22 views

Uploaded by Alexander Ramos

aeroelasticity, analysis, wings, aircrafts

- EUROPLEXU
- BX-Flow.pdf
- Aerodynamics - Wikipedia, The Free Encyclopedia
- MP Filtri ฟิลเตอร์ - FilterTH.com
- Aeroelasticity of Wind Turbines Blades Using Numerical Simulation
- Fluids
- CFX12_03_Physics1
- Control of Nonlinear Dynamic Systems
- Ias
- 0c960517961c8ef915000000.pdf
- COMPUTATIONAL ANALYSIS OF A SEDAN USING COMPUTATIONAL FLUID DYNAMICS SOFTWARE
- BABCOCK_J
- SINGLE TRAIN PASSING THROUGH A TUNNEL (Jakub Novak)
- 16667
- Controlling loads in a wind turbine drivetrain
- 11837_002912154103.pdf
- FUNDAMENTALS OF FLUID MECHANICS, Cengel Cimbala Solutions Chap01.pdf
- 80802
- SYLLABUS-2183-222 - 2558 (Second Semester - AERO) - Course Syllabus
- Tian 1994

You are on page 1of 408

A V Balakrishnan

Aeroelasticity

The Continuum Theory

123

A V Balakrishnan

Department of Electrical Engineering

Department of Mathematics

University of California

Westwood Blvd.

Los Angeles, California

USA

DOI 10.1007/978-1-4614-3609-6

Springer New York Heidelberg Dordrecht London

Library of Congress Control Number: 2012938709

This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of

the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation,

broadcasting, reproduction on microfilms or in any other physical way, and transmission or information

storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology

now known or hereafter developed. Exempted from this legal reservation are brief excerpts in connection

with reviews or scholarly analysis or material supplied specifically for the purpose of being entered

and executed on a computer system, for exclusive use by the purchaser of the work. Duplication of

this publication or parts thereof is permitted only under the provisions of the Copyright Law of the

Publishers location, in its current version, and permission for use must always be obtained from Springer.

Permissions for use may be obtained through RightsLink at the Copyright Clearance Center. Violations

are liable to prosecution under the respective Copyright Law.

The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication

does not imply, even in the absence of a specific statement, that such names are exempt from the relevant

protective laws and regulations and therefore free for general use.

While the advice and information in this book are believed to be true and accurate at the date of

publication, neither the authors nor the editors nor the publisher can accept any legal responsibility for

any errors or omissions that may be made. The publisher makes no warranty, express or implied, with

respect to the material contained herein.

Foreword

The author is a noted applied mathematician and control theorist who has in recent

years turned his attention to the subject of aeroelasticity. The present volume is

the result of his research and provides an illuminating new view of the field.

As he emphasizes in his introduction, his approach is one of continuum models

of the aerodynamic flow interacting with a flexible structure whose behavior

is governed by partial differential equations. Both linear and nonlinear models

are considered although much of the book is concerned with the former while

keeping the latter clearly in view and indeed a complete chapter is devoted to

nonlinear theory. The author has provided new insights into the classical inviscid

aerodynamics described by the celebrated Possios equation and raises novel and

interesting questions on fundamental issues that have too often been neglected or

forgotten in the development of the early history of the subject. The author contrasts

his approach with discrete models that have gained enormous popularity in the

practice of aeroelasticity such as the doublet lattice model of Rodden et al. for

unsteady aerodynamic flow and the finite element model for the structure. Much

of aeroelasticity has been developed with applications firmly in mind because of

its enormous consequences for the safety of aircraft. Aeroelastic instabilities such

as divergence and flutter and aeroelastic responses to gusts can pose a significant

hazard to the aircraft and affect its performance. Yet it is now recognized that

there are many other physical phenomena that have similar characteristics ranging

from flows around flexible tall buildings and long span bridges to flows internal

to the human body such as blood flows through arteries and air flow over the

tongue, and the author touches on these topics as well. For the theorist and applied

mathematician who wishes an introduction to this fascinating subject as well as for

the experienced aeroelastician who is open to new challenges and a fresh viewpoint,

this book and its author have much to offer the reader.

v

Preface

Aeroelasticity deals with the dynamics of an elastic structure in airflow with primary

focus on the endemic instability of the structure called flutter that occurs at

high enough speed. This book presents the continuum theory in contrast to

extant literature that is largely computational; where typically one starts with the

basic continuum model, a partial differential equation usually highly nonlinear

but omitting the all important boundary conditions and disregarding the question

of existence of solution; going immediately to the discretized approximation;

presenting charts and figures for a confluence of numerical values for the parameters

and conclusions drawn from them.

Here we stay with the basic continuum model theory until the very end, where

constructive methods are developed for calculating physical quantities of interest,

such as the flutter speed. Indeed this is considered mission impossible because it

is nonlinear and complex.

As in any scientific discipline, continuum theory provides answers to what if

questions which numerical codes cannot. It makes possible precise definitions

such as what is flutter speed. Physical phenomenasuch as transonic dip, for

examplecan be captured by simple closed-form formulae. And above all it can

help develop intuition based on a better understanding of the phenomena of interest.

As with any mathematical theory it enables a degree of generality and qualitative

conclusions, increasing insight.

But the use of continuum models comes with a price: it requires a high

level of abstract mathematics. For a precise statement of the problem, however,

the language of modern analysisdeveloped in the latter half of the twentieth

centuryabstract functional analysis, in particular, the theory of boundary value

problems of partial differential equations, is unavoidable. Indeed the aeroelastic

problem, the structure dynamics in normal air flow-formulates as a nonlinear

convolution/evolution equation in a Hilbert space.

On the other hand the numerical range of the physical parameters plays an

important role in being able to generate constructive solutions otherwise impossible

vii

viii Preface

sense that we use mathematics to solve todays engineering problems addressed

to engineers as well as mathematicians.

And now for some points of view, points of departure, of this book closer to the

subject matter. Aeroelasticity is concerned with the stability of the structure in air

flow. The air flow per se is of less interest. Thus we are not concerned, for example,

whether there are shocks in the flow or not, in itself a controversial matter. The faith

of the aeroelasticians in shocks, it turns out, is not substantiated by the mathematical

theory (2D or 3D flow). It may be heresy to the clan but shocks may exist that do

not affect the stability (or rather the instability) of the structure. Another and more

significant view concerns the interaction between Lagrangian structure dynamics

and Eulerian fluid dynamics, often the most mysterious part of computational work.

Here we take the simple engineering inputoutput point of view where the

velocity of the structure is the input and the pressure jump across the structure is the

output. The inputoutput relation is the integral equation of Possio that does not get

any mention in as recent a work as [17] which features partial differential equations.

The Possio Integral Equation can be looked as an illustration of the Duhamel

principle and we make systematic use of itlinear and nonlinearthroughout the

book. We show that flutter speed is simply the smallest speed at which the structure

becomes unstable; it is a Hopf bifurcation point determined completely by the

linearized model about the steady state. In turn this means incidentally that the

control for extending the flutter speed need not be nonlinear, contrary to current

wisdom.

The mathematical style of the book is largely imitated/borrowed from that of

R.E. Mayer [14], and ChorinMarsden [4] where they claim to Present basic ideas

in a mathematically attractive manner (which does not mean fully rigorous).

In this sense although we use abstract functional analysis, we try to reduce the

abstraction and sacrifice mathematical generality, preferring to emphasize construc-

tive solutions and basic ideas rather than get lost in Sobolev spaces and weak

solutions. Quoting another pioneer in this style: I shall not be guilty of artificially

complicating simple matters. A phenomenon that sometimes occurs in mathematical

writing. Tricomi in his book Integral Equation, 1957 [11].

We should caution that there are many problems that mathematical theory cannot

currently answer especially in viscous flow and as a result also in aeroelasticity. We

invoke the Prandtl boundary layer theory, for example, with this caveat.

We should also note a price to be paid for mixing the abstract with the concrete,

saying too much or too little at either end.

Acknowledgments

Earl Dowell (who was my chief mentor and also opened my eyes to the nonaircraft

applications)

John Edwards

Oddvar Bendiksen

Dewey Hodges

Ciprian Preda

Marianna Shubov

Roberto Triggiani

Amjad Tuffaha

My former students:

Jason Lin

Oscar Alvarez Salazar

Irena Lasiecka (Post Doc)

And finally the NSF monitor Dr. Kishan Baheti for his unfailing belief in the value

of the work and his help and encouragement throughout.

ix

Contents

1 Introduction .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 1

2 Dynamics of Wing Structures . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 9

2.1 Introduction .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 9

2.2 The Goland Beam Model . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 9

2.3 Time Domain Analysis.. . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 11

2.4 Structure Modes and Mode Shapes. . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 18

2.5 Robust Feedback Control Theory: Stability Enhancement .. . . . . . . . 26

2.6 Nonfixed Wing Models: Flying Wings . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 35

2.7 Nonlinear Structure Models . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 40

2.8 Beams of Infinite Length.. . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 43

3 The Air Flow Model/Boundary Fluid Structure

Interaction/The Aeroelastic Problem .. . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 47

3.1 Introduction .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 47

3.2 Notation/Physical Constants .. . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 47

3.3 Nonviscous Flow: The Euler Full Potential Equation . . . . . . . . . . . . . . 49

3.4 Problem Statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 58

4 The Steady-State (Static) Solution of the Aeroelastic Equation . . . . . . . 65

4.1 Introduction .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 65

4.2 Goland Structure Model . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 66

4.3 Linear Aeroelasticity Theory: The Finite Plane Case . . . . . . . . . . . . . . 72

4.4 The General Nonlinear Case . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 80

4.5 Nonlinear Structure Models . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 96

5 Linear Aeroelasticity Theory/ The Possio Integral Equation .. . . . . . . . . 103

5.1 Introduction .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 103

5.2 Power Series Expansion of the 3D Potential . . . .. . . . . . . . . . . . . . . . . . . . 103

5.3 The Linear Possio Integral Equation . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 106

5.4 Linear Possio Equation: Compressible Flow: M > 0 . . . . . . . . . . . . . . 122

5.5 Linear Time Domain Airfoil Dynamics .. . . . . . . .. . . . . . . . . . . . . . . . . . . . 155

5.6 State Space Theory .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 184

xi

xii Contents

5.8 Nonlinear Structure Models . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 249

5.9 Appendix: Computer Program for Flutter Speed

in Incompressible Flow: Goland Model .. . . . . . . .. . . . . . . . . . . . . . . . . . . . 260

6 Nonlinear Aeroelasticity Theory in 2D Aerodynamics:

Flutter Instability as an LCO .. . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . 269

6.1 Introduction .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 269

6.2 The Aeroelastic Equations: Linear Structure Model.. . . . . . . . . . . . . . . 270

6.3 The Nonlinear Possio Integral Equation . . . . . . . .. . . . . . . . . . . . . . . . . . . . 296

6.4 Nonlinear Aeroelastic Dynamics . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 298

6.5 Stability .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 315

6.6 Limit Cycle Oscillation . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 318

6.7 The Air Flow Decomposition Theory . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 323

7 Viscous Air flow Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 327

7.1 Introduction .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 327

7.2 The Field Equation/Conservation Laws . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 327

8 Optimal Control Theory: Flutter Suppression . . . . . .. . . . . . . . . . . . . . . . . . . . 337

8.1 Introduction .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 337

9 Aeroelastic Gust Response . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 349

9.1 Introduction .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 349

9.2 The Turbulence Field. . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 349

9.3 The Gust Forced Aeroelastic Equations .. . . . . . . .. . . . . . . . . . . . . . . . . . . . 351

9.4 Structure Response to Turbulence .. . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 362

9.5 Illustrative Example .. . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 364

10 Addendum: Axial Air flow TheoryContinuum Models. . . . . . . . . . . . . . 367

10.1 Introduction .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 367

10.2 The Aeroelastic Equations . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 367

10.3 Steady-State Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 370

10.4 Power Series Expansion . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 370

10.5 Linear Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 376

10.6 Stability: Aeroelastic Modes .. . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 379

10.7 Linear Time Domain Theory: The Convolution/Evolution

Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 384

10.8 Nonlinear Stability Theory: Hopf Bifurcation/Flutter LCO .. . . . . . . 386

References .. .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 387

Index . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 393

Chapter 1

Introduction

traditionally the wing of an aircraft in flight. There are important nonaircraft

applications of more recent interest as well which we treat in Chap. 10. Thus it

involves both structural dynamics and fluid dynamics without being either. In this

sense it has an inherent identity complex but still has developed as a separate

discipline over the many years since flight began. The central problem is flutter,

a potentially destructive instability that occurs at any altitude as the speed is

increased to the flutter speed, the flutter boundary. Indeed the Federal Aviation

Administration mandates that a margin of 15% be maintained in all flight. And thus

flutter analysis is an ongoing activity in flight centers as well as in aircraft design.

Although in the aircraft context ultimately a flight test would be needed to verify

the flutter phenomena, flight tests are costly and need to be kept at a minimum. Wind

tunnel tests are much less expensive but paperwork is the least expensive. Hence the

importance of analysis! The ultimate objective is of course to minimize the number

of flight tests needed.

Here one has to distinguish between analytical (continuum) theory and (digital)

computation (FEM/CFD) which is predominant in all the current work. Both

structure and aerodynamic models originate as continuum models and digital

computation needs perforce to approximate the continuum: convert the distributed

parameter system to a lumped parameter system and replace the continuum

partial differential equations by ordinary differential equations, however high the

order. And of course digitization discretizes time and amplitude. Furthermore the

system parameters must be specified numerically, numeric computation as opposed

to symbolic computation. This restricts generality and can inhibit qualitative

understanding of phenomena. Perhaps another way of putting it is to say that

in this work we do not truncate the model at the beginning; we work with

the continuum models or symbolic software and approximation as needed only

eventually at the end-level calculation of specific functionals (such as flutter speed

as a function of system parameters). We approximate the solution, rather than

approximate the equations. Phenomena occur at the continuum level that do not

occur in any solution of the discretized equations, however fine the discretization.

DOI 10.1007/978-1-4614-3609-6 1, Springer Science+Business Media, LLC 2012

2 1 Introduction

In the computational approach the discretization is made without any concern about

whether the continuum equations have a solution and whether are unique. Moreover

there is no attempt to prove that there is a solution to the problem in the language

of the starting continuum formulation. The digital computer solution is accepted as

a good enough approximation to the solution. This can be particularly misleading

when the continuum equations are shown to have more than one solution or no

solution at all. Indeed it is typical in extant literature (see the List of Papers) in

aeroelasticity where complex continuum structure models are introduced without

any consideration of whether they have any solution and are immediately converted

to discretized computational models (FEM). This is also true of the fluid dynamic

models where the partial differential equations are introduced without any precise

statement of the (fluid/structure) boundary conditions.

Thus we emphasize that a distinguishing feature of this treatment is that a precise

statement of the aeroelastic problem is made especially at the LagrangianEuler

fluid structure boundary; the dynamic equations are shown to have a unique solution

and qualitative results are deduced and then only numerical computation is made for

the functionals of interest, without discretizing the equations at any level.

The material has been distributed into 10 chapters in reasonable logical

progression and are listed in the Table of Contents. Here we provide an overall

preview of the various chapters and how they interlink.

We begin in Chap. 2 with the dynamics of structures, where the important

notion of modes and mode shapes is described in the language of function spaces,

specifically L2 -spaces. (It is interesting to note that the need for infinite-dimensional

spaces is underscored by a recent publication Dynamics of Very High Dimensional

Systems [24].) The primary reason for this level of abstraction is that a precise

mathematical statement of the structure dynamics in air flow is not possible

without it.

The first flexible wing modelthe Goland modelwas proposed by Goland

[78] in 1945, a uniform rectangular beam of zero thickness with two degrees of

freedom, plunge and pitch. Similar equations with higher degrees of freedom are

employed in applications to spacecraft structures [54, 61]. It is a fixed wing

model cantilevered to the fuselage with the wing tip (clamped/free end conditions).

This is our canonical model; (all later versions still anchored essentially on this)

including nonlinear models. A drawback is of course that the thickness or the wing

camber is neglected. However, extending it to a plate model turns out to be too

complex to handle and it is questionable whether the results justify the analytical

complexity. There are no existence theorems, neither where the models include

wing camber nor the wing section as a tear drop [18]. Another peculiarity is that

the cantilever beam model is used only for the structure dynamics corresponding

to the aerodynamic loads, and to calculate the latter one takes advantage of the

high aspect ratio of the wing which must needs take into account the flexibility

of the structure, long slender wing. Some authors even further qualify it by very

flexible wing [82]. The flexibility is only along the span; we have thus a beam

instead of a thin plate. Furthermore, this allows the enormous simplification that

we may neglect the dependence on the span axis in describing the aerodynamic field

1 Introduction 3

equation. This is referred to as the strip theory or the typical section theory. We

can also see this as a first approximation of the finite plane theory (cf. Chap. 3).

In other words the structure beam model is consistent with the aerodynamic

strip [6] theory.

The Hilbert space for structure state variables presented is standard and essential

use is made of the notion of elastic energy, including a characterization of the

structure deflections for which the energy can be defined. This defines the modes

and mode shapes as eigenvalues and eigenvectors respectively. In continuum theory

the number of modes is infinite. In practice we know we will only deal with the

first few modes, so that much of the mathematical theory of asymptotic modes

is irrelevant. In the approximated computer model, the number of modes is indeed

finite, but then we can always add one more! Indeed we construct a continuum

model in which we match the first few modes of interest. This is because of the

inherent damping of the structure that eventually increases as the mode number

increases. But truncating the model at the beginning one loses physical phenomena

that only the continuum model can capture.

Control theory is needed for our objective of stabilizing or enhancing the stability

of the structure and here the truncated numerical model is useless. We need the

continuum model to generate feedback control laws. One difficulty is in not knowing

the inherent damping. The few models we have are not faithful enough. Hence in

robust control we assume a model in which there is no damping (as in our models)

but devise a feedback control which is guaranteed not to decrease the inherent

damping.

Current control design for distributed structures uses colocated point controllers

where the sensor and actuator are colocated. An inherent limitation here is that the

damping will decrease (not necessarily monotonically) to zero as the mode number

increases. An alternative is to use the self-straining actuators [70] which can

ideally achieve super stability. So they are attractive to try in the aeroelastic case.

Finally we consider nonlinear structure models (they may be considered exten-

sions of the Goland model) still beam models with zero thickness. The models

suggested [75, 76] unfortunately have not been analyzed in the continuum version

unlike the linear case. There is no longer the notion of modes (eigenvalues) which

makes the determination of stability difficult.

To make a mathematical statement of the structure dynamics we first need to

invoke aerodynamics to calculate the aerodynamic loads. Chapter 3 thus begins

with the aerodynamics part of aeroelasticity, the theory leading to the calculation

of the aerodynamic lift and moment. We limit consideration to nonviscous flow

where the viscosity of the air is neglected. Thermodynamics is involved and the

assumption thruout is that the perfect gas law holds. In the absence of viscosity we

need to assume that the entropy is a constant so that the heat exchange processes

are adiabatic. The assumption of constant entropy enables us to relate the pressure

to density:

p D A

4 1 Introduction

and show that the flow is vortex-free and we have potential flow which can be

then characterized by the Euler full potential equation. However, what distinguishes

the aeroelastic problem from being simply fluid dynamics is the fluidstructure

interaction, the boundary conditions.

The assumption here is flow tangency, the wing velocity normal to the plane is

equal to the fluid velocity normal to the wing plane. In addition we have to impose

the conditions discovered by Joukowsky and the condition named after Kutta which

is necessary for uniqueness of solution.

We are then able to come up with a complete statement of the aeroelastic

equations. Here we also specialize to 2D flow, and zero angle of attack in which

we ignore the dependence on the wing span (abandon the Finite Plane model) and

consider the typical section approximation justified by the highaspect ratio of

the winga raison detrealso for the flexible PDE model. Finally here we also

extend the formulation to include nonlinear structure models.

The study of stability of the wing structure (which for us now depends on the

assumed far-field speed that we consider a parameter) requires that we specify

stability about the state of the system. One usually thinks of the equilibrium state

to which the system is attracted or returns in the absence of any disturbance.

We need to determine the solution to the aeroelastic equation where we set all time

derivatives to zero. We call this the static solution. This is the objective in Chap. 4.

The technique for solution is to exploit the analyticity in terms of the structure state

variables so that we can make a Taylor series expansion about the zero or rest

structure state. We show that the solution is nonunique for a sequence of values of

the far-field speed parameter for which we have two different solutions, even if we

may call one of them trivial.

The main tool is the static version of the Possio integral equation which turns

out to be the finite Hilbert transform solved by Tricomi, who showed the need for

the Kutta condition for uniqueness of solution; see [11, and therein to the work

of Sohngen, a German pioneer in the theory]. See also [31]. Here we are able to

show the convergence of the power series pointwise. There are no discontinuities

except on the boundary, for z D 0. Here we also derive a simple formula for the

Transonic Dip for nonzero angle of attack which in numerical computation has to

be extrapolated from a point on the graph. See [65] as well as the references therein.

The nonzero static solution can be considered as an eigen value problemalbeit

nonlinear. The nonlinearity is primarily in the pitch or torsion variable.

The flow has no discontinuities in the field and the convergence is pointwisein

sharp contrast to the time varying case in Chap. 6.

We conclude the chapter with an example which turns out to be relevant to

viscous flow treated in Chap. 7.

One of the major results of the continuum theory is the characterization

of the smallest speed at which the wing is unstablethe Flutter speed

can be characterized as a Hopf bifurcation pointan illustration of the

Hopf bifurcation theory the successor to the classical stability theory of

Liapunov [42]. Here the stability is determined completely by the linear system

obtained by linearizing the nonlinear system about the steady state. Naturally the

1 Introduction 5

bulk of the aeroelastic theory deals with the Linear case, and there is no exception

here. By the linear case we mean the solution to the equations linearized about

a steady state. The term linear model is often used in the aeroelastic literature

without specifying the steady state about which it is linearized. Of course the

natural steady state would be that of constant flow and zero structure state. The

corresponding linear theory is treated in Chap. 5 the longest chapter in the book,

with eight sections. This provides the bread-and-butter of aeroelasticity and the

reason for its life in the sun. Here we study linear flutter analysis with continuum

models without approximation of any kind. (No Pade approximation or the Rodden

computer algorithm [17, 36]). The main concept is that of aeroelastic modes and

the linear structure dynamics is formulated as a convolution/evolution (semigroup)

equation in a Hilbert space. What is also new here is the development of a state space

theory which in particular shows that the aeroelastic modes are the eigenvalues of

the statespace stability operator. We also show that the case M D 0 is a special

one for which we have second degree noncirculatory terms not present for M > 0.

Also we are able to characterize the mode shapes explicitly and their properties

as eigenfunctions of the stability operator. The central theme is the linear Possio

integral equation [56, 89] which connects the Lagrangian dynamics to the Eulerian.

More specifically it provides the inputoutput relation between the structure velocity

as the input and the pressure jump across the wing as the output, so that we can

express the aerodynamic force and moment in terms of the structure state variables.

The key concept is the Mikhlin Multiplier and the Balakrishnan formula based

on it. The original version of the integral by Possio [89] was in the frequency

domain and is transported here to the Laplace domain. This suffices for calculating

the flutter mode and frequency. The aeroelastic modes are shown to be the zeros

of the determinant of a 3 by 3 matrix function of a complex variable which is

analytic except for the logarithmic singularity along the negative real axis for all

values of M . The aeroelastic modes are shown to be in a bounded vertical strip.

Zero is an aeroelastic mode for all M < 1. The associated flutter speed is known

as the divergence speed and is shown to be finite. The latter result is essential in

showing the existence of the flutter speed. It is discontinuous in M as a function

of the angle of attack and an explicit formula exhibits a transonic dip for a high

enough value of M , hitherto deduced from graphical computation. The modes are

characterized as the eigenvalues of the stability operator in the state space theory.

The uniqueness of solution of the Possio equation requires the Kutta condition

and in fact uniqueness implies existence. Various expressions are deduced for the

solution.

Key to the state space representation is the characterization of the aeroelastic

structure equation as a linear convolution evolution equation in a Hilbert space. The

convolution part plays the essential role in the flutter phenomena. We also derive an

explicit time domain solution for

M D0 and M D 1:

6 1 Introduction

We note that all the results obtained in the classical treatise [6] based on the

Theodorsen approximation can be deduced from the solution to the Possio equation

as in fact we do here. The sonic case .M D 1/ is interesting in that the role of the

Theodorsen function is taken over by the error function.

Essential for the stability theory is the fact that the slope of the stability curve is

negative for small speeds for all modes. The system becomes more stable initially

with airflow and then becomes unstable at some point as we increase speed, which is

the flutter speed. A relatively simple algorithm for it based on the continuum model

is given in the appendix to the chapter. One of the interesting by-products of the

continuum theory is the existence of evanescent states which decay too fast to be

captured in the CFD computer calculations.

All of this is only a prelude to the stability theory for the nonlinear system

developed in Chap. 6. The main result here is that the system is stable for all

values of speed less than the flutter speed and at the flutter speed the asymptotic

instability can be characterized as an LCO, limit cycle oscillation. It is consistent

with the Hopf bifurcation theory but the details of the proof have had to be developed

independently. The analysis turns out to be tedious and complicated involving many

asymptotic estimates. We show that the limiting response is periodic with the period

equal to the inverse of the linear flutter frequency with the preponderant harmonic

being the third harmonic.

We also derive an expression for the amplitude of the LCO rather involved

in terms of the various parameters. It is valid strictly speaking only for small

initial amplitude and is shown to be proportional to it. Although probably of less

importance in practice, the question of what happens for large initial amplitudes

is left open here as in computation. Heavy use is made throughout of Volterra

expansion. In fact in showing that the composite aeroelastic equation is a nonlinear

convolution/evolution equation, the nonlinearity is expressed in terms of a Volterra

series. And we note that in it the torsion variable is more dominant. We go with

the linear structure model, the extension to the nonlinear case being straightforward

based on the prequel. It is possible to examine the role of the structure in contrast to

the aerodynamics although we dont go into it.

As noted, our primary interest is in the structure but we also obtain en route a

revealing decomposition of the airflowthe velocity potentialinto two parts. One

part produces all the lift and can be linearized about the steady state whereas the

second part is continuous across the wing and cannot be so linearized. Interpreted

in the weak sense (which we do not attempt here), it may contain shocks, which is

controversial.

In Chap. 7 we venture even more into controversial territory where we allow the

flow to be viscous.

We assume small viscosity consistent with the fluid now being air. And we now

have to refresh the conservation laws. The basic field equation now becomes the

NavierStokes equation for incompressible flow. This is a much trodden area in fluid

dynamics and yet still with many unresolved questions, which are then reflected

in the structure dynamics as well. For example, there is no proof yet whether we

return to the Euler equation as the viscosity becomes vanishingly small. The big

1 Introduction 7

yet of shocks in 2D flow. The boundary conditions now require that the structure

velocity be completely equal to the fluid velocity. To make any progress at all we

have to invoke the still unproven Prandtl boundary layer hypothesis, and even then

our treatment is not as complete as we would like.

The question of whether we can altogether suppress flutter by means of control

actuators (flutter suppression as it is called) is the theme of the eighth chapter. We

do have a model, a state space model, that makes it possible for us to consider

control design. Stabilizability is a central question in optimal control theory but

the application to state spaces of nonfinite dimension essential to exhibit the flutter

phenomena as here poses greater complexity. The main result here is that no actuator

on the structure can stabilize the system; we simply cannot suppress flutter. It will

always occur at some speed. However, we can design controls that enhance the

stability in the sense that we can increase the speed at which flutter occurs. But this

is no longer a crisp optimization problem and comparison of control performance

is probably not possible. One fallout of our theory is that the control need not be

nonlinear. Indeed as we have shown, the flutter speed is completely determined by

the linearized system.

One of the important safety issues for aircraft in flight is the effect of wind gust

which still continues to be of interest and is a familiar topic in flight dynamics [9,33].

Here we consider the problem with the 3D random field Kolmogorov model of air

turbulence [35] as opposed to the deterministic gust models studied in [6, 84]. We

calculate the spectral density of the structure response showing the role of the flutter

speed. Illustrative examples are given for both bending flutter and torsion flutter.

Finally in Chap. 10 we provide an addendum on the latest area of research in

flutter systems, nonaircraft and nonflight applications, for example to Piezoelectric

power generation [106,109] and the biomedical problem of palatal flutter [102,105].

Here the main difference is in the nature of the air flow. It is no longer normal to the

structure but is axial, axial flow. Our treatment is again based on continuum models

as opposed to the regulation computational as in extant literature [50, 102]. This

brings profound differences; the theory is more complicated and so this chapter is in

the nature of an Addendum confined largely to problem formulation, emphasizing

the difference from normal flow. It turns out that there is one simplification. In the

application we need only consider incompressible flow which is then characterized

simply by the linear Laplace equation. The nonlinearity is thus only on the boundary

dynamics. We consider here again the Goland beam model, and in particular the

symmetric case where the cg is on the elastic axis, and as a result the plunge and

pitch dynamics decouple. However, the mathematics is more complex in that the

Hilbert space has to be generalized to a Banach space: from an L2 space to Lp ,

1 < p < 2. We are limited by space alone to problem formulation with details

deferred to a forthcoming second volume.

8 1 Introduction

continuum models were used by the German pioneers: Kushner, Schwarz,

Sohngen, and Wagner among others. The convolution aspect was already shown by

Wagner. But the Allies after WWII decreed a moratorium on any further research

by Germany from 1945 until 1954. This was a death blow to German work in

aeroelasticity from which it never really recovered. The advent of FEM/CFD can be

traced to the late 1970s. A landmark is the appearance in 1982 of Transonic Shock

and Multidimensional Flows: Advances in Scientific Computing edited by Meyer.

It is ironic that this volume contains an article by G Moretti pleading for Closer

Cooperation between Theoretical and Numerical Analysis in Gas Dynamics. The

Aeroelasticians combined the discrete approximation in structures FEM with CFD

for the aerodynamics. But the integration of the two via the boundary conditions was

always an art, often the more mysterious part. The battle between continuum theory

and computation is still enjoined. And as late as 2010 we see one unresolved point:

see [98] where the author writes: verification of numerical solutions is a step in

their scientific acceptability and a formal requirement of many engineering related

enquiries. . . However, these equations are known to have nonunique solutions. If the

true answer is not unique, what does correctness of approximation mean? As we

show in Chap. 7 that in viscous flow even existence of a solution is not proved

compounding this difficulty, because now we can ask: approximation to what? So

the conclusions from the computation have to be taken on faith.

It should also be noted that time is also discretized in the approximation but

there are phenomena, for example, in stochastic process theory, Gaussianness of the

innovation process in nonLinear filtering, which holds only in the continuous time

model (Girsanovs theorem: see, for example, [26]).

Chapter 2

Dynamics of Wing Structures

2.1 Introduction

This chapter deals with the dynamics of elastic structures: unswept fixed-wing and

free-free wings, modeled as beams of zero thickness, linear as well as nonlinear.

The main concern is with spectral analysis, modes and mode shapes as a means to

study stability. The language is that of abstract functional analysis: Hilbert spaces

and semigroup theory of operators for time domain description.

1945. See Fig. 2.1. It is a uniform rectangular beam. The co-ordinate axes are chosen

so that X -axis is the (rigid body) airplane axis also called the Chord-axis with the

width or chord length 2b:

b < x < b:

The one-sided wing span is ` along the Y -axis:

0<y<`

`=b

is high, justifying the flexible beam model. It is endowed with two degrees of

freedom: bending displacement h./ in a plane normal to the beam (plunge in

aeroelastic parlance) and the torsion angle ./ in radians (pitch in aeroelastic

DOI 10.1007/978-1-4614-3609-6 2, Springer Science+Business Media, LLC 2012

10 2 Dynamics of Wing Structures

TRAILING EDGE

U b

CG bx PITCH AXIS

ba

0

SPAN

b LEADING EDGE

SPAN = ; HALFCHORD = b

ANGLE OF ATTACK =

THICKNESS =

BEAM MODEL: = 0

parlance) about an axis parallel to the wing axis at distance bx from the cg as

shown in Fig. 2.1. The bending dynamics is described by an Euler equation:

R y/ C S .t;

mh.t; R y/ C EI h0000 .t; y/ D L.t; y/; t > 0I 0<y<1<1

(2.1)

R y/ C S h.t;

I .t; R y/ GJ 00 .t; y/ D M.t; y/; t > 0; 0 < y < `; (2.2)

L.t; y/ is the aerodynamic force per unit length (lift);

M.t; y/ is the applied aerodynamic.

The moment about the elastic axis per unit length, are both determined later from

the airflow model.

m D Mass per unit length

I D Crosssectional moment of inertia

EI D Bending stiffness

GJ D Torsional stiffness

S D Coupling constant D mbx , jx j < 1,

see Fig. 2.1 for x ; all constant, justifying the term uniform.

See [5] for more details on the physical units.

Here x can be positive or negative, and it is required that S 2 < mI .

The convention throughout is that the superdots denote the partial derivatives

with respect to time t and the superprimes the partial derivatives with respect to the

spatial co-ordinate y. Note that the structure variables do not depend on the chord

variable. To complete the dynamics we need to specify the end conditions.

2.3 Time Domain Analysis 11

End Conditions

1. For fixed-wing aircraft where the wing is attached to the fuselage (as in

Fig. 2.1) we add the Clamped-Free or CF end conditions:

.t; 0/ D 0I 0 .t; `/ D 0 t > 0: (2.3)

2. FF Free-free condition

This is typical of the recent UAV aircraft that have no fuselage and resemble a

flying wing. Here both ends are free.

0 .t; 0/ D 0I 0 .t; `/ D 0 t > 0: (2.4)

3. CC Clamped-clamped condition

This is relevant to non-aircraft applicationsee Chap. 10

h.t; `/ D 0I h0 .t; `/ D 0I .t; `/ D 0:

Our main concern is the stability of the structure. Stability of motion is a time

domain concept even if all the techniques involve Laplace transformation, Laplace

domain.

For an elementary treatment of vibration of beams reference may be made to

the classic treatise of Timoshenko [32] and the textbook [34], and the more recent

treatment relevant to aeroelasticity by Hodges and Pierce [5].

Here of course we use abstract functional analysis, albeit rudimentary, following

[61] where a model with three degrees of freedom with damping is analyzed and

[56] where specifically the Goland model is treated. Both of these follow the general

Hilbert Space formulation in [16, 55]. Not only does it make the presentation

compact and elegant, this level of abstraction is actually necessary for a precise

mathematical formulation of the structure dynamics in airflow, the subject matter of

this book.

Let H denote the Hilbert space:

L2 0; ` L2 0; `

with elements

h./

xD :

./

12 2 Dynamics of Wing Structures

This choice of space is motivated by the fact that the elastic potential energy is

given by

Z Z !

` `

1 00 0

EI jh .y/j dy C GJ

2

j .y/j dy :

2

(2.5)

2 0 0

The Domain of A (denoted D(A)): where we need to distinguish between CF

clampedfree, FF freefree and CC clampedclamped end conditions:

For CF

h

x j h0000 ./ 2 L2 0; `I 00 ./ 2 L2 0; `; h .0/ D 0I h0 .0/ D 0I

i

.0/ D 0 h000 .`/ D 0I h00 .`/ D 0I 0 .`/ D 0 :

For FF

h

x j h0000 ./ 2 L2 0; `I 00 ./ 2 L2 0; `; h000 .0/ D 0I h00 .0/ D 0I

i

0 .0/ D 0 h000 .`/ D 0I h00 .`/ D 0I 0 .`/ D 0 :

For CC

h

x j h0000 ./ 2 L2 0; `I 00 ./ 2 L2 0; `; h .0/ D 0I h0 .0/ D 0I

i

.0/ D 0 h .`/ D 0I h0 .`/ D 0I .`/ D 0 :

principle, it is less messy to go ahead with one operator in the abstract theory, but

making the necessary changes in the concrete calculations.

EI h0000

Ax D : (2.6)

GJ 00

Thus defined it is standard [16] to verify that the domain is dense in H and that A

is closed on its domain. Note further that for x in D(A):

Ax; x D 2 potential energy 0.

2.3 Time Domain Analysis 13

Suppose for some x in the domain of A, we have

Ax D 0:

Then

Ax; x D 0:

Or the potential energy is zero, and hence

x D 0:

Ax D x; x nonzero

The set of eigenvalues is countable; the eigenvalues fk g of A are positive and

the corresponding eigenfunction spaces are each of dimension 1. The eigenfunctions

fk g are orthogonal, complete, span the space, and are orthonormalized for our

purposes here. The resolvent, denoted R.; A/, is compact, because the beam is

finite. Any complex number not equal to k for any k is in the resolvent set. Also

the resolvent is HilbertSchmidt [16]

1

X 1

X 1

R.; A/ k D2

<1

j k j2

kD1 kD1

1

X x; k

R.; A/x D k : (2.7)

k

kD1

Moreover the eigenfunctions are complete in H ; see [16]. Any element in H has

the modal representation

1

X 1

X

xD x; k k ; jx; k j2 < 1:

kD1 kD1

The potential energy can be defined for a largerpclass of elements in H than

D.A/. For this we need the square root of A denoted A. It can be defined in terms

of the eigenfunctions as

1

X

p

x2D A if k jx; k j2 < 1

kD1

14 2 Dynamics of Wing Structures

and we define

1 p

X

p

Ax D k x; k k (2.8)

kD1

p

p that zero is not an eigenvalue of A.

from which it follows

The domain of A is thus larger than that of A. For x in D.A/

hp p i

Ax; Ax D Ax; x D 2 Potential energy :

p

We then define potential energy for all elements in D A as:

1 p

jj Axjj2 :

2

This is the largest domain for which p

the potential energy can be defined. We rarely

need to know the precise domain of A (see [55] for details on the domain) or the

precise form; we should note, however, that it may not be a differential operator. See

[63] and below.

The time domain solution: S D 0.

We begin by rewriting the dynamic equation (2.1) and (2.2) in abstract form as

L.t/

MxR C Ax D ; (2.9)

M.t/

where M is the nonsingular nonnegative definite 2 2 matrix

ms

s I

and is the mass/inertia operator; the operator A can be called the stiffness operator,

extending the finite-dimensional definitions.

Let us introduce next the energy space H consisting of elements of the form:

p

xl

Y D x1 2 D A ; x2 2 H

x2

hp p i

Y; ZE D Ax1 ; Az1 C Mx2 ; z2 ; (2.10)

2.3 Time Domain Analysis 15

where

Zl p

ZD Z1 2 D A ; z2 2 H:

Z2

Theorem 2.1. H is a Hilbert space.

Proof. We need to show that every Cauchy sequence converges to an element in the

space.

Let

xn

Yn D

yn

p

Axn

p

is a Cauchy sequence in H . Because 0 is not an eigenvalue of A,

p p

R 0; A Axn D xn ;

p p

where R ; A denotes the resolvent of A, converges to an element x in H ,

p p

actually in D A , because A is closed.

hp p i

My; y D M y; M y ;

follows that Yn converges to an element in H. Hence H is a Hilbert space. t

u

p

p Note that what we have proved is that the domain of A is closed in H because

A has a bounded inverse. This is the largest domain for which the potential energy

can be defined. p

We rarely need to know the precise domain of A (see [63] for details on

the domain) or the precise form; we should note, however, that it may not be a

differential operator. In our case we can see that

p

p h EI h00

A D :

0 0

But the square root for the torsion operator is no longer a differential operator.

16 2 Dynamics of Wing Structures

p Z

1 ` Sin s 0

A D gI g.s/ D `

./d

` 0 Cos s

`

Cos `

Note that the norm in H has a physical significance. jjY jj2 D 2 total energy D 2

(potential energy C kinetic energy).

Hence we refer to it as energy space.

As in finite dimensions, we go on to the state space formulation of the problem.

Let

x.t/

Y .t/ D :

P

x.t/

Then the second-order equation (2.5) goes over into the first-order in time equation:

where

vl .t/

V.t/ D ;

v2 .t/

where

0

v1 .t/ D ;

0

L .t/

v2 .t/ D ;

M .t/

where L.t/ denotes the element L.t; y/ and M.t/ the element M.t; y/ in H, and

the operator A is defined by:

yl 0 I yl y2

A D 1 D

y2 M A 0 y2 M1 Ay 1

with domain:

y1 2 D.A/;

p

y2 2 D A :

2.3 Time Domain Analysis 17

Moreover we can readily verify that the adjoint (in H) A is given by A:

A C A D 0

and hence [16] A generates a C0 semigroup S.t/; t 0. This is not true

incidentally if the space is H H . See [39]. The solution of the homogeneous

equation

YP .t/ D AY .t/

is given by

Y .t/ D S.t/ Y .0/ for Y .0/ in D.A/:

Lemma 2.2. The semigroup S./ is isometric:

Proof. For Y in H,

P

m.t/ D .A C A /S.t/Y; B.t/Y D 0

jjS.t/Y jj D jjY jj:

But the domain of A is dense in H, and hence (2.8) follows. t

u

An equivalent statement is that

S.t/1 D S.t/ :

YP .t/ D AY .t/

is satisfied only for Y .0/ in the domain of A.

The nonhomogeneous equation we started with:

Z t

Y .t/ D S.t/Y .0/ C S.t /./d t >0 (2.13)

0

18 2 Dynamics of Wing Structures

for Y .0/ in the domain of A, and may need to be interpreted in the weak sense

depending on the function ./ (as a function of time) [3] but we do not need to go

into this until later.

Our main concern is the stability of the solution or of the semigroup S./. This

means we need to study the spectrum of A. Because R.; A/ the resolvent of A is

compact, we need only consider the eigenvalues of A.

Unraveling

y1

.I A/Y D 0; Y D ;

y2

we have:

y1 y2 D 0;

y2 C M1 Ay1 D 0:

Hence

y2 D y1 ;

2 My1 C Ay1 D 0I Ay1 D 2 My1 ;

Ay1 ; y1 D 2 My1 ; y1 :

because if it is, so is y1 ; y2 .

The eigenvalues then are defined by

y2 D i!y1 ;

Ay1 D ! 2 My1 ;

y1 is an eigenvector of A but with respect to M rather than the identity. Again the

eigenvectors are countable and complete in the Hilbert space H with an equivalent

inner product:

x; yM D x; My:

See [2].

Hence the eigenvalues are purely imaginary:

k D i!k ; !k > 0:

2.4 Structure Modes and Mode Shapes 19

The !k are thus defined as the modes of the structure (in radians) and because

.1=!k / goes to zero, can be arranged in increasing order of magnitude so that we can

talk about the kth mode without ambiguity. We may note here that the asymptotic

behavior of the modes is of little interest to us in practice, where only the first

few modes play a role. Let k denote an eigenvector of A corresponding to the

eigenvalue k .

Note that k has the form

k

k D :

i !k k

In our particular case we can calculate the modes !k and the mode shapes k by

solving the differential equations

2 mh C EI h0000 D 0; (2.14)

2 I GJ 00 D 0: (2.15)

Note that this is no more than taking Laplace transforms of the time domain

equations, familiar in engineering.

Thus k is of the form:

hk 0

or

0 k

and we can distinguish between the bending modes and the torsion modes.

We have corresponding to the bending motion

k .y/ (2.16)

h.0/ D 0 D h0 .0/;

h00 .`/ D 0 D h000 .`/:

As may be expected, this is a classical result already found in Timoshenko, 1928

[32] and in textbooks [34]. The modes !k are determined by the equations:

and

1=4

1 EI

!k D k 0 < k " :

` m

20 2 Dynamics of Wing Structures

Mode shape:

Torsion modes: These are the zeros of

cos h

!k ` D cos

!k ` D 0;

s

GJ

!k D .2k 1/ ; k D 1; 2: (2.18)

2` I

Mode shape

.2k 1/

const sin y 0 < y < `: (2.19)

2`

An obvious comment here is that the modes decrease linearly in magnitude as the

span length increases. As a rule the structure also increases in flexibility so that the

density of modes also increases. There are more modes to be considered as length

increases.

We note here that for the clampedclamped case the modes are determined as the

roots of:

"p p # "p p #!

p 2 `m1=4 jj 2 `m1=4 jj

EI GJ 1 C cos cosh

EI 1=4 EI 1=4

p

` I

sinh p D0

GJ

The modes are undamped. The energy remains constant and does not increase or

decrease. The structure model is neutrally stable.

1

!

X 1

Y D Y; k k C Y; k k ; (2.20)

kD1

2!k2

2.4 Structure Modes and Mode Shapes 21

where

k

k D I Ak D !k2 k I Ak D i !k k I

i !k k

A k D i !k k ; !k > 0; (2.21)

X1

1

2

< 1I k ; k D !k2 I k ; k D 0 (2.22)

kD1

!k

k

Mk ; j D jk

D k

i !k k

Then

1

!

X 1 i!t

S.t/Y D Y; k e k C Y; k e

i!t

k ; (2.23)

kD1

2!k2

1

!

X 1 k k

R.; A/Y D Y; k C Y; k : (2.24)

2!k2 i !k C i !k

kD1

Proof. Follows [16, 54]. The modal expansion uses completeness of the fk g.

Otherwise, because, the eigenfunctions are orthogonal, the proof of the expansions

is straightforward. t

u

A useful alternate representation of the semigroup and the resolvent is through the

Greens function; see [44, vol. 1]. Here we develop it using the modal representation.

Thus S.t/Y is the function

1

X1

i!k t

Y; k e i!k t

k .s/ C Y; k e k .s/ D Y; G .t; s; :/; (2.25)

kD1

2!k2

where

1 i!k t

1

X

i!t

G.t; s; / D e k ./k .s/ C e k ./ k .s/ (2.26)

kD1

.2!k2 /

22 2 Dynamics of Wing Structures

is then the Greens function for S.t/. Similarly the resolvent is simply the Laplace

transform which is analytic in excepting the imaginary axis.

O

R.; A/Y D Y; G.; s; :/ (2.27)

X1

O 1 1 1

G.; s; / D k ./k .s/ C k ./k .s/ ; i !k

2!k2 i !k C i !k

kD1

(2.28)

Z ` Z `

O

jG.; s; /j2 dsd < 1: (2.29)

0 0

Next we consider the general case allowing for nonzero S . The main question is

how much the coupling changes the modes. We do expect the mode shapes to be

coupled. We limit consideration to the CF case here.

The equations corresponding to (2.1, 2.2) are:

To solve this set of equations we proceed to consider the state-space version.

Here we follow [56].

Thus let

Y .s/ D Col.h.s/; h0 .s/; h00 .s/; h000 .s/; .s/; 0 .s/:

Then we have

Y .0/ D Col 0; 0; h00 .0/; h000 .0/; 0; 0 .0/

0 1

0 100 0 0

B0 0C

B 010 0 C

B C

B0 001 0 0C

A./ D B C (2.32)

Bw1 000 w2 0C

B C

@0 000 0 1A

w3 000 w4 0

2.4 Structure Modes and Mode Shapes 23

m

w1 D 2 ;

EI

S

w2 D 2 ;

EI

S

w3 D 2 ;

GJ

I

w4 D 2 ;

GJ

where w2 and w3 are the coupling terms.

The eigenvalues in the CF case are the zeros of dc ./ D Det Dc ./.

We define the 3 3 matrix:

Dc ./ D P e `A./ Qc ;

0 1

00 0

0 1 B0 0 0C

B C

001000 B C

B1 0 0C

P D @0 0 0 1 0 0A Qc D B C:

B0 1 0C

000001 B C

@0 0 0A

00 1

Df ./ D P e `A./ Qf ;

where

0 1

10 0

B0 1 0C

B C

B C

B0 0 0C

Qf D B C

B0 0 0C

B C

@0 0 1A

00 0

where

0 1

100000

Pc D @0 1 0 0 0 0A :

000010

24 2 Dynamics of Wing Structures

Thus defined d./ (generic for both dc and df ) is an entire function of finite order

(see [10]).

For S D 0,

Dc ./DP e A./ Qc

0 1

1 sinw11=4 ` C sinhw11=4 `

B .cosw1 1=4

` C coshw1 1=4

`/ C

B 2 2w11=4 C

D B 1 1=4 C

@ w1 .sinw11=4 ` C sinhw11=4 `/ 12 .cosw11=4 ` C coshw11=4 `/ A

2

0 0

this yields:

1

dc ./ D .1 C cos `/cosh `cosh

`; (2.33)

2

where

1=4 1=2

D w1 I

D w4 ;

To make explicit our interest in the dependence on S, let us modify the

notation to:

A.; S / in place of A.)

d.; S / for d./

so that d.; 0/ is given by d./.

The function is clearly analytic in S.

Let

Ap .; S / D .A.; S / A.; 0//=S:

0 1

0 000 0 0

B C

B 0 000 0 0C

B C

B 0 0C

B 0 000 C

B C

B 2 C :

B 0 000 0C

B C

B EI C

B 0 000 0 0C

B C

@ 2 A

000 0 0

GJ

2.4 Structure Modes and Mode Shapes 25

Now

d tA.;S /

e D A.; S /etA.;S /

dt

D A.; 0/etA.;S / C SAp .; 0/etA.;S / :

Z t

etA.;S / D etA.;0/ C S e.t /A.;0/ Ap .; 0/eA.;S /d; t > 0:

0

We can treat this as a matrix Volterra equation for etA.;S / and the solution has the

expansion:

1

X

e1A.;S / D e1A.;0/ C S n Fn .1/;

nD1

where

Z 1

F1 .1/ D eA.;0/.1 / Ap .; 0/eA.;0/ d:

0

For terms of higher order, see [70]. As shown there, d.; S / is a function of S 2 and

using the first term in the series yields

where

1

d.; 0/ D .1 C cos ` cosh `/ cosh

`

2

1 p

d2 .; 0/ D . 2

2 .4. 2

2 / coshl

sinl

2

p p

.. 2 C

2 / sinhl

sinhl

/

p p

C coshl .. 2 C

2 /2 C . 2

2 /2 cos2 l

p

C 4 2

2 .2 cosl coshl

cosh2 l

/

p p p

4

. 2

2 / sinl sinhl

/ C 4

p p

coshl

.2

C . 2 C

2 / sinhl sinhl

/

26 2 Dynamics of Wing Structures

p p

C cosl .. 2

2 /2 . 2 C

2 /2 cosh2 l

p p

4 2

2 cosh2 l

C 4

. 2 C

2 / sinhl

sinhl

///=.8m. 4 2

4 /2 I /: (2.35)

root finding algorithm, we can see that for small S it can be approximated by

k C k ; (2.36)

where

k D S 2 d2 .k ; 0/=d 0 .k; 0/

and the main thing to note is that

The modes are damped, and the mode shapes are now coupled. The bending mode-

shape vector now has a nonzero torsion component proportional to S 2 , and similarly

for the torsion-mode shape there is a bending component. But the extra components

being small, we continue to identify them as bending or torsion modes. Calculating

these would take us too far from our main interest.

