
that in the current solution the drag is over-predicted by substantial amounts.
The methods for error estimation presented in Section 2.3.2 have been tested for
the small mesh variations without success. The steps in mesh size are too small.
The error predicted by Equation 2.8 in conjunction with Equation 2.9 yields for the
66
0.6435 0.65
0.643 0.645
0.6425 0.64
0.642 0.635
CL
L
C
0.6415 0.63
0.641 0.625
0.6405 0.62
0.64 0.615
0.9 0.95 1 1.05 1.1 1.15 0 0.5 1 1.5
Relative mesh size squared Relative mesh size squared
(a) Lift coefficient
0.0336 0.036
0.0334 0.034
0.0332 0.032
0.033 0.03
D
CD
C
0.0328 0.028
0.0326 0.026
0.0324 0.024
0.0322 0.022
0.9 0.95 1 1.05 1.1 1.15 0 0.5 1 1.5
Relative mesh size squared Relative mesh size squared
(b) Drag coefficient
Figure 5.12: Lift and drag vs. relative meshsize squared
lift coefficient an error of 4.3%, which appears sensible, but for the drag coefficent
an error of 33% is predicted. This is certainly too high and does not agree with the
findings of the other validations in this chapter and the grid convergence method
used above in Figure 5.12. This demonstrates, that these methods only work if
larger variations in the mesh size together with a uniform refinement are used.
Otherwise the underlying assumptions are violated.
Summarising, the used mesh size was just large enough to yield reasonable results.
The solutions converge monotonically with the grid size but the asymptotic region
has most likely not yet or just been reached. This is supported by the results of
the AIAA drag workshops, where the current grid size is just the coarsest size used
[24]. In these papers, solutions reach the asymptotic convergence region at about
8 million nodes and above for the transonic flow. Here, the incompressible case
and the fact that only the wing and not a wing fuselage combination was simulated,
should allow the conclusion, that the region of asymptotic convergence requires
less severe grid sizes. Nevertheless, it is necessary to double the mesh size at least
once to prove this.
67
21
Standard
20 Small
Large
19
18
L/D
17
16
15
14
13
0 1 2 3 4 5 6 7 8 9 10
α
Figure 5.13: Variation of L/D with domain size at α = 6◦
Domain Boundary Small Standard Large

front 5c 10c 20c
back 10c 20c 40c
top/bot 5c 8c 16c
side 10c 12c 24c
Table 5.2: Distances of domain boundaries relative to the MAC for do-
main size verification
5.3.2
Domain size
The size of the computational domain can influence the results of the CFD compu-
tations. If the domain is too small, the effect of the boundary conditions can change
the flow around the wing. Thus, it is necessary to investigate the sensitivity of the
solution to the domain size. Table 5.2 shows the three sizes tested. The smallest
domain has the minimum size usually recommended for a CFD simulation. The
standard domain was chosen larger to minimise the influence of the borders. It is
a compromise between the size required by the meshing guidelines from Section
2.4.3, where 100 chord lengths were specified, and the mesh size allowables. The
largest is even closer to the recommendations of the drag prediction workshops.
The standard Hoerner wing was used for these runs at 3,6,8 and 10 degrees angle
of attack as these runs were the fastest.
68
Figure 5.13 shows the resulting L/D curves for the three domains. There are only
limited results for the non-standard domains available, but the trend is clearly vis-
ible. There is no clear dependence of the results on the domain size, otherwise the
results would scale with the domain size in one way or another. Here, there is no
logical order, the standard domain gives the highest results with the two others
showing a reduction in L/D over the entire range. Therefore, it can be concluded
that the results are independent of the domain size and that the small differences
between the results are caused by the mesh dependency of the calculations. The
very large domains recommended for the drag prediction workshops are most likely
only required for compressible flow solutions, where shock waves are present.
5.3.3
Choice of Turbulence Model
An attempt to run a simulation at α = 3◦ with the k − turbulence model was

made to study the effect of the turbulence model on the solution. As shown in
Figure 5.14, this attempt failed. The solver terminated after 29 iterations due to
a divergence in the solution. The conditions were exactly the same as for the run
with the SST model shown in the figure on the right. Both cases were initialised
with a converged solution. A clean start with the k − model resulted in a solution
divergence after 12 iterations. Both, first and second order schemes were tested for
the turbulence equations.
−1 −2
10 10
Mass
U
V
−2 −3 W
10 10
−3 −4
10 10
Residuals
Residuals
−4 −5
10 10
−5 −6
10 10
−6 −7
10 10
0 10 20 30 0 50 100 150
Iterations Iterations
Figure 5.14: Residuals for k− turbulence model compared to SST model
69
Figure 5.15: SST first blending function for α = 5◦
Figure 5.16: Velocity contours at α = 5◦
Therefore, it appears that the k − model is not only less accurate [30], but also
introduces numerical instability into the current solution. The decision to use the
SST model exclusively was justified by this test. If the SST would not have been
available, the entire project would most likely have failed within the hardware re-
strictions.
5.3.4
SST Model Verification
To investigate, if the blending functions in the SST turbulence model actually work,
they were plotted on a cut plane across the mid-wing at α = 5◦ as shown in Figure
5.15 and 5.17.
As mentioned in Section 2.4.2, the first blending function F1 combines the k −

and k − ω turbulence models depending on the flow region. The k − ω model is
used in the boundary layer where the blending function is one (shown as red in
the figure). Figure 5.16 shows the corresponding velocity contours for comparison.
The boundary layer is very thin behind the leading edge on top of the wing and
expands to the trailing edge. Correspondingly, the region of F1 = 1 expands along
70
Figure 5.17: SST second blending function for α = 5◦
the wing chord. The change in F1 to zero away from the wing surface is sharp in the
first two thirds of the wing chord. Near the trailing edge, there is a smaller velocity
gradient due to the thickening boundary layer and the blending function reduces
to zero more slowly. Just below the leading edge is the stagnation point with a
region of low speed flow. In this region F1 is one, too, which is not desired. This
shows the difficulty to define a mathematical description of the boundary layer
edge. The actual blending function contains three conditions to ensure the best
possible prediction of this transition region. These seem to work well, but with
minor imperfections, as some other regions in the flow also fulfil these conditions.
The effect of using the k − ω model in the stagnation region further away from the
surface, will, however, be small compared to other errors in the solution.
The second blending function F2 , that blends the different formulations for the
eddy viscosity in the SST model, is shown in Figure 5.17. The region where it takes
the value one is similar to F1 , but the distance normal to the surface appears to be
larger. Also the fading to zero of F2 is much more gradual than for F1 , especially
near the trailing edge. Still, the second blending function is one only inside the
boundary layer, which shows the formulation to work as expected, with the same
limitations as described before.
A planned comparison of the prediction of the separation point could not be com-
pleted, because, as discussed in Section 5.3.3, the other turbulence models did not
run properly.
The independence of the SST model of the free stream value of ω has been verified
accidentally by making the wrong choice (automatic length scale) for the initial
condition of ω, as described in Section 4.4.2. As shown in Table 5.3, this resulted
in a very high value for ω, compared to the correct setting of ω = 1.509 s−1 .
71
This error, however, did not have any effect on the results as shown in the table.
Hence, the data also prove the independence of the SST model of the free stream
value for ω, as expected.
ω = 1.509 s−1 Auto ω = 1568 s−1

CD : 0.0281 0.0281
CL : 0.5692 0.5693
Table 5.3: Sensitivity to free stream ω for α = 5◦
5.3.5
Turbulence Intensity
One major quality criterion for wind tunnels is the turbulence level in the test sec-
tion. This implies that the results will depend on this flow characteristic. To eval-
uate the effect of the turbulence intensity for the current project, a series of runs
of the Hoerner wing at α = 6◦ with increasing levels of turbulence was made. The
results are plotted in Figure 5.18 in form of the L/D changes between the runs.
The results show a 0.4% difference between a turbulence intensity of 0.1% and 5%.
This is a small sensitivity over a large range of input values, comparable to a 0.2
degree change in angle of attack in Figure 5.13, for example. While certainly signif-
icant for high precision measurements, other sources of error in this project will
dominate the precision of the results. Further, any effects of this setting will cancel
during the comparison of the different cases.
19.54
19.53
19.52
19.51
L/D
19.5
19.49
19.48
19.47
19.46
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Turbulence Intensity [%]
Figure 5.18: L/D sensitivity to turbulence intensity for α = 6◦
72
First order Advection Second order advection

CD : 0.0495 0.0281
CL : 0.5450 0.5693
L/D : 11.01 20.26
Table 5.4: Advection scheme sensitivity for α = 5◦
5.3.6
Advection Scheme Selection
The choice of the advection schemes in the CFD solution has a large influence on
the final results. Usually, one would attempt to use second order schemes as they
are known to be more precise on a given mesh size. On the other hand, these
schemes are more likely to generate convergence problems and will take more time
and memory for the calculation of each iteration. If the mesh quality does not
allow the use of these second order schemes, one can use first order schemes or a
blending between the two to increase the stability of the solution.
These first order schemes were used in the early stages of the mesh development to
promote convergence. A major point of concern during this stage were unrealistic
drag values. Later in the project the mesh achieved a quality that allowed the use of
second order schemes. Immediately, the drag problems disappeared. To quantify
the difference in results between the two methods, a comparison was made on the
final mesh.
Table 5.4 shows the results for the Hoerner wing at α = 5◦ . The differences in lift
are much less significant than the errors for the drag. Overall there is a factor of
two between the two resulting values for the L/D. This impressive difference shows
clearly, that for aerodynamic calculations, where the drag is important, the use of
second order advection schemes is of paramount importance.
5.3.7
Dynamic Viscosity of Air
When setting the properties of air to the working conditions of standard atmo-
sphere sea level, it was overlooked, that the dynamic viscosity had to be changed,
too, due to a temperature change to 15◦ Celsius from 20◦ Celsius. For unknown
reasons is this setting very well hidden in CFX-Pre, so it was missed. This presented
no major concern as these effects cancel out when comparing cases. Nevertheless,
to estimate the error introduced, a validation run was made with the correct value
for µ = 1.79e−5 kg/(ms).
73
5. Verification of the CFD Simulation 5.4. Summary
µ = 1.831e−5 kg/(ms) µ = 1.79e−5 kg/(ms)

