SILVA, R. G. A., A Study On Correction Methods For Aeroelastic Analysis in Transonic Flow.

Thesis presented to the Faculty of the Division of Graduate Studies of the
Technological Institute of Aeronautics in partial fulfillment of the
requirements for the Degree of Doctor in Science in the Course of
Aeronautical and Mechanical Engineering, Area of Aerodynamics,
Propulsion and Energy.
Roberto Gil Annes da Silva
A STUDY ON CORRECTION METHODS FOR
AEROELASTIC ANALYSIS IN TRANSONIC FLOW
...................................................................
Prof. Dr. João Luiz Filgueiras de Azevedo
Advisor
...........................................................
Dr. Olympio Achilles de Faria Mello
Advisor
......................................................
Prof. Dr. Homero Santiago Maciel
Head of the Division of Graduate Studies
Campo Montenegro
São José dos Campos, SP – Brasil
2004
Dados Internacionais de Catalogação-na-Publicação (CIP)
Divisão Biblioteca Central do ITA/CTA
Silva, Roberto Gil Annes da Silva
A Study on Correction Methods for Aeroelastic Analysis in Transonic Flow /
Roberto Gil Annes da Silva.
São José dos Campos, 2004.
233f.
Tese de doutorado– Curso de Engenharia Aeronáutica e Mecânica, e área de Aerodinâmica, Propulsão

e Energia – Instituto Tecnológico de Aeronáutica, 2004. Orientadores: Prof. Dr. João Luiz Filgueiras
de Azevedo e Prof. Dr. Olympio Achilles de Faria Mello.
1. Aeroelasticidade . 2. Aerodinâmica não estacionária. 3. Escoamento transônico. 4. Correção. 5.

Dinâmica dos fluidos computacional. 6. Sustentação aerodinâmica. 7 Mecânica dos fluidos. 8 Física. I.
Centro Técnico Aeroespacial. Instituto Tecnológico de Aeronáutica. Divisão de Engenharia Aeronáutica.
II.Título
REFERÊNCIA BIBLIOGRÁFICA
SILVA, Roberto Gil Annes da. A Study on Correction Methods for Aeroelastic Analysis
in Transonic Flow. 2004. 233f. Tese de doutorado – Instituto Tecnológico de Aeronáutica,
São José dos Campos.
CESSÃO DE DIREITOS
NOME DO AUTOR : Roberto Gil Annes da Silva

TÍTULO DO TRABALHO : A Study on Correction Methods for Aeroelastic Analysis in
Transonic Flow
TIPO DO TRABALHO/ANO : Tese / 2004
É concedida ao Instituto Tecnológico de Aeronáutica permissão para reproduzir cópias desta

tese e para emprestar ou vender cópias somente para propósitos acadêmicos e científicos. O
autor reserva outros direitos de publicação e nenhuma parte desta tese pode ser reproduzida
sem a autorização do autor.
______________________
Rua Salto Grande, no 264, apto. 102, Perdizes, São Paulo,
SP – CEP: 01257-020, Brasil.
A STUDY ON CORRECTION METHODS FOR
AEROELASTIC ANALYSIS IN TRANSONIC FLOW
Thesis Committee Composition:
Prof. Dr. Maher Nasr Bismark-Nasr .........................................................Chairperson - ITA
Prof. Dr. João Luiz Filgueiras de Azevedo .....................................................Advisor - ITA
Prof. Dr. Olympio Achilles de Faria Mello .....................................................Advisor - IAE
Prof. Dr. Donizeti de Andrade .......................................................................Member - ITA
Prof. Dr. Armando Miguel Awruch ..........................................................Member - UFRGS
Prof. Dr. Flávio Donizeti Marques ................................................................Member - USP
ITA
DEDICATION
To my wife Michelle and my children Julia and Pedro.

ACKNOWLEDGEMENTS
I would like to express my sincere gratitude to my advisors, Prof. Dr. João Luiz
Filgueiras de Azevedo and Prof. Dr. Olympio Achilles de Faria Mello, for their guidance and
advice during the course of this work. Their encouragement, thoughtfulness, and supervision
are deeply acknowledge
I would like to thank Prof. Dr. Danny D. Liu of Arizona State University and Mr. P.C.
Chen of ZONA Technology for their valuable guidance, constructive ideas, and generosity in
sharing their experiences. I have been very fortunate to receive invaluable lessons especially
during my stay in Arizona, for their valuable support and assistance. ZONA technology has
been very kind to let me use their facilities during the final stage of the present work
I wish to express my sincere thanks to Dr. Lei Tang, Dr. Luciano Amaury dos Santos,
and Dr. Maurício Guimarães Silva in helping me processing some of the results presented in
this work.
I would like to thank Dr. Dario Baldelli, Dr. Reggie Chang, Dr. Xiaowei Gao, Mr.
Darius Sarhaddi, Ms. Jennifer Sherr for their personal attention and help in providing me the
best conditions during my stay in Arizona.
I would like to express my gratitude to my colleagues Major Antonio Carlos Ponce
Alonso, and Major Guilherme Augusto Vargas Cesar, who provide me the conditions for the
development this work
And finally I would to express my profound gratitude to my parents, Arnaldo and
Irene and to my wife Michelle, who guided me with encouragement.

ABSTRACT
The work presents a study of correction techniques to compute unsteady transonic
pressure distributions and aeroelastic stability in this flow regime. The methodologies herein
investigated are based on corrections of pressure distributions by the weighting of the lifting
surface self-induced downwash, resulting from aeroelastic structural displacements or
prescribed motions. A number approaches were investigated.
An investigation into the linear/nonlinear behavior of unsteady transonic flows was
also conducted. It was concluded from such investigation that unsteady transonic flows
present a linear behavior with respect to small aeroelastic structural displacements around a
steady nonlinear mean flow. Such behavior is the basis for further development of downwash
correction methods.
The correction of pressure distributions through the weighting of the lifting surface
self-induced downwash is also known as downwash weighting method. This method has been
enhanced leading to a new downwash correction technique. The procedure may be divided in
two steps, where the first step is a nonlinear steady mean flow correction, with nonlinear
pressure differences considered as reference conditions to correct the self induced downwash.
The second step is the correction of the unsteady component of the downwash, where the
corresponding reference unsteady pressure differences are predicted by a linear aerodynamic
model, based on the potential flow equations.

This extended downwash correction method led to a rational formulation named as
“successive kernel expansion method” (SKEM). The unsteady pressures and aeroelastic
stability boundaries computations using such method led to good agreement with
experimental measurements. This procedure is a rapid form to compute the transonic flutter
speed boundaries, compared to computational aeroelasticity and experimental techniques.

RESUMO
O trabalho apresenta um estudo sobre técnicas de correção para calcular distribuições
de pressão não lineares e não estacionárias, assim como a estabilidade aeroelástica em
escoamentos transônicos. As metodologias investigadas neste trabalho são baseadas na
correção de distribuições de pressão através da ponderação das velocidades normais (ou
downwash) induzidas pelos deslocamentos estruturais de natureza aeroelástica, ou devido a
movimentos prescritos. Diferentes formas de aproximação foram investigadas.
Investigou-se o comportamento linear/não-linear de escoamentos transônicos não
estacionários. Conclui-se de tal estudo que os escoamentos transônicos não estacionários
apresentam um comportamento linear, com respeito às pequenas deformações estruturais de
natureza aeroelástica, ao redor de um escoamento estacionário não linear. Tais observações
relacionadas à este comportamento serão base para os desenvolvimentos adicionais dos
métodos de correção.
A correção das pressões através da ponderação do vetor de velocidades normais
induzidas é também conhecido como um método de ponderação do downwash. Este método
foi aperfeiçoado, levando a um novo procedimento de correção de downwash. Tal
procedimento pode ser divido em dois passos, sendo o primeiro baseado em uma correção do
escoamento estacionário não linear médio, onde diferenças de pressão não lineares serão
consideradas condições de referência para corrigir o downwash. O segundo passo consiste na
correção da parcela não estacionária do downwash, onde as pressões de referência são não
estacionárias e calculadas a partir de um modelo aerodinâmico linear baseado na equação do
escoamento potencial.
Esta extensão do método de correção do downwash resulta em uma formulação
racional, denominada como “método de expansão sucessiva do kernel” (successive kernel
expansion method - SKEM). Tal método apresentou bons resultados tanto com relação ao
cálculo de pressões não estacionárias quanto à estabilidade aeroelástica, em concordância com
resultados experimentais. Este procedimento é uma forma rápida de calcular as fronteiras de
estabilidade aeroelástica no regime transônico, quando comparado com métodos de
aeroelasticidade computacional ou técnicas experimentais.

CONTENTS
List of Figures …………………………………………………………..……………… IV
List of Tables ….…………………………………...…………………………….….… XII
List of .Symbols ………………………………………………………………………. XIII
Chapter 1 – Introduction …………….………………………………..……………...…... 1
1.1 – Background ...........…….………………………………………………………. 1
1.1.1 Aeroelastic Stability Behavior in Transonic Flow................................ 1
1.1.2 Approximate Solutions.......................................................................... 5
1.1.3 Force Matching Methods....................................................................... 7
1.1.4 Pressure Matching Methods................................................................ 10
1.1.5 Dau-Garner Type Methods................................................................... 13
1.1.6 Modal Aerodynamic Influence Coefficient Matrix (MAIC) Methods .15
1.2 – Analytical Review of Correction Methods.......................................................... 19
1.2.1 Application of Approximate Solutions for Transonic Flow
Computation......................................................................................... 19
1.2.2 Proposal of a Correction Technique..................................................... 24
1.3 – Objective of the Present Work.....………………………………………...…… 26
1.4 – Outline of the Thesis...............................................…………………………… 27
Chapter 2 – Unsteady Aerodynamic Modeling and Aeroelastic Analysis.……………... 29
2.1 – Integral Solution of the Linearized Potential Flow Equation……..…………... 29
2.2 – Discrete Element Kernel Function Methods …………………………….…… 35
2.3 – The Aerodynamic Influence Coefficients Approach.......................................... 37
2.4 – Aeroelastic Model............................................................................................... 40
2.4.1 – Aeroelastic Equations of Motion......................................................... 40
2.4.2 – Modal Approach ................................................................................. 43
I
2.4.3 – Simple Harmonic Motion Approach.................................................... 44
2.4.4 – Interconnection Between Aerodynamics and Structures..................... 45
2.5 – Flutter Solution Techniques................................................................................ 47
2.5.1 – Flutter Computation Based on the k-method........................................ 48
2.5.2 – The p-k method..................................................................................... 49
2.5.3 – The g-method........................................................................................ 51
2.5.4 – p-method for the Solution of a State Space Aeroelastic System.......... 53
Chapter 3 – Unsteady Transonic Flow Behavior……………………..………………..... 57
3.1 – Linear/Nonlinear Behavior Investigation...………………..…………….…….. 57
3.2 – The Navier Stokes Model....................…………..…………………….…….... 58
3.2.1 – Governing Equations........................................................................... 58
3.2.2 – Numerical Method............................................................................... 60
3.2.3 – Turbulence Model................................................................................ 66
3.2.4 – Boundary Conditions........................................................................... 67
3.3 – Transonic Flow Investigations of a Low Aspect Ratio Wing............................. 69
3.3.1 – Preliminary Considerations..................................................................69
3.3.2 – Validation of the Computational Procedure for Steady Flow
Computations....................................................................................... 70
3.3.3 - Validation of the Computational Procedure for Unsteady Flow
Computations....................................................................................... 72
3.3.4 – Investigation Methodology.................................................................. 76
3.3.5 – Analysis of the Linear/Nonlinear Behavior......................................... 77
Chapter 4 – Correction Methods for Approximate Nonlinear Aerodynamic Modeling.... 85
4.1 – Developments on the Pressure Matching Approach................…………….….. 85
II
4.1.1 - Downwash Weighting Method Based on Steady Reference
Conditions............................................................................................. 85
4.1.2 - Downwash Weighting Method Based on Unsteady Reference
Conditions............................................................................................. 89
4.2 – Extension of the Downwash Weighting Method Based on the Kernel Expansion
Hypothesis.……………………………............................................................. 92
4.2.1 – Motivation............................................................................................ 92
4.2.2 – A Modified Downwash Correction Method........................................ 95
Chapter 5 – Results and Discussion .......………...……………...……………….…….. 101
5.1 – Evaluation of Giesing´s Method for Transonic Flutter Computation................ 101
5.2 – Downwash Weighting Methods........................................................................ 107
5.2.1 – Investigation of the AGARD Wing Unsteady Pressure Distribution .108
5.2.2 – AGARD Wing Aeroelastic Stability Analysis................................... 134
5.3 – Evaluation Of the Successive Kernel Expansion Method................................. 143
5.3.1 – Preliminary Considerations................................................................ 143
5.3.2 – Validation of Unsteady Pressures Computations............................... 144
5.3.3 – Validation of Flutter Boundary Computations ................................. 165
Chapter 6 – Conclusions and Final Remarks……...……………...……………….…… 176
6.1 – Conclusions....................................................................................................... 176
6.2 –Contributions...................................................................................................... 181
6.3 – Recommendations for Future Work.................................................................. 182
Chapter 7 – References ….……………………………………….…………………...... 184
Appendix A – Giesing´s Force Matching Method.........................……….……………. 199
Appendix B – The Dau-Garner Method .…………...................……………..……..….. 202
III
LIST OF FIGURES
Figure 1.1 The transonic dip phenomenon. (taken from Tijdeman, 1977)................................ 3
Figure 3.1- Finite differences computational mesh surrounding the F-5-NLR wing and its
profile...................................................................................................................................... 71
Figure 3.2: Steady pressure distribution over the F-5 NLR wing at M∞ = 0.95..................... 71
Figure 3.3: Real and imaginary parts of the pressure coefficients over the F-5 NLR wing
(M=0.95, 35.2% spanwise station, frequency = 40 Hz)......................................................... 74
Figure 3.6: Differential pressure coefficient time history over F-5 wing (M=0.95, ∆α=1.0º;
k=0.1)..................................................................................................................................... 78
Figure 3.7: Frequency contents (amplitudes and phases) of differential pressure coefficient
over F-5 wing (M=0.95, ∆α=1.0º; k=0.1).............................................................................. 78
Figure 3.8: Effect of dynamic angle of attack on unsteady aerodynamic coefficients and shock
motion (amplitudes and phases; station 35.5%; k=0.2)......................................................... 79
Figure 3.9: Effect of dynamic angle of attack on unsteady aerodynamic coefficients and shock
motion (amplitudes and phases; station 72.1%; k=0.2)......................................................... 81
Figure 3.10: Effect of dynamic angle of attack on unsteady aerodynamic coefficients and
shock motion (amplitudes and phases; station 97.7%; k=0.2).............................................. 82
IV
Figure 3.11: Boundaries for linear behavior based on the CM criteria.................................... 83
Figure 3.12: Boundaries for linear behavior based on the shock displacement criteria.......... 83
Figure 5.1: Sketch of the AGARD 445.6 wing and the corresponding discrete element
aerodynamic model................................................................................................................ 102
Figure 5.2 : Comparison between the unsteady pressures obtained from the Giesing´s method,
ZONA 6 and CFL3D - AGARD 445.6 wing, weakened model #3, M∞= 0.96..................... 104
Figure 5.3 : Flutter boundaries (dynamic pressure and flutter frequency) for the AGARD wing
445.6, weakened model #3..................................................................................................... 105
Figure 5.4 : Sketch of the PAPA wing and the corresponding discrete element aerodynamic
model...................................................................................................................................... 105
Figure 5.5 : Flutter boundaries (dynamic pressure and flutter frequency) for the PAPA wing,
at a reference angle of attack of α = 1.0o............................................................................... 106
at a reference angle of attack of α =-2.0o.............................................................................. 106
Figure 5.7: DLM paneling of the AGARD wing................................................................... 108
Figure 5.8 : AGARD wing CFD aerodynamic mesh............................................................. 109
Figure 5.9 : AGARD wing 445.6 NACA 65A004 profile.................................................... 110
Figure 5.10 : Unsteady pressure distributions for the wing station at 30.8% of the span,
M∞=0.678, ∆α = 0.25o.......................................................................................................... 111
Figure 5.11: Unsteady pressure distributions for the wing station at 30.8% of the span,
M∞=0.678, ∆α = 0.5o............................................................................................................ 112
V
M∞=0.678, ∆α = 1.0o............................................................................................................ 112
M∞=0.678, ∆α = 1.5o. .......................................................................................................... 113
M∞=0.678, ∆α = 2.0o............................................................................................................ 113
M∞=0.678, ∆α = 2.5o............................................................................................................. 114
M∞=0.678, ∆α = 3.0o............................................................................................................. 114
M∞=0.901, ∆α = 0.25o........................................................................................................... 115
M∞=0.901, ∆α = 0.5o............................................................................................................ 115
M∞=0.901, ∆α = 1.0o............................................................................................................ 116
station
Figure 5.20: Unsteady pressure distributions for the wing at 30.8% of the span, M∞=
0.901, ∆α = 1.5o.................................................................................................................... 116
Figure 5.21: Unsteady pressure distributions for the wing station at 30.8% of the span, M∞=
0.901, ∆α = 2.0o.................................................................................................................... 117
M∞=0.901, ∆α = 2.5o............................................................................................................ 117
VI
M∞=0.901, ∆α = 3.0o........................................................................................................... 118
0.96, ∆α = 0.25o................................................................................................................... 118
M∞=0.96, ∆α = 0.5o ............................................................................................................. 119
0.96, ∆α = 0.25o..................................................................................................................... 119
0.96, ∆α = 1.5o....................................................................................................................... 120
0.96, ∆α = 2.0o....................................................................................................................... 120
0.96, ∆α = 2.5o....................................................................................................................... 121
0.96, ∆α = 3.0o....................................................................................................................... 121
Figure 5.31 : Contour plot of instantaneous pressures, at spanwise station 30,8%, M∞=0.678,
α = 2.0o.....…………….......................................................................................................... 122
Figure 5.32 : Instantaneous pressure distributions, at spanwise station 30,8%, M∞=0.678,
∆α = 2.0o..........…….............................................................................................................. 123
Figure 5.33: Contour plot of instantaneous pressures, at spanwise station 30,8%, M∞=0.678,
α = 3.0o……………............................................................................................................... 123
VII
Figure 5.34: Instantaneous pressure distributions, at spanwise station 30,8%, M∞=0.678,
∆α = 3.0o..........…….............................................................................................................. 124
α = 2.0o……………............................................................................................................... 125
∆α = 2.0o................................................................................................................................ 126
α = 3.0o...................................................……………............................................................ 126
∆α = 3.0o...................................................................................................................……..... 127
α = 2.0o.............................................................................................................…………….. 128
∆α = 2.0o............................................................................................................................… 128
α = 3.0o...............................................................................................……………................ 129
∆α = 3.0o................................................................................................................................ 129
Figure 5.43: Instantaneous velocity vector fields over the leading edge of the AGARD wing
445.6....................................................................................................................................... 131
Figure 5.44. Non-dimensional phase angles of the pressure difference distributions for
spanwise station 30.8%, M∞=0.96. ....................................................................................... 132
VIII
Figure 5.45. Amplitudes of the pressure difference distributions for spanwise station 30.8 %,
Mach =0.96. .......................................................................................................................... 133
Figure 5.46: AGARD wing 445.6 results - comparison of the flutter computation results
between steady and unsteady downwash weighting methods............................................... 136
between the unsteady downwash weighting methods and other methods............................. 137
Figure 5.48: Comparison between the flutter speed plots as a function of the amplitude of the
motion.................................................................................................................................... 140
Figure 5.49: Flutter speed behavior as a function of the amplitude of the motion................ 141
Figure 5.50: Flutter frequency behavior as a function of the amplitude of the motion......... 142
Figure 5.51 : Sketch of the F-5 wing and the corresponding discrete element aerodynamic
model mesh. .......................................................................................................................... 145
Figure 5.52 : Steady pressure distribution for the F-5 wing at M = 0.9 and α = 0°.............. 146
Figure 5.53 : Pressure coefficient ratio for the F-5 wing at M = 0.9 and ∆α = 0.5°............. 147
Figure 5.54 : Real part of the pressure distribution for the F-5 wing at M = 0.9 and reduced
frequency k = 0.275............................................................................................................... 148
Figure 5.55 : Imaginary part of the pressure distribution for the F-5 wing at M = 0.9 and
reduced frequency k = 0.275.................................................................................................. 149
Figure 5.56 : Steady pressure distribution for the F-5 wing at M∞ = 0.948 and α = 0°......... 151
Figure 5.57 : Pressure coefficient ratio for the F-5 wing at M∞ = 0.948 and ∆α = 0.5°........ 152
Figure 5.58 : Real part of the pressure distribution for the F-5 wing at M∞ = 0.948 and reduced
frequency k = 0.264............................................................................................................... 153
IX
Figure 5.59 : Imaginary part of the pressure distribution for the F-5 wing at M∞ = 0.948 and
Figure 5.60 : Sketch of the LANN wing and the corresponding discrete element aerodynamic
model..................................................................................................................................... 155
Figure 5.61 : Steady pressure distribution for the LANN wing at M∞ =0.822 and α = 0.6°. 156
Figure 5.62 : Pressure coefficient ratio for the LANN wing at M∞ =0.822 and ∆α = 0.15°. 157
Figure 5.63 : Real part of the pressure distribution for the LANN wing at M∞ =0.822 and
Figure 5.64 : Imaginary part of the pressure distribution for the LANN wing at M∞ =0.822 and
Figure 5.65 : Sketch of the Lessing wing and the corresponding discrete element aerodynamic
model...................................................................................................................................... 160
Figure 5.66 : Steady pressure distribution for the Lessing wing at M∞ =0.9 and α = 0°....... 161
Figure 5.67 : Pressure coefficient ratio for the Lessing wing at M = 0.9 and ∆α = 0.2°....... 162
Figure 5.68 : Pressure distribution (amplitudes) for the Lessing wing at M∞ =0.9 and reduced
frequency k = 0.13................................................................................................................. 163
Figure 5.69 : Pressure distribution (phases) for the Lessing wing at M∞ =0.9 and reduced
frequency k = 0.13................................................................................................................. 164
Figure 5.70 - Flutter boundaries (dynamic pressure and flutter frequency) for the AGARD
wing 445.6, weakened model #3........................................................................................... 166
Figure 5.71 : Dynamic pressures and frequencies in the flutter boundaries of the PAPA wing,
at angle of attack, α = 1.0o.................................................................................................... 170
X
Figure 5.72 : Dynamic pressures and frequencies in the flutter boundaries of the PAPA wing,
at angle of attack, α = -2.0o................................................................................................... 171
Figure 5.73 : Sketch of the YXX wing and the corresponding discrete element aerodynamic
model...................................................................................................................................... 172
Figure 5.74 : Flutter dynamic pressure for the YXX wing, at a reference angle of attack
α=-1.0o................................................................................................................................... 174
Figure 5.75 : Flutter frequency for the YXX wing, at a reference angle of attack α=-1.0o.. 174
XI
LIST OF TABLES
Table 5.1: Flow conditions for AGARD wing 445.6 (weakened # 3 ) aeroelastic analysis.. 103
Table 5.2: Reduced frequencies for AGARD wing 445.6 aeroelastic analysis..................... 110
Table 5.3: Flutter speeds and frequencies for AGARD wing 445.6...................................... 135
Table 5.4: Flutter speeds and frequencies for AGARD wing 445.6...................................... 136
Table 5.5: Flutter speeds and frequencies for AGARD wing 445.6. until ∆α=1.5o............. 140
Table 5.6: Flutter speeds and frequencies for AGARD wing 445.6, until ∆α=3.0o............. 140
XII
LIST OF SYMBOLS
a Speed of the sound
A+ = 26.0 van Driest constant
A = ∂Eˆ ∂Qˆ Flux Jacobean matrix
A (U ∞ ) State space matrix of an aeroelastic system
A1 , A4 , A6 Auxiliary metric terms (Eq. 3.24)
[ AIC ] Aerodynamic influence coefficients matrix
b Reference length equals to the semi-chord
c = 2b Lifting surface streamwise section chord
CL Sectional lift coefficient
CM Sectional moment coefficient about the quarter-chord
cp (in Prandtl number) Specific heat at constants pressure
Cp Pressure coefficient
cr Root chord
d Distance from the wall
[ D] Kernel function matrix
[ D ]n , n = 1, m Constant coefficient matrices, obtained from the asymptotic series
expansion
Dij Element of the kernel function matrix
Dˆ en ,n +1 Explicit fourth order dissipation
e Total energy per unit volume
Eˆ , Fˆ , Gˆ Inviscid flux vectors
XIII
[F ]. Substantial derivative operator
g =γ k Damping factor
[G ] Spline matrix operator
gs . artificial structural damping
h Displacement mode shape vector
[H ] Aerodynamic transfer function matrix
[I ] Identity matrix
i Index corresponding to a receiving point in the discrete element
kernel function aerodynamic model
i Imaginary number ( i = − 1 )
j Index corresponding to a sending point in the discrete element
kernel function aerodynamic model
J Jacobian transformation between the curvilinear and cartesian
systems
[K ] Structural stiffness matrix
 K  = [ Φ ] [ K ][ Φ ] Structural modal stiffness matrix

T
k = ωb U ∞ Reduced frequency based on the free-stream flow speed
k (in Prandtl number) Thermal conductivity
kr = ω c a∞ = k ⋅ ( M ∞ c b ) Reduced frequency based on the speed of the sound
Kϕ Kernel of the integral relation in term of the velocity potential
Kψ Kernel of the integral relation in term of the acceleration potential
L Laplacian operator
XIV
{L} Total aerodynamic load vector
{La } Induced aerodynamic load vector
{Le } External aerodynamic load vector
m Mixing length
[M ] Structural mass matrix
 M  = [ Φ ] [ M ][ Φ ] Structural modal mass matrix

T
M∞ Free-stream Mach number
O ( ∆τ ) Order of error in finite difference
approximation
p Pressure
p (in flutter solution) = γ k + ik , aeroelastic eigenvalue
Pr = µ c p k Prandtl number
PrT = µT c p kT Turbulent Prandtl number
[Q ] Aerodynamic loads influence coefficients matrix
Q̂ vector of unknown flow properties
q dynamic pressure
{q} Generalized coordinates
Qap  Aerodynamic loads influence coefficients matrix approximated by
a Padé polynomial
Q j  Matrix coefficients of the Padé polynomial
{q } lag
Lag states of an aeroelastic system
XV
R = X 2 + β 2 (Y 2 + z 2 ) distances between the sending and receiving points in the discrete
element kernel function model
Rˆ , Sˆ , Tˆ Viscous flux vectors
R5, S5, T5 Auxiliary variables for vicous fluxes, given in, Eq. (3.7)
Re = ρ∞ a∞ c µ∞ Reynolds number based on free-stream speed of sound
Re1 = ρ∞U ∞ cr µ ∞ Reynolds number based on free-stream flow speed
Re( ), Im( ) Real and imaginary parts of a complex number
s = iω . Laplace variable
sb Nondimensional Laplace variable

s=
U∞
S Lifting surface area
[S ] Integration matrix
t time
Tξ , Tζ . Eigenvectors (transformation) matrices
{u } physical displacement vector of the structure
u,v,w cartesian components of velocity
Uc, , Vc , Wc Contravariant components of velocity
{w} Downwash vector
{w} Downwash vector divided by amplitude of the quasi-steady motion
amplitude ∆α .
W Wake surface area
[WT ] Weighting matrix operator
x Cartesian coordinate, aligned with the undisturbed flow speed
xs , ys ,zs Cartesian coordinates in the structural model coordinate system
XVI
y Cartesian coordinate, aligned with the spanwise direction
z Cartesian coordinate, normal to the lifting surface
α Angle of attack
β = 1− M ∞2 Prandtl- Glauert constant
βn Lag parameter
γ Ratio of specific heats
γ Damping decaying rate
δ Difference operator corresponding to standard second order central
difference
δ Difference operator corresponding to half-point central difference
{δα∆C p } Dimensionless pressure coefficient disturbance rate vector
{δ h} Vector of the virtual displacement at the aerodynamic control
points
{δ h } = ∆α
x
Vector of the quasi-steady disturbance in effective angle of attack
{δu } Vector of the virtual displacement at the structural nodes
∆α Amplitude of the dynamic angle of attack
{δ∆C } nl
p
Vector of the quasi-steady pressure coefficient differences
∆ξ , ∇ξ Forward and backward difference operators
∆ϕ Doublet singularity strength
∆τ Computational non-dimensional time step
εE Coefficient of explicit numerical dissipation
εI Coefficient of implicit numerical dissipation
XVII
ζ Coordinate in curvilinear system, aligned with the spanwise
direction
η Coordinate in curvilinear system, normal to the lifting surface,
κ von Kármán constant,
{λ} State space variables vector of an aeroelastic system
Λξ , Λζ contents the eigenvalues of the matrices A and C
µ Molecular viscosity
µT Eddy viscosity
ξ Coordinate in curvilinear system, tangent to the lifting surface
boundary, in the streamwise direction.
ξ ,η Planar position of doublet (discrete kernel function method)
ρ Density
σ Point source singularity strength
τ = a∞t c non-dimensional time
τ max Maximum shear stress in the boundary layer
τ wall Shear stress in the wall
τ xx ,τ xy ,τ xz ,τ yz ,τ yy ,τ zz Shear stresses
φ Perturbation velocity potential
φ1 , φ2 , φ3 Auxiliary variables in numerical dissipation
{φ} Vector mode shape (columns of the of the modal matrix)
[Φ ] Modal transformation matrix, composed by mode shape vectors
ϕ Frequency domain velocity potential
ϕD Unsteady doublet singularity potential
XVIII
ϕS Unsteady source singularity potential
ψ Acceleration potential
ω Oscillation frequency associated to the harmonic motion.
ω mean vorticity
Subscripts
A aerodynamic
ap approximate
e external
i,j,k indexes indicatign a point in the computational fluid dynamic grid
i imaginary
I, and F refer to an initial position and a final position respectively
LE leading edge
nlag number of lag states
R real
s structural
T turbulent
t Partial derivative with respect to time
TE Trailing edge
x Partial derivative with respect to x
y Partial derivative with respect to y
z Partial derivative with respect to z
∞ condition at free-stream
XIX
Superscripts
aero Aerodynamic
l lower
n, n + 1 time levels
nl nonlinear
str structural
u upper
XX
Chapter 1 – Introduction 1
1 - INTRODUCTION
1.1 Background
1.1.1 Aeroelastic Stability Behavior in Transonic Flow
The most common way in performing aeroelastic analyses in the aeronautical industry
is by the use of the structural-dynamic/unsteady aerodynamic commercial codes based on
linear unsteady aerodynamic modeling techniques combined with a structural dynamics
solver. Usually, linear aerodynamic modeling techniques are based on discrete element
solutions of the linear equations of the fluid flow. Discrete element methods, also named as
panel methods, are defined as numerical procedures which require the body to be discretized
into many small elements, also named as panels. Each discrete element contains a control
point where the boundary condition is imposed.
The fluid flow is represented by the linearized potential flow equation (Landahl,
1961), which has an integral solution in terms of the source and doublet singularity
distribution over the body surface (Chen et al. 1993). The resulting velocity potentials over
each of the elements are related to control point displacements through the boundary
conditions which requires the flow to be tangential to the body at all times. This condition is
represented by the substantial derivative of the body displacements leading to induced
velocities, normal to the body, when the tangential flow condition is satisfied.
These velocities are named as downwash, which depends on the pattern of the
displacements of the body undergoing an unsteady motion. Once the velocity potentials are
computed by the integral solution, the derivatives of those potentials with respect to the
direction normal to the body surface, lead to induced downwash, which are related to the
downwash due to the body motion by the boundary condition. The remaining unknown is the
strength of the elementary singularity distribution along each of the elements.
However, the solution of the same linearized potential flow equation, in terms of the
acceleration potential instead of the velocity potential would greatly simplifies the
aerodynamic modeling. This is so because the acceleration potential, is the linearized
pressure, thus the body wake domain would not be discretized because there is no pressure
jump across the wake. Therefore, the solution of the aerodynamic model in terms of
elementary integral solution would be reduced to a pressure to downwash relationship, where
the assembly of these elementary integral solution gives a matrix whose coefficients represent
the aerodynamic influence of the discrete elements pulsating pressures to control point
displacements.
The resulting system of equations can be expressed as a linear operator applied to a
downwash vector resulted from the substantial derivatives of the body motion, in other words,
from the boundary conditions. The matrix which defines this linear operator is named as the
aerodynamic influence coefficient matrix (AIC) and its dimension is the same as the number
of panels. The body undergoing unsteady motions may be represented by the substantial
derivatives of control points displacements accordingly a mode shape pattern. Therefore, the
resulting downwash due to the body modal motion will be defined hereafter as downwash
mode shape. Summarizing, the mode shape displacements are related to known boundary
conditions, leading to a downwash vector. Furthermore, the AIC matrix is also known and is
obtained from the assembly of the integral solutions of the linearized potential flow equation.
Thus, the solution of the unsteady flow may be represented by the resulting pressure
distribution computed from the application of the AIC matrix operator to the downwash
vector.
Examples of discrete element methods are the doublet lattice method – DLM (Albano
and Rodden, 1969), the unsteady subsonic vortex lattice method – UVLM (Soviero and
Bortolus, 1992), the constant pressure method –CPM (Appa, 1987), the harmonic gradient
method –HGM Chen and Liu, 1985), and others.
The application of these aforementioned methods are limited to purely subsonic or
supersonic flows, since the governing equations over which the method were developed are
based on a linearized unsteady potential flow hypothesis. However, the aeroelastic behavior
of an aircraft is typically critical in the transonic flight regime, where nonlinear phenomena
related to embedded moving shock waves and viscosity, play an important role in aeroelastic
stability. As discussed by Ashley (1980), the shock wave movement and strength profoundly
affects the flexure-torsion flutter mechanism. One consequence of this behavior is the so
called “transonic dip”, which is graphically represented in Figure (1.1).
1
linear theory
measurements
Dimensionless Flutter Speed
0.95
0.9
0.85
0.8
0.75
0.7 0.8 0.9 1 1.1 1.2
Mach Number
Figure 1.1 Flutter speed boundary as a function of the freestream Mach number, exposing the
transonic dip phenomenon.

This phenomenon is characterized by a decrease of the slope of the flutter speed plot
as a function of the Mach number, when compared to the same plot obtained from a linear
aeroelastic analysis. Therefore, it is necessary to pay special attention to the flutter
phenomenon under these circumstances. This is especially significant as most modern aircraft
fly under transonic flow conditions.
One of the feasible alternatives for analyzing the aeroelastic stability in nonlinear flow
conditions is by the use of time accurate computational fluid dynamic (CFD) solutions of the
nonlinear fluid equations coupled with structural dynamic representation of the vehicle.
Another approach is obtained by wind tunnel testing of aeroelastic models, under tronsonic
flow conditions. However, wind tunnel testing for aeroelastic investigation regarding flutter
boundary computation is not usual because this class of experiments involves expensive
models and high operational costs. In most cases of experimental aeroelastic stability
investigation, the objective is to obtain accurate measurement data to validate computational
procedures applied to the solution of nonlinear equations. Examples of experiments on
unsteady aerodynamics regarding pressures measurements are reported by Lessing et al..
(1960), Tijdeman et al.. (1978), and Malone et al. (1983). Notable examples of experimental
flutter investigations were presented by Isogai (1983), Yates (1988), Farmer et al.. (1988),
and Hong et al. (2003). The other way to evaluate the transonic aeroelastic behavior, is from
flutter flight testing, which is the most hazardous and expensive option in terms of operational
costs. This approach may be used either for experimental flutter boundaries identification or
to verify the aeroelastic subcritical aerodynamic damping at specific flight envelope points to
validate aeroelastic numerical models.

1.1.2 - Approximate Solutions
There have been several attempts to solve the transonic aeroelastic problem using
combined procedures which relate linear models to measured data for the correction of
unsteady linear aerodynamic models. A good review on correction techniques was presented
by Palacios et al. (2001), describing the most employed methods concerning transonic flutter
prediction via combined procedures. Such approaches are named here as mixed procedures.
The purpose of such procedures is to correct the linear aerodynamic models to take into
account nonlinear effects, unpredictable by the linearized potential-based equations of the
fluid flow.
The methodologies for the solution of the transonic aeroelastic problem based on
mixed procedures are referred also as semi-empirical corrections. These corrections can be
performed by the multiplication, addition or the whole replacement of the AIC matrix. This
approach is an adequate tool for engineering applications, because the methodology employed
is less expensive than direct use of CFD techniques. The correction techniques, which have
been applied to unsteady loading calculation for static or dynamic stability analysis, are here
classified in four major groups:
– Force matching methods - (Yates, 1966; Giesing et al., 1976; Zwaan, 1985; Pitt
and Goodman, 1987; Brink-Spalink and Bruns, 2001 );
– Pressure matching methods - (Rodden and Revell, 1962; Bergh, 1964; Bergh and
Zwaan, 1966; Luber and Schmid, 1982; Baker, 1997; Baker et al., 2000; Jadic et
al., 1999; Jadic et al., 2000);
– Dau-Garner type procedures – (Garner, 1977; Dau, 1992; SenGupta, 1996;
Schulze and Vogel, 1989; Yonemoto, 1985);

– Modal aerodynamic influence coefficients (MAIC) matrix replacement – (Suciu et
al., 1990; Suciu, 2000; Suciu, 2003; Liu et al., 1988; Chen et al., 1996 ).
The main idea of the first procedure is to match reference nonlinear forces and
moments, which may be obtained from experiments or accurate CFD solutions of the
nonlinear flow governing equations. In this case, nonlinearities such as pressure jumps due to
shock waves and viscous effects are embedded in such reference quantities. The second form
to proceed with the correction is the matching of the pressures taken as reference conditions.
The same nonlinearities are present in the reference conditions, in this case pressure
distributions. The Dau-Garner class of methods are a procedures where the unsteadiness of
the resulting nonlinear corrected pressures are computed based on steady nonlinear
information with the aid of semi-empirical relations. The procedure which named this class of
methods will be further detailed in sub-section 1.1.5, and its formulation is presented in
Appendix B.
Finally, the MAIC procedures consist in generating a modal aerodynamic influence
coefficient matrix referred to measured or computed nonlinear pressures or loading due to
given modal displacements of the lifting surface. This new matrix is then substituted in the
aereolastic equations of motion, where the generalized unsteady aerodynamic forces are
related to the associated modal displacements of the lifting surface, or downwash mode
shapes. The aeroelastic analysis is performed taking into account transonic flow effects
embedded in such new modal aerodynamic influence coefficient matrix. In the next sub-
sections, brief descriptions of the aforementioned methods will be presented, so that it will be
clear how they compare to each other.

