Chair of Structural Mechanics

Department of Civil, Geo and Environmental Engineering

Technical University of Munich

Structural Dynamics
Prof. Dr.-Ing. Gerhard Müller

summer term 2019

Contents
List of Symbols XI

1 Introduction 1
1.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 General classification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2.1 Demands and loads . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2.2 Frequency ranges . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.2.3 Procedures for the prediction . . . . . . . . . . . . . . . . . . . . . . 5

2 Description of harmonic oscillations 7

2.1 Complex numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.2 Application to time harmonic oscillations . . . . . . . . . . . . . . . . . . . 10
2.2.1 Complex representation . . . . . . . . . . . . . . . . . . . . . . . . . 12

3 Fourier transformation 16
3.1 Periodic signals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
3.1.1 Representation with real coefficients . . . . . . . . . . . . . . . . . . 16
3.1.2 Representation with complex coefficients . . . . . . . . . . . . . . . 20
3.2 Fourier integral . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

4 Setting up the equation of motion 29

4.1 Singledegree of freedom system . . . . . . . . . . . . . . . . . . . . . . . . 29
4.1.1 Equilibrium of forces . . . . . . . . . . . . . . . . . . . . . . . . . . 29
4.1.2 Virtual work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
4.1.3 The principle of Hamilton . . . . . . . . . . . . . . . . . . . . . . . 37
4.1.3.1 Undamped systems . . . . . . . . . . . . . . . . . . . . . . 37
4.1.3.2 Introduction of non-conservative forces . . . . . . . . . . . 40
4.1.3.3 Lagrange multiplicators . . . . . . . . . . . . . . . . . . . . 44
4.2 Multiple degree of freedom systems . . . . . . . . . . . . . . . . . . . . . . 49
4.2.1 Equilibrium of forces . . . . . . . . . . . . . . . . . . . . . . . . . . 49
4.2.2 Principle of Hamilton . . . . . . . . . . . . . . . . . . . . . . . . . . 53
4.3 Damping models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
4.3.1 Linear viscoelastic damping . . . . . . . . . . . . . . . . . . . . . . . 55
4.3.2 Ideal hysteretic damping . . . . . . . . . . . . . . . . . . . . . . . . 62
4.3.3 Coulomb damping . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64

Structural Dynamics
summer term 2019
II
 4.3.4 Aeroelastic damping . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
4.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65

5 Free vibration of linear systems with single degree of freedom 66

5.1 The undamped case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
5.2 The damped case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
5.2.1 Velocity proportional (viscous) damping . . . . . . . . . . . . . . . . 68
5.2.1.1 Quasi periodic case (D < 1) . . . . . . . . . . . . . . . . . 70
5.2.1.2 Aperiodic limit case (D = 1) . . . . . . . . . . . . . . . . . 71
5.2.1.3 Aperiodic case (D > 1) . . . . . . . . . . . . . . . . . . . . 72
5.2.2 Relations between different damping measures . . . . . . . . . . . . 72

6 Classification of Excitations 74
6.1 Modelling of the excitation . . . . . . . . . . . . . . . . . . . .
. . . . . . . 74
6.1.1 Harmonic excitation . . . . . . . . . . . . . . . . . . . .
. . . . . . . 76
6.1.2 Periodic excitation . . . . . . . . . . . . . . . . . . . . .
. . . . . . . 76
6.1.3 Transient load . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . 78
6.1.4 Irregular, random loads . . . . . . . . . . . . . . . . . .
. . . . . . . 79
6.1.5 Loads influenced by the parameters of adjacent systems . . . . . . . 80
6.2 Frequency content . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
6.3 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
6.4 Classification in standards . . . . . . . . . . . . . . . . . . . . . . . . . . . 82

7 Forced vibration of linear systems with single degree of freedom 86

7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
7.1.1 “Quasi-static”and dynamic processes . . . . . . . . . . . . . . . . . 86
7.1.2 Time and frequency representation . . . . . . . . . . . . . . . . . . . 87
7.2 Harmonic excitation for systems with viscous damping . . . . . . . . . . . . 87
7.2.1 Harmonic forced excitation . . . . . . . . . . . . . . . . . . . . . . . 88
7.2.1.1 Steady state solution . . . . . . . . . . . . . . . . . . . . . 89
7.2.1.2 Transient response . . . . . . . . . . . . . . . . . . . . . . 97
7.2.1.3 Resonance case . . . . . . . . . . . . . . . . . . . . . . . . 98
7.2.1.4 Transmissibility . . . . . . . . . . . . . . . . . . . . . . . . 101
7.2.2 Harmonic root point excitation . . . . . . . . . . . . . . . . . . . . . 102
7.2.2.1 Equation of motion . . . . . . . . . . . . . . . . . . . . . . 102
7.2.2.2 Particular solution (forced vibration) . . . . . . . . . . . . 103
7.2.2.3 Transmissibility . . . . . . . . . . . . . . . . . . . . . . . . 104
7.3 Harmonic forced excitation for hysteretically damped systems . . . . . . . . 105
7.4 Arbitrary periodic excitation . . . . . . . . . . . . . . . . . . . . . . . . . . 108
7.4.1 Decomposition into harmonics . . . . . . . . . . . . . . . . . . . . . 108
7.5 Aperiodic excitation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
7.5.1 Time domain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
7.5.1.1 Unit impulse response function . . . . . . . . . . . . . . . 114

Structural Dynamics Contents

summer term 2019 III
 7.5.1.2 Superposition, Duhamel-Integral . . . . . . . . . . . . . . . 116
7.5.1.3 Response spectra . . . . . . . . . . . . . . . . . . . . . . . 118
7.5.2 Exkursus: Kelvin-Voigt model . . . . . . . . . . . . . . . . . . . . . 121
7.5.3 Frequency domain . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
7.5.3.1 Response to harmonic excitation . . . . . . . . . . . . . . . 124
7.5.3.2 Response to aperiodic excitation . . . . . . . . . . . . . . . 125

8 Free vibration of linear systems with multiple degrees of freedoms 128

8.1 Preliminary remark: Maxwell-Betti theorem . . . . . . . . . . . . . . . . . 128
8.2 Solution of the homogeneous system of equations . . . . . . . . . . . . . . . 129
8.2.1 Undamped MDOF systems . . . . . . . . . . . . . . . . . . . . . . . 130
8.2.2 Consideration of the initial conditions . . . . . . . . . . . . . . . . . 134
8.2.3 Damped MDOF systems . . . . . . . . . . . . . . . . . . . . . . . . 141
8.2.3.1 The quadratic eigenvalue problem . . . . . . . . . . . . . . 142
8.2.3.2 The standard form of the damped eigenvalue problem . . . 143
8.2.3.3 Rayleigh damping . . . . . . . . . . . . . . . . . . . . . . . 147
8.2.3.4 Caughey damping model using additional orthogonality rela-
tions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150

9 Forced vibration of linear systems with multiple degree of freedoms 151

9.1 Direct solution for harmonic loads . . . . . . . . . . . . . . . . . . . . . . . 151
9.2 Modal analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
9.3 Time integration methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
9.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160

10 Impedance models 162

10.1 Preliminary definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
10.2 Dynamic stiffness and impedance . . . . . . . . . . . . . . . . . . . . . . . 166
10.3 Dynamic stiffnesses and impedances in passive systems . . . . . . . . . . . 167
10.3.1 Mass impedance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
10.3.2 Conservative forces . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
10.3.2.1 Potential . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
10.3.2.2 Spring impedance . . . . . . . . . . . . . . . . . . . . . . . 170
10.3.3 Non-conservative forces . . . . . . . . . . . . . . . . . . . . . . . . . 171
10.3.3.1 Damping impedance . . . . . . . . . . . . . . . . . . . . . 171
10.3.3.2 Loss factor . . . . . . . . . . . . . . . . . . . . . . . . . . . 172
10.4 Parallel and serial connection . . . . . . . . . . . . . . . . . . . . . . . . . . 175
10.4.1 Parallel connnection . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
10.4.2 Serial connnection . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176
10.5 Arbitrary connections . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176
10.6 Impedances for selected systems . . . . . . . . . . . . . . . . . . . . . . . . 181

Structural Dynamics Contents

summer term 2019 IV
11 Solutions for selected continuous systems 183
11.1 String, longitudinal and torsional rod . . . . . . . . . . . . . . . . . . . . . 183
11.1.1 Equation of motion . . . . . . . . . . . . . . . . . . . . . . . . . . . 183
11.1.2 Free vibration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
11.1.2.1 Solution via a separation approach . . . . . . . . . . . . . 185
11.1.2.2 Direct solution via a wave approach . . . . . . . . . . . . . 190
11.2 Euler-Bernoulli beam . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
11.2.1 Equation of motion . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
11.2.2 Free vibration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
11.2.2.1 Solution via a separation approach . . . . . . . . . . . . . 198
11.2.2.2 Solution via a wave approach . . . . . . . . . . . . . . . . 202
11.2.3 Displacement approaches - spectral elements based on analytic solutions 205
11.2.3.1 Beams without loads and harmonic excitation at the bound-
aries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
11.2.3.2 Loaded beam elements - particular solution . . . . . . . . . 210
11.3 Extended beam theories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212
11.4 Thin plates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
11.5 Eigenfrequencies and mode shapes of selected systems . . . . . . . . . . . . 217
11.6 Modal analysis for loaded Euler-Bernoulli beam . . . . . . . . . . . . . . . 218
11.6.1 Euler-Bernoulli beam under step load . . . . . . . . . . . . . . . . . 218
11.7 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227

12 Approximate solutions of continuous systems 228

12.1 Discretization of continuous systems . . . . . . . . . . . . . . . . . . . . . . 228
12.1.1 The principle of virtual work . . . . . . . . . . . . . . . . . . . . . . 231
12.1.2 Lumped mass matrix and static condensation . . . . . . . . . . . . . 236
12.1.3 Principle of Hamilton . . . . . . . . . . . . . . . . . . . . . . . . . . 237

13 Random vibrations 238

13.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . 238
13.2 Description of random variables . . . . . . . . . . . . . .
. . . . . . . . . . 240
13.2.1 Single random variable . . . . . . . . . . . . . . .
. . . . . . . . . . 240
13.2.2 Two random variables . . . . . . . . . . . . . . . .
. . . . . . . . . . 243
13.3 Description of stochastic processes . . . . . . . . . . . . .
. . . . . . . . . . 246
13.3.1 Probability distribution . . . . . . . . . . . . . . .
. . . . . . . . . . 246
13.3.2 Moment functions . . . . . . . . . . . . . . . . . .
. . . . . . . . . . 247
13.3.3 Stationarity of stochastic processes . . . . . . . . .
. . . . . . . . . . 249
13.3.4 Ergodicity of stochastic processes . . . . . . . . .
. . . . . . . . . . 250
13.3.5 The Gaussian process . . . . . . . . . . . . . . . .
. . . . . . . . . . 251
13.3.6 Frequency domain analysis of stochastic processes . . . . . . . . . . 252
13.3.7 Stochastic response of linear dynamic system . . . . . . . . . . . . . 256

Structural Dynamics Contents

summer term 2019 V
14 Aeroelastic vibrations (wind) 259
14.1 General information . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 259
14.1.1 Loads . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 259
14.1.2 Description of wind . . . . . . . . . . . . . . . . . . . . . . . . . . . 260
14.1.2.1 Wind velocity . . . . . . . . . . . . . . . . . . . . . . . . . 260
14.1.2.2 Temporal and spatial structure of wind velocity measurements 260
14.1.3 Temporal structure of the wind velocity . . . . . . . . . . . . . . . . 261
14.1.3.1 Power spectral density . . . . . . . . . . . . . . . . . . . . 261
14.1.3.2 Turbulence intensity . . . . . . . . . . . . . . . . . . . . . 262
14.1.4 Spatial structure of the wind velocity . . . . . . . . . . . . . . . . . 262
14.1.4.1 Atmospheric boundary layer - wind profile . . . . . . . . . 262
14.1.4.2 Turbulences . . . . . . . . . . . . . . . . . . . . . . . . . . 264
14.2 Wind induced vibrations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 264
14.2.1 Gust induced vibrations . . . . . . . . . . . . . . . . . . . . . . . . . 265
14.2.2 Vortex induced vibrations . . . . . . . . . . . . . . . . . . . . . . . . 268
14.2.2.1 Occurence of vortices . . . . . . . . . . . . . . . . . . . . . 269
14.2.2.2 Frequency of vortices . . . . . . . . . . . . . . . . . . . . . 270
14.2.2.3 Forces related to vortices . . . . . . . . . . . . . . . . . . . 271
14.2.2.4 Assessment of vortex induced vibrations . . . . . . . . . . 271
14.2.3 Galloping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 273
14.2.4 Divergence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 276
14.2.5 Flutter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 276
14.2.6 Interference effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . 278

15 Earthquakes 279
15.1 General information and terms . . . . . . . . . . . . . . . . . . . . . . . . . 279
15.1.1 Magnitude . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 280
15.1.2 Intensity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 281
15.1.3 Calculation procedures . . . . . . . . . . . . . . . . . . . . . . . . . 282
15.2 Response spectrum analysis (RSA) . . . . . . . . . . . . . . . . . . . . . . . 283
15.2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 283
15.2.2 Design codes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 287
15.2.3 Constructive notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . 289
15.3 Capacity design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 290
15.3.1 Basics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 290
15.3.2 Reduction factors . . . . . . . . . . . . . . . . . . . . . . . . . . . . 292
15.3.3 The capacity design . . . . . . . . . . . . . . . . . . . . . . . . . . . 293

16 Numerical time step procedures 296

16.1 General remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 296
16.1.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 296
16.1.2 Explicit - Implicit . . . . . . . . . . . . . . . . . . . . . . . . . . . . 296
16.1.3 Errors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 296

Structural Dynamics Contents

summer term 2019 VI
 16.2 Newmark-β method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 297
16.2.1 General approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . 297
16.2.2 Constant acceleration over the time interval . . . . . . . . . . . . . . 297
16.2.3 Predictor/ corrector approach . . . . . . . . . . . . . . . . . . . . . 299
16.2.4 Stability considerations . . . . . . . . . . . . . . . . . . . . . . . . . 300
16.3 The central difference method . . . . . . . . . . . . . . . . . . . . . . . . . 301
16.3.1 Derivation of ẅ and ẇ out of w . . . . . . . . . . . . . . . . . . . . 301
16.3.2 General approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . 301
16.3.3 Start procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 302
16.4 Non-linear problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 302

Bibliography 305

Structural Dynamics Contents

summer term 2019 VII
List of Figures
1.1 Two dimensional probability density function for the demand S and the resis-
tance R and limit-state function . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Requirements for highly sensitive equipment given in VDI 2038 [2012]. . . 4
1.3 Prediction methods in structural dynamics . . . . . . . . . . . . . . . . . . 6

2.1 Description of a complex number . . . . . . . . . . . . . . . . . . . . . . . . 8

2.2 Visualization of multiplication of two complex numbers . . . . . . . . . . . 10
2.3 Point on unit circle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.4 Point on unit circle (f0 = 1) with phase shift . . . . . . . . . . . . . . . . . 13
2.5 Representation of complex pointers with different contributions of cosine and
sine part. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.6 The complex exponential function evolves harmonically in the complex plane
over time. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

3.1 Graphical representation of complex Fourier series coefficients.

  . . . . . . 21
3.2 Fourier transform of the rectangular function g(t) = g0 Π Tt0 . . . . . . . . 24
 
3.3 Fourier transform of the rectangular function g 0 (t) = h0 Π t
T00
with h0 = 2g0
and T00 = T20 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.4 Fourier transform of the impulse function g 0 (t) = g0 T0 δ(t). . . . . . . . . . 25
3.5 Fourier transform of the sine function g(t) = g0 sin(2πf0 t). . . . . . . . . . 26
3.6 Fourier transform of the cosine function g(t) = g0 cos(2πf0 t). . . . . . . . 27

4.1 External and reaction forces on single mass point. . . . . . . . . . . . . . . 29

4.3 Illustrations of different viscoelastic models. . . . . . . . . . . . . . . . . . 58

5.1 Free body cut for the unforced (free) SDOF system . . . . . . . . . . . . . 66
5.2 Homogeneous solution of SDOF-system . . . . . . . . . . . . . . . . . . . . 68
5.3 Damped oscillation over one period T with w0 = 1 and ϕ = 0. . . . . . . . 71
5.4 Homogeneous solution of the SDOF system for the aperiodic case D > 1. . 73

6.1 Transient, triangular load . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75

6.2 Representation of an aperiodic signal using an infinite return period . . . . 75
6.3 Periodic signal and harmonic decomposition . . . . . . . . . . . . . . . . . 77
6.4 Transient, triangular load . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
6.5 Representation of an aperiodic signal using an infinite return period . . . . 79
6.6 Illustration of a stochastic process . . . . . . . . . . . . . . . . . . . . . . . 79

Structural Dynamics
summer term 2019
VIII
 6.7 Illustration of the autocorrelation function . . . . . . . . . . . . . . . . . . 80
6.8 Synchronization of pedestrians with the structural vibration, courtesy of M.
Schneider . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
6.9 Frequency content characterization in a emission–transmission–imission sys-
tem. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
6.10 Common methods in strucutral dynamics and vibro-acoustics and their appli-
cability. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82

7.2 Displacement solution wp (t) in the complex plane. . . . . . . . . . . . . . . 91

7.4 Transfer function in the complex plane. . . . . . . . . . . . . . . . . . . . . 94
7.5 Transfer function of the SDOF system over frequency . . . . . . . . . . . . 95
7.6 Particular solution in the complex plane. . . . . . . . . . . . . . . . . . . . 96
7.8 Transient response in resonance for D = 0.1. . . . . . . . . . . . . . . . . . 100
7.13 Transfer function for linear hysteretic damping in the complex plane. . . . 106
7.14 Amplification function harmonically foced SDOF system with linear hysteretic
damping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
7.15 Example for an arbitrary periodic load time history. . . . . . . . . . . . . 108
7.16 Square wave with period T . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
7.18 Unit impulse response function for varying percentage of critical damping D 115
7.21 Response spectrum of SDOF system for rectangular excitation . . . . . . . 119
7.22 Triangular shock on column with varying length . . . . . . . . . . . . . . . 120
7.24 Response spectra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
7.25 Displacement response response of Kelvin-Voigt model. . . . . . . . . . . . 123
7.26 Response of SDOF system to arbitrary excitation. . . . . . . . . . . . . . . 127

8.1 Solution for the 2DOF system. . . . . . . . . . . . . . . . . . . . . . . . . 141

8.2 Solution for the 2DOF system. . . . . . . . . . . . . . . . . . . . . . . . . 147

9.1 Displacement w1+ for the 2DOF system under harmonic excitation, direct so-
lution, undamped system . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
9.2 Displacement w1+ for the 2DOF system under harmonic excitation, direct so-
lution, damped system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
9.3 Solution for the 2DOF system under two step loads . . . . . . . . . . . . . 160

10.3 Physical quantities f (t) and v(t). . . . . . . . . . . . . . . . . . . . . . . . 164

10.8 Voigt-Kelvin-Modell. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
10.9 SDOF impedance model for different values of damping ratio . . . . . . . . 178
10.10Elastic element impedance model for different values of damping ratio . . . 179
10.11Elastic element impedance model for different values of damping ratio . . . 181

11.1 Forces on differential elements. . . . . . . . . . . . . . . . . . . . . . . . . 184

11.2 Boundary conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 186
11.3 Part of the wave solution. . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
11.4 Wave traveling in the positive x-direction. . . . . . . . . . . . . . . . . . . 191

Structural Dynamics List of Figures

summer term 2019 IX
 12.1 Nodal forces and displacement degrees of freedom . . . . . . . . . . . . . . 229
12.2 Functions for unit displacements/rotations - cubic approach . . . . . . . . . 231
12.3 Nodal forces for the beam element . . . . . . . . . . . . . . . . . . . . . . . 234

13.5 Autocorrelation function φXX for the above process in terms of time shift. 249
13.6 Five realizations of the stochastic process with X0 following the standard normal
distribution. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 249
13.7 from [Petersen 2000, p.1220, Fig. 34] . . . . . . . . . . . . . . . . . . . . . 251
13.8 Comparison of different processes from [Petersen 2000, p. 1228, Fig. 43]) . . 255
13.9 Autocorreation function and spectral density for the white noise process [Pe-
tersen 2000, p. 1228, Fig. 43]) . . . . . . . . . . . . . . . . . . . . . . . . . 256

14.1 Measured wind velocities at three different heights [Holmes 2007, p. 56] . . 261
14.2 Frequenzspektrum der horizontalen Windgeschwindigkeit (1957), Messungen in
Brookhaven, New York [Burton et al 2001, p. 12], [Petersen 2000, p. 596] . 262
14.3 aus Ruscheweyh, Dynamische Windwirkung an Bauwerken Vol. 2 (1982), S.
16, Abb. 2.3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 264
14.4 Static and dynamic part of the wind velocity . . . . . . . . . . . . . . . . . 265
14.5 Comparison of different models: z = 10 m, v̄10 = 20 m s , z0 = 0.05 (dimen-
sionless), κ = 0.006 (open grassland, grassland); above: log-log illustration of
the dimensionless turbulence spectra; below: Comparison for low frequencies
[Runtemund 2013] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 266
14.6 Example for Davenport spectrum . . . . . . . . . . . . . . . . . . . . . . . 268
14.7 Vortices, from [Petersen 2000] . . . . . . . . . . . . . . . . . . . . . . . . . 269
14.8 Different states of vortex shedding . . . . . . . . . . . . . . . . . . . . . . . 270
14.9 vgl: Bachmann, Hugo, Vibration Problems in Structures (1997), in: Vibrations
in across-wind direction induced by vortex-shedding, S. 203, Abb. H.10 . . 273
14.10Examples for the aerodynamic coefficient cs (α) . . . . . . . . . . . . . . . . 274
14.11Profiles at which Galloping can occur . . . . . . . . . . . . . . . . . . . . . 275
14.12Flutter, [Ruscheweyh 1982, Vol. 2 (1982), S. 118, Abb. 4.67] . . . . . . . . 277

15.1 Terminology of an earthquake, from [Flesch 1993, p. 203, Fig. 7.1] . . . . . 280
15.2 Frequency of occurrence of magnitudes for the whole earth (a) and the Rhine
valley (b) from [Flesch 1993, p. 203, Fig. 7.2] . . . . . . . . . . . . . . . . . 281
15.3 Exemplary accelerogram given in terms of the acceleration relative to the grav-
itational acceleration on earth g. . . . . . . . . . . . . . . . . . . . . . . . 282
15.4 SDOF system with different natural frequencies and damping ratios for the
response spectrum analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . 285
15.5 Response spectra for absolute and pseudo-absolute acceleration, relative and
pseudo-relative velocity and relative displacement . . . . . . . . . . . . . . 286
15.6 Design response spectrum according to the DIN 4149. . . . . . . . . . . . 288

Structural Dynamics List of Figures

summer term 2019 X
List of Symbols

Greek Symbols

α, β angle; also coefficients (of the Rayleigh damping)

α vector of coefficients
γ shear strain; also coefficient (of the Newmark-β method)
δ(x) Dirac-delta function
δ decay constant
 vector of strains
 mean-square error; also strain
η loss factor; also kinematic viscosity
ϑ propagating angle
λ eigenvalue (scalar); also wave length
Λ diagonal matrix of squared eigenfrequencies, i.e., [Λ]ii = λi = ωi2
µ mass per unit length; also ductility
ν Poissons ratio
ξ dimensionless coordinate, i.e., ξ = xl ; also percentage of critical
damping
Π potential
ρ density
σ internal force vector; also vector of stresses
σ stress; also standard deviation
τ time; also shear stress
ϕ Phase
φ eigenvector
φ (cross-sectional) rotation
Φ matrix of eigenvectors
Ω circular frequency (of the load)
ω radial frequency i.e., ω = 2πf

Roman Symbols

A, B amplitudes

Structural Dynamics
summer term 2019
XI
 A cross section
A flexibility matrix
a acceleration
ak coefficient of the Fourier series
B0 bending stiffness
bk coefficient of the Fourier series
C damping matrix
c damping constant; also wave velocity
D damping ratio, i.e. percentage of critical damping D = 2√ckm
d diameter
E Young’s modulus; also energy
F {·} Fourier transformation
F −1 {·} inverse Fourier transformation
f (x), g(x) function/ signal of variable x
f force; also frequency f = T1
f force vector
G shear modulus
g gravitational acceleration
H horizontal force
H(t) Heaviside function
h(t) impulse response function
h̃(t) transfer function, Fourier transform of the impulse response func-
tion
h thickness
I rotatory inertia; also intensity
i imaginary unit, i.e., i2 = −1
i rotatory inertia per unit length
K stiffness matrix
k stiffness
L(t) Lagrange function
l length
M moment; also magnitude (of an earthquake)
M mass matrix
m mass
N normal force
N shape function vector
n number of DOFs
Pr(E) probability of event E
Q lateral force
q velocity pressure (dynamic pressure)
R Resistance
Re Reynolds number

Structural Dynamics List of Symbols

summer term 2019 XII
 r absolute value of a complex number; also reaction force; also radius
S demand; also spectral density; also Strouhal number
T time period
T (t) kinetic energy
t time
U (t) potential energy; also unit-step function
Ub transfer function
u displacement
V lateral force; also amplification function
v velocity
Var(X) variance of X
W (x) space function
W work
w displacement vector
w displacement
Z impedance
X random variable
x real part of a complex number; also Cartesian spatial coordinate
Y (x) space function
Y random variable
y imaginary part of a complex number; also Cartesian spatial coor-
dinate
Z impedance
z state space vector
z complex number; also Cartesian spatial coordinate
z∗ complex conjugate of complex number z

Mathematical Operations

∂
∂x
partial differential operator with respect to x
d
dx
differential operator with respect to x
Fourier transformation in frequency domain
inverse Fourier transformation in time domain
()T matrix transpose
ẇ the first derivative of w with respect to time
ẅ the second derivative of w with respect to time
w0 the first derivative of w with respect to the spatial coordinate
w00 ,w00 , . . . the second, third,. . . derivative of w with respect to the spatial
coordinate

Structural Dynamics List of Symbols

summer term 2019 XIII
Acronyms

BEM boundary element method

CDF cumulative distribution function
DOF degree of freedom
FEM finite element method
FT Fourier transformation
IFT inverse Fourier transformation
MDOF multi eegree of freedom
ODE ordinary differential equation
PDE partial differential equation
PDF probability density function
SDOF single degree of freedom
SEA statistical energy analysis
VDI Verein Deutscher Ingenieure

Structural Dynamics List of Symbols

summer term 2019 XIV
1 Introduction

1.1 Overview

With increasing slenderness of structures, growing comfort demands, higher demands for low
vibrational influence of machinery and a dense arrangement of buildings, the requirements of
structural dynamic investigations increase.
The foundations of structural dynamics were found by Newton, Euler, d’Alembert, Lagrange
and Hamilton in the 17th and 18th century.
The insight, that bridges demand special treatment, has surely been in the mind of the builders
before the collapse of the Braughton bridge in 1831. However, the influence of moving loads
on bridges has been treated since the middle of the 19th century. Noteworthy are the names
of Willis, Zimmermann and Timoshenko, Fryba, Bachmann, Grundmann.
The names Geiger und Rausch are connected with machinery foundations, that are supported
in order to reduce the elastic dynamic forces. Regarding the installation conditions of sensitive
equipment the relevant publications by Ungar, Heiland has to be mentioned.
The topic soil-structure interaction for linear loading processes was extensively treated, e.g.
by Wolf and Gazetas.
For the topic of earthquakes there are a number of research groups in the USA and Japan. This
field of work is closely connected to the names of Biot, Housner, Newmark and Rosenblueth.
In Germany also since 1951, after the earthquake of Euskirchen with the name of Meskouris,
and in Switzerland with the name of Bachmann.
Wind induced vibrations especially rose to general awareness since the collapse of the Tacoma
Narrows bridge in 1940. Here the name Ruscheweyh is to be mentioned. The treatment of
irregular and stochastic excitations is connected to the names Freudenthal, Shinuozuka, Lin,
Grundmann, Schueller, Soize.
Consideration of structure borne sound, i.e., higher frequency vibrations, were in particular
treated by Cremer, Heckl and Lyon.
The mentioning of these names is only exemplarily — as the overview over the technical rules
and literature — and shall facilitate the introduction of the various topics for the reader.

Structural Dynamics
summer term 2019
1
1.2 General classification

1.2.1 Demands and loads

Dynamic loads acting on buildings can be relevant with respect to structural safety, fatigue
strength, or structural serviceability, depending on the impact strength and the demand. The
demands and loads therefore depend on the specific task, distinguished as follows:
• Particularly strong dynamic loads, e.g. impact from
– earthquakes,
– explosions,
– airplane crash, or
– vehicle impact
have to be treated with respect to the damage and the collapse of the structures (struc-
tural safety).
• Dynamic loads that occur repeatedly regularly and do not cause damages are impacts
from

– wind,
– pedestrians,
– moving persons, or
– moving loads.

These have to be examined with respect to fatigue (fatigue strength) and serviceability of the
structure. Demands on the serviceability can also exist for various impacts with low strength,
e.g. for vibrations due to railway or street traffic, vibrations due to building operations,
vibrations due to moving persons, or due to machinery induced dynamic forces. They do not
have to be assessed with respect to fatigue or structural safety, but with respect to comfort
requirements or the useability of sensitive equipment.
Whereas non-linear approaches are necessary for the assessment of structural safety under
one-time extreme load impacts (that are partly rehashed by linear replacement procedures for
practical applications), for problems concerning the fatigue strength and the serviceability, in
general procedures linear material laws are applied.
The semi-probabilistic safety concept in the context of structural safety sets minimal proba-
bilities of failure Pr(R < S), with the resistance of the structure R and the demand S.
For the serviceability much lower vibration limits exist, for which the exceedance is usually less
critical, as it is not linked to danger to life, but to a non-compliance with comfort requirements

Structural Dynamics 1 Introduction

summer term 2019 2
of the people in the building or with requirements for the operation of sensitive equipment.
Thereby, larger exceedance probabilities compared to the structural safety concept can be
accepted (comp. Fig. 1.1b).

(a) Structural safety (b) Serviceability

Figure 1.1: Two dimensional probability density function for the demand S and the resistance R and
limit-state function (identity R = S ) from Müller [2009].

Since the approaches and results from dynamic prognoses can be tested via measurements of
the construction’s response especially for the assessment of fatigue and serviceability, mea-
surement procedures play a significant role for the demand S as well as the resistance R value.
Also it is often useful to increase the prediction accuracy by means additional measurements in
the planing and building phase. Thus, the uncertainties linked to the prediction for the proof
of structural safety under static loads are significantly less verifiable through measurements
than the uncertainties in structural dynamics.
Especially for problems concerning the serviceability structural dynamic investigations are
not necessary for all building projects. In VDI 2038 [2012] criteria are summarized in order to
decide exemplarily to what extent dynamic investigations are necessary. Fig. 1.2 exemplarily
depicts current requirements given in VDI 2038 [2012].

Structural Dynamics 1 Introduction

summer term 2019 3
 1000

VC-A

100 VC-B
VelocityH[µHm/s] VC-C

VC-D

10 VC-E
Nano-D

Nano-E

Nano E-F
1

0.1
1 10 100 1000
FrequencyH[Hz]
VC-A VC-B VC-C VC-D
VC-E Nano-D Nano-E Nano E-F

Figure 1.2: Requirements for highly sensitive equipment given in VDI 2038 [2012].

1.2.2 Frequency ranges

In structural dynamics the following frequency ranges are distinguished:

• vibrations: 0 Hz (static deformation) to 80 Hz
• structure borne sound: from around 16 Hz to multiple kHz
The frequency content of a time varying signal can be assessed with the help of signal analysis
(comp. Cha. 3)

Structural Dynamics 1 Introduction

summer term 2019 4
 sound waves z y

z Reradiated sound

x, y, z Vibrations

Waves in solids

Emission Transmission Immission

Vibrations Structure borne sound

1 2 5 10 20 5080100 1000
Frequency in Hz

1.2.3 Procedures for the prediction

For the numerical prediction of noise and vibration there are a variety of methods available,
which are depicted in Fig. 1.3. In engineering practice the different methods are typically
applied depending on the system, the frequency and the type of problem (Fig. 1.3). Espe-
cially uncertainties in the modelization and the necessary detailing of the response, influence
considerably the choice of the method. The methods are not competing with each other, but
are complementary, so that there is a "Marriage à la mode“ (Zienkiewicz et al [1977]) - "entre
plusiers personnes“ - not only applying jointly the Boundary and Finite Element method, but
also analytical methods, the discretization by the help of lumped mass systems, Structural
Fuzzy (Soize [1993]; Weaver [1997]) and the Statistical Energy Analysis SEA. Thus, there is a
demand for hybrid methods, bringing together the various approaches for either different types
of sub-systems or different frequency ranges (Langley and Bremner [1999]; Langley [1990]).

Structural Dynamics 1 Introduction

summer term 2019 5
 hybrid
analytical approaches
algorithms

unlimited

systems M A
Mu BE SE
osc ltib
limited illa ody
tor
M high
simple FE

boundary frequency
conditions
complex low
Figure 1.3: Prediction methods in structural dynamics

Structural Dynamics 1 Introduction

summer term 2019 6
2 Description of harmonic oscillations
This chapter gives an introduction to the description of harmonic oscillations. Arbitrary
harmonic functions with a phase shift can be described through sine and cosine functions.
Computations with those functions can be greatly simplified through the use of a complex
description. To this end, we first introduce complex numbers together with important rules of
computation and subsequently treat the description of harmonic oscillations.

2.1 Complex numbers

Consider a complex number z ∈ C

z = x + iy (2.1)

where x = Re (z) and y = Im (z) are the real and imaginary part of z. We define the complex
conjugate of z as

z ∗ = x − iy .

With this, we introduce the following identities, that define the real and imaginary part of a
complex number using complex conjugation

z + z∗
Re (z) = (2.2)
2
z − z∗
Im (z) = . (2.3)
2i

We can also state z in polar form

z = r (cos ϕ + i sin ϕ) (2.4)

Structural Dynamics
summer term 2019
7
 Im
z − z∗

Im
y z −z ∗ i Im z z

Re z + z ∗ Re
ϕ x ϕ
−ϕ −ϕ
Re z
−y z∗ z∗

Figure 2.1: Geometric description of a complex number. On the left hand side, the complex number
z with real part x and imaginary part y is depicted. It’s complex conjugate z ∗ can be
geometrically interpreted as the reflection of z about the real axis. The right hand side
gives a geometric interpretation of Eqs. (2.2) and (2.3). The sum z + z ∗ always results in
a real outcome, whereas z − z ∗ always gives a purely imaginary result.

and from comparison with Eq. (2.1) we find

x = r cos ϕ
y = r sin ϕ .

Now, consider Euler’s formula

eiϕ = cos ϕ + i sin ϕ (2.5)

which introduces the complex exponential function. With this, we can simply describe the
complex number z as

z = reiϕ (2.6)

where r is the absolute value of the complex number z = x + iy = reiϕ ,

q √
|z| = x2 + y 2 = zz ∗ = r (2.7)

and ϕ is the argument or phase of z and is given by the relation

Im z
tan(ϕ) = . (2.8)
Re z

Structural Dynamics 2 Description of harmonic oscillations

summer term 2019 8
As the inverse relation is not uniquely defined throughout the range iof ϕ, using h the arctan-
function, the phase of z is just uniquely defined within the interval ϕ ∈ − π2 , π2 . To circumvent
this, we introduce the function atan2,

ϕ = arg z = atan2 (Im z, Re z) = atan2 (y,x) , (2.9)

with
 
arctan y
for x > 0

x


 
arctan y + π for x < 0 and y > 0


x



arctan y − π for x < 0 and y ≤ 0


atan2 (y,x) = x (2.10)

 π
for x = 0 and y > 0
2


− 2
π
for x = 0 and y < 0




undefined for x = 0 and y = 0 .



From Eq. (2.5) we can further deduce

 
cos ϕ = Re eiϕ
 
sin ϕ = Im eiϕ .

Through the symmetry properties cos (−ϕ) = cos (ϕ) and sin (−ϕ) = − sin (ϕ) and using (2.4),
we find for the complex conjugate
∗
z ∗ = reiϕ = r(cos ϕ + i sin ϕ)∗ = r (cos ϕ − i sin ϕ) = r (cos (−ϕ) + i sin (−ϕ)) = re−iϕ

which as a side result gives

∗
(ez )∗ = e(z) .

A major advantage of the exponential notation in Eq. (2.6) is the simplification of multipli-
cation of complex numbers

z1 z2 = r1 r2 eiϕ1 eiϕ2 = r1 r2 ei(ϕ1 +ϕ2 ) . (2.11)

From this we conclude, that multiplication with eiϕ2 corresponds to a rotation of the vector
in the complex plane by ϕ2 . Therefore, the multiplication of a complex number z with the
π
imaginary unit i = ei 2 results in

iz = ireiϕ = rei 2 eiϕ = ei(ϕ+ 2 )

π π
(2.12)

Structural Dynamics 2 Description of harmonic oscillations

summer term 2019 9
which can be interpreted as a rotation of π
2
or 90◦ in the complex plane.

Im Im
π
i = ei 2
z1 = eiϕ1 z = eiϕ
π
iz = ei(ϕ+ 2 )
ϕ1 + π
z2 = eiϕ2 π
2
ϕ2
ϕ1 Re 2 ϕ1 Re

ϕ1 + ϕ2

z1 z2

Figure 2.2: Visualization of multiplication of two complex numbers. The left hand side shows the
multiplication of two complex numbers with unit amplitude |z1 | = |z2 | = 1. The resulting
complex number z1 · z2 is the complex number that is rotated by ϕ1 + ϕ2 in the complex
plane. The special case of z2 = i is depicted on the right hand side. In the complex plane,
the purely imaginary number z2 = i is rotated by π2 with respect to the real axis. Thus,
multiplication with i always results in rotation of π2 .

For the multiplication of two complex numbers, we find the following identities for the absolute
value and argument of the resulting complex number
 
|z1 z2 | = r1 r2 eiϕ1 eiϕ2  = r1 r2 = |z1 ||z2 | (2.13)
 

arg z1 z2 = arg r1 r2 eiϕ1 eiϕ2 = arg r1 r2 ei(ϕ1 +ϕ2 ) = ϕ1 + ϕ2 . (2.14)

2.2 Application to time harmonic oscillations

We now want to use the above results to find descriptions for arbitrary time-harmonic signals.
From the harmonic addition theorem we know that sums of sine and cosine functions with
arbitrary amplitudes and phase shifts but the same frequency can be reduced to a single cosine
or sine function. To reduce complexity, we consider a signal f (t) that is composed of a cosine
and sine function with amplitudes f01 and f02 , respectively.

f (t) = f01 cos (ωt) − f02 sin (ωt) (2.15)

where t denotes time, and ω is the radial frequency ω = 2πf . The subtraction of the sine term
is explained later on.

Structural Dynamics 2 Description of harmonic oscillations

summer term 2019 10
 y x,y

yP = sin(ωt)
yP P

π 3
P0 x 2 π 2
π 2π ωt
xP
ωt
xP = cos(ωt)

Figure 2.3: Point on unit circle. The location of a point on the unit circle can be given in Cartesian
coordinates (xP ,yP ) or polar coordinates through the angle ωt. If we express (xP ,yP ) in
terms of ωt, we note that the coordinates xP and yP vary harmonically.

Consider now a unit-circle in the x-y-plane. As the circle has the radius 1, every point on the
circle can be solely described by an angle ωt. Furthermore, a point P can also be characterized
in terms of its Cartesian coordinates (xP ,yP ). This is depicted in Fig. 2.3, from where we also
deduce

xP = cos (ωt) (2.16)

yP = sin (ωt) . (2.17)

The x- and y-coordinates are also depicted in Fig. 2.3 in dependency on the angle ωt. We can
also interpret the coordinates as the projection of the vector P on x- and y-axis, respectively.
Here, we find the two parts, that make up our signal from Eq. (2.15), up to the amplitudes
f01 and f02 . Using the harmonic addition theorems

cos (α ± β) = cos (α) cos (β) ∓ sin (α) sin (β) (2.18)
sin (α ± β) = sin (α) cos (β) ± cos (α) sin (β) (2.19)

we further introduce the identities

f01 cos (ωt) − f02 sin (ωt) = f0 [cos (ωt) cos (ϕ) − sin (ωt) sin (ϕ)] = f0 cos (ωt + ϕ) (2.20)
f02 cos (ωt) + f01 sin (ωt) = f0 [cos (ωt) sin (ϕ) + sin (ωt) cos (ϕ)] = f0 sin (ωt + ϕ) (2.21)

with

f01 = f0 cos (ϕ) (2.22)

f02 = f0 sin (ϕ) (2.23)

Structural Dynamics 2 Description of harmonic oscillations

summer term 2019 11
or inversely
q
f0 = 2
f01 + f02
2
(2.24)
ϕ = atan2 (f02 ,f01 ) . (2.25)

With this, we can equivalently state a harmonic signal using a single phase-shifted sine or
cosine term. Hereinafter, we choose the representation using f0 cos (ωt + ϕ) as it complies
with Eq. (2.15). Further, we define ϕ∗ as the complementary angle to ϕ, such that ϕ + ϕ∗ = π2
and state Eq. (2.21) in terms of ϕ∗ as

f0 sin (ωt + ϕ∗ ) = f0 [cos (ωt) sin (ϕ∗ ) + sin (ωt) cos (ϕ∗ )] (2.26)

Using the identities

π
 
sin ϕ = sin
∗
− ϕ = cos ϕ (2.27)
2
π
 
cos ϕ = cos
∗
− ϕ = sin ϕ (2.28)
2

to further transform Eq. (2.26), it reads

f0 sin (ωt + ϕ∗ ) = f0 [cos (ωt) sin (ϕ∗ ) + sin (ωt) cos (ϕ∗ )] = (2.29)
= f0 [cos (ωt) cos (ϕ) + sin (ωt) sin (ϕ)] = (2.30)
= f01 cos (ωt) + f02 sin (ωt) (2.31)
= f0 cos (ωt − ϕ) . (2.32)

The same can be stated for the following.

f0 cos (ωt + ϕ∗ ) = −f0 sin (ωt − ϕ) . (2.33)

Both, a sine and cosine with negative phase shift ϕ are depicted in Fig. 2.4. In the next step,
we represent the harmonic signals using complex arithmetic. This shall be discussed in the
following section.

2.2.1 Complex representation

We now consider the complex number z = x+iy. From Sec. 2.1 we know that we can represent
z using Euler’s formula. In the following we want to show that we can represent the signal
from Eq. (2.15) by a complex exponential and its complex conjugate. For that we separately

Structural Dynamics 2 Description of harmonic oscillations

summer term 2019 12
 y x,y
yP = sin(ωt − ϕ)

xP 0

ϕ π 3
xP x 2 π 2
π 2π ωt

yP P
ωt
yP0
P0
ϕ
xP = cos(ωt − ϕ)
Figure 2.4: Point on unit circle (f0 = 1) with phase shift

apply some operations to the sine and cosine term. First, the term 12 f01 (i sin (ωt) − i sin (ωt))
is added to f01 cos (ωt), as it does not alter the equation.

1 1
 
f01 cos (ωt) = f01 (cos (ωt) + i sin (ωt)) + (cos (ωt) − i sin (ωt)) (2.34)
2 2
1 h i
= f01 e iωt + e −iωt . (2.35)
2

We note, that we can easily apply Euler’s formula to convert the expression to the complex
exponential form. The same is attempted for the sine part. To this end, we add 12 f02 (i cos(ωt)−
i cos(ωt)) to f02 sin (ωt).

1 1
 
f02 sin (ωt) = f02 (sin (ωt) + i cos (ωt)) + (sin (ωt) − i cos (ωt)) . (2.36)
2 2

Now, the conversion to the complex exponential form requires some more alterations to the
above equation. For this purpose, we factor the imaginary unit i out of the right hand side.

1 sin (ωt) 1 sin (ωt)

" ! !#
= if02 + cos (ωt) + − cos (ωt) . (2.37)
2 i 2 i

We further use that division by the complex number i can be rewritten by multiplication with
−i due to the fact that 1i = ii2 = −1
i
= −i.

1 1
 
= if02 (cos (ωt) − i sin (ωt)) − (cos (ωt) + i sin (ωt)) . (2.38)
2 2

Structural Dynamics 2 Description of harmonic oscillations

summer term 2019 13
Now, we can again apply Euler’s formula to obtain

1 h i
= −i f02 e iωt − e −iωt . (2.39)
2

To obtain a full representation of Eq. (2.15), we combine results from Eqs. (2.35) and (2.39).

f (t) = f01 cos (ωt) − f02 sin (ωt) (2.40)

1 h i 1 h i
= f01 e iωt + e −iωt + i f02 e iωt − e −iωt (2.41)
2 2
1 1
= [f01 + if02 ] e iωt + [f01 − if02 ] e −iωt (2.42)
|2 {z } |2 {z }
fb+ fb−

f0 iϕ iωt f0 −iϕ −iωt

= fb+ e iωt + fb− e −iωt = e e + e e . (2.43)
2 2

We note, that an arbitrary harmonic signal with circular frequency ω is composed of two
complex exponential parts, that are complex conjugates of each other
 ∗  ∗
fb+ e iωt = fb+∗ e iωt = fb− e −iωt (2.44)

as

fb+∗ = fb− (2.45)

 ∗
e iωt = e −iωt . (2.46)

Depending on whether we have a pure sine or cosine signal we can state for the coefficients fb+
and fb− :
pure cosine signal → fb+ and fb− are real
pure sine signal → fb+ and fb− are imaginary

Structural Dynamics 2 Description of harmonic oscillations

summer term 2019 14
 Im Im Im

fb+ = f01 + if02

fb+ = if02
fb+ = f01 Re Re Re

(a) Pure cosine, real (b) Pure sine, imagi- (c) Mixed, arbitrary
fb+ nary fb+ complex fb+
Figure 2.5: Representation of complex pointers with different contributions of cosine and sine part.
The left hand side complex number fb+ only has contributions of the cosine part f01 and
is therefore purely real. Vice versa the middle illustration shows the complex number that
only consists of a sine contribution f02 and is therefore purely imaginary. In a general
case, the complex pointer shows an arbitrary orientation, depending on f01 and f02 .

Im

1
z1
Re z2
7π
1 π
3π
2 3π 2 ωt
π 2π 5π 4π
2 2
−1
z3

−1
Figure 2.6: The complex exponential function evolves harmonically in the complex plane over time. It
is z1 = e iωt (—); z2 = Re(z1 ) = cos(ωt), (—); z3 = i Im(z1 ) = i sin(ωt), (—).

Structural Dynamics 2 Description of harmonic oscillations

summer term 2019 15
3 Fourier transformation

3.1 Periodic signals

At first deterministic signals are considered. We consider random signals in Cha.13. We

distinguish between transient signals that are only present within a limited time windows, and
periodic signals that have a repetition frequency (period). After the time T a periodic signal
f (t) repeats, such that it holds:

f (t) = f (t ± nT ) . (3.1)

This family of signals can be represented by a series of sine and cosine functions. This repre-
sentation is called Fourier series. Subsequently we discuss how to determine such representa-
tions.

3.1.1 Representation with real coefficients

An arbitrary periodic function can be represented by a Fourier series under the condition that
it is integrable on one period [0,T ], i.e. the integral

ZT
f (t) dt (3.2)
0

exists. Then the Fourier series reads

∞
a0 X
f (t) = + (ak cos (kω0 t) + bk sin (kω0 t)) (3.3)
2 k=1

with ωk = kω0 and ω0 = 2π T

. For the above equality the series has to converge for k → ∞.
The coefficients ak , bk are determined such that the mean-square error ε between the given

Structural Dynamics
summer term 2019
16
function f (t) and the series representation is minimal. The error reads

ZT ∞
!2
a0 X
ε= f (t) − − (ak cos (ωk t) + bk sin (ωk t)) dt . (3.4)
2 k=1
0

In order to find an optimal solution we differentiate the error ε with respect to the unknown
coefficients and set the resulting expressions to zero:

∂ε !
=0 (3.5)
∂aj
∂ε !
= 0. (3.6)
∂bj

Applying the chain rule, and neglecting the case j = 0 at first, we find

T ∞
!
∂ε Z
a0 X
= −2 f (t) − − (ak cos (ωk t) + bk sin (ωk t)) cos (ωj t) dt = 0 (3.7)
∂aj 2 k=1
0
T ∞
!
∂ε Z
a0 X
= −2 f (t) − − (ak cos (ωk t) + bk sin (ωk t)) sin (ωj t) dt = 0 . (3.8)
∂bj 2 k=1
0

For j,k > 0, we find that

ZT

0 j=6 k
sin(ωk t) sin(ωj t) dt =  T (3.9)
0 2
j = k 6= 0,
ZT

0 j 6= k
cos(ωk t) cos(ωj t) dt =  T (3.10)
0 2
j = k 6= 0,
ZT
cos(ωk t) sin(ωj t) dt = 0, (3.11)
0

ZT ZT
cos(ωj t) dt = sin(ωj t) dt = 0. (3.12)
0 0

Thus, Eqs. (3.7) and (3.8) reduce to

 T 
Z
T
−2  f (t) cos (ωk t) dt − ak  =0 (3.13)
2
0

Structural Dynamics 3 Fourier transformation

summer term 2019 17
  T 
Z
T
−2  f (t) sin (ωk t) dt − bk  = 0. (3.14)
2
0

The coefficients ak , bk can then be determined:

T
2Z
ak = f (t) cos (ωk t) dt (3.15)
T
0
T
2Z
bk = f (t) sin (ωk t) dt . (3.16)
T
0

For the case k = 0, we have

T ∞
1
!
∂ε Z
a0 X
=2 f (t) − − (ak cos (ωk t) + bk sin (ωk t)) dt = 0 . (3.17)
∂a0 2 k=1 2
0

The integrals over the sine and cosine functions vanish, so that it follows

ZT
a0
f (t) dt − T = 0. (3.18)
2
0

The coefficient a0 then reads

T
2Z
a0 = f (t) dt . (3.19)
T
0

The Fourier coefficients are completely determined by Eqs. (3.19), (3.15) and (3.16). Since
Eq. (3.15) is equivalent to Eq. (3.19) for k = 0 since cos(0) = 1, the expression for a0 is often
not explicitly stated.

Structural Dynamics 3 Fourier transformation

summer term 2019 18
 Example 3.1

Consider the following wave f (t) with amplitude f0 and period T .

f
f0

The considered signal is an odd function, i.e., f (t) = −f (−t). Through this, we can
immediately state that all ak = 0 and thus all cosine terms in the series vanish. This
leaves us with the evaluation of the coefficients bk . Using Eq. (3.8), we find:
 T 
T
2 Z2 Z
bk =  f0 sin(ωk t) dt + (−f0 ) sin(ωk t) dt
 
T  
0 T
2
T
4 1 2
= − f0 cos(ωk t)

T ωk 
0

 4f0 for k = 1,3,5, . . .
=  kπ
0 for k = 0,2,4, . . .

Then, the Fourier series representation of f (t) reads

4f0 ∞
1
f (t) = sin (ωk t) .
X
π k=1,3,... k

Structural Dynamics 3 Fourier transformation

summer term 2019 19
3.1.2 Representation with complex coefficients

In general it is easier to represent a periodic function by a Fourier series in complex notation

∞
f (t) = cbk e iωk t . (3.20)
X

k=−∞

Therein the coefficients cbk are determined by

1ZT
cbk = f (t)e −iωk t dt . (3.21)
T 0

By application of Euler’s formula e iϕ = cos(ϕ) + i sin(ϕ), we find

1ZT 1ZT 1
cbk = f (t)e −iωk t dt = f (t) (cos(iωk t) − i sin(iωk t)) dt = (ak − ibk ) (3.22)
T 0 T 0 2

where the last step can be found by comparing with Eqs. (3.15) and (3.16). Furthermore for
the coefficient with negative k we find:

1
cb−k = (ak + ibk ) . (3.23)
2

Thus, cbk and cb−k are complex conjugate:

cb−k = cb∗k . (3.24)

For k = 0 it holds
a0
cb0 = . (3.25)
2

We can check the identity of the real and complex Fourier representations by inserting
Eqs. (3.22) and (3.23) into Eq. (3.20):
∞ ∞
f (t) = cbk e iωk t = cbk e iωk t + cb∗k e −iωk t . (3.26)
X X

k=−∞ k=1

Structural Dynamics 3 Fourier transformation

summer term 2019 20
We split the summation from −∞ to ∞ by applying the knowledge that the coefficients for
negative k are the complex conjugate values of the coefficients with positive k. 1

∞
1
f (t) = cb0 e 0t + (abk − ibk ) (cos(ωk t) + i sin(ωk t)) +
X

k=1 2
∞
1
+ (abk + ibk ) (cos(ωk t) − i sin(ωk t)) . (3.27)
X

k=1 2

This can finally be shown to be

∞
a0 X
f (t) = (ak cos(ωk t) + bk sin(ωk t)) (3.28)
2 k=1

which is equal to Eq. (3.22).

Im(cbk ) Re(cbk ) Im(cbk )

cb−2
cb4 cb−3
cb−1
Re cbk r r

cb1
cb−4 cb3
cb2
even odd

Figure 3.1: Graphical representation of complex Fourier series coefficients. Here, exemplarily three
terms are depicted in the complex plane, and the corresponding discrete spectrum over
r for real and imaginary part of the b
ck are given. Due to the complex conjugate property
of the complex coefficients, we can easily deduce that the real parts are always be even
with respect to r while the imaginary parts are always odd with respect to r.

The following expression summarize the relations between the coefficients of the real and
complex representation of the Fourier series.

a0 1 1
cb0 = cbk = (ak − ibk ) cb−k = (ak + ibk ) (3.29)
2 2 2
a0 = 2cb0 ak = cbk + cb−k bk = i (cbk − cb−k ) . (3.30)
1
Here we see that due to the form of the Fourier series, where the sine term is added rather than subtracted,
as was introduced in the representation of harmonic signals in Cha. 2, the coefficient of the positive time
exponential function has a negative inaginary part.

Structural Dynamics 3 Fourier transformation

summer term 2019 21
3.2 Fourier integral

The Fourier transformation can be interpreted as a limiting case of the Fourier series for
signals with an infinite period T → ∞. By this non periodic functions the representation of
non-periodic functions can be approached. We insert the definition of the coefficients cbk from
Eq. (3.21) into the Fourier series Eq. (3.20) and obtain

∞
1
Z T !
f (t) = e ikω0 t
f (t)e −ikω0 t
dt (3.31)
X

k=−∞ 0 T

and introduce ∆ω = ω0 = 2π
T
. With

2π
lim ∆ω = lim = dω (3.32)
T →∞ T →∞ T

k∆ω → ω (3.33)

one finds for the limiting case T → ∞:

∞ ∞
 
1 Z Z
f (t) = f (t)e −iωt dt e iωt dω
2π
−∞ −∞

and further
Z∞
f˜(ω) = f (t)e −iωt dt (3.34)
−∞
∞
1 Z ˜
f (t) = f (ω)e iωt dω . (3.35)
2π
−∞

Eq. (3.34) is called Fourier transform (FT), Eq. (3.35) is called inverse Fourier transform
(IFT).
An alternative representation using f = ω
2π
is given by

Z∞
g̃(f ) = g(t)e −i2πf t dt (3.36)
−∞
Z∞
g(t) = g̃(f )e i2πf t df . (3.37)
−∞

Structural Dynamics 3 Fourier transformation

summer term 2019 22
The functions f (t) and f˜(ω) and g(t) and g̃(f ) are called a Fourier transform pair.
The Fourier ˜
 transform can also be written in operator notation as F (f (t)) = f (ω) and
F −1 f˜(ω) = f (t). Often, the Fourier transform of f (t) is denoted by an uppercase F {f } =
F (ω). We use the introduced notation to avoid confusion with the convention that random
variables are denoted by an uppercase letter (comp. Chapter 13).

Example 3.2: Fourier transform of the rectangular function

The following rectangular function is given


g0
 −T0 < t < T0
t
  
g(t) = g0 Π = g0 /2 t = ±T0
T0 
0 else



where Π(t) denotes the unit rectangular function. By applying Eq. (3.34), we find the
Fourier transform of g̃(t).

