Structural Dynamics and

Aeroelasticity
Notes from prof. Quaranta’s lessons

Jacopo Alberti, Luigi Maria Boccia and Lorenzo Lucatello

2023 - 2024
Preface
Questa dispensa è stata scritta sulla base degli appunti presi a lezione.
Non è in alcun modo pensata per poter sostituire un libro di testo o il seguire le lezioni
né se ne assicura la totale correttezza.

Questa dispensa viene distribuita gratuitamente e non deve essere venduta.

Nel caso di errori da correggere è possibile scrivere a [email protected] o

[email protected]

This book has been written using our notes taken in class.
This is not meant to substitute the following classes or the book Professor Quaranta
suggested.

This book is free. It’s not meant to be sold.

If you find errors please write at [email protected] or [email protected]

First edition, 2024

Second edition, 2024. Correction of errors in 2.5.3, 3.4, 3.5.3, 4.1.1, 4.2

0 Structural Dynamics and Aeroelasticity

Index
1 Introduction 8
1.1 What is structural dynamics? . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.2 What is Aeroelasticity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.3 Divergence and Flutter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.3.1 Divergence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.3.2 Flutter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.4 Brief history of Aeroelasticity . . . . . . . . . . . . . . . . . . . . . . . . . 10

2 Static Aerolasticity 11
2.1 Rigid wing model: typical section . . . . . . . . . . . . . . . . . . . . . . . 11
2.1.1 Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.2 Divergence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.2.1 Equilibrium approach . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.2.2 Comparison of aeroelastic and rigid lift . . . . . . . . . . . . . . . 13
2.2.3 Torsional divergence as a stability problem . . . . . . . . . . . . . 14
2.2.4 Eigenvalue approach problem . . . . . . . . . . . . . . . . . . . . . 14
2.2.5 Divergence and Mach number . . . . . . . . . . . . . . . . . . . . . 15
2.3 Control surface effectiveness . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.3.1 Control Elastic efficiency and control reversal . . . . . . . . . . . . 16
2.3.2 Plot control surface elastic efficiency . . . . . . . . . . . . . . . . . 16
2.3.3 Control Surface Load Distribution . . . . . . . . . . . . . . . . . . 17
2.3.4 Matrix form of control reversal problem . . . . . . . . . . . . . . . 17
2.4 Control surface stiffness . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.5 Rolling of a straight wing . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.5.1 Aerodynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.5.2 Twist . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.5.3 Control reversal problem . . . . . . . . . . . . . . . . . . . . . . . . 22
2.5.4 Consistent performance problems . . . . . . . . . . . . . . . . . . . 22

3 Multi degree of freedom torsional divergence model 23

3.1 Principle of virtual works - PVW . . . . . . . . . . . . . . . . . . . . . . . 23
3.2 MDOF model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
3.2.1 Power method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.3 Equivalent stiffness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.4 Integral formulation - Flexibility influence functions . . . . . . . . . . . . 25
3.5 Slender wing: analytical solution for torsion . . . . . . . . . . . . . . . . . 27
3.5.1 Wing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.5.2 Solution of the homogeneous . . . . . . . . . . . . . . . . . . . . . 28
3.5.3 Solution of the forced problem . . . . . . . . . . . . . . . . . . . . 29
3.6 Ritz-Galerkin approssimation . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.6.1 Requirements for Ritz in weak formulation . . . . . . . . . . . . . 33

1 Structural Dynamics and Aeroelasticity

4 Swept wings 34
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
4.1.1 Reason to use a sweep angle . . . . . . . . . . . . . . . . . . . . . . 34
4.1.2 Torsion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
4.1.3 Bending . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
4.1.4 Total change of angle of attack . . . . . . . . . . . . . . . . . . . . 36
4.1.5 Forces and moments . . . . . . . . . . . . . . . . . . . . . . . . . . 36
4.2 Divergence - typical section . . . . . . . . . . . . . . . . . . . . . . . . . . 37
4.3 Lift effectiveness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
4.4 Control reversal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
4.5 Tailoring . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
4.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

5 Dynamic Aeroelasticity: introduction to flutter 42

5.1 Homogenous problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
5.1.1 Associated eigenvector . . . . . . . . . . . . . . . . . . . . . . . . . 43
5.2 Steady aerodynamics forces . . . . . . . . . . . . . . . . . . . . . . . . . . 43
5.2.1 Logarithmic decrement . . . . . . . . . . . . . . . . . . . . . . . . . 44
5.2.2 Stability of the typical section with steady aero . . . . . . . . . . . 45
5.2.3 Limit of stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
5.2.4 Divergence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
5.2.5 Steady model: graphics . . . . . . . . . . . . . . . . . . . . . . . . 46
5.2.6 Effect of phase shift . . . . . . . . . . . . . . . . . . . . . . . . . . 47
5.3 Quasi-Steady aerodynamics forces . . . . . . . . . . . . . . . . . . . . . . 47
5.3.1 Quasi-steady: graphics . . . . . . . . . . . . . . . . . . . . . . . . . 48
5.3.2 Quasi-steady model: pitch rate . . . . . . . . . . . . . . . . . . . . 49
5.4 Unsteady aerodynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
5.4.1 Wave length . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51

6 Structural dynamics 52
6.0.1 Continuous deformable structure: hypothesis . . . . . . . . . . . . 52
6.1 Introduction and beam modeling . . . . . . . . . . . . . . . . . . . . . . . 52
6.2 Euler-Bernoulli’s beam model . . . . . . . . . . . . . . . . . . . . . . . . . 52
6.2.1 Axial load . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
6.2.2 Bending . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
6.2.3 Torsion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
6.3 Timoshenko’s beam model . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
6.4 Free vibration for the beam bending . . . . . . . . . . . . . . . . . . . . . 54
6.5 Free vibration for the beam torsion . . . . . . . . . . . . . . . . . . . . . . 55
6.6 Properties of the modal forms . . . . . . . . . . . . . . . . . . . . . . . . . 55
6.6.1 Orthogonality with respect to mass distribution . . . . . . . . . . . 56
6.6.2 Orthogonality with respect to stiffness distribution . . . . . . . . . 56
6.6.3 Inner product . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
6.7 Modal decomposition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57

2 Structural Dynamics and Aeroelasticity

 6.8 Ritz-Galerkin approximation . . . . . . . . . . . . . . . . . . . . . . . . . 58
6.8.1 Decoupling in Ritz . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
6.8.2 FEM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
6.8.3 Ritz-Gakerkin vs FEM . . . . . . . . . . . . . . . . . . . . . . . . . 59

7 Modal Analysis 60
7.1 State-space form . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
7.2 Normalization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
7.3 Decoupling through eigenmodes . . . . . . . . . . . . . . . . . . . . . . . . 61
7.4 State space form . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
7.4.1 Decoupling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
7.5 Coincident eigenvalues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
7.6 Rigid body modes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
7.7 Gram-Schmidt orthogonalization . . . . . . . . . . . . . . . . . . . . . . . 63
7.8 Computation of the solution to initial conditions . . . . . . . . . . . . . . 63
7.9 Rayleigh quotient . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
7.9.1 Error on the eigenvalue . . . . . . . . . . . . . . . . . . . . . . . . 65
7.10 Computations of a group of eigenvectors & eigenvalues . . . . . . . . . . . 65
7.10.1 Bloch-Stodola block iteration . . . . . . . . . . . . . . . . . . . . . 66

8 Damping models 67
8.1 Kevin-Voigd model (Parallel) . . . . . . . . . . . . . . . . . . . . . . . . . 67
8.2 Maxwell model (Series) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
8.3 Equivalent Damping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
8.4 Energy analysis - meaning of equivalent damping ratio . . . . . . . . . . . 69
8.4.1 Energy dissipation . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
8.5 Modal damping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
8.6 Rayleigh damping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
8.7 Equivalent damping: Coulomb friction . . . . . . . . . . . . . . . . . . . . 71
8.8 Hysteretic damping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71

9 Dynamic Response 73
9.1 Laplace transform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
9.1.1 State-space Format . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
9.2 Frequency response . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
9.2.1 Dynamic stiffness . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
9.2.2 One DoF . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
9.2.3 Modal computation of the response . . . . . . . . . . . . . . . . . . 76
9.2.4 Contribution of different terms . . . . . . . . . . . . . . . . . . . . 77
9.3 Fourier Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
9.3.1 Non-periodic but periodic-sizable signals . . . . . . . . . . . . . . . 78
9.4 Fourier Transform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
9.5 Limited band excitation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
9.6 Inertia Relief . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81

3 Structural Dynamics and Aeroelasticity

 9.7 Anti-resonance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
9.8 Solution through convolution . . . . . . . . . . . . . . . . . . . . . . . . . 83
9.8.1 Convolution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
9.8.2 Response to the impulse . . . . . . . . . . . . . . . . . . . . . . . . 83
9.9 Response using the step response . . . . . . . . . . . . . . . . . . . . . . . 84
9.10 Truncation of the modal basis . . . . . . . . . . . . . . . . . . . . . . . . . 85
9.11 Recovery of internal forces . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
9.11.1 Direct recovery . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
9.11.2 Acceleration modes . . . . . . . . . . . . . . . . . . . . . . . . . . . 86

10 Random Vibrations 88
10.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
10.2 Statistic quantities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
10.2.1 Probability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
10.2.2 Probability distribution . . . . . . . . . . . . . . . . . . . . . . . . 88
10.2.3 Cumulative probability function (CPF) . . . . . . . . . . . . . . . 89
10.3 Probability density function . . . . . . . . . . . . . . . . . . . . . . . . . . 89
10.3.1 Expected values . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
10.3.2 Variance and standard deviation . . . . . . . . . . . . . . . . . . . 89
10.3.3 Gaussian process . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
10.3.4 Gaussian process: joint distributions . . . . . . . . . . . . . . . . . 89
10.4 Process function of time . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
10.4.1 Auto-correlation and cross-correlation . . . . . . . . . . . . . . . . 90
10.4.2 Stationary process . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
10.4.3 Time shift . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
10.4.4 Properties of the correlation . . . . . . . . . . . . . . . . . . . . . . 91
10.5 Ergodic process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
10.5.1 Auto-covariance and Cross-covariance . . . . . . . . . . . . . . . . 92
10.6 Auto-covariance of a difference . . . . . . . . . . . . . . . . . . . . . . . . 92
10.6.1 Auto-covariance as a mean to de-noise a signal . . . . . . . . . . . 93
10.7 Power Spectral Density (PSD) . . . . . . . . . . . . . . . . . . . . . . . . 93
10.7.1 Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
10.8 White noise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
10.8.1 Colored Noise (Band-limited white noise) . . . . . . . . . . . . . . 94
10.9 Exponential correlation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94

11 Random response 95
11.1 Computation of the mean . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
11.1.1 Computation of the mean knowing the mean of the input . . . . . 95
11.1.2 Computation of the mean with the second order formulation . . . 95
11.1.3 State space formulation . . . . . . . . . . . . . . . . . . . . . . . . 96
11.2 Computation of the variance of the output using the impulse response . . 96
11.3 Computation of the response in frequency domain . . . . . . . . . . . . . 96
11.4 White noise approximation . . . . . . . . . . . . . . . . . . . . . . . . . . 97

4 Structural Dynamics and Aeroelasticity

 11.5 Lyapunov Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
11.6 Shape filters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
11.7 Risk assessment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
11.8 Fatigue failure: Rice formula . . . . . . . . . . . . . . . . . . . . . . . . . 101
11.8.1 Computation of the probability of crossing a threshold level b in
a finite time interval T . . . . . . . . . . . . . . . . . . . . . . . . . 102

12 Unsteady aerodynamics 104

12.1 Interface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
12.2 Unsteady aerodynamics phenomena . . . . . . . . . . . . . . . . . . . . . 104
12.3 Spatial Filtering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
12.4 Material derivative . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
12.5 Reynolds transport theorem . . . . . . . . . . . . . . . . . . . . . . . . . . 105
12.6 Euler equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
12.7 Rate of change of kinetic energy . . . . . . . . . . . . . . . . . . . . . . . 106
12.8 Entropy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
12.9 Acceleration and Vorticity . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
12.10Circulation and Kevin’s theorem . . . . . . . . . . . . . . . . . . . . . . . 107
12.11Velocity potential . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
12.12Unsteady Bernoulli equation . . . . . . . . . . . . . . . . . . . . . . . . . . 108
12.12.1 Unsteady Bernoulli . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
12.12.2 Unsteady Bernoulli for adiabatic flow . . . . . . . . . . . . . . . . 108
12.12.3 Pressure coefficient . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
12.13Full potential equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
12.14Kutta condition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
12.15Linearized potential . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
12.15.1 Linearized potential - equation . . . . . . . . . . . . . . . . . . . . 110
12.15.2 Linearized potential - boundary conditions . . . . . . . . . . . . . 111
12.15.3 Linearized potential - pressure coefficient . . . . . . . . . . . . . . 111
12.16Linearized boundary conditions . . . . . . . . . . . . . . . . . . . . . . . . 111
12.17Solution using Biot-Savart . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
12.17.1 Straight vortex in 3D . . . . . . . . . . . . . . . . . . . . . . . . . 112
12.17.2 Line of infinite length vortices . . . . . . . . . . . . . . . . . . . . . 113

13 2D unsteady airfoil theory 114

13.1 Hypothesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
13.2 Boundary conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
13.3 Wake vorticity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
13.4 Method of solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
13.4.1 Change of coordinates . . . . . . . . . . . . . . . . . . . . . . . . . 116
13.4.2 Decomposition of bound vorticity . . . . . . . . . . . . . . . . . . . 117
13.5 Computation of load . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
13.5.1 Loads . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
13.6 Solution in frequency domain . . . . . . . . . . . . . . . . . . . . . . . . . 119

5 Structural Dynamics and Aeroelasticity

 13.6.1 Wake vorticity in frequency domain . . . . . . . . . . . . . . . . . 119
13.6.2 Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
13.7 Theodorsen model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
13.7.1 Theodorsen function . . . . . . . . . . . . . . . . . . . . . . . . . . 121
13.7.2 Lift variation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
13.8 Cicala, Küssner-Schwarz approach for arbitrary deformations . . . . . . . 123
13.9 Wagner approach: time domain solution . . . . . . . . . . . . . . . . . . . 124
13.10Lumped vortex panel method . . . . . . . . . . . . . . . . . . . . . . . . . 125
13.11Frozen gust . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
13.11.1 Frequency domain . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
13.11.2 Gust . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
13.11.3 Sears function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
13.11.4 Final result . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128

14 Morino’s method 129

14.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
14.2 Green’s identity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
14.3 Incompressible, steady flow . . . . . . . . . . . . . . . . . . . . . . . . . . 130
14.3.1 Analysis of the wake . . . . . . . . . . . . . . . . . . . . . . . . . . 131
14.3.2 Discretization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
14.3.3 System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
14.3.4 Computation of loads . . . . . . . . . . . . . . . . . . . . . . . . . 133
14.4 Incompressible, unsteady case . . . . . . . . . . . . . . . . . . . . . . . . . 134
14.4.1 Wake analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
14.4.2 Discretization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
14.4.3 Loads . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
14.5 Compressible flow, steady case . . . . . . . . . . . . . . . . . . . . . . . . 135
14.6 Compressible flow, unsteady case . . . . . . . . . . . . . . . . . . . . . . . 136
14.6.1 Subsonic flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
14.6.2 Supersonic flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
14.6.3 Supersonic flow panelization . . . . . . . . . . . . . . . . . . . . . . 138
14.7 Piston theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138

15 The Aeroelastic problem 140

15.1 Computation of flutter speed . . . . . . . . . . . . . . . . . . . . . . . . . 140
15.1.1 Flutter envelope . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
15.2 Assembly of the aeroelastic system . . . . . . . . . . . . . . . . . . . . . . 141
15.3 Aeroelastic equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
15.4 Quasy-steady approximation . . . . . . . . . . . . . . . . . . . . . . . . . 142
15.4.1 Computation of the QS approximation from the frequency response143
15.4.2 Meaning of QS approximation in time domain . . . . . . . . . . . 143
15.4.3 Problems with this method . . . . . . . . . . . . . . . . . . . . . . 144
15.5 Direct solution for flutter . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
15.6 Flutter: P-K method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144

6 Structural Dynamics and Aeroelasticity

 15.6.1 P-K method algorithm . . . . . . . . . . . . . . . . . . . . . . . . . 145
15.7 Continuation approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146

16 Response problems 148

16.1 Equilibrium manouvres . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
16.2 Dynamic manouvers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
16.2.1 Modeling of a control surface . . . . . . . . . . . . . . . . . . . . . 150
16.2.2 Ideal servo actuator . . . . . . . . . . . . . . . . . . . . . . . . . . 151
16.2.3 Modeling of a servo-controlled surface . . . . . . . . . . . . . . . . 151
16.2.4 Modeling of dynamic compliance surface . . . . . . . . . . . . . . . 152
16.2.5 Complete model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
16.3 Gust response . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
16.4 Discrete gust . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
16.4.1 Stochastic Gust . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154

17 Useful formulas 155

17.1 Formulas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
17.1.1 Computation of variance . . . . . . . . . . . . . . . . . . . . . . . . 155
17.1.2 Lyapunov formula . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
17.1.3 Rice formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
17.1.4 Compressible unsteady Bernoulli theorem . . . . . . . . . . . . . . 155
17.2 Cp through Bernoulli . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
17.2.1 Cp in Morino . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
17.2.2 Linearized BCs (Theodorsen and Morino case) . . . . . . . . . . . 155
17.2.3 Theodorsen model . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
17.3 Theodorsen and HAM Matrix . . . . . . . . . . . . . . . . . . . . . . . . . 156

7 Structural Dynamics and Aeroelasticity

1 Introduction
1.1 What is structural dynamics?
Structural dynamics is concerned with the vibrations and dynamic response of the ele-
ments of a structure.

Aerospace structures differ from other structures due to the combined effect of the re-
quest to be lightweight and at the same time extremely performing. This, as a conse-
quence, leads to thin-walled structures, often based on the usage of composite or other
multi-functional materials, characterized by large flexibility.

1.2 What is Aeroelasticity

Aeroelasticity is the study of the dynamics of such structures considering the influence
of the fluid flows that surround them. It studies the interactions between the inertial,
elastic, and aerodynamic forces occurring while an elastic body is exposed to a fluid flow.
The most general phenomena are dynamic by nature but static aeroelastic phenomena
(steady-state response) can also have a significant impact on the design. Aeroelastic and
structural dynamic phenomena can sometimes result in extremely dangerous loading
conditions and/or instabilities: divergence and flutter.

Figure 1: Aerodynamics-Inertia-Elastic relationships

Considering also controls and thermal effects we have the so-called ”Aero-Servo-Thermo-
Elasticity”.

1.3 Divergence and Flutter

One of the simplest interactions found in a fixed-wing aircraft is the flexure of the wing
relative to the rigid fuselage. For aircraft with slender straight cantilever wings, two
typical modes of motion exist. The first is a bending mode where the wing tip flexes
up and down relative to the fixed wing root. The second is a twisting mode where the
wing rotates about its stiffness axis.

8 Structural Dynamics and Aeroelasticity

 Figure 2: Aero-Servo-Thermo-Elasticity

Normally there is minimal effect of these two modes on structural behavior, with only a
slight vibration being seen for each motion. The bending mode shows up as a relatively
low-frequency flapping effect while the twisting mode is found to be a much higher
frequency vibration. However, with the application of high-speed airflow as a source of
excitation energy, these two modes can produce motions with will severely distort or
break the wing.

1.3.1 Divergence
In the divergence case, the moment produced by the air load is greater than the structural
torsional stiffness and thus it will be twisted off the vehicle. The threshold speed for this
type of failure to occur is called divergence speed, particular problems occur with swept
forward wings as these have a relatively low divergence speed.

1.3.2 Flutter
In this case, there is a synchronized interaction between both modes so that energy
is absorbed from the airflow in one mode to increase the amplitude of the other. At
this point, the frequency of each mode has converged to the same value so that only
one combined mode is possible. The wing will absorb energy from the airflow and will
behave as an ever-increasing bending and torsion flexure until sufficient displacement is
reached and the wing breaks. When the airflow is increased to the critical point to cause
this failure, it is called the flutter speed.
Still today Control surface flutter is the most common problem found in design. i.e.
right and left stabilizers are connected to avoid flutter.

9 Structural Dynamics and Aeroelasticity

Important: flutter is an instability characterized by oscillatory behavior, and
divergence is characterized by exponential behavior.

1.4 Brief history of Aeroelasticity

• 1903-1930: First flight. The main problem is the torsional divergence on a single-
wing aircraft. Wright brothers solved the problem by building a biplane: no tor-
sional divergence.

Monoplanes were made of wood and fabric: with very low torsional stiffness. i.e.
Fokker D-VIII had wing failures at high speeds.

• 1916-now: Control surface flutter. i.e. H.P. Type 0 had a tail with twist moment:
oscillatory divergence. This was solved by joining the two surfaces.

• 1930-now: Flutter. Oscillatory instability from a certain speed. This problem

has been present since the beginning of flight. It can be fatal. We have to prevent
flutter inside the flight envelope of the aircraft.

• 1940-now: Control reversal. The presence of a flexible wing and flexible aileron
means that there is a certain speed at which the rotation of the control surfaces
produces no effects. This is a very dangerous situation since the pilot can’t control
the plane anymore.

• 1959-now: Panel flutter, stall flutter. This behavior is connected to a supersonic

or hypersonic flight condition. It is a local phenomenon connected to a single
panel. It can lead to fatigue breaking, i.e. North American X-15.

10 Structural Dynamics and Aeroelasticity

2 Static Aerolasticity
2.1 Rigid wing model: typical section
• Rigid with a constant section;

• Span perpendicular to wing speed;

• Deformability concentrated in a torsional spring root;

• 2D strip theory;

• linear behavior since only low amplitude pitch oscillations are considered

Figure 3: Rigid wing mounted on an elastic lumped support

2.1.1 Definitions
Shear center (SC)
It is a point of a 2D section where if you apply a transverse shear load it produces zero
change (i.e. rate) of twist of the section. It is a property of the cross-section

Elastic axis (EA):

It is the locus of shear centers along the span of the wing (may not be a straight line).
It is typically between 35-45% of the chord for semi-monocoque structures (with no
balancing ballast).

Flexural Center (FC or center of twist):

the point where the shear load P applied to the cantilever where it does not cause torsion
(i.e., rotation) of the section ACB, but not necessarily elsewhere in the wing.
In general, the flexural center depends on the load distribution along the beam. The
location of the flexural axis may be important for the assessment of bending-twisting

11 Structural Dynamics and Aeroelasticity

coupling of structures.

Aerodynamic Center (AC):

the point of the section around which the steady aerodynamic moment coefficient is
constant with respect to changes of angle of attack

2.2 Divergence

Figure 4: Typical section

2.2.1 Equilibrium approach

• AC (aerodynamic center): point of the section around which CM doesn’t depend
on the variation of the AOA CMACα
• EA (elastic axis): locus of shear centers along the span of the wing.
• e: distance between EA and AC
• Structural stiffness: kα
• Aerodynamic stiffness: kA = eSCLα
• Equivalent stiffness: K̃ = kα − qKa
From moment equilibrium in AC:
kα θ = Le + MAC
eCLα + cCMAC M0
θ = qS =
kα + qeSCLα kα − qKA
Elastic deformation:
M0
θ0 =
Kα
Aeroelastic deformation:
M0
θ=
Kα − qKa
Imposing null the denominator we find the divergence pressure

12 Structural Dynamics and Aeroelasticity

 Figure 5: Aeroelastic loop

kα kα
qD = =
eSCLα kA

When I reach a certain dynamic pressure it is impossible to reach a θ of equilibrium, so

the airfoil twists infinitely as an unconstrained body until it breaks.

With a negative e we can prevent divergence. This is structurally difficult to obtain.

We can also calculate the divergence speed:

s
2kα
UD =
ρeSCLα

Aeroelastic performance index

θ 1
=
θ0 1 − qqD

The closer q is to qD the higher will be the incidence of aeroelastic effects.

2.2.2 Comparison of aeroelastic and rigid lift

LR = qSCL0 = qSCLα α0
L = qSCLα (α0 + θ)
1
θ = θ0
1 − qqD
!
L − LR θ0 1
=
LR α0 1 − qqD

13 Structural Dynamics and Aeroelasticity

 Figure 6: Rigid and elastic lift

2.2.3 Torsional divergence as a stability problem

Taken a system that is in equilibrium, the equilibrium is said statically stable if given
a slight perturbation there is a tendency to return to the equilibrium. Considering the
equilibrium solution θ = θT we can analyze the evolution by expanding.
∂f
MAE = f (θ, q) ≃ f (θT , q) + ∆θ
∂θ
MAE = qSeCL0 + qScCMAC + qSeCLα θ − kα θ
But in the equilibrium:
MAE (θT , q) = 0
The first derivative:
∂f
∆θ = (qSeCLα − kα )∆θ
∂θ
Stable for: qSeClα − kα < 0
Unstable for: qSeClα − kα > 0

The critical condition is found by imposing the derivative null. We get q= qd So qd is

the dynamic pressure for which the system turns from stable to unstable.