CD : 0.0281 0.0281
CL : 0.5693 0.5695
Table 5.5: Sensitivity to dynamic viscosity
Table 5.5 shows, that there is no appreciable effect on the results. The change in
dynamic viscosity is too small to cause a measurable effect.
5.4 Summary
This chapter contained a variety of methods to gain confidence in the CFD simula-
tion results. All these investigations showed a high level of agreement with the CFD
data, within the limitations of the methods. Further, except for the mesh size con-
straints, no other critical sensitivity on a solution parameter was identified. This
validates the setup of the experiments. Overall, the results of this chapter allow the
study the results in the next chapter with good confidence in their quality.
74
6
Results and Discussion of the Original
Winglet
6.1 Introduction
This chapter assesses the changes in performance of the Merlin III wing with mod-
ifications to the wing tip. Firstly, the square wing tip is compared to the Hoerner
tip. Then the winglet will be characterised extensively, investigating the flow field,
the performance changes, possible effects on the handling qualities and the impli-
cations on the wing structure.
It will be shown that the current winglet design is less than optimal in terms of
performance. Its shortcomings will be shown and interpreted, while it has to be
kept in mind that it was designed purely for aesthetic reasons. Therefore, the main
goal is not to introduce any negative effects. Any performance improvement is of
extra benefit. This discussion then leads to the next chapter, where an improved
design is presented.
The main source of data for this chapter are CFD results, reported for the 1/10th
scale model at the simulation Reynolds number. These will be complemented by
inviscid results where appropriate.
6.2 Square Wingtip

As mentioned in Section 3.3, there is no apparent advantage of the complicated
Hoerner wing tip over a square tip in the literature. To investigate the motivations
behind this design choice, a square tip of the same span has been tested on the
Merlin III wing. This tip was modelled by simply extending the wing tip section by
300mm, while maintaining leading and trailing edge continuity. This results in a
slightly more tapered wing, but this difference was insignificant. The edges of the
square tip were modelled as sharp corners to simulate an extreme case.
75
6. Results and Discussion of the Original Winglet 6.2. Square Wingtip
21
20
Hoerner
19
square
18
L/D
17
16
15
14
13
0 2 4 6 8 10 12 14
α
Figure 6.1: L/D for the square tip versus the Hoerner tip
These sharp edges required a relatively fine mesh, as the changes in flow direction
are very pronounced in these areas. This is especially true near the leading edge of
the wing.
In Figure 6.1, the resulting L/D curve is plotted in comparison to the Hoerner tip
results. Figure 6.2 shows some flow features of the square tip versus the Hoerner
tip. The L/D graph proves the results from the literature [33]. There is no advantage
of the Hoerner tip over the square tip. Surprisingly, the square tip appears to be a
better choice at angles of attack larger than three degrees. This last result has to
be treated with some caution, since the differences are small and they could well
be caused by the mesh dependency of the solution. There is no practical way of
verifying a wing tip performance result with the methods available to this project.
The only way of confirming the superiority of the square tip would be a test in a
wind tunnel or a flight test. The results in the literature are based on these tests
(although the test conditions, especially the Reynolds number, are unknown), with
both tip designs performing equally.
The flow features in Figure 6.2 illustrate the different mechanisms in forming the
wing tip vortices. While the Hoerner tip sheds the vortices from its sharp, but
rounded edge in a seemingly smooth way, the flow around the square tip appears
much more untidy. Here there are three distinct vortices, one starting near the
sharp lower leading edge corner, another, smaller one starting at the upper leading
edge and moving about 20 cm inboard (arrows) and finally a vortex starting near
the maximum thickness of the upper wing. These three vortices merge into a single
tip vortex near the trailing edge.
76
6. Results and Discussion of the Original Winglet 6.2. Square Wingtip
(a) Square tip (b) Hoerner tip
(c) Square tip (d) Hoerner tip
Figure 6.2: Vortex comparison on square and Hoerner tip at α = 6◦
The pressure distributions shown in Figure 6.3 may explain why this more compli-
cated flow around the square tip is equal or even better in terms of drag than the
cleaner flow on the Hoerner tip. The Hoerner tip shows a larger region of very low
pressure, and thus high velocities, where the vortex core builds up along the edge
of the tip. This will produce high surface friction and increase this drag component.
Also, the vortex core appears to be very strong so that further losses due to friction
in the vortex itself will create a drag increment. The square tip only has a small
region of very low pressure near the leading edge corner and two less pronounced
low pressure areas along the edges of the upper and lower surfaces. It appears,
that the three vortices generated are weaker along the edge of the tip before they
combine into the tip vortex. This may lead to less internal energy loss and less
surface friction, although the flow looks more chaotic.
Hoerner in [33] also argues, that a measure for wing tip performance is the effec-
tive wing span, meaning the spanwise location, where the tip vortex is shed from
the wing. Both tip designs have the same effective span in his book and this was
confirmed visually by inspecting the vortex graphs created by the CFD solution.
77
6. Results and Discussion of the Original Winglet 6.3. Winglet
(a) Square tip (b) Hoerner tip
Figure 6.3: Pressure comparison on square and Hoerner tip at α = 6◦
Concluding, with the data available, there is no reason for using the more compli-
cated Hoerner tip design, compared to a simple square tip. The changes in wing
performance due to a change in tip shape are very small and therefore hard to mea-
sure, especially during the 1960s, when this wing was designed. Most likely this
choice is simply a standard practice solution by the designers which is based on a
‘it has worked before’ approach.
6.3 Winglet
The addition of a winglet has influence on three main characteristics of an aero-
plane. It is added to improve the flight performance, while also improving or at
least not deteriorating the handling qualities. Finally, the impact on the structural
design of the wing must be considered and minimised. The following sections looks
into each of these fields to assess the impact of adding the winglet design to the
Merlin III. The findings are interpreted using selected flow features to explain the
reasons for the results.
6.3.1
Effect on Performance
The main reason to fit a winglet to an existing wing is to reduce the lift dependent
drag and thus to improve the performance. Figure 6.4 shows the results for lift,
drag and L/D for the three tip shapes. The square tip has been included for com-
parison. Inspecting the figure, the differences in results for CL and CD are small.
The winglet increases the lift by a small amount, mainly due to the increased wing
area. At high angles of attack, the reduction in lift curve slope is similar for the two
wing tip designs. This indicates no major changes to the stalling characteristics.
78
1.2 0.09
Hoerner
1.1 square
0.08
winglet
1
0.07
0.9
0.06
0.8
CD
CL
0.7 0.05
0.6
0.04
0.5
0.03
0.4 Hoerner
square 0.02
0.3 winglet
0.2 0.01
0 2 4 6 8 10 12 14 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1.1 1.2
α CL
(a) CFD solutions for CL (b) CFD solutions for CD

21 −3
x 10
10
Hoerner
20 square 9.5
Winglet
9
19
8.5
18
8
L/D
17
CD
7.5
16 7
6.5
15
6
14
5.5 Hoerner
Winglet
13 5
0 2 4 6 8 10 12 14 0 2 4 6 8 10 12 14
α α
(c) CFD solutions for L/D (d) Difference in viscous drag
Figure 6.4: Results for aerodynamic performance of the wing with differ-
ent tip designs
The drag polar shows only slight improvements as some of the benefits of the
winglet on the pressure drag are consumed by the increase in viscous drag. This
increase in viscous drag is shown in Figure 6.4(d). It is caused by the larger wetted
area of the wing with the winglet, in addition to the strong viscous effects of the
leading edge vortex as discussed in Section 6.3.2 and 7.2.3.
An interesting feature is the small dent in the viscous drag curve at α = 5◦ . Initially,
this was thought to be a calculation error, but it shows up in all results for the
Hoerner tip as well as for the original winglet, but not for the new winglet in Section
7.2.3. Some minor changes to the Hoerner tip at 40% chord were necessary to
accommodate the new winglet. These must have changed the beginning tip vortex
in that region (see Figure 6.2(d)) for this specific angle of attack and removed the
effect.
The effects of the three tips are best seen in the L/D graph in Figure 6.4(c). The
maximum L/D for the standard wing is at α = 4◦ , which is near the cruise condi-
tion as calculated in Section 4.2.1. The winglet produces an increase in maximum
79
L/D of 0.28, which corresponds to 1.37% at the same angle of attack. This is clearly
not a significant improvement but also not a bad result, considering the design in-
tend. The winglet compensates for the addition of the viscous drag component and
therefore is at least performance neutral. The benchmark, however, for a winglet
design on a general aviation aircraft will be a L/D improvement of 10% or more as
demonstrated in [18] and [16]. This shows, that the design is far from being opti-
mum. Interestingly, the square tip is as good as the winglet at α > 5◦ . This could
again be a result of the mesh dependency of the solution.
6.3.2
Flow Features
In this section an attempt will be made to explain the poor performance of the
current winglet design by interpreting several flow features around the winglet.
CFD solutions contain a wealth of imformation, which makes them very useful as
a design tool, even if the numerical results are not completely accurate. The CFX
manual [23] states, that for a detailed flow visualisation a reduction of the solution
residuals to 10−4 is sufficient. This is considerably less than the required thresholds
for accurate numerical predictions (see Section 4.4.4), and should be possible even
under difficult conditions. Visualisations of the flow field presented in this section
are helpful in understanding the physics of the flow. Some conclusions can be made
simply by inspection, which is difficult if not impossible with most other methods
for aerodynamic predictions. The methods used are stream lines, contour plots and
iso-surfaces of variables that indicate the vorticity in the flow.
Flow Vorticity Figure 6.5 shows the extensive vortex generation of the winglet at
◦
α = 6 and its effects on the surface shear stress. The most striking feature is the
strong vortex that runs along the leading edge and then across the winglet to form
a trailing vortex. We can see in Figure 6.5(d), that this vortex is not the winglet tip
vortex and therefore some of the span of the winglet is not effective. There is a
small tip vortex, too. Its intensity is considerably less than the main vortex and is
not displayed with the settings used to create Figure 6.5(d).
The vortex is created by the large leading edge sweep of 61 degrees of the winglet.
This introduces a strong spanwise flow and causes the winglet to behave like a delta
wing. Comparing this flow field to images of delta wings in [3] shows the similar
shapes of vortices near the leading edge. It is commonly known, that delta wings
are not very effective lifting surfaces at low speeds and low angles of attack. Their
main advantage in this flow regime is the tolerance of high angles of attack without
stalling.
80
(a) (b)
(c) (d)
Figure 6.5: Wall shear and vortices on the winglet at α = 6◦ and zero
sideslip
The stream lines in Figs. 6.5(a) and 6.5(b) show that the origin of the leading edge
vortex is in the wing-winglet joint. This vortex is a common feature of non-optimal
discontinuities in wing chord, which is why fully blended winglets work best. One
might think, that here simply the joint vortex is deflected along the leading edge
of the winglet and the remaining flow is more aligned with the free stream, flowing
across the vortex as visible in Figure 6.5(a). In Figure 6.5(b), a different set of stream
lines is deflected around the leading edge of the winglet further up. These stream
lines originate from below the set in Figure 6.5(a), which shows the strong vertical
flow components. Some of these stream lines also join the vortex. This proves, that
this leading edge vortex is fed by the flow around the leading edge along the entire
span. Consequently, the delta wing theory is reinforced by these findings.
81
The wall shear stress in Figure 6.5(c) illustrates the path of the vortex core across
the winglet surface. High stresses, which correspond to high drag, are visible in the
path of the vortex. The upper edge of the winglet has no surface shear, the flow
is fully separated. The shear on the leading edge is very high, corresponding to
the high velocities around the edge. The sudden decrease in shear right behind the
leading edge (the colour changes from red to green) is caused by the non-optimal
shape of the winglet section. As discussed in Section 3.4, the winglet is basically a
flat plate with a rounded leading edge. The curvature of the leading edge is quite
blunt as shown in Figure 6.7(c), which requires strong directional changes of the
flow over a short distance. The drop in wall shear shows, that the gradients are
too strong and the flow locally separates to flow around the leading edge vortex
before aligning with the winglet surface. A more gradual increase in thickness of
the aerofoil (as featured by any ‘proper’ section) is required to improve the leading
edge flow.
Pressure Distribution Figures 6.6 and 6.7 shows some images of the pressure
distribution on the winglet at α = 6◦ . Note, that the data shows only the variation
from the ambient pressure. Figure 6.6(a) corresponds to the wall shear from above.
Very low pressure regions can be found on the leading edge and in the vortex core,
where the velocities are large as they correspond via Bernoulli’s equation at this
flow regime. Compared to a normal aerofoil, the low leading edge pressure on
the leading edge creates a suction force approximately straight forward without a
notable component normal to the winglet. This suction force could be regarded as
a thrust, but the area it acts on is too small to create a positive impact. A stronger
(a) (b)
Figure 6.6: Pressure distributions on the winglet at α = 6◦ and zero