Enhancement

We have seen in Sect. 2.4 that the structure model with zero coupling is neutrally

stable. The system energy neither decreases or increases whatever the initial

condition. Experience shows that all modes decay to zero and the higher the

damping the higher the mode. However, the lower modes may not decay as fast

as we would like. Hence we need to instrument controls to increase the damping

without, however, destabilizing the structure and without increasing the damping in

any mode. Unfortunately, the inherent damping is difficult to model and we need

to design a controller without knowing the damping and yet without degrading the

inherent damping. Fortunately we can indeed design such a control, a triumph of

control theory of structures. In any real system we can only provide for a finite

number of control inputs. Hence we may assume that the controls denoted u./,

Let B denote the control operator so that (2.5) now includes controls and we have

Mx.t/

R C Bu.t/ C Ax.t/ D v.t/:

2.5 Robust Feedback Control Theory: Stability Enhancement 27

note that D is linear bounded self-adjoint and nonnegative definite. Thus we have

finally:

MxR C D x.t/

P C Ax.t/ C Bu.t/ D 0: (2.38)

An example of a damping operator D is

p p p p p p

2 M T M; T D M1 A M1 ; jj < 1;

velocity, going back to Timoshenko [32].

Here we simply assume it is compact and nonnegative definite, zero not in the

spectrum.

With H as before we go to the state-space form:

where

0 I

Ad D ;

M1 A M1 D

0

BD ;

Ml B

.Ad C Ad /Y; Y E D Dx 2 ; x2 ;

where

xl

Y D :

x2

Sd Y; Y Y; Y

with compact resolvent. Our main result is to show that we can design a feedback

controller that is robust in that it is not required to know either D or A and is such

that the structure is strongly stable.

Controllability

the original notion in finite dimensions extended to infinite dimensions. Let R

28 2 Dynamics of Wing Structures

space H. We say that the semigroup is strongly stable if for each Y in H,

jjS.t/Y jj ! 0 as t ! 1:

We say that the pair R B is controllable if t 0 S.t/Bu; uRm is dense in H.

Here we can state a basic result of control theory.

Theorem 2.4. Suppose the semigroup S./ is a contraction, the generator has a

compact resolvent, and .A B/ is controllable. Then the semigroup generated by

A BB

is strongly stable.

Proof. See [16]. t

u

Remark. It is known (see [16]) that a finite-dimensional control cannot guarantee

a uniform decay rate for all modes (exponential stability). The rate of decay will

eventually go to zero as the mode number increases indefinitely. Strong stability

means that the energy in any initial state will eventually decay to zero.

Let us apply this to our case, to the semigroup S./ with generator A and control

operator B.

Theorem 2.5. The pair A; B is controllable if and only if for any nonzero eigen-

vector of A,

B 0:

Proof. Suppose contrarywise B D 0 for some nonzero eigenvector with

eigenvalue i!.

Then

B ; u D ; B u D 0 for every u in Rm :

Now

; S.t/B u D S.t/ ; B u D eit ! ; B u D 0:

Hence the set [

S.t/Bu; u Rm is not dense in H;

t 0

which is a contradiction. t

u

And the argument can be retraced for the only if part readily.

We assume now that A, B is controllable. The resolvent of A is compact and B

is finite-dimensional and hence the semigroup generated by

A BB

is strongly stable.

2.5 Robust Feedback Control Theory: Stability Enhancement 29

Let us see what this implies. By modal stability we mean that each mode decays

to zero, actually exponentially, with the rate determined by the real part of the

eigenvalue. In the case where the dimension of the space is not finite this does not

mean strong stability. It only implies that the system is damped but the rate of decay

depends on the element. Exponential or uniform stability is defined in terms of the

semigroup S./ by requiring that the stability index:

1

Inf LogjjS.t/jj D !o < 0;

t

logjjS.t/jj

! !o as t ! 1:

t

And given any > 0, we can find M such that

In our present structure dynamics context we may interpret this as guaranteeing

a uniform exponential decay rate for all initial states. Here we have a negative result

[16] that this cannot be achieved by finite-dimensional controllers, hence not by

means of active controllers.

Let us now return to show that the feedback control

u.t/ D B Y .t/

does not destabilize the structure model whatever the damping operator D is, where

R C D x.t/ P D0

D C BB * ;

which is

D:

If is any mode of this system we see that the damping

.D C BB /; D; :

30 2 Dynamics of Wing Structures

1

X

T rBB D jjB k jj2 < 1;

kD1

where k denote the modes, so that the active damping introduced goes to zero as

the mode number increases. It is usually much smaller than any natural damping.

Another point is that one cannot guarantee a specific value of damping for any mode.

Also, few reliable models of damping are known, so designing controls based on any

damping model can be hazardous. For a successful use of this control law see [103].

Self-Straining Actuators

we would need an infinite-dimensional controldistributed control as opposed to

point controllers on the boundaryto achieve this, which is of course not physically

realizable.

A class of actuators using piezzo strips that are self-sensing and self-straining

is described in (see [79] and the references therein) with the potential to yield

exponential stability.

We can extend our theory to investigate this class. We begin with a beam model.

The displacement of the piezzo-electric strip charges a condenser which is then bled

as a current for actuation. The differentiator circuit is an integral component, and

we have really rate feedback. We begin with the torsion dynamics:

R s/ GJ 00 .t; s/ D 0

I .t; 0 < s < `;

.t; 1/ D 0;

GJ 0 .t; 0/ C g P .t; 0/ D 0;

where g is the actuator gain, and we need to study how the system stability depends

on the gain. We may consider this as a singular mass matrix version of the boundary-

control problem:

R `/ GJ 0 .t; 0/ C g .t;

m.t; P 0/ D 0;

The novelty here is the inclusion of the values at the ends as separate from the

functions.

Thus let H1 denote the Hilbert Space:

L2 .0; `/ R1 ;

2.5 Robust Feedback Control Theory: Stability Enhancement 31

with elements

f ./

xD :

b

Let A denote the operator with domain in H1 :

0 1

f ./jf 00 ./ L2 .0; `/;

D.A/ D @ f .`/ D 0 A

f .0/

and

GJ f 00 ./

Af D :

GJ f 0 .0/

Z `

Af ; f D GJ jf 0 .s/j2 ds;

0

p

which is twice the elastic energy. Denoting by A the positive square root of A, it

can be seen following an argument similar to the one given in [11] that the domain:

" #

p f ./ 0

D. A/ D ; where f .`/ D 0 f ./ L2 .0; `/ ;

f .0/

p

but A is not a differential

p operator. We note that zero is not in the spectrum of A

and hence that of A . Next, as before we define the energy space:

p

HE D D. A/ L2 .0; `/;

hp

x y p i

; D Ax; Ay C I f; g

f g E

and

x x

; D 2 (total energy: kinetic C potential):

f f E

We define next the operator A with domain in HE with domain that is a little

complicated:

32 2 Dynamics of Wing Structures

y fl ./

D.A/ D Y D with y D.A/; D ;

f2 ./ fl .0/

! #

f2 ./ p

and GJ 0 D. A/ :

f .0/

g l

f2 .`/ D 0;

f20 ./ L2 .0; `/:

0 1

f2 ./

B GJ 0 C

AY D B f .0/ C :

@ g 1 A

GJ 00

f

I 1

./

Then A is closed linear with dense domain and compact resolvent. Also

.GJ / 0

AY; Y E D GJ f2 ./; f100 ./ jf1 .0/j2 C GJ f100 ./; f2 ./:

g

Hence

jf10 .0/j2

Re AY; Y D .GJ /2 0:

g

This is enough to prove that A generates a C0 semigroup S.t/; t 0, a contraction

semigroup, but actually exponentially stable, as we show presently. Of greater

interest to us are the eigenvalues and how they depend on the gain g.

Spectrum of A

Let A D which with

0 1

f1 ./

D @ f1 .0/ A

f2 ./

yields

where

I

2 D ;

GJ

2.5 Robust Feedback Control Theory: Stability Enhancement 33

Now the function on the left is an entire function of order one and of completely

regular growth [6] with a sequence of zeros given by

p

GJ I

tanh .`/ C D 0:

g

Or

p

g C GJI

e 2`

D p ;

g GJI

k D jk j i !k ;

s

g C pGJI

1 GJ

jk j D log p ;

2` I g GJI

s

.2k C 1/ GJ p

!k D ; g < GJI

2` I

s

GJ p

Dk ; g > GJI ;

` I

k D nonnegative integers:

p

gc D GJI

A plot of the relative damping constant ` versus the relative gain g=gc is

given in Fig. 2.2.

1 1Cx

log abs :

2 1x

As can be seen from Fig. 2.2, the limiting damping becomes infinite at the critical

gain.

Let us explore further what happens at the critical gain.

We have

d./ D cosh` C sinh` D e`

34 2 Dynamics of Wing Structures

(relative) gain

2.0

1.5

1.0

0.5

D 1;

in this case is that the response vanishes in finite time T determined by gc . All

states are evanescent. This has been given the rather naive if picturesque name: a

disappearing solution, in another context, scattering theory. See [80].

But this is of course impossible in practice for structure response, for many

reasons; see [62] for details. Primarily it is the degradation in the operational

amplifier differentiator as the gain increases. The gain gc is not physically attainable.

See [79].

Although our primary interest is in the modal response, we complete the analysis

for any noncritical gain by showing that we have exponential stability, that any initial

state will decay at an exponential rate. The modes are given by

0 1

k ./

k D @ k .0/ A ;

k k ./

where

k .s/ D ak sinhk v.1 s/ 0<s<`

are not orthogonal. But we can create a biorthogonal system; see [1] for more.

Define

0 1

k ./

k D @ k .0/ A ;

k k ./

2.6 Nonfixed Wing Models: Flying Wings 35

k ; j D 0 for k j:

Hence if

X Y; k

Y D bk k ; then bk D

k ; k

and

X

S.t/Y D e t ei!k t bk k

X

D e t ei!k t bk k ;

The corresponding theory for bendingEuler beams with self-straining

actuatorsis treated in [70], but there is no analagous superstability at the gain

corresponding to maximum damping. This would indicate that we can attain more

damping by torsion actuators, a fact corroborated by experiment.

Our interest is of course in the stabilization of the structure subject to aerody-

namic loading. We would expect that if we have a controller that does well in still

air, it should be a candidate for use in airflow as well. This is studied in Chap. 8.

We again consider the basic uniform Goland model, but it is no longer attached at

one end point to the fuselage; it is simply a flying wing. Both ends are free, FF.

Hence we use the matrix Qf in what follows. But it is articulated with discrete

masses placed along points (nodes) on the beam. This is typical of the recent

Helios UAV Flying Wing. See [81]. Such a model for a fixed-wing aircraft was

also considered earlier by Goland and Luke [77]. There is only a single wing

span, 0 < s < `. The masses mi , are at s D si ; i D 0; 1; : : : ; m C 1, with

s0 D 0; smC1 D 1. There are thus .m C 2/ masses maximum. We have only to

set the mass to be zero if there is none at si for any i ! See Fig. 2.3.

40 ft

60 lb

articulated beam

36 2 Dynamics of Wing Structures

Between nodes: si < s < si C1 , we have the Goland equations (cf. (2.1) and

(2.2)):

R s/ C S .t;

mh.t; R s/ C EI h0000 .t; s/ D L.t; s/; t > 0I si < s < si C1 ; (2.39)

I R .t; s/ C S h.t;

R s/ GJ 00 .t; s/ D M.t; s/; t > 0; si < s < si C1 : (2.40)

To allow for the discontinuities at the nodes we follow the technique in [2]. We

essentially incorporate the nodes into the function space: let

H D L2 .0; `/ RmC2 :

0 1

f ./

B f .0/ C

B C

B C

B f .s1 / C

B C

f DB f .s / C

B : i C:

B : C

B : C

B C

@ f .sm / A

f .`/

0 00 1

f ./; si < s < si C1

g D GJ @ f 0 .si C/ f 0 .si / A ; (2.41)

i D 0; 1; : : : ; m; m C 1

where

si C D limit si C ; 0 < ! 0;

si D limit si ; 0<!0

0 D 0I `C D `:

domain, nonnegative definite, and

Z 1

A f; f D GJ jf 0 .s/j2 ds 0; (2.42)

0

2.6 Nonfixed Wing Models: Flying Wings 37

achieve this. The moment-balance end conditions we need at each node are:

ri2 mi C mi `i h.t; (2.43)

where ri is the radius of gyration of the mass mi about the elastic axis at normal

distance `i . We note that if there are no masses at the ends 0 or `, then

Hh D L2 .0; `/ RmC4

0 1

f ./

B f 0 .0/ C

B C

B f .0/ C

B C

B C

B f .si / C

D.Ah / D B C 0000

B :: C with f ./ L2 .0; `/

B : C

B C

B f .sm / C

B C

@ f .`/ A

f 0 .`/

and

0 1

f 0000 ; .s/i < s < si C1

B f 00 .0/ C

B C

B 000 C

Ah f D gI g D EI B f .si C/ f 000 .si /; C (2.45)

B C

@ i D 0; : : : ; m; m C 1 A

f 00 .`/

nonnegative definite and

Z 1

Ah f; f D EI jf 00 .s/j2 ds (2.46)

0

the elastic energy in the bending mode. In addition we have to impose the nodal end

conditions

R si / C `i .t; (2.47)

38 2 Dynamics of Wing Structures

at the nodes s D si , and if there are no masses at the ends, the freefree conditions:

h00 .t; `/ D 0 D h00 .t; 0/:

H D Hh H

Ah 0

As D

0 A

on H with domain:

D.As / D D.Ah A /

and

Ah h h

As x D for x D in D.As /:

A

Z ` Z `

As x; x D EI jh00 .s/j2 ds C GJ j 0 .s/j2 ds;

0 0

p in the structure. As before we note that

As has apositive square root denoted As , and the elastic energy is defined on

p

D As as

hp p i

E.x/ D As x; As x ;

where we use the same inner-product notation for all three spaces we have

introduced.

Finally the structure dynamic equations can now be expressed:

R C As x.t/ C Bu.t/ D 0

M x.t/ (2.48)

allowing for an actuator on the structure and control input u./ as in the fixed-wing

model, and M as before.

2.6 Nonfixed Wing Models: Flying Wings 39

Structure Modes

We now calculate the structure modes for the uncoupled case S D 0. We use the

same notation as before in Sect. 2.4, including A() but we need to introduce more

to account for the fact that now we have several interconnected sections. Thus let

Ei denote the 6 6 matrix

0 1

1 000 0 0

B 0 100 0 0C

B C

B 0C

B 0 010 0 C

B C

B 2 mi mi `i C

2

B 0 0 1 0 C: (2.49)

B EI C

B EI C

B 0 0 0 0 1 0 C

B C

@ 2 m ` 2 mi ri2 A

i i

000 1

GJ GJ

Then the modes are the zeros of

If mi ; ri , and `i are zero, then we revert to d./ in Sect. 2.4. An important case is

the symmetric case where all the masses are on the elastic axis so that ri ; `i are all

zero.

F1 . ;

/sin

`;

where

1 2 m 1

F1 . ;

/ D 4

.1 C cosh ` cos `/ .1 C cosh ` cos `/

2 EI

!

1Ci .1 C i / ` .1 C i / `

sin sinh :

4 2 2

40 2 Dynamics of Wing Structures

As we expect, the mass at the center does not affect the torsion modes. Hence we

call the roots of F1 . ;

/ D 0 the bending modes which now depend on

as

well. Obviously we get back the beam modes for large mass m1 .

2. Several Point Masses.

This result readily generalizes to the case of several point masses, but none at the

end points. Following the Helios model [81], we consider (N 1) masses mi at si ,

placed symmetrically:

`

si D i ; i D 1; : : : ; N 1;

N

d./ D Det P eA./.`=N / EN 1 : : : : eA./.`=N / E N : : : eA./.`=N / Q which again

2

factors as

FN . ;

/ sin

`;

FN . ;

/ D 0:

So far we have only considered linear models, where the dynamics are characterized

by a linear equation. One may argue that real life structures are nonlinear. One

can of course come up with nonlinear models but the question of whether they are

well formulated in terms of existence and uniqueness of solutions is often ignored.

See for example [75, 81, 82] where in fact the elastic part is eventually truncated

lumpedto yield ordinary equations in place of partial differential equations. One

immediate difficulty with the nonlinear model is that there is no notion of modes; we

have no concept of spectrum of the nonlinear operator. And even if one could define

the notion, it is not clear what role it plays in the stability of the system, which is

our primary concern; see below.

Also we can no longer define an inner product based on the energy. We consider

two models that are essentially extensions of the generic Goland beam model.

The first one is due to Beran et al. [75]. We call it the BeranStraganac model.

This has only two degrees of freedom but is highly nonlinear.

2.7 Nonlinear Structure Models 41

We state this in our notation where we replace their w./ by h./ and ./ by ./.

We have, including the discrete masses, in their delta-function notation:

mhR C S R C EI h0000 C Ms hR xs .t;

R y/ .y ys / D Dx .h0 .h0 h00 /0 /0 C L.t/

(2.52)

C cross-product terms involving time derivatives which we omit

C M.t/: (2.53)

For the derivation and details we have omitted see [75]. Our main point here is that

the equation is nonlinear so that there is no notion of eigenvalues, of the spectrum

of the differential operator.

Of course the authors do not consider the question of whether these continuum

equations have a unique solution, and all calculations are based on discretized

models. Indeed, it would be a major task to establish this. We return to this model

in the succeeding chapters.

The second example we consider is the DowellHodges model [76].

This is a nonlinear model with three degrees of freedom as in [61]: the torsion angle

./ about the elastic axis and two bending variables, the plunge h./ in the xz plane

and an additional in-plane bending in the structure xy plane. We show existence and

uniqueness by a constructive method of solution; our emphasis is on the differences

from the linear Goland model.

The continuum equations (consistent with our notation) are:

mh.t;

D mg sin' C L.t; y/; (2.54)

mRv.t; y/ C EI 2 v0000 .t; y/ C .EI 2 EI 1 /..t; y/h.t; y/00 /00 D mg cos'; (2.55)

42 2 Dynamics of Wing Structures

R y/GJ 00 .t; y/C.EI 2 EI 1 /h.t; y/00 v.t; y/00 D M.t; y/; 0 < tI 0 < y < `;

I .t;

(2.56)

where we have omitted the tip masses in [76] but retained the gravity terms as part

of the forcing terms on the right. The angle ' is the angle between the x-axis and

the gravity vector. For the derivation and other details reference should be made to

the original paper [76].

The in-plane (in the xy-plane) displacement is denoted v./ and does not impact

the aerodynamics directly. This is a nonlinear system of equations (including the

airflow, based on the nonlinear aerodynamics described in Chap. 3).

We shall first consider the case where the airspeed is zero so that there is

no aerodynamic loading, as in the linear case. There is no notion of modes that

determines stability any more. We include the gravity terms in the aerodynamic

loading and hence omit them for the pure structure case. Then we have:

mh.t;

mRv.t; y/ C EI 2 v0000 .t; y/ C .EI 2 EI 1 /..t; y/h.t; y/00 /00 D 0;

R y/ GJ 00 .t; y/ C .EI 2 EI 1 /v.t; y/00 h.t; y/00 D 0;

I .t;

0 < t; 0 < y < `; (2.57)

plus CF end conditions, with the notation:

0 1

h .t; y/

x.t; y/ D @ v t; y/ A t > 0; 0 < y < `:

.t; y/

One surprising feature of the model although still nonlinear, and not typical, is that

we do have modes; in fact those of the linear system.

h.t; y/ D 0;

v.t; y/ D 0;

I R .t; y/ GJ 00 .t; y/ D 0

plus CF or FF end conditions, yields all the pitching modes (cf. (2.5)). In similar

fashion, we also get all the bending modes. Thus we have:

.t; y/ D 0;

v.t; y/ D 0;

R y/ C EI 1 h0000 .t; y/ D 0

mh.t;

2.8 Beams of Infinite Length 43

.t; y/ D 0;

h.t; y/ D 0;

mRv.t; y/ C EI 2 v0000 .t; y/ D 0

But of course we do not have the modal superposition property, except for the

class of functions for which only one co-ordinate is nonzero for all functions. Hence

it does not tell us much about the stability of the system. We return to this model in

succeeding chapters.

So far we have not taken in-thickness into account. Ideally of course one would

want a plate model, however thin. For aircraft wings we have to model the tear-drop

shapethe camber; see [18].

This brings considerable complication in the continuum model, explaining why

all current works immediately discretize to finite dimensions. An approach to

include camber is presented in [107] for flutter analysis.

beam, where the beam length is no longer finite. The main thing here is that we

need to continue with the same notion of energy:

Z 1 Z 1

1 1

ED EI jh00 .s/j2 ds C GJ j 0 .s/j2 ds:

2 0 2 0

then it would need to satisfy the same conditions at the end s D 1. Thus let H D

L2 0; 1 and let h./ and h0 ./; h00 ./, h000 be in H with

h.0/ D 0;

h0 .0/ D 0:

Then

d 0

jh.s/j2 D h0 .s/h.s/ C h.s/ h.s/

ds

44 2 Dynamics of Wing Structures

and hence

Z L Z L

d

jh.s/j2 ds D jh.L/j2 D .h0 .s/h.s/ C h.s/.h.s//0 /ds;

0 ds 0

which converges as L ! 1 to

2 Reh; h0 :

L!1

but the limit must be zero because h.0/ is in L2 0; 1. In as much as h00 is also in

L2 0; 1, we have that h and h0 are continuous with

h.1/ D 0 D h0 .1/ D 0:

.0/ D 0 D .1/:

Thus C at one end implies C at the other end also. Thus we have the case CC.

The modes are inversely proportional to the length, therefore we see that the

modes are not defined. Or, the point spectrum of A is empty.

Our interest is in the stability of the structure. Thus we can proceed as we did in

Sect. 2.3 with CC end conditions, and ` replaced by 1. The first problem then is the

definition of the square root of A. Here we need to invoke Fourier transforms. Thus

Q as the Fourier transform of h; Q ./ as the Fourier transform of ./.

we denote h./

h hQ

x./ D I x./

Q D Q ;

Z 1 Z 1

Q

h./ D Q

e2 i s h.s/dsI ./ D e2 i .s/ .s/ds; 1 < < 1:

0 0

Q and ./

Thus defined, we have that h./ Q are in L2 .1; 1/ with the properties

Z 1 Z 1

jh.s/j2 ds D Q

jh./j 2

d:

0 1

hQ D F h:

2.8 Beams of Infinite Length 45

Q g:

h; g D h; Q (2.58)

PaleyWiener theorem [51]:

Z 1 Q

jlogjjh.!/jjj

d!:

1 1 C !2

.Ax/ D mx;

Q

!

EI .2 i /4 0

D m; (2.59)

0 GJ .2 i /2

p !

p p EI .2v/4 0

A m D p : (2.60)

0 GJ .2v/2

0 I2

(2.61)

M l m 0

0 I2

MatrixExp t ; (2.62)

M 1 m 0

which then yields an isometric group for 1 < t < 1. This should help answer

the question of what happens as the beam length is infinite.

The wing model we use is perhaps the simplest. It represents an unswept wing so

that it is rectangular; we neglect wing camber (see [18] for a detailed wing-shape

description) and assume zero thickness.

46 2 Dynamics of Wing Structures

The use of functional analysis is now common in applied mathematics but has

yet to reach engineering and there is a disconnect here at present. Indeed, because of

software packages the level of analytical skill is actually decreasing! It is interesting

that the need for infinite-dimensional spaces is underscored by a recent publication,

Dynamics of Very High Dimensional Systems [76].

For the articulated FF case the derivation of the nodal conditions is novel in that it

is derived simply on the basis that the differential (stiffness) operator be nonnegative

definite, and not from physical principles. It also requires the inclusion of the nodal

values in the definition of the structure state. This technique was employed for the

first time in [54]. This may be considered as a generalization of the GolandLuke

model [77].

It should be noted that feedback controllers are all rate controllers well known in

classical control for stabilization.

For a detailed description of self-straining controllers, reference should be made

to [79] where the performance limitation due to operational amplifiers is included.

There are, of course, a great many nonlinear structures too numerous for

inclusion here even restricted to aircraft wings, neglecting thickness. Here we have

chosen two that relate closely to the Goland model. There is no general theory

for nonlinear models such as that for the linear case. The notion of elastic energy

does not seem adequate. We are unable to provide a time-domain solution at the

level of the Goland model. As we have noted, a generic feature here is that the

need to show that they have unique solutions seems not to bother the originators

who go on immediately to discretization of the model after the elaborate effort

for constructing the model; see, for example, [76]. Of course the problem here is

much less complicated than the viscous flow case (Chap. 7) and we only need the

linearized model (linearized about the steady state) for flutter analysis, as we show

in Chap. 6. Hence we resort to a perturbation technique leading to a Volterra integral

equation bootstrapping on the linear equation; see Chaps. 46.

All the beam models we consider have zero thickness. However, we can include

wing camber and we do so briefly.

Finally we consider the case where the beam length is allowed to be infinite and

this case is of mathematical interest in that there are no discrete modes any more, and

the Fourier transform theory and the notion of multipliers provide the appropriate

techniques to this case.

Chapter 3

The Air Flow Model/Boundary Fluid Structure

Interaction/The Aeroelastic Problem

3.1 Introduction

that we wish to solve. Having described the structure model, we turn to the air flow

model simplifying it to the most used case where we neglect viscosity and consider

nonviscous flow but more importantly assume that the entropy is constant. This

makes the flow vortex free so that the flow can be described in terms of the potential.

Our concern is again more the structure response in air flowaeroelasticityand

hence the fluidstructure boundary conditions play the dominant role in determining

the aerodynamic loading on the wing structure.

Starting with the three basic conservation laws, we derive the fundamental field

equation describing the air flow, The Eulerfull potential equation with the Kutta

Joukowsky boundary conditions. We present a complete statement of the aeroelastic

problem at the end of the chapter for nonviscous flow and nonlinear structure

models, including the simplification to Strip theory, the typical section theory.

We begin with the notation we use for the basic parameters necessary to describe the

flow generally throughout the book from now on. We also list the relevant physical

constants we need in the process:

Density

q Fluid velocity vector

q1 Far field velocity

U1 D jq1 j This is a free parameter, the far field air speed

a1 D Speed of sound

p Pressure

Viscosity

DOI 10.1007/978-1-4614-3609-6 3, Springer Science+Business Media, LLC 2012

48 3 The Air Flow Model/Boundary Fluid Structure Interaction

S Entropy

T Temperature

e Energy per unit volume

Subscript 1 denotes far field values

p D RT

R D cp cv

D cp =cv Ratio of specific heats

Thermodynamic Relation:

p es=cv D Const.

h enthalpy per unit mass D cp T

E internal energy per unit mass D cv T

All are functions of time t and the spatial coordinates x; y; z:

f D f .t; x; y; z/

The scalar functions are all positive.

Physical Constants:

All at standard air 59 F

0.00238 slug/ft3

1.23 kg=m3

0:372 106 slug=ftsec

17:8 106 kg=msec

p 2; 116; lb=ft2

1:0312 105 N=m2

a1 336 m=sec

Re Reynolds Number U 1= 106

R Gas Constant 287 kg=msecunits

cv 717

cp 1,004

1. 4

k Diffusivity D cp

17:8 103

Prandtl no. 0:7 (here taken as 1)

k cp

D D

R1 R1 1 1

3.3 Nonviscous Flow: The Euler Full Potential Equation 49

The field equation of fluid flow in 3D space (R3 , orthogonal coordinates x; y; z)

is derived from three basic laws of conservation which we state here in differential

form (as opposed to the integral form) with t representing the time co-ordinate.

We begin with the premordial.

Conservation Laws

@

C r .q/ D 0: (3.1)

@t

2. Conservation of momentum. The Euler momentum equation (nonviscous flow,

no heat conduction)

@q

C .q r/q C rp D 0 (3.2)

@t

in the usual notation [4, 12], where .q r/q is the vector:

Note that second spatial derivatives of q are not involved, unlike the viscous

case (see Chap. 7).

3. Conservation of energy: First law of thermodynamics in Eulerian form for perfect

gas [14]

1 p

ED ;

1

DE

C pr q D 0: (3.3)

Dt

Or

@E

C q rE C p rq D 0: (3.4)

@t

Note that we need three equations: conservation of mass, conservation of

momentum, and conservation of energy. All are expressed in differential form

rather than in integral form. This makes it possible to state the dynamic boundary

conditions crucial for aeroelasticity.

50 3 The Air Flow Model/Boundary Fluid Structure Interaction

This equation (3.4) implies [12, 14] that the total derivative of the entropy S is

zero:

DS

D 0; (3.5)

Dt

or the flow is homentropic [14].

Note that:

pT S

are thermodynamic state variables any one of which is determined by the other

three. Thus, under our assumption that the specific heats are constants, we have

[14]:

p D constant es=cv : (3.6)

Isentropic Flow

S D constant in t; x; y; z:

Homentropy does not imply isentropy. But we now assume that the flow is

isentropic. In particular then the energy equation is satisfied. And from (3.6) we have

the important conclusion that the pressure is a function of density. More specifically:

p D A ; (3.7)

p1 1 D A:

Next

rp A

D r

D A 1 rlog

D A 1 rlog

DA 1 rlog 1

1

DA r 1 : (3.8)

1

3.3 Nonviscous Flow: The Euler Full Potential Equation 51

Hence the Euler equation can be purged of the pressure variable p to yield:

@q

C .q r/q C A r 1 D 0: (3.9)

@t 1

This is what enables us to deduce that the flow is curl or vortex free. Thus let

D r q:

@q

r C r .q r/q D 0

@t

Hence

@

C r .q r/q D 0:

@t

But

1

.q r/q D r.q : q/ q : (3.10)

2

Hence finally:

@

C r .q / D 0: (3.11)

@t

Consistent with our assumption that the far field is q1 is that

q.0; x; y; z/ D q1 D r1

and hence

.0; x; y; z/ D 0:

Now for any (smooth enough) solution q.t; x; y; z/, we may consider ./ as a

solution of (3.9) which is a linear equation with zero initial condition and has the

identically zero solution

q D r:

And .t; x; y; z/ is the velocity potential such that the far field potential

p

.t; x; y; z/ where r D z2 C y 2 C x 2 ! 1;

D x.q1 i / C y.q1 j / C z.q1 k/

D 1 .0; x; y; z/:

52 3 The Air Flow Model/Boundary Fluid Structure Interaction

Assuming entropic flow, we are now ready to derive the field equation for the

velocity potential ./ using the continuity equation. First though we rewrite the

Euler equation using (3.7) and (3.8) as:

@r 1 A

C rr; r C r 1 D 0: (3.12)

@t 2 1

Hence

@ 1

C r; r C A 1 D constant D far field values

@t 2 1

1

D q1 ; q1 C A 1 :

2 1 1

Now because p is a function of ,

dp

D A 1 D a2 ;

d

where a is the local speed of sound with the far field value a1 , so that

2 1

a1 D A 1 ;

or

2

a1

A D 1

: (3.13)

1

Hence we have

!

@ 1 2 a2 1

C 2

U U1 D 1 1 1 ;

@t 2 1 1

where we use U for the flow speed. Hence we have for the density:

1 1 1 1 2 2

D 1 1 2

U U1 C @t : (3.14)

a1 2

@ 1 @

D . 1/ 2

@t @t

D . 1/ 2 r .r/

D . 1/ 2 . C r r/

D . 1/ 1 r r 1

3.3 Nonviscous Flow: The Euler Full Potential Equation 53

using

r r 1

. 1/ D 1 :

From (3.14) we have:

@ 1 1 @2 1

D 1 C r @tr/

2

@t @t 2 a1

1 1 1 2

r 1 r D 1 2

r r C @tr r:

a1 2

Hence

1 1 @2 @r

1 2

C r

a1 @t 2 @t

1 1 1 2 2

@

D . 1/1 1 U U C

2

a1 2 1 @t

1 1 1 2 @r

1 2

rjjrjj C r:

a1 2 @t

Hence finally:

@2 @r 2 1 1 2 2

@

C r D a1 1 U U 1 C

@t 2 @t 2

a1 2 @t

1 @r

r r.jjrjj2 / r;

2 @t

which we can rewrite in the form:

2

@2 @ 2 2 1 U1 jjrjj2 @

C jjrjj D a1 1 C 2

@t 2 @t a1 2 2 @t

jjrjj2

r r : (3.15)

2

This is the 3D Euler full potential equation valid except on the boundary specified

later.

We can now derive an explicit expression for the pressure in terms of the fluid

velocity. The acceleration is defined by the total derivative of the velocity:

54 3 The Air Flow Model/Boundary Fluid Structure Interaction

Dq @q

a.t/ D D C .q r/q

Dt @t

and for potential flow by (3.10)

1

.q r/q D r.q q/:

2

Hence

1

a.t/ D r @t C jrj2 D r .t/;

2

jrj2

.t/ D @t C (3.16)

2

and the far field value

2

U1

1 D :

2

The flow being isentropic

p D A

and by (3.14)

1=. 1/

1

D 1 1 . 1/ 2

(3.17)

a1

and hence =. 1/

. 1/

pD A1 1 . 1/ 2

: (3.18)

a1

And a very reasonable (and universal) approximation here is to take:

. 1/

p D A1 1 2

(3.19)

a1

consistent with

U1

0<M D < 1;

a1

where M is the Mach number, and that the perturbation of the flow by the airplane

is small compared to the far field speed. In any event we use (3.19) for p throughout

for isentropic flow.

The main relations we have are thus (3.19) and (3.15).

What distinguishes aeroelasticity from aerodynamics is of course the interaction

with the structure dynamics on the boundary, a singular boundary that complicates

the problem further.

3.3 Nonviscous Flow: The Euler Full Potential Equation 55

Boundary Conditions

As with any field equation, the conditions on the boundary determine the

solution. Here it is further complicated by the fact we have taken into account the

structure dynamics as well.

Flow-Structure Interaction:

Hence we first need to specify the structure model.

The Simplest Wing Structure Model

The simplest wing model is a slender thin plate whose thickness is then taken to

be zero, rectangular in shape (unswept wing) uniform, with ` denoting the half wing

span and 2b the width or chord length, so that b is the halfchord.

We only consider wings of high-aspect ratio:

` b;

which would justify the flexible model. However, we do consider a finite rectangular

plane (Finite plane) as the boundary for the aerodynamic equations. Thus we have

3D aerodynamics and a 2D wing boundary.

We choose the spatial co-ordinate system consistent with the aircraft rigid body

dynamics. Thus the x-axis is along the airplane axis, the y-axis is the span axis and

the negative z-axis is the plunge axis. Thus the boundary for the field equation is

the rectangle in the xy-plane described by:

D fb x bI 0 y `g:

For the structure dynamics, however, we specialize to a beam model, ignoring the

dependence on the chord variable. Such a model was described by Goland, as we

have seen in Chap. 2 where the structure is endowed with two degrees of freedom:

plunge and pitch.

The plunge displacement denoted h .t; y/ is then the displacement of the wing

along the negative z-axis. It is uniform along the chord. The pitch #.t; y/ is the twist

angle in radians about an axis parallel to the y-axis at distance ab from the x-axis

and again does not depend on the chord variable x. Hence h.t; y/ and #.t; y/ are

defined on:

0 tI 0 y `:

The wing is fixed to the fusilage. Thus we have a clampedfree (CF) model with:

56 3 The Air Flow Model/Boundary Fluid Structure Interaction

force/momentum balance equations.

The resulting structure dynamic equations are described in Chap. 2. Here we shall

describe the airwing interaction dynamics.

Thus we have two sets of conditions.

The normal velocity of the air along the wing is equal to the normal velocity of

the wing. We may need to distinguish between the top and bottom of the wing if

we allow for discontinuity in the fluid velocity across the wing, even though the

thickness is zero.

The total displacement of the wing is given by:

!

k z.t; x; y/ where z.t; x; y/ D .h.t; y/ C .x ab/ # .t; y//

!

Dz.t/ ! @z.t/

k D k C q.t; x; y; 0/ rz.t/

D.t/ @t

P

! P y/ C .q.t; x; y; 0/ !

D k h.t; y/ C .x a/#.t; i /#.t; y/

! @

C q.t; x; y; 0/ j .h.t; y/ C .x a/#.t; y// :

@y

Hence allowing for discontinuity in the flow, the flow tangency conditions become:

@.t; x; y; 0C/ @1 P y/ C .x a/#.t;

P y/

D C .1/ h.t;

@z @z

!

!

C .q.t; x; y; 0C/ i /#.t; y/ C .q.t; x; y; 0C/ j /

@

.h.t; y/ C .x a/#.t; y// ; x; y

; (3.20)

@y

3.3 Nonviscous Flow: The Euler Full Potential Equation 57

D P y/ C .x a/#.t;

C .1/ h.t; P y/.q.t; x; y; 0/ !

i/

@z @z

! @

#.t; y/C .q.t; x; y; 0/ j / .h.t; y/C.xa/#.t; y// ;

@y

x; y

: (3.21)

These conditions are peculiar to aeroelasticity and are given in terms of the pressure

which is discontinuous across the wing. We define the pressure Jump by p:

Defining

.t; x; y/ D .t; x; y; 0C/ .t; x; y; 0/;

we have from (3.20) that

p D A1 D 1 : (3.23)

a21

Or

.t; x; y/ D 0 for x; y not in : (3.25)

Or

.t; x; y/ D 0 as x ! b; x; y in : (3.27)

Structure Dynamics

Chap. 2 by including the aerodynamic lift and moment which are expressed in

terms of the structure state variables.

58 3 The Air Flow Model/Boundary Fluid Structure Interaction

Goland Model

Z b

R y/ C S #.t;

mh.t; R y/ C EI h0000 .t; y/ D L.t; y/ D p.t; x; y/dx;

b

Z b

R y/ C S h.t;

I# #.t; R y/ GJ # 00 .t; y/ D M.t; y/ D .x ab/p.t; x; y/dx

b

DowellHodges Model

R y/ C EI 1 h0000 .t; y/ C .EI 2 EI 1 /.#.t; y/v.t; y/00 /00 D mg sin' C L.t; y/;

mh.t;

mRv.t; y/ C EI 2 v0000 .t; y/ C .EI 2 EI 1 /.#.t; y/h.t; y/00 /00 D mgcos ';

R y/ GJ # 00 .t; y/ C .EI 2 EI 1 /h.t; y/00 v.t; y/00 D M.t; y/;

I# #.t;

0 < tI 0 < y < `:

Boundary

by the following equations.

Field equation

2

@2 2 2 1 U1 jjrjj2

C @t jjrjj D a1 1 C @t

@t 2 a12 2 2

jjrjj2

r r : (3.30)

2

3.4 Problem Statement 59

Far field

Boundary conditions

@.t; x; y; 0C/ @1 h

D P y/ C .x a/#.t;

C .1/ h.t; P y/

@z @z

!

!

C .q.t; x; y; 0C/ i /#.t; y/ C .q.t; x; y; 0C/ j /

@

.h.t; y/ C .x a/#.t; y// ; x; y

: (3.31)

@y

And

D P y/ C .x a/#.t;

C .1/ h.t; P y/

@z @z

!

!

.q.t; x; y; 0/ i /#.t; y/ .q.t; x; y; 0/ j /

@

.h.t; y/ C .x a/#.t; y// ; x; y

: (3.32)

@y

KuttaJoukowski conditions

@ jjrjj2

.t/ D C ; (3.33)

@t 2

.t; x; y/ D 0 for x; y not in ; (3.34)

.t; x; y/ D 0 as x ! b; x; y in : (3.35)

Z b

R y/ C S #.t;

h.t; R y/ C EI h0000 .t; y/ D 1 .t; x; y/ dx; 0 < y < `;

b

(3.36)

60 3 The Air Flow Model/Boundary Fluid Structure Interaction

Z b

R y/ C S h.t;

I# #.t; R y/ GJ # 00 .t; y/ D 1 .x ab/ .t; x; y/dx 0 < y < `;

b

(3.37)

h000 .t; `/ D 0 D h00 .t; `/ D 0I # 0 .t; `/ D 0;

or FF end conditions:

h000 .t; `/ D 0 D h00 .t; 1/ D 0I # 0 .t; `/ D 0:

Beran Straganac

The state variables are the same as in the Goland. For the nonlinearities see Chap. 2,

Sect. 2.7. The end conditions are also the same as in the Goland model.

DowellHodges

mh.t;

Z b

D mg sin' 1 .t; x; y/dx; 0 < y < `; (3.38)

b

mRv.t; y/ C EI 2 v0000 .t; y/ C .EI 2 EI 1 /.#.t; y/h.t; y/00 /00 D mg cos'; (3.39)

R y/ GJ # 00 .t; y/ C .EI 2 EI 1 /h.t; y/00 v.t; y/00

I# #.t;

Z b

D 1 .x ab/ .t; x; y/dx0 < tI 0 < y < `: (3.40)

b

The end conditions for v.t; :/ are the same as for h.t; :/.

3.4 Problem Statement 61

ordinate in consequence of the assumed high-aspect ratio `=b of the wing.

Problem Statement

2D Aerodynamics 1D Structure

2 2 !

@2 @ @ @

C C

@t 2 @t @x @z

0 0 2 2 11

@ @

B C

2 B 1 B

BU

2

@x @z @ C

CC

C

D a1 B1 C 2 B 1 CC

@ a1 @ 2 2 @t AA

0 2 2 1

@ @

B C C

B @x @z C

r r B C 1 < x; z < 1: (3.41)

@ 2 A

Boundary conditions

is now just the Chord: b < x < b and the flow tangency condition becomes:

!

Total Displacement D k z.t; x/ where

62 3 The Air Flow Model/Boundary Fluid Structure Interaction

Dz.t/ !

! @z.t/

k D k C q.t; x; y; 0/ rz.t/ ;

D.t/ @t

P

!

!

D k h .t; y/ C .x ab/#P .t; y/ C q .t; x; y; 0/ i #.t; y/ :

Hence allowing for discontinuity in the flow, the flow tangency conditions become:

D C.1/ h.t;P y/C.x a/#.t;

P y/

@z @z

! i

C q.t; x; 0C/ i #.t; y/ : (3.42)

And

D C .1/ h.t;P y/ C .x a/#.t;

P y/

@z @z

! i

C q.t; x; 0/ i #.t; y/ x

: (3.43)

KuttaJoukowski Conditions

.t; x/ D .t; x; 0C/ .t; x; 0/;

p D 1 :

Or

Kutta condition

p.t; x/ ! 0 as x ! b; in:

Or,

The structure equations with the indicated aerodynamic loading remain the same.

3.4 Problem Statement 63

This is then the precise statement of the aeroelastic problem continuum equa-

tions. The objective is to determine the stability of the structure state as a function

of U1 , the air speed.

To anticipate the theory that follows, the main conclusion is that for a given value

of M (equivalently, speed of sound, equivalently altitude) there is a speed, called

flutter speed, denoted UF , for U1 < UF , the structure is stable (see later for precise

definition of stability), and for U > UF the structure is unstable.

et al. [17], Hodges and Pierce [5] or Bisplinghof, Ashley, and Halfman [6] care

to make a precise statement of the aeroelastic problem as we do in this chapter.

Indeed without such a statement it is not clear what it is that the computer codes

used universally (see the many recent papers on aeroelasticity, for example [75, 81

83, 93]) are providing the (approximate) solution to, even omitting the cases where

the solvability of the problem cannot be established.

Indeed without such a formulation it is not possible to define the Flutter Speed

calculating which is a main objective of the theory.

We should note that most progress has been made for the typical section case (2D

aerodynamics) and it is fortunate that flexibility is consistent with high-aspect ratio

so that the typical section approximation is reasonable (without necessarily being

very high).

Regarding the foundational conservation laws, following [4,12] we have invoked

three of them rather than the first two as in [4,17], for example. The triad is essential

for aeroelasticity. As Meyer [14] notes the third is the Euler version of the first law

of thermodynamics.

Chapter 4

The Steady-State (Static) Solution

of the Aeroelastic Equation

4.1 Introduction

equations of Chap. 3 where we set all the time derivatives to zero and there is no

input. It is called the static solution in that there is no change with time. It is of

interest on its ownit is in fact central to the study of stabilitybut it also serves

to illustrate the solution techniques used for the general case in Chap. 6.

Often this solution is also referred to as the steady-state solution, and so we need

to clarify the terminology. The term steady-state response is used to indicate the

response of a linear system to a steady input, such as a sinusoid input which is then

not dependent on the initial condition. Here we have the case of zero input. So we

take an arbitrary initial condition (and far field speed) and let time march (literally

in computational programs) until there is no longer any change with time; in other

words we consider the asymptotic response in time. Physically we are considering

a pointwise limit of the potential as well as the structure state. Of course depending

on the assumed speed parameter and the initial conditions there may not be such a

solution. We largely follow [67]. To begin with, we can state the following.

Theorem 4.1. Every steady-state solution is a static solution.

Proof. If a steady-state limit exists, and there is no change in time, then it must

satisfy the static equation. t

u

In our case in addition to the initial condition we have a parameter to specify the

far field speed. And we show that for a sequence of speeds there is indeed more than

one solution to the static aeroelastic equation. However, as we show below, there

is one static solution for all speeds which is also a steady-state solution as well.

This provides us a simple example where the continuum equations need not have a

unique solution depending on the far field speed.

The steady-state solution depends obviously on the structure model used.

We begin with the workhorse model, the linear Goland model.

DOI 10.1007/978-1-4614-3609-6 4, Springer Science+Business Media, LLC 2012

66 4 The Steady-State (Static) Solution of the Aeroelastic Equation

We first consider the Goland linear structure model. We set all time derivatives in

the field and boundary equations to zero. And all functions are independent of time

so we drop the time variable. Thus we begin with the velocity potential .x; y; z/

which satisfies the 3D time-invariant Euler full potential equation, purging the time

derivatives in (3.15), yielding now:

1 1

a1 C

2

.jr1 j jrj / r r.jrj2 / D 0;

2 2

2 2

" 2 2 2 !#

. 1/ @ @ @

0D 2

a11C 2

U1

2

.4/

2a1 @x @y @z

1 @ @ 2 @ @ 2 @ @ 2

r C r C r : (4.1)

2 @x @x @y @y @z @z

Angle of Attack

We assume that the far field velocity vector is in the XZ plane making an angle

with the X -axis. This angle in flight control rigid body model [9] is called the angle

of attack. Hence

1 D U1 .x cos C z sin /;

q1 D U1 .i cos C k sin /:

@4 h.y/

EI D L.y/; (4.2)

@y 4

@2 .y/

GJ D M.y/ (4.3)

@y 2

4.2 Goland Structure Model 67

00

h .1/ D 0I h000 .1/ D 0I 0 .1/ D 0:

I. Flow Tangency Condition

@

.x; y; 0C/ D U1 sin .q.x; y; 0C/ i /.y/ .q.x; y; 0C/ j /.h0 .y/

@z

C.x ab/ 0 .y//; (4.4)

@

.x; y; 0/ D U1 sin .q.x; y; 0/ i /.y/ .q.x; y; 0/ j /.h0 .y/

@z

C.x ab/ `0 .y//: (4.5)

equilibrium solution:

.x; y; z/ D 1 .x; y; z/:

Structure at rest:

for all far field speeds.

In this case the fluid velocity and pressure are continuous across the wing.

Proof. Verified by direct substitution into the equations. The solution holds for

every value of the far field speed. u

t

We now show the existence of a nonzero static solution.

To find the nonzero static solution we need to solve the nonlinear static Euler

equation. We show that because of the nature of the boundary conditions, we can

obtain a series expansion, a solution technique that works also in the dynamic case,

but is of course less complex here.

68 4 The Steady-State (Static) Solution of the Aeroelastic Equation

In the structure equations we need to calculate L.y/ and M.y/ for each

y; 0 < y < `. If we fix y, however, the structure variables h.y/ and .y/ may

be looked upon as parameters which enter linearly in fact in (4.4) and (4.5) and

we may postulate that the solution is analytic in these parameters, at least locally.

The solution in other words can be expanded in a power series (for some nonzero

radius of convergence) about

h.y/ D 0I .y/ D 0:

This is conveniently done by introducing the complex number parameter and

considering the response to

h.y/; .y/

for given h.y/; .y/. More specifically, let the solution be denoted .; x; y; z/

corresponding to h.y/; .y/, so that

1

X k

.; x; y; z/ D 1 .x; y; z/ C k .x; y; z/; (4.7)

k

kD1

where

@k

k .x; y; z/ D .0; x; y; z/: (4.8)

@k

Before discussing the sense in which the series converges, let us first see how we

can calculate the derivatives for each point .x; y; z/. We go back to the field equation

(4.1) and differentiating it once with respect to and setting D 0, we obtain for

the first term therein:

2

a1 1

@2 1 2

2 @ 1 @2

U1

2

cos2 U 2

1 sin 2U 2

1 cos sin :

@x 2 @z2 @z@x

.1 / D 0;

where

@2 @2 @2

./ D .1 M 2 cos2 / C C .1 M 2

sin 2

/

@x 2 @y 2 @z2

@2

2M 2 sin cosff ; (4.9)

@x@z

4.2 Goland Structure Model 69

U1

M D < 1:

a1

scribed presently. Note that the thermodynamic constant does not appear.

For the higher-order potentials k we need to work a little harder. Let us fix the

spatial coordinates and let:

1 2

f ./ D a1 2

C .U1 jr./j2 / ; (4.10)

2

g./ D ./: (4.11)

f ./g./

1

X k

g./ D 4k :

k

kD1

Hence

1

X " X 1

!

k 1 k

f ./g./ D 2

a1 4k C U1 q1 C

2

rk

k 2 k

kD1 kD1

1

!# 1

!

X k X k

q1 C rk 4k

k k

kD1 kD1

1 1

!" 1

X k X k X k

D a1

2

4k C 4k .1 / q1 qk

k k k

kD1 kD1 kD1

1

! 1

!#

1 X k X k

C qk qk

2 k k

kD1 kD1

0 1

X 1 k X1 X 1 j k

D a1

2

4k C .1 / @ 4j q1 qk A

k j D1

j k

kD1 kD1

1 1 1

1 X X X i j k

C qi qj 4k ; (4.12)

2 i D1 j D1 i j k

kD1

70 4 The Steady-State (Static) Solution of the Aeroelastic Equation

qk D rk :

Next the second term in (4.1) is best handled using x1 ; x2 ; x3 , respectively, in place

of x; y; z. Thus it can be expressed:

X 3

@ X @ @2

3

i D1

@xi j D1 @xj @xi @xj

X

3 X 1 X

3 X 1 X

1

1 @n @m @2 p

D .nCmCp/ ;

i D1 j D1 nD0 mD0 pD1

nmp @xi @xj @xi @xj

where

0 D 1

so that

@k 0

D 0 k > 1:

@x k

Hence we have the expansion for (4.1):

0 1

X1 k X1 X 1 j k

2

a1 4k C .1 / @ 4j q1 qk A

k j D1

j k

kD1 kD1

1 1 1

1 X X X i j k

C qi qj 4k

2 i D1 j D1 i j k

kD1

X

3 X 1 X

3 X 1 X

1

1 @n @m @2 p

.nCmCp/ D 0; (4.13)

i D1 j D1 nD0 mD0 pD1

nmp @xi @xj @xi @xj

X

3 X 1 X

3 X 1 X

1

1 @n @m @2 p

.nCmCp/

i D1 j D1 nD0 mD0 pD1

nmp @xi @xj @xi @xj

X 3 X 3 X 1

1 p @0 @0 @2 p

D

i D1 j D1 pD1

p @xi @xj @xi @xj

X

3 X 1

3 X

1 mCp @0 @m @2 p

2

i D1 j D1 mD1

mp @xi @xj @xi @xj

X

3 X 1 X

3 X 1 X

1

1 @n @m @2 p

.nCmCp/ :

i D1 j D1 nD1 mD1 pD1

nmp @xi @xj @xi @xj

4.2 Goland Structure Model 71

@h @

C @y .; x; y; 0C/ C .x ab/ :

@y @y

Hence

@1 .x; y; 0C/ @1 .x; y; 0/

D U1 cos .y/ D (4.14)

@z @z

and for k 2:

@k .x; y; 0C/ @k1 .x; y; 0C/

D k .y/ C .@k1 .x; y; 0C//

@z @x

. @h @

@y C .x ab/ : (4.15)

@y @y

KuttaJoukowsky Condition

0 1

1 X

X 1 1

X

1@ 2 kCj k

D U1 C rk rj C 2 r1 rk A : (4.16)

2 kj k

kD1 j D1 kD1

2 X k 1

U1

.; x; y; z/ D C k .x; y; z/;

2 k

kD1

And hence

using (4.15) and (4.16), and as we have noted D 0 off the wing.

