P1: KAE 0521880428pree CUFX159-Virgin
0 521 85648 5
August 14, 2007
This page intentionally left blank
19:48
P1: ...

Author:
Lawrence N. Virgin

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!

P1: KAE 0521880428pree CUFX159-Virgin

0 521 85648 5

August 14, 2007

This page intentionally left blank

19:48

P1: KAE 0521880428pree CUFX159-Virgin

0 521 85648 5

VIBRATION OF AXIALLY LOADED STRUCTURES This book concerns the vibration and the stability of slender structural components. The loss of stability of structures is an important aspect of structural mechanics and is presented here in terms of dynamic behavior. A variety of structural components are analyzed with a view to predict their response to various (primarily axial) loading conditions. A number of different techniques are presented, with experimental verification from the laboratory. Practical applications are widespread, ranging from cables to space structures. The book presents methods by which the combined effects of vibration and buckling on various structures can be assessed. Vibrations and buckling are usually treated separately, but in this book their influence on each other is examined together, with examples when a combined approach is necessary. The avoidance of instability is the primary goal of this material. Dr. Lawrence N. Virgin completed his doctorate in structural mechanics in 1986 at University College London. Since 1988, he has been at Duke University, where he teaches and conducts research in engineering mechanics. His interests are centered on the instability behavior of nonlinear dynamics systems in the context of experimental vibrations, with applications including aeroelasticity, systems with discontinuities (impact and friction), fluid–structure interaction, and buckling. He is currently Gardner Professor and Chair of the Department of Civil and Environmental Engineering and holds a secondary appointment in the Department of Mechanical Engineering and Materials Science.

August 14, 2007

19:48

P1: KAE 0521880428pree CUFX159-Virgin

0 521 85648 5

August 14, 2007

19:48

P1: KAE 0521880428pree CUFX159-Virgin

0 521 85648 5

Vibration of Axially Loaded Structures LAWRENCE N. VIRGIN Duke University

August 14, 2007

19:48

CAMBRIDGE UNIVERSITY PRESS

Cambridge, New York, Melbourne, Madrid, Cape Town, Singapore, São Paulo Cambridge University Press The Edinburgh Building, Cambridge CB2 8RU, UK Published in the United States of America by Cambridge University Press, New York www.cambridge.org Information on this title: www.cambridge.org/9780521880428 © Lawrence N. Virgin 2007 This publication is in copyright. Subject to statutory exception and to the provision of relevant collective licensing agreements, no reproduction of any part may take place without the written permission of Cambridge University Press. First published in print format 2007

ISBN-13

978-0-511-46328-0

eBook (EBL)

ISBN-13

978-0-521-88042-8

hardback

Cambridge University Press has no responsibility for the persistence or accuracy of urls for external or third-party internet websites referred to in this publication, and does not guarantee that any content on such websites is, or will remain, accurate or appropriate.

P1: KAE 0521880428pree CUFX159-Virgin

0 521 85648 5

This book is dedicated to my wife Lianne, my children Elliot and Hayley, and my parents Margaret and Alan

August 14, 2007

19:48

P1: KAE 0521880428pree CUFX159-Virgin

0 521 85648 5

August 14, 2007

19:48

P1: KAE 0521880428pree CUFX159-Virgin

0 521 85648 5

August 14, 2007

Contents

Foreword Preface

page xiii xv

1 Context: The Point of Departure . . . . . . . . . . . . . . . . . . . . . . . . . 1 2 Elements of Classical Mechanics . . . . . . . . . . . . . . . . . . . . . . . . . 8 2.1 2.2 2.3 2.4 2.5

Introduction Newton’s Second Law Energy and Work Virtual Work and D’Alembert’s Principle Hamilton’s Principle and Lagrange’s Equations 2.5.1 Constraints 2.5.2 Conservation Laws 2.6 Nonconservative Forces and Energy Dissipation 2.6.1 Damping 2.6.2 Time-Dependent Forces 2.7 Strain Energy

8 8 10 11 13 15 15 17 18 19 20

references

21

3 Dynamics in the Vicinity of Equilibrium . . . . . . . . . . . . . . . . . . . . 22 3.1 3.2 3.3

The Linear Oscillator Oscillator with a Slow Sweep of Frequency Dynamics and Stability 3.3.1 Stability Concepts 3.4 Bifurcations 3.4.1 The Saddle-Node Bifurcation 3.4.2 Bifurcations from a Trivial Equilibrium 3.4.3 Initial Imperfections 3.4.4 Bifurcations of Maps 3.5 A Simple Demonstration Model 3.6 Experiments

22 28 30 30 32 33 34 35 37 37 41

references

43

vii

19:48

P1: KAE 0521880428pree CUFX159-Virgin

0 521 85648 5

viii

August 14, 2007

Contents

4 Higher-Order Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45 4.1 4.2

Introduction Multiple-Degree-of-Freedom Systems 4.2.1 The Algebraic Eigenvalue Problem 4.2.2 Normal Modes 4.2.3 Equilibrium, Linearization, and Stability 4.2.4 Routh–Hurwitz Criterion 4.2.5 Lyapunov Functions 4.2.6 Rayleigh’s Quotient 4.3 Distributed Systems 4.3.1 The Differential Eigenvalue Problem 4.3.2 Solution Methods 4.3.3 Context Revisited

45 45 46 47 48 54 55 56 57 59 60 64

references

64

5 Discrete-Link Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66 5.1 5.2

5.3 5.4 5.5 5.6 5.7 5.8 5.9 5.10

Introduction An Inverted Pendulum 5.2.1 Static Behavior 5.2.2 Geometric Imperfections 5.2.3 Dynamic Behavior 5.2.4 A Note on Inertia A Discrete-Strut Model An Asymmetric Model A Three-Bar Model A Snap-Through Model Augusti’s Model Multiple Loads Load-Dependent Supports Path Following and Continuation

references

66 66 67 69 70 74 75 80 82 84 88 91 93 94 95

6 Strings, Cables, and Membranes . . . . . . . . . . . . . . . . . . . . . . . . 97 6.1 6.2

Introduction The Stretched String 6.2.1 The Wave Equation 6.2.2 Traveling-Wave Solution 6.2.3 Energy Considerations and Rayleigh’s Principle 6.3 A Suspended Cable 6.3.1 The Hanging Chain 6.4 A Rectangular Membrane

97 97 97 100 101 102 106 108

references

109

19:48

P1: KAE 0521880428pree CUFX159-Virgin

0 521 85648 5

August 14, 2007

Contents

ix

7 Continuous Struts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111 7.1 7.2

7.3 7.4 7.5 7.6 7.7 7.8 7.9

7.10 7.11

Introduction Basic Formulation 7.2.1 The Response 7.2.2 The Temporal Solution 7.2.3 The Spatial Solution Rayleigh’s Quotient Rayleigh–Ritz Analysis A Galerkin Approach Higher Modes Rotating Beams A Strut with a Tangential Load Self-Weight 7.9.1 A Hanging Beam 7.9.2 Experiments Thermal Loading Other Effects

references

111 111 113 114 116 120 120 124 126 131 134 136 138 139 142 143 143

8 Other Column-Type Structures . . . . . . . . . . . . . . . . . . . . . . . . . 147 8.1 8.2 8.3 8.4 8.5 8.6 8.7

A Beam on an Elastic Foundation Elastically Restrained Supports Beams with Variable Cross Section Modal Coupling Flexural–Torsional Buckling and Vibration Type of Loading A Continuous Arch

references

147 149 150 154 157 161 162 164

9 Frames . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166 9.1 9.2 9.3 9.4 9.5 9.6

A Beam with General Boundary Conditions The Stiffness Method A Self-Strained Frame Example Modal Analysis Large-Deflection Analysis A Tubular Structure

references

10

166 168 172 174 177 178 181

Plates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183 10.1 Introduction 10.1.1 Brief Review of the Classical Theory 10.1.2 Strain Energy 10.1.3 Boundary and Initial Conditions

183 183 187 188

19:48

P1: KAE 0521880428pree CUFX159-Virgin

0 521 85648 5

x

August 14, 2007

Contents

10.2 10.3 10.4 10.5 10.6

10.7

10.1.4 The Simplest Case 10.1.5 Initial Imperfections The Ritz and Finite-Element Approaches A Fully Clamped Plate Moderately Large Deflections Postbuckling Mode Jumping 10.6.1 Introduction 10.6.2 The Analytic Approach 10.6.3 Finite-Element Transient Results Cylindrical Shells

references

11

12

13

190 193 193 196 198 199 205 205 205 209 209 212

Nondestructive Testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216 11.1 Introduction 11.1.1 The Southwell Plot 11.1.2 Examples 11.2 Some Background 11.3 Snap-Through Revisited 11.4 Range of Prediction 11.5 A Box Column 11.6 Plates and Shells

216 217 219 222 225 228 230 231

references

234

Highly Deformed Structures . . . . . . . . . . . . . . . . . . . . . . . . . . 237 12.1 Introduction to the Elastica 12.2 The Governing Equations 12.3 Case Study A: Self-Weight Loading Revisited 12.3.1 Numerical Results 12.3.2 Experiments 12.4 Case Study B: A Heavy Beam 12.4.1 Numerical Results 12.4.2 Experiments 12.5 Case Study C: A Pinched Loop 12.6 Case Study D: A Beam Loaded by a Cable 12.7 The Softening Loop Revisited

237 239 240 241 242 243 244 245 248 251 256

references

259

Suddenly Applied Loads . . . . . . . . . . . . . . . . . . . . . . . . . . . . 261 13.1 13.2 13.3 13.4 13.5

Load Classification Back to Link Models Dynamic Buckling of a Plate A Type of Escaping Motion Impulsive Loading

261 262 267 268 272

19:48

P1: KAE 0521880428pree CUFX159-Virgin

0 521 85648 5

August 14, 2007

Contents

14

13.5.1 Equilibrium Behavior 13.5.2 Behavior under Sudden Loading 13.6 Snap-Through of a Curved Panel

273 274 275

references

279

Harmonic Loading: Parametric Excitation . . . . . . . . . . . . . . . . . . 282 14.1 14.2 14.3 14.4

An Oscillating End Load The Variational Equation Mathieu’s Equation Pulsating Axial Loads on Shells 14.4.1 A Curved Panel 14.4.2 A Cylindrical Shell

references

15

xi

282 283 286 288 289 289 292

Harmonic Loading: Transverse Excitation . . . . . . . . . . . . . . . . . . 294 15.1 Introduction: Resonance Effects 15.1.1 A Single-Mode Approximation 15.1.2 Beyond Buckling 15.2 The Poincare´ Section 15.3 Continuous Systems 15.4 An Application to Vibration Isolation 15.4.1 Postbuckling of a Strut Revisited 15.4.2 Experimental Verification 15.4.3 The Forced Response 15.5 Forced Excitation of the Thermally Buckled Plate

294 295 296 297 299 304 305 306 307 308

references

310

16 Nonlinear Vibration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 312 PART I: FREE VIBRATION 16.1 Introduction 16.2 Abstract Models 16.3 A Mass Between Stretched Springs 16.4 Nonlinear Vibration of Strings 16.5 Nonlinear Vibration of Beams 16.6 Nonlinear Vibration of a Plate 16.7 Nonlinear Vibration in Cylindrical Shells PART II: FORCED VIBRATION 16.8 Nonlinear Forced Vibration of Strings 16.9 Nonlinear Forced Vibration of Beams 16.10 Persistent Snap-Through Behavior in a Plate 16.11 A Panel in Supersonic Flow 16.12 Chaotic Behavior

312 312 313 315 319 320 322 324 325 325 326 330 334 337

references

344

Index

347

19:48

P1: KAE 0521880428pree CUFX159-Virgin

0 521 85648 5

August 14, 2007

19:48

P1: KAE 0521880428pree CUFX159-Virgin

0 521 85648 5

August 14, 2007

Foreword

The concept of stability is intrinsically a dynamical one. This is recognized even by the simplistic classical definition, which ignores the random disturbances of the real world and just inquires what would happen if a system were displaced to an adjacent position in phase space. So we are lucky, indeed, to have this well-conceived book written by a leading researcher who has mastered both nonlinear dynamics and the static bifurcations of elastic stability theory. The latter theory works well for conservative systems, for which powerful energy theorems are available, but needs augmenting by dynamical methods in the presence of loading that is either nonconservative or time dependent. Lawrence Virgin has of course just the right background, having chosen (in his usual thoughtful way) to work first at University College London, then with Earl Dowell at Duke University. He is currently the Chair of the Department of Civil and Environmental Engineering at Duke (which has an active interdisciplinary program in nonlinear dynamics) and has enjoyed productive collaborations with Raymond Plaut (Virginia Polytechnic Institute and State University). His previous book, Introduction to Experimental Nonlinear Dynamics (also published by Cambridge University Press), brought a welcome sense of realism into the often esoteric field of nonlinear dynamics by focusing on experimental investigations, and I am delighted to see a similar emphasis in this new book titled Vibration of Axially Loaded Structures. Understanding the buckling and vibration of structures under axial compression is of very great importance to structural and aerospace engineers, to whom this book is primarily addressed. They, together with readers from many other areas of mechanics, will be well served by Lawrence’s latest offering. The book covers a wide field, including buckling, dynamics (both linear and nonlinear), theory, and experiments, all explained in a clear and lucid style. Especially valuable are the comprehensive lists of references, which nicely complement the text. I can heartily recommend this book to all who want to see a wide-ranging and scholarly treatment that brings new insights to an important long-standing but still emerging field. Michael Thompson, FRS Cambridge, England xiii

19:48

P1: KAE 0521880428pree CUFX159-Virgin

0 521 85648 5

August 14, 2007

19:48

P1: KAE 0521880428pree CUFX159-Virgin

0 521 85648 5

August 14, 2007

Preface

General Comments Rationale and Scope r The material covered by this book spans the areas of vibration and buckling. Both of these areas can be considered as subsets of structural mechanics and play a central role in the disciplines of civil, mechanical, and aerospace engineering. r Although vibration and buckling are key elements in the teaching of advanced engineering, they are typically taught separately. However, the interplay of dynamics and stability in structural mechanics and its coverage in a single text provide an opportunity to present material in an interesting way. r The quest for stronger, stiffer, and more lightweight structural systems is making the material covered in this book increasingly important in practical applications. r By using axially loaded structures as a consistent theme, the book covers a wide variety of types of structure, methods of analysis, and potential applications without trying to cover too much. Experimental verification appears throughout. r The level of material is appropriate for upper-level, advanced undergraduate classes, and graduate students, but researchers and practicing engineers will find plenty of interest too. r The text is liberally illustrated by figures, and close to 500 technical references are given.

Acknowledgments The material presented in this book contains a synthesis of material from the general literature together with results from my own research program. In terms of the latter, this is by no means a solo endeavor, and there are a number of people I would like to thank. First, and foremost, much of the work I have conducted in this area in the past 20 years or so has been done with Raymond Plaut from Virginia Polytechnic Institute and State University. I have learned a considerable amount from his deep xv

19:48

P1: KAE 0521880428pree CUFX159-Virgin

xvi

0 521 85648 5

August 14, 2007

Preface

understanding of theoretical and applied mechanics as well as his attention to detail and meticulous approach to research. He also proofread this book, making useful suggestions and providing invaluable guidance. My path along this road goes back to Terry Roberts in Cardiff, Michael Thompson in London, and Earl Dowell here in North Carolina. I have benefited immeasurably from their influence as mentors during my formative years (and beyond). In addition to my family, of course, I’d like to thank my friends and colleagues at Duke who have contributed to a supportive environment: Tod Laursen, John Dolbow, Henri Gavin, Ken Hall, Josiah Knight, and Bob Kielb. I have had the privilege of working with many talented graduate students over a period of almost 20 years, and those whose research contributed directly or indirectly to material in this book include Phil Bayly, Kevin Murphy, Mike Todd, Kara Slade, Hui Chen, David Holland, Mike Hunter, Ilinca Stanciulescu, Sophia Santillan, and Ben Davis (who also diligently proofread the manuscript). Thanks to them all. Lawrie Virgin Durham, North Carolina

19:48

P1: RTT Chapter-01

CUFX159-Virgin

0 521 85648 5

August 14, 2007

1 Context: The Point of Departure

In the engineered world (and in a good deal of the natural world), stable equilibrium, or some kind of stationary or steady-state behavior, is the order of the day. Systems are designed to operate in a predictable fashion to fulfill their intended functions despite disturbances and changing conditions. Control systems have been spectacularly successful in maintaining a desirable (stable1 ) response given inevitable uncertainty in modeling system physics. However, there are plenty of examples of systems becoming unstable – and often the consequences of instability are severe. This book looks at the interplay between vibrations and stability in elastic structures. A brief view of an ecological system provides an effective analogy. The competition between certain species can be viewed as a coupled dynamic system in a slowly changing environment. External influences are provided by various factors including the climate, disease, and human influence. The delicate interaction is played out as conditions evolve and populations respond accordingly – usually in a correspondingly slow way also. However, an instability may occur leading to extinction on a relatively short time scale, perhaps when a disease (or massive meteorite) wipes out an entire population. This situation is not that dissimilar to the fluctuations of the stock markets (in which prediction of sudden changes is of concern to individuals and governments). In an engineering context, we typically have considerable knowledge about the underlying physics and governing equations of our systems, are able to test a system both analytically and in the laboratory, and thus have a much better chance of assessing the robustness of a system, especially its propensity to failure. However, unforeseen circumstances do occur, and it would, of course, be remiss in a book concerning stability in engineering mechanics not to mention the Tacoma Narrows suspension bridge disaster. But many other bridges and buildings have collapsed, aircraft wings and rotorblades have a tendency to flutter, ships sometimes capsize, the tracks of a railroad will buckle from time to time, electric circuits sporadically exhibit unintended feedback, machine parts are prone to fatigue, and once in a while satellites disappear into deep space. What most of these systems have in common is that they were either subject to external influences with which they could not cope 1

Some aeronautics control systems take advantage of a brief loss of stability for enhanced maneuverability.

1

15:5

P1: RTT Chapter-01

CUFX159-Virgin

0 521 85648 5

2

August 14, 2007

Context: The Point of Departure

Gravity

Figure 1.1. A deteriorating scenario.

or they changed. Perhaps an encounter with a rogue wave in the case of a ship, or collision with space debris in the case of the satellite. In this last instance an error in the units used in trajectory calculations may cause disaster but in the sense that the system was designed correctly but for the wrong conditions. Of course, there are always practical limits to how much safety or redundancy can be built into a system; the World Trade Center provided a sobering example. But it is also likely that a system is subject to slowly changing conditions, which may, of course, lead to catastrophe, but in a gradual deteriorating sense. It is with these systems that we have scope for monitoring and prediction, as their (dynamic) response may give clues about future performance. Hence, given a (structural) system in some state of rest (equilibrium) or steadystate motion (an oscillation), we seek to understand those conditions that cause a change in the nominal response, and especially where such a change is large (and instability falls squarely into this category). The theoretical framework underlying this statement is of course based on Newton’s laws and subsequent developments especially concerning concepts of energy. To crystallize this approach, consider the schematic diagram shown in Fig. 1.1. Here we might consider the behavior of a small ball allowed to roll (under the influence of gravity) on a curved surface to represent a generic structural or mechanical system. The analogy is really brought into focus if we further assume that the curve is actually associated with the underlying potential energy of the system and that the surface causes a little energy dissipation as the ball rolls. Hence the bottom of the energy “well” is identified as a position of stable equilibrium, with linear theory based on a locally quadratic minimum. Linear stability theory will also tell us that the “hilltops” are points of unstable equilibrium. In both cases, the ball will remain at rest at these extremum values of the potential energy surface. However, the important behavior is observed if the system is subject to a disturbance. In the stable case, the ball might begin to oscillate but typically return to rest at the bottom of the well. In the unstable case, the ball picks up speed and

15:5

P1: RTT Chapter-01

CUFX159-Virgin

0 521 85648 5

August 14, 2007

Context: The Point of Departure

3

F(t) P(t)

Figure 1.2. A slender axially loaded structure and its dynamic response.

departs the local neighborhood of the hilltop. These situations are well covered by linear stability theory providing the size of the perturbation is small. Extending this concept further, it is natural to ask what happens r if the morphology of the potential energy surface changes (typically slowly) such that the potential energy at a stable equilibrium position ceases to be a minimum, r or if the ball is subject to a relatively large perturbation or disturbance that may push the ball well beyond the local neighborhood of the minimum. These are the two situations depicted in Fig. 1.1. The former case is the basis of most studies in classical buckling. The application of an external axial load is assumed to take place quasi-statically, and buckling occurs (typically leading to large deflections) as the ball can no longer maintain its position. Many practical examples like this can be handled very effectively by use of statics. Most interest is naturally focused on the behavior of the system prior to buckling when the system is changing sufficiently slowly that kinetic energy can safely be ignored in the Lagrangian description (although it may still be useful to gain information based on dynamics). However, in the latter case, the application of a large (say, sudden or periodic) perturbation inevitably leads to a dynamic, perhaps unbounded, response. In fact, even in those cases in which a static approach works well, if we want to track the postcritical behavior, we may still need to use a dynamic approach, for example, one in which a system subject to a slowly increasing load results in a fast dynamic jump at buckling. Figure 1.2 adds some specificity to the scope of the material covered in this book using the behavior exhibited by a vibrating thin beam: r Figure 1.2 illustrates a beam undergoing small-amplitude free vibrations, that is, with P(t) = F (t) = 0. This is a thoroughly linear situation, with the straight configuration the only equilibrium and damping causing dynamic behavior to decay. Exact solutions are available; natural frequencies are constant and scale with the stiffness of the beam. For example, a longer beam is less stiff and thus natural frequencies are lower. Clamped boundary conditions lead to higher natural frequencies than simply supported, and so on. r The presence of a constant axial load [but with F (t) = 0] tends to reduce the natural frequencies if the load is compressive and below its critical value. If the axial load is sufficiently large (i.e., greater than critical), postbuckled (nontrivial)

15:5

P1: RTT Chapter-01

CUFX159-Virgin

4

0 521 85648 5

August 14, 2007

Context: The Point of Departure

r r

r r r

equilibria exist, and natural frequencies can be computed about these nontrivial equilibria. For laterally excited systems (F (t) / 0 but with P = 0), we can have resonance. This may also occur about postbuckled equilibria when P > Pcr . If the axial load is a function of time (say, periodic), then the system may also lose stability (depending on the frequency of excitation) through parametric resonance. If the ends of the beam are both constrained against moving (in-plane) then membrane, or stretching, forces arise. In each of the preceding scenarios the vibration may have large amplitude. Many of these scenarios might occur simultaneously. For example, a postbuckled beam might snap through if excited laterally.

Thus this range of behavior encompasses both small-amplitude and largeamplitude motion about both trivial and nontrivial equilibria. Access to analytic solutions becomes restricted as the complexity (and nonlinearity) of the system increases. Damping oftens needs to be considered also. Although the example of the prismatic beam has been used here, extensions to other types of axially loaded structures, like plates and shells is easy to envision. Furthermore, some of these situations may lead to instability (both static and dynamic), which is of particular concern to engineers. It is worth mentioning that aerospace structures provide a natural context for much of this material; the continual quest for lighter vehicles naturally brings with it issues of vibration and stability. Some practical examples of slender structures in aerospace engineering in which axial loads and dynamics may need to be considered are shown in Fig. 1.3. These images all portray aerospace systems. Spacecraft applications tend to be very lightweight: Thin-film solar sails designed for deep-space propulsion; highaltitude unmanned surveillence craft like the Predator; lightweight solar-powered high-endurance aircraft like the Pathfinder; the shuttle; international space station; rotorcraft; and military aircraft all possess slender structural components subject to a variety of loading conditions including vibration and axial-load effects. Figure 1.4 shows some other examples of slender structures. They range from bridges to pipelines, telescopes to submarines, oil tankers to high-rise buildings. The vibrations of axially loaded structures also occur at very small scales, including the increasingly important range of applications in nanotechnology. The guitar string is an obvious case. The axial load in this case can only be tensile, but it is interesting to note the slightly angled bridge of the guitar – this accounts for the slight amount of bending stiffness in the thicker strings. Hence this book is broadly divided into two main parts to cover these rather wide-ranging scenarios. A conventional division in the presentation of vibration problems is between free and forced vibration. That convention is somewhat followed here in the development of the material. However, there are occasions for which this division is not clear (e.g., an impulsive force can also be viewed as an initial velocity), but in terms of organizing the material, this seemed to be a natural

15:5

P1: RTT Chapter-01

CUFX159-Virgin

0 521 85648 5

August 14, 2007

Context: The Point of Departure

Figure 1.3. Examples of slender structures in an aerospace context. Courtesy of NASA. See color plates I–IV following page xvi.

5

15:5

P1: RTT Chapter-01

CUFX159-Virgin

6

0 521 85648 5

August 14, 2007

Context: The Point of Departure

Figure 1.4. More examples of slender structures. See color plates V–VIII following page xvi.

15:5

P1: RTT Chapter-01

CUFX159-Virgin

0 521 85648 5

August 14, 2007

Context: The Point of Departure

Figure 1.4. (continued) More examples of slender structures. See color plates V–VIII following page xvi.

choice. The next chapter will provide a brief overview of basic mechanics (which can be omitted by the more advanced reader), followed by a treatment of the interplay of dynamics and stability, without introducing too much in the way of mathematics, but still providing a flavor of the types of more practical structural systems considered later in the book.

7

15:5

P1: RTT Chapter-02

CUFX159-Virgin

0 521 85648 5

August 14, 2007

2 Elements of Classical Mechanics

2.1 Introduction This chapter develops the theoretical basis for the derivation of governing equations of motion. It starts with Newton’s second law and then uses Hamilton’s principle to derive Lagrange’s equations. A number of conservation laws are introduced. The theory is developed initially for a single particle and extended to systems of particles where appropriate. The emphasis is placed on building the theory relevant to the types of physical system of interest in structural dynamics. Other than the usual limitations regarding relativistic and quantum effects, we also restrict ourselves to translational (rather than rotational) systems, which is largely a matter of coordinates. The majority of problems in this book involve systems in which the forces developed during elastic deformation play a crucial role. Certain standard problems in classical mechanics, for example the central force motion leading to the two-body problem or particle scattering, are not relevant here and are not considered. We shall see the important role played by energy methods in studying the dynamics of structures. Classical mechanics has a long history and in-depth treatment of the subject can be found in Goldstein [1], Whittaker [2], and Synge and Griffith [3] and, of course, going back to the early developments of Newton [4], Euler [5], and Lagrange [6].

2.2 Newton’s Second Law The natural starting point in any text covering an aspect of classical mechanics are Newton’s laws of motion. They date back to 1686, with the second being the most important: A body acted upon by a force moves in such a manner that the time rate of change of momentum equals the force. Mathematically we introduce the concept of a linear momentum vector p defined as the product of mass and velocity: p = mv,

8

(2.1)

17:48

P1: RTT Chapter-02

CUFX159-Virgin

0 521 85648 5

August 14, 2007

2.2 Newton’s Second Law

9

where m is the mass and v is the velocity vector. We can thus write Newton’s second law as F=

dp d = (mv), dt dt

(2.2)

in which F is the force vector. To apply this law we need to specify motion relative to a reference frame. If we define an absolute position vector, r, in an inertial frame (i.e., a frame at rest or moving with a constant velocity relative to the “fixed” stars), then the corresponding absolute velocity vector is given by v=

dr = r˙ , dt

(2.3)

where an overdot signifies a time derivative. Thus we can further express Newton’s second law in its more familiar form as F=m

dv = m¨r = ma, dt

(2.4)

where a is an absolute acceleration vector and we have assumed m does not vary with time. Equation (2.4) is a (set of) second-order ordinary differential equation fundamental to the study of mechanics. In general, F = F(r, r˙ , t),

(2.5)

and a solution r(t) that satisfies this equation can be obtained given appropriate initial conditions r(t0 ) and r˙ (t0 ). For the types of systems of relevance to the material covered in this book, these solutions are unique. The forces entering Eq. (2.5) arise from a number of different sources in structural dynamics: stiffness, inertia, excitation and damping being the most important. The SI units of force are newtons (N), where 1 N = 1 kg m/s2 . Clearly, if F = F(t), then it would be a straightforward task to integrate Eq. (2.4) directly to obtain v(t) and then r(t). However, this will not typically be the case (as elastic forces tend to depend on the change in position), and a variety of techniques can be called on to solve differential equations. We observe at this point that solutions to equations of the type (2.4) will often involve oscillations, and also that there may not be analytic solutions available, especially in those situations in which nonlinear terms are present. Further discussion of nonlinearity and other aspects of differential equations are left to later chapters. However, the concept of stability (which will be developed continuously throughout this book) involves considering the manner in which closely adjacent solutions of Eq. (2.4) behave as a function of time, and specifically, when one of those solutions represents some kind of steady or equilibrium solution.

17:48

P1: RTT Chapter-02

CUFX159-Virgin

0 521 85648 5

August 14, 2007

10

Elements of Classical Mechanics

2.3 Energy and Work Now suppose F = F(r). We can obtain information about the solution to Eq. (2.4) by performing a path integral with respect to r along the trajectory: r(t) t t 2 d r dr F(r) · dr = F(r) · r˙ dt = m · dt (2.6) 2 dt r(t0 ) t0 t0 dt t d 2 1 1 1 (˙r )dt = mv2 (t) − mv2 (t0 ), (2.7) = m 2 dt 2 2 t0 which gives the magnitude of the velocity [rather than r(t)] provided the integral on the left-hand side of Eq. (2.6) can be performed. This is not a straightforward matter because r(t) (which is unknown) appears in the upper limit and a path integral depends on the path of integration. However, if we let the path of this integral [in Eq. (2.6)] be called C, then we can introduce the work done by the force F moving along this path as WC = F · dr, (2.8) C

and, defining the kinetic energy as T=

1 2 mv , 2

(2.9)

we can rewrite Eq. (2.7) as WC = T2 − T1 ,

(2.10)

which is a statement of the work – energy theorem. It turns out that there is a relatively large class of problems for which the work done for any admissible path between points 1 and 2 depends on only the end points of the path. In these cases forces are called conservative, and they play a dominant role in the static analysis of buckling, for example. For a conservative force F(r), consider two paths C1 and C2 connecting two points r1 and r2 . In this case we can write F · dr = F · dr, (2.11) C1

which implies that

C2

F · dr = 0,

(2.12)

where the closed integral is performed from r1 to r2 and back again. We define the work done by a conservative force in moving a particle from a reference point, r0 , to an arbitrary position r as the potential energy, r0 V(r) = Fc · dr, (2.13) r

17:48

P1: RTT Chapter-02

CUFX159-Virgin

0 521 85648 5

August 14, 2007

2.4 Virtual Work and D’Alembert’s Principle

11

and the work done in terms of the potential energy of end points we can write as r0 r0 r2 Fc · dr = Fc · dr − Fc · dr, (2.14) r1

and therefore

r2

r1

r2

Fc · dr = V(r1 ) − V(r2 ) = V1 − V2 .

(2.15)

r1

The same result can be obtained if we write F as the gradient of the scalar function: F = −∇V,

(2.16)

where, for example, in Cartesian coordinates we have ∇≡

∂ ∂ ∂ i + j + k. ∂x ∂y ∂z

(2.17)

The potential energy is defined to within an additive constant, but because the important behavior depends on the change in potential energy, this constant is usually chosen to facilitate the solution procedure (and often zero is a convenient choice). Conservation of Energy. In the absence of external forcing or damping, the con-

cept of conservation of total mechanical energy provides a useful framework for analyzing a dynamic system. Equating Eqs. (2.10) and (2.15) we have T2 − T1 = V1 − V2 ,

(2.18)

and because we can assign r2 as any point on the path, then we obtain the conservation of energy T + V = E,

(2.19)

where E is a constant and represents the total (mechanical) energy of the system. We can thus make this statement: If the forces acting on a particle are conservative, then the total energy of the particle (T + V) is conserved. These concepts are easily extended to include systems of particles, and a number of other conservation theorems can be developed. For example, if a particle is free from the effects of any force, then the linear momentum p˙ = 0 and thus p is a constant. A similar expression can be developed in terms of angular momentum. Clearly these conserved quantities can play a significant role in facilitating a solution r(t) to a physical problem.

2.4 Virtual Work and D’Alembert’s Principle In practical situations it may be quite difficult to describe all the forces acting on a system in a vectorial context. We will see that this is one of the reasons that conducting an energy approach is often easier than using Newton’s laws directly. However,

17:48

P1: RTT Chapter-02

CUFX159-Virgin

0 521 85648 5

August 14, 2007

12

Elements of Classical Mechanics

it is possible to make use of a variational principle in mechanics to facilitate the solution procedure, and this involves the concept of virtual displacements. Suppose we have a particle in equilibrium (described by a position vector r) under a set of forces F. If the position of the particle is subject to infinitesimal changes (i.e., small variations in the systems coordinates, compatible with any system constraints) then the total virtual work done is n δW = Fi δr = 0 (2.20) i=1

for a system in equilibrium, where the symbol δ is given to instantaneous, virtual variations. This can be generalized for a number of particles and, indeed, for elastic bodies, which comprise the largest interest in this book. We can thus state the principle of virtual work: For a system of forces acting on a particle, the particle is in statical equilibrium if, when it is given any virtual displacement, the net work done by the forces is zero. There are a number of ways in which this statement [and Eq. (2.20)] can be put to practical use. We can divide the forces into two categories: applied forces and constraint forces. It can be shown that the virtual work that is due to constraint forces acting through small virtual (termed reversible) displacements is zero, and the principle of virtual work is adjusted accordingly. In applications to structural mechanics, it is convenient to also divide the work into two parts: that due to external loads and that due to internal forces, and thus δWe + δWi = 0.

(2.21)

We can incorporate dynamics into the framework of virtual work by using D’Alembert’s principle. We achieve this by writing Newton’s second law as F − m¨r = 0,

(2.22)

in which m¨r is called the inertia force. Therefore we can view this as a statement of dynamic equilibrium, and in simple structural dynamics problems this is often the easiest means of obtaining the equations of motion. The statement of virtual work can thus be written in a more general form for a system of N particles of mass mi acted on by forces Fi as N (Fi − mi δ¨ri ) · δri = 0,

(2.23)

i=1

and D’Alembert’s principle may be stated thus: The virtual work performed by the effective forces through infinitesimal virtual displacements compatible with the system constraints is zero. Here, “effective forces” refers to the combination of regular and inertia forces.

17:48

P1: RTT Chapter-02

CUFX159-Virgin

0 521 85648 5

August 14, 2007

2.5 Hamilton’s Principle and Lagrange’s Equations

13

However, in contrast to the energy approaches of the next section, Eq. (2.23) still describes motion in terms of physical, vectorial coordinates. A number of issues surround the independence of coordinates, for example, in systems possessing m equations of constraint the number of degrees of freedom (DOFs) is three fewer than the number of rectangular coordinates needed to describe the positions of all the particles. Suppose we have already eliminated the forces of constraint (because they do no work) and rewrite Eq. (2.23) as N

Fi · δri = 0,

(2.24)

i=1

where Fi is now a combination of the applied and inertia forces. To satisfy equilibrium, however, we need independent coordinates for Fi = 0 (i = 1, 2, . . . , N). It can be shown that transforming from the ri coordinates to generalized coordinates qj and then taking infinitesimal virtual displacements leads to the virtual work being written in the form ⎛ ⎞ n δW = ⎝ Qj ⎠ δqj , (2.25) j =1

where Qj =

N

Fi ·

i=1

∂ri , ∂qj

j = 1, 2, . . . n,

(2.26)

and the variations in r are in the q directions. The Qj are called the generalized forces. Equilibrium is thus given by Qj = 0,

j = 1, 2, . . . , n.

(2.27)

2.5 Hamilton’s Principle and Lagrange’s Equations Although Newton’s laws are remarkably useful, there are a number of limitations. These concern systems comprising particles at very small distances and also systems in which very high velocities are involved. These types of systems are of no concern in this book, but there are many circumstances in the macromechanical world for which determining all the forces present in a system is a challenging or even impossible task. An alternative approach is based on Hamilton’s principle, which can be used to derive equations of motion via Lagrange’s equations. Although they can be shown to be equivalent to Newton’s second law, they provide a more powerful and global approach to solving problems in mechanics. A particular advantage is the flexibility in choosing coordinate systems. Attention is focused primarily on conservative systems, with a more thorough discussion of nonconservative forces left until later.

17:48

P1: RTT Chapter-02

CUFX159-Virgin

0 521 85648 5

August 14, 2007

14

Elements of Classical Mechanics

Hamilton’s principle, for conservative systems, states Of all possible paths along which a dynamical system may move from one point to another within a specified time interval (consistent with any constraints), the actual path followed is that which minimizes the time integral of the difference between the kinetic and potential energies. The integral referred to in this statement is often called the action I and can be written as t2 t2 I= (T − V)dt = Ldt, (2.28) t1

t1

where L is the Lagrangian. Thus the issue is to find the minimum of this integral, a classic problem in the calculus of variations. If L depends on a single coordinate, say q [together with its time derivative q(t) ˙ and time t], then we need to find the trajectory q(t) that minimizes t2 I= L [q(t), q(t), ˙ t] dt. (2.29) t1

To do this we need to consider what other permissible trajectories do in comparison with q(t): Any neighboring trajectory must make I increase relative to the minimum. We consider the behavior of a close-by trajectory given by q(t) + φ(t),

(2.30)

where we suppose q(t) is the path corresponding to the minimum, is a small value, and the function φ(t) is zero at the end points t1 and t2 but is otherwise any function of time. Equation (2.29) is thus transformed to t2

˙ I= L q(t) + φ(t), q(t) ˙ + φ(t), t dt. (2.31) t1

Mathematically we express the condition for a minimum as dI = 0, d which then leads to

t2 t1

∂L ∂L ˙ φ+ φ dt = 0. ∂q ∂q˙

The second term in the integrand can be integrated by parts, leaving

t2 ∂L d ∂L − φ(t)dt = 0. ∂q dt ∂q˙ t1

(2.32)

(2.33)

(2.34)

For this to be a minimum for arbitrary φ(t), the term in the large parentheses must be zero, which gives us Lagrange’s equation: ∂L d ∂L − = 0. ∂q dt ∂q˙

(2.35)

17:48

P1: RTT Chapter-02

CUFX159-Virgin

0 521 85648 5

August 14, 2007

2.5 Hamilton’s Principle and Lagrange’s Equations

15

This can be extended to incorporate situations in which the trajectory depends on a number of independent coordinates qj (t): ∂L d ∂L − = 0, ∂qj dt ∂q˙ j

j = 1, 2, . . . , n.

(2.36)

There is considerable advantage in the Lagrangian approach in that the coordinates need not be physically meaningful. Generalized coordinates consist of any set of quantities that fully describes the state of a system, and we may view our dynamic system as evolving in this configuration space. The generalized coordinates are often referred to as qj , they are not unique, and their time derivatives are the generalized velocities q˙ j .

2.5.1 Constraints In many physical situations the motion of a system is subject to constraints. That is, there is some kind of kinematic restriction on the motion, usually involving a relation between coordinates, their rates of change, or time. Forces arise because of the constraints, but because they depend on the motion itself, they are not known a priori. However, if the constraint can be expressed as position coordinate relations (or just involve time explicitly) then it can be expressed in a differential form, is termed holonomic, and can be incorporated into the Lagrangian description without much difficulty. That the motion is restricted leads naturally to a reduced number of DOFs; that is, we seek to select independent generalized coordinates that do not violate the constraints, and, because the constraint forces do no virtual work, they do not appear in the equations of motion. However, another class of problem involves constraints that influence the rates of change of generalized coordinates. They may be expressed as inequalities or as nonintegrable differential relations and are termed nonholonomic. They cannot be reduced to independent generalized coordinates, and appropriate equations of motion must include the constraints. In practice, the method of Lagrange multipliers is used [7], in which the generalized coordinates and constraint forces are obtained simultaneously. Holonomic constraint forces can also be handled in this way, although they are not of direct interest to the material covered in this book.

2.5.2 Conservation Laws We have seen how certain quantities (e.g., the total mechanical energy) may be conserved. This was developed from basic definitions of work and energy and their relation to Newton’s second law. We can show that similar relations can be developed by using the Lagrangian description. There may often occur instances in which a symmetry property enables a considerable simplification to be made, which leads to the absence of a particular

17:48

P1: RTT Chapter-02

CUFX159-Virgin

0 521 85648 5

August 14, 2007

16

Elements of Classical Mechanics

coordinate in the Lagrangian. Suppose the missing (or ignorable) coordinate is qj ; then its Lagrange’s equations will be d ∂L = 0, (2.37) dt ∂q˙ j which implies ∂L = constant, (2.38) ∂q˙ i and because ∂L/∂q˙ j = m˙qj = p j , this means the generalized momentum is conserved, that is, we effectively have a constant of the motion. Another class of problem involves those in which time does not appear explicitly in the Lagrangian, and thus ∂L = 0. (2.39) ∂t In this case, we write the total derivative (L can change in time only through its dependence on the coordinates and velocities) as dL ∂L dq ∂L d˙q = + = 0. dt ∂q dt ∂q˙ dt

(2.40)

From Lagrange’s equations, we have ∂L d ∂L = , ∂q dt ∂q˙ and therefore dL = dt

∂L d˙q d ∂L q˙ + . dt ∂q˙ ∂q˙ dt

We recognize that this is the derivative of a product, that is,

dL d ∂L = q˙ , dt dt ∂q˙ which can be written as d dt

∂L q˙ − L = 0. ∂q˙

(2.41)

(2.42)

(2.43)

(2.44)

This can now be integrated, and the term in parentheses is therefore constant in time: ∂L q˙ − L = H = constant. (2.45) ∂q˙ If the potential energy does not depend explicitly on the velocities or time, we have V = V(q), and, using L = T − V, we obtain ∂L ∂(T − V) ∂T = = , ∂q˙ ∂q˙ ∂q˙

(2.46)

∂T − (T − V) = H. ∂q˙

(2.47)

and Eq. (2.45) becomes q˙

17:48

P1: RTT Chapter-02

CUFX159-Virgin

0 521 85648 5

August 14, 2007

2.6 Nonconservative Forces and Energy Dissipation

17

Assuming the kinetic energy depends on only the generalized velocities and that this relation is the standard quadratic form, then we obtain q˙

∂T = 2T, ∂q˙

(2.48)

and Eq. (2.47) gives us the total energy of the system: H = T + V.

(2.49)

Therefore, we have the result that H (termed the Hamiltonian) is equal to the total energy of the system if the Lagrangian does not depend explicitly on time, and the potential energy does not depend on velocity. These concepts can easily be extended to include systems of particles, provided the equations of transformation relating regular and generalized coordinates is also independent of time. It is possible to express the Lagrangian (and specifically the velocities associated with the generalized coordinates) in terms of generalized momenta. An advantage of doing this includes the fact that it is often the momentum that is a conserved quantity, and the Hamiltonian also has more physical meaning (through its relation with energy for conserved systems) than the Lagrangian. It also results in a set of 2n first-order equations rather than the n second-order Lagrange’s equations, and this may assist the development of numerical solutions. It turns out that the resulting Hamilton’s equations have certain symmetric features that render them unchanged under transformation of coordinates and momenta, and they are often referred to as the canonical equations of motion.

2.6 Nonconservative Forces and Energy Dissipation Not all forces are derivable from a potential, that is, there may not be a potential energy function V that satisfies Eq. (2.16) for a particular system. We can write the external forces acting on the system in the form Fi = FPi + FDi ,

(2.50)

where FPi is derivable from a potential V = V(qi ) and FDi is not. Thus we can also divide the virtual work into conservative and nonconservative parts, δW = δWP + δWD.

(2.51)

The first term on the right-hand side of Eq. (2.51) is defined by Eq. (2.16), and, by virtue of Eq. (2.26), we can write Lagrange’s equation as ∂L d ∂L − = QDj . ∂qj dt ∂q˙ j

(2.52)

We can also rearrange Eq. (2.51) and write δWD = δW − δWP = F · dr − FP · dr = dT − (−dV) = d(T + V) = dE.

(2.53)

17:48

P1: RTT Chapter-02

CUFX159-Virgin

0 521 85648 5

August 14, 2007

18

Elements of Classical Mechanics

Integrating over the path, we get r2 FD · dr = r1

E2

dE = E2 − E1 .

(2.54)

E1

Thus the change in total energy is equal to the work done by the nonconservative forces. In structural mechanics it is often found that there are two general classes of forces that do not arise from a potential function. In these cases, the total mechanical energy is not conserved. We will occasionally consider a load that is nonconservative in nature because of its direction changing (for example, following the slope at the end of a beam), but in general there are two classes of nonconservative forces encountered in mechanical systems. In cases in which the energy decreases we use the term dissipative forces. The main example is the loss of energy through damping, which in many cases relates to a force proportional to velocity. The other main type of nonconservative force is time dependent and often associated with the external driving of a system. 2.6.1 Damping In mechanical systems, energy dissipation is inevitable. If we assume that a certain class of nonconservative force acting on a single particle is a function of velocity only, then FDi = −g(vi )vi ,

(2.55)

where we also assume that the force is directed opposite to the velocity. Very often this relation will describe linear-viscous damping in unidirectional motion: δW = Fix = −cxi x˙ i ,

(2.56)

where c is the damping coefficient. The virtual work done by this dissipative force is F · δr δW = i

=−

n

cxi x˙ i δxi

i=1

⎡ ⎤ n n ∂xi ⎣ =− cxi x˙ i δqj ⎦ ∂qj i=1

j =1

n n 1 ∂ 2 =− cx x˙ δqj . 2 ∂q˙ j i i j =1

(2.57)

i=1

Thus the corresponding generalized force is given by QDj = −

n n 1 ∂ 1 ∂ cxi x˙ i = − cxi x˙ 2i . 2 ∂q˙ j 2 ∂q˙ j i=1

i=1

(2.58)

17:48

P1: RTT Chapter-02

CUFX159-Virgin

0 521 85648 5

August 14, 2007

2.6 Nonconservative Forces and Energy Dissipation

19

Introducing Rayleigh’s dissipation function D, 1 cxi x˙ 2i , 2 n

D=

(2.59)

i=1

we can obtain the dissipative generalized forces from δW =

n

QDj δqj = −

j =1

n ∂D δqj , ∂q˙ j

(2.60)

j =1

and thus a more general form of Lagrange’s equation: ∂L ∂D d ∂L − + = 0, dt ∂q˙ j ∂qj ∂q˙ j

j = 1, 2, . . . , n.

(2.61)

2.6.2 Time-Dependent Forces We now consider another group of nonconservative forces, namely, time-dependent forces, as they crop up quite naturally in structural systems subject to periodic excitation or impulses, for example. Suppose a system is subject to forces F (t) that depend on time (but are independent of the generalized coordinates), then a timedependent Lagrangian will have a term in the form qF (t). Clearly, this will then lead to the appearance of F (t) in the resulting equation of motion. Furthermore, the principle of impulse and momentum can be used via the Lagrangian approach. Consider an impulsive force t0 +t Fˆ = F(t)dt. (2.62) t0

Intgerating Lagrange’s equation from t1 = t0 to t2 = t0 + t and allowing t → 0 leads to the impulsive form of Lagrange’s equation: ∂T ∂T ˆ j, − =Q j = 1, 2, . . . , n, (2.63) ∂q˙ j 2 ∂q˙ j 1 ˆ j is a generalized impulse. where Q It is also possible to derive a velocity-dependent potential for some problems, although this will not be encountered in the types of applications in this book. Finally, we note that extension of Lagrangian mechanics to continuous systems (with an infinite number of DOFs) will be dealt with in a later chapter. In summary, we will often be in a position to write potential and kinetic energies and make use of Lagrange’s equation: d ∂L ∂L ∂D − + = qj Fj (t), dt ∂q˙ j ∂qj ∂q˙ j

j = 1, 2, . . . , n.

(2.64)

Given the types of axially loaded elastic structures of primary interest in this book, we inevitably focus quite heavily on equations of motion of the type Mq¨ j + C˙qj + f (qj , λ) = F (t),

j = 1, 2, . . . , n,

(2.65)

17:48

P1: RTT Chapter-02

CUFX159-Virgin

0 521 85648 5

August 14, 2007

20

Elements of Classical Mechanics

in which we have a mass matrix M, damping matrix C, the stiffness matrix f , which depends (often nonlinearly) on external axial loads (λ), and both free [F (t) = 0] and forced vibrations are examined for a wide variety of slender structures, and where we often begin by assuming C = 0.

2.7 Strain Energy We now introduce a brief description of strain energy, as this is a fundamental aspect of structural mechanics that we will repeatedly encounter throughout this book. In Section 2.3 the concepts of energy and work were introduced. Because many of the specific physical systems to be considered later in this book involve deformable bodies, it is important to understand the manner in which (elastic) energy is stored as strain energy, especially in bending. We start with a basic definition of strain energy per unit volume, or strainenergy density U0 , for a uniaxially loaded system given by 1 U0 = σx dx , (2.66) 0

in which σx is stress and x is strain. We see that the strain-energy density is equal to the area under the stress–strain from x = 0 to x = 1 . A thorough background to elasticity and the general description of fundamental issues in solid mechanics can be found in Langhaar [8]. The total strain energy stored in the solid is U= U0 dV. (2.67) V

For a linear isotropic elastic material, we have σx = Ex , where E is Young’s modulus, and, in this case, expression (2.66) becomes U0 =

σx2 . 2E

(2.68)

The strain energy U0 is always positive-definite, and the conservation of energy introduced in Subsection 2.5.2 can also be stated in terms of strain energy: If an elastic body is in equilibrium under an external force system, then the internal strain energy that is due to deformation is equal to the work of the externally applied force system. Application to Beams. We finish this chapter by briefly focusing on the strain en-

ergy associated with slender beams because they represent an important element in this book, but these concepts can be easily extended to strings, plates, and so on. A prismatic bar of cross-sectional area A and length L is subjected to an axial load P (which passes through the centroid of the cross section). The stress σ = P/A and thus the total strain energy in the bar is U=

P2 L . 2AE

(2.69)

17:48

P1: RTT Chapter-02

CUFX159-Virgin

0 521 85648 5

August 14, 2007

References

Suppose the beam is subjected to an applied lateral load rather than axial and has a second moment of area I. In this case, a bending moment causes a stress σx = My/I (where y is the distance from the centroid) and thus a total strain energy of L 2 M U= dx. (2.70) 0 2EI When standard beam theory [9] for a prismatic beam is used, this simplifies to the well-known expression L 2 2 1 ∂ w U = EI dx. (2.71) 2 ∂x2 0 The interaction between axial and bending effects will be a central theme. Throughout this book extensive use will be made of energy concepts [10]. Kinetic energy and the work done by external loads are added to the consideration of strain energy for a variety of structural systems. In simpler cases, we will make direct use of Newton’s laws, but for complex systems, energy will provide a powerful (equilibrium and stability) framework in which to study the dynamics of axially loaded structures. References [1] [2]

H. Goldstein. Classical Mechanics. Addison-Wesley, 1980. E.T. Whittaker. A Treatise on the Analytical Dynamics of Particles and Rigid Bodies. Cambridge Mathematical Library, 1988. [3] J.L. Synge and B. A. Griffith. Principles of Mechanics. McGraw-Hill, 1959. [4] I. Newton. Philosophiæ Naturalis Principia Mathematica. London, 1687. [5] L. Euler. Methodus Inveniendi Lineas Curvas Maximi Minimive Proprietate Gaudentes. Marcum Michaelem Bousquet, 1744. [6] J.L. Lagrange. Mecanique Analytique. Courier, 1788. [7] L. Meirovitch. Principles and Techniques of Vibrations. Prentice Hall, 1997. [8] H.L. Langhaar. Energy Methods in Applied Mechanics. Wiley, 1962. [9] S.P. Timoshenko and J.M. Gere. Theory of Elastic Stability, 2nd ed. McGraw-Hill, 1961. [10] J.H. Argyris and S. Kelsey. Energy Theorems and Structural Analysis. Butterworth, 1960.

21

17:48

P1: RTT Chapter-03

CUFX159-Virgin

0 521 85648 5

August 14, 2007

3 Dynamics in the Vicinity of Equilibrium

We are primarily interested in the concept of a slow evolution toward, and through, instability. The quasi-static change of a system parameter (typically the axial load) allows any instability mechanisms to unfold, especially if transient dynamics is present. Before moving on to consider a range of structural components, we investigate the response of a linear oscillator under various conditions, including a continuously deteriorating stiffness. This is followed by consideration of some aspects of the Lagrangian approach and the interaction of dynamics and stability. The final part of this chapter then uses a phenomenological model to describe the dynamics and stability of a simple physical system. Again, there is a large literature concerning the linear oscillations of mechanical systems. The interested reader can find good coverage of basic material in Refs. [1–7].

3.1 The Linear Oscillator Consider the continuous-time evolution of a dynamical system governed by x˙ = F(x, t),

x ∈ Rn ,

t ∈ R,

(3.1)

where x is a state vector that describes the evolution of the system under the vector field F. Given an initial condition, typically the values of the state vector prescribed at t = 0, that is, x(0), we can seek to solve system (3.1) to obtain a trajectory x(t), or orbit, along which the solution evolves with time. We then seek to ascertain the stability of the system, generally as a function of a (control) parameter µ, and thus consideration of x˙ = F(x, µ, t)

x ∈ Rn ,

t ∈ R,

(3.2)

will play a central role in the material contained in this book [8, 9]. Application of Newton’s second law relates acceleration and force (and hence position), and thus often results in a second-order ordinary differential equation of the form d2 x = −ω2n x, dt2 22

(3.3)

17:50

P1: RTT Chapter-03

CUFX159-Virgin

0 521 85648 5

August 14, 2007

3.1 The Linear Oscillator

23

z(t) x(t) k(x,µ,t) F(t)

m c

Figure 3.1. A spring–mass–damper system.

where ωn is a constant (the natural frequency), and with x˙ ≡ dx/dt, we obtain the nondimensional governing equation of motion, x¨ + ω2n x = 0,

(3.4)

subject to the two initial conditions, x(0), and x(0). ˙ This is the equation of motion governing the dynamic response of the spring–mass system shown in Fig. 3.1 with √ ωn = k/m (k and m constant) and all other parameters set equal to zero, that is, c = F (t) = z(t) = 0. Because Eq. (3.4) is a linear, homogeneous, ordinary differential equation with constant coefficients, we can write the solution as x(t) = Aest .

(3.5)

Placing Eq. (3.5) into Eq. (3.4), we find that s = ±iωn , and thus the general form of the solution is given by x(t) = Aeiωn t + Be−iωn t .

(3.6)

Alternatively, using Euler’s identities, we can write x(t) = C cos(ωn t) + D sin(ωn t).

(3.7)

To determine A and B (or C and D), we make use of the initial conditions to get x(t) = x(0) cos(ωn t) +

x(0) ˙ sin(ωn t). ωn

(3.8)

We can convert this system into a pair of coupled, first-order ordinary differential equations (in state-variable format) by introducing a new variable y = x, ˙

(3.9)

and substituting in Eq. (3.4) gives x˙ = y, In matrix notation,

y˙ = −ω2n x.

x˙ 0 = y˙ −ω2n

1 0

x . y

(3.10)

(3.11)

17:50

P1: RTT Chapter-03

CUFX159-Virgin

0 521 85648 5

August 14, 2007

24

Dynamics in the Vicinity of Equilibrium

2

2

(a)

x

0

0

-2

(b)

x

0

5

10

15

t

20

-2

0

5

10

15

t

20

Figure 3.2. (a) Two trajectories exhibiting simple harmonic motion: x(0) = 1, x(0) ˙ = 0, and x(0) = 0, x(0) ˙ = 1; (b) solutions to Eq. (3.16). (i) x(0) = −1, x(0) ˙ = 1 (solid curve); (ii) x(0) = 0.0001, x(0) ˙ = 0 (dotted curve); and (iii) x(0) = −0.99, x(0) ˙ = 1 (dot–dashed curve).

A plot of Eq. (3.8) (with ωn = 1) is shown in Fig. 3.2(a) for two typical sets of initial conditions. At this point, we simply note that, from Eq. (3.8) and its derivative (to obtain velocity), we can envision the trajectory evolving with time in a repeating manner. Plotting position versus velocity (the phase plane) will be a useful way of displaying dynamic behavior later in this book, and in this (undamped) case it is apparent that the motion is described by ellipses. This is, of course, the periodic behavior we would expect for a simple spring–mass system with ωn (assumed real, i.e., ω2n > 0) identified as the natural frequency of the oscillation. In terms of a heuristic concept of stability, we might consider this behavior to be neither stable or unstable, as any motion we might initiate does not decay or grow, but simply persists. Solution (3.7) can also be written as x(t) = A¯ cos(ωn t + φ),

(3.12) √ where A¯ = C2 + D2 is the amplitude and φ = arctan (C/D) is the phase. Thus we see that the larger the initial conditions, the larger the area enclosed by the ellipses, that is, 2 ¯2. x2 (t) + [x(t)/ω ˙ n] = A

(3.13)

The two trajectories shown in Fig. 3.2(a) differ by a phase φ = π/2, and thus the dashed curve can be viewed as the corresponding velocity time series. Later, we will see how this relates to energy. However, the form of the resulting motion is independent of the initial conditions. Suppose we have ω2n < 0. This is a situation that is difficult to envision, physically, but can occur, for example, in a nonlinear system if the spring stiffness becomes negative. Then the motion is governed by x¨ − ω2n x = 0.

(3.14)

Now adopting the solution x(t) = Aest leads to s = ±ωn , and thus x(t) = Aeωn t + Be−ωn t .

(3.15)

Using the definition of hyperbolic functions and the initial conditions, we also have x(t) = x(0) cosh ωn t + [x(0)/ω ˙ n ] sinh ωn t.

(3.16)

17:50

P1: RTT Chapter-03

CUFX159-Virgin

0 521 85648 5

August 14, 2007

3.1 The Linear Oscillator

25

In this case, we do not have a periodic solution: The positive exponent indicates that typically x → ∞ as t → ∞. Hence, our heuristic concept of stability indicates that this behavior is unstable. However, we also observe that we can choose very specific initial conditions (unlikely but nevertheless important cases) in which the trajectory will end up at the origin, that is, where the positive exponential term is completely suppressed, as well as the case in which the negative exponential term in Eq. (3.15) dominates for a short time before the trajectory is swept away. These cases are illustrated in Fig. 3.2(b). For all practical purposes, that is, arbitrary initial conditions, the motion is clearly unstable. The meaning of the special trajectory will be discussed at length later. Damping. The preceding examples are somewhat unrealistic in terms of practical mechanics because they do not include energy dissipation [10]. With the inevitable presence of damping the question of stability becomes less ambiguous. Typical motion will then consist of a transient followed by some kind of recurrent long-term behavior. This brings us to the fundamentally important concept of attractors. These are the special solutions alluded to earlier, and they play a key role in organizing dynamic behavior in phase space (the space of the state variables). We shall also see that for nonlinear systems unstable solutions have an important influence on the general nature of solutions. Suppose we now allow for some energy dissipation in the form of linear viscous damping, that is, c = 0 in Fig. 3.1. The equation of motion is now

x¨ + 2ζ ωn x˙ + ω2n x = 0,

(3.17)

into which a nondimensional damping ratio, ζ ≡ c/(2mωn ), has been introduced. Solutions to this equation now depend on the value of ζ. For underdamped systems, we have ζ < 1 and solutions of the form x(0) ˙ + ζ ωn x(0) x(t) = e−ζ ωn t sin ωd t + x(0) cos ωdt , (3.18) ωd where the damped natural frequency ωd is given by ωd = ωn 1 − ζ2 .

(3.19)

A typical underdamped response (ζ = 0.1) is shown in Figs. 3.3(a) and 3.3(b) as a time series and phase portrait, respectively. The origin in Fig. 3.3(b) indicates a position of asymptotically stable equilibrium; that is, any disturbance leads to a dynamic response that moves smoothly back to equilibrium. The trajectory gradually spirals down to this rest state: We can imagine a family of trajectories forming a flow as time evolves. Because this equilibrium is unique, the whole of the phase space is the attracting set for all initial conditions and disturbances [11]. Damping in this range, e.g., ζ ≈ 0.1, is quite typical for mechanical and structural systems. For a heavily (or overdamped) system ζ > 1, and in this case the form of the solution is √

x(t) = Ae(−ζ+

ζ2 −1)ωn t

√

+ Be(−ζ−

ζ2 −1)ωn t

,

(3.20)

17:50

P1: RTT Chapter-03

CUFX159-Virgin

0 521 85648 5

August 14, 2007

26

Dynamics in the Vicinity of Equilibrium

(a)

x(t) . x(t)

. 1.5 (b) x/ω n

x

.

1

x

0.5 0 −0.5 −

−1 −

−1.5 −1.5

t

x(t)1.2 . x(t) 1 (c)

−0.5

0

0.5

1

x

1.5

. 1.2 x/ω n (d) 1

.

x

0.8

−1

0.6

0.8

x

0.6

0.4

0.4

0.2

0.2

0

0

− 0.2

0

2

4

6

8

10

t

12

− 0.2 − 0.05

0

0.05

0.1

0.15

0.2

0.25

x

0.3

Figure 3.3. (a) Time series and (b) phase portraits for underdamped (oscillatory) motion, x(0) = 1.0, x(0) ˙ = 0.0, ζ = 0.1; (c) and (d), overdamped (nonoscillitory) motion, x(0) = 0.0, x(0) ˙ = 1.0, ζ = 1.5.

where A=

√ ζ2 − 1)ωn x(0) , √ 2ωn ζ2 − 1

x(0) ˙ + (ζ+

(3.21)

and √ −x(0) ˙ − (ζ− ζ2 − 1)ωn x(0) . B= √ 2ωn ζ2 − 1

(3.22)

The motion is a generally monotonically decreasing function of time and may take a relatively long time to overcome rather heavy damping forces on the way to equilibrium. A typical case is also shown in Figs. 3.3(c) and 3.3(d). The boundary between these two cases is the critically damped case, i.e., ζ = 1. In the context of this book, we will regularly encounter the situation in which the stiffness of a system degrades, and given the definition of ζ we expect not only a reduction in the natural frequency but also an increase in the damping ratio. Returning to the state-variable-matrix format of the linear oscillator, we therefore have x˙ 0 1 x = . (3.23) y˙ −ω2n −2ζωn y

17:50

P1: RTT Chapter-03

CUFX159-Virgin

0 521 85648 5

August 14, 2007

3.1 The Linear Oscillator

27

We can also write the solution in terms of the eigenvalues of the state matrix, that is, the roots of the characteristic equation λ2 + 2ζωn λ + ω2n = 0.

(3.24)

Critical damping thus relates to the discriminant’s being equal to zero. Given the scenario of a system losing stability, we can usefully view all the response possibilities of this type of linear system according to the location of the roots in the complex plane. For example, having two complex roots with negative real parts corresponds to an exponentially decaying oscillation. Summarizing these outcomes leads to Fig. 3.4 [12]. In general, we will have a system with positive stiffness and damping and thus a root structure corresponding to the upper-right-hand quadrant. Critical damping corresponds to the parabola, and phase portraits and eigenvalues are indicated for various combinations of the natural frequency and damping. The system eigenvectors organize the transient behavior in the phase portrait. Some useful terminology here includes the spiral or focus for decaying oscillatory motion (also called a sink), the node for overdamped motion, the inflected node for equal eigenvalues (and thus including the critically damped case), and the saddle for the motion characterized by having both a stable and unstable direction I R

2

n

STABLE

2 n

Figure 3.4. Phase portraits and root structure of a linear oscillator (after Thompson [12]).

17:50

P1: RTT Chapter-03

CUFX159-Virgin

0 521 85648 5

August 14, 2007

28

Dynamics in the Vicinity of Equilibrium

(eigenvector) with instability becoming dominant. We can also view the undamped case as a center. We shall focus more on higher-order dynamical systems in the next chapter, but more formally we state (see Jordan and Smith [13]) that if we have a dynamical system x˙ = Ax, where A is constant with eigenvalues λi , i = 1, 2, ..., n, then (1) if the system is stable, Re{λi } ≤ 0, i = 1, 2, . . . , n; (2) if either Re{λi } < 0, i = 1, 2, . . . , n, or if Re{λi } ≤ 0, i = 1, 2, . . . , n and there is no repeated eigenvalue, the system is uniformly stable; (3) the system is asymptotically stable if and only if Re{λi } < 0, i = 1, 2, . . . , n [and then it is also uniformly stable, by (2)]; (4) if Re{λi } > 0 for any i, the solution is unstable. We thus observe what will typically happen when the stiffness of the system degrades (e.g., because of an axial load acting on a slender structure), and this is indicated by the large arrow in Fig. 3.4. For a small amount of damping, the eigenvalues start off as a complex-conjugate pair with negative real parts. As the stiffness (and hence natural frequency) reduces, the eigenvalues merge on the negative real axis, and then their magnitudes diverge, with one entering the positive half-plane. Thus instability occurs, and solutions grow without bound. Although the preceding description relates to a single-degree-of-freedom (SDOF) linear oscillator, this type of scenario is encountered to a large extent within a variety of high-order and nonlinear systems. The geometric view afforded by a consideration of the root structure and phase portraits of families of solutions about equilibrium points is very useful. We will make extensive use of linearization to utilize this view locally to equilibrium within a nonlinear context [14].

3.2 Oscillator with a Slow Sweep of Frequency Consider again the spring–mass system shown in Fig. 3.1. Again assume that there is no damping or external forcing (c = F (t) = z(t) = 0), and that the spring stiffness decays linearly (in time) from a base value k = 1 at t = 0. We assume this decay is very slow, and characterized by a small parameter . In this case we can write the governing equation of motion as x¨ + µ2 (t)x = 0,

(3.25)

in which µ2 (t) =

k (1 − t), m

(3.26)

that is, the system will lose stability when t → 1/. If we assume that the evolution of the stiffness change is very slow ( 0 such that if r(t0 ) < δ then r(t0 ) < for all t ≥ t0 . (2) The equilibrium point xe is asymptotically stable if it is Lyapunov stable and limt→∞ r(t) = 0. (3) The equilibrium point xe is unstable if there is an a > 0 such that for arbitrarily small δ > 0 there is a motion xa (t) for which ra (t0 ) < δ and ra (t1 ) > at some time t1 > t0 . Thus we see that with these definitions instability occurs when the perturbed trajectory reaches the sphere of radius in finite time. We can view this situation graphically as shown in Fig. 4.1. Clearly, for conservative systems we expect Lyapunov definition (1) to apply [5]. In typical structural systems, we would expect

17:51

P1: KAE Chapter˙04

CUFX159-Virgin

0 521 85648 5

August 14, 2007

4.2 Multiple-Degree-of-Freedom Systems

51

some damping to be present, and assuming it is of the form of linear viscous damping (derivable from Rayleigh’s dissipation function) we might expect asymptotic stability to represent typical dynamical behavior for an autonomous structural system. Hence, after a small disturbance, a transient slowly decays back to the rest position. The presence of damping can considerably complicate the dynamic analysis of MDOF systems, although the adoption of proportional damping, as mentioned earlier, facilitates the modal analysis approach [6, 7]. We can return to some of the responses of the SDOF system to see how this more general definition encompasses the simple cases. In Figure 3.4, we showed how the response of a linear oscillator depended on the stiffness and damping. Thus we see the upper-right quadrant as corresponding to areas of aysmptotic stability. Since the primary focus of this book is axially loaded structures we will repeatedly encounter a decay in stiffness and thus a transition to instability (crossing the y axis). Although a decay in damping leads to instability as well, this type of flutter instability (usually initiated via a Hopf bifurcation [8]) is often associated with nonconservative forces (e.g., in aeroelasticity [9]) and is not a feature central to the dynamics of axially loaded structures of the type considered in this book, although one or two examples are given. Considering disturbances about an equilibrium state, we write x(t) = xe + η(t),

(4.30)

which can be substituted back into Eq. (4.27) to give d [xe (t) + η(t)] = F[xe (t) + η(t), t], dt

(4.31)

d xe (t) = F[xe (t), t], dt

(4.32)

and because

we can write d η(t) = F[xe (t) + η(t), t] − F[xe (t), t]. (4.33) dt This is the variational equation, and it plays a crucial role in determining the stability associated with certain solutions of dynamical systems. We can expand the right-hand side of Eq. (4.33) by using F[xe (t) + η(t), t] = F[xe (t), t] + DF(xe , t)η(t) + · · · +,

(4.34)

where DF(xe , t) is the matrix of first derivatives (the Jacobian) of F(xe , t) evaluated at xe : ⎤ ⎡ ∂F1 /∂x1 ∂F1 /∂x2 · · · ∂F1 /∂xN ⎢ ∂F2 /∂x1 ∂F2 /∂x2 · · · ∂F2 /∂xN ⎥ ⎥ ⎢ ⎢ . .⎥ ⎥ ⎢ . (4.35) DF(xe , t) = ⎢ ⎥ ⎢ . .⎥ ⎥ ⎢ ⎣ . .⎦ ∂FN /∂x1 · · · ∂FN /∂xN x=x e

17:51

P1: KAE Chapter˙04

CUFX159-Virgin

0 521 85648 5

August 14, 2007

52

Higher-Order Systems

Now, assuming the perturbations are small, we can truncate the higher-order terms in Eq. (4.34) and thus we can write Eq. (4.33) as dη = DF(xe , t)η(t). dt

(4.36)

This is called the linear variational equation for the system. In general, the coefficients of the Jacobian are time varying, and thus we have a non-autonomous system. We shall pay specific attention to systems in which the coefficients are periodic in Chapter 14, as this is a case commonly encountered in forced vibrations. For now, we focus on the stability of equilibrium, that is, xe (t) = xe represents a fixed point in phase space, and the coefficients in the Jacobian matrix are constant, that is, an autonomous system. Thus the focus on linear systems is justified on the grounds that nonlinear systems can often be linearized in the vicinity of equilibrium. By restricting ourselves to small neighborhoods about an equilibrium point, we can make use of much of the preceding linear theory. In terms of energy, and within a free-vibration context, we thus consider the behavior resulting from a small disturbance from an equilibrium point, q(t) = q0 + δ(t),

˙ ˙ = δ(t), q(t)

(4.37)

where δ and δ˙ are perturbation vectors. Because the nonlinearity will typically appear in the generalized positions (and assuming a smooth nonlinearity), we expand the potential energy as a Taylor series: ∂U 1 U(q1 , q2 , . . . , qn ) = U(q0 ) + δT + δT Kδ, (4.38) ∂q q=q0 2 where higher-order terms have been neglected. K is the stiffness matrix (evaluated at equilibrium). The first term on the right-hand side is an arbitrary constant scalar (which we can choose to be zero), the second term is equal to zero by virtue of the equilibrium condition, and hence we can follow the earlier theory to arrive at the linearized equations of motion: Mδ¨ + Kδ = 0,

(4.39)

where K is related to the change in potential energy about an equilibrium position and is generally a function of a set of external (axial) loads, and M is a mass matrix. Now suppose we revisit the stability of a set of ordinary differential equations in terms of the phase space. That is, we solve the algebraic eigenvalue problem arising from the equations of motion in first-order state-variable format. We also add a little damping (with a matrix of damping coefficients C) and introduce the perturba˙ T ], so that the set of equations (4.5) can be written tion state-vector as x(t) = [qT q(t) (following Meirovitch [2]) as x(t) ˙ = Ax(t),

(4.40)

17:51

P1: KAE Chapter˙04

CUFX159-Virgin

0 521 85648 5

August 14, 2007

4.2 Multiple-Degree-of-Freedom Systems

where

A=

53

I . −M−1 C

0 −M−1 K

(4.41)

We see that this is the matrix generalization of Eq. (3.11). Assuming an exponential form for the solution x(t) = eλt x

(4.42)

and placing in Eq. (4.40) (canceling eλt ), we get Ax = λx.

(4.43)

This is again the algebraic eigenvalue problem. A set of n second-order ordinary differential equations can always be converted to a set of 2n first-order ordinary differential equations. The general solution is given by x(t) =

2n

cj eλj t xj ,

(4.44)

j =1

where the coefficients cj are generally complex and depend on the initial conditions. Specifically, the eigenvalues, which can be written as λj = αj + iβj ,

j = 1, 2, . . . , 2n,

(4.45)

govern both the frequency and stability of the response by means of their imaginary and real parts, respectively. With positive stiffness, and in the absence of damping, we expect purely imaginary eigenvalues, that is, αj = 0 (which is a stable situation according to part 2 of the Lyapunov stability definition). We can think of eigenvalues distributed along the imaginary axis in the complex plane (akin to the positive x axis in Fig. 3.4). With the addition of damping, we expect to have αj < 0 and asymptotic stability (including both the focus and node illustrated in Fig. 3.4, which depend on the level of damping). Analyzing vibration problems by solving a set of secondorder systems is equivalent to converting and then solving a set of (twice as many) first-order systems. The eigenvalues of the two approaches are linked by Eq. (4.19). Although the second-order approach allows a more physical feel (in terms of natural frequencies and damping ratios), the first-order approach is popular in control theory, is well suited to systems in which coupled modes are present, is an approach very well suited to software packages like MATLAB [10], and, as we have seen, allows a more direct assessment of stability. We shall often encounter a situation in which axial loading reduces the lateral stiffness of a structure such that the potential energy can be written as 1 (aij − Pbij )qi qj , 2 n

V = U−W =

n

(4.46)

i=1 j =1

where the strain energy U (and matrix aij ) are positive definite and the work done by the axial loads (and bij ) are either indefinite or positive definite (semi-definite).

17:51

P1: KAE Chapter˙04

CUFX159-Virgin

0 521 85648 5

August 14, 2007

54

Higher-Order Systems

Placing this in the context of a set of lightly damped oscillators (and assuming a single load for simplicity) we then have Mq¨ + Cq˙ + (KL − PKG )q = 0,

(4.47)

where C is the damping matrix and KG is a matrix relating to the (prestressing) effect of the axial load. We typically start with P = 0 and thus have initial stability, and then a major question is to determine the specific (or critical) values of P that cause instability, and how this is reflected in dynamic behavior. 4.2.4 Routh–Hurwitz Criterion In the previous subsection, we saw how the stability of a system depended on the sign of the real parts of the eigenvalues [Eq. (4.45)]. To determine these signs, we need not solve the complete eigenvalue problem. Rather, a technique, referred to as the Routh–Hurwitz criterion, can be used [11]. Writing the characteristic equation in the form λn + a1 λn−1 + a2 λn−2 + · · · + an = 0,

(4.48)

it can be shown that the system is asymptotically stable (i.e., the real parts of the eigenvalues are negative) if and only if the principal minors of the n × n matrix ⎡ a1 ⎢ ⎢a3 ⎢ ⎢a ⎢ 5 ⎢ ⎢. ⎢ ⎢ ⎢. ⎢ ⎢ ⎣. .

1

0

0

···

a2

a1

1

···

a4

a3

a2

a1

.

.

0

1 ···

⎤

⎥ 0⎥ ⎥ 0⎥ ⎥ ⎥ . ⎥ ⎥ ⎥ . ⎥ ⎥ ⎥ . ⎦

(4.49)

an

.

are all positive, that is, 1 = a1 > 0, a1 1 > 0, 2 = a3 a2 a1 1 0 3 = a3 a2 a1 > 0, a a a 5

4

(4.50)

3

and so on. A necessary condition for asymptotic stability is that all the a’s be positive. Computationally, this criterion may be more convenient to evaluate than the full determinant.

17:51

P1: KAE Chapter˙04

CUFX159-Virgin

0 521 85648 5

August 14, 2007

4.2 Multiple-Degree-of-Freedom Systems

55

4.2.5 Lyapunov Functions As seen in the previous subsection, we can draw important conclusions about the stability of a dynamical system without obtaining a complete solution to the equations of motion; computing the Lyapunov stability of a system directly may not be the most straightforward approach. Also, much of the preceding stability theory was based on linear behavior (in the vicinity of equilibrium). Nonlinearity and global stability can be assessed by use of the concept of Lyapunov functions [7, 12, 13]. This is sometimes referred to as Lyapunov’s direct, or second, method [7], and it has been found to be especially useful in applications involving nonconservative forces and in control theory [14]. The concept is again based on assessing stability without solving the equations of motion and is closely related to energy for conservative systems. We introduce the Lyapunov function E(x), which we assume is a continuous and differentiable scalar function of the vector x(t) and satisfies the following conditions: (1) E(x) > 0, for all values of x(t) = 0, ˙ (2) E(x) ≤ 0, for all values of x(t) = 0. Now, if it can be shown that a Lyapunov function exists for a system, then that system is stable and is aymptotically stable if the time derivative of E is less than zero. Finding a Lyapunov function is not necessarily easy, and even if one cannot be found it does not imply that the system is unstable. The total energy of a system is often a useful place to start when searching for a Lyapunov function. With this in mind we take a brief look at our general damped MDOF system: Mq¨ + Cq˙ + Kq = 0,

(4.51)

and assuming the mass, damping, and stiffness matrices are all symmetric and positive-definite, we can confirm the asymptotic stability of the system. Let a Lyapunov function be based on the total energy for this system as ˙ E[q(t), q(t)] =

1 T ˙ + qT (t)Kq(t)]. [q˙ (t)Mq(t) 2

(4.52)

Clearly, this satisfies the first condition of Lyapunov’s direct method, that is, E(q) > 0. For the second part, we have ˙ ˙ E[q(t) q(t)] = q˙ T Mq¨ + q˙ T Kq,

(4.53)

which, by premultiplying Eq. (4.51) by q˙ T (t), we obtain ˙ ˙ E[q(t) q(t)] = −q˙ T Cq˙ ≤ 0,

(4.54)

where C is positive-definite. That a damped spring–mass system is stable is no surprise, but for highly nonlinear and nonconservative problems this is often a convenient approach.

17:51

P1: KAE Chapter˙04

CUFX159-Virgin

0 521 85648 5

August 14, 2007

56

Higher-Order Systems

4.2.6 Rayleigh’s Quotient The previous subsections showed how it may be possible to estimate the stability of a system without having to solve the equations of motion. In a similar vein, it is possible to estimate the natural frequencies of a system (and hence stability) based on an approach developed by Lord Rayleigh for eigenvalue problems [15]. We used this approach in Chapter 3 for a SDOF system, but extend it here to MDOF systems, where it finds most utility in estimating the lowest natural frequency. Given the standard eigenvalue problem, Avr = λr vr ,

r = 1, 2, . . . , n,

(4.55)

where A is a real symmetric positive-definite n × n matrix, we can express the eigenvalues in the form λr =

vTr Avr , vTr vr

r = 1, 2, . . . , n,

(4.56)

in which the eigenvalues are ordered λ1 ≤ λ2 ≤ · · · λn . The Rayleigh quotient is λ(v) = R(v) =

vT Av . vT v

(4.57)

Again for a MDOF system [Eq. (4.5)] we have the eigenvalue problem Ku = λMu,

(4.58)

and thus Rayleigh’s quotient in this case is R(λ, u) =

uT Ku . uT Mu

(4.59)

We note that the numerator is the maximum potential energy and the denominator is sometimes referred to as the zero-frequency kinetic energy during an oscillation [16]. What makes this approach especially useful in a practical context is that the assumed vector need not be terribly accurate to produce a reasonable eigenvalue estimate. This is partly due to the fact that it is stationary when perturbed around any of the actual system eigenvectors. It can be shown that this corresponds to an upper bound (and hence if an alternative vector is chosen that gives a eigenvalue lower than the previous one, this is necessarily a better estimate). This approach will be generalized in the next section, but suffice to say here that if the chosen vectors u satisfy certain boundary conditions and resemble the actual mode shapes, then estimates of the lowest natural frequency (usually the most important) are often reasonably accurate. A few other issues to consider before moving on to continuous systems are worth mentioning at this point. The definitions of stability introduced here are largely based (or at least specialized) for conservative systems (with the additon of a little damping) and the question of stability of an equilibrium position. Toward the end of this book we will focus on forced vibration problems, in which the stability of periodic behavior will need to be considered. In this

17:51

P1: KAE Chapter˙04

CUFX159-Virgin

0 521 85648 5

August 14, 2007

4.3 Distributed Systems

57

case we will need to extend the concepts of stability to include orbital stability. Another aspect of the general eigenvalue problem is how the eigenvalues change as a system parameter changes. Of course, in the types of problems encountered in this book we will often have the case that the stiffness matrix degrades in some sense, but there are a number of instances (including in control theory) for which a minor perturbation of the eigenvalue problem is a key issue. Finally, computational aspects of the algebraic eigenvalue problem are cornerstones of efficient algorithms in software applications [17]. This will also be touched on later.

4.3 Distributed Systems We have gone from SDOF to MDOF systems and now this is extended further to continuous systems, which can be thought of as having an infinite number of DOFs. Much of the preceding theory can be extended to these distributed systems, and indeed many methods have been developed to reduce the number of DOFs [2]. We start this section by returning to Hamilton’s principle in order to derive Lagrange’s equations and hence the equations of motion. In contrast to finitedimensional systems we will see that this spatially continuous analysis results in a boundary-value problem characterized by a partial differential equation with appropriate boundary conditions. We focus our attention on beamlike structures characterized by deflection w(x, t) and signify the boundary conditions at 0 and L over the 1D domain x. Hamilton’s principle for conservative systems can be written (following Meirovitch [2]) as

t2

δLdt = 0,

δw(t = t1 ) = δw(t = t2 ) = 0,

(4.60)

t1

where L = T − V, and

L

T=

ˆ w) T( ˙ dx,

(4.61)

ˆ V(w, w , w ) dx.

(4.62)

0 L

V= 0

The hats on Tˆ and Vˆ imply an energy density, an overdot is a time derivative, a prime is a spatial derivative, and we assume that the potential energy in this case may be a function of the second derivative of the displacement w, which will be the case for the strain energy of systems in bending, for example. We need to take the variation of the Lagrangian in Eq. (4.60) and to do this (in terms of a Lagrangian density) we have ∂Lˆ ∂Lˆ ∂Lˆ ∂Lˆ δLˆ = δw + δw + δw + δw, ˙ ∂w ∂w ∂w ∂w˙

(4.63)

17:51

P1: KAE Chapter˙04

CUFX159-Virgin

0 521 85648 5

August 14, 2007

58

Higher-Order Systems

and thus, evaluating the integrals and placing in Eq. (4.60) we get

∂Lˆ ∂2 ∂Lˆ ∂ ∂Lˆ ∂ ∂Lˆ + − δwdx − ∂w ∂x ∂w ∂x2 ∂w ∂t ∂w˙ 0 t1 ˆ ∂L ∂ ∂Lˆ δw(0, t) − − ∂w ∂x ∂w x=0 ˆ ∂L ∂ ∂Lˆ δw(L, t) + − ∂w ∂x ∂w x=L ∂Lˆ ∂Lˆ − δw (0, t) + δw (L, t) dt = 0. ∂w x=0 ∂w x=L t2

L

We next choose the virtual displacements such that the variations at (0, t) and (L, t) are zero and then we are left with ∂Lˆ ∂2 ∂Lˆ ∂ ∂Lˆ ∂ ∂Lˆ + − − = 0. ∂w ∂x ∂w ∂x2 ∂w ∂t ∂w˙

(4.64)

This is called the Lagrange differential equation of motion. It is a partial differential equation with possible boundary conditions at x = 0, ˆ ∂L ∂ ∂Lˆ − − = 0, ∂w ∂x ∂w x=0 w(0, t) = 0, ∂Lˆ − = 0, ∂w x=0 w (0, t) = 0,

(4.65) (4.66) (4.67) (4.68)

and at x = L,

∂Lˆ ∂ ∂Lˆ − = 0, ∂w ∂x ∂w x=L w(L, t) = 0, ∂Lˆ = 0, ∂w x=L w (L, t) = 0.

(4.69) (4.70) (4.71) (4.72)

The system must satisfy two boundary conditions at each end (assuming the Lagrangian depends on the second spatial derivative of the displacement). Given homogeneous boundary conditions we can simplify Eq. (4.64) to the form of Eq. (2.36). Specific examples will be given in later chapters in which this type of boundary-value problem is solved by use of a variety of techniques.

17:51

P1: KAE Chapter˙04

CUFX159-Virgin

0 521 85648 5

August 14, 2007

4.3 Distributed Systems

59

4.3.1 The Differential Eigenvalue Problem It will often be the case that the partial differential of motion can be subject to the separation of variables (temporal and spatial), that is, by assuming the motion is a sum of terms of the form w(x, t) = F (t)W(x).

(4.73)

This allows for the consideration of ordinary differential equations. The temporal part of the solution will typically consist of harmonic motion (certainly for stable conservative systems) satisfying the initial conditions (see Chapter 2). The spatial part of the solution, together with the appropriate boundary conditions, constitutes a differential eigenvalue problem. The actual form of the partial differential equation depends of course on the specific physics of the problem (e.g., we have already developed the case typically encountered for Euler–Bernoulli beams) including the boundary conditions. However, there are a few general conclusions that we can draw by using the concept of differential operators [18]. The general form is given by 2 ∂ w(x, t) K[w(x, t)] + M = 0, (4.74) ∂t2 where K and M are (linear) homogeneous differential operators defined (for example) by K = a1 + a2

∂ ∂ ∂2 ∂2 + a3 + a4 2 + a5 + ··· + . ∂x ∂y ∂x ∂x∂y

(4.75)

M is very often a simple constant, and the spatial vector x is defined over a domain D. The form of the operator (i.e., the values of the coefficients) depends on the specific physical problem, for example, the general preceding form is appropriate for two-dimensional (2D) problems, including plates. Assuming harmonic motion we obtain the eigenvalue problem K[W(x)] = λM[W(x)],

(4.76)

where we recognize λ = ω2 , and the boundary conditions Bµ [W(x)] = 0,

(4.77)

where Bµ is also a linear differential operator. We expect the trivial solution W(x) = 0 to be present, but it is the values of the parameter λ (associated with the natural frequencies of the system) that gives rise to nontrivial solutions [W(x) = 0 and satisfying the boundary conditions] that are of central importance. There are infinitely many of these values, and they are the eigenvalues of the system and the corresponding W(x) are eigenfunctions. We observe certain similarities between the algebraic eigenvalue problem and the differential eigenvalue problem with the operators K and M having their relation with the stiffness and mass matrices of MDOF systems.

17:51

P1: KAE Chapter˙04

CUFX159-Virgin

0 521 85648 5

August 14, 2007

60

Higher-Order Systems

The boundary conditions did not appear explicitly in the algebraic eigenvalue problem for MDOF (matrix) systems. Essentially, they were contained in the matrices themselves, but for continuous systems they are a very important part of the problem and influence the functions W(x). Eigenfunctions satisfy both the differential equation and the boundary conditions exactly. Comparison (test) and admissible functions satisfy certain of the boundary conditions, that is, both geometric and natural, and geometric, respectively. They also form the basis of a number of solution methods to be considered later. Differential operators may also be selfadjoint (typically for linear conservative systems), which is analogous to symmetric matrices in the algebraic eigenvalue problem [18]. Similarly the eigenvalue problem is said to be positive-definite if all its eigenvalues are greater than zero and the eigenfunctions exhibit orthogonality (to be discussed shortly).

4.3.2 Solution Methods Modal Analysis and Truncation

For some differential eigenvalue problems, a closed-form solution may be available, and this will often be the case for linear systems with relatively simple geometry. The expansion theorem [2] allows the solution of a boundary-value problem by transforming it into an infinite set of ordinary differential equations—modal equations. Again we focus attention on self-adjoint systems, as well as on systems in which the eigenvalues do not depend on the boundary conditions. Earlier, we saw how it was possible to decouple the equations of motion for a MDOF system by transforming the problem into modal coordinates. This idea carries through to infinite-dimensional systems [19]. Using operator notation, we typically have the system 2 ∂w(x, t) ∂ w(x, t) K[w(x, t)] + C = 0. (4.78) +M ∂t ∂t2 Assuming harmonic motion and without damping, we get the eigenvalue problem K[W(x)] = ω2 M[W(x)],

(4.79)

with the boundary conditions Bµ [W(x)] = 0,

µ = 1, 2, . . . , p,

(4.80)

that is, assuming that the boundary conditions are homogeneous (every term involves W) and that the differential expression is of the order of 2p [2]. Let us consider two distinct solutions of the eigenvalue problem: (λr , wr ) and (λs , ws ). Because of the self-adjoint nature of the system, wr Mws dD = ws Mwr dD, (4.81) D

D

wr Kws dD = D

ws Kwr dD D

(4.82)

17:51

P1: KAE Chapter˙04

CUFX159-Virgin

0 521 85648 5

August 14, 2007

4.3 Distributed Systems

61

we have the conditions of orthogonality of the eigenfunctions: wr Mws dD = 0 for ω2r = ω2s ,

(4.83)

D

D

wr Kws dD = 0 for ω2r = ω2s .

(4.84)

The magnitude of the eigenvectors is arbitrary, and thus it makes sense to normalize them in a consistent way. This can be achieved with wr Mws dD = δrs , r, s = 1, 2, . . . , (4.85) D

D

wr Kws dD = δrs ω2r , r, s = 1, 2, . . . ,

(4.86)

where δrs is the Kronecker delta. The expansion theorem [20] tells us that the system response is given by a linear combination of the eigenfunctions: w(x, t) =

∞

ar (t)wr (x).

(4.87)

r=1

We furthermore assume that proportional damping is now included, that is, wr Cws dD = δrs 2ζr ωr , r, s = 1, 2, . . . , (4.88) D

where ζr contains the modal damping ratios. Placing Eq. (4.87) into Eq. (4.78) leads to

∞

∞ ∞ ∂ ∂2 K wr (x)ar (t) + C wr (x)ar (t) + M wr (x)ar (t) = 0. ∂t ∂t2 r=1 r=1 r=1 (4.89) Multiplying by wr (x) and integrating over the domain results in ∞ r=1

ar (t)δrs ω2r +

∞

a˙ r (t)δrs 2ζr ωr +

r=1

∞

a¨ r (t)δrs = 0.

(4.90)

r=1

But orthogonality tells us this equation holds only if r = s, and thus a¨ r (t) + 2ζr ωr a˙ r (t) + ω2r ar (t) = 0,

r = 1, 2, . . . .

(4.91)

Thus we arrive at an infinite set of (uncoupled) ordinary differential equations in terms of normal modes, assuming the diagonal form exists. This form is somewhat familiar from the consideration of principal coordinates in MDOF systems. Because, in practical situations, it is only the lowest few modes that are important (the higher modes typically do not contribute very much to the solution) it may be possible to use a truncated modal model. The following sections introduce a couple of popular (approximate) approaches to solving boundary-value problems that are developed from the concept of basis functions.

17:51

P1: KAE Chapter˙04

CUFX159-Virgin

0 521 85648 5

August 14, 2007

62

Higher-Order Systems Rayleigh–Ritz

For continuous systems we can write Rayleigh’s quotient (see Subsection 4.2.6) as w(x)K[w(x)]dD 2 , (4.92) ω = R[w(x)] = D D w(x)M[w(x)]dD which is stationary in the class of kinematically admissible functions, at the eigenvalues. This is a very useful property for estimating especially the lowest eigenvalue of a continuous system [21]. The Rayleigh–Ritz method consists of replacing the differential eigenvalue problem with a set of algebraic eigenvalue problems and is based on assuming a solution of the form w(x) =

N

an (t)θn (x).

(4.93)

n=1

If the functions θn (x) satisfy both the differential equation and the boundary conditions exactly, then we would have the eigenvalues, but in general we will be faced with choosing these as trial functions. Clearly, a trial function that satisfies all the boundary conditions (both geometric, involving displacements/slopes, and natural, involving forces/moments) is desirable. These are generally called comparison or test functions. However, in general it will be easier to come up with trial functions that satisfy only the geometric boundary conditions. These are typically called admissible functions and tend to result in quite accurate estimates for the eigenvalues, especially as more are taken in the assumed solution, and because the approach yields an upper bound, we know that the solution will tend to approach the exact answer from above. This is very similar to the case for MDOF systems whereas now we seek trial functions rather than trial vectors. We also note the relation with Rayleigh’s energy method used in the previous chapter for a SDOF system. Rayleigh’s principle states that the Rayleigh quotient has a minimum value, which is the square of the lowest frequency of vibration, in the neighborhood of the fundamental mode: ω21 = min R(w) = min

Vmax , Tref

(4.94)

where Tref is the reference kinetic energy (referred to as the zero-frequency kinetic energy in Chapter 3). The Rayleigh–Ritz method is also closely related to the method of assumed modes [1]. Weighted Residuals–Galerkin

An alternative approach to obtaining approximate solutions to distributed parameter systems is based on the method of weighted residuals, the best-known technique being the Galerkin method [22]. This approach is not restricted to self-adjoint systems and generally requires the use of comparison functions rather than admissible functions, but when used for self-adjoint systems is equivalent to the Rayleigh–Ritz method. The central idea is that the error between an approximate solution and the exact solution is minimized.

17:51

P1: KAE Chapter˙04

CUFX159-Virgin

0 521 85648 5

August 14, 2007

4.3 Distributed Systems

63

Given a linear differential operator L, we assume the solution to the differential equation L[w(x)] = 0 (with homogeneous boundary conditions) is based on w (x, t) = (r)

N

ar (t)φr (x),

(4.95)

r=1

where the functions φr (x) are linearly independent, form a complete set, satisfy the boundary conditions, and have unknown coefficients ar (t). For an infinite set of functions, we get the exact solution, with the requirement that the function L[w(r) (x, t)] be orthogonal to all the functions φr (x). However, because we have N functions rather than an infinite set, we know that Eq. (4.95) will not typically satisfy the differential equation exactly, and we will get a remainder (or residual) that is orthogonal to the space of the chosen functions: L w(r) (x, t) φs (x)dD = 0, r, s = 1, 2, . . . , N. (4.96) D

Thus, suppose we have again a typical (self-adjoint with homogeneous boundary conditions) system given by Eq. (4.78). Placing Eq. (4.95) in Eq. (4.78), multiplying by φs (x), and integrating over the whole domain, we get Ma(t) ¨ + C˙a(t) + Ka(t) = 0, where M is an N × N mass matrix of coefficients msr given by msr = φs (x)M[φr (x)]dD, r, s = 1, 2, . . . , N,

(4.97)

(4.98)

D

and where the stiffness K and (proportionally damped) C can be described similarly. As mentioned previously, the Rayleigh–Ritz and Galerkin methods produce identical results for conservative systems. Other techniques based on weighted residuals include collocation and least squares. Furthermore, a more sophisticated approach based on Lyapunov-Schmidt reduction can be used as an alternative, especially in certain situations [23]. Specific techniques have also been developed for bifurcation and nonlinear eigenvalue problems [24]. The Finite-Element Method

Perhaps the technique with the greatest utility for solving continuous structural systems is the finite-element (FE) method (FEM) [25–29]. This is a huge subject, and obviously only a brief introduction is given here. In fact, FE analysis (FEA) will be outlined in more detail for specific structural forms later in this book (e.g., plane frames and plates). It is especially powerful for complicated structures and forms the basis of a vast array of commercially available software. Here, a brief introduction is given within the context of general boundary-value problems. This technique really comes into its own for geometrically complex structures, because (unlike the Rayleigh–Ritz method) trial functions (often relatively loworder polynomials) are defined over subdomains of a structure in a process of

17:51

P1: KAE Chapter˙04

CUFX159-Virgin

0 521 85648 5

64

August 14, 2007

Higher-Order Systems

discretization. Certain continuity conditions are enforced between subdomains at nodes, and a process of assembly brings all the elements together to form stiffness and mass matrices in standard vibration problems. From this point, the lumpedparameter model of the structure can be handled as a MDOF system, albeit with typically very high-dimensional matrices.

4.3.3 Context Revisited Before we leave this chapter and embark on the study of slender, axially loaded structures, a few comments are in order about more specific issues covered in the rest of this book. We are primarily interested in structural systems characterized according to the following scheme. r Strain energy. This a quantity associated with structural deformation. In linear theory, this has a quadratic form (small elastic deformations) and leads to a symmetric stiffness matrix, or stiffness operator. For zero deformation we typically have the trivial equilibrium state. For large deflections, both stretching (membrane) and bending effects contribute to strain energy. r Work done by axial loads. External loads are often imparted to a slender structure axially, as well as laterally, and may be due to dead (gravity) loading, for example. They generally lead to a geometric stiffness matrix that tends to diminish lateral (bending) stiffness as a function of the loading. If compressive, then they may lead to instability (the stiffness matrix becomes singular) and the appearance of nontrivial equilibria. If time-periodic axial loads are present, then parametric resonance is possible. r Kinetic energy. Again this is typically a quadratic form, but does not affect equilibrium. It may be imparted to a system in the form of sudden loading. Vibration can be considered as an oscillatory exhange of kinetic and potential energy. r Damping. Small levels of damping can be modeled by Rayleigh’s dissipation function. Also, it does not affect equilibrium. Often it turns the stability of conservative systems into asymptotically stable systems. Damping is always present in mechanical systems, although its effect can sometimes be neglected. r External forcing. Many structures are subject to excitation, which is often periodic. The resulting problem of resonance provides an important motivation for study. Loading is sometimes suddenly applied and may cause instability in the large, even though the local equilibrium is stable. External forcing increases the phase space because governing equations become inhomogenous (nonautonomous) and may lead to a wider spectrum of behavior, especially for nonlinear systems.

References [1] [2]

D.J. Inman. Engineering Vibration. Prentice Hall, 2000. L. Meirovitch. Principles and Techniques of Vibrations. Prentice Hall, 1997.

17:51

P1: KAE Chapter˙04

CUFX159-Virgin

0 521 85648 5

August 14, 2007

References [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20] [21] [22] [23] [24]

[25] [26] [27] [28] [29]

Y.G. Panovko and I.I. Gubanova. Stability and Oscillations of Elastic Systems – Paradoxes, Fallacies and New Concepts. Consultants Bureau, New York, 1965. A.M. Lyapunov. The General Problem of the Stability of Motion. Princeton University Press, 1947. A.P. Seyranian and A.A. Mailybaev. Multiparameter Stability Theory with Mechanical Applications. World Scientific, 2003. K. Huseyin. Nonlinear Theory of Elastic Stability. Noordhoff, 1975. H.H.E. Leipholz. Stability Theory. Wiley, 1987. S.H. Strogatz. Nonlinear Dynamics and Chaos. Addison-Wesley, 1994. E.H. Dowell. Aeroelasticity of Plates and Shells. Noordhoff, 1975. MATLAB. User’s guide. Technical report, The Math Works, 1989. C. Hayashi. Nonlinear Oscillations in Physical Systems. Princeton University Press, 1964. J. LaSalle and S. Lefschetz. Stability by Liapunov’s Direct Method with Applications. Academic, 1961. A.M. Lyapunov. Stability of Motion (Collected Papers). Academic, 1966. K. Ogata. System Dynamics. Prentice Hall, 1998. Lord Rayleigh (John William Strutt). The Theory of Sound. Dover, 1945. B.H. Tongue. Principles of Vibration. Oxford University Press, 1996. F.W. Williams and W.H. Wittrick. Exact buckling and frequency calculations surveyed. Journal of Structural Engineering, 109:169–87, 1983. R. Courant and D. Hilbert. Methods of Mathematical Physics. Wiley Classics Library, 1989. E.H. Dowell and D.M. Tang. Dynamics of Very High Dimensional Systems. World Scientific, 2003. J.H. Argyris and S. Kelsey. Energy Theorems and Structural Analysis. Butterworth, 1960. G. Temple and W.G. Bickley. Rayleigh’s Principle and Its Applications to Engineering. Oxford University Press, 1933. W.J. Duncan. Galerkin’s method in mechanics and differential equations. Reports and Memoranda No. 1798, Aeronautical Research Council London (England), 1937. H. Troger and A. Steindl. Nonlinear Stability and Bifurcation Theory: An Introduction for Engineers and Applied Scientists. Springer-Verlag, 1991. H.B. Keller. Numerical solution of bifurcation and nonlinear eigenvalue problems. In P. Rabinowitz, editor, Applications of Bifurcation Theory. Academic Press, 1977, pp. 359–89. T.J.R. Hughes. Finite Element Method-Linear Static and Dynamic Finite Element Analysis. Prentice Hall, 2000. O.C. Zienkiewicz and R.L. Taylor. The Finite Element Method. McGraw Hill, 1989. K.J. Bathe. Finite Element Procedures. Prentice Hall, 1995. M.A. Crisfield. Nonlinear Finite Element Analysis of Solids and Structures, Vol. 1: Essentials. Wiley, 1997. M.A. Crisfield. Nonlinear Finite Element Analysis of Solids and Structures, Vol. 2: Advanced Topics. Wiley, 1997.

65

17:51

P1: KAE Chapter-05

CUFX159-Virgin

0 521 85648 5

August 14, 2007

5 Discrete-Link Models

5.1 Introduction We first consider a number of discrete-link models in which system properties are concentrated at specific locations. The motivation for considering simple mechanical models is that most of the concepts of dynamics and stability issues encountered with continuous systems (e.g., beams, plates) can be observed with discrete systems but are somewhat easier to analyze [1–7]. In fact the governing equations will tend to be algebraic rather than differential (at least in space), and it is natural to start with a look at systems in which the behavior of the system is completely described by just a single degree of freedom.

5.2 An Inverted Pendulum Consider the simple hinged cantilever illustrated in Fig. 5.1. This system consists of a concentrated mass supported by a massless but rigid bar of length L. A torsional spring supplies a linear restoring force that is proportional to the angle of rotation of the hinge (in either direction), with spring coefficient K. The angle of rotation θ thus describes the location of the mass at any given instant of time. Typically, the vertical force is simply P = Mg, but here we assume that an axial load of magnitude P acts independently of the constant mass M. This assumption will be reexamined later. For a simple system like this, we can easily use any of the fundamental approaches introduced in Chapter 2 for writing the governing equation of motion. Approaching this problem by using Lagrange’s equation, we can write the total potential energy of the system V as consisting of two parts: U, the strain energy stored in the spring as it deflects, and VP , the potential energy associated with the work done by the axial load as the mass moves through a vertical distance. Thus V = U + VP =

1 2 Kθ − PL(1 − cos θ). 2

(5.1)

Similarly, the kinetic energy T is given by T=

1 ML2 θ˙ 2 , 2

(5.2)

and placing these into Lagrange’s equations, we obtain ML2 θ¨ + Kθ − PL sin θ = 0, 66

(5.3)

17:53

P1: KAE Chapter-05

CUFX159-Virgin

0 521 85648 5

August 14, 2007

5.2 An Inverted Pendulum

67

P

M

L Figure 5.1. A simple hinged bar (the inverted pendulum model).

K

and using sin θ = θ for small θ we see an effective natural frequency ω2 = K/(ML2 ) − P/(ML). Equation (5.3) can be nondimensionalized by assuming ω2n = K/ML2 ,

p = PL/K,

(5.4)

to give θ¨ + ω2n (θ − p sin θ) = 0.

(5.5)

Equation (5.5) is a nonlinear second-order, homogeneous, ordinary differential equation with constant coefficients.

5.2.1 Static Behavior Let’s consider the underlying equilibrium behavior, which we can easily obtain by setting the time-dependent terms equal to zero: ω2n (θ − p sin θ) = 0.

(5.6)

This is the first variation of the potential energy and could have been obtained from a direct application of the principle of stationary potential energy. Clearly, we have the trivial (or fundamental) equilibrium state for the perfectly upright position θ = 0. However, we see that another (postbuckled) solution to Eq. (5.6) is (for p > 1) p=

θ . sin θ

(5.7)

Equation (5.7) is plotted together with the trivial solution in Fig. 5.2(a). These paths intersect at p = 1, that is, P = K/L. This is the critical load of the structure at which point the trivial equilibrium position gives way to an inclined position. We establish

17:53

P1: KAE Chapter-05

CUFX159-Virgin

0 521 85648 5

August 14, 2007

68

Discrete-Link Models

1.6

1.6

p

p 1.4

1.4

1.2

1.2

1

1

0.8

0.8

0.6

0.6

0.4

0.4

0.2

p

1.0

0.2

(a) 0 -1.5

(b) -1

-0.5

0

0.5

1

θ

1.5

0

0

0.2

(c) 0.4

0.6

0.8

ε

1

θ 0.0

Figure 5.2. (a) The inverted pendulum and (b) equilibrium paths; (c) potential-energy contours (truncated for extreme levels).

that the trivial equilibrium path is unstable for loads greater than p = 1 by examining the curvature of the total potential energy in the vicinity of equilibrium. For stability it is sufficient that d 2V > 0, dθ 2

(5.8)

and thus for θ = 0 the preceding condition is satisfied only if p < 1. To test the stability of the secondary (postbuckled) solution, we evaluate Eq. (5.8) for the system along the secondary path described by Eq. (5.7), which results in d 2V = 1 − p cos θ. dθ 2

(5.9)

This is clearly positive (and indicative of stability) provided that p

1). These are termed complementary and would not normally be observed in a natural loading history. 1 .6

p

p

1 .4 1 .2

1.0

1 0 .8 0 .6

0

0 .4

0

0 .2

0

(a) 0 -1.5

-1

-0.5

0

0.5

1

= 0.1 = 0.2 = 0.3

(b) 1.5

0.0

Figure 5.3. (a) The equilibrium paths and (b) potential energy of the imperfect inverted pendulum model (truncated for extreme levels).

17:53

P1: KAE Chapter-05

CUFX159-Virgin

0 521 85648 5

August 14, 2007

70

Discrete-Link Models

The stability of the imperfect equilibrium paths is still obtained from the second derivative of the total potential energy: d 2V = 1 − p cos θ, dθ2

(5.15)

which, when evaluated on the primary (i.e., from zero-load) equilibrium curve, indicates stability [because p < 1/(cos θe )]. For the complementary path, part of the curve is stable and part unstable. In Fig. 5.3(a), they are identified by different dashed curves for different initial imperfections. Their stability of course can be determined from the potential energy, but stability considerations are deferred to the dynamic criterion discussed in the next subsection. Also plotted in this figure [part (b)] is a contour plot of the total potential energy for the specific case of θ0 = 0.1. The darker shades correspond to lower levels of potential energy for a given loading. Following the minimum as a function of load would thus lead to gradually increasing deflection θ in the positive direction, that is, the same direction as the initial deflection. We also see how the complementary path is represented by a (remote) minimum beyond a local maximum.

5.2.3 Dynamic Behavior We now turn to the dynamic response of the inverted pendulum model. We know that the trivial equilibrium solution is stable provided the magnitude of the axial load is less than its critical value, and thus we would expect small oscillations about the origin given some initial disturbance from a stable equilibrium state. Expanding the sine term in Eq. (5.5) about zero and dropping higher-order terms leads to θ¨ + ω2n (1 − p)θ = 0.

(5.16)

Here, we are assuming that although the mass provides the axial load the inertia is independent. We shall lift this restriction a little later. The linearized (effective) natural frequency drops toward zero as the critical load is approached, ωn → 0,

p → 1,

(5.17)

and, with harmonic motion θ(t) = c sin ωt, there is a linear relationship between the applied load and the square of the natural frequency ω2 = ω2n (1 − p), where p = PL/K and ω2n = K/ML2 . We can thus observe this decay if we plot a time series in which the load is made a slowly increasing function of time in exactly the same way that was considered in Chapter 3. We would expect to have oscillatory behavior about the stable postbuckled paths. We return to Eq. (5.5) and expand about a general equilibrium path θ = θe + δ.

(5.18)

δ¨ + ω2n [(θe + δ) − p sin(θe + δ)] = 0.

(5.19)

Placing this in Eq. (5.5) we have

17:53

P1: KAE Chapter-05

CUFX159-Virgin

0 521 85648 5

August 14, 2007

5.2 An Inverted Pendulum

71

Assuming δ is small such that cos δ ≈ 1 and sin δ ≈ δ, and because θe − p sin θe = 0, we are then left with the linearized (variational) equation of motion: δ¨ + ω2n [1 − p cos θe ]δ = 0.

(5.20)

For example, suppose p = 1.1; then we have θe = ±0.75 rad, which corresponds to ω2 = 0.35. Hence we obtain the result that although the natural frequency drops to zero as the critical load is approached, it then starts to increase as the load is increased into the (stiffening) postbuckled range. We shall return to this type of behavior in the next section. If we are not to be restricted to relatively small-amplitude oscillations about equilibrium then we must solve the nonlinear equation of motion, and this can be easily accomplished numerically [9]. We can also relate time and stiffness linearly— this provides a useful way to visualize the dynamics of the system in the vicinity of equilibrium while the load is slowly increased (as done in the introduction). Such a scheme can be incorporated numerically by θ¨ + θ − 0.01τ sin θ = 0,

(5.21)

where the critical load is reached after 100 time units have elapsed. We have also scaled the time according to τ = ωn t,

(5.22)

and hence the overdots in Eq. (5.21) signify θ˙ ≡ d θ/d τ. Initial conditions were chosen as θ(0) = 0.1, θ˙ (0) = 0.0 to start not too far away from the equilibrium, and the result of a numerical simulation is shown in Fig. 5.4. Another conceptual aid in understanding this dynamic behavior is to imagine a small ball rolling on the potential-energy surface given by Eq. (5.1). For small

p (t)

2 1.5 1 0.5

0.0 25

50

75

100

125

150

175

Figure 5.4. A time series evolving through increasing axial load.

17:53

P1: KAE Chapter-05

CUFX159-Virgin

0 521 85648 5

August 14, 2007

72

Discrete-Link Models

0

2

V -0.2

1

-0.4 0

0 0.5

-1 1 1.5 -2

p

Figure 5.5. Potential energy surface as a function of axial load.

oscillations (for which the linearized equation of motion is appropriate), we see how the evolution of the ball motion in this slowly evolving environment follows the local minima of the underlying potential energy function (see Fig. 5.5). The direction (i.e., positive or negative) followed after criticality is quite arbitrary and rotations to the left, that is, for negative θ, are just as likely for other initial conditions. In this figure, the potential-energy contours are cropped for large positive values to aid the view. For a conservative system such as this, we can also gain insight from plotting contours of constant total energy. Two such plots are shown in Fig. 5.6 for p = 0.8 in part (a) and p = 1.2 in part (b), with again the darker shades indicating lower levels of total potential energy. When p = 1.2 we have symmetric stable equilibria at θe = ±1.027. The origin is of course a saddle point at this level of load. Initial conditions close to the stable locations would result in approximately linear vibrations according to the solutions of Eq. (5.20). For initial conditions far away we might (a)

.

0.4

(b)

.

0.4

θ

θ 0.2

0.2

0

0

-0.2

-0.2

-0.4

-0.4

-1.5

-1

-0.5

0

0.5

1

θ

1.5

-1.5

-1

-0.5

Figure 5.6. Contours of total energy: (a) p = 0.8 and (b) p = 1.2.

0

0.5

1

θ

1.5

17:53

P1: KAE Chapter-05

CUFX159-Virgin

0 521 85648 5

August 14, 2007

5.2 An Inverted Pendulum

θ

73

.

(a)

1.5 1

0.2

0.5

0.1 20

40

60

80

100

τ -1.5

-1

-0.5

0.5

-0.5

-0.1

-1

-0.2

-1.5

-0.3

θ

(b)

0.3

1.4

(c)

1.2

1

1.5

.

(d)

0.3 0.2

1

0.1

0.8 0.6

0.2

0.4

-0.1

0.2

-0.2

20

40

60

80

100

τ

0.4

0.6

0.8

1

1.2

-0.3

Figure 5.7. Typical large-amplitude oscillations. Both cases correspond to p = 1.2: (a) and (b) θ(0) = 1.5, θ˙ (0) = 0.0, (c) and (d) θ(0) = 0.01, θ˙ (0) = 0.0.

expect large-amplitude oscillations that traverse the origin (slowing down when passing over the potential ridge). An example of such a periodic (but not simple harmonic) oscillation is shown in Fig. 5.7(a). Indeed, for moderately large amplitudes we would find markedly asymmetric oscillations without a traverse of the upright position. A limited amount of analytic progress can be made in such a situation if we retain higher-order terms in the Taylor series expansion about equilibrium. An example of this type of motion (which is very close to the homoclinic solution starting and finishing at the saddle) is shown in Fig. 5.7(c). The conservation of energy can be obtained directly from Eqs. (5.1) and (5.2) or we can alternatively use the simple relation θ¨ = θ˙

dθ˙ dθ

(5.23)

in Eq. (5.5) to separate variables and obtain the velocity as a function of position: θ˙ = ± C − θ2 − 2p cos θ,

(5.24)

where the constant C depends on the initial conditions. For example, in Fig. 5.7(b) we can use the initial conditions to derive a constant C = 2.42, and, using Eq. (5.24), we can confirm, for example, that as the bar passes through its upright position it will be doing so at a nondimensional velocity of 0.141 in either direction. The phase portrait (a plot of velocity versus position) can thus be viewed as a trajectory

1.4

17:53

P1: KAE Chapter-05

CUFX159-Virgin

0 521 85648 5

August 14, 2007

74

Discrete-Link Models

following one of these contours of constant total energy. The trajectories corresponding to the two preceding examples are also included in Fig. 5.7. Before leaving this section it is briefly shown how the (nonlinear) natural frequency of large-amplitude vibration can be extracted from the phase trajectories. We can separate variables in Eq. (5.5) and integrate to obtain t θmax dθ dt = , (5.25) 2 C − θ − 2p cos θ 0 0 where the period is equal to four times the time it takes to go from 0 to θmax : θmax −1/2 τ=4 C − θ2 − 2p cos θ d θ. (5.26) 0

Thus taking the initial conditions (and the value of C) corresponding to the trajectory in Fig. 5.7 we can evaluate the preceding integral (numerically) to confirm the (near homoclinic) period of 29.34 units. We shall return to this type of (elliptic integral) approach when we deal with continuous (elastica) systems. 5.2.4 A Note on Inertia Before moving on it is useful to mention here that in a realistic situation we might expect to increase the axial load in the inverted pendulum model by increasing the mass, that is, P = Mg. However, suppose the pendulum arm has some mass m associated with it (that does not change and is not sufficient to cause self-weight buckling). In this case, the energy expressions [from Eqs. (5.1) and (5.2)] change to V=

1 2 1 Kθ − MgL(1 − cos θ) − mgL(1 − cos θ) 2 2

(5.27)

1 1 ML2 θ˙ 2 + mL2 θ˙ 2 . 2 6

(5.28)

and T=

Thus the equation of motion becomes (M + m/3)L2 θ¨ + Kθ − (M + m/2)gL sin θ = 0.

(5.29)

It is a simple matter to linearize this equation to obtain an expression for the natural frequency (of small-amplitude oscillations) ω2 =

K − (M + m/2)gL , (M + m/3)L2

(5.30)

which vanishes when the end mass achieves its critical value, Mcr =

m K − . gL 2

(5.31)

The natural frequency and critical mass now provide a suitable means of nondimensionalizing, 2 = ω2 L/g,

k = K/(mgL),

p = M/m,

(5.32)

17:53

P1: KAE Chapter-05

CUFX159-Virgin

0 521 85648 5

August 14, 2007

5.3 A Discrete-Strut Model

75

0.5

(a)

t

p 0.4

0.0 20

0.3

40

60

80

100

-0.5

0.2 -1 0.1

p (t)

-1.5 0.0

0.2

0.4

0.6

0.8

1

1.2

1.4 2

Figure 5.8. (a) The mass–frequency relation for the inverted pendulum model and (b) a numerical time series resulting from a linear increase in mass with time.

which can then be used in the relation between the natural frequency and end load: ω2 =

3(2k − 1 − 2p) . 2(1 + 3p)

(5.33)

Given the nondimensionalization, the critical load is now p cr = k − 0.5, and a plot of mass (load) versus natural frequency squared is shown in Fig. 5.8(a) for k = 1. We can evolve the magnitude of the end mass as a linear function according to M = 0.01t, which gives the time series shown in Fig. 5.8(b). The critical mass is reached after 50 time units, after which the (undamped) system starts to oscillate as it follows one of its (nontrivial) postbuckled paths.

5.3 A Discrete-Strut Model We now move on to consider another rigid-link model, but this time the model is a little more general in the types of behavior it can exhibit. It is also a step closer to the continuous structures we will focus on later. The approach adopted in this section is based on energy considerations rather than the underlying differential equations, and some experimental data are also included. The model under consideration is shown in Fig. 5.9(a). A mechanical model was built to mimic this system by Croll and Walker at University College London [10]. A photographic image is shown in part (b) of Fig. 5.9. It consists of two rigid (massless) links hinged at their supports and in the center where a concentrated mass M is located. Two linear springs provide a restoring force, one against lateral deflection with modulus K, and the other against rotation with modulus C. An axial load of magnitude P acts at the left-hand ends that is unrestrained against horizontal movement. The coordinate Q describes the deflected position of the mass, and we can think of the schematic as providing a plan view; that is, we assume gravity is already taken into account. The total potential and kinetic energies for this system are given by [5] 1 V = 2Cθ2 + KL2 sin2 θ − 2PL(1 − cos θ) 2

(5.34)

(b)

17:53

P1: KAE Chapter-05

CUFX159-Virgin

0 521 85648 5

August 14, 2007

76

Discrete-Link Models P θ C L

L M

K

(a)

(b) Figure 5.9. (a) Schematic of the two-bar link model and (b) physical realization.

and T=

1 ML2 θ˙ 2 . 2

(5.35)

Equilibrium solutions are found from dV = 4Cθ + KL2 sin θ cos θ − 2PL sin θ = 0. dθ

(5.36)

We immediately see the trivial solution, θe = 0, together with the nontrivial solutions given by = α cos θ +

(1 − α)θ , sin θ

(5.37)

where = P/Pcr ,

(5.38)

Pcr =

KL + 4C , 2L

(5.39)

α=

KL2 . KL2 + 4C

(5.40)

2

The parameter α is a ratio of spring stiffnesses. For example, if α = 0 (i.e., K = 0), then we have exactly the same type of equilibrium curves as for the model in the previous section. The stability of the equilibrium paths can again be established from the sign of the second derivative of the total potential-energy function. However, let us assume initially that α = 1 (i.e., C = 0) so that we have only a lateral (translational) spring acting. The equilibrium expression simplifies to = cos θ.

(5.41)

In the presence of initial imperfections, the total potential energy becomes V=

1 KL2 (sin θ − sin θ0 )2 − 2PL(cos θ0 − cos θ) 2

(5.42)

17:53

P1: KAE Chapter-05

CUFX159-Virgin

0 521 85648 5

August 14, 2007

5.3 A Discrete-Strut Model 1.5

77 1.5

(a)

m

0

1

(b)

(c)

90

1

P(N)

60 0

0

0.5

0.5 < 0 0

0 −1

−0.5

0

0

30

0

> 0

0.5

0

0 −0.1

1

0.0 0

0.1 −0.8

−0.4

0

0.4

Figure 5.10. (a) Equilibrium paths for the two-bar link model with a lateral spring and initial geometric imperfections. θ0 ranges from 0.001 closest to the bifurcation to 0.1, (b) imperfection sensitivity and (c) measured data (adapted from [2]).

with the first derivative dV = KL2 (sin θ − sin θ0 ) cos θ − 2PL sin θ. dθ

(5.43)

Using the nondimensionalization, we obtain =

(sin θ − sin θ0 ) cos θ . sin θ

(5.44)

In the literature, it is sometimes observed that the trigonometric terms are replaced with their series expansion, and retaining the first few terms results in equilibrium paths [2]: =

θ − θ0 − 23 θ3 + 12 θ2 θ0 θ−

1 3 θ 6

≈1−

θ0 θ0 θ θ2 + − + ··· + . θ 3 2

(5.45)

These nontruncated [Eq. (5.44)], paths have the form shown in Fig. 5.10(a), in which we recognize the characteristic subcritical pitchfork bifurcation. Complementary paths are also present in this example. However, they prove to be unstable and have little to do with a natural loading path starting near the origin and hence are not included in the plot. For the perfect geometry, it is simple to see that the stability is governed by the coefficient of the second derivative of the potential energy. For the primary path, it is easy to show that equilibrium is unstable if p > 1. Similarly for the secondary path it can be shown that the coefficient is always negative and hence the postbuckled behavior is unstable (sometimes called unstable-symmetric branching behavior in the literature [4]). In practice this means that, if the load were gradually increased from zero, the strut model would buckle when p = 1 and the system would collapse completely (within the confines of the mathematical modeling). It should be mentioned here that experimental data taken from the system shown in Fig. 5.9(b) very closely match the theory shown in Fig. 5.10(a) [2] and is plotted in Fig. 5.10(c).