ZT0
g̃(f ) = g0 e −i2πf t dt =
−T0
   
ZT0 ZT0
= g0  cos(2πf t) dt − ig0  sin(2πf t) dt
   

−T0 −T0
| {z }
=0

sin(2πf T0 )
= 2g0 T0 .
2πf T0

The function sin(2πf T0 )

2πf T0
is also known as the sinus cardinalis function sinc(2πf T0 ). The
value at f = 0, g̃(0) = 2g0 T0 corresponds to the area under the signal in the time
domain. The Fourier transform pair g(t) and g̃(f ) is depicted in Fig. 3.2

Structural Dynamics 3 Fourier transformation

summer term 2019 23
 g(t) g̃(f )
I = 2g0 T0
2g0 T0
g0

t f
−T0 T0 1
2T0
1
T0
3
2T0
 
t
Figure 3.2: Fourier transform of the rectangular function g(t) = g0 Π T0 . The Fourier
transform g̃(f ) has the maximum amplitude 2g0 T0 at f = 0. Zero crossings
n
are given by f = 2T 0
.

If we now alter the function g(t) such that the signal is half as long with T00 = T0
2
, but
with double amplitude h0 = 2g0 , such that

2g0

− T20 < t < T0
2t

2
!
t
  
g 0 (t) = h0 Π = 2g0 Π = g0 t = ± T20
T00 T0 
0 else



we find the Fourier transform g̃ 0 (f ), which is depicted in Fig. Eq. (3.3).

I 0 = 2h0 T00
= 2g0 T0
g 0 (t) g̃ 0 (f )

h0 = 2g0 2h0 T00 = 2g0 T0

t f

T00 = T0
2
1
2T00
= 2
2T0
= 1
T0
 
Figure 3.3: Fourier transform of the rectangular function g 0 (t) = h0 Π t
T00 with h0 = 2g0
and T00 = T20 . The Fourier transform g̃ 0 (f ) has the maximum amplitude I 0 =
2h0 T00 = 2g0 T0 = I at f = 0. Zero crossings are given by f = Tn0 .

Through shortening the signal in the time domain, the Fourier transform becomes
wider in the frequency domain. The first zero crossing of g̃ 0 (f ) is twice as high as for

Structural Dynamics 3 Fourier transformation

summer term 2019 24
 the original signal g̃(f ). The maximum amplitude does not change, as the area under
the signal does not change either. In the limit case T0 → 0 the Fourier transform
results in
Z∞
I = 2g0 T0 δ (t) dt = 2g0 T0
−∞

with the Dirac-delta function defined, such that

δ(t) = 0 for t 6= 0
Z∞
δ(t) dt = 1 .
−∞

The Fourier transform of a Dirac impulse becomes a constant function in the frequency
domain. This is depicted in Fig. 3.4.

g 0 (t) g̃ 0 (f )

2g0 T0 δ(t) 2g0 T0

t f

Figure 3.4: Fourier transform of the impulse function g 0 (t) = g0 T0 δ(t). The Fourier trans-
form g̃ 0 (f ) is the constant function with g̃ 0 (f ) = 2g0 T0 .

Example 3.3: Fourier transform of the sine function

We consider the sine function

g(t) = g0 sin(2πf0 t) (3.38)

and its Fourier transform

g0
g̃(f ) = i [−δ(f − f0 ) + δ(f + f0 )] (3.39)
2

Structural Dynamics 3 Fourier transformation

summer term 2019 25
 which is taken without further proof. The Fourier transform pair g(t), g̃(f ) is depicted
in Fig. 3.6.

g̃(f ) g(t)
i g20 δ(f + f0 )
sin(2πf0 t)
g0
f0
f t
−f0

1
−i g20 δ(f − f0 ) f0

Figure 3.5: Fourier transform of the sine function g(t) = g0 sin(2πf0 t).

We can show that g̃(f ) (Eq. (3.39)) is the Fourier transform of g(t) (Eq. (3.38)), by
applying the inverse Fourier transformation (Eq. (3.37)) to Eq. (3.39).

Z∞
g(t) = g̃(f )e i2πf t df =
−∞
∞ Z∞
 
ig0  Z
= −δ(f − f0 )e i2πf t
df + δ(f + f0 )e i2πf t df  =
2
−∞ −∞

ig0 h i
= −e i2πf0 t + e −i2πf0 t =
2
ig0
= [− cos(2πf0 t) − i sin(2πf0 t) + cos(2πf0 t) − i sin(2πf0 t)] =
2
ig0
= [−2i sin(2πf0 t)] = g0 sin(2πf0 t)
2

which is the signal we defined in Eq. (3.38).

Structural Dynamics 3 Fourier transformation

summer term 2019 26
 Example 3.4: Fourier transform of the cosine function

We consider the cosine function

g(t) = g0 cos(2πf0 t) (3.40)

and its Fourier transform

g0
g̃(f ) = i [δ(f − f0 ) + δ(f + f0 )] (3.41)
2

which is taken without further proof. The Fourier transform pair g(t), g̃(f ) is depicted
in Fig. 3.6.

g̃(f ) g(t)
g0
2
δ(f + f0 )
cos(2πf0 t)
g0
δ(f − f0 ) g0
2

f t
−f0 f0

1
f0

Figure 3.6: Fourier transform of the cosine function g(t) = g0 cos(2πf0 t).

We can show that g̃(f ) (Eq. (3.39)) is the Fourier transform of g(t) (Eq. (3.38)), by
applying the inverse Fourier transformation (Eq. (3.37)) to Eq. (3.41).

Z∞
g(t) = g̃(f )e i2πf t df =
−∞
∞ Z∞
 
ig0  Z
= δ(f − f0 )e i2πf t
df + δ(f + f0 )e i2πf t df  =
2
−∞ −∞

ig0 h i2πf0 t i
= e + e −i2πf0 t =
2
ig0
= [cos(2πf0 t) + i sin(2πf0 t) + cos(2πf0 t) − i sin(2πf0 t)] =
2

Structural Dynamics 3 Fourier transformation

summer term 2019 27
 ig0
= [2 cos(2πf0 t)] = g0 cos(2πf0 t)
2

which is the signal we defined in Eq. (3.40).

Structural Dynamics 3 Fourier transformation

summer term 2019 28
4 Setting up the equation of motion

4.1 Single degree of freedom system

The single degree of freedom (SDOF) system plays a significant role in many problems in
structural dynamics. This importance stems from the fact, that many engineering systems
or system components can be approximated by SDOF systems. The mechanical behaviour is
described by the equation of motion, which is an ordinary differential equation. We summarize
important methods to obtain this differential equation in the following.

4.1.1 Equilibrium of forces

The equation of motion of the SDOF system follows from the equilibrium of forces. They are
composed out of the external forces, the gravity force, and the reaction forces, that result from
the motion of the system.
Additional to the static reaction forces (restoring force), we consider the d’Alembert inertia
forces and, possibly, damping forces. The equations of motion are determined upon the de-
flected system applying the principle of d’Alembert. To this end, a free body cut is applied
around the mass point of the SDOF system, as depicted in Fig. 4.1.

fe fe - external force

fg fg - gravity force
w,ẇ, ẅ

fK fK - elastic restoring force

fC - damper force
fC
(formulated as velocity-proportional in the following)
fI fI - d’Alembert inertia force

Figure 4.1: External and reaction forces on single mass point.

Structural Dynamics
summer term 2019
29
Thus, the equilibrium results in

−fK − fC − fI + fe + fg = 0. (4.1)

Inserting the expressions for the single forces, we obtain

mẅ (t) + cẇ (t) + kw (t) = fe (t) + fg . (4.2)

For the free vibration problem, i.e., fe (t) = 0 and neglecting the gravity force, the equation of
motion simplifies to

mẅ (t) + cẇ (t) + kw (t) = 0. (4.3)

We obtain a differential equation, where together with the displacement w(t), also the first two
derivatives with respect to time, namely the velocity ẇ(t) and the acceleration ẅ(t), appear.

Example 4.1: Examples for SDOF systems

«««< .mine «««< .mine In the following figure, some examples for SDOF models de-
scribing real systems are depicted.

w
m
m
µ=0 µ=0
EI l
wc k m EI

w
a) b) c)

w l GIT
m

EI µ=0 Iθ
kφ
φ

d) e)

Structural Dynamics 4 Setting up the equation of motion

summer term 2019 30
 Cases a) to d): Case e):

mẅ + cẇ + kw = f (t) Iθ φ̈ + cφ φ̇ + kφ φ = m(t)

with w(t) the displacement. The char- with φ(t) the rotation. The character-
acteristic parameters are istic parameters are

m . . . mass Iθ . . . Rotatory intertia

c . . . damping constant cφ . . . torsional damping constant
k . . . stiffness kφ . . . torsional stiffness

For example, for c), the stiffness term For example, for e), the stiffness term
k is k = l3 .
3EI
kφ is kφ = GIl t .
||||||| .r59 In the following figure, some examples for SDOF models describing real sys-
tems are depicted.

w
m
m
µ=0 µ=0
EI l
wc k m EI

w
a) b) c)

w l GIT
m

EI µ=0 Iθ
kφ
φ

d) e)

Structural Dynamics 4 Setting up the equation of motion

summer term 2019 31
 Cases a) to d): Case e):

mẅ + cẇ + kw = f (t) Iθ φ̈ + cφ φ̇ + kφ φ = m(t)

with w(t) the displacement. The char- with φ(t) the rotation. The character-
acteristic parameters are istic parameters are

m . . . mass Iθ . . . Rotatory intertia

c . . . damping constant cφ . . . torsional damping constant
k . . . stiffness kφ . . . torsional stiffness

For example, for c), the stiffness term For example, for e), the stiffness term
k is k = l3 .
3EI
kφ is kφ = GIl t .
======= In the following figure, some examples for SDOF models describing real
systems are depicted.

w
m
m
µ=0 µ=0
EI l
wc k m EI

w
a) b) c)

w l GIT
m

EI µ=0 Iθ
kφ
φ

d) e)

Structural Dynamics 4 Setting up the equation of motion

summer term 2019 32
 Cases a) to d): Case e):

mẅ + cẇ + kw = f (t) Iθ φ̈ + cφ φ̇ + kφ φ = m(t)

with w(t) the displacement. The char- with φ(t) the rotation. The character-
acteristic parameters are istic parameters are

m . . . mass Iθ . . . Rotatory intertia

c . . . damping constant cφ . . . torsional damping constant
k . . . stiffness kφ . . . torsional stiffness

For example, for c), the stiffness term For example, for e), the stiffness term
k is k = l3 .
3EI
kφ is kφ = GIl t .
»»»> .r61 ||||||| .r51 In the following figure, some examples for SDOF systems are
depicted.

w
m
m
µ=0 µ=0
EI l
wc k m EI

w
a) b) c)

w l GIT
m

EI µ=0 Iθ
kφ
φ

d) e)

Structural Dynamics 4 Setting up the equation of motion

summer term 2019 33
 Cases a) to d): Case e):

mẅ + cẇ + kw = F (t) Iθ φ̈ + cφ φ̇ + kφ φ = M (t)

with w(t) the displacement. The char- with φ(t) the rotation. The character-
acteristic parameters are istic parameters are

m . . . mass Iθ . . . Rotatory intertia

c . . . damping constant cφ . . . torsional damping constant
k . . . stiffness kφ . . . torsional stiffness

For example, for c), the stiffness term For example, for e), the stiffness term
k is k = l3 .
3EI
kφ is kφ = GIl t .
======= In the following figure, some examples for SDOF models describing real
systems are depicted.

w
m
m
µ=0 µ=0
EI l
wc k m EI

w
a) b) c)

w l GIT
m

EI µ=0 Iθ
kφ
φ

d) e)

Structural Dynamics 4 Setting up the equation of motion

summer term 2019 34
 Cases a) to d): Case e):

mẅ + cẇ + kw = f (t) Iθ φ̈ + cφ φ̇ + kφ φ = m(t)

with w(t) the displacement. The char- with φ(t) the rotation. The character-
acteristic parameters are istic parameters are

m . . . mass Iθ . . . Rotatory intertia

c . . . damping constant cφ . . . torsional damping constant
k . . . stiffness kφ . . . torsional stiffness

For example, for c), the stiffness term For example, for e), the stiffness term
k is k = 3EI
l3
. kφ is kφ = GIl t .
»»»> .r61

4.1.2 Virtual work

Analogously the equations of motions can be obtained using the principle of d’Alembert, where
a suitable virtual displacement is imposed on the system. Then the principle of virtual work
is written. The general procedure is shown by means of the following example.

Example 4.2: Setting up the ODE using the principle of virtual work

The following system is excited by a dynamic force F (t). Gravity forces are not con-
sidered.

f (t)
m
starr

c k

l/2 l/2 l/2 l/2 l/2

The system is kinematic and has one degree of freedom. The system in the deformed
state is depicted below.

Structural Dynamics 4 Setting up the equation of motion

summer term 2019 35
 wm
wc ϕ2
ϕ1 wk
wG

The kinematic relations are:

l ϕ1
wG = ϕ1 = ϕ2 l → = ϕ2
2 2
wm = ϕ1 l
l wm
wc = ϕ 1 =
2 2
l wm
wF = wG = ϕ1 =
2 2
l ϕ1 l 1 wm
wk = ϕ2 = = ϕ1 l = .
2 2 2 4 4

Hence the following forces act on the system:

f (t)
wm
wc
wk
wf
c ẇc
m ẅm k wk

Next, a virtual displacement is superimposed upon the deformed system.

δwm
δwc
f (t)
wm
wc
wk
wf
c ẇc
δwk
m ẅm
δwG k wk

Structural Dynamics 4 Setting up the equation of motion

summer term 2019 36
 For the system to be in equilibrium, the sum of virtual work of all forces has to vanish,
i.e. δW = 0. Thus,

−mẅm δwm − cẇc δwc − kwk δwk + f (t)δwF = 0

ẇm δwm wm δwm δwm
mẅm δwm + c +k = f (t) .
2 2 4 4 2

With δwm 6= 0 it follows

1 1 1
mẅm + cẇm + kwm = f (t) .
4 16 2

This equation corresponds to the differential equation of the system with respect to the
vertical displacement of the point mass m.

4.1.3 The principle of Hamilton

In the following we shall introduce the principle of Hamilton. Also refer to the lecture on
Continuum Mechanics.

4.1.3.1 Undamped systems

Alternatively to the principle of virtual work, the equations of motion can be derived through
the principle of Hamilton. As derived in the course Continuum Mechanics, Hamilton’s principle
states that for conservative systems (where forces have a potential) among all kinematic-
ally possible movements the movement will arise that minimizes the time integral over the
Lagrange-function

L(t) = T (t) − U (t). (4.4)

with the kinetic energy Ekin (t) = T (t) and the potential energy Epot (t) = U (t). For fixed
conditions at t0 und t1 ,

Zt1
I= L (w,ẇ) dt = extremum . (4.5)
t0

Structural Dynamics 4 Setting up the equation of motion

summer term 2019 37
 w(t)

δw

t
t0 t1

Example 4.3

An undamped SDOF system is deflected from its static position to initial conditions, and
is then released. There are infinitely many possibilities, which can satisfy the boundary
conditions. For each of these possibilities, the velocity ẇ, can be determined at each time
t. Thereby also the Lagrange function can be determined and the above integral can be
evaluated. The function, for which the integral gives an absolute extremum (minimum),
i.e., the integral that is smaller than any other possible integrals, is the function that
will really occur. As a second boundary condition we consider the observation that the
SDOF system shows the static displacement after the time t = T /4, i.e., w( T4 ) = 0.
The system parameters are chosen as follows.

m = 1, k = 1, ω = 1, T = 2π .

Initial conditions:

w0 = w(t = 0) = −1 vstart = v(t = 0) = 0 .

We investigate three different approaches for the displacement function w(t):

• 1. approach: quadratic parabola
• 2. approach: cosine function
We now evaluate the Lagrange function for the three approaches:
• 1. approach. We choose the following quadratic parabola, complying with our
observation at t = T4 ,

w0
 
w(t) = w0 − 4 2 t2 ,
π

Structural Dynamics 4 Setting up the equation of motion

summer term 2019 38
 and derive the velocity, kinetic and potential energy, and the Lagrange funtion.

dw(t) w0
 
v(t) = =− 8 2 t
dt π
1 1 w0 2
 
T (t) = mv(t)2 = m 8 2 t2
2 2 π
1 1 w0 2 2
   
U (t) = kw(t) = k w0 − 4 2 t
2
2 2 π
L(t) = T (t) − U (t) .

The evaluation of Eq. (4.5) thus reads

T
Z4
L(t) dt = 5,534 · 10−3 .
0

1 2
T(t) 0.5 1
w(t)
U(t) 0 v(t) 0
L(t)
-0.5 -1
0 0.5 1 1.5 2 0 0.5 1 1.5 2
• 2. approach. We choose the following cosine function, again complying with our
observation at t = T4 ,
s 
k 
w(t) = w0 cos  t
m

and derive the velocity, kinetic and potential energy, and the Lagrange function.
s s 
dw(t) k k 
v(t) = = −w0 sin  t
dt m m
s 
1 1 k k 
T (t) = mv(t)2 = mw02 sin2  t
2 2 m m
s 
1 1 k 
U (t) = kw(t)2 = kw02 cos2  t
2 2 m

Structural Dynamics 4 Setting up the equation of motion

summer term 2019 39
 L(t) = T (t) − U (t) .

The evaluation of Eq. (4.5) thus reads

T
Z4
L(t) dt = 0 .
0

0.5 1
T(t)
0 w(t) 0
U(t)
v(t)
L(t)
-0.5 -1
0 0.5 1 1.5 2 0 0.5 1 1.5 2
t t

The computation calculus for determining the function w(t) is the variational calcu-
lus (comp. continuum mechanics). From the variational calculus Eulers differential
equation results as

Zt1
∂U (w) d ∂T (ẇ)
!
δI = − − δw dt = 0 .
∂w dt ∂ ẇ
t0

As δw can be chosen arbitrary, e.g., always such that it has the same sign as the
expression in the braces, the integral can only be equal to zero, if the brace itself
becomes zero. Thus, the differential equation for the SDOF system follows to:
h i h i
1 1
∂U (w) d ∂T (ẇ) ∂ 2
kw2 d ∂ 2 mẇ
2
+ = + = kw + mẅ = 0 .
∂w dt ∂ ẇ ∂w dt ∂ ẇ

4.1.3.2 Introduction of non-conservative forces

Without non-conservative forces (e.g., damping forces or exterior forces that cannot be de-
scribed by a potential), the Hamilton’s principle states

Zt1
I= (T − U ) dt = extremum
t0

Structural Dynamics 4 Setting up the equation of motion

summer term 2019 40
 Zt1 
1 1

mẇ2 − kw2 dt .
2 2
t0

The variation δw (t) can be chosen arbitrary but needs to obey the boundary conditions:

Zt1
(mẇ · δ ẇ − kw · δw) dt .
t0

Integrating parts ( u0 v = − uv 0 + [uv]) results in:

R R

Zt1
(−mẅ · δw − kw · δw) dt + [mẇδw]tt10
| {z }
t0 Principle of virtual work

where δw is chosen to vanish at t = t0 and t = t1 (δw can be chosen arbitrarily).

Apparently, the virtual work, done by inertia and spring forces, can be described via the
variation of kinetic and potential energy T and U . The difference T −U is called the Lagrangian
L of the system.
Thus, the well-known formulation for the equilibrium can be derived:

−mẅ − kw = 0 .

This expression can be expanded by non conservative forces in the context of the principle of
virtual work and the equilibrium:

−mẅ − kw − cẇ + f = 0 .

Principle of virtual work:

(−mẅ − kw − cẇ + f )δw(t) = 0 .

For arbitrary δw(t), one obtains:

Zt2
(−mẅ − kw − cẇ + f )δw dt + [mẇ · δw]tt21 = 0
| {z }
t1
=0
Zt2
[(mẇ · δ ẇ − kw · δw) − cẇ · δw + f · δw] dt = 0
t1

Structural Dynamics 4 Setting up the equation of motion

summer term 2019 41
 Zt2
(δ(T − U ) − cẇδw + f δw) dt = 0
t1

where δw is the variation applied to the system. Thus, Hamilton’s principle can be extended
by the work Wnc of non-conservative forces:

w(t)

δw

t
t0 t1

Zw1
Wnc = (−cẇ + f ) dw .
w0

The variation of Wnc results in:

δWnc = (−cẇ + f )δw

which permits to extend the Lagrangian as:

Zt1
I= (T − U + Wnc ) dt = extremum .
t0

This implies:

Zt1 Zt1
δ L dt + δ Wnc dt = 0 .
t0 t0

Structural Dynamics 4 Setting up the equation of motion

summer term 2019 42
 Example 4.4

f (t)

m
w(t)
c k

Zt1 Zt1
δ L dt + δ Wnc dt = 0
t0 t0

with

L=T −U

and

T kinetic energy
U potential energy
Wnc non-conservative (external forces, damping) work .

• Conservative forces

1
U = kw2 − mgw
2
1
T = mẇ2
2
1 1
L = T − U = mw˙2 − kw2 + mgw
2 2
δWnc = f (t)δw − cẇδw .

Applying Hamilton’s principle delivers:

Zt1 Zt1
1 ˙2 1 2
δ L dt = δ mw − kw + mgw dt
2 2
t0 t0

Structural Dynamics 4 Setting up the equation of motion

summer term 2019 43
 Zt1
= mẇδ ẇ − kwδw + mgδw dt .
t0

• Expanded with non-conservative forces

Zt1 Zt1 Zt1

mẇδ ẇ dt − kwδw dt + (f (t) − cẇ + mg) δw dt = 0
t0 t0 t0

t1 Zt1 Zt1 Zt1

⇒ mẇδw − mẅδw dt − kwδw dt + (f (t) − cẇ + mg) δw dt = 0

t0
t0 t0 t0

t1 Zt1  
⇒ mẇδw + − mẅ − kw + f (t) − cẇ + mg δw dt = 0 .

t0
| {z } t0 | {z }
(a) (b)

The variation δw can be chosen such that (a) disappears, i.e. δw(t0 ) = δw(t1 ) = 0,
according to the derivation of Hamilton’s principle. Furthermore (b) must be zero
independent from δw in order to satisfy the equation.
Thus (b), states the equation of motion.
By applying the variation calculus, the same is obtained by the Euler-Lagrange-equation:

d ∂L(ẇ, w) ∂L(ẇ, w) δWnc

− + + =0
dt ∂ ẇ ∂w δw
d
− mẇ − kw + mg + f (t) − cẇ = 0
dt
− mẅ − kw + mg + f (t) − cẇ = 0 .

4.1.3.3 Lagrange multiplicators

The principle of Hamilton is especially advantageous, as further unknowns can be introduced

in the equations very easily. Often, the number of unknowns does not correspond to the
number of independent degrees of freedom. In these cases, further constraints are formulated,
that can then be considered through the method of Lagrange multipliers.

Structural Dynamics 4 Setting up the equation of motion

summer term 2019 44
 Example 4.5: Chain structure

l1 l5
l2 l3 l4

It is advantageous, to describe the kinetic and potential energy through the nodal
degrees of freedom (4 · 2 = 8 displacements ui , vi , or q1 . . . q8 ). Through the bearing not
all degrees of freedom can be chosen independently, as the lengths of the single chain
links l1 , l2 , . . . , l5 stay constant. Thus, it holds

L = T − U = L(q1 , . . . , q8 , q̇1 , . . . , q̇8 ) .

The geometrical constraints

l(q1 , q2 ) = l1
l(q1 , q2 , q3 , q4 ) = l2
...
l(q7 , q8 ) = l5

are written in the form

f (q1 , q2 ) = l(q1 , q2 ) − l1 = 0
f (q1 , q2 , q3 , q4 ) = l(q1 , q2 , q3 , q4 ) − l2 = 0
...
f (q7 , q8 ) = l(q7 , q8 ) − l5 = 0.

The principle of variation with the additional constraints is formulated as follows:

L̃ = L(q1 , . . . , q8 , q̇1 , . . . , q̇8 ) + λ1 f (q1 , q2 ) + λ2 f (q1 , q2 , q3 , q4 ) + . . . + λ5 f (q7 , q8 )

d ∂ L̃ ∂ L̃
− + = 0, with i = 1 . . . 8
dt ∂ q̇i ∂qi
d ∂ L̃ ∂ L̃ ∂ L̃
− + =0 → = 0, with j = 1 . . . 5
dt ∂ λ̇j ∂λj ∂λj

Structural Dynamics 4 Setting up the equation of motion

summer term 2019 45
 Example 4.6: Rigid beam, one degree of freedom, two unknowns w1 and w2

m
l1 l2
w1 w2
k starr

1 1
T = mẇ22 U = kw12
2 2

Geometric constraint:
w2 w1
=− → f = w2 l1 + w1 l2 = 0
l2 l1
Zt1 
1 1

δ mẇ22 − kw12 + λ(w2 l1 + w1 l2 ) dt = 0
2 2
t0

d ∂ L̃ ∂ L̃
Variation δw2 : − + =0 → −mẅ2 + λl1 = 0 (4.6)
dt ∂ ẇ2 ∂w2
d ∂ L̃ ∂ L̃
Variation δw1 : − + =0 → kw1 + λl2 = 0 (4.7)
dt ∂ ẇ1 ∂w1
d ∂ L̃ ∂ L̃
Variation δλ : − + =0 → w2 l1 + w1 l2 = 0 (4.8)
dt ∂ λ̇ ∂λ

Multiplication of Eq. (4.6) with Eq. (4.7) results in

X 
−ml2 ẅ2 + kl1 w1 = 0 M =0 .

With Eq. (4.8)

w1 l2
w2 = −
l1

follows

ml22 ẅ1 + kl12 w1 = 0.

Structural Dynamics 4 Setting up the equation of motion

summer term 2019 46
 Example 4.7: Elastic beam, Ritz-approach

The application of the principle of Hamilton for discretized, continuous systems is ex-
plained by means of the following example.

m
x
EI,µ EI,µ
k

l
xi

Kinetic energy:

Zl
1 2 1
T = µẇ (x) dx + mẇ2 (xi ) .
2 2
0

Potential energy (without consideration of the gravitational potential energy):

Zl
1 1
U= EI(w00 (x))2 dx + kwi2 .
2 2
0

Boundary condition:

wi = w(xi ) → f = wi − w(xi ) = 0 .

Lagrangian:

Zl 
1 2 1 1 1

L̃ = µẇ (x) − EI(w00 (x))2 dx + mẇi2 − kwi2 + λ (wi − w(xi )) .
2 2 2 2
0

Ritz-approach:
πx
w(A, x, t) = A sin sin (ωt)
l

Structural Dynamics 4 Setting up the equation of motion

summer term 2019 47
 wi = B sin (ωt)
λ = λ0 sin (ωt) .

Furthermore it holds:

Zl
πx 1
ẇ(x) = ωA sin cos (ωt) → ẇ2 dx = lω 2 A2 cos2 (ωt)
l 2
0

Zl
π2 πx 2 1 π4 2 2
w (x) = − 2 A sin
00
sin (ωt) → (w00 (x)) dx = A sin (ωt)
l l 2 l3
0

ẇi = ωB cos (ωt) .

With the help of the Ritz approach, the integral over the Lagrange function (depending
on A, B, and λ0 ) can be expressed through

ZT
L̃ = L̃(A, B, λ0 ) dt =
0

1 1 π4 1 1
" #
πxi T

= µlω 2 A2 − EI 3 A2 + mω 2 B 2 − kB 2 + λ0 B − A sin .
4 4 l 2 2 l 2

Minimization of the integral (principle of Hamilton) is then done by

∂ L̃ 1 1 π4 πxi
= µlω A − EI 3 A − λ0 sin
2
=0 (4.9)
∂A 2 2 l l
∂ L̃
= mω 2 B − kB + λ0 = 0 (4.10)
∂B
∂ L̃ πxi
= B − A sin = 0. (4.11)
∂λ0 l

Inserting Eq. (4.10) into Eq. (4.9) results in

1 1 π4 πxi πxi
µlω 2 A − EI 3 A + mω 2 B sin − kB sin = 0.
2 2 l l l

B can be expressed using Eq. (4.11) (B = A sin πxl i ):

1 1 π4 πxi πxi
µlω 2 − EI 3 + mω 2 sin2 − k sin2 = 0.
2 2 l l l

Structural Dynamics 4 Setting up the equation of motion

summer term 2019 48
 Solving for ω 2 returns the Rayleigh quotient
4
1
EI πl3 + k sin2 πxl i Epot, max
ω ≤2 2
ωest = 2
= .
1
2
µl + m sin2 πxl i Ekin, max/ω 2

4.2 Multiple degree of freedom systems

Hereafter different possibilities for setting up the equations of motion of multiple degree of
freedom (MDOF) systems are discussed. Here, the equilibrium of forces, the principle of
Hamilton, and the principle of virtual work are applied.

4.2.1 Equilibrium of forces

The equations of motion for MDOF systems can be obtained through cutting the single system
components free and writing the corresponding equilibrium conditions. This is presented
hereafter for a oscillator with a mass damper.

Example 4.8: 2DOF system - equilibrium

m1 ẅ1

m1
w1
k2 (w2 − w1 )
k2
k2 (w1 − w2 ) m2 ẅ2
k1 k1 k1 k1
2
m2 2 2 1
w 2 1
w
w2

The equilibrium of vertical forces reads:

m1 ẅ1 + k1 w1 + k2 (w1 − w2 ) = 0
m2 ẅ2 + k2 (w2 − w1 ) = 0 .

Structural Dynamics 4 Setting up the equation of motion

summer term 2019 49
 In matrix notation the equilibrium reads

m1 0 k + k2 −k2
" #" # " #" #
ẅ1 w1
+ 1 =0
0 m2 ẅ2 −k2 k2 w2

or

Mẅ + Kw = 0 .

Example 4.9: Frame structure with rigid bars and massless columns

This example gives an introduction to a simplified modeling of continuous frame struc-

tures as MDOF systems. For the depicted frame structure we assume that the vertical
columns are massless (µ = 0). The total floor masses are concentrated in the bars and
the bars are assumed to be rigid. All external forces act on the bars.

xi i+1 mi+1 üi+1

Element

i mi üi
wi

i−1 mi−1 üi−1

x1
µ=0

w1

The vertical columns are described by a Euler-Bernoulli beam approach:

EIw00000 = 0 as µ = 0 .

Structural Dynamics 4 Setting up the equation of motion

summer term 2019 50
 The displacement of the respective beam is then given by (with the local coordinate x):

w(x) = C0 + C1 x + C2 x2 + C3 x3

w0 (x) = C1 + 2C2 x + 3C3 x2

w00 (x) = 2C2 + 6C3 x

w000 (x) = 6C3 .

The four unknown of the displacement C0 , . . . , C3 can be expressed by the displacements

and rotations at the boundaries of the vertical beam element, i.e. through the nodal
unknowns wl , φl , wu and φu (l for “lower” side, u for “upper” side).
This results in

wu − wl 2 φu + 2φl 2 φu + φl 3 wu − wl 3
w(x) = wl + φl x + 3 2
x − x + 2
x −2 x
l l l l3

and with ξ = xl :
       
w(x) = 1 − 3 ξ 2 + 2 ξ 3 wl + l ξ − 2 ξ 2 + ξ 3 φl + 3 ξ 2 − 2 ξ 3 wu + l −ξ 2 + ξ 3 φu .

The nodal forces can be obtained from the derivatives of the displacement field as
M = −EIw00 and Q = −EIw000 . Under consideration of the definition of positive forces
corresponding to the following figure:

node

Vr
Mu

beam element
x

Ml
Vl w

node

Structural Dynamics 4 Setting up the equation of motion

summer term 2019 51
 the nodal forces result to:

6 4 6 2
 
Ml = [−EIw00 (0)] = EI w l + φl − w u + φu ,
l2 l l2 l
6 2 6 4
 
Mr = − [−EIw00 (l)] = EI 2 wl + φl − 2 wu + φu ,
l l l l
12 6 12 6
 
Vl = − [−EIw (0)] = EI 3 wl + 2 φl − 3 wu + 2 φu ,
000
l l l l
12 6 12 6
 
Vr = [−EIw (l)] = EI − 3 wl − 2 φl + 3 wu − 2 φu ,
000
l l l l

and for the special case of φl = φu = 0 to:

EI
Vl = 12 (wl − wu )
l3
EI
Vr = 12 (−wl + wu ) .
l3

With the full description of the beam elements, the interaction within the complete
structure can be described by the equilibrium conditions at the nodes.
From the assumption of rigid horizontal system components (floors), we can conclude,
that the rotations at the nodes vanish and that the horizontal displacements on the
left and right side are the same. For that reason the whole rigid horizontal system
component can be cut free and described by the equilibrium condition. In order to
consider the interaction of all elements, the displacement field and respective nodal end
forces have to considered in global coordinates.
(i+1)
ui = wu(i) = wl .

In the equilibrium Fx = 0 for each floor i, the inertia forces of the bar and external
P

forces are considered:

(i+1) (i+1)
Vl mi üi Vl

fi

Vr(i) Vr(i)

Structural Dynamics 4 Setting up the equation of motion

summer term 2019 52
 Transition condition:

wi = ui .

Then the equilbirium condition reads:

EI EI EI
 
H=0: 2 −12 3 ui−1 + 2 · 12 3 ui − 12 3 ui+1 + müi = fi
X
l l l
Ki,i−1 ui−1 + Kii ui + Ki,i+1 ui+1 + müi = fi
Ku + Mü = f

with

EI EI EI
Ki,i−1 = −24 Kii = 48 Ki,i+1 = −24 .
l3 l3 l3

Analogously this can be done for all floors. We need to consider that for the highest
floor no contribution V (i+1) appears.

4.2.2 Principle of Hamilton

Analogously to the SDOF systems discussed in Sec. 4.1, the principle of Hamilton can also be
applied to MDOF systems. This is presented in the following example.

Example 4.10: 2DOF - Hamilton’s principle

0 1 2
l1 kφ l2 m
φ1 φ2 Iθ
k

The kinematic relations read

w1 = φ1 l1
w2 = φ1 l1 + φ2 l2 .

Structural Dynamics 4 Setting up the equation of motion

summer term 2019 53
 The kinetic and potential energy can thus be expressed as

1 1 1  2 1
T = mẇ22 + Iθ φ̇22 = m φ̇1 l1 + φ̇2 l2 + Iθ φ̇22
2 2 2 2
1 1 1 1
U = kw12 + kφ (φ2 − φ1 )2 = k(φ1 l1 )2 + kφ (φ2 − φ1 )2
2 2 2 2
1 ˙ 1 1 1
L = T − U = m φ1 l1 + φ˙2 l2 + Iθ φ˙22 − kl12 φ21 − kφ (φ2 − φ1 )2 .
2
2 2 2 2

The Euler-Lagrange equations read

d ∂L(φ1 ,φ2 ,φ˙1 ,φ˙2 ) ∂L(φ1 ,φ2 ,φ˙1 ,φ˙2 )

− + =0
dt ∂ φ˙1 ∂φ1
d ∂L(φ1 ,φ2 ,φ˙1 ,φ˙2 ) ∂L(φ1 ,φ2 ,φ˙1 ,φ˙2 )
− + = 0.
dt ∂ φ˙2 ∂φ2

Under consideration of

∂L(φ1 ,φ2 ,φ̇1 ,φ̇2 )

= −kl12 φ1 + kφ (φ2 − φ1 )
∂φ1
∂L(φ1 ,φ2 ,φ̇1 ,φ̇2 ) h i
= m φ̇ l + φ̇2 2 l1
l
∂ φ˙1
1 1

∂L(φ1 ,φ2 ,φ̇1 ,φ̇2 )

= −kφ (φ2 − φ1 )
∂φ2
∂L(φ1 ,φ2 ,φ̇1 ,φ̇2 ) h i
= m φ̇1 l1 + φ̇2 l2 · l2 + Iφ φ̇2
∂ φ̇2

the Euler-Lagrange equations result in

−ml1 (φ̈1 l1 + φ̈2 l2 ) − kl12 φ1 + kφ (φ2 − φ1 ) = 0

−ml2 (φ̈1 l1 + φ̈2 l2 ) − Iθ φ¨2 − kφ (φ2 − φ1 ) = 0.

Again, the equations are written in matrix notation.

φ¨1 kl12 + kφ −kφ

" #" # " #" #
ml12 ml1 l2 φ1
+ =0
ml1 l2 ml22 + Iθ φ¨2 −kφ kφ φ2

Structural Dynamics 4 Setting up the equation of motion

summer term 2019 54
 or

Mẅ + Kw = 0 .

In the preceding examples, damping forces were neglected. As will be discussed later on,
the treatment of damping is often not straightforward (comp. Sec. 4.3). In general, some
assumptions are made that lead to a damping matrix C (comp. Chap. 8). Then the equation
of motion reads

Mẅ + Cẇ + Kw = 0 (4.12)

4.3 Damping models

4.3.1 Linear viscoelastic damping

In order to know how typically damping models are applied in engineering applications, we
first start by briefly looking at the transition from the theory of elasticity to the theory of
viscoelasticity. For this, we consider a spring that perfectly follows Hooke’s law, i.e., force and
displacement are dependent in a linear relationship:

f (t) = ku(t) (4.13)

with the proportionality constant k. Under a harmonically varying displacement u(t) =

u0 cos(ωt), the force is given by:

f (t) = ku0 cos(ωt) = f0 cos(ωt) (4.14)

and thus, force and displacement oscillate in phase. The force-displacement curve is depicted
in Fig. 4.2a. However, in real applications we observe that there is a phase shift between force
and displacement, showing that the model of a spring is not sufficiently describing the real
behavior. The force-displacement curve results in figures like 4.2b, also known as hysteresis
curveRor loop. With the knowledge, that the area within the hysteresis loop is a measure of the
work F (t)v(t) dt done by the oscillating force, we can conclude that for the ideal elastic spring
no energy is dissipated during a harmonic cycle. As the dissipation of energy (or rather the
conversion of mechanical into thermal energy or heat) is inherent to all mechanical phenomena,
the elastic model is not sufficient for an accurate description of the physical processes, when
it comes to the dynamic responses. A more realistic force-displacement curve also considering
non-linearities under cyclic oscillation is given in Fig. 4.2b.

Structural Dynamics 4 Setting up the equation of motion

summer term 2019 55
 f f
slightly nonlinear

highly nonlinear
u u

(a) Ideal linear elastic spring, Hooke’s law (b) More realistic hysteresis loops, considere-
ing non-linearities, adapted from Nashif
et al [1985]

Figure 4.2: Force-displacement curves, hysteresis loops

In general, the processes that lead to damping require more complex constitutive models,
because usually the governing effects are highly nonlinear.
Two categories of damping are commonly defined: external and internal damping. External
damping stems from boundary effects and is often associated with friction, such as between
structural components. Internal damping on the other hand is induced by the structural mate-
rial. For internal damping, the damping mechanisms can be linked to internal reconstructions,
due to, e.g. magnetic or thermal effects Nashif et al [1985]. An extension of the above men-
tioned theory of elasticity to cope with such phenomena is the theory of viscoelasticity, which
we will introduce briefly.
For a linear viscoelastic material that is subject to time-dependent variations of stress and
strain, the most general relation between stress and strain is given by the following partial
differential equation Snowdon [1968]

d d2 dn
!
a0 + a1 + a2 2 + . . . + an n + . . . f =
dt dt dt
d d2 dn
!
= b0 + b1 + b2 2 + . . . + bn n + . . . u. (4.15)
dt dt dt

Structural Dynamics 4 Setting up the equation of motion

summer term 2019 56
This relation is also known as the generalized standard model Nashif et al [1985], which relates
forces and displacements or, in an alternative setup, stresses and strains. It holds for arbitrary
orders and describes every system consisting of springs and viscous dampers. If we consider
all terms up to order n = 1, we obtain the adapted standard linear model:

df du
a0 f + a1 = b0 u + b1 . (4.16)
dt dt

We now consider a few simple models, that are all derived from Eq. (4.15), where we also give
the corresponding form with the spring constant k and the viscous damping coefficient c.
• Hooke model. This model corresponds to the purely elastic Hooke’s law with b0
a0
=k
and all ai = bi = 0 for i ≥ 1.

a0 f = b 0 u → f = ku . (4.17)

In this case all time derivatives vanish.

• Newton model. This model corresponds to the viscous dashpot with a0 , b1 6= 0 and
all other ai , bi equal to zero.

du
a0 f = b 1 → f = cu̇ . (4.18)
dt

• Maxwell model. This model corresponds to the a spring and dashpot in serial connec-
tion with a0 , a1 , b1 6= 0 and all other ai , bi equal to zero. The forces in the spring and
the damper are equal, thus

df du c
a0 f + a1 = b1 → f + f˙ = cu̇ . (4.19)
dt dt k

• Kelvin-Voigt model. This model corresponds to the a spring and dashpot in parallel
connection with a0 , b0 , b1 6= 0 and all other ai , bi equal to zero. The displacements in
the spring and the damper are equal, thus

du
a0 f = b 0 u + b 1 → f = ku + cu̇ . (4.20)
dt

Structural Dynamics 4 Setting up the equation of motion

summer term 2019 57
 c
k c k c
k

(a) Hooke (b) New- (c) Maxwell (d) Kelvin-

model ton model Voigt
model model
Figure 4.3: Illustrations of different viscoelastic models derived from the generalized standard model.

We now consider the generalized standard model for the case of harmonic oscillations. To this
end, we insert f (t) = fbe iωt and u(t) = ube iωt into Eq. (4.15):

d d2 dn
!
a0 + a1 + a2 2 + . . . + an n + . . . fbe iωt =
dt dt dt
d d2 dn
!
= b0 + b1 + b2 2 + . . . + bn n + . . . ube iωt . (4.21)
dt dt dt

We can cancel e iωt on both sides. Furthermore, we can write the n-th derivative of e iωt as

dn iωt
e = (iω)n e iωt (4.22)
dtn

and thus find for Eq. (4.21):

 
a0 + a1 (iω) + a2 (iω)2 + . . . + an (iω)n + . . . fb =
 
= b0 + b1 (iω) + b2 (iω)2 + . . . + bn (iω)n + . . . ub. (4.23)

Then, the ratio of force and displacement in the frequency domain

fb b0 + b1 (iω) + b2 (iω)2 + . . . + bn (iω)n + . . .

= = kb = k (1 + iη) . (4.24)
ub a0 + a1 (iω) + a2 (iω) + . . . + an (iω) + . . .
2 n

Here, we made use of the fact, that the ratio of any two arbitrary complex numbers can again
be written as a complex number. In general, now k(ω) and η(ω) are frequency dependent
values. The relationship between force and displacement is now, analogously to Hooke’s law,
interpreted as a complex spring, that represents the complete viscoelastic behavior. This
important result shows, that for harmonic oscillations, any connection of spring and damper
elements can be represented by a frequency dependent complex spring.

Structural Dynamics 4 Setting up the equation of motion

summer term 2019 58
 Example 4.11: Complex spring of Kelvin-Voigt model

Consider again the Kelvin-Voigt model, with the differential equation

cu̇ + ku = f .

For time harmonic oscillations f (t) = fbe iωt and u(t) = ube iωt , it results

c(iω)ube iωt + k ube iωt = fbe iωt .

Then, the complex spring kb is found as

fb cω
 
kb = =k 1+i .
ub k

Thus, the imaginary part η(ω) = cω E

is a linear function of the frequency, whereas the
real part k is constant throughout the frequency range. It is preempted here, that this is
often not found to be a realistic approximation to the behavior observed in engineering
materials.

The Newton model Consider the Newton model depicted in Fig. 4.3b. The governing
equation is

f = cu̇ . (4.25)

Assuming time harmonic oscillation u(t) = u0 sin(ωt), it is u̇(t) = u0 ω cos(ωt) and we obtain

f = cωu0 cos(ωt) . (4.26)

The work dissipated in one cycle of oscillation then is

ZT ZT
T
Wd = f (t)v(t) dt = cω 2 u20 cos2 (ωt) = cω 2 u20 = πcωu20 . (4.27)
2
0 0

Structural Dynamics 4 Setting up the equation of motion

summer term 2019 59
Furthermore, we derive the hysteresis loop for
q the Newton model. To this end, we start with
Eq. (4.26) and insert the identity cos(ωt) = 1 − sin2 (ωt):
q
f (t) = cωu0 1 − sin2 (ωt) . (4.28)

Taking the square on both sides:

 
f 2 (t) = c2 ω 2 u20 1 − sin2 (ωt) . (4.29)

As sin(ωt) = u(t)
u0
, we can write

!2 !2
f (t) u(t)
=1− (4.30)
cωu0 u0

which can be further written as

!2 !2
f (t) u(t)
+ = 1. (4.31)
cωu0 u0

This equation describes an ellipse and represents the hysteresis loop, i.e., the relationship
between instantaneous force versus instantaneous displacement, of the single dashpot element.
Furthermore, the area within an ellipse, described by (x/a)2 + (y/b)2 = 1 is given by A = πab,
thus for the hysteresis loop, we find

A = πcωu0 u0 = πcωu20 = Wd . (4.32)

We note, that the area within the hysteresis loop corresponds to the work dissipated within
one cycle of motion (Fig. 4.4a).

The Kelvin-Voigt model Furthermore, consider the Kelvin-Voigt model depicted in Fig. 4.3d.
The governing equation is

f = ku + cu̇ . (4.33)

Assuming time harmonic oscillation u(t) = u0 cos(ωt + ϕ), it is u̇(t) = u0 ω sin(ωt + ϕ) and we
obtain

f = ku0 cos(ωt + ϕ) + cωω sin(ωt + ϕ). (4.34)

Structural Dynamics 4 Setting up the equation of motion

summer term 2019 60
 f
f ku0
cωu0 cωu0

u u
u0 u0

(a) Dashpot, Newton model (b) nonlinear hysteresis loops, adapted

from Nashif et al [1985]

Figure 4.4: Force-displacement curves, hysteresis loops

To find an expression for the work dissipated in one cycle of motion, we separate the force
terms and first investigate the work done by the spring force fk (t) = ku0 cos(ωt + ϕ)

ZT ZT
Wd,k = fk (t)v(t) dt = −kωu20 cos(ωt + ϕu ) sin(ωt + ϕu ) dt = 0 . (4.35)
0 0

No energy is dissipated in the spring, which is as expected, since the spring behaves ideally
linear elastic. Furthermore the work done by the damping force fd (t) = cωu0 sin(ωt + ϕ) is

ZT ZT
T
Wd,d = fd (t)v(t) dt = cω 2 u20 sin2 (ωt + ϕ) = cω 2 u20 = πcωu20 . (4.36)
2
0 0

We obtain the same solution as for the Newton model in Eq. (4.27). Analogously, we derive
the hysteresis loop for the Kelvin-Voigt
q model. To this end, we start with Eq. (4.34) and insert
the identity cos(ωt + ϕ) = 1 − sin (ωt):
2

q
f (t) = ku0 cos(ωt + ϕ) + cωu0 1 − cos(ωt + ϕ) . (4.37)

Rearranging the equation, and taking the square on both sides leads to
!2
f (t) − ku0 cos(ωt + ϕ)
= 1 − cos(ωt + ϕ). (4.38)
cωu0

Structural Dynamics 4 Setting up the equation of motion

summer term 2019 61
Because cos(ωt + ϕ) = u(t)
u0
, we can write

!2 !2
f (t) − ku(t) u(t)
=1− (4.39)
cωu0 u0

which can be further written as

!2 !2
f (t) − ku(t) u(t)
+ = 1. (4.40)
cωu0 u0

This equation again describes an ellipse and represents the hysteresis loop, i.e., the relationship
between the instantaneous force versus the instantaneous displacement of the single Kelvin-
Voigt element. The ellipse is now tilted by an angle, which is given in Fig. 4.4b. Furthermore,
the area within this ellipse still corresponds to the work dissipated in the damper element
during one cycle of motion.

4.3.2 Ideal hysteretic damping

For many engineering materials, experiments show, that the amount of dissipated energy under
harmonic cycling does not depend on the frequency of oscillation, but rather is a function of
the maximum displacement. This motivated the creation of the concept of linear hysteretic
damping, often just termed hysteretic damping or structural damping. The term hysteretic
damping can be misleading, in the sense that also other forms of damping produce hysteresis
loops as was shown for viscous damping in the preceding section. The linear hysteretic damping
force is defined in the frequency domain by

f˜D (ω) = iηk sgn(ω)ũ(ω) . (4.41)

We observe that the damping force is now proportional to the displacement ũ(ω) but in phase
with the velocity due to the factor i. It has been found that the linear hysteretic damping
model is noncausal (Crandall [1961]), that is, the system responds before it is excited by a
force, which leads to non-physical behavior. Nevertheless, for many applications, e.g. stead-
state vibration, this error can be accepted. The damping force in the time domain is found by

fD (t) = F −1 f˜D (ω) = ηkF −1 {i sgn(ω)ũ(ω)} = ηkH {u(t)} = ηk ub(t)

n o
(4.42)

where xb(t) = H {u(t)} is the Hilbert transform of u(t), which we will not discuss further in
the scope of this lecture. Nevertheless, it can be found that the energy dissipated by the ideal

Structural Dynamics 4 Setting up the equation of motion

summer term 2019 62
linear hysteretic damping element is given by (Kausel [2017]):

ZT
Wd,h = fD (t)v(t) dt = πηku20 . (4.43)
0

We compare this to the energy dissipated in the viscous damper element (Newton model,
Eq. (4.27)), where we insert c = ηmωn and mωn = ωkn , and obtain

k ω
Wd,v = πcωu20 = πηmωn ωu20 = πη ωu20 = πηk u20 . (4.44)
ωn ωn
q
Here, we introduce the natural frequency ωn = m k
of the single degree of freedom system,
which will be introduced in Chap. 5. The amount of energy dissipated in one cycle of harmonic
motion differs by the factor ωωn for the hysteretic and the viscous damping element. In the
case ωωn = 1, i.e., for resonance, the dissipated energies are equal. This relation is depicted in
Fig. 4.5

Wd
viscous
hysteretic
πηku20
ω
−ωn ωn
−πηku20

Figure 4.5: Dissipated energy for viscous and hysteretic damper element

Structural Dynamics 4 Setting up the equation of motion

summer term 2019 63
4.3.3 Coulomb damping

For the Coloumb damping model we use the model of a rigid block on a rigid support, where
the sliding friction force in the interface is equal to the maximum static friction force.

w,ẇ

f (t)

|fD | = µ N
N

N
µ
|fD | = µ N

Figure 4.6: Damping due to friction

For this model, the absolute value of the reaction force fD (t) due to sliding friction does not
depend on the absolute value of the vibration velocity v(t). The direction of the reaction force
fD (t) follows directly from the direction of the velocity v(t) and the direction of the external
force f (t) in the steady state is the same as the direction of the velocity. This indicates that
a positive work is done on the system. This model is called St. Venant-model. It is sketched
in Fig. 4.7, where also the respective hysteresis is depicted.

f (t)

(a) System (b) Hysteresis

Figure 4.7: St. Venant model

Structural Dynamics 4 Setting up the equation of motion

summer term 2019 64
4.3.4 Aeroelastic damping

Using a quadratic relation between force and velocity, the reaction forces in an adjacent fluid
(gas or fluid) can be approximated. According to the Bernoulli-equation, the pressure depends
quadratically on the flow velocity.

D = −αv 2 (t) sgn (v(t)) (4.45)

4.4 Summary

• Energy and equilibrium methods enable the determination of the governing differential
equations. With the help of the Hamilton principle, the relevant equations can be ele-
gantly stated. As energy quantities are scalar values the signs in the governing equations
result from the derivation, which is less error prone and thus advantageous. Furthermore,
additional constraints can be taken into account using Lagrange multipliers.
• If forces have a potential, they can be derived from it. Therein, potential means, that the
potential energy of the system can be uniquely determined from the current deformation
state of the system. If all forces have a potential, under the deformation and subsequent
reverse deformation of the system no energy is lost.
• For damping and frictional forces as well as all external forces no potential exists. The
Hamilton principle can then be extended for these non-conservative forces.

Structural Dynamics 4 Setting up the equation of motion

summer term 2019 65
5 Free vibration of linear systems with
single degree of freedom
In the following we will look at mechanical systems with one degree of freedom, also termed
single degree of freedom (SDOF) systems. These systems are characterized by three coefficients
describing the inertia force fI , the elastic restoring force fK and the damping force fC :
m . . . mass
k . . . stiffness
c . . . damping coefficient (here, viscous damping)
We assume, that the equation of motion is found by any of the methods in section 4.1, e.g. the
equilibrium of forces, repeated here for convenience. All forces acting on a moving mass point
in the free state (no external force) at equilibrium are considered, namely spring, inertia, and
damping forces. The equation of motion is

mẅ

m w m
k c
kw cẇ

Figure 5.1: Free body cut for the unforced (free) SDOF system

mẅ(t) + cẇ(t) + kw(t) = 0 . (5.1)

In the following the free vibration of the SDOF-system shall be studied.

Structural Dynamics
summer term 2019
66
5.1 The undamped case

For a free vibration without damping, the equation of motion simplifies to

mẅ(t) + kw(t) = 0 . (5.2)

As no forcing term is present, no particular solution is needed to satisfy the differential equation
and the homogeneous solution describes the complete solution w(t) = wh (t). This homoge-
neous differential equation can be solved using the following equivalent approaches.

w(t) = wb+ eiωt + wb− e−iωt = w01 cos(ωt) − w02 sin(ωt) = w0 cos(ωt + ϕw ) . (5.3)

We decide to apply the latter approach, using a single cosine function with phase shift. We
now derive the second derivative with respect to time ẅ(t).

ẇ(t) = −w0 ω sin(ωt + ϕw ) (5.4)

ẅ(t) = −w0 ω 2 cos(ωt + ϕw ) . (5.5)

After inserting (5.4), (5.2) can be expressed as

 
−mω 2 + k w0 cos(ωt + ϕw ) = 0 . (5.6)

As in general w0 cos(ωt + ϕw ) 6= 0, the above equation is satisfied, if

−mω 2 + k = 0 (5.7)

which results in
s
k
ω = ωn = = 2πfn . (5.8)
m

ωn constitutes the natural (circular) frequency or eigenfrequency of the system. The solution
is depicted in Fig. 5.2 over time and as rotation in the plane, as was introduced in Chap. 2.

1 2π
T = = (natural) period of the motion
fn ωn
w02
tan ϕw = phase angle
w01
q
w0 = 2
w01 + w02
2
amplitude of vibration .

Structural Dynamics 5 Free vibration of linear systems with single degree of freedom
summer term 2019 67
 w w
w0

ϕ π 3
2 π 2
π 2π ωt
∗ ∗
ωt ωt
w = w0 cos(ωt + ϕ)

ϕ
ωT

Figure 5.2: Homogeneous solution of SDOF-system

w0 and ϕ are determined to satisfy the initial conditions w(0) = w0 and ẇ(0) = v0 , respectively.
The solution can then be given (without derivation):

v(t = 0)
w (t) = w(t = 0) cos (ωn t) + sin (ωn t) . (5.9)
ω

5.2 The damped case

5.2.1 Velocity proportional (viscous) damping

To solve the equation of motion considering velocity proportional damping

mẅ(t) + cẇ(t) + kw(t) = 0 (5.10)

the approach for a linear homogeneous differential equation with constant coefficients

w(t) = we
b λt (5.11)

is used.

⇒ ẇ(t) = v(t) = λwe

b λt (5.12)
⇒ ẅ(t) = a(t) = λ2 we
b λt (5.13)

Structural Dynamics 5 Free vibration of linear systems with single degree of freedom
summer term 2019 68
If (5.12) and (5.13) are introduced into (5.10) as

(mλ2 + cλ + k)we
b λt = 0 (5.14)

the characteristic polynom

c k
λ2 + λ+ =0 (5.15)
m m

with the non-trivial solution

s
2
c c k

λ1,2 =− ± −
2m 2m m
s
2
c c

=− ± − ωn2
2m 2m
s 2
c c
=− ± ωn −1 (5.16)
2m 2mωn

is obtained. We introduce
c
δ= decay constant,
2m
ccrit = 2mωn critical damping,
c
D= (= ξ) percentage of critical damping.
ccrit

Thus, Eq. (5.16) can be written as

s 2
c
λ1,2 = −δ ± ωn −1
ccrit
√
= −δ ± iωn 1 − D2 (5.17)

Depending on the value D, this equation has different types of solutions.