1.1.3 - Force Matching Methods
One of the first approaches for developing correction techniques for matching
experimental forces was introduced by Yates (1966). His method is based on a modification
of the strip theory, which is based on Theodorsen’s (1934) typical section unsteady loading
solution. The objective is to take into account nonlinear effects, due to both viscous and
transonic phenomena. This is performed by the replacement of the lifting and moment
coefficient derivatives, obtained from Theodorsen’s (1934) method by the same quantities
determined from measurements. This method is named as the modified strip theory (Yates,
1966) and the published results were in good agreement with the experiments for the case of a
wing with high aspect ratio.
Giesing et al.(1976) developed a method to correct the subsonic discrete lifting surface
theory based on a reference experimental or nonlinear computed loading. Their method
included a procedure to obtain a pre- or post-multiplicative operator to correct the AIC matrix
by the use of weighting factors. The purpose of such operators is to match a single reference
steady spanwise loading of the lifting surface.
Each spanwise station corresponds to a set of panels aligned in the streamwise
direction. However, there are less known quantities, i.e., section loads, than the necessary
weighting factors since the number of the spanwise stations is smaller than the number of
panels. In order to circumvent this problem, the weighting factors are obtained using the least
squares method. With such method, the pressures associated to the reference nonlinear
loading are not restored after the multiplication of the AIC matrix by the weighting factors.
However, the corrected loading condition will be identical to the reference loading condition.
The motivation for the development of Giesing et al.procedure (1976) is the
requirement of a safe prediction of the control surface flutter, because the employment of
potential theory based methods usually underestimates this type of flutter. The reason for this
underestimation is related to absence of representing nonlinear effects, such as, the boundary
layer thickening. This correction factor technique was successfully applied for this class of
problems in subsonic cases. In the transonic flow situation, there are important effects yet to
be considered, such as the control surface buzz and the transonic flutter on the control surface
itself. However, the authors of the cited reference did not present results regarding those
situations.
Zwaan (1985) presented an evaluation of methods to calculate unsteady transonic
loads for aircraft design and flutter analysis. He reviewed a set of correction methods based
on the matching of experimental or computed spanwise lift and moment distributions. All of
them are based on matching those spanwise loading quantities. The differences between the
evaluated methods are related to the correction of the three-dimensional effects, the use of
two-dimensional and three-dimensional reference data and the origin of the nonlinear data,
i.e., either from experiments or from nonlinear computations. Most of the reference nonlinear
loadings were based on wind tunnel measurements. However, Zwann also employed
nonlinear loading resulting from steady or unsteady finite difference solutions of the two-
dimensional transonic small disturbance equation. Zwaan concluded from the evaluation of
such procedures that they require further improvement. The aeroelastic stability behavior, in
some cases, did not agree with the test results in a quantitative sense. However, qualitatively,
all the methods predicted the flutter speed dip. All of them were clearly dependent on the
reference flowfield conditions. These conditions are effects such as reference angle of attack,
wind tunnel test section interference, and whether the computed reference nonlinear loading is
of three-dimensional or two-dimensional nature.
Pitt and Goodman (1987) presented a method for correcting a linear stripwise AIC
matrix, which relates the strip displacements to their loads. Their procedure is based on a
post-multiplication of the stripwise AIC matrix by weighting factors. The weighting factors
are obtained from the inversion of this AIC matrix multiplied by the reference loading
condition. In turn, the reference loading is obtained from the solution of the transonic small
disturbance equations in a steady flow condition. Pitt and Goodman (1987) presented results
in which the transonic dip phenomenon was captured, and the computed flutter speeds were in
agreement with experiment. Some small differences between computed and experimental
results were attributed to viscous effects by the authors.
Brink-Spalink and Bruns (2001) presented an extension of the method of Giesing et
al.(1976). Instead of using a simple least square procedure to match desired loads, as
performed by Giesing et al.(1976), these authors introduced a weighted least squares
procedure, based on a linear quadratic optimization (Brink-Spalink and Bruns, 2001). The
objective of such optimization procedure is to minimize the least squares deviation from the
theoretical linear AIC matrix, while matching multiple loading conditions. Therefore, another
feature added in their procedure, when compared to Giesing’s et al. (1976) method, is the
possibility to include different sets of loads due to distinct displacement conditions. These
displacement conditions can range from a frozen elastic mode shape to a control surface rigid
rotation, for example. Brink-Spalink and Bruns’ (2001) procedure presents better chances to
guarantee the matching of the desired loading condition without severe distortion in the
pressure distributions. This is so, because the uncorrected AIC matrix is taken as reference for
the minimization condition. In other words, the goal of the minimization process is to
compute a weighting matrix which does not distort severely the AIC, preserving interference
effects as much as possible. However, the shock jump conditions for the case of transonic
flow situations cannot be restored, because the AIC matrices computed from a linear
aerodynamic model does not have the capability to predict nonlinearities. The resulting
weighting operator will be a fully populated matrix, instead of the diagonal operator obtained
by Giesing et al.(1976) procedure. The authors did not present conclusive results with regard
to flutter computation. They only presented comparisons between the resulting corrected AIC
matrix and the original one.
1.1.4 - Pressure Matching Methods
Rodden and Revell (1962) presented a procedure based on the pre-multiplication of
the AIC matrix by a diagonal weighting operator. This operator is obtained from the ratio
between the experimental and theoretical steady pressures. In a similar approach, Bergh(1964)
and Bergh and Zwaan (1966) correct the AIC matrix by the use of a diagonal pre-multiplying
weighting operator. However, these authors extended Rodden and Revell´s procedure to
compute the weighting operator based on the ratio between measured and theoretical unsteady
pressures, which are related to a given downwash mode shape. Their main concern is to verify
if is possible to extent the use of a unique weighting matrix to restore unsteady pressures for
different mode shapes and reduced frequencies. Their conclusions are that this extension is
admissible for different mode shapes, however, around a small reduced frequency variation.
Luber and Schmid (1982) presented a method for the aeroelastic analysis in transonic
flow based on a procedure similar to the one presented by Rodden and Revell (1962).
However, Luber and Schmid (1982) employ the ratio between the nonlinear and the linear
quasi-steady pressure slopes as opposed to the ratio between the pressures, as performed by
Rodden and Revell (1962). The resulting ratio of the pressure slopes will compose a diagonal
weighting operator, which role is to match a nonlinear pressures distribution correcting the
linear pressures distribution. Luber and Schmid (1982) perform this correction in two forms,
as a multiplicative weighting operator or as an additive weighting operator. These weighting
operators are used to corrected the AIC matrix as a multiplying diagonal operator or as an
incremental contribution to be added to the AIC matrix, respectively. Luber and Schmid
(1982) presented results regarding unsteady pressures and flutter computations. The computed
unsteady pressures results showed a wrong trend when compared to experimental data. The
flutter computation also did not present good results. The computed flutter speeds were
nonconservative when compared to experimental measurements.
Another recent approach for transonic flutter analysis was presented by Baker (1997),
and it was later extended for static aeroelastic analysis by Baker et al. (2000). Baker and his
co-workers perform an additive correction based on the increment of the linear AIC matrix by
a frequency dependent nonlinear term. This nonlinear term is obtained from a nonlinear
unsteady CFD computation of the lifting surface undergoing a impulse-type displacement.
The pressures are computed from a discrete Fourier transform of the aerodynamic response in
the time domain. The resulting frequency dependent unsteady pressures are used to obtain the
nonlinear term of the AIC matrix, at each of the reduced frequencies of interest. Each of the
frequency dependent nonlinear terms are obtained by the scaling of those pressures by the
chosen reference mode shape, resulting in an additive diagonal operator. This method requires
a refined time marching CFD solution to get accurate frequency response results. Since
Baker´s (1997) procedure employs an unsteady CFD solution, it is a more expensive approach
to correct the linear aerodynamics. Furthermore, Baker’s (1997) results present a fairly good
agreement with the experiments in the case of flutter computations for a standard aeroelastic
wing configuration.
Another extension of the Giesing et al. (1976) correction method was presented by
Jadic et al.. (1999, 2000), who developed a procedure which takes pressures instead of loads
as reference conditions. Furthermore, their procedure admits the employment of multiple
reference pressure conditions, such as the method of Brink-Spalink and Bruns (2001). Each of
these reference pressure conditions is associated to a reference mode shape. The main
difference between Jadic’s et al. (1999, 2000) and Brink-Spalink and Bruns’ (2001) methods
is how the weighting operator is obtained.
The main concern of Jadic et al.. (1999, 2000) procedure is to avoid the use of
minimization techniques for the determination of the weighting factors. However, usually the
number of the given reference steady pressures is smaller than the order of the AIC matrix.
Therefore, the resulting pressure matrix, which contains the given pressures due to prescribed
mode shapes will be a rectangular matrix. At this step, the computation of the weighing
operator is performed by the inversion of the matrix of given pressures. However, since it is a
rectangular matrix, the computation of the inversion requires a minimization technique.
Jadic’s et al.. (1999, 2000) suggestion to avoid the minimization technique is based on the
computation of a set of linear steady pressures based on a linearly independent artificial mode
shape basis. Since these complementary pressure conditions are known, they will be included
in the given reference pressure matrix. Then, a modified square pressure matrix is
constructed, in which each of its columns are related to a given modal displacement. These
displacements, on the other hand, can be either associated to the reference mode shapes or to
the artificially generated ones. The resulting pressure matrix is inverted and multiplied by an
exclusively linear pressure matrix, constructed based on the same modal basis of the modified
pressure matrix. The resulting weighting operator will be a fully populated matrix to be
multiplied to the AIC matrix to match the reference pressure conditions.
As discussed by Jadic et al. (1999, 2000), their procedure presents a good strategy to
avoid the severe pressure distortion typical of the Giesing et al. (1976) method, because the
DLM interference effects are not strongly modified. Some subsonic flutter analysis results
were presented by the authors showing a good agreement between the computed data and the
experiment. However, further extension for solving transonic flutter problems (Jadic et al..
2001), based on steady nonlinear given pressure modes, did not present good results. The
reason for the bad results are because transonic flow nonlinearities cannot be restored, since
the additional artificial mode shapes are generated from the DLM method.
A disadvantage of this method, either for static aeroelasticity or flutter calculations is
the need of additional CFD computations on a set of artificially deformed lifting surface mode
shapes, in order to generate sufficient nonlinear reference conditions to compute an effective
weighting matrix. On the other hand, only a single simulation is necessary to compute the
reference conditions for obtaining a diagonal weighting matrix, which would be sufficiently
effective to perform the correction.
1.1.5 - Dau-Garner Type Methods
This class of correction methods has been developed with the objective of computing
approximate nonlinear unsteady pressures in the frequency domain to be used for flutter
calculation in transonic flow. The computation of these nonlinear pressures is performed with
the aid of semi-empirical relations, taking as reference steady state nonlinear pressure
distributions.
The first approach in this class of procedures was introduced by Garner (1977), who
presented the theoretical background of such methods. The extension of the theory was
performed by Dau (1992), resulting in the well-known Dau-Garner method (SenGupta, 1996).
The objective of Dau (1992) is to develop an expedient procedure for transonic aeroelastic
stability analysis and unsteady loads calculation for industrial applications. The mathematical
formulation of the method is presented in Appendix B.
The basic hypothesis of the Dau-Garner method (Dau, 1992) is an empirical relation
introduced by Garner (1977), where it is assumed that the ratio between the unsteady and
steady velocity potential gradients remains the same whether the flow is transonic or purely
subsonic. Once this hypothesis is assumed, it is possible to express the unsteady velocity
potential gradient as a function of the given steady state velocity potential gradient. The ratio
between the linear unsteady and steady velocity potential gradients can be computed from a
linear theory such as the doublet lattice method.
The additional condition to be considered is the relation between unsteady pressures
and the unsteady velocity potential, which may be derived from the Euler momentum
equations, in a small disturbances context. The complete derivation of such relation is
presented by Dau (1992). When the hypothesis due to Garner (1977) is considered, one
should note that the unsteady velocity potential gradient is a function of the steady state
velocity potential gradient. Likewise, the steady velocity potential gradient can be written as a
function of a steady state pressure distribution, as given by the pressure to velocity potential
relation, which was derived from the Euler momentum equations. Therefore, the unsteady
velocity potential distribution can be written as a function of steady pressure distribution
times the ratio between the unsteady and steady velocity potential gradients.
The introduction of the nonlinearities associated with transonic flow is performed by
the substitution of the steady state linear pressures in the resulting equation by the nonlinear
ones, in such way that both real and imaginary parts of the pressures are modified with respect
to a nonlinear steady state pressure distribution.
SenGupta (1996) and Schulze and Vogel (1989) presented results of the application of
this method for unsteady pressure computations. In some of the unsteady pressure
distributions, the phase components present a discrepant behavior, when compared with
experimental measurements. In most of the results obtained by SenGupta (1996), the
computed phases were different from the experimental ones by approximately about 50
degrees. This fact indicate a shortcoming in the pressure phase calculation. Schulze and Vogel
(1989) applied the Dau-Garner method for unsteady pressures calculation of the LANN wing.
Again, when their solution was compared to a finite difference computation of the transonic
small disturbance equation and also to experimental data, the computed pressures presented
wrong trends with regard to the imaginary pressure components.
Another different approach to compute unsteady pressures in transonic flow was
presented by Yonemoto (1985). His procedure is divided in two steps: an amplitude
correction, such as the procedure presented by Luber and Schmid (1982), and a phase
correction based on a semi-empirical relationship. The phase correction relationship is the
same presented by Tijdeman (1977), and it was obtained from experimental observations.
Yonemoto’s procedure was applied for flutter computation of the YXX wing model
(Yonemoto, 1985). This wing has a supercritical profile and it was tested in a wind tunnel at
transonic flow conditions. The computed flutter boundaries agree qualitatively with the
experimental ones, including the capture of the transonic dip for the test case studied.
Yonemoto suggested that further refinement of this procedure was needed for better
quantitative results.
1.1.6 - Modal Aerodynamic Influence Coefficient Matrix (MAIC) Methods
The linear operator which relates pressures to the resulting downwash due to body
motions is the aerodynamic influence coefficient matrix (AIC). However, when these body
motions are written in terms of modal generalized coordinates instead of physical
displacements coordinates the matrix operator, which would satisfy the relation between the
resulting pressures modes to displacement mode shapes, will be the modal aerodynamic
influence coefficient matrix (MAIC). These pressure mode shapes are integrated along the
body area result in a generalized aerodynamic loading. The idea of correction procedures
based on the MAIC concept consists in its replacement of the modal aerodynamic influence
coefficients matrix, computed from linear theory, by a externally computed one which takes
into account a nonlinear aerodynamic behavior.
The aerodynamic derivative factoring procedure from Suciu et al., (1990) is classified
as a MAIC correcting procedure. This method is a stripwise force matching procedure such as
presented in the methods of Pitt and Goodman’s (1987) and Zwaan’s (1985). However,
differently from these two methods, Suciu et al. replace the modal aerodynamic influence
coefficients matrix by an externally computed one.
The main objective of the work of Suciu et al. (1990) is the correction of control
surface and tab hinge moments of a wing based on steady state experimental measurements of
these quantities. Each of these loading quantities are related to a given mode shape pattern
associated to the lifting surface displacement. Suciu et al. assumed that, any lifting surface
displacement mode shape can be decomposed in a linear combination of the typical section
degrees of freedom (Theodorsen, 1934). Therefore, the vector form of the downwash is
transformed into a matrix form, where each of the columns is associated to one of the chosen
typical section degrees of freedom. After the multiplication of this new downwash matrix by
the AIC matrix it is possible to restore a pressure matrix due to the displacements of those
degrees of freedom. The result is that each of the lifting surface elements will include
contributions due the typical section degrees of freedom. By integrating all the pressures
related to elements which belong to a given strip, it is possible to quantify the strip loading,
that is, the lift, moment and controls (aileron and tab) hinge moments.
Correction factors are reduced from the relation between the decomposed linear
loading and the corresponding nonlinear reference data. This is performed from the ratio
between each of the stripwise loading derivative, due to a given typical section degree of
freedom, and the measured experimental loading derivatives. The following step is to use
these correction factors to multiply the decomposed linear loading, which are generated for
each of the reduced frequencies set of the aeroelastic stability analysis to be performed. The
addition of the lift and moments derivatives will result in the lift and moments due to the
typical section degree of freedom displacements. The generalized aerodynamic loading
matrix, for each of those reduced frequencies are computed from the multiplication of the
decomposed mode shape matrix times the corrected stripwise loading. One should note that
there is one correction factor set for each of the mode shapes taken as reference.
The results concerning this method are restricted to subsonic aeroelastic stability
analysis, and present improved prediction of flutter speeds with respect to flutter results using
an uncorrected linear model. Suciu (2000) presented some comparisons between the results
from the above procedure and the Giesing et al.(1976) procedure.. Further generalization of
this procedure, to take into account the interference effects between two or more lifting
surfaces are presented in Suciu (2003). It should be noted that no transonic flutter analysis has
been presented using such method.
An interesting approach for the transonic unsteady aerodynamic loading for flutter
analysis in this regime was presented by Liu et al. (1988), named as Transonic Equivalent
Strip (TES) method. The procedure is divided in two consecutive correction steps, a stripwise
correction followed by a spanwise correction. In the first step, a chordwise mean flow
correction is performed based on steady pressures inputs from CFD codes or measured data.
An inverse airfoil design procedure is employed to generate a new profile of an equivalent
airfoil, for each of the spanwise stations of the lifting surface, being the reference conditions
for the design algorithm the given nonlinear steady pressure distribution. The computation of
unsteady pressures in the profile, now subjected to new boundary conditions, is based on a
time domain impulse response finite difference solution of the transonic small disturbance
equation. The unsteady pressures in the frequency domain are computed from the Fourier
transform of the time response of these pressures. The second step consists in a spanwise
correction of these two-dimensional unsteady pressures to take into account the three-
dimensional effects. This correction is a phase adjustment based on the three-dimensional
linear wave propagation analogy, referred to a given downwash mode. This three-dimensional
wave analogy may be understood as the interference effect computed by the DLM.
The computation of the unsteady pressures distribution is performed through the
application of the spanwise phase correction to the two-dimensional computed pressures.
Since the downwash modes and the associated unsteady pressures modes are known, it is
possible to construct a modal aerodynamic influence coefficients matrix (MAIC) (Chen et al..
1996). Therefore, the generalized aerodynamic forces are computed and replaced in the
aeroelastic equations of motion in terms of modal coordinates, for further application in
dynamic stability analysis or unsteady pressures calculation.
The results regarding the application of this method for both unsteady pressures
computation or flutter calculation are in good agreement with the experimental data, in the
cases of isolated lifting surfaces. One disadvantage of this procedure resides in the
performance of the inverse airfoil design algorithm, when handling complex typical section
profiles such as a combination of wing-pylon-nacelles or external stores with the wing profile.
However, this method is not so expensive like three-dimensional time domain computational
aeroelasticity simulation, since it depends on two-dimensional finite difference solutions of
the transonic small disturbances equation.

1.2 Analytical Review of Correction Methods
The correction techniques reviewed in the last section present advantages and
disadvantages regarding the application context. Most of them employ steady state reference
data, which may be loads or pressures. Some of them are based on the computation of
corrected unsteady pressures from semi-empirical relations, or with the aid of computational
fluid dynamics simulations. The objective of this section is to present an analytical review of
the most applied correction methodologies, with special attention to the ones developed for
transonic flow computation. The emphasis will be to identify their failures, in order to look
for guidelines which can help the development or the enhancement of correction procedures
for approximate modeling of the transonic unsteady flow.
1.2.1 Application of Approximate Solutions for Transonic Flow Computation.
The Giesing et al.(1976) procedure depends on a least squares technique. Thus, there
is no guarantee that the modified pressure distribution, resulting from the corrected AIC
matrix, will be the same as the experimental or the ones computed with nonlinear method. In
the presence of shock waves, this pressure distortion can modify severely the aeroelastic
behavior. Furthermore, the nature of the transonic dip depends on the shock wave
characteristics with regard to its positioning and strength. Nevertheless, at least the modified
pressure distribution will guarantee that the sectional lift and moment will be the same. But
this fact is not sufficient to yield a complete restoration of the nonlinear quasi-steady flow
shock behavior as it would not be the same as the reference condition regarding the shock
dynamics contribution. Care must be taken when using such methods to unsteady transonic
flow correction because the shock wave dynamics plays an important role on the transonic
flutter. These considerations are valid for all force matching procedures, which correct the
discrete element solution AIC matrix.
In the case of the application of correction factors directly to the strip loads, the
nonlinear contribution is implicitly included in reference loading condition. This strategy is
done in Yates (1966) strip theory, and in some of the correction methods presented by Zwaan
(1985). Special attention need to be paid with regard to Yates’ (1966) method, since his
procedure is best suited for wing with high aspect ratio, because the three-dimensional effects
are not taken into account in unsteady loads computation.
The method of Brink-Spalink and Bruns (2000) is based on an optimization technique,
where the difference of the theoretical linear AIC matrix, and the one resulting from its
product by the weighting matrix is forced to be a minimum. This fact represents an
improvement when comparing with the method of Giesing et al.(1976). Brink-Spalink and
Bruns’ (2000) method presents better chances to guarantee the matching of the desired
loading condition without severe distortion the pressure distribution. However, is not possible
to guarantee that shock jump conditions present in transonic flow conditions can be restored
since the reference AIC matrix, taken from the DLM method does not have the capability to
model such transonic flow nonlinearity.
The best way to preserve the reference nonlinear flow conditions regarding the shock
structure is the direct matching of pressures by the use of a weighting operator. However, the
pressure weighting methods introduced by Rodden and Revell (1962), Bergh (1964), Bergh
and Zwaan (1966), and Luber and Schmid (1982) present problems with regard to the
evaluation of the pressure ratios necessary for the computation of the correction factors. If in
any case there is a zero or very low pressure value in the denominator, the correction will
present a tendency to blow up to very high correction factors values implying in a bad
conditioning of the resulting corrected AIC matrix.

A multiple reference pressure modes condition procedure is presented by Jadic et al.
(1999,2000). Their method is based on the introduction of a fully populated pre-multiplying
correction matrix as a pressure weighting method. The use of reference pressures associated
multiple mode shapes is a form to avoid the problem of the scaling of the pressures by small
displacements, even though the weighting operator is not applied to the downwash. But the
problem regarding shock wave characteristics of the reference flow remains the same, since
the additional pressures due to the artificial mode shapes are obtained from a linear theory.
This fact is the reason for the method to fail in the evaluation of the transonic flutter, as
presented in the results of Jadic et al.. (2001).
The other way to reformulate the adjustment of the pressures is to correct the
downwash vector, defined as the substantial derivative of the displacements of the control
points. A work which explores a downwash correction methodology is the procedure
developed by Pitt and Goodman (1987). These authors presented in their work a procedure
based on the post-multiplication of a modified AIC matrix, which relate the strip loads to its
degrees of freedom. This procedure is a force matching method because the reference
conditions are strip loads. The choice of the correction of the downwash is based on
considerations highlighted by MacCain’s (1985). This author discussed in his work that the
post-multiplication weighting operators allows modifying both the real and imaginary parts of
the downwash, thus changing the pressures amplitude and phase angles.
The results presented by Pitt and Goodman (1987) are in agreement with experimental
measured flutter speeds. However, there is no indications in the work of Pitt and Goodman
(1987) with regard to the computation of the loading phases, which can confirm MacCains’s
(1985) observations. The discrepancies found with regard to the flutter speed computation
were attributed by the authors to viscous effects. The reference loading is obtained from the
numerical solution of the transonic small disturbance equation. Indeed, viscous effects alter
the strength and location of shock waves over the wing surface, which in its turn, may also
have a significant effect on the flutter computations.
Another approach is the computation of the nonlinear unsteady pressures based on
semi-empirical hypothesis or relations. The Dau-Garner method (Dau, K, 1992) is a procedure
based on a rational formulation, whose nonlinear unsteady pressures are computed with the
aid of semi-empirical relations for unsteady transonic flows derived by Garner (1977). This
method may be applied, in an aeroelastic analysis context, by the inclusion of downwash
weighting matrices referred to these new computed unsteady nonlinear pressures. Each of
these new reference pressure conditions is related to a given reduced frequency. Therefore,
the corresponding frequency dependent AIC matrix is multiplied by a weighting operator,
which is dependent of the same frequency, to match those new computed reference
conditions.
This method presents a failure with regard to the pressures phase computation,
identified by SenGunpta (1996). The correction of the imaginary component is not effective
because for low reduced frequency value it is simply reduced to a phase shift of 180 degrees.
A mathematical explanation of this behavior is presented in Appendix B. This is probably the
main cause of the pressure phases differences detected by SenGupta (1996), in about 50o ,
when comparing with experimental results.
The correction methods based on the modal aerodynamic influence coefficient
(MAIC) matrix replacement are presented by Suciu et al., (1990) and Liu et al., (1988). It is
possible to understand the Dau-Garner method (Dau, 1992) as a MAIC approach when its
computed pressures are used to construct the generalized aerodynamic loads matrices. The
aerodynamic derivative factoring scheme presented by Suciu et al., (1990) is based on the
correction of the loading derivatives of a lifting surface divided in typical sections. The
reference conditions used to generate a corrected spanwise loading are: steady state lift and
moment derivatives associated to each of the typical section degrees of freedom. Suciu et al.,
(1990) did not present results regarding transonic aeroelastic stability investigations, since
their main concern is the control surface flutter investigation in subsonic flow. However their
procedure allows the use of reference steady or unsteady transonic loading derivatives,
resulting in MAIC matrices, which can be defined for each of the reduced frequencies of
interest. It is an interesting approach which is similar of the work presented by Liu et al.
(1988), where the MAIC approach is also adopted (Chen et al. 2000). The difference between
these two approaches is that one uses pressures distributions to construct the MAIC matrix
(Liu et al. 1988), and the other uses the spanwise loading decomposed in the typical section
degrees of freedom (Suciu et al., 1990) .
The transonic equivalent strip method (Liu et al. 1983) is procedure based on spanwise
computations of two-dimensional unsteady pressures distribution along the chord length in the
frequency domain. This computation is performed by solving the two-dimensional form of the
transonic small disturbance equations for each of the strips of a lifting surface. The results
presented by Liu et al. (1983) show a good agreement of the computed unsteady pressures
and flutter speeds when compared to the experimental measurements. The efficiency of this
procedure is due to the fact that the unsteady pressure computation is performed based on the
solution of a nonlinear equation. However, transonic equivalent strip method and a possible
extension of Suciu’s et al. (1990) procedure for transonic flutter computation are dependent
on nonlinear unsteady finite difference solutions or measured unsteady pressures.

1.2.2 – Proposal of a Correction Technique
From the analytical review of the correction methods, presented in section 1.2.1, one
should observe that the pressure correction based on the downwash weighting was not
explored by other authors. The combination of the matching of the pressures through the
downwash weighting presents good robustness and the preservation of the mean steady flow
nonlinear characteristics. These features indicate that this approach may be developed for the
prediction of the approximate unsteady transonic aerodynamic loading. Other advantages with
regard to the application of this approach are mentioned in what follows.
The matching of the pressures based on downwash correction consists in the
modification of the control points displacements vector to satisfy a given reference pressure
distribution. The advantage of this method when comparing with the pressures correction, is
that the weighting factors computation depends on a linear system equation solution instead of
a simple pressure ratio. For this reason, even though there are null displacements or pressures
given as reference conditions, the solution of the linear system of equation will not imply in
weighting factors blow up, as it happens in the pressures matching procedures at those
conditions.
Another advantage to be pointed out is related to the use of pressures instead of loads
as reference conditions. The use of loads to correct an AIC matrix which relates downwash to
pressures, may lead to distortions of the pressure distributions regarding shock wave
positioning and strength. This problem does not happen when one uses a properly defined
AIC matrix, which relates loads to the corresponding displacements in the typical section
degrees of freedom. However, the AIC matrices which results from the conventional
modeling based on discrete element solutions of the linearized potential flow equations relates
pressures to displacements ant control points. Therefore, it is natural to employ the pressure
matching for the correction of the discrete element models. The nonlinear reference pressures
may be computed from steady state CFD solutions or from experimental measurements.
The computation of the weighting function is based on quasi-steady pressure slopes, as
performed by Luber and Schmid (1988). The choice of the pressure rates instead of their
absolute values is made in order to have displacement-independent weighting functions. The
following step is the setting of a prescribed quasi steady downwash mode shape. One
suggestion is the use of a pitch mode, for example. However, it is possible to employ other
pattern of displacements, such as a frozen structural mode shape. The resulting weighting
function will post-multiply the AIC matrix, that is, it will operate over the downwash vector.
The aeroelastic analysis is performed considering such weighting operator as a post-multiplier
of the AIC matrices, for all reduced frequencies to be explored in an aeroelastic analysis.
Generally, the pressure or load matching procedures, based on steady reference
conditions, fail in obtaining unsteady pressures at higher frequencies. The magnitude of these
pressures are well approximated, but the phases computation presents wrong trends. This is so
because the pressure amplitudes do not present a great variation with the reduced frequencies
as it happens with the pressure phases (Ashley, 1980). The reason for this fact is the absence
of unsteady transonic flow information in the reference conditions (Lui et al., 1983; Ashley,
1980).
The conclusion of this analytical review is that the downwash correction procedure
based on steady reference pressures may be a good option for the discrete element
aerodynamic approximate modeling for transonic flow. Another advantage of such procedure
lies in the fact that they are inexpensive in the computational point of view, since it depends
on steady state computed pressures. The other advantage is that the matching of the pressures
does not distort the nonlinear steady flow characteristics, and also is sufficiently robust in
terms of the computation of the correcting weighting functions. However, the procedure based
on weighting functions computed from steady pressures does not present good results with
regard to the approximation of the imaginary counterpart of the nonlinear unsteady pressures.
Therefore, there is the need to improve such procedure, to adequately compute unsteady
transonic pressures.
1.3 Objectives of the Present Work
The objective of the present work is to study correction techniques applied to unsteady
transonic flow computation. Downwash weighting procedures are the chosen methods for the
investigation of correction procedures, based on their robustness and advantageous features
that were identified, such as cost and compatibility with the physics of the problem. They also
present the advantage in providing a simple way to modify the pressures obtained from the
linear theory by the replacement of externally computed or measured pressures distributions.
These procedures will be investigated using either steady or unsteady pressures. An extension
of this formulation will also be developed to circumvent the problem of the wrong pressure
phases computation, without the use of unsteady reference pressures data. The methodology
of the present work is summarized below:
- Understand the small disturbance nonlinear transonic flow characteristic with
regard to its behavior when subjected to small disturbance conditions;
- Evaluate the downwash correction techniques using either steady or unsteady
reference pressures, and also to understand its behavior based on the nature of those reference
conditions;
- Enhance the downwash correction procedure, to obtain the correct transonic
unsteady pressures based on steady reference conditions, independently of the reduced

frequency of interest. This development should be performed accordingly to a rational
formulation based on the governing equations of the unsteady linear potential flow.
1.4 Outline of the Thesis.
The present chapter presented a review of correction techniques and the statement of
the objectives of the current investigation. Chapter 2 presents the theoretical background of
unsteady linear aerodynamic modeling based on discrete element solutions, and the
aeroelastic modeling for stability analysis. A linear/nonlinear behavior investigation based on
a finite difference solution technique of the Navier-Stokes equations for unsteady transonic
flow computation is presented in chapter 3. This investigation is performed for the better
understanding of the transonic unsteady flow in a small disturbance context. Some results
with regard to this investigation are also presented in the same chapter. Chapter 4 presents
the development of the downwash correction procedure applied to the discrete element
solutions of the linearized potential flow equation. The same chapter presents the
development of an extension of the downwash correction method, which will be named as the
kernel expansion procedure. This method is based on a rational formulation, which goal is to
search for a mathematically and physically supported correction procedure which bypass the
failure and disadvantages of the downwash correction methods, with regard to unsteady
transonic pressures computation.
The results of the downwash correction and the kernel expansion methods, as well as
the comments and discussion about them are presented in chapter 5. The overall discussion
and conclusions with regard to the studies and developments are presented in chapter 6. The
same chapter presents the contribution and suggestions for future works. The bibliography is
presented in chapter 7, and the Giesing et al. (1976) and the Dau-Garner methods (Dau,
1992) are revisited in the A and B appendices respectively.

2 – UNSTEADY AERODYNAMIC MODELING AND AEROELASTIC ANALYSIS
2.1 Integral Solution of the Linearized Potential Flow Equation.
The aerodynamic modeling of unsteady linear potential flows is performed based on
methods such as the discrete kernel function approach. This section presents a brief review of
the discrete kernel function theory applied to subsonic unsteady aerodynamic modeling. The
basic governing equation over which the discrete element methods are developed is the small
disturbances linearized potential flow equation (Landahl, 1961).
φ ( x, y , z , t ) = β 2φxx + φ yy + φzz − 2 ( M ∞2 U ∞ )φxt − ( M ∞2 U ∞ )φtt = 0 (2.1)
∂2 ∂2 ∂2 ∂2 ∂2
where ≡ β2 + +
∂x 2 ∂ y 2 ∂z 2
− 2 ( M 2
∞ U ∞ ) ∂x∂t
− ( M 2
∞ U ∞ ) ∂t 2
is a linear operator
applied to the small disturbances potential.
Under the assumption of harmonic motion, the potential φ is related to a frequency
domain potential ϕ by:
φ ( x, y, z, t ) = ϕ ( x, y, z, k ) eikt , (2.2)
and Eq. (2.1) is written in the frequency domain as:
ϕ ( x, y , z , k ) = β 2ϕ xx + ϕ yy + ϕ zz − 2ikM ∞2 ϕ x + k 2 M ∞2 ϕ = 0 . (2.3)
where is a linear operator applied to the small disturbance potential in frequency domain,
(x,y,z) are the cartesian coordinates, M ∞ the free-stream Mach number, β 2 = 1− M ∞2 ,
k = ωb U ∞ is the nondimensional reduced frequency referred to a free-stream flow speed
U ∞ and a reference length b, and ω is the oscillation frequency associated to the harmonic
motion.
Chapter 2 – Unsteady Aerodynamic Modeling and Aeroelastic Analysis 30
The development of discrete element kernel function methods is based on integral
solutions of Eq. (2.1) for a configuration of interest. The integral solution is obtained by the
application of Green’s theorem to Eq. (2.1) (Chen et al., 1993), leading to:
ϕ ( x, y , z , k ) = −∫∫ ϕS dS + ∫∫ ϕD d ( S + W ) , (2.4)
S S +W
in terms of unsteady source and doublet singularity distributions ϕS and ϕD , respectively,
over a surface S and its associated wake surfaces W. The harmonic oscillating potential, for
source singularity distribution with superficial density σ , is given by:
 ikM ∞ ( M ∞ R− X ) 
σ (ξ , η , 0, k )  e β 2 
ϕS =   , (2.5)
4π  R 
 
which is a function of geometrical parameters, reduced frequency and Mach number. The
coordinates (ξ , η , 0) define the position of an equivalent point source of strength σ dS
resulting from the density integration over the element surface S; and ( x, y, z ) are the
corresponding coordinates of a receiving point in the field to be influenced by the singularity.
The distances between these two distinct points are given by R = X 2 + β 2 (Y 2 + z 2 ) , where
X = x − ξ and Y = y − η . The corresponding doublet potential is given by the derivative of
the source potential with respect to the normal direction :
 ikM ∞ ( M ∞ R− X ) 
∆ϕ (ξ , η , 0, k ) ∂  e β 2 
ϕD =   . (2.6)
4π ∂n  R 
 
One should note that there is a common term inside the parenthesis of Eq. (2.5) and Eq. (2.6).
This term is named the kernel of the integral relation, denoted as Kϕ and given by:
ikM ∞
( M ∞ R− X )
β2
e
ϕ0 = = Kϕ (2.7)
R
The kernel is the free space Green’s function of Eq. (2.3). It represents the solution of the Eq.
(2.3) over an unbounded three-dimensional domain with a point unsteady source of unitary
intensity, concentrated in the point (ξ , η , 0) (Chen et al., 1993). The notation for the doublet
singularity as ∆ϕ is used because the doublet strength can be understood as velocity
potential jump across the surface over which the doublets are distributed. Rewriting Eq. (2.4)
in terms of the kernel function one obtains:
1 1 ∂K
ϕ ( x, y , z , ik ) = − ∫∫ σ (ξ , η , 0, ik )Kϕ dS + ∫∫ ∆ϕ (ξ , η , 0, ik ) ϕ d ( S + W ) (2.8)
4π S 4π S +W ∂n
The change of the notation of the variable argument k to ik is introduced to reinforce the
simple harmonic motion nature associated to the unsteady perturbation potential.
The objective of discrete kernel function methods is to solve the problem of unsteady
aerodynamic modeling of general aircraft configurations, by subdividing its surface, as well
as the associated wakes, in discrete elements. It is assumed that each of these elements contain
elementary solutions of the governing equations. The main concern of this brief review is the
unsteady aerodynamic modeling of planar lifting surfaces, which was chosen for simplicity.
This is the reason why the out of plane coordinate of the source point is set to zero in Eq.
(2.8).
In the case of linear unsteady aerodynamics of thin lifting surfaces, such as a wing, the
thickness effects are of second order. Therefore, it is assumed that only a doublet singularity
distributed on the lifting surface plane may be used to represent the wing. The source
singularity is usually employed to model thickness effects whereas doublet singularity is used
to generate lift. This fact reduces Eq. (2.8) to a simpler form given by:
1 ∂K
ϕ ( x, y , z , ik ) = ∫∫ ∆ϕ (ξ , η , 0, ik ) ϕ d ( S + W ) . (2.9)
4π (S +W ) ∂n
The integral relation presented in Eq. (2.9) represents the doublet singularity distribution over
the lifting surface and its wake (S+W). The discretization of wing-like components and their
associated wake surfaces imply that discrete element aerodynamic modeling is required not
only for the lifting surface, but also for the wake surfaces. This fact greatly increases the
computational effort, because the wake needs to be discretized from the lifting surface trailing
edge downstream to a position sufficiently far from the wing, to simulate far field conditions.
Therefore, the acceleration potential, which is directly proportional to a perturbation pressure
field, is introduced in order to avoid the need of wake modeling, since there is no pressure
jump across wake surfaces. Consequently, the integration limits of Eq. (2.9) are restricted to
the lifting surface domain S.
The linearized potential flow equation problem can be formulated either based on
velocity potential or acceleration potential approaches. The acceleration potential ψ is
defined as the linearized unsteady pressure derived from the linearized unsteady Bernoulli
equation (Bisplinghoff et al., 1951; Liu, 1988), and is given by :
∂∆ϕ ∆C p (ik )
∆ψ ( x, y, z , ik ) = + ik ∆ϕ = − . . (2.10)
∂x 2
Furthermore, quantities such as the velocity potential and its derivative with respect to
the flow direction “x”, and the proper acceleration potential satisfy (∆ϕ) = 0 , (∆ϕ ) = 0 x
and ψ = 0 , respectively, since is a linear operator. Eq. (2.10) is an ordinary differential
equation. Therefore, the velocity potential is obtained by the integration of the acceleration
potential as:
x
ϕ ( x, y, z, ik ) = e ikx
∫e
ikx0
ψ ( x0 , y, z , ik ) dx0 , (2.11)
−∞
where xo is an auxiliary variable. The transformations between the unsteady doublet
singularity and the pressures are given by Eq. (2.10) and (2.11). The choice of the lower limit
of the integral is to satisfy the condition that ϕ vanishes as x → −∞ , that is, ahead of the
lifting surface. When the transformation from the velocity to the acceleration potential is
performed, by the substitution of Eq. (2.11) in Eq. (2.9), the velocity potential is given as a
function of the perturbation pressure field by :
1
ϕ ( x, y , z , ik ) = − ∆C p (ξ , η , 0, ik )K ψ ( X , Y , 0, ik ) dS .
8π ∫∫
(2.12)
S
The remaining function inside the integrand of Eq. (2.12) to be rewritten in terms of
the acceleration potential is a new kernel function K ψ of the integral relation given by Eq.
(2.12). This new kernel function is obtained as a function of the acceleration potential from
the relationship between the harmonically oscillating velocity potential and the acceleration
potential as:
X
∂Kϕ
K ψ ( X , Y , 0) = e−ikX ∫e
ikx0
( x0 , Y , 0) dx0 (2.13)
−∞
∂n
and the final expression for the new kernel function is given by:
 X  ikM ∞  
  ( M ∞ R− x0 )
 
∂  −ikX β2
∫
e
K ψ ( X , Y , 0, ik ) = e eikx0   dx0  (2.14)
 
 x0 + β (Y + z )  
∂n  2 2 2 2
 −∞
The objective at the present step is to establish a relationship between the boundary
conditions and the solution for the velocity potential given by Eq. (2.12), which is a function
of the acceleration potential, or in other words, the linearized pressures. The downwash may
be understood either as the normal component of the induced flow due to a doublet potential
field or as the induced velocities due to displacements of the lifting surface, when the flow
tangency condition over the lifting surface is satisfied. This tangential flow condition in a
small disturbances context, suggests representing the downwash as the substantial derivatives
of the body displacements. On the other hand, the downwash may also be represented as the
derivative of the potential with respect to the normal direction, satisfying the following
relation:
∂h
ϕn ( x, y , 0, ik ) = w ( x, y , 0, ik ) = ( x, y, 0) + ikh ( x, y , 0) , (2.15)
∂x
where the right hand side of Eq. (2.15) results from the application of a substantial derivative
applied to a displacement mode shape vector h, in the reduced frequency domain context.
This is a Neumann boundary value problem, since the velocities are related to potential
derivatives with respect to the normal direction n.
Since a planar lifting surface has been assumed, the normal direction n is aligned with
the “z” axis. Thus, the gradient of the velocity potential field ϕ with respect to the “z”
coordinate, which is normal to the flow direction, will be the downwash. The evaluation of
Eq. (2.12) for the doublet potential over the lifting surface, that is, when z → 0 , leads to the
downwash rewritten as:
 ∂  1 
ϕz ( x, y, 0, ik )z =0 = lim  − ∫∫ ∆C p (ξ , η , 0, ik )K ψ ( X , Y , 0, ik ) dS  (2.16)
z → 0  ∂z
  8π S 
Replacing Eq. (2.14) in Eq. (2.16) yields:
    ikM ∞     
    ( M ∞ R− x0 )
   