0.8

17:53

P1: KAE Chapter-05

CUFX159-Virgin

0 521 85648 5

August 14, 2007

78

Discrete-Link Models

For the imperfect geometry the limit of the stability of the paths coincides with the maximum load m (the horizontal tangency) for a given path. It can be shown 1/3 that this occurs when θ = θ0 , and placing this back into Eq. (5.45) results in the cusp geometry shown in Fig. 5.10(b). This displays an important characteristic of some axially loaded structures: the load-carrying capacity of the structure is reduced when initial imperfections are present, i.e., it is imperfection sensitive. This type of subcritical behavior may then be viewed as a potentially dangerous consequence when compared with the supercritical behavior described in the previous section. Following a similar line of reasoning to Section 3.3 we can show that the frequency of small oscillations can be obtained by using Rayleigh’s method: 2 = 1 − cos θ − α(1 − cos 2θ),

(5.46)

where ω is the effective natural frequency, = ω2n =

ω , ωn

(5.47)

4C + KL2 , ML2

(5.48)

and again setting α = 1, incorporating initial imperfection θ0 , and simplifying, we get θ2 2 = (1 − 2θ2 + θθ0 ) − 1 − . (5.49) 2 We can then plot Eq. (5.49), incorporating Eq. (5.45) for various initial imperfections, and this is shown in Fig. 5.11 for θ0 = 0.001, 0.01, and 0.1. Thus we see that the often linear relation between the natural frequency squared and the axial load 1.5 0 0 0

1

0.5

0

0

0.2

0.4

0.6

0.8

2

1

Figure 5.11. The relation between the natural frequency and axial load for the imperfect unstable symmetric model. θ0 = 0.001, 0.01, 0.1.

17:53

P1: KAE Chapter-05

CUFX159-Virgin

0 521 85648 5

August 14, 2007

5.3 A Discrete-Strut Model

79

p 0.5 20 0.1

40

τ 60

-0.2

V 0

2

-0.1 -0.2

1

0

θ

-0.4

0 0.5

(a)

-0.6

θ

-1

p

1

-0.8 1.5 -2

(b)

Figure 5.12. (a) The potential-energy surface with an initial imperfection of θ0 = −0.1 and (b) a typical time series evolving through the p axis, θ(0) = −0.2.

is not true for initially imperfect geometries, and in fact here the natural frequency drops to zero before the critical load is reached. The potential-energy surface is plotted as a function of load and deflection in Fig. 5.12(a) together with an evolving time series under linearly increasing end load in part (b). We see how the oscillation continues until the limit point is reached and the trajectory slides off to infinity (actually to large oscillations about π but we are not interested in such solutions). According to Fig. 5.10, we expect a maximum load of about p m ≈ 0.7 for this imperfection and hence given a ramped load of p = 0.01t we would expect the solution to lose stability after about τ = 70 time units. We observe the decay in natural frequency as a gradual lengthening of the period of oscillation as the instability is approached. Now, let’s assume that α = 0 (or K = 0), which in fact produces a situation qualitatively similar to the response of the inverted pendulum. In this case, the equilibria (for the initially perfect geometry) are given by θ = 0,

(5.50)

= θ/ sin θ

(5.51)

for θ = 0 and

for θ = 0, and a frequency–load relation 2 = 1 − cos θ,

(5.52)

which, for the prebuckled (trivial) equilibrium path gives 2 = 1 − .

(5.53)

80

17:53

P1: KAE Chapter-05

CUFX159-Virgin

0 521 85648 5

August 14, 2007

80

Discrete-Link Models

For the postbuckled (nontrivial) equilibrium path, we substitute Eq. (5.51) into Eq. (5.52) to get 2 = 1 − θ cot θ ≈

θ2 . 3

(5.54)

Expanding Eq. (5.51), =

θ2 θ ≈1+ , sin θ 6

(5.55)

and combining Eqs. (5.54) and (5.55) leads to 2 = 2( − 1).

(5.56)

Thus we see the interesting result that, as a function of axial load, the postbuckled linear frequency changes at half the rate of the prebuckled frequency, a result also contained in Eq. (5.20). Of course, this relation is true only for the moderately buckled structure because the trigonometric power series was truncated. It is also mentioned here that these types of link model can be used to illustrate the effect of thermal loading. For example, Croll and Walker [2] showed that an increase in temperature will result in stable-symmetric buckling (supercritical pitchfork bifurcation). It can also be shown that the natural frequency will decay in the usual way as the critical temperature is approached. Thermal buckling is an important consideration for plated structures and will be considered in more detail in a later chapter.

5.4 An Asymmetric Model The models described in the previous sections of this chapter were symmetric in the absence of initial imperfections. There are a number of structural systems that behave quite differently according to the direction of the deformation [1, 3, 11]. In a buckling context, this is characterized by the asymmetric (or transcritical) point of bifurcation: A structure exhibiting this behavior is shown in Fig. 5.13. The deflection of this SDOF system is described by the coordinate X, the horizontal distance of the top of the (massless) bar. It is convenient to nondimensionalize this by the length of the bar: x = X/L. A spring of modulus K provides the restoring force, with x0 denoting the initial imperfection (and thus x0 = X0 /L). However, the dead load (the point mass) is further offset by a fixed amount l (measured at right angles to the bar and set equal to αx0 ). The equilibrium paths are obtained from the first derivative of the total potential energy and are given by p=

2[(1 − x2 )1/2 − (1 − x)1/2 (1 + x0 )1/2 ] , [x + αx0 (1 − x2 )1/2 ]

(5.57)

where p = P/(KL). Some typical paths are shown in Fig. 5.14(a) for a variety of initial imperfections. It is clear that the behavior (including instability behavior) is strongly influened by the sign of the initial imperfection. The stability of these paths

17:53

P1: KAE Chapter-05

CUFX159-Virgin

0 521 85648 5

August 14, 2007

5.4 An Asymmetric Model

81 X X0 l

M P = Mg K L

L

Figure 5.13. Schematic of an asymmetric link model [11].

can be determined from the second derivative of the total potential energy. However, we can use the alternative criterion of requiring real frequencies for stability. These are obtained from Lagrange’s equation and are given by 2 = (1 + x0 )1/2 (1 + x)−3/2 − p(1 − x2 )−3/2 ,

(5.58)

where p is determined from Eq. (5.57) and is nondimensionalized with respect to (KL2 )/(2M). Some typical frequency–load plots are shown for the same set of 2

p

x0 > 0

(a)

2

(b)

p 1.5

1.5 x0 < 0 1

1

x0 < 0

x0 = 0 0.5

0.5 x0 > 0 0 −1

− 0.5

0

0.5

x 1

0

0

0.5

1

1.5

2

Figure 5.14. (a) Equilibrium paths and (b) frequency–load relation. α = 5, x0 = 0, ±0.005, ±0.015.

2

17:53

P1: KAE Chapter-05

CUFX159-Virgin

0 521 85648 5

August 14, 2007

82

Discrete-Link Models P L

Y1L

m

L

Y2L L m

k

m k

Figure 5.15. A schematic of a three-bar link model.

initial imperfections in Fig. 5.14(b). We see how for negative values of x0 the natural frequency decays to zero as the system buckles at a limit point. However, when the initial imperfection is in the other direction no instability is encountered and the frequencies start to increase as the structure moves beyond the critical value for the corresponding perfect system. We again note the presence of the complementary equilibrium paths that would not be ordinarily encountered under a monotonic increase in axial load, that is, the natural loading path. The sign of the initial imperfection also has a strong effect on the load-carrying capacity of the structure, and, given the often arbitrary nature of the initial imperfection, this system presents obvious concern in a design context. This type of behavior is sometimes encountered in frame structures. We again note that the axial load is assumed to be an independent parameter (see the discussion in Subsection 5.2.4).

5.5 A Three-Bar Model Now let us consider the dynamics and stability of a simple mechanism made up of three rigid links of length L and mass per unit length m. They are hinged at their connections, and linear rotational springs of stiffness coefficient k are placed at the two internal joints, as shown in Fig. 5.15 [4]. It is assumed that the structure is in equilibrium in its undeflected (straight) configuration and that an axial load of magnitude P is acting. This model has two degrees of freedom, that is, an equilibrium configuration is determined if two coordinate values are specified. In contrast to the simple examples outlined earlier, there is some flexibility in the way we choose the coordinates to describe the deflected configuration of the system. Suppose we choose the lateral deflections as the coordinates (as shown in the figure). In this case, we can write down the strain energy stored in the springs as U=

1 1 k[sin−1 Y1 − sin−1 (Y2 − Y1 )]2 + k[sin−1 Y2 − sin−1 (Y2 − Y1 )]2 , 2 2

(5.59)

which can be expanded to give U=

1 k[5Y 21 − 8Y1 Y2 + 5Y 22 + · · · +]. 2

(5.60)

17:53

P1: KAE Chapter-05

CUFX159-Virgin

0 521 85648 5

August 14, 2007

5.5 A Three-Bar Model

83

We can also write the potential energy associated with the movement of the axial load: 1/2 1/2 V = −PL 3 − 1 − Y12 (5.61) − 1 − Y22 − [1 − (Y2 − Y1 )2 ]1/2 , which can also be expanded to give V = −PL[Y 21 − Y1 Y2 + Y 22 + · · · +].

(5.62)

The kinetic energy for a typical link with displacement a and b at each end is L 1 2 ˙ [a(1 ˙ − x/L) + b(x/L)] d x, (5.63) T= m 2 0 which, after substitution and adding the effects of all three links, leads to

2 1 1 T = mL2 Y˙ 21 + Y˙1Y˙2 + Y˙ 22 . 2 3 2

(5.64)

Thus we have our Lagrangian L = T − U + VP ,

(5.65)

and a direct application of Lagrange’s equation will lead to the equations of motion by use of the dummy suffix notation Tij eY¨j + Vij e Yj = 0,

(5.66)

and, assuming harmonic oscillations Yj = Aj sin ωt, we obtain Vij e Aj − ω2 Tij e Aj = 0, thus leading to the characteristic equation e Vij − ω2 Tij e = 0.

(5.68)

For the specific case at hand, we then have 5k − 2PL − ω2 2 mL3 − 4k + PL − ω2 16 mL3 3 = 0, − 4k + PL − ω2 1 mL3 5k − 2PL − ω2 2 mL3 6

(5.67)

(5.69)

3

and, by using p= 2 =

PL , k

(5.70)

ω2 mL3 , k

(5.71)

we get 6 [(8 − 3p) ± (2p − 7)] , 5 and thus the two natural frequencies 2 =

(5.72)

6 (1 − p), 5

(5.73)

22 = 6(3 − p).

(5.74)

21 =

17:53

P1: KAE Chapter-05

CUFX159-Virgin

0 521 85648 5

August 14, 2007

84

Discrete-Link Models

Each of these eigenvalues corresponds to an eigenvector that describes both the mode of buckling and vibration: The first corresponds to a mode in which both generalized coordinates are equal, the second to a mode in which the coordinates are equal but opposite in sign. From Eqs. (5.73) and (5.74), we can set the natural frequencies equal to zero to obtain the critical loads p1 = 1

(P1 = k/L),

(5.75)

p2 = 3

(P2 = 3k/L).

(5.76)

And setting the axial load equal to zero, we get the natural frequencies 2

ω1 = 6k/(5mL3 ) , 21 = 6/5 2

ω2 = 18k/(mL3 ) . 22 = 18

(5.77) (5.78)

From Eqs. (5.73) and (5.74), we again see the linear relation between the square of the natural frequency and the axial load. The linearity of this relation occurs because of the equivalence of the buckling modes and the natural modes of vibration, having a finite set of generalized coordinates, and the fact that the frequencies depend on a single parameter. We would not necessarily expect this relation to be exactly linear for the general continuous case, as explained earlier. This example can also provide a powerful illustration of the utility of choosing principal coordinates. For this two-DOF (2DOF) system, it is a simple matter to expand the determinant of Eq. (5.69), and, of course, there are a myriad of techniques for achieving this numerically for higher-order systems [9]. However, our earlier theory illustrated how coordinate transformations can be useful in the setting up of the equations of motion in going from physical or generalized to principal coordinates. Thompson and Hunt [4] show that the simple transformation Y1 + Y2 , 2 Y1 − Y2 u2 = 2 u1 =

(5.79) (5.80)

decouples the equations, enabling the critical loads and natural frequencies to be written immediately [see Eq. (3.43)]. However, this type of direct transformation is usually not obvious a priori in a typical analysis. A large variety of numerical algorithms are available to compute this transformation efficiently.

5.6 A Snap-Through Model In this section, we look at a simple example of snap-through buckling associated with a saddle-node bifurcation [12, 13]. The link model shown in Fig. 5.16 consists of two rigid massless links of length L, with a linear spring K allowing horizontal movement at one end. A point mass M is located at the center where a point load of magnitude P acts in the vertical direction. Again we assume that P acts independently of the mass, and we can also think of Fig. 5.16 as a plan view, such that gravity

17:53

P1: KAE Chapter-05

CUFX159-Virgin

0 521 85648 5

August 14, 2007

5.6 A Snap-Through Model

85

M L

K

L P

Figure 5.16. Schematic of a simple arch link model.

is not an issue. The generalized coordinate for this SDOF system is again θ, which is measured from the initial rise of the structure that is fixed at α = π/8 in keeping with earlier work on this model. We can write the total potential energy for this model as V = 2KL2 [cos (α − θ) − cos α]2 − PL [sin α − sin (α − θ)]

(5.81)

and a kinetic energy of T=

1 ML2 θ˙ 2 . 2

(5.82)

A direct application of Lagrange’s equation, and defining p = P/(4KL),

ω2n = 4K/M,

(5.83)

leads to the nondimensional equation of motion: θ¨ + ω2n [cos (α − θ) − cos α] sin (α − θ) − p cos (α − θ) = 0.

(5.84)

The equilibrium path for an initial rise of α = π/8 is plotted in Fig. 5.17, where it is shown superimposed on contours of the total potential energy. It follows the

p 0.02 0.01 0 -0.01

-0.2

0

0.2 0.4 0.6 0.8

1

θ

1.2

Figure 5.17. The potential-energy surface as a function of axial load with the equilibrium path superimposed.

17:53

P1: KAE Chapter-05

CUFX159-Virgin

0 521 85648 5

August 14, 2007

86

Discrete-Link Models run down

(rad) 1 0.8 0.6 0.4

sweep up

0.2 0.0 –0.2

100

200

300

400 t = 1,000p

Figure 5.18. Evolution through snap-through illustrating hysteresis. Initial conditions: For both θ˙ = 0.0, θ(0) = −0.1 at t = 0 and θ(0) = 0.8654 at t = 400.

stationary points of the potential energy. The second derivative can be used to determine which of these stationary points is a minimum, and we know that a critical point is reached when the equilibrium curve passes through a horizontal tangency. For this system, the limit point is reached when p = 0.0116, which corresponds to a deflection of θ = 0.166 rad, that is, when the load reaches this value the structure has deflected to nearly 10 deg and then suddenly snaps through to an inverted position (which is about θ = 0.844 rad, that is, about 25 deg below the horizontal). Note that in this figure the energy is cropped for large values (white) and low values (black) of the potential energy, but in general the darker the shade, the deeper the potential-energy well at that load. Conducting a numerical integration, we can sweep through the instability by using the following loading function: p = −0.02 + 0.0001t.

(5.85)

An example is shown in Fig. 5.18, in which an initial condition is used that is 0.01 rad away from stable equilibrium, and ωn = 1 for convenience. Using the ramp function of Eq. (5.85) thus converts to an anticipated critical time of tcr = 316. The loss of stability is abrupt (albeit slightly delayed) under the gradual increase in load. The postbuckled behavior is characterized by relatively large-amplitude oscillation about the inverted position because of the large transient motion initiated by the jump. Note the asymmetric nature of the waveform as the trajectory oscillates along its decidedly asymmetric potential-energy surface. We can conduct a reverse sweep by changing the direction of the load, and again starting from an initial condition adjacent to the equilibrium we again observe the snap back to the original branch, in which case the load evolution has been scaled such that a jump occurs at about t = 84. This evolving trajectory is shown in gray in Fig. 5.18, and thus a region of hysteresis is revealed. The loading described in this example can be considered as “dead” or “force loaded.” An alternative, which can often be the most practical approach in a

17:53

P1: KAE Chapter-05

CUFX159-Virgin

0 521 85648 5

August 14, 2007

5.6 A Snap-Through Model

87

laboratory context, is to use a “displacement-loaded” device. In the former case, it is straightforward to prescribe the load (usually because of gravity) and then measure deflection. In the latter, we prescribe deflection and measure the resulting load (using a load cell for example). In the displacement-controlled approach, the hysteresis is manifested in terms of displacement, and thus the nature of the stability of equilibrium changes [2]. This also brings with it an issue of extensibility, because of a SDOF model, for example, will not be able to oscillate if subject to displacement-controlled loading. The ability to follow equilibrium paths that have bifurcations or turning points is a subject of considerable importance and will be dealt with in more detail in a later section. We note finally that, unlike for symmetric systems possessing a trivial equilibrium solution, this type of limit point buckling is not sensitive to initial imperfections. It is affected by the presence of a small initial imperfection (in fact linearly) but not in a disproportionate sense. We shall also look at the analog of this system in a continuous (arch) structure later. Before leaving this section, we again take a brief look at the relation between the load and the natural frequency. The local stiffness of the force–deflection curve can be obtained from the derivative of the equilibrium condition, that is, dp cos α − cos3 (α − θ) = , dθ cos2 (α − θ)

(5.86)

which in turn is linearly related to the square of the effective linear natural frequency in the usual way. Hence we can plot the square root of the right-hand side of Eq. (5.86) against load, as shown in Fig. 5.19. An alternative view of stiffness would relate load and vertical deflection rather than angle but the relation is very nearly 0.015 p cr = 0.0116

p

ω 0.01 ω2

0.005 ω4 0 0

0.1

0.2

0.3

0.4 Frequency

0.5

Figure 5.19. The relation between the natural frequency and load for the snap-through model.

17:53

P1: KAE Chapter-05

CUFX159-Virgin

0 521 85648 5

August 14, 2007

88

Discrete-Link Models

the same [2]. It is actually the natural frequency raised to the fourth power that varies linearly with load when close to the instability [14, 15]. This is a typical result in the vicinity of a saddle-node bifurcation and will be touched upon again in the chapter on nondestructive testing.

5.7 Augusti’s Model In this section, we again focus attention on an inverted pendulum model but now replace the torsional spring with a universal joint, that is, a hinge that is not confined to a plane. This is sometimes referred to as Augusti’s model in the literature [1, 16, 17]. Thus the deflection of the system needs two coordinates for a complete description. This model shows some interesting behavior when modes interact. Specifically, we outline a bifurcation from nontrivial equilibrium: the secondary bifurcation. The model is shown in Fig. 5.20. A slender, rigid (but massless) bar of length L is pinned at its base, where rotational springs with constant stiffnesses C1 and C2 (C2 > C1 ) initially act in perpendicular planes and rotate with the bar. The corresponding angles of rotation with respect to two horizontal, perpendicular axes are α1 (t) and α2 (t), and the angles θ1 (t) and θ2 (t) are defined as θ1 (t) = (π/2) − α1 (t),

θ2 (t) = (π/2) − α2 (t),

(5.87)

with θ1 (t) = θ10 and θ2 (t) = θ20 when the springs are unstretched. A downward vertical load P is applied at the top of the bar. A concentrated mass M is attached at the top of the bar.

P P M M m

L

L

m C2 2

C1

1

x2 x1 Figure 5.20. Geometry of the Augusti model.

x2 x1

17:53

P1: KAE Chapter-05

CUFX159-Virgin

0 521 85648 5

August 14, 2007

5.7 Augusti’s Model

89

The potential energy V is given by [18] 1 1 C1 (θ1 − θ10 )2 + C2 (θ2 − θ20 )2 2 2

V=

−PL[(1 − sin2 θ10 − sin2 θ20 )1/2 −(1 − sin2 θ1 − sin2 θ2 )1/2 ], and the kinetic energy T is 2 2 d θ 1 d θ 1 2 T = ML2 cos2 θ1 + cos2 θ2 2 dt dt 2 2 [(d θ1 /d t) sin 2θ1 + (d θ2 /dt) sin 2θ2 ] + 2ML . (1 − sin2 θ1 − sin2 θ2 )

(5.88)

(5.89)

We can again use Lagrange’s equations to obtain the equations of motion. The nonlinear inertia terms in the resulting equations do not affect small vibrations of the system about an equilibrium state. Hence, for simplicity, only the linearized inertia terms are used. The analysis is conducted in terms of the following nondimensional quantities, where is a dimensional vibration frequency: c = C2 /C1 ,

p = PL/C1 ,

τ = t(C1 /ML2 )1/2 ,

ω = (ML2 /C1 )1/2 ,

(5.90)

with c > 1. It can be shown (from equating the second derivative of the potential energy to zero) that p = 1 is the critical buckling load, that is, for the perfect system (the bar is vertical and the springs unstretched). The coupled, nonlinear equations of motion are obtained as p (1 − sin2 θ1 − sin2 θ2 )−1/2 sin θ1 = 0, 2 p θ2 + cθ2 − cθ20 − (1 − sin2 θ1 − sin2 θ2 )−1/2 sin θ2 = 0. 2 θ1 + θ1 − θ10 −

(5.91) (5.92)

Again we focus initially on the underlying equilibria of the perfect system geometry (θ10 = θ20 = 0). The four solutions are θ1 = 0,

θ2 = 0,

θ2 = 0,

p = θ1 / sin θ1 ,

θ1 = 0,

p = cθ2 / sin θ2 , p = (θ1 / sin 2θ1 ) (1 − sin2 θ1 − sin2 θ2 ) = (cθ2 / sin 2θ2 ) (1 − sin2 θ1 − sin2 θ2 ).

(5.93)

These curves are plotted in Fig. 5.21(a), with the secondary bifurcation occurring at c sin 2θ1∗ = 2θ1∗ ,

p = c cos 2θ1∗ ,

(5.94)

17:53

P1: KAE Chapter-05

CUFX159-Virgin

0 521 85648 5

August 14, 2007

90

Discrete-Link Models

1.1

1.1

1.05

(b)

p

(a)

p Primary path

1.05

1

2 B

1

0.95

0.95

Secondary path ( 1* = 0.3745, p* = 1.024)

0.9 0.85

0.9

(c = 1.1) 2

Fundamental (trivial) path

A

0.85 (c = 1.1)

0.8

0

0.2

0.4

0.6

0.8

0.8

1

0

0.05

0.1

0.15

0.2

0.25

1

2

0.3

Figure 5.21. (a) Equilibrium paths for the perfect model and (b) corresponding characteristic curves. 1.1 (c = 1.1, θ10 = 0.01) p 1.05

1.1

(a)

(c = 1.1, θ10 = 0.01)

p

(b)

1.05

1

1

ω B2

(θb = 0.4183, pb = 1.0052) 0.95

0.95

0.9

0.9

0.85

0.85

0.8

0

0.2

0.4

0.6

0.8

θ1

1

0.8

0

ωA2 0.05

0.1

0.15

0.2

0.25

ω2

0.3

Figure 5.22. (a) Equilibrium paths for the imperfect model and (b) corresponding characteristic curves.

which for the specific case shown (with c = 1.1) is at (θ1∗ , p ∗ ) = (0.3745, 1.024). Shown in part (b) is the dependence of the natural frequencies on the axial load. These are evaluated on the equilibrium paths. For the trivial equilibrium, we have ω2A = 1 − p,

ω2B = c − p,

(5.95)

and for the primary (postbuckled, i.e., p > 1) branch, we have ω2A = 1 − p cos θ1 = 1 − θ1 / tan θ1 ,

(5.96)

ω2B = c − p cos θ1 = c − 2θ1 / sin 2θ1 .

(5.97)

For the case in which there is a small amount of initial geometric imperfection, the results shown in Fig. 5.22 are obtained. The initial angle θ10 = 0.01 is chosen. The secondary bifurcation now occurs directly from the primary path. It is interesting to note that previous studies of this system have often used truncation to ease some

17:53

P1: KAE Chapter-05

CUFX159-Virgin

0 521 85648 5

August 14, 2007

5.8 Multiple Loads

0.4

91

θ1

0.2 0.0 250

−0.2

500

750

1000

1250

τ

1500

1000

1250

τ

1500

−0.4

0.4

θ2

p(τ )

0.2 0.0 −0.2

250

500

750

−0.4

Figure 5.23. A slow sweep through secondary bifurcation.

of the computations, and in that case the load–frequency (squared) relations are exactly linear. We again close this section by conducting a numerical simulation as the system is swept through the bifurcation(s). Figure 5.23 was based on the following load evolution, p = 0.95 + 0.00005τ,

(5.98)

where the initial geometry is perfect and a very small amount of damping was added to the system. We observe a gradual decrease in the θ1 natural frequency as the system approaches the initial bifurcation at τ = 1000 and then a gradual increase in the natural frequency. However, in the lower part of this figure we also see that it is the natural frequency associated with θ2 that decreases as the secondary bifurcation is approached at τ = 1480. This system will be revisited a couple of times later in this book: as an example of a path-following algorithm and in cases in which oscillations are not necessarily small.

5.8 Multiple Loads In all the link model examples so far there has been a single axial load. A monotonic increase in this parameter resulted in a linear decay in the natural frequency (squared). However, if more than one independent axial load is present, then the natural frequencies will still decay if either or both of these loads are increased,

17:53

P1: KAE Chapter-05

CUFX159-Virgin

0 521 85648 5

August 14, 2007

92

Discrete-Link Models

P2 m2 2

P1 m1

L

k2

Figure 5.24. A 2DOF model under the action of two axial loads.

L 1

k1

but a linear relation is no longer to be expected. To illustrate this effect, consider the link model shown in Fig. 5.24. Following [19, 20], and assuming k1 = 2k, k2 = k, m1 = 2m, m2 = m, we can write the energy terms: 1 U = kθ12 + k(θ2 − θ1 )2 , 2 1 1 VP = − LP1 θ12 − LP2 θ12 + θ22 , 2 2 1 T = mL2 3θ˙ 12 + θ˙ 22 + 2θ˙ 1 θ˙ 2 . 2

(5.99) (5.100) (5.101)

We can nondimensionalize by using the following parameters, 2 =

ω2 mL2 , k

pi =

Pi L , k

(5.102)

and by using Lagrange’s equations obtain the characteristic equation 24 + 2 (p 1 + 4p 2 − 8) + p 22 + p 1 p 2 − p 1 − 4p 2 + 2 = 0.

(5.103)

Setting 2 = 0 gives the two critical load conditions from the quadratic p 22 + p 1 p 2 − p 1 − 4p 2 + 2 = 0,

(5.104)

which can be solved for the lowest critical loads acting separately of p 1 = 2 and p 2 = 0.586. Setting both axial loads equal to zero leads to the natural frequencies 21 = 0.268 and 21 = 3.732. Thus we can plot the roots of Eq. (5.103) as two surfaces, as shown in Fig. 5.25. The curve on the left is the most relevant given the typical situation of a monotonic increase in the loads, that is, it will be encountered first.

17:53

P1: KAE Chapter-05

CUFX159-Virgin

0 521 85648 5

August 14, 2007

5.9 Load-Dependent Supports

2

93

3

0.2

4

2 2 2

1 0.1 0.4 0 0.5

0.2

p1

1

3

1 0 0

p2

2 1

p1

1.5 2 0

1

2 3 40

Figure 5.25. The characteristic surface plotting the natural frequency squared in terms of the two axial loads.

5.9 Load-Dependent Supports It may happen that the stiffness of a structure or its supports is a function of the applied axial load [21, 22]. To illustrate this situation, we briefly return to the inverted pendulum model (see Fig. 5.1), but now we assume a distributed mass such that the moment of inertia of the bar about its base is I. We assume a fixed vertical end load P and ignore gravity. We assume a spring stiffness, rather than a constant torsional stiffness k, that increases linearly with the end load from a baseline value of K0 . The equation of motion is given by θ¨ − p sin θ + k(p)θ = 0, where the following nondimensional parameters have been used, p = PL/K0 , k(p) = K(pK0 /L)/K0 , t = T (K0 /I),

(5.105)

(5.106)

and the derivatives in Eq. (5.105) are with respect to the scaled time t. Equilibrium conditions are obtained from −p sin θ + k(p)θ = 0,

(5.107)

and the frequencies of small vibrations about these equilibria are given by ω2 = k(p) − p cos θ.

(5.108)

Now, assume the spring stiffness is related to load in the following way: k(p) = 1 + γ p,

(5.109)

where γ is taken as positive. Equilibrium paths and load–frequency relations are plotted in Fig. 5.26 for a number of different values of the parameter γ. We see an increase in the critical load and an increase in the natural frequency at a given load as a function of spring stiffness. Further examples of rigid-link models and their stability can be found in Seyranian and Mailybaev [23].

p2

17:53

P1: KAE Chapter-05

CUFX159-Virgin

0 521 85648 5

August 14, 2007

94

Discrete-Link Models 3

3 (a)

p

p

0.4

2.5

(b)

= 0.6

2.5 = 0.6

2 1.5

0.2

0.4

2

0.2

1.5

0.0

0.0 1

1

0.5

0.5

0

-2

-1.5

-1

-0.5

0

0.5

1

1.5

0

2

0

0.2

0.4

0.6

0.8

2

1

Figure 5.26. (a) Equilibrium paths and (b) load–frequency relation for a link model with a stiffening spring.

5.10 Path Following and Continuation In a number of places in this chapter, we have had cause to solve nonlinear algebraic equations, and this will be the case throughout this book. The standard technique for solving sets of nonlinear algebraic equations is Newton–Raphson [9, 24] and this is a standard feature in Mathematica [25] and MATLAB [26]. However, the solution path, as a parameter is changed, may be quite complicated (e.g., including turning points), and some difficulty may be encountered. A number of specialized techniques have been developed based on augmenting Newton–Raphson such that a solution path is followed. These are predictor–corrector techniques and work for differential as well as algebraic equations, and are typically called continuation methods. Because these techniques typically involve the rates of change of the response as a function of a parameter, they obtain information about stability (based on the evaluation of the Jacobian) without too much difficulty. A popular and efficient algorithm is contained in the software package auto.

1.8 1.6

1.6 p

1.4 p 1.2

1.4 1.2

1

1 0.8

0.8

0.6

0.6

0.4

0.4

0.2 0 -1.5

0.2

-1

-0.5

0 1

0.5

1

1.5

-2

-1

0 2

1

1

2 0 0

0.5 0.5 00 0.2 0.4 0.4 0.6 0.6 0.8 0.8

0.5 11.0 1.2 1.2 1.4

1.6

1

2

1

Figure 5.27. Path following results for the Augusti model: (a) geometrically perfect initial configuration and (b) with initial imperfections [29].

17:53

P1: KAE Chapter-05

CUFX159-Virgin

0 521 85648 5

August 14, 2007

References

Rather than detail specific algorithmic features being given here, readers are referred to Doedel [27] and Doedel et al. [28] for more details. But an example is given based on the solution of the equilibrium equations for the Augusti model from Section 5.7. Using the same parameters as those for Figs. 5.21(a) and 5.22(a), auto was used to generate the results shown in Fig. 5.27 [29].

References [1] [2] [3] [4] [5] [6] [7] [8]

[9] [10]

[11] [12] [13] [14] [15] [16]

[17]

[18] [19] [20]

J.M.T. Thompson and G.W. Hunt. A General Theory of Elastic Stability. Wiley, 1973. J.G.A. Croll and A.C. Walker. Elements of Structural Stability. Wiley, 1972. G.J. Simitses. An Introduction to the Elastic Stability of Structures. Prentice Hall, 1976. J.M.T. Thompson and G.W. Hunt. Elastic Instability Phenomena. Wiley, 1984. L.N. Virgin. The dynamics of symmetric postbuckling. International Journal of Mechanical Sciences, 27:235–48, 1985. A.N. Kounadis. Nonlinear dynamic buckling of discrete dissipative or nondissipative systems under step loading. AIAA Journal, 29:280–9, 1991. A.N. Kounadis. Nonlinear dynamic buckling and stability of autonomous structural systems. International Journal of Mechanical Sciences, 35:643–56, 1993. I. Elishakoff, S. Marcus, and J.H. Starnes. On vibrational imperfection sensitivity of Augusti’s model structure in the vicinity of a nonlinear static state. International Journal of Non-Linear Mechanics, 31:229–36, 1996. W.H. Press, B.P. Flannery, S.A. Teukolsky, and W.T. Vetterling. Numerical Recipes in Fortran. Cambridge University Press, 1992. A.C. Walker, J.G.A. Croll, and E. Wilson. Experimental models to illustrate the nonlinear behavior of elastic structures. Bulletin of Mechanical Engineering Education, 10:247–59, 1971. M.A. Souza. Vibration of thin-walled structures with asymmetric post-buckling characteristics. Thin-Walled Structures, 14:45–57, 1992. P.X. Bellini. The concept of snap-buckling illustrated by a simple model. International Journal of Nonlinear Mechanics, 7:634–50, 1972. D.A. Pecknold, J. Ghaboussi, and T.J. Healey. Snap-through and bifurcation in a simple structure. Journal of Engineering Mechanics (ASCE), 111:909–22, 1985. L.N. Virgin. Parametric studies of the dynamic evolution through a fold. Journal of Sound and Vibration, 110:99–109, 1986. J.M.T. Thompson and H.B. Stewart. Nonlinear Dynamics and Chaos, 2nd ed. Wiley, 2002. G. Augusti, V. Sepe, and A. Paolone. An introduction to compound and coupled buckling and dynamic bifurcations. In J. Rondal, editor, Coupled Instabilities in Metal Structures: Theoretical and Design Aspects. Springer-Verlag, 1998, pp. 1–27. N. Challamel. Softening branches of a two-degree-of-freedom system induced by spatial buckling. International Journal of Structural Stability and Dynamics, 6:493–512, 2006. L.N. Virgin and R.H. Plaut. Use of frequency data to predict secondary bifurcation. Journal of Sound and Vibration, 251:919–26, 2002. K. Huseyin and J. Roorda. The loading-frequency relationship in multiple eigenvalue problems. Journal of Applied Mechanics, 38:1007–11, 1971. K. Huseyin. Multiple Parameter Stability Theory and Its Applications. Oxford University Press, 1986.

95

17:53

P1: KAE Chapter-05

CUFX159-Virgin

96

0 521 85648 5

August 14, 2007

Discrete-Link Models [21] R.H. Plaut. Column buckling when support stiffens under compression. Journal of Applied Mechanics, 56:484, 1989. [22] R.H. Plaut. Stability and vibration of a column model with load-dependent support stiffness. Dynamics and Stability of Systems, 6:79–88, 1991. [23] A.P. Seyranian and A.A. Mailybaev. Multiparameter Stability Theory with Mechanical Applications. World Scientific, 2003. [24] T.S. Parker and L.O. Chua. Practical Numerical Algorithms for Chaotic Systems. Springer-Verlag, 1989. [25] S. Wolfram. The Mathematica Book. Cambridge University Press, 1996. [26] MATLAB. User’s guide. Technical report, The Math Works, 1989. [27] E.J. Doedel. AUTO—Software for continuation and bifurcation problems in ordinary differential equations. California Institute of Technology, 1986. [28] E.J. Doedel, A.R. Champneys, T.F. Fairgrieve, Y.A. Kuznetsov, B. Sandstede, and X.J. Wang. Auto97: Continuation and bifurcation software for ordinary differential equations. Technical report, Department of Computer Science, Concordia University, Montreal, Canada, 1997 (available by FTP from ftp.cs.concordia.ca in directory pub/doedel/auto). [29] H. Chen. Nonlinear analysis of post-buckling dynamics and higher order instabilities of flexible structures. Ph.D. dissertation, Duke University, 2004.

17:53

P1: KAE Chapter-06

CUFX159-Virgin

0 521 85648 5

August 14, 2007

6 Strings, Cables, and Membranes

6.1 Introduction In transitioning from discrete to continuous systems, we will naturally encounter partial differential equations; that is, the link models of the previous chapter lead to an algebraic eiegnvalue problem whereas the systems considered in this chapter lead naturally to a differential eigenvalue problem, even though we will find utility in approximations leading back to a discrete description. Although the main focus in this book is the behavior of structures subject to compressive axial load, it is instructive to consider systems subject to tensile loads as an introduction. Because these systems tend to be used in a linear context, they typically do not suffer the instability phenomena associated with buckling. However, they do provide a relatively gentle introduction to the behavior and methods of analysis associated with distributed systems subject to axial loading. They also provide a compelling analogy with an everyday example of the relation between axial loading (tension) and natural frequency (pitch): tuning a stringed musical instrument.

6.2 The Stretched String 6.2.1 The Wave Equation We start by considering the undamped, small-amplitude motion of a stretched string under a tension τ. By considering a small element dx, as shown in Fig. 6.1, we have for horizontal equilibrium τ(x + dx) cos [θ(x + dx)] − τ(x) cos θ(x) = 0,

(6.1)

and because for small slopes cos θ ≈ 1, τ is constant. For vertical equilibrium τ sin [θ(x + dx)] − τ sin θ(x) = ρdx

∂2 w , ∂t2

(6.2)

where ρ is the mass per unit length of the string. Expanding the sine terms in Eq. (6.2) gives ∂2 w ∂θ(x) + · · · + − τ [θ(x) + · · · +] = ρdx 2 , τ θ(x) + dx ∂x ∂t

(6.3) 97

17:54

P1: KAE Chapter-06

CUFX159-Virgin

0 521 85648 5

August 14, 2007

98

Strings, Cables, and Membranes

w

τ(x+dx) θ(x+dx)

w+dw

w

θ(x)

ds

(x) x x

x+dx

Figure 6.1. Forces acting on a segment of a taut string when undergoing transverse vibration.

and because ∂θ/∂x = ∂2 w/∂x2 we obtain the governing equation of (lateral motion) for a string: ∂2 w 1 ∂2 w = , ∂x2 c2s ∂t2

(6.4)

√ where cs = τ/ρ is a constant, and later to be identified with the speed of lateral motion. It also relates to the velocity of wave propagation along the string, although we are primarily interested in transverse (lateral) effects here. Equation (6.4) is a partial differential equation of the type introduced in Chapter 4, is called the wave equation, and occurs in many branches of the physical sciences [1]. A standard approach to solving partial differential equations of this type is based on the separation of variables. This approach will also be used when we consider the dynamic behavior of structures with bending stiffness later. We assume that the displacement of the string can be written as w(x, t) = φ(x)q(t),

(6.5)

which, on substitution in Eq. (6.4) leads to 1 d2 φ 1 1 d2 q = . φ dx2 q c2s dt2

(6.6)

We see that the left-hand side is a function of the position and the right-hand side is a function of time. In this case both sides of the equation must be equal to the same constant, which we set as −(ω/cs )2 (with the negative sign and square chosen with forethought). Now we have the two uncoupled ordinary differential equations, d2 φ + dx2

ω cs

2 φ = 0,

(6.7)

d2 q + ω2 q = 0. dt2

(6.8)

17:54

P1: KAE Chapter-06

CUFX159-Virgin

0 521 85648 5

August 14, 2007

6.2 The Stretched String

99

These linear equations have very familiar solutions ω ω x + B cos x, φ(x) = A sin cs cs

(6.9)

q(t) = C sin ωt + D cos ωt,

(6.10)

where the constants (A, B) and (C, D) are obtained from the boundary and initial conditions [2]. Suppose we have a string that is stretched between fixed points. In this case, the boundary conditions can be written as w(0, t) = 0,

(6.11)

w(L, t) = 0.

(6.12)

Evaluating the spatial part of the solution [Eq. (6.9)] under these circumstances, we get B = 0 from Eq. (6.11), and from Eq. (6.12), ω A sin L = 0. (6.13) cs The solutions to this (in addition to the trivial solution A = 0) are ωn L = nπ, cs

n = 1, 2, . . . ,

(6.14)

that is, there are an infinite number of natural frequencies: nπcs nπ τ ωn = = . L L ρ

(6.15)

And thus we have a linear relation between tension τ and the square of the natural frequencies ωn with corresponding mode shapes, ωn x nπx . (6.16) = sin φn (x) = sin cs L The mode shapes satisfy the conditions of orthogonality described earlier in this book (see Section 4.3). Thus we have an equation for the lateral displacement of the string given by w(x, t) =

∞

qn (t) sin

n=1

=

∞

nπx L

(6.17)

(Cn sin ωn t + Dn cos ωn t) sin

n=1

nπx , L

(6.18)

or, using complex notation, w(x, t) =

∞ n=1

βn eiωn t sin

nπx , L

(6.19)

where the βn are complex (and it is the real part that corresponds to physically meaningful solutions). The constants associated with the temporal part of the solution are

17:54

P1: KAE Chapter-06

CUFX159-Virgin

0 521 85648 5

August 14, 2007

100

Strings, Cables, and Membranes

determined from the initial conditions w(x, 0) and ∂w(x, 0)/∂t and correspond to the Fourier coefficients. As an example, consider a taut string that is displaced by an amount H (not too large) at its center and then released (i.e., with an initial velocity of zero). The string has the initial form 2H x, L 2H w(x, 0) = (L − x), L

w(x, 0) =

0 ≤ x ≤ L/2, (6.20) L/2 ≤ x ≤ L,

and evaluating C and D leads to a response πx 3πx 1 8H cos ω1 t − sin cos ω3 t + · · · + . w(x, t) = 2 sin π L 9 L

(6.21)

That only odd harmonics are excited is due to the symmetric nature of the initial disturbance, which leads to symmetric motion. The resulting triangular wave is dominated by the first (fundamental) mode, as expected. Thus we see that the boundary conditions influence the mode shapes and natural frequencies and the initial conditions determine the contribution of each mode. We observe our familiar linear relation between the square of the natural frequency and the tension. This closed-form solution would not have been available if the tension in the string had not been constant. Later in this book we will use a somewhat similar approach for beams and plates. The major differences are that systems with bending stiffness will tend to lead to higher-order differential equations (and hence more boundary conditions) and the forces of interest will primarily be compressive, which allows for bifurcational phenomena to appear.

6.2.2 Traveling-Wave Solution At this point we might wonder why Eq. (6.4) has the name it does. We now show that an alternative, but equivalent, description can be obtained by writing the general solution in the form w(x, t) = F1 (x − cs t) + F2 (x + cs t).

(6.22)

We can think of the first term on the right-hand side of Eq. (6.22) as representing a wave moving in the positive x direction with constant velocity cs , with the second term corresponding to a similar wave but moving in the opposite direction. In both cases the functions F represent the (non-changing) shape, or profile, of the wave. Because we have already seen that sinusoidal motion is typical in the vibration of strings, let’s consider a wave of the form w(x, t) = A sin

2π (x − cs t), λ

(6.23)

17:54

P1: KAE Chapter-06

CUFX159-Virgin

0 521 85648 5

August 14, 2007

6.2 The Stretched String

101

that is, a wave of amplitude A and wavelength λ traveling in the positive x direction. Introducing the wavenumber k = 1/λ, we can rewrite Eq. (6.23) as w(x, t) = A sin (2πkx − ωt),

(6.24)

in which ω = cs (2π)/λ is the frequency of the wave. However, if we have two equal waves but traveling in opposite directions, we then have w(x, t) = A sin (2πkx − ωt) + A sin (2πkx + ωt),

(6.25)

which can be rewritten as w(x, t) = 2A sin (2πkx) cos (ωt).

(6.26)

Thus we see that the two waves in this case together respresent a standing wave but with an oscillating profile. The two components cancel at those points where x = nλ/2, and these are called node points. If x = λ(2n + 1)/4, then we have a reinforcement, or antinodes. To apply this approach to a specific (finite) string, we note that, for example, for a string stretched between two points we will have nodes at the end points, and thus 2kL = r where r is an integer, and therefore τ cs ωr = 2πkcs = rπ = rπ (6.27) L ρL2 with the frequencies obtained previously. We note that, in the presence of damping, the traveling waves will decay, and although it is possible to solve the wave equation with damping, the solution becomes considerably more involved. 6.2.3 Energy Considerations and Rayleigh’s Principle The wave equation can also be obtained by use of Hamilton’s principle. The kinetic energy of the string is given by L 2 1 ∂w T= ρ dx (6.28) 2 0 ∂t 2 L ∞ 1 nπx = ρ q˙ sin dx (6.29) 2 0 L n=1

=

ρL 4

∞

q˙ 2n ,

(6.30)

n=1

and the potential energy associated with the stretching is given by ⎞ ⎛ 2 L ⎝ 1 + ∂w − 1⎠ dx U=τ ∂x 0 ≈

1 τ 2

0

L

∂w ∂x

(6.31)

2 dx

(6.32)

17:54

P1: KAE Chapter-06

CUFX159-Virgin

0 521 85648 5

August 14, 2007

102

Strings, Cables, and Membranes

1 = τ 2

L 0

∞ nπ

nπx qn cos L L

n=1

2 dx

(6.33)

∞

=

ρL 2 2 ωn qn , 4

(6.34)

n=1

where we have used the fact that, from Eq. (6.15), ωn L 2 τ=ρ . nπ

(6.35)

We can then invoke Hamilton’s principle and, by taking variations in the energy terms, arrive at the wave equation. Assuming there is no external forcing or damping, we also have the conservation of total energy, that is, ∞

E = T+U=

ρL 2 ωn |βn |2 , 4

(6.36)

n=1

where the βn [from Eq. (6.19)] depends on the initial conditions. We can also use an approximate mode shape (although it is not really necessary in this instance as the exact solution is well known), such as a parabola: w = 4H(t)x(L − x)/L2 .

(6.37)

Plugging this into the expression for the kinetic and potential energy expressions [Eqs. (6.28) and (6.31)] and using Rayleigh’s method gives ω2 =

10τ , ρL2

(6.38)

which is less than 1% greater than the exact value [i.e., ω2 = π2 τ/(ρL2 )]. Rayleigh devised a method of improving this result by incorporating an adjustable constant into the displacement function [3]. It is interesting to note that when a string is excited it often exhibits non-planar motion. This type of whirling motion (familiar from a child’s skipping rope) is just one type of complicated motion found in the forced string problem. In this chapter, we have assumed that the tension in the string remains constant during motion. However, the tension must fluctuate, and it is the subtle interaction between longitudinal and transverse motion that underlies much of the interesting (nonlinear) behavior. In fact, the greater the tension in the string, the relatively less the tension changes during motion. This problem will be revisited later when we consider large-amplitude vibration.

6.3 A Suspended Cable A natural extension to the study of the dynamics of a taut string is to consider the behavior of a string that is not taut, but rather sags because of the effect of gravity [4]. Without bending stiffness, it is again the axial load that provides the restoring

17:54

P1: KAE Chapter-06

CUFX159-Virgin

0 521 85648 5

August 14, 2007

6.3 A Suspended Cable

103

ds

dy

g

H dx

V

T l

x d

y

L

Figure 6.2. The cable geometry and the forces acting on a segment.

force for this important class of practical problems. We focus on symmetric cables having suspension points located at the same vertical elevation, and consider only modes of transverse vibration taking place in-plane [5]. If the cable configuration is shallow, then certain simplifying assumptions can be made, but, in general, the sag may have a relatively profound effect on dynamic behavior. Because of the role of gravity, it is convenient to use the coordinate system shown in Fig. 6.2. Assuming the cable is inextensional, we can write the equilibrium in the vertical direction as d dy T = −mg, (6.39) ds ds and in the horizontal direction we have d dx T = 0, ds ds

(6.40)

which can be integrated to give T

dx = H, ds

(6.41)

in which H is the horizontal component of the cable tension. Thus Eq. (6.39) can be rewritten as H

d2 y ds = −mg . 2 dx dx

Using horizontal equilibrium and using ds2 = dx2 + dy2 , we can then write 2 d dy mg dy =− 1+ . dx dx H dx

(6.42)

(6.43)

The solution to Eq. (6.43) is the well-known catenary, and given that the end points of the cable have the same vertical elevation, we can write the solution in terms of the length-to-span ratio [5, 6], L sinh [lmg/(2H)] = , l lmg/(2H)

(6.44)

17:54

P1: KAE Chapter-06

CUFX159-Virgin

0 521 85648 5

August 14, 2007

104

Strings, Cables, and Membranes

and the sag-to-length ratio H d = {cosh [lmg/(2H)] − 1}. l lmg

(6.45)

For a shallow cable (with a small sag), we can expand the hyperbolic functions in Eqs. (6.44) and (6.45) as Taylor series or simplify governing equation (6.42): d2 y mg . =− 2 dx H It can easily be shown that the deflected shape of the cable is then given by x x x l2 mg x 1− = 4d 1 − , y= 2H l l l l

(6.46)

(6.47)

where the vertical sag in the center of the cable d is given by d 1 lmg = . l 8 H

(6.48)

We can now consider oscillations about equilibrium. In addition to an inertia force (by use of D’Alembert’s principle), the added motion w(x, t) will induce a varying horizontal force h(t), and thus the equation of vertical motion is (H + h)

∂2 ∂2 w (y + w) = −mg + m , ∂x2 dt2

(6.49)

which, for small-amplitude motion, can be simplified to m ∂2 w mg h ∂2 w . − = 2 ∂x H dt2 H H

(6.50)

It can also be shown [6] that for inextensional cables (based on linearized theory) l h(t) depends on the displacement function w(x, t) such that 0 wdx = 0. If we then seek harmonic oscillations of the form w(x, t) = w(x) ˜ cos ωt,

h(t) = h˜ cos ωt,

(6.51)

then substituting these into Eq. (6.50) and using β2 = (mω2 )/H results in ∂2 w˜ mg h˜ . + β2 w˜ = 2 ∂x H H

(6.52)

Solving Eq. (6.52) by using the boundary conditions w(0) = w(l) = 0 results in the following solutions: r The symmetric modes: w˜ cos [βn (x − (1/2)l)] h˜ 8Cn 1− , (6.53) = = Cn , 2 d (βn l) cos[(1/2)βn l] H l with 0 wdx ˜ = 0 leading to the lowest root (1/2)β1 l = 4.493. Therefore the lowest (symmetric mode) frequency is ω1 = 8.99(H/ml2 )1/2 .

17:54

P1: KAE Chapter-06

CUFX159-Virgin

0 521 85648 5

August 14, 2007

6.3 A Suspended Cable

105

(a)

(b)

Figure 6.3. The first inextensional symmetric and the first antisymmetric modes.

r The antisymmetric modes: w˜ 2π = Cn sin n x , d l

2π ωn = n l

H m

(6.54)

with h˜ = 0 leading to the lowest (antisymmetric mode) frequency ω1 = 2π(H/ml2 )1/2 . The lowest symmetric and antisymmetric modes are shown in Figs. 6.3(a) and 6.3(b), respectively. In the former case, it is interesting to see that the wavelength of the mode is shorter than the span and the cable straightens out near the supports. In the latter case, we recognize the form from the taut string, as one side moves up and the other side moves down. It is also interesting to note that these antisymmetric modes would not be influenced by any elasticity of the cable (whereas the symmetric modes may be quite influenced). For example, by introducing the parameter λ, accounting for the elastic flexibility of the cable, EA 8d 2 λ2 = , (6.55) H l we can incorporate this effect into the preceding theory [6]. As the cable flexibility decreases (λ → ∞), we approach the inextensible case already considered. At the other end of the spectrum, we have λ → 0 and a frequency that tends to decrease, and reaches the value of a taut string ω = π H/ml2 . This effect is summarized in Fig. 6.4. We observe that when λ2 < 4π2 , the lowest mode is symmetric (with no internal nodes), but at the transition point λ2 = 4π2 the in-plane frequencies are equal. This type of modal exchange will also be encountered later in the behavior of a pin-ended column with an elastic restraint at midspan, as well as other structural systems in which axial load is a consideration. When λ2 > 4π2 , the lowest mode

17:54

P1: KAE Chapter-06

CUFX159-Virgin

0 521 85648 5

August 14, 2007

106

Strings, Cables, and Membranes (a) 2

3

2

λ =

1

symmetric

10 0

10 1

(b)

3

2

antisymmetric

1

0

4π2

10 2

10 3

0

0.025

0.05

0.075

d/L

0.1

Figure 6.4. Dynamic response of an extensible cable with sag: (a) two lowest natural frequencies and (b) experimental verification (adapted from [7]).

is antisymmetric. Part (b) of Fig. 6.4 also shows this crossover of modes but now as a function of the sag d/l, and some experimental data points are superimposed [6, 7]. The lowest frequency here corresponds to a sway mode. Thus we see that the dynamic response of suspended cables depend on a variety of factors including sag (and hence mass), vertical distance between the support points, flexibility of the cable itself, and so on. The symmetric modes of vibration of the cable are heavily influenced by the cable stiffness (characterized by λ2 ) and thus depend on sag and cable flexibility. 6.3.1 The Hanging Chain The oscillations of a flexible cable suspended vertically from a single fixed point represent a problem that occupies an important place in the historical development of mechanics [8, 9]. It was one of the first systems in which the normal modes of vibration were identified and provided an initial motivation for the development of Bessel functions [10]. In a gravitational field, the weight of a vertically suspended slender beam becomes increasingly important, that is, as the length of the beam increases the bending stiffness becomes negligible in comparsion with gravitational effects. The hanging chain has no bending stiffness. At the opposite end of the flexibility spectrum is the rigid-arm pendulum. Rather than the classical analysis being described here (we shall of course focus on beams in bending later) a relatively simple approximate analysis based on Rayleigh’s method is shown. Assuming the origin is placed at the top end of a vertically hanging chain, of length L, a reasonable first-mode shape in terms of lateral deflection w is given by 2 x x w=Q , (6.56) +β L L in which x is measured downward from the top and β is a constant to be determined. Note that, in contrast to a hanging beam to be considered later, a linear term is

17:54

P1: KAE Chapter-06

CUFX159-Virgin

0 521 85648 5

August 14, 2007

6.3 A Suspended Cable

107

Figure 6.5. The lowest four vibration mode shapes for an experimental hanging beam.

included to allow for a nonzero slope at the clamped end. The kinetic energy is 1 m 2

T=

L 0

w˙ 2 dx =

1 ˙ 2 (10 + 15β + 6β2 ). mLQ 60

(6.57)

1 mgQ2 (3 + 4β + 2β2 ), 12

(6.58)

The potential energy is V=

1 mg 2

0

L x

w2 dxdx =

0

and application of Rayleigh’s method (see Section 3.3) gives 5(3 + 4β + 2β2 ) g ω = . 10 + 15β + 6β2 L 2

(6.59)

The value of β (= 0.289206) that minimizes the frequency results in ω = √ 1.2025 g/L. Because of the extremum nature of Rayleigh’s method, we know that the lower the frequency estimate is, the closer it will be to the exact answer. How√ ever, the exact value (obtained in this case with Bessel functions) is 1.2025 g/L [11], and thus the mode shape described in Eq. (6.56) is obviously very accurate. It is in√ teresting to note that for a rigid bar of the same length the frequency is 1.22474 g/L [12]. Figure 6.5 shows some experimental snapshots of a hanging axisymmteric chain suspended from a spinning shaft. These mode shapes correspond closely to those obtained with Bessel functions for bending motion with nodal points easily observed.

17:54

P1: KAE Chapter-06

CUFX159-Virgin

0 521 85648 5

August 14, 2007

108

Strings, Cables, and Membranes

y

a

y

S

W 22 S S w

b

x 0

x S

Figure 6.6. (a) A rectangular stretched membrane and (b) the (2,2) mode of vibration.

6.4 A Rectangular Membrane A stretched membrane may exhibit transverse vibrations in much the same way as the string, but we now need two dimensions to describe the geometry [2, 13]. We take a brief look at the simple case of a rectangular membrane of uniform thickness, with mass density ρ, which is stretched such that the tension S can be assumed to be constant over the membrane. A schematic of the membrane is shown in Fig. 6.6(a). The lateral deflection is w, and the membrane has length a in the x direction and width b in the y direction. After harmonic motion is assumed, the governing equation is given by ∇ 2 W(x, y) + β2 W(x, y) = 0,

(6.60)

where β2 = ρω2 /S and ∇ 2 is the Laplacian [2], that is, ∇ 2 = ∂2 /∂x2 + ∂2 /∂y2 , which we shall also make use of in the chapter on plates. We can separate variables, apply the boundary conditions, and obtain the natural frequencies 2 S m 2 n ωmn = π , m, n = 1, 2, . . . . (6.61) + ρ a b The corresponding (normalized) mode shapes are given by Wmn = √

nπy 2 mπx sin , sin a b ρab

m, n = 1, 2, . . . .

(6.62)

An example of themode corresponding to the fourth lowest vibration (with frequency ω22 = 2π (S/ρ)(1/a2 + 1/b2 ) is shown in Figure 6.6(b). Hence we again see that the square of the natural frequencies increases linearly with the tension. The lowest frequency for a square membrane is thus given by ω = π 2S/(ρa2 ). Alternatively, we can write down the kinetic energy associated with the vibration of the membrane as a b 1 T= ρ w˙ 2 dxdy, (6.63) 2 0 0

17:54

P1: KAE Chapter-06

CUFX159-Virgin

0 521 85648 5

August 14, 2007

References

109

and the change in potential energy that is due to the tension as the membrane deflects as a b 2 2 ∂w 1 ∂w dxdy. (6.64) V= S + 2 0 0 ∂x ∂y The deflection can be represented by the function w=

∞ ∞

φmn (t) sin

m=1 n=1

nπy mπx sin , a b

(6.65)

which is then used to evaluate the energy terms ∞

T=

∞

1 φ˙ 2mn ρab 8

(6.66)

m=1 n=1

and ∞

∞

1 V = Sab φ2mn 8 m=1 n=1

mπ a

2

+

nπ b

2 .

We again use Lagrange’s equation (2.35) to obtain the equations of motion 2 2 m S n ¨φmn + π2 φmn = 0, + m, n = 1, 2, . . . , ρ a b

(6.67)

(6.68)

with a set of natural frequencies obtained previously [Eq. (6.61)]. Later, we shall compare this behavior with that of systems in which there is also bending stiffness, that is, plates. It turns out that the analysis of membranes depends very much on the shape of the boundary. For example, a circular membrane is more conveniently analyzed by use of polar coordinates and results in Bessel functions [10], and an irregularly shaped membrane would typically require a FEA. The statements made earlier regarding the large-amplitude motion of strings also apply to membranes but we now move on to consider the much wider class of problem in which the structure subject to axial loading possesses bending stiffness and the axial loads are often compressive.

References [1] [2] [3] [4] [5] [6] [7]

R. Courant and D. Hilbert. Methods of Mathematical Physics. Wiley Classics Library, 1989. L. Meirovitch. Principles and Techniques of Vibrations. Prentice Hall, 1997. Lord Rayleigh (John William Strutt). The Theory of Sound. Dover, 1945. H.M. Irvine and T.K. Caughey. The linear theory of free vibrations of a suspended cable. Proceedings of the Royal Society of London Series A, 341:299–315, 1974. S. Krenk. Mechanics and Analysis of Beams, Columns and Cables. Springer, 2001. H.M. Irvine. Cable Structures. MIT Press, 1981. S.E. Ramberg and O.M. Griffin. Free vibration of taut and slack marine cables. Journal of the Structural Division, Proc. ASCE, 103:2079–92, 1977.

17:54

P1: KAE Chapter-06

CUFX159-Virgin

110

0 521 85648 5

August 14, 2007

Strings, Cables, and Membranes [8] H. Lamb. Higher Mechanics. Cambridge University Press, 1929. [9] D. Yong. Strings, chains, and ropes. SIAM Review, 48:771–81, 2006. [10] G.N. Watson. A Treatise on the Theory of Bessel Functions. Cambridge University Press, 1966. [11] H. Lamb. The Dynamical Theory of Sound. Arnold, 1910. [12] R.D. Blevins. Formulas for Natural Frequencies and Mode Shapes. Van Nostrand Rheinhold, 1979. [13] A.D. Dimarogonas. Vibration for Engineers. Prentice Hall, 1996.

17:54

P1: KAE Chapter-07

CUFX159-Virgin

0 521 85648 5

August 14, 2007

7 Continuous Struts

7.1 Introduction In this chapter, we consider the dynamics of a thin elastic strut including axial effects, arising primarily from one of two situations: r the axial load is applied externally (including postbuckling), or r deformation is sufficient to cause coupling between axial and bending behavior (the membrane effect). In the first section, attention is focused on a traditional approach to setting up the equations of motion (by means of D’Alembert’s principle) for the simple case based on engineering beam theory (Euler–Bernoulli) with the addition of axial loads. The resulting partial differential equation of motion is then separated into temporal and spatial ordinary differential equations and the response analyzed for various magnitudes of the axial load [1, 2]. Then an energy approach is used together with Rayleigh’s method [3]. In this case, additional terms are retained in the potential energy to allow postbuckled effects to be analyzed [4] and the effect of initial geometric imperfections are included. An alternative approach is developed based on a simple application of Hamilton’s principle, and in this instance stretching effects are also included and a solution developed by use of Galerkin’s method. This approach will be similar to that used in the previous chapter on strings but now bending strain energy enters into the analysis (as well as compressive axial loading). In the final part of the chapter we consider the dynamics of struts that are loaded by gravity through self-weight. The next chapter will then continue the study of axially loaded members but with the scope opened to include a wider class of problem.

7.2 Basic Formulation In this section, we develop the governing equation of motion for a thin, elastic, prismatic beam subject to a constant axial force. In Fig. 7.1(a) a schematic of the beam is shown. It has mass per unit length m, constant flexural rigidity EI, and is subject to an axial load P. The length is L, the coordinate along the beam is x, and the lateral (transverse) deflection is w(x, t). In part (b) is shown an element of the beam between locations x and x + x, which is subject to the D’Alembert forces R(x, t) = m∂2 w/∂t2 .

(7.1) 111

17:55

P1: KAE Chapter-07

CUFX159-Virgin

0 521 85648 5

August 14, 2007

112

Continuous Struts

(a) L x

EI, m

P

w(x,t)

(b)

x w1

M

Figure 7.1. A schematic of a thin elastic beam.

w2

R(x,t) x

P M+ M

0

S

P Q(x,t) x x/2

F(x,t) x

S+ S

At this point, we make the assumption that the angle through which the section rotates (w/x) is small and we neglect axial and rotary inertia [5]. Taking moments about 0 we get x x − (S + S) + M − (M + M) + Pw = 0, 2 2 and as x → 0 we have ∂M ∂w S− +P = 0. ∂x ∂x Summing forces in the vertical direction we also have S

(S + S) − S + F (x, t)x − R(x, t)x = 0,

(7.2)

(7.3)

(7.4)

and again passing to the limit x → 0, we have ∂2 w ∂S + F (x, t) − m 2 = 0. ∂x ∂t

(7.5)

Differentiating Eq. (7.3) we obtain ∂S ∂2 M ∂2 w − + P 2 = 0, 2 ∂x ∂x ∂x

(7.6)

and eliminating ∂S/∂x between Eqs. (7.5) and (7.6) gives m

∂2 w ∂2 M ∂2 w − + P = F (x, t), ∂t2 ∂x2 ∂x2

(7.7)

and finally, using the familiar expression from engineering beam theory [6, 7], ∂2 w = −M, ∂x2

(7.8)

∂2 w ∂2 w ∂4 w + P + m = F (x, t). ∂x4 ∂x2 ∂t2

(7.9)

EI we arrive at the governing equation: EI

17:55

P1: KAE Chapter-07

CUFX159-Virgin

0 521 85648 5

August 14, 2007

7.2 Basic Formulation

113

This linear partial differential equation is a key equation in this book and can be rewritten in shorthand as EIw + Pw + mw¨ = F (x, t)

(7.10)

EIwxxxx + Pwxx + mwtt = F (x, t),

(7.11)

or

where derivatives are signified by primes and subscripts in Eqs. (7.10) and (7.11), respectively, and, given appropriate boundary conditions at x = 0, L and initial conditions, can be solved by use of standard methods as outlined in Chapter 4. Before we consider the specific aspects of the solution, the utility of nondimensionalizing the governing equation of motion is mentioned. Nondimensionalization is useful because it reduces the number of parameters and allows a more consistent comparison of behavior. In this chapter, we are focused on the unforced problem (F = 0), and, by defining the following parameters, (7.12) x¯ = x/L, w¯ = w/L, t¯ = t EI/(mL4 ), p = PL2 /(EI), we can rewrite Eq. (7.9) as ∂4 w¯ ∂2 w¯ ∂2 w¯ + p + = 0. ∂x¯ 4 ∂x¯ 2 ∂t¯2

(7.13)

Throughout this book we shall look at solutions to this type of equation and often incorporate nonlinearities into the analysis, in which case a variety of approximate techniques (of the type outlined in Chapter 4) will be utilized. Sometimes the free parameters will be chosen such that they are further normalized (e.g., p might be related to a critical buckling load). However, we start by looking at the simple freevibration case for which there is good access to analytical solutions. 7.2.1 The Response We initially consider the free-vibration problem and assume that F = 0 and that the motion consists of a function W(x) that varies with Y(t) such that w(x, t) = W(x)Y(t).

(7.14)

Placing this back into Eq. (7.9) leads to EI

d4 W d2 W d2 Y Y + P Y = −mW . dx4 dx2 dt2

(7.15)

That is, after dividing by WY (which is not zero if w is not zero), we have an equation separated into spatial (x) and temporal (t) parts, and thus the ratio on each side must be a constant (which we label −ω2 ) m d2 Y EI d4 W P d2 W = − − = −ω2 . Y dt2 W dx4 W dx2

(7.16)

17:55

P1: KAE Chapter-07

CUFX159-Virgin

0 521 85648 5

August 14, 2007

114

Continuous Struts

Thus we have two ordinary differential equations—one in space that we can solve by using the appropriate boundary conditions and that describes the mode shapes (which are half sine waves in the simply supported case); the other in time that we can solve with the appropriate initial conditions and contains the frequency information (simple harmonic motion). It is not uncommon for the vibration of mechanical systems to be dominated by the mode with the lowest frequency, but we consider the full system at first before developing approximate techniques. 7.2.2 The Temporal Solution We might expect the second-order ordinary differential equation in time to have oscillatory solutions (given positive values of flexural rigidity, etc.). However, we anticipate that the dependence of the form of the temporal solution will depend on the magnitude of the axial load [8, 9]. To be a little more specific, before going on to consider the more general boundary conditions, let us suppose we have ends that are pinned (no deflection and no resistance to rotation), that is, the deflection (w) and bending moment (−EI∂2 w/∂2 x) are zero at x = 0 and x = L. In the general case, we would assume an exponential form for the solution, but with these relatively convenient boundary conditions, we can take w(x, t) =

∞

Y(t) sin

n=1

nπx . L

(7.17)

We can obtain the temporal part of the solution by assuming Yn (t) = An eiωn t ,

(7.18)

and substituting into Eq. (7.9) leads to 2 2 ∞ n2 π2 nπ nπx iωn t 2 EI 2 − P e − mωn An sin = 0. L L2 L

(7.19)

n=1

Clearly, the term in the square brackets must vanish for a nontrivial solution so that PL2 n4 EIπ4 ω2n = 1 − . (7.20) mL4 n2 EIπ2 If we define the following parameters p n = n2 EIπ2 /L2 , Eq. (7.20) becomes

ω¯ 2n = n4 EIπ4 /mL4 ,

ωn = ±ω¯ n 1 − p, ¯

(7.21)

(7.22)

where p¯ = P/p n , and we see that the nature of the solution depends crucially on the discriminant. Making use of the Euler identities, we consider the following four

17:55

P1: KAE Chapter-07

CUFX159-Virgin

0 521 85648 5

August 14, 2007

7.2 Basic Formulation

115

cases, in which An and Bn are constants obtained from the initial conditions: r If p¯ = 0, then Yn (t) = An cos ω¯ n t + Bn sin ω¯ n t,

(7.23)

and we observe simple harmonic motion, a familiar result from linear vibration theory. r If 0 < p¯ < 1, then Yn (t) = An cos ωn t + Bn sin ωn t,

(7.24)

where ωn is given by Eq. (7.22) and is real, and simple harmonic motion results. Any perturbation will induce oscillatory motion about equilibrium. Assuming no damping, the response neither grows nor decays. This includes the response for a tensile axial load, that is, p¯ < 0. r If p¯ = 1, Eq. (7.22) has a double-zero root, and then the solution can be written as Yn (t) = An + Bn t,

(7.25)

and the motion grows linearly with time (this is a special case). r If p¯ > 1, the roots of (7.22) are purely imaginary and then Yn (t) = An cosh ω¯ n t + Bn sinh ω¯ n t,

(7.26)

and the motion grows exponentially with time. Typical examples of these cases are shown in Fig. 7.2(a) in which the natural frequency in the absence of axial load was taken as unity [10], and hence a natural period of 2π. Also shown in this figure are the stability of equilibrium, part (b), and effective natural frequency (squared), part (c), as a function of axial load. Let us focus attention on the lowest natural frequency and its corresponding mode (n = 1). With no axial load (p¯ = 0) we obtain ω1 = ω¯ 1 . However, as the axial load increases the natural frequency decreases according to Eq. (7.22), that is, we observe a linear relationship between the magnitude of the axial load and the square of the natural frequency [see Fig. 7.2(c)]. Any nonzero initial conditions result in bounded motion, and we may consider this to be a stable situation (at least in the sense of Lyapunov). When p¯ → 1, ω1 vanishes and the solution ceases to be oscillatory [the linearly increasing (constant-velocity) solution shown in Fig. 7.2]. Any inevitable perturbation will cause the system to become unstable. This type of instability is monotonic because, locally, the deflections grow in one direction (determined by the initial conditions). This type of behavior is sometimes referred to as divergence. The higher modes (n > 1) will exhibit oscillations (in theory) but the important practical information has been gained, that is, typical behavior is dominated by the lowest mode.

17:55

P1: KAE Chapter-07

CUFX159-Virgin

0 521 85648 5

August 14, 2007

116

Continuous Struts 10 (a)

Y

_

p >1

8 6

_

p =1

4 2

_

p=0 _

0 −2

p 10, the beamlike bending modes ¨ start to dominate, and in this region Flugge shell theory provides more accurate solutions [46]. In the other extreme, the behavior tends to be more platelike. Other studies have included the effects of initial imperfections [53] and composites [54, 55]. Equation (10.101) thus gives the familiar-looking result shown in Fig. 10.16 in which both the frequency and axial load are nondimensionalized. The linearity in the axial-force–frequency (squared) relation was encountered earlier in a variety of systems for which the buckling and vibration modes were similar. It is the relative simplicity of this relation that provides compelling motivation for nondestructive testing purposes, that is, using frequencies to predict buckling, and this is the subject of the next chapter.

18:0

P1: KAE Chapter-10

CUFX159-Virgin

0 521 85648 5

August 14, 2007

212

Plates

1

−Nx

0 n = 11

0

1

2

Figure 10.16. Fundamental natural frequency for a specific cylinder as a function of axial load.

Other Issues

The stiffening effect that is due to a spinning circular plate [56] is a problem of practical interest and is related to the spinning beam analysis in Chapter 7. Certain plate problems are conveniently solved by the methods of Lagrange multipliers [57] and finite differences [58]. Later chapters will also revisit shells in terms of step loading and parametric excitation.

References [1]

S.P. Timoshenko and S. Woinowsky-Krieger. Theory of Plates and Shells, 2nd ed. McGraw-Hill, 1968. [2] A.W. Leissa. Vibration of plates. Technical Report SP–160, NASA, 1969. [3] T. von Karman, E.E. Sechler, and L.H. Donnell. The strength of thin plates in compression. Transactions of ASME, 54:53–7, 1932. [4] S. Levy. Bending of rectangular plates with large deflections. Technical Report 737, NACA, 1942. [5] S. Levy. Buckling of rectangular plates with built-in edges. Journal of Applied Mechanics, 9:171–4, 1942. [6] H.-N. Chu and G. Herrmann. Influence of large amplitudes on free flexural vibrations of rectangular elastic plates. Journal of Applied Mechanics, 23:532–40, 1956. [7] S.F. Bassily and S.M. Dickinson. Buckling and lateral vibration of rectangular plates subject to in-plane loads—a Ritz approach. Journal of Sound and Vibration, 24:219–39, 1972. [8] S.M. Dickinson. The buckling and frequency of flexural vibration of rectangular isotropic and orthotropic plates using Rayleigh’s method. Journal of Sound and Vibration, 61:1–8, 1978. [9] C.F. Ng and R.G. White. Dynamic behavior of postbuckled isotropic plates under in-plane compression. Journal of Sound and Vibration, 120:1–18, 1988. [10] G.H. Bryan. On the stability of a plane plate under thrusts in its own plane with applications to the buckling of the sides of a ship. Proceedings of the London Mathematical Society, 22:54–67, 1891.

18:0

P1: KAE Chapter-10

CUFX159-Virgin

0 521 85648 5

August 14, 2007

References [11] S. Ilanko. Vibration and post-buckling of in-plane loaded rectangular plates using a multiterm Galerkin’s method. Journal of Applied Mechanics, 69:589–92, 2002. [12] A.C. Ugural. Stresses in Plates and Shells. McGraw-Hill, 1999. [13] G. Herrmann and J. Shaw. Vibration of thin shells under initial stress. Journal of Engineering Mechanics, 91:37–59, 1965. [14] J.F. Doyle. Nonlinear Analysis of Thin-Walled Structures. Springer, 2001. [15] K.K. Kapur and B.J. Hartz. Stability of plates using the finite element method. Journal of Engineering Mechanics, 92:177–95, 1966. [16] R.G. Anderson, B.M. Irons, and O.C. Zienkiewicz. Vibration and stability of plates using finite elements. International Journal of Solids and Structures, 4:1031–55, 1968. [17] B.A. Boley and J.H. Weiner. Theory of Thermal Stresses. Wiley, 1960. [18] D.J. Johns. Thermal Stress Analysis. Pergamon, 1965. [19] J. Marcinowski. Postbuckling behaviour of rectangular plates in axial compression. Archives of Civil Engineering, 45:275–88, 1999. [20] K.D. Murphy. Theoretical and experimental studies in nonlinear dynamics and stability of elastic structures. Ph.D. dissertation, Duke University, 1994. [21] R.E. Kielb and L.S. Han. Vibration and buckling of rectangular plates under in-plane hydrostatic loading. Journal of Sound and Vibration, 70:543–55, 1980. [22] R.E. Kielb. Thermal buckling of uniform rectangular plates. Technical Report, U.S. Air Force Wright-Patterson, ASD-TR-75-37, 1976. [23] T.R. Tauchert. Thermally induced flexure, buckling, and vibration of plates. Applied Mechanics Reviews, 44:347–60, 1991. [24] K.D. Murphy, L.N. Virgin, and S.A. Rizzi. The effect of thermal prestress on the free vibration characteristics of clamped rectangular plates: Theory and experiment. Journal of Vibration and Acoustics, 119:243–9, 1997. [25] M. Stein. Loads and deformation of buckled rectangular plates. Technical Report R–40, NASA, 1959. [26] D.G. Schaeffer and M. Golubitsky. Boundary conditions and mode jumping in the buckling of rectangular plates. Communications in Mathematics and Physics, 69:209–36, 1979. [27] R. Maaskant and J. Roorda. Mode jumping in biaxially compressed plates. International Journal of Solids and Structures, 29:1209–19, 1991. [28] H. Chen and L.N. Virgin. Finite element analysis of postbuckling dynamics in plates: Part I: An asymptotic approach. International Journal of Solids and Structures, 43:3983–4007, 2006. [29] H. Chen and L.N. Virgin. Finite element analysis of postbuckling dynamics in plates: Part II: A nonstationary analysis. International Journal of Solids and Structures, 43:4008–27, 2006. [30] K.D. Murphy, L.N. Virgin, and S.A. Rizzi. Characterizing the dynamic response of a thermally loaded, acoustically excited plate. Journal of Sound and Vibration, 196:635–58, 1996. [31] K.D. Murphy, L.N. Virgin, and S.A. Rizzi. Experimental snap-through boundaries for acoustically excited, thermally buckled plates. Experimental Mechanics, 36:312–7, 1996. [32] L.N. Virgin. Parametric studies of the dynamic evolution through a fold. Journal of Sound and Vibration, 110:99–109, 1986. [33] R.V. Southwell. On the analysis of experimental observations in problems of elastic stability. Proceedings of the Royal Society of London, 135A:601–16, 1932. [34] K.D. Murphy, L.N. Virgin, and S.A. Rizzi. Free vibration of thermally loaded panels including initial imperfections and post-buckling effects. Technical Memorandum 109097, NASA, 1994.

213

18:0

P1: KAE Chapter-10

CUFX159-Virgin

214

0 521 85648 5

August 14, 2007

Plates [35] M. Uemura and O. Byon. Secondary buckling of a flat plate under uniaxial compression – Part 1: Theoretical analysis of simply supported flat plate. International Journal of Non-Linear Mechanics, 12:355–70, 1977. [36] E. Riks, C.C. Rankin, and F.A. Brogan. On the solution of mode jumping phenomena in thin-walled shell structures. Computer Methods in Applied Mechanics and Engineering, 36:59–92, 1996. [37] H. Troger and A. Steindl. Nonlinear Stability and Bifurcation Theory: An Introduction for Engineers and Applied Scientists. Springer-Verlag, 1991. [38] P.R. Everall and G.W. Hunt. Mode jumping in the buckling of struts and plates: A comparative study. International Journal of Non-Linear Mechanics, 35:1067–79, 2000. [39] G.W. Hunt and P.R. Everall. Arnold tongues and mode-jumping in the supercritical post-buckling of an archetypal elastic structure. Proceedings of the Royal Society of London A, 445:125–40, 1999. [40] T. Nakamura and K. Uetani. The secondary buckling and post-buckling behaviors of rectangular plates. International Journal of Mechanical Sciences, 21:265–86, 1979. [41] E.J. Doedel. AUTO – Software for continuation and bifurcation problems in ordinary differential equations. California Institute of Technology, 1986. [42] E.J. Doedel, A.R. Champneys, T.F. Fairgrieve, Y.A. Kuznetsov, B. Sandstede, and X.J. Wang. Auto97: Continuation and bifurcation software for ordinary differential equations. Technical Report, Department of Computer Science, Concordia University, Montreal, Canada, 1997 (available by FTP from ftp.cs.concordia.ca in directory pub/doedel/auto). [43] W.J. Supple. On the change in buckle pattern in elastic structures. International Journal of Mechanical Sciences, 10:737–45, 1968. [44] J.M.T. Thompson and H.B. Stewart. Nonlinear Dynamics and Chaos, 2nd ed. Wiley, 2002. [45] H. Chen. Nonlinear analysis of post-buckling dynamics and higher order instabilities of flexible structures. Ph.D. dissertation, Duke University, 2004. [46] A.W. Leissa. Vibration of shells. Technical Report SP-288, NASA, 1973. [47] E.H. Dowell. Aeroelasticity of Plates and Shells. Noordhoff, 1975. [48] T. von Karman and H.S. Tsien. The buckling of thin cylindrical shells under axial compression. Journal of the Aeronautical Sciences, 8:303–12, 1941. [49] L.H. Donnell. A new theory for the buckling of thin cylinders under axial compression and bending. Transactions of ASME, 56:796–806, 1934. [50] R.D. Blevins. Formulas for Natural Frequencies and Mode Shapes. Van Nostrand Rheinhold, 1979. [51] K. Forsberg. A review of analytical methods used to determine the modal characteristics of cylindrical shells. NASA Report CR-613, Lockheed Aircraft Company, CA, September 1966. [52] S.B. Batdorf. A simplified method of elastic-stability analysis for thin cylindrical shells. Technical Report 874, NACA, 1947. [53] A.E. Armenakas. Influence of initial stress on the vibrations of simply supported circular cylindrical shells. AIAA Journal, 2:1607–12, 1964. [54] H.S. Shen. Thermomechanical post-buckling analysis of imperfect laminated plates using a higher-order shear-deformation theory. Computers and Structures, 66:395–409, 1998.

18:0

P1: KAE Chapter-10

CUFX159-Virgin

0 521 85648 5

August 14, 2007

References [55] C.-S. Chen and C.-P. Fung. Non-linear vibration of initially stressed hybrid composite plates. Journal of Sound and Vibration, 274:1013–29, 2004. [56] R.G. Parker and C.D. Mote. Tuning of the natural frequency spectrum of a circular plate by in-plane stress. Journal of Sound and Vibration, 145:95–110, 1991. [57] J.H. Ginsberg. Advanced Engineering Dynamics. Cambridge University Press, 1995. [58] F. Bleich. Buckling Strength of Metal Structures. McGraw-Hill, 1952.

215

18:0

P1: KAE Chapter-11

CUFX159-Virgin

0 521 85648 5

August 14, 2007

11 Nondestructive Testing

11.1 Introduction Previous chapters have repeatedly illustrated the often well-defined relation between axial load and natural frequency. In some cases, for example, a simply supported strut, the equivalence of the buckling and vibration modes results in an exactly linear relation between the axial load (providing it is less than critical) and square of the effective natural frequency: ω2 P =1− . 2 Pcr ω0

(11.1)

In other cases, this relation is very nearly linear. For example, consider a simple cantilever. The fundamental frequencies in bending have the mode shapes λi x λi x λi x λi x W(x) = cosh − cos − σi sinh − sin , (11.2) L L L L with σ1 = 0.7341, λ1 = 1.8751 for the lowest mode and corresponding frequency [i.e., ω1 = 3.516 EI/(mL4 )]. The buckling mode for a cantilever with an end load is given by πx W(x) = 1 − cos , (11.3) 2L with a critical load of Pcr = π2 EI/(4L2 ). These normalized shapes are plotted in Fig. 11.1. They are close, but unlike the pinned–pinned (and some sliding boundary conditions) case, they are not equal. However, also plotted in this figure (as the dashed curve) is the buckling mode shape corresponding to the cantilever subject to self-weight. This shape is computed numerically and the critical parameter was established as hcr = 1.986 in Section 7.9 (and equivalent to α = −7.837). This is much closer to the fundamental mode of vibration, and in fact, the difference between them is never more than 1%. The equivalence of the vibration and buckling mode shapes results in the linear relation between axial load and frequency, that is, the extent to which the vibration mode shape is changed by axial loading. Thus we have a frequency (squared) versus load relation that is closer to linearity for the vibrations of a cantilever subject to self-weight than an end load. However, even for the end-loaded cantilever case a simple use of abaqus shows that when (P/Pcr ) = 0.51875 we obtain a lowest natural 216

18:2

P1: KAE Chapter-11

CUFX159-Virgin

0 521 85648 5

August 14, 2007

11.1 Introduction

217

1 0.8

Buckling mode (self-weight)

0.6

First vibration mode

0.4 0.2 0

0

Buckling mode (end load)

0.2

0.4

0.6

0.8

1

Figure 11.1. Comparison of vibration and buckling mode shapes for a uniform cantilever.

frequency of (ω/ω0 )2 = 0.43804 compared with an estimate of (ω/ω0 )2 = 0.41825 suggested by Eq. (11.1). Thus Eq. (11.1) brings the possibility of using dynamics as a means of assessing axial-load effects, including the prediction of buckling [1–3]. In static buckling tests it is often unavoidable that specimens are destroyed during the experimental procedure (often the result of plastic deformation during large deflections). The Southwell plot is a related static approach that also exploits a linear extrapolation to predict buckling nondestructively [4]. Correlation studies between dynamic response and stiffness are also used to determine the actual boundary conditions as well [5, 6]. The simplicity of this relation can be used to nondestructively test axially loaded slender structural elements through monitoring of dynamic response [7–12].

11.1.1 The Southwell Plot In Eq. (7.57) we saw how a small initial geometric imperfection tended to amplify the lateral deflections of a strut, especially as the buckling load is approached. Suppose we have a simply supported beam as shown in Fig. 11.2(a). We can measure the lateral deflections from the initially bent configuration, w0 , but here we measure the total lateral deflection, w, from the straight configuration. If we assume the initial deflection is in the form of a half-sine wave of amplitude Q0 , we can plot the amplification effect [Eq. (7.57)] as shown in Fig. 11.2(b). This, of course, assumes small deflections. However, in an experimental context what we would actually measure would typically be the lateral deflection over and above the initial deflection, which we can call δ = w − w0 and thus (at the midpoint of the strut) δ=

Q0 P/PE − Q0 = Q0 . 1 − P/PE 1 − P/PE

(11.4)

Equation (11.4) can be arranged in the form δ δ Q0 = + , P PE PE

(11.5)

18:2

P1: KAE Chapter-11

CUFX159-Virgin

0 521 85648 5

August 14, 2007

218

Nondestructive Testing (a) w w0

P

P

P/PE

(b)

0.8

(c)

/P

1/ PE

0.6

}

0.4 0.2

Q 0 /PE

Q 0 = 0.05 0 0.1

0

0.2

0.3

0.4

0.5

Q

Figure 11.2. The Southwell plot: (a) strut geometry with an initial imperfection, (b) axial-load– lateral-deflection relation, and (c) Southwell plot.

so that if we plot δ/P as a function of δ we get a straight line in which the intercept is given by Q0 /PE and the slope is given by 1/PE . Southwell [4] recognized the usefulness of this approach to determine both the critical load and initial imperfection, and this is shown schematically in Fig. 11.2(c). A Southwell plot based on experimental data is shown in Fig. 11.3 [13]. Here, the data suggest a critical load (slope) in the vicinity of 87 N and an imperfection of ≈ 0.05 or about 3 deg; values not unreasonable when compared with the data presented in Fig. 5.10(c). Although there are limitations to this approach, the key utility here is that the linear relation allows for extrapolation. This provides some

δ (radian)

0.4

0.2 δ/P (radian N−1) ε = 0.05

0

0.002

0.004

0.006

Figure 11.3. The Southwell plot obtained with experimental data taken from the system shown in Fig. 5.9. Adapted from Croll and Walker [13].

18:2

P1: KAE Chapter-11

CUFX159-Virgin

0 521 85648 5

August 14, 2007

11.1 Introduction

219

(a)

Figure 11.4. The effect of axial-load direction on natural frequencies.

(b)

1

g

motivation for exploring related concepts in the dynamic testing of structures, in which vibration testing in situ is a well-established procedure (e.g., in structural health monitoring [14]).

11.1.2 Examples The relation between axial load and lateral vibrations and its potential use for nondestructive evaluation purposes goes back to Sommerfeld [15]. He made the simple observation that the fundamental natural frequencies of the two systems shown in Fig. 11.4 were quite different (with ω2 > ω1 ). He concluded that the greater the compressive stress, the lower the natural frequency of lateral vibration. With tensile stress, an increase in natural frequency was observed. Furthermore, in the former case it was noted that the frequency dropped to zero as the compressive load approached its critical value. We can conduct a simple analysis of this system by using the methods developed earlier in this book. Suppose the strut has a length l, end mass m, flexural rigidity EI, √ and oscillates in gravity g. Introducing the nondimensional parameter α = mg/EI, we can readily show that when the strut has the mass placed at its top [Fig. 11.4(a)], the natural frequency is given by [16] ω = gα/(tan αl − αl), (11.6) which remains positive until buckling occurs at mc = π2 EI/4gl 2 . When the strut is turned upside down [Fig. 11.4(b)] the natural frequency becomes ω = gα/(αl − tanh αl). (11.7) Thus, suppose we have a mass that corresponds to about the half the critical load, √ 2 2 that is, m = (π EI)/(8gl ); then α = π/(2 2l) and the natural frequency of the sys√ √ tem in part (a) would be 1.106 g/l, as opposed to 1.904 g/l for the system in part (b). In fact, even a mass that causes buckling in part (a) would result in oscillations √ of frequency 1.55 g/l in the inverted system [part (b)]. It is quite easy to demonstrate this experimentally [16]. Consider a simple polycarbonate cantilever strip, as shown in the inset to Fig. 11.5. If we consider the beam mass as being negligible compared with the concentrated mass added to its free end, then the theoretical results given by Eqs. (11.6) and (11.7) apply. For the specific case of a strip with L = 0.181 m, a second moment of area I = 1.903 × 10−12 m4 , and

2

18:2

P1: KAE Chapter-11

CUFX159-Virgin

0 521 85648 5

August 14, 2007

220

Nondestructive Testing 2 2

1/2

(mgL /EI)

downward horizontal upright

1.5

1

0.5

0 0

0.2

0.4

0.6

0.8 1 1/2 f/(g/L)

Figure 11.5. A simple experimental cantilever and its frequency variation with end load for different orientations.

Young’s modulus E = 2.4 × 109 N/m2 , we add increments of end mass and measure the resulting fundamental frequency. The nondimensionalized results are shown in Fig. 11.5 together with the theoretical results. Also included is the simple horizontal √ cantilever result based on the lumped stiffness approximation, that is, ω = Ke /m in which Ke = 3EI/L3 [17]. The effect of the mass of the beam itself can be included either in this lumped analysis or by use of a more sophisticated approach (adding a little distributed mass would tend to shift the data points up slightly), but the trend describing the effect of gravity (and hence axial loading) is clear. We also note at this point that experiments on cantilevers with self-weight loading are relatively easy to set up. The results from tests with other boundary and loading conditions, for example, in a testing machine, need more careful interpretation, as discussed in Section 7.2. As a reference point the typical amount of end mass the strut was able to withstand before appreciably starting to droop to one side was about 27 g. The Euler load for a cantilever is EIπ2 /(4L2 ), which gives a value of mc = 35 g, and given the inevitable initial imperfections in the system this magnitude is not unreasonable. Furthermore, when no end mass was added the strut vibrated with a measured natural frequency of a little over 6 Hz (in fact 6.075, 6.2375, and 6.4 in its upright, horizontal, and downward orientations). The theory of continuous elastic beams coveredin Chapter 7 listed a fundamental natural frequency for a cantilever of ω = 3.52 EI/mL4 and with the total mass of the strip measured at mL = 5.94 × 10−3 kg this corresponds to a predicted frequency of 6.38 Hz. We can also reinterpret Fig. 7.18 at this point. Recall that this plot referred to a slender continuous strut subject to self-weight loading (rather than a concentrated end mass). Plotting the original “weight” parameter |α| as a function of frequency squared gives the results shown in Fig. 11.6(a). In the absence of gravity we would

18:2

P1: KAE Chapter-11

CUFX159-Virgin

0 521 85648 5

August 14, 2007

11.1 Introduction

221 (b)

(a) 25

L /Lc

1.4 1.3

Downward

1.2

20

1.1 g

1

15 upright

0.9 0.8

10 downward

0.7 0.6

5

upright

0.5 0

0

5

10

15

20

25

30

35

40

45

50

0.4

0

0.5

1

1.5

2

2.5

3

3.5

4

4.5

Frequency (Hz)

Figure 11.6. The frequencies of a simple but heavy experimental cantilever. The solid line represents the upright case, and the dashed line represents the hanging-down orientation: (a) |α| versus the fundamental frequency squared and (b) alternative plot of the same results.

expect the frequency to be proportional to the inverse of the length squared, and this case is shown too. However, in the upright configuration, as the critical length is approached the stiffness is diminished such that the frequency drops to zero at the critical length. If we plot the dimensional frequency versus the length (normalized by the critical length), we get the results shown in Fig. 11.6(b). However, not all the data from part (a) are included because of different thicknesses. The near linear relationship is, of course, a suitable form for extrapolation. Thus we might measure the fundamental natural frequency for a number of different α values (specifically changing the length L) and fitting a straight line to this data [Fig. 11.6(b)] we would predict buckling in the vicinity of α ≈ 7.8. Recall that in this plot the “weight” α is a nondimensional parameter given by α = mgL3 /(EI), and hence with mass per unit length of 0.0147 kg/m, cross-sectional dimensions of 25.4 mm × 0.508 mm, and Young’s modulus of 2.4 GPa we get the actual length at buckling of about 0.33 m. The cantilever that hangs down never buckles as the length increases of course. We expect the natural frequency of a system to reduce if more mass is added to it. But what the preceding setup shows is that if the mass acts through gravity then it may reduce the stiffness of the system, and it is this tendency that can be exploited in terms of nondestructive (stability) testing. In other words, if we refer back to Fig. 11.5 and consider a fixed value of the end mass toward higher values (where gravity has more effect) then the effect of orientation (and whether the effective axial load is compressive or tensile) is apparent. A good deal of the earlier material in this book has highlighted ways in which this trend may be relatively simple [18]. If the trend is linear then it also provides the possibility of predicting the elastic buckling not only from, in principle, measurement of the lower natural frequency at two distinct axial-loading conditions but even when one or more of these loads is tensile. In a practical (experimental) context, the boundary conditions may not be

5

18:2

P1: KAE Chapter-11

CUFX159-Virgin

0 521 85648 5

August 14, 2007

222

Nondestructive Testing

Phase 7.00 102 Accelerometer/ force

7.00 10-2 2.00 101

Frequency (Hz)

2.56 102

Figure 11.7. Frequency content of a prismatic beam showing the shift in resonant frequencies under the application of axial loading. Reproduced with permission from Elsevier [19].

known. Consider the results shown in Fig. 11.7, which were described in Livingston et al. [19]. This frequency spectrum was obtained from a prismatic beam by experimental modal analysis as part of a larger study in the context of system identification and parameter estimation. Over this frequency range the lowest three frequencies are quite distinct. The solid line corresponds to (practically) zero axial loading, with the dotted line showing the shift to higher frequencies when the beam is subject to a tensile axial load (approximately of a similar magnitude to that of the Euler buckling load with boundary conditions somewhat intermediate between clamped and pinned) [20, 21].

11.2 Some Background As mentioned in the introduction to this chapter, the idea of using dynamic behavior to predict buckling has received attention from a number of researchers over the years. However, major contributions were made by Massonet [1], who considered a variety of structural systems from a theoretical standpoint, and Lurie [2], who showed the utility of this approach including experiments. Even when the mode of vibration and buckling mode are not identical the load–frequency (-squared) relation may be almost linear. Lurie used an energy approach to show that an upper limit for the frequency of axially loaded thin beams resulted in a relation l 2 l dx mω2 0 w2 dx P 0 dw dx 1≥ l d2 w 2 + l d2 w 2 . EI dx EI dx 2 2 0

dx

0

dx

(11.8)

18:2

P1: KAE Chapter-11

CUFX159-Virgin

0 521 85648 5

August 14, 2007

11.2 Some Background

223

1

P/Pcr 0.8

0.6

0.4

0.2

0

0

0.2

0.4

0.6

0.8

2

(ω/ω1)

1

Figure 11.8. Upper and lower bounds of the frequency–load relation for a clamped–clamped beam.

For example, consider a clamped–clamped beam for which we know that Pcr = 2 2 4 4π EI/l and ωn (P = 0) = 22.373 EI/(ml ), and using the buckling mode shape 2πx w = A 1 − cos (11.9) l results in an expression

ω 1 ≥ 0.9635 ωn

2 +

P . Pcr

(11.10)

P . Pcr

(11.11)

Using the lowest mode of vibration [22] results in 1≥

ω ωn

2 + 0.9704

Relations (11.10) and (11.11) are plotted as inequalities in Fig. 11.8 together with the linear relation [Eq. (11.1)]. Thus we see the possibility of exploiting the linear relation between the square of the lowest natural frequency and the level of axial loading to extrapolate critical conditions [23]. Underlying General Theory. We have repeatedly looked at systems with a stiffness

that tended to be diminished by the presence of (compressive) axial loading. In terms of potential energy, we can write this as V = U(qi ) − ηkEk(q)i ,

(11.12)

18:2

P1: KAE Chapter-11

CUFX159-Virgin

0 521 85648 5

August 14, 2007

224

Nondestructive Testing

with the quadratic approximation in the form of the inner product V=

1 < q, (U − ηkEk)q >, 2

(11.13)

where the ηk (k = 1, 2, . . . , m) are independent parameters and U is the strain energy (symmetric and positive-definite). In terms of the equations of motion, we use Lagrange’s equation to obtain Mq¨ + (U − ηkEk)q = 0,

(11.14)

and assuming harmonic motion in the usual way, q = ueλt , we obtain the characteristic equation |Mλ2 + U − ηkEk| = 0.

(11.15)

For conservative systems, we have λ = iω with λ2 identified as the negative of the square of the natural frequencies (see Chapter 4). We are, of course, primarily interested in systems for which q = 0 represents a stable system but may become unstable (at buckling), and this occurs when one of the eigenvalues vanishes. Although instability may occur by means of a complex pair of eigenvalues in nonconservative systems (flutter, e.g., Beck’s problem, Section 7.8), in this book we remain primarily focused on the conditions under which a real eigenvalue vanishes at the divergence boundary. The relation between ω2 and ηk constitutes the characteristic curve (or surface, when more than one parameter is present). It has been proven [24, 25] that, for conservative systems with a trivial equilibrium state, any number of degrees of freedom, and equations of motion that are linear in the parameters ηk, the surface involving the fundamental frequency cannot have convexity toward the origin. Furthermore, it also follows that the fundamental surface is a plane (or straight line for a system with a single parameter) if the matrices M, U, and Ek can be reduced to a diagonal form simultaneously. A useful implication of this convexity property (and of obvious usefulness in the context of the present chapter) can be concluded. The divergence boundary is contained in the characteristic curve, and we obtain it by setting ω2 = 0. If a single parameter ξ (load) is acting on the system, then it is possible to obtain an (upper bound) estimate of the critical value. From Fig. 11.9, we see that if we know the frequencies at two values of the loading parameter ξ, ω211 (ξ1 ) and ω212 (ξ2 ), we can gain an estimate of frequencies at other loading values. Of course, if the characteristic curve is a straight line (e.g., if the equations uncouple) then this estimate will be exact. By extrapolating a line joining them, we obtain an upper bound on the critical value of ξ from the intersection with the ξ axis. Clearly, the accuracy of the estimate also depends on the location of the two reference points. This is an issue that will be discussed later.

18:2

P1: KAE Chapter-11

CUFX159-Virgin

0 521 85648 5

August 14, 2007

11.3 Snap-Through Revisited

225

u cr 2

Figure 11.9. Convexity of the characteristic curve and its implication for providing a lower bound.

1

2

2

2

12

11

11.3 Snap-Through Revisited In the previous section, we considered systems for which the fundamental equilibrium path was the trivial one. In snap-through buckling, we might still expect to monitor the lowest natural frequency to predict instability, but the nonlinearity of the underlying equilibrium curve can also have an influence. To quantify this, we go back to one of our standard forms from Chapter 3 in which we consider the dynamics of a system in the vicinity of a saddle-node bifurcation: ¨ − X2 − λ = 0, X

(11.16)

where both the deflection X and the load parameter λ are measured from the origin. Now suppose we have an equilibrium position (Xe ), and we wish to study the behavior of small oscillations about it. We can expand Eq. (11.16) in the usual way by replacing X with Xe + x that leads to x¨ − X2e − 2Xe x − x2 − λ = 0.

(11.17)

The x2 term can be dropped because it is small, and because of equilibrium we also have −X2e − λ = 0, and thus we are left with x¨ − 2Xe x = 0.

(11.18)

This system has the natural frequency ω=

−2Xe = +2(−λ)1/2 ,

(11.19)

and thus, for large negative λ say, we observe a linear relation between the loading parameter and the fourth power of the natural frequency [26]. Effect of Damping. So far we have concentrated mainly on undamped systems.

In most of the mechanical systems of interest, there is usually a little energy

18:2

P1: KAE Chapter-11

CUFX159-Virgin

0 521 85648 5

August 14, 2007

226

Nondestructive Testing −4

0

−2

2

0

ω2

2

4

ω4 β=2

−0.5

−0.5

β=1 β=0

−1.0

β=1 −1.5

λ −2.0

β=2

−1.0

β=0

−1.5

λ −2.0

Figure 11.10. The effect of damping on the frequency–load-parameter relation.

dissipation, and we shall assume that this takes the form of a linear-viscous damping (see Section 3.5). Thus we consider X¨ + βX˙ − X2 − λ = 0.

(11.20)

Conducting an analysis similar to that of the previous section we arrive at relationships between the natural frequency (the harmonic factor in the decaying, oscillating motion) and load parameter of ω2 = ±2(−λ)1/2 − (β/2)2 , ω4 = 4(−λ) ± 4(−λ)1/2 (β/2)2 + (β/2)4 .

(11.21)

These expressions are plotted in Fig. 11.10 for three values of damping including the undamped case. We see that damping has the effect of causing the natural frequency to diminish to zero prior to buckling. For example, with a damping level of β = 2 (and assuming the damping coefficient is constant), we observe that oscillations will cease when the load reaches a value of about λ = −0.25. One way of thinking about this is to recall the standard expression for a damping ratio: ζ = c/(2mωn ), but now the stiffness is reducing and thus, although the damping coefficient is constant, the damping ratio increases such that damping effectively becomes critically damped (to use the definitions introduced in Section 3.1) just prior to the stiffness dropping to zero. We can again integrate the equation of motion while slowly sweeping through the load parameter (as was done in Chapter 3). For example, Fig. 11.11 shows nine trajectories generated for system equation (11.20) and with a constant value of the initial total energy with β = 0.5, λ evolved √ at the rate 30t, and the initial conditions prescribed by x(0) ˙ = 0.0, x(0) = − 300 + A, where A varied between −8 and 8 in increments of 2. A number of interesting features can be seen in this figure. Damping does indeed appear to make the oscillations die out prior to instability, although this drifting system is never, of course, quite in equilibrium. We also see that

18:2

P1: KAE Chapter-11

CUFX159-Virgin

0 521 85648 5

August 14, 2007

11.3 Snap-Through Revisited

227

x 5

–10

–8

–6

–4

t

–2 –5 –10 –15 –20 –25

Figure 11.11. Some trajectories plotted as time series as the system is swept toward the saddle-node bifurcation.

the larger-amplitude oscillations display a degree of asymmetry—this feature is not unexpected because for a given amplitude the motion evolves toward the instability as the potential-energy well shrinks, from one side. A simple experimental verification of this situation is shown in Fig. 11.12. Here, a flexible rod with an end mass in a heavily postbuckled configuration was subject to base rotation such that a saddle-node bifurcation was encountered. The base rotation can be thought of as control in our standard system of gravity acting on the mass. The system follows its complementary equilibrium path during which time the frequency of natural (superimposed) oscillations are measured. The jump at the saddle-node bifurcation is represented schematically as A–B in Fig. 11.12(b). The measured frequencies and their relation to the control parameter are shown plotted in Fig. 11.13. Part (b) shows some times series in which the parameter r is a measure of the rate at which the base is rotated. We see that raising the frequency to the fourth power provides a more linear relationship than for the second power with which to predict criticality [26].

(a)

(b) Snap-through

m

Deflection to left

A Complementary post-critical state

B g

Natural post-critical state

End mass m (held constant throughout)

A Base rotation

Unstable region B

Control parameter

Figure 11.12. (a) A flexible strut with an end mass and (b) control surface showing a transition through bifurcation [26].

18:2

P1: KAE CUFX159-Virgin

0 521 85648 5

August 14, 2007

228

Nondestructive Testing

r3 = 4 r1

ω4

Freq. (arbitrary units)

0

ω2

Control parameter (rotation) Λ

Displacement

Chapter-11

r2 = 2 r1

Decreasing frequency

r1

Control parameter (= base rotation, a linear function of time)

Figure 11.13. (a) Measured frequencies for the flexible strut rotated through a saddle-node bifurcation and (b) some sample time series with different rates of rotation [26].

The concept of using the reduction in frequency as a stability predictor has also been used in the context of secondary buckling [27]. Referring back to Fig. 5.21, it is apparent that once the Augusti model has buckled into its primary mode, it is the frequency of the second mode that can be used to infer the approach of the secondary bifurcation as this frequency tends toward zero.

11.4 Range of Prediction The previous section indicated that rather than there being a universal relation between frequency and load, which would make extrapolation fairly straightforward, we see that damping, changing boundary conditions, initial imperfections, and type of instability, all conspire to make predictions more difficult. For example, Lurie [2] pointed out the increasingly important role played by initial imperfections as the critical load is approached (especially for plates). An experimental example of this effect [28] is shown in Fig. 11.14(a). Part (b) shows the results from a laminated composite column in which Chailleux et al. [28] identified three distinct regions, with region II providing the most useful (linear) relation for prediction purposes. Also in part (b), the authors noted that with very low load levels they experienced some clearance in the boundary conditions. As pointed out by Lurie [2], it may be possible to test specimens in tension, in which case initial imperfections will have minimal influence. In both of these sets of results, we observe that the linear relation breaks down near buckling (a feature first observed in Fig. 7.6). Given that raising the frequency to either the second or fourth power might be a more appropriate predictor, it seems reasonable to raise the frequency to various powers in order to see how the subsequent curve might change from concave to convex, which then has clear implications for lower bound estimates. For example, the frequency–load relations for an elastic arch were given in Eqs. (8.76) and (8.77) for a shallow, and less shallow arch, respectively, and shown in Fig. 8.11. Plaut and Virgin [29] studied the effect of extrapolating frequency raised to various powers

18:2

P1: KAE Chapter-11

CUFX159-Virgin

0 521 85648 5

August 14, 2007

11.4 Range of Prediction

(a)

229

(b)

Pdyn = 117.5 kg 100

P (kg)

III

P (kg)

Pdyn = 45.7 kg II

10

0

2

f (Hz 2 )

5 103

I

1 104

f 2 (Hz 2)

Figure 11.14. The load–frequency (-squared) relation for (a) rectangular duraluminum plate and (b) laminated composite column. Adapted from Chailleux et al. [28].

with a special reference for the range over which data were measured. That is, by raising the frequency to various powers a value is reached whereby the relation shown in Fig. 8.11 changes from convex to concave, with the transition point providing a close-to-linear relationship. For example, Plaut and Virgin [29] used the criterion suggested by Singer et al. [12] using (numerical) data from the simple elastic arch considered earlier. Over a broad range of loading conditions, i.e, not necessarily close to buckling, we typically do not observe a (near) linear-frequency-squared–load relation. However, we can fit data to a relation of the form p = C − D r ,

(11.22)

in which p is the load and is the frequency ωf [from Eq. (8.76) or (8.77)] nondimensionalized by the frequency under zero load. They showed that the values of r varied according to the range of load levels considered but that a change in curvature of the (frequency)r versus load relation could be extracted to provide reasonable upper and lower bound estimates of the buckling load. That is, a value for r is sought such that it results in the most linear relation between p and (frequency)r . The following table shows an example for the case λ = 3 (i.e., the less shallow arch). N 4 5 6 7 8 9 10 11 12

C

D

r

11.04 9.30 9.25 9.35 9.38 9.46 9.48 9.53 9.59

11.04 9.32 9.27 9.37 9.40 9.47 9.48 9.52 9.57

2.15 2.69 2.71 2.69 2.65 2.61 2.60 2.57 2.52

18:2

P1: KAE Chapter-11

CUFX159-Virgin

0 521 85648 5

August 14, 2007

230

Nondestructive Testing

(a)

(b)

p/pb

p/pb

f

f

Figure 11.15. Frequency–load relations for an elastic arch: (a) λ = 2 and (b) λ = 3.

In this table, N represents the number of data points used in the fit and spreads roughly evenly from zero to critical loading, and the last column gives the exponent. In this nondimensionalization, the critical load (which corresponds to a bifurcation rather than to a limit point) occurs at p = 9.71. The data were subject to a nonlinearleast-squares fit and we observe a range of optimal exponents depending on the number of data points used. The shallow arch is a useful device for illustrating frequency–load effects and applicability in terms of predicting instability because the λ parameter (i.e., the rise of the arch), can change the nature of the critical point [30]. For example, in Fig. 8.11, when λ = 2, the lowest frequency drops to zero when the arch experiences a saddle-node bifurcation and the arch snaps through to its inverted position. For a less shallow arch (e.g., when λ = 3), there is bifurcation and the arch experiences an asymmetric buckling (with a full-sine mode). However, the relations between frequencies and the type of instability (for example, as described in Section 3.4), and the material earlier in this section, were based on local generic behavior. If we take a closer look at Fig. 8.11 and focus in on a close proximity to buckling we get the results shown in Fig. 11.15. Here, the load has been normalized such that buckling corresponds to p/p b = 1. Thus in the vicinity of the instability it is the frequency raised to the fourth power that is more useful for predicting the saddle-node [part (a)], and frequency squared for the branching bifurcation [part (b)].

11.5 A Box Column A notable piece of work on the dynamic nondestructive evaluation of structures can be found in Jubb et al. [31]. The authors conducted some tests on box columns, which provided a clever way of studying plates by incorporating simply supported edges. One of their main goals was to establish a means of assessing the effects of

18:2

P1: KAE Chapter-11

CUFX159-Virgin

0 521 85648 5

August 14, 2007

11.6 Plates and Shells

residual stresses on the stiffness, dynamics, and stability of a typical structure. They suggested using the following variation on the frequency ( f )–load (σ) theme: 2 σr σa f k + + = 1. (11.23) σcr σcr f0 In this expression, the residual stress, σr , is added to the external stress, and k is a constant (less than unity) that takes account of stress distribution. We have already encountered how the (ordering of) modes of vibration of a plate depend on axial load (see Fig. 10.3), and in the experiments of Jubb et al. [31] they chose an aspect ratio of 4. In the absence of axial load, if the frequency corresponding to four half-sine waves in the longitudinal direction (m = 4) is denoted by 1.0, the lower modes turn out to have relative frequencies of 0.610 (m = 3), 0.391 (m = 2), and 0.282 (m = 1). As the axial force increases, these frequencies decrease (linearly with frequency squared) but at different rates such that it is the m = 4 mode that drops to zero at buckling (i.e., P/Pcr = 1). The experimental results are shown in Fig. 11.16. The lowest four natural frequencies are plotted as functions of axial load. Two important points can be extracted from the results. First, the welded corners induce a degree of residual stress such that the initiation of an applied axial load does not correspond to zero axial load in the frequency–load relation. Second, a degree of postbuckling stiffening is apparent in each of the modes—this is a feature anticipated by the analysis of Section 10.4. Therefore it may be important to monitor the first few lowest natural frequencies in order to capture the appropriate buckling mode.

11.6 Plates and Shells A thorough body of work on nondestructive testing of cylindrical shells under axial loading by use of dynamic (vibration) characteristics is due to Singer and his colleagues [11, 12]. An example of this type of research is shown in Fig. 11.17. Here a series of tests was conducted on cylindrical shells to see if the lowest frequency of lateral vibrations could be useful in predicting buckling. The cylinder included rib stiffeners, which had the effect of reducing some of the imperfection sensitivity typically encountered in axially loaded shells. Figure 11.17(a) shows a conventional frequency-squared versus load plot. Part (b) shows the same data with the frequency raised to the 2.9th power [12]. Figure 11.18 shows a plot in which (lower-load-level) frequencies are raised to various powers and then extrapolated as suggested in Plaut and Virgin [29]. The simplicity of the frequency–load relationship (or a variation thereof) can also be exploited when more complex structures are studied. For example, consider the panel (representative of a typical aeronautical configuration) shown in the lower part of Fig. 11.19. An eigenanalysis was conducted by Williams et al. [32] using the computer program viconopt. This structure, which might be a typical component in an aircraft fuselage, has ends that are simply supported such that sinusoidal modes result in the longitudinal direction. These are characterized by half-wavelengths λ,

231

18:2

P1: KAE Chapter-11

CUFX159-Virgin

0 521 85648 5

August 14, 2007

232

Nondestructive Testing Appled axial load (kN) 0

200

600

400

800

Theoretical elastic buckling load = 1070 kN

1.0

(f m/f 0)

Actual buckling load = 754 kN

2

1000

0.5

m=4 0

0.5

m=3

0 0.5

m=2

0 0.5

m=1 0

0

0.4

0.2

0.6

0.8

1.0

P/Pcr

Figure 11.16. The four lowest frequencies for the box column plotted as functions of the applied axial load. Adapted from Jubb et al. [31].

2

5

f × 10 2 (Hz) 4

q

9

f × 10 q (Hz)

3

2

2 1 1 0 0

1000

2000

3000

4000

P(Kg)

0 0

1000

2000

3000

4000 P(Kg)

Figure 11.17. The lowest vibration frequency of an axially loaded cylinder: (a) frequency squared and (b) frequency raised to the power q = 2.9. Adapted from Singer et al. [12].

18:2

P1: KAE Chapter-11

CUFX159-Virgin

0 521 85648 5

August 14, 2007

11.6 Plates and Shells

233

q=1 q=2 q=3

1.0

f

q

0.5 q=4 q=5

0

1000

3000

2000

4000

P (kg) Pexp

Figure 11.18. The same frequencies as in Fig. 11.17 but raised to various powers and subsequently extrapolated (linearly) to the buckling load. Adapted from Singer et al. [12].

which are integral fractions, l/i, of the length of the structure, l. A set of eigenvalues (natural frequencies) can be computed for values of λ. Suppose we have a structure that is isotropic (but need not necessarily be so) and with the geometric properties shown in the lower part of the figure. The nota tion α = 1000/E and β = (4000ρ/E) are introduced, and Poisson’s rato is ν = 0.3. Suppose we wish to extract the natural frequencies below 6/βl when l = 6b. The computed characteristic curves shown in the upper part of Fig. 11.19 indicate natural frequencies where the vertical dashed lines intercept the curves, with the small markers indicating a frequency of 6/βl. Thus we observe the frequencies listed in the top rows of the following table (the entries are in terms of the reciprocal of βl): nj nσj nj (cont’d) nσj (cont’d)

1.37(1) 0.82(1)

1.66(1) 1.25(1)

2.02(1) 1.70(1)

2.29(1) 2.01(1)

4.43(2) 3.85(2)

4.82(2) 4.29(2)

5.25(1) 5.13(1)

5.40(2) 4.94(2)

5.99(2) 5.57(2)

5.99(1) 5.89(1)

− 5.76(3)

Now, if the panel is subjected to a compressive load σ = 0.0012E we can exploit the equivalence of the vibration and buckling modes to compute the reduced natural frequencies that are due to the presence of the axial load. That is, we make use of the relation [of the form of Eq. (11.1)]: ασ 1/2 2 nσj = nj − 2 2 . βλ

(11.24)

Thus (ασ)1/2 ≈ 1.1 and for the various λ we get the altered natural frequencies as shown in the lower rows of the preceding table. We note how the order has changed, the equal values have separated, and a new natural frequency has fallen into the range of interest because of the presence of the axial load.

18:2

P1: KAE Chapter-11

CUFX159-Virgin

0 521 85648 5

234

August 14, 2007

Nondestructive Testing

Figure 11.19. Eigenvalue curves for an axially loaded, prismatic, stiffened, plated structure. Reproduced with permission from Elsevier [32].

We note here also that the linearity in the axial load versus frequency-squared relation can also be used to infer the percentage error in neglecting shear effects in the modeling [33]. Another important practical application related to this approach is determining the level of stresses in pressure vessels [34]. References [1]

C. Massonnet. Les relations entre les modes normaux de vibration et la stabilite´ des ´ systemes elastiques. Technical Report, Bulletin des cours et des laboratoires d’essais des constructions du genie civil et d’hydraulique fluviale, Brussels, Belgium, I (1,2), 1– 353, 1940.

18:2

P1: KAE Chapter-11

CUFX159-Virgin

0 521 85648 5

August 14, 2007

References [2] [3] [4] [5] [6]

[7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20] [21] [22] [23]

[24] [25] [26] [27]

H. Lurie. Lateral vibrations as related to structural stability. Journal of Applied Mechanics, 19:195–204, 1952. C. Sundararajan. Frequency analysis of axially loaded structures. AIAA Journal, 30:1139–41, 1992. R.V. Southwell. On the analysis of experimental observations in problems of elastic stability. Proceedings of the Royal Society of London, 135A:601–16, 1932. J. Ari-Gur, T. Weller, and J. Singer. Experimental and theoretical studies of columns under axial loading. International Journal of Solids and Structures, 18:619–41, 1982. M.A. Souza. The effects of initial imperfection and changing support conditions on the vibration of structural elements liable to buckling. Thin-Walled Structures, 5:411–23, 1987. R.G. White. Evaluation of the dynamic characteristics of structures by transient testing. Journal of Sound and Vibration, 15:147–61, 1971. A. Segall and M. Baruch. A nondestructive dynamic method for the determination of the critical load of elastic columns. Experimental Mechanics, 20:285–8, 1980. P.M. Mujumdar and S. Suryanarayan. Nondestructive techniques for prediction of buckling loads – a review. Journal of the Aeronautical Society of India, 41:205–23, 1989. M.A. Souza and L.M.B. Assaid. A new technique for the prediction of buckling loads from nondestructive vibration tests. Experimental Mechanics, 31:93–7, 1991. J. Singer, J. Arbocz, and T. Weller. Buckling Experiments, Vol. 1. Wiley, 1998. J. Singer, J. Arbocz, and T. Weller. Buckling Experiments, Vol. 2. Wiley, 2002. J.G.A. Croll and A.C. Walker. Elements of Structural Stability. Wiley, 1972. C.R. Farrar, S.W. Doebling, and D.A. Nix. Vibration-based structural damage identification. Philosophical Transactions of the Royal Society, A359:131–49, 2001. A. Sommerfeld. Eine einfache Vorrichtung zur veranschaulichung des Knickungsvorganges. Zeitschrift des Verein deutscher Ingenieure, pp. 1320–3, 1905. A.B. Pippard. Response and Stability. Cambridge University Press, 1985. T.D. Burton. Introduction to Dynamic Systems Analysis. McGraw-Hill, 1994. A. Segall and G.S. Springer. A dynamic method for measuring the critical loads of elastic flat panels. Experimental Mechanics, 26:354–9, 1986. ´ T. Livingston, J.G. Beliveau, and D.R. Huston. Estimation of axial load in prismatic members using flexural vibrations. Journal of Sound and Vibration, 179:899–908, 1995. A.L. Sweet and J. Genin. Identification of a model for predicting elastic buckling. Journal of Sound and Vibration, 14:317–24, 1971. A.L. Sweet, J. Genin, and P.F. Mlakar. Determination of column-buckling criteria using vibratory data. Experimental Mechanics, 17:385–91, 1977. D.J. Inman. Engineering Vibration. Prentice Hall, 2000. P.-Y. Shih and H.L. Schreyer. Lower bounds to fundamental frequencies and buckling loads of columns and plates. International Journal of Solids and Structures, 14:1013–26, 1978. K. Huseyin and J. Roorda. The loading–frequency relationship in multiple eigenvalue problems. Journal of Applied Mechanics, 38:1007–11, 1971. K. Huseyin. Multiple Parameter Stability Theory and Its Applications. Oxford University Press, 1986. L.N. Virgin. Parametric studies of the dynamic evolution through a fold. Journal of Sound and Vibration, 110:99–109, 1986. L.N. Virgin and R.H. Plaut. Use of frequency data to predict secondary bifurcation. Journal of Sound and Vibration, 251:919–26, 2002.