√
D < 1, complex root: λ1,2 = −δ ± iωn 1 − D2 = −δ ± iωD (5.18)
D = 1, root = 0 : λ1,2 = −δ (5.19)
√
D > 1, real root: λ1,2 = −δ ± ωn D2 − 1 = −δ ± ω ∗ (5.20)

The solution for the three cases is given subsequently.

Structural Dynamics 5 Free vibration of linear systems with single degree of freedom
summer term 2019 69
5.2.1.1 Quasi periodic case (D < 1)

After inserting Eq. (5.18) into the initial approach in Eq. (5.11) the solution is obtained as
a sum of conjugate complex solutions with conjugate complex coefficients. We denote the
respective amplitudes by w+ and w− .

w(t) = wb+ e (−δ+iωD )t + wb− e (−δ−iωD )t

 
= e −δt wb+ eiωD t + wb− e−iωD t (5.21)
= e −δt (wb+ (cos(ωD t) + i sin(ωD t)) + wb− (cos(ωD t) − i sin(ωD t)))
= e −δt ((wb+ + wb− ) cos(ωD t) + i (wb+ − wb− ) sin(ωD t))
| {z } | {z }
w01 =2 Re(w
b+ ) iw02 =2i Im(w
b+ )

= e −δt (w01 cos(ωD t) − w02 sin(ωD t)) (5.22)

or

w(t) = w0 e −δt cos(ωD t + ϕ)

with
s
√ k
ωD = 1 − D2 damped natural (circular) frequency
m
1
TD = √ (natural) period of the motion
f 1 − D2
w2
tan ϕw = phase angle
w1
q
w0 = 2
w01 + w02
2
initial amplitude at the beginning of the vibration .

w0 and ϕw , or w01 and w02 , or wb+ and wb+ are determined such that the solution satisfies the
initial conditions w(t = 0) = w0 and ẇ(t = 0) = v0 . The resulting formulation in form of
Eq. (5.22) is given without further derivation as

v0 + δw0
" #
w (t) = e −δt
w0 cos (ωD t) + sin (ωD t) . (5.23)
ωD

The resulting vibration is depicted exemplarily in Fig. 2.4 for the case w0 = 1 and v0 = 0.

Structural Dynamics 5 Free vibration of linear systems with single degree of freedom
summer term 2019 70
 Im(2w+ ) w

1 − e−δT

ωt Re(2w+ ) π
2 π
3
2
π 2π ωD t

e−δt
1 − e−δT
Figure 5.3: Damped oscillation over one period T with w0 = 1 and ϕ = 0.

In this App the time response of a damped single degree of freedom system is
visualized. The mass m, stiffness k, damping constant c and initial velocity
v0 are parameters which can be changed and the initial displacement w0 = 0.
By starting the oscillation, the system oscillates about the static equilibrium
position.
http://apps.bm.bgu.tum.de:5006/Damped_oscillator

From the definition of the damped natural frequency, we find (from Chopra [1995])
ωD
ω
√ 1
ωD = 1 − D2 ω ωD 2
 
ω
+ D2 = 1
ωD = (1 − D2 )ω 2
2
ωD

+ D2 = 1 .
ωn D
1
For small values√of damping D < 0.2, the ratio of damped and undamped natural frequency
is close to one ( 1 − 0.22 ≈ 0.98).

5.2.1.2 Aperiodic limit case (D = 1)

The case D = ωδn is the so-called critical damping case, for which c = ccrit = 2mωn . This is
the limit above which the system does not oscillate but decays in an aperiodic movement. As
the two eigenvalues of the problem in Eq. (5.19), λ1,2 = −δ, coincide, the solution approach

Structural Dynamics 5 Free vibration of linear systems with single degree of freedom
summer term 2019 71
has to be altered. In general, the solution for the SDOF with eigenvalues of multiplicity 2 is
given as

w(t) = w1 e −δt + w2 te −δt

= e−δt (w1 + w2 t) . (5.24)

The constant coefficients w1 and w2 are determined to fulfill the initial conditions w(0) = w0
and ẇ(0) = v0 .

If the default parameters are used, i.e. the mass m = 8 kg and the stiffness
k = 50 mN
, then ωn = 2,5 rad
s
and ccrit = 40 Ns
m
. Thus the aperiodic limit case
can be visualized when the damping constant is set to c = ccrit = 40 Ns
m
.
http://www.bm.bgu.tum.de/lehre/interactive-apps/
damped-oscillator/

5.2.1.3 Aperiodic case (D > 1)

For the aperiodic case with D > 1, the solution reads

∗ )t ∗ )t
w(t) = w1 e (−δ+ω + w2 e (−δ−ω
∗ ∗t
= w1 e −δt e ω t + w2 e −δt e −ω
∗ ∗t
 
= e−δt w1 e ω t + w2 e −ω (5.25)

with
s
√ k
ω∗ = D2 − 1 .
m

The constant coefficients w1 and w2 are determined to fulfill the initial conditions w(0) = w0
and ẇ(0) = v0 . The solution is exemplarily depicted in Fig. 5.4 for the aperiodic and aperiodic
limit case.

5.2.2 Relations between different damping measures

The values describing the damping of a system can be expressed via a series of measures.
Tab. 5.1 provides an overview of commonly used values and their relations.

Structural Dynamics 5 Free vibration of linear systems with single degree of freedom
summer term 2019 72
 w (t) v0 > 0
v0 = 0
v0 = −w0 (δ + ω ∗ )
w0

t
v0 < −w0 (δ + ω ) ∗

Figure 5.4: Homogeneous solution of the SDOF system for the aperiodic case D > 1. The solution
is depicted for different initial conditions. For all these cases, no oscillatory is present.

Table 5.1: Overview over the commonly used damping values and their relation.

percentage of
mechanical decay logarithmic
critical loss factor
resistance constant decrement
damping
c δ D Λ η
√ √
= 2mδ
q
c c 2D km 2Λ km
4π 2 +Λ2
η km
c
δ = δ ωn D √ ωn Λ η
ω
2 n
2m 4π 2 +Λ2

D = √c
2 km
δ
ωn
D √ Λ
4π 2 +Λ2
η
2

Λ = √ 2πc
4km−c2
√ 2πδ √2πD
1−D2
Λ √ πη
ωn −δ 2
2 1−(η/2)2
η = √c
km
δ
πf0
2D √ 2Λ
4π 2 +Λ2
η

Structural Dynamics 5 Free vibration of linear systems with single degree of freedom
summer term 2019 73
6 Classification of Excitations
It is of benefit to group the various excitations in order to be able to find favorable solu-
tion strategies for each of those problems. Hereafter, types of excitation that are treated in
structural dynamics problems are introduced.

6.1 Modelling of the excitation

In structural dynamics, excitations from time varying forces or displacements are causal for
occurring motions. Nevertheless, the excitations—depending on the the considered system
that is “cut free” from the physical environment—are not always modeled as forces.
Relevant excitations in structural dynamics from earthquakes, traffic induced vibrations, vibra-
tions from industrial facilities in the neighborhood, building work or detonations, are usually
applied via root-point excitation, i.e., displacements (comp. Sec. 7.2.2). Thereby, the causal
(force) excitation, e.g., induced by dynamic forces (emission) between a moving train wheel
and the track including the transmission into the soil and to the foundation, is described
indirectly via the motion of the soil, e.g. at the foundation, by a displacement excitation.
The concept of impedances, with which also large substructures can be described and possibly
measured, thereby helps for the consideration of the coupling (e.g., Tab. 10.1).
Nevertheless, often the excitation is directly modeled by dynamic forces, e.g. resulting from
the operation of fixed or moving machinery with direct influence on the structure, from building
work, traffic, wind, shock waves, from explosions, or from moving persons.
Vibrations, that result from time varying or, for moving loads, spatially varying parameters
(e.g., a periodically varying stiffness under a train wheel on and between the railway sleepers),
are termed as parametric excitations, as they depend on the systems’ parameters (especially
the stiffness) of the excited structure. They are either described by differential equations with
time dependent stiffness and/or damping values (solution in the time domain), indirectly by
substitute forces, obtained from simple models, or by time varying systems, for calculations
in the time domain.
Furthermore, the temporal structure of the excitation influences the modeling. Depending
on it, it is distinguished between transient loads (e.g. explosion loads, falling weights,
impact loads, jumping persons), periodic loads (e.g. machines with rotating parts, effect of
unroundness of railway wheels, loads from moving persons, stick-slip vibrations, which appear,

Structural Dynamics
summer term 2019
74
e.g., for friction processes, pile shaking, dynamic building ground compression, rotating rollers,
etc.) and stochastic loads (e.g. wind loads, earthquake loads, traffic induced vibrations,
loads from mills and crushers).
A non-periodic, transient load, can be replaced by a periodic load with an infinite period T ,
moving from the Fourier series to the Fourier integral (comp. Chap. 3). However, due to
the inevitable discretization in the time and frequency representation, an infinite period is not
suitable. For transient signals described in the frequency domain, the computational temporal
repetition length T can be chosen such that the response between the repeating transient
events has decayed. Then, discrete spectra may be applied.

Figure 6.1: Transient, triangular load

t
t=0
T →∞ T →∞

Figure 6.2: Representation of an aperiodic signal using an infinite return period

For linear time invariant system—or system which allow linearization around a static initial
state—the advantage of a representation in the frequency domain can be made use of. The
periodic excitation can the be represented by a sum of harmonic loads.
The description of the vibration response to stochastic loads is done for a few problems, e.g.,
for earthquake engineering, by means of response spectra. (comp. Sec. 7.5.1.1). They process
the stochastic excitation for engineering applications, such that approximate deterministic
calculations can be carried out.
Independently, treatment of the stochastic processes can be done with the help of the proba-
bility theory (comp. Chap. 13)
In VDI 2038 [2012] the most important dynamic excitations for buildings, components and
other plant components are classified.
Subsequently, we give a short overview over the different categories of temporal structures.

Structural Dynamics 6 Classification of Excitations

summer term 2019 75
6.1.1 Harmonic excitation

Within the model of a harmonic excitation it is assumed that the load has been acting on the
system since infinite time. This requirement can be relaxed by requiring the time since the
start of the excitation to be much larger than the eigenperiod of the system T , so that Tt  1.
Then it can be assumed, that the system is in the steady state vibration. This means, that
the transient response, which is present after the beginning of the excitation, has sufficiently
decayed.
Examples for harmonic excitations are:
• Machines with rotating components
• Paper factories (large rolls)
• Rotating machinery
• Wind (to certain extent)
A harmonic load f (t) can be given as f (t) = f0 cos(Ωt + ϕ), with
f0 amplitude of load in N

Ω = 2πf = 2π
T
circular frequency (of the load) in rad
s
f= Ω
2π
frequency (of the load) in Hz = s1
T = 1
f
= 2π
Ω
period of the motion in s
ϕ phase shift angle in rad

6.1.2 Periodic excitation

In the model representation of a periodic excitation we also assume that the load has been
acting on the system since infinite time. This type of excitation is, however, not characterized
through a single excitation frequency, but a superposition of a number of harmonic excita-
tions. Thus, harmonic excitations can be seen as a special case of a periodic excitation. The
Calculation of the individual harmonics carried out by a Fourier transformation as explained
in Chap. 3.
Examples for periodic excitations are:
• Rotating machinery
• Propellers
• Surface irregularities of wheels of trains or out-of-roundness of truck tires
• Pedestrians

Structural Dynamics 6 Classification of Excitations

summer term 2019 76
 • Stick–slip vibrations
• Sheet pile shaking, dynamic building ground compression

Example 6.1: Decomposition of periodic load into harmonics

Figure 6.3 shows a load record and its decomposition in individual harmonic signals.

f (t)

f=
T T
=
f1
t f1 sin (Ω0 t + ϕ1 )
T

+
f2
t f2 sin (2Ω0 t + ϕ2 )
T T
2 2

+
f3
t f3 sin (3Ω0 t + ϕ3 )
T T T
3 3 3

+ ...
Figure 6.3: Periodic signal and harmonic decomposition

Example 6.2: Short-time Fourier transformation

In the below figure, a short-time Fourier transformation of the acceleration at the axle-
bearing of an ICE is depicted. The single lines in the colored representation show
the excitation frequencies. In this illustration a line pattern with distinct excitation
frequencies becomes apparent (e.g. for around 60, 80, 100, 120 Hz). This excitation
stems from the unroundness of the wheels. Periodic loads then act on the ground as

Structural Dynamics 6 Classification of Excitations

summer term 2019 77
 well as the vehicle.

Time [s]

Frequency [Hz]

6.1.3 Transient load

Transient loads are only present in the system during a limited time window.
Examples for transient excitations are:
• Explosions
• Shock or crash
• Impact
• Hammers
• Jumping
Fig. 6.4 shows a triangular load signal over time as an example for a transient excitation.
Aperiodic transient loads can be substituted by periodic loads with an infinite period. This is
linked to switching from the Fourier-series to the Fourier-integral (comp. Chap. 3).

Structural Dynamics 6 Classification of Excitations

summer term 2019 78
 f

Figure 6.4: Transient, triangular load

t
t=0
T →∞ T →∞

Figure 6.5: Representation of an aperiodic signal using an infinite return period

6.1.4 Irregular, random loads

A great number of loading processes show neither periodic parts nor can they be described
through a transient time function exactly. The randomness inherent in the excitation requires
a different description. Examples for irregular, random excitations are:
• Wind (only to certain extent)
• Train-induced vibration in the soil (to certain extent)
• Vibrations in vehicles through irregularities on the ground
• Earthquakes

Structural Dynamics 6 Classification of Excitations

summer term 2019 79
 Figure 6.6: Illustration of a stochastic process

Figure 6.7: Illustration of the autocorrelation function

This type of excitation can be characterized by a probabilistic description (comp. Chap. 13).
Typically the load is described through the autocorrelation function, its power density through
the Power Spectral Density.
Here, it is of great importance to to identify and describe the significant characteristics of the
excitation as basis for suitable methods for the acquisition (measurement) as well as for the
description of the excitation function.

6.1.5 Loads influenced by the parameters of adjacent systems

Some excitation processes result from the interaction of two systems, e.g. the load of vehicles,
that drive over a (flexible) bridge or over a (flexible) railway track. Some load components can
be neglected, if e.g. the flexibility of the bridge or the railway would be infinitely small, i.e.
the stiffness infinitely large. Vibration processes resulting from this type of coupling are also
called parametric vibrations, because they depend on the parameters of the adjoining coupled
systems. Examples for loads associated with interaction are:
• Vehicles on a bridge or railway track
• Pedestrians that synchronize with the structural vibration

Figure 6.8: Synchronization of pedestrians with the structural vibration, courtesy of M. Schnei-
der

• Fluid-Structure-Interaction

Structural Dynamics 6 Classification of Excitations

summer term 2019 80
6.2 Frequency content

The signals that typically occur in engineering problems can have various frequency charac-
teristics. Fig. 6.9 depicts a typical scenario. A train induces oscillation in the ground and the
surrounding air. Through wave motion the disturbances radiate from the track as airborne
and solid waves. Vibrations are induced into adjacent structures, like the building depicted.
Linked with these vibrations sound is radiated, termed reradiated sound. Sound waves that
directly cause vibrations inside the house cause primary sound.
To summarize, multiple paths occur. Depending on the frequency range of the occurring
signals, the oscillations are termed differently, depending on the specific background (e.g.
structure-borne sound, vibrations).

sound waves z y

z Reradiated sound

x, y, z Vibrations

Waves in solids

Emission Transmission Immission

Vibrations Structure borne sound

1 2 5 10 20 5080100 1000
Frequency in Hz
Figure 6.9: Frequency content characterization in a emission–transmission–imission system.

6.3 Methods

The choice of methods for structural dynamic or vibro-acoustic predictions depends on the
specific problem setting. Fig. 6.10 gives an classification of commonly used methods.

Structural Dynamics 6 Classification of Excitations

summer term 2019 81
 hybrid
analytical approaches
algorithms

unlimited

systems M A
Mu BE SE
osc ltib
limited illa ody
tor
M high
simple FE

boundary frequency
conditions
complex low
Figure 6.10: Common methods in strucutral dynamics and vibro-acoustics and their applicability.

6.4 Classification in standards

The dynamic load through machinery with rotating masses (engines, fans, turbines, vibra-
tors, centrifuges, generators, rotary pumps, etc.) can be well represented well in the frequency
domain with respect to their excitation frequency that depends on the rotational speed. To
some extent, from the moving masses the excitation forces can directly be concluded (e.g.,
DIN [2004a]). If this is not possible, the excitation forces can be obtained via measurements
at the machine foundations, together with impedance (comp. Chap. 10) measurements at
the support points. Hints are given in DIN 4024-1 [1988]; DIN 4024-2 [1991]. The indirect
evaluation of the forces by measurements is described in VDI 2038 [2012].
Machinery with oscillating masses (piston engines and pumps, saws, but also bells) are
partly well describable with respect to the induced, periodic forces (e.g., DIN 4178 [2005]).
The measurement based determination of forces is, e.g., described in DIN [2004b]; VDI 2038
[2012]. The excitation is usually described in the frequency domain, whereby also natural
vibrations (homogeneous solutions) at impact processes, e.g., for fast (emergency) switch off,
have to be considered.
Machinery with non-periodically moving masses (forging hammers, presses, jacking
systems, packing machines, placement machines, grinding machines, milling machines) lead to
transient excitations, that can be either represented in the time or frequency domain. Espe-
cially for hammer and falling weights, the excitation can be deduced from simple mechanical
models considering the masses and geometries. For machinery with complex inner movement
processes usually measurements are necessary.

Structural Dynamics 6 Classification of Excitations

summer term 2019 82
In the context of traffic induced loads the investigations, carried out for railway induced ex-
citations, are gathered in simplified substitute models (see e.g. Thompson [2008]; Wettschureck
et al [2012]). Here, usually measurements are carried out, but are interpreted in knowledge
of the mechanical phenomena—from the parametric excitation resulting from the periodic
changes of the track stiffness due to discrete supports of the tracks by sleepers, over the
displacement induced periodic excitation from the unroundness of the wheels, the stochas-
tic excitation from the roughness of the track surface and the track geometry, the excitation
from vibrations of the vehicle and the track, up to the excitation from the moved self weight
(z.B. Wettschureck et al [2012]; Müller [2011]). Remarks for road traffic induced vibrations are
given, e.g in Melke [1995]; VDI 2038 [2012].
Pedestrian induced excitations can be modeled by periodic loads. An overview can be
found in Butz and Distl [2008]. In Sahnaci [2013]; VDI 2038 [2012] a comprehensive overview
over the state of the art is given.
Dynamic excitations for building work and detonations include impact-like and pe-
riodic processes, for which an overview with mechanical models can be found in VDI 2038
[2012]; Müller-Boruttau and Breitsamter [2000].
Dynamic wind excitations range from stochastic broadband loads from turbulences (gust
induced vibrations) over periodic loads from vortex excitation to self-excited vibrations. For
and overview it is referred to, e.g., Peil and Clobes [2008]; Ruscheweyh [1982].
The single load types are normatively edited in DIN EN 1991-1-4:2010-129 [2010]; ISO 4354
[2009].
For the modeling of the highly dynamic, transient excitations in case of shock loads and
earthquake loads, it is referred to the corresponding contributions in this volume.
In the following table, the most important dynamic excitations for buildings, components and
other plant components are summarized, that are of of relevance in practical applications (from
VDI 2038:2010-11):

Structural Dynamics 6 Classification of Excitations

summer term 2019 83
 Excitation Remark Code
1 Operational
• Machinery with rotating masses periodic DIN 4024/
• Machinery with oscillating masses periodic VDI 3838
• from fluids in components periodic - pressure oscillations VDI 3842
Transient - shocks
• transport in buildings, floor conveyors transient
• bell ringing periodic DIN 4178
2 Traffic-induced
• railway traffic transient DIN 4150-2
• road traffic transient DIN 4150-2
• ship traffic periodic - under water sound
• air traffic periodic – low frequency
sound
3 Pedestrian-induced ISO 10137
• Walking, running periodic - pedestrian bridges
• jumping, dancing periodic - ceilings, corridors
• seesawing periodic – stands
4 building work DIN 4150
• shaking, compressing periodic/ impacts
• vibratory driving periodic
• impact driving impacts / periodic
• chiselling work impacts / periodic
• blasts impacts
• demolition, load crash impacts
• work traffic transient / impacts
5 Wind-induced DIN 1055-4
• gusts stochastic ISO 4354
• vortex shedding periodic
• galloping (self-excited vibration) amplifying vibration
• flutter, interference amplifying vibration
6 Incidents, special loads transient
• earthquakes DIN 4149
• blast waves
• impact from airplanes / flying object
• impact from ships / vehicles
• explosions of highly energetic tanks
• machinery accident

Structural Dynamics 6 Classification of Excitations

summer term 2019 84
 railway induced pedes- machi- wind constr.
vibrations trians nery work
force excitation vehicle vibration
description
by interaction excita- vehicle-track interaction
tion wheel motion from
sleeper
base point excita- vibrations in buildings
tion on the basis of measure-
ments in the free field
harmonic
time function periodic wheel roughness
moving load model (axle
distance, wagon length)
transient track impacts

stochastic track unroundness

point source vibrations in the near

field
spatial funct. line sources vibrations in the far field

area source

stochastic distribu- spread of the qualitiy of

tion wheels/bogies

Structural Dynamics 6 Classification of Excitations

summer term 2019 85
7 Forced vibration of linear systems with
single degree of freedom

7.1 Introduction

7.1.1 “Quasi-static” and dynamic processes

The system response to dynamic loads depends on the temporal structure of the load process.
Very slow processes can be treated as “quasi-static” like “infinitely” slowly applied loads. In
this case, the forces caused by the temporal changes of the processes (comp. Chap. 10, inertia
and damping forces, but not forces from sliding friction), are small compared to the static
reaction forces of the structure and can be neglected. For “quickly” preceding processes, the
forces caused by the temporal changes in those processes have to be considered.
The distinction between “quasi-static” and dynamic processes depends on the ratio between
the displacement dependent reaction forces and the inertia, damping, and friction forces caused
by the temporal changes. As shown in Chap. 10, the lowest eigenfrequencies of a structure
provides insight about the ratio of inertia and damping forces related to the static reaction
forces under time varying loads. Whereas for very soft and heavy structures, i.e., for structures
with comparably low displacement dependent reaction forces and high inertia (i.e., very small
lowest eigenfrequency) also relatively slow excitations, e.g., slow walking of a person, already
have to modeled as dynamic loading, for stiffer and lighter structures (e.g. a steel girder with
short span width and comparably high lowest eigenfrequency) it is sufficient, to consider the
same load as “quasi-static”. Thus, for an engineering evaluation of the influence of the temporal
structure on the vibration response, the temporal structure of the load has to be assessed
considering the dynamic properties of the system. This is usually done via a representation in
the frequency domain (spectral representation).
If a load at rest is applied very fast and “abruptly”, larger inertia forces occur compared to
a slower, “softer” loading. Here the time interval, within which the load is applied, plays a
significant role. This can be either considered by a spectral approach (comp. Chap. 3) or by
a representation of the response vibration with respect to the ratio of the characteristic time
intervals of the loading and the eigenperiods of the system, obtained from the eigenfrequencies
(comp. Sec. 7.5.1.1; response spectra: Figs. 7.21 and 7.24).

Structural Dynamics
summer term 2019
86
7.1.2 Time and frequency representation

Every representation of the excitation as a function of time, i.e., in the time domain, can,
without loss of information, be transformed in an equivalent representation in the frequency
domain (spectral representation).
The representation in the frequency domain simplifies the solutions of the underlying partial
differential equations considerably. Mathematically, this corresponds with solutions of the
equations of motion obtained by sine and cosine approaches. In many practical cases it is suf-
ficient, to limit the evaluation of the structural response to the frequency domain (to spectra).
Every part of the spectrum can thereby be described as a harmonic function (by a sine or
cosine function over time with the radial frequency Ω = 2πf ).
In the frequency domain, the temporal structure of the excitation for dynamic problems with
respect to the expected phenomena can usually be characterized more clearly in a mechanical
and physical sense. Thus, for example, resonance effects can be immediately analyzed and
the sensitivity of the result to changes in the load and system’s parameters can be narrowed
down. However, for the determination of the time function, e.g., for assessing the maximum
displacement over time, a transformation into the time domain is necessary, i.e., a superposition
of the single spectral parts.
For a consideration in the time domain, the state variables (displacements, velocities, strains,
stresses) are directly represented as functions over time. Maximum state variables are thereby
directly obtained.
Material laws and kinematic relations describe the relations between stresses and strains and
displacement and strains, respectively, whereby the state variables are considered as time
dependent. If the material and/or the kinematic is described by non-linear relations (e.g.,
non-linear stress-strain relationship or theory of 2-nd order), the linear superposition of the
responses to different load parts is not valid and the principle of superposition cannot be ap-
plied. In the case of non-linear systems, therefore, a treatment in the time domain is necessary.
If the influence of the non-linearities can be neglected with respect to the vibration response,
e.g., due to very small vibration values, superposing the static initial state, a linearization
is possible. Then the principle of superposition can still be applied and the solution of the
dynamic process can be determined in the frequency domain, after the static initial state has
been described by non-linear analysis. The solution is then finally given by the superposition
of the single spectral parts. As well as representations in the time as well as the frequency
domain have great significance and therefore considered parallelly.

7.2 Harmonic excitation for systems with viscous damping

In this Section we will deduce the solutions for harmonically forced SDOF-systems with viscous
damping. Two cases are of particular interest here, namely the SDOF-system with external,

Structural Dynamics 7 Forced vibration of linear systems with single degree of freedom
summer term 2019 87
harmonic loading, and the base point excited SDOF-system.

7.2.1 Harmonic forced excitation

In the following we investigate the SDOF-system under a harmonically varying load as a basis
for the representation in the frequency domain.

f (t)
m ẅ f01 cos Ωt − f02 sin Ωt
m w m
k c
kw c ẇ

Figure 7.1: Single degree of freedom system with harmonic load

For the depicted harmonically excited SDOF-system the equation of motion is given by (comp.
Sec. 4.1)

mẅ + cẇ + kw = f (t) = f01 cos (Ωt) − f02 sin (Ωt) = f0 cos(Ωt + ϕf ). (7.1)

This linear, inhomogeneous differential equation is solved by a combination of the homogeneous

solution wh (t), which is introduced in chapter 5, and a particular (inhomogeneous) solution
wp (t) as

w (t) = wh (t) + wp (t) (7.2)

As introduced in chapter 2, the description of harmonic oscillations can be done in various

ways. We choose to replace the right hand side of the differential equation (7.1) by a complex
representation, as

f (t) = f01 cos (Ωt) − f02 sin (Ωt)

1 1
= (f01 + if02 ) e iΩt + (f01 − if02 ) e −iΩt
|2 {z } |2 {z }
fb+ fb−

= fb+ e iΩt + fb− e −iΩt (7.3)

Structural Dynamics 7 Forced vibration of linear systems with single degree of freedom
summer term 2019 88
A detailed derivation of the above was given in section 2.2.1. The resulting excitation is
now a sum of two complex terms fb+ eiΩt and fb− e−iΩt . These are complex conjugate as are
their amplitudes, and have the same absolute value of the frequency with opposite signs. The
absolute value of the amplitude of the force f (t), f0 , therefore results in
r
q  2  2
f0 = 2
f01 + 2
f02 = 2 Re fb+ + Im fb+ (7.4)

7.2.1.1 Steady state solution

For the given linear SDOF-system, a harmonic excitation leads to harmonic particular solution,
given as

wp (t) = wbp+ e iΩt + wbp− e −iΩt (7.5)

= wp01 cos (Ωt) − wp02 sin (Ωt)
= wp0 cos(Ωt + ϕw ) (7.6)

where

1 1
wp+ = (wp01 + iwp02 ) = wp0 e iϕw (7.7)
2 2
1 1
wp− = (wp01 − iwp02 ) = wp0 e −iϕw (7.8)
2 2

We continue the derivation with the complex representation of wp (t) in Eq. (7.5), as it simplifies
the calculations. The first two derivatives of (7.5) with respect to t are given by

dwp (t)
ẇp (t) = = iΩwbp+ e iΩt − iΩwbp− e −iΩt (7.9)
dt
d wp (t)
2
ẅp (t) = = −Ω2 wbp+ e iΩt − Ω2 wbp− e −iΩt . (7.10)
dt2

If we now insert approach (7.5) together with (7.9) and (7.10) into the equation of motion
(7.1) and separate the equation for the parts with e iΩt and e −iΩt , we obtain

(−mΩ2 + icΩ + k)wbp+ e iΩt = fb+ e iΩt (7.11)

(−mΩ2 − icΩ + k)wbp− e −iΩt = fb− e −iΩt (7.12)

Structural Dynamics 7 Forced vibration of linear systems with single degree of freedom
summer term 2019 89
After canceling the exponential on both sides of Eqs. (7.11) and (7.12), we can recast the
formulation to

wbp+ = Ub (+Ω)fb+ = Ub+ fb+ (7.13)

wbp− = Ub (−Ω)fb− = Ub− fb− (7.14)

with U (Ω) the spectral transfer function (dynamic flexibility)

1
Ub (Ω) = . (7.15)
k − mΩ2 + icΩ

The inverse of the transfer function

h i−1
k̂+ = Ub (Ω) = (k − mΩ2 + icΩ) (7.16)

and
h i−1
k̂− = Ub (−Ω) = (k − mΩ2 − icΩ) (7.17)

∗
are called dynamic stiffnesses. Note that k̂− = k̂+
∗
as well as Ub (−Ω) = Ub (Ω) are pairs of
complex conjugate values. Furthermore, it holds

wbp− = wbp+
∗
. (7.18)

The solution in the complex plane is depicted in Fig. 7.2. We further continue with

1
wbp+ = fˆ+
k − mΩ2 + ic Ω
k − mΩ2 − ic Ω (f01 + i f02 )
=
(k − mΩ2 )2 + (c Ω)2 2
[(k − mΩ2 )f01 + c Ωf02 ] + i [(k − mΩ2 ) f02 − c Ωf01 ]
= .
2 [(k − mΩ2 )2 + (c Ω)2 ]

With this, we can then isolate real and imaginary part

1 (k − mΩ2 )f01 + c Ωf02

Re (wbp+ ) = (7.19)
2 [(k − mΩ2 )2 + (c Ω)2 ]
1 (k − mΩ2 ) f02 − c Ωf01
Im (wbp+ ) = (7.20)
2 [(k − mΩ2 )2 + (c Ω)2 ]

Structural Dynamics 7 Forced vibration of linear systems with single degree of freedom
summer term 2019 90
 Im

2wp+

2wp+ e iΩt

wp+
wp+ e iΩt
2 Re(wp+ e iΩt ) Ωt
ϕw Re
−ϕw
wp+ + wp− = 2 Re(wp+ )
−Ωt
wp− e −iΩt
wp−

2wp− e −iΩt

2wp−

Figure 7.2: Displacement solution wp (t) in the complex plane. The solution wp (t) is the sum of two
complex conjugate values wp+ e iΩt and wp− e −iΩt . At the time t = 0, the exponential
function reduces to one, and the solution is solely the sum wp (t = 0) = wp+ + wp− . As
time progresses the two complex pointers get rotated in the complex plane by the angle
Ωt and −Ωt, respectively. The solution is always real, as the imaginary parts of wp+ e iΩt
and wp− e −iΩt cancel out for all t.

Correspondingly, for wbp− we obtain

1
wbp− = fb−
k − mΩ2 − ic Ω
k − mΩ2 + ic Ω (f01 − i f02 )
=
(k − mΩ2 )2 + (c Ω)2 2
(k − mΩ2 )f01 − c Ωf02 − i((k − mΩ2 )f02 − c Ωf01 )
= ,
2 [(k − mΩ2 )2 + (c Ω)2 ]

Structural Dynamics 7 Forced vibration of linear systems with single degree of freedom
summer term 2019 91
As the solution wp (t) - just like the force f (t) - consists of two terms that are complex conjugate,
it finally results in (physically correct) real values.
Using Eq. (2.13) and Eq. (2.14), we can decompose the absolute value and argument of |wbp+ |
and |wbp− | into force and transfer function contributions
  
|wbp+ | = Ub+ fb+  (7.21)
  
|wbp− | = Ub− fb−  (7.22)
  

       
Because Ub+  = Ub−  and fb+  = fb− , also |wbp+ | = |wbp− | holds. The corresponding amplitude
       

wp0 in the representation of Eq. (7.6) reads

q q
wp0 = 2
wp01 + wp02
2
= 2 Re(wbp+ )2 + Im(wbp+ )2 = 2|wbp+ | (7.23)

Furthermore, the response phase is found by

ϕw = arg wbp+ = arg Ub+ + arg fb+ = ϕU+ + ϕf (7.24)

The response wp (t) thus shows a phaseshift ∆ϕ = ϕU+ = ϕf − ϕw with respect to the force
f (t). In case of a positive phase shift the response follows behind the force. Using Eqs. (7.19)
and (7.19), absolute value and argument of the transfer function read
    1
U+  = Ub−  = q (7.25)
b   
(k − mΩ2 )2 + (cΩ)2
 
  Im Ub+ −cΩ
tan (∆ϕ) = tan ϕU+ =   = (7.26)
Re Ub+ k − mΩ2

Thus, the amplitude of the response (Eq. (7.23), considering Eq. (7.21)) can be written as:

b  1 f0
     
wp0 = 2|wbp+ | = 2Ub+ fb+  = 2U+  f0 =q (7.27)
  
2 (k − mΩ2 )2 + (cΩ)2

The transfer function is illustrated in the complex plane as shown in Fig. 7.4. In practice, the
magnitude of the response is usually given by the frequency-dependent and real amplification
function V (Ω, ωn ) = V (η) = k|U (Ω)|.

k 1 1
V (η) = q = s =q (7.28)
(k − mΩ2 )2 + (cΩ)2  2  2
(1 − η 2 )2 + (2Dη)2
1− Ω
ωn
+ ( 2D
ωn
Ω)2

Structural Dynamics 7 Forced vibration of linear systems with single degree of freedom
summer term 2019 92
with the frequency ratio η = ωΩn . The amplification function gives the ratio between the
maximum amplitude and the static displacement wstat = fk0 . The phase-shift ∆ϕ describes
by how much the response of the system lags behind the force. For ∆ϕ = 0 the response of
the system is in phase with the load. For ∆ϕ < 0 the oscillation reaches its maximum (or
minimum, zero crossing) at a later time instance relative to the load. For ∆ϕ = −π = −180◦
the response and the excitation oscillate in counterphase.
|V |
5.0
4.5 D=0
D = 0.2
4.0 D = 0.4
3.5
3.0 ∆ϕ
2.5 Vmax = √1
2 D 1−D2 −π
2.0
1.5 D=0
1.0 − π2 D = 0.2
D = 0.4
0.5
0.0 0.0
0.5 1.0 √ 1.5 2.0 2.5 3.0 0.5 1.0 1.5 2.0 2.5 3.0
ηmax = 1 − 2 D2 η η
Figure 7.3: Amplification function for force excitation

For Ω → 0 the dynamic stiffness converges to the static spring stiffness k. For very large values
of Ω the mass term −mΩ2 dominates the dynamic stiffness and the real part of the dynamic
stiffness becomes negative. At resonance (η = 1, i.e. Ω = ωn ), the spring stiffness k and the
mass term −mΩ2 cancel out. The response is then determined by the damping constant c. In
this case the dynamic stiffness becomes purely imaginary.

Example 7.1: SDOF system with cosine load

In the following, we load the SDOF system with a harmonic load that is described by
a cosine function.

0
f (t) = f01 cos (Ωt) −  >sin (Ωt) = f cos (Ωt)
f02

0

which implies, that the phase ϕL = 0 and f0 = f01 , as the amplitude of the sine part
f02 is zero. The resulting response amplitude is given by Eq. (7.23) in terms of the real

Structural Dynamics 7 Forced vibration of linear systems with single degree of freedom
summer term 2019 93
 Im
Im
Re
Im(U− ) U−

D = 0.5
ϕU− Re
ϕU+
Re(U+ ) = Re(U− ) D = 0.2

Im(U+ ) U+
D = 0.1

(a) Complex pointers of the transfer (b) Nyquist plot of the transfer function
function. for positive frequencies and differ-
ent values of damping.
Figure 7.4: Transfer function in the complex plane.

and imaginary part of wbp+ . Under consideration of f02 = 0, it reads

1 (k − mΩ2 )
Re (wbp+ ) = f0 (7.29)
2 [(k − mΩ2 )2 + (c Ω)2 ]
1 −c Ω
Im (wbp+ ) = f0 . (7.30)
2 [(k − mΩ2 )2 + (c Ω)2 ]

Inserting Eqs. (7.29) and (7.30) into Eq. (7.23), we obtain

q q
wp0 = 2
wp01 + wp02
2
= 2 Re(wbp+ )2 + Im(wbp+ )2
v
u (k − mΩ2 )2 + (cΩ)2
u
= 2f0 u
t  2 =
4 (k − mΩ2 )2 + (cΩ)2
1 f0 1
=q f0 = q =
(k − mΩ2 )2 + (cΩ)2 k (1 − η 2 )2 + (2Dη)2
| {z } | {z }
=|U (Ω)| =V (η)

f0
= |U (Ω)| f0 = V (η)
k

Structural Dynamics 7 Forced vibration of linear systems with single degree of freedom
summer term 2019 94
 Ω

ωn

ωt
|U |
Re(U )
1
k

Im(U )
Figure 7.5: Transfer function of the SDOF system over frequency (—) and real (—) and imaginary
(—) part.

where η = ωΩn is the frequency ratio and V (η) is the amplification function. We note that
the amplification function is linked to the spectral transfer function via k |U (Ω)| = V (η).
Thus, the response amplitude can be calculated from the force amplitude by a multi-
plication with the quotient of the amplification function V (η) and the stiffness k. The
response lags behind the excitation, as the imaginary part of wbp+ is negative. As for
the cos-excitation the phase of the force ϕf is zero, the phase shift is equal to the phase
of the response, which follows to

Im(wbp+ ) −cΩ 2Dη

tan (ϕw ) = = =−
Re(wbp+ ) k − mΩ 2 1 − η2

with c = 2Dωn m (comp. Tab. 5.1). The equations of motion (7.11) and (7.12) consti-
tute a dynamic equilibrium and can be illustrated using pointer diagrams.

Structural Dynamics 7 Forced vibration of linear systems with single degree of freedom
summer term 2019 95
 t
−Ω2 mwp+
Im Im iΩcwp+
t

F+t
t
kwp+
Ωt Re
F+ e iΩt
Im(wp+ e iΩt ) wp+ e iΩt

∆ϕ = ϕwp+

Ωt Ωt − ∆ϕ F+ = F− Re
−Ωt −Ωt + ∆ϕ

∆ϕ = ϕwp−
Im

Im(wp− e −iΩt ) Re
wp− e −iΩt
F− e −iΩt Ωt
t
kwp−
F−t

t
−Ω2 mwp− t
−iΩcwp−
Figure 7.6: Particular solution in the complex plane. The illustration shows the two pairs
of complex conjugate force and displacement values f+ and f− as well as wp+
and wp− at time t. As the sine term is zero, the force term lies on the real axis
at time t = 0. For the sake of clarity the pointers wp+ and wp− at time t = 0
are not depicted. Both force and response rotate in the complex plane with the
rotation angle Ωt. The positive term rotates counterclockwise (mathematically
positive) whereas the negative term rotates clockwise (mathematically negative).
The response lags behind the force by an angle of ∆ϕ. Again the sum of both
parts results in real output values. The equilibrium of external force f and internal
forces fk , fc and fm in the spring, damper and mass, respectively, is depicted
for the positive and negative pointer at time t (indicated by superscript (·)t , such
t
that, i.e. wp− = wp− e −iΩt ).

Structural Dynamics 7 Forced vibration of linear systems with single degree of freedom
summer term 2019 96
7.2.1.2 Transient response

We are now interested in obtaining the full response of a system, on which a harmonic load
starts acting at time t = 0. We assume, that for time t = 0, the system is at rest or
has otherwise defined initial conditions w (t = 0) and ẇ (t = 0) = v (t = 0). Then, the full
solution is given by the superposition of homogeneous solution wh (t) (depending on damping,
Eqs. (5.22), (5.24), (5.25)) and the particular solution wp (t).

w (t) = wh (t) + wp (t) (7.31)

ẇ (t) = ẇh (t) + ẇp (t) (7.32)

The constants w01 and w02 of the homogeneous solution are determined from the initial con-
ditions w (t = 0) and v (t = 0) = ẇ (t = 0). After the homogeneous solution has decreased, in
the steady state only the particular solution wp (t) needs to the be taken into account. Due to
damping the homogeneous part of the solution wh (t) decays for increasing time. In the limit
t → ∞, we observe wh (t) → 0 for the homogeneous part.
Vibrations with large amplitudes can occur in the starting phase of the vibration. This holds
especially for systems with low eigenfrequencies and is especially important for high frequency
excitation on systems at rest. The transient response of a system initially at rest over time is
exemplarily shown in Fig. 7.7.

w(t)
w(t) = wh (t) + wp (t)
wp (t)

wh (t)

Figure 7.7: Transient response of forced SDOF system. We observe that the homogeneous part of
the solution has almost completely decayed out after 5 periods. Nevertheless, the full
solution shows a higher maximum response amplitude than would have been predicted
by the particular solution only.

Structural Dynamics 7 Forced vibration of linear systems with single degree of freedom
summer term 2019 97
The full solution is given without further derivation:

f0
w(t) = e −δt (w01 cos (ωD t) − w02 sin (ωD t)) + V (η) cos (Ωt + ϕw ) (7.33)
k

with:

f0
w01 = w(t = 0) − V (η) cos (ϕw ) (7.34)
k
1
" # !
f0 f0
w02 =− v(t = 0) + w(t = 0) − V (η) cos (ϕw ) + V (η)Ω sin (ϕw ) . (7.35)
ωD k k

In this App the response due to a harmonic load is visualized. The homoge-
nous and particular solution as well as the total solution are plotted. One can
visualize that the homogenous solution is important for the initial conditions
and over time its influence tends to disappear according to the damping
constant.
http://www.bm.bgu.tum.de/lehre/interactive-apps/
single-degree-of-freedom-system/

7.2.1.3 Resonance case

For η = 1, i.e., when excitation frequency and eigenfrequency coincide (Ω = ωn ), weakly

damped systems undergo very large amplitudes. In an undamped system the dynamic stiffness
kb+ (comp. Eq. (7.16)) would tend to zero, as k and −mΩ2 cancel each other out, and the
resulting response amplitude would become infinitely large. Often, it is of interest to know
about the system response, when the excitation is only in the resonance frequency for a short
time, e.g. when the rotational speed of a machine is increased and the region of resonance is
passed.
We begin with the derivation of the solution of the damped SDOF system in the case of
resonant excitation, i.e., Ω = ωn and η = 1. Here, we assume that the force has zero phase,
i.e., ϕf = 0, and it follows a cosine function f (t) = cos(ωn t). With this, the particular solution
reads

f0
wp (t) = V (1) cos(ωn t + ∆ϕ). (7.36)
k
 
For η = 1 the phase shift ∆ϕ is π2 . Using the identity cos α + π
2
= − sin (α), and inserting

Structural Dynamics 7 Forced vibration of linear systems with single degree of freedom
summer term 2019 98
V (1) = 1
2D
, Eq. (7.36) reads

f0
wp (t) = − sin(ωn t). (7.37)
2Dk

The homogeneous solution to the problem reads

wh (t) = e −δt (wh1 cos(ωD t) − wh2 sin(ωD t)) . (7.38)

The total solution

f0
w(t) = wh (t) + wp (t) = e −δt (wh1 cos(ωD t) − wh2 sin(ωD t)) − sin(ωn t) (7.39)
2Dk

has to satisfy the boundary conditions. We assume that at t = 0 the system is at rest, i.e.,
w0 = v0 = 0. It follows

!
w(t = 0) = wh1 = 0 (7.40)

With wh1 = 0, the velocity reads

f0
ẇ(t) = −wh2 δe −δt sin(ωD t) − wh2 e −δt ωD cos(ωD t) − ωn cos(ωn t) (7.41)
2Dk

Under the condition ẇ(t = 0) = 0, we obtain

ωn f0
wh2 = . (7.42)
ωD 2Dk
√
We insert ωD = 1 − D2 ωn and δ = Dωn , and write the full solution as

1 f0 f0
w(t) = e −Dωn t √ sin(ωD t) − sin(ωn t) (7.43)
1 − D 2Dk
2 2Dk
f  √ 
= √0 e −Dωn t
sin(ωD t) − 1 − D 2 sin(ω t)
n (7.44)
2Dk 1 − D2
(7.45)
√
For small damping ratios D << 1, we assume 1 − D2 ≈ 1, and ωD ≈ ωn , and obtain

f0  −Dωn t 
w(t) = e − 1 sin(ωn t) (7.46)
2Dk

Structural Dynamics 7 Forced vibration of linear systems with single degree of freedom
summer term 2019 99
The resulting response is illustrated in Fig. 7.8.

w(t)
F0
2Dk

ωt

F0
− 2Dk

Figure 7.8: Transient response

f0 −Dωn t

in resonance for D = 0.1. The outer lines show the envelope function
2Dk e − 1 . The response amplitude is bounded by the maximum amplitude of
f0
the particular solution 2Dk .

For undamped systems with D = 0, Eq. (7.46) is not defined, as a division with D occurs. We
can find the limiting value

f0  −Dωn t 
lim w(t) = lim e − 1 sin(ωn t) (7.47)
D→0 D→0 2Dk

by application of the rule of l’Hospital. We define w(t) = f (D)

g(D)
with

f0  −Dωn t 
f (D) = e − 1 sin(ωn t) (7.48)
2k
g(D) = D (7.49)

and find

df f0
= −ωn t e −Dωn t sin(ωn t) (7.50)
dD 2k
dg
= 1. (7.51)
dD

Thus, the limiting value in Eq. (7.47) reads

f0
lim w(t) = − ωn t sin(ωn t) . (7.52)
D→0 2k

Structural Dynamics 7 Forced vibration of linear systems with single degree of freedom
summer term 2019 100
The resulting response is illustrated in Fig. 7.9).

w(t)

ωt

Figure 7.9: Transient response in resonance in the limit case D → ∞. Under the assumption of
vanishing damping, the response is unbounded and tends to ∞ as t → ∞. The envelope
function is linear. The dash-dotted curve shows the response for a damping ratio D = 0.1
from Fig. 7.8.

7.2.1.4 Transmissibility

In the design of elastic bearings, the forces fk (t) and fd (t) that are transmitted over the spring
and the damper, play an significant role. In the steady state (stationary solution) they are
given by

fk (t) = k w(t) = k wp0 cos (Ωt + ϕw ) (7.53)

fd (t) = c ẇ(t) = −c Ω wp0 sin(Ωt + ϕw ) (7.54)

The amplitude of the sum of both forces |fU max | can be calculated using the harmonic addition
theorem (see 2.2).
s
q q c2 Ω2
|fUmax | = fk2max + fd2max = 2
k 2 wp0 + c2 Ω2 wp0
2
= kwp0 1 + (7.55)
k2

m2
and with c2 = (2mDωn )2 , k2
= 1
ωn4, it follows
v
ωn2 2
u
|fU max | = kwp0 t1 + (2 D)2 Ω =
u
ωn4
q
= kwp0 1 + (2Dη)2 =

Structural Dynamics 7 Forced vibration of linear systems with single degree of freedom
summer term 2019 101
We further insert wp0 = f0
k
V (η) and obtain
q
f0 1 + (2Dη)2
=k q =
k (1 − η 2 )2 + (2Dη)2
q
1 + (2Dη)2
= f0 q . (7.56)
(1 − η 2 )2 + (2Dη)2

The ratio of the force it is transmitted into the ground and the force that acts on the system
is called transmissibility T (η). It follows to
q q
fk2 + fd2 1 + (2Dη)2
T (η) = = q . (7.57)
f0 (1 − η 2 )2 + (2Dη)2

7.2.2 Harmonic root point excitation

In structural dynamics often root point excited vibrations occur, e.g. for earthquakes or the
excitation through vibrations in the soil (cf. lecture Soil Dynamics). In the following we
assume a harmonic motion we (t) of the system’s support.

mẅ

m
w(t)
m
k c
k(w − we ) c(ẇ − ẇe )
we (t)

Figure 7.10: SDOF system with root point excitation.

7.2.2.1 Equation of motion

The harmonic motion of the support is described through

we (t) = we01 cos (Ωt) − we02 sin (Ωt) = we0 cos(Ωt + ϕe ) (7.58)
ẅe (t) = −Ω2 we (t) . (7.59)

Structural Dynamics 7 Forced vibration of linear systems with single degree of freedom
summer term 2019 102
The equation of motion is obtained from the equilibrium (see Fig. 7.10)

mẅ + c (w − we ) + k (w − we ) = 0 . (7.60)

We introduce the relative displacement wr = w − we (distance between mass and support) and
find for the SDOF acceleration

ẅ = ẅr + ẅe = ẅr − we0 Ω2 cos(Ωt + ϕe ) . (7.61)

We choose the relative displacement wr as unknown in the equation of motion Eq. (7.60) and
obtain

mẅr + cẇr + kwr = −mẅe = we0 mΩ2 cos(Ωt + ϕe ) . (7.62)

7.2.2.2 Particular solution (forced vibration)

The structure of the differential equation of the root-point excited SDOF system corresponds
to the structure of differential equation of the SDOF system under harmonic force (Eq. (7.1)).
The solution for the relative displacement wr can thus be given analogously to the particular
solution for the forced vibration (Eqs. (7.6), (7.26) and (7.27)):

wrp = wrp0 cos (Ωt + ϕwr ) Eq. (7.6): wp = wp0 cos (Ωt + ϕw ) (7.63)
we0 mΩ2 f0
wrp0 = q Eq. (7.27): wp0 = q (7.64)
(k − Ω2 m) + Ω2 c2
2
(k − Ω2 m)2 + Ω2 c2
−Ωc −Ωc
tan ∆ϕr = Eq. (7.26): tan ∆ϕ = (7.65)
k − Ω2 m k − Ω2 m

and

ϕwr = ϕe + ∆ϕr . (7.66)

The displacement of the mass w(t) thus reads

w(t) = wrh (t) + wrp (t) + we (t) = wrh (t) + wrp0 cos (Ωt + ϕwr ) + we0 cos(Ωt + ϕe ) . (7.67)

The results can again be illustrated as a function of the excitation frequency Ω. To this end,
the amplification function Vr . It is defined as the ratio of the relative displacement wr0 and

Structural Dynamics 7 Forced vibration of linear systems with single degree of freedom
summer term 2019 103
the amplitude of the support motion we0 , i.e.,

 wrp0  Ω2 m η2
 
Vr =   =q =q . (7.68)
we0 (k − Ω2 m)2 + Ω2 c2 (1 − η 2 )2 + (2Dη)2
 

The phase shift of the mass motion to the support motion ϕwr is then given as

2Dη
tan ∆ϕr = − . (7.69)
1 − η2

|Vr |
5.0
4.5 D=0
D = 0.2
4.0 D = 0.4
3.5
3.0 ∆ϕr
2.5 Vmax = √1
2 D 1−D2 −π
2.0
1.5 D=0
1.0 − π2 D = 0.2
0.5 D = 0.4
0.0 0.0
0.5 1.0 1.5 2.0 2.5 3.0 0.5 1.0 1.5 2.0 2.5 3.0
ηmax = 1−2 D2
√ 1 η η

Figure 7.11: Amplification function and phase shift (root point excitation)

Fig. 7.11 shows that for η → 0 Vr → 0 (and thus wr0 → 0) and ϕwr → 0 holds. For very
low frequencies and thus slow motions no relative displacements occur, which means that the
mass moves like the support. For very high frequencies and thus fast motion, i.e., η → ∞, it
is Vr → 1, ϕwr → π, and wrp0 → we0 . In this case, the mass stays at rest. Due to its inertia it
cannot follow the fast motion of the ground. For η → 1 resonance occurs.

7.2.2.3 Transmissibility

Die maximum spring and damping forces fkmax and fcmax are given by

fkmax = kwr0 (7.70)

fcmax = cΩwr0 . (7.71)

Structural Dynamics 7 Forced vibration of linear systems with single degree of freedom
summer term 2019 104
Thus, the maximum force transmiited to the support fUmax is
q q
fUmax = fk2max + fc2max = 1 + (2Dη)2 kwe0 Vr . (7.72)

Also for the root point excited system, during the transient response, much higher forces can
occur from the superposition with the homogeneous solution (comp. 7.2.1.3).

7.3 Harmonic forced excitation for hysteretically damped

systems

In this Section we will deduce the solutions for harmonically externally forced SDOF-systems
with ideal hysteretic damping, which is introduced in Section 4.3.2. As the ideal hysteretic
damping model is defined in the frequency domain a time domain representation is not straight-
forward. Nevertheless, an equilibrium of forces in the frequency domain is depicted in the
following for better understanding. Note that only the part for the positive time exponential
are considered for clarity. In Eq. (4.41), the damping force in the frequency domain is defined
as

f˜D (ω) = iηk sgn(ω)ũ(ω) . (7.73)

Combined with the real spring stiffness, the total force of elastic restoring force and damping
force is often given as the resulting force of a complex spring:

f˜
−mω 2 w̃ f˜

m w̃ m
k(1 + i sgn(ω)η)
k(1 + i sgn(ω)η)w̃

Figure 7.12: Forced SDOF system with linear hysteretic damping

For the depicted harmonically excited SDOF-system the equation of motion in the frequency
domain is given by (comp. Sec. 4.1):

k(1 + i sgn(ω)η) − mω 2 w̃(ω) = f˜(ω) .

 
(7.74)

Structural Dynamics 7 Forced vibration of linear systems with single degree of freedom
summer term 2019 105
Thus, the following transfer function can be derived:

1  1
k(1+iη)−mω 2
= U+ for ω > 0
U (ω) = = . (7.75)
k(1 + i sgn(ω)η) − mω 2 1
k(1−iη)−mω 2
= U− for ω < 0

Despite the fact that there is no obvious physical interpretation, the loss factor η can still be
given in terms of the damping ratio D as

η = 2D. (7.76)

The resulting Nyquist plot (representation in the complex plane) is given in the following
Figure.

Im
Im
Re
Im(U− ) U−

D = 0.5
ϕU− Re
ϕU+ D = 0.2
Re(U+ ) = Re(U− )

Im(U+ ) U+ D = 0.1

(a) Complex pointers of the transfer (b) Nyquist plot of the transfer function
function. for positive frequencies and differ-
ent values of damping.
Figure 7.13: Transfer function for linear hysteretic damping in the complex plane.

Structural Dynamics 7 Forced vibration of linear systems with single degree of freedom
summer term 2019 106
 |V |
5.0
4.5 D=0
D = 0.2
4.0 D = 0.4
3.5
3.0 ∆ϕ
2.5 Vmax = 1
2D −π
2.0
1.5 D=0
1.0 − π2 D = 0.2
D = 0.4
0.5
0.0 0.0
0.5 1.0 1.5 2.0 2.5 3.0 0.5 1.0 1.5 2.0 2.5 3.0
ηmax = 1 η η
Figure 7.14: Amplification function harmonically forced SDOF system with linear hysteretic damping.

Structural Dynamics 7 Forced vibration of linear systems with single degree of freedom
summer term 2019 107
7.4 Arbitrary periodic excitation

7.4.1 Decomposition into harmonics

According to the principle of superposition, the steady state response (particular solution) of
linear systems to a dynamic loading—that consists of different sine and cosine parts—can be
obtained by the superposition of the different partial solutions. Assuming an excitation with

nF
F (t) = F0,i cos(Ωi t + ϕF,i ) = F0,1 cos(Ω1 t + ϕF,1 ) + F0,2 cos(Ω2 t + ϕF,2 ) + . . . , (7.77)
X

the particular solution of the system reads

nF
wp (t) = w0,i cos(Ωi t + ϕw,i ) = w0,1 cos (Ω1 t + ϕw,1 ) + w0,2 cos (Ω2 t + ϕw,2 ) + . . . (7.78)
X

with

F0,i
w0,i = q (7.79)
2
(k − mΩ2i ) + (cΩi )2
ϕw,i = ϕF,i + ϕU,i (7.80)
cΩi
tan ϕU,i = − . (7.81)
k − mΩ2i

Fig. 7.15 shows an example of an arbitrary periodic force.