We pause here at this point to examine in detail the linear case:

k D 1:

72 4 The Steady-State (Static) Solution of the Aeroelastic Equation

Linear theory has an importance all its own and in addition we actually bootstrap

about it for the nonlinear case. And we can carry the solution some distance for

the general 3D aerodynamics. We do not discretize it now (as in most work in

aeroelasticity, as we have noted already). The field equation is, as we have seen:

.1 / D 0:

We rewrite this as an equation in z because z D 0 contains the boundary.

@2 1 @2 1

.1 M 2 sin2 / 2M 2

sin cos

@z2 @x@z

@2 1 @2 1

D .1 M 2 cos2 / : (4.19)

@x 2 @y 2

The first step is to choose the function space for the solution.

For each z we seek a solution in

Lp R2 ; 1 < p < 2:

Z 1 Z 1

O1 .z; !1 ; !2 / D 1 .x; y; z/e.i!1 Ci !2 / dxdy !1 ; !2 2 R2 (4.20)

1 1

and note that the Fourier transform of the right side of (4.21) yields

@2 O1 @2 O 1

.1 M 2 sin2 / C 2M 2

sin cos

@z2 @x@z

D .!12 .1 M 2 cos2 / C !22 /O1 .z; !1 ; !2 /: (4.21)

Let

@ O

vO .i !1 ; i !2 / D 1 .0; i !1 ; i !2 /:

@z

Then we can show that the only solution which goes to zero (in the Lp norm) as

jzj ! 1 is given by

1

O1 .z; !1 ; !2 / D ezr1 vO .i !1 ; i !2 / for z > 0 (4.22)

r1

1 zr2

D e vO .i !1 ; i !2 / for z < 0; (4.23)

r2

4.3 Linear Aeroelasticity Theory: The Finite Plane Case 73

where

p 2

r1 D .M 2 .i !1 /sin cos .!2 .1 M 2 sin2 /

C !12 .1 M 2 ///=.1 M 2 Sin2 /; (4.24)

p 2

r2 D .M 2 .i !1 /sin cos C .!2 .1 M 2 sin2 /

C !12 .1 M 2 ///=.1 M 2 sin2 / D r1 : (4.25)

p D 1 :

Let

1

A1 .x; y/ D (4.26)

U1

D cos 1 :

Z b Z `

AO1 .i !1 ; i !2 / D eix!1 iy!2 A.x; y/dxdy

b 0

1 1

D i !1 cos vO 1 .i !1 ; i !2 / ;

r1 r2

or,

O !1 ; i !2 /

1 A.i

vO 1 .i !1 ; i !2 / D .!12 .1 M 2 cos2 /

2 i !1 cos

p

C !22 /=. .!12 .1 M 2 / C !22 //

and by (4.15)

O 2 /;

D U1 cos .i! (4.27)

74 4 The Steady-State (Static) Solution of the Aeroelastic Equation

where

Z `

O

.i!/ D `.y/eiy! dy:

0

O !1 ; i !2 / D U1 cos O .i !2 /;

PO .i !1 ; i !2 /A.i

where

1 1 h

PO .i !1 ; i !2 / D !12 .1 M 2 cos2

2 i !1 cos

. i

p 2

C !22 .!1 .1 M 2 C !22 (4.28)

Z 1 Z 1

PO .i !1 ; i !2 / D P .x; y/eix!1 iy!2 dxdy:

1 1

This is the static version of the 2D integral equation of Possio which we describe

later and which plays a crucial role in our theory.

Z b Z `

P .x

; y /A.

; /d

d D U1 .y/; 0 < y < 1; jxj < b; (4.29)

b 0

Let us assume that this equation has a solution in Lp b; b. Then

Z `

A.x; y/ D L.x; y; /./d; x; y on the wing (4.32)

0

and hence, getting back to the structure equations (4.2) and (4.3), we have:

Z b Z `

GJ 00 .y/ D .x ab/ L.x; y; /./ddx

b 0

Z `

D M.y; /./d: (4.33)

0

4.3 Linear Aeroelasticity Theory: The Finite Plane Case 75

And with the CF end conditions we have thus a linear eigenvalue problem that would

have a solution at most for a sequence of values of the far field speeds which we may

call divergence speeds. But not being able to calculate the kernel M (., .), we cannot

do much more. So we proceed to simplify the problem where we can say more.

The first and most important simplification is to consider the following.

We assume that the aspect ratio `=b is high enough so we may consider the air flow

on the wing to be uniform so that there is no dependence on the span variable y,

also called the strip theory. In this case the linear field equation (4.20) simplifies to:

@2 1 @2 1 @2 1

.1 M 2 sin2 / 2

2M 2 sin cos D .1 M 2 cos2 / 2 (4.34)

@z @x@z @x

and we define, modifying (4.21)

Z 1

O 1 .z; i !/ D 1 .z; x/eix! dx

1

O !/ D 1 .0; i !/:

.i

1

@1 .0; x/ D U1 cos .y/; jxj < b;

@z

where y is fixed.

We want

p.x/ D 1 .x/

and

We define:

1

A1 .x/ D ; b < x < b

U1

so that

p D 1 U1 A1 :

We now turn to Fourier transforms which are obtained by simply setting in the

previous expressions:

!1 D !I !2 D 0:

76 4 The Steady-State (Static) Solution of the Aeroelastic Equation

Then

1 1

AO1 .i !/ D i ! O !/:

.i

r1 r2

Hence

1 A.iO !/ h .

p

i

O !/ D

.i ! 2 .1 M 2 cos2 1 M 2 !2 :

2 i ! cos

Let

h . i

1 1 p

PO .i !/ D ! 2 1 M 2 cos2 1 M 2 !2 I

2 i ! cos

Z 1

PO .i !/ D P .x/eix! dx; 1 < x < 1

1

and we have:

Z b

P .x

/A.

/d D U1 .y/; jxj < b

b

A.b/ D 0: (4.35)

We should note that (4.36) is a special case of the Possio integral equation which

we go into in far more detail in Chap. 5, where it is now the airfoil equation or the

finite Hilbert transform equation. PO .i !/ can be expressed as

1 j!j

PO .i !/ D const ; (4.36)

2 i!

where

1 M 2 cos2

const D p sec

1 M2

and

j!j

i!

4.3 Linear Aeroelasticity Theory: The Finite Plane Case 77

distribution along chord

4

Z b

1 1

const A.s/ds D U .y/; A.b/ D 0; jxj < b: (4.37)

2 b xs

And we have the explicit solution, due to Tricomi (see [11], and the more general

treatment in [31]).

s s

Z

1 1 bx b

bC

1

A.x/ D U1 .y/ d

; jxj < b:

Const
bCx b b

x

In as much as

Z s

1 b

bC

1

d

D b;

b b

x

s p

b x 2 1 M2

A.x/ D bU 1 .y/ cos ; jxj < b: (4.38)

b C x 1 M 2 cos2

Note that A.b/ D 0. The function A.:/ is in Lp b; b for 1 < p < 2 but Not for

p D 2. In other words this is Not an L2 theory, as opposed to structure dynamics,

where it is and we have a notion of energy.

See Fig. 4.1 for the canonical static pressure distribution along the chord for

b D 1.

Note the discontinuity with respect to the angle of attack. For D 0, the pressure

distribution at midchord increases without bound as M increases to 1. But for small

nonzero , the pressure peaks at some point and then decreases to zero. We have a

78 4 The Steady-State (Static) Solution of the Aeroelastic Equation

distribution at midchord

D 0 and D 1 degree 70

60

50

40

30

transonic peak. Here the word transonic simply means that M is close to 1. See

Fig. 4.2.

Structure Response: Divergence Speed

Because the pressure jump depends only on the pitch angle the structure response

equations reduce to

h p i

GJ 00 .y/ DU1

2

.y/ 2 1 M 2 = 1 M 2 cos2

s

Z b

bx

1 cos .x ab/ dx.0/ D 0 0 .1/ D 0; (4.39)

b b Cx

s r

Z b Z 1

bx 1x

2 .x ab/ dx D 2b 2

.x a/ dx D b 2 .1 C 2a/ ;

b b C x 1 1 C x

1

jaj < :

2

The structure equation for the torsion angle .y/,

Z b

GJ 00 .y/ D .x ab/p.x/dx;

b

becomes .y/ D .y/; 0 < y < 1, with .0/ D 0 .0/ D 0 and has the

00 2

must have

B cos` D 0:

Thus B D 0 is always a solution, whatever U1 .

If we want a nonzero solutionB is not zerowe have an eigenvalue problem

and thus we have a nonzero solution only for

4.3 Linear Aeroelasticity Theory: The Finite Plane Case 79

D .2n 1/ ; n1

2`

and

h
y i

.y/ D B sin .2n 1/ 0 < y < `;

2`

where:

.

2 D U1

2

1 cos .1 M 2 /1=2 .1 M 2 cos2 /b 2 .1 C 2a/;

. p 1=2

U1 D .2n 1/ .1 M 2 cos2 / 1 M2

1 p .

p

.
GJ / . 1 .1 C 2a// sec ;

2b1

n positive integer.

The smallest value is called the divergence Speed:

.p 1=2

Ud D 1 M 2 cos2 1 M2

1 p .

p

GJ . 1 .1 C 2a// sec : (4.40)

2b1

It is interesting to see the dependence on M , and the difference between

D0 and 0:

For D 0, we have

1 p .

p

Ud D .1 M 2 /1=4
GJ . 1 .1 C 2a// (4.41)

2b`

and Ud decreases monotonically to zero.

On the other hand for nonzero we have a nonzero minimum and Ud increases

to infinity at M D 1. We have thus what is referred to as the transonic dip which

occurs at

p

M D .1 tan2 /;

Ud =Ud .0/:

computation; see [99] where it has to be extrapolated from literally a few points

on the graph.

80 4 The Steady-State (Static) Solution of the Aeroelastic Equation

D 1o

0.30

0.29

0.28

s

p . Z

bx b

D 1 Ud2 2.y/ 1 M2 .1 M 2 cos2 /

b b Cx

p . h y i

D 2 b 1 M 2 .1 M 2 cos2 / 1 Ud2 sin ; 0 < y < `;

2`

a nonhomogeneous equation with the end conditions:

Getting back now to the general case, let D denote the derivative with respect to .

Then with f ./ given by (4.11) and g./ by (4.12), we have:

X

k1

D k .f ./g.// D f ./D k g./ C g./D k f ./ C Cjk D kj g./D j f ./;

j D1

1 X k X

k1 j

@j m @m

k C cj kj Cmj ;

2 j D1 mD0

@xi @xi

4.4 The General Nonlinear Case 81

Next in (4.14), the coefficient of k is given by

X

3 X X

k2 X 3 X

@2 k1 X 1 @1m @m

3 3 1

@2 k @0 @0

c ;

j D1 i D1

@xi @xj @xi @xj i D1 j D1

@xi @xj mD0 m @xi @xj

lD1

(4.42)

Hence we obtain the nonhomogeneous linear equation for k :

where

1 XX k X

3 k1 j

@j m @m

gk1 D C kj Cmj

2 i D1 j D1 j mD1

@xi @xi

X

k2 X 3 X 3

@2 k1 X 1 @1m @m

1

C C

i D1 j D1

@xi @xj mD0 m @xi @xj

lD1

X

3 X

3 X

k1

@0 @m @2 km

C2 Cmk

i D1 j D1 mD1

@xi @xi @xi @xj

X

k1

C .1 / j q1 qkj

j D1

g0 D 0:

And in particular

X 3 X 3

@0 @1 @2 1

g1 D 2.1 /1 q1 q1 C 4 ;

i D1 j D1

@xi @xi @xi @xj

illustrating the nature of the functions gk . Note that the derivatives are well defined

for any k.

82 4 The Steady-State (Static) Solution of the Aeroelastic Equation

Next we need to go deeper into the boundary conditions. We begin with the typical

section case where the flow is 2D and brings considerable simplification well

justified for high-aspect ratio wings and is the best developed part of the theory.

We specialize to typical section theory, but keep a nonzero angle of attack.

D C1k D k1 .y/

@z @x

@k1 .x; 0C/

D k .y/ (4.44)

@x

D C1k D k1 .y/

@z @x

@k1 .x; 0/

D k .y/: (4.45)

@x

Flow Decomposition

To solve (4.43), we decompose the flow field as the sum of two interdependent

terms: for each k

k .x; z/ D kL .x; z/ C k0 .x; z/; (4.46)

.kL / D 0

subject to the boundary conditions (4.44) and (4.45) and k0 satisfies the

nonhomogeneous equation

.k0 / D gk1

4.4 The General Nonlinear Case 83

We define

1

X k

0 D 0 C k0 ; (4.48)

k

kD2

1

X k

L D kL (4.49)

k

kD1

so that

D 0 C L ; (4.50)

where 0 has no discontinuities on z D 0 and hence produces no lift, and L

produces all the lift.

Our main concern is of course the pressure jump across the wing, or equivalently

the acceleration potential jump. From (4.7)

2 2 !

@ @

D C D 0 z D 0; jxj > b:

@x @z

By analyticity therefore

Again our main concern is to evaluate this on the structure. We begin with

Theorem 4.3. Let ' denote the disturbance potential

' D 0 (4.51)

Then

@k @k

k D U1 cos C U1 sin ; b < x < b (4.52)

@x @z

Proof. We calculate that:

jrj2 D U1

2

C 2r0 r' C jr'j2 : (4.53)

Hence

1

D jrj2 D r0 r':

2

84 4 The Steady-State (Static) Solution of the Aeroelastic Equation

Hence

@k @k

k D r0 rk D U1 cos C sin : t

u

@x @z

Note that ' is the disturbance potential and we expect that jr'j2 is small

2

compared to U1 so that we may neglect it in (4.53) anyway. The angle of attack

is usually less than a degree and can be neglected in the first look, simplifying the

analysis. And sensitivity to the angle of attack is analyzed subsequently.

Let us now specialize to the case where the angle of attack is zero.

Then we have the following remarkable result.

Theorem 4.4. Suppose D 0. Then

Proof. First we note that

1

X k

r' D rk :

k

kD1

From

1 .x; z/ D 1 .x; z/

we have:

1 .x; 0C/ C 1 .x; 0/ D 0:

Hence

@1 .x; 0C/ @1 .x; 0/

C D 0:

@x @x

Hence

2

@1

D 0:

@x

Also from

@1

D0

@z

4.4 The General Nonlinear Case 85

we have

2

@1

D 0:

@z

Hence:

jr1 j2 D 0: t

u

Lemma 4.5. Suppose

jrj j2 D 0; j k 1:

Then

jrk j2 D 0:

@k @k1

D k.y/ jxj < b

@z @x

and hence

2 2

@k @k1

D k 2 .y/2 jxj < b:

@z @x

Hence

2

@k

D0 jxj < b:

@z

@j

j D

@x

along with

@j

j for

@z

and the Fourier transforms by Oj ; O j , respectively. We are only concerned with j L .

t

u

Following the proof for j D 1, we take Fourier transforms and obtain

i!

Oj .i !; 0C/ D O j .i !; 0C/; (4.55)

r1

i!

Oj .i !; 0/ D O j .i !; 0/; (4.56)

r1

86 4 The Steady-State (Static) Solution of the Aeroelastic Equation

where

p

r1 D j!j 1 M2 :

r1

O j .i !; 0C/ C O j .i !; 0/ D O j (4.57)

i!

and adding, we get

r1

.i !/ D .O j .i !; 0C/ O j .i !; 0//; (4.58)

i! j

where we use the notation j D j .x; 0C/ C j .x; 0/; the super bar does Not

denote complex conjugate.

Next we use the boundary condition (4.44) and (4.45)

By Theorem 4.2,

k .x/ D Uk .x/; jxj < b

and hence

k .x/

Ak .x/ D D k .x/; jxj < b:

U

Now

j .x; 0C/ C j .x; 0/ D j.y/ j 1 .x/; jxj < b:

But

r1

O j .i !; 0C/ C O j .i !; 0/ D Oj .i !/:

i!

Hence

r1 O

Aj .!/ D j.y/O j 1 .i !/: (4.59)

i!

Next

D j.y/Aj 1 .x/; jxj < b:

4.4 The General Nonlinear Case 87

and

i!

j .i !/ D .O j .i !; 0C/ O j .i !; 0//;

r1

hence

i! O

Aj 1 .i !/j.y/ D j .i !/;

r1

or

j!j

p AOj 1 .i !/j.y/ D j .i !/: (4.60)

i! 1 M 2

In (4.60) and (4.59) we have a pair of Possio equations for the time-invariant case,

similar to (4.38). This time it is convenient to define the Tricomi transform:

s r

Z

1 bx b

b C s f .s/

T f D hP; h.x/ D ds; jxj < b;

bCx b bs sx

p

1 M 2 Aj D j.y/ T P j 1

j!j

p AOj 2 .i !/.j 1/.y/ D O j 1 .i !/:

i! 1 M 2

Hence

p

Aj 2 .j 1/.y/ D 1 M 2 T P j 1 ;

1 r i! p

p D 1 M 2:

1 M 2 i! r

Hence

p

j.y/Aj 2 .j 1/.y/ D j.y/ 1 M 2 T P j 1 D .1 M 2 /Aj :

88 4 The Steady-State (Static) Solution of the Aeroelastic Equation

Aj D 0 for j even

and hence

j D 0 for j odd:

Ak

D .y/k1 .1 M 2 /..k1/=2/ A1

k

and

A2k D 0:

Hence

1

X 1

X

Ak 1

AD D .y/2k2 .1 M 2 /..2k1/=2/ A1 D .y/2

A1 : (4.62)

k 1C

kD1 kD1 1M 2

It is important to note that the series converges for small enough .y/, but the

closed-form solution holds without any smallness assumption on .y/. Also the

convergence is uniform in x and z.

We can now consider the convergence of the power series expansion:

1

X 2kL .x; z/

1

X .

C ..2kC1/L .x; z// .2k C 1/ (4.63)

.2k/

kD0 kD0

1 O .k 1/

kL .i !; z/ D ejzjr1 .y/k1 .1 M 2 /..k2/=2/ AO1 .i !/:

k k

kL .x; z/ 1

D 1 .x; z/ .k 1/.y/k1 .1 M 2 /..k2/=2/ (4.64)

k k

and we see that the convergence of the first series in (4.57) is uniform in x and z.

For k odd similarly we obtain using (4.61):

4.4 The General Nonlinear Case 89

Again we see that the convergence of the second series in (4.57) is uniform in x and

z. And we have an explicit solution that we can now describe. First of all we see that

for each y it is of the form:

where

1

X .2k/.y/k1

N.M; U1 ; .y// .1 M 2 /.k1/

.2k/

kD0

1

X 1

C `.y/2k .1 M 2 /k ;

.2k C 1/

kD0

Thus in particular the differentiability properties of L are those of 1 .x; z/ and

there are no discontinuities in the velocity off the wing.

Let us next calculate the steady-state structure response. We have for the aerody-

namic moment:

Z b

1

M.y/ D .y/2

U1 .x ab/A1 .x/dx;

1C 1M 2

b

1

D .y/ b 2 U1

2

.1 C 2a/ p :

1 M2

Hence

.y/ 1

M.y/ D .y/2

b 2 U1

2

.1 C 2a/ p :

1C 1M 2

1 M2

Let

.1 C 2a/ 1

2 D
b 2 p :

GJ 1 M2

90 4 The Steady-State (Static) Solution of the Aeroelastic Equation

.y/

00 .y/ D 2 U1

2

.y/2

0 < y < 1; (4.66)

1C 1M 2

.0/ D 0I 0 .1/ D 0:

We have the trivial solution

To find the nonzero solution we see that we have to solve a nonlinear eigenvalue

problem. Making a change of variable

.y/

.y/ D p ;

1 M2

this becomes

.y/

00 .y/ D 2 U1

2

0<y<1 (4.67)

1 C .y/2

.0/ D 0 0 .1/ D 0:

We now establish that (4.67) has a solution only for a countable values of U1 .

Multiplying (4.67) by 0 (y) on both sides we obtain:

2 d log.1 C .y/ /

2

d 0

.y/ D 2 U1 :

d d 2

Or

log.1 C .y/2 /

0 .y/ C 2 U1

2

D const D 0 .0/: (4.68)

2

Putting y D 1 we get:

log.1 C .1/2 /

2 U1

2

D 0 .0/; (4.69)

2

which is an homogeneous boundary condition. Now the solution of the first-order

differential equation (4.68) is analytic in the coefficient

2 U1

2

4.4 The General Nonlinear Case 91

2

. Hence the zeros of

log.1 C .1/2 /

2 U1

2

0 .0/ D 0

2

are at most countable and cannot have a finite accumulation point. Thus we have

at most a sequence of values of U1 or which (4.67) holds. An approximation is

provided by the divergence speeds.

We could also solve the nonlinear Volterra equation:

Z y

log.1 C .s/2 /

.y/ C 2 U1

2

ds D 0 .0/y; 0 < y < 1;

0 2

which follows from (4.68). But the sequence of speeds does not play a role in

stability analysis, so we dont pursue this any more.

Z b

.y/

L.y/ D U1 A.x/dx D 2 bU 21 p

b 1 M2

Next the plunge equation is Eulerian:

@4 h.y/

EI D 2
bU 21 .y/ 0<y<1 (4.70)

@y 4

and is thus determined completely by the pitch, and we do not continue the

calculation because the solution is of little interest.

part 0 in (4.50), even though it plays no role in the stability of the structure.

We first need to solve the nonhomogeneous equation (4.43) for D 0:

@2 k @2 k

.1 M 2 / 2

C D gk1 :

@x @z2

@1

g1 .x; z/ D 2. 1/1 ;

@x

92 4 The Steady-State (Static) Solution of the Aeroelastic Equation

where

Z b p .

@1

D z 1 M 2 ..x

/2 C z2 .1 M 2 //A.

/d

@x b

s

2 bx

A.x/ D U1 `.y/ p ;

1 M2 bCx

Z 1 Z 1

20 .x; z/ D L.x

; z /g1 .

; /d

d;

1 1

where

p

1 z 1 M

2

L.x; z/ D :

x 2 C .1 M 2 /z2

It follows that

1 Z

X 1 Z 1

0 .x; z/ D 0 .x/ C L.x

; z /gk1 d

d: (4.71)

kD2 1 1

We stop here because as before our interest is again only in the structure dynamics

which does not depend on this part of the flow.

Note, however, we have deduced a constructive solution to the Euler potential

equation in the form of a convergent power series.

It is expressible as the sum of two terms neither of which satisfies the equation.

But only one is required for the structure response. Hence there is no point in

calculating the second part as in CFD codes.

is to consider a virtual structure with no dynamics, but we prescribe a normal

velocity on the boundary and seek the solution to the steady state 2D solution for

zero angle of attack aerodynamic field equations. Thus we specify, with the same

notation as before:

@.x; 0/

D w.x/; jxj < b (4.72)

@z

and the KuttaJoukowsky conditions in addition. Proceeding as before we denote

the solution corresponding to w./ by .; x; z/, assumed analytic in with a

nonzero radius of convergence uniformly in x; z.

4.4 The General Nonlinear Case 93

.0; x; z/ D x U1 ;

@.; x; 0/

D w.x/; jxj < b;

@z

For k D 1, we obtain as before the Fourier transform

ejzjr1

O 1 .i !; z/ D O 1 .i !; 0C/; z>0

r1

ejzjr1

D O 1 .i !; 0/; z < 0: (4.73)

r1

In particular then

O 1 D 2O 1 .i !; 0/:

@1

1 .x; z/ D U1 .x; z/:

@x

Hence with

1

A.x/ D ;

U1

A .b/ D 0:

O !/ D 2 1 i ! O 1 .i !/;

A.i

r1

or

1 O j!j p

O 1 .i !/ D A.i!/ 1 M 2:

2 i!

Hence using the Tricomi operator we have:

s

Z s

2 1 bx b bC

A.x/ D p w.

/d

; jxj < b; (4.74)

1 M 2
b C x b b

where to satisfy the Kutta condition we assume that w ./ satisfies a Lipschitz

condition of order ; 12 < 1.

94 4 The Steady-State (Static) Solution of the Aeroelastic Equation

The corresponding solution for the velocity potential is then given by:

jzjr1

e

O1 .i !; z/ D O !/

A.i z>0

2i !

ejzjr1 O

DC A.i !/ z < 0; (4.75)

2i !

where

p Z 1 p .

jzjr1 j!jjzj 1M 2

e De D ei!x 2jzj 1 M 2 .z2 .1 M 2 / C x 2 /dx;

1

z 0:

Hence

p Z b

@1 .x; z/

D z 1 M2 .A .

/d

/=z2 .1 M 2 / C .x

/2 ; jxj < 1:

@x b

(4.76)

Next

D r1 A.i !/ D ejzjr1 1 M 2 A.i

@z 2i ! 2 i!

p Z 1

@1 .x; z/ 1 M2

Dz .B.

//=.z2 .1 M 2 / C .x

/2 /d

; (4.77)

@z 2 1

where

Z b

1 A.

/

B D HA./I B.x/ D d

; jxj < 1;

b x

H being the Hilbert transform. This is an Lp Lq transform, 1 < p < 2, A./ being

in Lp ; 1 p < 2.

A canonical example is:

1

w.x/ D p ; jxj < b; in which case; (4.78)

bCx

s s

Z

2 bx b

1 1

A.x/ D p d

; jxj < b: (4.79)

1 M2 bCx b b

x

The integral does not have a closed form and has to be evaluated numerically.

4.4 The General Nonlinear Case 95

distribution along chord

10

Figure 4.4 shows the plot for b D 1 (within a gain factor). For k 2, abbreviating

the notation for this section k D k0 satisfies the nonhomogeneous equation

@2 k @2 k

.1 M 2 / 2

C D gk1 : (4.80)

@x @z2

For k D 2,

@1

g1 D U1 . 1/1 (4.81)

@x

The nonhomogeneous equation (4.80) is readily solved by taking Fourier trans-

forms. We have

Z 1Z 1

k .x; z/ D L.x

; z /gk1 .

; /d

d; x; z in R2 ; (4.82)

1 1

where

1 p .

L.

; / D 2jj 1 M 2 .2 .1 M 2 / C

2 /: (4.83)

2

derivative as well, no discontinuities in flow velocity. Hence the series:

1

X 1

0 .x; z/ D 0 .x; z/ C k;0 .x; z/ (4.84)

k

kD2

D 0 C L :

The flow thus has no discontinuities. And hence there are no shocks in the flow.

96 4 The Steady-State (Static) Solution of the Aeroelastic Equation

However complex the nonlinearity we do have the zero or rest solution as the static

solution. In turn the linearization gives us back the Goland equation.

Z b

EI 1 h0000 .y/ C .EI 2 EI 1 /..y/v.y/00 /00 D mg cos ' .x; y/dx; (4.85)

b

Z b

GJ 00 .y/ C .EI 2 EI1 /v.y/00 h00 .y/ D .x ab/ .x; y/dx; (4.87)

b

Field Equation

" 2 2 !#

. 1/ @ @

0D 1C

2

a1 2

U1

2

./

2a1 @x @z

1 @ @ @ @

jrj C

2

jrj 2

(4.88)

2 @x @x @z @z

jrj2

D :

2

FluidStructure Boundary Conditions

@.x; 0/ @.x; 0/

D .y/:

@z @x

The gravity terms on the rightmg cos '; mg sin 'contribute to the

complexity.

4.5 Nonlinear Structure Models 97

If we set g in the gravity term to be zero, we see that we have the zero solution

for the structure and constant flow for the aerodynamic equation is a static solution

for all U1 . In particular we need to set g to be zero for determining the ground

modes corresponding to U D 0, and we see that we go back to the linear Goland

model.

If the difference term

EI 2 EI 1

is set to zero, then of course, the equations become linear, and we go back to the

solution for the Goland model we started with, except for the addition of the v./

equation:

EI 2 v0000 .y/ D 0;

Let us examine then the dependence on

.EI 2 EI 1 /:

.EI 2 EI 1 /;

satisfying:

.; y/

D mg cos' 2 bU 21 p

1 M2

EI 2 v0000 .; y/ C .EI 2 EI 1 /..; y/h.; y/00 /00 D mg sin '

.; y/

GJ 00 .; y/ C .EI 2 EI 1 /v.; y/00 h00 .; y/ D ;

.; y/2

1C

1 M2

where

1

D b 2 U1

2

.1 C 2a/ p :

1 M2

Let

dk

hk .y/ D h.; y/ D 0

dk

98 4 The Steady-State (Static) Solution of the Aeroelastic Equation

1

X hk .y/k

h.; y/ D

k

kD0

We can calculate that

.0; y/

EI 1 h0000 00 00

1 .y/ C .EI 2 EI 1 /..0; y/v.0; y/ / D 2
bU 1 p

2

1 M2

EI 2 v0000 00 00

1 .y/ C .EI 2 EI 1 /..0; y/h.0; y/ / D 0

More generally we have for k 2:

EI 1 h0000

k .y/ D rh; k1 .y/

EI 2 v0000

k .; y/ D rv; k1 .y/

with zero end conditions, because the derivatives with respect to are zero.

The functions

;k1 .y/ h;k1 .y/ v;k1 .y/

Theorem 4.6. The pitch equation

FF W k0 .0/ D 0I k0 .`/ D 0

Z `

k .y/ D G .y; s/;k1 .s/ds; 0 < y < `;

0

4.5 Nonlinear Structure Models 99

Proof. r

Let ! D :

GJ

Then

Z ! Z

1 ` s

sin !.s /

k0 .s/ D cos !.` /;k1 ./d C `;k1 ./d

cos !` 0 0 !

u

The solution procedure is similar for the other two equations.

Finally then we have the series expansion for the solution which we can express

in the form

X1

xQ k .y/

x.y/ D .EI 2 EI 1 /k ; 0 < y < `; (4.89)

k

kD0

where

xk .y/

xQ k .y/ D ;

.EI 2 EI 1 /k

0 1

h .0; y/

x.y/ D @ v .0; y/ A

.0; y/

is then the static solution to the problem (CF end conditions)

Z `

xk .y/ D G.y; s/xk1 .s/ds; 0 < y < `;

0

0 1

Gh .y; s/

G.y; s/ D @ Gv .y; s/ A :

G .y; s/

EI 1 h1 .y/0000 D 0;

EI 2 v0000

1 .y/ D 0;

2 !

2 2 y2

D .EI 2 EI l / m g cos ' sin ' 2`y C .6` 3`2 /

4

D .y/

100 4 The Steady-State (Static) Solution of the Aeroelastic Equation

and

Z ! Z

1 ` s

sin !.s /

10 .y/ D k0 .s/ D cos !.` /./d C ./d:

cos !` 0 0 !

And

h1 .y/ D h.0; y/I v1 .y/ D v.0; y/;

which gives some idea about the nature of the solution.

Linearization

Having evaluated the static solution, we now go on to linearize the solution to the

aeroelastic equations about this solution for U nonzero so that g has to be nonzero.

We anticipate that it is more complex than for the Goland model.

Thus let

v.; t; y/ D v.0; y/ C v1 .t; y/;

.; t; y/ D .0; y/ C 1 .t; y/;

.; t; x; z/ D .0; x; z/ C 1 .t; x; z/;

@.; t; x; 0/ @.0; x; 0C/ @1 .t; x; 0C/

D C ;

@z @z @z

where

0 1

h .0; y/

@ v .0; y/ A D x.y/;

.0; y/

M.; t; y/ D M.y/ C M1 .t; y/;

mh.;

D m g sin ' C L.; t; y/; (4.90)

mRv.; t; y/ C EI 2 v0000 .; t; y/ C .EI 2 EI 1 /..; t; y/h.; t; y/00 /00 D mg cos '

(4.91)

4.5 Nonlinear Structure Models 101

R

I .; t; y/ GJ `00 .; t; y/ C .EI 2 EI 1 /h.; t; y/00 v.; t; y/00

D M.; t; y/; 0 < tI 0 < y < `: (4.92)

1 .t; y/ C .EI 2 EI 1 /.1 .t; y/v.0; y/

00

1 .t; y/ C .EI 2 EI 1 /.1 .t; y/h.0; y/

00

I R1 .t; y/ GJ 100 .t; y/ C .EI 2 EI 1 /.h1 .t; y/00 v.0; y/00

C h.0; y/00 v1 .t; y// D M1 .t; y/; (4.95)

d

L1 .t; y/ D L.0; t; y/;

d

d

M1 .t; y/ D M.0; t; y/:

d

This is the linearized equation that will determine system stability and is now more

complex than the Goland case because the coefficients in the equation are no longer

constants. It also then requires an existence and uniqueness proof. We consider this

as part of the convolution/evolution equation in Chap. 5.

The crucial result used here is the explicit solution to the finite Hilbert transform

given by Tricomi [11] who notes also the work of Sohngen, the German pioneer in

aeroelasticity theory, and shows the Kutta condition as necessary for uniqueness of

solution. There is also a large body of related Russian work (mainly that of Gohberg;

see [31]) which seems to ignore the Tricomi work.

Note that the aerodynamics is replaced by the Hilbert transform, which we look

upon as the static case of the Possio integral equation. A feature of our entire theory

is the systematic use of the Possio equation. As here and more generally, the best-

developed results are for the typical section case where the air flow is 2D and the

Possio equation is 1D. The angle of attack becomes important in that the functionals

turn out to be discontinuous at zero for large M . The point here is that for large

enough M there may be regions in the interior for which the local flow is supersonic,

102 4 The Steady-State (Static) Solution of the Aeroelastic Equation

in which case the flow is termed transonic. The precise value of M for which this

happens is not clear. Meyer [14, p 146] gives an inequality for the interior Mach

number. One author [48] suggests M 0:8 as transonic.

We have in this chapter a simple instance of nonunique solutions for the static

solution for a sequence of values of the far field speed determined by a nonlinear

eigenvalue problem. We could argue that we wont see this nonuniqueness in time

marching because we will not have the exact values for the sequence. In any case

it does raise a question for CFD solutions. Our main result concerning the air flow

would appear to be new: that the potential can be expressed as the sum of two,

one which produces lift and the other that does not, consistent with our theme that

we are concerned only with the structure dynamics whereas the second part holds

little interest for us.

Chapter 5

Linear Aeroelasticity Theory/

The Possio Integral Equation

5.1 Introduction

In this chapter we present the linear theory which plays an essential role in the

nonlinear Hopf bifurcation stability theory to come in Chap. 6. This is the most

studied part of the theory beginning with the pioneers: Kussner, Sohngen, Wagner,

Garrick, and Theodorsen; see the classic treatise [6]. This is also the longest chapter

in the book and covers the theory buttressing a wide variety of topics loosely referred

to as Flutter Analysis: the every day bread-and-butter part of the subject.

We begin with a power series expansion of the solution of (3.15) with the associated

boundary dynamics about the rest or equilibrium aeroelastic state solution (Chap. 4):

h.t; y/ D 0I .t; y/ D 0:

We call this an Attractor and note that it is unique in that it holds for every value

of U1 unlike the Divergence speed (Chap. 4).

To determine the Linear potential field equation, we consider the

structure variables as parameters and expand the solution in a power series

about zero. Thus with denoting a complex variable we consider the solution

.; t; x; y; z/ corresponding to h.t; y/; .t; y/:

1

X k

.; t; x; y; z/ D 0 .t; x; y; z/ C k .t; x; y; z/; (5.1)

k

kD1

A V Balakrishnan, Aeroelasticity: The Continuum Theory, 103

DOI 10.1007/978-1-4614-3609-6 5, Springer Science+Business Media, LLC 2012

104 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

Leaving for later the sense in which the convergence is taken, as in the steady

state case in Chap. 4, we first calculate the kth order potentials k given by

k .t; x; y; z/ D @k .0; t; x; y; z/ =@k :

Differentiating (3.15) with respect to we get the field equation for each k:

@2 @2 @2

./ D C 2U 1 Cos C Sin ff

@t 2 @t@x @t@z

@2 @2

a1 2

1 M 2 Cos2 2

C 1 M 2 Sin2

@x @z2

@2 @2

C 2M 2

Cos Sin (5.3)

@y 2 @x@z

and gk involves the potentials up to the kth order, and g0 D 0. We begin with

1 the Linear potential. The boundary dynamics are determined from (3.19): Flow

Tangency:

"

@0 P y/ C .x a/.t;

P y/

.@.; t; x; y; 0//=@z D C .1/ h.t;

@z

C ..@.; t; x; y; 0//=@y/

#

@

.h.t; y/ C .x a/.t; y// ; x; y 2 :

@y

(5.4)

h

P y/ C .x a/.t;

.@1 .t; x; y; 0//=@z D .1/ h.t; P y/

i

C U1 Cos .t; y/ ; x; y 2 : (5.5)

downwash in the aeroelasticity parlance.

In particular we see that the normal derivative of the linear potential is continuous

across the wing. Also no spanwise derivatives are involved.

5.2 Power Series Expansion of the 3D Potential 105

where

2

1

.; t; x; y; z/ D .@.; t; x; y; z//=@t C r.; t; x; y; z/ : (5.6)

2

1

X k

1 2

.; t; x; y; z/ D U1 C k .t; x; y; z/; (5.7)

2 k

kD1

we need to calculate

k

@ .0; t; x; y; z/ =@k D k .t; x; y; z/:

C U1 Sin .@1 .t; x; y; z/=@z/:

we have:

@ @

A1 .t; x; y/ D 1 .t; x; y/ C U1 Cos 1 .t; x; y/: (5.8)

@t @x

And the KuttaJoukowsky conditions are:

which are then the boundary conditions for the linear equation (5.3), along with

(5.5). Thus we can state the

106 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

Field Equation

@2 @2 @2 @2

C 2U 1 Cos C Sin a1

2

1 M 2

Cos2

@t 2 @t@x @t@z @x 2

@2 @2 @2

C 1 M 2 Sin2 C 2M 2

Cos Sin D 0:

@z2 @y 2 @x@z

Flow Tangency:

h i

P y/ C .x a/.t;

D .1/ h.t; P y/ C U1 Cosff .t; y/ ; x; y 2 :

KuttaJoukowsky:

A.t; b; y/ D 0; 0 < y < `: (5.11)

equation. This is not too surprising because basically we have a Neumann-type

boundary value problem which is usually treated this way (cf [1]).

See [55, 56, 72, 96]. The Possio Integral equation plays a central role in our theory

and we shall revisit it in many places in many forms, in both Laplace Domain and

Time Domain.

5.3 The Linear Possio Integral Equation 107

We may look at what we need from the air flow model as an inputoutput problem

where the downwash is the Input and the pressure jump is the Output. The

InputOutput Relation is then the Possio Integral Equation.

We may also consider it as providing the link between Lagrangian Dynamics and

the Fluid Eulerian Dynamics. This gluing is usually the most mysterious part in

computational aeroelasticity.

We take the Laplace Transform of (5.3) in the time domain and the Lp Lq

Fourier Transforms, 1 < p < 2, in the spatial x; y coordinates. We use super to

denote Laplace Transform and super ^ to denote the Fourier Transform:

Z 1 Z 1 Z 1

QO

.; !1 ; !2 ; z/ D et eix!1 eiy!2 .t; x; y; z/dx dy dt

1 1 0

Re > a 0I !1 ; !2 " R2

It is convenient from now on to use the reduced frequency

b

kD

U1

D for b D 1:

U1

Then taking the Laplace Transform of the field equation (5.3), we obtain:

Z 1

O @

.; x; y; z/ D et .t; x; y; z/dt I .0; x; y; z/ D 0I .0; x; y; z/ D 0

0 @t

! "

@O @O @2 O

0 D O C 2U1

2

Cos C Sin

1 M 2 Cos2 2

a1

@x @z @x 2

#

@2 O @2 O @2 O

C .1 M Sin / 2 C 2 2M Cos Sin

2 2 2

: (5.12)

@z @y @x@z

108 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

QO

Using '.; !1 ; !2 ; z/ in place of .; !1 ; !2 ; z/; we have, taking Fourier

2

Transforms in (5.12) and dividing by a1 :

@2 '

@'

1 M 2 Sin2 2 M 2 i !1 Cos Sin C M 2 k Sin

@z2 @z

2 2

M k C 2M 2 k Cos i !1 C !12 .1 M 2 Cos2 / C !22 ' D 0: (5.13)

We note that

' D c erz ;

where c is not a function of z, is a solution of this equation provided r satisfies the

algebraic equation

ar 2 2br c D 0;

where

a D .1 M 2 Sin2 /;

b D M 2 .k C i !1 Cos /Sin ;

c D k 2 M 2 C 2M 2i !1 Cos C !12 .1 M 2 Cos2 / C !22 :

1 2

r1 D M .k C i !1 Cos /Sin

.1 M Sin /

2 2

p 2 2

k M C 2kM 2 i !1 Cos C !12 .1 M 2 / C !22 .1 M 2 Sin2 / ;

(5.14)

1 2

r2 D M .k C i !1 Cos ff/Sin

1 M Sin

2 2

p 2 2

C k M C 2kM 2 i !1 Cos C !12 .1 M 2 / C !22 .1 M 2 Sin2 / :

(5.15)

Before we proceed further we need to clarify the definition of the square root in

(5.14) and (5.15).

Lemma 5.1.

is never equal to a negative number for any complex number k which is not pure

imaginary.

5.3 The Linear Possio Integral Equation 109

Proof. If M D 0 or k D 0

2M 2 . C !1 Cos / and the real part D . 2 2 /M 2 2M 2 !1 Cos C !12 .1

M 2 / C !22 .1 M 2 Sin2 /:

We need to consider the case

. C !1 Cos / D 0;

u

Hence the usual principal value of the square root will be taken throughout.

As noted in Chap. 4, we can again show that the only solution that vanishes as jzj

goes to zero is given by:

QO

.; i !1 ; i !2 ; z/ D AC er1 ; z > 0;

O r1 z

D e ; z > 0;

r1

D A er2 z ; z < 0;

O r2 z

D e ; z < 0;

r2

where we define

@QO

O D z D 0;

@z

which as we have noted is continuous at z D 0.

Now by (5.8)

QO i ! ; i ! / D . C U i ! Cos /

O .; i ! ; i ! /

A.;

1

1

: (5.16)

1 2 1 1 1 2

r1 r2

Hence

O

.; QO i ! ; i ! /

i !1 ; i !2 / D A.;

r1 r2 1

; (5.17)

1 2

r2 r1 . C U1 i !1 Cos /

110 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

where

r1 r2 1 1

D k 2 M 2 C 2kM 2 i !1 Cos

r2 r1 2 . C U1 i !1 Cos /

. p 2 2

C!12 .1 M 2 Cos2 / C !22 k M C 2kM 2 i !1 Cos

C!12 .1 M 2 / C .1 M 2 Sin2 /!22 : (5.18)

1 1

D k 2 M 2 C 2kM 2 i !1 Cos

2 . C U1 i !1 Cos /

. p

C!12 .1 M 2 Cos2 / C !22 k 2 M 2 C 2kM 2 i !1 Cos

C!12 .1 M 2 / C .1 M 2 Sin2 /!22 :

Z b Z `

O

.; x; y/ D PO .; x ; y /A.;

O ; /dd; 1 < x; y < 1

b 0

O ; / ! 0 as ! b;

A.;

where the kernel is specified in terms of its Lp Lq spatial double Fourier Transform

Z 1 Z 1

1 1

PO .; x; y/eix!1 iy!2 dxdy D

1 1 2 . C U1 i !1 Cos /

.

p 2 2

k 2 M 2 C 2kM 2 i !1 Cos C !12 1 M 2 Cos2 C !22 k M

C 2kM 2 i !1 Cos C !12 .1 M 2 / C .1 M 2 Sin2 /!22 ; (5.19)

In particular, specializing to , and this is Possios simple yet profound idea, we

get the integral equation named after him:

5.3 The Linear Possio Integral Equation 111

Z b Z `

O

.; x; y/ D PO .; x ; y /A.;

O ; /d d; x; y
; (5.20)

b 0

where

O ; / ! 0 as ! b;

A.;

and where

Z 1Z 1

1 1

PO .; x; y/eix!1 iy!2 dx dy D

1 1 2 C U1 i !1 Cos ff

.

p 2 2

k 2 M 2 C 2kM 2 i !1 Cos C !12 .1 M 2 Cos2 / C !22 k M

C 2kM 2 i !1 Cos C !12 .1 M 2 / C .1 M 2 Sin2 /!22 :

O :; :/.

Or, in other words this is enough to determine A.;

In particular we note that if the Linear Aeroelastic problem has a solution, then

the Laplace Transform of the Kussner Doublet Function defined in (5.8) will satisfy

the Possio Equation (5.20).

And conversely, if the Possio equation has a solution, then the Laplace/

Fourier Transform of the solution of the potential equation (5.13) is given by:

QO

.; i !1 ; i !2 ; z/ D PQO .; i !1 ; i !2 /A.;

QO i ! ; i ! / 1 er1 z

1 2 z > 0; (5.21)

r1

Q QO 1

PO .; i !1 ; i !2 /A.; i !1 ; i !2 / er2 z z < 0: (5.22)

r2

coordinates that we started with. The Laplace/Fourier Transformation has to be in-

verted. But the Transforms by themselvesthe Laplace Transforms in particular

are of independent interest, specifically in studying system stability assuming the

Transform is invertible!that it is the Transform of a time domain function which

satisfies (5.3).

The existence and uniqueness of the Possio equation (5.20) is of primary concern to

us and we shall need to spread this over many stages.

112 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

First let us note that if the LDPLaplace Domain Possio equation (5.20)has two

solutions and thus also two time-domain solutions of (5.3), then the difference will

satisfy the equation with

wa .t; x; y/ D 0:

But then the difference will also yield a nonzero solution of the potential flow

equation with zero pressure jump across the wing. Hence existence and uniqueness

of solution of the Linear Euler equation with the KuttaJoukowsky boundary

condition is equivalent to the existence and uniqueness of solution to the LDP

Equation with the extra condition that the solution is a Laplace Transform:

Z 1

O x; y/ D

A.; et A.t; x; y/dt Re: > a :

0

Let us specialize now to the Typical Section or Airfoil case, where we set the y-

coordinate to zero simplifying to 2D aerodynamicswhich is equivalent to setting

!2 to zero in the Fourier Transforms. Thus the 1D Possio Equation becomes:

Z b

O a .; x/ D

w PO .; x /A.;

O /d; jxj < b; (5.23)

b

O b/ D 0;

A.; (5.24)

O x/ goes to a finite limit

which may be weakened to A.;

2 2

Q 1 1 M k C 2kM 2 i !Cos C ! 2 .1 M 2 Cos /

O

P .; i !/ D ;

2 .k C i !Cos / p .M 2 k 2 C 2kM 2 i !Cos C ! 2 .1 M 2 //

And an immediate observation from (5.23) is that if A.; O :/ is the solution corre-

O

sponding to w.; O O

:/, then f ./A.; :/ is the solution corresponding to f ./w.; :/.

In particular this implies that the solution has as many time derivatives as wa .t; :/

has.

5.3 The Linear Possio Integral Equation 113

If we set in addition

D0

the kernel P becomes

1 1 p 2 2

PQO .; i !/ D M k C 2kM 2 i ! C ! 2 .1 M 2 / : (5.27)

2 k C i!

The next result is crucial and is the point of departure from the extant aeroelasticity

literature.

Balakrishnan Formula

Let D 0.

Then in (5.27) PNO .; !/ can be expressed as

1 j!j p

PNO .; !/ D NO i !/ ;

1 M 2 1 C B.k; (5.28)

2 i!

NO i !/ is the Fourier Transform

where B.k;

Z 1

NO i !/ D

B.k; O x/ei !x dx;

B.k; 1 < ! < 1;

1

where

kx Z 1

O x/ D k p e

B.k; k ekxs a.M; s/ds x > 0;

1 M2 0

Z 2

k ekxs a.M; s/ds; x < 0;

0

where

M M

1 D I 2 D I

1CM 1M

1 p

a.M; s/ D ..1 s/.2 C s// =.1 s/; 2 < s < 1 ;

and B.k;

114 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

Proof. First let us prove that B.k;

We note that ja.M; s/j is bounded in 2 s 1 for each M . Let c.M /

denote this bound. Since we have:

O x/ D k B.1;

B.k; O kx/ (5.29)

Z 1 1 ex1

exs a.M; s/ds c.M / ; x>0

0 x

Z 2 .

exs a.M; s/ds c.M / 1 ejxj2 .jxj/ x<0

0

and Z 1

1 ex1 P

dx < 1 for p > 1:

0 x

The representation (5.28) was discovered by the author [56] by contour

integration (see [71]) but proved by direct integration, which we follow here.

First let us note that by (5.33) we may take

k D 1:

Let

1 < Re: z < 2 z 0:

Then

Z 1 Z

O x/dx D p 1 1 1 1 1 p

ezx B.1; ..1 s/

1 1 M2 1 C z 0 s C z

Z

1 2 1 p

.2 C s/// .1 s/ds C ..2 s/

0 sz

.1 C s/// .1 C s/ds;

which by changing variables in the integrals (see [4] for details) can be

expressed as:

z p 1

D .z C 1 /.z 2 / 1

zC1 z2

and for

z D i!

this is

j!j 1 p

D ..i ! C 1 /.2 i !// 1:

i! 1 C i!

5.3 The Linear Possio Integral Equation 115

Hence

B.k; 1 2

i! k C i!

j!j 1 1 p 2 2

D p .M k C 2kM 2 i ! C ! 2 .1 M 2 // 1

i! k C i! 1 M 2

Finally let us note that for Typical Section and zero angle of attack, (5.22) and (5.23)

specialize to

QO

.; i !; z/ D

1 QO i !/erz z > 0

A.;

k C i!