∂ 1 ∂  X
e β2
   
ϕz ( x, y, 0, ik )z =0 = lim  − ∫∫ ∆C p (ξ , η , 0, ik ) e−ikX ∫ eikx0   dx0  dS  
z → 0  ∂z
  8π S ∂z   x02 + β 2 (Y 2 + z 2 )    
   −∞
    


(2.17)
Since pressure distributions over lifting surfaces are not functions of the normal coordinate
“z”, the final expression for the downwash is given by:
   
  2   ikM ∞
 
X
 ( M ∞ R− x0 )
∂  β2
∫∫ ∫
1 e
ϕz ( x, y, 0, ik )z=0 = − ∆C p (ξ , η , 0, ik )lim  2 e−ikX ikx0 
e  
 dx0  dS

8π  z→0 ∂ z 
  x + β (Y + z )  
2 2 2 2 
 
S 


  −∞  0   

(2.18)
This solution is also known as the Küssner relation between the acceleration potential
(linearized pressure) and the downwash between two distinct points, where ∆C p (ξ , η , 0, ik )
is the pressure coefficient difference associated to doublet strength (Küssner, 1940).
2.2 – Discrete Element Kernel Function Methods
The formulation presented in section 2.1 is the basic development of the discrete
kernel function methods. The evaluation of the kernel function is a difficult task due to the
complexity of the solution of the resulting integral relations. There are several ways to solve
the integral inside the kernel given by Eq (2.14), considering a number of approaches. These
approaches have been developed to simplify the solution of the kernel function and
correspond to the well-known methods available in the literature.
Sulaeman (2001) presented an evaluation of discrete element kernel function methods
for linear unsteady potential flow modeling. There are several methods that have been
developed and implemented, including the ones due to Landahl and Stark (1968), Albano and
Rodden (1969), Jordan (1978), and Giesing et al. (1971), Ueda and Dowell (1982) Ichikawa
et al. (1985), Rodden et al.(1997,1998), and van Zyl (1998,1999). Sulaeman (2001) reviewed
the formulations of the doublet lattice method of Rodden et al. (1997,1998), the doublet point
method of Ueda and Dowell (1982) and the hybrid doublet lattice/doublet point method of
Eversman and Pitt (1991). He performed an evaluation of these formulations and also
presented expedient approaches to solve the kernel function integrals.
The main differences between these discrete kernel function approaches concern how
the assumed singularity is distributed over an element. The integration of the associated
kernel over each of these elements is also performed in different ways for each method. The
doublet lattice method, for example, assumes that elementary doublets are distributed over the
¼ chord line of each of the panels. An additional point known as the control point needs to be
considered, and is the location over which the downwash induced by the pulsating unsteady
pressures will be computed. This location is the receiving point which is over the intersection
point between the ¾ chord line and the panel midspan line. Therefore, all elements are
influenced by the line of doublets over each of the ¼ chords of each of the panels, yet at the
same time, the downwash is computed in each of the control points.
The doublet point method of Ueda and Dowell (1982) assumes a doublet singularity
concentrated over the intersection point of the ¼ chord line of each panel and its midspan line,
and the receiving point at the same position as defined for the doublet lattice method.
Eversman and Pitt (1991) developed a combined approach between the doublet lattice and the
doublet point methods, named as the doublet hybrid method. The objective of this procedure
is to extend the doublet point method to the nonplanar case. These authors concluded that the
best approach is to utilize the nonplanar doublet point method only for sending and receiving
pairs which are distant. If the sending and receiving points are close, they suggest the use of
the nonplanar doublet lattice method. This combined technique is the basis of their doublet
hybrid method.
In subsonic aeroelastic analysis, the most widely-used unsteady aerodynamic method
is the doublet lattice method of Rodden and Johnson (1994), for its ready applicability to
complex nonplanar configurations. In that method, the integral relation given by Eq. (2.18) is
solved after the application of a variable transformation based in Richardson-Landahl’s
transform (Rodden et al, 1970), which greatly simplifies the evaluation of the integral by a
exponential substitution approach, proposed by Laschka (1963), also used by Giesing et al.,
(1971). Further refinement, with regard to kernel function integration along the element area,
was presented by Rodden et al. (1997,1998). Another discrete element kernel function method
approach was presented by Chen et al. (1993). These authors perform the unsteady subsonic
aerodynamics modeling for bodies and wings with external stores, using a method named
ZONA 6.
One should note that both source and doublet distributions are enclosed in the integral
solutions of the linearized potential flow given by Eq. (2.4). However, the evaluation of the
kernel function needs to be performed distinctly for the case of a wing-like lifting surface or a
body element. This is performed assuming either a surface doublet or a source distribution,
over the lifting surface panel and body panel respectively.
In all the aforementioned methods, it is assumed that the doublet strength is constant
over a line, a surface or concentrated in a point. The way one quantifies the pressures for the
downwash computation is not relevant, since all methods result in algebraic systems of
equations. However, a pressure distribution along each element area is conveniently
considered. This feature greatly simplifies obtaining a linear system of equations that
represents the unsteady aerodynamic model.
2.3 – The Aerodynamic Influence Coefficients Approach
The basic concept of discrete element methods is the subdivision of the lifting surface
(or body) domain into discrete elements, named also as panels. The configuration of interest
discretized into panels is known as a panel model. Each panel contains a control point where
the boundary condition is imposed. The integral equation is approximated by the summation
of elementary integrals associated with each panel.
These elementary integrals at each panel, as well as the aerodynamic interference of
one panel onto others leads to a linear algebraic system of equations relating the pressure
coefficient differences to downwash. The algebraic form of the resulting system of equations
does not depend on the discrete element kernel function method. The assembly of the
elementary integral solutions results in a matrix which elements represent the aerodynamic
influence of the panels into the control points. This matrix is named as the aerodynamic
influence coefficients matrix, which relates the structural deformations to the aerodynamic
forces. This linear relationship between the panels and the control points is justified by the
elementary solutions superposition principle (Bismarck-Nasr, 1993).
The approach used here is based on assuming constant pressure distribution along each
of the elements which discretizes the lifting surfaces and bodies (Chen et al., 1993). In other
words, it is assumed that the pressures associated to the singularity strengths are uniformly
distributed over each of the lifting surface elements Chen et al., 1993). Since these pressures
are assumed constant along the element surfaces, Eq. (2.18) is rewritten as:
∆C pj  ∂  
ϕzi ( x, y, 0, ik )z=0 = −
8π
(ξ , η , 0, ik )
∫∫ Sj
lim

 ( K ψ ) dS
 z→0  ∂z  

(2.19)
The integral relationship (2.19) between the downwash at a receiving point “i”, and
the pressures at a sending panel “j’ can be written as a system of equations as:
ϕ zi = wi = Dij ∆C p j , (2.20)
where each of the matrix elements Dij is the result of the integration of the kernel function
over the given “jth” lifting surface element geometry (Chen, 1993):
ξj ηj
1   ∂  
Dij = − ∫
8π ξ ∫ lim  ∂z K ( x − ξ , y − η , 0, M
z →0
ψ i n i n ∞ , k ) d ξn d ηn ,

(2.21)
j−1 η j−1
In the doublet lattice method this integration is performed as a line integral, since the
doublet distribution is over a line along the ¼ chord length. In the ZONA 6 method, the
integration of the kernel function is performed along the panel area. This feature turns the
ZONA 6 method of higher order than the doublet lattice method.
The inverse of the matrix operator D multiplied by the downwash vector yields the
pressure distribution. In other words, the solution of the system of equations gives the doublet
strength at each panel referred to a given downwash which is related to a displacement mode
shape. The resulting inverse matrix is named as the aerodynamic influence coefficients matrix
AIC, which is a function of the reduced frequency, and is related to the pressure distribution
by:
{∆C p (ik )} =  AIC (ik ) {w(ik )} . (2.22)
One should recall that a simple harmonic motion is assumed, hence the dependence on
ik. The coefficients of this matrix may be interpreted as rates of pressure variation due to a
given displacement amplitude input associated to the boundary conditions. Then, the
determination of the pressure coefficient vector in Eq. (2.22) is performed from the known
downwash, which is related to the amplitude of the pitch and plunge motion at each element.
The substantial derivative of a given displacement mode is composed of a derivative of the
normal direction displacement with respect to the main flow direction plus the associated
velocity scaled by the undisturbed flow speed which, in a small disturbance sense, represents
an angle of attack. Therefore, from the boundary conditions for those small perturbations, the
relationship between the normalwash and a solid boundary displacement is rewritten as:
∂h ( x , y , 0)
{w ( x, y, 0, ik )} = + ikh ( x, y , 0) =  F (ik ) {h ( x, y , 0)} (2.23)
∂x
The substantioal dierivative applied to a given modal displacement vector {h}, which
appears in Eq. (2.23) is denoted by thematrix operator  F (ik ) . The resulting aerodynamic
loading vector, {La (ik )} may be expressed as the multiplication of the pressures by an
integration matrix, [S], which is constructed from the panel elements geometrical
characteristics. The resulting final expression for the unsteady loading over the lifting surface
is given by:
{La (ik )} = q∞ [ S ]  AIC (ik )  F (ik ) {h} (2.24)

where the matrix  F (ik ) is given by:
∂ ⋅ 
 F (ik ) (⋅) =  ( ) + ik (⋅) (2.25)
   ∂x 
 
The subsonic discrete kernel function approach will be further employed as the
unsteady aerodynamic theory for computation of unsteady pressures and loads for aeroelastic
response and stability analysis, to be discussed in the next section.
2.4 – Aeroelastic Model
The aeroelastic response of a given airframe is a result of the mutual interaction of
inertial, elastic and aerodynamic forces induced by static or dynamic deformations of the
structure and other disturbances, such as gusts, atmospheric turbulence, control surface
actuation, etc.
2.4.1 – Aeroelastic Equations of Motion
The equations of motion of a discrete aeroelastic system can be represented as an
equilibrium relation between the structural and aerodynamic forces as:
[ M ]{u ( xs , ys , zs , t )} + [ K ]{u ( xs , ys , zs , t )} = {L ( u ( xs , ys , zs , t ) , u ( xs , ys , zs , t ) )} , (2.26)
where {u ( xs , ys ,zs ,t )} is the physical displacement vector, which is associated to the
displacements of the structure in a structural coordinate system ( xs , ys , zs ) . This vector is a
function of the geometric coordinates of the system and it will be denoted hereafter as {u (t )}
for simplicity. The matrices [M] and [K] are the mass and stiffness matrices respectively,
{ }
which represents the structural dynamic properties of the structure, and L ( u ( t ) , u ( t ) ) is the
aerodynamic load vector.
The matrices [M] and [K] are usually obtained from a finite element model (Bismarck-
Nasr, 1993) of the airframe. Such model is based on the discretization of the structural domain
in elements connected by nodes. At these nodes, the loading is applied and the resulting
displacements due to the structural deformation are computed. The aerodynamic loading
present in Eq. (2.26) is applied at the nodes of the structure, and the displacement vector
{u ( x, y,z,t )} is a function of the nodal displacements in the cartesian coordinates ( x, y,z ) .
The aerodynamic load vector is a function of the structural displacements and it can be
{ }
split in two parts, the induced aerodynamic forces La ( u ( t ) , u ( t ) ) and the external forces
{Le ( t )} . The external forces are usually due to gust loads, atmospheric turbulence, control
surface forces or store ejection loads. They are considered in most cases for the aeroelastic
dynamic response investigation. The induced aerodynamic load vector may be obtained from
any unsteady aerodynamic theory. These loads depends on the structural deformations
{u (t )} as an aerodynamic feedback system. Considering the dependence of the induced
aerodynamic loading with the structural displacements, Eq. (2.26) is rewritten as :
[ M ]{u ( t )} + [ K ]{u ( t )} − {La ( u ( t ) , u ( t ) )} = {Le ( t )} (2.27)
The investigation of the dynamic stability of this aeroelastic system is known as the
flutter problem. This problem consists in a investigation of the stability behavior withe respect
to parameters, such as dynamic pressure, density or flight speed. Usually the induced
{ }
aerodynamic load vector La ( u ( t ) , u ( t ) ) is a nonlinear function of the displacements of the
structure, specially in transonic flow situations. Therefore, in such cases, the flutter analysis
should be performed by a time marching procedure solving the homogeneous form of Eq.
(2.27), based on known initial conditions at t = 0. The stability boundary is obtained by
examining the decay or growth of the structural response with respect to any chosen
parameter variation.
Time marching simulations require the solution of the nonlinear aerodynamic
equations of the flow. These solutions are based on computational fluid dynamic procedures,
where the nonlinear governing equations are solved by numerical methods, such as the finite
difference, finite volume or finite elements approaches. Since the solution of the nonlinear
governing equations is performed in the time domain, in most cases this approach to solve
flutter problems is computationally expensive. However, it is the most accurate way in
obtaining the stability behavior of an aeroelastic system, because of the high fidelity
aerodynamic modeling.
The mathematical structural dynamic model, which is the left hand side of Eq. (2.26)
is in most cases linear, since in aeroelastic phenomena the amplitudes of deformations are
small (Bisplinghoff et al., 1996). When the aerodynamic response may be assumed linear
with respect to the structural deformations, the system of equations (2.27) is linear. The flutter
boundary computation should be performed by solving the complex eigenvalue problem of
this system, for a given parameter variation.
The solution of the eigenvalue problem requires that the homogeneous aeroelastic
system of equations :
[ M ]{u ( t )} + [ K ]{u ( t )} − {La ( u ( t ) , u ( t ) )} = 0 , (2.28)
be written into the Laplace domain. The transformation of the induced aerodynamic loads to
the Laplace domain is performed by a convolution integral as :
∫
 U 
La ( u ( t ) , u ( t ) ) = q∞  H  ∞ ( t − τ )   {u (τ )} dτ
  b  (2.29)
0
where [ H ] represents the aerodynamic transfer function matrix, q∞ is the dynamic pressure, b
is a reference length equal to half the reference chord, and U ∞ is the undisturbed flow speed.
The aerodynamic loading in the Laplace domain is written as :
{L ( u ( s ) )} = q
a ∞
H
 ( ) {u(s)}
sb
U∞
(2.30)
with  H
 ( ) as the Laplace domain counterpart of  H ( ) , and sb U
sb
U∞
U ∞t
b ∞ being the non-
dimensional Laplace variable. The resulting eigenvalue problem is obtained by transforming
Eq. (2.28) to the Laplace domain as:
 2 
 s [ M ] + [ K ] − q∞  H ( )  {u(s)} = 0
sb
U∞
(2.31)
2.4.2 – Modal Approach
In some cases, there is a large number of degrees of freedom associated with the
discretization of the structure, which implies in large mass and stiffness matrices. The most
adequate approach to be adopted is the transformation of the structural dynamic system model
to a modal space. This transformation is performed by the eigensolution of the homogeneous
form of Eq. (2.26) in the Laplace domain. The resulting eigenvalues are known as the natural
frequencies, and the eigenvectors are the associated mode shape vectors. Therefore, in a
modal representation, the structural displacements of the structure are written as a linear
combination of the generalized coordinates {q ( t )} and the physical displacements of the
structure {u ( t )} as:
{u ( t )} = [Φ ]{q ( t )} , (2.32)
where [ Φ ] is the modal transformation matrix, composed by mode shape vectors, each of
them associated to a given natural frequency of the structure.
The critical flutter modes usually result from the coupling of low order structural
modes, leading to a mode coalescense. The low order modes are also associated to the lower
natural frequencies of the structure. Therefore, the structural deformation of the airframe in
the flutter mode can be represented by the superposition of the lower order structural modes.
For this reason it is possible to reduce the size of the matrix system of equations by employing
the modal approach. This is performed by the choice of the lower order modes to represent the
structural dynamic behavior of the structure. The resulting aeroelastic system of equations
written in the modal space is given by:
 2    
 s  M  +  K  − q∞ Q ( )  {q(s)} = 0
sb
U∞
(2.33)
where :
 M  = [ Φ ] [ M ][ Φ ]  K  = [ Φ ] [ K ][ Φ ] ,
T T
and (2.34)
are the generalized mass and stiffness matrices respectively, and
Q
 ( ) = [Φ]  H ( ) [Φ] ,
sb
U∞
T sb
U∞
(2.35)
is the generalized aerodynamic forces matrix, with {q ( s )} being the generalized coordinates
vector in the Laplace domain.
2.4.3 Simple Harmonic Motion Approach
The computation of the aerodynamic transfer functions in the Laplace domain is a
complicated task. For this reason, unsteady aerodynamic methods are often formulated in the
frequency domain assuming simple harmonic motion for the body and a simple harmonic
aerodynamic model in the frequency domain. The aerodynamic influence coefficients

approach may be understood as an aerodynamic transfer function, however, it is computed in
a simple harmonic reduced frequency domain rather than in the Laplace domain.
Previously in this chapter, the discrete kernel function approach was presented, where
the aerodynamic transfer function was obtained in the form of the aerodynamic influence
coefficients matrix, based on the simple harmonic hypothesis. Its role is to relate the structural
deformations to the discrete element aerodynamic pressures. The induced aerodynamic loads
in the frequency domain are obtained after the multiplication of the pressure coefficients by
the dynamic pressures and by an integration matrix, resulting in Eq. (2.24).
The induced aerodynamic loading is obtained by a convolution integral which
transforms the aerodynamic loading from the time domain to the Laplace domain. Since the
AIC matrix is only available in the simple harmonic reduced frequency domain, the imaginary
counterpart of aeroelastic system of equations, given by Eq. (2.33), is obtained replacing s by
iω. leading to:
 −ω 2  M  +  K  − q∞ Q ( ik )   {q (iω )} = 0 (2.36)

 
with {u ( iω )} = [ Φ ]{q ( iω )} .
The assumption of simple harmonic motion is valid for a stability analysis of the
aeroelastic system, since damping at this condition is zero. Moreover, this hypothesis is
mathematically consistent with the assumption of the simple harmonic motion introduced by
the unsteady aerodynamic formulation.
2.4.4 – Interconnection Between Aerodynamics and Structures
It should be pointed out that the aerodynamic discretization usually does not match the
structural dynamic model discretization. The aerodynamic degrees of freedom are the
displacements at the panel control points. On the other hand, the structural degrees of freedom
are the modal displacements associated with the finite element model. Furthermore, each of
the models can be defined based on a particular coordinate system, which need not to be
coincident. Therefore, the exchange of displacement information between the structural and
the aerodynamic degrees of freedom requires an interpolation procedure, which may be
regarded as an operator [G] applied to the following transformations:
{h ( x, y,0)} = [G ]{u ( xs , ys ,zs )} and {φis } = [G ]{φia } (2.37)
where {u ( xs , ys , zs )} is the finite element model nodal displacements vector and {h ( x, y, 0)} ,
is the displacement vector at the aerodynamic control points. [G] is a general matrix
{ }
transformation operator, φis is the mode shape vector of the modal matrix [ Φ s ] , referred to
the structural degrees of freedom, as well as the vector φia { } and the modal matrix [Φ ] are a
referred to the aerodynamic degrees of freedom.
The aerodynamic loading must be also transferred from the aerodynamic to the
structural grid points. For this case, the loading transfer is performed based on the virtual
work principle, which implies that the distributed loading applied to the aerodynamic and
structural reference points must be equivalent. In other words, this equivalence relationship is
given by:
{δ h ( x, y,0)}
T
{Laero
a ( x, y, 0)} = {δu ( xs , ys ,z s )} { La ( xs , ys ,z s )}
T str (2.38)
T T
where, {δ h ( x, y, 0)} and {δu ( xs , ys ,zs )} are the virtual displacements, and the aerodynamic
loading at the structural grid points and at the aerodynamic control points are denoted as
{Lstr
a ( xs , ys ,zs )} and, {Laero
a ( x, y,0)} , respectively. The resulting relationship between the loads
in these two reference systems is given by:
{L str
a ( xs , ys , zs )} = [G ]T {Laero
a ( x, y,0)} (2.39)
Replacing Eq. (2.25) in Eq. (2.39) and using Eq. (2.37) leads to:
{L str
a ( xs , ys ,zs ,ik )} = q∞ [G ]T [ S ]  AIC (ik )  F (ik ) [G ]{u ( xs , ys ,zs )} . (2.40)
The generalized aerodynamic forces matrix results from the transformation:
Q (ik ) = [Φ a ]T [G ]T [ S ]  AIC (ik )  F (ik ) [G ][Φ a ] (2.41)

    
The most adequate interpolation operator for the connection between the structural
grid points and aerodynamic control points is based on spline function approximations. The
resulting operator G is named here as the spline matrix, which coefficients are obtained from
the interpolation functions. A description of the spline matrix interpolation methods is
presented in Zona Technology (2003), and Rodden and Johnson (1994). Further information
about the procedure for obtaining the coefficients of the spline matrix operator is presented by
Harder and Desmarais (1972), and the mathematical procedure is presented by Silva (1994),
Zona Technology (2003), and Rodden and Johnson (1994).
2.5 – Flutter Solution Techniques
The stability of an aeroelastic system is evaluated by using flutter solution techniques.
These methods are based on the solution of the eigenvalue problem with respect to a given
parameter variation. Some examples of flutter solutions techniques are the well known “k”
method due to Theodorsen (1935), also presented by Rodden and Johnson, (1994), and
Rodden and Berlinger, (1982); the “pk” method - Irwin and Guyett (1965) and Hassig (1971);
and the “g” method (Chen, 1999). The latter will be applied for the flutter computations in the
present work.
A brief review of such flutter solution techniques will be presented. Further
information about the solution algorithm, with regard to the k and pk-methods are presented
by Rodden and Johnson (1994). The detailed procedure of the g method is presented in the
works of Chen (1999) and Zona Technology (2003).
2.5.1 – Flutter Computation Based on the k-method
The flutter solution of an aeroelastic system is defined as the stability analysis of the
linear system of equations (2.36), which results from the assumption of a purely harmonic
motion, assuming s = iω . In order to evaluate the stability behavior with respect to a given
parameter variation, it is introduced an artificial damping, since the eigensolution of Eq.
(2.36), in its original form, results in imaginary eigenvalues. Therefore, Eq. (2.36), is
rewritten as:
 −ω 2  M  + (1 + ig s )  K  − q∞ Q ( ik )   {q (iω )} = 0 , (2.42)

 
where the new term ig s is an added artificial structural damping, originally introduced by
Theodorsen (1935).
This approach is known as the k-method flutter solution. It is mathematically
consistent with the assumption of simple harmonic motion introduced in the unsteady
aerodynamic formulation. The damping is only an artifice used to seek out the flutter stability
margin, and cannot be interpreted as having physical significance as a measure of the decay
rate of the aeroelastic system response. However, the method is capable of providing the
correct prediction of the flutter boundary, which, in other words, is the flutter speed at zero
damping.
2.5.2 – The pk-method
The damping values at flow conditions below the flutter speed can serve as guidelines
for conducting wind tunnel tests or flight flutter tests. For this reason, it is necessary to
consider a different approach for the flutter solution with regard to the computation of the true
damping of the aeroelastic system. Recalling Eq. (2.33) one should note that it can be
rewritten as a function of a non-dimensional Laplace parameter as:
 b  2 
  p 2
 M  + 
    ∞K  − q  Q ( p )   {q ( p )} = 0
 (2.43)
 U ∞  
where p = γ k + ik = sb U ∞ , with γ as the transient decay rate coefficient. This is the so-
called p-method equation. Its solution can provide the true damping of the aeroelastic system.
The remaining problem is the availability of the aerodynamic transfer function, which
is readily available for simple harmonic motion only. One way to circumvent this problem is a
method originally proposed by Irwin and Guyett (1965), known as the p-k flutter solution
technique. The basic assumption of this approach is the replacement of the simple harmonic
motion aerodynamic transfer function in the p-method equation (2.33). The resulting
aeroelastic system to be solved is given by:
 b  2 
  p  M  +  K  − q∞ Q ( ik )   {q ( p )} = 0
2
. (2.44)
 U ∞  
The p-k method equation is mathematically inconsistent since the eigenvalue p is
complex, (corresponding to damped oscillatory motion), while Q ( ik )  is obtained based on
undamped harmonic motion (hence the purely harmonic imaginary argument). Nevertheless,
it is believed that this method is a good approximation in finding the damped solution The
rationale for the p-k method is that for oscillatory motion with slowly increasing or decreasing
amplitude the generalized aerodynamic force based on constant amplitude harmonic motion is
a good approximation. A detailed description, with regard to the superiority of the p-k method
in comparison with the k-method, is presented by Hassig (1972).
Rodden et al. (1979) presented a modification of the original p-k method. Their
formulation shows the effectiveness of the p-k method in comparison with the k method. This
modified method is based on the introduction of an aerodynamic damping matrix into Eq.
(2.44) resulting in:
 U ∞  2 2 
  p  M  +  K  − q∞ QI ( ik ) k  p − q∞ QR ( ik )   {q ( p )} = 0 (2.45)
 b  
where the complex aerodynamic influence coefficients matrix Q ( ik )  is written as:
Q ( ik )  = QR ( ik )  + i QI ( ik ) (2.46)
with QR ( ik )  and QI ( ik )  as the real and imaginary parts of Q ( ik )  , respectively.
Substituting the non-dimensional Laplace variable, which multiplies the imaginary part of
Q ( ik )  in (2.45), yields:
 U ∞  2 2 
  p  M  +  K  − q∞ QI ( ik ) k  γ k − q∞ Q ( ik )   {q ( p )} = 0 (2.47)
 b  
One should note that Eq. (2.47) presents an extra term, when comparing with Eq. (2.44). This
term is the added damping matrix.
The eigensolution of the system of equations (2.47) is performed by varying the
parameter q∞ , or the associate flow speed U ∞ or the density ρ , for each of the modes of
interest. The resulting eigenvalues are associated to each of the chosen modes, and their
imaginary parts are compared with the reduced frequency by an iterative procedure. This
procedure requires the repeated interpolation of the matrices Q ( ik )  from a given reduced
frequency set, during the process of eigenvalue extraction. As soon as the imaginary part of p
matches the associated reduced frequency value used to compute Q ( ik )  , the associated
frequency and damping of the eigenvalue are computed and stored as a flutter solution point.
2.5.3 – The g-method
Another recent approach for the flutter solution of the aeroelastic system is presented
by Chen (2000), named as the g-method. It is a damping perturbation method, where it is
suggested that a first order term, in terms of the derivative of the aerodynamic influence
coefficients matrix with respect to the damping g, could be included in the flutter equation in
the Laplace domain. This method is rigorously derived from the Laplace domain
aerodynamics, where it is assumed the aerodynamic transfer function matrix Q ( p )  can be
defined for the whole Laplace domain. This assumption is based on the analytic continuation
of the aerodynamic transfer function matrix for decayed motion, since it is analytic for
divergent or constant amplitude motion. Therefore, assuming that Q ( p )  is analytic, it can
be expanded along the imaginary axis, around small values of the damping g as:
 ∂ Q ( p )  
Q ( p )  Q ( ik )  + g    (2.48)
 ∂ g 
  g =0
The derivative or the aerodynamic transfer function Q ( p )  with respect to the
damping is not available from the unsteady aerodynamic theory. Since the p-domain
aerodynamic transfer function is analytic, it must satisfy the Cauchy-Riemann relations such
that the derivative of Q ( p )  with respect to the damping can be written as the derivative of
Q ( p )  with respect to the complex reduced frequency ik, leading to:
Q ( p )  Q ( ik )  + g Q ' ( ik )  (2.49)

where g = γ k . Substituting Eq. (2.49 in (2.43) yields the g method equation :
 U ∞  2 2 
 p  M  +  K  − q∞ Q ( ik )  γ k − q∞ Q ( ik )   {q ( p )} = 0
'
 (2.50)
 b  
One should note that the difference between the pk method flutter equation (2.47) and
the g method equation lies in the damping matrices, Q ' ( ik )  and QI ( ik ) k  , respectively.
At zero reduced frequency, both methods will provide the same flutter boundary. In fact the
damping matrix in the pk-method is a special case of the damping matrix in the g method.
This fact can be shown by expanding Q ( ik )  by a Taylor’s expansion as:
1
Q ( ik )  = Q ( 0 )  + ik Q ' ( 0 )  + ( ik ) Q '' ( 0 )  + …
2
(2.51)
2
Since the derivative terms at ik = 0 are real, the aerodynamic transfer function for
simple harmonic motion can be split into a real and imaginary part as follows:
 1   1 
Q ( ik )  =  Q ( 0 )  − k 2 Q '' ( 0 )  + …  + i  k Q ' ( 0 )  − k 3 Q ''' ( 0 )  + …  ,(2.52)
 2   6 
dividing the imaginary part of Eq. (2.52) by k leads to the damping matrix introduced in the
pk method:
1
QI ( ik ) k  = Q ' ( 0 )  − k 2 Q ''' ( 0 )  + … (2.53)
6
Differentiating the Taylor expansion series with respect to ik gives the damping matrix
introduced by the g method as:
Q ' ( ik )  = Q ' ( 0 )  + ik Q '' ( 0 ) + … (2.54)
A comparison between Eqs. (2.53) and (2.54) indicates that QI ( ik ) k  , and
Q ' ( ik )  are equal if Q ( ik )  is a linear function of k, at k = 0 . However, the aerodynamic
transfer function matrix for simple harmonic motions is a highly nonlinear function of the
reduced frequency. The frequency dependent terms resulting form the expansion of the pk-
method aerodynamic damping matrix may introduce errors in its approximation. The
consequence of this fact is the production of unrealistic roots, during the extraction of the
aeroelastic system eigenvalues. The aforementioned comments indicate the advantage of the
g-method when comparing it with the pk-method.
2.5.4 – p-method for the Solution of a State Space Aeroelastic System.
The flutter solutions techniques presented in subsections 2.5.1 to 2.5.3 are based on an
unsteady aerodynamic model which is valid for simple harmonic motion of the body. This
approach is valid for flutter stability boundary computation. However, at conditions other than
the flutter boundary, the body is undergoing aeroelastic motion which is not purely harmonic.
Therefore, rigorously speaking, the aerodynamic damping computed with those methods does
not have a physical significance. The p-k and g methods are good solutions for computing an
estimate of the damping for arbitrary motion, but still are approximate techniques.
In order to seek out a solution which can provide the true damping of the aeroeastic
system, the p-method is based on unsteady aerodynamics approximated by rational functions
in terms of the Laplace variable s, that is, for arbitrary motions. Looking at Eq. (2.33), one
should observe that the aerodynamic influence coefficients matrix is a function of the
nondimensional Laplace variable p = sb U ∞ . The p-method consists in approximating a set
of aerodynamic influence coefficient matrices, computed for simple harmonic motion, by a
representative function of the Laplace variable s. This function is known as a Padé polynomial
(Vepa 1977), and has zero to second order terms in the Laplace variable s, as well as a series
of poles, that is:
s [Qn + 2 ]
2
 b  2 b 
nlag
Qap ( s )  = [Q0 ] + [Q1 ] s   [ 2] 
+ Q s  ∑
+ (2.55)
 U∞  ( )
 U ∞  n =1 s + b∞ β n
U
It should be recalled that sb U ∞ = ik on the imaginary axis. The βn are called lag
parameters since they are physically related to the inherent lag between motion and
aerodynamic loading in unsteady phenomena, and nlag is the number if these terms. The
values of these parameters should be obtained either by optimization procedures or by user's
experience. The coefficients Q j  , j = 1, , 2 + nlag of the Padé polynomial are obtained using
the least-square method in order to fit a set of tabulated matrices Q ( km )  . This
approximation is performed for each of the coefficients of the aerodynamic influence
coefficient matrices, independently of the other matrix elements. A more detailed description
of such procedure may be found in the work of Silva (1994).
One should note that the rational function approximation was performed based on
simple harmonic motion unsteady aerodynamics. However, the resulting approximate
aerodynamic influence coefficient matrix is valid for unsteady arbitrary motion. This fact is
justified by the analytical extension principle associated to the rational function, which now is
defined for the entire complex plane. Therefore, the resulting approximation leads to an
unsteady aerodynamic model for arbitrary motions. Once the rational approximation has been
obtained (Eq. 2.55), one can rewrite Eq. (2.33) as :
 s 2  M  +  K   {q ( s )} = q∞ Qap ( s )  {q ( s )} (2.56)
 
The most adequate approach to compute the aeroelastic stability of the system, where
the aerodynamic influence coefficient matrix is approximated by Padé polynomials, is by
transforming the system of equations from the frequency domain to the time domain. Thus, an
inverse Laplace transform should be performed leading to Eq. (2.56) written in the time
domain as :
 b 
 M  {q (t )} +  K  {q (t )} = q∞ [Q0 ]{q (t )} + q∞ [Q1 ]{q (t )}  +
 U∞ 
 b 
2
nlag (2.57)
+ q∞ [Q2 ]{q (t )}   + q∞ ∑ [Qn + 2 ]{qlag (t )}n
 U∞  n =1
The new generalized coordinates qlag (t ) are the so-called aerodynamically augmented
or lag states, and they appear because of the rational form of the poles series. They may be
obtained by applying the convolution theorem to the product of functions:
 s [Q ] 
 n+2
 {q ( s )} , (2.58)
 ( )
 s + b βn 
U∞

where the inverse Laplace transforms are known:
 [Q ] 
L −1  s {q ( s )} = {q ( t )}  = [Qn + 2 ] e −(U ∞ b ) β nt
and L −1  n+2
. (2.59)
 ( )
 s + b βn 
U∞

Therefore,
t
{q ( t )} = ∫ {q (τ )}e (
lag n
− U ∞ b ) β n ( t −τ )
dτ . (2.60)
0
The differentiation of Eq. (2.60) gives a set of ordinary differential equations for the lag
states, which can be included in the set of equations of the aeroelastic system:
{q ( t )} = {q ( t )} − Ub
lag n
β n {qlag ( t )}n , (2.61)
∞
and a vector of state space variables of the aereolastic system may be written as :
T
{λ ( t )} = {q ( t )} {q ( t )} {qlag ( t )}1 …{qlag ( t )}n  ,
T T T T
(2.62)
 lag 
The resulting aeroelastic system of equations is represented by an unique matrix with
only the freestream flow speed as argument:
{λ ( t )} = A (U ) {λ ( t )} ,
∞ (2.63)
where the state space matrix  A (U ∞ )  is the state matrix of the state space aeroelastic
representation of the system:
 − M −1 B − M −1 K − M −1Q3 − M −1Qnlag + 2 
 
 I 0 0 0 

A (U ∞ ) =  I 0 − ( )β I
U∞
b 3 0 
 , (2.64)
 
 
 I

0 0 ( )
− b β nlag + 2 I 
U∞

with,
2
 b   K  =  K  − q∞ [Q0 ]
 M  =  M  − q∞   [Q2 ] ,
 U∞  . (2.65)
 b 
 B  =  B  − q∞   [Q1 ] , Qn  = q∞ Qn 
 lag   lag 
U
 ∞
The stability investigation of such system of equations should be performed through
its eigensolutions. The resulting eigenvalues lead to aeroelastic frequency and aerodynamic
damping, which in this method has physical significance. Silva (1994) pointed out that the
differences between the computed aerodynamic damping based on the p-k method and p-
method are small. However, the expediency of state space aeroelastic modeling lies in the fact
that it is the most adequate formulation for modeling aeroservoelasticity problems.
Another fact to be pointed out is that the g-method formulation is based on an
expansion of the aerodynamic influence coefficient matrix as a first order series, with respect
to its derivative in relation to the damping g. Furthermore, the resulting approximation for the
aerodynamic influence coefficient matrix for arbitrary motion is valid based on the same
analytic continuation principle. This allows for a comparison between the p and g methods,
where p-method employs a rational function to compute the unsteady aerodynamics for
arbitrary motion, while for the g-method the unsteady aerodynamics is computed from the
approximation of the aerodynamic influence coefficients matrix by a first order series
expansion in terms of the derivative with respect to g.

3 –UNSTEADY TRANSONIC FLOW BEHAVIOR
3.1 - Linear/Nonlinear Behavior Investigation
Approximate flutter analyses in transonic flow are typically based on linear solutions
of an aeroelastic system of equations, with the inclusion of the nonlinear transonic effects, by
applying corrections to the linear aerodynamic model. However, such analyses implicitly
require that transonic aerodynamic loads, taken as reference conditions, be “locally” linear
with respect to structural deformation of a give lifting surface.
In the work of Dowell et al. (1983), some studies in the linear behavior of the
nonlinear unsteady transonic flow were performed in order to identify the range of
parameters, such as phase angle and shock strength, related to reduced frequency and angle of
attack, over which the linear behavior occurs. The objective in this case is to understand the
limits of linear behavior for aeroelastic applications, where most analyses apply linear
equations and the motions are small. Some conclusions arose from the results presented by
Dowell et al. (1983), such as, the linearity of shock wave motion with respect to a small angle
of attack change, which is the case of aeroelastic applications, and the importance of an
accurate mean flow calculation to take into account the correct shock wave position and
strength.
This chapter presents an extension of the linear/nonlinear behavior investigation
presented by Dowell et al. (1983). The nonlinear aerodynamic model over which their
investigation was performed is based on finite difference solutions of the transonic small
disturbances equations. The analysis test cases were restricted to computational simulations of
two-dimensional profiles oscillating in harmonic motion. In the present investigation,
unsteady transonic flow computations are performed based on solutions of the Navier-Stokes
Chapter 3 – Unsteady Transonic Flow Behavior 58
equations, using the finite difference procedure to be presented in the next section. The results
with regard to time domain simulations of the Navier-Stokes equations are used to verify the
“locally” linear behavior of aerodynamic loads with respect to the dynamic angle of attack.
With the present numerical technique, viscous and three-dimensional effects are taken into
account and shown to be significant.
3.2 – The Navier Stokes Nonlinear Model
The nonlinear aerodynamic computations were based in the finite difference solutions
of the Navies Stokes equations in order to have steady and unsteady pressures distribution
over the lifting surfaces. The Navier-Stokes solver used for unsteady flow computations
(Mello, 1994) is a modified version of a code developed by Sankar and Kwon (1990). It is a
finite difference solution of the vector form of the full Reynolds-averaged, three-dimensional
Navier-Stokes equations based on an arbitrary curvilinear coordinate system The time
marching consists in a two-point backward difference at the new time level n + 1 . All spatial
derivatives are approximated by standard second-order central differences (Mello, 1994). The
mathematical formulation and the description of the numerical method are here reproduced
from the work of Mello (1994).
3.2.1 – Governing Equations
The vector form of the full Reynolds-averaged, three-dimensional Navier-Stokes
equations based on an arbitrary curvilinear coordinate system can be written in non-
dimensional form as:
Qˆτ + Eˆξ + Fˆη + Gˆ ς =

Re
(
1 ˆ
Rξ + Sˆη + Tˆς ) (3.1)
where Q̂ is the vector of unknown flow properties; Ê , F̂ , Ĝ are the inviscid flux vectors;
R̂ , Ŝ , Tˆ are the viscous flux vectors and Re = ρ∞ a∞ c µ∞ is the Reynolds number based on the
free-stream speed of sound a∞ , density ρ∞ , viscosity µ ∞ and reference chord c . The
resulting non-dimensional flux vectors are given by,
 ρ   ρU c   ρVc 
 ρu   ρ uU + ξ p   ρ uV + η p 
1   1   1  
c x c x
Qˆ =  ρ v  ; ˆ
E =  ρ vU c + ξ y p  ; ˆ
F =  ρ vVc + η y p 
J  J J
ρw ρ wU c + ξ z p  ρ wVc + η z p 
     
 e  ( e + p )U c − ξt p  ( e + p ) Vc − ηt p 
 ρWc   0 
 ρ uW + ζ p  ξ τ + ξ τ + ξ τ 
1 
z xz 
1  
c x x xx y xy
ˆ ˆ 
G =  ρ vWc + ζ y p  ; R = ξ xτ xy + ξ yτ yy + ξ zτ xz 
J J
ρ wWc + ζ z p  ξ τ + ξ yτ yz + ξ zτ zz 
   x xz 
( e + p ) Wc − ζ t p   ξ x R5 + ξ y S5 + ξ zT5 
 0   0 
η τ + η τ + η τ  ζ τ + ζ τ + ζ τ 
1 
z xz 
1 
x xx y xy x xx y xy z xz 
ˆ  ˆ 
S = η xτ xy + η yτ yy + η zτ xz  ; T = ζ xτ xy + ζ yτ yy + ζ zτ xz  (3.2)
J J
η τ + η yτ yz + η zτ zz  ζ τ + ζ yτ yz + ζ zτ zz 
 x xz   x xz 
 η x R5 + η y S5 + η zT5   ζ x R5 + ζ y S5 + ζ zT5 
with J as a jacobian transformation between the curvilinear and cartesian systems written as:
J =  yξ ( xζ zη − xη zζ ) + yη ( xξ zζ − xζ zξ ) + yζ ( xη zξ − xξ zη )  .
−1
(3.3)
Uc, Vc and Wc are the contravariant velocities defined as :
U c = ξt + ξ x u + ξ y v + ξ z w
Vc = ηt + η x u + η y v + η z w (3.4)
Wc = ζ t + ζ x u + ζ y v + ζ z w
and the pressure p is related to the total energy and kinetic energy per unit volume by:
 1 
 2
(
p = (γ + 1) e − ρ u 2 + v 2 + w2  .