Figure 7.15: Example for an arbitrary periodic load time history. The signal is repeated with period T ,
which corresponds to a fundamental radial frequency Ω0 = 2π T

Structural Dynamics 7 Forced vibration of linear systems with single degree of freedom
summer term 2019 108
 Reminder 7.1: Fourier series

A short repetition of Fourier series (comp. Section 3.1.1) shall be given in the following.
An arbitrary periodic loading can be represented by a Fourier series
∞
a0 X
f (t) = + (ak cos (kω0 t) + bk sin (kω0 t))
2 k=1

with ωk = kω0 and ω0 = 2π T

. The coefficients of the individual parts are determined
in such a way, that the mean-square error between the function F (t) and the Fourier
series is minimal.

ZT ∞
!2
a0 X
ε= f (t) − − (ak cos (ωk t) + bk sin (ωk t)) dt
2 k=1
0

The Fourier-coefficients follow to

T
2Z
ak = f (t) cos (ωk t) dt for k = 0,1,2, . . . (7.82)
T
0
T
2Z
bk = f (t) sin (ωk t) dt for k = 1,2, . . . (7.83)
T
0

Structural Dynamics 7 Forced vibration of linear systems with single degree of freedom
summer term 2019 109
 Example 7.2: Response of SDOF system to square-wave

In this example we will derive the Fourier series of the square-wave function depicted
in Fig. 7.16.

f
f0

Figure 7.16: Square wave with period T .

The Fourier series representation of this signal is derived in Example 3.1. The consid-
ered signal is an odd function, i.e., f (t) = −f (−t), which results in all aj = 0 and thus
all cosine terms in the series vanish. The Fourier coefficients bj are given by:
 T 
T
2 Z2 Z
bj =  f0 sin(ωj t) dt + (−f0 ) sin(ωj t) dt
 
T  
0 T
2
T
4 1 2
= − f0 cos(ωj t)
T ωj 
0

 4f0 for j = 1,3,5, . . .
= jπ
0 for j = 0,2,4, . . .

Then, the Fourier series representation of f (t) reads

4f0 X∞
1
f (t) = sin (ωj t) .
π j=1,3,... j

Using the amplification function (comp. Eq. (7.28)), the amplitude of the response of

Structural Dynamics 7 Forced vibration of linear systems with single degree of freedom
summer term 2019 110
 the SDOF system can be given as

f0
w0 = V (η)wstat = V (η) .
k

Further considering the phase shift between load and response, we obtain for the dis-
placement response:

4F0 X 1 1
w(t) = q sin(jω0 t + ϕw,j ) .
π j=1,3,... j (k − m(jω0 )2 )2 + (cjω0 )2

with

ckω0
tan ϕw,j = − .
k − m(kω0 )2

For the superposition of the individual harmonic parts the phase angle of each harmonic
has be taken into account. For a damped system with c 6= 0 they will vary between
0 and π. The following illustrations show the decomposition of load and response into
the individual harmonics.
Below, the resulting series representation is depicted for maximum j of 7. We assume
that no damping is present, thus the phase shift can easily be incorporated as being
either 0 or −π. Furthermore, the system is such that the eigenfrequency ωn corresponds
to four times the fundamental frequency ω0 .

Structural Dynamics 7 Forced vibration of linear systems with single degree of freedom
summer term 2019 111
 f (t)
(1 + 3 + 5) (1) (1 + 3)
(1 + 3 + 5 + 7)
f0

t
T
2

w(t) (1 + 3) (1 + 3 + 5 + 7)
(1) (1 + 3 + 5)

t
T
2

Structural Dynamics 7 Forced vibration of linear systems with single degree of freedom
summer term 2019 112
 Fourier spectrum
of the external force
f (ωj )
with eigenfrequency
4F0
π (1) ωn = 4ω0

( 13 )
( 51 ) ( 17 )
ωj
j= ω0
1 3 5 7

V (Ω)
Amplification function

V (Ω = ωj ) = √(k−mωk2 )2 +(cω 2
j j)

2,908
1,216
1,079
1 0,389
j
1 3 5 7
4 (resonance)

f (Ω)
k
V (ωj ) F ourier spectrum
of the response
4F0
(1,079)
π (0,969) (only amplitude spectrum)

(0,234)
(0,056)
j
1 3 5 7

Structural Dynamics 7 Forced vibration of linear systems with single degree of freedom
summer term 2019 113
7.5 Aperiodic excitation

7.5.1 Time domain

7.5.1.1 Unit impulse response function

Additionally to the response of the SDOF-system to harmonic loads, we can also directly
determine the response to an impulse. The impulse of a force that acts upon the system
during a time frame ∆t reads

− ∆t
2
Z
I0 = f (t) dt (7.84)
− ∆T
2

f (t)

f (t) = I0 δ(t)

t
∆t

Figure 7.17: Force impulse

If the limit ∆t tends to be zero, while the integral retains a finite value, the force must tend
to infinity, which can be written in a symbolic manner using the δ-Dirac-distribution.

f (t) = I0 δ(t) . (7.85)

The impulse then follows to

− ∆t
2
− ∆t
2
Z Z
I = lim f (t) dt = lim I0 δ(t) dt = I0 . (7.86)
∆t→0 ∆t→0
− ∆t
2
− ∆t
2

At the start of vibration (undeformed configuration), all other forces in a SDOF system (the
spring force kw, the damping force cẇ) are negligible in comparison to f (t). The system is at

Structural Dynamics 7 Forced vibration of linear systems with single degree of freedom
summer term 2019 114
rest, i.e., w(t = 0) = 0. The conservation of momentum reads

− ∆t
2
Z
mv0+ = 
m +
0− f (t) dt = I0 ,
− ∆t
2

with the initial velocity before the impact v0− and the velocity after the impact v0+ . If
additionally v(t− = 0) = v0− = 0, then

I0
mv0+ = I0 → v0 = v0+ = .
m

For the case of the undercritically damped SDOF system with the initial velocity v0 = I0
m
, we
obtain the following homogeneous response

w(t) = I0 h(t) (7.87)

where h(t) is the unit impulse response function, defined as

1 √ 
h(t) = q e−Dωn t sin 1 − D 2 ωn t . (7.88)
km(1 − D2 )

Fig. 7.18 shows the unit impulse response function for varying percentage of critical damping
D.
√
kmh(t)
1 D=0
D = 0.05
D = 0.7
D = 0.2
ωn t

−1

Figure 7.18: Unit impulse response function for varying percentage of critical damping D; D = 0 (—),
D = 0.05 (–·–), D = 0.2 (– –), D = 0.7 (· · ·). For larger D the response decays more
quickly.

Structural Dynamics 7 Forced vibration of linear systems with single degree of freedom
summer term 2019 115
7.5.1.2 Superposition, Duhamel-Integral

For linear systems, the response to an arbitrary loading can be calculated by the superposition
of the responses to a sequence of impulses. In order to determine the response at the time t
the responses caused by all impulses f (τ ) dτ are summed up. At time t the response to an
impulse at time τ is equal to dI h(t − τ ) where dI = f (τ ) dτ . The total response can finally
be obtained as the sum of all responses that have acted until time t.

τ =t
Z
w(t) = f (τ )h(t − τ ) dτ .
τ =0

The principle is depited in Fig. 7.19.

f
f (τ )dτ w h(t − τ )

f (t)
t
τ t−τ
τ dτ t
t

Figure 7.19: Superposition of responses to single impulses. The left hand side figure shows the
impulse dI acting on the system at time τ , dI = f (τ ) dτ . This impulse causes the
response that is given by Eq. (7.87) and depicted on the right hand side. The unit
impulse response function h(t) is shifted to the time τ and scaled with F (τ ). Due to this
we subtract τ in the argument of h(t − τ ).

Due to the fact, that h(t) = 0 for t ≤ 0 1 , we can rewrite Eq. (??) as follows .

Z∞
w(t) = f (τ )h(t − τ ) dτ (7.89)
τ =0

Furthermore, we consider f (τ ) = 0 for τ ≤ 0, and thus

Z∞
w(t) = f (τ )h(t − τ ) dτ (7.90)
−∞

1
This follows from causality. An impulse cannot cause a response before its impact.

Structural Dynamics 7 Forced vibration of linear systems with single degree of freedom
summer term 2019 116
The given mathematical operation in the equation is called convolution. For the viscously
damped SDOF system, the Duhamel integral reads:

Zt
1 √ 
w(t) = q f (τ )e−Dωn (t−τ ) sin 1 − D2 ωn (t − τ ) dτ . (7.91)
km(1 − D2 ) −∞

For the undamped SDOF system Eq. (7.91) simplifies to

t
1 Z
w(t) = √ f (τ ) sin (ωn (t − τ )) dτ . (7.92)
km −∞

Note that by entering the impulse response function of the SDOF system, we can adjust the
limits of integration. This is because h(t) = 0 for t < 0, i.e. no response can occur before
an excitation (causality). In the definition of the impulse response function this is implicitly
assumed without further specification. 2

Example 7.3: Undamped SDOF system under step load

We calculate the response of a linear SDOF system subjected to the step load depicted
in Fig. 7.20.

f (t)

f0

t0 t

Figure 7.20: Step load for a SDOF system

The Duhamel integral reads

Zt t
f0 Z
t < t0 : w(t) = f0 h(t − τ ) dτ = √ sin (ωn (t − τ )) dτ
0
km 0

2
A more rigorous definition could be done by the consideration of the cases t < 0 and t ≥ 0 or by multiplication
with the Heaviside function.

Structural Dynamics 7 Forced vibration of linear systems with single degree of freedom
summer term 2019 117
 t
f0 1  f0
=√ cos (ωn (t − τ ))  = (1 − cos (ωn t)) (7.93)
km ωn 
0
k
Zt0
f0
t > t0 : w(t) = f0 h(t − τ ) dτ = [cos ωn (t − t0 ) − cos (ωn t)] . (7.94)
k
0

In general, the solution can also be found by already found solutions to the differential
equation of the SDOF system. For this the response is determined superimposing results
from two time-shifted opposite load jumps. In principle the solution can also be found
by separating the ranges, and solving the system separately under consideration of the
initial and transfer conditions. In the analysis often only the maximum displacment
wmax is of interest. Once the duration of the excitation exceeds half the length of the
period, which implies t0 > T2n , wmax reaches the value wmax = 2fk0 within the interval
0 < t ≤ t0 . In all other cases, the maximum value occurs at t1 > t0 and can be found
by finding the maximum in Eq. (7.94). Thus, we state

f0
t > t0 : w = [cos (ωn (t − t0 )) − cos (ωn t)] (7.95)
k
!
ẇ = 0 → t1 → wmax = w(t1 ) (7.96)
2f0 t0
 
wmax = sin ωn . (7.97)
k 2

Note that as the system is undamped, the corresponding maximum value may be found
repeatedly.

7.5.1.3 Response spectra

The results to peak value searches are presented in so called response-, impact-, or shock
response-spectra. The response spectrum shows the maximum response value depending on
the eigenfrequency or eigenperiod of the system, without noting the corresponding time of
occurrence. Typically, a characteristic time of excitation, relative to the eigenperiod Tn = f1n , is
shown dimensionless on the abscissa. The ordinate then usually shows the maximum dynamic
amplitude relative to the static displacement. With this the maximum response quantity can
be solely deduced in dependence on eigenfrequency and damping. Figure 7.21 exemplarily
shows the response spectrum of the undamped SDOF system under rectangular excitation of
Fig. 7.20 in example 7.3.

Structural Dynamics 7 Forced vibration of linear systems with single degree of freedom
summer term 2019 118
 kwmax
F0

t > t0 t < t0
2

t0
Tn
0,5 1

Figure 7.21: Response spectrum of SDOF system for rectangular excitation with duration t0 and am-
plitude f0 using Eq. (7.94) and Eq. (7.97). The graph depicts the maximum response
value as a function of the ratio of load duration and eigenperiod of the system. For a
load duration t0 shorter than half of the eigenperiod of the system, the maximum value
will be reached after the load has stopped acting and vice versa, for t0 longer than half
of the eigenperiod of the system, the maximum value will be reached before the load
has stopped acting.

Response spectra are also used for the characterization of stochastic excitations, e.g. in earth-
quake engineering (DIN [2006]). In the case of root point excitation the force term f (t) needs
to be replaced withqthe load term of −m ẅe (t) of Eq. (7.62) in the Duhamel integral in Eq. (??).
Considering ωn = m k
, the integral for determining the realtive displacement simplifies to

τ =t
Z
1 √ 
wr (t) = −ẅe (τ ) q e−Dωn (t−τ ) sin 1 − D2 ωn (t − τ ) dτ . (7.98)
τ =0 ωn (1 − D2 )

Structural Dynamics 7 Forced vibration of linear systems with single degree of freedom
summer term 2019 119
 Example 7.4: Triangular shock

The response spectrum can also be given as a function of Tn/t0 . As an example consider
a massless, flexible column with varying length and attached mass at its top end. A
load acts on the mass point. For increasing length of the system, the stiffness decreases,
leading to decreasing eigenfrequencies and thus, increasing eigenperiods.

f (t)
f0
t
t0

f (t)

increasing Tn

Figure 7.22: Triangular shock on column with varying length

Depending on the ratio of eigenperiod to load duration, the maximum response value
is given in the following diagram.

V = max w
f0

max V = 1,25
k

V3 V1
V2
Tn = 2π
ωn

0
0,1 0,56 1 5 10 Tn
t0
[log]

Figure 7.23: Response spectrum of SDOF system for triangular excitation with duration t0
and maximum amplitude f0 . The graph depicts the maximum response value
as a function of the ratio of the eigenperiod of the system Tn and the load
duration t0 .

Structural Dynamics 7 Forced vibration of linear systems with single degree of freedom
summer term 2019 120
Response spectra for the SDOF system under different impact time histories are depicted in
Fig. 7.24 (cf. Petersen [1996]).

V V V
2.0 2.0 2.0

1.5 1.5 1.5

1.0 1.0 1.0

0.5 t
t0
0.5 t
t0
0.5 t
t0

0 0 0
0 1.0 2.0 3.0 4.0 0 1.0 2.0 3.0 4.0 0 1.0 2.0 3.0 4.0
t0 t0 t0
Tn Tn Tn
V V V
2.0 2.0 2.0

1.5 1.5 1.5

1.0 1.0 1.0

0.5 t0
t 0.5 t0
t 0.5 t0
t

0 0 0
0 1.0 2.0 3.0 4.0 0 1.0 2.0 3.0 4.0 0 1.0 2.0 3.0 4.0
t0 t0 t0
Tn Tn Tn

Figure 7.24: Response spectra - overview from Petersen [1996]

7.5.2 Exkursus: Kelvin-Voigt model

In Section 7.1.1, we discuss the distinction between “quasi-static” and dynamic processes.
To consider dynamic processes, the influence of inertia and damping forces are considered.
However, also for processes, which vary very slowly in time, a static consideration of the
problem does not lead to a full understanding of the process. This is for example the case for
creep processes. To illustrate this, we derive the response of the Kelvin-Voigt model (spring
and damper in parallel) to a step load.

Structural Dynamics 7 Forced vibration of linear systems with single degree of freedom
summer term 2019 121
 Example 7.5: Kelvin-Voigt model under step load

Consider the Kelvin-Voigt model governed by the differential equation (4.20), slightly
reformulated to

k f
u̇ + u =
c c

where the stress is applied abruptly at t = 0 and then held constant, thus

f
0 for t ≥ 0
f (t) = U (t) =  .
0 for t < 0

In order to obtain the homogeneous solution, the following approach is given

uh = uh0 e λt .

Inserting into the ODE (f = 0) gives

k
λuh0 e λt + uh0 e λt = 0
c
!
k
⇔ λ+ uh0 e λt = 0
c

Thus,

k
λ=−
c

and
k
uh = uh0 e − c t

with uh0 being undetermined for now. As f (t) is constant for t ≥ 0, the particular
solution up (t) is the constant function

up (t) = up0 .

Structural Dynamics 7 Forced vibration of linear systems with single degree of freedom
summer term 2019 122
 Inserting into the ODE gives

k f0
up0 =
c c
f0
⇔ up0 = .
k

The full solution then reads

k f0
u(t) = uh (t) + up (t) = uh0 e − c t + .
k

The full solution now has to satisfy the initial condition, chosen here as u(0) = 0. Thus,

f0
u(0) = uh0 + =0
k
f0
⇔ uh0 = − .
k

Finally, we then obtain

f0 − k t f0 f0  k

u(t) = − e c + = 1 − e−ct .
k k k

The resulting displacement solution is depicted in the following.

u
f0
k

Figure 7.25: Displacement response response of Kelvin-Voigt model to applied unit-step

stress.

Due to the neglection of inertia forces in the equation of motion, this model does not
give a oscillatory response.

Structural Dynamics 7 Forced vibration of linear systems with single degree of freedom
summer term 2019 123
7.5.3 Frequency domain
Reminder 7.2: Fourier transformation

The Fourier transformation is defined by

Z∞
f˜(ω) = f (t)e −iωt dt
−∞
∞
1 Z ˜
f (t) = f (ω)e iωt dω .
2π
−∞

7.5.3.1 Response to harmonic excitation

For harmonic excitation f (t) = fb+ e iΩt + fb− e iΩt , the steady state response is

wb = wb+ e iΩt + wb− e −iΩt = Ub (Ω)fb+ e iΩt + Ub (−Ω)fb− e iΩt (7.99)

wb+ = Ub (Ω)fb+ (7.100)
wb− = Ub (−Ω)fb− . (7.101)

Alternatively, we can also determine the response using the Duhamel integral, as

Z∞ −∞
Z Z∞
w(t) = f (τ )h(t − τ ) dτ = − f (t − ρ)h(ρ) dρ = f (t − ρ)h(ρ) dρ (7.102)
−∞ ∞ −∞

where we substituted τ = t − ρ or ρ = t − τ and used the fact that dρ

dτ
= −1. For an excitation
that is acting since τ = −∞, it holds

Z∞
w(t) = f (t − ρ)h(ρ) dρ
−∞
Z∞  
= fb+ e iΩ(t−ρ) + fb− e −iΩ(t−ρ) h(ρ) dρ
−∞
Z∞ Z∞
= +e
fb iΩt
h(ρ)e −iΩρ
dρ + fb −e
−iΩt
h(ρ)e iΩρ dρ
−∞ −∞

Structural Dynamics 7 Forced vibration of linear systems with single degree of freedom
summer term 2019 124
We substitute τ = −ρ in the second part of the preceeding equation and obtain

Z∞ Z∞
= +e
fb iΩt
h(ρ)e −iΩρ
dρ + fb
−e
−iΩt
h(−τ )e −iΩτ dτ
−∞ −∞

= F {h(t)} fb+ e iΩt + F {h(−t)} fb− e −iΩt

= h̃(Ω)fb+ e iΩt + h̃(−Ω)fb− e −iΩt (7.103)

h̃(Ω) is the Fourier transform of the unit impulse response function h(t). Comparing Eq. (7.103)
with Eq. (7.100), we note that

Z∞
U (Ω) = h̃(Ω) = h(t)e−iΩt dt = F {h(t)}
−∞

Thus, the transfer function for harmonic excitation is equal to the Fourier transform of the
unit impulse response function.

7.5.3.2 Response to aperiodic excitation

We consider the general, aperiodic load f (t) and its Fourier transform
∞
1 Z ˜
f (t) = f (ω)e iωt dω
2π
−∞
Z∞
f˜(ω) = f (t)e −iωt dt .
−∞

The response of an SDOF system can then be written as

∞
1 Z
w(t) = w̃(ω)e iωt dω . (7.104)
2π
−∞

This representation gives the response in the time domain as the inverse Fourier transform of
the response in the frequency domain.
Independently, we can also state the response as the Duhamel integral

Z∞
w(t) = f (t − τ )h(τ ) dτ (7.105)
−∞

Structural Dynamics 7 Forced vibration of linear systems with single degree of freedom
summer term 2019 125
We know replace f (t − τ ) in Eq. (7.105) by its Fourier transform

Z∞ ∞
 
1 Z ˜
w(t) = f (ω)e iω(t−τ ) dω  h(τ ) dτ
2π

−∞ −∞
Z∞
 ∞ 
1 ˜
Z
= f (ω) h(τ )e −iωτ dτ  e iωt dω
2π

−∞ −∞
| {z }
h̃(ω)
∞
1 Z ˜
= f (ω)h̃(ω)e iωt dω . (7.106)
2π
−∞

By comparing Eq. (7.104) with Eq. (7.106), we find that

w̃(ω) = h̃(ω)f˜(ω) .

with

w̃(ω) : Fourier transform of the response

f˜(ω) : Fourier transform of the excitation
h̃(ω) : Frequency domain transfer function = U (ω)

Thus, the system response in the frequency domain w̃(ω) can simply be calculated by mul-
tiplying the transfer function h̃(ω) with the Fourier transform of the excitation f˜(ω). The
system response in the time domain w(t) is given by the inverse Fourier transform of w̃(ω).

w(t) = F −1 {w̃(ω)}

The relations are further depicted in Fig. 7.26.

Structural Dynamics 7 Forced vibration of linear systems with single degree of freedom
summer term 2019 126
 f (t)
m
w(t)

k c

convolution
R∞
w(t) = f (τ )h(t − τ ) dτ
−∞
f (t) w(t)
h(t)

∗ =

f˜(ω) w̃(ω)
h̃(ω)

· =

multiplication
w̃(ω) = h̃(ω) · f˜(ω)

Figure 7.26: Response of SDOF system to arbitrary excitation. The force signal f (t) is given. To
calculate the response signal w(t), the problem can be considered in the time or fre-
quency domain. In the time domain, the excitation signal is convoluted with the impulse
response function h(t), and the resulting integral has to be solved. For a frequency
domain calculation, the force signal f (t) is transformed into the frequency domain by a
Fourier transformation, resulting in f˜(ω). The response in the frequency domain w̃(ω)
is then found by multiplication of the force f˜(ω) with the systems’ transfer function h̃(ω).
The resulting response in the time domain is then given by an inverse Fourier transform
of w̃(ω). The two approaches yield the same result for the steady-state response.

Structural Dynamics 7 Forced vibration of linear systems with single degree of freedom
summer term 2019 127
8 Free vibration of linear systems with
multiple degrees of freedoms
This section deals with the free vibration of systems with multiple degrees of freedom. We
assume that the equation of motion is found by any of the methods discussed in chapter 4.

8.1 Preliminary remark: Maxwell-Betti theorem

For a discrete, linear elastic system it holds

w = Af

with the flexibility matrix A, the displacement vector w, and the force vector f . We now
consider a system, where two forces are applied one after the other, first at position k and
secondly at position i. This case is called I. The reverse is called case II. The displacements
in both cases are

wi = aii fi + aik fk

wk = aki fi + akk fk

where aij is the (i,j) element of the flexibility matrix A. After applying the loads, the dis-
placements are equal in both cases In a linear system, the work performed on the system by
the forces has to be equal independently from the sequence of the loads. For case I the work
is

1 1
WI = akk fk2 + aii fi2 + aki fi Fk . (8.1)
2 2 | {z }
wki

The last term corresponds to the work performed by load fk on the displacement wki = aki fi .
For case II the work is

1 1
WII = aii fi2 + akk fk2 + aik fk fi , (8.2)
2 2 | {z }
wik

Structural Dynamics
summer term 2019
128
where now the last term corresponds to the work performed by load fi on the displacement
wik = aik fk . As the work performed in both cases is the same, we equate Eqs. (8.1) and (8.2)
and obtain

wki fk = wik fi . (8.3)

It follows that the work performed by force fk on the displacement wki (displacement at
position k due to fi ) is the same as the work performed by force fi on the displacement wik
(displacement at position i due to fk ). Furthermore, from

aki fi · fk = aik fk · fi (8.4)

we obtain

aki = aik .

Here, aki corresponds to the displacement at position k due to a unit load 1 at position i and
aik corresponds to the displacement at position i due to a unit load 1 at position k. Thus, we
can state for the flexibility matrix

AT = A.

From this it follows that the transpose of the flexibility matrix is equal to the flexibility matrix.
Thus, the matrix is symmetric. The stiffness matrix K is the inverse of the flexibility matrix
and by this also symmetric, i.e.

KT = K .

8.2 Solution of the homogeneous system of equations

The general equation of motion for unforced MDOF systems with velocity proportional damp-
ing reads

Mẅ + Cẇ + Kw = 0. (8.5)

Even though the system is not excited by an external force, it can have non-trivial solutions
w(t) 6= 0 responding to the initial conditions that are not in equilibrium. We first discuss the
case of undamped MDOF systems and then generalize for the case of damped systems. In the
following n denotes the number of DOFs.

Structural Dynamics 8 Free vibration of linear systems with multiple degrees of freedoms
summer term 2019 129
8.2.1 Undamped MDOF systems

The derivation of the eigensolutions (f = 0, natural vibrations) of a MDOF is first given for
systems without damping (C = 0). The system of equations thus reads

Mẅ + Kw = 0 (8.6)

We use an exponential approach to solve Eq. (8.6)

w = φe iωt (8.7)
ẅ = −ω 2 φe iωt , (8.8)

and insert Eqs. (8.7) and (8.8) into Eq. (8.6) to obtain a homgoeneous linear system of equa-
tions representing a matrix eigenvalue problem.
 
K − ω2M φ = 0 (8.9)

with λ = ω 2 . Nontrivial solutions for the displacements w are only possible if the determinant
of the matrix (K − λM) is zero. The only unknown in Eq. (8.9) is the value λ = ω 2 (repre-
senting the square of the circular natural frequency in Eq. (8.7)). Thus, we find the circular
natural frequencies of the system by

det (K − λM) = 0. (8.10)

Eq. (8.10) leads to an equation for λ = ω 2 with the order n (n = number of DOFs) and therefore
has n solutions for the eigenvalues ω12 , . . . , ωn2 . The solutions of the system of equations for a
vanishing determinant are called mode shapes or eigenvectors. The eigenvector φi related to
λi = ωi2 is obtained by inserting ωi2 in eq. (8.9).
 
K − ωi2 M φi = 0 (8.11)

For a vanishing determinant the above equation system becomes linearly dependent. Therefore,
the eigenvectors φi (also called modes, mode shapes) are only determined up to a constant
factor (hence also α φi is an eigenvector of the system). In general, various options can be
chosen for the normalization of the eigenvectors:
• The eigenvector is scaled, such that one vector component, e.g. the largest, is chosen
arbitrarily. Usually this value is set equal to one.
• The eigenvector is scaled, such that the L2 -norm is one, i.e., the eigenvector has the
length one.

Structural Dynamics 8 Free vibration of linear systems with multiple degrees of freedoms
summer term 2019 130
 • The eigenvector is scaled, such that the generalized mass matrix is the identity matrix
(explained in the following), i.e., the norm of the eigenvectors with respect to the mass
matrix is equal to one.
A solution is finally given by the superposition of all the eigenvectors of the system weighted
by time-dependent weighting functions. The solution w(t) to Eq. (8.6) can be written as
n   n
w(t) = φi ŷi+ e iωi t + ŷi− e −iωi t = φi yi (t) = Φy(t) . (8.12)
X X

i=1 i=1

The solution can also be written in sine/cosine notation, as discussed in section 2.2. Then,
n
w(t) = φi (yi,01 cos (ωi t) − yi,02 sin (ωi t)) = Φy(t) . (8.13)
X

i=1

In order, to discuss the important characteristics of the eigenvalue solution, we write Eq. (8.9)
for any two eigenvectors φi and φj with corresponding eigenvalues λi and λj

Kφi = λi Mφi (8.14)

Kφj = λj Mφj . (8.15)

for all i,j = 1,2, . . . , n. We now premultiply on both sides of Eq. (8.14) by φTj and obtain

φTj Kφi = λi φTj Mφi . (8.16)

Considering the symmetry of the mass and (due to the law of Maxwell-Betti, see 8.1) the
 T
stiffness matrix MT = M and KT = K and using AT BC = CT BT A, we find the transpose
of Eq. (8.16) as

φTi Kφj = λi φTi Mφj . (8.17)

Analogously Eq. (8.15) is premultiplied on both sides by φTi such that

φTi Kφj = λj φTi Mφj . (8.18)

If we substract Eq. (8.18) from Eq. (8.17), we obtain:

0 = (λi − λj ) φTi Mφj .

Structural Dynamics 8 Free vibration of linear systems with multiple degrees of freedoms
summer term 2019 131
From this we find the orthogonality relation

0 for i 6= j
φTi Mφj = . (8.19)
Mi for i = j

Eq. (8.19) shows that the eigenvectors are orthogonal with respect to the mass matrix. We call
M∗ = ΦT MΦ the generalized mass matrix and Mi = φTi Mφi the generalized or modal mass
of mode i. Here, Φ is the modal matrix that contains the eigenvectors column-wise. Since
the eigenvectors are only determined up to a constant factor, for i = j the value Mi can take
arbitrary values. If we now insert Eq. (8.19) into Eq. (8.17), we obtain

0 for i 6= j
φTi Kφj = . (8.20)
K i = Mi λi = Mi ωi2 for i = j

We furthermore call K∗ = φTi Kφj the generalized stiffness matrix and Ki = φTi Kφi the
generalized or modal stiffness of mode i. The eigenfrequencies ωi can be calculated from the
generalized stiffnesses and masses as

Ki
ωi2 = . (8.21)
Mi

As was discussed before, the eigenvectors need to be normalized. We can now choose the
normalization of φi such that the generalized mass matrix is the identity matrix, i.e.,

φTi Mφj = δij (8.22)

holds. This choice is also known as “unit modal mass”. It is Ai = 1 and thus we can write for
Eq. (8.20)

φTi Kφj = λi δij = ωi2 δij . (8.23)

We obtain a compact notation of these n matrix equations by introducing the diagonal matrix

   
λ1 ω12
[Λ] =  .. = .. (8.24)
. .
   
 ,
   
λn ωn2

containing all the eigenvalues and the matrix

Φ = [φ1 , φ2 , . . . , φn ] , (8.25)

Structural Dynamics 8 Free vibration of linear systems with multiple degrees of freedoms
summer term 2019 132
containing all the eigenvectors. With this, we can write Eq. (8.22) and Eq. (8.23) as

ΦT MΦ = I (8.26)
ΦT KΦ = Λ . (8.27)

Note that this only holds for mass normalized eigenvectors. Applying the derived properties
of the solution, we can find a simple expression for the equation of motion (8.6) in terms of
the modal coordinates y(t). The solution given in 8.12 is repeated her for convenience in the
general form:

w(t) = Φy(t) . (8.28)

The time functions y(t) are time harmonic functions, that can be either expressed as complex
exponential functions or sine/cosine functions. We now enter Eq. (8.28) into the equation of
motion Eq. (8.6). It follows

M Φÿ(t) + KΦy(t) = 0 . (8.29)

By multiplying Eq. (8.29) with the transpose of the matrix of the eigenvectors ΦT , we obtain

ΦT MΦÿ(t) + ΦT KΦy(t) = 0 . (8.30)

If the eigenvectors are normalized with respect to the mass matrix, i.e. ΦT MΦ = I, after
inserting Eqs. (8.22) and (8.23) it follows

Iÿ(t) + Λy(t) = 0 . (8.31)

As I and Λ are diagonal matrices the system of equations become decoupled and we obtain n
equations

ÿi (t) + ωi2 yi (t) = 0 (8.32)

for i = 1, . . . ,n. The decoupled equations of motion correspond to the equation of motion of
the SDOF system.

Structural Dynamics 8 Free vibration of linear systems with multiple degrees of freedoms
summer term 2019 133
 Example 8.1: Two degree of freedom system

In the general case, for n = 2, Eq. (8.30) reads

φT1 Mφ1 φT1 Mφ2 φT1 Kφ1 φT1 Kφ2

" # " #
ÿ(t) + T y(t) = 0 .
φT2 Mφ1 φT2 Mφ2 φ2 Kφ1 φT2 Kφ2

Through the orthogonality relations for M and K this leads to a decoupled system of
equations

m∗1 0 k∗ 0
" #" # " #" #
ÿ1 y1
+ 1 ∗ =0
0 m∗2 ÿ2 0 k2 y2

with mi ∗ = φTi Mφi and ki ∗ = φTi Kφi . The above holds for any normalization of the
eigenvectors. The response yi (t) can now be determined for every mode shape φi as
the solution of the corresponding SDOF system:

m∗1 ÿ1 (t) + k1∗ y1 (t) = 0

m∗2 ÿ2 (t) + k2∗ y2 (t) = 0.

8.2.2 Consideration of the initial conditions

Furthermore the initial conditions of the MDOF system shall be given as

w(t = 0) = w0 (8.33)
ẇ(t = 0) = v(t = 0) = v0 (8.34)

Eqs. (8.33) and (8.34) are now represented by the initial conditions of the modal responses

w(t = 0) = Φy(t = 0) (8.35)

ẇ(t = 0) = v(t = 0) = Φẏ(t = 0) (8.36)

The initial conditions w(t = 0) and ẇ(t = 0) are projected on the modal basis. We multiply
Eqs, (8.35) and (8.36) by ΦT M and obtain

ΦT MΦy(0) = ΦT Mw(0) (8.37)

Structural Dynamics 8 Free vibration of linear systems with multiple degrees of freedoms
summer term 2019 134
 ΦT MΦẏ(0) = ΦT Mẇ(0) . (8.38)

Considering the orthogonality relation (Eq. (8.22)) under the assumption of unit modal mass
normalization it follows

Iy(0) = ΦT Mw(0) (8.39)

Iẏ(0) = ΦT Mẇ(0) . (8.40)

Finally the values for y(0) and ẏ(0) are

y(0) = Tw(0) (8.41)

ẏ(0) = Tẇ(0) (8.42)

with T = ΦT M.

Example 8.2: Two degree of freedom system (cont.)

The following MDOF system with two degrees of freedom w1 und w2 is considered. The
homogeneous solution shall be determined.

m1

w1 m1 = 80 kg
m2 = 8 kg
k1 N
k1 = 200 m
N
k2 = 125 m
m2
w2
k2

For this purpose the eigenfrequencies and mode shapes need to be calculated in a
first step. In a second step, the solution can be obtained for the decoupled system of
equations.

a) First, we calculate the eigenfrequencies ω1 and ω2 as well as the corresponding

Structural Dynamics 8 Free vibration of linear systems with multiple degrees of freedoms
summer term 2019 135
 mode shapes φ1 and φ2 . From Eq. (8.6) and Eq. (8.8) it follows

MΦ(−ω 2 )y(t) + KΦy(t) = 0

and thus
 
−ω 2 M + K = 0.

A solution for the above system of equations exists if

 
det −ω 2 M + K = 0 .

With

m1 0
" #
M=
0 m2
" #
k1 −k1
K=
−k1 k1 + k2

it yields
" #!
k1 − m1 ω 2 −k1
det = 640 ω 4 − 27.600 ω 2 + 25.000 = 0 .
−k1 k1 + k2 − m2 ω 2

Two solutions are obtained for ω 2

ω12 = 0.93
ω22 = 42.20 .

The eigenfrequencies are

ω1 = 0.96
ω2 = 6.50 .

The corresponding eigenvectors can be determined from

0
" # ! !
k1 − m1 ωi 2 −k1 φ1i
= (8.43)
−k1 k1 + k2 − m2 ωi2 φ2i 0

for i = 1,2. As the system of equations is underdetermined we solve the first

Structural Dynamics 8 Free vibration of linear systems with multiple degrees of freedoms
summer term 2019 136
 equation for φ2i
 
k1 − m1 ωi2 Φ1i − k1 Φ2i = 0
(k1 − m1 ωi 2 )Φ1i
⇒ Φ2i =
k1

and choose Φ1i = 1, thus

k1 − m1 ω1 2
Φ21 = = 0.63
k1
k1 − m1 ω2 2
Φ22 = = −15.88 .
k1

The matrix of eigenvectors is then given as

1 1
" # " #
φ φ
Φ = [φ1 φ2 ] = 11 12 = .
φ21 φ22 0.63 −15.88

Note that the eigenvectors are not normalized with respect to the mass matrix,
i.e., ΦT MΦ 6= [I]. To obtain the mass normalized eigenvectors, we calculate the
modal mass of each eigenvector

m∗i = φTi Mφi .

For the first eigenvector, we obtain

80 0 1
" #" #
h i
m∗1 = φT1 Mφ1 = 1 0.63 = 83.17
0 8 0.63
80 0 1
" #" #
h i
m∗2 = φT2 Mφ2 = 1 −15.88 = 2 097.33 .
0 8 −15.88

Thus, we obtain the mass normalized eigenvectors by dividing each eigenvector

by the square-root of its modal mass

φ φ
m
φi = √ i ∗ = q i .
mi T
φi Mφi

The mass normalized eigenvectors are indicated by a left superscript m in this

Structural Dynamics 8 Free vibration of linear systems with multiple degrees of freedoms
summer term 2019 137
 example. Thus,

1 1 1 0.110
" # " #
m
φ1 = √ ∗ φ1 = √ =
m1 83.17 0.63 0.069
1 1 1 0.022
" # " #
m
φ2 = √ ∗ φ2 = √ = .
m2 2 097.33 −15.88 −0.347

b) In the following we choose to use the mass normalized eigenvectors m φi . With

the eigenfrequencies and eigenvectors determined, we can now solve the decoupled
equation system

1 0 ω2 0
" # " #
ÿ(t) + 1 y(t) = 0
0 1 0 ω22

for each eigenform separately:

ÿi (t) + ωi2 yi (t) = 0 . (8.44)

The solution to the homogeneous differential equation for each yi (t) is given by

yi (t) = yi1 cos (ωi t) − yi2 sin (ωi t) .

For the given 2DOF system with eigenfrequencies ω1 and ω2 the solution reads

y1 (t) = y11 cos (ω1 t) − y12 sin (ω1 t)

y2 (t) = y21 cos (ω2 t) − y22 sin (ω2 t) .

The four unknowns y11 , y12 , y21 , y22 can be determined from the initial conditions.
Here, we choose

w1 (0) = −1 w2 (0) = 1 ẇ1 (0) = 0 ẇ2 (0) = 0.

For that purpose the initial conditions need to be transformed to modal coordi-
nates according to Eqs. (8.41) and (8.42) using the transformation matrix

0.110 0.069 80 0 8,772 0,552

" #" # " #
T= Φ M=m T
=
0.022 −0.347 0 8 1,747 −2,774

Hence the initial conditions y(0) und ẏ(0) relating to the eigenvectors are found

Structural Dynamics 8 Free vibration of linear systems with multiple degrees of freedoms
summer term 2019 138
 by using Eqs. (8.41) and (8.42).
" #
−8,220
y(0) = Tw(0) =
−4,521
0
" #
ẏ(0) = Tẇ(0) = .
0

Note, that this transformation only holds, when the mass normalized eigenvectors
are used. With the transformed initial conditions in the modal space, Eq. (8.44)
can be solved for each mode shape separately for y(t). For the first mode shape, we
determine the unknowns y11 und y12 by inserting the transformed initial conditions
y(0) und ẏ(0) into the solution of the homogeneous differential equation Eq. (5.9).

y11 = −8.220
y12 = 0 .

Analogously the coefficients C21 und C22 for the second mode shape are deter-
mined.

y21 = −4.521
y22 = 0 .

Thus the modal solutions read

y1 (t) = y11 cos (ω1 t) + y12 sin (ω1 t) = −8.220 cos (0.962 t)
y2 (t) = y21 cos (ω2 t) + y22 sin (ω2 t) = −4.521 cos (6.496 t) .

The solution w(t) is obtained by

w (t) 0.110 0.022 −8.220 cos (0.962 t)

" # " #" #
w(t) = 1 = m Φy(t) =
w2 (t) 0.069 −0.347 −4.521 cos (6.496 t)
−0.901 cos (0.962 t) − 0.099 cos (6.496 t)
" #
= .
−0.568 cos (0.962 t) + 1.568 cos (6.496 t)

We can easily identify the contributions of each mode shape to the displacement
solution of the individual degree of freedom. The first summand in the first
and second entry is the contribution of the first mode shape to the first and
second degree of freedom, respectively. The second summand corresponds to the
contributions of the second mode shape.

Structural Dynamics 8 Free vibration of linear systems with multiple degrees of freedoms
summer term 2019 139
 c) Alternative solution
The oscillation of the degrees of freedom w1 (t) und w2 (t) can also be directly
given by Eq. (8.13). This procedure does not require the eigenvectors to be mass
normalized. We still denote the unknowns as yij .

w (t) Φ Φ12 y1 (t)

" # " #" #
w(t) = 1 = Φy(t) = 11
w2 (t) Φ21 Φ22 y2 (t)
Φ (y cos (ω1 t) − y12 sin (ω1 t)) + Φ12 (y21 cos (ω2 t) − y22 sin (ω2 t))
" #
= 11 11 .
Φ21 (y11 cos (ω1 t) − y12 sin (ω1 t)) + Φ22 (y21 cos (ω2 t) − y22 sin (ω2 t))

The unknowns y11 , y12 , y21 , y22 are found using the initial conditions w(t = 0)
and ẇ(t = 0). The resulting equation systems read

w1 (0) Φ11 −Φ12

" # " #" #
y11
=
w2 (0) −Φ21 Φ22 y21
ẇ1 (0) Φ ω Φ12 ω2
" # " #" #
y12
= − 11 1 .
ẇ2 (0) Φ21 ω1 Φ22 ω2 y22

Inserting the initial conditions

w1 (0) = −1 w2 (0) = 1 ẇ1 (0) = 0 ẇ2 (0) = 0,

we can solve for the unknowns using eiter of the normalized eigenvectors. We
again choose the mass normalized eigenvectors and obtain
#−1 "
0.110 −0.022
" # " #
y11 −1
=
y21 −0.069 −0.347 1
#−1 " #
0.110 · 0.69 0.022 · 42.2 Φ11 ωE1 Φ12 ωE2 0 0
" # " #" " #
y12
=− = .
y22 0.069 · 0.69 −0.347 · 42.2 Φ21 ωE1 Φ22 ωE2 0 0

Since we used the mass normalized in both approaches we obtain the same results
for the coefficients yij . Nevertheless, the second approach would also permit the
use of otherwise scaled eigenvectors. Then the coefficients yij would differ, but
the same result for w(t) would be achieved.
d) The resulting amplitude is now depicted with respect to time for both degrees of
freedom separately between t = 0 s (initial conditions) and t = 15 s.

Structural Dynamics 8 Free vibration of linear systems with multiple degrees of freedoms
summer term 2019 140
 Figure 8.1: Solution for the 2DOF system. The left figure and right figure show the
displacement solution for degree of freedom 1 and 2, respectively (—).
Each figure gives the separate contributions from mode shape 1 (–·–) and
mode shape 2 (- - -). It can be observed that the solution for DOF 1 is
mainly governed by the contributions from mode shape 1, whereas the
second DOF is predominantly governed by the second mode shape. The
2π
periods are T1 = ω 1
= 2π = 6.53 s and T2 = ω
0.96 Hz
2π
2
= 2π = 0.97 s.
6.5 Hz

8.2.3 Damped MDOF systems

Until now, we discussed the free vibration for the undamped MDOF system. In the following
we investigate the influence of a damping matrix C 6= 0 on the solution strategy. Consider
the equation of motion for the damped MDOF system without external loads:

Mẅ + Cẇ + Kw = 0. (8.45)

A major concern is the behavior of the damping matrix under the transformation to modal
coordinates using the eigenvectors of the undamped problem φ, i.e., the form of the generalized
damping matrix

C∗ = ΦT CΦ. (8.46)

For an arbitrary damping matrix C, in general, the generalized damping matrix C∗ is not
diagonal, and thus the equations of motion do not decouple.
Three strategies are usually introduced to cope with this problem. These are:
• Solve the problem with the general damping matrix. This most general approach
leads to a quadratic eigenvalue problem, which can be further augmented to a standard
eigenvalue problem. The resulting eigenvectors and eigenvalues are complex. This ap-
proach has to be chosen, especially, if the spatial distribution of the damping differs

Structural Dynamics 8 Free vibration of linear systems with multiple degrees of freedoms
summer term 2019 141
 significantly from the distribution of mass or stiffness (this holds especially for discrete
damping elements like tuned mass dampers for example).
• Assume a damping matrix that is diagonalizable. Whenever the generalized
damping matrix C∗ = ΦT CΦ becomes diagonal, the equation system becomes decoupled
and the single equations correspond to the equations of motion of the damped SDOF
system. The simplest form of this kind of damping is called Rayleigh damping.
• Project the damping matrix on each mode. With this approach, the damping
matrix C is projected onto the modal space and only the diagonal terms are retained,
i.e., only the terms

ci = φTi Cφi (8.47)

are used for the generalized damping matrix. All cross terms φTj Cφi for j 6= i are
neglected.
• Assume modal damping. With this approach, the generalized (modal) damping
matrix C∗ is given directly. Each mode is assigned a damping value, thereby C∗ is
directly diagonal. Often it is not possible to construct a damping matrix from a physical
model. In this case it might be beneficial to assume damping values for the single modes
directly.

8.2.3.1 The quadratic eigenvalue problem

In the case of a damped MODF system, the system of equations reads

Mẅ + Cẇ + Kw = 0. (8.48)

In the general case, the following ansatz holds

w = φe λt , (8.49)

and thus the first two derivatives with respect to time are

ẇ = λφe λt (8.50)
ẅ = λ2 φe λt . (8.51)

We know insert Eq. (8.49) to Eq. (8.51) into Eq. (8.45) and obtain
 
Mλ2 + Cλ + K φ = 0, (8.52)

Structural Dynamics 8 Free vibration of linear systems with multiple degrees of freedoms
summer term 2019 142
which is commonly referred to as a quadratic eigenvalue problem. A nontrivial solution for φ
is found for the case that the matrix (Mλ2 + Cλ + K) becomes singular and the determinant
is zero
 
det Mλ2 + Cλ + K = 0. (8.53)

Eq. (8.53) is the characteristic equation of the problem. It is a polynomial with order 2n and
gives 2n solutions for λ which are either all real or occur as complex conjugate pairs. Thus,
we obtain 2n eigenvalues λ and corresponding eigenvectors φ. As for the undamped case, the
eigenvectors are only determined up to an arbitrary factor.

8.2.3.2 The standard form of the damped eigenvalue problem

This derivation follows Humar [2012]. We introduce the a vector z of displacements and
velocities (state space formulation)
!
w
z= (8.54)
ẇ

and write the equations of motion (8.45) as

Mẅ + Cẇ + Kw = 0 (8.55)

Mẇ − Mẇ = 0. (8.56)

In matrix notation, this reads

" #" # " #" # " #
C M ẇ K 0 w 0
+ = (8.57)
M 0 ẅ 0 −M ẇ 0
| {z } | {z }
:=A :=B

Hence, Eq. (8.57) is written as

Aż(t) + Bz(t) = 0. (8.58)

A solution approach is given by z = φe λt . Inserted into Eq. (8.58) it reads

(λA + B) φe λt = 0. (8.59)

Structural Dynamics 8 Free vibration of linear systems with multiple degrees of freedoms
summer term 2019 143
Eq. (8.58) is a linear eigenvalue problem. This can further be transformed to a standard
eigenvalue problem, by multiplying with the inverse of A, such that
 
λI + A−1 B φe λt = 0. (8.60)

Thereby Eq. (8.60) can be written as

(λI − D) φe λt = 0. (8.61)

with
" #
0 I
D= . (8.62)
−M K −M−1 C
−1

We finally obtain the homogeneous equation system

" #!
0 I
λI − φ = 0. (8.63)
−M−1 K −M−1 C

This formulation is now a common eigenvalue problem. The resulting eigenvalues and eigen-
vectors are in general complex. Here, we get complex values for both the eigenvalues and the
coordinates of the eigenvectors.
Whereas vibrations in real mode shapes show for specific points in time for all nodes of the
system zero crossings or maximum values at the same time, this is not the case for complex
mode shapes.
Since many available finite element codes cannot solve complex eigenvalue problems, in general,
a direct integration using a time step method hast to be applied in this case (comp. Chapter
16).

Structural Dynamics 8 Free vibration of linear systems with multiple degrees of freedoms
summer term 2019 144
 Example 8.3

The given 2DOF system is extended by one additional, viscous damper c2 , as depicted
in the following figure.

m1

w1 m1 = 80 kg
m2 = 8 kg
k1 c1 N
k1 = 200 m
N
k2 = 125 m
m2
w2 c1 = 0
c2 = 0.6 Ns
m
k2 c2

The differential equation of the MDOF system with n degrees of freedom reads

Mẅ(t) + Cẇ(t) + Kw(t) = 0. (8.64)

For the given system the differential equation for the two degrees of freedom w1 and w2
are

m1 0
" # " # " #
c1 −c1 k1 −k1
ẅ(t) + ẇ(t) + w(t) = 0.
0 m2 −c1 c1 + c2 −k1 k1 + k2

Again, the following initial conditions are given.

w1 (0) = −1 w2 (0) = 1 ẇ1 (0) = 0 ẇ2 (0) = 0

We use the approach in Eq. (8.63), and introduce the vector of unknown state variables
z, that includes displacements and velocities.

w1
 
!
w w 
z= =  2
 
ẇ ẇ1 
ẇ2

Structural Dynamics 8 Free vibration of linear systems with multiple degrees of freedoms
summer term 2019 145
 Hence (8.64) can be rewritten as

ż(t) = Dz(t),

or equivalently

ẇ1 # w1
   
"
ẇ  0 I w 
 =
 2  2
−1 −1  .
ẅ1  −M K −M C ẇ1 
ẅ2 ẇ2

With the approach z = φe λt the following homogeneous equation system is obtained

" " ##
0 I
λI − φ = 0.
−M K −M−1 C
−1

The problem can be solved using MATLAB® via the polyeig command. The solution
to the above equation system is given by the following four eigenvalues.

λ1 = −3.61 · 10−2 + 6,50i

λ2 = −3.61 · 10−2 − 6,50i
λ3 = −1.43 · 10−3 + 0,96i
λ4 = −1.43 · 10−3 − 0,96i .

The eigenvalues are complex conjugate pairs. The corresponding eigenvectors can be
calculated using Eq. (8.63). They are also complex conjugate pairs. The resulting
eigenvectors from the polyeig command in MATLAB® are normalized such that they
have unit-length. Here, we choose to give the eigenvectors such that the first entry is
equal to one for better readability:

1,00 1,00
   
 −15,88 + 0,12i   −15,88 − 0,12i 
φ1 =  φ2 = 
   
 −0,04 + 6,50i   −0,04 − 6,50i 
 

−0,11 − 103,15i −0,11 + 103,15i

1,00 1,00
   
 0,63 + 0,69 · 10−3 i   0,63 − 0,69 · 10−3 i 
φ3 =  φ4 =  .
   
−0,50 · 10−3 + 0,96i −0,50 · 10−3 − 0,96i


−0,97 · 10−3 + 0,61i −0,97 · 10−3 − 0,61i

Structural Dynamics 8 Free vibration of linear systems with multiple degrees of freedoms
summer term 2019 146
 The solution for the augmented state vector z is then written as the sum z = bj φj e .
P2n λj t
j=1 z
The unknowns zbj can be determined from the initial conditions:

w1 (0) zb1 Φ11 e λ1 0 + zb2 Φ12 e λ2 0 + zb3 Φ13 e λ3 0 + zb4 Φ14 e λ4 0 −1

     
w (0) zb Φ e λ1 0 + zb Φ e λ2 0 + zb Φ e λ3 0 + zb Φ e λ4 0   1 
z(0) =  2
=
  1 21 2 22 3 23 4 24
=
 
 .

ẇ1 (0) zb1 Φ31 e λ1 0 + zb2 Φ32 e λ2 0 + zb3 Φ33 e λ3 0 + zb4 Φ34 e λ4 0   0 


ẇ2 (0) zb1 Φ41 e λ1 0 + zb2 Φ42 e λ2 0 + zb3 Φ43 e λ3 0 + zb4 Φ44 e λ4 0 0

The resulting vibration is finally depicted with respect to time for both degrees of
freedom separately between t = 0 s (initial conditions) and t = 100 s. Note that due to
the calculation approach the mode shape with higher eigenvalue is now the first mode
shape as compared to the undamped case in Example 8.2.

Figure 8.2: Solution for the damped 2DOF system. The left figure and right figure show
the displacement solution for degree of freedom 1 and 2, respectively (—). Each
figure gives the separate contributions from mode shape 1 (–·–) and mode shape
2 (- - -). It can be observed that the solution for DOF 1 is mainly governed by the
contributions from mode shape 2, whereas the second DOF is predominantly
governed by the second mode shape for small values of t. Nevertheless, for
larger t the response of mode shape 1 decays and the response is governed
by mode shape 1. Due to the applied damping the response of mode shape 1
2π
decays much more rapidly than for mode shape 2. The periods are T1 = ω 2
=
2π 2π 2π
6.5 rad Hz
= 0.97 s and T2 = ω1 = 0.96 rad Hz
= 6.53 s.

8.2.3.3 Rayleigh damping

The equation of motion of an unloaded system with multiple degrees of freedom and viscous
damping reads

Mẅ + Cẇ + Kw = 0. (8.65)

Structural Dynamics 8 Free vibration of linear systems with multiple degrees of freedoms
summer term 2019 147
We insert as ansatz the jth mode shape of the undamped MDOF system (modal ansatz),

wj (t) = φj yj (t) (8.66)

into Eq. (8.65) and obtain

Mφj ÿj (t) + Cφj ẏj (t) + Kφj yj (t) = 0. (8.67)

Furthermore, we multiply with the transpose of the i-th eigenvector φi of the undamped
system, from the left,

φTi Mφj ÿj (t) + φTi Cφj ẏj (t) + φTi Kφj yj (t) = 0. (8.68)
| {z } | {z }
δij ωj δij

For an arbitrary damping matrix the above equations are not decoupled anymore, since in
general it holds

φTi Cφj 6= Ci δij . (8.69)

Only in the special case of

C = αM + βK (8.70)

the modal ansatz of the undamped system leads to decoupled equations in the damped case
and thus a diagonalizable modal damping matrix:

Mj ÿj + Cj ẏj + Kj yj = 0 (8.71)

with

Cj = αφTj Mφj + βφTj Kφj = αMj∗ + βωj2 Mj∗ . (8.72)

In this case the damping is mass and/or stiffness proportional. It is called Rayleigh damping.
The relationship between the Rayleigh damping coefficients α and β and the modal damping
ratio Dj of the j-th eigenmode is given by (compare Tab. 5.1):

1
!
Cj
Cj α
Dj = q = = + βωj . (8.73)
2 Kj Mj 2Mj ωj 2 ωj

The stiffness proportional part increases for increasing eigenfrequencies ωj , whereas the mass
proportional part decreases for increasing eigenfrequencies ωj , as it is proportional to its inverse

Structural Dynamics 8 Free vibration of linear systems with multiple degrees of freedoms
summer term 2019 148
1
ωj
. In the Rayleigh-damping model the free parameters α and β have to be chosen. A
possibility is to choose desired damping values for two of the mode shapes φi and φj with
eigenfrequencies ωi and ωj , respectively, say Di and Dj , and adjust the values for α and β
such that

1 α
 
Di = + βωi (8.74)
2 ωi
1
!
α
Dj = + βωj . (8.75)
2 ωj

This leads to the following equation solution for the unknowns α and β.
" # "1 #−1 " #
α ωi Di
=2 ωi
1 . (8.76)
β ωj
ωj Dj

Dj Dj
α
2 ωj
β
ω
2 j

ωj

Figure 8.3: Rayleigh damping

The decoupled equations of motion for systems with Rayleigh-damping can be solved with the
methods already introduced for SDOF systems.