2.2.4 Eigenvalue approach problem

In equilibrium, as before:
MAE = f (θT , q) + (kα − qSeCLα )∆θ = 0
(kα − qSeCLα )∆θ = 0
The trivial solution is ∆θ = 0, otherwise:
(kα − qSeCLα )∆θ = 0
(kα − qkA )∆θ = 0
This looks like an eigenvalue problem.

14 Structural Dynamics and Aeroelasticity

2.2.5 Divergence and Mach number
Compressibility enters into the aeroelastic problem in two ways:
1. Variation of Clα

2. Variation of the position of the aerodynamic center

 M =0 2  M =0 2
4 qD 2 qD
MD + MD − =0
A A
Taking into account compressibility the equivalent stiffness of the system is reduced,
consequently, the system diverges earlier. This equation doesn’t take into account the
travel of the AC due to compressibility. This hypothesis is valid if M is small. A is a
function of the altitude so for every altitude is possible to identify several divergence
Mach numbers, only the lowest is meaningful.

Remark: divergence is NOT a supersonic problem if divergence speed belongs to

the supersonic regime it’s only theoretical.
1
A = γP∞
2
Through the Prandtl-Glauer relation (valid only for subsonic and not transonic speeds):
q
M =0 2
qD = qD 1 − MD

2.3 Control surface effectiveness

Figure 7: Typical section & flap

Let’s now add another degree of freedom: the flap deflection β.

Note: from now on θ will be short for ∆θ, we are not interest in the equilibrium condition.

Equilibrium equations (two d.o.f):

L = qS(CL0 + CLα θ + CLβ β)

MAC = qSe(CMAC + CMβ β)

Note: typically CMβ < 0.

15 Structural Dynamics and Aeroelasticity

Twist due to flap rotation
From equilibrium:
kα θ = Le + MAC
We get:
eCLβ + cCMβ
θ(β) = qS β
kα − qeSCLα
Same divergence dynamic pressure as before.

2.3.1 Control Elastic efficiency and control reversal

L = qS(CLα θ + CLβ β)
Substituting:
eCLβ + cCMβ
 
L = qS CLα qS β + CLβ β
kα − qeSCLα
qSCLα (eCLβ + cCMβ )
 
L = qS + CLβ β
kα − qkA
CLα CMβ
(CLβ )e 1 + qSc kα CLβ
EC = = q
CLβ 1− qD

It is possible to reach a dynamic pressure such that the numerator goes to zero: it means
that the rotation of the flap isn’t producing any change of lift. The rotation of the flap
produces an increase of lift CLβ and consequently negative pitching, the wing submitted
to CMβ twist with θ < 0, this causes a reduction of the angle of attack and consequently
a reduction of lift that (in this specific case), balances the increase of lift due to the flap.
Control reversal condition:

kα CLβ
qR = −
ScCLα CMβ

In the control reversal condition, the flap deflection produces an effect that is opposite
to the one for which the flap was designed. The more I increase q the less efficient is the
control. If you want an aircraft that is more controllable at a higher speeds, you must
meet divergence earlier.

2.3.2 Plot control surface elastic efficiency

q
1− qR
Ec = q
1− qD

q < qR
 EC > 0 −→ The higher is q, the more efficient
q = qR EC = 0 −→ Zero efficiency, no controllability

q > qR EC < 0 −→ Opposite behaviour than expected


16 Structural Dynamics and Aeroelasticity

 Figure 8: Surface elastic efficiency

2.3.3 Control Surface Load Distribution

When the aileron is activated the center of pressure is moved backward (because the
increase of lift generated by the flap is behind the center of pressure of the wing shifts
backward). The shift on the center of pressure generates a negative (nose down) pitching
moment.

The elastic twist reduces the incidence and torsional loads, favorable redistribution of
loads.

2.3.4 Matrix form of control reversal problem

Starting from the system:

L = qS(CL0 + CLα θ + CLβ β)

(kα − qKA )θ = qSe(CMAC + CMβ β)

We can retrieve the control reversal pressure through two different approaches, choosing
different unknowns.

Matrix approach 1: lift increment

Let’s determine the lift increment and twist due to the flap deflection β for an assigned
dynamic pressure:
    
0 kα − qKA L eCLβ + cCMβ
= qS β
1 −qSCLα θ CLβ
    
L qs −qSCLα −Kα + qKA eCLβ + cCMβ
= β
θ Kα − qkA −1 0 CLβ

17 Structural Dynamics and Aeroelasticity

To identify the control reversal condition we can use the first equation L = 0.
kα CLβ
qR = −
ScCLα CMβ

Matrix approach two eigenvalue problem

Given a target variation of lift let’s compute the required rotation of the flap and the
resulting twist:     
−qSeCLβ + cCMβ kα − qKA θ 0
= L
−qSCLβ −qSCLα β −1
   
θ 0
A = L
β −1
When L=0 the homogeneous problem can be seen as an eigenvalue problem in which
the eigenvalues are the qi for which a non-zero control rotation is possible even if L=0.
 
θ
A =0
β

Find qR by imposing:
det(A) = 0
This is again an eigenvalues problem that gives us an additional solution at zero
dynamic pressure any rotation of beta produces no change of lift.

2.4 Control surface stiffness

Figure 9: With the added stiffness

We have to introduce another stiffness to model the control chain.

Let’s define the aerodynamic hinge moment H:

H = qSf cf (CHα α + CHβ β)

We are adding another equation. Our system is now in equilibrium around EA and flap
hinge.

18 Structural Dynamics and Aeroelasticity

 (
qSe (CLα θ + CLβ β) + qScCmβ β − kα θ = 0
qSf cf (CHα θ + CHβ β) − kβ (β − β0 ) = 0
       
kα 0 SeCLα SeCLβ + ScCmβ θ 0
−q = β0
0 kβ Sf cf CHα Sf cf CHβ β kβ
Ax = F =⇒ x = A−1 F
By solving it:

kβ
θ = qS (eCLβ + cCmβ )β0
detA
kβ
β= (kα + qSeCLα )β0
detA
Substituting in the lift L = qS(CLα θ + CLβ β):

L = qS (CLα θ + CLβ β)
kβ
L = qS (CLα qS (eCLβ + cCmβ ) + CLβ kα − CLβ qSeCLα ) β0
det A  
kβ kα qSc
L = qS CLβ + Cmβ CLα β0
det A kα

Calling LR = qSCLβ β0 , then:

 
L kβ kα qSc Cmβ CLα
= 1+
LR det(A) kα CLβ

Reversal is possible for the same dynamic pressure as for the infinite flap hinge
stiffness. There is a change in divergence speed.

2.5 Rolling of a straight wing

Figure 10: Model for the rolling wing

19 Structural Dynamics and Aeroelasticity

We would like to represent a dynamic maneuver of the aircraft considering the effect of
flexibility.

We allow the simple typical section model to rotate freely about the roll axis. The
proposed model allows the dynamics of the rigid roll movement caused by a rotation of
the aileron to be investigated

Hypothesis:

1. The motion is considered the sum of a rigid movement plus the deformation of the
wing structure.

2. It is supposed that the structure deformation behaves statically, i.e.

• The characteristic time required by the deformable structure to adapt to a

change of loading is much smaller than the characteristic time of the rigid
movement.
• The structure is deformed as if the load is applied statically, i.e.the different
loading conditions during the maneuver are equivalent to a sequence of static
deformations.

3. Aerodynamics too behaves statically.

2.5.1 Aerodynamics
 
py py
∆α = tan ≈
U∞ U∞
The lift is:    
py
L = qc(y) CLα θ − + CLβ β
U∞
MAC = qc2 (y)Cmβ β
We have also inertia forces in G:
Fi = −mṗy
The twist generated around the EA is:

Mi = mdṗy

Integrating:
Z b Z b
Ly dy − my 2 dy ṗ = 0
0 0
Z b   
py
−Ixx ṗ + qc CLα θ − + CLβ β dy = 0
0 U∞

20 Structural Dynamics and Aeroelasticity

For a rigid system θ = 0, then:
 
py
Ixx ṗ = qSb Clp + Cl β β
U∞
Where:
CLα CLβ
Cl p = − Cl β =
3 2
For the elastic wing instead:
 
py CLα
Ixx ṗ = qSb Clθ θ + Clp + Cl β β (1) Clθ =
U∞ 2
This model is still considered part of static aeroelasticity. The only inertia forces con-
sidered are those caused by rigid motion.

2.5.2 Twist
Z b Z b
kα θ = (Le + MAC dy + myd dy ṗ
0 0
Substituting:
mb2
   
1 pb

kα θ = qS eCLα θ − + eCLβ + Cmβ β + dṗ
2 U∞ 2
Or also:
b2
 
eCLα pb

(kα − qSeCLα )θ = qS − + eCLβ + Cmβ β + mT dṗ
2 U∞ 2
We have a system of two equations:
 
py

Ixx ṗ = qSb Clθ θ + Clp U + Clβ β (1)


∞
b2
 
eC
(kα − qSeCLα )θ = qS − Lα f racpbU∞ + eCLβ + Cmβ β + mT dṗ (2)



2 2
Matrix form of the problem:
" # 
C    
Ixx −qSb L2 α ṗ bCℓβ bCℓp pb
= qS β − qS eCLα
−mT db
2 (k α − qkA ) θ eC Lβ
+ cCm β 2 U∞

If d = 0, then:
(eCLβ + cCmβ ) eCLα pb
θ = qS β − qS
kα − qkA 2(kα − qkA ) U∞
Otherwise:
(eCLβ + cCmβ ) eCLα pb mT db
θ = qS β − qS + ṗ
kα − qkA 2(kα − qkA ) U∞ 2(kα − qkA )

21 Structural Dynamics and Aeroelasticity

Substituting in (1) we get:
!
eCL2 α CLα (eCLβ + cCmβ )
 
pb
Ixx ṗ = −qSn Clp + qS + qSb Clβ + qS β
4(kα − qkA U∞ 2 kα − qkA

Or  
pb
Ixx ṗ = qSb −(Clp )e + (Clβ )e β
U∞

2.5.3 Control reversal problem

Find q = qR such that (Clp )e = 0.

Kα CLβ
qR = −
ScCLα Cmβ

It is the same reversal pressure.

2.5.4 Consistent performance problems

We can use this formulation to calculate several consistent performance problems.
Consider:

• Angular speed generated by a rotation of β a regime roll speed p̄ when ṗ = 0.

p̄b (Clβ )e
= β
U∞ (Clp )e

• Initial acceleration ṗ0 generated by β. At t = 0, p0 = 0.

qSb(Clβ )e
ṗ0 = β
Ixx

22 Structural Dynamics and Aeroelasticity

3 Multi degree of freedom torsional divergence model

Figure 11: Multi dof typical section model

The model is composed of several rigid wing portions connected through torsional
springs. For each rigid portion the simple pseudo-2D aerodynamic model is used. No
mutual aerodynamic influence between portions is considered.

3.1 Principle of virtual works - PVW

n
X n
X
δW = Fi · δri + Mj · δφj
i=0 j=0

Where δri and φj are virtual movements: do not violate any constraints.
Another way to write the PWC is:

δWi = δWe

3.2 MDOF model

N −1
1X 1
U= ki+1 (θi+1 − θi )2 + k1 θ12
2 2
i=1
∂U
= ki (θi − θi−1 ) − ki+1 (θi+1 − θi )
∂θi
And the i-th aerodynamic moments:

MiAER = Mi0 + qSi ei CLα θi

Putting all together:

X
δW N C = MiAER δθi −→ Qi = MiAER
i

∂U
= Qi
∂θi

23 Structural Dynamics and Aeroelasticity

The resulting model has the same structure as the typical section case but is matricial.

(Ks − qKA )θ = M0

To find the divergence conditions consider the homogenous problem.

(Ks − qKA )θ = 0

The non-trivial solution is determined as:

det(Ks − qKA ) = 0

Note: in general, we will have as many divergences dynamic pressures as many degrees
of freedom.

3.2.1 Power method

This method can be used to find the biggest eigenvalue.

For our purposes we are actually interested in the lowest one (the most restricting dy-
namic pressure).
1
We can simply define λ̂ = λ and look for the biggest λ̂.

3.3 Equivalent stiffness

Equivalent stiffness is defined as the ratio between the twist moment applied divided by
the twist at two subsequent sections.
Mt (yi+1 )
ki+1 =
θi+1 − θi

Let’s compute them using the PCVW (Principle of Complementary Virtual Works)
through a virtual load δMt applied in yi+1 :
Z b Z yi+1
∗ Mt Mt Mt δMt
δ Wi = δMt dξ = δMt dξ = yi+1
0 GJ 0 GJ GJ

δ ∗ We = θi+1 δMt
Mt δMt
yi+1 = θi+1 δMt
GJ
 
Mt
δMt θi+1 − yi+1 = 0 ∀δMt (1)
GJ
And a virtual load δ M̂t applied in yi :
 
Mt
δ M̂t θi − yi = 0 ∀δ M̂t (2)
GJ

24 Structural Dynamics and Aeroelasticity

Subtracting (1) and (2):
yi+1 − yi
∆θ = Mt
GJ
Substituting in the definition we get:
GJ
ki+1 =
yi+1 − yi

3.4 Integral formulation - Flexibility influence functions

Define θ̂i as the relative angle between two sections :
1
θ̂n = θn − θn−1 = mn
kn
n n
1 X X
θ̂n−1 = mj = Cn−1 mj
kn−1
j=n−1 j=n−1

Go on until:
n n
1 X X
θ̂1 = mj = C1 mj
k1
j=1 j=1

Where Ci are influence functions.

If the number of sections n → ∞ then we have:

Z L
θ(y) = C θθ (y, ξ)m(ξ) dξ
0

Note: if a unit point moment is applied as y = γ i.e m(ξ) = δ(ξ − γ) where δ is the Dirac
delta we have: Z L
θ(y) = C θθ (y, ξ)δ(ξ − γ) dξ = C θθ (y, γ)
0
So C θθ represents the twist at y due to a unit moment in γ. How do we proceed?

1. Find C θθ (yi , yj ), They could be evaluated experimentally: apply a moment in yi

and measure the twist in every other section. (Th. Maxwell-Betti)

C θθ (yi , yj ) = C θθ (yj , yi )

2. Use the PVW, compute θ(y) due to a moment in ξ.

Z L Z min(y,ξ)
θθ Mt 1
C (y, ξ)δMt = δMt dη = dηδMt
0 GJ 0 GJ

ξ

 y>ξ
GJ
 y y≤ξ

GJ

25 Structural Dynamics and Aeroelasticity

Now the aerodynamic moment in the AC is:

mt = qc(eCL0 + eCLα θ + cCmAC )

(cCL )e
But changing the unknowns: θ =
cCLα

mt = q(ecCL0 + e(cCL )e + c2 CmAC )

Substituting:
Z L
(cCL )e
=q C θθ (y, ξ)(ecCL0 + e(cCL )e + c2 CmAC ) dξ
cCLα 0

(cCL )e is taken as the unknown and discretized as constant values on wing strips, making
the hypothesis that the deformation does not depend on how loads are distributed on a
strip.

Figure 12: Constant distribution of cCL

(cCL )ej X Z yi+1/2

=q C θθ (yj , ξ) dξ(ei ci CL0i + ei (cCL )ei + c2 CmACi )
cj CLαj yi−1/2
i

And
Cij = C θθ (yj , yi )
The integral is:
C θθ (yj , yi )∆bi
Where ∆bi = yi+1/2 − yi−1/2 . The sum can be also written as:
  
C11 C12 . . . ∆b1 0 0
C21 C22 . . .  0 ∆b 0
F=  2 
.. .. .. ..
. . . 0 0 .

Note: F = K−1

26 Structural Dynamics and Aeroelasticity

Putting everything together we have :
.
.. .. ..
     
.. .
.  .   ..   . 
 1    
 cj CLα  (cCL )e  = qF m0   ej  (cCL )e 
j j  j   j 
. . ..
  
.. .. .. ..
. . .
 .
.. ..
      
.. ..
 1   .   .   . 
 cj CLα  − qF  ej  (cCL )e  = qF m0 
j j  j
.. ..
    
.. ..
. . . .
Or
(A − qB)(cCL )e = qQ
Where:
Q = Fm0 A = Diag(1/cj CLαj ) B = FDiag(ej )
From here we can find the divergence speed by solving the associated homogeneous
problem, through the eigenvalue approach.

3.5 Slender wing: analytical solution for torsion

Let’s now consider a slender beam constrained in one end and a distributed moment mt .
Take an infinitesimal portion of the beam of length dy. We can write the equilibrium

Mt − (Mt + dMt ) = mt dy
dMt
+ mt = 0
dy
Then the constitutive law for torsion can be expressed as GJθ′ = Mt .
Substituting:
(GJθ′ )′ + mt = 0
Apply BCs: θ(0) = 0 and θ′ (L) = 0.

We can also get the same result by using the PVW. Let’s write the PVW for torsion:
Z Z L Z L
′
δWi = σij δεij dv = δθ Mt dy = δθ′ GJθ′ dy
V 0 0

Integrate by parts:
Z L Z L
δWi = [δθGJθ′ ]L
0 − δθ(GJθ ) dy ′ ′
δWe = δθmt dy
0 0

Since this has to be valid ∀δθ it results:

(GJθ′ )′ + mt = 0

27 Structural Dynamics and Aeroelasticity

3.5.1 Wing

Write the differential equation for the wing case (all properties uniform):

(GJθ′ )′ = −qc(e(CL0 + CLα θ) + cCmAC ) + dN mg

Deriving:
qceCLα qc dN mg
θ′′ + =− (e(CL0 + cCmAC ) +
GJ GJ GJ
Let’s non-dimensionalize the problem by defining:

y d2 θ d2 θ d2 ŷ θ∗∗
ŷ = −→ θ′′ = 2 = 2 2 = 2
L dy dŷ dy L
Our equation then becomes:

θ∗∗ + µ2 θ = QA + Qm , θ(0) = 0, θ∗ (1) = 0

Where:
qecCLα L2 qcL2 dN mgL2
µ2 = QA = − (eCL0 + cCmAC )) Qm =
GJ GJ GJ

3.5.2 Solution of the homogeneous

This is an easy equation to solve.

θ(ŷ)) = A sin µŷ + B cos µŷ

Applying the BCs we have a system:

    
0 1 A 0
=
µ cos µ −µ sin µ B 0

Except for the trivial solution, A = B = 0 the other solution is:

µ cos µ = 0

This happens for µ = 0 and µ = ±(2n + 1) π2 , n ∈ N. From here we can retrieve the n
divergence pressure as:
 π 2 GJ
qD = (2n + 1)2
2 ecL2 CLα
We have an infinite number of eigenvalues or divergence dynamic pressures.

From the same result, we can also find the associated deformation modes or eigenfunc-
tions:  πy 
θ̂n = sin (2n + 1)
2L

28 Structural Dynamics and Aeroelasticity

 Figure 13: Modes of the divergence

Comparison with the typical section

 π 2 GJ kα
qD0 = qD T S =
2 ecL2 CLα ecLCLα

The comparison says that:

π 2 GJ
kα =
4 L

3.5.3 Solution of the forced problem

θ∗∗ + µ2 θ = QA + Qm = QT , θ(0) = 0, θ∗ (1) = 0
The solution is easily found as:
QT
θ(ŷ) = A sin µŷ + B cos µŷ +
µ2
    " QT #
0 1 A − µ2
=
µ cos µ −µ sin µ B 0
That results in:
QT QT
B= A= tan µ
µ2 µ2
Substituting:
QT
θ(ỹ) = (1 − tan µ sin µỹ − cos µỹ)
µ2
Case 1: variation of lift distribution at constant angle of attack

If we neglect the inertial effects (i.e. d = 0) and consider a symmetric airfoil (CmAC = 0),
then:
qcL2 eCL0 QT CL
QT = QA = − =⇒ 2 = − 0
GJ µ CLα

29 Structural Dynamics and Aeroelasticity

So:
(CL )e = CLα θ = −CL0 (1 − tan µ sin µỹ − cos µỹ)
CL CL0 + (CL )e
= = tan µ sin µỹ + cos µỹ
CL0 CL0

Figure 14: Constant angle of attack lift distribution

Due to the twist of the wing, we will have a higher angle of attack at the tip of the wing.
This means that we will have an overall higher lift than the one expected for the rigid
wing.

This will also change the shape of our distribution. i.e. if we wanted an elliptical lift
distribution for better performances the twist would change this behaviour.

Case 2: variation of lift distribution at constant lift

Write the equilibrium equation in the z direction for the rigid aircraft with W (weight)
For simplicity define α0 trim angle of rigid aircraft and α̂0 trim angle of flexible aircraft.
For the rigid a/c
Z L
WN
q cCLα α0 dy =
0 2

For the elastic A/C

Z L Z L
WN
q c(CL )e dy + q cCLα α̂0 dy =
0 0 2
This imposes that: Z L Z L
q c(CL )e dy = cCLα (α0 − α̂0 ) dy
0 0
So the variation of the trim angle of attack to obtain the same total lift is ∆α = α0 − α̂0 .
Write: Z L
cCLα (θ − ∆α0 ) dy = 0
0

30 Structural Dynamics and Aeroelasticity

 Z 1  
QT
cL CLα (1 − tan µ sin µỹ − cos µỹ) − ∆α0 dy = 0
0 µ2
 
QT tan µ cos µ tan µ sin µ
1 − − − − ∆α0 = 0
µ2 µ µ µ
Assuming CmAC = 0 and d = 0:
 
QT tan µ
1− = ∆α0
µ2 µ

(CL )e = CLα (θ − ∆α0 )

 
QT tan µ
(CL )e = CLα 2 1 − tan µ sin µỹ − cos µỹ − 1 +
µ µ
The ratio is:
CL tan µ
=1− + tan µ sin µỹ − cos µỹ
(CL )e µ
To keep the same total lift we have to decrease the angle of attack. We have the same
integral but the function is lower in the middle and higher in the extremity.

Figure 15: Constant lift, lift distribution

3.6 Ritz-Galerkin approssimation

Define:
∞
X
θ(y) ≈ θ̃(y) = Ni (y)qi
i=1

Where:

• Ni (y) are known and called Shape functions

• qi are the generalized coordinates

31 Structural Dynamics and Aeroelasticity

Important notes:
• Express the unknown displacement as a complete linear combination of known
space functions.
• The shape functions chosen must belong to a complete set, meaning that in
the domain considered, taking a sufficient number of terms it must be possible
to represent the exact solution to any degree of accuracy (convergence). Typical
complete sets are polynomials, trigonometric functions, etc. . .
• The coefficient of the series qi is a new set of coordinates (i.e. degrees of freedom)
used to describe the displacement field. In practice we use a finite set of functions:

N
X
θ(y) ≈ θ̃(y) = Ni (y)qi = N(y)q
i=1
Where:
N = [N1 , N2 . . . , NN ] q = [q1 , q2 . . . , qN ]T
Note: If the method is applied to a strong formulation, (i.e. differential) the chosen
shape functions must satisfy all boundary conditions.

Each shape function should be p times differentiable where p is the highest spatial
derivative present in the formulation. At least one of the shape function p-th order
derivatives should be not zero.

If F is the functional that describes our problem then:

F (θ(y)) = 0
But
F (θ̃(y)) = ε(y) ̸= 0
We have an error.

To get a solvable system with N equations we can adopt two different strategies:
1. Weighted integrals: Z L
wi (y)ε(y) dy = 0
0
Where wi are weight functions.
2. Collocation (spectral method): imposing null the error in N points:
ε(Yi ) = 0 i = 1, 2, 3, . . . , m

Usually, the Ritz approach is applied to the weak formulation. In our case the PVW.
In this case there are two functionals to be approximated: the displacement field (trial
functions), and the virtual displacement field (test function). If the same series (func-
tional space) is used for both, the approach is called Ritz-Galerkin.

32 Structural Dynamics and Aeroelasticity

3.6.1 Requirements for Ritz in weak formulation
• Regularity required to shape function is lower since the maximum p derivative is
lower.

• The shape functions must satisfy boundary conditions on displacements (Dirichlet

BCs) (and rotations) often called “geometric” or “kinematics” boundary condi-
tions.

• The boundary conditions on forces and moments, called “natural” (Neumann BCs)
are not necessary. Their satisfaction will be obtained. naturally at convergence
(by using a sufficiently large number of shape functions). However, satisfying also
these conditions can speed up convergence.

33 Structural Dynamics and Aeroelasticity

4 Swept wings

Figure 16: X-29: forward swept wings

4.1 Introduction
4.1.1 Reason to use a sweep angle
There are many reasons to use swept wings in an aircraft: firstly, we want to reduce the
local Mach. Secondly, we want to increase stability for subsonic flight. We know that
in the transonic range, we have a higher CD due to compressibility effects. By sweeping
the wing the velocity seen by the wing will be reduced by a factor of cos Λ. This means
we can go faster and still have a lower local Mach on the wing, preventing in this way
the added drag.

Let’s look at the image below (fig 17) (source: Hoerner’s Fluid Dynamic Drag, section
XV).

Figure 17: Drag increase in transonic and swept wing

34 Structural Dynamics and Aeroelasticity

Aerodynamics-wise, the same effect can be obtained with backward-swept wings (Λ > 0),
as the majority of aircraft, or with forward-swept wings Λ < 0. We will see that there
is an aeroelastic difference.

The second effect is stability:

XCG − XAC
e=
c
Pushing backward the AC with respect to the CG will increase the longitudinal stability
of the aircraft (e > 0 for stability). This can be obtained by sweeping the wing.