DC

1 QO i !/erz z < 0;

A.;

k C i!

where

p

r D .k 2 M 2 C 2kM 2 i ! C ! 2 .1 M 2 //

so that, in particular

.@.t; x; z//=@z D .@.t; x; z//=@z;

.@.t; x; z//=@x D .@.t; x; z//=@z:

The case M D 0 corresponds to Incompressible Flow for reasons explained

below. The importance for us is that we can obtain an explicit solution for the Possio

equation and a systematic use of which yields an alternate theory as compared with

the classical theory in [6]. M D 0 is an outlier in that it is atypical in comparison

to the solution for nonzero M. t

u

Incompressible Flow

First we have:

Theorem 5.3. For each and M; 0 M 1, (5.27) defines an Lp Lq

Mikhlin multiplier on Lp .R1 / and hence a linear bounded operator on Lp .R1 /

into itself. Let .; M; k/ denote the operator. It is convenient to regard Lp .b; b/

as a subspace of Lp .b; b/ Let P denote the projection operator on Lp .R1 / into

Lp .b; b/. Then the Possio Equation can be expressed in operator form as an

equation in Lp .b; b/, 1 p < 2:

O :/

wO a .; :/ D P.; M; k/A.; (5.30)

116 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

O b/ D 0:

A.;

Proof. By definition (see [28, 41, 56]), a scalar valued function ./ de-

fined on R1 is called a Mikhlin multiplier if g.!/ Q D .!/fQ.!/;

1 < ! < 1; excepting ! D 0 is the Fourier Transform of a function in

Lp .R1 / whenever fQ. : / is. A sufficient condition for this due to Mikhlin [1] is that

. : / be continuously differentiable in .1; 1/ and

This condition is readily verified for (5.27). The statement of the theorem follows

from Mikhlins theory [28] who shows that if T denotes the corresponding operator,

then T is linear bounded with norm:

That the space has to be Lp , 1 < p < 2 can be seen by considering the

special case where k D 0, which corresponds to the steady state case treated

in Chap. 4. Let us look at some special cases where an explicit solution can be

constructed. t

u

Note that for zero angle of attack, the kernel PQO in (5.27) becomes:

1p 2 2

.M k C 2kM 2 i ! C ! 2 .1 M 2 //=.k C i !/: (5.31)

2

Theorem 5.4. Let D 0 and M D 0 (Incompressible Case). Then the Possio

Equation in this case:

O :/

wO a .; :/ D P.0; 0; k/A.; (5.32)

has a unique solution in Lp which we can express explicitly. Here we follow [4].

Proof. Let .0; 0; k/ be simplified to .k/. For D 0 D M the multiplier is

1 j!j

;

2 k C i!

5.3 The Linear Possio Integral Equation 117

1 k j!j

1 :

2 k C i! i!

1

:

k C i!

Z x

R.k/f D gI g.x/ D ek.xs/ f .s/ds; 1 < x < 1: (5.33)

1

j!j

Next the multiplier i! is recognized as representing the Hilbert Transform, denoted

H, given by

Z 1

1 f ./

Hf D gI g.x/ D d; 1 < x < 1: (5.34)

1 x

HR.k/ D R.k/H:

Then P.k/ can be expressed:

1

P.k/Pf D P.I kR.k//HPf for f in Lp .R1 /:

2

Hence (5.33) becomes:

O :/ kPR.k/HP A.;

O a .; :/ D PHP A.;

2w O :/

O :/ kPHR.k/P A.;

D PHP A.; O :/

D PHP A.; O :/

D PHP.I kR.k/P/A.; O :/: (5.35)

O :/:

PH.I P/R.k/P A.;

118 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

Denoting it by q. : /, we have:

Z 1 Z b

1 1 O :/dsd;

q.x/ D ek.s/ A.; jxj < b

b x b

Z 1

1 1 O :/ ;

D ek d L k; A.;

0 xb

where Z b

L.k; f / D ek.bs/ f .s/ds:

b

Hence we have

O

2w.; O :/ kq:

:/ D PHP.I kR.k/P/A.;

Hence

O

2T w.; O :/ kT q;

:/ D .I kL.k//A.; (5.36)

where L.k/ for each k is a Volterra operator on Lp .b; b/ into itself, defined by:

Z x

L.k/f D gI g.x/ D ek.xs/ f .s/ds; jxj < b

b

O ://h;

kL.k; A.;

where

s " r #

Z Z 1

1 bx 1 b

bCs 1 1 k

h.x/ D ds e d ;

bCx b bs sb sx 0

Z r

1 b

bCs p 2b C 1

1=..s b /.s x//ds D : (5.37)

b bs xb

Here T is recognized as the Tricomi Operator which we treat more fully in the

sequel.

Hence

s

Z

1 b x 1 k p 2b C 1

h.x/ D e d jxj < b: (5.38)

bCx 0 xb

(Note that . : / is NOT defined for k negative. An important point for us in the

sequel, in Sect. 5.5.) Hence (5.37) becomes:

5.3 The Linear Possio Integral Equation 119

O

2T w.; O :/ kL.; A.;

:/ D .I kL.k//A.; O ://h: (5.39)

t

u

Lemma 5.5.

.I kL.k//1 D .I C kL.0//: (5.40)

Proof. Either by the Volterra Expansion

1

X

.I kL.k//1 D .kL .k//n :

nD0

Or, solving

.I kL.k//1 f D g;

Z x

f .x/ D g.x/ k ek.xs/ g.s/ds:

b

Hence

g0 .x/ D f 0 .x/ C kf .x/:

Hence Z x

g.x/ D f .x/ C k f .s/ds; since f .b/ D g.b/:

b

Hence

.I kL .k//1 D .I C kL.0//: t

u

Hence

O :/ kL.k; A.;

2.I C kL.0//T wO a .; :/ D A.; O ://.I C kL.0//h (5.41)

O ://:

L.k; A.;

For which we take the functional L.k; :/ on both sides of (5.41). L.k; lhs/ D

L.k; rhs/ which yields

L.k; A.; O :///=.1kL.k; .I CkL.0//h//: (5.42)

However we still need to verify that the denominator in (5.42) is not zero. Here we

have a remarkable result due to Sears [110].

120 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

Sears Formula

For Re: k > 0:

Z 1

p 2b C

.1 kL.k; .I C kL.0//h// D k ek d:

0

kL.k; .I C kL.0//h/:

First, integrating by parts in

Z b Z x

k.bx/

L.k; kL.0/h/ D e k h.s/dsdx

b b

yields

Z b

D h.s/ds L.k; h/:

b

Hence

0 s 1

Z 1 Z

p 2b C bx

b

kL.k; .I Ck L.0/h// D k ek @1 1

dx A d;

0 b bCx xb

p

D 1:

2b C

Z 1

p 2b C

1 kL.k; .I C kL.0//h/ D k ek d: (5.43)

0

That it is analytic in k in the entire plane except for k 0 where it has a logarithmic

singularitytypifying the analyticity properties functions of k we shall use. This

can be seen from rewriting (5.43) in the formby a change of variable:

Z 1

p 2bk C t

et dt: (5.44)

0 t

Z 1

k p 2b C

k e d D bkK0 .bk/ C K1 .bk/ebk : (5.45)

0

5.3 The Linear Possio Integral Equation 121

, Z

1

k t p 2b C t

1 k e dt : (5.46)

0 t

Z 1

c1 .t/ek t dt; c1 . : / 0:

0

gives an approximation to the function c1 . : /, which is referred to as the Sears

function. This is necessary to obtain the time domain structure response. But it is

easier of course to show just that (5.44) is bounded away from zero.

In the form (5.44) we see that as k goes to zero in the right-half-plane, the

function has the limit equal to 1, and goes to infinity as Re k goes to infinity. In

fact we have

Z 1 Z 1

p 2bk C t p 2bk

et dt 1 D et C 1 1 dt 0 for k 0

0 t 0 t

p

and Re 1 C z 1 0 for Re z 0. t

u

2 3

A.; 42T wO a .; :/ C h R1 q 5:

e t 2bkCt

dt

0 t

Or

O :/ D .I C kL.0// 2T wO a .; :/ C h .L.k; .I C kL.0//2T wO a .; :/// :

A.;

bk.K0 .bk/ C K1 .bk//ebk

(5.47)

To show that

O b/ D 0;

A.;

we make a change of variable

D t 2 .b x/

122 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

in s r

Z 1

1 bx k 2b C 1

h.x/ D e d

bCx 0 xb

yielding

r Z 1

1 1 p 2dt

ek t

2 .bx/

D .1/ 2b C t 2 .b x/ ;

bCx 0 1 C t2

and in this form it follows readily that h.x/ ! 1 as x ! b. Now from (5.34),

evaluating the limit as x ! b, we see that the left side goes to zero, while the right

side

O :/ kL.; A.;

D .I kL.k//A.; O ://h

Z b Z b

O

0 D A.; b/ k e k.bs/ O

A.; s/ds C k O s/ds:

ek.bs/ A.;

b b

O b/ D 0.

Or A.; t

u

Remarks: It is of interest in the sequel to determine the range of values of k beyond

Re k > 0 for which the solution continues to be valid.

Here we use the solution in the form (5.47) involving the Bessel K

functions. They are defined and analytic and nonzero except for k < 0I and we

have already covered k D 0: Hence the solution is valid for all k omitting the

negative real axis where we have a line of singularity. What about the kernel

in the equation? Here the problem is that 1=.k C i !/ is no longer the Laplace

transform of a function in the positive time domain for Re k negative. This ends the

incompressible case where we have presented a closed-form solution to the Possio

equation in the Laplace domain. We consider the timedomain solution later. Next

we go on to consider M > 0 where we no longer have the luxury of a closed-form

solution of the Possio equation (not as yet, of course).

form solution available (except for M D 1). Nor does the technique of solution for

M D 0 carry over.

In fact we show that M D 0 is typical and may be labeled singular. We call

the flow compressible if the far field Mach number is bigger than 0. There is a

further qualification: transonic is somewhat vague, for M > 0:8 (this depends

on the Mach number of the flow in the interior; see [49]). This further distinction

plays no role in our theory and we do not invoke it. Our first objective is to develop

an abstract version of the equation.

5.4 Linear Possio Equation: Compressible Flow: M > 0 123

throughout that

wa .t; :/ is in C1 b; b for every t

and Z 1

e t jjwa .t; :/jjdt < 1 > a > 0:

0

For most if not all of our purposes we may take the abscissa a to be zero. The C1

condition is indeed true for the downwash.

Let .!/ denote the Lp Lq Mikhlin multiplier

1 j!j p

.!/ D NO i !/ ;

1 M 2 1 C B.k;

2 i!

where

Z

NO i !/ D k 1 1 1

1

B.k; p a.M; s/ds :

1 M 2 k C i! 2 ks C i !

1p

1 M 2 H.I C B.k//

2

so that the Possio equation (5.23) in abstract form is:

1p O

wO a .; :/ D 1 M 2 PH.I C B.k//PA.; :/; (5.48)

2

where H is the Hilbert transform

Z 1

1

Hf D gI g.x/ D f .s/ds; 1 < x < 1

1 .x s/

Z

kR.k/A 1

B.k/A D p kR.ks/Aa.M; s/ds; A Lp .b; b/;

1 M2 2

where R.k/ is the bounded linear operator on Lp .R1 / into itself corresponding to

the multiplier

1

;

k C i!

124 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

and is the resolvent of the generator D, defined for all k off the imaginary axis.

Df D f 0

S.t/f D f .: t/ t 0

R.k/ is short for R.k; D/, the resolvent of D defined for off the imaginary axis:

Z 1

R.; D/f D et S.t/f dt; Re > 0;

0

Z 1

R.; D/f D et S.t/f dt; Re > 0:

0

Again P denotes the projection operator on Lp .R1 / into Lp b; b 1 < p < 2. The

Hilbert transform can be expressed [10] as

Z 1

1 .S.t/f S.t/f /

Hf D dt;

0 t

where

HR.; D/ D R.; D/H

and we note that Z 2b

PR.; D/Pf D et PS.t/Pf dt

0

Lemma 5.7. As Re k goes to infinity, B.k/ converges strongly over Lp .R1 / to

M

I p H;

1 M2

M j!j

1 p :

1 M2 i!

s

NO i !/ D i ! kM 2 .k C 2i !/

B.k; 1C 1;

k C i! ! 2 .1 M 2 /

5.4 Linear Possio Equation: Compressible Flow: M > 0 125

M j!j

1 p :

1 M 2 i!

This is enough but we establish this directly using elementary semigroup theory:

Z

kR.k; D/f 1

B.k/f D p kR.ks; D/f a.M; s/ds:

1 M2 2

We note that

kR.k; D/f ! 0 as Re k ! 1:

Hence the first term goes to zero. But jjkR.k; D/jj does not. kR.k; D/ converges

only strongly. However, we do have that

jkj

jjksR.ks; D/jj ; Re k > 0; 2 < s < 1 :

Re k

And the integral can be expressed:

Z 1 Z 1

kR.ks; D/f a.M; s/ds a.M; 0/ kR.ks; D/f ds

2 2

Z 1

C kR.ks; D/f .a .M; s/ a.M; 0//ds

2

where

M

a.M; 0/ D p

1 M2

and

Z 1 Z 1Z 1

kR.ks; D/f ds D ke kst ds S.t/f dt

2 0 0

Z 1Z 2

ke kst ds S.t/f dt

0 0

Z 1Z 1

D ekst ds k.S.t/f S.t/f /dt

0 0

Z 1Z 2

ke kst ds S.t/f dt

0 1

Z 1

.S.t/f S.t/f /

D .1 ek1 t / dt

0 t

Z 1

.ek1 t ek2 t /

S.t/f dt;

0 t

126 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

Z 1

.S.t/f S.t/f /

dt D Hf:

0 t

Now Z 1

a.M; s/ a.M; 0/ kR.k; D/f

ksR.ks; D/f ds p

2 s 1 M2

converges boundedly in any sector of opening less than =2 to

Z

1

a.M; s/ a.M; 0/ f

f ds p :

2 s 1 M2

Z 1 Z 1

k k

a.M; s/ds D .a.M; s/ a.M; 0//ds

2 ks C i ! 2 ks C i !

Z 1

k

C a.M; 0/ ds;

2 ks C i !

where the second term goes to the multiplier corresponding to the Hilbert transform:

M j!j

p :

1 M i!

2

But

Z 1

N

O k 1 k

B.k; i !/ D p .a.M; s/ a.M; 0//ds

1M 2 k C i ! 2 ks C i!

Z 1 Z 1

M k k

Cp ds .a .M; s/ a.M; 0//ds

1 M 2

2 ks C i ! 2 ks C i!

Z 1

NO i !/ C

D B.k;

k 1

p

M k

p ds :

1 M 2 k C i! 1 M 2 2 ks C i w

Z 1

1 1

.a .M; s/ a.M; 0// ds p D 1;

2 s 1 M2

which establishes the result. But we emphasize that the convergence is only

strong. t

u

5.4 Linear Possio Equation: Compressible Flow: M > 0 127

Using this result we can express the Possio equation in abstract form as an

equation in Lp b; b:

p

2wO a .; :/ D O

1 M 2 PH.I C B.k//A.; :/

p

D O

1 M 2 PH.I C B.k/ B.1//A.; :/

p

M O

C 1 M 2 PH I p H A.; :/:

1 M2

Or

p

2 O 1 M2 O

wO a .; :/ D A.; :/ C PH.B.k/ B.1//A.; :/; 0 < M < 1;

M M

(5.49)

which is defined for all off the imaginary axis where R.k; D/ is defined. It is

important to note that in (5.49), we are restricted to 0 < M < 1. The cases M D 0

and M D 1 are treated separately below. Note that (5.49) does not impose the Kutta

condition.

It is not evident from (5.49) that the solution will satisfy the Kutta condition: For

this we go to an equivalent but different version of the LDP (Laplace domain Possio)

by invoking the Tricomi operator.

Theorem 5.8. The Possio equation can be expressed in the form:

2T wO a .; :/ O

p D ..I C PB .k/P/ C T PH.I P/B.k/P/A.; :/; (5.50)

1 M2

Proof. For nonzero M we multiply both sides of (5.48) by M to obtain:

p

2w O

O a .; :/ D M A.; :/ C O

1 M 2 PH.B.k/ B.1//A.; :/;

where

p p

1 M 2 PH.B.k/ B.1// D 1 M 2 PH.B.k/ C I / MI:

Hence

O a .; :/

2w O O :/;

p D PHA.; :/ C PHB.k/A.;

1 M2

128 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

O

D PHPA.; O

:/ C PHPB.k/PA.; O

:/ C PH.I P/B.k/A.; :/: (5.51)

t

u

Lemma 5.9. Suppose f is in Lp b; b; 1 < p < 2 and

T f D 0:

Then

f D 0:

Proof. Let

T f D g;

s r

Z

1 bx b

b C s f .s/

g.x/ D ds;

bCx b bs sx

which is given to be

D 0; jxj < b:

Hence Z r

b

b C s f .s/

ds D 0; jxj < b:

b bs sx

Hence (see [11]):

r

bCs const

f .s/ D p ;

bs b2 s2

const

f .s/ D ; jsj < b;

bCs

1 < p < 2. t

u

Hence

O a .; :/

2w O O :/ D 0

T p PHA.; :/ C PHB.k/A.;

1 M2

is equivalent to

O a .; :/

2w O :/ C PHB.k/A.;

O :/ D 0;

p PHA.;

1 M2

u

5.4 Linear Possio Equation: Compressible Flow: M > 0 129

Lemma 5.10. PB.k/P is a Volterra Operator on Lp .b; b/ into itself, and its

range is contained in C1 b; b. It is defined for all k and is an entire function.

Proof.

Z 1

kPR.k/PA

PB.k/PA D p kPR.ks/PA a.M; s/ds

1 M2 1

Z 1

kPR .ks/PA a.M; s/ds;

2

where Z 1

PR.k/Pf D gI gD ek t PS.t/ P f dt:

0

And because

PS.t/P D 0 for t > b;

we have: Z x

g.x/ D ek.xs/ f .s/ ds; jxj < b:

b

Let us use the notation:

L.k/ D PR.k/P

L.k/ is then an entire function of k. And

Z

kL.k/ 1

PB.k/P D p kL.ks/ a .M; s/ds

1 M2 2

is a Volterra operator on Lp .b; b/ into itself, and L.k/ is defined for every k in

the plane. In fact we have for

A D PB.k/Pf

Rx

k ek.x/ f ./d

b

A.x/ D p

1 M2

Z x Z 1

k f ./ eks.x/ a.M; s/ ds d; jxj < b:

b 2

into C1 b; b. t

u

Getting back to (5.51), we see that by the Tricomi theorem

h i

O

T PHPA.; O

:/ C PHPB.k/PA.; O

:/ D .I C PB.k/P/A.; :/

130 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

and is in Lp b; b. Next note that the third term on the right in (5.51),

O

PH.I P/B.k/A.; :/;

O

T .PH.I P/B.k/A.; ://

is in Lp b; b for 1 < p < 4=3: We can prove more by actual calculation.

Lemma 5.11. Let W .k/ D T PH.I P/B.k/P which by (5.50) is defined for every

k off the imaginary axis. However for Re k > 0 by [5, 8] for A in Lp b; b; 1 <

p < 2,

Z 1

k

W .k/A D p h .k/L .k; A/ C k h .ks/L .ks; A/a.M; s/ds

1 M2 0

Z 2

Ck .hC .ks/ C j.ks//LC .ks; A/a .M; s/ds; (5.52)

0

Z b

LC .k; A/ D ek.bC/ A./d;

b

Z b

L .k; A/ D ek.b/ A./d

b

s r

Z 1

1 bx 1

hC .k; x/ D ek d;

bCx 0 2b C b C C x

s r

Z 1

1 bx 2b C 1

h .k; x/ D ek d;

bCx 0 bC x

Each function on the right in (5.52) is in Lp b; b; 1 < p < 2. Furthermore,

hC .k; b/ D 0;

h .k; b/ D 1:

5.4 Linear Possio Equation: Compressible Flow: M > 0 131

Proof.

Z

kR.k/A. : / 1

D T PH.I P/ p kR.ks/A. : /a.M; s/ds :

1 M2 2

s s

Z

k bx b

bC 1

T PH.I P/kR.k/A. : / D

bCx b b x

Z 1 Z Z s

1 1

b ek.s/ .PA/./d ds d:

1 b s 1

We therefore have

s s

Z

k bx b

bC 1

k T PH.I P/R.k/A. : / D

bCx b b x

"Z Z # Z

b 1 s

1 1

C ek.s/ .PA/./ d ds d

1 b s b

s s

Z

k bx b

bC 1

D

bCx b b x

Z 1 Z b

1 1

ek.s/ .A/./d ds d:

b s b

obtain

s

Z s

k b x b b C 1

T PH.I P/kR.k/A. : / D

b C x b b x

Z 1 Z b

1 ek

ek.b/ A./d d d

0 . C b/ b

Z Z

k 1 b k.bs/

D e A.s/ds ek

0 b

Z s

1 b bC 1 1

d d:

b b x . C b/

132 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

Z s

1 b

bC 1 1

I D d;

b b x . C b/

we get that

Z s

1 b

bC 1 1 1

I D d:

b b x . C b/ x . C b/

Z s

1 b bC 1

D 1 for jxj < b and the integration identity;

b b x

Z s

1 b bC 1 bCz

d D p C 1 for z > b and therefore;

b b z z2 b 2

1 2b C

I D 1 p 1

x . C b/ 2 C 2b

p

1 2b C

D p :

. C b/ x

s p

Z 1

k bx ek 2b C

T PH.I P/kR.k/A./ D p

bCx 0 . C b/ x

Z b

ek.b/ A./d d

b

Z 1

T PHk .I P/R.ks/A. : /a.M; s/ds

0

Z 1

D kh .ks; x/L .ks; A/a.M; s/ds :

0

Z 2

T PHk .I P/R.ks/A. : /a.M; s/ds

0

5.4 Linear Possio Equation: Compressible Flow: M > 0 133

starting with

Z 2

T PHk R.ks/A. : /a.M; s/ds

0

s s "Z #

Z b Z Z 1

k bx b

bC 1 1 b

1

D C C

bCx b b x 1 b b z

Z 2 Z 1

e ks.z/

.PA/./d a.M; s/ds dz d;

0 z

where we used the fact that R.k/ is the operator corresponding to the multiplier

1

ki !

which when Re k < 0 is

Z 1

R.k/f D ek.z/ f ./d

z

Z 2

T PHk R.ks/A. : /a.M; s/ds

0

s s "Z

Z b Z b#

k bx b

bC 1 1 1

D C

bCx b b x 1 b z

"Z Z #

2 b

eks.z/ A./d a.M; s/ds dz d: (5.53)

0 z

s s

Z Z b

k bx b

bC 1 1 1

I1 D

bCx b b x 1 z

"Z Z #

2 b

e ks.z/

A./d a.M; s/ds dz d

0 b

s "Z s #

Z Z 1

k 2

bx b

bC 1 1 1

D d

0 bCx 0 b b x C Cb

"Z #

b

ks. CbC/

e A./d a.M; s/ d ds:

b

134 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

Evaluating

Z s

1 b

bC 1 1

d

b b x C . C b/

Z s

1 b bC 1 1 1

D d

b b . C b/ C x x C . C b/

1

D 1 p 1

. C b/ C x 2 C 2b

p

1

D p ;

x C . C b/ C 2b

Z s

1 b bC 1

D 1; for jxj < b and the integration identity;

b b x

Z s

1 b bC 1 bz

d D p C 1 for jzj > 1:

b b z z2 b 2

Thus, I1 becomes

s p #

Z Z 1 Z

k 2 bx 1

I1 D p d

0 bCx 0 0 x C . C b/ C 2b

"Z #

b

eks. CbC/ A./d a.M; s/ d ds

b

Z 2

D k hC .ks; x/LC .ks; A/a.M; s/ds:

0

s

Z s Z

k b x b b C 1 1 b 1

I2 D

b C x b b x b z

"Z Z #

2 b

eks.z/ A./d a.M; s/ds dz d

0 z

Z "Z #

2 b

D k T PHP e ks.:/

A./d a.M; s/ds

0 :

5.4 Linear Possio Equation: Compressible Flow: M > 0 135

Z 2 Z b

D k eks.x/ A./d a.M; s/ds:

0 x

Z 2

I3 D k T PHPR.ks/A. : /a.M; s/ds

0

Z 2

Dk L.ks/PA. : /a.M; s/ds

0

Z 2 Z x

Dk eks.x/ A./d a.M; s/ds:

0 b

Therefore, we get

Z 2

I1 C I 2 I 3 D k hC .ks; x/LC .ks; A/a.M; s/ds

0

Z 2 Z b

k eks.x/ A./d a.M; s/ds

0 b

Z 2

D k hC .ks; x/LC .ks; A/a.M; s/ds

0

Z 2 Z b

k eks.xCb/ eks.bC/ A./da.M; s/ds

0 b

Z 2

D k .hC .ks; x/ C eks.xCb/ /LC .ks; A/a.M; s/ds

0

s r

Z 1

1 bx 2b C 1

h .ks; x/ D eks d; s > 0; (5.54)

bCx 0 bC x

Z r

1 2b C

1

eks d ; s>0

0 bC x

is Z r

1

1 2b C Re ks

e d

bx 0

136 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

and s

bx 1 1

Dp ; jxj < b

bCx bx b x2

2

s r

Z 1

1 bx 1

hC .ks; x/ D eks d; s > 0;

bCx 0 2b C b C C x

s r

Z 1

1 bx 1 Re ks

jhC .ks; x/j e d; s > 0; (5.55)

bCx 0 2b C

and hence hC .ks; :/ is in Lp b; b; 1 < p < 2 for s > 0, Re k > 0, and

Hence

Z 1 R b jA./jd r 1

h .ks; x/L .ks; A/a.M; s/ds b

0 b2 x2

Z 1r Z 1

1

a.M; s/eRe k s ds d; (5.57)

0 2b C 0

verifying that the second term in (5.52) is in Lp b; b; 1 < p < 2. Similarly

Z 2

.hC .ks; x/ C j.ks; x//LC .ks; A/a.M; s/ds

0

Rb s

Z r Z 2

b jA./jd bx 1 1

a.M; s/eRe k s ds d

bCx 0 2b C 0

Z b Z 2

C jA./jd eRe k sx a.M; s/ds;

b 0

and

u

5.4 Linear Possio Equation: Compressible Flow: M > 0 137

Theorem 5.12. Suppose the (Possio) equation given by (5.50) for any has a

unique solution. Then it is given by

A.; p

1 M2

O :/ satisfies the Kutta condition:

and A.;

O x/ ! 0

A.; as x ! b:

1 < p < 2.

Proof. Follows immediately from (5.51) because it is defined in terms of the

continuous linear functionals LC .k; A/ and L .k; A/0 . t

u

It follows from the Lemma 5.13 that

where C.k/ is compact on Lp b; b into itself, 1 < p < 2. Hence either there is a

nonzero A such that

.I C C.k//A D 0

or

.I C C.k//

has a bounded inverse. Hence if the solution is unique, there is no nonzero A such

that

.I C C.k//A D 0

and the theorem follows.

Theorem 5.14. Suppose (5.50) has a unique solution. Then the solution will satisfy

the Kutta condition.

Proof. Let

O :/:

g D ..I C PB.k/P/ C T PH.I P /B.k/P /A.;

Then

Rx

k O /d

ek.x/ A.;

O x/ C

g.x/ D A.; b

p

1 M2

Z x Z 1

k O /

A.; eks.x/ a.M; s/dds; jxj < b

b 2

Z 1

k

Cp h .k; x/L .k; A/ C k h .ks; x/L .ks; A/a.M; s/ds

1 M2 0

Z 2

Ck .hC .ks; x/ C j.ks; x//LC .ks; A/a.M; s/ds:

0

138 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

Rb

k O /d

ek .b /A.;

O b/ C

g.b/ D A.; b

p

1 M2

Z b Z 1

k O /

A.; eks.b/ a.M; s/dds; jxj < b

b 2

Z 1

k

Cp C L .k; A/ C k L .ks; A/a.M; s/ds

1 M2 0

Z 2

Ck j.ks; b/LC .k; A/a.M; s/ds

0

O b/:

D A.;

u

Remarks: What is new is the introduction of the operator W .k/ which is defined

for all k, except for the negative real axis where it has a logarithmic singularity. See

below on the latter.

Existence Theorem

In lieu of a blanket existence theorem for every M and every k, we offer two results,

each with some limitation.

Existence Theorem for k in Bounded Vertical Strip

Theorem 5.15. For values of k in a bounded vertical strip, that is,

we can find M0 such that the LDP has a unique solution for all M < M0 ;

M0 > 0.

Proof. We begin with the equation in the form:

2T wO a .; :/ O

p D .I C PB.k/P C W .k//A.; :/:

1 M2

5.4 Linear Possio Equation: Compressible Flow: M > 0 139

where

Z 1

1

PB M .k/P D k L.k/ p 1 k L.ks/a.M; s/ds;

1 M2 2

where

1

WM .k/A D .kh .k/L .k; A// p 1

1 M2

Z 1

Ck h .ks/L .ks; A/a.M; s//ds

0

Z 2

Ck .hC .ks/ C j.ks//LC .ks; A/a.M; s/ds;

0

so that

(Note P ! P B ! B here and below.)

2T wO a .; :/ O :/:

p D .I C PB0 .k/P C W0 .k/ C PB M .k/P C WM .k//A.;

1 M2

For M D 0; we have already proved that

O :/

.I C .I C PB0 .k/P C W0 .k///1 .PB M .k/P C WM .k//A.;

2T wO a .; :/

D .I C PB0 .k/P C W0 .k//1 p ;

1 M2

Next we show that

can be made less than 1 in operator norm for all k, Re k > a for 0 < M < M0 for

some M0 .

140 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

We calculate

Z 1

1

D kL.k/A p 1 kL.ks/Aa.M; s/ds

1 M2 2

1

C.kh .k/L .k; A// p 1

1 M2

Z 1

Ck h .ks/L .ks; A/ a.M; s/ds

0

Z 2

Ck .hC .ks/ C j.ks//LC .ks; A/a.M; s/ds

0

1

D p 1 .kL.k/A C kh .k/L .k; A//

1 M2

Z 1 Z 1

kL.ks/Aa.M; s/ds C k h .ks/L .ks; A/a.M; s/ds

2 0

Z 2

Ck .hC .ks/ C j.ks//LC .ks; A/ a.M; s/ds;

0

where Z 1

kL.ks/Aa.M; s/ds

2

can be expressed

Z Z 1

1

.a.M; s/ a.M; 0//

D skL.ks/A ds kL.ks/Aa.M; 0/ds:

2 s 2

1

p 1 ! 0 as M !0

1 M2

and

p

1 .1 s/.2 C s/

a.M; s/ D ; 2 < s < 1

1s

q

2M 2 M2

1 s C 1M 2 s C 1M 2

2

D ;

1s

r

1 M2

a.M; 0/ D :

1 M2

5.4 Linear Possio Equation: Compressible Flow: M > 0 141

Hence

2q 3

2M 2 M2

a.M; s/ a.M; 0/ 1 6 s 2 C 1M 2s C 1M 2 M 7

D 4 p 5

s s 1s 1 M2

"p #

1 s 2 .1 M 2 / 2M 2s C M 2 M.1 s/

D p

1 M2 .1 s/s

"p #

1 M 2 .1 s 2 / s 2 M.1 s/

D p

1 M2 .1 s/s

q

s2

M 1 1 M.1s/

D p :

1M 2 s

p

Hence for jsj < M 2 C 4M M; this is in absolute magnitude

p

1 s

< 1 M 2 :

2 .1 s/

Z M

1CM

ja.M; s/jds ! 0 with M;

M

1M

Z M

1CM

ja.M; 0/jds ! 0 with M;

M

1M

Z

a.M; s/ a.M; 0/

M

1CM

ds ! 0 with M:

M s

1M

Hence

1

X 1

D .1/n ..I C PB0 .k/P C W0 .k//1 .PB M .k/P C WM .k///n A;

nD0

142 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

A.;

2T wO a .; :/

.I C PB0 .k/P C W0 .k//1 p

1 M2

2T wO a .; :/

D .I C PB 0 .k/P C W0 .k//1 .I PB M .k/P C WM .k// p

1 M2

u

Theorem 5.16. The Possio equation has a unique solution for small enough k,

given by the Neumann expansion:

X1

O :/ D 2T wO a .; :/

A.; .1/n .PB.k/P C W .k//n p :

nD0 1 M2

O :/ D 2T wO a .0; :/

A.0; p

1 M2

Proof. We dont need to invoke the fact that W .k/ is compact. We prove that

PB.k/P C W .k/ goes to zero in operator norm with k.

Now jj.I C PB.k/P/1 -I jj goes to zero as k goes to zero, because jjPB.k/Pjj

goes to zero as k goes to zero. Hence it is enough to show that jjW .k/jj is small for

small k.

Now

s

Z r

1 bx 1 1

hC .k; x/ D ek d

bCx 0 2b C b C C x

s r

Z 1

1 bx 1

D e d;

bCx 0 2bk C C k.b C x/

5.4 Linear Possio Equation: Compressible Flow: M > 0 143

s r

Z 1

1 bx 1

hC .i !; x/ D e d;

bCx 0 2bi ! C C i !.b C x/

s r

Z 1

1 1 bx 1

jhC .i !; x/j p e d:

2bj!j bCx 0

s r

Z 1

1 bx 1

e d

bCx 0 2bk C

s r

Z 1

1 bx 1

e d

bCx 0 2bk

r s

Z 1

1 1 bx 1

p e d:

2bk bCx 0

p !

jkj

jjkhC .k/jj D 0 p

j cos j

p

D0 jkj if D :

2

Hence the theorem holds as k goes to zero in such a way that the real part is bounded

away from zero, or equal to zero. We note that the larger the M , the smaller the k

we will need. The Neumann expansion follows as soon as k is small enough so that

the operator norm

u

144 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

p

2 O :/ C 1 M2 O :/

wO a .; :/ D A.; PH.B.k/ B.1//A.;

M M

whereas Re k goes to infinity. Here PH.B.k/B.1// goes to zero strongly only, not

O :/ converges

in operator norm! Hence, for example, if wO a .; : / D f . : / and if A.;

then the limit will be .2=M /f . : /:

Note this cannot hold for M D 0: The limit blows up as we show in that case.

We show that it does hold for M D 1: Note also that in (5.50) we cannot assert that

PH.B.k/ B.1//

p

1 M2

IC PH.B.k/ B.1//

M

Theorem 5.17. Suppose the Possio equation has a unique solution.

Then p

1 M2

IC PH.B.k/ B.1//P

M

has a bounded inverse on Lp .b; b/.

Proof.

A.; p :

1 M2

Hence by (5.50)

p !

1 M2

IC PH.B.k/ B.1// .I C PB.k/P

M

2T wO a .; :/ 2

C T PH.I P/B.k/P/1 p D wO a .; :/:

1 M2 M

Here wO a .; :/ is any element of C1 b; b, therefore we have (for nonzero M):

p !

1 M2

I PH.B.k/ B.1// .I C PB.k/P

M

M

CT PH.I P/B.k/P/1 p T D I;

1 M2

5.4 Linear Possio Equation: Compressible Flow: M > 0 145

" p #1

1 M2

IC PH.B.k/ B.1// f

M

M

D .I C PB.k/P C T PH.I P/B.k/P/1 p Tf

1 M2

u

The convergence for large Re k being only in the strong sense, we cannot assert

that either side converges. However, assuming both sides converge strongly, we can

check whether:

M

f D .I C PB.k/P C T PH.I P/B.k/P/1 p T f:

1 M2

MH

.I C PB.1/P/f D I P I C p Pf

1 M2

M

D p PHPf;

1 M2

M

T PH.I P/B.1/Pf D T PH.I P/ P p HP f

1 M2

M

Dp T PH.I P/HPf

1 M2

M

Dp .T C T PHPH/f:

1 M2

Because

T PHPHf D PHf; for f in C1 b; b;

we have

M

.I C PB.k/P C T PH.I P/B.k/P/1 p Tf

1 M2

T 1 T f D f:

Hence at least under the assumption that the Possio equation has a solution we do

have that p !1

1 M2

IC PH.B.k/ B.1// f Df

M

for f in C1 b; b as Re k goes to infinity.

146 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

LDP.

We turn next to the time domain solutions.

So far we have followed the traditional (and Possios own original formulation in

the frequency domain where the main concern is to calculate the aeroelastic modes.

However, as we delve more into the problem of stabilityour main concernwe

need the time-domain solution.

For incompressible flow we simply set M D 0 in the Possio equation. Let us see

what the corresponding linear potential equation is.

Going back to (5.3) we set M D 0 therein. This yields

2

@2 @2 @2 @ @2 @2

C 2U 1 cos C sin a1

2

C C D 0:

@t 2 @t@x @t@z @x 2 @z2 @y 2

2

Hence dividing both sides by a1 and taking the limit, we obtain

@2 @2 @2

2

C 2 C 2 D 0;

@x @z @y

celebrated Laplace equation and time does not appear in it but does go into the

boundary conditions:

Far Field Velocity D i U 1 :

Flow Tangency

@1 .t; x; y; 0/

D wa .t; x/

@z

P y/ C .x a/.t;

D .1/h.t; P y/ C U1 cos .t; y/;

x; y 2 : (5.58)

5.4 Linear Possio Equation: Compressible Flow: M > 0 147

KuttaJoukowsky

1 .t; x; y/

A1 .t; x; y/ D (5.59)

U1

D 0 off the wing ! 0 as x ! b as defined before:

We note that we can go over to the finite plane model, with the nonzero angle of

attack making no difference except for replacing U1 by U1 cos , and small at

that.

The Mikhlin multiplier (5.14) becomes

q

1 1

!12 C !22 : (5.60)

2 C i !1 U1 cos

At the present time we have no solution technique for the corresponding Possio

equation.

Note that we get the beam or high aspect ratio solution if we set !2 D 0 which

we may consider as a first approximation and try a power series expansion

q

1 1 1 j!1 j 1 !22

!12 C !22 D 1C :. :

2 C i !1 U1 cos 2 C i !1 U1 cos 2 !12

Hence we specialize to the following:

Theory/Zero Angle of Attack

We can rewrite the potential field equation (5.43) now as simply that the divergence

of the flow is zero:

r:q D 0;

where q is the velocity vector and this is the defining property of incompressible

flow.

The potential flow is now 2D:

148 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

which is then the boundary; we call attention to the fact that it is a singular boundary

which makes all the difference and characterizes aeroelasticity.

We start with the Laplace Fourier transforms we have calculated. Thus

1 1 O i !/ejzj! ;

O1 .; i !; z/ D A.;

2 k C i!

O i !/ D .I C kL.0//.2T w

A.; O a .; :/

3

L.k; .I C kL.0//2T wO a .; :// O 7

C R1 q h5 ;

k t 2bCt

0 e t

dt

s r

Z

O 1 b x 1 k 2b C 1

h.x/ D e d;

bCx 0 xb

jxj < bk D=U1 : (5.61)

To get the time domain response we need to find the inverse Laplace/Fourier

transforms.

Theorem 5.18. Let u.t; :/ D 2T wa .t; :/, where wa .t; :/ is given by (5.54) and is in

C1 .b; b/, as before.

We assume that the initial structure state is zero so that

wa .0; :/ D 0:

Z 1

1

R1 q D e.t =U1 / c1 .t/dt

k 0 ek 2bC

d

0

Z 1

D et U1 c1 .tU1 /dt:

0

1

A.t; x/ D .t; x/ C P .t; s/ds;

U1 b

where

Z Z !

t b

.t; x/ D u.t; x/ C hc .t ; x/ .Pu.; s/ds d;

0 b

5.4 Linear Possio Equation: Compressible Flow: M > 0 149

where

R1 q p

hO ek 1 bCx

bx 2bC 1

xb d

hO c D

0

R1 q D R1 q (5.62)

k 0 ek 2bC

d k 0 ek 2bC d

Z t

hc .t; x/ D U1 h.U1 ; x/U1 c1 .U1 .t //d;

0

where s r

1 bx 2b C 1

h.; x/ D ; jxj < b:

bCx xb

Proof.

2 3

Z

O i !/ D .I C kL.0// 6

b

hO 7

A.; 4uO .; :/ C uO .; x/dx R q 5;

b 1 k t 2bCt

0 e t dt

Z b Z x

L.k; .I C kL.0//Ou.; :// D ek.bx/ uO .; x/ C k uO .; /d dx;

b b

where

Z b Z x Z b Z b

k.bx/

e k uO .; /ddx D uO .; /d ek.bx/ uO .; x/dx:

b b b b

Hence

Z b

L.k; .I C kL.0//Ou.; :// D uO .; /d:

b

Hence Z !

b

O i !/ D .I C kL.0// uO .; :/ C k

A.; uO .; x/dx hO c :

b

Let Z b

hO c

O .; :/ D uO .; :/ C k uO .; x/dx :

b k

150 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

Then Z x

1

kL.0/ O .; :/ D O .; /d:

U1 b

Hence taking inverse transforms we have:

Z b Z t

.t; :/ D u.t; :/ C u.t ; x/dxhc .; :/d;

b 0

Rx

which is (5.62). and kL.0/ O .; :/ is the transform of .1=U1 / b P .t; x/dx; which

proves the theorem. t

u

Next using this result, we can derive the solution to the potential equation obtained

a different way in [6].

Z b Z t

z 1

1 .t; x; z/ A.t ; y/ ddy: (5.63)

b 0 ..x y U1 /2 C z2 /

Proof. Use

Z 1

.k C i !/1 D U1 et i !t U1 dt;

0

Z 1

1 2z

ej!jz D ei !x dx; z > 0;

2 1 x2 C z2

Z 1 Z 1

ej!jz t i wx 1 z

D e e dxdt

.k C i !/ 0 1 .x U1 t/2 C z2

yielding

Z 1

O 1 .; i !; z/ D eix!

1

Z 1 Z t Z b

z 1

et dt A.t ; x y/ddydx:

0 0 b ..x y U1 /2 C z2 /

5.4 Linear Possio Equation: Compressible Flow: M > 0 151

Hence:

Z b Z t

z 1

1 .t; x; z/ D A.t ; y/ ddy: t

u

b 0 ..x y U1 /2 C z2 /

D ;

@x @x

@1 .t; x; z/ @1 .t; x; z/

D :

@z @z

@1 .t; x; z/

D i !:

@x

Our next case where we are able to obtain a closed-form solution to the Possio

equation is for M D 1, which is referred to as the sonic case.

Zero Angle of Attack

s

1 k 2 C 2ki ! O

O

.; i !/ D A.; i !/; (5.64)

2 k C i!

s

1 k 2 C 2ki ! 1 p k 1

.1; k/ D D . k/ 2 p : (5.65)

2 k C i! 2 k C i! k C 2i !

Then let as before L./ denote the operator to the multiplier 1=Ci !

corresponding

p

and L2 .k/ correspond to the multiplier 1=. k C 2i !/

Z x

f .s/

L2 .k/f D gI g.x/ D e.k=2/.xs/ p ds; jxj < b: (5.66)

b 2.x s/

152 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

Thus we have

p

P.k; 1/PA.; O :/ D k.2I k.k//L2 .k/A.;

O :/

D w.;

O :/: (5.67)

The Tricomi operator plays no role at all! .2 I kL.k// has a bounded inverse

but not of course L2 .k/. Thus we have

L2 .k/A.; O :/; (5.68)

k

where

1 k k

.2I kL.k//1 D IC L :

2 2 2

But

L2 .k/f D g

is essentially the Abel integral equation to find f given g. But we have to restrict g.

Thus following Tricomi [11] we require that g have an Lp .b; b/ derivative also.

This defines a Sobolev space; we denote it Wp1 . We can verify that the right side of

(5.68) is indeed contained in this space in as much as

O

w.; :/ is in C1 .b; b/:

Theorem 5.20. The Possio equation (5.67) has the unique solution

O :/ D p2 w

A.;

2

O a .; b/ C p L2 .k/.wO a 0.; :/ C k wO a .; ://: (5.69)

k k

Lemma 5.21. For g in Wp1 the integral equation L2 .k/f D g has a unique

solution in Lp .b; b/ given explicitly by:

where

Z x

k g.s/

f1 .x/ D p e..k.xs//=2/ p ds;

2 b xs

Z

1 x

g 0 .s/

f2 .x/ D p e..k.xs//=2/ p ds;

b xs

1 ek..xCb/=2 /

f3 .x/ D p g.b/ p :

xCb

5.4 Linear Possio Equation: Compressible Flow: M > 0 153

We note that

p 0 k

f1 C f2 D 2L2 .k/ g C g :

2

Proof. First we show that zero is not in the point spectrum of L2 .k/: For if

L2 .k/f D 0 for f in Lp .b; b/ by (5.61) we must have

Z x

h.s/

p ds D 0; jxj < b; where h.x/ D ek.x=2/ f .x/:

b xs

Next we follow the ingenious argument by Tricomi [11] to show that h. : / must

be zero.

Z x Z y Z x Z x

1 h.s/ 1

p p dsdy D h./ p dyd;

b x y b y s b .x y/.y /

Next (5.61) is also proved by Tricomi [11] and we can use the notation:

f D L2 .k/1 g

Next we go back to (5.68):

!

1 kL. k2 /

O :/ D L2 .k/

A.; 1

p IC O a .; :/:

w

2 k 2

!

kL. k2 /

gD IC f; then

2

f .b/ D g.b/

and

k

g0 C g D f 0 C kf

2

from which (5.69) follows. t

u

It is interesting to look at some special cases first.

154 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

Special Cases

is

p Z

2 ek..bCx/=2/ x

1

AO1 .; :/ D p p C2 k e.k=2/.xs/ p ds; jxj < b:

k 2.b C x/ b 2.x s/

Note that the solution does not converge as k ! 0, and goes to the value 2, as

Rek ! 1.2=M more generally).

This is consistent with our general theory for nonzero M .

2. Case 2:

wO a .; x/ D f2 .x/ D x; jxj < b;

left as an exercise.

To get the field equation we have only to make M D 1 in the general form (5.3):

@2 @2 2

2 @

2

C 2a1 a1 D 0: (5.71)

@t @t@x @z2

1 D xa 1 because U1 D a1 :

The boundary dynamics are the same as for every M , except that now

U1 D a1 .

Hence to get the time domain response we have only to take inverse Laplace

transforms in (5.70) which here unlike the case M D 0, where we needed the Sears

theorem, is more straightforward.

Here we merely state it, referring to [19] for more details as necessary.

Z t

g1 .t; x/ D 2 wa .t ; b/G.; x/d;

0

5.5 Linear Time Domain Airfoil Dynamics 155

1 .b C x/ .1=2/ .b C x/

G.t; x/ D p t ; t> ;

2.b C x/ 2 2

Z tZ x

1 .x / .1=2/ 0

g2 .x/ D 2 p wa .t ; /dd;

0 b 2.x / 2

Z Z

d t x 1 .x / .1=2/

g3 .t; x/ D 2 p wa .t ; /dd

dt 0 b 2.x / 2

Having calculated explicitly the pressure jump across the wing, for M D 0 and

M D 1 we now go on to consider the time domain structure response to the

aerodynamic lift and moment.

We begin with incompressible flow: M D 0:

It turns out that we can save a lot of tedious analysis by first taking the Laplace

transforms of the structure dynamics equations and waiting until the end before

inverting to the time domain. Here we follow [21] and consider Re > 0, unless

otherwise specified.

First note that now for us

O s/ C .x ab/.;

wO a .; x/ D h.; O O

s/ C U1 .; s/ jxj < b: (5.72)

normalize the half-chord b to be equal to 1. Thus we redefine k as

b

kD

U1

and define

O

w.; x/ D wO a .; bx/; jxj < 1;

which can be expressed as

O s/ C .k.x a/ C 1/.;

D U1 k h.; O s/; jxj < 1: (5.73)

156 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

Z 1

O

w.; x/ D PO .; x /A.;

O /d; jxj < 1;

1

O x/ D A.; bx/;

A.; jxj < 1:

So we may proceed with using our formulae for the solution just taking b D 1;

with k now being taken as b.=U1 /. In particular let AOi .; x/ be the solution

O

corresponding to w.; x/ D fi .x/; i D 1; 2 where

and Z 1

wij D fi .x/AOj .; x/dx; i D 1; 2I j D 1; 2: (5.74)

1

To calculate the lift and moment we take:

O s/ C .x ab/.;

wa .; x/ D h.; O O

s/ C U1 .; s/ jxj < b (5.75)

O s/ C .k.x a/ C 1/.;

D U1 k h.; O s/; jxj < 1; (5.76)

Z 1

O

L.; s/ D bU 1 O x/dx

A.;

1

k O

O s/ C .kw12 C .1 ka/w11 /.;

D bU1

2

w11 h.; s/ (5.77)

b

Z 1

MO .; s/ D b 2 U1 O x/dx

.x a/A.;

1

k O s/

D b 2 U1

2

.w21 aw11 /h.;

b

O

C .kw22 C .1 ak/w21 akw12 a.1 ak/w11 /.; s/ : (5.78)

5.5 Linear Time Domain Airfoil Dynamics 157

Thus for AO1 .; :/; we go back to (5.61) and note that where now

u.; :/ D 2T wO a .; :/

r Z s

1x 2 1

1C 1

u.; x/ D d

1Cx 1 1 x

r

1x

D2 ; jxj < 1

1Cx

and hence

Z 1

u.; x/dx D 2:

1

Hence

AO1 .; :/ D .I C kL.0//.u.; :/ C 2 hO c .; ://

and in turn

Z 1 Z 1

w11 D 2 C 2 hO c .; x/dx C k .1 x/.u.; x/ C 2 hO c .; x//dx;

1 1

where

Z 1

1

hO c .; x/dx D q 1

1 R1

k 0 ek 2C

d

1 1

D 1;

k .K0 .k/ C K1 .k//ek

where K0 .:/ : K1 .:/ are Bessel K functions of order zero and one.

Z Z r

1 1

1x

k .1 x/u.; x/dx D 2k .1 x/ dx

1 1 1Cx

D 3k:

And

Z 1

1

k .1 x/hOc .; x/dx D ek K1 .k/ ..K0 .k/ C K1 .k//ek / 1:

1 k

158 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

Using

Z 1

1 1

.1 x/3=2 p dx

1 1 C x.x 1 /

r r !

2C

D 1 1

2C

and

Z r ! , Z r !

1

k

p 2C 1

k t 2Ct

e .2 C / d e dt

0 0 t

R1 p

ek t . .2 C / /d 1

D 0

R1 q ;

k t 2Ct k

0 e t

dt

where

Z 1 p

ek . .2b C / /d

0

Z r !

1

k 1

D e .2b C / 1 d

0

Z 1 r !

d k 1

D e .2b C / 1 d; bD1

dk 0

d 1

D .ek .K0 .k/ C K1 .k//

dk k

D ek .K0 .k/ C K1 .k// C .ek /

K1 .k/ 1

K1 .k/ K0 .k/ C 2

k k

K1 .k/ 1

D ek 2:

k k

5.5 Linear Time Domain Airfoil Dynamics 159

Hence

1 1

w11 D 2 C 2 1 C 3k

k .K0 .k/ C K1 .k//ek

1

C 2 e K1

k

..K0 .k/ C K1 .k//ek /1 2k

k

D 2T .k/ C k;

where

K1 .k/

T .k/ D

K0 .k/ C K1 .k/

is called the Theodorsen function [6].