Structural Dynamics 8 Free vibration of linear systems with multiple degrees of freedoms
summer term 2019 149
8.2.3.4 Caughey damping model using additional orthogonality relations

According to Humar [2012], the following orthogonality conditions


 b 0 i 6= j
φTi M M−1 K φj = (8.77)
λb
i i=j

hold for all b ∈ Z. Thus, for any damping matrix C that is represented by the following linear
combination
 b
C= αb M M−1 K (8.78)
X

the equations of motion become decoupled under the transformation into modal coordinates.
This damping model is called Caughey damping model. The Rayleigh damping model is
included as a special case in Eq. (8.78) for b = 0,1, i.e
 0
C = α0 M M−1 K + α1 MM−1 K = α0 M + α1 K (8.79)

and α0 = α and α1 = β. With this model, damping ratios can be specified for more than two
modes, as is the case for the Rayleigh damping model.

Structural Dynamics 8 Free vibration of linear systems with multiple degrees of freedoms
summer term 2019 150
9 Forced vibration of linear systems with
multiple degree of freedoms
Depending on the time structure of the load and the assumed damping model, different meth-
ods are applicable to determine the response of the forced multiple degree of freedom system.
We introduce the direct solution, modal analysis and time step procedures hereinafter.

9.1 Direct solution for harmonic loads

In case of an harmonic excitation, that is an excitation varying with a sine/cosine function in

time, a direct solution for the response can be obtained by inversion of the dynamic stiffness
matrix. This procedure is applicable for cases, where either, the solution is only needed for a
single frequency, or a more general time dependence can be represented by the sum of harmonic
functions (Fourier series, Fourier integral).
Consider the general linear, damped MDOF system with the equation of motion.

Mẅ + Cẇ + Kw = f (9.1)

These are n coupled differential equation given in terms of the n unknown displacements w
and their corresponding derivatives. Under the assumption of time harmonic excitation, we
obtain

f (t) = f 01 cos(Ωt) − f 02 sin(Ωt) = (9.2)

 
= f 0 cos Ωt + ϕf = (9.3)

= bf + e iΩt + bf − e −iΩt . (9.4)

Using the complex exponential approach in Eq. (9.4) simplifies the following derivations. For
that reason, the same approach is made for the unknown displacements

w(t) = w
b + e iΩt + w
b − e −iΩt . (9.5)

Structural Dynamics
summer term 2019
151
The first two derivatives with respect to time are then given by

ẇ = (iΩw
b + ) e iΩt + (−iΩw
b − ) e −iΩt (9.6)
   
ẅ = −Ω2 w
b + e iΩt + −Ω2 w
b − e −iΩt . (9.7)

Inserting Eqs. (9.5) to (9.7) and (9.4) into Eq. (9.1), we obtain

   
b + e iΩt + −Ω2 w
M −Ω2 w b − e −iΩt + C iΩw
b + e iΩt − iΩw
b − e −iΩt +
 
+K w
b + e iΩt + w
b − e −iΩt = f + e iΩt + f − e −iΩt . (9.8)

The parts with positive and negative frequencies are then treated separately, thus
     
b + e iΩt + C iΩw
M −Ω2 w b + e iΩt + K w
b + e iΩt = f + e iΩt (9.9)
     
b − e −iΩt + C −iΩw
M −Ω2 w b − e −iΩt + K w
b − e −iΩt = f − e −iΩt . (9.10)

Reorganizing the different terms leads to

 
−Ω2 M + iΩC + K w
b + e iΩt = f + e iΩt (9.11)
 
−Ω2 M − iΩC + K w
b − e −iΩt = f − e −iΩt . (9.12)

The time functions e iΩt and e −iΩt can be cancelled from both sides in both the equations. As
b + and w
w b − are complex conjugate, the solution only needs to be calculated for one of the
two. The solution for the complex coefficients is given by
 −1
b + = −Ω2 M + iΩC + K
w f+ (9.13)
 −1
b − = −Ω2 M − iΩC + K
w f− = w
b ∗+ . (9.14)

The matrix −Ω2 M + iΩC + K is called the dynamic stiffness matrix Kdyn . Its inverse K−1 dyn =
Rw is called the receptance matrix (flexibility matrix). If the solution is sought for different
frequencies, the receptance matrix has to be calculated for each frequency separately. This
can become very time consuming for large problems, especially because the dynamic stiffness
matrix is complex. A different approach to the harmonic vibration problem is the application
of the modal superposition.

Structural Dynamics 9 Forced vibration of linear systems with multiple degree of freedoms
summer term 2019 152
 Example 9.1: Two degree of freedom system (cont)

We continue example 8.2 and apply a harmonic forces to the degrees of freedom.

m1 f1 (t)
w1 m1 = 80 kg
m2 = 8 kg
k1 c1 N
k1 = 200 m
f2 (t) N
k2 = 125 m
m2
w2 c1 = 0
c2 = 0.6 Ns
m
k2 c2

The loads p1 (t) and p2 (t) are now harmonic forces, such that

f (t)
! ! !
f f
f (t) = 1 = 1+ e iΩt + 1− e −iΩt = f + e iΩt + f − e −iΩt
f2 (t) f2+ f2−

We only consider the parts with positive frequency f + e iΩt since the parts with negative
frequencies can be found as the complex conjugates of the former. In this example
we choose f1+ = 1 and f2+ = 0. We consider the system without and with damping.
Subsequently, the solution is depicted for both degrees of freedom. Real part, imaginary
part, absolute value and phase are given.

Structural Dynamics 9 Forced vibration of linear systems with multiple degree of freedoms
summer term 2019 153
 Figure 9.1: Displacement w1+ (—) and w2+ (–·–) for the 2DOF system under harmonic
excitation, direct solution, undamped system.

Figure 9.2: Displacement w1+ (—) and w2+ (–·–) for the 2DOF system under harmonic
excitation, direct solution, damped system.

Structural Dynamics 9 Forced vibration of linear systems with multiple degree of freedoms
summer term 2019 154
9.2 Modal analysis

The modal analysis is based on the superposition of eigenmodes (eigenvectors) φi , that are
scaled with time-dependent factors yi (t). It is only applicable for linear vibrations, since only
then the principle of superposition is valid. For systems with a diagonalizable damping matrix,
the displacement vector of a MDOF system is then written as
n
w(t) = yi (t)φi = Φy(t) (9.15)
X

with the matrix of eigenvectors

Φ = [φ1 , φ2 , . . . , φn ] (9.16)

obtained from the free vibration problem of the undamped system and the coefficients

y(t) = [y1 (t), y2 (t), . . . , yn (t)]T . (9.17)

The equation of motion for the MDOF system including external forces reads

Mẅ + Cẇ + Kw = f (t) . (9.18)

We assume that the eigenvectors and -values are known from a previous solution of the un-
damped, homogeneous problem (comp. section 8.2). Inserting Eq. (9.15) into Eq. (9.18), we
obtain

MΦÿ + CΦẏ + KΦy = f . (9.19)

Then, multiplying Eq. (9.19) with ΦT , it follows

ΦT MΦÿ + ΦT CΦẏ + ΦT KΦy = ΦT f (9.20)

or written separately for each equation

n n n
! ! !
φTj M φi ÿi + φTj C φi ẏi + φTj K φi yi = φTj f . (9.21)
X X X

i i i

Structural Dynamics 9 Forced vibration of linear systems with multiple degree of freedoms
summer term 2019 155
 Reminder 9.1

From section 8.2 we know that the following relations hold

ΦT MΦ = M∗
ΦT KΦ = K∗ = ΛM∗

where Λ is the diagonal matrix of squared eigenfrequencies. In the case of mass nor-
malized eigenvectors this further reduces to

ΦT MΦ = I
ΦT KΦ = Λ

We furthermore require that the damping matrix is diagonalizable, i.e., C∗ is a diagonal matrix,
which can, e.g., be obtained by Rayleigh damping or a modal damping approach. With this
Eq. (9.20) reads

M∗ ÿ + C∗ ẏ + K∗ y = ΦT f .

Thereby all mixed terms vanish from Eq. (9.20), and the equations of motion become decou-
pled. We can obtain a notation for the separate equations by writing Eq. (9.21) as

φTj Mφj ÿj + φTj Cφj ẏj + φTj [K] φj yj = φTj p .

| {z } | {z } | {z } | {z }
Mj Cj Kj fj

Thus, n different and decoupled ordinary differential equations in the form of the SDOF
differential equation are obtained.

Mj ÿj (t) + Cj ÿj (t) + Kj yj (t) = fj .

Therein are

Mj = φTj Mφj generalized (modal) mass

Cj = φTj Cφj = 2Dj ωj Mj generalized (modal) damping
Kj = φTj Kφj = ωj2 Mj generalized (modal) stiffness
pj = φTj f generalized load

Structural Dynamics 9 Forced vibration of linear systems with multiple degree of freedoms
summer term 2019 156
Remember that this derivation of decoupled equations only holds under the assumption that
the damping matrix is diagonalizable. In the special case of mass-normalized eigenvectors the
differential equations simplify to

ÿj + 2Dj ωj ẏj + ωj2 yj = fj .

Depending on the form of fj (t), the solution for each yj (t) can then be found by any of
the methods for the forced SDOF  systemdiscussed in chapter 7. For example, in case of a
harmonic excitation f (t) = f 0 cos Ωt + ϕf , all modal forces will also be harmonic. Then, the
steady state (particular) solutions for single modes can be easily obtained by application of
the amplification function V (η). In the case of transient loads, the response can be determined
by the Duhamel integral. The physical solution w is finally obtained by inserting the solution
y for in Eq. (9.15).

Example 9.2: Two degree of freedom system (cont.)

We continue example 8.2 and apply two step loads to the two degrees of freedom,
respectively.

m1 f1 (t)
w1 m1 = 80 kg
m2 = 8 kg
k1 N
k1 = 200 m
f2 (t) N
k2 = 125 m
m2
w2
k2

MΦÿ(t) + KΦy(t) = f (t)

By left multiplication with ΦT we obtain

ÿ(t) + Λy(t) = ΦT f (t)

Structural Dynamics 9 Forced vibration of linear systems with multiple degree of freedoms
summer term 2019 157
 The load is given by

f1 (t) f U (t − t1 )
" # " #
= 10
f2 (t) f20 U (t − t2 )

where U (t) is the unit-step function


1 for t ≥ 0
U (t) =  .
0 for t < 0

The mass normalized eigenvectors of the system that are derived in example 8.2 read

0.110
" #
m
φ1 =
0.069
0.022
" #
m
φ2 = .
−0.347

The generalized load thus results in

#T "
Φ Φ12 f1 (t) Φ11 p1 (t) + m Φ21 f2 (t)
" # " #
m m m
f (t) = Φ f (t) = m 11
∗ m T
= .
Φ21 m
Φ22 f2 (t) m
Φ12 p1 (t) + m Φ22 f2 (t)

Since the equations of motion decouple under the transformation to the modal coor-
dinates, the Duhamel-integral can be applied. As the decoupled equations correspond
the the differential equation of the SDOF system, the impulse response function of the
undamped SDOF system, given by

1 1
hj (t) = q sin (ωj t) = sin (ωj t)
kj mj ωj

can be used in the integral equation directly:

Z∞ ∞
1 Z
yj (t) = fj (τ )hj (t − τ ) dτ = fj (τ ) sin(t − τ ) dτ .
ωj
−∞ −∞

Thus:
∞
1 Z
y1 (t) = f1 (τ ) sin(ω1 (t − τ )) dτ =
ω1
−∞

Structural Dynamics 9 Forced vibration of linear systems with multiple degree of freedoms
summer term 2019 158
 ∞
1 Z m
= ( Φ11 f1 (t) + m Φ21 f2 (t)) sin(ω1 (t − τ )) dτ =
ω1
−∞
∞
1 Z m
= ( Φ11 f10 U (t − t1 ) + m Φ21 f20 U (t − t2 )) sin(ω1 (t − τ )) dτ =
ω1
−∞
∞
1 Z m
= Φ11 f10 U (t − t1 ) sin(ω1 (t − τ )) dτ + .
ω1
−∞
∞
1 Z m
+ Φ21 f20 U (t − t2 ) sin(ω1 (t − τ )) dτ
ω1
−∞

Since U (t − t1 ) = 0 and U (t − t2 ) = 0 for t < t1 and t < t2 , respectively, it holds

∞ ∞
1 Z m 1 Z m
y1 (t) = Φ11 f10 sin(ω1 (t − τ )) dτ + Φ21 f20 sin(ω1 (t − τ )) dτ =
ω1 ω1
t1 t2

Φ11 f10 1 Φ21 f20 1

m  t m  t
= − cos(ω1 (t − τ )) + − cos(ω1 (t − τ )) =
ω1 ω1 t1 ω1 ω1 t2
m
Φ11 f10 m
Φ21 f20
= 2
[1 − cos(ω1 (t − t1 ))] + [1 − cos(ω1 (t − t2 ))] .
ω1 ω12

For the second modal coordinate y2 (t) we obtain analogously

m
Φ12 f10 m
Φ22 f20
y2 (t) = 2
[1 − cos(ω2 (t − t1 ))] + [1 − cos(ω2 (t − t2 ))] .
ω2 ω22

The solution in terms of the degrees of freedom w(t) is then given by

y (t)
" #
w(t) = Φy(t) = Φ 1 = φ1 y1 (t) + φ2 y2 (t) .
y2 (t)

We calculate the response for f10 = 10, f20 = −5, t1 = 1 t2 = 3. The results are
depicted in the following figure:

Structural Dynamics 9 Forced vibration of linear systems with multiple degree of freedoms
summer term 2019 159
 Figure 9.3: Solution for the 2DOF system under two step loads. The left figure and right
figure show the displacement solution for degree of freedom 1 and 2, respectively
(—). Each figure gives the separate contributions from mode shape 1 (–·–) and
mode shape 2 (- - -). The load application times are marked by vertical lines
(|) It can be observed that the solution for both DOFs is mainly governed by the
2π
contributions from mode shape 1. The periods are T1 = ω 1
= 2π = 6.53 s 0.96 Hz
2π 2π
and T2 = = = 0.97 s.
ω2 6.5 Hz

9.3 Time integration methods

If the application of a diagonal damping matrix is not valid and the excitation cannot be
split in single harmonic parts, the two methods presented above cannot be applied. Then,
numerical time integration methods have to be applied. Different methods can be applied,
such as difference methods, Crank-Nicholson, Houboldt-, Newmark, or generalized α methods.
Some of these are presented in chapter 16. These procedures can become computationally
very expensive, as the full system has to be solved in every time step.

9.4 Summary

• Vibrations of system with n degrees of freedom under arbitrary, time dependent ex-
citation can be described with the help of the modal analysis. Thereby the system is
decoupled into n SDOF systems, which can be solved for separately and then superposed
to obtain the complete solution. The resulting parameters of the single SDOF systems
are called generalized quantities (generalized stiffness, mass, damping, and load). They
can be interpreted by the kinetic and potential energy and the externally introduced
work.
• The number of degrees of freedom that describe the systems’ behavior is significantly
smaller than the absolute number of DOFS n. This can be explained by the fact, that

Structural Dynamics 9 Forced vibration of linear systems with multiple degree of freedoms
summer term 2019 160
 the stiffness term ωj2 Mj strongly increases for higher frequent modes. An exception to
this is the case, where the system is excited at resonance.
• For arbitrary damping, the generalized damping matrix is not diagonal in the modal co-
ordinates. Only in the case, where the damping matrix is given as a specific combination
of mass and stiffness matrix, diagonality is given.
 b
C= αb M M−1 K
X

An often applied form of this damping model is the Rayleigh damping, which is achieved
for b = {0,1}:

CRayleigh = αM + βK

The damping factors α and β of the Rayleigh damping are adopted to realistic damping,
as far as the damping is distributed continuously over the system. In many cases, e.g.
for single damper elements, an adoption via the Rayleigh damping is not possible. Then,
a complex eigenvalue analysis can be performed.
• With the help of the Fourier transformation, arbitrary loads, as presented for the SDOF
system, can be transformed to harmonic loads, that then act on the different eigenmodes.

Structural Dynamics 9 Forced vibration of linear systems with multiple degree of freedoms
summer term 2019 161
10 Impedance models
With the improvement of measurement capabilities, it has become common to characterize
parts of a dynamic system by means of their dynamic stiffness (or impedance in general) and
receptance (or admittance in general) matrix and to use a dynamic substructuring approach
for the analysis of the dynamic behavior of the parts separately prior to the coupling. This
reduces the computational time and improves the repeatability of numerical analysis. In
vibroacoustics, the most used quantity to describe the vibrational behavior of a system or a
subsystem is the impedance, which is defined as the ratio of the force spectrum to the resulting
velocity spectrum at the loading point and describes how much a structural part resists to
motions when subjected to a harmonic force.
Impedance models provide an alternative simplified technique to FEM or analytical solutions
for vibroacoustic predictions. Impedance modeling is a powerful tool for the simplified pre-
diction of traffic-induced vibrations, or for the design with respect to acoustic requirements
that require sophisticated regarding vibrations. They are also used in noise, vibration, and
harshness control for cars and airplanes.

10.1 Preliminary definitions

In the following we assume small amplitudes and linear systems. Time harmonic (sine/cosine)
displacements w(t) result from harmonic forces f (t). The system response will be characterized
by the velocity v(t) at the point of excitation. Furthermore, the force is defined in the same
direction as the velocity (comp. Fig. 10.1a). A dynamic reaction force r(t) occurs in the
system, which may result from different mechanisms. It is in equilibrium with the load f (t).

Structural Dynamics
summer term 2019
162
 f (t)

w,v
r(t)

(a) Dynamic Equilibrium (b) Impedance hammer

Figure 10.1: Reaction force under harmonic load

For the following considerations we choose the advantageous description using complex num-
bers

f (t) = fb+ e iΩt + fb− e −iΩt

1 1
= (f01 + if02 )e iΩt + (f01 − if02 )e −iΩt = f01 cos (Ωt) − f02 sin (Ωt) (10.1)
2 2
v(t) = ẇ(t) = vb+ e + vb− e −iΩt
iΩt

1 1
= (v01 + iv02 )e iΩt + (v01 − iv02 )e −iΩt = v01 cos (Ωt) − v02 sin (Ωt) (10.2)
2 2

where Ω = 2π T
is the circular frequency and fb+ and fb− as well as vb+ and vb− are complex
conjugate to each other.

1 Im Im
Im Im v
2 02 vb+
1 fb+
f
2 02

−ϕf 1
f ϕv 1
Re 2 01 Re Re v
2 01 Re
1 1
ϕf f
2 01
v
2 01 −ϕv
− 12 f02 fb−
− 12 v02 vb−

Figure 10.2: Complex pointers fˆ+ and fˆ− as well as v̂+ and v̂−

Structural Dynamics 10 Impedance models

summer term 2019 163
 fmax f (t)
vmax

v(t)

Figure 10.3: Physical quantities f (t) and v(t). One period of the harmonic oscillation is depicted,
where f (t) and v(t) show an arbitrary phase shift ∆ϕ.

After applying an integration or differentiation to the velocity v(t), the vibration can also be
expressed in terms of the displacement w(t) or the acceleration a(t) = ẅ(t), respectively.

1 1
 
w(t) = vb+ e iΩt + − vb− e −iΩt = wb+ e iΩt + wb− e −iΩt (10.3)
iΩ iΩ
a(t) = ẅ(t) = iΩvb+ e iΩt + (−iΩvb− ) e −iΩt = ab+ e iΩt + ab− e −iΩt (10.4)

where the complex pointers are

1 1
ŵ+ = v̂+ = −i v̂+ (10.5)
iΩ Ω
1 1
ŵ− = − v̂− = i v̂+ (10.6)
iΩ Ω
â+ = iΩv̂+ (10.7)
â− = −iΩv̂− . (10.8)

The input power inserted at the the excitation point is given by

P (t) = f (t)v(t). (10.9)

Thus, the work within one period (one cycle of the load) is
Z T
W = P (t) dt
0
Z T
= (fˆ+ e iΩt + fˆ− e −iΩt )(v̂+ e iΩt + v̂− e −iΩt ) dt
0

Structural Dynamics 10 Impedance models

summer term 2019 164
 Z T
= fb+ vb+ e 2iΩt + fb− vb− e −2iΩt + fb+ vb− e 0 + fb− vb+ e 0 dt . (10.10)
0

The terms with e 2iΩt and e −2iΩt integrate to zero. Thus it follows
 
W = fb+ vb− + fb− vb+ T. (10.11)

We can also represent the complex numbers fb+ and vb+ as

   
fb+ = fb+ e iϕf , fb− = fb− e −iϕf (10.12)
   

vb+ = |vb+ |e iϕv , vb− = |vb− |e −iϕv . (10.13)

Inserting Eqs. (10.12) and (10.13) into Eq. (10.11), we obtain

   
W = fb+ |vb− |e iϕF e −iϕv + fb− |vb+ |e −iϕF e iϕv T
   
= fb+ |vb+ | e i(ϕv −ϕF ) + e −i(ϕv −ϕF ) T. (10.14)
 
| {z }
=2 cos(ϕv −ϕF )

   
Here we use the fact that fb+  = fb−  and |vb+ | = |vb− |. Finally, we obtain

Z T  
W = P (t) dt = 2fb+ |vb+ |T cos (ϕv − ϕF ) . (10.15)
0

This result for the work W is depicted in Fig. 10.4 in dependency of ϕv −ϕF . We can distinguish
between regions where the work performed is positive or negative

W
positive work

π 2π 3π 4π ϕv − ϕF
negative work

Figure 10.4: Work performed by the force f (t).

π π
 
ϕv − ϕF ∈ − , → positive work is performed on the system
2 2

Structural Dynamics 10 Impedance models

summer term 2019 165
 π 3π
 
ϕv − ϕF ∈ , → negative work is performed on the system
2 2

We can thus state: hIf the iphase shift between velocity v(t) and the force f (t) is within the
range (ϕv − ϕF ) ∈ − π2 , π2 the force, which is in equilibrium with the reaction force in the
system, carries out positive work. This work has got a maximum for ϕv = ϕF and it is equal
to zero for ϕv − ϕF = ± π2 . Due to the requirement of a stationary harmonic vibration, which
is defined in Eq. (10.2), this work has to hbe “consumed”
i by the system in each period. For
the phase shift in the range (ϕv − ϕF ) ∈ 2 , 2 , we observe negative work. Thus, an energy
π 3π

source would have to exist within the system.

10.2 Dynamic stiffness and impedance

In statics the quotient of exciting force and displacement corresponds to the stiffness (“spring”).
A dynamic equivalent is found for

fb+
kb+ = (10.16)
wb+
fb−
kb− = . (10.17)
wb−

Here, k̂+ and k̂− is the dynamic stiffnesses, also denoted by kdyn (Ω). The quotient of the force
and the velocity at the excitation point is called impedance Z(Ω).

fb+
Z(Ω) = zb+ = (10.18)
vb+
fb−
Z(−Ω) = zb− = . (10.19)
vb−

It can be shown that both parts are complex conjugates of each other.
!∗
F− F+∗ F+
z− =
b = ∗ = = zb+
∗
. (10.20)
v− v+ v+

Impedance and dynamic stiffness are, in general, quantities that depend on the frequency and
can be determined with the help of measurements. For this purpose we have to simultaneously
measure force and velocity, displacement, or acceleration, respectively. Similarly, the quotient
of force and acceleration can be computed, which is called dynamic mass (in German, “Reak-
tanz”). The inverse of the impedance is called mobility (in German, “Admittanz”); the inverse

Structural Dynamics 10 Impedance models

summer term 2019 166
of the dynamic stiffness is called receptance (flexibility) (in German, “Nachgiebigkeit”); the
inverse of the dynamic mass is called accelerance (in German, “Inertanz”).

10.3 Dynamic stiffnesses and impedances in passive

systems

In the following we only consider system, where forces insert mechanical work, that is dissipated
by the (passive) system. The system can only dissipate energy, but not cannot generate energy,
e.g., by an internal energy source. Thus, no work can be done on the exciting forceh by the i
system, and the phase shift between velocity and force is in the range (ϕv − ϕF ) ∈ − π2 , π2 .
In the following the reaction forces r(t) that are in equilibrium with the external load f (t)
are discussed depending on the displacements at the excitation point (comp. Fig. 10.1). In
contrast to statics it is important, which phase shifts occur between force and velocity. These
relations are expressed with dynamic stiffnesses or impedances:

f (t) = fb+ e iΩt + fb− e −iΩt = kb+ wb+ e iΩt + kb− wb− e −iΩt = zb+ vb+ e iΩt + zb− vb− e −iΩt . (10.21)

10.3.1 Mass impedance

According to Newton’s law, the inertial force of a single mass point is

mẅ = f = (10.22)
X
fn .
n

Here, f is the resultant of all forces f n acting on a point with a mass m. This equation can be
rearranged as

f n − mẅ = 0. (10.23)
X

This equation constitutes the “dynamic equilibrium” for an accelerated system, as the mass
is accelerated. This equilibrium corresponds to the static equilibrium equation for masses
moving at a constant speed except for the term −mẅ.

fn = 0 . (10.24)
X

The reaction force −mẅ is directed in the opposite direction of the acceleration ẅ.

Structural Dynamics 10 Impedance models

summer term 2019 167
 f (t)

m
w,ẇ,ẅ

m ẅ(t)

Figure 10.5: Mass excited by dynamic force.

In case of a harmonic force f (t) = fb+ e iΩt + fb− e −iΩt acting on the mass Eq. (10.23) results in

f (t) = mẅ(t) . (10.25)

Thus, with Eq. (10.4), it follows

 
f (t) = fb+ e iΩt + fb− e −iΩt = m iΩvb+ e iΩt + (−iΩvb− ) e −iΩt . (10.26)

Thus, fb+ and fb− are given by

fb+ = imΩvb+ (10.27)

fb− = −imΩvb− . (10.28)

The force with positive exponential undergoes a phase shift relative to the velocity due to the
multiplication with the imaginary unit i, which is ϕf+ − ϕv+ = π2 corresponding to 90◦ . For the
part with negative exponential the phase shift is ϕf− − ϕv− = − π2 due to multiplication with
−i Thus, the force f (t) is “hurrying on ahead” of the velocity v(t). For this case, according to
Eq. (10.14), no work is done on the system, and thus no energy is consumed. This can also be
observed in Fig. 10.4.
The dynamic stiffnesses in Eqs. (10.16) and (10.17) therefore read under consideration of
Eqs. (10.3), (10.5), and (10.6),

fb+ imΩvb+
kb+ = = = −mΩ2 (10.29)
wb+ −i Ω1 vb+
fb− −imΩvb−
kb− = = = −mΩ2 . (10.30)
wb− i Ω1 vb−

Structural Dynamics 10 Impedance models

summer term 2019 168
The impedances in Eqs. (10.18) and (10.19) under consideration of Eq. (10.3) follow to

fb+ imΩvb+
zb+ = = = iΩm (10.31)
vb+ vb+
fb− −imΩvb−
zb− = = = −iΩm. (10.32)
vb− vb−

10.3.2 Conservative forces

10.3.2.1 Potential

Forces performing work that can be explicitly characterized by the initial and the endpoint of
the performance, have a potential Π , i.e., the work, which is done by the load on the system,
is independent from the loading process and it only depends on the initial and the endpoint
of the loading process. In case of forces that can be defined by means of a potential, the work
due to a loading and subsequent relaxation to the initial state is equal to zero. An example of
a force, which can be defined with the help of a potential, is the resulting force Fk of a linear
elastic spring

Fk = −kw w (10.33)

where kw is the spring-stiffness. Another example is the resulting moment of a torsional

spring

Mγ = −kγ γ

where kγ is the stiffness of the rotational spring. The minus sign indicates (comp. Fig. 10.1)
that the spring force is acting against the displacement at the excitation point of the external
load (the moment of the rotational spring is acting against the rotation) and is in equilibrium
with the external load (with the external moment):

fk + f = 0 (10.34)
−kw w + f = 0 . (10.35)

In case of a longitudinal spring, the potential is defined as

1
Π = kw w2 . (10.36)
2

Structural Dynamics 10 Impedance models

summer term 2019 169
The reaction force can be deduced from the potential

δΠ
f =− = −kw w . (10.37)
δw

10.3.2.2 Spring impedance

fb+ e iΩt + fb− e−iΩt

kw
wb+ e iΩt + wb− e −iΩt

Figure 10.6: Spring under harmonic force.

In case of an external harmonic force f (t) = fb+ e iΩt + fb− e −iΩt acting on a spring, the resulting
force is obtained from the linear relation:

f (t) = kw w(t) . (10.38)

Thus, with Eq. (10.3), it follows

1 1
   
f (t) = +e
fb iΩt
+ fb
−e
−iΩt
= kw vb+ e iΩt + − vb− e −iΩt . (10.39)
iΩ iΩ

Then fb+ and fb− are given by:

kw
fb+ = −i vb+ (10.40)
Ω
kw
fb− = i vb+ . (10.41)
Ω

This implies that the force with positive exponential undergoes a phase shift with respect to
the velocity of ϕf+ − ϕv+ = − π2 corresponding to −90◦ . For the part with negative exponential
the phase shift is ϕf− − ϕv− = π2 . That is, the force f (t) is “lagging behind” the velocity v(t).
In this case, similar to the case where the force acts on a mass, no work W is done on the
system (according to Eq. (10.14)) and no energy is dissipated. The values for kb+ and kb− under

Structural Dynamics 10 Impedance models

summer term 2019 170
consideration of Eqs. (10.16) and (10.17) are given by

fb+ −i kΩw vb+

kb+ = = = kw (10.42)
wb+ −i Ω1 vb+
fb− i kw vb+
kb− = = Ω1 = kw . (10.43)
wb− i Ω vb+

They are equal to the spring stiffness kw and thus positive and real numbers. Analogously, we
obtain the impedances from Eqs. (10.18) and (10.19):

fb+ −i kΩw vb+ kw

zb+ = = = −i (10.44)
v+
b v+
b Ω
fb− i kw vb− kw
zb− = = Ω =i . (10.45)
vb− vb− Ω

10.3.3 Non-conservative forces

10.3.3.1 Damping impedance

For the viscous damper (Newton-Modell) a linear relation between force and velocity is given

f (t) = cv(t) = cẇ(t) (10.46)

where c is the constant of the viscous damper. The damping force is acting oppositely to the
velocity v(t).

fb+ e iΩt + fb− e−iΩt

c
wb+ e iΩt + wb− e −iΩt

Figure 10.7: Viscous damper under harmonic force.

For a stationary harmonic vibration it follows

 
f (t) = fb+ e iΩt + fb− e −iΩt = c vb+ e iΩt + vb− e −iΩt . (10.47)

Structural Dynamics 10 Impedance models

summer term 2019 171
Then, fb+ and fb− are given by

fb+ = cvb+ (10.48)

fb− = cvb− . (10.49)

For both force terms, no phase shift occurs between force and velocity. The values for kb+ and
kb− under consideration of Eqs. (10.16) and (10.17) follow to

fb+ cvb+
kb+ = = 1 = iΩc (10.50)
wb+ vb
iΩ +

fb− cvb−
kb− = = 1 = −iΩc . (10.51)
wb− − iΩ vb−

Hence for viscous damping without a spring frequency proportional, purely imaginary dynamic
stiffness is obtained. Analogously, we obtain the impedances from Eqs. (10.18) and (10.19):

fb+ cvb+
zb+ = = =c (10.52)
vb+ vb+
fb− cvb−
zb− = = = c. (10.53)
vb− vb−

The impedance is equal to the viscous damping coefficient and thus purely real.

10.3.3.2 Loss factor

For passive systems the complex dynamic stiffness

h ican only lead to a phase shift between
velocity and force in the range (ϕF − ϕv ) ∈ − 2 , 2 . Except for the limit cases − π2 and π2
π π

energy is dissipated.
The forces are written in terms of their dynamic stiffnesses and the velocity.

1
 
fb+ = = + −i vb+
kb+ wb+ kb (10.54)
Ω
1
 
fb− = kb− wb− = kb− i vb− . (10.55)
Ω

Structural Dynamics 10 Impedance models

summer term 2019 172
Thus, we obtain for the dynamic stiffness

f+
kb+ = iΩ
v+
f01 + if02
= iΩ
v01 + iv02
f01 v01 + f02 v02 + i (f02 v01 − f01 v02 )
= iΩ 2
v01 + v02
2

Ω
= 2 [(f01 v02 − f02 v01 ) + i (f01 v01 + f02 v02 )]
v01 + v022

f01 v01 + f02 v02

" #
f01 v02 − f02 v01
=Ω 1+i (10.56)
v01 + v02
2 2
f01 v02 − f02 v01

and

f−
kb− = −iΩ
v−
f01 − if02
= −iΩ
v01 − iv02
f01 v01 + f02 v02 + i (f01 v02 − f02 v01 )
= −iΩ 2
v01 + v02
2

Ω
=− 2 [(f02 v01 − f01 v02 ) + i (f01 v01 + f02 v02 )]
v01 + v022

f01 v01 + f02 v02

" #
f01 v02 − f02 v01
=Ω 1−i . (10.57)
v01 + v02
2 2
f01 v02 − f02 v01

The dynamic stiffness kdyn = kb+ is also termed as “complex spring”. It is often represented as
follows
     
kdyn = Re kb+ + i Im kb+ = Re kb+ (1 + iη) . (10.58)

The parameter

Im(kb+ ) f01 v01 + f02 v02

η= = . (10.59)
Re(kb+ ) f01 v02 − f02 v01

is called loss factor (comp. Table 5.1). For coupled systems, the real part of the complex spring
kdyn results from spring stiffnesses and masses. The imaginary part of kdyn is decisive for the
energy dissipation. The loss factor η can be explained with the help of the model depicted

Structural Dynamics 10 Impedance models

summer term 2019 173
in Fig. 10.8a. In a parallel arrangement of a spring with a stiffness k and a viscous damper
with a damping coefficient c a complex spring stiffness with a frequency independent real part
(resulting from the spring) as well as a frequency-dependent imaginary part (resulting from
the damper) results from the sum of both the dynamic stiffnesses. The hysteresis shows the
shape depicted in Fig. 10.8b.

f
fmax fmax f (t)
wmax

w(t)
w
kw c wmax t

f
2π
(a) System (b) Hysteresis und time history

Figure 10.8: Voigt-Kelvin-Modell, Petersen [1996].

According to Eq. (10.11), the work dissipated in one period is

 
Wdiss = fb+ vb− + fb− vb+ T. (10.60)
1 1 1 1
 
= (f01 + if02 ) (v01 − iv02 ) + (f01 − if02 ) (v01 + iv02 ) T (10.61)
2 2 2 2
1
= [f01 v01 − if01 v02 + if02 v01 + f01 v01 + if01 v02 − if02 v01 ] T (10.62)
4
1
= (f01 v01 + f02 v02 ) T . (10.63)
2

On the other hand, for a linear elastic spring with a spring stiffness kw = Re(kdyn ) =
Ω f01 vv02
2
−f02 v01
2 (Eq. (10.56)), the energy Welast stored in the spring is for the maximum dis-
01 +v02
placement:

1 1
Welast = kw wmax
2
= Re(kdyn )wmax
2
. (10.64)
2 2

Structural Dynamics 10 Impedance models

summer term 2019 174
Under consideration of Eq. (10.5) it follows

v 2|vb+ |
 
b+ 
wmax = 2|wb+ | = 2 
 = , (10.65)
iΩ Ω

and thus

1 f01 v02 − f02 v01 |vb+ |2

Welast = Ω 4 2 = (10.66)
2 (2|vb+ |)2 Ω
1
= (f01 v02 − f02 v01 ) . (10.67)
2Ω

Hence the ratio Wdiss

Welast
reads

Wdiss 1
(f01 v01 + f02 v02 ) T (f01 v01 + f02 v02 ) T f01 v01 + f02 v02
= 2
= T = 2π = 2πη (10.68)
Welast 1
2Ω
(f01 v02 − f02 v01 ) 2π
(f01 v02 − f02 v01 ) f01 v02 − f02 v01
| {z }
=η

where we identified η as in Eq. (10.59). Thus, the value 2πη gives the ratio of dissipated to
restorable mechanical energy.

10.4 Parallel and serial connection

The representation of mechanical quantities using complex numbers provides an elegant pos-
sibility to describe steady state vibrations of mechanical systems under harmonic loads. As
discussed in Section 7.2.1.2, steady state implies that the transient effects due to initial con-
ditions and activation operations already have decayed.
Eqs. (10.32), (10.45), and (10.53) define the impedances of a single mass, a spring, and a
viscous damper. For each of these elements the impedance describes the complex ratio of the
force fˆ+ and the velocity v̂+ at the excitation point. Assembling subsystems consisting of the
already mentioned elements mass, spring, and damper in serial and parallel connections, we
can deduce total impedances for complex systems (similar to the combination of springs in
statics). It is useful to compute the impedance of coupled systems starting from the opposite
side of the load.

10.4.1 Parallel connnection

For subsystems that are subjected to the same displacement, and thus, the same velocity v̂+
at the excitation point, the impedances are summed up because the total force fbtot+ results

Structural Dynamics 10 Impedance models

summer term 2019 175
from the individual resistant forces of the subsystems. That is, the force has to “work against”
the sum of the individual impedances. For a parallel connection composed of n subsystems,
we obtain:

fbtot+ = vb+ zb1+ + vb+ zb2+ + . . . + vb+ zbn+

= vb+ (zb1+ + zb2+ + . . . + zbn+ ) . (10.69)

Thus, the impedance is given by

fbtot+
zbtot+ = = zb+1 + zb2+ + . . . + zbn+ . (10.70)
vb+

10.4.2 Serial connnection

In case of a serial connection of subsystems, the individual displacements are summed up,
whereas the force remains the same in all the subsystems. Then, we obtain the following
equation for the total impedance (serial connection of impedances):

vbtot+ = vb1+ + vb2+ + ... + vbn+

fb+ fb+ fb+
= + + ... +
zb1+ zb2+ zbn+
1 1 1
!
= fb
+ + + ... + . (10.71)
zb1+ zb2+ zbn+

Thus, the impedance is given by

Fb+ 1
zbtot+ = = 1 1 . (10.72)
vbtot+ z1+
b + bz2+ + ... + bzn+
1

10.5 Arbitrary connections

Using the idea of impedances, one can easily assess, e.g., the effect of additional masses installed
in order to reduce vibrations (addition of a mass-impedance) or an elastic support (serial
connection with a spring-impedance).
Herafter, we renounce parts with negative frequencies, and only consider the part with positive
frequencies as in, e.g., f (t) = fb+ e iΩt + fb− e −iΩt . The parts with negative frequencies can be
obtained by complex conjugation, as the impedance for the negative frequency part is the
complex conjugate of the part for positive frequencies.

Structural Dynamics 10 Impedance models

summer term 2019 176
 Example 10.1: SDOF system

The following SDOF system with mass m , spring stiffness k and viscous damping c is
considered.

k m
fb+ e iΩt

c vb+ e iΩt

Here, the displacements and thus velocities of the three elements are equal. This corre-
sponds to a parallel connection of mass, spring and damper. Hence, we obtain for the
total impedance Z of the SDOF system

k
z+ = zm+ + zc+ + zk+ = iΩm + c + .
iΩ

The dynamic stiffness kdyn = wfb++ can be computed from the impedance by multiplication
b

with iΩ considering Eq. (10.3):

kdyn = −Ω2 m + iΩc + k.

Thus, we immediately obtain the well-known equation of the dynamic stiffness of the
harmonically forced SDOF system (cf. Section 7.2.1). The transfer function is given by
the inverse relation

1 1
U= = .
kdyn −Ω2 m + iΩc + k

The impedance and transfer function are plotted over the excitation frequency in the
following figure:

Structural Dynamics 10 Impedance models

summer term 2019 177
 (a) Impedance (b) Transfer function
Figure 10.9: SDOF impedance model for different values of damping ratio, D = 0.05 (—),
D = 0.2 (–·–), D = 0.7 (- - -).

Example 10.2: Inserting an elastic element

The following system is considered

m
fb+ e iΩt
k c
vb+ e iΩt

In this system the spring and the damper carry the same force. The displacements sum
up. Therefore these two elements can be interpreted as a serial connection.

zc,k+ vb+ e iΩt

+e
fb iΩt
=
ˆ f+ e
b iΩt

k c vb+ e iΩt vb+ e iΩt

Thus, the total impedance of these two subsystems at the interface to the mass is

1 1
zc,k+ = = .
1
zk+
+ 1
zc+
iΩ
k
+ 1
c

Structural Dynamics 10 Impedance models

summer term 2019 178
 The endpoint of the subsystem spring-damper and the mass are subjected to the same
displacement. Therefore, the connection of the spring-damper and mass subsystem is
a parallel connection.

m
zc,k+ vb+ e iΩt fb+ e iΩt

vb+ e iΩt

Hence, for the computation of the total impedance we have to add the mass impedance
as shown below:

1
z+ = iΩm + .
iΩ
k
+ 1
c

(a) Impedance (b) Transfer function

Figure 10.10: SDOF impedance model for different values of damping ratio, D = 0.05 (—),
D = 0.2 (–·–), D = 0.7 (- - -).

Structural Dynamics 10 Impedance models

summer term 2019 179
 Example 10.3: Undamped 2DOF system

The following undamped 2DOF system is considered.

m1 m2
fb+ e iΩt
k1 k2
vb+ e iΩt

We first represent the left subsystem by its impedance.

m1
zk1 ,m1 + vb+ e iΩt
fb+ e iΩt = fb+ e iΩt
k1
vb+ e iΩt vb+ e iΩt

Therefore, we calculate the impedance for the parallel connection of the spring k1 and
the mass m1 .

k1
zk1 ,m1 + = zm1 + + zk1 + = iΩm1 + .
iΩ

In the next step, the spring k2 is added. This spring carries the same force as that of the
first subsystem k1 , m1 , thus, the first subsystem and the second spring are connected
serially. Hence, we obtain the impedance of the system k1 , m1 , and k2 as shown below.

zk1 ,m1 ,k2 + vb+ e iΩt

zk1 ,m1 + vb+ e iΩt
+e
fb iΩt
=
ˆ fb+ e iΩt
k2 vb+ e iΩt vb+ e iΩt

It results

1
zk1 ,m1 ,k2 + = 1 1 .
+
k1 k2
iΩm1 +
iΩ iΩ

Structural Dynamics 10 Impedance models

summer term 2019 180
 Finally, the mass m2 is coupled to the subsystem k1 , m1 and k2 . The displacement of
the mass m2 is equal to the displacement at the endpoint of spring k2 . Thus, the mass
m2 is coupled to the subsystem k1 , m1 and k2 in a parallel connection.

m2
zk1 ,m1 ,k2 + vb+ e iΩt fb+ e iΩt

vb+ e iΩt

It follows

1
z+ = zk1 ,m1 ,k2 ,m2 + = 1 + iΩm2 .
+ k1
k1 2
iΩm1 +
iΩ iΩ

(a) Impedance (b) Transfer function

Figure 10.11: 2DOF impedance model.

10.6 Impedances for selected systems

The representation of substructures through impedances or dynamic stiffnesses, respectively,

enables the coupling to other substructures that can also be described via impedances. This
way also structures defined on infinitely extended domains can be represented. In Tab. 10.1
selected impedances at the excitation points of point forces for single or two sided infinitely
extended domains are given.

Structural Dynamics 10 Impedance models

summer term 2019 181
Furthermore, results from measured impedances can be directly considered. For example, the
dynamic stiffness of a foundation on an elastic isotropic halfspace can be coupled in serial or
parallel to the dynamic stiffness of a structure (e.g., see Studer et al [2007]).
√
rod Z = A Eρ

beam Z = 2ρAcB (1 + i)

beam Z = 12 ρAcB (1 + i)

√
plate Z = 8 B 0 ρh

√
plate Z = 3.5 B 0 ρh

Table 10.1: Selected impedances for continuous systems, from Cremer et al [2005].

Here, A denotes the cross-section, ρ is the density, cB is the bending wave velocity, B 0 denotes
the bending stiffness per length, h is the thickness of the plate, and E is the Young’s modulus.

Structural Dynamics 10 Impedance models

summer term 2019 182
11 Solutions for selected continuous
systems
Until now, we discussed the dynamic characteristics of discrete systems, such as the MODF
systems or continuous systems described by impedances. The systems’ description is reduced
to the parameters of stiffness, damping and mass or in the are assumed to act as discrete
elements of finite number. In example 4.9, this approach is used to model a three storey
frame, where floors are assumed as discrete or lumped masses that are infinitely stiff and
columns act as massless, but elastic springs. Through this, we reduce the governing equations
to simple ordinary differential equations, that can be easily solved. However, this modeling
approach is not always applicable and the continuous distribution of stiffness, damping and
mass have to be accounted for. Furthermore, the analysis of continuous dynamic systems is
helpful to gain a better understanding of physical phenomena.
In general, only very simple models can be solved analytically. Especially for structures of
complex shape, numeric methods have to be applied, that are based on the discretization of
the system, such as the finite or boundary element method.
In order to be able to interprete the behavior of continuous systems some analytical derivations
are shown subsequently.

11.1 String, longitudinal and torsional rod

11.1.1 Equation of motion

We derive the equations of motion for linear (one dimensional) elements with constant cross
sectional values over the length analogously to the derivation of the equation of motion of
a SDOF system (comp. Chapter 5). The formulation of the dynamic equilibrium of the
differential element is given for rods subjected to longitudinal motion, under St. Venant
torsion, as well as the taut string constrained to small sag. These formulations share the same
structure. Fig. 11.1 illustrates the forces that act on the differential element with length dx.
For the string in Fig. 11.1c we assume that the horizontal force H is constant over the length
of the string.

Structural Dynamics
summer term 2019
183
 px dx mT dx p dx
Hw0
cu̇ dx cφ̇ dx cẇ dx
N N + N 0 dx MT MT + MT0 dx
H
w H
µẅ dx
u µü dx φ Iφ φ̈ dx Hw0 + Hw00 dx
dx dx dx
(a) EA-rod (b) St.-Venant torsion (c) String with small sag
Figure 11.1: Forces on differential elements: axial and torsional rod, string, with normal force N ,
Youngs modulus E , cross-sectional area A, mass per unit length µ = ρA, the horizontal
displacement u, the torsional moment MT , the de St. Venant torsional stiffness GIT ,
the polar moment of inertia Iφ , the cross-sectional rotation φ, the horizontal tension force
H (assumed constant) and the transversal displacement w.

The normal force follows to N = EAu0 , and the de St. Venant torsional moment to MT =
GIT φ0 . Hence, the following identically structured equilibrium equations are found containing
an elastic reaction force FK , a damper force FC and a inertia force FI together with external
loading Fe .

fK + fC + fI = fe (11.1)

For the longitudinal vibration of the axial rod (Fig. 11.1a) it holds:

−EAu00 + cu̇ + µü = px . (11.2)

For the torsional vibration of the torsional rod (Fig. 11.1b) it holds:

−GIT φ00 + cφ̇ + Iφ φ̈ = mT . (11.3)

For the vibration of a string with small sag (neglecting displacement u, Fig. 11.1c) it holds:

−Hw00 + cẇ + µẅ = p . (11.4)

In the following we introduce y for any of the variables describing the deformation u, φ, or
w. The stiffness related coefficients EA, GIT and H are denoted ke and the inertia related
coefficients µ and Iφ are denoted m
f. We now consider the free vibration of the above systems.

Structural Dynamics 11 Solutions for selected continuous systems

summer term 2019 184
11.1.2 Free vibration

Neglecting all damping forces and external forces the partial differential equations read

e 00 + m
−ky fÿ = 0 (11.5)

with the variable of motion y, the stiffness ke and the inertia m

f. We divide Eq. (11.5) by m
f
and obtain

−c2 y 00 + ÿ = 0 (11.6)

e . This equation is in general known as the classical one dimensional wave

where c2 = m
k
e

equation.

11.1.2.1 Solution via a separation approach

We use the separation approach

y(x,t) = Y (x)f (t) (11.7)

with the space function Y (x) and time function f (t). Inserting Eq. (11.7) into the differential
equation (11.5) results in

−c2 Y 00 (x)f (t) + Y (x)f¨(t) = 0 . (11.8)

We divide Eq. (11.8) by Y (x)f (t), such that

f¨(t) Y 00 (x)
= c2 . (11.9)
f (t) Y (x)

Eq. (11.9) must hold at every instance in time and at every point along the system, i.e., both
sides have to be equal to a constant, say λ = −ω 2 .

f¨(t) Y 00 (x)
= c2 = −ω 2 . (11.10)
f (t) Y (x)

Thus, we can write the following for both terms in Eq. (11.10) separately to obtain the two
decoupled ordinary differential equations

f¨(t) = −ω 2 f (t) (11.11)

Structural Dynamics 11 Solutions for selected continuous systems

summer term 2019 185
 c2 Y 00 (x) = −ω 2 Y (x) . (11.12)

These are eigenvalues for the continuous system. The above differential equation with respect
to time is already known from the discussion on discrete SDOF systems. Additionally the
second equation now introduces the continuous relations for the spatial domain. Under the
assumption ω 2 > 0, we find the following solutions:

f (t) = fb+ e iωt + fb− e −iωt (11.13)

Y (x) = Yb+ e ikx + Yb− e −ikx . (11.14)

with

ω2
k2 = . (11.15)
c2

Rewritten in sine and cosine terms, it follows

f (t) = f01 cos (ωt) − f02 sin (ωt) (11.16)

Y (x) = Y01 cos (kx) − Y02 sin (kx) . (11.17)

The unknown coefficients f01 , f02 , Y01 and Y02 in Eqs. (11.16) and (11.17) can be found through
the spatial boundary conditions (at both ends) and the temporal initial conditions (initial
displacement and velocity). Possible spatial boundary conditions are depicted in Fig. 11.2.
The above solution is the homogeneous solution of the continuous problem related to the
given strucutres.

Fext
x,Y x,Y x,Y
0 0
(a) fixed, Y = 0 (Dirichlet) (b) free, SY = Fext (c) spring, SY = kY
(Neumann) (Robin)
Figure 11.2: Boundary conditions

Structural Dynamics 11 Solutions for selected continuous systems

summer term 2019 186
 Example 11.1: Mode shapes of rod-shaped systems

We consider the following system according to Fig. 11.1 with length l and fixed bound-
aries at both ends

x, Y (x)
ke = EA
l
f=µ
m

The free vibrations of the depicted system are given by Eqs. (11.16). The boundary
conditions are given as

x=0: Y =0
x=l: Y = 0.

These conditions inserted into Eqs. (11.16) lead to the following equation system for
the unknowns Y01 and Y02

1 0 0
" #" # " #
Y01
= . (11.18)
cos (kl) − sin (kl) Y02 0

This equation system only has a nontrivial solution (trivial would be Y01 = Y02 = 0), if
the determinant of the coefficient matrix is zero, i.e.,

sin (kl) = 0 .

Hence, the eigenvalues kj of the free vibration follow to

jπ
kj = .
l

Under consideration of Eq. (11.15) the radial eigenfrequencies ωj are

s
jπ jπ ke
ωj = kj c = c= .
l l m
f

The eigenvectors of the system can then be obtained by inserting the result for kj
into Eq. (11.18). From this we can deduce that Y01 = 0. As the equation system is

Structural Dynamics 11 Solutions for selected continuous systems

summer term 2019 187
 underdetermined, Y02 6= 0 is arbitrary and thus

jπ
 
φj (x) = −Y02,j sin x
l
∞
jπ
 
y(x,t) = sin x (f01,j cos (ωj t) − f02,j sin (ωj t)) .
X

j=1 l

The term −Y02,j can be neglected in the summation because the unknowns in the time
function are also not yet determined. The functions φj (x) describe the mode shapes of
the vibration with ωj as corresponding radial eigenfrequency. The system can vibrate
freely in these mode shapes with the related frequencies. The unknowns f01,j and f01,j
can be determined via the given temporal initial conditions y(x,t = 0) and ẏ(x,t = 0).
Note that the initial conditions are functions of x. In general the displacements and
velocities for t = 0 can be written as
∞
nπ
 
y (x, t = 0) = sin (11.19)
X
x f01,j
j=1 l
∞
nπ
 
ẏ (x, t = 0) = − sin (11.20)
X
x ωj f02,j .
j=1 l

Depending on the (spatial)

r
boundary conditions different eigenvalues kj , radial eigen-
frequencies ωj = kj me and mode shapes φj are obtained. The following table shows
k
e

selected cases.

boundary condition char. equation eigenvalues kj mode shape φj

fixed – fixed sin (kl) = 0 jπ

l
sin (kj x)
fixed – free cos (kl) = 0 (2j − 1) 2lπ sin (kj x)
free – free sin (kl) = 0 jπ
l
cos (kj x)

Table 11.1: Eigenvalues kj and mode shapes φj from Petersen [1996]

Consider the solutions to the Eqs. (11.13) and (11.14). The total solution w(x,t) is given as
the product of the two separate solutions as

w(t) = Y (x)f (t)

= f+ Y+ e iωt e ikx + f− Y+ e −iωt e ikx + f+ Y− e iωt e−ikx + f− Y− e −iωt e −ikx

Structural Dynamics 11 Solutions for selected continuous systems

summer term 2019 188
 = f+ Y+ e i(kx+ωt) + f− Y+ e i(kx−ωt) + f+ Y− e −i(kx−ωt) + f− Y− e −i(kx+ωt) . (11.21)

Inserting ω = kc, we obtain

h i h i
w(t) = f+ Y+ e ik(x+ct) + f− Y− e −ik(x+ct) + f− Y+ e ik(x−ct) + f+ Y− e −ik(x−ct) . (11.22)
| {z } | {z }
wave travelling in negative x-direction wave travelling in positive x-direction

The arguments of the full solution x− = x + ct and x+ = x − ct describe waves traveling in

the negative and positive x-direction, respectively, with the wave speed
s
ω ke
c= = . (11.23)
k m
f

The wave speed c only depends on the stiffness and inertia terms ke and m
f, respectively, and is
thus a mere system parameter. It is independent of the frequency or the initial conditions.
We now represent Eq. (11.22) using sine/cosine functions

w(t) = 2 [Re(f+ Y+ ) cos(kx + ωt) + Im(f+ Y+ ) sin(kx + ωt)] +

+ 2 [Re(f− Y+ ) cos(kx − ωt) + Im(f− Y+ ) sin(kx − ωt)] . (11.24)

We then denote amplitudes of the solution that travel in negative x-direction (argument k(x +
ωt)) by an uppercase −, e.g., w− , and the ones that travel in positive x-direction (argument
k(x − ωt)) by an uppercase +, e.g., w+ :

h i
w(t) = w01
−
cos(kx + ωt) + w02
−
sin(kx + ωt) +
h i
+ w01
+
cos(kx − ωt) + w02
+
sin(kx − ωt) . (11.25)

Applying Eqs. (2.18) and (2.19), we obtain

h i
w(t) = w01
+
(cos(kx) cos(ωt) − sin(kx) sin(ωt)) + w02
+
(cos(kx) sin(ωt) + sin(kx) cos(ωt)) +
h i
+ w01
−
(cos(kx) cos(ωt) + sin(kx) sin(ωt)) + w02
−
(cos(kx) sin(ωt) − sin(kx) cos(ωt)) .
(11.26)

Consider now the special case w02

−
= w02
+
and w01
−
= w01
+
. Eq. (11.26) then becomes
 
w(t) = 2 cos(kx) w01
+
cos(ωt) + w02
+
sin(ωt) (11.27)

Structural Dynamics 11 Solutions for selected continuous systems

summer term 2019 189
This special case describes a standing wave with harmonically varying amplitude.