235

18:2

P1: KAE Chapter-11

CUFX159-Virgin

236

0 521 85648 5

August 14, 2007

Nondestructive Testing [28] A. Chailleux, Y. Hans, and G. Verchery. Experimental study of the buckling of laminated composite columns and plates. International Journal of Mechanical Sciences, 17:489–98, 1975. [29] R.H. Plaut and L.N. Virgin. Use of frequency data to predict buckling. Journal of Engineering Mechanics, 116:2330–5, 1990. [30] R.H. Plaut and E.R. Johnson. The effect of initial thrust and elastic foundation on the vibration frequencies of a shallow arch. Journal of Sound and Vibration, 78:565–71, 1981. [31] J.E.M. Jubb, I.G. Phillips, and H. Becker. Interrelation of structural stability, stiffness, residual stress and natural frequency. Journal of Sound and Vibration, 39:121–34, 1975. [32] F.W. Williams, P.N. Bennett, and D. Kennedy. Curves for natural frequencies of axially compressed prismatic plate assemblies. Journal of Sound and Vibration, 194:13–24, 1996. [33] P.N. Bennett and F.W. Williams. Insight into the sensitivity to axial compressive load of the natural frequencies of structures which include shear deformation. Journal of Sound and Vibration, 209:707–22, 1998. [34] R.R. Archer. On the influence of uniform stress states on the natural frequencies of spherical shells. Journal of Applied Mechanics, pp. 502–5, September 1962.

18:2

P1: KAE Chapter˙12

CUFX159-Virgin

0 521 85648 5

August 14, 2007

12 Highly Deformed Structures

In most practical applications, Euler–Bernoulli beam theory is often sufficient to provide useful information about the relation between axial loading and free vibrations. However, there are a number of instances in which the axial loading, or some related effect, results in relatively highly deflected states of the system, especially when the structure under consideration is very slender. For example, a pipeline or cable is characterized by having one of its dimensions very much greater than the other two, and the loads to which it is subject may often result in large deflections, even in cases in which self-weight is the only appreciable loading [1, 2]. Elastic bending stiffness does not necessarily dominate the effects of gravity, for example. In these cases, a more sophisticated description of the geometry is needed, and it is these types of flexible structures that form the basis for this chapter. In Chapter 7, we saw how initial postbuckling could be handled by retaining extra terms in the various energy expressions. But now, we allow (static) deflections to become large by using an arc-length description of the geometry and then consider small-amplitude oscillations about these nonlinear equilibrium configurations. In the final section, a FE solution is also shown for a specific case (essentially with the same approach as used toward the end of Chapter 9). It also turns out that experimental verification is relatively easy, especially if thermoplastics like polycarbonate are used.

12.1 Introduction to the Elastica We start, in the usual way, by considering the behavior of an initially straight, inextensible, prismatic, thin elastic beam. The curvature of such a system is given by 1 dθ = , ρ dS

(12.1)

in which the deformed geometry of the system is described in terms of the arc-length coordinates S and θ, as shown for an axially loaded clamped beam in Fig. 12.1. In terms of Cartesian coordinates, we can write the angle as θ = tan−1 and, using dS = dX

dY , dX

1 + (dY/dX)2 ,

(12.2)

(12.3) 237

18:3

P1: KAE Chapter˙12

CUFX159-Virgin

0 521 85648 5

August 14, 2007

238

Highly Deformed Structures

Y S

X

L

Figure 12.1. A thin elastic beam with clamped end conditions subjected to end shortening.

we obtain the familiar expression for curvature dθ dθ dX d 2 Y/dX 2 1 = = = . 3/2 ρ dS dX dS [1 + (dY/dX)2 ]

(12.4)

Expanding the right-hand side as a Taylor series leads to 1 = (d 2 Y/dX 2 )[1 + (3/2)(dY/dX)2 + · · · +]−1 . ρ

(12.5)

Thus, if dY/dX is small, then we obtain the simple expression for curvature familiar from Euler–Bernoulli beam theory and generally used as an analytical basis in Chapters 7 and 8. In Eq. (7.40) the curvature was developed in terms of Lagrangian coordinates, and the next term in the expansion was retained for moderately large slopes. Both descriptions appear in the literature [3]. However, we now consider the fully (but still elastic) nonlinear system with no restriction on deflections. Returning to the example in Fig. 12.1 (and assuming no gravitational effects just yet) we can write the governing (elastica) equation in the relatively simple form d 2θ P sin θ, =− 2 dS EI

(12.6)

in which P is an axial load associated with the end shortening . This form is restricted to prismatic members with no forcing acting along its length but is a form familiar from the swings of a pendulum [4]. The boundary conditions must then be specified to obtain a solution to a specific problem. However, this equation is not easy to solve. Analytical solutions are available through elliptic integrals [5], and this form also allows mildly nonlinear solutions to be obtained with perturbation methods (see Naschie [3]), but we shall adopt numerical methods to solve this type of problem, that is, to obtain the deflected configuration under various axial loads and subsequent vibration properties about equilibrium configurations (which may be highly nontrivial) [6–8]. Four case studies will be described in which the primary difference between the cases concerns the boundary conditions. Experimental verification is included for each.

18:3

P1: KAE Chapter˙12

CUFX159-Virgin

0 521 85648 5

August 14, 2007

12.2 The Governing Equations

239

12.2 The Governing Equations If we return to a prismatic strip but now inclined to the horizontal by an angle β, and characterized by flexural rigidity EI, length L, and weight (per unit length) W, we can describe the geometry in terms of coordinates X(S, T ) and Y(S, T ), and rotation θ(S, T ) with respect to the X axis, where S is the arc length and T is time. The internal forces in the strip are denoted P(S, T ) and Q(S, T ) parallel to the X and Y axes, respectively, and the bending moment is M(S, T ). The governing equations can be written as [9] ∂X/∂S = cos θ,

∂Y/∂S = sin θ,

∂θ/∂S = M/EI,

∂M/∂S = Q cos θ − P sin θ,

(12.7)

and, in addition, we have dynamic equilibrium, ∂P/∂S = −(W/g)∂2 X/∂T2 − W sin β, ∂Q/∂S = −W cos β − (W/g)∂2 Y/∂T2 .

(12.8)

Damping can be added at this point, but relative to the experimental results to be discussed later the damping is very light and is neglected in the analytical description. We introduce convenient nondimensional quantities that are especially useful in the context of this kind of nonlinear formulation w = WL3 /EI,

x = X/L,

q = QL2 /EI,

m = ML/EI,

y = Y/L,

s = S/L, p = PL2 /EI, t = (T/L2 ) EIg/W, = ωL2 W/EIg, (12.9)

where ω is a dimensional vibration frequency. In nondimensional terms, equilibrium equations (12.8) thus become ∂x/∂s = cos θ,

∂y/∂s = sin θ,

∂θ/∂s = m,

∂m/∂s = q cos θ − p sin θ,

(12.10)

and for linear vibrations [Eqs. (12.8)] ∂ p/∂ s = −w sin β − ∂2 x/∂ t2 ,

∂ q/∂ s = −w cos β − ∂2 y/∂ t2 .

(12.11)

Assuming harmonic motion appropriate to small-amplitude vibration in the usual way, we can write the variables in the form x(s, t) = xe (s) + xd (s) sin t,

y(s, t) = ye (s) + yd (s) sin t,

θ(s, t) = θe (s) + θd(s) sin t,

m(s, t) = me (s) + md(s) sin t,

p(s, t) = p e (s) + p d(s) sin t,

q(s, t) = qe (s) + qd (s) sin t,

(12.12)

18:3

P1: KAE Chapter˙12

CUFX159-Virgin

0 521 85648 5

August 14, 2007

240

Highly Deformed Structures

where subscripts e and d denote equilibrium and dynamic quantities, respectively. At equilibrium, the equations are now given by xe = cos θe ,

ye = sin θe ,

θe = me ,

me = qe cos θe − p e sin θe ,

(12.13)

where the prime is used to denote the derivative with respect to s and where the internal forces can be written as p e (s) = p 0 − sw sin β,

qe (s) = q0 − sw cos β,

(12.14)

where p 0 and q0 are constants representing values at s = 0. We can determine equilibrium shapes by solving Eqs. (12.14), and then we obtain small vibrations about these equilibrium solutions by solving the resulting linear equations in the dynamic variables: xd = −θd sin θe ,

yd = θd cos θe ,

θd = md,

md = (qd − p e θd ) cos θe − (p d + qe θd ) sin θe ,

p d = 2 xd ,

qd = 2 yd .

(12.15)

The general approach used here for solving these types of nonlinear boundaryvalue problems is based on the shooting method [9, 10]. With this approach, the known boundary conditions at one end are used together with educated guesses of the unknown boundary conditions, and the nonlinear ordinary differential equations are solved numerically. However, the boundary conditions at the far end will not typically be satisfied, and the error is then used to iteratively re-solve the system until a tolerance has been achieved [11]. This is basically a root-finding approach and can be significantly simplified by use of some of the built-in capabilities of MATLAB or Mathematica, for example. Continuation is an alternative solution procedure [12].

12.3 Case Study A: Self-Weight Loading Revisited In Section 7.9, we considered the effect of self-weight on the dynamics of an upright cantilever by using a Rayleigh–Ritz approach and included some simple experimental results. In the experimental results shown in Fig. 7.15(a) a degree of postbuckled stiffness can be observed for moderately large deflections. We take another look at this system but now using the elastica approach outlined in the previous section (as originally shown in Fig. 7.13). Figure 12.2 shows the arc-length coordinates together with some typical equilibrium configurations. These, of course, go well beyond the range of validity of Euler–Bernoulli theory. Before some further results are presented, the appropriate nondimensionalization is mentioned. For the examples to be considered later in this chapter, it is natural to normalize the “weight” according to the first element in Eqs. (12.9) with the distance between the ends providing a natural control parameter. For the upright

18:3

P1: KAE Chapter˙12

CUFX159-Virgin

0 521 85648 5

August 14, 2007

12.3 Case Study A: Self-Weight Loading Revisited

241

Y

Y X

X

S

H

W

(c) (a)

(b)

Figure 12.2. The slender column subject to self-weight: (a) basic geometry, (b) arc-length coordinate system, and (c) some typical deflected shapes.

column (and also the pinched-loop configuration to be considered later), it is more convenient (especially for subsequent comparisons with experimental results) to use a = (EI/W)1/3 as the key nondimensional parameter. This is somewhat different from the scheme presented in Eqs. (12.9), which are used later. We use h = H/a as the control, where H is the height of the column (and equivalent to L), such that increasing the length of the system, in the presence of gravity, increases the effective axial loading and leads to instability.

12.3.1 Numerical Results For a vertical upright column (i.e., with the clamped end at the bottom), we can set β = π/2, and the gravity acts in the negative x direction. The boundary conditions are fully clamped at the base, that is, xe = ye = θe = 0 when s = 0, and free at the tip, that is, me = 0 when s = h, and the shooting method is used to determine the unknowns at the tip to within a prescribed accuracy. A summary of the equilibrium solutions for the heavy column are shown in Fig. 12.3(a) as a bifurcation diagram, that is, a measure of the nondimensional (lateral tip) deflection, y(h) (see Fig. 12.2) plotted against the control parameter (nondimensional column height h). A typical break in symmetry will be provided by some initial geometric imperfection, but this effect is not included here [9]. The trivial solution is ye (x) = 0. The first bifurcation point occurs at the critical height hcr = 1.986 (compare with the analyses described

18:3

P1: KAE Chapter˙12

CUFX159-Virgin

0 521 85648 5

August 14, 2007

242

Highly Deformed Structures 3 4 h 3

(a)

(b)

h 2.5

postbuckled frequencies

2

2 1.5 1 0 −3

1 −2

−1

0

1

2 3 y(h)

0.5 −2

prebuckled frequencies

0

2

4

2 6

Figure 12.3. (a) Equilibrium paths for the column including gravity and (b) corresponding frequencies of small-amplitude vibrations.

in Section 7.9 [13]), and the trivial solution is unstable for larger values of h. The nature of the bifurcation point is supercritical [i.e., stable symmetric; see Fig. 3.7(a)], and the column smoothly begins to droop as the height is increased past its critical value. Again, this is behavior somewhat familiar from the approximate analytical solution. The stability of these equilibria can be obtained from a linear-vibration analysis. Fundamental vibration frequencies are plotted in Fig. 12.3(b). Negative values of 2 (shaded gray) are associated with unstable equilibrium states and with motions that grow exponentially (see Section 3.1). The fundamental frequency is zero at the critical height (at least for the geometrically perfect, undamped case). The curve is convex toward the origin, unlike typical characteristic curves in which a loading parameter (rather than the height) is plotted versus the frequency squared [14]—see Section 11.3. Fundamental frequencies for vibrations about the stable postbuckled equilibrium path for the perfect column are plotted also.

12.3.2 Experiments We refer back to Section 7.9.2 that showed some experimental results for a (circularcross-section) cantilever [9]. The rod was placed in an upright position and the length was incrementally increased. The results from these experiments are shown in Fig. 7.15, which suggests buckling between H = 15 and 20 cm. In this case, an estimate of the flexural rigidity was obtained from the (gravitional) droop of a horizontal cantilever and using the theoretical critical length, 1.986(EI/W)1/3 resulted in an estimated critical height very close to 20 cm. This is close to the height associated with the minimum value of the fundamental frequency, shown in Fig. 7.15(b). The column clearly exhibits a supercritical bifurcation. A flat, slender strip of polycarbonate was used to take some additional data (natural frequencies) in which the geometry of the cross section ensured unambiguous deflection in a plane. The free end of the strut was subjected to a small perturbation and subsequent oscillations were monitored by a laser vibrometer. The fundamental frequency content was then extracted, with the results shown in

18:3

P1: KAE Chapter˙12

CUFX159-Virgin

0 521 85648 5

August 14, 2007

12.4 Case Study B: A Heavy Beam

243

150

H (cm) 125

100

75

50 0

1

2

3

4

5

2

6

Figure 12.4. Fundamental frequency for a polycarbonate strip.

Fig. 12.4. The reduction in the lowest natural frequency can clearly be seen as buckling is approached (the theoretical critical length is approximately 110 cm in this case) together with an increase in the postbuckling frequencies. The form of the length–frequency relation follows the theoretical curves depicted in Fig. 12.3(b) quite closely, especially if an initial imperfection had been incorporated into the analysis. Also, the inevitable presence of a little damping has a minor effect on the frequencies, but again, this is not considered in the analysis here.

12.4 Case Study B: A Heavy Beam Suppose the column is now rotated back to its horizontal configuration (β = 0) and both ends are constrained against lateral deflection and rotation. Rather than the column height (length), it is more convenient to use the axial (imposed) end shortening of the strip as the control parameter (see Fig. 12.1), and this can also be placed into nondimensional terms by use of δ = /L [15–17]. Now we use the nondimensionalization used in Eqs. (12.9) rather than a from the previous section. In Chapter 2 2 7, we obtained the critical buckling load for a clamped–clamped beam of 4π EI/L and lowest natural frequency of 22.37 EI/(ρAL4 ) for the straight beam (of length L and flexural rigidity EI) without including the effect of gravity (acting laterally on the beam and giving a constant weight W per unit length). Now the boundary conditions when s = 0 are xe = ye = θe = 0, and the quantities me (0), p 0 , and q0 are determined by the shooting method, based on satisfying the conditions xe = 1 − δ and ye = θe = 0 at s = 1 with sufficient accuracy. A Note on Lift-Off and Self-Contact. For a very long strip, subject to end short-

ening, only a central part of the beam will tend to lift-off [18]. As the ends of the strip are pushed toward each other, the length of the proportion of the beam that

18:3

P1: KAE Chapter˙12

CUFX159-Virgin

0 521 85648 5

August 14, 2007

244

Highly Deformed Structures

does not lift off diminishes until the whole strip is characterized by a nonzero lateral deflection (other than immediately at the clamped ends). For convenience, the strip is called “short” if it does not touch the foundation between its ends. If there is a flat section resting on the foundation at both ends, the strip is called “long.” The extent to which the beam is long or short depends to a large extent on its weight/stiffness ratio. A practical aspect of this is that for the long beam a zero-frequency traveling wave is observed [18]. Another interesting feature of this type of system is that, for sufficiently high values of end shortening, self-contact may occur, in which two points on the strip contact each other and the segment between has a teardrop (or pinched-loop) shape [19]. A geometry related to this specific case will be considered in a later section of this chapter.

12.4.1 Numerical Results Equilibrium paths are depicted in Fig. 12.5 in the plane of end shortening δ versus axial load p 0 , along with some corresponding equilibrium shapes, for weight parameters w = 0, 25, 125, 250, and 343. As a reference point, for w = 0 (no weight), the fixed–fixed strip buckles at p 0 = 4π2 , and shapes at points A, B, and C along the postbuckling path are shown. This result was essentially obtained in Chapter 7 in which the increase in deflection after buckling (indicating a degree of postbuckled stiffness) was first observed. Self-contact occurs at C, when δ = 0.849 and p 0 = 72.18 [18], with midpoint deflection y(0.5) = 0.403. Under increasing δ, the near-horizontal path to the right of C is followed, with large increases in p 0 associated with small increases in δ. For heavy strips (w > 0), as δ is increased from zero, the strip is initially long but then becomes short. For w = 25, symmetric self-contact occurs when δ = 0.845, 1

B 0

C

D 125

E 250

F

G 343

H 343

I

w=0

A

125

250

0

C E

0.8 G

I B

0.6

F

0.4 H

A D

250

0.2 0 −50

0

343

125

w=0

25

50

100

150

p0 200

Figure 12.5. Equilibrium shapes and end shortening as functions of axial load for horizontal strip with weights (from left to right) w = 0, 25, 125, 250, and 343.

18:3

P1: KAE Chapter˙12

CUFX159-Virgin

0 521 85648 5

August 14, 2007

12.4 Case Study B: A Heavy Beam

245

1 δ 0.8

0.6 250 125

w=0

0.4

0.2 25

0

0

10

20

30

40

Ω

50

Figure 12.6. End shortening as a function of fundamental frequency for short strips with w = 0, 25, 125, and 250, and mode shapes; equilibrium shapes are shown in gray.

p 0 = 71.65, and ye (0.5) = 0.401. For w = 125, the symmetric shape becomes unstable when δ = 0.820 and p 0 = 69.74, and the stable equilibrium associated with tilted (asymmetric) shapes bifurcates leftward in Fig. 12.5 toward point E, where self-contact occurs with δ = 0.820 and p 0 = 60.33. These equilibrium shapes correspond only to positive lateral deflection (lift-off). In the absence of a foundation a variety of other equilibrium configurations are possible (although most of these are unstable) [20]. Small vibrations about equilibrium are shown in Fig. 12.6 in terms of the fundamental frequency for w = 0, 25, 125, and 250, along with the mode shapes for four specific cases. As the end shortening δ is increased, the fundamental frequency is zero at the transition from the long to short equilibrium shape, and then is zero again when the symmetric shape becomes unstable (as seen for weights w of 125 and 250). For w = 0, the frequency at δ = 0 is = 44.36, corresponding to the second mode of a fixed–fixed column subjected to an axial load p 0 = 4π2 . When = 19.81 for w = 0, and also when = 5.35 for w = 25, symmetric self-contact occurs, and these two curves in Fig. 12.6 are ended. For w = 125, as δ is increased beyond the value 0.820 where the symmetric shape becomes unstable, the strip tilts and the frequency increases until self-contact occurs when = 0.155 (and δ = 0.820 still). An asymmetric mode along the path for w = 250 is shown in the top-left part of Fig. 12.6. 12.4.2 Experiments It turns out that although the types of structures described in this section exhibit very large deflections it is relatively straightforward to confirm much of this behavior experimentally. Throughout this chapter some experimental studies in which thin polycarbonate strips are used are described. The general approach to measuring

18:3

P1: KAE Chapter˙12

CUFX159-Virgin

246

0 521 85648 5

August 14, 2007

Highly Deformed Structures

frequencies and mode shapes under different levels of axial loading is similar for each of the systems in this chapter. For this very flexible elastic material, the specific weight was measured at 11.2 kN/m3 and Young’s modulus was 2.4 GPa. The strips were 7.62 mm wide. Two thicknesses (0.508 and 1.016 mm) and two lengths (0.532 and 0.832 m) were used in combinations to yield three different values for the nondimensional weight w. The strip was clamped such that one end could move toward the other in 6.3-mm increments to create the end shortening (see Fig. 12.1). This is the main control parameter in this example, that is, the actual (arc) length of the beam is held constant. An alternative means of changing the system would be to feed additional material in from one side, which was effectively done for the system shown in Fig. 3.9. The strip was deflected beyond the transition from long to short equilibrium before any measurements were taken for both the static and dynamic experiments. A point-to-point laser vibrometer was used to measure the velocity at a userprescribed point on the strip (avoiding any obvious node). For modal measurements, the velocity at multiple points was taken. For δ values ranging from 0.021 to 0.917, the first four frequencies were obtained by excitation of the strip by an impact hammer at different locations along the strip. Frequency measurements were also independently confirmed by measurement of the beam response to forced excitation, that is, use of a sine sweep (from 0.008 to 50 Hz) applied to the baseplate by an electromagnetic shaker. For the modal analysis, the strip was again excited with a modal impact hammer. Data were acquired by the vibrometer (utilizing Bruel and Kjaer pulse signalprocessing software) and analyzed with ME’scope VES to generate a frequencyresponse function for each measurement point. This same approach was then used to measure the response at 30 different points along the strip. Vibration modes associated with the first few frequencies were constructed. In Fig. 12.7, experimental and analytical results are compared, with frequencies given in hertz. The open circles are associated with tests on a strip of length 0.532 m and thickness 1.016 mm, giving w = 8.145. The solid circles in Fig. 12.7 correspond to tests on a strip of length 0.532 m again, but a thickness of 0.508 mm, so that w = 32.56. Several data points at high values of end shortening are associated with strips having self-contact. For the leftmost results, the open triangles were obtained experimentally for a longer strip, with length 0.832 m and thickness 0.508 mm (corresponding to w = 124.7). In terms of higher frequencies, Fig. 12.8 shows the first four frequencies (in hertz) for the specific case w = 32.56. Black circles correspond to data acquired from an impact test; open circles denote results obtained by forced vibration. The fundamental frequency is zero when the strip becomes short at δ = 0.017 (with p 0 = 80.86). This can be inferred from Fig. 12.5. The frequency increases and then decreases. Self-contact occurs when δ = 0.844 and p 0 = 71.47. Vibration mode shapes are shown in Fig. 12.9 for δ = 0.117, with the analytical shapes on the left and the experimental shapes on the right. Further results (including cases for β = 0) can be found in Santillan et al. [18].

18:3

P1: KAE Chapter˙12

CUFX159-Virgin

0 521 85648 5

August 14, 2007

12.4 Case Study B: A Heavy Beam

247

1 δ/L 0.9 0.8 0.7 w = 8.145 0.6 0.5 w = 32.56

0.4

w = 124.7

0.3 0.2 0.1 0

0

2

4

6

8

10 Frequency (Hz)

12

Figure 12.7. End shortening as a function of fundamental frequency for horizontal strips. Solid curves correspond to w = 124.7, 32.56, and 8.145; , •, and ◦, experiment.

1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0

0

5

10

15

20

25

30

F requency (Hz)

Figure 12.8. End shortening as a function of lowest four frequencies for horizontal strip with w = 32.56. experiment: ◦ (forced) and • (free); dashed curves, theory. Self-contact is indicated by the horizontal gray lines: continuous, theory; dashed, experiment.

18:3

P1: KAE Chapter˙12

CUFX159-Virgin

0 521 85648 5

August 14, 2007

248

Highly Deformed Structures 4.30 Hz

4.47 Hz

Mode 1

8.1 Hz

7.83 Hz

Mode 2

16.46 Hz

17.0 Hz

Mode 3

Experiment

L=

Theory

Figure 12.9. First three vibration modes from analysis and experiment for horizontal strip with w = 32.56 and δ = 0.117; equilibrium shape is dashed.

A relatively highly deflected (or prestressed) elastic beam with pinned ends can also be considered to be an arch. Such a structure was analyzed by Perkins [21] in which a slightly different analytical approach was used to determine the natural frequencies of small-amplitude vibrations about highly nonlinear equilibria. He used a variational formulation to obtain the results shown in Fig. 12.10, in which the four lowest natural frequencies are plotted as a function of the nondimensional end load n. Some mode shapes at specific values of the end load are superimposed (for n = 5, 15, 21.55). The lowest natural frequency dropping to zero at the Euler buckling load (corresponding to n = π2 ) is observed together with the subsequent jump to an asymmetric mode. He also conducted some experiments, and a comparison between theoretical and measured frequencies is summarized in the table following (in which H is the separation distance between the pinned ends, and L is arc length): ω1

ω2

ω3

H/L

n

Th.

Exp.

%

Th.

Exp.

%

Th.

Exp.

%

0.8 0.35 0.06

11 15 20

26.9 13.6 4.57

25.7 13.3 4.7

4.8 2.1 2.7

77.3 62.8 53.1

75.9 61.7 52.6

1.9 1.9 0.9

146 133.7 122.8

140 135 113.7

4.7 1.7 8.0

12.5 Case Study C: A Pinched Loop An interesting extension to the analysis of the first case study can be made if the ends of the clamped beam are rotated such that they are pressed flat together, as shown in Fig. 12.11. Also shown in this figure (part c) is a snapshot of a highly deflected shape corresponding to a length beyond the buckling point for the upright equilibrium configuration. Again the parameter a is used for the nondimensionalization and the behavior is investigated through the evolution of the nondimensional

18:3

P1: KAE Chapter˙12

CUFX159-Virgin

0 521 85648 5

August 14, 2007

12.5 Case Study C: A Pinched Loop

249

160

120

80

40

25

0 5

10

15

20

End Load

–20

Figure 12.10. Natural frequencies and mode shapes of an elastic arch as functions of end load n. Reproduced with permission from ASME [21].

control parameter l = L/a. The boundary conditions are basically the same as for the clamped beam, but now we have θe (l) = −π, and although it is easy to incline the pinched support [18], we focus attention on the upright system, that is, β = π/2 in the governing equations (and again include gravitational effects). (a)

(b) L

(c)

X g

S

Y

Figure 12.11. (a) Geometry of pinched loop, (b) photographic image of experimental setup, and (c) a highly deflected configuration.

18:3

P1: KAE CUFX159-Virgin

0 521 85648 5

August 14, 2007

250

Highly Deformed Structures

(b)

(a)

7

l 6 5 4 3 2 0

0.5

1

1.5 2 y(l/2)

0

1

2

3

ω

4

Figure 12.12. Horizontal midpoint deflection of upright loop as a function of length. Solid curve, numerical; •, experimental.

Because the theoretical approach is now well established, both theoretical and experimental results are plotted together in this section. The experimental results were based on a strip of cross-sectional dimensions 25.4 mm × 0.508 mm, which corresponds to the reference length a of 0.167 m, and the nondimensional vibration √ frequency ω of 0.130 times the dimensional frequency (where ω = a/g) that is √ due to the scaling of time: t = T g/a—again note the difference with Eqs. (12.9). In this section, most of the results are presented in terms of nondimensional quantities of frequency ω and the coordinates x and y of the midpoint (where s = 0.5). Equilibrium results for the loop are depicted in Fig. 12.12(a). The horizontal deflection of the midpoint is plotted as a function of the length of the loop, with the appearance of a critical length of l = 4.50 (signifying the onset of a supercritical pitchfork bifurcation). A typical experimental frequency spectrum is depicted in Fig. 12.13, obtained with the laser vibrometer discussed in the previous section. The strip from which rms Velocity (m/s)

Chapter˙12

0

10

20

30

40 50 Frequency (Hz)

Figure 12.13. Typical frequency spectrum for upright loop with l = 3.06; main peaks at 1.81, 10.73, 20.16, 30.53 Hz.

18:3

P1: KAE Chapter˙12

CUFX159-Virgin

0 521 85648 5

August 14, 2007

12.6 Case Study D: A Beam Loaded by a Cable

251

140 ω 120 100 80

Figure 12.14. Lowest four natural frequencies of upright loop as functions of length.

60 40 20 0 0.5

1

1.5

2

2.5

3

these data were collected is 0.5096 m long (l = 3.06). The results were averaged over four time series. The four lowest-dimensional (in-plane) frequencies are distinct at 1.81, 10.73, 20.16, and 30.53 Hz. This procedure was repeated for several thicknesses and a number of lengths. The four lowest measured frequencies are denoted by circles, squares, and triangles in Fig. 12.14 over a range of nondimensional lengths (these vibrations are all about the trivial equilibrium configuration for this range of l). Solid curves represent the analytical results, and in general we see the anticipated reduction in the natural frequency as the loop length increases, thus observing the softening effect of gravity (for the upright orientation). The mode shapes were also determined experimentally and analytically for some cases. The first four mode shapes for the upright case with l = 3.48 and a strip thickness of 0.508 mm are depicted in Fig. 12.15, along with the equilibrium shape. The corresponding measured dimensional frequencies are 1.3, 8.26, 15.8, and 23.9 Hz. As expected, the second and fourth modes are symmetric with respect to the vertical axis. Finally, the fundamental vibration frequency is plotted in Fig. 12.12(b) as a function of the nondimensional length for relatively long loop lengths (including postcritical drooping). The experimental data points were obtained as the average values from multiple-frequency spectra. As the length of the loop is increased, the fundamental frequency decreases until it is effectively zero at the critical length l ≈ 4.5. Extrapolation of measured fundamental frequencies at smaller lengths to the length at zero frequency could be used to predict the critical length (see Chapter 11). As the length is increased further and the loop droops (postbuckling deformation), the fundamental frequency increases.

12.6 Case Study D: A Beam Loaded by a Cable In the final section of this chapter, we again consider the free vibrations of an elastic structural system characterized by large deflection that is due primarily to axial-load

l

3.5

18:3

P1: KAE Chapter˙12

CUFX159-Virgin

0 521 85648 5

252

August 14, 2007

Highly Deformed Structures

Mode 1

Mode 3

Mode 2

Mode 4

Figure 12.15. First four mode shapes for upright loop. Dashed curve, equilibrium (analytical); solid curve, mode (analytical); •, equilibrium (experimental); ◦, mode (experimental).

effects. A prismatic cantilever beam has a cable attached to its free end that is then pulled in a direction depending on the location of its far-end attachment point [22]. This can be considered as somewhat intermediate between the cantilever subject to an increasing end load that maintains its direction and Beck’s problem (Chapter 7), although it can be shown that this is a conservative system. If the distant end of the cable is attached to the beam on its axis, then Timoshenko and Gere [5] showed that the critical load is equal to the Euler load (but for a pinned–pinned beam). If the far-end attachment point is offset, then symmetry is broken and the increase in load results in the nonlinear deflection of the cantilever in much the same way that initial geometric imperfections influence the behavior of axially loaded systems in general. The cable loading has a considerably greater effect than gravity and hence weight is not included in this analysis. A schematic of the system is shown in Fig. 12.16. The primary method of analysis is again based on a shooting method solution of the boundary-value problem. Again some experimental verification is presented and some FE solutions are also included. Here, the tension in the (axially very stiff) cable is the principal control parameter (as the far end is pulled through the attachment point), with natural frequencies and mode shapes again characterizing the vibration of small-amplitude motion about even highly deflected equilibria. We still basically have the same form of the governing static equations but with the addition of the horizontal component of the end force that complicates the

18:3

P1: KAE Chapter˙12

CUFX159-Virgin

0 521 85648 5

August 14, 2007

12.6 Case Study D: A Beam Loaded by a Cable

A θ

253

x

s

y

L

C (a, b)

P B Figure 12.16. Geometry of a slender cantilever column loaded by a cable passing from the free end to a point near the base.

boundary conditions. Given the x and y offset coordinates a and b, respectively, the boundary condition (moment) at the (s = 0) clamped end is Me (0) = Pve (L)b − Phe (L)a,

(12.16)

and the load is adjusted (by use of Newton’s method) until Me (L) = 0 is satisfied. In the results to be presented, some comparisons will be made with FE results obtained with abaqus. Beam elements (B31), suitable for large deflections, were used in which 1000 elements were employed to ensure spatial convergence. A truss element was used for modeling the cable, and an extreme negative thermal load was applied to drastically reduce the length of the cable and thus pull on the end of the cantilever. A path-following algorithm based on Riks method was employed [23, 24], that is, the same approach as used for the highly deflected cantilever in Chapter 9. Figure 12.17 shows an experimental setup for the cable–beam system. A thin polycarbonate beam of dimensions 0.762 m long with a rectangular cross section of 25.4 mm × 4.8 mm is configured as a cantilever. The elastic modulus and density were given earlier in this chapter. The cable was made of high-strength woven steel wire with a stiffness of 11.67 kN/m. The cable was then connected to the base Acelrometr Shak er Beam

Mounting Bloc

Adapter Plate

k Cab

Fle le Laser T arget

xib

le

Joint Load Cel

Figure 12.17. The experimental system of cantilever beam and end-loading cable.

18:3

P1: KAE Chapter˙12

CUFX159-Virgin

254

0 521 85648 5

August 14, 2007

Highly Deformed Structures

Figure 12.18. A typical frequency-response spectrum from the cable–beam system.

by a flexible joint with the other end attached to the tip of the cantilever. A load cell was incorporated into the wire connection in order to monitor the tensile force in the system. In some cases, the wire was replaced with shorter lengths to facilitate accurate tension measurements. The base of the cantilever was mounted to an electromagnetic shaker, with the excitation measured by an accelerometer. The response of the beam was again measured by a laser velocity vibrometer in the usual way. Standard data acquisition and signal-processing data were used (the Bruel and Kjaer pulse system), and a typical frequency response is shown in Fig. 12.18. This particular spectrum was taken from a system in which the nondimensional offset was a/L = 0.0375, b/L = 0.0167, and the applied load was P/Pcr = 1.12, in which Pcr corresponds to the elastic critical (Euler) load for the underlying case with no offset. The excitation employed here was a pseudorandom input signal over the range 0 to 100 Hz and with a sampling rate of f = 0.03125 Hz. Appropriate windowing and averaging were used to improve the quality of the data [25]. Vibration mode shapes were then extracted by the standard approach at a specific level of the cable tension. Figure 12.19 shows the static and dynamic response for the beam with a/L = 0.0375, b/L = 0.0167. Part (a) shows both shooting results and abaqus together with some experimental data points for the lateral deflection of the tip. The inset shows a couple of deflected shapes. Part (b) shows the corresponding fundamental frequency, which also changes as a function of the tension in the cable. The frequency does not exhibit the type of monotonic decay we observed in earlier sections of this book. The frequency in part (b) is nondimensionalized bythe fundamental frequency for a cantilever beam with no cable attached (ω = 3.516 EI/mL4 ), and it is interesting to see the effect the cable has (for both FE and shooting results) as the

18:3

P1: KAE Chapter˙12

CUFX159-Virgin

0 521 85648 5

August 14, 2007

12.6 Case Study D: A Beam Loaded by a Cable

255

1.5 P/Pcr

a/L = 0.0375 b/L = 0.0167

1.0

0.5

P/Pcr = 1.01

0

0.1

0.2

0.3

0.4

0.5

P/Pcr = 1.30

0.6

0.7

1.0

1.5

2.5

2.0

Ytip / L

1

1, P = k = 0

Figure 12.19. (a) Load-deflection characteristic with b/L = 0.0167 and (b) the corresponding fundamental natural frequency. The solid curves represent the FEA solution, dashed for analytical results obtained with the shooting method. 1.5 P/Pcr

a/L = 0.0375 b/L = 0.075

1.0

0.5

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7 Ytip / L

1.0

2.0

4.0

3.0 1

1, P = k = 0

Figure 12.20. (a) Load-deflection characteristic with b/L = 0.075 and (b) the corresponding fundamental natural frequency. The solid curves represent the FEA solution, dashed for analytical results obtained with the shooting method.

tension tends to zero, i.e., even at zero tension the cable provides some constraint at the “free” end. Figure 12.20 shows similar results but with a larger static offset: a/L = 0.0375, b/L = 0.075. The behavior is seen to be quite sensitive to the magnitude of the cable offset. A similar procedure was followed to obtain the higher frequencies and the first four, for the smaller offset, are shown in Fig. 12.21. The gray dashed curve at P/Pcr = 1.12 indicates the level at which the frequency spectrum illustrated in Fig. 12.18 was taken. Although the frequencies are separated in a somewhat typical

18:3

P1: KAE Chapter˙12

CUFX159-Virgin

0 521 85648 5

August 14, 2007

256

Highly Deformed Structures 1.5

P/Pcr

1.0

0.5

0

10

20

30

40

50 1

60

1, P = k = 0

Figure 12.21. The four lowest natural frequencies plotted as functions of the tension in the cable with b/L = 0.0167. The solid curves represent the FEA solution, dashed for analytical results obtained with the shooting method.

spread for beam vibrations the effect of the tension loading depends on the specific frequency. Some vibration modes shapes are shown in Fig. 12.22 for the larger offset value. There is good agreement between the results from a shooting analysis (eigenvectors), abaqus, and experimentally determined mode shapes. There is an arbitrary phase in some of these plots. Finally, one of the potential applications of this type of system is solar sails (see also Section 9.6). In these innovative structural systems, the idea is to use the Sun’s photons for propulsion based on a very lightweight but large surface area, rather like a kite. However, to keep such a membrane taut, it would need to be attached to relatively stiff but inevitably slender booms [26] that might typically lead to offset loading of the type considered in this section. There might also be some advantage to using tapered booms (see Section 8.3 and Holland et al. [27]).

12.7 The Softening Loop Revisited Before we leave this chapter, a brief result is given that uses the elastica to solve the generating example described in Section 3.5, again with the shooting method. By incorporating a geometrically nonlinear (softening) moment–curvature relation from experiments, we obtain the subcritical pitchfork bifurcation result shown in Fig. 12.23. The dashed curve is the result when the effect of a small initial geometric imperfection is included in the analysis (i.e., the symmetry is broken). The lower part

18:3

P1: KAE Chapter˙12

CUFX159-Virgin

0 521 85648 5

August 14, 2007

12.7 The Softening Loop Revisited

257 (a)

(b)

(c)

Figure 12.22. The lowest four vibration modes with b/L = 0.075: (a) shooting, (b) abaqus, and (c) experimental.

8

h 7 q = 0.01 A 6

q=0

5

4

−5

−2.5

0

1.5

y

y 1

0.5 0 −1

0

1

2

x

3

2.5

z(h)

5

1.5 1 0.5 0

0

1

2

3

4

z 5

Figure 12.23. Equilibrium paths for softening loop, where q is an initial imperfection, q = 0 or q = 0.01. Two views of the deflected configuration corresponding to point A are also shown [10].

18:3

P1: KAE Chapter˙12

CUFX159-Virgin

0 521 85648 5

August 14, 2007

258

Highly Deformed Structures

of this figure shows two views of the drooped configuration immediately after the initial instability. The transition between the upright and this drooped configuration is dynamic because there is no locally adjacent stable equilibrium, thus confirming the subcritical qualitative behavior from Figs. 3.10(c) and 3.10(d). Further details of the solution technique can be found in Plaut and Virgin [10]. In experiments on a loop with axisymmetric section properties, the natural frequencies consist of out-of-plane, in-plane, and twisting modes, and the lowest mode for each case is shown for oscillations about the (prebuckled) upright position in Fig. 12.24. Here h is a nondimensional parameter associated with one-half the length of the loop. Appropriately nondimensionalized, this also confirms part of the qualitative picture from Figs. 3.10(b) and 3.10(d), as well as the experimental data in Fig. 3.14, thus again confirming the trend of the lowest natural frequency toward zero at buckling.

h

(a)

6 5.5 5 4.5 4 3.5 3 2.5 2 0

(b)

1

2

(c)

3

4

5

(d)

Figure 12.24. (a) Half-length versus lowest frequencies of each type, and corresponding modes of vibration, (b) out-of-plane (——–); (c) in-plane (- - - - -); (d) twist (........).

18:3

P1: KAE Chapter˙12

CUFX159-Virgin

0 521 85648 5

August 14, 2007

References References [1] [2] [3] [4] [5] [6] [7] [8] [9]

[10]

[11] [12] [13]

[14] [15] [16] [17] [18] [19] [20] [21] [22]

[23]

J.F. Wilson. Dynamics of Offshore Structures. Wiley, 2002. Y. Bai and Q. Bai. Subsea Pipelines and Risers. Elsevier, 2005. M.S. El Naschie. Stress, Stability and Chaos in Structural Engineering: An Energy Approach. McGraw-Hill, 1990. D.W. Jordan and P. Smith. Nonlinear Ordinary Differential Equations. Oxford University Press, 1999. S.P. Timoshenko and J.M. Gere. Theory of Elastic Stability, 2nd ed. McGraw-Hill, 1961. J.P. Cusumano. Low-Dimensional, Chaotic, Nonplanar Motions of the Elastica: Experiment and Theory. Ph.D. dissertation, Cornell University, 1990. C. Gatti-Bono and N. C. Perkins. Dynamic analysis of loop formation in cables under compression. International Journal of Offshore and Polar Engineering, 12:217–22, 2002. D. Addessi, W. Lacarbonara, and A. Paolone. On the linear normal modes of planar pre-stressed curved beams. Journal of Sound and Vibration, 284:1075–97, 2005. L.N. Virgin and R.H. Plaut. Postbuckling and vibrations of linearly elastic and softening columns under self-weight. International Journal of Solids and Structures, 41:4989–5001, 2004. R.H. Plaut and L.N. Virgin. Three-dimensional postbuckling and vibration of vertical half-loop under self-weight. International Journal of Solids and Structures, 41:4975–88, 2004. C.J. Goh and C.M. Wang. Generalized shooting method for elastic stability analysis and optimization of structural members. Computers and Structures, 38:73–81, 1990. M.A. Crisfield. Nonlinear Finite Element Analysis of Solids and Structures, Vol. 2: Advanced Topics. Wiley, 1997. A.G. Greenhill. Determination of the greatest height consistent with stability that a vertical pole or mast can be made, and of the greatest height to which a tree of given proportions can grow. Proceedings of the Cambridge Philosophical Society, 4:65–73, 1881. K. Huseyin. Multiple Parameter Stability Theory and Its Applications. Oxford University Press, 1986. R. Schmidt and D.A. DaDeppo. Large deflection of heavy cantilever beams and columns. Quarterly Journal of Applied Mathematics, 28:441–4, 1970. C.Y. Wang. A critical review of the heavy elastica. International Journal of Mechanical Sciences, 28:549–59, 1986. S.-B. Hsu and S.-F. Hwang. Analysis of large deformation of a heavy cantilever. SIAM Journal on Mathematical Analysis, 19:854–66, 1988. S. Santillan, L.N. Virgin, and R.H. Plaut. Post-buckling and vibration of heavy beam on horizontal or inclined rigid foundation. Journal of Applied Mechanics, 73:664–71, 2006. S. Santillan, L.N. Virgin, and R.H. Plaut. Equilibria and vibration of a heavy pinched loop. Journal of Sound and Vibration, 288:81–90, 2005. J.G.A. Croll. Some comments on the mechanics of thermal buckling. The Structural Engineer, 83:127–32, 2005. N.C. Perkins. Planar vibration of an elastica arch: Theory and experiment. Journal of Vibration and Acoustics, 112:374–9, 1990. H.G. McComb. Large deflection of a cantilever beam under arbitrarily directed tip load. Technical Report, Technical Memorandum 86442, NASA Langley Research Center, 1985. E. Riks. The application of Newton’s method to the problem of elastic stability. Journal of Applied Mechanics, 39:1060–6, 1972.

259

18:3

P1: KAE Chapter˙12

CUFX159-Virgin

260

0 521 85648 5

August 14, 2007

Highly Deformed Structures [24] D.B. Holland, I. Stanciulescu, L.N. Virgin, and R.H. Plaut. Vibration and large deflection of cantilevered elastica compressed by angled cable. AIAA Journal, 44:1468–76, 2006. [25] T.G. Beckwith, R.D. Marangoni, and J.H. Lienhard. Mechanical Measurements. Addison-Wesley, 1993. [26] I. Stanciulescu, L.N. Virgin, and T.A. Laursen. Finite element analysis of slender solar sail booms. Journal of Spacecraft and Rockets, 44:528–37, 2007. [27] D.B. Holland, L.N. Virgin, and R.H. Plaut. Large deflections and vibration of a tapered cantilever pulled at its tip by a cable. Journal of Sound and Vibration, 2007, to appear.

18:3

P1: KAE Chapter-13

CUFX159-Virgin

0 521 85648 5

August 14, 2007

13 Suddenly Applied Loads

13.1 Load Classification The early parts of this book focused attention on the dynamics of structures in which axial loading was increased quasi-statically. Thus dynamic response was typically considered under effectively set axial-loading conditions and observed in terms of free vibrations. However, it is just as common for either the axial load to be applied dynamically or an axially loaded structure to be subjected to dynamic lateral loading as well [1]. In these cases, it is not uncommon for the maximum response to occur during transient motion. In this chapter, we look at a number of different scenarios in which a structure with a constant axial load is then subject to various types of (dynamic) disturbance forces. These will range from a slow, but nonnegligible, increase in axial loading, to suddenly applied loading (e.g., an impulse or step input [2–5]). We have already seen (e.g., Fig. 5.7) that the free response of an undamped nonlinear system is strongly influenced by the magnitude of the initial conditions. In some, a large initial velocity can be considered as an impulse. Before moving on to consider some specific structural systems, we return to the simple (underdamped) linear oscillator from Chapter 3 (Fig. 3.1). However, we now add a step input applied directly to the mass such that the governing equation of motion is given by m¨x + cx˙ + kx = F0 ,

(13.1)

in which we assume a system initially at rest, that is, x(0) = x(0) ˙ = 0. It is easy to show (by use of Laplace transforms for instance) that the response of this system is given by x(t) ωn −ζωn t = x(t) ¯ =1− e sin (ωd t + φ1 ), (13.2) F0 /k ωd √ where φ1 = cos−1 ζ, ω2n = k/m, ωd = ωn 1 − ζ 2 , ζ = c/(2mωn ). A typical response (normalized with respect to the magnitude of the input) is shown in Fig. 13.1 with ωn = 1, ζ = 0.1. This type of step response can be characterized by a number of metrics, for example, how long the transient takes to die out or by how much the response overshoots the final resting position initially. Clearly, damping has a key role to play here. It can be shown that the maximum percentage overshoot (OS, i.e., x¯ over and above unity) is √

OS = 100e−πζ/

1−ζ 2

,

(13.3) 261

18:8

P1: KAE Chapter-13

CUFX159-Virgin

0 521 85648 5

August 14, 2007

262

Suddenly Applied Loads tr

2 x1.5

= 0.1 OS

1 0.5

F0

ts

t

0 0

10

20

t

30

40

50

Figure 13.1. The step response of a linear system.

that is, twice the static response for an undamped (ζ = 0) system, and this occurs after the rise time (tr ). Thus we may envision a load applied suddenly to be stronger in its effect than the same load applied quasi-statically (i.e., on a much slower time scale than the flexural dynamics of the system). Similarly, the time taken for the response to decay to within 5% (say) of the final value is called the settling time (ts ) and is often approximated by 3/(ζωn ) [6]. So much for the step response of a linear system. Things become more interesting when we study the effect of a step input on the response of a nonlinear system, and specifically one in which the nonlinear system relates to the equilibrium of an axially loaded structure. Not only may a suddenly applied load cause additional axial loading through large-amplitude effects but it may also result in collapse because of the traversing of an adjacent underlying unstable equilibrium (local potentialenergy hilltop), or snap-through, for example.

13.2 Back to Link Models A good example of where stability in the large may be especially important is for imperfection-sensitive structures. If we reconsider the system shown in Fig. 5.9(a) with C = 0 and thus α = 1 we obtain the equilibrium condition =

(sin θ − sin θ0 ) cos θ . sin θ

(13.4)

Note that because we are now interested in relatively large excursions from equilibrium we do not use the approximation given by Eq. (5.45). The dynamic response is governed by θ¨ + ω2n [(sin θ − sin θ0 ) cos θ − sin θ] = 0,

(13.5)

where we have already established the relation between the linearized natural frequency and axial load (see Section 5.3). We now investigate the robustness of equilibrium to relatively large disturbances as a function of the initial imperfection. Without damping we can view

18:8

P1: KAE Chapter-13

CUFX159-Virgin

0 521 85648 5

August 14, 2007

13.2 Back to Link Models

.

263

1

(a)

θ

.

1

0.5

0.5

0

0

−0.5

−0.5

−1

−2

−1

0

1

θ

2

(b)

θ

−1

−2

−1

0

1

θ

2

Figure 13.2. Contours of constant total energy indicating the changing catchment region as a function of initial imperfection.

free oscillations within the context of constant total energy, the level of which is determined by the initial conditions. For example, in Fig. 13.2 are shown two contour plots of total mechanical energy when the axial load has been set to = 0.5, that is, at 50% of the critical load for the initially perfect geometry. In part (a), the initial imperfection is θ0 = 0.01, and in part (b) θ0 = 0.1. We see that, as expected for an imperfection-sensitive structure, the difference in potential energy between the stable and closest unstable equilibrium has diminished for the larger initial imperfection, and thus, given a certain initial velocity or step input, we would have a greater chance of an escaping solution. For example, when θ0 = 0.01 we have a stable equilibrium at θe = 0.020 with adjacent saddle points at θe = −1.0538 and 1.0404, and when θ0 = 0.1 we have a stable equilibrium at θe = 0.2055 with the adjacent saddle points at θe = −1.1043 and 0.9654. Going back to Fig. 5.7, we saw some large-amplitude motion that was due to initial conditions somewhat distant from equilibrium. These were, of course, highly nonlinear, and in general recourse is made to numerical integration to solve these types of equations of motion. However, the large excursions generated by the application of a sudden load (or relatively large initial conditions) may lead to escaping solutions, which correspond in some sense to a dynamic buckling load. Consider the link model shown in Fig. 5.16. Suppose the load is held fixed at p = 0.005. In this case, there are three relevant equilibria: θe = 0.036789 (stable), θe = 0.325084 (unstable), and θe = 0.813625 (stable). For a system at rest (equilibrium), we may prescribe an initial velocity such that enough energy is imparted to the system that the subsequent transient traverses the hilltop. We can extract the critical velocity from the total energy and find θ˙ 0 ≈ 0.0496 if θ(0) = 0.036789. This can be considered as an impulse given to the system, with an initial velocity of θ˙ (0) > 0.0497 resulting in motion that goes beyond the distant equilibrium—undergoing oscillations that are far from sinusoidal. The trajectory passes through the unstable equilibrium position: a potential-energy hilltop.

18:8

P1: KAE Chapter-13

CUFX159-Virgin

0 521 85648 5

August 14, 2007

264

Suddenly Applied Loads 1

(a)

1

λ f (t)

_ ξ+ξ M ω12 = K/M λ c = K/2

λc λ

F = K ( ξ + b ξ3 )

(b)

_

1

0.8

λS λC

_

(c) 1.0

0.9

λs

0.6 0.4

0.8

0.2 0.7 0.2

0.4

−0.06 −0.03

0.03

0.06

_

Figure 13.3. (a) A schematic of a link model, (b) equilibrium paths for the perfect and imperfect system illustrating an unstable-symmetric point of bifurcation, and (c) imperfection sensitivity [7].

An Approximate Analysis. An early study of link models subject to suddenly ap-

plied loading can be found in Budiansky [7]. In this paper, an approximate treatment was given based on perturbation theory for imperfection-sensitive structures (in which case instability is associated with some kind of severe collapse). We start by considering the case in which a step load is applied to a link model. Imperfection sensitivity was first encountered in Subsection 3.4.3 and a specific example given in Section 5.3. We again consider a simple model, shown in Fig. 13.3(a). Following the analysis in Budiansky [7] the nonlinearity in this system is solely due to the spring characteristic rather than to any kind of large-deflection effect encountered in Chapter 5. Also, the bars (of unit length) are assumed to be rigid but weightless, and all the mass is concentrated at the central hinge (to give a natural frequency of unity). Adopting an approximate (Galerkin) approach to this system, we obtain an equation of motion,

1 ¨ λ f (t) ¯ λ f (t) 3 ξ + bξ = ξ+ 1− ξ, λc λc ω21

(13.6)

18:8

P1: KAE Chapter-13

CUFX159-Virgin

0 521 85648 5

August 14, 2007

13.2 Back to Link Models

265

with underlying (cubic) equilibrium paths of the form [for f(t) = 1] ¯ (1 − λ/λc )ξ + bξ3 = (λ/λc )ξ.

(13.7)

Given b = −1, the solutions of Eq. (13.7) are plotted in Fig. 13.3(b) for both the perfect (ξ¯ = 0) and imperfect (ξ¯ = 0.025) cases. We thus confirm that in the presence of an initial imperfection (which may also be a small lateral load) the load-carrying capacity of the structure is somewhat diminished. The extent of the reduction of the maximum load is thus given by √ 3 3 3/2 ¯ s /λc ) = 0, (−b)1/2 |ξ|(λ (13.8) [1 − (λs /λc )] − 2 and this expression is plotted in part (c) of Fig. 13.3. Assuming the axial load λ is applied quasi-statically (as in Chapter 5) then these results clearly resemble those shown in Fig. 5.10. For the specific case shown (ξ¯ = 0.025), the maximum load (λs = 0.855) corresponds to a maximum deflection of ξ = 0.27. Now, suppose the load is applied suddenly as a step input, for example, by instantaneously removal of the support from a weight attached to the structure. In this case, a first integral can be performed (assuming there is no damping present) on Eq. (13.6), and from the resulting energy contours the maximum displacement (when ξ˙ = 0) is obtained from 3 ¯ /4 = (λ/λc )ξ. [1 − (λ/λc )] ξmax /2 + bξmax

(13.9)

This relation is shown in Fig. 13.4(a). The maximum deflection for which bounded solutions exist is defined as the dynamic buckling load λD and is determined from dλ/dξmax = 0, that is, √ ¯ D/λc ) (13.10) [1 − (λD/λc )]3/2 = 1.5 6(−b)1/2 |ξ|(λ now corresponds to a maximum (step) load of λD = 0.821, as shown in Fig. 13.4(a) with a maximum deflection of ξ = 0.345. (a)

c

(b)

_

1

D 0.8

s

_

0.6 0.4

D

0.2

0.2

0.4

0.6 max

s

c

Figure 13.4. (a) Step loading vs. maximum deflection and (b) magnitude of the dynamic buckling load relative to the static case [7].

18:8

P1: KAE Chapter-13

CUFX159-Virgin

0 521 85648 5

August 14, 2007

266

Suddenly Applied Loads (a)

0.5 0.4

0.4

0.3

0.3

0.2

0.2

0.1

0.1

5

10

15

20

25

(b)

0.5

200

30

400

600

1000

800

t

t

Figure 13.5. (a) Response of the structure that is due to a step load and (b) slow evolution of the load.

Equations (13.8) and (13.10) can now be combined to give the result

1 − (λD/λs )(λs /λc ) 1 − (λs /λc )

3/2 =

√ λD 2 , λs

(13.11)

which is plotted in Fig. 13.4(b), in which the same level of initial imperfection is used for both the static and dynamic cases. For example, if the structure has an initial imperfection of ξ¯ = 0.025, then there is a 0.821/0.855 = 0.96 reduction in the maximum step load (relative to maximum statically applied load), as shown by the dashed line in Fig. 13.4(b). Thus we see, as anticipated, that when the load is suddenly applied the (imperfection-sensitive) structure is able to withstand only a reduced loading condition, and this effect is proportionately lower for a structure with a larger initial imperfection. We finally conduct a couple of simple numerical simulations of this system. Applying the step load to Eq. (13.6) and using zero initial conditions leads to the results shown in Fig. 13.5(a). The preceding analysis gave a critical dynamic buckling load of λ = 0.821, and results are shown for the cases λ = 0.82, “stable,” and λ = 0.822, “unstable.” In part (b) the load parameter λ is very slowly evolved from zero (as was done in Chapter 5) until the system loses stability close to t ≈ 850. It is worth mentioning here that given the relatively small level of initial imperfection there is not a large difference between the static and dynamic conditions. Also, the trajectory shown in Fig. 13.5(b) is not quite in equilibrium but rather the load is incremented in a large number of small steps as a function of the time step. These results do not include the effect of damping, and, as such, the results of numerical error may be an issue [8]. An alternative approach to this type of problem was conducted in Thompson [9] with the introduction of the concept of an astatic buckling load. This approach was also tested against some experiments on a continuous type of thin elastic arch structure, an example of which is shown in Fig. 13.6. The top-right-hand portion of this figure shows an arch with pinned ends but a rigid connection at the center and also

18:8

P1: KAE Chapter-13

CUFX159-Virgin

0 521 85648 5

August 14, 2007

13.3 Dynamic Buckling of a Plate

267

Offset, ε (inches) Figure 13.6. Step loading of a shallow arch. Reprinted with permission from the author [9].

shown in its deflected configuration. This structure was subject to a number of previous static tests [10] [as an example of an unstable-symmetric (subcritical) point of bifurcation], but here the suspended weight is suddenly released. The magnitude of the mass () and the offset from the apex (initial imperfection ) are then related to the (dimensional) dynamic buckling results of Fig. 13.6, in which M represents the static load, D is the dynamic load, and N is the theoretical result from Thompson [9]. Again we observe the detrimental effect of applying the load suddenly. In general, the applied force may have a finite duration, and thus the structure may be subject to a pulse. For longer durations (relative to the natural dynamics of the structure), we approach the previous results of the step load. For relatively short pulses, we approach an impulse, which we have already shown has an equivalence to a nonzero initial velocity. The effect of duration length was also considered in Thompson [7] based on the earlier work described in Hutchinson and Budiansky [11] as well as extensive studies in Simitses [12].

13.3 Dynamic Buckling of a Plate In Subsection 10.1.4, we considered a simply supported rectangular plate subject to a uniaxial load. Now, suppose this load is applied dynamically. Such a case was considered in Zizicas [13] and reproduced in Bulson [14] based on small-deflection plate theory but included the effect of an initial imperfection. In this case, it can be shown that the governing equation of motion is D∇ 4 (w − w0 ) + ρ

∂ 2w ∂ 2w − N = 0, x ∂t 2 ∂x 2

(13.12)

18:8

P1: KAE Chapter-13

CUFX159-Virgin

0 521 85648 5

August 14, 2007

268

Suddenly Applied Loads 2

10 _ Q0

1.2

1.05

30

1.0 0.95

25 20 15 10

0.8 0.7

5 0.6

2

4

6

8

10

Figure 13.7. Deflection of a simply supported plate subject to a dynamic axial load [13].

in which w0 is given by Eq. (10.49). The critical load and lowest natural frequency are given by Eqs. (10.46) and (10.45) (with m = n = 1), and again these provide convenient values with which to nondimensionalize the equation of motion, that is, introducing Nx /Ncr = α and ωt = τ, we can rewrite Eq. (13.12) as d 2 [δ/Q0 ] δ + (1 − α) − 1 = 0, 2 dτ Q0

(13.13)

in which δ is the central deflection caused by the dynamic load and Q0 is the magnitude of the initial imperfection [see Eq. (10.49)]. Solutions to Eq. (13.13) are shown in Fig. 13.7. The axial load tends to magnify the initial imperfection according to Eq. (10.50), and it is apparent that the maximum response of the oscillation when the load is applied dynamically is larger than that for the quasi-static case. For example, when the magnitude of the axial load is 60% of the critical magnitude (α = 0.6), the static deflection is 2.5 that of the initial value [from Eq. (10.50)], whereas the suddenly applied load at this level results in a peak-to-peak response that oscillates between δ/Q0 = 1 and δ/Q0 = 4 (and centered on δ/Q0 = 2.5). This result was anticipated from the introduction to this chapter and the overshoot of 100% for an undamped linear system subject to a step input. Clearly, we have the result that if the magnitude of the suddenly applied load is equal to, or greater than, the critical static load, then a monotonic growth of deflection occurs without bound (keeping in mind the limitations imposed when small-deflection theory is used).

13.4 A Type of Escaping Motion In the next chapter, we shall focus attention on harmonically excited systems. In that case, most of the interest involves steady-state behavior. But even with harmonic excitation we can still expect a certain amount of transient behavior: For example, a system starting from rest will take awhile to settle down. It may very well be that the largest excursions (and hence proximity to instability) will occur during this transient stage [15, 16]. We go back to the basic form of an equation of motion in the

18:8

P1: KAE Chapter-13

CUFX159-Virgin

0 521 85648 5

August 14, 2007

13.4 A Type of Escaping Motion

269

vicinity of a transcritical bifurcation [Eq. (3.53)] and appeal to the analogy of a small ball rolling on the underlying potential-energy surface given by V=

x2 x3 −µ . 3 2

(13.14)

For negative values of the control parameter µ, we have a stable (trivial) equilibrium state and an unstable equilibrium state (a maximum of the potential energy). As µ becomes less negative, these equilibria approach each other and interchange stability at µ = 0 (see Fig. 3.6). However, it is important to realize that the domain of stability (against disturbances) changes. For an undamped system, we can construct a separatrix emanating from the hilltop (saddle) as a contour of constant total energy. Hence, any trajectory starting within this region will lead to constrained, or bounded, motion. But, again, we see the possibility of some trajectories escaping. This scenario is certainly complicated by the presence of damping. The area within the separatrix contains those initial conditions that do not lead to escape. However, this area changes (shrinks) as we approach the critical (buckling) condition. Thus we can imagine a situation in which a given step input for a relatively large negative level of µ will not cause escape (leading to infinity) whereas the same step applied to the system with a µ value less negative may very well lead to escape. Clearly, the magnitude of the step plays a crucial role as well, as seen in the previous section. Thus we have effectively described the two scenarios found in Fig. 1.1 at the start of this book. A structure may buckle because of: r a deteriorating stiffness caused by increasing axial loading (the natural frequency characterizes this essentially static behavior) or r an excessive disturbance applied to a structure with a given level of axial loading. This is essentially a transient, dynamic behavior. Returning to the case of a harmonically forced, axially loaded structure, we can again consider a system described by Eq. (13.1), but now, rather than a linear spring, we shall assume a force that is quadratically related to deflection. Changing the step input to a harmonic drive, we consider x¨ + ζ x˙ + x(1 − x) = F2 sin (t + φ).

(13.15)

Assuming the initial conditions are zero, we have the possibility that the forcing may be sufficient to cause the response to exceed x = 1, thus leading to solutions escaping to infinity. However, this is a more involved issue than finding the critical initial conditions of the previous case. We can make use of the ball rolling on the potential-energy surface analogy but now the surface itself is shaken harmonically (thus resulting in a 3D phase space). It turns out that horizontally shaking the potential-energy surface is akin to transmissibility [17], a related but slightly different case from direct forcing, and thus an additional factor of 2 also appears in the forcing magnitude [18]. That is, the mass is not subject to a direct force, but rather indirectly through a base displacement. Clearly, the larger F is, the greater the likelihood of escape, but we also anticipate a resonant type of effect if the forcing

18:8

P1: KAE Chapter-13

CUFX159-Virgin

0 521 85648 5

270

August 14, 2007

Suddenly Applied Loads

Figure 13.8. Parameter space indicating whether transients escape.

frequency is somewhat close to the linear natural frequency of the system (unity). Furthermore, it would not be unreasonable to expect a cosine function to generate larger transients starting from t = 0 than a sine function would, and thus φ has an effect. Rather than simply choose a number of forcing parameters and simulate Eq. (13.15) to determine whether steady-state oscillations persist, we can gain a more complete picture by conducting a thorough investigation of parameter space (F, ) by dividing it into a fine grid and labeling the outcome. An example of such a plot is shown in Fig. 13.8. What this figure describes is the result of many thousands of numerical simulations (all starting from zero initial conditions) and the areas shaded white indicate regions of parameter space that resulted in nonescaping behavior. The black-shaded regions led to escaping motion, which can be thought of as dynamic buckling. It can be seen that the boundary between escape and no-escape is not simple, and in fact exhibits certain fractal properties [19]. That is, on close inspection the border is nonsmooth (and self-similar at finer and finer scales) such that given any reasonable (small) uncertainty in the forcing parameters it may be difficult to tell whether the motion will escape or not, even though this is a thoroughly deterministic scenario. There is an increased likelihood of escape when the forcing frequency () is relatively close to one, as there is some associated softening resonant effect. This aspect of nonlinear behavior will be explained more throughly in later chapters.

18:8

P1: KAE Chapter-13

CUFX159-Virgin

0 521 85648 5

August 14, 2007

13.4 A Type of Escaping Motion

x

271 (a)

1

x

(b)

1 0.5

0.5

t

t 20

40

60

80

20

100

-0.5

-0.5

-1

-1

x

60

80

100

(d)

(c)

1

40

x

1

0.5

0.5

t 20

40

60

80

t 20

100

-0.5

-0.5

-1

-1

40

60

80

100

Figure 13.9. Some typical time series for parameter values spanning the escape boundary, = 0.9, ζ = 0.1: (a) F = 0.107, (b) F = 0.1075, (c) F = 0.108, (d) F = 0.1085.

Figure 13.9 shows four typical time series generated with the same (quiescent) initial conditions and forcing frequency except for a very slightly different forcing amplitude in each case. We see that whether the trajectory escapes or not is a sensitive function of the forcing parameters (at least in certain ranges of the parameter space). In part (a), the motion passes quite close to the potential hilltop (at x = 1) but does not lead to escape. There is a small possibility of escape after the time range of the simulation. An informative view of a trajectory can be found in the phase projection (the phase space is 3D) shown in Fig. 13.10. This picture corresponds to the time series in part (d) of Fig. 13.9. We see that these oscillations are far from sinusoidal (which would yield elliptical trajectories), and it is interesting to see that the trajectory even passes beyond x = 1 (briefly) before coming back

. x

1 0.5

Figure 13.10. A phase projection corresponding to the escaping trajectory when F = 0.1085.

-0.5

0.5 -0.5 -1

1

1.5

x

2

18:8

P1: KAE Chapter-13

CUFX159-Virgin

0 521 85648 5

August 14, 2007

272

Suddenly Applied Loads

and then finally escaping [15, 19–21]. A final feature worth mentioning here is that although Fig. 13.8 divides the parameter space into two regions, within the black areas it is not unreasonable to ask the question: How long does it take for a given trajectory to escape? It turns out that trajectories corresponding to relatively high forcing magnitudes (F ) tend to escape quickly (within a couple of forcing cycles or so), and trajectories corresponding to parameters very close to the boundary may take a relatively long time before finally escaping. This is also an aspect of nonlinear behavior that will be revisited in the final chapter.

13.5 Impulsive Loading In this section, we focus attention on what happens as the duration of a suddenly applied load approaches zero. In the limit, we deal with impulsive loading. For a SDOF system, it can easily be shown that this situation is equivalent to a system subjected to a nonzero initial velocity. The situation is a little more complicated for higher-order systems [22]. Here, we consider a two-DOF (2DOF) link model that is configured as an arch, such that the loading-deflection path has a bifurcation point before a limit point is reached, and the structure buckles into an asymmetric mode. Consider the system shown in Fig. 13.11. This system was analyzed extensively in [12, 22], and here we focus on impulsive loading and use a total energy approach following that reference. This system has two hinges of rotational stiffness β with concentrated masses m, and these are the locations at which two equal vertical loads P are applied. Deflection of the system is allowed by horizontal sliding at the righthand support, resisted by a linear spring of stiffness k. The initial rise of the arch is characterized by the angle α (initially equal at each end). The deflected state of the system is described in terms of the end angles θ and φ [22]. Other than the method of external forcing, this structure is similar to the one shown in Fig. 5.15. Assuming small angles, the total potential energy can be written as

P

P

m

L

β

m β

L L α

θ

φ k

Figure 13.11. A 2DOF link model [22].

18:8

P1: KAE Chapter-13

CUFX159-Virgin

0 521 85648 5

August 14, 2007

13.5 Impulsive Loading

UTP =

273

1 β(5θ + 5φ2 − 2α2 − 8φθ − 2θα − 2φα) 2 +

kL2 2 (α − θ2 − φ 2 + φθ)2 − PL(2α − θφ). 2

(13.16)

It is convenient to introduce new displacement variables ¯ r = (θ + φ)/2 β, ¯ s = (θ − φ)/2 β,

(13.17)

in which β¯ = β/(kL2 ). In terms of these new variables, we can write the total potential energy as √ √ 1 P U¯ T = (r2 + 9s2 − 2 r + ) + ( − r2 − 3s2 )2 − 2p( − r), 2

(13.18)

in which the following nondimensional parameters have also been used P U¯ T = UTP /(β¯ 2 kL2 ),

p = P/(kLβ¯ 3/2 )

¯ = α2 /β.

(13.19)

13.5.1 Equilibrium Behavior In the usual way, equilibrium is obtained from stationary values of the potential ¯ ¯ energy (∂U/∂r = 0 and ∂U/∂s = 0): 2(r −

√ ) − ( − r2 − 3s2 )2r + 2p = 0, 18s − ( − r2 − 3s2 )6s = 0.

(13.20)

The solutions to these equations give the symmetric response (s = 0) ( − 1 − r2 )r = p −

(13.21)

and the asymmetric response (s = 0) − 3 = r2 + 3s2 , 2r = p − .

(13.22)

The type of resulting behavior obviously depends on the magnitude of . It can be shown [12] that if > 4 then the system encounters an unstable point of bifurcation, and because this is a new feature we now focus on the specific case = 6. Figure 13.12(a) shows the potential-energy contours in terms of the two coordinates r and s. Five equilibrium points exist: Points 1 and 3 are stable equilibria, with an unstable point 2 in between; points 4 and 5 are unstable saddle points.

18:8

P1: KAE Chapter-13

CUFX159-Virgin

0 521 85648 5

August 14, 2007

274

Suddenly Applied Loads 1.5

(a)

s 1

4

0.5

3

1

0

2

–0.5

5

–1

–1.5 –2

–1

0

1

2

r (b)

10

p 7.5 5 2.5

r 1

2

Figure 13.12. (a) Contours of potential energy indicating five extremum points, = 6, (b) corresponding equilibrium curves [12].

13.5.2 Behavior under Sudden Loading If the system is initially located at point 1 in stable equilibrium and is then subject to a disturbance (impulse), it is apparent that a sufficiently large input may cause the system to traverse the unstable equilibrium (which is a local maximum of the potential energy). Furthermore, we can imagine the situation in which sufficient kinetic energy is imparted to the system such that either of the saddle points is traversed, typically leading to very large-amplitude oscillations. The equilibrium curves are projected in Fig. 13.12(b) as a function of p. Simitses [12] used an energy approach and the impulse-momentum theorem in which an impulse (PT0 ) associated with a load P over a short time duration (T0 ) is related to kinetic energy and the potential energy at the hilltop equilibrium to arrive

18:8

P1: KAE Chapter-13

CUFX159-Virgin

0 521 85648 5

August 14, 2007

13.6 Snap-Through of a Curved Panel

275

14

2( pτ 0)cr 12

barrier 1

10

8

6

4

barrier 2

2

4

6

8

10

Λ

12

Figure 13.13. Upper and lower bounds for the critical ideal impulse [22].

at analytical expressions for the two instability situations described in the previous ¯ paragraph. Using the nondimensional time τ = t (βk/2m), we find that the relevant expressions are 2(pτ0 )cr = 3 − 1,

2(pτ0 )cr = [5 − 2 − 3 2 − 4 + 0.25( + 2 + 2 − 4)2 ]1/2 ,

(13.23)

where τ0 is the nondimensional duration of the impulse. The expressions in Eqs. (13.23) are plotted in Fig. 13.13. The barriers represent levels of the dynamic forcing required to cause instability. Thus we see that when = 6, the critical impulse that causes dynamic buckling is given by 2(pτ0 )cr = 6.7. These results on discrete-link models can also be extended to continuous, suddenly loaded structures liable to snap-through [23]. For example, some early studies were conducted by Hsu (e.g., [24]) on shallow elastic arches under the action of various time-dependent lateral loads, including sinusoidal, arbitrary, concentrated, etc. The effects of initial thrust and elastic foundations were investigated [25] and interaction curves developed to assess the effect of various load combinations on the snap-buckling of shallow arches [23, 26–30]. Some interesting features associated with this type of problem are described in [31, 32] in which impulsive and harmonic loading may lead to counterintuitive behavior.

13.6 Snap-Through of a Curved Panel We considered the snap-through (saddle-node bifurcation) of a simple link model under the action of a quasi-statically increasing load in Section 5.6. For a continuous (shell-like) system under suddenly applied (step) loading, it is also possible to

18:8

P1: KAE Chapter-13

CUFX159-Virgin

0 521 85648 5

August 14, 2007

276

Suddenly Applied Loads 5.31

max/v/ (m) 0.3

(a) p

0.2 0.1

a

0.12

0.36

0.24

0.0937 p (kN/m2 )

t (s)

t=0

0.0

0.0

(b) 0.2954

p(t)

0.48

2.5

5.0

time, t (s)

−1 0 1

0.0 p = 2.5

5

7.5

. v (m/s)

5.29 5.2

0.08

2

0.0937

5.3 kN/m

5.31 kN/m 2

0.16

0.24

v (m) p=7

6

5.5 5.4 5.35 5.32

(c)

v (m)

Figure 13.14. Snap-through of a shallow panel subject to step loading: (a) the geometry and loading, (b) maximum response as a function of loading intensity, and (c) time series and phase projec¨ ¨ tion. Adapted from Kratzig and Nawrotzki [35] and Kratzig [36].

encounter snap-through [33]. However, for such a system the FE technique is the method of choice. ¨ An example of this behavior was given by Kratzig [34], in which a shallow curved panel is subject to a uniformly distributed vertical step load of magnitude p, as shown in Fig. 13.14(a). For this particular simulation, a square panel of width 10 m, thickness 0.1 m, and a radius of curvature of 100 m is considered. Young’s modulus is 3.4 × 107 kN/m2 and Poisson’s ratio is 0.2. With the rest position used as the initial condition, the results of numerical simulation shown in part (b) indicate that provided the magnitude of the step load is less than p = 5.31 kN/m2 the response is insufficient to traverse the potential-energy hilltop associated with the unstable equilibrium. Thus there is a considerable difference in the maximum response of the system depending on whether the system tends toward its inverted equilibrium or not. In both cases damped oscillations characterize the system response as energy is dissipated. A set of superimposed times series and phase projections is shown in Fig. 13.14(c) for various values of the applied load. The phase projection shows two trajectories, one on each side of the critical value. Unlike the simple link models of Chapter 5, this type of modeling requires special numerical procedures and is

18:8

P1: KAE Chapter-13

CUFX159-Virgin

0 521 85648 5

August 14, 2007

13.6 Snap-Through of a Curved Panel

277

the subject of current research [35, 36], including similar sensitivity in spherical caps [37]. Suddenly Loaded Column. We conclude this chapter by considering an axially

loaded column in which the end load is applied suddenly [1, 2, 38–41]. The following chapter will look at a pulsating end load in which case it is possible to get buckling for loads lower than the static buckling load. In this final section, we see that it is possible to have stability for loads far in excess of the static buckling load depending on the duration of the load. We focus attention on pulse loading of the type shown schematically (but with finite duration) in the inset of Fig. 13.1 [42, 43], and this is closely related to the impact loading of a bar, for example, a hammer hitting a nail [44]. Although the load is transmitted as an axial stress wave, it has been shown that typically buckling motion as the wave passes can be neglected, and the total length of the column is relatively unimportant; that is, the duration of the axial loading is relatively large compared with the period of longitudinal vibration of the bar. One consequence of this is that the buckling modes can be quite complicated (and associated with short buckling wavelengths), with divergent (hyperbolic) and bounded (trigonometric) modes both present and their separation depending on various characteristics of the system. A subtlety associated with this problem (other than the direct influence of dynamics of course) is that initial imperfections are necessary, and thus, rather than having a distinct bifurcational event, the loss of stability is most appropriately couched in terms of a dynamic growth, or amplification, of the initial geometric imperfection. We assume in the analysis that the behavior is elastic, although of course in practice very often plastic deformation is encountered [45]. We briefly discuss the simple case of a uniform, simply supported bar, subject to a suddenly applied load: ∂ 4w ∂2 ∂ 2w + P 2 (w + w0 ) + m 2 = 0. (13.24) 4 ∂x ∂x ∂t Note that in contrast to the initial imperfection encountered in Section 7.3, here, we measure the deflection w(x, t) that is due to axial loading from the initial imperfection w0 (x). In light of the earlier comments about the relative unimportance of the column length (see Lindberg and Florence [46]) we introduce a characteristic length of 1/k (where k 2 = P/EI) and then nondimensionalize Eq. (13.24) by using EI

x¯ = kx,

w¯ = w/r,

t¯ = trk 2 c,

where c = E/m is the speed of wave propagation and r = ration. The result is w¯ + w¯ + w¨¯ = −w¯ 0 .

(13.25) √

I/A is the radius of gy(13.26)

Application of the simply supported boundary conditions (w¯ = w¯ = 0) at (x¯ = 0 and x¯ = l = kL) leads to a solution of the form w( ¯ x, ¯ t¯) =

∞ n=1

gn (t¯) sin

nπx¯ , l

(13.27)

18:8

P1: KAE Chapter-13

CUFX159-Virgin

0 521 85648 5

August 14, 2007

278

Suddenly Applied Loads

and we assume a spatial distribution of initial imperfection according to w¯ 0 (x) ¯ =

∞

an sin

n=1

where 2 an = l

l

w¯ 0 (x) ¯ sin 0

nπx¯ , l

(13.28)

nπx¯ d¯x. l

(13.29)

Thus the solution for the amplitudes is obtained from g¨ n + η 2 (η 2 − 1)gn = η 2 an ,

(13.30)

where η = nπ/l is the wavenumber. We see the form of solution is quite different according to whether η is greater than, or less than, unity. This was a distinction in the form of the solution we first encountered at the start of Chapter 7, and assuming the bar is initially at rest (i.e., w¯ = w˙¯ when t = 0) we obtain the solution w( ¯ x, ¯ t¯) =

∞ n=1

an nπx¯ (cos p n t¯ − 1) sin 1 − η2 l

(13.31)

an nπx¯ (cosh p n t¯ − 1) sin 1 − η2 l

(13.32)

when η > 1, and w( ¯ x, ¯ t¯) =

∞ n=1

when η < 1, and where p n = η|1 − η 2 |1/2 .

(13.33)

In terms of amplification, or lateral growth of motion, it is convenient to scale Eqs. (13.31) and (13.32) according to the underlying static amplification of an imperfect simply supported, axially loaded bar, and thus gn (t¯) 1 nπx¯ cosh Gn (t¯) = p n t¯ − 1 sin , (13.34) = an 1 − η 2 cos l in which we take the cosine term for η > 1 and the hyperbolic cosine term for η < 1. Equation (13.34) is plotted in Fig. 13.15(a) for two values of the nondimensional time t¯. From this we can see that greatest amplification takes place in a narrow band of wavelengths (the preferred mode of buckling). We can then make use of the derivative to find the maximum value and then plot that maximum value versus time, as shown in Fig. 13.15(b). By use of an approximate analysis [46], it can be shown that under very high compression the bar will buckle into wavelengths close √ to 8.88r/ P/AE with very rapid growth in motion after approximately t¯ = 4. The effect of the velocity of impact loading was studied in Holzer and Eubanks [47] and Hayashi and Sano [48, 49]. We can also approach this type of problem by considering the dropping of a weight onto the end of a long strut, which would be an easy scenario to set up in

18:8

P1: KAE Chapter-13

CUFX159-Virgin

0 521 85648 5

August 14, 2007

References

279 200

G

14

(b)

(a)

12

150

t=5

10

Gmax

8

100

6 4

50

t=3

2 0.5

1

1.5

0

2

t Figure 13.15. Amplification vs. wavenumber, (b) corresponding maximum amplification as a function of time [46].

the laboratory. In this case, it can be shown that the buckling wavelength scales with the inverse square root of the impact speed [50]. In a practical sense, the distribution of initial geometric imperfections might have a random element, and this has also been treated in the literature [51]. Extensive studies using similar approaches have focused on rapid application of axial loading on cylinders and shells [51–53]. Finally, the case of a dynamic application of the end load that is relatively slow is mentioned. In this type of ramp function, which is somewhat representative of what would happen in a loading machine, there is also some interesting behavior [55].

References [1]

J.H. Meier. On the dynamics of elastic buckling. Journal of the Aeronautical Sciences, 12:433–40, 1945. [2] J.F. Davidson. Buckling of struts under dynamic loading. Journal of the Mechanics and Physics of Solids, 2:54–66, 1953. [3] G. Herrmann. Dynamic Stability of Structures. Pergamon, 1967. [4] G.J. Simitses, A.N. Kounadis, and J. Giri. Dynamic buckling of simple frames under a step load. Journal of Engineering Mechanics, 105:896–900, 1979. [5] A.T. Brewer and L.A. Godoy. On interaction between static and dynamic loads in instability of symmetric or asymmetric structural systems. Journal of Sound and Vibration, 147:105–14, 1991. [6] I. Cochin and H.P. Plass. Analysis and Design of Dynamic Systems. Harper & Row, 1990. [7] B. Budiansky. Dynamic buckling of elastic structures: Criteria and estimates. In G. Herrmann, editor, Dynamic Stability of Structures. Pergamon, 1967. [8] G.J. Simitses. Effect of static preloading on the dynamic stability of structures. AIAA Journal, 21:1174–80, 1983. [9] J.M.T. Thompson. Dynamic buckling under step loading. In G. Herrmann, editor, Dynamic Stability of Structures. Pergamon, 1967. [10] J. Roorda. Stability of structures with small imperfections. Journal of the Engineering Mechanics Division, ASCE, 91:87, 1965.

18:8

P1: KAE Chapter-13

CUFX159-Virgin

280

0 521 85648 5

August 14, 2007

Suddenly Applied Loads [11] J.W. Hutchinson and B. Budiansky. Dynamic buckling estimates. AIAA Journal, 4:525– 30, 1966. [12] G.J. Simitses. Instability of dynamically-loaded structures. Applied Mechanics Reviews, 40:1403–8, 1987. [13] G.A. Zizicas. Dynamic buckling of thin elastic plates. Transactions of the ASME, 74:1257–68, 1952. [14] P.S. Bulson. The Stability of Flat Plates. Chatto and Windus, 1970. [15] B. Budiansky and E.S. Roth. Axisymmetric dynamic buckling of clamped shallow spherical shells. Technical Report, NASA TND-1510, 1962. [16] M.W. Hilburger, A.M. Waas, and J.H. Starnes. Modeling the dynamic response and establishing post buckling snap-through equilibrium of discrete structures via a transient analysis. Journal of Applied Mechanics, 64:590–5, 1997. [17] D.J. Inman. Engineering Vibration. Prentice-Hall, 2000. [18] L.N. Virgin. Introduction to Experimental Nonlinear Dynamics: A Case Study in Mechanical Vibration. Cambridge University Press, 2000. [19] J.M.T. Thompson. Chaotic phenomena triggering escape from a potential well. Proceedings of the Royal Society of London, A421:195–225, 1989. [20] L.N. Virgin, R.H. Plaut, and C.-C. Cheng. Prediction of escape from a potential well under harmonic excitation. International Journal of Non-Linear Mechanics, 27:357–65, 1992. [21] J.A. Gottwald, L.N. Virgin, and E.H. Dowell. Routes to escape from an energy well. Journal of Sound and Vibration, 187:133–44, 1995. [22] G.J. Simitses. Dynamic Stability of Suddenly Loaded Structures. Springer-Verlag, 1989. [23] D.L.C. Lo and E.F. Masur. Dynamic buckling of shallow arches. Journal of the Engineering Mechanics Division, ASCE, 102:901–17, 1976. [24] C.S. Hsu. Stability of shallow arches against snap-through under timewise step loads. Journal of Applied Mechanics, 35:31–9, 1968. [25] R.H. Plaut and E.R. Johnson. The effect of initial thrust and elastic foundation on the vibration frequencies of a shallow arch. Journal of Sound and Vibration, 78:565–71, 1981. [26] N.J. Hoff and V.G. Bruce. Dynamic analysis of the buckling of laterally loaded flat arches. Journal of Mathematical Physics, 32:276–88, 1954. [27] A.M. Liapunov. Stability of Motion (Collected Papers). Academic, 1966. [28] R.E. Fulton and F.W. Barton. Dynamic buckling of shallow arches. Journal of Engineering Mechanics, 97:865–77, 1971. [29] K.-Y. Huang and R.H. Plaut. Snap-through of a shallow arch under pulsating load. In F.H. Schroeder, editor, Stability in the Mechanics of Continua. Springer, 1982, pp. 215– 33. [30] M.T. Donaldson and R.H. Plaut. Dynamic stability boundaries for a sinusoidal shallow arch under pulse loads. AIAA Journal, 21:469–71, 1983. [31] E.R. Johnson and I.K. McIvor. The effect of spatial distribution on dynamic snapthrough. Journal of Applied Mechanics, 45:612–18, 1978. [32] R.H. Plaut and J.-C. Hsieh. Oscillations and instability of a shallow arch under twofrequency excitation. Journal of Sound and Vibration, 102:189–201, 1985. [33] J.S. Humphreys and S.R. Bodner. Dynamic buckling of shallow shells under impulsive loading. Journal of Engineering Mechanics, 88:17–36, 1962. ¨ ¨ [34] W.B. Kratzig. Nonlinear responses. In A.N. Kounadis and W.B. Kratzig, editors, Nonlinear Stability of Structures (Theory and Computational Techniques). Springer-Verlag, 1995.

18:8

P1: KAE Chapter-13

CUFX159-Virgin

0 521 85648 5

August 14, 2007

References ¨ [35] W.B. Kratzig and P. Nawrotzki. Computational concepts for kinetic instability prob¨ lems. In A.N. Kounadis and W.B. Kratzig, editors, Nonlinear Stability of Structures (Theory and Computational Techniques). Springer-Verlag, 1995. ¨ [36] W.B. Kratzig. Eine einheitliche statische und dynamische Stabilitatstheorie fur Pfad¨ Angeverfolgungsalgorithmen in der numerischen Festkorpermechanik. Zeitschrift fur wandte Mathematik und Mechanik, 69:203–13, 1989. [37] D. Dinkler and J. Pontow. A model to evaluate dynamic stability of imperfection sensitive shells. Computational Mechanics, 37:523–9, 2006. [38] C. Koning and J. Taub. Impact buckling of thin bars in the elastic range hinged at both ends. Luftfahrforschung, 10:55–64, 1933. [39] G. Gerard and H. Becker. Column behavior under conditions of impact. Journal of the Aeronautical Sciences, 19:58–60, 1952. [40] E. Sevin. On the elastic bending of columns due to dynamic axial forces including effects of axial inertia. Journal of Applied Mechanics, 27:125–31, 1960. [41] R. Grybos. Impact stability of a bar. International Journal of Engineering Science, 13:463–77, 1975. [42] N.J. Huffington. Response of elastic columns to axial pulse loading. AIAA Journal, 1:2099–2104, 1963. [43] I.K. McIvor and J.E. Bernard. The dynamic response of columns under short duration axial loads. Journal of Applied Mechanics, 40:688–92, 1973. [44] J.M. Housner and N.F. Knight. The dynamic collapse of a column impacting a rigid surface. AIAA Journal, 21:1187–95, 1983. [45] W. Abramowicz and N. Jones. Dynamic progressive buckling of circular and square tubes. International Journal of Impact Engineering, 4:247–70, 1986. [46] H.E. Lindberg and A.L. Florence. Dynamic Pulse Buckling: Theory and Experiment. Nijhoff, 1987. [47] S.M. Holzer and R.A. Eubanks. Stability of columns subjected to impulsive loading. Journal of Engineering Mechanics, 95:897–920, 1969. [48] T. Hayashi and Y. Sano. Dynamic buckling of elastic bars (the case of low velocity impact). Bulletin of the Japanese Society of Mechanical Engineering, 15:1167–75, 1972. [49] T. Hayashi and Y. Sano. Dynamic buckling of elastic bars (the case of high velocity impact). Bulletin of the Japanese Society of Mechanical Engineering, 15:1176–84, 1972. [50] J.R. Gladden, N.Z. Handzy, A. Belmonte, and E. Villermaux. Dynamic buckling and fragmentation in brittle rods. Physical Review Letters, 94(035503), 2005. [51] I. Elishakoff. Axial impact buckling of a column with random initial imperfections. Journal of Applied Mechanics, 45:361–5, 1978. [52] H.E. Lindberg and R.E. Herbert. Dynamic buckling of a thin cylindrical shell under axial impact. Journal of Applied Mechanics, 33:105–12, 1966. [53] N. Jones and C.S. Ahn. Dynamic elastic and plastic buckling of complete spherical shells. International Journal of Solids and Structures, 10:1357–74, 1974. [54] R. Kao. Nonlinear dynamic buckling of spherical caps with initial imperfections. Computers and Structures, 12:49–63, 1980. [55] N.J. Hoff. The dynamics of the buckling of elastic columns. Journal of Applied Mechanics, 17:68–74, 1953.

281

18:8

P1: KAE Chapter-14

CUFX159-Virgin

0 521 85648 5

August 14, 2007

14 Harmonic Loading: Parametric Excitation

14.1 An Oscillating End Load In Chapter 7 we considered the free vibrations of a simply supported beam subject to an axial load of constant magnitude, and then in Chapter 13 the load was applied suddenly. Now suppose the end load is pulsating harmonically, that is, we replace P in Fig. 7.1 with P + S cos t. Because the forcing will appear in the stiffness term we are dealing with parametric excitation, and this is sometimes referred to as vibration buckling in the literature [1]. The governing equation of motion is thus m

∂2 w ∂4 w ∂2 w + EI + (P + S cos t) = 0. ∂t2 ∂x4 ∂x2

(14.1)

With simply supported boundary conditions we take the solution in the form of a single half-sine wave, πx w = f (t) sin . (14.2) L Substituting this back into Eq. (14.1), reintroducing the term from Eq. (7.20), that is, PL2 EIπ4 ω2 = 1 − , (14.3) mL4 EIπ2 again normalizing the axial load and natural frequency by the Euler load (Pcr = π2 EI/L2 ) and frequency without load [ω20 = π4 EI/(mL4 )], respectively, and rescaling time according to τ = t, we arrive at f (τ) + (α + cos τ)f (τ) = 0,

(14.4)

where α = ω20 2 (1 − p),

= − ω20 2 s.

(14.5)

Although Eq. (14.4), which is called Mathieu’s equation, is linear, it is by no means easy to solve, and the character of the solutions (specifically their stability) depends in a nonsimple way on the parameters α and . This equation is also encountered in the dynamics of a pendulum with a harmonically shaken pivot [2], and it is related to the stability of forced oscillations which will be studied in more detail in the final chapter. Sinha [3] also considers the problem of pulsating axial load but applied to Timoshenko beams. To consider the stability of this type of system (which can be 282

18:9

P1: KAE Chapter-14

CUFX159-Virgin

0 521 85648 5

August 14, 2007

14.2 The Variational Equation

283

viewed as a natural extension of the stability of equilibria introduced in Chapter 3), we again make use of linearization before proceeding to Floquet theory [4–11].

14.2 The Variational Equation In this chapter, we consider nonautonomous dynamical systems of the form x˙ = f(x, t),

(14.6)

and proceed to consider the stability of solutions based on the behavior of small perturbations. In Chapter 3 we also considered the behavior of transients in the vicinity of equilibria (point attractors) in autonomous systems, whereas now we focus on transients in the vicinity of periodic solutions in nonautonomous (especially periodically forced) systems. Again linearization is a key concept, and use will be made of Poincare´ sampling and fixed points of maps. Equation (14.6) is in general a nonlinear system. Following the development in Subsection 4.2.3, we obtain the linear variational equation η˙ = DF(t)η.

(14.7)

It is this equation that governs the stability of solutions in the vicinity of special solutions (which will be steady-state oscillations in this context). Equations with Periodic Coefficients. Equation (14.4) is a specific case of the wider

class of forced vibration problems, and is clearly related to Eq. (14.7) in which the coefficients of the Jacobian DF(t) are periodic. Consider Hill’s equation [12], x¨ + G(t)x = 0,

(14.8)

where G(t + T) = G(t). Although this is a rather benign-looking (linear) equation, it does not, in general, submit to closed-form analytical solutions, and hence a variety of approximate techniques have been developed. We also note that a simple transformation allows the related case of damped parametric oscillations to be brought into the standard form of Eq. (14.8). Hence this equation represents quite a wide class of problems and has received considerable scrutiny over the years [12–15]. Placing Eq. (14.7) in a little more general (state-variable) context, and simplifying the notation by using A ≡ DF and x ≡ η, we consider the solutions of x˙ = A(t)x,

t ∈ R,

(14.9)

where x is an n-dimensional state vector and A(t) is a continuous T-periodic n × n matrix, that is, A(t + T ) = A(t). This system has n linearly independent fundamental solutions φi , where i = 1, 2, . . . , n, which can be expressed as a fundamental solution matrix (t) [15]. Shifting in time by T, we see that (t + T ) is also a fundamental matrix solution, and because they are linearly independent there is a nonsingular n × n matrix C such that

(t + T ) = (t)C.

(14.10)

18:9

P1: KAE Chapter-14

CUFX159-Virgin

0 521 85648 5

284

August 14, 2007

Harmonic Loading: Parametric Excitation

This constant matrix C (which depends on T but not on t) is called the monodromy matrix of Eq. (14.9) and contains the key stability information. The eigenvalues ρ of C are called characteristic (or Floquet) multipliers (CMs). They are uniquely determined and govern local divergence or convergence about a periodic orbit [16]. The matrix C can also be expressed as C = eBT ,

(14.11)

where B is a constant matrix. The Floquet theorem then states that the fundamental matrix (t) of Eq. (14.9) can be written as

(t) = P(t)eBt ,

(14.12)

where P(t) is T periodic and B is a constant n × n matrix. The eigenvalues γ of B are the characteristic (or Floquet) exponents (CEs) and are essentially the same as encountered earlier for equilibria, and they govern the stability of the trivial solution of Eq. (14.9). They are unique only to within an integer multiple of 2πi/T. It is helpful to consider the monodromy matrix C as a Floquet operator, which maps (t) onto (t + T ), and taking the initial condition as the identity vector

(0) = I, we then have from Eq. (14.10), C = (T ). Furthermore, if the eigenvalues ρi , i = 1, 2, . . . , n, of C (the CMs) are distinct, then Eq. (14.9) has n linearly independent normal solutions of the form xi = pi (t)eγi t ,

(14.13)

where the pi (t) are periodic functions with period T. Thus we see the fundamental relationship between the CEs and the CMs: ρ = eγt .

(14.14)

Now, again with distinct eigenvalues we can diagonalize C; that is, we define a transformed system

(t + T ) = (t)J,

(14.15)

where = M−1 CM, M is a nonsingular n × n constant matrix (chosen to simplify J ), and φ(t) = P−1 . In this case, we can again consider the relation between the CEs and the CMs but in component form, ψi (t + T ) = ρi ψ(t),

(14.16)

ψi (t) = eγi t φi (t).

(14.17)

with the Mathieu functions

The stability of the periodic solutions is now emerging. Equation (14.16) can be extended to ψi (t + NT ) = ρN i ψ(t),

(14.18)

where N is an integer. Therefore we see that it is the magnitude of ρ that determines stability:

18:9

P1: KAE Chapter-14

CUFX159-Virgin

0 521 85648 5

August 14, 2007

14.2 The Variational Equation

285

Loss of stability of a cycle Saddle node

Period doubling (Flip)

I

Neimark

I

I

R

R

R

Characteristic multipliers

Figure 14.1. The generic routes to instability (in terms of |ρi |) for a system under the action of a single control parameter.

r For stability of a periodic orbit, we have ψi (t) → 0 as t (and hence N) → ∞ if |ρi | < 1 (i.e., the real part of γi is negative). r For instability, we have ψi (t) → ∞ as t (and hence N) → ∞ if |ρi | > 1 (i.e., the real part of γi is positive). In practice, the most difficult part of any analysis of this kind is determining the matrix (t). We shall see that this can be achieved numerically by means of the Poincare´ map or by using various approximate analytical schemes. These will be dealt with in more detail in later chapters but we will see that the three typical ways in which multipliers leave the unit circle are shown in Fig. 14.1 [15, 17], with R and I signifying real and imaginary, respectively. We note that the Neimark bifurcation is less commonly encountered in the types of structural system encountered in this book. We can gain some stability insight by considering certain constraints, akin to the Routh–Hurwitz criterion for the stability of equilibria, placed on the eigenvalues. To do this, use is made of the Wronskian determinant of the fundamental matrix corresponding to Eq. (14.9): t Det (t) = exp Tr A(s) ds , (14.19) 0

where Tr A(s) is the trace of A(s). The Floquet theorem [Eq. (14.12)] tells us that Det (t) = Det[P(t)eBt ], which leads to

Det(eBT ) = exp

T

(14.20)

Tr A(t) dt .

(14.21)

0

Now we are in a position to state that ρ1 ρ2 . . . ρn = exp

T

Tr A(t) dt 0

(14.22)

18:9

P1: KAE Chapter-14

CUFX159-Virgin

0 521 85648 5

August 14, 2007

286

Harmonic Loading: Parametric Excitation

and n

γi =

i=1

1 T

T 0

2πi . Tr A(t) dt mod T

(14.23)

The procedure is quite straightforward for obtaining stability information provided the fundamental matrix of normal solutions, , is known. This is seldom the case, and a variety of approximate analytical and numerical techniques have been developed to account for this. Fortunately, there are a number of shortcuts that can be taken to determine stability without needing the full solution.

14.3 Mathieu’s Equation Using the information concerning the stability of periodic motion, we now return to Mathieu’s equation [Eq. (14.4)], where f is replaced with x: x (τ) + (α + cos τ)x(τ) = 0.

(14.24)

This can be expressed in state matrix terms (with x˙ ≡ y, replacing the primes):

x˙ 0 1 x = . (14.25) y˙ −α − cos t 0 y We see that the trace of the matrix in the preceding equation is zero, and using Eq. (14.22) applied to Eq. (14.25) we have ρ1 ρ 2 = e0 = 1, and thus the roots satisfy the quadratic equation ρ2 − φ(α, )ρ + 1 = 0.

(14.26)

The solutions are given by

1 φ(α, ) ± φ(α, )2 − 4 . (14.27) 2 In general, we might be more interested in determining whether the motion is bounded or not, rather than in being able to write the specific form of the solution, and hence the transition curves between stable and unstable behavior are of paramount importance, and these occur when φ(α, ) = ±2. For Mathieu’s equation, these transition curves correspond to the specific combinations of α and for which periodic solutions, with period 2π or 4π, occur [12]. The standard analytical approach to obtaining the transition curves involves a Hill determinant [12, 17]. However, a useful approximate technique based on the perturbation method is introduced that will also be useful when we consider largeamplitude vibration in the final chapter [2, 15]. ρ1 , ρ2 =

A Perturbation Solution. For relatively small values of the parameter , the transition curves can be computed with a perturbation method [15]. The solutions to Mathieu’s equation are assumed to be of the form

x(t) = x0 (t) + x1 (t) + · · · +,

(14.28)

18:9

P1: KAE Chapter-14

CUFX159-Virgin

0 521 85648 5

August 14, 2007

14.3 Mathieu’s Equation

287

in which x0 , x1 , . . . , have period 2π or 4π, and the transition curves are given by α = α() = α0 + α1 + · · · + .

(14.29)

On substituting these expressions into Mathieu’s equation, we obtain (x¨ 0 + α0 x0 ) + (x¨ 1 + α1 x0 + x0 cos t + α0 x1 ) + 2 (x¨ 2 + α2 x0 + α1 x1 + x1 cos t + α0 x2 ) + · · · + = 0,

(14.30)

and setting the coefficients of each power of equal to zero we obtain x¨ 0 + α0 x0 = 0,

(14.31)

x¨ 1 + α0 x1 = −(α1 + cos t)x0 ,

(14.32)

x¨ 2 + α0 x2 = −x0 α2 − (α1 + cos t)x1 , . . . ,

(14.33)

and so on. If we consider the solutions of Eq. (14.31) first, we see that we have simple harmonic motion of period 2π or 4π if α0 = (1/4)n2 with n = 0, 1, . . . . With n = 0, we have α0 = 0, x0 = 1 (assuming a unit displacement as the initial condition), and Eq. (14.32) becomes x¨ 1 = −α1 − cos t,

(14.34)

and for periodic solutions, we require α1 = 0 and thus, x1 (t) = cos t + c,

(14.35)

where c is a constant. Equation (14.33) then becomes x¨ 2 = −α2 − 1/2 − c cos t − 1/2 cos 2t,

(14.36)

and again, for periodic solutions we require α2 = −1/2, and thus, up to terms of second order in we have α = −1/22 + · · · + .

(14.37)

Repeating the analysis for n = 1, we arrive at α = 1/4 ± (1/2) − (1/8)2 + · · · +,

(14.38)

and for n = 2, α = 1 + (5/12)2 + · · · +,

(14.39)

α = 1 − (1/12)2 + · · · + .

(14.40)

These transition curves are plotted in Fig. 14.2, with the shaded regions indicating regions of unbounded growth of motion. A couple of numerical simulations show the form of the stable and unstable motion. The dashed curves within the unstable zones indicate the transition curves when a small amount of damping is added. This diagram is symmetric about the α axis but is plotted only for positive here. We are now in a position to interpret the dynamic response of the beam with a pulsating end load [18, 19]. The relation between α and and the forcing characteristics of the end load lead to the plot shown in Fig. 14.3. Here, we have focused on

18:9

P1: KAE Chapter-14

CUFX159-Virgin

0 521 85648 5

August 14, 2007

288

Harmonic Loading: Parametric Excitation

0.8 x

x

0.6 t

t 0.4 0.2

–0.5

0.5

0.0

1.0

Figure 14.2. The response of Mathieu’s equation in terms of the parameters α and , and based on a perturbation solution.

the instability arising from α = 0.25. When the static component of the axial load is zero, we see that an instability occurs if the forcing frequency of the oscillating part of the load is close to twice the natural frequency of the system. With either additional static compression or tension the parametric instability shifts according to the fundamental frequency of the beam with an axial load—a situation described at length in earlier parts of this book. Again, damping tends to have a stabilizing effect, such that even when the forcing frequency is exactly twice the natural frequency, there needs to be a certain amount of forcing magnitude to cause instability, and this tends to make the higher-order zones of instability practically disappear. We also note that this behavior is related to the issue of quasi-perodicity and Arnold tongues, found for example in the sine map [20–22].

14.4 Pulsating Axial Loads on Shells In much the same way that an oscillating axial load produces some interesting dynamic behavior in a column, a similar effect occurs in plates, panels, and shells [23, 24].

s 0.6

p = 0.5

p = −0.5

p=0

0.4

Figure 14.3. The stability of a prismatic beam with end load in which a portion of the load is oscillating. Principal parametric resonance.

0.2

0

1

2

3

4

5

7

6 2

18:9

P1: KAE Chapter-14

CUFX159-Virgin

0 521 85648 5

August 14, 2007

14.4 Pulsating Axial Loads on Shells

289

(b)

p(t)

(a)

0.6

unstable unstable

d cr

a

s

0.3

a R

0.0 0.5

1.0 1

Figure 14.4. Regions of instability computed for a cylindrical panel with a periodically varying axial load. Adapted from [25].

14.4.1 A Curved Panel We now consider a shallow panel that has some unidirectional curvature in a direction perpendicular to the loading. The analysis of such a system becomes increasingly complicated and recourse to numerical (FE) techniques is often used. ¨ Figure 14.4(a) shows an example taken from Kratzig and Nawrotzki [25] in which a shallow cylindrical panel segment is subject to a harmonically oscillating axial load. The geometry of the shell is defined by a = 10 m, R = 83.33 m, thickness = 0.1 m, and material properties for mild steel were used in the authors’ time integration. They produced both stable and unstable time series, depending on the parameter values used, specifically the constant (static) load (λS ), the forcing amplitude (λD), and the forcing frequency (), with the results normalized with respect to the elastic critical buckling load (λcr ) and linear natural frequency (ω1 ), as shown in Fig. 14.4(b). The ¨ dashed curve within each instability zone indicates the effect of damping. Kratzig and Nawrotzki [25] also use a similar FE technique to assess the parametric instability of a truncated conical shell and compute the magnitude of the Floquet multipliers.

14.4.2 A Cylindrical Shell When the shell is a complete cylinder, rather than a segment, we may still get parametric instability. This subsection describes the behavior reported in Popov et al. [26], based on an approximate analytical treatment. Given a cylindrical shell of the type shown in Fig. 10.14, we suppose that instead of a fixed axial load of magnitude Nx we now have a pulsating axial load of the form p(t) = p 1 cos ωt. We then make a single-mode Galerkin analysis (based on the Donnell shell theory) for the shell with the following properties: R/h = 100, L/R = 2, and a little damping added. With these parameters, it is appropriate to take an assumed form for the solution of w(x, y, t) = f 1 (t)h cos

πx 2πx 5y cos + f 2 (t)h cos . L R L

(14.41)

18:9

P1: KAE Chapter-14

CUFX159-Virgin

0 521 85648 5

August 14, 2007

290 (a)

Harmonic Loading: Parametric Excitation (b)

L

Figure 14.5. (a) Axisymmetric, or concertina mode of vibration, (b) asymmetric, or checkerboard, mode of vibration. Reproduced with permission from [27].

The two modes are shown in Fig. 14.5. Popov et al. [28] show that the dynamic response in mode 1 (for example) is related to Mathieu’s equation in the form d2 f 1 ω + 1 − 2µ cos τ f 1 = 0. (14.42) dτ2 ω1 Furthermore, the control parameter µ = p 1 /(2p c ) is introduced, in which p c is the linear-elastic (static) buckling load. Popov et al. [28] use a continuation technique to track the loss of stability of the trivial solution, that is, the transition from purely extensional to bending behavior. In the first part of this chapter we considered the solutions of Mathieu’s equation in terms of whether the motion grew with time (or not). In a comprehensive analysis, it is possible to more fully characterize the instability phenomena, as shown in Fig. 14.6(a). Here the regions of principal parametric (ω/ω1 ≈ 2) and fundamental resonance (ω/ω1 ≈ 1) can again be observed, with the transition curves labeled Sp1 and so on. These bifurcations indicate the nature of the instability, with the superscript 1 corresponding to a flip bifurcation and the superscript 2 indicating a pitchfork bifurcation. The former leads to the buildup of motion at twice the period of the external excitation, with the latter leading to motion at the same period as the forcing. Furthermore, we see the appearance of a couple of additional transition curves (B) that correspond to saddle-node bifurcations. All these transitions occur when a Floquet multiplier is equal to one in magnitude. For example, the flip bifurcation is characterized by a Floquet multiplier = −1 (see Fig. 14.1), which will be identified with the onset of period doubling in the final chapter. This diagram was based on a single-mode solution [i.e., with f 2 (t) = 0]. Also shown in Fig. 14.6 are the bifurcation diagrams at a number of frequency ratios [and indicated by the vertical dashed lines in part (a)]. Starting at ω/ω1 = 0.9 we see that as µ is increased the trivial solution loses stability (the dashed lines in these plots indicate unstable paths) and oscillations occur, which gradually grow as the unstable region is further penetrated. The subscript p indicates that this pitchfork bifurcation is supercritical (see Subsection 3.4.2). For ω/ω1 = 1.1 the bifurcation is now subcritical and leads to a sudden jump in response. An interesting example is found when ω/ω1 = 1.27 in which the system initially loses stability by means of a supercritical flip but then restabilizes and a subcritical pitchfork is encountered. Other modes can be included in the analysis [e.g., the f 2 (t) term in Eq. (14.41)] and although they have no effect on the initial loss of stability of the trivial state, they can

18:9

P1: KAE Chapter-14

CUFX159-Virgin

0 521 85648 5

August 14, 2007

14.4 Pulsating Axial Loads on Shells

291

0.9 1.1 1.27

1.9 2.1

1.5

1.0

2

2

Sp

Sb

1

Sb 1

0.5

B

Sp

2

B 0

0.5

1.0

2.0

1.5

1

2.5

3.0 1

2.5

f1

0.5

2.5

0.9

T

1.1

f1

B

2

2

S

Sp1

2 b

0.0

0.0

1.5

2.5

2.5

f1

S2 b

−0.5

−2.5 1.5

0.0

T 2T

Sp −2.5

1.27

f1

T

1.9

2.1

f1

2T

2.0

2T

B1 1

1

Sb

Sp −2.5

−2.5 0.0

1.5

0.0

1.5

Figure 14.6. Regions of instability computed for a circular cylinder with a periodically varying axial load. Adapted from Popov et al. [28].

play an important role in subsequent postcritical behavior (for example the B curves tend to emanate from the bottom of the transition curves toward lower frequency ratios), and significant modal interactions can occur [28, 29]. Popov et al. [28] also consider a shallow panel by using basically the same technique, with the major difference indicating loss of stability by means of a transcritical bifurcation. Also, the proportion of static load to the magnitude of the pulsating part is an interesting parameter that relates back to the underlying buckling behavior. The behavior of these types of system may be especially complicated when parametric and direct external excitation take place simultaneously [30, 31], and combination resonances can also occur [32, 33]. At the other end of the loading regime, we have creep buckling. A comprehensive account of this phenomenon can be found in Bazant and Cedolin [18]. This chapter finishes with a mention of Meissner’s problem. This is similar to the Mathieu equation but in this case the periodic

18:9

P1: KAE Chapter-14

CUFX159-Virgin

0 521 85648 5

292

August 14, 2007

Harmonic Loading: Parametric Excitation

variation in the stiffness is piecewise constant (a square wave) and is amenable to analytic treatment [34, 35]. References [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20] [21]

[22]

[23]

[24]

J. Singer, J. Arbocz, and T. Weller. Buckling Experiments, Vol. 2. Wiley, 2002. A.H. Nayfeh and D.T. Mook. Nonlinear Oscillations. Wiley, 1979. S.K. Sinha. Dynamic stability of a timoshenko beam subjected to an oscillating axial force. Journal of Sound and Vibration, 131:509–14, 1989. R.M. Evan-Iwanowski. On the parametric response of structures. Applied Mechanics Reviews, 18:699–702, 1965. D. Krajcinovic and G. Herrmann. Parametric resonance of straight bars subjected to repeated impulsive compression. AIAA Journal, 6:2025–7, 1968. A.D.S. Barr and G.T.S. Done. Parametric oscillations in aircraft structures. The Aeronautical Journal, 75:654–8, 1971. C.E. Hammond. An application of Floquet theory to prediction of mechanical instability. Journal of the American Helicopter Society, 4:14–23, 1974. R.A. Ibrahim and A.D.S. Barr. Parametric resonance, part I: Mechanics of linear problems. Shock and Vibration Digest, 10(1):15–29, 1978. R.A. Ibrahim and A.D.S. Barr. Parametric resonance, part II: Mechanics of nonlinear problems. Shock and Vibration Digest, 10(2):9–24, 1978. G.J. Simitses. Dynamic Stability of Suddenly Loaded Structures. Springer-Verlag, 1989. J.P. Cusumano. Low-Dimensional, Chaotic, Nonplanar Motions of the Elastica: Experiment and Theory. Ph.D. dissertation, Cornell University, 1990. C. Hayashi. Nonlinear Oscillations in Physical Systems. Princeton University Press, 1964. V.V. Bolotin. The Dynamic Stability of Elastic Systems. Holden-Day, 1964. M.A. Souza. Vibration of thin-walled structures with asymmetric post-buckling characteristics. Thin-Walled Structures, 14:45–57, 1992. D.W. Jordan and P. Smith. Nonlinear Ordinary Differential Equations. Oxford University Press, 1999. J. Guckenheimer and P.J. Holmes. Nonlinear Oscillations, Dynamical Systems, and Bifurcations of Vector Fields. Springer-Verlag, 1983. N.W. McLachlan. Theory and Applications of Mathieu Functions. Dover, 1964. Z.P. Bazant and L. Cedolin. Stability of Structures. Oxford University Press, 1991. J.F. Doyle. Nonlinear Analysis of Thin-Walled Structures. Springer, 2001. E. Ott. Chaos in Dynamical Systems. Cambridge University Press, 1993. P.R. Everall and G.W. Hunt. Arnold tongue predictions of secondary buckling in thin elastic plates. Journal of the Mechanics and Physics of Solids, 47:2187–2206, 1999. P.R. Everall and G.W. Hunt. Quasi-periodic buckling of an elastic structure under the influence of changing boundary conditions. Proceedings of the Royal Society of London A, 455:3041–51, 1999. K.K.V. Devarakonda and C.W. Bert. Flexural vibration of rectangular plates subjected to sinusoidal distributed compressive loading on two opposite sides. Journal of Sound and Vibration, 283:749–63, 2005. ¨ Y. Basar, C. Eller, and W.B. Kratzig. Finite element procedures for parametric phenomena of arbitrary elastic shell structures. Computational Mechanics, 2:89–98, 1987.

18:9

P1: KAE Chapter-14

CUFX159-Virgin

0 521 85648 5

August 14, 2007

References ¨ [25] W.B. Kratzig and P. Nawrotzki. Computational concepts for kinetic instability prob¨ lems. In A.N. Kounadis and W.B. Kratzig, editors, Nonlinear Stability of Structures (Theory and Computational Techniques). Springer-Verlag, 1995. [26] A.A. Popov, J.M.T. Thompson, and J.G.A. Croll. Bifurcation analyses in the parametrically excited vibrations of cylindrical panels. Nonlinear Dynamics, 17:205–25, 1998. [27] A.A. Popov. Parametric resonance in cylindrical shells: A case study in the nonlinear vibration of structural shells. Engineering Structures, 25:789–99, 2003. [28] A.A. Popov, J.M.T. Thompson, and F.A. McRobie. Low dimensional models of shell vibrations: Parametrically excited vibrations of cylindrical shells. Journal of Sound and Vibration, 209:163–86, 1998. [29] F.A. McRobie, A.A. Popov, and J.M.T. Thompson. Auto-parametric resonance in cylindrical shells using geometric averaging. Journal of Sound and Vibration, 227:65–84, 1999. [30] C.S. Hsu. Impulsive parametric excitation: Theory. Journal of Applied Mechanics, 39:551–8, 1972. [31] N. HaQuang, D.T. Mook, and R.H. Plaut. A non-linear analysis of the interactions between parametric and external excitations. Journal of Sound and Vibration, 118:425–39, 1987. [32] T. Iwatsubo, Y. Sugiyama, and S. Ogino. Simple and combination resonances of columns under periodic axial loads. Journal of Sound and Vibration, 33:211–21, 1974. [33] R.H. Plaut, N. HaQuang, and D.T. Mook. Simultaneous resonances in non-linear structural vibrations under two-frequency excitation. Journal of Sound and Vibration, 106:361–76, 1986. [34] A.P. Seyranian and A.A. Mailybaev. Multiparameter Stability Theory with Mechanical Applications. World Scientific, 2003. [35] C.-H. Xei. Dynamic Stability of Structures. Cambridge University Press, 2006.

293

18:9

P1: KAE Chapter-15

CUFX159-Virgin

0 521 85648 5

August 14, 2007

15 Harmonic Loading: Transverse Excitation

15.1 Introduction: Resonance Effects We have already seen many examples of how the presence of axial load tends to reduce the lateral stiffness and hence natural frequencies. In this chapter, we shall consider the effect of axial loads on the steady-state response of forced structural systems. This section will focus on an important class of forcing functions, that is, harmonic excitation. Thus, in Fig. 3.1 we might have F (t) = F0 sin ωt, or z(t) = z0 sin ωt, say. In the former case, we have a governing equation of motion of the form Mx¨ + C˙x + K(1 − p)x = F0 sin ωt,

(15.1)

in which M, K, and C represent physical properties associated with a slender structural system. We again assume that the spring stiffness is reduced by the presence of a parameter p, later to be identified with axial load [1]. The solution of Eq. (15.1) consists of the summation of two parts. First, the homogeneous solution is obtained from the free vibration and was derived previously. For typical damping values, it consists of an exponentially decaying oscillation (assuming p < 1). Second, the particular solution consists of a steady-state oscillation, X0 eiωt , where the magnitude of the steady-state response (relative to the forcing magnitude) is given by X0 1 = R(ω) = , F0 [K(1 − p) − ω2 M]2 + (Cω)2

(15.2)

and is often referred to as the receptance, amplitude-response, or frequencyresponse function (FRF) [2]. We observe the important resonant effect if the driv√ ing frequency ω is close to the natural frequency ωn = K/M of the system, or = ω/ωn = 1. Figure 15.1 shows the receptance for the parameter values K = M = 1 and C = 0.2 as a function of the destabilizing parameter p. We note that increasing p tends to shift the resonant peaks toward lower frequencies [with negative p (tensile) having the opposite effect]. The receptance could also have been nondimensionalized with respect to the effective stiffness in which case the curves would have emanated from a common point on the y axis. 294

12:33

P1: KAE Chapter-15

CUFX159-Virgin

0 521 85648 5

August 14, 2007

15.1 Introduction: Resonance Effects

295

20

Receptance R (ω)

p = 1.0 p = 0.9

15

p = 0.6 10

p = 0.3 p = 0.0 5

0

0

0.5

1

0.5

2

Figure 15.1. Effect of diminished stiffness on the receptance of a spring–mass–damper.

15.1.1 A Single-Mode Approximation We have seen a number of times how it is possible to make a relatively accurate single-mode analysis of beam dynamics. Consider a thin elastic beam of length L, flexural rigidity EI, and mass m (per unit length), which is clamped at both ends and subject to a compressive axial load of magnitude P. A single-mode energy analysis of this system, assuming a mode shape of the form Q(t) 2πx w(x, t) = 1 − cos , (15.3) 2 L can be conducted along the lines of Chapter 7, resulting in a natural frequency of 2 2π 2π 4 1 2 −P ω = . (15.4) EI 3m L L From this we immediately see that buckling occurs when Pcr = EI(2π/L)2 (exact), and in the absence of the axial load, we obtain a natural frequency of ω0 = 22.79 EI/(mL4 ) (exact coefficient = 22.37). Using these to nondimensionalize ( p¯ = P/Pcr and ω¯ = ω/ω0 ) we have ¯ ω¯ 2 = 1 − p.

(15.5)

Now, subjecting the system to a transverse harmonic point force at mid-span, F (t), and assuming a small amount of linear-viscous damping, C, we have the equation of motion given by Eq. (15.1) in which x = Q,

M = 3m,

K = EI(2π/L)4 ,

p = P/Pcr ,

and the corresponding receptance is still given by Eq. (15.2).

(15.6)

12:33

P1: KAE Chapter-15

CUFX159-Virgin

0 521 85648 5

August 14, 2007

296

Harmonic Loading: Transverse Excitation

1

p 0.8

0.6

0.4

0.2

0 0

0.5

1

1.5

ω

2

Figure 15.2. Effect of axial load on the receptance of a damped, SDOF model of a beam as a contour plot. Darker shades are higher in magnitude. C = 0.1.