We may note also that

r

1x

AO1 .; x/ D 2

1Cx

r Z r !

1

1x 2 C k d

C 2 e ek .K0 .k/ C K1 .k//

1Cx 0 x1

Z r Z r

x

1s x

1s

C 2k ds C 2k ds

1 1Cs 1 1Cs

Z 1r

2 C k d

e .ek .K0 .k/ C K1 .k///:

0 s1

Note that in this form we can show

O x/ D 0:

Ltx ! 1 A.;

Next Z 1

w21 D x AO1 .; x/dx D T .k/

1

omitting the details of the calculation, which are similar to the previous case.

We need next

AO2 .; x/:

Here

p

u.; x/ D 2 1 x 2 ;

160 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

Z 1

u.; x/dx D :

1

Hence

p

AO2 .; x/ D 2 1 x 2

r Z r !

1

1x 2 C k d

C e ek .K0 .k/ C K1 .k//

1Cx 0 x1

Z x p

C 2k 1 s 2 ds

1

"Z r Z r

#

1

x

1s 2 C k d

C k ds e e .K0 .k/ C K1 .k// :

k

1 1Cs 0 s1

w12 D T .k/;

k

w22 D 1 C T .k/ ;

2 4

w11 D k C 2T .k/;

w21 D T .k/: (5.79)

b

kD :

U

Next let us calculate the lift:

O k O s/

L.; s/ D bU 1 2

.k C 2T .k//h.;

b

O

C .kT .k/ C .1 ak/.k C 2T .k///.; s/

O s/ C .1 ak/k O .; s/

D bU 21 k 2 h.;

b

O s/ C b 1 a O .; s/U1 .;

C T .k/.2bU 1 / h.; O s/

2

5.5 Linear Time Domain Airfoil Dynamics 161

O s/ C 2 b 3 a.;

D 2 b 2 h.; O O

s/ b 2 U1 .; s/

O 1 O O

T .k/2bU 1 h.; s/ C b a .; s/ C U1 .; s/ :

2

(5.80)

Note the appearance of terms containing 2 , and the term containing the

Theodorsen function. We note first that

Z 1

dt

K0 .k/ D ek t p ;

1 t2 1

K1 .k/ D K00 .k/

Z 1

tdt

D ek t p :

1 t2 1

Hence

Z 1

r

k t t C1

K0 .k/ C K1 .k/ D e dt;

1 t 1

Z 1

r

k t t C2

e .K0 .k/ C K1 .k// D

k

e dt;

0 t

Z 1

p

ek K0 .k/ D ek t .1=.t.t C 2///dt;

0

Z 1

p

ek K1 .k/ D ek t ..t C 1/=. .t.t C 2////dt

0

and

T .k/

D ek K1 .k/=ke k .K0 .k/ C K1 .k//

k

Z 1

D ek t L.t/dt;

0

where

Z t

C1

L.t/ D c1 .t / p d t > 0;

0 . C 2/

known as the Wagner function [6], where

162 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

and Z 1

c1 .t/dt D 1

0

because

1=.k.K0 .k/ C K1 .k//ek / ! 1 as k ! 0:

Here it is known also (KussnerSears [6]) that

and we see so is

t C1

p :

t.t C 2/

Hence L.t/ does not go to zero as t goes to zero!

To find the limit let us note that T .k/ has the following expansion.

At zero

C .EulerGamma2 2EulerGamma log2

C log22 C 2EulerGamma logk 2log2logk

C logk2 /k 2 C Ok3

showing the logarithmic essential singularity along the negative real line.

Also

kT 0 .k/ ! 0 as k ! 0:

At infinity:

3

1 1 1 1

T .k/ D C 2

CO :

2 8k 16k k

Note that

kT 0 .k/ ! 0 at infinity:

The inverse Laplace transform of T .k/ contains a delta function at the origin and

hence

T .k/ 1

k goes to D L.0C/ as Re k ! 1:

k 2

Now

1 1

T .k/ D .K1 .k/ K0 .k//=.K1 .k/ C K0 .k//

2 2

1

D k.ek K1 .k/ ek K0 .k//=.ke k .K1 .k/ C K0 .k///

2

5.5 Linear Time Domain Airfoil Dynamics 163

and

Z 1

r

t k t

k.e K1 .k/ e K0 .k// D

k k

ke dt

0 t C2

Z 1 r

k t d t

D e dt

0 dt t C 2

Z 1

p

D ek t 1=. .t.t C 2/3 // dt:

0

Hence Z 1

1

T .k/ D ek t `.t/dt;

2 0

where Z t

p

`.t/ D c1 .t /1=. .. C 2/3 //d;

0

where Z 1

1

`.t/ 0I `.t/dt D :

0 2

Hence we have that

Z t

1

L.t/ D C `.t/dt;

2 0

L.1/ D1 D T .0C/;

Z 1 Z 1

1

ek t `.t/dt Dk ek t L.t/dt : (5.81)

0 0 2

Hence

as jkj ! 0; Re k > 0;

Z 1

k ek t L.t/dt ! 0: (5.82)

0

k O s/

MO .; s/ D U1

2 2

b .T .k/ a.k C 2T .k// h.;

b

k

C k 1 C T .k/ .1 ak/T .k/ akT .k/

2 4

O

a.1 ak/.k C 2T .k// .; s/

164 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

O 1 O

D b a h.; s/ b a C

3 2 4 2

2 .; s/

8

C.U1 b 3 / O

a .; s/ C T .k/U1 2 2

b

2

k k O O

2a h.; s/ C k a.1 ak/2 .; s/ ;

b b 2

where the last term containing the Theodorsen function can be expressed as

kO O

2 2

T .k/U1 b h.; s/ C k .; s/

b 2

O O kO

2a .; s/ C a2 k2 .; s/ 2a h.; s//

b

1 O s/Cb.12a/.;

O O

D T .k/U1 b 2 Ca 2h.; s/C2U1 .; s/ :

2

O

w.; O s/ C b.1 2a/.;

s/ D 2h.; O O

s/ C 2U1 .; s/

P s/ C b.1 2a/.t;

w.t; s/ D 2 h.t; P s/ C 2U1 .t; s/; t > 0:

O

L.; s/ D 2 b 2 h.; O

O s/ C 2 b 3 a.; O

s/ b 2 U1 .; s/

T .k/bU 1 w.;

O s/

O s/ b 4 a2 C 1 2 O .; s/ C .U1 b 3 /

MO .; s/ D b 3 2 a h.;

8

O 1

a .; s/ C T .k/U1 b

2

C a w.;

O s/

2 2

O 1

D b a h.; s/ b a C

3 2 4 2

2 O .; s/

8

b 2 U1 O

O s/ C b 2 U 2 .;

O

w.; s/ C b 2 U1 h.; 1 s/

2

1

CT .k/U1 b

2

C a w.;

O s/:

2

5.5 Linear Time Domain Airfoil Dynamics 165

Using

1 O

.U1 b 3 / a .; s/

2

b 2 U1 O s/ C b.1 2a/.;

O

D .2h.; s/

2

O

C2U1 .; 2 O

s// C b 2 U1 .; s/

b 2 U1 b 2 U1 O

O s/ C b 2 U 2 .;

D O

w.; s/ C 2h.; 1 s/:

2 2

We recall now our Hilbert space formulation of the structure dynamics in Chap. 2

with the Hilbert space H as therein and M and A defined as therein along with the

notation:

h.t/

x.t/ D ;

.t/

L.t/

R C Ax.t/ D

M x.t/ :

M.t/

We take Laplace transforms on both sides, with x.0/ D 0 and using the notation

O

x./ for the Laplace transform we have

O

L.; :/

2

O

M x./ C Ax./

O D O :

M .; :/

Substituting for

O

L.; :/

O

M .; :/

and collecting terms containing 2 and we have

2 Mx./

O C Dx./

O C Kx./

O C Ax./

O D FO ./x.;

O :/; (5.83)

where

!

m C b 2 s b 3 a

MD ;

S b 3 a I C b 4 .a2 C 18 /

166 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

1 1 2

b 2 b 4 a2 C .b 3 a/2 D b 3 :

8 8

Next

!

0 b 2 U1

DD ;

b 2 U1 0

!

0 0

KD ;

0 b 2 U1

2

0 T .k/

1

k bU1

FO ./ D @ T .k/

b 2 U1

A:

k

U1 b 2 1

2

C a 2

We have thus obtained the structure dynamics under aerodynamic loads in the

Laplace transform domain.

The next step is to take the inverse Laplace transforms. Here the only slight

complication is that the inverse transform of T ./ contains a delta function. Thus

in taking the inverse transform of

we note that we are already assuming that wa .t; :/ is differentiable in t and hence

we may write:

T .k/

O

T .k/w.; s/ D k wO a .; s/

k

b T .k/

D .wO a .; s//

U1 k

Z t

b

L.t /wP a .; s/d;

U1 0

Z t

p

L.t/ D c1 .t /. C 1/=. .. C 2///d:

0

5.5 Linear Time Domain Airfoil Dynamics 167

Or as Z t

b 1

wa .t; s/ C `.t /w.; s/d :

U1 2 0

The first form is preferred traditionally [6] and so we use it for the time domain

version.

Hence with !

x.t/

Y .t/ D

P

x.t/

with range in H H and defining

8 9

0 >

>

< =

2U1

BD

2 >

>

: ;

b .1 2a/

B1

D ;

B2

we have

wa .t; :/ D B Y .t/ D B1 x.t/ C B2 x.t/:

P

We can now rewrite the structure equations in the time domain as

Z t

Mx.t/

R C Dx.t/

P C Ax.t/ C Kx.t/ D F .t /B YP ./d;

0

where !

bU1 L.U1 t/

F .t/ D :

b 2 a C 2l U1 .L U1 t/ 12

As usual we convert this to a first-order equation in time:

Z t

YP .t/ D AS Y .t/ C T Y .t/ C L.t /YP ./d

0

0 I

As D

M1 A 0

0 0

T D

M1 K M1 D

168 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

(not to be confused with the Tricomi operator) the subscript S denoting structure.

The operator T reflects the so-called non-circulatory terms in the literature; see [6].

0 1

0

@L.t/Y D 0 A:

M1 F .t/B Y

We now modify the energy space H using M in place of M . Thus modifying (5.3)

we have:

The energy space H consists of elements of the form:

x

Y D 1 ;

x2

p

x1 2 D. A/; x2 2 H

endowed with the energy inner product:

p p

Y; ZE D Ax1 ; Az1 C Mx2 ; z2 ; (5.84)

where

z D z1

z2 ;

p

z1 2 D. A/; z2 2 H:

Note that if we set U1 to be zero, we get back to the pure en vacuo structure

equations of Chap. 2 without the aerodynamic forcing terms.

Hereinafter we simply continue to use ; and drop the subscript E, because the

context should make clear which inner product is being used. Also we use:

x h1 h2

Y D 1 I x1 D I x2 D :

x2 1 2

A D As C T

as is well known [11, 16]. Let us denote the semigroup by S.t/; t 0. Unlike the

semigroup generated by As , this semigroup is not necessarily a contraction. In fact

we have:

ReAY; Y D ReT Y; Y ;

because

ReAS Y; Y D 0I ReDx2 ; x2 D 0

5.5 Linear Time Domain Airfoil Dynamics 169

and

ReT Y; Y D ReKx1 ; x2 D CU1

2 2

b Re1 ; 2 ;

which shows that the semigroup S./ may not be a contraction.

But the growth bound is of course finite because the generator is a perturbation of

As by a bounded operator; and the precise growth bound is not needed here anyway.

What is important, however, is that the resolvent of A, denoted R.; A/ is

compact because the resolvent of As is compact.

What can we say about the convolution part?

The properties of L./ are readily obtained from those of the numerical function

L./. We have seen that

1

L.0/ D I L.1/ D 1

2

and the derivative of L.t/ is given by `.t/ which is continuous and is integrable

0; 1: L./ is nonnegative and

Z 1

ek t L.t/dt < 1 Re k > 0:

0

Hence we define Z 1

O

L./ D et L.t/dt Re > 0:

0

Again thus defined, its properties depend upon .T .//= which is not defined along

the negative real axis, including zero, and can be extended analytically to the whole

plane, except the negative real axis and D 0. On the other hand L./ O which

involves only T ./ is now defined at D 0 by continuity. Hence L./ O is defined

on the whole finite plane excepting < 0:

The calculation of the aeroelastic modes is the central part of Flutter analysis,

which is really part of the spectral theory of the stability operator.

M D 0I D 0:

But before we prove existence and uniqueness of the solution of the

convolution/evolution equation we need to first study spectral theory, which also

plays a crucial role in stability theory.

Thus we go back to the Laplace transform theory.

We have: for every except the negative real axis:

2 Mx./

O C Dx./

O C Kx./ C Ax./

O D FO ./x.;

O :/;

170 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

Z t

YP .t/ D AS Y .t/ C T Y .t/ C L.t /YP ./d

0

O

.I A L.// YO ./ D .I L.//Y

O .0/;

Z 1

YO ./ D et Y .t/dt

0

and

O

L./ O

D M1 F./B

;

which is defined in the whole plane except for Re < 0. Hence, following

semigroup theory techniques, we consider the generalized resolvent equation:

O

.I A L.//Y DZ

in H:

Exploiting the compactness of R.; AS / we have the next theorem.

Theorem 5.22. Given any in the resolvent set of As (not an en vacuo structure

mode in other words), and omitting the negative half-line, either

O

.I A L.//Y D0 Y 0

O

for some nonzero Y or .I A L.// has a bounded inverse, which we call the

generalized resolvent.

Proof.

O

I A L./ O //:

D I As .T C L.Y

Hence

O

R.; AS /.I A L.//Y O

D .I R.; AS /. C L.///Y;

where

O

R.; As /.T C L.//

is compact because R.; As / is.

Hence either

O

.I R.; AS /.T C L.///Y D0

5.5 Linear Time Domain Airfoil Dynamics 171

O

.I A L.//Y D0

or

O

I R.; AS /.T C L.//

and equivalently

O

.I A L.//

has a bounded inverse. t

u

We call the bounded operator

O

.I A L.// 1

Note, however, that R./ does not satisfy the resolvent equation [10, 41].

Aeroelastic Modes

The zeros of I A L./ O are called the aeroelastic modes. They are not

identified as eigenvalues at the expense of changing the space, as we show later.

The zeros of the adjoint:

O

.I A L.//

Y D0

The main objective in flutter analysis is to calculate these modes and the

corresponding mode shapes Y .:/.

We begin with M D 0: Here we recall the analysis in Chap. 2, including now the

aerodynamic loading. Thus solving

O

.I A L.//Y D0

O

L.; :/

2

O

M x./ C Ax./

O D O ;

M .; :/

172 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

which unravels to

1 h 2 O i

hO .; s/ D O

0000

.mh.; s/ C S .; O

s// L.; s/ ;

EI

1 h 2 i

O .; s/ D O s// MO .; s/ ; 0 < s < `

00

.I C S h.;

GJ

with the end conditions:

O 0/ D 0I

h.; hO .; `/ D 0I

000

hO .; `/ D 0;

00

O

.; 0/ D 0I O

.; `/ D 0:

This is recognized as a two-point boundary value problem for ordinary differen-

tial equations for each .

0 1

O s/

h.;

B hO 0 .; s/ C

B C

B O 00 C

B h .; s/ C

Y .; s/ D B O 000 C:

Bh .; s/ C

B C

O

@ .; s/ A

O 0 .; s/

d

y.; s/ D A./Y .; s/;

ds

where now (cf. (2.13)) 0 1

0 10 0 0 0

B0 0C

B 01 0 0 C

B C

B0 00 1 0 0C

A./ D B C;

Bw1 00 0 w2 0C

B C

@0 00 0 0 1A

w3 00 0 w4 0

5.5 Linear Time Domain Airfoil Dynamics 173

where now:

1

w1 D .m2 C U1 bw11 /;

EI

1 h 2 i

w2 D S C bU 21 ..1 ak/w11 C kw12 / ;

EI

1 h 2 i

w3 D S C bU 21 .w21 aw11 / ;

GJ

1 h 2 i

w4 D I C b 2 U1

2

.w21 C kw22 a.1 ak/w11 ak.w21 C w12 //

GJ

and is defined for every except the negative half-line. The wij are given in (5.79)

for M D 0.

The solution is given by

y.; 0/ D Qz./;

and

P y.; `/ D P e `A./ Qz./; z./ 0: (5.85)

Hence we must have

0 1

001000

B C

P D @0 0 0 1 0 0A;

00000 1

0 1

0 00

B0 0 0C

B C

B C

B1 0 0C

QDB

B0

C

B 1 0C

C

B C

@0 0 0A

001

D P :

174 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

Theorem 5.23. The function

d.0; ; U /

for each U positive is an analytic function of except for a branch cut along the

negative real axis, logarithmic singularity along Re 0, with at most a countable

number of isolated zeros in a half-plane that do not have an accumulation in the

finite part of the plane.

The aeroelastic modes are the roots of

d.0; ; U1 / D 0:

Proof. The analyticity properties are inherited from those of the matrix function

A./ and in turn by those of T .k/; and ultimately the analyticity properties of the

Bessel K functions. Hence the stated analyticity properties follow. t

u

Let us explore the aeroelastic modes.

Let

x1 h1 h2

Y D ; x1 D ; x2 D I Y:Y D 1:

x2 1 2

Then

O

.I A L.//Y D0

or, equivalently:

x1 D x2 ;

Mx2 D Ax1 Kx1 Dx1

0 1

2

T .k/U1

C@ b 2 U1 A .2U1 1 C2h2 Cb.12a/2 /;

T .k/b.aC 12 /U1

2

2

where

h2 D h1 I 2 D 1 :

We begin with the special case D 0; of interest because it is an unstable mode

even though it is not a structure mode.

Theorem 5.24. Zero is an aeroelastic mode for a sequence Un of values of U1

given by p p

Un D .2n C 1/. GJ /=.2b` ..1 C 2a///:

5.5 Linear Time Domain Airfoil Dynamics 175

Proof. We have:

x2 D 0;

3

2U1 1

0D Ax1 :

2b .a C 12 / U1 31

Or

!

EI h1 0000 .s/ 2

2U1 1 .s/

D ;

GJ 1 00 .s/ 2b.a C 12 / U1

2

1 .s/

where we see that

1

GJ 1 00 .s/ D 2b a C 2

U1 1 .s/ 0<s<`

2

U1 given by

p

p

Un D .2 n C 1/ GJ .2b` ..1 C 2a/// ;

s

1 .s/ D const sin 0 < s < ` for all n: t

u

2`

The minimal value is called the divergence speed. See Sect. 5.5; of course the speeds

are also determined from solving

d.0; 0; U / D 0;

number of zeros.

Theorem 5.25. The aeroelastic modes are confined to a finite vertical strip in the

complex plane; the real part is bounded.

Proof. Let be an aeroelastic mode and Y the mode shape, where we normalize so

that

Y; Y D 1:

176 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

Hence

O

Y; Y D D AY; Y C L./Y; Y

O

D 2Rex1 ; Ax 1 jj2 Dx1 ; x1 jj2 Kx1 ; x1 C L./Y; Y ;

where:

x1

Y D ;

x1

Y; Y D x1 ; Ax 1 C jj2 Mx1 ; x1 D 1

Dx1 ; x1 D 0I Kx1 ; x1 D b 2 U1

2

1 ; 1

Re.1 C 2x1 ; Ax 1 C b 2 U1

2

1 ; 1 /

O

D ReL./Y; Y ;

where the right-hand side goes to zero as Re ! 1. Hence Re is bounded. The

aeroelastic modes are confined to a vertical strip in the complex plane. Furthermore

there can only be a finite number of zeros with positive real part [8]. Indeed if we

have a sequence of the form: k D C i !k and !k ! 1 we would contradict the

asymptotic properties of the roots. t

u

Mode Shapes

component xk;1 of Yk . : /:

xk; 1

Yk D ;

xk; 2

hk .s/

xk;1 D ; 0 < s < `:

k .s/

Let us calculate Yk .

O k/ D 0

k Yk AYk k L.

and hence we have:

xk;2 D k xk;1 ;

2k Mxk;1 D Axk;1 Kxk;1 k Dxk;1

5.5 Linear Time Domain Airfoil Dynamics 177

0 1

k b 2

B T U k C

Uk

CB

@ k b C

1 k b Uk A

2

T b a C Uk

2

Uk 2 2

.2U1 k C 2 hk C b.1 2a/k k /:

Recall that

Det:P e `A.k / Q D 0: (5.86)

Hence there is a nonzero zk such that

Note that the dimension of the nullspace of the matrix is at most two. Define

k .s/; k .s/; k .s/;

zk .0/ D Qzk :

Next define

P e`A.k / Qzk D 0;

0 1

10 00 00

B0 0 00 1 0C

T46 ./ D B

@ 0

C:

00 0 0A

00 00 0

xk;1

Yk D :

xk;2

178 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

Generalized Resolvent

The generalized resolvent equation

O

I A L./ Y D Yg ;

where

x1;g hg . : /

Yg D I x2;g D ;

x2;g g . : /

xl

Y D (5.88)

x2

x1 x2 D x1;g ;

1 h 2 O i h .s/

hO 0000 .; s/ D O O g

mh.; s/ C S .; s/ L.; s/ C ; (5.89)

EI EI

1 h 2 i

O s/ MO .; s/ C g .s/ ;

O 00 .; s/ D I C S h.;

GJ GJ

0 < s < `; (5.90)

d

y.; s/ D A./y.; s/ C yg .s/; (5.91)

ds

hg .s/ g .s/

yg .s/ D col 0; 0; 0; ; 0; ;

EI GJ

y.; 0/ D Qz./;

Z s

y.; s/ D e sA./

Qz./ C e.s /A./ yg ./d; (5.92)

0

Z

`

P e`A./ Qz./ C v./ D 0I v./ D e.` /A./ yg ./d;

0

or,

D./z./ D P v./I D./ D P e `A./ Q:

So, omitting the aeroelastic modes this has the solution:

5.5 Linear Time Domain Airfoil Dynamics 179

and in turn:

Z s

1

y.; s/ D e sA./

QD./ P v./ C e.s /A./ yg ./d;

0

0<s<` (5.93)

x1 ./

R./Yg D ;

x2 ./

x1 ./ D T26 y.; :/I x2 ./ D x1 ./ x1;g ; (5.94)

where

h .; s/

T26 y.; s/ D 0 < s < `:

.; s/

Let us now get back to the solution to the initial value problem:

O

I .A/ L./ O

YO ./ D I L./ Y .0/;

YO ./ D R./ I L./

O Y .0/; k :

Because D./ has only simple zeros at k , it follows that R./ has only isolated

simple poles at k , confined to a finite strip.

Hence the inverse Laplace transform

Z L

1

W .t/ D lim et . Ci !/ R. C i !/d!; t > 0:

L!1 2 L

Following [33, 44] we deform the contour of the inverse Laplace transform to

obtain [8]

1 Z 1 !

X

rt

W .t/Y D Pk Y e C

k t

e J .r/Y dr

kD1 0

180 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

I

1

Pk Y D R./Y d; D k C rei ; r small enough

2 i

O k / Pk Y D 0:

k I A k L.

k I A k L.O k/ :

only l and we may assume this for simplicity.

Next note that the generalized resolvent is defined for every except fk g and

the negative half-line. We also exclude k D 0.

Then with r > 0 we have to consider the integral along the bounding lines:

D r C i ! and D r i !;

can be calculated in many ways, our main interest being in the limit as r goes to

zero.

We now define

1

J .r/Y D lim R.r i !/ R.r C i !/Y; r > 0;

2 i !!0

where we need to first show that the limits exist. Now

where

0 1

0

L.r C i !/Y D @ 0 A;

Ml FO .r C i !/B Y

where

0 T .k/ 1

bU

B k C

FO .r C i !/ D @ T .k/ 1 b 2 U A ;

2

b U Ca

k 2 2

5.5 Linear Time Domain Airfoil Dynamics 181

k D .r C i !/:

Note the appearance of the scale factor b=U for the first time.

We need next to calculate the jump. Now

z 2m

1

X

K0 .z/ D I0 .z/logz C I0 .z/log2 C 2 :

mD0

.m/2

So in the definition we need to go to the Riemann sheet where the value of the

logarithm jumps by 2 i every time we cross the negative half-line. Using for r >

0; ! > 0,

p !

logr C i ! D log r 2 C ! 2 C i. /; Tan D ;

r

p

logr i ! D log r 2 C ! 2 i. /;

logr C i ! logr C i ! D 2 i 2i;

1

,

X 2k

C .r C i !/ 2k

2 .k/2 .k C 1/;

kD0

We can define:

Similarly:

Let

182 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

D

K1 .r/ C i I1 .r/ C K0 .r/ C i I0 .r/

:

K1 .r/ i I1 .r/ C K0 .r/ i I0 .r/

And

lim ! ! 0 T .r C i !/ D D TC .r/;

K1 .r/ i I1 .r/ C K0 .r/ i I0 .r/

lim ! ! 0 T .r i !/ D C JT .r/:

K1 .r/ i I1 .r/ C K0 .r/ i I0 .r/

Then

1

J .r/Y D lim R.r i !/ R .r C i !/Y; r >0

2 i !!0

1

D lim ..r i !/I A .r i !/L.r i !//1 Y

2 i !!0

..r C i !/I A .r C i !/L.r C i !//1 Y

1

D .rI A C rL .r//1 .r.L .r/

2 i

LC .r///.rI A C rLC .r//1 Y;

where

LC .r/ D lim ! ! 0 L.r C i !/:

We are most interested in the limit as r ! 0, because of the familiar relation for

Laplace transforms: following (5.20)

Z 1 Z 1

rt

lim t ! 1 t e J .r/Y dr D eR J .R=t/Y dR

0 0

! J .0C/Y as t ! 1

and

J .0C/Y D 0:

Also we have: Z 1

1

X Pk Y J .r/

R./Y D C Y dt:

k 0 Cr

kD1

5.5 Linear Time Domain Airfoil Dynamics 183

Z 1

QY D J .r/Y dr;

0

we have that

lim R./Y D W .0/Y D Y

Re !1

and hence

1

X

Y D Pk Y C QY;

kD1

where

QP k D 0 D Pk QI Q2 D Q:

The mode shape vectors do not span the space H, nor are they orthogonal.

Evanescent States

What is novel here is the existence of evanescent states that decay faster than any

modal response, a phenomenon that becomes apparent only in continuum theory

and lost in the discretized models of CFD.

The evanescent states are the states in the range space of the operator Q. We note

that the response to evanescent states is given by

Z 1

W .t/QY D ert J .r/QY dr

0

and are thus Laplace transforms in t, and thus are stable for all values of U . In

[33] the structure is discretized and the aerodynamics is also approximated (Pade

approximation) and the authors call this part the nonrational component. Here we

see the exact solution and in particular it is stable for all values of the speed U .

Next we derive the time domain solution of the convolution/evolution equation

using state space theory.

184 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

We are ready to consider now the time domain solution of the convolution/evolution

equation for given initial conditions. The solution given in [8, 22] involves a fair

degree of operator theory.

Here we use a different and more self-contained technique: state space represen-

tation which is of interest on its own for us; but requires some knowledge of the

theory of semigroups of operators.

Here we follow [16]. Our time domain equation is the initial value problem in H:

Z t

YP .t/ D AS Y .t/ C T Y .t/ C L.t /YP ./d (5.95)

0

We show that this can be expressed in the state space form:

P

Z.t/ D AZ.t/;

Y .t/ D PZ.t/ (5.96)

a projection operator.

We make strong use of the fact that L.:/ is in C 0; 1 (the space of

continuous functions with finite limit at 1) and is absolutely continuous therein. In

particular:

L. : / 2 ;

where is the Banach space C 0; 1 of continuous 4 by 4 matrix-valued functions.

It is also absolutely continuous with derivatives therein.

Z t Z t

L.t /YP ./d D P /Y ./d L.t/Y .0/

L.t

0 0

but this makes it awkward, bringing in the initial value in the state equation which

we want to avoid!

We begin with the convolution part which is defined by the function L. : /.

Consider the system inputoutput relation where u . : / is the input and v. : / is

the output

5.6 State Space Theory 185

Z t

v.t/ D L.t /u./d t > 0; (5.97)

0

RT

where the input u. : / takes its values in a separable Hilbert space and 0 jju.t/jjdt <

1 over every finite T . For representation theory, it is convenient to embed this class

of inputs in a larger class where u. : / is defined on .1; 1/ but vanishes on a

negative half-line:

Z t

v.t/ D L.t /u./d:

1

Then of course we can extend this definition to any t, not necessarily positive. But

we are only interested in the t > 0 with which we started. The implication is that we

embed this in a controllable system [16] but this is irrelevant in our current context.

Theorem 5.26. Let X denote the Banach space C 0; 1I H . The inputoutput

relation Z t

v.t/ D L.t /u./d t >0

1

can be represented as

v.t/ D C x.t/;

P

x.t/ D Ax.t/ C Lu.t/ (5.98)

for t > 0; where x.t/ is in X for each t > 0; and A is the infinitesimal generator of

a C0 semigroup over X , and L is a linear bounded operator on H into X , and C is

a linear bounded operator on X into H .

Proof. Define L by

Lu D xI

x.t/ D L .t/u; t 0

t

u

and in as much as

jj x.t/ jj jj L .t/ jj jjujj;

it follows that x is in X .

Next define C on X by

C x D x.0/

186 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

jj x.0/ jj jjxjj:

defined by

D.A/ D x. : / in X such that x 0 . : / is in X ;

Ax D yI y.t/ D x 0 .t/ t 0:

Next let denote the closed linear subspace generated by elements in X of the

form:

S.t/Lu u in H and t 0:

Then is an invariant subspace for the semigroup: S.t/ is contained in for

every t: We may take our state space to be and let Sc . : / denote the semigroup

restricted to , and let Ac denote the generator, being then a restriction of A.

Then we have the next lemma.

Lemma 5.27. The nonhomogeneous Cauchy problem:

P

x.t/ D Ac x.t/ C Lu.t/ t >0

jjx.t/ x.0/jj ! 0 as t ! 0C

given by Z t

x.t/ D Sc .t/x.0/ C Sc .t /Lu./d:

0

Proof. The proof is standard; see, for example, [16]. t

u

Next let us calculate v. : /. We have

Z t

C x.t/ D CS c .t/x.0/ C CS c .t /Lu./d

0

Z t

D x.0; t/ C L.t /u./d

0

D v.t/

5.6 State Space Theory 187

x.t; s/ s 0

to denote x.t/.

Remarks: As is well known (see [16]) the method of variation of parameters

solution holds in the generalized or weak sense even if x.0/ is not in the domain

of the generator, with the continuity at the origin, but of course the solution need

not be in the domain of the generator.

Resolvent of Ac

Let R.; Ac / denote the resolvent of Ac . Then for Re > 0 we have

Z 1

R.; Ac /x D et Sc .t/xdt

0

and

.I Ac /x D LY; Y in H

has the unique solution for Re > 0 given by

x D R.; Ac /LY

and Z 1

Cx D C O

et Sc .t/LY dt D L./Y:

0

and

O

C x D L./Y

O

for in the region of analyticity of L./:

Proof. For Re > 0

Z 1

x./ D et Sc .t/LY dt

0

is an element of .

188 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

Z 1

x.; s/ D et L.t C s/Y dt s0

0

Z 1

D e s

e L./Y d

s

Z s

O

D es L./Y es e L./Y d (5.99)

0

0 s 1;

x.; 1/ D L.1/Y; (5.100)

O

x.; 0/ D C x.; : / D L./Y: (5.101)

The main point here is that thus defined for Re > 0 x.; s/ can be continued

O

analytically as an analytic function of in the region of analyticity of L./Y

which includes the whole plane except 0. Hence x.; : / can be continued

analytically satisfying

.I Ac /x.; : / D LY (5.102)

as required. t

u

Let us use x. to denote the state space, the product space xH which is then a

Banach space and denote the elements therein as

x

ZD x"; Y "H;

Y

PZ D Y:

XP .t/ D Ac x.t/ C LYP .t/: (5.103)

x.t/

Z.t/ D

Y .t/

and defining the linear bounded operator B on X. into itself by

LY

BZ D ;

0

5.6 State Space Theory 189

we have:

P Ac 0 P

Z.t/ D Z.t/ C B Z.t/:

C A

Or

P Ac 0

.I B/Z.t/ D Z.t/:

C A

.I B/ D .I C B/1

P Ac 0

Z.t/ D .I C B/ Z.t/

C A

D A.Z.t/; (5.104)

where

Ac 0 Ac C LC LA

A. D .I C B/ D :

C A C A

P

Z.t/ D A.Z.t/; (5.105)

where Z. : / is the state, the state space is X.; and A. is the system

stability operator. To prove existence and uniqueness of the convolution/evolution

equation (5.95) we only need to show that A. generates a C0 semigroup.

We begin with the spectrum of A.. In turn, we begin with the point spectrum

(eigenvalues).

The point spectrum of A. consists of such that

x D .Ac C LC /x C LAY;

Y D C x C AY:

190 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

Suppose

Y D 0:

Then

Cx D 0

and hence

x D Ac x:

But then

x.s/ D es C x D 0:

Hence Y cannot be zero.

Suppose

x D 0:

Then we must have

LAY D 0

and the second equation yields

Y D0

and hence neither x nor Y can be zero.

Next, the first equation can be rewritten as:

x Ac x D LY

and can have a solution even if is not in the resolvent set of Ac , and the second as

Y AY D C x;

where

O

C x D L./Y;

so that

O

Y AY D C x D L./Y:

Or

O

Y AY L./Y D0

and thus is an aeroelastic mode.

We can say more.

Theorem 5.29. The point spectrum of the system stability operator A. is the set of

all aeroelastic modes.

Proof. Let be an aeroelastic mode, defined by

O

.I A L.//Y D 0; Y 0:

5.6 State Space Theory 191

Or

O

.I A/Y D L./Y:

O

Recall that L./ is defined and analytic in the whole plane except the

negative real axis.

Let us next consider the equation in ,

x Ac x D LY I

O

C x D L./Y:

O

.I A/Y D L./Y:

Hence let

x

ZD :

Y

x D Ac x C LY

D Ac x C L.AY C C x/;

Y D AY C C x: t

u

Resolvent of A.

\

D fresolvent set of Ag

\

fComplement of the set of aeroelastic modesg fRe > 0g:

Then

is contained in the resolvent set of A.:

192 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

Moreover, for in

O

.I R.; A/L.// has a bounded inverse.

Define 1

O

R./ D I R.; A/L./ R.; A/:

x

ZD ;

Y

R.; AC / .L LY C x/

R.; A./Z D ;

where

O

D R./.Y C CR.; Ac /x L./Y /:

Proof. Note first that includes a right-half-plane. For in , we note that

R.; A/ is compact, and hence so is

O

R.; A/L./:

Hence either

O

I R.; A/L./ Y D 0; Y 0;

or

O

I R.; A/L./

O

.I R.; A/L.//Y D0

O

.I A L.//Y D 0;

or is in the point spectrum of A.: Hence

O

.I R.; A/L.//

O

I A L./

5.6 State Space Theory 193

1

O

R ./ D I R.; A/L./ R.; A/

.I A./Z D Zg

for given

xg

Zg D :

Yg

Let

x

ZD :

Y

Then

x Ac x L.AY C C x/ D xg ;

Y AY C x D Yg :

O

CR.; Ac /L D L./:

Hence

O

C x D CR.; Ac /xg C L./Y O

L./Y g;

O

Y AY L./Y O

D Yg C CR.; Ac /xg L./Y g:

Hence

O

Y D R./ Yg C CR.; Ac /xg L./Y g ;

u

194 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

Remarks: The resolvent above not being compact implies that the resolvent set is

properly contained in the complement of the point spectrum.

Note also that

0 O

PR.; A./ D R./.I L.//Y:

Y

Semigroup Generated by A

.

Rather than estimating the resolvent, we simply construct the semigroup explicitly.

First let us note that

O

jjR.; A/L./jj !0 as Re ! 1

and hence

O

jjR.; A/L./jj <1

in a right-half-plane contained in . Hence we have the Neumann expansion

1 1

X k

O

I R.; A/L./ DI C O

R.; A/L./ ;

kD1

where Z 1

O

R.; A/L./Y D et J.t/Y dt

0

and Z t

J.t/Y D S.t /./;

0

O

where . : / is the inverse Laplace transform of L./:

Hence it follows that

1

O

I R.; A/L./ D I C rO ./;

Z 1

rO ./ D et r.t/dt;

0

5.6 State Space Theory 195

where

1

O

rO ./ D I R.; A/L./ I

1

O

D I R.; A/L./ O

I I R.; A/L./

O

D R./L./

and hence Z 1

R./Y D et W .t/Y dt

0

and

Z t

W .t/Y D S.t/Y C S.t /r./Y d;

0

Here we have obtained it in a more constructive way that also displays explicitly the

nature of the function. In particular

W .0/ D I

Armed with this function we can now proceed to construct the semigroup, by simply

taking the inverse Laplace transform of the resolvent. Thus, given

x

ZD ;

Y

Z t

Y .t/ D W .t/Y C W .t /.CS c ./x L./Y /d

0

and

0 Z t 1

@Sc .t/x C Sc .t /LYP ./d A

Z.t/ D Sz .t/Z D 0 ;

Y .t/

196 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

where we note that the derivative is well defined for x in the domain of Ac and Y in

the domain of A.

Thus defined, the Laplace transform of the derivative

Z 1

et SPz .t/Zdt D R.; A./Z Z D A.R.; A./Z;

0

Z 1 Z 1

t

D A. e Sz .t/Zdt D et A.Sz .t/Zdt ;

0 0

or

P

S.t/Z D A.S.t/Z;

which is enough to prove that A. generates the C0 semigroup Sz . : /:

And hence we have proved the state space representation

P

Z.t/ D A.Z.t/;

Y .t/ D PZ.t/: t

u

Next we consider the case M D 1 which differs radically from the case M D 0:

Here we follow [71]. We begin by evaluating the coefficients wij defined as

before.

k p

e

w11 D 4 p C erf k ;

k

p

erf k ek

w12 D p ;

k2 k3

p !

1 p k k

w21 D 2 erf k C 2e 1 ;

k

" p r #

2 3 k k 1 2 3

w22 D 2 C 3 erf k C 2e :

3 k k k2 k3

Note that in contrast with the case M D 0, there are no polynomials in k. In fact

wij .k/ does not converge to a finite limit as k goes to zero.

5.6 State Space Theory 197

The role of the Theodorsen function is now played by the erf function

Z z

2

et dt

2

erfz D p

0

p

and erf z has a logarithmic singularity along the negative real axis and hence so

does wij . In fact:

p 1

erf s 2 X

p D p .1/k s k =.2k C 1/ entire function;

s kD0

Z 1

p

erf s D est ..t/ f .t//dt;

0

where

1 . p

D 1 .t .t 1// ; t > 1:

The Laplace domain equations are

O s/ C 2 S .;

2 m h.; O O s/000

s/ C EI h.;

O

D L.; s/

k O s/ C .kw12 C .1 ka/w11 /O .; s/

D bU 21 w11 h.;

b

k p

k e O s/

D bU 21 4 p C erf k h.;

b k

p ! k ! #

erf k ek e p

C p C .1 ka/4 p C erf k O

.; s/ ;

k k k

O

2 I .; O s/ GJ .;

s/ C S 2 h.; O s/00 D MO .; s/

k O s/

D b 2 U1

2

.w21 aw11 /h.;

b

i

C.kw22 C .1 ak/w21 akw12 a.1 ak/w11 /O .; s/

198 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

" p !

k 1 p k

D b 2 2

U1 erf k C 2ek 1

b k2

p p

ek O 2 3

a4 p C erf k h.; s/ C k 2 C 3 erf k

k 3 k

r p !

k 1 2 3 1 p k

C 2ek C .1 ak/ 2 erf k C 2ek 1

k k2 k3 k

p k ! #

erf k ek e p

a p 4a.1 ak/ p C erf k O

.; s/ :

k k3 k

Let us next isolate the non-circulatory terms by taking the nonzero limits as Re

goes to infinity.

O

M 2 x.; s/ C Ax.;

O s/ D F ./x.;

O s/;

where

F11 ./ F12 ./

F ./ D ;

F21 ./ F22 ./

where

p

ek

F11 ./ D 4U1

2

k p C erf k ;

k

p ! k !

erf k ek e p

F12 ./ D bU 12

p C 4.1 ka/ p C erf k ;

k k k

" p ! k !#

k 1 p k e p

F21 ./ D bU 12

erf k C 2ek 1 4a p C erf k ;

b k2 k

p r

2 3 k k 1 2 3

F22 ./ D b U1 k

2 2

2 C 3 erf k C 2e

3 k k k2 k3

p ! p

1 p k erf k ek

k

C .1 ak/ 2 erf k C 2e 1 a p

k k k3

k p

e

4a.1 ak/ p C erf k :

k

5.6 State Space Theory 199

Hence

F ./

lim D 0;

!1 2

0 1

F ./ 4U1

2

4bU 2

1a

lim D@ 4 A;

!1 k 0 b 2 U1

2

C 4a2

3

0 1

4U1

2

4bU 2

1a 0 4bU1 2

lim .F ./ k/ @ 4 A D :

!1 0 b 2 U1

2

C 4a2 0 4ab 2 U1 2

3

Having evaluated the non-circulatory terms we can formulate the equations in the

Laplace domain as

2 Mx./

O C Dx./

O C Kx./ C Ax./

O D FO ./x.;

O :/;

where

0 1

4bU1 4b2 U1 a

DD@ 4 A;

0 b 2 U1 2

C 4a2

3

0 4bU1 2

KD ;

0 4ab 2 U1 2

0 1

4U1 2

4bU 2

1a 0 4bU1 2

O

F ./ D F ./ k @ 4 A

0 b 2 U1

2

C 4a2 0 4ab 2 U1 2

3

Z 1

D et L1 .t/dt Re > a :

0

Aeroelastic Modes

As before

D./ D P e `A./t Q;

d.1; ; U / D det D./:

200 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

d.1; k ; Uk / D 0:

They are again confined to a finite strip. The corresponding mode shape vector is

given by

xl;k

Yk D ;

x2;k

x2;k D k x1;k ;

D.k /zk D 0; zk 0:

x1;k .s/ D T26 yk .s/ 0 < s < `:

We can now go on, as in the case M D 0, to define the generalized resolvent R./

and the modal representation.

The big difference now, unlike M D 0, is that

The unique feature here is that zero is not a mode; indeed D./ is not defined at

zero. There is no continuity at zero.

We now get back to the general case 0 < M < 1 having treated separately the cases

M D 0 and M D 1 where we exploited the luxury of explicit solution of the Possio

equation. We continue with the notation as in Sect. 5.3 and the normalization there,

so we may take b D 1; with the normalized frequency k D b=U .

The Possio equation for 0 < M < 1 in the abstract form is given by:

p

2 O 1 M2 O

wO a .; : / D A.; :/ C PH.B.k/ B.1//A.; :/ (5.106)

M M

and we simply assume that it has a unique solution, and the solution is then given by

5.6 State Space Theory 201

p !1

O 1 M2 2

A.; :/ D I C PH.B.k/ B.1// wO a .; : /: (5.107)

M M

(We should recall here that we do have in Sect. 5.4 a constructive solution for

small , which happens to be the region of major interest in practice.)

We define the coefficients wij ./ as in Sect. 5.4 (5.3), which are now recognized

as functions of , and this is all we need from the solution to the Possio equation for

all M.

Note that for every M:

wij ./ D wij ./;

implying that the aeroelastic modes come in conjugate pairs. The main

difference for M > 0 from M D 0 is that from (5.72) we have that

lim Re ! 1

Z 1

2

wij D d D 0; i j;

M 1

4

w11 D ;

M

Z 1

2 4

w22 D 2 d D :

M 1 3M

non-polynomial term the polynomial term is a constant (of zero degree), unlike

the case for M D 0, where it is of degree 1. The polynomial terms are called

the non-circulatory terms in [6]. To see the implication of this let us consider the

Laplace transform version of the structure dynamics:

O s/ C 2 S .;

2 m h.; O O s/0000 D L.;

s/ C EI h.; O s/

k O

O s/ C .kw12 C .1 ka/w11 /.;

D bU 21 w11 h.; s/ :

b

Or, collecting the non-circulatory terms into the structure state terms:

O s/ C 2 S .;

O 4 O s/ ab .;

O

2

m h.; s/ C bU1 .h.; s//

M

O s/

C bU 21 .; s/ C EI h.; 1

M b

O

C .kw12 C .1 ka/wM 11 /.; s/;

202 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

where

4

wM 11 ./ D w11 ./ and ! 0 as Re ! 1:

M

Similarly

O

2 I .; O s/ GJ O .; s/00

s/ C S 2 h.;

O k O s/

D M .; s/ D b U12 2

.w21 aw11 /h.;

b

i

O

C.kw22 C .1 ak/w21 akw12 a.1 ak/w11 /.; s/ :

O 2O 4ab O 4ab O

I .; s/CS h.; s/ C bU1

2

h.; s/ C bU 1 h.; s/

M M

4 4 2 O 4 O O

C bU 1 a2 b 2 C b .; s/ bU 21 ab .; s/GJ .; s/00

M 3M M

k O

D b U1

2 2

.w21 awM 11 /h.; s/ C .k wM 22 C .1 ak/w21 akw12

b

i

a.1 ak/wM 11 / O .; s/ ;

where

4

wM 22 ./ D w22 ./

3M

and goes to zero as Re ! 1.

The main thing is now to note that the structure inertia matrix is not changed

unlike in the case where M D 0. But damping and stiffness terms are added which

are reflected in the abstract version as bounded operators. Thus collecting terms

containing 2 and we can now write the structure with the aerodynamic loading

Laplace domain equation:

2 Mx./

O C Dx./

O C Kx./

O C Ax./

O D F ./x.;

O : /; (5.108)

5.6 State Space Theory 203

where

m s

MD ;

s I

0 1

4 4

B bU1 ab bU1 C

B M M C

DDB C ;

@ 4 4 4 A

ab bU1 2 2

bU 1 a b C b 2

M M 3M

0 1

4 2

0 bU1

B M C

KD@ A; (5.109)

4

0 bU1 ab

2

M

where MDK are constant 2 2 matrices and FO ./ is the Laplace transform:

0 1

k

bU 2

M

w h C .kw C .1 ka/ M

w /

B 1 C

11 12 11

B

b C

h B C

FO ./ DB k C (5.110)

Bb 2 U1 2

.w21 awM 11 /h C

@ b A

C .k wM 22 C .1 ak/w21 akw12 a.1 ak/wM 11 /

and A is the structure differential operator as defined in Chap. 2.

Next we take inverse Laplace transforms

Z t

Mx.t/

R C Dx.t/ P C Kx.t/ C Ax.t/ D F ./x.t /d; (5.111)

0

Z 1

FO ./ D et F .t/dt Re > 0: (5.112)

0

Next with

x.t/

Y .t/ D ;

P

x.t/

Z t

YP .t/ D AY .t/ C L.t /Y ./d;

0

where

0 0

L.t/ D 1 4 4 matrix function ; (5.113)

M F .t/ 0

204 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

0 I

AD l 1 (5.114)

M .A C K/ M D

Here as before, A generates a C0 semigroup with compact resolvent; and the

analysis is similar to that for M D 1: M D 0 is a special case, whereas M D 1 is

typical of M > 0.

Taking Laplace transforms in

Z t

YP .t/ D AY .t/ C L.t /Y ./d;

0

we have:

O

.I A L.// YO ./ D Y .0/:

Thus we define the aeroelastic modes as the zeros:

O

I A L./ Y D0

1

R./ D I ALO ./ :

Note that the change from the M D 0 case is that (as in the case M D 1/

O

L./ O

is replaced by L./ O

(the two L./ are not the same functions, of course!)

which makes the problem less complex in that now

O

I A L./ YO ./ D Y:

Or,

YO ./ D R./Y:

The aeroelastic mode fk g are of course obtained as the zeros of the

determinant

D.M; ; U / D P e A./ Q

with A./ just as defined in Sect. 5.4 except that the wij are no longer expressible

explicitly.

We assume that AOi .; M / is the solution of the Possio equation (5.2) special-

ized to

5.6 State Space Theory 205

p !1

1 M2 2

AOi .; : / D I C PH.B.k/ B.1// fi ;

M M

where

D s; b < s < b

and note that the solution is defined for all omitting the modes k and the negative

half-line. Moreover the solution is analytic in this open set and hence so are the wij

and so is A./.

Note that the roots again come in conjugate pairs, if complex. There can be real

roots for some speeds; zero is a root for some speeds.

For the existence of the roots for U positive, we need to draw on the

analyticity properties of D.M; ; U /. These are determined by those of fwij ./g

which in turn are dependent on those of the solution to the Possio equation.

However, we no longer have the luxury of an explicit solution. And thus we have to

go back to (5.49), the abstract form of LDP for values other than Rek > 0, assumed

there. The Hilbert transform operator H is what causes trouble. The right-hand side

of (5.49) is

b

PHPB.k/PA C T PH.I P/B.k/PA; k D ;

U

where the first term is an entire function of and the second term

WA D T PB.I P/B.k/PA;

we now show is analytic except for the logarithmic singularity along the negative

half-line, of the same kind as we have seen for M D 0. Thus we have

kR.k/A. : /

T PH.I P/B.k/PA D T PH.I P/ p

1 M2

Z 1

kR.ks/A. : /a.M; s/ds ;

2

where the second term is kind of an average of the first term and inherits the

analyticity properties of the first term. Denoting the first term by g. : / we have:

s p

Z 1 Z b

k bx ek 2b C

g.x/ D p d ek.b/ A./d;

bCx 0 . C b x/ b

jxj < b

and although this function is defined for Re k < 0, the integral is not defined for

Re k < 0.

206 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

Z 1

p

ek 2b C

p d D F .k/; Re k > 0; take b D 1;

0 . C s/

dF

D sF .z/ .K0 .z/ C K1 .z// Re z > 0

dz

and because .K0 .z/ C K1 .z// is analytic except for z 0; F .z/ can be extended

analytically to the left-half-plane, but will have an essential singularitybranch

cutalong the negative axis exactly as is the case for M D 0.

Thus the operator PH.B.k/B.1// in the Possio equation can be extended to be

analytic in the whole plane except for the branch cut along the negative axis, and this

property then extends to the solution, and hence to wij , and finally to D.M; ; U /

and d.M; ; U /.

Further more they are limited to a finite strip (we call it the spectral strip) because

the contribution of the circulatory terms:

O

L./ !0 as Re ! 1 (5.115)

and hence we may imitate the arguments for M D 0 (cf. Sect. 5.3) where really only

this argument is needed.