) (3.5)
The shear stresses are :
4 2 
τ xx = µ 
3
( uξ ξ x + uηη x + uζ ζ x ) − ( vξ ξ y + vηη y + vζ ζ y + wξ ξ z + wηη z + wζ ζ z ) 
3 
τ xy = µ ( uξ ξ y + uηη y + uζ ζ y ) + ( vξ ξ x + vηη x + vζ ζ x ) 
τ xz = µ ( uξ ξ z + uηη z + uζ ζ z ) + ( wξ ξ x + wηη x + wζ ζ x ) 
4 2 
τ yy = µ 
3
( vξ ξ y + vηη y + vζ ζ y ) − ( uξ ξ x + uηη x + uζ ζ x + wξ ξ z + wηη z + wζ ζ z ) 
3 
τ yz = µ ( vξ ξ z + vηη z + vζ ζ z ) + ( wξ ξ y + wηη y + wζ ζ y ) 
4 2  (3.6)
τ zz = µ  ( wξ ξ z + wηη z + wζ ζ z ) − ( uξ ξ x + uηη x + uζ ζ x + vξ ξ y + vηη y + vζ ζ y ) 
3 3 
with the R5, S5, and T5 variables defined as:
µ  ∂a 2 ∂a 2 ∂a 2 
R5 = uτ xx + vτ xy + wτ xz + ξ
 x + η + ζ 
Pr (γ + 1)  ∂ξ ∂η ∂ζ 
x x
µ  ∂a 2 ∂a 2 ∂a 2 
S5 = uτ xy + vτ yy + wτ yz + ξ y +ηy +ζ y 
Pr (γ + 1)  ∂ξ ∂η ∂ζ  (3.7)
µ  ∂a ∂a
2
∂a  2 2
T5 = uτ xz + vτ yz + wτ zz + ξz + ηz +ζ z 
Pr (γ + 1)  ∂ξ ∂η ∂ζ 
where Pr = µ c p k is the Prandtl number and a is the speed of the sound.
In turbulent flows, the molecular viscosity µ appearing in Eqs. (3.6) and (3.7) is
replaced by µ + µT , and the quantity µ Pr in Eq. (3.7) is replaced by µ Pr + µT PrT , where
µT is an eddy viscosity and PrT is the turbulent Prandtl number.
3.2.2 – Numerical Method
The time derivative, Q̂τ , of Eqn. (3.1) is approximated using two-point backward difference
at the new time level n + 1 ,
Qˆ n +1 − Qˆ n ∆Qˆ n +1
Qˆτ = + O ( ∆τ ) = + O ( ∆τ ) . (3.8)
∆τ ∆τ
All spatial derivatives are approximated by standard second-order central differences and are
represented by the difference operators δ,
Eˆ i +1, j ,k − Eˆ i −1, j ,k
( Eˆ )ξ
i , j ,k
( )
= δ ξ Eˆ
i , j ,k
(
+ O ∆ξ 2 = ) 2∆ξ
+ O ( ∆ξ 2 ) . (3.9)
The streamwise and normal derivatives, Êξ and Ĝζ , are evaluated implicitly at the new
time level n + 1 . The spanwise derivative, F̂η , is evaluated explicitly at the old time level n
but uses the n + 1 values as soon as they become available.
n +1 n n +1 n +1 n +1 n +1 n +1
Qˆ i , j , k − Qˆ i , j ,k
n
Eˆ − Eˆ i −1, j , k Fˆ i , j +1, k − Fˆ i , j −1, k Gˆ i , j , k +1 − Gˆ i , j ,k −1
+ i +1, j ,k + + (3.10)
∆τ 2 2 2
This semi-explicit treatment of the spanwise derivative enables the scheme to solve
implicitly for ∆Qˆ n +1 at all points at a given spanwise station at a time. To eliminate any
dependency the solution may have on the sweeping direction, the solver reverses the direction
of spanwise sweeping with every sweep.
n +1 n n +1 n +1 n +1 n +1 n +1
Qˆ i , j , k − Qˆ i , j ,k
n
Eˆ − Eˆ i −1, j , k Fˆ i , j +1, k − Fˆ i , j −1, k Gˆ i , j , k +1 − Gˆ i , j ,k −1
+ i +1, j ,k + + (3.11)
∆τ 2 2 2
The viscous terms R̂ξ , Ŝη and Tˆζ are evaluated explicitly, using half-point central
differences denoted here by the difference operator δ , so that the computational stencil for
the stress terms uses only three nodes in each of the three directions. Explicit treatment of the
stress terms still permits the use of large time steps since the Reynolds numbers of interest
here are fairly large. With the above described time and space discretizations, Eqn. (3.1)
becomes:
(
∆Qˆ n +1 + ∆τ δ ξ Eˆ n +1 + δη Fˆ n ,n +1 + δ ς Gˆ n +1 = ) 1
Re
(
δ ξ Rˆ n ,n+1 + δη Sˆ n ,n +1 + δ ς Tˆ n ,n+1 ) (3.12)
Note that all viscous terms include η-derivatives, for which known values at the new time
level “n+1” are used, hence the notation δ ξ Rˆ n ,n +1 .

Application of Eq. (3.12) to the grid points leads to a system of non-linear, block
penta-diagonal matrix equations for the unknown ∆Qˆ n +1 = Qˆ n +1 − Qˆ n . The convection fluxes
Ê , F̂ , and Ĝ are non-linear functions of the vector of unknown flow properties Q̂ .
Therefore, Eq. (3.12) is then linearized using the Jacobean matrices A = ∂Eˆ ∂Qˆ and
C = ∂Gˆ ∂Qˆ in order to write it as a linear system of equations. Since the spanwise derivative
term is obtained explicitly, it is not necessary to compute the Jacobean matrix B = ∂Fˆ ∂Qˆ .
The Jacobean matrices A and C are presented in the work of Mello (1994). The linearization
of each term of the Eq. (3.12) is obtained as follows:
n
 ∂Eˆ  ˆ n +1 ˆ n
Eˆ n +1
= Eˆ + 
n
ˆ ( 2
) ( )
 Q − Q + O ∆τ = Eˆ + A ∆Qˆ + O ∆τ
n n n +1 2
( )
 ∂Q 
n +1
n ∆ Qˆ (3.13)
A
n
 ∂Gˆ  ˆ n +1 ˆ n
Gˆ n +1
= Eˆ + n
ˆ ( 2
) ( )
 Q − Q + O ∆τ = Gˆ + C ∆Qˆ + O ∆τ
n n n +1 2
( )
 ∂Q  n +1
n ∆ Qˆ (3.14).
C
Replacing the linearized flux vectors in the left hand side of Eq. (3.12), and setting all the
semi-explicit terms in the right hand side yields:
 (  )
 I + ∆τ δ ξ An + δ ζ C n  ∆Qˆ n +1 = RHS ˆ n ,n +1
∆τ
(
= −∆τ δ ξ Eˆ n + δη Fˆ n ,n +1 + δ ζ Gˆ n + )
Re
(
δ ξ Rˆ n ,n +1 + δη Sˆ n ,n +1 + δ ζ Tˆ n ,n +1 ) (3.15)
The resulting system of linear equation has a block pentadiagonal matrix structure,
which is considerably expensive to solve. A solution to circumvent the computational cost is
the use of an approximate factorization algorithm. The chosen approximate factorization is
presented by Beam and Warming (1978), where the system of equation reduces from a block
pentadiagonal to a block tridiagonal structure:
 (  )
 I + ∆τ δ ξ An + δ ζ C n  ∆Qˆ n +1 =  I + ∆τδ ξ An   I + ∆τδ ζ C n  ∆Qˆ n +1 + O ∆τ 3
   ( ) (3.16)
which, allows the system of equations (3.15) to be written as:
 I + ∆τδ ξ An   I + ∆τδ ζ C n  ∆Qˆ n +1 = RHS

ˆ n ,n +1 (3.17)
  
Note that there is no loss of temporal accuracy, because the error incurred due to the
( )
approximate factorization is of order O ∆τ 3 . The system of equations (3.17) may now be
solved in two steps, each involving only a block-tridiagonal system of equations:
 I + ∆τδ ξ An  ∆Qˆ *n +1 = RHSˆ n ,n +1

 
 I + ∆τδ ζ C n  ∆Qˆ n +1 = ∆Qˆ *n +1
  (3.18)
However, the computational effort may be reduced by employing a diagonalization
algorithm, which reduce the block tridiagonal matrices to single diagonal ones. The procedure
to be employed is the Pulliam and Chausse (1981) algorithm. Since the Jacobean matrices A
and C have a complete set of eigenvalues and the corresponding eigenvectors, the following
similarity transformation may be used to diagonalize A and C :
A = Tξ Λξ Tξ−1 ;and C = Tζ Λζ Tζ−1 (3.19)
where the diagonal matrices Λξ and Λζ contents the eigenvalues of the matrices A and C
respectively, as far as the associated eigenvectors (transformation) matrices Tξ and Tζ . These
matrices are presented in the work of Mello (1994). Applying the similarity transformations
(2.22) to Eq. (2.20) and using the identities [ I ] = Tk Tk−1 yields:
(
 T T −1
) ( ) (
+ ∆τδ ξ Tξ Λξ Tξ−1   Tζ Tζ−1 ) ( )
+ ∆τδ ζ Tζ Λζ Tζ−1  ∆Qˆ n +1 = RHS
n n n n
ˆ n ,n +1 (3.20)
 ξ ξ    .
The fundamental simplification of the diagonal algorithm is obtained by moving Tξ
and Tζ outside the difference operators δ ξ and δ ζ , respectively:
( )
Tξn [ I ] + ∆τδ ξ Λξn  N n [ I ] + ∆τδ ς Λςn  Tτ−1 n
∆Qˆ n +1 = RHS
ˆ n ,n +1 (3.21)
where
ˆ n , n +1 = ∆τ δ Eˆ n + δ Fˆ n ,n +1 + δ Gˆ n + ∆τ δ Rˆ n ,n +1 + δ Sˆ n , n +1 + δ Tˆ n ,n +1
RHS ( ) ( ) , (3.21)
ξ η ς ξ η ς
Re
and N = Tξ−1Tζ . This simplification introduces and error because Tξ and Tζ are functions of
(ξ ,η , ζ ) and cannot be arbitrarily brought in or out the derivatives. It is however believed that
the errors introduced are of order O ∆τ 2 ( ) and have the effect of making the scheme first
order accurate in time, the same order as the previous approximations.
The solution of Eqn. (3.20) involves two block-tridiagonal systems where the blocks
are diagonal matrices. The solution of Eq. (3.21) is obtained through the following steps:
( ) RHS
[ I ] + ∆τδ ξ Λξn  ∆Qˆ *n +1 = Tξ−1
n
ˆ n , n +1
[ I ] + ∆τδ ζ Λζn  ∆Qˆ **n +1 = ( N ) ∆Qˆ

n
−1 * n +1
(3.22)
∆Qˆ n +1 = T −1∆Qˆ **n +1
ζ .
The use of standard central differences to approximate the spatial derivatives can give
rise to the growth of high frequency errors in the numerical solution with time. To control this
growth, a set of 2nd/4th difference nonlinear, spectral radius based, explicit artificial
dissipation terms are added to the discretized equations. A second order implicit dissipation is
used to help the overall numerical stability of the scheme (Pulliam and Steger, 1986). The
systems of equations in (3.22) are modified as follows:
[ I ] + ∆τδ ξ Λξn − ∆τε I ∇ξ φ1∆ξ J  ∆Qˆ *n +1 = Tξ−1 ( ) ( RHS )

n
ˆ n , n +1
− Dˆ en ,n +1
(3.23)
[ I ] + ∆τδ ζ Λζ − ∆τε I ∇ζ φ3∆ζ J  ∆Qˆ **n +1 = ( N ) ∆Qˆ
n
n −1 * n +1
,
where ε I is the coefficient of implicit numerical dissipation, ∇ξ and ∆ξ denote backward
and forward difference operators, respectively, that is., ∇ξ ( i )i , j ,k = ( i )i , j ,k − ( i )i −1, j ,k  ∆ξ and

 
∆ξ ( i )i , j ,k = ( i )i +1, j ,k − ( i )i , j ,k  ∆ξ . The quantities φ1 and φ3 are defined as:

 
 
V +a A4 Wc + a A6 
(φ1 )i , j ,k =  J −1
( )
 i + 2 , j ,k

U c + a A1  1 1 + max  c
U + a
, 

A1 U c + a A1 i + 1 , j ,k 

  c
 2  , (3.24)
 
(φ3 )i , j ,k =  J −1 ( 1 + max  U c + a A1 Vc + a A4 
Wc + a A6  )
 i , j ,k + 12  W + a
, 
A6 Wc + a A6 i , j ,k + 1 

  c 2 
with A1 = ξ x2 + ξ y2 + ξ z2 , A4 = η x2 + η y2 + η z2 and A6 = ζ x2 + ζ y2 + ζ z2 as auxiliary metric terms.
Dˆ en ,n +1 is the explicit fourth order dissipation, given by:
Dˆ en ,n +1 = ∆τε E ∇ξ φ1 ( ∆ξ ∇ξ ∆ξ ) + φ2 ( ∇η ∆η ) + ∇ζ φ3 ( ∆ζ ∇ζ ∆ζ )  JQˆ n

2
, (3.25)
 
where ε E is the coefficient of explicit numerical dissipation and
 
 U c + a A1 Wc + a A6 
( φ 2 )i , j , k
1
(
= J −1 Vc + a A4
2
) 1
i , j + ,k

1 + max  V + a A , V + a A 

 . (3.26)
2   c 4 c 4 i , j + 1 , k 
 2 
For points adjacent to the computational boundaries, second-order explicit dissipation is used
instead of Eq. (3.25).
The above described fourth order dissipation may lead to oscillations near shocks. To
avoid this problem, a switching function based on the second normalized streamwise
derivative of pressure :
pi +1 − 2 pi + pi −1
, (3.27)
pi +1 − 2 pi + pi −1
is used to replace the fourth order dissipation with second order dissipation near shocks
(Jameson et al., 1981).

3.2.3 Turbulence Model
A slightly modified version of the Baldwin-Lomax algebraic turbulence model
(Baldwin and Lomax, 1978), is used, where the maximum shear stress is used instead of the
wall shear stress because in the vicinity of separation points, the shear stress values approach
zero at the wall. In this model, two layers are considered; in the inner layer, µT is given by:
µTinnerlayer = ρ 2
m ω (3.28)
where ω is the mean vorticity, given by:
2 2 2
 ∂w ∂v   ∂u ∂w   ∂v ∂u 
ω =  −  + −  + −  (3.29)
 ∂y ∂z   ∂z ∂x   ∂x ∂y 
and m is the mixing length, given by:
m =κd (1− e − d + A+
) (3.30)
where κ = 0.41 is the von Kármán constant, d is the distance from the wall, A+ = 26.0 is the
van Driest constant, and
ρτ max
d+ = d (3.31)
µ∞
The modification with respect to the original Baldwin-Lomax model is apparent in Eq.
(3.31), where τ max is used instead of τ wall . In the outer layer, µT is given by:
( µT )outerlayer = K c ρ c1Fw Fk (3.32)
where K c = 0.0168 is Clauser's constant, c1 = 1.6 is an empirical constant, Fw is given by:
 2
d maxU dif 
Fw = min  d max Fmax , 0.25  (3.33)
 Fmax
 
where
 ω
Fmax = max  m  , (3.34)
 κ 
U dif = max ( )
u 2 + v 2 + w2 − min ( u 2 + v 2 + w2 ) , (3.35)
and d max is the distance from the wall where Fmax occurs. Also in Eq. (3.32) Fκ is given by:
−1
  0.3d  
6
Fκ = 1 + 5.5    (3.36)
  d max  
The switch between inner and outer zones occurs at the distance d c , defined as the smallest
distance from the wall for which ( µT )innerlayer = ( µT )outerlayer , i.e., the values from Eqs. (3.28)
and (3.32) are the same.
It should be noted that this change to the Baldwin-Lomax model allows the method to
treat mild separation, but it is not clear to what extent the results would be valid for massive
separation. Since the scope of the present investigation is the computation of unsteady flow
behavior of oscillating wings at moderate amplitudes of angle of attack, this model is
adequate and will be applied to the problems at hand (Mello and Azevedo, 2000).
3.2.4 Boundary Conditions
The boundary conditions must be specified along the solid surface, at the wing root
and the far field boundaries. The solid surface corresponds to the plane k=1. The unknown
conservative variables ( Q̂ ) are obtained for values between k=2 and k=(kmax-1). At the end
of each iteration, the new conservative variables values are computed at the solid surface, that
is, for k=1 where the densities and pressures are computed assuming a adiabatic wall
condition, in other words the normal derivatives ∂p ∂n = ∂ρ ∂n = 0 . For the case of near
orthogonal grids this conditions are satisfied as :
4 ρi , j ,2 − ρi , j ,3 4 pi , j ,2 − pi , j ,3
ρi , j ,1 = and pi , j ,1 = . (3.37)
3 3
The velocities at the surface are computed from the non-slip condition as, ui , j ,1 = ( xτ )i , j ,1 ,
vi , j ,1 = ( yτ )i , j ,1 and wi , j ,1 = ( zτ )i , j ,1 .
The wing root corresponds to the plane j = 1. The unknown conservable variables
vector includes values from j = 2 to j = (jmax-1). Their values at k = 1, that is, at the root, are
not updated; when computing the residual at the j = 2. The fluxes at j = 1 are computed with
the symmetry condition that the contravariant velocity normal to the boundary vanishes, i.e.,
V = 0. The pressure values at j = 1 are computed using zeroth order extrapolation, so that
pi ,1,k = pi ,2,k .
The far field boundaries are located outboard of the wing tip, downstream beyond the
trailing edge. The downstream (i = 1 and i = imax) and outboard (j = jmax) boundaries are
treated in the same way. The velocity normal to the boundary is computed. Then, the
boundary conditions are imposed depending on whether it is an inflow or outflow and
whether it is subsonic or supersonic:
•Supersonic outflow: all variables are extrapolated from the interior of the domain;
•Subsonic outflow: the pressure is fixed to be the free-stream value and the other
variables are extrapolated;
•Subsonic inflow: the density is extrapolated from the interior of the domain and the
other variables are fixed from the free-stream;
•Supersonic inflow: all variables are fixed to be the free-stream values.

3.3 – Transonic Flow Investigations of a Low Aspect Ratio Wing
3.3.1 - Preliminary Considerations
The finite difference solution of the Navier Stokes equations has been applied to the
simulation of the F-5 wing, since it is a low aspect ratio wing, where the three-dimensional
effects are more significant. The numerical method used in the present work has been
previously validated (Mello, 1994) against experimental data (Tijdemann et al., 1978) for the
case of a model of the F-5 wing at several Mach numbers up to 0.95, in steady state
conditions, and oscillating at frequencies of 20 and 40 Hz. This wing was tested in wind
tunnel, under transonic flow conditions, oscillating in rigid pitch around an axis perpendicular
to the root midchord, around a steady state angle α o = 0.0 0 , and in the steady state situation.
In the work of Mello (1994) the computational simulations were performed employing
a hybrid Navier-Stokes/full potential method. This method is based on the subdivision of the
computational domain in two zones. In the inner zone, where the viscous effects are more
relevant, the Navier Stokes equations are solved, at the same time that in the outer zone, the
solution is based on the full-potential equations. Further descriptions about this numerical
approach are found in the work of Mello (1994). Even though of the hybrid method presents
good results and a lower computational cost, in the present investigations, the Navier-Stokes
solution has been employed because, at the present time, its computational cost is not
prohibitive, even for unsteady flow computations.

3.3.2 - Validation of the Computational Procedure for Steady Flow Computations
In a first step, a steady state simulation of the F-5 wing is performed for the validation
of the Navier-Stokes method. The chosen undisturbed flow conditions to perform the desired
validation are: M ∞ = 0.95 , and Reynolds number, here defined as Re1 = ρ∞U ∞ cr µ ∞ is 11
million, scaled by the wing root chord cr = 0.6396 meters. This wing was tested in the NLR
laboratories, under these same transonic flow conditions, at steady state angle of attack
α o = 0.0 0 . This undisturbed flow speed is chosen because, at this situation, the strongest shock
waves occur, that is, where the transonic effects are more pronounced. The numerical method
used here does not include any prediction of transition location, which may be arbitrarily set.
For the conditions under consideration the flow is considered completely turbulent.
The computational mesh here employed was an algebraic generated “C” type
topology, with 141 points in the ξ direction, tangent to the lifting surface boundary, where
121 points are over the lifting surface. In the η direction, normal to the wing surface, there are
41 points between the solid surface and the limit of the computational mesh. And finally, in
the ζ direction there are 22 points aligned with the spanwise direction, where 17 points are
over the lifting surface, and the remaining are between the wing tip and the computational
domain limit. The computational mesh, which surrounds the lifting surface, as well as its
profile are presented in Fig. (3.1).

0.2
upper surface
lower surface
0.1
t/c 0.0
0 0.2 0.4 0.6 0.8 1
-0.1
-0.2
x/c
Figure 3.1- Finite difference computational mesh surrounding the F-5-NLR wing and
its profile.
For consistency, the same boundary conditions were employed in the validation,
namely no-slip at the solid surface, approximate non-reflective conditions at the far field and
no-slip wall boundary conditions at the wing root, since this validation is based on
experimental measurements for a wind tunnel model. A comparison between the computed
steady pressures and corresponding experimental measurement are presented in Fig. (3.2) for
three spanwise stations.
-0.8 -0.8 -0.8

-0.6 -0.6 -0.6
-0.4 -0.4 -0.4
-0.2 -0.2 -0.2
0 0 0
C C C (calc) C C (calc)
P P 0.2 P,L P 0.2
0.2 C
P,L
(calc) P,L
C (calc) C (calc)
C (calc) P,U P,U
0.4 P,U 0.4 C (exp) 0.4 C (exp)
C (exp) P,L P,L
P,L 0.6 C (exp) 0.6 C (exp)
0.6 C (exp) P,U P,U
P,U
0.8 0.8 0.8
1 1 1
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
x/c x/c x/c
35,2 % 72,1 % 97,7 %
Figure 3.2: Steady pressure distribution over the F-5 NLR wing at M∞ = 0.95.
The dominating feature to be noted is the formation of a shock wave in the upper and
lower surfaces. The matching of the pressures in both surfaces are well predicted, except at
station 97,7 %, where the suction peak is overestimated. Nevertheless, the results are in good
agreement with the experimental data, qualifying the numerical method for the unsteady
computation of this wing at the same undisturbed flow conditions.
3.3.3 – Validation of the Computational Procedure for Unsteady Flow
Computations
The time domain unsteady flow simulations were performed from a steady state
converged solution. The unsteady flow simulation initiates imposing a harmonic pitching
motion to the wing, around the root midchord axis. Under these pitching oscillation, the lifting
surface deforms aeroelastically, since the experimentally tested wing presents flexibility.
Therefore, the lifting surface will harmonically move at a given vibration mode shape. The
test condition to be simulated is presented by Tijdeman et al. (1978), for Mach number 0.95,
and the harmonic motion frequency is 40 Hz.
The computation of the first harmonic components of the pressures is performed by
Fourier transformations applied to pressures time histories, given by:

τ1 + 2π / kr
k
( )
Re C p (kr ) =
π ∫ C p (τ ) sin ( krτ ) dτ , and (3.38)
τ 1
τ1 + 2π / kr
k
( )
Im C p ( kr ) =
π ∫ C p (τ ) cos ( krτ ) dτ , (3.39)
τ1
where kr is the reduced frequency of a harmonic motion, and τ is the non-dimensional time,
as defined in the non-dimensionalization of the Navier-Stokes equations. Here, the reduced
frequency is defined as kr = ω c a∞ , where ω is the circular frequency c , is a reference chord,

and a∞ is the speed of the sound, in conformity with the non-dimensional time defined by
τ = a∞t c . Since the time domain simulation of the Navier-Stokes equations is based on a
discrete time marching algorithm, the Fourier integrals, presented in relations (3.38) and
(3.39), are approximated by discrete Fourier transforms as follows:
kr M ∞ ∆τ m1 + mT
( )
Re C p ( kr ) =
2π∆α
∑C pm sin ( kr m∆τ ) , (3.40)
m = m1
( )
Im C p ( kr ) =
2π∆α
∑ C p m cos ( kr m∆τ ) . (3.41)
m = m1
where ∆τ is the computational non-dimensional time step over which the time marching
simulation was performed. Therefore, the complex first harmonic components of the pressure
coefficient differences C p ( kr ) are obtained from the real and imaginary part of the first
harmonic components, given by Eqs (3.40), and (3.41) as:
{C ( ik )} = ( Re {C ( ik )}) + i ( Im {C ( ik )}) .
p r p r p r
(3.42)
The time simulation period is twice the time expended in one complete harmonic
motion cycle, in order to avoid undesired transient effects due to the transition from a steady
state converged solution to the unsteady one. Hence, unsteady pressures to be acquired will be
related to the second simulation cycle. This needs to be done in order to avoid spurious effects
related to undesired frequency contents, when the transformation from the time to the
frequency domain via Fourier analysis of the acquired pressured coefficients time history is
performed. In Figs. (3.3) through (3.5) the first harmonic components of the pressure
coefficients at given spanwise stations are presented for the aforementioned test conditions.
The time domain unsteady flow simulations were performed for two time steps, in order to
evaluate its effect on the unsteady flow solution.

20 4
Re C (calc, ∆t=.005)
P,i,L
Re C (calc, ∆t=.005) 2
15 Re C
P,i,U
(exp)
P,i,L
Re C (exp)
Re C
P,i,U
(calc, ∆t=.002)
0
10 P,i,L
ReCpi Re C (calc, ∆t=.002)
P,i,L -2
ImCpi
5 Im C (calc, ∆t=.005)
P,i,L
Im C (calc, ∆t=.005)
-4 P,i,U
Im C (exp)
P,i,L
0 Im C (exp)
-6 P,i,U
P,i,L
-5 Im C (calc, ∆t=.002)
P,i,U
-8
-10 -10
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
x/c x/c
Figure 3.3: Real and imaginary parts of the pressure coefficients over the F-5 NLR
wing (M∞ =0.95, 35.2% spanwise station, frequency = 40 Hz).
30 3
25 Re C
P,i,L
(calc, ∆t=.005)
Re C (calc, ∆t=.005) 0
P,i,U
20 Re C
P,i,L
(exp)
Re C (exp) -3
P,i,U
15 Re C (calc, ∆t=.002)
P,i,L
ReCpi Re C (calc, ∆t=.002) Im C
P,i,L
(calc, ∆t=.005)
10 P,i,U ImCpi -6 Im C (calc, ∆t=.005)
P,i,U
Im C (exp)
P,i,L
5 Im C (exp)
-9 P,i,U
P,i,L
0 Im C (calc, ∆t=.002)
P,i,U
-12
-5
-10 -15
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
x/c x/c

3
40 P,i,L
Re C
P,i,U
(calc, ∆t=.005) 0
Re C (exp)
P,i,L -3
30 Re C (exp)
P,i,U
Re C
P,i,L
(calc, ∆t=.002)
-6
P,i,U
20 -9 Im C (calc, ∆t=.005)
ImCpi Im C
P,i,L
(calc, ∆t=.005)
ReCpi -12 P,i,U
Im C (exp)
10 P,i,L
Im C (exp)
-15 P,i,U
P,i,L
0 -18 Im C (calc, ∆t=.002)
P,i,U
-21
-10
-24
0 0.2 0.4 0.6 0.8 1
x/c 0 0.2 0.4 0.6 0.8 1
x/c
A good agreement in unsteady pressure coefficients was observed over all wing
stations, although a few discrepancies were noted at spanwise station 97.7 %, with respect to
difficulties in obtaining a smooth grid around the it within reasonably economical grid sizes.
The computations with a smaller time step show some improvement in the real part of the
pressures, mainly with regard to the shock capturing, as shown in Fig. (3.4). However, further
reduction of the time step does not guarantee a good shock capturing, since this capability
mainly depends on the discretization of the domain around the shock position. Overall, the
computed unsteady pressures correlate well with experimental measurements, validating the
numerical method to be used for the linear/nonlinear behavior investigation to be further
developed.
3.3.4 –Investigation Methodology
The linear/nonlinear behavior investigation is performed based on the same test case
employed to validate the numerical procedure. The undisturbed flow conditions, as well as the
mean angle of attack position, are the same which the validation of the numerical procedure
was based on. However, the time domain simulation of the F-5 wing is performed based on a
rigid body oscillation in pitch, that is, without considering the lifting surface flexibility,
around an axis perpendicular to the root midchord. The rigid body pitching amplitudes is also
denoted here as “dynamic angles of attack”, were 0.125o to 1.5o , varying in equally spaced
steps of 0.125o. The simulations were conducted for reduced frequencies of 0.1, 0.15, 0.2,
0.25 and 0.3.
The objective is to evaluate the behavior of the pressures and the associated loading,
as a function of the displacements of the wing and the reduced frequency. The aim was to
establish the limit for linear behavior with dynamic angle of attack for the range of reduced
frequencies tested, and at each spanwise station, so that three-dimensional effects could be
identified. From the application of the discrete Fourier transformations to the pressure time
histories, given by Eqs. (3.40) and (3.41), the same transformations are applied to the lift
coefficient CL and moment coefficient CM about the quarter-chord. These loading derivatives
are computed by the integration of the pressures along the non-dimensional chord length as:
(τ ) − C pupper (τ ) d 
1 x
CL (τ ) =
π ∫
C lower
p  (3.43)
0 c
(τ ) − C plower (τ ) 
2 x 1  x
CM (τ ) =
π ∫
C upper
p − d   (3.44)
0  c 4  c
The shock positioning along the wing section chord is determined by using the
criterion of maximum pressure coefficient gradient along the chord length. However, the grid
resolution in around the shock positions is coarse, thus a polynomial interpolation is needed in
order to find the maximum gradient position. A polynomial interpolation of third degree was
chosen, in order to yield a linear expression for the second derivative with respect to the non-
dimensional chordwise coordinate, from which the maximum gradient position can be found.
The points employed in the interpolation are chosen to be nearby the shock location, taking
into account the imposed mesh displacements required by the unsteady computations. After
the shock location time history is determined, the same discrete Fourier transformations are
applied leading to first harmonic counterparts of the shock displacement as a function of the
reduced frequency.
3.3.5 Analysis of the Linear/Nonlinear Behavior
In the present investigation, it was found that obtaining frequency content for the
aerodynamic coefficients Cp, CL and CM gave more consistent results than the ones for the
shock position, which depends on how good the polynomial interpolation is. Inadequate
polynomials would easily result in spurious shock positions. Three representative spanwise
stations at 35.5%, 72.1% and 97.7% were chosen for the present investigation, in order to
evaluate the three-dimensional effects. In the following, some results concerning the linearity
studies will be presented.
Figure (3.6) presents a sample time history for the pressure coefficient differential at
the three chosen spanwise stations. The associated corresponding frequency contents, for the
three spanwise stations under consideration for a dynamic angle of attack ∆α=1.0º and
reduced frequency k=0.1 are presented tin Fig (3.7).

0.4
35.5%
0.3 72.1%
97.7%
0.2
∆Cp
0.1
0
0.0 20.0 40.0 60.0 80.0
-0.1
-0.2 τ (non-dimensional)
Figure 3.6 : Pressure difference coefficient time history over F-5 wing (M∞ =0.95,
∆α=1.0º; k=0.1).
0.30 300 35.5%

35.5% 72.1%
0.25 250
Phase angles (deg)
72.1% 97.7%
Amplitudes
0.20 97.7% 200
0.15 150
0.10 100
0.05 50
0.00 0
0.1 0.2 0.3 0.4 0.1 0.2 0.3 0.4
k k
Figure 3.7: Frequency contents (amplitudes and phases) of differential pressure
coefficient over F-5 wing (M∞ = 0.95, ∆α=1.0º; k=0.1).
It may be observed that the pressure coefficient time response is not in a single
harmonic of the movement frequency, especially near the wing tip. This indicates a non-
linear, hysteresis-like behavior. Also, three-dimensional effects are clearly present, as a one-
harmonic sinusoidal behavior is evident at the inboard station.. Even though the first harmonic
is dominant for all stations, the ones near the tip have significant components in the second
and third harmonics. Also, these higher harmonics are out of phase with respect to the wing
motion.
It should be noted that these higher harmonic effects are not included in linear
methods such as the doublet-lattice method and correction methods based on modification of
the aerodynamic influence coefficient matrices would not be readily adaptable to incorporate
higher-harmonic corrections. However, there are situations where a second mode is close to a
multiple of a first mode. In such case, the presence of a second harmonic in the lower mode
aerodynamic response may facilitate the exchange of energy between the modes and be a
contributing factor to coalescence
At reduced frequency 0.2, the first harmonic of the response in lift and moment
coefficients, upper and lower surface shock motion were computed for the various dynamic
angles of attack. The results for station 35.5% span are shown in Fig. (3.8) for the amplitudes
and phase angles, of aerodynamic coefficients and shock motion. Linear trend lines are placed
in the amplitude plot to illustrate deviations from linear behavior. Those are not drawn in the
phase plot, because during the analysis of the phase lag variation a basic linear behavior could
not be established and thus a deviation from linearity could not be characterized.
0.35
CL
0.30
CM x10 30
CL
0.25 DXsu x10
Amplitudes
CM
Phase (deg)
DXsl x10
0.20 20 DXsu
DXsl /10
0.15
0.10 10
0.05
0.00 0
0 0.5 1 1.5 2 0 0.5 1 1.5 2
∆α (deg) ∆α (deg)
Figure 3.8: Effect of dynamic angle of attack on unsteady aerodynamic coefficients
and shock motion (amplitudes and phases; station 35.5%; k=0.2).

In order to identify the limit for each case, a linear equation was estimated so that it
would start at the origin and pass through the first and second points, which corresponded to a
dynamic angles of attack of 0.125 and 0.250 degrees. A linear equation was then reduced for
these first values, at which the linear behavior would be expected. Then this linear
relationship is used to extrapolate to the following points and a percent deviation of the actual
unsteady coefficient or shock displacement to the extrapolated value was computed. The first
point where the deviation exceeded 5% and the point before that were used in an interpolation
to find the dynamic angle of attack where the linear limit was.
Two criteria were used to establish the linear boundary: 5% deviation in pitching
moment amplitude and 5% deviation in shock motion amplitude. For the latter, the deviation
is in the upper surface case only. The lower surface shock is weaker than the upper surface
one, because the assimetry of the wing profile (see Fig. (3.1) ). The shock movement then
presents a significant dispersion probably promoted by the turbulent boundary layer.
Using the above criteria, the linear limits for the test cases corresponding to Fig. (3.8),
that is, reduced frequency 0.2 and station 35.5%, were found to be ∆α=0.61º for the pitching
moment criterion and ∆α=0.58º for the shock displacement criterion. The results for the same
reduced frequency of 0.2 at station 72.1% span are shown in Fig. (3.9) for the amplitudes and
phase angles of aerodynamic coefficients and shock motion. Linear trend lines are placed in
the amplitude plot to illustrate deviations from linear behavior. It is apparent from Fig. (3.9)
that the deviation from linear behavior occur at smaller dynamic angles. Indeed, using the
above described procedure to compute the linear limits yields linear limits of ∆α=0.34º for the
pitching moment criterion and ∆α=0.50º for the shock displacement criterion.
0.25
50
0.20 CL 40
CL
Amplitudes
Phase (deg)
0.15 CM 30 CM
x10
DXsu
DXsu
0.10 20
x10 DXsl /10
DXsl
0.05 x10
10
0
0.00
0 0.5 1 1.5 2
0 0.5 1 1.5 2 ∆α (deg)
∆α (deg)
The results for the same reduced frequency of 0.2 at station 97.7% span are shown in
Fig. (3.10) for the amplitudes and phase angles of aerodynamic coefficients and shock
motion. Again, linear trend lines are placed in the amplitude plot to illustrate deviations from
linear behavior. The deviation from linear behavior seem to occur even earlier now in terms
of dynamic angles. For this case, the computed linear limits are ∆α=0.32º for the pitching
moment criterion and ∆α=0.27º for the shock displacement criterion. Even though the results
at the outboard station should be regarded with caution, the overall presented results show that
the linear boundary is very dependent on the spanwise location, outboard being more severe.
Three-dimensional effects are therefore significant and using two-dimensional linear
boundaries in a general problem would seem to be questionable. Another feature to be noted
is that the shock strength increases with the proximity to the wing tip. Then, the linear
boundary could be affected by the upper surface shock strength.

0.25 30
CL CL
CM
0.20 CM 20 DXsu
Phase (deg)
Amplitudes
x10 DXsl /10

0.15 DXsu
10
DXsl
0.10
x10
0
0.05
0.00 -10
0 0.5 1 1.5 2 0 0.5 1 1.5 2
∆α (deg) ∆α (deg)
This linear/nonlinear behavior investigation presented some indications that unsteady
transonic flow is strongly dependent of the amplitude of the motion. However, in aeroelastic
stability analysis the local mean angle of attack variations are small, since the linear unsteady
aerodynamic models are developed considering small disturbances of the boundary
conditions. Then, the associated displacements are usually less than the aforementioned linear
limit values, indicating that, the unsteady character of the transonic flow can be considered as
a contribution predicted by a small disturbance linear aerodynamic model, to be added to a
transonic flow steady pressures distribution. Otherwise, as soon as the amplitude of the
aeroelastic deformations grows up, other non-linear phenomena can appear, and the suggested
linear aerodynamic model representation need to be further investigated.
Simulations for the several reduced frequencies with the corresponding determination
of linear limits allow for a comparison of these limits at different spanwise stations using both
linearity criteria. This is illustrated in Fig. (3.11), where the linear boundaries are plotted
against reduced frequency for the three spanwise stations (35.5%; 72.1% and 97.7%) for the
pitching moment and Fig. (3.12) for the upper surface shock displacement criteria.
0.30
0.25
k Non-linear
Linear
0.20
35.5% span
0.15 72.1% span
97.7% span
0.10
0.0 0.2 0.4 0.6 0.8 1.0
∆α (deg)
Figure 3.11: Boundaries for linear behavior based on the CM criteria.
0.30
0.25
k Non-linear
0.20
Linear 35.5% span

0.15 72.1% span
97.7% span
0.10
0.0 0.2 0.4 0.6 0.8 1.0
∆α (deg)
Figure 3.12: Boundaries for linear behavior based on the shock displacement criteria.
The results shown in Figs. (3.10) and (3.11) indicate that the shock displacement
criterion is more conservative than the pitching moment criterion, as found by Dowell et al.
(1983). In addition, the linear boundaries based on both pitching moment and shock
displacement criteria clearly depend on the spanwise station. For the first criterion, at the
inner station the behavior is similar to the two–dimensional, inviscid case as presented by
Dowell et al (1983). For outer spanwise stations, the shock is stronger and three-dimensional
effects are more pronounced. Consequently, the linear behavior is affected, as may be noted in
Fig. (3.10). A somewhat surprising result is that the shock movement-based linear limit does
not seem to increase with reduced frequency for the inner stations, and even decreases with
reduced frequency for the outer station
The linear limits calculated using the moment coefficient criterion were found to
depend significantly on the spanwise station. This indicates that simply using two-
dimensional linear limits, as presented by Dowell et al. (1983), would not be appropriate
when dealing with correction methods. Notwithstanding the dependence of the linear limits on
reduced frequency and spanwise station, it is clear that some degree of linear behavior may be
assumed allowing the approximation of an unsteady transonic flow by means of small
disturbance linear governing equations.