11.1.2.2 Direct solution via a wave approach

In the preceeding section, the solution to the differential equation (11.5), which reads

∂ 2y ∂ 2y
−c2 + 2 =0 (11.28)
∂x2 ∂t

was derived. It can be interpreted as a sum of two waves that travel in opposite directions.
We now discuss a different solution approch, that was introduced by d’Alembert. To this end,
we introduce the new variables

x+ = x − ct (11.29)
x− = x + ct . (11.30)

Applying the chain rule of differentiation we can write the partial differential operators ∂
∂t
and
∂
∂x
as

∂ ∂x+ ∂ ∂x− ∂ ∂ ∂
= +
+ −
=c + −c − (11.31)
∂t ∂t ∂x ∂t ∂x ∂x ∂x
+ −
∂ ∂x ∂ ∂x ∂ ∂ ∂
= +
+ −
= +
+ −, (11.32)
∂x ∂x ∂x ∂x ∂x ∂x ∂x

and furthermore
!
∂2 ∂2 ∂2 ∂2
= c 2
− 2 + (11.33)
∂t2 ∂x+ 2 ∂x+ ∂x− ∂x− 2
!
∂2 ∂2 ∂2 ∂2
= + 2 + . (11.34)
∂x2 ∂x+ 2 ∂x+ ∂x− ∂x− 2

Eq. (11.28) can then be written as

2 ∂ y(x ,x )
+ −
∂ 2y ∂ 2y 2
−c2 + = −4c . (11.35)
∂x2 ∂t2 ∂x+ ∂x−

A solution to Eq. (11.35) is given by

y(x+ ,x− ) = f1 (x+ ) + f2 (x− ) = f1 (x − ct) + f2 (x + ct) . (11.36)

Structural Dynamics 11 Solutions for selected continuous systems

summer term 2019 190
This is the sum of two functions f1 and f2 , that move in opposite directions (f1 in the positive,
f2 in the negative x-direction) with the velocity c.

y(x,t)

f1 (x,t = 0) f1 (x,t = ∆t)

c ∆t x+ = x − c∆t

Figure 11.3: Wave solution, f1 (x − ct), Eq. (11.36).

t
t3

t2

y(x,t)
t1

x
x1 x2 x3

Figure 11.4: Wave traveling in the positive x-direction.

This approach solves for arbitrary functions and f1 and f2 Eq. (11.5):
!
d 2 f1 d 2 f2 d 2 f1 d 2 f2
ÿ = c 2
+ y 00 = + (11.37)
dx2+ dx2− dx2+ dx2−

The unknown functions f1 (x+ ) and f2 (x− ) have to satisfy displacement and velocity initial

Structural Dynamics 11 Solutions for selected continuous systems

summer term 2019 191
conditions respecting given boundary conditions. The initial conditions are given by

y(x,t = 0) = y0 (x) = f1 (x) + f2 (x) (11.38)

ẏ(x,t = 0) = v0 (x) = f˙1 (x − ct) + f˙2 (x + ct) = −cf10 (x) + cf20 (x). (11.39)

Eq. (11.39) is rewritten as

1
−f10 (x) + f20 (x) = v0 (x) (11.40)
c

Integrating Eq. (11.40) leads to

1Z x
−f1 (x) + f2 (x) = v0 (ξ) dξ − f1 (0) + f2 (0) (11.41)
c 0

We now add Eq. (11.38) to Eq. (11.40) and obtain

1 Zx y0 (x) f1 (0) f2 (0)

f2 (x) = v0 (ξ)dξ + − + . (11.42)
2c 0 2 2 2

Analogously, we substract Eq. (11.40) from Eq. (11.38). Thus, it holds

1 Zx y0 (x) f1 (0) f2 (0)

f1 (x) = − v0 (ξ)dξ + + − . (11.43)
2c 0 2 2 2

Inserting Eqs. (11.42) and (11.43) into Eq. (11.36), the displacement y(x,t) reads

y(x,t) = f1 (x − ct) + f2 (x + ct)

y0 (x − ct) + y0 (x + ct) 1 x+ct Z x−ct
Z 
= + v0 (ξ) dξ − v0 (ξ) dξ =
2 2c 0 0

y0 (x − ct) + y0 (x + ct) 1 Z x+ct

= + v0 (ξ)ξ . (11.44)
2 2c x−ct

In cases, where the initial velocity is equal to zero, i.e., v0 (x) = 0, the initial displacement is
split into two oppositely traveling wave packets with half of the initial amplitude.

Structural Dynamics 11 Solutions for selected continuous systems

summer term 2019 192
 Example 11.2: Wave approach

The depicted system with given initial conditions shall be solved through a wave ap-
proach according to Eq. (11.36).

y0 (x) v0 (x) = 0

x x

ymax
l l l l

Figure 11.5: Initial conditions, e.g. deflection of string.

As the initial velocity is v0 (x) = 0, the displacement y(x,t) is given as

y0 (x − ct) + y0 (x + ct)
y(x,t) = .
2

For t = 0, it holds

y0 (x) + y0 (x)
y(x,0) = .
2

This solution corresponds to two oppositely traveling waves with the individual ampli-
tudes ymax
2
. In Fig. 11.6 both waves are depicted for different time instances t0 , t1 , t2 ,
t3 .

Structural Dynamics 11 Solutions for selected continuous systems

summer term 2019 193
 t = t0 so that ct0 = 0

ymax
2

ymax
2

x
l l
ct0 = 0
t = t1 so that 0 < ct1 < l

ymax ymax
2 x 2
l l
ct1 ct1
t = t2 so that ct2 = l

ymax x ymax
2 2
l l
ct2 ct2

t = t3 so that ct3 = 2l

x
ymax ymax
2 2
l l
ct3 ct3

Figure 11.6: Oppositely travelling waves at times t0 , t1 , t2 , t3 .

Additionaly to the temporal initial conditions also the spatial boundary conditions have
to be satisfied (Fig. 11.7).

Structural Dynamics 11 Solutions for selected continuous systems

summer term 2019 194
 x = −l y(−l,t) = 0 x=l y(l,t) = 0

x = −l y 0 (−l,t) = 0 x=l y 0 (l,t) = 0

Figure 11.7: Spatial boundary conditions at the supports.

The boundary conditions at the bearings can be satisfied through oppositely traveling
waves. The displacement at the respective side is given by the sum of the both waves.
For the left side, the boundary condition is exemplarily given in Fig. 11.8.
The illustration is valid for the time t = t0 = 0. Here, a wave is traveling to the left
with wave speed c with known shape and amplitude. A second (reflected) wave with
the same wave speed c but unknown amplitude AR , that travels to the right and is
shifted to the left by 2l, is added.

t = t0 so that ct0 = 0

AR fR (x + 2l − ct0 ) Af (x + ct0 )
l l l l
ct0 ct0

Figure 11.8: Boundary condition at the left support

Fig. 11.9 illustrates the situation for a later point in time t = t1 .

Structural Dynamics 11 Solutions for selected continuous systems

summer term 2019 195
 t = t1 so that 0 < ct1 < l

AR fR (x + 2l − ct1 )Af (x + ct1 )

l l l l
ct1 ct1

Figure 11.9: Situation at the left support for time t = t1

The displacement at the boundary is given by the superposition of both waves

y(x,t) = AR fR (x + 2l − ct) + Af (x + ct).

The ratio of the amplitudes AR and A follows from the boundary conditions. For a
support, that is fixed in x-direction at the location x = −l, the condition y(−l,t) = 0
results in

y(−l,t) = AR fR (l − ct) + Af (−l + ct) = 0.

Hence,

fR (l − ct) = f (−l + ct),

and thus,

AR = −A.

A boundary that is fixed in x-direction, is hence represented by two oppositely traveling,

mirrored waves with opposite amplitudes.
In the case of a free end at the position x = −l, the boundary condition of vanishing
forces at the end, i.e., y 0 (−l,t) = 0, leads to the amplitude AR = A.

Structural Dynamics 11 Solutions for selected continuous systems

summer term 2019 196
11.2 Euler-Bernoulli beam

11.2.1 Equation of motion

We derive the equation of motion of the Euler-Bernoulli beam via the equilibrium of forces
on a differential element.

p dx
µẅdx cẇdx

M M + M 0 dx
w
Q dx Q + Q0 dx

Figure 11.10: Equilibrium on a differential element of the Euler -Bernoulli beam.

We assume the material properties to be independent of time. The dynamic equilibrium for
the vibration of a beam with the bending stiffness EI and the mass per unit length µ, and
under the assumption of Euler-Bernoulli theory (i.e., neglecting rotational inertia and shear
deformation), reads

−Q0 (x,t) + c(x)ẇ(x,t) + µ(x)ẅ(x,t) − p(x,t) = 0 . (11.45)

We assume the material and cross-sectional properties to be time-independent. The shear

force is Q(x,t) = (−EI(x)w00 (x,t))0 . Thus, we obtain
00
(EI(x)w00 (x,t)) + c(x)ẇ(x,t) + µ(x)ẅ(x,t) − p(x,t) = 0 . (11.46)

11.2.2 Free vibration

Neglecting the external loading as well as damping forces, and assuming a constant bending
stiffness EI and mass density µ, the partial differential equation for the free vibration is
obtained.

EIw0000 (x,t) + µẅ(x,t) = 0 (11.47)

Structural Dynamics 11 Solutions for selected continuous systems

summer term 2019 197
11.2.2.1 Solution via a separation approach

The partial differential equation (11.47) of the free vibration can be solved by a separation
approach with space function W (x) and time function f (t) 1 .

w(x,t) = W (x)f (t) . (11.48)

We insert Eq. (11.48) into Eq. (11.47) and obtain

EIW 0000 (x)f (t) + µW (x)f¨(t) = 0 . (11.49)

Dividing Eq. (11.49) by µ and W (x)f (t), it follows

f¨(t) EI W 0000 (x)

=− . (11.50)
f (t) µ W (x)

Eq. (11.50) must hold at every instance in time and space, i.e., both sides have to be equal to
a constant, also called separation constant, say λ = −ω 2 . This at first not obvious choice will
become apparent when discussing the solution characteristics.

f¨(t) EI W 0000 (x)

=− = −ω 2 . (11.51)
f (t) µ W (x)

Thus, we can write the following for both terms in Eq. (11.51) separately to obtain the two
decoupled ordinary homogeneous differential equations

f¨(t) + ω 2 f (t) = 0 (11.52)

W 0000 (x) − k 4 W (x) = 0 (11.53)

with
µ 2
k4 = ω . (11.54)
EI

We find the following solutions under the assumption ω 2 > 0.

W (x) = K
c e kx + K
x1
c e −kx + K
x2
c e ikx + K
x3
c e −ikx
x4 (11.55)
f (t) = K
c e iωt + K
t1
c e −iωt .
t2 (11.56)

1
The solution to the free vibration problem of the unloaded beam with the partial differential equation
EIw0000 + µẅ = 0 is also called the homogeneous solution.

Structural Dynamics 11 Solutions for selected continuous systems

summer term 2019 198
The solution in time now corresponds to oscillating waves as the exponent of the exponential
function is complex. Now, reconsider our choice for λ in Eq. (11.51). If the constant λ was
assumed to be ω 2 for ω 2 > 0 we would have obtained solutions that correspond to decaying
and growing exponential functions, which are not admissible in our problems. Eqs. (11.55)
and (11.56) can alternatively be written as

W (x) = Cx1 cosh(kx) + Cx2 sinh(kx) + Cx3 cos(kx) − Cx4 sin(kx) (11.57)
f (t) = Ct1 cos(ωt) − Ct2 sin(ωt) . (11.58)

Thus, the solution contains six unknown coefficients in Eqs. (11.55) and (11.56), or Eqs. (11.57)
and (11.58), respectively. These can be found by inserting the spatial boundary conditions
and the temporal initial conditions. Possible spatial boundary conditions are depicted in
Fig. 11.11.

hinged support: M = 0, w = 0

clamped support: w = 0, φ = 0

free end: M = 0, Q = 0

dilational or rotational spring: Q = kw or M = −kφ φ

Figure 11.11: Spatial boundary conditions.

Through the relations w(x,t) = W (x)f (t), w0 (x,t) = W 0 (x)f (t), M (x,t) = −EIW 00 (x)f (t),
and Q(x,t) = −EIW 000 (x)f (t) all boundary conditions can be represented by derivatives of the
displacement. At each boundary of a beam two boundary conditions can be found, i.e., at each
boundary two equations for the unknown coefficients are given. This results in a 4 × 4 system
of equations for the unknown coefficients Cxi . A solution is found by excluding the trivial
solution and setting the determinant of the equation system to zero. As the solution for the
time function f (t) is decoupled, the unknowns Cti are found through the initial displacement
and velocities for t = 0.

Structural Dynamics 11 Solutions for selected continuous systems

summer term 2019 199
 Example 11.3: Mode shapes of the Euler-Bernoulli beam

In order to find the unknowns of the solution to the Euler-Bernoulli beam vibration,
the derivatives of the space function W (x) with respect to x and time function f (t)
with respect to t are needed. Under consideration of Eqs. (11.57) and (11.58) it follows.

W 0 (x) = k [Cx1 sinh(kx) + Cx2 cosh(kx) − Cx3 sin(kx) − Cx4 cos(kx)]

W 00 (x) = k 2 [Cx1 cosh(kx) + Cx2 sinh(kx) − Cx3 cos(kx) + Cx4 sin(kx)]
W 000 (x) = k 3 [Cx1 sinh(kx) + Cx2 cosh(kx) + Cx3 sin(kx) + Cx4 cos(kx)]
f˙(t) = ω [−Ct1 sin(ωt) − Ct2 cos(ωt)] .

The unknowns Cx1 to Cx4 are determined from the spatial boundary conditions. In the
following this will be performed exemplarily for the hinged Euler-Bernoulli beam.

µ = const., EI = const.
w(x,t) x
l

Figure 11.12: Hinged Euler-Bernoulli beam.

Since cos (ωt) and sin (ωt) change periodically, it follows, that w(x,t) and M (x,t =
−EIW 00 (x,t) have to be equal to zero for all times t. Thus,

x = 0 : W (x = 0) = 0 W 00 (x = 0) = 0 (11.59)
x = l : W (x = l) = 0 W 00 (x = l) = 0 . (11.60)

These four equations lead to the following equation system for the unknowns Cx1 to
Cx4 in matrix notation

1 0 1 0 Cx1 0
    
 k 2
0 −k 2
0  C  0
= .
   x2   
 cosh(kl) sinh(kl) cos(kl) − sin(kl)  Cx3  0
 

k 2 cosh(kl) k 2 sinh(kl) −k 2 cos(kl) k 2 sin(kl) Cx4 0

From the first two lines, it immediately follows Cx1 = Cx3 = 0. Hence, we write the

Structural Dynamics 11 Solutions for selected continuous systems

summer term 2019 200
 reduced equation system

sinh(kl) − sin(kl) 0
" # ! !
Cx2
= . (11.61)
sinh(kl) sin(kl) Cx4 0

This homogeneous equation system only permits a non-trivial solution, if its determi-
nant is zero, thus

sinh(kl) − sin(kl)
" #
det = 2 sinh(kl) sin(kl) = 0 .
sinh(kl) sin(kl)

As sinh (k l) is only equal to zero for kl = 0, which is another trivial solutiona , the
above characteristic equation for kl reduces to

sin(kl) = 0 .

Thus, we obtain the eigenvalues of the system

kj l = jπ (11.62)

for j ∈ N+ . The eigenvalue problem thus has an infinite number of solutions. Under
consideration of Eq. (11.54), the eigenfrequencies for which non-trivial solutions are
obtained, follows to
s 2 s
EI jπ EI

ωj = kj2 =± . (11.63)
µ l µ

These eigenfrequencies are associated with mode shapes φj . They follow from the
solution of Eq. (11.61) after inserting the previously obtained values for k.

nπ
 
φj (x) = Cx4,j sin x .
l

The beam vibrates in the mode shapes φj (x) with the corresponding eigenfrequencies
ωj . The complete solution therefore follows as the addition of the individual solutions.

wj (x,t) = φj (x)fj (t)

∞ 
jπ jπ
    
w(x,t) = Ct1,j sin x cos (ωj t) − Ct2,j sin x sin (ωj t)
X

j=1 l l

Structural Dynamics 11 Solutions for selected continuous systems

summer term 2019 201
 ∞
jπ jπ
     
ẇ(x,t) = ωj −Ct1,j sin x sin (ωj t) − Ct2,j sin x cos (ωj t)
X

j=1 l l

The coefficients Ct1,j and Ct2,j then depend on the initial conditions.
Depending on the spatial boundary conditions, different eigenvalues kj , eigenfrequencies
ωj and mode shapes φj (x) result. Selected results are given in the following table.

Table 11.2: Eigenfrequencies of the Euler-Bernoulli beam for different boundary conditions
from Petersen [1996]

system λj Aj Eigenform Φj
 
λj = j π − sin λj xl
   
sinh(λj )−sin(λj ) sin λj xl − sinh λj x
+
λj = 2j+1
2
π cosh(λj )−cos(λj )
  l  
+Aj [cosh λj xl − cos λj x
l
]
   
sinh(λj )+sin(λj ) sin λj xl − sinh λj xl +
λj = 2j−1
2
π cosh(λj )+cos(λj )
   
+Aj [cosh λj xl − cos λj xl ]
   
sinh(λj )−sin(λj ) sin λj xl − sinh λj xl +
λj = 4j+1
4
π cosh(λj )−cos(λj )
   
+Aj [cosh λj xl − cos λj xl ]

a
This trivial solution corresponds to the static limit case, for which another solution is valid.

11.2.2.2 Solution via a wave approach

The solutions given in Eqs. (11.55) and (??) can be interpreted by waves travelling in both
directions and decaying near fields.
A multiplication of K c e ikx and K
x3
c e −ikx with f (t) = K
x4
c e iωt + K
t1
c e −iωt results in functions
t2
±ik(x− ω t) ±ik(x+ ω t)
that depend on e k or e k . The argument x+ = x − ωk t is linked with a wave
travelling in the positive x-direction, the argument x− = x + ωk t with a wave travelling in the
negative x-direction with the velocity of propagation cB = ωk . Using Eq. (11.54), we find the
bending wave velocity
s s
ω EI √ EI 2
cB = = 4
ω i.e. ω= k . (11.64)
k µ µ

Structural Dynamics 11 Solutions for selected continuous systems

summer term 2019 202
In contrast to the wave propagation velocity from the solution of the differential equation of
2nd order (comp. Eq. (??), this so-called bending wave phase velocity only has the meaning
of a propagation velocity for a harmonic (sinusoidal or cosinusoidal) waves. Waves, which are
composed of different sine or cosine parts, distort themselves in the propagation, since portions
with higher frequencies ω (or smaller wave numbers k), propagate faster than lower frequency
components (with larger wavenumbers). Thereby the propagation on beams (differential equa-
tion of 4-th order) differs from the propagation the on rod-shaped systems described by the
differential equation of 2nd order. The frequency dependence of the bending wave is called
dispersion.
It has special significance for the sound radiation of structures. For frequencies, for which
the bending wave phase velocity cB takes values below the sound velocity of an adjacent fluid
(air or water) cf luid, in the stationary case and with infinitely extended structures, the sound
radiation is zero, because the wave pattern on the structure does not “couple” to a wave
pattern in the adjacent fluid (see Fig. 11.13).

c Fluid

λLuft ϑ

x
λLuft
λB =
sin(ϑ) z

Figure 11.13: Spuranpassung, Biegewellenlänge λB > Wellenlänge Fluid λf luid

The frequency, for which the bending wave phase velocity cB is equal to the sound wave
velocity cf luid (speed of sound) of the adjacent fluid, is called coincidence frequency.
The other parts of the solution in Eq. (11.55) and (??) that result from the multiplication of the
terms Kc e kx and K
x1
c e −kx with f (t) = K
x2
c e iωt + K
t1
c e −iωt can be interpreted as harmonically
t2
oscillating near fields. The arise at the boundaries of the beam and decay exponentially
from there.

Structural Dynamics 11 Solutions for selected continuous systems

summer term 2019 203
 Example 11.4: Semi-infinite beam for harmonic boundary moment

We consider the following semi-infinite Euler-Bernoulli beam with a harmonically vary-

ing moment at the boundary x = 0.

M0 cos Ωt
EI, µ

Figure 11.14: Beam under harmonic boundary moment

The general solution reads:

w(x,t) =K
c e kx+iωt + K
1
c e −kx+iωt + K
2
c e i(kx+ωt) + K
3
c e −i(kx−ωt) +
4

+K
c e kx−iωt + K
5
c e −kx−iωt + K
6
c e i(kx−ωt) + K
7
c e −i(kx+ωt)
8

Since the domain of the beam extends to infinity in the positive x-direction, we can
assume that no waves propagate from the infinite towards the origin, and the positive
exponential near field cannot exist, since they would lead to infinitely high amplitudes
for x → ∞. These conditions are also called the Sommerfeld radiation conditions. We
furthermore insert the relationship ω = kcB , and obtain
 
w(x,t) =
ce  B t) + K
c e −k(x−icB t) + K
c e  B t) + K
c e −ik(x−cB t) +
k(x+ic  ik(x+c 
K 1 2 3 4

+
ce  B t) + K
c e −k(x+icB t) + K
c e ik(x−cB t) + (
c(e(

k(x−ic −ik(x+c((
(( B t)
K 5 6 7 K 8

where we skipped the above mentioned corresponding solution terms. The solution is
thus given by a wave traveling in positive x-direction (parts with K
c and K
4
c ) and a
7
decaying near-field (parts with K
c and K
2
c ). Considering the moment and displacement
6
boundary condition at x = 0:

EIw00 (0,t) = −M0 cos (Ωt)

w(0,t) = 0

Structural Dynamics 11 Solutions for selected continuous systems

summer term 2019 204
 the solution is found by

M0 h    i
w(x,t) = cos k(x − c B t) − e −kx
cos k c B t . (11.65)
2EI k 2

It is easy to show, that this solution satisfies the homogeneous PDE as well as the
boundary conditions. This example considered a time harmonic boundary moment.
If, for example, the boundary moment was modeled as a Dirac impulse, it could be
represented as the superposition of single cosine terms by a Fourier transformation.

The dispersion of the beam wave leads to the fact, that vibrations with higher frequency
propagate faster than the ones with lower frequency. If such a beam is excited with a Dirac
impulse (composed of all frequency components with equal amplitude), at a finite distance
from the excitation point the signal “shape” will be altered.

11.2.3 Displacement approaches - spectral elements based on analytic

solutions

(For information only, treated in “Ergänzungskurs Technische Mechanik”.)

11.2.3.1 Beams without loads and harmonic excitation at the boundaries

The free vibration of a system composed from individual beams or impedances at nodes can
be determined from the solution of the individual elements (e.g., columns and bars) under the
assumption of a harmonic motion. The procedure corresponds to the displacement approaches
in statics. Every beam element is described by solution to Eqs. (11.57) and (11.58).

W (x) = Φ (x) α (11.66)

with

Φ (x) = {cosh(kx), sinh(kx), cos(kx), sin(kx)} (11.67)

and

αT = {Cx1 , Cx2 , Cx3 , Cx4 } (11.68)

Note that the sign of the sine function in Eq. (11.67) differs from the convention introduced in
section 2.2. This is to keep consistency with literature on the topic. The unknown coefficients

Structural Dynamics 11 Solutions for selected continuous systems

summer term 2019 205
Cx1 to Cx4 can be determined via boundary and transition conditions. Corresponding to a
finite element formulation, in a first step, the displacement field W (x) is represented through
the ansatz functions Φ (x) and the nodal displacements we (comp. Fig. 11.15).

wTe = {wi , φi , wk , φk } (11.69)

The relationship between the coefficients α and the nodal displacements we is given by

we = Aα (11.70)

where we find through inserting

1 0 1 0
 
 0 k 0 k 
[A] =  (11.71)
 
 cosh(kle ) sinh(kle ) cos(kle ) sin(kle ) 


k sinh(kle ) k cosh(kle ) −k sin(kle ) k cos(kle )

Hence we can express the unknwon coefficents α by the nodal displacements we through the
inverse relation

α = A−1 we (11.72)

Inserting Eq. (11.72) into Eq. (11.66), we obtain

W (x) = ΦA−1 we = Nwe (11.73)

with the vector of shape functions N (the so called spectral ansatz).

In order to obtain the element stiffness matrix the principle of virtual work is applied, analo-
gously to section 12.1.1. Considering the appearing integrals that have to be solved under this
approach, in the present case it is advantageous, to determine the nodal forces directly from
the corresponding derivatives. With the vector
" #
w000
= (11.74)
−w00

and the Matrix of stiffnesses

EI 0
" #
D= (11.75)
0 EI

Structural Dynamics 11 Solutions for selected continuous systems

summer term 2019 206
 the internal forces
" #
−Q
σ= (11.76)
M

can be represented as

σ = D (11.77)

i le k
wi wk
µ ω2 4
λ4 = EI
l
+Mik
φi +Mki
+Vik φk
+Vki

Figure 11.15: Beam element, sectional forces/moments and degrees of freedom

The sectional forces at the boundaries of the beam

qT = {Vik , Mik , Vki , Mki } (11.78)

result from inserting the coordinates x = 0 and x = le , respectively and considering the
definition of positive directions in Fig. 11.15 to

0 k 0 −k
 

0 1 0
( )
EIi  −1 
q= = EIk 2   α = Sα.

−EIk −k sinh(kle ) −k cosh(kle ) −k sin(kle ) k cos(kle ) 


cosh(kle ) sinh(kle ) − cos(kle ) − sin(kle )

(11.79)

With Eq. (11.72) we obtain

q = Sα = SA−1 we = Ke we , (11.80)

with the element stiffness matrix Ke . Introducing the auxiliary functions Fi (λ) introduced

Structural Dynamics 11 Solutions for selected continuous systems

summer term 2019 207
by Kolousek [1973], Ke reads

1 
1 1 1 
 l2 F6 (λ) le
F4 (λ) − 2 F5 (λ)
le le
F3 (λ) 
 e
 1 1


F (λ) F (λ) − F (λ) F (λ)
 
4 2 3 1
EI 
 
[Ke ] =  le le
(11.81)

le − F (λ) − 1 F (λ) 1 F (λ) − 1 F (λ)
 1


 2 5 3 2 6 4
 le l l l

e e e 
 1 1
 
F3 (λ) F1 (λ) − F4 (λ) F2 (λ)

le le

with λ = kle . The auxiliary functions F1 (λ) to F6 (λ) read:

λ (sinh λ − sin λ)
F1 (λ) =
N1 (λ)
λ (cosh λ sin λ − sinh λ cos λ)
F2 (λ) =
N1 (λ)
λ2 (cosh λ − cos λ)
F3 (λ) =
N1 (λ)
λ2 sinh λ sin λ
F4 (λ) =
N1 (λ)
λ3 (sinh λ + sin λ)
F5 (λ) =
N1 (λ)
λ3 (cosh λ sin λ + sinh λ cos λ)
F6 (λ) = .
N1 (λ)

with

N1 (λ) = 1 − cosh λ cos λ

Further derivation of element stiffness matrices can be, e.g., found in Petersen [1996]; Grund-
mann et al [1983].

Structural Dynamics 11 Solutions for selected continuous systems

summer term 2019 208
 Example 11.5: Free vibration of a frame

We consider the following one-storey frame with flexible but longitudinally rigid bars.
The system is depicted in Fig. 11.16 together with the identified degrees of freedom and
the corresponding individual displacement curves for each degree of freedom.
u
1 2
φ1 φ2

0 3

u=0 u=0 u 6= 0

φ1 6= 0 φ2 = 0 φ1 = 0 φ2 6= 0 φ1 = 0 φ2 = 0

Figure 11.16: System and degrees of freedom

Now, the three states 1 to 3 have to be superposed, such that all nodal forces are in
equilibrium.

a) Free body diagram at node 1 (only nodal moments are depicted)

M12 due to φ1 und φ2

M10 due to φ1 und u

b) Free body diagram at node 2 (only nodal moments are depicted)

M21 due to φ1 , φ2

M23 due to φ2 und u

Structural Dynamics 11 Solutions for selected continuous systems

summer term 2019 209
 c) Free body diagram of the bar (only nodal forces are depicted; the equilibrium
would have to be dealt with separately for nodes 1 and 2 if the bar would not be
modeled as longitudinally rigid)

−µ12 l12 ω 2 u

V10 due to φ1 and u V23 due to φ2 and u

The nodal forces are expressed in terms of λ or ω, respectively, via the auxiliary func-
tions F (λ) (see Eq. (11.81)). The equilibrium conditions have to be expressed at the
basis of those nodal forces. Then, we obtain a system of equations for the unknown
nodal displacement values.

11.2.3.2 Loaded beam elements - particular solution

In the case of harmonically excited elements, the sectional forces and moments for a clamped
beam can be calculated as follows:

p(x,t)

Figure 11.17: Assessment of the sectional forces and moments

The equation of motion

EIw0000 + µẅ = p+ (x) e iΩt + p− (x) e −iΩt (11.82)

is solved by the approach

w (x) = W+ (x) e iΩt + W− (x) e iΩt (11.83)

We now assume Im(p+ ) = 0 (then also Im(p− ) = 0), such that the load p(x,t) = p0 cos(Ωt)
oscillates in phase and with constant amplitude for all x. This supposes that all motions, that
result due to the nodal displacements and rotations, vary harmonically with the excitation
frequency Ω. By that, the arguments λ of the functions Fj (λ) are known. Since the system

Structural Dynamics 11 Solutions for selected continuous systems

summer term 2019 210
is modeled without damping, the phase shift between the load and the displacement can only
be 0 or π. Both phase shifts are covered by the cos function via the sign.
Thus, with w(x,t) = W (x) cos (Ωt), the differential equation (11.82) reads

EIW 0000 (x) − µΩ2 W (x) = p0 (11.84)

The partial differential equation is reduced to a ordinary differential equation in the variable
x. The solution is composed of a homogeneous and a particular part.

W = Wh + Wp (11.85)

The homogeneous solution (part without load) is described by Eq. (11.73) For the particular
solution to the right hand side p(x) = p0 = const., we use the approach

p0 p0
Wp (x) = − 2
=− (11.86)
µΩ EIk 4

The full solution Wh + Wp has to satisfy the boundary conditions wTe = {wi , φi , wk , φk } = 0.
The corresponding system of equations reads

 C   p0 
1 0 1 0


 x1 
 
 µΩ2 


 
 
 

  
 
 

0 0 0
    

 k k  
 C
  x2 












= (11.87)
 
cosh(k l) sinh(k l) cos(k l) sin(k l) 
 
    p0 
Cx3 
   
 

  
 
 µΩ 2

k sinh(k l) k cosh(k l) −k sin(k l) k cos(k l) 
 
 
 
 

 
 
 

0
   
Cx4
   

The solution gives the constants Cx1 to Cx4 . From the second derivative of the displacement
function, we find the moment M = −EIw00 under consideration of the sign convention and
k = lλe as

p0 l2 cosh λ cos λ − sinh λ sin λ

Mi0 = −Mk0 = (11.88)
λ2 cosh λ cos λ − 1

Analogously the nodal forces can be determined. Further loading conditions can be considered
in a similar manner (see Grundmann et al [1983]; Petersen [1996]).

Structural Dynamics 11 Solutions for selected continuous systems

summer term 2019 211
11.3 Extended beam theories

The dispersion equation (11.64) theoretically leads to arbitrarily high bending wave phase
velocities. Physically the maximum propagation velocity in the elastic continuum is given by
the compresssion wave velocity cp (comp. Continuum Mechanics, Soil Dynamics). This is not
a contradiction, since for short wave lengths and corresponding high wave numbers k the as-
sumptions of the Euler-Bernoulli beam theory are not valid anymore. Important assumptions
herein are the neglection of rotatory inertia and shear deformation.
In certain bounds the range of applicability can then be extended towards the Timoshenko
beam theory. For shorter wave lengths, additional terms are then considered. Then, depending
on these additional terms describing the rotatory inertia per unit length iθ (for the Euler-
Bernoulli beam theory iθ = 0) and shear stiffness GAG (for the Euler-Bernoulli beam theory
GAG → ∞), modified bending wave phase velocities are obtained (see Petersen [1996]; Cremer
et al [2005]). Based on this—following Petersen [1996]; Cremer et al [2005]—the general relation
between the wave number k und the frequency ω for a beam with uniform mass and stiffness
distribution over its length, can be be written as

" ! # !
H kφ EI kφ
 
1+ EIk 4 + 1 + bett H + kwbett + kφbett k 2 + 1 + bett kwbett −
GAG GAG GAG GAG
" ! #
H EI kφ kw µ iθ 4
 
− 1+ iθ + µ k 2 + 1 + bett µ + bett iθ ω 2 + ω = 0 (11.89)
GAG GAG GAG GAG GAG

with the longitudinal force H (positive in tension), the rotational bedding kφbett and the transla-
tional bedding kwbett describing the attached subgrade. Neglecting the term GA µ iθ
G
ω 4 , a solution
is directly found.
v
kφbett kwbett kφbett kφbett kwbett
u    
u 1 + 1 + GA 2 + GA k 2 + EI k 2 + 1 + GA
s H
EI
u
EI k EI k4
ω= k2u h G  G
i k k
G
(11.90)
µ 1+ 1+ H iθ EI
+ k +
φ w
+
bett iθ
t
2 bett
GAG µ GAG GAG GAG µ

Hence, for the single span beam the changes in the eigenfrequency compared to the Euler-
Bernoulli beam theory can be easily estimated by inserting the wave numbers k = kn (comp.
Eq. (11.62)). For the case of a beam hinged at both ends it follows

nπ
kn = . (11.91)
l

Often, merely the influence of a normal force or a elastic bedding under the assumption

Structural Dynamics 11 Solutions for selected continuous systems

summer term 2019 212
GAG → ∞ is of interest. For this case we obtain from Eq. (11.90)
v
k kwbett
u 1 + EIk2 + EIk 2 +
s u
H φbett
EI 2 u
ω= k t EIk4
(11.92)
µ 1+ µk
iθ 2

The eigenfrequencies of a single span beam subject to normal force, bedding, shear deformation
and rotatory inertia are summarized in Tab. 11.3 (comp. Grundmann et al [1983]).

EI, GA∗ , µ, iθ H

k
x l

EI: bending stiffness

GA∗ : shear stiffness GAκ
µ: mass per unit length
iθ : rotatory inertia per unit length i2 µ
q
n2 π 2
ω0 = l2
EI
µ

EI GA µ iθ H k radial eigenfrequncy

6= 0 ∞ 6= 0 6= 0 0 0 q ω0
n2 π 2
1+i2
l2

6= 0 6= 0 6= 0 0 0 0 q ω0
2 2
1+i2 n 2π E G
κ
q l
n2 π 2
6= 0 ∞ 6= 0 0 6= 0 0 ω0 1+ H
PK
with PK = l2
EI
s
nπ H
0 0 6= 0 0 6= 0 0
lq µ
6= 0 ∞ 6= 0 0 0 6 0 ω0 1 +
= k l4
EI n4 π 4
!
ω0 µ2 i2
6= 0 6= 0 6= 0 6= 0 0 0 q for κ ω4 ∼
=0
1 + i2 n2 π 2
(1 + E
κ) GA
l2 G

Table 11.3: Radial eigenfrequncies of the single span beam with normal force, bedding, shear defor-
mation and rotatory inertia.

In Kreutz [2013], the beam theory is extended for higher cross-section deformation. There—
depending on the frequency—dispersion diagrams for different wave numbers are generated.
Fig. 11.18 illustrates the results for a beam with rectangular cross-section. The limits of the
corresponding theories can be identified compared to the finite element calculations.

Structural Dynamics 11 Solutions for selected continuous systems

summer term 2019 213
 10.00
8.00 FT FEM
compression
torsion (St.-Venant)
6.00 y-bending (Euler-Bernoulli)
z-bending (Euler-Bernoulli)

4.00 y-bending (Timoshenko)

kx [ rad
m
]

z-bending (Timoshenko)

2.00 y-bending 2 (Timoshenko)

z-bending 2 (Timoshenko)

0.00
0 4000 8000 12000 16000 20000
ω[ rad
s
]

Figure 11.18: Dispersion diagram for different wave numbers for a rectangular cross-section (from
Kreutz [2013])

11.4 Thin plates

Analogously to beam structures, the dispersion equations can also be determined for the
vibration of thin plates. The differential equations of a elastically supported plate according
to Kirchhoff theory in Cartesian coordinates with the normal forces Hx and Hy in x- and y-
direction, respectively, per unit length (in [N/m]) and the spring stiffness per unit area kw0 bedd
(in [N/m3 ]) is given as
!
∂ 4w ∂ 4w ∂ 4w ∂ w 2 2
0∂ w ∂ 2w
B 0
+2 2 2 + + Hx0 + Hy 2 + kwbedd w + ρh 2 = p(x,y,t)
0
(11.93)
∂x4 ∂x ∂y ∂y 4 ∂x2 ∂y ∂t

3
with the plate stiffness B 0 = 12(1−ν
Eh
2 ) , the plate thickness h, the Poisson ratio ν, the dis-

placement perpenicular to the central plate plane w = w(x,y,t), and the density ρ. Under
consideration of shear deformations and rotatory inertia according to the Mindlin theory,
Eq. (11.93) can be extended (comp. Cremer et al [2005]).

B0ρ
! ! !
∂ 4w ∂ 4w ∂ 4w h3 ∂ 4w ∂ 4w h3 ρ ∂ 4 w
B 0
+2 2 2 + − +ρ + +ρ =
∂x4 ∂x ∂y ∂y 4 G 12 ∂x2 ∂t2 ∂y 2 ∂t2 12 G ∂t4
B0
!
∂ 2p ∂ 2p ρI ∂ 2 p
= p− + + (11.94)
Gh ∂x2 ∂y 2 Gh ∂t2

Here, influences resulting from the normal forces and an elastic bedding are neglected. The
solutions for the free vibration problem (p = 0) can be obtained by applying an exponential

Structural Dynamics 11 Solutions for selected continuous systems

summer term 2019 214
approach to Eqs. (11.93) and (11.94) similar to the procedure for the beam in Eqs. 11.55
and 11.56. Hence, it follows from Eq. (11.93) with the wave numbers kx and ky in x- and
y-direction, respectively, and Hx0 = Hy0 = H 0
 2  
B 0 kx2 + ky2 + H 0 kx2 + ky2 + kw0 bett − ρhω 2 = 0. (11.95)

Introducing the wave number kr2 = kx2 + ky2 , it follows

B 0 kr4 + H 0 kr2 + kw0 bett − ρhω 2 = 0. (11.96)

The resulting dispersion relation thus corresponds to the relations for the beam in Eq. (11.92).
 
The wave number kr can be interpreted with a wave, inclined by the angle α = arcsin kkxr
(comp. Fig. 11.19) in the x-y-plane.

λr
λy
α

x
λx = λr
sin(α)

Figure 11.19: Einfallswinkel der Biegewelle

Further relations are covered in Cremer et al [2005]. Tab. 11.4 shows the radial eigenfrequencies
for rectangular plates under various boundary conditions.

Structural Dynamics 11 Solutions for selected continuous systems

summer term 2019 215
 2π
s
E d3
ωi = βi 2
a 12 (1 − ν 2 ) µ
ν: Poissons ratio β1 = β1,1
µ: mass per unit length β2 = min(β1,2 , β2,1 )

hinged
γ= a
b
clamping

β1,1 = 1.57 (1 + γ 2 )
b
β2,1 = 6.28 (1 + 0.25 γ 2 )

a β1,2 = 1.57 (1 + 4 γ 2 )

√
β1,1 = 1.57 1 + 2.5 γ 2 + 5.14 γ 4
b √
β2,1 = 6.28 1 + 0.625 γ 2 + 0.321 γ 4
√
a β1,2 = 1.57 1 + 9.32 γ 2 + 39.06 γ 4

√
β1,1 = 1.57 5.14 + 2.92 γ 2 + 2.44 γ 4
b √
β2,1 = 9.82 1 + 0.266 γ 2 + 0.0625 γ 4
√
a β1,2 = 1.57 5.14 + 10.86 γ 2 + 25.63 γ 4

√
β1,1 = 1.57 1 + 2.33 γ 2 + 2.44 γ 4
b √
β2,1 = 6.28 1 + 0.582 γ 2 + 0.152 γ 4
√
a β1,2 = 1.57 1 + 8.69 γ 2 + 25.63 γ 4

√
β1,1 = 1.57 2.44 + 2.72 γ 2 + 2.44 γ 4
b √
β2,1 = 7.95 1 + 0.395 γ 2 + 0.095 γ 4
√
a β1,2 = 1.57 2.44 + 10.12 γ 2 + 25.63 γ 4

√
β1,1 = 1.57 5.14 + 3.13 γ 2 + 5.14 γ 4
b √
β2,1 = 9.82 1 + 0.298 γ 2 + 0.132 γ 4
√
a β1,2 = 1.57 5.14 + 11.65 γ 2 + 39.06 γ 4

Table 11.4: Eigenfrequencies of rectangular plates (Kirchhoff, Eq. (11.93)), without elastic bedding
and without horizontal forces) from Grundmann et al [1983]

Structural Dynamics 11 Solutions for selected continuous systems

summer term 2019 216
11.5 Eigenfrequencies and mode shapes of selected systems

In the following eigenfrequencies and mode shapes for two selected examples are given. Further
examples can be found in, e.g., Müller and Möser [2004]; Cremer et al [2005].

Circular ring (from Grundmann et al [1983])

w EA: axial stiffness

EI: bending stiffness
r
ϕ µ: mass per unit length
r: radius
w: deflection in radial direction

mode shape φj radial eigenfrequency ωj

j (j 2 −1) 1 q EI
j≥2 wj cos(jϕ) √ 2
j 2 +1 r µ
q
j=0 wj 1
r
EA
µ

Cavity filled with air or fluid (from Müller and Möser [2004])

V : volume
c: compresssional wave velocity in the fluid
lx , ly , lz : dimensions in the spatial directions
lz

y x
ly lx

boundary pressure mode shape φj radial eigenfrequency ωj

r
j1 2 j2 2 j3 2
          
reverberant cos j1 π x
lx
cos j2 π y
ly
cos j3 π z
lz
πc lx
+ ly
+ lz
r
j1 2 j2 2 j3 2
          
absorbent sin j1 π x
lx
sin j2 π y
ly
sin j3 π z
lz
πc lx
+ ly
+ lz

Structural Dynamics 11 Solutions for selected continuous systems

summer term 2019 217
11.6 Modal analysis for loaded Euler-Bernoulli beam

The modal superposition method can be applied to continuous systems analogously to the
application for MDOF systems (comp. Section 9.2). We present the method exemplarily for
an Euler-Bernoulli beam hereinafter.

11.6.1 Euler-Bernoulli beam under step load

For every time t the vibration of a continuous system can be described by a superposition of
mode shapes φj (x) (analogously to the eigenvectors φj for discrete systems) multiplied with
time dependent factors yj (t), often called modal coordinates.

∞
w(x, t) = φj (x)yj (t) = φ1 (x)y1 (t) + φ2 (x)y2 (t) + . . . + φn (x)yn (t) + . . . (11.97)
X

j=1

φ1 (x) φ2 (x) φ3 (x)

φ1,1 φ ...
1,2 φ1,3

Analogously to the approach for MDOF systems (comp. Chapter 8, Eqs. 8.19 and 8.20),
orthogonality conditions for the mode shapes are given. For an undamped Euler-Bernoulli
beam, we find:

Zl
φi (x)µ(x)φj (x) dx = mj δij (11.98)
0

Zl
φ00i (x)EI(x)φ00j (x) dx = kj δij (11.99)
0

A more formal introduction concerning properties of mode shapes for different differential
equations can be found, e.g. in Humar [2012].

Structural Dynamics 11 Solutions for selected continuous systems

summer term 2019 218
 p(x)
Iθ Ps
ms MS
EI(x), µ(x)
kw kϕ

x
x1 x2 x3 x4 x5 x6
l
l

For arbitrary systems the generalized quantities then are:

Zl
2
mj = µ(x)φ2j (x) dx + Ms φ2j (x1 ) + Iθ φ0j (x2 ) (11.100)
0

Zl
2 2
kj = ωj2 mj = EI(x)φ00j (x) dx + kw φ2j (x3 ) + kϕ φ0j (x4 ) (11.101)
0

Zl
pj = p(x)Φj (x) dx + Ps φj (x5 ) + Ms φ0j (x6 ) (11.102)
0

Remark

If the eigenfrequencies of a system are not known, the relation in Eq. (11.101) can be
used for estimating the eigenfrequency ωj . For this a mode shape φj (x) is inserted
into Eqs. (11.100) and (11.101) and the generalized quantities are evaluated. Then the
Rayleigh quotient can be determined (here neglecting rotational inertia and attached
springs):

Rl
EI(x)φ2 (x)00 dx
0
ω2 ≤
Rl
µ(x)φ2 (x) dx
0

Structural Dynamics 11 Solutions for selected continuous systems

summer term 2019 219
With the generalized quantities the differential equations can be decoupled, which leads to the
known form of the SDOF system

mj ÿj + kj yj = pj

The time dependent functions yj (t) are generalized or modal coordinates that give the contribu-
tion of each mode shape. The accuracy of the solution in Eq. (11.97) improves with the number
of considered mode shapes (eigenvectors). In practice, the representation in Eq. (11.97) has
to be truncated. For this reason, only the contributions of n mode shapes are considered.
n
φj (x)yj (t) = φ1 (x)y1 (t) + φ2 (x)y2 (t) + . . . + φn (x)yn (t)
X
w(x, t) ≈
j=1

Then sequentially modal contributions can be added until the solution has converged. Theo-
retically the exact is obtained by adding all terms (infinitely many), however, considering the
Bernoulli-theory.
In the case of continuous systems, e.g. beams, it has to be verified that the underlying theory
(Euler-Bernoulli- hypothesis) is still valid. Often, a low number of mode shapes is sufficient
to obtain an accurate solution. Frequently, only considering the first mode shape is sufficient
to describe low frequency vibrations accurately enough.
Depending on the form of the load, the different methods for solving a SDOF system can be
applied (comp. Chapter 7.4).

Example 11.6

F F (t)

F
l l t
2 2

w
The depicted single span beam is excited by the point force p(x,t), modeled by a step
function at t = 0. We want to determine the displacement and internal forces response,
whereby we neglect any damping influences.
First, we derive the eigenfrequencies and -vectors. The homogeneous equation of motion

Structural Dynamics 11 Solutions for selected continuous systems

summer term 2019 220
 of the Euler-Bernoulli beam with constant bending stiffness reads

EIw0000 + µẅ = 0

with can be solved by a separation approach w(t) = W (x)f (t) (comp. Section 11.1.2.1)
that leads to the following general solution (comp. Eq. (11.57)):

W (x) = Cx1 cosh(kx) + Cx2 sinh(kx) + Cx3 cos(kx) − Cx4 sin(kx)

f (t) = Ct1 cos(ωt) − Ct2 sin(ωt)

For the single span beam, the mode shapes (eigenfunctions) are derived in Ex. 11.3.
They read

x
 
φj (x) = sin jπ
l

for j ∈ N+ . We can check that the boundary conditions are satisfied:

x = 0 : φj (0) = 0, φ00j (0) = 0

x = l : φj (l) = 0, φ00j (l) = 0

The corresponding eigenfrequencies are given in Eq. (11.63), repeated here:

4
π EI

ωj2 = j = j 4 ω12 .
l µ

This can be solved for ω 2 , which are the eigenfrequencies given in Eq. (11.63). We can
now derive the generalized quantities Mj , Kj , pj of the non-homogeneous problem by:

Zl Zl
x 1
 
mj = µ(x)φ2j (x) dx =µ sin 2
j·π dx = µl
l 2
0 0
4
π EI
kj = ωj2 Mj = j 4
2l3
xp π
   
pj = F φj (xp ) = F sin j · π = F sin j
l 2

Structural Dynamics 11 Solutions for selected continuous systems

summer term 2019 221
 For the generalized load pj we find

for j = 1, 5, 9, . . .

F

π
  
F sin j = 0 for j = 2, 4, 6, . . .
2 
for j = 3, 7, 11, . . .


−F

This leads to decoupled differential equations that correspond to the differential equa-
tion of the SDOF system:

mj ÿj + kj yj = pj

or
pj
ÿj + ωj2 yj =
mj

The solution to the non-homogeneous differential equation consists of the sum of the
homogeneous and particular solution. The general solution for the j-th mode shape
reads:

yj,h = yj,h01 cos(ωj t) − yj,h02 sin(ωj t)

The following heuristic approach is made, since the load term is a constant function

yj,p = yj,p0

Inserted into the equation of motion, it reads

pj
ωj2 yj,p0 =
Mj

The particular solution finally reads

pj pj
yj,p0 = 2
=
ωj mj kj

Thus, the full solution is given by

pj
yj (t) = yj,01 cos(ωj t) − yj,02 sin(ωj t) +
kj
ẏj (t) = −ωj (yj,01 sin(ωj t) + yj,02 cos(ωj t))

Structural Dynamics 11 Solutions for selected continuous systems

summer term 2019 222
 We assume a system at rest at time t = 0. Thus the initial conditions are given by:

w(x, 0) = 0
ẇ(x, 0) = 0

The first condition leads to

∞
w(x, 0) = φj (x)yj (t = 0) = 0
X

j=1

Thus, all yj (t = 0) have to be zero.

pj pj
yj,01 + =0 → yj,01 = −
kj kj

The initial condition with respect to velocity results in

∞
ẇ(x, 0) = φj (x)ẏj (t = 0) = 0
X

j=1

Thus, all ẏj (t = 0) have to be zero. This leads to

−ωj yj,02 = 0 → yj,02 = 0

Then, the response of the j-th mode shape follows to

pj π 2l3
 
yj (t) = (1 − cos(ωj t)) = F sin j (1 − cos(ωj t))
kj 2 j 4 π 4 EI

The solution in the physical space is found by superposition:

∞
w(x,t) = φj (x)yj (t)
X

j=1
∞
x π 2l3
   
= sin jπ F sin j (1 − cos(ωj t))
X

j=1 l 2 j 4 π 4 EI
2F l3 x 1 x
    
= sin π (1 − cos(ω 1 t)) − sin 3π (1 − cos(ω3 t))+
π 4 EI l 34 l
1 x
  
+ 4 sin 5π (1 − cos(ω5 t)) − . . .
5 l

Structural Dynamics 11 Solutions for selected continuous systems

summer term 2019 223
 The internal forces can be derived from the derivatives of w(x,t), as EI is constant,

M (x, t) = −EIw00 (x, t)

Q(x, t) = −EIw000 (x, t)

or through integration using the equilibrium of forces, where the inertia forces are
considered as external loads

pI (x, t) = −mẅ(x, t),

with

2F l3
"
x ω2 x
   
ẅ(x, t) = 4 ω12 sin π cos(ω1 t) − 43 sin 3π cos(ω3 t)+
π EI l 3 l
#
ω52 x
 
+ 4 sin 5π cos(ω5 t) − . . .
5 l
2F x x
    
= sin π cos(ω1 t) − sin 3π cos(ω3 t)+
ml l l
x
  
+ sin 5π cos(ω5 t) − . . .
l

The internal forces follow from the derivatives to:

2F l x 1 x
    
M (x, t) = 2 sin π (1 − cos(ω1 t)) − 2 sin 3π (1 − cos(ω3 t))+
π l 3 l
1 x
  
+ 2 sin 5π (1 − cos(ω5 t)) − . . .
5 l
2F x 1 x
    
Q(x, t) = cos π (1 − cos(ω1 t)) − cos 3π (1 − cos(ω3 t))+
π l 3 l
1 x
  
+ cos 5π (1 − cos(ω5 t)) − . . .
5 l

or through integration

2F x x
    
pI (x, t) = − sin π cos(ω1 t) − sin 3π cos(ω3 t)+
l l l
x
  
+ sin 5π cos(ω5 t) − . . .
l
F 2F x 1 x
    
Q(x, t) = − cos π cos(ω1 t) − cos 3π cos(ω3 t)+
2 π l 3 l

Structural Dynamics 11 Solutions for selected continuous systems

summer term 2019 224
 1 x

+ cos 5(π ) cos(ω5 t) − . . .
5 l
F x 2F l x 1 x
    
M (x, t) = − 2 sin π cos(ω1 t) − 2 sin 3π cos(ω3 t)+
2 π l 3 l
1 x
  
+ 2 sin 5π cos(ω5 t) − . . .
5 l

where, due to symmetry, we 

only

consider the range

0 ≤ x < 2l . The following

table

shows the displacement w 2l , t0 , the moment M 2l , t0 and shear force Q 2l , t0 for
the time t0 = ωπ1 relative to the corresponding static value and for a different number
of considered modes.

j 1 3 5 limit
2 · 96 int. forces
w 1,971 + 0,024 1,995 + 0,003
= 1,971 2,0 derived by
wst π4 = 1,995 = 1,998
M 16 1,621 + 0,180 1,801 + 0,065
= 1,621 2,0 derivative
Mst π2 = 1,801 = 1,866
EIw00
M 8 1,811 + 0,090 1,901 + 0,032
1+ = 1,811 2,0 equilibrium
Mst π2 = 1,901 = 1,933
pI = −µẅ
Q 8 2,546 − 0,848 1,698 + 0,509
= 2,546 2,0 derivative
Qst π = 1,698 = 2,207
EIw000
Q 2,273 − 0,424 1,849 + 0,254
1+ 4
= 2,273 2,0 equilibrium
Qst π = 1,849 = 2,103
pI = −µẅ

The results show that under consideration of a few modes only, already a good approx-
imation of the response is obtained. Note though that this is even valid for the applied
load. The generalized load usually decays with increasing frequency, if, e.g. in the case
of an earthquake, the load acts evenly distributed on the system. An exception are
single loads, for which the generalized load stays more or less equal for increasing j.

- - -
+ + + +
+
+ +

Furthermore, a Fourier transform reveals that the amplitude of the force in the frequency

Structural Dynamics 11 Solutions for selected continuous systems

summer term 2019 225
 domain f˜(ω) decays with ∼ ω1 . For this reason, the modal forces pj for higher j decay
rapidly for the applied step load.
Furthermore we note that the convergence of the internal forces is slower than for the
displacement. This is due to the fact that the higher frequency mode shapes exhibit
larger bending terms, whereas they give small contributions to the displacement.
The accuracy of the internal forces is better when they are determined by the equilib-
rium method, where the inertia forces are assumed as external loading, rather when
determined by the derivatives of the displacements. This leads to the conclusion that
dynamic internal forces for structures with known mode shapes should be calculated
preferably by the method of external inertia forces.

load moment tranvsversal forces

F
Fl
4

F
2
= − −
2F
l 2F l
π2

2F
π
− − −
2F
l

2F
3π
2F l
9π 2
+ + +
. . .
. . .
. .

limit: Fl
2
F

The presented method can also be applied to different structures, e.g. plates or shells.
Nevertheless the modal analysis is restricted to linear systems, as here the superposition
principle is valid.

Structural Dynamics 11 Solutions for selected continuous systems

summer term 2019 226
11.7 Summary

For the elements in the technical bending theory different differential equations apply. Therein,
in general, the beam differential equation as differential equation of fourth order has to be
distinguished from the differential equations of the torsional, shear or longitudinal rod and
the string, those being of second order. Whereas no unique wave velocity results for the beam
differential equation due to dispersion of the beam wave, for the differential equations of second
order frequency-independent wave velocities can be derived.

Structural Dynamics 11 Solutions for selected continuous systems

summer term 2019 227
12 Approximate solutions of continuous
systems

12.1 Discretization of continuous systems

Systems with distributed elasticities and masses have an infinite number of degrees of freedom.
In some rare cases, differential equations are found, that can be solved analytically (see Chapter
11). In general this is not possible, and the motion of such systems is approximated by
appropriate shape functions. The task to be solved is then described by a finite number of
discrete unknowns. Often the unknowns are then the nodal displacements and/or rotations.
The system is thus transformed into a discrete system, and corresponds to a MDOF system.
For the finite element method, the field of the displacement vector

uT (x,y,z,t) = [u(x,y,z,t), v(x,y,z,t), w(x,y,z,t)] (12.1)

is approximated through spatial shape functions φj (x,y,z). Thereby an approximate solution

e (x,y,z,t) is obtained. The problem is hence reduced to a determination of the unknown
u
time-dependent coefficients αj (t).

n
e (x,y,z,t) = αj (t)φj (x,y,z) (12.2)
X
u(x,y,z,t) ≈ u
j=1

Herafter, the notation e· is dropped for the sake of clarity. The applied shape function have
to be able to describe the significant properties of the solution, especially the wavelengths,
typically described by the frequency dependent wave numbers k. Therefore knowledge about
the analytic solutions is necessary. The single shape functions can extend over large or infinite
domains of the system or be limited to finite elements. Especially in dynamics, global shape
functions that are directly obtained from analytic solutions (comp. Chapter 11) or that are
based on analytic solutions for idealized boundary conditions (e.g. Component Mode Synthesis
Buchschmid [2011]; Buchschmid et al [2010]), can be very efficient.