Again define xk by:

D./ being a 3 3 matrix the null space of the matrix is at most 2, and we assume

it to be 1, as is the case for U D 0. Similarly we assume that the multiplicity is also

1, so that

D 0 .k /xk 0: (5.117)

Lemma 5.31. Let

D.k /? zk D 0 zk 0:

Then

zk ; xk 0: (5.118)

Proof. The 3-space has the orthogonal decomposition:

5.6 State Space Theory 207

Hence if

zk ; xk D 0;

we must have that xk is in the range space of D.k /. Or,

xk D D.k /h for h 0:

Hence

D.k /2 h D 0:

Taking the derivative with respect to , we have

D 0 .k /xk D 0;

u

The corresponding Yk satisfies

O k //Yk D 0:

.k I A L.

Or, with

xl k

Yk D ;

x2 k

we have

k x1;k D x2;k ;

Hence

x2;k D k x1;k : (5.120)

Define the 2 6 matrix:

100000

T26 D : (5.121)

000100

Then

x1;k .s/ D T26 eA.k /s Qx k ; 0 < s < `k

and xk is defined by (5.10).

We may normalize xk by:

" Z ! #

`

A? .k /s

Mx1k ; xlk D 1I Q? e T26 ? MT26 eA.k /s Qds xk ; xk D 1: (5.122)

0

208 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

T26 eA.k /s Qxk

Yk .s/ D ; 0 < s < `:

k T26 eA.k /s Qxk

fl

Lk x D I f1 .s/ D T26 eA.k /s Qx; 0 < s < `;

k fl

Yk D Lk xk ;

and

Yk ; Yk D Lk ? Lk xk ; xk D j k j2 :

The energy space norm of Yk is not equal

to 1.

O k / is the dimension of the

The dimension of the null space of k I A L.

null space of P e A.k /` Q and is 1, with multiplicity 1.

We note that in as much as the conjugates

N

D./ N

D D./I N

A./ N

D A./;

Yk YNk

R./Yk D I R./YNk D

k N k

corresponding to

xk

D./1 xk D :

k

The case where zero is an aeroelastic mode is treated separately below, as it is

involved in defining the divergence speed.

We note that zero is an aeroelastic mode only for a sequence of speeds. The

functions (for M 1) are defined as the limit from above (Im nonzero) or from

the right (Re > 0).

And

d.0; 0C; 0C/ D 1:

Generalized Resolvent

The generalized resolvent is defined for excepting the aeroelastic modes and the

negative half-line:

5.6 State Space Theory 209

1

O

R./ D I A L./

O 1 xl;g

R./Yg D .I A L.// Yg W Yg D :

x2;g

Let

hg

D Mx2;g C .D C K/x1;g :

g

hg .s/ g .s/

yg .s/ D col 0; 0; 0; ; 0; ;

EI GJ

Z s

y.; s/ D esA./ Qz./ C e.s /A./ yg ./d;

0

Z `

P .e`A./ Qz./ C vg .// D 0I vg ./ D e.` /A./ yg ./d:

0

Or,

D./z./ D P vg ./I D./ D P e `A./ Q:

Hence

z./ D D./1 P vg ./;

which is defined where wij are defined excepting the modes fk g and

Z `

y.; s/ D esA./ QD./1 P e.` /A./ yg ./d

0

Z s

C e.s /A./ yg ./d; (5.123)

0

and finally:

xl ./

R./Yg D ;

x2 ./

h.; : /

x1 ./ D T26 y.; : / D ;

.; : /

x2 ./ D x1 ./ x1;g :

210 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

We proceed first to deduce the Greens function for the generalized resolvent. For

this purpose it is convenient to use the notation:

h. : / . : /

C6 ./Y D col 0; 0; 0; ; 0; ;

EI GJ

where

h. : / xl

D Mx2 C .D C K/x1 I Y D :

. : / x2

Z `

x1 .; s/

D R.; s; /Y ./d; 0 < s < `; (5.124)

x2 .; s/ 0

Z `

x1 .; s/ D T26 esA./ QD./1 P e.` /A./ C6 ./Y .s/ds

0

Z s

C e.s /A./ C6 ./Y ./dx2 .; s/ D x1 .; s/ x1 .s/:

0

Z ` Z `

jjR.; s; /jj2 dsd < 1:

0 0

X 1

< 1;

j k j2

where the summation is over all roots, the index k being chosen such that the

imaginary part

j !kC1 j j !k j:

Our convention for the modes is:

k D k C i !k ; !k > 0

and the conjugate is of course always a mode as well, if complex. And in particular

5.6 State Space Theory 211

X 1

< 1 nonzero k : (5.125)

jk j2

Equation (5.9) shows that the generalized resolvent has simple poles at the

aeroelastic modes confined to a finite strip. We can use this to determine the inverse

Laplace transform by the inversion formula

Z L

1

W .t/Y0 D lim e. Ci !/t R. C i !/Y0 d!; (5.126)

L!1 2 i L

x1 .t/

W .t/Y0 D ;

x2 .t/

x1;0

Y0 D ;

x2;0

W .0/Y0 D Y0 in D.A/;

and

d

x2 .t/ D x1 .t/;

dt

h0 ./

D Mx2;0 C .D C K/x1;0 ;

0 ./

h0 .; s/ 0 .; s/

y0 .; s/ D col 0; 0; 0; ; 0; ;

EI GJ

Z L

1

x1 .t; s/ D lim et T26

L!1 2 i L

Z `

esA./ QD./1 P e.`/A./ y0 .; /d

0

Z s

C e .s/A./

y0 .; /d id!jD Ci ! 0 < s < `;

0

where we have seen that D./1 has simple poles at D k with residue

212 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

I

1

Dk1 D D./1 d D k C rei :

2 i

(The convergence of the infinite series is assumed here for the moment; the proof is

given below.)

Hence we deform the contour as before following [33,44] where the functions are

meromorphic but with a branch cut along the negative axis.

Thus we have

1 Z ` !

X

1

x1 .t; s/ D T26 e sA.k /

QD k P e .` /A.k /

y0 ./d ek t C x1;R .t; s/;

kD1 0

where the second term x1;R .t; : / is the evanescent term we have already treated

in Sect. 5.5.

The evanescent term can be expressed:

Z 1

U

x1;R .t; : / D eRU=bt RJ .R/Yg dR; (5.128)

b 0

where RJ . : / is the jump across the line of singularity: 1 < 0 and for all

M 0 including M D 0.

Note the slight difference in notation from Sect. 5.5. Thus we define

1

RJ .r/Y D lim R.r C i !/ R.r i !/Y; ! 0;

2 i !!0

x

Y D l

x2

" Z

sA./ `

.R./Y /.s/ D T26 e QD./1 P e.` /A./ y./d

0

Z s

C e.s /A./ y./d ; 0 < s < `;

0

h.; s/ .; s/

y.s/ D col 0; 0; 0; ; 0; ;

EI GJ

h./

D Mx2 C .D C K/x1

./

5.6 State Space Theory 213

" Z

sA./ `

.R./Y /.s/ D T26 e QD./1 P e.` /A./ y./d

0

Z s

C e.s /A./ y./d ; 0<s<`

0

branch cut.

Moreover the jump R.r C i !/ R.r i !/Y; ! 0 converges strongly as

we approach the line of singularity as ! goes to zero for each r > 0. And the limit

RJ .r/Y goes to zero as r goes to zero.

Hence finally:

x1 .t/ x1;0

W .t/Y0 D I Y0 D ;

x2 .t/ x2;0

1 Z !

X `

x1 .t; s/ D T26 e sA.k /

QD 1

k P e .` /A.k /

yg ./d ek t

kD1 0

Z 1

C ert .RJ .r/Y0 /.s/dr;

0

Z 1

xO 1 .; : / D et x1 .t; : /dt

0

Z !

X `

1

D T26 e sA.k /

QD 1

k P e .` /A.k /

yg ./d

0 k

Z 1

1 .RJ .r/Yg /.s/

dr; 0 < s < `;

0 Cr

Z 1

et x2 .t; : /dt D xO 2 ./ D xO 1 ./ x1;g :

0

214 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

0P R ` .` /A. / 1

T26 esA.k / QD 1

1

k k P 0 e

k y ./d

g

B C

B R C

B C 1 01 .RJ .r/Y /.s/

dr C

B Cr C

DB R C;

BP 1 ` C

B T26 e sA.k /

QD k P 0 e .` /A.k /

yg ./d k C

@ A

R

1 1 .RJ .r/Y /.s/

C 0 Cr

dr x1;g .s/

Let us define the linear bounded operator on H into H by the contour integral

Z

Pk Y D R./dI .k / D f D k C rei ; 0 2g

.k /

Z 2

1

D R k C rei Y d for 0 < r;

0 2

0 R ` .` /A. / 1

T26 esA.k / QD 1

k P 0 e

k y ./d

g

@ R ` .` /A. / A ; 0 < s < `: (5.131)

k T26 esA.k / QD 1

k P 0 e k y ./d

g

Z

f ./R./d D f .k /Pk :

.k /

xl

For Y D ;

x2

0 Z 1 1

l RJ .r/Y

X Pk Y dr

B Cr C

R./Y D CB Z

0 C:

k @ 1 1

RJ .r/Y A

C dr x1 .s/

0 Cr

Now

O

X I A L./ PkY

O

Y D I A L./ R./Y

k

5.6 State Space Theory 215

0 Z 1 1

1 RJ .r/Y

B dr

Cr C

O

C I A L./ B Z

0 C:

@ 1 1

RJ .r/Y A

dr x1 .s/

0 Cr

H

Then the contour integral .1=2 i /Yd; D k C rei yields

O k / Pk Y:

0 D k I A L.

Hence Pk Y is the slant (not orthogonal because Pk Pk ) projection on the null

the dimension of the range of Pk is 1. Or

Pk Y; Yk

Pk Y D k Yk D Yk ;

Yk ; Yk

Pk Yk D Yk ;

Z 1 0 1

l RJ .r/

X Pk Y B Y dr C

R./Y D CB Z 01 C r C:

k @ l rRJ .r/Y dr

A

k

0

Note that this runs over all roots: the conjugates are treated as separate roots.

Next let Q denote the linear bounded operator

0 R1 1

1

0 RJ .r/Y dr

QY D @ R A: (5.132)

1 1

0 rRJ .r/Y dr

The range of Q are again the evanescent states as in the incompressible case, stable

for all U . From (6.9)

Pk Y

R./Pk Y D :

k

Now

lim R./Y D Y:

Re!1

X

Y D Pk Y C QY (5.133)

216 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

Lemma 5.32.

Pk Pj D Pj Pk D 0; j k

D Pk j D k;

Q2 D Q;

Pk Q D 0 D QP k :

Proof. Here we essentially follow the theory of pseudoresolvents in [10, p. 208

et seq].

Thus we note that for k j ,

Z Z

1

Pk Pj x D R./R./dd;

.2 i /2 .k / .j /

where h i

O

R./ R./ D . /I L./ O

L./ R./R./

and hence

!

R./ R./ O

L./ O

L./

R./R./ D C R./R./ (5.134)

and

Z Z

1 R./

dd D 0; k j D Pk ; k D j;

.2 i /2 .k / .j /

Z Z

1 R./

dd D 0; k j D Pk ; k D j;

.2 i /2 .k / .j /

Z Z !

1 O

L./ O

L./

R./R./dd

.2 i /2 .k / .j /

O k / L.

L. O j/

D Pk Pj k j:

j k

Or

O k / L.

L. O j/

Pk Pj D Pk Pj k j:

j k

5.6 State Space Theory 217

Hence

Pk Pj D 0 k j:

Also by a similar integration procedure:

Pk Qx D 0 D QP k :

t

u

Next, using (5.22)

0 1

X

Pk x D Pk @ Pj x C Qx A D Pk Pk x;

j

0 1

X

Qx D Q @ Pj x C Qx A D QQx:

j

projections because not self-adjoint.

To proceed further we need to go on to the operator adjoints.

Adjoint Theory

Let us consider now the properties of the adjoint operator: (Note that the adjoint

now is with respect to the energy inner product.) We calculate

O

.I A L.// ?

ZD0

by

O

Z; .I A L.//Y D0 for every Y:

Or

Az1 ; x1 x2

h i

C z2 ; 2 Mx1 C Dx1 C Kx1 C Ax1 C Dx1 C Kx1 FO ./x1 D 0

with

h

x1 D I

218 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

h i

O

D h2 ; 2 mh C 2 S C EI h0000 L.; :/

h i

C 2 ; 2 I C S 2 h GJ 00 MO .; :/ ;

where

O k

L.; s/ D bU1 2

w11 h.s/ C .kw12 C .1 ka/w11 /.s/ MO .; s/

b

k

D b 2 U1 2

.w21 aw11 /h.s/

b

C .kw22 C .1 ak/w21 akw12 a.1 ak/w11 / .s/ :

z k

zk D 1 ;

z2 k

Q

z2k .s/ D T26 eA.k /s Qzk 0 < s < `; (5.135)

Q k /zk D 0;

D. zk 0; (5.136)

Q Q

D./ D P e A./` ;

0 1

0 100 0 0

B0 010 0C

B 0 C

B C

Q B0 001 0 0C

A./ DB C;

Bw1 0 0 0 w3 0C

B C

@0 000 0 1A

w2 0 0 0 w4 0

N 2k Mz2k C Az2k D K? C FO .k / N k D? z2k ; (5.137)

N k z1k D I A1 K? C A1 F.

O k /? z2k : (5.138)

Q

This time z1k is determined from z2k . The zeros of det D./ are the conjugates of

those of det D./.

5.6 State Space Theory 219

and asymptotically in k,

x1k ; Ax1k

!1

jk j2

and

jjYk jj2

! 2:

jk j2

We would want a similar normalization for Zk . We now normalize so that

Then " Z ! #

`

Q k /? s Q k /s

? A.

Q e .T26 / MT26 e A.

Qds zk ; zk D jk j2 ;

0

Then we calculate

1 h 1 ? 1 O

i

O k /? z2k :

z1k ; Az1k D I C A K A F . k / ?

z2k ; A C K ?

F.

jk j2

Now

N 2k Mz2k C Az2k D K? C F.

O k / N k D z2k

,

1 ?

z1k ; Az1k D 2k z2k ; Mz2k C D z2k Mz2k ; z2k

N k

1 ? 1 O 1 ?

k A K A F .k / z2k ; Mz2k C

2 ?

D z2k

N k

,

Mz2k ; z2k :

Hence asymptotically in k

j ! 1; noting that ! 1:

jk j2 jk j2

220 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

.

Zk ; Zk jk j2 ! 2:

Second Normalization

0, therefore we define

,q

Q

Yk ; Zk ZQ k ; ZQ k Yk ; Yk D 1: (5.139)

We simply keep using Zk again in this renormalized version. Again we can write

zk D LQ k zk ;

Q g1 Q

Lk z D I g2 .s/ D T26 eA.k /s Qz; 0 < s < `;

g2

N k g1 D g2 A1 K ? A1 FO .k /? g2 ;

LQ k MLQ k g2 ; g2 D jk j2 :

Note that

Pk? Z; Pj Y D 0 j k:

Hence

Yk ; Zj D 0 for kj

and similarly:

QY; Zj D 0 for everyj:

The sequences fYk g; fZk g are biorthogonal. Note the set of eigenvalues is the same.

Then we have the representations:

X

Y D k Yk C QY (5.140)

X

D k Zk C Q Y: (5.141)

5.6 State Space Theory 221

Pk QY D 0 D QP k Y:

We use the representation:

Y; Zj D Pj Y; Zj

D j Yj ; Zj :

If

1

X

Yj ; Y D Yj ; k zk C Yj ; Q Y D 0;

kD1

Y; Zk

k D :

Yk ; Zk

Next

p

jYk ; Zk j jjY jj Zk ; Zk :

Next

p p

Yk ; Zk D .1/ Yk ; Yk Zk ; Zk :

Hence

jjY jj

j k jjjYk jj p ;

Yk ; Yk

which asymptotically

jjY jj

:

jk j

Because

1

X 1

< 1;

jk j2

kD1

it follows that

1

X

jk j2 jjYk jj2 < 1: (5.142)

kD1

222 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

1

X

jk j2 jjZk jj2 < 1: (5.143)

kD1

This is called the Riesz property of the biorthogonal sequence proved in a more

abstract way in [22].

Note that

X

Y; Y D k Nk Yk ; Zk : (5.144)

This is enough to prove the convergence of all the infinite sums considered above.

Next we can calculate the Zk in a different way. Thus

Ak Pk Y D k Pk Y:

Ak Pk Y; Z D k Pk Y; Z D Y; N k Pk Z:

Hence

Pk? Z
D.Ak /

and

N k I A?k Pk Z D 0:

We define:

Pk? Zk D Zk :

z x h

Pk Y; Z; Z D l I Y D l I C62 x D col 0; 0; 0; ; 0;

z2 x2 EI GJ

Z `

D T26 esA.k / .QD 1

k P /C6 Y .s/; .Az1 /.s/ ds

0

Z `

C k T26 esA.k / .QD 1

k P /C6 Y .s/; Mz2 .s/ ds

0

Z ` h i

D x1 .s/; .C6 /? .QD 1 sA.k /?

k P/ e .T26 /? .Az1 /.s/ ds

0

Z ` h i

C x2 .s/; N k .C6 /? .QD 1 sA.k /?

k P/ e .T26 /? Mz2 .s/ ds

0

5.6 State Space Theory 223

Z ` h i

D Ax1 .s/; A1 .C6 / .QD 1

k P / ? sA.k /?

e .T 26 / ?

.Az1 /.s/ ds

0

Z ` h i

C Mx2 .s/; N k M1 .C6 /? .QD 1

k P / sA.k /?

e .T 26 / ?

Mz2 .s/ ds:

0

Hence

!

A1 .C6 / .QD l .T26 / .Azl /.s/

? sA.k / ?

k P/ e

Pk Z D ;

N k Ml .C6 / .QD l ? sA.k /?

k P/ e .T26 /? Mz2 .s/

QD 1

k P D P .Dk1 /? Q? ; : : : ; a 6 6 matrix;

I

1

.Dk1 /? D D ./1 d D N k C r ei :

2 i

Finally let

0 Z 1 1

1

B ert RJ .r/Y dr C

Q.t/Y D B

@1 Z 1

0 C

A (5.145)

rt

.r/e RJ .r/Y dr

0

Furthermore, QY; Zk D 0 for every k. Hence the range of Qthe space of

evanescent statesis orthogonal to the modal space spanned by the sequence fZk g.

And similarly

Q? Z; Yk D 0:

Then finally the solution to the aeroelastic equation can be expressed:

X

Y .t/ D ek t Pk Y .0/ C Q.t/Y .0/; (5.146)

where

Y; Zk

Pk Y D Yk :

Yk ; Zk

Here the most intriguing part is the term that cannot be expressed in terms of modes!

Now we can return to the properties of the nonrational component: the evanescent

states:

Q.t/Y; t 0:

224 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

S.t/Y D W .t/.Y QY /; t 0

X

D ek t Pk Y

k

D W .t/Y Q.t/Y

modes fYk g.

S.t/QY D QS.t/Y D 0

Q.t/Y is stable for all U:

u

From (5.35) or directly from (5.16)we have the Greens function representation for

the time domain solution:

W .t/Y .0/ D Y .t/

is the function: Z `

Y .t; y/ D W .t; y; s/Y .0; s/ds; (5.147)

0

where Z 1

et W .t; y; s/dt D R.; y; s/: (5.148)

0

Next we extend the state space theory for M > 0.

In view of the fact that the convolution/evolution is different from that for M D 0:

Z t

YP .t/ D AY .t/ C L.t /Y ./d

0

(no derivative in the integral) we need to examine the changes needed in the state

space representation (5.2)

P

Z.t/ D A.Z.t/ Y .t/ D PZ.t/:

5.6 State Space Theory 225

Thus we begin with the analogue of Theorem 5.22. The main question is the nature

of the function L. : /. We begin with:

2

AOi .0; :/ D p T fi ;

1 M2

r

2 1x

AO1 .0; x/ D p ;

1 M2 1 C x

2 p

AO2 .0; x/ D p 1 x2 :

1 M2

Hence

2

w11 .0/ D p ;

1 M2

2 4

wM 11 .0/ D p ;

1 M2 M

w22 .0/ D 0;

4

wM 22 .0/ D :

3M

Which is enough to show that L. : / here enjoys the same sort of properties as in

P M: Hence Theorem 5.22 holds in this case as well and so does the

the case for zero

definition of and that in turn of X.; of C and Ac , and Z denoting the elements

therein, as there. But now with

x.t/

Z.t/ D ;

Y .t/

we have:

P

x.t/ D Ac x.t/ C LY .t/;

YP .t/ D AY .t/ C C x.t/

P Ac L

Z.t/ D A.Z.t/I A. D ;

C A

PZ.t/ D Y .t/: (5.149)

Let us examine the changes needed with this definition. The domain of A. remains

the same: i.e., Domain of Ac D Domain of A.

226 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

Spectrum of A.

x D Ac x C LY;

Y D C x C AY:

Or

x Ac x D LY:

Proof. For Re > 0; is in the resolvent set of Ac so that given any Y in H, we

have:

Z 1

D et Sc .t/LY dt:

0

Z 1

x.; s/ D et L.s C t/Y dt; 0s1

0

and hence Z 1

C x D x.; 0/ D O

et L.t/Y dt D L./Y:

0

Hence we have the representation:

Z s

O

x.; s/ D e L./Y

s

e.s / L./Y d; (5.150)

0

where

L.1/Y

x.; 1/ D :

O

But this defines an analytic function of in the region of analyticity of L./Y =

which is the whole plane omitting the branch cut along < 0: In other words we

have proved that given Y , we can find x./ in such that

5.6 State Space Theory 227

x Ac x D LY

even if is not in the resolvent set of Ac , and thus not necessarily unique. Hence

O

Y AY D L./Y

Next suppose is an aeroelastic mode. Then

O

Y AY D L./Y:

But given Y we have shown that we can find x./ such that

u

Resolvent of A.

R xg

Z Az D ;

Yg

x Ac x LY D xg ;

Y AY C x D Yg : (5.151)

Theorem 5.35. Let denote the region in the complex plane: D

fresolvent set of Ag \ fcomplement of the point spectrum of A.g \ fRe > 0g.

Then is contained in the resolvent set of A.. Moreover, it has the representation

R.; Ac /.x C L/

. /Z D

R.; A

;

D R./.CR.; Ac /x C Y /;

Proof. For in we can solve (5.40) as:

x R.; Ac /.xg C LY /

D :

Y R.; A/.C x C Yg /

228 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

But

O

C x D CR.; Ac /xg C L./Y

O

D CR.; Ac /xg C L./Y;

O

Y AY CR.; Ac /xg L./Y D Yg ;

O

Y AY L./Y D Yg C CR.; Ac /xg :

LY D LR./.CR.; Ac /xg C Yg /:

Hence

x R.; Ac /.xg C L/

D ;

Y

D R./.CR.; Ac /xg C Yg /

as required. t

u

Semigroup Generated by A.

We show next that A. generates a C0 semigroup. The main step here is again to

evaluate the inverse Laplace transform of the generalized resolvent. We note that

O

jjR.; A/L./jj !0 as Re ! 1:

O

jjR.; A/L./jj <1

in a right-half-plane and furthermore, also the Neumann expansion therein

1

X k

O

I R.; A/L./ 1

D O

R.; A/L./ ;

kD0

5.6 State Space Theory 229

where

Z 1

O

R.; A/L./Y D et J1 .t/Y dt;

0

Z t

J1 .t/ D S.t /L./d;

0

whose Laplace transform is defined for Re > the growth bound of the semi-

group S .

Define

Z t

Jn .t/Y D J1 .t /Jn1 ./d; n 2;

0

n

O

R.; A/L./ :

Let

1

O

rO ./ D I R.; A/L./ I;

Then

1

O

rO ./ D I R.; A/L./ O

R.; A/L./ O

D R./L./:

Let

1

X

r.t/Y D Jn .t/Y; t 0:

nD1

Z 1 1

X k 1

et r.t/dt D O

R.; A/L./ O

D I R.; A/L./ I

0 kD1

D rO ./:

Next let

Z t

W .t/Y D S.t/Y C S.t /r./Y d t > 0: (5.152)

0

230 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

Z 1

et W .t/Y dt D R.; A/ .I C rO .// Y

0

1

D R.; A/ I R.; A/L./O Y

D R./Y:

Define

Rt

Sc .t/x C 0 Sc .t /L./d x

Sz .t/Z D ; ZD ;

.t/ Y

Z t

.t/ D W .t/Y C W .t /CS c ./ xd:

0

The Laplace transform is the resolvent of A., proving that it is the semigroup

generated by A.. Thus we have the state space representation:

P

Z.t/ D A.Z.t/;

aeroelastic equations. This result is essential for control theory treated in Chap. 8.

We can now go on to the main objective of the theory: flutter analysis next.

Divergence Speed

We begin with the divergence speed (4.42) we have already seen in connection with

the steady-state or static solution in Chap. 4. Here we examine it from the dynamic

side as corresponding to zero frequency.

Thus here we consider U such that d.M; 0; U / D 0. Setting D 0 is equivalent

to setting all the time derivative terms to be zero.

Theorem 5.36.

p

d.M; 0; U / D cosh ` w4 ;

5.7 Flutter Analysis 231

where

b 2 .1 C 2a/ 2

w4 D p U :

GJ 1 M 2

Proof. We calculate wij .0/: Using:

O :/ D p 2

A.0; T wO a .0; :/;

1 M2

r

O 2 1x

A.0; :/ D p jxj < 1:

1M 2 1Cx

Hence

2

w11 .0/ D p ;

1 M2

w12 .0/ D p ;

1 M2

w21 .0/ D p ;

1 M2

w22 .0/ D 0:

w1 D 0;

w3 D 0;

2

w2 D bU 2 p ;

1 M2

b 2 .1 C 2a/ 2

w4 D p U ;

GJ 1 M 2

s

b 2 .1 C 2a/ 2

d.M; 0; U / D cosh ` p U : t

u

GJ 1 M 2

p

Hence d.M; 0; U / D 0 yields cosh ` w4 D 0:

Or s !

b 2 .1 C 2a/ 2

cos ` p U D 0:

GJ 1 M 2

232 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

Hence p

2 1=4 1 GJ

U D .2n C 1/ 1 M p ;

2b` .1 C 2a/

n positive integer and hence the divergence speed:

p

1=4 1 GJ

Ud D 1 M 2 p ; (5.154)

2b` .1 C 2a/

which of course checks our previous result based on the time-invariant solution

(5.42).

speed at the divergence speed. We want to calculate the nearest aeroelastic mode as

we increase the speed. We show that the derivative with respect to U is positive at

this point, thus as we increase speed the system becomes unstable, with a positive

aeroelastic mode.

d.M; ; Ud C U / D 0:

Because

p

d.M; 0; U / D cosh ` w4 ;

we see that s

@d.M; 0; Ud / b 2 .1 C 2a/

D ` p :

@U GJ 1 M 2

However we run into the difficulty that @d.M; 0; Ud /=@ is not defined at D 0

singularity of the function at D 0! We hence use the perturbation formula: for

small

Z !

`

D det P e A.0/

QCP e A.0/.`s/

.A./ A.0//e A.0/s

Qds ;

0

where we may use the approximation for small for the wij ./ given in [21],

following the expansion in Sect. 5.4.

Here, however, we follow a less computation-intense type of approximation.

Thus note that d.M; ; U / is a function of the variables, w1 ; : : : w4 and we may

denote it d.w1 ; : : : w4 / We note that

p

d.0; w2 ; 0; w4 / D cosh ` w4

5.7 Flutter Analysis 233

d.M; ; U / D w1 C w3

@w1 @w3

p

C cosh ` w4 : (5.155)

This is in fact a relation we use many times. And as a further approximation we may

simply use

p

d.M; ; U / D cosh ` w4

D cosh

" r #

1

` 2 I C b 2 U 2 .w21 C kw22 a.1 ak/w11 ak.w21 C w12 //

GJ

b

kD :

U

U D U C Ud ; U > 0;

4 2

2 I C b 2 U 2 .w21 C kw22 a.1 ak/w11 ak.w21 C w12 // D GJ ;

`2

where we know that

4 2

b 2 Ud2 .w21 .0/ aw11 .0// D GJ :

`2

Subtracting the bottom from the top we have

2 I C b 2 U 2 w21 w21 .0/ C kw22 a2 kw11 a.w11 w11 .0//

C.w21 .0/ aw11 .0// b 2 U 2 Ud2 ak.w21 C w12 / D 0:

We can approximate the wij here by their limit values at D 0, and obtain the

quadratic equation in :

p

2 1 M 2 I b 2 Ua2 2b .1 C 2a/b 2 .U 2 Ud2 / D 0:

But for U > Ud we see that the term independent of is negative, and the roots are

1 2 2

p b Ua b

1 M I

2

p 2 2 2 p

b Ua b C 1 M 2 I .1 C 2a/b 2 U 2 Ud2 ;

234 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

and we note that there is a positive root which also depends on M and I . This

verifies our claim. The main point is that we wind up with a quadratic equation in

, where the term independent of is negative; this is what continues to hold even

with higher-order approximations.

The divergence speed for nonzero and the transonic dip has been covered in

Chap. 4. So far we have considered the CF case. Following similar lines it is easy to

see that for the FF or CC case the speeds are determined by

p

sinh` w4 D 0

p

2 1=4 1 GJ

Ud D 1 M p ;

4b` .1 C 2a/

Our first objective is to make precise the notion of flutter speed. For this we begin

with a closer examination of the aeroelastic modes. The function d.M; ; U / fixed

M , is analytic in both and U (even though we are interested only in positive values

of U ), and for fixed U , with a branch cut along the negative axis, including zero,

in . For each U there is a countable number of roots of d.M; ; U / D 0 which

we denote by .U /. We consider only nonzero roots, and assume that there are no

multiple roots. If there are, we need to work with each branch in a similar way. We

have in fact seen that there are two near the divergence speed.

We can apply the implicit function theorem by which we see that .M; U / is an

analytic function of U , omitting isolated singularities where

@d.M; ; U /

D 0I d.M; ; U / D 0

@

and in particular we see that:

,

d @d.M; ; U / @d.M; ; U /

.M; U / D j D .M; U /:

d @U @

5.7 Flutter Analysis 235

Also

@2 d.:/ @2 d.:/ d 2 @2 d.:/

C 2

d2 @U 2 @@U dU @2

.M; U / D j D .M; U /;

d2 @d.M; ; U /

@

where of course the denominator cannot be zero. At U D 0, the aeroelastic modes

are the structure modes, and we assume that the damping can be neglected, actually

that the coupling S can be neglected. As before, let us order these modes in terms of

increasing magnitudes, within each classbending or torsionso we can talk for

instance about the first few modes: k .0/ where k .0/ D i !k .0/. For nonzero U

we have difficulty in classifying them. So we use root locus. Thus we consider the

function k .U / starting with the value at zero, the structure mode i !k .0/. Hence we

can talk about a bending mode, or torsion mode. Of course the mode and the

mode shape will change as U increases. We call this function a root locus starting

at i !k .0/. We use the notation

k .U / D k .U / C i !k .U /

and refer to k .U / as the damping termstable mode if the real part is strictly

negative.

We note that if complex, the modes occur in complex conjugate pairs; so we may

only consider the ones with the positive imaginary part for most purposes.

By a stability curve we mean the function k .U /; U > 0 so that we have

instability if it is nonnegative, an unstable mode. Our first result is that the damping

decreases at U D 0, whatever the mode.

Theorem 5.37.

D at every bending mode (5.156)

@U mM

2b 3 a2 C 13

D at every torsion mode; (5.157)

MI

@!k .0/

D 0: (5.158)

@U

Remarks: Note that the slope does not depend on the mode number. This result does

not appear to be in the aeroelastic lore.

Z `

Pe `.ACA/

Q D Pe Q C P

`A

e.`s/A AesA dsQ;

0

236 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

have already considered in Chap. 2

0 1

0 1 0 0 0 0

B 0 0 1 0 0 0C

B C

B 0C

B 0 0 0 1 0 C

A./ D B m2 C

B EI 0 0 0 0 0C

B C

@ 0 0 0 0 0 1A

2 I

0 0 0 0 GJ

0

and A is the increment to get back to A./. Now, recall from Chap. 2:

P e `A Q D

0
sinw11=4 `Csinhw11=4 `

1

1

cosw11=4 ` C cosh w11=4 ` 0

1=4
1=4
1=4 2w1

2 1=4

B1 C

@ 2 w1 1=4

sin w1 ` C sinh w1 ` 1

cos w1 ` C cosh wl 1=4 ` 0 A

2

p

0 0 cosh w4`

and

d0 D determinant P e`A Q

1
hp i

D 1 C cos w11=4 ` cosh w11=4 ` cosh w4`

2

and the perturbation term (see [21])

Z `

P e.`s/A A esA dsQ D

0

0 1

0

B C

B 0 C

B p C

B w3 2e w4` w11=4 pw42epw4` ;w11=4 pw4ew11=4 ` .pw1Cw4/Cew11=4` .pw1Cw4/2.pw1w4/sinw11=4` C

B C

B C

B 4w11=4 .w1w42 /

C

B C

B 0 C

B C

B C

B 0 C

B p p p p p p p C

B w3 2e w4` w12e w4` w1Cew1 ` . w1Cw4/Cew1 ` . w1Cw4/C2. w1w4/cosw11=4` C

1=4 1=4

B p C

B .4 w1.w1w42 // C

B C

B p p p p p p p C

B w2 2e w4` w11=4 w4C2e w4` w11=4 w4Cew11=4 ` . w1Cw4/ew11=4 ` . w1Cw4/C2. w1w4/sinw11=4` C

B C

B .4w11=4 .w1w42 // C

B C

B p p p C

B w2.. w1Cw4/cosw11=4` C. w1Cw4/coshw11=4` 2w4 cosh w4`/ C

@ 2.w1w42 / A

0

5.7 Flutter Analysis 237

where . : ; : / is a 3 by 3 matrix depending only on ; .

From (5.6) we can calculate that the derivatives we need:

@d @d0 @ @d0 @ @

D C C .w2 w3 Det:. ; //;

@U @ @U @ @U @U

@d @d0 @ @d0 @ @

D C C .w2 w3 Det:. ; //:

@ @ @ @ @ @

And hence at U D 0:

@w4

d.U /

D @U at a pitching (torsion) mode

dU @w4

@

@wr1

D @U at a bending (plunge) mode:

@w1

@

Using:

@wij @wij b

D ;

@U @k U2

@wij @wij b

D ;

@ @k U

we have:

@w1 1 @w11

D bw11 kb ;

@U EI @k

@w1 1 @w11

D b C bU w11 C 2m ;

2

@ EI @k

@w1 @w1

b w11 k

@U @k

D ;

@w1 w11 @w11

2m C C b 2

@ k @k

238 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

@w4 3 @c.k/

2b 3 c.k/

k b

@U D @k ;

@w1 b 2 @c.k/

2I C

@ k @k

where

c.k/ D w21 C kw22 a.1 ak/w11 ak.w21 C w12 /

and again the ratio depends only on k. Note that U D 0 means that k D 1.

For M > 0 we see from (5.1) that

2

AOi .; : / ! fi as k ! 1

M

and hence we see that

4

w11 .1/ D ;

M

wij .1/ D 0; i j;

4

w22 .1/ D ;

3M

@wij .k/

k !0 as k ! 1:

@k

Hence

w2 ! 0 as U ! 0.S D 0/;

w3 ! 0 as U ! 0.S D 0/:

Hence

@w1

@U U D 0 D b 4 ;

@w1 2m M

@

which yields (5.3).

Next:

c.k/ 4 4

! C a2 as U ! 0;

k 3M M

@c.k/ 4 4

! C a2 as U ! 0:

@k 3M M

Hence (5.4) and (5.5) follow. t

u

5.7 Flutter Analysis 239

are not continuous with respect to M at zero. In fact we have to deal with the case

M D 0 separately.

Noting that

T .k/ ! 1 as k ! 1;

we have:

@.U / b

U D0 D at every bending mode

@U 2.m C b 2 /

2

b 3 a 12

D at every pitching mode: (5.160)

2 I C .a2 C 18 /

We begin with the stability curve:

Theorem 5.37 shows that the slope of this curve is strictly negative at U D 0. Hence

the speed U at which k .U / becomes zero again for the first time we call the flutter

speed for this mode denoted UF .k/. Thus UF .k/ is defined by:

d

k .UF .k// > 0:

dU

The second condition means that the system becomes more unstable as we increase

the speed. Of course at this point, Im k .UF / is not necessarily D !k and may have

changed considerably.

Flutter Speed

inf

k UF .k/ D UF ; (5.161)

where the infimum is taken over all structure modes.

240 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

infinity.

The situation changes if we modify the definition (as we do) to be instead the

infimum over all aeroelastic modes (rather than merely structure modes) because

this would then include zero. First we prove that the flutter speed in either definition

is positive.

Corollary 5.38. The flutter speed is positive. The infimum in (5.8) is attained.

Proof. Suppose the infimum in (5.8) is zero. Then we can find a sequence of mode

frequencies !n and corresponding speeds UF n such that UF n converges to zero.

Suppose the infimum is attained for some n. Then the corresponding flutter speed

must be positive because the slope at zero speed has to be negative by Theorem

5.37. Suppose then it is not attained. Consider the stability curve for each n:

d

n .0/ < < 0

dU

and by definition

d

n .UF n / 0:

dU

The function .d=dU /n.U /is continuous on finite intervals. Hence for some U

denoted U n the slope

d

n .Un / D 0; where Un < UF n

dU

and Un goes to zero by hypothesis.

Let us look at the corresponding mode sequence !n . Suppose the sequence is

bounded. Then we can find a convergent subsequence with finite limit !1 which

would then attain the infimum. Hence the sequence must be unbounded.

Then

n b

Kn D I d.M; n; Un / D 0

Un

and in as much as !n ! 1; we must have that

jn j ! 1:

,

@ @d @d

D

@U @U @

5.7 Flutter Analysis 241

and

@d @2 d.:/ @2 d.:/ @d

@2

D @U @@U @U 2 @ D .M; U /

2 2

@U @d

@

Now for each n we have, with primes denoting derivative with respect to U :

Z Un

0 D n0 .Un / D n0 .0/ C n00 .U / dU ;

0

where the integral ! 0 and the first term goes to a nonzero value, leading to a

contradiction.

Hence the infimum is attained and is positive. u

t

A computer program for generating stability curves in incompressible flow is

given in the Appendix of this chapter with numerical calculations for the Goland

model. We note that the first torsion mode flutters.

Theorem 5.39. Zero is an aeroelastic mode for 0 M < 1.

Proof. For M D 0:

We note that from Sect. 5.4

O

L./ D0 at D 0:

Hence to prove that zero is an aeroelastic mode we need to find Y such that

AY D 0; Y 0:

x

Y D l x2 D 0;

x2

Ax 1 C Kx1 D 0:

h

Let x1 D ,

EI h0000 .y/ D 0 plus end conditions: (5.162)

But this is precisely the linear steady-state aeroelastic equation treated in Chap. 3.

For a nonzero solution we have an eigenvalue problem, which has a solution for

only a sequence of far field speeds: under CF end conditions

242 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

p

1 GJ

U1 D Un D .2n C 1/ p ;

2b` .1 C 2a/

with .y/ determined by (5.9.) and h.y/ D 0 and also, of course d.0; 0; Un/ D 0.

For 0 < M < 1 we have

AY L.0/Y D 0

x1 h

Y D I x1 D ;

0

4

EI h0000 .y/ C bU 2 .y/ D 0;

M

1

GJ 00 .y/ D b 2 U 2 p .y/; (5.163)

1 M2

plus end conditions. Thus for nonzero solution we have an eigenvalue problem with

solution (under CF end conditions):

p

1 GJ

U1 D Un D .2n C 1/.1 M / 2 1=4

p

2b` .1 C 2a/

u

Remarks: Note that it is required that M < 1. For M D 1; the formula for Un

yields zero. Indeed we have seen that for M D 1 zero is not an aeroelastic mode.

So what is new is that we have evaluated the mode shape corresponding to the

aeroelastic mode equal to zero. And we see that Un is the sequence of flutter speeds

corresponding to the flutter mode zero, which however is not a structure mode.

Moreover if we consider this as corresponding to k D 0, the stability curve

starting at 0 .Un /, we see that the slope is nonnegative, by the Remark under

Theorem 5.37.

In particular this shows that there are zeros other than those given by the root loci

of the structure modes.

Which definition do we use? Note that we do not know whether UF .k/ is < Ud :

This depends on whether a is positive or negativewhether the cg is above

or below the elastic axis; see [6]but of course we have offered no mathematical

proof. We use k D 0 to indicate that the mode frequency is zero.

Note that for the mode zero we were essentially calculating the mode shape .x1

in our notation above).

Note that by our definition, if we include k D 0 in (5.9) we have

5.7 Flutter Analysis 243

The appendix presents a computer program for calculating the flutter speed in

incompressible flow for the Goland model. Here the first torsion mode flut-

ters.

At the present time we have no idea, no formula for determining which mode

will flutter and which wont, without numerical calculation. Fortunately we are

interested in practice only in the first few modes, which makes it doable by

computation.

Now we return to proving the convergence of the many series representations for

the solution of the linear equation. The basic idea here is that as the mode number

increases in the limit they all are like the structure modes fk .0/g. Also we know that

physically the modes cannot grow indefinitely (there are no microwaves) but the

model shows that they are of diminishing importance as the mode number increases.

The roots fk g depend on U1 which we shorten to simply U and use the notation

k .U / when we need to emphasize the dependence on U .

Thus

Normalization

x1 k

Mx1k; x1k D 1I Yk D :

i !k x1 k

Similarly we now use R.; U / to emphasize the dependence on U.

We can express

O

A C L./

O U /;

A.U / D A.0/ C T .U / C L.;

0 0

T .U / D

M1 K.U / Ml D .U /

244 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

O U/ D 0 0

L.; ;

M1 FO .; U / 0

norm as jj goes to infinity in the spectral strip. Our main result is the asymptotic

equivalence of the aeroelastic modes with the modes of the structure.

O

Note that it holds in particular if the convolution term L./ is absent.

Theorem 5.40. For each U

n .U /

! 1 as n ! 1:

n .0/

p

1 M 2 2 C 2M i ! C .1 M 2 /! 2

2 C i!

we exploit.

As jj ! 1 in any strip that houses the roots, the multiplier goes to

M=2 and

2

.; M / ! ;

M

2

w11 ! ;

M

wij !0 i j;

w22 !

We use the same technique as in proving Lemma 5.41.

The (nonzero) roots fn .U /g are defined along the root locus and we have:

Z U

@n .u/

n .U / D n .0/ C du: (5.165)

0 @u

Z

U

@n .u/

du goes to a finite limit as n ! 1:

@u

0

5.7 Flutter Analysis 245

This follows from the calculations above. Dividing by n .0/ in (5.8), we obtain

RU @n .u/

n .U / du

D1C 0 @u

; (5.166)

n .0/ n .0/

where the second term goes to zero with n. Hence the ratio

n .U /

! 1 as n ! 1:

n .0/

t

u

In particular we see that

Re n .U / ! 0 as n ! 1 (5.167)

and that

n .U / D O.n/ (5.168)

because this is true of n .0/.

What we have proved is that the sequences fk .U /g as U varies are

asymptotically equivalent to fk .0/g; see [2] for similar property in control

problems.

Actually we are interested in the flutter speed at a given altitude. The speed of

sound a1 depends on the altitude. Thus the system is stable if

U

U < UF :

a1

The Federal Aviation Administration (FAA) in the United States mandates a 15%

margin. The question of whether the Flutter speed decreases or increases with M

has been of interest from the early days of aeroelasticity. The first attempt at this was

by the pioneer Garrick [38] where he determines an empirical formula drawing on

the Possio integral equation for nonzero M for M up to 0:6 or so. We may express

this as the Garrick formula:

UF .M / D .1 M 2 /1=4 UF .0/:

246 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

Note that according to this, the flutter speed decreases as M increases, and goes to

zero as M goes to 1. This is of course for zero angle of attack. It is remarkable that

there is a marked change for nonzero angle of attack as seen for D 0 in Chap. 4.

Because we are interested in the dependence on M , we now write .M; U / in

place of .U /I and similarly k .M; U /; UF .M; k/ for the kth mode.

It has been observed [19] that a mode which flutters at M D 0 does not flutter

at M D 1, and vice versa. This would indicate that for some value of M there are

multiple roots for the same U . This is true near the flutter speed for k D 0; as we

have already noted. We return to this question in Chap. 6.

Numerical calculations indicate that typically for current commercial aircraft

(heavy wings) the value of k for flutter is small, around 0.1 or so. In as much as

we know that

p

d.M; 0; U / D cosh` w4 ;

we may use the perturbation formula to obtain:

1

d.M; ; U / D .2.2880.4w32w33

5760w49

4w1 w22 w23 w4 C w94 C 2w2 w3 w44 .w1 w24 //

C 1440w2 w3 w4 .4w22 w23 C w64 2w2 w3 w4 .2w1 C w24 //`2

C 120w24 .w2 w3 C w1 w4 /.12w22 w23 C 2w64 C w2 w3 w4 .2w1 5w24 //`4

4w2 w3 w34 .w2 w3 w1 w4 /.24w2 w3 4w1 w4 5w34 /`6

p

C w44 .w2 w3 w1 w4 /2 .w2 w3 w34 /`8 /cosh w4 `

C w2 w3 .1440.12w22w23 12w1 w2 w3 w4 C 5w2 w3 w34 C 4w1 w44 8w64 /

C 2880w4 .3w22 w23 2w1 w44 C 3w2 w3 w4 .w1 C w24 //`2

C 60w24 .w2 w3 C w1 w4 /.5w2 w3 C 4.3w1 w4 C w34 //`4

C 8w34 .w2 w3 C w1 w4 /.21w2 w3 C 26w1 w4 /`6 C 6w44 .w2 w3 w1 w4 /2 `8

60.4w1 w44 .24 C w1 `4 / C w22 w23 .96 C w4 `2 .48 C 5w4 `2 //

p

3w2 w3 w4 .40w24 C w1 .32 C w4 `2 .16 C 3w4 `2 ////cosh2 w4 `

p

C w4 `..2880.4w22w23 C 2w1 w44 C 3w64 C 4w2 w3 w4 .w1 2w24 //

480w4 .17w22 w23 C 2w1 w24 .w1 C 3w24 / w2 w3 w4 .19w1 C 3w24 //`2

48w24 .w2 w3 C w1 w4 /.21w2 w3 C w1 w4 C 10w34 /`4

C 4w34 .w2 w3 C w1 w4 /.w2 w3 C 4wlw4 /`6 C w44 .w2 w3 w1 w4 /2 `8 /

p

sinh w4 ` C 24.w2 w3 w1 w4 /.w24 .120w4 C 20w1 `2 C w1 w4 `4 /

p

C w2 w3 .240 C w4 `2 .50 C w4 `2 /// sinh2 w4 `///: (5.169)

5.7 Flutter Analysis 247

In addition we may use the approximation for wij following [21] using the Neumann

expansion in Sect. 5.4. Thus we have:

2P i k

w11 D p 2 p BesselK.0; k/

1 M2 1 M2

C BesselK.1; k//.BesselI.0; k/

Z 1

BesselI.1; k// k a.s/.BesselK.0; ks/ C BesselK.1; ks//

0

Z 2

C a.s/.BesselK.0; ks/

0

BesselK.1; ks//.BesselI.0; ks/ C BesselI.1; ks//ds ;

Pi k

w12 Dp 1C p

1M 2 1 M2

1 2BesselI.1 ; k/

.BesselK.0; k/ C BesselK.1; k//

k k

Z 1

1 2BesselI.1; ks/

Ck a.s/ .BesselK.0; ks/

0 ks ks

Z 2

C BesselK.1; ks// ds C k a.s/

0

1 2BesselI.1; ks/

C .BesselK.1; ks/ BesselK.0; ks// ds ;

ks ks

2P i 1 k

w21 D p Cp

1M 2 2 1 M2

1 1 1 BesselK.1; k/

BesselI.0; k/ BesselI.1; k//

k k 2 k

Z 1

1 1 1

Ck a.s/

0 ks ks 2

BesselK.1; ks/

BesselI.0; ks/ BesselI.1; ks// ds

ks

Z 2

1 1 1 BesselK.1; ks/

Ck a.s/ C

0 ks ks 2 ks

.BesselI.0; ks/ C BesselI.1; ks// ds;

248 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

1 1 Pi 1

w22 D p p 2P i BesselI.1; k/BesselK.1; k/

1 M2 1 M2 k k

Z 1

1

C a.M; s/ .P i 2P i BesselI.1; ks/BesselK.1; ks//ds

0 ks 2

Z 2

1

C a.M; s/ .P i 2P i BesselI.1; ks/BesselK.1; ks//ds:

0 ks 2

1 p 1

d.M; ; U / D w1 `4 cosh w4 ` C w3 w2 .11520w64

12 5760w94

p

2.5760 w64 C 1440 w74`2 240w84 `4 /cosh w4 `

13=2 p p

C 8640 w4 ` sinh w4 `/ C cosh` w4 :

@d

@UF .0; / @M

D ;

@M @d

@U

b 2 .1 C 2a/ 2

w4 D p U ;

GJ 1 M 2

and hence

@w4

@UF .0; / M Ud

D @M D

@M @w4 2.1 M 2 /

@U

as obtained in [21].

This is consistent with the Garrick formula above but of course it is limited to

small . It is also consistent with divergence as a flutter speed for zero frequency,

except that zero is not a structure mode as we have noted.

5.8 Nonlinear Structure Models 249

Chap. 2.

We are fortunate that for this model in spite of its extreme complexity, the static

solution is the zero solution for the structure and constant air flow. Hence the

linearization yields exactly the same equations as in the case of the linear Goland

model.

Here as we have noted the presence of the gravity terms complicates matters. Of

course they have to be considered only along with nonzero far field velocity, as we

have noted before.

The static solution which is nonzero, is given in Chap. 4, Sect. 4.5. Here we go

on to linearize the solution to the aeroelastic equations about this solution. This

requires substantial effort because the static structure solution is no longer zero. An

important question is the role played by the gravity terms.

Thus we seek the solution to the structure equations in the power series form in

the structure state variables. Using the parameter , we define

0 1

h

x D @v A (5.170)

and determine the linear equation characterizing x.t; :/ by retaining only linear

terms in . Correspondingly .; t; x; z/ satisfies the field equation for every :

@2 .:; / @

C jjr.; :/jj2

@t 2 @t

2

1 U1 jjr.; :/jj2 @.; :/

D a1 2

1C 2 .; :/

a1 2 2 @t

jjr.; :/jj2

r.; : / r (5.171)

2

250 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

and

@ jrj2

.; t; x; z/ D C ;

@t 2

p.; : / D .; : :/

with the boundary conditions detailed in Chap. 3 plus the flow tangency condition

and the Kutta Joukowsky condition.