4 - CORRECTION METHODS FOR APPROXIMATE NONLINEAR
AERODYNAMIC MODELING
4.1 Developments on the Pressure Matching Approach
The main purpose of the correction methods to be developed here is the computation
of unsteady transonic flows for dynamic aeroelastic response and aeroelastic stability
analysis. Among the correction methods reviewed in the first chapter, the chosen procedure to
be developed, and further extended, is the pressure matching based on the downwash
weighting approach. The main idea of such procedure is the modification of the downwash
vector inside the pressure to downwash linear relationship. The transonic flow reference
conditions can be based on, either computational fluid dynamics (CFD) solutions of the
nonlinear fluid dynamic governing equations, or experimental data.
4.1.1 – Downwash Weighting Method Based on Steady Reference Conditions
The basic equation over which this procedure is developed, is the algebraic pressure to
downwash relationship, derived from the application of discrete element kernel function
methods to model the linearized potential flow equation. This relation is rewritten in a
simplified form as a function only of the reduced frequency as:
{∆C p (ik )} =  AIC (ik ) {w(ik )} . (4.1)
The downwash vector is related to the lifting surface displacements by boundary conditions
defined in a small disturbances context, and may be regarded as an effective dynamic angle of
attack at each of the lifting surface elements. In the frequency domain, these boundary
conditions are rewritten here as:

Chapter 4 – Correction Methods for Approximate Nonlinear Aerodynamic Modeling 86
∂
{w ( x, y, 0, ik )} = {h ( x, y, 0)} + ik {h ( x, y, 0)} , (4.2)
∂x
recalling that {h ( x, y, 0)} is an out of plane lifting surface displacement mode shape, and
{w ( x, y, 0, ik )} is the resulting downwash with respect to the modal motion. In steady state
conditions, this downwash is reduced to the derivative of the associated mode shape
displacements with respect to the flow direction, “x” as:
∂
{w ( x, y, 0)} = {h ( x, y, 0)} , (4.3)
∂x
leading to a steady state pressure to downwash relationship given by:
{∆Cp (ik = 0)} = [ AIC (ik = 0)]{hx } . (4.4)
The downwash correction method to be developed is based on the matching of
externally generated nonlinear quasi-steady pressure differences. The quasi-steady
approximation is defined here as the variation between two states of any given property at a
infinitely large time step. The computation of the nonlinear quasi-steady pressure differences
is performed based on a quasi-steady modal displacement, where the pressure coefficient
differences are computed at these two distinct mode shape deformation states, independently
of any time variation. For the present investigations, the nonlinear pressure differences are
obtained from CFD solutions of the Navier-Stokes equations by the use of the numerical
method presented in chapter 3. The chosen displacement of the lifting surface is a rigid body
pitching rotation mode, where the associated quasi-steady displacement is the same as a
variation in angle of attack given by ∆α = α F − α I . The indexes “I”, and “F” refer to an initial
and a final mode shape displacement state, respectively.
The nonlinear pressure coefficient differences are here denoted by the superscript “nl”.
They are computed from the solutions of the Navier-Stokes equations, as the difference of the
pressures between lower {C lp } and upper {C up } surfaces of the lifting surface as:
{∆C } = {C } − {C }
nl
p
l
p
u
p
. (4.5)
The resulting quasi-steady pressure coefficient difference contributions are defined as the
subtraction between these quantities at a final and a initial position of the lifting surface
displacement as:
{δ∆C } = {∆C } − {∆C }

nl
p
nl
p F
nl
p I
, (4.6)
which may be regarded as pressure coefficients disturbances due to the quasi steady motion.
The resulting pressure disturbance can be related to the lifting surface quasi-steady motion, in
the same way as the absolute pressure coefficients differences are related to the lifting surface
displacements, accordingly with Eq. (4.4), as:
{δ∆C p (ik = 0)} = [ AIC (ik = 0)]{αF − αI } (4.7)
However, the pressure disturbance to downwash relationship, given by Eq. (4.7), is a
result from a linear unsteady aerodynamic model. Thus, the downwash needs to be weighted
by the introduction of a multiplying matrix operator, here denoted as [WT (ik = 0)] . Its role is
to match the nonlinear quasi-steady reference conditions. The argument of the weighting
operator, denoted as ik = 0 , indicates that, its role is the matching of quasi-steady pressure
differences. Therefore, the resulting equation for the quasi-steady downwash correction is
given by:
{δ∆C pnl (ik = 0)} = [ AIC (ik = 0)][WT (ik = 0)]{αF − αI } . (4.8)
In order to have a weighting function, which is independent of the amplitude of the
quasi-steady motion, a non-dimensional pressure coefficient disturbance rate is defined,
obtained from the division of the quasi-steady pressure disturbances by the corresponding
displacements as:
  
 ∂ {δ∆C nl }   {∆C p }F − {∆C p }I  
 nl nl
 p  
  =  = {δα ∆C pnl (ik = 0)} . (4.9)
 ∂α 
 (ik =0)  (αF − αI ) 

 (ik =0)

These pressure disturbance rates are used as new reference conditions for the
weighting operator computation. Likewise, the downwash vector needs to be divided by the
amplitude of the displacement in order to have Eq. (4.8) mathematically consistent. The
downwash vector is the same as the variation in angle of attack ∆α = α F − α I . Thus, the
remaining term in the right hand side of Eq. (4.8) is the desired operator resulting from the
solution of the linear system of equations given by:
[ AIC (ik = 0)]

−1
{δα∆C pnl (ik = 0)} = [WT (ik = 0)]{1} . (4.10)
One should note that the left-hand-side of Eq. (4.10) may be regarded as the ratio of a
modified disturbance downwash mode {δ h } ,

nl
x by the quasi–steady mode
{δ h } = ∆α = α
x F − α I . This modified disturbance downwash vector, multiplied by the
aerodynamic influence coefficients matrix generates the nonlinear pressure coefficients
difference rates. The resulting downwash weighting coefficients result from the ratio between
each of the elements of both downwash vectors as:
(hxnl )i
(WT )ii = , (4.11)
(hx )i
which may compose the elements of a matrix operator [WT (ik = 0)] . This matrix is
diagonal in order to satisfy the dimensions of the matrix multiplication inside Eq. (4.10).
The pressure differences associated to the reference conditions can be obtained from
different sources, such as CFD computations or experimental measurements. In both cases,
the data acquisition points usually do not agree with the panels control points. Hence, an
interpolation is necessary for the determination of those quasi-steady pressure differences at
the desired locations, in agreement with the discrete element method aerodynamic mesh.
These interpolations can be since surface splines (Zona Technology, 2003) to linear
chordwise interpolations (Silva et al. 1999).

The subsequent step is the modification of the unsteady aerodynamic loading (Eq.
2.25) by the introduction of the weighting operator, leading to a approximate nonlinear
aerodynamic loading given by:
{Lnl
a (ik )} = q∞ [ S ]  AIC (ik ) [WT (ik = 0)]{w (ik )} (4.12)
In steady state conditions, the nonlinear aerodynamic loading is completely restored
by the application of the weighting operator. Looking at Eq. (4.12), one should note that, even
though the weighting matrix is defined based on quasi-steady conditions, it is assumed that
the weighting of the downwash will be performed for all the set of reduced frequencies to be
considered in the solution of the aeroelastic stability problem.
4.1.2 – Downwash Weighting Method Based on Unsteady Reference Conditions
Another capability of downwash correction procedures to be investigated is the use of
computed or experimental unsteady pressure differences. In such cases, the reference
conditions are based on frequency-dependent pressures, computed from the time domain
aerodynamic response. Pressure time histories, for example, may result from the lifting
surface displacements due to a prescribed motion. These displacements are arbitrary, as they
may be, for example, an impulse-type or an oscillating harmonic motions. For the present
downwash correction investigation, the unsteady pressure differences computation was
performed based on a time domain CFD solution of the Navier-Stokes equations, using the
finite differences approach presented in Chapter 3.
The chosen displacement may be any unsteady motion associated to a displacement
mode shape of the lifting surface. In the present investigations, it is chosen a rigid body
harmonic pitching rotation around the root midchord axis of the lifting surface, at a preset
reduced frequency kr . This motion leads to an unsteady downwash, which also may be
regarded as an unsteady perturbation in angle of attack, with amplitude equals to the
amplitude of the pitching motion ∆α.
Since the linear aerodynamic model to be corrected is defined in the frequency
domain, it shall be necessary to transform the reference time domain pressure differences to
the frequency domain. The most adequate approach is the first harmonic components
computation of the pressure differences using a Fourier transform algorithm applied to their
time histories. The Fourier transformations used herein to obtain the frequency domain
components of the pressure differences are given by:

τ1 + 2π / kr
k
(
Re ∆C (ikr ) =
nl
p ) π ∫ ∆C p (τ ) sin ( krτ ) dτ , and (4.13)
τ1
τ1 + 2π / kr
k
(
Im ∆C nl
p ( ikr ) ) =
π ∫ ∆C p (τ ) cos ( krτ ) dτ , (4.14)
τ1
where kr is given by kr = ω c a∞ , τ = a∞t c is the non-dimensional time, as defined in the
non-dimensionalization of the Navier-Stokes equations. Since the time domain simulation of
the Navier-Stokes equations is based on a discrete time marching algorithm, the Fourier
integrals, presented in relations (4.13) and (4.14), shall be approximated by discrete Fourier
transforms as follows:
k M ∆τ m1 + mT
(
Re ∆C nl
p ( ikr ) ) = r ∞
2π∆α
∑ ∆C p m sin ( kr m∆τ ) , (4.15)
m = m1
(
Im ∆C pnl ( ikr ) = ) 2π∆α
∑ ∆C p m cos ( kr m∆τ ) . (4.16)
m = m1
where ∆τ , is the computational non-dimensional time step over which the time marching
simulation was performed. Therefore, the complex first harmonic components of the pressure
coefficient differences ∆C pnl ( ik ) are obtained from the real and imaginary part of the first
harmonic components, given by Eqs (4.15), and (4.16) as:

{∆C ( ik )} = ( Re {∆C ( ik )}) + i ( Im {∆C ( ik )})

nl
p r
nl
p r
nl
p r
. (4.17)
{ }
Since the unsteady pressure differences ∆C pnl ( ikr ) are computed, it is possible to determine
correction factors, which satisfy a system of equations analogue to Eq. (4.10). Special care
needs to be taken regarding definition of the reduced frequency for the Navier-Stokes
simulations. The reduced frequency differs from that defined for the discrete element kernel
function method by a factor of kr = k ⋅ ( M ∞c b) , since k = ωb U ∞ . Thus, in order to have
unsteady pressures computed by the Navier-Stokes simulation and kernel function methods at
the same reduced frequency, the aforementioned relation needs to be satisfied. Another
feature to be noted is that the unsteady pressure coefficients are divided by the amplitude of
the motion ∆α , as indicated in Eqs (4.15) and (4.16).
The system of equations (4.8) is now rewritten as function of a frequency dependent
weighting operator given by:
{∆C ( ik )} =  AIC ( ik ) WT ( ik ) {w ( ik )}

nl
p r r r r , (4.18)
where {w ( ik )} ,
r is the downwash vector divided by amplitude of the motion ∆α . The
diagonal weighting matrix coefficients are obtained from the ratio between a modified
unsteady downwash, computed by:
{wnl (ikr )} =  AIC (ikr ) {∆C pnl (ikr )}

−1
, (4.19)
and the known downwash associated to the prescribed lifting surface motion, leading to:
(wnl (ikr ))i

(WT (ikr ))ii = . (4.20)
(w (ikr ))i
The nonlinear unsteady aerodynamic loading computation is performed by introducing in Eq.
(2.25), the correcting weighting operator, which multiplies the downwash vector, leading to
an approximate aerodynamic loading given by:

{Lnl
a (ikr )} = q∞ [ S ]  AIC (ikr ) [WT (ikr )]{w (ikr )} (4.21)
The same comments regarding the steady loading calculation are valid for the
unsteady case. The choice of the reduced frequency over which the nonlinear unsteady
pressure differences are referred is arbitrary, however a good suggestion is the flutter reduced
frequency predicted from the purely linear stability analysis of the aeroelastic model. This
assumption allows the best correction of the unsteady aerodynamic loading at the critical
reduced frequency, representing an improvement on the prediction of transonic effects near
the flutter boundary. However, for different reduced frequency values, there is no guarantee
that the computation of the unsteady transonic loading has physical significance. This is so
because the weighting function is computed based on a given reduced frequency value, which
shall be different from the one in this different unsteady flow condition.
4.2 – Extension of the Downwash Weighting Method Based on the Kernel
Expansion Hypothesis
4.2.1 – Motivation
An alternative for the enhancement of the nonlinear unsteady pressure based
procedure is the generation of weighting operators computed from reference pressure
differences at different reduced frequency values. These pressure differences in the frequency
domain may be computed by two possible approaches, as the Fourier transform of an impulse
or a harmonic time domain aerodynamic responses. These resulting pressure differences are
either computed for frequency spectrum, in the case of the impulse type aerodynamic
response, or at a single frequency value for the case of the harmonic aerodynamic response.
The computation of the pressure differences based on impulse type aerodynamic
response requires accurate time domain CFD simulation. This approach may be regarded as
an extension of the previous unsteady pressure based downwash correction method, where the
harmonic nature of the aerodynamic response restricts the correction to single frequency
dependent reference pressure differences. Therefore, the resulting weighting function will be
consistent for the correction of the aerodynamic model computed for the same reduced
frequency associated to the reference data.
The frequency contents of the impulse aerodynamic response is theoretically infinite,
however, its time domain computational simulation restricts the frequency response to a finite
range of values (Ballhaus and Goorjian, 1978). The suggested approach is to compute a set of
weighting functions, based on unsteady pressure differences at each of the reduced frequency
identified by the Fourier transformation applied to the time domain aerodynamic impulse
response. Therefore, the correction of the linear aerodynamic model is performed based on the
same algorithm employed for the correction based on unsteady pressure differences at a single
reduced frequency. The difference is that, now each weighting function is applied to correct
the corresponding pressure to downwash relation resulting from the linear unsteady
aerodynamic modeling for the same reduced frequency.
The disadvantage of approaches based on unsteady reference pressures is the
computational cost, since it depends on time domain CFD computation to simulate the
nonlinear unsteady aerodynamic loading. The costs are prohibitive, when the aerodynamic
configuration and governing equations are more complex.
Another possibility to minimize the computational cost is to perform the downwash
correction based on nonlinear steady pressure differences. However, in this case the
procedure, as it is defined, is not capable to take into account properly unsteady transonic
effects, since they are not embedded in the steady state solutions of the nonlinear flow.
Furthermore, the use of a single frequency dependent condition for the correction of the
aerodynamic model for different reduced frequency values is inconsistent. This fact can be
better understood considering the splitting of the aerodynamic influence coefficients matrix,
denoted as [AIC], in real and imaginary parts as [ AICR ] and [ AICI ] , respectively. The
computation of the weighting function is based on pressure differences computed at a reduced
frequency kr , where for the steady state, this value is zero. Then, the resulting pressure to
downwash relationship, is rewritten as:
{∆C nl
p (ik ) } = [ AIC R (ik ) + AICI (ik ) ][WT (ikr ) ]{hx + ikh} =
= [ AICR (ik ) ][WT (ik r ) ]{hx } + ik [ AICR (ik ) ][WT (ik r ) ]{h} + (4.22)
+ [ AICI (ik ) ][WT (ikr ) ]{hx } + ik [ AICI (ik ) ][WT (ikr ) ]{h}
One should note that none of the parcels of the relation above present a physical meaning,
since they are arbitrary, unless for the case when the reduced frequency associated to the
reference unsteady pressures is the same of the pressure to downwash equation, that is,
kr = k . Thus, a correct pressure distribution recovery is possible only by the following
relationships:
{∆C p (ik = 0)} = [ AIC (ik = 0)][WT (ik = 0)]{h }

nl
x
, (4.23)
{∆C p (ik )} = [ AIC (ik )][WT (ik )]{h + ik h}
nl
r r r x r
where it is considered the corresponding frequency dependent aerodynamic influence
coefficients matrices relating the downwash to the steady and unsteady nonlinear pressure
differences, respectively.
The above discussion suggests that the imaginary part of the pressure differences
cannot be obtained from the given steady pressure differences. A suggested approach is to
obtain the unsteady counterpart of the reference pressures, for example based on the subsonic
equation for the linearized unsteady potential flow. That is, it shall be necessary to search a
way to correct the linear unsteady aerodynamic model rewriting the downwash correction in
such way that, the nonlinear steady component contributes to the correction of nonlinear
unsteady pressure differences.
4.2.2 A Modified Downwash Correction Method
Linear unsteady aerodynamic theories, applied to the aeroelastic modeling and
analysis, are developed considering small disturbances around mean angle of attack variations
of the lifting surface. In chapter 3, a linear/nonlinear behavior investigation was presented,
indicating that unsteady transonic flow behavior is strongly dependent on the amplitude of the
motion. However, aeroelastic deformations are usually smaller than the computed linear
limits, in terms of amplitudes of the motion, which were identified in chapter 3. Based on this
observation, it is inferred that, for aeroelastic analysis in a small disturbance context,
unsteadiness of transonic flows present a linear behavior with regard to aerodynamic
derivatives and shock dynamics, when the amplitude of lifting surfaces undergoing unsteady
motion are below the linear limits.
In other words, the unsteadiness of transonic flow may be understood as the
superposition of a linear contribution to a mean steady state nonlinear pressure distribution. A
linear unsteady pressure small disturbance contribution, for example, can be computed as a
contribution predicted by a small disturbance linear aerodynamic model. The same
conclusions were indicated by Dowell et al. (1983), where they pointed out that the unsteady
pressure disturbances behavior, is linear with respect to the aeroelastic deformations around
the mean steady state pressure distribution, at sub-transonic flow conditions.
From these observations, it is possible to assume the hypothesis that an unsteady
contribution to a steady nonlinear mean flow, in small disturbance transonic flows, can be
represented by a subsonic linear model. Thus, a correction procedure based on steady mean
flow pressure distribution as reference condition, added by an unsteady contribution predicted
by the linearized potential flow equation, has a great chance to present good results regarding
unsteady aerodynamic loading prediction for aeroelastic stability analysis.
The extension of the standard downwash correction method is performed considering
the hypothesis mentioned above. The procedure is divided in two steps, the first being a
nonlinear steady mean flow correction as is performed for the correction of the steady
downwash, when nonlinear pressure differences are considered as reference conditions. The
second step is the correction of the unsteady downwash, where the unsteady counterpart of the
reference pressures to be added to the steady nonlinear reference pressures will compose new
reference pressure differences. These unsteady pressure contributions are predicted by a linear
unsteady flow aerodynamic model. Recalling that the downwash is related to the pressure
differences by an integral equation, which is written in a general form as:
1
w ( x, y, 0, ik ) = − ∆Cp (ξ , η ) K ψ ( x − ξ , y − η , 0, ik , M ) d ξ d η ,
8π ∫∫
(4.24)
A
it may also be written in an algebraic form as:
{w ( ik )} = [ D(ik )]{∆C ( ik )}
p . (4.25)
The kernel function integral relations between the sending and receiving points are
represented by the matrix operator [ D(ik ) ] , leading to a resulting equation which relates the
downwash to the pressure difference.
The matrix [ D(ik ) ] , hereafter named as the kernel function matrix, is computed for
subsonic flows and is a function of the reduced frequency. Therefore, it is assumed, as a
second hypothesis that, it is possible to expand the subsonic kernel function matrix as an
asymptotic series around small reduced frequencies. This kernel is a function of the reduced
frequency, and the domain of dependency is continuos and analytic at this circumstances.
Consequently, (Miles, 1950) the function may be expanded in an asymptotic series around
reduced frequency values smaller than 1.0. Therefore it is possible to express all dependent
variables in the equations in terms of the reduced frequency as the expansion parameter (Liu,
1977; Hui, 1969). The resulting expanded kernel matrix is given by:
[ D(ik )] ≅ ( ik ) [ D ]0 + ( ik ) [ D ]1 + ( ik ) [ D ]2 + + ( ik ) [ D ]n
0 1 2 n
n , (4.26)
= [ D ]0 + ∑ ( ik ) [ D ]m
m
m =1
where k < 1.0 , and [ D ]n , n = 1, m , are constant coefficient matrices, obtained from the
asymptotic series expansion.
The transonic flutter reduced frequency is normally less than 1.0. This assumption is
justified based on Ashley (1980) and Tijdemann (1977) observations. Both authors pointed
out that flutter instability usually results from the coalescence of low frequency aeroelastic
modes, such as the coupling of a first structural bending and a first structural torsion mode.
Since under those conditions, the flutter flow speed is high and the associated frequency is
low, the resulting flutter reduced frequency is low.
The first term of the expansion series (4.26) is the steady state kernel function matrix
([ D ] = [ D (ik = 0)]) , since this term is the unique matrix which is independent of the reduced
0
frequency. The purpose of the kernel expansion is to allow the introduction of nonlinear
steady state reference conditions in place of the original steady state kernel function. This may
be performed by the replacement of the original kernel by a non linear steady state kernel
matrix in Eq. (4.26), which was computed from the downwash weighting based on the
nonlinear steady pressure differences using the following equation:
{∆C pnl (ik = 0)} = [ AIC (ik = 0)][WT (ik = 0)]{hx } (4.27)
One should observe that the downwash correction may be understood either as a left
multiplication of the downwash, or a right multiplication of aerodynamic influence
coefficients matrix. The right multiplication results in a aerodynamic influence coefficients

matrix, in such way that its multiplication by the downwash results in steady nonlinear
pressure difference distribution. Thus, the downwash for a known steady displacement of the
lifting surface is given by:
−1
{hx } = {w(ik = 0)} =  AIC nl (ik = 0) {∆C pnl (ik = 0)}
(4.28)
=  D nl (ik = 0) {∆C pnl (ik = 0)}
−1
where  AIC nl (ik = 0) is a modified steady state aerodynamic influence coefficients
 
matrix, corrected by nonlinear pressure differences distribution. The inverse of this matrix is
denoted as  D nl (ik = 0) corresponding to a modified kernel function matrix, computed from
the weighting procedure based on the nonlinear steady pressure differences distribution.
Therefore, the resulting kernel expansion modified by the replacement of the steady
nonlinear kernel is given by:
 D nl (ik )  ≅  D nl ( ik = 0 )  + ( ik ) [ D ]1 + ( ik ) [ D ]2 + + ( ik ) [ D ]n
1 2 n
n . (4.29)
=  D nl ( ik = 0 )  + ∑ ( ik ) [ D ]m
m
m =1
The remaining term of the expansion is a function of the reduced frequency and may be
written as:
∑ ( ik ) [ D ] ≅ [ D(ik )] − [ D(ik = 0)]

m
m
(4.30)
m =1
After replacing Eq. (4.30) into the right hand side of Eq. (4.29), a modified kernel function
matrix, which takes into account the mean flow nonlinear contribution, is given by:
 D nl (ik )  =  D(ik ) − D(ik = 0) + D nl (ik = 0)  (4.31)
The inversion of  D nl (ik )  results in a modified aerodynamic influence coefficients matrix
 AIC nl (ik )  . Then, the nonlinear unsteady pressure difference is given by:
{∆C ( ik )} =  AIC
nl
p
nl
(ik )  {w(ik )} (4.32)
Once the nonlinear unsteady pressures are established, it is possible to obtain a
frequency dependent correction matrix satisfying the above relation, written in downwash
correction form as:
{∆C ( ik )} = [ AIC (ik )] WT ( ik ) {w(ik )}

nl
p (4.33)
for a given reduced frequency of interest. Then the downwash correcting matrix is obtained
from:
 D (ik ) {∆C pnl (ik )} = WT (ik ) {w(ik )} . (4.34)

   
One should note that that the left hand side represents the modified downwash related to the
unsteady non-linear pressure differences. Therefore, the weighting function will be a diagonal
matrix whose elements are obtained from the relation between both computed downwash :
WT (ik ) =
{wnl (ik )}i
  ii . (4.35)
{w(ik )}i
WT (ik ) is a complex matrix and carries the information about the nonlinear unsteady
 
pressure difference contribution. It can be used to right multiply the aerodynamic influence
coefficients matrices for the unsteady aerodynamic loading computation as:
{L nl
a (ik )} = q∞ [ S ]  AIC (ik ) [WT (ik )]{w (ik )} (4.36)
This extension of the downwash weighting method, based on the steady kernel matrix
replacement inside the series expansion of the kernel function will be denominated hereafter
as the successive kernel expansion method (SKEM). This procedure can be applied to any
discrete element kernel function method. However, considering the hypothesis
aforementioned in the text, i.e., its application is restricted to the computation of nonlinear
flow in sub-transonic situations. The extension of such procedure for the full transonic flow
range, that is, for M ∞ ≥ 1.0 , should be regarded as a future development to be performed.
The choice of the denomination of the proposed procedure as successive refers to its
applicability for aeroelastic stability analysis. In such situation, the method is applied in a
successive manner to correct each of the linear aerodynamic influence coefficients defined for
a reduced frequency domain set, defined for the computation of the flutter boundaries as a
function of a given parameter of interest. The investigation to be performed about the
efficiency of such method will be presented in the results chapter both for unsteady pressures
computation and flutter calculations.

5 – RESULTS AND DISCUSSION
The main concern of this Chapter is to present results from the application of
correction methods which were presented in Chapter 4. However, the first results to be
presented herein are obtained from the force matching method of Giesing et al. (1976). The
objective of this first investigation is to analyze the performance of such method, when
handling transonic aeroelastic stability analysis. The description of Giesing’s method is
presented in Appendix A.
Downwash correction methods based on steady and unsteady data (Sub-sect. 4.1.1 and
4.1.2, respectively) will be applied for aeroelastic stability analysis, in transonic flow. For
better understanding of the results with regard to these correction approaches, the unsteady
flow behavior over the prospective test case is investigated, in order to correlate unsteady
pressure distributions with the flutter boundaries calculated by such procedures.
Furthermore, the successive kernel expansion method (SKEM), which development is
presented in the Sect. 4.2 of Chapter 4, will be validated for unsteady pressure computations
and stability aeroelastic analysis. Comparisons with experimental data will be performed to
show the performance of such method with regard to the prediction of unsteady transonic
flow and flutter computation.
5.1 - Evaluation of Giesing´s Method for Transonic Flutter Computation
This section presents an evaluation of Giesing’s method (Giesing et al., 1976) with
respect to its performance in transonic aeroelastic stability analysis. This is a force matching
method, where the computation of the weighting matrix is based on strip loads. Each strip is
composed by a set of panels along the streamwise direction. As the reference loads are taken
Chapter 5 – Results and Discussion 102
for each strip, while correction factors are to be calculated for each panel, it is clear that there
are fewer available data (reference loads) than unknowns (correction factors). Therefore the
correction factors need to be computed by a minimization technique, which is described in
Appendix A, where the Giesing’s method is briefly descripted.
Giesing´s method was tested for two standard aeroelastic configurations, the AGARD
wing 445.6 weakened model #3 (Yates, 1988) and the PAPA wing (Pitch And Plunge
Apparatus, Farmer and Rivera., 1988). The first test case under investigation, is well known
as a standard aeroelastic configuration. It has an aspect ratio of 4 and a NACA 65A004 airfoil
section.
Figure 5.1 : Sketch of the AGARD 445.6 wing and the corresponding discrete element
aerodynamic model.
The AGARD 445.6 wing is modeled by ZONA 6 method, implemented in the ZAERO
software system (Zona Technology, 2003), as an isolated wing. The flow conditions which
were used for analysis belong to a subset of those presented in the work of Yates (1988),
which consists in an experimental investigation of the aeroelastic behavior of this wing under
transonic flow conditions. The model under investigation is described in Yates (1988),
regarding its structural and geometrical characteristics. The Mach numbers and air densities
for the cases considered here are presented in Table 5.1.
Table 5.1: Flow conditions for AGARD wing 445.6 (weakened # 3) aeroelastic
analysis.
Mach Reynolds Density (kg/m3) Density (slug/ft3)

0.678 1,410×106 0.2082 0,000404
0.901 0,911×106 0.0995 0,000193
0.960 0,627×106 0.0634 0,000123
The discrete element aerodynamic model, based on the ZONA 6 method, is comprised
of 220 panels, uniformly distributed along the chord length. The flutter computations were
performed using the g-method. Since the objective is to identify flutter boundaries, the use of
the k, pk or g methods is indifferent, because for that purpose, they give the same results. The
advantage of the g-method, in comparison with the others, is a better prediction of the sub-
critical aeroelastic response, since the predicted damping is more realistic.
The stripwise lifting surface loading was computed based on the integration of
nonlinear steady pressure distributions, computed by the CFL3D Navier-Stokes
computational fluid dynamic solver (Thomas et al., 1990, and Rumsey et al., 1997), The
nonlinear flow simulations were performed at two distinct angles of attack of 0.0o and 0.5o.
Quasi-steady pressure disturbances are computed and scaled by the amplitude of the motion,
turning them independent of the amplitude of the disturbances, if one considers that the flow
behaves linearly in this range. Once these loads are obtained, the procedure allows the
computation of a force matching weighting matrix which restores exactly the nonlinear
reference loading condition. However, the pressures are not restored since the force matching
is a least squares based procedure, that is, there are less unknowns, which are the correction
factors, than equations, since the weighting matrix needs to have the same dimension of the
AIC matrix. Then, this weighting matrix does not guarantee a correct pressure distribution, as
it only assures the load matching. In Figure (5.2) the complex pressure distribution, after the
multiplication of the weighting matrix from Giesing´s method, is compared with the nonlinear
pressures obtained by the Navier-Stokes solution, and with the ones obtained from ZONA 6
linear solution.
14.0 0.5
12.0 Giesing's Method 0

10.0 Navier-Stokes 0 0.2 0.4 0.6 0.8 1
-0.5
Imag (∆ Cp)
Real ( ∆ Cp)
ZONA 6 (linear)
8.0
-1
6.0
-1.5 Giesing's Method
4.0
-2
Navier-Stokes
2.0
ZONA 6 (linear)
0.0 -2.5
0 0.2 0.4 0.6 0.8 1
X/C X/C
Figure 5.2 : Comparison between the unsteady pressures obtained from the Giesing´s method,
ZONA 6 and CFL3D - AGARD 445.6 wing, weakened model #3, M∞ =0.96.
The results from the flutter solution using Giesing´s correction procedure are
presented in Figure (5.3). In the same figure are presented results for the downwash weighting
method (DWM) based on the same reference flow conditions used for Giesing’s method, i.e.
for the same quasi-steady pressure disturbances. One should observe that the flutter stability
margin is over-predicted in terms of the flutter dynamic pressure. This is an indication that
there is a failure in such procedure when handling transonic effects, which may be justified by
the anomalous shock wave dynamics behavior observed in Fig. (5.2).

140.0 24.0
130.0 AGARD 445.6 ∆α = 0.5 AGARD 445.6 ∆α = 0.5

22.0
120.0
Frequency (Hz)
Dyn. Press.
20.0
110.0
100.0 18.0
90.0 DWM
Giesing
16.0 DWM
80.0 Giesing
ZONA 6 14.0
70.0 ZONA 6
experiment
experiment
60.0 12.0
0.6 0.7 0.8 0.9 1 0.6 0.7 0.8 0.9 1
Mach Mach
Figure 5.3 : Flutter boundaries (dynamic pressure and flutter frequency) for the AGARD wing
445.6, weakened model #3.
The other test case is the flutter computation of the PAPA wing (Farmer and Rivera,
1988) at two distinct reference mean angles of attack. It has a cambered supercritical airfoil
with maximum thickness of 12% and a rectangular planform with chord of 16 inches (0.4064
m) and semispan of 32 inches (0.8128 m). The structural support provides two rigid body
modes: a plunge mode (3.43 Hz) and a pitch mode (5.44 Hz). The ZONA 6 aerodynamic
model is comprised of 200 panels, equally distributed along the chord length (Fig. 5.4).
Figure 5.4 : Sketch of the PAPA wing and the corresponding discrete element aerodynamic
model.
The flutter of this wing results from the coalescence of these two modes, depending on
the mean angle of attack of the wing, and from the undisturbed flow conditions. The flutter
boundaries, for a few Mach numbers near the transonic flow regime are presented in Figures
(5.5) and (5.6).
180.0 4.7
papa +1 deg papa +1 deg
4.65
175.0
4.6
170.0
Dyn. Press.
DWM 4.55
Freq (Hz)
Giesing
165.0 ZONA 6 4.5
experimental
4.45
160.0
DWM
4.4
Giesing
155.0
4.35 ZONA 6
experimental
150.0 4.3
0.6 0.65 0.7 0.75 0.8 0.85 0.6 0.65 0.7 0.75 0.8 0.85
Mach Mach
Figure 5.5 : Flutter boundaries (dynamic pressure and flutter frequency) for the PAPA wing at
a reference angle of attack of α = 1.0o.
230.0 4.65
DWM DWM
Giesing 4.6 Giesing
210.0 ZONA 6 ZONA 6
4.55
experimental
Dyn. Press.
experimental
Freq (Hz)
4.5
190.0
papa -2 deg 4.45
4.4
170.0
4.35 papa -2 deg
150.0 4.3
0.65 0.7 0.75 0.8 0.85 0.65 0.7 0.75 0.8 0.85
Mach Mach
at a reference angle of attack of α =-2.0o.

These results are in better agreement with the experimental ones when comparing the
dynamic pressure in the case of the PAPA wing at α=1.0o . One should note that the results of
the steady downwash weighting method and from Giesing´s method are nearly coincident for
both the dynamic pressure and flutter frequency. However the flutter frequencies still remain
underestimated, when compared to the experimental measurements. At higher Mach numbers,
Giesing’s method results are worse in the case of the PAPA wing at α=-2.0o, when compared
to the results at the same Mach number and α = 1.0o. Such behavior is expected, because
transonic effects at higher angle of attack amplitudes are stronger than at lower amplitudes,
and this method does not guarantee that shock positioning and strength be restored, thus
giving worse results for flow conditions with stronger transonic effects. Certainly this method
will fail on the prediction of transonic flutter speeds, turning it inadequate for such
computations.
5.2 –Downwash Weighting Methods
The results to be presented in this section are obtained using downwash correction
methods, based on steady and unsteady pressures as the nonlinear reference conditions. The
test case to be investigated is the AGARD wing 445.6 weakened model #3 (Yates, 1988), in
subsonic to transonic flow conditions. As a first step, the unsteady pressure distribution over
the AGARD 445.6 wing are investigated in order to understand the fluid dynamic behavior
and its relation with the results with regard to the flutter computation for this wing.
5.2.1 – Investigation of the AGARD Wing Unsteady Pressure Distributions.
The test case under investigation is the AGARD wing 445.6, weakened model # 3
standard aeroelastic configuration. It has an aspect ratio of 4.0 and a NACA 65A004 airfoil
section. This wing is modeled by the doublet lattice method, implemented in
MSC/NASTRANTM software system, as an isolated wing in different flow conditions, where
the discrete element model is comprised of 240 panels (Fig. 5.7).
Figure 5.7: DLM paneling of the AGARD 445.6 wing.
The test conditions belong to a subset of those presented in the work of Yates (1988),
which consists in an experimental investigation of the aeroelastic behavior of this wing under
transonic flow conditions. The model under investigation is described in Yates (1988),
regarding its structural and geometrical characteristics. The Mach numbers and air densities
for the cases considered here are presented in Table (5.1).

The computation of the nonlinear unsteady pressures is performed from the finite
differences solution of the Navier-Stokes equations, using the method described in Chapter
3. The computational mesh surrounding the wing is an algebraic generated “C” type topology,
with 141 points in the ξ direction, tangent to the lifting surface boundary, where 121 points
are over the lifting surface solid surface. In the η direction, normal to the wing surface, there
are 41 points between the solid surface and the limit of the computational mesh. And finally,
in the ζ direction there are 25 points aligned with the spanwise direction, where 17 points are
over the lifting surface, and the remaining are between the wing tip and the computational
domain limit. The computational mesh which surrounds the wing, as well as its profile at
spanwise station 30.8 % are presented in Fig. (5.8) and (5.9) respectively.
1.1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1.1 1.2
Figure 5.8 : AGARD wing CFD aerodynamic mesh.