4.1.2 Torsion

Figure 18: Swept wings reference systems

Aerodynamic forces are developed by airfoils aligned with the asymptotic speed.

Define a structural reference system frame (x̄, ȳ) aligned with the axis of the wing.
Rotated by a Λ angle wrt the aerodynamic reference (x, y). Torsion and bending are
along the structural reference system. We can write the rotation matrix as:
    
x cos Λ − sin Λ x̂
=
y sin Λ cos Λ ŷ

Because of this only a portion of the torsion θ results in a change in angle of attack:
∆αT = θ̄ cos Λ
φ = θ̄ sin Λ
This effect can be neglected since is small.

4.1.3 Bending
When a swept-back wing bends up them angle of attack of streamwise sections will
reduce.
∆ȳ = c sin Λ
z(ȳ1 ) − z(ȳ2 ) 1 dz
∆αB = =− ∆ȳ = −z ′ sin Λ
c c dy

35 Structural Dynamics and Aeroelasticity

4.1.4 Total change of angle of attack
The change of the streamwise angle of attack resulting from the elastic deformation is
made up of a component coming from the structure twist and a component coming from
the slope caused by wing bending.

∆α = ∆αT + ∆αB = θ̄ cos Λ − z ′ sin Λ

α = α0 + ∆α = α0 + θ̄ cos Λ − z ′ sin Λ

4.1.5 Forces and moments

To solve the bending and torsional problems it is necessary to compute the shear S and
torsional m load along the structure.

S(y) = qcCL − mN g

N mg: inertia forces

m(y) = qecCL + qc2 CmAC − N mgd
The total shear is: Z b Z b̄
ST = S(y) dy = S(ȳ) dȳ
0 0
If we consider dy = dȳ cos Λ, then:

S(ȳ) = (qcCL − mN g) cos Λ

m(ȳ) = (qecCL + qc2 CmAC − N gd) cos Λ

The aerodynamic moment could be decomposed in a torsional component along ȳ and
a bending component along x̄.
mȳ = m(ȳ) cos Λ
mx̄ = m(ȳ) sin Λ
Note: mx̄ is negative.

The wing bending is caused by the shear force load S(ȳ) and the moment mx̄ . This
second term may be neglected (as a first approximation).

S(ȳ) = (qcCL − mN g) cos Λ = (qcCLα α − mN g) cos Λ

S(ȳ) = qcCLα (α0 + θ̄ cos Λ − z ′ sin Λ) cos Λ − mN g cos Λ

Then:
mb (ȳ) = mx̄ (ȳ) = ((qecCLα α0 + qc2 CmAC − N gd) cos Λ) sin Λ + . . .
The torsion is caused by the moment mȳ .

mT (ȳ) = mȳ = ((qecCL + qc2 CmAC − N gd) cos2 Λ)

36 Structural Dynamics and Aeroelasticity

4.2 Divergence - typical section
Consider a rigid model with:

• spring at root to represent TORSIONAL STIFFNESS

• another spring to represent BENDING STIFFNESS

Figure 19: Swept wings stiffness

Write the equilibrium using the PVW:

δWi = δ θ̄kθ θ̄ + δ φ̄kφ φ̄

The lift L works for the virtual displacement of the AC: δwAC = δ(θ̄e + φ̄y).
Z b̄ Z b̄
δWe = δ θ̄T eL d(ȳ) + δ φ̄T eȳL d(ȳ)
0 0

Where:
1  α
0

α0 + θ̄ cos Λ − z ′ sin Λ = q̄c̄C̄Lα + θ̄ − z ′ tan Λ


L = qcCLα α = q̄c̄C̄Lα
cos Λ cos Λ
Note:
c̄ α CLα
dy = dȳ cos Λ , c = , q̄ = q cos2 Λ , ᾱ = , C̄Lα =
cos Λ cos Λ cos Λ
We obtain:    
θ̄ Q̄ ē
(Ks − Q̄KA ) = α0
φ̄ cos λ b̄/2
Where:    
k 0 ē − tan Λē
Ks = θ̄ KA =
0 kφ̄ b/2 − tan Λb̄/2
And Q̄ = q̄c̄b̄C̄Lα = q̄S C̄Lα = qS cos2 ΛC̄Lα

37 Structural Dynamics and Aeroelasticity

Note: det(KA ) = 0 the aerodynamic stiffness is singular!! This means that even if we
have two dofs we have only one divergence dynamic pressure.

Compute the divergence speed as usual:

det(Ks − Q̄KA ) = 0
We get:
1 kθ 1
qD = 2
 
cos Λ SēC̄Lα 1 − kθ b tan Λ
kϕ e 2

It depends on Λ:
1. Λ > 0 (backward sweep angle). It is possible to identify a critical sweep angle
ΛCRIT for which no divergence exists, because qD < 0.
2. Λ < 0 (forward sweep angle). The higher is |Λ| the lower is qD
for example below −30° divergence speed is close to zero.

4.3 Lift effectiveness

Figure 20: Lift effectiveness for swept wings

Compute θ̄ and φ̄ and plug them in:

 α   α 
0 0
L = q̄S C̄Lα + θ̄ − φ̄ tan Λ LR = q̄S C̄Lα
cos Λ cos Λ
L 1
=
LR 1 − q¯q̄D

Λ < ΛCRIT , qD > 0 L/LR > 1

Λ = ΛCRIT , qD = ∞ L/LR = 1

Λ > ΛCRIT , qD < 0 L/LR < 1


When Λ > ΛCRIT we have a loss in lift (we have to go faster) BUT it is more stable.

When Λ < 0 the ratio grows very fast: we have a lot of lift at higher speed BUT it
is less stable. For large positive sweep angles, the divergence disappears but there is a
reduction of the lift effectiveness.

38 Structural Dynamics and Aeroelasticity

4.4 Control reversal
Consider now a rigid aileron on the rigid wing:

Figure 21: Swept wing with aileron model

CL
L = q̄S C̄Lα θ̄ − φ̄ tan Λ + q̄sCLβ β = Q̄(θ̄ − φ̄ tan Λ) + Q̄ β β


C̄Lα

Cm β
MAC = q̄Sc̄Cmβ = Q̄c̄ β
C̄Lα
 
Cm
CL 1 + ēc̄ CL β
 
θ̄
(Ks − Q̄KA ) = Q̄ē β  β β
φ̄ C̄Lα b̄
2ē

Now solve and retrieve θ̄ and φ̄.

CLβ
LR = Q̄ β
C̄Lα
And
Cm Q̄c̄ Cmβ
L 1 1 Q̄c̄ CL β 1+ kθ̄ CLβ
β
= + =
LR 1 − q¯q̄D 1 − q¯q̄D kθ̄ 1− q̄
q¯D

Substitute Q̄ = qS cos2 ΛC̄Lα :

S C̄Lα c̄ Cmβ
L 1+ kθ̄ C Lβ cos2 Λ
= q̄
LR 1− q¯D

Imposing the numerator null will give us the Control reversal speed:

kθ̄ CLβ 1
qR = −
S C̄Lα c̄ Cmβ cos2 Λ

39 Structural Dynamics and Aeroelasticity

4.5 Tailoring
We want to find a way to reduce the problem of torsional divergence for forward-swept
wings. Is it possible to obtain a torsional structural moment when the wing bends and
vice-versa?

Consider the case:

Figure 22: Easy tailoring example for the typical section

 
kθ̄ k
Ks =
k kφ̄
We are actively coupling the problem:
    
θ̄ cos Λ sin Λ θ
=
φ̄ − sin Λ cos Λ φ

We can use again the PVW to get the equilibrium equation in the coupled case.
Find then the divergence pressure and impose the denominator null: the critical condi-
tion corresponds to QD = ∞.
 
−1 kφ̄ ē − k b̄/2
ΛCRIT = tan
b̄/2kθ̄ − kē

It is possible to find a values of k that lead to a negative Λ.

In this way, for a negative swept wing (i.e. forward swept), it is possible to obtain
a negative critical sweep angle if k > 0 and large enough in modulus. It means to
implement a washout effect on the forward swept wing.

More generally using composite material with appropriate direction for the deposition
of the fibers, it is possible to obtain the level of coupling required.

40 Structural Dynamics and Aeroelasticity

 Figure 23: Laminate re-orientation for shape control (tailoring)

4.6 Summary
Positive sweep:

• Increases divergence speed (that often disappears).

• Decreases lift effectiveness.

• Decreases controllability.

The opposite is obtained for a negative swept.

We can use the aeroelastic tailoring to change this behavior.

41 Structural Dynamics and Aeroelasticity

5 Dynamic Aeroelasticity: introduction to flutter

Figure 24: Dynamic aeroelasticity model

Let’s introduce a new parameter: time. Considering the typical section we have two
d.o.f: pitch θ(t) and plunge h(t) both measured at the elastic axis.
The equation of motion is:
       
m md ḧ kh 0 h Qh
2 + =
md md + Iθ θ̈ 0 kθ θ Qθ
Where:
Qh = −L
Qθ = eL + MAC
Or briefly:
Mq̈ + Kq̇ = Q

5.1 Homogenous problem

Mq̈ + Kq̇ = 0
The general solution is: q = q0 eλt .
(λ2 M + K)q0 eλt = 0
Solving the generalized eigenvalue problem:
det(λ2 M + K) = 0
s
1 I θ 1 Iθ2 2
2 Iθ
λ2± = − ωh2 + ωθ2 ± ωh + ωθ2 − ωh2 ωθ2


2 I0 4 I02 I0
 q
λ12 = ±j λ2+ = ±jω1 ∈ C

q
λ34 = ±j λ2 = ±jω2 ∈ C

−

Re(λ) = 0
We have purely imaginary eigenvalues:
The system is stable but not asymptotically.

42 Structural Dynamics and Aeroelasticity

 Figure 25: Eigenvalues sign and stability

5.1.1 Associated eigenvector

 
h0 d
=
θ0 ωh2
1−
ωθ2

Figure 26: Plunge and twist

5.2 Steady aerodynamics forces

L = qSCLα
MAC = qScCMAC
Steady approximation α = θ:
      
Qh 0 −1 h 0
= qSCLα + qS
Qθ 0 e θ cCMAC

So:          
m Sθ ḧ kh 0 0 −1 h
+ + qSCLα =0
S θ Iθ θ̈ 0 kθ 0 e θ

43 Structural Dynamics and Aeroelasticity

As before:
Mλ2 + Ks − qKA q0 eλt = 0

This is a non-linear eigenvalue problem or generalized eigenvalue problem.

det(A(λ)) = 0 =⇒ aλ4 + bλ2 + c = 0

The solutions of this algebraic equation are 4 complex numbers.

λi = σ + jω

Where we can define:

• Damped natural frequency: ω = Im(λ)

• Damping ratio: ξ = − Re(λ)

|λ|

From which: 
ξ > 0 Asympthotically stable

ξ = 0 Stable

ξ < 0 Unstable


For a single dof:

ẍ + 2ξωn ẋ + ωn2 x = 0
p
λ1,2 = ξωn ± jωn 1 − ξ 2
Natural frequency:
Im(λ)
ωn = p
1 − ξ2
Note: In aeroelasticity problems, it is often used a different parameter: ”g”.

Re(λ)
g=
Im(λ)

WARNING: this is not meaningful when the eigenvalue is real. g → ∞

But if
ξ << 1 =⇒ g ≈ −ξ

5.2.1 Logarithmic decrement

This (fig: 27) is a diagram of a damped response that we could get by measuring
amplitude and time and then plotting. This diagram is easily obtainable.Can we get
useful information from it?

44 Structural Dynamics and Aeroelasticity

 Figure 27: Damped oscillation

We know that each peak is going down following a damped exponential term:

x(t + T ) = x(t)e−ξωn T

So we can easily compute the quantity δ by inverting the expression:

x(t)
δ = ln
x(t + T )

But we can also write:

δ = ξωn T
Since:
2π ωd
T = ωn = p
ωd 1 − ξ2
We can easily write:
2π
δ = ξp
1 − ξ2

5.2.2 Stability of the typical section with steady aero

 2
Sθ λ2 + qSCLα
 
mλ + kh h0
=0
Sθ λ2 Iθ λ2 + kθ − qSeCLα θ0
aλ4 + bλ2 + c = 0
Where:
a = mIθ − Sθ2 > 0 always
b = m(kθ − qSeCLα ) + Iθ kh − sθ qSCLα
c = kh (kθ − qSeCLα ) > 0 if q > qD
b can be negative for q > q̃.

45 Structural Dynamics and Aeroelasticity

 • (b2 − 4ac) > 0 and b > 0
We have four purely complex eigenvalues:

λi = ±j − ωi ∈ C

The system is STABLE

√
• (b2 −4ac) < 0 =⇒ b− 4ac At least one of the eigenvalues λ will have a Re(λ) > 0.
The system is UNSTABLE.

5.2.3 Limit of stability

The limit of stability represents the condition where the system goes from STABLE into
UNSTABLE. This condition is called FLUTTER.

To identify it, we have to find the dynamic pressure at which the quantity (b2 − 4ac)
gets negative.

Flutter: find qF s.t. (b2 − 4ac) < 0, where qF is the flutter dynamic pressure.

We can also write this problem as f (qF ) = 0. We can verify that

c1 qF2 + c2 qF + c3 = 0

If qF < 0 ∀i or qF ∈ C there is no meaningful flutter dynamic pressure.

This happens is Sθ ≤ 0

5.2.4 Divergence
If we increment further q we get to a point where c = 0

kθ − qSeCLα = 0 =⇒ q = qD

We can also note that:

aλ4 + bλ2 = λ2 (aλ2 + b) = 0
We have two null eigenvalues (rigid movement). This is the divergence condition we
studied.

5.2.5 Steady model: graphics

The flutter condition arrives when ξ crosses the real axis, at ≈ 1.18 (this is an adimen-
sionalized velocity). In the same condition, we have a coalescence of modes, and both
frequencies (plunge and pitch) are the same (see Figure (a)).

It’s also interesting to look at the phase shift of the eigenvector.

46 Structural Dynamics and Aeroelasticity

 (a) ω over V (b) ξ over V

Figure 29: Eigenvectors for steady aerodynamics

5.2.6 Effect of phase shift

From the eigenvector graph, we can see that from a certain velocity we start to have a
phase shift between the lift and the plunge movement.

Under the flutter condition (V ≈ 1.18) the phase shift is 0. Where are moving on a
straight line: the work done is null. We are keeping all our energy. Above the flutter
condition we have a phase shift that grows up to 180°. This means that the airflow
pumps energy into the structure that necessarily amplifies its oscillations. The air is
injecting energy into the system. We have to slow down.

5.3 Quasi-Steady aerodynamics forces

When the airfoil is plunging it is subjected to a vertical speed ḣ that change effectively
the angle of attack. The plunge speed is defined at the elastic axis.

ḣ
α=θ+
U
Positive in the upward direction increases the angle of attack.
            
m Sθ ḧ qSCLα 1 0 ḣ kh 0 0 −1 h
+ + + qSCLα =0
Sθ Iθ θ̈ U −e 0 θ̇ 0 kθ 0 e θ

47 Structural Dynamics and Aeroelasticity

 (a) Before the flutter q < qD (b) After flutter q > qD

Mq̈ + qCA q̇ + (Ks − qKA )q = 0

We can compute the eigenvalues as:

det(λ2 M + λqCA + (MS − qKA )) = 0

This is generally not solvable. We have to rely on numerical methods.

5.3.1 Quasi-steady: graphics

Figure 31: Eigenvectors for quasi-steady aerodynamics

48 Structural Dynamics and Aeroelasticity

The flutter condition occurs at V ≈ 0.47. Way different from the steady model. Note
that we don’t need the coincidence of frequency to see flutter.

We have a model that says that the flutter occurs at 1.18 and the other at 0.47. Which
one is correct?

They are actually both wrong.

Let’s now look at an actual flutter model, confirmed experimentally. The graph is from
Dowell’s book. (Fig:32). The actual flutter is at 0.86.

Figure 32: Actual flutter

Another interesting phenomenon that we can observe in this graph is how the damping ξ
(lower part of the graph) tends to become more negative as the velocity increases. This
is dangerous behavior since it looks like it will get more stable. Suddenly one of the two
ξ increases and becomes positive. This is known as hard flutter.

5.3.2 Quasi-steady model: pitch rate

Why is the quasi-steady model less correct than the steady one? The reason is that we
are considering only a portion of the vertical velocity. Since the airfoil is rotating with
a speed θ̇ we should consider this effect as an added vertical speed.

This leads to a butterfly speed distribution along the chord that has an analogous effect
of a cambered profile.

49 Structural Dynamics and Aeroelasticity

 Figure 33: Plunge and pitch rate effect

We can correct the model by writing:

!
ḣ  c ac  α̇
CL = 2π α + −
U 4 2 U

We generally do not consider this effect alone. We will study the complete unsteady
model in chapter 13.

5.4 Unsteady aerodynamics

The variation of lift on the airfoil causes a variation of vorticity released in the wake. The
wake will be composed of vortexes of variable intensity that are convected in the flow
field. Those vortexes induce effects on the airfoil that cause a change in lift distribution
on it. There is an internal feedback loop in the aerodynamics too.
2π
The structure oscillates at frequency ω. The time to complete a cycle is T = ω .

A fluid particle interacts with the airfoil for a time that is :

c
τ=
2U
If τ << T the interaction could be considered static.

Introduce the reduced frequency:

τ ωc
k= 2π =
T 2U
We can now discriminate cases where a static interaction can be expected (k << 1) and
where we have to consider them.

Note: to not consider a static model we need to have at least k < 0.001.

50 Structural Dynamics and Aeroelasticity

5.4.1 Wave length
Define η the wavelength of the wake:
2π c
η = UT = U = U 2π
ω 2U k
πc
η=
k
When k = 0.1 −→ η ≈ 31.5c. So not to have dynamic effects we should have

k = 0.001 −→ η ≈ 3141.5c

We almost always have dynamic effects.

51 Structural Dynamics and Aeroelasticity

6 Structural dynamics
It is necessary to understand how the elastic structure of an aircraft behaves when
subject to loads. We are interested in the entire structure’s behaviour but particularly
the elements that may cause a significant change in aerodynamic forces. A very detailed
FEM (Finite Element Modeling) may be helpful for the determination of internal stresses
but not appropriate for the determination of dynamic loads.

6.0.1 Continuous deformable structure: hypothesis

Deformable structures are arbitrarily shaped bodies which are acted by a distribution of
surface forces applied to the boundaries and body forces acting over their volume:

• Small displacement

• Equilibrium equations could be written in the initial (undeformed) configuration

6.1 Introduction and beam modeling

• A beam is a slender structural component, essentially mono-dimensional.

• Geometric and structural properties are assumed to change way more regularly
span-wise than section-wise. This means that abrupt section changes or abrupt
bends are not allowed.

• To be representative, it is necessary to reduce the complex behaviour of a 3D

structure to these simple, and rather schematic, elements.

• From the forces point of view, the beam is represented by the resultant stresses
acting on its section.

• From the kinematic point of view, the beam is described by the behaviour of a
reference line; generalized stain measures must be defined.

6.2 Euler-Bernoulli’s beam model

1. The cross-section is infinitely rigid in its own plane (does not change shape)

2. The cross section of a beam remains plain after deformation

3. The cross-section remains normal to the deformed axis of the beam

Thanks to assumption 1 the displacements in the plane of the section are rigid, i.e. not
function of the position on the section

u2 (x1 , x2 , x3 ) = v(x1 ), u3 (x1 , x2 , x3 ) = w(x1 )

52 Structural Dynamics and Aeroelasticity

Thanks to assumption 2 the axial displacement could be only composed by one transla-
tion and two rotations:

u1 (x1 , x2 , x3 ) = u(x1 ) + x3 φ2 (x1 ) − x2 φ3 (x1 )

Thanks to assumption 3:
dw
φ2 = − = −w′
dx1
dv
φ3 = = v′
dx1

6.2.1 Axial load

∗ ′ ′

(EA u ) − mü = −q1

2BC

2IC


6.2.2 Bending
∗ ′′ ′′ ′

(EI2 w ) + mẅ = q3 + m2

4BC

2IC


∗ ′′ ′′ ′

(EI3 v ) + mv̈ = q2 + m3

4BC

2IC


Note: This is valid only for the principal reference system, placed in the centroid of the
section. In general we have three coupled equations.
   ′ 
N EA V3 −V2 u
M2   V3 EI2 −EI23  −w′′ 
M3 −V2 −EI23 EI3 v ′′

For the principal reference system, instead:

   ′ 
N EA 0 0 u
M2   0 EI2 0  −w′′ 
M3 0 0 EI3 v ′′

6.2.3 Torsion
∗ ′ ′

(GJ θ ) − I θ̈ = −qt

2BC

2IC


53 Structural Dynamics and Aeroelasticity

6.3 Timoshenko’s beam model
The Euler-Bernoulli is not a perfectly accurate model. One of its main defects is not
considering the section’s warping. Real beams are free to warp to comply with boundary
conditions. The shear stress at the corner must go to zero. No local twist at the corner
is possible. Remember that τst = Gγst .

Warping depends on the geometry of the section. When an idealized 1D description of

the beam is considered, the info related to section geometry is lost.

The simplest model to recover an “average” effect of warping is to consider that sections
do not remain perpendicular to the beam axis.

Figure 34: Timoshenko model bending vs Euler

An independent rotation angle for the section is defined θy/z .

The shear stress will be different: S3 = GA∗ (w′ + θy ).

Figure 35: Timoshenko model on the right, free to warp end

6.4 Free vibration for the beam bending

Consider the homogenous problem for a uniform beam:

EIw′′′′ + mẅ = 0

Solve this equation by separating variables and then adding BCs and ICs.

w(x, t) = f (x)q(t)

54 Structural Dynamics and Aeroelasticity

The solution is:
EIf ′′′′ (x)q(t) + mf (x)q̈(t) = 0
EI f ′′′′ (x) q̈(t)
=−
m f (x) q(t)
Since we have on the left something that only depends on x and on the right something
that only depends on t the only possible solution ∀x, t is a constant.

EI f ′′′′ (x) q̈(t)

=− = const = ω 2
m f (x) q(t)

In particular, we can write:

f ′′′′ (x) = −β 4 f (x)
And
q̈(t) = −ω 2 q(t)
f (x) = a sinh βx + b cosh βx + c sin βx + d cos βx
q(t) = g sin ωt + h cos ωt
To find the free vibrations we have to impose the boundary conditions, create the matrix
H that expresses all the boundary conditions and find the solution in terms of β or ω of
the equation det(H) = 0.

6.5 Free vibration for the beam torsion

GJθ′′ − Iθ θ̈ = 0
Separation of variables:
q̈ GJ f ′′
θ = f (x)g(t) = = ω2
q Iθ f
The solution:
f (x) = a sinh βx + b cosh βx
q(t) = c sin ωt + d cos ωt
Where:
Iθ ω 2
β2 = −
GJ

6.6 Properties of the modal forms

The generic bending modal solution is:

w(x, t) = wi (x)ejωi t

Where ωi is modal frequency and wi is the modal shape (eigenfunction).

55 Structural Dynamics and Aeroelasticity

Let’s take two modal solutions wi and wj :

(EI2 wi′′ )′′ − mωi2 wi = 0 (1)

(EI2 wj′′ )′′ − mωj2 wj = 0 (2)

Multiply (1) by wj and integrate over the domain:
Z L
wj (EI2 wi′′ )′′ − mωi2 wi dx1 = 0


0
Z L
(EI2 wj′′ wi′′ − mωi2 wj wi dx1 = 0


0
Z L Z L
EI2 wi′′ wj′′ dx1 = ωi2 wj mwi dx1
0 0
Repeat the same process with inverted order of i and j.
Z L Z L
′′ ′′ 2
EI2 wj wi dx1 = ωi wi mwj dx1
0 0

6.6.1 Orthogonality with respect to mass distribution

Subtracting the two equations:
Z L
=⇒ (ωi2 − ωj2 ) mwi wj dx1 = 0 =⇒
0
Z L
mwi wj dx1 = 0 ∀i ̸= j
0

6.6.2 Orthogonality with respect to stiffness distribution

Summing the two equations:
Z L Z L
′′ ′′ 2 2
=⇒ 2 EI2 wi wj dx1 = (ωi + ωj ) mwi wj dx1 = 0 ∀i ̸= j
0 0
Z L
EI2 wi′′ wj′′ dx1 = 0 ∀i
0
What does this mean?

Z L
mwi wj dx1 = µi δij
0
Z L
EI2 wi′′ wj′′ dx1 = ωi2 µi δij
0

56 Structural Dynamics and Aeroelasticity

Where δij is the Kronecker’s delta and µi is the modal mass.

Note:
• The mode shapes corresponding to distinct frequencies are orthogonal with respect
to the mass distribution.

• The mode shapes are also orthogonal with respect to the stiffness distribution.
Define the Rayleigh quotient
Z L
EI2 (wi′′ )2 dx1
ωi2 = Z0 L
m(wi )2 dx1
0

6.6.3 Inner product

Proper orthogonal modes are form functions that are mutually orthogonal with respect
to mass and stiffness distributions, i.e. to scalar products defined as:
Z L
< wi , wj >m = wi mwj dx1
0
Z L
< wi , wj >s = wi′′ EJwj′′ dx1
0
Consequently, proper orthogonal models can be used to decouple the equations
of motion for systems where only elastic and inertia forces are considered.