sideslip, extreme values limited to show the important fea-
tures more clearly
82
(a) Top view onto winglet near half span (b) Front view
Figure 6.7: Pressure distributions on the winglet continued at α = 6◦ and

zero sideslip, extreme values limited to show the important
features more clearly
normal component is required for a winglet to operate effectively as discussed in

Section 2.2.3.
A comparison of the pressure distribution in Figure 6.7(a) and the section distri-
butions in Appendix B with the pressure distribution on a thin flat plate (see [3]
for example) shows similar very low pressure peaks right at the leading edge with
steep adverse pressure gradients that lead to the observed flow separations. On
the inside of the winglet low pressure regions and thus a normal force is generated
only near the leading edge vortex. The rest of the winglet surface is not adding
substantial force increments.
Figure 6.7(b) shows the low pressure region in the vortex core in a lateral cut. This
image also shows the strong inboard flow across the leading edge, which appears
to be spilling across the sudden barrier in the path.
3-D Flow Features Figure 6.8 shows some images of the strongly 3-dimensional
flow on the winglet at α = 6◦ . The vector plot in Figure 6.8(a) confirms the strong
spanwise flow near the leading edge, also shown in Figure 6.5(a). The vortex forms
a pocket behind the leading edge, effectively thickening the winglet section at this
area. As mentioned before, the flow around the leading edge cannot negotiate the
sharp corner and runs around the vortex before aligning with the winglet surface.
Without the vortex, the flow would separate right behind the leading edge.
The vertical velocity components in Figure 6.8(b) reinforce the interpretation of the
flow. Here we can see the air ‘spilling’ around the leading edge, actually gaining
momentum in a downwards direction on the inside of the winglet. This is also
visible in the vector plot, where the flow around the vortex is directed downwards.
Opposed to that, the leading edge vortex has a strong upwards velocity component.
83
(a) Velocity vectors (b) Vertical velocity components
(c) (d)
Figure 6.8: 3-d flow and tip vortices on the winglet at α = 6◦ and zero
sideslip
This will create strong shear in the regions where the two flow directions meet.
Increases in drag are the consequence, as momentum is lost in that process.
Figures 6.8(c) and 6.8(d) confirm the findings about the trailing edge vortices in
[18]. In the near field, behind the wing tip, two distinct vortices are present (see
also Fig. 6.5(d)). The directions of the vortices were made visible using the swirling
vector components. We can see, that the two vortices rotate in the same direction
and merge in the far field into one larger vortex. The observations in [18] about the
twisting path of the two vortices and the vortex diffusion observed in the towing
tank could not be confirmed. Possibly, the much lower Reynolds number of these
tests and a larger winglet caused the observed effects.
84
6.3.3
Effect on Handling Qualities
The addition of winglets to an existing wing will have several effects on the han-
dling qualities of the aeroplane [14] [16]. The main influences, according to these
references, are:
1. The stalling behaviour is frequently improved due to the interaction of the

winglet with the main wing flow field.
2. Changes to lateral stability and modes of motion (Dutch roll and Spiral mode)
take place. Winglets will affect mainly the two sideslip derivatives Cy,β and
Cl,β . Both become more stable. The directional stability derivative Cn,β is usu-
ally not strongly affected. As a result, the spiral mode becomes more stable
and the dutch roll will increase its natural frequency with a small reduction
in damping. These changes may have a critical effect on the dutch roll fre-
quency response, resulting in deteriorating handling qualities. According to
[16], these effects depend on the overall aircraft configuration and no clear
rules to estimate the exact changes due to adding a winglet exist yet. Testing
is required on a case to case basis.
3. The maximum crosswind landing capabilities will be reduced due to the in-
crease in effective dihedral. This reduces the aileron control power available
for a steady heading sideslip manoeuvre.
These effects can be very pronounced on highly manoeuvrable aeroplanes like agri-
cultural aircraft [16]. For a lower performance aeroplane, that is designed for pas-
senger transport, these effects are usually less of a concern, but nevertheless need
to be evaluated to ensure safe operation.
Stalling Behaviour The stalling of the Merlin III wing has been investigated in its
original configuration and with the winglet fitted. Figure 6.9 shows a plot of the
wall shear at 13 degrees angle of attack. This shows the flow separation near the
trailing edge of the wing. The extend of the stall reduces near the tip due to the
washout of the wing. Inspecting Figure 6.9 and Figure 6.4(a), there is no appreciable
effect on the stalling pattern when the winglet is installed. The lift curves show the
typical reduction in slope due to the stalling, but remain parallel for both wing tip
designs. A very small area of low shear near the winglet is visible in Figure 6.9, that
is not present for the Hoerner tip (see arrow). This is caused by separation in the
winglet junction and may lead to a more severe separation near the tip as the angle
of attack increases further. As the aileron is already in the fully separated region,
it is unlikely that this will cause any major problems.
85
Figure 6.9: Stall comparison of Hoerner wing and winglet at α = 13◦
It is interesting to note that the current winglet does not improve the overall stalling
behaviour as observed in [16]. This presents further evidence that this winglet
operates less than optimal and that there is room for improvements.
Another unwanted effect might occur, if vortices generated by the winglet interact
with the aileron. This could severely affect the effectiveness of the roll control
and therefore lead to unacceptable handling conditions. This may be especially
significant as the winglet begins to stall at high sideslip angles. The aircraft has a
demonstrated cross-wind capability of 10.7 degrees, where stalling will inevitably
occur on the winglet.
Figures 6.10 and 6.11 show the vortex development on the winglet at the landing
flight condition up to 15 degrees sideslip. The winglet fully stalls between 6 and 10
degrees at the experiment Reynolds number. More exact results were not possible
due to time limitations. A run at the flight Reynolds number confirmed a full stall
at 10 degrees sideslip.
Inspecting the figures we find, that at no sideslip condition the winglet vortices or
separation zones present any problem for the ailerons. The only reason for concern
due to the stalled winglet will be structural vibrations, which will be discussed
in the next section. The effect of inverting the direction of the sideslip has been
investigated using TORNADO. These preliminary results showed a strong reduction
of loading on the winglet, as the sideslip and the spanwise velocity component Vy
86
(a) α = 6◦ , β = 0◦ (b) α = 6◦ , β = 3◦
(c) α = 6◦ , β = 6◦
Figure 6.10: Vortex development with increasing sideslip, winglet not yet
fully stalled
due to the tip vortex cancel. In addition to that, any separated flow will be on the
outside of the winglet and will not cause any problems for the control systems.
Changes to Lateral Derivatives An attempt was made to investigate the effect

of the winglet on the lateral derivatives of the aircraft, mainly Cl,β and Cy,β as
suggested in [16]. The CFD results cannot be used for this purpose, because these
derivatives are a results of asymmetric flow over both wings. This is especially true
for the dihedral effect Cl,β . As only one wing was modelled in the CFD solution, no
sensible results can be obtained from this method.
TORNADO offers the complete set of derivatives in its results. They are calculated
by a perturbation method. Unfortunately, the accuracy of this method has been
found not to be adequate for the two derivatives of concern. The dihedral deriva-
tive for a single wing with positive dihedral should always be stable and therefore
87
(a) α = 6◦ , β = 10.7◦
(b) α = 5◦ , β = 15◦
Figure 6.11: Vortex development with increasing sideslip, winglet fully