Structural Dynamics
summer term 2019
228
 Example 12.1: Shape functions Euler-Bernoulli beam

We consider a planar problem and uniaxial bending. The displacement w(x) of a beam
element perpendicular to the beam axis over the element length le is described by shape
functions

w (x) = Φ (x) α (12.3)

with the row vector of the shape functions Φ (x) and the column vector of coefficients
α. The displacements in longitudinal direction directly follow from the Euler-Bernoulli
beam theroy. The spatial displacement field can thereby be completely described by
w(x).
In the following, a cubic approach
h i
Φ (x) = 1, x, x2 , x3

with the four coefficients

αT = [C0 , C1 , C2 , C3 ]

is exemplarily chosen. Practically the coefficients are expressed in terms of the unknown
nodal displacements and roations of the beam element

wTe = [wi , φi , wk , φk ]

with φ = dw
dx
.

i le k
wi wk
µ ω2 4
λ4 = EI
l
+Mik
φi +Mki
+Vik φk
+Vki
Figure 12.1: Nodal forces and displacement degrees of freedom

Structural Dynamics 12 Approximate solutions of continuous systems

summer term 2019 229
 Thus, it reads

we = Aα

with

1 0 0 0
 
0 1 0 0
A= 
1 le le2 le3 
 

0 1 2le 3le2

For the coefficients α it holds

α = A−1 we

with

1 0 0 0
 

 0 1 0 0 

A−1 =

− 32 − 2 3 1
− le 
 l le e le2 
2 1
le3 le2
− l23 1
le2
e

Using Eq. (12.3), the displacements along the beam axis can be expressed through the
nodal displacements we .

w (x) = Φ (x) A−1 we = N (x) we (12.4)

With ξ = x
le
we obtain
h    i
N (ξ) = 1 − 3ξ 2 + 2ξ 3 , le ξ − 2ξ 2 + ξ 3 , 3ξ 2 − 2ξ 3 , le −ξ 2 + ξ 3 (12.5)

Structural Dynamics 12 Approximate solutions of continuous systems

summer term 2019 230
 1 1
ξ 1 ξ
1
N1 (ξ) N2 (ξ)

1 1
ξ ξ
1 1
N3 (ξ) N4 (ξ)
Figure 12.2: Functions for unit displacements/rotations - cubic approach

12.1.1 The principle of virtual work

In the dynamic equilibrium, the sum of internal and external virtual work δW of a finite
element is given by
Z Z Z Z
δW = − δε σ dΩ −
T
δu ρü dΩ +
T
δu p dΩ +
T
δuT t dΓ = 0 (12.6)
(Ω) (Ω) (Ω) (Γ)

with the domain of the structure Ω, the boundary of the structure Γ, the density ρ, the stresses
in the interior of the domain σ, the corresponding vector of stresses on the boundary t, and
the vector of external forces p. The vector field δu describes a virtual displacement field, δε
the corresponding virtual strains.

Reminder 12.1

For the derivation of the equations of motion for a linear elastic isotropic material
(Lamé equation, refer to the lecture Continuum Mechanics) the relation between the
displacements u, v, and w (in x-, y-, and z-direction) and the resulting strains for
linearized systems (small deformations) are:

∂u ∂v ∂u
εx = γxy = +
∂x ∂x ∂y
∂v ∂w ∂v
εy = γyz = +
∂y ∂y ∂z
∂w ∂u ∂w
εz = γzx = +
∂z ∂z ∂x

Structural Dynamics 12 Approximate solutions of continuous systems

summer term 2019 231
 with the strains εk and the shear deformations γmn .
Furthermore, for a linear elastic isotropic material the stress-strain relationship is given
by

σx = λ(εx + εy + εz ) + 2Gεx τxy = τyx = Gγxy

σy = λ(εx + εy + εz ) + 2Gεy τyz = τzy = Gγyz
σz = λ(εx + εy + εz ) + 2Gεz τzx = τxz = Gγzx

The relations given in Eq. (??) can be applied to the n shape functions in Eq. (12.2) (small
deformations are assumed, that allow linearization, Bathe [2007]). Hence the strains result in

n
ε(x,y,z) = Bj (x,y,z)αj = Bα (12.7)
X

j=1

with

εT = [εx , εy , εz , γxy , γyz , γzx ] (12.8)

Thereof n different virtual displacements and the corresponding virtual strains can be derived.

δuj (x,y,z) = φj (x,y,z)δαj , δεj (x,y,z) = Bj (x,y,z)δαj (12.9)

The acceleration field is found analogously:

n
ü(x,y,z) = φj (x,y,z)α̈j = Φα̈ (12.10)
X

j=1

The stresses can be found under consideration of Eq. (??). Thus,

n
σ(x,y,z) = Eε(x,y,z) = E Bj (x,y,z)αj = EBα (12.11)
X

j=1

with

σ T = [σx , σy , σz , τxy , τyz , τzx ] (12.12)

Structural Dynamics 12 Approximate solutions of continuous systems

summer term 2019 232
and

λ + 2G 0 0 0
 
λ λ
 λ

λ + 2G λ 0 0 0
 λ λ λ + 2G 0 0 0
 
[E] =  (12.13)
0 0 0 0 0


 G 
0 0 0 0 G 0
 

0 0 0 0 0 G

The vector fields Bj , ε, ü, φj and σ are column vectors that depend on x, y, and z. α is the
vector of the coefficients αj .
In the matrices B and Φ these are contained column-wise. It results a system of equations
with n equations, in which every equation corresponds to one single virtual displacement state.
Thereby Eq. (12.6) results in

Z Z
δW = −δα T
B EB dΩ α − δα
T T
ρΦT Φ dΩ α̈+
(Ω) (Ω)
Z Z
+ δα T
Φ p dΩ + δα
T T
ΦT t dΩ = 0 (12.14)
(Ω) (Γ)

Example 12.2: Virtual work at the Euler-Bernoulli beam

This example is a continuation of Ex. 12.1.

First, the strains in the Euler-Bernoulli beam for uni-axial bending without normal
force are approximated by:

εx = −zw00 (12.15)

The virtual work in Eq. (12.6) can then be written as:

Zle Z Zle Z Zle

δW = − σx δεx dA dx − ρẅδw dA dx + p(x,t)δw dx + Vik δw(x = 0)
0 (A) 0 (A) 0

+ Vki δw(x = le ) + Mik δw0 (x = 0) + Mki δw0 (x = l) = 0 (12.16)

with the cross sectional are A and the outer load p(x,t). With Eq. (12.15), the inner

Structural Dynamics 12 Approximate solutions of continuous systems

summer term 2019 233
 virtual work linked with the stresses follows to:

Zle Z Zle Z Zle Zle

σx δεx dA dx = Ez w δw dA dx =
2 00 00
EIy w δw dx =
00 00
M δw00 dx .
0 (A) 0 (A) 0 0

Further the work related to the d’ Alembert inertia forces reads:

Zle Z Zle
ρẅδw dA dx = µẅδw dx .
0 (A) 0

The nodal forces are gathered in the vector qT = [Vik , Mik , Vki , Mki ] (comp. Fig. 12.3).

node

Vki
Mki
beam element
x

Mik
Vik w

node
Figure 12.3: Nodal forces for the beam element

For determining the inner virtual work using the shape functions in both, the actual
and virtual displacement fields, the second derivative of the displacement in Eq. (12.5)
has to be determined.

d2 1
B (x) = − N (x) = 2 [6 − 12ξ, le (4 − 6ξ) , −6 + 12ξ, le (2 − 6ξ)]
dx 2 le
d2
−w (x) = − 2 N (x) we = B (x) we
00
dx

The bending stiffness D = EI is assumed constant over the element length. The
d’Alembert inertia forces result in

pi (x,t) = µẅ(x) = µN (x) ẅe (12.17)

Structural Dynamics 12 Approximate solutions of continuous systems

summer term 2019 234
 For the application of the principle of virtual work, here, the shape functions are also
used for the virtual displacements. Thus, it follows

δw (x) = N (x) δwe , −δw00 (x) = B (x) δwe

Hence we can write for the work expression in Eq. (12.16) (also comp. Eq. (12.14)):

Zle Zle
δW = − EIy w (x) δw (x) dx −
00 00
pi (x) δw (x) dx +
| {z } | {z }
0 T 0 T
δwT
e B (x)[D]B(x)we δwT
e N (x)µN(x)ẅe

Zle
p (x) δw (x) dx + Vik δwi + Mik δφi + Vki δwk + Mki δφk = 0 (12.18)
| {z } | {z }
0 T δwT
δwT
e N (x) eq

Out of this result the stiffness- and mass-matrix:

Zle


δW = δwe  T
− BT (x) DB (x) dx we + q−
0
| {z }
Ke

Zle Zle


− µNT (x) N (x) dx ẅe + p (x) NT (x) dx  = 0 (12.19)

0 0
| {z } | {z }
Me p

Thus, the equation of motion for a single beam element is given in terms of the nodal
displacements and rotations as

Me ẅe + Ke we = q + p

For cubic shape functions applied here, the mass- and stiffness-matrices Me and Ke for
an element of length le with constant stiffness EI and mass distribution µ are

12 6l −12 6l 156 22l 54 −13l

   

EI  6l 4l 2
−6l 2l2  µl  22l 4l 2
13l −3l2 
Ke = 3   ,

Me =  
l −12 −6l 12 −6l 420  54 13l 156 −22l
  

6l 2l2 −6l 4l2 −13l −3l2 −22l 4l2

Me is termed as consistent mass matrix of the beam element. Depending on the ratio
of wave length and element length, the mass is often concentrated at the nodes (lumped

Structural Dynamics 12 Approximate solutions of continuous systems

summer term 2019 235
 mass) and the inertia forces and moments are applied as single forces and moments. In
the case of a viscous damping distributed like the mass µ, the damping matrix cẇ (x)
is treated like mẅ (x) (comp. Eqs. (12.17) and (12.18)):

pD (x) = cẇ(x) = c N (x) ẇe

and

Zl
Ce = c NT (x) N (x) dx .
0

This leads to the equation of motion of the beam element with distributed damping.
Further reference is given in section 8.2.3.3.

Me ẅe + Ce ẇe + Ke we = q + p (12.20)

12.1.2 Lumped mass matrix and static condensation

In general, the dynamic analysis of a consistent mass system considerably requires more com-
putational effort than a lumped mass description. At the same time, the accuracy is often
only slightly improved. Therefore, it can be advantageous to substitute the general consistent
mass system by a lumped mass system.
In the lumped mass description, the masses are concentrated in the nodes i.e., for the node i.

1 1
Mi = mi−1 · li−1 + mi+1 · li+1
2 2

where the subscripts i−1 and i+1 are related to the mass or length of the neighboring elements,
respectively. The rotatory inertia is neglected in this case. Accordingly, the mass matrix is
simply a diagonal matrix, with zero elements in the position referring to the rotational degrees
of freedom (The damping matrix can be brought into a corresponding form.).
If the nodal rotations do not lead to contributions to the mass and damping matrix, it appears
advantageous to eliminate them by a procedure called static condensation. To derive this, the
unknowns are rearranged as follows:
h i
wT = [w1 , w2 , . . . , wn , φ1 , φ2 , . . . , φn ] = wTT , wTφ .

Structural Dynamics 12 Approximate solutions of continuous systems

summer term 2019 236
The equations of motion are then written as
" #" # " #" # " #" # " #
Kww Kwφ wT Mw 0 ẅT Cw 0 ẇT p
+ + = T
Kφw Kφφ wφ 0 0 ẅφ 0 0 ẇφ 0

with the nodal forces pT . The second row describes the static equilibrium without inertia
forces and gives a dependence between wφ and wT :

Kφw wT + Kφφ wφ = 0 → wφ = −K−1

φφ Kφw wT

If this is substituted into the first row, the equation of motion with the order reduced by half

Mw ẅT + Cw ẇT + Kw wT = pT

is obtained, with the modified stiffness matrix

Kw = Kww − Kwφ K−1

φφ Kφw

12.1.3 Principle of Hamilton

Alternatively the equations of motion can be determined from the principle of Hamilton (comp.
4.1.3.1). The term extends the approach with the work of the non-conservative forces (external
forces, damping forces, general forces, that cannot be described by a potential, comp. section
??).

Zt2 Zt2
δ Ekin − Epot dt + δWnc dt = 0 (12.21)
t1 t1

The procedure is analogous to the example 4.10.

Structural Dynamics 12 Approximate solutions of continuous systems

summer term 2019 237
13 Random vibrations

13.1 Introduction

Most of the excitation processes in structural dynamics are of random nature, respectively have
random components. However, many of them are modeled by deterministic loads. With the
help of the probability theory we can deduce properties of the stochastic response under the
knowledge of the random properties of the excitation process. Thereby an uncertain quantity,
e.g. an excitation force or displacement, is modeled as a random variable. Using the theory
of probability certain characteristics of a random response can then be determined, which
allow to assess probabilities of the underlying events, e.g. probability of exceeding a certain
threshold. This procedure is outlined briefly in the following. Further insights can be found
in [Wirsching et al 2006; Roberts and Spanos 2003; Petersen 1996]. [Li and Chen 2009] gives
an overview over models for stochastic processes for typical excitations. The derivation in this
chapter follow Lutes and Sarkani [2004].
Excitation and response time histories that are random in nature can be modeled as a stochas-
tic (or random) processes. A stochastic process can be understood as a set of random vari-
ables.
Consider for example some process x(t), such as an excitation or response that evolves over
time. At every instance of time t the outcome of the process is uncertain and x(t) is modeled
as a random variable X(t). The whole stochastic process is then denoted by {X(t)}1 . The
parameter t can be viewed as the index parameter for the process. Usually this index set is
continuous.
A stochastic process can also be interpreted in terms of all possible realizations. Note though
that this concept only holds in the limit for an infinite number of observations. Neverthe-
less, through a sufficiently large number of realizations (observations) of the process a good
understanding of the likelihood of certain outcomes can be obtained.
The collection of realizations, also called time histories, of a stochastic process is then called
ensemble.
From m observations discrete histograms can be derived, from which the relative frequency of
occurence nmk emerges. The histogram describes how often the process {X(t)} at time ti takes
a certain value x. For this, m time histories xi (t) have to be evaluated. In the limit ∆x → 0
1
This follows the common practice to denote a set

Structural Dynamics
summer term 2019
238
and m → ∞ the histogram corresponds to the probability density function fX(t) (x). In most
applications however, it is not feasible to collect a sufficient number of process time histories
in order to obtain an accurate description of the process {X(t)}

xm (t) xm (t2 )=bm

xm (t1 )=am

t
x2 (t)
x2 (t1 )=a2 x2 (t2 )=b2
Ensemble
x1 (t) t
x1 (t1 )=a1 x1 (t2 )=b1

t
τ
t = t1 t = t2

Figure 13.1: Stochastic process with m realizations

Structural Dynamics 13 Random vibrations

summer term 2019 239
 x1 (t)
realization

t
histogram at time ti
∆x
(for the total ensemble)
x2 (t)
ensemble
∆x
nk
t m

xm (t)
x
∆x

∆x t

ti tj
Figure 13.2: Histogramm X(ti )

13.2 Description of random variables

13.2.1 Single random variable

A real single random variable is a function, X : S → R that maps events in S to intervals in

R. In the scope of this lecture we are usually concerned with continuous random variables.
Through random variables we can identify the probability of occurence of certain events. To
this end, we describe the probability Pr of occurrence of the event {X ≤ x}, i.e., the event that
the random variable X takes a value smaller or equal than x, by the cumulative distribution
function (CDF) FX in terms of the outcome x. Thus,

FX (x) = Pr (X ≤ x) . (13.1)

Structural Dynamics 13 Random vibrations

summer term 2019 240
For continuous random variables we define the probability density function (PDF) fX (x) by

fX (x) dx = Pr (x < X ≤ x + dx) . (13.2)

It gives the probability that the outcome of X takes a value in the interval x < X ≤ x +
dx (comp. Fig. 13.3). We note, that the PDF of a random variable itself does not give a
probability, the probability in the above equation is given for the infinitesimally small interval
dx. From differentiation of the CDF, we find the PDF and vice versa, from the integration of
the PDF we find the CDF,

dFX (x)
fX (x) = (13.3)
dx
Zx
FX (x) = fX (x) dx (13.4)
−∞

Especially it holds

Z∞
fX (x) dx = FX (∞) = 1. (13.5)
−∞

fX (x)

fX (x) dx = Pr(x < X < x + dx)

µX x
dx

Figure 13.3: Probability density function fX (x) of the random variable X

Information on the random variable X can also be given in terms of its moments. To this end,
we define the expected value of any function g (X) of the random variable X as

Z∞
E [g (X)] = g (x) fX (x) dx (13.6)
−∞

Structural Dynamics 13 Random vibrations

summer term 2019 241
We define the n-th moment of random variable X as follows:
Z∞
E [X ] =
n
xn fX (x) dx (13.7)
−∞

For the case n = 1, the first order moment is obtained. It is called the mean or expected value
of X:
Z∞
µX = E [X] = xfX (x) dx (13.8)
−∞

For n = 2, the second order moment, also termed mean square of X, is obtained. Furthermore,
the n-th central moment of X is defined as
Z∞
E [(X − µX ) ] = n
(x − µX )n fX (x) dx (13.9)
−∞

It holds

E [(X − µX )] = 0 (13.10)

For the case n = 2, we obtain the variance of X.

h i Z∞
Var [X] = E (X − µX ) 2
= (x − µX )2 fX (x) dx (13.11)
−∞

It gives a characterization of the deviation of the outcomes of X around the mean value µX .
The standard deviation σX is then defined as
q
σX = Var [X] (13.12)

Most commonly the normal distribution (Gaussian distribution) is used as a probability dis-
tribution. Its PDF is given by

(x−µX )2
1 −
fX (x) = √ e 2σ 2
X (13.13)
σX 2π

This distribution is fully characterized by two parameters, which can be derived from its first
two moments, i.e., mean value µX and standard deviation σX .

Structural Dynamics 13 Random vibrations

summer term 2019 242
 σ2
fX (x) 2σ2
1 fX (x) = probability density function
µX = mean value
2 σ = standard deviation
µX1 = µX2
2σ1 x σ1 < σ2
µX σ 1

Figure 13.4: Normal distribution

13.2.2 Two random variables

Two random variables X and Y are described by the joint PDF fXY (x,y):

fXY = fXY (x,y)

y
fXY (x,y)

x
x

Now, we have two mean values and variances, each for the single random variables X and Y .
The mean values are given by

Z∞ Z∞
µX = E [X] = xfXY (x,y) dx dy (13.14)
−∞ −∞

Structural Dynamics 13 Random vibrations

summer term 2019 243
 Z∞ Z∞
µY = E [Y ] = yfXY (x, y) dx dy (13.15)
−∞ −∞

and the variances by:

h i
Var [X] = σX
2
= E (X − µX )2 (13.16)
h i
Var [Y ] = σY2 = E (Y − µY )2 . (13.17)

Analogously to Section 13.2.1, we furthermore introduce the joint moments of X and Y of

order (nx + ny )

Z∞
E [X Y nx ny
]= xnx y ny fXY (x,y) dx dy (13.18)
−∞

and the joint central moments of X and Y of order (nx + ny )

Z∞
E [(X − µX ) nx
(Y − µY ) ] =
ny
(x − µX )nx (y − µY )ny fXY (x,y) dx dy (13.19)
−∞

Of most interest are lowest order joint moments, i.e., nx = ny = 1. For this case the joint
central moment is called covariance, given by

Z∞ Z∞
Cov [X,Y ] = 2
σXY = E [(X − µX ) (Y − µY )] = (x − µX ) (y − µY ) fXY (x,y) dx dy
−∞ −∞

Furthermore the correlation coefficient ρXY is defined as

Cov [X,Y ] 2
σXY
ρXY = q q =
Var [X] Var [Y ] σX σY

The covariance and the correlation coefficient can be interpreted as a measure of linear depen-
dence between X and Y .

Structural Dynamics 13 Random vibrations

summer term 2019 244
 ρXY > 0 ρXY < 0 ρXY = 1

ρXY ≈ 0 ρXY ≈ 0 ρXY ≈ 0

For the case that X and Y are completely independent, covariance and correlation coefficient
vanish.

Example 13.1

Consider the two random variables U and F , with mean value µF and standard deviation
σF , and where additionally it holds

U = kF + u0 .

k and u0 are deterministic. This relation describes, e.g. the random displacement
response of a linear system (spring) subject to a random static load. Thus, F and U
are perfectly linearly dependent. The expected value of U is given by

µU = kµF + u0

Then, we find for the variance σU2

σF2 U = E [(F − µF )(kF + u0 − c1 µF − u0 )]

= E [(F − µF )k(F − µF )]
h i
= k E (F − µF )2
= kσF2

and the covariance σF2 U

h i
σU2 = E (kF + u0 − kµF − u0 )2
h i
= k 2 E (F − µF )2

Structural Dynamics 13 Random vibrations

summer term 2019 245
 = k 2 σF2

The correlation coefficient then reads

σF2 U kσF2
ρF U = = = ±1
σF σU σF |k| σF

Because spring stiffness are greater than zero here, the case −1 can be omitted.

13.3 Description of stochastic processes

As noted in Section 13.1, a stochastic process is a set of random variables. It can thus be
seen as a generalization of random vectors, where the random vector consists of infinitely
many random variables. In the following we introduce common descriptions of a stochastic
process.

13.3.1 Probability distribution

In order to have a complete probabilistic description of a stochastic process, the joint prob-
ability of all single random variables {X(t1 ), X(t2 ), . . . , X(tn )} needs to be known. This can
be given by the joint probability density function

fX(t1 )X(t2 )···X(tn ) (x1 , x2 , . . . , xn ) (13.20)

In general, the information available on a random process is at most its first or second order
marginal distribution, e.g.

fX(t1 ) (x1 ) (13.21)

fX(t1 )X(t2 ) (x1 , x2 ) (13.22)

The marginal distributions can be easily derived from the more general description in Eq. (13.20).
Thereby fX(t1 ) (x1 ) describes the probability density for the random variable X(t1 ) at a single
time instance only, and fX(t1 )X(t2 ) (x1 , x2 ) describes the joint distribution of any two random
variables X(t1 ) and X(t2 ) at two time instances. An equivalent description can also be given in
terms of the joint cumulative distribution function. Often, descriptions of stochastic processes
are not given in terms of the joint probability, but in terms of their moment functions.

Structural Dynamics 13 Random vibrations

summer term 2019 246
13.3.2 Moment functions

The mean or expected value of a stochastic process is defined by

Z ∞
µX (t) = E [X(t)] = xfX(t) (x) dx . (13.23)
−∞

Note that now the mean value is a function of time. The autocorrelation function of the
stochastic process {X(t)} is defined as
Z ∞ Z ∞
φXX (t1 , t2 ) = E [X(t1 )X(t2 )] = x1 x2 fX(t1 )X(t2 ) (x1 ,x2 ) dx1 dx2 (13.24)
−∞ −∞

Correspondingly, we also define the autocovariance function of {X(t)}

KXX (t1 , t2 ) = E [[X(t1 ) − µX (t1 )] [X(t2 ) − µX (t2 )]] (13.25)

Mean, autocorrelation and autocovariance are related by

KXX (t1 , t2 ) = φXX (t1 , t2 ) − µX (t1 )µX (t2 ) (13.26)

Mean-square and variance function are obtained, by evaluating autocorrelation and autoco-
variance function, respectively, at the same time t = t1 = t2 :
h i
E X 2 (t) = E [X(t)X(t)] = φXX (t, t) (13.27)
2
σX (t) = KXX (t,t) (13.28)

Similarly, higher order moments can be defined, which is not further illustrated in the scope of
this lecture. It is noted that a full probabilistic description requires full knowledge about all
moments of the process. In this case the information contained in the full PDF of the process
and all the moment functions is the same.

Example 13.2: Cosine function with random amplitude and phase

Consider the following stochastic process

X(t) = X0 cos (ωt + Φ)

where X0 and Φ are random variables and ω is the deterministic frequency. Furthermore
X0 and Φ are assumed to be independent and Φ is assumed to be uniformly distributed

Structural Dynamics 13 Random vibrations

summer term 2019 247
 over the interval [0,2π]. The mean of the process is given by

µX (t) = E [X0 cos (ωt + Φ)]

Due to the independence we can write

Z2π
µX (t) = E [X0 ] E [cos (ωt + Φ)] = µX0 cos (ωt + ϕ) fΦ (ϕ) dϕ
0

The integral over the cosine function equals zero, thus it is µX (t) = 0. This holds for
any µX0 < ∞. Furthermore, since the mean value is zero, autocorrelation φXX and
covariance function KXX coincide. We write

φXX (t1 ,t2 ) = E [(X0 cos (ωt1 + Φ)) (X0 cos (ωt2 + Φ))]

Again, due to the independence of X0 and Φ it follows

h i
φXX (t1 ,t2 ) = E X02 E [cos (ωt1 + Φ) cos (ωt2 + Φ)]

We introduce the identity cos(x) cos(y) = 1

2
(cos(x + y) + cos(x − y)) and rewrite the
second part of the preceding equation as

1
E [cos (ωt1 + Φ) cos (ωt2 + Φ)] = E [cos (ω(t1 + t2 ) + 2Φ) + cos (ω(t1 − t2 ))]
2
2π
1Z 1
= (cos (ω(t1 + t2 ) + 2ϕ) + cos (ω(t1 − t2 ))) dϕ
2 2π
0
1
= cos (ω(t1 − t2 ))
2

Then, the autocorrelation function reads

E [X02 ] E [X02 ]
φXX (t1 ,t2 ) = cos (ω(t1 − t2 )) = cos (ωτ )
2 2

with τ = t1 − t2 . Fig. 13.5 depicts the autocorrelation function in terms of the time
difference τ and Fig. 13.6 shows five different realizations of the process, where addi-
tionally it is assumed that the amplitude X0 follows a standard normal distribution.
Subsequently we will see that a process, for which the autocorrelation only depends on
the time shift between two points, is called stationary. Thus, the stochastic process
introduced in this example is stationary.

Structural Dynamics 13 Random vibrations

summer term 2019 248
 φXX x(t)
E[X02 ]
2

−3π −π π 3π ωτ 4π ωt
−2π 2π

E[X02 ]
− 2
Figure 13.5: Autocorrelation function Figure 13.6: Five realizations of the
φXX for the above pro- stochastic process with
cess in terms of time X0 following the standard
shift. normal distribution.

13.3.3 Stationarity of stochastic processes

Stationarity of a stochastic process describes the property that some description of the stochas-
tic process is unchanged by an arbitrary shift along the time axis. A stochastic process {X(t)}
is called mean value stationary, if

µX (t + τ ) = µX (t). (13.29)

For this to hold, it must be

µX (t) = µX , (13.30)

i.e., the mean value of the process is constant over time. The process {X(t)} is called second-
moment stationary if the second-moment function, i.e., the autocorrelation function φXX is
invariant under a time shift:

φXX (t1 + r, t2 + r) = φXX (t1 , t2 ) (13.31)

This results in the fact that the autocorrelation function only depends on one time argument,
which is the time difference τ = t1 − t2 , or t1 = t2 + τ . Thus,

RXX (τ ) = φXX (t + τ, t) = E [X(t + τ )X(t)] . (13.32)

Structural Dynamics 13 Random vibrations

summer term 2019 249
Here, we set t2 = t. Furthermore, the process {X(t)} is called covariant stationary if the
second central moment function, i.e., the covariance function KXX is invariant under a time
shift:

GXX (τ ) = KXX (t + τ, t) = E [(X(t + τ ) − µX (t + τ )) (X(t) − µX (t))] . (13.33)

A process that is both mean and second-moment stationary can be shown to also be covariant
stationary. Processes with this type of stationarity are also termed weakly stationary or wide-
sense stationary. Strict stationarity implies stationarity in the probability density function.
For a covariant stationary process the variance of the process is found by evaluation of the
covariance function for the time shift τ = 0 (Eqs. ??,):
h i
Var [X] = E (X(t) − µX (t))2 = KXX (t, t) = GXX (0) (13.34)

xm (t) xm (t2 )=bm

xm (t1 )=am

t
x2 (t)
x2 (t1 )=a2 x2 (t2 )=b2
Ensemble
x1 (t) t
x1 (t1 )=a1 x1 (t2 )=b1

t
τ
t = t1 t = t2

13.3.4 Ergodicity of stochastic processes

Ergodicity describes the property of a stochastic process that a mathematical expectation is

equally described by a time average over an infinitely long time history, i.e., one realization of
the process. Analogously to the definition of stationarity, different types of ergodicity exist.
The process {X(t)} is called ergodic in mean if it is mean-value stationary and it holds

1 Z
2

µX ≡ lim x(t) dt (13.35)

T →∞ T
T
−2

Structural Dynamics 13 Random vibrations

summer term 2019 250
 Figure 13.7: from [Petersen 2000, p.1220, Fig. 34]

Furthermore, the process is called ergodic in second moment if it is second-moment stationary

and it holds
T

1 Z
2

RXX (τ ) ≡ lim x(t + τ )x(t) dt (13.36)

T →∞ T
T
−2

Also the process is called ergodic in covariance if it is covariant stationary and it holds
T

1 Z2
GXX (τ ) ≡ lim [x(t + τ ) − µX ] [x(t) − µX ] dt (13.37)
T →∞ T
− T2

For this case, we find the variance of the process by

T

1 Z2
Var [X(t)] = σX
2
= GXX (0) ≡ lim [x(t) − µX ]2 dt (13.38)
T →∞ T
− T2

13.3.5 The Gaussian process

The Gaussian process is a stochastic process {X(t)} for which every finite subset of random
variables that constitute the process {X(t1 ), X(t2 ), . . . , X(tn )} has a multivariate normal dis-
tribution. A special property of the Gaussian process is that it is fully defined by its mean
vector and covariance matrix. The PDF of the collection X = {X(t1 ), X(t2 ), . . . , X(tn )} for
any n can then be written as

1 1
 
fX (x) = q exp − (x − µX )T Σ−1
XX (x − µX ) (13.39)
(2π)d det (ΣXX ) 2

Structural Dynamics 13 Random vibrations

summer term 2019 251
where µX is the vector containing the mean values of all X(ti ) and ΣXX is the covariance
matrix.
Many physical phenomena are the result of the superposition of a large number of stochastic
processes, e.g. wind turbulences result from the superposition of different vortices or the
excitation due to waves results from the superposition of an infinite number of different surface
waves with different directions, phases and amplitudes.
The central limit theorem [Maybeck 1979] states (in short) that the sum of independent and
identically distributed random variables can be described by a normal distribution.
An important property is that a Gaussian load process [Bendat and Piersol 2010; Runtemund
2013] results in a Gaussian response process for linear systems. Then by the knowledge of
mean and covariance function, the stochastic response process is completely defined.

13.3.6 Frequency domain analysis of stochastic processes

Subsequently we are interested in a frequency domain analysis of stochastic processes. In order

to obtain a frequency domain representation we can apply the Fourier transformation. For
convenience we repeat the definition of the Fourier transformation, given in Eq. (3.34) and
Eq. (3.35).

Z∞
f˜(ω) = f (t)e −iωt dt (13.40)
−∞
∞
1 Z ˜
f (t) = f (ω)e iωt dω (13.41)
2π
−∞

We now apply the Fourier transformation to the process {X(t)} to define the new process
{X̃(ω)}.

Z∞
X̃(ω) = X(t)e −iωt dt (13.42)
−∞

The mean value of the Fourier transformed process {X̃(ω)} is given by

 ∞
Z∞

h i Z
µX̃ (ω) = E X̃(ω) = E  X(t)e −iωt
dt = E [X(t)] e −iωt dt = µ̃X (ω) (13.43)
| {z }
−∞ −∞ µX (t)

We note that the mean value of the Fourier transformed process is simply the Fourier trans-
form of the mean value function of {X(t)}. This only holds, if the Fourier transformation

Structural Dynamics 13 Random vibrations

summer term 2019 252
of µX (t) exist, which is not the case for stationary signals, because then the mean value is a
constant and thus, not absolute-value integrable. We will consider the analysis for stationary
signals hereafter. Furthermore, we consider the second moment function of the process in
the frequency domain. To this end, we modify the definition of the autocorrelation function,
because the process in the frequency domain is now in general complex:
h i
φX̃ X̃ (ω1 ,ω2 ) = E X̃(ω1 )X̃ ∗ (ω2 )
 ∞  ∞ ∗ 
Z Z
= E X(t1 )e −iω1 t1 dt1  X(t2 )e −iω2 t2 dt2  
−∞ −∞
 ∞ ∞ 
Z Z
= E X(t1 )X(t2 )e −iω1 t1 e iω2 t2 dt1 dt2 
−∞ −∞
Z∞ Z∞
= E [X(t1 )X(t2 )] e −iω1 t1 e iω2 t2 dt1 dt2
−∞ −∞
Z∞ Z∞
= φXX (t1 ,t2 )e −i(ω1 t1 −ω2 t2 ) dt1 dt2 (13.44)
−∞ −∞

Analogously, we find the expression for the covariance function

Z∞ Z∞
KX̃ X̃ (ω1 ,ω2 ) = KXX (t1 ,t2 )e −i(ω1 t1 −ω2 t2 ) dt1 dt2 (13.45)
−∞ −∞

This expression will also not exist, if the process is not second-moment stationary. In order
to circumvent the problem of undefined Fourier transforms for stationary signals, we present
an alternative procedure that applies a slight modification to the considered signals. To this
end, we consider a truncated version of the process {X(t)}, denoted by {XT (t)}:

T T
    
XT (t) = X(t) H t + −H t− (13.46)
2 2

where H(t) is the Heaviside function. This modified process coincides with the original process
in the limit T → ∞. For the truncated stochastic process, the covariance function reads
T T
Z2 Z2
KX̃T X̃T (ω1 ,ω2 ) = GXX (t1 − t2 )e −i(ω1 t1 −ω2 t2 ) dt1 dt2 (13.47)
− T2 − T2

Structural Dynamics 13 Random vibrations

summer term 2019 253
Here we made use of the knowledge that the covariance of the stationary process depends
only on the time difference τ = t1 − t2 and is given by GXX . According to Lutes and Sarkani
[2004], Eq. (13.47) can be reformulated, which yields an expression for the covariance function
KX̃T X̃T (ω1 ,ω2 ) that is bounded for all T including the limit T → ∞, but ω1 6= ω2 . It can be
further derived that for the case ω = ω1 = ω2 and in the limit T → ∞ the following holds

Z∞
lim KX̃T X̃T (ω,ω) = T GXX (τ )e −iωτ dτ (13.48)
T →∞
−∞

This expression grows proportionally with T for large T . To obtain an expression that exists
in the limit T → ∞, we define a new function through dividing KX̃T X̃T (ω,ω) by T :

1
SXX (ω) = lim KX̃T X̃T (ω,ω) (13.49)
T →∞ T

The function SXX (ω) is called autospectral density function or power spectral density or just
spectral density. By inserting Eq. (13.48) into Eq. (13.49), we find:

Z∞
SXX (ω) = GXX (τ )e −iωτ dτ (13.50)
−∞

Thus, the power spectral density is given by the Fourier transform of the autocovariance
function of the process in the time domain. For the zero mean process Z(t) = X(t) − µX
(µZ = 0), covariance function and autocorrelation function coincide and the spectral density
is equivalently given by

Z∞
SZZ (ω) = RZZ (τ )e −iωτ dτ . (13.51)
−∞

The inverse relation to Eq. (13.50) is given by

∞
1 Z
GXX (τ ) = SXX (ω)e iωτ dω (13.52)
2π
−∞

Structural Dynamics 13 Random vibrations

summer term 2019 254
 sinusoidal sinusoidal with stochastic
superimposed process with with broad band
random narrow band
fluctuations

Figure 13.8: Comparison of different processes from [Petersen 2000, p. 1228, Fig. 43])

Example 13.3: White noise process

A special case of a stochastic process is the white noise process. The white noise process
{X(t)} is the weakly stationary process with zero mean µX = 0, and the following
autocorrelation function

RXX (τ ) = GXX (τ ) = σX
2
δ(τ )

where δ(τ ) is the Dirac delta function that is equal to zero for all τ 6= 0. We can

Structural Dynamics 13 Random vibrations

summer term 2019 255
 calculate the spectral density of this process by evaluating Eq. (13.50):

Z∞ Z∞ Z∞
SXX (ω) = GXX (τ )e −iωτ
dτ = 2
σX δ(τ )e −iωτ dτ = 2
σX δ(τ )e −iωτ dτ
−∞ −∞ −∞

By application of the sampling property of the Dirac delta function, δ(x−a)f (x) dx =
R∞
−∞
f (a), we obtain

SXX (ω) = σX
2 −iω0
e = σX
2
,

which is constant throughout the entire frequency range and has an amplitude that
corresponds to the variance of the stochastic process. It should be noted that this
process is unphysical as it has an infinite energy content. Nevertheless it is often used
as an approximation in practical applications.

Figure 13.9: Autocorreation function and spectral density for the white noise process [Pe-
tersen 2000, p. 1228, Fig. 43])

13.3.7 Stochastic response of linear dynamic system

In the preceding section we derived expressions to describe the first two moment functions
of stochastic processes. Consider now the deterministic linear problem, where usually an
excitation is given from which we want to calculate the response of a system. In order to do
so, we describe the relation between excitation and response by differential equations. As was
discussed, e.g. in Chapter 7 for the linear SDOF system, the deterministic system response in
the frequency domain can be described by its transfer function H(ω),

x̃(ω) = H(ω)f˜(ω). (13.53)

This relationship can also be stated for the stochastic excitation and response processes {F̃ (t)}
and {X̃(t)}:

X̃(ω) = h̃(ω)F̃ (ω) (13.54)

Structural Dynamics 13 Random vibrations

summer term 2019 256
For a known excitation process, we now want to derive the first two moment functions of the
response process. Taking the expectation of Eq. (13.54) gives
h i h i h i
µX̃ (ω) = E X̃(ω) = E h̃(ω)F̃ (ω) = h̃(ω) E F̃ (ω) = h̃(ω)µF̃ (13.55)

For the case of covariant stationary processes, we now derive the spectral density function for
the response process. For this we first rewrite Eq. (13.49):

1 1 h  ∗ i
SXX (ω) = lim KX̃T X̃T (ω,ω) = lim E X̃T − µX̃ (ω) X̃T − µX̃ (ω) (13.56)
T →∞ T T →∞ T

Rephrasing the multiplication with the complex conjugate as the square of absolute values,
we find

1
 2 
SXX (ω) = lim E X̃T − µX̃ (ω) (13.57)
 
T →∞ T

Inserting Eqs. (13.54) and (13.55) into Eq. (13.57), we obtain

1
 2 
SXX (ω) = lim E h̃(ω)F̃ (ω) − h̃(ω)µF̃ 
T →∞ T

1
 2  2 
= h̃(ω) lim E F̃ (ω) − µF̃ 
   
T →∞ T
| {z }
KF̃ F̃ (ω,ω)
| {z }
SF F (ω)
 2
= h̃(ω) SF F (ω) (13.58)
 

The result shows that the spectral density of the response process is obtained by multiplication
of the excitation spectral density by the square of the absolute value of the transfer function.
From this, further response characteristics can be obtained, e.g. the response variance by
consideration of Eq. (13.52):
∞ ∞
1 Z 1 Z
Var [X] = GXX (τ = 0) = SXX (ω)e dω =
iω0
|H(ω)|2 SF F (ω) dω (13.59)
2π 2π
−∞ −∞

If the stochastic load process is a Gaussian process then the response process is also Gaussian
and fully characterized by the mean and covariance function derived in this chapter. The
following figure exemplarily shows the work flow for a SDOF system. Here we see how the
system “filters” the significant contributions of the load via its transfer function.

Structural Dynamics 13 Random vibrations

summer term 2019 257
 F (t)

m
w(t)
c k

time domain
F (t) w(t)

σF
t convolution σw
t

T T
1
RT
w(t)= m·ω F (t) sin[ωD (t−τ )]e−δ(t−τ ) dτ
D

frequency domain
T +∞
R2 R +∞
R
R(τ )= lim F (t)·F (t+τ )dt F̂ F (f )= F (t)e−i2πtf dt 2=
σw Sw (f )df
T →∞ −∞ −∞
−T
2
+∞ |F̂ F (f )|2
SF (f )= lim
R
SF (f )= R(τ )e−i2πτ f dτ T →∞ T
−∞
multiplication
2
Sw (f )=SF (f )·|Û (f )|
2
SF (f ) Sw (f )

Û (f )
 

1
f f f
fe fe

Structural Dynamics 13 Random vibrations

summer term 2019 258
14 Aeroelastic vibrations (wind)
Different types of wind excited vibrations are distinguished:
• By gusts induced vibrations (vibrations in wind direction)
• Vibrations due to vortex shedding (vibrations transversely to the wind direction)
• Self excited vibrations (stability problems)
• Galloping (vibrations transversely to the wind direction)
• Divergence and fluttering (vibrations transversely to the wind direction)
• Interference phenomena (interaction phenomena between different structures)
First the wind effects will be discussed phenomenologically. Then current standards are ad-
dressed.

14.1 General information

14.1.1 Loads

The basic value of all flow induced forces is the velocity pressure q (also termed dynamic
pressure):

ρ
q = v2 (14.1)
2

with the density of air ρ = 1,25 mkg and the wind velocity v. The force values W are obtained
3

from the velocity pressure through multiplication with the aerodynamic coefficients. These
coefficients are found in the literature or can be determined from the following relation.

W
cf = (14.2)
qA

Structural Dynamics
summer term 2019
259
where W is the measured resulting force of all flow induced pressure, pull and friction forces,
acting on the structure, and A is the loaded area. Analogously the moment coefficient is found.

M
cM = (14.3)
qAl

14.1.2 Description of wind

14.1.2.1 Wind velocity

The effect of different wind velocities is given through the Beaufort scale.

Beaufort Description Average wind speed

Land conditions
number of wind in m/s
0 calm Smoke rises vertically 0 - 0,4
1 light air Smoke rises almost vertically 0,4 - 1,8
2 light breeze almost felt on face 1,8 - 3,6
3 gentle breeze moves leaves and small twigs 3,6 - 5,6
4 moderate breeze raises dust and loos paper, 5,6 - 7,9
moves small branches
5 fresh breeze moves larger branches, inconve- 7,9 - 10,4
nience felt
6 strong breeze whistling heard 10,4 - 13,1
7 near gale whole trees in motion, 13,1 - 16,0
8 gale twigs break of trees, impedes 16,0 - 19,1
pedestrians
9 strong gale slight structural damage 19,1 - 22,1
10 storm trees uprooted, considerable 22,1 - 26,0
structural damage
11 violent storm widespread damage 26,0 - 29,7
12 hurricane devastation > 29,7

14.1.2.2 Temporal and spatial structure of wind velocity measurements

Fig. 14.1 depicts typical measurements of the wind speed at three different heights of a mast.

Structural Dynamics 14 Aeroelastic vibrations (wind)

summer term 2019 260
 u [m/s]
153 m 64 m 12 m

30
25
20
15
10
t [min.]
1 2 3 4 5
Figure 14.1: Measured wind velocities at three different heights [Holmes 2007, p. 56]

From this it can be deduced, e.g. [Holmes 2007, p. 55-56]:

• The average wind velocity increases with the height up to a limit value.
• Turbulences/gusts occur at all heights.
• Gusts are broadband, i.e., slow changes are superimposed with highly frequent vibrations.
• Low frequency vibrations occur similarly at all heights.
• The gustiness over smooth terrain (sea, flat land) is smaller than for rugged terrain
(cities) und decreases for increasing height [Stathopoulos and Baniotopoulos 2007, p.
3-4].

14.1.3 Temporal structure of the wind velocity

14.1.3.1 Power spectral density

[Van der Hoven 1957] investigated the frequency content of the wind and the weather system
in the atmospheric boundary layer on the basis of differently long measurements of the wind
velocity in the height of 80 m at the tower of the Brookhaven National Laboratory, New York.
Fig. 14.2 shows a typical wind spectrum, that is gained by assembling the different time scales.
A logarithmic frequency scale is chosen to cover the large frequency domain. In order to obtain
a measure of energy per frequency by the area under the depicted power spectral density, the
y-axis is scaled via the following relation:
Z ∞ Z ∞
σv2 = Gv (f ) df = f Gv (f ) d(ln f ) (14.4)
0 0

with the one sided power spectral density function Gv (f ) = 2S(f ). The wind spectrum can be
split into the “mesometeorological” and the “micrometeorological” range, with periods of over
and below one hour, respectively. In between lies the “spectral gap” with low energy input
(periods of under a day to around ten minutes).

Structural Dynamics 14 Aeroelastic vibrations (wind)

summer term 2019 261
 mean wind velocity turbulence
Anual peak
f S(f) (speculativ not
measured)
Synoptic peak
(passage of weather
systems)

Turbulent peak
Diurnal peak
(day-night)

Spectral gap
log f
-4 -3 -2 -1
frequency 1/h 10 10 10 10 1 10 100 1000
period in h 10000 1000 100 10 1 0,1 0,01 0,001

1 year 4 days 0,5 days 5 min. 1 min. 5 sec.

mesometeorological micrometeorological

Figure 14.2: Frequenzspektrum der horizontalen Windgeschwindigkeit (1957), Messungen in

Brookhaven, New York [Burton et al 2001, p. 12], [Petersen 2000, p. 596]

In wind engineering, in particular the higher frequency parts in the “micrometeorological”

range are of interest. These can be represented as stochastic processes with zero mean value
on the basis of the power spectral density [Burton et al 2001, p. 11ff].

14.1.3.2 Turbulence intensity

The fluctuating wind velocity v is described by the turbulence intensity, the power spectral
density and the correlation lengths. The turbulence intensity is a measure for the standard
deviation σ in the “micrometeorological” range.
s
1ZT
sZ
∞
σV = v(x,y,z,t)2 dt = Gv (f ) df (14.5)
T 0 0

with the averaging time T , which results in the turbulence intensity

σV
IV = . (14.6)
µV (z)

14.1.4 Spatial structure of the wind velocity

14.1.4.1 Atmospheric boundary layer - wind profile

Due to the “spectral gap”, for sufficiently long averaging times, e.g., T = [10,60] min, the
average value µV (z,t) becomes time invariant (mean value stationary), and thus only depends

Structural Dynamics 14 Aeroelastic vibrations (wind)

summer term 2019 262
on the height z. The mean wind velocity increases with height up to a surface independent
limit value. This is due to the friction on the ground surface.
The “gradient wind velocity” thus obtained results from air pressure differences in the atmo-
sphere, that are linked to the different solar heating of the earth [Mendis et al 2007]. The
sphere of influence of the surface friction is called atmospheric boundary layer. The height
of this layer zg (gradient height) depends on the surface roughness. The dependence of the
average wind velocity on the height can be either represented through a logarithmic relation
or through an empirical exponential approach. The first one can be analytically derived from
boundary layer considerations (z.B. [Holmes 2007, p. 56ff.]). The exponential approach results
in
!α
z
µV (z) = µV (zref ) (14.7)
zref

with the reference height zref = 10 m. The exponent α [Balendra 1993] depends on the surface
roughness. It is related to the roughness length z0 of the logarithmic approach α = ln(zref/z0 )−1
[Holmes 2007, p. 58-59]. The dimensionless “surface drag coefficient” κ relates
√ the velocity at
the ground v∗ to the average wind speed at the height of 10 m via v∗ = κµV (10).

ground conditions roughness length z0 [m] surface drag coefficient κ [-]

very flat ground (snow, desert) 0.001 - 0.005 0.002 - 0.003
open ground (grasland, few trees) 0.01 - 0.05 0.003 - 0.006
suburban areas (buildings 3 - 5 m) 0.1 - 0.5 0.0075 - 0.02
dense cities (buildings 10 - 30 m) 1-5 0.03 - 0.3
Table 14.1: Surfaces, roughness lengths and surface drag coefficient [Holmes 2007, p. 58]

The distribution of the temporal mean value of the velocity is thus described, e.g. by the
following relation:

z [m]

I III
100

50

10
5
h i
m
v s
20 40 60

Structural Dynamics 14 Aeroelastic vibrations (wind)

summer term 2019 263
 α
z

v(z) = v10m ≥ 0.9v10m
10

Here, v10m is the 10 minute time average, that is reached or exceeded in the statistical average
in 50 years and occurs at 10 m height.
Fig. 14.3 shows the the decay for different surface roughnesses.

z[m]
α=0.4 %
100
k=0.05
%
400 100
89
α=0.28
%
k=0.015 α=0.22 100 %
77 90 k=0.009
94 α=0.16 100
200 k=0.005

61 76 82 91
42 59 67 79

Figure 14.3: aus Ruscheweyh, Dynamische Windwirkung an Bauwerken Vol. 2 (1982), S. 16,
Abb. 2.3

14.1.4.2 Turbulences

Turbulences are caused by differently large vortices, that are transported with the mean wind
velocity and showing a time structure with higher frequencies. Geoffrey I. Taylor (1939)
proposed, that–under certain conditions–the statistical composition of the turbulences can be
understood as a “frozen pattern”, that is transported with the mean wind velocity (Taylor’s
frozen-in turbulence hypothesis [Holm 2005]). It is based on the assumption, that the changes
of the vortices are slow compared to the wind velocity. On this basis, the size (wave length λ)
of the vortices is related to the excitation frequency f and the mean wind velocity µV (z):

λ = µV (z)/f [m]. (14.8)

14.2 Wind induced vibrations

Along with the static wind force, various dynamic effects have to be considered.

Structural Dynamics 14 Aeroelastic vibrations (wind)

summer term 2019 264
14.2.1 Gust induced vibrations

Due to included gusts within the flow, the wind velocities are distributed unevenly over the
height. It can then be composed of two parts. One part is the static part vm (z) that is constant
over time according to Fig. 14.4 and a superimposed, randomly variable part v(z,t).

vges (z,t) = vm (z) + v(z,t)

vm (z)

v
Figure 14.4: Static and dynamic part of the wind velocity

This distribution can be used for slender buildings. For a structural element with sufficiently
low width in the flow1 , the corresponding wind force per unit length is:
ρ 2
q(z, t) = cf vges (z,t)
2
ρ 2
= cf (vm (z) + 2vm (z)v(z, t) + v 2 (z, t))
2
ρ 2 ρ
= cf vm (z) + cf ρvm (z)v(z,t) + cf v 2 (z, t) (14.9)
2 2

with

cf ρ2 vm (z)2 static term

cf ρvm (z) v(z,t) linear dynamic term

cf ρ2 v(z,t)2 non-linear dynamic term (is usually neglected)

The frequency content of the excitation is given by the power spectral density S(f ) or S(T ) of
the wind. Fig. 14.2 exemplarily shows such a power spectral density. For the “mesometeoro-
logical” range (comp. Fig. 14.2) there are different approaches. Some of those are summarized
in Tab. 14.2. The differences are exemplarily depicted in Fig. 14.5.

1
Sufficiently small in the sense, that the same wind speed can be assumed over the whole width.

Structural Dynamics 14 Aeroelastic vibrations (wind)

summer term 2019 265
 2
Non−dimensional wind velocity spectra fG(f)/σ , z = 10 [m]

Kolmogorov (z)
Kolmogorov (L)
Kareem
Simiu
Harris
Davenport
Kaimal
−1
10 von Karman
ESDU

−2
10
−2 −1 0 1 2
10 10 10 10 10
2
Wind velocity spectra G(f) in [(m/s) /Hz], z = 10 [m]
700

600

500

400

300

200

100

0
0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04 0.045 0.05
f [Hz]

Figure 14.5: Comparison of different models: z = 10 m, v̄10 = 20 m s , z0 = 0.05 (dimensionless),

κ = 0.006 (open grassland, grassland); above: log-log illustration of the dimensionless
turbulence spectra; below: Comparison for low frequencies [Runtemund 2013]

Exemplarily the approach of Davenport is explained. The dynamic additional velocities for
the gusts are here given by

2 ξ2
Sv (f ) = I(z)2 v 2 (z) , (14.10)
3 (1 + ξ 2 )4/3

with the frequency f in Hz, the average wind velocity v in m

s , the dimensionless frequency

Structural Dynamics 14 Aeroelastic vibrations (wind)

summer term 2019 266
 Kolmogorov
application limitations basis
limit

off-shore structures; land empirical, strong wind

Kaimal based structures with low z < 200 [m] fulfilled conditions, flat homoge-
surface roughness neous terrain

Kareem off-shore structures adjustable parameters empirical

atmospheric turbulences
von
(e.g. airplane construc- z ≥ 200 [m] fulfilled theoretically justified
Kármán
tions), wind tunnel tests

ESDU empirical modification of the von Kármán spectrum with adjustable parameters

off-shore Strukturen;
landgestützte Strukturen
Simiu z < 50 [m] underestimated empirisch
mit geringer Geländer-
auigkeit
empirical, mean values
Davenport land based structures overestimated
based on different mea-
f > 0.01 [Hz] surements on land at dif-
Harris land based structures underestimatedferent heights

Table 14.2: Comparison of different approaches from [Runtemund 2013]

coefficient

1200
ξ= f, (14.11)
v10m

and the dimensionless turbulence intensity

−2α
z

I (z) = 6k
2
. (14.12)
10

Here z is the height in m and k is the roughness coefficient. It ranges from k = 0,05 in cities
to k = 0,005 in open land.

Structural Dynamics 14 Aeroelastic vibrations (wind)

summer term 2019 267
 Sv (f )

f [Hz]
10 −2
10 −1
1 101

Figure 14.6: Example for Davenport spectrum.

The power spectral density of the dynamic load, neglecting the non-linear term in Eq. (14.9)
is given by:
2
ρ

Sq (f ) = cf 2vm,z Sv (f ) (14.13)
2

For a small structure, with equally distributed wind pressure, the vibration response can be
determined from this excitation (Eq. (13.58)).
 2
Sw (f ) = h̃(f ) Sq (f ) (14.14)
 

Under the assumption of a stationary Gauß process, by the above a probabilistic description
of the response is given. It allows to develop design concepts, e.g. for a gust factor. For this,
a target probability for the case, that during an intended usage period a specified response
value is not exceeded, is specified. For linearly shaped or extensive structures it further needs
to be considered, that pressures are unevenly distributed over length- and cross-dimension of
the structure. With the help of a modal superposition, design calculation concepts of practical
applictions, e.g., cantilever systems, have been derived.

14.2.2 Vortex induced vibrations

Due to vortices, periodic forces result at structures that form obstacles in the flow, which are
initially at rest and loaded by a flow. These forces predominantly act transverse to the flow
direction. They can induce forced vibrations. For coincidence of the vortex frequency and
the eigenfrequency of the structures, the risk of resonance exists. This is often the case for
chimneys, masts, ropes, and pipes. There is also the risk of resonant deformations, as far as
mode shapes of the structures, that are connected to ovalization, are excited in resonance, e.g.
ovalization of not sufficiently stiff cylinders.

Structural Dynamics 14 Aeroelastic vibrations (wind)

summer term 2019 268
 Figure 14.7: Vortices, from [Petersen 2000]

14.2.2.1 Occurence of vortices

In order to describe the character of the flow and the occurrence of vortices, the Reynolds
number is introduced:

vd
Re =
η

which is dimensionless, with

m
v wind velocity in
s
d diameter of the structure in m
η = 15 · 10−6 m2 /s kinematic visosity in air,

The following cases can be distinguished:

• Subcritical: 40 < Re < 3,5 · 105
laminar flow in the boundary layer, before the cross section maximum, i.e., widest pos-
sible cut through the cross section, orthognoal to the flow direction, stable vortex trail
⇒ large transverse forces
• Supercritical: 3,5 · 105 < Re < 7 · 106
turbulent flow in the boundary layer, vortex shedding due to shedding and re-application
on the cross-section far behind the maximum of the cross-section, disordered wake flow
with irregular vortices and low correlation of the lateral pressures
⇒ low transverse forces

Structural Dynamics 14 Aeroelastic vibrations (wind)

summer term 2019 269
 Re < 5 5 < Re < 40

laminare Strömung kleine symmetrische Wirbel

keine Wirbel keine Ablösung

40 < Re < 3,5 · 105 Re > 3,5 · 105

Stabile unsymmetrische Wirbel hochturbulente Wirbel mit unregelmäßiger

mit periodischer Ablösung Ablösung, für Re > 3,5 · 105 − 7 · 106 nahezu
(unterkritisch) periodischer Nachlauf (über-/transkritisch)

Figure 14.8: Different states of vortex shedding

• Transcritical: 7 · 106 < Re

Vortex shedding again close to the cross-section maximum, but again periodic vortices
in the wake flow
⇒ again increasing transverse forces
Below a Reynolds number of 40, no significant vortex shedding occurs.
Fig. 14.8 shows states of vortex shedding, from the laminar to the transcritical flow.