6.7 Modal decomposition

Consider the case where the solution is represented as:
∞
X
w(x, t) = wi (x)qi (t)
i=1

The PVW tells us:

Z L Z L
′′T ′′
δw EI2 w dx1 = δwT (−mẅ + q3 ) , dx1
0 0

Substituting we get:
 Z L 
2
δqi µi (q̈i + ωi qi ) − wi q3 dx1 = 0 ∀i
0

Note that the forcing term for each equation is the work of distributed eternal forces
along each modal shape.

57 Structural Dynamics and Aeroelasticity

Since this has to be true ∀δqi we get an infinite amount of decoupled equations, one for
each mode.

Each mode is energetically decoupled from the others. No energy transmission from one
to the other is possible.

Physically, this equation states that every mode behaves like a single d.o.f. oscillator of
mass µi and frequency ωi ; the generalized force is once again the work of the external
distributed force p on the mode wi .

6.8 Ritz-Galerkin approximation

We can now move to a finite dimension space using the Ritz-Galerkin approximation.
For any unknow displacement field w(x, t) we can write:
n
X
w(x, t) ≈ ŵ(x, t) = Ni (x)qi (t)
i=1

Remeber:

• Ni must be linear independent (they have to be a base of the functional space).

• They have to be complete (the method has to be convergent).

Defining the error:
1/2
εn = ||w(x, t) − ŵ(x, t)||, where ||v|| =< v, v >m,s

lim εn = 0
n→∞

Note: To be complete the set of functions Ni must satisfy the kinematic boundary
conditions (Essential or Dirichlet).

Let’s now take a particular set of ortogonal shape function so that:

< Ni , Nj > = ||Ni ||2 δij

Then we can write: X

w(x) = αi Ni (x)
i

Premultiply by Ni :
< Ni (x), w(x) >= αi < Ni (x), Ni (x) >
< Ni , w >
αi =
||Ni ||2

58 Structural Dynamics and Aeroelasticity

6.8.1 Decoupling in Ritz
Let’s now apply this to the PVW. Since we have a finite number of variable we can
organize them in a matrix system, as done in the DOFs model.

Mq̈ + Kq = Fp

By using the orthogonality we have only diagonal matrices:

..
     
.. ..
 .   .  R . 
M= µ  K= µ ω 2  Fp =  Ni qi dxq 
i i i
..
     
.. ..
. . .

6.8.2 FEM
We can subdivide the domain and apply for each the Ritz formulation. Then we can
proceed to assemble the matrix of the system starting from all the matrices of the single
elements.

6.8.3 Ritz-Gakerkin vs FEM

Ritz Galerkin

• Shape functions with global support

• Matrices M,K are full

• When geometry is complex the definition of appropriate function may be difficult

• The prediction of local effects may require the inclusion of local shape functions
who may be difficult to define

• With well defined shape functions the convergence is fast (with few dofs)

FEM

• Shape functions with local support

• Matrices M,K are sparse

• Easy(-er) to apply to complex geometries

• The prediction of local effects may be obtained with local refinement of the number
of elements

• Availability of lots of tool (commercial/free/opensource software)

59 Structural Dynamics and Aeroelasticity

7 Modal Analysis
In a general case we are not able to find an analytical solution. In general:

Mq̈ + Kq = 0, q = 0 ∈ RN

Where:
q = ϕi ejωi t
ϕi i-th eigenvector (eigenmode)

(−Mωi2 + K)ϕi ) = 0

Kϕi = Mϕi ωi2

Note: The eigenmode is a vector that multiplied by K is equal to itself multiplied by
matrix M and the eigenvalue λ = ωi2 . Since it’s true ∀i:

KΦ = MΦΛ

Φ = [ϕ1 , ...., ϕN ]
Λ = Diag(ωi2 )
Note: The eigenvector is a direction for which the elastic force is equal and opposite to
the inertia force. If we move along the eigenvector we generate a movement with zero
work.

7.1 State-space form

˜ = Aq̃
q̇
(A − λI)ϕ̃i ) = 0
Aϕ̃i = λi ϕ̃i
AΦ = ΦΛ
Φ = [ϕ̃1 , ...., ϕ̃2N ]
Λ = Diag(λi )
We can diagonalize A by:
Λ = Φ−1 AΦ

7.2 Normalization
We can normalize a modal shape unless for a constant:

1. Maximum displacement max(ϕi ) = 1

2. Modal mass ϕTi M ϕi = µi = 1

60 Structural Dynamics and Aeroelasticity

7.3 Decoupling through eigenmodes
Mϕi ωi2 = Kϕi
Multiplying:
ϕTj Mϕi ωi2 = ϕTj Kϕi
Or
ϕTi Mϕj ωj2 = ϕTi KϕJ
Subtracting:
Note: ϕTj Mϕi is a scalar
(ωi2 − ωj2 )ϕTj Mϕi = 0
From which: orthogonality wrt mass matrix

ϕTj Mϕi = 0 if i ̸= j

And summing: orthogonality wrt stiffness matrix

ϕTj Kϕi = 0 if i ̸= j

Then we can define a modal coordinate:

q = Φz

Rewriting:
MΦz̈ + KΦz = 0
And
ΦT (MΦz̈ + KΦz) = 0
But these are all diagonal matrices. Therefore:

ΦT MΦ = Diag(µi )

ΦT KΦ = Diag(µi ωi2 )
We get N decoupled scalar equations:

µi (z̈i + ωi2 zi ) = 0 ∀i

7.4 State space form

Define:
q̇ = qd
We have:
Mq̇d + Kq = 0

61 Structural Dynamics and Aeroelasticity

But we can rewrite:
˜ = K̃q̃
M̃q̇
Since M̃ is diagonal and M > 0 is positive defined: ∃M̃−1
˜ = M̃−1 K̃q̃
q̇
Or
˜ = Aq̃,
q̇ q̃ ∈ R2N

7.4.1 Decoupling
Just as before we can write (droppings the tilde)
ϕTj Aϕi = λi ϕTj ϕi

ϕTi Aϕj = λj ϕTi ϕj

And
(λi − λj )ϕTi ϕj = 0
That means:
ϕTi ϕj = 0, i ̸= j
We can also write:
ΦT Φ = I
Φ is orthogonal.
ΦT = Φ−1
So we diagonalize like:
Λ = ΦT AΦ
This means that:
q = Φz =⇒ ż = ΦT AΦz
We get N different scalar equations:
żi = λzi ∀i

7.5 Coincident eigenvalues

Typically we have them where we have symmetry.

We define:
• Algebraic multiplicity: the number of times an eigenvalue is repeated
• Geometric multiplicity: is the dimension of the eigenspace Ev associated with an
eigenvalue λ
When the AM and GM are the same we can still say that the eigenvalues are orthogonal
and we can decouple the problem through diagonalization.

62 Structural Dynamics and Aeroelasticity

7.6 Rigid body modes
Characterized by:
ωi = 0
φTj Mφi ωi2 = φTj Kφi = 0
So there is NO DEFORMATION (and no energy associated): φTj Kφi = 0.

For a free-flying aircraft the K matrix is 6 times singular: 6 rigid body modes.

It is very important to note that because ωi = 0, we cannot say anything about φTj Mφi ,
therefore rigid body modes may not be orthogonal with respect to mass. The only
condition for which the modes are orthogonal is when we consider as degrees of freedom
the rotation and translation of the CG.

7.7 Gram-Schmidt orthogonalization

The initially computed rigid body modes may not be linearly independent, therefore we
want to compute modes that diagonalize the matrices through the GRAM-SCHMIDT
algorithm.

If φTj Mφi ̸= 0, define:

φ˜j = φj − αφi
So that:
φ̃Tj Mφi = 0 =⇒ φTj Mφi − αφTi Mφi = 0

Substituting we can find α a:

φTj Mφi
α=
φTi Mφi

7.8 Computation of the solution to initial conditions

ˆ
Mq̈ + Kq = 0, q ∈ RN , q(0) = q̂, q̇(0) = q̇
The general solution is the superimposition of modes: φi ejωi t .
N
X
q= φi (Ai cos ωi t + Bi sinωi t)
i=1

Exploiting the definition of modal mass and the orthogonality of the modal shapes with
respect to mass we can write:
N
X
φTk Mq = φTk M φi (Ai cos(ωi t) + Bi sin(ωi t)) = φTk Mφk (Ak sin(ωk t) + Bk cos(ωk t))
i=1

63 Structural Dynamics and Aeroelasticity

Applying the IC t = 0:
φTk Mq̂0 φTk Mq̂˙ 0
Ak = Bk =
µk ωk µ k
So:
N
!
X φTi Mq̂0 φT Mq̂˙ 0
q= φi cos(ωi t) + i sin(ωi t)
µi ωi µi
i=1

For a rigid body mode ωi = 0 therefore cos(ωi t) → 1 and sin(ω

ωi
i t)
→ t. Therefore we can
decompose the solution in rigid modes and flexible modes:
r N
! !
X φTk Mq̂0 φTk Mq̂˙ 0 X φTi Mq̂0 φTi Mq̂˙ 0
q= φk + t + φi cos ωi t + sin ωi t
µk µk µi ωi µi
k=1 i=r+1

Note: The rigid body modes make the system unstable because the solution grows
indefinitely with t.

7.9 Rayleigh quotient

We can extend the concept of the Rayleigh quotient to the discrete case:
Z L
EI2 (wi′′ )2 dx1
φTi Kφi
ωi2 = Z 0
L
−→ ωi2 =
φTi Mφi
m(wi )2 dx1
0

We then consider a generic vector u and we consider it to be a linear combination of

eigenvectors that represent the basis of the vector space of the solution.
n
X
u= αk φk = φα
k=1

We then compute the Rayleigh quotient of the vector u.

uTi Kui (Φα)T KΦα

R(ui ) = =
uTi Mui (Φα)T MΦα

Supposing ω1 > ω2 > ... > ωN , then:

n
X α 2 ω 2 µk
k k
1+
α12 ω12 µ1
k=2
R(ui ) = ω12 n
X αk2 µk
1+
α12 µ1
k=2

Therefore if u is close to φ1 , then we are going to obtain an approximation of ω1 .

64 Structural Dynamics and Aeroelasticity

7.9.1 Error on the eigenvalue
If u = φi + δu is an approximation of the true eigenvector. Substituting inside the
definition of R.Q. .
φT Kφi
R(ui ) = Ti + O(||δu||2 )
φi Mφi
If the approximated eigenvector has an error of O(||δu||) then the eigenvalue calculated
with the Rayleigh Quotient has an error of O(||δu||2 ).

This means that the Rayleigh Quotient is a good approximation of the associated eigen-
value ωi .

7.10 Computations of a group of eigenvectors & eigenvalues

We can use this to write an algorithm to compute the approximation of associated
eigenvalues:
u = Φ̃α u = φi + δu

R(u) = R(φi ) + O(∥δu∥2 )

Where Φ̃ is an approximation of the proper orthogonal modes matrix (it is a limited set
of approximated modal shapes).

Because the Rayleigh quotient is quadratic in the error, then the first derivative of the
quotient with respect to u will be null. Therefore:
dR(u) dR(u) dR(u) du
= 0 =⇒ = =0
du dα du dα

uT Ku αT Φ̃T K Φ̃α αT K̃α

R(u) = = =
uT M u αT Φ̃T M Φ̃α αT M̃ α
Because Φ̃ is a limited set of approximated modal shapes, therefore it does not represent
exactly the modal basis, the tilde matrices will be full matrices.

dR(u) 2K̃ααT M̃ α − 2αT K̃αM̃ α

= =0
dα (αT M̃ α)2
!
1 αT K̃α
K̃α − M̃ α =0
αT M̃ α αT M̃ α
Since αT M̃ α ̸= 0:
αT K̃α
λ̃i = =⇒ K̃α = λi M̃ α
αT M̃ α
Computing the eigenvalues of the reduced and approximated mass and stiffness matrix
it is possible to obtain an approximation of the eigenvalues.

65 Structural Dynamics and Aeroelasticity

7.10.1 Bloch-Stodola block iteration
1. Start with an initial guess Φ̄0 that is of size [N × p] where p is the number of
eigenvalues that we want to approximate.

2. for k = 0, . . . , m, we have KΦ̄k+1 = MΦ̄k . This corresponds to the application of

the power method. We find Φ̄k+1

3. Find: K̄ = Φ̄Tk+1 KΦ̄k+1 ; M̄ = Φ̄Tk+1 MΦ̄k+1

4. Solve K̄α = λM̄α → (λi , αi ) i = 1, . . . , p

5. φ̄newi = Φ̄k+1 αi =⇒ Φ̄new = [φ̄new1 , φ̄new2 , φ̄new3 , . . . , φ̄newp ]

6. Restart

In this way we are able to calculate a set of eigenvalues and associated eigenvectors,
in this case, we are interested in studying all of the modes and not just evaluating the
divergence pressure value.

66 Structural Dynamics and Aeroelasticity

8 Damping models
All structures have a mechanism that leads to the dissipation of energy. The dissipation
is responsible for the decay of harmonic oscillations when set in motion.
There are several sources of dissipation:

• Dissipation associated with the straining of the material, due to viscoelastic be-
haviour of the constitutive law (very limited in a metallic material, more relevant
in composite structures due to matrix and to micro-sliding of fibers).

• Dissipation due to friction for sliding inner joints, support connections, etc. This
is extremely important for complex structures made of several parts.

8.1 Kevin-Voigd model (Parallel)

Spring and damper in parallel.

Figure 36: Kelvin-Voigdt damper model

F = Fe + Fd = ku + cu̇

ue = ud = u
Solution of the equation:
If F = F̄ = const.
F̄ −k/ct F̄
u(t) = − e +
k k
If u = ū = const.
F = kū

8.2 Maxwell model (Series)

Spring and damper in series.
F = Fe = Fd
Ḟ F
u̇ = ue + ud =⇒ u̇ = +
k c

67 Structural Dynamics and Aeroelasticity

 Figure 37: Maxwell damper model

Solution of the equation:

If F = F̄ = const.  
1 t
u(t) = F̄ +
k c
If u = ū = const.
kt
F = kūe− c
Typically materials have a mixed behavior in parallel and series.

8.3 Equivalent Damping

The behavior of dissipation is complex, nonlinear, not-deterministic, and characterized

by a dependence on many parameters and operative conditions, consequently, it is ex-
tremely difficult to be modeled starting from first principles.

For small amplitude oscillations, the mechanism can be represented through a linear
contribution proportional to the time derivative of the free coordinates.

Fd = Cu̇

68 Structural Dynamics and Aeroelasticity

Find the equivalent damping by similitude with a harmonic damped oscillator.
Apply a harmonic input at the natural frequency ωn .

mẍ + cẋ + kx = F0 sinωn t =⇒ mẍ + 2ξmωn ẋ + ωn2 x = F0 sinωn t

In resonance:
x = x0 cosωn t
Substituting we get:
|F0 | |F0 |
|x0 | = =
ωn c 2kξ
Knowing ωn , we can find ξ:
|F0 |
ξ=
2k|x0 |

Figure 38: Hysteretic behavior of the structural damping

8.4 Energy analysis - meaning of equivalent damping ratio

Let’s see a procedure to compute the equivalent damping with an experiment.

Let’s consider the work of the reaction force R on the ground.

R = kx + cẋ
p
R = kx − cωn x0 quad(ωn t) 1 − cos2 (ωn t)
s  2
x
R = kx − cωn x0 quad(ωn t) 1 −
0

The work: I I I
Wc = R dx = kx dx + cẋ dx

69 Structural Dynamics and Aeroelasticity

Note: kx is conservative, its work in a closed path is null:
s  2
Z x0
x
Wc = 4 cωn x0 ωn 1 − dx
0 x 0

Wc = πcx20 ωn
Or we consider the elastic work from x to x0 , which is a quarter cycle.
Z x0
1
Wk = kx dx = kx20
x 2
Since we can measure both the WC and the Wk experimentally we can find the value of
ξeq .

8.4.1 Energy dissipation

Wc 1 πcx20 ωn cωn
eD = = = = 2ξ
2πWk 2π 0.5kx20 k
From which:
Wc
ξ=
4πWk

8.5 Modal damping

Mq̈ + Cq̇ + Kq, q = Φz
ΦT MΦz̈ + ΦT CΦż + ΦT KΦz = 0
The NORMAL DAMPING is defined as the case where the modal damping matrix is
diagonal.
ΦT CΦ = diag[2ξi µi ωi ]
In general, we are always going to consider this kind of damping as a first approximation.

8.6 Rayleigh damping

The Rayleigh damping is a particular case of normal damping where we say that the
matrix C is a linear combination of the mass and stiffness matrix:

C = αM + βK
 
1 α
=⇒ ξi = + βωi
2 ωi
The constants α and β could be computed starting from the experimental evaluation of
modal damping factors ξi .

The two constants are sufficient to obtain a perfect modal diagonal damping only if

70 Structural Dynamics and Aeroelasticity

two modes are considered, otherwise, a “best fit” that minimizes the error (eventually
weighted to better approximate the most important modes) should be used.

In many cases we can consider it being diagonal. Despite it being heuristic, it may be
sufficiently accurate.

8.7 Equivalent damping: Coulomb friction

Figure 39: Coulomb damping

This model is related to the pure friction-damping phenomenon.

Fd = sign(ẋ)µN
1
Wk = kx20 Wc = 4µN x0
2
2µN
=⇒ ξ =
πkx0
In order to verify if the damping ratio behaves as such, we can experimentally increase
the amplitude of the oscillation.

We can see how the damping depends on the amplitude of the oscillation x0 in this
model. This is, in practice, not true. We have to correct ”manually” this behavior by
removing this dependency.

8.8 Hysteretic damping

The viscous damping does not simulate satisfactorily actual engineering materials. Many
materials, when subjected to cyclic loading, exhibit a type of internal damping causing
energy losses per cycle that are proportional to the square of the amplitude and inde-
pendent of the frequency.

71 Structural Dynamics and Aeroelasticity

If Wc = πhx20 , then:
Wc h
ξ= = ··· =
4πWk 2k
We don’t have any more dependency on the amplitude of oscillation.

And
h
Fc = 2mξωn ẋ = ... = ẋ
2ωn
In frequency domain:
Fc = jhx
ξ
mẍ + k(1 + jη)x = 0, η =
2
Where η is the loss factor.

72 Structural Dynamics and Aeroelasticity

9 Dynamic Response
We are considering the system:

Mq̈ + Cq̇ + Kq, q = Φz

q = q0

q̇ = q̇0


The solution of this system will be in the form of:

Homogeneous Solution (H) + Particular Solution (P)

or

Free Response (X) + Forced Response (F)

Note: In general, H ̸= X and P ̸= F. With no forcing term then H ≡ X and at steady

state then P ≡ F

We are dealing with a system of differential equations in matricial form. The direct
solution in the time domain is generally impossible. We have to find a way to solve our
problem.

9.1 Laplace transform

Definition: Z ∞
L[f (t)] = F = f (t)e−st dt s = σ + jω, s ∈ C
0
Note:
L[f˙(t)] = sF (s) − f (0)
L[f¨(t)] = s2 F (s) − sf (0) − f˙(0)
Applying it to our system we get (If q0 = q̇0 = 0 then we can solve the forced response):

(s2 M + sC + K)q(s) = F(s)

We can rewrite:
q(s) = (s2 M + sC + K)−1 F(s)
Where we define the transfer function G(s) as:

G(s) = (s2 M + sC + K)−1

73 Structural Dynamics and Aeroelasticity

9.1.1 State-space Format
We can write the problem in state-space form therefore transforming the problem into
a first-order differential equation.
(
q̇ = qd
Mq̇d + Cqd + Mq = F

Calling:  
q
z= , F = Tu
qd
We can rewrite it as:
   
0 I 0
ż = z+ u(t) ⇐⇒ ż = Az + Bu(t)
−M−1 K M−1 C M−1 T

And the OUTPUT:

y(t) = Cz + Du(t)
Note:

• In order to retrieve a specific generalized coordinate. I am going to set:

D = 0 ; C = [0 . . . 0 1 0 . . . 0].

• In order to retrieve the displacement at a specific location xi I am going to set:

D = 0 ; C = [N(xi ), 0 . . . 0]

• In order to retrieve the velocity at a specific location xi I am going to set:

D = 0 ; C = [0 . . . 0, N(xi )]

• In order to retrieve the acceleration at a specific location xi I am going to write:

y(xi ) = a(xi ) = N(xi )q̈(t) = [−NM−1 K − NM−1 C]z + NM−1 Tu = Cz + Du

We have the space-state system:

(
ż = Az + Bu(t)
y(t) = Cz + Du(t)

Similarly as before we can apply the Laplace transform and get:

y
G(s) = = C(sI − A)−1 B + D
u
Define:

• POLES: eigenvalues of A. Found for the denominator of G(s) going to zero.

• ZEROES: yi = 0 even with ui ̸= 0. Found for null numerator of G(s)

74 Structural Dynamics and Aeroelasticity

Each coefficient of the transfer function is a ratio of polynomials and they all have the
same denominator.

In general, applying the Laplace transform is not very useful because it is difficult to anti-
transform and go back to the time domain. We generally cannot say that the function
in the Laplace domain is a linear combination of notable solutions of which we know the
inverse.

9.2 Frequency response

Note: There is a major distinction concerning what we did with the Laplace transform.
In this case, we are limiting the forcing term to a harmonic function and we are implicitly
assuming that the system is asymptotically stable (in the long run the free response tends
to zero and the particular integral is going to be meaningless). Let’s go back to our forced
system.

Let’s take a harmonic forcing term:

F (t) = F1 cos ωt + F2 sin ωt = F0 cos(ωt + φ0 )

Since:
ejα + e−jα
cos α =
2
=⇒ F (t) = F̂ ejωt + F̂ ∗ e−jωt , F̂ ∈ C
I can therefore compute the generic complex response to a generic complex input. Then
by summing the response with its complex conjugate I can retrieve the correct real
response.

9.2.1 Dynamic stiffness

The general solution is a harmonic function q(t) = q̂ejωt .

(−ω 2 M + jωC + K)q̂ejωt = F̂ejωt

Or
q̂ = (−ω 2 M + jωC + K)−1 F̂
Where we call dynamic stiffness:

Kdyn (ω) = (−ω 2 M + jωC + K)

In this case, the analogue of the transfer function is the dynamic compliance:

Ĥ(ω) = K−1
dyn (ω)

75 Structural Dynamics and Aeroelasticity

 (a) Amplitude - log scale (b) Frequency

9.2.2 One DoF

mẍ + xẋ + kx = f
k
Ĥ(ω) =
−ω 2 m + jωc + k

9.2.3 Modal computation of the response

We need to understand how to efficiently compute the inverse of the dynamic stiffness
matrix. To do so we are going to exploit a change of coordinates from generalized
coordinates to modal coordinates. Let’s start with the system:

(−ω 2 M + jωC + K)q̂ = F̂

Going back to the modal coordinates q = Φz and pre-multiply by ΦT :

ΦT (−ω 2 M + jωC + K)Φẑ = ΦT F̂

So:
Diag(µi (−ω 2 + jω2ξi ωi + ωi2 ))ẑ = ΦT F̂
By inverting and changing variable again back to q̂:
 
1
q̂ = Φẑ = Φ ΦT F̂
µi (−ω 2 + jω2ξi ωi + ωi2 )

But we said that:

q̂ = Ĥ(ω)F̂

76 Structural Dynamics and Aeroelasticity

It derives that:
N
X φi φTi
Ĥ(ω) =
i=1
µi (−ω 2 + jω2ξi ωi + ωi2 )

We can also define the dynamic amplification factor D̂ :

1
D̂i (ω) =  2
1 + j2ξi ωωi − ω
ωi

Figure 41: Dynamic amplification effect for for different ξ

Getting:
N
X φi φT i
Ĥ(ω) = D̂i (ω)
i=1
µi ωi2

9.2.4 Contribution of different terms

Fixed ωi (natural frequency of the i-th mode) evaluate the contribution of each mode to
the solution.

• for terms where ω << ωi the contribute is almost static (left of graphic).

• for terms where ω ≈ ωi the contribute is subject to a dynamic amplification (the

lower the damping the higher the amplification: see fig. 41.

• for terms where ω >> ωi the contribute is not significant.

We can cut the modal form up to a certain frequency.

77 Structural Dynamics and Aeroelasticity

9.3 Fourier Series
Consider a periodic f (t) signal for which:
Z t+T
|f (τ )| dτ < ∞
t

We can represent using the Fourier Series:

∞
a0 X 2π
f (t) = + (an cos nΩt + bn sin nΩt), Ω =
2 T
n=1

Or
∞
X
f (t) = ĉn ejΩnt , ĉ−n = ĉ∗−n
n=−∞

Where: Z T /2
1
ĉn = f (t)e−jΩnt dt
T −T /2

We can apply this to our case:

∞
X ∞
X
q(t) = q̂n ejΩnt F(t) = F̂ejΩnt
n=−∞ n=−∞

q̂n = Ĥn (nΩ)F̂n

We then know that for nΩ → ∞, Ĥ(nΩ) → 0, therefore we can truncate the series at N
such that Ĥ(N Ω) > toll.
∞
X N
X
q̂ = Ĥ(nΩ)F̂n ejΩt ≃ Ĥ(nΩ)F̂n ejΩt
n=−∞ n=−∞

9.3.1 Non-periodic but periodic-sizable signals

Sometimes we have a particular kind of signal that decays after a certain time. Define
td the decay time, i.e. the time after which the structural dofs q will decay below a set
threshold After time td we can restart the excitation without introducing errors.