stalled
negative. TORNADO produces positive values of a similar order of magnitude as

would be expected. The developer claims in his verification [42], that this is caused
by the choice of reference points. This has to be doubted, as no sensible reference
location could be found to produce a negative result for Cl,β . It is more likely, that
there is simply a sign error in the code as it uses a different co-ordinate system
than the standard flight mechanics convention1 . If this is assumed true, the two
derivatives become more stable as mentioned in [16]. The changes, according to
TORNADO, are:
1
It is, of course, also possible that the precision of TORNADO in predicting these derivatives is
simply not accurate enough.
88
∆Cy,β = −0.0349 rad−1
∆Cl,β = −0.0139 rad−1
An assessment of the implications on the handling qualities can be done using

the standard approximations for the lateral modes of motion. The effect on the
sideslip landing capabilities can be estimated using the lateral stability equations.
Both methods require the knowledge of the aerodynamics characteristics of the
entire aircraft. These are usually known only by the manufacturer and are not
available for this project. Methods for estimating these data are beyond the scope
of this project. These methods include the DATCOM database, more sophisticated
simulations, as well as a wind tunnel test.
A second method of determining the effects of the winglets on the lateral behaviour
of the aircraft is a flight test programme. A prototype of the winglet could be
installed to the aircraft and carefully tested in flight, relying on pilot input for the
handling qualities and on measurements for the cross wind landing capabilities.
In the current case it will most likely be easier and cheaper to perform the flight
testing, compared to further simulations, as these results will be required by the
airworthiness authorities anyway. These flight tests can also help to solve some of
the questions regarding the structural influences of the winglets, as discussed in
Section 6.3.4.
Changes to Wing Pitching Moment As mentioned in Section 2.2.3, the addition

of a winglet will have an influence on the pitching moment of the wing. This may
lead to extra trim drag, as the tail may have to create larger forces to trim the
aeroplane.
To assess the implications of these changes in pitching moment, they would have
to be compared to the pitching moment changes due to a moving CG of the aircraft
and the required tailplane deflections. This information is not available for this
project, so the changes are simply stated to be used on further assessment.
Tip α PM Change
Hoerner: 3deg −10.47 Nm -
6deg −12.31 Nm -
Org. Winglet: 3deg −10.89 Nm -3.86%
6deg −12.80 Nm -3.83%
Table 6.1: Effects of the winglet on the wing 1/4 chord pitching moment
89
Case Winglet normal force BM at Tip joint BM Wing root

◦
Hoe, α = 3 - - 120.32 Nm
Wlet, α = 3◦ 6.17 N 0.54 Nm 126.99 Nm
Change 5.25%
Hoe, α = 6◦ - - 188.92 Nm
◦
Wlet, α = 6 8.30 N 0.73 Nm 196.63 Nm
Change 3.92%
◦
Hoe, α = 10 - - 272.06 Nm
◦
Wlet, α = 10 10.45 N 0.92 Nm 279.75 Nm
Change 2.75%
◦ ◦
Wlet, α = 6 , β = 6 10.88 N - ≈ 207.11 Nm
Wlet, α = 6◦ , β = 10◦ 10.88 N - ≈ 216.03 Nm
Table 6.2: Load summary for the main flight conditions on the 1/10th
scale wing at Re=1.6 million
6.3.4
Effect on Wing Structure
The installation of a winglet to an existing wing will change the loads the wing
structure has to withstand. One main argument for a winglet as compared to a
simple span extension is the reduced effect on the wing root bending moment (BM)
because the lift vector of the winglet is in spanwise direction. This results in a
smaller moment arm about the wing root. Nevertheless there will be an increase
in bending moment with the winglet installed. This has to be assessed against
the wing loading capabilities to determine if it is safe or if reinforcements will be
required.
The current winglet design was shown in the last sections only to be moderately
effective. Therefore, it is expected, that the additional loads generated by it will be
relatively small. Table 6.2 lists some results for the main flight conditions.
The loads on the real wing are expected to be approximately 5 times larger, as
the difference in Reynolds number is about a factor of 5. A test run at the flight
Reynolds number from Section 5.2.6 confirms this.
When comparing the increase in bending moment with the L/D plot in Figure 6.4,
it correlates with the findings there. The winglet is most effective at low angles of
attack producing an increase in L/D of 1.37%. The corresponding increase in wing
root bending moment is 5.25%. At higher angles of attack the L/D gains due to the
winglet reduce, as does the effect on the bending moment. This is supported by
Figure 6.12, where the pressure distribution on the wing along the quarter chord
line is plotted for α = 3◦ and 10◦ . The figure also shows, that only the outboard
90
1.5
1
−Cp
0.5
Hoerner tip
Winglet
0
−0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
Span station, α=10deg
0.7
0.6
0.5
0.4
−Cp
0.3
0.2
Hoerner tip
0.1 Winglet
0
−0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
Figure 6.12: Pressure distribution on the wing at 1/4 chord
wing loading is affected by the winglet, which in turn produces the largest increases
in bending moment due to the large moment arm. In any case, simply by inspection,
the loads generated by the winglet will never exceed the reserves in the structure
for the 1.82 m extension that is installed on the larger versions of the aircraft (see
Section 3.1).
The normal force on the winglet itself is small with about 1.5% of the total lift forces
generated by the wing for all cases. Incidentally, the winglet area is approximately
1.6% of the total wing area. These small loads will not present any difficulties in
designing a structure for the winglet, even with conservative load factors.
The sideslip causes only moderate changes to the bending moment compared to
the cases without sideslip (5.6% for α = 6◦ , β = 6◦ ) due to the higher loading of the
winglet. All values here are approximate as the wing area is larger for the sideslip
cases. As the wing was rotated inside the domain about the Z-axis, the gap at the
back between the symmetry plane and the wall had to be filled. This fillet is larger
than the area lost at the front because the wing was rotated about the quarter chord
point. There is no further increase in winglet normal load at high sideslip because
of the stalling of the winglet.
The final question regards the effect of a stalled winglet on the wing structure. The
stalling at high sideslip angles will most likely cause buffeting at the wing tip. This
may excite vibrations in the wing, if the frequencies are close to the natural fre-
91
6. Results and Discussion of the Original Winglet 6.4. Summary
quencies of the wing. In the literature, there is no evidence of this being a problem,
but for a possible certification this has to be investigated. A theoretical assessment
is very complicated, as it requires the frequencies of the vibrations in the winglet as
well as the magnitude. Also a complete structural model of the wing is required to
determine the effects. As before, it will be easier to flight test a prototype winglet
and carefully monitor the effects as stalling is approached. These tests will also be
required by the certifying authority.
6.4 Summary
The impact of the added winglet on the aircraft characteristics has been assessed.
It was found that the design will not add any substantial benefit to the airframe.
The consequences of the strong leading edge sweep in combination with the rather
small dimensions of the winglet limit the induced drag improvements to 1.3%.
Cleaning the aircraft has probably the same effect. On the other hand, neither the
stalling characteristics nor the handling properties seem to be strongly affected.
The impact on the wing structure is moderate, considering that the wing can han-
dle a 1.8m extension for the larger Metro variant. Therefore, it can be concluded
with the data available, that the addition of the winglet for aesthetic reasons should
not produce a safety hazard for the aircraft. This assessment excludes the possi-
ble generation of flutter at high sideslip flight due to winglet stalling, which could
not be investigated with the current methods. Finally, the clear indications of the
reasons for the limited performance of the winglet, as presented in this chapter,
should allow for an improved design. This will be attempted in the next chapter.
92
7
Improved Winglet Design
7.1 Introduction
In the last chapter it was determined, that the original winglet design does not
create a hazard to the aircraft performance. On the other hand, it also generated
only minimal improvements. Several design flaws were identified to be responsible
for this. In the following section, an attempt will be made to use the design tools
from Section 2.2.4 to improve the winglet design. Four iterations were briefly tested
before arriving at the current new winglet design. Only this ‘final’ design will be
discussed together with the reasoning behind it. Due to time limitations several
issues remain (mainly the shape of the junction region), but these have to be left
for further work.
Additionally, the replacement of the Hoerner wing tip with a square tip, while re-
taining the original winglet, was investigated. This was done to assess the possibil-
ity of removing the complicated geometry of the Hoerner tip and to replace it with
a simple square tip. If successful, this would reduce cost and effort during con-
struction of the new tip. The square tip showed a higher drag in the order of 1%,
so it was decided not to change too many variables for this project and to retain
the Hoerner tip. This keeps the meshes and the results more comparable to the
original winglet.
7.2 Design Improvements