The response can also be plotted as a contour plot in terms of axial load and forcing frequency. This is shown in Fig. 15.2. In this plot, we can observe how the resonant peaks spread as the axial load is increased. This can be viewed as an increase in the damping ratio (as this is relative to the natural frequency and hence stiffness). A closely related circumstance is what happens when the support upon which the mass is supported is excited [e.g., z(t) = z0 sin ωt]. This is called transmissibility, and will be studied in detail a little later in relation to vibration isolation. We conclude that an axial load not only has the effect of shifting resonant peaks to lower frequencies but also increases the effective damping in the system. We can again obtain a useful physical sense of the effect of the changing axial load on the forced vibration problem by evolving the axial load as a linear function of time: p = 0.002t. In this way, the stiffness of the system will reduce to zero when t = 500. Figure 15.3 shows an example based on numerical simulation of the governing equation of motion, including the diminishing stiffness. The forcing parameters are fixed at F0 = 1 and ω = 0.5, and hence resonance should occur when t ≈ 375. Note that there is again a small amount of overshoot in the nonstationary (slowly evolving, or swept) response.

15.1.2 Beyond Buckling We can extend the single-mode energy analysis of this system by including higherorder terms in the potential energy of this system (truncated after the second term): V=

1 EI 2

L 0

1 [w2 + w2 w2 ] dx − P 2

L

0

1 w2 + w4 dx, 4

(15.7)

12:33

P1: KAE Chapter-15

CUFX159-Virgin

0 521 85648 5

August 14, 2007

15.2 The Poincare´ Section

297

10 x(t)

7.5 5.0 2.5 0.0

−2.5 −5.0 −7.5 −10 0

100

200

300

400

500

600 t

Figure 15.3. A sweep through decaying stiffness and passing through resonance.

and the kinetic energy remains the same. Evaluating the potential-energy expression now gives 4 6 1 2π 2π 1 2 V= EIL EIL Q + Q4 16 L 256 L 2 2π 1 1 3L 2π 4 4 P − PL Q2 − Q . 16 L 128 8 L

(15.8)

In addition to the trivial (Q = 0) solution, we now have a nontrivial (post-buckled) path given by 8(p¯ − 1) Q2 = 2

, 2π 3 1 − p ¯ L 4

(15.9)

where p¯ =

EI

P 2π 2 .

(15.10)

L

After buckling, the nondimensional natural frequency is given by 9 ( p ¯ − 1) 3( p ¯ − 1) ω¯ 2 = 1 +

− p¯ 1 + 4

= ω¯ 2 = 2(p¯ − 1), 1 − 34 p¯ 1 − 34 p¯

(15.11)

that is, half the prebuckling slope of the load–frequency (-squared) relation [Eq. (15.5)]. This is a result anticipated from the normal form of the supercritical pitchfork bifurcation (Section 7.4).

15.2 The Poincare´ Section Before moving on to consider the resonance response of axially loaded continuous systems we introduce the concept of Poincare´ sampling. The response of the forced nonlinear oscillator of the type of Eq. (15.1) with p = 0 is typically given in terms of transient (see Chapter 3) and steady-state parts. In the framework of

12:33

P1: KAE Chapter-15

CUFX159-Virgin

0 521 85648 5

August 14, 2007

298

Harmonic Loading: Transverse Excitation

dynamical system theory, we can view the steady state as a periodic attractor for the surrounding transients. For a linear system, the periodic attractor is unique. We shall see in the next chapter that this is not necessarily the case for nonlinear oscillators. A typical engineering approach is then to plot the maximum amplitude of response as a function of the forcing frequency (see Fig. 15.1). Often the phase difference between the forcing function and the response is also plotted and a sudden shift in phase is associated with resonance [3, 4]. However, an alternative description of the response is to reduce the 3D phase space in continuous time to a 2D phase space in discrete time by Poincare´ sampling [5, 6]. The complete solution to Eq. (15.1) with p = 0 can be written as x(t) =

F0 sin(ωt − φ) + X1 e−ζωn t sin( 1 − ζ 2 ωn t + φ1 ), K 2 [1 − (ω/ωn )2 ] + [2ζω/ωn ]2 (15.12)

and focusing on the steady-state solution we ignore the second term and write Eq. (15.12) in the alternative form x(t) = a cos(ωt) + b sin(ωt).

(15.13)

Differentiating this to get the velocity, we have y(t) ≡ x˙ = −aω sin(ωt) + bω cos(ωt),

(15.14)

and setting t = 0 (which effectively fixes the initial forcing phase), we simply get x = a and y = bω as the fixed point location, where a= b=

(1 − 2 )f , (1 − 2 )2 + (2ζ) 2

(15.15)

2ζ f , + (2ζ, )2

(15.16)

(1 −

2 )2

and = ω/ωn . The Poincare´ section can thus be considered as an alternative to the more conventional (amplitude–phase) representation of the response of an oscillator. The complementary function (the transient solution) can also be included in the following way to give a discrete mapping: the Poincare´ map P [6], ⎡ ⎤ 1 n S S C + ζω −2πζωn x x ωd ωd ⎣ ⎦ →e ω 2 ω ζω y y − ωnd S C − ωdn S ⎤ ⎡ −aC + − ζωωnda − bω S −2πζωn a ωd ⎦+ 2 +e ω ⎣ , (15.17) aωn nω bω S −bωC + ωd + ζbω ωd √ where C ≡ cos(2πωd /ω), S ≡ sin(2πωd /ω), and ωd = ωn 1 − ζ 2 . Thus, given some initial conditions, this set of difference equations will map out the transient at

12:33

P1: KAE Chapter-15

CUFX159-Virgin

0 521 85648 5

August 14, 2007

15.3 Continuous Systems

299

intervals of the forcing period until converging on the fixed point (x, y) = (a, bω).

(15.18)

This mapping is exact for a linear oscillator (and is related to the Z transform [7]) but cannot usually be easily obtained for nonlinear systems. However, importantly, this complete mapping contains the stability information regarding the fixed point and relates back to the section on Floquet theory and characteristic multipliers (see Section 14.2). We have already introduced the concept of characteristic eigenvalues (exponents) (CEs) for determining stability of equilibria in unforced systems. Now we see that it is the eigenvalues of the map (characteristic multipliers, or CMs) that determine the stability of cycles. The eigenvalues of the Jacobian, that is, the first partial derivatives of the map given by Eq. (15.17), DP(a, ωb), are given by λ1,2 = e−

2πζωn ω

±i

2πωd ω

,

(15.19)

which confirms that the fixed point is asymptotically stable, because the damping and natural frequency are positive numbers. This is why consideration of discrete maps plays a useful role in the study of flows. Despite the fact that this approach tends to hide the usually important engineering aspects of amplitude and phase, it does provide a very convenient means of assessing stability, and the evolution of the system responses under the slow change in a parameter. For the types of nonlinear system to be considered later, Poincare´ sampling provides a powerful tool in the numerical and experimental investigation of periodically excited nonlinear oscillators.

15.3 Continuous Systems We now turn to consideration of axially loaded, transversely forced, continuous beams. The forced string and membrane can also easily be handled by use of these techniques. A beam of length L, mass per unit length m, uniform flexural rigidity EI, viscous damping coefficient C, axial force P, and transverse load Q0 F (x) cos t, respond as w(x, t). The governing equation of motion is mwtt + Cwt + EIwxxxx + Pwxx = Q0 F (x) cos T,

(15.20)

in which subscripts on w reflect partial derivatives. This equation can be put in the nondimensional form ∂4 w¯ ∂2 w¯ ∂w¯ ∂2 w¯ ¯ t¯, + p + c + = f (x) cos (15.21) ∂x¯ 4 ∂t¯ ∂x¯ 2 ∂t¯2 with x¯ = x/L, w¯ = w/L, t¯ = t EI/(mL4 ), √ ¯ = mL4 /(EI), c = CL2 / mEI, (15.22) p = PL2 /(EI), f (x) = Q0 F (X)L3 /(EI),

ζ = c/(21 ),

12:33

P1: KAE Chapter-15

CUFX159-Virgin

0 521 85648 5

August 14, 2007

300

Harmonic Loading: Transverse Excitation L Q cos T

X

EI, m

P W(X,T)

Figure 15.4. A cantilever beam subject to a constant axial load and a harmonically varying uniformly distributed lateral force.

where 1 is the fundamental natural frequency and its value depends on the boundary conditions. For convenience, we now drop the overbar notation, and assuming a steady-state harmonic response of the form w(x, t) = Re y(x)eit

(15.23)

y (x) + py (x) + (ic − 2 )y(x) = f (x).

(15.24)

leads to

Equation (15.24) can then be solved for a specific set of boundary conditions and transverse forcing types. A number of examples with distributed, transverse, harmonic forcing can be found in Virgin and Plaut [1]. Here, two cases will be considered: r an axially loaded cantilever beam with a uniformly distributed harmonic load, and r an axially loaded, clamped–clamped beam with a harmonic central point load. Considering the first case as shown in Fig. 15.4 we have a governing equation of motion given by Eq. (15.24) but now with unity on the right-hand side. The general solution is given by y(x) = (ic − 2 )−1

2 (aj cosh λj x + bj sinh λj x), j =1

where λj = (νj /γj ) + i(γj /2),

1/2 γj = 2 − j + 2j + ν2j ,

ν1 = φ/2,

ν2 = −φ/2,

1 = ( − p)/2,

2 = −( + p)/2, 1/2 φ = − δ + δ2 + (c)2 /2 ,

=−

c , 2φ

δ = p 2 + 42 .

(15.25)

12:33

P1: KAE Chapter-15

CUFX159-Virgin

0 521 85648 5

August 14, 2007

15.3 Continuous Systems

301

0.4

A 0.3

p = π2 / 5 p=0

0.2

p = - π2 / 10 0.1

0 0

5

10

15

20

Ω

25

Figure 15.5. The central amplitude versus the forcing frequency for a cantilever beam with a uniformly distributed force.

The boundary conditions in this case are y(0) = 0, y (0) = 0, y (1) = 0, and y (1) + py (1) = 0. The elastic critical load is p cr = π2 /4, and the fundamental natural frequency is 1 = 3.516. Applying the boundary conditions and solving the resulting simultaneous equations (in aj , bj ) leads to the results shown in Fig. 15.5 in which a damping ratio of ζ = 0.142 was used. The amplitude A is the magnitude of the maximum central deflection, which depends on the static axial load and the forcing frequency, i.e., A(, p) = |y(0.5)|. Note the presence of the second resonant peak in the vicinity of = 22. It is interesting to see how the amplitude and the corresponding resonant frequency vary with axial load. Figure 15.6 shows the relation AR (p) = max A(, p), ≥0

(15.26)

and the values of (squared) for which this condition occurs. Three representative damping values are used. We again get a near-linear relation between frequency squared and the axial load. Now consider the second case. Figure 15.7(a) shows the amplitude response (receptance) when ζ = 0.02. We again see the anticipated increase in natural frequency for a tensile axial force, and reduction for compression. Figure 15.7(b) shows how the receptance associated with the third (second symmetric) mode is affected by the presence of an axial load. In this case, the undamped third frequency occurs at = 120.9 for the unloaded case. The peaks are also shifted slightly from those of the undamped case because of the presence of a little damping.

12:33

P1: KAE Chapter-15

CUFX159-Virgin

0 521 85648 5

August 14, 2007

302

Harmonic Loading: Transverse Excitation

2.5

2.5

(a)

(b)

p 2

2

p

ζ = 0.2 1.5

1.5

ζ = 0.05 1

1

ζ = 0.1

0.5

ζ = 0.1

0.5

ζ = 0.05 0

−0.5 0

ζ = 0.2

0

0.2

0.4

0.6

0.8

AR

1

−0.5

0

5

10

2

ΩR

15

Figure 15.6. (a) The resonant amplitude, (b) the frequency squared versus the axial load for a cantilever beam with a uniformly distributed force.

In both of the preceding examples, it can be shown that the resonant amplitude and frequency are given approximately by AR (p) ≈ AR (0)/ 1 − (p/p cr ), 2R

≈

21 [1

− (p/p cr )],

(15.27) (15.28)

and we can also relate this back to the material presented in Chapter 7, for example, the p = 0 resonant peak for the beam in Fig. 15.7 occurs at the natural frequency coefficient of 22.4. Experimental Verification. A thin steel strip was clamped between blocks at its ends

and placed in a displacement-controlled testing machine. In this configuration, the end shortening is prescribed, and the resulting axial force is measured with a load cell. With the standard expressions based on the earlier analysis, the critical load was computed at 1235 N and the lowest natural frequency (with no axial load) at 44 Hz. The strut was struck by an impact hammer and a laser velocity vibrometer was used to measure the response [8]. The data were then acquired and analyzed with the Bruel and Kjaer pulse system. Velocity time series were then subject to a Hann window and a fast Fourier transform algorithm to extract the frequency

12:33

P1: KAE Chapter-15

CUFX159-Virgin

0 521 85648 5

August 14, 2007

15.3 Continuous Systems

303

0.3

0.02

y(0.5) f0 0.25

p=4

(a)

2

p=3

2

p=2 0.2

2

p = 0

2

p=0

0.15

p = 2

0.015

2

p=

(b)

y(0.5) f0

p=-

0.01 2

0.1 0.005 0.05

0

0

5

10

15

20

25

30

0 110

115

120

125

130

Figure 15.7. (a) Amplitude response of an axially loaded, clamped–clamped beam subject to a central point harmonic point force and (b) in the vicinity of the third mode.

content. This process was repeated 10 times at each axial-load level. The results were averaged and displayed as normalized mobility, that is, normalized with respect to the force of the impact hammer. A photograph of the experimental system is shown in Fig. 15.8(a). Figure 15.8(b) shows a summary of how the peaks shift to lower (higher) frequency as the compressive (tensile) axial load is increased. However, experimental results from this type of system need careful consideration because testing machines are sometimes referred to as semirigid loading devices, membrane effects in the beam may occur, and even the boundary conditions can be a function of loading [9, 10].

(a)

(b) 1

p 0.5

0 −0.5 −1 −1.5 −2 −2.5 0

0.5

1

1.5

2

2.5

3

3.5

( ω/ω ) 0

Figure 15.8. (a) A photograph of the clamped beam in the testing machine and (b) the axial load plotted as a function of the natural frequencies (squared).

2

12:33

P1: KAE Chapter-15

CUFX159-Virgin

0 521 85648 5

August 14, 2007

304

Harmonic Loading: Transverse Excitation m X

c

k Z

Figure 15.9. Schematic of a mass isolated from the motion of a foundation.

15.4 An Application to Vibration Isolation In many instances, buckling is viewed as an undesirable occurrence, in particular when it precipitates a total loss of stiffness and collapse (see, for example, Section 3.4). However, there are ways in which postbuckled stiffness can be exploited, and this section introduces an example. The concept of vibration isolation, that is, how to reduce the force or motion transmitted to a device from a source of vibration, is well established [11–14]. This section describes an approach to effective vibration isolation by use of the subtle interplay of axial loads, dynamics, and stability [15]. Consider a simple mechanical system consisting of a mass (in a gravitational field) supported by a spring and damper, which are themselves supported on a base, as shown in Fig. 15.9. If the motion of the base is harmonic, for example, Z(t) = Z0 sin ωt, then it can be shown that the steady-state displacement transmissibility, X/Z (where X is the response amplitude of the mass), is given by the expression 1/2 1 + (2ζ)2 X = . Z (1 − 2 )2 + (2ζ)2

(15.29)

This is plotted in Fig. 15.10 as a function of the frequency ratio = ω/ωn , where √ ωn = k/m. Damping tends to severely reduce the resonant peak. This was also true for the cases examined earlier in this chapter in which the force was applied to the mass directly. In these earlier cases, damping tended to reduce the magnitude of the response at all frequencies. Here, we notice an interesting feature in which (linear-viscous) damping results in slightly larger responses at higher √ frequencies. We also observe that damping has no influence when is exactly 1/ 2. In addition to direct mass and base excitation, a third type of resonance can also be found in a system with a rotating unbalance [4]. However, overall, we see that√the transmissibility is small for relatively highfrequency ratios, that is, for > 2, X/Z < 1.0. Given a forcing frequency ω, a typical design option would be to mount the device on a soft spring to induce a low natural frequency, ωn . But if the spring has a low stiffness, there is a danger that

12:33

P1: KAE Chapter-15

CUFX159-Virgin

0 521 85648 5

August 14, 2007

15.4 An Application to Vibration Isolation

305 8 X/Z

ζ = 0.0

6

ζ = 0.1

Figure 15.10. Displacement transmissibility for a SDOF oscillator for four typical damping values.

4

ζ = 0.2 ζ = 0.3

2

0 0.5

1

1.5

2

it will deflect statically too much (because mg = kx), and this often places practical limits on the spring stiffness.

15.4.1 Postbuckling of a Strut Revisited In earlier chapters, we have seen how axially loaded structures typically possess nonlinear characteristics, especially close to, or beyond, initial buckling. Often this takes the form of additional postbuckled stiffness (e.g., in plates). In their postcritical state, they exhibit relatively low stiffness in the axial direction and yet they carry axial loads above their buckling load. As indicated schematically in Fig. 15.9 we see a potential opportunity in the context of vibration isolation. Let’s return to the simply supported strut as shown in Fig. 7.1. Any structure exhibiting stable postbuckled behavior can be used in this situation, but the pinned beam is easiest to analyze. Thus, we are interested in a structural system of the supercritical type. We have already seen [see Eq. (7.51)] that the initial postbuckled equilibrium configuration is described by P π2 Q 2 =1+ , (15.30) Pe 8 L where Pe = EI(π/L)2 is the classical Euler critical load. In Chapter 7 we were primarily interested in lateral stiffness effects, but now we need to consider stiffness in the axial direction as this is the direction in which the force acts. The geometric relation between the lateral deflection Q and the end shortening δ can be established as [15] δ π2 Q 2 3π4 Q 4 = + . (15.31) L 4 L 64 L The end shortening is approximately related to the square of the lateral deflection, and eliminating Q in Eqs. (15.30) and (15.31) leads to P 1 δ 1 δ . (15.32) = 2+ 1+3 ≈1+ Pe 3 L 2 L

Ω

2.5

12:33

P1: KAE Chapter-15

CUFX159-Virgin

0 521 85648 5

August 14, 2007

306

Harmonic Loading: Transverse Excitation 1.2

1.2

(a)

P/P e

(b)

P/Pe

1.1

_ p

1.1

_ δ

1

1

0.9

0.9

0.8

0.8

0.7

0.7

0.6

0

0.05

0.1

0.15

0.2 0.25 Q/L

0.6

0

0.05

0.1

δ /L

0.15

Figure 15.11. Deflection of the strut as a function of axial load: (a) central lateral deflection, (b) end shortening.

This is a result based on moderate lateral deflections, i.e., up to about 20% of the length. Recall that this is a nonlinearity induced by the curvature expression [Eq. (7.40)], rather than the membrane effect discussed in Section 7.5 for example, that is, the ends are free to move toward each other. Equations (15.30) and (15.32) are shown by the solid curves in Figs. 15.11(a) and 15.11(b), respectively. The postbuckled stiffness is only mildly affected by initial imperfections (unlike in the vicinity of the critical point), and when a single mode is adopted, representative initial geometric imperfection leads to the dashed curves in Fig. 15.11. If we load the strut axially to slightly above its elastic critical load, for example, P/Pe = 1.05, then (for Q0 = 0) we have Q/L ≈ 0.2, δ/L ≈ 0.1. This specific load-deflection condition furnishes an equilibrium position from which incremental coordinates are measured: p¯ = P/Pe − 1.05, δ¯ = δ/L − 0.1, and indicated in Fig. 15.11(b). This strut, then, is able to support a relatively high axial load (sufficient to cause buckling) but exhibits the desirable soft spring characteristic.

15.4.2 Experimental Verification An experimental verification is considered in this section and configured as shown in Fig. 15.12(a). Part (b) shows an alternative configuration in which four postbuckled panels provide the support. The vertical shaker was connected to a cam-shaft attached to a variable-speed motor. The forcing amplitude was fixed at 3 mm, and the shaker then imparted an almost sinusoidal motion through the isolation system (consisting of two steel struts made of spring steel) to the mass, which moved

12:33

P1: KAE Chapter-15

CUFX159-Virgin

0 521 85648 5

August 14, 2007

15.4 An Application to Vibration Isolation

307

Figure 15.12. (a) Photograph of the experimental setup and (b) alternative configuration in which panels re used.

(a)

(b)

vertically, guided by low-friction linear bearings. The frequency of excitation was then varied over an appropriate range, and the vertical motion of the mass was measured [15]. Plots of axial load versus lateral deflection and end shortening are shown in Fig. 15.13, where the deflections were measured with a linear-variable displacement transformer. The two struts were 268 mm long, 19 mm wide, 0.66 mm thick (and thus I = 4.55 × 10−13 m4 ), and taking a typical value for Young’s modulus of 200 GPa, we anticipate an Euler load in the vicinity of 25 N ≡ 2.55 kg. A Southwell plot can be used to recast the data from Fig. 15.13(a) to estimate a critical load of approximately 23 N. Because of initial geometric imperfections, the “critical load” is again manifest as a relatively rapid increase in the deflection that is due to additional load. The axial load versus lateral deflection, and versus end shortening, results are shown in Figs. 15.13(a) and 15.13(b), respectively, and these relations illustrate a good correlation with the corresponding theoretical curves (including an initial imperfection) given in Fig. 15.11. By the choice of an appropriate point on the curve in Fig. 15.13(b) as the fundamental equilibrium position, for example, P = 23.5N → (P/Pe ≈ 1.0), δ = 15.2 mm → (δ/L = 0.057), the transmissibility can be assessed over a range of excitation frequencies. Locally, the stiffness is approximately 195 N/m (i.e., the slope of P/Pe versus δ/L about the chosen operating point), and because the mass is 2.4 kg, we would thus expect a natural frequency of free vibration close to 9 rad/s ≡ 1.43 Hz. A free decay of this system gives a natural period of approximately 0.68 s (and hence ωn =√1.47 Hz), and thus, we anticipate effective isolation for forcing frequencies ω > 2ωn ≈ 2.2 Hz. If a conventional linear (helical) spring had been used instead of the buckled struts, a static deflection of approximately four times the deflection of the struts would have resulted from this load level.

15.4.3 The Forced Response The transmissibility √ratio [given by Eq. (15.29)] should be low, more specifically less than one, for > 2. Subjecting the system to a range of excitation frequencies (at

12:33

P1: KAE Chapter-15

CUFX159-Virgin

0 521 85648 5

August 14, 2007

308

Harmonic Loading: Transverse Excitation 30

30

(a)

Force (N)

(b)

Force (N)

25

25

20

20

15

15

10

10

5

5

_

_

( δ , p)

0

0 0

10

20

30

40

50

0

5

10

15

Q (mm)

20 25 δ (mm)

Figure 15.13. Axial load versus (a) midpoint lateral deflection and (b) end shortening for the experimental system.

constant forcing amplitude) leads to the experimental results shown in Fig. 15.14. Three typical time series are shown for the frequency ratios indicated. The response when the forcing frequency is exactly twice the natural frequency (indicated by the open circle) shows an interesting subharmonic of order two. This is a consequence of combined parametric and external forcing terms in the governing dynamic equations (see Chapter 14). In general we see a highly attenuated response for higher frequencies; that is, in this frequency range, the mass is effectively isolated from the motion of the base. This concept can be extended in a number of ways [16], and the usual issues of avoiding stroke-out, nonlinear behavior, and fatigue still apply. Here some research is mentioned in which axially loaded structures are taken advantage of to amplify, rather than reduce, motion [17, 18], and how axial load can be used to tune resonant frequencies for the purposes of energy scavenging [19].

15.5 Forced Excitation of the Thermally Buckled Plate In Chapter 10 we saw how axial loading (thermal) effects influenced the free vibration of thin plates. At ambient temperatures, the plate would exhibit a periodic response if subject to a periodic lateral excitation, and increasing the thermal loading would result in the now-familiar shift in resonance characteristics. For temperatures above the critical buckling level, two coexisting equilibria appear, that is,

12:33

P1: KAE CUFX159-Virgin

0 521 85648 5

August 14, 2007

15.5 Forced Excitation of the Thermally Buckled Plate

309

1 X/Y 0.8

0.6

0.4

0.2

0 1

2

3

4

5

6

7

8

Figure 15.14. Experimental transmissibility for the displacement of the strut-supported mass.

supercritical behavior. For relatively low levels of external excitation, we still expect periodic response but now offset about some nontrivial equilibrium configuration. We shall use a panel system similar to the one described in Section 10.5 to illustrate this effect experimentally (the only difference being that now the panel is thinner). In the final chapter, we shall look at more intense excitation that can result in large-amplitude responses, including chaos and intermittent snapping between these equilibria [20]. Typical periodic responses are shown in Fig. 15.15 in which two alternative coexisting oscillations are superimposed, that is, motion about both static (postbuckled) equilibrium configurations within the corresponding potential energy wells. This is the situation shown schematically in Fig. 10.6. The measurand in this case is strain and is plotted against strain a quarter of a cycle later in part (b), by use of a standard embedding technique in nonlinear dynamics [21]. These results were recorded

PSD

(t +T/4)

(c)

10 2 0

10 -1

y

0

10 5

(b)

(a)

y

Chapter-15

10 -4

0

0.05 t (s)

0.1

10 -7

y

0

(t)

0

(Hz)

Figure 15.15. Small-amplitude periodic behavior about both postbuckled positions using microstrain and time-lag embedding: (a) time series, (b) phase projections, and (c) frequency spectrum.

500

12:33

P1: KAE Chapter-15

CUFX159-Virgin

0 521 85648 5

310

August 14, 2007

Harmonic Loading: Transverse Excitation f

at (ζ, η) = (0.583, 0.416) with the temperature set at T/T cr = 1.76, with an excitation level1 of 130 dB at 120 Hz (quite near the lowest natural frequency at this temperature). The second attractor (which has a slightly smaller basin of attraction) is attained by giving a variety of perturbations access to different areas of the initial condition space. Because of the inevitable initial geometric imperfections there is a mild asymmetry, such that the postbuckled plate has a slightly tilted underlying potential-energy function [see Fig.3.8(a)]. This asymmetry is also reflected in a slightly different location in the phase projection of the two responses as well as a slight difference in the period in their time series. This kind of small-amplitude periodic behavior possesses a power spectrum with a dominant spike at the forcing frequency [see Fig. 15.15(c)]. The sharpness of the resonant peak is a standard means (the half-power method) by which the damping can be estimated [22, 23]. A good correlation with a theoretical analysis is described in Murphy et al. [24].

References [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17]

1

L.N. Virgin and R.H. Plaut. Effect of axial load on forced vibrations of beams. Journal of Sound and Vibration, 168:395–405, 1993. D.J. Ewins. Modal Testing: Theory and Practice. Research Studies Press, 1984. W.T. Thomson. Theory of Vibration with Applications. Prentice Hall, 1981. D.J. Inman. Engineering Vibration. Prentice Hall, 2000. D.W. Jordan and P. Smith. Nonlinear Ordinary Differential Equations. Oxford University Press, 1999. J. Guckenheimer and P.J. Holmes. Nonlinear Oscillations, Dynamical Systems, and Bifurcations of Vector Fields. Springer-Verlag, 1983. K. Ogata. System Dynamics. Prentice Hall, 1998. G.C. Goodwin and R.L. Payne. Dynamics System Identification: Experiment Design and Data Analysis. Academic, 1977. A. Picard, D. Beaulieu, and B. Perusse. Rotational restraint of a simple column base connection. Canadian Journal of Civil Engineering, 14:49–57, 1987. R.H. Plaut. Column buckling when support stiffens under compression. Journal of Applied Mechanics, 56:484, 1989. F.C. Nelson. Vibration isolation: A review, I. Sinusoidal and random excitations. Shock and Vibration, 1:485–93, 1994. R.H. Racca. Characteristics of vibration isolators and isolation systems. In Shock and Vibration Handbook, 4th ed. McGraw-Hill, 1996, Chapter 32. J. Winterflood, T. Barber, and D.G. Blair. High performance vibration isolation using spring in Euler column buckling mode. Physics Letters A, 19:1639–45, 2002. E.I. Rivin. Passive Vibration Isolation. ASME, 2003. L.N. Virgin and R.B. Davis. Vibration isolation using buckled struts. Journal of Sound and Vibration, 260:965–73, 2003. R.H. Plaut, J.E. Sidbury, and L.N. Virgin. Analysis of buckled and pre-bent fixed-end columns used as vibration isolators. Journal of Sound and Vibration, 283:1216–28, 2005. J. Jiang and E. Mockensturm. A motion amplifier using an axially driven buckling beam: I. Design and experiments. Nonlinear Dynamics, 43:391–409, 2006. The reference value for the decibel scale is the rms value 20 µN/m2 for sound pressure level.

12:33

P1: KAE Chapter-15

CUFX159-Virgin

0 521 85648 5

August 14, 2007

References [18] J. Jiang and E. Mockensturm. A motion amplifier using an axially driven buckling beam: II. Modeling and analysis. Nonlinear Dynamics, 45:1–14, 2006. [19] E.S. Leland and P.K. Wright. Resonance tuning of piezoelectric vibration energy scavenging generators using compressive axial load. Smart Materials and Structures, 15:1413– 20, 2006. [20] K.D. Murphy, L.N. Virgin, and S.A. Rizzi. Experimental snap-through boundaries for acoustically excited, thermally buckled plates. Experimental Mechanics, 36:312–17, 1996. [21] L.N. Virgin. Introduction to Experimental Nonlinear Dynamics: A Case Study in Mechanical Vibration. Cambridge University Press, 2000. [22] D.E. Newland. An Introduction to Random Vibrations and Spectral Analysis. Longman, 1984. [23] J.S. Bendat and A.G. Piersol. Random Data: Analysis and Measurement Procedures. Wiley, 1986. [24] K.D. Murphy, L.N. Virgin, and S.A. Rizzi. Characterizing the dynamic response of a thermally loaded, acoustically excited plate. Journal of Sound and Vibration, 196:635– 58, 1996.

311

12:33

P1: RTT Chapter˙16

CUFX159-Virgin

0 521 85648 5

August 14, 2007

16 Nonlinear Vibration

PART I: FREE VIBRATION 16.1 Introduction This last chapter considers the dynamic response of axially loaded structural systems in which the motion is not necessarily confined to the local vicinity of an underlying equilibrium position and dynamic behavior is not necessarily harmonic. In a number of places throughout this book, a statement has been made to the effect that largeamplitude behavior will be described later. We now finally consider such situations, largely in terms of revisiting examples detailed in earlier examples, but now, not relying on certain restrictions, for example, linear, or small-amplitude, behavior. Both free and forced vibrations will be considered [1, 2]. By way of a simple introduction, we go back to the softening cable example described in Section 3.5, and specifically consider the context of Fig. 3.12. This is a free vibration started (with initial conditions) some distance from any of the three available stable equilibrium points present at this level of loading. Because there is no damping, the total energy is conserved, and thus phase trajectories can be viewed as contours of constant total energy. Figure 16.1(a) illustrates the energy levels as a contour plot, and thus we can view the phase trajectory of Fig. 3.12 living in the second darkest shade within the contours of Fig. 16.1(a). Parts (b)–(d) give specific examples of time series generated (numerically) by different initial conditions. We see that the time series in part (b), which corresponds to the phase trajectory shown in Fig. 3.12, is far from sinusoidal. The time series shown in part (c) has relatively small amplitude with a natural frequency close to that predicted by linear theory; see Fig. 3.2 (but still slightly nonlinear; note the expanded y-axis). The time series shown in part (d) is initiated from a position very slightly removed from one of the unstable equilibria. After remaining in the local vicinity of the unstable point (previously calculated to be at q = 1.92), the trajectory moves away, undergoing a long-period motion, which again is far from sinusoidal. Parts (b) and (d) illustrate how the frequency of the response is highly dependent on the amplitude of motion and hence on the initial conditions. Other types of motion are exhibited following different initial conditions and values of the parameter (which is assumed to be maintained at the same level during the motion). The next section will briefly describe a couple of approximate analytical techniques primarily designed to assess the effects of moderate nonlinearity on the 312

18:14

P1: RTT Chapter˙16

CUFX159-Virgin

0 521 85648 5

August 14, 2007

16.2 Abstract Models

313 (a)

. q

2

(b)

q4

t

2 −2 −4

1

10

20

30

40

50

(c)

q 0.4 t

0.2 0 −0.2 −0.4 −1

q

10

20

30

40

50

(d)

2 1

−2

−1 −4

−2

0

2

4

q

t 10

20

30

40

−2

Figure 16.1. (a) A contour plot of total energy versus the state variables. λ = −0.5. Time series started from q˙ = 0, and (b) q(0) = 4.8; (c) q(0) = 0.5; (d) q(0) = 1.91.

system response, for example, in going from the type of motion in part (c) to that in part (d).

16.2 Abstract Models At various points in this book, including the previous section, we have seen how the natural frequency of a system may sometimes depend on the amplitude of the motion. It is straightforward to integrate the nonlinear equation of motion numerically, but it also useful to be able to obtain an analytical relation between frequency and amplitude, for example. In free vibrations of undamped systems, it may be possible to obtain an exact relation based on elliptic integrals [3], or use can be made of the conservation of total energy to facilitate a solution [4]. A variety of approximate analytical techniques have also been developed, typically applicable to moderately large (oscillatory) behavior. Suppose we go back to the simplest link model, first considered in Section 5.2, and set the axial load level at p = 1.2 corresponding to the equation of motion θ¨ + ω2n (θ − 1.2 sin θ) = 0.

(16.1)

Because there is no damping, we expect trajectories to trace phase trajectories at constant values of the total energy. Previously we expanded the sine term as a Taylor series, retaining just the first term for a standard linearization. Suppose we keep the next term in the series expansion and shift the origin to the positive equilibrium position (i.e., at the bottom of the right-hand potential-energy well with p = 1.2, we have an equilibrium at θe = 1.02674). The potential energy corresponding to this situation is shown in Fig. 16.2(a) as the solid curve and corresponds to the cubic

50

18:14

P1: RTT Chapter˙16

CUFX159-Virgin

0 521 85648 5

August 14, 2007

314

Nonlinear Vibration

(a)

0.2

V 0.1

x −3

−2

−1

1 −0.1 0.2

R (b) 0.1

x −3

−2

−1

1 −0.1

Figure 16.2. (a) Potential-energy functions for the inverted pendulum and (b) corresponding restoring force functions (x = θ − 1.02674 for the specific non-truncated system).

restoring force (R = Vdx) in part (b). Superimposed is the potential energy associated with linearization (long-dashed curve) together with that for a quadratic restoring force (and thus cubic potential, the short-dashed curve). There are many ways in which these curves may be fit, but the quadratic restoring force reflects the nonlinear relation between force and displacement as well as the asymmetry. We thus see how the linearization corresponds to small-amplitude motion about the equilibrium position, a distorted egg-shaped phase trajectory corresponds to motion barely contained in one of the local potential-energy wells and can be captured by the quadratic fit, and a thoroughly nonlinear cross-well behavior needs at least a cubic restoring force for global containment, or boundedness. To consider the growth from linear to mildly nonlinear motion we can thus study an equation of the form x¨ + x + x2 = 0,

(16.2)

in which x is measured from the shifted origin and the restoring force has been changed for convenience such that the unstable equilibrium position occurs at negative one. We recognize this as one of the standard forms from Chapter 3, and this system was also subject to a sudden load in Section 13.4. This system possesses a stable equilibrium at the origin and a saddle point at x = −1. Elliptic integral solutions are available, as well as approximate solutions based on the techniques of harmonic balance and perturbation methods. Further details of these methods can

18:14

P1: RTT Chapter˙16

CUFX159-Virgin

0 521 85648 5

August 14, 2007

16.3 A Mass Between Stretched Springs

315

.

x

0.4 x(0) 0.3

x

.

Figure 16.3. Natural frequency versus amplitude of motion.

x(0) = 0.4 , x(0) = 0

0.2

0.1

.

x(0) = 0.2 , x(0) = 0

0.7

0.8

0.9

1.0

be found in [4–8]. Numerical integration of Eq. (16.2) shows a softening nonlinearity, that is, the frequency tends to diminish with amplitude. This relation is shown in Fig. 16.3. Also shown are a couple of phase trajectories for the specific cases generated by the initial conditions x(0) = 0.2, x(0) ˙ = 0.0, and x(0) = 0.4, and x(0) ˙ = 0.0. The egg-shaped, asymmetric behavior of the phase trajectory started farther away from equilibrium reflects the underlying potential energy as anticipated by the form of the dashed curve in Fig. 16.2. Of course, an initial condition started slightly below x(0) = −1.0 results in a response in which the period is very long. In terms of dynamical systems theory the separatrix is a homoclinic orbit (i.e., an orbit that starts at the unstable equilibrium and ends there after infinite time) and separates bounded from unbounded (escaping) motion. An experimental analog of Eq. (16.2) based on the concept of a rolling point mass on a curved surface is described in Virgin [8] and Gottwald et al. [9]. The preceding section shows how the range of dynamic behavior is much broader when not confined to small amplitudes, especially when the axial load is somewhat higher than the initial buckling load. The membrane, or stretching, effect that was encountered earlier for axially constrained systems very easily leads to nonlinear vibrations.

16.3 A Mass Between Stretched Springs Consider the analogy between a string and a point mass supported by two identical springs. One motivation for doing this is that the effects of large amplitude and of varying axial loads can be introduced without too much mathematical sophistication. Such a system is shown schematically (and in a highly deflected configuration) in Fig. 16.4. Although the springs are linear in a direction perpendicular to the SDOF, they provide a nonlinear restoring force in the x direction (the only allowable direction), and this increases in a disproportionate sense with displacement [10–12]. If we suppose that the natural length of the springs is less than L, that is, each was stretched by an amount d (put in tension by T = kd), then we have one equilibrium position at the origin. In this case, the force acting on the mass in the x

1.1

18:14

P1: RTT Chapter˙16

CUFX159-Virgin

0 521 85648 5

August 14, 2007

316

Nonlinear Vibration

k L/2

x

m

Figure 16.4. A mass supported by two stretched springs. A single-mode analog of a string.

direction (up to cubic terms in x) is 4k(L − d) 3 kd x− x, L L3

(16.3)

4k(L − d) 3 kd x+ x = 0. L L3

(16.4)

F (x) = −2 and thus the equation of motion is m¨x + 2

Note that the linearization at x = 0 depends on d. Clearly, if the motion of the mass is small then the cubic term is negligibly small and we have a linear oscillator with natural frequency (2kd)/(mL), that is, the square of the natural frequency is linearly related to the tension in the springs. In fact, if the natural length of the springs is L (and therefore d = 0) then there is no linear restoring force (in the x direction). However, we also see that for moderately large oscillations (in the x direction) the natural frequency depends on the amplitude as well. This is an example of a system with a hardening spring stiffness in which the tension in the springs exerts a nonlinear restoring force in the direction of the motion [12]. The effective natural frequency can be obtained in a number of ways. We can again assume a harmonic form for the solution [5] x = A cos ωt,

(16.5)

which can then be placed into Eq. (16.4), and balancing (i.e., equating the coefficients of) the cosine terms (after expanding the cubic term and ignoring the third harmonic), we find the expression 3 ¯ ω¯ 2 = d¯ + A¯ 2 (1 − d), 2

(16.6)

where ω¯ 2 = mω2 /(2k),

d¯ = d/L,

A¯ = A/L.

(16.7)

18:14

P1: RTT Chapter˙16

CUFX159-Virgin

0 521 85648 5

August 14, 2007

16.3 A Mass Between Stretched Springs

317

d d

d

A

Figure 16.5. The natural frequency plotted as a function of response amplitude for a number of different pre-stress levels.

These results are plotted in Fig. 16.5. We see that if there is no initial stretching of the springs (d¯ = 0) then the natural frequency is proportional to the amplitude of the motion (within the confines of the harmonic balance approximation). It should be borne in mind that this outcome is the result of a number of approximations. First, the restoring force was expanded as a power series up to, and including, cubic terms, and second, the harmonic balance solution procedure ignored the higher harmonic terms. Despite the fact that this might not be considered an obviously axially loaded structure, it does show how the dynamic response of a system is affected by the axial forces, and in this sense it can be compared with the largeamplitude oscillations of a stretched (continuous) string [11]. Finally, it is worth noting that in a practical context damping would tend to mitigate against a system’s operating in large-amplitude motion, unless of course, the system were also subject to external forcing. This type of situation will be considered later in this chapter. Augusti’s Model Revisted. The 2DOF system considered in Section 5.7 is now

revisited. We start by numerically integrating the equations of motion for the case in which there are no initial geometric imperfections and the axial load is fixed at a level of p = 1.01134; thus the equilibrium of the system is given by (θ1e = 0.2598, θ2e = 0.0), and the two linear natural frequencies are given by ω21 = 0.0226, ω22 = 0.0535. If initial conditions are chosen with zero initial velocity but very close to equilibrium [θ1 (0) = 0.25, θ2 (0) = 0.001], then the resulting motion is harmonic. This is shown in Fig. 16.6(a) as a phase projection (in terms of one of the two angles and its rate of change). The frequency spectrum in part (b) shows a spike at the natural frequency corresponding to motion in the θ1 direction (ω21 = 0.0226, f = 0.0239 Hz) together with a small spike at twice this frequency [13]. On increasing the distance of the initial condition from equilibrium [θ1 (0) = 0.027, θ2 (0) = 0.0001], we

18:14

P1: RTT CUFX159-Virgin

0 521 85648 5

August 14, 2007

318

Nonlinear Vibration

.

x 10 2

-3

(a)

0

1 1 (dB)

1

PSD for

0 Ð1 Ð2

0.25

0.26

f

Ð 20 Ð 40

2f

0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2

Frequency (Hz)

1

.

0.02

20

(c)

1

1

(dB)

0.01

PSD for

0 Ð0.01 Ð0.02

0

0.1

0.2

0.3

0.4

(d)

0 Ð 20 Ð 40 Ð 60 Ð 80

f

0

0.03

1

(e)

0.01

PSD for

0

0

0.2

(f)

20 1 (dB)

1 0.02

Ð 0.01 Ð 0.02 Ð 0.03 Ð 0.4 Ð 0.2

0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2

Frequency (Hz)

1

.

1

Ð 80

Ð 100 0

0.27

(b)

1

Ð 60

0 Ð 20 Ð 40 Ð 60 Ð 80

0.4

0

0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2

Frequency (Hz)

1

.

0.03

(dB)

(h)

0 Ð 20

1

0.01 0 Ð 0.01 Ð 0.02 Ð 0.03 Ð 0.4 Ð 0.2

20

(g)

1 0.02

PSD for

Chapter˙16

0

0.2

0.4 1

Ð 40 Ð 60 Ð 80 0

0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2

Frequency (Hz)

Figure 16.6. Sample numerical simulation results for the Augusti model [13].

get the results shown in parts (c) and (d). Now the asymmetry in the motion is apparent, and this is reflected in the higher harmonics in the frequency spectrum. The fundamental frequency has also shifted to a lower value, again reflecting a softening spring characteristic. Now with the initial conditions set at θ1 (0) = 0.37 and θ2 (0) = 0.0015, the motion has sufficient energy to traverse the potential-energy hilltop as shown in parts (e) and (f) but is still periodic as it passes around the remote equilibrium as well.

18:14

P1: RTT Chapter˙16

CUFX159-Virgin

0 521 85648 5

August 14, 2007

16.4 Nonlinear Vibration of Strings

319

In all three of these cases, the motion in the θ2 direction was minimal. However, now suppose the initial conditions are θ1 (0) = 0.027 and θ2 (0) = 0.05, that is, similar to those used in part (c) but now with a larger initial value in the θ2 direction. This results in the behavior shown in parts (g) and (h). This motion is far from periodic as the trajectory appears to wander around and between the two stable equilibria. The frequency spectrum is now broadband with energy contributions at all frequencies. There is also now significant dynamic behavior in the θ2 direction (not shown). This is an example of chaotic behavior. Despite the apparent randomlike behavior, it is characterized by some interesting underlying order. However, this behavior is occurring in a 4D phase space and is somewhat nonrepresentative because no damping is present. In fact, for this type of behavior to occur in a continuous time (as opposed to a discrete-map) system, it must be nonlinear and have at least a 3D phase space. This is the circumstance for a forced SDOF nonlinear system, and we shall come back to it a little later in this chapter.

16.4 Nonlinear Vibration of Strings Returning now to the stretched string first considered in Section 6.2, we can restate the equation of motion but without resorting to the small-amplitude assumption (while still assuming planar motion) and thus consider 2 c21 ∂2 w L ∂w 2 ∂2 w 2∂ w − c = dx, (16.8) s ∂t2 ∂x2 2L ∂x2 0 ∂x in which c1 is the longitudinal wave speed. Given the boundary conditions w(0, t) = w(L, t) = 0, and the natural frequencies and mode shapes ωn = nπcs /L,

φn (x) = sin (nπx/L)

(16.9)

from Eqs. (6.15) and (6.16), we seek a solution by using an expansion of the linear modes: w(x, t) =

∞

n (t) sin (nπx/L).

(16.10)

n=1

Substituting Eq. (16.10) into Eq. (16.8), we get [14] ∞

¨ n + ω2n n = −

c21 n2 π4 n m2 2m. 4 4L

(16.11)

n=1

Assuming a single-mode response (with appropriate initial conditions), we set n = 1 to get ¨ 1 + ω21 1 +

c21 π4 3 = 0, 4L4 1

(16.12)

which is Duffing’s equation [15] with a hardening spring characteristic (because the cubic coefficient is positive).

18:14

P1: RTT Chapter˙16

CUFX159-Virgin

0 521 85648 5

August 14, 2007

320

Nonlinear Vibration

We can solve Eq. (16.12) by using elliptic integrals, but we can easily obtain an approximate analytic solution by using the method of harmonic balance. By assuming = A cos ωt, we find an amplitude-dependent frequency 2 2 π c1 3 ω2 = ω2n 1 + A2 , (16.13) 16 cs L and the anticipated stiffening effect of large-amplitude motion. Sagging cables can also exhibit nonlinear vibrations [16], including interesting nonplanar behavior [17].

16.5 Nonlinear Vibration of Beams In this section, we investigate the large-amplitude free oscillations of a clamped– clamped beam following the work of Yamaki [18, 19]. The beam behavior is described by the lateral deflection w(x, t), and the beam has an initial axial displacement U0 , such that the governing equation of motion is given by ∂4 w EA 1 L ∂w 2 ∂2 w ∂2 w EI 4 − dx + ρA = 0. (16.14) U0 + ∂x L 2 0 ∂x ∂x2 ∂t2 It is relatively easy to incorporate an initial geometric imperfection into the analysis, although this is not undertaken here. Again, it is convenient to put Eq. (16.14) in nondimensional form (akin to the procedure of Section 15.3), using w¯ = w/h, x¯ = x/L, and T = t EI/mL4 to give 1 w¯ − u0 + 6 w¯ 2 d¯x w¯ + w¨¯ = 0, (16.15) 0

where a prime denotes differentiation with respect to x¯ and an overdot with respect to T, with a natural frequency ω = mL4 /EI and u0 = (LA/I)U0 . Note that the displacement is normalized with respect to the thickness of the beam rather than the length as used to develop Eq. (15.21), and this form of equation was also encountered in Section 7.4 before the small-deflection approximation was applied. From Eq. (16.15), we see how the effective axial load depends on the lateral deflection [20–22]. Once the boundary conditions have been specified, the solution can be assumed to take the form w¯ = Ym(t)Wm(x), m = 1, 2, 3, . . . , (16.16) m

where we seek to solve for the unknown time functions Ym(t), and where Wm(x) are the solutions of the underlying linear-eigenvalue problem, that is, the orthonormal modes of vibration of the beam (which satisfy the geometric boundary conditions). Use is then made of Galerkin’s method to obtain a set of ordinary differential (Duffing-like) equations [23]. A few more details of this approach will be given later for the harmonically forced system.

18:14

P1: RTT Chapter˙16

CUFX159-Virgin

0 521 85648 5

August 14, 2007

16.5 Nonlinear Vibration of Beams

321

(wm)rms 2.0 Figure 16.7. The nonlinear fundamental natural frequency as a function of response amplitude. Static component not included.

1.0 u0 = 0 u0 = -60 u0 = 40

0

50

These equations can then be solved with an approximate analytical method (e.g., harmonic balance). For a single-mode analysis with small-amplitude oscillations we obtain the results discussed in Section 7.3. In Fig. 16.7 are shown some results for moderately large-amplitude oscillations. Here the response of the center of the clamped beam is plotted as an rms measure of a three-mode analysis that is due to Yamaki and Mori [18]. First consider the case with u0 = 0. For linear vibrations, we have the coefficient 22.37 [24, 25] for the fundamental natural frequency, shown as the solid curve. However, this increases to a value of 60 when the response magnitude (wm)rms is 2.4248, for example. When the initial end displacement is u0 = 40 (stretching), we obtain a shift to higher frequencies as expected and shown as the long-dashed curve. For a typical compressive end displacement (u0 = −60), we find some interesting behavior. In this case the beam is buckled and we see the appearance of two branches (both shown as dotted curves). The first branch is close to the origin and corresponds to oscillations about the postbuckled equilibrium position with the natural frequency corresponding to vanishingly small oscillations coincidentally close to the u0 = 0 case (see Fig. 7.6). However, there is also another branch and this gives finite frequencies starting from (Wm)rms ≈ 0.5. This corresponds to motion that traverses the remote equilibrium position [in the manner of Fig. 16.1(d)]. The participation of higher modes is also an important issue and will be discussed later when we deal with the nonlinear response of beams to harmonic excitation.

A Simplified Energy Approach. As we saw in earlier chapters, a useful means of

estimating the fundamental frequency of vibration by Rayleigh’s quotient can be based on the fact that the frequency corresponds to a stationary value in the neighborhood of a natural mode. Using an assumed displacement function, we can show that for an inexact eigenvector we get an eigenvalue that differs from the true value to the second order. This concept was introduced in Subsection 4.2.6 for discrete systems and in Subsection 4.3.2 for continuous systems in terms of the Rayleigh quotient, and we apply it here to a thin beam, including stretching effects [26].

18:14

P1: RTT Chapter˙16

CUFX159-Virgin

0 521 85648 5

August 14, 2007

322

Nonlinear Vibration

Even without an externally applied end shortening, as we have seen, if the ends of the member are constrained in the axial direction then membrane effects can occur for deflections that are not especially large. We have just seen how strings may have this effect, and membrane effects are also important in plates, which shall be considered a little later. We can take an approximate energy approach here for a simply supported thin beam of the type shown in Fig. 7.1 [27, 28]. The energy contributions are [29] L 1 T= m (w) ˙ 2 dx, (16.17) 2 0 L 1 (w )2 dx, (16.18) U = EI 2 0 L 1 2 1 (w ) dx, (16.19) VP = P 2 0 2 in which it can be shown that the induced axial load is given by L EI P= (w )2 dx, 2Lr2 0

(16.20)

where, for a beam of cross-sectional area A and radius of gyration r, we have I = Ar2 . Suppose the ends are fully pinned (i.e., immovable); then the lowest mode for the linear problem is simply a half-sine wave w = Q(t) sin (πx/L). We are primarily interested in the maximum amplitude of motion Qm. Evaluating the energy terms and adding them, we obtain the total energy constant C=

π4 EI 2 π4 EI Q + Q4 , m mL4 8mL4 r2 m

(16.21)

which can be used to obtain the phase trajectory as a function of the initial conditions. The equation of motion can be subjected to separation of variables and integrated (numerically) to obtain the natural period (and hence frequency) as a function of maximum amplitude. The frequency is normalized with respect to the linear natural frequency (ωL). The result is shown in Fig. 16.8 together with a Galerkin approach (the dashed curve [30, 31]) that yields 2 ω 3 Qm 2 =1+ . (16.22) ωL 16 r FEA can also be used to solve this type of problem [28].

16.6 Nonlinear Vibration of a Plate We return to consider the case in which the lateral deflections (and boundary conditions) of a plate are such that midplane stretching, or membrane response, occurs. This is somewhat similar in effect to externally applied axial loads, which will be considered subsequently. In general, these stretching effects depend on the in-plane

18:14

P1: RTT Chapter˙16

CUFX159-Virgin

0 521 85648 5

August 14, 2007

16.6 Nonlinear Vibration of a Plate

323

L

2.2

Galerkin dashed

2.0 1.8 1.6 1.4 1.2

1.0 0

1

2

3

4

Qm/r

Figure 16.8. The increase in natural frequency with amplitude for a simply supported strut that is constrained from moving axially at the ends.

boundary conditions but need not involve amplitudes that are considered large. No exact solutions are available, but we can again make use of an approximate approach based on assumed deflection shapes (i.e., a Galerkin approach). In general we seek solutions by using the form w(ξ, η, t) = cA(t) (x)(y),

(16.23)

where (x) and (y) are spatial mode shapes that satisfy the boundary conditions of the panel, and c is a constant. We will consider the example from Section 10.1, namely, the free-vibration behavior of a simply supported square panel, by using a single-mode approximation. Other boundary conditions will also be considered. An exact solution to the corresponding linear problem exists if two opposite sides are simply supported [32]. Importantly, we consider the case in which in-plane axial deformation is prevented along the edges. Chu and Herrmann [33] obtained a solution for the simply supported case, and Wah [3] extended their results (using a slightly simpler approach) to include other boundary conditions. Application of the Galerkin procedure leads to an ordinary differential equation of the form (very similar to the large-deflection equation for beams) d2 A c2 4 + µ A + 6 λA3 = 0, dζ2 h2

(16.24)

√ where ζ = [t/(ab)] D/ρ, µ4 = (ρ/D)a2 b2 p 2 , λ = η2 /ξ2 , where η and ξ come from the √ Airy stress function. Other plate characteristics include p = µ2 /(ab) D/ρ, the natural frequency of the linear problem, h, the plate thickness, and ρ, the mass per unit area. We recognize the equation of motion as a form of Duffing’s equation with a hardening spring characteristic. A solution to the period of the response can be obtained in terms of elliptic integrals, that is, T∗ 2K(k) = 1/2 , 2 T π 1 + 6 Ah2 µλ4

(16.25)

18:14

P1: RTT Chapter˙16

CUFX159-Virgin

0 521 85648 5

August 14, 2007

324

Nonlinear Vibration T */T 1 0.8 0.6 0.4 0.2

0

0.5

1

1.5

A/h

Figure 16.9. Effect of amplitude on the natural period of square plates: (a) continuous curve, SSSS-SS-SS; (b) dashed curve, SS-C-SS-SS; (c) dotted curve, SS-C-SS-C [3].

where K[k] is the complete elliptic integral of the first kind, and where k=

3A2 /h2 , 6A2 /h2 + µ4 /λ

(16.26)

and T is the period of the linear system, that is, T = 2π/µ2 . The values of µ4 /λ depend on the boundary conditions and three types are given here (following Wah [3]): case (a), simply supported on all four sides (SS-SS-SS-SS); case (b), simply supported on three sides and clamped on the fourth (SS-C-SS-SS); and case (c) simply supported on two opposite sides and clamped on the others (SS-C-SS-C). In each case, there is also a dependency on the aspect ratio but we restrict ourselves to square plates here (a/b = 1). Some results are shown in Fig. 16.9. We see that with these in-plane boundary conditions the period is reduced by approximately 25% when the deflections reach the thickness of the plate (and thus the frequency increases).

16.7 Nonlinear Vibration in Cylindrical Shells With reference to Fig. 10.14, we begin by looking at free vibrations under so-called shear diaphragm end conditions (but without end loading, i.e., Nx = 0). The equations of motion can be developed along similar lines to plate theory with appropriate extensions incorporating the additional geometrical effects of shells. A popular theory, mentioned in Chapter 10, is based on the Donnell–Mushtari equations [31]. Assuming a single mode of displacement in the radial direction, we have w(x, y, t) = Amn (t) cos

mπx ny mπx n2 2 sin + Amn (t) sin2 , R l 4R l

(16.27)

which, on substitution into the (partial differential) equation of motion and using a Galerkin procedure, results in the nonlinear ordinary differential equation 2 d2 ζ d2 ζ 3 dζ + ζ + ζ ζ 2 + (16.28) − γζ3 + 2 δζ5 = 0, dτ2 8 dτ dτ

18:14

P1: RTT Chapter˙16

CUFX159-Virgin

0 521 85648 5

August 14, 2007

16.8 Nonlinear Forced Vibration of Strings

1.8

325

0

1.4

1

1

2

3

_ A

4

0.6

Figure 16.10. The relation between the natural frequency and amplitude of vibration for a cylindrical shell. Solid curve, ξ = 0.5; dashed curve, ξ = 2.0 [34].

where ζ = Amn /h, τ = ωmn t, and = (n2 h/R)2 . The coefficents γ and δ are complicated geometrical expressions that depend on the aspect ratio, ξ = mπR/nl, of the mode and Poisson’s ratio, µ (see Leissa [31] for details). Evensen and Fulton [34] used the method of averaging to solve Eq. (16.28). The response is given by ζ(τ) = A¯ cos ω∗ τ,

(16.29)

in which the natural frequency depends on the amplitude in the following way: ω∗2 =

ω ωmn

=

1 − 34 γ A¯ 2 + 58 2 δA¯ 4 . 1 + 3 A¯ 2

(16.30)

16

The presence of cubic and quintic terms in Eq. (16.28) alerts us to the possibility of interesting behavior, and indeed, plotting Eq. (16.30) for a sample of geometries gives the results shown in Fig. 16.10. In this figure, the solid and dashed curves relate to ξ = 0.5 and ξ = 2.0, respectively, and we see that whether the nonlinear free vibrations can be classified as hardening or softening depends on and hence on geometry.

PART II: FORCED VIBRATION 16.8 Nonlinear Forced Vibration of Strings In this section, we add external excitation (and damping) and consider the nonlinear vibrations of a number of axially loaded systems. For small-amplitude oscillations the response is typically linear. This situation was covered in Chapter 15. However, in going from free to forced vibration, we are increasing the phase space from 2D to 3D, and then with nonlinear effects we encounter a host of (possibly very complicated) behavior. Again, the steady state is of particular interest but we shall also see that the long-term dynamic behavior may no longer be independent of initial conditions. We shall start by looking at a stretched string before moving on to consider the nonlinear forced vibration of beams.

18:14

P1: RTT Chapter˙16

CUFX159-Virgin

0 521 85648 5

August 14, 2007

326

Nonlinear Vibration

If we subject the mass-suspended model from Section 16.3, that is, a single-mode approximation of the stretched string, to harmonic excitation then the equation of motion becomes m¨x + cx˙ + 2

kd 4k(L − d) 3 x = F (t), x+ L L3

(16.31)

in which c is a linear-viscous damping coefficient. Using the notation of Tufillaro et al. [11], we can also write Eq. (16.31) as a special case of r¨ + λ˙r + ω20 (1 + Kr2 )r = f(t),

(16.32)

2(L − d) . L2 d

(16.33)

where ω20 =

2kd , mL

K=

Here, r is the radial distance from the origin, and thus r2 = x2 + y2 , in which the motion in the y direction allows for the possibility of whirling motion [35]. The (unidirectional) forcing is f(t) = [A cos(ωt), 0] and in relation to the continuous string we see by analogy with Eq. (16.12) that ω0 =

cs π , L

K=

π2 , 4L

(16.34)

where is the longitudinal extension of the string and (cs /c1 )2 = /L is assumed to be small. Finally, we scale time and transverse deflection according to τ = ω0 t and s = r/(L − d), and, assuming planar motion [s(t) = x(t)], we arrive at the forced Duffing equation: x + αx + (1 + βx2 )x = G(γ, τ),

(16.35)

in which α = λ/ω0 , β = 2(L − d)3 /(dL2 ), G = Lf /[2kd(L − d)], γ = ω/ω0 , and f is a harmonic forcing term. Again an approximate solution to this nonlinear ordinary differential equation can be obtained by a number of methods, but we shall leave consideration of specific responses until the next section when we look at the class of axially loaded structures associated with beams. It is also worth mentioning here that for typical nonlinear string vibrations the response often involves the onset of nonplanar (or whirling) motion. In this case, the problem can be solved by use of two coupled Duffing oscillators, with the suspended-mass analogy provided by a second DOF (in a direction into the page in Fig. 16.4 [6, 11, 35]).

16.9 Nonlinear Forced Vibration of Beams Returning to the response of a beam for which the ends are held a fixed distance apart, we now add the effect of an external (harmonic), lateral excitation, and add a small amount of linear-viscous damping. In this case, the terms −P cos t + C∂w/∂t are added to the left-hand side of Eq. (16.14), which corresponds to −p cos ωt +

18:14

P1: RTT Chapter˙16

CUFX159-Virgin

0 521 85648 5

August 14, 2007

16.9 Nonlinear Forced Vibration of Beams

327

(a)

|A|

(b)

|A|

c = 0.0

1.5

2.0 1.5

1.0

c = 0.4

1.0

0.5

F = 2.0 F = 1.0 F = 0.5

0.5

c=1 0.0

0.5

1.0

1.5

2.0

0

0.5

1.0

1.5

2.0

2.5

Figure 16.11. Resonance response curve based on a harmonic balance solution of Duffing’s equation with α = β = F = 1: (a) Various damping levels for fixed forcing magnitude (F = 1), (b) various forcing magnitudes for a fixed level of damping (c = 0.2).

c∂w/∂ ¯ t¯ in nondimensional terms, in which p = PL4 /(EIh) = 12PL4 /(Ebh4 ) and c = √ CL2 / mEI, and a rectangular beam of cross section bh has been assumed. Solutions to the governing equation of motion are obtained in the usual way, but before looking at the detailed response we consider a simplified single-mode analysis. Using the expansion from Eq. (16.16) with m = 1, we again obtain the forced form of Duffing’s equation, x¨ + cx˙ + αx + βx3 = F sin ωt,

(16.36)

in which α, β, and so on, depend on the various physical parameters of the beam [23]. This has the same form as Eq. (16.35), and we now consider its steady-state solution, with special attention paid to resonance phenomena. The method of harmonic balance can again be used, based on the assumed solution x = A cos ωt, to give an approximate relation between the various parameters:

3 3 2 2 (α − ω )A + βA + c2 A2 ω2 = F 2 . (16.37) 4 In Eq. (16.37) for the unforced (F = 0), undamped (c = 0) case, we retain the hardening backbone curve (i.e., ω ≈ 1 + (3/8)A2 + · · · +), obtained earlier in this chapter for free vibration (assuming both α and β are positive unity). Furthermore, with β = 0 we get the linear resonance response curve described in Fig. 15.1. Some typical nonlinear response curves are shown in Fig. 16.11. Here, in part (a), the dashed curve corresponds to the free- (undamped) vibration case. The solid curves correspond to different levels of damping (for a fixed level of forcing). With c = 0 we get the outer curve. When the damping is increased to a level of c = 0.4, we obtain a response that has a maximum amplitude close to A = 1.5 that occurs in the vicinity of ω = 1.6. For the more heavily damped case (c = 1), the response hardly exhibits resonance at all and is similar to the linear response. This can still be considered underdamped (ζ = 0.5) according to the linear description from Section 3.1.

18:14

P1: RTT Chapter˙16

CUFX159-Virgin

0 521 85648 5

August 14, 2007

328

Nonlinear Vibration (a)

|A|

(b) Response

3

2

1

0

Forcing Amplitude

0.5

1.0

1.5

2.0

Forcing Frequency

2.5

Figure 16.12. (a) Resonance response indicating stability, α = β = F = 1, c = 0.2 and (b) the folded surface.

However, the lightly damped (large-amplitude) behavior is generally quite different from the linear response. Consider Fig. 16.12(b) in which c = 0.2, and take F = 1. We note that the response does not scale linearly with forcing magnitude (for a fixed level of damping). Suppose the forcing frequency were very slowly increased from an initially low value. The response would gradually grow to a maximum (close to ω ≈ 2.2) but then suddenly drop to the lower-amplitude branch. On subsequent slow reduction in the forcing frequency, the response would again slowly grow but with a sudden jump up (close to ω ≈ 1.6) before following the original path. Over the range 1.6 < ω < 2.2, we have two stable solutions (the curve between is unstable). Thus there is a certain dependence on initial conditions and hence path-dependent behavior. Another kind of hysteresis was encountered earlier in this book with regard to snap-through buckling. The behavior here is also characterized by sudden dynamic jumps in the response, although we now have instabilities during the evolution of oscillatory behavior rather than equilibria. However, the instability phenomena are basically the same—a saddle-node bifurcation. The link between them is in going from a time-continuous system to a time-discrete system, as discussed in Section 3.4. The hysteresis can be envisioned as passing along a folded surface [8], as shown in Fig. 16.12. The points of instability correspond to vertical tangencies in the response curve and can thus be obtained from dω/dA = 0 to give 3 2 9 2 2 2 α − ω + βA α − ω + βA + c2 ω2 = 0. (16.38) 4 4 The roots of this equation define the region of multiple solutions [shown shaded in Fig. 16.12(b)] with one of the curves coinciding with the backbone curve when there is no damping. In the region of hysteresis, the relative dominance of the two possible stable steady-state oscillations is reflected in their domains of attraction, that is, which initial conditions lead to which solution. In some ways, this is similar to a standard application of Newton’s method as successive estimates of a root iterate toward a

18:14

P1: RTT Chapter˙16

CUFX159-Virgin

0 521 85648 5

August 14, 2007

16.9 Nonlinear Forced Vibration of Beams

329

converged value that depends to an extent on the initial guess. To determine the relative dominance of the two stable solutions, it would be necessary to divide the space of initial position and initial velocity into a fine grid and then numerically solve the governing equation from these starting conditions, subsequently labeling the final outcome. Outside the region of hysteresis the final outcome does not depend on the initial conditions. We shall return to this issue a little later, and also note that there are a variety of other ways in which nonlinear dynamical systems can lose stability, especially when the nonlinearity is of the softening variety [8]. And this is where perturbations of the steady-state response lead to the variational equational (with periodic coefficients) and use of Floquet theory (see Jordan and Smith [4] for more details). Having seen how lateral (bending) and axial (stretching) behavior, based on a single-mode analysis of the governing equation, are characterized by a bending over of the main resonance curve, consider the results of a multi-mode analysis of a clamped beam based on the work of Yamaki and Mori [18], in which they solved 1

2 2 (w¯ − w¯ 0 ) − u0 + 6 (16.39) w¯ − w¯ 0 d¯xx w¯ + w¨¯ = f¯ cos ωt, 0

where an initial imperfection w¯ 0 was included, although we just consider the perfect geometry here. Assuming a separable solution w¯ = m Ym(t)Wm(x) and imposing clamped boundary conditions, they arrived at a set of coupled nonlinear Duffing oscillators: Y¨n + ω2n Yn − u0 βmn Ym + 6 βkl βmn YkYl Ym m

= γn f¯ cos ωt,

k

l

m

k, l, m, n = 1, 2, 3, . . . ,

(16.40)

where the β terms are obtained from orthogonality. For a three-mode analysis for symmetric vibrations (which are most accessible experimentally), this reduces to Y¨n + ω2n Yn − u0 (β1n Y1 + β3n Y3 + β5n Y5 ) + 6 βkl βmn YkYl Ym k=1,3,5 l=1,3,5 m=1,3,5

= γn f¯ cos ωt,

n = 1, 3, 5.

(16.41)

Finally, assuming the solutions are given by Ym(t) =

3

dmj µ cos (j µωt),

µ = 1, 1/2, 1/3,

(16.42)

j =0

the method of harmonic balance can again be applied, and the results represented by the rms value of the response at the center of the beam:

(wm)rms

⎧ 2 2 2 ⎫1/2 ⎬ 1 1 1 1 ⎨ 1 2 3 = √ dmφm dm φm dm φm , ⎭ 2 2 2 2⎩ m m m

m = 1, 3, 5. (16.43)

18:14

P1: RTT Chapter˙16

CUFX159-Virgin

0 521 85648 5

330

August 14, 2007

Nonlinear Vibration

This was the approach used to determine the free-vibration curves shown in Fig. 16.7 in which p¯ = 0. Determinig the solution involves using Newton’s method to solve the 12 simultaneous cubic equations. Figure 16.13 shows typical responses when the beam is not initially subject to any axial displacement (u0 = 0) but rather axial effects are induced because of midplane stretching. The forcing magnitude is set at f¯ = 200. We see the complicated nature of the response, including various subharmonic and superharmonic waveforms. Again, regions of unstable behavior, resonant jumps, and hysteresis are apparent. In Yamaki and Mori [18], equivalent results are also shown for the buckled beam (e.g., u0 = −60), which brings into play the frequencies associated with postbuckled equilibria (see Fig. 16.7) as well as snap-through (which will be considered in the next section with regard to a panel). A companion paper [19] shows remarkable experimental confirmation of this type of behavior based on the excitation of a small duralumin test specimen in which the test frame was subjected to a constant peak acceleration with a measurement of relative displacement. Some intriguing applications of this theory at very small scales are also mentioned [36].

16.10 Persistent Snap-Through Behavior in a Plate Next, we return to the thermally loaded plate first considered in Section 10.5 in terms of free vibration and also subject to relatively mild harmonic (narrowband acoustic) excitation in Section 15.5. Now suppose that the magnitude of excitation is increased (from 130 to 155 dB). In this case sufficient energy is applied to the system so that the plate can exhibit snap-through behavior, i.e., both stable equilibria are traversed (see Fig. 10.6). Figure 16.14 shows a typical response (in terms of a strain measurement), with the temperature change held fixed at 32 ◦ F above ambient. At this post-critical level of thermal loading the lateral deflection at the center of the panel is of the order of the panel thickness. The excitation magnitude is 155 dB and is narrowly focused in the vicinity of 115 Hz. Figure 16.14(a) shows an experimental time series of strain-gauge data taken from a point on the panel (see the photo in Fig. 10.9) at (ξ, η) = (0.583, 0.416), where ξ = x/a, η = y/b. Note the contrast with Fig. 15.15. Part (b) again shows a time-lag embedded phase projection that also contrasts with the closed orbit corresponding to periodic motion, and the frequency spectrum in part (c) is decidedly broadband. This type of intermittent snap-through behavior can also be observed in the vibrations of a buckled beam and has clear implications for fatigue [37]. Figure 16.15 shows a time series in terms of lateral deflection but based on a nine- (cosine) mode Galerkin analysis for nominally the same parameter values as those of the experimental data. More details of this analysis can be found in Murphy [12]. Another way of encountering snap-through is through a slow sweep of a system parameter. This was a technique used earlier in this book to allow the evolution of a buckling instability. Figure 16.16 shows two examples of nonstationary snap-through

18:14

P1: RTT Chapter˙16

CUFX159-Virgin

0 521 85648 5

August 14, 2007

16.10 Persistent Snap-Through Behavior in a Plate

Figure 16.13. Frequency-response characteristics for a clamped beam including typical waveforms. Reproduced with permission from Elsevier [18].

behavior where the sound pressure level (SPL) is gradually ramped up and then down to produce a nonstationary transition through large-amplitude behavior. The f parameters used to generate these experimental results correspond to T/cr = 1.76 with a baseline forcing of 130 dB at 120 Hz. In both parts, the SPL is ramped

331

18:14

P1: RTT CUFX159-Virgin

0 521 85648 5

August 14, 2007

332

Nonlinear Vibration

yy (t +T/4)

500

0

-500 0

0.05 t (s)

0.1

(b)

10 3

(c)

10 0

PSD

(a)

500

yy

Chapter˙16

0

10 -3

-500 -500

0

500

10 -6

0

yy (t)

500

1000

(Hz)

Figure 16.14. Large-amplitude snap-through behavior: (a) strain time series, (b) time-lag embedded phase projection, and (c) frequency spectrum.

from 130 dB up to 150 dB and then back down to 130 dB, at roughly the same rate. In part (a), the motion is initiated as a small-amplitude periodic oscillation that grows very gradually until it is sufficiently large that the motion escapes the confines of its local potential-energy well and snap-through occurs, that is, an erratic snapping organized around the two stable equilibria. The motion then reduces in amplitude as the SPL is ramped down and settles to small-amplitude motion about the other equilibrium configuration (in the adjacent potential-energy well). In part (b) the system is initiated with nominally the same conditions. After the burst of cross-well motion, the system, on subsequent reduction in the applied SPL, settles back to small-amplitude motion about the original equilibrium. In Section 13.4, and specifically in Fig. 13.8, it was shown how the strength (in terms of forcing amplitude and frequency) of harmonic excitation could lead to a transient resonantlike condition in which the motion might escape the confines of its local potential-energy well. In essence, this is the situation with the snap-through behavior shown in Fig. 16.16. Fixing the (postcritical) temperature at T/Tcr thus determines the fundamental natural frequency. For a given frequency of excitation, the SPL is then quasistatically increased until the first occurrence of snap-through behavior occurs, that is, the system goes beyond the potential-energy hilltop associated with the unstable (almost flat) equilibrium configuration. The frequency is then incremented, and the procedure is repeated. Figure 16.17 shows a summary of 30 such runs plotted against nondimensional frequency, clearly separating regions of parameter space into snapthrough and no-snap-through regions [38].

w/h

2 1 0 -1 -2

0

0.05

0.1

0.15

0.2

0.25

t 0.3

Figure 16.15. A numerical simulation based on using nine (cosine) modes [12].

18:14

P1: RTT Chapter˙16

CUFX159-Virgin

0 521 85648 5

August 14, 2007

16.10 Persistent Snap-Through Behavior in a Plate 500 400

333

500

(a)

(b) 400

300

0 3 0

200

200

100

100

0

0

−100

−100

−200

−200

−300

−300

−400

−400

−500 0

1

2

3

4

5

6

t (s)

7

−500 0

1

2

3

4

5

6

t (s)

Figure 16.16. Evolving time series showing transient snap-through caused by sweeping the SPL from 130 dB → 150 dB → 130 dB. (a) The motion was initiated around the secondary equilibrium but then settled around the primary equilibrium. (b) The motion was initiated around the secondary equilibrium and returned there after a burst of transient snap-through.

The similarity between Figs. 16.17 and 13.8, at least in terms of the boundary between escape (snap-through) and no escape (no snap-through) is noted. Although there are many details associated with this behavior, we can envision this situation as the growth of a softening resonance effect that encounters instability and subsequent transition beyond the locally bounded area of phase space.

SPL (dB)

154 Snap-Through 152 150 148 146 144 No Snap-Through 142 140

0.75

0.8

0.85

0.9

0.95 n

Figure 16.17. Snap-through boundary plotted in parameter space using tonal inputs T/Tcr = 1.95, ωn ≈ 111 Hz at this temperature.

7

8

18:14

P1: RTT Chapter˙16

CUFX159-Virgin

0 521 85648 5

August 14, 2007

334

Nonlinear Vibration

16.11 A Panel in Supersonic Flow Some axially loaded slender structures are also subject to nonconservative forces, e.g., in fluid–structure interaction, or Beck’s problem (see Section 7.8), and some of this behavior can lead to nonlinear oscillations [39, 40]. We shall look at a typical situation in which axial load and fluid loading conspire to produce some complex dynamic behavior. This will also provide an example of chaotic oscillations that are not generated by periodic forcing [41]. Only a simplified analysis will be conducted in which a number of assumptions are made in order to ease the analysis [42, 43]. Consider the thin elastic simply supported panel shown in Fig. 16.18. It has a length L, is subject to a constant axial load Nx , and has flow velocity U that produces a dynamic pressure p. We assume that there is no structural damping (although some energy dissipation will be introduced by the aerodynamic modeling). If we assume that the width of this panel (into the page) is infinite then we can use a 2D form of Von Karman’s plate theory [39] to describe lateral deflections w(x, t) by means of solutions of ∂2 w ∂2 w Eαh a ∂w 2 ∂4 w ρmh 2 − Nx + dx + D + pˆ = 0. (16.44) ∂t 2a 0 ∂x ∂x2 ∂x4 The parameters used are panel thickness h, Young’s modulus E, material density ρm, in-plane stiffness α, D is the same as in Chapter 10, that is, D = Eh3 /(12(1 − ν2 ), and pˆ is the aerodynamic pressure loading. The spatial parameters and time can be nondimensionalized following the procedure of Section 7.2, that is, D w¯ = w/h, x¯ = x/L, τ = t , (16.45) ρmhL4 such that Eq. (16.44) can be written as 1 2 2 ∂ w¯ ∂2 w¯ ∂w¯ ∂4 w¯ 2 − R + 6α(1 − ν ) d¯ x + + Pˆ = 0, x ∂τ2 ∂x¯ ∂x¯ 2 ∂x¯ 4 0

(16.46)

where we now have scaled axial load Rx = Nx a2 /D and scaled dynamic pressure that can be obtained from piston theory for supersonic flow speeds, that is, Mach number M 1. This is a linear aerodynamic theory [39, 42], and in nondimensional terms, U x P ED w

Figure 16.18. A slender panel subject to both axial loading and supersonic flow over one face.

18:14

P1: RTT Chapter˙16

CUFX159-Virgin

0 521 85648 5

August 14, 2007

16.11 A Panel in Supersonic Flow

335

causes a pressure on the panel of the form

µ ∂w¯ ∂w¯ + , Pˆ = λ ∂x¯ Mλ ∂τ

(16.47)

where µ and λ are nondimensional flow density and flow velocity, respectively. We next conduct a standard Galerkin procedure and assume a series expansion for the motion w(τ, ¯ x) ¯ =

N

qm(τ) sin mπx. ¯

(16.48)

m

Placing this expression back in Eq. (16.46), we obtain a set of N coupled ordinary differential equations: N ∂2 qm + (mπ)2 qm Rx + (mπ)2 + 3α(1 − ν2 ) (iπ)2 q2i + 2Pˆ m = 0, (16.49) ∂τ2 i in which

1

Pˆ m =

Pˆ sin mπx¯ d¯x

(16.50)

0

=λ

N i

qi

mi m2 − i2

[1 − (−1)m+i ] +

µλ dqm . M dτ

(16.51)

Previous studies on this system have indicated that assuming two terms in the solution [Eq. (16.48)] gives the correct qualitative results, although six or eight terms should be retained for accurate quantitative results [39]. It is interesting to note that these equations may be numerically “stiff,” which requires special care in their solution [44]. In relating this analysis to a more practical setting it may be necessary to incorporate some in-plane stiffness at the boundaries [45]. Taking the first two harmonics leads to the coupled equations

q¨ 1 + U˙q1 − U2 q2 + q1 (1 − P) + 4q1 q21 + 4q22 = 0, (16.52)

q¨ 2 + U˙q2 + U2 q1 + 4q2 (4 − P) + 16q2 q21 + 4q22 = 0, where some additional nondimensionalization has been undertaken, and we focus on the role of two nondimensional parameters: a flow velocity U and an axial load P. In general these equations can be solved only numerically, and we might anticipate quite complicated solutions depending on the parameters. Before conducting some numerical experiments, it is instructive to consider the loss of stability from the trivial (flat) equilibrium configuration. In this case, if the axial loading is gradually introduced we encounter our familiar transition to buckling. Given the scaling in Eqs. (16.52) we see the Euler load of P = 1 that corresponds to Rx = −π2 . But a loss of stability can also be induced by the fluid loading through increasing the parameter U. Suppose we have no axial loading, and we restrict ourselves to relatively small-amplitude oscillations. In this (linear) case we can assume that the solution consists of exponential terms of the form r = Aeλt , which leads

18:14

P1: RTT Chapter˙16

CUFX159-Virgin

0 521 85648 5

August 14, 2007

336

Nonlinear Vibration

directly to the state matrix ⎤ 0 U2 0 0 ⎥ ⎥ ⎥ [r], −U −16⎦ 1 0

⎡

−U −1 ⎢ 1 0 ⎢ [˙r] = ⎢ −U2 ⎣ 0 0 0

(16.53)

in which r = (dq1 /dt, q1 , dq2 /dt, q2 )T , the eigenvalues of which determine the form of the motion. Specifically, we have the possibility that positive eigenvalues lead to a local growth of motion and hence instability. In general, in this book, the monitoring of an eigenvalue has typically been associated with real eigenvalues (which is proportional to the square of the frequency) and their sign changing from negative to positive. However, we now consider another route to instability for the case in which the system eigenvalues are complex, and the imaginary part is associated with frequency. Setting U = 0 in Eq. (16.53) leads to purely imaginary eigenvalues in this state-variable format. With Euler identities this means oscillatory motion (which is what we would expect for a plate without external forcing or damping—recall that with our simplified modeling the aerodynamics is the only source of energy dissipation). However, as the flow rate (described by U) increases, the system eigenvalues change and we can track their movement in the complex plane as shown in Fig. 16.19(a). Part (b) shows the variation of the real part of the system eigenvalues with changing U, and instability occurs when the critical value Uc = 3.59 is reached. This is a Hopf bifurcation (and quite different from Beck’s problem) and is the other way in which a system under the action of a single control parameter might generically lose its stability (along with the saddle-node). In aeroelasticity, this phenomenon is known as flutter (and the static instability that is due to the axial load is often called divergence). Both of these instabilities can also be traced back to the root structure of the linear oscillator (Fig. 3.4), and thus we identify the Hopf bifurcation with the onset of negative damping, in addition to the loss of stiffness encountered throughout this book. 5 Im

(a)

1 Re(λ) 0

(b)

−1

0

Uc = 3.59

−2

−5

−3

−3

−2

−1

0

Re 1

0

1

2

3

4

U

5

Figure 16.19. (a) A root locus of the state eigenvalues, (b) the real part of the eigenvalues plotted as a function of the control parameter U.

18:14

P1: RTT Chapter˙16

CUFX159-Virgin

0 521 85648 5

August 14, 2007

16.12 Chaotic Behavior

When the two control parameters operate together on this moderately highorder system we can expect a myriad of types of response. The following presents a summary of typical behavior. In Fig. 16.20 a variety of time-series responses are shown for the panel with no axial load (on the left) together with a similar set when a constant axial load of twice the buckling load is present, that is, P = 2. In all these cases, the inital conditions consist of a half unit in each of the two modal amplitudes but with zero initial velocity. As the flow velocity is increased, we have increased damping and the initial condition leads to a stable equilibrium for the flat panel. The damping increases with flow rate up to a certain extent and then starts to decrease. When U = 4.2, the flow speed is greater than the critical value corresponding to Fig. 16.19, and we observe the appearance of a limit cycle oscillation (LCO). The amplitude of this periodic motion grows with flow rate and follows a classic (nonlinear) Hopf bifurcation scenario [46]. The right-hand column indicates roughly similar behavior but now the stable equilibrium has shifted to a nontrival (buckled) value. However, for a certain range of flow rates (including U = 2.2), the panel returns to a stable equilibrium at the origin, that is, the flat panel. Subsequently, flutter occurs again, but this time at a lower flow rate than was the case for the unloaded panel. Figure 16.21 shows a more complicated periodic response when the axial load is P = 5 and the flow rate is U = 1.85. Here the periodic, but non-simple, trajectory seems to dwell in the vicinity of the positive and negative underlying equilibria. The phase projection in part (b) shows an alternative view. This is a relatively high-order (phase-space) system (depending on how many modes we retain in the expansion), and thus we should not be surprised to learn that chaos is not uncommon in this particular system. It is instructive to summarize these responses in term of the control parameters P and U. Figure 16.22 shows how the response of the panel depends on the combination of control parameters. We can thus locate the specific responses illustrated from Figs. 16.20 and 16.21 as the black data points, as well as the critical flow rate when no axial load is acting. It is interesting to note that it is possible for the panel to be flat and stable for certain flow rates even though the panel is buckled and stable for lower flow rates. To the left of the wide-dashed gray curve in the LCO region is where nonsimple oscillations may typically occur. The figure is the result of many numerical simulations but with the same initial conditions, and different responses are possible depending on the choice of initial conditions. Previous studies have included the effect of a static pressure differential across the panel (e.g., the top surface of an airplane wing), and a distributed stiffness in the form of a supporting elastic foundation [39].

16.12 Chaotic Behavior In a number of situations in this final chapter, we have observed behavior that appears to be erratic in nature, for example, Figs. 16.6(g) and 16.14. Both of these randomlike responses are characterized by broadband power spectra, that is,

337

18:14

P1: RTT Chapter˙16

CUFX159-Virgin

0 521 85648 5

August 14, 2007

338

Nonlinear Vibration P=0

P=2

w

w

0.5

0.5

0.0

10

20

30

40

U = 0.2

0.0

w

w

0.5

0.5

10

20

30

40

U = 1.2

0.0

−0.5

−0.5

w

w

0.5

0.5

0.0

10

20

30

40

U = 2.2

w

10

20

30

40

0.0

10

20

30

40

10

20

30

40

0.5

10

20

30

40

U = 3.2

−0.5

0.0 −0.5

w

_

0.5

−0.5

40

w

0.5

0.0

30

−0.5

−0.5

0.0

20

−0.5

−0.5

0.0

10

10

20

30

40

U = 4.2

x = 0.75 q1(0) = 0.5 q1(0) = 0.0 q2(0) = 0.5 q (0) = 0.0

.

.

2

Figure 16.20. A summary of transient responses of the infinite panel with different combinations of the control parameters U and P.

18:14

P1: RTT Chapter˙16

CUFX159-Virgin

0 521 85648 5

August 14, 2007

16.12 Chaotic Behavior

339 (a)

w

(b)

w 1

1

0.5

0.5 60

70

80

90

−1

100

−0.5

−0.5

0.5

1

−0.5

−1

-1

Figure 16.21. A nonsimple periodic response when P = 5 and U = 1.85.

motion in which a multitude of frequencies actively participate. However, although a similar response might be obtained from a very high-order system or one in which noise were present, these types of responses are examples of low-order deterministic chaos. This is a feature of nonlinear dynamical systems that has received considerable attention over recent times, and we finish this final chapter by examining an abstract model of an axially loaded structure that exhibits chaotic behavior. We start by recalling the forced form of Duffing’s equation: x¨ + cx˙ + αx + βx3 = F sin ωt.

(16.54)

Depending on the parameters, this nonlinear oscillator with its 3D phase space, that is, requiring the position, velocity, and forcing phase to uniquely determine the solution, is capable of exhibiting a vast array of behavior. Earlier in this chapter we saw how a region of hysteresis was possible, including a dependence on initial conditions. We now look at some typical responses of Eq. (16.54) in which the forcing is such that chaos occurs. Suppose we set the parameters as α = β = 1, c = 0.3, ω = 1.2, and F = 0.5. Numerically integrating Eq. (16.54) leads to the results shown in Fig. 16.23. The time

P 5

Buckled, but dynamically stable

4

Limit cycle oscillations (LCO)

3 2 1

Flat and stable

0 0

1

2

3

4

U

Figure 16.22. Dependence of the panel response as a function of axial load P and flow rate U.

w

18:14

P1: RTT Chapter˙16

CUFX159-Virgin

0 521 85648 5

August 14, 2007

340

Nonlinear Vibration

x

(a)

(b)

1.5

x

1

1 0.5

.

0.5

x

t 0 −0.5 −1

20

40

60

80

100

−1

−0.5

0.5

1

−0.5 −1 −1.5

Figure 16.23. A typical chaotic response, c = 0.3, ω = 1.2, F = 0.5: (a) time series; (b) phase projection.

series in part (a) shows a randomlike hopping around and between the equilibria (at ±1). Clearly, this behavior bears a strong similarity to the intermittent (postbuckled) snap-through motion shown in Fig. 16.15. Again, a convenient alternative form for displaying this response is the phase projection (velocity versus position), and this is shown for the same data in part (b). However, despite the apparent randomness of these responses, they are deterministic and there is a good degree of order underlying this behavior. A useful technique (especially for SDOF oscillators) is the Poincare´ section. This was introduced in Section 15.2 in terms of an analytical expression for a linear forced oscillator. This can still be obtained for nonlinear dynamical systems but not typically in closed form. If a time series is sampled at intervals of the forcing period, as shown in Fig. 16.24, then a periodic orbit would penetrate this section at the same location [47]. In this illustration, the periodic orbit repeats itself very two forcing cycles. However, if the response is chaotic then an interesting sequence of points is mapped because of the folding and stretching evolution of the chaotic attractor. This is persistent behavior and not associated with any initial transients that may be present. It is also in stark contrast to a damped unforced oscillator in which equilibrium represents a point attractor, or a damped forced oscillator in which a steady-state motion is represented by a periodic attractor. An example of a chaotic attractor is shown in Fig. 16.25 for the same motion as in Fig. 16.23. This fine structure shows some fractal characteristics and displays an extreme sensitivity to initial conditions (about 10,000 points after transients have been allowed to die out are plotted). Many numerical tools have been developed to shed light on chaos. We have already noted the broadband nature of the frequency spectrum [48, 49], but the complex geometry of the attractor can be described in terms of dimension [50] (and this is where certain fractal features are apparent). Initial conditions do not affect the qualitative nature of behavior in the vicinity of isolated point and periodic attractors, but we saw earlier how hysteresis allowed for some dependence of the final outcome on initial conditions. For a chaotic response, this dependence is extreme. There is a local exponential divergence of adjacent points on a trajectory and this feature is described by a positive Lyapunov

18:14

P1: RTT Chapter˙16

CUFX159-Virgin

0 521 85648 5

August 14, 2007

16.12 Chaotic Behavior

341

y = dx/dt T = period

T = period

x B

B

B

t A

A

A y

section n + 1

section n

section n + 2 B x

Steady states Tranisents

A

Figure 16.24. Schematic of a stroboscopic trajectory sampling: the Poincare´ section. Reproduced with permission from Thompson and Stewart [47].

exponent (LE). The LEs are related to the characteristic eigenvalues from Chapter 3 [51]. There are other measures, including the autocorrelation function [8]. The reader is referred to specialized texts for more details [47, 52]. The sensitivity to initial conditions is illustrated in Fig. 16.26. Here, Duffing’s equation, with the parameters set for periodic motion, is numerically integrated by a fine grid of initial conditions, and the black and white regions correspond to those initial conditions (basins of attraction) that lead to periodic motion about the +1

xp

Figure 16.25. A chaotic attractor based on a numerical simulation of Duffing’s equation. c = 0.3, α = −1, β = 1, ω = 1.2, F = 0.5.

xp

18:14

P1: RTT Chapter˙16

CUFX159-Virgin

0 521 85648 5

August 14, 2007

342

Nonlinear Vibration 2 . 1.5 x(0) 1 0.5 0 -0.5 -1 -1.5 -2 -2

-1.5

-1

-0.5

0

0.5

1

1.5 x(0)

2

Figure 16.26. Fractal basin boundaries based on the numerical simulation of Duffing’s equation from a fine grid of initial conditions. c = 0.168, α = −0.5, β = 0.5, ω = 1, F = 0.15.

and −1 equilibrium positions, respectively [53–55]. Hence there is a degree of uncertainty about the exact location of an initial condition; it may be very difficult to say which of the possible steady states the transient will be attracted to. The fractal nature of these basin boundaries remains no matter how fine the grid, and of course, in an experimental context there is always a degree of imprecision. Thus we see that sensitivity to initial conditions in terms of basin boundaries may occur even when steady-state chaos is not present, as only periodic solutions are present in Fig. 16.26. There are some other universal features of chaos that have made its study fascinating. Many nonlinear structures can exhibit chaos, including, for example, the shallow arch [56, 57], although it should be mentioned that most practical designs would not typically encounter this type of thoroughly nonlinear behavior. Often the broad characteristics of chaotic attractors are quite similar. For example, Fig. 16.27 shows a typical Poincare´ section taken from the model of the forced suspended mass from Section 16.3 (with d = 0), and thus we have a purely cubic oscillator that can be thought of as analogous to a laterally excited strut in which an axial load is maintained at its critical buckling value. It should be pointed out that to induce chaos in this particular case the forcing needs to be relatively large. Also, chaos can often occur after a sequence of period-doubling bifurcations [58] or can be manifest in other standard sequences including intermittency and quasi-periodicity [59]. Clearly, the study of chaos relies heavily on numerical simulation (and graphics), but some progress has been made analytically, for example, in the development of Melnikov theory to predict the onset of strange attractors, in which use is made of perturbation theory [60–62]. Finally, some experimental evidence for chaos in axially loaded structures is described. Figure 16.28 is a Poincare´ section based on the response of a magnetoelastic thin strip (a buckled beam) [63], which can be modeled by Duffing’s equation [64, 65]. The double-well shape of the underlying potential energy function is a form we have come to know well in this book.

18:14

P1: RTT Chapter˙16

CUFX159-Virgin

0 521 85648 5

August 14, 2007

16.12 Chaotic Behavior

343

. xp

xp

Figure 16.27. A Poincare´ section from the suspended mass problem, d = 0.

Epilogue We have now covered all the scenarios laid out in the point of departure at the start of this book. Vibration and buckling play out in a variety of interesting ways, ranging from linear free vibrations of axially loaded rigid-link models all the way to large-amplitude forced vibrations of axially loaded continuous systems. On this journey, we have come to rely on approximate techniques or numerical simulation as access to exact solutions has become limited. However, the dynamic behavior of structures in which there is a degree of axial loading is important and has widespread

Figure 16.28. A Poincare´ section taken from an experimental nonlinear beam. Reproduced with the permission from Elsevier [63].

18:14

P1: RTT Chapter˙16

CUFX159-Virgin

0 521 85648 5

344

August 14, 2007

Nonlinear Vibration

application. The quest for lighter aerospace structures is just one example of how the types of problem discussed in this book have an increasingly key role to play. References [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18]

[19]

[20] [21] [22] [23]

P. Hagedorn. Nonlinear Oscillations. Clarendon, 1981. N. Krylov and N. Bogoliubov. Introduction to Non-Linear Mechanics. Princeton University Press, 1949. T. Wah. Large amplitude flexural vibration of rectangular plates. International Journal of Mechanical Sciences, 5:425–38, 1963. D.W. Jordan and P. Smith. Nonlinear Ordinary Differential Equations. Oxford University Press, 1999. C. Hayashi. Nonlinear Oscillations in Physical Systems. Princeton University Press, 1964. A.H. Nayfeh and D.T. Mook. Nonlinear Oscillations. Wiley, 1979. J. Kevorkian and J.D. Cole. Perturbation Methods in Applied Mathematics. SpringerVerlag, 1981. L.N. Virgin. Introduction to Experimental Nonlinear Dynamics: A Case Study in Mechanical Vibration. Cambridge University Press, 2000. J.A. Gottwald, L.N. Virgin, and E.H. Dowell. Experimental mimicry of Duffing’s equation. Journal of Sound and Vibration, 158:447–67, 1992. G.H. Argyris and H.-P. Mlejnek. Dynamics of Structures. North-Holland, 1991. N.B. Tufillaro, T. Abbott, and J. Reilly. An Experimental Approach to Nonlinear Dynamics and Chaos. Addison-Wesley, 1992. K.D. Murphy. Theoretical and experimental studies in nonlinear dynamics and stability of elastic structures. Ph.D. dissertation, Duke University, 1994. H. Chen. Nonlinear analysis of post-buckling dynamics and higher order instabilities of flexible structures. Ph.D. dissertation, Duke University, 2004. J.W. Miles. Resonant, nonplanar motion of a stretched string. Journal of the Acoustical Society of America, 75:1505–10, 1984. G. Duffing. Erzwungene Schwingungen bei veranderlicher Eigenfrequenz. F. Vieweg u. Sohn, 1918. H.M. Irvine. Cable Structures. MIT Press, 1981. O.M. O’Reilly and P.J. Holmes. Non-linear, non-planar and non-periodic vibrations of a string. Journal of Sound and Vibration, 153:413–35, 1992. N. Yamaki and A. Mori. Non-linear vibrations of a clamped beam with initial deflection and initial axial displacement, part I: Theory. Journal of Sound and Vibration, 71:333– 46, 1980. N. Yamaki, K. Otomo, and A. Mori. Non-linear vibrations of a clamped beam with initial deflection and initial axial displacement, part II: Experiment. Journal of Sound and Vibration, 71:347–60, 1980. J.G. Eisley. Large amplitude vibration of buckled beams and rectangular plates. AIAA Journal, 2:2207–9, 1964. G.-B. Min and J.G. Eisley. Nonlinear vibration of buckled beams. ASME Journal of Engineering for Industry, 94:637–46, 1972. A.H. Nayfeh, W. Kreider, and T.J. Anderson. Investigation of natural frequencies and mode shapes of buckled beams. AIAA Journal, 33:1121–6, 1995. W.Y. Tseng and J. Dugundji. Nonlinear vibrations of a buckled beam under harmonic excitation. Journal of Applied Mechanics, 38:467–76, 1971.

18:14

P1: RTT Chapter˙16

CUFX159-Virgin

0 521 85648 5

August 14, 2007

References [24] W.T. Thomson. Theory of Vibration with Applications. Prentice Hall, 1981. [25] D.J. Inman. Engineering Vibration. Prentice Hall, 2000. [26] G.V. Rao and K.K. Raju. Large amplitude free vibration of beams – an energy approach, Zeitschrift fur ¨ Angewandte Mathematik und Mechanik 83:493–8, 2003. [27] A.V. Srinivasan. Large amplitude free oscillations of beams and plates. AIAA Journal, 3:1951–3, 1965. [28] C. Mei. Nonlinear vibration of beams by matrix displacement method. AIAA Journal, 10:355–7, 1972. [29] J.M.T. Thompson and G.W. Hunt. Elastic Instability Phenomena. Wiley, 1984. [30] H. Wagner. Large-amplitude free vibrations of a beam. Journal of Applied Mechanics, 82:887–90, 1965. [31] A.W. Leissa. Vibration of plates. Technical Report SP–160, NASA, 1969. [32] S.P. Timoshenko and S. Woinowsky-Krieger. Theory of Plates and Shells, 2nd ed. McGraw-Hill, 1968. [33] H.-N. Chu and G. Herrmann. Influence of large amplitudes on free flexural vibrations of rectangular elastic plates. Journal of Applied Mechanics, 23:532–40, 1956. [34] D.A. Evensen and R.E. Fulton. Some studies on the nonlinear dynamic response of shell-type structures. Technical Report, NASA TMX 56843, 1965. [35] J.M. Johnson and A.K. Bajaj. Amplitude modulated and chaotic dynamics in resonant motion of strings. Journal of Sound and Vibration, 128:87–107, 1989. [36] L. Nicu and C. Bergaud. Experimental and theoretical investigations on nonlinear resonances of composite buckled microbridges. Journal of Applied Physics, 86:5835–40, 1999. [37] B.L. Clarkson. Review of sonic fatigue technology. Technical Report, NASA Contract Report 4587, 1994. [38] K.D. Murphy, L.N. Virgin, and S.A. Rizzi. Experimental snap-through boundaries for acoustically excited, thermally buckled plates. Experimental Mechanics, 36:312–17, 1996. [39] E.H. Dowell. Aeroelasticity of Plates and Shells. Noordhoff, 1975. [40] P.J. Holmes. Bifurcations to divergence and flutter in flow-induced oscillations. Journal of Sound and Vibration, 53:471–503, 1977. [41] B. van der Pol. The nonlinear theory of electric oscillations. Proceedings of the Institute of Radio Engineers, 22:1051–86, 1934. [42] R.L. Bisplinghoff, H. Ashley, and R.L. Halfman. Aeroelasticity. Addision-Wesley, 1955. [43] J.M.T. Thompson. Instabilities and Catastrophes in Science and Engineering. Wiley, 1982. [44] C.W. Gear. Numerical Initial Value Problems in Ordinary Differential Equations. Prentice Hall, 1971. [45] C.S. Ventress and E.H. Dowell. Comparison of theory and experiment for nonlinear flutter of loaded plates. AIAA Journal, 8:2022–30, 1970. [46] P.J. Holmes. Nonlinear dynamics, chaos, and mechanics. Applied Mechanics Reviews, 43:23–39, 1990. [47] J.M.T. Thompson and H.B. Stewart. Nonlinear Dynamics and Chaos, 2nd ed. Wiley, 2002. [48] D.E. Newland. An Introduction to Random Vibrations and Spectral Analysis. Longman, 1984. [49] V. Brunsden, J. Cortell, and P.J. Holmes. Power spectra of chaotic vibrations of a buckled beam. Journal of Sound and Vibration, 130:1–25, 1989.

345

18:14

P1: RTT Chapter˙16

CUFX159-Virgin

346

0 521 85648 5

August 14, 2007

Nonlinear Vibration [50] P. Grassberger and I. Procaccia. Measuring the strangeness of strange attractors. Physica D, 9:189–208, 1983. [51] A. Wolf, J.B. Swift, H.L. Swinney, and J.A. Vastano. Determining Lyapunov exponents from a time series. Physica D, 16:285–317, 1985. [52] C. Grebogi, E. Ott, and J.A. Yorke. Chaos, strange attractors, and fractal basin boundaries in nonlinear dynamics. Science, 238:632–8, 1987. [53] C.S. Hsu. Cell-to-Cell Mapping: A Method of Global Analysis for Nonlinear for Nonlinear Systems. Springer-Verlag, 1987. [54] E. Eschenazi, H.G. Solari, and R. Gilmore. Basins of attraction in driven dynamical systems. Physical Review A, 39:2609–27, 1989. [55] H.E. Nusse and J.A. Yorke. Basins of attraction. Science, 271:1376–80, 1996. [56] N. Sri Namachchivaya and M.M. Doyle. Chaotic motion of a shallow arch. In Proceedings of the 29th AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynamics, and Materials Conference. AIAA, New York, 1988, pp. 198–209. [57] J.J. Thomsen. Chaotic vibrations of non-shallow arches. Journal of Sound and Vibration, 153:239–58, 1992. [58] M.J. Feigenbaum. Quantitative universality for a class of nonlinear transformations. Journal of Statistical Physics, 19:25–32, 1978. [59] S.H. Strogatz. Nonlinear Dynamics and Chaos. Addison-Wesley, 1994. [60] V.K. Melnikov. On the stability of the center for time periodic solutions. Transactions of the Moscow Mathematics Society, 12:1–57, 1963. [61] J. Guckenheimer and P.J. Holmes. Nonlinear Oscillations, Dynamical Systems, and Bifurcations of Vector Fields. Springer-Verlag, 1983. [62] S. Wiggins. An Introduction to Applied Dynamical Systems Theory and Chaos. SpringerVerlag, 1990. [63] F.C. Moon and P.J. Holmes. A magneto elastic strange attractor. Journal of Sound and Vibration, 65:275–96, 1979. [64] P.J. Holmes and F.C. Moon. Strange attractors and chaos in nonlinear mechanics. Journal of Applied Mechanics, 50:1021–32, 1983. [65] F.C. Moon. Chaotic and Fractal Dynamics, An Introduction for Applied Scientists and Engineers. Wiley, 1992.

18:14

P1: KAE 0521880428ind CUFX159-Virgin

0 521 85648 5

August 14, 2007

Index

abaqus finite-element package, 171, 177, 195, 253, 254, 256 Action, 14 Admissible function, 60, 62, 137 Aeroelasticity, 51, 336 Airy function, 29, 199, 201, 323 Angular momentum, 11 ansys finite-element package, 142, 171, 195 Antinode, 101 Antisymmetric mode, 105 Arc-length coordinates, 237, 240 Arnold tongues, 288 Astatic buckling load, 266 Asymmetric mode, 167, 245, 248 Attracting set, 25 Attractor, see Equilibrium point Augusti’s model, 88, 95, 207, 228, 317 auto continuation package, 94, 205, 206 Autocorrelation function, 341 Autonomous system, 52, 283 Axial stress wave, 277 Axial–torsional coupling, 158 Aysmptotic stability, 51 Basin of attraction, 310, 341 Beck’s problem, 134, 224, 334, 336 Bessel function, 106, 109 Bifurcation Flip, 290 Fold, 34 Hopf bifurcation, 51, 336 Indeterminate, 209 Period doubling, 290, 342 Pitchfork, 123, 290 Saddle-node, 33, 34, 36, 84, 88, 163, 201, 225, 227, 230, 275, 290, 328, 336 Secondary, 88, 90, 205, 206, 228 Subcritical pitchfork, 35, 37, 41, 77, 256, 267, 290 Supercritical pitchfork, 35, 68, 123, 137, 140, 242, 250, 290, 297, 305, 309 Transcritical, 34, 80, 269, 291 Bifurcation theory, 32 Biharmonic equation, 187 Bounded motion, 115, 269 Broadband excitation, 140, 174

Cable elasticity, 105 Cable sag, 104, 106 Calculus of variations, 14 Campbell diagram, 134 Canonical equations of motion, 17 Center, see Equilibrium point Center manifold theory, 32 Centrifugal forces, 131 Chaotic behavior, 319, 334, 337, 339 Characteristic eigenvalue (exponent) (CE), 284, 299 Characteristic equation, 27, 40, 47, 54, 83, 117, 126, 129, 135, 150, 156, 161, 166, 224 Characteristic multiplier (CM), 284, 299 Circular membrane, 109 Collocation method, 63 Combination resonances, 291 Comparison function, 60, 62 Complementary path, 69, 77, 82 Configuration space, 15 Conservation laws, 8, 15 Conservation of energy, 11, 20, 73, 102, 269, 313 Conservative force, 10, 30 Conservative system, 10, 72 Consistent mass matrix, 170, 195 Constant total energy, 72, 74, 312, 322 Constraint forces, 12, 15 Constraints, 15 Continuation, see Path following Coordinate transformation, 84 Coupled-mode flutter, 136 Creep buckling, 291 D’Alembert’s principle, 12, 104, 111, 154, 195 Damped natural frequency, 25 Damping, 9, 18, 64 Critical damping, 26, 40 Damping ratio, 25, 42, 226, 296 Dissipative forces, 18 Energy dissipation, 18, 25, 334 Half-power method, 310 Overdamped, 25, 27 Proportional damping, 48, 51, 61, 170 Underdamped, 25 Viscous damping, 25, 40, 226, 299, 304, 326

347

12:31

P1: KAE 0521880428ind CUFX159-Virgin

0 521 85648 5

348

August 14, 2007

Index Damping ratio, see Damping Dead loading, 86, 161 Difference equation, 298 Differential eigenvalue problem, 46, 59 Differential operator, 59 Dimension (attractor), 340 Displacement function, 102, 104, 121, 137, 169, 321 Displacement loading, 87 Dissipative forces, see Damping Divergence instability, 115, 178, 224, 336 Domain of attraction, 328 Donnell shell theory, 210, 289, 324 Double-well potential, 35 Drifting, 226 Driving force, 18 Duffing’s equation, 319, 323, 326, 327, 339, 341, 342 Dummy suffix notation, 128 Duncan polynomial, 129, 130, 197 Dynamic buckling load, 263 Dynamic equilibrium, 12 Dynamic loading, 261 Dynamic stiffness method, 133, 166 Effective force, 12 Effective stiffness matrix, 31 Eigenfunction, 59, 61, 129 Eigenvalue problem, 45 Elastic foundation, 147, 275, 337 Elastica, 74, 137, 177, 237, 240, 256 Element stiffness matrix, 169 Elliptic integrals, 126, 238, 313, 320, 323 End shortening, 46, 121, 125, 162, 238, 243, 246 Energy dissipation, see Damping Energy methods, 8 Equilibrium, 12, 13, 20, 25, 30, 31, 33–36, 38, 42, 45, 49, 50, 52, 55, 64, 67, 71, 77, 82, 93, 104, 115, 128, 180, 184, 202, 206, 225, 226, 240, 245, 258, 262, 263, 266, 273, 306, 312, 315, 332, 341 Equilibrium point Center, 28 Focus, 27, 53 Inflected node, 27 Node, 27, 34, 53 Point attractor, 41, 340 Saddle point, 27, 72, 263, 273 Sink, 27 Spiral, 27 Equivalent beam, 180 Escape, 263, 271, 332 Euclidian norm, 50 Euler identities, 48, 114, 336 Euler load, 123, 126, 143, 160, 220, 282, 305, 335 Euler–Bernoulli theory, 59, 111, 177, 237 Expansion theorem, 61 Experimental modal analysis, 174, 177, 222 Finite differences, 212 Finite-element method, 63, 168

Fixed point, 283 ¨ Flugge shell theory, 210, 211 Flapping motion, 132 Flexural rigidity, 186 Flexural–torsional buckling, 161 Flexural–torsional coupling, 157 Flip bifurcation, see Bifurcation Floquet exponents, 284 Floquet multiplier, 284, 290 Floquet theory, 284, 285, 299 Flow, 25 Flutter instability, 51, 178, 224, 336 Focus, see Equilibrium point Fold, see Saddle-node bifurcation Follower force, 136, 178 Fourier coefficients, 100 Fractal, 270, 342 Frame structure, 82, 166 Frequency-response function, 246, 294 Frequency spectrum, 222, 250, 319, 330, 340 Galerkin’s method, 62, 111, 124, 200, 206, 264, 289, 320, 322–324, 330, 335 Generalized coordinates, 13, 15, 30, 32, 46, 84, 128, 194 Generalized force, 13, 18 Generalized impulse, 19 Generalized momenta, 16, 17 Generalized velocity, 15, 30, 128 Geometric stiffness matrix, 170, 195 Half-power method, see Damping Hamiltonian, 17 Hamilton’s equations, 17 Hamilton’s principle, 8, 13, 57, 101, 111, 124, 184, 188 Hanging chain, 106, 139 Hann window, 302 Hardening spring characteristic, 316, 319, 323, 327 Harmonic balance, 314, 320, 321, 327, 329 Harmonic excitation, 268, 269, 294, 320, 321, 326, 332 Hill’s determinant, 286 Hill’s equation, 283 Hingeless blade, 132 Holonomic, 15 Homoclinic orbit, 315 Homoclinic solution, 73 Homogeneous solution, 294 Hooke’s Law, 49 Hopf bifurcation, see Bifurcation Hydrostatic pressure, 138 Hysteresis, 41, 86, 205, 206, 209, 328, 339, 340 Ideal stiffness, 155 Impact loading, 174, 277, 278 Imperfection sensitivity, 36, 78, 231, 262, 264, 266 Impulse and momentum, principle of, 19 Impulse input, 261, 267

12:31

P1: KAE 0521880428ind CUFX159-Virgin

0 521 85648 5

August 14, 2007

Index Impulse momentum theorem, 274 Impulsive force, 19 Indeterminate bifurcation, see Bifurcation Inertia, 9, 13 Inertia force, 12, 104 Inertial frame, 9 Inextensional beam theory, 121 Inextensional cable, 103 Inflected node, see Equilibrium point Initial imperfections, 32, 35, 69, 122, 177, 183, 193, 200, 228, 263, 277, 306, 310 Initial postbuckling, 122, 177 Intermittency, 342 Intermittent behavior, 330 Internal forces, 12, 239, 240 Isogrid structure, 178 Isotropic elastic material, 20, 183, 185, 233 Jacobian, 51, 52, 202, 283, 299 Kinetic energy, 17, 19, 21, 30, 36, 46, 64, 83, 101, 108, 121, 128, 138, 152, 159, 188, 196, 274 Kronecker delta, 61 Lagrange multipliers, 15, 212 Lagrange’s equation, 8, 13, 14, 19, 30, 32, 45, 57, 58, 66, 81, 83, 89, 92, 109, 120, 130, 152, 159, 224 Lagrangian, 14, 17, 83, 120, 121 Laplace operator, 187 Laplace transforms, 261 Laplacian, 108 Lateral buckling, 161 Least squares, 63, 230 Levy solution, 193 Lift-off, 243 Limit cycle oscillation (LCO), 337 Limit point, 34, 79, 86, 87, 162, 163, 230, 272 Linear momentum, 8, 11 Linear oscillator, 22, 26, 32, 261, 316, 336, 340 Linearization, 28, 39, 49, 202, 283, 313, 314, 316 Longitudinal wave speed, 319 Lumped parameter, 45, 64, 220 Lyapunov exponent (LE), 340 Lyapunov function, 55 Lyapunov stability, 50, 53, 115 Lyapunov’s direct method, 55 Lyapunov–Schmidt reduction, 63 Mass matrix, 46, 52, 59, 64, 170 Mathematica software package, 94, 240 Mathieu functions, 284 Mathieu’s equation, 282, 286, 290 MATLAB software package, 94, 240 Meissner’s equation, 291 Melnikov theory, 342 Membrane, 108 Membrane effect, 111, 117, 125, 183, 187, 202, 303, 306, 322

349 Modal analysis, 47, 51, 174, 246 Modal assurance criterion, 177 Modal interaction, 133 Modal interactions, 291 Mode jumping, 200, 205 Mode shape, 47, 114–116, 118, 122, 126, 127, 130, 154, 161, 166, 174, 180, 191, 192, 200, 204, 206, 216 Mode veering, 155 Moment frame, 166 Monodromy matrix, 284 Multiple scales, 29 Narrowband excitation, 203, 330 Neimark bifurcation, 285 Newton’s laws, 8, 11, 12, 22, 32 Newton’s method, 253, 328, 330 Nodal point, 101, 155 Node, see Equilibrium point Nonautonomous system, 52, 283 Nonconservative, 17 Nonconservative forces, 18, 51, 134, 334 Nondestructive testing, 219 Nongyroscopic forces, 30 Nonholonomic, 15 Nonlinear boundary-value problem, 240 Nonstationarity, 296, 330 Normal modes, 47, 61, 106, 127 Normal solutions, 284 Numerically stiff systems, 335 Ordinary differential equations, 45 Orthogonality, 60, 63, 99, 201, 329 Orthonormal modes, 320 Overdamped, see Damping Overshoot, 40, 261, 296 Parametric excitation, 282 Parametric resonance, 64, 288, 290, 308 Partial differential equations, 45, 97, 98, 111, 147, 201, 324 Particular solution, 294 Path following, 91, 177, 253 Continuation, 240, 290 Path integral, 10 Path-dependent behavior, 328 Period-doubling bifurcation, see Bifurcation Periodic attractor, 298, 340 Periodic behavior, 24, 56, 341 Perturbation methods, 28, 238, 264, 286 Phase portrait, 25, 27, 32, 49, 73 Phase projection, 49, 310, 330, 337 Phase space, 25, 49, 52, 64, 298, 319, 333, 339 Piston theory, 334 Pitchfork bifurcation, see Bifurcation Plate aspect ratio, 190–192, 197, 202, 206, 210 Poincare´ map, 285 Poincare´ sampling, 37, 283, 297 Poincare´ section, 299, 341, 342

12:31

P1: KAE 0521880428ind CUFX159-Virgin

0 521 85648 5

350

August 14, 2007

Index Point attractor, see Equilibrium point Positive-definiteness, 31, 46, 48, 55, 60, 169 Postbuckled stiffness, 183, 198, 205, 240, 244, 304, 306 Potential energy, 11, 16, 19, 30, 33, 36, 38, 46, 49, 52, 53, 56, 57, 66, 69, 72, 79, 83, 86, 89, 101, 107, 111, 122, 128, 137, 152, 160, 162, 169, 196, 201, 202, 223, 227, 262, 269, 272, 276, 313, 314, 332, 342 Preferred mode of buckling, 278 Pressure vessels, 234 Principal coordinates, 32, 48, 61, 84 Prismatic, 20, 111, 120, 147, 152, 158, 237, 252, 288 Proportional damping, see Damping Pulse loading, 267, 277 pulse software system, 246, 254 Quadratic form, 17, 30, 31, 36, 38, 46, 49 Quasi-periodicity, 288, 342 Quasi-static forcing, 261 Ramp input, 279 Random initial imperfections, 279 Rayleigh’s dissipation function, 19, 51, 64 Rayleigh’s method, 78, 102, 106, 111 Rayleigh’s principle, 62, 101 Rayleigh’s quotient, 56, 62, 120, 138 Rayleigh–Ritz, 62, 120, 132, 137, 139, 150, 240 Receptance, 294, 301 Resonance, 133, 143, 294, 297, 308, 327, 329 Riks’ method, 177, 180, 253 Rise of an arch, 162, 272 Rise time, 262 Ritz analysis, 195 Rotary inertia, 143, 188 Rotor blades, 131 Routh–Hurwitz criterion, 54, 285 Saddle-node bifurcation, see Bifurcation Saddle point, see Equilibrium point Sagging cables, 320 Secondary bifurcation, see Bifurcation Self-adjoint, 60, 62 Self-contact, 244 Self-weight, 111, 136, 139 Semirigid loading, 162, 303 Separation of variables, 59, 98, 322 Separatrix, 269, 315 Settling time, 262 Shallow cable, 103 Shallow elastic arch, 275 Shape function, 130, 169, 194 Shear deformation, 143, 172, 189, 234 Shear diaphragm, 210 Shear diaphragm end conditions, 324 Shear modulus, 185 Shooting method, 240, 243, 256 Sink, see Equilibrium point

Sinusoidal sweep excitation, 174 Slenderness ratio, 177, 180 Slope-deflection equations, 167 Snap-through buckling, 86, 162, 225, 262, 275, 328, 331, 332, 340 Snapdown, 162 Softening spring characteristic, 37, 315, 318, 329, 333 Solar sails, 256 Southwell plot, 201, 203, 217, 307 Southwell’s method, 132 Spectrogram, 134 Spiral, see Equilibrium point Spring–mass system, 23, 28, 295 St. Venant’s torsion, 158 Stable-symmetric branching, see Supercritical pitchfork bifurcation State variable, 23, 26, 52, 283, 313, 336 State vector, 22, 50, 52, 283 Statically indeterminate frame, 172 Stationary value, 30, 49, 56, 62, 273, 321 Step input, 261, 267 Stiffness matrix, 49, 52, 59, 64, 194 Stiffness method, 168 Strain-energy, 20, 21, 31, 53, 64, 66, 82, 111, 121, 124, 128, 148, 151, 158, 169, 187, 193, 194, 196, 224 Strain energy density, 20 Strain tensor, 187 Strange attractor, 342 Stress tensor, 187 Stretched string, 97, 319, 326 Stretching, see Membrane effect Subcritical pitchfork bifurcation, see Bifurcation Supercritical pitchfork bifurcation, see Bifurcation Sylvester’s criterion, 46 Symmetric mode, 104, 106, 167, 301 Tapered section, 150, 256 Tensor summation notation Dummy suffix, 31 Thermal buckling, 80, 199, 200 Thermal loading, 80, 142, 197, 330 Thin-walled bars, 157 Time-lag embedding, 309, 330 Timoshenko beam theory, 143, 172 Total mechanical energy, 11, 263 Traction force, 138, 188 Trajectory, 10, 15, 22, 24, 25, 34, 41, 49, 50, 73, 79, 86, 312, 319, 337, 340 Transcritical bifurcation, see Bifurcation Transient, 22, 25, 51, 86, 261, 263, 268, 270, 283, 333, 340, 342 Transient behavior, 268 Transition curves, 286, 290 Transmissibility, 269, 296, 307 Traveling wave, 101, 244 Trial function, 62 Truss structure, 166

12:31

P1: KAE 0521880428ind CUFX159-Virgin

0 521 85648 5

August 14, 2007

Index Underdamped, see Damping Unstable-symmetric branching, see Subcritical pitchfork bifurcation Variational equation, 51, 71, 283 Vector field, 22 Vibration isolation, 304 Virtual displacements, 12, 58 Virtual work, 12, 13, 18 Virtual work, principle of, 12, 169 Viscous damping, see Damping von Karman theory, 183, 198, 334 Waterfall plot, 134

351 Wave equation, 98, 102 Wavelength, 101, 105, 231, 277, 279 Wavenumber, 101 Weighted residual, 62 Whirling motion, 102, 326 Work done, 10, 46, 53, 64, 124, 129 Work-energy theorem, 10 Wronskian determinant, 285 Young’s modulus, 20 Z transform, 299 Zero-frequency kinetic energy, 56, 62

12:31

0 521 85648 5

August 14, 2007

This page intentionally left blank

19:48

P1: KAE 0521880428pree CUFX159-Virgin

0 521 85648 5

VIBRATION OF AXIALLY LOADED STRUCTURES This book concerns the vibration and the stability of slender structural components. The loss of stability of structures is an important aspect of structural mechanics and is presented here in terms of dynamic behavior. A variety of structural components are analyzed with a view to predict their response to various (primarily axial) loading conditions. A number of different techniques are presented, with experimental verification from the laboratory. Practical applications are widespread, ranging from cables to space structures. The book presents methods by which the combined effects of vibration and buckling on various structures can be assessed. Vibrations and buckling are usually treated separately, but in this book their influence on each other is examined together, with examples when a combined approach is necessary. The avoidance of instability is the primary goal of this material. Dr. Lawrence N. Virgin completed his doctorate in structural mechanics in 1986 at University College London. Since 1988, he has been at Duke University, where he teaches and conducts research in engineering mechanics. His interests are centered on the instability behavior of nonlinear dynamics systems in the context of experimental vibrations, with applications including aeroelasticity, systems with discontinuities (impact and friction), fluid–structure interaction, and buckling. He is currently Gardner Professor and Chair of the Department of Civil and Environmental Engineering and holds a secondary appointment in the Department of Mechanical Engineering and Materials Science.