It is possible to apply the “Fourier series approach” to the ”now-period” signal.

9.4 Fourier Transform

Can we apply what we studied here to a non-periodic or harmonic signal?
Let f (t) be a non periodic but s.t. f (t) ∈ L1 or
Z ∞
|f (τ )| dτ < ∞
∞

78 Structural Dynamics and Aeroelasticity

Note: physically this means that we want a stable system.

If so we can define the Fourier Transform of f

Z ∞
F[f (t)] = f (t)e−jωt dt
−∞

Note: the Fourier transform is a restriction of the Laplace transform where s = jΩ:

F[f (t)] = f (ω) = f (ω)|s=jω = L[f (t)]|s=jω

We can then demonstrate that:

F[f˙(t)] = jωF (ω)

F[f¨(t)] = −ω 2 F (ω)
Note: We are losing every information on the initial conditions. This is not a problem
if the system is stable (supposed a priori), since the homogeneous term dies out.
This is a very important distinction from the Laplace transform. This is why we require
that f (t) ∈ L1 .

Applying it to our system we get (remember that a forced response has q0 = q̇0 = 0):

(−ω 2 M + jωC + K)q(ω) = F(ω)

We can rewrite:
q(ω) = (−ω 2 M + ωC + K)−1 F(ω)
Where we define the frequency response function H(ω) as:

H(ω) = (−ω 2 M + jωC + K)−1

And
q(ω) = H(ω)F(ω)
We can now anti-transform the solution:
Z ∞
−1 1
q(t) = F [H(ω)] = H(ω)F(ω)ejωt dω
2π −∞

This integral is easily solved using a quadrature formula, limiting the domain up to a
significant frequency ωmax and exploiting the symmetry of the domain of integration.

1 ωmax
Z
q(t) ≃ H(ω)F(ω)ejωt dω
π 0

79 Structural Dynamics and Aeroelasticity

 Figure 42: Example of limited bandwidth excitation

9.5 Limited band excitation

Let’s suppose that the forcing term is limited in bandwidth. As told in 9.2.4 we can cut
off modes from a certain point (let’s say from m + 1), considering that they are going to
respond statically.
m N
X φi φTi X φi φTi
Ĥ ≃ D̂ i (ω) + 1
µ ω2
i=1 i i
µ ω2
i=m+1 i i

Remember:
   
T −1 1 −1 1
Φ KΦ = Diag(µi ωi2 ) =⇒ (Φ KΦ)T
= Diag =⇒ K = ΦDiag ΦT
µi ωi2 µi ωi2

Since we call this the static gain, then:

N
X φi φT
Ĥ(0) = i
1 = K−1
i=1
µi ωi2

Therefore we can rewrite H as:

m m
X φi φT i −1
X φi φT i
Ĥ ≃ 2 D̂i (ω) + K − 1
i=1
µ i ωi i=1
µi ωi2

So our response q(ω) is:

m m
!
X φi φT X φi φT
q(ω) = ĤF(ω) ≃ i
D̂i (ω)F(ω) + K−1 − i
1 F(ω)
i=1
µi ωi2 i=1
µi ωi2

80 Structural Dynamics and Aeroelasticity

Calling:
m
!
X φi φT
R= K−1 − i
1
i=1
µi ωi2
Or
m
X φi φT  
q(ω) ≃ i
2 D̂i (ω) − 1 F(ω) + K−1 F(ω)
i=1
µ i ω i

Note: qqs = RF(ω) is called the quasi-static correction and takes into account the
behavior of the high-frequency modes that behave statically.

The system’s response could be approximated by accounting for the dynamic response
of low-order modes included in the bandwidth plus the static response of the others.

9.6 Inertia Relief

When there is one or more rigid body modes this procedure does not work because the
matrix K is singular and not invertible (it’s singular).

Consider simplifying the expression taking C = 0 (no damping) and the modal forms
normalized at unit mass (so, µi = 1∀i). Separate the DoFs into rigid and elastic DoFs.
q = qe + qR = Φe ze + ΦR zR
It is possible to compute the acceleration that is necessary to satisfy the equilibrium of
rigid body equations (pre-multiplying the structural equation for the transposed of the
rigid body modes).
MΦR z̈R + MΦe z̈e + KΦe ze = F
No elastic energy is associated with rigid body modes.
ΦTR MΦR z̈R = ΦTR F
Unit mass normalization ΦTR MΦR = 1. Therefore we obtain the following important
result:
z̈R = ΦTR F
Substituting this in the original equation:
MΦe z̈e + KΦe ze = F − MΦR z̈R
MΦe z̈e + KΦe ze = F − MΦR ΦTR F
MΦe z̈e + KΦe ze = I − MΦR ΦTR F = PT F

This is the projection matrix:

P = I − ΦR ΦTR M

MΦe z̈e + KΦe ze = PT F

The projection matrix has the two following properties:

81 Structural Dynamics and Aeroelasticity

 1. The term P T F works only for the elastic modes: ΦTR PT F = 0, indeed:
PΦR = ΦR I − ΦR ΦTR MΦR = ΦR − ΦR = 0 =⇒ ΦR ⊥ PT F

2. The work of P T F for the elastic modes is the same as F for the same modes.
ΦTe PT F = ΦTe F =⇒ ΦTe PT = ΦTe =⇒ PΦe = Φe

So we can just say:

ΦTe MΦe z̈e + ΦTe KΦe ze = ΦTe PT F = ΦTe F

How does this help us? We are now able to constrain the system to the ground with
dummy constraints without affecting the elastic modes.
m
!
−1
X φi φTi
RF = K − 2 1 F(ω) = q0e + qi
µω
i=1 i i

Since the system is in equilibrium as a rigid body it is possible to add dummy constraints
to the ground to remove rigid body modes. They are not working anyway and their
reaction on ground is always null.
K̃ = K + ΦR Kg ΦR
Now K̃ is not singular so ∃K̃−1 It can be demonstrated that it is valid ∀K̃
At this point we can find:
q0e = K̃−1 F = K̃−1 PT F
The term q0e corresponds to the static elastic modes which appear in the quasi-static
correction approach.
q0e = PK̃−1 PT F = PK−1 PT F

9.7 Anti-resonance
Consider an undamped system ξi = 0. The transfer matrix is going to be:
m
X φi φT i
Ĥ ≃ D̂i (ω) + R
i=1
µi ωi2
With:
1
D̂i (ω) =  2
1 + j2ξi ωωi − ω
ωi

Of such transfer function we want to study the collocated transfer function which relates
the displacement of node k with the force applied to the same node. Therefore of the
vector φi we just consider the k th component.
m
X φ2ik 1
Hkk =  2 + Rkk
µ ω2
i=1 i i
ω
1− ωi

82 Structural Dynamics and Aeroelasticity

When ω → ωi+ then Hkk → +∞ and ω → ωi− then Hkk → −∞. Between two consecutive
natural frequencies, there is going to be always a zero of Hkk , and such zero is called an
anti-resonance because to every input the output is null. Anti-resonance depends only
on the frequency and not on how the forces are applied. For damped systems, the poles
and zeros are on the imaginary axis.

Figure 43: Resonance and anti-resonance

9.8 Solution through convolution

9.8.1 Convolution
The convolution technique allows us to solve the problem directly in the time domain
when the Fourier transform is not applicable, as an example in the case of the response
to a step input. Let’s define the operation ′ ∗ ′ as:
Z ∞
p(t) = f (t) ∗ g(t) = f (τ )g(t − τ )dτ
−∞

It is possible to demonstrate that:

F[f (t) ∗ g(t)] = F (ω)G(ω)
And
p(t) = F −1 [F (ω)G(ω)]
This is useful since we know that Q(ω) = H(ω)F(ω) So:
Z ∞
q(t) = F −1 [H(ω)F(ω)] = h(t) ∗ f (t) = h(τ )f (t − τ ) dτ
−∞

Where:
h(t) = F −1 [H(ω)]

9.8.2 Response to the impulse

Let f (t) be the Dirac delta δ(t). Then:
Z ∞
q(t) = f (τ )δ(τ ) dτ = f (0)
−∞

83 Structural Dynamics and Aeroelasticity

 Z ∞
q(t) = h(τ )δ(t − τ ) dτ = h(t)
−∞

So the function h(t) is the response of the system to an impulse in time t=0.

What is the impulse response?

Let’s use the Laplace transform on 1 DoF system
(
mẍ + cẋ + kx = f0 δ(t)
x0 = ẋ0 = 0

This is:
(s2 m + sc + k)x = f0 =⇒ (s2 m + sc + k)x − f0 = 0
Remember:
L[f¨(t)] = s2 F (s) − sf (0) − f˙(0)
So we can also think of this system as a free response with a non-null initial condition:

mẍ + cẋ + kx = 0
x0 = 0, ẋ0 = f0
m

Time response using the impulse response

The computation of the response using the convolution integral of the time history of the
input force times the impulse response is an application of the superimposition principle.
The final response is the superimposition of an infinite sequence of impulse responses
with impulses of different intensities.
Z t
q(t) = f (τ )h(t − τ ) dτ
0

9.9 Response using the step response

Let’s define the step function:
(
1 t>0
s(t) =
0 t<0
Note that: Z t
s(t) = δ(t) dτ =⇒ ṡ(t) = δ(t)
0
Define now the indicial function a(t) as:
Z +∞
a(t) = h(τ )s(t − τ ) dτ
−∞

84 Structural Dynamics and Aeroelasticity

Note: Z +∞
ȧ(t) = h(τ )δ(t − τ ) dτ = h(t)
−∞

Duhamel integral: response using the indicial function

Z ∞ Z ∞
q(t) = h(t − τ )f (τ ) dτ = ȧ(t − τ )f (τ ) dτ
−∞ −∞

Integrating by parts:
Z t
f
q(t) = a(t)f (0) + a(t − τ ) (τ ) dτ
0 dτ

9.10 Truncation of the modal basis

Consider simplifying the expression taking C = 0 (no damping) and modal forms nor-
malized at unit mass. µi = 1∀i. We know that (remember 9.2.3):
 
1
q = Φz = Φ ΦT F
µi (−ω 2 + jω2ξi ωi + ωi2 )
But ξi = 0∀i, therefore:  
1
q = ΦDiag 2 ΦT F
ωi − ω 2
We also know that high modes give very little contribution. Divide the matrix into fast
and slow modes.
Φ = [ΦS | ΦF ]
It is reasonable to retain only the modes that are within the bandwidth of the input plus
a few “guard” modes to ensure enough precision (between 5 and 10 “guard” modes).
 
1
q ≃ ΦS Diag ΦTS F
ωi2 − ω 2

9.11 Recovery of internal forces

The recovery of internal stresses (or internal loads) is important to employ the results
of simulations for the sizing of structures.

σ = Cε
d
ε= u = Du ≃ DNq = Bq
dt
σ ≃ CBq
The internal loads are the integral of stresses. As such they are proportional to the
elastic forces in the structure K.q

85 Structural Dynamics and Aeroelasticity

9.11.1 Direct recovery
In this case, computed the displacements q we proceed directly to the computation of
internal loads.
     
1 T 1 T
Kq = K ΦS Diag 2 − ω 2 ΦS F + ΦF Diag ω 2 − ω 2 ΦF F
ωiS iF

Or
Kq = (Kq)S + (Kq)F
If the bandwidth is limited then we could think of neglecting the fast modes. Kq ≈
(Kq)S . We need to assess the validity of such an approximation.

Remember that the following is the solution to the homogeneous problem (7.3):

KΦ = MΦDiag(ωi2 )

For large ωi we know that ωi2 /(ωi2 − ω 2 ) → 1 then:

 2
ωiS

T T
Kq ≈ MΦS Diag 2 − ω 2 ) ΦS F + MΦF ΦF F
(ωiS

This shows that with the direct recovery, we cannot use the slow modes for the compu-
tation of the loads as the fast modes behave statically and this equation shows it.
The only particular case where the fast modes can be used is when the excitation F is
bandwidth-limited and F is spatially distributed in a regular way (as in lift loads). In
this way, the overall work done by the loads is going to be null.

9.11.2 Acceleration modes

The elastic loads are equal to the sum of all other loads applied to the system (external,
inertial and eventually damping if present:

Kq = F − Mq̈

In this equation we substitute q̈:

ω2
   
2 2 2 1 T
q̈ = ω q = ω HF = ω ΦDiag 2 Φ F = ΦDiag ΦT F
ωi − ω 2 ωi2 − ω 2

Therefore:

ω2
 
Kq = F + MΦDiag ΦT F
ωi2 − ω 2

86 Structural Dynamics and Aeroelasticity

 ω2
For large ωi we have that: → 0.
ωi2 − ω 2

ω2
   
T ω
Kq = F + MΦS Diag 2 − ω 2 ΦS F + MΦF Diag ω 2 ΦTF F
ωiS iF

ω2
 
≈ F + MΦS Diag 2 − ω2 ΦTS F
ωiS
This method allows for recovery of the static deflection (and so internal loads) of fast
modes because it is equivalent to applying back to the full stiffness matrix the instanta-
neous loads as static loads. What is missing is only inertia (and damping) forces due to
fast modes.

87 Structural Dynamics and Aeroelasticity

10 Random Vibrations
10.1 Motivation
Up to now, we considered only deterministic excitations to the considered structures,
i.e., excitation for whom at an assigned position in space and at a precise instant in time
the value of the excitation signal is perfectly known

There are cases where the excitation signaling terms of value assumed in space and time
are random and can only be defined through a probability of occurrence. In this case,
the excitation is called non-deterministic, and it is defined through a stochastic pro-
cess. The physical characteristics of a stochastic process are described by its statistical
properties.

An attempt to predict its value at a certain position and time can only be performed in
a statistical sense. An observed set of records cannot precisely be repeated, but it will
follow a certain pattern that may only be mathematically represented by statistics.

An observation of a stochastic process is called a realization,i.e., a sample signal none

of which can, with certainty, be repeated even if the conditions are seemingly the same,
and so it must be considered random. The collection of all possible realizations is called
ensemble or Stochastic (random) process.

If the excitation random process is the cause of another process, this will also be a
stochastic process. A stochastic input will provide a stochastic output.

10.2 Statistic quantities

10.2.1 Probability
Consider a finite set of events Li . ni is the number of times the event Li is repeated.
ni ni
The relative frequency is: fi = P = .
ni N
N : number of trials

Define the probability of Li as: Pi = P(Li ) = limN →∞ fi .

10.2.2 Probability distribution

This probability definition satisfies the three Probability Axioms defined by Kolmogorov:
1. P(Li ) ≥ 0 ∀Xi

2. Call Ω the whole space. Then P(Ω) = 1

3. If L1 , L2 , L3 , . . . are mutually exclusive events then:

P(L1 + L2 + L3 + . . . ) = P(L1 ) + P(L2 ) + . . .

88 Structural Dynamics and Aeroelasticity

10.2.3 Cumulative probability function (CPF)
10.3 Probability density function
The continuous counterpart of the relative frequency.
We have that:
dF (x)
p(x) =
dx
Z x1
F (x1 ) = p(x) dx
−∞
Probability on the whole domain:
Z ∞
p(x) dx = 1
−∞

10.3.1 Expected values

The expected values are equal to the ensemble or statistical average, also known as
mean. Z ∞
E[X] = µX = xp(x) dx
−∞

10.3.2 Variance and standard deviation

Variance:
2
σX = E[(X − µx )2 ] =⇒ σX
2
= E[X 2 ] − µ2x
Standard deviation: q
σX = 2
σX

10.3.3 Gaussian process

1 1 x−µ 2
p(x) = √ e− 2 ( σ )
σ 2π
Note: The Gaussian distribution only depends on two parameters: mean value and
variance.

10.3.4 Gaussian process: joint distributions

d2 P(X ≤ x, Y ≤ y) d2 F (x, y)
p(x, y) = =
dxdy dxdy
The two processes are independent if p(x, y) = p(x)p(y).

89 Structural Dynamics and Aeroelasticity

10.4 Process function of time
For a specific time t the random signal X(t) is a random variable with a cumulative
probability function that depends on t.

FX (x, t) = P(X(t) ≤ x)

The corresponding PDF is:

∂P(X(t) ≤ x)
p(x, t) =
∂t
The mean and the variance are defined as:
Z ∞
µx (t) = E[X(t)] = xp(x, t) dxx
−∞

2
σXX (t) = E[(X(t) − µX (t))2 ]

10.4.1 Auto-correlation and cross-correlation

It is defined as the joint moment as the joint moment of the random variable at two
instant in time t1 and t2 .

RXX (t1 , t2 ) = E[X(t1 ), X(t2 )]

Z ∞
RXX (t1 , t2 ) = x1 x2 p(x1 , x2 ; t1 , t2 ), dx1 dx2
−∞
Note
RXX (t1 , t1 ) = RXX (t1 ) = E[X 2 (t1 )]
It allows us to understand how statistics of the time history of a variable is related to
itself, shifted in time:

Cross correlation
RXY (t1 , t2 ) = E[X(t1 ), Y (t2 )]

10.4.2 Stationary process

Consider a process X(t). The process is said to be stationary if the probability density
doesn’t depend on time.
p(x; t) → p(x)
It is easy to see that the same is true for the mean:

µX (t) = µX

90 Structural Dynamics and Aeroelasticity

10.4.3 Time shift
Taking two instants of time:
p(x1 , x2 ; t1 , t2 ) = p(x1 , x2 ; t1 + ε, t2 + ε)
If we call ε = −t1 and call τ = t2 − t1 , then:
p(x1 , x2 ; t1 , t2 ) = p(x1 , x2 ; τ )
The correlation becomes only a function of τ .
RXX (t1 , t2 ) = RXX (τ )

10.4.4 Properties of the correlation

Symmetry:
RXX (τ ) = RXX (−τ )
Maximum in τ = 0.
RXX (0) ≥ RXX (τ ) ∀τ, τ ̸= 0
The auto-correlation is a function that has its maximum at τ = 0 and then decays
toward infinity.

10.5 Ergodic process

The determination of the statistics from an ensemble poses the same practical problems
related to the repetition of experiments. It is often difficult to have sufficient repetitions
to obtain an accurate estimation of probability densities. However, if the process is
stationary, it seems reasonable to take one realization and replace ensemble averages
with time averages. A stochastic process for which this exchange is possible is called
ergodic.

It is possible to start by taking the time history dividing it in several portions of size T
and use them as realizations.

Changing T it is possible to generate a sequence of random variables and the statistics

could be computed taking the limit of this sequence:
Z T
1
E[X] = lim x(t) dt
T →∞ 2T −T

Define this integral as:

Z T Z
1
E[X] = lim . . . dt = − . . . dt
T →∞ 2T −T

So: Z
RX (τ ) = E[X(t + τ )X(t)] = − x(t + τ )x(t), dt

91 Structural Dynamics and Aeroelasticity

10.5.1 Auto-covariance and Cross-covariance
The auto-covariance is the equivalent of the autocorrelation taking off the mean by the
signal.
kXX (τ ) = E[(X(t + τ ) − µX )(X(t) − µX ]
Z Z Z
kXX (τ ) = − x(t + τ )x(t)dt − µx − x(t + τ )dt − µx − x(t)dt + µ2x

kXX (τ ) = RXX (τ ) − µ2X

Note:
kXX (0) = RXX (0) − µ2X = σX
2

Cross-covariance

kXY (τ ) = E[(X(t + τ ) − µX )(Y (t + τ ) − µY ]

kXY (τ ) = RXY (τ ) − µX µY
Define the coherence:
kXY (0)
ρXY =
σx σY
We have that:
−1 ≤ ρXY ≤ 1
Note: If X and Y are Uncorrelated then ρXY = 0.

10.6 Auto-covariance of a difference

Let’s evaluate the square of the difference between the stochastic process x and the same
process shifted by t.
∆X(t) = X(t) − µx
E[(∆X(t) − ∆X(t + τ ))2 ] = · · · = 2 σXX
2


− kXX (τ )
This expected value will be null if the two-time histories are coincident. The larger the
value larger the difference.

The more rapidly the auto-covariance decays with τ the less the shifted time histories
are similar.

So more rapidly the auto-covariance decays and more casual or chaotic is the signal.

The auto-covariance allows us to characterize exactly how a stochastic process behaves,

in a sort of deterministic way through a function.

92 Structural Dynamics and Aeroelasticity

10.6.1 Auto-covariance as a mean to de-noise a signal
If Y (t) is a noisy signal so that y(t) = x(t) + n(t):

RY Y (τ ) = RX X(τ ) + 2RN X (τ ) + RN N (τ )

Since N is noise and it is random, we have that:

RN N → 0 RN X → 0

So:
RY Y ≈ RXX (τ )

10.7 Power Spectral Density (PSD)

If we are interested in processing the stochastic response in the frequency domain, it
is now possible to do the Fourier transform of an ergodic signal. So, the frequency
content of an ergodic signal may be found by looking at the Fourier transform of the
autocovariance: Z ∞
ΦXX (ω) = F[kXX (τ )] = kXX (τ )e−jωτ dτ
−∞
Since it is related to the square of a signal (E[X2 ])it is considered a measure of energy
content.

10.7.1 Properties
ΦXX ∈ R if kXX is EV EN
ΦXX ≥ 0
Z +∞
1
kXX = F −1 [ΦXX (ω)] = ΦXX (ω)ejωt dω
2π −∞
Since KXX (0) = σ 2 :
Z +∞
2 1
σXX = ΦXX (ω) dω
π 0

10.8 White noise

White noise is a process with uniform PSD. The autocovariance is the Dirac delta. This
process is not physically realizable because there is an infinite area (power) under the
spectrum.
ΦXX (ω) = 2πS0 − ∞ < ω < ∞
Z +∞
1
kXX = 2πS0 ejωt dt = S0 δ(τ )
2π −∞

93 Structural Dynamics and Aeroelasticity

 Figure 44: Different kinds of PSDs

Figure 45: White noise PSD and its autocovariance

10.8.1 Colored Noise (Band-limited white noise)

ΦXX (ω) = 2πS0 − ω0 < ω < ω0
Z +ω0
1 sin ω0 τ
kXX = 2πS0 ejωt dt = 2S0
2π −ω0 τ

Figure 46: Colored noise PSD and its autocovariance

10.9 Exponential correlation

β2
ΦXX (ω) = S0
β 2 + ω2
S0 β −βτ
kXX = e
2

94 Structural Dynamics and Aeroelasticity

11 Random response
Let’s remember that the response of an ergodic input is still an ergodic output.
Consider a SISO system: Z ∞
q(t) = h(τ ))F (t − τ ) dτ
0

11.1 Computation of the mean

11.1.1 Computation of the mean knowing the mean of the input
Z ∞ 
mq = E[q(t)] = E h(τ ))F (t − τ ) dτ
0
Z Z ∞ 
=− h(τ ))F (t − τ ) dτ dt
0
Z ∞
=⇒ mq = h(τ ))E [F (t − τ )] dτ
0
Z ∞
mq = mF h(τ )) dτ
0
Since: Z ∞ Z ∞
−st
H(s) = h(τ )e dτ =⇒ H(0) = h(τ ) dτ
0 0
Therefore:
=⇒ mq = H(0)mF

11.1.2 Computation of the mean with the second order formulation

Mq̈ + Cq̇ + Kq = F
Taking the mean:
ME[q̈] + CE[q̇] + KE[q] = E[F]
Defining:
E[q] = mq E[F] = mF
For an ergodic process q(t)the mean of a derivative of the process is null. This means
that:

Kmq = mF
So:
mq = K−1 mF

95 Structural Dynamics and Aeroelasticity

11.1.3 State space formulation
(
ẋ = Ax + Bu
y = Cx + Du
E[x] = mx E[y] = my E[u] = mu
(
0̇ = Amx + Bmu
my = Cmx + Dmu

=⇒ my = (−CA−1 B + D)mu = H(0)mu

11.2 Computation of the variance of the output using the impulse

response
Define:
∆q = q − mq , ∆F = F − mF
Z ∞
∆q = h(v)∆F dv
0
The auto-covariance of q is:
Z
kqq (τ ) = − ∆q(t)∆qT (t + τ ) dt

Substituting the expression for ∆q we can retrieve that (slide 11.7):

Z ∞Z ∞
kqq (τ ) = h(v)kF F(τ + v − w)hT (w) dv dw
0 0

The variance is the auto-covariance evaluated in τ = 0:

Z ∞Z ∞
σqq = kqq (0) = h(v)kF F(v − w)hT (w) dv dw
0 0

Note: To compute the variance of the output we need the impulse response and the
auto-covariance of the input.

11.3 Computation of the response in frequency domain

Z ∞
Φqq = kqq (τ )e−jωτ dτ
∞
By using the definition above of kqq (τ ) and substituting.

Φqq (ω) = H(−ω)ΦFF (ω)HT (ω)

This is the complex conjugate.