This section introduces the new design proposal. It will be shown, that it deliv-
ers a markedly improved performance over the original design. Firstly, the de-
sign methodology is discussed and then a detailed performance assessment is pre-
sented. Possible influences on the handling qualities could not be investigated in
detail due to time limitations, this will have to be left for future work.
93
7. Improved Winglet Design 7.2. Design Improvements
7.2.1
Design Methodology
The following factors were identified in Chapter 6 to cause the relatively poor per-
formance of the original winglet:
1. Missing aerofoil section
2. High leading edge sweep
3. Small winglet size
Additionally, the low outboard loading of main wing limits the effectiveness of the
winglet as mentined in Section 2.2.4. This is a design feature of the aircraft, caused
by the high taper ratio and the wing washout and cannot be changed. The other
three factors are accessible to design improvements and were altered during a few
iterations to the final design concept. These modifications were:
1. A NACA 64009 symmetrical aerofoil was used on the winglet. In [18], a sym-
metrical section was shown to be as effective as a cambered section for a gen-
eral aviation wing. The section was chosen from the same family as the main
wing sections (NACA 6 series) with 9% thickness and the winglet was installed
to the tip with zero toe angle, both as proposed in the reference. To create
an usable CFD model that did not require major changes in meshing strategy,
the aerofoil section was modified with a blunt trailing edge. The width of this
blunt edge was chosen to be 1% of the thickness of the section to fit into the
current trailing edge mesh sizes.
2. The leading edge sweep was reduced to 29 degrees and the trailing edge was
made vertical. This should improve the leading edge vortex problem of the
original winglet. Technically, at the low flight Mach number no sweep is re-
quired [18], but for aesthetic reasons it has been retained.
3. The winglet design principles state, that the winglet efficiency depends on its
height. Therefore, to achieve further performance gains, the height of the
winglet was increased by 20cm. This also enlarged the aspect ratio from 1.4
to 1.75, which will reduce the induced drag of the winglet in the drag polar
equation (2.1). The amount of height increment was based on some testing
of the resulting bending moment gains, where the chosen amount presented
an optimum within the tested size range (maximum winglet height of 80cm).
The winglet size was limited arbitrarily to honour the original design intend
and to limit the structural impact. The resulting larger bending moment in the
wing will be shown later to be small, compared to the positive influence of the
larger winglet on the efficiency of the entire wing.
94
4. The cant angle was increased to 20 degrees to achieve a flight cant angle of 15
degrees due to the wing dihedral. This corrects a design error of the original
winglet, which did not account for the wing dihedral. The cant angle change
will result in the normal force vector of the winglet being tilted upwards by the
same amount and thus the lift component of this force will be increased. It ap-
pears, that a cant angle of 15◦ is the optimum combination between efficiency
gains and bending moment increases [46].
7.2.2
Geometry of the New Winglet
The new winglet design was frozen with the dimensions listed in Table 7.1. The new
geometry is shown in Figure 7.1, where it is also compared to the original design.
The figure also shows a different design of the model in the wing-winglet junction
area. To be able to guarantee the integrity of the aerofoil section in SOLIDWORKS,
it was necessary to build a socket onto the Hoerner tip (in blue, and also shown by
the spanwise contours in Figure 7.1(c)), on which the new winglet could be lofted (in
red and with horizontal contours). This method of construction means, that a small
portion of the winglet (blue section) is added to the wing region of the CFD mesh,
because the mesh regions are determined by parts in the geometry definition. Any
force calculations on the winglet alone will therefore yield a smaller value in the
order of 10%, which is the area not included in the winglet mesh region. This does
not affect the overall wing forces, which are a sum of all mesh regions.
Section: NACA 64009 Cant angle: 15◦

Root chord: 540 mm Root toe angle: 0◦
Tip chord: 200 mm Tip toe angle: 0◦
Span: 650 mm Leading edge sweep Λ: 29.5◦
Area: 0.25 m2 Taper ratio λ: 0.335
Aspect ratio: 1.75 MAC: 390 mm
Table 7.1: New Winglet data
95
(a) New winglet design (top) and original shape (bottom)
(b) top view (c) Tip-Winglet Junction
Figure 7.1: New Winglet Geometry (Model 1/10th scale)
7.2.3
Performance Assessment
The new winglet geometry generates a significant performance improvement. Table

7.2 and Figure 7.2 show a more than threefold increase in maximum L/D at α = 4◦ ,
while the bending moment at the wing root grows by only 1.5%. Most interestingly,
the viscous drag for the new winglet is only slightly higher than the original winglet,
despite almost twice the surface area. The drop in viscous drag at α = 5◦ has also
disappeared.
96
−3
x 10
22 8.6
Hoerner
8.5 Hoerner
New Design
New Design
21 Org. Winglet
8.4 Org. Winglet
8.3
20
8.2
L/D
CD
19 8.1
18
7.9
7.8
17
7.7
16 7.6
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10
α CL
(a) L/D (b) Viscous drag coefficient
Figure 7.2: Performance comparison, the data for the new design is af-
fected by stalling due to the low Reynolds number above
α = 6◦
Tip Area L/Dmax Change Root BM Change

Hoerner - 20.48 - 143.83 Nm -
2
Original Wlet ≈ 0.14 m 20.76 1.37% 150.98 Nm 4.97%
2
New Winglet ≈ 0.25 m 21.52 5.08% 153.37 Nm 6.63%
Table 7.2: Wing tip performance comparison for the 1/10th scale model
(Changes are relative to Hoerner wing tip)
These results compare well with literature [18][14][16]. There, the designs achieve
gains in L/D of 10% using winglets about twice the size (0.47 m2 [14] vs. ≈ 0.25 m2
for this project). Therefore, the efficiency of the new winglet design is in line with
the industry standard. As mentioned before, the area of the winglet has been kept
smaller to honour the initial design intentions of a purely aesthetic improvement.
Further, the non-optimal wing-winglet juncture has room for improvement as dis-
cussed below.
Next, the flow fields around the two winglets will be compared, using similar meth-
ods as in Section 6.3.2. An attempt will be made to explain the performance gains
and the findings for the viscous drag component.
Flow Field The flow fields around the two winglets are compared in Figure 7.3.
Figures 7.3(a) and 7.3(b) contain exactly the same streamline source. It produces
very different flow patterns. The original winglet created a strong spanwise flow,
deflecting the streamlines upwards on both sides and forming the dominating lead-
ing edge vortex. In contrast, the new winglet shows only a small vortex, which is
generated at the edge of the wing-winglet joint and otherwise no significant verti-
cal flow deflections. The reduction in sweep clearly worked as intended. In Figure
97
(a) (b)
(c) (d)
Figure 7.3: Vortex comparison on the two winglets at α = 6◦ and zero

sideslip. In (d) the flight Reynolds number was used to avoid
the stalling near the junction vortex. The pattern in (c) is not
affected by Reynolds number effects.
7.3(d) we can see that the full span of the winglet is active, with a tip vortex pro-
duced right at the outer edge. On the original winglet in Figure 7.3(c), significant
parts of the outer span were stalled in the vicinity of the leading edge vortex (see
Section 6.3.2). Figure 7.3(d) also shows the junction vortex caused by the non-
blended junction. This junction vortex appears to be more defined for the new
design in Figure 7.3(d), as compared to Figure 7.3(c).
98
(a) (b)
(c) (d)
Figure 7.4: Pressure and wall shear comparison on the two winglets at
α = 6◦ and zero sideslip. The new winglet is plotted at the
flight Reynolds number to avoid the stalling. The limits for
the pressure were scaled by a factor of 5 to result in compa-
rable images.
Figure 7.4 shows comparisons between the two designs in terms of pressure and
wall shear distribution. Note, the oscillatory structures near the leading edge of the
wing tip are a result of the poor geometry quality produced by SOLIDWORKS. The
surface of the tip in this region is deformed, which causes these changes in pressure
and shear. No method exists in SOLIDWORKS to improve on that, as discussed in
Section 4.3.1.
Comparing the pressure fields in Figures 7.4(a) and 7.4(b), the new design has a
spanwise uniform distribution with smaller chordwise gradients, as it would be
expected for a conventional wing with limited sweep. Only the junction vortex
causes minor disturbances near the root of the winglet.
99
0.9
0.8
0.7
0.6
0.5
−Cp
0.4
0.3
0.2
0.1
0
0.7 0.8 0.9 1 1.1 1.2 1.3 1.4
Figure 7.5: Spanwise pressure distribution on the new winglet at quarter