And

Z b

L.; t; y/ D 1 .; x; y/dx;

b

Z b

M.; t; y/ D 1 .x ab/ .; x; y/dx

b

mh.;

0000 00 00

mRv.; t; y/ C EI 2 v .; t; y/ C .EI 2 EI1 /..; t; y/h.; t; y/ /

R

I .; 00

t; y/ GJ .; t; y/ C .EI 2 EI 1 /h.; t; y/ v.; t; y/00 00

Taking the derivative with respect to at zero (or equivalently equating coefficients

of the first power of ), we see that x.t; :/ satisfies the linear equation:

mh.t;

C o .y/v.t; y/00 /00 D L1 .t; y/;

mRv.t; y/ C EI 2 v0000 .t; y/ C .EI 2 EI 1 /..t; y/ho .y/00

R y/ GJ 00 .t; y/

I .t;

5.8 Nonlinear Structure Models 251

where

Z b

L1 .t; y/ D p1 .t; x; y/dx;

b

Z b

M1 .t; y/ D .x ab/p1 .t; x; y/dx;

b

d

p1 D p.; :/ D 0 0 < tI 0<y<`

d

and

o .y/ D .0; y/: (5.175)

Consider first the case where the air speed is zero, U1 D 0, but we retain the

gravity terms, so that we have that x.0; s/ is the solution of:

A power series solution is given Chap. 4, Sect. 4.5. From which we can see in

particular that the functions are real-valued.

The linear structure dynamics equations then become:

mh.t;

mRv.t; y/ C EI 2 v00 .t; y/ C .EI 2 EI 1 /..t; y/h0 .y/00 C 0 .y/h.t; y/00 /00 D 0;

I .t;

This can be given a Hilbert space formulation as in the case of the Goland model in

Chap. 2.

252 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

0 1

h

x D @vA :

0 1

EI1 h0000

Ax D @ EI2 v0000 A :

GJ 00

Then because the coefficient functions h0 .y/; v0 .y/, and 0 .y/ have continuous

fourth derivatives in 0; `, it follows that

0 1

.v000 C 0 v00 /00

Bx D .EI2 EI1 / @.h000 C 0 h00 /00 A

h00 v000 C h000 v00

B D B D B

.A C B/ D .A C B/

and is closed on the domain of A. Hence the equations can be written in abstract

form:

Mx.t/

R C Ax.t/ C Bx.t/ D 0;

where

M D diagm; m; I :

We note that, as in Chap. 2, A is self-adjoint and nonnegative definite, and Ax D 0

implies x D 0.

Hence we may, as in Chap. 2, introduce the energy norm space

p

H D D. A/H

5.8 Nonlinear Structure Models 253

p p

Y; Z D Ax1 ; Az1 C Mx2 ; z2

where

0 I

AD ;

Ml .A C B/ 0

where we note that D.A/ is contained in D.B/, and

xl

D.A/ D ; x2
H; x1
D.A/ :

x2

xl z

Y AY D ZI XD DI ZD l :

x2 z2

Or

Lemma 5.41.

B.2 M C A/1

is bounded (but not necessarily compact) and the norm

1

jjB.2 M C A/1 jj D 0 for > 0:

2

Proof. Only the last part is new. Here we note B is closed and that the range of

.2 M C A/1 is contained in the domain of A which is contained in the domain of

B. And we can write

1 1

B 2 M C A D BA1 2 MA1 C I ;

2 1 2 2

2 X k

MA1 C I 1 x D x; k ;

C

2

k

kD1

1 1

jj 2 MA1 C I jj D 0

2

254 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

1

1

jjB. MCA/ jj D 0

2

: t

u

2

Lemma 5.42. The resolvent R.; A/ is compact for all in the resolvent set that

includes a half-plane

X D R.; A/Z;

1 1

x1 D 2 M C A I C B 2 M C A M.z2 C z1 /;

x2 D x1 z1 (5.177)

Proof.

The compactness follows from (5.8). t

u

We next proceed to calculate the modes following the procedure outlined in

Chap. 2, Sect. 2.4, for the Goland beam. Here we simplify the analysis by taking

advantage of the fact that mg.EI 2 EI1 / is small.

Let

Y D Colh; h0 ; h00 ; h000 ; v; v0 ; v00 ; v000 ; ; 0 :

Then (5.7) can be expressed:

where

0 1

0 100 0 0 00 0 0

B0 010 0 0 00 0 0C

B C

B0 001 0 0 00 0 0C

B C

Bw 0 0C

B 1 000 0 0 00 C

B C

B0 000 0 1 00 0 0C

A../ D B C;

B0 000 0 0 10 0 0C

B C

B0 000 0 0 01 0 0C

B C

B0 000 w2 0 00 0 0C

B C

@0 000 0 0 00 1 0A

0 000 0 0 00 0 w3

5.8 Nonlinear Structure Models 255

m

w1 D 2 ;

EI 1

m

w2 D 2 ;

EI2

I

w3 D2 ;

GJ

00 00

.h0 C 0 h / ; 0; .h v0 C h0 00 v00 /, which we solve as

00 00 00

Z s

Y .s/ D eA. ./s Y .0/ C eA. ./.s / B. Y ./d;

0

which is a Volterra integral equation that we solve by retaining only the linear term

in B. to obtain:

Z s

Y .s/ D eA. ./s Y .0/ C eA. ./.s / B. eA. ./ Y .0/d;

0

so that

Z !

`

Y .`/ D eA. ./` C eA. ./.` / B eA. ./ d

. Y .0/:

0

The eigenvalues for the CF end conditions then are the zeros of

Z !

`

d./ D det P eA. ./` C eA. ./.` / B. eA. ./ d Q;

0

0 1

000 00

B0 0 0 0 0C

B C

B1 0 0 0 0C

B C

B0 1 0 0 0C

B C

B C

B0 0 1 0 0C

P DB C:

B0 0 0 1 0C

B C

B0 0 0 0 0C

B C

B0 0 0 0 0C

B C

@0 0 0 0 0A

000 01

The main point to note here is that d./ is an entire function of and has a countable

number of zeros with no limit point in the finite plane.

256 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

Let us see what we can say next about the perturbed eigenvalues.

First we see that

D i !k D .1 C cosh k cos k /.1 C coshk cosk /cos!k` ;

1=4

1 EI1

!k D k

` m

or

1=4

1 EI2

!k D k

` m

or

s

GJ

!k D .2k C 1/ :

2` I

Let the perturbation be denoted:

Z `

d1 ./ D Det P eA. ./.` / B. eA. ./ dQ:

0

derivative is imaginary. Hence a one Newton step shows that the root continues to be

imaginary. Hence it follows that the perturbed eigenvalues are also imaginary. Usu-

ally for perturbation by a controller, for example, we assume that the eigenfunctions

are approximately the same and only the eigenvalues change. If we do that here we

would find no change in the eigenvalues because Bk; k D 0.

Next let us linearize the potential equation about the static solution. Let 0 .:; :/

denote the static solution, and .t; :; :/ the linearized potential. Then .t; :/ satisfies

the linear equation with nonconstant coefficients:

5.8 Nonlinear Structure Models 257

2

@2 @r 1 U1 jro j2

C 2r0 D a1 1 C 2

2

@t 2 @t a1 2 2

@

. 1/0 r0 r C

@t

jr0 j2

r0 r.r0 r/ r r

2

@.t; x; 0/ P y/ C .x ab/.t;P y/ @0 .t; y/

D h.t;

@z @x

0 .y/.@.t; x; 0// @x; jxj < b:

no longer constant. So we seek only an approximate solution, as in the case above

for the structure. Thus we break up the linearized equation with

00 .x; z/ D xU

and correspondingly

@.t; x; 0/ P y/ C .x ab/.t;P y/ @0 .t; y/

D h.t;

@z @x

C 0 .y/.@ .t; x; 0//=@x : (5.178)

Now .t; : / satisfies the field equation:

@2 @2 2

2 @ @2

C 2U C a1

2

.1 M / C

@t 2 @x@t @x 2 @z2

2

@ U1 jr0 j2

D '.t; x; z/ D .r00 r0 /: r C . 1/

@t 2 2

258 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

@

. 1/0 r0 r C .r0 r00 /

@t

jr0 j2

r..r0 r00 / r/ r r:

2

So we express the solution as the sum of two: One (denote it F ) that satisfies

the nonhomogeneous field equation with zero boundary conditions and the other

(denote it B ) that satisfies the homogeneous field equation:

@2 @2 2

2 @ @2

C 2U C a1 .1 M / 2 C 2 D 0

2

@t 2 @x@t @x @z

The solution to the first is given by (as in [7])

Z t Z

F .t; x; z/ D d L.t ; x ; z /'.; ; /dd;

0 R2

where

1 1

L.t; x; z/ D p s ;

2

2a1 1 M2 2 1 x2 z2

t U x

C 2

c12 1 M2 c12 c2

where

c12 D a12

.1 M 2 /I c22 D a1

2

;

Z t

z A.t ; /d

B .t; x; z/ D d

0 ..x U/2 C z2 /

and

A .t; :/ is the solution of the Possio equation:

Z b

P y/ C .x ab/.t;

P .t ; x /A.; /dd D h.t; P y/

b

Z

@0 d t

.t; y/ C 0 .y/ A.t; x/ C A.t ; x U/d ;

@x dt 0

5.8 Nonlinear Structure Models 259

This equation is a little bit more involved than the standard TDP, because of the

function 0 .y/ but we can still take advantage of our solution of the Possio equation

and rewrite the equation as

where

Z

d t

LA D A S.U/A.t ; : /d;

dt 0

w.t; x/ D h.t; P y/ C @0 .t; y/

P y/ C .x ab/.t; for fixed y:

@x

And hence

A D .I C 0 .y/ .M /L/1 .M /w;

which may be approximated by

Hence we have:

Z tZ

.t; x; z/ D B .t; x; z/ C dL.t ; x ; z /'.t; ; /dd;

0 R2

which is a Volterra equation in the time domain, and has the solution (up to the first

term):

Z t Z

.t; x; z/ D B .t; x; z/ C d L.t ; x ; z /:

0 R2

2

@ U1 jr0 j2

.r00 r0 / . rB C . 1/ B

@t 2 2

. 1/0 .r0 :rB / .r0 r00 / r..r0 r00 / rB /

r.jr0 j / rB dd:

2

What is unique about this example is that we have modes (eigenvalues) even

though the system is not linear which continue to be the modes for the linearized

approximation.

We stop here because further analysis would involve explicit use of the function

0 .., . /.

260 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

in Incompressible Flow: Goland Model

Unsteady Aerodynamics Using Full 2D Continuum Model

(*Initialization & Output Log File*)

Off [General::"spelll", General :: "spell"];

(*Initialize*)

ClearGlobal[ ] :=ClearAll "Global0 "I Clear Derivative];

Remove Global0 I /I

ClearGlobal[ ];

sciNumut[num ] := ToString [MantissaExponent

[Abs[num]] [[1]]] <>

"E"<> ToString" [MantissaExponent[Abs[num]] [[2]]];

writeNum name ; num; desc W "" WD name <>" = n t"<>

ToString[N[num]] <> "n t" <> desc <>"n n";

sLogFileName =

"z. RunLog."<> StringReplaceList Extract ["FileName"/.

nbInfo, 2],

". nb" ! ""];

nbFileName = Extract["FileName"/. nbInfo, 2];

nbDir = DirectoryName "ToFileName["FileName"/. nbInfo]];

nbFilePath =

"FileName"/. nbInfo /. FrontEnd FileName dir ;

fname ; $ ToFileName [dir, fname];

Setoptions [EvaluationNotebook[ ], WindowTitle ! Dynamic

[nbFileName <> sWinTtl]];

SetDirectory [nbDir];

sTimeStampForm = f"Year", "-", "Month", "-", "Day",

",", "Hour", ".", "Minute", ".", "Second",",",

"Millisecond"g,

sOutFile = DateString Join [fsLogFileName, ", "g,

sTimeStampForm]];

sLogFile = sOutFile <>".xls";

sPDFFile = sOutFile <>". pdf";

sStartTime = DateString Join [f"Start time: ntg,

sTimeStampForm, f"nng]];

5.9 Appendix: Computer Program for Flutter Speed in Incompressible Flow... 261

dtStartTime = DateList [ ]; (*raw time stamp for run

time

calculation *)

fTimeUsedO = TimeUsed [ ]; (*total CPU processing time

used by

Mathematica Kernel *)

m = 0.7; (* mass per unit length *)

a = -0.3; (* location of elastic axis *)

b = 3; (* half-chord length *)

1 = 20; (* span *)

S = 0.447;

I D 1:943I

GJ D 2:39 106

EI D 23:6 106 I

mt D 0I

st D 0I

Iyt D 0I

It D 0I

ghi D 0I gi D 0I steu D 010I numu D 100I

= . 0022973;

b2

D I

m

2:39 10 6

Structure *)

sWinTtl =": Calc lst bending & 1st torsion of

structure...";

s

xD I

q

mb

r D mbI2 I

Iyt 2

n1 D EI C gh

EI I n2 D mt=EI2I n3 D St =EI2 I

g

n4 D St =GJ2 I n5 D It =GJ2 C I

GJ

m

tw1 D 2 .1 C /I

EI

m

tw2 D b2 .x a/I

EI

262 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

m

tw3 D b2 .x a/I

GJ

m 2 2 2 1

tw4 D b r C Ca 2

I

0 GJ 1 8

0 100 0 0

B 0 0 1 0 0 0C

B C

B C

B 0 0 0 1 0 0C

ADB C =: U > 0I

Btw1 0 0 0 tw2 0C

B C

@ 0 0 0 0 0 1A

tw3 0 0 0 tw4 0

0 1

0 n1 1 0 0 0

p3 D @n2 0 0 1 n3 0A I

n4 0 0 0 n5 1

0 1

000

B0 0 0 C

B C

B C

B1 0 0 C

q3 D B CI

B0 1 0 C

B C

@0 0 0 A

001

2

r

2 EI

!b1 D .0:597/ 2 I

r ` m

GJ

!t1 D I

I 21 r

2 EI

!b2 D .1:49/2 2 I

r ` m

GJ 3

!t2 D I

I 21

!2 D !=:FindRootg DD 0;f!; !b1; !b1 C 0:01gI

!3 D !=:FindRootg DD 0;f!; !b2; !b2 C 0:1gI

!1 D !=:FindRootg DD 0;f!; !t1; !t1 C 0:1gI

!4 D !=:FindRootg DD 0;f!; !t2; !t2 C 0:01gI

mode D Chopf!1; !2; !3; !4gI

mode=2= I

(*System Equations*)

sWinTtl D W Calc system equations : : : I

ClearAllA; p3; q3I

z D b=UI

w11 D Pi z C 2 Pi T;

w12 D Pi TI

w21 D Pi TI

w22 D Pi=2.1 C z=4 T/I

5.9 Appendix: Computer Program for Flutter Speed in Incompressible Flow... 263

w1 D 1=EI.2 m C Ubw11/I

w2 D 1=EI.2 S C bU2 ..1 az/w11 C zw12//I

w3 D 1=GJ.2 S C b2 U.w21 aw12//I

w4 D01=GJ.2 I C b 1 U .w21 C zw22 a.1 az/w11 az.w21 C w12///I

2 2

0 100 0 0

B 0 0 1 0 0 0C

B C

B C

B 0 0 0 1 0 0C

ADB C

Bw1 0 0 0 w2 0C

B C

@ 0 0 0 0 0 1A

w3 0 0 0 w4 0

0 1

0 n1 1 0 0 0

p3 D @n2 0 0 1 n3 0A I

n4 0 0 0 n5 1

0 1

000

B0 0 0 C

B C

B C

B1 0 0 C

q3 D B CI

B0 1 0 C

B C

@0 0 0 A

001

1

d0 D Detp3:MatrixExpA1:q3=:Cz > I

2

d1 D Detp3:MatrixExpA1:q3=:gh > ghi=:g > gi=:

BesselK1; z

Cz > I

BesselK0; z C BesselK1; z

(*Root Loci*)

sWinTt1 =": Calc Root Loci w/ param U.";

ClearAll; gh; g; UI gr D fgI tp D fgI

Plot Titles*)

sPlotTtl = fst0 <> "1st Torsion"; st0 <> "1st Bending";

st0 <> "2nd Bending"; st0 <> "3rd Bending"; st0 <> "2nd

T orsi on"I gI

10 D "; Listing for"I

11 D f10 <> "1st Torsion"; 10 <> "1st Bending"; 10 <>

"2nd Bending";

10 <> "2nd Torsion gI

Do U D 0I ht D fgI rp D fgI ip D fgI tmp D modekiI

WriteString [strm, "UntSigma=Re(L)ntf=Im(L) ntRe/Imnt

|dL|n

tmoden n"];

264 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

( column headers )

Do U D U C steuI D =:FindRootd1 DD 0, f;

tmp; tmp C 0:010 igI tmp D I Um D UI

sWinTt1 =": Calc Root Loci w/ param U = "<> ToString[U];

WriteString [strm, StringForm ["" n t"n t"n t"n t"n t"n

n",

NumberForm[Um, 10], NumberForm[Re[], 10], NumberForm

[Im[], 10],

NumberFormRe=Im; 10; sciNumOutAbsd1,

NumberForm[mode[[k]],

10]]];

Re

rp = Append rp; Um; I

Im

Im

ip = Append ip; Um; I

2

Re Im

ht = Append ht; ; ,

Im 2

fj,

1, numug

;

sTimeUsed = "CPU time used (min): nt"<> ToString

[(TimeUsed

[]-- fTimeUsed0)/60.0]<> "nnnn"; WriteString [strm ,

sTimeUsed];

tp = Append [tp, frp, ipg];

p1 = ListLinePlot [rp, AxesLabel ! f"U"; "Re"gI

p2 = ListLinePlot [ip, AxesLabel ! f"U"; "Im"gI

p3 = ListLinePlot [ht, AxesLabel ! f"Re"; "Im"gI

Print [GraphicsGrid ffp1; p2; p3gg, ImageSize ! Full,

Frame ! True,

FrameStyle ! Directive [Blue, Dotted], PlotLabel !

sPlotTt1

k;

fk; 1; 4g

I

sEndTime = DateString[Join[f"End time: nt"g,

sTimeStampForm, f"nn"gI

sTimeUsed =

"CPU time used (min): nt" <> ToString

[(TimeUsed[]-fTimeUsed0)/60.0] <> "nn"IWriteString[strm,

sStartTime, sEndTime, sTimeUsed];

(*Print time stamps for performance review*)

WriteString[strm, writeNum["m", m, "mass per unit

length"] <>

writeNum["a", a, "location of elastic axis"]<>

writeNum["b", b, "half-chord length"] <> writeNum["1",

1,

"wing span"]<>

5.9 Appendix: Computer Program for Flutter Speed in Incompressible Flow... 265

["GJ", GJ]<>

writeNum["EI", sciNumOut[EI]] <> writeNum["mt", mt] <>

writeNum

"S t"; S t<>

writeNum "I yt"; Iyt <> writeNum "I t"; It <> writeNum

["ghi", ghi]<>

writeNum ["g-alpha-i", gi] <> writeNum ["steu", steu]

<>

writeNum ["numu", numu] <> writeNum ["rho", , ""]<>

writeNum

["kappa",]];

Close [strm];

NotebookPrint [EvaluationNotebook[], nbDir <> sPDFFile];

(*Save nb file with graphs as PDF*)

sWinTt1 =": Done! See log file.";

0.4

14 14

0.3 13 13

0.2 12 12

0.1 11 11

U U Re[]

200 400 600 800 1000 200 400 600 800 1000 0.1 0.2 0.3 0.4

Re[] Im[] Im[]

U

200 400 600 800 1000 8.8 8.8

0.5 8.6 8.6

8.4 8.4

1.0 8.2 8.2

8.0 8.0

1.5 7.8 7.8

U Re[]

2.0 200 400 600 800 1000 2.0 1.5 1.0 0.5

U 54.4 54.4

200 400 600 800 1000 54.2 54.2

0.005

0.010 54.0 54.0

53.8 53.8

0.015

53.6 53.6

0.020

53.4 53.4

0.025 U Re[]

0.030 200 400 600 800 1000 0.030 0.025 0.020 0.015 0.010 0.005

U 37.6 37.6

200 400 600 800 1000

37.4 37.4

0.05 37.2 37.2

37.0 37.0

0.10 36.8 36.8

36.6 36.6

U Re[]

0.15 200 400 600 800 1000 0.15 0.10 0.05

266 5 Linear Aeroelasticity Theory/ The Possio Integral Equation

It is humbling to note that in spite of all the theory developed, we still cannot answer

some of the simplest of questions. For example: which mode is going to flutter first

without carrying out a computer program for the given parameters? The dependence

of the flutter speed on the parameters is just a little too complicated. We do have,

however, a closed-form formula for the divergence speed which can give some idea

of the range of the flutter speed but not much more. This is further confounded by

the fact that in the case of axial flow treated in Chap. 10 the divergence speed is

simply not defined!

The Possio integral equation, which is the heart and soul of our theory, was

derived by Possio in 1938 [37] for the oscillating wing; it was customary at

that time to distinguish between the steady oscillatory motion and the transient

unsteady motion, a distinction that persists in the aeroelastic literature even today.

The Laplace Transform version that encompasses both was given in 2003 in [4].

It is not merely a matter of replacing i ! by . The Fourier transform integrals in the

original Possio version are not convergent whereas the Laplace transform integrals

are shown to be convergent and the Fourier transform version obtained by taking

limits as in modern theory [10, 41]. The well-known book by Fung [47] claimed

that the existence of a solution was proved but no reference was given. It was not

until 1976 that an unequivocal statement appeared in [49] in the negative but the

Fung assertion was believed by most of the aeroelasticians. In fact when I began

my research many of my colleagues in aeroelasticty would ask me, Why are you

doing this? It is all known! Moreover, a successful computational algorithm by

Rodden [36] has replaced the analytical version in practice even though no proof of

convergence has been given.

A radically new version that employs Fourier transforms in the space domain was

given in [4] with a simple explicit function (in contrast to the two-page statement

involving Hankel functions in [6]). In particular this shows the lack of analyticity

in M and k (the normalized frequency) thereby questioning many of the early

approximations in M and in k reported in [6] and references therein. In [4] it is

proved that there is a unique solution in Lp , 1 < p < 2 for each M for small

enough jkj. A time domain abstract version was given in 2007 in [5]. Finally the

nonlinear version was given in [14].

The basic idea is that the calculation of the pressure jump given the structure

normal velocity is embodied in the Possio equation and effectively replaces the

Euler field equation insofar as the structure dynamics is concerned.

We have thus an inputoutput problem characterized by the Possio equation,

the recurrent theme throughout this work.

The Possio equation and its various extensions including the nonlinear time

domain version are used systematically in this work, even for the incompressible

case in place of the classical theory in [6] which draws on the pioneering work of

Theodorsen. The equation was mostly ignored in the literature, with no mention

Notes and Comments 267

in the recent standard references [17] or [5], for example. It is ironic that Possio

was killed in the last Allied air raid on Turin, Italy, in World War II. Indeed even

his name would seem to be largely forgotten in his country of birth because of the

perception that he was a collaborator!

Chapter 6

Nonlinear Aeroelasticity Theory in 2D

Aerodynamics: Flutter Instability as an LCO

6.1 Introduction

Chap. 3, with the structure models both linear and nonlinear, described in Chap. 2

and the isentropic aerodynamics as treated in Chap. 3 with the flow tangency and

the KuttaJoukowsky boundary conditions. Recall that we use continuum models

without immediately approximating them by finite-dimensional models as in all the

current aeroelastic literature.

Our major result is a characterization of Flutter as a limit cycle oscillation with

the flutter speed as a Hopf bifurcation point determined by the linear equations,

obtained by linearizing the nonlinear equations about the equilibrium or rest

structure state and constant air flow. The point of departure is the inputoutput

point of view and the key role is played by the Possio equation which we need to

generalize to the nonlinear case, perforce in the time domain rather than the Laplace

domain. We limit the theory to 2D air flow because almost no results are available

for the Possio equation except in this case. For the same reason the angle of attack

is taken to be zero. Furthermore we consider only M such that 0 < M < 1: The

case M D 0 is treated in Chap. 10.

A basic assumption is that the structure displacements are neglible compared to

the air displacement in the same time. This means that the structure is essentially not

moving compared to the air, which is implicit in the statement of the fluid-structure

boundary conditions.

As in the time invariant case in Chap. 4, which we follow closely, we provide a

constructive existence theorem where the main tool is the power series expansion

in terms of the structure variables. This is unique to our approach and is consistent

with our view that our interest is in the structure dynamics and how it is affected by

the air flow primarily, and in the air flow per se only secondarily.

DOI 10.1007/978-1-4614-3609-6 6, Springer Science+Business Media, LLC 2012

270 6 Nonlinear Aeroelasticity Theory in 2D Aerodynamics

We begin with a statement of the full dynamics with the linear Goland structure

model. Thus we have:

R y/ C S .t;

mh.t; R y/ C EI h0000 .t; y/ D L.t; y/

Z b

D p.t; x; y/dx; 0 < y < `; (6.1)

b

R y/ C S h.t;

I .t; R y/ GJ 00 .t; y/ DM.t; y/

Z b

D .x a/p.t; x; y/dx

b

D @ 1

.t; x; z/ D D C r r; (6.4)

Dt @t 2

@2 @ @

C jrj2 C . 1/

@t 2 @t @t

1 2 jrj2

D a1 1 C

2

2

U jrj 2

r r ;

2a1 2

1 < x; z < 1; excepting z D 0; (6.5)

U1

0 < M < 1I M D :

a1

The boundary conditions are:

1. Flow Tangency.

6.2 The Aeroelastic Equations: Linear Structure Model 271

Allowing for flow discontinuity between the top and bottom of the wing and again

assuming wing displacement to be small compared to that of the air.

@ @1 h

.t; x; 0C/ D P y/ C .x a/P .t; y/

C .1/ h.t;

@z @z

@

C .t; x; 0C/.t; y/ ; jxj < b: (6.6)

@y

And

@ @1 P y/ C .x a/.t;

P y/

.t; x; 0/ D C .1/ h.t;

@z @z

@

C .t; x; 0/.t; y/ ; jxj < b: (6.7)

@x

2. KuttaJoukowski Conditions:

The pressure jump is zero off the wing:

.t; x/ D 0 as x ! b :

an inputoutput problem where the structure velocity wa .t; :/ (the downwash)

can be considered the input and the pressure jump p.t; :/ as the output. The lift and

moment are just linear functionals of p.t; :/.

Heuristically, we can then invoke the Duhamel principle, familiar in electro-

magnetic theory or more generally (see [1, 44]), as the response to a boundary

input in wave motion. Thus p.t; :/ must be a physically realizable response, albeit

nonlinear, extending the theory from the steady-state case in Chap. 4.

272 6 Nonlinear Aeroelasticity Theory in 2D Aerodynamics

1 .t; x; y; z/ D xU ;

where U is the speed parameter, U 0. We omit the subscript 1.

As in Chap. 4, we consider the solution potential .; t; x; y; z/ corresponding

to h.t; y/; .t; y/ where is a scalar-scale parameter, and make it 0 jj 1;

yielding the expansion:

1

X k

.; t; x; z/ D k .t; x; z/; (6.9)

k

kD0

where

0 .t; x; z/ D 1 .t; x; z/ D xU ;

@k

k .t; x; z/ D .0; t; x; y; z/; k 1;

@k

We calculate these by taking derivatives in the field equation with respect to .

We see that 1 .t; x; z/ satisfies the linear field equation:

where

2

@2 @2 2

2 @ 2 @

./ D C 2U a1

2

1 M C a1 : (6.10)

@t 2 @t@x @x 2 @z2

It is interesting to note that the constant does not appear in this equation, and

hence if we are interested only in the linear equation we may take D 1:

More generally, for k ; k > 1, as in Chap. 4, we expand both sides of (6.5) in a

power series, obtaining:

0 1

1

X k 2 1 1 1 X1

@ k @ @ X X k j A

X k j @j

C qk qj C . 1/ k

k @t 2 @t j D0

k j j D1

k j @t

kD1 kD0 kD1

1

!0 0

1 X 1 1

11

X k 1 X i j qi :qj X i

@

D a1

2

k @1 C @ U

i AA

k 2a12 i j @x

kD1 i D1 j D1 i D1

6.2 The Aeroelastic Equations: Linear Structure Model 273

X 1 X

1 X 1

1 @n @m @2 p @m @2 p

.nCmCp/ C

nD0 mD0 pD1

nmp @x @x @x 2 @z @x@z

@n @m @2 p @m @2 p

C C ; (6.11)

@z @z @z2 @x @x@z

where

@2 @2

D C ;

@x 2 @z2

qk D rk ;

@i @j @i @j

qi :qj D C :

@x @x @z @z

These of course reduce to the equations in Chap. 4 upon setting the time derivatives

to zero.

Theorem 6.1. The kth-order potential satisfies the linear nonhomogeneous equa-

tion

.k / D gk1 ;

where gk only involves potentials of order k, and g0 D 0.

0 1

@ @X X

k1 k1

@j

gk1 D Ck;j qkj qj A Ck;j kj

@t j D1 j D1

@t

X

k1

@kj

C .1 /U Ck;j j

j D1

@x

1 XXX 1 1 1

C k qi qj m; i Cj CmD k

2 i D1 j D1

i j m

kD1

XXX

k @n @m @2 p

C

nD1 mD1 pD1

nmp @x @x @x 2

C C

@x @z @x@z @z @x @z@x

274 6 Nonlinear Aeroelasticity Theory in 2D Aerodynamics

@n @m @2 p

C ; nCmCp Dk

@z @z @z2

X

k1

@kp @2 p @kp @2 p

2U Ck;p 2

C :

pD1

@x @x @z @z2

0 1 0 1

1 1 1 1 X1

@ @X X k j X k 2

@ @ X k j

qk qj A D 2U @ qk qj A :

k

C

@t j D0

k j k @x@t @t j D1

k j

kD0 kD1 kD1

And

X 1 X

1 X 1

1 .nCmCp/ @n @m @2 p @m @2 p

C

nD0 mD0 pD1

nmp @x @x @x 2 @z @x@z

@n @m @2 p @m @2 p

C 2

C

@z @z @z @x @x@z

as

1

X X1

@2 p @1 @2 p @1 @2 p

U2 p 2U C pC1

pD1

@x 2 pD1

@x @x 2 @z @x@z

X 1 X

1 X 1

1 .nCmCp/ @n @m @2 p @m @2 p

C

nD1 mD1 pD1

nmp @x @x @x 2 @z @x@z

@n @m @2 p @m @2 p

C 2

C :

@z @z @z @x @x@z

0 1

1

X k 2 1

X k 2 1 X

X 1 m j

@ k @ k @ @

C 2U C qm qj A

k @t 2 k @x@t @t mD1 j D1 m j

kD1 kD1

1 X

X 1

m j @j

C . 1/ 4m

mD1 j D1

m j @t

1

X 1

X 1

X

k k

@2 k

2 kC1 @1 @2 k @1 @2 k

D 2

a1 4k U 2U C

k k @x 2 k @x @x 2 @z @x@z

kD1 kD1 kD1

6.2 The Aeroelastic Equations: Linear Structure Model 275

X1 X 1 1 1 1

j k 1 X X X i j k

C .1 / j q1 qk C qi

j D1

j k 2 i D1 j D1 i j k

kD1 kD1

X 1 X

1 X 1

1

qj k .nCmCp/

nD1 mD1 pD1

nmp

@n @m @2 p @n @m @2 p @n @m @2 p

2

C C

@x @x @x @x @z @x@z @z @x @z@x

1 1

X X n p @n @2 p

@n @2 p

C 2U C : (6.12)

pD1

n p @x @x 2 @z @z2

kD1

where

0 1

@ @X X

k1 k1

@j

gk1 D ck;j qkj qj A Ck;j kj

@t j D1 j D1

@t

X

k1

@kj 1

C .1 /U Ck;j j C k

j D1

@x 2

XXX 1 1 1

qi qj m ; i Cj CmDk

i D1 j D1

i j m

kD1

XXX

k @n @m @2 p @n @m @2 p

C 2

C

nD1 mD1 pD1

nmp @x @x @x @x @z @x@z

@n @m @2 p @n @m @2 p

C C

@z @x @z@x @z @z @z2

X

k1

@kp @2 p @kp @2 p

2U Ck;p C ; n C m C p D k:

pD1

@x @x 2 @z @z2

(6.14)

276 6 Nonlinear Aeroelasticity Theory in 2D Aerodynamics

In particular for k D 1,

@ @1 @1

g1 D 2 jq1 j2 2 1 C 2.1 /U 1

@t @t @x

@1 @2 1 @1 @2 1

4U C :

@x @x 2 @z @z2

t

u

Next we deduce the corresponding boundary conditions.

Boundary Conditions

Flow Tangency

@ @

.; t; x; y; 0C/ D .h.t; y/ C .x a/.t; y//

@z @t

X1

@.; t; x; y; 0C/ @

C .t; y/ k k .t; x; 0C/;

@x @z

kD1

@

D .h.t; y/ C .x a/.t; y//

@t

1

X @

k k k1 .t; x; y; 0C/.t; y/:

@x

kD1

Hence

@ @ @

1 .t; x; 0C/ D .h.t; y/ C .x a/.t; y// U.t; y/ k .t; x; y; 0C/

@z @t @z

@

Dk k1 .t; x; 0C/.t; y/ for k 2;

@x

@ @ @

1 .t; x; y; 0/ D .h.t; y/ C .x a/.t; y// U.t; y/ k .t; x; y; 0/

@z @t @z

Dk .t; y/ for k 2:

@x

6.2 The Aeroelastic Equations: Linear Structure Model 277

Hence

@ @

1 .t; x; y; 0/ D .h.t; y/ C .x a/.t; y// U.t; y/; (6.15)

@z @t

@ @k1 .t; x; y; 0/

k .t; x; y; 0/ D k .t; y/ for k 2: (6.16)

@z @x

KuttaJoukowsky Conditions

case as well for uniqueness of solution.

For this we need to consider as before the acceleration potential first.

We have

@ 1

.:/ D C r r ;

@t 2

1

X @k 1

X 1 1

k

U2 k @k X X j k

.; t; x; y; z/ D C CU C qk qj :

k @t 2 k @x j D1

j k

kD1 kD1 kD1

With

@k .0; t; x; z/

k .t; x; z/ D ;

@k

we have the expansion

1

U 2 X k

.; t; x; z/ D C k .t; x; z/; (6.17)

2 k

kD1

@k X

k1

@k

k .t; x; z/ D CU C Ck;j qkj qj ; (6.18)

@t @x j D1

1

X k

.; t; x; 0/ D k .t; x; 0/; (6.19)

k

kD1

278 6 Nonlinear Aeroelasticity Theory in 2D Aerodynamics

implies that

k .t; x; 0/ D0 for jxj > b for every k:

Next define: (Kussner doublet function)

Ak .t; x; y/ D k .t; x; 0/=U: (6.21)

Then

@1 @1

.U /A1 .t; x/ D 1 .t; x; 0/ D CU : (6.22)

@t @x

And from the boundary conditions

D2 ;

@t @t

@1 @1 .t; x; 0C/

D2 :

@x @x

As in the time invariant case we consider next

Flow Decomposition

The next important step is to take advantage of the fact that (6.14) is a linear

nonhomogeneous equation with nonzero boundary conditions and decompose the

potentials as

k D L;k C 0;k ;

where 0;k satisfies the nonhomogeneous equation (6.14),

z D 0. Although L;k satisfies the homogeneous equation,

.L;k / D 0 (6.24)

particular that 1 D L;1 .

6.2 The Aeroelastic Equations: Linear Structure Model 279

@2 L;k @2 L;k 2

2 @ L;k

2

2 @ L;k

C 2U a1

2

.1 M / C a1 D 0;

@t 2 @t@x @x 2 @z2

we deduce that

@ @

L;k .t; x; 0/ D L;k .t; x; 0C/: (6.25)

@z @z

Hence

@ @

L;k .t; x; 0/ D L;k .t; x; 0C/: (6.26)

@x @x

Hence from (6.27)

2

@L;k

D 0:

@x

In as much as

2

@L;k @L;k .:; 0C/ @L;k .:; 0/

D C

@z @z @z

@L;k .:; 0C/ @L;k .:; 0/

;

@z @z

we have

2

@L;k

D 0 by .6:26/:

@z

Hence we define

@

k .t; x/ D L;k .t; x; 0/:

@z

280 6 Nonlinear Aeroelasticity Theory in 2D Aerodynamics

Theorem 6.2.

X

k1

k .t; x; y; 0/ D @k =@t C U @k =@x C Ck;j .qkj qj /:

j D1

X

k1

Ck;j .qkj qj / D 0; (6.28)

j D1

or

.qkj qj / D 0:

Now the left side

D qkj .:; 0C/ qj .:; 0C/ qkj .:; 0/ qj .:; 0/: (6.29)

Then (6.29),

D kj .:; 0C/ j .:; 0C/ kj .:; 0/ j .:; 0/

C .kj .:; 0C/j .:; 0C/ kj .:; 0/j .:; 0//:

And

kj .:; 0C/ j .:; 0C/ kj .:; 0/ j .:; 0/

D 0 by .6:25/;

and

.kj .:; 0C/j .:; 0C/ kj .:; 0/ j .:; 0//

D .kj .:; 0C/ kj .:; 0//j .:; 0C/ C .j .:; 0C/ j .:; 0//kj .:; 0/;

which by (6.26)D 0.

Hence (6.29) holds, proving the theorem. t

u

6.2 The Aeroelastic Equations: Linear Structure Model 281

Remarks: The importance of (6.27) is, among others, that the formula for the

nonlinear case is the same as for the linear case. We have seen this already in the

time invariant case in Chap. 4 for zero angle of attack. In particular we use it to

calculate the pressure jump:

1

X

p D 1 I D k =k.@k =@t C U @k @x/:

kD1

@ @L;k1 .t; x; 0/

L;k .t; x; 0/ D k .t; y/ for k 2; (6.30)

@z @x

Ak D L;k =U

and

1

X

AD Ak =k (6.31)

kD1

which we show satisfy the linear time domain Possio equation.

2

.1 M 2 /@2 1 =@x 2 C a1 @ 1 =@z2

D 0

2 2

with the flow tangency boundary condition and the KuttaJoukowsky conditions.

The solution technique was to take the Laplace transform in the time domain and

the Fourier transform in the space domain and obtain the Laplace transform version

of the Possio integral equation. Showing that the solution is the Laplace Transform

of a time domain function can be nontrivialsee [10, 41]. Indeed the basic problem

of when a function analytic in a right-half-plane is the Laplace transform of a time

domain function is still largely an open question. For M D 0, for example, we had

to invoke a special result due to Sears.

282 6 Nonlinear Aeroelasticity Theory in 2D Aerodynamics

Here we obtain the time domain version by formally inverting the Laplace

domain equation. This, it should be noted, would not be possible without the use of

the function a.M; s/ introduced in Chap. 5. We begin with the version:

p

2WO a .; :/ D .1 M 2 /PH.I C B.k//A.;

O :/; (6.33)

Z 1

p

B.k/A D .kR.k; D/A/=. .1 M 2 // kR.ks; D/A a.M; s/ds;

2

Z 1

O x/ D

A.; et A.t; x/dt Re: > a ; jxj < b;

0

R.k; D/A.;

Z x

O /d;

e U .x/ A.; 1 < x < 1;

1

whose inverse Laplace transform is

Z x Z t

. .x /=U /A.t ; /d d

1 0

Z tZ x

D .
.x /=U /A.t
; /dd

0 1

Z t

D A.t ; x U/d: (6.34)

0

We note the mixing up here between the time and space variables.

O :/ is the Laplace transform of

Hence R.k; D/A.;

Z t

S.U/A.t ; :/d:

0

Next we need to take the inverse transform of kR.k; D/A.;

the candidate would be

Z t

d=dt S.U/A.t ; :/d in 0 < t; (6.35)

0

6.2 The Aeroelastic Equations: Linear Structure Model 283

which if A.t; :/ is in the domain of D for t > 0 and DA.t; :/ is in X can be defined

as:

Z t

D A.t/ C U PS.U t U/DA.; :/d; 0<t

0

and would be ok. But in our case, we simply cannot assume that A.t; :/ is in

the domain of D. Instead we require that A.t/ be absolutely continuous in t with

derivative in X .

We can then state the time domain Possio integral (TDP) as an equation in X :

Z b h

p

2=. .1 M 2 //wa .t; x/ D 1= 1=.x / A.t; /

b

Z t

p

@=@t ..A.t ; U//=. .1 M 2 //

0

Z 1 i

C a.M; s/A.t ; U=s/ds=s/d d;

2

jxj < b:

h

p

wa .t; :/ D . .1 M 2 //=2PH A1 .t; :/

Z th

p

d=dt .PS.U.t //PA1 .; ://=. .1 M 2 //

0

Z 1 ii

C PS..U.t //=s/PA1 .; :/a.M; s/ds=s d; (6.36)

2

where the sense in which the integral is defined to take care of the singularity at

s D 0 needs elaboration:

The integral

Z tZ 1

S..U.t //=s/A1 .; :/a.M; s/ds=s d

0 2

Lim ! 0

284 6 Nonlinear Aeroelasticity Theory in 2D Aerodynamics

Z tZ

S..U.t //=s/A1 .; :/a.M; s/ds=s d

0

Z tZ 1

M

C p S.U=s/A1 .t ; :/ds=s d

.1 M 2 / 0

Z tZ

M

C p S.U=s/A1 .t ; :/ds=s d;

.1 M 2 / 0 2

so that

Z t Z 1

1= S.U=s/A1 .t ; :/ds=s d

0 2

Z 1 Z t =s

D 1= ds 1=
.S.U
/ S.U
//A1 .t
s; :/d

0 0

Z 1 Z 1

D HA1 .t/ 1= 1=
.S.U
/ S.U
//.A1 .t/ A1 .t
s//dt ds

0 t =s

Z t

D PHA.t; :/ d=dt PH B.t /A./d;

0

where

Z 1

p

B.t/f D .S.U t/f /=. .1 M 2 // C S.U t=s/f a.M; s/ds=s:

2

.M /A D w

Z b h Z t

p

1M =.2/ 2

1=.x / A.t; / @=@t .A.t ; U//

b 0

. Z 1

p

. .1 M 2 // C .a.M; s/ a.M; 0//A.t s; U/ds

1

Z 2 i

C a.M; 0/A.t s; U/ds d d

1

D wa .t; :/ (6.37)

with domain in X and range in X , with the Kutta condition in addition (implicit

everywhere, if not specified explicitly otherwise).

6.2 The Aeroelastic Equations: Linear Structure Model 285

If we assume the time domain version has a solution, then of course we may take

the Laplace transform which will then satisfy the LDP.

Let us explore next how far we can go in terms of solving the time domain Possio

(TDP), inverting the Laplace Transform versions in Chap. 5.

.M /wa where we have seen that in Sect. 5.5 the Laplace transform of the solution

has the form

p

AO D .I C PB.k/P C W .k//1 2=. .1 M 2 //T wO a : (6.38)

p ! !1

.1 M 2 / 2

0 < M < 1 W AO D I C PH.B.k/ B.1// wO a : (6.39)

M M

Lemma 6.3. For A in Lp b; b

; 1 < p < 2: .I C PB.k/P C W .k//A is the

Laplace transform:

Z 1

D ek t d=dtP.B.t/ C W.t//Adt;

0

p

where PW.t/A is the function 1= ..b x/=.b C x//

h Z t

p

1=. .1 M 2 // A.b t C /d

0

Z tZ 1

p

C a.M; s/A.b .t /=s/ .=.2bs C //ds=..b x/s C /

0 0

p

C a.M; s/A..t /=s b/ ..2bs C /=/ds=..b C x/s C /

Z 2

p

C a.M; s/A..t /=s b/ ..2bs C /=/ds=..b C x/s C /d

1

Z 2

C A.t=s .2b C x//a.M; s/ds; jxj < b

0

286 6 Nonlinear Aeroelasticity Theory in 2D Aerodynamics

and

Z 1

p

P B.t/PA D 1=. .1 M 2 //P S.t/PA C P S.t=s/PAa.M; s/ds=s :

2

Proof. To derive Z 1

W .k/A D P

ek t W.t/Adt

0

t

u

we use

p

1=. .1 M 2 //h .k/L .k/

Z 1 . h

p p

D ek t dt 1 . .1 M 2 // 1= ..b x/=.b C x//

0

Z b i

p

..t b C /=.t C b C //1=.t .x // A./d;

b

Z 1 Z b Z t

D ek t A./d .t s.b C //1=

0 b 0

p p

..b x/=.b C x// .=.2bs C //d=..b C x/s C /dt

Z 1 Z b

p

D ek t A./d 1= ..b x/=.b C x//

0 b

p

..t s.b C //=.2bs C t s.b C ///1=..b C x/s C t s.b C //

Z 1 Z b

k t p

D e A./d 1= ..b x/=.b C x//

0 b

p

..t s.b C //=.t C s.b ///dt=.t C s.x //:

Z 2

LC .ks; A/hC .ks; x/a.M; s/ds

0

Z 1 Z b Z 2

p

D ek t dt A./d 1= ..b x/=.b C x//

0 b 0

p

..t s.b C //=.t C s.b ///.a.M; s/ds/=.t C s.x //

Z 1 Z b

k t p

D e dt 1= ..b x/=.b C x// A./d

0 b

Z 2

p

..t s.b C //=.t C s.b ///.a.M; s/ds/=.t C s.x //;

0

6.2 The Aeroelastic Equations: Linear Structure Model 287

and

Z 1

L .ks; A/h .ks; x/a.M; s/ds

0

Z 1 h Z b Z 1

p p

D ek t dt 1= ..b x/=.b C x// ..t sb C s/

0 b 0

. i

.t C sb C s//.a.M; s/ds/=.t s.x // A./d:

Using

Z 1

p p

h .ks; x/ D ek t 1= ..b x/=.b C x// .t=.2bs C t//dt=..b x/s C t/;

0

Z 1 Z b

L .ks; A/ D ek t dt .t s.b //A./d

0 b

and

Z b

j.ks/LC .ks; A/ D eks.bCx/ eks.bC/ A./d

b

Z b

D eks . x/A./d

b

Z 1 Z b

k t

D e dt .t s. x//A./d;

0 b

we have

Z 2

j.ks/LC .ks; A/a.M; s/ds

0

Z 1 Z b Z 2

D ek t dt .t s. x//a.M; s/ds A./d:

0 b 0

Z b Z 2

D .t s. x//a.M; s/ds A./d

b 0

Z b

p

C1= ..b x/=.b C x// A./d

b

288 6 Nonlinear Aeroelasticity Theory in 2D Aerodynamics

Z 2

p

..t s.b C //=.t C s.b ///

0

p

.a.M; s/ds/=.t C s.x // C 1= ..b x/=.b C x//

Z b Z 1

p

..t sb C s/=.t C sb C s//.a.M; s/ds/=.t s.x //A./d

b 0

p p

C1=. .1 M 2 //1= ..b x/=.b C x//

Z b

p

..t b C /=.t C b C //1=.t .x //A./d:

b

,

Z 1

PB.k/PA D ek t P B.t/PAdt;

0

Z 1

p

PB.t/PA D 1=. .1 M 2 //PS.t/PA PS.t=s/PA a.M; s/ds=s:

2

Z t

d=dt P B.t /PA.; :/d

0

Z Z Z !

t b 2

C d=dt d . s. x//a.M; s/dsA.t ; /d

0 b 0

Z b Z 2

Cd=dt a.M; s/A.t s. x/; /dsd

b 0

Z tZ b Z 2

p p

C1= ..b x/=.b C x//d=dt .. s.b C //=. C s.b ///

0 b 0

Z t Z b Z 1

p p

C1= ..b x/=.b C x//d=dt d .. sb C s/=. C sb C s//

0 b 0

p p

C1=. .1 M 2 //1= ..b x/=.b C x//d=dt

Z t Z b

p

d .. b C /=. C b C //1=. .x //A.t ; /d:

0 b

(6.40)

6.2 The Aeroelastic Equations: Linear Structure Model 289

Hence

.I C PB.k/P C W .k//w.k;

O :/

Z 1 Z t

D ek t w.t; :/ C d=dt .P B.
/P C W.
//w.t
; :/d
dt:

0 0

(6.41)

Z t

2

A.t; :/ C d=dt .P B. /P C W. //A.t ; :/d D p T wa .t; :/: (6.42)

0 1 M2

Theorem 6.4. The TDP (6.42) has at most one solution. Any solution of the TDP

(6.42) will satisfy the Kutta condition.

Proof. The Laplace transform of any solution satisfies the LDP which we have

shown has a unique solution for small enough (Theorem 5.16).

Furthermore we have shown that the Laplace transform satisfies the Kutta

condition. t

u

Finally, from the form (6.39):

p !

.1 M 2 /

IC O :/ D 2 w.;

PH.B.k/ B.1// A.; O :/;

M M

p

we have (strong) limit Re ! 1 I C . .1 M 2 / =M PH.B.k/ B.1/// D I .

Hence the time domain transform has a delta function at the origin for nonzero M .

(For M D 0 we also have the delta function derivative. Thus M D 0 is special, not

obtained as the 1imit for small M .) Hence, if the TDP has a solution for M > 0; it

is of the form

Z t

2

A.t; :/ D wa .t/ C P .M; t /wa .; :/d

; (6.43)

M 0

Z 1

p

ek t P .M; t/wdt D .I C . .1 M 2 //=M PH.B.k/ B.1///1 I

0

and P .M; :/ does not have a delta function at the origin. This is useful to us below

in isolating the non-circulatory terms in the aeroelastic structure equations.

290 6 Nonlinear Aeroelasticity Theory in 2D Aerodynamics

p !1

.1 M 2 /

IC PH.B.1// I

M

(because B.0/ D 0)

p

D M=. .1 M 2 //T I;

which checks with the solution in the time invariant case

@L;k1 .t; x; y; 0/

D k .t; y/: (6.44)

@x

The potential 1 .:/ is given by specializing (5.18) to the typical section and

D0W QO

.; QO i !/erz=.k C i !/

i !; z/ D 1=2A.; z > 0;

where

p

r D .M 2 k 2 C 2M 2 ki ! C .1 M 2 /! 2 /:

Hence the LaplaceFourier transform of .@L;1 .t; x; y; 0C//=@x is

Hence

Z t

.@L;1 .t; x; 0C//=@x D 1=2 A1 .t; x/ d=dt A1 .t ; x U/d ;

0

Z 1

1=. C i U !/ D 1=U e U x ei !x dx:

0

Hence Z t

@1 =@x D A1 .t; x/ C d=dt A1 .t ; x U/d: (6.45)

0

1 @

Ak .t; x/ D L;k C L ; jxj < b; (6.46)

U @x

6.2 The Aeroelastic Equations: Linear Structure Model 291

Or

@k =@t D U @k =@x UAk .t; x/; 0 < t; jxj < b

D U @k =@x jxj > b: (6.47)

Z t

k D U S.U/Ak .t /d

0

and Z t

@k =@x D Ak .t; x/ C d=dt Ak .t ; x U/d: (6.48)

0

In (6.47), because

A.t; x/ D 0 for jxj > b;

we have that the integral is zero for U t > .b C x/ and hence for t > .2b/=U

Z .bCx/=U

@k =@x D Ak .t; x/ C d=dt Ak .t ; x U/d: (6.49)

0

.; i !; 0C/ z > 0; (6.50)

where

p

r D .M 2 2 C 2M 2 i! C .1 M 2 /! 2 /;

D =U

and

QOk .; i !; z/ D erz =r QO k .; i !; 0/ z < 0; (6.51)

where

@=@zOk .; x; z/ D O k .; x; z/:

Hence with

@=@x Ok .; x; z/ D Ok .; x; z/;

we have

292 6 Nonlinear Aeroelasticity Theory in 2D Aerodynamics

Again:

D0 jxj > b:

AMOk .; i !/ D . C i !/ QOk .; i !; 0C/ QOk .; i !; 0/

D . C i !/=r MO k .; i !; 0C/ C MO k ; .; i !; 0/ :

Or

1 r MO

Ak .; i !/ D 1=2 MO k .; i !; 0C/ C MO k .; i !; 0/ : (6.54)

2 C i!

Equation (6.54) is then the version of the linear Possio equation valid for k 2,

recalling that .1 r/=.2 C i !/ is the multiplier corresponding to the time domain

operator .M / on into .