August 14, 2007

19:48

P1: KAE 0521880428pree CUFX159-Virgin

0 521 85648 5

August 14, 2007

19:48

P1: KAE 0521880428pree CUFX159-Virgin

0 521 85648 5

Vibration of Axially Loaded Structures LAWRENCE N. VIRGIN Duke University

August 14, 2007

19:48

CAMBRIDGE UNIVERSITY PRESS

Cambridge, New York, Melbourne, Madrid, Cape Town, Singapore, São Paulo Cambridge University Press The Edinburgh Building, Cambridge CB2 8RU, UK Published in the United States of America by Cambridge University Press, New York www.cambridge.org Information on this title: www.cambridge.org/9780521880428 © Lawrence N. Virgin 2007 This publication is in copyright. Subject to statutory exception and to the provision of relevant collective licensing agreements, no reproduction of any part may take place without the written permission of Cambridge University Press. First published in print format 2007

ISBN-13

978-0-511-46328-0

eBook (EBL)

ISBN-13

978-0-521-88042-8

hardback

Cambridge University Press has no responsibility for the persistence or accuracy of urls for external or third-party internet websites referred to in this publication, and does not guarantee that any content on such websites is, or will remain, accurate or appropriate.

P1: KAE 0521880428pree CUFX159-Virgin

0 521 85648 5

This book is dedicated to my wife Lianne, my children Elliot and Hayley, and my parents Margaret and Alan

August 14, 2007

19:48

P1: KAE 0521880428pree CUFX159-Virgin

0 521 85648 5

August 14, 2007

19:48

P1: KAE 0521880428pree CUFX159-Virgin

0 521 85648 5

August 14, 2007

Contents

Foreword Preface

page xiii xv

1 Context: The Point of Departure . . . . . . . . . . . . . . . . . . . . . . . . . 1 2 Elements of Classical Mechanics . . . . . . . . . . . . . . . . . . . . . . . . . 8 2.1 2.2 2.3 2.4 2.5

Introduction Newton’s Second Law Energy and Work Virtual Work and D’Alembert’s Principle Hamilton’s Principle and Lagrange’s Equations 2.5.1 Constraints 2.5.2 Conservation Laws 2.6 Nonconservative Forces and Energy Dissipation 2.6.1 Damping 2.6.2 Time-Dependent Forces 2.7 Strain Energy

8 8 10 11 13 15 15 17 18 19 20

references

21

3 Dynamics in the Vicinity of Equilibrium . . . . . . . . . . . . . . . . . . . . 22 3.1 3.2 3.3

The Linear Oscillator Oscillator with a Slow Sweep of Frequency Dynamics and Stability 3.3.1 Stability Concepts 3.4 Bifurcations 3.4.1 The Saddle-Node Bifurcation 3.4.2 Bifurcations from a Trivial Equilibrium 3.4.3 Initial Imperfections 3.4.4 Bifurcations of Maps 3.5 A Simple Demonstration Model 3.6 Experiments

22 28 30 30 32 33 34 35 37 37 41

references

43

vii

19:48

P1: KAE 0521880428pree CUFX159-Virgin

0 521 85648 5

viii

August 14, 2007

Contents

4 Higher-Order Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45 4.1 4.2

Introduction Multiple-Degree-of-Freedom Systems 4.2.1 The Algebraic Eigenvalue Problem 4.2.2 Normal Modes 4.2.3 Equilibrium, Linearization, and Stability 4.2.4 Routh–Hurwitz Criterion 4.2.5 Lyapunov Functions 4.2.6 Rayleigh’s Quotient 4.3 Distributed Systems 4.3.1 The Differential Eigenvalue Problem 4.3.2 Solution Methods 4.3.3 Context Revisited

45 45 46 47 48 54 55 56 57 59 60 64

references

64

5 Discrete-Link Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66 5.1 5.2

5.3 5.4 5.5 5.6 5.7 5.8 5.9 5.10

Introduction An Inverted Pendulum 5.2.1 Static Behavior 5.2.2 Geometric Imperfections 5.2.3 Dynamic Behavior 5.2.4 A Note on Inertia A Discrete-Strut Model An Asymmetric Model A Three-Bar Model A Snap-Through Model Augusti’s Model Multiple Loads Load-Dependent Supports Path Following and Continuation

references

66 66 67 69 70 74 75 80 82 84 88 91 93 94 95

6 Strings, Cables, and Membranes . . . . . . . . . . . . . . . . . . . . . . . . 97 6.1 6.2

Introduction The Stretched String 6.2.1 The Wave Equation 6.2.2 Traveling-Wave Solution 6.2.3 Energy Considerations and Rayleigh’s Principle 6.3 A Suspended Cable 6.3.1 The Hanging Chain 6.4 A Rectangular Membrane

97 97 97 100 101 102 106 108

references

109

19:48

P1: KAE 0521880428pree CUFX159-Virgin

0 521 85648 5

August 14, 2007

Contents

ix

7 Continuous Struts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111 7.1 7.2

7.3 7.4 7.5 7.6 7.7 7.8 7.9

7.10 7.11

Introduction Basic Formulation 7.2.1 The Response 7.2.2 The Temporal Solution 7.2.3 The Spatial Solution Rayleigh’s Quotient Rayleigh–Ritz Analysis A Galerkin Approach Higher Modes Rotating Beams A Strut with a Tangential Load Self-Weight 7.9.1 A Hanging Beam 7.9.2 Experiments Thermal Loading Other Effects

references

111 111 113 114 116 120 120 124 126 131 134 136 138 139 142 143 143

8 Other Column-Type Structures . . . . . . . . . . . . . . . . . . . . . . . . . 147 8.1 8.2 8.3 8.4 8.5 8.6 8.7

A Beam on an Elastic Foundation Elastically Restrained Supports Beams with Variable Cross Section Modal Coupling Flexural–Torsional Buckling and Vibration Type of Loading A Continuous Arch

references

147 149 150 154 157 161 162 164

9 Frames . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166 9.1 9.2 9.3 9.4 9.5 9.6

A Beam with General Boundary Conditions The Stiffness Method A Self-Strained Frame Example Modal Analysis Large-Deflection Analysis A Tubular Structure

references

10

166 168 172 174 177 178 181

Plates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183 10.1 Introduction 10.1.1 Brief Review of the Classical Theory 10.1.2 Strain Energy 10.1.3 Boundary and Initial Conditions

183 183 187 188

19:48

P1: KAE 0521880428pree CUFX159-Virgin

0 521 85648 5

x

August 14, 2007

Contents

10.2 10.3 10.4 10.5 10.6

10.7

10.1.4 The Simplest Case 10.1.5 Initial Imperfections The Ritz and Finite-Element Approaches A Fully Clamped Plate Moderately Large Deflections Postbuckling Mode Jumping 10.6.1 Introduction 10.6.2 The Analytic Approach 10.6.3 Finite-Element Transient Results Cylindrical Shells

references

11

12

13

190 193 193 196 198 199 205 205 205 209 209 212

Nondestructive Testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216 11.1 Introduction 11.1.1 The Southwell Plot 11.1.2 Examples 11.2 Some Background 11.3 Snap-Through Revisited 11.4 Range of Prediction 11.5 A Box Column 11.6 Plates and Shells

216 217 219 222 225 228 230 231

references

234

Highly Deformed Structures . . . . . . . . . . . . . . . . . . . . . . . . . . 237 12.1 Introduction to the Elastica 12.2 The Governing Equations 12.3 Case Study A: Self-Weight Loading Revisited 12.3.1 Numerical Results 12.3.2 Experiments 12.4 Case Study B: A Heavy Beam 12.4.1 Numerical Results 12.4.2 Experiments 12.5 Case Study C: A Pinched Loop 12.6 Case Study D: A Beam Loaded by a Cable 12.7 The Softening Loop Revisited

237 239 240 241 242 243 244 245 248 251 256

references

259

Suddenly Applied Loads . . . . . . . . . . . . . . . . . . . . . . . . . . . . 261 13.1 13.2 13.3 13.4 13.5

Load Classification Back to Link Models Dynamic Buckling of a Plate A Type of Escaping Motion Impulsive Loading

261 262 267 268 272

19:48

P1: KAE 0521880428pree CUFX159-Virgin

0 521 85648 5

August 14, 2007

Contents

14

13.5.1 Equilibrium Behavior 13.5.2 Behavior under Sudden Loading 13.6 Snap-Through of a Curved Panel

273 274 275

references

279

Harmonic Loading: Parametric Excitation . . . . . . . . . . . . . . . . . . 282 14.1 14.2 14.3 14.4

An Oscillating End Load The Variational Equation Mathieu’s Equation Pulsating Axial Loads on Shells 14.4.1 A Curved Panel 14.4.2 A Cylindrical Shell

references

15

xi

282 283 286 288 289 289 292

Harmonic Loading: Transverse Excitation . . . . . . . . . . . . . . . . . . 294 15.1 Introduction: Resonance Effects 15.1.1 A Single-Mode Approximation 15.1.2 Beyond Buckling 15.2 The Poincare´ Section 15.3 Continuous Systems 15.4 An Application to Vibration Isolation 15.4.1 Postbuckling of a Strut Revisited 15.4.2 Experimental Verification 15.4.3 The Forced Response 15.5 Forced Excitation of the Thermally Buckled Plate

294 295 296 297 299 304 305 306 307 308

references

310

16 Nonlinear Vibration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 312 PART I: FREE VIBRATION 16.1 Introduction 16.2 Abstract Models 16.3 A Mass Between Stretched Springs 16.4 Nonlinear Vibration of Strings 16.5 Nonlinear Vibration of Beams 16.6 Nonlinear Vibration of a Plate 16.7 Nonlinear Vibration in Cylindrical Shells PART II: FORCED VIBRATION 16.8 Nonlinear Forced Vibration of Strings 16.9 Nonlinear Forced Vibration of Beams 16.10 Persistent Snap-Through Behavior in a Plate 16.11 A Panel in Supersonic Flow 16.12 Chaotic Behavior

312 312 313 315 319 320 322 324 325 325 326 330 334 337

references

344

Index

347

19:48

P1: KAE 0521880428pree CUFX159-Virgin

0 521 85648 5

August 14, 2007

19:48

P1: KAE 0521880428pree CUFX159-Virgin

0 521 85648 5

August 14, 2007

Foreword

The concept of stability is intrinsically a dynamical one. This is recognized even by the simplistic classical definition, which ignores the random disturbances of the real world and just inquires what would happen if a system were displaced to an adjacent position in phase space. So we are lucky, indeed, to have this well-conceived book written by a leading researcher who has mastered both nonlinear dynamics and the static bifurcations of elastic stability theory. The latter theory works well for conservative systems, for which powerful energy theorems are available, but needs augmenting by dynamical methods in the presence of loading that is either nonconservative or time dependent. Lawrence Virgin has of course just the right background, having chosen (in his usual thoughtful way) to work first at University College London, then with Earl Dowell at Duke University. He is currently the Chair of the Department of Civil and Environmental Engineering at Duke (which has an active interdisciplinary program in nonlinear dynamics) and has enjoyed productive collaborations with Raymond Plaut (Virginia Polytechnic Institute and State University). His previous book, Introduction to Experimental Nonlinear Dynamics (also published by Cambridge University Press), brought a welcome sense of realism into the often esoteric field of nonlinear dynamics by focusing on experimental investigations, and I am delighted to see a similar emphasis in this new book titled Vibration of Axially Loaded Structures. Understanding the buckling and vibration of structures under axial compression is of very great importance to structural and aerospace engineers, to whom this book is primarily addressed. They, together with readers from many other areas of mechanics, will be well served by Lawrence’s latest offering. The book covers a wide field, including buckling, dynamics (both linear and nonlinear), theory, and experiments, all explained in a clear and lucid style. Especially valuable are the comprehensive lists of references, which nicely complement the text. I can heartily recommend this book to all who want to see a wide-ranging and scholarly treatment that brings new insights to an important long-standing but still emerging field. Michael Thompson, FRS Cambridge, England xiii

19:48

P1: KAE 0521880428pree CUFX159-Virgin

0 521 85648 5

August 14, 2007

19:48

P1: KAE 0521880428pree CUFX159-Virgin

0 521 85648 5

August 14, 2007

Preface

General Comments Rationale and Scope r The material covered by this book spans the areas of vibration and buckling. Both of these areas can be considered as subsets of structural mechanics and play a central role in the disciplines of civil, mechanical, and aerospace engineering. r Although vibration and buckling are key elements in the teaching of advanced engineering, they are typically taught separately. However, the interplay of dynamics and stability in structural mechanics and its coverage in a single text provide an opportunity to present material in an interesting way. r The quest for stronger, stiffer, and more lightweight structural systems is making the material covered in this book increasingly important in practical applications. r By using axially loaded structures as a consistent theme, the book covers a wide variety of types of structure, methods of analysis, and potential applications without trying to cover too much. Experimental verification appears throughout. r The level of material is appropriate for upper-level, advanced undergraduate classes, and graduate students, but researchers and practicing engineers will find plenty of interest too. r The text is liberally illustrated by figures, and close to 500 technical references are given.

Acknowledgments The material presented in this book contains a synthesis of material from the general literature together with results from my own research program. In terms of the latter, this is by no means a solo endeavor, and there are a number of people I would like to thank. First, and foremost, much of the work I have conducted in this area in the past 20 years or so has been done with Raymond Plaut from Virginia Polytechnic Institute and State University. I have learned a considerable amount from his deep xv

19:48

P1: KAE 0521880428pree CUFX159-Virgin

xvi

0 521 85648 5

August 14, 2007

Preface

understanding of theoretical and applied mechanics as well as his attention to detail and meticulous approach to research. He also proofread this book, making useful suggestions and providing invaluable guidance. My path along this road goes back to Terry Roberts in Cardiff, Michael Thompson in London, and Earl Dowell here in North Carolina. I have benefited immeasurably from their influence as mentors during my formative years (and beyond). In addition to my family, of course, I’d like to thank my friends and colleagues at Duke who have contributed to a supportive environment: Tod Laursen, John Dolbow, Henri Gavin, Ken Hall, Josiah Knight, and Bob Kielb. I have had the privilege of working with many talented graduate students over a period of almost 20 years, and those whose research contributed directly or indirectly to material in this book include Phil Bayly, Kevin Murphy, Mike Todd, Kara Slade, Hui Chen, David Holland, Mike Hunter, Ilinca Stanciulescu, Sophia Santillan, and Ben Davis (who also diligently proofread the manuscript). Thanks to them all. Lawrie Virgin Durham, North Carolina

19:48

P1: RTT Chapter-01

CUFX159-Virgin

0 521 85648 5

August 14, 2007

1 Context: The Point of Departure

In the engineered world (and in a good deal of the natural world), stable equilibrium, or some kind of stationary or steady-state behavior, is the order of the day. Systems are designed to operate in a predictable fashion to fulfill their intended functions despite disturbances and changing conditions. Control systems have been spectacularly successful in maintaining a desirable (stable1 ) response given inevitable uncertainty in modeling system physics. However, there are plenty of examples of systems becoming unstable – and often the consequences of instability are severe. This book looks at the interplay between vibrations and stability in elastic structures. A brief view of an ecological system provides an effective analogy. The competition between certain species can be viewed as a coupled dynamic system in a slowly changing environment. External influences are provided by various factors including the climate, disease, and human influence. The delicate interaction is played out as conditions evolve and populations respond accordingly – usually in a correspondingly slow way also. However, an instability may occur leading to extinction on a relatively short time scale, perhaps when a disease (or massive meteorite) wipes out an entire population. This situation is not that dissimilar to the fluctuations of the stock markets (in which prediction of sudden changes is of concern to individuals and governments). In an engineering context, we typically have considerable knowledge about the underlying physics and governing equations of our systems, are able to test a system both analytically and in the laboratory, and thus have a much better chance of assessing the robustness of a system, especially its propensity to failure. However, unforeseen circumstances do occur, and it would, of course, be remiss in a book concerning stability in engineering mechanics not to mention the Tacoma Narrows suspension bridge disaster. But many other bridges and buildings have collapsed, aircraft wings and rotorblades have a tendency to flutter, ships sometimes capsize, the tracks of a railroad will buckle from time to time, electric circuits sporadically exhibit unintended feedback, machine parts are prone to fatigue, and once in a while satellites disappear into deep space. What most of these systems have in common is that they were either subject to external influences with which they could not cope 1

Some aeronautics control systems take advantage of a brief loss of stability for enhanced maneuverability.

1

15:5

P1: RTT Chapter-01

CUFX159-Virgin

0 521 85648 5

2

August 14, 2007

Context: The Point of Departure

Gravity

Figure 1.1. A deteriorating scenario.

or they changed. Perhaps an encounter with a rogue wave in the case of a ship, or collision with space debris in the case of the satellite. In this last instance an error in the units used in trajectory calculations may cause disaster but in the sense that the system was designed correctly but for the wrong conditions. Of course, there are always practical limits to how much safety or redundancy can be built into a system; the World Trade Center provided a sobering example. But it is also likely that a system is subject to slowly changing conditions, which may, of course, lead to catastrophe, but in a gradual deteriorating sense. It is with these systems that we have scope for monitoring and prediction, as their (dynamic) response may give clues about future performance. Hence, given a (structural) system in some state of rest (equilibrium) or steadystate motion (an oscillation), we seek to understand those conditions that cause a change in the nominal response, and especially where such a change is large (and instability falls squarely into this category). The theoretical framework underlying this statement is of course based on Newton’s laws and subsequent developments especially concerning concepts of energy. To crystallize this approach, consider the schematic diagram shown in Fig. 1.1. Here we might consider the behavior of a small ball allowed to roll (under the influence of gravity) on a curved surface to represent a generic structural or mechanical system. The analogy is really brought into focus if we further assume that the curve is actually associated with the underlying potential energy of the system and that the surface causes a little energy dissipation as the ball rolls. Hence the bottom of the energy “well” is identified as a position of stable equilibrium, with linear theory based on a locally quadratic minimum. Linear stability theory will also tell us that the “hilltops” are points of unstable equilibrium. In both cases, the ball will remain at rest at these extremum values of the potential energy surface. However, the important behavior is observed if the system is subject to a disturbance. In the stable case, the ball might begin to oscillate but typically return to rest at the bottom of the well. In the unstable case, the ball picks up speed and

15:5

P1: RTT Chapter-01

CUFX159-Virgin

0 521 85648 5

August 14, 2007

Context: The Point of Departure

3

F(t) P(t)

Figure 1.2. A slender axially loaded structure and its dynamic response.

departs the local neighborhood of the hilltop. These situations are well covered by linear stability theory providing the size of the perturbation is small. Extending this concept further, it is natural to ask what happens r if the morphology of the potential energy surface changes (typically slowly) such that the potential energy at a stable equilibrium position ceases to be a minimum, r or if the ball is subject to a relatively large perturbation or disturbance that may push the ball well beyond the local neighborhood of the minimum. These are the two situations depicted in Fig. 1.1. The former case is the basis of most studies in classical buckling. The application of an external axial load is assumed to take place quasi-statically, and buckling occurs (typically leading to large deflections) as the ball can no longer maintain its position. Many practical examples like this can be handled very effectively by use of statics. Most interest is naturally focused on the behavior of the system prior to buckling when the system is changing sufficiently slowly that kinetic energy can safely be ignored in the Lagrangian description (although it may still be useful to gain information based on dynamics). However, in the latter case, the application of a large (say, sudden or periodic) perturbation inevitably leads to a dynamic, perhaps unbounded, response. In fact, even in those cases in which a static approach works well, if we want to track the postcritical behavior, we may still need to use a dynamic approach, for example, one in which a system subject to a slowly increasing load results in a fast dynamic jump at buckling. Figure 1.2 adds some specificity to the scope of the material covered in this book using the behavior exhibited by a vibrating thin beam: r Figure 1.2 illustrates a beam undergoing small-amplitude free vibrations, that is, with P(t) = F (t) = 0. This is a thoroughly linear situation, with the straight configuration the only equilibrium and damping causing dynamic behavior to decay. Exact solutions are available; natural frequencies are constant and scale with the stiffness of the beam. For example, a longer beam is less stiff and thus natural frequencies are lower. Clamped boundary conditions lead to higher natural frequencies than simply supported, and so on. r The presence of a constant axial load [but with F (t) = 0] tends to reduce the natural frequencies if the load is compressive and below its critical value. If the axial load is sufficiently large (i.e., greater than critical), postbuckled (nontrivial)

15:5

P1: RTT Chapter-01

CUFX159-Virgin

4

0 521 85648 5

August 14, 2007

Context: The Point of Departure

r r

r r r

equilibria exist, and natural frequencies can be computed about these nontrivial equilibria. For laterally excited systems (F (t) / 0 but with P = 0), we can have resonance. This may also occur about postbuckled equilibria when P > Pcr . If the axial load is a function of time (say, periodic), then the system may also lose stability (depending on the frequency of excitation) through parametric resonance. If the ends of the beam are both constrained against moving (in-plane) then membrane, or stretching, forces arise. In each of the preceding scenarios the vibration may have large amplitude. Many of these scenarios might occur simultaneously. For example, a postbuckled beam might snap through if excited laterally.

Thus this range of behavior encompasses both small-amplitude and largeamplitude motion about both trivial and nontrivial equilibria. Access to analytic solutions becomes restricted as the complexity (and nonlinearity) of the system increases. Damping oftens needs to be considered also. Although the example of the prismatic beam has been used here, extensions to other types of axially loaded structures, like plates and shells is easy to envision. Furthermore, some of these situations may lead to instability (both static and dynamic), which is of particular concern to engineers. It is worth mentioning that aerospace structures provide a natural context for much of this material; the continual quest for lighter vehicles naturally brings with it issues of vibration and stability. Some practical examples of slender structures in aerospace engineering in which axial loads and dynamics may need to be considered are shown in Fig. 1.3. These images all portray aerospace systems. Spacecraft applications tend to be very lightweight: Thin-film solar sails designed for deep-space propulsion; highaltitude unmanned surveillence craft like the Predator; lightweight solar-powered high-endurance aircraft like the Pathfinder; the shuttle; international space station; rotorcraft; and military aircraft all possess slender structural components subject to a variety of loading conditions including vibration and axial-load effects. Figure 1.4 shows some other examples of slender structures. They range from bridges to pipelines, telescopes to submarines, oil tankers to high-rise buildings. The vibrations of axially loaded structures also occur at very small scales, including the increasingly important range of applications in nanotechnology. The guitar string is an obvious case. The axial load in this case can only be tensile, but it is interesting to note the slightly angled bridge of the guitar – this accounts for the slight amount of bending stiffness in the thicker strings. Hence this book is broadly divided into two main parts to cover these rather wide-ranging scenarios. A conventional division in the presentation of vibration problems is between free and forced vibration. That convention is somewhat followed here in the development of the material. However, there are occasions for which this division is not clear (e.g., an impulsive force can also be viewed as an initial velocity), but in terms of organizing the material, this seemed to be a natural

15:5

P1: RTT Chapter-01

CUFX159-Virgin

0 521 85648 5

August 14, 2007

Context: The Point of Departure

Figure 1.3. Examples of slender structures in an aerospace context. Courtesy of NASA. See color plates I–IV following page xvi.

5

15:5

P1: RTT Chapter-01

CUFX159-Virgin

6

0 521 85648 5

August 14, 2007

Context: The Point of Departure

Figure 1.4. More examples of slender structures. See color plates V–VIII following page xvi.

15:5

P1: RTT Chapter-01

CUFX159-Virgin

0 521 85648 5

August 14, 2007

Context: The Point of Departure

Figure 1.4. (continued) More examples of slender structures. See color plates V–VIII following page xvi.

choice. The next chapter will provide a brief overview of basic mechanics (which can be omitted by the more advanced reader), followed by a treatment of the interplay of dynamics and stability, without introducing too much in the way of mathematics, but still providing a flavor of the types of more practical structural systems considered later in the book.

7

15:5

P1: RTT Chapter-02

CUFX159-Virgin

0 521 85648 5

August 14, 2007

2 Elements of Classical Mechanics

2.1 Introduction This chapter develops the theoretical basis for the derivation of governing equations of motion. It starts with Newton’s second law and then uses Hamilton’s principle to derive Lagrange’s equations. A number of conservation laws are introduced. The theory is developed initially for a single particle and extended to systems of particles where appropriate. The emphasis is placed on building the theory relevant to the types of physical system of interest in structural dynamics. Other than the usual limitations regarding relativistic and quantum effects, we also restrict ourselves to translational (rather than rotational) systems, which is largely a matter of coordinates. The majority of problems in this book involve systems in which the forces developed during elastic deformation play a crucial role. Certain standard problems in classical mechanics, for example the central force motion leading to the two-body problem or particle scattering, are not relevant here and are not considered. We shall see the important role played by energy methods in studying the dynamics of structures. Classical mechanics has a long history and in-depth treatment of the subject can be found in Goldstein [1], Whittaker [2], and Synge and Griffith [3] and, of course, going back to the early developments of Newton [4], Euler [5], and Lagrange [6].

2.2 Newton’s Second Law The natural starting point in any text covering an aspect of classical mechanics are Newton’s laws of motion. They date back to 1686, with the second being the most important: A body acted upon by a force moves in such a manner that the time rate of change of momentum equals the force. Mathematically we introduce the concept of a linear momentum vector p defined as the product of mass and velocity: p = mv,

8

(2.1)

17:48

P1: RTT Chapter-02

CUFX159-Virgin

0 521 85648 5

August 14, 2007

2.2 Newton’s Second Law

9

where m is the mass and v is the velocity vector. We can thus write Newton’s second law as F=

dp d = (mv), dt dt

(2.2)

in which F is the force vector. To apply this law we need to specify motion relative to a reference frame. If we define an absolute position vector, r, in an inertial frame (i.e., a frame at rest or moving with a constant velocity relative to the “fixed” stars), then the corresponding absolute velocity vector is given by v=

dr = r˙ , dt

(2.3)

where an overdot signifies a time derivative. Thus we can further express Newton’s second law in its more familiar form as F=m

dv = m¨r = ma, dt

(2.4)

where a is an absolute acceleration vector and we have assumed m does not vary with time. Equation (2.4) is a (set of) second-order ordinary differential equation fundamental to the study of mechanics. In general, F = F(r, r˙ , t),

(2.5)

and a solution r(t) that satisfies this equation can be obtained given appropriate initial conditions r(t0 ) and r˙ (t0 ). For the types of systems of relevance to the material covered in this book, these solutions are unique. The forces entering Eq. (2.5) arise from a number of different sources in structural dynamics: stiffness, inertia, excitation and damping being the most important. The SI units of force are newtons (N), where 1 N = 1 kg m/s2 . Clearly, if F = F(t), then it would be a straightforward task to integrate Eq. (2.4) directly to obtain v(t) and then r(t). However, this will not typically be the case (as elastic forces tend to depend on the change in position), and a variety of techniques can be called on to solve differential equations. We observe at this point that solutions to equations of the type (2.4) will often involve oscillations, and also that there may not be analytic solutions available, especially in those situations in which nonlinear terms are present. Further discussion of nonlinearity and other aspects of differential equations are left to later chapters. However, the concept of stability (which will be developed continuously throughout this book) involves considering the manner in which closely adjacent solutions of Eq. (2.4) behave as a function of time, and specifically, when one of those solutions represents some kind of steady or equilibrium solution.

17:48

P1: RTT Chapter-02

CUFX159-Virgin

0 521 85648 5

August 14, 2007

10

Elements of Classical Mechanics

2.3 Energy and Work Now suppose F = F(r). We can obtain information about the solution to Eq. (2.4) by performing a path integral with respect to r along the trajectory: r(t) t t 2 d r dr F(r) · dr = F(r) · r˙ dt = m · dt (2.6) 2 dt r(t0 ) t0 t0 dt t d 2 1 1 1 (˙r )dt = mv2 (t) − mv2 (t0 ), (2.7) = m 2 dt 2 2 t0 which gives the magnitude of the velocity [rather than r(t)] provided the integral on the left-hand side of Eq. (2.6) can be performed. This is not a straightforward matter because r(t) (which is unknown) appears in the upper limit and a path integral depends on the path of integration. However, if we let the path of this integral [in Eq. (2.6)] be called C, then we can introduce the work done by the force F moving along this path as WC = F · dr, (2.8) C

and, defining the kinetic energy as T=

1 2 mv , 2

(2.9)

we can rewrite Eq. (2.7) as WC = T2 − T1 ,

(2.10)

which is a statement of the work – energy theorem. It turns out that there is a relatively large class of problems for which the work done for any admissible path between points 1 and 2 depends on only the end points of the path. In these cases forces are called conservative, and they play a dominant role in the static analysis of buckling, for example. For a conservative force F(r), consider two paths C1 and C2 connecting two points r1 and r2 . In this case we can write F · dr = F · dr, (2.11) C1

which implies that

C2

F · dr = 0,

(2.12)

where the closed integral is performed from r1 to r2 and back again. We define the work done by a conservative force in moving a particle from a reference point, r0 , to an arbitrary position r as the potential energy, r0 V(r) = Fc · dr, (2.13) r

17:48

P1: RTT Chapter-02

CUFX159-Virgin

0 521 85648 5

August 14, 2007

2.4 Virtual Work and D’Alembert’s Principle

11

and the work done in terms of the potential energy of end points we can write as r0 r0 r2 Fc · dr = Fc · dr − Fc · dr, (2.14) r1

and therefore

r2

r1

r2

Fc · dr = V(r1 ) − V(r2 ) = V1 − V2 .

(2.15)

r1

The same result can be obtained if we write F as the gradient of the scalar function: F = −∇V,

(2.16)

where, for example, in Cartesian coordinates we have ∇≡

∂ ∂ ∂ i + j + k. ∂x ∂y ∂z

(2.17)

The potential energy is defined to within an additive constant, but because the important behavior depends on the change in potential energy, this constant is usually chosen to facilitate the solution procedure (and often zero is a convenient choice). Conservation of Energy. In the absence of external forcing or damping, the con-

cept of conservation of total mechanical energy provides a useful framework for analyzing a dynamic system. Equating Eqs. (2.10) and (2.15) we have T2 − T1 = V1 − V2 ,

(2.18)

and because we can assign r2 as any point on the path, then we obtain the conservation of energy T + V = E,

(2.19)

where E is a constant and represents the total (mechanical) energy of the system. We can thus make this statement: If the forces acting on a particle are conservative, then the total energy of the particle (T + V) is conserved. These concepts are easily extended to include systems of particles, and a number of other conservation theorems can be developed. For example, if a particle is free from the effects of any force, then the linear momentum p˙ = 0 and thus p is a constant. A similar expression can be developed in terms of angular momentum. Clearly these conserved quantities can play a significant role in facilitating a solution r(t) to a physical problem.

2.4 Virtual Work and D’Alembert’s Principle In practical situations it may be quite difficult to describe all the forces acting on a system in a vectorial context. We will see that this is one of the reasons that conducting an energy approach is often easier than using Newton’s laws directly. However,

17:48

P1: RTT Chapter-02

CUFX159-Virgin

0 521 85648 5

August 14, 2007

12

Elements of Classical Mechanics

it is possible to make use of a variational principle in mechanics to facilitate the solution procedure, and this involves the concept of virtual displacements. Suppose we have a particle in equilibrium (described by a position vector r) under a set of forces F. If the position of the particle is subject to infinitesimal changes (i.e., small variations in the systems coordinates, compatible with any system constraints) then the total virtual work done is n δW = Fi δr = 0 (2.20) i=1

for a system in equilibrium, where the symbol δ is given to instantaneous, virtual variations. This can be generalized for a number of particles and, indeed, for elastic bodies, which comprise the largest interest in this book. We can thus state the principle of virtual work: For a system of forces acting on a particle, the particle is in statical equilibrium if, when it is given any virtual displacement, the net work done by the forces is zero. There are a number of ways in which this statement [and Eq. (2.20)] can be put to practical use. We can divide the forces into two categories: applied forces and constraint forces. It can be shown that the virtual work that is due to constraint forces acting through small virtual (termed reversible) displacements is zero, and the principle of virtual work is adjusted accordingly. In applications to structural mechanics, it is convenient to also divide the work into two parts: that due to external loads and that due to internal forces, and thus δWe + δWi = 0.

(2.21)

We can incorporate dynamics into the framework of virtual work by using D’Alembert’s principle. We achieve this by writing Newton’s second law as F − m¨r = 0,

(2.22)

in which m¨r is called the inertia force. Therefore we can view this as a statement of dynamic equilibrium, and in simple structural dynamics problems this is often the easiest means of obtaining the equations of motion. The statement of virtual work can thus be written in a more general form for a system of N particles of mass mi acted on by forces Fi as N (Fi − mi δ¨ri ) · δri = 0,

(2.23)

i=1

and D’Alembert’s principle may be stated thus: The virtual work performed by the effective forces through infinitesimal virtual displacements compatible with the system constraints is zero. Here, “effective forces” refers to the combination of regular and inertia forces.

17:48

P1: RTT Chapter-02

CUFX159-Virgin

0 521 85648 5

August 14, 2007

2.5 Hamilton’s Principle and Lagrange’s Equations

13

However, in contrast to the energy approaches of the next section, Eq. (2.23) still describes motion in terms of physical, vectorial coordinates. A number of issues surround the independence of coordinates, for example, in systems possessing m equations of constraint the number of degrees of freedom (DOFs) is three fewer than the number of rectangular coordinates needed to describe the positions of all the particles. Suppose we have already eliminated the forces of constraint (because they do no work) and rewrite Eq. (2.23) as N

Fi · δri = 0,

(2.24)

i=1

where Fi is now a combination of the applied and inertia forces. To satisfy equilibrium, however, we need independent coordinates for Fi = 0 (i = 1, 2, . . . , N). It can be shown that transforming from the ri coordinates to generalized coordinates qj and then taking infinitesimal virtual displacements leads to the virtual work being written in the form ⎛ ⎞ n δW = ⎝ Qj ⎠ δqj , (2.25) j =1

where Qj =

N

Fi ·

i=1

∂ri , ∂qj

j = 1, 2, . . . n,

(2.26)

and the variations in r are in the q directions. The Qj are called the generalized forces. Equilibrium is thus given by Qj = 0,

j = 1, 2, . . . , n.

(2.27)

2.5 Hamilton’s Principle and Lagrange’s Equations Although Newton’s laws are remarkably useful, there are a number of limitations. These concern systems comprising particles at very small distances and also systems in which very high velocities are involved. These types of systems are of no concern in this book, but there are many circumstances in the macromechanical world for which determining all the forces present in a system is a challenging or even impossible task. An alternative approach is based on Hamilton’s principle, which can be used to derive equations of motion via Lagrange’s equations. Although they can be shown to be equivalent to Newton’s second law, they provide a more powerful and global approach to solving problems in mechanics. A particular advantage is the flexibility in choosing coordinate systems. Attention is focused primarily on conservative systems, with a more thorough discussion of nonconservative forces left until later.

17:48

P1: RTT Chapter-02

CUFX159-Virgin

0 521 85648 5

August 14, 2007

14

Elements of Classical Mechanics

Hamilton’s principle, for conservative systems, states Of all possible paths along which a dynamical system may move from one point to another within a specified time interval (consistent with any constraints), the actual path followed is that which minimizes the time integral of the difference between the kinetic and potential energies. The integral referred to in this statement is often called the action I and can be written as t2 t2 I= (T − V)dt = Ldt, (2.28) t1

t1

where L is the Lagrangian. Thus the issue is to find the minimum of this integral, a classic problem in the calculus of variations. If L depends on a single coordinate, say q [together with its time derivative q(t) ˙ and time t], then we need to find the trajectory q(t) that minimizes t2 I= L [q(t), q(t), ˙ t] dt. (2.29) t1

To do this we need to consider what other permissible trajectories do in comparison with q(t): Any neighboring trajectory must make I increase relative to the minimum. We consider the behavior of a close-by trajectory given by q(t) + φ(t),

(2.30)

where we suppose q(t) is the path corresponding to the minimum, is a small value, and the function φ(t) is zero at the end points t1 and t2 but is otherwise any function of time. Equation (2.29) is thus transformed to t2

˙ I= L q(t) + φ(t), q(t) ˙ + φ(t), t dt. (2.31) t1

Mathematically we express the condition for a minimum as dI = 0, d which then leads to

t2 t1

∂L ∂L ˙ φ+ φ dt = 0. ∂q ∂q˙

The second term in the integrand can be integrated by parts, leaving

t2 ∂L d ∂L − φ(t)dt = 0. ∂q dt ∂q˙ t1

(2.32)

(2.33)

(2.34)

For this to be a minimum for arbitrary φ(t), the term in the large parentheses must be zero, which gives us Lagrange’s equation: ∂L d ∂L − = 0. ∂q dt ∂q˙

(2.35)

17:48

P1: RTT Chapter-02

CUFX159-Virgin

0 521 85648 5

August 14, 2007

2.5 Hamilton’s Principle and Lagrange’s Equations

15

This can be extended to incorporate situations in which the trajectory depends on a number of independent coordinates qj (t): ∂L d ∂L − = 0, ∂qj dt ∂q˙ j

j = 1, 2, . . . , n.

(2.36)

There is considerable advantage in the Lagrangian approach in that the coordinates need not be physically meaningful. Generalized coordinates consist of any set of quantities that fully describes the state of a system, and we may view our dynamic system as evolving in this configuration space. The generalized coordinates are often referred to as qj , they are not unique, and their time derivatives are the generalized velocities q˙ j .

2.5.1 Constraints In many physical situations the motion of a system is subject to constraints. That is, there is some kind of kinematic restriction on the motion, usually involving a relation between coordinates, their rates of change, or time. Forces arise because of the constraints, but because they depend on the motion itself, they are not known a priori. However, if the constraint can be expressed as position coordinate relations (or just involve time explicitly) then it can be expressed in a differential form, is termed holonomic, and can be incorporated into the Lagrangian description without much difficulty. That the motion is restricted leads naturally to a reduced number of DOFs; that is, we seek to select independent generalized coordinates that do not violate the constraints, and, because the constraint forces do no virtual work, they do not appear in the equations of motion. However, another class of problem involves constraints that influence the rates of change of generalized coordinates. They may be expressed as inequalities or as nonintegrable differential relations and are termed nonholonomic. They cannot be reduced to independent generalized coordinates, and appropriate equations of motion must include the constraints. In practice, the method of Lagrange multipliers is used [7], in which the generalized coordinates and constraint forces are obtained simultaneously. Holonomic constraint forces can also be handled in this way, although they are not of direct interest to the material covered in this book.

2.5.2 Conservation Laws We have seen how certain quantities (e.g., the total mechanical energy) may be conserved. This was developed from basic definitions of work and energy and their relation to Newton’s second law. We can show that similar relations can be developed by using the Lagrangian description. There may often occur instances in which a symmetry property enables a considerable simplification to be made, which leads to the absence of a particular

17:48

P1: RTT Chapter-02

CUFX159-Virgin

0 521 85648 5

August 14, 2007

16

Elements of Classical Mechanics

coordinate in the Lagrangian. Suppose the missing (or ignorable) coordinate is qj ; then its Lagrange’s equations will be d ∂L = 0, (2.37) dt ∂q˙ j which implies ∂L = constant, (2.38) ∂q˙ i and because ∂L/∂q˙ j = m˙qj = p j , this means the generalized momentum is conserved, that is, we effectively have a constant of the motion. Another class of problem involves those in which time does not appear explicitly in the Lagrangian, and thus ∂L = 0. (2.39) ∂t In this case, we write the total derivative (L can change in time only through its dependence on the coordinates and velocities) as dL ∂L dq ∂L d˙q = + = 0. dt ∂q dt ∂q˙ dt

(2.40)

From Lagrange’s equations, we have ∂L d ∂L = , ∂q dt ∂q˙ and therefore dL = dt

∂L d˙q d ∂L q˙ + . dt ∂q˙ ∂q˙ dt

We recognize that this is the derivative of a product, that is,

dL d ∂L = q˙ , dt dt ∂q˙ which can be written as d dt

∂L q˙ − L = 0. ∂q˙

(2.41)

(2.42)

(2.43)

(2.44)

This can now be integrated, and the term in parentheses is therefore constant in time: ∂L q˙ − L = H = constant. (2.45) ∂q˙ If the potential energy does not depend explicitly on the velocities or time, we have V = V(q), and, using L = T − V, we obtain ∂L ∂(T − V) ∂T = = , ∂q˙ ∂q˙ ∂q˙

(2.46)

∂T − (T − V) = H. ∂q˙

(2.47)

and Eq. (2.45) becomes q˙

17:48

P1: RTT Chapter-02

CUFX159-Virgin

0 521 85648 5

August 14, 2007

2.6 Nonconservative Forces and Energy Dissipation

17

Assuming the kinetic energy depends on only the generalized velocities and that this relation is the standard quadratic form, then we obtain q˙

∂T = 2T, ∂q˙

(2.48)

and Eq. (2.47) gives us the total energy of the system: H = T + V.

(2.49)

Therefore, we have the result that H (termed the Hamiltonian) is equal to the total energy of the system if the Lagrangian does not depend explicitly on time, and the potential energy does not depend on velocity. These concepts can easily be extended to include systems of particles, provided the equations of transformation relating regular and generalized coordinates is also independent of time. It is possible to express the Lagrangian (and specifically the velocities associated with the generalized coordinates) in terms of generalized momenta. An advantage of doing this includes the fact that it is often the momentum that is a conserved quantity, and the Hamiltonian also has more physical meaning (through its relation with energy for conserved systems) than the Lagrangian. It also results in a set of 2n first-order equations rather than the n second-order Lagrange’s equations, and this may assist the development of numerical solutions. It turns out that the resulting Hamilton’s equations have certain symmetric features that render them unchanged under transformation of coordinates and momenta, and they are often referred to as the canonical equations of motion.

2.6 Nonconservative Forces and Energy Dissipation Not all forces are derivable from a potential, that is, there may not be a potential energy function V that satisfies Eq. (2.16) for a particular system. We can write the external forces acting on the system in the form Fi = FPi + FDi ,

(2.50)

where FPi is derivable from a potential V = V(qi ) and FDi is not. Thus we can also divide the virtual work into conservative and nonconservative parts, δW = δWP + δWD.

(2.51)

The first term on the right-hand side of Eq. (2.51) is defined by Eq. (2.16), and, by virtue of Eq. (2.26), we can write Lagrange’s equation as ∂L d ∂L − = QDj . ∂qj dt ∂q˙ j

(2.52)

We can also rearrange Eq. (2.51) and write δWD = δW − δWP = F · dr − FP · dr = dT − (−dV) = d(T + V) = dE.

(2.53)

17:48

P1: RTT Chapter-02

CUFX159-Virgin

0 521 85648 5

August 14, 2007

18

Elements of Classical Mechanics

Integrating over the path, we get r2 FD · dr = r1

E2

dE = E2 − E1 .

(2.54)

E1

Thus the change in total energy is equal to the work done by the nonconservative forces. In structural mechanics it is often found that there are two general classes of forces that do not arise from a potential function. In these cases, the total mechanical energy is not conserved. We will occasionally consider a load that is nonconservative in nature because of its direction changing (for example, following the slope at the end of a beam), but in general there are two classes of nonconservative forces encountered in mechanical systems. In cases in which the energy decreases we use the term dissipative forces. The main example is the loss of energy through damping, which in many cases relates to a force proportional to velocity. The other main type of nonconservative force is time dependent and often associated with the external driving of a system. 2.6.1 Damping In mechanical systems, energy dissipation is inevitable. If we assume that a certain class of nonconservative force acting on a single particle is a function of velocity only, then FDi = −g(vi )vi ,

(2.55)

where we also assume that the force is directed opposite to the velocity. Very often this relation will describe linear-viscous damping in unidirectional motion: δW = Fix = −cxi x˙ i ,

(2.56)

where c is the damping coefficient. The virtual work done by this dissipative force is F · δr δW = i

=−

n

cxi x˙ i δxi

i=1

⎡ ⎤ n n ∂xi ⎣ =− cxi x˙ i δqj ⎦ ∂qj i=1

j =1

n n 1 ∂ 2 =− cx x˙ δqj . 2 ∂q˙ j i i j =1

(2.57)

i=1

Thus the corresponding generalized force is given by QDj = −

n n 1 ∂ 1 ∂ cxi x˙ i = − cxi x˙ 2i . 2 ∂q˙ j 2 ∂q˙ j i=1

i=1

(2.58)

17:48

P1: RTT Chapter-02

CUFX159-Virgin

0 521 85648 5

August 14, 2007

2.6 Nonconservative Forces and Energy Dissipation

19

Introducing Rayleigh’s dissipation function D, 1 cxi x˙ 2i , 2 n

D=

(2.59)

i=1

we can obtain the dissipative generalized forces from δW =

n

QDj δqj = −

j =1

n ∂D δqj , ∂q˙ j

(2.60)

j =1

and thus a more general form of Lagrange’s equation: ∂L ∂D d ∂L − + = 0, dt ∂q˙ j ∂qj ∂q˙ j

j = 1, 2, . . . , n.

(2.61)

2.6.2 Time-Dependent Forces We now consider another group of nonconservative forces, namely, time-dependent forces, as they crop up quite naturally in structural systems subject to periodic excitation or impulses, for example. Suppose a system is subject to forces F (t) that depend on time (but are independent of the generalized coordinates), then a timedependent Lagrangian will have a term in the form qF (t). Clearly, this will then lead to the appearance of F (t) in the resulting equation of motion. Furthermore, the principle of impulse and momentum can be used via the Lagrangian approach. Consider an impulsive force t0 +t Fˆ = F(t)dt. (2.62) t0

Intgerating Lagrange’s equation from t1 = t0 to t2 = t0 + t and allowing t → 0 leads to the impulsive form of Lagrange’s equation: ∂T ∂T ˆ j, − =Q j = 1, 2, . . . , n, (2.63) ∂q˙ j 2 ∂q˙ j 1 ˆ j is a generalized impulse. where Q It is also possible to derive a velocity-dependent potential for some problems, although this will not be encountered in the types of applications in this book. Finally, we note that extension of Lagrangian mechanics to continuous systems (with an infinite number of DOFs) will be dealt with in a later chapter. In summary, we will often be in a position to write potential and kinetic energies and make use of Lagrange’s equation: d ∂L ∂L ∂D − + = qj Fj (t), dt ∂q˙ j ∂qj ∂q˙ j

j = 1, 2, . . . , n.

(2.64)

Given the types of axially loaded elastic structures of primary interest in this book, we inevitably focus quite heavily on equations of motion of the type Mq¨ j + C˙qj + f (qj , λ) = F (t),

j = 1, 2, . . . , n,

(2.65)

17:48

P1: RTT Chapter-02

CUFX159-Virgin

0 521 85648 5

August 14, 2007

20

Elements of Classical Mechanics

in which we have a mass matrix M, damping matrix C, the stiffness matrix f , which depends (often nonlinearly) on external axial loads (λ), and both free [F (t) = 0] and forced vibrations are examined for a wide variety of slender structures, and where we often begin by assuming C = 0.

2.7 Strain Energy We now introduce a brief description of strain energy, as this is a fundamental aspect of structural mechanics that we will repeatedly encounter throughout this book. In Section 2.3 the concepts of energy and work were introduced. Because many of the specific physical systems to be considered later in this book involve deformable bodies, it is important to understand the manner in which (elastic) energy is stored as strain energy, especially in bending. We start with a basic definition of strain energy per unit volume, or strainenergy density U0 , for a uniaxially loaded system given by 1 U0 = σx dx , (2.66) 0

in which σx is stress and x is strain. We see that the strain-energy density is equal to the area under the stress–strain from x = 0 to x = 1 . A thorough background to elasticity and the general description of fundamental issues in solid mechanics can be found in Langhaar [8]. The total strain energy stored in the solid is U= U0 dV. (2.67) V

For a linear isotropic elastic material, we have σx = Ex , where E is Young’s modulus, and, in this case, expression (2.66) becomes U0 =

σx2 . 2E

(2.68)

The strain energy U0 is always positive-definite, and the conservation of energy introduced in Subsection 2.5.2 can also be stated in terms of strain energy: If an elastic body is in equilibrium under an external force system, then the internal strain energy that is due to deformation is equal to the work of the externally applied force system. Application to Beams. We finish this chapter by briefly focusing on the strain en-

ergy associated with slender beams because they represent an important element in this book, but these concepts can be easily extended to strings, plates, and so on. A prismatic bar of cross-sectional area A and length L is subjected to an axial load P (which passes through the centroid of the cross section). The stress σ = P/A and thus the total strain energy in the bar is U=

P2 L . 2AE

(2.69)

17:48

P1: RTT Chapter-02

CUFX159-Virgin

0 521 85648 5

August 14, 2007

References

Suppose the beam is subjected to an applied lateral load rather than axial and has a second moment of area I. In this case, a bending moment causes a stress σx = My/I (where y is the distance from the centroid) and thus a total strain energy of L 2 M U= dx. (2.70) 0 2EI When standard beam theory [9] for a prismatic beam is used, this simplifies to the well-known expression L 2 2 1 ∂ w U = EI dx. (2.71) 2 ∂x2 0 The interaction between axial and bending effects will be a central theme. Throughout this book extensive use will be made of energy concepts [10]. Kinetic energy and the work done by external loads are added to the consideration of strain energy for a variety of structural systems. In simpler cases, we will make direct use of Newton’s laws, but for complex systems, energy will provide a powerful (equilibrium and stability) framework in which to study the dynamics of axially loaded structures. References [1] [2]

H. Goldstein. Classical Mechanics. Addison-Wesley, 1980. E.T. Whittaker. A Treatise on the Analytical Dynamics of Particles and Rigid Bodies. Cambridge Mathematical Library, 1988. [3] J.L. Synge and B. A. Griffith. Principles of Mechanics. McGraw-Hill, 1959. [4] I. Newton. Philosophiæ Naturalis Principia Mathematica. London, 1687. [5] L. Euler. Methodus Inveniendi Lineas Curvas Maximi Minimive Proprietate Gaudentes. Marcum Michaelem Bousquet, 1744. [6] J.L. Lagrange. Mecanique Analytique. Courier, 1788. [7] L. Meirovitch. Principles and Techniques of Vibrations. Prentice Hall, 1997. [8] H.L. Langhaar. Energy Methods in Applied Mechanics. Wiley, 1962. [9] S.P. Timoshenko and J.M. Gere. Theory of Elastic Stability, 2nd ed. McGraw-Hill, 1961. [10] J.H. Argyris and S. Kelsey. Energy Theorems and Structural Analysis. Butterworth, 1960.

21

17:48

P1: RTT Chapter-03

CUFX159-Virgin

0 521 85648 5

August 14, 2007

3 Dynamics in the Vicinity of Equilibrium

We are primarily interested in the concept of a slow evolution toward, and through, instability. The quasi-static change of a system parameter (typically the axial load) allows any instability mechanisms to unfold, especially if transient dynamics is present. Before moving on to consider a range of structural components, we investigate the response of a linear oscillator under various conditions, including a continuously deteriorating stiffness. This is followed by consideration of some aspects of the Lagrangian approach and the interaction of dynamics and stability. The final part of this chapter then uses a phenomenological model to describe the dynamics and stability of a simple physical system. Again, there is a large literature concerning the linear oscillations of mechanical systems. The interested reader can find good coverage of basic material in Refs. [1–7].

3.1 The Linear Oscillator Consider the continuous-time evolution of a dynamical system governed by x˙ = F(x, t),

x ∈ Rn ,

t ∈ R,

(3.1)

where x is a state vector that describes the evolution of the system under the vector field F. Given an initial condition, typically the values of the state vector prescribed at t = 0, that is, x(0), we can seek to solve system (3.1) to obtain a trajectory x(t), or orbit, along which the solution evolves with time. We then seek to ascertain the stability of the system, generally as a function of a (control) parameter µ, and thus consideration of x˙ = F(x, µ, t)

x ∈ Rn ,

t ∈ R,

(3.2)

will play a central role in the material contained in this book [8, 9]. Application of Newton’s second law relates acceleration and force (and hence position), and thus often results in a second-order ordinary differential equation of the form d2 x = −ω2n x, dt2 22

(3.3)

17:50

P1: RTT Chapter-03

CUFX159-Virgin

0 521 85648 5

August 14, 2007

3.1 The Linear Oscillator

23

z(t) x(t) k(x,µ,t) F(t)

m c

Figure 3.1. A spring–mass–damper system.

where ωn is a constant (the natural frequency), and with x˙ ≡ dx/dt, we obtain the nondimensional governing equation of motion, x¨ + ω2n x = 0,

(3.4)

subject to the two initial conditions, x(0), and x(0). ˙ This is the equation of motion governing the dynamic response of the spring–mass system shown in Fig. 3.1 with √ ωn = k/m (k and m constant) and all other parameters set equal to zero, that is, c = F (t) = z(t) = 0. Because Eq. (3.4) is a linear, homogeneous, ordinary differential equation with constant coefficients, we can write the solution as x(t) = Aest .

(3.5)

Placing Eq. (3.5) into Eq. (3.4), we find that s = ±iωn , and thus the general form of the solution is given by x(t) = Aeiωn t + Be−iωn t .

(3.6)

Alternatively, using Euler’s identities, we can write x(t) = C cos(ωn t) + D sin(ωn t).

(3.7)

To determine A and B (or C and D), we make use of the initial conditions to get x(t) = x(0) cos(ωn t) +

x(0) ˙ sin(ωn t). ωn

(3.8)

We can convert this system into a pair of coupled, first-order ordinary differential equations (in state-variable format) by introducing a new variable y = x, ˙

(3.9)

and substituting in Eq. (3.4) gives x˙ = y, In matrix notation,

y˙ = −ω2n x.

x˙ 0 = y˙ −ω2n

1 0

x . y

(3.10)

(3.11)

17:50

P1: RTT Chapter-03

CUFX159-Virgin

0 521 85648 5

August 14, 2007

24

Dynamics in the Vicinity of Equilibrium

2

2

(a)

x

0

0

-2

(b)

x

0

5

10

15

t

20

-2

0

5

10

15

t

20

Figure 3.2. (a) Two trajectories exhibiting simple harmonic motion: x(0) = 1, x(0) ˙ = 0, and x(0) = 0, x(0) ˙ = 1; (b) solutions to Eq. (3.16). (i) x(0) = −1, x(0) ˙ = 1 (solid curve); (ii) x(0) = 0.0001, x(0) ˙ = 0 (dotted curve); and (iii) x(0) = −0.99, x(0) ˙ = 1 (dot–dashed curve).

A plot of Eq. (3.8) (with ωn = 1) is shown in Fig. 3.2(a) for two typical sets of initial conditions. At this point, we simply note that, from Eq. (3.8) and its derivative (to obtain velocity), we can envision the trajectory evolving with time in a repeating manner. Plotting position versus velocity (the phase plane) will be a useful way of displaying dynamic behavior later in this book, and in this (undamped) case it is apparent that the motion is described by ellipses. This is, of course, the periodic behavior we would expect for a simple spring–mass system with ωn (assumed real, i.e., ω2n > 0) identified as the natural frequency of the oscillation. In terms of a heuristic concept of stability, we might consider this behavior to be neither stable or unstable, as any motion we might initiate does not decay or grow, but simply persists. Solution (3.7) can also be written as x(t) = A¯ cos(ωn t + φ),

(3.12) √ where A¯ = C2 + D2 is the amplitude and φ = arctan (C/D) is the phase. Thus we see that the larger the initial conditions, the larger the area enclosed by the ellipses, that is, 2 ¯2. x2 (t) + [x(t)/ω ˙ n] = A

(3.13)

The two trajectories shown in Fig. 3.2(a) differ by a phase φ = π/2, and thus the dashed curve can be viewed as the corresponding velocity time series. Later, we will see how this relates to energy. However, the form of the resulting motion is independent of the initial conditions. Suppose we have ω2n < 0. This is a situation that is difficult to envision, physically, but can occur, for example, in a nonlinear system if the spring stiffness becomes negative. Then the motion is governed by x¨ − ω2n x = 0.

(3.14)

Now adopting the solution x(t) = Aest leads to s = ±ωn , and thus x(t) = Aeωn t + Be−ωn t .

(3.15)

Using the definition of hyperbolic functions and the initial conditions, we also have x(t) = x(0) cosh ωn t + [x(0)/ω ˙ n ] sinh ωn t.

(3.16)

17:50

P1: RTT Chapter-03

CUFX159-Virgin

0 521 85648 5

August 14, 2007

3.1 The Linear Oscillator

25

In this case, we do not have a periodic solution: The positive exponent indicates that typically x → ∞ as t → ∞. Hence, our heuristic concept of stability indicates that this behavior is unstable. However, we also observe that we can choose very specific initial conditions (unlikely but nevertheless important cases) in which the trajectory will end up at the origin, that is, where the positive exponential term is completely suppressed, as well as the case in which the negative exponential term in Eq. (3.15) dominates for a short time before the trajectory is swept away. These cases are illustrated in Fig. 3.2(b). For all practical purposes, that is, arbitrary initial conditions, the motion is clearly unstable. The meaning of the special trajectory will be discussed at length later. Damping. The preceding examples are somewhat unrealistic in terms of practical mechanics because they do not include energy dissipation [10]. With the inevitable presence of damping the question of stability becomes less ambiguous. Typical motion will then consist of a transient followed by some kind of recurrent long-term behavior. This brings us to the fundamentally important concept of attractors. These are the special solutions alluded to earlier, and they play a key role in organizing dynamic behavior in phase space (the space of the state variables). We shall also see that for nonlinear systems unstable solutions have an important influence on the general nature of solutions. Suppose we now allow for some energy dissipation in the form of linear viscous damping, that is, c = 0 in Fig. 3.1. The equation of motion is now

x¨ + 2ζ ωn x˙ + ω2n x = 0,

(3.17)

into which a nondimensional damping ratio, ζ ≡ c/(2mωn ), has been introduced. Solutions to this equation now depend on the value of ζ. For underdamped systems, we have ζ < 1 and solutions of the form x(0) ˙ + ζ ωn x(0) x(t) = e−ζ ωn t sin ωd t + x(0) cos ωdt , (3.18) ωd where the damped natural frequency ωd is given by ωd = ωn 1 − ζ2 .

(3.19)

A typical underdamped response (ζ = 0.1) is shown in Figs. 3.3(a) and 3.3(b) as a time series and phase portrait, respectively. The origin in Fig. 3.3(b) indicates a position of asymptotically stable equilibrium; that is, any disturbance leads to a dynamic response that moves smoothly back to equilibrium. The trajectory gradually spirals down to this rest state: We can imagine a family of trajectories forming a flow as time evolves. Because this equilibrium is unique, the whole of the phase space is the attracting set for all initial conditions and disturbances [11]. Damping in this range, e.g., ζ ≈ 0.1, is quite typical for mechanical and structural systems. For a heavily (or overdamped) system ζ > 1, and in this case the form of the solution is √

x(t) = Ae(−ζ+

ζ2 −1)ωn t

√

+ Be(−ζ−

ζ2 −1)ωn t

,

(3.20)

17:50

P1: RTT Chapter-03

CUFX159-Virgin

0 521 85648 5

August 14, 2007

26

Dynamics in the Vicinity of Equilibrium

(a)

x(t) . x(t)

. 1.5 (b) x/ω n

x

.

1

x

0.5 0 −0.5 −

−1 −

−1.5 −1.5

t

x(t)1.2 . x(t) 1 (c)

−0.5

0

0.5

1

x

1.5

. 1.2 x/ω n (d) 1

.

x

0.8

−1

0.6

0.8

x

0.6

0.4

0.4

0.2

0.2

0

0

− 0.2

0

2

4

6

8

10

t

12

− 0.2 − 0.05

0

0.05

0.1

0.15

0.2

0.25

x

0.3

Figure 3.3. (a) Time series and (b) phase portraits for underdamped (oscillatory) motion, x(0) = 1.0, x(0) ˙ = 0.0, ζ = 0.1; (c) and (d), overdamped (nonoscillitory) motion, x(0) = 0.0, x(0) ˙ = 1.0, ζ = 1.5.

where A=

√ ζ2 − 1)ωn x(0) , √ 2ωn ζ2 − 1

x(0) ˙ + (ζ+

(3.21)

and √ −x(0) ˙ − (ζ− ζ2 − 1)ωn x(0) . B= √ 2ωn ζ2 − 1

(3.22)

The motion is a generally monotonically decreasing function of time and may take a relatively long time to overcome rather heavy damping forces on the way to equilibrium. A typical case is also shown in Figs. 3.3(c) and 3.3(d). The boundary between these two cases is the critically damped case, i.e., ζ = 1. In the context of this book, we will regularly encounter the situation in which the stiffness of a system degrades, and given the definition of ζ we expect not only a reduction in the natural frequency but also an increase in the damping ratio. Returning to the state-variable-matrix format of the linear oscillator, we therefore have x˙ 0 1 x = . (3.23) y˙ −ω2n −2ζωn y

17:50

P1: RTT Chapter-03

CUFX159-Virgin

0 521 85648 5

August 14, 2007

3.1 The Linear Oscillator

27

We can also write the solution in terms of the eigenvalues of the state matrix, that is, the roots of the characteristic equation λ2 + 2ζωn λ + ω2n = 0.

(3.24)

Critical damping thus relates to the discriminant’s being equal to zero. Given the scenario of a system losing stability, we can usefully view all the response possibilities of this type of linear system according to the location of the roots in the complex plane. For example, having two complex roots with negative real parts corresponds to an exponentially decaying oscillation. Summarizing these outcomes leads to Fig. 3.4 [12]. In general, we will have a system with positive stiffness and damping and thus a root structure corresponding to the upper-right-hand quadrant. Critical damping corresponds to the parabola, and phase portraits and eigenvalues are indicated for various combinations of the natural frequency and damping. The system eigenvectors organize the transient behavior in the phase portrait. Some useful terminology here includes the spiral or focus for decaying oscillatory motion (also called a sink), the node for overdamped motion, the inflected node for equal eigenvalues (and thus including the critically damped case), and the saddle for the motion characterized by having both a stable and unstable direction I R

2

n

STABLE

2 n

Figure 3.4. Phase portraits and root structure of a linear oscillator (after Thompson [12]).

17:50

P1: RTT Chapter-03

CUFX159-Virgin

0 521 85648 5

August 14, 2007

28

Dynamics in the Vicinity of Equilibrium

(eigenvector) with instability becoming dominant. We can also view the undamped case as a center. We shall focus more on higher-order dynamical systems in the next chapter, but more formally we state (see Jordan and Smith [13]) that if we have a dynamical system x˙ = Ax, where A is constant with eigenvalues λi , i = 1, 2, ..., n, then (1) if the system is stable, Re{λi } ≤ 0, i = 1, 2, . . . , n; (2) if either Re{λi } < 0, i = 1, 2, . . . , n, or if Re{λi } ≤ 0, i = 1, 2, . . . , n and there is no repeated eigenvalue, the system is uniformly stable; (3) the system is asymptotically stable if and only if Re{λi } < 0, i = 1, 2, . . . , n [and then it is also uniformly stable, by (2)]; (4) if Re{λi } > 0 for any i, the solution is unstable. We thus observe what will typically happen when the stiffness of the system degrades (e.g., because of an axial load acting on a slender structure), and this is indicated by the large arrow in Fig. 3.4. For a small amount of damping, the eigenvalues start off as a complex-conjugate pair with negative real parts. As the stiffness (and hence natural frequency) reduces, the eigenvalues merge on the negative real axis, and then their magnitudes diverge, with one entering the positive half-plane. Thus instability occurs, and solutions grow without bound. Although the preceding description relates to a single-degree-of-freedom (SDOF) linear oscillator, this type of scenario is encountered to a large extent within a variety of high-order and nonlinear systems. The geometric view afforded by a consideration of the root structure and phase portraits of families of solutions about equilibrium points is very useful. We will make extensive use of linearization to utilize this view locally to equilibrium within a nonlinear context [14].

3.2 Oscillator with a Slow Sweep of Frequency Consider again the spring–mass system shown in Fig. 3.1. Again assume that there is no damping or external forcing (c = F (t) = z(t) = 0), and that the spring stiffness decays linearly (in time) from a base value k = 1 at t = 0. We assume this decay is very slow, and characterized by a small parameter . In this case we can write the governing equation of motion as x¨ + µ2 (t)x = 0,

(3.25)

in which µ2 (t) =

k (1 − t), m

(3.26)

that is, the system will lose stability when t → 1/. If we assume that the evolution of the stiffness change is very slow ( 0 such that if r(t0 ) < δ then r(t0 ) < for all t ≥ t0 . (2) The equilibrium point xe is asymptotically stable if it is Lyapunov stable and limt→∞ r(t) = 0. (3) The equilibrium point xe is unstable if there is an a > 0 such that for arbitrarily small δ > 0 there is a motion xa (t) for which ra (t0 ) < δ and ra (t1 ) > at some time t1 > t0 . Thus we see that with these definitions instability occurs when the perturbed trajectory reaches the sphere of radius in finite time. We can view this situation graphically as shown in Fig. 4.1. Clearly, for conservative systems we expect Lyapunov definition (1) to apply [5]. In typical structural systems, we would expect

17:51

P1: KAE Chapter˙04

CUFX159-Virgin

0 521 85648 5

August 14, 2007

4.2 Multiple-Degree-of-Freedom Systems

51

some damping to be present, and assuming it is of the form of linear viscous damping (derivable from Rayleigh’s dissipation function) we might expect asymptotic stability to represent typical dynamical behavior for an autonomous structural system. Hence, after a small disturbance, a transient slowly decays back to the rest position. The presence of damping can considerably complicate the dynamic analysis of MDOF systems, although the adoption of proportional damping, as mentioned earlier, facilitates the modal analysis approach [6, 7]. We can return to some of the responses of the SDOF system to see how this more general definition encompasses the simple cases. In Figure 3.4, we showed how the response of a linear oscillator depended on the stiffness and damping. Thus we see the upper-right quadrant as corresponding to areas of aysmptotic stability. Since the primary focus of this book is axially loaded structures we will repeatedly encounter a decay in stiffness and thus a transition to instability (crossing the y axis). Although a decay in damping leads to instability as well, this type of flutter instability (usually initiated via a Hopf bifurcation [8]) is often associated with nonconservative forces (e.g., in aeroelasticity [9]) and is not a feature central to the dynamics of axially loaded structures of the type considered in this book, although one or two examples are given. Considering disturbances about an equilibrium state, we write x(t) = xe + η(t),

(4.30)

which can be substituted back into Eq. (4.27) to give d [xe (t) + η(t)] = F[xe (t) + η(t), t], dt

(4.31)

d xe (t) = F[xe (t), t], dt

(4.32)

and because

we can write d η(t) = F[xe (t) + η(t), t] − F[xe (t), t]. (4.33) dt This is the variational equation, and it plays a crucial role in determining the stability associated with certain solutions of dynamical systems. We can expand the right-hand side of Eq. (4.33) by using F[xe (t) + η(t), t] = F[xe (t), t] + DF(xe , t)η(t) + · · · +,

(4.34)

where DF(xe , t) is the matrix of first derivatives (the Jacobian) of F(xe , t) evaluated at xe : ⎤ ⎡ ∂F1 /∂x1 ∂F1 /∂x2 · · · ∂F1 /∂xN ⎢ ∂F2 /∂x1 ∂F2 /∂x2 · · · ∂F2 /∂xN ⎥ ⎥ ⎢ ⎢ . .⎥ ⎥ ⎢ . (4.35) DF(xe , t) = ⎢ ⎥ ⎢ . .⎥ ⎥ ⎢ ⎣ . .⎦ ∂FN /∂x1 · · · ∂FN /∂xN x=x e

17:51

P1: KAE Chapter˙04

CUFX159-Virgin

0 521 85648 5

August 14, 2007

52

Higher-Order Systems

Now, assuming the perturbations are small, we can truncate the higher-order terms in Eq. (4.34) and thus we can write Eq. (4.33) as dη = DF(xe , t)η(t). dt

(4.36)

This is called the linear variational equation for the system. In general, the coefficients of the Jacobian are time varying, and thus we have a non-autonomous system. We shall pay specific attention to systems in which the coefficients are periodic in Chapter 14, as this is a case commonly encountered in forced vibrations. For now, we focus on the stability of equilibrium, that is, xe (t) = xe represents a fixed point in phase space, and the coefficients in the Jacobian matrix are constant, that is, an autonomous system. Thus the focus on linear systems is justified on the grounds that nonlinear systems can often be linearized in the vicinity of equilibrium. By restricting ourselves to small neighborhoods about an equilibrium point, we can make use of much of the preceding linear theory. In terms of energy, and within a free-vibration context, we thus consider the behavior resulting from a small disturbance from an equilibrium point, q(t) = q0 + δ(t),

˙ ˙ = δ(t), q(t)

(4.37)

where δ and δ˙ are perturbation vectors. Because the nonlinearity will typically appear in the generalized positions (and assuming a smooth nonlinearity), we expand the potential energy as a Taylor series: ∂U 1 U(q1 , q2 , . . . , qn ) = U(q0 ) + δT + δT Kδ, (4.38) ∂q q=q0 2 where higher-order terms have been neglected. K is the stiffness matrix (evaluated at equilibrium). The first term on the right-hand side is an arbitrary constant scalar (which we can choose to be zero), the second term is equal to zero by virtue of the equilibrium condition, and hence we can follow the earlier theory to arrive at the linearized equations of motion: Mδ¨ + Kδ = 0,

(4.39)

where K is related to the change in potential energy about an equilibrium position and is generally a function of a set of external (axial) loads, and M is a mass matrix. Now suppose we revisit the stability of a set of ordinary differential equations in terms of the phase space. That is, we solve the algebraic eigenvalue problem arising from the equations of motion in first-order state-variable format. We also add a little damping (with a matrix of damping coefficients C) and introduce the perturba˙ T ], so that the set of equations (4.5) can be written tion state-vector as x(t) = [qT q(t) (following Meirovitch [2]) as x(t) ˙ = Ax(t),

(4.40)

17:51

P1: KAE Chapter˙04

CUFX159-Virgin

0 521 85648 5

August 14, 2007

4.2 Multiple-Degree-of-Freedom Systems

where

A=

53

I . −M−1 C

0 −M−1 K

(4.41)

We see that this is the matrix generalization of Eq. (3.11). Assuming an exponential form for the solution x(t) = eλt x

(4.42)

and placing in Eq. (4.40) (canceling eλt ), we get Ax = λx.

(4.43)

This is again the algebraic eigenvalue problem. A set of n second-order ordinary differential equations can always be converted to a set of 2n first-order ordinary differential equations. The general solution is given by x(t) =

2n

cj eλj t xj ,

(4.44)

j =1

where the coefficients cj are generally complex and depend on the initial conditions. Specifically, the eigenvalues, which can be written as λj = αj + iβj ,

j = 1, 2, . . . , 2n,

(4.45)

govern both the frequency and stability of the response by means of their imaginary and real parts, respectively. With positive stiffness, and in the absence of damping, we expect purely imaginary eigenvalues, that is, αj = 0 (which is a stable situation according to part 2 of the Lyapunov stability definition). We can think of eigenvalues distributed along the imaginary axis in the complex plane (akin to the positive x axis in Fig. 3.4). With the addition of damping, we expect to have αj < 0 and asymptotic stability (including both the focus and node illustrated in Fig. 3.4, which depend on the level of damping). Analyzing vibration problems by solving a set of secondorder systems is equivalent to converting and then solving a set of (twice as many) first-order systems. The eigenvalues of the two approaches are linked by Eq. (4.19). Although the second-order approach allows a more physical feel (in terms of natural frequencies and damping ratios), the first-order approach is popular in control theory, is well suited to systems in which coupled modes are present, is an approach very well suited to software packages like MATLAB [10], and, as we have seen, allows a more direct assessment of stability. We shall often encounter a situation in which axial loading reduces the lateral stiffness of a structure such that the potential energy can be written as 1 (aij − Pbij )qi qj , 2 n

V = U−W =

n

(4.46)

i=1 j =1

where the strain energy U (and matrix aij ) are positive definite and the work done by the axial loads (and bij ) are either indefinite or positive definite (semi-definite).

17:51

P1: KAE Chapter˙04

CUFX159-Virgin

0 521 85648 5

August 14, 2007

54

Higher-Order Systems

Placing this in the context of a set of lightly damped oscillators (and assuming a single load for simplicity) we then have Mq¨ + Cq˙ + (KL − PKG )q = 0,

(4.47)

where C is the damping matrix and KG is a matrix relating to the (prestressing) effect of the axial load. We typically start with P = 0 and thus have initial stability, and then a major question is to determine the specific (or critical) values of P that cause instability, and how this is reflected in dynamic behavior. 4.2.4 Routh–Hurwitz Criterion In the previous subsection, we saw how the stability of a system depended on the sign of the real parts of the eigenvalues [Eq. (4.45)]. To determine these signs, we need not solve the complete eigenvalue problem. Rather, a technique, referred to as the Routh–Hurwitz criterion, can be used [11]. Writing the characteristic equation in the form λn + a1 λn−1 + a2 λn−2 + · · · + an = 0,

(4.48)

it can be shown that the system is asymptotically stable (i.e., the real parts of the eigenvalues are negative) if and only if the principal minors of the n × n matrix ⎡ a1 ⎢ ⎢a3 ⎢ ⎢a ⎢ 5 ⎢ ⎢. ⎢ ⎢ ⎢. ⎢ ⎢ ⎣. .

1

0

0

···

a2

a1

1

···

a4

a3

a2

a1

.

.

0

1 ···

⎤

⎥ 0⎥ ⎥ 0⎥ ⎥ ⎥ . ⎥ ⎥ ⎥ . ⎥ ⎥ ⎥ . ⎦

(4.49)

an

.

are all positive, that is, 1 = a1 > 0, a1 1 > 0, 2 = a3 a2 a1 1 0 3 = a3 a2 a1 > 0, a a a 5

4

(4.50)

3

and so on. A necessary condition for asymptotic stability is that all the a’s be positive. Computationally, this criterion may be more convenient to evaluate than the full determinant.

17:51

P1: KAE Chapter˙04

CUFX159-Virgin

0 521 85648 5

August 14, 2007

4.2 Multiple-Degree-of-Freedom Systems

55

4.2.5 Lyapunov Functions As seen in the previous subsection, we can draw important conclusions about the stability of a dynamical system without obtaining a complete solution to the equations of motion; computing the Lyapunov stability of a system directly may not be the most straightforward approach. Also, much of the preceding stability theory was based on linear behavior (in the vicinity of equilibrium). Nonlinearity and global stability can be assessed by use of the concept of Lyapunov functions [7, 12, 13]. This is sometimes referred to as Lyapunov’s direct, or second, method [7], and it has been found to be especially useful in applications involving nonconservative forces and in control theory [14]. The concept is again based on assessing stability without solving the equations of motion and is closely related to energy for conservative systems. We introduce the Lyapunov function E(x), which we assume is a continuous and differentiable scalar function of the vector x(t) and satisfies the following conditions: (1) E(x) > 0, for all values of x(t) = 0, ˙ (2) E(x) ≤ 0, for all values of x(t) = 0. Now, if it can be shown that a Lyapunov function exists for a system, then that system is stable and is aymptotically stable if the time derivative of E is less than zero. Finding a Lyapunov function is not necessarily easy, and even if one cannot be found it does not imply that the system is unstable. The total energy of a system is often a useful place to start when searching for a Lyapunov function. With this in mind we take a brief look at our general damped MDOF system: Mq¨ + Cq˙ + Kq = 0,

(4.51)

and assuming the mass, damping, and stiffness matrices are all symmetric and positive-definite, we can confirm the asymptotic stability of the system. Let a Lyapunov function be based on the total energy for this system as ˙ E[q(t), q(t)] =

1 T ˙ + qT (t)Kq(t)]. [q˙ (t)Mq(t) 2

(4.52)

Clearly, this satisfies the first condition of Lyapunov’s direct method, that is, E(q) > 0. For the second part, we have ˙ ˙ E[q(t) q(t)] = q˙ T Mq¨ + q˙ T Kq,

(4.53)

which, by premultiplying Eq. (4.51) by q˙ T (t), we obtain ˙ ˙ E[q(t) q(t)] = −q˙ T Cq˙ ≤ 0,

(4.54)

where C is positive-definite. That a damped spring–mass system is stable is no surprise, but for highly nonlinear and nonconservative problems this is often a convenient approach.

17:51

P1: KAE Chapter˙04

CUFX159-Virgin

0 521 85648 5

August 14, 2007

56

Higher-Order Systems

4.2.6 Rayleigh’s Quotient The previous subsections showed how it may be possible to estimate the stability of a system without having to solve the equations of motion. In a similar vein, it is possible to estimate the natural frequencies of a system (and hence stability) based on an approach developed by Lord Rayleigh for eigenvalue problems [15]. We used this approach in Chapter 3 for a SDOF system, but extend it here to MDOF systems, where it finds most utility in estimating the lowest natural frequency. Given the standard eigenvalue problem, Avr = λr vr ,

r = 1, 2, . . . , n,

(4.55)

where A is a real symmetric positive-definite n × n matrix, we can express the eigenvalues in the form λr =

vTr Avr , vTr vr

r = 1, 2, . . . , n,

(4.56)

in which the eigenvalues are ordered λ1 ≤ λ2 ≤ · · · λn . The Rayleigh quotient is λ(v) = R(v) =

vT Av . vT v

(4.57)

Again for a MDOF system [Eq. (4.5)] we have the eigenvalue problem Ku = λMu,

(4.58)

and thus Rayleigh’s quotient in this case is R(λ, u) =

uT Ku . uT Mu

(4.59)

We note that the numerator is the maximum potential energy and the denominator is sometimes referred to as the zero-frequency kinetic energy during an oscillation [16]. What makes this approach especially useful in a practical context is that the assumed vector need not be terribly accurate to produce a reasonable eigenvalue estimate. This is partly due to the fact that it is stationary when perturbed around any of the actual system eigenvectors. It can be shown that this corresponds to an upper bound (and hence if an alternative vector is chosen that gives a eigenvalue lower than the previous one, this is necessarily a better estimate). This approach will be generalized in the next section, but suffice to say here that if the chosen vectors u satisfy certain boundary conditions and resemble the actual mode shapes, then estimates of the lowest natural frequency (usually the most important) are often reasonably accurate. A few other issues to consider before moving on to continuous systems are worth mentioning at this point. The definitions of stability introduced here are largely based (or at least specialized) for conservative systems (with the additon of a little damping) and the question of stability of an equilibrium position. Toward the end of this book we will focus on forced vibration problems, in which the stability of periodic behavior will need to be considered. In this

17:51

P1: KAE Chapter˙04

CUFX159-Virgin

0 521 85648 5

August 14, 2007

4.3 Distributed Systems

57

case we will need to extend the concepts of stability to include orbital stability. Another aspect of the general eigenvalue problem is how the eigenvalues change as a system parameter changes. Of course, in the types of problems encountered in this book we will often have the case that the stiffness matrix degrades in some sense, but there are a number of instances (including in control theory) for which a minor perturbation of the eigenvalue problem is a key issue. Finally, computational aspects of the algebraic eigenvalue problem are cornerstones of efficient algorithms in software applications [17]. This will also be touched on later.

4.3 Distributed Systems We have gone from SDOF to MDOF systems and now this is extended further to continuous systems, which can be thought of as having an infinite number of DOFs. Much of the preceding theory can be extended to these distributed systems, and indeed many methods have been developed to reduce the number of DOFs [2]. We start this section by returning to Hamilton’s principle in order to derive Lagrange’s equations and hence the equations of motion. In contrast to finitedimensional systems we will see that this spatially continuous analysis results in a boundary-value problem characterized by a partial differential equation with appropriate boundary conditions. We focus our attention on beamlike structures characterized by deflection w(x, t) and signify the boundary conditions at 0 and L over the 1D domain x. Hamilton’s principle for conservative systems can be written (following Meirovitch [2]) as

t2

δLdt = 0,

δw(t = t1 ) = δw(t = t2 ) = 0,

(4.60)

t1

where L = T − V, and

L

T=

ˆ w) T( ˙ dx,

(4.61)

ˆ V(w, w , w ) dx.

(4.62)

0 L

V= 0

The hats on Tˆ and Vˆ imply an energy density, an overdot is a time derivative, a prime is a spatial derivative, and we assume that the potential energy in this case may be a function of the second derivative of the displacement w, which will be the case for the strain energy of systems in bending, for example. We need to take the variation of the Lagrangian in Eq. (4.60) and to do this (in terms of a Lagrangian density) we have ∂Lˆ ∂Lˆ ∂Lˆ ∂Lˆ δLˆ = δw + δw + δw + δw, ˙ ∂w ∂w ∂w ∂w˙

(4.63)

17:51

P1: KAE Chapter˙04

CUFX159-Virgin

0 521 85648 5

August 14, 2007

58

Higher-Order Systems

and thus, evaluating the integrals and placing in Eq. (4.60) we get

∂Lˆ ∂2 ∂Lˆ ∂ ∂Lˆ ∂ ∂Lˆ + − δwdx − ∂w ∂x ∂w ∂x2 ∂w ∂t ∂w˙ 0 t1 ˆ ∂L ∂ ∂Lˆ δw(0, t) − − ∂w ∂x ∂w x=0 ˆ ∂L ∂ ∂Lˆ δw(L, t) + − ∂w ∂x ∂w x=L ∂Lˆ ∂Lˆ − δw (0, t) + δw (L, t) dt = 0. ∂w x=0 ∂w x=L t2

L

We next choose the virtual displacements such that the variations at (0, t) and (L, t) are zero and then we are left with ∂Lˆ ∂2 ∂Lˆ ∂ ∂Lˆ ∂ ∂Lˆ + − − = 0. ∂w ∂x ∂w ∂x2 ∂w ∂t ∂w˙

(4.64)

This is called the Lagrange differential equation of motion. It is a partial differential equation with possible boundary conditions at x = 0, ˆ ∂L ∂ ∂Lˆ − − = 0, ∂w ∂x ∂w x=0 w(0, t) = 0, ∂Lˆ − = 0, ∂w x=0 w (0, t) = 0,

(4.65) (4.66) (4.67) (4.68)

and at x = L,

∂Lˆ ∂ ∂Lˆ − = 0, ∂w ∂x ∂w x=L w(L, t) = 0, ∂Lˆ = 0, ∂w x=L w (L, t) = 0.

(4.69) (4.70) (4.71) (4.72)

The system must satisfy two boundary conditions at each end (assuming the Lagrangian depends on the second spatial derivative of the displacement). Given homogeneous boundary conditions we can simplify Eq. (4.64) to the form of Eq. (2.36). Specific examples will be given in later chapters in which this type of boundary-value problem is solved by use of a variety of techniques.

17:51

P1: KAE Chapter˙04

CUFX159-Virgin

0 521 85648 5

August 14, 2007

4.3 Distributed Systems

59

4.3.1 The Differential Eigenvalue Problem It will often be the case that the partial differential of motion can be subject to the separation of variables (temporal and spatial), that is, by assuming the motion is a sum of terms of the form w(x, t) = F (t)W(x).

(4.73)

This allows for the consideration of ordinary differential equations. The temporal part of the solution will typically consist of harmonic motion (certainly for stable conservative systems) satisfying the initial conditions (see Chapter 2). The spatial part of the solution, together with the appropriate boundary conditions, constitutes a differential eigenvalue problem. The actual form of the partial differential equation depends of course on the specific physics of the problem (e.g., we have already developed the case typically encountered for Euler–Bernoulli beams) including the boundary conditions. However, there are a few general conclusions that we can draw by using the concept of differential operators [18]. The general form is given by 2 ∂ w(x, t) K[w(x, t)] + M = 0, (4.74) ∂t2 where K and M are (linear) homogeneous differential operators defined (for example) by K = a1 + a2

∂ ∂ ∂2 ∂2 + a3 + a4 2 + a5 + ··· + . ∂x ∂y ∂x ∂x∂y

(4.75)

M is very often a simple constant, and the spatial vector x is defined over a domain D. The form of the operator (i.e., the values of the coefficients) depends on the specific physical problem, for example, the general preceding form is appropriate for two-dimensional (2D) problems, including plates. Assuming harmonic motion we obtain the eigenvalue problem K[W(x)] = λM[W(x)],

(4.76)

where we recognize λ = ω2 , and the boundary conditions Bµ [W(x)] = 0,

(4.77)

where Bµ is also a linear differential operator. We expect the trivial solution W(x) = 0 to be present, but it is the values of the parameter λ (associated with the natural frequencies of the system) that gives rise to nontrivial solutions [W(x) = 0 and satisfying the boundary conditions] that are of central importance. There are infinitely many of these values, and they are the eigenvalues of the system and the corresponding W(x) are eigenfunctions. We observe certain similarities between the algebraic eigenvalue problem and the differential eigenvalue problem with the operators K and M having their relation with the stiffness and mass matrices of MDOF systems.

17:51

P1: KAE Chapter˙04

CUFX159-Virgin

0 521 85648 5

August 14, 2007

60

Higher-Order Systems

The boundary conditions did not appear explicitly in the algebraic eigenvalue problem for MDOF (matrix) systems. Essentially, they were contained in the matrices themselves, but for continuous systems they are a very important part of the problem and influence the functions W(x). Eigenfunctions satisfy both the differential equation and the boundary conditions exactly. Comparison (test) and admissible functions satisfy certain of the boundary conditions, that is, both geometric and natural, and geometric, respectively. They also form the basis of a number of solution methods to be considered later. Differential operators may also be selfadjoint (typically for linear conservative systems), which is analogous to symmetric matrices in the algebraic eigenvalue problem [18]. Similarly the eigenvalue problem is said to be positive-definite if all its eigenvalues are greater than zero and the eigenfunctions exhibit orthogonality (to be discussed shortly).

4.3.2 Solution Methods Modal Analysis and Truncation

For some differential eigenvalue problems, a closed-form solution may be available, and this will often be the case for linear systems with relatively simple geometry. The expansion theorem [2] allows the solution of a boundary-value problem by transforming it into an infinite set of ordinary differential equations—modal equations. Again we focus attention on self-adjoint systems, as well as on systems in which the eigenvalues do not depend on the boundary conditions. Earlier, we saw how it was possible to decouple the equations of motion for a MDOF system by transforming the problem into modal coordinates. This idea carries through to infinite-dimensional systems [19]. Using operator notation, we typically have the system 2 ∂w(x, t) ∂ w(x, t) K[w(x, t)] + C = 0. (4.78) +M ∂t ∂t2 Assuming harmonic motion and without damping, we get the eigenvalue problem K[W(x)] = ω2 M[W(x)],

(4.79)

with the boundary conditions Bµ [W(x)] = 0,

µ = 1, 2, . . . , p,

(4.80)

that is, assuming that the boundary conditions are homogeneous (every term involves W) and that the differential expression is of the order of 2p [2]. Let us consider two distinct solutions of the eigenvalue problem: (λr , wr ) and (λs , ws ). Because of the self-adjoint nature of the system, wr Mws dD = ws Mwr dD, (4.81) D

D

wr Kws dD = D

ws Kwr dD D

(4.82)

17:51

P1: KAE Chapter˙04

CUFX159-Virgin

0 521 85648 5

August 14, 2007

4.3 Distributed Systems

61

we have the conditions of orthogonality of the eigenfunctions: wr Mws dD = 0 for ω2r = ω2s ,

(4.83)

D

D

wr Kws dD = 0 for ω2r = ω2s .

(4.84)

The magnitude of the eigenvectors is arbitrary, and thus it makes sense to normalize them in a consistent way. This can be achieved with wr Mws dD = δrs , r, s = 1, 2, . . . , (4.85) D

D

wr Kws dD = δrs ω2r , r, s = 1, 2, . . . ,

(4.86)

where δrs is the Kronecker delta. The expansion theorem [20] tells us that the system response is given by a linear combination of the eigenfunctions: w(x, t) =

∞

ar (t)wr (x).

(4.87)

r=1

We furthermore assume that proportional damping is now included, that is, wr Cws dD = δrs 2ζr ωr , r, s = 1, 2, . . . , (4.88) D

where ζr contains the modal damping ratios. Placing Eq. (4.87) into Eq. (4.78) leads to

∞

∞ ∞ ∂ ∂2 K wr (x)ar (t) + C wr (x)ar (t) + M wr (x)ar (t) = 0. ∂t ∂t2 r=1 r=1 r=1 (4.89) Multiplying by wr (x) and integrating over the domain results in ∞ r=1

ar (t)δrs ω2r +

∞

a˙ r (t)δrs 2ζr ωr +

r=1

∞

a¨ r (t)δrs = 0.

(4.90)

r=1

But orthogonality tells us this equation holds only if r = s, and thus a¨ r (t) + 2ζr ωr a˙ r (t) + ω2r ar (t) = 0,

r = 1, 2, . . . .

(4.91)

Thus we arrive at an infinite set of (uncoupled) ordinary differential equations in terms of normal modes, assuming the diagonal form exists. This form is somewhat familiar from the consideration of principal coordinates in MDOF systems. Because, in practical situations, it is only the lowest few modes that are important (the higher modes typically do not contribute very much to the solution) it may be possible to use a truncated modal model. The following sections introduce a couple of popular (approximate) approaches to solving boundary-value problems that are developed from the concept of basis functions.

17:51

P1: KAE Chapter˙04

CUFX159-Virgin

0 521 85648 5

August 14, 2007

62

Higher-Order Systems Rayleigh–Ritz

For continuous systems we can write Rayleigh’s quotient (see Subsection 4.2.6) as w(x)K[w(x)]dD 2 , (4.92) ω = R[w(x)] = D D w(x)M[w(x)]dD which is stationary in the class of kinematically admissible functions, at the eigenvalues. This is a very useful property for estimating especially the lowest eigenvalue of a continuous system [21]. The Rayleigh–Ritz method consists of replacing the differential eigenvalue problem with a set of algebraic eigenvalue problems and is based on assuming a solution of the form w(x) =

N

an (t)θn (x).

(4.93)

n=1

If the functions θn (x) satisfy both the differential equation and the boundary conditions exactly, then we would have the eigenvalues, but in general we will be faced with choosing these as trial functions. Clearly, a trial function that satisfies all the boundary conditions (both geometric, involving displacements/slopes, and natural, involving forces/moments) is desirable. These are generally called comparison or test functions. However, in general it will be easier to come up with trial functions that satisfy only the geometric boundary conditions. These are typically called admissible functions and tend to result in quite accurate estimates for the eigenvalues, especially as more are taken in the assumed solution, and because the approach yields an upper bound, we know that the solution will tend to approach the exact answer from above. This is very similar to the case for MDOF systems whereas now we seek trial functions rather than trial vectors. We also note the relation with Rayleigh’s energy method used in the previous chapter for a SDOF system. Rayleigh’s principle states that the Rayleigh quotient has a minimum value, which is the square of the lowest frequency of vibration, in the neighborhood of the fundamental mode: ω21 = min R(w) = min

Vmax , Tref

(4.94)

where Tref is the reference kinetic energy (referred to as the zero-frequency kinetic energy in Chapter 3). The Rayleigh–Ritz method is also closely related to the method of assumed modes [1]. Weighted Residuals–Galerkin

An alternative approach to obtaining approximate solutions to distributed parameter systems is based on the method of weighted residuals, the best-known technique being the Galerkin method [22]. This approach is not restricted to self-adjoint systems and generally requires the use of comparison functions rather than admissible functions, but when used for self-adjoint systems is equivalent to the Rayleigh–Ritz method. The central idea is that the error between an approximate solution and the exact solution is minimized.

17:51

P1: KAE Chapter˙04

CUFX159-Virgin

0 521 85648 5

August 14, 2007

4.3 Distributed Systems

63

Given a linear differential operator L, we assume the solution to the differential equation L[w(x)] = 0 (with homogeneous boundary conditions) is based on w (x, t) = (r)

N

ar (t)φr (x),

(4.95)

r=1

where the functions φr (x) are linearly independent, form a complete set, satisfy the boundary conditions, and have unknown coefficients ar (t). For an infinite set of functions, we get the exact solution, with the requirement that the function L[w(r) (x, t)] be orthogonal to all the functions φr (x). However, because we have N functions rather than an infinite set, we know that Eq. (4.95) will not typically satisfy the differential equation exactly, and we will get a remainder (or residual) that is orthogonal to the space of the chosen functions: L w(r) (x, t) φs (x)dD = 0, r, s = 1, 2, . . . , N. (4.96) D

Thus, suppose we have again a typical (self-adjoint with homogeneous boundary conditions) system given by Eq. (4.78). Placing Eq. (4.95) in Eq. (4.78), multiplying by φs (x), and integrating over the whole domain, we get Ma(t) ¨ + C˙a(t) + Ka(t) = 0, where M is an N × N mass matrix of coefficients msr given by msr = φs (x)M[φr (x)]dD, r, s = 1, 2, . . . , N,

(4.97)

(4.98)

D

and where the stiffness K and (proportionally damped) C can be described similarly. As mentioned previously, the Rayleigh–Ritz and Galerkin methods produce identical results for conservative systems. Other techniques based on weighted residuals include collocation and least squares. Furthermore, a more sophisticated approach based on Lyapunov-Schmidt reduction can be used as an alternative, especially in certain situations [23]. Specific techniques have also been developed for bifurcation and nonlinear eigenvalue problems [24]. The Finite-Element Method

Perhaps the technique with the greatest utility for solving continuous structural systems is the finite-element (FE) method (FEM) [25–29]. This is a huge subject, and obviously only a brief introduction is given here. In fact, FE analysis (FEA) will be outlined in more detail for specific structural forms later in this book (e.g., plane frames and plates). It is especially powerful for complicated structures and forms the basis of a vast array of commercially available software. Here, a brief introduction is given within the context of general boundary-value problems. This technique really comes into its own for geometrically complex structures, because (unlike the Rayleigh–Ritz method) trial functions (often relatively loworder polynomials) are defined over subdomains of a structure in a process of

17:51

P1: KAE Chapter˙04

CUFX159-Virgin

0 521 85648 5

64

August 14, 2007

Higher-Order Systems

discretization. Certain continuity conditions are enforced between subdomains at nodes, and a process of assembly brings all the elements together to form stiffness and mass matrices in standard vibration problems. From this point, the lumpedparameter model of the structure can be handled as a MDOF system, albeit with typically very high-dimensional matrices.

4.3.3 Context Revisited Before we leave this chapter and embark on the study of slender, axially loaded structures, a few comments are in order about more specific issues covered in the rest of this book. We are primarily interested in structural systems characterized according to the following scheme. r Strain energy. This a quantity associated with structural deformation. In linear theory, this has a quadratic form (small elastic deformations) and leads to a symmetric stiffness matrix, or stiffness operator. For zero deformation we typically have the trivial equilibrium state. For large deflections, both stretching (membrane) and bending effects contribute to strain energy. r Work done by axial loads. External loads are often imparted to a slender structure axially, as well as laterally, and may be due to dead (gravity) loading, for example. They generally lead to a geometric stiffness matrix that tends to diminish lateral (bending) stiffness as a function of the loading. If compressive, then they may lead to instability (the stiffness matrix becomes singular) and the appearance of nontrivial equilibria. If time-periodic axial loads are present, then parametric resonance is possible. r Kinetic energy. Again this is typically a quadratic form, but does not affect equilibrium. It may be imparted to a system in the form of sudden loading. Vibration can be considered as an oscillatory exhange of kinetic and potential energy. r Damping. Small levels of damping can be modeled by Rayleigh’s dissipation function. Also, it does not affect equilibrium. Often it turns the stability of conservative systems into asymptotically stable systems. Damping is always present in mechanical systems, although its effect can sometimes be neglected. r External forcing. Many structures are subject to excitation, which is often periodic. The resulting problem of resonance provides an important motivation for study. Loading is sometimes suddenly applied and may cause instability in the large, even though the local equilibrium is stable. External forcing increases the phase space because governing equations become inhomogenous (nonautonomous) and may lead to a wider spectrum of behavior, especially for nonlinear systems.

References [1] [2]

D.J. Inman. Engineering Vibration. Prentice Hall, 2000. L. Meirovitch. Principles and Techniques of Vibrations. Prentice Hall, 1997.

17:51

P1: KAE Chapter˙04

CUFX159-Virgin

0 521 85648 5

August 14, 2007

References [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20] [21] [22] [23] [24]

[25] [26] [27] [28] [29]

Y.G. Panovko and I.I. Gubanova. Stability and Oscillations of Elastic Systems – Paradoxes, Fallacies and New Concepts. Consultants Bureau, New York, 1965. A.M. Lyapunov. The General Problem of the Stability of Motion. Princeton University Press, 1947. A.P. Seyranian and A.A. Mailybaev. Multiparameter Stability Theory with Mechanical Applications. World Scientific, 2003. K. Huseyin. Nonlinear Theory of Elastic Stability. Noordhoff, 1975. H.H.E. Leipholz. Stability Theory. Wiley, 1987. S.H. Strogatz. Nonlinear Dynamics and Chaos. Addison-Wesley, 1994. E.H. Dowell. Aeroelasticity of Plates and Shells. Noordhoff, 1975. MATLAB. User’s guide. Technical report, The Math Works, 1989. C. Hayashi. Nonlinear Oscillations in Physical Systems. Princeton University Press, 1964. J. LaSalle and S. Lefschetz. Stability by Liapunov’s Direct Method with Applications. Academic, 1961. A.M. Lyapunov. Stability of Motion (Collected Papers). Academic, 1966. K. Ogata. System Dynamics. Prentice Hall, 1998. Lord Rayleigh (John William Strutt). The Theory of Sound. Dover, 1945. B.H. Tongue. Principles of Vibration. Oxford University Press, 1996. F.W. Williams and W.H. Wittrick. Exact buckling and frequency calculations surveyed. Journal of Structural Engineering, 109:169–87, 1983. R. Courant and D. Hilbert. Methods of Mathematical Physics. Wiley Classics Library, 1989. E.H. Dowell and D.M. Tang. Dynamics of Very High Dimensional Systems. World Scientific, 2003. J.H. Argyris and S. Kelsey. Energy Theorems and Structural Analysis. Butterworth, 1960. G. Temple and W.G. Bickley. Rayleigh’s Principle and Its Applications to Engineering. Oxford University Press, 1933. W.J. Duncan. Galerkin’s method in mechanics and differential equations. Reports and Memoranda No. 1798, Aeronautical Research Council London (England), 1937. H. Troger and A. Steindl. Nonlinear Stability and Bifurcation Theory: An Introduction for Engineers and Applied Scientists. Springer-Verlag, 1991. H.B. Keller. Numerical solution of bifurcation and nonlinear eigenvalue problems. In P. Rabinowitz, editor, Applications of Bifurcation Theory. Academic Press, 1977, pp. 359–89. T.J.R. Hughes. Finite Element Method-Linear Static and Dynamic Finite Element Analysis. Prentice Hall, 2000. O.C. Zienkiewicz and R.L. Taylor. The Finite Element Method. McGraw Hill, 1989. K.J. Bathe. Finite Element Procedures. Prentice Hall, 1995. M.A. Crisfield. Nonlinear Finite Element Analysis of Solids and Structures, Vol. 1: Essentials. Wiley, 1997. M.A. Crisfield. Nonlinear Finite Element Analysis of Solids and Structures, Vol. 2: Advanced Topics. Wiley, 1997.

65

17:51

P1: KAE Chapter-05

CUFX159-Virgin

0 521 85648 5

August 14, 2007

5 Discrete-Link Models

5.1 Introduction We first consider a number of discrete-link models in which system properties are concentrated at specific locations. The motivation for considering simple mechanical models is that most of the concepts of dynamics and stability issues encountered with continuous systems (e.g., beams, plates) can be observed with discrete systems but are somewhat easier to analyze [1–7]. In fact the governing equations will tend to be algebraic rather than differential (at least in space), and it is natural to start with a look at systems in which the behavior of the system is completely described by just a single degree of freedom.

5.2 An Inverted Pendulum Consider the simple hinged cantilever illustrated in Fig. 5.1. This system consists of a concentrated mass supported by a massless but rigid bar of length L. A torsional spring supplies a linear restoring force that is proportional to the angle of rotation of the hinge (in either direction), with spring coefficient K. The angle of rotation θ thus describes the location of the mass at any given instant of time. Typically, the vertical force is simply P = Mg, but here we assume that an axial load of magnitude P acts independently of the constant mass M. This assumption will be reexamined later. For a simple system like this, we can easily use any of the fundamental approaches introduced in Chapter 2 for writing the governing equation of motion. Approaching this problem by using Lagrange’s equation, we can write the total potential energy of the system V as consisting of two parts: U, the strain energy stored in the spring as it deflects, and VP , the potential energy associated with the work done by the axial load as the mass moves through a vertical distance. Thus V = U + VP =

1 2 Kθ − PL(1 − cos θ). 2

(5.1)

Similarly, the kinetic energy T is given by T=

1 ML2 θ˙ 2 , 2

(5.2)

and placing these into Lagrange’s equations, we obtain ML2 θ¨ + Kθ − PL sin θ = 0, 66

(5.3)

17:53

P1: KAE Chapter-05

CUFX159-Virgin

0 521 85648 5

August 14, 2007

5.2 An Inverted Pendulum

67

P

M

L Figure 5.1. A simple hinged bar (the inverted pendulum model).

K

and using sin θ = θ for small θ we see an effective natural frequency ω2 = K/(ML2 ) − P/(ML). Equation (5.3) can be nondimensionalized by assuming ω2n = K/ML2 ,

p = PL/K,

(5.4)

to give θ¨ + ω2n (θ − p sin θ) = 0.

(5.5)

Equation (5.5) is a nonlinear second-order, homogeneous, ordinary differential equation with constant coefficients.

5.2.1 Static Behavior Let’s consider the underlying equilibrium behavior, which we can easily obtain by setting the time-dependent terms equal to zero: ω2n (θ − p sin θ) = 0.

(5.6)

This is the first variation of the potential energy and could have been obtained from a direct application of the principle of stationary potential energy. Clearly, we have the trivial (or fundamental) equilibrium state for the perfectly upright position θ = 0. However, we see that another (postbuckled) solution to Eq. (5.6) is (for p > 1) p=

θ . sin θ

(5.7)

Equation (5.7) is plotted together with the trivial solution in Fig. 5.2(a). These paths intersect at p = 1, that is, P = K/L. This is the critical load of the structure at which point the trivial equilibrium position gives way to an inclined position. We establish

17:53

P1: KAE Chapter-05

CUFX159-Virgin

0 521 85648 5

August 14, 2007

68

Discrete-Link Models

1.6

1.6

p

p 1.4

1.4

1.2

1.2

1

1

0.8

0.8

0.6

0.6

0.4

0.4

0.2

p

1.0

0.2

(a) 0 -1.5

(b) -1

-0.5

0

0.5

1

θ

1.5

0

0

0.2

(c) 0.4

0.6

0.8

ε

1

θ 0.0

Figure 5.2. (a) The inverted pendulum and (b) equilibrium paths; (c) potential-energy contours (truncated for extreme levels).

that the trivial equilibrium path is unstable for loads greater than p = 1 by examining the curvature of the total potential energy in the vicinity of equilibrium. For stability it is sufficient that d 2V > 0, dθ 2

(5.8)

and thus for θ = 0 the preceding condition is satisfied only if p < 1. To test the stability of the secondary (postbuckled) solution, we evaluate Eq. (5.8) for the system along the secondary path described by Eq. (5.7), which results in d 2V = 1 − p cos θ. dθ 2

(5.9)

This is clearly positive (and indicative of stability) provided that p

1). These are termed complementary and would not normally be observed in a natural loading history. 1 .6

p

p

1 .4 1 .2

1.0

1 0 .8 0 .6

0

0 .4

0

0 .2

0

(a) 0 -1.5

-1

-0.5

0

0.5

1

= 0.1 = 0.2 = 0.3

(b) 1.5

0.0

Figure 5.3. (a) The equilibrium paths and (b) potential energy of the imperfect inverted pendulum model (truncated for extreme levels).

17:53

P1: KAE Chapter-05

CUFX159-Virgin

0 521 85648 5

August 14, 2007

70

Discrete-Link Models

The stability of the imperfect equilibrium paths is still obtained from the second derivative of the total potential energy: d 2V = 1 − p cos θ, dθ2

(5.15)

which, when evaluated on the primary (i.e., from zero-load) equilibrium curve, indicates stability [because p < 1/(cos θe )]. For the complementary path, part of the curve is stable and part unstable. In Fig. 5.3(a), they are identified by different dashed curves for different initial imperfections. Their stability of course can be determined from the potential energy, but stability considerations are deferred to the dynamic criterion discussed in the next subsection. Also plotted in this figure [part (b)] is a contour plot of the total potential energy for the specific case of θ0 = 0.1. The darker shades correspond to lower levels of potential energy for a given loading. Following the minimum as a function of load would thus lead to gradually increasing deflection θ in the positive direction, that is, the same direction as the initial deflection. We also see how the complementary path is represented by a (remote) minimum beyond a local maximum.

5.2.3 Dynamic Behavior We now turn to the dynamic response of the inverted pendulum model. We know that the trivial equilibrium solution is stable provided the magnitude of the axial load is less than its critical value, and thus we would expect small oscillations about the origin given some initial disturbance from a stable equilibrium state. Expanding the sine term in Eq. (5.5) about zero and dropping higher-order terms leads to θ¨ + ω2n (1 − p)θ = 0.

(5.16)

Here, we are assuming that although the mass provides the axial load the inertia is independent. We shall lift this restriction a little later. The linearized (effective) natural frequency drops toward zero as the critical load is approached, ωn → 0,

p → 1,

(5.17)

and, with harmonic motion θ(t) = c sin ωt, there is a linear relationship between the applied load and the square of the natural frequency ω2 = ω2n (1 − p), where p = PL/K and ω2n = K/ML2 . We can thus observe this decay if we plot a time series in which the load is made a slowly increasing function of time in exactly the same way that was considered in Chapter 3. We would expect to have oscillatory behavior about the stable postbuckled paths. We return to Eq. (5.5) and expand about a general equilibrium path θ = θe + δ.

(5.18)

δ¨ + ω2n [(θe + δ) − p sin(θe + δ)] = 0.

(5.19)

Placing this in Eq. (5.5) we have

17:53

P1: KAE Chapter-05

CUFX159-Virgin

0 521 85648 5

August 14, 2007

5.2 An Inverted Pendulum

71

Assuming δ is small such that cos δ ≈ 1 and sin δ ≈ δ, and because θe − p sin θe = 0, we are then left with the linearized (variational) equation of motion: δ¨ + ω2n [1 − p cos θe ]δ = 0.

(5.20)

For example, suppose p = 1.1; then we have θe = ±0.75 rad, which corresponds to ω2 = 0.35. Hence we obtain the result that although the natural frequency drops to zero as the critical load is approached, it then starts to increase as the load is increased into the (stiffening) postbuckled range. We shall return to this type of behavior in the next section. If we are not to be restricted to relatively small-amplitude oscillations about equilibrium then we must solve the nonlinear equation of motion, and this can be easily accomplished numerically [9]. We can also relate time and stiffness linearly— this provides a useful way to visualize the dynamics of the system in the vicinity of equilibrium while the load is slowly increased (as done in the introduction). Such a scheme can be incorporated numerically by θ¨ + θ − 0.01τ sin θ = 0,

(5.21)

where the critical load is reached after 100 time units have elapsed. We have also scaled the time according to τ = ωn t,

(5.22)

and hence the overdots in Eq. (5.21) signify θ˙ ≡ d θ/d τ. Initial conditions were chosen as θ(0) = 0.1, θ˙ (0) = 0.0 to start not too far away from the equilibrium, and the result of a numerical simulation is shown in Fig. 5.4. Another conceptual aid in understanding this dynamic behavior is to imagine a small ball rolling on the potential-energy surface given by Eq. (5.1). For small

p (t)

2 1.5 1 0.5

0.0 25

50

75

100

125

150

175

Figure 5.4. A time series evolving through increasing axial load.

17:53

P1: KAE Chapter-05

CUFX159-Virgin

0 521 85648 5

August 14, 2007

72

Discrete-Link Models

0

2

V -0.2

1

-0.4 0

0 0.5

-1 1 1.5 -2

p

Figure 5.5. Potential energy surface as a function of axial load.

oscillations (for which the linearized equation of motion is appropriate), we see how the evolution of the ball motion in this slowly evolving environment follows the local minima of the underlying potential energy function (see Fig. 5.5). The direction (i.e., positive or negative) followed after criticality is quite arbitrary and rotations to the left, that is, for negative θ, are just as likely for other initial conditions. In this figure, the potential-energy contours are cropped for large positive values to aid the view. For a conservative system such as this, we can also gain insight from plotting contours of constant total energy. Two such plots are shown in Fig. 5.6 for p = 0.8 in part (a) and p = 1.2 in part (b), with again the darker shades indicating lower levels of total potential energy. When p = 1.2 we have symmetric stable equilibria at θe = ±1.027. The origin is of course a saddle point at this level of load. Initial conditions close to the stable locations would result in approximately linear vibrations according to the solutions of Eq. (5.20). For initial conditions far away we might (a)

.

0.4

(b)

.

0.4

θ

θ 0.2

0.2

0

0

-0.2

-0.2

-0.4

-0.4

-1.5

-1

-0.5

0

0.5

1

θ

1.5

-1.5

-1

-0.5

Figure 5.6. Contours of total energy: (a) p = 0.8 and (b) p = 1.2.

0

0.5

1

θ

1.5

17:53

P1: KAE Chapter-05

CUFX159-Virgin

0 521 85648 5

August 14, 2007

5.2 An Inverted Pendulum

θ

73

.

(a)

1.5 1

0.2

0.5

0.1 20

40

60

80

100

τ -1.5

-1

-0.5

0.5

-0.5

-0.1

-1

-0.2

-1.5

-0.3

θ

(b)

0.3

1.4

(c)

1.2

1

1.5

.

(d)

0.3 0.2

1

0.1

0.8 0.6

0.2

0.4

-0.1

0.2

-0.2

20

40

60

80

100

τ

0.4

0.6

0.8

1

1.2

-0.3

Figure 5.7. Typical large-amplitude oscillations. Both cases correspond to p = 1.2: (a) and (b) θ(0) = 1.5, θ˙ (0) = 0.0, (c) and (d) θ(0) = 0.01, θ˙ (0) = 0.0.

expect large-amplitude oscillations that traverse the origin (slowing down when passing over the potential ridge). An example of such a periodic (but not simple harmonic) oscillation is shown in Fig. 5.7(a). Indeed, for moderately large amplitudes we would find markedly asymmetric oscillations without a traverse of the upright position. A limited amount of analytic progress can be made in such a situation if we retain higher-order terms in the Taylor series expansion about equilibrium. An example of this type of motion (which is very close to the homoclinic solution starting and finishing at the saddle) is shown in Fig. 5.7(c). The conservation of energy can be obtained directly from Eqs. (5.1) and (5.2) or we can alternatively use the simple relation θ¨ = θ˙

dθ˙ dθ

(5.23)

in Eq. (5.5) to separate variables and obtain the velocity as a function of position: θ˙ = ± C − θ2 − 2p cos θ,

(5.24)

where the constant C depends on the initial conditions. For example, in Fig. 5.7(b) we can use the initial conditions to derive a constant C = 2.42, and, using Eq. (5.24), we can confirm, for example, that as the bar passes through its upright position it will be doing so at a nondimensional velocity of 0.141 in either direction. The phase portrait (a plot of velocity versus position) can thus be viewed as a trajectory

1.4

17:53

P1: KAE Chapter-05

CUFX159-Virgin

0 521 85648 5

August 14, 2007

74

Discrete-Link Models

following one of these contours of constant total energy. The trajectories corresponding to the two preceding examples are also included in Fig. 5.7. Before leaving this section it is briefly shown how the (nonlinear) natural frequency of large-amplitude vibration can be extracted from the phase trajectories. We can separate variables in Eq. (5.5) and integrate to obtain t θmax dθ dt = , (5.25) 2 C − θ − 2p cos θ 0 0 where the period is equal to four times the time it takes to go from 0 to θmax : θmax −1/2 τ=4 C − θ2 − 2p cos θ d θ. (5.26) 0

Thus taking the initial conditions (and the value of C) corresponding to the trajectory in Fig. 5.7 we can evaluate the preceding integral (numerically) to confirm the (near homoclinic) period of 29.34 units. We shall return to this type of (elliptic integral) approach when we deal with continuous (elastica) systems. 5.2.4 A Note on Inertia Before moving on it is useful to mention here that in a realistic situation we might expect to increase the axial load in the inverted pendulum model by increasing the mass, that is, P = Mg. However, suppose the pendulum arm has some mass m associated with it (that does not change and is not sufficient to cause self-weight buckling). In this case, the energy expressions [from Eqs. (5.1) and (5.2)] change to V=

1 2 1 Kθ − MgL(1 − cos θ) − mgL(1 − cos θ) 2 2

(5.27)

1 1 ML2 θ˙ 2 + mL2 θ˙ 2 . 2 6

(5.28)

and T=

Thus the equation of motion becomes (M + m/3)L2 θ¨ + Kθ − (M + m/2)gL sin θ = 0.

(5.29)

It is a simple matter to linearize this equation to obtain an expression for the natural frequency (of small-amplitude oscillations) ω2 =

K − (M + m/2)gL , (M + m/3)L2

(5.30)

which vanishes when the end mass achieves its critical value, Mcr =

m K − . gL 2

(5.31)

The natural frequency and critical mass now provide a suitable means of nondimensionalizing, 2 = ω2 L/g,

k = K/(mgL),

p = M/m,

(5.32)

17:53

P1: KAE Chapter-05

CUFX159-Virgin

0 521 85648 5

August 14, 2007

5.3 A Discrete-Strut Model

75

0.5

(a)

t

p 0.4

0.0 20

0.3

40

60

80

100

-0.5

0.2 -1 0.1

p (t)

-1.5 0.0

0.2

0.4

0.6

0.8

1

1.2

1.4 2

Figure 5.8. (a) The mass–frequency relation for the inverted pendulum model and (b) a numerical time series resulting from a linear increase in mass with time.

which can then be used in the relation between the natural frequency and end load: ω2 =

3(2k − 1 − 2p) . 2(1 + 3p)

(5.33)

Given the nondimensionalization, the critical load is now p cr = k − 0.5, and a plot of mass (load) versus natural frequency squared is shown in Fig. 5.8(a) for k = 1. We can evolve the magnitude of the end mass as a linear function according to M = 0.01t, which gives the time series shown in Fig. 5.8(b). The critical mass is reached after 50 time units, after which the (undamped) system starts to oscillate as it follows one of its (nontrivial) postbuckled paths.

5.3 A Discrete-Strut Model We now move on to consider another rigid-link model, but this time the model is a little more general in the types of behavior it can exhibit. It is also a step closer to the continuous structures we will focus on later. The approach adopted in this section is based on energy considerations rather than the underlying differential equations, and some experimental data are also included. The model under consideration is shown in Fig. 5.9(a). A mechanical model was built to mimic this system by Croll and Walker at University College London [10]. A photographic image is shown in part (b) of Fig. 5.9. It consists of two rigid (massless) links hinged at their supports and in the center where a concentrated mass M is located. Two linear springs provide a restoring force, one against lateral deflection with modulus K, and the other against rotation with modulus C. An axial load of magnitude P acts at the left-hand ends that is unrestrained against horizontal movement. The coordinate Q describes the deflected position of the mass, and we can think of the schematic as providing a plan view; that is, we assume gravity is already taken into account. The total potential and kinetic energies for this system are given by [5] 1 V = 2Cθ2 + KL2 sin2 θ − 2PL(1 − cos θ) 2

(5.34)

(b)

17:53

P1: KAE Chapter-05

CUFX159-Virgin

0 521 85648 5

August 14, 2007

76

Discrete-Link Models P θ C L

L M

K

(a)

(b) Figure 5.9. (a) Schematic of the two-bar link model and (b) physical realization.

and T=

1 ML2 θ˙ 2 . 2

(5.35)

Equilibrium solutions are found from dV = 4Cθ + KL2 sin θ cos θ − 2PL sin θ = 0. dθ

(5.36)

We immediately see the trivial solution, θe = 0, together with the nontrivial solutions given by = α cos θ +

(1 − α)θ , sin θ

(5.37)

where = P/Pcr ,

(5.38)

Pcr =

KL + 4C , 2L

(5.39)

α=

KL2 . KL2 + 4C

(5.40)

2

The parameter α is a ratio of spring stiffnesses. For example, if α = 0 (i.e., K = 0), then we have exactly the same type of equilibrium curves as for the model in the previous section. The stability of the equilibrium paths can again be established from the sign of the second derivative of the total potential-energy function. However, let us assume initially that α = 1 (i.e., C = 0) so that we have only a lateral (translational) spring acting. The equilibrium expression simplifies to = cos θ.

(5.41)

In the presence of initial imperfections, the total potential energy becomes V=

1 KL2 (sin θ − sin θ0 )2 − 2PL(cos θ0 − cos θ) 2

(5.42)

17:53

P1: KAE Chapter-05

CUFX159-Virgin

0 521 85648 5

August 14, 2007

5.3 A Discrete-Strut Model 1.5

77 1.5

(a)

m

0

1

(b)

(c)

90

1

P(N)

60 0

0

0.5

0.5 < 0 0

0 −1

−0.5

0

0

30

0

> 0

0.5

0

0 −0.1

1

0.0 0

0.1 −0.8

−0.4

0

0.4

Figure 5.10. (a) Equilibrium paths for the two-bar link model with a lateral spring and initial geometric imperfections. θ0 ranges from 0.001 closest to the bifurcation to 0.1, (b) imperfection sensitivity and (c) measured data (adapted from [2]).

with the first derivative dV = KL2 (sin θ − sin θ0 ) cos θ − 2PL sin θ. dθ

(5.43)

Using the nondimensionalization, we obtain =

(sin θ − sin θ0 ) cos θ . sin θ

(5.44)

In the literature, it is sometimes observed that the trigonometric terms are replaced with their series expansion, and retaining the first few terms results in equilibrium paths [2]: =

θ − θ0 − 23 θ3 + 12 θ2 θ0 θ−

1 3 θ 6

≈1−

θ0 θ0 θ θ2 + − + ··· + . θ 3 2

(5.45)

These nontruncated [Eq. (5.44)], paths have the form shown in Fig. 5.10(a), in which we recognize the characteristic subcritical pitchfork bifurcation. Complementary paths are also present in this example. However, they prove to be unstable and have little to do with a natural loading path starting near the origin and hence are not included in the plot. For the perfect geometry, it is simple to see that the stability is governed by the coefficient of the second derivative of the potential energy. For the primary path, it is easy to show that equilibrium is unstable if p > 1. Similarly for the secondary path it can be shown that the coefficient is always negative and hence the postbuckled behavior is unstable (sometimes called unstable-symmetric branching behavior in the literature [4]). In practice this means that, if the load were gradually increased from zero, the strut model would buckle when p = 1 and the system would collapse completely (within the confines of the mathematical modeling). It should be mentioned here that experimental data taken from the system shown in Fig. 5.9(b) very closely match the theory shown in Fig. 5.10(a) [2] and is plotted in Fig. 5.10(c).