chord for α = 3◦
Figure 7.5 shows the quarter chord pressure distribution along the winglet. The low
aspect ratio and the reducing strength of the main wing tip vortex along the span
of the winglet results in a constant drop in span loading to the winglet tip. These
losses are more pronounced than these of a conventional, high aspect ratio wing
(see Fig. 6.12), but this is expected for a small winglet. As a result, the entire new
winglet contributes to the generation of the normal force and its pressure distribu-
tion fulfils the design objectives from Section 2.2.4, where a higher loading at the
root was specified. Compared to this, the original winglet only creates significant
amounts of normal force near the leading edge vortex, which only acts on about
a quarter of the entire surface. This is the first reason for the performance gains
shown above.
The second factor is the wall shear distribution in Figures 7.4(c) and 7.4(d). Here
the original winglet shows extended regions of high shear forces, where the leading
edge vortex is in contact with the surface. The new design has high shear only on
the leading edge and low, uniform values on the main surfaces. This explains the
almost identical viscous drag for both designs as shown in Figure 7.2(b), despite
the large difference in wetted area.
The design objectives from Section 2.2.4 state, that the winglet should produce
the required normal force as efficiently as possible near the design conditions and
should not stall at off design situations. For the current case, landing is an off-
design condition. At this flight condition, with α = 6◦ , the new winglet works at
high lift coefficients, as can be seen from the strong low pressure peaks near the
leading edge in Figure 7.6 1 . This graph can be used to estimate the angles of
attack on the winglet by comparing the section pressure distributions with inviscid
solutions, as discussed in Section 5.2.1. A very rough estimate for the angle of
attack near the winglet root (black curve in Fig. 7.6) is a section angle of attack
of 4.7 degrees (from JAVAfoil, based on the minimum Cp ), which corresponds to
1
These pressure distributions are also given in separate plots in Appendix C
100
−3
z=0.07m
z=0.086m
−2.5 z=0.105m
z=0.125m
−2
−1.5
Cp
−1
−0.5
0.5
1
0.05 0.06 0.07 0.08 0.09 0.1 0.11
x−station
Figure 7.6: Pressure distributions on the new winglet for α = 6◦
≈ 11◦ on the 3-d winglet 2 . The peaks in minimum Cp reduce along the span of
the winglet, which reinforces the theory of a reduction in tip vortex strength (and
corresponding angle of attack on the winglet) with increasing distance from the
wing surface.
The new winglet fulfils the no-stall requirements, although it will not be able to
handle any significant increases in angle of attack due to additional sideslip without
stalling. This is discussed further in the next section.
For cruise we can extrapolate from the pressure distributions in Figure 7.6 by as-
suming a linear lift curve. It results in a winglet angle of attack of 3-5 degrees
depending on the span station for a wing angle of attack of 3 degrees. This puts
the winglet into the region of maximum L/D for a typical wing [3] and fulfils the
objective of highly efficiency normal force generation.
Stalling Behaviour The original winglet has one advantage: Due to its delta wing
behaviour, it is capable of tolerating higher angle of attack during sideslip manoeu-
vres and stalls later. In Section 6.3.3 it was shown that at β = 6◦ the original winglet
was not yet fully stalled. The new design, with its more conventional flow field, is
fully stalled in this condition at both the simulation- and the flight Reynolds num-
ber. As with the original winglet, this flow separation does not influence the flow
over the wing, such that the ailerons will not be affected by this stalling. Nev-
ertheless, this off-design behaviour needs to be further investigated in terms of
structural flutter.
2
This estimate contains large uncertainties, as the lifting line equation is very inaccurate below an
aspect ratio of 3. Nevertheless, it should be precise enough to illustrate the general trend
101
7. Improved Winglet Design 7.3. Summary
If required, it will be possible to trade off some efficiency for a delay in stalling by
toeing the winglet out slightly. This will be a topic for future work.
Effect on Wing Pitching Moment The changes in wing pitching moment are listed
in Table 7.3 together with the data for the original winglet. The larger size and the
more efficient force generation of the new winglet result in a 1-2% larger variation
in pitching moment. As before these values are simply stated for future use.
Tip α PM Change
Hoerner: 3deg -10.47Nm -
6deg -12.31Nm -
Org. Winglet: 3deg -10.89Nm 3.86%
6deg -12.80Nm 3.83%
New Winglet: 3deg -11.04Nm 5.16%
6deg -13.14Nm 6.32%
Table 7.3: Effects of the new winglet on the wing 1/4 chord pitching mo-
ment (Changes relative to Hoerner tip)
7.3 Summary
This chapter presented a new, improved design for the winglet. It was possible to
achieve a performance gain over the original winglet by a factor of three. At the
same time, the wing bending moment increased by a factor of only 0.3. The new
design also has a similar skin friction drag as the original winglet, despite twice
the surface area. This data shows, that careful application of the design objectives
presented in Section 2.2.4 will result in a well performing geometry. Fine-tuning of
the wing-winglet juncture will further improve the overall design.
102
8
Future Work
A few topics could not be completed in the limited time frame of this project,
because they would require a whole new line of studies or methods that are not
easily accessible, as discussed in the previous chapters. These topics,
• Wing-winglet junction design
• Winglet stalling and flutter
• Trim drag behaviour and handling qualities
• Further CFD verifications
have to be addressed before one can be completely confident about the character-
istics of the original and new winglet.
To optimise the wing-winglet joint, a new model in a more flexible CAD application
is necessary. This will allow to control the geometry of the joint more effectively
and produce an overall better quality model. Then the shape of the junction should
be critically investigated with regards to a fully blended design and how it will
further improve the performance. Ideas about possible shapes can be found on
current business jets and other high performance aircraft. Also, the shape of the
wing tip itself should be re-visited to determine, if a simpler, but rounded design
as proposed in Section 7.1 can yield a further drag reduction and at the same time
a cost improvement for the manufacturing process.
Next, it has to be established if any flutter due to stalling of the winglet occurs and
if it will have any negative effects on the aircraft. This is probably the most involved
task ahead, as it will require flight testing of a prototype. If there are problems at
high sideslip angles, the winglet will have to be re-designed by adjusting the toe
angle to reduce the angle of attack at these flight conditions. This will also have
an impact on the performance gains as the winglet will not operate at optimum
conditions during cruise. During these tests, the effect on the trim drag and the
other handling qualities can be assessed as discussed in Section 6.3.3.
103
8. Future Work
To further improve the confidence in the CFD results a full mesh convergence
study is required. This will have to include the preparation of a mesh with half
the grid spacing to assess the level of grid convergence achieved as discussed in
Section 5.3.1. Also an attempt should be made to run the cases at the correct flight
Reynolds number, which will require a rework of the boundary layer mesh. Fi-
nally, if the hardware resources are available, a full half model of the aircraft could
be simulated to obtain results for the trim drag and other possible interferences,
which could not be investigated with the wing model alone.
104
9
Conclusion
This report discussed the aerodynamic characteristics of a winglet design for a

Fairchild Merlin III 8-seat, twin turboprop aircraft. The winglet was designed by
the aircraft owner to improve the aesthetics of the aeroplane. To ensure, that the
winglet will not have any negative effect on the aircraft, a series of comparative CFD
simulations was carried out. A hybrid mesh with unstructured surface- and volume
grids, in addition to a structured boundary layer mesh was used on the wing, which
was isolated from the fuselage.
Confidence was gained by comparing parts of the results with accepted compu-
tational and empirical methods. Good agreement was achieved for 2-d pressure
distributions, as well as for the 3-d lift- and drag curves. The predictions for the
friction drag were within one drag count of the corresponding empirical methods.
Further, no significant problems with solution sensitivity on various parameters,
except the mesh size, was found. The solutions were mesh dependent, but con-
verged with mesh size. More work on a larger computer is required to carry out a
full mesh dependency study.
The results for the winglet showed a 1.3% increase in L/D of the wing at α = 4◦ .
A strong leading edge vortex created by the high sweep of the winglet (comparable
to a delta wing) severely limited the performance gains by inefficient lift genera-
tion and high friction drag. Further the missing aerofoil section resulted in high
flow gradients with separation near the leading edge. Stalling of the winglet at high
sideslip angles did not cause any interference with the ailerons. The stalling be-
haviour of the main wing was not significantly changed at high angles of attack.
The limited efficiency of the winglet resulted in small air loads on it and only mod-
erate wing bending moment gains. The question, whether stalling of the winglet
will cause any structural flutter, could not be answered with the methods available.
Based on the findings for the original winglet, a new, improved design was tested.
It featured less sweep and twice the surface area, with a NACA 64009 aerofoil.
This new design improved the cruise L/D by 5%, which is a threefold improvement
over the original winglet. At the same time, the skin friction drag was found to be
similar, despite the large change in area. Finally, the wing bending moment was
only increased by an additional 2% in cruise. The performance gains of the new
winglet design were comparable to industry standards reported in literature.
105
Bibliography
[1] Ilan Kroo. Drag due to lift: Concepts for prediction and reduction. Annual
Review of Fluid Mechanics, 33:587, 2001.
[2] C. D. Perkins and R. E. Hage. Airplane Performance Stability and Control. John
Wiley & Sons, Inc, New York, 1967.
[3] John D. Anderson. Fundamentals of Aerodynamics. McGraw-Hill, 2007.
[4] R Eppler. Induced drag and winglets. Aerospace Science and Technology, (1):3–
15, 1997.
[5] John E. Yates and Coleman duP. Donaldson. A fundamental study of drag
and an assessment of conventional drag-due-to-lift reduction devices. NASA
Contractor Report 4004, 1986.
[6] S.A.Kravchenko. Wing tip lifting surfaces - aerodynamic design and compara-
tive analysis. Aircraft Engineering, Technology, and Operations Congress, 1st,
Los Angeles, CA, Sept 19-21, AIAA-1995-3909, 1995.
[7] Aviation Partners INC. Company web page, 2010. Available from World Wide
Web: http://www.aviationpartners.com/future.html.
[8] Ilan Kroo. Nonplanar wing concepts for increased aircraft efficiency. VKI lec-
ture series on Innovative Configurations and Advanced Concepts for Future
Civil Aircraft, 2005.
[9] NASA. Kc-135 winglet progam review. NASA Conference Publication 2211,
1982.
[10] Staff of Boeing Commercial Airplane Company. Selected advanced aerodynam-

ics and active controls technology concepts development on a derivative b-747
aircraft. NASA Contractor Report 3295, 1980.
[11] Staff of the Douglas Aircraft Company. Dc- 10 winglet flight evaluation. NASA
Contractor Report 3704, 1983.
[12] Keisuke Asai. Theoretical considerations in the aerodynamic effectiveness of

winglets. JOURNAL OF AIRCRAFT, 22(7), 1985.
[13] Richard T. Whitcomb. A design approach and selected wind-tunnel results at

high subsonic speeds for wingtip mounted winglets. NASA TN D-8260, 1976.
106
Bibliography Bibliography
[14] Bruce J. Holmes, Cornelis P. van Dam, Philip W. Brown Deal, and Perry L. Flight
evaluation of the effect of winglets on performance and handling qualities of a
single-engine general aviation airplane. NASA Technical Memorandum 81892,
1980.
[15] ESDU. Aerodynamic principles of winglets, 1998. Available from World Wide
Web: www.demec.ufmg.br/Cea/Disciplinas/AerodinamicaAplicada/
artigo28.pdf.
[16] Cornells P. van Dam, Bruce J. Holmest, and Calvin Pitts. Effect of winglets on
performance and handling qualities of general aviation aircraft. JOURNAL OF
AIRCRAFT, VOL. 18, NO. 7, 1981.
[17] Krzysztof Kubrynski. Wing-winglet design methodology for low speed appli-
cations. 41st Aerospace Sciences Meeting and Exhibit, Reno, Nevada, Jan. 6-9,
AIAA-2003-215, 2003.
[18] III Marchman, J. F., D. Manor, and H. F. Faery. Whitcomb winglet applications
to general aviation aircraft. AIAA, Aircraft Systemsand Technology Conference,
Los Angeles, Calif., AIAA-1478, 1978.
[19] Mark D. Maughmer. Design of winglets for high-performance sailplanes. JOUR-