Φqq (ω) = H∗ (ω)ΦFF (ω)HT (ω)

96 Structural Dynamics and Aeroelasticity

For a SISO input:
Φqq (ω) = |H(ω)|2 ΦF F
Back-transforming:
Z ∞ Z ∞
1 1
2
σqq = Φqq (ω) dω = H∗ (ω)ΦFF (ω)HT (ω) dω
π 0 π 0

11.4 White noise approximation

Linear oscillator
mẍ + cẏ + ky = 0
Where x = x0 + y.
ÿ + 2ξωn ẏ + ωn2 = −ẍ0
The transfer function is easily computed in the time domain:
y 1
H(ω) = = 2
ẍ0 ω − ωn2 + j2ξωn ω
1
|H(ω)|2 =
(ω − ωn + 4ξ 2 ω 2 ωn2 )2
Consider the PSD of the base acceleration x0 as a white noise of intensity 2πS0 .
Φẍ0 ẍ0 = 2πS0
2πS0
Φyy (ω) = |H(ω)|2 ΦF F =
(ω − ωn )2 + 4ξ 2 ω 2 ωn2
And the variance:
1 ∞
Z
2
= σyy Φqq (ω) dω
π 0
This integral can only be computed numerically as there is no closed-form solution.
In order to verify the validity of the numerical result we exploit the impulse response
technique to compute the variance. The frequency response h(t) is computed with the
inverse Fourier transform of the transfer function H(ω).
sin ωd t
h(t) = e−ξωn t
ωd
S0
2
σyy =
4ξωn3
Note: Even if we have an input with infinite energy, the output has finite energy provided
that the system is damped.

Can we always use this approximation?

Shortly, no. We can only use it when the actual input PSD is almost constant and
reaches the maximum when |H(ω)|2 has a spike (which acts as a weighting function). If
the real PSD varies in the critical region (where |H(ω)|2 has a spike), then we are going
to make major errors by using the white noise approximation.

97 Structural Dynamics and Aeroelasticity

 (a) Applicable (b) NOT applicable

11.5 Lyapunov Equation

This method is used when we don’t want to pass into the frequency domain as we know
the PSD and the transfer function. This alternative approach to calculating the variance
of the output is based on the state space approach.
(
ẋ = Ax + Bu
y = Cx + Du

Defining as before:

∆x(t) = x(t) − mx ∆u(t) = u(t) − mu ∆y(t) = y(t) − my

(
˙ = A∆x + B∆u
∆x
∆y = C∆x + D∆u

Multiply by ∆xT and calculate the mean integral. We get:

2 2 2
σẋx = Aσxx + Bσux
2 = 0. The same CANNOT be said for multiple
Note: for 1 DoF we can say that σẋx
DoFs. The matrix is going to be skew-symmetric.

We can say that:

2
σẋx = −σx2ẋ =⇒ σẋx
2
+ σx2ẋ = 0

By transposing the initial first equation and by applying the same procedure we obtain:

σx2ẋ = σxx
2
AT + σxu
2
BT (2)

98 Structural Dynamics and Aeroelasticity

Summing (1) and (2) it results, and remembering that σx2ẋ is skew-symmetric, then:
2 2
Aσxx + σxx AT + Bσux
2 2
+ σxu BT = 0 (3)
2 :
We then proceed to investigate the term σxu

2 2 T
σxu = σux

Exploiting the impulse response:

Z ∞
∆x(t) = h(τ )Bδu(t − τ ) dτ
−∞

Inserting the impulse response and taking the cross-variance:

Z Z ∞ Z ∞
2 T
σxu = − h(τ )Bδu(t − τ ) dτ δu (t) dt = h(τ )Bkuu (τ ) dτ
−∞ −∞

If u is a white noise then kuu = Wδ(τ ), then:

2 1
σxu = BW
2

2 T 2 1
=⇒ σxu = σux = WBT
2
Substituting in (3) we get:
2 2
Aσxx + σxx AT = BWBT (4)

This is the Lyapunov equation and it is a matrix equation that allows us to compute
the cross-variance if the states are subject to a white noise input.
2 we can compute the variance of the output:
Once found σxx

∆y = C∆x + D∆u
2 2
σyy = Cσxx CT + Cσxu
2
DT + Dσux
2
CT + Dσuu
2
DT
If the input is a white noise W, then:
   
2 2 T 1 T 1
σyy = Cσxx C + C BW D + D BW CT + DWDT δ(0)
2 2

Note: since the white noise has an infinite amount of energy (remember it’s only the-
oretical) we cannot have a direct connection from input to output. Basically we need
that D = 0. The relation becomes:
2 2
σyy = Cσxx CT

99 Structural Dynamics and Aeroelasticity

11.6 Shape filters
When the input is not a white noise, we can say that the real input is the result of a
shape filter system subject to a white noise as input.

Given an input vector u with its assigned PSD Φuu , find the transfer matrix of the
state-space filter Hf that returns the same PSD subjected to a white noise input w.

Φuu = |Hf (ω)|2 w

The transfer function Hf can be determined in the following way. The transfer function
will be governed by the following state space system:
(
ẋf = Af xf + Bf uf
u = Cf xf

Therefore the transfer function:

Hf (ω) = Cf (jωI − Af )−1 Bf

By adjusting the coefficients we can create the filter. Combining in a single state-space
state where we have an additional state which is the filter state xf . The input of the
system u is equal to yf = Cf xf .
      
ẋ A BCf x 0
 ẋ = 0 A + n


xf B

f f
 
 x
y = [C DCf ] x


f

We now have a new state-space system:

(
ẋa = Aa xa + Ba n
y = Ca xa

Apply the Lyapunov equation to this system to obtain the variance of the output.

Note: if the PSD is not a polynomial ratio we CANNOT COMPUTE the needed
shape filter. We must use the direct method.

11.7 Risk assessment

The information on the variance of the output can be used to assess the probability
of the output reaching a certain threshold. We can build a matrix of probability and
consequences of events. Different variance intervals represent different probabilities for
an event to happen.

In this way we can determine the likelihood of an event and evaluate its consequences.

100 Structural Dynamics and Aeroelasticity

 Figure 48: Risk matrix

Figure 49: Probability based on the distance kσ from the mean

11.8 Fatigue failure: Rice formula

When a structure is subject to random variation for a large number of cycles failure may
come from fatigue damage.

In order to compute the damage, we have to count the number of times the quantity of
interest crosses a certain threshold b.

We have to define a counting statistical process N(b). To do so it is possible to use the

Step (Heaviside) function H. (
1 x≥b
y(t) =
0 x<b
Take the derivative:
ẏ(t) = Ḣ(x(t) − b) = δ(x(t) − b)ẋ

101 Structural Dynamics and Aeroelasticity

 Figure 50: Events surpassing the threshold

The derivative of H will be different from zero every time H is discontinuous. We are
interested in the rate of threshold crossing is independent of the direction of crossing.
Therefore:
|ẋ|δ(x(t) − b)
The number of threshold crossing is equal to the integral of:
Z t2
N (b, t2 , t2 ) = |ẋ|δ(x(t) − b) dt
tq

We can compute the expected value for the number of threshold crossings per unit time.
Z +∞ Z +∞
N̄ = E[N (b, t)] = |ẋ|δ(x(t) − b)p(x, ẋ, t) = dx dẋ
−∞ −∞

The signal is ergodic and 2

σẋx= 0. We are also interested only in the positive crossing.
2 Z ∞ 2
1 − 21 b2 − 12 ẋ2
N̄+ (b) = e σ xx |ẋ|e σẋẋ dẋ
2πσxx σẋẋ 0
With a change of variable:
2
1 σẋẋ − 21 σb2
N̄+ (b) = e xx
2π σxx
If we compute the variance of the output signal x and the variance of its derivative
it is possible to compute the expected number of crossing per unit of time using this
expression that is called the Rice Formula.

11.8.1 Computation of the probability of crossing a threshold level b in a

finite time interval T
Every aircraft has a lifetime and we need to assess the probability of encountering loads
that are above a certain threshold during the lifetime.
N̄+ (b) dt → Probability of crossing in a infinitesimal time dt

102 Structural Dynamics and Aeroelasticity

 1 − N̄+ (b) dt → Probability of NOT crossing in a infinitesimal time dt
Considering:
 
T T
dt = lim =⇒ P (x < b, dt) = lim 1 − N̄+ (b)
N →∞ n N →∞ n

The probability in a time interval is:

T n
 
P (x < b, t ∈ [0, T ]) = lim 1 − N̄+ (b) = e−N̄+ (b)T
N →∞ n

The probability of crossing the level is going to be:

1 − e−N̄+ (b)T

The n exponent comes from the fact that if the events are independent then the combined
probability is the product of all probabilities for every event.

Clearly we are making the strong hypothesis that what happens at a certain time is
independent of what happens at other time instances.

103 Structural Dynamics and Aeroelasticity

12 Unsteady aerodynamics
The objective is to build a tool that can perform aeroelastic predictions. The structural
and aerodynamic operator models are not geometrically compatible therefore we require
an interface operator. The interface operator transfers the information from one model
to another.

12.1 Interface
To create the loop between the structural dynamics and aerodynamics we need to first
translate the information of the displacement of the structural nodes into the displace-
ment of the aerodynamic nodes to compute the aerodynamic loads.

dA = I(dB ) = I(N(x)q(t)) = I(N(x))q(t) = N (x)q(t)

Once computed the displacements we need to convert back the information of the loads
into concentrated loads applied to the structural nodes.
 
N
 Sy 
  Z 
 Sz 
  = IT P (x, t)A dS
T 
  S
My 
Mz

12.2 Unsteady aerodynamics phenomena

We need to understand how to model the unsteady aerodynamics.

The use of CFD is generally not the best option as it is very computationally expen-
sive. Furthermore, during the early design phase, there is no detailed information on
the geometry therefore the method is excessive. Another important drawback is that
the method depends on time marching integration therefore to pass into the frequency
domain we need to apply heavy post-processing. The use of CFD is therefore limited
only to viscous phenomena such as aileron buzz, buffeting, stall, etc.

In all other cases, we can rely on simplified models that are going to use simplified ge-
ometries (from full 3D geometry to 2D lifting surface) and simplified equations (from
NS to Euler or potential equations).


 Incompressible RANS

Compressible RANS

NS → (
 Incompressible Euler
Euler equations →


 
Full potential flow → Linearized potential

104 Structural Dynamics and Aeroelasticity

12.3 Spatial Filtering
In aeroelasticity, we are interested in the generalized loads which are the product of
distributed forces over the modal shapes. Indeed what counts when we are weighting
the distribution of forces by the change of shapes are the global effects and not the local
effects.

In aerodynamics instead the local effects are extremely relevant. As an example, the
local pressure distribution may have major consequences on the rest of the airfoil (BL
stability).

This implies that in the aeroelastic models when we approximate the aerodynamic ge-
ometry as a flat plate and therefore lose the local effects, we need to be careful that we
work within the hypothesis of the approximation.

12.4 Material derivative

Let’s consider a Lagrangian setting.
If f = f (x(t), t) then it’s derivative:

df ∂f X ∂f ∂xi ∂f
= + = + u · ∇f
dt ∂t ∂xi ∂t ∂t
i

Often called material derivative.

Df ∂f
= + u · ∇f
Dt ∂t

Note that df Df
dt = Dt is true only in a Lagrangian setting. Otherwise, the material deriva-
tive is not strictly a derivative.

12.5 Reynolds transport theorem

Z Z Z
d ∂f
f dV = dV + f u · n dS
dt V V ∂t ∂V

12.6 Euler equations

Euler equations are obtained from the conservation laws by applying the hypothesis of
negligible viscosity and negligible heat transfer effects.
dρ


 + ρ∇ · u = 0


 dt
 du 1

+ ∇P = 0

 dt ρ


 dE t 1

 + ∇ · (P u) = 0
dt ρ

105 Structural Dynamics and Aeroelasticity

In order to have a solvable system we need to add the two thermodynamic equations of
state for the gas. Hypothesizing an ideal polytropic gas:
(
P = ρRT
e = cv T

12.7 Rate of change of kinetic energy

From the momentum balance:
 
du 1
u· + ∇P = 0
dt ρ

d u2
 
1
= − u · ∇P
dt 2 ρ
For inviscid flows, the rate of change of kinetic energy is equal to the work done by
pressure forces (there is no dissipation through viscosity).

12.8 Entropy
Recall the 1st principle of thermodynamics:
 
1
de = T ds − P
ρ

We use the first principle of thermodynamics inside the equation of energy balance
(Euler).
Ds
= 0 =⇒ s = s0 = const
Dt

We find that in an inviscid fluid flow, the total time derivative of entropy in the flow is
null. This means that if there is no initial gradient of entropy in the fluid then the flow
will remain isoentropic.

12.9 Acceleration and Vorticity

Vorticity is defined as:
ω =∇×u
We can rewrite the momentum balance as:
u2
 
∂u 1
=ω×u+∇ = − ∇P
∂t 2 ρ

106 Structural Dynamics and Aeroelasticity

12.10 Circulation and Kevin’s theorem
When we have vorticity then we can define a quantity called circulation:
Z
Γ= ω · n̂ dS
S

Applying Stokes theorem:

Z I
Γ= ω · n̂ dS = u · dl
S ∂S

Now let’s differentiate Γ and exploit the relation between acceleration and vorticity and
using the momentum balance equation we get:
∇P ∇P
Z Z Z I
dΓ d
= ∇ × u · n̂ dS = ∇ × a · n̂ dS = − ∇ × · n̂ dS = · dl
dt dt S S S ρ ∂S ρ

Therefore:
∇P
I I
dΓ dP
= · dl =
dt ∂S ρ ∂S ρ
This means that an isoentropic flow that is also barotropic. In fact, P = P (ρ, s0 ) = P (ρ).
Then the flow will be irrotation, given an initial null vorticity.
I
dP dΓ dω dΓ
= 0 =⇒ = 0 =⇒ = lim =0
∂S ρ dt dt S→0 dt

dΓlm
=0
dt

Figure 51: Sudden start vortex

Kelvin’s theorem has a major implication in the phenomenon of the sudden start of a 2D
airfoil. Indeed for Kelvin’s theorem to be valid, the overall vorticity inside the boundary
∂S must be null therefore it must generate a vorticity that is equal and opposite to the
one created by the 2D airfoil, Γw .

107 Structural Dynamics and Aeroelasticity

12.11 Velocity potential
Excluding the airfoil from the domain, we can consider the flow to be irrotational and
given the irrotationality, we can define a velocity potential. u = ∇Φ. By defining the
velocity potential in this way, irrotationality will always be guaranteed as rot(grad( )) =
0.
∇ × ∇Φ = 0

12.12 Unsteady Bernoulli equation

12.12.1 Unsteady Bernoulli
We start from the Euler momentum balance equation plus the vorticity equation:
 2
∂u u ∇P
+∇ =−
∂t 2 ρ
To this equation, we add the hypothesis of irrotationality.
 2
∂∇Φ u ∇P
+∇ =−
∂t 2 ρ
We then add the hypothesis of inviscid flow (isoentropic) which means that:
Z
dP dP
T ds = dh − =⇒ h =
ρ ρ
Therefore:  
∂Φ 1 2 ∂Φ 1 2
∇ + u + h = 0 =⇒ + u + h = F (t)
∂t 2 ∂t 2

12.12.2 Unsteady Bernoulli for adiabatic flow

Given the hypothesis of isoentropic flow:
 
P dP γP
d γ
= 0 =⇒ γ − γ−1 dρ = 0
ρ ρ ρ
Therefore: Z    
dP γ P P∞
= ... = −
ρ γ−1 ρ ρ∞
Recalling that a2 = γRT and substituting this expression inside the unsteady Bernoulli
equation we get:
∂Φ u2 a2 u2 a2
+ + = ∞+ ∞
∂t 2 γ−1 2 γ−1

12.12.3 Pressure coefficient

2 ∂Φ |∇Φ|2
 
P − P∞
Cp = 1 2
=− 2 ∂t
+ 2
−1
2 ρ∞ U∞
U∞ U∞

108 Structural Dynamics and Aeroelasticity

12.13 Full potential equation
With Bernoulli, we have the expression which expresses the aerodynamic loads as a
function of the velocity potential. We now need to find the relation that governs the
velocity potential.

We start from the continuity equation:

dρ
+ ρ∇2 Φ = 0
dt
To this equation, we add two thermodynamic relations and the Bernoulli equation:
dP dP dρ
a2 = =⇒ dP = a2 dρ =⇒ = a2
dρ dt dt

d ∂Φ |∇Φ|2
 
dP dP dh
dh = =⇒ =ρ =ρ +
ρ dt dt dt ∂t 2
Recalling again the Bernoulli equation we can define an as:

∂Φ |∇Φ|2 − U∞ 2
 
2
a (Φ) = −(γ − 1) + + a2∞
∂t 2
Therefore by combining the two together, we get the full velocity potential equation:

ρ d ∂Φ |∇Φ|2
 
 2
 ρ∇ Φ = +
a2 (Φ) dt ∂t

 2
2 − U2
 
2 ∂Φ |∇Φ| ∞
+ a2∞

a (Φ) = −(γ − 1)
 +
∂t 2


In order to solve the problem we need to ass the boundary conditions, by defining the
lifting surface B(x, t) = 0, we then impose that the velocity of the fluid normal to the
body boundary is equal to the velocity of the boundary:

u · nˆB = ∇Φ · nˆB = vB · nˆB

For the condition at infinity, we just impose:

lim u = lim ∇Φ = U∞
x→∞ x→∞

Note: This equation is very complicated to solve and it results even more difficult than
resolving directly the Euler equations

12.14 Kutta condition

The potential problem with Neumann boundary conditions admits a unique solution only
in a simply connected domain. In the case of the airfoil problem, then, the potential
cannot be uniquely defined. We are going to need a condition which is able to define

109 Structural Dynamics and Aeroelasticity

uniquely the potential and such a condition is based on the recovery of the viscous effects:
the Kutta condition.

The Kutta condition is a condition that says that the flow exits the trailing edge regularly
without curving around it as it would be non-physical. This same condition can be seen
also as a condition on the regularity of pressure as the pressure at the TE is equal
between the top and bottom and it can also be seen as a condition that the vorticity at
the TE is null.

The Kutta condition allows us to impose a vorticity on the airfoil such that the trailing
edge becomes a stagnation point.

12.15 Linearized potential

12.15.1 Linearized potential - equation
We start by defining the linearized variables:

a = a∞ + δa
P = P∞ + δP
ρ = ρ∞ + δρ
u = U∞ + δu

Therefore:
Φ = U∞ x + φ
This directly implies that:
∇2 Φ = ∇2 φ
∂Φ ∂φ
=
∂t ∂t
Furthermore, we can exploit the following two approximations:
|u|2 U2 ∂φ
≈ ∞ + U∞
2 2 ∂x
d ∂ ∂
= + U∞
dt ∂t ∂x
Therefore using these approximations inside the potential equation, we get the linearized
potential equation:
 2
∂2φ 2

2 1 ∂ φ 2 ∂ φ
∇ φ= 2 + 2U∞ + U∞ 2
a∞ ∂t2 ∂t∂x ∂x
This equation is not valid only in the presence of strong shockwaves which are going to
break the irrotationality hypothesis.

110 Structural Dynamics and Aeroelasticity

Note: In the case of an incompressible flow a∞ → ∞ as perturbations propagate instan-
taneously over the whole domain therefore the equation becomes:

∇2 φ = 0

12.15.2 Linearized potential - boundary conditions

∂φ
∇Φ · n̂B ≈ U∞ nBx + = vB · n̂B
∂ n̂ B

12.15.3 Linearized potential - pressure coefficient

γ !
γ − 1 ∂Φ |∇Φ|2 − U∞ 2
   γ−1
1
Cp = 1 2
1− 2 + −1
2 γM∞
a∞ ∂t 2
 
2 ∂φ ∂φ 2 ∂φ
Cp ≈ − 2 + U∞ =− 2
U∞ ∂t ∂x U∞ dt

12.16 Linearized boundary conditions

We shall investigate more thoroughly the expression of the linearized boundary condi-
tions. First of all, we consider two instances in time t and t + dt. Therefore:

B(x, t) = 0 and B(x + dx, t + dt) = 0

We can then use Taylor expansion to say:

∂B
B(x + dx, t + dt) = 0 ≈ B(x, t) + ∇B · dx + dt = 0
∂t
Therefore:
∂B ∂B
∇ · B dx + dt = 0 =⇒ + ∇B · vB = 0
∂t ∂t
Furthermore:
∇B ∂B 1
vb · n̂B = u · ·n̂B =⇒ vb · = u · n̂B =⇒ − · = u · n̂B
|∇B| ∂t |∇B|
The final way to write the boundary condition is as follows:
∂B
+ u · ∇B = 0 on B = 0
∂t
At this point, we can divide the boundary into two parts: the bottom half and the upper
half. In this way, we can write the boundary conditions as:
∂b ∂b ∂φ

 u + U∞ u =

∂t ∂x ∂z
 ∂b l ∂bl ∂φ
 + U∞ =
∂t ∂x ∂z

111 Structural Dynamics and Aeroelasticity

In this way, the equation becomes fully linear as we have a linear equation and linear
boundary conditions in the variable bu and bl . This is a major result as at this point we
can apply to the system the superimposition principle. Indeed we can write the overall
perturbation potential as the superimposition of various potentials related to the chord,
the chamber line, the thickness, and deformations. The first three are going to impact
only the steady aerodynamics while in our case we are interested only in the last which
describes the unsteady aerodynamics.

We can write:
bu = N (x)q(t)
Therefore the linearized boundary condition becomes:
∂N ∂φd
N (x)q̇(t) + U∞ q(t) = =W
∂x ∂z
Note: We can even subdivide the deformation potential into sub-components which are
related to the individual shape functions.

Note: By dividing the boundary conditions by U∞ we can see that the first term relates
to the change in AoA due to kinematics while the second relates to a local change in
AoA due to the geometric deformation.
N (x)q̇(t) ∂N W
+ q(t) =
U∞ ∂x U∞

12.17 Solution using Biot-Savart

When dealing with incompressible irrotational flows we can use another technique to
solve the problem. This technique involves using the vector potential defined as:
⃗ :u=∇×ψ
ψ ⃗
Using the vector relations:
⃗ = ∇(∇ · ψ
ω =∇×∇×ψ ⃗ − ∇2 ψ
⃗ =⇒ ∇2 ψ
⃗ = −ω
Exploiting the Biot-Savart equation we are able to recover the velocity in every point in
the domain just by solving one integral. The real advantage of this technique is that we
can restrict the domain of integration only to where the vorticity is not null, therefore
just on the airfoil and on the wake. Therefore we just need to discretize the surface of
the airfoil and of the wake and not the whole domain.
ω(ξ) × (x − ξ)
Z
1
u(x) = dξ
4π V |x − ξ|3

12.17.1 Straight vortex in 3D

Consider a straight vortex of intensity Γ so that ω = Γ dl, and we consider the induced
velocity at point P which is identified by the vector r = x − ξ:
Γ
vP = k̂
2πd

112 Structural Dynamics and Aeroelasticity

12.17.2 Line of infinite length vortices

Figure 52: Line of distributed vortices

Let’s consider a line that goes from x = xLE , z = 0 to x = xT E , z = 0 of vortices of

infinite length with axis aligned with y. The vorticity per unit length is γ. Each vortex
generates the following induced speed on a plane perpendicular to the vortex axis:
1 ∆z
u= γ
2π ∆x + ∆z 2
2

1 −∆x
w= γ
2π ∆x2 + ∆z 2
Since there is a continuous array of vortices, the sum of all the effects is:
Z xT E
1 z
u(x, z) = γ(ξ) dξ
2π xLE (x − ξ)2 + z 2
Z xT E
1 x−ξ
w(x, z) = γ(ξ) dξ
2π xLE (x − ξ)2 + z 2
The induced speed along the vortical line z = 0 is:

u(x, 0) = 0
Z xT E
1 γ(ξ)
w(x, 0) = dξ
2π xLE x − ξ

113 Structural Dynamics and Aeroelasticity

13 2D unsteady airfoil theory

Figure 53: Evolution of the unsteady wake

13.1 Hypothesis
• 2D irrotational, incompressible flow undergoing unsteady motion in uniform flow

• Airfoil represented by a thin flap-plate (no thickness or camber effects)

• The shed wake is geometrically represented by a straight line from the TE to

infinity in the direction of the flow field

• Both airfoil and wake are planar sheets of vorticity

• Airfoil chord: 2b

• The position of the elastic axis is at ab (a non-dimensional position of the EA)

13.2 Boundary conditions

The normal velocity of the body should be equal to the direction of the flow in the z
direction.
∂η ∂η
vbz = + U∞ = wa
∂t ∂x
The velocity of the air wa is the sum of the velocity induced by wake vortexes λ and the
one induced by the bound vortex wb .

wa (x) = λ(x) + wb (x) ∀x

Where, applying the Biot-Savart rule:

Z b Z b
1 γb (ξ, t) 1 γw (ξ, t)
wb = dξ λ= dξ
2π −b x−ξ 2π −b x−ξ

114 Structural Dynamics and Aeroelasticity

 Figure 54: Sketch for the unsteady profile

13.3 Wake vorticity

We need to determine the two components of vorticity that influence the air velocity wa .
The bound circulation Γ over the airfoil responsible for lift is defined as:
Z b
Γ= γb dξ
−b

From the Kelvin theorem:

dΓT dΓ + Γw dΓw dΓ
= 0 =⇒ = = 0 =⇒ =−
dt dt dt dt
But take the definition of Γw and derivative with respect to time.
dΓw
dΓw = γw dx =⇒ = γ w U∞
dt
Putting it all together:
At every instant at the TE a vortex γw is released by the wake. The intensity is equal
and opposite to the bound vorticity variation according to Kelvin’s theorem.
This vortex is convicted at speed U∞ .
 
dΓ x−b
U∞ γw (x, t) = − t−
dt U∞
The vorticity in the wake at (x, t) is a function of the variation of vorticity on the airfoil
at time t − x−b
U∞ ,

115 Structural Dynamics and Aeroelasticity

i.e. when the vortex that traveled at U∞ speed was at the TE. This is a direct conse-
quence of Kelvin’s theorem.