The investigation of the unsteady transonic flow behavior around the AGARD wing
445.6 is here performed examining pressure distributions at spanwise station 30.8 % Fig (5.9).
The main concern of this investigation is to verify the influence of the amplitude of the
motion on pressure distributions. Thus, this station was chosen because the flow is less
subjected to three-dimensional effects as indicated by Silva et al. (2001).
Three reduced frequencies are chosen from the test conditions presented by Yates
(1988). These frequencies are presented in Table 5.2, where, k denotes the reduced frequency
k = ωb U ∞ , while kr = ω c a∞ is another way of representing the reduced frequency, used in
the Navier-Stokes code. For both linear and nonlinear computations, c= 2b = 1.0 (m).
Table 5.2: Reduced frequencies for AGARD wing 445.6 aeroelastic analysis.
Mach kr k
0,678 0.48839 0.36016
0,901 0.34076 0.18910
0,960 0.28252 0.14714
0.4
0.3
0.2
0.1
-0.1
-0.2
-0.3
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
Figure 5.9 : AGARD wing 445.6 NACA 65A004 profile, spanwise station 30.8 %.
Pressure distributions are computed from the doublet lattice method and the unsteady
Navier-Stokes simulation for the reduced frequencies and corresponding Mach numbers,
presented in Table (5.2). In the case of the nonlinear solution, they were obtained for different
amplitudes of oscillation, leading to pressure coefficient differences properly scaled by the
magnitude of the disturbance. The computations were performed for different amplitudes in
order to quantify the influence of the amplitude of the motion on the pressure ratios, and to
compare these to the ones computed from the linear model.
The computed unsteady pressure distributions along the chord at spanwise station 30,8
%, are presented in Figs. (5.10) through (5.30). Both real and imaginary parts of these
pressures are presented at a given Mach number (0.678, 0.901 and 0.960), the corresponding
reduced frequency k given in Table (5.2), and amplitude of oscillation. The chosen amplitudes
were between 0.25o to 3.0o in angle of attack rotation around the root midchord axis.
0.8 0.02
0.7 Doublet Lattice 0.01
0
0.6 Navier-Stokes
-0.01 0 0.2 0.4 0.6 0.8 1
Real (∆ Cp)
0.5
Imag (∆ Cp)
-0.02
0.4
-0.03
0.3 -0.04
0.2 -0.05 Doublet Lattice
0.1 -0.06 Navier-Stokes
0 -0.07
0 0.2 0.4 0.6 0.8 1 -0.08
-0.09
x/c x/c
Figure 5.10 : Unsteady pressure distributions for the wing station at 30.8% of the span,
M∞=0.678, ∆α = 0.25o.
1.6 0.04
1.4 0.02
Doublet Lattice 0
1.2
-0.02 0 0.2 0.4 0.6 0.8 1
Real (∆ Cp)
1 Navier-Stokes
Imag (∆ Cp)
-0.04
0.8 -0.06
0.6 -0.08
Doublet Lattice
0.4 -0.1
-0.12 Navier-Stokes
0.2
-0.14
0
-0.16
0 0.2 0.4 0.6 0.8 1
-0.18
x/c x/c
M∞=0.678, ∆α = 0.5o
3 0.1
Doublet Lattice 0.05
2.5
Navier-Stokes 0
Real (∆ Cp)
2
-0.05 0 0.2 0.4 0.6 0.8 1
Imag (∆ Cp)
1.5 -0.1
1 -0.15
-0.2
0.5 Doublet Lattice
-0.25
0
Navier-Stokes
-0.3
0 0.2 0.4 0.6 0.8 1
-0.35
x/c x/c
M∞=0.678, ∆α = 1.0o
4.5 0.2
4 Doublet Lattice
0.1
3.5 Navier-Stokes
0
Real (∆ Cp)
3
0 0.2 0.4 0.6 0.8 1
Imag (∆ Cp)
2.5 -0.1
2
-0.2
1.5
1 -0.3
0.5 -0.4
0 Doublet Lattice
-0.5
0 0.2 0.4 0.6 0.8 1 Navier-Stokes
-0.6
x/c x/c
M∞=0.678, ∆α = 1.5o.
6 0.2
5 0.1
Doublet Lattice
0
Real (∆ Cp)
4
Navier-Stokes 0 0.2 0.4 0.6 0.8 1
Imag (∆ Cp)
-0.1
3
-0.2
2
-0.3 Doublet Lattice
1
-0.4 Navier-Stokes
0 -0.5
0 0.2 0.4 0.6 0.8 1
-0.6
x/c x/c
M∞=0.678, ∆α = 2.0o
8 0.2
7 Doublet Lattice 0.1
6 Navier-Stokes 0
Real (∆ Cp)
5 -0.1 0 0.2 0.4 0.6 0.8 1
Imag(∆ Cp)
4 -0.2
-0.3
3
2
-0.5 Navier-Stokes
1
-0.6
0
-0.7
0 0.2 0.4 0.6 0.8 1
-0.8
x/c x/c
M∞=0.678, ∆α = 2.5o
9 0.4
8
Doublet Lattice 0.2
7
Real (∆ Cp)
6 Navier-Stokes 0
Imag (∆ Cp)
5 0 0.2 0.4 0.6 0.8 1

-0.2
4
3 -0.4
2 Doublet Lattice
1 -0.6
Navier-Stokes
0 -0.8
0 0.2 0.4 0.6 0.8 1
-1
x/c x/c
M∞=0.678, ∆α = 3.0o
0.8 0.02
0.7 Doublet Lattice
0
0.6 Navier-Stokes 0 0.2 0.4 0.6 0.8 1
Real (∆ Cp)
-0.02
0.5
Imag (∆ Cp)
0.4 -0.04
0.3
-0.06
0.2
-0.08
0.1
Doublet Lattice
0 -0.1
0 0.2 0.4 0.6 0.8 1 Navier-Stokes
-0.12
x/c x/c
M∞=0.901, ∆α = 0.25o
1.6 0.05
1.4 Doublet Lattice
0
1.2 Navier-Stokes
0 0.2 0.4 0.6 0.8 1
Real (∆ Cp)
1 -0.05
Imag (∆ Cp)
0.8
-0.1
0.6
0.4 -0.15
0.2
0
0 0.2 0.4 0.6 0.8 1 Navier-Stokes
-0.25
x/c x/c
M∞=0.901, ∆α = 0.5o
3 0.1
0.05
2.5
Doublet Lattice 0
-0.05 0 0.2 0.4 0.6 0.8 1
Real (∆ Cp)
2 Navier-Stokes
Imag (∆ Cp)
-0.1
1.5 -0.15
1 -0.2
Doublet Lattice
-0.25
0.5
Navier-Stokes
-0.3
-0.35
0
-0.4
0 0.2 0.4 0.6 0.8 1
-0.45
x/c x/c
M∞=0.901, ∆α = 1.0o
4.5 0.2
4 0.1
3.5 0
Doublet Lattice
Real (∆ Cp)
3
Navier-Stokes -0.1 0 0.2 0.4 0.6 0.8 1
Imag (∆ Cp)
2.5
-0.2
2
-0.3
1.5
1 -0.4 Doublet Lattice
0 -0.6
0 0.2 0.4 0.6 0.8 1 -0.7
x/c x/c
M∞= 0.901, ∆α = 1.5o

6 0.2
5
Doublet Lattice
0
Navier-Stokes 0 0.2 0.4 0.6 0.8 1
Real (∆ Cp)
4
-0.2
Imag (∆ Cp)
3
-0.4
2
1 Navier-Stokes
-0.8
0
0 0.2 0.4 0.6 0.8 1 -1
x/c x/c
M∞= 0.901, ∆α = 2.0o
8 0.2
7 Doublet Lattice 0
6 Navier-Stokes 0 0.2 0.4 0.6 0.8 1
Real (∆ Cp)
-0.2
5
Imag(∆ Cp)
4 -0.4
3
-0.6
2 Doublet Lattice
1 -0.8 Navier-Stokes
0 -1
0 0.2 0.4 0.6 0.8 1
-1.2
x/c x/c
M∞=0.901, ∆α = 2.5o
9 0.4
8 Doublet Lattice 0.2
7 0
Navier-Stokes
Real (∆ Cp)
6
-0.2 0 0.2 0.4 0.6 0.8 1
Imag (∆ Cp)
5
-0.4
4
3 -0.6

1 -1 Navier-Stokes
0 -1.2
0 0.2 0.4 0.6 0.8 1
-1.4
x/c x/c
M∞=0.901, ∆α = 3.0o
0.8 0.02
0.7 Doublet Lattice
0
0.6 Navier-Stokes 0 0.2 0.4 0.6 0.8 1
Real (∆ Cp)
0.5 -0.02
Imag (∆ Cp)
0.4
-0.04
0.3
Doublet Lattice
0.2 -0.06
Navier-Stokes
0.1
-0.08
0
0 0.2 0.4 0.6 0.8 1 -0.1
x/c x/c
M∞= 0.96, ∆α = 0.25o

1.6 0.05
1.4 Doublet Lattice
0
1.2 Navier-Stokes
0 0.2 0.4 0.6 0.8 1
Real (∆ Cp)
Imag (∆ Cp)
-0.05
0.8
0.6
-0.1
0.4
Doublet Lattice
0
0 0.2 0.4 0.6 0.8 1 -0.2
x/c x/c
M∞=0.96, ∆α = 0.5o
3 0.1
Doublet Lattice 0.05

2.5
0
Navier-Stokes
Real (∆ Cp)
2 -0.05 0 0.2 0.4 0.6 0.8 1

Imag (∆ Cp)
-0.1
1.5
-0.15
1 -0.2
0.5 -0.25
0 -0.35
0 0.2 0.4 0.6 0.8 1
Navier-Stokes
-0.4
x/c x/c
0.96, ∆α = 0.25o
4.5 0.1
4
Doublet Lattice 0
3.5
0 0.2 0.4 0.6 0.8 1
Navier-Stokes -0.1
Real (∆ Cp)
Imag (∆ Cp)
2.5 -0.2
2
1.5 -0.3

0.5 Navier-Stokes
-0.5
0
0 0.2 0.4 0.6 0.8 1
-0.6
x/c x/c
M∞= 0.96, ∆α = 1.5o
6 0.2
0.1
5 Doublet Lattice
0
Navier-Stokes
Real (∆ Cp)
4 -0.1 0 0.2 0.4 0.6 0.8 1

Imag (∆ Cp)
-0.2
3
-0.3
2 -0.4

-0.6 Navier-Stokes
0 -0.7
0 0.2 0.4 0.6 0.8 1
-0.8
x/c x/c
M∞= 0.96, ∆α = 2.0o

8 0.2
7 Doublet Lattice 0
6 Navier-Stokes 0 0.2 0.4 0.6 0.8 1
Real (∆ Cp)
5 -0.2
Imag (∆ Cp)
4
-0.4
3
1 Navier-Stokes
-0.8
0
0 0.2 0.4 0.6 0.8 1
-1
x/c x/c
M∞= 0.96, ∆α = 2.5o
9 0.2
8
0
7 Doublet Lattice
0 0.2 0.4 0.6 0.8 1
Navier-Stokes -0.2
Real (∆ Cp)
6
Imag (∆ Cp)
5 -0.4
4
3
2 -0.8 Navier-Stokes
1
-1
0
0 0.2 0.4 0.6 0.8 1 -1.2
x/c x/c
M∞= 0.96, ∆α = 3.0o.

The results presented in Figs. (5.10) through (5.16) are the complex pressure
distributions for a subsonic (M∞ = 0.678) flow condition. One should note that there is a good
agreement between the results from the Navier-Stokes and the doublet lattice computations,
up to ∆α = 1.5o. For these low amplitudes of motion, there is a distinct leading edge suction
peak. However, examining the results presented in Figs (5.14) through (5.16), it is possible to
note that there is a deviation near the leading edge, for both real and imaginary parts of the
pressures. In order to further investigate this behavior, a contour plot of the pressures at the
maximum angle α = 2.0o is shown in Fig (5.31), while instantaneous pressures at several
angle of attack are plotted in Fig. (5.32) for ∆α = 2.0o. Similar plots for the motion amplitude
∆α= 3.0o and maximum angle α = 3.0o are shown in Figs. (5.34) and (5.33).
Figure 5.31 : Contour plot of instantaneous pressures at spanwise station 30,8%, M∞=0.678,
α = 2.0o.
1.80
α =2.0
1.60 α =1.8
α =1.6
1.40
α =1.4
1.20 α =1.2
1.00
α =1.0
α =0.8
Cp
0.80
α =0.6
0.60 α =0.4
α =0.2
0.40
0.20
0.00
0.0 0.2 0.4 0.6 0.8 1.0
-0.20 x/c
Figure 5.32 : Instantaneous pressure distributions at spanwise station 30,8%, M∞=0.678,
∆α = 2.0o.
Figure 5.33: Contour plot of instantaneous pressures at spanwise station 30,8%, M∞=0.678,
α = 3.0o.
1.80
α =3.0
1.60 α =2.7
α =2.4
1.40
α =2.1
1.20 α =1.8
1.00
α =1.5
α =1.2
Cp
0.80
α =0.9
0.60 α =0.6
α =0.3
0.40
0.20
0.00
0.0 0.2 0.4 0.6 0.8 1.0
-0.20 x/c
Figure 5.34: Instantaneous pressure fields and distributions at spanwise station 30,8%,
M∞=0.678, ∆α = 3.0o.
For both angles of attack, one should observe the appearance of a secondary suction
peak which moves towards the trailing edge with the increase in motion amplitude. The
displacement of this secondary pressure suction peak is associated to the aerodynamic lag of
the flow around the finite thickness profile, computed by the Navier-Stokes solution. The
surface pressure distribution plots in Figs (5.32) and (5.34), indicate that the pressure
distribution is quite influenced by the motion amplitude. For example, for the same
instantaneous angle of attack of 1.8o, it is possible to note the dependency of the aerodynamic
lag on the motion amplitude. For ∆α = 2.00 (Fig. 5.32), the position of the secondary pressure
peak is around x c ≅ 0.1 , while for ∆α = 3.00 (Fig. 5.34), the position of the secondary
pressure peak is x c ≅ 0.2 . Since these secondary pressure peaks are moving with time, and
its displacements increase with the amplitude of the motion, then the deviation from the linear
complex pressure distributions are larger for the higher angle of attack amplitudes (Figs 5.14
through 5.16).
The downstream displacement of the secondary pressure peak is more pronounced in
the subsonic Mach number because, under this condition, the aerodynamic lag effects are
more relevant. However, examining instantaneous pressure plots (Fig 5.36 and 5.38) and the
pressure contours at the maximum angle of attack (Fig 5.36 and 5.38) for M∞=0.901, it is
possible to note a behavior similar to the subsonic case,i.e., M∞=0.678, but with less intensity,
as expected.
α = 2.0o.
1.80
α =2.0
1.60 α =1.8
α =1.6
1.40
α =1.4
1.20 α =1.2
1.00
α =1.0
α =0.8
Cp
0.80
α =0.6
0.60 α =0.4
α =0.2
0.40
0.20
0.00
0.0 0.2 0.4 0.6 0.8 1.0
-0.20 x/c
Figure 5.36: Instantaneous pressure distributions at spanwise station 30,8%, M∞=0.901,
∆α = 2.0o.
α = 3.0o.
1.80
α =3.0
1.60 α =2.7
α =2.4
1.40
α =2.1
1.20 α =1.8
1.00
α =1.5
α =1.2
Cp
0.80
α =0.9
0.60 α =0.6
α =0.3
0.40
0.20
0.00
0.0 0.2 0.4 0.6 0.8 1.0
-0.20 x/c
∆α = 3.0o.
The differences that have been observed between the doublet lattice method and the
Navier-Stokes results at M∞=0.96 (Figs. 5.24 through 5.30) are associated to a shock wave
formation. Contour plots of the instantaneous pressures at maximum angle of attack situations
are presented in Figs. (5.39) and (5.41) for this Mach number. In latter situation, there is a
slight trend for appearance of a secondary peak (Figs. 5.42), because the aerodynamic lag is
suppressed by the high undisturbed flow speed energy.

α = 2.0o.
1.60
α =2.0
1.40 α =1.8
α =1.6
1.20
α =1.4
1.00
α =1.2
α =1.0
0.80 α =0.8
Cp
α =0.6
0.60
α =0.4
0.40 α =0.2
0.20
0.00
0.0 0.2 0.4 0.6 0.8 1.0
-0.20 x/c
∆α = 2.0o.
α = 3.0o.
1.80
α =3.0
1.60 α =2.7
α =2.4
1.40
α =2.1
1.20 α =1.8
1.00
α =1.5
α =1.2
Cp
0.80
α =0.9
0.60 α =0.6
α =0.3
0.40
0.20
0.00
0.0 0.2 0.4 0.6 0.8 1.0
-0.20 x/c
∆α = 3.0o.
One should observe in Figs (5.40 and 5.42) that there is a displacement of the shock
wave towards the trailing edge, from x c ≅ 0.50 to x c ≅ 0.57 for the angles of attack
α = 2.0o and α = 3.0o respectively. The transonic effects are more pronounced for this Mach
number, and are especially evident in the imaginary part of the pressures, leading to strong
differences between the linear and nonlinear predicted phases. As the pressure phases are
more sensitive to the shock movement than the pressure amplitudes, the main differences
when comparing the potential and the Navier-Stokes results will be in the imaginary part of
the pressure. At M∞=0.901, the transonic effects are weaker than the aforementioned
situation, as one should observe in Figs. (5.17) through (5.23). This fact can be explained
examining the airfoil profile maximum thickness. The AGARD wing has a 65A004 airfoil
with a maximum thickness ratio (t/c) equal to 4.0 %. Hence, the shock wave formation will
only occur at higher Mach numbers.
Another feature to be noted is the influence of the amplitude of the motion on the
shock wave position. In Figs (5.27) through (5.30), one may notice that the shock appears to
move aft as the amplitude is increased. The reason for this behavior is that, in steady flow
conditions, the shock wave position moves towards the trailing edge as the angle of attack is
increased, as shown by Silva et al., (2002 A). This behavior is also noted for unsteady flows,
as indicated by Dowell et al. (1983). Recalling that the pressures are scaled by the amplitude
of the motion, one should note that, for smaller angles of attack, the shock position is about
the same. This is so because int the low angle of attack range the transonic flow behaves
linearly, with respect to angle of attack, as concluded in Chapter 3.
Another aspect to be highlighted is that the formation of the moving secondary suction
peak is apparently not associated to flow separations. This is illustrated by the instantaneous
velocity vector plots presented in Figs (5.43) for Mach numbers 0.678 to 0.960 and maximum
angles α = 2.00 and α = 3.00 .

M∞ = 0.678, α =2.0o M∞ = 0.678, α =3.0o
M∞ = 0.901, α =2.0o M∞ = 0.901, α =3.0o
M∞ = 0.960, α =2.0o M∞ = 0.960, α =3.0o
Figure 5.43: Instantaneous velocity vector fields over the leading edge of the AGARD 445.6
wing.
Examining the velocity vector plots in Fig (5.43), one should observe that for
∆α = 2.00 and ∆α = 3.00 , there is no evidence of flow separation. However, the mesh around
the leading edge is not sufficiently refined to capture mild separation. Neither it was
guaranteed that the maximum displacement positions are the ones where the dynamic angle of
attack derivative is null, because the time domain Navier-Stokes simulation is performed in
discrete time steps. On the other hand, since the identified unsteady flow behavior regards
aerodynamic lag associated to the wing motion, the aforementioned modeling restrictions do
no impact on the conclusions of the present investigation.
The behavior of the phase angles of the scaled complex pressure difference along the
chord for M∞=0.96 is summarized in Fig (5.44), where Θ is the non-dimensional phase angle
resulting from the scaling of the phase angle θ and the amplitude ∆α , as Θ = θ ∆α .
50
40 0.25 deg
0.5 deg
30 1.0 deg
1.5 deg
2.0 deg
Θ (deg)
20
2.5 deg
10 3.0 deg
0
0 0.2 0.4 0.6 0.8 1
-10
-20
x/c
Figure 5.44: Non-dimensional phase angles of the pressure difference distributions for
spanwise station 30.8%, M∞=0.96.

As may be observed in Fig. (5.44), for this specific case (i.e. M∞=0.96, 30,8%
spanwise station) pressure phases remain around –8o up to x c ≅ 0.55 . Aft of that location,
the pressures phase advances as the trailing edge is approached. The corresponding pressure
difference amplitudes, scaled by the amplitude of the motion are summarized in Fig. (5.45). It
may be noticed that the scaled pressure difference amplitude changes only slightly as the
angle of attack amplitude is increased.
2.5
0.25 deg
0.5 deg
2 1.0 deg
1.5 deg
|∆Cp|/Da
1.5 2.0 deg

2.5 deg
3.0 deg
1
0.5
0
0 0.2 0.4 0.6 0.8 1
x/c
Figure 5.45. Amplitudes of the pressure difference distributions for spanwise station 30.8%,
M∞=0.96.
The pressure distribution behavior summarized in the results of this section suggest
that the amplitude of the motion plays an important role in the computation of the complex
unsteady pressure distributions. The amplitude of the pressures gives the lift, while the
pressure phases may be understood as a form to quantify the lag between the center of
pressure displacement and the lifting surface motion. Since the center of pressure position is
governed by the shock displacement, as indicated by Ashley (1980) and Tijdeman (1977), the
moment coefficient derivatives will be more sensitive to the nonlinearities of the flow over
the lifting surface. Therefore, from these observations it is suggested that the influence of the
nonlinearities will promote significant changes in the computation of the aeroelastic stability,
since flutter depends on the relative position of the center of pressure with respect to the
elastic axis position. In the next sub-section an investigation of the aeroelastic stability
behavior will be performed to evaluate the influence of the amplitude of the angle of attack on
the computation of flutter speeds based on nonlinear unsteady pressures.
5.2.2 AGARD 445.6 Wing Aeroelastic Stability Analysis
The computation of the aeroelastic stability of the AGARD wing is performed in order
to evaluate the efficiency of the downwash correction method, using either nonlinear steady
or unsteady pressure distribution as reference conditions. The unsteady aerodynamic
modeling is based on the doublet lattice method, implemented in the MSC/NASTRANTM
software system. The chosen flutter solution technique is the k-method, which is
mathematically consistent for the computation of the flutter boundary. The weighting
operators are computed to correct the pressure to downwash relation, resulting from the
modeling of the AGARD wing using the doublet lattice method.
The nonlinear unsteady pressures were obtained under quasi-steady motion and
unsteady motions to provide the corresponding reference conditions for the steady and
unsteady downwash weighting methods, respectively. In the latter case, the pressures were
computed under harmonic motion oscillations of the wing with amplitude ∆α = 2.00 , where
the nonlinear contribution due to unsteady transonic effects are more relevant. Otherwise, in
the former case, the chosen amplitude of the quasi steady motion is ∆α = 0.50 because the
nonlinear contribution comes from the steady mean transonic flow. Indeed, this amplitude is
sufficient because it will be computed quasi-steady pressure rates, which would be
independent of the amplitude of the motion. Furthermore, it was identified by Silva et al.
(2002 A) that for small angles of attack (up to 2.0o ) the flow behaves linearly with the
variation in angle of attack with regard to the shock wave displacement in steady flow
conditions. However, for larger angles of attack it is possible to have shock induced
separations, which are not desirable for the computation of the quasi-steady reference
pressure, because this phenomenon is out of the scope of the present investigation. The main
concern is to have a nonlinear steady mean flow distribution with regard to the shock
positioning and strength.
The weighting operators are introduced in the aeroelastic analysis as correction
factors, yielding the computed flutter speeds shown in Tables (5.3) and (5.4). Table (5.3)
presents a comparison between the transonic flutter computation based on the correction using
steady and unsteady nonlinear pressures and experimental results (Yates, 1988). Table (5.4)
includes comparisons with some well known aeroelastic analysis codes (Chen et al, 2000)
with unsteady downwash correction procedure results.
Table 5.3: Flutter speeds and frequencies for AGARD wing 445.6.
Experimental Linear Steady Correction Unsteady Correction

M∞ VF [m/s] ωF [Hz] VF [m/s] ωF [Hz] VF [m/s] ωF [Hz] VF [m/s] ωF [Hz]
0.678 231,37 17,98 239,89 20,18 213,82 21,25 239,30 23,76
0.901 296,69 16,09 299,30 16,38 275,65 17,35 284,76 17,12
0.960 309,01 13,89 329,18 14,57 315,97 15,65 298,63 15,41
Table 5.4: Flutter speeds and frequencies for AGARD wing 445.6.
Unsteady Correction CAPTSD (nonlinear) ZTAIC

M∞ (Bennet et al.,1989) (Chen et al., 2000)
VF [m/s] ωF [Hz] VF [m/s] ωF [Hz] VF [m/s] ωF [Hz]
0.678 239,30 23,76 234,09 19,2 231,95 19,30
0.901 284,76 17,12 290,17 15,8 294,19 16,38
0.95 N/A N/A 291,39 12,8 287,91 13,46
0.96 298,63 15,41 N/A N/A N/A N/A
The ZTAIC method is a modal aerodynamic influence coefficiet matrix correction
method, based on the transonic equivalent strip method developed by Liu et al. (1983) and
further extended by (Chen et al.,2000). The CAP-TSD method is a time domain finite
difference solution of the transonic small disturbance equation (Bennet et al.,1989) coupled
with a finite element structural dynamic model. The results shown in Tables (5.3) and (5.4)
are also represented in graphical form in Figs. (5.46) and (5.47), respectively, in terms of the
flutter speed index (FSI) as a function of the Mach number.
0.5 24 Experimental
0.48 Linear
0.46 22
Flutter Frequency (Hz)
Unsteady
0.44 C ti
20
0.42
FSI
0.4 18
0.38
Unst. Correction 16
0.36
Experimental
0.34
Linear 14
0.32 Steady correction
0.3 12
0.6 0.7 0.8 0.9 1 0.6 0.7 0.8 0.9 1
Mach Mach
between steady and unsteady downwash weighting methods.

0.5 24 Experimental
0.48 Unsteady

0.46 22 Correction
ZTAIC
0.44
20
0.42
FSI
0.4 18
0.38 Unst. Correction
Experimental 16
0.36
0.34 ZTAIC
14
CAP-TSD
0.32
0.3 12
0.6 0.7 0.8 0.9 1 0.6 0.7 0.8 0.9 1
Mach Mach
between the unsteady downwash weighting methods and other methods.
The flutter speed index is a nondimensional flutter speed defined as
FSI = V f ( b ωα µ ) ,
s where V f is the dimensional flutter speed, bs is the root semi-chord
length, and ωα is a reference frequency. The term inside the square root is the mass ratio
defined as µ = m ρϑ , where m is the mass of the wing, ρ is the flow density if the test
medium, ϑ is the volume of a conical frustum having streamwise root chord as a lower base
diameter, streamwise tip as the upper base diameter, and the wing span as height. The flutter
speed index computation is performed for each Mach number, depending on the freestream
flow properties and geometrical properties of the wing, as FSI M ∞ =0.678 = V f 1662.40 ,
FSI M ∞ =0.901 = V f 2405.18 , and FSI M ∞ =0.96 = V f 3012.81 .
One can observe in Fig. (5.46) that the flutter speed indexes computed from the
correction based on unsteady pressures indicates the presence of the transonic dip. This
phenomenon is characterized by a decrease of the slope of the flutter speed plot as a function
of the Mach number, when comparing with the linearly predicted one. In the same Figure are
also shown results from the correction based on steady pressures. One may notice that there is
a good agreement when comparing the dip slope between the unsteady pressures based
correction procedure and the experimental data. The steady pressures based correction
underestimates some of the flutter speeds, as well as the dip slope. The reason for these
discrepancies is related to the absence of a nonlinear unsteady pressures contribution, because
the reference pressures, over which the correction factors are computed, are from steady
nature.
One feature to be noted in the same results (Fig. 5.46) is that in the subsonic Mach
number case (M∞=0.678), the correction based on unsteady pressures did not change
significantly the linear prediction. This fact can be understood by resorting to Figs. (5.14),
(5.21) and (5.28), where one should observe the differences in the pressure phases from the
linear and nonlinear calculations, for each Mach number and for the same amplitude of the
motion ∆α = 2.00 . Considering those differences between the linear and nonlinear pressure
distributions, the resulting correction factors for the transonic flow conditions will properly
introduce the necessary changes in the downwash vector to take into account the nonlinear
behavior. However, at the subsonic flow condition the differences between the amplitude of
the linear and the nonlinear computed pressures are very small and not sufficient to introduce
changes in the computation of the corrected flutter speed. The same behavior is observed in
the frequency plots shown in Fig. (5.46). Note that at M∞=0.678 the flutter frequency is not
subjected to significant changes, when comparing it with the uncorrected results. On the other
hand, in the case of the computed flutter frequencies at M∞=0.901, and M∞=0.96, the
frequencies are subjected to important changes resulting in corrected values which are in
better agreement with the experimental ones.
The results presented by Chen et al. (2000), as indicted in Fig. (5.47), show that the
transonic dip phenomenon is well characterized in the solution with the ZTAIC method, and
the computed flutter speed also presents good agreement with the experimental data. The
ZTAIC procedure is based in a more comprehensive theory of AIC matrix correction. This
method employs two-dimensional finite differences solutions for the computation of three-
dimensional nonlinear unsteady pressures distributions, for a set of different downwash
modes, by the use of the transonic equivalent strip method (Liu et al., 1983). If one considers
that the downwash correction method is based on unsteady pressures computed for a single
pitch mode, it is clear that one should expect the ZTAIC method to yield a better correlation
with the experimental data. In Fig. (5.47), the CAP-TSD code calculations (Bennet et al.,
1989) also yields good results, since its formulation is based on a finite difference nonlinear
solution of the three dimensional form of the transonic small disturbance equations.
Furthermore, the ZTAIC and CAP-TSD methods represent adequately the severity in the
flutter dip phenomenon, which in this case is a desirable feature in transonic flutter
prediction. A disadvantage regarding such procedures is the dependency on two and three-
dimensional unsteady finite difference solutions of the nonlinear equations, respectively,
increasing the computational cost in comparison with the downwash correction method.
The next step is to perform a sensitivity analysis with regard to the variation of the
dynamic amplitude input, which is used to generate the nonlinear unsteady pressure
distribution, taken as reference conditions for the computation of the correction factors. The
objective is to understand the sensitivity of the computed aeroelastic system stability margins
with respect to the amplitude of the motion, which may be related to the linear/nonlinear
behavior. Tables (5.5) and (5.6) present the computed flutter speeds, based on the unsteady
downwash correction method, using the resulting pressures with respect to a set of
displacement amplitudes.
Table 5.5: Flutter speeds and frequencies for AGARD wing 445.6. until ∆α=1.5o.
∆α=0.25o ∆α=0.5o ∆α=1.0o ∆α=1.5o

M∞ VF [ft/s] ωF [Hz] VF [ft/s] ωF [Hz] VF [ft/s] ωF [Hz] VF [ft/s] ωF [Hz]
0.678 776.86 23.49 777.62 23.52 772.97 23.39 774.67 23.44
0.901 954.32 17.26 945.92 17.96 934.21 16.71 929.16 16.93
0.960 1046.1 15.36 1039.9 15.19 1020.5 15.47 999.35 15.26
Table 5.6: Flutter speeds and frequencies for AGARD wing 445.6, until ∆α=3.0o.
uncorrected ∆α=2.0o ∆α=2.5o ∆α=3.0o experimental

M∞ VF [ft/s] ωF [Hz] VF [ft/s] ωF [Hz] VF [ft/s] ωF [Hz] VF [ft/s] ωF [Hz] VF [ft/s] ωF [Hz]
0.678 787.03 17.54 785.09 23.76 818.58 20.97 861.69 28.93 759.1 17.98
0.901 981.95 15.28 934.24 17.12 983.37 17.01 1086.22 18.03 973.4 16.09
0.960 1080.0 14.35 979.72 15.41 1021.1 15.77 1167.15 17.17 1013.8 13.89
Tables (5.5) and (5.6) include comparisons with the results computed from the
uncorrected aeroelastic model and experimental measurements by Yates (1988). One should
observe that the flutter speeds present significant variation with the nature of the unsteady
pressure data used to compute the correction factors. These results are graphically represented
in Figs. (5.48), as the variation of the non-dimensional flutter speed (FSI) with the freestream
Mach number.
0.5 0.55
0.5
0.45
0.45
FSI
0.4
FSI
Experimental ∆α = 1.5
0.4 ∆α = 2.0
Linear
∆α = 0.25 ∆α = 3.0
0.35
0.35 Experimental
∆α = 0.5
∆α = 2.5
∆α = 1.0
Linear
0.3
0.3
0.6 0.7 0.8 0.9 1
Mach 0.6 0.7 0.8 0.9 1
Mach
Figure 5.48: Comparison between the flutter speed plots as a function of the amplitude of the
motion.
In Fig. (5.48), it is possible to notice that the behavior of the non-dimensional flutter
speed with the Mach number presents a certain proportionality in the slopes of the curves up
to ∆α=1.0o. As the motion amplitude approaches to this value, the predicted flutter speed for
M∞=0.96 approaches the corresponding experimental value. However, the transonic dip,
which is characterized as the slope of the non-dimensional flutter speed curve as a function of
the Mach number, is less pronounced than the experimental one. Also in Fig (5.48), another
set of computed flutter speeds is presented for greater amplitudes of the dynamic angle of
attack. The best results in approaching the transonic dip slope is when one considers a
dynamic angle amplitude of ∆α=2.0o, at the same time as the flutter speeds are slightly
underestimated. Above this value, an interesting result should be noted. The computed flutter
speeds at 2.5o are nearly coincident with the experimental ones in the transonic Mach number
range. Otherwise, it is possible to note an increase of the flutter speeds computed at 3.0o. In
Figs. (5.49) and (5.50) the behavior of the aeroelastic stability of tha AGARD 445.6 wing is
summarized graphically representing the flutter speeds and frequencies as functions of the
amplitude of the motion.
1200
Mach=0.678
Mach=0.901
1100
Mach=0.96
1000
Vf (ft/s)
900
800
700
0.00 0.25 0.50 0.75 1.00 1.25 1.50 1.75 2.00 2.25 2.50 2.75 3.00
∆α
Figure 5.49: Flutter speed behavior as a function of the amplitude of the motion.
35.0
Mach=0.678
30.0 Mach=0.901
Mach=0.96
25.0
Frequency (Hz)
20.0
15.0
10.0
5.0
0.0
0.00 0.25 0.50 0.75 1.00 1.25 1.50 1.75 2.00 2.25 2.50 2.75 3.00
∆α
Figure 5.50: Flutter frequency behavior as a function of the amplitude of the motion.
It should be noted that up to ∆α= 2.0o, the flutter speed decreases, and also the
transonic dip is more evident, reinforcing the fact that the transonic dip phenomenon is mainly
governed by the nonlinear unsteady flow conditions. However, as soon as the angle of attack
increases beyond 2.0o, it should be noted that the flutter speed increases with the amplitude of
the motion. On possible reason for this behavior may be related to the differences in the
complex pressures which may be observed in the pressure plots for the amplitudes 2.5o and
3.0o, given in Figs (5.15), (5.16), (5.22), (5.23), (5.29), and (5.30). Looking at these plots, one
may notice that there are significant differences between the linear and the nonlinear
pressures, with regard to the prediction of the leading edge suction peak. These differences
are resulting from the increase aerodynamic lag associated to the displacement of a secondary
pressure peak present due to the thickness of the profile, as it was identified in Figs (5.31)
through (5.42). Thus, the center of pressure position is changed in such way that it modifies
the moment on the wing. Consequently, its flutter characteristics will be a function of the
amplitudes of the motion, as one should observe in Fig (5.49) and (5.50), confirming that the
aerodynamic lag effect is playing and important role in the stability of the aeroelastic system.
Furthermore, it is possible to note that, up to the amplitude of 2.0o, the transonic dip is more
evident as the angle of attack increases, indicating the nonlinearity of such phenomenon, with
regard to the aerodynamic lag due to the thickness effect.
The present investigation indicates that computed flutter speeds using unsteady
downwash correction method depend on amplitudes of the motion.. In the linear/non-linear
investigations presented in Chapter 3, the transonic flow linear behavior limit could be
established for disturbances in angle of attack bellow 0.35o, assuming the moment coefficient
criterion. One should recall that moments play an important role in the flutter phenomenon.
Therefore, the variation of the flutter speeds and frequencies as a function of amplitudes of
the motion results from the nonlinear flow behavior, regarding thickness effects.
5.3 - Evaluation of the Successive Kernel Expansion Method (SKEM)
5.3.1 – Preliminary Considerations
The scope of this section is to present the results regarding application of the extension
of the downwash weighting method, which had its development was presented in Chapter 4.
This procedure is based on a nonlinear steady kernel matrix replacement inside the series
expansion of the kernel function. The procedure was denominated as the successive kernel
expansion method (SKEM). The discrete element kernel function method to be corrected is
presented in the work of Chen et al. (1993) and in Zona Technology (2003), named as ZONA
6 method, implemented in the ZAERO software system. This code was selected because it
allows comprehensive modifications in terms of programming, with regard to the
implementation of the correction procedure to be here investigated.

The validation of the successive kernel expansion method consists basically in three
test cases regarding unsteady pressure calculation, and three flutter cases. A broad variety of
wings and profiles were considered for the present studies, ranging from thin profiles, as the
F-5 wing and the AGARD 445.6 wing, to thick supercritical profiles, like the LANN, YXX
and PAPA wings, and also parabolic arc profiles in the Lessing wing case. The wing
planforms were also different for several cases, unswept (PAPA wing, Lessing wing) or swept
(AGARD 445.6, YXX, LANN wings), and with low aspect ratio (F-5, PAPA and Lessing
wings) or high aspect ratio (LANN and YXX wings). Complex configurations, as wing store
configurations and complete aircraft models were not considered in the present investigation,
since the purpose is to evaluate, as a first step, the efficiency of this new correction method,
based on wing alone standard aeroelastic configurations.
5.3.2 - Validation of Unsteady Pressure Computations
The validation to be here performed consists in the computation of unsteady pressure
distributions using the successive kernel expansion method in order to confront it with
experimental data. The nonlinear steady pressures distributions were computed by the CFL3D
Navier-Stokes computational fluid dynamic solver (Thomas et al., 1990, and Rumsey et al.,
1997), whose results for each test configuration were taken from the work of Chen et al.
(2003). The test cases under investigation are:
A) - F-5 wing pitching about 50% root chord at M∞ = 0.9 and k = 0.275;
B) - F-5 wing pitching about 50% root chord at M∞ = 0.948 and k = 0.264;
C) - LANN wing in pitch mode about 62% root chord at M∞ = 0.822 and k = 0.105;
D) - Lessing wing in first bending mode at M∞ = 0.9 and k = 0.13.

A) F-5 Wing Pitching About 50% Root Chord at M∞ = 0.9 and k = 0.275 :
Figure 5.51 - Sketch of the F-5 wing and the corresponding discrete element aerodynamic
model mesh.
The ZONA 6 aerodynamic model is comprised of 200 panels, uniformly distributed
along the chord length, over the wing surface (Fig. 5.51).The reference conditions for the
computation of the nonlinear steady kernel are based on steady state nonlinear pressure
distributions, which were computed using the CFL3D Navier-Stokes solution. At M∞ = 0.9
and α = 0° the pressure coefficients differences correlates very well with the test data
(Tijdeman et al., 1978), as one should note in Fig (5.52). Quasi-steady pressure disturbances
ratios are computed in order to obtain the downwash weighting operator, following the steady
downwash correction procedure, presented in Chapter 4. Therefore, the ratios δ∆C pnl need to
be computed by ( ∆C p (α = 0.5 ) − ∆C p (α = 0.0 ) ) ( ∆α = 0.5 ) ,

nl o nl o o
leading to the nonlinear
kernel function matrix  D nl (ik = 0)  , obtained from the post-multiplication of the AIC matrix
by the weighting operator. For the present test case investigation, the represented quasi-steady
motion is a rigid body pitch displacement around the 50% chord location.
-0.8 -0.8
y/2b=51.5% y/2b=87.5%
-0.4 -0.4
Steady Cp
Steady Cp
0 0
0.4 Experiment (Upper) 0.4 Experiment (Upper)

Experiment (Lower) Experiment (Lower)
0.8 CFL3D (Upper) CFL3D (Upper)
0.8
CFL3D (Lower) CFL3D (Lower)
0 0.2 0.4 0.6 0.8 1
0 0.2 0.4 0.6 0.8 1
X/C
X/C
-0.8
y/2b=97.7%
-0.4
Steady Cp
0.4 Experiment (Upper)

Experiment (Lower)
0.8
CFL3D (Upper)
CFL3D (Lower)
0 0.2 0.4 0.6 0.8 1
X/C
Figure 5.52 : Steady pressure distribution for the F-5 wing at M∞ = 0.9 and α = 0°.
The quasi steady pressure ratios are represented in Fig. (5.53), where, it is possible to
observe that the pressure jump with respect to the shock wave is preserved. The subsequent
step is to compute the unsteady pressures for a known reduced frequency, using the
successive kernel expansion procedure.

Y/2b=51.5 % Y/2b=87.5 %
40 25
35
20
∆ Cp/∆α (steady)
30
25 15
20
10
15
10 5
5
0
0
0.0 0.2 0.4 0.6 0.8 1.0
0.0 0.2 0.4 0.6 0.8 1.0
-5 -5
X/C X/C
60
Y/2b=97.7 %
50
40
30
20
10
0
0.0 0.2 0.4 0.6 0.8 1.0
-10
X/C
Figure 5.53 : Pressure coefficient ratio for the F-5 wing at M∞ = 0.9 and ∆α = 0.5°.
In Fig. (5.54) and (5.55) the real and imaginary parts of the pressures respectively, are
compared with the steady downwash correction method (DWM), as well as the experimental
measurements. Fig (5.53) shows a good agreement between the results.

35 30
Experiment y/2b=51.5% Experiment
30 25 y/2b=87.5%
SKEM SKEM
25 20 DWM
DWM
Re (∆ Cp)
20 15
Re (∆ Cp)
15 10
10 5
5 0
0 0.2 0.4 0.6 0.8 1
0 -5
0 0.2 0.4 0.6 0.8 1
-5 -10
X/C X/C
60
Experiment
50 SKEM
DWM
40
y/2b=97.7%
Re (∆ Cp)
30
20
10
0
0 0.2 0.4 0.6 0.8 1
-10
X/C
frequency k = 0.275.
15 10
10
5
5
Im (∆ Cp)
0
Im (∆ Cp)
0
0 0.2 0.4 0.6 0.8 1
-5 0 0.5 1
-5 y/2b=87.5%
-10
Experiment Experiment
-15 -10
SKEM SKEM
-20 y/2b=51.5% DWM
DWM
-15
-25
X/C X/C
15
10
5
Im (∆ Cp)
0
0 0.2 0.4 0.6 0.8 1
-5
y/2b=97.7%
-10
Experiment
-15 SKEM
DWM
-20
X/C
reduced frequency k = 0.275.
However, the results presented in Fig. (5.55), for the imaginary part show that, the
steady pressures-based downwash correction leads to an erroneous shock jump behavior,
when comparing with the experimental measurements, whereas the results from the
successive kernel expansion method agree with the experimental measurements.