NAL OF AIRCRAFT, Vol. 40, No. 6, 2003.
[20] Mark D. Maughmer and Peter J. Kunz. Sailplane winglet design. Available
from World Wide Web: pagesperso-orange.fr/air-club-adour/modele/
.../about_winglets_design.pdf.
[21] John D. Anderson. Computational Fluid Dynamics. McGraw-Hill, 1995.
[22] Antony Jameson and Massimiliano Fatica. Using computational fluid dynamics
for aerodynamics. Stanford University, 2003.
[23] Ansys. Ansys cfx 12 manuals, 2009. Available from World Wide Web:
CFXhelpfiles.
[24] Dimitri J. Mavriplis, John C. Vassberg, Edward N. Tinoco, Mori Mani, Olaf P.
Brodersen, Bernhard Eisfeld, Richard A. Wahls, Joseph H. Morrison, Tom Zick-
uhr, David Levy, and Mitsuhiro Murayama. Grid quality and resolution issues
from the drag prediction workshop series. JOURNAL OF AIRCRAFT, Vol. 46(3),
2009.
[25] J. M. Luckring, M. J. Hemsch, and J. H. Morrison. Uncertainty in computational

aerodynamics. 41st AIAA Aerospace Sciences Meeting, AIAA-0409, 2003.
[26] AIAA. Aiaa drag prediction workshops, 2009. Available from World Wide Web:
http://aaac.larc.nasa.gov/tsab/cfdlarc/aiaa-dpw/index.html.
107
[27] Amarvir Chilka and Ashish Kulkarni. Modeling turbulent flows in fluent, 2010.
Available from World Wide Web: www.fluent.com/software/university/
blog/turbulent.pdf.
[28] Elizabeth Lee-Rausch, Neal Frink, Dimitri Mavriplis, Russ Rausch, and William
Milholen. Transonic drag prediction on a dlr-f6 transport configuration using
unstructured grid solvers. 42nd AIAA Aerospace Sciences Meeting and Exhibit,
Reno, Nevada, AIAA-554, 2004.
[29] David C. Wilcox. Turbulence Modeling for CFD. DCW Industries, 2006.
[30] F. R. Menter. Two-equation eddy-viscosity turbulence models for engineering

applications. AIAA JOURNAL, 32(8), 1994.
[31] S. V. Utyuzhnikov. Generalized wall functions and their application for simula-
tion of turbulent flows. International Journal for Numerical Methods in Fluids,
47, 2005.
[32] J. Schlueter. Toward the prediction of turbulent boundary layers using a cou-
pled rans les method. Center for Turbulence Research Proceedings of the Sum-
mer Program, 2006.
[33] S. Hoerner. Aerodynamic shape of wing tips, 1952. Available

from World Wide Web: http://oai.dtic.mil/oai/oai?verb=getRecord\
&metadataPrefix=html\&identifier=ADA800374.
[34] S. Hoerner. Fluid-Dynamic Drag. self published, 1965.
[35] Ira H. Abbott and Albert E. von Doenhoff. Theory of Wing Sections. Dover,
1958.
[36] David Lednicer. The incomplete guide to airfoil usage, 2010. Available from
World Wide Web: http://www.kams.net.au/AirfoilsUsed2.htm.
[37] Wikipedia. Murphys law, 2010. Available from World Wide Web: http://en.
wikipedia.org/wiki/Murphy27s_law.
[38] Timothy J. Baker. Mesh generation: Art or science? Progress in Aerospace

Sciences, 41:29–63, 2005.
[39] Ansys. Ansys workbench meshing helpfiles, 2009. Available from World Wide
Web: Workbenchhelpfiles.
[40] Martin Hepperle. Javafoil, 2007. Available from World Wide Web: www.
mh-aerotools.de/airfoils/javafoil.htm.
[41] Thomas Melin. Tornado vortex lattice method, 2000. Available from World
Wide Web: www.redhammer.se.
108
[42] Thomas Melin. A Vortex Lattice MATLAB Implementation for Linear Aerody-
namic Wing Applications. PhD thesis, 2000.
[43] Kai Lehmkuehler. Aerdynamic design of a next generation supersonic trans-

port. Technical report, Aerospace Design2, University of Sydney, 2009.
[44] James L. Thomas, Boris Diskin, and Christopher L. Rumsey. Toward verification
of unstructured-grid solvers. JOURNAL OF AIRCRAFT, Vol. 46, No. 12, 2008.
[45] J. B. Barlow, Jr. W. H. Rae, and A. Pope. Low speed wind tunnel testing. Wiley,
3rd edition, 1999.
[46] Harry H. Heyson, Gregory D. Riebe, and Cynthia L. Fulton. Theoretical para-
metric study of the relative advantages of winglets and wing-tip extensions.
NASA Technical Paper 1020, 1977.
109
A
List of Mesh Settings
Sizing:
Size function: Curvature
Curvature Normal Angle: 10◦
Min. size: 1.5e−4 m
Max face size: 0.2 m
Growth rate: 1.25
Inflation layers:
Option: First layer thickness
First layer height: 6e−6 m (y + ≈ 3)
Max. layers: 22
Growth rate: 1.28
Algorithm: Pre
Maximum Angle: 140◦
Table A.1: Global mesh settings for all cases
Region Size Other

−4
Leading edge 8.0e m Curvature Angle: 18◦
p0 2.6e−3 m -
−3
p1 4.2e m -
p2 6.0e−3 m -
−3
p3 4.2e m -
p4 2.6e−3 m -
−4
Trailing edge main wing 3.0e m -
Table A.2: Common mesh settings for all cases
110
Appendix A. List of Mesh Settings
Region Size Other

Wingtip 2.0e−3 m Curvature Angle: 18◦
Wingtip TE 2.0e−4 m Curvature Angle: 0.1◦
Wingtip p4 1.5e−3 m -
Table A.3: Mesh sizing for Hoerner wing
Region Size Other

Wingtip 2.0e−3 m Curvature Angle: 18◦
Winglet 2.0e−3 m Curvature Angle: 18◦
Winglet trailing edge 3.0e−4 m -
−3
Winglet top 2.5e m Growth rate: 1.22
Table A.4: Mesh sizing for wing with original winglet
Region Size Other

−3
Wingtip 2.0e m Curvature Angle: 18◦
Winglet 1.2e−3 m Curvature Angle: 15◦
Winglet trailing edge 2.0e−4 m -
Winglet top 2.5e−3 m Growth rate: 1.22
Table A.5: Mesh sizing for wing with new winglet design
111
B
Selected Pressure Distributions
on the Original Winglet
Figure B.1: Locations of pressure distributions
−1.5 −2.5
−2
−1
−1.5
Cp
Cp
−0.5 −1
−0.5
0
0
0.5 0.5
0.04 0.06 0.08 0.1 0.12 0.06 0.08 0.1 0.12
x−station, z=ref x−station, z=0.1m
−2 −1
−1.5
−0.5
−1
Cp
Cp
−0.5
0
0
0.5 0.5
0.08 0.09 0.1 0.11 0.12 0.1 0.11 0.12 0.13
x−station, z=0.2m x−station, z=0.3m
Figure B.2: Winglet pressure distributions for α = 3◦ , β = 0◦
112
Appendix B. Selected Pressure Distributions
on the Original Winglet
−3 −3.5
−2.5 −3
−2.5
−2
−2
−1.5
Cp
Cp
−1.5
−1
−1
−0.5
−0.5
0 0
0.5 0.5
−0.02 0 0.02 0.04 0.06 0 0.01 0.02 0.03 0.04 0.05
−2 −1
−1.5 −0.8
−1 −0.6
Cp
Cp
−0.5 −0.4
0 −0.2
0.5 0
0.02 0.03 0.04 0.05 0.06 0.04 0.045 0.05 0.055 0.06
−3.5 −3.5
−3 −3
−2.5 −2.5
−2 −2
Cp
Cp
−1.5 −1.5
−1 −1
−0.5 −0.5
0 0
0.5 0.5
−0.08 −0.06 −0.04 −0.02 0 −0.05 −0.04 −0.03 −0.02 −0.01 0
−2 −1.4
−1.2
−1.5
−1
−1 −0.8
Cp
Cp
−0.5 −0.6
−0.4
0
−0.2
0.5 0
−0.04 −0.03 −0.02 −0.01 0 −20 −15 −10 −5 0 5
x−station, z=0.2m x−station, z=0.3m −3
x 10
113
C
Selected Pressure Distributions on the
New Winglet
−3 −2.5
−2.5 −2
−2
−1.5
−1.5
−1
Cp
−1
C
−0.5
−0.5
0
0
0.5 0.5
1 1
0.05 0.06 0.07 0.08 0.09 0.1 0.11 0.06 0.07 0.08 0.09 0.1 0.11
−2 −1.5
−1.5
−1
−1
−0.5
Cp
−0.5
C
0
0
0.5
0.5
1 1
0.08 0.09 0.1 0.11 0.12 0.08 0.085 0.09 0.095 0.1 0.105 0.11
Figure C.1: New winglet pressure distributions for α = 3◦ , β = 0◦
114
D
Limitations of inviscid methods
for winglet design
The strongly 3-dimensional flow on the original winglet demonstrated some limita-
tions of linear, inviscid panel methods for this kind of flow field. Figure D.1 shows
the L/D predictions of TORNADO for the three main wing tips investigated in this
project. The absolute values are distorted by the differences in drag prediction, but
this is not of major concern. The L/D improvement between the Hoerner tip and
the new winglet is predicted as about 4.9%, which is very close to the CFD results.
On the other hand, the results for the highly swept original winglet are strongly
over-predicted. According to Tornado, there is hardly any difference between the
two winglets.
The reason for this is shown in Figure D.2. The figure shows the computed inviscid
pressure distribution for both cases. For the new winglet it looks very similar to
the the CFD results. On the original winglet, however, a ’well behaved’ pressure
distribution is predicted, completely missing out on the leading edge vortex created
by the highly swept winglet. It appears in this solution, that the original winglet is
as efficient in creating a normal force as the new design, with differences only due
to the size.
This shows that a panel method introduces large errors for this kind of highly
swept winglets. Therefore, care must be taken when interpreting results for ex-
treme geometries without any alternative prediction method. The CFD solution
is clearly superior for these cases, while for the more conventional new winglet a
panel method solution is sufficient for preliminary design studies.
115
Appendix D. Limitations of inviscid methods
for winglet design
24
23
22
L/D
21
20
19 Hoerner
Original Wlet
New Wlet
18
1 2 3 4 5 6 7 8 9
α
Figure D.1: Tornado L/D comparison
Figure D.2: Tornado Cp distributions
116
E
CFX Solver settings
The complete solver settings are given for α = 3◦ for reference below:
LIBRARY: EXTERNAL SOLVER COUPLING:

MATERIAL: Air at 15 C STD sea level Option = None
Material Description=Std sea level,15 deg,101325pa END
Material Group = Air Data,Constant Property Gases END
Option = Pure Substance DOMAIN: Domain
Thermodynamic State = Gas Coord Frame = Coord 0
PROPERTIES: Domain Type = Fluid
Option = General Material Location = B625
EQUATION OF STATE: BOUNDARY: bot
Density = 1.225 [kg m^-3] Boundary Type = INLET
Molar Mass = 28.96 [kg kmol^-1] Location = bot
Option = Value BOUNDARY CONDITIONS:
END FLOW REGIME:
SPECIFIC HEAT CAPACITY: Option = Subsonic
Option = Value END
Specific Heat Capacity=1.0044E+03[J kg^-1 K^-1] MASS AND MOMENTUM:
Specific Heat Type = Constant Pressure Option = Cartesian Velocity Components
END U = 124.829 [m s^-1]
REFERENCE STATE: V = 0 [m s^-1]
Option = Specified Point W = 6.542 [m s^-1]
Reference Pressure = 101325 [Pa] END
Reference Specific Enthalpy = 0. [J/kg] TURBULENCE:
Reference Specific Entropy = 0. [J/kg/K] Eddy Length Scale = 0.1851 [m]
Reference Temperature = 15 [C] Fractional Intensity = 0.001
END Option = Intensity and Length Scale
DYNAMIC VISCOSITY: END
Dynamic Viscosity = 1.831E-05 [kg m^-1 s^-1] END
Option = Value END
END BOUNDARY: in
THERMAL CONDUCTIVITY: Boundary Type = INLET
Option = Value Location = in
Thermal Conductivity = 2.61E-02 [W m^-1 K^-1] BOUNDARY CONDITIONS:
END FLOW REGIME:
ABSORPTION COEFFICIENT: Option = Subsonic
Absorption Coefficient = 0.01 [m^-1] END
Option = Value MASS AND MOMENTUM:
END Option = Cartesian Velocity Components
SCATTERING COEFFICIENT: U = 124.829 [m s^-1]
Option = Value V = 0 [m s^-1]
Scattering Coefficient = 0.0 [m^-1] W = 6.542 [m s^-1]
END END
REFRACTIVE INDEX: TURBULENCE:
Option = Value Eddy Length Scale = 0.1851 [m]
Refractive Index = 1.0 [m m^-1] Fractional Intensity = 0.001
END Option = Intensity and Length Scale
THERMAL EXPANSIVITY: END
Option = Value END
Thermal Expansivity = 0.003356 [K^-1] END
END BOUNDARY: out
END Boundary Type = OUTLET
END Location = out
END BOUNDARY CONDITIONS:
FLOW: Flow Analysis 1 FLOW REGIME:
SOLUTION UNITS: Option = Subsonic
Angle Units = [rad] END
Length Units = [m] MASS AND MOMENTUM:
Mass Units = [kg] Option = Average Static Pressure
Solid Angle Units = [sr] Pressure Profile Blend = 0.05
Temperature Units = [K] Relative Pressure = 0 [Pa]
Time Units = [s] END
END PRESSURE AVERAGING:
ANALYSIS TYPE: Option = Average Over Whole Outlet
Option = Steady State END
117
Appendix E. CFX Solver settings
END END
END TURBULENCE MODEL:
BOUNDARY: side Option = SST
Boundary Type = SYMMETRY END
Location = side TURBULENT WALL FUNCTIONS:
END Option = Automatic
BOUNDARY: sym END
Boundary Type = SYMMETRY END
Location = sym INITIALISATION:
BOUNDARY: top INITIAL CONDITIONS:
Boundary Type = OUTLET Velocity Type = Cartesian
Location = top CARTESIAN VELOCITY COMPONENTS:
BOUNDARY CONDITIONS: Option = Automatic
FLOW REGIME: END
Option = Subsonic STATIC PRESSURE:
MASS AND MOMENTUM: END
Option = Average Static Pressure TURBULENCE INITIAL CONDITIONS:
Pressure Profile Blend = 0.05 Option = Intensity and Length Scale
Relative Pressure = 0 [Pa] EDDY LENGTH SCALE:
END Eddy Length Scale = 0.1851 [m]
PRESSURE AVERAGING: Option = Automatic with Value
Option = Average Over Whole Outlet END
END FRACTIONAL INTENSITY:
END Fractional Intensity = 0.001
END Option = Automatic with Value
BOUNDARY: wing END
Boundary Type = WALL END
Location = wing,wingtip END
BOUNDARY CONDITIONS: END
MASS AND MOMENTUM: END
Option = No Slip Wall OUTPUT CONTROL:
END BACKUP RESULTS: Backup Results 1
WALL ROUGHNESS: File Compression Level = Default
Option = Smooth Wall Option = Standard
END OUTPUT FREQUENCY:
END Elapsed Time Interval = 2 [h]
END Option = Elapsed Time Interval
BOUNDARY: winglet END
Boundary Type = WALL END
Location = winglet MONITOR OBJECTS:
BOUNDARY CONDITIONS: MONITOR BALANCES:
MASS AND MOMENTUM: Option = Full
Option = No Slip Wall END
END MONITOR FORCES:
WALL ROUGHNESS: Option = Full
Option = Smooth Wall END
END MONITOR PARTICLES:
END Option = Full
END END
DOMAIN MODELS: MONITOR RESIDUALS:
BUOYANCY MODEL: Option = Full
Option = Non Buoyant END
END MONITOR TOTALS:
DOMAIN MOTION: Option = Full
Option = Stationary END
END END
MESH DEFORMATION: RESULTS:
Option = None File Compression Level = Default
END Option = Standard
REFERENCE PRESSURE: END
Reference Pressure = 101325 [Pa] END
END SOLVER CONTROL:
END Turbulence Numerics = High Resolution
FLUID DEFINITION: Fluid 1 ADVECTION SCHEME:
Material = Air at 15 C STD sea level Option = High Resolution
Option = Material Library END
MORPHOLOGY: CONVERGENCE CONTROL:
Option = Continuous Fluid Length Scale Option = Aggressive
END Maximum Number of Iterations = 120
END Minimum Number of Iterations = 10
FLUID MODELS: Timescale Control = Auto Timescale
COMBUSTION MODEL: Timescale Factor = 1.0
Option = None END
END CONVERGENCE CRITERIA:
HEAT TRANSFER MODEL: Residual Target = 2e-06
Fluid Temperature = 15 [C] Residual Type = RMS
Option = Isothermal END
END END
THERMAL RADIATION MODEL: END
Option = None
118
View publication stats

Merlin300 Winglet Kai - Thesis

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Merlin300 Winglet Kai - Thesis

Uploaded by

Copyright:

Available Formats

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

Winglet Design for a Fairchild Merlin III using CFD Analysis

Research · July 2015

UAV flight testing and system identification View project

The user has requested enhancement of the downloaded file.

List of Tables viii

5 Verification of the CFD Simulation 53

6 Results and Discussion of the Original Winglet 75

7 Improved Winglet Design 93

8 Future Work 103

A List of Mesh Settings 110

B Selected Pressure Distributions

C Selected Pressure Distributions on the New Winglet 114

D Limitations of inviscid methods

E CFX Solver settings 117

2.1 Velocities on wingtip projected in the XY-plane . . . . . . . . . . . . . . . 7

3.1 Fairchild Merlin III VH-SSM . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

4.1 Estimate of the effect of flaps on wing CL . . . . . . . . . . . . . . . . . . 34

5.15 SST first blending function for α = 5◦ . . . . . . . . . . . . . . . . . . . . . 70

7.1 New Winglet Geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96

B.1 Locations of pressure distributions . . . . . . . . . . . . . . . . . . . . . . 112

C.1 New winglet pressure distributions for α = 3◦ , β = 0◦ . . . . . . . . . . . 114

D.1 Tornado L/D comparison . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116

3.1 Merlin III data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

4.1 Mesh data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46

5.1 Parameters for mesh size verification . . . . . . . . . . . . . . . . . . . . . 66

7.1 New Winglet data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95

A.1 Global mesh settings for all cases . . . . . . . . . . . . . . . . . . . . . . . 110

a = Lift curve slope CD = Coefficienct of drag: D/(qS)

The purpose of this thesis is the aerodynamic characterisation of a winglet design

In Chapter 7, a new and improved design is introduced to resolve some shortcom-

2.2 Aircraft Drag

This leads to the standard drag equation:

CD = CD,0 + KCL2 (2.1)

Equation (2.1) can be written as

for a twisted wing contains an additional linear term in CL ,

CD = CD,0 + K1 CL + K2 CL2 (2.4)

Further, several concepts are studied theoretically or experimentally without being

Figure 2.1: Velocities on wingtip projected in the XY-plane

in Y-direction by the presence of the winglet. The spanwise velocity components

• To generate this lift distribution as efficiently as possible.

• To minimize the effects if interference at the junction. For transsonic con-

• To avoid separated flow at all flight conditions. This objective is questionable

• to minimise the structural impact of the winglet.

The induced drag of a wing-winglet combination decreases nearly linearly with

A winglet should have the largest possible aspect ratio to be as effective in

A limited amount of research literature is available for winglets on general aviation

2.3 Computational Fluid Dynamics

“CFD is the art of replacing the partial differentials or integrals in the

An overview over the applications of CFD for aerodynamic predictions is given in

The solution error is defined as [23]

Thus the derivative δu/δx can be represented as

These spatial discretisation errors contribute to a phenomenon called numerical

In the asymptotic convergence region ‘the principle concept of grid convergence

An indicator if the solutions are in the asymptotic range is given as

and should be close to constant.

A method for managing CFD errors based on quality management principles is

2.4 Drag Prediction with CFD

“A very satisfactory explanation of the physical processes in the bound-

Figure 2.2: Structure of a boundary layer. Source: [27], modified

(a) Near leading edge (b) Mid-wing

(c) Near trailing edge (d) Separation at α = 10◦

Any flow above a Reynolds number of approximately 500,000 is turbulent. Thus,

η = (ν 3 /)1/4 τ = (ν/)1/2 v = (ν)1/4 (2.11)