13.4 Method of solution

We have found the relation between the potential jump at the TE and the potential
jump in the wake. We still have to find the overall solution. Therefore we first recall the
equation for wB :
Z b
1 γb (ξ, t)
wB = dξ = wA − λ
2π −b x − ξ
Plus the Kutta condition:
γB (b) = 0 ∀t
The solution is going to be a Fedholm integral equation of the first kind:

r s
b
b−x b + ξ wA (ξ, t) − λ(ξ, t)
Z
2
γB (x, t) = − dξ
π b+x −b b−ξ x−ξ
This, though, is not a proper solution as we still don’t know how to express λ which is
a function of γB . Therefore this is an implicit equation in γB .

13.4.1 Change of coordinates

We can further simplify the expression by expressing the velocities as a Fourier series.
Let’s use the following transformation.

x = b cos θ

Write:
∞
X ∞
X
λ(θ, t) = λn cos nθ wa (θ, t) = wan cos nθ
n=0 n=0

Where: Z π Z π
1 1
λn = λ(θ, t) cos nθ dθ wan = wa (θ, t) cos nθ dθ
π 0 π 0
Plugging these equations in (1) we can get:
∞
X
γb = 2 (wan − λn )fn (θ)
n=0

Where:   
 tan θ

n=1
fn (θ) = 2

sin(nθ), n ≥ 1

116 Structural Dynamics and Aeroelasticity

So that when we integrate it to find Γ:
Z b Z π
Γ= γb dξ = γb (θ) sin θ dθ
−b 0

Only the 0 and 1 components have a non-zero contribution to circulation.

   
1 1
Γ = 2πb wa0 + wa1 + λ0 + λ1
2 2

13.4.2 Decomposition of bound vorticity

It is possible to decompose the bound vorticity into:
- γbC : Circulatory bound vorticity
- γbN C : Non circulatory bound vorticity

γb = γbC + γbN C

In particular, γbC is responsible for Γ (0 and 1 terms) but with zero induced speed. So
that when we integrate it to find Γ:
Z b
Γ= γbC dξ
−b
Z b
1 γbC (ξ, t)
wb = dξ = 0 ∀x ∈ [−b, b]
2π −b x − ξ
   
2 1 1
γbC = wa0 + wa1 + λ0 + λ1
sin θ 2 2
On the other hand, for the non-circulatory component, it is exactly the opposite: it will
have no effect on the circulation but it will create an induced speed.
Z b
γbN C dξ = 0
−b
Z b
1 γbC (ξ, t)
wb = dξ = (wa − λ) ∀x ∈ [−b, b]
2π −b x−ξ
  ∞
2 1 X
γb N C = − (wa0 − λ0 ) cos θ + (wa1 − λ1 ) cos 2θ + 2 (wa n − λn ) sin nθ
sin θ 2
n=2

Note: The non-circulatory component depends on all components of w and λ with respect
to the circulatory component that depends only on the first two components.

117 Structural Dynamics and Aeroelasticity

13.5 Computation of load
From the unsteady Bernoulli theorem:
 
∂φ ∂φ
p(x, t) = −ρ U∞ +
∂x ∂t

We are interested in the jump of potential ∆p = pU − pL .

 
∂∆φ ∂∆φ
∆p(x, t) = −ρ U∞ + (1)
∂x ∂t

Where ∆φ is the jump of potential across the airfoil, due to the distributed vorticity.

Remember that:
γb = ∆u(x) = u(x, 0+ ) − u(x, 0− )
Then, we know that:
∂φ
u=
∂x
So:
∂∆φ
γb =
∂x
But: Z x Z x
∂∆φ
Γx = γb dξ = dξ = ∆φ =⇒ Γx = ∆φ
−b −b ∂ξ
This also implies that:
∂∆φ ∂Γx
=
∂t ∂t
Substituting in (1):  
∂Γx
∆p(x, t) = −ρ U∞ γb +
∂t
And  Z x 
∂
−∆p(x, t) = ρ U∞ γb + γb (ξ) dξ
∂t −b
We have the expression of γb :
∞
X
γb = 2 (wan − λn )fn (θ)
n=0

So:
∞
X
−∆p = pn fn (θ)
n=0

p0 = 2ρU∞ (wa0 − λ0 )

118 Structural Dynamics and Aeroelasticity

  
b b
pn = 2ρ U∞ (wan − λn ) + (ẇan−1 − λ̇n−1 ) − (ẇan+1 − λ̇n+1 )
n 2n
This dynamic system connects the distributed pressure to the dynamic of the velocity
on the airfoil (the sum of the velocity of the body and the velocity induced by the wake
or inflow velocity).

13.5.1 Loads
Now that we have the pressure along the chord we can calculate everything.
Z b ∞
X Z π
L(x, t) = − ∆p(ξ, t) dξ =⇒ L(x, t) = pn fn (θ)dθ
−b n=0 0

Note: Only the components multiplied by f0 and f1 have integrals over the chord different
from zero!

We can divide the lift as ;

L = Lw + LQ + LN C
Where Lw is the wake-induced lift, LQ is the quasi-static lift (the one we are used
to) and LN C is the non-circulatory lift, the one related to Γ = 0, also called added
mass as it behaves like an inertial force.

Note: It is possible to verify that only the non-circulatory components of bound vorticity
led to contributions to terms depending on derivatives with respect to time.

13.6 Solution in frequency domain

13.6.1 Wake vorticity in frequency domain
 
dΓ x−b
U∞ γw (x, t) = − t−
dt U∞
In frequency domain:
x−b
U∞ γw = −jωΓe−jω U∞
Remember the definition of reduced frequency:
ωb
k=
U∞
Introduce a change of variable x̃ = x/b. We know that:

∂∆φ
γb =
∂x
And
Γx = ∆φ

119 Structural Dynamics and Aeroelasticity

So:
∂∆φw ωb 1 ωb
= −j ∆φT E e−j U∞ (x̃−1)
∂x U∞ b
∂∆φw
= −jk∆φT E e−jk(x̃−1)
∂ x̃
The solution is easily computed.

∆φw = ∆φT E e−jk(x̃−1)

The jump of potential along the wake is the jump of potential in the trailing edge
computed with a delay in time: when the wake element was at the TE.

13.6.2 Solution
Z ∞
1 1
λ(x̃, t) = ∆φw (ξ, t) dξ
2bπ 0 x̃ − ξ
∆φw = Γe−jk(x̃−1)
Take now:  
∂∆φ ∂∆φ
∆p = −ρ U∞ + (1)
∂x ∂t
Transform it:  
−ρU∞ ∂∆φ
∆p = + jk∆φ
b ∂ x̃
The lift is:
LC = LQ + Lw
Where:
LC = ρU∞ G(k)Γ(k)
This is a generalization to the frequency domain of the Kutta-Joukowsky formula.

We define the Kutta-Joukowsky frequency response function G(k) as:

π (2)
G(k) = jkej(π+k )H1 (k)
2
(2)
Where H1 (k) is a Hankel function of the first time: see here: https://mathworld.
wolfram.com/HankelFunctionoftheFirstKind.html

120 Structural Dynamics and Aeroelasticity

13.7 Theodorsen model
This model is a simple case: harmonic plunge and pitch.

Remember η is the position of the elastic axis.

η(x, t) = h(t) + α(t)(x − ab)
∂η ∂η
wa = + U∞
∂t ∂
wa = ḣ + α̇(x − ab) + U∞ α
Change of coordinates:
wa = ḣ + α̇(b cos θ − ab) + U∞ α
Divide it into its components:
wa1 = α̇ = ḣ + α̇ab + U∞ α = α0 U∞
wan = 0 ∀n n ≥ 2
wa1 = α̇
wan = 0 ∀n ≥ 2
The movement is harmonic.
h(t) = ĥejωt α(t) = α̂ejωt
We can compute the integrals and end up with:
   
2
  1
L(k) = πρb ḧ + U∞ α̇ − baα̈ + 2πρU∞ bC(k) ḣ + U∞ α + b − a α̇
2
   
1 1 a
Mc/4 (k) = −πρb3 ḧ + U∞ α̇ + b − α̈
2 8 2
Note that we have a dependency of the C(k) function: the Theodorsen function.
(2)
H1 (k)
C(k) = F (k) + jG(k) = (2) (2)
H1 (k) + jkH0 (k)

13.7.1 Theodorsen function

The Theodorsen function is defined as:
C(k) = F (k) + jG(k)
• k = 0: We have a real value of the function therefore the phase shift is zero. This
indicates steady aerodynamics.
• k = 0.2: We have the maximum phase shift which indicates the maximum time
delay to change in boundary conditions.
• k → ∞: The phase shift will go to zero therefore behaving statically yet the
amplitude tends to 0.5.

121 Structural Dynamics and Aeroelasticity

 (a) G(k) (b) F(k)

Figure 56: Real and imaginary part

13.7.2 Lift variation

The introduction of an imaginary part means that the lift will respond with a certain
delay. The bigger the imaginary part, the higher the delay.

The maximum shift is at k = 0.2.

• For k = 0 Lift oscillates along a line: we only have a static behavior.

• For higher k the oscillation has a hysteretic behavior, as seen before. The CL lags
after the instantaneous angle of attack. This results in energy being pumped into
the system: we have the flutter.

• For k → ∞ we have a purely real C(k). The ellipsis collapses into a line again
with half the slope.

122 Structural Dynamics and Aeroelasticity

 Figure 57: Phase shift

This is what happens ONLY to the circulatory part

When we combine the circulatory component with the non-circulatory component we

obtain the following two graphs:

Figure 58: Effects on the normalized CL for a pure pitch oscillation around the LE

The non-circulatory part starts from zero amplitude and then increases almost quadrat-
ically (∝ ω 2 ) and the phase shift is always +90° because acceleration is 90° in advance
of speed. The anticipation of lift with respect to wing movement determines if we are
injecting or absorbing energy inside the system therefore crucial for flutter identification.

13.8 Cicala, Küssner-Schwarz approach for arbitrary deformations

The Theodorsen model is only valid for harmonic pitch and plunging. In reality, we can
compute the change of lift due to any input.

Cicala, Küssner, and Schwarz models can be used to obtain the response for an arbitrary
movement.

123 Structural Dynamics and Aeroelasticity

We have the same set of equations just with different inputs and outputs.
∞
X
wa (θ, t) = wan cos nθ
n=0
Z π
1
wan = wa (θ, t) cos nθ dθ
π 0
It is possible to verify that the jump of pressure coefficient across the airfoil is equal to:
∞
2 X
∆Cp (θ, t) = Cpn fn (θ)
U∞
n=0

With: 
Cp0 = −wa1 + C(k)(wa0 − wa1 n = 0
Cpn = wan + jk (wan+1 − wan−1
2n
 
2 jk
L = 2πρU∞ (wa0 − wa1 )C(k) + (wa0 − wa2 )
2
 
2 jk
M0 (k) = πρU∞ (wa0 C(k) − wa1 (1 − C(k)) − (wa1 − wa3 ) − wa2
4
This approach could be used if the airfoil changes shape. For instance, a change of
camber is the one that is generated when a movable surface is rotated by an angle β.

13.9 Wagner approach: time domain solution

In parallel to Cicala and others, Wagner studied the direct solution in time domain.
The solution is perfectly compatible with the anti-transformation of the Theodorsen,
etc. models.

Response to a step change in α. Define the indicial function (or Wagner function) ϕ:
Z +∞
2 C(k) jkτ 2
L(τ ) = ρbU∞ α e τ = 2πρbU∞ ϕ(τ )α
−∞ jk

With τ = U∞ t/b the non-dimensional time and ϕ(τ ) Wagner function.

An approximation of ϕ is from RT Jones (1940)

ϕ(τ ) = 1 − 0.165e−0.455τ − 0.335e−0.30τ

To get the overall lift and sum the circulatory and non-circulatory components, we
exploit the Duhamel convolution for the circulatory part.
 Z τ 
2 dα
L(τ ) = LN C (τ ) + 2πρbU α0 ϕ(τ ) + ϕ(τ − σ) dσ
0 dσ

124 Structural Dynamics and Aeroelasticity

13.10 Lumped vortex panel method
We are going to replace the distributed vorticity over the airfoil with a single lumped
vortex at c/4 and the collocation point at 3c/4 (heuristic considerations). At every instant
in time, we are going to release a vortex in the stream which is proportional to the
variation of bound vorticity (the one on the profile):
i
dΓ X
= 0 =⇒ Γ(ti ) + Γwk = 0
dt
k=1

To do so, at every instant in time we are going to find the speed at the collocation point
by imposing the boundary condition we impose:
Γi
|vi | =
2πd
We then compute lift as:
Z b  
∂Γ
L= −∆P dx = ρ U∞ Γ(t) + 2b
−b ∂t

This is going to be the overall lift.

The airfoil can be split into portions called panels and the same procedure can be applied
to each panel. Therefore we are going to impose as many boundary conditions on the
collocation points as is the number of panels (and therefore of lumped vortices).

Another thing we can do is to compute the induced velocity in the location of every
lumped vortex in the stream and therefore consider also the motion of the vortex. This
is called the ”free wake vortical method”.

Figure 59: Lumped vortex scheme

125 Structural Dynamics and Aeroelasticity

13.11 Frozen gust
An external velocity could also cause a change of velocity along the profile. The normal
velocity is usually called gust. Generally the vertical speed is a function of space. Define
the gust speed vG as a frozen distribution of speed fixed in space.

vG = vG (ξ)

The aircraft, initially at a distance x0 from the gust front, travels at U∞ . Call x the
position of a point of the aircraft on the local reference frame.

ξA = U∞ t − (x − x0 )

Consequently, the gust speed in the local reference system of the a/c is:

vG (x, t) = vG (ξA ) = vG (U∞ t − (x − x0 ))

13.11.1 Frequency domain

Transform: Z +∞
vG (x, ω) = vG (U∞ t − (x − x0 ))e−jωt dt
−∞

Change of coordinates: t → ξ, t = ξ+x+x

U∞
0

So: Z +∞ ξ+x+x0 1
vG (x, ω) = vG (ξ))e−jω U∞ dt
−∞ U∞
Defining now x̃ = x/b and ξ¯ = ξ/b:
Z +∞
−jk(x̃+x̃0 ) b ¯ −jkξ̄ dξ¯ =
vG (x̃, ω) = e vG (ξ))e
U∞ −∞

vG (x̃, k) = e−jk(x̃+x̃0 ) W (k)

Note that wee have the gust speed spectrum W (k) multiplied by a spatial delay e−jk(x̃+x̃0 ) .

13.11.2 Gust
The normal velocity of the body minus the gust should be equal to the velocity of the
flow in z;
vbz − vG = wa
Since the problem is linear we can say:

wa = −vG

Note the minus sign due to the different reference system.

126 Structural Dynamics and Aeroelasticity

Consider the case where the gust spectrum is W (K):

vG (x̃, k) = e−jkx̃ W (k)e−jkx̃0

Making the usual change of variable x̃ = cos θ:

vG (x̃, k) = e−jk cos θ W (k)e−jkx̃0

We can express the first term as a Fourier series through a Bessel Function of I kind:
∞
X
ejz cos θ = J0 (z) + 2 j n Jn (k) cos nθ
n=1

We can represent the gust speed where wn is composed of different Fourier components
due to the gust speed.
∞
!
X
vG (θ, k) = −wa (θ, k) = −W (k)e−jkx̃0 J0 (k) + 2 (−j)n Jn (k) cos nθ
n=1

13.11.3 Sears function

This way, we can express the total lift as dependent on the gust speed.

wa0 = −W (k)J0 (k)

wan = −2W (k)(−j)n Jn (k), ∀n > 0

L(k) = πρ2bU∞ W (k)ψ(k)

b
M0 (k) = L(k)
2
The function ψ (Sears function) is defined as:

ψ(k) = [J0 (k) − jJ1 (k)]C(k) + jJ1 (k)

It starts as C(k) and then is transformed into a more difficult expression due to the
exponential term of delay. Combining the movement in the circle of the exponential and
the C(k) function we get the spiral movement in the figure.

127 Structural Dynamics and Aeroelasticity

 Figure 60: Sears Function

We get that the response to the plunge and the gust differ due to the ex-
ponential term. Physically what’s changing is the so-called penetration effect: the
gust speed perceived at the TE is different than the gust speed in the other points of
the profile.

When k is small the exponential term is not very important and we can consider every-
thing without the exponential: the gust response is equivalent to the plunge response.

At high k the perceived effect is as if the TE is plunging in a different way than the LE
edge.

13.11.4 Final result

For every lifting surface airfoil, we can compute the Lift and Moments generated due
to the movement of the airfoil. The frequency response function for these forces and
moments is:    
L h
 M  = Ham (k) α + Hag (k)vg
Mh β
Where Ham (k) is associated with the Theodorsen model and Hag is associated with the
gust.

128 Structural Dynamics and Aeroelasticity

14 Morino’s method
14.1 Introduction
• The methodology stems from a smart usage of Green’s theorem (or divergence
theorem).

• The advantage is to lead to a very simple integral expression that can be easily
solved using a numerical approach

• Several authors worked on this formulation. However, the most complete work was
developed by Prof. Morino who applied it to subsonic and supersonic flows. This
is why here it will be nominated as “Morino’s method”

• This method represents a prototype for all possible panel methods in 2D/3D com-
pressible/incompressible subsonic/supersonic flows

14.2 Green’s identity

Starting from the divergence theorem
Z Z
∇ · b dv = n · b ds
V ∂V

In general, we can write

b = φ∇G
G is a generic scalar function
Applying the Divergence theorem
Z Z
∇ · φ∇G dv = n · φ∇G ds
V ∂V
Z Z
∇ · G∇φ dv = n · G∇φ ds
V ∂V
We get the Green’s identity
Z Z  
2 2 ∂φ ∂G
(φ∇ G − G∇ φ) dV = G −φ dS
V ∂V ∂n ∂n

Consider now to take as function G the fundamental solution to a Laplace problem:

∇2 G = δ(|x − x0 |)

The solution in 3D is
1
G=−
4π|x − x0 |

129 Structural Dynamics and Aeroelasticity

Using this particular choice of G we can verify that
Z Z
2
φ∇ G dV = φδ(|x − x0 |) dV = E(x0 )φ(x0 )
V V

Where 
0
 if x0 ∈
/ V,
E(x0 ) = 1 if x0 ∈ V and x0 ∈
/ ∂V,

1
2 if x0 ∈ ∂V.
Using this in the Green’s identity we get
Z   Z
∂φ ∂G
E(x0 )φ(x0 ) = −G +φ dS + φ∇2 G dV
∂V ∂n ∂n V

This expression can be used to compute the potential φ when the solution G is known.

14.3 Incompressible, steady flow

Firstly since the flow it’s incompressible

∇2 φ = 0

Then, the boundary can be divided in

∂V = Σ ∪ B ∪ W
Or

Figure 61: Boundaries for Morino’s method

Z Z Z
Eφ(r0 ) = . . . dS + . . . dS + . . . dS
∂Σ ∂B ∂W
The integral in the volume disappears as the integral on the external boundary Σ.
Z   Z  
∂G ∂φ ∂G ∂φ
E(x0 )φ(x0 ) = φ −G dS + φ −G dS
B ∂n ∂n W ∂n ∂n

130 Structural Dynamics and Aeroelasticity

Boundary conditions
∂φ ∂η ∂η
= + U∞
∂n B ∂t ∂x
The wake is aligned with the flow

∂φ
uW · nW = 0 −→ =0
∂n W

We get Z Z Z
∂G ∂φ ∂G
E(x0 )φ(x0 ) = φ − G dS + φ dS
B ∂n B ∂n W ∂n

14.3.1 Analysis of the wake

The wake is unloaded: null jump of pressure.

∆Cp |W = 0

Through Bernoulli
 
2 ∂∆φ ∂∆φ
∆Cp |W =− 2 U∞ + =0
U∞ ∂x ∂t W

The flow is steady: time derivatives are null.

2 ∂
(φU − φL )W = 0
U∞ ∂x
The jump of potential is null along x direction: The potential is constant on the wake

(φU − φL )W (x, y) = (φU − φL )W (x = T E, y) = ∆φT E (y)

The jump of potential on the wake at the TE is equal to the circulation at the span
station y as the Kutta condition says

∆φT E (y) = Γ(y) ∝ L(y)

For x0 ∈ V and x ∈ R3
Z   Z Z  
1 1 ∂ 1 1 1 ∂φ 1 ∂ 1
φ(x0 ) = − φ + dS − ∆φ dS
2 4π B ∂n r 4π B r ∂n 4π W ∂n r

The obtained integral expression can be used to compute the potential on every point in
the flow field when we know the potential on B or to compute the potential on B itself.

131 Structural Dynamics and Aeroelasticity

 Figure 62: Discretization for a steady flow

14.3.2 Discretization
1. Divide the wing surface in N = Nx × Ny panels and the wake in Ny semi-infinite
strips.

2. Define the type of approximation of φ over each panel (i.e., 0th-order: φ constant,
1st-order: φ linear, . . . ).

3. Decide how and where to impose the BCs on each panel (i.e., for the 0th-order
panel, the best collocation point (the point where the boundary condition is ap-
plied) is at the center of the panel.

N Z   N Z Ny Z  
X ∂ 1 X ∂φk 1 X ∂ 1
2πφi = − φk ds + ds − ∆φT Ej ds
Bk ∂nk rik ∂nk Bk rik Wj ∂nj rij
k=1 k=1 j=1

Ny Z   N Z  
X ∂ 1 X ∂ 1
∆φT Ej ds = hk φ k ds
Wj ∂nj rij Wj ∂nk rik
j=1 k=1

0
 k is not a TE panel
hk = 1 k is a TE upper panel

−1 k is a TE lower panel


Note: It is not necessary to have potential values associated with the wake panels. The
jump of the potential of the wake panel is equal to the difference between the potential
of the corresponding TE panels (imposition of the Kutta condition).

14.3.3 System
We can write everything in a linear system

132 Structural Dynamics and Aeroelasticity

 ∂φ
Yφ = Z
∂n
Where Z   Z  
∂ 1 ∂ 1
Yij = 2πδij + dS + hj dS
Bj ∂n̂ j rij Wj ∂n̂ j rij
Z
1
Zij = dS
Bj rij

14.3.4 Computation of loads

2 ∂φ
Cp = −
U∞ ∂x
Approximate the potential as
X
φ ≈ φ̃ Ni (x, y)φ = Nφ
i

Define
∂ φ̃ ∂Nφ
= = Pφ
∂x ∂x
Remember the BC on the body (steady)

∂φ ∂η
= U∞
∂n B ∂x

So
∂η
Yφ = ZU∞
∂x
Then calling A = Z−1 Y
Aφ = U∞ α
Where
∂η
α=
∂x
Then
φ = A−1 U∞ α
Substituting in CP
2 ∂ φ̃ 2 2
CP = − =− Pφ = − PA−1 U∞ α
U∞ ∂x U∞ U∞
The pressure distribution is
Cp = −2PA−1 α

133 Structural Dynamics and Aeroelasticity

14.4 Incompressible, unsteady case
14.4.1 Wake analysis
The wake is unloaded: null jump of pressure.

∆Cp |W = 0

Through Bernoulli
 
2 ∂∆φ ∂∆φ
∆Cp |W =− 2 U∞ + =0
U∞ ∂x ∂t W

The flow is unsteady: time derivatives are NOT null.

Transform in the frequency domain

∂∆φ jω∆φ
+ =0
∂x U∞
The solution is easily found + BCS
jω
∆φ = Γ(y)e− U∞ x

Substitute: x̃ = x/b
∆φ = Γ(y)e−k(x̃−1)
The jump of potential along the wake panels is not constant anymore: it is equal to
the product of the jump of potential at the trailing edge multiplied by a delay operator
(the exponential term) that expresses the fact that the vorticity travels in the wake at
the asymptotic speed.