0.8

17:53

P1: KAE Chapter-05

CUFX159-Virgin

0 521 85648 5

August 14, 2007

78

Discrete-Link Models

For the imperfect geometry the limit of the stability of the paths coincides with the maximum load m (the horizontal tangency) for a given path. It can be shown 1/3 that this occurs when θ = θ0 , and placing this back into Eq. (5.45) results in the cusp geometry shown in Fig. 5.10(b). This displays an important characteristic of some axially loaded structures: the load-carrying capacity of the structure is reduced when initial imperfections are present, i.e., it is imperfection sensitive. This type of subcritical behavior may then be viewed as a potentially dangerous consequence when compared with the supercritical behavior described in the previous section. Following a similar line of reasoning to Section 3.3 we can show that the frequency of small oscillations can be obtained by using Rayleigh’s method: 2 = 1 − cos θ − α(1 − cos 2θ),

(5.46)

where ω is the effective natural frequency, = ω2n =

ω , ωn

(5.47)

4C + KL2 , ML2

(5.48)

and again setting α = 1, incorporating initial imperfection θ0 , and simplifying, we get θ2 2 = (1 − 2θ2 + θθ0 ) − 1 − . (5.49) 2 We can then plot Eq. (5.49), incorporating Eq. (5.45) for various initial imperfections, and this is shown in Fig. 5.11 for θ0 = 0.001, 0.01, and 0.1. Thus we see that the often linear relation between the natural frequency squared and the axial load 1.5 0 0 0

1

0.5

0

0

0.2

0.4

0.6

0.8

2

1

Figure 5.11. The relation between the natural frequency and axial load for the imperfect unstable symmetric model. θ0 = 0.001, 0.01, 0.1.

17:53

P1: KAE Chapter-05

CUFX159-Virgin

0 521 85648 5

August 14, 2007

5.3 A Discrete-Strut Model

79

p 0.5 20 0.1

40

τ 60

-0.2

V 0

2

-0.1 -0.2

1

0

θ

-0.4

0 0.5

(a)

-0.6

θ

-1

p

1

-0.8 1.5 -2

(b)

Figure 5.12. (a) The potential-energy surface with an initial imperfection of θ0 = −0.1 and (b) a typical time series evolving through the p axis, θ(0) = −0.2.

is not true for initially imperfect geometries, and in fact here the natural frequency drops to zero before the critical load is reached. The potential-energy surface is plotted as a function of load and deflection in Fig. 5.12(a) together with an evolving time series under linearly increasing end load in part (b). We see how the oscillation continues until the limit point is reached and the trajectory slides off to infinity (actually to large oscillations about π but we are not interested in such solutions). According to Fig. 5.10, we expect a maximum load of about p m ≈ 0.7 for this imperfection and hence given a ramped load of p = 0.01t we would expect the solution to lose stability after about τ = 70 time units. We observe the decay in natural frequency as a gradual lengthening of the period of oscillation as the instability is approached. Now, let’s assume that α = 0 (or K = 0), which in fact produces a situation qualitatively similar to the response of the inverted pendulum. In this case, the equilibria (for the initially perfect geometry) are given by θ = 0,

(5.50)

= θ/ sin θ

(5.51)

for θ = 0 and

for θ = 0, and a frequency–load relation 2 = 1 − cos θ,

(5.52)

which, for the prebuckled (trivial) equilibrium path gives 2 = 1 − .

(5.53)

80

17:53

P1: KAE Chapter-05

CUFX159-Virgin

0 521 85648 5

August 14, 2007

80

Discrete-Link Models

For the postbuckled (nontrivial) equilibrium path, we substitute Eq. (5.51) into Eq. (5.52) to get 2 = 1 − θ cot θ ≈

θ2 . 3

(5.54)

Expanding Eq. (5.51), =

θ2 θ ≈1+ , sin θ 6

(5.55)

and combining Eqs. (5.54) and (5.55) leads to 2 = 2( − 1).

(5.56)

Thus we see the interesting result that, as a function of axial load, the postbuckled linear frequency changes at half the rate of the prebuckled frequency, a result also contained in Eq. (5.20). Of course, this relation is true only for the moderately buckled structure because the trigonometric power series was truncated. It is also mentioned here that these types of link model can be used to illustrate the effect of thermal loading. For example, Croll and Walker [2] showed that an increase in temperature will result in stable-symmetric buckling (supercritical pitchfork bifurcation). It can also be shown that the natural frequency will decay in the usual way as the critical temperature is approached. Thermal buckling is an important consideration for plated structures and will be considered in more detail in a later chapter.

5.4 An Asymmetric Model The models described in the previous sections of this chapter were symmetric in the absence of initial imperfections. There are a number of structural systems that behave quite differently according to the direction of the deformation [1, 3, 11]. In a buckling context, this is characterized by the asymmetric (or transcritical) point of bifurcation: A structure exhibiting this behavior is shown in Fig. 5.13. The deflection of this SDOF system is described by the coordinate X, the horizontal distance of the top of the (massless) bar. It is convenient to nondimensionalize this by the length of the bar: x = X/L. A spring of modulus K provides the restoring force, with x0 denoting the initial imperfection (and thus x0 = X0 /L). However, the dead load (the point mass) is further offset by a fixed amount l (measured at right angles to the bar and set equal to αx0 ). The equilibrium paths are obtained from the first derivative of the total potential energy and are given by p=

2[(1 − x2 )1/2 − (1 − x)1/2 (1 + x0 )1/2 ] , [x + αx0 (1 − x2 )1/2 ]

(5.57)

where p = P/(KL). Some typical paths are shown in Fig. 5.14(a) for a variety of initial imperfections. It is clear that the behavior (including instability behavior) is strongly influened by the sign of the initial imperfection. The stability of these paths

17:53

P1: KAE Chapter-05

CUFX159-Virgin

0 521 85648 5

August 14, 2007

5.4 An Asymmetric Model

81 X X0 l

M P = Mg K L

L

Figure 5.13. Schematic of an asymmetric link model [11].

can be determined from the second derivative of the total potential energy. However, we can use the alternative criterion of requiring real frequencies for stability. These are obtained from Lagrange’s equation and are given by 2 = (1 + x0 )1/2 (1 + x)−3/2 − p(1 − x2 )−3/2 ,

(5.58)

where p is determined from Eq. (5.57) and is nondimensionalized with respect to (KL2 )/(2M). Some typical frequency–load plots are shown for the same set of 2

p

x0 > 0

(a)

2

(b)

p 1.5

1.5 x0 < 0 1

1

x0 < 0

x0 = 0 0.5

0.5 x0 > 0 0 −1

− 0.5

0

0.5

x 1

0

0

0.5

1

1.5

2

Figure 5.14. (a) Equilibrium paths and (b) frequency–load relation. α = 5, x0 = 0, ±0.005, ±0.015.

2

17:53

P1: KAE Chapter-05

CUFX159-Virgin

0 521 85648 5

August 14, 2007

82

Discrete-Link Models P L

Y1L

m

L

Y2L L m

k

m k

Figure 5.15. A schematic of a three-bar link model.

initial imperfections in Fig. 5.14(b). We see how for negative values of x0 the natural frequency decays to zero as the system buckles at a limit point. However, when the initial imperfection is in the other direction no instability is encountered and the frequencies start to increase as the structure moves beyond the critical value for the corresponding perfect system. We again note the presence of the complementary equilibrium paths that would not be ordinarily encountered under a monotonic increase in axial load, that is, the natural loading path. The sign of the initial imperfection also has a strong effect on the load-carrying capacity of the structure, and, given the often arbitrary nature of the initial imperfection, this system presents obvious concern in a design context. This type of behavior is sometimes encountered in frame structures. We again note that the axial load is assumed to be an independent parameter (see the discussion in Subsection 5.2.4).

5.5 A Three-Bar Model Now let us consider the dynamics and stability of a simple mechanism made up of three rigid links of length L and mass per unit length m. They are hinged at their connections, and linear rotational springs of stiffness coefficient k are placed at the two internal joints, as shown in Fig. 5.15 [4]. It is assumed that the structure is in equilibrium in its undeflected (straight) configuration and that an axial load of magnitude P is acting. This model has two degrees of freedom, that is, an equilibrium configuration is determined if two coordinate values are specified. In contrast to the simple examples outlined earlier, there is some flexibility in the way we choose the coordinates to describe the deflected configuration of the system. Suppose we choose the lateral deflections as the coordinates (as shown in the figure). In this case, we can write down the strain energy stored in the springs as U=

1 1 k[sin−1 Y1 − sin−1 (Y2 − Y1 )]2 + k[sin−1 Y2 − sin−1 (Y2 − Y1 )]2 , 2 2

(5.59)

which can be expanded to give U=

1 k[5Y 21 − 8Y1 Y2 + 5Y 22 + · · · +]. 2

(5.60)

17:53

P1: KAE Chapter-05

CUFX159-Virgin

0 521 85648 5

August 14, 2007

5.5 A Three-Bar Model

83

We can also write the potential energy associated with the movement of the axial load: 1/2 1/2 V = −PL 3 − 1 − Y12 (5.61) − 1 − Y22 − [1 − (Y2 − Y1 )2 ]1/2 , which can also be expanded to give V = −PL[Y 21 − Y1 Y2 + Y 22 + · · · +].

(5.62)

The kinetic energy for a typical link with displacement a and b at each end is L 1 2 ˙ [a(1 ˙ − x/L) + b(x/L)] d x, (5.63) T= m 2 0 which, after substitution and adding the effects of all three links, leads to

2 1 1 T = mL2 Y˙ 21 + Y˙1Y˙2 + Y˙ 22 . 2 3 2

(5.64)

Thus we have our Lagrangian L = T − U + VP ,

(5.65)

and a direct application of Lagrange’s equation will lead to the equations of motion by use of the dummy suffix notation Tij eY¨j + Vij e Yj = 0,

(5.66)

and, assuming harmonic oscillations Yj = Aj sin ωt, we obtain Vij e Aj − ω2 Tij e Aj = 0, thus leading to the characteristic equation e Vij − ω2 Tij e = 0.

(5.68)

For the specific case at hand, we then have 5k − 2PL − ω2 2 mL3 − 4k + PL − ω2 16 mL3 3 = 0, − 4k + PL − ω2 1 mL3 5k − 2PL − ω2 2 mL3 6

(5.67)

(5.69)

3

and, by using p= 2 =

PL , k

(5.70)

ω2 mL3 , k

(5.71)

we get 6 [(8 − 3p) ± (2p − 7)] , 5 and thus the two natural frequencies 2 =

(5.72)

6 (1 − p), 5

(5.73)

22 = 6(3 − p).

(5.74)

21 =

17:53

P1: KAE Chapter-05

CUFX159-Virgin

0 521 85648 5

August 14, 2007

84

Discrete-Link Models

Each of these eigenvalues corresponds to an eigenvector that describes both the mode of buckling and vibration: The first corresponds to a mode in which both generalized coordinates are equal, the second to a mode in which the coordinates are equal but opposite in sign. From Eqs. (5.73) and (5.74), we can set the natural frequencies equal to zero to obtain the critical loads p1 = 1

(P1 = k/L),

(5.75)

p2 = 3

(P2 = 3k/L).

(5.76)

And setting the axial load equal to zero, we get the natural frequencies 2

ω1 = 6k/(5mL3 ) , 21 = 6/5 2

ω2 = 18k/(mL3 ) . 22 = 18

(5.77) (5.78)

From Eqs. (5.73) and (5.74), we again see the linear relation between the square of the natural frequency and the axial load. The linearity of this relation occurs because of the equivalence of the buckling modes and the natural modes of vibration, having a finite set of generalized coordinates, and the fact that the frequencies depend on a single parameter. We would not necessarily expect this relation to be exactly linear for the general continuous case, as explained earlier. This example can also provide a powerful illustration of the utility of choosing principal coordinates. For this two-DOF (2DOF) system, it is a simple matter to expand the determinant of Eq. (5.69), and, of course, there are a myriad of techniques for achieving this numerically for higher-order systems [9]. However, our earlier theory illustrated how coordinate transformations can be useful in the setting up of the equations of motion in going from physical or generalized to principal coordinates. Thompson and Hunt [4] show that the simple transformation Y1 + Y2 , 2 Y1 − Y2 u2 = 2 u1 =

(5.79) (5.80)

decouples the equations, enabling the critical loads and natural frequencies to be written immediately [see Eq. (3.43)]. However, this type of direct transformation is usually not obvious a priori in a typical analysis. A large variety of numerical algorithms are available to compute this transformation efficiently.

5.6 A Snap-Through Model In this section, we look at a simple example of snap-through buckling associated with a saddle-node bifurcation [12, 13]. The link model shown in Fig. 5.16 consists of two rigid massless links of length L, with a linear spring K allowing horizontal movement at one end. A point mass M is located at the center where a point load of magnitude P acts in the vertical direction. Again we assume that P acts independently of the mass, and we can also think of Fig. 5.16 as a plan view, such that gravity

17:53

P1: KAE Chapter-05

CUFX159-Virgin

0 521 85648 5

August 14, 2007

5.6 A Snap-Through Model

85

M L

K

L P

Figure 5.16. Schematic of a simple arch link model.

is not an issue. The generalized coordinate for this SDOF system is again θ, which is measured from the initial rise of the structure that is fixed at α = π/8 in keeping with earlier work on this model. We can write the total potential energy for this model as V = 2KL2 [cos (α − θ) − cos α]2 − PL [sin α − sin (α − θ)]

(5.81)

and a kinetic energy of T=

1 ML2 θ˙ 2 . 2

(5.82)

A direct application of Lagrange’s equation, and defining p = P/(4KL),

ω2n = 4K/M,

(5.83)

leads to the nondimensional equation of motion: θ¨ + ω2n [cos (α − θ) − cos α] sin (α − θ) − p cos (α − θ) = 0.

(5.84)

The equilibrium path for an initial rise of α = π/8 is plotted in Fig. 5.17, where it is shown superimposed on contours of the total potential energy. It follows the

p 0.02 0.01 0 -0.01

-0.2

0

0.2 0.4 0.6 0.8

1

θ

1.2

Figure 5.17. The potential-energy surface as a function of axial load with the equilibrium path superimposed.

17:53

P1: KAE Chapter-05

CUFX159-Virgin

0 521 85648 5

August 14, 2007

86

Discrete-Link Models run down

(rad) 1 0.8 0.6 0.4

sweep up

0.2 0.0 –0.2

100

200

300

400 t = 1,000p

Figure 5.18. Evolution through snap-through illustrating hysteresis. Initial conditions: For both θ˙ = 0.0, θ(0) = −0.1 at t = 0 and θ(0) = 0.8654 at t = 400.

stationary points of the potential energy. The second derivative can be used to determine which of these stationary points is a minimum, and we know that a critical point is reached when the equilibrium curve passes through a horizontal tangency. For this system, the limit point is reached when p = 0.0116, which corresponds to a deflection of θ = 0.166 rad, that is, when the load reaches this value the structure has deflected to nearly 10 deg and then suddenly snaps through to an inverted position (which is about θ = 0.844 rad, that is, about 25 deg below the horizontal). Note that in this figure the energy is cropped for large values (white) and low values (black) of the potential energy, but in general the darker the shade, the deeper the potential-energy well at that load. Conducting a numerical integration, we can sweep through the instability by using the following loading function: p = −0.02 + 0.0001t.

(5.85)

An example is shown in Fig. 5.18, in which an initial condition is used that is 0.01 rad away from stable equilibrium, and ωn = 1 for convenience. Using the ramp function of Eq. (5.85) thus converts to an anticipated critical time of tcr = 316. The loss of stability is abrupt (albeit slightly delayed) under the gradual increase in load. The postbuckled behavior is characterized by relatively large-amplitude oscillation about the inverted position because of the large transient motion initiated by the jump. Note the asymmetric nature of the waveform as the trajectory oscillates along its decidedly asymmetric potential-energy surface. We can conduct a reverse sweep by changing the direction of the load, and again starting from an initial condition adjacent to the equilibrium we again observe the snap back to the original branch, in which case the load evolution has been scaled such that a jump occurs at about t = 84. This evolving trajectory is shown in gray in Fig. 5.18, and thus a region of hysteresis is revealed. The loading described in this example can be considered as “dead” or “force loaded.” An alternative, which can often be the most practical approach in a

17:53

P1: KAE Chapter-05

CUFX159-Virgin

0 521 85648 5

August 14, 2007

5.6 A Snap-Through Model

87

laboratory context, is to use a “displacement-loaded” device. In the former case, it is straightforward to prescribe the load (usually because of gravity) and then measure deflection. In the latter, we prescribe deflection and measure the resulting load (using a load cell for example). In the displacement-controlled approach, the hysteresis is manifested in terms of displacement, and thus the nature of the stability of equilibrium changes [2]. This also brings with it an issue of extensibility, because of a SDOF model, for example, will not be able to oscillate if subject to displacement-controlled loading. The ability to follow equilibrium paths that have bifurcations or turning points is a subject of considerable importance and will be dealt with in more detail in a later section. We note finally that, unlike for symmetric systems possessing a trivial equilibrium solution, this type of limit point buckling is not sensitive to initial imperfections. It is affected by the presence of a small initial imperfection (in fact linearly) but not in a disproportionate sense. We shall also look at the analog of this system in a continuous (arch) structure later. Before leaving this section, we again take a brief look at the relation between the load and the natural frequency. The local stiffness of the force–deflection curve can be obtained from the derivative of the equilibrium condition, that is, dp cos α − cos3 (α − θ) = , dθ cos2 (α − θ)

(5.86)

which in turn is linearly related to the square of the effective linear natural frequency in the usual way. Hence we can plot the square root of the right-hand side of Eq. (5.86) against load, as shown in Fig. 5.19. An alternative view of stiffness would relate load and vertical deflection rather than angle but the relation is very nearly 0.015 p cr = 0.0116

p

ω 0.01 ω2

0.005 ω4 0 0

0.1

0.2

0.3

0.4 Frequency

0.5

Figure 5.19. The relation between the natural frequency and load for the snap-through model.

17:53

P1: KAE Chapter-05

CUFX159-Virgin

0 521 85648 5

August 14, 2007

88

Discrete-Link Models

the same [2]. It is actually the natural frequency raised to the fourth power that varies linearly with load when close to the instability [14, 15]. This is a typical result in the vicinity of a saddle-node bifurcation and will be touched upon again in the chapter on nondestructive testing.

5.7 Augusti’s Model In this section, we again focus attention on an inverted pendulum model but now replace the torsional spring with a universal joint, that is, a hinge that is not confined to a plane. This is sometimes referred to as Augusti’s model in the literature [1, 16, 17]. Thus the deflection of the system needs two coordinates for a complete description. This model shows some interesting behavior when modes interact. Specifically, we outline a bifurcation from nontrivial equilibrium: the secondary bifurcation. The model is shown in Fig. 5.20. A slender, rigid (but massless) bar of length L is pinned at its base, where rotational springs with constant stiffnesses C1 and C2 (C2 > C1 ) initially act in perpendicular planes and rotate with the bar. The corresponding angles of rotation with respect to two horizontal, perpendicular axes are α1 (t) and α2 (t), and the angles θ1 (t) and θ2 (t) are defined as θ1 (t) = (π/2) − α1 (t),

θ2 (t) = (π/2) − α2 (t),

(5.87)

with θ1 (t) = θ10 and θ2 (t) = θ20 when the springs are unstretched. A downward vertical load P is applied at the top of the bar. A concentrated mass M is attached at the top of the bar.

P P M M m

L

L

m C2 2

C1

1

x2 x1 Figure 5.20. Geometry of the Augusti model.

x2 x1

17:53

P1: KAE Chapter-05

CUFX159-Virgin

0 521 85648 5

August 14, 2007

5.7 Augusti’s Model

89

The potential energy V is given by [18] 1 1 C1 (θ1 − θ10 )2 + C2 (θ2 − θ20 )2 2 2

V=

−PL[(1 − sin2 θ10 − sin2 θ20 )1/2 −(1 − sin2 θ1 − sin2 θ2 )1/2 ], and the kinetic energy T is 2 2 d θ 1 d θ 1 2 T = ML2 cos2 θ1 + cos2 θ2 2 dt dt 2 2 [(d θ1 /d t) sin 2θ1 + (d θ2 /dt) sin 2θ2 ] + 2ML . (1 − sin2 θ1 − sin2 θ2 )

(5.88)

(5.89)

We can again use Lagrange’s equations to obtain the equations of motion. The nonlinear inertia terms in the resulting equations do not affect small vibrations of the system about an equilibrium state. Hence, for simplicity, only the linearized inertia terms are used. The analysis is conducted in terms of the following nondimensional quantities, where is a dimensional vibration frequency: c = C2 /C1 ,

p = PL/C1 ,

τ = t(C1 /ML2 )1/2 ,

ω = (ML2 /C1 )1/2 ,

(5.90)

with c > 1. It can be shown (from equating the second derivative of the potential energy to zero) that p = 1 is the critical buckling load, that is, for the perfect system (the bar is vertical and the springs unstretched). The coupled, nonlinear equations of motion are obtained as p (1 − sin2 θ1 − sin2 θ2 )−1/2 sin θ1 = 0, 2 p θ2 + cθ2 − cθ20 − (1 − sin2 θ1 − sin2 θ2 )−1/2 sin θ2 = 0. 2 θ1 + θ1 − θ10 −

(5.91) (5.92)

Again we focus initially on the underlying equilibria of the perfect system geometry (θ10 = θ20 = 0). The four solutions are θ1 = 0,

θ2 = 0,

θ2 = 0,

p = θ1 / sin θ1 ,

θ1 = 0,

p = cθ2 / sin θ2 , p = (θ1 / sin 2θ1 ) (1 − sin2 θ1 − sin2 θ2 ) = (cθ2 / sin 2θ2 ) (1 − sin2 θ1 − sin2 θ2 ).

(5.93)

These curves are plotted in Fig. 5.21(a), with the secondary bifurcation occurring at c sin 2θ1∗ = 2θ1∗ ,

p = c cos 2θ1∗ ,

(5.94)

17:53

P1: KAE Chapter-05

CUFX159-Virgin

0 521 85648 5

August 14, 2007

90

Discrete-Link Models

1.1

1.1

1.05

(b)

p

(a)

p Primary path

1.05

1

2 B

1

0.95

0.95

Secondary path ( 1* = 0.3745, p* = 1.024)

0.9 0.85

0.9

(c = 1.1) 2

Fundamental (trivial) path

A

0.85 (c = 1.1)

0.8

0

0.2

0.4

0.6

0.8

0.8

1

0

0.05

0.1

0.15

0.2

0.25

1

2

0.3

Figure 5.21. (a) Equilibrium paths for the perfect model and (b) corresponding characteristic curves. 1.1 (c = 1.1, θ10 = 0.01) p 1.05

1.1

(a)

(c = 1.1, θ10 = 0.01)

p

(b)

1.05

1

1

ω B2

(θb = 0.4183, pb = 1.0052) 0.95

0.95

0.9

0.9

0.85

0.85

0.8

0

0.2

0.4

0.6

0.8

θ1

1

0.8

0

ωA2 0.05

0.1

0.15

0.2

0.25

ω2

0.3

Figure 5.22. (a) Equilibrium paths for the imperfect model and (b) corresponding characteristic curves.

which for the specific case shown (with c = 1.1) is at (θ1∗ , p ∗ ) = (0.3745, 1.024). Shown in part (b) is the dependence of the natural frequencies on the axial load. These are evaluated on the equilibrium paths. For the trivial equilibrium, we have ω2A = 1 − p,

ω2B = c − p,

(5.95)

and for the primary (postbuckled, i.e., p > 1) branch, we have ω2A = 1 − p cos θ1 = 1 − θ1 / tan θ1 ,

(5.96)

ω2B = c − p cos θ1 = c − 2θ1 / sin 2θ1 .

(5.97)

For the case in which there is a small amount of initial geometric imperfection, the results shown in Fig. 5.22 are obtained. The initial angle θ10 = 0.01 is chosen. The secondary bifurcation now occurs directly from the primary path. It is interesting to note that previous studies of this system have often used truncation to ease some

17:53

P1: KAE Chapter-05

CUFX159-Virgin

0 521 85648 5

August 14, 2007

5.8 Multiple Loads

0.4

91

θ1

0.2 0.0 250

−0.2

500

750

1000

1250

τ

1500

1000

1250

τ

1500

−0.4

0.4

θ2

p(τ )

0.2 0.0 −0.2

250

500

750

−0.4

Figure 5.23. A slow sweep through secondary bifurcation.

of the computations, and in that case the load–frequency (squared) relations are exactly linear. We again close this section by conducting a numerical simulation as the system is swept through the bifurcation(s). Figure 5.23 was based on the following load evolution, p = 0.95 + 0.00005τ,

(5.98)

where the initial geometry is perfect and a very small amount of damping was added to the system. We observe a gradual decrease in the θ1 natural frequency as the system approaches the initial bifurcation at τ = 1000 and then a gradual increase in the natural frequency. However, in the lower part of this figure we also see that it is the natural frequency associated with θ2 that decreases as the secondary bifurcation is approached at τ = 1480. This system will be revisited a couple of times later in this book: as an example of a path-following algorithm and in cases in which oscillations are not necessarily small.

5.8 Multiple Loads In all the link model examples so far there has been a single axial load. A monotonic increase in this parameter resulted in a linear decay in the natural frequency (squared). However, if more than one independent axial load is present, then the natural frequencies will still decay if either or both of these loads are increased,

17:53

P1: KAE Chapter-05

CUFX159-Virgin

0 521 85648 5

August 14, 2007

92

Discrete-Link Models

P2 m2 2

P1 m1

L

k2

Figure 5.24. A 2DOF model under the action of two axial loads.

L 1

k1

but a linear relation is no longer to be expected. To illustrate this effect, consider the link model shown in Fig. 5.24. Following [19, 20], and assuming k1 = 2k, k2 = k, m1 = 2m, m2 = m, we can write the energy terms: 1 U = kθ12 + k(θ2 − θ1 )2 , 2 1 1 VP = − LP1 θ12 − LP2 θ12 + θ22 , 2 2 1 T = mL2 3θ˙ 12 + θ˙ 22 + 2θ˙ 1 θ˙ 2 . 2

(5.99) (5.100) (5.101)

We can nondimensionalize by using the following parameters, 2 =

ω2 mL2 , k

pi =

Pi L , k

(5.102)

and by using Lagrange’s equations obtain the characteristic equation 24 + 2 (p 1 + 4p 2 − 8) + p 22 + p 1 p 2 − p 1 − 4p 2 + 2 = 0.

(5.103)

Setting 2 = 0 gives the two critical load conditions from the quadratic p 22 + p 1 p 2 − p 1 − 4p 2 + 2 = 0,

(5.104)

which can be solved for the lowest critical loads acting separately of p 1 = 2 and p 2 = 0.586. Setting both axial loads equal to zero leads to the natural frequencies 21 = 0.268 and 21 = 3.732. Thus we can plot the roots of Eq. (5.103) as two surfaces, as shown in Fig. 5.25. The curve on the left is the most relevant given the typical situation of a monotonic increase in the loads, that is, it will be encountered first.

17:53

P1: KAE Chapter-05

CUFX159-Virgin

0 521 85648 5

August 14, 2007

5.9 Load-Dependent Supports

2

93

3

0.2

4

2 2 2

1 0.1 0.4 0 0.5

0.2

p1

1

3

1 0 0

p2

2 1

p1

1.5 2 0

1

2 3 40

Figure 5.25. The characteristic surface plotting the natural frequency squared in terms of the two axial loads.

5.9 Load-Dependent Supports It may happen that the stiffness of a structure or its supports is a function of the applied axial load [21, 22]. To illustrate this situation, we briefly return to the inverted pendulum model (see Fig. 5.1), but now we assume a distributed mass such that the moment of inertia of the bar about its base is I. We assume a fixed vertical end load P and ignore gravity. We assume a spring stiffness, rather than a constant torsional stiffness k, that increases linearly with the end load from a baseline value of K0 . The equation of motion is given by θ¨ − p sin θ + k(p)θ = 0, where the following nondimensional parameters have been used, p = PL/K0 , k(p) = K(pK0 /L)/K0 , t = T (K0 /I),

(5.105)

(5.106)

and the derivatives in Eq. (5.105) are with respect to the scaled time t. Equilibrium conditions are obtained from −p sin θ + k(p)θ = 0,

(5.107)

and the frequencies of small vibrations about these equilibria are given by ω2 = k(p) − p cos θ.

(5.108)

Now, assume the spring stiffness is related to load in the following way: k(p) = 1 + γ p,

(5.109)

where γ is taken as positive. Equilibrium paths and load–frequency relations are plotted in Fig. 5.26 for a number of different values of the parameter γ. We see an increase in the critical load and an increase in the natural frequency at a given load as a function of spring stiffness. Further examples of rigid-link models and their stability can be found in Seyranian and Mailybaev [23].

p2

17:53

P1: KAE Chapter-05

CUFX159-Virgin

0 521 85648 5

August 14, 2007

94

Discrete-Link Models 3

3 (a)

p

p

0.4

2.5

(b)

= 0.6

2.5 = 0.6

2 1.5

0.2

0.4

2

0.2

1.5

0.0

0.0 1

1

0.5

0.5

0

-2

-1.5

-1

-0.5

0

0.5

1

1.5

0

2

0

0.2

0.4

0.6

0.8

2

1

Figure 5.26. (a) Equilibrium paths and (b) load–frequency relation for a link model with a stiffening spring.

5.10 Path Following and Continuation In a number of places in this chapter, we have had cause to solve nonlinear algebraic equations, and this will be the case throughout this book. The standard technique for solving sets of nonlinear algebraic equations is Newton–Raphson [9, 24] and this is a standard feature in Mathematica [25] and MATLAB [26]. However, the solution path, as a parameter is changed, may be quite complicated (e.g., including turning points), and some difficulty may be encountered. A number of specialized techniques have been developed based on augmenting Newton–Raphson such that a solution path is followed. These are predictor–corrector techniques and work for differential as well as algebraic equations, and are typically called continuation methods. Because these techniques typically involve the rates of change of the response as a function of a parameter, they obtain information about stability (based on the evaluation of the Jacobian) without too much difficulty. A popular and efficient algorithm is contained in the software package auto.

1.8 1.6

1.6 p

1.4 p 1.2

1.4 1.2

1

1 0.8

0.8

0.6

0.6

0.4

0.4

0.2 0 -1.5

0.2

-1

-0.5

0 1

0.5

1

1.5

-2

-1

0 2

1

1

2 0 0

0.5 0.5 00 0.2 0.4 0.4 0.6 0.6 0.8 0.8

0.5 11.0 1.2 1.2 1.4

1.6

1

2

1

Figure 5.27. Path following results for the Augusti model: (a) geometrically perfect initial configuration and (b) with initial imperfections [29].

17:53

P1: KAE Chapter-05

CUFX159-Virgin

0 521 85648 5

August 14, 2007

References

Rather than detail specific algorithmic features being given here, readers are referred to Doedel [27] and Doedel et al. [28] for more details. But an example is given based on the solution of the equilibrium equations for the Augusti model from Section 5.7. Using the same parameters as those for Figs. 5.21(a) and 5.22(a), auto was used to generate the results shown in Fig. 5.27 [29].

References [1] [2] [3] [4] [5] [6] [7] [8]

[9] [10]

[11] [12] [13] [14] [15] [16]

[17]

[18] [19] [20]

J.M.T. Thompson and G.W. Hunt. A General Theory of Elastic Stability. Wiley, 1973. J.G.A. Croll and A.C. Walker. Elements of Structural Stability. Wiley, 1972. G.J. Simitses. An Introduction to the Elastic Stability of Structures. Prentice Hall, 1976. J.M.T. Thompson and G.W. Hunt. Elastic Instability Phenomena. Wiley, 1984. L.N. Virgin. The dynamics of symmetric postbuckling. International Journal of Mechanical Sciences, 27:235–48, 1985. A.N. Kounadis. Nonlinear dynamic buckling of discrete dissipative or nondissipative systems under step loading. AIAA Journal, 29:280–9, 1991. A.N. Kounadis. Nonlinear dynamic buckling and stability of autonomous structural systems. International Journal of Mechanical Sciences, 35:643–56, 1993. I. Elishakoff, S. Marcus, and J.H. Starnes. On vibrational imperfection sensitivity of Augusti’s model structure in the vicinity of a nonlinear static state. International Journal of Non-Linear Mechanics, 31:229–36, 1996. W.H. Press, B.P. Flannery, S.A. Teukolsky, and W.T. Vetterling. Numerical Recipes in Fortran. Cambridge University Press, 1992. A.C. Walker, J.G.A. Croll, and E. Wilson. Experimental models to illustrate the nonlinear behavior of elastic structures. Bulletin of Mechanical Engineering Education, 10:247–59, 1971. M.A. Souza. Vibration of thin-walled structures with asymmetric post-buckling characteristics. Thin-Walled Structures, 14:45–57, 1992. P.X. Bellini. The concept of snap-buckling illustrated by a simple model. International Journal of Nonlinear Mechanics, 7:634–50, 1972. D.A. Pecknold, J. Ghaboussi, and T.J. Healey. Snap-through and bifurcation in a simple structure. Journal of Engineering Mechanics (ASCE), 111:909–22, 1985. L.N. Virgin. Parametric studies of the dynamic evolution through a fold. Journal of Sound and Vibration, 110:99–109, 1986. J.M.T. Thompson and H.B. Stewart. Nonlinear Dynamics and Chaos, 2nd ed. Wiley, 2002. G. Augusti, V. Sepe, and A. Paolone. An introduction to compound and coupled buckling and dynamic bifurcations. In J. Rondal, editor, Coupled Instabilities in Metal Structures: Theoretical and Design Aspects. Springer-Verlag, 1998, pp. 1–27. N. Challamel. Softening branches of a two-degree-of-freedom system induced by spatial buckling. International Journal of Structural Stability and Dynamics, 6:493–512, 2006. L.N. Virgin and R.H. Plaut. Use of frequency data to predict secondary bifurcation. Journal of Sound and Vibration, 251:919–26, 2002. K. Huseyin and J. Roorda. The loading-frequency relationship in multiple eigenvalue problems. Journal of Applied Mechanics, 38:1007–11, 1971. K. Huseyin. Multiple Parameter Stability Theory and Its Applications. Oxford University Press, 1986.

95

17:53

P1: KAE Chapter-05

CUFX159-Virgin

96

0 521 85648 5

August 14, 2007

Discrete-Link Models [21] R.H. Plaut. Column buckling when support stiffens under compression. Journal of Applied Mechanics, 56:484, 1989. [22] R.H. Plaut. Stability and vibration of a column model with load-dependent support stiffness. Dynamics and Stability of Systems, 6:79–88, 1991. [23] A.P. Seyranian and A.A. Mailybaev. Multiparameter Stability Theory with Mechanical Applications. World Scientific, 2003. [24] T.S. Parker and L.O. Chua. Practical Numerical Algorithms for Chaotic Systems. Springer-Verlag, 1989. [25] S. Wolfram. The Mathematica Book. Cambridge University Press, 1996. [26] MATLAB. User’s guide. Technical report, The Math Works, 1989. [27] E.J. Doedel. AUTO—Software for continuation and bifurcation problems in ordinary differential equations. California Institute of Technology, 1986. [28] E.J. Doedel, A.R. Champneys, T.F. Fairgrieve, Y.A. Kuznetsov, B. Sandstede, and X.J. Wang. Auto97: Continuation and bifurcation software for ordinary differential equations. Technical report, Department of Computer Science, Concordia University, Montreal, Canada, 1997 (available by FTP from ftp.cs.concordia.ca in directory pub/doedel/auto). [29] H. Chen. Nonlinear analysis of post-buckling dynamics and higher order instabilities of flexible structures. Ph.D. dissertation, Duke University, 2004.

17:53

P1: KAE Chapter-06

CUFX159-Virgin

0 521 85648 5

August 14, 2007

6 Strings, Cables, and Membranes

6.1 Introduction In transitioning from discrete to continuous systems, we will naturally encounter partial differential equations; that is, the link models of the previous chapter lead to an algebraic eiegnvalue problem whereas the systems considered in this chapter lead naturally to a differential eigenvalue problem, even though we will find utility in approximations leading back to a discrete description. Although the main focus in this book is the behavior of structures subject to compressive axial load, it is instructive to consider systems subject to tensile loads as an introduction. Because these systems tend to be used in a linear context, they typically do not suffer the instability phenomena associated with buckling. However, they do provide a relatively gentle introduction to the behavior and methods of analysis associated with distributed systems subject to axial loading. They also provide a compelling analogy with an everyday example of the relation between axial loading (tension) and natural frequency (pitch): tuning a stringed musical instrument.

6.2 The Stretched String 6.2.1 The Wave Equation We start by considering the undamped, small-amplitude motion of a stretched string under a tension τ. By considering a small element dx, as shown in Fig. 6.1, we have for horizontal equilibrium τ(x + dx) cos [θ(x + dx)] − τ(x) cos θ(x) = 0,

(6.1)

and because for small slopes cos θ ≈ 1, τ is constant. For vertical equilibrium τ sin [θ(x + dx)] − τ sin θ(x) = ρdx

∂2 w , ∂t2

(6.2)

where ρ is the mass per unit length of the string. Expanding the sine terms in Eq. (6.2) gives ∂2 w ∂θ(x) + · · · + − τ [θ(x) + · · · +] = ρdx 2 , τ θ(x) + dx ∂x ∂t

(6.3) 97

17:54

P1: KAE Chapter-06

CUFX159-Virgin

0 521 85648 5

August 14, 2007

98

Strings, Cables, and Membranes

w

τ(x+dx) θ(x+dx)

w+dw

w

θ(x)

ds

(x) x x

x+dx

Figure 6.1. Forces acting on a segment of a taut string when undergoing transverse vibration.

and because ∂θ/∂x = ∂2 w/∂x2 we obtain the governing equation of (lateral motion) for a string: ∂2 w 1 ∂2 w = , ∂x2 c2s ∂t2

(6.4)

√ where cs = τ/ρ is a constant, and later to be identified with the speed of lateral motion. It also relates to the velocity of wave propagation along the string, although we are primarily interested in transverse (lateral) effects here. Equation (6.4) is a partial differential equation of the type introduced in Chapter 4, is called the wave equation, and occurs in many branches of the physical sciences [1]. A standard approach to solving partial differential equations of this type is based on the separation of variables. This approach will also be used when we consider the dynamic behavior of structures with bending stiffness later. We assume that the displacement of the string can be written as w(x, t) = φ(x)q(t),

(6.5)

which, on substitution in Eq. (6.4) leads to 1 d2 φ 1 1 d2 q = . φ dx2 q c2s dt2

(6.6)

We see that the left-hand side is a function of the position and the right-hand side is a function of time. In this case both sides of the equation must be equal to the same constant, which we set as −(ω/cs )2 (with the negative sign and square chosen with forethought). Now we have the two uncoupled ordinary differential equations, d2 φ + dx2

ω cs

2 φ = 0,

(6.7)

d2 q + ω2 q = 0. dt2

(6.8)

17:54

P1: KAE Chapter-06

CUFX159-Virgin

0 521 85648 5

August 14, 2007

6.2 The Stretched String

99

These linear equations have very familiar solutions ω ω x + B cos x, φ(x) = A sin cs cs

(6.9)

q(t) = C sin ωt + D cos ωt,

(6.10)

where the constants (A, B) and (C, D) are obtained from the boundary and initial conditions [2]. Suppose we have a string that is stretched between fixed points. In this case, the boundary conditions can be written as w(0, t) = 0,

(6.11)

w(L, t) = 0.

(6.12)

Evaluating the spatial part of the solution [Eq. (6.9)] under these circumstances, we get B = 0 from Eq. (6.11), and from Eq. (6.12), ω A sin L = 0. (6.13) cs The solutions to this (in addition to the trivial solution A = 0) are ωn L = nπ, cs

n = 1, 2, . . . ,

(6.14)

that is, there are an infinite number of natural frequencies: nπcs nπ τ ωn = = . L L ρ

(6.15)

And thus we have a linear relation between tension τ and the square of the natural frequencies ωn with corresponding mode shapes, ωn x nπx . (6.16) = sin φn (x) = sin cs L The mode shapes satisfy the conditions of orthogonality described earlier in this book (see Section 4.3). Thus we have an equation for the lateral displacement of the string given by w(x, t) =

∞

qn (t) sin

n=1

=

∞

nπx L

(6.17)

(Cn sin ωn t + Dn cos ωn t) sin

n=1

nπx , L

(6.18)

or, using complex notation, w(x, t) =

∞ n=1

βn eiωn t sin

nπx , L

(6.19)

where the βn are complex (and it is the real part that corresponds to physically meaningful solutions). The constants associated with the temporal part of the solution are

17:54

P1: KAE Chapter-06

CUFX159-Virgin

0 521 85648 5

August 14, 2007

100

Strings, Cables, and Membranes

determined from the initial conditions w(x, 0) and ∂w(x, 0)/∂t and correspond to the Fourier coefficients. As an example, consider a taut string that is displaced by an amount H (not too large) at its center and then released (i.e., with an initial velocity of zero). The string has the initial form 2H x, L 2H w(x, 0) = (L − x), L

w(x, 0) =

0 ≤ x ≤ L/2, (6.20) L/2 ≤ x ≤ L,

and evaluating C and D leads to a response πx 3πx 1 8H cos ω1 t − sin cos ω3 t + · · · + . w(x, t) = 2 sin π L 9 L

(6.21)

That only odd harmonics are excited is due to the symmetric nature of the initial disturbance, which leads to symmetric motion. The resulting triangular wave is dominated by the first (fundamental) mode, as expected. Thus we see that the boundary conditions influence the mode shapes and natural frequencies and the initial conditions determine the contribution of each mode. We observe our familiar linear relation between the square of the natural frequency and the tension. This closed-form solution would not have been available if the tension in the string had not been constant. Later in this book we will use a somewhat similar approach for beams and plates. The major differences are that systems with bending stiffness will tend to lead to higher-order differential equations (and hence more boundary conditions) and the forces of interest will primarily be compressive, which allows for bifurcational phenomena to appear.

6.2.2 Traveling-Wave Solution At this point we might wonder why Eq. (6.4) has the name it does. We now show that an alternative, but equivalent, description can be obtained by writing the general solution in the form w(x, t) = F1 (x − cs t) + F2 (x + cs t).

(6.22)

We can think of the first term on the right-hand side of Eq. (6.22) as representing a wave moving in the positive x direction with constant velocity cs , with the second term corresponding to a similar wave but moving in the opposite direction. In both cases the functions F represent the (non-changing) shape, or profile, of the wave. Because we have already seen that sinusoidal motion is typical in the vibration of strings, let’s consider a wave of the form w(x, t) = A sin

2π (x − cs t), λ

(6.23)

17:54

P1: KAE Chapter-06

CUFX159-Virgin

0 521 85648 5

August 14, 2007

6.2 The Stretched String

101

that is, a wave of amplitude A and wavelength λ traveling in the positive x direction. Introducing the wavenumber k = 1/λ, we can rewrite Eq. (6.23) as w(x, t) = A sin (2πkx − ωt),

(6.24)

in which ω = cs (2π)/λ is the frequency of the wave. However, if we have two equal waves but traveling in opposite directions, we then have w(x, t) = A sin (2πkx − ωt) + A sin (2πkx + ωt),

(6.25)

which can be rewritten as w(x, t) = 2A sin (2πkx) cos (ωt).

(6.26)

Thus we see that the two waves in this case together respresent a standing wave but with an oscillating profile. The two components cancel at those points where x = nλ/2, and these are called node points. If x = λ(2n + 1)/4, then we have a reinforcement, or antinodes. To apply this approach to a specific (finite) string, we note that, for example, for a string stretched between two points we will have nodes at the end points, and thus 2kL = r where r is an integer, and therefore τ cs ωr = 2πkcs = rπ = rπ (6.27) L ρL2 with the frequencies obtained previously. We note that, in the presence of damping, the traveling waves will decay, and although it is possible to solve the wave equation with damping, the solution becomes considerably more involved. 6.2.3 Energy Considerations and Rayleigh’s Principle The wave equation can also be obtained by use of Hamilton’s principle. The kinetic energy of the string is given by L 2 1 ∂w T= ρ dx (6.28) 2 0 ∂t 2 L ∞ 1 nπx = ρ q˙ sin dx (6.29) 2 0 L n=1

=

ρL 4

∞

q˙ 2n ,

(6.30)

n=1

and the potential energy associated with the stretching is given by ⎞ ⎛ 2 L ⎝ 1 + ∂w − 1⎠ dx U=τ ∂x 0 ≈

1 τ 2

0

L

∂w ∂x

(6.31)

2 dx

(6.32)

17:54

P1: KAE Chapter-06

CUFX159-Virgin

0 521 85648 5

August 14, 2007

102

Strings, Cables, and Membranes

1 = τ 2

L 0

∞ nπ

nπx qn cos L L

n=1

2 dx

(6.33)

∞

=

ρL 2 2 ωn qn , 4

(6.34)

n=1

where we have used the fact that, from Eq. (6.15), ωn L 2 τ=ρ . nπ

(6.35)

We can then invoke Hamilton’s principle and, by taking variations in the energy terms, arrive at the wave equation. Assuming there is no external forcing or damping, we also have the conservation of total energy, that is, ∞

E = T+U=

ρL 2 ωn |βn |2 , 4

(6.36)

n=1

where the βn [from Eq. (6.19)] depends on the initial conditions. We can also use an approximate mode shape (although it is not really necessary in this instance as the exact solution is well known), such as a parabola: w = 4H(t)x(L − x)/L2 .

(6.37)

Plugging this into the expression for the kinetic and potential energy expressions [Eqs. (6.28) and (6.31)] and using Rayleigh’s method gives ω2 =

10τ , ρL2

(6.38)

which is less than 1% greater than the exact value [i.e., ω2 = π2 τ/(ρL2 )]. Rayleigh devised a method of improving this result by incorporating an adjustable constant into the displacement function [3]. It is interesting to note that when a string is excited it often exhibits non-planar motion. This type of whirling motion (familiar from a child’s skipping rope) is just one type of complicated motion found in the forced string problem. In this chapter, we have assumed that the tension in the string remains constant during motion. However, the tension must fluctuate, and it is the subtle interaction between longitudinal and transverse motion that underlies much of the interesting (nonlinear) behavior. In fact, the greater the tension in the string, the relatively less the tension changes during motion. This problem will be revisited later when we consider large-amplitude vibration.

6.3 A Suspended Cable A natural extension to the study of the dynamics of a taut string is to consider the behavior of a string that is not taut, but rather sags because of the effect of gravity [4]. Without bending stiffness, it is again the axial load that provides the restoring

17:54

P1: KAE Chapter-06

CUFX159-Virgin

0 521 85648 5

August 14, 2007

6.3 A Suspended Cable

103

ds

dy

g

H dx

V

T l

x d

y

L

Figure 6.2. The cable geometry and the forces acting on a segment.

force for this important class of practical problems. We focus on symmetric cables having suspension points located at the same vertical elevation, and consider only modes of transverse vibration taking place in-plane [5]. If the cable configuration is shallow, then certain simplifying assumptions can be made, but, in general, the sag may have a relatively profound effect on dynamic behavior. Because of the role of gravity, it is convenient to use the coordinate system shown in Fig. 6.2. Assuming the cable is inextensional, we can write the equilibrium in the vertical direction as d dy T = −mg, (6.39) ds ds and in the horizontal direction we have d dx T = 0, ds ds

(6.40)

which can be integrated to give T

dx = H, ds

(6.41)

in which H is the horizontal component of the cable tension. Thus Eq. (6.39) can be rewritten as H

d2 y ds = −mg . 2 dx dx

Using horizontal equilibrium and using ds2 = dx2 + dy2 , we can then write 2 d dy mg dy =− 1+ . dx dx H dx

(6.42)

(6.43)

The solution to Eq. (6.43) is the well-known catenary, and given that the end points of the cable have the same vertical elevation, we can write the solution in terms of the length-to-span ratio [5, 6], L sinh [lmg/(2H)] = , l lmg/(2H)

(6.44)

17:54

P1: KAE Chapter-06

CUFX159-Virgin

0 521 85648 5

August 14, 2007

104

Strings, Cables, and Membranes

and the sag-to-length ratio H d = {cosh [lmg/(2H)] − 1}. l lmg

(6.45)

For a shallow cable (with a small sag), we can expand the hyperbolic functions in Eqs. (6.44) and (6.45) as Taylor series or simplify governing equation (6.42): d2 y mg . =− 2 dx H It can easily be shown that the deflected shape of the cable is then given by x x x l2 mg x 1− = 4d 1 − , y= 2H l l l l

(6.46)

(6.47)

where the vertical sag in the center of the cable d is given by d 1 lmg = . l 8 H

(6.48)

We can now consider oscillations about equilibrium. In addition to an inertia force (by use of D’Alembert’s principle), the added motion w(x, t) will induce a varying horizontal force h(t), and thus the equation of vertical motion is (H + h)

∂2 ∂2 w (y + w) = −mg + m , ∂x2 dt2

(6.49)

which, for small-amplitude motion, can be simplified to m ∂2 w mg h ∂2 w . − = 2 ∂x H dt2 H H

(6.50)

It can also be shown [6] that for inextensional cables (based on linearized theory) l h(t) depends on the displacement function w(x, t) such that 0 wdx = 0. If we then seek harmonic oscillations of the form w(x, t) = w(x) ˜ cos ωt,

h(t) = h˜ cos ωt,

(6.51)

then substituting these into Eq. (6.50) and using β2 = (mω2 )/H results in ∂2 w˜ mg h˜ . + β2 w˜ = 2 ∂x H H

(6.52)

Solving Eq. (6.52) by using the boundary conditions w(0) = w(l) = 0 results in the following solutions: r The symmetric modes: w˜ cos [βn (x − (1/2)l)] h˜ 8Cn 1− , (6.53) = = Cn , 2 d (βn l) cos[(1/2)βn l] H l with 0 wdx ˜ = 0 leading to the lowest root (1/2)β1 l = 4.493. Therefore the lowest (symmetric mode) frequency is ω1 = 8.99(H/ml2 )1/2 .

17:54

P1: KAE Chapter-06

CUFX159-Virgin

0 521 85648 5

August 14, 2007

6.3 A Suspended Cable

105

(a)

(b)

Figure 6.3. The first inextensional symmetric and the first antisymmetric modes.

r The antisymmetric modes: w˜ 2π = Cn sin n x , d l

2π ωn = n l

H m

(6.54)

with h˜ = 0 leading to the lowest (antisymmetric mode) frequency ω1 = 2π(H/ml2 )1/2 . The lowest symmetric and antisymmetric modes are shown in Figs. 6.3(a) and 6.3(b), respectively. In the former case, it is interesting to see that the wavelength of the mode is shorter than the span and the cable straightens out near the supports. In the latter case, we recognize the form from the taut string, as one side moves up and the other side moves down. It is also interesting to note that these antisymmetric modes would not be influenced by any elasticity of the cable (whereas the symmetric modes may be quite influenced). For example, by introducing the parameter λ, accounting for the elastic flexibility of the cable, EA 8d 2 λ2 = , (6.55) H l we can incorporate this effect into the preceding theory [6]. As the cable flexibility decreases (λ → ∞), we approach the inextensible case already considered. At the other end of the spectrum, we have λ → 0 and a frequency that tends to decrease, and reaches the value of a taut string ω = π H/ml2 . This effect is summarized in Fig. 6.4. We observe that when λ2 < 4π2 , the lowest mode is symmetric (with no internal nodes), but at the transition point λ2 = 4π2 the in-plane frequencies are equal. This type of modal exchange will also be encountered later in the behavior of a pin-ended column with an elastic restraint at midspan, as well as other structural systems in which axial load is a consideration. When λ2 > 4π2 , the lowest mode

17:54

P1: KAE Chapter-06

CUFX159-Virgin

0 521 85648 5

August 14, 2007

106

Strings, Cables, and Membranes (a) 2

3

2

λ =

1

symmetric

10 0

10 1

(b)

3

2

antisymmetric

1

0

4π2

10 2

10 3

0

0.025

0.05

0.075

d/L

0.1

Figure 6.4. Dynamic response of an extensible cable with sag: (a) two lowest natural frequencies and (b) experimental verification (adapted from [7]).

is antisymmetric. Part (b) of Fig. 6.4 also shows this crossover of modes but now as a function of the sag d/l, and some experimental data points are superimposed [6, 7]. The lowest frequency here corresponds to a sway mode. Thus we see that the dynamic response of suspended cables depend on a variety of factors including sag (and hence mass), vertical distance between the support points, flexibility of the cable itself, and so on. The symmetric modes of vibration of the cable are heavily influenced by the cable stiffness (characterized by λ2 ) and thus depend on sag and cable flexibility. 6.3.1 The Hanging Chain The oscillations of a flexible cable suspended vertically from a single fixed point represent a problem that occupies an important place in the historical development of mechanics [8, 9]. It was one of the first systems in which the normal modes of vibration were identified and provided an initial motivation for the development of Bessel functions [10]. In a gravitational field, the weight of a vertically suspended slender beam becomes increasingly important, that is, as the length of the beam increases the bending stiffness becomes negligible in comparsion with gravitational effects. The hanging chain has no bending stiffness. At the opposite end of the flexibility spectrum is the rigid-arm pendulum. Rather than the classical analysis being described here (we shall of course focus on beams in bending later) a relatively simple approximate analysis based on Rayleigh’s method is shown. Assuming the origin is placed at the top end of a vertically hanging chain, of length L, a reasonable first-mode shape in terms of lateral deflection w is given by 2 x x w=Q , (6.56) +β L L in which x is measured downward from the top and β is a constant to be determined. Note that, in contrast to a hanging beam to be considered later, a linear term is

17:54

P1: KAE Chapter-06

CUFX159-Virgin

0 521 85648 5

August 14, 2007

6.3 A Suspended Cable

107

Figure 6.5. The lowest four vibration mode shapes for an experimental hanging beam.

included to allow for a nonzero slope at the clamped end. The kinetic energy is 1 m 2

T=

L 0

w˙ 2 dx =

1 ˙ 2 (10 + 15β + 6β2 ). mLQ 60

(6.57)

1 mgQ2 (3 + 4β + 2β2 ), 12

(6.58)

The potential energy is V=

1 mg 2

0

L x

w2 dxdx =

0

and application of Rayleigh’s method (see Section 3.3) gives 5(3 + 4β + 2β2 ) g ω = . 10 + 15β + 6β2 L 2

(6.59)

The value of β (= 0.289206) that minimizes the frequency results in ω = √ 1.2025 g/L. Because of the extremum nature of Rayleigh’s method, we know that the lower the frequency estimate is, the closer it will be to the exact answer. How√ ever, the exact value (obtained in this case with Bessel functions) is 1.2025 g/L [11], and thus the mode shape described in Eq. (6.56) is obviously very accurate. It is in√ teresting to note that for a rigid bar of the same length the frequency is 1.22474 g/L [12]. Figure 6.5 shows some experimental snapshots of a hanging axisymmteric chain suspended from a spinning shaft. These mode shapes correspond closely to those obtained with Bessel functions for bending motion with nodal points easily observed.

17:54

P1: KAE Chapter-06

CUFX159-Virgin

0 521 85648 5

August 14, 2007

108

Strings, Cables, and Membranes

y

a

y

S

W 22 S S w

b

x 0

x S

Figure 6.6. (a) A rectangular stretched membrane and (b) the (2,2) mode of vibration.

6.4 A Rectangular Membrane A stretched membrane may exhibit transverse vibrations in much the same way as the string, but we now need two dimensions to describe the geometry [2, 13]. We take a brief look at the simple case of a rectangular membrane of uniform thickness, with mass density ρ, which is stretched such that the tension S can be assumed to be constant over the membrane. A schematic of the membrane is shown in Fig. 6.6(a). The lateral deflection is w, and the membrane has length a in the x direction and width b in the y direction. After harmonic motion is assumed, the governing equation is given by ∇ 2 W(x, y) + β2 W(x, y) = 0,

(6.60)

where β2 = ρω2 /S and ∇ 2 is the Laplacian [2], that is, ∇ 2 = ∂2 /∂x2 + ∂2 /∂y2 , which we shall also make use of in the chapter on plates. We can separate variables, apply the boundary conditions, and obtain the natural frequencies 2 S m 2 n ωmn = π , m, n = 1, 2, . . . . (6.61) + ρ a b The corresponding (normalized) mode shapes are given by Wmn = √

nπy 2 mπx sin , sin a b ρab

m, n = 1, 2, . . . .

(6.62)

An example of themode corresponding to the fourth lowest vibration (with frequency ω22 = 2π (S/ρ)(1/a2 + 1/b2 ) is shown in Figure 6.6(b). Hence we again see that the square of the natural frequencies increases linearly with the tension. The lowest frequency for a square membrane is thus given by ω = π 2S/(ρa2 ). Alternatively, we can write down the kinetic energy associated with the vibration of the membrane as a b 1 T= ρ w˙ 2 dxdy, (6.63) 2 0 0

17:54

P1: KAE Chapter-06

CUFX159-Virgin

0 521 85648 5

August 14, 2007

References

109

and the change in potential energy that is due to the tension as the membrane deflects as a b 2 2 ∂w 1 ∂w dxdy. (6.64) V= S + 2 0 0 ∂x ∂y The deflection can be represented by the function w=

∞ ∞

φmn (t) sin

m=1 n=1

nπy mπx sin , a b

(6.65)

which is then used to evaluate the energy terms ∞

T=

∞

1 φ˙ 2mn ρab 8

(6.66)

m=1 n=1

and ∞

∞

1 V = Sab φ2mn 8 m=1 n=1

mπ a

2

+

nπ b

2 .

We again use Lagrange’s equation (2.35) to obtain the equations of motion 2 2 m S n ¨φmn + π2 φmn = 0, + m, n = 1, 2, . . . , ρ a b

(6.67)

(6.68)

with a set of natural frequencies obtained previously [Eq. (6.61)]. Later, we shall compare this behavior with that of systems in which there is also bending stiffness, that is, plates. It turns out that the analysis of membranes depends very much on the shape of the boundary. For example, a circular membrane is more conveniently analyzed by use of polar coordinates and results in Bessel functions [10], and an irregularly shaped membrane would typically require a FEA. The statements made earlier regarding the large-amplitude motion of strings also apply to membranes but we now move on to consider the much wider class of problem in which the structure subject to axial loading possesses bending stiffness and the axial loads are often compressive.

References [1] [2] [3] [4] [5] [6] [7]

R. Courant and D. Hilbert. Methods of Mathematical Physics. Wiley Classics Library, 1989. L. Meirovitch. Principles and Techniques of Vibrations. Prentice Hall, 1997. Lord Rayleigh (John William Strutt). The Theory of Sound. Dover, 1945. H.M. Irvine and T.K. Caughey. The linear theory of free vibrations of a suspended cable. Proceedings of the Royal Society of London Series A, 341:299–315, 1974. S. Krenk. Mechanics and Analysis of Beams, Columns and Cables. Springer, 2001. H.M. Irvine. Cable Structures. MIT Press, 1981. S.E. Ramberg and O.M. Griffin. Free vibration of taut and slack marine cables. Journal of the Structural Division, Proc. ASCE, 103:2079–92, 1977.

17:54

P1: KAE Chapter-06

CUFX159-Virgin

110

0 521 85648 5

August 14, 2007

Strings, Cables, and Membranes [8] H. Lamb. Higher Mechanics. Cambridge University Press, 1929. [9] D. Yong. Strings, chains, and ropes. SIAM Review, 48:771–81, 2006. [10] G.N. Watson. A Treatise on the Theory of Bessel Functions. Cambridge University Press, 1966. [11] H. Lamb. The Dynamical Theory of Sound. Arnold, 1910. [12] R.D. Blevins. Formulas for Natural Frequencies and Mode Shapes. Van Nostrand Rheinhold, 1979. [13] A.D. Dimarogonas. Vibration for Engineers. Prentice Hall, 1996.

17:54

P1: KAE Chapter-07

CUFX159-Virgin

0 521 85648 5

August 14, 2007

7 Continuous Struts

7.1 Introduction In this chapter, we consider the dynamics of a thin elastic strut including axial effects, arising primarily from one of two situations: r the axial load is applied externally (including postbuckling), or r deformation is sufficient to cause coupling between axial and bending behavior (the membrane effect). In the first section, attention is focused on a traditional approach to setting up the equations of motion (by means of D’Alembert’s principle) for the simple case based on engineering beam theory (Euler–Bernoulli) with the addition of axial loads. The resulting partial differential equation of motion is then separated into temporal and spatial ordinary differential equations and the response analyzed for various magnitudes of the axial load [1, 2]. Then an energy approach is used together with Rayleigh’s method [3]. In this case, additional terms are retained in the potential energy to allow postbuckled effects to be analyzed [4] and the effect of initial geometric imperfections are included. An alternative approach is developed based on a simple application of Hamilton’s principle, and in this instance stretching effects are also included and a solution developed by use of Galerkin’s method. This approach will be similar to that used in the previous chapter on strings but now bending strain energy enters into the analysis (as well as compressive axial loading). In the final part of the chapter we consider the dynamics of struts that are loaded by gravity through self-weight. The next chapter will then continue the study of axially loaded members but with the scope opened to include a wider class of problem.

7.2 Basic Formulation In this section, we develop the governing equation of motion for a thin, elastic, prismatic beam subject to a constant axial force. In Fig. 7.1(a) a schematic of the beam is shown. It has mass per unit length m, constant flexural rigidity EI, and is subject to an axial load P. The length is L, the coordinate along the beam is x, and the lateral (transverse) deflection is w(x, t). In part (b) is shown an element of the beam between locations x and x + x, which is subject to the D’Alembert forces R(x, t) = m∂2 w/∂t2 .

(7.1) 111

17:55

P1: KAE Chapter-07

CUFX159-Virgin

0 521 85648 5

August 14, 2007

112

Continuous Struts

(a) L x

EI, m

P

w(x,t)

(b)

x w1

M

Figure 7.1. A schematic of a thin elastic beam.

w2

R(x,t) x

P M+ M

0

S

P Q(x,t) x x/2

F(x,t) x

S+ S

At this point, we make the assumption that the angle through which the section rotates (w/x) is small and we neglect axial and rotary inertia [5]. Taking moments about 0 we get x x − (S + S) + M − (M + M) + Pw = 0, 2 2 and as x → 0 we have ∂M ∂w S− +P = 0. ∂x ∂x Summing forces in the vertical direction we also have S

(S + S) − S + F (x, t)x − R(x, t)x = 0,

(7.2)

(7.3)

(7.4)

and again passing to the limit x → 0, we have ∂2 w ∂S + F (x, t) − m 2 = 0. ∂x ∂t

(7.5)

Differentiating Eq. (7.3) we obtain ∂S ∂2 M ∂2 w − + P 2 = 0, 2 ∂x ∂x ∂x

(7.6)

and eliminating ∂S/∂x between Eqs. (7.5) and (7.6) gives m

∂2 w ∂2 M ∂2 w − + P = F (x, t), ∂t2 ∂x2 ∂x2

(7.7)

and finally, using the familiar expression from engineering beam theory [6, 7], ∂2 w = −M, ∂x2

(7.8)

∂2 w ∂2 w ∂4 w + P + m = F (x, t). ∂x4 ∂x2 ∂t2

(7.9)

EI we arrive at the governing equation: EI

17:55

P1: KAE Chapter-07

CUFX159-Virgin

0 521 85648 5

August 14, 2007

7.2 Basic Formulation

113

This linear partial differential equation is a key equation in this book and can be rewritten in shorthand as EIw + Pw + mw¨ = F (x, t)

(7.10)

EIwxxxx + Pwxx + mwtt = F (x, t),

(7.11)

or

where derivatives are signified by primes and subscripts in Eqs. (7.10) and (7.11), respectively, and, given appropriate boundary conditions at x = 0, L and initial conditions, can be solved by use of standard methods as outlined in Chapter 4. Before we consider the specific aspects of the solution, the utility of nondimensionalizing the governing equation of motion is mentioned. Nondimensionalization is useful because it reduces the number of parameters and allows a more consistent comparison of behavior. In this chapter, we are focused on the unforced problem (F = 0), and, by defining the following parameters, (7.12) x¯ = x/L, w¯ = w/L, t¯ = t EI/(mL4 ), p = PL2 /(EI), we can rewrite Eq. (7.9) as ∂4 w¯ ∂2 w¯ ∂2 w¯ + p + = 0. ∂x¯ 4 ∂x¯ 2 ∂t¯2

(7.13)

Throughout this book we shall look at solutions to this type of equation and often incorporate nonlinearities into the analysis, in which case a variety of approximate techniques (of the type outlined in Chapter 4) will be utilized. Sometimes the free parameters will be chosen such that they are further normalized (e.g., p might be related to a critical buckling load). However, we start by looking at the simple freevibration case for which there is good access to analytical solutions. 7.2.1 The Response We initially consider the free-vibration problem and assume that F = 0 and that the motion consists of a function W(x) that varies with Y(t) such that w(x, t) = W(x)Y(t).

(7.14)

Placing this back into Eq. (7.9) leads to EI

d4 W d2 W d2 Y Y + P Y = −mW . dx4 dx2 dt2

(7.15)

That is, after dividing by WY (which is not zero if w is not zero), we have an equation separated into spatial (x) and temporal (t) parts, and thus the ratio on each side must be a constant (which we label −ω2 ) m d2 Y EI d4 W P d2 W = − − = −ω2 . Y dt2 W dx4 W dx2

(7.16)

17:55

P1: KAE Chapter-07

CUFX159-Virgin

0 521 85648 5

August 14, 2007

114

Continuous Struts

Thus we have two ordinary differential equations—one in space that we can solve by using the appropriate boundary conditions and that describes the mode shapes (which are half sine waves in the simply supported case); the other in time that we can solve with the appropriate initial conditions and contains the frequency information (simple harmonic motion). It is not uncommon for the vibration of mechanical systems to be dominated by the mode with the lowest frequency, but we consider the full system at first before developing approximate techniques. 7.2.2 The Temporal Solution We might expect the second-order ordinary differential equation in time to have oscillatory solutions (given positive values of flexural rigidity, etc.). However, we anticipate that the dependence of the form of the temporal solution will depend on the magnitude of the axial load [8, 9]. To be a little more specific, before going on to consider the more general boundary conditions, let us suppose we have ends that are pinned (no deflection and no resistance to rotation), that is, the deflection (w) and bending moment (−EI∂2 w/∂2 x) are zero at x = 0 and x = L. In the general case, we would assume an exponential form for the solution, but with these relatively convenient boundary conditions, we can take w(x, t) =

∞

Y(t) sin

n=1

nπx . L

(7.17)

We can obtain the temporal part of the solution by assuming Yn (t) = An eiωn t ,

(7.18)

and substituting into Eq. (7.9) leads to 2 2 ∞ n2 π2 nπ nπx iωn t 2 EI 2 − P e − mωn An sin = 0. L L2 L

(7.19)

n=1

Clearly, the term in the square brackets must vanish for a nontrivial solution so that PL2 n4 EIπ4 ω2n = 1 − . (7.20) mL4 n2 EIπ2 If we define the following parameters p n = n2 EIπ2 /L2 , Eq. (7.20) becomes

ω¯ 2n = n4 EIπ4 /mL4 ,

ωn = ±ω¯ n 1 − p, ¯

(7.21)

(7.22)

where p¯ = P/p n , and we see that the nature of the solution depends crucially on the discriminant. Making use of the Euler identities, we consider the following four

17:55

P1: KAE Chapter-07

CUFX159-Virgin

0 521 85648 5

August 14, 2007

7.2 Basic Formulation

115

cases, in which An and Bn are constants obtained from the initial conditions: r If p¯ = 0, then Yn (t) = An cos ω¯ n t + Bn sin ω¯ n t,

(7.23)

and we observe simple harmonic motion, a familiar result from linear vibration theory. r If 0 < p¯ < 1, then Yn (t) = An cos ωn t + Bn sin ωn t,

(7.24)

where ωn is given by Eq. (7.22) and is real, and simple harmonic motion results. Any perturbation will induce oscillatory motion about equilibrium. Assuming no damping, the response neither grows nor decays. This includes the response for a tensile axial load, that is, p¯ < 0. r If p¯ = 1, Eq. (7.22) has a double-zero root, and then the solution can be written as Yn (t) = An + Bn t,

(7.25)

and the motion grows linearly with time (this is a special case). r If p¯ > 1, the roots of (7.22) are purely imaginary and then Yn (t) = An cosh ω¯ n t + Bn sinh ω¯ n t,

(7.26)

and the motion grows exponentially with time. Typical examples of these cases are shown in Fig. 7.2(a) in which the natural frequency in the absence of axial load was taken as unity [10], and hence a natural period of 2π. Also shown in this figure are the stability of equilibrium, part (b), and effective natural frequency (squared), part (c), as a function of axial load. Let us focus attention on the lowest natural frequency and its corresponding mode (n = 1). With no axial load (p¯ = 0) we obtain ω1 = ω¯ 1 . However, as the axial load increases the natural frequency decreases according to Eq. (7.22), that is, we observe a linear relationship between the magnitude of the axial load and the square of the natural frequency [see Fig. 7.2(c)]. Any nonzero initial conditions result in bounded motion, and we may consider this to be a stable situation (at least in the sense of Lyapunov). When p¯ → 1, ω1 vanishes and the solution ceases to be oscillatory [the linearly increasing (constant-velocity) solution shown in Fig. 7.2]. Any inevitable perturbation will cause the system to become unstable. This type of instability is monotonic because, locally, the deflections grow in one direction (determined by the initial conditions). This type of behavior is sometimes referred to as divergence. The higher modes (n > 1) will exhibit oscillations (in theory) but the important practical information has been gained, that is, typical behavior is dominated by the lowest mode.

17:55

P1: KAE Chapter-07

CUFX159-Virgin

0 521 85648 5

August 14, 2007

116

Continuous Struts 10 (a)

Y

_

p >1

8 6

_

p =1

4 2

_

p=0 _

0 −2

p 10, the beamlike bending modes ¨ start to dominate, and in this region Flugge shell theory provides more accurate solutions [46]. In the other extreme, the behavior tends to be more platelike. Other studies have included the effects of initial imperfections [53] and composites [54, 55]. Equation (10.101) thus gives the familiar-looking result shown in Fig. 10.16 in which both the frequency and axial load are nondimensionalized. The linearity in the axial-force–frequency (squared) relation was encountered earlier in a variety of systems for which the buckling and vibration modes were similar. It is the relative simplicity of this relation that provides compelling motivation for nondestructive testing purposes, that is, using frequencies to predict buckling, and this is the subject of the next chapter.

18:0

P1: KAE Chapter-10

CUFX159-Virgin

0 521 85648 5

August 14, 2007

212

Plates

1

−Nx

0 n = 11

0

1

2

Figure 10.16. Fundamental natural frequency for a specific cylinder as a function of axial load.

Other Issues

The stiffening effect that is due to a spinning circular plate [56] is a problem of practical interest and is related to the spinning beam analysis in Chapter 7. Certain plate problems are conveniently solved by the methods of Lagrange multipliers [57] and finite differences [58]. Later chapters will also revisit shells in terms of step loading and parametric excitation.

References [1]

S.P. Timoshenko and S. Woinowsky-Krieger. Theory of Plates and Shells, 2nd ed. McGraw-Hill, 1968. [2] A.W. Leissa. Vibration of plates. Technical Report SP–160, NASA, 1969. [3] T. von Karman, E.E. Sechler, and L.H. Donnell. The strength of thin plates in compression. Transactions of ASME, 54:53–7, 1932. [4] S. Levy. Bending of rectangular plates with large deflections. Technical Report 737, NACA, 1942. [5] S. Levy. Buckling of rectangular plates with built-in edges. Journal of Applied Mechanics, 9:171–4, 1942. [6] H.-N. Chu and G. Herrmann. Influence of large amplitudes on free flexural vibrations of rectangular elastic plates. Journal of Applied Mechanics, 23:532–40, 1956. [7] S.F. Bassily and S.M. Dickinson. Buckling and lateral vibration of rectangular plates subject to in-plane loads—a Ritz approach. Journal of Sound and Vibration, 24:219–39, 1972. [8] S.M. Dickinson. The buckling and frequency of flexural vibration of rectangular isotropic and orthotropic plates using Rayleigh’s method. Journal of Sound and Vibration, 61:1–8, 1978. [9] C.F. Ng and R.G. White. Dynamic behavior of postbuckled isotropic plates under in-plane compression. Journal of Sound and Vibration, 120:1–18, 1988. [10] G.H. Bryan. On the stability of a plane plate under thrusts in its own plane with applications to the buckling of the sides of a ship. Proceedings of the London Mathematical Society, 22:54–67, 1891.

18:0

P1: KAE Chapter-10

CUFX159-Virgin

0 521 85648 5

August 14, 2007

References [11] S. Ilanko. Vibration and post-buckling of in-plane loaded rectangular plates using a multiterm Galerkin’s method. Journal of Applied Mechanics, 69:589–92, 2002. [12] A.C. Ugural. Stresses in Plates and Shells. McGraw-Hill, 1999. [13] G. Herrmann and J. Shaw. Vibration of thin shells under initial stress. Journal of Engineering Mechanics, 91:37–59, 1965. [14] J.F. Doyle. Nonlinear Analysis of Thin-Walled Structures. Springer, 2001. [15] K.K. Kapur and B.J. Hartz. Stability of plates using the finite element method. Journal of Engineering Mechanics, 92:177–95, 1966. [16] R.G. Anderson, B.M. Irons, and O.C. Zienkiewicz. Vibration and stability of plates using finite elements. International Journal of Solids and Structures, 4:1031–55, 1968. [17] B.A. Boley and J.H. Weiner. Theory of Thermal Stresses. Wiley, 1960. [18] D.J. Johns. Thermal Stress Analysis. Pergamon, 1965. [19] J. Marcinowski. Postbuckling behaviour of rectangular plates in axial compression. Archives of Civil Engineering, 45:275–88, 1999. [20] K.D. Murphy. Theoretical and experimental studies in nonlinear dynamics and stability of elastic structures. Ph.D. dissertation, Duke University, 1994. [21] R.E. Kielb and L.S. Han. Vibration and buckling of rectangular plates under in-plane hydrostatic loading. Journal of Sound and Vibration, 70:543–55, 1980. [22] R.E. Kielb. Thermal buckling of uniform rectangular plates. Technical Report, U.S. Air Force Wright-Patterson, ASD-TR-75-37, 1976. [23] T.R. Tauchert. Thermally induced flexure, buckling, and vibration of plates. Applied Mechanics Reviews, 44:347–60, 1991. [24] K.D. Murphy, L.N. Virgin, and S.A. Rizzi. The effect of thermal prestress on the free vibration characteristics of clamped rectangular plates: Theory and experiment. Journal of Vibration and Acoustics, 119:243–9, 1997. [25] M. Stein. Loads and deformation of buckled rectangular plates. Technical Report R–40, NASA, 1959. [26] D.G. Schaeffer and M. Golubitsky. Boundary conditions and mode jumping in the buckling of rectangular plates. Communications in Mathematics and Physics, 69:209–36, 1979. [27] R. Maaskant and J. Roorda. Mode jumping in biaxially compressed plates. International Journal of Solids and Structures, 29:1209–19, 1991. [28] H. Chen and L.N. Virgin. Finite element analysis of postbuckling dynamics in plates: Part I: An asymptotic approach. International Journal of Solids and Structures, 43:3983–4007, 2006. [29] H. Chen and L.N. Virgin. Finite element analysis of postbuckling dynamics in plates: Part II: A nonstationary analysis. International Journal of Solids and Structures, 43:4008–27, 2006. [30] K.D. Murphy, L.N. Virgin, and S.A. Rizzi. Characterizing the dynamic response of a thermally loaded, acoustically excited plate. Journal of Sound and Vibration, 196:635–58, 1996. [31] K.D. Murphy, L.N. Virgin, and S.A. Rizzi. Experimental snap-through boundaries for acoustically excited, thermally buckled plates. Experimental Mechanics, 36:312–7, 1996. [32] L.N. Virgin. Parametric studies of the dynamic evolution through a fold. Journal of Sound and Vibration, 110:99–109, 1986. [33] R.V. Southwell. On the analysis of experimental observations in problems of elastic stability. Proceedings of the Royal Society of London, 135A:601–16, 1932. [34] K.D. Murphy, L.N. Virgin, and S.A. Rizzi. Free vibration of thermally loaded panels including initial imperfections and post-buckling effects. Technical Memorandum 109097, NASA, 1994.

213

18:0

P1: KAE Chapter-10

CUFX159-Virgin

214

0 521 85648 5

August 14, 2007

Plates [35] M. Uemura and O. Byon. Secondary buckling of a flat plate under uniaxial compression – Part 1: Theoretical analysis of simply supported flat plate. International Journal of Non-Linear Mechanics, 12:355–70, 1977. [36] E. Riks, C.C. Rankin, and F.A. Brogan. On the solution of mode jumping phenomena in thin-walled shell structures. Computer Methods in Applied Mechanics and Engineering, 36:59–92, 1996. [37] H. Troger and A. Steindl. Nonlinear Stability and Bifurcation Theory: An Introduction for Engineers and Applied Scientists. Springer-Verlag, 1991. [38] P.R. Everall and G.W. Hunt. Mode jumping in the buckling of struts and plates: A comparative study. International Journal of Non-Linear Mechanics, 35:1067–79, 2000. [39] G.W. Hunt and P.R. Everall. Arnold tongues and mode-jumping in the supercritical post-buckling of an archetypal elastic structure. Proceedings of the Royal Society of London A, 445:125–40, 1999. [40] T. Nakamura and K. Uetani. The secondary buckling and post-buckling behaviors of rectangular plates. International Journal of Mechanical Sciences, 21:265–86, 1979. [41] E.J. Doedel. AUTO – Software for continuation and bifurcation problems in ordinary differential equations. California Institute of Technology, 1986. [42] E.J. Doedel, A.R. Champneys, T.F. Fairgrieve, Y.A. Kuznetsov, B. Sandstede, and X.J. Wang. Auto97: Continuation and bifurcation software for ordinary differential equations. Technical Report, Department of Computer Science, Concordia University, Montreal, Canada, 1997 (available by FTP from ftp.cs.concordia.ca in directory pub/doedel/auto). [43] W.J. Supple. On the change in buckle pattern in elastic structures. International Journal of Mechanical Sciences, 10:737–45, 1968. [44] J.M.T. Thompson and H.B. Stewart. Nonlinear Dynamics and Chaos, 2nd ed. Wiley, 2002. [45] H. Chen. Nonlinear analysis of post-buckling dynamics and higher order instabilities of flexible structures. Ph.D. dissertation, Duke University, 2004. [46] A.W. Leissa. Vibration of shells. Technical Report SP-288, NASA, 1973. [47] E.H. Dowell. Aeroelasticity of Plates and Shells. Noordhoff, 1975. [48] T. von Karman and H.S. Tsien. The buckling of thin cylindrical shells under axial compression. Journal of the Aeronautical Sciences, 8:303–12, 1941. [49] L.H. Donnell. A new theory for the buckling of thin cylinders under axial compression and bending. Transactions of ASME, 56:796–806, 1934. [50] R.D. Blevins. Formulas for Natural Frequencies and Mode Shapes. Van Nostrand Rheinhold, 1979. [51] K. Forsberg. A review of analytical methods used to determine the modal characteristics of cylindrical shells. NASA Report CR-613, Lockheed Aircraft Company, CA, September 1966. [52] S.B. Batdorf. A simplified method of elastic-stability analysis for thin cylindrical shells. Technical Report 874, NACA, 1947. [53] A.E. Armenakas. Influence of initial stress on the vibrations of simply supported circular cylindrical shells. AIAA Journal, 2:1607–12, 1964. [54] H.S. Shen. Thermomechanical post-buckling analysis of imperfect laminated plates using a higher-order shear-deformation theory. Computers and Structures, 66:395–409, 1998.

18:0

P1: KAE Chapter-10

CUFX159-Virgin

0 521 85648 5

August 14, 2007

References [55] C.-S. Chen and C.-P. Fung. Non-linear vibration of initially stressed hybrid composite plates. Journal of Sound and Vibration, 274:1013–29, 2004. [56] R.G. Parker and C.D. Mote. Tuning of the natural frequency spectrum of a circular plate by in-plane stress. Journal of Sound and Vibration, 145:95–110, 1991. [57] J.H. Ginsberg. Advanced Engineering Dynamics. Cambridge University Press, 1995. [58] F. Bleich. Buckling Strength of Metal Structures. McGraw-Hill, 1952.

215

18:0

P1: KAE Chapter-11

CUFX159-Virgin

0 521 85648 5

August 14, 2007

11 Nondestructive Testing

11.1 Introduction Previous chapters have repeatedly illustrated the often well-defined relation between axial load and natural frequency. In some cases, for example, a simply supported strut, the equivalence of the buckling and vibration modes results in an exactly linear relation between the axial load (providing it is less than critical) and square of the effective natural frequency: ω2 P =1− . 2 Pcr ω0

(11.1)

In other cases, this relation is very nearly linear. For example, consider a simple cantilever. The fundamental frequencies in bending have the mode shapes λi x λi x λi x λi x W(x) = cosh − cos − σi sinh − sin , (11.2) L L L L with σ1 = 0.7341, λ1 = 1.8751 for the lowest mode and corresponding frequency [i.e., ω1 = 3.516 EI/(mL4 )]. The buckling mode for a cantilever with an end load is given by πx W(x) = 1 − cos , (11.3) 2L with a critical load of Pcr = π2 EI/(4L2 ). These normalized shapes are plotted in Fig. 11.1. They are close, but unlike the pinned–pinned (and some sliding boundary conditions) case, they are not equal. However, also plotted in this figure (as the dashed curve) is the buckling mode shape corresponding to the cantilever subject to self-weight. This shape is computed numerically and the critical parameter was established as hcr = 1.986 in Section 7.9 (and equivalent to α = −7.837). This is much closer to the fundamental mode of vibration, and in fact, the difference between them is never more than 1%. The equivalence of the vibration and buckling mode shapes results in the linear relation between axial load and frequency, that is, the extent to which the vibration mode shape is changed by axial loading. Thus we have a frequency (squared) versus load relation that is closer to linearity for the vibrations of a cantilever subject to self-weight than an end load. However, even for the end-loaded cantilever case a simple use of abaqus shows that when (P/Pcr ) = 0.51875 we obtain a lowest natural 216

18:2

P1: KAE Chapter-11

CUFX159-Virgin

0 521 85648 5

August 14, 2007

11.1 Introduction

217

1 0.8

Buckling mode (self-weight)

0.6

First vibration mode

0.4 0.2 0

0

Buckling mode (end load)

0.2

0.4

0.6

0.8

1

Figure 11.1. Comparison of vibration and buckling mode shapes for a uniform cantilever.

frequency of (ω/ω0 )2 = 0.43804 compared with an estimate of (ω/ω0 )2 = 0.41825 suggested by Eq. (11.1). Thus Eq. (11.1) brings the possibility of using dynamics as a means of assessing axial-load effects, including the prediction of buckling [1–3]. In static buckling tests it is often unavoidable that specimens are destroyed during the experimental procedure (often the result of plastic deformation during large deflections). The Southwell plot is a related static approach that also exploits a linear extrapolation to predict buckling nondestructively [4]. Correlation studies between dynamic response and stiffness are also used to determine the actual boundary conditions as well [5, 6]. The simplicity of this relation can be used to nondestructively test axially loaded slender structural elements through monitoring of dynamic response [7–12].

11.1.1 The Southwell Plot In Eq. (7.57) we saw how a small initial geometric imperfection tended to amplify the lateral deflections of a strut, especially as the buckling load is approached. Suppose we have a simply supported beam as shown in Fig. 11.2(a). We can measure the lateral deflections from the initially bent configuration, w0 , but here we measure the total lateral deflection, w, from the straight configuration. If we assume the initial deflection is in the form of a half-sine wave of amplitude Q0 , we can plot the amplification effect [Eq. (7.57)] as shown in Fig. 11.2(b). This, of course, assumes small deflections. However, in an experimental context what we would actually measure would typically be the lateral deflection over and above the initial deflection, which we can call δ = w − w0 and thus (at the midpoint of the strut) δ=

Q0 P/PE − Q0 = Q0 . 1 − P/PE 1 − P/PE

(11.4)

Equation (11.4) can be arranged in the form δ δ Q0 = + , P PE PE

(11.5)

18:2

P1: KAE Chapter-11

CUFX159-Virgin

0 521 85648 5

August 14, 2007

218

Nondestructive Testing (a) w w0

P

P

P/PE

(b)

0.8

(c)

/P

1/ PE

0.6

}

0.4 0.2

Q 0 /PE

Q 0 = 0.05 0 0.1

0

0.2

0.3

0.4

0.5

Q

Figure 11.2. The Southwell plot: (a) strut geometry with an initial imperfection, (b) axial-load– lateral-deflection relation, and (c) Southwell plot.

so that if we plot δ/P as a function of δ we get a straight line in which the intercept is given by Q0 /PE and the slope is given by 1/PE . Southwell [4] recognized the usefulness of this approach to determine both the critical load and initial imperfection, and this is shown schematically in Fig. 11.2(c). A Southwell plot based on experimental data is shown in Fig. 11.3 [13]. Here, the data suggest a critical load (slope) in the vicinity of 87 N and an imperfection of ≈ 0.05 or about 3 deg; values not unreasonable when compared with the data presented in Fig. 5.10(c). Although there are limitations to this approach, the key utility here is that the linear relation allows for extrapolation. This provides some

δ (radian)

0.4

0.2 δ/P (radian N−1) ε = 0.05

0

0.002

0.004

0.006

Figure 11.3. The Southwell plot obtained with experimental data taken from the system shown in Fig. 5.9. Adapted from Croll and Walker [13].

18:2

P1: KAE Chapter-11

CUFX159-Virgin

0 521 85648 5

August 14, 2007

11.1 Introduction

219

(a)

Figure 11.4. The effect of axial-load direction on natural frequencies.

(b)

1

g

motivation for exploring related concepts in the dynamic testing of structures, in which vibration testing in situ is a well-established procedure (e.g., in structural health monitoring [14]).

11.1.2 Examples The relation between axial load and lateral vibrations and its potential use for nondestructive evaluation purposes goes back to Sommerfeld [15]. He made the simple observation that the fundamental natural frequencies of the two systems shown in Fig. 11.4 were quite different (with ω2 > ω1 ). He concluded that the greater the compressive stress, the lower the natural frequency of lateral vibration. With tensile stress, an increase in natural frequency was observed. Furthermore, in the former case it was noted that the frequency dropped to zero as the compressive load approached its critical value. We can conduct a simple analysis of this system by using the methods developed earlier in this book. Suppose the strut has a length l, end mass m, flexural rigidity EI, √ and oscillates in gravity g. Introducing the nondimensional parameter α = mg/EI, we can readily show that when the strut has the mass placed at its top [Fig. 11.4(a)], the natural frequency is given by [16] ω = gα/(tan αl − αl), (11.6) which remains positive until buckling occurs at mc = π2 EI/4gl 2 . When the strut is turned upside down [Fig. 11.4(b)] the natural frequency becomes ω = gα/(αl − tanh αl). (11.7) Thus, suppose we have a mass that corresponds to about the half the critical load, √ 2 2 that is, m = (π EI)/(8gl ); then α = π/(2 2l) and the natural frequency of the sys√ √ tem in part (a) would be 1.106 g/l, as opposed to 1.904 g/l for the system in part (b). In fact, even a mass that causes buckling in part (a) would result in oscillations √ of frequency 1.55 g/l in the inverted system [part (b)]. It is quite easy to demonstrate this experimentally [16]. Consider a simple polycarbonate cantilever strip, as shown in the inset to Fig. 11.5. If we consider the beam mass as being negligible compared with the concentrated mass added to its free end, then the theoretical results given by Eqs. (11.6) and (11.7) apply. For the specific case of a strip with L = 0.181 m, a second moment of area I = 1.903 × 10−12 m4 , and

2

18:2

P1: KAE Chapter-11

CUFX159-Virgin

0 521 85648 5

August 14, 2007

220

Nondestructive Testing 2 2

1/2

(mgL /EI)

downward horizontal upright

1.5

1

0.5

0 0

0.2

0.4

0.6

0.8 1 1/2 f/(g/L)

Figure 11.5. A simple experimental cantilever and its frequency variation with end load for different orientations.

Young’s modulus E = 2.4 × 109 N/m2 , we add increments of end mass and measure the resulting fundamental frequency. The nondimensionalized results are shown in Fig. 11.5 together with the theoretical results. Also included is the simple horizontal √ cantilever result based on the lumped stiffness approximation, that is, ω = Ke /m in which Ke = 3EI/L3 [17]. The effect of the mass of the beam itself can be included either in this lumped analysis or by use of a more sophisticated approach (adding a little distributed mass would tend to shift the data points up slightly), but the trend describing the effect of gravity (and hence axial loading) is clear. We also note at this point that experiments on cantilevers with self-weight loading are relatively easy to set up. The results from tests with other boundary and loading conditions, for example, in a testing machine, need more careful interpretation, as discussed in Section 7.2. As a reference point the typical amount of end mass the strut was able to withstand before appreciably starting to droop to one side was about 27 g. The Euler load for a cantilever is EIπ2 /(4L2 ), which gives a value of mc = 35 g, and given the inevitable initial imperfections in the system this magnitude is not unreasonable. Furthermore, when no end mass was added the strut vibrated with a measured natural frequency of a little over 6 Hz (in fact 6.075, 6.2375, and 6.4 in its upright, horizontal, and downward orientations). The theory of continuous elastic beams coveredin Chapter 7 listed a fundamental natural frequency for a cantilever of ω = 3.52 EI/mL4 and with the total mass of the strip measured at mL = 5.94 × 10−3 kg this corresponds to a predicted frequency of 6.38 Hz. We can also reinterpret Fig. 7.18 at this point. Recall that this plot referred to a slender continuous strut subject to self-weight loading (rather than a concentrated end mass). Plotting the original “weight” parameter |α| as a function of frequency squared gives the results shown in Fig. 11.6(a). In the absence of gravity we would

18:2

P1: KAE Chapter-11

CUFX159-Virgin

0 521 85648 5

August 14, 2007

11.1 Introduction

221 (b)

(a) 25

L /Lc

1.4 1.3

Downward

1.2

20

1.1 g

1

15 upright

0.9 0.8

10 downward

0.7 0.6

5

upright

0.5 0

0

5

10

15

20

25

30

35

40

45

50

0.4

0

0.5

1

1.5

2

2.5

3

3.5

4

4.5

Frequency (Hz)

Figure 11.6. The frequencies of a simple but heavy experimental cantilever. The solid line represents the upright case, and the dashed line represents the hanging-down orientation: (a) |α| versus the fundamental frequency squared and (b) alternative plot of the same results.

expect the frequency to be proportional to the inverse of the length squared, and this case is shown too. However, in the upright configuration, as the critical length is approached the stiffness is diminished such that the frequency drops to zero at the critical length. If we plot the dimensional frequency versus the length (normalized by the critical length), we get the results shown in Fig. 11.6(b). However, not all the data from part (a) are included because of different thicknesses. The near linear relationship is, of course, a suitable form for extrapolation. Thus we might measure the fundamental natural frequency for a number of different α values (specifically changing the length L) and fitting a straight line to this data [Fig. 11.6(b)] we would predict buckling in the vicinity of α ≈ 7.8. Recall that in this plot the “weight” α is a nondimensional parameter given by α = mgL3 /(EI), and hence with mass per unit length of 0.0147 kg/m, cross-sectional dimensions of 25.4 mm × 0.508 mm, and Young’s modulus of 2.4 GPa we get the actual length at buckling of about 0.33 m. The cantilever that hangs down never buckles as the length increases of course. We expect the natural frequency of a system to reduce if more mass is added to it. But what the preceding setup shows is that if the mass acts through gravity then it may reduce the stiffness of the system, and it is this tendency that can be exploited in terms of nondestructive (stability) testing. In other words, if we refer back to Fig. 11.5 and consider a fixed value of the end mass toward higher values (where gravity has more effect) then the effect of orientation (and whether the effective axial load is compressive or tensile) is apparent. A good deal of the earlier material in this book has highlighted ways in which this trend may be relatively simple [18]. If the trend is linear then it also provides the possibility of predicting the elastic buckling not only from, in principle, measurement of the lower natural frequency at two distinct axial-loading conditions but even when one or more of these loads is tensile. In a practical (experimental) context, the boundary conditions may not be

5

18:2

P1: KAE Chapter-11

CUFX159-Virgin

0 521 85648 5

August 14, 2007

222

Nondestructive Testing

Phase 7.00 102 Accelerometer/ force

7.00 10-2 2.00 101

Frequency (Hz)

2.56 102

Figure 11.7. Frequency content of a prismatic beam showing the shift in resonant frequencies under the application of axial loading. Reproduced with permission from Elsevier [19].

known. Consider the results shown in Fig. 11.7, which were described in Livingston et al. [19]. This frequency spectrum was obtained from a prismatic beam by experimental modal analysis as part of a larger study in the context of system identification and parameter estimation. Over this frequency range the lowest three frequencies are quite distinct. The solid line corresponds to (practically) zero axial loading, with the dotted line showing the shift to higher frequencies when the beam is subject to a tensile axial load (approximately of a similar magnitude to that of the Euler buckling load with boundary conditions somewhat intermediate between clamped and pinned) [20, 21].

11.2 Some Background As mentioned in the introduction to this chapter, the idea of using dynamic behavior to predict buckling has received attention from a number of researchers over the years. However, major contributions were made by Massonet [1], who considered a variety of structural systems from a theoretical standpoint, and Lurie [2], who showed the utility of this approach including experiments. Even when the mode of vibration and buckling mode are not identical the load–frequency (-squared) relation may be almost linear. Lurie used an energy approach to show that an upper limit for the frequency of axially loaded thin beams resulted in a relation l 2 l dx mω2 0 w2 dx P 0 dw dx 1≥ l d2 w 2 + l d2 w 2 . EI dx EI dx 2 2 0

dx

0

dx

(11.8)

18:2

P1: KAE Chapter-11

CUFX159-Virgin

0 521 85648 5

August 14, 2007

11.2 Some Background

223

1

P/Pcr 0.8

0.6

0.4

0.2

0

0

0.2

0.4

0.6

0.8

2

(ω/ω1)

1

Figure 11.8. Upper and lower bounds of the frequency–load relation for a clamped–clamped beam.

For example, consider a clamped–clamped beam for which we know that Pcr = 2 2 4 4π EI/l and ωn (P = 0) = 22.373 EI/(ml ), and using the buckling mode shape 2πx w = A 1 − cos (11.9) l results in an expression

ω 1 ≥ 0.9635 ωn

2 +

P . Pcr

(11.10)

P . Pcr

(11.11)

Using the lowest mode of vibration [22] results in 1≥

ω ωn

2 + 0.9704

Relations (11.10) and (11.11) are plotted as inequalities in Fig. 11.8 together with the linear relation [Eq. (11.1)]. Thus we see the possibility of exploiting the linear relation between the square of the lowest natural frequency and the level of axial loading to extrapolate critical conditions [23]. Underlying General Theory. We have repeatedly looked at systems with a stiffness

that tended to be diminished by the presence of (compressive) axial loading. In terms of potential energy, we can write this as V = U(qi ) − ηkEk(q)i ,

(11.12)

18:2

P1: KAE Chapter-11

CUFX159-Virgin

0 521 85648 5

August 14, 2007

224

Nondestructive Testing

with the quadratic approximation in the form of the inner product V=

1 < q, (U − ηkEk)q >, 2

(11.13)

where the ηk (k = 1, 2, . . . , m) are independent parameters and U is the strain energy (symmetric and positive-definite). In terms of the equations of motion, we use Lagrange’s equation to obtain Mq¨ + (U − ηkEk)q = 0,

(11.14)

and assuming harmonic motion in the usual way, q = ueλt , we obtain the characteristic equation |Mλ2 + U − ηkEk| = 0.

(11.15)

For conservative systems, we have λ = iω with λ2 identified as the negative of the square of the natural frequencies (see Chapter 4). We are, of course, primarily interested in systems for which q = 0 represents a stable system but may become unstable (at buckling), and this occurs when one of the eigenvalues vanishes. Although instability may occur by means of a complex pair of eigenvalues in nonconservative systems (flutter, e.g., Beck’s problem, Section 7.8), in this book we remain primarily focused on the conditions under which a real eigenvalue vanishes at the divergence boundary. The relation between ω2 and ηk constitutes the characteristic curve (or surface, when more than one parameter is present). It has been proven [24, 25] that, for conservative systems with a trivial equilibrium state, any number of degrees of freedom, and equations of motion that are linear in the parameters ηk, the surface involving the fundamental frequency cannot have convexity toward the origin. Furthermore, it also follows that the fundamental surface is a plane (or straight line for a system with a single parameter) if the matrices M, U, and Ek can be reduced to a diagonal form simultaneously. A useful implication of this convexity property (and of obvious usefulness in the context of the present chapter) can be concluded. The divergence boundary is contained in the characteristic curve, and we obtain it by setting ω2 = 0. If a single parameter ξ (load) is acting on the system, then it is possible to obtain an (upper bound) estimate of the critical value. From Fig. 11.9, we see that if we know the frequencies at two values of the loading parameter ξ, ω211 (ξ1 ) and ω212 (ξ2 ), we can gain an estimate of frequencies at other loading values. Of course, if the characteristic curve is a straight line (e.g., if the equations uncouple) then this estimate will be exact. By extrapolating a line joining them, we obtain an upper bound on the critical value of ξ from the intersection with the ξ axis. Clearly, the accuracy of the estimate also depends on the location of the two reference points. This is an issue that will be discussed later.

18:2

P1: KAE Chapter-11

CUFX159-Virgin

0 521 85648 5

August 14, 2007

11.3 Snap-Through Revisited

225

u cr 2

Figure 11.9. Convexity of the characteristic curve and its implication for providing a lower bound.

1

2

2

2

12

11

11.3 Snap-Through Revisited In the previous section, we considered systems for which the fundamental equilibrium path was the trivial one. In snap-through buckling, we might still expect to monitor the lowest natural frequency to predict instability, but the nonlinearity of the underlying equilibrium curve can also have an influence. To quantify this, we go back to one of our standard forms from Chapter 3 in which we consider the dynamics of a system in the vicinity of a saddle-node bifurcation: ¨ − X2 − λ = 0, X

(11.16)

where both the deflection X and the load parameter λ are measured from the origin. Now suppose we have an equilibrium position (Xe ), and we wish to study the behavior of small oscillations about it. We can expand Eq. (11.16) in the usual way by replacing X with Xe + x that leads to x¨ − X2e − 2Xe x − x2 − λ = 0.

(11.17)

The x2 term can be dropped because it is small, and because of equilibrium we also have −X2e − λ = 0, and thus we are left with x¨ − 2Xe x = 0.

(11.18)

This system has the natural frequency ω=

−2Xe = +2(−λ)1/2 ,

(11.19)

and thus, for large negative λ say, we observe a linear relation between the loading parameter and the fourth power of the natural frequency [26]. Effect of Damping. So far we have concentrated mainly on undamped systems.

In most of the mechanical systems of interest, there is usually a little energy

18:2

P1: KAE Chapter-11

CUFX159-Virgin

0 521 85648 5

August 14, 2007

226

Nondestructive Testing −4

0

−2

2

0

ω2

2

4

ω4 β=2

−0.5

−0.5

β=1 β=0

−1.0

β=1 −1.5

λ −2.0

β=2

−1.0

β=0

−1.5

λ −2.0

Figure 11.10. The effect of damping on the frequency–load-parameter relation.

dissipation, and we shall assume that this takes the form of a linear-viscous damping (see Section 3.5). Thus we consider X¨ + βX˙ − X2 − λ = 0.

(11.20)

Conducting an analysis similar to that of the previous section we arrive at relationships between the natural frequency (the harmonic factor in the decaying, oscillating motion) and load parameter of ω2 = ±2(−λ)1/2 − (β/2)2 , ω4 = 4(−λ) ± 4(−λ)1/2 (β/2)2 + (β/2)4 .

(11.21)

These expressions are plotted in Fig. 11.10 for three values of damping including the undamped case. We see that damping has the effect of causing the natural frequency to diminish to zero prior to buckling. For example, with a damping level of β = 2 (and assuming the damping coefficient is constant), we observe that oscillations will cease when the load reaches a value of about λ = −0.25. One way of thinking about this is to recall the standard expression for a damping ratio: ζ = c/(2mωn ), but now the stiffness is reducing and thus, although the damping coefficient is constant, the damping ratio increases such that damping effectively becomes critically damped (to use the definitions introduced in Section 3.1) just prior to the stiffness dropping to zero. We can again integrate the equation of motion while slowly sweeping through the load parameter (as was done in Chapter 3). For example, Fig. 11.11 shows nine trajectories generated for system equation (11.20) and with a constant value of the initial total energy with β = 0.5, λ evolved √ at the rate 30t, and the initial conditions prescribed by x(0) ˙ = 0.0, x(0) = − 300 + A, where A varied between −8 and 8 in increments of 2. A number of interesting features can be seen in this figure. Damping does indeed appear to make the oscillations die out prior to instability, although this drifting system is never, of course, quite in equilibrium. We also see that

18:2

P1: KAE Chapter-11

CUFX159-Virgin

0 521 85648 5

August 14, 2007

11.3 Snap-Through Revisited

227

x 5

–10

–8

–6

–4

t

–2 –5 –10 –15 –20 –25

Figure 11.11. Some trajectories plotted as time series as the system is swept toward the saddle-node bifurcation.

the larger-amplitude oscillations display a degree of asymmetry—this feature is not unexpected because for a given amplitude the motion evolves toward the instability as the potential-energy well shrinks, from one side. A simple experimental verification of this situation is shown in Fig. 11.12. Here, a flexible rod with an end mass in a heavily postbuckled configuration was subject to base rotation such that a saddle-node bifurcation was encountered. The base rotation can be thought of as control in our standard system of gravity acting on the mass. The system follows its complementary equilibrium path during which time the frequency of natural (superimposed) oscillations are measured. The jump at the saddle-node bifurcation is represented schematically as A–B in Fig. 11.12(b). The measured frequencies and their relation to the control parameter are shown plotted in Fig. 11.13. Part (b) shows some times series in which the parameter r is a measure of the rate at which the base is rotated. We see that raising the frequency to the fourth power provides a more linear relationship than for the second power with which to predict criticality [26].

(a)

(b) Snap-through

m

Deflection to left

A Complementary post-critical state

B g

Natural post-critical state

End mass m (held constant throughout)

A Base rotation

Unstable region B

Control parameter

Figure 11.12. (a) A flexible strut with an end mass and (b) control surface showing a transition through bifurcation [26].

18:2

P1: KAE CUFX159-Virgin

0 521 85648 5

August 14, 2007

228

Nondestructive Testing

r3 = 4 r1

ω4

Freq. (arbitrary units)

0

ω2

Control parameter (rotation) Λ

Displacement

Chapter-11

r2 = 2 r1

Decreasing frequency

r1

Control parameter (= base rotation, a linear function of time)

Figure 11.13. (a) Measured frequencies for the flexible strut rotated through a saddle-node bifurcation and (b) some sample time series with different rates of rotation [26].

The concept of using the reduction in frequency as a stability predictor has also been used in the context of secondary buckling [27]. Referring back to Fig. 5.21, it is apparent that once the Augusti model has buckled into its primary mode, it is the frequency of the second mode that can be used to infer the approach of the secondary bifurcation as this frequency tends toward zero.

11.4 Range of Prediction The previous section indicated that rather than there being a universal relation between frequency and load, which would make extrapolation fairly straightforward, we see that damping, changing boundary conditions, initial imperfections, and type of instability, all conspire to make predictions more difficult. For example, Lurie [2] pointed out the increasingly important role played by initial imperfections as the critical load is approached (especially for plates). An experimental example of this effect [28] is shown in Fig. 11.14(a). Part (b) shows the results from a laminated composite column in which Chailleux et al. [28] identified three distinct regions, with region II providing the most useful (linear) relation for prediction purposes. Also in part (b), the authors noted that with very low load levels they experienced some clearance in the boundary conditions. As pointed out by Lurie [2], it may be possible to test specimens in tension, in which case initial imperfections will have minimal influence. In both of these sets of results, we observe that the linear relation breaks down near buckling (a feature first observed in Fig. 7.6). Given that raising the frequency to either the second or fourth power might be a more appropriate predictor, it seems reasonable to raise the frequency to various powers in order to see how the subsequent curve might change from concave to convex, which then has clear implications for lower bound estimates. For example, the frequency–load relations for an elastic arch were given in Eqs. (8.76) and (8.77) for a shallow, and less shallow arch, respectively, and shown in Fig. 8.11. Plaut and Virgin [29] studied the effect of extrapolating frequency raised to various powers

18:2

P1: KAE Chapter-11

CUFX159-Virgin

0 521 85648 5

August 14, 2007

11.4 Range of Prediction

(a)

229

(b)

Pdyn = 117.5 kg 100

P (kg)

III

P (kg)

Pdyn = 45.7 kg II

10

0

2

f (Hz 2 )

5 103

I

1 104

f 2 (Hz 2)

Figure 11.14. The load–frequency (-squared) relation for (a) rectangular duraluminum plate and (b) laminated composite column. Adapted from Chailleux et al. [28].

with a special reference for the range over which data were measured. That is, by raising the frequency to various powers a value is reached whereby the relation shown in Fig. 8.11 changes from convex to concave, with the transition point providing a close-to-linear relationship. For example, Plaut and Virgin [29] used the criterion suggested by Singer et al. [12] using (numerical) data from the simple elastic arch considered earlier. Over a broad range of loading conditions, i.e, not necessarily close to buckling, we typically do not observe a (near) linear-frequency-squared–load relation. However, we can fit data to a relation of the form p = C − D r ,

(11.22)

in which p is the load and is the frequency ωf [from Eq. (8.76) or (8.77)] nondimensionalized by the frequency under zero load. They showed that the values of r varied according to the range of load levels considered but that a change in curvature of the (frequency)r versus load relation could be extracted to provide reasonable upper and lower bound estimates of the buckling load. That is, a value for r is sought such that it results in the most linear relation between p and (frequency)r . The following table shows an example for the case λ = 3 (i.e., the less shallow arch). N 4 5 6 7 8 9 10 11 12

C

D

r

11.04 9.30 9.25 9.35 9.38 9.46 9.48 9.53 9.59

11.04 9.32 9.27 9.37 9.40 9.47 9.48 9.52 9.57

2.15 2.69 2.71 2.69 2.65 2.61 2.60 2.57 2.52

18:2

P1: KAE Chapter-11

CUFX159-Virgin

0 521 85648 5

August 14, 2007

230

Nondestructive Testing

(a)

(b)

p/pb

p/pb

f

f

Figure 11.15. Frequency–load relations for an elastic arch: (a) λ = 2 and (b) λ = 3.

In this table, N represents the number of data points used in the fit and spreads roughly evenly from zero to critical loading, and the last column gives the exponent. In this nondimensionalization, the critical load (which corresponds to a bifurcation rather than to a limit point) occurs at p = 9.71. The data were subject to a nonlinearleast-squares fit and we observe a range of optimal exponents depending on the number of data points used. The shallow arch is a useful device for illustrating frequency–load effects and applicability in terms of predicting instability because the λ parameter (i.e., the rise of the arch), can change the nature of the critical point [30]. For example, in Fig. 8.11, when λ = 2, the lowest frequency drops to zero when the arch experiences a saddle-node bifurcation and the arch snaps through to its inverted position. For a less shallow arch (e.g., when λ = 3), there is bifurcation and the arch experiences an asymmetric buckling (with a full-sine mode). However, the relations between frequencies and the type of instability (for example, as described in Section 3.4), and the material earlier in this section, were based on local generic behavior. If we take a closer look at Fig. 8.11 and focus in on a close proximity to buckling we get the results shown in Fig. 11.15. Here, the load has been normalized such that buckling corresponds to p/p b = 1. Thus in the vicinity of the instability it is the frequency raised to the fourth power that is more useful for predicting the saddle-node [part (a)], and frequency squared for the branching bifurcation [part (b)].

11.5 A Box Column A notable piece of work on the dynamic nondestructive evaluation of structures can be found in Jubb et al. [31]. The authors conducted some tests on box columns, which provided a clever way of studying plates by incorporating simply supported edges. One of their main goals was to establish a means of assessing the effects of

18:2

P1: KAE Chapter-11

CUFX159-Virgin

0 521 85648 5

August 14, 2007

11.6 Plates and Shells

residual stresses on the stiffness, dynamics, and stability of a typical structure. They suggested using the following variation on the frequency ( f )–load (σ) theme: 2 σr σa f k + + = 1. (11.23) σcr σcr f0 In this expression, the residual stress, σr , is added to the external stress, and k is a constant (less than unity) that takes account of stress distribution. We have already encountered how the (ordering of) modes of vibration of a plate depend on axial load (see Fig. 10.3), and in the experiments of Jubb et al. [31] they chose an aspect ratio of 4. In the absence of axial load, if the frequency corresponding to four half-sine waves in the longitudinal direction (m = 4) is denoted by 1.0, the lower modes turn out to have relative frequencies of 0.610 (m = 3), 0.391 (m = 2), and 0.282 (m = 1). As the axial force increases, these frequencies decrease (linearly with frequency squared) but at different rates such that it is the m = 4 mode that drops to zero at buckling (i.e., P/Pcr = 1). The experimental results are shown in Fig. 11.16. The lowest four natural frequencies are plotted as functions of axial load. Two important points can be extracted from the results. First, the welded corners induce a degree of residual stress such that the initiation of an applied axial load does not correspond to zero axial load in the frequency–load relation. Second, a degree of postbuckling stiffening is apparent in each of the modes—this is a feature anticipated by the analysis of Section 10.4. Therefore it may be important to monitor the first few lowest natural frequencies in order to capture the appropriate buckling mode.

11.6 Plates and Shells A thorough body of work on nondestructive testing of cylindrical shells under axial loading by use of dynamic (vibration) characteristics is due to Singer and his colleagues [11, 12]. An example of this type of research is shown in Fig. 11.17. Here a series of tests was conducted on cylindrical shells to see if the lowest frequency of lateral vibrations could be useful in predicting buckling. The cylinder included rib stiffeners, which had the effect of reducing some of the imperfection sensitivity typically encountered in axially loaded shells. Figure 11.17(a) shows a conventional frequency-squared versus load plot. Part (b) shows the same data with the frequency raised to the 2.9th power [12]. Figure 11.18 shows a plot in which (lower-load-level) frequencies are raised to various powers and then extrapolated as suggested in Plaut and Virgin [29]. The simplicity of the frequency–load relationship (or a variation thereof) can also be exploited when more complex structures are studied. For example, consider the panel (representative of a typical aeronautical configuration) shown in the lower part of Fig. 11.19. An eigenanalysis was conducted by Williams et al. [32] using the computer program viconopt. This structure, which might be a typical component in an aircraft fuselage, has ends that are simply supported such that sinusoidal modes result in the longitudinal direction. These are characterized by half-wavelengths λ,

231

18:2

P1: KAE Chapter-11

CUFX159-Virgin

0 521 85648 5

August 14, 2007

232

Nondestructive Testing Appled axial load (kN) 0

200

600

400

800

Theoretical elastic buckling load = 1070 kN

1.0

(f m/f 0)

Actual buckling load = 754 kN

2

1000

0.5

m=4 0

0.5

m=3

0 0.5

m=2

0 0.5

m=1 0

0

0.4

0.2

0.6

0.8

1.0

P/Pcr

Figure 11.16. The four lowest frequencies for the box column plotted as functions of the applied axial load. Adapted from Jubb et al. [31].

2

5

f × 10 2 (Hz) 4

q

9

f × 10 q (Hz)

3

2

2 1 1 0 0

1000

2000

3000

4000

P(Kg)

0 0

1000

2000

3000

4000 P(Kg)

Figure 11.17. The lowest vibration frequency of an axially loaded cylinder: (a) frequency squared and (b) frequency raised to the power q = 2.9. Adapted from Singer et al. [12].

18:2

P1: KAE Chapter-11

CUFX159-Virgin

0 521 85648 5

August 14, 2007

11.6 Plates and Shells

233

q=1 q=2 q=3

1.0

f

q

0.5 q=4 q=5

0

1000

3000

2000

4000

P (kg) Pexp

Figure 11.18. The same frequencies as in Fig. 11.17 but raised to various powers and subsequently extrapolated (linearly) to the buckling load. Adapted from Singer et al. [12].

which are integral fractions, l/i, of the length of the structure, l. A set of eigenvalues (natural frequencies) can be computed for values of λ. Suppose we have a structure that is isotropic (but need not necessarily be so) and with the geometric properties shown in the lower part of the figure. The nota tion α = 1000/E and β = (4000ρ/E) are introduced, and Poisson’s rato is ν = 0.3. Suppose we wish to extract the natural frequencies below 6/βl when l = 6b. The computed characteristic curves shown in the upper part of Fig. 11.19 indicate natural frequencies where the vertical dashed lines intercept the curves, with the small markers indicating a frequency of 6/βl. Thus we observe the frequencies listed in the top rows of the following table (the entries are in terms of the reciprocal of βl): nj nσj nj (cont’d) nσj (cont’d)

1.37(1) 0.82(1)

1.66(1) 1.25(1)

2.02(1) 1.70(1)

2.29(1) 2.01(1)

4.43(2) 3.85(2)

4.82(2) 4.29(2)

5.25(1) 5.13(1)

5.40(2) 4.94(2)

5.99(2) 5.57(2)

5.99(1) 5.89(1)

− 5.76(3)

Now, if the panel is subjected to a compressive load σ = 0.0012E we can exploit the equivalence of the vibration and buckling modes to compute the reduced natural frequencies that are due to the presence of the axial load. That is, we make use of the relation [of the form of Eq. (11.1)]: ασ 1/2 2 nσj = nj − 2 2 . βλ

(11.24)

Thus (ασ)1/2 ≈ 1.1 and for the various λ we get the altered natural frequencies as shown in the lower rows of the preceding table. We note how the order has changed, the equal values have separated, and a new natural frequency has fallen into the range of interest because of the presence of the axial load.

18:2

P1: KAE Chapter-11

CUFX159-Virgin

0 521 85648 5

234

August 14, 2007

Nondestructive Testing

Figure 11.19. Eigenvalue curves for an axially loaded, prismatic, stiffened, plated structure. Reproduced with permission from Elsevier [32].

We note here also that the linearity in the axial load versus frequency-squared relation can also be used to infer the percentage error in neglecting shear effects in the modeling [33]. Another important practical application related to this approach is determining the level of stresses in pressure vessels [34]. References [1]

C. Massonnet. Les relations entre les modes normaux de vibration et la stabilite´ des ´ systemes elastiques. Technical Report, Bulletin des cours et des laboratoires d’essais des constructions du genie civil et d’hydraulique fluviale, Brussels, Belgium, I (1,2), 1– 353, 1940.

18:2

P1: KAE Chapter-11

CUFX159-Virgin

0 521 85648 5

August 14, 2007

References [2] [3] [4] [5] [6]

[7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20] [21] [22] [23]

[24] [25] [26] [27]

H. Lurie. Lateral vibrations as related to structural stability. Journal of Applied Mechanics, 19:195–204, 1952. C. Sundararajan. Frequency analysis of axially loaded structures. AIAA Journal, 30:1139–41, 1992. R.V. Southwell. On the analysis of experimental observations in problems of elastic stability. Proceedings of the Royal Society of London, 135A:601–16, 1932. J. Ari-Gur, T. Weller, and J. Singer. Experimental and theoretical studies of columns under axial loading. International Journal of Solids and Structures, 18:619–41, 1982. M.A. Souza. The effects of initial imperfection and changing support conditions on the vibration of structural elements liable to buckling. Thin-Walled Structures, 5:411–23, 1987. R.G. White. Evaluation of the dynamic characteristics of structures by transient testing. Journal of Sound and Vibration, 15:147–61, 1971. A. Segall and M. Baruch. A nondestructive dynamic method for the determination of the critical load of elastic columns. Experimental Mechanics, 20:285–8, 1980. P.M. Mujumdar and S. Suryanarayan. Nondestructive techniques for prediction of buckling loads – a review. Journal of the Aeronautical Society of India, 41:205–23, 1989. M.A. Souza and L.M.B. Assaid. A new technique for the prediction of buckling loads from nondestructive vibration tests. Experimental Mechanics, 31:93–7, 1991. J. Singer, J. Arbocz, and T. Weller. Buckling Experiments, Vol. 1. Wiley, 1998. J. Singer, J. Arbocz, and T. Weller. Buckling Experiments, Vol. 2. Wiley, 2002. J.G.A. Croll and A.C. Walker. Elements of Structural Stability. Wiley, 1972. C.R. Farrar, S.W. Doebling, and D.A. Nix. Vibration-based structural damage identification. Philosophical Transactions of the Royal Society, A359:131–49, 2001. A. Sommerfeld. Eine einfache Vorrichtung zur veranschaulichung des Knickungsvorganges. Zeitschrift des Verein deutscher Ingenieure, pp. 1320–3, 1905. A.B. Pippard. Response and Stability. Cambridge University Press, 1985. T.D. Burton. Introduction to Dynamic Systems Analysis. McGraw-Hill, 1994. A. Segall and G.S. Springer. A dynamic method for measuring the critical loads of elastic flat panels. Experimental Mechanics, 26:354–9, 1986. ´ T. Livingston, J.G. Beliveau, and D.R. Huston. Estimation of axial load in prismatic members using flexural vibrations. Journal of Sound and Vibration, 179:899–908, 1995. A.L. Sweet and J. Genin. Identification of a model for predicting elastic buckling. Journal of Sound and Vibration, 14:317–24, 1971. A.L. Sweet, J. Genin, and P.F. Mlakar. Determination of column-buckling criteria using vibratory data. Experimental Mechanics, 17:385–91, 1977. D.J. Inman. Engineering Vibration. Prentice Hall, 2000. P.-Y. Shih and H.L. Schreyer. Lower bounds to fundamental frequencies and buckling loads of columns and plates. International Journal of Solids and Structures, 14:1013–26, 1978. K. Huseyin and J. Roorda. The loading–frequency relationship in multiple eigenvalue problems. Journal of Applied Mechanics, 38:1007–11, 1971. K. Huseyin. Multiple Parameter Stability Theory and Its Applications. Oxford University Press, 1986. L.N. Virgin. Parametric studies of the dynamic evolution through a fold. Journal of Sound and Vibration, 110:99–109, 1986. L.N. Virgin and R.H. Plaut. Use of frequency data to predict secondary bifurcation. Journal of Sound and Vibration, 251:919–26, 2002.