In the station nearest the tip the discrepancies between the computed and experimental
pressures are more pronounced. This fact is related to the quality of the nonlinearly computed
steady pressures from the Navier Stokes simulation. Near the tip, in most cases the mesh
presents some problems in representing the tip geometry and this fact can introduce some
discrepancies in computing the flow in this region. Therefore, problems regarding the
computation of pressure distributions lead to incorrect unsteady pressures distributions
predicted by the successive kernel expansion method, because this procedure depends on
those steady pressures distributions.
B) F-5 Wing Pitching About 50% Root Chord at M = 0.948 and k = 0.264
The next results to be presented are analogous with the previous ones, but for different
Mach number and reduced frequency. In this case, the Mach number is 0.948, and the reduced
frequency is 0.264. The same comparison was performed against experimental data (Tijdeman
et al., 1978), for a steady state zero angle of attack situation, in order to evaluate the quality of
the CFL3D solution, as presented in Fig. (5.56).
The quasi-steady pressure ratios are plotted in Fig. (5.57). The shock jump is well
characterized, however, at spanwise station 64.1%, there is a second peak near the leading
edge, caused by a distortion on the prediction of the leading edge suction peak at angle of
attack of 0.5o.
-0.8 -0.8
y/2b=35.2% y/2b=64.1%
-0.4 -0.4
Steady Cp
Steady Cp
0 0
0.4 Experiment (Upper) 0.4 Experiment (Upper)

Experiment (Lower) Experiment (Lower)
CFL3D (Upper) CFL3D (Upper)
0.8 0.8
CFL3D (Lower) CFL3D(Lower)
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
X/C X/C
-0.8
y/2b=81.7%
-0.4
Steady Cp

Experiment (Lower)
0.8
CFL3D (Upper)
CFL3D (Lower)
0 0.2 0.4 0.6 0.8 1
X/C
Figure 5.56 : Steady pressure distribution for the F-5 wing at M∞ = 0.948 and α = 0°.
Y/2b=35.2 % Y/2b=64.1 %
10 25
9
8 20
7
6 15
5
4 10
3
2 5
1
0 0
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
X/C X/C
Y/2b=81.7 %
35
30
25
20
15
10
0
0.0 0.2 0.4 0.6 0.8 1.0
X/C
Figure 5.57 : Pressure coefficient ratio for the F-5 wing at M∞ = 0.948 and ∆α = 0.5°
In this higher Mach number, one should notice that the results regarding successive
kernel expansion method are also in good agreement with the experimental data (Fig. 5.58),
even though with the presence of strong shock waves. The shock wave jump behavior is well
represented for both real and imaginary part of the pressure coefficients, as one should
observe in Figs. (5.58) and (5.59). However, near the leading edge at the spanwise station
64.1%, a discrepant behavior of the computed complex pressures is noted, since the same
distortion is embedded in the reference quasi-steady pressures.
25 20
Experiment
15
20 SKEM
15
DWM 10
Re (∆ Cp)
Re ( ∆ Cp)
5
y/2b=35.2%
10
0
5 0 0.2 0.4 0.6 0.8 1
-5 Experiment
0
SKEM y/2b=64.1%
-10
0 0.2 0.4 0.6 0.8 1 DWM
-5 -15
X/C X/C
25
Experiment
20 SKEM
DWM
Re (∆ Cp)
15
y/2b=81.7%
10
0
0 0.2 0.4 0.6 0.8 1
X/C
frequency k = 0.264.
6 6
4 y/2b=35.2% 4 y/2b=64.1%
2 2
0 0
Im (∆ Cp)
Im (∆ Cp)
0 0.2 0.4 0.6 0.8 1

-2 -2 0 0.5 1
-4 -4 Experiment
-6 Experiment -6 SKEM
SKEM -8
DWM
-8
DWM
-10 -10
X/C X/C
4 y/2b=81.7%
2
0
0 0.2 0.4 0.6 0.8 1
Im (∆ Cp)
-2
-4
-6
-8
Experiment
-10
SKEM
-12
DWM
-14
X/C
The main cause for the differences between the pressures computed using the
successive kernel expansion method, near the leading edge of spanwise station 64.1 % (Figs.
5.58 and 5.59), is related to problems regarding the nonlinear quasi-steady pressures
computation at this position. Therefore, from the results presented above, it is concluded that
an accurate unsteady pressure prediction depends strongly on a correct quasi-steady mean
flow pressure distribution.

C)- LANN Wing in Pitch About 62% Root Chord at M∞ = 0.822 and k = 0.105.
Figure 5.60 : Sketch of the LANN wing and the corresponding discrete element aerodynamic
model.
The LANN wing is a supercritical wing with aspect ratio of 7.92 and 25o of sweep
angle along ¼ chord axis. It consists in a good test case for evaluation of the present
procedure, since it is a supercritical profile configuration. The wing was discretized in 280
panels in order to perform unsteady pressures computation based on the ZONA 6 linear
aerodynamic model. The same steps were repeated with regard to the nonlinear pressure
distribution computation. A steady CFL3D Navier-Stokes solution at M∞ = 0.822 and α = 0.6°
was simulated (Fig. 5.61) and the pressures correlate very well with test data (Malone et al.,
1960). The unsteady pressure distributions presented by Malone et al., (1960) correspond to
the wing undergoing a rigid body pitch oscillation around the 62% of the root chord location.
The quasi-steady pressure coefficient ratio δ∆C pnl ∆α was computed for a smaller angle of
attack difference, by ( ∆C p (α = 0.75 ) − ∆C p (α = 0.6 ) ) ( ∆α = 0.15 ) ,

nl o nl o o
since this profile is
supercritical, and smaller angle of attack changes can strongly modify the transonic flow
behavior. The quasi-steady pressure differences ratios are plotted in Fig. (5.62), where it is
possible to note that, at spanwise station 82.5%, there are again two peaks, since this station is
near from the tip, and the same aforementioned meshing problem arose.
Cp at Y/L=0.475 Cp at Y/L=0.65
1.2 1.2
1 1
0.8 0.8
0.6 0.6
Cp (steady)
Cp (steady)
0.4 0.4
0.2 0.2
0 0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
-0.2 -0.2
Experiment (upper) Experiment (upper)
-0.4 -0.4
Experiment (lower) Experiment (lower)
-0.6 -0.6
CFL3D CFL3D
-0.8 -0.8
X/C
x/c
Cp at Y/L = 0.825
-1.2
y/2b=82.5%
-0.8
Steady Cp
-0.4
0.4
Experiment (Upper)
Experiment (Lower)
0.8 CFL3D (Upper)
CFL3D (Lower)
0 0.2 0.4 0.6 0.8 1
X/C
Figure 5.61: Steady pressure distribution for the LANN wing at M∞ =0.822 and α = 0.6°.
Y/L=0.475 Y/L=0.65
80 50
70
40
60
50 30
40
20
30
20 10
10
0
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
-10 -10
X/C X/C
Y/L=0.825
50
40
30
20
10
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
-10
-20
X/C
Figure 5.62 : Pressure coefficient ratio for the LANN wing at M∞ =0.822 and ∆α = 0.15°.
In Figs. (5.63) and (5.64) one should observe a good agreement with test data is
obtained at the spanwise stations y/2b = 47.5% and 65% by the successive kernel expansion
method, while the steady downwash correction method gives incorrect shock jump in the
imaginary pressure component (Fig. 5.64). The oscillating pressure behavior near the shock
jump at y/2b = 82.5% computed by the successive kernel expansion method is due to
oscillations δ∆C pnl ∆α computed by CFL3D simulation. However, the shock positioning and
strength ate well predicted for the three spanwise stations.
80 50
70
SKEM 40 SKEM
60 DWM DWM
50 30
y/2b=47.5% y/2b=65%
Re (∆ Cp)
Re (∆ Cp)
40
20
30
20 10
10
0
0
0 0.2 0.4 0.6 0.8 1
0 0.2 0.4 0.6 0.8 1
-10 -10
X/C X/C
50
Experiment
40
SKEM
30 DWM
20
Re (∆ Cp)
10
0
0 0.2 0.4 0.6 0.8 1
-10
-20
y/2b=82.5%
-30
X/C
Figure 5.63 : Real part of the pressure distribution for the LANN wing at M∞ =0.822 and

10 10
5
5
0
0
0 0.2 0.4 0.6 0.8 1
-5 0 0.5 1
Im (∆ Cp)
Im (∆ C p)
-5
-10
-10
-15 y/2b=47.5% y/2b=65%
-15 Experiment
-20 Experiment
-20 SKEM
-25 SKEM
DWM DWM
-30 -25
X/C X/C
10
8 y/2b=82.5%
6
2
Im (∆ Cp)
0
0 0.2 0.4 0.6 0.8 1
-2
-4
Experiment
-6
SKEM
-8
DWM
-10
X/C
Figure 5.64 : Imaginary part of the pressure distribution for the LANN wing at M∞=0.822 and

D) Lessing Wing in First Bending Mode at M = 0.9 and k = 0.13
Figure 5.65 : Sketch of the Lessing wing and the corresponding discrete element aerodynamic
model.
The Lessing wing has a rectangular planform with an aspect ratio equal to 3.0, and a
5% parabolic arc profile section. Fig. (5.65) presents the sketch of the wing as well as the
linear aerodynamic model mesh employed in the linear unsteady pressures computation via
the ZONA 6 aerodynamic theory, comprised by 220 panels, equally distributed along the
chord length. Unsteady pressure measurements were performed for this test wing, considering
a bending mode displacement of the lifting surface (Lessing et al., 1960). The objective of
this test case is to demonstrate that the successive kernel expansion method can give accurate
unsteady pressure of a bending mode even if δ∆C pnl ∆α is computed based on a quasi steady
pitch mode at k = 0. Steady pressures distribution were computed by the CFL3D Navier
Stokes code, and the pressures distribution for M∞ =0.9 and α = 0° are presented in Fig.
(5.66), showing a good agreement with experimental data.
The computation of the quasi steady pressure ratios δ∆C pnl ∆α was performed as
( ∆C p (α = 0.2 ) − ∆C p (α = 0.0 ) ) ( ∆α = 0.2 ) , and the resulting distributions are presented in

nl o nl o o
Fig. (5.67). One should note that in the station near the wing tip there is an oscillation of the
computed quasi-steady pressures ratios due to the same reason mentioned in the previous test
cases, regarding the CFD aerodynamic meshing problems.
-0.4
-0.5
y/2b=50% -0.3
-0.4
-0.3 -0.2
Steady Cp
-0.2 -0.1
Steady Cp
-0.1 0 y/2b=70%
0
0.1
Experiment (Upper) 0.2
Experiment (Lower)
0.2 Experiment (Lower)
0.3 CFL3D (Upper)
0.3 CFL3D (Upper)
CFL3D (Lower) CFL3D (Lower)
0.4
0.4
0 0.2 0.4 0.6 0.8 1
0 0.2 0.4 0.6 0.8 1
X/C X/C
-0.4
-0.3
-0.2
Steady Cp
-0.1
0
y/2b=90%
0.1
Experiment (Upper)
0.2
Experiment (Lower)
0.3 CFL3D (Upper)
CFL3D (Lower)
0.4
0 0.2 0.4 0.6 0.8 1
X/C
Figure 5.66 : Steady pressure distribution for the Lessing wing at M∞ =0.9 and α = 0°.
Y/2b=50 % Y/2b=70 %
20 18
16
15 14
12
10
10
8
6
5
4
2
0
0
0.0 0.2 0.4 0.6 0.8 1.0
-2 0.0 0.2 0.4 0.6 0.8 1.0
-5 -4
X/C X/C
Y/2b=90 %
14
12
10
0
0.0 0.2 0.4 0.6 0.8 1.0
-2
X/C
Figure 5.67 : Pressure coefficient ratio for the Lessing wing at M = 0.9 and ∆α = 0.2°.
In the midspan station (50%), there is also a slight oscillation immediately after the
shock wave position, towards to the leading edge. This oscillation is reflected in the
computation of the unsteady pressures amplitudes for the bending displacement of the wing,
as one should observe in the corresponding plot in Fig. (5.68).

14 16
Experiment (Test 1) Experiment (Test 1)
14
12 Experiment (Test 2) Experiment (Test 2)
ZONA6 (Linear) 12 ZONA6 (Linear)
10 SKEM SKEM
Magnitude
Magnitude
10
8
y/2b=50 % 8 y/2b= 70%
6
6
4 4
2 2
0
0
0 0.2 0.4 0.6 0.8 1
0 0.2 0.4 0.6 0.8 1
X/C X/C
16
Experiment (Test 1)
14
Experiment (Test 2)
12
ZONA6 (Linear)
Magnitude
10 SKEM
8
y/2b=90%
6
0
0 0.2 0.4 0.6 0.8 1
X/C
Figure 5.68 : Pressure distribution amplitudes for the Lessing wing at M∞ =0.9 and reduced
frequency k = 0.13.
300 300
Experiment (Test 1) Experiment (Test 1)
250 Experiment (Test 2) 250 Experiment (Test 2)
ZONA6 (Linear) ZONA6 (Linear)
Phase Angle (deg)
Phase Angle (deg)

200 SKEM 200 SKEM
150 150
100 100
50 50
y/2b=50% y/2b=70%
0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
X/C X/C
300
Experiment (Test 1)
250 Experiment (Test 2)
Phase Angle (deg)
ZONA6 (Linear)
200
SKEM
150
100
50
y/2b=90%
0
0 0.2 0.4 0.6 0.8 1
X/C
Figure 5.69 : Pressure distribution phases for the Lessing wing at M∞ =0.9 and reduced
frequency k = 0.13.
However, an excellent agreement with test data is obtained by the successive kernel
expansion method, as presented in the results summarized in the plots of Figs. (5.68), and
(5.69), whereas the linear method such as ZONA 6 fails to predict the unsteady shock effects,
as expected. These results are encouraging, since it is possible to note that this method,
employing a quasi steady pitch motion, has the capability of correcting pressures with respect
to a bending type motion.

In all of the test cases it was observed that the successive kernel expansion method
was capable to restore the nonlinear unsteady pressure distribution and the shock wave
dynamics. Therefore, from these results, it is possible to conclude that the unsteady character
of the small disturbance transonic flow can be computed as a contribution predicted by a
small disturbance linear aerodynamic model.
However, it is clear that the successive kernel expansion method performance is
strongly dependent on the quality of the computed steady mean flow pressure distribution.
Any distortion in the computation of the pressures difference coefficient ratios will be
reflected in the computation of the approximate unsteady aerodynamic loading.
5.3.3 - Validation of the Successive Kernel Expansion Method for Flutter
Boundaries Computation
Flutter calculations were performed taking three standard aeroelastic wing
configurations, for which the aeroelastic stability was evaluated in wind tunnel tests. The
following test cases were selected:
A) - AGARD 445.6 wing - weakened model # 3 (Yates, 1988);
B) - PAPA wing (Pitch And Plunge Apparatus, Farmer and Rivera., 1988);
C) - YXX wing (Isogai, 1983).
The successive kernel expansion method (SKEM) is adopted for unsteady pressures
computation in order to provide the unsteady aerodynamic loading for flutter computations.
The nonlinear pressure distributions were computed for these wings undergoing quasi-steady
motion, as it was performed for the cases of transonic unsteady pressures prediction,
investigated in the last section.

A) Flutter Boundary of the AGARD 445.6 Weakened Wing
The investigation of the AGARD wing 445.6 aeroelastic stability is revisited, with the
objective to evaluate the performance of the successive kernel expansion procedure for the
transonic flutter prediction.. The reference nonlinear steady pressure distribution was
computed by the numerical method for solving the Navier-Stokes solver presented in
Chapter 3 at the angles of attack 0.0o and 0.5o , in order to compute the reference nonlinear
quasi-steady pressures disturbances. The discrete element aerodynamic model, based on the
ZONA 6 method, is comprised of 220 panels, uniformly distributed along the chord length
(Fig 5.1). The flutter computations were performed using the g-method. Flutter boundaries at
M∞ = 0.678, 0.901 and 0.95 are computed using ZONA6 (linear method), the steady
downwash weighting method (DWM) and the successive kernel expansion method, and the
results are presented in Fig. (5.70).
0.5 24.0
0.48 DWM
0.46 22.0 ZONA 6
Experiment
0.44
Frequency (Hz)
20.0 SKEM
0.42
FSI
0.4 18.0
0.38 DWM
0.36 ZONA 6 16.0
Experiment
0.34 SKEM
14.0
0.32
0.3 12.0
0.6 0.7 0.8 0.9 1 0.6 0.7 0.8 0.9 1
Mach Mach
Figure 5.70 : Flutter boundaries (dynamic pressure and flutter frequency) for the AGARD
wing 445.6, weakened model #3.

Among these three methods, the results obtained by the latter procedure present the
best correlation with the test data, when observing the capability in capturing the transonic
dip, and in decreasing the flutter speed, approaching the experimental results. Note that the
flutter speeds predicted by the steady downwash correction method applied to the ZONA 6
aerodynamic model lead to flutter speeds higher than the ones computed by the same
correction procedure using the doublet lattice method, for which the results were presented in
Fig. (5.46).
Previous computations indicated that large correction factors, resulting from the
correction of the pressures near the leading edge, lead to distortions of the computed unsteady
pressures. These large correction factors, which are computed by the steady downwash
correction, are the reason for the large differences between the computed flutter speed by the
steady downwash correction procedure with respect to the other results, already presented in
Fig (5.46). Those large correction factors would tend to increase the unsteady aerodynamic
loading in such way that the flutter speeds are underestimated, because there is an
overestimation of the magnitude of the unsteady loading.
A strategy to circumvent the problem is to suppress the correction of the pressures at
panels near the leading edge. This approach is consistent, because the objective of the present
investigation is to correct an unsteady aerodynamic loading to take into account transonic
effects, instead of potential effects such as the differences in the computation of the leading
edge suction peak. Furthermore, the main cause for these differences is related to a mesh
interpolation problem. One should observe that the control point is at the centroid of the
panel. Near the leading edge, the pressure distribution along the chord, presents higher
gradients when comparing to the remaining portion of the airfoil section. Thus, the
interpolated pressures at the control point are lower in magnitude than the pressures between
the control point and the leading edge of the wing, predicted at the mesh points of the finite
difference Navier Stokes numerical model. Since the purpose of the correction method is to
take into account the transonic flow behavior in the sense of the contribution of the shock
wave positioning and strength, the suppression of correction effects of potential nature does
not impair the procedure.
In the steady downwash correction the pressure phases are not corrected, resulting in
more conservative flutter dynamic pressures, when comparing with strictly linear results. In
addition, one should note that the results of this method, for transonic Mach numbers, have
the same behavior of the uncorrected linear ones, regarding the transonic dip curve slope. This
is an indication that the transonic dip behavior is closely related to the contribution of the
imaginary part of the pressures.
In the case of the successive kernel expansion procedure, the computed correction
factors result from the matching of nonlinear unsteady pressures, composed by nonlinear
steady mean flow pressures and a linear unsteady pressures contributions. Therefore, the
correction factor will take into account in an approximate form the unsteady transonic flow
behavior, since its computation is referred to those nonlinear unsteady pressures. In summary,
the dip slope, predicted by the successive kernel expansion method is increased, because the
correction procedure takes into account the influence of the real and imaginary part of the
transonic unsteady pressures. Thus, this improvement in the reference conditions leads to
better results in approaching the experimental measurements
The plots of the flutter frequency as a function of the Mach number, (Fig. 5.70)
indicate that the successive kernel expansion method overpredicts the flutter frequencies. The
reason for this distortion is that a larger change in the flutter frequency is expected since the
successive kernel expansion procedure introduces an important contribution to the imaginary
part of the complex eigenvalue problem involved in the flutter solution. Furthermore, the
correction factors are computed based on pressure disturbances ratios with respect to a given
pitch amplitude of the motion. However, the flutter mode is not a single pitch displacements,
as it consists in a flexure-torsion mode shape displacement, in the case of this wing under
investigation. Therefore, there is no guarantee that the computed correction factors will
properly restore the pressures distribution at the flutter mode, even considering that these
correction factors are computed independently of the motion nature.
It should be observed that the aerodynamic interference effects are computed by the
discrete kernel function method, which in the present test case is the ZONA 6. Interference
effects obtained from a pitch downwash mode are different from the interference effects due
to a flexure-torsion mode, even if the reduced frequency is the same. This fact suggest the use
of a “frozen” flutter mode or any other elastic mode, instead of a single pitch mode, for the
computation of the nonlinear quasi-steady pressures conditions to obtain the correction
factors. This approach allows to take into account the nonlinear interference effects, which are
embedded in the reference quasi-steady pressure distribution.
These observations are also valid for the explanation of the results with regard to the
flutter speeds. The differences of the computed flutter speeds using the successive kernel
expansion method with respect to the experimental results (Fig. 5.70) are justified by the same
reason, that is, the absence of a correction which takes into account the interference effects
may lead to differences between the computed and the experimental flutter speeds.
B) Flutter Boundary of PAPA Wing at α= 1°
The next test case is the investigation of the aeroelastic stability margins of the PAPA
wing. This wing is a good test case to evaluate the capability in correcting the unsteady
aerodynamics based on a supercritical flow pressure distribution, and for a low aspect ratio
wing. The ZONA 6 aerodynamic model is comprised of 200 panels, equally distributed along
the chord length. The same modeling features, employed in the AGARD wing 445.6 case are
considered, such as the suppression of correction factors for the leading edge panels. The
flutter solution technique employed is the g-method. The results of the flutter computation of
the PAPA wing, at two distinct angles of attack, are presented in Figs. (5.71) and (5.72).
200 6
Experiment α = +1 (deg) Experiment α = +1 (deg)
190 Zona 6 Zona 6

Dyn. Pressure (psf)
DWM 5.5 DWM

180 SKEM SKEM
170
5
160
150
4.5
140
130 4
0.3 0.4 0.5 0.6 0.7 0.8 0.9 0.3 0.4 0.5 0.6 0.7 0.8 0.9
Mach Mach
Figure 5.71 : Dynamic pressures and frequencies in the flutter boundaries of the PAPA wing
at angle of attack, α = 1.0o

200 6
190 Zona 6 Zona 6

DWM DWM
Dyn. Pressure (psf)
SKEM 5.5 SKEM

180
170
5
160
150
4.5
140
α = -2 (deg) α = -2 (deg)
130 4
0.3 0.4 0.5 0.6 0.7 0.8 0.9 0.3 0.4 0.5 0.6 0.7 0.8 0.9
Mach Mach
Figure 5.72 : Dynamic pressures and frequencies in the flutter boundaries of the PAPA wing
at angle of attack, α = -2.0o.
Another feature to be highlighted, with regard to the successive kernel expansion
method, is the capability to predict the flutter of the wing at different initial angles of attack,
instead of the linear theory, which can only predict the flutter margins for a zero angle of
attack. This is so, because the steady mean flow can be computed at a given steady state angle
of attack, and the nonlinear unsteady pressures, used to compute the correction factors, will be
obtained from the contribution of the nonlinear steady mean flow pressures added to unsteady
linear pressures. Therefore, the angle of attack contribution is included in the mean flow
conditions, because unsteady linear pressures, predicted by the linear unsteady aerodynamic
model is independent of the angle of attack.
One should observe that the downwash correction method largely underpredicts the
flutter dynamic pressure while the flutter results of the successive kernel expansion method
correlate well with test data, both for the subsonic and the transonic Mach numbers. This is so
because the unsteady components of the nonlinear pressures, introduced by the correction
procedure, play an important role in the flutter computation. However, the flutter frequencies
are underestimated, as one should observe in Figs. (5.71) and (5.72). The reason for those
differences are the same presented in the case of the AGARD wing 445.6. In the present test
case, the flutter results from the coalescence of the pitch an plunge mode, while the nonlinear
mean flow reference pressures are obtained for a quasi steady rigid body pitch mode.
C) - Flutter Boundary of the YXX Wing
The YXX wing is an elastic swept wing with supercritical profile, which was tested in
wind tunnel under transonic flow conditions. It has an aspect ratio of 10.0, a 17o quarter chord
sweep angle, and supercritical airfoil tapered from 16% of thickness at the root section to 12%
at the tip section. Its planform presents a crank in the trailing edge, and demands a more
detailed linear aerodynamic model. The wing was modeled by the ZONA 6 method, where
the lifting surface was subdivided in 360 panels, equally spaced along the chordwise
direction.
Figure 5.73 : Sketch of the YXX wing and the corresponding discrete element aerodynamic
model.
The flutter boundaries were measured by Yonemoto (1984), and also computed by
Isogai (1983). This last author identified that the flutter boundaries are extremely sensitive to
the location of the center of the pressure; a shift in 2% of the elastic axis causes a 70% of
change in the flutter dynamic pressure, at M∞ =0.7. Furthermore, this wing presents a large
aeroelastic static deformation, even at low angle of attack. Therefore, such static aeroelastic
deformation may alter significantly the position of the center of the pressure, indicating the
flutter boundary depends on the static aeroelastic behavior of the structure.
Chen et al. (2003) present an overset filed-panel method for unsteady transonic
aerodynamic computations. The YXX wing test case was also investigated, with regard to its
aeroelastic stability. The results of the ZTRAN method show that the variation of the static
angle of attack promotes a strong impact on the wing flutter boundary at transonic flow
conditions. The discrepancies of the ZTRAN method results and the experimental data are
believed to be due to the exclusion of the static aeroelastic deformation in the CFL3D
aerodynamic mesh, since it is considered in the Navier-Stokes computations that the wing is
rigid. In addition, Chen et al. (2003) concluded in their work that the reference angle of attack
condition, which presents the best correlation with experimental data is α = -1.0o. This is so,
because among the angles of attack of 2.0o, 0.0o, and –1.0o the steady background flow at α=-
1.0o , computed by the CFL3D should be the closest one to that of a static aeroelastic
deformed wing.
The quasi-steady pressure distribution to be used in the successive kernel expansion
method, and the steady downwash weighting method, was computed with the CFL3D Navier
Stokes code, for two angles of attack : -0.8o and –1.0o. The selection of the reference angle of
attack as –1.0o is made in order to compare the results of the present investigation with the
ones from the ZTRAN method. The flutter results are presented in Figs. (5.74), and (5.75).
10
Dyn. Pressure (KPa)
6
Experiment (AOA=0 deg)
ZONA6 (Linear)
4 SKEM (AOA=-1 deg)
ZTRAN (AOA=-1 deg)
DWM (AOA=-1 deg)
2
0.6 0.65 0.7 0.75 0.8 0.85 0.9
Mach
Figure 5.74 : Flutter dynamic pressure for the YXX wing, at angle of attack α = -1.0o.
250
200
150

100 ZONA6 (Linear)
SKEM (AOA=-1 deg)
ZTRAN (AOA=-1 deg)
DWM (AOA=-1 deg)
50
0.6 0.65 0.7 0.75 0.8 0.85 0.9
Mach
Figure 5.75 : Flutter frequency for the YXX wing, at angle of attack α=-1.0o.
One should notice that these results presents a large overprediction of the transonic dip
phenomenon, with respect to the experiment, for both methods under investigation (DWM
and SKEM). In addition these results are also different from the ones computed by the
ZTRAN method, at α = -1.0o. However, the transonic dip behavior is represented in these
results. The overall quality of this solution is questionable, since the pressure ratios due to the
quasi-steady motion were based on a variation in angle of attack amplitude, which play an
essential role for supercritical wing profiles. At angles of attack of –0.8o and –1.0o, one should
observe that differences between the shock wave structure in the steady nonlinear pressure
field are quite significant. Probably the expected linear behavior concerning the pressures
under quasi steady motion is failing, and the changes in the flow are unrealistic.
6 – CONCLUSIONS AND FINAL REMARKS
6.1 Conclusions
A study on correction methods for aeroelastic stability analysis in transonic flow has
been developed. This investigation can be outlined in three major steps. First, a
linear/nonlinear behavior investigation of unsteady transonic flows has been conducted, in
order to allow a better understanding of the linear limits. Such knowledge, in turn, allows
correction methods to be developed and investigated. Second, downwash correction methods
for transonic flutter computation were evaluated, including a sensitivity analysis of the
computed transonic flutter speeds as a function of unsteady reference pressures resulting from
different amplitudes of the lifting surface motion. Finally, an extension of downwash
correction methods was developed, leading to a new formulation named as “successive kernel
expansion method”, for which the theoretical background and hypotheses are supported by the
subsonic discrete kernel function method and observations on the transonic flow behavior,
respectively. The conclusions that can be drawn from the present investigation are outlined in
the forthcoming paragraphs.
Simulations of unsteady transonic flow over the NLR F-5 wing model were performed
in order to study the linearity of unsteady loads and shock displacement with respect to
dynamic angle of attack amplitude and frequency. A linear behavior was quantified in terms
of the variation of lift and moment coefficient derivatives and shock displacement as a
function of the amplitude of the motion of the NLR F-5 wing.
The aerodynamic response and shock displacement were found to include higher
harmonics, which are usually not considered in correction procedures applied to linear
aeroelastic methods. In flutter analysis, there are situations in which a second mode is close to
Chapter 6 – Conclusions and Final Remarks 177
a multiple of a first mode. In such case, the presence of a second harmonic in the lower mode
aerodynamic response may facilitate the exchange of energy between the modes and be a
contributing factor to coalescence. For this reason, the present results seem to indicate the
need for considering such harmonics in future correction methods.
Linear boundaries were computed for a few reduced frequencies and spanwise stations
along the NLR F-5 wing, using two criteria, based on moment coefficient and shock
displacement. These boundaries were found to be more conservative for the shock
displacement criterion. The linear limits calculated using the moment coefficient criterion
were found to depend significantly on the spanwise station. This indicates that simply using
two-dimensional linear limits would not be appropriate when dealing with correction
methods.
The linear/nonlinear behavior investigation indicated that the unsteady transonic flow
is strongly dependent on the amplitude of the disturbances it is subjected. Notwithstanding the
dependence of the linear limits on reduced frequency and spanwise station, it is clear that
some degree of linear behavior may be assumed for aeroelastic methods that employ corrected
coefficients around mean flow reference conditions. However, this assumption of linearity
should be used with caution and would not apply to problems such as aeroelastic response or
limit cycle oscillations, unless the amplitude of deformation is kept below previously
identified linear limits that take into account reduced frequency and spanwise station. In a
context of small disturbances associated to lifting surface motion, unsteady transonic flow
behaves linearly around a nonlinear mean flow. The identified linear limits were higher than
the small amplitudes associated to aeroelastic dynamic deformations of the structure, which
ensures the feasibility of applying correction methods in this context.
The evaluation of the downwash weighting method (DWM) confirmed that the
matching of the pressures, based on the correction of the linear discrete element kernel
function method (ZONA 6, doublet lattice method), gives better results than force matching
methods for approximate transonic flow computations. This is due to the fact that the steady
mean flow nonlinear pressure distribution is preserved, whereas the force matching methods
do not present this capability. One should observe that in force matching procedures, there are
fewer unknowns (weighting factors) than the order of the AIC matrix. Thus, the use of the
least squares method is required to compute the correction factors. The resulting corrected
pressures computed by such procedure present a wrong trend regarding the shock wave
dynamics. Consequently, flutter boundary computations using such procedures do not present
good results and thus are not adequate for the approximate modeling of unsteady transonic
flow for aeroelastic applications.
Downwash weighting methods, using either steady or unsteady pressures as reference
conditions were investigated. The transonic dip was captured with application of both
correction approaches. However, the decrease of the flutter speed curve slope was more
evident and closer to experiments when the unsteady downwash correction method was
employed. The flutter speed curve slope obtained by the steady downwash weighting method
was about the same as the one obtained by the uncorrected aeroelastic stability analysis. It
may be concluded that the unsteady flow contribution, embedded in the unsteady reference
pressures, plays an important role in the transonic flutter phenomenon. However, the unsteady
downwash weighting method requires unsteady CFD computations, increasing its cost with
respect to the steady downwash weighting procedure. Furthermore, in a strict sense, it should
be applied only at the specific reduced frequency for which the unsteady pressures were
obtained. Also, the unsteady downwash weighting method depends on the nature of the
unsteady reference pressures, as indicated by the sensitivity analysis to be commented in what
follows.
A sensitivity analysis was performed in order to investigate the role of the unsteady
reference pressure disturbances, which are referred to a given amplitude of the lifting surface
motion on the prediction of the transonic flutter by the use of the downwash correction
method. This investigation indicated that the flutter boundaries computed by unsteady
downwash correction methods are significantly dependent on the unsteady pressures taken as
reference conditions.
The above is consistent with the observations regarding the linear/non-linear
investigations, where it was found that the linear limit associated to the lifting moment is
more sensitive to the amplitude of the motion. The flutter instability is mainly governed by
the lifting moment, which is associated to the streamwise displacement of the center of
pressures. Investigation of the local two dimensional flow over the AGARD 445.6 wing
indicated that there is a formation of a secondary suction peak which displacement may alter
the position of the center of pressures. The appearance of this secondary suction peak results
from the Navier-Stokes computation of the flow around the leading edge of a profile with
finite thickness. Therefore, aerodynamic lag appears, which is dependent on the amplitude of
the motion at a given reduced frequency. These observations explain the dependence of the
computed flutter speed on these amplitudes. The overall results confirm that the nonlinear
effects associated to the profile thickness play an important role in flutter prediction by
correction methods based on nonlinear unsteady pressure distributions.
The development of the successive kernel expansion method arose from the
aforementioned observations regarding the unsteady transonic flow behavior. The advantages
of such procedure is the dependency on a nonlinear steady state pressure distribution, instead
of nonlinear unsteady transonic pressures computation or measurement. This method was
validated for unsteady pressures calculation and flutter computations. The results showed that
the quality of the unsteady transonic pressure distributions computed by the successive kernel
expansion methods depends on the quality of the steady pressures taken as reference
conditions. The shock wave dynamics was restored when compared with experimental
measurements, independently of shock strength. Some discrepancies were observed in the
unsteady pressures computation near the lifting surface tip. These are due to the CFD method
failing to accurately predict the viscous flow field there. Care must be taken in order to have a
good CFD solution because its quality has an important role in the computation of the ratio
between the steady pressure and the quasi steady motion amplitude.
The flutter computation using the successive kernel expansion method also led to good
results when predicting the transonic dip. However, some slight discrepancies were found,
mainly related to the flutter frequency prediction. The reason for these discrepancies is related
to the nature of the steady reference condition, regarding the absence of correction of
interference effects at the flutter mode, which may justify the over prediction of those flutter
frequencies. In general, the transonic dip, as well as the computed flutter speeds and
frequencies are in good agreement with experimental measurements. This procedure is a rapid
form to compute the transonic flutter speed boundaries.
The downwash weighing method based on unsteady reference pressures requires time
accurate CFD simulation. For that method, The unsteady flow Navier-Stokes solution,
expended approximately 8.2 CPU hours for the simulation of two cycles of pitching motion of
the AGARD 445.6 wing, in a Silicon GraphicsTM Octane II workstation, with two R12000
RISC architecture processors. On the other hand, the successive kernel expansion method
employs a steady nonlinear pressure distribution. The time expended for a steady flow
simulation of the same wing is approximately 1.3 hours elapsed time. The additional
computational effort expended by the ZONA 6 method modified by the successive kernel
expansion method is less than five minutes of elapsed time, on a computer with a single 2.8
GHz PentiumTM IV processor.

In summary, there is a restriction in the unsteady downwash correction method, for it
is not suitable for computing unsteady transonic flow for reduced frequencies other than the
one corresponding to the reference unsteady pressures. The successive kernel expansion
method does not have the same restriction, for it has the capability of computing unsteady
pressures based on steady state pressures, for any reduced frequency of interest. Thus, if one
were to conduct computations for a number of reduced frequencies, the time required by the
unsteady downwash correction method would be the number of reduced frequencies of
interest multiplied by the expended time for the unsteady Navier-Stokes CFD solution, while
the time require by the successive kernel expansion method would essentially not depend on
the number of reduced frequencies.
6.2 Contributions
The first contribution of the present work is the linear/nonlinear investigation in terms
of Navier-Stokes solutions. The improvement when comparing with the existing literature is
the use of three-dimensional Navier-Stokes solutions instead of two-dimensional transonic
small disturbances solutions, for the identification of the linear limits.
Another contribution is a comprehensive study of the downwash correction techniques
extending their application to the matching of pressures instead of strip loads, as it was
originally proposed by Pitt and Goodman (1987). Furthermore, the capability to correct the
pressures to downwash linear relation using either steady or unsteady reference nonlinear
pressure conditions was introduced, as was the idea to correct the downwash using unsteady
pressures computed for the flutter reduced frequency predicted by the uncorrected linear
aeroelastic model, in order to evaluate the stability of the aeroelastic system.
An expedient way to compute unsteady transonic flow was developed, named in this
work as the successive kernel extension method (SKEM). The application of such procedure
can be performed either for aeroelastic response or stability analysis, without the need of
unsteady CFD computations or measurements. The expediency of such method lies in the idea
of employing nonlinear steady state mean flow conditions, which are inexpensive to compute,
when compared with the unsteady ones, which were applied in the downwash correction
procedure. This feature makes this method suitable for application in industrial developments,
due to its computational efficiency.
6.3 – Recommendations for Future Work
The successive kernel expansion method may be further extended to cover full
transonic ranges including Mach numbers M∞ ≥ 1.0. The employment of different quasi-
steady patterns of motion to generate the steady non-linear reference field is another approach
to be investigated. In order to improve the performance of the procedure, a next step in the
continuation of the present effort might be to use a correction matrix which is constructed
from a set of steady downwash modes. However, such approach increases the computational
cost of the aeroelastic analysis since more steady CFD simulations would need to be
performed in order to create the correction factors. As the transonic dip phenomenon is
predicted, and the computed flutter speeds are conservative, the important aspect which
remains to be analyzed, hence, is whether the additional costs of such an approach are
worthwhile. Otherwise, the computational cost might be increased so much that it would be
near that of full computational aeroelasticity.
Additional investigation on the unsteady downwash correction method needs to be
performed in order to explain nonlinear behavior regarding the variation of the flutter speed
with the amplitude of the motion. It is suggested further investigations on viscous effects,
increasing the aerodynamic mesh refinement and confronting the results against Euler
computations.
Another approach which might be investigated is the use of a set of nonlinear unsteady
pressures computed from the transformation to the frequency domain of a impulse-type
aerodynamic time response. Such approach would circumvent the disadvantage in having a
single set of unsteady pressures for the correction of the pressure to downwash linear relation,
for different reduced frequency values.
The successive kernel expansion method results are encouraging, suggesting its further
extension. Complex configurations and high frequency tests cases were not performed. It is
suggested that the method be evaluated under these conditions in order to extend its
applicability .
7 – REFERENCES
1. Albano, E., and Rodden, W. P., “A Doublet Lattice Method for Calculating
Lift Distributions on Oscillating Surfaces in Subsonic Flows”, AIAA Journal, vol. 7, No. 2,
1969 pp 279-285.
2. Appa, K., “Constant Pressure Panel Method for Supersonic Unsteady Airload
Analysis”, Journal of Aircraft, vol. 24, No. 10, 1987, pp 696-702.
3. Ashley, H., “Role of Shocks in the ‘Sub-Transonic’ Flutter Phenomenon”,
Journal of Aircraft, Vol. 17, No. 3, March, 1980, pp 187-195.
4. Baker, M. L., "CFD Based Corrections for Linear Aerodynamic Methods".
AGARD R822, October 1997
5. Baker, M. L., Yuan, K., and Goggin, P.J., “Calculation of Corrections to Linear
Aerodynamic Methods for Static and Dynamic Analysis and Design”, Proc. of the 39th AIAA/
ASME/ASCE/AHS/ASC Structures, Structural Dynamics, and Materials Conference and
Exhibit, Paper AIAA-98-2072, Long Beach, California, April 20-23, 1998.
6. Baldwin, B., and Lomax, H., “Thin Layer Approximation and Algebraic Model
for Separated Turbulent Flows”, AIAA-Paper-78-257, AIAA 16th Aerospace Sciences
Meeting, Huntsville Alabama, January, 1978.
7. Ballhaus, W.F., and Goorjian, P.M., “Computation of Unsteady Transonic
Flows by the Indicial Method,” AIAA Journal, Vol. 16, No. 2, Feb. 1978, pp. 117-124.
8. Beam, R. M., and Warming, R. F., “An Implicit Factored Scheme for the
Compressible Navier-Stokes Equations”, AIAA Journal, Vol. 16, April 1978, pp. 393-402.
Chapter 7 – References 185
9. Bennet, R. M., and Edwards, J. W., “An Overview of Recent Developments in
Computational Aeroelasticity”, AIAA Paper 98-2421, 29th Fluid Dynamics Conference, June
15-18, Albuquerque, NM, 1998.
10. Bennet, R. M., Batina, J. T., and Cunningham, H. J., “Wing Flutter
Calculations with the CAP-TSD Unsteady Transonic Small-Disturbance Program”, Journal of
Aircraft, Vol. 26, No, 9, 1989, pp. 876-882.
11. Bergh, H., and Zwaan, R. J., “A Method for Estimating Unsteady Pressure
Distributions for Arbitrary Vibration Modes from Theory and form Measured Distributions
for One Single Mode”, NLR-TR F.250,1966.
12. Bertin, J.J. and Smith, M.L., Aerodynamics for Engineers, Prentice-Hall, New
Jersey, 1979, USA.
13. Bismark-Nasr, M. N., “Finite Elements in Applied Mechanics”, São Paulo,
Abete Gráfica, v.1 p. 591, 1993.
14. Bisplinghoff, R., L., Ashley, H., and Halfman, R. L., Aeroelasticity, Addison-
Wesley, 1955, Reprint Dover Publications, Inc., Mineola, N.Y., 1996.
15. Bisplinghoff, R., L., and Ashley, H., Principles of Aeroelasticity. John Wiley &
Sons, N.Y., 1962.
16. Brink-Spalink, J., and Bruns, J. M., “Correction Of Unsteady Aerodynamic
Influence Coefficients Using Experimental or CFD Data”, AIAA-2000-1498, 41st
AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynamics and Materials Conference
and Exhibit, Atlanta, Georgia, 3-6 April, 2000.