14.4.2 Discretization

Figure 63: Discretization for an unsteady flow

134 Structural Dynamics and Aeroelasticity

We now have to divide the wake into panels. We get:
Z   Z  
∂ 1 −jk(x̃−1) ∂ 1
Yij = 2πδij + dS + hj e dS
Bj ∂n̂ j rij Wj ∂n̂ j rij
Z
1
Zij = dS
Bj rij

The system is
 
∂φ ∂η jω
Yφ = Z = ZU∞ + η
∂n ∂x U∞

14.4.3 Loads
 
2 jω
Cp = − Pφ + φ
U∞ U∞

14.5 Compressible flow, steady case

Prandtl-Glauert transformation:
∂2φ
∇ 2 φ = M∞
2
∂x2
With an intelligent change of variables, we can get:

∇2 φ̂ = 0

With: φ̂ = φ(x̂, ŷ, ẑ). Where

x
x̂ = , ŷ = y, ẑ = z
β
p
Having defined β = 2
1 − M∞

Problem formulation
We have to define the radius:
rβ2
r̂2 = (x̂ − x̂0 )2 + (ŷ − ŷ0 )2 + (ẑ − ẑ0 )2 =
β2
The sign of rβ2 depends on the nature of the flow: positive if subsonic, negative if super-
sonic.
Z   Z Z  
∂ 1 1 ∂φ ∂ 1
2πφ(x0 ) = φ ds − ds + ∆φT E ds
B ∂n rβ r
B β ∂n W ∂n rβ

Z   Z   Z
∂ 1 ∂ 1 1
Yik = 2πδik + ds + hk ds Zik = ds
Bk ∂nk rβik Wk ∂nk rβik Bk rβik

135 Structural Dynamics and Aeroelasticity

14.6 Compressible flow, unsteady case
1 d2 φ
∇2 φ − =0
a2∞ dt2
This is a modified Laplace problem.
The fundamental solution is:
1 d2 G
∇2 G − = δ(|x − x0 |)δ(t)
a2∞ dt2

In frequency domain
ω2
∇2 G + G = δ(|x − x0 |)δ(t)
a2∞
The general solution:
1 jωτ
G=− e
4πrβ
Where τ is the time required by the perturbation to travel from x0 to x:

Explanation of τ

For a fluid at rest s

r2
τ=
a2∞
If the fluid is at rest independently from the relative position between x and x0, the time
delay is the ratio between the distance and the speed of sound

For a fluid moving at U∞

If the fluid is moving at a constant speed, after the time τ x0 will move to a different
place so the delay will be different in the different directions.
In general, we have two solutions
1 −M∞ ∆x ∓ rβ
τ=
a∞ β2

14.6.1 Subsonic flow

For M∞ < 1 (fig: 64) we always have two solutions τ1 > 0 and τ2 < 0 not acceptable.

1 rβ − M∞ ∆x D + M∞
τ = τ1 = =
a∞ β2 U∞
So
1 −j aω D+
G=− e ∞
4πrβ

136 Structural Dynamics and Aeroelasticity

 Figure 64: Subsonic τ

We get:
+
Z   +
Z  
jω D ∂ 1 D
−jk(x̃−1) jω a∞ ∂ 1
Yik = 2πδij + e a
∞ dS + hj e e dS
Bj ∂n̂ j rij Wj ∂n̂ j rij
Z
D+ 1
Zik = ejω a∞ dS
Bj rij

14.6.2 Supersonic flow

Figure 65: Mach cone

For M∞ < 1 we could have:

• τ1 , τ2 < 0: no solution acceptable. The point is outside of the Mach cone and is
not reached by the perturbation
• τ1 , τ2 > 0: two acceptable solutions. The point is within the Mach cone, and it is
reached by two waves that started at different times from different positions

137 Structural Dynamics and Aeroelasticity

14.6.3 Supersonic flow panelization

Figure 66: Discretization for a supersonic flow: cone of influence

Consider a panel A in the wing. Define the domain of dependenceand domain of

influence. Panel A influences the panels within the Mach cone directed downward and
is influenced only by the panels included in the Mach cone directed upward.

From a panelization point of view, the perturbation will not completely affect all of the
panels. Indeed the cone of influence will affect partially some panels therefore in order
to discretize correctly we are going to need small enough panels.

Additionally, if we use a 0-order approximation because we have jumps in potential

across each panel, every jump in potential is going to create spurious shock waves. In
order to avoid this, we need to use a 1st order approximation, where the potential varies
linearly across the panel.

14.7 Piston theory

When the Mach number is very large the region of influence and dependence are very
small, so it is reasonable to consider the variation of pressure on every point only function
of the movement of that point (not influenced by others).
Every local point functions as a “piston”.

If M >> 1 2γ
 
p γ−1 w γ−1
= 1+
p∞ 2 a∞
w
If a∞ >> 1 then with a first order expansion

1 2 2 w
p − p∞ = ρ∞ a∞ w = U∞
2 M ∞ M∞
2
Cp = α
M∞

138 Structural Dynamics and Aeroelasticity

And the ∆p
4
∆p = q α(x)
M∞

139 Structural Dynamics and Aeroelasticity

15 The Aeroelastic problem
We now want to solve the complete aeroelastic problem in this order:
1. Discover if the system is dynamically stable and assesses stability and no flutter
2. Investigate further to find the response and the divergence condition

15.1 Computation of flutter speed

Dynamic pressure q is the main scaling factor for dynamic loads: it depends on the speed
but also on the altitude.
We need to use the Equivalent airspeed to consider the density too.
r
ρh
UEAS = UT AS
ρ0
And
UT2 AS ρh UT2 AS 2q
MT AS = = · · · = 2 =⇒ = Ph
a∞2 γMT AS γMT2 AS
We can define the flutter envelope.

Figure 67: Flight envelope and flutter boundaries

15.1.1 Flutter envelope

An aircraft has its flight envelope. We have to make sure that an aircraft does not show
flutter inside the flight envelope.
To keep into account the uncertainty, we have to increase the flutter boundary by 15%
from the manufacturer flight envelope.

Note that we have a negative altitude. It could happen that we are flying in a higher-
density air. This can be represented in the standard atmosphere by defining negative
altitude. We have to prove that within the green envelope, we have no flutter.

Where does it typically have flutter? In the high EAS low altitude (high q)

140 Structural Dynamics and Aeroelasticity

15.2 Assembly of the aeroelastic system
Consider a structural model where a set of generalized coordinates describe the displace-
ment of any structural note η
η = Nq
The structural model will be:

Mq̈ + Cq̇ + Kq = qFa

The vector Fa . We need to understand how to represent this vector.

Using the PVW Z
δWa = q δηaT · na Cp (xa ) dSa
As
Where As is the aerodynamic surface where the aerodynamic grid xa is defined.
The vector ηa is the vector of the displacement if the aerodynamic nodes.

Using the defined interface scheme it will be possible to express:

ηa = Hηs

So it will result: Z
δWa = q δ(H∗ηs )T · na Cp (xa ) dSa
As
Z
δWa = qδqT NT ηsT · na Cp (xa ) dSa
As
Or:
XZ
T
δWa = qδq NT ηsT · na Cp (xai ) dSa
i A si

δWa = qδqT DCp

Cp = P(k)φ

∂φ
φ = Y(k, M )−1 Z(M )
∂n
But:      
∂φ ∂ηa jω ∂ηs jω jk
= + ηa = H + ηs = H N/x + N q
∂n ∂x U∞ ∂x U∞ b
So:
∂φ
= (A + jkB)q
∂n
Substituting:
δWa = qδqT DP(k)Y−1 (k, M )Z(M )(A + jkB)q
=⇒ Fa (k) = Ham (k, M )q

141 Structural Dynamics and Aeroelasticity

15.3 Aeroelastic equation
We may want to solve the problem in the time domain and to do so we can use the
impulse response with the convolution integral:
Z t
Mq̈ + Cq̇ + Kq = q ham (t − τ )q(τ )dτ
0

This is theoretically feasible yet in practice not applicable.

The other technique is to apply the work in the frequency domain using the Laplace
transformation:
(Ms2 + sC + K − qHam (p, M ))q = 0
sb σb jωb
Where p = U∞ = U∞ + U∞

In general we don’t have the full transfer function Ham (p, M ) but we have only available
the frequency response i.e. Ham (k, M ).

We need to find methods to compute the correct solution to the eigenvalue problem
having this partial information.

15.4 Quasy-steady approximation

Using the McLaurin series around s = 0.

∂H s2 ∂ 2 H
H(s) = H|s=0 + s + + ...
∂s s=0 2 ∂s2 s=0

We get
s2 ∂ 2 H
 
∂H
Fa (s) = H|s=0 + s + + ... q(s)
∂s s=0 2 ∂s2 s=0
Using the inverse Laplace transformation:

∂H(0) −1 1 ∂ 2 H(0) −1 2
L−1 → Fa (t) = H(0)L−1 [q(s)] + L [sq(s)] + L [s q(s)]
∂s 2 ∂s2
∂H(0) 1 ∂ 2 H(0)
Fa (t) = H(0)q(t) + q̇(t) + q̈(t)
∂s 2 ∂s2
We know that
∂H(0) 1 ∂ 2 H(0)
H(0) = Ka = CA = Ma
∂p 2 ∂p2
Since H = H(p, M ) we have

∂H ∂H ∂p ∂H b
= =
∂s ∂p ∂s ∂s U∞

142 Structural Dynamics and Aeroelasticity

 ∂2H ∂ 2 H b2
=
∂s2 ∂s2 U∞ 2

Substituting in the equation we get

b2
   
b
M − q 2 Ma q̈ + C − q Ca q̇ + (K − qKa )q = 0
U∞ U∞

15.4.1 Computation of the QS approximation from the frequency response

Since the function is analytic the derivative can be taken along any direction.
Restricting the complex variable on the null real part
∂H ∂H ∂Im(H) ∂Re(H)
= = −j +
∂p ∂jω ∂ω ∂ω
∂2H ∂2H ∂ 2 Im(H) ∂ 2 Re(H)
= − = −j −
∂p2 ∂ω 2 ∂ω 2 ∂ω 2
But, since the real part of the frequency response is even at zero the derivative has to
be null.
∂Re(H)
=0
∂ω ω=0
And the imaginary part is odd so the second derivative has to have a flex in zero.
∂ 2 Im(H)
=0
∂ω 2 ω=0

We get that
∂H ∂Im(H) ∂2H ∂ 2 Re(H)
= = −
∂p ∂ω ∂p2 ∂ω 2
We can now plug these results in the equation above and find the flutter equation.

15.4.2 Meaning of QS approximation in time domain

Suppose we are able to write the system as a state space approach:
(
ẋ = Ax + Bu
y = Cx + Du

Now lets differentiate the function x and impose ẍ = 0:

ẍ = 0 = Aẋ + B u̇ = A(Ax + Bu) + B u̇ =⇒ x = −A−1 Bu − A−2 B u̇

Therefore:
y = (CA−1 B + D)u − CA−2 B u̇
The QS approximation of order n is equivalent to saying that the derivative (n+1) in
time is negligible.

143 Structural Dynamics and Aeroelasticity

15.4.3 Problems with this method
The McLaurin approach around s = 0 is ok for small structural frequency. For higher
frequency, anyway, this kind of approximation leads to a big error.
Theoretically we could increase the order of approximation and not stop at the second
term, but this is generally not done since we do not have the analytic expression of H
and we usually are not interested in measuring the III or IV order derivatives (that are
needed in structural analysis).

In general the McLaurin approach is not used for s >> 1.

What can we do?

15.5 Direct solution for flutter

We know that the flutter phenomenon occurs when the real part of the eigenvalue goes
to zero. We can impose it.
s = 0 + jωF
p = 0 + jkF
In the equation

det(−ωF2 M + jωF C + K − qF Ham (kF , MF )) = 0

This is a non-linear imaginary equation that can be solved by imposing MF and qF .

In principle this is always solvable, in practice, it is not for more than two dofs.

In practice, it is more convenient to compute all the (desired) eigenvalues for a range of
EAS velocities (or dynamic pressures) starting from zero and evaluate their real part.
Flutter is detected as the lowest airstream speed at which the real part of at least
one eigenvalue crosses the zero axis from negative to positive. The knowledge of all
eigenvalues from zero to the desired airstream velocity not only allows the analyst to
check for flutter clearance but also to determine how the eigenvalues evolve with the
EAS velocity allowing an experimental verification of the aircraft aeroelastic stability.
See Figure 68 below.

Another problem is that in this way we are not investigating any other condition than
the flutter one. This leads to a validation problem. Normative imposes to asses the
model by using q < qF (that are measurable too). The certification process consists
of comparing the damping obtained by the model (at q << qf ) and the one measured.
Using the direct approach we don’t have access to these results. Hence it cannot be used
to validate the mode.

15.6 Flutter: P-K method

The P-K method is the method that is generally used in every flutter calculator software.
The idea behind it is pretty simple.

144 Structural Dynamics and Aeroelasticity

 Figure 68: Flutter V-f V-g diagrams

Since we are not interested in modes with high damping, associated with big Real part
Re(λ) ̸= 0 (they don’t have an effect on the flutter speed) we can think about linearizing
with Taylor’s series around s = jk in this way we are only restricting the real part around
zero. But this is exactly what we are looking for: low damping modes are associated
with Re(λ) ≈ 0.
p = jk
Expanding through a Taylor series

∂Ham 1 ∂ 2 Ham
Ham (p) ≈ Ham (k) + (p − k) + (p − k)2 + . . .
∂p 2 ∂p2

So at first order (we can also use higher orders)

Im(Ham )(k)
Ham (p) ≈ Ham (k) = Re(Ham )(k) + jk
k

Im(Ham )(k)
Ham (p) ≈ Re(Ham )(k) + p
k
On this formula, we can build an algorithm to calculate the flutter speed

15.6.1 P-K method algorithm

For an initial given set of parameters: U∞ , ρ, h, M, ...

1. Compute a preliminary guess for one eigenvalue λ0i . i.e. we can use a structural
eigenvalue for small speeds

2. Compute ωi = Im(λ0i ) and ki = ωi b/U∞

145 Structural Dynamics and Aeroelasticity

 3. solve the equation
 
2 λb Im(Ham (k)
det λ M + λC − q + K − qRe(Ham )(k) = 0
U∞ k

4. Identify the closest eigenvalue λi to λ0i

5. if |λi − λ0i | < ϵ stop, else set λ0i = λi and iterate

The algorithm is repeated for every eigenvalue of interest.

Note 1: There is no mathematical proof that the algorithm converges.

Note 2: The smaller the damping, the better the approximation. When we are at flutter
speed then the approximation is exact. This is the reason why a zero-order is sufficient
to have a good approximation.

Note 3: This method is robust as the error becomes smaller and smaller the closer we
get to the flutter condition.

15.7 Continuation approach

The eigenvalue problem could be seen as a nonlinear algebraic problem

f (s)q = 0 =⇒ F(s, q) = 0

f (s, q) = (s2 M + xC + K − qHam (p, M ))q = 0

s∈C q ∈ Cn
We have n+1 complex unknowns and 2(n+1) real unknowns.
However, the eigenvector is defined but for a constant, so the unknowns are 2n + 1 and
there are 2n equations.
q∗ q = 1
We have a system
(
f (s, q) = (s2 M + xC + K − qHam (p, M ))q = 0
q∗ q = 1

This is a non-linear problem. We can solve it numerically through Newton’s method.

Consider q = qi + ∆q and s = si + ∆s

 ∂f ∆s + ∂f ∆q = −f (si , qi )

∂s ∂q
 ∗
2qi ∆q = 1 − q∗ q

146 Structural Dynamics and Aeroelasticity

The linear system we have
 ∂f ∂f     
∂s ∂q ∆s −f (si , qi )
=
0 2q∗i ∆q 1 − q∗ q

Iterate until ||∆q|| < ε and |∆| < ε to find the solution.
We start from a very low speed (close to zero) and as an estimate of the eigenvalues
we use the structural dynamics ones. To obtain an initial guess at a new speed it is
possible to use the value at the previous speed or try to estimate the eigenvalues and
eigenvectors at the new speed.

147 Structural Dynamics and Aeroelasticity

16 Response problems
• The analysis of the response is meaningful only for flight conditions where the
aircraft is STABLE. The types of input:

• Controls (i.e.movable surfaces)

• Gusts

• Exciters (vibrating masses, bonkers, etc. . . )

Figure 69: Typical aeronautic bonker

16.1 Equilibrium manouvres

The aircraft is in a steady maneuver, so it is balanced, or trimmed, and so no changes
in times must be considered. Examples: steady level flight, pull-up push-down, banked
turn, ...

EASA CS-25 says “If deflection under load would significantly change the distribution
of internal loads, this re-distribution must be kept into account”.

Consequently, the problem could be approached using static aeroelasticity methods.

The problem could be approached:

a) Retaining a rigid aircraft model and modifying the aerodynamics to keep flexibility
into account.

b) Using a fully flexible model.

Generally, a rotation of a control surface generates a change in forces. If the reaction

is slow enough, we can consider the aircraft as if it were going from an equilibrium
condition to another in a quasi-steady way. We can use static dynamic methods.

148 Structural Dynamics and Aeroelasticity

For aerodynamics, it is possible to use the quasi-steady behavior since the frequency of
the input forces is basically 0, which means that the structure responds instantaneously
to the forcing(if this is not the case then a dynamic maneuver approach should be
followed for the evaluations)
Z t
Mq̈ + Cq̇ + Kq = q ham (t − τ )q(τ )dτ = Fe
0
Z t
q∞
ham (t − τ )q(τ )dτ = Ca q̇ + q∞ Ka q
0 U∞
q∞
Mq̈ + Cq̇ + Kq − Ca q̇ − q∞ Ka q = Fe
U∞
We can divide our dofs id rigid and elastic.
       
MRR MRE 0 0 0 0 q
M= C= K= q= R
MER MEE 0 CEE 0 KEE qE
Note that the elastic dof is faster than the rigid. We can safely assume that in these
conditions

q̈E ≈ 0, q̇E ≈ 0
So from the first equation
q∞
MRR q̈R + MRE q̈E − (CaRR q̇R + CaRE q̇E ) − q∞ (KaRR qR + KaRE qE ) = FeR
U∞
We get
q∞
MRR q̈R + − CaRR q̇R − q∞ (KaRR qR + KaRE qE ) = FeR
U∞
From the second equation
q∞
MER q̈R +MEE q̈E +CEE q̇E +KEE qE − (CaER q̇R +CaEE q̇E )−q∞ (KaER qR +KaEE qE ) = FeE
U∞
As before
q∞
MER q̈R + KEE qE − CaER q̇R − q∞ (KaER qR + KaEE qE ) = FeE
U∞
Define as usual KAE = q∞ KaEE .
KAE is the aeroelastic stiffness (that is singular at the divergence dynamic pressure qD)

Note that for

q∞ ≈ qD =⇒ KAE ≈ KAA
It is possible to see the deformation of the structure as the combined effect of external
loads, inertia forces due to rigid movements and aerodynamic loads due to rigid move-
ments.

149 Structural Dynamics and Aeroelasticity

For low q∞ the effect is negligible.
For high q∞ the aeroelastic deformation is no longer negligible.

We can consider this effect by adding a new term in the rigid equation as below
 
−1

q∞ −1


MRR + q∞ KaRE KAE MER q̈R + − CaRR − q∞ KaRE KAE CaER q̇R +
U∞

+ −q∞ KaRR − q∞ KaRE K−1 qR = FeR + q∞ KaRE K−1

AE KaER AE FeE

The equation has the same format as the rigid ones, but the matrices contain new
aerodynamic derivative terms (green).

If MER = 0 then there are no aerodynamic stability derivatives that are proportional to
rigid accelerations (as usually in Flight Mechanics).

Note: if the proper orthogonal modes of a free-free structure are used, MER = 0 by
default.

16.2 Dynamic manouvers

When we are interested in dynamic maneuvers, we will consider the application of a
time history of a control surface movement to see what happens to loads.

It is essential to model the behavior of a control surface and its actuation behavior.
Consider the case where a servo-system is applied. We will consider the approach
based on Laplace transform only for easiness of the exposition, but the same steps could
be followed in any approach.

16.2.1 Modeling of a control surface

Figure 70: Control chain (rigid)

Specific shape functions used to represent the rigid rotation of the movable surface will be
added to the usual shape functions. In this way, a specific degree of freedom associated
with the rotation of the movable surface will be defined

150 Structural Dynamics and Aeroelasticity

The set of shape functions will be the union of blocked-surface shape functions + movable
surface shape functions
ηs = Nq + Nδ δ
δ angle of deflection of the control surface

16.2.2 Ideal servo actuator

Considering an ideal servo actuator (that can keep exactly any required position, inde-
pendently from the load applied), it is possible to consider the control rotation δ as an
input.

The second equation could be used to identify the load Fδ that must be applied by the
actuator/pilot to keep the position δ. This load is in general, a function of time (so a
control system is required to adapt it to keep the delta constant.
   
2 q Feq
(s M + xC + K − qHam (p, M )) =
δ Feδ

Write     
Zqq Zqδ q Feq
=
Zδq Zδδ δ Feδ

16.2.3 Modeling of a servo-controlled surface

Figure 71: Servo-actuator scheme

To improve the approximation, it is possible to include the dynamics of the actuator

through a servo-system transfer function.

Keep into account that this model is still ideal, because also, in this case, the force Fδ
must change instantaneously to adapt to changes of v.

δ = Hservo (s)δc

151 Structural Dynamics and Aeroelasticity

 Zqq q = Feq − Zqδ Hservo (s)δc
Feδ = Zδq + Zδδ Hservo (s)δc
Define: v = valve opening, Q = flux, P and G = gains

v = G(δc − δ)

δ̇ = P v
In order to be more accurate we should include the deformability of the control chain.
So:
PG
δ̇ = P Gδc − P Gδ −→ δ = deltac
s + PG
More general format
PG
δ= δC
(s + P G)(s + 2ξω0 s + ω02
2

We have to consider that the pilot has a transfer function too

 
q
δc = Hp (s)δpilot + HF CS (s)Hsensors (s) q̇
q̈
The transfer function of the pilot may be not so easy to identify. HSN is the transfer
function of sensors.

The dynamics of the control system may become coupled with the aeroelastic dynamics
and we need to assess the stability of the overall systems. Indeed an elastic deformation
of the wing may be seen as a rolling action by the sensor therefore the control input may
worsen the situation. The sensor must filter out the aeroelastic dynamics. The analysis
of the modal shapes may help the decision on where to locate the sensors.

16.2.4 Modeling of dynamic compliance surface

Figure 72: Elastic control chain model

In the case where there is no servo-actuator too (Hservo = 1), it is possible to have δ
and δc different because the control chain is, in reality, an elastic system.

152 Structural Dynamics and Aeroelasticity

It is possible to see that δ = δc only if k → ∞ orFδ = 0

Both conditions are ideal, so it is necessary to consider dynamic admittance in practice.

Feδ Feδ
= (δ − δc ) =⇒ δ = δc +
k k
So
Feδ
δ = Hs (s)δc + Hc (s)
k
Where Hc is the admittance or Dynamic compliance.

16.2.5 Complete model

    
Zqq Zqδ 0 q Feq
 Zδq Zδδ −1   δ  =  Feδ 
0 1 −HC Feδ Hs δc

16.3 Gust response

Figure 73: Aircraft encountering turbulence

For design purposes, atmospheric change of wind speed is divided into two ideal cate-
gories
1. discrete gusts, where the gust velocity varies in a deterministic manner, usually
in the form of a ‘1-cosine’ shape
2. continuous turbulence, where the gust velocity is assumed to vary in a random
manner

16.4 Discrete gust

The length and the intensity of the Gusts are assigned by the certification standards.
Constant gust (
0 xg < 0
wg (xg ) =
wg0 xg ≥ 0

153 Structural Dynamics and Aeroelasticity

1-cosine gust
Where

Figure 74: 1-cosine type of gust

 
wg0 2πxg
wg (xg ) = 1 − cos 0 ≤ xg ≤ Lg
2 Lg

16.4.1 Stochastic Gust

At this point the official model to be considered by certification organizations is the
Von-Karman PSD for turbulence:

Ω is the scaled frequency

σg gust intensity
L characteristic scale
L 1 + (8/3)(1.339ΩL)2
Φgg (Ω) = σg2
π (1 + 1.339ΩL)2 )11/6

154 Structural Dynamics and Aeroelasticity

17 Useful formulas
17.1 Formulas
17.1.1 Computation of variance
1 ∞ ∗
Z
2
σqq = H (ω)ΦFF (ω)HT (ω) dω
π 0

17.1.2 Lyapunov formula

2 2
Aσxx + σxx AT = BWBT

17.1.3 Rice formula

The probability of crossing the level is going to be:

1 − e−N̄+ (b)T

Where 2
1 σẋẋ − 21 σb2
N̄+ (b) = e xx
2π σxx

17.1.4 Compressible unsteady Bernoulli theorem

 
∂Φ 1 2
∇ + u +h =0
∂t 2

17.2 Cp through Bernoulli

2 ∂Φ |∇Φ|2
 
P − P∞
Cp = 1 2
=− 2 ∂t
+ 2
−1
2 ρ∞ U∞
U∞ U∞

17.2.1 Cp in Morino
 
2 ∂∆φ ∂∆φ
∆Cp |W =− 2 U∞ + =0
U∞ ∂x ∂t W

17.2.2 Linearized BCs (Theodorsen and Morino case)

∂η ∂η ∂φ
+ U∞ =
∂t ∂x ∂z

17.2.3 Theodorsen model

   
2
  1
L(k) = πρb ḧ + U∞ α̇ − baα̈ + 2πρU∞ bC(k) ḣ + U∞ α + b − a α̇
2
   
3 1 1 a
Mc/4 (k) = −πρb ḧ + U∞ α̇ + b − α̈
2 8 2

155 Structural Dynamics and Aeroelasticity

17.3 Theodorsen and HAM Matrix
Z b 
T Ham11
T Ham12
δWa = δq q Q Q dy q = δqT HAM (k)q
−b Ham21 Ham22
Where Q is the rotation matrix from structural coordinates and aerodynamic dofs
 
h
= Qq
α

156 Structural Dynamics and Aeroelasticity