235

18:2

P1: KAE Chapter-11

CUFX159-Virgin

236

0 521 85648 5

August 14, 2007

Nondestructive Testing [28] A. Chailleux, Y. Hans, and G. Verchery. Experimental study of the buckling of laminated composite columns and plates. International Journal of Mechanical Sciences, 17:489–98, 1975. [29] R.H. Plaut and L.N. Virgin. Use of frequency data to predict buckling. Journal of Engineering Mechanics, 116:2330–5, 1990. [30] R.H. Plaut and E.R. Johnson. The effect of initial thrust and elastic foundation on the vibration frequencies of a shallow arch. Journal of Sound and Vibration, 78:565–71, 1981. [31] J.E.M. Jubb, I.G. Phillips, and H. Becker. Interrelation of structural stability, stiffness, residual stress and natural frequency. Journal of Sound and Vibration, 39:121–34, 1975. [32] F.W. Williams, P.N. Bennett, and D. Kennedy. Curves for natural frequencies of axially compressed prismatic plate assemblies. Journal of Sound and Vibration, 194:13–24, 1996. [33] P.N. Bennett and F.W. Williams. Insight into the sensitivity to axial compressive load of the natural frequencies of structures which include shear deformation. Journal of Sound and Vibration, 209:707–22, 1998. [34] R.R. Archer. On the influence of uniform stress states on the natural frequencies of spherical shells. Journal of Applied Mechanics, pp. 502–5, September 1962.

18:2

P1: KAE Chapter˙12

CUFX159-Virgin

0 521 85648 5

August 14, 2007

12 Highly Deformed Structures

In most practical applications, Euler–Bernoulli beam theory is often sufficient to provide useful information about the relation between axial loading and free vibrations. However, there are a number of instances in which the axial loading, or some related effect, results in relatively highly deflected states of the system, especially when the structure under consideration is very slender. For example, a pipeline or cable is characterized by having one of its dimensions very much greater than the other two, and the loads to which it is subject may often result in large deflections, even in cases in which self-weight is the only appreciable loading [1, 2]. Elastic bending stiffness does not necessarily dominate the effects of gravity, for example. In these cases, a more sophisticated description of the geometry is needed, and it is these types of flexible structures that form the basis for this chapter. In Chapter 7, we saw how initial postbuckling could be handled by retaining extra terms in the various energy expressions. But now, we allow (static) deflections to become large by using an arc-length description of the geometry and then consider small-amplitude oscillations about these nonlinear equilibrium configurations. In the final section, a FE solution is also shown for a specific case (essentially with the same approach as used toward the end of Chapter 9). It also turns out that experimental verification is relatively easy, especially if thermoplastics like polycarbonate are used.

12.1 Introduction to the Elastica We start, in the usual way, by considering the behavior of an initially straight, inextensible, prismatic, thin elastic beam. The curvature of such a system is given by 1 dθ = , ρ dS

(12.1)

in which the deformed geometry of the system is described in terms of the arc-length coordinates S and θ, as shown for an axially loaded clamped beam in Fig. 12.1. In terms of Cartesian coordinates, we can write the angle as θ = tan−1 and, using dS = dX

dY , dX

1 + (dY/dX)2 ,

(12.2)

(12.3) 237

18:3

P1: KAE Chapter˙12

CUFX159-Virgin

0 521 85648 5

August 14, 2007

238

Highly Deformed Structures

Y S

X

L

Figure 12.1. A thin elastic beam with clamped end conditions subjected to end shortening.

we obtain the familiar expression for curvature dθ dθ dX d 2 Y/dX 2 1 = = = . 3/2 ρ dS dX dS [1 + (dY/dX)2 ]

(12.4)

Expanding the right-hand side as a Taylor series leads to 1 = (d 2 Y/dX 2 )[1 + (3/2)(dY/dX)2 + · · · +]−1 . ρ

(12.5)

Thus, if dY/dX is small, then we obtain the simple expression for curvature familiar from Euler–Bernoulli beam theory and generally used as an analytical basis in Chapters 7 and 8. In Eq. (7.40) the curvature was developed in terms of Lagrangian coordinates, and the next term in the expansion was retained for moderately large slopes. Both descriptions appear in the literature [3]. However, we now consider the fully (but still elastic) nonlinear system with no restriction on deflections. Returning to the example in Fig. 12.1 (and assuming no gravitational effects just yet) we can write the governing (elastica) equation in the relatively simple form d 2θ P sin θ, =− 2 dS EI

(12.6)

in which P is an axial load associated with the end shortening . This form is restricted to prismatic members with no forcing acting along its length but is a form familiar from the swings of a pendulum [4]. The boundary conditions must then be specified to obtain a solution to a specific problem. However, this equation is not easy to solve. Analytical solutions are available through elliptic integrals [5], and this form also allows mildly nonlinear solutions to be obtained with perturbation methods (see Naschie [3]), but we shall adopt numerical methods to solve this type of problem, that is, to obtain the deflected configuration under various axial loads and subsequent vibration properties about equilibrium configurations (which may be highly nontrivial) [6–8]. Four case studies will be described in which the primary difference between the cases concerns the boundary conditions. Experimental verification is included for each.

18:3

P1: KAE Chapter˙12

CUFX159-Virgin

0 521 85648 5

August 14, 2007

12.2 The Governing Equations

239

12.2 The Governing Equations If we return to a prismatic strip but now inclined to the horizontal by an angle β, and characterized by flexural rigidity EI, length L, and weight (per unit length) W, we can describe the geometry in terms of coordinates X(S, T ) and Y(S, T ), and rotation θ(S, T ) with respect to the X axis, where S is the arc length and T is time. The internal forces in the strip are denoted P(S, T ) and Q(S, T ) parallel to the X and Y axes, respectively, and the bending moment is M(S, T ). The governing equations can be written as [9] ∂X/∂S = cos θ,

∂Y/∂S = sin θ,

∂θ/∂S = M/EI,

∂M/∂S = Q cos θ − P sin θ,

(12.7)

and, in addition, we have dynamic equilibrium, ∂P/∂S = −(W/g)∂2 X/∂T2 − W sin β, ∂Q/∂S = −W cos β − (W/g)∂2 Y/∂T2 .

(12.8)

Damping can be added at this point, but relative to the experimental results to be discussed later the damping is very light and is neglected in the analytical description. We introduce convenient nondimensional quantities that are especially useful in the context of this kind of nonlinear formulation w = WL3 /EI,

x = X/L,

q = QL2 /EI,

m = ML/EI,

y = Y/L,

s = S/L, p = PL2 /EI, t = (T/L2 ) EIg/W, = ωL2 W/EIg, (12.9)

where ω is a dimensional vibration frequency. In nondimensional terms, equilibrium equations (12.8) thus become ∂x/∂s = cos θ,

∂y/∂s = sin θ,

∂θ/∂s = m,

∂m/∂s = q cos θ − p sin θ,

(12.10)

and for linear vibrations [Eqs. (12.8)] ∂ p/∂ s = −w sin β − ∂2 x/∂ t2 ,

∂ q/∂ s = −w cos β − ∂2 y/∂ t2 .

(12.11)

Assuming harmonic motion appropriate to small-amplitude vibration in the usual way, we can write the variables in the form x(s, t) = xe (s) + xd (s) sin t,

y(s, t) = ye (s) + yd (s) sin t,

θ(s, t) = θe (s) + θd(s) sin t,

m(s, t) = me (s) + md(s) sin t,

p(s, t) = p e (s) + p d(s) sin t,

q(s, t) = qe (s) + qd (s) sin t,

(12.12)

18:3

P1: KAE Chapter˙12

CUFX159-Virgin

0 521 85648 5

August 14, 2007

240

Highly Deformed Structures

where subscripts e and d denote equilibrium and dynamic quantities, respectively. At equilibrium, the equations are now given by xe = cos θe ,

ye = sin θe ,

θe = me ,

me = qe cos θe − p e sin θe ,

(12.13)

where the prime is used to denote the derivative with respect to s and where the internal forces can be written as p e (s) = p 0 − sw sin β,

qe (s) = q0 − sw cos β,

(12.14)

where p 0 and q0 are constants representing values at s = 0. We can determine equilibrium shapes by solving Eqs. (12.14), and then we obtain small vibrations about these equilibrium solutions by solving the resulting linear equations in the dynamic variables: xd = −θd sin θe ,

yd = θd cos θe ,

θd = md,

md = (qd − p e θd ) cos θe − (p d + qe θd ) sin θe ,

p d = 2 xd ,

qd = 2 yd .

(12.15)

The general approach used here for solving these types of nonlinear boundaryvalue problems is based on the shooting method [9, 10]. With this approach, the known boundary conditions at one end are used together with educated guesses of the unknown boundary conditions, and the nonlinear ordinary differential equations are solved numerically. However, the boundary conditions at the far end will not typically be satisfied, and the error is then used to iteratively re-solve the system until a tolerance has been achieved [11]. This is basically a root-finding approach and can be significantly simplified by use of some of the built-in capabilities of MATLAB or Mathematica, for example. Continuation is an alternative solution procedure [12].

12.3 Case Study A: Self-Weight Loading Revisited In Section 7.9, we considered the effect of self-weight on the dynamics of an upright cantilever by using a Rayleigh–Ritz approach and included some simple experimental results. In the experimental results shown in Fig. 7.15(a) a degree of postbuckled stiffness can be observed for moderately large deflections. We take another look at this system but now using the elastica approach outlined in the previous section (as originally shown in Fig. 7.13). Figure 12.2 shows the arc-length coordinates together with some typical equilibrium configurations. These, of course, go well beyond the range of validity of Euler–Bernoulli theory. Before some further results are presented, the appropriate nondimensionalization is mentioned. For the examples to be considered later in this chapter, it is natural to normalize the “weight” according to the first element in Eqs. (12.9) with the distance between the ends providing a natural control parameter. For the upright

18:3

P1: KAE Chapter˙12

CUFX159-Virgin

0 521 85648 5

August 14, 2007

12.3 Case Study A: Self-Weight Loading Revisited

241

Y

Y X

X

S

H

W

(c) (a)

(b)

Figure 12.2. The slender column subject to self-weight: (a) basic geometry, (b) arc-length coordinate system, and (c) some typical deflected shapes.

column (and also the pinched-loop configuration to be considered later), it is more convenient (especially for subsequent comparisons with experimental results) to use a = (EI/W)1/3 as the key nondimensional parameter. This is somewhat different from the scheme presented in Eqs. (12.9), which are used later. We use h = H/a as the control, where H is the height of the column (and equivalent to L), such that increasing the length of the system, in the presence of gravity, increases the effective axial loading and leads to instability.

12.3.1 Numerical Results For a vertical upright column (i.e., with the clamped end at the bottom), we can set β = π/2, and the gravity acts in the negative x direction. The boundary conditions are fully clamped at the base, that is, xe = ye = θe = 0 when s = 0, and free at the tip, that is, me = 0 when s = h, and the shooting method is used to determine the unknowns at the tip to within a prescribed accuracy. A summary of the equilibrium solutions for the heavy column are shown in Fig. 12.3(a) as a bifurcation diagram, that is, a measure of the nondimensional (lateral tip) deflection, y(h) (see Fig. 12.2) plotted against the control parameter (nondimensional column height h). A typical break in symmetry will be provided by some initial geometric imperfection, but this effect is not included here [9]. The trivial solution is ye (x) = 0. The first bifurcation point occurs at the critical height hcr = 1.986 (compare with the analyses described

18:3

P1: KAE Chapter˙12

CUFX159-Virgin

0 521 85648 5

August 14, 2007

242

Highly Deformed Structures 3 4 h 3

(a)

(b)

h 2.5

postbuckled frequencies

2

2 1.5 1 0 −3

1 −2

−1

0

1

2 3 y(h)

0.5 −2

prebuckled frequencies

0

2

4

2 6

Figure 12.3. (a) Equilibrium paths for the column including gravity and (b) corresponding frequencies of small-amplitude vibrations.

in Section 7.9 [13]), and the trivial solution is unstable for larger values of h. The nature of the bifurcation point is supercritical [i.e., stable symmetric; see Fig. 3.7(a)], and the column smoothly begins to droop as the height is increased past its critical value. Again, this is behavior somewhat familiar from the approximate analytical solution. The stability of these equilibria can be obtained from a linear-vibration analysis. Fundamental vibration frequencies are plotted in Fig. 12.3(b). Negative values of 2 (shaded gray) are associated with unstable equilibrium states and with motions that grow exponentially (see Section 3.1). The fundamental frequency is zero at the critical height (at least for the geometrically perfect, undamped case). The curve is convex toward the origin, unlike typical characteristic curves in which a loading parameter (rather than the height) is plotted versus the frequency squared [14]—see Section 11.3. Fundamental frequencies for vibrations about the stable postbuckled equilibrium path for the perfect column are plotted also.

12.3.2 Experiments We refer back to Section 7.9.2 that showed some experimental results for a (circularcross-section) cantilever [9]. The rod was placed in an upright position and the length was incrementally increased. The results from these experiments are shown in Fig. 7.15, which suggests buckling between H = 15 and 20 cm. In this case, an estimate of the flexural rigidity was obtained from the (gravitional) droop of a horizontal cantilever and using the theoretical critical length, 1.986(EI/W)1/3 resulted in an estimated critical height very close to 20 cm. This is close to the height associated with the minimum value of the fundamental frequency, shown in Fig. 7.15(b). The column clearly exhibits a supercritical bifurcation. A flat, slender strip of polycarbonate was used to take some additional data (natural frequencies) in which the geometry of the cross section ensured unambiguous deflection in a plane. The free end of the strut was subjected to a small perturbation and subsequent oscillations were monitored by a laser vibrometer. The fundamental frequency content was then extracted, with the results shown in

18:3

P1: KAE Chapter˙12

CUFX159-Virgin

0 521 85648 5

August 14, 2007

12.4 Case Study B: A Heavy Beam

243

150

H (cm) 125

100

75

50 0

1

2

3

4

5

2

6

Figure 12.4. Fundamental frequency for a polycarbonate strip.

Fig. 12.4. The reduction in the lowest natural frequency can clearly be seen as buckling is approached (the theoretical critical length is approximately 110 cm in this case) together with an increase in the postbuckling frequencies. The form of the length–frequency relation follows the theoretical curves depicted in Fig. 12.3(b) quite closely, especially if an initial imperfection had been incorporated into the analysis. Also, the inevitable presence of a little damping has a minor effect on the frequencies, but again, this is not considered in the analysis here.

12.4 Case Study B: A Heavy Beam Suppose the column is now rotated back to its horizontal configuration (β = 0) and both ends are constrained against lateral deflection and rotation. Rather than the column height (length), it is more convenient to use the axial (imposed) end shortening of the strip as the control parameter (see Fig. 12.1), and this can also be placed into nondimensional terms by use of δ = /L [15–17]. Now we use the nondimensionalization used in Eqs. (12.9) rather than a from the previous section. In Chapter 2 2 7, we obtained the critical buckling load for a clamped–clamped beam of 4π EI/L and lowest natural frequency of 22.37 EI/(ρAL4 ) for the straight beam (of length L and flexural rigidity EI) without including the effect of gravity (acting laterally on the beam and giving a constant weight W per unit length). Now the boundary conditions when s = 0 are xe = ye = θe = 0, and the quantities me (0), p 0 , and q0 are determined by the shooting method, based on satisfying the conditions xe = 1 − δ and ye = θe = 0 at s = 1 with sufficient accuracy. A Note on Lift-Off and Self-Contact. For a very long strip, subject to end short-

ening, only a central part of the beam will tend to lift-off [18]. As the ends of the strip are pushed toward each other, the length of the proportion of the beam that

18:3

P1: KAE Chapter˙12

CUFX159-Virgin

0 521 85648 5

August 14, 2007

244

Highly Deformed Structures

does not lift off diminishes until the whole strip is characterized by a nonzero lateral deflection (other than immediately at the clamped ends). For convenience, the strip is called “short” if it does not touch the foundation between its ends. If there is a flat section resting on the foundation at both ends, the strip is called “long.” The extent to which the beam is long or short depends to a large extent on its weight/stiffness ratio. A practical aspect of this is that for the long beam a zero-frequency traveling wave is observed [18]. Another interesting feature of this type of system is that, for sufficiently high values of end shortening, self-contact may occur, in which two points on the strip contact each other and the segment between has a teardrop (or pinched-loop) shape [19]. A geometry related to this specific case will be considered in a later section of this chapter.

12.4.1 Numerical Results Equilibrium paths are depicted in Fig. 12.5 in the plane of end shortening δ versus axial load p 0 , along with some corresponding equilibrium shapes, for weight parameters w = 0, 25, 125, 250, and 343. As a reference point, for w = 0 (no weight), the fixed–fixed strip buckles at p 0 = 4π2 , and shapes at points A, B, and C along the postbuckling path are shown. This result was essentially obtained in Chapter 7 in which the increase in deflection after buckling (indicating a degree of postbuckled stiffness) was first observed. Self-contact occurs at C, when δ = 0.849 and p 0 = 72.18 [18], with midpoint deflection y(0.5) = 0.403. Under increasing δ, the near-horizontal path to the right of C is followed, with large increases in p 0 associated with small increases in δ. For heavy strips (w > 0), as δ is increased from zero, the strip is initially long but then becomes short. For w = 25, symmetric self-contact occurs when δ = 0.845, 1

B 0

C

D 125

E 250

F

G 343

H 343

I

w=0

A

125

250

0

C E

0.8 G

I B

0.6

F

0.4 H

A D

250

0.2 0 −50

0

343

125

w=0

25

50

100

150

p0 200

Figure 12.5. Equilibrium shapes and end shortening as functions of axial load for horizontal strip with weights (from left to right) w = 0, 25, 125, 250, and 343.

18:3

P1: KAE Chapter˙12

CUFX159-Virgin

0 521 85648 5

August 14, 2007

12.4 Case Study B: A Heavy Beam

245

1 δ 0.8

0.6 250 125

w=0

0.4

0.2 25

0

0

10

20

30

40

Ω

50

Figure 12.6. End shortening as a function of fundamental frequency for short strips with w = 0, 25, 125, and 250, and mode shapes; equilibrium shapes are shown in gray.

p 0 = 71.65, and ye (0.5) = 0.401. For w = 125, the symmetric shape becomes unstable when δ = 0.820 and p 0 = 69.74, and the stable equilibrium associated with tilted (asymmetric) shapes bifurcates leftward in Fig. 12.5 toward point E, where self-contact occurs with δ = 0.820 and p 0 = 60.33. These equilibrium shapes correspond only to positive lateral deflection (lift-off). In the absence of a foundation a variety of other equilibrium configurations are possible (although most of these are unstable) [20]. Small vibrations about equilibrium are shown in Fig. 12.6 in terms of the fundamental frequency for w = 0, 25, 125, and 250, along with the mode shapes for four specific cases. As the end shortening δ is increased, the fundamental frequency is zero at the transition from the long to short equilibrium shape, and then is zero again when the symmetric shape becomes unstable (as seen for weights w of 125 and 250). For w = 0, the frequency at δ = 0 is = 44.36, corresponding to the second mode of a fixed–fixed column subjected to an axial load p 0 = 4π2 . When = 19.81 for w = 0, and also when = 5.35 for w = 25, symmetric self-contact occurs, and these two curves in Fig. 12.6 are ended. For w = 125, as δ is increased beyond the value 0.820 where the symmetric shape becomes unstable, the strip tilts and the frequency increases until self-contact occurs when = 0.155 (and δ = 0.820 still). An asymmetric mode along the path for w = 250 is shown in the top-left part of Fig. 12.6. 12.4.2 Experiments It turns out that although the types of structures described in this section exhibit very large deflections it is relatively straightforward to confirm much of this behavior experimentally. Throughout this chapter some experimental studies in which thin polycarbonate strips are used are described. The general approach to measuring

18:3

P1: KAE Chapter˙12

CUFX159-Virgin

246

0 521 85648 5

August 14, 2007

Highly Deformed Structures

frequencies and mode shapes under different levels of axial loading is similar for each of the systems in this chapter. For this very flexible elastic material, the specific weight was measured at 11.2 kN/m3 and Young’s modulus was 2.4 GPa. The strips were 7.62 mm wide. Two thicknesses (0.508 and 1.016 mm) and two lengths (0.532 and 0.832 m) were used in combinations to yield three different values for the nondimensional weight w. The strip was clamped such that one end could move toward the other in 6.3-mm increments to create the end shortening (see Fig. 12.1). This is the main control parameter in this example, that is, the actual (arc) length of the beam is held constant. An alternative means of changing the system would be to feed additional material in from one side, which was effectively done for the system shown in Fig. 3.9. The strip was deflected beyond the transition from long to short equilibrium before any measurements were taken for both the static and dynamic experiments. A point-to-point laser vibrometer was used to measure the velocity at a userprescribed point on the strip (avoiding any obvious node). For modal measurements, the velocity at multiple points was taken. For δ values ranging from 0.021 to 0.917, the first four frequencies were obtained by excitation of the strip by an impact hammer at different locations along the strip. Frequency measurements were also independently confirmed by measurement of the beam response to forced excitation, that is, use of a sine sweep (from 0.008 to 50 Hz) applied to the baseplate by an electromagnetic shaker. For the modal analysis, the strip was again excited with a modal impact hammer. Data were acquired by the vibrometer (utilizing Bruel and Kjaer pulse signalprocessing software) and analyzed with ME’scope VES to generate a frequencyresponse function for each measurement point. This same approach was then used to measure the response at 30 different points along the strip. Vibration modes associated with the first few frequencies were constructed. In Fig. 12.7, experimental and analytical results are compared, with frequencies given in hertz. The open circles are associated with tests on a strip of length 0.532 m and thickness 1.016 mm, giving w = 8.145. The solid circles in Fig. 12.7 correspond to tests on a strip of length 0.532 m again, but a thickness of 0.508 mm, so that w = 32.56. Several data points at high values of end shortening are associated with strips having self-contact. For the leftmost results, the open triangles were obtained experimentally for a longer strip, with length 0.832 m and thickness 0.508 mm (corresponding to w = 124.7). In terms of higher frequencies, Fig. 12.8 shows the first four frequencies (in hertz) for the specific case w = 32.56. Black circles correspond to data acquired from an impact test; open circles denote results obtained by forced vibration. The fundamental frequency is zero when the strip becomes short at δ = 0.017 (with p 0 = 80.86). This can be inferred from Fig. 12.5. The frequency increases and then decreases. Self-contact occurs when δ = 0.844 and p 0 = 71.47. Vibration mode shapes are shown in Fig. 12.9 for δ = 0.117, with the analytical shapes on the left and the experimental shapes on the right. Further results (including cases for β = 0) can be found in Santillan et al. [18].

18:3

P1: KAE Chapter˙12

CUFX159-Virgin

0 521 85648 5

August 14, 2007

12.4 Case Study B: A Heavy Beam

247

1 δ/L 0.9 0.8 0.7 w = 8.145 0.6 0.5 w = 32.56

0.4

w = 124.7

0.3 0.2 0.1 0

0

2

4

6

8

10 Frequency (Hz)

12

Figure 12.7. End shortening as a function of fundamental frequency for horizontal strips. Solid curves correspond to w = 124.7, 32.56, and 8.145; , •, and ◦, experiment.

1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0

0

5

10

15

20

25

30

F requency (Hz)

Figure 12.8. End shortening as a function of lowest four frequencies for horizontal strip with w = 32.56. experiment: ◦ (forced) and • (free); dashed curves, theory. Self-contact is indicated by the horizontal gray lines: continuous, theory; dashed, experiment.

18:3

P1: KAE Chapter˙12

CUFX159-Virgin

0 521 85648 5

August 14, 2007

248

Highly Deformed Structures 4.30 Hz

4.47 Hz

Mode 1

8.1 Hz

7.83 Hz

Mode 2

16.46 Hz

17.0 Hz

Mode 3

Experiment

L=

Theory

Figure 12.9. First three vibration modes from analysis and experiment for horizontal strip with w = 32.56 and δ = 0.117; equilibrium shape is dashed.

A relatively highly deflected (or prestressed) elastic beam with pinned ends can also be considered to be an arch. Such a structure was analyzed by Perkins [21] in which a slightly different analytical approach was used to determine the natural frequencies of small-amplitude vibrations about highly nonlinear equilibria. He used a variational formulation to obtain the results shown in Fig. 12.10, in which the four lowest natural frequencies are plotted as a function of the nondimensional end load n. Some mode shapes at specific values of the end load are superimposed (for n = 5, 15, 21.55). The lowest natural frequency dropping to zero at the Euler buckling load (corresponding to n = π2 ) is observed together with the subsequent jump to an asymmetric mode. He also conducted some experiments, and a comparison between theoretical and measured frequencies is summarized in the table following (in which H is the separation distance between the pinned ends, and L is arc length): ω1

ω2

ω3

H/L

n

Th.

Exp.

%

Th.

Exp.

%

Th.

Exp.

%

0.8 0.35 0.06

11 15 20

26.9 13.6 4.57

25.7 13.3 4.7

4.8 2.1 2.7

77.3 62.8 53.1

75.9 61.7 52.6

1.9 1.9 0.9

146 133.7 122.8

140 135 113.7

4.7 1.7 8.0

12.5 Case Study C: A Pinched Loop An interesting extension to the analysis of the first case study can be made if the ends of the clamped beam are rotated such that they are pressed flat together, as shown in Fig. 12.11. Also shown in this figure (part c) is a snapshot of a highly deflected shape corresponding to a length beyond the buckling point for the upright equilibrium configuration. Again the parameter a is used for the nondimensionalization and the behavior is investigated through the evolution of the nondimensional

18:3

P1: KAE Chapter˙12

CUFX159-Virgin

0 521 85648 5

August 14, 2007

12.5 Case Study C: A Pinched Loop

249

160

120

80

40

25

0 5

10

15

20

End Load

–20

Figure 12.10. Natural frequencies and mode shapes of an elastic arch as functions of end load n. Reproduced with permission from ASME [21].

control parameter l = L/a. The boundary conditions are basically the same as for the clamped beam, but now we have θe (l) = −π, and although it is easy to incline the pinched support [18], we focus attention on the upright system, that is, β = π/2 in the governing equations (and again include gravitational effects). (a)

(b) L

(c)

X g

S

Y

Figure 12.11. (a) Geometry of pinched loop, (b) photographic image of experimental setup, and (c) a highly deflected configuration.

18:3

P1: KAE CUFX159-Virgin

0 521 85648 5

August 14, 2007

250

Highly Deformed Structures

(b)

(a)

7

l 6 5 4 3 2 0

0.5

1

1.5 2 y(l/2)

0

1

2

3

ω

4

Figure 12.12. Horizontal midpoint deflection of upright loop as a function of length. Solid curve, numerical; •, experimental.

Because the theoretical approach is now well established, both theoretical and experimental results are plotted together in this section. The experimental results were based on a strip of cross-sectional dimensions 25.4 mm × 0.508 mm, which corresponds to the reference length a of 0.167 m, and the nondimensional vibration √ frequency ω of 0.130 times the dimensional frequency (where ω = a/g) that is √ due to the scaling of time: t = T g/a—again note the difference with Eqs. (12.9). In this section, most of the results are presented in terms of nondimensional quantities of frequency ω and the coordinates x and y of the midpoint (where s = 0.5). Equilibrium results for the loop are depicted in Fig. 12.12(a). The horizontal deflection of the midpoint is plotted as a function of the length of the loop, with the appearance of a critical length of l = 4.50 (signifying the onset of a supercritical pitchfork bifurcation). A typical experimental frequency spectrum is depicted in Fig. 12.13, obtained with the laser vibrometer discussed in the previous section. The strip from which rms Velocity (m/s)

Chapter˙12

0

10

20

30

40 50 Frequency (Hz)

Figure 12.13. Typical frequency spectrum for upright loop with l = 3.06; main peaks at 1.81, 10.73, 20.16, 30.53 Hz.

18:3

P1: KAE Chapter˙12

CUFX159-Virgin

0 521 85648 5

August 14, 2007

12.6 Case Study D: A Beam Loaded by a Cable

251

140 ω 120 100 80

Figure 12.14. Lowest four natural frequencies of upright loop as functions of length.

60 40 20 0 0.5

1

1.5

2

2.5

3

these data were collected is 0.5096 m long (l = 3.06). The results were averaged over four time series. The four lowest-dimensional (in-plane) frequencies are distinct at 1.81, 10.73, 20.16, and 30.53 Hz. This procedure was repeated for several thicknesses and a number of lengths. The four lowest measured frequencies are denoted by circles, squares, and triangles in Fig. 12.14 over a range of nondimensional lengths (these vibrations are all about the trivial equilibrium configuration for this range of l). Solid curves represent the analytical results, and in general we see the anticipated reduction in the natural frequency as the loop length increases, thus observing the softening effect of gravity (for the upright orientation). The mode shapes were also determined experimentally and analytically for some cases. The first four mode shapes for the upright case with l = 3.48 and a strip thickness of 0.508 mm are depicted in Fig. 12.15, along with the equilibrium shape. The corresponding measured dimensional frequencies are 1.3, 8.26, 15.8, and 23.9 Hz. As expected, the second and fourth modes are symmetric with respect to the vertical axis. Finally, the fundamental vibration frequency is plotted in Fig. 12.12(b) as a function of the nondimensional length for relatively long loop lengths (including postcritical drooping). The experimental data points were obtained as the average values from multiple-frequency spectra. As the length of the loop is increased, the fundamental frequency decreases until it is effectively zero at the critical length l ≈ 4.5. Extrapolation of measured fundamental frequencies at smaller lengths to the length at zero frequency could be used to predict the critical length (see Chapter 11). As the length is increased further and the loop droops (postbuckling deformation), the fundamental frequency increases.

12.6 Case Study D: A Beam Loaded by a Cable In the final section of this chapter, we again consider the free vibrations of an elastic structural system characterized by large deflection that is due primarily to axial-load

l

3.5

18:3

P1: KAE Chapter˙12

CUFX159-Virgin

0 521 85648 5

252

August 14, 2007

Highly Deformed Structures

Mode 1

Mode 3

Mode 2

Mode 4

Figure 12.15. First four mode shapes for upright loop. Dashed curve, equilibrium (analytical); solid curve, mode (analytical); •, equilibrium (experimental); ◦, mode (experimental).

effects. A prismatic cantilever beam has a cable attached to its free end that is then pulled in a direction depending on the location of its far-end attachment point [22]. This can be considered as somewhat intermediate between the cantilever subject to an increasing end load that maintains its direction and Beck’s problem (Chapter 7), although it can be shown that this is a conservative system. If the distant end of the cable is attached to the beam on its axis, then Timoshenko and Gere [5] showed that the critical load is equal to the Euler load (but for a pinned–pinned beam). If the far-end attachment point is offset, then symmetry is broken and the increase in load results in the nonlinear deflection of the cantilever in much the same way that initial geometric imperfections influence the behavior of axially loaded systems in general. The cable loading has a considerably greater effect than gravity and hence weight is not included in this analysis. A schematic of the system is shown in Fig. 12.16. The primary method of analysis is again based on a shooting method solution of the boundary-value problem. Again some experimental verification is presented and some FE solutions are also included. Here, the tension in the (axially very stiff) cable is the principal control parameter (as the far end is pulled through the attachment point), with natural frequencies and mode shapes again characterizing the vibration of small-amplitude motion about even highly deflected equilibria. We still basically have the same form of the governing static equations but with the addition of the horizontal component of the end force that complicates the

18:3

P1: KAE Chapter˙12

CUFX159-Virgin

0 521 85648 5

August 14, 2007

12.6 Case Study D: A Beam Loaded by a Cable

A θ

253

x

s

y

L

C (a, b)

P B Figure 12.16. Geometry of a slender cantilever column loaded by a cable passing from the free end to a point near the base.

boundary conditions. Given the x and y offset coordinates a and b, respectively, the boundary condition (moment) at the (s = 0) clamped end is Me (0) = Pve (L)b − Phe (L)a,

(12.16)

and the load is adjusted (by use of Newton’s method) until Me (L) = 0 is satisfied. In the results to be presented, some comparisons will be made with FE results obtained with abaqus. Beam elements (B31), suitable for large deflections, were used in which 1000 elements were employed to ensure spatial convergence. A truss element was used for modeling the cable, and an extreme negative thermal load was applied to drastically reduce the length of the cable and thus pull on the end of the cantilever. A path-following algorithm based on Riks method was employed [23, 24], that is, the same approach as used for the highly deflected cantilever in Chapter 9. Figure 12.17 shows an experimental setup for the cable–beam system. A thin polycarbonate beam of dimensions 0.762 m long with a rectangular cross section of 25.4 mm × 4.8 mm is configured as a cantilever. The elastic modulus and density were given earlier in this chapter. The cable was made of high-strength woven steel wire with a stiffness of 11.67 kN/m. The cable was then connected to the base Acelrometr Shak er Beam

Mounting Bloc

Adapter Plate

k Cab

Fle le Laser T arget

xib

le

Joint Load Cel

Figure 12.17. The experimental system of cantilever beam and end-loading cable.

18:3

P1: KAE Chapter˙12

CUFX159-Virgin

254

0 521 85648 5

August 14, 2007

Highly Deformed Structures

Figure 12.18. A typical frequency-response spectrum from the cable–beam system.

by a flexible joint with the other end attached to the tip of the cantilever. A load cell was incorporated into the wire connection in order to monitor the tensile force in the system. In some cases, the wire was replaced with shorter lengths to facilitate accurate tension measurements. The base of the cantilever was mounted to an electromagnetic shaker, with the excitation measured by an accelerometer. The response of the beam was again measured by a laser velocity vibrometer in the usual way. Standard data acquisition and signal-processing data were used (the Bruel and Kjaer pulse system), and a typical frequency response is shown in Fig. 12.18. This particular spectrum was taken from a system in which the nondimensional offset was a/L = 0.0375, b/L = 0.0167, and the applied load was P/Pcr = 1.12, in which Pcr corresponds to the elastic critical (Euler) load for the underlying case with no offset. The excitation employed here was a pseudorandom input signal over the range 0 to 100 Hz and with a sampling rate of f = 0.03125 Hz. Appropriate windowing and averaging were used to improve the quality of the data [25]. Vibration mode shapes were then extracted by the standard approach at a specific level of the cable tension. Figure 12.19 shows the static and dynamic response for the beam with a/L = 0.0375, b/L = 0.0167. Part (a) shows both shooting results and abaqus together with some experimental data points for the lateral deflection of the tip. The inset shows a couple of deflected shapes. Part (b) shows the corresponding fundamental frequency, which also changes as a function of the tension in the cable. The frequency does not exhibit the type of monotonic decay we observed in earlier sections of this book. The frequency in part (b) is nondimensionalized bythe fundamental frequency for a cantilever beam with no cable attached (ω = 3.516 EI/mL4 ), and it is interesting to see the effect the cable has (for both FE and shooting results) as the

18:3

P1: KAE Chapter˙12

CUFX159-Virgin

0 521 85648 5

August 14, 2007

12.6 Case Study D: A Beam Loaded by a Cable

255

1.5 P/Pcr

a/L = 0.0375 b/L = 0.0167

1.0

0.5

P/Pcr = 1.01

0

0.1

0.2

0.3

0.4

0.5

P/Pcr = 1.30

0.6

0.7

1.0

1.5

2.5

2.0

Ytip / L

1

1, P = k = 0

Figure 12.19. (a) Load-deflection characteristic with b/L = 0.0167 and (b) the corresponding fundamental natural frequency. The solid curves represent the FEA solution, dashed for analytical results obtained with the shooting method. 1.5 P/Pcr

a/L = 0.0375 b/L = 0.075

1.0

0.5

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7 Ytip / L

1.0

2.0

4.0

3.0 1

1, P = k = 0

Figure 12.20. (a) Load-deflection characteristic with b/L = 0.075 and (b) the corresponding fundamental natural frequency. The solid curves represent the FEA solution, dashed for analytical results obtained with the shooting method.

tension tends to zero, i.e., even at zero tension the cable provides some constraint at the “free” end. Figure 12.20 shows similar results but with a larger static offset: a/L = 0.0375, b/L = 0.075. The behavior is seen to be quite sensitive to the magnitude of the cable offset. A similar procedure was followed to obtain the higher frequencies and the first four, for the smaller offset, are shown in Fig. 12.21. The gray dashed curve at P/Pcr = 1.12 indicates the level at which the frequency spectrum illustrated in Fig. 12.18 was taken. Although the frequencies are separated in a somewhat typical

18:3

P1: KAE Chapter˙12

CUFX159-Virgin

0 521 85648 5

August 14, 2007

256

Highly Deformed Structures 1.5

P/Pcr

1.0

0.5

0

10

20

30

40

50 1

60

1, P = k = 0

Figure 12.21. The four lowest natural frequencies plotted as functions of the tension in the cable with b/L = 0.0167. The solid curves represent the FEA solution, dashed for analytical results obtained with the shooting method.

spread for beam vibrations the effect of the tension loading depends on the specific frequency. Some vibration modes shapes are shown in Fig. 12.22 for the larger offset value. There is good agreement between the results from a shooting analysis (eigenvectors), abaqus, and experimentally determined mode shapes. There is an arbitrary phase in some of these plots. Finally, one of the potential applications of this type of system is solar sails (see also Section 9.6). In these innovative structural systems, the idea is to use the Sun’s photons for propulsion based on a very lightweight but large surface area, rather like a kite. However, to keep such a membrane taut, it would need to be attached to relatively stiff but inevitably slender booms [26] that might typically lead to offset loading of the type considered in this section. There might also be some advantage to using tapered booms (see Section 8.3 and Holland et al. [27]).

12.7 The Softening Loop Revisited Before we leave this chapter, a brief result is given that uses the elastica to solve the generating example described in Section 3.5, again with the shooting method. By incorporating a geometrically nonlinear (softening) moment–curvature relation from experiments, we obtain the subcritical pitchfork bifurcation result shown in Fig. 12.23. The dashed curve is the result when the effect of a small initial geometric imperfection is included in the analysis (i.e., the symmetry is broken). The lower part

18:3

P1: KAE Chapter˙12

CUFX159-Virgin

0 521 85648 5

August 14, 2007

12.7 The Softening Loop Revisited

257 (a)

(b)

(c)

Figure 12.22. The lowest four vibration modes with b/L = 0.075: (a) shooting, (b) abaqus, and (c) experimental.

8

h 7 q = 0.01 A 6

q=0

5

4

−5

−2.5

0

1.5

y

y 1

0.5 0 −1

0

1

2

x

3

2.5

z(h)

5

1.5 1 0.5 0

0

1

2

3

4

z 5

Figure 12.23. Equilibrium paths for softening loop, where q is an initial imperfection, q = 0 or q = 0.01. Two views of the deflected configuration corresponding to point A are also shown [10].

18:3

P1: KAE Chapter˙12

CUFX159-Virgin

0 521 85648 5

August 14, 2007

258

Highly Deformed Structures

of this figure shows two views of the drooped configuration immediately after the initial instability. The transition between the upright and this drooped configuration is dynamic because there is no locally adjacent stable equilibrium, thus confirming the subcritical qualitative behavior from Figs. 3.10(c) and 3.10(d). Further details of the solution technique can be found in Plaut and Virgin [10]. In experiments on a loop with axisymmetric section properties, the natural frequencies consist of out-of-plane, in-plane, and twisting modes, and the lowest mode for each case is shown for oscillations about the (prebuckled) upright position in Fig. 12.24. Here h is a nondimensional parameter associated with one-half the length of the loop. Appropriately nondimensionalized, this also confirms part of the qualitative picture from Figs. 3.10(b) and 3.10(d), as well as the experimental data in Fig. 3.14, thus again confirming the trend of the lowest natural frequency toward zero at buckling.

h

(a)

6 5.5 5 4.5 4 3.5 3 2.5 2 0

(b)

1

2

(c)

3

4

5

(d)

Figure 12.24. (a) Half-length versus lowest frequencies of each type, and corresponding modes of vibration, (b) out-of-plane (——–); (c) in-plane (- - - - -); (d) twist (........).

18:3

P1: KAE Chapter˙12

CUFX159-Virgin

0 521 85648 5

August 14, 2007

References References [1] [2] [3] [4] [5] [6] [7] [8] [9]

[10]

[11] [12] [13]

[14] [15] [16] [17] [18] [19] [20] [21] [22]

[23]

J.F. Wilson. Dynamics of Offshore Structures. Wiley, 2002. Y. Bai and Q. Bai. Subsea Pipelines and Risers. Elsevier, 2005. M.S. El Naschie. Stress, Stability and Chaos in Structural Engineering: An Energy Approach. McGraw-Hill, 1990. D.W. Jordan and P. Smith. Nonlinear Ordinary Differential Equations. Oxford University Press, 1999. S.P. Timoshenko and J.M. Gere. Theory of Elastic Stability, 2nd ed. McGraw-Hill, 1961. J.P. Cusumano. Low-Dimensional, Chaotic, Nonplanar Motions of the Elastica: Experiment and Theory. Ph.D. dissertation, Cornell University, 1990. C. Gatti-Bono and N. C. Perkins. Dynamic analysis of loop formation in cables under compression. International Journal of Offshore and Polar Engineering, 12:217–22, 2002. D. Addessi, W. Lacarbonara, and A. Paolone. On the linear normal modes of planar pre-stressed curved beams. Journal of Sound and Vibration, 284:1075–97, 2005. L.N. Virgin and R.H. Plaut. Postbuckling and vibrations of linearly elastic and softening columns under self-weight. International Journal of Solids and Structures, 41:4989–5001, 2004. R.H. Plaut and L.N. Virgin. Three-dimensional postbuckling and vibration of vertical half-loop under self-weight. International Journal of Solids and Structures, 41:4975–88, 2004. C.J. Goh and C.M. Wang. Generalized shooting method for elastic stability analysis and optimization of structural members. Computers and Structures, 38:73–81, 1990. M.A. Crisfield. Nonlinear Finite Element Analysis of Solids and Structures, Vol. 2: Advanced Topics. Wiley, 1997. A.G. Greenhill. Determination of the greatest height consistent with stability that a vertical pole or mast can be made, and of the greatest height to which a tree of given proportions can grow. Proceedings of the Cambridge Philosophical Society, 4:65–73, 1881. K. Huseyin. Multiple Parameter Stability Theory and Its Applications. Oxford University Press, 1986. R. Schmidt and D.A. DaDeppo. Large deflection of heavy cantilever beams and columns. Quarterly Journal of Applied Mathematics, 28:441–4, 1970. C.Y. Wang. A critical review of the heavy elastica. International Journal of Mechanical Sciences, 28:549–59, 1986. S.-B. Hsu and S.-F. Hwang. Analysis of large deformation of a heavy cantilever. SIAM Journal on Mathematical Analysis, 19:854–66, 1988. S. Santillan, L.N. Virgin, and R.H. Plaut. Post-buckling and vibration of heavy beam on horizontal or inclined rigid foundation. Journal of Applied Mechanics, 73:664–71, 2006. S. Santillan, L.N. Virgin, and R.H. Plaut. Equilibria and vibration of a heavy pinched loop. Journal of Sound and Vibration, 288:81–90, 2005. J.G.A. Croll. Some comments on the mechanics of thermal buckling. The Structural Engineer, 83:127–32, 2005. N.C. Perkins. Planar vibration of an elastica arch: Theory and experiment. Journal of Vibration and Acoustics, 112:374–9, 1990. H.G. McComb. Large deflection of a cantilever beam under arbitrarily directed tip load. Technical Report, Technical Memorandum 86442, NASA Langley Research Center, 1985. E. Riks. The application of Newton’s method to the problem of elastic stability. Journal of Applied Mechanics, 39:1060–6, 1972.

259

18:3

P1: KAE Chapter˙12

CUFX159-Virgin

260

0 521 85648 5

August 14, 2007

Highly Deformed Structures [24] D.B. Holland, I. Stanciulescu, L.N. Virgin, and R.H. Plaut. Vibration and large deflection of cantilevered elastica compressed by angled cable. AIAA Journal, 44:1468–76, 2006. [25] T.G. Beckwith, R.D. Marangoni, and J.H. Lienhard. Mechanical Measurements. Addison-Wesley, 1993. [26] I. Stanciulescu, L.N. Virgin, and T.A. Laursen. Finite element analysis of slender solar sail booms. Journal of Spacecraft and Rockets, 44:528–37, 2007. [27] D.B. Holland, L.N. Virgin, and R.H. Plaut. Large deflections and vibration of a tapered cantilever pulled at its tip by a cable. Journal of Sound and Vibration, 2007, to appear.

18:3

P1: KAE Chapter-13

CUFX159-Virgin

0 521 85648 5

August 14, 2007

13 Suddenly Applied Loads

13.1 Load Classification The early parts of this book focused attention on the dynamics of structures in which axial loading was increased quasi-statically. Thus dynamic response was typically considered under effectively set axial-loading conditions and observed in terms of free vibrations. However, it is just as common for either the axial load to be applied dynamically or an axially loaded structure to be subjected to dynamic lateral loading as well [1]. In these cases, it is not uncommon for the maximum response to occur during transient motion. In this chapter, we look at a number of different scenarios in which a structure with a constant axial load is then subject to various types of (dynamic) disturbance forces. These will range from a slow, but nonnegligible, increase in axial loading, to suddenly applied loading (e.g., an impulse or step input [2–5]). We have already seen (e.g., Fig. 5.7) that the free response of an undamped nonlinear system is strongly influenced by the magnitude of the initial conditions. In some, a large initial velocity can be considered as an impulse. Before moving on to consider some specific structural systems, we return to the simple (underdamped) linear oscillator from Chapter 3 (Fig. 3.1). However, we now add a step input applied directly to the mass such that the governing equation of motion is given by m¨x + cx˙ + kx = F0 ,

(13.1)

in which we assume a system initially at rest, that is, x(0) = x(0) ˙ = 0. It is easy to show (by use of Laplace transforms for instance) that the response of this system is given by x(t) ωn −ζωn t = x(t) ¯ =1− e sin (ωd t + φ1 ), (13.2) F0 /k ωd √ where φ1 = cos−1 ζ, ω2n = k/m, ωd = ωn 1 − ζ 2 , ζ = c/(2mωn ). A typical response (normalized with respect to the magnitude of the input) is shown in Fig. 13.1 with ωn = 1, ζ = 0.1. This type of step response can be characterized by a number of metrics, for example, how long the transient takes to die out or by how much the response overshoots the final resting position initially. Clearly, damping has a key role to play here. It can be shown that the maximum percentage overshoot (OS, i.e., x¯ over and above unity) is √

OS = 100e−πζ/

1−ζ 2

,

(13.3) 261

18:8

P1: KAE Chapter-13

CUFX159-Virgin

0 521 85648 5

August 14, 2007

262

Suddenly Applied Loads tr

2 x1.5

= 0.1 OS

1 0.5

F0

ts

t

0 0

10

20

t

30

40

50

Figure 13.1. The step response of a linear system.

that is, twice the static response for an undamped (ζ = 0) system, and this occurs after the rise time (tr ). Thus we may envision a load applied suddenly to be stronger in its effect than the same load applied quasi-statically (i.e., on a much slower time scale than the flexural dynamics of the system). Similarly, the time taken for the response to decay to within 5% (say) of the final value is called the settling time (ts ) and is often approximated by 3/(ζωn ) [6]. So much for the step response of a linear system. Things become more interesting when we study the effect of a step input on the response of a nonlinear system, and specifically one in which the nonlinear system relates to the equilibrium of an axially loaded structure. Not only may a suddenly applied load cause additional axial loading through large-amplitude effects but it may also result in collapse because of the traversing of an adjacent underlying unstable equilibrium (local potentialenergy hilltop), or snap-through, for example.

13.2 Back to Link Models A good example of where stability in the large may be especially important is for imperfection-sensitive structures. If we reconsider the system shown in Fig. 5.9(a) with C = 0 and thus α = 1 we obtain the equilibrium condition =

(sin θ − sin θ0 ) cos θ . sin θ

(13.4)

Note that because we are now interested in relatively large excursions from equilibrium we do not use the approximation given by Eq. (5.45). The dynamic response is governed by θ¨ + ω2n [(sin θ − sin θ0 ) cos θ − sin θ] = 0,

(13.5)

where we have already established the relation between the linearized natural frequency and axial load (see Section 5.3). We now investigate the robustness of equilibrium to relatively large disturbances as a function of the initial imperfection. Without damping we can view

18:8

P1: KAE Chapter-13

CUFX159-Virgin

0 521 85648 5

August 14, 2007

13.2 Back to Link Models

.

263

1

(a)

θ

.

1

0.5

0.5

0

0

−0.5

−0.5

−1

−2

−1

0

1

θ

2

(b)

θ

−1

−2

−1

0

1

θ

2

Figure 13.2. Contours of constant total energy indicating the changing catchment region as a function of initial imperfection.

free oscillations within the context of constant total energy, the level of which is determined by the initial conditions. For example, in Fig. 13.2 are shown two contour plots of total mechanical energy when the axial load has been set to = 0.5, that is, at 50% of the critical load for the initially perfect geometry. In part (a), the initial imperfection is θ0 = 0.01, and in part (b) θ0 = 0.1. We see that, as expected for an imperfection-sensitive structure, the difference in potential energy between the stable and closest unstable equilibrium has diminished for the larger initial imperfection, and thus, given a certain initial velocity or step input, we would have a greater chance of an escaping solution. For example, when θ0 = 0.01 we have a stable equilibrium at θe = 0.020 with adjacent saddle points at θe = −1.0538 and 1.0404, and when θ0 = 0.1 we have a stable equilibrium at θe = 0.2055 with the adjacent saddle points at θe = −1.1043 and 0.9654. Going back to Fig. 5.7, we saw some large-amplitude motion that was due to initial conditions somewhat distant from equilibrium. These were, of course, highly nonlinear, and in general recourse is made to numerical integration to solve these types of equations of motion. However, the large excursions generated by the application of a sudden load (or relatively large initial conditions) may lead to escaping solutions, which correspond in some sense to a dynamic buckling load. Consider the link model shown in Fig. 5.16. Suppose the load is held fixed at p = 0.005. In this case, there are three relevant equilibria: θe = 0.036789 (stable), θe = 0.325084 (unstable), and θe = 0.813625 (stable). For a system at rest (equilibrium), we may prescribe an initial velocity such that enough energy is imparted to the system that the subsequent transient traverses the hilltop. We can extract the critical velocity from the total energy and find θ˙ 0 ≈ 0.0496 if θ(0) = 0.036789. This can be considered as an impulse given to the system, with an initial velocity of θ˙ (0) > 0.0497 resulting in motion that goes beyond the distant equilibrium—undergoing oscillations that are far from sinusoidal. The trajectory passes through the unstable equilibrium position: a potential-energy hilltop.

18:8

P1: KAE Chapter-13

CUFX159-Virgin

0 521 85648 5

August 14, 2007

264

Suddenly Applied Loads 1

(a)

1

λ f (t)

_ ξ+ξ M ω12 = K/M λ c = K/2

λc λ

F = K ( ξ + b ξ3 )

(b)

_

1

0.8

λS λC

_

(c) 1.0

0.9

λs

0.6 0.4

0.8

0.2 0.7 0.2

0.4

−0.06 −0.03

0.03

0.06

_

Figure 13.3. (a) A schematic of a link model, (b) equilibrium paths for the perfect and imperfect system illustrating an unstable-symmetric point of bifurcation, and (c) imperfection sensitivity [7].

An Approximate Analysis. An early study of link models subject to suddenly ap-

plied loading can be found in Budiansky [7]. In this paper, an approximate treatment was given based on perturbation theory for imperfection-sensitive structures (in which case instability is associated with some kind of severe collapse). We start by considering the case in which a step load is applied to a link model. Imperfection sensitivity was first encountered in Subsection 3.4.3 and a specific example given in Section 5.3. We again consider a simple model, shown in Fig. 13.3(a). Following the analysis in Budiansky [7] the nonlinearity in this system is solely due to the spring characteristic rather than to any kind of large-deflection effect encountered in Chapter 5. Also, the bars (of unit length) are assumed to be rigid but weightless, and all the mass is concentrated at the central hinge (to give a natural frequency of unity). Adopting an approximate (Galerkin) approach to this system, we obtain an equation of motion,

1 ¨ λ f (t) ¯ λ f (t) 3 ξ + bξ = ξ+ 1− ξ, λc λc ω21

(13.6)

18:8

P1: KAE Chapter-13

CUFX159-Virgin

0 521 85648 5

August 14, 2007

13.2 Back to Link Models

265

with underlying (cubic) equilibrium paths of the form [for f(t) = 1] ¯ (1 − λ/λc )ξ + bξ3 = (λ/λc )ξ.

(13.7)

Given b = −1, the solutions of Eq. (13.7) are plotted in Fig. 13.3(b) for both the perfect (ξ¯ = 0) and imperfect (ξ¯ = 0.025) cases. We thus confirm that in the presence of an initial imperfection (which may also be a small lateral load) the load-carrying capacity of the structure is somewhat diminished. The extent of the reduction of the maximum load is thus given by √ 3 3 3/2 ¯ s /λc ) = 0, (−b)1/2 |ξ|(λ (13.8) [1 − (λs /λc )] − 2 and this expression is plotted in part (c) of Fig. 13.3. Assuming the axial load λ is applied quasi-statically (as in Chapter 5) then these results clearly resemble those shown in Fig. 5.10. For the specific case shown (ξ¯ = 0.025), the maximum load (λs = 0.855) corresponds to a maximum deflection of ξ = 0.27. Now, suppose the load is applied suddenly as a step input, for example, by instantaneously removal of the support from a weight attached to the structure. In this case, a first integral can be performed (assuming there is no damping present) on Eq. (13.6), and from the resulting energy contours the maximum displacement (when ξ˙ = 0) is obtained from 3 ¯ /4 = (λ/λc )ξ. [1 − (λ/λc )] ξmax /2 + bξmax

(13.9)

This relation is shown in Fig. 13.4(a). The maximum deflection for which bounded solutions exist is defined as the dynamic buckling load λD and is determined from dλ/dξmax = 0, that is, √ ¯ D/λc ) (13.10) [1 − (λD/λc )]3/2 = 1.5 6(−b)1/2 |ξ|(λ now corresponds to a maximum (step) load of λD = 0.821, as shown in Fig. 13.4(a) with a maximum deflection of ξ = 0.345. (a)

c

(b)

_

1

D 0.8

s

_

0.6 0.4

D

0.2

0.2

0.4

0.6 max

s

c

Figure 13.4. (a) Step loading vs. maximum deflection and (b) magnitude of the dynamic buckling load relative to the static case [7].

18:8

P1: KAE Chapter-13

CUFX159-Virgin

0 521 85648 5

August 14, 2007

266

Suddenly Applied Loads (a)

0.5 0.4

0.4

0.3

0.3

0.2

0.2

0.1

0.1

5

10

15

20

25

(b)

0.5

200

30

400

600

1000

800

t

t

Figure 13.5. (a) Response of the structure that is due to a step load and (b) slow evolution of the load.

Equations (13.8) and (13.10) can now be combined to give the result

1 − (λD/λs )(λs /λc ) 1 − (λs /λc )

3/2 =

√ λD 2 , λs

(13.11)

which is plotted in Fig. 13.4(b), in which the same level of initial imperfection is used for both the static and dynamic cases. For example, if the structure has an initial imperfection of ξ¯ = 0.025, then there is a 0.821/0.855 = 0.96 reduction in the maximum step load (relative to maximum statically applied load), as shown by the dashed line in Fig. 13.4(b). Thus we see, as anticipated, that when the load is suddenly applied the (imperfection-sensitive) structure is able to withstand only a reduced loading condition, and this effect is proportionately lower for a structure with a larger initial imperfection. We finally conduct a couple of simple numerical simulations of this system. Applying the step load to Eq. (13.6) and using zero initial conditions leads to the results shown in Fig. 13.5(a). The preceding analysis gave a critical dynamic buckling load of λ = 0.821, and results are shown for the cases λ = 0.82, “stable,” and λ = 0.822, “unstable.” In part (b) the load parameter λ is very slowly evolved from zero (as was done in Chapter 5) until the system loses stability close to t ≈ 850. It is worth mentioning here that given the relatively small level of initial imperfection there is not a large difference between the static and dynamic conditions. Also, the trajectory shown in Fig. 13.5(b) is not quite in equilibrium but rather the load is incremented in a large number of small steps as a function of the time step. These results do not include the effect of damping, and, as such, the results of numerical error may be an issue [8]. An alternative approach to this type of problem was conducted in Thompson [9] with the introduction of the concept of an astatic buckling load. This approach was also tested against some experiments on a continuous type of thin elastic arch structure, an example of which is shown in Fig. 13.6. The top-right-hand portion of this figure shows an arch with pinned ends but a rigid connection at the center and also

18:8

P1: KAE Chapter-13

CUFX159-Virgin

0 521 85648 5

August 14, 2007

13.3 Dynamic Buckling of a Plate

267

Offset, ε (inches) Figure 13.6. Step loading of a shallow arch. Reprinted with permission from the author [9].

shown in its deflected configuration. This structure was subject to a number of previous static tests [10] [as an example of an unstable-symmetric (subcritical) point of bifurcation], but here the suspended weight is suddenly released. The magnitude of the mass () and the offset from the apex (initial imperfection ) are then related to the (dimensional) dynamic buckling results of Fig. 13.6, in which M represents the static load, D is the dynamic load, and N is the theoretical result from Thompson [9]. Again we observe the detrimental effect of applying the load suddenly. In general, the applied force may have a finite duration, and thus the structure may be subject to a pulse. For longer durations (relative to the natural dynamics of the structure), we approach the previous results of the step load. For relatively short pulses, we approach an impulse, which we have already shown has an equivalence to a nonzero initial velocity. The effect of duration length was also considered in Thompson [7] based on the earlier work described in Hutchinson and Budiansky [11] as well as extensive studies in Simitses [12].

13.3 Dynamic Buckling of a Plate In Subsection 10.1.4, we considered a simply supported rectangular plate subject to a uniaxial load. Now, suppose this load is applied dynamically. Such a case was considered in Zizicas [13] and reproduced in Bulson [14] based on small-deflection plate theory but included the effect of an initial imperfection. In this case, it can be shown that the governing equation of motion is D∇ 4 (w − w0 ) + ρ

∂ 2w ∂ 2w − N = 0, x ∂t 2 ∂x 2

(13.12)

18:8

P1: KAE Chapter-13

CUFX159-Virgin

0 521 85648 5

August 14, 2007

268

Suddenly Applied Loads 2

10 _ Q0

1.2

1.05

30

1.0 0.95

25 20 15 10

0.8 0.7

5 0.6

2

4

6

8

10

Figure 13.7. Deflection of a simply supported plate subject to a dynamic axial load [13].

in which w0 is given by Eq. (10.49). The critical load and lowest natural frequency are given by Eqs. (10.46) and (10.45) (with m = n = 1), and again these provide convenient values with which to nondimensionalize the equation of motion, that is, introducing Nx /Ncr = α and ωt = τ, we can rewrite Eq. (13.12) as d 2 [δ/Q0 ] δ + (1 − α) − 1 = 0, 2 dτ Q0

(13.13)

in which δ is the central deflection caused by the dynamic load and Q0 is the magnitude of the initial imperfection [see Eq. (10.49)]. Solutions to Eq. (13.13) are shown in Fig. 13.7. The axial load tends to magnify the initial imperfection according to Eq. (10.50), and it is apparent that the maximum response of the oscillation when the load is applied dynamically is larger than that for the quasi-static case. For example, when the magnitude of the axial load is 60% of the critical magnitude (α = 0.6), the static deflection is 2.5 that of the initial value [from Eq. (10.50)], whereas the suddenly applied load at this level results in a peak-to-peak response that oscillates between δ/Q0 = 1 and δ/Q0 = 4 (and centered on δ/Q0 = 2.5). This result was anticipated from the introduction to this chapter and the overshoot of 100% for an undamped linear system subject to a step input. Clearly, we have the result that if the magnitude of the suddenly applied load is equal to, or greater than, the critical static load, then a monotonic growth of deflection occurs without bound (keeping in mind the limitations imposed when small-deflection theory is used).

13.4 A Type of Escaping Motion In the next chapter, we shall focus attention on harmonically excited systems. In that case, most of the interest involves steady-state behavior. But even with harmonic excitation we can still expect a certain amount of transient behavior: For example, a system starting from rest will take awhile to settle down. It may very well be that the largest excursions (and hence proximity to instability) will occur during this transient stage [15, 16]. We go back to the basic form of an equation of motion in the

18:8

P1: KAE Chapter-13

CUFX159-Virgin

0 521 85648 5

August 14, 2007

13.4 A Type of Escaping Motion

269

vicinity of a transcritical bifurcation [Eq. (3.53)] and appeal to the analogy of a small ball rolling on the underlying potential-energy surface given by V=

x2 x3 −µ . 3 2

(13.14)

For negative values of the control parameter µ, we have a stable (trivial) equilibrium state and an unstable equilibrium state (a maximum of the potential energy). As µ becomes less negative, these equilibria approach each other and interchange stability at µ = 0 (see Fig. 3.6). However, it is important to realize that the domain of stability (against disturbances) changes. For an undamped system, we can construct a separatrix emanating from the hilltop (saddle) as a contour of constant total energy. Hence, any trajectory starting within this region will lead to constrained, or bounded, motion. But, again, we see the possibility of some trajectories escaping. This scenario is certainly complicated by the presence of damping. The area within the separatrix contains those initial conditions that do not lead to escape. However, this area changes (shrinks) as we approach the critical (buckling) condition. Thus we can imagine a situation in which a given step input for a relatively large negative level of µ will not cause escape (leading to infinity) whereas the same step applied to the system with a µ value less negative may very well lead to escape. Clearly, the magnitude of the step plays a crucial role as well, as seen in the previous section. Thus we have effectively described the two scenarios found in Fig. 1.1 at the start of this book. A structure may buckle because of: r a deteriorating stiffness caused by increasing axial loading (the natural frequency characterizes this essentially static behavior) or r an excessive disturbance applied to a structure with a given level of axial loading. This is essentially a transient, dynamic behavior. Returning to the case of a harmonically forced, axially loaded structure, we can again consider a system described by Eq. (13.1), but now, rather than a linear spring, we shall assume a force that is quadratically related to deflection. Changing the step input to a harmonic drive, we consider x¨ + ζ x˙ + x(1 − x) = F2 sin (t + φ).

(13.15)

Assuming the initial conditions are zero, we have the possibility that the forcing may be sufficient to cause the response to exceed x = 1, thus leading to solutions escaping to infinity. However, this is a more involved issue than finding the critical initial conditions of the previous case. We can make use of the ball rolling on the potential-energy surface analogy but now the surface itself is shaken harmonically (thus resulting in a 3D phase space). It turns out that horizontally shaking the potential-energy surface is akin to transmissibility [17], a related but slightly different case from direct forcing, and thus an additional factor of 2 also appears in the forcing magnitude [18]. That is, the mass is not subject to a direct force, but rather indirectly through a base displacement. Clearly, the larger F is, the greater the likelihood of escape, but we also anticipate a resonant type of effect if the forcing

18:8

P1: KAE Chapter-13

CUFX159-Virgin

0 521 85648 5

270

August 14, 2007

Suddenly Applied Loads

Figure 13.8. Parameter space indicating whether transients escape.

frequency is somewhat close to the linear natural frequency of the system (unity). Furthermore, it would not be unreasonable to expect a cosine function to generate larger transients starting from t = 0 than a sine function would, and thus φ has an effect. Rather than simply choose a number of forcing parameters and simulate Eq. (13.15) to determine whether steady-state oscillations persist, we can gain a more complete picture by conducting a thorough investigation of parameter space (F, ) by dividing it into a fine grid and labeling the outcome. An example of such a plot is shown in Fig. 13.8. What this figure describes is the result of many thousands of numerical simulations (all starting from zero initial conditions) and the areas shaded white indicate regions of parameter space that resulted in nonescaping behavior. The black-shaded regions led to escaping motion, which can be thought of as dynamic buckling. It can be seen that the boundary between escape and no-escape is not simple, and in fact exhibits certain fractal properties [19]. That is, on close inspection the border is nonsmooth (and self-similar at finer and finer scales) such that given any reasonable (small) uncertainty in the forcing parameters it may be difficult to tell whether the motion will escape or not, even though this is a thoroughly deterministic scenario. There is an increased likelihood of escape when the forcing frequency () is relatively close to one, as there is some associated softening resonant effect. This aspect of nonlinear behavior will be explained more throughly in later chapters.

18:8

P1: KAE Chapter-13

CUFX159-Virgin

0 521 85648 5

August 14, 2007

13.4 A Type of Escaping Motion

x

271 (a)

1

x

(b)

1 0.5

0.5

t

t 20

40

60

80

20

100

-0.5

-0.5

-1

-1

x

60

80

100

(d)

(c)

1

40

x

1

0.5

0.5

t 20

40

60

80

t 20

100

-0.5

-0.5

-1

-1

40

60

80

100

Figure 13.9. Some typical time series for parameter values spanning the escape boundary, = 0.9, ζ = 0.1: (a) F = 0.107, (b) F = 0.1075, (c) F = 0.108, (d) F = 0.1085.

Figure 13.9 shows four typical time series generated with the same (quiescent) initial conditions and forcing frequency except for a very slightly different forcing amplitude in each case. We see that whether the trajectory escapes or not is a sensitive function of the forcing parameters (at least in certain ranges of the parameter space). In part (a), the motion passes quite close to the potential hilltop (at x = 1) but does not lead to escape. There is a small possibility of escape after the time range of the simulation. An informative view of a trajectory can be found in the phase projection (the phase space is 3D) shown in Fig. 13.10. This picture corresponds to the time series in part (d) of Fig. 13.9. We see that these oscillations are far from sinusoidal (which would yield elliptical trajectories), and it is interesting to see that the trajectory even passes beyond x = 1 (briefly) before coming back

. x

1 0.5

Figure 13.10. A phase projection corresponding to the escaping trajectory when F = 0.1085.

-0.5

0.5 -0.5 -1

1

1.5

x

2

18:8

P1: KAE Chapter-13

CUFX159-Virgin

0 521 85648 5

August 14, 2007

272

Suddenly Applied Loads

and then finally escaping [15, 19–21]. A final feature worth mentioning here is that although Fig. 13.8 divides the parameter space into two regions, within the black areas it is not unreasonable to ask the question: How long does it take for a given trajectory to escape? It turns out that trajectories corresponding to relatively high forcing magnitudes (F ) tend to escape quickly (within a couple of forcing cycles or so), and trajectories corresponding to parameters very close to the boundary may take a relatively long time before finally escaping. This is also an aspect of nonlinear behavior that will be revisited in the final chapter.

13.5 Impulsive Loading In this section, we focus attention on what happens as the duration of a suddenly applied load approaches zero. In the limit, we deal with impulsive loading. For a SDOF system, it can easily be shown that this situation is equivalent to a system subjected to a nonzero initial velocity. The situation is a little more complicated for higher-order systems [22]. Here, we consider a two-DOF (2DOF) link model that is configured as an arch, such that the loading-deflection path has a bifurcation point before a limit point is reached, and the structure buckles into an asymmetric mode. Consider the system shown in Fig. 13.11. This system was analyzed extensively in [12, 22], and here we focus on impulsive loading and use a total energy approach following that reference. This system has two hinges of rotational stiffness β with concentrated masses m, and these are the locations at which two equal vertical loads P are applied. Deflection of the system is allowed by horizontal sliding at the righthand support, resisted by a linear spring of stiffness k. The initial rise of the arch is characterized by the angle α (initially equal at each end). The deflected state of the system is described in terms of the end angles θ and φ [22]. Other than the method of external forcing, this structure is similar to the one shown in Fig. 5.15. Assuming small angles, the total potential energy can be written as

P

P

m

L

β

m β

L L α

θ

φ k

Figure 13.11. A 2DOF link model [22].

18:8

P1: KAE Chapter-13

CUFX159-Virgin

0 521 85648 5

August 14, 2007

13.5 Impulsive Loading

UTP =

273

1 β(5θ + 5φ2 − 2α2 − 8φθ − 2θα − 2φα) 2 +

kL2 2 (α − θ2 − φ 2 + φθ)2 − PL(2α − θφ). 2

(13.16)

It is convenient to introduce new displacement variables ¯ r = (θ + φ)/2 β, ¯ s = (θ − φ)/2 β,

(13.17)

in which β¯ = β/(kL2 ). In terms of these new variables, we can write the total potential energy as √ √ 1 P U¯ T = (r2 + 9s2 − 2 r + ) + ( − r2 − 3s2 )2 − 2p( − r), 2

(13.18)

in which the following nondimensional parameters have also been used P U¯ T = UTP /(β¯ 2 kL2 ),

p = P/(kLβ¯ 3/2 )

¯ = α2 /β.

(13.19)

13.5.1 Equilibrium Behavior In the usual way, equilibrium is obtained from stationary values of the potential ¯ ¯ energy (∂U/∂r = 0 and ∂U/∂s = 0): 2(r −

√ ) − ( − r2 − 3s2 )2r + 2p = 0, 18s − ( − r2 − 3s2 )6s = 0.

(13.20)

The solutions to these equations give the symmetric response (s = 0) ( − 1 − r2 )r = p −

(13.21)

and the asymmetric response (s = 0) − 3 = r2 + 3s2 , 2r = p − .

(13.22)

The type of resulting behavior obviously depends on the magnitude of . It can be shown [12] that if > 4 then the system encounters an unstable point of bifurcation, and because this is a new feature we now focus on the specific case = 6. Figure 13.12(a) shows the potential-energy contours in terms of the two coordinates r and s. Five equilibrium points exist: Points 1 and 3 are stable equilibria, with an unstable point 2 in between; points 4 and 5 are unstable saddle points.

18:8

P1: KAE Chapter-13

CUFX159-Virgin

0 521 85648 5

August 14, 2007

274

Suddenly Applied Loads 1.5

(a)

s 1

4

0.5

3

1

0

2

–0.5

5

–1

–1.5 –2

–1

0

1

2

r (b)

10

p 7.5 5 2.5

r 1

2

Figure 13.12. (a) Contours of potential energy indicating five extremum points, = 6, (b) corresponding equilibrium curves [12].

13.5.2 Behavior under Sudden Loading If the system is initially located at point 1 in stable equilibrium and is then subject to a disturbance (impulse), it is apparent that a sufficiently large input may cause the system to traverse the unstable equilibrium (which is a local maximum of the potential energy). Furthermore, we can imagine the situation in which sufficient kinetic energy is imparted to the system such that either of the saddle points is traversed, typically leading to very large-amplitude oscillations. The equilibrium curves are projected in Fig. 13.12(b) as a function of p. Simitses [12] used an energy approach and the impulse-momentum theorem in which an impulse (PT0 ) associated with a load P over a short time duration (T0 ) is related to kinetic energy and the potential energy at the hilltop equilibrium to arrive

18:8

P1: KAE Chapter-13

CUFX159-Virgin

0 521 85648 5

August 14, 2007

13.6 Snap-Through of a Curved Panel

275

14

2( pτ 0)cr 12

barrier 1

10

8

6

4

barrier 2

2

4

6

8

10

Λ

12

Figure 13.13. Upper and lower bounds for the critical ideal impulse [22].

at analytical expressions for the two instability situations described in the previous ¯ paragraph. Using the nondimensional time τ = t (βk/2m), we find that the relevant expressions are 2(pτ0 )cr = 3 − 1,

2(pτ0 )cr = [5 − 2 − 3 2 − 4 + 0.25( + 2 + 2 − 4)2 ]1/2 ,

(13.23)

where τ0 is the nondimensional duration of the impulse. The expressions in Eqs. (13.23) are plotted in Fig. 13.13. The barriers represent levels of the dynamic forcing required to cause instability. Thus we see that when = 6, the critical impulse that causes dynamic buckling is given by 2(pτ0 )cr = 6.7. These results on discrete-link models can also be extended to continuous, suddenly loaded structures liable to snap-through [23]. For example, some early studies were conducted by Hsu (e.g., [24]) on shallow elastic arches under the action of various time-dependent lateral loads, including sinusoidal, arbitrary, concentrated, etc. The effects of initial thrust and elastic foundations were investigated [25] and interaction curves developed to assess the effect of various load combinations on the snap-buckling of shallow arches [23, 26–30]. Some interesting features associated with this type of problem are described in [31, 32] in which impulsive and harmonic loading may lead to counterintuitive behavior.

13.6 Snap-Through of a Curved Panel We considered the snap-through (saddle-node bifurcation) of a simple link model under the action of a quasi-statically increasing load in Section 5.6. For a continuous (shell-like) system under suddenly applied (step) loading, it is also possible to

18:8

P1: KAE Chapter-13

CUFX159-Virgin

0 521 85648 5

August 14, 2007

276

Suddenly Applied Loads 5.31

max/v/ (m) 0.3

(a) p

0.2 0.1

a

0.12

0.36

0.24

0.0937 p (kN/m2 )

t (s)

t=0

0.0

0.0

(b) 0.2954

p(t)

0.48

2.5

5.0

time, t (s)

−1 0 1

0.0 p = 2.5

5

7.5

. v (m/s)

5.29 5.2

0.08

2

0.0937

5.3 kN/m

5.31 kN/m 2

0.16

0.24

v (m) p=7

6

5.5 5.4 5.35 5.32

(c)

v (m)

Figure 13.14. Snap-through of a shallow panel subject to step loading: (a) the geometry and loading, (b) maximum response as a function of loading intensity, and (c) time series and phase projec¨ ¨ tion. Adapted from Kratzig and Nawrotzki [35] and Kratzig [36].

encounter snap-through [33]. However, for such a system the FE technique is the method of choice. ¨ An example of this behavior was given by Kratzig [34], in which a shallow curved panel is subject to a uniformly distributed vertical step load of magnitude p, as shown in Fig. 13.14(a). For this particular simulation, a square panel of width 10 m, thickness 0.1 m, and a radius of curvature of 100 m is considered. Young’s modulus is 3.4 × 107 kN/m2 and Poisson’s ratio is 0.2. With the rest position used as the initial condition, the results of numerical simulation shown in part (b) indicate that provided the magnitude of the step load is less than p = 5.31 kN/m2 the response is insufficient to traverse the potential-energy hilltop associated with the unstable equilibrium. Thus there is a considerable difference in the maximum response of the system depending on whether the system tends toward its inverted equilibrium or not. In both cases damped oscillations characterize the system response as energy is dissipated. A set of superimposed times series and phase projections is shown in Fig. 13.14(c) for various values of the applied load. The phase projection shows two trajectories, one on each side of the critical value. Unlike the simple link models of Chapter 5, this type of modeling requires special numerical procedures and is

18:8

P1: KAE Chapter-13

CUFX159-Virgin

0 521 85648 5

August 14, 2007

13.6 Snap-Through of a Curved Panel

277

the subject of current research [35, 36], including similar sensitivity in spherical caps [37]. Suddenly Loaded Column. We conclude this chapter by considering an axially

loaded column in which the end load is applied suddenly [1, 2, 38–41]. The following chapter will look at a pulsating end load in which case it is possible to get buckling for loads lower than the static buckling load. In this final section, we see that it is possible to have stability for loads far in excess of the static buckling load depending on the duration of the load. We focus attention on pulse loading of the type shown schematically (but with finite duration) in the inset of Fig. 13.1 [42, 43], and this is closely related to the impact loading of a bar, for example, a hammer hitting a nail [44]. Although the load is transmitted as an axial stress wave, it has been shown that typically buckling motion as the wave passes can be neglected, and the total length of the column is relatively unimportant; that is, the duration of the axial loading is relatively large compared with the period of longitudinal vibration of the bar. One consequence of this is that the buckling modes can be quite complicated (and associated with short buckling wavelengths), with divergent (hyperbolic) and bounded (trigonometric) modes both present and their separation depending on various characteristics of the system. A subtlety associated with this problem (other than the direct influence of dynamics of course) is that initial imperfections are necessary, and thus, rather than having a distinct bifurcational event, the loss of stability is most appropriately couched in terms of a dynamic growth, or amplification, of the initial geometric imperfection. We assume in the analysis that the behavior is elastic, although of course in practice very often plastic deformation is encountered [45]. We briefly discuss the simple case of a uniform, simply supported bar, subject to a suddenly applied load: ∂ 4w ∂2 ∂ 2w + P 2 (w + w0 ) + m 2 = 0. (13.24) 4 ∂x ∂x ∂t Note that in contrast to the initial imperfection encountered in Section 7.3, here, we measure the deflection w(x, t) that is due to axial loading from the initial imperfection w0 (x). In light of the earlier comments about the relative unimportance of the column length (see Lindberg and Florence [46]) we introduce a characteristic length of 1/k (where k 2 = P/EI) and then nondimensionalize Eq. (13.24) by using EI

x¯ = kx,

w¯ = w/r,

t¯ = trk 2 c,

where c = E/m is the speed of wave propagation and r = ration. The result is w¯ + w¯ + w¨¯ = −w¯ 0 .

(13.25) √

I/A is the radius of gy(13.26)

Application of the simply supported boundary conditions (w¯ = w¯ = 0) at (x¯ = 0 and x¯ = l = kL) leads to a solution of the form w( ¯ x, ¯ t¯) =

∞ n=1

gn (t¯) sin

nπx¯ , l

(13.27)

18:8

P1: KAE Chapter-13

CUFX159-Virgin

0 521 85648 5

August 14, 2007

278

Suddenly Applied Loads

and we assume a spatial distribution of initial imperfection according to w¯ 0 (x) ¯ =

∞

an sin

n=1

where 2 an = l

l

w¯ 0 (x) ¯ sin 0

nπx¯ , l

(13.28)

nπx¯ d¯x. l

(13.29)

Thus the solution for the amplitudes is obtained from g¨ n + η 2 (η 2 − 1)gn = η 2 an ,

(13.30)

where η = nπ/l is the wavenumber. We see the form of solution is quite different according to whether η is greater than, or less than, unity. This was a distinction in the form of the solution we first encountered at the start of Chapter 7, and assuming the bar is initially at rest (i.e., w¯ = w˙¯ when t = 0) we obtain the solution w( ¯ x, ¯ t¯) =

∞ n=1

an nπx¯ (cos p n t¯ − 1) sin 1 − η2 l

(13.31)

an nπx¯ (cosh p n t¯ − 1) sin 1 − η2 l

(13.32)

when η > 1, and w( ¯ x, ¯ t¯) =

∞ n=1

when η < 1, and where p n = η|1 − η 2 |1/2 .

(13.33)

In terms of amplification, or lateral growth of motion, it is convenient to scale Eqs. (13.31) and (13.32) according to the underlying static amplification of an imperfect simply supported, axially loaded bar, and thus gn (t¯) 1 nπx¯ cosh Gn (t¯) = p n t¯ − 1 sin , (13.34) = an 1 − η 2 cos l in which we take the cosine term for η > 1 and the hyperbolic cosine term for η < 1. Equation (13.34) is plotted in Fig. 13.15(a) for two values of the nondimensional time t¯. From this we can see that greatest amplification takes place in a narrow band of wavelengths (the preferred mode of buckling). We can then make use of the derivative to find the maximum value and then plot that maximum value versus time, as shown in Fig. 13.15(b). By use of an approximate analysis [46], it can be shown that under very high compression the bar will buckle into wavelengths close √ to 8.88r/ P/AE with very rapid growth in motion after approximately t¯ = 4. The effect of the velocity of impact loading was studied in Holzer and Eubanks [47] and Hayashi and Sano [48, 49]. We can also approach this type of problem by considering the dropping of a weight onto the end of a long strut, which would be an easy scenario to set up in

18:8

P1: KAE Chapter-13

CUFX159-Virgin

0 521 85648 5

August 14, 2007

References

279 200

G

14

(b)

(a)

12

150

t=5

10

Gmax

8

100

6 4

50

t=3

2 0.5

1

1.5

0

2

t Figure 13.15. Amplification vs. wavenumber, (b) corresponding maximum amplification as a function of time [46].

the laboratory. In this case, it can be shown that the buckling wavelength scales with the inverse square root of the impact speed [50]. In a practical sense, the distribution of initial geometric imperfections might have a random element, and this has also been treated in the literature [51]. Extensive studies using similar approaches have focused on rapid application of axial loading on cylinders and shells [51–53]. Finally, the case of a dynamic application of the end load that is relatively slow is mentioned. In this type of ramp function, which is somewhat representative of what would happen in a loading machine, there is also some interesting behavior [55].

References [1]

J.H. Meier. On the dynamics of elastic buckling. Journal of the Aeronautical Sciences, 12:433–40, 1945. [2] J.F. Davidson. Buckling of struts under dynamic loading. Journal of the Mechanics and Physics of Solids, 2:54–66, 1953. [3] G. Herrmann. Dynamic Stability of Structures. Pergamon, 1967. [4] G.J. Simitses, A.N. Kounadis, and J. Giri. Dynamic buckling of simple frames under a step load. Journal of Engineering Mechanics, 105:896–900, 1979. [5] A.T. Brewer and L.A. Godoy. On interaction between static and dynamic loads in instability of symmetric or asymmetric structural systems. Journal of Sound and Vibration, 147:105–14, 1991. [6] I. Cochin and H.P. Plass. Analysis and Design of Dynamic Systems. Harper & Row, 1990. [7] B. Budiansky. Dynamic buckling of elastic structures: Criteria and estimates. In G. Herrmann, editor, Dynamic Stability of Structures. Pergamon, 1967. [8] G.J. Simitses. Effect of static preloading on the dynamic stability of structures. AIAA Journal, 21:1174–80, 1983. [9] J.M.T. Thompson. Dynamic buckling under step loading. In G. Herrmann, editor, Dynamic Stability of Structures. Pergamon, 1967. [10] J. Roorda. Stability of structures with small imperfections. Journal of the Engineering Mechanics Division, ASCE, 91:87, 1965.

18:8

P1: KAE Chapter-13

CUFX159-Virgin

280

0 521 85648 5

August 14, 2007

Suddenly Applied Loads [11] J.W. Hutchinson and B. Budiansky. Dynamic buckling estimates. AIAA Journal, 4:525– 30, 1966. [12] G.J. Simitses. Instability of dynamically-loaded structures. Applied Mechanics Reviews, 40:1403–8, 1987. [13] G.A. Zizicas. Dynamic buckling of thin elastic plates. Transactions of the ASME, 74:1257–68, 1952. [14] P.S. Bulson. The Stability of Flat Plates. Chatto and Windus, 1970. [15] B. Budiansky and E.S. Roth. Axisymmetric dynamic buckling of clamped shallow spherical shells. Technical Report, NASA TND-1510, 1962. [16] M.W. Hilburger, A.M. Waas, and J.H. Starnes. Modeling the dynamic response and establishing post buckling snap-through equilibrium of discrete structures via a transient analysis. Journal of Applied Mechanics, 64:590–5, 1997. [17] D.J. Inman. Engineering Vibration. Prentice-Hall, 2000. [18] L.N. Virgin. Introduction to Experimental Nonlinear Dynamics: A Case Study in Mechanical Vibration. Cambridge University Press, 2000. [19] J.M.T. Thompson. Chaotic phenomena triggering escape from a potential well. Proceedings of the Royal Society of London, A421:195–225, 1989. [20] L.N. Virgin, R.H. Plaut, and C.-C. Cheng. Prediction of escape from a potential well under harmonic excitation. International Journal of Non-Linear Mechanics, 27:357–65, 1992. [21] J.A. Gottwald, L.N. Virgin, and E.H. Dowell. Routes to escape from an energy well. Journal of Sound and Vibration, 187:133–44, 1995. [22] G.J. Simitses. Dynamic Stability of Suddenly Loaded Structures. Springer-Verlag, 1989. [23] D.L.C. Lo and E.F. Masur. Dynamic buckling of shallow arches. Journal of the Engineering Mechanics Division, ASCE, 102:901–17, 1976. [24] C.S. Hsu. Stability of shallow arches against snap-through under timewise step loads. Journal of Applied Mechanics, 35:31–9, 1968. [25] R.H. Plaut and E.R. Johnson. The effect of initial thrust and elastic foundation on the vibration frequencies of a shallow arch. Journal of Sound and Vibration, 78:565–71, 1981. [26] N.J. Hoff and V.G. Bruce. Dynamic analysis of the buckling of laterally loaded flat arches. Journal of Mathematical Physics, 32:276–88, 1954. [27] A.M. Liapunov. Stability of Motion (Collected Papers). Academic, 1966. [28] R.E. Fulton and F.W. Barton. Dynamic buckling of shallow arches. Journal of Engineering Mechanics, 97:865–77, 1971. [29] K.-Y. Huang and R.H. Plaut. Snap-through of a shallow arch under pulsating load. In F.H. Schroeder, editor, Stability in the Mechanics of Continua. Springer, 1982, pp. 215– 33. [30] M.T. Donaldson and R.H. Plaut. Dynamic stability boundaries for a sinusoidal shallow arch under pulse loads. AIAA Journal, 21:469–71, 1983. [31] E.R. Johnson and I.K. McIvor. The effect of spatial distribution on dynamic snapthrough. Journal of Applied Mechanics, 45:612–18, 1978. [32] R.H. Plaut and J.-C. Hsieh. Oscillations and instability of a shallow arch under twofrequency excitation. Journal of Sound and Vibration, 102:189–201, 1985. [33] J.S. Humphreys and S.R. Bodner. Dynamic buckling of shallow shells under impulsive loading. Journal of Engineering Mechanics, 88:17–36, 1962. ¨ ¨ [34] W.B. Kratzig. Nonlinear responses. In A.N. Kounadis and W.B. Kratzig, editors, Nonlinear Stability of Structures (Theory and Computational Techniques). Springer-Verlag, 1995.

18:8

P1: KAE Chapter-13

CUFX159-Virgin

0 521 85648 5

August 14, 2007

References ¨ [35] W.B. Kratzig and P. Nawrotzki. Computational concepts for kinetic instability prob¨ lems. In A.N. Kounadis and W.B. Kratzig, editors, Nonlinear Stability of Structures (Theory and Computational Techniques). Springer-Verlag, 1995. ¨ [36] W.B. Kratzig. Eine einheitliche statische und dynamische Stabilitatstheorie fur Pfad¨ Angeverfolgungsalgorithmen in der numerischen Festkorpermechanik. Zeitschrift fur wandte Mathematik und Mechanik, 69:203–13, 1989. [37] D. Dinkler and J. Pontow. A model to evaluate dynamic stability of imperfection sensitive shells. Computational Mechanics, 37:523–9, 2006. [38] C. Koning and J. Taub. Impact buckling of thin bars in the elastic range hinged at both ends. Luftfahrforschung, 10:55–64, 1933. [39] G. Gerard and H. Becker. Column behavior under conditions of impact. Journal of the Aeronautical Sciences, 19:58–60, 1952. [40] E. Sevin. On the elastic bending of columns due to dynamic axial forces including effects of axial inertia. Journal of Applied Mechanics, 27:125–31, 1960. [41] R. Grybos. Impact stability of a bar. International Journal of Engineering Science, 13:463–77, 1975. [42] N.J. Huffington. Response of elastic columns to axial pulse loading. AIAA Journal, 1:2099–2104, 1963. [43] I.K. McIvor and J.E. Bernard. The dynamic response of columns under short duration axial loads. Journal of Applied Mechanics, 40:688–92, 1973. [44] J.M. Housner and N.F. Knight. The dynamic collapse of a column impacting a rigid surface. AIAA Journal, 21:1187–95, 1983. [45] W. Abramowicz and N. Jones. Dynamic progressive buckling of circular and square tubes. International Journal of Impact Engineering, 4:247–70, 1986. [46] H.E. Lindberg and A.L. Florence. Dynamic Pulse Buckling: Theory and Experiment. Nijhoff, 1987. [47] S.M. Holzer and R.A. Eubanks. Stability of columns subjected to impulsive loading. Journal of Engineering Mechanics, 95:897–920, 1969. [48] T. Hayashi and Y. Sano. Dynamic buckling of elastic bars (the case of low velocity impact). Bulletin of the Japanese Society of Mechanical Engineering, 15:1167–75, 1972. [49] T. Hayashi and Y. Sano. Dynamic buckling of elastic bars (the case of high velocity impact). Bulletin of the Japanese Society of Mechanical Engineering, 15:1176–84, 1972. [50] J.R. Gladden, N.Z. Handzy, A. Belmonte, and E. Villermaux. Dynamic buckling and fragmentation in brittle rods. Physical Review Letters, 94(035503), 2005. [51] I. Elishakoff. Axial impact buckling of a column with random initial imperfections. Journal of Applied Mechanics, 45:361–5, 1978. [52] H.E. Lindberg and R.E. Herbert. Dynamic buckling of a thin cylindrical shell under axial impact. Journal of Applied Mechanics, 33:105–12, 1966. [53] N. Jones and C.S. Ahn. Dynamic elastic and plastic buckling of complete spherical shells. International Journal of Solids and Structures, 10:1357–74, 1974. [54] R. Kao. Nonlinear dynamic buckling of spherical caps with initial imperfections. Computers and Structures, 12:49–63, 1980. [55] N.J. Hoff. The dynamics of the buckling of elastic columns. Journal of Applied Mechanics, 17:68–74, 1953.

281

18:8

P1: KAE Chapter-14

CUFX159-Virgin

0 521 85648 5

August 14, 2007

14 Harmonic Loading: Parametric Excitation

14.1 An Oscillating End Load In Chapter 7 we considered the free vibrations of a simply supported beam subject to an axial load of constant magnitude, and then in Chapter 13 the load was applied suddenly. Now suppose the end load is pulsating harmonically, that is, we replace P in Fig. 7.1 with P + S cos t. Because the forcing will appear in the stiffness term we are dealing with parametric excitation, and this is sometimes referred to as vibration buckling in the literature [1]. The governing equation of motion is thus m

∂2 w ∂4 w ∂2 w + EI + (P + S cos t) = 0. ∂t2 ∂x4 ∂x2

(14.1)

With simply supported boundary conditions we take the solution in the form of a single half-sine wave, πx w = f (t) sin . (14.2) L Substituting this back into Eq. (14.1), reintroducing the term from Eq. (7.20), that is, PL2 EIπ4 ω2 = 1 − , (14.3) mL4 EIπ2 again normalizing the axial load and natural frequency by the Euler load (Pcr = π2 EI/L2 ) and frequency without load [ω20 = π4 EI/(mL4 )], respectively, and rescaling time according to τ = t, we arrive at f (τ) + (α + cos τ)f (τ) = 0,

(14.4)

where α = ω20 2 (1 − p),

= − ω20 2 s.

(14.5)

Although Eq. (14.4), which is called Mathieu’s equation, is linear, it is by no means easy to solve, and the character of the solutions (specifically their stability) depends in a nonsimple way on the parameters α and . This equation is also encountered in the dynamics of a pendulum with a harmonically shaken pivot [2], and it is related to the stability of forced oscillations which will be studied in more detail in the final chapter. Sinha [3] also considers the problem of pulsating axial load but applied to Timoshenko beams. To consider the stability of this type of system (which can be 282

18:9

P1: KAE Chapter-14

CUFX159-Virgin

0 521 85648 5

August 14, 2007

14.2 The Variational Equation

283

viewed as a natural extension of the stability of equilibria introduced in Chapter 3), we again make use of linearization before proceeding to Floquet theory [4–11].

14.2 The Variational Equation In this chapter, we consider nonautonomous dynamical systems of the form x˙ = f(x, t),

(14.6)

and proceed to consider the stability of solutions based on the behavior of small perturbations. In Chapter 3 we also considered the behavior of transients in the vicinity of equilibria (point attractors) in autonomous systems, whereas now we focus on transients in the vicinity of periodic solutions in nonautonomous (especially periodically forced) systems. Again linearization is a key concept, and use will be made of Poincare´ sampling and fixed points of maps. Equation (14.6) is in general a nonlinear system. Following the development in Subsection 4.2.3, we obtain the linear variational equation η˙ = DF(t)η.

(14.7)

It is this equation that governs the stability of solutions in the vicinity of special solutions (which will be steady-state oscillations in this context). Equations with Periodic Coefficients. Equation (14.4) is a specific case of the wider

class of forced vibration problems, and is clearly related to Eq. (14.7) in which the coefficients of the Jacobian DF(t) are periodic. Consider Hill’s equation [12], x¨ + G(t)x = 0,

(14.8)

where G(t + T) = G(t). Although this is a rather benign-looking (linear) equation, it does not, in general, submit to closed-form analytical solutions, and hence a variety of approximate techniques have been developed. We also note that a simple transformation allows the related case of damped parametric oscillations to be brought into the standard form of Eq. (14.8). Hence this equation represents quite a wide class of problems and has received considerable scrutiny over the years [12–15]. Placing Eq. (14.7) in a little more general (state-variable) context, and simplifying the notation by using A ≡ DF and x ≡ η, we consider the solutions of x˙ = A(t)x,

t ∈ R,

(14.9)

where x is an n-dimensional state vector and A(t) is a continuous T-periodic n × n matrix, that is, A(t + T ) = A(t). This system has n linearly independent fundamental solutions φi , where i = 1, 2, . . . , n, which can be expressed as a fundamental solution matrix (t) [15]. Shifting in time by T, we see that (t + T ) is also a fundamental matrix solution, and because they are linearly independent there is a nonsingular n × n matrix C such that

(t + T ) = (t)C.

(14.10)

18:9

P1: KAE Chapter-14

CUFX159-Virgin

0 521 85648 5

284

August 14, 2007

Harmonic Loading: Parametric Excitation

This constant matrix C (which depends on T but not on t) is called the monodromy matrix of Eq. (14.9) and contains the key stability information. The eigenvalues ρ of C are called characteristic (or Floquet) multipliers (CMs). They are uniquely determined and govern local divergence or convergence about a periodic orbit [16]. The matrix C can also be expressed as C = eBT ,

(14.11)

where B is a constant matrix. The Floquet theorem then states that the fundamental matrix (t) of Eq. (14.9) can be written as

(t) = P(t)eBt ,

(14.12)

where P(t) is T periodic and B is a constant n × n matrix. The eigenvalues γ of B are the characteristic (or Floquet) exponents (CEs) and are essentially the same as encountered earlier for equilibria, and they govern the stability of the trivial solution of Eq. (14.9). They are unique only to within an integer multiple of 2πi/T. It is helpful to consider the monodromy matrix C as a Floquet operator, which maps (t) onto (t + T ), and taking the initial condition as the identity vector

(0) = I, we then have from Eq. (14.10), C = (T ). Furthermore, if the eigenvalues ρi , i = 1, 2, . . . , n, of C (the CMs) are distinct, then Eq. (14.9) has n linearly independent normal solutions of the form xi = pi (t)eγi t ,

(14.13)

where the pi (t) are periodic functions with period T. Thus we see the fundamental relationship between the CEs and the CMs: ρ = eγt .

(14.14)

Now, again with distinct eigenvalues we can diagonalize C; that is, we define a transformed system

(t + T ) = (t)J,

(14.15)

where = M−1 CM, M is a nonsingular n × n constant matrix (chosen to simplify J ), and φ(t) = P−1 . In this case, we can again consider the relation between the CEs and the CMs but in component form, ψi (t + T ) = ρi ψ(t),

(14.16)

ψi (t) = eγi t φi (t).

(14.17)

with the Mathieu functions

The stability of the periodic solutions is now emerging. Equation (14.16) can be extended to ψi (t + NT ) = ρN i ψ(t),

(14.18)

where N is an integer. Therefore we see that it is the magnitude of ρ that determines stability:

18:9

P1: KAE Chapter-14

CUFX159-Virgin

0 521 85648 5

August 14, 2007

14.2 The Variational Equation

285

Loss of stability of a cycle Saddle node

Period doubling (Flip)

I

Neimark

I

I

R

R

R

Characteristic multipliers

Figure 14.1. The generic routes to instability (in terms of |ρi |) for a system under the action of a single control parameter.

r For stability of a periodic orbit, we have ψi (t) → 0 as t (and hence N) → ∞ if |ρi | < 1 (i.e., the real part of γi is negative). r For instability, we have ψi (t) → ∞ as t (and hence N) → ∞ if |ρi | > 1 (i.e., the real part of γi is positive). In practice, the most difficult part of any analysis of this kind is determining the matrix (t). We shall see that this can be achieved numerically by means of the Poincare´ map or by using various approximate analytical schemes. These will be dealt with in more detail in later chapters but we will see that the three typical ways in which multipliers leave the unit circle are shown in Fig. 14.1 [15, 17], with R and I signifying real and imaginary, respectively. We note that the Neimark bifurcation is less commonly encountered in the types of structural system encountered in this book. We can gain some stability insight by considering certain constraints, akin to the Routh–Hurwitz criterion for the stability of equilibria, placed on the eigenvalues. To do this, use is made of the Wronskian determinant of the fundamental matrix corresponding to Eq. (14.9): t Det (t) = exp Tr A(s) ds , (14.19) 0

where Tr A(s) is the trace of A(s). The Floquet theorem [Eq. (14.12)] tells us that Det (t) = Det[P(t)eBt ], which leads to

Det(eBT ) = exp

T

(14.20)

Tr A(t) dt .

(14.21)

0

Now we are in a position to state that ρ1 ρ2 . . . ρn = exp

T

Tr A(t) dt 0

(14.22)

18:9

P1: KAE Chapter-14

CUFX159-Virgin

0 521 85648 5

August 14, 2007

286

Harmonic Loading: Parametric Excitation

and n

γi =

i=1

1 T

T 0

2πi . Tr A(t) dt mod T

(14.23)

The procedure is quite straightforward for obtaining stability information provided the fundamental matrix of normal solutions, , is known. This is seldom the case, and a variety of approximate analytical and numerical techniques have been developed to account for this. Fortunately, there are a number of shortcuts that can be taken to determine stability without needing the full solution.

14.3 Mathieu’s Equation Using the information concerning the stability of periodic motion, we now return to Mathieu’s equation [Eq. (14.4)], where f is replaced with x: x (τ) + (α + cos τ)x(τ) = 0.

(14.24)

This can be expressed in state matrix terms (with x˙ ≡ y, replacing the primes):

x˙ 0 1 x = . (14.25) y˙ −α − cos t 0 y We see that the trace of the matrix in the preceding equation is zero, and using Eq. (14.22) applied to Eq. (14.25) we have ρ1 ρ 2 = e0 = 1, and thus the roots satisfy the quadratic equation ρ2 − φ(α, )ρ + 1 = 0.

(14.26)

The solutions are given by

1 φ(α, ) ± φ(α, )2 − 4 . (14.27) 2 In general, we might be more interested in determining whether the motion is bounded or not, rather than in being able to write the specific form of the solution, and hence the transition curves between stable and unstable behavior are of paramount importance, and these occur when φ(α, ) = ±2. For Mathieu’s equation, these transition curves correspond to the specific combinations of α and for which periodic solutions, with period 2π or 4π, occur [12]. The standard analytical approach to obtaining the transition curves involves a Hill determinant [12, 17]. However, a useful approximate technique based on the perturbation method is introduced that will also be useful when we consider largeamplitude vibration in the final chapter [2, 15]. ρ1 , ρ2 =

A Perturbation Solution. For relatively small values of the parameter , the transition curves can be computed with a perturbation method [15]. The solutions to Mathieu’s equation are assumed to be of the form

x(t) = x0 (t) + x1 (t) + · · · +,

(14.28)

18:9

P1: KAE Chapter-14

CUFX159-Virgin

0 521 85648 5

August 14, 2007

14.3 Mathieu’s Equation

287

in which x0 , x1 , . . . , have period 2π or 4π, and the transition curves are given by α = α() = α0 + α1 + · · · + .

(14.29)

On substituting these expressions into Mathieu’s equation, we obtain (x¨ 0 + α0 x0 ) + (x¨ 1 + α1 x0 + x0 cos t + α0 x1 ) + 2 (x¨ 2 + α2 x0 + α1 x1 + x1 cos t + α0 x2 ) + · · · + = 0,

(14.30)

and setting the coefficients of each power of equal to zero we obtain x¨ 0 + α0 x0 = 0,

(14.31)

x¨ 1 + α0 x1 = −(α1 + cos t)x0 ,

(14.32)

x¨ 2 + α0 x2 = −x0 α2 − (α1 + cos t)x1 , . . . ,

(14.33)

and so on. If we consider the solutions of Eq. (14.31) first, we see that we have simple harmonic motion of period 2π or 4π if α0 = (1/4)n2 with n = 0, 1, . . . . With n = 0, we have α0 = 0, x0 = 1 (assuming a unit displacement as the initial condition), and Eq. (14.32) becomes x¨ 1 = −α1 − cos t,

(14.34)

and for periodic solutions, we require α1 = 0 and thus, x1 (t) = cos t + c,

(14.35)

where c is a constant. Equation (14.33) then becomes x¨ 2 = −α2 − 1/2 − c cos t − 1/2 cos 2t,

(14.36)

and again, for periodic solutions we require α2 = −1/2, and thus, up to terms of second order in we have α = −1/22 + · · · + .

(14.37)

Repeating the analysis for n = 1, we arrive at α = 1/4 ± (1/2) − (1/8)2 + · · · +,

(14.38)

and for n = 2, α = 1 + (5/12)2 + · · · +,

(14.39)

α = 1 − (1/12)2 + · · · + .

(14.40)

These transition curves are plotted in Fig. 14.2, with the shaded regions indicating regions of unbounded growth of motion. A couple of numerical simulations show the form of the stable and unstable motion. The dashed curves within the unstable zones indicate the transition curves when a small amount of damping is added. This diagram is symmetric about the α axis but is plotted only for positive here. We are now in a position to interpret the dynamic response of the beam with a pulsating end load [18, 19]. The relation between α and and the forcing characteristics of the end load lead to the plot shown in Fig. 14.3. Here, we have focused on

18:9

P1: KAE Chapter-14

CUFX159-Virgin

0 521 85648 5

August 14, 2007

288

Harmonic Loading: Parametric Excitation

0.8 x

x

0.6 t

t 0.4 0.2

–0.5

0.5

0.0

1.0

Figure 14.2. The response of Mathieu’s equation in terms of the parameters α and , and based on a perturbation solution.

the instability arising from α = 0.25. When the static component of the axial load is zero, we see that an instability occurs if the forcing frequency of the oscillating part of the load is close to twice the natural frequency of the system. With either additional static compression or tension the parametric instability shifts according to the fundamental frequency of the beam with an axial load—a situation described at length in earlier parts of this book. Again, damping tends to have a stabilizing effect, such that even when the forcing frequency is exactly twice the natural frequency, there needs to be a certain amount of forcing magnitude to cause instability, and this tends to make the higher-order zones of instability practically disappear. We also note that this behavior is related to the issue of quasi-perodicity and Arnold tongues, found for example in the sine map [20–22].

14.4 Pulsating Axial Loads on Shells In much the same way that an oscillating axial load produces some interesting dynamic behavior in a column, a similar effect occurs in plates, panels, and shells [23, 24].

s 0.6

p = 0.5

p = −0.5

p=0

0.4

Figure 14.3. The stability of a prismatic beam with end load in which a portion of the load is oscillating. Principal parametric resonance.

0.2

0

1

2

3

4

5

7

6 2

18:9

P1: KAE Chapter-14

CUFX159-Virgin

0 521 85648 5

August 14, 2007

14.4 Pulsating Axial Loads on Shells

289

(b)

p(t)

(a)

0.6

unstable unstable

d cr

a

s

0.3

a R

0.0 0.5

1.0 1

Figure 14.4. Regions of instability computed for a cylindrical panel with a periodically varying axial load. Adapted from [25].

14.4.1 A Curved Panel We now consider a shallow panel that has some unidirectional curvature in a direction perpendicular to the loading. The analysis of such a system becomes increasingly complicated and recourse to numerical (FE) techniques is often used. ¨ Figure 14.4(a) shows an example taken from Kratzig and Nawrotzki [25] in which a shallow cylindrical panel segment is subject to a harmonically oscillating axial load. The geometry of the shell is defined by a = 10 m, R = 83.33 m, thickness = 0.1 m, and material properties for mild steel were used in the authors’ time integration. They produced both stable and unstable time series, depending on the parameter values used, specifically the constant (static) load (λS ), the forcing amplitude (λD), and the forcing frequency (), with the results normalized with respect to the elastic critical buckling load (λcr ) and linear natural frequency (ω1 ), as shown in Fig. 14.4(b). The ¨ dashed curve within each instability zone indicates the effect of damping. Kratzig and Nawrotzki [25] also use a similar FE technique to assess the parametric instability of a truncated conical shell and compute the magnitude of the Floquet multipliers.

14.4.2 A Cylindrical Shell When the shell is a complete cylinder, rather than a segment, we may still get parametric instability. This subsection describes the behavior reported in Popov et al. [26], based on an approximate analytical treatment. Given a cylindrical shell of the type shown in Fig. 10.14, we suppose that instead of a fixed axial load of magnitude Nx we now have a pulsating axial load of the form p(t) = p 1 cos ωt. We then make a single-mode Galerkin analysis (based on the Donnell shell theory) for the shell with the following properties: R/h = 100, L/R = 2, and a little damping added. With these parameters, it is appropriate to take an assumed form for the solution of w(x, y, t) = f 1 (t)h cos

πx 2πx 5y cos + f 2 (t)h cos . L R L

(14.41)

18:9

P1: KAE Chapter-14

CUFX159-Virgin

0 521 85648 5

August 14, 2007

290 (a)

Harmonic Loading: Parametric Excitation (b)

L

Figure 14.5. (a) Axisymmetric, or concertina mode of vibration, (b) asymmetric, or checkerboard, mode of vibration. Reproduced with permission from [27].

The two modes are shown in Fig. 14.5. Popov et al. [28] show that the dynamic response in mode 1 (for example) is related to Mathieu’s equation in the form d2 f 1 ω + 1 − 2µ cos τ f 1 = 0. (14.42) dτ2 ω1 Furthermore, the control parameter µ = p 1 /(2p c ) is introduced, in which p c is the linear-elastic (static) buckling load. Popov et al. [28] use a continuation technique to track the loss of stability of the trivial solution, that is, the transition from purely extensional to bending behavior. In the first part of this chapter we considered the solutions of Mathieu’s equation in terms of whether the motion grew with time (or not). In a comprehensive analysis, it is possible to more fully characterize the instability phenomena, as shown in Fig. 14.6(a). Here the regions of principal parametric (ω/ω1 ≈ 2) and fundamental resonance (ω/ω1 ≈ 1) can again be observed, with the transition curves labeled Sp1 and so on. These bifurcations indicate the nature of the instability, with the superscript 1 corresponding to a flip bifurcation and the superscript 2 indicating a pitchfork bifurcation. The former leads to the buildup of motion at twice the period of the external excitation, with the latter leading to motion at the same period as the forcing. Furthermore, we see the appearance of a couple of additional transition curves (B) that correspond to saddle-node bifurcations. All these transitions occur when a Floquet multiplier is equal to one in magnitude. For example, the flip bifurcation is characterized by a Floquet multiplier = −1 (see Fig. 14.1), which will be identified with the onset of period doubling in the final chapter. This diagram was based on a single-mode solution [i.e., with f 2 (t) = 0]. Also shown in Fig. 14.6 are the bifurcation diagrams at a number of frequency ratios [and indicated by the vertical dashed lines in part (a)]. Starting at ω/ω1 = 0.9 we see that as µ is increased the trivial solution loses stability (the dashed lines in these plots indicate unstable paths) and oscillations occur, which gradually grow as the unstable region is further penetrated. The subscript p indicates that this pitchfork bifurcation is supercritical (see Subsection 3.4.2). For ω/ω1 = 1.1 the bifurcation is now subcritical and leads to a sudden jump in response. An interesting example is found when ω/ω1 = 1.27 in which the system initially loses stability by means of a supercritical flip but then restabilizes and a subcritical pitchfork is encountered. Other modes can be included in the analysis [e.g., the f 2 (t) term in Eq. (14.41)] and although they have no effect on the initial loss of stability of the trivial state, they can

18:9

P1: KAE Chapter-14

CUFX159-Virgin

0 521 85648 5

August 14, 2007

14.4 Pulsating Axial Loads on Shells

291

0.9 1.1 1.27

1.9 2.1

1.5

1.0

2

2

Sp

Sb

1

Sb 1

0.5

B

Sp

2

B 0

0.5

1.0

2.0

1.5

1

2.5

3.0 1

2.5

f1

0.5

2.5

0.9

T

1.1

f1

B

2

2

S

Sp1

2 b

0.0

0.0

1.5

2.5

2.5

f1

S2 b

−0.5

−2.5 1.5

0.0

T 2T

Sp −2.5

1.27

f1

T

1.9

2.1

f1

2T

2.0

2T

B1 1

1

Sb

Sp −2.5

−2.5 0.0

1.5

0.0

1.5

Figure 14.6. Regions of instability computed for a circular cylinder with a periodically varying axial load. Adapted from Popov et al. [28].

play an important role in subsequent postcritical behavior (for example the B curves tend to emanate from the bottom of the transition curves toward lower frequency ratios), and significant modal interactions can occur [28, 29]. Popov et al. [28] also consider a shallow panel by using basically the same technique, with the major difference indicating loss of stability by means of a transcritical bifurcation. Also, the proportion of static load to the magnitude of the pulsating part is an interesting parameter that relates back to the underlying buckling behavior. The behavior of these types of system may be especially complicated when parametric and direct external excitation take place simultaneously [30, 31], and combination resonances can also occur [32, 33]. At the other end of the loading regime, we have creep buckling. A comprehensive account of this phenomenon can be found in Bazant and Cedolin [18]. This chapter finishes with a mention of Meissner’s problem. This is similar to the Mathieu equation but in this case the periodic

18:9

P1: KAE Chapter-14

CUFX159-Virgin

0 521 85648 5

292

August 14, 2007

Harmonic Loading: Parametric Excitation

variation in the stiffness is piecewise constant (a square wave) and is amenable to analytic treatment [34, 35]. References [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20] [21]

[22]

[23]

[24]

J. Singer, J. Arbocz, and T. Weller. Buckling Experiments, Vol. 2. Wiley, 2002. A.H. Nayfeh and D.T. Mook. Nonlinear Oscillations. Wiley, 1979. S.K. Sinha. Dynamic stability of a timoshenko beam subjected to an oscillating axial force. Journal of Sound and Vibration, 131:509–14, 1989. R.M. Evan-Iwanowski. On the parametric response of structures. Applied Mechanics Reviews, 18:699–702, 1965. D. Krajcinovic and G. Herrmann. Parametric resonance of straight bars subjected to repeated impulsive compression. AIAA Journal, 6:2025–7, 1968. A.D.S. Barr and G.T.S. Done. Parametric oscillations in aircraft structures. The Aeronautical Journal, 75:654–8, 1971. C.E. Hammond. An application of Floquet theory to prediction of mechanical instability. Journal of the American Helicopter Society, 4:14–23, 1974. R.A. Ibrahim and A.D.S. Barr. Parametric resonance, part I: Mechanics of linear problems. Shock and Vibration Digest, 10(1):15–29, 1978. R.A. Ibrahim and A.D.S. Barr. Parametric resonance, part II: Mechanics of nonlinear problems. Shock and Vibration Digest, 10(2):9–24, 1978. G.J. Simitses. Dynamic Stability of Suddenly Loaded Structures. Springer-Verlag, 1989. J.P. Cusumano. Low-Dimensional, Chaotic, Nonplanar Motions of the Elastica: Experiment and Theory. Ph.D. dissertation, Cornell University, 1990. C. Hayashi. Nonlinear Oscillations in Physical Systems. Princeton University Press, 1964. V.V. Bolotin. The Dynamic Stability of Elastic Systems. Holden-Day, 1964. M.A. Souza. Vibration of thin-walled structures with asymmetric post-buckling characteristics. Thin-Walled Structures, 14:45–57, 1992. D.W. Jordan and P. Smith. Nonlinear Ordinary Differential Equations. Oxford University Press, 1999. J. Guckenheimer and P.J. Holmes. Nonlinear Oscillations, Dynamical Systems, and Bifurcations of Vector Fields. Springer-Verlag, 1983. N.W. McLachlan. Theory and Applications of Mathieu Functions. Dover, 1964. Z.P. Bazant and L. Cedolin. Stability of Structures. Oxford University Press, 1991. J.F. Doyle. Nonlinear Analysis of Thin-Walled Structures. Springer, 2001. E. Ott. Chaos in Dynamical Systems. Cambridge University Press, 1993. P.R. Everall and G.W. Hunt. Arnold tongue predictions of secondary buckling in thin elastic plates. Journal of the Mechanics and Physics of Solids, 47:2187–2206, 1999. P.R. Everall and G.W. Hunt. Quasi-periodic buckling of an elastic structure under the influence of changing boundary conditions. Proceedings of the Royal Society of London A, 455:3041–51, 1999. K.K.V. Devarakonda and C.W. Bert. Flexural vibration of rectangular plates subjected to sinusoidal distributed compressive loading on two opposite sides. Journal of Sound and Vibration, 283:749–63, 2005. ¨ Y. Basar, C. Eller, and W.B. Kratzig. Finite element procedures for parametric phenomena of arbitrary elastic shell structures. Computational Mechanics, 2:89–98, 1987.

18:9

P1: KAE Chapter-14

CUFX159-Virgin

0 521 85648 5

August 14, 2007

References ¨ [25] W.B. Kratzig and P. Nawrotzki. Computational concepts for kinetic instability prob¨ lems. In A.N. Kounadis and W.B. Kratzig, editors, Nonlinear Stability of Structures (Theory and Computational Techniques). Springer-Verlag, 1995. [26] A.A. Popov, J.M.T. Thompson, and J.G.A. Croll. Bifurcation analyses in the parametrically excited vibrations of cylindrical panels. Nonlinear Dynamics, 17:205–25, 1998. [27] A.A. Popov. Parametric resonance in cylindrical shells: A case study in the nonlinear vibration of structural shells. Engineering Structures, 25:789–99, 2003. [28] A.A. Popov, J.M.T. Thompson, and F.A. McRobie. Low dimensional models of shell vibrations: Parametrically excited vibrations of cylindrical shells. Journal of Sound and Vibration, 209:163–86, 1998. [29] F.A. McRobie, A.A. Popov, and J.M.T. Thompson. Auto-parametric resonance in cylindrical shells using geometric averaging. Journal of Sound and Vibration, 227:65–84, 1999. [30] C.S. Hsu. Impulsive parametric excitation: Theory. Journal of Applied Mechanics, 39:551–8, 1972. [31] N. HaQuang, D.T. Mook, and R.H. Plaut. A non-linear analysis of the interactions between parametric and external excitations. Journal of Sound and Vibration, 118:425–39, 1987. [32] T. Iwatsubo, Y. Sugiyama, and S. Ogino. Simple and combination resonances of columns under periodic axial loads. Journal of Sound and Vibration, 33:211–21, 1974. [33] R.H. Plaut, N. HaQuang, and D.T. Mook. Simultaneous resonances in non-linear structural vibrations under two-frequency excitation. Journal of Sound and Vibration, 106:361–76, 1986. [34] A.P. Seyranian and A.A. Mailybaev. Multiparameter Stability Theory with Mechanical Applications. World Scientific, 2003. [35] C.-H. Xei. Dynamic Stability of Structures. Cambridge University Press, 2006.

293

18:9

P1: KAE Chapter-15

CUFX159-Virgin

0 521 85648 5

August 14, 2007

15 Harmonic Loading: Transverse Excitation

15.1 Introduction: Resonance Effects We have already seen many examples of how the presence of axial load tends to reduce the lateral stiffness and hence natural frequencies. In this chapter, we shall consider the effect of axial loads on the steady-state response of forced structural systems. This section will focus on an important class of forcing functions, that is, harmonic excitation. Thus, in Fig. 3.1 we might have F (t) = F0 sin ωt, or z(t) = z0 sin ωt, say. In the former case, we have a governing equation of motion of the form Mx¨ + C˙x + K(1 − p)x = F0 sin ωt,

(15.1)

in which M, K, and C represent physical properties associated with a slender structural system. We again assume that the spring stiffness is reduced by the presence of a parameter p, later to be identified with axial load [1]. The solution of Eq. (15.1) consists of the summation of two parts. First, the homogeneous solution is obtained from the free vibration and was derived previously. For typical damping values, it consists of an exponentially decaying oscillation (assuming p < 1). Second, the particular solution consists of a steady-state oscillation, X0 eiωt , where the magnitude of the steady-state response (relative to the forcing magnitude) is given by X0 1 = R(ω) = , F0 [K(1 − p) − ω2 M]2 + (Cω)2

(15.2)

and is often referred to as the receptance, amplitude-response, or frequencyresponse function (FRF) [2]. We observe the important resonant effect if the driv√ ing frequency ω is close to the natural frequency ωn = K/M of the system, or = ω/ωn = 1. Figure 15.1 shows the receptance for the parameter values K = M = 1 and C = 0.2 as a function of the destabilizing parameter p. We note that increasing p tends to shift the resonant peaks toward lower frequencies [with negative p (tensile) having the opposite effect]. The receptance could also have been nondimensionalized with respect to the effective stiffness in which case the curves would have emanated from a common point on the y axis. 294

12:33

P1: KAE Chapter-15

CUFX159-Virgin

0 521 85648 5

August 14, 2007

15.1 Introduction: Resonance Effects

295

20

Receptance R (ω)

p = 1.0 p = 0.9

15

p = 0.6 10

p = 0.3 p = 0.0 5

0

0

0.5

1

0.5

2

Figure 15.1. Effect of diminished stiffness on the receptance of a spring–mass–damper.

15.1.1 A Single-Mode Approximation We have seen a number of times how it is possible to make a relatively accurate single-mode analysis of beam dynamics. Consider a thin elastic beam of length L, flexural rigidity EI, and mass m (per unit length), which is clamped at both ends and subject to a compressive axial load of magnitude P. A single-mode energy analysis of this system, assuming a mode shape of the form Q(t) 2πx w(x, t) = 1 − cos , (15.3) 2 L can be conducted along the lines of Chapter 7, resulting in a natural frequency of 2 2π 2π 4 1 2 −P ω = . (15.4) EI 3m L L From this we immediately see that buckling occurs when Pcr = EI(2π/L)2 (exact), and in the absence of the axial load, we obtain a natural frequency of ω0 = 22.79 EI/(mL4 ) (exact coefficient = 22.37). Using these to nondimensionalize ( p¯ = P/Pcr and ω¯ = ω/ω0 ) we have ¯ ω¯ 2 = 1 − p.

(15.5)

Now, subjecting the system to a transverse harmonic point force at mid-span, F (t), and assuming a small amount of linear-viscous damping, C, we have the equation of motion given by Eq. (15.1) in which x = Q,

M = 3m,

K = EI(2π/L)4 ,

p = P/Pcr ,

and the corresponding receptance is still given by Eq. (15.2).

(15.6)

12:33

P1: KAE Chapter-15

CUFX159-Virgin

0 521 85648 5

August 14, 2007

296

Harmonic Loading: Transverse Excitation

1

p 0.8

0.6

0.4

0.2

0 0

0.5

1

1.5

ω

2

Figure 15.2. Effect of axial load on the receptance of a damped, SDOF model of a beam as a contour plot. Darker shades are higher in magnitude. C = 0.1.