17. Chen, P. C., Gao, X. W., and Tang, L., “An Overset Field Panel Method for
Unsteady Transonic Aerodynamic Influence Coefficients Matrix Generation”, AIAA Paper-
2004-1512, In: Proceedings of the 45th AIAA/ASME/ASCE/AHS/ASC Structures, Structural
Dynamics & Materials Conference 12th AIAA/ASME/AHS Adaptive Structures Conference,
6th AIAA Non-Deterministic Approaches Forum, 5th AIAA Gossamer Spacecraft Forum,
Palm Springs, California, USA, 19 - 22 April, 2004.
18. Chen, P.C., and Liu, D.D., “Unsteady Supersonic Computation of Arbitrary
Wing-Body Configurations Including External Stores,” Journal of Aircraft, Vol. 27, No. 2,
Feb. 1990,pp. 108-116.
19. Chen, P.C., “A Damping Perturbation Method for Flutter Solution: The g-
Method,” International Forum on Aeroelasticity and Structural Dynamics, Hampton, VA, Jun.
22- 25, 1999.
20. Chen, P.C., Lee, H.W., and Liu, D.D., “Unsteady Subsonic Aerodynamics for
Bodies and Wings with External Stores Including Wake Effect”, Journal of Aircraft, Vol. 30,
No. 5, Sep.-Oct. 1993, pp. 618-628.
21. Chen, P.C., and Liu, D.D., “A Harmonic Gradient Method for Unsteady
Supersonic Flow Calculations”, Journal of Aircraft, Vol. 22, No. 5, Sep.-Oct. 1985, pp. 371-
379.
22. Chen. P. C., Sarhaddi, D., and Liu, D. D.,2000, “Transonic Aerodynamic
Influence Coefficient Approach for Aeroelastic and MDO Applications”, Journal of Aircraft,
Vol. 37, No. 1, pp. 85-94.

23. Dau, K., “A Semi-Empirical Method for Calculating Pressures on Oscillating
Wings in Unsteady Transonic Flow”, Report No. DA-EF24-B08/92, Deutche Airbus,
October, 1992.
24. Dowell, E. H. , Bland, S. R., and Williams, M. H., “Linear/Nonlinear Behavior
in Unsteady Transonic Aerodynamics”, AIAA Journal, Vol. 21, No. 1, January, 1983, pp. 38-
46.
25. Eversman, W., and Pitt, D.M., “Hybrid Doublet Lattice/Doublet Point Method
for Lifting Surfaces in Subsonic Flow, “ Journal of Aircraft, Vol. 28, No. 9, Sept. 1991, pp.
572 – 578.
26. Farmer, M. G., and Rivera, J. A., “Experimental Flutter results of a Cambered
Supercritical Wing on a Pitch and Plunge Apparatus”, presented at Aerospace Flutter and
Dynamics Council Meeting, NASA Langley Research Center, May, 1988.
27. Fung, K.Y., and Chung, A., “Computations of Unsteady Transonic
Aerodynamics Using Prescribed Steady Pressures,” Journal of Aircraft, Vol. 20, no. 12,
December 1983, pp. 1058-1061.
28. Garner, H. C., "A practical framework for evaluation of oscillating
aerodynamic loadings on wings in supercritical flows", RAE Technical Memorandum
Structures 900, United Kingdom, 1977.
29. Giesing, J. P. , Kalman, T. P., and Rodden, W. P. “Subsonic Unsteady
Aerodynamics for General Configurations, Air Force Flight Dynamics Laboratory, AFFDL-
TR-71-5, Part 1, Vol. 1, Wright Patterson Air Force Base, Ohio, Nov. 1971, Appendix A
30. Giesing, J. P., Kalman, T. P., and Rodden, W. P., “Correction Factor
Techniques for Improving Aerodynamic Prediction Methods”, NASA-CR-144967, 1976.
31. Gordnier, R. E., and Melville, R. B., “Transonic Flutter simulations Using an
Implicit Aeroelastic Solver”, Journal of Aircraft, Vol. 37, No. 5, 2000, pp. 872-879.
32. Harder, R.L., and Desmarais, R.N., “Interpolation Using Surface Splines,”
AIAA Journal, Vol. 9, No. 2, 1972, pp. 189-191.
33. Hassig, H., J., “An Approximate True Damping Solution of the Flutter
Equation By Determinant Iteration”, Journal of Aircraft, Vol. 8, No. 11, pp. 885-889., 1971
34. Hong, M. S., Bhatia, K. G., SenGupta, G., Kim, Taehyoun, Kuruvila, G., Silva,
W. A., Bartels, R., and Biedron, R., “Simulations of a Twin-Engine Transport Flutter Model
in the Transonic Dynamics Tunnel”, Paper No. US-44 In: Proceedings of the International
Forum on Aeroelasticity and Structural Dynamics, Amsterdam, Netherlands, June 2003.
35. Houwink, R., Kraan, A. N., and Zwaan, R. J., “Wind-Tunnel Study of the
Flutter Characteristics of a Supercritical Wing”, Journal of Aircraft, Vol. 19, No. 5, 1982.
36. Hui, W.H. " Stability of Oscillating Wedges and Caret Wings in Hypersonic
and Supersonic Flows" AIAA Journal. Vol. 7, No 8, Aug 1969, pp. 1524
37. Ichikawa, I., “Doublet Strip Method for Oscillating Swept Tapered Wings in
Incompressible Flow,” Journal of Aircraft, Vol. 22, No. 11, Nov. 1985, pp. 1008 –1012.
38. Isogai, K., “Numerical Simulation of Transonic Flutter of a High-Aspect-Ratio
Transport Wing”, National Aerospace Laboratory, Japan, Report TR-776T, August, 1983
39. Jadic I, Hartley D, Giri J. “An Enhanced Correction Factor Technique for
Aerodynamic Influence Coefficient Methods”. MacNeal-Schwendler Corporation’s
Proceedings for 1999 Aerospace User’s Conference, 1999.
40. Jadic, I., Hartley, D., and Giri, J. “Generalized Aerodynamic Forces Based on
CFD and Correction Factor Techniques for AGARD wing 445.6”. AIAA-Paper-2001-1208,
In: Proceedings of the 42nd AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynamics,
and Materials Conference and Exhibit, Seattle, Washington, April. 16-19, 2001.
41. Jadic, I., Hartley, D., and Giri, J. “Improving the Linear Aerodynamic
Approximation in Linear Aeroelasticity”, AIAA-Paper-2000-1450, In: Proceedings of the 41st
Structures, Structural Dynamics and Materials Conference and Exhibit, Atlanta, Georgia, 3-6
April 2000.
42. Jameson, A., Schmidt, W. and Turkel, E., “Numerical Solutions of the Euler
Equations by Finite Volume Method Using Runge-Kutta Time-Stepping Schemes”, AIAA
Paper 81-1259, In: Proceedings of the AIAA 14th Fluid and Plasma Dynamics Conference,
Palo Alto, USA, 1981.
43. Jordan, P. F., “Aerodynamic Flutter Coefficients for Subsonic, Sonic and
Supersonic Flow (Linear Two-Dimensional Theory),” British Aeronautical Research Council,
London, ARCTR R&M 2932, 1957.
44. Jordan, P. F., ‘Reliable Lifting Surface Solutions for Unsteady Flow,” Journal
of Aircraft, Vol. 15, September 1978, pp. 626-633.
45. Kalman, T.P., Rodden, W.P., and Giesing, J.P., “Application of the Doublet
Lattice Method to Nonplanar Configurations in Subsonic Flow,” Journal of Aircraft, Vol. 8,
No. 6, June 1971, pp. 406 – 413.

46. Küssner, H. G., “General Airfoil Theory”, NACA – TM – 979, 1940.
47. Küssner, H. G., “General Lifting Surface Theory”, Luftfahrtförschung, Vol. 17,
1940, pp. 370-378.
48. Landahl, M. T., Unsteady Transonic Flow, Pergamon Press, New York, 1951
49. Landahl, M., T., and Stark, V.J.E., “Numerical Lifting Surface Theory –
Problems and Progress,” AIAA Journal, Vol. 6, No. 11, Nov. 1968, pp. 2049 – 2000.
50. Laschka, B., ‘Zur Theorie der Harmonishe Schwindingen Tragenden Flache
bei Unterschallanstromung”, Zetschrift für Flugwissenschaften, Vol. 11, No. 7, July 1963,
pp.265-292.
51. Lee-Rausch, E. M., and Batina, J., “Calculation of AGARD Wing 445.6 Flutter
Using Navier-Stokes Aerodynamics,” Paper AIAA-93-3476, In Proceedings. of the 11th
AIAA Applied Aerodynamics Conference, Monterey, California, 1983
52. Lessing, H. C., Troutman, J. L., and Menees, G. P., “Experimental
Determination on a Rectangular Wing Oscillating in the First Bending Mode for Mach
Numbers from 0.24 to 1.30,” NASA TND-33, December 1960.
53. Liu, D. D, “Lifting Surface Theory and Panel Methods”, Lecture Notes,
C.S.I.R., May, 1988.
54. Liu, D. D, and Winther, B. A., “Towards a Mixed Kernel Function Approach
for Unsteady Transonic Flow Analysis”, Presented at the AGARD Symposium on Unsteady
Aerodynamics, Ottawa, CA, September 26-28, 1977 ( Paper No. 12, AGARD-CPP-227).
55. Liu, D. D., “A Mixed Kernel Function Method for Unsteady Transonic Flow
Calculations”, Northrop Tech-Brief PC-20-59, March, 1976.
56. Liu, D. D., “Panel Method Lectures”, Lecture Notes, Arizona State University,
Tempe, Arizona, 1990.
57. Liu, D. D., “Computational Transonic Equivalent Strip Method for
Applications to Unsteady 3D Aerodynamics,” AIAA-83-0261, In: Proceedings of the 21st
AIAA Aerospace Sciences Meeting, Reno, Nevada, January 10-13, 1983.
58. Liu, D.D., Chen, P.C., Yao, Z.X., and Sarhaddi, D., “Recent Advances in
Lifting Surface Methods,” the Aeronautical Journal of the Royal Aeronautical Society, Vol.
100, No. 958, Oct. 1996, pp. 327-339.
59. Liu, D. D., and Hui, W.H. " Oscillating Delta Wings with Attached Shock
Waves" AIAA Journal., vol. 15, No 6, June, 1977, pp 804-811.
60. Liu, D. D., Kao, Y. F., and Fung, K. Y., “An Efficient Method for Computing
Unsteady Transonic Aerodynamics of Swept Wings with Control Surfaces”, Journal of
Aircraft, Vol. 25, No. 1, 1988, pp. 25-31.
61. Luber, W. Schmid, H., "Flutter Investigations in the Transonic Flow Regime
for a Fighter Type Aircraft", AGARD RN 703, September 1982.
62. Mabey, D., “The Physical Phenomena Associated with Unsteady Transonic
Flow”, in “Unsteady Transonic Aerodynamics”, Edited by David Nixon, American Institute
of Aeronautics and Astronautics, Washington, D.C., 1989, Chapter 1, pp. 1-55.
63. Malone, J.B., and Ruo, S.Y., “LANN Wing Test Program: Acquisition and
Application of Unsteady Transonic Data for Evaluation of Three-Dimensional Computational

Methods,” Air Force Flight Dynamics Laboratory, AFWAL-TR-83-3006, Wright Patterson
Air Force Base, Ohio, February 1983.
64. McCain, W. E., “Measured and Calculated Airloads on a Transport Wing
Model”, Journal of Aircraft, Vol. 22, No. 4,1985 .
65. Mello, O. A. F., and Azevedo, J. L. F., “Simulation of High Angle-of-Attack
Flow over a Hemisphere-Cylinder”, In: Proceedings of the 8th Brazilian Congress of
Engineering and Thermal Sciences, ENCIT 2000 -, Porto Alegre, Rio Grande do Sul, 2000.
66. Mello, O. A. F., and Sankar, L. N., “Computation of Unsteady Transonic Flow
over a Fighter Wing Using a Zonal Navier-Stokes/Full-Potential Method,” International
Journal for Numerical Methods in Fluids, Vol. 29, No. 5, March 1999, pp. 575-585
67. Mello, O., A., F., “An Improved Hybrid Navier-Stokes/Full-Potential Method
for Computation of Unsteady Compressible Viscous flows”, PhD Thesis, Georgia Institute of
Technology, November, 1994
68. Miles, J.W., “Potential Theory of Unsteady Supersonic Flow,” Cambridge
Univ. Press, London, 1959.
69. Palacios, R., Climent, H., Karlsson, A., and Winzell, B., “Assessment of
Strategies for Correcting Linear Unsteady Aerodynamics Using CFD or Test Results”, Proc.
of the CEAS/AIAA International Forum on Aeroelasticity and Structural Dynamics, Madrid,
Spain, June 2001, pp. 195-210
70. Pitt, D.M., and Goodman, C.E., “Flutter Calculations Using Doublet Lattice
Aerodynamics Modified by the Full Potential Equations”, Proc. of the 28th AIAA/
ASME/ASCE/AHS Structures, Structural Dynamics, and Materials Conference, Paper AIAA-
87-0882-CP, Monterey, California, USA, 1987.
71. Pulliam, T. H., and Chaussee, D. S., “A Diagonal Form of an Implicit
Approximate-Factorization Algorithm”, Journal of Computational Physics, Vol. 39, 1981,
pp.347-363.
72. Pulliam, T. H., and Steger, J. L., “Implicit Finite-Difference Simulations of
Three-Dimensional Compressible Flow”, AIAA Journal, Vol. 18, 1980, pp. 159-167.
73. Rodden, W. P., and Revell, J. D., “Status of Unsteady Aerodynamic Influence
Coefficients”, Institute of Aeronautical Sciences, Fairchild Fund Paper FF-33, 1962.
74. Rodden, W. P., Giesing, J. P., and Kalman, T. P., “New Developments and
Applications of the Subsonic Doublet Lattice Method for Non-Planar Configurations”,
AGARD Symposium – Unsteady Aerodynamics for Aeroelastic Analysis of Interfering
Surfaces, Paper no 4, Tonsberg, OsloFjorden, Norway, November 4-3, 1970. (see also
Douglas Paper 5826)
75. Rodden, W. P., and Bellinger, E. D., “Aerodynamic Lag Functions, Divergence
and the British Flutter Method,” Journal of Aircraft, Vol. 19, No. 7, July 1982.
76. Rodden, W. P., and Bellinger, E. D., “Unrestrained Aeroelastic Divergence in a
Dynamic Stability Analysis,” Journal of Aircraft, Vol. 19, No. 9, September 1982.
77. Rodden, W. P., Harder, R.L., and Bellinger, E., D., “Aeroelastic Addition to
NASTRAN,” NASA CR 3094, March 1979.
78. Rodden, W. P., and Johnson, E. H., “MSC/NASTRAN Aeroelastic Analysis -
User’s Guide Version 68”, The MacNeal-Schwendler Corporation, 1994, USA.

79. Rodden, W. P., Giesing, J. P., and Kalman, T.P., “New Development and
Applications of the Subsonic Doublet-Lattice Method for Nonplanar Configurations,”
AGARD Conf. Proc. No. 80-71, Part II, No. 4, 1971.
80. Rodden, W.P., Giesing, J.P., and Kalman, T.P., “Refinement of the Nonplanar
Aspects of the Subsonic Doublet Lattice Lifting Surface Method,” Journal of Aircraft, Vol. 9,
Jan 1972, pp. 69 – 73.
81. Rodden, W.P., Taylor, P.F., and McIntosh, S.C., Jr., “Further Refinement of
the Subsonic Doublet Lattice Method,” Journal of Aircraft, Vol. 35, No. 5, 1998, pp. 720–
727.
82. Rodden, W.P., Taylor, P.F., McIntosh, S.C., Jr., and Baker, M.L, “Further
Convergence Studies of the Enhanced Subsonic Doublet Lattice Oscillatory Lifting Surface
Method,” In: Proceedings of the International Forum on Aeroelasticity and Structural
Dynamics, Vol. III, Rome, Italy, 1997, pp. 401-408.
83. Rumsey, C., Biedron, R., Thomas, J., “CFL3D: Its History and some Recent
Applications”, NASA-TM 112861, May, 1997; presented at the “Godunov’s Method for Gas
Dynamics” Symposium, Ann Arbor, MI, May, 1-2, 1997.
84. Sankar, L. N, Bharadvaj, B. K., and Tsung, F.-L., “A Three Dimensional
Navier-Stokes/Full-Potential Coupled Analysis for Viscous Transonic Flows,” AIAA Journal,
Vol. 31, No. 10, October 1993 .
85. Sankar, L. N., and Kwon, O. J., “High-Alpha Simulation of Fighter Aircraft,”
Proceedings of the NASA High Angle-of-Attack Technology Conference, Vol. 1, NASA
Langley Research Center, Hampton, VA, NASA CP-3149, Pt. 2, 1990, pp. 689–702
86. Schulze, E., and Vogel, S., “Comparison of Two Potential Flow Methods for
Transonic Flutter Analysis”, Rept - 89-005, 1989
87. SenGupta, G., “Evaluation of the Dau-Garner Method for Predicting Unsteady
Pressures in Transonic Flow”, Paper presented at the International Symposium on
Aeronautical Science & Technology, Jakarta, Indonesia, June 24-27, 1996 – Paper ISASTI-
96-1.3.4
88. Silva, R. G. A., “State Space Aeroelastic Analysis Applied to Fixed Wing
Aircraft”, [“Análise Aeroelástica no Espaço de Estados Aplicada a Aeronaves de Asa Fixa”],
Master Thesis, University of São Paulo, São Carlos, 1994. (in Portuguese).
89. Silva, R. G. A., and Mello, O. A. F., “ Prediction of Transonic Flutter Using
NASTRANTM with Aerodynamic Coefficients Tuned to Navier-Stokes Computations” ,
CEAS/AIAA/ICASE/NASA Langley International Forum on Aeroelasticity and Structural
Dynamics, Williamsburg, Virginia, 1999.
90. Silva, R. G. A., Mello, O. A. F., and Azevedo, J. L. F “A Navier-Stokes Based
Study into Linearity in Transonic Flow fort Flutter Analysis, AIAA Paper 2002-2971, In:
Proceedings of the 32nd AIAA Fluid Dynamics Conference, St. Louis, USA, June 2002.
91. Silva, R. .G. A., Mello, O. A. F. and, Azevedo, J. L. F. , ”A Numerical Study
into Shock Displacement Effects for Aeroelastic Analysis in Transonic Flow” In: Proceedings
of the 9th Brazilian Congress of Thermal Engineering and Sciences - ENCIT 2002 ,
Caxambú, Minas Gerais, October 2002.
92. Silva, R. G. A., Mello, O. A. F., and Azevedo, J. L. F., “Transonic Flutter
Calculations Based on Assumed Mode Shapes Corrections”, In: Proceedings. of the

CEAS/AIAA International Forum on Aeroelasticity and Structural Dynamics, Madrid, Spain,
2001, pp. 183-194.
93. Soviero, P. A. O., and Bortolus, M. V., “Generalized Vortex Lattice Method
for Oscillating Lifting Surfaces in Subsonic Flow”, AIAA Jounal, Vol. 30, No. 11, 1992, pp.
2723-2729.
94. Spalart, P., and Allmaras, S., “A One Equation Turbulence Model for
Aerodynamic Flows”, AIAA-Paper-92-0439, In: Proceedings of the AIAA 30th Aerospace
Sciences Meeting and Exhibit, Reno, Nevada, January 6-9, 1992.
95. Suciu, E., “A Comparison Between the Aerodynamic Pressures Factoring and
the Aerodynamic Derivatives factoring Methods for the Doublet lattice Program”, In:
Proceedings of the 3rd Worldwide Aerospace Conference and technology Showcase ,
September, 24-26, Toulouse, France, 2001.
96. Suciu, E., “A General Aerodynamic Derivatives Factoring Method for the
MSC.Nastran DLM Capable of Controlling Al Lifting Surfaces Aerodynamic Forces and
Moments, Including all Interference Effects”, Presented at the Aerospace Flutter and
Dynamics Council Spring 2003 Meeting, Dayton, Ohio, 7-9 May, 2003
97. Suciu, E., Glaser, J., and Coll, R., “Aerodynamic Derivatives Factoring
Scheme for the MSC/NASTRAN Doublet Lattice Program”, presented at the
MSC/NASTRAN Worlds Users’ Conference, Los Angeles – CA ,1990.
98. Sulaeman, E., “Effect of Compressive Force on Aeroelastic Stability of a Strut-
Braced Wing”, PhD thesis, Virginia Polytechnic Institute and State University, Blacksburg,
Virginia, USA, November 2001.

99. Theodorsen, T., “General Theory of Aerodynamic Instability and the
Mechanism of Flutter,” NACA Report No. 496, 1940.
100. Thomas, J. P., Dowell, E. H., and Hall, K. C., “A Harmonic Balance Approach
for Modeling Three-Dimensional Non-Linear Unsteady Aerodynamic and Aeroelasticity”,
Proceedings of the ASME International Mechanical Engineering Conference and Exposition,
November, 17-22, 2002, New Orleans, Louisiana, USA. Paper No. IMECE-2002-32535.
101. Thomas, J. L., Krist, S. L., and Anderson, W. K., “Navier-Stokes
Computations of Vortical Flows over Low-Aspect-Ratio Wings”, AIAA Journal, Vol. 28, No,
2, pp. 205-212, 1990.
102. Tijdeman, H., van Nunen, J. W. G., Kraan, A. N., Persoon, A. J., Poestkoke,
R., Roos, R., Schippers, P., and Siebert, C. M., “Transonic Wind Tunnel Tests on an
Oscillating Wing with External Stores, Part II: Clean Wing,” Air Force Flight Dynamics
Laboratory, AFFDL-TR-78-194, Part II, Wright-Patterson AFB, Ohio, 1979. (Also NLR-TR-
78106U, Part II)
103. Ueda, T., and Dowell, E. H., “A New Subsonic Method for Lifting Surfaces in
Subsonic Flow,” AIAA Journal, Vol. 20, No. 3, March 1982, pp. 348 – 355.
104. Van Zyl, L.H., “Arbitrary Accuracy Integration scheme for the Subsonic
Doublet Lattice Method,” Journal of Aircraft, Vol. 35, No. 6, 1998, pp. 975-977.
105. Van Zyl, L.H., “Application of the Subsonic Doublet Lattice Method to Delta
Wings,” Journal of Aircraft, Vol. 36, No. 3, 1999, pp. 609-611.
106. Vepa, R., “Finite State Modeling of Aeroelastic System”, NASA CR-2779,
Feb. 1977.
107. Watkins, C. E., Runyan, H. L., and Woolston, D. S., “On the Kernel Function
of the Integral Equation Relating the Lift and Downwash Distributions of Oscillating Finite
Wings in Subsonic Flow”, NACA Report – 1234, 1955, pp. 1-11.
108. Whitlow, W., Jr., “Computational Unsteady Aerodynamics for Aeroelastic
Analysis,” NASA TM-100523, December, 1987.
109. Winzel, B., "Recent Applications of Linear and Nonlinear Unsteady
Aerodynamic Aerodynamics for Aeroelastic Analysis", AGARD CP 507 - Transonic
Unsteady Aerodynamics and Aeroelasticity, San Diego, California, October 7-11, 1991
110. Yates Jr., E. C., “AGARD Standard Aeroelastic Configurations for Dynamic
response I-Wing 445.6”, AGARD Report No. 765, 1988.
111. Yates, E. C. Jr., “Modified Strip Analysis Method for Predicting Wing Flutter
at Subsonic to hypersonic Speeds”, Journal of Aircraft, Vol. 3, No. 1, Jan-Feb, 1966
112. Yates, E.C., Jr., Land, N.S., and Foughner, J.T., Jr., “Measured and Calculated
Subsonic and Transonic Flutter Characteristics of a 45º Sweptback Wing Planform in Air and
in Freon-12 in the Langley Transonic Dynamics Tunnel,” NASA TN-D-1616, Mar. 1963.
113. Yonemoto, K., "A practical method predicting transonic wing flutter
phenomena", 14th International Council of the Aeronautical Sciences Congress (ICAS-1984),
Toulouse, France, 1984, pp 724-732.
114. Zona Technology, “ZAERO Software System- Theoretical Manual – V. 6.5”,
Scottsdale, Arizona, October, 2003.
115. Zwaan, R. J., “Verification of Calculation Methods for Unsteady Airloads in
the Prediction of Transonic Flutter”, Journal of Aircraft, Vol. 22, No. 10, 1985.
APPENDIX A
GIESING´S FORCE MATCHING METHOD
The procedure to be revisited herein is the force matching method presented by
Giesing et al.(1976). The objective of this method is to perform the correction of the AIC
matrix to match experimental loading distributions of the lifting surface under investigation.
This method was originally proposed based on the requirement of a safe prediction of the
control surface flutter.
This method assumes that there is a weighting function operator which matches the
experimental pressures based on the theoretical ones as:
{∆C p }e = Wp  {∆C p } . (A.1)
In many cases what is available is experimental loads instead of the pressure distributions. In
such case, the experimental loads can be written as follows:
{Ce } = [ S ]{∆C pe } , (A.2)
where {Ce } is the load vector and [S] the integration matrix which relates the pressures to
the loads. Replacing (A1) in (A2) yields:
{∆Ce } = [ S ] W p  {∆C p } . (A.3)
This equation will be solved for the correction matrix [Wp].
As there are many more unknowns than equations, the way to solve this system is
using the Least Squares Method. Considering that the deviation can be written as :
{ε p } = W p − I  . (A.4)
The weighted sum of the square of the deviation {ε p } is:

Appendix A - Giesing´s Force Matching Method. 200
= {ε p } Tp  {ε p }
H
∑T ε 2
p p , (A.5)
shall be a minimum, and the error frictional fp is defined as:
1
{ε p } Tp  {ε p }
H
fp = . (A.6)
2
Replacing eqn. (A4) into eqn. (A1), the loads vector can be written as:
{Ce } = [ S ]{∆C p } + [ S ] ∆C p  {ε p } = {C } + [ S ] ∆C p  {ε p } . (A.7)
One should note that the integration matrix [S] relates pressure to loads, and these
loads can be either panel or strip loads. In a practical sense, the experimental loading data is
acquired in a stripwise way; the integration matrix needs to be properly defined as an operator
which relates panel pressures to strip loads. Then, rewriting the pressures vector as a diagonal
matrix, integrated load matrix will be defined as:
 S p  = [ S ]  ∆C p  . (A.8)
   
Recalling equation (A7), one should note that
{∆Ce } = {Ce } − {C } , (A.9)
{∆Ce } =  S p  {ε p } . (A.10)
since the variation of the error function as:
δ f p = {ε p } Tp  {δε p }

H
, (A.11)
the variation of the experimental constraints is
{δ∆Ce } =  S p  {δε p } = 0 . (A.12)
One should note that equations (A11) and (A12) compose a linear system of equations,
δ f p + {λ p } {∆Ce } = 0
H
, (A.13)
with the vector {λ p} as the associated Lagrange multipliers. Thus, rewriting this system of
equations substituting (A11) and (A12) into (A13) we get:

Appendix A - Giesing´s Force Matching Method. 201
(Tp ) + {λ p }  S p  {δε p } = 0
H H
δε p  , (A.14)
   
as δε p is an arbitrary non zero value, we have:
{λ p } = 0
H H
 ε p  Tp  +  S p  , (A.15)
     
or
{λp } = − Tp  ε p  .

H
S p  . (A.16)
 
Recalling equation (A10) given by {∆Ce } =  S p  {ε p } , the Lagrange rule replace is given by:
{λ p } = −( S p  Tp  )

−1 H −1
S p  {∆Ce } , (A.17)
 
and the {Wp} is obtained from

−1
{ε p } = − Tp  {λ p }
H
S p  , (A.18)
 
W p  = [ I ] +  ε p  . (A.19)
   
The weighting function Tp is arbitrary, at least near to the positive. The authors suggested in
their work that
Tp  = [ AIC ][ I ] , (A.20)

 
with [Wp] matrix defined, it is possible to restore the loads, but not necessarily the pressures.
APPENDIX B
THE DAU-GARNER METHOD
The Dau-Garner method (Dau, 1992, and Garner, 1977) is a procedure whose main
concern is to look for a rational form to compute the transonic unsteady pressure amplitudes
and phases. This method is one of the most employed in the aeronautical industry. However
here the objective is to show mathematically its failing with regard to the pressures phase
computation.
This method is based on a set of assumptions added by semi-empirical hypothesis,
which were introduced by Garner (1977). It is assumed that the flow around the lifting surface
can be described with sufficient accuracy by perturbation velocity potentials, and adiabatic
relationships. These assumption are valid for the cases of transonic flows with weak shock
waves. The displacement amplitudes are sufficiently small, thus the time dependent
perturbations of the velocities are much smaller that the steady state streamwise component of
the velocity, in the “x” direction. Furthermore, it is assumed that the relationship between the
velocity potentials, remains the same whether the flow is transonic or purely subsonic. Thus,
the ratio between the unsteady and steady gradients of the velocity potential should be written
as:
φ xnl ( ik ) φ x ( ik )
R ( ik ) = = . . (B.1)
φ xnl ( 0 ) φ x ( 0 )
And finally, the ratio of the quasi-steady transonic pressures to its corresponding linear
pressures is the same for all comparable vibration modes of the lifting surface.
Appendix B - The Dau-Garner Method 203
With the aid of these assumptions and considering the Euler momentum equation it is
possible to write the unsteady pressure distribution as a function of the velocity potential
gradients, after linearization for small disturbances, leading to:
C pnl ( ik ) = −2 (φ xnl ( ik ) + ikφ nl ( ik ) ) , (B.2)
where C pnl ( ik ) and φ xnl ( ik ) are the complex amplitudes of the first harmonic of the pressure
and velocity potential gradient, respectively. The complete derivation of Eq. (B.2), departing
from the Euler momentum equations is found in the work of Dau (1992). Recalling the Euler
momentum equation, one should note that the steady pressures C pnl ( 0 ) are related to thee
steady velocity potential φ xnl ( 0 ) as:
C pnl ( 0 ) C pnl ( 0 )
φ (0) = −
nl
x ⇒ φ (ik ) = −
nl
x R(ik ) . (B.3)
2 2
Looking at Eq. (E.2), is possible to see that the computation of the unsteady pressure
distribution should be performed from the following relationship:

X
( ik ) = R ( ik ) C ( 0 ) + ik ∫ R ( ik ) C pnl ( 0 )dξ
nl nl
C p p . (B.4)
X L. E .
Finally, the unsteady pressures computation is performed by the Dau-Garner method (Dau,
1992) satisfying relation (B.4).
The failing with regard to the pressures phase computation, identified by SenGunpta
(1996), has as explanation. Introducing the approximation of the potential for low reduced
frequencies as φ = φR + ikφ I , that is, a simple first order expansion series, the unsteady
pressure relationship between the potential and the pressures can be written as
C pnl ( ik ) = −2 φ xnlR + ikφ xnlI + ik (φRnl + ikφInl )  ⇒

(B.5)
( )
C pnl ( ik ) = −2 φ xnlR + ikφ xnlI + ikφ Rnl − k 2φInl  = −2 φ xnlR + ik φ xnlI + φ Rnl 
 
Appendix B - The Dau-Garner Method 204
∫
where φR = φ xR dx . That is, the out of phase component contribution vanishes because the
low reduced frequency value and the out of phase correction is simply reduced to a phase shift
of 180 degrees, since the velocity potential gradient ratio does not change the imaginary part
of the potential. This is probably the main cause of the pressure phases differences detected
by SenGupta (1996), in about 50 degree.

FOLHA DE REGISTRO DO DOCUMENTO
1. 2. 3. 4.
CLASSIFICAÇÃO/TIPO DATA DOCUMENTO N° N° DE PÁGINAS
TD 02 de Julho de 2004 CTA/ITA-IEA/TD-001/2004 233

5.
TÍTULO E SUBTÍTULO:
A study on correction methods for aeroelastic analysis in transonic flow
6.
AUTOR(ES):
7. INSTITUIÇÃO(ÕES)/ÓRGÃO(S) INTERNO(S)/DIVISÃO(ÕES):
Instituto Tecnológico de Aeronáutica. Divisão de Engenharia Aeronáutica – ITA/IEA
8.
PALAVRAS-CHAVE SUGERIDAS PELO AUTOR:
1. Aeroelasticidade . 2. Aerodinâmica não estacionária. 3. Escoamento transônico. 4. Métodos de

correção. 5. Aerodinâmica computacional.
9.PALAVRAS-CHAVE RESULTANTES DE INDEXAÇÃO:
Aeroelasticidade; Aerodinâmica não-estacionária; Escoamento transônico; Modelos matemáticos;
Correção; Dinâmica dos fluidos computacional; Sustentação aerodinâmica; Mecânica dos fluidos; Física
10.
APRESENTAÇÃO: X Nacional Internacional
ITA, São José dos Campos, 2004, 233 páginas
11.
RESUMO:
The work presents a study of correction techniques to compute unsteady transonic pressure
distributions and aeroelastic stability in this flow regime. The methodologies herein investigated are
based on corrections of pressure distributions by the weighting of the lifting surface self-induced
downwash, resulting from aeroelastic structural displacements or prescribed motions. A number
approaches were investigated.
An investigation into the linear/nonlinear behavior of unsteady transonic flows was also conducted. It
was concluded from such investigation that unsteady transonic flows present a linear behavior with
respect to small aeroelastic structural displacements around a steady nonlinear mean flow. Such behavior
is the basis for further development of downwash correction methods.
The correction of pressure distributions through the weighting of the lifting surface self-induced
downwash is also known as downwash weighting method. This method has been enhanced leading to a
new downwash correction technique. The procedure may be divided in two steps, where the first step is a
nonlinear steady mean flow correction, with nonlinear pressure differences considered as reference
conditions to correct the self induced downwash. The second step is the correction of the unsteady
component of the downwash, where the corresponding reference unsteady pressure differences are
predicted by a linear aerodynamic model, based on the potential flow equations.
This extended downwash correction method led to a rational formulation named as “successive
kernel expansion method” (SKEM). The unsteady pressures and aeroelastic stability boundaries
computations using such method led to good agreement with experimental measurements. This procedure
is a rapid form to compute the transonic flutter speed boundaries, compared to computational
aeroelasticity and experimental techniques.
12.
GRAU DE SIGILO:
(X ) OSTENSIVO ( ) RESERVADO ( ) CONFIDENCIAL ( ) SECRETO

SILVA, R. G. A., A Study On Correction Methods For Aeroelastic Analysis in Transonic Flow.

Enviado por

Dados do documento

Descrição original:

Título original

Direitos autorais

Formatos disponíveis

Compartilhar este documento

Compartilhar ou incorporar documento

Opções de compartilhamento

Você considera este documento útil?

Este conteúdo é inapropriado?

Direitos autorais:

Formatos disponíveis

SILVA, R. G. A., A Study On Correction Methods For Aeroelastic Analysis in Transonic Flow.

Enviado por

Direitos autorais:

Formatos disponíveis

Thesis presented to the Faculty of the Division of Graduate Studies of the

Technological Institute of Aeronautics in partial fulfillment of the

requirements for the Degree of Doctor in Science in the Course of

Aeronautical and Mechanical Engineering, Area of Aerodynamics,

Propulsion and Energy.

Roberto Gil Annes da Silva

A STUDY ON CORRECTION METHODS FOR

AEROELASTIC ANALYSIS IN TRANSONIC FLOW

Tese de doutorado– Curso de Engenharia Aeronáutica e Mecânica, e área de Aerodinâmica, Propulsão

1. Aeroelasticidade . 2. Aerodinâmica não estacionária. 3. Escoamento transônico. 4. Correção. 5.

NOME DO AUTOR : Roberto Gil Annes da Silva

É concedida ao Instituto Tecnológico de Aeronáutica permissão para reproduzir cópias desta

AEROELASTIC ANALYSIS IN TRANSONIC FLOW

Roberto Gil Annes da Silva

Thesis Committee Composition:

Prof. Dr. Maher Nasr Bismark-Nasr .........................................................Chairperson - ITA

Prof. Dr. João Luiz Filgueiras de Azevedo .....................................................Advisor - ITA

Prof. Dr. Olympio Achilles de Faria Mello .....................................................Advisor - IAE

Prof. Dr. Donizeti de Andrade .......................................................................Member - ITA

Prof. Dr. Armando Miguel Awruch ..........................................................Member - UFRGS

Prof. Dr. Flávio Donizeti Marques ................................................................Member - USP

To my wife Michelle and my children Julia and Pedro.

are deeply acknowledge

best conditions during my stay in Arizona.

I would like to express my gratitude to my colleagues Major Antonio Carlos Ponce

development this work

And finally I would to express my profound gratitude to my parents, Arnaldo and

Irene and to my wife Michelle, who guided me with encouragement.

The work presents a study of correction techniques to compute unsteady transonic

surface self-induced downwash, resulting from aeroelastic structural displacements or

prescribed motions. A number approaches were investigated.

An investigation into the linear/nonlinear behavior of unsteady transonic flows was

corresponding reference unsteady pressure differences are predicted by a linear aerodynamic

model, based on the potential flow equations.

speed boundaries, compared to computational aeroelasticity and experimental techniques.

O trabalho apresenta um estudo sobre técnicas de correção para calcular distribuições

de pressão não lineares e não estacionárias, assim como a estabilidade aeroelástica em

escoamentos transônicos. As metodologias investigadas neste trabalho são baseadas na

correção de distribuições de pressão através da ponderação das velocidades normais (ou

downwash) induzidas pelos deslocamentos estruturais de natureza aeroelástica, ou devido a

movimentos prescritos. Diferentes formas de aproximação foram investigadas.

Investigou-se o comportamento linear/não-linear de escoamentos transônicos não

estacionários. Conclui-se de tal estudo que os escoamentos transônicos não estacionários

apresentam um comportamento linear, com respeito às pequenas deformações estruturais de

natureza aeroelástica, ao redor de um escoamento estacionário não linear. Tais observações

relacionadas à este comportamento serão base para os desenvolvimentos adicionais dos

A correção das pressões através da ponderação do vetor de velocidades normais

induzidas é também conhecido como um método de ponderação do downwash. Este método

foi aperfeiçoado, levando a um novo procedimento de correção de downwash. Tal

consideradas condições de referência para corrigir o downwash. O segundo passo consiste na

Esta extensão do método de correção do downwash resulta em uma formulação

racional, denominada como “método de expansão sucessiva do kernel” (successive kernel

cálculo de pressões não estacionárias quanto à estabilidade aeroelástica, em concordância com

resultados experimentais. Este procedimento é uma forma rápida de calcular as fronteiras de

estabilidade aeroelástica no regime transônico, quando comparado com métodos de

aeroelasticidade computacional ou técnicas experimentais.

List of Figures …………………………………………………………..……………… IV

List of Tables ….…………………………………...…………………………….….… XII

List of .Symbols ………………………………………………………………………. XIII

Chapter 1 – Introduction …………….………………………………..……………...…... 1

1.1 – Background ...........…….………………………………………………………. 1

1.1.1 Aeroelastic Stability Behavior in Transonic Flow................................ 1