Hence we have the time domain equation in

.M /Ak D gk ;

gk .t; x/ D 1=2. k .t; x; 0C/ C k .t; x; 0// jxj < b:

The presence of the factor .t/ which depends on t makes use of the Laplace

transform difficult unlike the time invariant case in Chap. 4 where is a constant.

Hence we need to work in the time domain from now on. But first we have the

frequency domain relations.

By (6.52) and (6.53)

Similarly

.r/= i ! MOk1 .; i !; 0/ D MO k1 .; i !; 0/: (6.56)

Hence, adding, we have:

r= i ! MOk .; i !; 0C/ MOk .; i !; 0/ D MO k .; i !; 0C/ C MO k .; i !; 0/ ;

(6.57)

6.2 The Aeroelastic Equations: Linear Structure Model 293

D0 for jxj > b:

And by (6.54)

i ! r MO i ! MO

MOk D Ak .; i !/ D Ak .; i !/: (6.58)

r C i! C i!

Hence

AMOk .; i !/:

MOk D 1 C

C i!

Or in the time domain:

Z t

k .t; x/ D Ak .t; x/ C d=dt Ak .t ; x U/d; jxj < b

0

D0 jxj > b;

We now define the time domain operator on the subdomain of which is

absolutely continuous in t, with derivative in , into by

Z t

f D gI g.t; x/ D f .t; x/ d=dt f .t ; x U/d; jxj < b:

0

Then we have

k D Ak : (6.59)

Note that

f D 0 implies f D 0:

Hence we can define the inverse 1 by

Z x

1 d

1 f D gI g.t; x/ D f .t; x/ f .t; s/ds; (6.60)

U dt b

Z x

Ak .t; x/ D k .t; x/ C 1=U d=dtk .t; s/ds; jxj < b;

b

294 6 Nonlinear Aeroelasticity Theory in 2D Aerodynamics

Also we need now to define the linear operator with domain and range in

f D hI h.t; x/ D .t/f .t; x/; jxj < b where .t/ is the given torsion angle that

does not depend on the chord variable x.

We only need to consider the case where the torsion angle is bounded (actually

< 1) that is actually linear bounded.

Then we have

2.M /Ak D kP k1 ; (6.61)

where we use

k .t; x/ D 1=2.k .t; x; 0C/ C k .t; x; 0//

and similarly we use the notation:

MOk1 .; i !; 0C/ C MOk1 .; i !; 0/

D i !=r MO k1 .; i !; 0C/ MO k1 .; i !; 0/ : (6.62)

ing; see [55].

Lemma 6.5.

Z 1

i !=.r/ D O

ei !x L.; x/dx; 1 < ! < 1;

1

where

O

L.; O

x/ D L.; x/;

O M M

L.; x/ D K1 x x>0

.1 M 2 /3=2 1 M2

6.2 The Aeroelastic Equations: Linear Structure Model 295

" 2 !#

M .

d p M x

L.t; x/ D t t

2

;

.1 M / dt

2 2 1M U

2

M x M x

t> D0 0<t < x > 0;

1 M2 U 1 M2 U

Let L.t/ denote the bounded linear operator on Lp .R1 / into itself defined by

Z 1

L.t/f D gI g.x/ D L.t; x s/f .s/ds; jxj < 1:

1

Z t

L.M /A D gI g.t; :/ D PL.t /A.; M /d: (6.63)

0

Then we have:

k1 D L.M /.k 1/k2 : (6.64)

Next by (6.58) and (6.59),

i ! MO i!

Ak .; i !/ D M k .; i !; 0C/ C MO k .; i !; 0/ :

C i! r

Or

AMOk .; i !/ D MO k .; i !; 0C/ C MO k .; i !; 0/ :

r

C i!

Hence 2.M /Ak D P k D kP k1 which by (6.64) D kL.M /.k

1/k2 . Hence using (6.61)

Hence we obtain

296 6 Nonlinear Aeroelasticity Theory in 2D Aerodynamics

This generalizes the time invariant case in Chap. 4, where 1=2.M /L.M /

corresponds to multiplication by

A2k D 0

and hence

A2k D 0: (6.66)

Let

1

J D .M /L.M /: (6.67)

2

Then (6.51) yields:

A2nC1 =.2n C 1/ D J n A1

and finally

1

X 1

X

AD Ak =k D J k A1 : (6.68)

kD0 kD0

Assuming that

jjJ jj < 1: (6.69)

Or

Sup j.t; y/j

y

A D .I J /1 A1 : (6.70)

practice it is indeed much less than one radian, so the convergence holds.

We can express (6.70) as the nonlinear version of the time domain Possio

equation.

; 1 < p < 2, t > 0 for determining the pressure

doublet function A.t; x; y/, given the structure state variables, h.t; y/; .t; y/. In

other words it replaces the aerodynamics insofar as the structure dynamics is

concerned in isentropic flow. It is the nonlinear inputoutput relation for the system,

the structure dynamics in air flow.

6.3 The Nonlinear Possio Integral Equation 297

A JA D A1 :

Hence

.M /A .M /JA D wa ;

where

.M /J D L.M /:

Hence

.M /A 1=2L.M /A D wa : (6.71)

Then the nonlinear time domain Possio equation is given by

p Z Z t

.1 M 2 / b

1 @ p

A.t; / .A.t ; U//=. .1 M 2 //

2 b x @t 0

Z 1

C .a.M; s/ a.M; 0//A.t s; U/ds

1

Z 2

C a.M; 0/A.t s; U/ds d d

1

Z tZ b

1=2.t; y/ L.t ; x s/.; y/D.; s/dsd

0 b

where

Z t

d

D.t; x/ D A.t; x/ A.t ; x U/d (6.73)

dt 0

and

. .

d t p

L.t; x/ D.M / ..1 M 2 /2 / q t . .t 2 .M 2 x 2 /

dt t 2 Mx

. Mx

..1 M 2 /2 U 2 ///; for t > ; and D 0

U.1 M 2 /

Mx

0<t < x>0

U.1 M 2 /

L.t; x/ D L.t; x/:

298 6 Nonlinear Aeroelasticity Theory in 2D Aerodynamics

equationthe nonlinear Euler equation. We are not concerned with the air flow

any more. Note also that the nonlinearity in .:; :/ is physically realizable.

Remarks: We have existence and uniqueness for the solution of (6.72) assuming the

regular Possio equation has a unique solution and the torsion angle is small enough.

The solution is given by (6.53).

We use the qualifier nonlinear to indicate that the solution is no longer linear

in the structure state variables, h.t; y/; .t; y/.

Having determined the pressure jump, we can now go on to full nonlinear aeroelastic

dynamics: specialized to the typical section, zero angle of attack, 0 < M < 1, and

linear structure dynamics. We show first that the nonlinear aeroelastic dynamics

formulates as a nonlinear convolution/evolution equation in a Hilbert space, in fact

in the energy space we have already discussed in the previous chapters.

We begin with the series (6.68):

1

X

A D .M /wa C J k .M /wa : (6.74)

kD1

We have already studied the first term leading to the linear convolution evolution

equation in Chap. 5 for nonzero M . Now we have an additional component. We

verify first that there are no additional non-circulatory terms.

Lemma 6.6. There are no non-circulatory terms in

1

X

J k .M /wa : (6.75)

kD1

O

L.k; :/ ! 0 as Re: k ! 1

and therefore there are no delta functions in L.t; :/ and hence J .M /wa has no

non-circulatory components and neither does the sum in (6.76). u

t

6.4 Nonlinear Aeroelastic Dynamics 299

We next calculate the lift and moment corresponding to A(.,.) given by (6.75).

Here we need to bring back the dependence on the structure span variable y.

Thus we have:

x and y.

Hence we need to introduce the space C of functions of the form:

with range in the scalar field such that for each y, 0 < y < `,

And

Z `

.jjf .:; y/jjp /2 dy < 1;

0

where jj:jjp denotes the norm in Lp .b; b/. This is a Banach space with norm:

s

Z `

.jjf .:; y/jjp /2 dy:

0

denote the space of functions such that f .t; :/ is in HE

for each t 0 and

RT

1: 0 jjf .t; :/jjE dt < 1I for each T; 0 < T < 1

and Laplace transformable

R 1 t

2: 0 e jjf .t; :/jjE dt < 1 for > a 0.

We note that the function wa .t; :; y/ defined in (6.4) is in L1 . Define now on

Lp .b; b/:

Z b

LL .f / D U1 f .x/dx;

b

Z b

LM .f / D U1 .x ab/f .x/dx:

b

300 6 Nonlinear Aeroelasticity Theory in 2D Aerodynamics

are in L1 .

Recall now the structure state vector

x .t; y/

Y .t; y/ D ; 0<y<`

xP .t; y/

b1? D 0; U; 1; ab

I b2? D 0; 0; 0; 1

:

And hence

.M /wa D .M /.f1 b1? C f2 b2? /Y .:; y/

;

which yields a linear bounded operator denoted L12 on L1 into :

Let

g.t; x; y/; jxj < b; 0 < y < `

denote the function L12 Y . Then .M /wa is the function

Z t

h.t; x; y/ D g.t; x; y/ d=dt g.t ; x U; y/d;

0

where

Z t

d=dt g.t ; x U; y/d

0

Z bCx

U bCx

D P ; x U; y/d;

g.t t>

0 U

Z t

bCx

D P x U.t
/; y/d
;

g.
; t>

t bCx

U

U

and can be neglected for large t, compared to the first term. And in as much as our

interest is primarily in the asymptotic response we omit it, and thus in what follows

6.4 Nonlinear Aeroelastic Dynamics 301

we set .M / D .M /; so that

1

Jg D .M /L.M /L12Y .:/;

2

where now

and

1

D

.M /.L.M /L12Y .:// (6.80)

2

from now on. With this qualification let us evaluate Jg. We have: L.M /g is the

function

Z t

h.t; :; y/ D .t; y/ L.t /.; y/g.; :; y/d in Lp .b; b/

0

Z t

D B2 .Y .t; y/; Y .; y//L.t /g.; :; y/d;

0

b ? Y1 Y2 b:

We have already used this for h D wa . Recall (6.67) .M /h is the function:

Z t

2

hC P .M; t /h.; :/d;

M 0

Hence we find it convenient now to express it as

Z t

'.M; t /h.; :y/d

0

Z t Z

1

'.M; t / B2 .Y .; y/; Y . ; y//L. /g. ; :; y/d d (6.81)

2 0 0

302 6 Nonlinear Aeroelasticity Theory in 2D Aerodynamics

valid for large t, because we are taking as the identity, and is recognized as a

second-degree polynomial in Y .:; :/ the structure state.

Also

Z Z

1 t

J .M /wa D '.M; t / B2 .Y .; y/; Y .
; y//L.
/

2 0 0

Z

'.M;
s/wa .s; :; y/d
d: (6.82)

0

Z t Z

1

LL .Jg/LL '.M; t /B2 .Y .; y/; Y . ; y// L. /g. ; :; y/d d

2 0 0

can be expressed as

Z tZ

1

D LL .'.M; t /B2 .Y .; y/; Y .
; y//L.
/g.
; :; y//d
d:

2 0 0

(6.83)

And

1

LL .M /L.M / .M /wa.:; :; y/ ;

2

where .M /wa .:; :; y/ is the function (fixing M; 0 < M < 1; in what follows in

this chapter)

Z

g0 .
; :; y/ D '.
s/.f1 b1? C f2 b2? /Y .s; y/ds; (6.84)

0

Z tZ Z

1

LL .'.t /B2 .Y .; y/; Y . ; y//

0 0 0 2

L. /'.M; s/.f1 b1? C f2 b2 //Y .s; y/dsd d:

More generally:

LL .J k g0 /; k 1; (6.85)

6.4 Nonlinear Aeroelastic Dynamics 303

Z tZ

1

LL .'.t /B2 .Y .; y/; Y .
; y//L.
/gk1 .
; :; y//d
d (6.86)

2 0 0

Hence we can express:

as

Z tZ 1 Z 2k

1

:: LL '.t 1 /b ? Y .1 ; y/b ? Y .2 ; y/L.1 2 /

2k 0 0 0

::::::::::::::::::::::::

'.2k2 2k1 /b ? Y .2k1 ; y/b ? Y .2k ; y/L.2k1 2k /

:::::::::::::::::::::::: (6.87)

Z tZ 1 Z 2k

1

as D k :: LM .'.t 1 /B2 .Y .1 ; y/; Y .2 ; y//L.1 2 /:

2 0 0 0

::::::::::::::::::::::::

'.2k2 2k1 /B2 .Y .2k1 ; y/; Y .2k ; y//L.2kl 2k /

This completes the calculation of the lift and moment. So we can proceed to the

following.

linear case in Chap. 5, using the same notation therein.

We state this in the form of a theorem.

304 6 Nonlinear Aeroelasticity Theory in 2D Aerodynamics

Theorem 6.7. The dynamics of the structure under aerodynamic loading can be

expressed as

YP .t/ D AY .t/ C N .t; Y .// t > 0; (6.89)

where Y .t/ is the structure state at time t, N .t; Y .:// defines a physically realizable

nonlinear convolution, and for Y .:/ such that for each T; 0 < T < 1,

is

m.T / < 1; 0 t T; 0<y<` (6.90)

is defined by the Volterra expansion: valid for 0 < t < T

Z t

N .t; Y .:// D L.t /Y ./d (6.91)

0

1 Z t Z

X 1 Z 2k

C L2k .t 1 ; 1 2 ; :; 2k 2kC1 I

kD1 0 0 0

Y .1 /; :Y .2kC1 /d1 : : : d2kC1 ;

where

0

D ; (6.92)

M1 F2k .t1 ; t2 ; t2kC1 I Y1 ; Y2 ; Y2kC1 /

where

2 3

LL .'.t1 /B2 .Y1 ; Y2 /L.t2 /:'.t3 /B2 .Y3 ; Y4 /L.t4 /

6 7

6 7

6 7

6 '.t2k1 /B2 .Y2k1 ; Y2k /L.t2k / 7

6 7

1 6 6

'.t2kC1 /.f1 ; b1? C f2 ; b2? /Y2kC1 7

7:

D k6 (6.93)

2 6 LM .'.t1 /B2 .Y1 ; Y2 /L.t2 / '.t3 /B2 .Y3 ; Y4 /L.t4 / 7 7

6 7

6 7

6 7

4 '.t2k1 /B2 .Y2k1 ; Y2k /L.t2k / 5

'.t2kC1 /.f1 ; b1 C f2 ; b2 /Y2kC1

? ?

Proof. In what follows we fix T and so abbreviate m.T / to simply m. We start with

the linear part of (6.90) given in Sect. 5.6:

6.4 Nonlinear Aeroelastic Dynamics 305

Z t

YP .t/ D AY .t/ C L.t /Y ./d; (6.94)

0

where

x .t/

Y .t/ D ;

xP .t/

m S

MD ;

S I

0 I

AD :

M1 .A C K/ M1 D

0

L.t/Y D for Y in R4 :

M1 F .t/B Y

The nonlinear part is determined by the lift and moment calculated for the pressure

doublet A.:; :/ determined as a solution of the nonlinear Possio equation (6.73).

Under condition (6.90) we can use the expansion (6.75) for the pressure doublet

and our calculation of the corresponding lift and moment above.

This yields the kth term for k 1 in the Volterra expansion (6.90):

1

X

N .t; Y .:// D gk .t; Y .://; (6.95)

kD1

Z tZ 1 Z 2k

gk .t; Y .:// D ::: L2k .t 1 ; 1 2 ; : : : 2k 2kC1 I (6.96)

0 0 0

where

0

D

M1 F2k .t1 ; t2 ; : : : t2kC1 I Y1 ; Y2 : : : Y2kC1 /

and

306 6 Nonlinear Aeroelasticity Theory in 2D Aerodynamics

2 3

LL .'.t1 /b ? Y1 b ? Y2 L.t2 / '.t3 /b ? Y3 b ? Y4 /L.t4 /

6 7

6 7

6 7

6 ? ?

'.t2k1 /.b Y2k1 b Y2k /L.t2k / 7

6 7

1 6 '.t2kC1 /.f1 b1 C f2 b2 /Y2kC1

? ? 7

D k6 7

2 6 LM .'.t1 /b Y1 b Y2 L.t2 / '.t3 /b Y3 b Y4 L.t4 / 7

6 ? ? ? ?

7

6 7

6 7

6 7

4 '.t2k1 /b ? Y2k1 b ? Y2k L.t2k / 5

'.t2kC1 /.f1 ; b1? C f2 ; b2? /Y2kC1

and

0

gk .t; y/ D ; (6.98)

M1 F2k .t; y/

where

F2 kL .t; y/

F2k .t; y/ D

F2 kM .t; y/

Z tZ 1 Z 2k

1

F2kL .t; y/ D k :: LL .'.t 1 /b Y .1 ; y/b Y .2 ; y/L.1 2 /:

2 0 0 0

::::::::::::::::::

Z tZ 1 Z 2k

1

F2kM .t; y/ D k :: `M .'.t 1 /b Y .1 ; y/b Y .2 ; y/L.1 2 /:

2 0 0 0

::::::::::::::::::

'.2k2 2k1 /

b Y .2k1 ; y/b Y .2k ; y/L.2k1 2k /

'.2k 2kC1 /.f1 b1 C f2 b2 /Y .2kC1 //d1 : : : : : : d2kC1 :

6.4 Nonlinear Aeroelastic Dynamics 307

To prove convergence of the series in (6.91), we need to get bounds for the terms.

Let us look at the case k D 1. We have

Z t

1

F2L .t; y/ D `L '.t 1 / b Y .1 ; y/v1 .1 ; :; y/d1 ;

2 0

where

Z t

v1 .t; :; y/ D L.t s/b Y .s; y/v0 .s; :; y/ds;

0

Z t

v0 .t; :; y/ D '.t s/.f1 . : /b1 C f2 . : /b2 /Y .s; y/ds

0

and

Z t

jjv1 .t; :; y/jj m l.t s/a0 .s; :; y/ds;

0

1

jF2L .t; y/j jj`L jjm2 .a.:/ l.:/ v1 .:; y//.t/; (6.99)

2

where denotes convolution, and

.t/ D jj'.t/jj t 0;

l.t/ D jjL.t/jj t 0;

Z t

a0 .t; y/ D .t s/jj.f1 b1 C f2 b2 /Y .s; y/jjp ds; t 0:

0

1

F2kL .t; y/ k jj`L jjm2k ..:/ l.://k a0 .:; y/.t/

2

with a similar estimate for the moment.

Hence

1

X X 1

1

F2kL .t; y/ jj`L jjm2k .a.:/ l.://k .t/ a0 .t; y/; (6.100)

2k

kD1 kD1

308 6 Nonlinear Aeroelasticity Theory in 2D Aerodynamics

X

N

1

SN .t/ D jj`L jjm2k .a.:/ l.://k .t/

2k

kD1

u

Lemma 6.8. The series

1

X 1 2k

m ..:/ l.://k .t/ (6.101)

2k

kD1

Proof. Let

Z 1

O

./ D e t v.t/dt 0:

0

The transform goes to zero as ! 1, thus we see that for sufficiently large

1 2

0< O

m ./ < 1:

2

Now

Z T Z 1 X

N

1 2k

e t SN .t/dt e t SN .t/dt D O

m ..// k

;

0 0 2k

kD1

X1

1 2k 1 1

D O //k D m2 ./

m .. O < 1:

kD1

2 k 2 1 2 m2 ./

1

O

X

N

1 2k

e t m .a.:/ l.://k .t/

2k

kD1

to see that it converges to a finite limit a.e. in t, 0 < t < T , because the integrals

converge. t

u

6.4 Nonlinear Aeroelastic Dynamics 309

P

Hence 1 kD1 jF2kL .t; y/j converges for every t, rather than a.e, because of the

convolution. With similar results for the moment, we see that we have convergence

of the series (6.16):

1

X

gk .t; Y .:// D N .t; Y .://

kD1

However, in this generality there is no implication that the function is bounded

in 0; 1

. Indeed one has only to take 12 m2 v.t/ D 1 which yields the limit: et .

Series Solution

Next we develop a constructive solution of the initial value problem for the aeroe-

lastic equation (6.89) under some restrictive conditions that imply, in particular, the

torsion angle is small enough.

We can use the state space theory developed in Chap. 5 which would lead to the

theory of nonlinear semigroups (see [45]) but to minimize abstraction we take a

more direct and constructive route, especially because our main interest is in the

steady-state response.

Let S.t/ t 0 denote the semigroup generated by A. Let Y .0/ denote the initial

structure state. Then the method of variation of parameters yields:

Z t

Y .t/ D S.t/Y .0/ C S.t /N .; Y .://d; t 0; (6.102)

0

Z t

Y .t/ S.t /N .; Y .://d D S.t/Y .0/ t 0: (6.103)

0

The solution then can be expressed as a Volterra series [8], which will then be our

constructive solution.

Here, however, we can take advantage of the fact that we have already con-

structed the solution to the linear equation (6.15). Thus

Z t

YP .t/ D AY .t/ C L.t /Y ./d C u.t/; (6.104)

0

where

Z t

u.t/ D N .t; Y .:// L.t /Y ./d

0

310 6 Nonlinear Aeroelasticity Theory in 2D Aerodynamics

1 Z t Z

X 1 Z 2k

Nnl .t; Y .:// D L2k .t 1 ; 1 2 ; :; 2k 2kC1 I

kD1 0 0 0

The initial value problem for the linear equation

Z t

YP .t/ D AY .t/ C L.t /Y ./d

0

Y .t/ D W .t/Y .0/;

where W (.) is given by (5.15) and the solution of (6.102) can be expressed:

Z t

Y .t/ D W .t/Y .0/ C W .t s/Nnl .s; Y .://ds: (6.106)

0

However, to get a constructive solution we need to constrain the initial condition.

We first consider the case where the linear system is (strongly) stable.

Theorem 6.9. The Volterra equation (6.105) has the constructive Volterra series

solution for each T; 0 t < T , given by

1

X

Y .t; y/ D Y0 .t; y/ C Yk .t; y/ (6.107)

kD1

uniformly in y.

where

Z t

Y1 .t/ D W .t s/Nnl .s; Y0 .://ds;

0

Z t

Yk .t/ D W .t s/Nnl .s; Yk1 .://ds k1 (6.108)

0

for

m0 D maxjY .0; y/j; 0<y<` (6.109)

small enough.

6.4 Nonlinear Aeroelastic Dynamics 311

Proof. We use the standard technique for solving Volterra equations [8] which

yields the series in (6.106) and the problem is to prove convergence.

For this we need to relate the properties of Yk .:/ to those of Yk1 .:/. Hence we

start with the properties of Y1 .:/ in terms of the function

Y0 .t; y/ W

1

X

Nnl .t; Y0 .:// D gk .t; Y0 .://: (6.110)

kD1

Z t

v0 .t; y/ D '.t s/.f1 b1 C f2 b2 /Y0 .s; y/ds t 0; 0 < y < `;

0

Z t Z s

v1 .t; y/ D '.t s/b Y0 .s; y/ L.s /b Y0 . ; y/v0 . ; y/d ;

0 0

Z t Z s

vk .t; y/ D '.t s/b Y0 .s; y/ L.s /b Y0 . ; y/vk1 . ; y/d ; k 1:

0 0

(6.111)

1

F2kL .t; y/ D LL .vk .t; y// k 1;

2k

1

F2kM .t; y/ D LM .vk .t; y// k 1:

2k

0

gk .t; y/ D 1 :

M F2k .t; y/

Z t

Y0 .t; y/ D W .t; y; s/Y0 .s/ds;

0

we have that

312 6 Nonlinear Aeroelasticity Theory in 2D Aerodynamics

where

m.Y0 .:/; T / D MaxjY0 .t; y/j; 0 < y < `:

Then

Z t Z s

jv1 .t; y/j m 2

jj'.t s/jjds jjL.s /jjd ;

0 0

Z

jj '. /.f1 b1 C f2 b2 /Y0 .; y/djj; 0t T

0

.t/ D jj'.t/jj;

l.t/ D jjL.t/jj;

Z t

c.t; Y0 .:/; y/ D .jj'.t s/f1 jj jb1 Y0 .s; y/j

0

k c.:; Y0 .:/; y/.t/:

Hence

XN X N

vk .t; y/ vk .t; y/

kD1 kD1

! !

X

N

k

< m .:/ l.:/

2k

c.:; Y0 .:/; y/ .t/ t T; (6.112)

kD1

where

Z t

c.t; Y0 .:/; :/ ..t s/jjf1 jjj b1Y0 .s; y/j C a.t s/jjf2 jjj b2Y0 .s; y/j/ds

0

Z t

f .t s/jY0 .s; y/jds;

0

6.4 Nonlinear Aeroelastic Dynamics 313

Let

X

N

N D m2k .:/ l.:/

k :

kD1

Z 1 X

N

O

et N .t/dt D m2k O ./k ./k :

0 kD1

O D r < 1:

m2 O .0/l.0/

0, whose Laplace transform is

O

m2 O ./l./ r

which is bounded by

O

1 m O ./l./

2 1r

Z 1

r

.t/dt < : (6.114)

0 1r

Hence

X1 Z t

vk .t; y/ f . /.t
/Y0 .
; y/ d
(6.115)

kD1 0

r

m O .0/ 0 < t < T:

1r

Hence it follows that

r

Nnl .t; Y0 .:/; y/ < const. f O .0/m 0<t <T (6.116)

1r

and

p r

jjNnl .t; Y0 .://jj < ` const. f O .0/m: (6.117)

1r

314 6 Nonlinear Aeroelasticity Theory in 2D Aerodynamics

Next Z t

Y1 .t/ D W .t s/Nnl .s; Y0 .://ds 0 < t; (6.118)

0

which is recognized as the response of the linear system to the input

Z ` Z t

Y1 .t; y/ D W .t ; y; s/Nnl .; Y0 .:/; s/dds; (6.119)

0 0

where

jjW .t; y; s/jj < w < 1 for 0 < t < T; and 0 < y; s < `:

Hence

Z ` Z t

jY1 .t; y/j < w jNnl .; Y0 .:/; s/jdds (6.120)

0 0

Z ` Z T

<w jNnl .; Y0 .:/; s/ jdds 0 < t < T:

0 0

Or

r

m.Y1 .0/; T / w ` T const. f O .0/ m.Y0 .:/; T /:

1r

Then it follows that

r k

m.Yk .0/; T / w ` T const. f O .0/ m .Y0 .:/; T /;

1r

Next we constrain r so that

r

w ` T const.f O .0/ D 2 < 1; (6.121)

1r

so that

jY1 .t; y/j < 2 m for t < 1:

And iteratively:

6.5 Stability 315

And hence

1

X

Yk .t; y/ converges for t < T

kD1

as required. t

u

6.5 Stability

Our main interest is the stability of the solution (6.106) and its dependence on the

far field speed. Assuming zero angle of attack, as we do, the far field flow velocity

is q1 D Ei U1 and stability thus depends on U1 , which in this section we simply

denote U , the speed parameter.

This presents an excellent illustration of Hopf bifurcation theory (see [4, 7, 37]),

although our presentation is independent, not invoking that theory. In fact a direct

application is not possible, in the sense that we have a single mathematical theorem

we can quote from which the result follows.

We start with the steady-state solution of (6.97) given in Chap. 4: Y .t/ D 0 valid

for all speeds, noting that N .t; Y .:// D 0 for Y .:/ D 0.

Linearizing the equation about the zero structure state we get the linear equation:

Z t

YP .t/ D AY .t/ C L.t /Y ./d: (6.122)

0

The stability theory for this equation has been treated in detail in Chap. 5 and is

determined in terms of the aeroelastic modes.

The linear system (6.1) is modally stable for all speeds U such that 0 < U < UF

where UF is the flutter velocity. Each modeexcept for the zero frequencydecays

to zero in time.

There are stronger notions of stability. A linear system is strongly stable if

every initial state decays to zero; the elastic energy goes to zero.

Lemma 6.10. For U < UF the linear system (6.120) is strongly stable.

Proof. Actually we prove that the response can be characterized in terms of

semigroup theory.

Semigroup on Riesz Space:

Recall the time domain solution of the linearized equation (6.1) for M > 0 given by

X

W .t/Y D ek t Pk Y C Q.t/Y; (6.123)

316 6 Nonlinear Aeroelasticity Theory in 2D Aerodynamics

depends on the speed U , let R.U / denote the subspace generated by the Riesz

vectors fYk g; we call it the Riesz subspace, which is then also the span of the vectors

fZk g. Then W .t/ maps R.U / into itself. Call the restriction of W .t/ to R by WR .t/.

t

u

SubLemma 6.10. WR .t/ t 0, defines a C0 semigroup over R. It is a contraction

with a compact resolvent. For U < UF it is strongly stable. Furthermore, it has the

representation

X X

WR .t/Y D ek t Pk Y I D WR .t/ Y D ek t Pk Y:

X

AR Y D k Pk Y

X Pk Y

R.; AR /Y D for k :

k

k

Proof of Sublemma. We only show the semigroup property of WR .t/. And this is

immediate.

X X

WR .t/WR .t/Y D ek t Pk ej s Pj Y

X

D ek t Pk ek s Pk Y

X

D ek .sCt / Pk Y

D WR .s C t/Y:

Xh X i

WR .t/Y; WR .t/Y

D ek t Pk Y; ej t Qj Y

X

D e2k t Pk Y; Qk Y

X

Pk Y; Qk Y

D Y; Y

for every t 0.

6.5 Stability 317

X

e2k t Pk Y; Qk Y

; k > N2 is < 2:

X

e2k t Pk Y; Qk Y

; k < N2 is < 2;

u

Let us see what we can say about stability in the nonlinear case for (6.104).

We say that the nonlinear system is strongly stable if every initial condition

bounded in the norm by some nonzero constant decays to zero.

Theorem 6.11. For U < UF , the nonlinear system is strongly stable for a small

enough initial condition, by which we mean that jY .t; y/j ! 0 for all maxy jY0 .y/j

small enough.

Proof. For U < UF the linear system is strongly stable. In particular

jjW .t/Y0 jj ! 0 as t ! 0:

This is a new condition that we exploit in extending Theorem 6.7. Thus choose

Then

maxjY0 .t; y/j; 0 y `; m < 1 for all t > 0:

Next choose m so that

Z 1 Z 1

m 2

jj'.t/jjdt jjL.t/jjdt < r < 1:

0 0

Note that jY0 .t; y/j m for all t, and in fact ! 0 as t ! 1 even if not

exponentially.

Hence we can replace T by infinity in Theorem 6.7, and say more.

Thus in the same notation as there we have

r

jNnl .t; Y0 .:/; y/j < const. f O .0/m for all t; 0<t <1 (6.124)

1r

318 6 Nonlinear Aeroelasticity Theory in 2D Aerodynamics

r

const. f O .0/ D 2 < 1:

1r

And the iteration:

jYk .t; y/j < 2k jY0 .t; y/j:

Hence

1

X

Y .t; y/ D Yk .t; y/;

kD0

where

1

jY .t; y/j jY .t; y/j ! 0 uniformly in y; as t ! 1:

12

t

u

Next let us consider U D UF .

d.M; 1 ; U / D 0I Re 1 D 0:

However, there can be more than one despite popular belief to the contrary.

Remember that when we say the kth mode flutters, the frequency Imk .UF / is quite

different from Imk .0/. So Imk .UF / can be the same for different values of k. But

then for how many values? How many modes can flutter at the same flutter speed?

Indeed it may be nonfinite as, for example, for zero speed, if S is zero. Note that

these are questions that no computer program can answer.

Theorem 6.12. The number of modes that flutter at a given speed U > 0 is finite.

Proof. Suppose now that the number of modes that flutter is infinite. Then denoting

them fi !k g; !k > 0, we must have that

!k ! 1; k ! 1I k D i !k ;

in as much as they cannot have a finite accumulation point. Thus we have that

Z UF

@k

k .UF / D 0 D k .0/ C dU D 0:

0 @U

6.6 Limit Cycle Oscillation 319

Because

Z UF

@n

dU D 0 for every n; and n ! 1; (6.125)

0 @U

P4

@d @wi

@n i D1

@wi @U

D : (6.126)

@U P4 @d @wi

i D1

@wi @

Now as !n ! 1, so does n . And taking limits in (6.124), and using our previous

estimates for M D 0 and M nonzero, we can see that the limit is a constant.

But it is negative at U D 0, as we have shown in Sect. 5.6, and hence we reach

a contradiction showing that the number of fluttering modes for any speed cannot

be infinite. u

t

Remarks: Note the implication of this result: as the mode number increases they

eventually stop fluttering!

Let us first consider the case where exactly one mode has zero real part and all others

have strictly negative real parts. Hence let i ! D i !F ; where ! is the fluttering mode

@

d.M; i !; UF / D 0; D i ! > 0;

@U

At U D UF , let us denote W . : / by WF . : /. Then

1

X

Y .t/ D Y0 .t/ C Yk .t/; (6.127)

kD1

Y0 .t; y/ D e Yi ! .y/;

i !t

Z t

Y1 .t/ D WF .t s/Nnl .s; Y0 . : //ds;

0

Z t

Yk .t; y/ D WF .t s/Nnl .s; Yk1 .://ds;

0

1

X

Nnl .t; Y . : // D gk .t; y/;

kD1

0

gk .t; y/ D ;

M1 Fk .t; y/

320 6 Nonlinear Aeroelasticity Theory in 2D Aerodynamics

1 LL .vk .t; y//

Fk .t; y/ D k ;

2 LM .vk .t; y//

where

Z t

v0 .t; y/ D '.t s/.f1 b1 C f2 b2 /Y0 .s; y/ds t 0; 0 < y < `;

0

Z t Z s

v1 .t; y/ D '.t s/b Y0 .s; y/ L.s
/b Y0 .
; y/v0 .
; y/d
:

0 0

O !/'.i

O !/L.i

D e3i !t '.i O !/.f1 . : /b1 C f2 . : /b2 /Yi ! .y/.b Yi ! .y//2 :

O !/L.i O !/

where

k

jj '.i O !/ .b Yi ! .y//2k jj

O !/L.i

k

jj'.i

O !/jj jjL.i O

O !/jj jb Yi ! .y/j2 D .O .0/l.0// k

:

Let

and we assume

O

r D m2 O .0/l.0/ is < 1: (6.129)

Hence the series

1

X

v.t; y/ D vk .t; y/

kD1

O !/L.i O !//1

O !/L.i

'.i

6.6 Limit Cycle Oscillation 321

Approximation

For m small enough we may approximate the sum just by the first term, just the third

harmonic. Denote the approximation by va .:/:

O !/L.i O !/.f1 .:/b1

where

0

g1 .y/ D

M1 Fl .y/

1 `L .va .0; y//

F1 .y/ D

2 `M .va .0; y//

O !/L.i O !/.f1 . : /b1 C f2 . : /b2 /Yi ! .y/.b Yi ! .y//2 : (6.132)

Note that

r

jg1 .y/j const: O !/.f1 . : /b1 C f2 . : /b2 /Yi ! .y/j

j'.i

1r

r

const: O .0/f jYi ! .y/j; (6.133)

1r

Z t

Y1 .t/ D WF .t s/Nnl .s; Y0 .://ds

0

Z 1

Y1 .t/ D e 3i !t

WF .s/ei 3!s ds g1 .:/

0

322 6 Nonlinear Aeroelasticity Theory in 2D Aerodynamics

r

Y1 .0; y/ const. O .0/f Y0 .0; y/: (6.135)

1r

And similarly by iteration we have

where

k X

1

r

Yk .0; y/ const. O .0/f Yk .0; y/: (6.137)

1r

kD1

Hence

1

X 1

X

Y .t; y/ D e3i !t Yk .0; y/ C ei !t Yk .0; y/; (6.138)

kD1 kD1

r

const. O .0/f D < 1;

1r

and

X1

Yk .0; y/ < Yi ! .y/: (6.139)

1

kD1

We have thus a first approximation to the response at the flutter given by (6.13)

which is thus the LCO.

It is periodic with the same period as that of the flutter mode frequency predicted

by the linear model, with a predominant third harmonic. It is thus an illustration of

the Hopf bifurcation theory.

Higher-order approximation would involve the .3 C 2n/ harmonics. If we have

more than one mode fluttering, we would then have to consider .3 C 2n/ harmonics

of each mode and intermodulation, a messy calculation at best!

But the main point is again that we have a response which is periodic in the steady

state with harmonics of the fundamental frequency given by the linear model.

s

Z T

Lim T ! 1 .1=T / jjY .t; y/jj2 dt;

0

6.7 The Air Flow Decomposition Theory 323

AWGLELS ALPHA

60 3

40 2

20 1

0 0

20 1

40 2

60 3

11:01:05 11:01:10 11:01:15 11:01:20 11:01:25 11:01:30

which within the approximation (6.12) is given by the Parseval formula for the

Fourier series:

v !2

u 2

u X1

t

LCO RMS amplitude D Yi ! .y/ C Yk .0; y/ (6.140)

kD1

s

2

Yi ! .y/ 1 C : (6.141)

1

In particular we see that for initial amplitudes small enough, the LCO amplitude is

proportional to the initial amplitude. Further investigation is required to determine

whether this continues to be true for larger initial amplitudes.

Figure 6.1 shows the incidence of flutter LCO encountered in a specially

designed Flight Test; see [108] for more details.

Finally we come to the solution to the potential field equation, the air flow.

As we have noted, our primary concern is the stability of the structure. The air

flow per se is of secondary importance. However, as we have seen, we are able to

characterize the flow as well. We consider only 2D flow with zero angle of attack.

From Sect. 6.2, we have the decomposition of the kth-order potential

k D L;k C 0k

324 6 Nonlinear Aeroelasticity Theory in 2D Aerodynamics

and

1

X

D k =k

kD0

1

X

L D L;k =k

kD1

1

X

0 D 0;k =k

kD0

The solution of

.0;k / D gk1

following ([45, p. 692]) is given by

Z tZ 1 Z 1

0;k .t; x; z/ D F .t ; x ; z /gk1 .; ; /ddd;

0 1 1

p p 2

F .t; x; z/ D 1= 2 1 M 2 1= t Ux=c12

1= 1 M 2 x 2 =c12 C z2 =c22 ;

c12 D a1

2

1 M2 ;

c22 D a1

2

:

And defining

1

X

gD gk =k

kD1

Z tZ 1 Z 1

0 .t; x; z/ D F .t ; x ; z /g.; ; /ddd;

0 1 1

6.7 The Air Flow Decomposition Theory 325

the typical section case and zero angle of attack, we need to take the inverse

Laplace/Fourier transform of

p

1=.k C i !/ez k 2 M 2 C2kM 2 i !C! 2 .1M 2 /

AQOk .; i !/; z > 0:

is Z t

k .t; x/ D Ak .t ; x U/d; 1 < x < 1:

0

The inverse Fourier transform of

p

ez k M C2kM i !C! .1M /

2 2 2 2 2

is

p p

L.t; x; z/ D 1= 2 .1 M 2 / 1= ..t Ux=c12 /2 r 2 / ;

where

r 2 D .1 M 2 /.x 2 =c12 C z2 =c22 /:

Hence we have

Z tZ bCUt

L;k .t; x; z/ D L.t ; x ; z/k .; /dd: (6.144)

0 b

Z tZ bCUt

L .t; x; z/ D L.t ; x ; z/.; /dd; (6.145)

0 b

where

Z t

.t; x/ D A.t ; x U/d

0

.

p p

L.t; x; z/ D 1=.2 .1 M 2 //1=. ..t Ux c12 /2 r 2 //; (6.146)

Thus finally we have for the time domain potential flow decomposition

326 6 Nonlinear Aeroelasticity Theory in 2D Aerodynamics

where L .t; x; z/ provides the lift, has no discontinuities in the velocity, and can be

linearized, and the linearized equation determines the LCO speed and period. We

are not interested in 0 .t; x; z/ because it does not affect the structure stability. It

cannot be linearized as observed in [86], and may have discontinuities in the flow

velocity. But whether there are shocks (see, e.g., [14] for definition of shocks in

the flow) is not settled by this theory. Shocks imply a change in entropy [14] and

our key assumption is that the flow is isentropic. Also, even if there are shocks, it is

not clear [86] that they affect the structural stability, not withstanding the generally

held beliefs by aeroelasticians.

This chapter illustrates the need for analytical theory. For example, a precise

definition of the crucial concept of flutter speed is not possible by numerical

computation, which is probably why there is no formal definition in all of the

aeroelastic literature.

The continuum formulation and solution of the aeroelastic equations in isentropic

flow appears here for the first time. The structure dynamics is assumed to be linear

and although the extension to the nonlinear case would be of interest, the extra

complications it would bring in even in the domain specifications would be too

much without proportional gain.

And of course our main interest is in the nonlinear aerodynamics anyhow. We

have also limited the treatment to the case of zero angle of attack because the

extension to the nonzero case would involve too much effort without justifying the

returns. Our main objective here is in showing the Hopf bifurcation and the LCO.

We have provided the means, however complex, for calculating the LCO amplitude.

This is of course also a tedious process in CFD requiring time marching.

The existence of shocks in isentropic flow continues to be controversial. It is

indeed established in 1D flow (the Riemann problem, see, e.g., [4]). However, there

is no mathematical proof (yet) in 2D or higher dimensions [private communication:

A. Chorin, 2010]. And furthermore, ours can be termed a singular case because of

the finite boundary.

Chapter 7

Viscous Air flow Theory

7.1 Introduction

In this chapter we extend the theory allowing for nonzero viscosity. The field

equations then are the NavierStokes equations, a subject in fluid dynamics of

intense research activity, but still replete with many open problems. Our interest

here is again on the effect of viscosity on the wing-structure response rather than

the flow itself. The extant aeroelastic literature on this aspect is all computational

[82, 83, 93].

We begin as in Chap. 3, with the governing conservation laws for fluid flow in

their differential form, but now including viscosity. We follow the notation there

for pressure, density, and other thermodynamic variables, with q.t; x; y; z/ in

particular denoting the fluid velocity.

1. Conservation of mass

@

C r .q/ D 0: (7.1)

@t

2. Conservation of momentum (NavierStokes equation)

@q 2

C .q: r/q D q rp C C r.r q/; (7.2)

@t 3

where

is the coefficient of viscosity. .1:78106 (for air)).

is the second coefficient of viscosity; more on this later.

q is the 3 1 vector with components.

r rqi i D 1; 2; 3, and qi are the components of q.

DOI 10.1007/978-1-4614-3609-6 7, Springer Science+Business Media, LLC 2012

328 7 Viscous Air flow Theory

q:rqi i D 1; 2; 3;

E 3;

q D Ei q1 C jEq2 C kq

And finally the law that is of crucial importance in our theory:

3. Energy-Flux Equation

As noted by Meyer [14], this is no more than the first law of thermodynamics in

a Eulerian version.

The energy per unit volume

p

E.t/ D cv T D

. 1/

X 3 X 3

DE

D pr q C i k ei k C T; (7.3)

Dt i D1 kD1

2

i k D i k r:q C 2ei k ;

3

1 @qk @qi

ei k D C

2 @xi @xk

3 X

X 3

D i k ei k

i D1 kD1

is called the dissipation function and we have the thermodynamic relation [14]

DS

T D C T; (7.4)

dt

where S is the entropy. (assumed constant) is the coefficient of thermal

diffusivity 1; 000 ; D cp =cv 1:4. The flow is incompressible if the

flow divergence

r:q D 0:

These are the field equations. But what distinguishes the aeroelastic problem are

the boundary conditions. We consider again the finite plane structure model as in

Chap. 3. The big difference from the nonviscous flow is:

7.2 The Field Equation/Conservation Laws 329

Dz.t; x; y/

q.t; x; y; 0C/ D kE ; jxj < b; 0 < y < `; (7.5)

Dt

where z.. , . , . / is the structure displacement along the z-axis, as in Chap. 3. In

other words, there is no slip between the air and the wing motion, the latter being

normal to the wing plane in our beam model. This is of course more restrictive

than just flow tangency, as in the nonviscous model.

2. The KuttaJoukowsky conditions remain the same as described in Chap. 3.

However, whether the Kutta condition is needed is examined later. The far field is

Ei U where we omit the subscript infinity. Our main interest again is the structure

stability as a function of U > 0.

The basic relevant references for viscous flow for our needs are: Meyer [14];

LandauLifschitz [12], SchlictingKersten [15], 2003 edition of the 1965 classic;

ChorinMarsden [4], R. Temam [19]; and OleinikSamokhin [13].

The main difficulty here starts with the basic question of existence and unique-

ness whatever the function space and sense (weak) of convergence chosen for the

flow solution.

Of the huge readily available literature on NavierStokes equations, it is fortunate

that we are only concerned with aspects of the theory that affect the boundary-

structure stability.

It should be noted that if we set D 0P0 D 0 in the NavierStokes equation

(7.2), it does formally reduce to the Euler equation (3.2) we start with. But the

further condition of isentropy is required to reach the full-potential equation, and

the fluid-structure boundary conditions imposed are of course quite different.

Incompressible Flow

in the sense that the flow divergence can be neglected

r: q D 0:

coefficient of viscosity. As a result, the NavierStokes equation (7.2) simplifies to:

@q

C .q:r/q q C rp D 0 (7.6)

@t

and the energy-flux equation to:

X 3 X3

DE

D 2 ei k 2 C T; (7.7)

Dt i D1 kD1

330 7 Viscous Air flow Theory

1 @qk @qi

ei k D C (7.8)

2 @xi @xk

Before we can consider the stability of the structure we need to consider the

following.

3 X

X 3

2 ei k 2 C T D 0: (7.10)

i D1 kD1

But this condition by (7.4) is equivalent to saying that the entropy is constant in time.

Hence in the time-invariant case this equation is automatically satisfied because

entropy is taken to be time invariant. Also the continuity equation (7.1) using the

incompressibility condition yields

q:r D 0;

so that we may take to be constant. Thus (7.9) holds with a constant. The surprise

here is that the static solution for the nonviscous case, where the structure is at rest

and the flow vector is a constant, does not hold here because of the no-slip boundary

condition that the structure and the air move together so that the air velocity has

to be zero over the structure. Hence the flow cannot be a constant if the far field

velocity is nonzero.

specialize to the typical section/zero angle of attack, as in the previous chapters.

Thus we have the field equation

.q:r/q q D 0;

7.2 The Field Equation/Conservation Laws 331

where is a constant.

The 2D NS in the xz-plane can now be written out as

2

@q1 @q1 1 @p @ q1 @2 q1

q1 C q3 C Dv C q1 .x; 0/ D 0; (7.11)

@x @z @x @x 2 @z2

2

@q3 @q3 1 @p @ q3 @2 q3

q1 C C q3 C Dv C ; (7.12)

@x @z @z @x 2 @z2

@q1 @x

r:q D C q3 D 0;

@x @z

v D :

Boundary Conditions

Far Field

q3 .x; 1/ D 0 D q3 .1; z/;

q1 .x; 1/ D U D q1 .1; z/:

pe .s=cv / D constant :

So the first task is to determine the static solution. At the present time the only way

is to invoke the following celebrated theory.

Proposed by Prandtl in 1904 [15] for fluids of small viscosity (vanishingly small

and air qualifies with . 10/6 ); but more important, of large Reynolds number;

see [4]. The concept of the boundary layer is to subdivide the flow region into

two parts: a thin-boundary layer at the wing boundary where the viscosity must

332 7 Viscous Air flow Theory

be taken into account where the no-slip boundary condition holds, and the outer

layer which is the bulk of the region where we may neglect viscosity and assume

isentropic flow. For the matching problems here see [3, 4].

A word of caution! It should be noted right away that Oleinik writing in 1999

includes this problem in [13] in the section on Some Open Problems: Is it

possible to give a strict mathematical justification of this procedure and find the

limits of applicability of the Prandtl equations?

In other words, there is no mathematical justification for this procedure, yet.

Here we follow LandauLifschitz [12]. The Kutta condition is not invoked. In the

usual formulation, as in [12], the boundary is allowed to be infinite. But here we do

not have that luxury; the boundary is finite. There is another additional complication.

In reference to (7.2) we need to make additional simplifications in the inner layer:

@p @p

D0D

@z @x

p

vD :

Next we assume:

@2 q3 @2 q3

D 0 P

0 D 0:

@z2 @x 2

Hence the NavierStokes equation (7.2) becomes

@q1 @q1 @2 q1

q1 C q3 Dv 2 ; (7.13)

@x @z @z

@q3 @q3

q1 C q3 D 0;

@x @z

which using

@q1 @q3

C D0

@x @z

becomes

@q3 @q1

q1 D q3 : (7.14)

@x @x

Here we can state the next theorem.

7.2 The Field Equation/Conservation Laws 333

@q1 @q1 @2 q1

q1 C q3 Dv 2 ; (7.15)

@x @z @z

@q1 @q3

C D0 (7.16)

@x @z

q1 .x; 1/ D u; b < x < b

@q1 @2 q1

and

@z @z2

@q1 @q3

;

@x @z

We do not have a constructive solution technique. We do have the following

however.

Theorem 7.2 (Blasius [cf. 12, 14, 15]). If the chord is infinite, that is, we have in

our notation b < x < 1, the solution is unique and the limit value as z goes to

infinity r

vU

q3 .x; 1/ D 0:43 b < x < 1: (7.17)

xCb

Proof. See [14]. t

u

We use this as the boundary condition to calculate the steady-state potential flow in

the outer layer.

Thus we have the steady-state problem as in Chap. 4, (4.1), now specialized

to typical section and zero angle of attack. The field equation for the potential

.x; z/ is:

" 2 2 !#

2 . 1/ 2 @ @

0D a11C 2

U . /

2a1 @x @z

1 @ @ 2 @ @ 2

jr j C jr j ; (7.18)

2 @x @x @z @z

334 7 Viscous Air flow Theory

@2 @2

D C ;

@x 2 @z2

2 2

2 @ @

jr j D C

@x @z

@ .x; 0/

D w.x/; jxj < b;

@z

where w./ is given by (7.9), and the far field condition that

q1 D i U

This problem has already been treated in Chap. 4. The solution for the potential

as calculated there is given by

1

X 1

.x; z/ D xU C k .x; z/; (7.19)

k