The response can also be plotted as a contour plot in terms of axial load and forcing frequency. This is shown in Fig. 15.2. In this plot, we can observe how the resonant peaks spread as the axial load is increased. This can be viewed as an increase in the damping ratio (as this is relative to the natural frequency and hence stiffness). A closely related circumstance is what happens when the support upon which the mass is supported is excited [e.g., z(t) = z0 sin ωt]. This is called transmissibility, and will be studied in detail a little later in relation to vibration isolation. We conclude that an axial load not only has the effect of shifting resonant peaks to lower frequencies but also increases the effective damping in the system. We can again obtain a useful physical sense of the effect of the changing axial load on the forced vibration problem by evolving the axial load as a linear function of time: p = 0.002t. In this way, the stiffness of the system will reduce to zero when t = 500. Figure 15.3 shows an example based on numerical simulation of the governing equation of motion, including the diminishing stiffness. The forcing parameters are fixed at F0 = 1 and ω = 0.5, and hence resonance should occur when t ≈ 375. Note that there is again a small amount of overshoot in the nonstationary (slowly evolving, or swept) response.

15.1.2 Beyond Buckling We can extend the single-mode energy analysis of this system by including higherorder terms in the potential energy of this system (truncated after the second term): V=

1 EI 2

L 0

1 [w2 + w2 w2 ] dx − P 2

L

0

1 w2 + w4 dx, 4

(15.7)

12:33

P1: KAE Chapter-15

CUFX159-Virgin

0 521 85648 5

August 14, 2007

15.2 The Poincare´ Section

297

10 x(t)

7.5 5.0 2.5 0.0

−2.5 −5.0 −7.5 −10 0

100

200

300

400

500

600 t

Figure 15.3. A sweep through decaying stiffness and passing through resonance.

and the kinetic energy remains the same. Evaluating the potential-energy expression now gives 4 6 1 2π 2π 1 2 V= EIL EIL Q + Q4 16 L 256 L 2 2π 1 1 3L 2π 4 4 P − PL Q2 − Q . 16 L 128 8 L

(15.8)

In addition to the trivial (Q = 0) solution, we now have a nontrivial (post-buckled) path given by 8(p¯ − 1) Q2 = 2

, 2π 3 1 − p ¯ L 4

(15.9)

where p¯ =

EI

P 2π 2 .

(15.10)

L

After buckling, the nondimensional natural frequency is given by 9 ( p ¯ − 1) 3( p ¯ − 1) ω¯ 2 = 1 +

− p¯ 1 + 4

= ω¯ 2 = 2(p¯ − 1), 1 − 34 p¯ 1 − 34 p¯

(15.11)

that is, half the prebuckling slope of the load–frequency (-squared) relation [Eq. (15.5)]. This is a result anticipated from the normal form of the supercritical pitchfork bifurcation (Section 7.4).

15.2 The Poincare´ Section Before moving on to consider the resonance response of axially loaded continuous systems we introduce the concept of Poincare´ sampling. The response of the forced nonlinear oscillator of the type of Eq. (15.1) with p = 0 is typically given in terms of transient (see Chapter 3) and steady-state parts. In the framework of

12:33

P1: KAE Chapter-15

CUFX159-Virgin

0 521 85648 5

August 14, 2007

298

Harmonic Loading: Transverse Excitation

dynamical system theory, we can view the steady state as a periodic attractor for the surrounding transients. For a linear system, the periodic attractor is unique. We shall see in the next chapter that this is not necessarily the case for nonlinear oscillators. A typical engineering approach is then to plot the maximum amplitude of response as a function of the forcing frequency (see Fig. 15.1). Often the phase difference between the forcing function and the response is also plotted and a sudden shift in phase is associated with resonance [3, 4]. However, an alternative description of the response is to reduce the 3D phase space in continuous time to a 2D phase space in discrete time by Poincare´ sampling [5, 6]. The complete solution to Eq. (15.1) with p = 0 can be written as x(t) =

F0 sin(ωt − φ) + X1 e−ζωn t sin( 1 − ζ 2 ωn t + φ1 ), K 2 [1 − (ω/ωn )2 ] + [2ζω/ωn ]2 (15.12)

and focusing on the steady-state solution we ignore the second term and write Eq. (15.12) in the alternative form x(t) = a cos(ωt) + b sin(ωt).

(15.13)

Differentiating this to get the velocity, we have y(t) ≡ x˙ = −aω sin(ωt) + bω cos(ωt),

(15.14)

and setting t = 0 (which effectively fixes the initial forcing phase), we simply get x = a and y = bω as the fixed point location, where a= b=

(1 − 2 )f , (1 − 2 )2 + (2ζ) 2

(15.15)

2ζ f , + (2ζ, )2

(15.16)

(1 −

2 )2

and = ω/ωn . The Poincare´ section can thus be considered as an alternative to the more conventional (amplitude–phase) representation of the response of an oscillator. The complementary function (the transient solution) can also be included in the following way to give a discrete mapping: the Poincare´ map P [6], ⎡ ⎤ 1 n S S C + ζω −2πζωn x x ωd ωd ⎣ ⎦ →e ω 2 ω ζω y y − ωnd S C − ωdn S ⎤ ⎡ −aC + − ζωωnda − bω S −2πζωn a ωd ⎦+ 2 +e ω ⎣ , (15.17) aωn nω bω S −bωC + ωd + ζbω ωd √ where C ≡ cos(2πωd /ω), S ≡ sin(2πωd /ω), and ωd = ωn 1 − ζ 2 . Thus, given some initial conditions, this set of difference equations will map out the transient at

12:33

P1: KAE Chapter-15

CUFX159-Virgin

0 521 85648 5

August 14, 2007

15.3 Continuous Systems

299

intervals of the forcing period until converging on the fixed point (x, y) = (a, bω).

(15.18)

This mapping is exact for a linear oscillator (and is related to the Z transform [7]) but cannot usually be easily obtained for nonlinear systems. However, importantly, this complete mapping contains the stability information regarding the fixed point and relates back to the section on Floquet theory and characteristic multipliers (see Section 14.2). We have already introduced the concept of characteristic eigenvalues (exponents) (CEs) for determining stability of equilibria in unforced systems. Now we see that it is the eigenvalues of the map (characteristic multipliers, or CMs) that determine the stability of cycles. The eigenvalues of the Jacobian, that is, the first partial derivatives of the map given by Eq. (15.17), DP(a, ωb), are given by λ1,2 = e−

2πζωn ω

±i

2πωd ω

,

(15.19)

which confirms that the fixed point is asymptotically stable, because the damping and natural frequency are positive numbers. This is why consideration of discrete maps plays a useful role in the study of flows. Despite the fact that this approach tends to hide the usually important engineering aspects of amplitude and phase, it does provide a very convenient means of assessing stability, and the evolution of the system responses under the slow change in a parameter. For the types of nonlinear system to be considered later, Poincare´ sampling provides a powerful tool in the numerical and experimental investigation of periodically excited nonlinear oscillators.

15.3 Continuous Systems We now turn to consideration of axially loaded, transversely forced, continuous beams. The forced string and membrane can also easily be handled by use of these techniques. A beam of length L, mass per unit length m, uniform flexural rigidity EI, viscous damping coefficient C, axial force P, and transverse load Q0 F (x) cos t, respond as w(x, t). The governing equation of motion is mwtt + Cwt + EIwxxxx + Pwxx = Q0 F (x) cos T,

(15.20)

in which subscripts on w reflect partial derivatives. This equation can be put in the nondimensional form ∂4 w¯ ∂2 w¯ ∂w¯ ∂2 w¯ ¯ t¯, + p + c + = f (x) cos (15.21) ∂x¯ 4 ∂t¯ ∂x¯ 2 ∂t¯2 with x¯ = x/L, w¯ = w/L, t¯ = t EI/(mL4 ), √ ¯ = mL4 /(EI), c = CL2 / mEI, (15.22) p = PL2 /(EI), f (x) = Q0 F (X)L3 /(EI),

ζ = c/(21 ),

12:33

P1: KAE Chapter-15

CUFX159-Virgin

0 521 85648 5

August 14, 2007

300

Harmonic Loading: Transverse Excitation L Q cos T

X

EI, m

P W(X,T)

Figure 15.4. A cantilever beam subject to a constant axial load and a harmonically varying uniformly distributed lateral force.

where 1 is the fundamental natural frequency and its value depends on the boundary conditions. For convenience, we now drop the overbar notation, and assuming a steady-state harmonic response of the form w(x, t) = Re y(x)eit

(15.23)

y (x) + py (x) + (ic − 2 )y(x) = f (x).

(15.24)

leads to

Equation (15.24) can then be solved for a specific set of boundary conditions and transverse forcing types. A number of examples with distributed, transverse, harmonic forcing can be found in Virgin and Plaut [1]. Here, two cases will be considered: r an axially loaded cantilever beam with a uniformly distributed harmonic load, and r an axially loaded, clamped–clamped beam with a harmonic central point load. Considering the first case as shown in Fig. 15.4 we have a governing equation of motion given by Eq. (15.24) but now with unity on the right-hand side. The general solution is given by y(x) = (ic − 2 )−1

2 (aj cosh λj x + bj sinh λj x), j =1

where λj = (νj /γj ) + i(γj /2),

1/2 γj = 2 − j + 2j + ν2j ,

ν1 = φ/2,

ν2 = −φ/2,

1 = ( − p)/2,

2 = −( + p)/2, 1/2 φ = − δ + δ2 + (c)2 /2 ,

=−

c , 2φ

δ = p 2 + 42 .

(15.25)

12:33

P1: KAE Chapter-15

CUFX159-Virgin

0 521 85648 5

August 14, 2007

15.3 Continuous Systems

301

0.4

A 0.3

p = π2 / 5 p=0

0.2

p = - π2 / 10 0.1

0 0

5

10

15

20

Ω

25

Figure 15.5. The central amplitude versus the forcing frequency for a cantilever beam with a uniformly distributed force.

The boundary conditions in this case are y(0) = 0, y (0) = 0, y (1) = 0, and y (1) + py (1) = 0. The elastic critical load is p cr = π2 /4, and the fundamental natural frequency is 1 = 3.516. Applying the boundary conditions and solving the resulting simultaneous equations (in aj , bj ) leads to the results shown in Fig. 15.5 in which a damping ratio of ζ = 0.142 was used. The amplitude A is the magnitude of the maximum central deflection, which depends on the static axial load and the forcing frequency, i.e., A(, p) = |y(0.5)|. Note the presence of the second resonant peak in the vicinity of = 22. It is interesting to see how the amplitude and the corresponding resonant frequency vary with axial load. Figure 15.6 shows the relation AR (p) = max A(, p), ≥0

(15.26)

and the values of (squared) for which this condition occurs. Three representative damping values are used. We again get a near-linear relation between frequency squared and the axial load. Now consider the second case. Figure 15.7(a) shows the amplitude response (receptance) when ζ = 0.02. We again see the anticipated increase in natural frequency for a tensile axial force, and reduction for compression. Figure 15.7(b) shows how the receptance associated with the third (second symmetric) mode is affected by the presence of an axial load. In this case, the undamped third frequency occurs at = 120.9 for the unloaded case. The peaks are also shifted slightly from those of the undamped case because of the presence of a little damping.

12:33

P1: KAE Chapter-15

CUFX159-Virgin

0 521 85648 5

August 14, 2007

302

Harmonic Loading: Transverse Excitation

2.5

2.5

(a)

(b)

p 2

2

p

ζ = 0.2 1.5

1.5

ζ = 0.05 1

1

ζ = 0.1

0.5

ζ = 0.1

0.5

ζ = 0.05 0

−0.5 0

ζ = 0.2

0

0.2

0.4

0.6

0.8

AR

1

−0.5

0

5

10

2

ΩR

15

Figure 15.6. (a) The resonant amplitude, (b) the frequency squared versus the axial load for a cantilever beam with a uniformly distributed force.

In both of the preceding examples, it can be shown that the resonant amplitude and frequency are given approximately by AR (p) ≈ AR (0)/ 1 − (p/p cr ), 2R

≈

21 [1

− (p/p cr )],

(15.27) (15.28)

and we can also relate this back to the material presented in Chapter 7, for example, the p = 0 resonant peak for the beam in Fig. 15.7 occurs at the natural frequency coefficient of 22.4. Experimental Verification. A thin steel strip was clamped between blocks at its ends

and placed in a displacement-controlled testing machine. In this configuration, the end shortening is prescribed, and the resulting axial force is measured with a load cell. With the standard expressions based on the earlier analysis, the critical load was computed at 1235 N and the lowest natural frequency (with no axial load) at 44 Hz. The strut was struck by an impact hammer and a laser velocity vibrometer was used to measure the response [8]. The data were then acquired and analyzed with the Bruel and Kjaer pulse system. Velocity time series were then subject to a Hann window and a fast Fourier transform algorithm to extract the frequency

12:33

P1: KAE Chapter-15

CUFX159-Virgin

0 521 85648 5

August 14, 2007

15.3 Continuous Systems

303

0.3

0.02

y(0.5) f0 0.25

p=4

(a)

2

p=3

2

p=2 0.2

2

p = 0

2

p=0

0.15

p = 2

0.015

2

p=

(b)

y(0.5) f0

p=-

0.01 2

0.1 0.005 0.05

0

0

5

10

15

20

25

30

0 110

115

120

125

130

Figure 15.7. (a) Amplitude response of an axially loaded, clamped–clamped beam subject to a central point harmonic point force and (b) in the vicinity of the third mode.

content. This process was repeated 10 times at each axial-load level. The results were averaged and displayed as normalized mobility, that is, normalized with respect to the force of the impact hammer. A photograph of the experimental system is shown in Fig. 15.8(a). Figure 15.8(b) shows a summary of how the peaks shift to lower (higher) frequency as the compressive (tensile) axial load is increased. However, experimental results from this type of system need careful consideration because testing machines are sometimes referred to as semirigid loading devices, membrane effects in the beam may occur, and even the boundary conditions can be a function of loading [9, 10].

(a)

(b) 1

p 0.5

0 −0.5 −1 −1.5 −2 −2.5 0

0.5

1

1.5

2

2.5

3

3.5

( ω/ω ) 0

Figure 15.8. (a) A photograph of the clamped beam in the testing machine and (b) the axial load plotted as a function of the natural frequencies (squared).

2

12:33

P1: KAE Chapter-15

CUFX159-Virgin

0 521 85648 5

August 14, 2007

304

Harmonic Loading: Transverse Excitation m X

c

k Z

Figure 15.9. Schematic of a mass isolated from the motion of a foundation.

15.4 An Application to Vibration Isolation In many instances, buckling is viewed as an undesirable occurrence, in particular when it precipitates a total loss of stiffness and collapse (see, for example, Section 3.4). However, there are ways in which postbuckled stiffness can be exploited, and this section introduces an example. The concept of vibration isolation, that is, how to reduce the force or motion transmitted to a device from a source of vibration, is well established [11–14]. This section describes an approach to effective vibration isolation by use of the subtle interplay of axial loads, dynamics, and stability [15]. Consider a simple mechanical system consisting of a mass (in a gravitational field) supported by a spring and damper, which are themselves supported on a base, as shown in Fig. 15.9. If the motion of the base is harmonic, for example, Z(t) = Z0 sin ωt, then it can be shown that the steady-state displacement transmissibility, X/Z (where X is the response amplitude of the mass), is given by the expression 1/2 1 + (2ζ)2 X = . Z (1 − 2 )2 + (2ζ)2

(15.29)

This is plotted in Fig. 15.10 as a function of the frequency ratio = ω/ωn , where √ ωn = k/m. Damping tends to severely reduce the resonant peak. This was also true for the cases examined earlier in this chapter in which the force was applied to the mass directly. In these earlier cases, damping tended to reduce the magnitude of the response at all frequencies. Here, we notice an interesting feature in which (linear-viscous) damping results in slightly larger responses at higher √ frequencies. We also observe that damping has no influence when is exactly 1/ 2. In addition to direct mass and base excitation, a third type of resonance can also be found in a system with a rotating unbalance [4]. However, overall, we see that√the transmissibility is small for relatively highfrequency ratios, that is, for > 2, X/Z < 1.0. Given a forcing frequency ω, a typical design option would be to mount the device on a soft spring to induce a low natural frequency, ωn . But if the spring has a low stiffness, there is a danger that

12:33

P1: KAE Chapter-15

CUFX159-Virgin

0 521 85648 5

August 14, 2007

15.4 An Application to Vibration Isolation

305 8 X/Z

ζ = 0.0

6

ζ = 0.1

Figure 15.10. Displacement transmissibility for a SDOF oscillator for four typical damping values.

4

ζ = 0.2 ζ = 0.3

2

0 0.5

1

1.5

2

it will deflect statically too much (because mg = kx), and this often places practical limits on the spring stiffness.

15.4.1 Postbuckling of a Strut Revisited In earlier chapters, we have seen how axially loaded structures typically possess nonlinear characteristics, especially close to, or beyond, initial buckling. Often this takes the form of additional postbuckled stiffness (e.g., in plates). In their postcritical state, they exhibit relatively low stiffness in the axial direction and yet they carry axial loads above their buckling load. As indicated schematically in Fig. 15.9 we see a potential opportunity in the context of vibration isolation. Let’s return to the simply supported strut as shown in Fig. 7.1. Any structure exhibiting stable postbuckled behavior can be used in this situation, but the pinned beam is easiest to analyze. Thus, we are interested in a structural system of the supercritical type. We have already seen [see Eq. (7.51)] that the initial postbuckled equilibrium configuration is described by P π2 Q 2 =1+ , (15.30) Pe 8 L where Pe = EI(π/L)2 is the classical Euler critical load. In Chapter 7 we were primarily interested in lateral stiffness effects, but now we need to consider stiffness in the axial direction as this is the direction in which the force acts. The geometric relation between the lateral deflection Q and the end shortening δ can be established as [15] δ π2 Q 2 3π4 Q 4 = + . (15.31) L 4 L 64 L The end shortening is approximately related to the square of the lateral deflection, and eliminating Q in Eqs. (15.30) and (15.31) leads to P 1 δ 1 δ . (15.32) = 2+ 1+3 ≈1+ Pe 3 L 2 L

Ω

2.5

12:33

P1: KAE Chapter-15

CUFX159-Virgin

0 521 85648 5

August 14, 2007

306

Harmonic Loading: Transverse Excitation 1.2

1.2

(a)

P/P e

(b)

P/Pe

1.1

_ p

1.1

_ δ

1

1

0.9

0.9

0.8

0.8

0.7

0.7

0.6

0

0.05

0.1

0.15

0.2 0.25 Q/L

0.6

0

0.05

0.1

δ /L

0.15

Figure 15.11. Deflection of the strut as a function of axial load: (a) central lateral deflection, (b) end shortening.

This is a result based on moderate lateral deflections, i.e., up to about 20% of the length. Recall that this is a nonlinearity induced by the curvature expression [Eq. (7.40)], rather than the membrane effect discussed in Section 7.5 for example, that is, the ends are free to move toward each other. Equations (15.30) and (15.32) are shown by the solid curves in Figs. 15.11(a) and 15.11(b), respectively. The postbuckled stiffness is only mildly affected by initial imperfections (unlike in the vicinity of the critical point), and when a single mode is adopted, representative initial geometric imperfection leads to the dashed curves in Fig. 15.11. If we load the strut axially to slightly above its elastic critical load, for example, P/Pe = 1.05, then (for Q0 = 0) we have Q/L ≈ 0.2, δ/L ≈ 0.1. This specific load-deflection condition furnishes an equilibrium position from which incremental coordinates are measured: p¯ = P/Pe − 1.05, δ¯ = δ/L − 0.1, and indicated in Fig. 15.11(b). This strut, then, is able to support a relatively high axial load (sufficient to cause buckling) but exhibits the desirable soft spring characteristic.

15.4.2 Experimental Verification An experimental verification is considered in this section and configured as shown in Fig. 15.12(a). Part (b) shows an alternative configuration in which four postbuckled panels provide the support. The vertical shaker was connected to a cam-shaft attached to a variable-speed motor. The forcing amplitude was fixed at 3 mm, and the shaker then imparted an almost sinusoidal motion through the isolation system (consisting of two steel struts made of spring steel) to the mass, which moved

12:33

P1: KAE Chapter-15

CUFX159-Virgin

0 521 85648 5

August 14, 2007

15.4 An Application to Vibration Isolation

307

Figure 15.12. (a) Photograph of the experimental setup and (b) alternative configuration in which panels re used.

(a)

(b)

vertically, guided by low-friction linear bearings. The frequency of excitation was then varied over an appropriate range, and the vertical motion of the mass was measured [15]. Plots of axial load versus lateral deflection and end shortening are shown in Fig. 15.13, where the deflections were measured with a linear-variable displacement transformer. The two struts were 268 mm long, 19 mm wide, 0.66 mm thick (and thus I = 4.55 × 10−13 m4 ), and taking a typical value for Young’s modulus of 200 GPa, we anticipate an Euler load in the vicinity of 25 N ≡ 2.55 kg. A Southwell plot can be used to recast the data from Fig. 15.13(a) to estimate a critical load of approximately 23 N. Because of initial geometric imperfections, the “critical load” is again manifest as a relatively rapid increase in the deflection that is due to additional load. The axial load versus lateral deflection, and versus end shortening, results are shown in Figs. 15.13(a) and 15.13(b), respectively, and these relations illustrate a good correlation with the corresponding theoretical curves (including an initial imperfection) given in Fig. 15.11. By the choice of an appropriate point on the curve in Fig. 15.13(b) as the fundamental equilibrium position, for example, P = 23.5N → (P/Pe ≈ 1.0), δ = 15.2 mm → (δ/L = 0.057), the transmissibility can be assessed over a range of excitation frequencies. Locally, the stiffness is approximately 195 N/m (i.e., the slope of P/Pe versus δ/L about the chosen operating point), and because the mass is 2.4 kg, we would thus expect a natural frequency of free vibration close to 9 rad/s ≡ 1.43 Hz. A free decay of this system gives a natural period of approximately 0.68 s (and hence ωn =√1.47 Hz), and thus, we anticipate effective isolation for forcing frequencies ω > 2ωn ≈ 2.2 Hz. If a conventional linear (helical) spring had been used instead of the buckled struts, a static deflection of approximately four times the deflection of the struts would have resulted from this load level.

15.4.3 The Forced Response The transmissibility √ratio [given by Eq. (15.29)] should be low, more specifically less than one, for > 2. Subjecting the system to a range of excitation frequencies (at

12:33

P1: KAE Chapter-15

CUFX159-Virgin

0 521 85648 5

August 14, 2007

308

Harmonic Loading: Transverse Excitation 30

30

(a)

Force (N)

(b)

Force (N)

25

25

20

20

15

15

10

10

5

5

_

_

( δ , p)

0

0 0

10

20

30

40

50

0

5

10

15

Q (mm)

20 25 δ (mm)

Figure 15.13. Axial load versus (a) midpoint lateral deflection and (b) end shortening for the experimental system.

constant forcing amplitude) leads to the experimental results shown in Fig. 15.14. Three typical time series are shown for the frequency ratios indicated. The response when the forcing frequency is exactly twice the natural frequency (indicated by the open circle) shows an interesting subharmonic of order two. This is a consequence of combined parametric and external forcing terms in the governing dynamic equations (see Chapter 14). In general we see a highly attenuated response for higher frequencies; that is, in this frequency range, the mass is effectively isolated from the motion of the base. This concept can be extended in a number of ways [16], and the usual issues of avoiding stroke-out, nonlinear behavior, and fatigue still apply. Here some research is mentioned in which axially loaded structures are taken advantage of to amplify, rather than reduce, motion [17, 18], and how axial load can be used to tune resonant frequencies for the purposes of energy scavenging [19].

15.5 Forced Excitation of the Thermally Buckled Plate In Chapter 10 we saw how axial loading (thermal) effects influenced the free vibration of thin plates. At ambient temperatures, the plate would exhibit a periodic response if subject to a periodic lateral excitation, and increasing the thermal loading would result in the now-familiar shift in resonance characteristics. For temperatures above the critical buckling level, two coexisting equilibria appear, that is,

12:33

P1: KAE CUFX159-Virgin

0 521 85648 5

August 14, 2007

15.5 Forced Excitation of the Thermally Buckled Plate

309

1 X/Y 0.8

0.6

0.4

0.2

0 1

2

3

4

5

6

7

8

Figure 15.14. Experimental transmissibility for the displacement of the strut-supported mass.

supercritical behavior. For relatively low levels of external excitation, we still expect periodic response but now offset about some nontrivial equilibrium configuration. We shall use a panel system similar to the one described in Section 10.5 to illustrate this effect experimentally (the only difference being that now the panel is thinner). In the final chapter, we shall look at more intense excitation that can result in large-amplitude responses, including chaos and intermittent snapping between these equilibria [20]. Typical periodic responses are shown in Fig. 15.15 in which two alternative coexisting oscillations are superimposed, that is, motion about both static (postbuckled) equilibrium configurations within the corresponding potential energy wells. This is the situation shown schematically in Fig. 10.6. The measurand in this case is strain and is plotted against strain a quarter of a cycle later in part (b), by use of a standard embedding technique in nonlinear dynamics [21]. These results were recorded

PSD

(t +T/4)

(c)

10 2 0

10 -1

y

0

10 5

(b)

(a)

y

Chapter-15

10 -4

0

0.05 t (s)

0.1

10 -7

y

0

(t)

0

(Hz)

Figure 15.15. Small-amplitude periodic behavior about both postbuckled positions using microstrain and time-lag embedding: (a) time series, (b) phase projections, and (c) frequency spectrum.

500

12:33

P1: KAE Chapter-15

CUFX159-Virgin

0 521 85648 5

310

August 14, 2007

Harmonic Loading: Transverse Excitation f

at (ζ, η) = (0.583, 0.416) with the temperature set at T/T cr = 1.76, with an excitation level1 of 130 dB at 120 Hz (quite near the lowest natural frequency at this temperature). The second attractor (which has a slightly smaller basin of attraction) is attained by giving a variety of perturbations access to different areas of the initial condition space. Because of the inevitable initial geometric imperfections there is a mild asymmetry, such that the postbuckled plate has a slightly tilted underlying potential-energy function [see Fig.3.8(a)]. This asymmetry is also reflected in a slightly different location in the phase projection of the two responses as well as a slight difference in the period in their time series. This kind of small-amplitude periodic behavior possesses a power spectrum with a dominant spike at the forcing frequency [see Fig. 15.15(c)]. The sharpness of the resonant peak is a standard means (the half-power method) by which the damping can be estimated [22, 23]. A good correlation with a theoretical analysis is described in Murphy et al. [24].

References [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17]

1

L.N. Virgin and R.H. Plaut. Effect of axial load on forced vibrations of beams. Journal of Sound and Vibration, 168:395–405, 1993. D.J. Ewins. Modal Testing: Theory and Practice. Research Studies Press, 1984. W.T. Thomson. Theory of Vibration with Applications. Prentice Hall, 1981. D.J. Inman. Engineering Vibration. Prentice Hall, 2000. D.W. Jordan and P. Smith. Nonlinear Ordinary Differential Equations. Oxford University Press, 1999. J. Guckenheimer and P.J. Holmes. Nonlinear Oscillations, Dynamical Systems, and Bifurcations of Vector Fields. Springer-Verlag, 1983. K. Ogata. System Dynamics. Prentice Hall, 1998. G.C. Goodwin and R.L. Payne. Dynamics System Identification: Experiment Design and Data Analysis. Academic, 1977. A. Picard, D. Beaulieu, and B. Perusse. Rotational restraint of a simple column base connection. Canadian Journal of Civil Engineering, 14:49–57, 1987. R.H. Plaut. Column buckling when support stiffens under compression. Journal of Applied Mechanics, 56:484, 1989. F.C. Nelson. Vibration isolation: A review, I. Sinusoidal and random excitations. Shock and Vibration, 1:485–93, 1994. R.H. Racca. Characteristics of vibration isolators and isolation systems. In Shock and Vibration Handbook, 4th ed. McGraw-Hill, 1996, Chapter 32. J. Winterflood, T. Barber, and D.G. Blair. High performance vibration isolation using spring in Euler column buckling mode. Physics Letters A, 19:1639–45, 2002. E.I. Rivin. Passive Vibration Isolation. ASME, 2003. L.N. Virgin and R.B. Davis. Vibration isolation using buckled struts. Journal of Sound and Vibration, 260:965–73, 2003. R.H. Plaut, J.E. Sidbury, and L.N. Virgin. Analysis of buckled and pre-bent fixed-end columns used as vibration isolators. Journal of Sound and Vibration, 283:1216–28, 2005. J. Jiang and E. Mockensturm. A motion amplifier using an axially driven buckling beam: I. Design and experiments. Nonlinear Dynamics, 43:391–409, 2006. The reference value for the decibel scale is the rms value 20 µN/m2 for sound pressure level.

12:33

P1: KAE Chapter-15

CUFX159-Virgin

0 521 85648 5

August 14, 2007

References [18] J. Jiang and E. Mockensturm. A motion amplifier using an axially driven buckling beam: II. Modeling and analysis. Nonlinear Dynamics, 45:1–14, 2006. [19] E.S. Leland and P.K. Wright. Resonance tuning of piezoelectric vibration energy scavenging generators using compressive axial load. Smart Materials and Structures, 15:1413– 20, 2006. [20] K.D. Murphy, L.N. Virgin, and S.A. Rizzi. Experimental snap-through boundaries for acoustically excited, thermally buckled plates. Experimental Mechanics, 36:312–17, 1996. [21] L.N. Virgin. Introduction to Experimental Nonlinear Dynamics: A Case Study in Mechanical Vibration. Cambridge University Press, 2000. [22] D.E. Newland. An Introduction to Random Vibrations and Spectral Analysis. Longman, 1984. [23] J.S. Bendat and A.G. Piersol. Random Data: Analysis and Measurement Procedures. Wiley, 1986. [24] K.D. Murphy, L.N. Virgin, and S.A. Rizzi. Characterizing the dynamic response of a thermally loaded, acoustically excited plate. Journal of Sound and Vibration, 196:635– 58, 1996.

311

12:33

P1: RTT Chapter˙16

CUFX159-Virgin

0 521 85648 5

August 14, 2007

16 Nonlinear Vibration

PART I: FREE VIBRATION 16.1 Introduction This last chapter considers the dynamic response of axially loaded structural systems in which the motion is not necessarily confined to the local vicinity of an underlying equilibrium position and dynamic behavior is not necessarily harmonic. In a number of places throughout this book, a statement has been made to the effect that largeamplitude behavior will be described later. We now finally consider such situations, largely in terms of revisiting examples detailed in earlier examples, but now, not relying on certain restrictions, for example, linear, or small-amplitude, behavior. Both free and forced vibrations will be considered [1, 2]. By way of a simple introduction, we go back to the softening cable example described in Section 3.5, and specifically consider the context of Fig. 3.12. This is a free vibration started (with initial conditions) some distance from any of the three available stable equilibrium points present at this level of loading. Because there is no damping, the total energy is conserved, and thus phase trajectories can be viewed as contours of constant total energy. Figure 16.1(a) illustrates the energy levels as a contour plot, and thus we can view the phase trajectory of Fig. 3.12 living in the second darkest shade within the contours of Fig. 16.1(a). Parts (b)–(d) give specific examples of time series generated (numerically) by different initial conditions. We see that the time series in part (b), which corresponds to the phase trajectory shown in Fig. 3.12, is far from sinusoidal. The time series shown in part (c) has relatively small amplitude with a natural frequency close to that predicted by linear theory; see Fig. 3.2 (but still slightly nonlinear; note the expanded y-axis). The time series shown in part (d) is initiated from a position very slightly removed from one of the unstable equilibria. After remaining in the local vicinity of the unstable point (previously calculated to be at q = 1.92), the trajectory moves away, undergoing a long-period motion, which again is far from sinusoidal. Parts (b) and (d) illustrate how the frequency of the response is highly dependent on the amplitude of motion and hence on the initial conditions. Other types of motion are exhibited following different initial conditions and values of the parameter (which is assumed to be maintained at the same level during the motion). The next section will briefly describe a couple of approximate analytical techniques primarily designed to assess the effects of moderate nonlinearity on the 312

18:14

P1: RTT Chapter˙16

CUFX159-Virgin

0 521 85648 5

August 14, 2007

16.2 Abstract Models

313 (a)

. q

2

(b)

q4

t

2 −2 −4

1

10

20

30

40

50

(c)

q 0.4 t

0.2 0 −0.2 −0.4 −1

q

10

20

30

40

50

(d)

2 1

−2

−1 −4

−2

0

2

4

q

t 10

20

30

40

−2

Figure 16.1. (a) A contour plot of total energy versus the state variables. λ = −0.5. Time series started from q˙ = 0, and (b) q(0) = 4.8; (c) q(0) = 0.5; (d) q(0) = 1.91.

system response, for example, in going from the type of motion in part (c) to that in part (d).

16.2 Abstract Models At various points in this book, including the previous section, we have seen how the natural frequency of a system may sometimes depend on the amplitude of the motion. It is straightforward to integrate the nonlinear equation of motion numerically, but it also useful to be able to obtain an analytical relation between frequency and amplitude, for example. In free vibrations of undamped systems, it may be possible to obtain an exact relation based on elliptic integrals [3], or use can be made of the conservation of total energy to facilitate a solution [4]. A variety of approximate analytical techniques have also been developed, typically applicable to moderately large (oscillatory) behavior. Suppose we go back to the simplest link model, first considered in Section 5.2, and set the axial load level at p = 1.2 corresponding to the equation of motion θ¨ + ω2n (θ − 1.2 sin θ) = 0.

(16.1)

Because there is no damping, we expect trajectories to trace phase trajectories at constant values of the total energy. Previously we expanded the sine term as a Taylor series, retaining just the first term for a standard linearization. Suppose we keep the next term in the series expansion and shift the origin to the positive equilibrium position (i.e., at the bottom of the right-hand potential-energy well with p = 1.2, we have an equilibrium at θe = 1.02674). The potential energy corresponding to this situation is shown in Fig. 16.2(a) as the solid curve and corresponds to the cubic

50

18:14

P1: RTT Chapter˙16

CUFX159-Virgin

0 521 85648 5

August 14, 2007

314

Nonlinear Vibration

(a)

0.2

V 0.1

x −3

−2

−1

1 −0.1 0.2

R (b) 0.1

x −3

−2

−1

1 −0.1

Figure 16.2. (a) Potential-energy functions for the inverted pendulum and (b) corresponding restoring force functions (x = θ − 1.02674 for the specific non-truncated system).

restoring force (R = Vdx) in part (b). Superimposed is the potential energy associated with linearization (long-dashed curve) together with that for a quadratic restoring force (and thus cubic potential, the short-dashed curve). There are many ways in which these curves may be fit, but the quadratic restoring force reflects the nonlinear relation between force and displacement as well as the asymmetry. We thus see how the linearization corresponds to small-amplitude motion about the equilibrium position, a distorted egg-shaped phase trajectory corresponds to motion barely contained in one of the local potential-energy wells and can be captured by the quadratic fit, and a thoroughly nonlinear cross-well behavior needs at least a cubic restoring force for global containment, or boundedness. To consider the growth from linear to mildly nonlinear motion we can thus study an equation of the form x¨ + x + x2 = 0,

(16.2)

in which x is measured from the shifted origin and the restoring force has been changed for convenience such that the unstable equilibrium position occurs at negative one. We recognize this as one of the standard forms from Chapter 3, and this system was also subject to a sudden load in Section 13.4. This system possesses a stable equilibrium at the origin and a saddle point at x = −1. Elliptic integral solutions are available, as well as approximate solutions based on the techniques of harmonic balance and perturbation methods. Further details of these methods can

18:14

P1: RTT Chapter˙16

CUFX159-Virgin

0 521 85648 5

August 14, 2007

16.3 A Mass Between Stretched Springs

315

.

x

0.4 x(0) 0.3

x

.

Figure 16.3. Natural frequency versus amplitude of motion.

x(0) = 0.4 , x(0) = 0

0.2

0.1

.

x(0) = 0.2 , x(0) = 0

0.7

0.8

0.9

1.0

be found in [4–8]. Numerical integration of Eq. (16.2) shows a softening nonlinearity, that is, the frequency tends to diminish with amplitude. This relation is shown in Fig. 16.3. Also shown are a couple of phase trajectories for the specific cases generated by the initial conditions x(0) = 0.2, x(0) ˙ = 0.0, and x(0) = 0.4, and x(0) ˙ = 0.0. The egg-shaped, asymmetric behavior of the phase trajectory started farther away from equilibrium reflects the underlying potential energy as anticipated by the form of the dashed curve in Fig. 16.2. Of course, an initial condition started slightly below x(0) = −1.0 results in a response in which the period is very long. In terms of dynamical systems theory the separatrix is a homoclinic orbit (i.e., an orbit that starts at the unstable equilibrium and ends there after infinite time) and separates bounded from unbounded (escaping) motion. An experimental analog of Eq. (16.2) based on the concept of a rolling point mass on a curved surface is described in Virgin [8] and Gottwald et al. [9]. The preceding section shows how the range of dynamic behavior is much broader when not confined to small amplitudes, especially when the axial load is somewhat higher than the initial buckling load. The membrane, or stretching, effect that was encountered earlier for axially constrained systems very easily leads to nonlinear vibrations.

16.3 A Mass Between Stretched Springs Consider the analogy between a string and a point mass supported by two identical springs. One motivation for doing this is that the effects of large amplitude and of varying axial loads can be introduced without too much mathematical sophistication. Such a system is shown schematically (and in a highly deflected configuration) in Fig. 16.4. Although the springs are linear in a direction perpendicular to the SDOF, they provide a nonlinear restoring force in the x direction (the only allowable direction), and this increases in a disproportionate sense with displacement [10–12]. If we suppose that the natural length of the springs is less than L, that is, each was stretched by an amount d (put in tension by T = kd), then we have one equilibrium position at the origin. In this case, the force acting on the mass in the x

1.1

18:14

P1: RTT Chapter˙16

CUFX159-Virgin

0 521 85648 5

August 14, 2007

316

Nonlinear Vibration

k L/2

x

m

Figure 16.4. A mass supported by two stretched springs. A single-mode analog of a string.

direction (up to cubic terms in x) is 4k(L − d) 3 kd x− x, L L3

(16.3)

4k(L − d) 3 kd x+ x = 0. L L3

(16.4)

F (x) = −2 and thus the equation of motion is m¨x + 2

Note that the linearization at x = 0 depends on d. Clearly, if the motion of the mass is small then the cubic term is negligibly small and we have a linear oscillator with natural frequency (2kd)/(mL), that is, the square of the natural frequency is linearly related to the tension in the springs. In fact, if the natural length of the springs is L (and therefore d = 0) then there is no linear restoring force (in the x direction). However, we also see that for moderately large oscillations (in the x direction) the natural frequency depends on the amplitude as well. This is an example of a system with a hardening spring stiffness in which the tension in the springs exerts a nonlinear restoring force in the direction of the motion [12]. The effective natural frequency can be obtained in a number of ways. We can again assume a harmonic form for the solution [5] x = A cos ωt,

(16.5)

which can then be placed into Eq. (16.4), and balancing (i.e., equating the coefficients of) the cosine terms (after expanding the cubic term and ignoring the third harmonic), we find the expression 3 ¯ ω¯ 2 = d¯ + A¯ 2 (1 − d), 2

(16.6)

where ω¯ 2 = mω2 /(2k),

d¯ = d/L,

A¯ = A/L.

(16.7)

18:14

P1: RTT Chapter˙16

CUFX159-Virgin

0 521 85648 5

August 14, 2007

16.3 A Mass Between Stretched Springs

317

d d

d

A

Figure 16.5. The natural frequency plotted as a function of response amplitude for a number of different pre-stress levels.

These results are plotted in Fig. 16.5. We see that if there is no initial stretching of the springs (d¯ = 0) then the natural frequency is proportional to the amplitude of the motion (within the confines of the harmonic balance approximation). It should be borne in mind that this outcome is the result of a number of approximations. First, the restoring force was expanded as a power series up to, and including, cubic terms, and second, the harmonic balance solution procedure ignored the higher harmonic terms. Despite the fact that this might not be considered an obviously axially loaded structure, it does show how the dynamic response of a system is affected by the axial forces, and in this sense it can be compared with the largeamplitude oscillations of a stretched (continuous) string [11]. Finally, it is worth noting that in a practical context damping would tend to mitigate against a system’s operating in large-amplitude motion, unless of course, the system were also subject to external forcing. This type of situation will be considered later in this chapter. Augusti’s Model Revisted. The 2DOF system considered in Section 5.7 is now

revisited. We start by numerically integrating the equations of motion for the case in which there are no initial geometric imperfections and the axial load is fixed at a level of p = 1.01134; thus the equilibrium of the system is given by (θ1e = 0.2598, θ2e = 0.0), and the two linear natural frequencies are given by ω21 = 0.0226, ω22 = 0.0535. If initial conditions are chosen with zero initial velocity but very close to equilibrium [θ1 (0) = 0.25, θ2 (0) = 0.001], then the resulting motion is harmonic. This is shown in Fig. 16.6(a) as a phase projection (in terms of one of the two angles and its rate of change). The frequency spectrum in part (b) shows a spike at the natural frequency corresponding to motion in the θ1 direction (ω21 = 0.0226, f = 0.0239 Hz) together with a small spike at twice this frequency [13]. On increasing the distance of the initial condition from equilibrium [θ1 (0) = 0.027, θ2 (0) = 0.0001], we

18:14

P1: RTT CUFX159-Virgin

0 521 85648 5

August 14, 2007

318

Nonlinear Vibration

.

x 10 2

-3

(a)

0

1 1 (dB)

1

PSD for

0 Ð1 Ð2

0.25

0.26

f

Ð 20 Ð 40

2f

0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2

Frequency (Hz)

1

.

0.02

20

(c)

1

1

(dB)

0.01

PSD for

0 Ð0.01 Ð0.02

0

0.1

0.2

0.3

0.4

(d)

0 Ð 20 Ð 40 Ð 60 Ð 80

f

0

0.03

1

(e)

0.01

PSD for

0

0

0.2

(f)

20 1 (dB)

1 0.02

Ð 0.01 Ð 0.02 Ð 0.03 Ð 0.4 Ð 0.2

0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2

Frequency (Hz)

1

.

1

Ð 80

Ð 100 0

0.27

(b)

1

Ð 60

0 Ð 20 Ð 40 Ð 60 Ð 80

0.4

0

0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2

Frequency (Hz)

1

.

0.03

(dB)

(h)

0 Ð 20

1

0.01 0 Ð 0.01 Ð 0.02 Ð 0.03 Ð 0.4 Ð 0.2

20

(g)

1 0.02

PSD for

Chapter˙16

0

0.2

0.4 1

Ð 40 Ð 60 Ð 80 0

0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2

Frequency (Hz)

Figure 16.6. Sample numerical simulation results for the Augusti model [13].

get the results shown in parts (c) and (d). Now the asymmetry in the motion is apparent, and this is reflected in the higher harmonics in the frequency spectrum. The fundamental frequency has also shifted to a lower value, again reflecting a softening spring characteristic. Now with the initial conditions set at θ1 (0) = 0.37 and θ2 (0) = 0.0015, the motion has sufficient energy to traverse the potential-energy hilltop as shown in parts (e) and (f) but is still periodic as it passes around the remote equilibrium as well.

18:14

P1: RTT Chapter˙16

CUFX159-Virgin

0 521 85648 5

August 14, 2007

16.4 Nonlinear Vibration of Strings

319

In all three of these cases, the motion in the θ2 direction was minimal. However, now suppose the initial conditions are θ1 (0) = 0.027 and θ2 (0) = 0.05, that is, similar to those used in part (c) but now with a larger initial value in the θ2 direction. This results in the behavior shown in parts (g) and (h). This motion is far from periodic as the trajectory appears to wander around and between the two stable equilibria. The frequency spectrum is now broadband with energy contributions at all frequencies. There is also now significant dynamic behavior in the θ2 direction (not shown). This is an example of chaotic behavior. Despite the apparent randomlike behavior, it is characterized by some interesting underlying order. However, this behavior is occurring in a 4D phase space and is somewhat nonrepresentative because no damping is present. In fact, for this type of behavior to occur in a continuous time (as opposed to a discrete-map) system, it must be nonlinear and have at least a 3D phase space. This is the circumstance for a forced SDOF nonlinear system, and we shall come back to it a little later in this chapter.

16.4 Nonlinear Vibration of Strings Returning now to the stretched string first considered in Section 6.2, we can restate the equation of motion but without resorting to the small-amplitude assumption (while still assuming planar motion) and thus consider 2 c21 ∂2 w L ∂w 2 ∂2 w 2∂ w − c = dx, (16.8) s ∂t2 ∂x2 2L ∂x2 0 ∂x in which c1 is the longitudinal wave speed. Given the boundary conditions w(0, t) = w(L, t) = 0, and the natural frequencies and mode shapes ωn = nπcs /L,

φn (x) = sin (nπx/L)

(16.9)

from Eqs. (6.15) and (6.16), we seek a solution by using an expansion of the linear modes: w(x, t) =

∞

n (t) sin (nπx/L).

(16.10)

n=1

Substituting Eq. (16.10) into Eq. (16.8), we get [14] ∞

¨ n + ω2n n = −

c21 n2 π4 n m2 2m. 4 4L

(16.11)

n=1

Assuming a single-mode response (with appropriate initial conditions), we set n = 1 to get ¨ 1 + ω21 1 +

c21 π4 3 = 0, 4L4 1

(16.12)

which is Duffing’s equation [15] with a hardening spring characteristic (because the cubic coefficient is positive).

18:14

P1: RTT Chapter˙16

CUFX159-Virgin

0 521 85648 5

August 14, 2007

320

Nonlinear Vibration

We can solve Eq. (16.12) by using elliptic integrals, but we can easily obtain an approximate analytic solution by using the method of harmonic balance. By assuming = A cos ωt, we find an amplitude-dependent frequency 2 2 π c1 3 ω2 = ω2n 1 + A2 , (16.13) 16 cs L and the anticipated stiffening effect of large-amplitude motion. Sagging cables can also exhibit nonlinear vibrations [16], including interesting nonplanar behavior [17].

16.5 Nonlinear Vibration of Beams In this section, we investigate the large-amplitude free oscillations of a clamped– clamped beam following the work of Yamaki [18, 19]. The beam behavior is described by the lateral deflection w(x, t), and the beam has an initial axial displacement U0 , such that the governing equation of motion is given by ∂4 w EA 1 L ∂w 2 ∂2 w ∂2 w EI 4 − dx + ρA = 0. (16.14) U0 + ∂x L 2 0 ∂x ∂x2 ∂t2 It is relatively easy to incorporate an initial geometric imperfection into the analysis, although this is not undertaken here. Again, it is convenient to put Eq. (16.14) in nondimensional form (akin to the procedure of Section 15.3), using w¯ = w/h, x¯ = x/L, and T = t EI/mL4 to give 1 w¯ − u0 + 6 w¯ 2 d¯x w¯ + w¨¯ = 0, (16.15) 0

where a prime denotes differentiation with respect to x¯ and an overdot with respect to T, with a natural frequency ω = mL4 /EI and u0 = (LA/I)U0 . Note that the displacement is normalized with respect to the thickness of the beam rather than the length as used to develop Eq. (15.21), and this form of equation was also encountered in Section 7.4 before the small-deflection approximation was applied. From Eq. (16.15), we see how the effective axial load depends on the lateral deflection [20–22]. Once the boundary conditions have been specified, the solution can be assumed to take the form w¯ = Ym(t)Wm(x), m = 1, 2, 3, . . . , (16.16) m

where we seek to solve for the unknown time functions Ym(t), and where Wm(x) are the solutions of the underlying linear-eigenvalue problem, that is, the orthonormal modes of vibration of the beam (which satisfy the geometric boundary conditions). Use is then made of Galerkin’s method to obtain a set of ordinary differential (Duffing-like) equations [23]. A few more details of this approach will be given later for the harmonically forced system.

18:14

P1: RTT Chapter˙16

CUFX159-Virgin

0 521 85648 5

August 14, 2007

16.5 Nonlinear Vibration of Beams

321

(wm)rms 2.0 Figure 16.7. The nonlinear fundamental natural frequency as a function of response amplitude. Static component not included.

1.0 u0 = 0 u0 = -60 u0 = 40

0

50

These equations can then be solved with an approximate analytical method (e.g., harmonic balance). For a single-mode analysis with small-amplitude oscillations we obtain the results discussed in Section 7.3. In Fig. 16.7 are shown some results for moderately large-amplitude oscillations. Here the response of the center of the clamped beam is plotted as an rms measure of a three-mode analysis that is due to Yamaki and Mori [18]. First consider the case with u0 = 0. For linear vibrations, we have the coefficient 22.37 [24, 25] for the fundamental natural frequency, shown as the solid curve. However, this increases to a value of 60 when the response magnitude (wm)rms is 2.4248, for example. When the initial end displacement is u0 = 40 (stretching), we obtain a shift to higher frequencies as expected and shown as the long-dashed curve. For a typical compressive end displacement (u0 = −60), we find some interesting behavior. In this case the beam is buckled and we see the appearance of two branches (both shown as dotted curves). The first branch is close to the origin and corresponds to oscillations about the postbuckled equilibrium position with the natural frequency corresponding to vanishingly small oscillations coincidentally close to the u0 = 0 case (see Fig. 7.6). However, there is also another branch and this gives finite frequencies starting from (Wm)rms ≈ 0.5. This corresponds to motion that traverses the remote equilibrium position [in the manner of Fig. 16.1(d)]. The participation of higher modes is also an important issue and will be discussed later when we deal with the nonlinear response of beams to harmonic excitation.

A Simplified Energy Approach. As we saw in earlier chapters, a useful means of

estimating the fundamental frequency of vibration by Rayleigh’s quotient can be based on the fact that the frequency corresponds to a stationary value in the neighborhood of a natural mode. Using an assumed displacement function, we can show that for an inexact eigenvector we get an eigenvalue that differs from the true value to the second order. This concept was introduced in Subsection 4.2.6 for discrete systems and in Subsection 4.3.2 for continuous systems in terms of the Rayleigh quotient, and we apply it here to a thin beam, including stretching effects [26].

18:14

P1: RTT Chapter˙16

CUFX159-Virgin

0 521 85648 5

August 14, 2007

322

Nonlinear Vibration

Even without an externally applied end shortening, as we have seen, if the ends of the member are constrained in the axial direction then membrane effects can occur for deflections that are not especially large. We have just seen how strings may have this effect, and membrane effects are also important in plates, which shall be considered a little later. We can take an approximate energy approach here for a simply supported thin beam of the type shown in Fig. 7.1 [27, 28]. The energy contributions are [29] L 1 T= m (w) ˙ 2 dx, (16.17) 2 0 L 1 (w )2 dx, (16.18) U = EI 2 0 L 1 2 1 (w ) dx, (16.19) VP = P 2 0 2 in which it can be shown that the induced axial load is given by L EI P= (w )2 dx, 2Lr2 0

(16.20)

where, for a beam of cross-sectional area A and radius of gyration r, we have I = Ar2 . Suppose the ends are fully pinned (i.e., immovable); then the lowest mode for the linear problem is simply a half-sine wave w = Q(t) sin (πx/L). We are primarily interested in the maximum amplitude of motion Qm. Evaluating the energy terms and adding them, we obtain the total energy constant C=

π4 EI 2 π4 EI Q + Q4 , m mL4 8mL4 r2 m

(16.21)

which can be used to obtain the phase trajectory as a function of the initial conditions. The equation of motion can be subjected to separation of variables and integrated (numerically) to obtain the natural period (and hence frequency) as a function of maximum amplitude. The frequency is normalized with respect to the linear natural frequency (ωL). The result is shown in Fig. 16.8 together with a Galerkin approach (the dashed curve [30, 31]) that yields 2 ω 3 Qm 2 =1+ . (16.22) ωL 16 r FEA can also be used to solve this type of problem [28].

16.6 Nonlinear Vibration of a Plate We return to consider the case in which the lateral deflections (and boundary conditions) of a plate are such that midplane stretching, or membrane response, occurs. This is somewhat similar in effect to externally applied axial loads, which will be considered subsequently. In general, these stretching effects depend on the in-plane

18:14

P1: RTT Chapter˙16

CUFX159-Virgin

0 521 85648 5

August 14, 2007

16.6 Nonlinear Vibration of a Plate

323

L

2.2

Galerkin dashed

2.0 1.8 1.6 1.4 1.2

1.0 0

1

2

3

4

Qm/r

Figure 16.8. The increase in natural frequency with amplitude for a simply supported strut that is constrained from moving axially at the ends.

boundary conditions but need not involve amplitudes that are considered large. No exact solutions are available, but we can again make use of an approximate approach based on assumed deflection shapes (i.e., a Galerkin approach). In general we seek solutions by using the form w(ξ, η, t) = cA(t) (x)(y),

(16.23)

where (x) and (y) are spatial mode shapes that satisfy the boundary conditions of the panel, and c is a constant. We will consider the example from Section 10.1, namely, the free-vibration behavior of a simply supported square panel, by using a single-mode approximation. Other boundary conditions will also be considered. An exact solution to the corresponding linear problem exists if two opposite sides are simply supported [32]. Importantly, we consider the case in which in-plane axial deformation is prevented along the edges. Chu and Herrmann [33] obtained a solution for the simply supported case, and Wah [3] extended their results (using a slightly simpler approach) to include other boundary conditions. Application of the Galerkin procedure leads to an ordinary differential equation of the form (very similar to the large-deflection equation for beams) d2 A c2 4 + µ A + 6 λA3 = 0, dζ2 h2

(16.24)

√ where ζ = [t/(ab)] D/ρ, µ4 = (ρ/D)a2 b2 p 2 , λ = η2 /ξ2 , where η and ξ come from the √ Airy stress function. Other plate characteristics include p = µ2 /(ab) D/ρ, the natural frequency of the linear problem, h, the plate thickness, and ρ, the mass per unit area. We recognize the equation of motion as a form of Duffing’s equation with a hardening spring characteristic. A solution to the period of the response can be obtained in terms of elliptic integrals, that is, T∗ 2K(k) = 1/2 , 2 T π 1 + 6 Ah2 µλ4

(16.25)

18:14

P1: RTT Chapter˙16

CUFX159-Virgin

0 521 85648 5

August 14, 2007

324

Nonlinear Vibration T */T 1 0.8 0.6 0.4 0.2

0

0.5

1

1.5

A/h

Figure 16.9. Effect of amplitude on the natural period of square plates: (a) continuous curve, SSSS-SS-SS; (b) dashed curve, SS-C-SS-SS; (c) dotted curve, SS-C-SS-C [3].

where K[k] is the complete elliptic integral of the first kind, and where k=

3A2 /h2 , 6A2 /h2 + µ4 /λ

(16.26)

and T is the period of the linear system, that is, T = 2π/µ2 . The values of µ4 /λ depend on the boundary conditions and three types are given here (following Wah [3]): case (a), simply supported on all four sides (SS-SS-SS-SS); case (b), simply supported on three sides and clamped on the fourth (SS-C-SS-SS); and case (c) simply supported on two opposite sides and clamped on the others (SS-C-SS-C). In each case, there is also a dependency on the aspect ratio but we restrict ourselves to square plates here (a/b = 1). Some results are shown in Fig. 16.9. We see that with these in-plane boundary conditions the period is reduced by approximately 25% when the deflections reach the thickness of the plate (and thus the frequency increases).

16.7 Nonlinear Vibration in Cylindrical Shells With reference to Fig. 10.14, we begin by looking at free vibrations under so-called shear diaphragm end conditions (but without end loading, i.e., Nx = 0). The equations of motion can be developed along similar lines to plate theory with appropriate extensions incorporating the additional geometrical effects of shells. A popular theory, mentioned in Chapter 10, is based on the Donnell–Mushtari equations [31]. Assuming a single mode of displacement in the radial direction, we have w(x, y, t) = Amn (t) cos

mπx ny mπx n2 2 sin + Amn (t) sin2 , R l 4R l

(16.27)

which, on substitution into the (partial differential) equation of motion and using a Galerkin procedure, results in the nonlinear ordinary differential equation 2 d2 ζ d2 ζ 3 dζ + ζ + ζ ζ 2 + (16.28) − γζ3 + 2 δζ5 = 0, dτ2 8 dτ dτ

18:14

P1: RTT Chapter˙16

CUFX159-Virgin

0 521 85648 5

August 14, 2007

16.8 Nonlinear Forced Vibration of Strings

1.8

325

0

1.4

1

1

2

3

_ A

4

0.6

Figure 16.10. The relation between the natural frequency and amplitude of vibration for a cylindrical shell. Solid curve, ξ = 0.5; dashed curve, ξ = 2.0 [34].

where ζ = Amn /h, τ = ωmn t, and = (n2 h/R)2 . The coefficents γ and δ are complicated geometrical expressions that depend on the aspect ratio, ξ = mπR/nl, of the mode and Poisson’s ratio, µ (see Leissa [31] for details). Evensen and Fulton [34] used the method of averaging to solve Eq. (16.28). The response is given by ζ(τ) = A¯ cos ω∗ τ,

(16.29)

in which the natural frequency depends on the amplitude in the following way: ω∗2 =

ω ωmn

=

1 − 34 γ A¯ 2 + 58 2 δA¯ 4 . 1 + 3 A¯ 2

(16.30)

16

The presence of cubic and quintic terms in Eq. (16.28) alerts us to the possibility of interesting behavior, and indeed, plotting Eq. (16.30) for a sample of geometries gives the results shown in Fig. 16.10. In this figure, the solid and dashed curves relate to ξ = 0.5 and ξ = 2.0, respectively, and we see that whether the nonlinear free vibrations can be classified as hardening or softening depends on and hence on geometry.

PART II: FORCED VIBRATION 16.8 Nonlinear Forced Vibration of Strings In this section, we add external excitation (and damping) and consider the nonlinear vibrations of a number of axially loaded systems. For small-amplitude oscillations the response is typically linear. This situation was covered in Chapter 15. However, in going from free to forced vibration, we are increasing the phase space from 2D to 3D, and then with nonlinear effects we encounter a host of (possibly very complicated) behavior. Again, the steady state is of particular interest but we shall also see that the long-term dynamic behavior may no longer be independent of initial conditions. We shall start by looking at a stretched string before moving on to consider the nonlinear forced vibration of beams.

18:14

P1: RTT Chapter˙16

CUFX159-Virgin

0 521 85648 5

August 14, 2007

326

Nonlinear Vibration

If we subject the mass-suspended model from Section 16.3, that is, a single-mode approximation of the stretched string, to harmonic excitation then the equation of motion becomes m¨x + cx˙ + 2

kd 4k(L − d) 3 x = F (t), x+ L L3

(16.31)

in which c is a linear-viscous damping coefficient. Using the notation of Tufillaro et al. [11], we can also write Eq. (16.31) as a special case of r¨ + λ˙r + ω20 (1 + Kr2 )r = f(t),

(16.32)

2(L − d) . L2 d

(16.33)

where ω20 =

2kd , mL

K=

Here, r is the radial distance from the origin, and thus r2 = x2 + y2 , in which the motion in the y direction allows for the possibility of whirling motion [35]. The (unidirectional) forcing is f(t) = [A cos(ωt), 0] and in relation to the continuous string we see by analogy with Eq. (16.12) that ω0 =

cs π , L

K=

π2 , 4L

(16.34)

where is the longitudinal extension of the string and (cs /c1 )2 = /L is assumed to be small. Finally, we scale time and transverse deflection according to τ = ω0 t and s = r/(L − d), and, assuming planar motion [s(t) = x(t)], we arrive at the forced Duffing equation: x + αx + (1 + βx2 )x = G(γ, τ),

(16.35)

in which α = λ/ω0 , β = 2(L − d)3 /(dL2 ), G = Lf /[2kd(L − d)], γ = ω/ω0 , and f is a harmonic forcing term. Again an approximate solution to this nonlinear ordinary differential equation can be obtained by a number of methods, but we shall leave consideration of specific responses until the next section when we look at the class of axially loaded structures associated with beams. It is also worth mentioning here that for typical nonlinear string vibrations the response often involves the onset of nonplanar (or whirling) motion. In this case, the problem can be solved by use of two coupled Duffing oscillators, with the suspended-mass analogy provided by a second DOF (in a direction into the page in Fig. 16.4 [6, 11, 35]).

16.9 Nonlinear Forced Vibration of Beams Returning to the response of a beam for which the ends are held a fixed distance apart, we now add the effect of an external (harmonic), lateral excitation, and add a small amount of linear-viscous damping. In this case, the terms −P cos t + C∂w/∂t are added to the left-hand side of Eq. (16.14), which corresponds to −p cos ωt +

18:14

P1: RTT Chapter˙16

CUFX159-Virgin

0 521 85648 5

August 14, 2007

16.9 Nonlinear Forced Vibration of Beams

327

(a)

|A|

(b)

|A|

c = 0.0

1.5

2.0 1.5

1.0

c = 0.4

1.0

0.5

F = 2.0 F = 1.0 F = 0.5

0.5

c=1 0.0

0.5

1.0

1.5

2.0

0

0.5

1.0

1.5

2.0

2.5

Figure 16.11. Resonance response curve based on a harmonic balance solution of Duffing’s equation with α = β = F = 1: (a) Various damping levels for fixed forcing magnitude (F = 1), (b) various forcing magnitudes for a fixed level of damping (c = 0.2).

c∂w/∂ ¯ t¯ in nondimensional terms, in which p = PL4 /(EIh) = 12PL4 /(Ebh4 ) and c = √ CL2 / mEI, and a rectangular beam of cross section bh has been assumed. Solutions to the governing equation of motion are obtained in the usual way, but before looking at the detailed response we consider a simplified single-mode analysis. Using the expansion from Eq. (16.16) with m = 1, we again obtain the forced form of Duffing’s equation, x¨ + cx˙ + αx + βx3 = F sin ωt,

(16.36)

in which α, β, and so on, depend on the various physical parameters of the beam [23]. This has the same form as Eq. (16.35), and we now consider its steady-state solution, with special attention paid to resonance phenomena. The method of harmonic balance can again be used, based on the assumed solution x = A cos ωt, to give an approximate relation between the various parameters:

3 3 2 2 (α − ω )A + βA + c2 A2 ω2 = F 2 . (16.37) 4 In Eq. (16.37) for the unforced (F = 0), undamped (c = 0) case, we retain the hardening backbone curve (i.e., ω ≈ 1 + (3/8)A2 + · · · +), obtained earlier in this chapter for free vibration (assuming both α and β are positive unity). Furthermore, with β = 0 we get the linear resonance response curve described in Fig. 15.1. Some typical nonlinear response curves are shown in Fig. 16.11. Here, in part (a), the dashed curve corresponds to the free- (undamped) vibration case. The solid curves correspond to different levels of damping (for a fixed level of forcing). With c = 0 we get the outer curve. When the damping is increased to a level of c = 0.4, we obtain a response that has a maximum amplitude close to A = 1.5 that occurs in the vicinity of ω = 1.6. For the more heavily damped case (c = 1), the response hardly exhibits resonance at all and is similar to the linear response. This can still be considered underdamped (ζ = 0.5) according to the linear description from Section 3.1.

18:14

P1: RTT Chapter˙16

CUFX159-Virgin

0 521 85648 5

August 14, 2007

328

Nonlinear Vibration (a)

|A|

(b) Response

3

2

1

0

Forcing Amplitude

0.5

1.0

1.5

2.0

Forcing Frequency

2.5

Figure 16.12. (a) Resonance response indicating stability, α = β = F = 1, c = 0.2 and (b) the folded surface.

However, the lightly damped (large-amplitude) behavior is generally quite different from the linear response. Consider Fig. 16.12(b) in which c = 0.2, and take F = 1. We note that the response does not scale linearly with forcing magnitude (for a fixed level of damping). Suppose the forcing frequency were very slowly increased from an initially low value. The response would gradually grow to a maximum (close to ω ≈ 2.2) but then suddenly drop to the lower-amplitude branch. On subsequent slow reduction in the forcing frequency, the response would again slowly grow but with a sudden jump up (close to ω ≈ 1.6) before following the original path. Over the range 1.6 < ω < 2.2, we have two stable solutions (the curve between is unstable). Thus there is a certain dependence on initial conditions and hence path-dependent behavior. Another kind of hysteresis was encountered earlier in this book with regard to snap-through buckling. The behavior here is also characterized by sudden dynamic jumps in the response, although we now have instabilities during the evolution of oscillatory behavior rather than equilibria. However, the instability phenomena are basically the same—a saddle-node bifurcation. The link between them is in going from a time-continuous system to a time-discrete system, as discussed in Section 3.4. The hysteresis can be envisioned as passing along a folded surface [8], as shown in Fig. 16.12. The points of instability correspond to vertical tangencies in the response curve and can thus be obtained from dω/dA = 0 to give 3 2 9 2 2 2 α − ω + βA α − ω + βA + c2 ω2 = 0. (16.38) 4 4 The roots of this equation define the region of multiple solutions [shown shaded in Fig. 16.12(b)] with one of the curves coinciding with the backbone curve when there is no damping. In the region of hysteresis, the relative dominance of the two possible stable steady-state oscillations is reflected in their domains of attraction, that is, which initial conditions lead to which solution. In some ways, this is similar to a standard application of Newton’s method as successive estimates of a root iterate toward a

18:14

P1: RTT Chapter˙16

CUFX159-Virgin

0 521 85648 5

August 14, 2007

16.9 Nonlinear Forced Vibration of Beams

329

converged value that depends to an extent on the initial guess. To determine the relative dominance of the two stable solutions, it would be necessary to divide the space of initial position and initial velocity into a fine grid and then numerically solve the governing equation from these starting conditions, subsequently labeling the final outcome. Outside the region of hysteresis the final outcome does not depend on the initial conditions. We shall return to this issue a little later, and also note that there are a variety of other ways in which nonlinear dynamical systems can lose stability, especially when the nonlinearity is of the softening variety [8]. And this is where perturbations of the steady-state response lead to the variational equational (with periodic coefficients) and use of Floquet theory (see Jordan and Smith [4] for more details). Having seen how lateral (bending) and axial (stretching) behavior, based on a single-mode analysis of the governing equation, are characterized by a bending over of the main resonance curve, consider the results of a multi-mode analysis of a clamped beam based on the work of Yamaki and Mori [18], in which they solved 1

2 2 (w¯ − w¯ 0 ) − u0 + 6 (16.39) w¯ − w¯ 0 d¯xx w¯ + w¨¯ = f¯ cos ωt, 0

where an initial imperfection w¯ 0 was included, although we just consider the perfect geometry here. Assuming a separable solution w¯ = m Ym(t)Wm(x) and imposing clamped boundary conditions, they arrived at a set of coupled nonlinear Duffing oscillators: Y¨n + ω2n Yn − u0 βmn Ym + 6 βkl βmn YkYl Ym m

= γn f¯ cos ωt,

k

l

m

k, l, m, n = 1, 2, 3, . . . ,

(16.40)

where the β terms are obtained from orthogonality. For a three-mode analysis for symmetric vibrations (which are most accessible experimentally), this reduces to Y¨n + ω2n Yn − u0 (β1n Y1 + β3n Y3 + β5n Y5 ) + 6 βkl βmn YkYl Ym k=1,3,5 l=1,3,5 m=1,3,5

= γn f¯ cos ωt,

n = 1, 3, 5.

(16.41)

Finally, assuming the solutions are given by Ym(t) =

3

dmj µ cos (j µωt),

µ = 1, 1/2, 1/3,

(16.42)

j =0

the method of harmonic balance can again be applied, and the results represented by the rms value of the response at the center of the beam:

(wm)rms

⎧ 2 2 2 ⎫1/2 ⎬ 1 1 1 1 ⎨ 1 2 3 = √ dmφm dm φm dm φm , ⎭ 2 2 2 2⎩ m m m

m = 1, 3, 5. (16.43)

18:14

P1: RTT Chapter˙16

CUFX159-Virgin

0 521 85648 5

330

August 14, 2007

Nonlinear Vibration

This was the approach used to determine the free-vibration curves shown in Fig. 16.7 in which p¯ = 0. Determinig the solution involves using Newton’s method to solve the 12 simultaneous cubic equations. Figure 16.13 shows typical responses when the beam is not initially subject to any axial displacement (u0 = 0) but rather axial effects are induced because of midplane stretching. The forcing magnitude is set at f¯ = 200. We see the complicated nature of the response, including various subharmonic and superharmonic waveforms. Again, regions of unstable behavior, resonant jumps, and hysteresis are apparent. In Yamaki and Mori [18], equivalent results are also shown for the buckled beam (e.g., u0 = −60), which brings into play the frequencies associated with postbuckled equilibria (see Fig. 16.7) as well as snap-through (which will be considered in the next section with regard to a panel). A companion paper [19] shows remarkable experimental confirmation of this type of behavior based on the excitation of a small duralumin test specimen in which the test frame was subjected to a constant peak acceleration with a measurement of relative displacement. Some intriguing applications of this theory at very small scales are also mentioned [36].

16.10 Persistent Snap-Through Behavior in a Plate Next, we return to the thermally loaded plate first considered in Section 10.5 in terms of free vibration and also subject to relatively mild harmonic (narrowband acoustic) excitation in Section 15.5. Now suppose that the magnitude of excitation is increased (from 130 to 155 dB). In this case sufficient energy is applied to the system so that the plate can exhibit snap-through behavior, i.e., both stable equilibria are traversed (see Fig. 10.6). Figure 16.14 shows a typical response (in terms of a strain measurement), with the temperature change held fixed at 32 ◦ F above ambient. At this post-critical level of thermal loading the lateral deflection at the center of the panel is of the order of the panel thickness. The excitation magnitude is 155 dB and is narrowly focused in the vicinity of 115 Hz. Figure 16.14(a) shows an experimental time series of strain-gauge data taken from a point on the panel (see the photo in Fig. 10.9) at (ξ, η) = (0.583, 0.416), where ξ = x/a, η = y/b. Note the contrast with Fig. 15.15. Part (b) again shows a time-lag embedded phase projection that also contrasts with the closed orbit corresponding to periodic motion, and the frequency spectrum in part (c) is decidedly broadband. This type of intermittent snap-through behavior can also be observed in the vibrations of a buckled beam and has clear implications for fatigue [37]. Figure 16.15 shows a time series in terms of lateral deflection but based on a nine- (cosine) mode Galerkin analysis for nominally the same parameter values as those of the experimental data. More details of this analysis can be found in Murphy [12]. Another way of encountering snap-through is through a slow sweep of a system parameter. This was a technique used earlier in this book to allow the evolution of a buckling instability. Figure 16.16 shows two examples of nonstationary snap-through

18:14

P1: RTT Chapter˙16

CUFX159-Virgin

0 521 85648 5

August 14, 2007

16.10 Persistent Snap-Through Behavior in a Plate

Figure 16.13. Frequency-response characteristics for a clamped beam including typical waveforms. Reproduced with permission from Elsevier [18].

behavior where the sound pressure level (SPL) is gradually ramped up and then down to produce a nonstationary transition through large-amplitude behavior. The f parameters used to generate these experimental results correspond to T/cr = 1.76 with a baseline forcing of 130 dB at 120 Hz. In both parts, the SPL is ramped

331

18:14

P1: RTT CUFX159-Virgin

0 521 85648 5

August 14, 2007

332

Nonlinear Vibration

yy (t +T/4)

500

0

-500 0

0.05 t (s)

0.1

(b)

10 3

(c)

10 0

PSD

(a)

500

yy

Chapter˙16

0

10 -3

-500 -500

0

500

10 -6

0

yy (t)

500

1000

(Hz)

Figure 16.14. Large-amplitude snap-through behavior: (a) strain time series, (b) time-lag embedded phase projection, and (c) frequency spectrum.

from 130 dB up to 150 dB and then back down to 130 dB, at roughly the same rate. In part (a), the motion is initiated as a small-amplitude periodic oscillation that grows very gradually until it is sufficiently large that the motion escapes the confines of its local potential-energy well and snap-through occurs, that is, an erratic snapping organized around the two stable equilibria. The motion then reduces in amplitude as the SPL is ramped down and settles to small-amplitude motion about the other equilibrium configuration (in the adjacent potential-energy well). In part (b) the system is initiated with nominally the same conditions. After the burst of cross-well motion, the system, on subsequent reduction in the applied SPL, settles back to small-amplitude motion about the original equilibrium. In Section 13.4, and specifically in Fig. 13.8, it was shown how the strength (in terms of forcing amplitude and frequency) of harmonic excitation could lead to a transient resonantlike condition in which the motion might escape the confines of its local potential-energy well. In essence, this is the situation with the snap-through behavior shown in Fig. 16.16. Fixing the (postcritical) temperature at T/Tcr thus determines the fundamental natural frequency. For a given frequency of excitation, the SPL is then quasistatically increased until the first occurrence of snap-through behavior occurs, that is, the system goes beyond the potential-energy hilltop associated with the unstable (almost flat) equilibrium configuration. The frequency is then incremented, and the procedure is repeated. Figure 16.17 shows a summary of 30 such runs plotted against nondimensional frequency, clearly separating regions of parameter space into snapthrough and no-snap-through regions [38].

w/h

2 1 0 -1 -2

0

0.05

0.1

0.15

0.2

0.25

t 0.3

Figure 16.15. A numerical simulation based on using nine (cosine) modes [12].

18:14

P1: RTT Chapter˙16

CUFX159-Virgin

0 521 85648 5

August 14, 2007

16.10 Persistent Snap-Through Behavior in a Plate 500 400

333

500

(a)

(b) 400

300

0 3 0

200

200

100

100

0

0

−100

−100

−200

−200

−300

−300

−400

−400

−500 0

1

2

3

4

5

6

t (s)

7

−500 0

1

2

3

4

5

6

t (s)

Figure 16.16. Evolving time series showing transient snap-through caused by sweeping the SPL from 130 dB → 150 dB → 130 dB. (a) The motion was initiated around the secondary equilibrium but then settled around the primary equilibrium. (b) The motion was initiated around the secondary equilibrium and returned there after a burst of transient snap-through.

The similarity between Figs. 16.17 and 13.8, at least in terms of the boundary between escape (snap-through) and no escape (no snap-through) is noted. Although there are many details associated with this behavior, we can envision this situation as the growth of a softening resonance effect that encounters instability and subsequent transition beyond the locally bounded area of phase space.

SPL (dB)

154 Snap-Through 152 150 148 146 144 No Snap-Through 142 140

0.75

0.8

0.85

0.9

0.95 n

Figure 16.17. Snap-through boundary plotted in parameter space using tonal inputs T/Tcr = 1.95, ωn ≈ 111 Hz at this temperature.

7

8

18:14

P1: RTT Chapter˙16

CUFX159-Virgin

0 521 85648 5

August 14, 2007

334

Nonlinear Vibration

16.11 A Panel in Supersonic Flow Some axially loaded slender structures are also subject to nonconservative forces, e.g., in fluid–structure interaction, or Beck’s problem (see Section 7.8), and some of this behavior can lead to nonlinear oscillations [39, 40]. We shall look at a typical situation in which axial load and fluid loading conspire to produce some complex dynamic behavior. This will also provide an example of chaotic oscillations that are not generated by periodic forcing [41]. Only a simplified analysis will be conducted in which a number of assumptions are made in order to ease the analysis [42, 43]. Consider the thin elastic simply supported panel shown in Fig. 16.18. It has a length L, is subject to a constant axial load Nx , and has flow velocity U that produces a dynamic pressure p. We assume that there is no structural damping (although some energy dissipation will be introduced by the aerodynamic modeling). If we assume that the width of this panel (into the page) is infinite then we can use a 2D form of Von Karman’s plate theory [39] to describe lateral deflections w(x, t) by means of solutions of ∂2 w ∂2 w Eαh a ∂w 2 ∂4 w ρmh 2 − Nx + dx + D + pˆ = 0. (16.44) ∂t 2a 0 ∂x ∂x2 ∂x4 The parameters used are panel thickness h, Young’s modulus E, material density ρm, in-plane stiffness α, D is the same as in Chapter 10, that is, D = Eh3 /(12(1 − ν2 ), and pˆ is the aerodynamic pressure loading. The spatial parameters and time can be nondimensionalized following the procedure of Section 7.2, that is, D w¯ = w/h, x¯ = x/L, τ = t , (16.45) ρmhL4 such that Eq. (16.44) can be written as 1 2 2 ∂ w¯ ∂2 w¯ ∂w¯ ∂4 w¯ 2 − R + 6α(1 − ν ) d¯ x + + Pˆ = 0, x ∂τ2 ∂x¯ ∂x¯ 2 ∂x¯ 4 0

(16.46)

where we now have scaled axial load Rx = Nx a2 /D and scaled dynamic pressure that can be obtained from piston theory for supersonic flow speeds, that is, Mach number M 1. This is a linear aerodynamic theory [39, 42], and in nondimensional terms, U x P ED w

Figure 16.18. A slender panel subject to both axial loading and supersonic flow over one face.

18:14

P1: RTT Chapter˙16

CUFX159-Virgin

0 521 85648 5

August 14, 2007

16.11 A Panel in Supersonic Flow

335

causes a pressure on the panel of the form

µ ∂w¯ ∂w¯ + , Pˆ = λ ∂x¯ Mλ ∂τ

(16.47)

where µ and λ are nondimensional flow density and flow velocity, respectively. We next conduct a standard Galerkin procedure and assume a series expansion for the motion w(τ, ¯ x) ¯ =

N

qm(τ) sin mπx. ¯

(16.48)

m

Placing this expression back in Eq. (16.46), we obtain a set of N coupled ordinary differential equations: N ∂2 qm + (mπ)2 qm Rx + (mπ)2 + 3α(1 − ν2 ) (iπ)2 q2i + 2Pˆ m = 0, (16.49) ∂τ2 i in which

1

Pˆ m =

Pˆ sin mπx¯ d¯x

(16.50)

0

=λ

N i

qi

mi m2 − i2

[1 − (−1)m+i ] +

µλ dqm . M dτ

(16.51)

Previous studies on this system have indicated that assuming two terms in the solution [Eq. (16.48)] gives the correct qualitative results, although six or eight terms should be retained for accurate quantitative results [39]. It is interesting to note that these equations may be numerically “stiff,” which requires special care in their solution [44]. In relating this analysis to a more practical setting it may be necessary to incorporate some in-plane stiffness at the boundaries [45]. Taking the first two harmonics leads to the coupled equations

q¨ 1 + U˙q1 − U2 q2 + q1 (1 − P) + 4q1 q21 + 4q22 = 0, (16.52)

q¨ 2 + U˙q2 + U2 q1 + 4q2 (4 − P) + 16q2 q21 + 4q22 = 0, where some additional nondimensionalization has been undertaken, and we focus on the role of two nondimensional parameters: a flow velocity U and an axial load P. In general these equations can be solved only numerically, and we might anticipate quite complicated solutions depending on the parameters. Before conducting some numerical experiments, it is instructive to consider the loss of stability from the trivial (flat) equilibrium configuration. In this case, if the axial loading is gradually introduced we encounter our familiar transition to buckling. Given the scaling in Eqs. (16.52) we see the Euler load of P = 1 that corresponds to Rx = −π2 . But a loss of stability can also be induced by the fluid loading through increasing the parameter U. Suppose we have no axial loading, and we restrict ourselves to relatively small-amplitude oscillations. In this (linear) case we can assume that the solution consists of exponential terms of the form r = Aeλt , which leads

18:14

P1: RTT Chapter˙16

CUFX159-Virgin

0 521 85648 5

August 14, 2007

336

Nonlinear Vibration

directly to the state matrix ⎤ 0 U2 0 0 ⎥ ⎥ ⎥ [r], −U −16⎦ 1 0

⎡

−U −1 ⎢ 1 0 ⎢ [˙r] = ⎢ −U2 ⎣ 0 0 0

(16.53)

in which r = (dq1 /dt, q1 , dq2 /dt, q2 )T , the eigenvalues of which determine the form of the motion. Specifically, we have the possibility that positive eigenvalues lead to a local growth of motion and hence instability. In general, in this book, the monitoring of an eigenvalue has typically been associated with real eigenvalues (which is proportional to the square of the frequency) and their sign changing from negative to positive. However, we now consider another route to instability for the case in which the system eigenvalues are complex, and the imaginary part is associated with frequency. Setting U = 0 in Eq. (16.53) leads to purely imaginary eigenvalues in this state-variable format. With Euler identities this means oscillatory motion (which is what we would expect for a plate without external forcing or damping—recall that with our simplified modeling the aerodynamics is the only source of energy dissipation). However, as the flow rate (described by U) increases, the system eigenvalues change and we can track their movement in the complex plane as shown in Fig. 16.19(a). Part (b) shows the variation of the real part of the system eigenvalues with changing U, and instability occurs when the critical value Uc = 3.59 is reached. This is a Hopf bifurcation (and quite different from Beck’s problem) and is the other way in which a system under the action of a single control parameter might generically lose its stability (along with the saddle-node). In aeroelasticity, this phenomenon is known as flutter (and the static instability that is due to the axial load is often called divergence). Both of these instabilities can also be traced back to the root structure of the linear oscillator (Fig. 3.4), and thus we identify the Hopf bifurcation with the onset of negative damping, in addition to the loss of stiffness encountered throughout this book. 5 Im

(a)

1 Re(λ) 0

(b)

−1

0

Uc = 3.59

−2

−5

−3

−3

−2

−1

0

Re 1

0

1

2

3

4

U

5

Figure 16.19. (a) A root locus of the state eigenvalues, (b) the real part of the eigenvalues plotted as a function of the control parameter U.

18:14

P1: RTT Chapter˙16

CUFX159-Virgin

0 521 85648 5

August 14, 2007

16.12 Chaotic Behavior

When the two control parameters operate together on this moderately highorder system we can expect a myriad of types of response. The following presents a summary of typical behavior. In Fig. 16.20 a variety of time-series responses are shown for the panel with no axial load (on the left) together with a similar set when a constant axial load of twice the buckling load is present, that is, P = 2. In all these cases, the inital conditions consist of a half unit in each of the two modal amplitudes but with zero initial velocity. As the flow velocity is increased, we have increased damping and the initial condition leads to a stable equilibrium for the flat panel. The damping increases with flow rate up to a certain extent and then starts to decrease. When U = 4.2, the flow speed is greater than the critical value corresponding to Fig. 16.19, and we observe the appearance of a limit cycle oscillation (LCO). The amplitude of this periodic motion grows with flow rate and follows a classic (nonlinear) Hopf bifurcation scenario [46]. The right-hand column indicates roughly similar behavior but now the stable equilibrium has shifted to a nontrival (buckled) value. However, for a certain range of flow rates (including U = 2.2), the panel returns to a stable equilibrium at the origin, that is, the flat panel. Subsequently, flutter occurs again, but this time at a lower flow rate than was the case for the unloaded panel. Figure 16.21 shows a more complicated periodic response when the axial load is P = 5 and the flow rate is U = 1.85. Here the periodic, but non-simple, trajectory seems to dwell in the vicinity of the positive and negative underlying equilibria. The phase projection in part (b) shows an alternative view. This is a relatively high-order (phase-space) system (depending on how many modes we retain in the expansion), and thus we should not be surprised to learn that chaos is not uncommon in this particular system. It is instructive to summarize these responses in term of the control parameters P and U. Figure 16.22 shows how the response of the panel depends on the combination of control parameters. We can thus locate the specific responses illustrated from Figs. 16.20 and 16.21 as the black data points, as well as the critical flow rate when no axial load is acting. It is interesting to note that it is possible for the panel to be flat and stable for certain flow rates even though the panel is buckled and stable for lower flow rates. To the left of the wide-dashed gray curve in the LCO region is where nonsimple oscillations may typically occur. The figure is the result of many numerical simulations but with the same initial conditions, and different responses are possible depending on the choice of initial conditions. Previous studies have included the effect of a static pressure differential across the panel (e.g., the top surface of an airplane wing), and a distributed stiffness in the form of a supporting elastic foundation [39].

16.12 Chaotic Behavior In a number of situations in this final chapter, we have observed behavior that appears to be erratic in nature, for example, Figs. 16.6(g) and 16.14. Both of these randomlike responses are characterized by broadband power spectra, that is,

337

18:14

P1: RTT Chapter˙16

CUFX159-Virgin

0 521 85648 5

August 14, 2007

338

Nonlinear Vibration P=0

P=2

w

w

0.5

0.5

0.0

10

20

30

40

U = 0.2

0.0

w

w

0.5

0.5

10

20

30

40

U = 1.2

0.0

−0.5

−0.5

w

w

0.5

0.5

0.0

10

20

30

40

U = 2.2

w

10

20

30

40

0.0

10

20

30

40

10

20

30

40

0.5

10

20

30

40

U = 3.2

−0.5

0.0 −0.5

w

_

0.5

−0.5

40

w

0.5

0.0

30

−0.5

−0.5

0.0

20

−0.5

−0.5

0.0

10

10

20

30

40

U = 4.2

x = 0.75 q1(0) = 0.5 q1(0) = 0.0 q2(0) = 0.5 q (0) = 0.0

.

.

2

Figure 16.20. A summary of transient responses of the infinite panel with different combinations of the control parameters U and P.

18:14

P1: RTT Chapter˙16

CUFX159-Virgin

0 521 85648 5

August 14, 2007

16.12 Chaotic Behavior

339 (a)

w

(b)

w 1

1

0.5

0.5 60

70

80

90

−1

100

−0.5

−0.5

0.5

1

−0.5

−1

-1

Figure 16.21. A nonsimple periodic response when P = 5 and U = 1.85.

motion in which a multitude of frequencies actively participate. However, although a similar response might be obtained from a very high-order system or one in which noise were present, these types of responses are examples of low-order deterministic chaos. This is a feature of nonlinear dynamical systems that has received considerable attention over recent times, and we finish this final chapter by examining an abstract model of an axially loaded structure that exhibits chaotic behavior. We start by recalling the forced form of Duffing’s equation: x¨ + cx˙ + αx + βx3 = F sin ωt.

(16.54)

Depending on the parameters, this nonlinear oscillator with its 3D phase space, that is, requiring the position, velocity, and forcing phase to uniquely determine the solution, is capable of exhibiting a vast array of behavior. Earlier in this chapter we saw how a region of hysteresis was possible, including a dependence on initial conditions. We now look at some typical responses of Eq. (16.54) in which the forcing is such that chaos occurs. Suppose we set the parameters as α = β = 1, c = 0.3, ω = 1.2, and F = 0.5. Numerically integrating Eq. (16.54) leads to the results shown in Fig. 16.23. The time

P 5

Buckled, but dynamically stable

4

Limit cycle oscillations (LCO)

3 2 1

Flat and stable

0 0

1

2

3

4

U

Figure 16.22. Dependence of the panel response as a function of axial load P and flow rate U.

w

18:14

P1: RTT Chapter˙16

CUFX159-Virgin

0 521 85648 5

August 14, 2007

340

Nonlinear Vibration

x

(a)

(b)

1.5

x

1

1 0.5

.

0.5

x

t 0 −0.5 −1

20

40

60

80

100

−1

−0.5

0.5

1

−0.5 −1 −1.5

Figure 16.23. A typical chaotic response, c = 0.3, ω = 1.2, F = 0.5: (a) time series; (b) phase projection.

series in part (a) shows a randomlike hopping around and between the equilibria (at ±1). Clearly, this behavior bears a strong similarity to the intermittent (postbuckled) snap-through motion shown in Fig. 16.15. Again, a convenient alternative form for displaying this response is the phase projection (velocity versus position), and this is shown for the same data in part (b). However, despite the apparent randomness of these responses, they are deterministic and there is a good degree of order underlying this behavior. A useful technique (especially for SDOF oscillators) is the Poincare´ section. This was introduced in Section 15.2 in terms of an analytical expression for a linear forced oscillator. This can still be obtained for nonlinear dynamical systems but not typically in closed form. If a time series is sampled at intervals of the forcing period, as shown in Fig. 16.24, then a periodic orbit would penetrate this section at the same location [47]. In this illustration, the periodic orbit repeats itself very two forcing cycles. However, if the response is chaotic then an interesting sequence of points is mapped because of the folding and stretching evolution of the chaotic attractor. This is persistent behavior and not associated with any initial transients that may be present. It is also in stark contrast to a damped unforced oscillator in which equilibrium represents a point attractor, or a damped forced oscillator in which a steady-state motion is represented by a periodic attractor. An example of a chaotic attractor is shown in Fig. 16.25 for the same motion as in Fig. 16.23. This fine structure shows some fractal characteristics and displays an extreme sensitivity to initial conditions (about 10,000 points after transients have been allowed to die out are plotted). Many numerical tools have been developed to shed light on chaos. We have already noted the broadband nature of the frequency spectrum [48, 49], but the complex geometry of the attractor can be described in terms of dimension [50] (and this is where certain fractal features are apparent). Initial conditions do not affect the qualitative nature of behavior in the vicinity of isolated point and periodic attractors, but we saw earlier how hysteresis allowed for some dependence of the final outcome on initial conditions. For a chaotic response, this dependence is extreme. There is a local exponential divergence of adjacent points on a trajectory and this feature is described by a positive Lyapunov

18:14

P1: RTT Chapter˙16

CUFX159-Virgin

0 521 85648 5

August 14, 2007

16.12 Chaotic Behavior

341

y = dx/dt T = period

T = period

x B

B

B

t A

A

A y

section n + 1

section n

section n + 2 B x

Steady states Tranisents

A

Figure 16.24. Schematic of a stroboscopic trajectory sampling: the Poincare´ section. Reproduced with permission from Thompson and Stewart [47].

exponent (LE). The LEs are related to the characteristic eigenvalues from Chapter 3 [51]. There are other measures, including the autocorrelation function [8]. The reader is referred to specialized texts for more details [47, 52]. The sensitivity to initial conditions is illustrated in Fig. 16.26. Here, Duffing’s equation, with the parameters set for periodic motion, is numerically integrated by a fine grid of initial conditions, and the black and white regions correspond to those initial conditions (basins of attraction) that lead to periodic motion about the +1

xp

Figure 16.25. A chaotic attractor based on a numerical simulation of Duffing’s equation. c = 0.3, α = −1, β = 1, ω = 1.2, F = 0.5.

xp

18:14

P1: RTT Chapter˙16

CUFX159-Virgin

0 521 85648 5

August 14, 2007

342

Nonlinear Vibration 2 . 1.5 x(0) 1 0.5 0 -0.5 -1 -1.5 -2 -2

-1.5

-1

-0.5

0

0.5

1

1.5 x(0)

2

Figure 16.26. Fractal basin boundaries based on the numerical simulation of Duffing’s equation from a fine grid of initial conditions. c = 0.168, α = −0.5, β = 0.5, ω = 1, F = 0.15.

and −1 equilibrium positions, respectively [53–55]. Hence there is a degree of uncertainty about the exact location of an initial condition; it may be very difficult to say which of the possible steady states the transient will be attracted to. The fractal nature of these basin boundaries remains no matter how fine the grid, and of course, in an experimental context there is always a degree of imprecision. Thus we see that sensitivity to initial conditions in terms of basin boundaries may occur even when steady-state chaos is not present, as only periodic solutions are present in Fig. 16.26. There are some other universal features of chaos that have made its study fascinating. Many nonlinear structures can exhibit chaos, including, for example, the shallow arch [56, 57], although it should be mentioned that most practical designs would not typically encounter this type of thoroughly nonlinear behavior. Often the broad characteristics of chaotic attractors are quite similar. For example, Fig. 16.27 shows a typical Poincare´ section taken from the model of the forced suspended mass from Section 16.3 (with d = 0), and thus we have a purely cubic oscillator that can be thought of as analogous to a laterally excited strut in which an axial load is maintained at its critical buckling value. It should be pointed out that to induce chaos in this particular case the forcing needs to be relatively large. Also, chaos can often occur after a sequence of period-doubling bifurcations [58] or can be manifest in other standard sequences including intermittency and quasi-periodicity [59]. Clearly, the study of chaos relies heavily on numerical simulation (and graphics), but some progress has been made analytically, for example, in the development of Melnikov theory to predict the onset of strange attractors, in which use is made of perturbation theory [60–62]. Finally, some experimental evidence for chaos in axially loaded structures is described. Figure 16.28 is a Poincare´ section based on the response of a magnetoelastic thin strip (a buckled beam) [63], which can be modeled by Duffing’s equation [64, 65]. The double-well shape of the underlying potential energy function is a form we have come to know well in this book.

18:14

P1: RTT Chapter˙16

CUFX159-Virgin

0 521 85648 5

August 14, 2007

16.12 Chaotic Behavior

343

. xp

xp

Figure 16.27. A Poincare´ section from the suspended mass problem, d = 0.

Epilogue We have now covered all the scenarios laid out in the point of departure at the start of this book. Vibration and buckling play out in a variety of interesting ways, ranging from linear free vibrations of axially loaded rigid-link models all the way to large-amplitude forced vibrations of axially loaded continuous systems. On this journey, we have come to rely on approximate techniques or numerical simulation as access to exact solutions has become limited. However, the dynamic behavior of structures in which there is a degree of axial loading is important and has widespread

Figure 16.28. A Poincare´ section taken from an experimental nonlinear beam. Reproduced with the permission from Elsevier [63].

18:14

P1: RTT Chapter˙16

CUFX159-Virgin

0 521 85648 5

344

August 14, 2007

Nonlinear Vibration

application. The quest for lighter aerospace structures is just one example of how the types of problem discussed in this book have an increasingly key role to play. References [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18]

[19]

[20] [21] [22] [23]

P. Hagedorn. Nonlinear Oscillations. Clarendon, 1981. N. Krylov and N. Bogoliubov. Introduction to Non-Linear Mechanics. Princeton University Press, 1949. T. Wah. Large amplitude flexural vibration of rectangular plates. International Journal of Mechanical Sciences, 5:425–38, 1963. D.W. Jordan and P. Smith. Nonlinear Ordinary Differential Equations. Oxford University Press, 1999. C. Hayashi. Nonlinear Oscillations in Physical Systems. Princeton University Press, 1964. A.H. Nayfeh and D.T. Mook. Nonlinear Oscillations. Wiley, 1979. J. Kevorkian and J.D. Cole. Perturbation Methods in Applied Mathematics. SpringerVerlag, 1981. L.N. Virgin. Introduction to Experimental Nonlinear Dynamics: A Case Study in Mechanical Vibration. Cambridge University Press, 2000. J.A. Gottwald, L.N. Virgin, and E.H. Dowell. Experimental mimicry of Duffing’s equation. Journal of Sound and Vibration, 158:447–67, 1992. G.H. Argyris and H.-P. Mlejnek. Dynamics of Structures. North-Holland, 1991. N.B. Tufillaro, T. Abbott, and J. Reilly. An Experimental Approach to Nonlinear Dynamics and Chaos. Addison-Wesley, 1992. K.D. Murphy. Theoretical and experimental studies in nonlinear dynamics and stability of elastic structures. Ph.D. dissertation, Duke University, 1994. H. Chen. Nonlinear analysis of post-buckling dynamics and higher order instabilities of flexible structures. Ph.D. dissertation, Duke University, 2004. J.W. Miles. Resonant, nonplanar motion of a stretched string. Journal of the Acoustical Society of America, 75:1505–10, 1984. G. Duffing. Erzwungene Schwingungen bei veranderlicher Eigenfrequenz. F. Vieweg u. Sohn, 1918. H.M. Irvine. Cable Structures. MIT Press, 1981. O.M. O’Reilly and P.J. Holmes. Non-linear, non-planar and non-periodic vibrations of a string. Journal of Sound and Vibration, 153:413–35, 1992. N. Yamaki and A. Mori. Non-linear vibrations of a clamped beam with initial deflection and initial axial displacement, part I: Theory. Journal of Sound and Vibration, 71:333– 46, 1980. N. Yamaki, K. Otomo, and A. Mori. Non-linear vibrations of a clamped beam with initial deflection and initial axial displacement, part II: Experiment. Journal of Sound and Vibration, 71:347–60, 1980. J.G. Eisley. Large amplitude vibration of buckled beams and rectangular plates. AIAA Journal, 2:2207–9, 1964. G.-B. Min and J.G. Eisley. Nonlinear vibration of buckled beams. ASME Journal of Engineering for Industry, 94:637–46, 1972. A.H. Nayfeh, W. Kreider, and T.J. Anderson. Investigation of natural frequencies and mode shapes of buckled beams. AIAA Journal, 33:1121–6, 1995. W.Y. Tseng and J. Dugundji. Nonlinear vibrations of a buckled beam under harmonic excitation. Journal of Applied Mechanics, 38:467–76, 1971.

18:14

P1: RTT Chapter˙16

CUFX159-Virgin

0 521 85648 5

August 14, 2007

References [24] W.T. Thomson. Theory of Vibration with Applications. Prentice Hall, 1981. [25] D.J. Inman. Engineering Vibration. Prentice Hall, 2000. [26] G.V. Rao and K.K. Raju. Large amplitude free vibration of beams – an energy approach, Zeitschrift fur ¨ Angewandte Mathematik und Mechanik 83:493–8, 2003. [27] A.V. Srinivasan. Large amplitude free oscillations of beams and plates. AIAA Journal, 3:1951–3, 1965. [28] C. Mei. Nonlinear vibration of beams by matrix displacement method. AIAA Journal, 10:355–7, 1972. [29] J.M.T. Thompson and G.W. Hunt. Elastic Instability Phenomena. Wiley, 1984. [30] H. Wagner. Large-amplitude free vibrations of a beam. Journal of Applied Mechanics, 82:887–90, 1965. [31] A.W. Leissa. Vibration of plates. Technical Report SP–160, NASA, 1969. [32] S.P. Timoshenko and S. Woinowsky-Krieger. Theory of Plates and Shells, 2nd ed. McGraw-Hill, 1968. [33] H.-N. Chu and G. Herrmann. Influence of large amplitudes on free flexural vibrations of rectangular elastic plates. Journal of Applied Mechanics, 23:532–40, 1956. [34] D.A. Evensen and R.E. Fulton. Some studies on the nonlinear dynamic response of shell-type structures. Technical Report, NASA TMX 56843, 1965. [35] J.M. Johnson and A.K. Bajaj. Amplitude modulated and chaotic dynamics in resonant motion of strings. Journal of Sound and Vibration, 128:87–107, 1989. [36] L. Nicu and C. Bergaud. Experimental and theoretical investigations on nonlinear resonances of composite buckled microbridges. Journal of Applied Physics, 86:5835–40, 1999. [37] B.L. Clarkson. Review of sonic fatigue technology. Technical Report, NASA Contract Report 4587, 1994. [38] K.D. Murphy, L.N. Virgin, and S.A. Rizzi. Experimental snap-through boundaries for acoustically excited, thermally buckled plates. Experimental Mechanics, 36:312–17, 1996. [39] E.H. Dowell. Aeroelasticity of Plates and Shells. Noordhoff, 1975. [40] P.J. Holmes. Bifurcations to divergence and flutter in flow-induced oscillations. Journal of Sound and Vibration, 53:471–503, 1977. [41] B. van der Pol. The nonlinear theory of electric oscillations. Proceedings of the Institute of Radio Engineers, 22:1051–86, 1934. [42] R.L. Bisplinghoff, H. Ashley, and R.L. Halfman. Aeroelasticity. Addision-Wesley, 1955. [43] J.M.T. Thompson. Instabilities and Catastrophes in Science and Engineering. Wiley, 1982. [44] C.W. Gear. Numerical Initial Value Problems in Ordinary Differential Equations. Prentice Hall, 1971. [45] C.S. Ventress and E.H. Dowell. Comparison of theory and experiment for nonlinear flutter of loaded plates. AIAA Journal, 8:2022–30, 1970. [46] P.J. Holmes. Nonlinear dynamics, chaos, and mechanics. Applied Mechanics Reviews, 43:23–39, 1990. [47] J.M.T. Thompson and H.B. Stewart. Nonlinear Dynamics and Chaos, 2nd ed. Wiley, 2002. [48] D.E. Newland. An Introduction to Random Vibrations and Spectral Analysis. Longman, 1984. [49] V. Brunsden, J. Cortell, and P.J. Holmes. Power spectra of chaotic vibrations of a buckled beam. Journal of Sound and Vibration, 130:1–25, 1989.

345

18:14

P1: RTT Chapter˙16

CUFX159-Virgin

346

0 521 85648 5

August 14, 2007

Nonlinear Vibration [50] P. Grassberger and I. Procaccia. Measuring the strangeness of strange attractors. Physica D, 9:189–208, 1983. [51] A. Wolf, J.B. Swift, H.L. Swinney, and J.A. Vastano. Determining Lyapunov exponents from a time series. Physica D, 16:285–317, 1985. [52] C. Grebogi, E. Ott, and J.A. Yorke. Chaos, strange attractors, and fractal basin boundaries in nonlinear dynamics. Science, 238:632–8, 1987. [53] C.S. Hsu. Cell-to-Cell Mapping: A Method of Global Analysis for Nonlinear for Nonlinear Systems. Springer-Verlag, 1987. [54] E. Eschenazi, H.G. Solari, and R. Gilmore. Basins of attraction in driven dynamical systems. Physical Review A, 39:2609–27, 1989. [55] H.E. Nusse and J.A. Yorke. Basins of attraction. Science, 271:1376–80, 1996. [56] N. Sri Namachchivaya and M.M. Doyle. Chaotic motion of a shallow arch. In Proceedings of the 29th AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynamics, and Materials Conference. AIAA, New York, 1988, pp. 198–209. [57] J.J. Thomsen. Chaotic vibrations of non-shallow arches. Journal of Sound and Vibration, 153:239–58, 1992. [58] M.J. Feigenbaum. Quantitative universality for a class of nonlinear transformations. Journal of Statistical Physics, 19:25–32, 1978. [59] S.H. Strogatz. Nonlinear Dynamics and Chaos. Addison-Wesley, 1994. [60] V.K. Melnikov. On the stability of the center for time periodic solutions. Transactions of the Moscow Mathematics Society, 12:1–57, 1963. [61] J. Guckenheimer and P.J. Holmes. Nonlinear Oscillations, Dynamical Systems, and Bifurcations of Vector Fields. Springer-Verlag, 1983. [62] S. Wiggins. An Introduction to Applied Dynamical Systems Theory and Chaos. SpringerVerlag, 1990. [63] F.C. Moon and P.J. Holmes. A magneto elastic strange attractor. Journal of Sound and Vibration, 65:275–96, 1979. [64] P.J. Holmes and F.C. Moon. Strange attractors and chaos in nonlinear mechanics. Journal of Applied Mechanics, 50:1021–32, 1983. [65] F.C. Moon. Chaotic and Fractal Dynamics, An Introduction for Applied Scientists and Engineers. Wiley, 1992.

18:14

P1: KAE 0521880428ind CUFX159-Virgin

0 521 85648 5

August 14, 2007

Index

abaqus finite-element package, 171, 177, 195, 253, 254, 256 Action, 14 Admissible function, 60, 62, 137 Aeroelasticity, 51, 336 Airy function, 29, 199, 201, 323 Angular momentum, 11 ansys finite-element package, 142, 171, 195 Antinode, 101 Antisymmetric mode, 105 Arc-length coordinates, 237, 240 Arnold tongues, 288 Astatic buckling load, 266 Asymmetric mode, 167, 245, 248 Attracting set, 25 Attractor, see Equilibrium point Augusti’s model, 88, 95, 207, 228, 317 auto continuation package, 94, 205, 206 Autocorrelation function, 341 Autonomous system, 52, 283 Axial stress wave, 277 Axial–torsional coupling, 158 Aysmptotic stability, 51 Basin of attraction, 310, 341 Beck’s problem, 134, 224, 334, 336 Bessel function, 106, 109 Bifurcation Flip, 290 Fold, 34 Hopf bifurcation, 51, 336 Indeterminate, 209 Period doubling, 290, 342 Pitchfork, 123, 290 Saddle-node, 33, 34, 36, 84, 88, 163, 201, 225, 227, 230, 275, 290, 328, 336 Secondary, 88, 90, 205, 206, 228 Subcritical pitchfork, 35, 37, 41, 77, 256, 267, 290 Supercritical pitchfork, 35, 68, 123, 137, 140, 242, 250, 290, 297, 305, 309 Transcritical, 34, 80, 269, 291 Bifurcation theory, 32 Biharmonic equation, 187 Bounded motion, 115, 269 Broadband excitation, 140, 174

Cable elasticity, 105 Cable sag, 104, 106 Calculus of variations, 14 Campbell diagram, 134 Canonical equations of motion, 17 Center, see Equilibrium point Center manifold theory, 32 Centrifugal forces, 131 Chaotic behavior, 319, 334, 337, 339 Characteristic eigenvalue (exponent) (CE), 284, 299 Characteristic equation, 27, 40, 47, 54, 83, 117, 126, 129, 135, 150, 156, 161, 166, 224 Characteristic multiplier (CM), 284, 299 Circular membrane, 109 Collocation method, 63 Combination resonances, 291 Comparison function, 60, 62 Complementary path, 69, 77, 82 Configuration space, 15 Conservation laws, 8, 15 Conservation of energy, 11, 20, 73, 102, 269, 313 Conservative force, 10, 30 Conservative system, 10, 72 Consistent mass matrix, 170, 195 Constant total energy, 72, 74, 312, 322 Constraint forces, 12, 15 Constraints, 15 Continuation, see Path following Coordinate transformation, 84 Coupled-mode flutter, 136 Creep buckling, 291 D’Alembert’s principle, 12, 104, 111, 154, 195 Damped natural frequency, 25 Damping, 9, 18, 64 Critical damping, 26, 40 Damping ratio, 25, 42, 226, 296 Dissipative forces, 18 Energy dissipation, 18, 25, 334 Half-power method, 310 Overdamped, 25, 27 Proportional damping, 48, 51, 61, 170 Underdamped, 25 Viscous damping, 25, 40, 226, 299, 304, 326

347

12:31

P1: KAE 0521880428ind CUFX159-Virgin

0 521 85648 5

348

August 14, 2007

Index Damping ratio, see Damping Dead loading, 86, 161 Difference equation, 298 Differential eigenvalue problem, 46, 59 Differential operator, 59 Dimension (attractor), 340 Displacement function, 102, 104, 121, 137, 169, 321 Displacement loading, 87 Dissipative forces, see Damping Divergence instability, 115, 178, 224, 336 Domain of attraction, 328 Donnell shell theory, 210, 289, 324 Double-well potential, 35 Drifting, 226 Driving force, 18 Duffing’s equation, 319, 323, 326, 327, 339, 341, 342 Dummy suffix notation, 128 Duncan polynomial, 129, 130, 197 Dynamic buckling load, 263 Dynamic equilibrium, 12 Dynamic loading, 261 Dynamic stiffness method, 133, 166 Effective force, 12 Effective stiffness matrix, 31 Eigenfunction, 59, 61, 129 Eigenvalue problem, 45 Elastic foundation, 147, 275, 337 Elastica, 74, 137, 177, 237, 240, 256 Element stiffness matrix, 169 Elliptic integrals, 126, 238, 313, 320, 323 End shortening, 46, 121, 125, 162, 238, 243, 246 Energy dissipation, see Damping Energy methods, 8 Equilibrium, 12, 13, 20, 25, 30, 31, 33–36, 38, 42, 45, 49, 50, 52, 55, 64, 67, 71, 77, 82, 93, 104, 115, 128, 180, 184, 202, 206, 225, 226, 240, 245, 258, 262, 263, 266, 273, 306, 312, 315, 332, 341 Equilibrium point Center, 28 Focus, 27, 53 Inflected node, 27 Node, 27, 34, 53 Point attractor, 41, 340 Saddle point, 27, 72, 263, 273 Sink, 27 Spiral, 27 Equivalent beam, 180 Escape, 263, 271, 332 Euclidian norm, 50 Euler identities, 48, 114, 336 Euler load, 123, 126, 143, 160, 220, 282, 305, 335 Euler–Bernoulli theory, 59, 111, 177, 237 Expansion theorem, 61 Experimental modal analysis, 174, 177, 222 Finite differences, 212 Finite-element method, 63, 168

Fixed point, 283 ¨ Flugge shell theory, 210, 211 Flapping motion, 132 Flexural rigidity, 186 Flexural–torsional buckling, 161 Flexural–torsional coupling, 157 Flip bifurcation, see Bifurcation Floquet exponents, 284 Floquet multiplier, 284, 290 Floquet theory, 284, 285, 299 Flow, 25 Flutter instability, 51, 178, 224, 336 Focus, see Equilibrium point Fold, see Saddle-node bifurcation Follower force, 136, 178 Fourier coefficients, 100 Fractal, 270, 342 Frame structure, 82, 166 Frequency-response function, 246, 294 Frequency spectrum, 222, 250, 319, 330, 340 Galerkin’s method, 62, 111, 124, 200, 206, 264, 289, 320, 322–324, 330, 335 Generalized coordinates, 13, 15, 30, 32, 46, 84, 128, 194 Generalized force, 13, 18 Generalized impulse, 19 Generalized momenta, 16, 17 Generalized velocity, 15, 30, 128 Geometric stiffness matrix, 170, 195 Half-power method, see Damping Hamiltonian, 17 Hamilton’s equations, 17 Hamilton’s principle, 8, 13, 57, 101, 111, 124, 184, 188 Hanging chain, 106, 139 Hann window, 302 Hardening spring characteristic, 316, 319, 323, 327 Harmonic balance, 314, 320, 321, 327, 329 Harmonic excitation, 268, 269, 294, 320, 321, 326, 332 Hill’s determinant, 286 Hill’s equation, 283 Hingeless blade, 132 Holonomic, 15 Homoclinic orbit, 315 Homoclinic solution, 73 Homogeneous solution, 294 Hooke’s Law, 49 Hopf bifurcation, see Bifurcation Hydrostatic pressure, 138 Hysteresis, 41, 86, 205, 206, 209, 328, 339, 340 Ideal stiffness, 155 Impact loading, 174, 277, 278 Imperfection sensitivity, 36, 78, 231, 262, 264, 266 Impulse and momentum, principle of, 19 Impulse input, 261, 267

12:31

P1: KAE 0521880428ind CUFX159-Virgin

0 521 85648 5

August 14, 2007

Index Impulse momentum theorem, 274 Impulsive force, 19 Indeterminate bifurcation, see Bifurcation Inertia, 9, 13 Inertia force, 12, 104 Inertial frame, 9 Inextensional beam theory, 121 Inextensional cable, 103 Inflected node, see Equilibrium point Initial imperfections, 32, 35, 69, 122, 177, 183, 193, 200, 228, 263, 277, 306, 310 Initial postbuckling, 122, 177 Intermittency, 342 Intermittent behavior, 330 Internal forces, 12, 239, 240 Isogrid structure, 178 Isotropic elastic material, 20, 183, 185, 233 Jacobian, 51, 52, 202, 283, 299 Kinetic energy, 17, 19, 21, 30, 36, 46, 64, 83, 101, 108, 121, 128, 138, 152, 159, 188, 196, 274 Kronecker delta, 61 Lagrange multipliers, 15, 212 Lagrange’s equation, 8, 13, 14, 19, 30, 32, 45, 57, 58, 66, 81, 83, 89, 92, 109, 120, 130, 152, 159, 224 Lagrangian, 14, 17, 83, 120, 121 Laplace operator, 187 Laplace transforms, 261 Laplacian, 108 Lateral buckling, 161 Least squares, 63, 230 Levy solution, 193 Lift-off, 243 Limit cycle oscillation (LCO), 337 Limit point, 34, 79, 86, 87, 162, 163, 230, 272 Linear momentum, 8, 11 Linear oscillator, 22, 26, 32, 261, 316, 336, 340 Linearization, 28, 39, 49, 202, 283, 313, 314, 316 Longitudinal wave speed, 319 Lumped parameter, 45, 64, 220 Lyapunov exponent (LE), 340 Lyapunov function, 55 Lyapunov stability, 50, 53, 115 Lyapunov’s direct method, 55 Lyapunov–Schmidt reduction, 63 Mass matrix, 46, 52, 59, 64, 170 Mathematica software package, 94, 240 Mathieu functions, 284 Mathieu’s equation, 282, 286, 290 MATLAB software package, 94, 240 Meissner’s equation, 291 Melnikov theory, 342 Membrane, 108 Membrane effect, 111, 117, 125, 183, 187, 202, 303, 306, 322

349 Modal analysis, 47, 51, 174, 246 Modal assurance criterion, 177 Modal interaction, 133 Modal interactions, 291 Mode jumping, 200, 205 Mode shape, 47, 114–116, 118, 122, 126, 127, 130, 154, 161, 166, 174, 180, 191, 192, 200, 204, 206, 216 Mode veering, 155 Moment frame, 166 Monodromy matrix, 284 Multiple scales, 29 Narrowband excitation, 203, 330 Neimark bifurcation, 285 Newton’s laws, 8, 11, 12, 22, 32 Newton’s method, 253, 328, 330 Nodal point, 101, 155 Node, see Equilibrium point Nonautonomous system, 52, 283 Nonconservative, 17 Nonconservative forces, 18, 51, 134, 334 Nondestructive testing, 219 Nongyroscopic forces, 30 Nonholonomic, 15 Nonlinear boundary-value problem, 240 Nonstationarity, 296, 330 Normal modes, 47, 61, 106, 127 Normal solutions, 284 Numerically stiff systems, 335 Ordinary differential equations, 45 Orthogonality, 60, 63, 99, 201, 329 Orthonormal modes, 320 Overdamped, see Damping Overshoot, 40, 261, 296 Parametric excitation, 282 Parametric resonance, 64, 288, 290, 308 Partial differential equations, 45, 97, 98, 111, 147, 201, 324 Particular solution, 294 Path following, 91, 177, 253 Continuation, 240, 290 Path integral, 10 Path-dependent behavior, 328 Period-doubling bifurcation, see Bifurcation Periodic attractor, 298, 340 Periodic behavior, 24, 56, 341 Perturbation methods, 28, 238, 264, 286 Phase portrait, 25, 27, 32, 49, 73 Phase projection, 49, 310, 330, 337 Phase space, 25, 49, 52, 64, 298, 319, 333, 339 Piston theory, 334 Pitchfork bifurcation, see Bifurcation Plate aspect ratio, 190–192, 197, 202, 206, 210 Poincare´ map, 285 Poincare´ sampling, 37, 283, 297 Poincare´ section, 299, 341, 342

12:31

P1: KAE 0521880428ind CUFX159-Virgin

0 521 85648 5

350

August 14, 2007

Index Point attractor, see Equilibrium point Positive-definiteness, 31, 46, 48, 55, 60, 169 Postbuckled stiffness, 183, 198, 205, 240, 244, 304, 306 Potential energy, 11, 16, 19, 30, 33, 36, 38, 46, 49, 52, 53, 56, 57, 66, 69, 72, 79, 83, 86, 89, 101, 107, 111, 122, 128, 137, 152, 160, 162, 169, 196, 201, 202, 223, 227, 262, 269, 272, 276, 313, 314, 332, 342 Preferred mode of buckling, 278 Pressure vessels, 234 Principal coordinates, 32, 48, 61, 84 Prismatic, 20, 111, 120, 147, 152, 158, 237, 252, 288 Proportional damping, see Damping Pulse loading, 267, 277 pulse software system, 246, 254 Quadratic form, 17, 30, 31, 36, 38, 46, 49 Quasi-periodicity, 288, 342 Quasi-static forcing, 261 Ramp input, 279 Random initial imperfections, 279 Rayleigh’s dissipation function, 19, 51, 64 Rayleigh’s method, 78, 102, 106, 111 Rayleigh’s principle, 62, 101 Rayleigh’s quotient, 56, 62, 120, 138 Rayleigh–Ritz, 62, 120, 132, 137, 139, 150, 240 Receptance, 294, 301 Resonance, 133, 143, 294, 297, 308, 327, 329 Riks’ method, 177, 180, 253 Rise of an arch, 162, 272 Rise time, 262 Ritz analysis, 195 Rotary inertia, 143, 188 Rotor blades, 131 Routh–Hurwitz criterion, 54, 285 Saddle-node bifurcation, see Bifurcation Saddle point, see Equilibrium point Sagging cables, 320 Secondary bifurcation, see Bifurcation Self-adjoint, 60, 62 Self-contact, 244 Self-weight, 111, 136, 139 Semirigid loading, 162, 303 Separation of variables, 59, 98, 322 Separatrix, 269, 315 Settling time, 262 Shallow cable, 103 Shallow elastic arch, 275 Shape function, 130, 169, 194 Shear deformation, 143, 172, 189, 234 Shear diaphragm, 210 Shear diaphragm end conditions, 324 Shear modulus, 185 Shooting method, 240, 243, 256 Sink, see Equilibrium point

Sinusoidal sweep excitation, 174 Slenderness ratio, 177, 180 Slope-deflection equations, 167 Snap-through buckling, 86, 162, 225, 262, 275, 328, 331, 332, 340 Snapdown, 162 Softening spring characteristic, 37, 315, 318, 329, 333 Solar sails, 256 Southwell plot, 201, 203, 217, 307 Southwell’s method, 132 Spectrogram, 134 Spiral, see Equilibrium point Spring–mass system, 23, 28, 295 St. Venant’s torsion, 158 Stable-symmetric branching, see Supercritical pitchfork bifurcation State variable, 23, 26, 52, 283, 313, 336 State vector, 22, 50, 52, 283 Statically indeterminate frame, 172 Stationary value, 30, 49, 56, 62, 273, 321 Step input, 261, 267 Stiffness matrix, 49, 52, 59, 64, 194 Stiffness method, 168 Strain-energy, 20, 21, 31, 53, 64, 66, 82, 111, 121, 124, 128, 148, 151, 158, 169, 187, 193, 194, 196, 224 Strain energy density, 20 Strain tensor, 187 Strange attractor, 342 Stress tensor, 187 Stretched string, 97, 319, 326 Stretching, see Membrane effect Subcritical pitchfork bifurcation, see Bifurcation Supercritical pitchfork bifurcation, see Bifurcation Sylvester’s criterion, 46 Symmetric mode, 104, 106, 167, 301 Tapered section, 150, 256 Tensor summation notation Dummy suffix, 31 Thermal buckling, 80, 199, 200 Thermal loading, 80, 142, 197, 330 Thin-walled bars, 157 Time-lag embedding, 309, 330 Timoshenko beam theory, 143, 172 Total mechanical energy, 11, 263 Traction force, 138, 188 Trajectory, 10, 15, 22, 24, 25, 34, 41, 49, 50, 73, 79, 86, 312, 319, 337, 340 Transcritical bifurcation, see Bifurcation Transient, 22, 25, 51, 86, 261, 263, 268, 270, 283, 333, 340, 342 Transient behavior, 268 Transition curves, 286, 290 Transmissibility, 269, 296, 307 Traveling wave, 101, 244 Trial function, 62 Truss structure, 166

12:31

P1: KAE 0521880428ind CUFX159-Virgin

0 521 85648 5

August 14, 2007

Index Underdamped, see Damping Unstable-symmetric branching, see Subcritical pitchfork bifurcation Variational equation, 51, 71, 283 Vector field, 22 Vibration isolation, 304 Virtual displacements, 12, 58 Virtual work, 12, 13, 18 Virtual work, principle of, 12, 169 Viscous damping, see Damping von Karman theory, 183, 198, 334 Waterfall plot, 134

351 Wave equation, 98, 102 Wavelength, 101, 105, 231, 277, 279 Wavenumber, 101 Weighted residual, 62 Whirling motion, 102, 326 Work done, 10, 46, 53, 64, 124, 129 Work-energy theorem, 10 Wronskian determinant, 285 Young’s modulus, 20 Z transform, 299 Zero-frequency kinetic energy, 56, 62

12:31

Our partners will collect data and use cookies for ad personalization and measurement. Learn how we and our ad partner Google, collect and use data. Agree & close