14.2.2.2 Frequency of vortices

The frequency f of the alternating force resulting from the vortices acting on the structure is
described by the dimensionless Strouhal number S:

fd
S= (14.15)
v
Sv
→ f= . (14.16)
d

Thereby f corresponds to number of vortices created per second at one side of the obstacle.

Structural Dynamics 14 Aeroelastic vibrations (wind)

summer term 2019 270
14.2.2.3 Forces related to vortices

The force in transverse direction can be determined with the help of a lateral aerodynamic
coefficient cs . It results to
ρ
F (t) = v 2 cs dg(t) (14.17)
2

The lateral aerodynamic coefficient cs can, e.g. be determined in wind tunnel testing. The
function g(t) is a periodic function with amplitude 1 and the fundamental excitation frequency
f.

14.2.2.4 Assessment of vortex induced vibrations

Through measurements the dependency of the Strouhal and the Reynolds number can be
determined. For cylindrical structures, the Strouhal number takes values of around 0.2 and is
almost independent from the Reynolds number.

St, clat

irregular shedding
0,6

clat
0,4
St
0,2

102 103 104 105 106 107 Re

subcritical supercritical transcritical

Typically, the Strouhal number and the lateral aerodynamic coefficient cs are illustrated in
simplified diagrams. The following figure shows the relation exemplarily for a cylindrical
profile.

S,cs

0.7 cs
0.2 S
Re
10 4
105
10 6 10 7

Structural Dynamics 14 Aeroelastic vibrations (wind)

summer term 2019 271
By setting the excitation frequency f equal to a specific natural frequency fn , the critical wind
velocity can be determined.

dfn
vcrit =
S

The corresponding amplitude of the transverse force per unit length of a line structure follows
to:
ρ
p k = v 2 cs d
2

Through the resonance, the “static” response is amplified by the factor (comp. Tab. 5.1):

π
V =
δ

The risk of resonace exists, whenever the critical wind velocity falls within the range of possible
wind velocities. If this is the case, the resulting response has to be determined and the
structural reliablity has to investigated with respect to the response.
If the structure loaded by wind is not held at rest, i.e., vibrations of the structure occur,
a so called lock-in effect can be of significance. Here, an influence of the periodic motion
of the structure on the shedding frequency is observed. If a structure is excited by a wind
velocity, for which correspondence between the shedding frequency of the vortices and system
eigenfrequency (resonance) exists, and the wind velocity changes, the resonance state is not
lost. The lock-in effect causes (as long as the change in wind velocity is not too large) that
the shedding frequency of the vortices for larger vibration amplitudes is fixed at the previous
frequency of the structure, i.e., at the eigenfrequency of the structure.
Thereby, the excitation frequency and thus the resonance effect is maintained independently
of wind velocities within a certain range.
Constructive measures can be applied, e.g., the arrangement of spirals. These disturb the
regularity of the the shedding and simultaneously lead to an increase of the force coefficient
in the flow direction.

Structural Dynamics 14 Aeroelastic vibrations (wind)

summer term 2019 272
 S·v
d
a) b) amplification

vibration amplitude y/d

function

fe vortex excitation
region of
synchronization

Related
Λ = 0,01

vcrit
Wind speed v Reduced wind speed vr
Figure 14.9: vgl: Bachmann, Hugo, Vibration Problems in Structures (1997), in: Vibrations in across-
wind direction induced by vortex-shedding, S. 203, Abb. H.10

wire rings scruton

helix helix
disturbance shrouds
stripes

14.2.3 Galloping

Galloping describes self excited vibrations. It only occurs, when the system is already in
motion due to inevitable flows. Thereby flow induced forces are caused that can stimulate the
motion further, until a non-linear increasing damping is able to dissipate the energy introduced
within on period of motion. This is then typically linked with large amplitudes.
For circular cross-sections due to the full symmetry, such an excitation will not occur. Excep-
tions from this are asymmetries that occur, e.g. through ice cover or the flow of water in case
of rain.
The situation is depicted here for the example of a half circle cross section. In a coordinate
system attached to the moving structure the flow is acting at the angle α.

Structural Dynamics 14 Aeroelastic vibrations (wind)

summer term 2019 273
 mẅ
kw
cẇ
vwind d Fh
α
ẇ
Fv Faero (t) flow force
vres
vres closer streamlines:
α
ẇ higher velocities,
vwind thereby lower pressures

The force component in vertical direction depends on the relative flow angle α. It is determined,
e.g. in wind tunnel testing and given in terms of a aerodynamic coefficient cs (α) that depends
on the flow angle α:
ρ 2
Faero (α) = qdcs (α) = vwind dcs (α) (14.18)
2

The aerodynamic coefficient cs (α) is shown for two typical cross-section in the following figure.
Therein it is assumed that for α = 0 no transversal force Faero occurs.

cs cs

dcs
|
dα α=0
>0

α α
dcs
|
dα α=0
<0
Figure 14.10: Examples for the aerodynamic coefficient cs (α)

If the oscillation is idealized by a SDOF system, the governing differential equation are given
as:

mẅ + cẇ + kw = Faero (α) = cs (α)qd (14.19)

As cs (α) is a nonlinear function, the above equation is also non-linear. To determine, whether
a critical stimulation of vibrations can occur, in a first step the system’s behavior for small
deformations is investigated. Then, the coefficient cs (α) can be approximated by a Taylor

Structural Dynamics 14 Aeroelastic vibrations (wind)

summer term 2019 274
series. Then only the linear term is retained and cs (α) is linearized at α = 0:

dcs 

Faero (α) ≈ αqd (14.20)
dα α=0


With

ẇ
α ≈ tan α = (14.21)
vwind

we obtain

dcs 

ẇ ρ 2
mẅ + cẇ + kw = v d (14.22)
dα α=0 vwind 2 wind



dcs 
 !
ρ
→ mẅ + c − vwind d ẇ + kw = 0 (14.23)
dα α=0 2


| {z }
c̃

The expression c̃ characterizes the “damping”. Whenever c̃ results in positive values, a dif-
ferential equation results, that corresponds to a damped SDOF system. If c̃ takes negative
values, the motion is not attenuated but amplified. Then the vibration is self excited and
galloping occurs.


If dcs 
dα 
 ≤ 0, the modified damping c̃ cannot result in negative values and galloping can be
α=0 

excluded. If dcs 
dα 
 > 0, stimulation occurs occurs from a critical wind velocity on.
α=0

2c

dcs  ρ
c≥ vcrit d → vcrit = (14.24)
dα α=0 2
 
 dcs 
 ρd
dα  α=0

Galloping can occur for different cross-sections.

turbulent

Figure 14.11: Profiles at which Galloping can occur

Structural Dynamics 14 Aeroelastic vibrations (wind)

summer term 2019 275
In [DIN 1055-4 2005; EN 2010] galloping vibrations are treated in appendix E.

14.2.4 Divergence

Divergence is not a vibration problem. This phenomenon is described by “static tilting” that
occurs, whenever the torsional moment, induced by the flow on the structure, linked with
the rotation of the cross-section exceeds the reaction moment of the structure. In this case a
stability loss can occur. Usually, the critical wind speed for static divergence in civil engineering
applications is very high. It is higher than the velocities that are usually considered in the
design process.

14.2.5 Flutter

Fluttering of bridges is a stability problem, just as galloping. Small deformations of the system
(often caused by vortex shedding) are a precondition for the stimulation.

w
α

The relevant wind forces result from flow induced deformations due to bending and torsional
motion of a bridge.
Here, the flutter phenomenon is depicted in the following figure, showing the cross-section
of a bridge vibrating simultaneously in a torsional and bending mode hape (two degrees of
freedom). If the phase of the two degree of freedoms are coupled in such a way that the integral
of the work done on the self-excited, vibrating system over one period is larger than zero, an
amplification of the vibration occurs.

Structural Dynamics 14 Aeroelastic vibrations (wind)

summer term 2019 276
 Figure 14.12: Flutter, [Ruscheweyh 1982, Vol. 2 (1982), S. 118, Abb. 4.67]

In the above illustration translatory and rotatory motions are in phase. In the first quarter of
the motion the wind does positive work on the system, as the direction of force and motion
coincide. In the second quarter of the motion force and motion are counter-directed, such that
the force does negative work on the system. Over the whole period, the energy in the system
is thus not increased.
In the lower illustration the translatory and the rotatory motions are phase shifted, such
that for every instance in time the wind direction corresponds to the direction of motion and
positive work is done. Further information can be found in Klöppel / Thiele 2 .
Investigations based on potential theory, as for aircrafts, are hardly accurate for typical bridge
cross-sections. Therefore, usually wind tunnel testing is necessary. Instead of a model of a
miniature bridge, only a bridge segment, supported by springs (displacement and rotation), is
investigated. In order to obtain valid results, it is necessary that the vibration of the bridge
girder is not significantly affected by the motion of other structural parts, e.g. the pylons.
Necessary data, such as the mass, the rotatory inerty, the bending and torsonial frequencies
and the detailed cross-section design are prerequisites for the investigation.
Compared to the experimental values from Klöppel / Thiele significant differences in the critical
wind velocity can result from details in the constructive detailing, such as edge smoothing and
railing perforation, etc.. Therefore the data given in these publications are especially useful
for a predesign. As long as the risk of flutter cannot be excluded, wind tunnel test have to be
performed.
2
K. Klöppel/F. Thiele, Modellversuche im Windkanal zur Bemessung von Brücken gegen die Gefahr winder-
regter Schwingungen, 1967, Der Stahlbau, 32, 353-365

Structural Dynamics 14 Aeroelastic vibrations (wind)

summer term 2019 277
14.2.6 Interference effects

Vortices behind structures can induce excitations to the structures behind those buildings.
The details can only be determined in wind tunnel tests.
For a series or parallel arrangement, the proof has to be done that the following interference
effects do not occur:
• interference effects for transverse vibrations due to vortices,
• interference galloping,
• classical galloping for coupled cylinders.

Structural Dynamics 14 Aeroelastic vibrations (wind)

summer term 2019 278
15 Earthquakes

15.1 General information and terms

Earthquakes result from fractures in the lithosphere (earth crust). The thick, brittle plates
with a thickness ranging from a few to 200 km float on the comparably “liquid” asthenosphere.
The plates are in constant movement, which is termed continental drift. The fracture zone of
the lithosphere is called hypocenter. The projected point on the surface is called epicenter, as
shown in Fig. 15.1.
Often the hypocenter is covered by a sediment layer that is many kilometers thick. Caused
by the fracture movement of the crust, compression waves (P-waves), shear waves (S-waves)
and Rayleigh waves (R-waves) occur. These propagate with different propagation velocities
(cP > cS > cR ) and are responsible for the movement of structures on the earth surface. The
epicenter depth can be estimated from the arrival time difference between the P - and the
S-waves.
The waves can be attenuated or even amplified as they propagates through the soil toward the
structures. In particluar, the properties of top 30 meters of the soil can strongly influence the
amplitude and the frequency content of the seismic excitation. This was evident during the
earthquake in Mexico City in 1985. Even though the hypocenter was 400 km far away from
Mexico City, enormous damages occurred. The city is built upon a 30 − 40 m thick, very soft
clay layer. The amplification of the earthquake acceleration due to the presence of the soft clay
layer especially in the range of around 2 Hz corresponded very well with the eigenfrequency
of the buildings with 8-12 floors. On the stiff soil outside of the inner city, almost no damages
occurred

Structural Dynamics
summer term 2019
279
 epicenter epicentral
intensity I0 distance
∆ building
intensity I

depth of
sediment layer

focus h
fault
hypocentral
hypocenter (focus) distance s
depth of focus

isoseismal curves epicenter

I0 earth surface

I=I0 −1
I=I0 −2
I=I0 −3

Figure 15.1: Terminology of an earthquake, from [Flesch 1993, p. 203, Fig. 7.1]

The magnitude and intensity are used to characterize the strength of an earthquake.

15.1.1 Magnitude

The energy release of an earthquake can be described by the magnitude M on the Richter scale.
The largest magnitude, that was ever measured, was 9,5 (Validivia, Chile). The magnitude is
described by the maximum displacement smax that is measured 100km away from the epicenter.
An empirical relation between the energy released in an earthquake and the magnitude is given
by

lg(E) = 11.8 + 1.5M

The difference ∆M = 1 then corresponds approximately to a factor 31 in the energy release.

Here, the energy is given in the non SI-conform unit [E] = 10−7 J = 2,8 · 10−14 kWh. An
earthquake of magnitude 9 thus has an energy release of around 6 · 1011 kWh. Considering the
average energy consumption of 50.000 kWh per person and year in highly civilized countries,
this means, that with the energy of such an earthquake around 10 million people could be
supplied for one year. Furthermore, a relation between the magnitude and the length of the
fracture zone (in km) can be empirically given as

M = 5.65 + 0.98 lg(l0 ).

Structural Dynamics 15 Earthquakes

summer term 2019 280
Then, M = 9 corresponds to a length of l0 > 1000 km.
Fig. 15.2 shows the frequency of occurrence of earthquakes with a certain magnitudes per
year.

103
102 a
101
Earthquake/year

100 b
10−1
10−2
−4 max M ≈8.75
10−3 10
10−4 max M ≈6.25
10−5
10−6
2 3 4 5 6 7 8 9 10
Magnitude M
Figure 15.2: Frequency of occurrence of magnitudes for the whole earth (a) and the Rhine valley (b)
from [Flesch 1993, p. 203, Fig. 7.2]

15.1.2 Intensity

Earthquakes can classified according to their consequences on nature, humans and structures
(similar to the Beaufort scale for wind). This is done by the modified Mercalli-scale (MM
scale) or the Medvedev, Sponheuer, Karnik scale (MSK scale). An earthquake thus has only
one magnitude but different intensities from place to place. The intensity is a measure for the
perception and damaging effect of earthquakes. The relations between magnitude M , intensity
I and soil acceleration a can be approximately described empirically. For Germany the relation
between the horizontal soil acceleration a at the surface of the earth and the intensity I can
be described as follows

8 − I cm
lg a = 2 −
2.4 s2

There are a number of empirical relations between the magnitude M and the epicenter intensity
I0 , e.g., for Austria:

M = 0.7 I0 − 0.1

Structural Dynamics 15 Earthquakes

summer term 2019 281
and Switzerland

M = 0.67I0 + 2.3 log (focal depth in km) − 2.0

15.1.3 Calculation procedures

For the dynamic investigation of a structure subjected to an earthquake excitation, the soil
acceleration is applied as root point excitation. This typically shows an arbitrary time history,
which is referred to as accelerogram.

Figure 15.3: Exemplary accelerogram given in terms of the acceleration relative to the gravitational
acceleration on earth g .

Horizontal and vertical soil accelerations occur simultaneously. Usually, the horizontal move-
ments are more important, as building structures are typically designed for comparatively low
horizontal forces. The maximum vertical accelerations can be assumed to the in the order of
0,3 - 1,0 times the maximum horizontal acceleration. In the Swiss code SIA 160 a ratio of 2/3
is given, the German code DIN 4149 gives the ratio 0.7.
Essential for the demand, are the acceleration, the frequency spectrum of the accelerogram and
the duration of the loading. Exemplarily, a few numbers are given from Bachmann [2013]
Maximum values of the horizontal acceleration:
· Friaul, 1976, 15 km distance from epicenter 0,37 g
· Mexico, 1985, on rock 0,04 g
· Mexico, 1985, in the city (soft soil layers) 0,17 g
· Nortridge, 1994, Southern California 1,82 g
Frequency ranges of the soil acceleration:

Structural Dynamics 15 Earthquakes

summer term 2019 282
 · stiff soil, rock 3 - 10 Hz
· medium stiff soil 2 - 8 Hz
· soft soil 0,5 - 2 Hz
· Mexico, very soft soil 0,3 - 0,5 Hz
Duration:
·M=6-7 5 - 20 s
· Mexiko 1985 80 s
Different calculation methods can be used depending on the task to investigate the seismic
response of structures
• Approaches based on accelerograms:
– Frequency-domain analysis (only linear)
– Time-domain analysis (time history method; also non-linear)
Using a statistically relevant number of accelerograms, several nonlinear transient
analysis are carried out and the maxima of the forces and displacements are eval-
uated with averaging methods. The Eurocode 8 suggests a minimum number of
analysis equal to 8.
• Linear static methods
– Equivalent static design
– Response spectrum analysis (RSA)
• Nonlinear methods:
– Nonlinear static analysis
– Response history analysis (RHA)
(via modal analysis or direct integration methods)
– Capacity design based the hierarchy of strengths

15.2 Response spectrum analysis (RSA)

15.2.1 Introduction

In practical applications often the response spectrum analysis is applied in the earthquake
design of structures. This approach is only valid for linear system, nevertheless it is also often
applied for non-linear problems.

Structural Dynamics 15 Earthquakes

summer term 2019 283
 w

fg (t)
wg wr

The linear equation of motion for the root-point excited SDOF system with the relative dis-
placement wr and the soil displacement wg is given by:

m(ẅg (t) + ẅr (t)) + c ẇr (t) + k wr (t) = 0 (15.1)

or, equivalently

m ẅr (t) + c ẇr (t) + k wr (t) = −m ẅg (t) . (15.2)

| {z }
Fg (t)

The right hand side can be seen as seismic excitation Fg (t). The system response can now be
determined numerically using the Duhamel-integral

Zt
wr (t) = − mẅg (τ )h(t − τ ) dτ ,
0

or with other time integration methods. After dividing Eq. (15.2) by the mass m, it can be
observed that the system
q properties can be completely described using the definition of the
natural frequency (ω = k/m and δ = c/2m) and the damping.

ẅr (t) + 2δ ẇr (t) + ωE2 wr (t) = −ẅg (t)

With this approach, the maximum relative response values for different systems can be ob-
tained in dependence of the natural frequency ωE and the damping ratio ς = D = δ/ωE for a
given accelerogram ẅg (t).

wr,max (ωE , ς) = Swr (ωE , ς)

For this, the Duhamel integral (or equivalent time integration methods) is applied to evaluate
the maximum response of the SDOF system for different natural frequencies and damping
ratios, as shown in Fig. 15.4.

Structural Dynamics 15 Earthquakes

summer term 2019 284
 Swr (ωE,i ,ςi )

ẅg
Figure 15.4: SDOF system with different natural frequencies and damping ratios for the response
spectrum analysis

Under the assumption of harmonic responses, ẅr,max occurs at the same time as wr,max = Swr 1 ,
that is when ẇr = 0, and the following holds:

ẅr,max = −ωE2 Swr

and

|ẅr,max | = Sa (ωE , ς) = ωE2 Swr (ωE , ς)

or

2π
 2
|ẅr,max | = Sa (TE , ς) = Swr (TE , ς)
TE

The latter is also called “pseudo absolute acceleration”. The word “pseudo” refers to the
fact that the used formulas are actually only valid for linear harmonic vibrations, but are
applied here for non-harmonic vibrations introducing an approximation. It is also important
to notice, that only the absolute values of the maximum accelerations are accounted for,
therefore the information about the phase is lost. The following figures shows response spectra
for absolute and pseudo-absolute acceleration, relative and pseudo-relative velocity and relative
displacement. Absolute and pseudo-absolute acceleration differ only slightly, whereas relative
and pseudo-relative velocity show larger differences.

1
This assumptions is an approximation and holds for linear and harmonic, i.e., sinusoidal vibrations.

Structural Dynamics 15 Earthquakes

summer term 2019 285
 pseudo acc.
acc.

m/s2

pseudo relative vel.

relative vel.

m/s

relative displ.

cm

Figure 15.5: Response spectra for absolute and pseudo-absolute acceleration, relative and pseudo-
relative velocity and relative displacement

For SDOF systems, the response spectra can be directly used. Under certain conditions on
geomtery and material properties, buildings can be idealized as SDOF system. For more
complicated system–when applying the linear elastic response spectra–the modal superposi-
tion analysis can be used. A direct superposition of the modal responses would imply, that
the maxima of all modal vibrations occur at the same time, thus leading to unrealistically
high displacement and internal forces. Therefore the maximum value of quantities such as
displacements and forces are calculated as the root of the square sum:
r
Smax = (15.3)
X
Si2

with

Smax Resultant
Si Resultant corresponding to the -thi eigenform

Structural Dynamics 15 Earthquakes

summer term 2019 286
In reality loads from earthquakes lead to plastic deformations in structures, which dissipate
energy when limited to a certain controlled value. Therefore the application of the linear RSA
is a very conservative method. For a more realistic analyis, the response spectrum is reduced
to a factor that accounts for the structure ductility.2 .

This app visualizes the response spectrum method according to DIN 4149. It
calculates the maximum displacement of a building with three floors under
earthquake excitation.
http://www.bm.bgu.tum.de/lehre/interactive-apps/
response-spectrum-analysis/

15.2.2 Design codes

Relevant codes are:

• Eurocode 8 - Seismic Design of Buildings
• DIN 4149 (April 2005) - Building in German seimisc areas
• KTA - Guidelines for nuclear power plants
DIN 4149 allows calculations based on a SDOF system and the response spectrum method
using a linear elastic structural model and the design spectrum for the design of building
constructions, if the building has a regular structure in the layout and elevation and a soil
dependent, maximum fundamental period between 0,8s and 2s (DIN 4149, chapter 6.2.2). The
subsurface conditions are specified in DIN 4149, chapter 5.2.
The design codes made available design response spectra for seismic design purposes. The
design response spectra are simplified functions that envelope a large number of response
spectra for recorded seismic input motions. For each specific site, the probabilistic evaluation
of the seismic history, also called Probabilistic Seismic Hazard Analysis (PSHA), provides
a statistically significant number of response spectra for events with a certain probability
of occurrence in 50 year. From this response spectra population, an enveloping function is
estimated and the design spectra is obtained. Therefore, for each probability of occurrence,
a design spectrum can be derived. The smaller the probability of occurrence, the stronger is
the event for which the structure is design. The probability of occurrence can be expressed
also in terms of return period of the event. According to the DIN 4149, the probability of
occurrence of the design earthquake for residential buildings is 10 % within 50
years. This corresponds to a return period of 475 years.

2
In DIN 4149 the non-linear effects of the building material, load-bearing system and constructive details are
considered by a coefficient q.

Structural Dynamics 15 Earthquakes

summer term 2019 287
The elastic response spectrum from DIN 4149 for the linear calculation under consideration
of the hysteretic damping has the following form3 :

Se (T )
β0 B C
ag · γI · S · q

D
ag · γI · S A

T
TB TC TD
Figure 15.6: Design response spectrum according to the DIN 4149.

" !#
T β0
0 < T ≤ TB : Se (T ) = ag · γI · S · 1 + −1
TB q
β0
TB < T ≤ TC : Se (T ) = ag · γI · S ·
q
β0 TC
 
TC < T ≤ TD : Se (T ) = ag · γI · S · ·
q T
β0 TC · TD
TD ≤ T : Se (T ) = ag · γI · S · ·
q T2

with

Se (T ) ordinates of the design spectrum as a function of the structure natural period T

ag design ground acceleration on type A ground 0 ≤ ag ≤ 0,8 sm2

(which depends on the design seismic zone)

γI importance factor of the building 0,8 ≤ γI ≤ 1,4 (high value for buildings of
importance for the general safety, such as hospitals or emergency centers)

S soil factor 0,75 ≤ S ≤ 1,5

3
In DIN 4149 two expressions for damping effects are chosen. Here, we restrict to the presentation of the
behavior factor.

Structural Dynamics 15 Earthquakes

summer term 2019 288
 TB ,TC ,TD corner periods
(depends on the soil class)

β0 damping correction factor with a reference value of β0 = 2,5 für 5 % viscous damping

q behavior factor (according to DIN 4149, 8 to 12, depends on the regularity

of the building, on the structural type (frame/walls), on the material behavior):
Concrete 1,0 ≤ q ≤ 3,0
Steel 1,0 ≤ q ≤ 8,0
Timber 1,0 ≤ q ≤ 4,0

The energy dissipated due to the formation of plastic hinges cannot be

represented with the linear approach. As an approximation the factor q
is used to describes the amplitude reduction due to favorable dissipative
effects. With sufficient structural ductility, the design forces are smaller than
in the linear-elastic case (which is for q = 1) due to the plastic behavior.

The seismic excitation Fb is estimated according to the spectral value Sd (T ) of the design
response spectrum, depending on the natural period and structural damping:

Fb = Sd (T ) · M · λ

with

M toal mass of the building

λ 0,85 ≤ λ ≤ 1

The factor λ is introduced to consider the difference between effective modal mass and building
mass.

15.2.3 Constructive notes (short excerpt from DIN 4149 and EC 8)

• Structural forms with stable box-behavior

• Simple geamotries, uniformity, symmetry, redundancy
• Resolution of angled structural parts in cuboid

Structural Dynamics 15 Earthquakes

summer term 2019 289
 • No large lumped masses on the extremity of slender structures
• Even distribution of stiffening elements
• Prevention of torsional vibrations → center of mass near center of rigidity
• Connections and joints carefully done, including ring anchors
• Mindestauflagertiefen
• No discontinuities in stiffness and mass distribution
• Joints sufficiently wide, i. Joint width> 1.5 of the max. displ.; at least 2cm
• Foundation should not be placed at different depths using different foundation elements

A sufficient ductility needs to be ensured. Particular attention needs to be paid to avoiding

concrete compressive fracture and to keeping the foundation in the elastic range.

15.3 Capacity design

15.3.1 Basics

For a given earthquake, a structure can be designed differently. A purely elastic design of
structures for stronger earthquakes can lead to over-conservative constructions. Alternatively,
also an elastic-plastic design with lower structural resistance but planned plastic deformations,
is possible. This assumes, that damages are acceptable–without collapse–and requires very
careful planing and constructive design of the building.
The essential prerequisite for the reliability of the utilization of elastic-plastic reaction of the
structure is a sufficiently high structural resistance against horizontal forces and a sufficient
ductility. For general orientation the following rule holds:
“Quality” of earthquake behavior ≈ load resistance · ductility
Qualitatively “equal” possibilities are shown in the following sketch:

Structural Dynamics 15 Earthquakes

summer term 2019 290
 Resistance high resistance
low ductility,
no plastic deformation
medium resistance
A1 medium ductility
low resistance
A2
high ductility, large
A3 plastic deformation
Displacement
(global displacement)

Force Ductility

] Fracture µ= (..)u
(..)y

Displacement
(..)y (..)u

The ductility gives the ratio of the total elastic-plastic deformation to the elastic deformation
at the beginning of the yield process.
It can be introduced for different values, e.g. displacement ductility (global ductility), strain
ductility, rotational ductility, or curvature ductility (local ductility). Due to the narrow local
limitation of the plasticized region, the curvature ductility of a structure exceeds the displace-
ment ductility.
In the case of earthquakes the structure needs to behave sufficiently ductile even for repeated
cyclic loading. Here is a difference to inelastic states, that are due to static loads. For steel
structures a high ductility can be obtained, due to the good deformability of the material.
However, this requires a careful constructive design to avoid local instabilities (as buckling or
tipping). For concrete structures the plastic deformability is limited, nevertheless for a suitable
reinforcement design, limited ductilities can be achieved. As unreinforced masonry structures
show almost no ductility, in earthquake regions, a limitiation to few floors is necessary. For
high earthquake danger, the ductility can be increased through reinforcement. The ductility
for timber constructions is mainly given by the plastic deformability of the joints, the material
itself shows almost no plastic deformability.

Structural Dynamics 15 Earthquakes

summer term 2019 291
15.3.2 Reduction factors

For the reduction of the stress resultatnts corresponding the to the effect of a plastification,
different approaches can be found in the literature. The approximate treatment of the non-
linearity is proposed “statically” by a comparison of the energy contents. Taking as a basis
ideally elastic-plastic behavior, a linear calculation leads to a fictitious stress resultant Felmax ,
that would have to be resisted under earthquake loading.

F
upl
Fel,max µ= uel

Fpl

u
uel upl

Actually, the force cannot occur because of the plastification, but instead just Fpl as an approx-
imation of the stress resultant for elastic-plastic behavior. The requirement of equal deforma-
tion energy in both cases leads to very rough and mechanically hardly convincing estimation
u
for the internal forces under plastification. The decisive value is the ductility µ = upl el
.

1
!
Fel,max
(Fel, max + Fpl ) · uel − 1 ≈ Fpl (upl − uel )
2 Fpl
Fel,max
→ Fpl = √
2µ − 1

It needs to be considered, that with this static “energy balance”, the dynamic behavior can
only be predicted in a relative rough approximation.
There is furthermore the possibility to work with inelastic design response spectra, which al-
ready incorporate the corresponding reductions. Apparently the aforementioned procedures
lead to plausible, but calculatory not exactly verifiable values (there it needs to be consid-
ered, that an “exact calculation” can nevertheless only give an illusory precision, given the
uncertainty with respect to the uncertainties in the loading).

Structural Dynamics 15 Earthquakes

summer term 2019 292
15.3.3 The capacity design
Example 15.1: Chain under static loading

To justify the capacity design, a chain under static loading is considered exemplarily.

gleiche Glieder
F1
F1 >> F2
ein duktiles Glied
F2
i

With similarly elastic chain links, the necessary elongation would only be achievable by
a relatively large force F1 . Every link in the chain is loaded by the same force. Under
the assumption of brittle chain elements, all chain elements must remain in the linear
domain and their resistance had to be set high above this force F1 . With a ductile
chain link i, the same given end displacement ∆ can be achieved with a much lower
force F2 , compared to the case where all links, including i, behave elastically. All links
then have a much lower stress level and can be designed with less resistance. In the
case of dynamic loading, the corresponding correctly designed and constructed ductile
element, results in an additional high damping, thanks to the hysteretic behavior.

The reduction of the loading is only then permissible, if the structure has the necessary ductil-
ity. This requires, to already define the locations, where plastification shall occur purposefully.
At the same time, all other structural elements shall still behave elastically and only fail, when
the plastic deformation capacity of the dissipative elements is reached. This means, that a
dissipative structural element has to be designed such that it deforms sufficiently and thereby
dissipates energy. For the desired ductility, limit values have to be fulfilled for certain con-
struction forms. The plastification should be preferably distributed over various locations, as
depicted in the system below.

Structural Dynamics 15 Earthquakes

summer term 2019 293
 Example 15.2: Choice of mechanisms

ungeeignet geeignet

The procedure of the planing for the capacity design, is as follows:

1. design of the building

2. derivation of the structural model and calculation of the internal forces, e.g. with
the response spectrum method
3. dimensioning of the structural elements
• choice of a mechanism (where should the structure plasticize and where not!)
• constructive design of the plasticizing regions. The ductile structural element
must be designed constructively, such that it has the required high plastic
deformation capability under alternating loading. The reduction of bearing
capacity has to be considered. For example, it has to be taken care, that
the concrete compression zone cannot burst (Intention is to create inelastic
alternating demands).
• constructive design of the remaining elastic parts using the rules of elastic
design and considering possible post-limit stiffnesses of the plasticized re-
gions. The elastic, comparably “brittle” elements have to designed such hat
their behavior stays elastic with sufficient certainty, even if the yield strength
of the plastic elements exceed their planned values. Therefore, a “strength-
ening factor” (between 1,5 and 2,0) is introduced. This relatively high value
results, because the material strength of the ductile elements can be higher
than the nominal values. .

For the “capacity design” further questions concerning the constructive design are significant,
comp.,e.g., Bachmann [2013]. The procedure with elastic response spectra and damping coef-
ficients is based on the “natural” ductility, that is specifically used by the capacity design.
For important buildings under extreme earthquake loads very complex non-linear investigation
can thus become necessary. For the necessary detailed dynamic calculations with time step

Structural Dynamics 15 Earthquakes

summer term 2019 294
procedures, accelerograms are needed. If appropriate recorded accelerograms are not available,
artificial accelerograms can be created numerically, and scaled to match a specific design
response spectrum.

Structural Dynamics 15 Earthquakes

summer term 2019 295
16 Numerical time step procedures

16.1 General remarks

16.1.1 Overview

Different time step procedures (step by step Methods), such as the Newmark-β method, the
Wilson-θ method or the Houbold method, and the Generalized-α method and the central dif-
ferences method can be applied. In the scope of these methods, the time domain is discretized
into finite time intervals and the equations of motion are then only fulfilled in an approximative
manner.
The accuracy of the solution can in general be increased, by applying smaller time intervals.
However, the computing time strongly increases with the number of time steps. For large
calculation tasks, it is necessary, to find an optimum between the requirements with respect
to the accuracy and the required computation time.

16.1.2 Explicit - Implicit

The approaches can—in general—be classified as explicit or implicit. With an explicit method,
the response values are calculated only on the basis of the known values of the immediately
preceding time step. These procedures move from one time step to the next. With an implicit
procedure, e.g. the Newmark-β method, the response values are obtained from both, the
immediately preceding time step and the following time step.

16.1.3 Errors

Errors can occur due to rounding or too large time intervals, that are not able to capture
the time dependency of the response. These errors can introduce phase errors, changes in the
frequencies and also artificial (positive or negative) damping.

Structural Dynamics
summer term 2019
296
16.2 Newmark -β method

16.2.1 General approach

The general implicit description reads:

ẇn+1 = ẇn + (1 − γ) ∆tẅn + γ∆tẅn+1 (16.1)

1
 
wn+1 = wn + ∆tẇn + − β ∆t2 ẅn + β ∆t2 ẅn+1 (16.2)
2

16.2.2 Constant acceleration over the time interval

In the scope of this method, two parameters β und γ have to be chosen. For a simplified
variant, they are chosen to be β = 14 und γ = 21 , which correspond to a constant acceleration
over the time interval.

ẅ(τ )

Acceleration ẅn+1
(constant)
w¨n ẅaverage = 21 (ẅn + ẅn+1 )

ẇ(τ )
ẇ(τ ) = ẇn + τ2 (ẅn + ẅn+1 )
Velocity
(linear) ẇn+1 = ẇn + ∆t
(ẅn + ẅn+1 )
2
ẇn

w(τ ) w(τ ) = wn + ẇn τ + τ2

(ẅn + ẅn+1 )
4

Displacement
∆t2
(quadratic) wn+1 = wn + ẇn ∆t + 4
(ẅn + ẅn+1 )
wn
t=n t + ∆t = n + 1
∆t

Structural Dynamics 16 Numerical time step procedures

summer term 2019 297
With γ = 21 , the acceleration can be written as

1
ẅ(τ ) = (ẅn + ẅn+1 ).
2

This leads to the velocity

1
ẇ(τ ) = ẇn + τ (ẅn + ẅn+1 ),
2

and the displacement

1
w(τ ) = wn + ẇn τ + τ 2 (ẅn + ẅn+1 ).
4

These relations can be put into a explicit from (comp. with the expression in the above figure):

4 4
ẅn+1 = (w n+1 − w n ) − ẇn − ẅn (16.3)
∆t2 ∆t

After inserting Eq. (16.3) into Eq. (16.1), it follows for the velocity:

2
ẇn+1 = (wn+1 − wn ) − ẇn (16.4)
∆t

Further inserting the so obtained expressions for acceleration and velocity at time tn+1 into
the dynamic equilibrium, it reads

mẅn+1 + cẇn+1 + kwn+1 = fn+1 (16.5)

This equation can be transformed into a linear equation, where all elements, that belong the
the preceeding time step tn , are brought to the right hand side:

K̄wn+1 = f¯n+1

with

2c 4m
K̄ = k + + 2
∆t ∆t

and

2wn 4 4
   
f¯n+1 = fn+1 + c + ẇn + m wn + ẇn + ẅn .
∆t ∆t 2 ∆t

Structural Dynamics 16 Numerical time step procedures

summer term 2019 298
Thus, K̄, is the same for all time steps.
The so obtained linear equation can be solved for each time step. In order to calculate the
generalized force f¯n+1 , the velocity ẇn and the acceleration ẅn in the preceding time step are
determined according to Eqs. (16.3) and (16.4), respectively. The acceleration ẅn+1 is then
calculated from the equation of motion (Eq. (16.5)). In the first time step, w0 and ẇ0 are
determined from the initial conditions and ẅ0 is calculated from the equation of motion.

1
ẅ0 = (f0 − cẇ0 − kw0 ) .
m

16.2.3 Predictor/ corrector approach

Additionally a so called predictor/corrector approach is applicable. In the scope of this ap-

proach we start with the initial values for the velocity and the displacement (Eqs. (16.1) and
(16.2)), whereby the terms, that correspond to the following time step (underlined terms), are
neglected at first.

1
 
wen+1 = wn + ∆tẇn + − β ∆t2 ẅn + β ∆t2 ẅn+1 (16.6)
2
ẇ n+1 = w˙n + (1 − γ) ∆tẅn + γ∆tẅn+1 (16.7)
e

With these predictor values, the spring and damping forces are determined:

f¯i,n+1 = c ẇ˜n+1 + k w̃n+1

With these values, the acceleration at time tn+1 can be calculated. With

mẅn+1 + cẇ˜n+1 + k ẅ˜n+1 = fn+1

follows:

1 
fn+1 − f¯i,n+1

ẅn+1 =
m

The predictor values are used for the corrector step. For this purpose, the underlined terms
in Eqs. (16.1) and (16.2) are added.

ẇn+1 = ẇ˜n+1 + γ∆tẅn+1 and wn+1 = w̃n+1 + β∆t2 ẅn+1

Structural Dynamics 16 Numerical time step procedures

summer term 2019 299
16.2.4 Stability considerations

For the investigation of possibly occurring numerical instabilities, the homogeneous solution
fo the undamped motion is considered:

ẅn+1 + ωE2 wn+1 = 0 (16.8)

Again assuming a constant acceleration (corresponding to Eq. (16.3)), it follows

4wn+1 4wn 4ẇn

ẅn+1 = − − − ẅn (16.9)
∆t2 ∆t2 ∆t

and with Eq. (16.4)

2wn+1 2wn
ẇn+1 = − − w˙n (16.10)
∆t ∆t

and finally, inserting Eq. (16.9) into Eq. (16.8), we obtain

4 4wn 4ẇn
 
+ ωE2 wn+1 − − − ẅn = 0.
∆t 2 ∆t2 ∆t

This can alternatively be written as

1 4ẇn 4wn
 
wn+1 =   ẅn + + . (16.11)
4
+ ωE
2 ∆t ∆t2
∆t2

with ωE = 2π
TE
. The expression in Eq. (16.11), has the form

wn+1 = b31 ẅn + b32 ẇn + b33 wn .

Inserting the above into Eqs. (16.10) and (16.9), it follows

ẇn+1 = b21 ẅn + b22 ẇn + b23 wn

ẅn+1 = b11 ẅn + b12 ẇn + b13 wn .

In matrix notation this relationship is

Zn+1 = BZn

Structural Dynamics 16 Numerical time step procedures

summer term 2019 300
with
   
ẅn b11 b12 b13
Zn = ẇn  B = b21 b22 b23 
.
  

wn b31 b32 b33

After n time steps, the solution is Zn = [B]n Zo . The stability of the method depends on
the behavior of Bn for n → ∞. Since the coefficients of Bn depend on ωE and ∆t or TE ,
respectively, limit values can be defined, for which infinite growth of [B]n is excluded. The
Newmark-β method with β = 41 and γ = 21 is (absolute) stable, independently on the choice
of the time step ∆t. Absolute stability implies, that the result is limited independently on the
choice of the time step. i.e., the calculated values lie in the range of the static deformation of
the system, thtat result from wstat = fmaxk
.
For β = 61 and γ = 21 (linear acceleration) the Newmark-β method is only conditionally stable.
The linear acceleration approach leads for the same choice of ∆t und under consideration of
the stability criteria to better results. Nevertheless, these conclusions are only valid for linear
systems.
In general, the time step size has to be chosen such that the load characteristic und the
characteristic of the free vibration can be represented accurately enough.

16.3 The central difference method

16.3.1 Derivation of ẅ and ẇ out of w

With the central difference method, the numerical approaches for the discrete computation of
the first and second derivative are used:

1
ẇn = (−wn−1 + wn+1 ) (16.12)
2∆t
1
ẅn = (wn−1 − 2wn + wn+1 ) (16.13)
∆t2

For both expansions, the error is of order (∆t)2 .

16.3.2 General approach

The displacement for the time step n + 1 is obtained from the equilibrium at time step n:

mẅn + cẇn + kwn = f . (16.14)

Structural Dynamics 16 Numerical time step procedures

summer term 2019 301
Inserting Eqs. (16.12) and (16.13) into Eq. (16.14), it follows:

1 1 2 1 1
     
m+ c wn+1 = f − k − 2 m wn − m− c wn−1 .
∆t 2 2∆t ∆t ∆t 2 2∆t

This equation can be solved for wn+1 . The solution wn+1 thus depends on the application
of the equilibrium condition or equations of motion at the time step n. For this reason, the
method is referred to as an explicit integration method.

16.3.3 Start procedure

From the previous derivation it is apparent, that for the calculation of wn+1 always wn and
wn−1 are needed. For the computation at time step n = 0, thus, a separate start procedure
is necessary, because w0 as well as ẇ0 und ẅ0 are known, but w−1 not. w−1 can then be
calculated as follows:

∆t2
w−1 = w0 − ∆tẇ0 + ẅ0 .
2

It should be mentioned, that ẅ0 can be calculated from w0 and ẇ0 with the help of the
equilibrium equation.

16.4 Non-linear problems

In the following figures, a non-linear SDOF system is depicted. The free body cut shows the
corresponding forces. Non-linear dependencies between the forces and the displacements or
the velocities are depicted below.

f (t)
F (t)

m
FI

Fc Fk

Structural Dynamics 16 Numerical time step procedures

summer term 2019 302
 Fk Fc

w ẇ
wn wn+1 ẇn ẇn+1

The dynamic equilibrium for the time steps tn+1 and tn are

fIn+1 + fcn+1 + fkn+1 = fn+1 , (16.15)

f In + f cn + f k n = f n , (16.16)

with

fIn+1 − fIn = m∆ẅ (16.17)

dfc
fcn+1 − fcn ≈ ∆ẇ = cn ∆ẇ(t) (16.18)
dẇ
dfc
!
cn = (16.19)
dẇ n
dfk
fkn+1 − fkn ≈ ∆w = kn ∆w(t) (16.20)
dw
dfk
!
kn = (16.21)
dw n

fn+1 − fn = ∆fn (16.22)

Subtracting Eq. (16.15) from Eq. (16.16), an incremental representation is obtained:

m∆ẅ + cn ∆ẇ + kn ∆w = ∆fn . (16.23)

On this basis, now the Newmark-β method can be applied. For β = 1

4
and γ = 1
2
and with

wn+1 = wn + ∆w (16.24)

and

k̄wn+1 = f¯n+1 (16.25)

k̄wn = f¯n (16.26)

Structural Dynamics 16 Numerical time step procedures

summer term 2019 303
it follows:

k̄∆w = ∆f¯, (16.27)

with

2cn 4m
k̄ = kn + + 2
∆t ∆t
4w˙n
 
∆f¯ = ∆f + m + ẅn + cn (w˙n ) .
∆t

Structural Dynamics 16 Numerical time step procedures

summer term 2019 304
Bibliography
[Bachmann 2013] Bachmann, Hugo: Erdbebensicherung von Bauwerken. Springer-
Verlag, 2013

[Balendra 1993] Balendra, Thambirajah: Vibration of buildings to wind and earth-

quake loads. Springer-Verlag, 1993

[Bathe 2007] Bathe, K.J.: Finite-Elemente-Methoden. Springer, 2007

[Bendat and Piersol 2010] Bendat, Julius S. ; Piersol, Allan G.: Random Data -
Analysis and Measurement Procedures. John Wiley & Sons, 2010 (Wiley Series in
Probability and Statistics)

[Buchschmid 2011] Buchschmid, M.: ITM-Based FSI-Models for Rooms with Ab-
sorptive Boundaries, TU München, Ph.D. thesis, 2011

[Buchschmid 2013a] Buchschmid, M.: Integral Transform Methods - Theory and

Application (lecture notes). TU München, 2013

[Buchschmid 2013b] Buchschmid, M.: Vibroacoustics Lab (lecture notes). TU

München, 2013

[Buchschmid et al 2010] Buchschmid, M. ; Pospiech, M. ; Müller, G.: Coupling

Impedance Boundary Conditions for Absorptive Structures with Spectral Finite Elements
in Room Acoustical Simulations. In: Computing and Visualization in Science 13 (7)
(2010), p. 355–363

[Burton et al 2001] Burton, Tony ; Jenkins, Nick ; Sharpe, David ; Bossanyi, Ervin:
Wind Energy Handbook. Wiley, 2001

[Butz and Distl 2008] Butz, C. ; Distl, J.: Stahlbau-Kalender 2008. Chap. Person-
eninduzierte Schwingungen von Fußgängerbrücken, p. 699–744, Ernst & Sohn, 2008

[Chopra 1995] Chopra, Anil K.: Dynamics of structures: theory and applications
to earthquake engineering. Volume 2. Prentice Hall Englewood Cliffs, NJ, 1995

[Crandall 1961] Crandall, Stephen H.: Dynamic response of systems with structural
damping. 1961. – Research Report

Structural Dynamics
summer term 2019
305
[Cremer et al 2005] Cremer, L. ; Heckl, M. ; Petersson, B.A.T.: Structure-Borne
Sound. Springer, 2005

[DIN 2004a] DIN: DIN ISO 1940-1:2004-04 Mechanische Schwingungen; An-

forderungen an die Auswuchtgüte von Rotoren in konstantem (starrem) Zu-
stand; Teil 1: Festlegung und Nachprüfung der Unwuchttoleranz (ISO 1940-
1:2003). Beuth Verlag Berlin, 2004

[DIN 2004b] DIN ; DIN (Editor.): VDI 3838:2004-05 Messung und

Beurteilung mechanischer Schwingungen von Hubkolbenmotoren und -
kompressorenompressoren mit Leistungen über 100 kW; Ergänzung von DIN
ISO 10816-6. Berlin: Beuth Verlag, 2004

[DIN 2006] DIN: DIN EN 1998-1:2006-04 Eurocode 8: Auslegung von Bauwerken

gegen Erdbeben; Teil 1: Grundlagen, Erdbebeneinwirkungen und Regeln für
Hochbauten; Deutsche Fassung EN 1998-1:2004. Beuth Verlag Berlin, 2006

[DIN 1055-4 2005] DIN 1055-4: DIN 1055-4:2005-03 Einwirkungen auf Tragwerke;
Teil 4: Windlasten. Beuth Verlag Berlin, 2005

[DIN 4024-1 1988] DIN 4024-1: DIN 4024-1:1988-04 Maschinenfundamente;

Elastische Stützkonstruktionen für Maschinen mit rotierenden Massen. Beuth
Verlag Berlin, 1988

[DIN 4024-2 1991] DIN 4024-2: DIN 4024-2:1991-04 Maschinenfundamente; Steife

(starre) Stützkonstruktionen für Maschinen mit periodischer Erregung. Beuth
Verlag Berlin, 1991

[DIN 4178 2005] DIN 4178: DIN 4178:2005-04 Glockentürme. Beuth Verlag Berlin,
2005

[DIN EN 1991-1-4:2010-129 2010] Eurocode 1:Einwirkung auf Tragwerke; Teil 1-4: Wind-
lasten; Deutsche Fassung EN 1991-1-4:2005. 2010. – Standard

[EN 2010] EN, DIN: Eurocode 1: Einwirkungen auf Tragwerke–Teil 1-4: Allgemeine
Einwirkungen–Windlasten. (2010)

[Flesch 1993] Flesch: Baudynmik. 1993

[Grundmann et al 1983] Grundmann, H. ; Kreutzinger, H. ; Baumgärnter, W. ;

Albertshofer, K. ; Konrad, A. ; Müller, F.: Einführung in die Baudynamik. In:
Mitteilungen aus dem Institut für Bauingenieurwesen I der Technischen Uni-
verstität München Heft 9 (1983)

[Holm 2005] Holm, D. D.: Taylor0 s hypothesis, Hamilton0 s principle and the LANS-alpha
model for computing turbulence. In: Los Alamos Science 29 (2005), p. 172–180

Structural Dynamics Bibliography

summer term 2019 306
[Holmes 2007] Holmes, John D.: Wind Loading of Structures. Taylor & Francis, 2007

[Humar 2012] Humar, Jagmohan: Dynamics of structures. CRC press, 2012

[ISO 4354 2009] ISO 4354: ISO 4354:2009-06 Wind actions on structures. Genf:
ISO, 2009

[Kausel 2017] Kausel, Eduardo: Advanced structural dynamics. Cambridge University

Press, 2017

[Kolousek 1973] Kolousek, V.: Dynamics in Engineering Structures. Butterworths,

1973

[Kreutz 2013] Kreutz, J.: Augmented Beam Elements Using Unit Deflection
Shapes Together with a Finite Element Discretisation of the Cross Section.,
Techische Universität München, Ph.D. thesis, 2013

[Langley and Bremner 1999] Langley, Robin S. ; Bremner, Paul: A hybrid method for
the vibration analysis of complex structural-acoustic systems. In: The Journal of the
Acoustical Society of America 105 (1999), Nr. 3, p. 1657–1671

[Langley 1990] Langley, RS: Analysis of power flow in beams and frameworks using the
direct-dynamic stiffness method. In: Journal of Sound and Vibration 136 (1990), Nr. 3,
p. 439–452

[Li and Chen 2009] Li, J. ; Chen, J.: Stochastic Dynamics of Structures. Wiley, 2009

[Lutes and Sarkani 2004] Lutes, Loren D. ; Sarkani, Shahram: Random Vibrations:
Analysis of Structural and Mechanical Systems. Elsevier Science, 2004

[Maybeck 1979] Maybeck, P. S.: Stochastic Models, Estimation and Control, Vol-
ume 1. Academic Press, Mathematics in Science and Engineering, 1979

[Melke 1995] Melke, J: Erschütterungen und Körperschall des landgebunde-

nen Verkehrs, Prognose und Schutzmaßnahmen, Materialien Nr. 22,. Lan-
desumweltamt, Essen, 1995

[Mendis et al 2007] Mendis, P. ; Ngo, T. ; Haritos, N. ; Hira, A. ; Samali, B. ; Cheung,

J.: Wind loading on tall buildings. In: EJSE Special Issue: Loading on Structures 3
(2007), p. 41–54

[Müller 2009] Müller, G.: Abläufe bei baudynamischen Prognosen zur Gebrauch-
stauglichkeit. In: Baudynamik, VDI Bericht 2063, VDI Verlag, Baudynamik Düsseldorf,
2009, p. 627–637. – Print-ISBN: 978-3-18-092063-4

Structural Dynamics Bibliography

summer term 2019 307
[Müller 2011] Müller, G.: Erschütterungsentstehung und -ausbreitung. In: Lutzen-
berger, S. (Editor.) ; Müller, G. (Editor.) ; Lölgen, Th. (Editor.) ; Eichenlaub,
Ch. (Editor.): Fachtagung Bahnakustik 2011 Infrastruktur, Fahrzeuge, Betrieb,
München, MüllerBBM. Planegg : Bahn Fachverlag GmbH, 2011, p. 74–83

[Müller 2013a] Müller, G.: Baudynamik (lecture notes). TU München, 2013

[Müller 2013b] Müller, G.: Continuum Mechanics (lecture notes). TU München,

2013

[Müller 2013c] Müller, G.: Soil Dynamics (lecture notes). TU München, 2013

[Müller 2013d] Müller, G.: Technical Acoustics (lecture notes). TU München, 2013

[Müller and Möser 2004] Müller, G. ; Möser, M.: Taschenbuch der Technischen
Akustik. 3. Berlin Heidelberg New York : Springer Verlag, 2004

[Müller-Boruttau and Breitsamter 2000] Müller-Boruttau, F.H. ; Breitsamter, N.:

Elastische Elemente im Gleis verringern die Beanspruchung von Oberbau, Unterbau und
Untergrund. In: Eisenbahntechnische Rundschau 49 (2000), p. 587–596

[Nashif et al 1985] Nashif, Ahid D. ; Jones, David I. ; Henderson, John P. et al: Vi-
bration damping. John Wiley & Sons, 1985

[Peil and Clobes 2008] Peil, U. ; Clobes, M.: Stahlbau-Kalender. Chap. Dynamische
Windwirkungen, p. 441–478, Ernst & Sohn, 2008

[Petersen 1996] Petersen, C.: Dynamik der Baukonstruktionen. Vieweg, 1996

[Petersen 2000] Petersen, Christian: Dynamik der Baukonstruktionen. Vieweg &

Teubner Verlag, 2000

[Roberts and Spanos 2003] Roberts, J. B. ; Spanos, P. D.: Random Vibration and
Statistical Linearization. Dover Publication, 2003

[Runtemund 2013] Runtemund, K.: Output-only Measurement-Based Parameter

Identification of Dynamic Systems Subjected to Random Load Processes, TU
München, Ph.D. thesis, 2013

[Ruscheweyh 1982] Ruscheweyh, H.: Dynamische Windwirkung an Bauwerken,

Bd. 1 und 2. Bauverlag Wiesbaden/Berlin, 1982

[Sahnaci 2013] Sahnaci, C.: Menscheninduzierte Einwirkungen auf Tragwerke in-

folge der Lokomotionsformen Gehen und Rennen: Analyse und Modellierung,
Ruhr-Universität Bochum, Ph.D. thesis, 2013

[Snowdon 1968] Snowdon, John C.: Vibration and shock in damped mechanical
systems. J. Wiley, 1968

Structural Dynamics Bibliography

summer term 2019 308
[Soize 1993] Soize, Christian: A model and numerical method in the medium frequency
range for vibroacoustic predictions using the theory of structural fuzzy. In: The Journal
of the Acoustical Society of America 94 (1993), Nr. 2, p. 849–865

[Stathopoulos and Baniotopoulos 2007] Stathopoulos, Ted ; Baniotopoulos, Char-

alambos C.: Wind Effects on Buildings and Design of Wind-Sensitive Structures.
Springer, 2007 (International Centre for Mechanical Sciences, Courses and Lectures)

[Studer et al 2007] Studer, J. A. ; Laue, J. ; Koller, M. G.: Bodendynamik Grundla-

gen, Kennziffern, Probleme und Lösungsansätze. Springer Verlag Berlin Heidelberg
New York, 2007

[Thompson 2008] Thompson, D.: Railway Noise and Vibration: Mechanisms, Mod-
elling and Means of Control. Oxford UK Elsevier Science, 2008

[Van der Hoven 1957] Van der Hoven, Isaac: Power spectrum of horizontal wind speed
in the frequency range from 0,0007 to 900 cycles per hour. In: Journal of Meteorology
14 (1957), p. 160–164

[VDI 2038 2012] VDI 2038: VDI 2038: Gebrauchstauglichkeit von Bauwerken
bei dynamischen Einwirkungen - Untersuchungsmethoden und Beurteilungsver-
fahren der Baudynamik - Teil 1 Grundlagen - Methoden, Vorgehensweisen und
Einwirkungen. Verein Deutscher Ingenieure, 2012

[Weaver 1997] Weaver, RL: Mean and mean-square responses of a prototypical master/-
fuzzy structure. In: The Journal of the Acoustical Society of America 101 (1997),
Nr. 3, p. 1441–1449

[Wettschureck et al 2012] Wettschureck, R. G. ; Hauck, G. ; Diehl, R. ; Willenbrink,

Ludger: Handbook of Engineering Acoustics. Chap. Noise and Vibration from Railroad
Traffic, p. 393–488, Springer Heidelberg New York Dordrecht London, 2012

[Wirsching et al 2006] Wirsching, P. H. ; Paez, T. L. ; Ortiz, K.: Random Vibrations:

Theory and Practice. Dover Publications, 2006

[Zienkiewicz et al 1977] Zienkiewicz, C ; Kelly, Donald W. ; Bettess, Peter: Marriage

a la mode: The best of both worlds- Finite elements and boundary integrals. In: Inter-
national Symposium on Innovative Numerical Analysis in Applied Engineering
Science, Versailles, France, 1977, p. 19–26

Structural Dynamics Bibliography

summer term 2019 309